Function "dma_map_sg" is entitled to merge adjacent entries
and return a value smaller than what was passed as "nents".
Subsequently "ib_map_mr_sg" needs to work with this value ("sg_dma_len")
rather than the original "nents" parameter ("sg_len").
This old RDS bug was exposed and reliably causes kernel panics
(using RDMA operations "rds-stress -D") on x86_64 starting with:
commit c588072bba ("iommu/vt-d: Convert intel iommu driver to the iommu ops")
Simply put: Linux 5.11 and later.
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Link: https://lore.kernel.org/r/60efc69f-1f35-529d-a7ef-da0549cad143@oracle.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since "rds_ib_free_frmr" and "rds_ib_free_frmr_list" simply put
the FRMR memory segments on the "drop_list" or "free_list",
and it is the job of "rds_ib_flush_mr_pool" to reap those entries
by ultimately issuing a "IB_WR_LOCAL_INV" work-request,
we need to trigger and then wait for all those memory segments
attached to a particular connection to be fully released before
we can move on to release the QP, CQ, etc.
So we make "rds_ib_conn_path_shutdown" wait for one more
atomic_t called "i_fastreg_inuse_count" that keeps track of how
many FRWR memory segments are out there marked "FRMR_IS_INUSE"
(and also wake_up rds_ib_ring_empty_wait, as they go away).
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix a bug where fr_state first goes to FRMR_IS_STALE, because of a failure
of operation IB_WR_LOCAL_INV, but then gets set back to "FRMR_IS_FREE"
uncoditionally, even though the operation failed.
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to:
1) avoid a silly bouncing between "clean_list" and "drop_list"
triggered by function "rds_ib_reg_frmr" as it is releases frmr
regions whose state is not "FRMR_IS_FREE" right away.
2) prevent an invalid access error in a race from a pending
"IB_WR_LOCAL_INV" operation with a teardown ("dma_unmap_sg", "put_page")
and de-registration ("ib_dereg_mr") of the corresponding
memory region.
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the context of FRMR (ib_frmr.c):
Memory regions make it onto the "clean_list" via "rds_ib_flush_mr_pool",
after the memory region has been posted for invalidation via
"rds_ib_post_inv".
At that point in time, "fr_state" may still be in state "FRMR_IS_INUSE",
since the only place where "fr_state" transitions to "FRMR_IS_FREE"
is in "rds_ib_mr_cqe_handler", which is triggered by a tasklet.
So in case we notice that "fr_state != FRMR_IS_FREE" (see below),
we wait for "fr_inv_done" to trigger with a maximum of 10msec.
Then we check again, and only put the memory region onto the drop_list
(via "rds_ib_free_frmr") in case the situation remains unchanged.
This avoids the problem of memory-regions bouncing between "clean_list"
and "drop_list" before they even have a chance to be properly invalidated.
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 5601245931.
RDS kept spinning inside function "rds_ib_post_reg_frmr", waiting for
"i_fastreg_wrs" to become incremented:
while (atomic_dec_return(&ibmr->ic->i_fastreg_wrs) <= 0) {
atomic_inc(&ibmr->ic->i_fastreg_wrs);
cpu_relax();
}
Looking at the original commit:
commit 5601245931 ("RDS: IB: split the mr registration and
invalidation path")
In there, the "rds_ib_mr_cqe_handler" was changed in the following
way:
void rds_ib_mr_cqe_handler(struct
rds_ib_connection *ic,
struct ib_wc *wc)
if (frmr->fr_inv) {
frmr->fr_state = FRMR_IS_FREE;
frmr->fr_inv = false;
atomic_inc(&ic->i_fastreg_wrs);
} else {
atomic_inc(&ic->i_fastunreg_wrs);
}
It looks like it's got it exactly backwards:
Function "rds_ib_post_reg_frmr" keeps track of the outstanding
requests via "i_fastreg_wrs".
Function "rds_ib_post_inv" keeps track of the outstanding requests
via "i_fastunreg_wrs" (post original commit). It also sets:
frmr->fr_inv = true;
However the completion handler "rds_ib_mr_cqe_handler" adjusts
"i_fastreg_wrs" when "fr_inv" had been true, and adjusts
"i_fastunreg_wrs" otherwise.
The original commit was done in the name of performance:
to remove the performance bottleneck
No performance benefit could be observed with a fixed-up version
of the original commit measured between two Oracle X7 servers,
both equipped with Mellanox Connect-X5 HCAs.
The prudent course of action is to revert this commit.
Signed-off-by: Gerd Rausch <gerd.rausch@oracle.com>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Keeping single line wrapper functions is not useful. Hence remove the
ib_sg_dma_address() and ib_sg_dma_len() functions. This patch does not
change any functionality.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
rdma.git merge resolution for the 4.19 merge window
Conflicts:
drivers/infiniband/core/rdma_core.c
- Use the rdma code and revise with the new spelling for
atomic_fetch_add_unless
drivers/nvme/host/rdma.c
- Replace max_sge with max_send_sge in new blk code
drivers/nvme/target/rdma.c
- Use the blk code and revise to use NULL for ib_post_recv when
appropriate
- Replace max_sge with max_recv_sge in new blk code
net/rds/ib_send.c
- Use the net code and revise to use NULL for ib_post_recv when
appropriate
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
-----BEGIN PGP SIGNATURE-----
iQFSBAABCAA8FiEEq68RxlopcLEwq+PEeb4+QwBBGIYFAltwm2geHHRvcnZhbGRz
QGxpbnV4LWZvdW5kYXRpb24ub3JnAAoJEHm+PkMAQRiGITkH/iSzkVhT2OxHoir0
mLVzTi7/Z17L0e/ELl7TvAC0iLFlWZKdlGR0g3b4/QpXLPmNK4HxiDRTQuWn8ke0
qDZyDq89HqLt+mpeFZ43PCd9oqV8CH2xxK3iCWReqv6bNnowGnRpSStlks4rDqWn
zURC/5sUh7TzEG4s997RrrpnyPeQWUlf/Mhtzg2/WvK2btoLWgu5qzjX1uFh3s7u
vaF2NXVJ3X03gPktyxZzwtO1SwLFS1jhwUXWBZ5AnoJ99ywkghQnkqS/2YpekNTm
wFk80/78sU+d91aAqO8kkhHj8VRrd+9SGnZ4mB2aZHwjZjGcics4RRtxukSfOQ+6
L47IdXo=
=sJkt
-----END PGP SIGNATURE-----
Merge tag 'v4.18' into rdma.git for-next
Resolve merge conflicts from the -rc cycle against the rdma.git tree:
Conflicts:
drivers/infiniband/core/uverbs_cmd.c
- New ifs added to ib_uverbs_ex_create_flow in -rc and for-next
- Merge removal of file->ucontext in for-next with new code in -rc
drivers/infiniband/core/uverbs_main.c
- for-next removed code from ib_uverbs_write() that was modified
in for-rc
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Fix a static code checker warning:
net/rds/ib_frmr.c:82 rds_ib_alloc_frmr() warn: passing zero to 'ERR_PTR'
The error path for ib_alloc_mr failure should set err to PTR_ERR.
Fixes: 1659185fb4 ("RDS: IB: Support Fastreg MR (FRMR) memory registration mode")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of declaring and passing a dummy 'bad_wr' pointer, pass NULL
as third argument to ib_post_(send|recv|srq_recv)().
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Remove two WARN_ON() statements that verify something that is guaranteed
by the RDMA API, namely that the failed_wr pointer is not touched if an
ib_post_send() call succeeds and that it points at the failed wr if an
ib_post_send() call fails.
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
MR invalidation in RDS is done in background thread and not in
data path like registration. So break the dependency between them
which helps to remove the performance bottleneck.
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Fastreg MR(FRMR) is another method with which one can
register memory to HCA. Some of the newer HCAs supports only fastreg
mr mode, so we need to add support for it to have RDS functional
on them.
Signed-off-by: Santosh Shilimkar <ssantosh@kernel.org>
Signed-off-by: Avinash Repaka <avinash.repaka@oracle.com>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Gerd Rausch
Max Gurtovoy
Bart Van Assche
Jason Gunthorpe
YueHaibing
Avinash Repaka
Bart Van Assche
Santosh Shilimkar
Christoph Hellwig