From d8c9166961437263b8dd97ff8341c3335d894740 Mon Sep 17 00:00:00 2001 From: Dotan Barak Date: Thu, 23 Aug 2012 14:09:02 +0000 Subject: [PATCH 01/45] IB/core: Remove unused variables in ucm/ucma Remove unused wait objects from ucm/ucma events flow. Signed-off-by: Dotan Barak Signed-off-by: Or Gerlitz Acked-by: Sean Hefty Signed-off-by: Roland Dreier --- drivers/infiniband/core/ucm.c | 1 - drivers/infiniband/core/ucma.c | 1 - 2 files changed, 2 deletions(-) diff --git a/drivers/infiniband/core/ucm.c b/drivers/infiniband/core/ucm.c index 06f08713f487..49b15ac1987e 100644 --- a/drivers/infiniband/core/ucm.c +++ b/drivers/infiniband/core/ucm.c @@ -397,7 +397,6 @@ static ssize_t ib_ucm_event(struct ib_ucm_file *file, struct ib_ucm_event_get cmd; struct ib_ucm_event *uevent; int result = 0; - DEFINE_WAIT(wait); if (out_len < sizeof(struct ib_ucm_event_resp)) return -ENOSPC; diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 055ed59838dc..7972bae2e9b3 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -310,7 +310,6 @@ static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf, struct rdma_ucm_get_event cmd; struct ucma_event *uevent; int ret = 0; - DEFINE_WAIT(wait); if (out_len < sizeof uevent->resp) return -ENOSPC; From 2a22fb8c69275903b8be4c6203aa08bfac374844 Mon Sep 17 00:00:00 2001 From: Dotan Barak Date: Thu, 30 Aug 2012 09:09:55 +0000 Subject: [PATCH 02/45] RDMA/cma: Use consistent component mask for IPoIB port space multicast joins CMA multicast joins for the IPoIB port space need to use the same component mask used by the ipoib driver. Otherwise, it's possible for the CMA to create a group to which a join made by ipoib will fail, or vise-versa. Some of the component mask fields set by ipoib weren't set by the CMA, fix that. Signed-off-by: Dotan Barak Reviewed-by: Jack Morgenstein Acked-by: Sean Hefty Signed-off-by: Or Gerlitz Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 7172559ce0c1..26b37603dcf1 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -3058,7 +3058,10 @@ static int cma_join_ib_multicast(struct rdma_id_private *id_priv, if (id_priv->id.ps == RDMA_PS_IPOIB) comp_mask |= IB_SA_MCMEMBER_REC_RATE | - IB_SA_MCMEMBER_REC_RATE_SELECTOR; + IB_SA_MCMEMBER_REC_RATE_SELECTOR | + IB_SA_MCMEMBER_REC_MTU_SELECTOR | + IB_SA_MCMEMBER_REC_MTU | + IB_SA_MCMEMBER_REC_HOP_LIMIT; mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device, id_priv->id.port_num, &rec, From c079c28714e4d1a0b7cad120f84217e0fcde09a6 Mon Sep 17 00:00:00 2001 From: Emil Goode Date: Sun, 19 Aug 2012 17:59:40 +0000 Subject: [PATCH 03/45] RDMA/cxgb4: Fix error handling in create_qp() The variable ret is assigned return values in a couple of places, but its value is never returned. This patch makes use of the ret variable so that the caller get correct error codes returned. The following changes are also introduced: - The alloc_oc_sq function can return -ENOSYS or -ENOMEM so we want to get the return value from it. - Change the label names to improve readability. Signed-off-by: Emil Goode Acked-by: Steve Wise Signed-off-by: Roland Dreier --- drivers/infiniband/hw/cxgb4/qp.c | 62 +++++++++++++++++++------------- 1 file changed, 38 insertions(+), 24 deletions(-) diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c index 45aedf1d9338..e2bf9c68cfc8 100644 --- a/drivers/infiniband/hw/cxgb4/qp.c +++ b/drivers/infiniband/hw/cxgb4/qp.c @@ -137,19 +137,25 @@ static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, return -ENOMEM; wq->rq.qid = c4iw_get_qpid(rdev, uctx); - if (!wq->rq.qid) - goto err1; + if (!wq->rq.qid) { + ret = -ENOMEM; + goto free_sq_qid; + } if (!user) { wq->sq.sw_sq = kzalloc(wq->sq.size * sizeof *wq->sq.sw_sq, GFP_KERNEL); - if (!wq->sq.sw_sq) - goto err2; + if (!wq->sq.sw_sq) { + ret = -ENOMEM; + goto free_rq_qid; + } wq->rq.sw_rq = kzalloc(wq->rq.size * sizeof *wq->rq.sw_rq, GFP_KERNEL); - if (!wq->rq.sw_rq) - goto err3; + if (!wq->rq.sw_rq) { + ret = -ENOMEM; + goto free_sw_sq; + } } /* @@ -157,15 +163,23 @@ static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, */ wq->rq.rqt_size = roundup_pow_of_two(wq->rq.size); wq->rq.rqt_hwaddr = c4iw_rqtpool_alloc(rdev, wq->rq.rqt_size); - if (!wq->rq.rqt_hwaddr) - goto err4; + if (!wq->rq.rqt_hwaddr) { + ret = -ENOMEM; + goto free_sw_rq; + } if (user) { - if (alloc_oc_sq(rdev, &wq->sq) && alloc_host_sq(rdev, &wq->sq)) - goto err5; + ret = alloc_oc_sq(rdev, &wq->sq); + if (ret) + goto free_hwaddr; + + ret = alloc_host_sq(rdev, &wq->sq); + if (ret) + goto free_sq; } else - if (alloc_host_sq(rdev, &wq->sq)) - goto err5; + ret = alloc_host_sq(rdev, &wq->sq); + if (ret) + goto free_hwaddr; memset(wq->sq.queue, 0, wq->sq.memsize); dma_unmap_addr_set(&wq->sq, mapping, wq->sq.dma_addr); @@ -173,7 +187,7 @@ static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, wq->rq.memsize, &(wq->rq.dma_addr), GFP_KERNEL); if (!wq->rq.queue) - goto err6; + goto free_sq; PDBG("%s sq base va 0x%p pa 0x%llx rq base va 0x%p pa 0x%llx\n", __func__, wq->sq.queue, (unsigned long long)virt_to_phys(wq->sq.queue), @@ -201,7 +215,7 @@ static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, skb = alloc_skb(wr_len, GFP_KERNEL); if (!skb) { ret = -ENOMEM; - goto err7; + goto free_dma; } set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0); @@ -266,33 +280,33 @@ static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq, ret = c4iw_ofld_send(rdev, skb); if (ret) - goto err7; + goto free_dma; ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, wq->sq.qid, __func__); if (ret) - goto err7; + goto free_dma; PDBG("%s sqid 0x%x rqid 0x%x kdb 0x%p squdb 0x%llx rqudb 0x%llx\n", __func__, wq->sq.qid, wq->rq.qid, wq->db, (unsigned long long)wq->sq.udb, (unsigned long long)wq->rq.udb); return 0; -err7: +free_dma: dma_free_coherent(&(rdev->lldi.pdev->dev), wq->rq.memsize, wq->rq.queue, dma_unmap_addr(&wq->rq, mapping)); -err6: +free_sq: dealloc_sq(rdev, &wq->sq); -err5: +free_hwaddr: c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size); -err4: +free_sw_rq: kfree(wq->rq.sw_rq); -err3: +free_sw_sq: kfree(wq->sq.sw_sq); -err2: +free_rq_qid: c4iw_put_qpid(rdev, wq->rq.qid, uctx); -err1: +free_sq_qid: c4iw_put_qpid(rdev, wq->sq.qid, uctx); - return -ENOMEM; + return ret; } static int build_immd(struct t4_sq *sq, struct fw_ri_immd *immdp, From bea1e22df494a729978e7f2c54f7bda328f74bc3 Mon Sep 17 00:00:00 2001 From: Patrick McHardy Date: Thu, 30 Aug 2012 07:01:30 +0000 Subject: [PATCH 04/45] IPoIB: Fix use-after-free of multicast object Fix a crash in ipoib_mcast_join_task(). (with help from Or Gerlitz) Commit c8c2afe360b7 ("IPoIB: Use rtnl lock/unlock when changing device flags") added a call to rtnl_lock() in ipoib_mcast_join_task(), which is run from the ipoib_workqueue, and hence the workqueue can't be flushed from the context of ipoib_stop(). In the current code, ipoib_stop() (which doesn't flush the workqueue) calls ipoib_mcast_dev_flush(), which goes and deletes all the multicast entries. This takes place without any synchronization with a possible running instance of ipoib_mcast_join_task() for the same ipoib device, leading to a crash due to NULL pointer dereference. Fix this by making sure that the workqueue is flushed before ipoib_mcast_dev_flush() is called. To make that possible, we move the RTNL-lock wrapped code to ipoib_mcast_join_finish(). Signed-off-by: Patrick McHardy Cc: Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/ipoib/ipoib_main.c | 2 +- .../infiniband/ulp/ipoib/ipoib_multicast.c | 19 ++++++++++--------- 2 files changed, 11 insertions(+), 10 deletions(-) diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c index 1e19b5ae7c47..ea0dfc77a7f4 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_main.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c @@ -150,7 +150,7 @@ static int ipoib_stop(struct net_device *dev) netif_stop_queue(dev); - ipoib_ib_dev_down(dev, 0); + ipoib_ib_dev_down(dev, 1); ipoib_ib_dev_stop(dev, 0); if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c index 75367249f447..cecb98a4c662 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c @@ -175,7 +175,9 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, mcast->mcmember = *mcmember; - /* Set the cached Q_Key before we attach if it's the broadcast group */ + /* Set the multicast MTU and cached Q_Key before we attach if it's + * the broadcast group. + */ if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4, sizeof (union ib_gid))) { spin_lock_irq(&priv->lock); @@ -183,10 +185,17 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, spin_unlock_irq(&priv->lock); return -EAGAIN; } + priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu)); priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey); spin_unlock_irq(&priv->lock); priv->tx_wr.wr.ud.remote_qkey = priv->qkey; set_qkey = 1; + + if (!ipoib_cm_admin_enabled(dev)) { + rtnl_lock(); + dev_set_mtu(dev, min(priv->mcast_mtu, priv->admin_mtu)); + rtnl_unlock(); + } } if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { @@ -574,14 +583,6 @@ void ipoib_mcast_join_task(struct work_struct *work) return; } - priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu)); - - if (!ipoib_cm_admin_enabled(dev)) { - rtnl_lock(); - dev_set_mtu(dev, min(priv->mcast_mtu, priv->admin_mtu)); - rtnl_unlock(); - } - ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n"); clear_bit(IPOIB_MCAST_RUN, &priv->flags); From 46db567debb2abe7c0c03a1bd97891cbc1a2470b Mon Sep 17 00:00:00 2001 From: Dotan Barak Date: Thu, 23 Aug 2012 14:09:03 +0000 Subject: [PATCH 05/45] IB/mlx4: Fill in sq_sig_type in query QP Signed-off-by: Dotan Barak Signed-off-by: Or Gerlitz Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/qp.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index f585eddef4b7..56e66a4c335c 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -2225,6 +2225,10 @@ done: if (qp->flags & MLX4_IB_QP_LSO) qp_init_attr->create_flags |= IB_QP_CREATE_IPOIB_UD_LSO; + qp_init_attr->sq_sig_type = + qp->sq_signal_bits == cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE) ? + IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR; + out: mutex_unlock(&qp->mutex); return err; From 426dd00d76ad7c96250f5b5f871847d545fef33c Mon Sep 17 00:00:00 2001 From: Dotan Barak Date: Thu, 23 Aug 2012 14:09:04 +0000 Subject: [PATCH 06/45] mlx4_core: Fix wrong offset in parsing query device caps response The wrong offset was used when parsing the number of XRCs in mlx4_QUERY_DEV_CAP(). Signed-off-by: Dotan Barak Signed-off-by: Or Gerlitz Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/fw.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c index c69648487321..7cf8ec5f3827 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.c +++ b/drivers/net/ethernet/mellanox/mlx4/fw.c @@ -559,7 +559,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap) dev_cap->max_pds = 1 << (field & 0x3f); MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_XRC_OFFSET); dev_cap->reserved_xrcds = field >> 4; - MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET); + MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_XRC_OFFSET); dev_cap->max_xrcds = 1 << (field & 0x1f); MLX4_GET(size, outbox, QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET); From a084feebd24764fab9657a586ea65484ddbd79cf Mon Sep 17 00:00:00 2001 From: Or Gerlitz Date: Thu, 23 Aug 2012 14:09:05 +0000 Subject: [PATCH 07/45] mlx4_core: Remove annoying debug message in the resource tracker This innocent print makes it very hard to actually use the mlx4_core debug messages -- for example, the module load sequence of a device with two VFs yielded 3200 debug prints, with 2800 of them being this one. Let's just remove it. Signed-off-by: Or Gerlitz Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/resource_tracker.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c index 293c9e820c49..362b25e1f2e7 100644 --- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c +++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c @@ -360,8 +360,6 @@ static int get_res(struct mlx4_dev *dev, int slave, u64 res_id, r->from_state = r->state; r->state = RES_ANY_BUSY; - mlx4_dbg(dev, "res %s id 0x%llx to busy\n", - ResourceType(type), r->res_id); if (res) *((struct res_common **)res) = r; From d2b57063e4aba51d3c49ec957607d2e4c9d5f29a Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:37 +0000 Subject: [PATCH 08/45] IB/core: Reserve bits in enum ib_qp_create_flags for low-level driver use Reserve bits 26-31 for internal use by low-level drivers. Two such bits are used in the mlx4_b driver SR-IOV implementation. These enum additions guarantee that the core layer will never use these bits, so that low level drivers may safely make use of them. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- include/rdma/ib_verbs.h | 3 +++ 1 file changed, 3 insertions(+) diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 07996af8265a..46bc045bbe15 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -614,6 +614,9 @@ enum ib_qp_type { enum ib_qp_create_flags { IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, + /* reserve bits 26-31 for low level drivers' internal use */ + IB_QP_CREATE_RESERVED_START = 1 << 26, + IB_QP_CREATE_RESERVED_END = 1 << 31, }; struct ib_qp_init_attr { From ff7166c447df23a61e4f51bf748319dc6728dc74 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:38 +0000 Subject: [PATCH 09/45] IB/core: Handle table with full and partial membership for the same P_Key Extend the cached and non-cached P_Key table lookups to handle limited and full membership of the same P_Key to co-exist in the P_Key table. This is necessary for SR-IOV, to allow for some guests would to have the full membership P_Key in their virtual P_Key table, while other guests on the same physical HCA would have the limited one. To support this, we need both the limited and full membership P_Keys to be present in the master's (hypervisor physical port) P_Key table. The algorithm for handling P_Key tables which contain both the limited and the full membership versions of the same P_Key works as follows: When scanning the P_Key table for a 15-bit P_Key: A. If there is a full member version of that P_Key anywhere in the table, return its index (even if a limited-member version of the P_Key exists earlier in the table). B. If the full member version is not in the table, but the limited-member version is in the table, return the index of the limited P_Key. Signed-off-by: Liran Liss Signed-off-by: Jack Morgenstein Signed-off-by: Or Gerlitz Signed-off-by: Roland Dreier --- drivers/infiniband/core/cache.c | 15 ++++++++++++--- drivers/infiniband/core/device.c | 16 +++++++++++++--- 2 files changed, 25 insertions(+), 6 deletions(-) diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c index 9353992f9eea..4da381b74f54 100644 --- a/drivers/infiniband/core/cache.c +++ b/drivers/infiniband/core/cache.c @@ -167,6 +167,7 @@ int ib_find_cached_pkey(struct ib_device *device, unsigned long flags; int i; int ret = -ENOENT; + int partial_ix = -1; if (port_num < start_port(device) || port_num > end_port(device)) return -EINVAL; @@ -179,11 +180,19 @@ int ib_find_cached_pkey(struct ib_device *device, for (i = 0; i < cache->table_len; ++i) if ((cache->table[i] & 0x7fff) == (pkey & 0x7fff)) { - *index = i; - ret = 0; - break; + if (cache->table[i] & 0x8000) { + *index = i; + ret = 0; + break; + } else + partial_ix = i; } + if (ret && partial_ix >= 0) { + *index = partial_ix; + ret = 0; + } + read_unlock_irqrestore(&device->cache.lock, flags); return ret; diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c index e711de400a01..18c1ece765f2 100644 --- a/drivers/infiniband/core/device.c +++ b/drivers/infiniband/core/device.c @@ -707,18 +707,28 @@ int ib_find_pkey(struct ib_device *device, { int ret, i; u16 tmp_pkey; + int partial_ix = -1; for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) { ret = ib_query_pkey(device, port_num, i, &tmp_pkey); if (ret) return ret; - if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { - *index = i; - return 0; + /* if there is full-member pkey take it.*/ + if (tmp_pkey & 0x8000) { + *index = i; + return 0; + } + if (partial_ix < 0) + partial_ix = i; } } + /*no full-member, if exists take the limited*/ + if (partial_ix >= 0) { + *index = partial_ix; + return 0; + } return -ENOENT; } EXPORT_SYMBOL(ib_find_pkey); From 73aaa7418f8069103ca56fc620b3cd16c5a37d6e Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:39 +0000 Subject: [PATCH 10/45] IB/core: Add ib_find_exact_cached_pkey() When P_Key tables potentially contain both full and partial membership copies for the same P_Key, we need a function to find the index for an exact (16-bit) P_Key. This is necessary when the master forwards QP1 MADs sent by guests. If the guest has sent the MAD with a limited membership P_Key, we need to to forward the MAD using the same limited membership P_Key. Since the master may have both the limited and the full member P_Keys in its table, we must make sure to retrieve the limited membership P_Key in this case. Signed-off-by: Jack Morgenstein Signed-off-by: Or Gerlitz Signed-off-by: Roland Dreier --- drivers/infiniband/core/cache.c | 32 ++++++++++++++++++++++++++++++++ include/rdma/ib_cache.h | 16 ++++++++++++++++ 2 files changed, 48 insertions(+) diff --git a/drivers/infiniband/core/cache.c b/drivers/infiniband/core/cache.c index 4da381b74f54..80f6cf2449fb 100644 --- a/drivers/infiniband/core/cache.c +++ b/drivers/infiniband/core/cache.c @@ -199,6 +199,38 @@ int ib_find_cached_pkey(struct ib_device *device, } EXPORT_SYMBOL(ib_find_cached_pkey); +int ib_find_exact_cached_pkey(struct ib_device *device, + u8 port_num, + u16 pkey, + u16 *index) +{ + struct ib_pkey_cache *cache; + unsigned long flags; + int i; + int ret = -ENOENT; + + if (port_num < start_port(device) || port_num > end_port(device)) + return -EINVAL; + + read_lock_irqsave(&device->cache.lock, flags); + + cache = device->cache.pkey_cache[port_num - start_port(device)]; + + *index = -1; + + for (i = 0; i < cache->table_len; ++i) + if (cache->table[i] == pkey) { + *index = i; + ret = 0; + break; + } + + read_unlock_irqrestore(&device->cache.lock, flags); + + return ret; +} +EXPORT_SYMBOL(ib_find_exact_cached_pkey); + int ib_get_cached_lmc(struct ib_device *device, u8 port_num, u8 *lmc) diff --git a/include/rdma/ib_cache.h b/include/rdma/ib_cache.h index 00a2b8ec327f..ad9a3c280944 100644 --- a/include/rdma/ib_cache.h +++ b/include/rdma/ib_cache.h @@ -100,6 +100,22 @@ int ib_find_cached_pkey(struct ib_device *device, u16 pkey, u16 *index); +/** + * ib_find_exact_cached_pkey - Returns the PKey table index where a specified + * PKey value occurs. Comparison uses the FULL 16 bits (incl membership bit) + * @device: The device to query. + * @port_num: The port number of the device to search for the PKey. + * @pkey: The PKey value to search for. + * @index: The index into the cached PKey table where the PKey was found. + * + * ib_find_exact_cached_pkey() searches the specified PKey table in + * the local software cache. + */ +int ib_find_exact_cached_pkey(struct ib_device *device, + u8 port_num, + u16 pkey, + u16 *index); + /** * ib_get_cached_lmc - Returns a cached lmc table entry * @device: The device to query. From 1ffeb2eb8be9936e9dc1f9af2d5f4c14d69a0d36 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:40 +0000 Subject: [PATCH 11/45] IB/mlx4: SR-IOV IB context objects and proxy/tunnel SQP support 1. Introduce the basic SR-IOV parvirtualization context objects for multiplexing and demultiplexing MADs. 2. Introduce support for the new proxy and tunnel QP types. This patch introduces the objects required by the master for managing QP paravirtualization for guests. struct mlx4_ib_sriov is created by the master only. It is a container for the following: 1. All the info required by the PPF to multiplex and de-multiplex MADs (including those from the PF). (struct mlx4_ib_demux_ctx demux) 2. All the info required to manage alias GUIDs (i.e., the GUID at index 0 that each guest perceives. In fact, this is not the GUID which is actually at index 0, but is, in fact, the GUID which is at index[] in the physical table. 3. structures which are used to manage CM paravirtualization 4. structures for managing the real special QPs when running in SR-IOV mode. The real SQPs are controlled by the PPF in this case. All SQPs created and controlled by the ib core layer are proxy SQP. struct mlx4_ib_demux_ctx contains the information per port needed to manage paravirtualization: 1. All multicast paravirt info 2. All tunnel-qp paravirt info for the port. 3. GUID-table and GUID-prefix for the port 4. work queues. struct mlx4_ib_demux_pv_ctx contains all the info for managing the paravirtualized QPs for one slave/port. struct mlx4_ib_demux_pv_qp contains the info need to run an individual QP (either tunnel qp or real SQP). Note: We made use of the 2 most significant bits in enum mlx4_ib_qp_flags (based on enum ib_qp_create_flags in ib_verbs.h). We need these bits in the low-level driver for internal purposes. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/cq.c | 31 +- drivers/infiniband/hw/mlx4/mlx4_ib.h | 128 +++++- drivers/infiniband/hw/mlx4/qp.c | 618 +++++++++++++++++++++++---- include/linux/mlx4/device.h | 1 + include/linux/mlx4/qp.h | 3 +- 5 files changed, 703 insertions(+), 78 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/cq.c b/drivers/infiniband/hw/mlx4/cq.c index 6d4ef71cbcdf..c9eb6a6815ce 100644 --- a/drivers/infiniband/hw/mlx4/cq.c +++ b/drivers/infiniband/hw/mlx4/cq.c @@ -547,6 +547,26 @@ static int mlx4_ib_ipoib_csum_ok(__be16 status, __be16 checksum) checksum == cpu_to_be16(0xffff); } +static int use_tunnel_data(struct mlx4_ib_qp *qp, struct mlx4_ib_cq *cq, struct ib_wc *wc, + unsigned tail, struct mlx4_cqe *cqe) +{ + struct mlx4_ib_proxy_sqp_hdr *hdr; + + ib_dma_sync_single_for_cpu(qp->ibqp.device, + qp->sqp_proxy_rcv[tail].map, + sizeof (struct mlx4_ib_proxy_sqp_hdr), + DMA_FROM_DEVICE); + hdr = (struct mlx4_ib_proxy_sqp_hdr *) (qp->sqp_proxy_rcv[tail].addr); + wc->pkey_index = be16_to_cpu(hdr->tun.pkey_index); + wc->slid = be16_to_cpu(hdr->tun.slid_mac_47_32); + wc->sl = (u8) (be16_to_cpu(hdr->tun.sl_vid) >> 12); + wc->src_qp = be32_to_cpu(hdr->tun.flags_src_qp) & 0xFFFFFF; + wc->wc_flags |= (hdr->tun.g_ml_path & 0x80) ? (IB_WC_GRH) : 0; + wc->dlid_path_bits = 0; + + return 0; +} + static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq, struct mlx4_ib_qp **cur_qp, struct ib_wc *wc) @@ -559,6 +579,7 @@ static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq, int is_error; u32 g_mlpath_rqpn; u16 wqe_ctr; + unsigned tail = 0; repoll: cqe = next_cqe_sw(cq); @@ -634,7 +655,8 @@ repoll: mlx4_ib_free_srq_wqe(srq, wqe_ctr); } else { wq = &(*cur_qp)->rq; - wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)]; + tail = wq->tail & (wq->wqe_cnt - 1); + wc->wr_id = wq->wrid[tail]; ++wq->tail; } @@ -717,6 +739,13 @@ repoll: break; } + if (mlx4_is_mfunc(to_mdev(cq->ibcq.device)->dev)) { + if ((*cur_qp)->mlx4_ib_qp_type & + (MLX4_IB_QPT_PROXY_SMI_OWNER | + MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) + return use_tunnel_data(*cur_qp, cq, wc, tail, cqe); + } + wc->slid = be16_to_cpu(cqe->rlid); g_mlpath_rqpn = be32_to_cpu(cqe->g_mlpath_rqpn); wc->src_qp = g_mlpath_rqpn & 0xffffff; diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index c136bb618e29..1248d576b031 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -133,8 +133,10 @@ struct mlx4_ib_wq { }; enum mlx4_ib_qp_flags { - MLX4_IB_QP_LSO = 1 << 0, - MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK = 1 << 1, + MLX4_IB_QP_LSO = IB_QP_CREATE_IPOIB_UD_LSO, + MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK, + MLX4_IB_SRIOV_TUNNEL_QP = 1 << 30, + MLX4_IB_SRIOV_SQP = 1 << 31, }; struct mlx4_ib_gid_entry { @@ -144,6 +146,68 @@ struct mlx4_ib_gid_entry { u8 port; }; +enum mlx4_ib_qp_type { + /* + * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries + * here (and in that order) since the MAD layer uses them as + * indices into a 2-entry table. + */ + MLX4_IB_QPT_SMI = IB_QPT_SMI, + MLX4_IB_QPT_GSI = IB_QPT_GSI, + + MLX4_IB_QPT_RC = IB_QPT_RC, + MLX4_IB_QPT_UC = IB_QPT_UC, + MLX4_IB_QPT_UD = IB_QPT_UD, + MLX4_IB_QPT_RAW_IPV6 = IB_QPT_RAW_IPV6, + MLX4_IB_QPT_RAW_ETHERTYPE = IB_QPT_RAW_ETHERTYPE, + MLX4_IB_QPT_RAW_PACKET = IB_QPT_RAW_PACKET, + MLX4_IB_QPT_XRC_INI = IB_QPT_XRC_INI, + MLX4_IB_QPT_XRC_TGT = IB_QPT_XRC_TGT, + + MLX4_IB_QPT_PROXY_SMI_OWNER = 1 << 16, + MLX4_IB_QPT_PROXY_SMI = 1 << 17, + MLX4_IB_QPT_PROXY_GSI = 1 << 18, + MLX4_IB_QPT_TUN_SMI_OWNER = 1 << 19, + MLX4_IB_QPT_TUN_SMI = 1 << 20, + MLX4_IB_QPT_TUN_GSI = 1 << 21, +}; + +#define MLX4_IB_QPT_ANY_SRIOV (MLX4_IB_QPT_PROXY_SMI_OWNER | \ + MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER | \ + MLX4_IB_QPT_TUN_SMI | MLX4_IB_QPT_TUN_GSI) + +struct mlx4_ib_tunnel_header { + struct mlx4_av av; + __be32 remote_qpn; + __be32 qkey; + __be16 vlan; + u8 mac[6]; + __be16 pkey_index; + u8 reserved[6]; +}; + +struct mlx4_ib_buf { + void *addr; + dma_addr_t map; +}; + +struct mlx4_rcv_tunnel_hdr { + __be32 flags_src_qp; /* flags[6:5] is defined for VLANs: + * 0x0 - no vlan was in the packet + * 0x01 - C-VLAN was in the packet */ + u8 g_ml_path; /* gid bit stands for ipv6/4 header in RoCE */ + u8 reserved; + __be16 pkey_index; + __be16 sl_vid; + __be16 slid_mac_47_32; + __be32 mac_31_0; +}; + +struct mlx4_ib_proxy_sqp_hdr { + struct ib_grh grh; + struct mlx4_rcv_tunnel_hdr tun; +} __packed; + struct mlx4_ib_qp { struct ib_qp ibqp; struct mlx4_qp mqp; @@ -159,6 +223,7 @@ struct mlx4_ib_qp { int sq_spare_wqes; struct mlx4_ib_wq sq; + enum mlx4_ib_qp_type mlx4_ib_qp_type; struct ib_umem *umem; struct mlx4_mtt mtt; int buf_size; @@ -174,6 +239,8 @@ struct mlx4_ib_qp { int mlx_type; struct list_head gid_list; struct list_head steering_rules; + struct mlx4_ib_buf *sqp_proxy_rcv; + }; struct mlx4_ib_srq { @@ -196,6 +263,55 @@ struct mlx4_ib_ah { union mlx4_ext_av av; }; +struct mlx4_ib_tun_tx_buf { + struct mlx4_ib_buf buf; + struct ib_ah *ah; +}; + +struct mlx4_ib_demux_pv_qp { + struct ib_qp *qp; + enum ib_qp_type proxy_qpt; + struct mlx4_ib_buf *ring; + struct mlx4_ib_tun_tx_buf *tx_ring; + spinlock_t tx_lock; + unsigned tx_ix_head; + unsigned tx_ix_tail; +}; + +struct mlx4_ib_demux_pv_ctx { + int port; + int slave; + int has_smi; + struct ib_device *ib_dev; + struct ib_cq *cq; + struct ib_pd *pd; + struct ib_mr *mr; + struct work_struct work; + struct workqueue_struct *wq; + struct mlx4_ib_demux_pv_qp qp[2]; +}; + +struct mlx4_ib_demux_ctx { + struct ib_device *ib_dev; + int port; + struct workqueue_struct *wq; + struct workqueue_struct *ud_wq; + spinlock_t ud_lock; + __be64 subnet_prefix; + __be64 guid_cache[128]; + struct mlx4_ib_dev *dev; + struct mlx4_ib_demux_pv_ctx **tun; +}; + +struct mlx4_ib_sriov { + struct mlx4_ib_demux_ctx demux[MLX4_MAX_PORTS]; + struct mlx4_ib_demux_pv_ctx *sqps[MLX4_MAX_PORTS]; + /* when using this spinlock you should use "irq" because + * it may be called from interrupt context.*/ + spinlock_t going_down_lock; + int is_going_down; +}; + struct mlx4_ib_iboe { spinlock_t lock; struct net_device *netdevs[MLX4_MAX_PORTS]; @@ -216,6 +332,7 @@ struct mlx4_ib_dev { struct ib_mad_agent *send_agent[MLX4_MAX_PORTS][2]; struct ib_ah *sm_ah[MLX4_MAX_PORTS]; spinlock_t sm_lock; + struct mlx4_ib_sriov sriov; struct mutex cap_mask_mutex; bool ib_active; @@ -231,6 +348,13 @@ struct ib_event_work { struct mlx4_eqe ib_eqe; }; +struct mlx4_ib_qp_tunnel_init_attr { + struct ib_qp_init_attr init_attr; + int slave; + enum ib_qp_type proxy_qp_type; + u8 port; +}; + static inline struct mlx4_ib_dev *to_mdev(struct ib_device *ibdev) { return container_of(ibdev, struct mlx4_ib_dev, ib_dev); diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index f585eddef4b7..a8622510de42 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -38,6 +38,7 @@ #include #include #include +#include #include @@ -110,16 +111,38 @@ static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp) return container_of(mqp, struct mlx4_ib_sqp, qp); } -static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) +static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) { - return qp->mqp.qpn >= dev->dev->caps.sqp_start && - qp->mqp.qpn <= dev->dev->caps.sqp_start + 3; + if (!mlx4_is_master(dev->dev)) + return 0; + + return qp->mqp.qpn >= dev->dev->caps.base_sqpn && + qp->mqp.qpn < dev->dev->caps.base_sqpn + + 8 + 16 * MLX4_MFUNC_MAX; } +static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) +{ + return ((mlx4_is_master(dev->dev) && + qp->mqp.qpn >= dev->dev->caps.base_sqpn && + qp->mqp.qpn <= dev->dev->caps.base_sqpn + 3) || + (qp->mqp.qpn >= dev->dev->caps.sqp_start && + qp->mqp.qpn <= dev->dev->caps.sqp_start + 3)); +} + +/* used for INIT/CLOSE port logic */ static int is_qp0(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) { - return qp->mqp.qpn >= dev->dev->caps.sqp_start && - qp->mqp.qpn <= dev->dev->caps.sqp_start + 1; + int qp0; + + /* qp0 is either the proxy qp0, or the real qp0 */ + qp0 = (qp->mqp.qpn >= dev->dev->caps.sqp_start && + qp->mqp.qpn <= dev->dev->caps.sqp_start + 1) || + (mlx4_is_master(dev->dev) && + qp->mqp.qpn >= dev->dev->caps.base_sqpn && + qp->mqp.qpn <= dev->dev->caps.base_sqpn + 1); + + return qp0; } static void *get_wqe(struct mlx4_ib_qp *qp, int offset) @@ -270,7 +293,7 @@ static void mlx4_ib_qp_event(struct mlx4_qp *qp, enum mlx4_event type) } } -static int send_wqe_overhead(enum ib_qp_type type, u32 flags) +static int send_wqe_overhead(enum mlx4_ib_qp_type type, u32 flags) { /* * UD WQEs must have a datagram segment. @@ -279,19 +302,29 @@ static int send_wqe_overhead(enum ib_qp_type type, u32 flags) * header and space for the ICRC). */ switch (type) { - case IB_QPT_UD: + case MLX4_IB_QPT_UD: return sizeof (struct mlx4_wqe_ctrl_seg) + sizeof (struct mlx4_wqe_datagram_seg) + ((flags & MLX4_IB_QP_LSO) ? MLX4_IB_LSO_HEADER_SPARE : 0); - case IB_QPT_UC: + case MLX4_IB_QPT_PROXY_SMI_OWNER: + case MLX4_IB_QPT_PROXY_SMI: + case MLX4_IB_QPT_PROXY_GSI: + return sizeof (struct mlx4_wqe_ctrl_seg) + + sizeof (struct mlx4_wqe_datagram_seg) + 64; + case MLX4_IB_QPT_TUN_SMI_OWNER: + case MLX4_IB_QPT_TUN_GSI: + return sizeof (struct mlx4_wqe_ctrl_seg) + + sizeof (struct mlx4_wqe_datagram_seg); + + case MLX4_IB_QPT_UC: return sizeof (struct mlx4_wqe_ctrl_seg) + sizeof (struct mlx4_wqe_raddr_seg); - case IB_QPT_RC: + case MLX4_IB_QPT_RC: return sizeof (struct mlx4_wqe_ctrl_seg) + sizeof (struct mlx4_wqe_atomic_seg) + sizeof (struct mlx4_wqe_raddr_seg); - case IB_QPT_SMI: - case IB_QPT_GSI: + case MLX4_IB_QPT_SMI: + case MLX4_IB_QPT_GSI: return sizeof (struct mlx4_wqe_ctrl_seg) + ALIGN(MLX4_IB_UD_HEADER_SIZE + DIV_ROUND_UP(MLX4_IB_UD_HEADER_SIZE, @@ -345,7 +378,7 @@ static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap, } static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap, - enum ib_qp_type type, struct mlx4_ib_qp *qp) + enum mlx4_ib_qp_type type, struct mlx4_ib_qp *qp) { int s; @@ -360,7 +393,8 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap, * For MLX transport we need 2 extra S/G entries: * one for the header and one for the checksum at the end */ - if ((type == IB_QPT_SMI || type == IB_QPT_GSI) && + if ((type == MLX4_IB_QPT_SMI || type == MLX4_IB_QPT_GSI || + type & (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) && cap->max_send_sge + 2 > dev->dev->caps.max_sq_sg) return -EINVAL; @@ -404,7 +438,9 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap, */ if (dev->dev->caps.fw_ver >= MLX4_FW_VER_WQE_CTRL_NEC && qp->sq_signal_bits && BITS_PER_LONG == 64 && - type != IB_QPT_SMI && type != IB_QPT_GSI) + type != MLX4_IB_QPT_SMI && type != MLX4_IB_QPT_GSI && + !(type & (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_PROXY_SMI | + MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER))) qp->sq.wqe_shift = ilog2(64); else qp->sq.wqe_shift = ilog2(roundup_pow_of_two(s)); @@ -476,6 +512,54 @@ static int set_user_sq_size(struct mlx4_ib_dev *dev, return 0; } +static int alloc_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp) +{ + int i; + + qp->sqp_proxy_rcv = + kmalloc(sizeof (struct mlx4_ib_buf) * qp->rq.wqe_cnt, + GFP_KERNEL); + if (!qp->sqp_proxy_rcv) + return -ENOMEM; + for (i = 0; i < qp->rq.wqe_cnt; i++) { + qp->sqp_proxy_rcv[i].addr = + kmalloc(sizeof (struct mlx4_ib_proxy_sqp_hdr), + GFP_KERNEL); + if (!qp->sqp_proxy_rcv[i].addr) + goto err; + qp->sqp_proxy_rcv[i].map = + ib_dma_map_single(dev, qp->sqp_proxy_rcv[i].addr, + sizeof (struct mlx4_ib_proxy_sqp_hdr), + DMA_FROM_DEVICE); + } + return 0; + +err: + while (i > 0) { + --i; + ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map, + sizeof (struct mlx4_ib_proxy_sqp_hdr), + DMA_FROM_DEVICE); + kfree(qp->sqp_proxy_rcv[i].addr); + } + kfree(qp->sqp_proxy_rcv); + qp->sqp_proxy_rcv = NULL; + return -ENOMEM; +} + +static void free_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp) +{ + int i; + + for (i = 0; i < qp->rq.wqe_cnt; i++) { + ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map, + sizeof (struct mlx4_ib_proxy_sqp_hdr), + DMA_FROM_DEVICE); + kfree(qp->sqp_proxy_rcv[i].addr); + } + kfree(qp->sqp_proxy_rcv); +} + static int qp_has_rq(struct ib_qp_init_attr *attr) { if (attr->qp_type == IB_QPT_XRC_INI || attr->qp_type == IB_QPT_XRC_TGT) @@ -486,10 +570,71 @@ static int qp_has_rq(struct ib_qp_init_attr *attr) static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, struct ib_qp_init_attr *init_attr, - struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp) + struct ib_udata *udata, int sqpn, struct mlx4_ib_qp **caller_qp) { int qpn; int err; + struct mlx4_ib_sqp *sqp; + struct mlx4_ib_qp *qp; + enum mlx4_ib_qp_type qp_type = (enum mlx4_ib_qp_type) init_attr->qp_type; + + /* When tunneling special qps, we use a plain UD qp */ + if (sqpn) { + if (mlx4_is_mfunc(dev->dev) && + (!mlx4_is_master(dev->dev) || + !(init_attr->create_flags & MLX4_IB_SRIOV_SQP))) { + if (init_attr->qp_type == IB_QPT_GSI) + qp_type = MLX4_IB_QPT_PROXY_GSI; + else if (mlx4_is_master(dev->dev)) + qp_type = MLX4_IB_QPT_PROXY_SMI_OWNER; + else + qp_type = MLX4_IB_QPT_PROXY_SMI; + } + qpn = sqpn; + /* add extra sg entry for tunneling */ + init_attr->cap.max_recv_sge++; + } else if (init_attr->create_flags & MLX4_IB_SRIOV_TUNNEL_QP) { + struct mlx4_ib_qp_tunnel_init_attr *tnl_init = + container_of(init_attr, + struct mlx4_ib_qp_tunnel_init_attr, init_attr); + if ((tnl_init->proxy_qp_type != IB_QPT_SMI && + tnl_init->proxy_qp_type != IB_QPT_GSI) || + !mlx4_is_master(dev->dev)) + return -EINVAL; + if (tnl_init->proxy_qp_type == IB_QPT_GSI) + qp_type = MLX4_IB_QPT_TUN_GSI; + else if (tnl_init->slave == mlx4_master_func_num(dev->dev)) + qp_type = MLX4_IB_QPT_TUN_SMI_OWNER; + else + qp_type = MLX4_IB_QPT_TUN_SMI; + qpn = dev->dev->caps.base_tunnel_sqpn + 8 * tnl_init->slave + + tnl_init->proxy_qp_type * 2 + tnl_init->port - 1; + sqpn = qpn; + } + + if (!*caller_qp) { + if (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI || + (qp_type & (MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_SMI_OWNER | + MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER))) { + sqp = kzalloc(sizeof (struct mlx4_ib_sqp), GFP_KERNEL); + if (!sqp) + return -ENOMEM; + qp = &sqp->qp; + } else { + qp = kzalloc(sizeof (struct mlx4_ib_qp), GFP_KERNEL); + if (!qp) + return -ENOMEM; + } + } else + qp = *caller_qp; + + qp->mlx4_ib_qp_type = qp_type; + + if (mlx4_is_mfunc(dev->dev) && + (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI)) { + qpn -= 8; + sqpn -= 8; + } mutex_init(&qp->mutex); spin_lock_init(&qp->sq.lock); @@ -550,7 +695,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO) qp->flags |= MLX4_IB_QP_LSO; - err = set_kernel_sq_size(dev, &init_attr->cap, init_attr->qp_type, qp); + err = set_kernel_sq_size(dev, &init_attr->cap, qp_type, qp); if (err) goto err; @@ -586,7 +731,13 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, } if (sqpn) { - qpn = sqpn; + if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER | + MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) { + if (alloc_proxy_bufs(pd->device, qp)) { + err = -ENOMEM; + goto err_wrid; + } + } } else { /* Raw packet QPNs must be aligned to 8 bits. If not, the WQE * BlueFlame setup flow wrongly causes VLAN insertion. */ @@ -595,7 +746,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, else err = mlx4_qp_reserve_range(dev->dev, 1, 1, &qpn); if (err) - goto err_wrid; + goto err_proxy; } err = mlx4_qp_alloc(dev->dev, qpn, &qp->mqp); @@ -613,13 +764,16 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, qp->doorbell_qpn = swab32(qp->mqp.qpn << 8); qp->mqp.event = mlx4_ib_qp_event; - + if (!*caller_qp) + *caller_qp = qp; return 0; err_qpn: if (!sqpn) mlx4_qp_release_range(dev->dev, qpn, 1); - +err_proxy: + if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI) + free_proxy_bufs(pd->device, qp); err_wrid: if (pd->uobject) { if (qp_has_rq(init_attr)) @@ -643,6 +797,8 @@ err_db: mlx4_db_free(dev->dev, &qp->db); err: + if (!*caller_qp) + kfree(qp); return err; } @@ -755,7 +911,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp, mlx4_qp_free(dev->dev, &qp->mqp); - if (!is_sqp(dev, qp)) + if (!is_sqp(dev, qp) && !is_tunnel_qp(dev, qp)) mlx4_qp_release_range(dev->dev, qp->mqp.qpn, 1); mlx4_mtt_cleanup(dev->dev, &qp->mtt); @@ -768,6 +924,9 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp, } else { kfree(qp->sq.wrid); kfree(qp->rq.wrid); + if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER | + MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) + free_proxy_bufs(&dev->ib_dev, qp); mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf); if (qp->rq.wqe_cnt) mlx4_db_free(dev->dev, &qp->db); @@ -780,21 +939,25 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *init_attr, struct ib_udata *udata) { - struct mlx4_ib_sqp *sqp; - struct mlx4_ib_qp *qp; + struct mlx4_ib_qp *qp = NULL; int err; u16 xrcdn = 0; /* - * We only support LSO and multicast loopback blocking, and - * only for kernel UD QPs. + * We only support LSO, vendor flag1, and multicast loopback blocking, + * and only for kernel UD QPs. */ - if (init_attr->create_flags & ~(IB_QP_CREATE_IPOIB_UD_LSO | - IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)) + if (init_attr->create_flags & ~(MLX4_IB_QP_LSO | + MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK | + MLX4_IB_SRIOV_TUNNEL_QP | MLX4_IB_SRIOV_SQP)) return ERR_PTR(-EINVAL); if (init_attr->create_flags && - (udata || init_attr->qp_type != IB_QPT_UD)) + (udata || + ((init_attr->create_flags & ~MLX4_IB_SRIOV_SQP) && + init_attr->qp_type != IB_QPT_UD) || + ((init_attr->create_flags & MLX4_IB_SRIOV_SQP) && + init_attr->qp_type > IB_QPT_GSI))) return ERR_PTR(-EINVAL); switch (init_attr->qp_type) { @@ -810,18 +973,17 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd, /* fall through */ case IB_QPT_RC: case IB_QPT_UC: - case IB_QPT_UD: case IB_QPT_RAW_PACKET: - { qp = kzalloc(sizeof *qp, GFP_KERNEL); if (!qp) return ERR_PTR(-ENOMEM); - - err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata, 0, qp); - if (err) { - kfree(qp); + /* fall through */ + case IB_QPT_UD: + { + err = create_qp_common(to_mdev(pd->device), pd, init_attr, + udata, 0, &qp); + if (err) return ERR_PTR(err); - } qp->ibqp.qp_num = qp->mqp.qpn; qp->xrcdn = xrcdn; @@ -835,21 +997,13 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd, if (udata) return ERR_PTR(-EINVAL); - sqp = kzalloc(sizeof *sqp, GFP_KERNEL); - if (!sqp) - return ERR_PTR(-ENOMEM); - - qp = &sqp->qp; - err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata, to_mdev(pd->device)->dev->caps.sqp_start + (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) + init_attr->port_num - 1, - qp); - if (err) { - kfree(sqp); + &qp); + if (err) return ERR_PTR(err); - } qp->port = init_attr->port_num; qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1; @@ -884,18 +1038,27 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp) return 0; } -static int to_mlx4_st(enum ib_qp_type type) +static int to_mlx4_st(struct mlx4_ib_dev *dev, enum mlx4_ib_qp_type type) { switch (type) { - case IB_QPT_RC: return MLX4_QP_ST_RC; - case IB_QPT_UC: return MLX4_QP_ST_UC; - case IB_QPT_UD: return MLX4_QP_ST_UD; - case IB_QPT_XRC_INI: - case IB_QPT_XRC_TGT: return MLX4_QP_ST_XRC; - case IB_QPT_SMI: - case IB_QPT_GSI: - case IB_QPT_RAW_PACKET: return MLX4_QP_ST_MLX; - default: return -1; + case MLX4_IB_QPT_RC: return MLX4_QP_ST_RC; + case MLX4_IB_QPT_UC: return MLX4_QP_ST_UC; + case MLX4_IB_QPT_UD: return MLX4_QP_ST_UD; + case MLX4_IB_QPT_XRC_INI: + case MLX4_IB_QPT_XRC_TGT: return MLX4_QP_ST_XRC; + case MLX4_IB_QPT_SMI: + case MLX4_IB_QPT_GSI: + case MLX4_IB_QPT_RAW_PACKET: return MLX4_QP_ST_MLX; + + case MLX4_IB_QPT_PROXY_SMI_OWNER: + case MLX4_IB_QPT_TUN_SMI_OWNER: return (mlx4_is_mfunc(dev->dev) ? + MLX4_QP_ST_MLX : -1); + case MLX4_IB_QPT_PROXY_SMI: + case MLX4_IB_QPT_TUN_SMI: + case MLX4_IB_QPT_PROXY_GSI: + case MLX4_IB_QPT_TUN_GSI: return (mlx4_is_mfunc(dev->dev) ? + MLX4_QP_ST_UD : -1); + default: return -1; } } @@ -1043,7 +1206,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp, return -ENOMEM; context->flags = cpu_to_be32((to_mlx4_state(new_state) << 28) | - (to_mlx4_st(ibqp->qp_type) << 16)); + (to_mlx4_st(dev, qp->mlx4_ib_qp_type) << 16)); if (!(attr_mask & IB_QP_PATH_MIG_STATE)) context->flags |= cpu_to_be32(MLX4_QP_PM_MIGRATED << 11); @@ -1121,13 +1284,16 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp, } if (attr_mask & IB_QP_PKEY_INDEX) { + if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV) + context->pri_path.disable_pkey_check = 0x40; context->pri_path.pkey_index = attr->pkey_index; optpar |= MLX4_QP_OPTPAR_PKEY_INDEX; } if (attr_mask & IB_QP_AV) { if (mlx4_set_path(dev, &attr->ah_attr, &context->pri_path, - attr_mask & IB_QP_PORT ? attr->port_num : qp->port)) + attr_mask & IB_QP_PORT ? + attr->port_num : qp->port)) goto out; optpar |= (MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH | @@ -1210,8 +1376,24 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp, if (attr_mask & IB_QP_RQ_PSN) context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn); + /* proxy and tunnel qp qkeys will be changed in modify-qp wrappers */ if (attr_mask & IB_QP_QKEY) { - context->qkey = cpu_to_be32(attr->qkey); + if (qp->mlx4_ib_qp_type & + (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) + context->qkey = cpu_to_be32(IB_QP_SET_QKEY); + else { + if (mlx4_is_mfunc(dev->dev) && + !(qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV) && + (attr->qkey & MLX4_RESERVED_QKEY_MASK) == + MLX4_RESERVED_QKEY_BASE) { + pr_err("Cannot use reserved QKEY" + " 0x%x (range 0xffff0000..0xffffffff" + " is reserved)\n", attr->qkey); + err = -EINVAL; + goto out; + } + context->qkey = cpu_to_be32(attr->qkey); + } optpar |= MLX4_QP_OPTPAR_Q_KEY; } @@ -1227,10 +1409,17 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp, ibqp->qp_type == IB_QPT_UD || ibqp->qp_type == IB_QPT_RAW_PACKET)) { context->pri_path.sched_queue = (qp->port - 1) << 6; - if (is_qp0(dev, qp)) + if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_SMI || + qp->mlx4_ib_qp_type & + (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) { context->pri_path.sched_queue |= MLX4_IB_DEFAULT_QP0_SCHED_QUEUE; - else + if (qp->mlx4_ib_qp_type != MLX4_IB_QPT_SMI) + context->pri_path.fl = 0x80; + } else { + if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV) + context->pri_path.fl = 0x80; context->pri_path.sched_queue |= MLX4_IB_DEFAULT_SCHED_QUEUE; + } } if (cur_state == IB_QPS_RTS && new_state == IB_QPS_SQD && @@ -1346,7 +1535,7 @@ int mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, } if ((attr_mask & IB_QP_PORT) && - (attr->port_num == 0 || attr->port_num > dev->dev->caps.num_ports)) { + (attr->port_num == 0 || attr->port_num > dev->num_ports)) { pr_debug("qpn 0x%x: invalid port number (%d) specified " "for transition %d to %d. qp_type %d\n", ibqp->qp_num, attr->port_num, cur_state, @@ -1400,6 +1589,115 @@ out: return err; } +static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp, + struct ib_send_wr *wr, + void *wqe, unsigned *mlx_seg_len) +{ + struct mlx4_ib_dev *mdev = to_mdev(sqp->qp.ibqp.device); + struct ib_device *ib_dev = &mdev->ib_dev; + struct mlx4_wqe_mlx_seg *mlx = wqe; + struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx; + struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah); + u16 pkey; + u32 qkey; + int send_size; + int header_size; + int spc; + int i; + + if (wr->opcode != IB_WR_SEND) + return -EINVAL; + + send_size = 0; + + for (i = 0; i < wr->num_sge; ++i) + send_size += wr->sg_list[i].length; + + /* for proxy-qp0 sends, need to add in size of tunnel header */ + /* for tunnel-qp0 sends, tunnel header is already in s/g list */ + if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER) + send_size += sizeof (struct mlx4_ib_tunnel_header); + + ib_ud_header_init(send_size, 1, 0, 0, 0, 0, &sqp->ud_header); + + if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER) { + sqp->ud_header.lrh.service_level = + be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 28; + sqp->ud_header.lrh.destination_lid = + cpu_to_be16(ah->av.ib.g_slid & 0x7f); + sqp->ud_header.lrh.source_lid = + cpu_to_be16(ah->av.ib.g_slid & 0x7f); + } + + mlx->flags &= cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE); + + /* force loopback */ + mlx->flags |= cpu_to_be32(MLX4_WQE_MLX_VL15 | 0x1 | MLX4_WQE_MLX_SLR); + mlx->rlid = sqp->ud_header.lrh.destination_lid; + + sqp->ud_header.lrh.virtual_lane = 0; + sqp->ud_header.bth.solicited_event = !!(wr->send_flags & IB_SEND_SOLICITED); + ib_get_cached_pkey(ib_dev, sqp->qp.port, 0, &pkey); + sqp->ud_header.bth.pkey = cpu_to_be16(pkey); + if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_TUN_SMI_OWNER) + sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn); + else + sqp->ud_header.bth.destination_qpn = + cpu_to_be32(mdev->dev->caps.base_tunnel_sqpn + + sqp->qp.port - 1); + + sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1)); + if (mlx4_get_parav_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey)) + return -EINVAL; + sqp->ud_header.deth.qkey = cpu_to_be32(qkey); + sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.mqp.qpn); + + sqp->ud_header.bth.opcode = IB_OPCODE_UD_SEND_ONLY; + sqp->ud_header.immediate_present = 0; + + header_size = ib_ud_header_pack(&sqp->ud_header, sqp->header_buf); + + /* + * Inline data segments may not cross a 64 byte boundary. If + * our UD header is bigger than the space available up to the + * next 64 byte boundary in the WQE, use two inline data + * segments to hold the UD header. + */ + spc = MLX4_INLINE_ALIGN - + ((unsigned long) (inl + 1) & (MLX4_INLINE_ALIGN - 1)); + if (header_size <= spc) { + inl->byte_count = cpu_to_be32(1 << 31 | header_size); + memcpy(inl + 1, sqp->header_buf, header_size); + i = 1; + } else { + inl->byte_count = cpu_to_be32(1 << 31 | spc); + memcpy(inl + 1, sqp->header_buf, spc); + + inl = (void *) (inl + 1) + spc; + memcpy(inl + 1, sqp->header_buf + spc, header_size - spc); + /* + * Need a barrier here to make sure all the data is + * visible before the byte_count field is set. + * Otherwise the HCA prefetcher could grab the 64-byte + * chunk with this inline segment and get a valid (!= + * 0xffffffff) byte count but stale data, and end up + * generating a packet with bad headers. + * + * The first inline segment's byte_count field doesn't + * need a barrier, because it comes after a + * control/MLX segment and therefore is at an offset + * of 16 mod 64. + */ + wmb(); + inl->byte_count = cpu_to_be32(1 << 31 | (header_size - spc)); + i = 2; + } + + *mlx_seg_len = + ALIGN(i * sizeof (struct mlx4_wqe_inline_seg) + header_size, 16); + return 0; +} + static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len) { @@ -1418,6 +1716,7 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr, int is_vlan = 0; int is_grh; u16 vlan; + int err = 0; send_size = 0; for (i = 0; i < wr->num_sge; ++i) @@ -1426,8 +1725,24 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr, is_eth = rdma_port_get_link_layer(sqp->qp.ibqp.device, sqp->qp.port) == IB_LINK_LAYER_ETHERNET; is_grh = mlx4_ib_ah_grh_present(ah); if (is_eth) { - ib_get_cached_gid(ib_dev, be32_to_cpu(ah->av.ib.port_pd) >> 24, - ah->av.ib.gid_index, &sgid); + if (mlx4_is_mfunc(to_mdev(ib_dev)->dev)) { + /* When multi-function is enabled, the ib_core gid + * indexes don't necessarily match the hw ones, so + * we must use our own cache */ + sgid.global.subnet_prefix = + to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1]. + subnet_prefix; + sgid.global.interface_id = + to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1]. + guid_cache[ah->av.ib.gid_index]; + } else { + err = ib_get_cached_gid(ib_dev, + be32_to_cpu(ah->av.ib.port_pd) >> 24, + ah->av.ib.gid_index, &sgid); + if (err) + return err; + } + vlan = rdma_get_vlan_id(&sgid); is_vlan = vlan < 0x1000; } @@ -1446,8 +1761,21 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr, sqp->ud_header.grh.flow_label = ah->av.ib.sl_tclass_flowlabel & cpu_to_be32(0xfffff); sqp->ud_header.grh.hop_limit = ah->av.ib.hop_limit; - ib_get_cached_gid(ib_dev, be32_to_cpu(ah->av.ib.port_pd) >> 24, - ah->av.ib.gid_index, &sqp->ud_header.grh.source_gid); + if (mlx4_is_mfunc(to_mdev(ib_dev)->dev)) { + /* When multi-function is enabled, the ib_core gid + * indexes don't necessarily match the hw ones, so + * we must use our own cache */ + sqp->ud_header.grh.source_gid.global.subnet_prefix = + to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1]. + subnet_prefix; + sqp->ud_header.grh.source_gid.global.interface_id = + to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1]. + guid_cache[ah->av.ib.gid_index]; + } else + ib_get_cached_gid(ib_dev, + be32_to_cpu(ah->av.ib.port_pd) >> 24, + ah->av.ib.gid_index, + &sqp->ud_header.grh.source_gid); memcpy(sqp->ud_header.grh.destination_gid.raw, ah->av.ib.dgid, 16); } @@ -1459,6 +1787,8 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr, (sqp->ud_header.lrh.destination_lid == IB_LID_PERMISSIVE ? MLX4_WQE_MLX_SLR : 0) | (sqp->ud_header.lrh.service_level << 8)); + if (ah->av.ib.port_pd & cpu_to_be32(0x80000000)) + mlx->flags |= cpu_to_be32(0x1); /* force loopback */ mlx->rlid = sqp->ud_header.lrh.destination_lid; } @@ -1667,6 +1997,63 @@ static void set_datagram_seg(struct mlx4_wqe_datagram_seg *dseg, memcpy(dseg->mac, to_mah(wr->wr.ud.ah)->av.eth.mac, 6); } +static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev, + struct mlx4_wqe_datagram_seg *dseg, + struct ib_send_wr *wr, enum ib_qp_type qpt) +{ + union mlx4_ext_av *av = &to_mah(wr->wr.ud.ah)->av; + struct mlx4_av sqp_av = {0}; + int port = *((u8 *) &av->ib.port_pd) & 0x3; + + /* force loopback */ + sqp_av.port_pd = av->ib.port_pd | cpu_to_be32(0x80000000); + sqp_av.g_slid = av->ib.g_slid & 0x7f; /* no GRH */ + sqp_av.sl_tclass_flowlabel = av->ib.sl_tclass_flowlabel & + cpu_to_be32(0xf0000000); + + memcpy(dseg->av, &sqp_av, sizeof (struct mlx4_av)); + dseg->dqpn = cpu_to_be32(dev->dev->caps.base_tunnel_sqpn + + qpt * 2 + port - 1); + /* use well-known qkey from the QPC */ + dseg->qkey = cpu_to_be32(0x80000000); +} + +static void build_tunnel_header(struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len) +{ + struct mlx4_wqe_inline_seg *inl = wqe; + struct mlx4_ib_tunnel_header hdr; + struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah); + int spc; + int i; + + memcpy(&hdr.av, &ah->av, sizeof hdr.av); + hdr.remote_qpn = cpu_to_be32(wr->wr.ud.remote_qpn); + hdr.pkey_index = cpu_to_be16(wr->wr.ud.pkey_index); + hdr.qkey = cpu_to_be32(wr->wr.ud.remote_qkey); + + spc = MLX4_INLINE_ALIGN - + ((unsigned long) (inl + 1) & (MLX4_INLINE_ALIGN - 1)); + if (sizeof (hdr) <= spc) { + memcpy(inl + 1, &hdr, sizeof (hdr)); + wmb(); + inl->byte_count = cpu_to_be32(1 << 31 | sizeof (hdr)); + i = 1; + } else { + memcpy(inl + 1, &hdr, spc); + wmb(); + inl->byte_count = cpu_to_be32(1 << 31 | spc); + + inl = (void *) (inl + 1) + spc; + memcpy(inl + 1, (void *) &hdr + spc, sizeof (hdr) - spc); + wmb(); + inl->byte_count = cpu_to_be32(1 << 31 | (sizeof (hdr) - spc)); + i = 2; + } + + *mlx_seg_len = + ALIGN(i * sizeof (struct mlx4_wqe_inline_seg) + sizeof (hdr), 16); +} + static void set_mlx_icrc_seg(void *dseg) { u32 *t = dseg; @@ -1748,6 +2135,13 @@ static __be32 send_ieth(struct ib_send_wr *wr) } } +static void add_zero_len_inline(void *wqe) +{ + struct mlx4_wqe_inline_seg *inl = wqe; + memset(wqe, 0, 16); + inl->byte_count = cpu_to_be32(1 << 31); +} + int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, struct ib_send_wr **bad_wr) { @@ -1806,9 +2200,9 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, wqe += sizeof *ctrl; size = sizeof *ctrl / 16; - switch (ibqp->qp_type) { - case IB_QPT_RC: - case IB_QPT_UC: + switch (qp->mlx4_ib_qp_type) { + case MLX4_IB_QPT_RC: + case MLX4_IB_QPT_UC: switch (wr->opcode) { case IB_WR_ATOMIC_CMP_AND_SWP: case IB_WR_ATOMIC_FETCH_AND_ADD: @@ -1869,7 +2263,25 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, } break; - case IB_QPT_UD: + case MLX4_IB_QPT_TUN_SMI_OWNER: + err = build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen); + if (unlikely(err)) { + *bad_wr = wr; + goto out; + } + wqe += seglen; + size += seglen / 16; + break; + case MLX4_IB_QPT_TUN_SMI: + case MLX4_IB_QPT_TUN_GSI: + /* this is a UD qp used in MAD responses to slaves. */ + set_datagram_seg(wqe, wr); + /* set the forced-loopback bit in the data seg av */ + *(__be32 *) wqe |= cpu_to_be32(0x80000000); + wqe += sizeof (struct mlx4_wqe_datagram_seg); + size += sizeof (struct mlx4_wqe_datagram_seg) / 16; + break; + case MLX4_IB_QPT_UD: set_datagram_seg(wqe, wr); wqe += sizeof (struct mlx4_wqe_datagram_seg); size += sizeof (struct mlx4_wqe_datagram_seg) / 16; @@ -1886,8 +2298,47 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, } break; - case IB_QPT_SMI: - case IB_QPT_GSI: + case MLX4_IB_QPT_PROXY_SMI_OWNER: + if (unlikely(!mlx4_is_master(to_mdev(ibqp->device)->dev))) { + err = -ENOSYS; + *bad_wr = wr; + goto out; + } + err = build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen); + if (unlikely(err)) { + *bad_wr = wr; + goto out; + } + wqe += seglen; + size += seglen / 16; + /* to start tunnel header on a cache-line boundary */ + add_zero_len_inline(wqe); + wqe += 16; + size++; + build_tunnel_header(wr, wqe, &seglen); + wqe += seglen; + size += seglen / 16; + break; + case MLX4_IB_QPT_PROXY_SMI: + /* don't allow QP0 sends on guests */ + err = -ENOSYS; + *bad_wr = wr; + goto out; + case MLX4_IB_QPT_PROXY_GSI: + /* If we are tunneling special qps, this is a UD qp. + * In this case we first add a UD segment targeting + * the tunnel qp, and then add a header with address + * information */ + set_tunnel_datagram_seg(to_mdev(ibqp->device), wqe, wr, ibqp->qp_type); + wqe += sizeof (struct mlx4_wqe_datagram_seg); + size += sizeof (struct mlx4_wqe_datagram_seg) / 16; + build_tunnel_header(wr, wqe, &seglen); + wqe += seglen; + size += seglen / 16; + break; + + case MLX4_IB_QPT_SMI: + case MLX4_IB_QPT_GSI: err = build_mlx_header(to_msqp(qp), wr, ctrl, &seglen); if (unlikely(err)) { *bad_wr = wr; @@ -1913,8 +2364,10 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, size += wr->num_sge * (sizeof (struct mlx4_wqe_data_seg) / 16); /* Add one more inline data segment for ICRC for MLX sends */ - if (unlikely(qp->ibqp.qp_type == IB_QPT_SMI || - qp->ibqp.qp_type == IB_QPT_GSI)) { + if (unlikely(qp->mlx4_ib_qp_type == MLX4_IB_QPT_SMI || + qp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI || + qp->mlx4_ib_qp_type & + (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER))) { set_mlx_icrc_seg(dseg + 1); size += sizeof (struct mlx4_wqe_data_seg) / 16; } @@ -2006,8 +2459,10 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, int err = 0; int nreq; int ind; + int max_gs; int i; + max_gs = qp->rq.max_gs; spin_lock_irqsave(&qp->rq.lock, flags); ind = qp->rq.head & (qp->rq.wqe_cnt - 1); @@ -2027,10 +2482,25 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, scat = get_recv_wqe(qp, ind); + if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER | + MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) { + ib_dma_sync_single_for_device(ibqp->device, + qp->sqp_proxy_rcv[ind].map, + sizeof (struct mlx4_ib_proxy_sqp_hdr), + DMA_FROM_DEVICE); + scat->byte_count = + cpu_to_be32(sizeof (struct mlx4_ib_proxy_sqp_hdr)); + /* use dma lkey from upper layer entry */ + scat->lkey = cpu_to_be32(wr->sg_list->lkey); + scat->addr = cpu_to_be64(qp->sqp_proxy_rcv[ind].map); + scat++; + max_gs--; + } + for (i = 0; i < wr->num_sge; ++i) __set_data_seg(scat + i, wr->sg_list + i); - if (i < qp->rq.max_gs) { + if (i < max_gs) { scat[i].byte_count = 0; scat[i].lkey = cpu_to_be32(MLX4_INVALID_LKEY); scat[i].addr = 0; diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index 6e1b0f973a03..07aa8232e631 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -389,6 +389,7 @@ struct mlx4_caps { enum mlx4_port_type possible_type[MLX4_MAX_PORTS + 1]; u32 max_counters; u8 port_ib_mtu[MLX4_MAX_PORTS + 1]; + u16 sqp_demux; }; struct mlx4_buf_list { diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h index 338388ba260a..4b4ad6ffef92 100644 --- a/include/linux/mlx4/qp.h +++ b/include/linux/mlx4/qp.h @@ -126,7 +126,8 @@ struct mlx4_rss_context { struct mlx4_qp_path { u8 fl; - u8 reserved1[2]; + u8 reserved1[1]; + u8 disable_pkey_check; u8 pkey_index; u8 counter_index; u8 grh_mylmc; From e2c76824ca16a3e8443cc7b26abcb21af7c27b10 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:41 +0000 Subject: [PATCH 12/45] mlx4_core: Add proxy and tunnel QPs to the reserved QP area In addition, pass the proxy and tunnel QP numbers to slaves so the driver can perform special QP paravirtualization. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/fw.c | 14 +++++++++ drivers/net/ethernet/mellanox/mlx4/fw.h | 3 ++ drivers/net/ethernet/mellanox/mlx4/main.c | 5 ++++ drivers/net/ethernet/mellanox/mlx4/qp.c | 29 +++++++++++++++++-- .../ethernet/mellanox/mlx4/resource_tracker.c | 3 +- include/linux/mlx4/device.h | 13 ++++++++- 6 files changed, 62 insertions(+), 5 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c index c69648487321..35be90178bef 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.c +++ b/drivers/net/ethernet/mellanox/mlx4/fw.c @@ -184,6 +184,8 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave, #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET 0x28 #define QUERY_FUNC_CAP_MAX_EQ_OFFSET 0x2c #define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET 0X30 +#define QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET 0x44 +#define QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET 0x48 #define QUERY_FUNC_CAP_FMR_FLAG 0x80 #define QUERY_FUNC_CAP_FLAG_RDMA 0x40 @@ -247,6 +249,12 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave, size = dev->caps.num_mgms + dev->caps.num_amgms; MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET); + size = dev->caps.base_tunnel_sqpn + 8 * slave; + MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET); + + size = dev->caps.sqp_start + 8 * slave; + MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET); + } else err = -EINVAL; @@ -312,6 +320,12 @@ int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap) MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET); func_cap->mcg_quota = size & 0xFFFFFF; + MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET); + func_cap->base_tunnel_qpn = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET); + func_cap->base_proxy_qpn = size & 0xFFFFFF; + for (i = 1; i <= func_cap->num_ports; ++i) { err = mlx4_cmd_box(dev, 0, mailbox->dma, i, 1, MLX4_CMD_QUERY_FUNC_CAP, diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.h b/drivers/net/ethernet/mellanox/mlx4/fw.h index 83fcbbf1b169..ced1de57c818 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.h +++ b/drivers/net/ethernet/mellanox/mlx4/fw.h @@ -134,6 +134,9 @@ struct mlx4_func_cap { int max_eq; int reserved_eq; int mcg_quota; + u32 base_qpn; + u32 base_tunnel_qpn; + u32 base_proxy_qpn; u8 physical_port[MLX4_MAX_PORTS + 1]; u8 port_flags[MLX4_MAX_PORTS + 1]; }; diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 2f816c6aed72..06ef3afbc49a 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -384,6 +384,7 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap) dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] + dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH]; + dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0; return 0; } /*The function checks if there are live vf, return the num of them*/ @@ -541,6 +542,10 @@ static int mlx4_slave_cap(struct mlx4_dev *dev) return -ENODEV; } + /* Calculate our sqp_start */ + dev->caps.sqp_start = func_cap.base_proxy_qpn; + dev->caps.base_tunnel_sqpn = func_cap.base_tunnel_qpn; + return 0; } diff --git a/drivers/net/ethernet/mellanox/mlx4/qp.c b/drivers/net/ethernet/mellanox/mlx4/qp.c index fb2b36759cbf..b8da72b29cf7 100644 --- a/drivers/net/ethernet/mellanox/mlx4/qp.c +++ b/drivers/net/ethernet/mellanox/mlx4/qp.c @@ -406,7 +406,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev) * We also reserve the MSB of the 24-bit QP number to indicate * that a QP is an XRC QP. */ - dev->caps.sqp_start = + dev->caps.base_sqpn = ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8); { @@ -437,13 +437,36 @@ int mlx4_init_qp_table(struct mlx4_dev *dev) } + /* Reserve 8 real SQPs in both native and SRIOV modes. + * In addition, in SRIOV mode, reserve 8 proxy SQPs per function + * (for all PFs and VFs), and 8 corresponding tunnel QPs. + * Each proxy SQP works opposite its own tunnel QP. + * + * The QPs are arranged as follows: + * a. 8 real SQPs + * b. All the proxy SQPs (8 per function) + * c. All the tunnel QPs (8 per function) + */ + err = mlx4_bitmap_init(&qp_table->bitmap, dev->caps.num_qps, - (1 << 23) - 1, dev->caps.sqp_start + 8, + (1 << 23) - 1, dev->caps.base_sqpn + 8 + + 16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev), reserved_from_top); + + /* In mfunc, sqp_start is the base of the proxy SQPs, since the PF also + * uses paravirtualized SQPs. + * In native mode, sqp_start is the base of the real SQPs. */ + if (mlx4_is_mfunc(dev)) { + dev->caps.sqp_start = dev->caps.base_sqpn + + 8 * (mlx4_master_func_num(dev) + 1); + dev->caps.base_tunnel_sqpn = dev->caps.sqp_start + 8 * MLX4_MFUNC_MAX; + } else + dev->caps.sqp_start = dev->caps.base_sqpn; + if (err) return err; - return mlx4_CONF_SPECIAL_QP(dev, dev->caps.sqp_start); + return mlx4_CONF_SPECIAL_QP(dev, dev->caps.base_sqpn); } void mlx4_cleanup_qp_table(struct mlx4_dev *dev) diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c index 293c9e820c49..3c57a83e6287 100644 --- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c +++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c @@ -1105,7 +1105,8 @@ static void res_end_move(struct mlx4_dev *dev, int slave, static int valid_reserved(struct mlx4_dev *dev, int slave, int qpn) { - return mlx4_is_qp_reserved(dev, qpn); + return mlx4_is_qp_reserved(dev, qpn) && + (mlx4_is_master(dev) || mlx4_is_guest_proxy(dev, slave, qpn)); } static int qp_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd, diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index 07aa8232e631..d5c82b7216de 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -693,7 +693,18 @@ static inline int mlx4_is_master(struct mlx4_dev *dev) static inline int mlx4_is_qp_reserved(struct mlx4_dev *dev, u32 qpn) { - return (qpn < dev->caps.sqp_start + 8); + return (qpn < dev->caps.base_sqpn + 8 + + 16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev)); +} + +static inline int mlx4_is_guest_proxy(struct mlx4_dev *dev, int slave, u32 qpn) +{ + int base = dev->caps.sqp_start + slave * 8; + + if (qpn >= base && qpn < base + 8) + return 1; + + return 0; } static inline int mlx4_is_mfunc(struct mlx4_dev *dev) From fc06573dfaf8a33bc0533bb70c49de13fa5232a4 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:42 +0000 Subject: [PATCH 13/45] IB/mlx4: Initialize SR-IOV IB support for slaves in master context Allocate SR-IOV paravirtualization resources and MAD demuxing contexts on the master. This has two parts. The first part is to initialize the structures to contain the contexts. This is done at master startup time in mlx4_ib_init_sriov(). The second part is to actually create the tunneling resources required on the master to support a slave. This is performed the master detects that a slave has started up (MLX4_DEV_EVENT_SLAVE_INIT event generated when a slave initializes its comm channel). For the master, there is no such startup event, so it creates its own tunneling resources when it starts up. In addition, the master also creates the real special QPs. The ib_core layer on the master causes creation of proxy special QPs, since the master is also paravirtualized at the ib_core layer. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 684 +++++++++++++++++++++++ drivers/infiniband/hw/mlx4/main.c | 80 ++- drivers/infiniband/hw/mlx4/mlx4_ib.h | 34 ++ drivers/net/ethernet/mellanox/mlx4/cmd.c | 3 + include/linux/mlx4/device.h | 3 +- include/linux/mlx4/driver.h | 2 + 6 files changed, 798 insertions(+), 8 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 9c2ae7efd00f..e98849338a94 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -44,6 +44,35 @@ enum { MLX4_IB_VENDOR_CLASS2 = 0xa }; +#define MLX4_TUN_SEND_WRID_SHIFT 34 +#define MLX4_TUN_QPN_SHIFT 32 +#define MLX4_TUN_WRID_RECV (((u64) 1) << MLX4_TUN_SEND_WRID_SHIFT) +#define MLX4_TUN_SET_WRID_QPN(a) (((u64) ((a) & 0x3)) << MLX4_TUN_QPN_SHIFT) + +#define MLX4_TUN_IS_RECV(a) (((a) >> MLX4_TUN_SEND_WRID_SHIFT) & 0x1) +#define MLX4_TUN_WRID_QPN(a) (((a) >> MLX4_TUN_QPN_SHIFT) & 0x3) + +struct mlx4_mad_rcv_buf { + struct ib_grh grh; + u8 payload[256]; +} __packed; + +struct mlx4_mad_snd_buf { + u8 payload[256]; +} __packed; + +struct mlx4_tunnel_mad { + struct ib_grh grh; + struct mlx4_ib_tunnel_header hdr; + struct ib_mad mad; +} __packed; + +struct mlx4_rcv_tunnel_mad { + struct mlx4_rcv_tunnel_hdr hdr; + struct ib_grh grh; + struct ib_mad mad; +} __packed; + int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey, int port, struct ib_wc *in_wc, struct ib_grh *in_grh, void *in_mad, void *response_mad) @@ -516,3 +545,658 @@ void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num, ib_dispatch_event(&event); } + +static void mlx4_ib_tunnel_comp_handler(struct ib_cq *cq, void *arg) +{ + unsigned long flags; + struct mlx4_ib_demux_pv_ctx *ctx = cq->cq_context; + struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev); + spin_lock_irqsave(&dev->sriov.going_down_lock, flags); + if (!dev->sriov.is_going_down && ctx->state == DEMUX_PV_STATE_ACTIVE) + queue_work(ctx->wq, &ctx->work); + spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags); +} + +static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx, + struct mlx4_ib_demux_pv_qp *tun_qp, + int index) +{ + struct ib_sge sg_list; + struct ib_recv_wr recv_wr, *bad_recv_wr; + int size; + + size = (tun_qp->qp->qp_type == IB_QPT_UD) ? + sizeof (struct mlx4_tunnel_mad) : sizeof (struct mlx4_mad_rcv_buf); + + sg_list.addr = tun_qp->ring[index].map; + sg_list.length = size; + sg_list.lkey = ctx->mr->lkey; + + recv_wr.next = NULL; + recv_wr.sg_list = &sg_list; + recv_wr.num_sge = 1; + recv_wr.wr_id = (u64) index | MLX4_TUN_WRID_RECV | + MLX4_TUN_SET_WRID_QPN(tun_qp->proxy_qpt); + ib_dma_sync_single_for_device(ctx->ib_dev, tun_qp->ring[index].map, + size, DMA_FROM_DEVICE); + return ib_post_recv(tun_qp->qp, &recv_wr, &bad_recv_wr); +} + +static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx, + enum ib_qp_type qp_type, int is_tun) +{ + int i; + struct mlx4_ib_demux_pv_qp *tun_qp; + int rx_buf_size, tx_buf_size; + + if (qp_type > IB_QPT_GSI) + return -EINVAL; + + tun_qp = &ctx->qp[qp_type]; + + tun_qp->ring = kzalloc(sizeof (struct mlx4_ib_buf) * MLX4_NUM_TUNNEL_BUFS, + GFP_KERNEL); + if (!tun_qp->ring) + return -ENOMEM; + + tun_qp->tx_ring = kcalloc(MLX4_NUM_TUNNEL_BUFS, + sizeof (struct mlx4_ib_tun_tx_buf), + GFP_KERNEL); + if (!tun_qp->tx_ring) { + kfree(tun_qp->ring); + tun_qp->ring = NULL; + return -ENOMEM; + } + + if (is_tun) { + rx_buf_size = sizeof (struct mlx4_tunnel_mad); + tx_buf_size = sizeof (struct mlx4_rcv_tunnel_mad); + } else { + rx_buf_size = sizeof (struct mlx4_mad_rcv_buf); + tx_buf_size = sizeof (struct mlx4_mad_snd_buf); + } + + for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) { + tun_qp->ring[i].addr = kmalloc(rx_buf_size, GFP_KERNEL); + if (!tun_qp->ring[i].addr) + goto err; + tun_qp->ring[i].map = ib_dma_map_single(ctx->ib_dev, + tun_qp->ring[i].addr, + rx_buf_size, + DMA_FROM_DEVICE); + } + + for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) { + tun_qp->tx_ring[i].buf.addr = + kmalloc(tx_buf_size, GFP_KERNEL); + if (!tun_qp->tx_ring[i].buf.addr) + goto tx_err; + tun_qp->tx_ring[i].buf.map = + ib_dma_map_single(ctx->ib_dev, + tun_qp->tx_ring[i].buf.addr, + tx_buf_size, + DMA_TO_DEVICE); + tun_qp->tx_ring[i].ah = NULL; + } + spin_lock_init(&tun_qp->tx_lock); + tun_qp->tx_ix_head = 0; + tun_qp->tx_ix_tail = 0; + tun_qp->proxy_qpt = qp_type; + + return 0; + +tx_err: + while (i > 0) { + --i; + ib_dma_unmap_single(ctx->ib_dev, tun_qp->tx_ring[i].buf.map, + tx_buf_size, DMA_TO_DEVICE); + kfree(tun_qp->tx_ring[i].buf.addr); + } + kfree(tun_qp->tx_ring); + tun_qp->tx_ring = NULL; + i = MLX4_NUM_TUNNEL_BUFS; +err: + while (i > 0) { + --i; + ib_dma_unmap_single(ctx->ib_dev, tun_qp->ring[i].map, + rx_buf_size, DMA_FROM_DEVICE); + kfree(tun_qp->ring[i].addr); + } + kfree(tun_qp->ring); + tun_qp->ring = NULL; + return -ENOMEM; +} + +static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx, + enum ib_qp_type qp_type, int is_tun) +{ + int i; + struct mlx4_ib_demux_pv_qp *tun_qp; + int rx_buf_size, tx_buf_size; + + if (qp_type > IB_QPT_GSI) + return; + + tun_qp = &ctx->qp[qp_type]; + if (is_tun) { + rx_buf_size = sizeof (struct mlx4_tunnel_mad); + tx_buf_size = sizeof (struct mlx4_rcv_tunnel_mad); + } else { + rx_buf_size = sizeof (struct mlx4_mad_rcv_buf); + tx_buf_size = sizeof (struct mlx4_mad_snd_buf); + } + + + for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) { + ib_dma_unmap_single(ctx->ib_dev, tun_qp->ring[i].map, + rx_buf_size, DMA_FROM_DEVICE); + kfree(tun_qp->ring[i].addr); + } + + for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) { + ib_dma_unmap_single(ctx->ib_dev, tun_qp->tx_ring[i].buf.map, + tx_buf_size, DMA_TO_DEVICE); + kfree(tun_qp->tx_ring[i].buf.addr); + if (tun_qp->tx_ring[i].ah) + ib_destroy_ah(tun_qp->tx_ring[i].ah); + } + kfree(tun_qp->tx_ring); + kfree(tun_qp->ring); +} + +static void mlx4_ib_tunnel_comp_worker(struct work_struct *work) +{ + /* dummy until next patch in series */ +} + +static void pv_qp_event_handler(struct ib_event *event, void *qp_context) +{ + struct mlx4_ib_demux_pv_ctx *sqp = qp_context; + + /* It's worse than that! He's dead, Jim! */ + pr_err("Fatal error (%d) on a MAD QP on port %d\n", + event->event, sqp->port); +} + +static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx, + enum ib_qp_type qp_type, int create_tun) +{ + int i, ret; + struct mlx4_ib_demux_pv_qp *tun_qp; + struct mlx4_ib_qp_tunnel_init_attr qp_init_attr; + struct ib_qp_attr attr; + int qp_attr_mask_INIT; + + if (qp_type > IB_QPT_GSI) + return -EINVAL; + + tun_qp = &ctx->qp[qp_type]; + + memset(&qp_init_attr, 0, sizeof qp_init_attr); + qp_init_attr.init_attr.send_cq = ctx->cq; + qp_init_attr.init_attr.recv_cq = ctx->cq; + qp_init_attr.init_attr.sq_sig_type = IB_SIGNAL_ALL_WR; + qp_init_attr.init_attr.cap.max_send_wr = MLX4_NUM_TUNNEL_BUFS; + qp_init_attr.init_attr.cap.max_recv_wr = MLX4_NUM_TUNNEL_BUFS; + qp_init_attr.init_attr.cap.max_send_sge = 1; + qp_init_attr.init_attr.cap.max_recv_sge = 1; + if (create_tun) { + qp_init_attr.init_attr.qp_type = IB_QPT_UD; + qp_init_attr.init_attr.create_flags = MLX4_IB_SRIOV_TUNNEL_QP; + qp_init_attr.port = ctx->port; + qp_init_attr.slave = ctx->slave; + qp_init_attr.proxy_qp_type = qp_type; + qp_attr_mask_INIT = IB_QP_STATE | IB_QP_PKEY_INDEX | + IB_QP_QKEY | IB_QP_PORT; + } else { + qp_init_attr.init_attr.qp_type = qp_type; + qp_init_attr.init_attr.create_flags = MLX4_IB_SRIOV_SQP; + qp_attr_mask_INIT = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_QKEY; + } + qp_init_attr.init_attr.port_num = ctx->port; + qp_init_attr.init_attr.qp_context = ctx; + qp_init_attr.init_attr.event_handler = pv_qp_event_handler; + tun_qp->qp = ib_create_qp(ctx->pd, &qp_init_attr.init_attr); + if (IS_ERR(tun_qp->qp)) { + ret = PTR_ERR(tun_qp->qp); + tun_qp->qp = NULL; + pr_err("Couldn't create %s QP (%d)\n", + create_tun ? "tunnel" : "special", ret); + return ret; + } + + memset(&attr, 0, sizeof attr); + attr.qp_state = IB_QPS_INIT; + attr.pkey_index = + to_mdev(ctx->ib_dev)->pkeys.virt2phys_pkey[ctx->slave][ctx->port - 1][0]; + attr.qkey = IB_QP1_QKEY; + attr.port_num = ctx->port; + ret = ib_modify_qp(tun_qp->qp, &attr, qp_attr_mask_INIT); + if (ret) { + pr_err("Couldn't change %s qp state to INIT (%d)\n", + create_tun ? "tunnel" : "special", ret); + goto err_qp; + } + attr.qp_state = IB_QPS_RTR; + ret = ib_modify_qp(tun_qp->qp, &attr, IB_QP_STATE); + if (ret) { + pr_err("Couldn't change %s qp state to RTR (%d)\n", + create_tun ? "tunnel" : "special", ret); + goto err_qp; + } + attr.qp_state = IB_QPS_RTS; + attr.sq_psn = 0; + ret = ib_modify_qp(tun_qp->qp, &attr, IB_QP_STATE | IB_QP_SQ_PSN); + if (ret) { + pr_err("Couldn't change %s qp state to RTS (%d)\n", + create_tun ? "tunnel" : "special", ret); + goto err_qp; + } + + for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) { + ret = mlx4_ib_post_pv_qp_buf(ctx, tun_qp, i); + if (ret) { + pr_err(" mlx4_ib_post_pv_buf error" + " (err = %d, i = %d)\n", ret, i); + goto err_qp; + } + } + return 0; + +err_qp: + ib_destroy_qp(tun_qp->qp); + tun_qp->qp = NULL; + return ret; +} + +/* + * IB MAD completion callback for real SQPs + */ +static void mlx4_ib_sqp_comp_worker(struct work_struct *work) +{ + /* dummy until next patch in series */ +} + +static int alloc_pv_object(struct mlx4_ib_dev *dev, int slave, int port, + struct mlx4_ib_demux_pv_ctx **ret_ctx) +{ + struct mlx4_ib_demux_pv_ctx *ctx; + + *ret_ctx = NULL; + ctx = kzalloc(sizeof (struct mlx4_ib_demux_pv_ctx), GFP_KERNEL); + if (!ctx) { + pr_err("failed allocating pv resource context " + "for port %d, slave %d\n", port, slave); + return -ENOMEM; + } + + ctx->ib_dev = &dev->ib_dev; + ctx->port = port; + ctx->slave = slave; + *ret_ctx = ctx; + return 0; +} + +static void free_pv_object(struct mlx4_ib_dev *dev, int slave, int port) +{ + if (dev->sriov.demux[port - 1].tun[slave]) { + kfree(dev->sriov.demux[port - 1].tun[slave]); + dev->sriov.demux[port - 1].tun[slave] = NULL; + } +} + +static int create_pv_resources(struct ib_device *ibdev, int slave, int port, + int create_tun, struct mlx4_ib_demux_pv_ctx *ctx) +{ + int ret, cq_size; + + ctx->state = DEMUX_PV_STATE_STARTING; + /* have QP0 only on port owner, and only if link layer is IB */ + if (ctx->slave == mlx4_master_func_num(to_mdev(ctx->ib_dev)->dev) && + rdma_port_get_link_layer(ibdev, ctx->port) == IB_LINK_LAYER_INFINIBAND) + ctx->has_smi = 1; + + if (ctx->has_smi) { + ret = mlx4_ib_alloc_pv_bufs(ctx, IB_QPT_SMI, create_tun); + if (ret) { + pr_err("Failed allocating qp0 tunnel bufs (%d)\n", ret); + goto err_out; + } + } + + ret = mlx4_ib_alloc_pv_bufs(ctx, IB_QPT_GSI, create_tun); + if (ret) { + pr_err("Failed allocating qp1 tunnel bufs (%d)\n", ret); + goto err_out_qp0; + } + + cq_size = 2 * MLX4_NUM_TUNNEL_BUFS; + if (ctx->has_smi) + cq_size *= 2; + + ctx->cq = ib_create_cq(ctx->ib_dev, mlx4_ib_tunnel_comp_handler, + NULL, ctx, cq_size, 0); + if (IS_ERR(ctx->cq)) { + ret = PTR_ERR(ctx->cq); + pr_err("Couldn't create tunnel CQ (%d)\n", ret); + goto err_buf; + } + + ctx->pd = ib_alloc_pd(ctx->ib_dev); + if (IS_ERR(ctx->pd)) { + ret = PTR_ERR(ctx->pd); + pr_err("Couldn't create tunnel PD (%d)\n", ret); + goto err_cq; + } + + ctx->mr = ib_get_dma_mr(ctx->pd, IB_ACCESS_LOCAL_WRITE); + if (IS_ERR(ctx->mr)) { + ret = PTR_ERR(ctx->mr); + pr_err("Couldn't get tunnel DMA MR (%d)\n", ret); + goto err_pd; + } + + if (ctx->has_smi) { + ret = create_pv_sqp(ctx, IB_QPT_SMI, create_tun); + if (ret) { + pr_err("Couldn't create %s QP0 (%d)\n", + create_tun ? "tunnel for" : "", ret); + goto err_mr; + } + } + + ret = create_pv_sqp(ctx, IB_QPT_GSI, create_tun); + if (ret) { + pr_err("Couldn't create %s QP1 (%d)\n", + create_tun ? "tunnel for" : "", ret); + goto err_qp0; + } + + if (create_tun) + INIT_WORK(&ctx->work, mlx4_ib_tunnel_comp_worker); + else + INIT_WORK(&ctx->work, mlx4_ib_sqp_comp_worker); + + ctx->wq = to_mdev(ibdev)->sriov.demux[port - 1].wq; + + ret = ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP); + if (ret) { + pr_err("Couldn't arm tunnel cq (%d)\n", ret); + goto err_wq; + } + ctx->state = DEMUX_PV_STATE_ACTIVE; + return 0; + +err_wq: + ctx->wq = NULL; + ib_destroy_qp(ctx->qp[1].qp); + ctx->qp[1].qp = NULL; + + +err_qp0: + if (ctx->has_smi) + ib_destroy_qp(ctx->qp[0].qp); + ctx->qp[0].qp = NULL; + +err_mr: + ib_dereg_mr(ctx->mr); + ctx->mr = NULL; + +err_pd: + ib_dealloc_pd(ctx->pd); + ctx->pd = NULL; + +err_cq: + ib_destroy_cq(ctx->cq); + ctx->cq = NULL; + +err_buf: + mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_GSI, create_tun); + +err_out_qp0: + if (ctx->has_smi) + mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_SMI, create_tun); +err_out: + ctx->state = DEMUX_PV_STATE_DOWN; + return ret; +} + +static void destroy_pv_resources(struct mlx4_ib_dev *dev, int slave, int port, + struct mlx4_ib_demux_pv_ctx *ctx, int flush) +{ + if (!ctx) + return; + if (ctx->state > DEMUX_PV_STATE_DOWN) { + ctx->state = DEMUX_PV_STATE_DOWNING; + if (flush) + flush_workqueue(ctx->wq); + if (ctx->has_smi) { + ib_destroy_qp(ctx->qp[0].qp); + ctx->qp[0].qp = NULL; + mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_SMI, 1); + } + ib_destroy_qp(ctx->qp[1].qp); + ctx->qp[1].qp = NULL; + mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_GSI, 1); + ib_dereg_mr(ctx->mr); + ctx->mr = NULL; + ib_dealloc_pd(ctx->pd); + ctx->pd = NULL; + ib_destroy_cq(ctx->cq); + ctx->cq = NULL; + ctx->state = DEMUX_PV_STATE_DOWN; + } +} + +static int mlx4_ib_tunnels_update(struct mlx4_ib_dev *dev, int slave, + int port, int do_init) +{ + int ret = 0; + + if (!do_init) { + /* for master, destroy real sqp resources */ + if (slave == mlx4_master_func_num(dev->dev)) + destroy_pv_resources(dev, slave, port, + dev->sriov.sqps[port - 1], 1); + /* destroy the tunnel qp resources */ + destroy_pv_resources(dev, slave, port, + dev->sriov.demux[port - 1].tun[slave], 1); + return 0; + } + + /* create the tunnel qp resources */ + ret = create_pv_resources(&dev->ib_dev, slave, port, 1, + dev->sriov.demux[port - 1].tun[slave]); + + /* for master, create the real sqp resources */ + if (!ret && slave == mlx4_master_func_num(dev->dev)) + ret = create_pv_resources(&dev->ib_dev, slave, port, 0, + dev->sriov.sqps[port - 1]); + return ret; +} + +void mlx4_ib_tunnels_update_work(struct work_struct *work) +{ + struct mlx4_ib_demux_work *dmxw; + + dmxw = container_of(work, struct mlx4_ib_demux_work, work); + mlx4_ib_tunnels_update(dmxw->dev, dmxw->slave, (int) dmxw->port, + dmxw->do_init); + kfree(dmxw); + return; +} + +static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev, + struct mlx4_ib_demux_ctx *ctx, + int port) +{ + char name[12]; + int ret = 0; + int i; + + ctx->tun = kcalloc(dev->dev->caps.sqp_demux, + sizeof (struct mlx4_ib_demux_pv_ctx *), GFP_KERNEL); + if (!ctx->tun) + return -ENOMEM; + + ctx->dev = dev; + ctx->port = port; + ctx->ib_dev = &dev->ib_dev; + + for (i = 0; i < dev->dev->caps.sqp_demux; i++) { + ret = alloc_pv_object(dev, i, port, &ctx->tun[i]); + if (ret) { + ret = -ENOMEM; + goto err_wq; + } + } + + snprintf(name, sizeof name, "mlx4_ibt%d", port); + ctx->wq = create_singlethread_workqueue(name); + if (!ctx->wq) { + pr_err("Failed to create tunnelling WQ for port %d\n", port); + ret = -ENOMEM; + goto err_wq; + } + + snprintf(name, sizeof name, "mlx4_ibud%d", port); + ctx->ud_wq = create_singlethread_workqueue(name); + if (!ctx->ud_wq) { + pr_err("Failed to create up/down WQ for port %d\n", port); + ret = -ENOMEM; + goto err_udwq; + } + + return 0; + +err_udwq: + destroy_workqueue(ctx->wq); + ctx->wq = NULL; + +err_wq: + for (i = 0; i < dev->dev->caps.sqp_demux; i++) + free_pv_object(dev, i, port); + kfree(ctx->tun); + ctx->tun = NULL; + return ret; +} + +static void mlx4_ib_free_sqp_ctx(struct mlx4_ib_demux_pv_ctx *sqp_ctx) +{ + if (sqp_ctx->state > DEMUX_PV_STATE_DOWN) { + sqp_ctx->state = DEMUX_PV_STATE_DOWNING; + flush_workqueue(sqp_ctx->wq); + if (sqp_ctx->has_smi) { + ib_destroy_qp(sqp_ctx->qp[0].qp); + sqp_ctx->qp[0].qp = NULL; + mlx4_ib_free_pv_qp_bufs(sqp_ctx, IB_QPT_SMI, 0); + } + ib_destroy_qp(sqp_ctx->qp[1].qp); + sqp_ctx->qp[1].qp = NULL; + mlx4_ib_free_pv_qp_bufs(sqp_ctx, IB_QPT_GSI, 0); + ib_dereg_mr(sqp_ctx->mr); + sqp_ctx->mr = NULL; + ib_dealloc_pd(sqp_ctx->pd); + sqp_ctx->pd = NULL; + ib_destroy_cq(sqp_ctx->cq); + sqp_ctx->cq = NULL; + sqp_ctx->state = DEMUX_PV_STATE_DOWN; + } +} + +static void mlx4_ib_free_demux_ctx(struct mlx4_ib_demux_ctx *ctx) +{ + int i; + if (ctx) { + struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev); + for (i = 0; i < dev->dev->caps.sqp_demux; i++) { + if (!ctx->tun[i]) + continue; + if (ctx->tun[i]->state > DEMUX_PV_STATE_DOWN) + ctx->tun[i]->state = DEMUX_PV_STATE_DOWNING; + } + flush_workqueue(ctx->wq); + for (i = 0; i < dev->dev->caps.sqp_demux; i++) { + destroy_pv_resources(dev, i, ctx->port, ctx->tun[i], 0); + free_pv_object(dev, i, ctx->port); + } + kfree(ctx->tun); + destroy_workqueue(ctx->ud_wq); + destroy_workqueue(ctx->wq); + } +} + +static void mlx4_ib_master_tunnels(struct mlx4_ib_dev *dev, int do_init) +{ + int i; + + if (!mlx4_is_master(dev->dev)) + return; + /* initialize or tear down tunnel QPs for the master */ + for (i = 0; i < dev->dev->caps.num_ports; i++) + mlx4_ib_tunnels_update(dev, mlx4_master_func_num(dev->dev), i + 1, do_init); + return; +} + +int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev) +{ + int i = 0; + int err; + + if (!mlx4_is_mfunc(dev->dev)) + return 0; + + dev->sriov.is_going_down = 0; + spin_lock_init(&dev->sriov.going_down_lock); + + mlx4_ib_warn(&dev->ib_dev, "multi-function enabled\n"); + + if (mlx4_is_slave(dev->dev)) { + mlx4_ib_warn(&dev->ib_dev, "operating in qp1 tunnel mode\n"); + return 0; + } + + mlx4_ib_warn(&dev->ib_dev, "initializing demux service for %d qp1 clients\n", + dev->dev->caps.sqp_demux); + for (i = 0; i < dev->num_ports; i++) { + err = alloc_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1, + &dev->sriov.sqps[i]); + if (err) + goto demux_err; + err = mlx4_ib_alloc_demux_ctx(dev, &dev->sriov.demux[i], i + 1); + if (err) + goto demux_err; + } + mlx4_ib_master_tunnels(dev, 1); + return 0; + +demux_err: + while (i > 0) { + free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1); + mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]); + --i; + } + + return err; +} + +void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev) +{ + int i; + unsigned long flags; + + if (!mlx4_is_mfunc(dev->dev)) + return; + + spin_lock_irqsave(&dev->sriov.going_down_lock, flags); + dev->sriov.is_going_down = 1; + spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags); + if (mlx4_is_master(dev->dev)) + for (i = 0; i < dev->num_ports; i++) { + flush_workqueue(dev->sriov.demux[i].ud_wq); + mlx4_ib_free_sqp_ctx(dev->sriov.sqps[i]); + kfree(dev->sriov.sqps[i]); + dev->sriov.sqps[i] = NULL; + mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]); + } +} diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index cc05579ebce7..3f7f77f93a1c 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -1357,11 +1357,14 @@ static void *mlx4_ib_add(struct mlx4_dev *dev) if (mlx4_ib_mad_init(ibdev)) goto err_reg; + if (mlx4_ib_init_sriov(ibdev)) + goto err_mad; + if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) { iboe->nb.notifier_call = mlx4_ib_netdev_event; err = register_netdevice_notifier(&iboe->nb); if (err) - goto err_reg; + goto err_sriov; } for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) { @@ -1379,6 +1382,12 @@ err_notif: pr_warn("failure unregistering notifier\n"); flush_workqueue(wq); +err_sriov: + mlx4_ib_close_sriov(ibdev); + +err_mad: + mlx4_ib_mad_cleanup(ibdev); + err_reg: ib_unregister_device(&ibdev->ib_dev); @@ -1407,6 +1416,7 @@ static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr) struct mlx4_ib_dev *ibdev = ibdev_ptr; int p; + mlx4_ib_close_sriov(ibdev); mlx4_ib_mad_cleanup(ibdev); ib_unregister_device(&ibdev->ib_dev); if (ibdev->iboe.nb.notifier_call) { @@ -1428,6 +1438,51 @@ static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr) ib_dealloc_device(&ibdev->ib_dev); } +static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init) +{ + struct mlx4_ib_demux_work **dm = NULL; + struct mlx4_dev *dev = ibdev->dev; + int i; + unsigned long flags; + + if (!mlx4_is_master(dev)) + return; + + dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC); + if (!dm) { + pr_err("failed to allocate memory for tunneling qp update\n"); + goto out; + } + + for (i = 0; i < dev->caps.num_ports; i++) { + dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC); + if (!dm[i]) { + pr_err("failed to allocate memory for tunneling qp update work struct\n"); + for (i = 0; i < dev->caps.num_ports; i++) { + if (dm[i]) + kfree(dm[i]); + } + goto out; + } + } + /* initialize or tear down tunnel QPs for the slave */ + for (i = 0; i < dev->caps.num_ports; i++) { + INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work); + dm[i]->port = i + 1; + dm[i]->slave = slave; + dm[i]->do_init = do_init; + dm[i]->dev = ibdev; + spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags); + if (!ibdev->sriov.is_going_down) + queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work); + spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags); + } +out: + if (dm) + kfree(dm); + return; +} + static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, enum mlx4_dev_event event, unsigned long param) { @@ -1435,22 +1490,23 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr); struct mlx4_eqe *eqe = NULL; struct ib_event_work *ew; - int port = 0; + int p = 0; if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE) eqe = (struct mlx4_eqe *)param; else - port = (u8)param; - - if (port > ibdev->num_ports) - return; + p = (int) param; switch (event) { case MLX4_DEV_EVENT_PORT_UP: + if (p > ibdev->num_ports) + return; ibev.event = IB_EVENT_PORT_ACTIVE; break; case MLX4_DEV_EVENT_PORT_DOWN: + if (p > ibdev->num_ports) + return; ibev.event = IB_EVENT_PORT_ERR; break; @@ -1472,12 +1528,22 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, handle_port_mgmt_change_event(&ew->work); return; + case MLX4_DEV_EVENT_SLAVE_INIT: + /* here, p is the slave id */ + do_slave_init(ibdev, p, 1); + return; + + case MLX4_DEV_EVENT_SLAVE_SHUTDOWN: + /* here, p is the slave id */ + do_slave_init(ibdev, p, 0); + return; + default: return; } ibev.device = ibdev_ptr; - ibev.element.port_num = port; + ibev.element.port_num = (u8) p; ib_dispatch_event(&ibev); } diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index 1248d576b031..137941d79870 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -176,6 +176,10 @@ enum mlx4_ib_qp_type { MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER | \ MLX4_IB_QPT_TUN_SMI | MLX4_IB_QPT_TUN_GSI) +enum { + MLX4_NUM_TUNNEL_BUFS = 256, +}; + struct mlx4_ib_tunnel_header { struct mlx4_av av; __be32 remote_qpn; @@ -263,6 +267,15 @@ struct mlx4_ib_ah { union mlx4_ext_av av; }; +struct mlx4_ib_demux_work { + struct work_struct work; + struct mlx4_ib_dev *dev; + int slave; + int do_init; + u8 port; + +}; + struct mlx4_ib_tun_tx_buf { struct mlx4_ib_buf buf; struct ib_ah *ah; @@ -278,9 +291,17 @@ struct mlx4_ib_demux_pv_qp { unsigned tx_ix_tail; }; +enum mlx4_ib_demux_pv_state { + DEMUX_PV_STATE_DOWN, + DEMUX_PV_STATE_STARTING, + DEMUX_PV_STATE_ACTIVE, + DEMUX_PV_STATE_DOWNING, +}; + struct mlx4_ib_demux_pv_ctx { int port; int slave; + enum mlx4_ib_demux_pv_state state; int has_smi; struct ib_device *ib_dev; struct ib_cq *cq; @@ -319,6 +340,13 @@ struct mlx4_ib_iboe { union ib_gid gid_table[MLX4_MAX_PORTS][128]; }; +struct pkey_mgt { + u8 virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS]; + u16 phys_pkey_cache[MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS]; + struct list_head pkey_port_list[MLX4_MFUNC_MAX]; + struct kobject *device_parent[MLX4_MFUNC_MAX]; +}; + struct mlx4_ib_dev { struct ib_device ib_dev; struct mlx4_dev *dev; @@ -340,6 +368,7 @@ struct mlx4_ib_dev { int counters[MLX4_MAX_PORTS]; int *eq_table; int eq_added; + struct pkey_mgt pkeys; }; struct ib_event_work { @@ -424,6 +453,9 @@ static inline struct mlx4_ib_ah *to_mah(struct ib_ah *ibah) return container_of(ibah, struct mlx4_ib_ah, ibah); } +int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev); +void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev); + int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt, struct mlx4_db *db); void mlx4_ib_db_unmap_user(struct mlx4_ib_ucontext *context, struct mlx4_db *db); @@ -515,4 +547,6 @@ int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp, void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num, enum ib_event_type type); +void mlx4_ib_tunnels_update_work(struct work_struct *work); + #endif /* MLX4_IB_H */ diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index c8fef4353021..cb9bebe28276 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -1340,6 +1340,8 @@ static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd, if (MLX4_COMM_CMD_FLR == slave_state[slave].last_cmd) goto inform_slave_state; + mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_SHUTDOWN, slave); + /* write the version in the event field */ reply |= mlx4_comm_get_version(); @@ -1376,6 +1378,7 @@ static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd, goto reset_slave; slave_state[slave].vhcr_dma |= param; slave_state[slave].active = true; + mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_INIT, slave); break; case MLX4_COMM_CMD_VHCR_POST: if ((slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_EN) && diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index d5c82b7216de..b6b8d341b6c8 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -54,7 +54,8 @@ enum { }; enum { - MLX4_MAX_PORTS = 2 + MLX4_MAX_PORTS = 2, + MLX4_MAX_PORT_PKEYS = 128 }; /* base qkey for use in sriov tunnel-qp/proxy-qp communication. diff --git a/include/linux/mlx4/driver.h b/include/linux/mlx4/driver.h index d813704b963b..c257e1b211be 100644 --- a/include/linux/mlx4/driver.h +++ b/include/linux/mlx4/driver.h @@ -45,6 +45,8 @@ enum mlx4_dev_event { MLX4_DEV_EVENT_PORT_DOWN, MLX4_DEV_EVENT_PORT_REINIT, MLX4_DEV_EVENT_PORT_MGMT_CHANGE, + MLX4_DEV_EVENT_SLAVE_INIT, + MLX4_DEV_EVENT_SLAVE_SHUTDOWN, }; struct mlx4_interface { From 54679e148287f0ca1bdd09264c908bacb9f19b3f Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:43 +0000 Subject: [PATCH 14/45] mlx4: Implement QP paravirtualization and maintain phys_pkey_cache for smp_snoop This requires: 1. Replacing the paravirtualized P_Key index (inserted by the guest) with the real P_Key index. 2. For UD QPs, placing the guest's true source GID index in the address path structure mgid field, and setting the ud_force_mgid bit so that the mgid is taken from the QP context and not from the WQE when posting sends. 3. For UC and RC QPs, placing the guest's true source GID index in the address path structure mgid field. 4. For tunnel and proxy QPs, setting the Q_Key value reserved for that proxy/tunnel pair. Since not all the above adjustments occur in all the QP transitions, the QP transitions require separate wrapper functions. Secondly, initialize the P_Key virtualization table to its default values: Master virtualized table is 1-1 with the real P_Key table, guest virtualized table has P_Key index 0 mapped to the real P_Key index 0, and all the other P_Key indices mapped to the reserved (invalid) P_Key at index 127. Finally, add logic in smp_snoop for maintaining the phys_P_Key_cache. and generating events on the master only if a P_Key actually changed. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 33 ++- drivers/infiniband/hw/mlx4/main.c | 35 +++ drivers/net/ethernet/mellanox/mlx4/cmd.c | 12 +- drivers/net/ethernet/mellanox/mlx4/main.c | 11 + drivers/net/ethernet/mellanox/mlx4/mlx4.h | 47 ++++ .../ethernet/mellanox/mlx4/resource_tracker.c | 217 +++++++++++++++++- include/linux/mlx4/device.h | 3 + 7 files changed, 344 insertions(+), 14 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index e98849338a94..318d5bcf821b 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -185,6 +185,10 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad, { struct ib_port_info *pinfo; u16 lid; + __be16 *base; + u32 bn, pkey_change_bitmap; + int i; + struct mlx4_ib_dev *dev = to_mdev(ibdev); if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED || @@ -209,8 +213,33 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad, break; case IB_SMP_ATTR_PKEY_TABLE: - mlx4_ib_dispatch_event(dev, port_num, - IB_EVENT_PKEY_CHANGE); + if (!mlx4_is_mfunc(dev->dev)) { + mlx4_ib_dispatch_event(dev, port_num, + IB_EVENT_PKEY_CHANGE); + break; + } + + bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod) & 0xFFFF; + base = (__be16 *) &(((struct ib_smp *)mad)->data[0]); + pkey_change_bitmap = 0; + for (i = 0; i < 32; i++) { + pr_debug("PKEY[%d] = x%x\n", + i + bn*32, be16_to_cpu(base[i])); + if (be16_to_cpu(base[i]) != + dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32]) { + pkey_change_bitmap |= (1 << i); + dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32] = + be16_to_cpu(base[i]); + } + } + pr_debug("PKEY Change event: port=%d, " + "block=0x%x, change_bitmap=0x%x\n", + port_num, bn, pkey_change_bitmap); + + if (pkey_change_bitmap) + mlx4_ib_dispatch_event(dev, port_num, + IB_EVENT_PKEY_CHANGE); + break; case IB_SMP_ATTR_GUID_INFO: diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index 3f7f77f93a1c..8e10ec2af7b6 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -1121,6 +1121,38 @@ static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event return NOTIFY_DONE; } +static void init_pkeys(struct mlx4_ib_dev *ibdev) +{ + int port; + int slave; + int i; + + if (mlx4_is_master(ibdev->dev)) { + for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) { + for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) { + for (i = 0; + i < ibdev->dev->phys_caps.pkey_phys_table_len[port]; + ++i) { + ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] = + /* master has the identity virt2phys pkey mapping */ + (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i : + ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1; + mlx4_sync_pkey_table(ibdev->dev, slave, port, i, + ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]); + } + } + } + /* initialize pkey cache */ + for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) { + for (i = 0; + i < ibdev->dev->phys_caps.pkey_phys_table_len[port]; + ++i) + ibdev->pkeys.phys_pkey_cache[port-1][i] = + (i) ? 0 : 0xFFFF; + } + } +} + static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev) { char name[32]; @@ -1375,6 +1407,9 @@ static void *mlx4_ib_add(struct mlx4_dev *dev) ibdev->ib_active = true; + if (mlx4_is_mfunc(ibdev->dev)) + init_pkeys(ibdev); + return ibdev; err_notif: diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index cb9bebe28276..662a3c5da739 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -950,7 +950,7 @@ static struct mlx4_cmd_info cmd_info[] = { .out_is_imm = false, .encode_slave_id = false, .verify = NULL, - .wrapper = mlx4_GEN_QP_wrapper + .wrapper = mlx4_INIT2INIT_QP_wrapper }, { .opcode = MLX4_CMD_INIT2RTR_QP, @@ -968,7 +968,7 @@ static struct mlx4_cmd_info cmd_info[] = { .out_is_imm = false, .encode_slave_id = false, .verify = NULL, - .wrapper = mlx4_GEN_QP_wrapper + .wrapper = mlx4_RTR2RTS_QP_wrapper }, { .opcode = MLX4_CMD_RTS2RTS_QP, @@ -977,7 +977,7 @@ static struct mlx4_cmd_info cmd_info[] = { .out_is_imm = false, .encode_slave_id = false, .verify = NULL, - .wrapper = mlx4_GEN_QP_wrapper + .wrapper = mlx4_RTS2RTS_QP_wrapper }, { .opcode = MLX4_CMD_SQERR2RTS_QP, @@ -986,7 +986,7 @@ static struct mlx4_cmd_info cmd_info[] = { .out_is_imm = false, .encode_slave_id = false, .verify = NULL, - .wrapper = mlx4_GEN_QP_wrapper + .wrapper = mlx4_SQERR2RTS_QP_wrapper }, { .opcode = MLX4_CMD_2ERR_QP, @@ -1013,7 +1013,7 @@ static struct mlx4_cmd_info cmd_info[] = { .out_is_imm = false, .encode_slave_id = false, .verify = NULL, - .wrapper = mlx4_GEN_QP_wrapper + .wrapper = mlx4_SQD2SQD_QP_wrapper }, { .opcode = MLX4_CMD_SQD2RTS_QP, @@ -1022,7 +1022,7 @@ static struct mlx4_cmd_info cmd_info[] = { .out_is_imm = false, .encode_slave_id = false, .verify = NULL, - .wrapper = mlx4_GEN_QP_wrapper + .wrapper = mlx4_SQD2RTS_QP_wrapper }, { .opcode = MLX4_CMD_2RST_QP, diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 06ef3afbc49a..2294b7173180 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -424,6 +424,17 @@ int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey) } EXPORT_SYMBOL(mlx4_get_parav_qkey); +void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val) +{ + struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev); + + if (!mlx4_is_master(dev)) + return; + + priv->virt2phys_pkey[slave][port - 1][i] = val; +} +EXPORT_SYMBOL(mlx4_sync_pkey_table); + int mlx4_is_slave_active(struct mlx4_dev *dev, int slave) { struct mlx4_priv *priv = mlx4_priv(dev); diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h index dba69d98734a..7d27c3158d0c 100644 --- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h +++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h @@ -807,6 +807,8 @@ struct mlx4_priv { struct io_mapping *bf_mapping; int reserved_mtts; int fs_hash_mode; + u8 virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS]; + }; static inline struct mlx4_priv *mlx4_priv(struct mlx4_dev *dev) @@ -1011,16 +1013,61 @@ int mlx4_RST2INIT_QP_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_cmd_mailbox *inbox, struct mlx4_cmd_mailbox *outbox, struct mlx4_cmd_info *cmd); +int mlx4_INIT2INIT_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); int mlx4_INIT2RTR_QP_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, struct mlx4_cmd_mailbox *outbox, struct mlx4_cmd_info *cmd); +int mlx4_RTR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); +int mlx4_RTS2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); +int mlx4_SQERR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); +int mlx4_2ERR_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); +int mlx4_RTS2SQD_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); +int mlx4_SQD2SQD_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); +int mlx4_SQD2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); int mlx4_2RST_QP_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, struct mlx4_cmd_mailbox *outbox, struct mlx4_cmd_info *cmd); +int mlx4_QUERY_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd); int mlx4_GEN_EQE(struct mlx4_dev *dev, int slave, struct mlx4_eqe *eqe); diff --git a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c index 3c57a83e6287..49e9de725d0a 100644 --- a/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c +++ b/drivers/net/ethernet/mellanox/mlx4/resource_tracker.c @@ -242,6 +242,15 @@ static int res_tracker_insert(struct rb_root *root, struct res_common *res) return 0; } +enum qp_transition { + QP_TRANS_INIT2RTR, + QP_TRANS_RTR2RTS, + QP_TRANS_RTS2RTS, + QP_TRANS_SQERR2RTS, + QP_TRANS_SQD2SQD, + QP_TRANS_SQD2RTS +}; + /* For Debug uses */ static const char *ResourceType(enum mlx4_resource rt) { @@ -308,14 +317,41 @@ void mlx4_free_resource_tracker(struct mlx4_dev *dev, } } -static void update_ud_gid(struct mlx4_dev *dev, - struct mlx4_qp_context *qp_ctx, u8 slave) +static void update_pkey_index(struct mlx4_dev *dev, int slave, + struct mlx4_cmd_mailbox *inbox) { - u32 ts = (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff; + u8 sched = *(u8 *)(inbox->buf + 64); + u8 orig_index = *(u8 *)(inbox->buf + 35); + u8 new_index; + struct mlx4_priv *priv = mlx4_priv(dev); + int port; + + port = (sched >> 6 & 1) + 1; + + new_index = priv->virt2phys_pkey[slave][port - 1][orig_index]; + *(u8 *)(inbox->buf + 35) = new_index; + + mlx4_dbg(dev, "port = %d, orig pkey index = %d, " + "new pkey index = %d\n", port, orig_index, new_index); +} + +static void update_gid(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *inbox, + u8 slave) +{ + struct mlx4_qp_context *qp_ctx = inbox->buf + 8; + enum mlx4_qp_optpar optpar = be32_to_cpu(*(__be32 *) inbox->buf); + u32 ts = (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff; if (MLX4_QP_ST_UD == ts) qp_ctx->pri_path.mgid_index = 0x80 | slave; + if (MLX4_QP_ST_RC == ts || MLX4_QP_ST_UC == ts) { + if (optpar & MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH) + qp_ctx->pri_path.mgid_index = slave & 0x7F; + if (optpar & MLX4_QP_OPTPAR_ALT_ADDR_PATH) + qp_ctx->alt_path.mgid_index = slave & 0x7F; + } + mlx4_dbg(dev, "slave %d, new gid index: 0x%x ", slave, qp_ctx->pri_path.mgid_index); } @@ -1109,6 +1145,11 @@ static int valid_reserved(struct mlx4_dev *dev, int slave, int qpn) (mlx4_is_master(dev) || mlx4_is_guest_proxy(dev, slave, qpn)); } +static int fw_reserved(struct mlx4_dev *dev, int qpn) +{ + return qpn < dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW]; +} + static int qp_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd, u64 in_param, u64 *out_param) { @@ -1146,7 +1187,7 @@ static int qp_alloc_res(struct mlx4_dev *dev, int slave, int op, int cmd, if (err) return err; - if (!valid_reserved(dev, slave, qpn)) { + if (!fw_reserved(dev, qpn)) { err = __mlx4_qp_alloc_icm(dev, qpn); if (err) { res_abort_move(dev, slave, RES_QP, qpn); @@ -1499,7 +1540,7 @@ static int qp_free_res(struct mlx4_dev *dev, int slave, int op, int cmd, if (err) return err; - if (!valid_reserved(dev, slave, qpn)) + if (!fw_reserved(dev, qpn)) __mlx4_qp_free_icm(dev, qpn); res_end_move(dev, slave, RES_QP, qpn); @@ -1939,6 +1980,19 @@ static u32 qp_get_srqn(struct mlx4_qp_context *qpc) return be32_to_cpu(qpc->srqn) & 0x1ffffff; } +static void adjust_proxy_tun_qkey(struct mlx4_dev *dev, struct mlx4_vhcr *vhcr, + struct mlx4_qp_context *context) +{ + u32 qpn = vhcr->in_modifier & 0xffffff; + u32 qkey = 0; + + if (mlx4_get_parav_qkey(dev, qpn, &qkey)) + return; + + /* adjust qkey in qp context */ + context->qkey = cpu_to_be32(qkey); +} + int mlx4_RST2INIT_QP_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, @@ -1991,6 +2045,8 @@ int mlx4_RST2INIT_QP_wrapper(struct mlx4_dev *dev, int slave, goto ex_put_scq; } + adjust_proxy_tun_qkey(dev, vhcr, qpc); + update_pkey_index(dev, slave, inbox); err = mlx4_DMA_wrapper(dev, slave, vhcr, inbox, outbox, cmd); if (err) goto ex_put_srq; @@ -2136,6 +2192,48 @@ static int get_containing_mtt(struct mlx4_dev *dev, int slave, int start, return err; } +static int verify_qp_parameters(struct mlx4_dev *dev, + struct mlx4_cmd_mailbox *inbox, + enum qp_transition transition, u8 slave) +{ + u32 qp_type; + struct mlx4_qp_context *qp_ctx; + enum mlx4_qp_optpar optpar; + + qp_ctx = inbox->buf + 8; + qp_type = (be32_to_cpu(qp_ctx->flags) >> 16) & 0xff; + optpar = be32_to_cpu(*(__be32 *) inbox->buf); + + switch (qp_type) { + case MLX4_QP_ST_RC: + case MLX4_QP_ST_UC: + switch (transition) { + case QP_TRANS_INIT2RTR: + case QP_TRANS_RTR2RTS: + case QP_TRANS_RTS2RTS: + case QP_TRANS_SQD2SQD: + case QP_TRANS_SQD2RTS: + if (slave != mlx4_master_func_num(dev)) + /* slaves have only gid index 0 */ + if (optpar & MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH) + if (qp_ctx->pri_path.mgid_index) + return -EINVAL; + if (optpar & MLX4_QP_OPTPAR_ALT_ADDR_PATH) + if (qp_ctx->alt_path.mgid_index) + return -EINVAL; + break; + default: + break; + } + + break; + default: + break; + } + + return 0; +} + int mlx4_WRITE_MTT_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, @@ -2623,16 +2721,123 @@ out: return err; } +int mlx4_INIT2INIT_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd) +{ + struct mlx4_qp_context *context = inbox->buf + 8; + adjust_proxy_tun_qkey(dev, vhcr, context); + update_pkey_index(dev, slave, inbox); + return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd); +} + int mlx4_INIT2RTR_QP_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, struct mlx4_cmd_mailbox *outbox, struct mlx4_cmd_info *cmd) { + int err; struct mlx4_qp_context *qpc = inbox->buf + 8; - update_ud_gid(dev, qpc, (u8)slave); + err = verify_qp_parameters(dev, inbox, QP_TRANS_INIT2RTR, slave); + if (err) + return err; + update_pkey_index(dev, slave, inbox); + update_gid(dev, inbox, (u8)slave); + adjust_proxy_tun_qkey(dev, vhcr, qpc); + + return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd); +} + +int mlx4_RTR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd) +{ + int err; + struct mlx4_qp_context *context = inbox->buf + 8; + + err = verify_qp_parameters(dev, inbox, QP_TRANS_RTR2RTS, slave); + if (err) + return err; + + update_pkey_index(dev, slave, inbox); + update_gid(dev, inbox, (u8)slave); + adjust_proxy_tun_qkey(dev, vhcr, context); + return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd); +} + +int mlx4_RTS2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd) +{ + int err; + struct mlx4_qp_context *context = inbox->buf + 8; + + err = verify_qp_parameters(dev, inbox, QP_TRANS_RTS2RTS, slave); + if (err) + return err; + + update_pkey_index(dev, slave, inbox); + update_gid(dev, inbox, (u8)slave); + adjust_proxy_tun_qkey(dev, vhcr, context); + return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd); +} + + +int mlx4_SQERR2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd) +{ + struct mlx4_qp_context *context = inbox->buf + 8; + adjust_proxy_tun_qkey(dev, vhcr, context); + return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd); +} + +int mlx4_SQD2SQD_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd) +{ + int err; + struct mlx4_qp_context *context = inbox->buf + 8; + + err = verify_qp_parameters(dev, inbox, QP_TRANS_SQD2SQD, slave); + if (err) + return err; + + adjust_proxy_tun_qkey(dev, vhcr, context); + update_gid(dev, inbox, (u8)slave); + update_pkey_index(dev, slave, inbox); + return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd); +} + +int mlx4_SQD2RTS_QP_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd) +{ + int err; + struct mlx4_qp_context *context = inbox->buf + 8; + + err = verify_qp_parameters(dev, inbox, QP_TRANS_SQD2RTS, slave); + if (err) + return err; + + adjust_proxy_tun_qkey(dev, vhcr, context); + update_gid(dev, inbox, (u8)slave); + update_pkey_index(dev, slave, inbox); return mlx4_GEN_QP_wrapper(dev, slave, vhcr, inbox, outbox, cmd); } diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index b6b8d341b6c8..9803fd5d3dba 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -940,6 +940,9 @@ int mlx4_flow_attach(struct mlx4_dev *dev, struct mlx4_net_trans_rule *rule, u64 *reg_id); int mlx4_flow_detach(struct mlx4_dev *dev, u64 reg_id); +void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, + int i, int val); + int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey); #endif /* MLX4_DEVICE_H */ From 37bfc7c1e83f1589bcdc5918c7216422662644ee Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:44 +0000 Subject: [PATCH 15/45] IB/mlx4: SR-IOV multiplex and demultiplex MADs Special QPs are paravirtualized. vHCAs are not given direct access to QP0/1. Rather, these QPs are operated by a special context hosted by the PF, which mediates access to/from vHCAs. This is done by opening a "tunnel" per vHCA port per QP0/1. A tunnel comprises a pair of UD QPs: a "Tunnel QP" in the PF-context and a "Proxy QP" in the vHCA. All vHCA MAD traffic must pass through the corresponding tunnel. vHCA QPs cannot be assigned to VL15 and are denied of the well-known QKey. Outgoing messages are "de-multiplexed" (i.e., directed to the wire via the real special QP). Incoming messages are "multiplexed" (i.e. steered by the PPF to the correct VF or to the PF) QP0 access is restricted to the PF vHCA. VF vHCAs also have (virtual) QP0s, but they never receive any SMPs and all SMPs sent are discarded. QP1 traffic is allowed for all vHCAs, but special care is required to bridge the gap between the host and network views. Specifically: - Transaction IDs are mapped to guarantee uniqueness among vHCAs - CM para-virtualization o Incoming requests are steered to the correct vHCA according to the embedded GID o Local communication IDs are mapped to ensure uniqueness among vHCAs (see the patch that adds CM paravirtualization.) - Multicast para-virtualization o The PF context aggregates membership state from all vHCAs o The SA is contacted only when the aggregate membership changes o If the aggregate does not change, the PF context will provide the requesting vHCA with the proper response. (see the patch that adds multicast group paravirtualization) Incoming MADs are steered according to: - the DGID If a GRH is present - the mapped transaction ID for response MADs - the embedded GID in CM requests - the remote communication ID in other CM messages Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 567 ++++++++++++++++++++++++++++++- 1 file changed, 565 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 318d5bcf821b..8dfbf69f8370 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -32,6 +32,8 @@ #include #include +#include +#include #include #include @@ -300,6 +302,254 @@ static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *ma } } +static int mlx4_ib_demux_sa_handler(struct ib_device *ibdev, int port, int slave, + struct ib_sa_mad *sa_mad) +{ + return 0; +} + +int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid) +{ + struct mlx4_ib_dev *dev = to_mdev(ibdev); + int i; + + for (i = 0; i < dev->dev->caps.sqp_demux; i++) { + if (dev->sriov.demux[port - 1].guid_cache[i] == guid) + return i; + } + return -1; +} + + +static int get_pkey_phys_indices(struct mlx4_ib_dev *ibdev, u8 port, u8 ph_pkey_ix, + u8 *full_pk_ix, u8 *partial_pk_ix, + int *is_full_member) +{ + u16 search_pkey; + int fm; + int err = 0; + u16 pk; + + err = ib_get_cached_pkey(&ibdev->ib_dev, port, ph_pkey_ix, &search_pkey); + if (err) + return err; + + fm = (search_pkey & 0x8000) ? 1 : 0; + if (fm) { + *full_pk_ix = ph_pkey_ix; + search_pkey &= 0x7FFF; + } else { + *partial_pk_ix = ph_pkey_ix; + search_pkey |= 0x8000; + } + + if (ib_find_exact_cached_pkey(&ibdev->ib_dev, port, search_pkey, &pk)) + pk = 0xFFFF; + + if (fm) + *partial_pk_ix = (pk & 0xFF); + else + *full_pk_ix = (pk & 0xFF); + + *is_full_member = fm; + return err; +} + +int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port, + enum ib_qp_type dest_qpt, struct ib_wc *wc, + struct ib_grh *grh, struct ib_mad *mad) +{ + struct ib_sge list; + struct ib_send_wr wr, *bad_wr; + struct mlx4_ib_demux_pv_ctx *tun_ctx; + struct mlx4_ib_demux_pv_qp *tun_qp; + struct mlx4_rcv_tunnel_mad *tun_mad; + struct ib_ah_attr attr; + struct ib_ah *ah; + struct ib_qp *src_qp = NULL; + unsigned tun_tx_ix = 0; + int dqpn; + int ret = 0; + int i; + int is_full_member = 0; + u16 tun_pkey_ix; + u8 ph_pkey_ix, full_pk_ix = 0, partial_pk_ix = 0; + + if (dest_qpt > IB_QPT_GSI) + return -EINVAL; + + tun_ctx = dev->sriov.demux[port-1].tun[slave]; + + /* check if proxy qp created */ + if (!tun_ctx || tun_ctx->state != DEMUX_PV_STATE_ACTIVE) + return -EAGAIN; + + /* QP0 forwarding only for Dom0 */ + if (!dest_qpt && (mlx4_master_func_num(dev->dev) != slave)) + return -EINVAL; + + if (!dest_qpt) + tun_qp = &tun_ctx->qp[0]; + else + tun_qp = &tun_ctx->qp[1]; + + /* compute pkey index for slave */ + /* get physical pkey -- virtualized Dom0 pkey to phys*/ + if (dest_qpt) { + ph_pkey_ix = + dev->pkeys.virt2phys_pkey[mlx4_master_func_num(dev->dev)][port - 1][wc->pkey_index]; + + /* now, translate this to the slave pkey index */ + ret = get_pkey_phys_indices(dev, port, ph_pkey_ix, &full_pk_ix, + &partial_pk_ix, &is_full_member); + if (ret) + return -EINVAL; + + for (i = 0; i < dev->dev->caps.pkey_table_len[port]; i++) { + if ((dev->pkeys.virt2phys_pkey[slave][port - 1][i] == full_pk_ix) || + (is_full_member && + (dev->pkeys.virt2phys_pkey[slave][port - 1][i] == partial_pk_ix))) + break; + } + if (i == dev->dev->caps.pkey_table_len[port]) + return -EINVAL; + tun_pkey_ix = i; + } else + tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0]; + + dqpn = dev->dev->caps.sqp_start + 8 * slave + port + (dest_qpt * 2) - 1; + + /* get tunnel tx data buf for slave */ + src_qp = tun_qp->qp; + + /* create ah. Just need an empty one with the port num for the post send. + * The driver will set the force loopback bit in post_send */ + memset(&attr, 0, sizeof attr); + attr.port_num = port; + ah = ib_create_ah(tun_ctx->pd, &attr); + if (IS_ERR(ah)) + return -ENOMEM; + + /* allocate tunnel tx buf after pass failure returns */ + spin_lock(&tun_qp->tx_lock); + if (tun_qp->tx_ix_head - tun_qp->tx_ix_tail >= + (MLX4_NUM_TUNNEL_BUFS - 1)) + ret = -EAGAIN; + else + tun_tx_ix = (++tun_qp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1); + spin_unlock(&tun_qp->tx_lock); + if (ret) + goto out; + + tun_mad = (struct mlx4_rcv_tunnel_mad *) (tun_qp->tx_ring[tun_tx_ix].buf.addr); + if (tun_qp->tx_ring[tun_tx_ix].ah) + ib_destroy_ah(tun_qp->tx_ring[tun_tx_ix].ah); + tun_qp->tx_ring[tun_tx_ix].ah = ah; + ib_dma_sync_single_for_cpu(&dev->ib_dev, + tun_qp->tx_ring[tun_tx_ix].buf.map, + sizeof (struct mlx4_rcv_tunnel_mad), + DMA_TO_DEVICE); + + /* copy over to tunnel buffer */ + if (grh) + memcpy(&tun_mad->grh, grh, sizeof *grh); + memcpy(&tun_mad->mad, mad, sizeof *mad); + + /* adjust tunnel data */ + tun_mad->hdr.pkey_index = cpu_to_be16(tun_pkey_ix); + tun_mad->hdr.sl_vid = cpu_to_be16(((u16)(wc->sl)) << 12); + tun_mad->hdr.slid_mac_47_32 = cpu_to_be16(wc->slid); + tun_mad->hdr.flags_src_qp = cpu_to_be32(wc->src_qp & 0xFFFFFF); + tun_mad->hdr.g_ml_path = (grh && (wc->wc_flags & IB_WC_GRH)) ? 0x80 : 0; + + ib_dma_sync_single_for_device(&dev->ib_dev, + tun_qp->tx_ring[tun_tx_ix].buf.map, + sizeof (struct mlx4_rcv_tunnel_mad), + DMA_TO_DEVICE); + + list.addr = tun_qp->tx_ring[tun_tx_ix].buf.map; + list.length = sizeof (struct mlx4_rcv_tunnel_mad); + list.lkey = tun_ctx->mr->lkey; + + wr.wr.ud.ah = ah; + wr.wr.ud.port_num = port; + wr.wr.ud.remote_qkey = IB_QP_SET_QKEY; + wr.wr.ud.remote_qpn = dqpn; + wr.next = NULL; + wr.wr_id = ((u64) tun_tx_ix) | MLX4_TUN_SET_WRID_QPN(dest_qpt); + wr.sg_list = &list; + wr.num_sge = 1; + wr.opcode = IB_WR_SEND; + wr.send_flags = IB_SEND_SIGNALED; + + ret = ib_post_send(src_qp, &wr, &bad_wr); +out: + if (ret) + ib_destroy_ah(ah); + return ret; +} + +static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port, + struct ib_wc *wc, struct ib_grh *grh, + struct ib_mad *mad) +{ + struct mlx4_ib_dev *dev = to_mdev(ibdev); + int err; + int slave; + u8 *slave_id; + + /* Initially assume that this mad is for us */ + slave = mlx4_master_func_num(dev->dev); + + /* See if the slave id is encoded in a response mad */ + if (mad->mad_hdr.method & 0x80) { + slave_id = (u8 *) &mad->mad_hdr.tid; + slave = *slave_id; + if (slave != 255) /*255 indicates the dom0*/ + *slave_id = 0; /* remap tid */ + } + + /* If a grh is present, we demux according to it */ + if (wc->wc_flags & IB_WC_GRH) { + slave = mlx4_ib_find_real_gid(ibdev, port, grh->dgid.global.interface_id); + if (slave < 0) { + mlx4_ib_warn(ibdev, "failed matching grh\n"); + return -ENOENT; + } + } + /* Class-specific handling */ + switch (mad->mad_hdr.mgmt_class) { + case IB_MGMT_CLASS_SUBN_ADM: + if (mlx4_ib_demux_sa_handler(ibdev, port, slave, + (struct ib_sa_mad *) mad)) + return 0; + break; + case IB_MGMT_CLASS_DEVICE_MGMT: + if (mad->mad_hdr.method != IB_MGMT_METHOD_GET_RESP) + return 0; + break; + default: + /* Drop unsupported classes for slaves in tunnel mode */ + if (slave != mlx4_master_func_num(dev->dev)) { + pr_debug("dropping unsupported ingress mad from class:%d " + "for slave:%d\n", mad->mad_hdr.mgmt_class, slave); + return 0; + } + } + /*make sure that no slave==255 was not handled yet.*/ + if (slave >= dev->dev->caps.sqp_demux) { + mlx4_ib_warn(ibdev, "slave id: %d is bigger than allowed:%d\n", + slave, dev->dev->caps.sqp_demux); + return -ENOENT; + } + + err = mlx4_ib_send_to_slave(dev, slave, port, wc->qp->qp_type, wc, grh, mad); + if (err) + pr_debug("failed sending to slave %d via tunnel qp (%d)\n", + slave, err); + return 0; +} + static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num, struct ib_wc *in_wc, struct ib_grh *in_grh, struct ib_mad *in_mad, struct ib_mad *out_mad) @@ -611,6 +861,216 @@ static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx, return ib_post_recv(tun_qp->qp, &recv_wr, &bad_recv_wr); } +static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port, + int slave, struct ib_sa_mad *sa_mad) +{ + return 0; +} + +static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave) +{ + int slave_start = dev->dev->caps.sqp_start + 8 * slave; + + return (qpn >= slave_start && qpn <= slave_start + 1); +} + + +int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port, + enum ib_qp_type dest_qpt, u16 pkey_index, u32 remote_qpn, + u32 qkey, struct ib_ah_attr *attr, struct ib_mad *mad) +{ + struct ib_sge list; + struct ib_send_wr wr, *bad_wr; + struct mlx4_ib_demux_pv_ctx *sqp_ctx; + struct mlx4_ib_demux_pv_qp *sqp; + struct mlx4_mad_snd_buf *sqp_mad; + struct ib_ah *ah; + struct ib_qp *send_qp = NULL; + unsigned wire_tx_ix = 0; + int ret = 0; + u16 wire_pkey_ix; + int src_qpnum; + u8 sgid_index; + + + sqp_ctx = dev->sriov.sqps[port-1]; + + /* check if proxy qp created */ + if (!sqp_ctx || sqp_ctx->state != DEMUX_PV_STATE_ACTIVE) + return -EAGAIN; + + /* QP0 forwarding only for Dom0 */ + if (dest_qpt == IB_QPT_SMI && (mlx4_master_func_num(dev->dev) != slave)) + return -EINVAL; + + if (dest_qpt == IB_QPT_SMI) { + src_qpnum = 0; + sqp = &sqp_ctx->qp[0]; + wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0]; + } else { + src_qpnum = 1; + sqp = &sqp_ctx->qp[1]; + wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][pkey_index]; + } + + send_qp = sqp->qp; + + /* create ah */ + sgid_index = attr->grh.sgid_index; + attr->grh.sgid_index = 0; + ah = ib_create_ah(sqp_ctx->pd, attr); + if (IS_ERR(ah)) + return -ENOMEM; + attr->grh.sgid_index = sgid_index; + to_mah(ah)->av.ib.gid_index = sgid_index; + /* get rid of force-loopback bit */ + to_mah(ah)->av.ib.port_pd &= cpu_to_be32(0x7FFFFFFF); + spin_lock(&sqp->tx_lock); + if (sqp->tx_ix_head - sqp->tx_ix_tail >= + (MLX4_NUM_TUNNEL_BUFS - 1)) + ret = -EAGAIN; + else + wire_tx_ix = (++sqp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1); + spin_unlock(&sqp->tx_lock); + if (ret) + goto out; + + sqp_mad = (struct mlx4_mad_snd_buf *) (sqp->tx_ring[wire_tx_ix].buf.addr); + if (sqp->tx_ring[wire_tx_ix].ah) + ib_destroy_ah(sqp->tx_ring[wire_tx_ix].ah); + sqp->tx_ring[wire_tx_ix].ah = ah; + ib_dma_sync_single_for_cpu(&dev->ib_dev, + sqp->tx_ring[wire_tx_ix].buf.map, + sizeof (struct mlx4_mad_snd_buf), + DMA_TO_DEVICE); + + memcpy(&sqp_mad->payload, mad, sizeof *mad); + + ib_dma_sync_single_for_device(&dev->ib_dev, + sqp->tx_ring[wire_tx_ix].buf.map, + sizeof (struct mlx4_mad_snd_buf), + DMA_TO_DEVICE); + + list.addr = sqp->tx_ring[wire_tx_ix].buf.map; + list.length = sizeof (struct mlx4_mad_snd_buf); + list.lkey = sqp_ctx->mr->lkey; + + wr.wr.ud.ah = ah; + wr.wr.ud.port_num = port; + wr.wr.ud.pkey_index = wire_pkey_ix; + wr.wr.ud.remote_qkey = qkey; + wr.wr.ud.remote_qpn = remote_qpn; + wr.next = NULL; + wr.wr_id = ((u64) wire_tx_ix) | MLX4_TUN_SET_WRID_QPN(src_qpnum); + wr.sg_list = &list; + wr.num_sge = 1; + wr.opcode = IB_WR_SEND; + wr.send_flags = IB_SEND_SIGNALED; + + ret = ib_post_send(send_qp, &wr, &bad_wr); +out: + if (ret) + ib_destroy_ah(ah); + return ret; +} + +static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc *wc) +{ + struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev); + struct mlx4_ib_demux_pv_qp *tun_qp = &ctx->qp[MLX4_TUN_WRID_QPN(wc->wr_id)]; + int wr_ix = wc->wr_id & (MLX4_NUM_TUNNEL_BUFS - 1); + struct mlx4_tunnel_mad *tunnel = tun_qp->ring[wr_ix].addr; + struct mlx4_ib_ah ah; + struct ib_ah_attr ah_attr; + u8 *slave_id; + int slave; + + /* Get slave that sent this packet */ + if (wc->src_qp < dev->dev->caps.sqp_start || + wc->src_qp >= dev->dev->caps.base_tunnel_sqpn || + (wc->src_qp & 0x1) != ctx->port - 1 || + wc->src_qp & 0x4) { + mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d\n", wc->src_qp); + return; + } + slave = ((wc->src_qp & ~0x7) - dev->dev->caps.sqp_start) / 8; + if (slave != ctx->slave) { + mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: " + "belongs to another slave\n", wc->src_qp); + return; + } + if (slave != mlx4_master_func_num(dev->dev) && !(wc->src_qp & 0x2)) { + mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: " + "non-master trying to send QP0 packets\n", wc->src_qp); + return; + } + + /* Map transaction ID */ + ib_dma_sync_single_for_cpu(ctx->ib_dev, tun_qp->ring[wr_ix].map, + sizeof (struct mlx4_tunnel_mad), + DMA_FROM_DEVICE); + switch (tunnel->mad.mad_hdr.method) { + case IB_MGMT_METHOD_SET: + case IB_MGMT_METHOD_GET: + case IB_MGMT_METHOD_REPORT: + case IB_SA_METHOD_GET_TABLE: + case IB_SA_METHOD_DELETE: + case IB_SA_METHOD_GET_MULTI: + case IB_SA_METHOD_GET_TRACE_TBL: + slave_id = (u8 *) &tunnel->mad.mad_hdr.tid; + if (*slave_id) { + mlx4_ib_warn(ctx->ib_dev, "egress mad has non-null tid msb:%d " + "class:%d slave:%d\n", *slave_id, + tunnel->mad.mad_hdr.mgmt_class, slave); + return; + } else + *slave_id = slave; + default: + /* nothing */; + } + + /* Class-specific handling */ + switch (tunnel->mad.mad_hdr.mgmt_class) { + case IB_MGMT_CLASS_SUBN_ADM: + if (mlx4_ib_multiplex_sa_handler(ctx->ib_dev, ctx->port, slave, + (struct ib_sa_mad *) &tunnel->mad)) + return; + break; + case IB_MGMT_CLASS_DEVICE_MGMT: + if (tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_GET && + tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_SET) + return; + break; + default: + /* Drop unsupported classes for slaves in tunnel mode */ + if (slave != mlx4_master_func_num(dev->dev)) { + mlx4_ib_warn(ctx->ib_dev, "dropping unsupported egress mad from class:%d " + "for slave:%d\n", tunnel->mad.mad_hdr.mgmt_class, slave); + return; + } + } + + /* We are using standard ib_core services to send the mad, so generate a + * stadard address handle by decoding the tunnelled mlx4_ah fields */ + memcpy(&ah.av, &tunnel->hdr.av, sizeof (struct mlx4_av)); + ah.ibah.device = ctx->ib_dev; + mlx4_ib_query_ah(&ah.ibah, &ah_attr); + if ((ah_attr.ah_flags & IB_AH_GRH) && + (ah_attr.grh.sgid_index != slave)) { + mlx4_ib_warn(ctx->ib_dev, "slave:%d accessed invalid sgid_index:%d\n", + slave, ah_attr.grh.sgid_index); + return; + } + + mlx4_ib_send_to_wire(dev, slave, ctx->port, + is_proxy_qp0(dev, wc->src_qp, slave) ? + IB_QPT_SMI : IB_QPT_GSI, + be16_to_cpu(tunnel->hdr.pkey_index), + be32_to_cpu(tunnel->hdr.remote_qpn), + be32_to_cpu(tunnel->hdr.qkey), + &ah_attr, &tunnel->mad); +} + static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx, enum ib_qp_type qp_type, int is_tun) { @@ -735,7 +1195,57 @@ static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx, static void mlx4_ib_tunnel_comp_worker(struct work_struct *work) { - /* dummy until next patch in series */ + struct mlx4_ib_demux_pv_ctx *ctx; + struct mlx4_ib_demux_pv_qp *tun_qp; + struct ib_wc wc; + int ret; + ctx = container_of(work, struct mlx4_ib_demux_pv_ctx, work); + ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP); + + while (ib_poll_cq(ctx->cq, 1, &wc) == 1) { + tun_qp = &ctx->qp[MLX4_TUN_WRID_QPN(wc.wr_id)]; + if (wc.status == IB_WC_SUCCESS) { + switch (wc.opcode) { + case IB_WC_RECV: + mlx4_ib_multiplex_mad(ctx, &wc); + ret = mlx4_ib_post_pv_qp_buf(ctx, tun_qp, + wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1)); + if (ret) + pr_err("Failed reposting tunnel " + "buf:%lld\n", wc.wr_id); + break; + case IB_WC_SEND: + pr_debug("received tunnel send completion:" + "wrid=0x%llx, status=0x%x\n", + wc.wr_id, wc.status); + ib_destroy_ah(tun_qp->tx_ring[wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1)].ah); + tun_qp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah + = NULL; + spin_lock(&tun_qp->tx_lock); + tun_qp->tx_ix_tail++; + spin_unlock(&tun_qp->tx_lock); + + break; + default: + break; + } + } else { + pr_debug("mlx4_ib: completion error in tunnel: %d." + " status = %d, wrid = 0x%llx\n", + ctx->slave, wc.status, wc.wr_id); + if (!MLX4_TUN_IS_RECV(wc.wr_id)) { + ib_destroy_ah(tun_qp->tx_ring[wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1)].ah); + tun_qp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah + = NULL; + spin_lock(&tun_qp->tx_lock); + tun_qp->tx_ix_tail++; + spin_unlock(&tun_qp->tx_lock); + } + } + } } static void pv_qp_event_handler(struct ib_event *event, void *qp_context) @@ -843,7 +1353,60 @@ err_qp: */ static void mlx4_ib_sqp_comp_worker(struct work_struct *work) { - /* dummy until next patch in series */ + struct mlx4_ib_demux_pv_ctx *ctx; + struct mlx4_ib_demux_pv_qp *sqp; + struct ib_wc wc; + struct ib_grh *grh; + struct ib_mad *mad; + + ctx = container_of(work, struct mlx4_ib_demux_pv_ctx, work); + ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP); + + while (mlx4_ib_poll_cq(ctx->cq, 1, &wc) == 1) { + sqp = &ctx->qp[MLX4_TUN_WRID_QPN(wc.wr_id)]; + if (wc.status == IB_WC_SUCCESS) { + switch (wc.opcode) { + case IB_WC_SEND: + ib_destroy_ah(sqp->tx_ring[wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1)].ah); + sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah + = NULL; + spin_lock(&sqp->tx_lock); + sqp->tx_ix_tail++; + spin_unlock(&sqp->tx_lock); + break; + case IB_WC_RECV: + mad = (struct ib_mad *) &(((struct mlx4_mad_rcv_buf *) + (sqp->ring[wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1)].addr))->payload); + grh = &(((struct mlx4_mad_rcv_buf *) + (sqp->ring[wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1)].addr))->grh); + mlx4_ib_demux_mad(ctx->ib_dev, ctx->port, &wc, grh, mad); + if (mlx4_ib_post_pv_qp_buf(ctx, sqp, wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1))) + pr_err("Failed reposting SQP " + "buf:%lld\n", wc.wr_id); + break; + default: + BUG_ON(1); + break; + } + } else { + pr_debug("mlx4_ib: completion error in tunnel: %d." + " status = %d, wrid = 0x%llx\n", + ctx->slave, wc.status, wc.wr_id); + if (!MLX4_TUN_IS_RECV(wc.wr_id)) { + ib_destroy_ah(sqp->tx_ring[wc.wr_id & + (MLX4_NUM_TUNNEL_BUFS - 1)].ah); + sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah + = NULL; + spin_lock(&sqp->tx_lock); + sqp->tx_ix_tail++; + spin_unlock(&sqp->tx_lock); + } + } + } } static int alloc_pv_object(struct mlx4_ib_dev *dev, int slave, int port, From 0a9a01884d447c216eff75f8f274a0a3e82c7cee Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:45 +0000 Subject: [PATCH 16/45] mlx4: MAD_IFC paravirtualization The MAD_IFC firmware command fulfills two functions. First, it is used in the QP0/QP1 MAD-handling flow to obtain information from the FW (for answering queries), and for setting variables in the HCA (MAD SET packets). For this, MAD_IFC should provide the FW (physical) view of the data. This is the view that OpenSM needs. We call this the "network view". In the second case, MAD_IFC is used by various verbs to obtain data regarding the local HCA (e.g., ib_query_device()). We call this the "host view". This data needs to be paravirtualized. MAD_IFC therefore needs a wrapper function, and also needs another flag indicating whether it should provide the network view (when it is called by ib_process_mad in special-qp packet handling), or the host view (when it is called while implementing a verb). There are currently 2 flag parameters in mlx4_MAD_IFC already: ignore_bkey and ignore_mkey. These two parameters are replaced by a single "mad_ifc_flags" parameter, with different bits set for each flag. A third flag is added: "network-view/host-view". Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 20 +-- drivers/infiniband/hw/mlx4/main.c | 64 ++++++--- drivers/infiniband/hw/mlx4/mlx4_ib.h | 14 +- drivers/net/ethernet/mellanox/mlx4/cmd.c | 162 +++++++++++++++++++++++ 4 files changed, 234 insertions(+), 26 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 8dfbf69f8370..ba2580693f79 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -75,7 +75,7 @@ struct mlx4_rcv_tunnel_mad { struct ib_mad mad; } __packed; -int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey, +int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags, int port, struct ib_wc *in_wc, struct ib_grh *in_grh, void *in_mad, void *response_mad) { @@ -102,10 +102,13 @@ int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey, * Key check traps can't be generated unless we have in_wc to * tell us where to send the trap. */ - if (ignore_mkey || !in_wc) + if ((mad_ifc_flags & MLX4_MAD_IFC_IGNORE_MKEY) || !in_wc) op_modifier |= 0x1; - if (ignore_bkey || !in_wc) + if ((mad_ifc_flags & MLX4_MAD_IFC_IGNORE_BKEY) || !in_wc) op_modifier |= 0x2; + if (mlx4_is_mfunc(dev->dev) && + (mad_ifc_flags & MLX4_MAD_IFC_NET_VIEW || in_wc)) + op_modifier |= 0x8; if (in_wc) { struct { @@ -138,10 +141,10 @@ int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey, in_modifier |= in_wc->slid << 16; } - err = mlx4_cmd_box(dev->dev, inmailbox->dma, outmailbox->dma, - in_modifier, op_modifier, + err = mlx4_cmd_box(dev->dev, inmailbox->dma, outmailbox->dma, in_modifier, + mlx4_is_master(dev->dev) ? (op_modifier & ~0x8) : op_modifier, MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C, - MLX4_CMD_NATIVE); + (op_modifier & 0x8) ? MLX4_CMD_NATIVE : MLX4_CMD_WRAPPED); if (!err) memcpy(response_mad, outmailbox->buf, 256); @@ -614,8 +617,9 @@ static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num, prev_lid = pattr.lid; err = mlx4_MAD_IFC(to_mdev(ibdev), - mad_flags & IB_MAD_IGNORE_MKEY, - mad_flags & IB_MAD_IGNORE_BKEY, + (mad_flags & IB_MAD_IGNORE_MKEY ? MLX4_MAD_IFC_IGNORE_MKEY : 0) | + (mad_flags & IB_MAD_IGNORE_BKEY ? MLX4_MAD_IFC_IGNORE_BKEY : 0) | + MLX4_MAD_IFC_NET_VIEW, port_num, in_wc, in_grh, in_mad, out_mad); if (err) return IB_MAD_RESULT_FAILURE; diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index 8e10ec2af7b6..45a6cc04036b 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -98,7 +98,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev, init_query_mad(in_mad); in_mad->attr_id = IB_SMP_ATTR_NODE_INFO; - err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad); + err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, + 1, NULL, NULL, in_mad, out_mad); if (err) goto out; @@ -182,11 +183,12 @@ mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num) } static int ib_link_query_port(struct ib_device *ibdev, u8 port, - struct ib_port_attr *props) + struct ib_port_attr *props, int netw_view) { struct ib_smp *in_mad = NULL; struct ib_smp *out_mad = NULL; int ext_active_speed; + int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; int err = -ENOMEM; in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); @@ -198,7 +200,10 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port, in_mad->attr_id = IB_SMP_ATTR_PORT_INFO; in_mad->attr_mod = cpu_to_be32(port); - err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, + if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view) + mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; + + err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL, in_mad, out_mad); if (err) goto out; @@ -211,7 +216,10 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port, props->state = out_mad->data[32] & 0xf; props->phys_state = out_mad->data[33] >> 4; props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20)); - props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port]; + if (netw_view) + props->gid_tbl_len = out_mad->data[50]; + else + props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port]; props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz; props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port]; props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46)); @@ -244,7 +252,7 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port, in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO; in_mad->attr_mod = cpu_to_be32(port); - err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, + err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL, in_mad, out_mad); if (err) goto out; @@ -270,7 +278,7 @@ static u8 state_to_phys_state(enum ib_port_state state) } static int eth_link_query_port(struct ib_device *ibdev, u8 port, - struct ib_port_attr *props) + struct ib_port_attr *props, int netw_view) { struct mlx4_ib_dev *mdev = to_mdev(ibdev); @@ -320,20 +328,27 @@ out: return err; } -static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port, - struct ib_port_attr *props) +int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port, + struct ib_port_attr *props, int netw_view) { int err; memset(props, 0, sizeof *props); err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ? - ib_link_query_port(ibdev, port, props) : - eth_link_query_port(ibdev, port, props); + ib_link_query_port(ibdev, port, props, netw_view) : + eth_link_query_port(ibdev, port, props, netw_view); return err; } +static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port, + struct ib_port_attr *props) +{ + /* returns host view */ + return __mlx4_ib_query_port(ibdev, port, props, 0); +} + static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, union ib_gid *gid) { @@ -350,7 +365,8 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, in_mad->attr_id = IB_SMP_ATTR_PORT_INFO; in_mad->attr_mod = cpu_to_be32(port); - err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad); + err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port, + NULL, NULL, in_mad, out_mad); if (err) goto out; @@ -360,7 +376,8 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, in_mad->attr_id = IB_SMP_ATTR_GUID_INFO; in_mad->attr_mod = cpu_to_be32(index / 8); - err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad); + err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port, + NULL, NULL, in_mad, out_mad); if (err) goto out; @@ -391,11 +408,12 @@ static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, return iboe_query_gid(ibdev, port, index, gid); } -static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, - u16 *pkey) +int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, + u16 *pkey, int netw_view) { struct ib_smp *in_mad = NULL; struct ib_smp *out_mad = NULL; + int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; int err = -ENOMEM; in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); @@ -407,7 +425,11 @@ static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE; in_mad->attr_mod = cpu_to_be32(index / 32); - err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad); + if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view) + mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; + + err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL, + in_mad, out_mad); if (err) goto out; @@ -419,6 +441,11 @@ out: return err; } +static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) +{ + return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0); +} + static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask, struct ib_device_modify *props) { @@ -849,6 +876,7 @@ static int init_node_data(struct mlx4_ib_dev *dev) { struct ib_smp *in_mad = NULL; struct ib_smp *out_mad = NULL; + int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; int err = -ENOMEM; in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); @@ -858,8 +886,10 @@ static int init_node_data(struct mlx4_ib_dev *dev) init_query_mad(in_mad); in_mad->attr_id = IB_SMP_ATTR_NODE_DESC; + if (mlx4_is_master(dev->dev)) + mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; - err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad); + err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad); if (err) goto out; @@ -867,7 +897,7 @@ static int init_node_data(struct mlx4_ib_dev *dev) in_mad->attr_id = IB_SMP_ATTR_NODE_INFO; - err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad); + err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad); if (err) goto out; diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index 137941d79870..ac71d56ffc7e 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -176,6 +176,14 @@ enum mlx4_ib_qp_type { MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER | \ MLX4_IB_QPT_TUN_SMI | MLX4_IB_QPT_TUN_GSI) +enum mlx4_ib_mad_ifc_flags { + MLX4_MAD_IFC_IGNORE_MKEY = 1, + MLX4_MAD_IFC_IGNORE_BKEY = 2, + MLX4_MAD_IFC_IGNORE_KEYS = (MLX4_MAD_IFC_IGNORE_MKEY | + MLX4_MAD_IFC_IGNORE_BKEY), + MLX4_MAD_IFC_NET_VIEW = 4, +}; + enum { MLX4_NUM_TUNNEL_BUFS = 256, }; @@ -512,7 +520,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, struct ib_recv_wr **bad_wr); -int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey, +int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags, int port, struct ib_wc *in_wc, struct ib_grh *in_grh, void *in_mad, void *response_mad); int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num, @@ -527,6 +535,10 @@ int mlx4_ib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list, int npages, u64 iova); int mlx4_ib_unmap_fmr(struct list_head *fmr_list); int mlx4_ib_fmr_dealloc(struct ib_fmr *fmr); +int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port, + struct ib_port_attr *props, int netw_view); +int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, + u16 *pkey, int netw_view); int mlx4_ib_resolve_grh(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah_attr, u8 *mac, int *is_mcast, u8 port); diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index 662a3c5da739..a13d8a69b3bc 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -40,6 +40,7 @@ #include #include +#include #include @@ -627,6 +628,149 @@ static int mlx4_ACCESS_MEM(struct mlx4_dev *dev, u64 master_addr, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); } +static int query_pkey_block(struct mlx4_dev *dev, u8 port, u16 index, u16 *pkey, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox) +{ + struct ib_smp *in_mad = (struct ib_smp *)(inbox->buf); + struct ib_smp *out_mad = (struct ib_smp *)(outbox->buf); + int err; + int i; + + if (index & 0x1f) + return -EINVAL; + + in_mad->attr_mod = cpu_to_be32(index / 32); + + err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, port, 3, + MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C, + MLX4_CMD_NATIVE); + if (err) + return err; + + for (i = 0; i < 32; ++i) + pkey[i] = be16_to_cpu(((__be16 *) out_mad->data)[i]); + + return err; +} + +static int get_full_pkey_table(struct mlx4_dev *dev, u8 port, u16 *table, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox) +{ + int i; + int err; + + for (i = 0; i < dev->caps.pkey_table_len[port]; i += 32) { + err = query_pkey_block(dev, port, i, table + i, inbox, outbox); + if (err) + return err; + } + + return 0; +} +#define PORT_CAPABILITY_LOCATION_IN_SMP 20 +#define PORT_STATE_OFFSET 32 + +static enum ib_port_state vf_port_state(struct mlx4_dev *dev, int port, int vf) +{ + /* will be modified when add alias_guid feature */ + return IB_PORT_DOWN; +} + +static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave, + struct mlx4_vhcr *vhcr, + struct mlx4_cmd_mailbox *inbox, + struct mlx4_cmd_mailbox *outbox, + struct mlx4_cmd_info *cmd) +{ + struct ib_smp *smp = inbox->buf; + u32 index; + u8 port; + u16 *table; + int err; + int vidx, pidx; + struct mlx4_priv *priv = mlx4_priv(dev); + struct ib_smp *outsmp = outbox->buf; + __be16 *outtab = (__be16 *)(outsmp->data); + __be32 slave_cap_mask; + port = vhcr->in_modifier; + + if (smp->base_version == 1 && + smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED && + smp->class_version == 1) { + if (smp->method == IB_MGMT_METHOD_GET) { + if (smp->attr_id == IB_SMP_ATTR_PKEY_TABLE) { + index = be32_to_cpu(smp->attr_mod); + if (port < 1 || port > dev->caps.num_ports) + return -EINVAL; + table = kcalloc(dev->caps.pkey_table_len[port], sizeof *table, GFP_KERNEL); + if (!table) + return -ENOMEM; + /* need to get the full pkey table because the paravirtualized + * pkeys may be scattered among several pkey blocks. + */ + err = get_full_pkey_table(dev, port, table, inbox, outbox); + if (!err) { + for (vidx = index * 32; vidx < (index + 1) * 32; ++vidx) { + pidx = priv->virt2phys_pkey[slave][port - 1][vidx]; + outtab[vidx % 32] = cpu_to_be16(table[pidx]); + } + } + kfree(table); + return err; + } + if (smp->attr_id == IB_SMP_ATTR_PORT_INFO) { + /*get the slave specific caps:*/ + /*do the command */ + err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, + vhcr->in_modifier, vhcr->op_modifier, + vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE); + /* modify the response for slaves */ + if (!err && slave != mlx4_master_func_num(dev)) { + u8 *state = outsmp->data + PORT_STATE_OFFSET; + + *state = (*state & 0xf0) | vf_port_state(dev, port, slave); + slave_cap_mask = priv->mfunc.master.slave_state[slave].ib_cap_mask[port]; + memcpy(outsmp->data + PORT_CAPABILITY_LOCATION_IN_SMP, &slave_cap_mask, 4); + } + return err; + } + if (smp->attr_id == IB_SMP_ATTR_GUID_INFO) { + /* compute slave's gid block */ + smp->attr_mod = cpu_to_be32(slave / 8); + /* execute cmd */ + err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, + vhcr->in_modifier, vhcr->op_modifier, + vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE); + if (!err) { + /* if needed, move slave gid to index 0 */ + if (slave % 8) + memcpy(outsmp->data, + outsmp->data + (slave % 8) * 8, 8); + /* delete all other gids */ + memset(outsmp->data + 8, 0, 56); + } + return err; + } + } + } + if (slave != mlx4_master_func_num(dev) && + ((smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) || + (smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED && + smp->method == IB_MGMT_METHOD_SET))) { + mlx4_err(dev, "slave %d is trying to execute a Subnet MGMT MAD, " + "class 0x%x, method 0x%x for attr 0x%x. Rejecting\n", + slave, smp->method, smp->mgmt_class, + be16_to_cpu(smp->attr_id)); + return -EPERM; + } + /*default:*/ + return mlx4_cmd_box(dev, inbox->dma, outbox->dma, + vhcr->in_modifier, vhcr->op_modifier, + vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE); +} + int mlx4_DMA_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, @@ -1060,6 +1204,24 @@ static struct mlx4_cmd_info cmd_info[] = { .verify = NULL, .wrapper = mlx4_GEN_QP_wrapper }, + { + .opcode = MLX4_CMD_CONF_SPECIAL_QP, + .has_inbox = false, + .has_outbox = false, + .out_is_imm = false, + .encode_slave_id = false, + .verify = NULL, /* XXX verify: only demux can do this */ + .wrapper = NULL + }, + { + .opcode = MLX4_CMD_MAD_IFC, + .has_inbox = true, + .has_outbox = true, + .out_is_imm = false, + .encode_slave_id = false, + .verify = NULL, + .wrapper = mlx4_MAD_IFC_wrapper + }, { .opcode = MLX4_CMD_QUERY_IF_STAT, .has_inbox = false, From b9c5d6a643589ad39064f652938baa698f0e884a Mon Sep 17 00:00:00 2001 From: Oren Duer Date: Fri, 3 Aug 2012 08:40:46 +0000 Subject: [PATCH 17/45] IB/mlx4: Add multicast group (MCG) paravirtualization for SR-IOV MCG paravirtualization support includes: - Creating multicast groups by VFs, and keeping accounting of them - Leaving multicast groups by VFs - Updating SM only with real changes in the overall picture of MCGs status - Creation of MGID=0 groups (let SM choose MGID) Note that the MCG module maintains its own internal MCG object reference counts. The reason for this is that the IB core is used to track only the multicast groups joins generated by the PF it runs over. The PF IB core layer is unaware of slaves, so it cannot be used to keep track of MCG joins they generate. Signed-off-by: Oren Duer Signed-off-by: Eli Cohen Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/Makefile | 2 +- drivers/infiniband/hw/mlx4/mad.c | 60 +- drivers/infiniband/hw/mlx4/main.c | 18 +- drivers/infiniband/hw/mlx4/mcg.c | 1187 ++++++++++++++++++++++++++ drivers/infiniband/hw/mlx4/mlx4_ib.h | 30 + 5 files changed, 1285 insertions(+), 12 deletions(-) create mode 100644 drivers/infiniband/hw/mlx4/mcg.c diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile index 70f09c7826da..20d627d1f046 100644 --- a/drivers/infiniband/hw/mlx4/Makefile +++ b/drivers/infiniband/hw/mlx4/Makefile @@ -1,3 +1,3 @@ obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o -mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o +mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index ba2580693f79..29ed3b43e4a6 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -75,6 +75,14 @@ struct mlx4_rcv_tunnel_mad { struct ib_mad mad; } __packed; +static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num); + +__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx) +{ + return cpu_to_be64(atomic_inc_return(&ctx->tid)) | + cpu_to_be64(0xff00000000000000LL); +} + int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags, int port, struct ib_wc *in_wc, struct ib_grh *in_grh, void *in_mad, void *response_mad) @@ -209,8 +217,7 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad, pinfo->neighbormtu_mastersmsl & 0xf); if (pinfo->clientrereg_resv_subnetto & 0x80) - mlx4_ib_dispatch_event(dev, port_num, - IB_EVENT_CLIENT_REREGISTER); + handle_client_rereg_event(dev, port_num); if (prev_lid != lid) mlx4_ib_dispatch_event(dev, port_num, @@ -308,7 +315,17 @@ static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *ma static int mlx4_ib_demux_sa_handler(struct ib_device *ibdev, int port, int slave, struct ib_sa_mad *sa_mad) { - return 0; + int ret = 0; + + /* dispatch to different sa handlers */ + switch (be16_to_cpu(sa_mad->mad_hdr.attr_id)) { + case IB_SA_ATTR_MC_MEMBER_REC: + ret = mlx4_ib_mcg_demux_handler(ibdev, port, slave, sa_mad); + break; + default: + break; + } + return ret; } int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid) @@ -768,6 +785,16 @@ void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev) } } +static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num) +{ + /* re-configure the mcg's */ + if (mlx4_is_master(dev->dev)) { + if (!dev->sriov.is_going_down) + mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0); + } + mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_CLIENT_REREGISTER); +} + void handle_port_mgmt_change_event(struct work_struct *work) { struct ib_event_work *ew = container_of(work, struct ib_event_work, work); @@ -797,8 +824,7 @@ void handle_port_mgmt_change_event(struct work_struct *work) mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE); if (changed_attr & MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK) - mlx4_ib_dispatch_event(dev, port, - IB_EVENT_CLIENT_REREGISTER); + handle_client_rereg_event(dev, port); break; case MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE: @@ -868,7 +894,17 @@ static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx, static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port, int slave, struct ib_sa_mad *sa_mad) { - return 0; + int ret = 0; + + /* dispatch to different sa handlers */ + switch (be16_to_cpu(sa_mad->mad_hdr.attr_id)) { + case IB_SA_ATTR_MC_MEMBER_REC: + ret = mlx4_ib_mcg_multiplex_handler(ibdev, port, slave, sa_mad); + break; + default: + break; + } + return ret; } static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave) @@ -1590,6 +1626,7 @@ static int mlx4_ib_tunnels_update(struct mlx4_ib_dev *dev, int slave, int ret = 0; if (!do_init) { + clean_vf_mcast(&dev->sriov.demux[port - 1], slave); /* for master, destroy real sqp resources */ if (slave == mlx4_master_func_num(dev->dev)) destroy_pv_resources(dev, slave, port, @@ -1643,10 +1680,16 @@ static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev, ret = alloc_pv_object(dev, i, port, &ctx->tun[i]); if (ret) { ret = -ENOMEM; - goto err_wq; + goto err_mcg; } } + ret = mlx4_ib_mcg_port_init(ctx); + if (ret) { + pr_err("Failed initializing mcg para-virt (%d)\n", ret); + goto err_mcg; + } + snprintf(name, sizeof name, "mlx4_ibt%d", port); ctx->wq = create_singlethread_workqueue(name); if (!ctx->wq) { @@ -1670,6 +1713,8 @@ err_udwq: ctx->wq = NULL; err_wq: + mlx4_ib_mcg_port_cleanup(ctx, 1); +err_mcg: for (i = 0; i < dev->dev->caps.sqp_demux; i++) free_pv_object(dev, i, port); kfree(ctx->tun); @@ -1705,6 +1750,7 @@ static void mlx4_ib_free_demux_ctx(struct mlx4_ib_demux_ctx *ctx) int i; if (ctx) { struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev); + mlx4_ib_mcg_port_cleanup(ctx, 1); for (i = 0; i < dev->dev->caps.sqp_demux; i++) { if (!ctx->tun[i]) continue; diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index 45a6cc04036b..b959fe4665dd 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -1628,18 +1628,28 @@ static int __init mlx4_ib_init(void) if (!wq) return -ENOMEM; + err = mlx4_ib_mcg_init(); + if (err) + goto clean_wq; + err = mlx4_register_interface(&mlx4_ib_interface); - if (err) { - destroy_workqueue(wq); - return err; - } + if (err) + goto clean_mcg; return 0; + +clean_mcg: + mlx4_ib_mcg_destroy(); + +clean_wq: + destroy_workqueue(wq); + return err; } static void __exit mlx4_ib_cleanup(void) { mlx4_unregister_interface(&mlx4_ib_interface); + mlx4_ib_mcg_destroy(); destroy_workqueue(wq); } diff --git a/drivers/infiniband/hw/mlx4/mcg.c b/drivers/infiniband/hw/mlx4/mcg.c new file mode 100644 index 000000000000..1ee2e3a3347a --- /dev/null +++ b/drivers/infiniband/hw/mlx4/mcg.c @@ -0,0 +1,1187 @@ +/* + * Copyright (c) 2012 Mellanox Technologies. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include +#include +#include +#include + +#include +#include +#include + +#include "mlx4_ib.h" + +#define MAX_VFS 80 +#define MAX_PEND_REQS_PER_FUNC 4 +#define MAD_TIMEOUT_MS 2000 + +#define mcg_warn(fmt, arg...) pr_warn("MCG WARNING: " fmt, ##arg) +#define mcg_error(fmt, arg...) pr_err(fmt, ##arg) +#define mcg_warn_group(group, format, arg...) \ + pr_warn("%s-%d: %16s (port %d): WARNING: " format, __func__, __LINE__,\ + (group)->name, group->demux->port, ## arg) + +#define mcg_error_group(group, format, arg...) \ + pr_err(" %16s: " format, (group)->name, ## arg) + + +static union ib_gid mgid0; + +static struct workqueue_struct *clean_wq; + +enum mcast_state { + MCAST_NOT_MEMBER = 0, + MCAST_MEMBER, +}; + +enum mcast_group_state { + MCAST_IDLE, + MCAST_JOIN_SENT, + MCAST_LEAVE_SENT, + MCAST_RESP_READY +}; + +struct mcast_member { + enum mcast_state state; + uint8_t join_state; + int num_pend_reqs; + struct list_head pending; +}; + +struct ib_sa_mcmember_data { + union ib_gid mgid; + union ib_gid port_gid; + __be32 qkey; + __be16 mlid; + u8 mtusel_mtu; + u8 tclass; + __be16 pkey; + u8 ratesel_rate; + u8 lifetmsel_lifetm; + __be32 sl_flowlabel_hoplimit; + u8 scope_join_state; + u8 proxy_join; + u8 reserved[2]; +}; + +struct mcast_group { + struct ib_sa_mcmember_data rec; + struct rb_node node; + struct list_head mgid0_list; + struct mlx4_ib_demux_ctx *demux; + struct mcast_member func[MAX_VFS]; + struct mutex lock; + struct work_struct work; + struct list_head pending_list; + int members[3]; + enum mcast_group_state state; + enum mcast_group_state prev_state; + struct ib_sa_mad response_sa_mad; + __be64 last_req_tid; + + char name[33]; /* MGID string */ + + /* refcount is the reference count for the following: + 1. Each queued request + 2. Each invocation of the worker thread + 3. Membership of the port at the SA + */ + atomic_t refcount; + + /* delayed work to clean pending SM request */ + struct delayed_work timeout_work; + struct list_head cleanup_list; +}; + +struct mcast_req { + int func; + struct ib_sa_mad sa_mad; + struct list_head group_list; + struct list_head func_list; + struct mcast_group *group; + int clean; +}; + + +#define safe_atomic_dec(ref) \ + do {\ + if (atomic_dec_and_test(ref)) \ + mcg_warn_group(group, "did not expect to reach zero\n"); \ + } while (0) + +static const char *get_state_string(enum mcast_group_state state) +{ + switch (state) { + case MCAST_IDLE: + return "MCAST_IDLE"; + case MCAST_JOIN_SENT: + return "MCAST_JOIN_SENT"; + case MCAST_LEAVE_SENT: + return "MCAST_LEAVE_SENT"; + case MCAST_RESP_READY: + return "MCAST_RESP_READY"; + } + return "Invalid State"; +} + +static struct mcast_group *mcast_find(struct mlx4_ib_demux_ctx *ctx, + union ib_gid *mgid) +{ + struct rb_node *node = ctx->mcg_table.rb_node; + struct mcast_group *group; + int ret; + + while (node) { + group = rb_entry(node, struct mcast_group, node); + ret = memcmp(mgid->raw, group->rec.mgid.raw, sizeof *mgid); + if (!ret) + return group; + + if (ret < 0) + node = node->rb_left; + else + node = node->rb_right; + } + return NULL; +} + +static struct mcast_group *mcast_insert(struct mlx4_ib_demux_ctx *ctx, + struct mcast_group *group) +{ + struct rb_node **link = &ctx->mcg_table.rb_node; + struct rb_node *parent = NULL; + struct mcast_group *cur_group; + int ret; + + while (*link) { + parent = *link; + cur_group = rb_entry(parent, struct mcast_group, node); + + ret = memcmp(group->rec.mgid.raw, cur_group->rec.mgid.raw, + sizeof group->rec.mgid); + if (ret < 0) + link = &(*link)->rb_left; + else if (ret > 0) + link = &(*link)->rb_right; + else + return cur_group; + } + rb_link_node(&group->node, parent, link); + rb_insert_color(&group->node, &ctx->mcg_table); + return NULL; +} + +static int send_mad_to_wire(struct mlx4_ib_demux_ctx *ctx, struct ib_mad *mad) +{ + struct mlx4_ib_dev *dev = ctx->dev; + struct ib_ah_attr ah_attr; + + spin_lock(&dev->sm_lock); + if (!dev->sm_ah[ctx->port - 1]) { + /* port is not yet Active, sm_ah not ready */ + spin_unlock(&dev->sm_lock); + return -EAGAIN; + } + mlx4_ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr); + spin_unlock(&dev->sm_lock); + return mlx4_ib_send_to_wire(dev, mlx4_master_func_num(dev->dev), ctx->port, + IB_QPT_GSI, 0, 1, IB_QP1_QKEY, &ah_attr, mad); +} + +static int send_mad_to_slave(int slave, struct mlx4_ib_demux_ctx *ctx, + struct ib_mad *mad) +{ + struct mlx4_ib_dev *dev = ctx->dev; + struct ib_mad_agent *agent = dev->send_agent[ctx->port - 1][1]; + struct ib_wc wc; + struct ib_ah_attr ah_attr; + + /* Our agent might not yet be registered when mads start to arrive */ + if (!agent) + return -EAGAIN; + + ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr); + + wc.pkey_index = 0; + wc.sl = 0; + wc.dlid_path_bits = 0; + wc.port_num = ctx->port; + wc.slid = ah_attr.dlid; /* opensm lid */ + wc.src_qp = 1; + return mlx4_ib_send_to_slave(dev, slave, ctx->port, IB_QPT_GSI, &wc, NULL, mad); +} + +static int send_join_to_wire(struct mcast_group *group, struct ib_sa_mad *sa_mad) +{ + struct ib_sa_mad mad; + struct ib_sa_mcmember_data *sa_mad_data = (struct ib_sa_mcmember_data *)&mad.data; + int ret; + + /* we rely on a mad request as arrived from a VF */ + memcpy(&mad, sa_mad, sizeof mad); + + /* fix port GID to be the real one (slave 0) */ + sa_mad_data->port_gid.global.interface_id = group->demux->guid_cache[0]; + + /* assign our own TID */ + mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux); + group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */ + + ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad); + /* set timeout handler */ + if (!ret) { + /* calls mlx4_ib_mcg_timeout_handler */ + queue_delayed_work(group->demux->mcg_wq, &group->timeout_work, + msecs_to_jiffies(MAD_TIMEOUT_MS)); + } + + return ret; +} + +static int send_leave_to_wire(struct mcast_group *group, u8 join_state) +{ + struct ib_sa_mad mad; + struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data; + int ret; + + memset(&mad, 0, sizeof mad); + mad.mad_hdr.base_version = 1; + mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM; + mad.mad_hdr.class_version = 2; + mad.mad_hdr.method = IB_SA_METHOD_DELETE; + mad.mad_hdr.status = cpu_to_be16(0); + mad.mad_hdr.class_specific = cpu_to_be16(0); + mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux); + group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */ + mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC); + mad.mad_hdr.attr_mod = cpu_to_be32(0); + mad.sa_hdr.sm_key = 0x0; + mad.sa_hdr.attr_offset = cpu_to_be16(7); + mad.sa_hdr.comp_mask = IB_SA_MCMEMBER_REC_MGID | + IB_SA_MCMEMBER_REC_PORT_GID | IB_SA_MCMEMBER_REC_JOIN_STATE; + + *sa_data = group->rec; + sa_data->scope_join_state = join_state; + + ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad); + if (ret) + group->state = MCAST_IDLE; + + /* set timeout handler */ + if (!ret) { + /* calls mlx4_ib_mcg_timeout_handler */ + queue_delayed_work(group->demux->mcg_wq, &group->timeout_work, + msecs_to_jiffies(MAD_TIMEOUT_MS)); + } + + return ret; +} + +static int send_reply_to_slave(int slave, struct mcast_group *group, + struct ib_sa_mad *req_sa_mad, u16 status) +{ + struct ib_sa_mad mad; + struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data; + struct ib_sa_mcmember_data *req_sa_data = (struct ib_sa_mcmember_data *)&req_sa_mad->data; + int ret; + + memset(&mad, 0, sizeof mad); + mad.mad_hdr.base_version = 1; + mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM; + mad.mad_hdr.class_version = 2; + mad.mad_hdr.method = IB_MGMT_METHOD_GET_RESP; + mad.mad_hdr.status = cpu_to_be16(status); + mad.mad_hdr.class_specific = cpu_to_be16(0); + mad.mad_hdr.tid = req_sa_mad->mad_hdr.tid; + *(u8 *)&mad.mad_hdr.tid = 0; /* resetting tid to 0 */ + mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC); + mad.mad_hdr.attr_mod = cpu_to_be32(0); + mad.sa_hdr.sm_key = req_sa_mad->sa_hdr.sm_key; + mad.sa_hdr.attr_offset = cpu_to_be16(7); + mad.sa_hdr.comp_mask = 0; /* ignored on responses, see IBTA spec */ + + *sa_data = group->rec; + + /* reconstruct VF's requested join_state and port_gid */ + sa_data->scope_join_state &= 0xf0; + sa_data->scope_join_state |= (group->func[slave].join_state & 0x0f); + memcpy(&sa_data->port_gid, &req_sa_data->port_gid, sizeof req_sa_data->port_gid); + + ret = send_mad_to_slave(slave, group->demux, (struct ib_mad *)&mad); + return ret; +} + +static int check_selector(ib_sa_comp_mask comp_mask, + ib_sa_comp_mask selector_mask, + ib_sa_comp_mask value_mask, + u8 src_value, u8 dst_value) +{ + int err; + u8 selector = dst_value >> 6; + dst_value &= 0x3f; + src_value &= 0x3f; + + if (!(comp_mask & selector_mask) || !(comp_mask & value_mask)) + return 0; + + switch (selector) { + case IB_SA_GT: + err = (src_value <= dst_value); + break; + case IB_SA_LT: + err = (src_value >= dst_value); + break; + case IB_SA_EQ: + err = (src_value != dst_value); + break; + default: + err = 0; + break; + } + + return err; +} + +static u16 cmp_rec(struct ib_sa_mcmember_data *src, + struct ib_sa_mcmember_data *dst, ib_sa_comp_mask comp_mask) +{ + /* src is group record, dst is request record */ + /* MGID must already match */ + /* Port_GID we always replace to our Port_GID, so it is a match */ + +#define MAD_STATUS_REQ_INVALID 0x0200 + if (comp_mask & IB_SA_MCMEMBER_REC_QKEY && src->qkey != dst->qkey) + return MAD_STATUS_REQ_INVALID; + if (comp_mask & IB_SA_MCMEMBER_REC_MLID && src->mlid != dst->mlid) + return MAD_STATUS_REQ_INVALID; + if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_MTU_SELECTOR, + IB_SA_MCMEMBER_REC_MTU, + src->mtusel_mtu, dst->mtusel_mtu)) + return MAD_STATUS_REQ_INVALID; + if (comp_mask & IB_SA_MCMEMBER_REC_TRAFFIC_CLASS && + src->tclass != dst->tclass) + return MAD_STATUS_REQ_INVALID; + if (comp_mask & IB_SA_MCMEMBER_REC_PKEY && src->pkey != dst->pkey) + return MAD_STATUS_REQ_INVALID; + if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_RATE_SELECTOR, + IB_SA_MCMEMBER_REC_RATE, + src->ratesel_rate, dst->ratesel_rate)) + return MAD_STATUS_REQ_INVALID; + if (check_selector(comp_mask, + IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME_SELECTOR, + IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME, + src->lifetmsel_lifetm, dst->lifetmsel_lifetm)) + return MAD_STATUS_REQ_INVALID; + if (comp_mask & IB_SA_MCMEMBER_REC_SL && + (be32_to_cpu(src->sl_flowlabel_hoplimit) & 0xf0000000) != + (be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0xf0000000)) + return MAD_STATUS_REQ_INVALID; + if (comp_mask & IB_SA_MCMEMBER_REC_FLOW_LABEL && + (be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x0fffff00) != + (be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x0fffff00)) + return MAD_STATUS_REQ_INVALID; + if (comp_mask & IB_SA_MCMEMBER_REC_HOP_LIMIT && + (be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x000000ff) != + (be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x000000ff)) + return MAD_STATUS_REQ_INVALID; + if (comp_mask & IB_SA_MCMEMBER_REC_SCOPE && + (src->scope_join_state & 0xf0) != + (dst->scope_join_state & 0xf0)) + return MAD_STATUS_REQ_INVALID; + + /* join_state checked separately, proxy_join ignored */ + + return 0; +} + +/* release group, return 1 if this was last release and group is destroyed + * timout work is canceled sync */ +static int release_group(struct mcast_group *group, int from_timeout_handler) +{ + struct mlx4_ib_demux_ctx *ctx = group->demux; + int nzgroup; + + mutex_lock(&ctx->mcg_table_lock); + mutex_lock(&group->lock); + if (atomic_dec_and_test(&group->refcount)) { + if (!from_timeout_handler) { + if (group->state != MCAST_IDLE && + !cancel_delayed_work(&group->timeout_work)) { + atomic_inc(&group->refcount); + mutex_unlock(&group->lock); + mutex_unlock(&ctx->mcg_table_lock); + return 0; + } + } + + nzgroup = memcmp(&group->rec.mgid, &mgid0, sizeof mgid0); + if (!list_empty(&group->pending_list)) + mcg_warn_group(group, "releasing a group with non empty pending list\n"); + if (nzgroup) + rb_erase(&group->node, &ctx->mcg_table); + list_del_init(&group->mgid0_list); + mutex_unlock(&group->lock); + mutex_unlock(&ctx->mcg_table_lock); + kfree(group); + return 1; + } else { + mutex_unlock(&group->lock); + mutex_unlock(&ctx->mcg_table_lock); + } + return 0; +} + +static void adjust_membership(struct mcast_group *group, u8 join_state, int inc) +{ + int i; + + for (i = 0; i < 3; i++, join_state >>= 1) + if (join_state & 0x1) + group->members[i] += inc; +} + +static u8 get_leave_state(struct mcast_group *group) +{ + u8 leave_state = 0; + int i; + + for (i = 0; i < 3; i++) + if (!group->members[i]) + leave_state |= (1 << i); + + return leave_state & (group->rec.scope_join_state & 7); +} + +static int join_group(struct mcast_group *group, int slave, u8 join_mask) +{ + int ret = 0; + u8 join_state; + + /* remove bits that slave is already member of, and adjust */ + join_state = join_mask & (~group->func[slave].join_state); + adjust_membership(group, join_state, 1); + group->func[slave].join_state |= join_state; + if (group->func[slave].state != MCAST_MEMBER && join_state) { + group->func[slave].state = MCAST_MEMBER; + ret = 1; + } + return ret; +} + +static int leave_group(struct mcast_group *group, int slave, u8 leave_state) +{ + int ret = 0; + + adjust_membership(group, leave_state, -1); + group->func[slave].join_state &= ~leave_state; + if (!group->func[slave].join_state) { + group->func[slave].state = MCAST_NOT_MEMBER; + ret = 1; + } + return ret; +} + +static int check_leave(struct mcast_group *group, int slave, u8 leave_mask) +{ + if (group->func[slave].state != MCAST_MEMBER) + return MAD_STATUS_REQ_INVALID; + + /* make sure we're not deleting unset bits */ + if (~group->func[slave].join_state & leave_mask) + return MAD_STATUS_REQ_INVALID; + + if (!leave_mask) + return MAD_STATUS_REQ_INVALID; + + return 0; +} + +static void mlx4_ib_mcg_timeout_handler(struct work_struct *work) +{ + struct delayed_work *delay = to_delayed_work(work); + struct mcast_group *group; + struct mcast_req *req = NULL; + + group = container_of(delay, typeof(*group), timeout_work); + + mutex_lock(&group->lock); + if (group->state == MCAST_JOIN_SENT) { + if (!list_empty(&group->pending_list)) { + req = list_first_entry(&group->pending_list, struct mcast_req, group_list); + list_del(&req->group_list); + list_del(&req->func_list); + --group->func[req->func].num_pend_reqs; + mutex_unlock(&group->lock); + kfree(req); + if (memcmp(&group->rec.mgid, &mgid0, sizeof mgid0)) { + if (release_group(group, 1)) + return; + } else { + kfree(group); + return; + } + mutex_lock(&group->lock); + } else + mcg_warn_group(group, "DRIVER BUG\n"); + } else if (group->state == MCAST_LEAVE_SENT) { + if (group->rec.scope_join_state & 7) + group->rec.scope_join_state &= 0xf8; + group->state = MCAST_IDLE; + mutex_unlock(&group->lock); + if (release_group(group, 1)) + return; + mutex_lock(&group->lock); + } else + mcg_warn_group(group, "invalid state %s\n", get_state_string(group->state)); + group->state = MCAST_IDLE; + atomic_inc(&group->refcount); + if (!queue_work(group->demux->mcg_wq, &group->work)) + safe_atomic_dec(&group->refcount); + + mutex_unlock(&group->lock); +} + +static int handle_leave_req(struct mcast_group *group, u8 leave_mask, + struct mcast_req *req) +{ + u16 status; + + if (req->clean) + leave_mask = group->func[req->func].join_state; + + status = check_leave(group, req->func, leave_mask); + if (!status) + leave_group(group, req->func, leave_mask); + + if (!req->clean) + send_reply_to_slave(req->func, group, &req->sa_mad, status); + --group->func[req->func].num_pend_reqs; + list_del(&req->group_list); + list_del(&req->func_list); + kfree(req); + return 1; +} + +static int handle_join_req(struct mcast_group *group, u8 join_mask, + struct mcast_req *req) +{ + u8 group_join_state = group->rec.scope_join_state & 7; + int ref = 0; + u16 status; + struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data; + + if (join_mask == (group_join_state & join_mask)) { + /* port's membership need not change */ + status = cmp_rec(&group->rec, sa_data, req->sa_mad.sa_hdr.comp_mask); + if (!status) + join_group(group, req->func, join_mask); + + --group->func[req->func].num_pend_reqs; + send_reply_to_slave(req->func, group, &req->sa_mad, status); + list_del(&req->group_list); + list_del(&req->func_list); + kfree(req); + ++ref; + } else { + /* port's membership needs to be updated */ + group->prev_state = group->state; + if (send_join_to_wire(group, &req->sa_mad)) { + --group->func[req->func].num_pend_reqs; + list_del(&req->group_list); + list_del(&req->func_list); + kfree(req); + ref = 1; + group->state = group->prev_state; + } else + group->state = MCAST_JOIN_SENT; + } + + return ref; +} + +static void mlx4_ib_mcg_work_handler(struct work_struct *work) +{ + struct mcast_group *group; + struct mcast_req *req = NULL; + struct ib_sa_mcmember_data *sa_data; + u8 req_join_state; + int rc = 1; /* release_count - this is for the scheduled work */ + u16 status; + u8 method; + + group = container_of(work, typeof(*group), work); + + mutex_lock(&group->lock); + + /* First, let's see if a response from SM is waiting regarding this group. + * If so, we need to update the group's REC. If this is a bad response, we + * may need to send a bad response to a VF waiting for it. If VF is waiting + * and this is a good response, the VF will be answered later in this func. */ + if (group->state == MCAST_RESP_READY) { + /* cancels mlx4_ib_mcg_timeout_handler */ + cancel_delayed_work(&group->timeout_work); + status = be16_to_cpu(group->response_sa_mad.mad_hdr.status); + method = group->response_sa_mad.mad_hdr.method; + if (group->last_req_tid != group->response_sa_mad.mad_hdr.tid) { + mcg_warn_group(group, "Got MAD response to existing MGID but wrong TID, dropping. Resp TID=%llx, group TID=%llx\n", + be64_to_cpu(group->response_sa_mad.mad_hdr.tid), + be64_to_cpu(group->last_req_tid)); + group->state = group->prev_state; + goto process_requests; + } + if (status) { + if (!list_empty(&group->pending_list)) + req = list_first_entry(&group->pending_list, + struct mcast_req, group_list); + if ((method == IB_MGMT_METHOD_GET_RESP)) { + if (req) { + send_reply_to_slave(req->func, group, &req->sa_mad, status); + --group->func[req->func].num_pend_reqs; + list_del(&req->group_list); + list_del(&req->func_list); + kfree(req); + ++rc; + } else + mcg_warn_group(group, "no request for failed join\n"); + } else if (method == IB_SA_METHOD_DELETE_RESP && group->demux->flushing) + ++rc; + } else { + u8 resp_join_state; + u8 cur_join_state; + + resp_join_state = ((struct ib_sa_mcmember_data *) + group->response_sa_mad.data)->scope_join_state & 7; + cur_join_state = group->rec.scope_join_state & 7; + + if (method == IB_MGMT_METHOD_GET_RESP) { + /* successfull join */ + if (!cur_join_state && resp_join_state) + --rc; + } else if (!resp_join_state) + ++rc; + memcpy(&group->rec, group->response_sa_mad.data, sizeof group->rec); + } + group->state = MCAST_IDLE; + } + +process_requests: + /* We should now go over pending join/leave requests, as long as we are idle. */ + while (!list_empty(&group->pending_list) && group->state == MCAST_IDLE) { + req = list_first_entry(&group->pending_list, struct mcast_req, + group_list); + sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data; + req_join_state = sa_data->scope_join_state & 0x7; + + /* For a leave request, we will immediately answer the VF, and + * update our internal counters. The actual leave will be sent + * to SM later, if at all needed. We dequeue the request now. */ + if (req->sa_mad.mad_hdr.method == IB_SA_METHOD_DELETE) + rc += handle_leave_req(group, req_join_state, req); + else + rc += handle_join_req(group, req_join_state, req); + } + + /* Handle leaves */ + if (group->state == MCAST_IDLE) { + req_join_state = get_leave_state(group); + if (req_join_state) { + group->rec.scope_join_state &= ~req_join_state; + group->prev_state = group->state; + if (send_leave_to_wire(group, req_join_state)) { + group->state = group->prev_state; + ++rc; + } else + group->state = MCAST_LEAVE_SENT; + } + } + + if (!list_empty(&group->pending_list) && group->state == MCAST_IDLE) + goto process_requests; + mutex_unlock(&group->lock); + + while (rc--) + release_group(group, 0); +} + +static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx *ctx, + __be64 tid, + union ib_gid *new_mgid) +{ + struct mcast_group *group = NULL, *cur_group; + struct mcast_req *req; + struct list_head *pos; + struct list_head *n; + + mutex_lock(&ctx->mcg_table_lock); + list_for_each_safe(pos, n, &ctx->mcg_mgid0_list) { + group = list_entry(pos, struct mcast_group, mgid0_list); + mutex_lock(&group->lock); + if (group->last_req_tid == tid) { + if (memcmp(new_mgid, &mgid0, sizeof mgid0)) { + group->rec.mgid = *new_mgid; + sprintf(group->name, "%016llx%016llx", + be64_to_cpu(group->rec.mgid.global.subnet_prefix), + be64_to_cpu(group->rec.mgid.global.interface_id)); + list_del_init(&group->mgid0_list); + cur_group = mcast_insert(ctx, group); + if (cur_group) { + /* A race between our code and SM. Silently cleaning the new one */ + req = list_first_entry(&group->pending_list, + struct mcast_req, group_list); + --group->func[req->func].num_pend_reqs; + list_del(&req->group_list); + list_del(&req->func_list); + kfree(req); + mutex_unlock(&group->lock); + mutex_unlock(&ctx->mcg_table_lock); + release_group(group, 0); + return NULL; + } + + atomic_inc(&group->refcount); + mutex_unlock(&group->lock); + mutex_unlock(&ctx->mcg_table_lock); + return group; + } else { + struct mcast_req *tmp1, *tmp2; + + list_del(&group->mgid0_list); + if (!list_empty(&group->pending_list) && group->state != MCAST_IDLE) + cancel_delayed_work_sync(&group->timeout_work); + + list_for_each_entry_safe(tmp1, tmp2, &group->pending_list, group_list) { + list_del(&tmp1->group_list); + kfree(tmp1); + } + mutex_unlock(&group->lock); + mutex_unlock(&ctx->mcg_table_lock); + kfree(group); + return NULL; + } + } + mutex_unlock(&group->lock); + } + mutex_unlock(&ctx->mcg_table_lock); + + return NULL; +} + +static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx, + union ib_gid *mgid, int create, + gfp_t gfp_mask) +{ + struct mcast_group *group, *cur_group; + int is_mgid0; + int i; + + is_mgid0 = !memcmp(&mgid0, mgid, sizeof mgid0); + if (!is_mgid0) { + group = mcast_find(ctx, mgid); + if (group) + goto found; + } + + if (!create) + return ERR_PTR(-ENOENT); + + group = kzalloc(sizeof *group, gfp_mask); + if (!group) + return ERR_PTR(-ENOMEM); + + group->demux = ctx; + group->rec.mgid = *mgid; + INIT_LIST_HEAD(&group->pending_list); + INIT_LIST_HEAD(&group->mgid0_list); + for (i = 0; i < MAX_VFS; ++i) + INIT_LIST_HEAD(&group->func[i].pending); + INIT_WORK(&group->work, mlx4_ib_mcg_work_handler); + INIT_DELAYED_WORK(&group->timeout_work, mlx4_ib_mcg_timeout_handler); + mutex_init(&group->lock); + sprintf(group->name, "%016llx%016llx", + be64_to_cpu(group->rec.mgid.global.subnet_prefix), + be64_to_cpu(group->rec.mgid.global.interface_id)); + group->state = MCAST_IDLE; + + if (is_mgid0) { + list_add(&group->mgid0_list, &ctx->mcg_mgid0_list); + goto found; + } + + cur_group = mcast_insert(ctx, group); + if (cur_group) { + mcg_warn("group just showed up %s - confused\n", cur_group->name); + kfree(group); + return ERR_PTR(-EINVAL); + } + +found: + atomic_inc(&group->refcount); + return group; +} + +static void queue_req(struct mcast_req *req) +{ + struct mcast_group *group = req->group; + + atomic_inc(&group->refcount); /* for the request */ + atomic_inc(&group->refcount); /* for scheduling the work */ + list_add_tail(&req->group_list, &group->pending_list); + list_add_tail(&req->func_list, &group->func[req->func].pending); + /* calls mlx4_ib_mcg_work_handler */ + if (!queue_work(group->demux->mcg_wq, &group->work)) + safe_atomic_dec(&group->refcount); +} + +int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave, + struct ib_sa_mad *mad) +{ + struct mlx4_ib_dev *dev = to_mdev(ibdev); + struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)mad->data; + struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1]; + struct mcast_group *group; + + switch (mad->mad_hdr.method) { + case IB_MGMT_METHOD_GET_RESP: + case IB_SA_METHOD_DELETE_RESP: + mutex_lock(&ctx->mcg_table_lock); + group = acquire_group(ctx, &rec->mgid, 0, GFP_KERNEL); + mutex_unlock(&ctx->mcg_table_lock); + if (IS_ERR(group)) { + if (mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP) { + __be64 tid = mad->mad_hdr.tid; + *(u8 *)(&tid) = (u8)slave; /* in group we kept the modified TID */ + group = search_relocate_mgid0_group(ctx, tid, &rec->mgid); + } else + group = NULL; + } + + if (!group) + return 1; + + mutex_lock(&group->lock); + group->response_sa_mad = *mad; + group->prev_state = group->state; + group->state = MCAST_RESP_READY; + /* calls mlx4_ib_mcg_work_handler */ + atomic_inc(&group->refcount); + if (!queue_work(ctx->mcg_wq, &group->work)) + safe_atomic_dec(&group->refcount); + mutex_unlock(&group->lock); + release_group(group, 0); + return 1; /* consumed */ + case IB_MGMT_METHOD_SET: + case IB_SA_METHOD_GET_TABLE: + case IB_SA_METHOD_GET_TABLE_RESP: + case IB_SA_METHOD_DELETE: + return 0; /* not consumed, pass-through to guest over tunnel */ + default: + mcg_warn("In demux, port %d: unexpected MCMember method: 0x%x, dropping\n", + port, mad->mad_hdr.method); + return 1; /* consumed */ + } +} + +int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port, + int slave, struct ib_sa_mad *sa_mad) +{ + struct mlx4_ib_dev *dev = to_mdev(ibdev); + struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)sa_mad->data; + struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1]; + struct mcast_group *group; + struct mcast_req *req; + int may_create = 0; + + if (ctx->flushing) + return -EAGAIN; + + switch (sa_mad->mad_hdr.method) { + case IB_MGMT_METHOD_SET: + may_create = 1; + case IB_SA_METHOD_DELETE: + req = kzalloc(sizeof *req, GFP_KERNEL); + if (!req) + return -ENOMEM; + + req->func = slave; + req->sa_mad = *sa_mad; + + mutex_lock(&ctx->mcg_table_lock); + group = acquire_group(ctx, &rec->mgid, may_create, GFP_KERNEL); + mutex_unlock(&ctx->mcg_table_lock); + if (IS_ERR(group)) { + kfree(req); + return PTR_ERR(group); + } + mutex_lock(&group->lock); + if (group->func[slave].num_pend_reqs > MAX_PEND_REQS_PER_FUNC) { + mutex_unlock(&group->lock); + mcg_warn_group(group, "Port %d, Func %d has too many pending requests (%d), dropping\n", + port, slave, MAX_PEND_REQS_PER_FUNC); + release_group(group, 0); + kfree(req); + return -ENOMEM; + } + ++group->func[slave].num_pend_reqs; + req->group = group; + queue_req(req); + mutex_unlock(&group->lock); + release_group(group, 0); + return 1; /* consumed */ + case IB_SA_METHOD_GET_TABLE: + case IB_MGMT_METHOD_GET_RESP: + case IB_SA_METHOD_GET_TABLE_RESP: + case IB_SA_METHOD_DELETE_RESP: + return 0; /* not consumed, pass-through */ + default: + mcg_warn("In multiplex, port %d, func %d: unexpected MCMember method: 0x%x, dropping\n", + port, slave, sa_mad->mad_hdr.method); + return 1; /* consumed */ + } +} + +int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx) +{ + char name[20]; + + atomic_set(&ctx->tid, 0); + sprintf(name, "mlx4_ib_mcg%d", ctx->port); + ctx->mcg_wq = create_singlethread_workqueue(name); + if (!ctx->mcg_wq) + return -ENOMEM; + + mutex_init(&ctx->mcg_table_lock); + ctx->mcg_table = RB_ROOT; + INIT_LIST_HEAD(&ctx->mcg_mgid0_list); + ctx->flushing = 0; + + return 0; +} + +static void force_clean_group(struct mcast_group *group) +{ + struct mcast_req *req, *tmp + ; + list_for_each_entry_safe(req, tmp, &group->pending_list, group_list) { + list_del(&req->group_list); + kfree(req); + } + rb_erase(&group->node, &group->demux->mcg_table); + kfree(group); +} + +static void _mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq) +{ + int i; + struct rb_node *p; + struct mcast_group *group; + unsigned long end; + int count; + + if (ctx->flushing) + return; + + ctx->flushing = 1; + for (i = 0; i < MAX_VFS; ++i) + clean_vf_mcast(ctx, i); + + end = jiffies + msecs_to_jiffies(MAD_TIMEOUT_MS + 3000); + do { + count = 0; + mutex_lock(&ctx->mcg_table_lock); + for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p)) + ++count; + mutex_unlock(&ctx->mcg_table_lock); + if (!count) + break; + + msleep(1); + } while (time_after(end, jiffies)); + + flush_workqueue(ctx->mcg_wq); + if (destroy_wq) + destroy_workqueue(ctx->mcg_wq); + + mutex_lock(&ctx->mcg_table_lock); + while ((p = rb_first(&ctx->mcg_table)) != NULL) { + group = rb_entry(p, struct mcast_group, node); + if (atomic_read(&group->refcount)) + mcg_warn_group(group, "group refcount %d!!! (pointer %p)\n", atomic_read(&group->refcount), group); + + force_clean_group(group); + } + mutex_unlock(&ctx->mcg_table_lock); + + if (!destroy_wq) + ctx->flushing = 0; +} + +struct clean_work { + struct work_struct work; + struct mlx4_ib_demux_ctx *ctx; + int destroy_wq; +}; + +static void mcg_clean_task(struct work_struct *work) +{ + struct clean_work *cw = container_of(work, struct clean_work, work); + + _mlx4_ib_mcg_port_cleanup(cw->ctx, cw->destroy_wq); + kfree(cw); +} + +void mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq) +{ + struct clean_work *work; + + if (destroy_wq) { + _mlx4_ib_mcg_port_cleanup(ctx, destroy_wq); + return; + } + + work = kmalloc(sizeof *work, GFP_KERNEL); + if (!work) { + mcg_warn("failed allocating work for cleanup\n"); + return; + } + + work->ctx = ctx; + work->destroy_wq = destroy_wq; + INIT_WORK(&work->work, mcg_clean_task); + queue_work(clean_wq, &work->work); +} + +static void build_leave_mad(struct mcast_req *req) +{ + struct ib_sa_mad *mad = &req->sa_mad; + + mad->mad_hdr.method = IB_SA_METHOD_DELETE; +} + + +static void clear_pending_reqs(struct mcast_group *group, int vf) +{ + struct mcast_req *req, *tmp, *group_first = NULL; + int clear; + int pend = 0; + + if (!list_empty(&group->pending_list)) + group_first = list_first_entry(&group->pending_list, struct mcast_req, group_list); + + list_for_each_entry_safe(req, tmp, &group->func[vf].pending, func_list) { + clear = 1; + if (group_first == req && + (group->state == MCAST_JOIN_SENT || + group->state == MCAST_LEAVE_SENT)) { + clear = cancel_delayed_work(&group->timeout_work); + pend = !clear; + group->state = MCAST_IDLE; + } + if (clear) { + --group->func[vf].num_pend_reqs; + list_del(&req->group_list); + list_del(&req->func_list); + kfree(req); + atomic_dec(&group->refcount); + } + } + + if (!pend && (!list_empty(&group->func[vf].pending) || group->func[vf].num_pend_reqs)) { + mcg_warn_group(group, "DRIVER BUG: list_empty %d, num_pend_reqs %d\n", + list_empty(&group->func[vf].pending), group->func[vf].num_pend_reqs); + } +} + +static int push_deleteing_req(struct mcast_group *group, int slave) +{ + struct mcast_req *req; + struct mcast_req *pend_req; + + if (!group->func[slave].join_state) + return 0; + + req = kzalloc(sizeof *req, GFP_KERNEL); + if (!req) { + mcg_warn_group(group, "failed allocation - may leave stall groups\n"); + return -ENOMEM; + } + + if (!list_empty(&group->func[slave].pending)) { + pend_req = list_entry(group->func[slave].pending.prev, struct mcast_req, group_list); + if (pend_req->clean) { + kfree(req); + return 0; + } + } + + req->clean = 1; + req->func = slave; + req->group = group; + ++group->func[slave].num_pend_reqs; + build_leave_mad(req); + queue_req(req); + return 0; +} + +void clean_vf_mcast(struct mlx4_ib_demux_ctx *ctx, int slave) +{ + struct mcast_group *group; + struct rb_node *p; + + mutex_lock(&ctx->mcg_table_lock); + for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p)) { + group = rb_entry(p, struct mcast_group, node); + mutex_lock(&group->lock); + if (atomic_read(&group->refcount)) { + /* clear pending requests of this VF */ + clear_pending_reqs(group, slave); + push_deleteing_req(group, slave); + } + mutex_unlock(&group->lock); + } + mutex_unlock(&ctx->mcg_table_lock); +} + + +int mlx4_ib_mcg_init(void) +{ + clean_wq = create_singlethread_workqueue("mlx4_ib_mcg"); + if (!clean_wq) + return -ENOMEM; + + return 0; +} + +void mlx4_ib_mcg_destroy(void) +{ + destroy_workqueue(clean_wq); +} diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index ac71d56ffc7e..01ba9f1692b1 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -37,9 +37,11 @@ #include #include #include +#include #include #include +#include #include #include @@ -329,7 +331,14 @@ struct mlx4_ib_demux_ctx { __be64 subnet_prefix; __be64 guid_cache[128]; struct mlx4_ib_dev *dev; + /* the following lock protects both mcg_table and mcg_mgid0_list */ + struct mutex mcg_table_lock; + struct rb_root mcg_table; + struct list_head mcg_mgid0_list; + struct workqueue_struct *mcg_wq; struct mlx4_ib_demux_pv_ctx **tun; + atomic_t tid; + int flushing; /* flushing the work queue */ }; struct mlx4_ib_sriov { @@ -553,6 +562,19 @@ static inline int mlx4_ib_ah_grh_present(struct mlx4_ib_ah *ah) return !!(ah->av.ib.g_slid & 0x80); } +int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx); +void mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq); +void clean_vf_mcast(struct mlx4_ib_demux_ctx *ctx, int slave); +int mlx4_ib_mcg_init(void); +void mlx4_ib_mcg_destroy(void); + +int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid); + +int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port, int slave, + struct ib_sa_mad *sa_mad); +int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave, + struct ib_sa_mad *mad); + int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp, union ib_gid *gid); @@ -561,4 +583,12 @@ void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num, void mlx4_ib_tunnels_update_work(struct work_struct *work); +int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port, + enum ib_qp_type qpt, struct ib_wc *wc, + struct ib_grh *grh, struct ib_mad *mad); +int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port, + enum ib_qp_type dest_qpt, u16 pkey_index, u32 remote_qpn, + u32 qkey, struct ib_ah_attr *attr, struct ib_mad *mad); +__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx); + #endif /* MLX4_IB_H */ From 3cf69cc8dbebf15b99deb342ea422105ae9c2774 Mon Sep 17 00:00:00 2001 From: Amir Vadai Date: Fri, 3 Aug 2012 08:40:47 +0000 Subject: [PATCH 18/45] IB/mlx4: Add CM paravirtualization In CM para-virtualization: 1. Incoming requests are steered to the correct vHCA according to the embedded GID. 2. Communication IDs on outgoing requests are replaced by a globally unique ID, generated by the PPF, since there is no synchronization of ID generation between guests (and so these IDs are not guaranteed to be globally unique). The guest's comm ID is stored, and is returned to the response MAD when it arrives. Signed-off-by: Amir Vadai Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/Makefile | 2 +- drivers/infiniband/hw/mlx4/cm.c | 437 +++++++++++++++++++++++++++ drivers/infiniband/hw/mlx4/mad.c | 16 +- drivers/infiniband/hw/mlx4/mlx4_ib.h | 15 + 4 files changed, 468 insertions(+), 2 deletions(-) create mode 100644 drivers/infiniband/hw/mlx4/cm.c diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile index 20d627d1f046..bf0aa901c313 100644 --- a/drivers/infiniband/hw/mlx4/Makefile +++ b/drivers/infiniband/hw/mlx4/Makefile @@ -1,3 +1,3 @@ obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o -mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o +mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c new file mode 100644 index 000000000000..e25e4dafb8a8 --- /dev/null +++ b/drivers/infiniband/hw/mlx4/cm.c @@ -0,0 +1,437 @@ +/* + * Copyright (c) 2012 Mellanox Technologies. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include + +#include +#include +#include +#include + +#include "mlx4_ib.h" + +#define CM_CLEANUP_CACHE_TIMEOUT (5 * HZ) + +struct id_map_entry { + struct rb_node node; + + u32 sl_cm_id; + u32 pv_cm_id; + int slave_id; + int scheduled_delete; + struct mlx4_ib_dev *dev; + + struct list_head list; + struct delayed_work timeout; +}; + +struct cm_generic_msg { + struct ib_mad_hdr hdr; + + __be32 local_comm_id; + __be32 remote_comm_id; +}; + +struct cm_req_msg { + unsigned char unused[0x60]; + union ib_gid primary_path_sgid; +}; + + +static void set_local_comm_id(struct ib_mad *mad, u32 cm_id) +{ + struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; + msg->local_comm_id = cpu_to_be32(cm_id); +} + +static u32 get_local_comm_id(struct ib_mad *mad) +{ + struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; + + return be32_to_cpu(msg->local_comm_id); +} + +static void set_remote_comm_id(struct ib_mad *mad, u32 cm_id) +{ + struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; + msg->remote_comm_id = cpu_to_be32(cm_id); +} + +static u32 get_remote_comm_id(struct ib_mad *mad) +{ + struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; + + return be32_to_cpu(msg->remote_comm_id); +} + +static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad) +{ + struct cm_req_msg *msg = (struct cm_req_msg *)mad; + + return msg->primary_path_sgid; +} + +/* Lock should be taken before called */ +static struct id_map_entry * +id_map_find_by_sl_id(struct ib_device *ibdev, u32 slave_id, u32 sl_cm_id) +{ + struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map; + struct rb_node *node = sl_id_map->rb_node; + + while (node) { + struct id_map_entry *id_map_entry = + rb_entry(node, struct id_map_entry, node); + + if (id_map_entry->sl_cm_id > sl_cm_id) + node = node->rb_left; + else if (id_map_entry->sl_cm_id < sl_cm_id) + node = node->rb_right; + else if (id_map_entry->slave_id > slave_id) + node = node->rb_left; + else if (id_map_entry->slave_id < slave_id) + node = node->rb_right; + else + return id_map_entry; + } + return NULL; +} + +static void id_map_ent_timeout(struct work_struct *work) +{ + struct delayed_work *delay = to_delayed_work(work); + struct id_map_entry *ent = container_of(delay, struct id_map_entry, timeout); + struct id_map_entry *db_ent, *found_ent; + struct mlx4_ib_dev *dev = ent->dev; + struct mlx4_ib_sriov *sriov = &dev->sriov; + struct rb_root *sl_id_map = &sriov->sl_id_map; + int pv_id = (int) ent->pv_cm_id; + + spin_lock(&sriov->id_map_lock); + db_ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_id); + if (!db_ent) + goto out; + found_ent = id_map_find_by_sl_id(&dev->ib_dev, ent->slave_id, ent->sl_cm_id); + if (found_ent && found_ent == ent) + rb_erase(&found_ent->node, sl_id_map); + idr_remove(&sriov->pv_id_table, pv_id); + +out: + list_del(&ent->list); + spin_unlock(&sriov->id_map_lock); + kfree(ent); +} + +static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id) +{ + struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; + struct rb_root *sl_id_map = &sriov->sl_id_map; + struct id_map_entry *ent, *found_ent; + + spin_lock(&sriov->id_map_lock); + ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_cm_id); + if (!ent) + goto out; + found_ent = id_map_find_by_sl_id(ibdev, ent->slave_id, ent->sl_cm_id); + if (found_ent && found_ent == ent) + rb_erase(&found_ent->node, sl_id_map); + idr_remove(&sriov->pv_id_table, pv_cm_id); +out: + spin_unlock(&sriov->id_map_lock); +} + +static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new) +{ + struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map; + struct rb_node **link = &sl_id_map->rb_node, *parent = NULL; + struct id_map_entry *ent; + int slave_id = new->slave_id; + int sl_cm_id = new->sl_cm_id; + + ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id); + if (ent) { + pr_debug("overriding existing sl_id_map entry (cm_id = %x)\n", + sl_cm_id); + + rb_replace_node(&ent->node, &new->node, sl_id_map); + return; + } + + /* Go to the bottom of the tree */ + while (*link) { + parent = *link; + ent = rb_entry(parent, struct id_map_entry, node); + + if (ent->sl_cm_id > sl_cm_id || (ent->sl_cm_id == sl_cm_id && ent->slave_id > slave_id)) + link = &(*link)->rb_left; + else + link = &(*link)->rb_right; + } + + rb_link_node(&new->node, parent, link); + rb_insert_color(&new->node, sl_id_map); +} + +static struct id_map_entry * +id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id) +{ + int ret, id; + static int next_id; + struct id_map_entry *ent; + struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; + + ent = kmalloc(sizeof (struct id_map_entry), GFP_KERNEL); + if (!ent) { + mlx4_ib_warn(ibdev, "Couldn't allocate id cache entry - out of memory\n"); + return ERR_PTR(-ENOMEM); + } + + ent->sl_cm_id = sl_cm_id; + ent->slave_id = slave_id; + ent->scheduled_delete = 0; + ent->dev = to_mdev(ibdev); + INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout); + + do { + spin_lock(&to_mdev(ibdev)->sriov.id_map_lock); + ret = idr_get_new_above(&sriov->pv_id_table, ent, + next_id, &id); + if (!ret) { + next_id = ((unsigned) id + 1) & MAX_ID_MASK; + ent->pv_cm_id = (u32)id; + sl_id_map_add(ibdev, ent); + } + + spin_unlock(&sriov->id_map_lock); + } while (ret == -EAGAIN && idr_pre_get(&sriov->pv_id_table, GFP_KERNEL)); + /*the function idr_get_new_above can return -ENOSPC, so don't insert in that case.*/ + if (!ret) { + spin_lock(&sriov->id_map_lock); + list_add_tail(&ent->list, &sriov->cm_list); + spin_unlock(&sriov->id_map_lock); + return ent; + } + /*error flow*/ + kfree(ent); + mlx4_ib_warn(ibdev, "No more space in the idr (err:0x%x)\n", ret); + return ERR_PTR(-ENOMEM); +} + +static struct id_map_entry * +id_map_get(struct ib_device *ibdev, int *pv_cm_id, int sl_cm_id, int slave_id) +{ + struct id_map_entry *ent; + struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; + + spin_lock(&sriov->id_map_lock); + if (*pv_cm_id == -1) { + ent = id_map_find_by_sl_id(ibdev, sl_cm_id, slave_id); + if (ent) + *pv_cm_id = (int) ent->pv_cm_id; + } else + ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, *pv_cm_id); + spin_unlock(&sriov->id_map_lock); + + return ent; +} + +static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id) +{ + struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; + unsigned long flags; + + spin_lock_irqsave(&sriov->going_down_lock, flags); + spin_lock(&sriov->id_map_lock); + /*make sure that there is no schedule inside the scheduled work.*/ + if (!sriov->is_going_down) { + id->scheduled_delete = 1; + schedule_delayed_work(&id->timeout, CM_CLEANUP_CACHE_TIMEOUT); + } + spin_unlock(&sriov->id_map_lock); + spin_unlock_irqrestore(&sriov->going_down_lock, flags); +} + +int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id, + struct ib_mad *mad) +{ + struct id_map_entry *id; + u32 sl_cm_id; + int pv_cm_id = -1; + + sl_cm_id = get_local_comm_id(mad); + + if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID || + mad->mad_hdr.attr_id == CM_REP_ATTR_ID) { + id = id_map_alloc(ibdev, slave_id, sl_cm_id); + if (IS_ERR(id)) { + mlx4_ib_warn(ibdev, "%s: id{slave: %d, sl_cm_id: 0x%x} Failed to id_map_alloc\n", + __func__, slave_id, sl_cm_id); + return PTR_ERR(id); + } + } else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID) { + return 0; + } else { + id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id); + } + + if (!id) { + pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n", + slave_id, sl_cm_id); + return -EINVAL; + } + + set_local_comm_id(mad, id->pv_cm_id); + + if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID) + schedule_delayed(ibdev, id); + else if (mad->mad_hdr.attr_id == CM_DREP_ATTR_ID) + id_map_find_del(ibdev, pv_cm_id); + + return 0; +} + +int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave, + struct ib_mad *mad) +{ + u32 pv_cm_id; + struct id_map_entry *id; + + if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID) { + union ib_gid gid; + + gid = gid_from_req_msg(ibdev, mad); + *slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id); + if (*slave < 0) { + mlx4_ib_warn(ibdev, "failed matching slave_id by gid (0x%llx)\n", + gid.global.interface_id); + return -ENOENT; + } + return 0; + } + + pv_cm_id = get_remote_comm_id(mad); + id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1); + + if (!id) { + pr_debug("Couldn't find an entry for pv_cm_id 0x%x\n", pv_cm_id); + return -ENOENT; + } + + *slave = id->slave_id; + set_remote_comm_id(mad, id->sl_cm_id); + + if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID) + schedule_delayed(ibdev, id); + else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID || + mad->mad_hdr.attr_id == CM_DREP_ATTR_ID) { + id_map_find_del(ibdev, (int) pv_cm_id); + } + + return 0; +} + +void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev) +{ + spin_lock_init(&dev->sriov.id_map_lock); + INIT_LIST_HEAD(&dev->sriov.cm_list); + dev->sriov.sl_id_map = RB_ROOT; + idr_init(&dev->sriov.pv_id_table); + idr_pre_get(&dev->sriov.pv_id_table, GFP_KERNEL); +} + +/* slave = -1 ==> all slaves */ +/* TBD -- call paravirt clean for single slave. Need for slave RESET event */ +void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave) +{ + struct mlx4_ib_sriov *sriov = &dev->sriov; + struct rb_root *sl_id_map = &sriov->sl_id_map; + struct list_head lh; + struct rb_node *nd; + int need_flush = 1; + struct id_map_entry *map, *tmp_map; + /* cancel all delayed work queue entries */ + INIT_LIST_HEAD(&lh); + spin_lock(&sriov->id_map_lock); + list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) { + if (slave < 0 || slave == map->slave_id) { + if (map->scheduled_delete) + need_flush &= !!cancel_delayed_work(&map->timeout); + } + } + + spin_unlock(&sriov->id_map_lock); + + if (!need_flush) + flush_scheduled_work(); /* make sure all timers were flushed */ + + /* now, remove all leftover entries from databases*/ + spin_lock(&sriov->id_map_lock); + if (slave < 0) { + while (rb_first(sl_id_map)) { + struct id_map_entry *ent = + rb_entry(rb_first(sl_id_map), + struct id_map_entry, node); + + rb_erase(&ent->node, sl_id_map); + idr_remove(&sriov->pv_id_table, (int) ent->pv_cm_id); + } + list_splice_init(&dev->sriov.cm_list, &lh); + } else { + /* first, move nodes belonging to slave to db remove list */ + nd = rb_first(sl_id_map); + while (nd) { + struct id_map_entry *ent = + rb_entry(nd, struct id_map_entry, node); + nd = rb_next(nd); + if (ent->slave_id == slave) + list_move_tail(&ent->list, &lh); + } + /* remove those nodes from databases */ + list_for_each_entry_safe(map, tmp_map, &lh, list) { + rb_erase(&map->node, sl_id_map); + idr_remove(&sriov->pv_id_table, (int) map->pv_cm_id); + } + + /* add remaining nodes from cm_list */ + list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) { + if (slave == map->slave_id) + list_move_tail(&map->list, &lh); + } + } + + spin_unlock(&sriov->id_map_lock); + + /* free any map entries left behind due to cancel_delayed_work above */ + list_for_each_entry_safe(map, tmp_map, &lh, list) { + list_del(&map->list); + kfree(map); + } +} diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 29ed3b43e4a6..2f13894299ee 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -544,6 +544,10 @@ static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port, (struct ib_sa_mad *) mad)) return 0; break; + case IB_MGMT_CLASS_CM: + if (mlx4_ib_demux_cm_handler(ibdev, port, &slave, mad)) + return 0; + break; case IB_MGMT_CLASS_DEVICE_MGMT: if (mad->mad_hdr.method != IB_MGMT_METHOD_GET_RESP) return 0; @@ -1076,6 +1080,11 @@ static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc (struct ib_sa_mad *) &tunnel->mad)) return; break; + case IB_MGMT_CLASS_CM: + if (mlx4_ib_multiplex_cm_handler(ctx->ib_dev, ctx->port, slave, + (struct ib_mad *) &tunnel->mad)) + return; + break; case IB_MGMT_CLASS_DEVICE_MGMT: if (tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_GET && tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_SET) @@ -1790,6 +1799,7 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev) dev->sriov.is_going_down = 0; spin_lock_init(&dev->sriov.going_down_lock); + mlx4_ib_cm_paravirt_init(dev); mlx4_ib_warn(&dev->ib_dev, "multi-function enabled\n"); @@ -1818,6 +1828,7 @@ demux_err: mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]); --i; } + mlx4_ib_cm_paravirt_clean(dev, -1); return err; } @@ -1833,7 +1844,7 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev) spin_lock_irqsave(&dev->sriov.going_down_lock, flags); dev->sriov.is_going_down = 1; spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags); - if (mlx4_is_master(dev->dev)) + if (mlx4_is_master(dev->dev)) { for (i = 0; i < dev->num_ports; i++) { flush_workqueue(dev->sriov.demux[i].ud_wq); mlx4_ib_free_sqp_ctx(dev->sriov.sqps[i]); @@ -1841,4 +1852,7 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev) dev->sriov.sqps[i] = NULL; mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]); } + + mlx4_ib_cm_paravirt_clean(dev, -1); + } } diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index 01ba9f1692b1..7476e2439f6b 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -348,6 +348,12 @@ struct mlx4_ib_sriov { * it may be called from interrupt context.*/ spinlock_t going_down_lock; int is_going_down; + + /* CM paravirtualization fields */ + struct list_head cm_list; + spinlock_t id_map_lock; + struct rb_root sl_id_map; + struct idr pv_id_table; }; struct mlx4_ib_iboe { @@ -591,4 +597,13 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port, u32 qkey, struct ib_ah_attr *attr, struct ib_mad *mad); __be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx); +int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave, + struct ib_mad *mad); + +int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id, + struct ib_mad *mad); + +void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev); +void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave_id); + #endif /* MLX4_IB_H */ From 993c401e207946fa56f69c51e39f015e7108497b Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:48 +0000 Subject: [PATCH 19/45] mlx4_core: Add IB port-state machine and port mgmt event propagation For an IB port, a slave should not show port active until that slave has a valid alias-guid (provided by the subnet manager). Therefore the port-up event should be passed to a slave only after both the port is up, and the slave's alias-guid has been set. Also, provide the infrastructure for propagating port-management events (client-reregister, etc) to slaves. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/eq.c | 249 ++++++++++++++++++++-- drivers/net/ethernet/mellanox/mlx4/mlx4.h | 1 + include/linux/mlx4/device.h | 28 +++ 3 files changed, 256 insertions(+), 22 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/eq.c b/drivers/net/ethernet/mellanox/mlx4/eq.c index 99a04648fab0..c425826b1fc0 100644 --- a/drivers/net/ethernet/mellanox/mlx4/eq.c +++ b/drivers/net/ethernet/mellanox/mlx4/eq.c @@ -200,6 +200,196 @@ static void mlx4_slave_event(struct mlx4_dev *dev, int slave, slave_event(dev, slave, eqe); } +int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port) +{ + struct mlx4_eqe eqe; + + struct mlx4_priv *priv = mlx4_priv(dev); + struct mlx4_slave_state *s_slave = &priv->mfunc.master.slave_state[slave]; + + if (!s_slave->active) + return 0; + + memset(&eqe, 0, sizeof eqe); + + eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT; + eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE; + eqe.event.port_mgmt_change.port = port; + + return mlx4_GEN_EQE(dev, slave, &eqe); +} +EXPORT_SYMBOL(mlx4_gen_pkey_eqe); + +int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port) +{ + struct mlx4_eqe eqe; + + /*don't send if we don't have the that slave */ + if (dev->num_vfs < slave) + return 0; + memset(&eqe, 0, sizeof eqe); + + eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT; + eqe.subtype = MLX4_DEV_PMC_SUBTYPE_GUID_INFO; + eqe.event.port_mgmt_change.port = port; + + return mlx4_GEN_EQE(dev, slave, &eqe); +} +EXPORT_SYMBOL(mlx4_gen_guid_change_eqe); + +int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port, + u8 port_subtype_change) +{ + struct mlx4_eqe eqe; + + /*don't send if we don't have the that slave */ + if (dev->num_vfs < slave) + return 0; + memset(&eqe, 0, sizeof eqe); + + eqe.type = MLX4_EVENT_TYPE_PORT_CHANGE; + eqe.subtype = port_subtype_change; + eqe.event.port_change.port = cpu_to_be32(port << 28); + + mlx4_dbg(dev, "%s: sending: %d to slave: %d on port: %d\n", __func__, + port_subtype_change, slave, port); + return mlx4_GEN_EQE(dev, slave, &eqe); +} +EXPORT_SYMBOL(mlx4_gen_port_state_change_eqe); + +enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port) +{ + struct mlx4_priv *priv = mlx4_priv(dev); + struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state; + if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS) { + pr_err("%s: Error: asking for slave:%d, port:%d\n", + __func__, slave, port); + return SLAVE_PORT_DOWN; + } + return s_state[slave].port_state[port]; +} +EXPORT_SYMBOL(mlx4_get_slave_port_state); + +static int mlx4_set_slave_port_state(struct mlx4_dev *dev, int slave, u8 port, + enum slave_port_state state) +{ + struct mlx4_priv *priv = mlx4_priv(dev); + struct mlx4_slave_state *s_state = priv->mfunc.master.slave_state; + + if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) { + pr_err("%s: Error: asking for slave:%d, port:%d\n", + __func__, slave, port); + return -1; + } + s_state[slave].port_state[port] = state; + + return 0; +} + +static void set_all_slave_state(struct mlx4_dev *dev, u8 port, int event) +{ + int i; + enum slave_port_gen_event gen_event; + + for (i = 0; i < dev->num_slaves; i++) + set_and_calc_slave_port_state(dev, i, port, event, &gen_event); +} +/************************************************************************** + The function get as input the new event to that port, + and according to the prev state change the slave's port state. + The events are: + MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN, + MLX4_PORT_STATE_DEV_EVENT_PORT_UP + MLX4_PORT_STATE_IB_EVENT_GID_VALID + MLX4_PORT_STATE_IB_EVENT_GID_INVALID +***************************************************************************/ +int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave, + u8 port, int event, + enum slave_port_gen_event *gen_event) +{ + struct mlx4_priv *priv = mlx4_priv(dev); + struct mlx4_slave_state *ctx = NULL; + unsigned long flags; + int ret = -1; + enum slave_port_state cur_state = + mlx4_get_slave_port_state(dev, slave, port); + + *gen_event = SLAVE_PORT_GEN_EVENT_NONE; + + if (slave >= dev->num_slaves || port > MLX4_MAX_PORTS || port == 0) { + pr_err("%s: Error: asking for slave:%d, port:%d\n", + __func__, slave, port); + return ret; + } + + ctx = &priv->mfunc.master.slave_state[slave]; + spin_lock_irqsave(&ctx->lock, flags); + + mlx4_dbg(dev, "%s: slave: %d, current state: %d new event :%d\n", + __func__, slave, cur_state, event); + + switch (cur_state) { + case SLAVE_PORT_DOWN: + if (MLX4_PORT_STATE_DEV_EVENT_PORT_UP == event) + mlx4_set_slave_port_state(dev, slave, port, + SLAVE_PENDING_UP); + break; + case SLAVE_PENDING_UP: + if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) + mlx4_set_slave_port_state(dev, slave, port, + SLAVE_PORT_DOWN); + else if (MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID == event) { + mlx4_set_slave_port_state(dev, slave, port, + SLAVE_PORT_UP); + *gen_event = SLAVE_PORT_GEN_EVENT_UP; + } + break; + case SLAVE_PORT_UP: + if (MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN == event) { + mlx4_set_slave_port_state(dev, slave, port, + SLAVE_PORT_DOWN); + *gen_event = SLAVE_PORT_GEN_EVENT_DOWN; + } else if (MLX4_PORT_STATE_IB_EVENT_GID_INVALID == + event) { + mlx4_set_slave_port_state(dev, slave, port, + SLAVE_PENDING_UP); + *gen_event = SLAVE_PORT_GEN_EVENT_DOWN; + } + break; + default: + pr_err("%s: BUG!!! UNKNOWN state: " + "slave:%d, port:%d\n", __func__, slave, port); + goto out; + } + ret = mlx4_get_slave_port_state(dev, slave, port); + mlx4_dbg(dev, "%s: slave: %d, current state: %d new event" + " :%d gen_event: %d\n", + __func__, slave, cur_state, event, *gen_event); + +out: + spin_unlock_irqrestore(&ctx->lock, flags); + return ret; +} + +EXPORT_SYMBOL(set_and_calc_slave_port_state); + +int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr) +{ + struct mlx4_eqe eqe; + + memset(&eqe, 0, sizeof eqe); + + eqe.type = MLX4_EVENT_TYPE_PORT_MNG_CHG_EVENT; + eqe.subtype = MLX4_DEV_PMC_SUBTYPE_PORT_INFO; + eqe.event.port_mgmt_change.port = port; + eqe.event.port_mgmt_change.params.port_info.changed_attr = + cpu_to_be32((u32) attr); + + slave_event(dev, ALL_SLAVES, &eqe); + return 0; +} +EXPORT_SYMBOL(mlx4_gen_slaves_port_mgt_ev); + void mlx4_master_handle_slave_flr(struct work_struct *work) { struct mlx4_mfunc_master_ctx *master = @@ -251,6 +441,7 @@ static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq) u32 flr_slave; u8 update_slave_state; int i; + enum slave_port_gen_event gen_event; while ((eqe = next_eqe_sw(eq))) { /* @@ -347,35 +538,49 @@ static int mlx4_eq_int(struct mlx4_dev *dev, struct mlx4_eq *eq) case MLX4_EVENT_TYPE_PORT_CHANGE: port = be32_to_cpu(eqe->event.port_change.port) >> 28; if (eqe->subtype == MLX4_PORT_CHANGE_SUBTYPE_DOWN) { - mlx4_dispatch_event(dev, - MLX4_DEV_EVENT_PORT_DOWN, + mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_DOWN, port); mlx4_priv(dev)->sense.do_sense_port[port] = 1; - if (mlx4_is_master(dev)) - /*change the state of all slave's port - * to down:*/ - for (i = 0; i < dev->num_slaves; i++) { - mlx4_dbg(dev, "%s: Sending " - "MLX4_PORT_CHANGE_SUBTYPE_DOWN" + if (!mlx4_is_master(dev)) + break; + for (i = 0; i < dev->num_slaves; i++) { + if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) { + if (i == mlx4_master_func_num(dev)) + continue; + mlx4_dbg(dev, "%s: Sending MLX4_PORT_CHANGE_SUBTYPE_DOWN" " to slave: %d, port:%d\n", __func__, i, port); - if (i == dev->caps.function) - continue; - mlx4_slave_event(dev, i, eqe); - } - } else { - mlx4_dispatch_event(dev, - MLX4_DEV_EVENT_PORT_UP, - port); - mlx4_priv(dev)->sense.do_sense_port[port] = 0; - - if (mlx4_is_master(dev)) { - for (i = 0; i < dev->num_slaves; i++) { - if (i == dev->caps.function) - continue; mlx4_slave_event(dev, i, eqe); + } else { /* IB port */ + set_and_calc_slave_port_state(dev, i, port, + MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN, + &gen_event); + /*we can be in pending state, then do not send port_down event*/ + if (SLAVE_PORT_GEN_EVENT_DOWN == gen_event) { + if (i == mlx4_master_func_num(dev)) + continue; + mlx4_slave_event(dev, i, eqe); + } } } + } else { + mlx4_dispatch_event(dev, MLX4_DEV_EVENT_PORT_UP, port); + + mlx4_priv(dev)->sense.do_sense_port[port] = 0; + + if (!mlx4_is_master(dev)) + break; + if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH) + for (i = 0; i < dev->num_slaves; i++) { + if (i == mlx4_master_func_num(dev)) + continue; + mlx4_slave_event(dev, i, eqe); + } + else /* IB port */ + /* port-up event will be sent to a slave when the + * slave's alias-guid is set. This is done in alias_GUID.c + */ + set_all_slave_state(dev, port, MLX4_DEV_EVENT_PORT_UP); } break; diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h index 7d27c3158d0c..23f74759f403 100644 --- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h +++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h @@ -452,6 +452,7 @@ struct mlx4_slave_state { /*initialized via the kzalloc*/ u8 is_slave_going_down; u32 cookie; + enum slave_port_state port_state[MLX4_MAX_PORTS + 1]; }; struct slave_list { diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index 9803fd5d3dba..380e01671d1e 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -192,6 +192,25 @@ enum { MLX4_FATAL_WARNING_SUBTYPE_WARMING = 0, }; +enum slave_port_state { + SLAVE_PORT_DOWN = 0, + SLAVE_PENDING_UP, + SLAVE_PORT_UP, +}; + +enum slave_port_gen_event { + SLAVE_PORT_GEN_EVENT_DOWN = 0, + SLAVE_PORT_GEN_EVENT_UP, + SLAVE_PORT_GEN_EVENT_NONE, +}; + +enum slave_port_state_event { + MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN, + MLX4_PORT_STATE_DEV_EVENT_PORT_UP, + MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID, + MLX4_PORT_STATE_IB_EVENT_GID_INVALID, +}; + enum { MLX4_PERM_LOCAL_READ = 1 << 10, MLX4_PERM_LOCAL_WRITE = 1 << 11, @@ -945,4 +964,13 @@ void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey); +int mlx4_is_slave_active(struct mlx4_dev *dev, int slave); +int mlx4_gen_pkey_eqe(struct mlx4_dev *dev, int slave, u8 port); +int mlx4_gen_guid_change_eqe(struct mlx4_dev *dev, int slave, u8 port); +int mlx4_gen_slaves_port_mgt_ev(struct mlx4_dev *dev, u8 port, int attr); +int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port, u8 port_subtype_change); +enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port); +int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave, u8 port, int event, enum slave_port_gen_event *gen_event); + + #endif /* MLX4_DEVICE_H */ From a0c64a17aba88c29d55ba989b96ac6ccb1268f0a Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:49 +0000 Subject: [PATCH 20/45] mlx4: Add alias_guid mechanism For IB ports, we paravirtualize the GUID at index 0 on slaves. The GUID at index 0 seen by a slave is the actual GUID occupying the GUID table at the slave-id index. The driver, by default, requests at startup time that subnet manager populate its entire guid table with GUIDs. These guids are then mapped (paravirtualized) to the slaves, and appear for each slave as its GUID at index 0. Until each slave has such a guid, its port status is DOWN. The guid table is cached to support special QP paravirtualization, and event propagation to slaves on guid change (we test to see if the guid really changed before propagating an event to the slave). To support this caching, add capability to __mlx4_ib_query_gid() to obtain the network view (i.e., physical view) gid at index X, not just the host (paravirtualized) view. Based on a patch from Erez Shitrit Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/Makefile | 2 +- drivers/infiniband/hw/mlx4/alias_GUID.c | 688 +++++++++++++++++++++++ drivers/infiniband/hw/mlx4/mad.c | 19 +- drivers/infiniband/hw/mlx4/main.c | 37 +- drivers/infiniband/hw/mlx4/mlx4_ib.h | 74 +++ drivers/net/ethernet/mellanox/mlx4/cmd.c | 6 +- 6 files changed, 816 insertions(+), 10 deletions(-) create mode 100644 drivers/infiniband/hw/mlx4/alias_GUID.c diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile index bf0aa901c313..31d4c8aac679 100644 --- a/drivers/infiniband/hw/mlx4/Makefile +++ b/drivers/infiniband/hw/mlx4/Makefile @@ -1,3 +1,3 @@ obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o -mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o +mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o alias_GUID.o diff --git a/drivers/infiniband/hw/mlx4/alias_GUID.c b/drivers/infiniband/hw/mlx4/alias_GUID.c new file mode 100644 index 000000000000..ef6d356927c3 --- /dev/null +++ b/drivers/infiniband/hw/mlx4/alias_GUID.c @@ -0,0 +1,688 @@ +/* + * Copyright (c) 2012 Mellanox Technologies. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + /***********************************************************/ +/*This file support the handling of the Alias GUID feature. */ +/***********************************************************/ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "mlx4_ib.h" + +/* +The driver keeps the current state of all guids, as they are in the HW. +Whenever we receive an smp mad GUIDInfo record, the data will be cached. +*/ + +struct mlx4_alias_guid_work_context { + u8 port; + struct mlx4_ib_dev *dev ; + struct ib_sa_query *sa_query; + struct completion done; + int query_id; + struct list_head list; + int block_num; +}; + +struct mlx4_next_alias_guid_work { + u8 port; + u8 block_num; + struct mlx4_sriov_alias_guid_info_rec_det rec_det; +}; + + +void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, int block_num, + u8 port_num, u8 *p_data) +{ + int i; + u64 guid_indexes; + int slave_id; + int port_index = port_num - 1; + + if (!mlx4_is_master(dev->dev)) + return; + + guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid. + ports_guid[port_num - 1]. + all_rec_per_port[block_num].guid_indexes); + pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes); + + for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) { + /* The location of the specific index starts from bit number 4 + * until bit num 11 */ + if (test_bit(i + 4, (unsigned long *)&guid_indexes)) { + slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ; + if (slave_id >= dev->dev->num_slaves) { + pr_debug("The last slave: %d\n", slave_id); + return; + } + + /* cache the guid: */ + memcpy(&dev->sriov.demux[port_index].guid_cache[slave_id], + &p_data[i * GUID_REC_SIZE], + GUID_REC_SIZE); + } else + pr_debug("Guid number: %d in block: %d" + " was not updated\n", i, block_num); + } +} + +static __be64 get_cached_alias_guid(struct mlx4_ib_dev *dev, int port, int index) +{ + if (index >= NUM_ALIAS_GUID_PER_PORT) { + pr_err("%s: ERROR: asked for index:%d\n", __func__, index); + return (__force __be64) ((u64) 0xFFFFFFFFFFFFFFFFUL); + } + return *(__be64 *)&dev->sriov.demux[port - 1].guid_cache[index]; +} + + +static ib_sa_comp_mask get_aguid_comp_mask_from_ix(int index) +{ + return IB_SA_COMP_MASK(4 + index); +} + +/* + * Whenever new GUID is set/unset (guid table change) create event and + * notify the relevant slave (master also should be notified). + * If the GUID value is not as we have in the cache the slave will not be + * updated; in this case it waits for the smp_snoop or the port management + * event to call the function and to update the slave. + * block_number - the index of the block (16 blocks available) + * port_number - 1 or 2 + */ +void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev, + int block_num, u8 port_num, + u8 *p_data) +{ + int i; + u64 guid_indexes; + int slave_id; + enum slave_port_state new_state; + enum slave_port_state prev_state; + __be64 tmp_cur_ag, form_cache_ag; + enum slave_port_gen_event gen_event; + + if (!mlx4_is_master(dev->dev)) + return; + + guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid. + ports_guid[port_num - 1]. + all_rec_per_port[block_num].guid_indexes); + pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes); + + /*calculate the slaves and notify them*/ + for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) { + /* the location of the specific index runs from bits 4..11 */ + if (!(test_bit(i + 4, (unsigned long *)&guid_indexes))) + continue; + + slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ; + if (slave_id >= dev->dev->num_slaves) + return; + tmp_cur_ag = *(__be64 *)&p_data[i * GUID_REC_SIZE]; + form_cache_ag = get_cached_alias_guid(dev, port_num, + (NUM_ALIAS_GUID_IN_REC * block_num) + i); + /* + * Check if guid is not the same as in the cache, + * If it is different, wait for the snoop_smp or the port mgmt + * change event to update the slave on its port state change + */ + if (tmp_cur_ag != form_cache_ag) + continue; + mlx4_gen_guid_change_eqe(dev->dev, slave_id, port_num); + + /*2 cases: Valid GUID, and Invalid Guid*/ + + if (tmp_cur_ag != MLX4_NOT_SET_GUID) { /*valid GUID*/ + prev_state = mlx4_get_slave_port_state(dev->dev, slave_id, port_num); + new_state = set_and_calc_slave_port_state(dev->dev, slave_id, port_num, + MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID, + &gen_event); + pr_debug("slave: %d, port: %d prev_port_state: %d," + " new_port_state: %d, gen_event: %d\n", + slave_id, port_num, prev_state, new_state, gen_event); + if (gen_event == SLAVE_PORT_GEN_EVENT_UP) { + pr_debug("sending PORT_UP event to slave: %d, port: %d\n", + slave_id, port_num); + mlx4_gen_port_state_change_eqe(dev->dev, slave_id, + port_num, MLX4_PORT_CHANGE_SUBTYPE_ACTIVE); + } + } else { /* request to invalidate GUID */ + set_and_calc_slave_port_state(dev->dev, slave_id, port_num, + MLX4_PORT_STATE_IB_EVENT_GID_INVALID, + &gen_event); + pr_debug("sending PORT DOWN event to slave: %d, port: %d\n", + slave_id, port_num); + mlx4_gen_port_state_change_eqe(dev->dev, slave_id, port_num, + MLX4_PORT_CHANGE_SUBTYPE_DOWN); + } + } +} + +static void aliasguid_query_handler(int status, + struct ib_sa_guidinfo_rec *guid_rec, + void *context) +{ + struct mlx4_ib_dev *dev; + struct mlx4_alias_guid_work_context *cb_ctx = context; + u8 port_index ; + int i; + struct mlx4_sriov_alias_guid_info_rec_det *rec; + unsigned long flags, flags1; + + if (!context) + return; + + dev = cb_ctx->dev; + port_index = cb_ctx->port - 1; + rec = &dev->sriov.alias_guid.ports_guid[port_index]. + all_rec_per_port[cb_ctx->block_num]; + + if (status) { + rec->status = MLX4_GUID_INFO_STATUS_IDLE; + pr_debug("(port: %d) failed: status = %d\n", + cb_ctx->port, status); + goto out; + } + + if (guid_rec->block_num != cb_ctx->block_num) { + pr_err("block num mismatch: %d != %d\n", + cb_ctx->block_num, guid_rec->block_num); + goto out; + } + + pr_debug("lid/port: %d/%d, block_num: %d\n", + be16_to_cpu(guid_rec->lid), cb_ctx->port, + guid_rec->block_num); + + rec = &dev->sriov.alias_guid.ports_guid[port_index]. + all_rec_per_port[guid_rec->block_num]; + + rec->status = MLX4_GUID_INFO_STATUS_SET; + rec->method = MLX4_GUID_INFO_RECORD_SET; + + for (i = 0 ; i < NUM_ALIAS_GUID_IN_REC; i++) { + __be64 tmp_cur_ag; + tmp_cur_ag = *(__be64 *)&guid_rec->guid_info_list[i * GUID_REC_SIZE]; + /* check if the SM didn't assign one of the records. + * if it didn't, if it was not sysadmin request: + * ask the SM to give a new GUID, (instead of the driver request). + */ + if (tmp_cur_ag == MLX4_NOT_SET_GUID) { + mlx4_ib_warn(&dev->ib_dev, "%s:Record num %d in " + "block_num: %d was declined by SM, " + "ownership by %d (0 = driver, 1=sysAdmin," + " 2=None)\n", __func__, i, + guid_rec->block_num, rec->ownership); + if (rec->ownership == MLX4_GUID_DRIVER_ASSIGN) { + /* if it is driver assign, asks for new GUID from SM*/ + *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE] = + MLX4_NOT_SET_GUID; + + /* Mark the record as not assigned, and let it + * be sent again in the next work sched.*/ + rec->status = MLX4_GUID_INFO_STATUS_IDLE; + rec->guid_indexes |= get_aguid_comp_mask_from_ix(i); + } + } else { + /* properly assigned record. */ + /* We save the GUID we just got from the SM in the + * admin_guid in order to be persistent, and in the + * request from the sm the process will ask for the same GUID */ + if (rec->ownership == MLX4_GUID_SYSADMIN_ASSIGN && + tmp_cur_ag != *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE]) { + /* the sysadmin assignment failed.*/ + mlx4_ib_warn(&dev->ib_dev, "%s: Failed to set" + " admin guid after SysAdmin " + "configuration. " + "Record num %d in block_num:%d " + "was declined by SM, " + "new val(0x%llx) was kept\n", + __func__, i, + guid_rec->block_num, + be64_to_cpu(*(__be64 *) & + rec->all_recs[i * GUID_REC_SIZE])); + } else { + memcpy(&rec->all_recs[i * GUID_REC_SIZE], + &guid_rec->guid_info_list[i * GUID_REC_SIZE], + GUID_REC_SIZE); + } + } + } + /* + The func is call here to close the cases when the + sm doesn't send smp, so in the sa response the driver + notifies the slave. + */ + mlx4_ib_notify_slaves_on_guid_change(dev, guid_rec->block_num, + cb_ctx->port, + guid_rec->guid_info_list); +out: + spin_lock_irqsave(&dev->sriov.going_down_lock, flags); + spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1); + if (!dev->sriov.is_going_down) + queue_delayed_work(dev->sriov.alias_guid.ports_guid[port_index].wq, + &dev->sriov.alias_guid.ports_guid[port_index]. + alias_guid_work, 0); + if (cb_ctx->sa_query) { + list_del(&cb_ctx->list); + kfree(cb_ctx); + } else + complete(&cb_ctx->done); + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1); + spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags); +} + +static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index) +{ + int i; + u64 cur_admin_val; + ib_sa_comp_mask comp_mask = 0; + + dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].status + = MLX4_GUID_INFO_STATUS_IDLE; + dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].method + = MLX4_GUID_INFO_RECORD_SET; + + /* calculate the comp_mask for that record.*/ + for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) { + cur_admin_val = + *(u64 *)&dev->sriov.alias_guid.ports_guid[port - 1]. + all_rec_per_port[index].all_recs[GUID_REC_SIZE * i]; + /* + check the admin value: if it's for delete (~00LL) or + it is the first guid of the first record (hw guid) or + the records is not in ownership of the sysadmin and the sm doesn't + need to assign GUIDs, then don't put it up for assignment. + */ + if (MLX4_GUID_FOR_DELETE_VAL == cur_admin_val || + (!index && !i) || + MLX4_GUID_NONE_ASSIGN == dev->sriov.alias_guid. + ports_guid[port - 1].all_rec_per_port[index].ownership) + continue; + comp_mask |= get_aguid_comp_mask_from_ix(i); + } + dev->sriov.alias_guid.ports_guid[port - 1]. + all_rec_per_port[index].guid_indexes = comp_mask; +} + +static int set_guid_rec(struct ib_device *ibdev, + u8 port, int index, + struct mlx4_sriov_alias_guid_info_rec_det *rec_det) +{ + int err; + struct mlx4_ib_dev *dev = to_mdev(ibdev); + struct ib_sa_guidinfo_rec guid_info_rec; + ib_sa_comp_mask comp_mask; + struct ib_port_attr attr; + struct mlx4_alias_guid_work_context *callback_context; + unsigned long resched_delay, flags, flags1; + struct list_head *head = + &dev->sriov.alias_guid.ports_guid[port - 1].cb_list; + + err = __mlx4_ib_query_port(ibdev, port, &attr, 1); + if (err) { + pr_debug("mlx4_ib_query_port failed (err: %d), port: %d\n", + err, port); + return err; + } + /*check the port was configured by the sm, otherwise no need to send */ + if (attr.state != IB_PORT_ACTIVE) { + pr_debug("port %d not active...rescheduling\n", port); + resched_delay = 5 * HZ; + err = -EAGAIN; + goto new_schedule; + } + + callback_context = kmalloc(sizeof *callback_context, GFP_KERNEL); + if (!callback_context) { + err = -ENOMEM; + resched_delay = HZ * 5; + goto new_schedule; + } + callback_context->port = port; + callback_context->dev = dev; + callback_context->block_num = index; + + memset(&guid_info_rec, 0, sizeof (struct ib_sa_guidinfo_rec)); + + guid_info_rec.lid = cpu_to_be16(attr.lid); + guid_info_rec.block_num = index; + + memcpy(guid_info_rec.guid_info_list, rec_det->all_recs, + GUID_REC_SIZE * NUM_ALIAS_GUID_IN_REC); + comp_mask = IB_SA_GUIDINFO_REC_LID | IB_SA_GUIDINFO_REC_BLOCK_NUM | + rec_det->guid_indexes; + + init_completion(&callback_context->done); + spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1); + list_add_tail(&callback_context->list, head); + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1); + + callback_context->query_id = + ib_sa_guid_info_rec_query(dev->sriov.alias_guid.sa_client, + ibdev, port, &guid_info_rec, + comp_mask, rec_det->method, 1000, + GFP_KERNEL, aliasguid_query_handler, + callback_context, + &callback_context->sa_query); + if (callback_context->query_id < 0) { + pr_debug("ib_sa_guid_info_rec_query failed, query_id: " + "%d. will reschedule to the next 1 sec.\n", + callback_context->query_id); + spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1); + list_del(&callback_context->list); + kfree(callback_context); + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1); + resched_delay = 1 * HZ; + err = -EAGAIN; + goto new_schedule; + } + err = 0; + goto out; + +new_schedule: + spin_lock_irqsave(&dev->sriov.going_down_lock, flags); + spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1); + invalidate_guid_record(dev, port, index); + if (!dev->sriov.is_going_down) { + queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq, + &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work, + resched_delay); + } + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1); + spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags); + +out: + return err; +} + +void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port) +{ + int i; + unsigned long flags, flags1; + + pr_debug("port %d\n", port); + + spin_lock_irqsave(&dev->sriov.going_down_lock, flags); + spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1); + for (i = 0; i < NUM_ALIAS_GUID_REC_IN_PORT; i++) + invalidate_guid_record(dev, port, i); + + if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down) { + /* + make sure no work waits in the queue, if the work is already + queued(not on the timer) the cancel will fail. That is not a problem + because we just want the work started. + */ + __cancel_delayed_work(&dev->sriov.alias_guid. + ports_guid[port - 1].alias_guid_work); + queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq, + &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work, + 0); + } + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1); + spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags); +} + +/* The function returns the next record that was + * not configured (or failed to be configured) */ +static int get_next_record_to_update(struct mlx4_ib_dev *dev, u8 port, + struct mlx4_next_alias_guid_work *rec) +{ + int j; + unsigned long flags; + + for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) { + spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags); + if (dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status == + MLX4_GUID_INFO_STATUS_IDLE) { + memcpy(&rec->rec_det, + &dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j], + sizeof (struct mlx4_sriov_alias_guid_info_rec_det)); + rec->port = port; + rec->block_num = j; + dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status = + MLX4_GUID_INFO_STATUS_PENDING; + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags); + return 0; + } + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags); + } + return -ENOENT; +} + +static void set_administratively_guid_record(struct mlx4_ib_dev *dev, int port, + int rec_index, + struct mlx4_sriov_alias_guid_info_rec_det *rec_det) +{ + dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].guid_indexes = + rec_det->guid_indexes; + memcpy(dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].all_recs, + rec_det->all_recs, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE); + dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].status = + rec_det->status; +} + +static void set_all_slaves_guids(struct mlx4_ib_dev *dev, int port) +{ + int j; + struct mlx4_sriov_alias_guid_info_rec_det rec_det ; + + for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT ; j++) { + memset(rec_det.all_recs, 0, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE); + rec_det.guid_indexes = (!j ? 0 : IB_SA_GUIDINFO_REC_GID0) | + IB_SA_GUIDINFO_REC_GID1 | IB_SA_GUIDINFO_REC_GID2 | + IB_SA_GUIDINFO_REC_GID3 | IB_SA_GUIDINFO_REC_GID4 | + IB_SA_GUIDINFO_REC_GID5 | IB_SA_GUIDINFO_REC_GID6 | + IB_SA_GUIDINFO_REC_GID7; + rec_det.status = MLX4_GUID_INFO_STATUS_IDLE; + set_administratively_guid_record(dev, port, j, &rec_det); + } +} + +static void alias_guid_work(struct work_struct *work) +{ + struct delayed_work *delay = to_delayed_work(work); + int ret = 0; + struct mlx4_next_alias_guid_work *rec; + struct mlx4_sriov_alias_guid_port_rec_det *sriov_alias_port = + container_of(delay, struct mlx4_sriov_alias_guid_port_rec_det, + alias_guid_work); + struct mlx4_sriov_alias_guid *sriov_alias_guid = sriov_alias_port->parent; + struct mlx4_ib_sriov *ib_sriov = container_of(sriov_alias_guid, + struct mlx4_ib_sriov, + alias_guid); + struct mlx4_ib_dev *dev = container_of(ib_sriov, struct mlx4_ib_dev, sriov); + + rec = kzalloc(sizeof *rec, GFP_KERNEL); + if (!rec) { + pr_err("alias_guid_work: No Memory\n"); + return; + } + + pr_debug("starting [port: %d]...\n", sriov_alias_port->port + 1); + ret = get_next_record_to_update(dev, sriov_alias_port->port, rec); + if (ret) { + pr_debug("No more records to update.\n"); + goto out; + } + + set_guid_rec(&dev->ib_dev, rec->port + 1, rec->block_num, + &rec->rec_det); + +out: + kfree(rec); +} + + +void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port) +{ + unsigned long flags, flags1; + + if (!mlx4_is_master(dev->dev)) + return; + spin_lock_irqsave(&dev->sriov.going_down_lock, flags); + spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1); + if (!dev->sriov.is_going_down) { + queue_delayed_work(dev->sriov.alias_guid.ports_guid[port].wq, + &dev->sriov.alias_guid.ports_guid[port].alias_guid_work, 0); + } + spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1); + spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags); +} + +void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev) +{ + int i; + struct mlx4_ib_sriov *sriov = &dev->sriov; + struct mlx4_alias_guid_work_context *cb_ctx; + struct mlx4_sriov_alias_guid_port_rec_det *det; + struct ib_sa_query *sa_query; + unsigned long flags; + + for (i = 0 ; i < dev->num_ports; i++) { + cancel_delayed_work(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work); + det = &sriov->alias_guid.ports_guid[i]; + spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags); + while (!list_empty(&det->cb_list)) { + cb_ctx = list_entry(det->cb_list.next, + struct mlx4_alias_guid_work_context, + list); + sa_query = cb_ctx->sa_query; + cb_ctx->sa_query = NULL; + list_del(&cb_ctx->list); + spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags); + ib_sa_cancel_query(cb_ctx->query_id, sa_query); + wait_for_completion(&cb_ctx->done); + kfree(cb_ctx); + spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags); + } + spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags); + } + for (i = 0 ; i < dev->num_ports; i++) { + flush_workqueue(dev->sriov.alias_guid.ports_guid[i].wq); + destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq); + } + ib_sa_unregister_client(dev->sriov.alias_guid.sa_client); + kfree(dev->sriov.alias_guid.sa_client); +} + +int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev) +{ + char alias_wq_name[15]; + int ret = 0; + int i, j, k; + union ib_gid gid; + + if (!mlx4_is_master(dev->dev)) + return 0; + dev->sriov.alias_guid.sa_client = + kzalloc(sizeof *dev->sriov.alias_guid.sa_client, GFP_KERNEL); + if (!dev->sriov.alias_guid.sa_client) + return -ENOMEM; + + ib_sa_register_client(dev->sriov.alias_guid.sa_client); + + spin_lock_init(&dev->sriov.alias_guid.ag_work_lock); + + for (i = 1; i <= dev->num_ports; ++i) { + if (dev->ib_dev.query_gid(&dev->ib_dev , i, 0, &gid)) { + ret = -EFAULT; + goto err_unregister; + } + } + + for (i = 0 ; i < dev->num_ports; i++) { + memset(&dev->sriov.alias_guid.ports_guid[i], 0, + sizeof (struct mlx4_sriov_alias_guid_port_rec_det)); + /*Check if the SM doesn't need to assign the GUIDs*/ + for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) { + if (mlx4_ib_sm_guid_assign) { + dev->sriov.alias_guid.ports_guid[i]. + all_rec_per_port[j]. + ownership = MLX4_GUID_DRIVER_ASSIGN; + continue; + } + dev->sriov.alias_guid.ports_guid[i].all_rec_per_port[j]. + ownership = MLX4_GUID_NONE_ASSIGN; + /*mark each val as it was deleted, + till the sysAdmin will give it valid val*/ + for (k = 0; k < NUM_ALIAS_GUID_IN_REC; k++) { + *(__be64 *)&dev->sriov.alias_guid.ports_guid[i]. + all_rec_per_port[j].all_recs[GUID_REC_SIZE * k] = + cpu_to_be64(MLX4_GUID_FOR_DELETE_VAL); + } + } + INIT_LIST_HEAD(&dev->sriov.alias_guid.ports_guid[i].cb_list); + /*prepare the records, set them to be allocated by sm*/ + for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) + invalidate_guid_record(dev, i + 1, j); + + dev->sriov.alias_guid.ports_guid[i].parent = &dev->sriov.alias_guid; + dev->sriov.alias_guid.ports_guid[i].port = i; + if (mlx4_ib_sm_guid_assign) + set_all_slaves_guids(dev, i); + + snprintf(alias_wq_name, sizeof alias_wq_name, "alias_guid%d", i); + dev->sriov.alias_guid.ports_guid[i].wq = + create_singlethread_workqueue(alias_wq_name); + if (!dev->sriov.alias_guid.ports_guid[i].wq) { + ret = -ENOMEM; + goto err_thread; + } + INIT_DELAYED_WORK(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work, + alias_guid_work); + } + return 0; + +err_thread: + for (--i; i >= 0; i--) { + destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq); + dev->sriov.alias_guid.ports_guid[i].wq = NULL; + } + +err_unregister: + ib_sa_unregister_client(dev->sriov.alias_guid.sa_client); + kfree(dev->sriov.alias_guid.sa_client); + dev->sriov.alias_guid.sa_client = NULL; + pr_err("init_alias_guid_service: Failed. (ret:%d)\n", ret); + return ret; +} diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 2f13894299ee..b8cb25ebce50 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -791,8 +791,10 @@ void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev) static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num) { - /* re-configure the mcg's */ + /* re-configure the alias-guid and mcg's */ if (mlx4_is_master(dev->dev)) { + mlx4_ib_invalidate_all_guid_record(dev, port_num); + if (!dev->sriov.is_going_down) mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0); } @@ -1808,9 +1810,20 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev) return 0; } + err = mlx4_ib_init_alias_guid_service(dev); + if (err) { + mlx4_ib_warn(&dev->ib_dev, "Failed init alias guid process.\n"); + goto paravirt_err; + } + mlx4_ib_warn(&dev->ib_dev, "initializing demux service for %d qp1 clients\n", dev->dev->caps.sqp_demux); for (i = 0; i < dev->num_ports; i++) { + union ib_gid gid; + err = __mlx4_ib_query_gid(&dev->ib_dev, i + 1, 0, &gid, 1); + if (err) + goto demux_err; + dev->sriov.demux[i].guid_cache[0] = gid.global.interface_id; err = alloc_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1, &dev->sriov.sqps[i]); if (err) @@ -1828,6 +1841,9 @@ demux_err: mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]); --i; } + mlx4_ib_destroy_alias_guid_service(dev); + +paravirt_err: mlx4_ib_cm_paravirt_clean(dev, -1); return err; @@ -1854,5 +1870,6 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev) } mlx4_ib_cm_paravirt_clean(dev, -1); + mlx4_ib_destroy_alias_guid_service(dev); } } diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index b959fe4665dd..7d97578fbbaa 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -59,6 +59,10 @@ MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver"); MODULE_LICENSE("Dual BSD/GPL"); MODULE_VERSION(DRV_VERSION); +int mlx4_ib_sm_guid_assign = 1; +module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444); +MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)"); + static const char mlx4_ib_version[] = DRV_NAME ": Mellanox ConnectX InfiniBand driver v" DRV_VERSION " (" DRV_RELDATE ")\n"; @@ -349,12 +353,15 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port, return __mlx4_ib_query_port(ibdev, port, props, 0); } -static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, - union ib_gid *gid) +int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, + union ib_gid *gid, int netw_view) { struct ib_smp *in_mad = NULL; struct ib_smp *out_mad = NULL; int err = -ENOMEM; + struct mlx4_ib_dev *dev = to_mdev(ibdev); + int clear = 0; + int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); @@ -365,18 +372,29 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, in_mad->attr_id = IB_SMP_ATTR_PORT_INFO; in_mad->attr_mod = cpu_to_be32(port); - err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port, - NULL, NULL, in_mad, out_mad); + if (mlx4_is_mfunc(dev->dev) && netw_view) + mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; + + err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad); if (err) goto out; memcpy(gid->raw, out_mad->data + 8, 8); + if (mlx4_is_mfunc(dev->dev) && !netw_view) { + if (index) { + /* For any index > 0, return the null guid */ + err = 0; + clear = 1; + goto out; + } + } + init_query_mad(in_mad); in_mad->attr_id = IB_SMP_ATTR_GUID_INFO; in_mad->attr_mod = cpu_to_be32(index / 8); - err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, port, + err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad); if (err) goto out; @@ -384,6 +402,8 @@ static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8); out: + if (clear) + memset(gid->raw + 8, 0, 8); kfree(in_mad); kfree(out_mad); return err; @@ -403,7 +423,7 @@ static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, union ib_gid *gid) { if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND) - return __mlx4_ib_query_gid(ibdev, port, index, gid); + return __mlx4_ib_query_gid(ibdev, port, index, gid, 0); else return iboe_query_gid(ibdev, port, index, gid); } @@ -1566,6 +1586,11 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, case MLX4_DEV_EVENT_PORT_UP: if (p > ibdev->num_ports) return; + if (mlx4_is_master(dev) && + rdma_port_get_link_layer(&ibdev->ib_dev, p) == + IB_LINK_LAYER_INFINIBAND) { + mlx4_ib_invalidate_all_guid_record(ibdev, p); + } ibev.event = IB_EVENT_PORT_ACTIVE; break; diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index 7476e2439f6b..f3f75f8229a7 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -42,6 +42,7 @@ #include #include #include +#include #include #include @@ -64,6 +65,9 @@ enum { #define MLX4_IB_SQ_HEADROOM(shift) ((MLX4_IB_MAX_HEADROOM >> (shift)) + 1) #define MLX4_IB_SQ_MAX_SPARE (MLX4_IB_SQ_HEADROOM(MLX4_IB_SQ_MIN_WQE_SHIFT)) +/*module param to indicate if SM assigns the alias_GUID*/ +extern int mlx4_ib_sm_guid_assign; + struct mlx4_ib_ucontext { struct ib_ucontext ibucontext; struct mlx4_uar uar; @@ -277,6 +281,57 @@ struct mlx4_ib_ah { union mlx4_ext_av av; }; +/****************************************/ +/* alias guid support */ +/****************************************/ +#define NUM_PORT_ALIAS_GUID 2 +#define NUM_ALIAS_GUID_IN_REC 8 +#define NUM_ALIAS_GUID_REC_IN_PORT 16 +#define GUID_REC_SIZE 8 +#define NUM_ALIAS_GUID_PER_PORT 128 +#define MLX4_NOT_SET_GUID (0x00LL) +#define MLX4_GUID_FOR_DELETE_VAL (~(0x00LL)) + +enum mlx4_guid_alias_rec_status { + MLX4_GUID_INFO_STATUS_IDLE, + MLX4_GUID_INFO_STATUS_SET, + MLX4_GUID_INFO_STATUS_PENDING, +}; + +enum mlx4_guid_alias_rec_ownership { + MLX4_GUID_DRIVER_ASSIGN, + MLX4_GUID_SYSADMIN_ASSIGN, + MLX4_GUID_NONE_ASSIGN, /*init state of each record*/ +}; + +enum mlx4_guid_alias_rec_method { + MLX4_GUID_INFO_RECORD_SET = IB_MGMT_METHOD_SET, + MLX4_GUID_INFO_RECORD_DELETE = IB_SA_METHOD_DELETE, +}; + +struct mlx4_sriov_alias_guid_info_rec_det { + u8 all_recs[GUID_REC_SIZE * NUM_ALIAS_GUID_IN_REC]; + ib_sa_comp_mask guid_indexes; /*indicates what from the 8 records are valid*/ + enum mlx4_guid_alias_rec_status status; /*indicates the administraively status of the record.*/ + u8 method; /*set or delete*/ + enum mlx4_guid_alias_rec_ownership ownership; /*indicates who assign that alias_guid record*/ +}; + +struct mlx4_sriov_alias_guid_port_rec_det { + struct mlx4_sriov_alias_guid_info_rec_det all_rec_per_port[NUM_ALIAS_GUID_REC_IN_PORT]; + struct workqueue_struct *wq; + struct delayed_work alias_guid_work; + u8 port; + struct mlx4_sriov_alias_guid *parent; + struct list_head cb_list; +}; + +struct mlx4_sriov_alias_guid { + struct mlx4_sriov_alias_guid_port_rec_det ports_guid[MLX4_MAX_PORTS]; + spinlock_t ag_work_lock; + struct ib_sa_client *sa_client; +}; + struct mlx4_ib_demux_work { struct work_struct work; struct mlx4_ib_dev *dev; @@ -349,6 +404,8 @@ struct mlx4_ib_sriov { spinlock_t going_down_lock; int is_going_down; + struct mlx4_sriov_alias_guid alias_guid; + /* CM paravirtualization fields */ struct list_head cm_list; spinlock_t id_map_lock; @@ -555,6 +612,9 @@ int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port, int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey, int netw_view); +int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, + union ib_gid *gid, int netw_view); + int mlx4_ib_resolve_grh(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah_attr, u8 *mac, int *is_mcast, u8 port); @@ -606,4 +666,18 @@ int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev); void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave_id); +/* alias guid support */ +void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port); +int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev); +void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev); +void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port); + +void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev, + int block_num, + u8 port_num, u8 *p_data); + +void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, + int block_num, u8 port_num, + u8 *p_data); + #endif /* MLX4_IB_H */ diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index a13d8a69b3bc..ea5c884ab899 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -674,8 +674,10 @@ static int get_full_pkey_table(struct mlx4_dev *dev, u8 port, u16 *table, static enum ib_port_state vf_port_state(struct mlx4_dev *dev, int port, int vf) { - /* will be modified when add alias_guid feature */ - return IB_PORT_DOWN; + if (mlx4_get_slave_port_state(dev, vf, port) == SLAVE_PORT_UP) + return IB_PORT_ACTIVE; + else + return IB_PORT_DOWN; } static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave, From 2a4fae148cf4b60e73faf0a427302697917409d9 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:50 +0000 Subject: [PATCH 21/45] IB/mlx4: Propagate P_Key and guid change port management events to slaves P_Key change and guid change events are not of interest to all slaves, but only to those slaves which "see" the table slots whose contents have change. For example, if the guid at port 1, index 5 has changed in the PPF, we wish to propagate the gid-change event only to the function which has that guid index mapped to its port/guid table (in this case it is slave #5). Other functions should not get the event, since the event does not affect them. Similarly with P_Keys -- P_Key change events are forwarded only to slaves which have that P_Key index mapped to their virtual P_Key table. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 162 +++++++++++++++++++++++- drivers/net/ethernet/mellanox/mlx4/fw.c | 6 + 2 files changed, 161 insertions(+), 7 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index b8cb25ebce50..591c2891159b 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -54,6 +54,15 @@ enum { #define MLX4_TUN_IS_RECV(a) (((a) >> MLX4_TUN_SEND_WRID_SHIFT) & 0x1) #define MLX4_TUN_WRID_QPN(a) (((a) >> MLX4_TUN_QPN_SHIFT) & 0x3) + /* Port mgmt change event handling */ + +#define GET_BLK_PTR_FROM_EQE(eqe) be32_to_cpu(eqe->event.port_mgmt_change.params.tbl_change_info.block_ptr) +#define GET_MASK_FROM_EQE(eqe) be32_to_cpu(eqe->event.port_mgmt_change.params.tbl_change_info.tbl_entries_mask) +#define NUM_IDX_IN_PKEY_TBL_BLK 32 +#define GUID_TBL_ENTRY_SIZE 8 /* size in bytes */ +#define GUID_TBL_BLK_NUM_ENTRIES 8 +#define GUID_TBL_BLK_SIZE (GUID_TBL_ENTRY_SIZE * GUID_TBL_BLK_NUM_ENTRIES) + struct mlx4_mad_rcv_buf { struct ib_grh grh; u8 payload[256]; @@ -76,6 +85,9 @@ struct mlx4_rcv_tunnel_mad { } __packed; static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num); +static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num); +static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num, + int block, u32 change_bitmap); __be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx) { @@ -220,8 +232,7 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad, handle_client_rereg_event(dev, port_num); if (prev_lid != lid) - mlx4_ib_dispatch_event(dev, port_num, - IB_EVENT_LID_CHANGE); + handle_lid_change_event(dev, port_num); break; case IB_SMP_ATTR_PKEY_TABLE: @@ -231,6 +242,9 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad, break; } + /* at this point, we are running in the master. + * Slaves do not receive SMPs. + */ bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod) & 0xFFFF; base = (__be16 *) &(((struct ib_smp *)mad)->data[0]); pkey_change_bitmap = 0; @@ -248,10 +262,13 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad, "block=0x%x, change_bitmap=0x%x\n", port_num, bn, pkey_change_bitmap); - if (pkey_change_bitmap) + if (pkey_change_bitmap) { mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_PKEY_CHANGE); - + if (!dev->sriov.is_going_down) + __propagate_pkey_ev(dev, port_num, bn, + pkey_change_bitmap); + } break; case IB_SMP_ATTR_GUID_INFO: @@ -259,12 +276,56 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad, if (!mlx4_is_master(dev->dev)) mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_GID_CHANGE); + /*if master, notify relevant slaves*/ + if (mlx4_is_master(dev->dev) && + !dev->sriov.is_going_down) { + bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod); + mlx4_ib_update_cache_on_guid_change(dev, bn, port_num, + (u8 *)(&((struct ib_smp *)mad)->data)); + mlx4_ib_notify_slaves_on_guid_change(dev, bn, port_num, + (u8 *)(&((struct ib_smp *)mad)->data)); + } break; + default: break; } } +static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num, + int block, u32 change_bitmap) +{ + int i, ix, slave, err; + int have_event = 0; + + for (slave = 0; slave < dev->dev->caps.sqp_demux; slave++) { + if (slave == mlx4_master_func_num(dev->dev)) + continue; + if (!mlx4_is_slave_active(dev->dev, slave)) + continue; + + have_event = 0; + for (i = 0; i < 32; i++) { + if (!(change_bitmap & (1 << i))) + continue; + for (ix = 0; + ix < dev->dev->caps.pkey_table_len[port_num]; ix++) { + if (dev->pkeys.virt2phys_pkey[slave][port_num - 1] + [ix] == i + 32 * block) { + err = mlx4_gen_pkey_eqe(dev->dev, slave, port_num); + pr_debug("propagate_pkey_ev: slave %d," + " port %d, ix %d (%d)\n", + slave, port_num, ix, err); + have_event = 1; + break; + } + } + if (have_event) + break; + } + } +} + static void node_desc_override(struct ib_device *dev, struct ib_mad *mad) { @@ -789,18 +850,90 @@ void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev) } } +static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num) +{ + mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_LID_CHANGE); + + if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down) + mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num, + MLX4_EQ_PORT_INFO_LID_CHANGE_MASK); +} + static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num) { /* re-configure the alias-guid and mcg's */ if (mlx4_is_master(dev->dev)) { mlx4_ib_invalidate_all_guid_record(dev, port_num); - if (!dev->sriov.is_going_down) + if (!dev->sriov.is_going_down) { mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0); + mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num, + MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK); + } } mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_CLIENT_REREGISTER); } +static void propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num, + struct mlx4_eqe *eqe) +{ + __propagate_pkey_ev(dev, port_num, GET_BLK_PTR_FROM_EQE(eqe), + GET_MASK_FROM_EQE(eqe)); +} + +static void handle_slaves_guid_change(struct mlx4_ib_dev *dev, u8 port_num, + u32 guid_tbl_blk_num, u32 change_bitmap) +{ + struct ib_smp *in_mad = NULL; + struct ib_smp *out_mad = NULL; + u16 i; + + if (!mlx4_is_mfunc(dev->dev) || !mlx4_is_master(dev->dev)) + return; + + in_mad = kmalloc(sizeof *in_mad, GFP_KERNEL); + out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); + if (!in_mad || !out_mad) { + mlx4_ib_warn(&dev->ib_dev, "failed to allocate memory for guid info mads\n"); + goto out; + } + + guid_tbl_blk_num *= 4; + + for (i = 0; i < 4; i++) { + if (change_bitmap && (!((change_bitmap >> (8 * i)) & 0xff))) + continue; + memset(in_mad, 0, sizeof *in_mad); + memset(out_mad, 0, sizeof *out_mad); + + in_mad->base_version = 1; + in_mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED; + in_mad->class_version = 1; + in_mad->method = IB_MGMT_METHOD_GET; + in_mad->attr_id = IB_SMP_ATTR_GUID_INFO; + in_mad->attr_mod = cpu_to_be32(guid_tbl_blk_num + i); + + if (mlx4_MAD_IFC(dev, + MLX4_MAD_IFC_IGNORE_KEYS | MLX4_MAD_IFC_NET_VIEW, + port_num, NULL, NULL, in_mad, out_mad)) { + mlx4_ib_warn(&dev->ib_dev, "Failed in get GUID INFO MAD_IFC\n"); + goto out; + } + + mlx4_ib_update_cache_on_guid_change(dev, guid_tbl_blk_num + i, + port_num, + (u8 *)(&((struct ib_smp *)out_mad)->data)); + mlx4_ib_notify_slaves_on_guid_change(dev, guid_tbl_blk_num + i, + port_num, + (u8 *)(&((struct ib_smp *)out_mad)->data)); + } + +out: + kfree(in_mad); + kfree(out_mad); + return; +} + void handle_port_mgmt_change_event(struct work_struct *work) { struct ib_event_work *ew = container_of(work, struct ib_event_work, work); @@ -808,6 +941,8 @@ void handle_port_mgmt_change_event(struct work_struct *work) struct mlx4_eqe *eqe = &(ew->ib_eqe); u8 port = eqe->event.port_mgmt_change.port; u32 changed_attr; + u32 tbl_block; + u32 change_bitmap; switch (eqe->subtype) { case MLX4_DEV_PMC_SUBTYPE_PORT_INFO: @@ -823,11 +958,16 @@ void handle_port_mgmt_change_event(struct work_struct *work) /* Check if it is a lid change event */ if (changed_attr & MLX4_EQ_PORT_INFO_LID_CHANGE_MASK) - mlx4_ib_dispatch_event(dev, port, IB_EVENT_LID_CHANGE); + handle_lid_change_event(dev, port); /* Generate GUID changed event */ - if (changed_attr & MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK) + if (changed_attr & MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK) { mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE); + /*if master, notify all slaves*/ + if (mlx4_is_master(dev->dev)) + mlx4_gen_slaves_port_mgt_ev(dev->dev, port, + MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK); + } if (changed_attr & MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK) handle_client_rereg_event(dev, port); @@ -835,11 +975,19 @@ void handle_port_mgmt_change_event(struct work_struct *work) case MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE: mlx4_ib_dispatch_event(dev, port, IB_EVENT_PKEY_CHANGE); + if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down) + propagate_pkey_ev(dev, port, eqe); break; case MLX4_DEV_PMC_SUBTYPE_GUID_INFO: /* paravirtualized master's guid is guid 0 -- does not change */ if (!mlx4_is_master(dev->dev)) mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE); + /*if master, notify relevant slaves*/ + else if (!dev->sriov.is_going_down) { + tbl_block = GET_BLK_PTR_FROM_EQE(eqe); + change_bitmap = GET_MASK_FROM_EQE(eqe); + handle_slaves_guid_change(dev, port, tbl_block, change_bitmap); + } break; default: pr_warn("Unsupported subtype 0x%x for " diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c index 35be90178bef..b78803937897 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.c +++ b/drivers/net/ethernet/mellanox/mlx4/fw.c @@ -729,6 +729,7 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_cmd_mailbox *outbox, struct mlx4_cmd_info *cmd) { + u64 flags; int err = 0; u8 field; @@ -737,6 +738,11 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave, if (err) return err; + /* add port mng change event capability unconditionally to slaves */ + MLX4_GET(flags, outbox->buf, QUERY_DEV_CAP_EXT_FLAGS_OFFSET); + flags |= MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV; + MLX4_PUT(outbox->buf, flags, QUERY_DEV_CAP_EXT_FLAGS_OFFSET); + /* For guests, report Blueflame disabled */ MLX4_GET(field, outbox->buf, QUERY_DEV_CAP_BF_OFFSET); field &= 0x7f; From c1e7e466120b80ce49e91af0c9da1ce6dee4844a Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:51 +0000 Subject: [PATCH 22/45] IB/mlx4: Add iov directory in sysfs under the ib device This directory is added only for the master -- slaves do not have it. The sysfs iov directory is used to manage and examine the port P_Key and guid paravirtualization. Under iov/ports, the administrator may examine the gid and P_Key tables as they are present in the device (and as are seen in the "network view" presented to the SM). Under the iov/ directories, the admin may map the index numbers in the physical tables (as under iov/ports) to the paravirtualized index numbers that guests see. For example, if the administrator, for port 1 on guest 2 maps physical pkey index 10 to virtual index 1, then that guest, whenever it uses its pkey index 1, will actually be using the real pkey index 10. Based on patch from Erez Shitrit Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/Makefile | 2 +- drivers/infiniband/hw/mlx4/alias_GUID.c | 6 +- drivers/infiniband/hw/mlx4/mad.c | 9 + drivers/infiniband/hw/mlx4/mcg.c | 67 ++ drivers/infiniband/hw/mlx4/mlx4_ib.h | 43 ++ drivers/infiniband/hw/mlx4/sysfs.c | 794 ++++++++++++++++++++++++ 6 files changed, 917 insertions(+), 4 deletions(-) create mode 100644 drivers/infiniband/hw/mlx4/sysfs.c diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile index 31d4c8aac679..f4213b3a8fe1 100644 --- a/drivers/infiniband/hw/mlx4/Makefile +++ b/drivers/infiniband/hw/mlx4/Makefile @@ -1,3 +1,3 @@ obj-$(CONFIG_MLX4_INFINIBAND) += mlx4_ib.o -mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o alias_GUID.o +mlx4_ib-y := ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o alias_GUID.o sysfs.o diff --git a/drivers/infiniband/hw/mlx4/alias_GUID.c b/drivers/infiniband/hw/mlx4/alias_GUID.c index ef6d356927c3..0fcd5cd6f3ee 100644 --- a/drivers/infiniband/hw/mlx4/alias_GUID.c +++ b/drivers/infiniband/hw/mlx4/alias_GUID.c @@ -113,7 +113,7 @@ static __be64 get_cached_alias_guid(struct mlx4_ib_dev *dev, int port, int index } -static ib_sa_comp_mask get_aguid_comp_mask_from_ix(int index) +ib_sa_comp_mask mlx4_ib_get_aguid_comp_mask_from_ix(int index) { return IB_SA_COMP_MASK(4 + index); } @@ -259,7 +259,7 @@ static void aliasguid_query_handler(int status, /* Mark the record as not assigned, and let it * be sent again in the next work sched.*/ rec->status = MLX4_GUID_INFO_STATUS_IDLE; - rec->guid_indexes |= get_aguid_comp_mask_from_ix(i); + rec->guid_indexes |= mlx4_ib_get_aguid_comp_mask_from_ix(i); } } else { /* properly assigned record. */ @@ -337,7 +337,7 @@ static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index) MLX4_GUID_NONE_ASSIGN == dev->sriov.alias_guid. ports_guid[port - 1].all_rec_per_port[index].ownership) continue; - comp_mask |= get_aguid_comp_mask_from_ix(i); + comp_mask |= mlx4_ib_get_aguid_comp_mask_from_ix(i); } dev->sriov.alias_guid.ports_guid[port - 1]. all_rec_per_port[index].guid_indexes = comp_mask; diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 591c2891159b..b689dbd6d8f6 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -1963,6 +1963,11 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev) mlx4_ib_warn(&dev->ib_dev, "Failed init alias guid process.\n"); goto paravirt_err; } + err = mlx4_ib_device_register_sysfs(dev); + if (err) { + mlx4_ib_warn(&dev->ib_dev, "Failed to register sysfs\n"); + goto sysfs_err; + } mlx4_ib_warn(&dev->ib_dev, "initializing demux service for %d qp1 clients\n", dev->dev->caps.sqp_demux); @@ -1989,6 +1994,9 @@ demux_err: mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]); --i; } + mlx4_ib_device_unregister_sysfs(dev); + +sysfs_err: mlx4_ib_destroy_alias_guid_service(dev); paravirt_err: @@ -2019,5 +2027,6 @@ void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev) mlx4_ib_cm_paravirt_clean(dev, -1); mlx4_ib_destroy_alias_guid_service(dev); + mlx4_ib_device_unregister_sysfs(dev); } } diff --git a/drivers/infiniband/hw/mlx4/mcg.c b/drivers/infiniband/hw/mlx4/mcg.c index 1ee2e3a3347a..3c3b54c3fdd9 100644 --- a/drivers/infiniband/hw/mlx4/mcg.c +++ b/drivers/infiniband/hw/mlx4/mcg.c @@ -110,6 +110,7 @@ struct mcast_group { __be64 last_req_tid; char name[33]; /* MGID string */ + struct device_attribute dentry; /* refcount is the reference count for the following: 1. Each queued request @@ -445,6 +446,8 @@ static int release_group(struct mcast_group *group, int from_timeout_handler) } nzgroup = memcmp(&group->rec.mgid, &mgid0, sizeof mgid0); + if (nzgroup) + del_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr); if (!list_empty(&group->pending_list)) mcg_warn_group(group, "releasing a group with non empty pending list\n"); if (nzgroup) @@ -769,6 +772,7 @@ static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx } atomic_inc(&group->refcount); + add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr); mutex_unlock(&group->lock); mutex_unlock(&ctx->mcg_table_lock); return group; @@ -796,6 +800,9 @@ static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx return NULL; } +static ssize_t sysfs_show_group(struct device *dev, + struct device_attribute *attr, char *buf); + static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx, union ib_gid *mgid, int create, gfp_t gfp_mask) @@ -830,6 +837,11 @@ static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx, sprintf(group->name, "%016llx%016llx", be64_to_cpu(group->rec.mgid.global.subnet_prefix), be64_to_cpu(group->rec.mgid.global.interface_id)); + sysfs_attr_init(&group->dentry.attr); + group->dentry.show = sysfs_show_group; + group->dentry.store = NULL; + group->dentry.attr.name = group->name; + group->dentry.attr.mode = 0400; group->state = MCAST_IDLE; if (is_mgid0) { @@ -844,6 +856,8 @@ static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx, return ERR_PTR(-EINVAL); } + add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr); + found: atomic_inc(&group->refcount); return group; @@ -969,6 +983,58 @@ int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port, } } +static ssize_t sysfs_show_group(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct mcast_group *group = + container_of(attr, struct mcast_group, dentry); + struct mcast_req *req = NULL; + char pending_str[40]; + char state_str[40]; + ssize_t len = 0; + int f; + + if (group->state == MCAST_IDLE) + sprintf(state_str, "%s", get_state_string(group->state)); + else + sprintf(state_str, "%s(TID=0x%llx)", + get_state_string(group->state), + be64_to_cpu(group->last_req_tid)); + if (list_empty(&group->pending_list)) { + sprintf(pending_str, "No"); + } else { + req = list_first_entry(&group->pending_list, struct mcast_req, group_list); + sprintf(pending_str, "Yes(TID=0x%llx)", + be64_to_cpu(req->sa_mad.mad_hdr.tid)); + } + len += sprintf(buf + len, "%1d [%02d,%02d,%02d] %4d %4s %5s ", + group->rec.scope_join_state & 0xf, + group->members[2], group->members[1], group->members[0], + atomic_read(&group->refcount), + pending_str, + state_str); + for (f = 0; f < MAX_VFS; ++f) + if (group->func[f].state == MCAST_MEMBER) + len += sprintf(buf + len, "%d[%1x] ", + f, group->func[f].join_state); + + len += sprintf(buf + len, "\t\t(%4hx %4x %2x %2x %2x %2x %2x " + "%4x %4x %2x %2x)\n", + be16_to_cpu(group->rec.pkey), + be32_to_cpu(group->rec.qkey), + (group->rec.mtusel_mtu & 0xc0) >> 6, + group->rec.mtusel_mtu & 0x3f, + group->rec.tclass, + (group->rec.ratesel_rate & 0xc0) >> 6, + group->rec.ratesel_rate & 0x3f, + (be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0xf0000000) >> 28, + (be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x0fffff00) >> 8, + be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x000000ff, + group->rec.proxy_join); + + return len; +} + int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx) { char name[20]; @@ -995,6 +1061,7 @@ static void force_clean_group(struct mcast_group *group) list_del(&req->group_list); kfree(req); } + del_sysfs_port_mcg_attr(group->demux->dev, group->demux->port, &group->dentry.attr); rb_erase(&group->node, &group->demux->mcg_table); kfree(group); } diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index f3f75f8229a7..e57a220a4d55 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -427,6 +427,35 @@ struct pkey_mgt { struct kobject *device_parent[MLX4_MFUNC_MAX]; }; +struct mlx4_ib_iov_sysfs_attr { + void *ctx; + struct kobject *kobj; + unsigned long data; + u32 entry_num; + char name[15]; + struct device_attribute dentry; + struct device *dev; +}; + +struct mlx4_ib_iov_sysfs_attr_ar { + struct mlx4_ib_iov_sysfs_attr dentries[3 * NUM_ALIAS_GUID_PER_PORT + 1]; +}; + +struct mlx4_ib_iov_port { + char name[100]; + u8 num; + struct mlx4_ib_dev *dev; + struct list_head list; + struct mlx4_ib_iov_sysfs_attr_ar *dentr_ar; + struct ib_port_attr attr; + struct kobject *cur_port; + struct kobject *admin_alias_parent; + struct kobject *gids_parent; + struct kobject *pkeys_parent; + struct kobject *mcgs_parent; + struct mlx4_ib_iov_sysfs_attr mcg_dentry; +}; + struct mlx4_ib_dev { struct ib_device ib_dev; struct mlx4_dev *dev; @@ -448,6 +477,10 @@ struct mlx4_ib_dev { int counters[MLX4_MAX_PORTS]; int *eq_table; int eq_added; + struct kobject *iov_parent; + struct kobject *ports_parent; + struct kobject *dev_ports_parent[MLX4_MFUNC_MAX]; + struct mlx4_ib_iov_port iov_ports[MLX4_MAX_PORTS]; struct pkey_mgt pkeys; }; @@ -680,4 +713,14 @@ void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, int block_num, u8 port_num, u8 *p_data); +int add_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num, + struct attribute *attr); +void del_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num, + struct attribute *attr); +ib_sa_comp_mask mlx4_ib_get_aguid_comp_mask_from_ix(int index); + +int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *device) ; + +void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device); + #endif /* MLX4_IB_H */ diff --git a/drivers/infiniband/hw/mlx4/sysfs.c b/drivers/infiniband/hw/mlx4/sysfs.c new file mode 100644 index 000000000000..5b2a01dfb907 --- /dev/null +++ b/drivers/infiniband/hw/mlx4/sysfs.c @@ -0,0 +1,794 @@ +/* + * Copyright (c) 2012 Mellanox Technologies. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +/*#include "core_priv.h"*/ +#include "mlx4_ib.h" +#include +#include +#include + +#include +/*show_admin_alias_guid returns the administratively assigned value of that GUID. + * Values returned in buf parameter string: + * 0 - requests opensm to assign a value. + * ffffffffffffffff - delete this entry. + * other - value assigned by administrator. + */ +static ssize_t show_admin_alias_guid(struct device *dev, + struct device_attribute *attr, char *buf) +{ + int record_num;/*0-15*/ + int guid_index_in_rec; /*0 - 7*/ + struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry = + container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry); + struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx; + struct mlx4_ib_dev *mdev = port->dev; + + record_num = mlx4_ib_iov_dentry->entry_num / 8 ; + guid_index_in_rec = mlx4_ib_iov_dentry->entry_num % 8 ; + + return sprintf(buf, "%llx\n", + be64_to_cpu(*(__be64 *)&mdev->sriov.alias_guid. + ports_guid[port->num - 1]. + all_rec_per_port[record_num]. + all_recs[8 * guid_index_in_rec])); +} + +/* store_admin_alias_guid stores the (new) administratively assigned value of that GUID. + * Values in buf parameter string: + * 0 - requests opensm to assign a value. + * 0xffffffffffffffff - delete this entry. + * other - guid value assigned by the administrator. + */ +static ssize_t store_admin_alias_guid(struct device *dev, + struct device_attribute *attr, + const char *buf, size_t count) +{ + int record_num;/*0-15*/ + int guid_index_in_rec; /*0 - 7*/ + struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry = + container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry); + struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx; + struct mlx4_ib_dev *mdev = port->dev; + u64 sysadmin_ag_val; + + record_num = mlx4_ib_iov_dentry->entry_num / 8; + guid_index_in_rec = mlx4_ib_iov_dentry->entry_num % 8; + if (0 == record_num && 0 == guid_index_in_rec) { + pr_err("GUID 0 block 0 is RO\n"); + return count; + } + sscanf(buf, "%llx", &sysadmin_ag_val); + *(__be64 *)&mdev->sriov.alias_guid.ports_guid[port->num - 1]. + all_rec_per_port[record_num]. + all_recs[GUID_REC_SIZE * guid_index_in_rec] = + cpu_to_be64(sysadmin_ag_val); + + /* Change the state to be pending for update */ + mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].status + = MLX4_GUID_INFO_STATUS_IDLE ; + + mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].method + = MLX4_GUID_INFO_RECORD_SET; + + switch (sysadmin_ag_val) { + case MLX4_GUID_FOR_DELETE_VAL: + mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].method + = MLX4_GUID_INFO_RECORD_DELETE; + mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership + = MLX4_GUID_SYSADMIN_ASSIGN; + break; + /* The sysadmin requests the SM to re-assign */ + case MLX4_NOT_SET_GUID: + mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership + = MLX4_GUID_DRIVER_ASSIGN; + break; + /* The sysadmin requests a specific value.*/ + default: + mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership + = MLX4_GUID_SYSADMIN_ASSIGN; + break; + } + + /* set the record index */ + mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].guid_indexes + = mlx4_ib_get_aguid_comp_mask_from_ix(guid_index_in_rec); + + mlx4_ib_init_alias_guid_work(mdev, port->num - 1); + + return count; +} + +static ssize_t show_port_gid(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry = + container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry); + struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx; + struct mlx4_ib_dev *mdev = port->dev; + union ib_gid gid; + ssize_t ret; + + ret = __mlx4_ib_query_gid(&mdev->ib_dev, port->num, + mlx4_ib_iov_dentry->entry_num, &gid, 1); + if (ret) + return ret; + ret = sprintf(buf, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x\n", + be16_to_cpu(((__be16 *) gid.raw)[0]), + be16_to_cpu(((__be16 *) gid.raw)[1]), + be16_to_cpu(((__be16 *) gid.raw)[2]), + be16_to_cpu(((__be16 *) gid.raw)[3]), + be16_to_cpu(((__be16 *) gid.raw)[4]), + be16_to_cpu(((__be16 *) gid.raw)[5]), + be16_to_cpu(((__be16 *) gid.raw)[6]), + be16_to_cpu(((__be16 *) gid.raw)[7])); + return ret; +} + +static ssize_t show_phys_port_pkey(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry = + container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry); + struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx; + struct mlx4_ib_dev *mdev = port->dev; + u16 pkey; + ssize_t ret; + + ret = __mlx4_ib_query_pkey(&mdev->ib_dev, port->num, + mlx4_ib_iov_dentry->entry_num, &pkey, 1); + if (ret) + return ret; + + return sprintf(buf, "0x%04x\n", pkey); +} + +#define DENTRY_REMOVE(_dentry) \ +do { \ + sysfs_remove_file((_dentry)->kobj, &(_dentry)->dentry.attr); \ +} while (0); + +static int create_sysfs_entry(void *_ctx, struct mlx4_ib_iov_sysfs_attr *_dentry, + char *_name, struct kobject *_kobj, + ssize_t (*show)(struct device *dev, + struct device_attribute *attr, + char *buf), + ssize_t (*store)(struct device *dev, + struct device_attribute *attr, + const char *buf, size_t count) + ) +{ + int ret = 0; + struct mlx4_ib_iov_sysfs_attr *vdentry = _dentry; + + vdentry->ctx = _ctx; + vdentry->dentry.show = show; + vdentry->dentry.store = store; + sysfs_attr_init(&vdentry->dentry.attr); + vdentry->dentry.attr.name = vdentry->name; + vdentry->dentry.attr.mode = 0; + vdentry->kobj = _kobj; + snprintf(vdentry->name, 15, "%s", _name); + + if (vdentry->dentry.store) + vdentry->dentry.attr.mode |= S_IWUSR; + + if (vdentry->dentry.show) + vdentry->dentry.attr.mode |= S_IRUGO; + + ret = sysfs_create_file(vdentry->kobj, &vdentry->dentry.attr); + if (ret) { + pr_err("failed to create %s\n", vdentry->dentry.attr.name); + vdentry->ctx = NULL; + return ret; + } + + return ret; +} + +int add_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num, + struct attribute *attr) +{ + struct mlx4_ib_iov_port *port = &device->iov_ports[port_num - 1]; + int ret; + + ret = sysfs_create_file(port->mcgs_parent, attr); + if (ret) + pr_err("failed to create %s\n", attr->name); + + return ret; +} + +void del_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num, + struct attribute *attr) +{ + struct mlx4_ib_iov_port *port = &device->iov_ports[port_num - 1]; + + sysfs_remove_file(port->mcgs_parent, attr); +} + +static int add_port_entries(struct mlx4_ib_dev *device, int port_num) +{ + int i; + char buff[10]; + struct mlx4_ib_iov_port *port = NULL; + int ret = 0 ; + struct ib_port_attr attr; + + /* get the physical gid and pkey table sizes.*/ + ret = __mlx4_ib_query_port(&device->ib_dev, port_num, &attr, 1); + if (ret) + goto err; + + port = &device->iov_ports[port_num - 1]; + port->dev = device; + port->num = port_num; + /* Directory structure: + * iov - + * port num - + * admin_guids + * gids (operational) + * mcg_table + */ + port->dentr_ar = kzalloc(sizeof (struct mlx4_ib_iov_sysfs_attr_ar), + GFP_KERNEL); + if (!port->dentr_ar) { + ret = -ENOMEM; + goto err; + } + sprintf(buff, "%d", port_num); + port->cur_port = kobject_create_and_add(buff, + kobject_get(device->ports_parent)); + if (!port->cur_port) { + ret = -ENOMEM; + goto kobj_create_err; + } + /* admin GUIDs */ + port->admin_alias_parent = kobject_create_and_add("admin_guids", + kobject_get(port->cur_port)); + if (!port->admin_alias_parent) { + ret = -ENOMEM; + goto err_admin_guids; + } + for (i = 0 ; i < attr.gid_tbl_len; i++) { + sprintf(buff, "%d", i); + port->dentr_ar->dentries[i].entry_num = i; + ret = create_sysfs_entry(port, &port->dentr_ar->dentries[i], + buff, port->admin_alias_parent, + show_admin_alias_guid, store_admin_alias_guid); + if (ret) + goto err_admin_alias_parent; + } + + /* gids subdirectory (operational gids) */ + port->gids_parent = kobject_create_and_add("gids", + kobject_get(port->cur_port)); + if (!port->gids_parent) { + ret = -ENOMEM; + goto err_gids; + } + + for (i = 0 ; i < attr.gid_tbl_len; i++) { + sprintf(buff, "%d", i); + port->dentr_ar->dentries[attr.gid_tbl_len + i].entry_num = i; + ret = create_sysfs_entry(port, + &port->dentr_ar->dentries[attr.gid_tbl_len + i], + buff, + port->gids_parent, show_port_gid, NULL); + if (ret) + goto err_gids_parent; + } + + /* physical port pkey table */ + port->pkeys_parent = + kobject_create_and_add("pkeys", kobject_get(port->cur_port)); + if (!port->pkeys_parent) { + ret = -ENOMEM; + goto err_pkeys; + } + + for (i = 0 ; i < attr.pkey_tbl_len; i++) { + sprintf(buff, "%d", i); + port->dentr_ar->dentries[2 * attr.gid_tbl_len + i].entry_num = i; + ret = create_sysfs_entry(port, + &port->dentr_ar->dentries[2 * attr.gid_tbl_len + i], + buff, port->pkeys_parent, + show_phys_port_pkey, NULL); + if (ret) + goto err_pkeys_parent; + } + + /* MCGs table */ + port->mcgs_parent = + kobject_create_and_add("mcgs", kobject_get(port->cur_port)); + if (!port->mcgs_parent) { + ret = -ENOMEM; + goto err_mcgs; + } + return 0; + +err_mcgs: + kobject_put(port->cur_port); + +err_pkeys_parent: + kobject_put(port->pkeys_parent); + +err_pkeys: + kobject_put(port->cur_port); + +err_gids_parent: + kobject_put(port->gids_parent); + +err_gids: + kobject_put(port->cur_port); + +err_admin_alias_parent: + kobject_put(port->admin_alias_parent); + +err_admin_guids: + kobject_put(port->cur_port); + kobject_put(port->cur_port); /* once more for create_and_add buff */ + +kobj_create_err: + kobject_put(device->ports_parent); + kfree(port->dentr_ar); + +err: + pr_err("add_port_entries FAILED: for port:%d, error: %d\n", + port_num, ret); + return ret; +} + +static void get_name(struct mlx4_ib_dev *dev, char *name, int i, int max) +{ + char base_name[9]; + + /* pci_name format is: bus:dev:func -> xxxx:yy:zz.n */ + strlcpy(name, pci_name(dev->dev->pdev), max); + strncpy(base_name, name, 8); /*till xxxx:yy:*/ + base_name[8] = '\0'; + /* with no ARI only 3 last bits are used so when the fn is higher than 8 + * need to add it to the dev num, so count in the last number will be + * modulo 8 */ + sprintf(name, "%s%.2d.%d", base_name, (i/8), (i%8)); +} + +struct mlx4_port { + struct kobject kobj; + struct mlx4_ib_dev *dev; + struct attribute_group pkey_group; + struct attribute_group gid_group; + u8 port_num; + int slave; +}; + + +static void mlx4_port_release(struct kobject *kobj) +{ + struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj); + struct attribute *a; + int i; + + for (i = 0; (a = p->pkey_group.attrs[i]); ++i) + kfree(a); + kfree(p->pkey_group.attrs); + for (i = 0; (a = p->gid_group.attrs[i]); ++i) + kfree(a); + kfree(p->gid_group.attrs); + kfree(p); +} + +struct port_attribute { + struct attribute attr; + ssize_t (*show)(struct mlx4_port *, struct port_attribute *, char *buf); + ssize_t (*store)(struct mlx4_port *, struct port_attribute *, + const char *buf, size_t count); +}; + +static ssize_t port_attr_show(struct kobject *kobj, + struct attribute *attr, char *buf) +{ + struct port_attribute *port_attr = + container_of(attr, struct port_attribute, attr); + struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj); + + if (!port_attr->show) + return -EIO; + return port_attr->show(p, port_attr, buf); +} + +static ssize_t port_attr_store(struct kobject *kobj, + struct attribute *attr, + const char *buf, size_t size) +{ + struct port_attribute *port_attr = + container_of(attr, struct port_attribute, attr); + struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj); + + if (!port_attr->store) + return -EIO; + return port_attr->store(p, port_attr, buf, size); +} + +static const struct sysfs_ops port_sysfs_ops = { + .show = port_attr_show, + .store = port_attr_store, +}; + +static struct kobj_type port_type = { + .release = mlx4_port_release, + .sysfs_ops = &port_sysfs_ops, +}; + +struct port_table_attribute { + struct port_attribute attr; + char name[8]; + int index; +}; + +static ssize_t show_port_pkey(struct mlx4_port *p, struct port_attribute *attr, + char *buf) +{ + struct port_table_attribute *tab_attr = + container_of(attr, struct port_table_attribute, attr); + ssize_t ret = -ENODEV; + + if (p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1][tab_attr->index] >= + (p->dev->dev->caps.pkey_table_len[p->port_num])) + ret = sprintf(buf, "none\n"); + else + ret = sprintf(buf, "%d\n", + p->dev->pkeys.virt2phys_pkey[p->slave] + [p->port_num - 1][tab_attr->index]); + return ret; +} + +static ssize_t store_port_pkey(struct mlx4_port *p, struct port_attribute *attr, + const char *buf, size_t count) +{ + struct port_table_attribute *tab_attr = + container_of(attr, struct port_table_attribute, attr); + int idx; + int err; + + /* do not allow remapping Dom0 virtual pkey table */ + if (p->slave == mlx4_master_func_num(p->dev->dev)) + return -EINVAL; + + if (!strncasecmp(buf, "no", 2)) + idx = p->dev->dev->phys_caps.pkey_phys_table_len[p->port_num] - 1; + else if (sscanf(buf, "%i", &idx) != 1 || + idx >= p->dev->dev->caps.pkey_table_len[p->port_num] || + idx < 0) + return -EINVAL; + + p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1] + [tab_attr->index] = idx; + mlx4_sync_pkey_table(p->dev->dev, p->slave, p->port_num, + tab_attr->index, idx); + err = mlx4_gen_pkey_eqe(p->dev->dev, p->slave, p->port_num); + if (err) { + pr_err("mlx4_gen_pkey_eqe failed for slave %d," + " port %d, index %d\n", p->slave, p->port_num, idx); + return err; + } + return count; +} + +static ssize_t show_port_gid_idx(struct mlx4_port *p, + struct port_attribute *attr, char *buf) +{ + return sprintf(buf, "%d\n", p->slave); +} + +static struct attribute ** +alloc_group_attrs(ssize_t (*show)(struct mlx4_port *, + struct port_attribute *, char *buf), + ssize_t (*store)(struct mlx4_port *, struct port_attribute *, + const char *buf, size_t count), + int len) +{ + struct attribute **tab_attr; + struct port_table_attribute *element; + int i; + + tab_attr = kcalloc(1 + len, sizeof (struct attribute *), GFP_KERNEL); + if (!tab_attr) + return NULL; + + for (i = 0; i < len; i++) { + element = kzalloc(sizeof (struct port_table_attribute), + GFP_KERNEL); + if (!element) + goto err; + if (snprintf(element->name, sizeof (element->name), + "%d", i) >= sizeof (element->name)) { + kfree(element); + goto err; + } + sysfs_attr_init(&element->attr.attr); + element->attr.attr.name = element->name; + if (store) { + element->attr.attr.mode = S_IWUSR | S_IRUGO; + element->attr.store = store; + } else + element->attr.attr.mode = S_IRUGO; + + element->attr.show = show; + element->index = i; + tab_attr[i] = &element->attr.attr; + } + return tab_attr; + +err: + while (--i >= 0) + kfree(tab_attr[i]); + kfree(tab_attr); + return NULL; +} + +static int add_port(struct mlx4_ib_dev *dev, int port_num, int slave) +{ + struct mlx4_port *p; + int i; + int ret; + + p = kzalloc(sizeof *p, GFP_KERNEL); + if (!p) + return -ENOMEM; + + p->dev = dev; + p->port_num = port_num; + p->slave = slave; + + ret = kobject_init_and_add(&p->kobj, &port_type, + kobject_get(dev->dev_ports_parent[slave]), + "%d", port_num); + if (ret) + goto err_alloc; + + p->pkey_group.name = "pkey_idx"; + p->pkey_group.attrs = + alloc_group_attrs(show_port_pkey, store_port_pkey, + dev->dev->caps.pkey_table_len[port_num]); + if (!p->pkey_group.attrs) + goto err_alloc; + + ret = sysfs_create_group(&p->kobj, &p->pkey_group); + if (ret) + goto err_free_pkey; + + p->gid_group.name = "gid_idx"; + p->gid_group.attrs = alloc_group_attrs(show_port_gid_idx, NULL, 1); + if (!p->gid_group.attrs) + goto err_free_pkey; + + ret = sysfs_create_group(&p->kobj, &p->gid_group); + if (ret) + goto err_free_gid; + + list_add_tail(&p->kobj.entry, &dev->pkeys.pkey_port_list[slave]); + return 0; + +err_free_gid: + kfree(p->gid_group.attrs[0]); + kfree(p->gid_group.attrs); + +err_free_pkey: + for (i = 0; i < dev->dev->caps.pkey_table_len[port_num]; ++i) + kfree(p->pkey_group.attrs[i]); + kfree(p->pkey_group.attrs); + +err_alloc: + kobject_put(dev->dev_ports_parent[slave]); + kfree(p); + return ret; +} + +static int register_one_pkey_tree(struct mlx4_ib_dev *dev, int slave) +{ + char name[32]; + int err; + int port; + struct kobject *p, *t; + struct mlx4_port *mport; + + get_name(dev, name, slave, sizeof name); + + dev->pkeys.device_parent[slave] = + kobject_create_and_add(name, kobject_get(dev->iov_parent)); + + if (!dev->pkeys.device_parent[slave]) { + err = -ENOMEM; + goto fail_dev; + } + + INIT_LIST_HEAD(&dev->pkeys.pkey_port_list[slave]); + + dev->dev_ports_parent[slave] = + kobject_create_and_add("ports", + kobject_get(dev->pkeys.device_parent[slave])); + + if (!dev->dev_ports_parent[slave]) { + err = -ENOMEM; + goto err_ports; + } + + for (port = 1; port <= dev->dev->caps.num_ports; ++port) { + err = add_port(dev, port, slave); + if (err) + goto err_add; + } + return 0; + +err_add: + list_for_each_entry_safe(p, t, + &dev->pkeys.pkey_port_list[slave], + entry) { + list_del(&p->entry); + mport = container_of(p, struct mlx4_port, kobj); + sysfs_remove_group(p, &mport->pkey_group); + sysfs_remove_group(p, &mport->gid_group); + kobject_put(p); + } + kobject_put(dev->dev_ports_parent[slave]); + +err_ports: + kobject_put(dev->pkeys.device_parent[slave]); + /* extra put for the device_parent create_and_add */ + kobject_put(dev->pkeys.device_parent[slave]); + +fail_dev: + kobject_put(dev->iov_parent); + return err; +} + +static int register_pkey_tree(struct mlx4_ib_dev *device) +{ + int i; + + if (!mlx4_is_master(device->dev)) + return 0; + + for (i = 0; i <= device->dev->num_vfs; ++i) + register_one_pkey_tree(device, i); + + return 0; +} + +static void unregister_pkey_tree(struct mlx4_ib_dev *device) +{ + int slave; + struct kobject *p, *t; + struct mlx4_port *port; + + if (!mlx4_is_master(device->dev)) + return; + + for (slave = device->dev->num_vfs; slave >= 0; --slave) { + list_for_each_entry_safe(p, t, + &device->pkeys.pkey_port_list[slave], + entry) { + list_del(&p->entry); + port = container_of(p, struct mlx4_port, kobj); + sysfs_remove_group(p, &port->pkey_group); + sysfs_remove_group(p, &port->gid_group); + kobject_put(p); + kobject_put(device->dev_ports_parent[slave]); + } + kobject_put(device->dev_ports_parent[slave]); + kobject_put(device->pkeys.device_parent[slave]); + kobject_put(device->pkeys.device_parent[slave]); + kobject_put(device->iov_parent); + } +} + +int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *dev) +{ + int i; + int ret = 0; + + if (!mlx4_is_master(dev->dev)) + return 0; + + dev->iov_parent = + kobject_create_and_add("iov", + kobject_get(dev->ib_dev.ports_parent->parent)); + if (!dev->iov_parent) { + ret = -ENOMEM; + goto err; + } + dev->ports_parent = + kobject_create_and_add("ports", + kobject_get(dev->iov_parent)); + if (!dev->iov_parent) { + ret = -ENOMEM; + goto err_ports; + } + + for (i = 1; i <= dev->ib_dev.phys_port_cnt; ++i) { + ret = add_port_entries(dev, i); + if (ret) + goto err_add_entries; + } + + ret = register_pkey_tree(dev); + if (ret) + goto err_add_entries; + return 0; + +err_add_entries: + kobject_put(dev->ports_parent); + +err_ports: + kobject_put(dev->iov_parent); +err: + kobject_put(dev->ib_dev.ports_parent->parent); + pr_err("mlx4_ib_device_register_sysfs error (%d)\n", ret); + return ret; +} + +static void unregister_alias_guid_tree(struct mlx4_ib_dev *device) +{ + struct mlx4_ib_iov_port *p; + int i; + + if (!mlx4_is_master(device->dev)) + return; + + for (i = 0; i < device->dev->caps.num_ports; i++) { + p = &device->iov_ports[i]; + kobject_put(p->admin_alias_parent); + kobject_put(p->gids_parent); + kobject_put(p->pkeys_parent); + kobject_put(p->mcgs_parent); + kobject_put(p->cur_port); + kobject_put(p->cur_port); + kobject_put(p->cur_port); + kobject_put(p->cur_port); + kobject_put(p->cur_port); + kobject_put(p->dev->ports_parent); + kfree(p->dentr_ar); + } +} + +void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device) +{ + unregister_alias_guid_tree(device); + unregister_pkey_tree(device); + kobject_put(device->ports_parent); + kobject_put(device->iov_parent); + kobject_put(device->iov_parent); + kobject_put(device->ib_dev.ports_parent->parent); +} From efcd235d736ab05ef2b29d7fe1493a2f52b07b66 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:52 +0000 Subject: [PATCH 23/45] net/mlx4_core: Adjustments to SET_PORT for IB SR-IOV 1. Slaves may not set the IS_SM capability for the port. 2. DEV_MGMT may not be set in multifunction mode. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/port.c | 10 ++++++++++ include/linux/mlx4/device.h | 5 +++++ 2 files changed, 15 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx4/port.c b/drivers/net/ethernet/mellanox/mlx4/port.c index e36dd0f2fa73..4c51b05efa28 100644 --- a/drivers/net/ethernet/mellanox/mlx4/port.c +++ b/drivers/net/ethernet/mellanox/mlx4/port.c @@ -732,6 +732,16 @@ static int mlx4_common_set_port(struct mlx4_dev *dev, int slave, u32 in_mod, new_cap_mask = ((__be32 *) inbox->buf)[1]; } + /* slave may not set the IS_SM capability for the port */ + if (slave != mlx4_master_func_num(dev) && + (be32_to_cpu(new_cap_mask) & MLX4_PORT_CAP_IS_SM)) + return -EINVAL; + + /* No DEV_MGMT in multifunc mode */ + if (mlx4_is_mfunc(dev) && + (be32_to_cpu(new_cap_mask) & MLX4_PORT_CAP_DEV_MGMT_SUP)) + return -EINVAL; + agg_cap_mask = 0; slave_cap_mask = priv->mfunc.master.slave_state[slave].ib_cap_mask[port]; diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index 380e01671d1e..b3f9b4500b9b 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -53,6 +53,11 @@ enum { MLX4_FLAG_SRIOV = 1 << 4, }; +enum { + MLX4_PORT_CAP_IS_SM = 1 << 1, + MLX4_PORT_CAP_DEV_MGMT_SUP = 1 << 19, +}; + enum { MLX4_MAX_PORTS = 2, MLX4_MAX_PORT_PKEYS = 128 From 980e90010f15362b3b02ed4875ef2758aee3de72 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:53 +0000 Subject: [PATCH 24/45] mlx4_core: INIT/CLOSE port logic for IB ports in SR-IOV mode Normally, INIT_PORT and CLOSE_PORT are invoked when special QP0 transitions to RTR, or transitions to ERR/RESET respectively. In SR-IOV mode, however, the master is also paravirtualized. This in turn requires that we not do INIT_PORT until the entire QP0 path (real QP0 and proxy QP0) is ready to receive. When the real QP0 goes down, we should indicate that the port is not active. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/fw.c | 71 ++++++++++++++++++++----- drivers/net/ethernet/mellanox/mlx4/qp.c | 38 +++++++++++-- 2 files changed, 92 insertions(+), 17 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c index b78803937897..bf159c0fe82b 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.c +++ b/drivers/net/ethernet/mellanox/mlx4/fw.c @@ -1365,6 +1365,19 @@ out: return err; } +/* for IB-type ports only in SRIOV mode. Checks that both proxy QP0 + * and real QP0 are active, so that the paravirtualized QP0 is ready + * to operate */ +static int check_qp0_state(struct mlx4_dev *dev, int function, int port) +{ + struct mlx4_priv *priv = mlx4_priv(dev); + /* irrelevant if not infiniband */ + if (priv->mfunc.master.qp0_state[port].proxy_qp0_active && + priv->mfunc.master.qp0_state[port].qp0_active) + return 1; + return 0; +} + int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, @@ -1381,14 +1394,29 @@ int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave, if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB) return -ENODEV; - /* Enable port only if it was previously disabled */ - if (!priv->mfunc.master.init_port_ref[port]) { - err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT, - MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); - if (err) - return err; + if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) { + /* Enable port only if it was previously disabled */ + if (!priv->mfunc.master.init_port_ref[port]) { + err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT, + MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); + if (err) + return err; + } + priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port); + } else { + if (slave == mlx4_master_func_num(dev)) { + if (check_qp0_state(dev, slave, port) && + !priv->mfunc.master.qp0_state[port].port_active) { + err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_INIT_PORT, + MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); + if (err) + return err; + priv->mfunc.master.qp0_state[port].port_active = 1; + priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port); + } + } else + priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port); } - priv->mfunc.master.slave_state[slave].init_port_mask |= (1 << port); ++priv->mfunc.master.init_port_ref[port]; return 0; } @@ -1463,13 +1491,30 @@ int mlx4_CLOSE_PORT_wrapper(struct mlx4_dev *dev, int slave, if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB) return -ENODEV; - if (priv->mfunc.master.init_port_ref[port] == 1) { - err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT, 1000, - MLX4_CMD_NATIVE); - if (err) - return err; + + if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) { + if (priv->mfunc.master.init_port_ref[port] == 1) { + err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT, + 1000, MLX4_CMD_NATIVE); + if (err) + return err; + } + priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port); + } else { + /* infiniband port */ + if (slave == mlx4_master_func_num(dev)) { + if (!priv->mfunc.master.qp0_state[port].qp0_active && + priv->mfunc.master.qp0_state[port].port_active) { + err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT, + 1000, MLX4_CMD_NATIVE); + if (err) + return err; + priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port); + priv->mfunc.master.qp0_state[port].port_active = 0; + } + } else + priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port); } - priv->mfunc.master.slave_state[slave].init_port_mask &= ~(1 << port); --priv->mfunc.master.init_port_ref[port]; return 0; } diff --git a/drivers/net/ethernet/mellanox/mlx4/qp.c b/drivers/net/ethernet/mellanox/mlx4/qp.c index b8da72b29cf7..436ef6c69add 100644 --- a/drivers/net/ethernet/mellanox/mlx4/qp.c +++ b/drivers/net/ethernet/mellanox/mlx4/qp.c @@ -67,10 +67,18 @@ void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type) complete(&qp->free); } -static int is_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp) +/* used for INIT/CLOSE port logic */ +static int is_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0) { - return qp->qpn >= dev->caps.sqp_start && + /* qp0 is either the proxy qp0, or the real qp0 */ + *proxy_qp0 = qp->qpn >= dev->caps.sqp_start && qp->qpn <= dev->caps.sqp_start + 1; + + *real_qp0 = mlx4_is_master(dev) && + qp->qpn >= dev->caps.base_sqpn && + qp->qpn <= dev->caps.base_sqpn + 1; + + return *real_qp0 || *proxy_qp0; } static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, @@ -122,6 +130,8 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, struct mlx4_priv *priv = mlx4_priv(dev); struct mlx4_cmd_mailbox *mailbox; int ret = 0; + int real_qp0 = 0; + int proxy_qp0 = 0; u8 port; if (cur_state >= MLX4_QP_NUM_STATE || new_state >= MLX4_QP_NUM_STATE || @@ -133,9 +143,12 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, MLX4_CMD_2RST_QP, MLX4_CMD_TIME_CLASS_A, native); if (mlx4_is_master(dev) && cur_state != MLX4_QP_STATE_ERR && cur_state != MLX4_QP_STATE_RST && - is_qp0(dev, qp)) { + is_qp0(dev, qp, &real_qp0, &proxy_qp0)) { port = (qp->qpn & 1) + 1; - priv->mfunc.master.qp0_state[port].qp0_active = 0; + if (proxy_qp0) + priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0; + else + priv->mfunc.master.qp0_state[port].qp0_active = 0; } return ret; } @@ -162,6 +175,23 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, new_state == MLX4_QP_STATE_RST ? 2 : 0, op[cur_state][new_state], MLX4_CMD_TIME_CLASS_C, native); + if (mlx4_is_master(dev) && is_qp0(dev, qp, &real_qp0, &proxy_qp0)) { + port = (qp->qpn & 1) + 1; + if (cur_state != MLX4_QP_STATE_ERR && + cur_state != MLX4_QP_STATE_RST && + new_state == MLX4_QP_STATE_ERR) { + if (proxy_qp0) + priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0; + else + priv->mfunc.master.qp0_state[port].qp0_active = 0; + } else if (new_state == MLX4_QP_STATE_RTR) { + if (proxy_qp0) + priv->mfunc.master.qp0_state[port].proxy_qp0_active = 1; + else + priv->mfunc.master.qp0_state[port].qp0_active = 1; + } + } + mlx4_free_cmd_mailbox(dev, mailbox); return ret; } From 992e8e6e8781b71fd475bd1fd0555da7dba59966 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:54 +0000 Subject: [PATCH 25/45] IB/mlx4: Miscellaneous adjustments for SR-IOV IB support 1. Allow only master to change node description. 2. Prevent AH leakage in send mads. 3. Take device part number from PCI structure, so that guests see the VF part number (and not the PF part number). 4. Place the device revision ID into caps structure at startup. 5. SET_PORT in update_gids_task needs to go through wrapper on master. 6. In mlx4_ib_event(), PORT_MGMT_EVENT needs be handled in a work queue on the master, since it propagates events to slaves using GEN_EQE. 7. Do not support FMR on slaves. 8. Add spinlock to slave_event(), since it is called both in interrupt context and in process context (due to 6 above, and also if smp_snoop is used). This fix was found and implemented by Saeed Mahameed Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 6 +++++- drivers/infiniband/hw/mlx4/main.c | 26 ++++++++++++++++------- drivers/net/ethernet/mellanox/mlx4/cmd.c | 1 + drivers/net/ethernet/mellanox/mlx4/eq.c | 8 +++++-- drivers/net/ethernet/mellanox/mlx4/mlx4.h | 1 + 5 files changed, 31 insertions(+), 11 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index b689dbd6d8f6..b91b4865d635 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -709,7 +709,9 @@ static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num, if (!out_mad->mad_hdr.status) { if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV)) smp_snoop(ibdev, port_num, in_mad, prev_lid); - node_desc_override(ibdev, out_mad); + /* slaves get node desc from FW */ + if (!mlx4_is_slave(to_mdev(ibdev)->dev)) + node_desc_override(ibdev, out_mad); } /* set return bit in status of directed route responses */ @@ -792,6 +794,8 @@ int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num, static void send_handler(struct ib_mad_agent *agent, struct ib_mad_send_wc *mad_send_wc) { + if (mad_send_wc->send_buf->context[0]) + ib_destroy_ah(mad_send_wc->send_buf->context[0]); ib_free_send_mad(mad_send_wc->send_buf); } diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index 7d97578fbbaa..46303b209ce6 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -138,7 +138,7 @@ static int mlx4_ib_query_device(struct ib_device *ibdev, props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) & 0xffffff; - props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30)); + props->vendor_part_id = dev->dev->pdev->device; props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32)); memcpy(&props->sys_image_guid, out_mad->data + 4, 8); @@ -478,6 +478,9 @@ static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask, if (!(mask & IB_DEVICE_MODIFY_NODE_DESC)) return 0; + if (mlx4_is_slave(to_mdev(ibdev)->dev)) + return -EOPNOTSUPP; + spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags); memcpy(ibdev->node_desc, props->node_desc, 64); spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags); @@ -493,7 +496,7 @@ static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask, memset(mailbox->buf, 0, 256); memcpy(mailbox->buf, props->node_desc, 64); mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0, - MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); + MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox); @@ -921,6 +924,7 @@ static int init_node_data(struct mlx4_ib_dev *dev) if (err) goto out; + dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32)); memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8); out: @@ -1009,7 +1013,7 @@ static void update_gids_task(struct work_struct *work) err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port, 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B, - MLX4_CMD_NATIVE); + MLX4_CMD_WRAPPED); if (err) pr_warn("set port command failed\n"); else { @@ -1400,10 +1404,12 @@ static void *mlx4_ib_add(struct mlx4_dev *dev) ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach; ibdev->ib_dev.process_mad = mlx4_ib_process_mad; - ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc; - ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr; - ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr; - ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc; + if (!mlx4_is_slave(ibdev->dev)) { + ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc; + ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr; + ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr; + ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc; + } if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) { ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd; @@ -1615,7 +1621,11 @@ static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, INIT_WORK(&ew->work, handle_port_mgmt_change_event); memcpy(&ew->ib_eqe, eqe, sizeof *eqe); ew->ib_dev = ibdev; - handle_port_mgmt_change_event(&ew->work); + /* need to queue only for port owner, which uses GEN_EQE */ + if (mlx4_is_master(dev)) + queue_work(wq, &ew->work); + else + handle_port_mgmt_change_event(&ew->work); return; case MLX4_DEV_EVENT_SLAVE_INIT: diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index ea5c884ab899..78fdbd653fa9 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -1757,6 +1757,7 @@ int mlx4_multi_func_init(struct mlx4_dev *dev) INIT_WORK(&priv->mfunc.master.slave_flr_event_work, mlx4_master_handle_slave_flr); spin_lock_init(&priv->mfunc.master.slave_state_lock); + spin_lock_init(&priv->mfunc.master.slave_eq.event_lock); priv->mfunc.master.comm_wq = create_singlethread_workqueue("mlx4_comm"); if (!priv->mfunc.master.comm_wq) diff --git a/drivers/net/ethernet/mellanox/mlx4/eq.c b/drivers/net/ethernet/mellanox/mlx4/eq.c index c425826b1fc0..51c764901ad2 100644 --- a/drivers/net/ethernet/mellanox/mlx4/eq.c +++ b/drivers/net/ethernet/mellanox/mlx4/eq.c @@ -164,13 +164,16 @@ static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe) { struct mlx4_priv *priv = mlx4_priv(dev); struct mlx4_slave_event_eq *slave_eq = &priv->mfunc.master.slave_eq; - struct mlx4_eqe *s_eqe = - &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)]; + struct mlx4_eqe *s_eqe; + unsigned long flags; + spin_lock_irqsave(&slave_eq->event_lock, flags); + s_eqe = &slave_eq->event_eqe[slave_eq->prod & (SLAVE_EVENT_EQ_SIZE - 1)]; if ((!!(s_eqe->owner & 0x80)) ^ (!!(slave_eq->prod & SLAVE_EVENT_EQ_SIZE))) { mlx4_warn(dev, "Master failed to generate an EQE for slave: %d. " "No free EQE on slave events queue\n", slave); + spin_unlock_irqrestore(&slave_eq->event_lock, flags); return; } @@ -183,6 +186,7 @@ static void slave_event(struct mlx4_dev *dev, u8 slave, struct mlx4_eqe *eqe) queue_work(priv->mfunc.master.comm_wq, &priv->mfunc.master.slave_event_work); + spin_unlock_irqrestore(&slave_eq->event_lock, flags); } static void mlx4_slave_event(struct mlx4_dev *dev, int slave, diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h index 23f74759f403..399793a23457 100644 --- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h +++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h @@ -473,6 +473,7 @@ struct mlx4_slave_event_eq { u32 eqn; u32 cons; u32 prod; + spinlock_t event_lock; struct mlx4_eqe event_eqe[SLAVE_EVENT_EQ_SIZE]; }; From 026149cbaada391d98f1cbec47c488cb548f753a Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:55 +0000 Subject: [PATCH 26/45] mlx4: Activate SR-IOV mode for IB Remove the error returns for IB ports from mlx4_ib_add, mlx4_INIT_PORT_wrapper, and mlx4_CLOSE_PORT_wrapper. Currently, SRIOV is supported only for devices for which the link layer is IB on all ports; RoCE support will be added later. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/main.c | 8 ++++++-- drivers/net/ethernet/mellanox/mlx4/fw.c | 6 ------ include/linux/mlx4/device.h | 4 ++++ 3 files changed, 10 insertions(+), 8 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index 46303b209ce6..e849347ef99a 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -1293,11 +1293,15 @@ static void *mlx4_ib_add(struct mlx4_dev *dev) pr_info_once("%s", mlx4_ib_version); - if (mlx4_is_mfunc(dev)) { - pr_warn("IB not yet supported in SRIOV\n"); + mlx4_foreach_non_ib_transport_port(i, dev) + num_ports++; + + if (mlx4_is_mfunc(dev) && num_ports) { + dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n"); return NULL; } + num_ports = 0; mlx4_foreach_ib_transport_port(i, dev) num_ports++; diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c index bf159c0fe82b..1d74e8541146 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.c +++ b/drivers/net/ethernet/mellanox/mlx4/fw.c @@ -1391,9 +1391,6 @@ int mlx4_INIT_PORT_wrapper(struct mlx4_dev *dev, int slave, if (priv->mfunc.master.slave_state[slave].init_port_mask & (1 << port)) return 0; - if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB) - return -ENODEV; - if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) { /* Enable port only if it was previously disabled */ if (!priv->mfunc.master.init_port_ref[port]) { @@ -1489,9 +1486,6 @@ int mlx4_CLOSE_PORT_wrapper(struct mlx4_dev *dev, int slave, (1 << port))) return 0; - if (dev->caps.port_mask[port] == MLX4_PORT_TYPE_IB) - return -ENODEV; - if (dev->caps.port_mask[port] != MLX4_PORT_TYPE_IB) { if (priv->mfunc.master.init_port_ref[port] == 1) { err = mlx4_cmd(dev, 0, port, 0, MLX4_CMD_CLOSE_PORT, diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index b3f9b4500b9b..9b243df789ef 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -697,6 +697,10 @@ struct mlx4_init_port_param { for ((port) = 1; (port) <= (dev)->caps.num_ports; (port)++) \ if ((type) == (dev)->caps.port_mask[(port)]) +#define mlx4_foreach_non_ib_transport_port(port, dev) \ + for ((port) = 1; (port) <= (dev)->caps.num_ports; (port)++) \ + if (((dev)->caps.port_mask[port] != MLX4_PORT_TYPE_IB)) + #define mlx4_foreach_ib_transport_port(port, dev) \ for ((port) = 1; (port) <= (dev)->caps.num_ports; (port)++) \ if (((dev)->caps.port_mask[port] == MLX4_PORT_TYPE_IB) || \ From afa8fd1db9f295a0c4130bc6d87bf8b05bdd0523 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:56 +0000 Subject: [PATCH 27/45] mlx4: Paravirtualize Node Guids for slaves This is necessary in order to support > 1 VF/PF in a VM for software that uses the node guid as a discriminator, such as librdmacm. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 14 ++++++++++++++ drivers/infiniband/hw/mlx4/mlx4_ib.h | 3 +++ drivers/net/ethernet/mellanox/mlx4/cmd.c | 11 +++++++++++ drivers/net/ethernet/mellanox/mlx4/main.c | 22 ++++++++++++++++++++++ drivers/net/ethernet/mellanox/mlx4/mlx4.h | 1 + include/linux/mlx4/device.h | 2 ++ 6 files changed, 53 insertions(+) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index b91b4865d635..603a114b3dfe 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -35,6 +35,7 @@ #include #include +#include #include #include #include @@ -89,6 +90,12 @@ static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num); static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num, int block, u32 change_bitmap); +__be64 mlx4_ib_gen_node_guid(void) +{ +#define NODE_GUID_HI ((u64) (((u64)IB_OPENIB_OUI) << 40)) + return cpu_to_be64(NODE_GUID_HI | random32()); +} + __be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx) { return cpu_to_be64(atomic_inc_return(&ctx->tid)) | @@ -1962,6 +1969,13 @@ int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev) return 0; } + for (i = 0; i < dev->dev->caps.sqp_demux; i++) { + if (i == mlx4_master_func_num(dev->dev)) + mlx4_put_slave_node_guid(dev->dev, i, dev->ib_dev.node_guid); + else + mlx4_put_slave_node_guid(dev->dev, i, mlx4_ib_gen_node_guid()); + } + err = mlx4_ib_init_alias_guid_service(dev); if (err) { mlx4_ib_warn(&dev->ib_dev, "Failed init alias guid process.\n"); diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index e57a220a4d55..e04cbc9a54a5 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -723,4 +723,7 @@ int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *device) ; void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device); +__be64 mlx4_ib_gen_node_guid(void); + + #endif /* MLX4_IB_H */ diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index 78fdbd653fa9..90774b7e47a4 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -696,6 +696,7 @@ static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave, struct ib_smp *outsmp = outbox->buf; __be16 *outtab = (__be16 *)(outsmp->data); __be32 slave_cap_mask; + __be64 slave_node_guid; port = vhcr->in_modifier; if (smp->base_version == 1 && @@ -755,6 +756,16 @@ static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave, } return err; } + if (smp->attr_id == IB_SMP_ATTR_NODE_INFO) { + err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, + vhcr->in_modifier, vhcr->op_modifier, + vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE); + if (!err) { + slave_node_guid = mlx4_get_slave_node_guid(dev, slave); + memcpy(outsmp->data + 12, &slave_node_guid, 8); + } + return err; + } } } if (slave != mlx4_master_func_num(dev) && diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 2294b7173180..76f69fdd01d5 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -435,6 +435,28 @@ void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int } EXPORT_SYMBOL(mlx4_sync_pkey_table); +void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid) +{ + struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev); + + if (!mlx4_is_master(dev)) + return; + + priv->slave_node_guids[slave] = guid; +} +EXPORT_SYMBOL(mlx4_put_slave_node_guid); + +__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave) +{ + struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev); + + if (!mlx4_is_master(dev)) + return 0; + + return priv->slave_node_guids[slave]; +} +EXPORT_SYMBOL(mlx4_get_slave_node_guid); + int mlx4_is_slave_active(struct mlx4_dev *dev, int slave) { struct mlx4_priv *priv = mlx4_priv(dev); diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h index 399793a23457..4b1fbd84a68f 100644 --- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h +++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h @@ -810,6 +810,7 @@ struct mlx4_priv { int reserved_mtts; int fs_hash_mode; u8 virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS]; + __be64 slave_node_guids[MLX4_MFUNC_MAX]; }; diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index 9b243df789ef..c0f8f74a0a5e 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -981,5 +981,7 @@ int mlx4_gen_port_state_change_eqe(struct mlx4_dev *dev, int slave, u8 port, u8 enum slave_port_state mlx4_get_slave_port_state(struct mlx4_dev *dev, int slave, u8 port); int set_and_calc_slave_port_state(struct mlx4_dev *dev, int slave, u8 port, int event, enum slave_port_gen_event *gen_event); +void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid); +__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave); #endif /* MLX4_DEVICE_H */ From 47605df953985c2b792ac9f3ddf70d270b89adb8 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:57 +0000 Subject: [PATCH 28/45] mlx4: Modify proxy/tunnel QP mechanism so that guests do no calculations Previously, the structure of a guest's proxy QPs followed the structure of the PPF special qps (qp0 port 1, qp0 port 2, qp1 port 1, qp1 port 2, ...). The guest then did offset calculations on the sqp_base qp number that the PPF passed to it in QUERY_FUNC_CAP(). This is now changed so that the guest does no offset calculations regarding proxy or tunnel QPs to use. This change frees the PPF from needing to adhere to a specific order in allocating proxy and tunnel QPs. Now QUERY_FUNC_CAP provides each port individually with its proxy qp0, proxy qp1, tunnel qp0, and tunnel qp1 QP numbers, and these are used directly where required (with no offset calculations). To accomplish this change, several fields were added to the phys_caps structure for use by the PPF and by non-SR-IOV mode: base_sqpn -- in non-sriov mode, this was formerly sqp_start. base_proxy_sqpn -- the first physical proxy qp number -- used by PPF base_tunnel_sqpn -- the first physical tunnel qp number -- used by PPF. The current code in the PPF still adheres to the previous layout of sqps, proxy-sqps and tunnel-sqps. However, the PPF can change this layout without affecting VF or (paravirtualized) PF code. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 12 +- drivers/infiniband/hw/mlx4/qp.c | 104 ++++++++---- drivers/net/ethernet/mellanox/mlx4/fw.c | 191 ++++++++++++---------- drivers/net/ethernet/mellanox/mlx4/fw.h | 14 +- drivers/net/ethernet/mellanox/mlx4/main.c | 61 +++++-- drivers/net/ethernet/mellanox/mlx4/qp.c | 75 ++++++--- include/linux/mlx4/device.h | 16 +- 7 files changed, 301 insertions(+), 172 deletions(-) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 603a114b3dfe..658a622791fb 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -505,7 +505,7 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port, } else tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0]; - dqpn = dev->dev->caps.sqp_start + 8 * slave + port + (dest_qpt * 2) - 1; + dqpn = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave + port + (dest_qpt * 2) - 1; /* get tunnel tx data buf for slave */ src_qp = tun_qp->qp; @@ -1074,9 +1074,9 @@ static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port, static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave) { - int slave_start = dev->dev->caps.sqp_start + 8 * slave; + int proxy_start = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave; - return (qpn >= slave_start && qpn <= slave_start + 1); + return (qpn >= proxy_start && qpn <= proxy_start + 1); } @@ -1191,14 +1191,14 @@ static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc int slave; /* Get slave that sent this packet */ - if (wc->src_qp < dev->dev->caps.sqp_start || - wc->src_qp >= dev->dev->caps.base_tunnel_sqpn || + if (wc->src_qp < dev->dev->phys_caps.base_proxy_sqpn || + wc->src_qp >= dev->dev->phys_caps.base_proxy_sqpn + 8 * MLX4_MFUNC_MAX || (wc->src_qp & 0x1) != ctx->port - 1 || wc->src_qp & 0x4) { mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d\n", wc->src_qp); return; } - slave = ((wc->src_qp & ~0x7) - dev->dev->caps.sqp_start) / 8; + slave = ((wc->src_qp & ~0x7) - dev->dev->phys_caps.base_proxy_sqpn) / 8; if (slave != ctx->slave) { mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: " "belongs to another slave\n", wc->src_qp); diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index a8622510de42..96fe103f5973 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -116,33 +116,57 @@ static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) if (!mlx4_is_master(dev->dev)) return 0; - return qp->mqp.qpn >= dev->dev->caps.base_sqpn && - qp->mqp.qpn < dev->dev->caps.base_sqpn + - 8 + 16 * MLX4_MFUNC_MAX; + return qp->mqp.qpn >= dev->dev->phys_caps.base_tunnel_sqpn && + qp->mqp.qpn < dev->dev->phys_caps.base_tunnel_sqpn + + 8 * MLX4_MFUNC_MAX; } static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) { - return ((mlx4_is_master(dev->dev) && - qp->mqp.qpn >= dev->dev->caps.base_sqpn && - qp->mqp.qpn <= dev->dev->caps.base_sqpn + 3) || - (qp->mqp.qpn >= dev->dev->caps.sqp_start && - qp->mqp.qpn <= dev->dev->caps.sqp_start + 3)); + int proxy_sqp = 0; + int real_sqp = 0; + int i; + /* PPF or Native -- real SQP */ + real_sqp = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) && + qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn && + qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 3); + if (real_sqp) + return 1; + /* VF or PF -- proxy SQP */ + if (mlx4_is_mfunc(dev->dev)) { + for (i = 0; i < dev->dev->caps.num_ports; i++) { + if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i] || + qp->mqp.qpn == dev->dev->caps.qp1_proxy[i]) { + proxy_sqp = 1; + break; + } + } + } + return proxy_sqp; } /* used for INIT/CLOSE port logic */ static int is_qp0(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp) { - int qp0; - - /* qp0 is either the proxy qp0, or the real qp0 */ - qp0 = (qp->mqp.qpn >= dev->dev->caps.sqp_start && - qp->mqp.qpn <= dev->dev->caps.sqp_start + 1) || - (mlx4_is_master(dev->dev) && - qp->mqp.qpn >= dev->dev->caps.base_sqpn && - qp->mqp.qpn <= dev->dev->caps.base_sqpn + 1); - - return qp0; + int proxy_qp0 = 0; + int real_qp0 = 0; + int i; + /* PPF or Native -- real QP0 */ + real_qp0 = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) && + qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn && + qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 1); + if (real_qp0) + return 1; + /* VF or PF -- proxy QP0 */ + if (mlx4_is_mfunc(dev->dev)) { + for (i = 0; i < dev->dev->caps.num_ports; i++) { + if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i]) { + proxy_qp0 = 1; + break; + } + } + } + return proxy_qp0; } static void *get_wqe(struct mlx4_ib_qp *qp, int offset) @@ -607,8 +631,10 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, qp_type = MLX4_IB_QPT_TUN_SMI_OWNER; else qp_type = MLX4_IB_QPT_TUN_SMI; - qpn = dev->dev->caps.base_tunnel_sqpn + 8 * tnl_init->slave + - tnl_init->proxy_qp_type * 2 + tnl_init->port - 1; + /* we are definitely in the PPF here, since we are creating + * tunnel QPs. base_tunnel_sqpn is therefore valid. */ + qpn = dev->dev->phys_caps.base_tunnel_sqpn + 8 * tnl_init->slave + + tnl_init->proxy_qp_type * 2 + tnl_init->port - 1; sqpn = qpn; } @@ -630,12 +656,6 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, qp->mlx4_ib_qp_type = qp_type; - if (mlx4_is_mfunc(dev->dev) && - (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI)) { - qpn -= 8; - sqpn -= 8; - } - mutex_init(&qp->mutex); spin_lock_init(&qp->sq.lock); spin_lock_init(&qp->rq.lock); @@ -935,6 +955,23 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp, del_gid_entries(qp); } +static u32 get_sqp_num(struct mlx4_ib_dev *dev, struct ib_qp_init_attr *attr) +{ + /* Native or PPF */ + if (!mlx4_is_mfunc(dev->dev) || + (mlx4_is_master(dev->dev) && + attr->create_flags & MLX4_IB_SRIOV_SQP)) { + return dev->dev->phys_caps.base_sqpn + + (attr->qp_type == IB_QPT_SMI ? 0 : 2) + + attr->port_num - 1; + } + /* PF or VF -- creating proxies */ + if (attr->qp_type == IB_QPT_SMI) + return dev->dev->caps.qp0_proxy[attr->port_num - 1]; + else + return dev->dev->caps.qp1_proxy[attr->port_num - 1]; +} + struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *init_attr, struct ib_udata *udata) @@ -998,9 +1035,7 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd, return ERR_PTR(-EINVAL); err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata, - to_mdev(pd->device)->dev->caps.sqp_start + - (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) + - init_attr->port_num - 1, + get_sqp_num(to_mdev(pd->device), init_attr), &qp); if (err) return ERR_PTR(err); @@ -1643,8 +1678,7 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp, sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn); else sqp->ud_header.bth.destination_qpn = - cpu_to_be32(mdev->dev->caps.base_tunnel_sqpn + - sqp->qp.port - 1); + cpu_to_be32(mdev->dev->caps.qp0_tunnel[sqp->qp.port - 1]); sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1)); if (mlx4_get_parav_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey)) @@ -2012,10 +2046,10 @@ static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev, cpu_to_be32(0xf0000000); memcpy(dseg->av, &sqp_av, sizeof (struct mlx4_av)); - dseg->dqpn = cpu_to_be32(dev->dev->caps.base_tunnel_sqpn + - qpt * 2 + port - 1); - /* use well-known qkey from the QPC */ - dseg->qkey = cpu_to_be32(0x80000000); + /* This function used only for sending on QP1 proxies */ + dseg->dqpn = cpu_to_be32(dev->dev->caps.qp1_tunnel[port - 1]); + /* Use QKEY from the QP context, which is set by master */ + dseg->qkey = cpu_to_be32(IB_QP_SET_QKEY); } static void build_tunnel_header(struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len) diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c index 1d74e8541146..e2ed36e2fb29 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.c +++ b/drivers/net/ethernet/mellanox/mlx4/fw.c @@ -184,8 +184,6 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave, #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET 0x28 #define QUERY_FUNC_CAP_MAX_EQ_OFFSET 0x2c #define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET 0X30 -#define QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET 0x44 -#define QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET 0x48 #define QUERY_FUNC_CAP_FMR_FLAG 0x80 #define QUERY_FUNC_CAP_FLAG_RDMA 0x40 @@ -196,21 +194,39 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave, #define QUERY_FUNC_CAP_RDMA_PROPS_OFFSET 0x8 #define QUERY_FUNC_CAP_ETH_PROPS_OFFSET 0xc +#define QUERY_FUNC_CAP_QP0_TUNNEL 0x10 +#define QUERY_FUNC_CAP_QP0_PROXY 0x14 +#define QUERY_FUNC_CAP_QP1_TUNNEL 0x18 +#define QUERY_FUNC_CAP_QP1_PROXY 0x1c + #define QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC 0x40 #define QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN 0x80 #define QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID 0x80 if (vhcr->op_modifier == 1) { - field = vhcr->in_modifier; - MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET); - field = 0; /* ensure force vlan and force mac bits are not set */ MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_ETH_PROPS_OFFSET); /* ensure that phy_wqe_gid bit is not set */ MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET); + field = vhcr->in_modifier; /* phys-port = logical-port */ + MLX4_PUT(outbox->buf, field, QUERY_FUNC_CAP_PHYS_PORT_OFFSET); + + /* size is now the QP number */ + size = dev->phys_caps.base_tunnel_sqpn + 8 * slave + field - 1; + MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_TUNNEL); + + size += 2; + MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_TUNNEL); + + size = dev->phys_caps.base_proxy_sqpn + 8 * slave + field - 1; + MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP0_PROXY); + + size += 2; + MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_QP1_PROXY); + } else if (vhcr->op_modifier == 0) { /* enable rdma and ethernet interfaces */ field = (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA); @@ -249,117 +265,124 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave, size = dev->caps.num_mgms + dev->caps.num_amgms; MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET); - size = dev->caps.base_tunnel_sqpn + 8 * slave; - MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET); - - size = dev->caps.sqp_start + 8 * slave; - MLX4_PUT(outbox->buf, size, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET); - } else err = -EINVAL; return err; } -int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap) +int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port, + struct mlx4_func_cap *func_cap) { struct mlx4_cmd_mailbox *mailbox; u32 *outbox; - u8 field; + u8 field, op_modifier; u32 size; - int i; int err = 0; + op_modifier = !!gen_or_port; /* 0 = general, 1 = logical port */ mailbox = mlx4_alloc_cmd_mailbox(dev); if (IS_ERR(mailbox)) return PTR_ERR(mailbox); - err = mlx4_cmd_box(dev, 0, mailbox->dma, 0, 0, MLX4_CMD_QUERY_FUNC_CAP, + err = mlx4_cmd_box(dev, 0, mailbox->dma, gen_or_port, op_modifier, + MLX4_CMD_QUERY_FUNC_CAP, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); if (err) goto out; outbox = mailbox->buf; - MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET); - if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) { - mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n"); - err = -EPROTONOSUPPORT; + if (!op_modifier) { + MLX4_GET(field, outbox, QUERY_FUNC_CAP_FLAGS_OFFSET); + if (!(field & (QUERY_FUNC_CAP_FLAG_ETH | QUERY_FUNC_CAP_FLAG_RDMA))) { + mlx4_err(dev, "The host supports neither eth nor rdma interfaces\n"); + err = -EPROTONOSUPPORT; + goto out; + } + func_cap->flags = field; + + MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET); + func_cap->num_ports = field; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET); + func_cap->pf_context_behaviour = size; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET); + func_cap->qp_quota = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET); + func_cap->srq_quota = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET); + func_cap->cq_quota = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET); + func_cap->max_eq = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET); + func_cap->reserved_eq = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET); + func_cap->mpt_quota = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET); + func_cap->mtt_quota = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET); + func_cap->mcg_quota = size & 0xFFFFFF; goto out; } - func_cap->flags = field; - MLX4_GET(field, outbox, QUERY_FUNC_CAP_NUM_PORTS_OFFSET); - func_cap->num_ports = field; + /* logical port query */ + if (gen_or_port > dev->caps.num_ports) { + err = -EINVAL; + goto out; + } - MLX4_GET(size, outbox, QUERY_FUNC_CAP_PF_BHVR_OFFSET); - func_cap->pf_context_behaviour = size; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP_QUOTA_OFFSET); - func_cap->qp_quota = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_SRQ_QUOTA_OFFSET); - func_cap->srq_quota = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_CQ_QUOTA_OFFSET); - func_cap->cq_quota = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_MAX_EQ_OFFSET); - func_cap->max_eq = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_RESERVED_EQ_OFFSET); - func_cap->reserved_eq = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_MPT_QUOTA_OFFSET); - func_cap->mpt_quota = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_MTT_QUOTA_OFFSET); - func_cap->mtt_quota = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_MCG_QUOTA_OFFSET); - func_cap->mcg_quota = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_TUNNEL_QPN_OFFSET); - func_cap->base_tunnel_qpn = size & 0xFFFFFF; - - MLX4_GET(size, outbox, QUERY_FUNC_CAP_BASE_PROXY_QPN_OFFSET); - func_cap->base_proxy_qpn = size & 0xFFFFFF; - - for (i = 1; i <= func_cap->num_ports; ++i) { - err = mlx4_cmd_box(dev, 0, mailbox->dma, i, 1, - MLX4_CMD_QUERY_FUNC_CAP, - MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED); - if (err) + if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_ETH) { + MLX4_GET(field, outbox, QUERY_FUNC_CAP_ETH_PROPS_OFFSET); + if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN) { + mlx4_err(dev, "VLAN is enforced on this port\n"); + err = -EPROTONOSUPPORT; goto out; - - if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH) { - MLX4_GET(field, outbox, QUERY_FUNC_CAP_ETH_PROPS_OFFSET); - if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_VLAN) { - mlx4_err(dev, "VLAN is enforced on this port\n"); - err = -EPROTONOSUPPORT; - goto out; - } - - if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC) { - mlx4_err(dev, "Force mac is enabled on this port\n"); - err = -EPROTONOSUPPORT; - goto out; - } - } else if (dev->caps.port_type[i] == MLX4_PORT_TYPE_IB) { - MLX4_GET(field, outbox, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET); - if (field & QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID) { - mlx4_err(dev, "phy_wqe_gid is " - "enforced on this ib port\n"); - err = -EPROTONOSUPPORT; - goto out; - } } - MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET); - func_cap->physical_port[i] = field; + if (field & QUERY_FUNC_CAP_ETH_PROPS_FORCE_MAC) { + mlx4_err(dev, "Force mac is enabled on this port\n"); + err = -EPROTONOSUPPORT; + goto out; + } + } else if (dev->caps.port_type[gen_or_port] == MLX4_PORT_TYPE_IB) { + MLX4_GET(field, outbox, QUERY_FUNC_CAP_RDMA_PROPS_OFFSET); + if (field & QUERY_FUNC_CAP_RDMA_PROPS_FORCE_PHY_WQE_GID) { + mlx4_err(dev, "phy_wqe_gid is " + "enforced on this ib port\n"); + err = -EPROTONOSUPPORT; + goto out; + } } + MLX4_GET(field, outbox, QUERY_FUNC_CAP_PHYS_PORT_OFFSET); + func_cap->physical_port = field; + if (func_cap->physical_port != gen_or_port) { + err = -ENOSYS; + goto out; + } + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_TUNNEL); + func_cap->qp0_tunnel_qpn = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP0_PROXY); + func_cap->qp0_proxy_qpn = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_TUNNEL); + func_cap->qp1_tunnel_qpn = size & 0xFFFFFF; + + MLX4_GET(size, outbox, QUERY_FUNC_CAP_QP1_PROXY); + func_cap->qp1_proxy_qpn = size & 0xFFFFFF; + /* All other resources are allocated by the master, but we still report * 'num' and 'reserved' capabilities as follows: * - num remains the maximum resource index diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.h b/drivers/net/ethernet/mellanox/mlx4/fw.h index ced1de57c818..85abe9c11a22 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.h +++ b/drivers/net/ethernet/mellanox/mlx4/fw.h @@ -134,11 +134,12 @@ struct mlx4_func_cap { int max_eq; int reserved_eq; int mcg_quota; - u32 base_qpn; - u32 base_tunnel_qpn; - u32 base_proxy_qpn; - u8 physical_port[MLX4_MAX_PORTS + 1]; - u8 port_flags[MLX4_MAX_PORTS + 1]; + u32 qp0_tunnel_qpn; + u32 qp0_proxy_qpn; + u32 qp1_tunnel_qpn; + u32 qp1_proxy_qpn; + u8 physical_port; + u8 port_flags; }; struct mlx4_adapter { @@ -195,7 +196,8 @@ struct mlx4_set_ib_param { }; int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap); -int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, struct mlx4_func_cap *func_cap); +int mlx4_QUERY_FUNC_CAP(struct mlx4_dev *dev, u32 gen_or_port, + struct mlx4_func_cap *func_cap); int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave, struct mlx4_vhcr *vhcr, struct mlx4_cmd_mailbox *inbox, diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 76f69fdd01d5..1e2ab9d00f1c 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -410,15 +410,16 @@ static int mlx4_how_many_lives_vf(struct mlx4_dev *dev) int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey) { u32 qk = MLX4_RESERVED_QKEY_BASE; - if (qpn >= dev->caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX || - qpn < dev->caps.sqp_start) + + if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX || + qpn < dev->phys_caps.base_proxy_sqpn) return -EINVAL; - if (qpn >= dev->caps.base_tunnel_sqpn) + if (qpn >= dev->phys_caps.base_tunnel_sqpn) /* tunnel qp */ - qk += qpn - dev->caps.base_tunnel_sqpn; + qk += qpn - dev->phys_caps.base_tunnel_sqpn; else - qk += qpn - dev->caps.sqp_start; + qk += qpn - dev->phys_caps.base_proxy_sqpn; *qkey = qk; return 0; } @@ -527,9 +528,10 @@ static int mlx4_slave_cap(struct mlx4_dev *dev) } memset(&func_cap, 0, sizeof(func_cap)); - err = mlx4_QUERY_FUNC_CAP(dev, &func_cap); + err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap); if (err) { - mlx4_err(dev, "QUERY_FUNC_CAP command failed, aborting.\n"); + mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d).\n", + err); return err; } @@ -557,12 +559,33 @@ static int mlx4_slave_cap(struct mlx4_dev *dev) return -ENODEV; } + dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + + if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy || + !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) { + err = -ENOMEM; + goto err_mem; + } + for (i = 1; i <= dev->caps.num_ports; ++i) { + err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap); + if (err) { + mlx4_err(dev, "QUERY_FUNC_CAP port command failed for" + " port %d, aborting (%d).\n", i, err); + goto err_mem; + } + dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn; + dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn; + dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn; + dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn; dev->caps.port_mask[i] = dev->caps.port_type[i]; if (mlx4_get_slave_pkey_gid_tbl_len(dev, i, &dev->caps.gid_table_len[i], &dev->caps.pkey_table_len[i])) - return -ENODEV; + goto err_mem; } if (dev->caps.uar_page_size * (dev->caps.num_uars - @@ -572,14 +595,20 @@ static int mlx4_slave_cap(struct mlx4_dev *dev) "PCI resource 2 size of 0x%llx, aborting.\n", dev->caps.uar_page_size * dev->caps.num_uars, (unsigned long long) pci_resource_len(dev->pdev, 2)); - return -ENODEV; + goto err_mem; } - /* Calculate our sqp_start */ - dev->caps.sqp_start = func_cap.base_proxy_qpn; - dev->caps.base_tunnel_sqpn = func_cap.base_tunnel_qpn; - return 0; + +err_mem: + kfree(dev->caps.qp0_tunnel); + kfree(dev->caps.qp0_proxy); + kfree(dev->caps.qp1_tunnel); + kfree(dev->caps.qp1_proxy); + dev->caps.qp0_tunnel = dev->caps.qp0_proxy = + dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL; + + return err; } /* @@ -2261,6 +2290,12 @@ static void mlx4_remove_one(struct pci_dev *pdev) if (!mlx4_is_slave(dev)) mlx4_free_ownership(dev); + + kfree(dev->caps.qp0_tunnel); + kfree(dev->caps.qp0_proxy); + kfree(dev->caps.qp1_tunnel); + kfree(dev->caps.qp1_proxy); + kfree(priv); pci_release_regions(pdev); pci_disable_device(pdev); diff --git a/drivers/net/ethernet/mellanox/mlx4/qp.c b/drivers/net/ethernet/mellanox/mlx4/qp.c index 436ef6c69add..81e2abe07bbb 100644 --- a/drivers/net/ethernet/mellanox/mlx4/qp.c +++ b/drivers/net/ethernet/mellanox/mlx4/qp.c @@ -68,15 +68,15 @@ void mlx4_qp_event(struct mlx4_dev *dev, u32 qpn, int event_type) } /* used for INIT/CLOSE port logic */ -static int is_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0) +static int is_master_qp0(struct mlx4_dev *dev, struct mlx4_qp *qp, int *real_qp0, int *proxy_qp0) { + /* this procedure is called after we already know we are on the master */ /* qp0 is either the proxy qp0, or the real qp0 */ - *proxy_qp0 = qp->qpn >= dev->caps.sqp_start && - qp->qpn <= dev->caps.sqp_start + 1; + u32 pf_proxy_offset = dev->phys_caps.base_proxy_sqpn + 8 * mlx4_master_func_num(dev); + *proxy_qp0 = qp->qpn >= pf_proxy_offset && qp->qpn <= pf_proxy_offset + 1; - *real_qp0 = mlx4_is_master(dev) && - qp->qpn >= dev->caps.base_sqpn && - qp->qpn <= dev->caps.base_sqpn + 1; + *real_qp0 = qp->qpn >= dev->phys_caps.base_sqpn && + qp->qpn <= dev->phys_caps.base_sqpn + 1; return *real_qp0 || *proxy_qp0; } @@ -143,7 +143,7 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, MLX4_CMD_2RST_QP, MLX4_CMD_TIME_CLASS_A, native); if (mlx4_is_master(dev) && cur_state != MLX4_QP_STATE_ERR && cur_state != MLX4_QP_STATE_RST && - is_qp0(dev, qp, &real_qp0, &proxy_qp0)) { + is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) { port = (qp->qpn & 1) + 1; if (proxy_qp0) priv->mfunc.master.qp0_state[port].proxy_qp0_active = 0; @@ -175,7 +175,7 @@ static int __mlx4_qp_modify(struct mlx4_dev *dev, struct mlx4_mtt *mtt, new_state == MLX4_QP_STATE_RST ? 2 : 0, op[cur_state][new_state], MLX4_CMD_TIME_CLASS_C, native); - if (mlx4_is_master(dev) && is_qp0(dev, qp, &real_qp0, &proxy_qp0)) { + if (mlx4_is_master(dev) && is_master_qp0(dev, qp, &real_qp0, &proxy_qp0)) { port = (qp->qpn & 1) + 1; if (cur_state != MLX4_QP_STATE_ERR && cur_state != MLX4_QP_STATE_RST && @@ -422,6 +422,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev) struct mlx4_qp_table *qp_table = &mlx4_priv(dev)->qp_table; int err; int reserved_from_top = 0; + int k; spin_lock_init(&qp_table->lock); INIT_RADIX_TREE(&dev->qp_table_tree, GFP_ATOMIC); @@ -436,7 +437,7 @@ int mlx4_init_qp_table(struct mlx4_dev *dev) * We also reserve the MSB of the 24-bit QP number to indicate * that a QP is an XRC QP. */ - dev->caps.base_sqpn = + dev->phys_caps.base_sqpn = ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8); { @@ -479,24 +480,54 @@ int mlx4_init_qp_table(struct mlx4_dev *dev) */ err = mlx4_bitmap_init(&qp_table->bitmap, dev->caps.num_qps, - (1 << 23) - 1, dev->caps.base_sqpn + 8 + + (1 << 23) - 1, dev->phys_caps.base_sqpn + 8 + 16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev), reserved_from_top); - - /* In mfunc, sqp_start is the base of the proxy SQPs, since the PF also - * uses paravirtualized SQPs. - * In native mode, sqp_start is the base of the real SQPs. */ - if (mlx4_is_mfunc(dev)) { - dev->caps.sqp_start = dev->caps.base_sqpn + - 8 * (mlx4_master_func_num(dev) + 1); - dev->caps.base_tunnel_sqpn = dev->caps.sqp_start + 8 * MLX4_MFUNC_MAX; - } else - dev->caps.sqp_start = dev->caps.base_sqpn; - if (err) return err; - return mlx4_CONF_SPECIAL_QP(dev, dev->caps.base_sqpn); + if (mlx4_is_mfunc(dev)) { + /* for PPF use */ + dev->phys_caps.base_proxy_sqpn = dev->phys_caps.base_sqpn + 8; + dev->phys_caps.base_tunnel_sqpn = dev->phys_caps.base_sqpn + 8 + 8 * MLX4_MFUNC_MAX; + + /* In mfunc, calculate proxy and tunnel qp offsets for the PF here, + * since the PF does not call mlx4_slave_caps */ + dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL); + + if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy || + !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) { + err = -ENOMEM; + goto err_mem; + } + + for (k = 0; k < dev->caps.num_ports; k++) { + dev->caps.qp0_proxy[k] = dev->phys_caps.base_proxy_sqpn + + 8 * mlx4_master_func_num(dev) + k; + dev->caps.qp0_tunnel[k] = dev->caps.qp0_proxy[k] + 8 * MLX4_MFUNC_MAX; + dev->caps.qp1_proxy[k] = dev->phys_caps.base_proxy_sqpn + + 8 * mlx4_master_func_num(dev) + MLX4_MAX_PORTS + k; + dev->caps.qp1_tunnel[k] = dev->caps.qp1_proxy[k] + 8 * MLX4_MFUNC_MAX; + } + } + + + err = mlx4_CONF_SPECIAL_QP(dev, dev->phys_caps.base_sqpn); + if (err) + goto err_mem; + return 0; + +err_mem: + kfree(dev->caps.qp0_tunnel); + kfree(dev->caps.qp0_proxy); + kfree(dev->caps.qp1_tunnel); + kfree(dev->caps.qp1_proxy); + dev->caps.qp0_tunnel = dev->caps.qp0_proxy = + dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL; + return err; } void mlx4_cleanup_qp_table(struct mlx4_dev *dev) diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index c0f8f74a0a5e..6d1acb04cd17 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -328,6 +328,9 @@ struct mlx4_phys_caps { u32 gid_phys_table_len[MLX4_MAX_PORTS + 1]; u32 pkey_phys_table_len[MLX4_MAX_PORTS + 1]; u32 num_phys_eqs; + u32 base_sqpn; + u32 base_proxy_sqpn; + u32 base_tunnel_sqpn; }; struct mlx4_caps { @@ -358,9 +361,10 @@ struct mlx4_caps { int max_rq_desc_sz; int max_qp_init_rdma; int max_qp_dest_rdma; - int sqp_start; - u32 base_sqpn; - u32 base_tunnel_sqpn; + u32 *qp0_proxy; + u32 *qp1_proxy; + u32 *qp0_tunnel; + u32 *qp1_tunnel; int num_srqs; int max_srq_wqes; int max_srq_sge; @@ -722,15 +726,15 @@ static inline int mlx4_is_master(struct mlx4_dev *dev) static inline int mlx4_is_qp_reserved(struct mlx4_dev *dev, u32 qpn) { - return (qpn < dev->caps.base_sqpn + 8 + + return (qpn < dev->phys_caps.base_sqpn + 8 + 16 * MLX4_MFUNC_MAX * !!mlx4_is_master(dev)); } static inline int mlx4_is_guest_proxy(struct mlx4_dev *dev, int slave, u32 qpn) { - int base = dev->caps.sqp_start + slave * 8; + int guest_proxy_base = dev->phys_caps.base_proxy_sqpn + slave * 8; - if (qpn >= base && qpn < base + 8) + if (qpn >= guest_proxy_base && qpn < guest_proxy_base + 8) return 1; return 0; From 3806d08cf6c42193c10c4963ca2d68f56a088668 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Fri, 3 Aug 2012 08:40:58 +0000 Subject: [PATCH 29/45] IB/mlx4: Create paravirt contexts for VFs when master IB driver initializes When we have VFs and PFs on same host, the VFs are activated within the mlx4_core module before the mlx4_ib kernel module is loaded. When the mlx4_ib module initializes the PF (master), it now creates MAD paravirtualization contexts for any VFs that already active. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/mad.c | 3 +++ drivers/infiniband/hw/mlx4/main.c | 11 +++++++++++ 2 files changed, 14 insertions(+) diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c index 658a622791fb..21a794152d15 100644 --- a/drivers/infiniband/hw/mlx4/mad.c +++ b/drivers/infiniband/hw/mlx4/mad.c @@ -1652,6 +1652,9 @@ static int create_pv_resources(struct ib_device *ibdev, int slave, int port, { int ret, cq_size; + if (ctx->state != DEMUX_PV_STATE_DOWN) + return -EEXIST; + ctx->state = DEMUX_PV_STATE_STARTING; /* have QP0 only on port owner, and only if link layer is IB */ if (ctx->slave == mlx4_master_func_num(to_mdev(ctx->ib_dev)->dev) && diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c index e849347ef99a..718ec6b2bad2 100644 --- a/drivers/infiniband/hw/mlx4/main.c +++ b/drivers/infiniband/hw/mlx4/main.c @@ -74,6 +74,8 @@ struct update_gid_work { int port; }; +static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init); + static struct workqueue_struct *wq; static void init_query_mad(struct ib_smp *mad) @@ -1470,6 +1472,15 @@ static void *mlx4_ib_add(struct mlx4_dev *dev) if (mlx4_is_mfunc(ibdev->dev)) init_pkeys(ibdev); + /* create paravirt contexts for any VFs which are active */ + if (mlx4_is_master(ibdev->dev)) { + for (j = 0; j < MLX4_MFUNC_MAX; j++) { + if (j == mlx4_master_func_num(ibdev->dev)) + continue; + if (mlx4_is_slave_active(ibdev->dev, j)) + do_slave_init(ibdev, j, 1); + } + } return ibdev; err_notif: From 69612b9ff44ddc305b727a1f78de2248f9d70bf0 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Sun, 23 Sep 2012 09:18:24 -0700 Subject: [PATCH 30/45] mlx4_core: Trivial readability fix: "0X30" -> "0x30" Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/fw.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c index e2ed36e2fb29..419914bf4834 100644 --- a/drivers/net/ethernet/mellanox/mlx4/fw.c +++ b/drivers/net/ethernet/mellanox/mlx4/fw.c @@ -183,7 +183,7 @@ int mlx4_QUERY_FUNC_CAP_wrapper(struct mlx4_dev *dev, int slave, #define QUERY_FUNC_CAP_MTT_QUOTA_OFFSET 0x24 #define QUERY_FUNC_CAP_MCG_QUOTA_OFFSET 0x28 #define QUERY_FUNC_CAP_MAX_EQ_OFFSET 0x2c -#define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET 0X30 +#define QUERY_FUNC_CAP_RESERVED_EQ_OFFSET 0x30 #define QUERY_FUNC_CAP_FMR_FLAG 0x80 #define QUERY_FUNC_CAP_FLAG_RDMA 0x40 From 84b1f1538fadd066cc9877c2276f3fefd8a79bc4 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Tue, 25 Sep 2012 17:09:42 -0700 Subject: [PATCH 31/45] mlx4_core: Trivial cleanups to driver log messages Also put format string onto one line. Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/main.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 1e2ab9d00f1c..296288c7cb37 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -2036,12 +2036,11 @@ static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id) } if (num_vfs) { - mlx4_warn(dev, "Enabling sriov with:%d vfs\n", num_vfs); + mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", num_vfs); err = pci_enable_sriov(pdev, num_vfs); if (err) { - mlx4_err(dev, "Failed to enable sriov," - "continuing without sriov enabled" - " (err = %d).\n", err); + mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d).\n", + err); err = 0; } else { mlx4_warn(dev, "Running in master mode\n"); @@ -2284,7 +2283,7 @@ static void mlx4_remove_one(struct pci_dev *pdev) if (dev->flags & MLX4_FLAG_MSI_X) pci_disable_msix(pdev); if (dev->flags & MLX4_FLAG_SRIOV) { - mlx4_warn(dev, "Disabling sriov\n"); + mlx4_warn(dev, "Disabling SR-IOV\n"); pci_disable_sriov(pdev); } From f3d4c89ee4a8c993cc334a67f84a3fb724a1dd35 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Tue, 25 Sep 2012 21:24:07 -0700 Subject: [PATCH 32/45] mlx4_core: Fix crash on uninitialized priv->cmd.slave_sem On an SR-IOV master device, __mlx4_init_one() calls mlx4_init_hca() before mlx4_multi_func_init(). However, for unlucky configurations, mlx4_init_hca() might call mlx4_SENSE_PORT() (via mlx4_dev_cap()), and that calls mlx4_cmd_imm() with MLX4_CMD_WRAPPED set. However, on a multifunction device with MLX4_CMD_WRAPPED, __mlx4_cmd() calls into mlx4_slave_cmd(), and that immediately tries to do down(&priv->cmd.slave_sem); but priv->cmd.slave_sem isn't initialized until mlx4_multi_func_init() (which we haven't called yet). The next thing it tries to do is access priv->mfunc.vhcr, but that hasn't been allocated yet. Fix this by moving the initialization of slave_sem and vhcr up into mlx4_cmd_init(). Also, since slave_sem is really just being used as a mutex, convert it into a slave_cmd_mutex. Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/cmd.c | 51 ++++++++++++++--------- drivers/net/ethernet/mellanox/mlx4/main.c | 11 ++--- drivers/net/ethernet/mellanox/mlx4/mlx4.h | 2 +- 3 files changed, 38 insertions(+), 26 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/cmd.c b/drivers/net/ethernet/mellanox/mlx4/cmd.c index 90774b7e47a4..3d1899ff1076 100644 --- a/drivers/net/ethernet/mellanox/mlx4/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx4/cmd.c @@ -395,7 +395,8 @@ static int mlx4_slave_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param, struct mlx4_vhcr_cmd *vhcr = priv->mfunc.vhcr; int ret; - down(&priv->cmd.slave_sem); + mutex_lock(&priv->cmd.slave_cmd_mutex); + vhcr->in_param = cpu_to_be64(in_param); vhcr->out_param = out_param ? cpu_to_be64(*out_param) : 0; vhcr->in_modifier = cpu_to_be32(in_modifier); @@ -403,6 +404,7 @@ static int mlx4_slave_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param, vhcr->token = cpu_to_be16(CMD_POLL_TOKEN); vhcr->status = 0; vhcr->flags = !!(priv->cmd.use_events) << 6; + if (mlx4_is_master(dev)) { ret = mlx4_master_process_vhcr(dev, dev->caps.function, vhcr); if (!ret) { @@ -439,7 +441,8 @@ static int mlx4_slave_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param, mlx4_err(dev, "failed execution of VHCR_POST command" "opcode 0x%x\n", op); } - up(&priv->cmd.slave_sem); + + mutex_unlock(&priv->cmd.slave_cmd_mutex); return ret; } @@ -1559,14 +1562,15 @@ static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd, if ((slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_EN) && (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_POST)) goto reset_slave; - down(&priv->cmd.slave_sem); + + mutex_lock(&priv->cmd.slave_cmd_mutex); if (mlx4_master_process_vhcr(dev, slave, NULL)) { mlx4_err(dev, "Failed processing vhcr for slave:%d," " resetting slave.\n", slave); - up(&priv->cmd.slave_sem); + mutex_unlock(&priv->cmd.slave_cmd_mutex); goto reset_slave; } - up(&priv->cmd.slave_sem); + mutex_unlock(&priv->cmd.slave_cmd_mutex); break; default: mlx4_warn(dev, "Bad comm cmd:%d from slave:%d\n", cmd, slave); @@ -1707,14 +1711,6 @@ int mlx4_multi_func_init(struct mlx4_dev *dev) struct mlx4_slave_state *s_state; int i, j, err, port; - priv->mfunc.vhcr = dma_alloc_coherent(&(dev->pdev->dev), PAGE_SIZE, - &priv->mfunc.vhcr_dma, - GFP_KERNEL); - if (!priv->mfunc.vhcr) { - mlx4_err(dev, "Couldn't allocate vhcr.\n"); - return -ENOMEM; - } - if (mlx4_is_master(dev)) priv->mfunc.comm = ioremap(pci_resource_start(dev->pdev, priv->fw.comm_bar) + @@ -1777,7 +1773,6 @@ int mlx4_multi_func_init(struct mlx4_dev *dev) if (mlx4_init_resource_tracker(dev)) goto err_thread; - sema_init(&priv->cmd.slave_sem, 1); err = mlx4_ARM_COMM_CHANNEL(dev); if (err) { mlx4_err(dev, " Failed to arm comm channel eq: %x\n", @@ -1791,8 +1786,6 @@ int mlx4_multi_func_init(struct mlx4_dev *dev) mlx4_err(dev, "Couldn't sync toggles\n"); goto err_comm; } - - sema_init(&priv->cmd.slave_sem, 1); } return 0; @@ -1822,6 +1815,7 @@ int mlx4_cmd_init(struct mlx4_dev *dev) struct mlx4_priv *priv = mlx4_priv(dev); mutex_init(&priv->cmd.hcr_mutex); + mutex_init(&priv->cmd.slave_cmd_mutex); sema_init(&priv->cmd.poll_sem, 1); priv->cmd.use_events = 0; priv->cmd.toggle = 1; @@ -1838,14 +1832,30 @@ int mlx4_cmd_init(struct mlx4_dev *dev) } } + if (mlx4_is_mfunc(dev)) { + priv->mfunc.vhcr = dma_alloc_coherent(&(dev->pdev->dev), PAGE_SIZE, + &priv->mfunc.vhcr_dma, + GFP_KERNEL); + if (!priv->mfunc.vhcr) { + mlx4_err(dev, "Couldn't allocate VHCR.\n"); + goto err_hcr; + } + } + priv->cmd.pool = pci_pool_create("mlx4_cmd", dev->pdev, MLX4_MAILBOX_SIZE, MLX4_MAILBOX_SIZE, 0); if (!priv->cmd.pool) - goto err_hcr; + goto err_vhcr; return 0; +err_vhcr: + if (mlx4_is_mfunc(dev)) + dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE, + priv->mfunc.vhcr, priv->mfunc.vhcr_dma); + priv->mfunc.vhcr = NULL; + err_hcr: if (!mlx4_is_slave(dev)) iounmap(priv->cmd.hcr); @@ -1868,9 +1878,6 @@ void mlx4_multi_func_cleanup(struct mlx4_dev *dev) } iounmap(priv->mfunc.comm); - dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE, - priv->mfunc.vhcr, priv->mfunc.vhcr_dma); - priv->mfunc.vhcr = NULL; } void mlx4_cmd_cleanup(struct mlx4_dev *dev) @@ -1881,6 +1888,10 @@ void mlx4_cmd_cleanup(struct mlx4_dev *dev) if (!mlx4_is_slave(dev)) iounmap(priv->cmd.hcr); + if (mlx4_is_mfunc(dev)) + dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE, + priv->mfunc.vhcr, priv->mfunc.vhcr_dma); + priv->mfunc.vhcr = NULL; } /* diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 296288c7cb37..1e9f816a94db 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -1159,10 +1159,10 @@ static void mlx4_slave_exit(struct mlx4_dev *dev) { struct mlx4_priv *priv = mlx4_priv(dev); - down(&priv->cmd.slave_sem); + mutex_lock(&priv->cmd.slave_cmd_mutex); if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME)) mlx4_warn(dev, "Failed to close slave function.\n"); - up(&priv->cmd.slave_sem); + mutex_unlock(&priv->cmd.slave_cmd_mutex); } static int map_bf_area(struct mlx4_dev *dev) @@ -1214,7 +1214,7 @@ static int mlx4_init_slave(struct mlx4_dev *dev) u32 slave_read; u32 cmd_channel_ver; - down(&priv->cmd.slave_sem); + mutex_lock(&priv->cmd.slave_cmd_mutex); priv->cmd.max_cmds = 1; mlx4_warn(dev, "Sending reset\n"); ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, @@ -1263,12 +1263,13 @@ static int mlx4_init_slave(struct mlx4_dev *dev) goto err; if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME)) goto err; - up(&priv->cmd.slave_sem); + + mutex_unlock(&priv->cmd.slave_cmd_mutex); return 0; err: mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0); - up(&priv->cmd.slave_sem); + mutex_unlock(&priv->cmd.slave_cmd_mutex); return -EIO; } diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h index 4b1fbd84a68f..71eed05426ec 100644 --- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h +++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h @@ -513,9 +513,9 @@ struct mlx4_cmd { struct pci_pool *pool; void __iomem *hcr; struct mutex hcr_mutex; + struct mutex slave_cmd_mutex; struct semaphore poll_sem; struct semaphore event_sem; - struct semaphore slave_sem; int max_cmds; spinlock_t context_lock; int free_head; From ef3d0c4a5e8c7ad3429b9f9ef66cf5a7563cd513 Mon Sep 17 00:00:00 2001 From: Tatyana Nikolova Date: Thu, 20 Sep 2012 19:23:14 +0000 Subject: [PATCH 33/45] RDMA/nes: Fix incorrect resolving of the loopback MAC address Signed-off-by: Tatyana Nikolova Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 32 +++++++----------------------- 1 file changed, 7 insertions(+), 25 deletions(-) diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 020e95c4c4b9..49a9383137db 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -1356,7 +1356,7 @@ static int nes_addr_resolve_neigh(struct nes_vnic *nesvnic, u32 dst_ip, int arpi else netdev = nesvnic->netdev; - neigh = dst_neigh_lookup(&rt->dst, &dst_ip); + neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, netdev); rcu_read_lock(); if (neigh) { @@ -1465,12 +1465,8 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core, cm_node->loopbackpartner = NULL; /* get the mac addr for the remote node */ - if (ipv4_is_loopback(htonl(cm_node->rem_addr))) { - arpindex = nes_arp_table(nesdev, ntohl(nesvnic->local_ipaddr), NULL, NES_ARP_RESOLVE); - } else { - oldarpindex = nes_arp_table(nesdev, cm_node->rem_addr, NULL, NES_ARP_RESOLVE); - arpindex = nes_addr_resolve_neigh(nesvnic, cm_info->rem_addr, oldarpindex); - } + oldarpindex = nes_arp_table(nesdev, cm_node->rem_addr, NULL, NES_ARP_RESOLVE); + arpindex = nes_addr_resolve_neigh(nesvnic, cm_info->rem_addr, oldarpindex); if (arpindex < 0) { kfree(cm_node); return NULL; @@ -3153,11 +3149,7 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) nesqp->nesqp_context->tcpPorts[1] = cpu_to_le16(ntohs(cm_id->remote_addr.sin_port)); - if (ipv4_is_loopback(cm_id->remote_addr.sin_addr.s_addr)) - nesqp->nesqp_context->ip0 = - cpu_to_le32(ntohl(nesvnic->local_ipaddr)); - else - nesqp->nesqp_context->ip0 = + nesqp->nesqp_context->ip0 = cpu_to_le32(ntohl(cm_id->remote_addr.sin_addr.s_addr)); nesqp->nesqp_context->misc2 |= cpu_to_le32( @@ -3182,10 +3174,7 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) memset(&nes_quad, 0, sizeof(nes_quad)); nes_quad.DstIpAdrIndex = cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24); - if (ipv4_is_loopback(cm_id->remote_addr.sin_addr.s_addr)) - nes_quad.SrcIpadr = nesvnic->local_ipaddr; - else - nes_quad.SrcIpadr = cm_id->remote_addr.sin_addr.s_addr; + nes_quad.SrcIpadr = cm_id->remote_addr.sin_addr.s_addr; nes_quad.TcpPorts[0] = cm_id->remote_addr.sin_port; nes_quad.TcpPorts[1] = cm_id->local_addr.sin_port; @@ -3538,11 +3527,7 @@ static void cm_event_connected(struct nes_cm_event *event) cpu_to_le16(ntohs(cm_id->local_addr.sin_port)); nesqp->nesqp_context->tcpPorts[1] = cpu_to_le16(ntohs(cm_id->remote_addr.sin_port)); - if (ipv4_is_loopback(cm_id->remote_addr.sin_addr.s_addr)) - nesqp->nesqp_context->ip0 = - cpu_to_le32(ntohl(nesvnic->local_ipaddr)); - else - nesqp->nesqp_context->ip0 = + nesqp->nesqp_context->ip0 = cpu_to_le32(ntohl(cm_id->remote_addr.sin_addr.s_addr)); nesqp->nesqp_context->misc2 |= cpu_to_le32( @@ -3571,10 +3556,7 @@ static void cm_event_connected(struct nes_cm_event *event) nes_quad.DstIpAdrIndex = cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24); - if (ipv4_is_loopback(cm_id->remote_addr.sin_addr.s_addr)) - nes_quad.SrcIpadr = nesvnic->local_ipaddr; - else - nes_quad.SrcIpadr = cm_id->remote_addr.sin_addr.s_addr; + nes_quad.SrcIpadr = cm_id->remote_addr.sin_addr.s_addr; nes_quad.TcpPorts[0] = cm_id->remote_addr.sin_port; nes_quad.TcpPorts[1] = cm_id->local_addr.sin_port; From 6ad1be814b106e182d9b12898280d0c115541b72 Mon Sep 17 00:00:00 2001 From: Tatyana Nikolova Date: Thu, 20 Sep 2012 19:34:20 +0000 Subject: [PATCH 34/45] RDMA/nes: Fix for incorrect MSS when TSO is on In TSO handling code, skb_shared_info() is used to get the MSS instead of the bool function skb_is_gso() (which always returns 1). Signed-off-by: Tatyana Nikolova Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_nic.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/drivers/infiniband/hw/nes/nes_nic.c b/drivers/infiniband/hw/nes/nes_nic.c index f3a3ecf8d09e..1c02ba787536 100644 --- a/drivers/infiniband/hw/nes/nes_nic.c +++ b/drivers/infiniband/hw/nes/nes_nic.c @@ -390,10 +390,10 @@ static int nes_nic_send(struct sk_buff *skb, struct net_device *netdev) tcph = tcp_hdr(skb); if (1) { if (skb_is_gso(skb)) { - /* nes_debug(NES_DBG_NIC_TX, "%s: TSO request... seg size = %u\n", - netdev->name, skb_is_gso(skb)); */ + /* nes_debug(NES_DBG_NIC_TX, "%s: TSO request... is_gso = %u seg size = %u\n", + netdev->name, skb_is_gso(skb), skb_shinfo(skb)->gso_size); */ wqe_misc |= NES_NIC_SQ_WQE_LSO_ENABLE | - NES_NIC_SQ_WQE_COMPLETION | (u16)skb_is_gso(skb); + NES_NIC_SQ_WQE_COMPLETION | (u16)skb_shinfo(skb)->gso_size; set_wqe_32bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_LSO_INFO_IDX, ((u32)tcph->doff) | (((u32)(((unsigned char *)tcph) - skb->data)) << 4)); @@ -597,10 +597,10 @@ tso_sq_no_longer_full: nes_debug(NES_DBG_NIC_TX, "ERROR: SKB header too big, headlen=%u, FIRST_FRAG_SIZE=%u\n", original_first_length, NES_FIRST_FRAG_SIZE); nes_debug(NES_DBG_NIC_TX, "%s Request to tx NIC packet length %u, headlen %u," - " (%u frags), tso_size=%u\n", + " (%u frags), is_gso = %u tso_size=%u\n", netdev->name, skb->len, skb_headlen(skb), - skb_shinfo(skb)->nr_frags, skb_is_gso(skb)); + skb_shinfo(skb)->nr_frags, skb_is_gso(skb), skb_shinfo(skb)->gso_size); } memcpy(&nesnic->first_frag_vbase[nesnic->sq_head].buffer, skb->data, min(((unsigned int)NES_FIRST_FRAG_SIZE), @@ -652,8 +652,8 @@ tso_sq_no_longer_full: } else { nesnic->tx_skb[nesnic->sq_head] = NULL; } - wqe_misc |= NES_NIC_SQ_WQE_COMPLETION | (u16)skb_is_gso(skb); - if ((tso_wqe_length + original_first_length) > skb_is_gso(skb)) { + wqe_misc |= NES_NIC_SQ_WQE_COMPLETION | (u16)skb_shinfo(skb)->gso_size; + if ((tso_wqe_length + original_first_length) > skb_shinfo(skb)->gso_size) { wqe_misc |= NES_NIC_SQ_WQE_LSO_ENABLE; } else { iph->tot_len = htons(tso_wqe_length + original_first_length - nhoffset); From 48a9956362c27e717a76435d450d906d5f49344a Mon Sep 17 00:00:00 2001 From: Tatyana Nikolova Date: Thu, 20 Sep 2012 20:53:27 +0000 Subject: [PATCH 35/45] RDMA/nes: Cosmetic changes - Remove unnecessary statement "if (1)" - Refactor a statement (wqe_misc |= NES_NIC_SQ_WQE_COMPLETION) out of if/else statement, because it is independant of the flow. - Define netdev->features in one line for clarity. Signed-off-by: Tatyana Nikolova Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_nic.c | 32 ++++++++++++----------------- 1 file changed, 13 insertions(+), 19 deletions(-) diff --git a/drivers/infiniband/hw/nes/nes_nic.c b/drivers/infiniband/hw/nes/nes_nic.c index 1c02ba787536..d0391ac21d9b 100644 --- a/drivers/infiniband/hw/nes/nes_nic.c +++ b/drivers/infiniband/hw/nes/nes_nic.c @@ -385,24 +385,20 @@ static int nes_nic_send(struct sk_buff *skb, struct net_device *netdev) /* bump past the vlan tag */ wqe_fragment_length++; /* wqe_fragment_address = (u64 *)&nic_sqe->wqe_words[NES_NIC_SQ_WQE_FRAG0_LOW_IDX]; */ + wqe_misc |= NES_NIC_SQ_WQE_COMPLETION; if (skb->ip_summed == CHECKSUM_PARTIAL) { - tcph = tcp_hdr(skb); - if (1) { - if (skb_is_gso(skb)) { - /* nes_debug(NES_DBG_NIC_TX, "%s: TSO request... is_gso = %u seg size = %u\n", - netdev->name, skb_is_gso(skb), skb_shinfo(skb)->gso_size); */ - wqe_misc |= NES_NIC_SQ_WQE_LSO_ENABLE | - NES_NIC_SQ_WQE_COMPLETION | (u16)skb_shinfo(skb)->gso_size; - set_wqe_32bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_LSO_INFO_IDX, - ((u32)tcph->doff) | - (((u32)(((unsigned char *)tcph) - skb->data)) << 4)); - } else { - wqe_misc |= NES_NIC_SQ_WQE_COMPLETION; - } + if (skb_is_gso(skb)) { + tcph = tcp_hdr(skb); + /* nes_debug(NES_DBG_NIC_TX, "%s: TSO request... is_gso = %u seg size = %u\n", + netdev->name, skb_is_gso(skb), skb_shinfo(skb)->gso_size); */ + wqe_misc |= NES_NIC_SQ_WQE_LSO_ENABLE | (u16)skb_shinfo(skb)->gso_size; + set_wqe_32bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_LSO_INFO_IDX, + ((u32)tcph->doff) | + (((u32)(((unsigned char *)tcph) - skb->data)) << 4)); } } else { /* CHECKSUM_HW */ - wqe_misc |= NES_NIC_SQ_WQE_DISABLE_CHKSUM | NES_NIC_SQ_WQE_COMPLETION; + wqe_misc |= NES_NIC_SQ_WQE_DISABLE_CHKSUM; } set_wqe_32bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_TOTAL_LENGTH_IDX, @@ -1679,12 +1675,10 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev, netdev->hard_header_len = ETH_HLEN; netdev->addr_len = ETH_ALEN; netdev->type = ARPHRD_ETHER; - netdev->features = NETIF_F_HIGHDMA; netdev->netdev_ops = &nes_netdev_ops; netdev->ethtool_ops = &nes_ethtool_ops; netif_napi_add(netdev, &nesvnic->napi, nes_netdev_poll, 128); nes_debug(NES_DBG_INIT, "Enabling VLAN Insert/Delete.\n"); - netdev->features |= NETIF_F_HW_VLAN_TX; /* Fill in the port structure */ nesvnic->netdev = netdev; @@ -1711,11 +1705,11 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev, netdev->dev_addr[5] = (u8)u64temp; memcpy(netdev->perm_addr, netdev->dev_addr, 6); - netdev->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM | - NETIF_F_HW_VLAN_RX; + netdev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_RXCSUM | NETIF_F_HW_VLAN_RX; if ((nesvnic->logical_port < 2) || (nesdev->nesadapter->hw_rev != NE020_REV)) netdev->hw_features |= NETIF_F_TSO; - netdev->features |= netdev->hw_features; + + netdev->features = netdev->hw_features | NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_TX; netdev->hw_features |= NETIF_F_LRO; nes_debug(NES_DBG_INIT, "nesvnic = %p, reported features = 0x%lX, QPid = %d," From fc4ba7291b3dc7ace34794f51c47f041949c7a92 Mon Sep 17 00:00:00 2001 From: Tatyana Nikolova Date: Thu, 20 Sep 2012 20:55:33 +0000 Subject: [PATCH 36/45] RDMA/nes: Fix for crash when TX checksum offload is off When TX checksum offload is disabled for an iWarp connection, skb->ip_summed needs to be set to CHECKSUM_NONE. Signed-off-by: Tatyana Nikolova Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 49a9383137db..cfaacaf6bf5f 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -430,6 +430,8 @@ static void form_cm_frame(struct sk_buff *skb, buf += sizeof(*tcph); skb->ip_summed = CHECKSUM_PARTIAL; + if (!(cm_node->netdev->features & NETIF_F_IP_CSUM)) + skb->ip_summed = CHECKSUM_NONE; skb->protocol = htons(0x800); skb->data_len = 0; skb->mac_len = ETH_HLEN; From 818216442b6dd49fa93a4caf9d7b56af364b502b Mon Sep 17 00:00:00 2001 From: Tatyana Nikolova Date: Thu, 20 Sep 2012 21:08:04 +0000 Subject: [PATCH 37/45] RDMA/nes: Print hardware resource type Hardware resource types are added and when a resource isn't available, its type is printed. Signed-off-by: Tatyana Nikolova Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes.h | 15 ++++++++++++--- drivers/infiniband/hw/nes/nes_utils.c | 2 +- drivers/infiniband/hw/nes/nes_verbs.c | 14 +++++++------- 3 files changed, 20 insertions(+), 11 deletions(-) diff --git a/drivers/infiniband/hw/nes/nes.h b/drivers/infiniband/hw/nes/nes.h index c438e4691b3c..0da62b904d00 100644 --- a/drivers/infiniband/hw/nes/nes.h +++ b/drivers/infiniband/hw/nes/nes.h @@ -399,11 +399,20 @@ static inline void nes_write8(void __iomem *addr, u8 val) writeb(val, addr); } - +enum nes_resource { + NES_RESOURCE_MW = 1, + NES_RESOURCE_FAST_MR, + NES_RESOURCE_PHYS_MR, + NES_RESOURCE_USER_MR, + NES_RESOURCE_PD, + NES_RESOURCE_QP, + NES_RESOURCE_CQ, + NES_RESOURCE_ARP +}; static inline int nes_alloc_resource(struct nes_adapter *nesadapter, unsigned long *resource_array, u32 max_resources, - u32 *req_resource_num, u32 *next) + u32 *req_resource_num, u32 *next, enum nes_resource resource_type) { unsigned long flags; u32 resource_num; @@ -414,7 +423,7 @@ static inline int nes_alloc_resource(struct nes_adapter *nesadapter, if (resource_num >= max_resources) { resource_num = find_first_zero_bit(resource_array, max_resources); if (resource_num >= max_resources) { - printk(KERN_ERR PFX "%s: No available resourcess.\n", __func__); + printk(KERN_ERR PFX "%s: No available resources [type=%u].\n", __func__, resource_type); spin_unlock_irqrestore(&nesadapter->resource_lock, flags); return -EMFILE; } diff --git a/drivers/infiniband/hw/nes/nes_utils.c b/drivers/infiniband/hw/nes/nes_utils.c index e98f4fc0b768..2042c0f29759 100644 --- a/drivers/infiniband/hw/nes/nes_utils.c +++ b/drivers/infiniband/hw/nes/nes_utils.c @@ -699,7 +699,7 @@ int nes_arp_table(struct nes_device *nesdev, u32 ip_addr, u8 *mac_addr, u32 acti arp_index = 0; err = nes_alloc_resource(nesadapter, nesadapter->allocated_arps, - nesadapter->arp_table_size, (u32 *)&arp_index, &nesadapter->next_arp_index); + nesadapter->arp_table_size, (u32 *)&arp_index, &nesadapter->next_arp_index, NES_RESOURCE_ARP); if (err) { nes_debug(NES_DBG_NETDEV, "nes_alloc_resource returned error = %u\n", err); return err; diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c index 8b8812de4b5c..1dadcf388c02 100644 --- a/drivers/infiniband/hw/nes/nes_verbs.c +++ b/drivers/infiniband/hw/nes/nes_verbs.c @@ -80,7 +80,7 @@ static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd) { next_stag_index %= nesadapter->max_mr; ret = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs, - nesadapter->max_mr, &stag_index, &next_stag_index); + nesadapter->max_mr, &stag_index, &next_stag_index, NES_RESOURCE_MW); if (ret) { return ERR_PTR(ret); } @@ -404,7 +404,7 @@ static struct ib_mr *nes_alloc_fast_reg_mr(struct ib_pd *ibpd, int max_page_list err = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs, nesadapter->max_mr, &stag_index, - &next_stag_index); + &next_stag_index, NES_RESOURCE_FAST_MR); if (err) return ERR_PTR(err); @@ -780,7 +780,7 @@ static struct ib_pd *nes_alloc_pd(struct ib_device *ibdev, netdev_refcnt_read(nesvnic->netdev)); err = nes_alloc_resource(nesadapter, nesadapter->allocated_pds, - nesadapter->max_pd, &pd_num, &nesadapter->next_pd); + nesadapter->max_pd, &pd_num, &nesadapter->next_pd, NES_RESOURCE_PD); if (err) { return ERR_PTR(err); } @@ -1157,7 +1157,7 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd, nes_debug(NES_DBG_QP, "RQ size=%u, SQ Size=%u\n", rq_size, sq_size); ret = nes_alloc_resource(nesadapter, nesadapter->allocated_qps, - nesadapter->max_qp, &qp_num, &nesadapter->next_qp); + nesadapter->max_qp, &qp_num, &nesadapter->next_qp, NES_RESOURCE_QP); if (ret) { return ERR_PTR(ret); } @@ -1546,7 +1546,7 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries, return ERR_PTR(-EINVAL); err = nes_alloc_resource(nesadapter, nesadapter->allocated_cqs, - nesadapter->max_cq, &cq_num, &nesadapter->next_cq); + nesadapter->max_cq, &cq_num, &nesadapter->next_cq, NES_RESOURCE_CQ); if (err) { return ERR_PTR(err); } @@ -2129,7 +2129,7 @@ static struct ib_mr *nes_reg_phys_mr(struct ib_pd *ib_pd, return ERR_PTR(-EINVAL); err = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs, nesadapter->max_mr, - &stag_index, &next_stag_index); + &stag_index, &next_stag_index, NES_RESOURCE_PHYS_MR); if (err) { return ERR_PTR(err); } @@ -2360,7 +2360,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, next_stag_index %= nesadapter->max_mr; err = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs, - nesadapter->max_mr, &stag_index, &next_stag_index); + nesadapter->max_mr, &stag_index, &next_stag_index, NES_RESOURCE_USER_MR); if (err) { ib_umem_release(region); return ERR_PTR(err); From bca1935ccdecf2354a55d9edab4c022a995ee490 Mon Sep 17 00:00:00 2001 From: Tatyana Nikolova Date: Thu, 20 Sep 2012 22:37:09 +0000 Subject: [PATCH 38/45] RDMA/nes: Fix compilation error when nes_debug is enabled Removing old variables caused a compile error from nes_debug(). Signed-off-by: Tatyana Nikolova Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_hw.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/hw/nes/nes_hw.c b/drivers/infiniband/hw/nes/nes_hw.c index d42c9f435b1b..78e936dde773 100644 --- a/drivers/infiniband/hw/nes/nes_hw.c +++ b/drivers/infiniband/hw/nes/nes_hw.c @@ -3577,10 +3577,10 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev, tcp_state = (aeq_info & NES_AEQE_TCP_STATE_MASK) >> NES_AEQE_TCP_STATE_SHIFT; iwarp_state = (aeq_info & NES_AEQE_IWARP_STATE_MASK) >> NES_AEQE_IWARP_STATE_SHIFT; nes_debug(NES_DBG_AEQ, "aeid = 0x%04X, qp-cq id = %d, aeqe = %p," - " Tcp state = %s, iWARP state = %s\n", + " Tcp state = %d, iWARP state = %d\n", async_event_id, le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]), aeqe, - nes_tcp_state_str[tcp_state], nes_iwarp_state_str[iwarp_state]); + tcp_state, iwarp_state); aeqe_cq_id = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]); if (aeq_info & NES_AEQE_QP) { From e20d583818a5d6fc052e59fe2345d82ffd089462 Mon Sep 17 00:00:00 2001 From: Dean Luick Date: Thu, 13 Sep 2012 17:19:02 +0000 Subject: [PATCH 39/45] IB/qib: Add a qib driver version Reviewed-by: Mike Marciniszyn Signed-off-by: Dean Luick Signed-off-by: Roland Dreier --- drivers/infiniband/hw/qib/qib_common.h | 14 +++++++++++++- drivers/infiniband/hw/qib/qib_driver.c | 3 ++- drivers/infiniband/hw/qib/qib_verbs.c | 2 +- 3 files changed, 16 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/qib/qib_common.h b/drivers/infiniband/hw/qib/qib_common.h index 145da4040883..d39e0183ff82 100644 --- a/drivers/infiniband/hw/qib/qib_common.h +++ b/drivers/infiniband/hw/qib/qib_common.h @@ -285,7 +285,6 @@ struct qib_base_info { #ifndef QIB_KERN_TYPE #define QIB_KERN_TYPE 0 -#define QIB_IDSTR "QLogic kernel.org driver" #endif /* @@ -301,6 +300,19 @@ struct qib_base_info { */ #define QIB_KERN_SWVERSION ((QIB_KERN_TYPE << 31) | QIB_USER_SWVERSION) +/* + * Define the driver version number. This is something that refers only + * to the driver itself, not the software interfaces it supports. + */ +#define QIB_DRIVER_VERSION_BASE "1.11" + +/* create the final driver version string */ +#ifdef QIB_IDSTR +#define QIB_DRIVER_VERSION QIB_DRIVER_VERSION_BASE " " QIB_IDSTR +#else +#define QIB_DRIVER_VERSION QIB_DRIVER_VERSION_BASE +#endif + /* * If the unit is specified via open, HCA choice is fixed. If port is * specified, it's also fixed. Otherwise we try to spread contexts diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c index e41e7f7fc763..5423edcab51f 100644 --- a/drivers/infiniband/hw/qib/qib_driver.c +++ b/drivers/infiniband/hw/qib/qib_driver.c @@ -46,7 +46,7 @@ * The size has to be longer than this string, so we can append * board/chip information to it in the init code. */ -const char ib_qib_version[] = QIB_IDSTR "\n"; +const char ib_qib_version[] = QIB_DRIVER_VERSION "\n"; DEFINE_SPINLOCK(qib_devs_lock); LIST_HEAD(qib_dev_list); @@ -65,6 +65,7 @@ MODULE_PARM_DESC(compat_ddr_negotiate, MODULE_LICENSE("Dual BSD/GPL"); MODULE_AUTHOR("QLogic "); MODULE_DESCRIPTION("QLogic IB driver"); +MODULE_VERSION(QIB_DRIVER_VERSION); /* * QIB_PIO_MAXIBHDR is the max IB header size allowed for in our diff --git a/drivers/infiniband/hw/qib/qib_verbs.c b/drivers/infiniband/hw/qib/qib_verbs.c index fc9b205c2412..ba51a4715a1d 100644 --- a/drivers/infiniband/hw/qib/qib_verbs.c +++ b/drivers/infiniband/hw/qib/qib_verbs.c @@ -2224,7 +2224,7 @@ int qib_register_ib_device(struct qib_devdata *dd) ibdev->dma_ops = &qib_dma_mapping_ops; snprintf(ibdev->node_desc, sizeof(ibdev->node_desc), - QIB_IDSTR " %s", init_utsname()->nodename); + "QLogic Infiniband HCA %s", init_utsname()->nodename); ret = ib_register_device(ibdev, qib_create_port_files); if (ret) From 9b796d06d5d1b1e85ae2316a283ea11dd739ef96 Mon Sep 17 00:00:00 2001 From: Bart Van Assche Date: Fri, 24 Aug 2012 10:27:54 +0000 Subject: [PATCH 40/45] IB/srp: Fix use-after-free in srp_reset_req() srp_free_req() uses the scsi_cmnd structure contents to unmap buffers, so we must invoke srp_free_req() before we release ownership of that structure. Signed-off-by: Bart Van Assche Acked-by: David Dillow Cc: Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index 1b5b0c730054..ac66e6b43ee2 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -638,9 +638,9 @@ static void srp_reset_req(struct srp_target_port *target, struct srp_request *re struct scsi_cmnd *scmnd = srp_claim_req(target, req, NULL); if (scmnd) { + srp_free_req(target, req, scmnd, 0); scmnd->result = DID_RESET << 16; scmnd->scsi_done(scmnd); - srp_free_req(target, req, scmnd, 0); } } From d8536670916a685df116b5c2cb256573fd25e4e3 Mon Sep 17 00:00:00 2001 From: Bart Van Assche Date: Fri, 24 Aug 2012 10:29:11 +0000 Subject: [PATCH 41/45] IB/srp: Avoid having aborted requests hang We need to call scsi_done() for commands after we abort them. Signed-off-by: Bart Van Assche Acked-by: David Dillow Cc: Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/srp/ib_srp.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c index ac66e6b43ee2..922d845f76b0 100644 --- a/drivers/infiniband/ulp/srp/ib_srp.c +++ b/drivers/infiniband/ulp/srp/ib_srp.c @@ -1687,6 +1687,7 @@ static int srp_abort(struct scsi_cmnd *scmnd) SRP_TSK_ABORT_TASK); srp_free_req(target, req, scmnd, 0); scmnd->result = DID_ABORT << 16; + scmnd->scsi_done(scmnd); return SUCCESS; } From 839f12434c7618d326b9d1ece5eca643e5e48d0a Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Thu, 27 Sep 2012 09:23:41 -0700 Subject: [PATCH 42/45] mlx4_core: Stash PCI ID driver_data in mlx4_priv structure That way we can check flags later on, when we've finished with the pci_device_id structure. Also convert the "is VF" flag to an enum: "Never do in the preprocessor what can be done in C." Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/main.c | 29 +++++++++++++---------- drivers/net/ethernet/mellanox/mlx4/mlx4.h | 6 +++++ 2 files changed, 22 insertions(+), 13 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 1e9f816a94db..c9732ca4191e 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -95,8 +95,6 @@ MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num" " Not in use with device managed" " flow steering"); -#define MLX4_VF (1 << 0) - #define HCA_GLOBAL_CAP_MASK 0 #define PF_CONTEXT_BEHAVIOUR_MASK 0 @@ -1916,7 +1914,7 @@ static void mlx4_free_ownership(struct mlx4_dev *dev) iounmap(owner); } -static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id) +static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data) { struct mlx4_priv *priv; struct mlx4_dev *dev; @@ -1939,12 +1937,11 @@ static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id) /* * Check for BARs. */ - if (((id == NULL) || !(id->driver_data & MLX4_VF)) && + if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) && !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) { dev_err(&pdev->dev, "Missing DCS, aborting." - "(id == 0X%p, id->driver_data: 0x%lx," - " pci_resource_flags(pdev, 0):0x%lx)\n", id, - id ? id->driver_data : 0, pci_resource_flags(pdev, 0)); + "(driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n", + pci_dev_data, pci_resource_flags(pdev, 0)); err = -ENODEV; goto err_disable_pdev; } @@ -2009,7 +2006,7 @@ static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id) dev->rev_id = pdev->revision; /* Detect if this device is a virtual function */ - if (id && id->driver_data & MLX4_VF) { + if (pci_dev_data & MLX4_PCI_DEV_IS_VF) { /* When acting as pf, we normally skip vfs unless explicitly * requested to probe them. */ if (num_vfs && extended_func_num(pdev) > probe_vf) { @@ -2156,6 +2153,7 @@ slave_start: mlx4_sense_init(dev); mlx4_start_sense(dev); + priv->pci_dev_data = pci_dev_data; pci_set_drvdata(pdev, dev); return 0; @@ -2225,7 +2223,7 @@ static int __devinit mlx4_init_one(struct pci_dev *pdev, { printk_once(KERN_INFO "%s", mlx4_version); - return __mlx4_init_one(pdev, id); + return __mlx4_init_one(pdev, id->driver_data); } static void mlx4_remove_one(struct pci_dev *pdev) @@ -2305,8 +2303,13 @@ static void mlx4_remove_one(struct pci_dev *pdev) int mlx4_restart_one(struct pci_dev *pdev) { + struct mlx4_dev *dev = pci_get_drvdata(pdev); + struct mlx4_priv *priv = mlx4_priv(dev); + int pci_dev_data; + + pci_dev_data = priv->pci_dev_data; mlx4_remove_one(pdev); - return __mlx4_init_one(pdev, NULL); + return __mlx4_init_one(pdev, pci_dev_data); } static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = { @@ -2335,11 +2338,11 @@ static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = { /* MT26478 ConnectX2 40GigE PCIe gen2 */ { PCI_VDEVICE(MELLANOX, 0x676e), 0 }, /* MT25400 Family [ConnectX-2 Virtual Function] */ - { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_VF }, + { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF }, /* MT27500 Family [ConnectX-3] */ { PCI_VDEVICE(MELLANOX, 0x1003), 0 }, /* MT27500 Family [ConnectX-3 Virtual Function] */ - { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_VF }, + { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF }, { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */ { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */ { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */ @@ -2368,7 +2371,7 @@ static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev, static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev) { - int ret = __mlx4_init_one(pdev, NULL); + int ret = __mlx4_init_one(pdev, 0); return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; } diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h index 71eed05426ec..ac58189ae6da 100644 --- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h +++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h @@ -768,6 +768,10 @@ struct _rule_hw { }; }; +enum { + MLX4_PCI_DEV_IS_VF = 1 << 0, +}; + struct mlx4_priv { struct mlx4_dev dev; @@ -775,6 +779,8 @@ struct mlx4_priv { struct list_head ctx_list; spinlock_t ctx_lock; + int pci_dev_data; + struct list_head pgdir_list; struct mutex pgdir_mutex; From ca3e57a599e1f3624a6164a5c3a655859368f7aa Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Thu, 27 Sep 2012 09:53:05 -0700 Subject: [PATCH 43/45] mlx4_core: Clean up enabling of SENSE_PORT for older (ConnectX-1/-2) HCAs Instead of having a hard-coded "PCI device ID != 0x1003" (which obviously breaks as newer devices with ID != 0x1003 become available), instead let's set a flag in our PCI device table for the older devices where we're supposed to force using SENSE_PORT. This also avoids enabling SENSE_PORT for virtual functions by mistake. Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/main.c | 28 +++++++++++------------ drivers/net/ethernet/mellanox/mlx4/mlx4.h | 1 + 2 files changed, 15 insertions(+), 14 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index c9732ca4191e..877d1122f787 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -297,8 +297,8 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap) mlx4_dbg(dev, "Steering mode is: %s\n", mlx4_steering_mode_str(dev->caps.steering_mode)); - /* Sense port always allowed on supported devices for ConnectX1 and 2 */ - if (dev->pdev->device != 0x1003) + /* Sense port always allowed on supported devices for ConnectX-1 and -2 */ + if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT) dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT; dev->caps.log_num_macs = log_num_mac; @@ -2314,29 +2314,29 @@ int mlx4_restart_one(struct pci_dev *pdev) static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = { /* MT25408 "Hermon" SDR */ - { PCI_VDEVICE(MELLANOX, 0x6340), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25408 "Hermon" DDR */ - { PCI_VDEVICE(MELLANOX, 0x634a), 0 }, + { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25408 "Hermon" QDR */ - { PCI_VDEVICE(MELLANOX, 0x6354), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25408 "Hermon" DDR PCIe gen2 */ - { PCI_VDEVICE(MELLANOX, 0x6732), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25408 "Hermon" QDR PCIe gen2 */ - { PCI_VDEVICE(MELLANOX, 0x673c), 0 }, + { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25408 "Hermon" EN 10GigE */ - { PCI_VDEVICE(MELLANOX, 0x6368), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25408 "Hermon" EN 10GigE PCIe gen2 */ - { PCI_VDEVICE(MELLANOX, 0x6750), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25458 ConnectX EN 10GBASE-T 10GigE */ - { PCI_VDEVICE(MELLANOX, 0x6372), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */ - { PCI_VDEVICE(MELLANOX, 0x675a), 0 }, + { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT26468 ConnectX EN 10GigE PCIe gen2*/ - { PCI_VDEVICE(MELLANOX, 0x6764), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */ - { PCI_VDEVICE(MELLANOX, 0x6746), 0 }, + { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT26478 ConnectX2 40GigE PCIe gen2 */ - { PCI_VDEVICE(MELLANOX, 0x676e), 0 }, + { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT }, /* MT25400 Family [ConnectX-2 Virtual Function] */ { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF }, /* MT27500 Family [ConnectX-3] */ diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4.h b/drivers/net/ethernet/mellanox/mlx4/mlx4.h index ac58189ae6da..1cf42036d7bb 100644 --- a/drivers/net/ethernet/mellanox/mlx4/mlx4.h +++ b/drivers/net/ethernet/mellanox/mlx4/mlx4.h @@ -770,6 +770,7 @@ struct _rule_hw { enum { MLX4_PCI_DEV_IS_VF = 1 << 0, + MLX4_PCI_DEV_FORCE_SENSE_PORT = 1 << 1, }; struct mlx4_priv { From aadf4f3f66a7b710b05b31ac0839fbbf59e41f7c Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Thu, 27 Sep 2012 10:01:19 -0700 Subject: [PATCH 44/45] mlx4_core: Disable SENSE_PORT for multifunction devices In the current driver, the SENSE_PORT firmware command is issued as a "wrapped" command, but the command handling code doesn't have a wrapper, so it will never do anything other than log an error message. The latest ConnectX-3 2.11.500 firmware reports the SENSE_PORT capability even in multi-function (SR-IOV) mode, so the driver will try to issue the command. At least until the driver has a proper wrapper for SENSE_PORT, make sure we disable the command for multi-function devices. Signed-off-by: Roland Dreier --- drivers/net/ethernet/mellanox/mlx4/main.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c index 877d1122f787..bc1e5d41c292 100644 --- a/drivers/net/ethernet/mellanox/mlx4/main.c +++ b/drivers/net/ethernet/mellanox/mlx4/main.c @@ -300,6 +300,9 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap) /* Sense port always allowed on supported devices for ConnectX-1 and -2 */ if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT) dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT; + /* Don't do sense port on multifunction devices (for now at least) */ + if (mlx4_is_mfunc(dev)) + dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT; dev->caps.log_num_macs = log_num_mac; dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS; From c00aaa1a0221a22153071fcb2f320e2ebaecc2fd Mon Sep 17 00:00:00 2001 From: Mike Marciniszyn Date: Fri, 28 Sep 2012 17:33:52 +0000 Subject: [PATCH 45/45] IB/qib: Fix local access validation for user MRs Commit 8aac4cc3a9d7 ("IB/qib: RCU locking for MR validation") introduced a bug that broke user post sends. The proper validation of the MR was lost in the patch. This patch corrects that validation. Reviewed-by: Dean Luick Signed-off-by: Mike Marciniszyn Cc: Signed-off-by: Roland Dreier --- drivers/infiniband/hw/qib/qib_keys.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/hw/qib/qib_keys.c b/drivers/infiniband/hw/qib/qib_keys.c index e9486c74c226..81c7b73695d2 100644 --- a/drivers/infiniband/hw/qib/qib_keys.c +++ b/drivers/infiniband/hw/qib/qib_keys.c @@ -186,8 +186,9 @@ int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd, goto bail; off = sge->addr - mr->user_base; - if (unlikely(sge->addr < mr->iova || off + sge->length > mr->length || - (mr->access_flags & acc) == 0)) + if (unlikely(sge->addr < mr->user_base || + off + sge->length > mr->length || + (mr->access_flags & acc) != acc)) goto bail; if (unlikely(!atomic_inc_not_zero(&mr->refcount))) goto bail;