Currently, buffers are cleared when smc connections are created and
buffers are reused. This slows down the speed of establishing new
connections. In most cases, the applications want to establish
connections as quickly as possible.
This patch moves memset() from connection creation path to release and
buffer unuse path, this trades off between speed of establishing and
release.
Test environments:
- CPU Intel Xeon Platinum 8 core, mem 32 GiB, nic Mellanox CX4
- socket sndbuf / rcvbuf: 16384 / 131072 bytes
- w/o first round, 5 rounds, avg, 100 conns batch per round
- smc_buf_create() use bpftrace kprobe, introduces extra latency
Latency benchmarks for smc_buf_create():
w/o patch : 19040.0 ns
w/ patch : 1932.6 ns
ratio : 10.2% (-89.8%)
Latency benchmarks for socket create and connect:
w/o patch : 143.3 us
w/ patch : 102.2 us
ratio : 71.3% (-28.7%)
The latency of establishing connections is reduced by 28.7%.
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Reviewed-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/20211203113331.2818873-1-kgraul@linux.ibm.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit 5ac4f180bd.
Sorry for taking no notice that the function devlink_register() has been
already declared as void, so it is needs to revert this patch.
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Link: https://lore.kernel.org/r/20211204012448.51360-1-huangguangbin2@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The iosm driver started using relayfs, but is missing the Kconfig
logic to ensure it's built into the kernel:
x86_64-linux-ld: drivers/net/wwan/iosm/iosm_ipc_trace.o: in function `ipc_trace_create_buf_file_handler':
iosm_ipc_trace.c:(.text+0x16): undefined reference to `relay_file_operations'
x86_64-linux-ld: drivers/net/wwan/iosm/iosm_ipc_trace.o: in function `ipc_trace_subbuf_start_handler':
iosm_ipc_trace.c:(.text+0x31): undefined reference to `relay_buf_full'
x86_64-linux-ld: drivers/net/wwan/iosm/iosm_ipc_trace.o: in function `ipc_trace_ctrl_file_write':
iosm_ipc_trace.c:(.text+0xd5): undefined reference to `relay_flush'
x86_64-linux-ld: drivers/net/wwan/iosm/iosm_ipc_trace.o: in function `ipc_trace_port_rx':
Fixes: 00ef32565b ("net: wwan: iosm: device trace collection using relayfs")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Sergey Ryazanov <ryazanov.s.a@gmail.com>
Reviewed-by: M Chetan Kumar <m.chetan.kumar@linux.intel.com>
Link: https://lore.kernel.org/r/20211204174033.950528-1-arnd@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Vladimir reported csum issues after my recent change in skb_postpull_rcsum()
Issue here is the following:
initial skb->csum is the csum of
[part to be pulled][rest of packet]
Old code:
skb->csum = csum_sub(skb->csum, csum_partial(pull, pull_length, 0));
New code:
skb->csum = ~csum_partial(pull, pull_length, ~skb->csum);
This is broken if the csum of [pulled part]
happens to be equal to skb->csum, because end
result of skb->csum is 0 in new code, instead
of being 0xffffffff
David Laight suggested to use
skb->csum = -csum_partial(pull, pull_length, -skb->csum);
I based my patches on existing code present in include/net/seg6.h,
update_csum_diff4() and update_csum_diff16() which might need
a similar fix.
I guess that my tests, mostly pulling 40 bytes of IPv6 header
were not providing enough entropy to hit this bug.
v2: added wsum_negate() to make sparse happy.
Fixes: 29c3002644 ("net: optimize skb_postpull_rcsum()")
Fixes: 0bd28476f6 ("gro: optimize skb_gro_postpull_rcsum()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Suggested-by: David Laight <David.Laight@ACULAB.COM>
Cc: David Lebrun <dlebrun@google.com>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20211204045356.3659278-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Eric Dumazet says:
====================
net: add preliminary netdev refcount tracking
Two first patches add a generic infrastructure, that will be used
to get tracking of refcount increments/decrements.
The general idea is to be able to precisely pair each decrement with
a corresponding prior increment. Both share a cookie, basically
a pointer to private data storing stack traces.
The third patch adds dev_hold_track() and dev_put_track() helpers
(CONFIG_NET_DEV_REFCNT_TRACKER)
Then a series of 20 patches converts some dev_hold()/dev_put()
pairs to new hepers : dev_hold_track() and dev_put_track().
Hopefully this will be used by developpers and syzbot to
root cause bugs that cause netdevice dismantles freezes.
With CONFIG_PCPU_DEV_REFCNT=n option, we were able to detect
some class of bugs, but too late (when too many dev_put()
were happening).
Another series will be sent after this one is merged.
v3: moved NET_DEV_REFCNT_TRACKER to net/Kconfig.debug
added "depends on DEBUG_KERNEL && STACKTRACE_SUPPORT"
to hopefully get rid of kbuild reports for ARCH=nios2
Reworded patch 3 changelog.
Added missing htmldocs (Jakub)
v2: added four additional patches,
added netdev_tracker_alloc() and netdev_tracker_free()
addressed build error (kernel bots),
use GFP_ATOMIC in test_ref_tracker_timer_func()
====================
Link: https://lore.kernel.org/r/20211205042217.982127-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a netdevice_tracker inside struct net_device, to track
the self reference when a device is in lweventlist.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Note that other ip_tunnel users do not seem to hold a reference
on tunnel->dev. Probably needs some investigations.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We want to track all dev_hold()/dev_put() to ease leak hunting.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We want to track all dev_hold()/dev_put() to ease leak hunting.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This helper might hold a netdev reference for a long time,
lets add reference tracking.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This will help debugging pesky netdev reference leaks.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net device are refcounted. Over the years we had numerous bugs
caused by imbalanced dev_hold() and dev_put() calls.
The general idea is to be able to precisely pair each decrement with
a corresponding prior increment. Both share a cookie, basically
a pointer to private data storing stack traces.
This patch adds dev_hold_track() and dev_put_track().
To use these helpers, each data structure owning a refcount
should also use a "netdevice_tracker" to pair the hold and put.
netdevice_tracker dev_tracker;
...
dev_hold_track(dev, &dev_tracker, GFP_ATOMIC);
...
dev_put_track(dev, &dev_tracker);
Whenever a leak happens, we will get precise stack traces
of the point dev_hold_track() happened, at device dismantle phase.
We will also get a stack trace if too many dev_put_track() for the same
netdevice_tracker are attempted.
This is guarded by CONFIG_NET_DEV_REFCNT_TRACKER option.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It can be hard to track where references are taken and released.
In networking, we have annoying issues at device or netns dismantles,
and we had various proposals to ease root causing them.
This patch adds new infrastructure pairing refcount increases
and decreases. This will self document code, because programmers
will have to associate increments/decrements.
This is controled by CONFIG_REF_TRACKER which can be selected
by users of this feature.
This adds both cpu and memory costs, and thus should probably be
used with care.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
ESL(Enhanced System Lockdown) was designed to lock PCI adapter firmware
images and prevent changes to critical non-volatile configuration data
so that uncontrolled, malicious or unintentional modification to the
adapters are avoided, ensuring it's operational state. Once this feature is
enabled, the device is locked, rejecting any modification to non-volatile
images. Once unlocked, the protection is off such that firmware and
non-volatile configurations may be altered.
Driver just reflects the capability and status of this through
the ethtool private flag.
Signed-off-by: Manish Chopra <manishc@marvell.com>
Signed-off-by: Prabhakar Kushwaha <pkushwaha@marvell.com>
Signed-off-by: Alok Prasad <palok@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch add some new qed APIs to query status block
info and report various data to MFW on tx timeout event
Along with that it enhances qede to dump more debug logs
(not just specific to the queue which was reported by stack)
on tx timeout which includes various other basic metadata about
all tx queues and other info (like status block etc.)
Signed-off-by: Manish Chopra <manishc@marvell.com>
Signed-off-by: Prabhakar Kushwaha <pkushwaha@marvell.com>
Signed-off-by: Alok Prasad <palok@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The devm_ioremap() function does not return error pointers. It returns
NULL.
Fixes: db8bcaad53 ("net: lan966x: add the basic lan966x driver")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Horatiu Vultur <horatiu.vultur@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rhashtable_lookup_fast() returns NULL pointer not ERR_PTR().
Return rhashtable_lookup_fast() directly to fix this.
Fixes: 47327e198d ("net: prestera: acl: migrate to new vTCAM api")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
GPIO library does copy the of_node from the parent device of
the GPIO chip, there is no need to repeat this in the individual
drivers. Remove assignment here.
For the details one may look into the of_gpio_dev_init() implementation.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Short-lived connections increase the insertion rate requirements,
fill the offload table and provide very limited offload value since
they process a very small amount of packets. The ct ASSURED flag is
designed to filter short-lived connections for early expiration.
Offload connections when they are ESTABLISHED and ASSURED.
Signed-off-by: Chris Mi <cmi@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Guangbin Huang says:
====================
net: hns3: some cleanups for -next
To improve code readability and simplicity, this series add some cleanup
patches for the HNS3 ethernet driver.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The hns3 driver header file uses the structure of other files, but does
not include corresponding file, which causes a check warning that the
header file is not self-contained.
Therefore, the required header file is included in the header file, and
the structure declaration is added to the header file to avoid cyclic
dependency of the header file.
Signed-off-by: Jie Wang <wangjie125@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace one tab with space between symbol ')' and '{' in for statement of
function hclge_map_tqp().
Signed-off-by: Hao Chen <chenhao288@hisilicon.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Return value judgment should follow the function call, so remove line
between them.
Signed-off-by: Hao Chen <chenhao288@hisilicon.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we use hclge_dbg_fill_content() to fill contents with
specific format according to struct hclge_dbg_item *items,
it may cause content cover due to unreasonable items.
So add comments to explain how to avoid it.
Signed-off-by: Hao Chen <chenhao288@hisilicon.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add void before function which don't receive ret to improve code
readability.
Signed-off-by: Hao Chen <chenhao288@hisilicon.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change output value type of atomic_read() from %u to %d.
Signed-off-by: Hao Chen <chenhao288@hisilicon.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The argument len will not be changed in hclge_ncl_config_data_print(), it
is no need to declare as a pointer, so modify it into int type.
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The c language has a set of implicit type conversions, when
two variables perform bitwise or arithmetic operations.
For example, variable A (type u16/u8) -1, its output is int type variable.
u16/u8 will convert to int type implicitly before it does arithmetic
operations. So, change 1 to unsigned type.
Signed-off-by: Hao Chen <chenhao288@hisilicon.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds print vport id when failed to get or set vlan
filter parameters.
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Function hclge_set_vlan_filter_hw() is a bit too long, so add a new
function hclge_need_update_port_vlan() to simplify code and improve
code readability.
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
hclge_cfg_common_loopback() is a bit too long, so
encapsulate hclge_cfg_common_loopback_cmd_send() and
hclge_cfg_common_loopback_wait() two functions to
improve readability.
Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The flow counters bulk query buffer is allocated once during
mlx5_fc_init_stats(). For PFs and VFs this buffer usually takes a little
more than 512KB of memory, which is aligned to the next power of 2, to
1MB. For SFs, this buffer is reduced and takes around 128 Bytes.
The buffer size determines the maximum number of flow counters that
can be queried at a time. Thus, having a bigger buffer can improve
performance for users that need to query many flow counters.
There are cases that don't use many flow counters and don't need a big
buffer (e.g. SFs, VFs). Since this size is critical with large scale,
in these cases the buffer size should be reduced.
In order to reduce memory consumption while maintaining query
performance, change the query buffer's allocation scheme to the
following:
- First allocate the buffer with small initial size.
- If the number of counters surpasses the initial size, resize the
buffer to the maximum size.
The buffer only grows and isn't shrank, because users with many flow
counters don't care about the buffer size and we don't want to add
resize overhead if the current number of counters drops.
This solution is preferable to the current one, which is less accurate
and only addresses SFs.
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Tony Lu
Guangbin Huang
José Expósito
Arnd Bergmann
Eric Dumazet
Jakub Kicinski
Manish Chopra
Dan Carpenter
Yang Yingliang
Andy Shevchenko
Chris Mi
David S. Miller
Jie Wang
Hao Chen
Yufeng Mo
Avihai Horon