While reading sysctl_tcp_keepalive_(time|probes|intvl), they can be changed
concurrently. Thus, we need to add READ_ONCE() to their readers.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extend SMC-R link group netlink attribute SMC_GEN_LGR_SMCR.
Introduce SMC_NLA_LGR_R_BUF_TYPE to show the buffer type of
SMC-R link group.
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On long-running enterprise production servers, high-order contiguous
memory pages are usually very rare and in most cases we can only get
fragmented pages.
When replacing TCP with SMC-R in such production scenarios, attempting
to allocate high-order physically contiguous sndbufs and RMBs may result
in frequent memory compaction, which will cause unexpected hung issue
and further stability risks.
So this patch is aimed to allow SMC-R link group to use virtually
contiguous sndbufs and RMBs to avoid potential issues mentioned above.
Whether to use physically or virtually contiguous buffers can be set
by sysctl smcr_buf_type.
Note that using virtually contiguous buffers will bring an acceptable
performance regression, which can be mainly divided into two parts:
1) regression in data path, which is brought by additional address
translation of sndbuf by RNIC in Tx. But in general, translating
address through MTT is fast.
Taking 256KB sndbuf and RMB as an example, the comparisons in qperf
latency and bandwidth test with physically and virtually contiguous
buffers are as follows:
- client:
smc_run taskset -c <cpu> qperf <server> -oo msg_size:1:64K:*2\
-t 5 -vu tcp_{bw|lat}
- server:
smc_run taskset -c <cpu> qperf
[latency]
msgsize tcp smcr smcr-use-virt-buf
1 11.17 us 7.56 us 7.51 us (-0.67%)
2 10.65 us 7.74 us 7.56 us (-2.31%)
4 11.11 us 7.52 us 7.59 us ( 0.84%)
8 10.83 us 7.55 us 7.51 us (-0.48%)
16 11.21 us 7.46 us 7.51 us ( 0.71%)
32 10.65 us 7.53 us 7.58 us ( 0.61%)
64 10.95 us 7.74 us 7.80 us ( 0.76%)
128 11.14 us 7.83 us 7.87 us ( 0.47%)
256 10.97 us 7.94 us 7.92 us (-0.28%)
512 11.23 us 7.94 us 8.20 us ( 3.25%)
1024 11.60 us 8.12 us 8.20 us ( 0.96%)
2048 14.04 us 8.30 us 8.51 us ( 2.49%)
4096 16.88 us 9.13 us 9.07 us (-0.64%)
8192 22.50 us 10.56 us 11.22 us ( 6.26%)
16384 28.99 us 12.88 us 13.83 us ( 7.37%)
32768 40.13 us 16.76 us 16.95 us ( 1.16%)
65536 68.70 us 24.68 us 24.85 us ( 0.68%)
[bandwidth]
msgsize tcp smcr smcr-use-virt-buf
1 1.65 MB/s 1.59 MB/s 1.53 MB/s (-3.88%)
2 3.32 MB/s 3.17 MB/s 3.08 MB/s (-2.67%)
4 6.66 MB/s 6.33 MB/s 6.09 MB/s (-3.85%)
8 13.67 MB/s 13.45 MB/s 11.97 MB/s (-10.99%)
16 25.36 MB/s 27.15 MB/s 24.16 MB/s (-11.01%)
32 48.22 MB/s 54.24 MB/s 49.41 MB/s (-8.89%)
64 106.79 MB/s 107.32 MB/s 99.05 MB/s (-7.71%)
128 210.21 MB/s 202.46 MB/s 201.02 MB/s (-0.71%)
256 400.81 MB/s 416.81 MB/s 393.52 MB/s (-5.59%)
512 746.49 MB/s 834.12 MB/s 809.99 MB/s (-2.89%)
1024 1292.33 MB/s 1641.96 MB/s 1571.82 MB/s (-4.27%)
2048 2007.64 MB/s 2760.44 MB/s 2717.68 MB/s (-1.55%)
4096 2665.17 MB/s 4157.44 MB/s 4070.76 MB/s (-2.09%)
8192 3159.72 MB/s 4361.57 MB/s 4270.65 MB/s (-2.08%)
16384 4186.70 MB/s 4574.13 MB/s 4501.17 MB/s (-1.60%)
32768 4093.21 MB/s 4487.42 MB/s 4322.43 MB/s (-3.68%)
65536 4057.14 MB/s 4735.61 MB/s 4555.17 MB/s (-3.81%)
2) regression in buffer initialization and destruction path, which is
brought by additional MR operations of sndbufs. But thanks to link
group buffer reuse mechanism, the impact of this kind of regression
decreases as times of buffer reuse increases.
Taking 256KB sndbuf and RMB as an example, latency of some key SMC-R
buffer-related function obtained by bpftrace are as follows:
Function Phys-bufs Virt-bufs
smcr_new_buf_create() 67154 ns 79164 ns
smc_ib_buf_map_sg() 525 ns 928 ns
smc_ib_get_memory_region() 162294 ns 161191 ns
smc_wr_reg_send() 9957 ns 9635 ns
smc_ib_put_memory_region() 203548 ns 198374 ns
smc_ib_buf_unmap_sg() 508 ns 1158 ns
------------
Test environment notes:
1. Above tests run on 2 VMs within the same Host.
2. The NIC is ConnectX-4Lx, using SRIOV and passing through 2 VFs to
the each VM respectively.
3. VMs' vCPUs are binded to different physical CPUs, and the binded
physical CPUs are isolated by `isolcpus=xxx` cmdline.
4. NICs' queue number are set to 1.
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch introduces a new SMC-R specific element buf_type
in struct smc_link_group, for recording the value of sysctl
smcr_buf_type when link group is created.
New created link group will create and reuse buffers of the
type specified by buf_type.
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch introduces the sysctl smcr_buf_type for setting
the type of SMC-R sndbufs and RMBs.
Valid values includes:
- SMCR_PHYS_CONT_BUFS, which means use physically contiguous
buffers for better performance and is the default value.
- SMCR_VIRT_CONT_BUFS, which means use virtually contiguous
buffers in case of physically contiguous memory is scarce.
- SMCR_MIXED_BUFS, which means first try to use physically
contiguous buffers. If not available, then use virtually
contiguous buffers.
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some CPU, such as Xeon, can guarantee DMA cache coherency.
So it is no need to use dma sync APIs to flush cache on such CPUs.
In order to avoid calling dma sync APIs on the IO path, use the
dma_need_sync to check whether smc_buf_desc needs dma sync when
creating smc_buf_desc.
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
smc_ib_sync_sg_for_cpu/device are the ops used for dma memory cache
consistency. Smc sndbufs are dma buffers, where CPU writes data to
it and PCIE device reads data from it. So for sndbufs,
smc_ib_sync_sg_for_device is needed and smc_ib_sync_sg_for_cpu is
redundant as PCIE device will not write the buffers. Smc rmbs
are dma buffers, where PCIE device write data to it and CPU read
data from it. So for rmbs, smc_ib_sync_sg_for_cpu is needed and
smc_ib_sync_sg_for_device is redundant as CPU will not write the buffers.
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Netdev reference helpers have a dev_ prefix for historic
reasons. Renaming the old helpers would be too much churn
but we can rename the tracking ones which are relatively
recent and should be the default for new code.
Rename:
dev_hold_track() -> netdev_hold()
dev_put_track() -> netdev_put()
dev_replace_track() -> netdev_ref_replace()
Link: https://lore.kernel.org/r/20220608043955.919359-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
"struct smc_cdc_tx_pend **" can not directly convert
to "struct smc_wr_tx_pend_priv *".
Fixes: 2bced6aefa ("net/smc: put slot when connection is killed")
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the process of checking whether RDMAv2 is available, the current
implementation first sets ini->smcrv2.ib_dev_v2, and then allocates
smc buf desc and register rmb, but the latter may fail. In this case,
the pointer should be reset.
Fixes: e49300a6bf ("net/smc: add listen processing for SMC-Rv2")
Signed-off-by: liuyacan <liuyacan@corp.netease.com>
Reviewed-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/20220525085408.812273-1-liuyacan@corp.netease.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In the process of checking whether RDMAv2 is available, the current
implementation first sets ini->smcrv2.ib_dev_v2, and then allocates
smc buf desc, but the latter may fail. Unfortunately, the caller
will only check the former. In this case, a NULL pointer reference
will occur in smc_clc_send_confirm_accept() when accessing
conn->rmb_desc.
This patch does two things:
1. Use the return code to determine whether V2 is available.
2. If the return code is NODEV, continue to check whether V1 is
available.
Fixes: e49300a6bf ("net/smc: add listen processing for SMC-Rv2")
Signed-off-by: liuyacan <liuyacan@corp.netease.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Same trigger condition as commit 86434744. When setsockopt runs
in parallel to a connect(), and switch the socket into fallback
mode. Then the sk_refcnt is incremented in smc_connect(), but
its state stay in SMC_INIT (NOT SMC_ACTIVE). This cause the
corresponding sk_refcnt decrement in __smc_release() will not be
performed.
Fixes: 86434744fe ("net/smc: add fallback check to connect()")
Signed-off-by: liuyacan <liuyacan@corp.netease.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rdma write with inline flag when sending small packages,
whose length is shorter than the qp's max_inline_data, can
help reducing latency.
In my test environment, which are 2 VMs running on the same
physical host and whose NICs(ConnectX-4Lx) are working on
SR-IOV mode, qperf shows 0.5us-0.7us improvement in latency.
Test command:
server: smc_run taskset -c 1 qperf
client: smc_run taskset -c 1 qperf <server ip> -oo \
msg_size:1:2K:*2 -t 30 -vu tcp_lat
The results shown below:
msgsize before after
1B 11.2 us 10.6 us (-0.6 us)
2B 11.2 us 10.7 us (-0.5 us)
4B 11.3 us 10.7 us (-0.6 us)
8B 11.2 us 10.6 us (-0.6 us)
16B 11.3 us 10.7 us (-0.6 us)
32B 11.3 us 10.6 us (-0.7 us)
64B 11.2 us 11.2 us (0 us)
128B 11.2 us 11.2 us (0 us)
256B 11.2 us 11.2 us (0 us)
512B 11.4 us 11.3 us (-0.1 us)
1KB 11.4 us 11.5 us (0.1 us)
2KB 11.5 us 11.5 us (0 us)
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Tested-by: kernel test robot <lkp@intel.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As cdc msg's length is 44B, cdc msgs can be sent inline in
most rdma devices, which can help reducing sending latency.
In my test environment, which are 2 VMs running on the same
physical host and whose NICs(ConnectX-4Lx) are working on
SR-IOV mode, qperf shows 0.4us-0.7us improvement in latency.
Test command:
server: smc_run taskset -c 1 qperf
client: smc_run taskset -c 1 qperf <server ip> -oo \
msg_size:1:2K:*2 -t 30 -vu tcp_lat
The results shown below:
msgsize before after
1B 11.9 us 11.2 us (-0.7 us)
2B 11.7 us 11.2 us (-0.5 us)
4B 11.7 us 11.3 us (-0.4 us)
8B 11.6 us 11.2 us (-0.4 us)
16B 11.7 us 11.3 us (-0.4 us)
32B 11.7 us 11.3 us (-0.4 us)
64B 11.7 us 11.2 us (-0.5 us)
128B 11.6 us 11.2 us (-0.4 us)
256B 11.8 us 11.2 us (-0.6 us)
512B 11.8 us 11.4 us (-0.4 us)
1KB 11.9 us 11.4 us (-0.5 us)
2KB 12.1 us 11.5 us (-0.6 us)
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Tested-by: kernel test robot <lkp@intel.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Connect with O_NONBLOCK will not be completed immediately
and returns -EINPROGRESS. It is possible to use selector/poll
for completion by selecting the socket for writing. After select
indicates writability, a second connect function call will return
0 to indicate connected successfully as TCP does, but smc returns
-EISCONN. Use socket state for smc to indicate connect state, which
can help smc aligning the connect behaviour with TCP.
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Non blocking sendmsg will return -EAGAIN when any signal pending
and no send space left, while non blocking recvmsg return -EINTR
when signal pending and no data received. This may makes confused.
As TCP returns -EAGAIN in the conditions described above. Align the
behavior of smc with TCP.
Fixes: 846e344eb7 ("net/smc: add receive timeout check")
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/20220512030820.73848-1-guangguan.wang@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
syzbot reported a slab-out-of-bounds/use-after-free issue,
which was caused by accessing an already freed smc sock in
fallback-specific callback functions of clcsock.
This patch fixes the issue by restoring fallback-specific
callback functions to original ones and resetting clcsock
sk_user_data to NULL before freeing smc sock.
Meanwhile, this patch introduces sk_callback_lock to make
the access and assignment to sk_user_data mutually exclusive.
Reported-by: syzbot+b425899ed22c6943e00b@syzkaller.appspotmail.com
Fixes: 341adeec9a ("net/smc: Forward wakeup to smc socket waitqueue after fallback")
Link: https://lore.kernel.org/r/00000000000013ca8105d7ae3ada@google.com/
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Both listen and fallback process will save the current clcsock
callback functions and establish new ones. But if both of them
happen, the saved callback functions will be overwritten.
So this patch introduces some helpers to ensure that only save
the original callback functions of clcsock.
Fixes: 341adeec9a ("net/smc: Forward wakeup to smc socket waitqueue after fallback")
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In the current implementation, when TCP initiates a connection
to an unavailable [ip,port], ECONNREFUSED will be stored in the
TCP socket, but SMC will not. However, some apps (like curl) use
getsockopt(,,SO_ERROR,,) to get the error information, which makes
them miss the error message and behave strangely.
Fixes: 50717a37db ("net/smc: nonblocking connect rework")
Signed-off-by: liuyacan <liuyacan@corp.netease.com>
Reviewed-by: Tony Lu <tonylu@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit e5d5aadcf3 ("net/smc: fix sk_refcnt underflow on linkdown
and fallback"), for a fallback connection, __smc_release() does not call
sock_put() if its state is already SMC_CLOSED.
When calling smc_shutdown() after falling back, its state is set to
SMC_CLOSED but does not call sock_put(), so this patch calls it.
Reported-and-tested-by: syzbot+6e29a053eb165bd50de5@syzkaller.appspotmail.com
Fixes: e5d5aadcf3 ("net/smc: fix sk_refcnt underflow on linkdown and fallback")
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Child sockets may inherit the af_ops from the parent listen socket.
When the listen socket is released then the af_ops of the child socket
points to released memory.
Solve that by restoring the original af_ops for child sockets which
inherited the parent af_ops. And clear any inherited user_data of the
parent socket.
Fixes: 8270d9c210 ("net/smc: Limit backlog connections")
Reviewed-by: Wenjia Zhang <wenjia@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Reviewed-by: D. Wythe <alibuda@linux.alibaba.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
dev_name() was called with dev.parent as argument but without to
NULL-check it before.
Solve this by checking the pointer before the call to dev_name().
Fixes: af5f60c7e3 ("net/smc: allow PCI IDs as ib device names in the pnet table")
Reported-by: syzbot+03e3e228510223dabd34@syzkaller.appspotmail.com
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Using snprintf() to convert not null-terminated strings to null
terminated strings may cause out of bounds read in the source string.
Therefore use memcpy() and terminate the target string with a null
afterwards.
Fixes: fa08666255 ("net/smc: add support for user defined EIDs")
Fixes: 3c572145c2 ("net/smc: add generic netlink support for system EID")
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The current autocork algorithms will delay the data transmission
in BH context to smc_release_cb() when sock_lock is hold by user.
So there is a possibility that when connection is being actively
closed (sock_lock is hold by user now), some corked data still
remains in sndbuf, waiting to be sent by smc_release_cb(). This
will cause:
- smc_close_stream_wait(), which is called under the sock_lock,
has a high probability of timeout because data transmission is
delayed until sock_lock is released.
- Unexpected data sends may happen after connction closed and use
the rtoken which has been deleted by remote peer through
LLC_DELETE_RKEY messages.
So this patch will try to send out the remaining corked data in
sndbuf before active close process, to ensure data integrity and
avoid unexpected data transmission after close.
Reported-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Fixes: 6b88af839d ("net/smc: don't send in the BH context if sock_owned_by_user")
Signed-off-by: Wen Gu <guwen@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/1648447836-111521-1-git-send-email-guwen@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Recently added smc_sysctl_net_exit() forgot to free
the memory allocated from smc_sysctl_net_init()
for non initial network namespace.
Fixes: 462791bbfa ("net/smc: add sysctl interface for SMC")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Tony Lu <tonylu@linux.alibaba.com>
Cc: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
when CONFIG_SYSCTL not set, smc_sysctl_net_init/exit
need to be static inline to avoid missing-prototypes
if compile with W=1.
Since __net_exit has noinline annotation when CONFIG_NET_NS
not set, it should not be used with static inline.
So remove the __net_init/exit when CONFIG_SYSCTL not set.
Fixes: 7de8eb0d90 ("net/smc: fix compile warning for smc_sysctl")
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Link: https://lore.kernel.org/r/20220309033051.41893-1-dust.li@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
kernel test robot reports multiple warning for smc_sysctl:
In file included from net/smc/smc_sysctl.c:17:
>> net/smc/smc_sysctl.h:23:5: warning: no previous prototype \
for function 'smc_sysctl_init' [-Wmissing-prototypes]
int smc_sysctl_init(void)
^
and
>> WARNING: modpost: vmlinux.o(.text+0x12ced2d): Section mismatch \
in reference from the function smc_sysctl_exit() to the variable
.init.data:smc_sysctl_ops
The function smc_sysctl_exit() references
the variable __initdata smc_sysctl_ops.
This is often because smc_sysctl_exit lacks a __initdata
annotation or the annotation of smc_sysctl_ops is wrong.
and
net/smc/smc_sysctl.c: In function 'smc_sysctl_init_net':
net/smc/smc_sysctl.c:47:17: error: 'struct netns_smc' has no member named 'smc_hdr'
47 | net->smc.smc_hdr = register_net_sysctl(net, "net/smc", table);
Since we don't need global sysctl initialization. To make things
clean and simple, remove the global pernet_operations and
smc_sysctl_{init|exit}. Call smc_sysctl_net_{init|exit} directly
from smc_net_{init|exit}.
Also initialized sysctl_autocorking_size if CONFIG_SYSCTL it not
set, this make sure SMC autocorking is enabled by default if
CONFIG_SYSCTL is not set.
Fixes: 462791bbfa ("net/smc: add sysctl interface for SMC")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Tested-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit a505cce6f7.
Leon says:
We already discussed that. SMC should be changed to use
RDMA CQ pool API
drivers/infiniband/core/cq.c.
ib_poll_handler() has much better implementation (tracing,
IRQ rescheduling, proper error handling) than this SMC variant.
Since we will switch to ib_poll_handler() in the future,
revert this patch.
Link: https://lore.kernel.org/netdev/20220301105332.GA9417@linux.alibaba.com/
Suggested-by: Leon Romanovsky <leon@kernel.org>
Suggested-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The problem of SMC_CLC_DECL_ERR_REGRMB on the server is very clear.
Based on the fact that whether a new SMC connection can be accepted or
not depends on not only the limit of conn nums, but also the available
entries of rtoken. Since the rtoken release is trigger by peer, while
the conn nums is decrease by local, tons of thing can happen in this
time difference.
This only thing that needs to be mentioned is that now all connection
creations are completely protected by smc_server_lgr_pending lock, it's
enough to check only the available entries in rtokens_used_mask.
Fixes: cd6851f303 ("smc: remote memory buffers (RMBs)")
Signed-off-by: D. Wythe <alibuda@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The main reason for this unexpected SMC_CLC_DECL_ERR_REGRMB in client
dues to following execution sequence:
Server Conn A: Server Conn B: Client Conn B:
smc_lgr_unregister_conn
smc_lgr_register_conn
smc_clc_send_accept ->
smc_rtoken_add
smcr_buf_unuse
-> Client Conn A:
smc_rtoken_delete
smc_lgr_unregister_conn() makes current link available to assigned to new
incoming connection, while smcr_buf_unuse() has not executed yet, which
means that smc_rtoken_add may fail because of insufficient rtoken_entry,
reversing their execution order will avoid this problem.
Fixes: 3e034725c0 ("net/smc: common functions for RMBs and send buffers")
Signed-off-by: D. Wythe <alibuda@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Send data all the way down to the RDMA device is a time
consuming operation(get a new slot, maybe do RDMA Write
and send a CDC, etc). Moving those operations from BH
to user context is good for performance.
If the sock_lock is hold by user, we don't try to send
data out in the BH context, but just mark we should
send. Since the user will release the sock_lock soon, we
can do the sending there.
Add smc_release_cb() which will be called in release_sock()
and try send in the callback if needed.
This patch moves the sending part out from BH if sock lock
is hold by user. In my testing environment, this saves about
20% softirq in the qperf 4K tcp_bw test in the sender side
with no noticeable throughput drop.
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we are handling softirq workload, enable hardirq may
again interrupt the current routine of softirq, and then
try to raise softirq again. This only wastes CPU cycles
and won't have any real gain.
Since IB_CQ_REPORT_MISSED_EVENTS already make sure if
ib_req_notify_cq() returns 0, it is safe to wait for the
next event, with no need to poll the CQ again in this case.
This patch disables hardirq during the processing of softirq,
and re-arm the CQ after softirq is done. Somehow like NAPI.
Co-developed-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rmbe_update_limit is used to limit announcing receive
window updating too frequently. RFC7609 request a minimal
increase in the window size of 10% of the receive buffer
space. But current implementation used:
min_t(int, rmbe_size / 10, SOCK_MIN_SNDBUF / 2)
and SOCK_MIN_SNDBUF / 2 == 2304 Bytes, which is almost
always less then 10% of the receive buffer space.
This causes the receiver always sending CDC message to
update its consumer cursor when it consumes more then 2K
of data. And as a result, we may encounter something like
"TCP silly window syndrome" when sending 2.5~8K message.
This patch fixes this using max(rmbe_size / 10, SOCK_MIN_SNDBUF / 2).
With this patch and SMC autocorking enabled, qperf 2K/4K/8K
tcp_bw test shows 45%/75%/40% increase in throughput respectively.
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit ea785a1a573b("net/smc: Send directly when
TCP_CORK is cleared"), we don't use delayed work
to implement cork.
This patch use the same algorithm, removes the
delayed work when setting TCP_NODELAY and send
directly in setsockopt(). This also makes the
TCP_NODELAY the same as TCP.
Cc: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This add a new sysctl: net.smc.autocorking_size
We can dynamically change the behaviour of autocorking
by change the value of autocorking_size.
Setting to 0 disables autocorking in SMC
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds autocorking support for SMC which could improve
throughput for small message by x3+.
The main idea is borrowed from TCP autocorking with some RDMA
specific modification:
1. The first message should never cork to make sure we won't
bring extra latency
2. If we have posted any Tx WRs to the NIC that have not
completed, cork the new messages until:
a) Receive CQE for the last Tx WR
b) We have corked enough message on the connection
3. Try to push the corked data out when we receive CQE of
the last Tx WR to prevent the corked messages hang in
the send queue.
Both SMC autocorking and TCP autocorking check the TX completion
to decide whether we should cork or not. The difference is
when we got a SMC Tx WR completion, the data have been confirmed
by the RNIC while TCP TX completion just tells us the data
have been sent out by the local NIC.
Add an atomic variable tx_pushing in smc_connection to make
sure only one can send to let it cork more and save CDC slot.
SMC autocorking should not bring extra latency since the first
message will always been sent out immediately.
The qperf tcp_bw test shows more than x4 increase under small
message size with Mellanox connectX4-Lx, same result with other
throughput benchmarks like sockperf/netperf.
The qperf tcp_lat test shows SMC autocorking has not increase any
ping-pong latency.
Test command:
client: smc_run taskset -c 1 qperf smc-server -oo msg_size:1:64K:*2 \
-t 30 -vu tcp_{bw|lat}
server: smc_run taskset -c 1 qperf
=== Bandwidth ====
MsgSize(Bytes) SMC-NoCork TCP SMC-AutoCorking
1 0.578 MB/s 2.392 MB/s(313.57%) 2.647 MB/s(357.72%)
2 1.159 MB/s 4.780 MB/s(312.53%) 5.153 MB/s(344.71%)
4 2.283 MB/s 10.266 MB/s(349.77%) 10.363 MB/s(354.02%)
8 4.668 MB/s 19.040 MB/s(307.86%) 21.215 MB/s(354.45%)
16 9.147 MB/s 38.904 MB/s(325.31%) 41.740 MB/s(356.32%)
32 18.369 MB/s 79.587 MB/s(333.25%) 82.392 MB/s(348.52%)
64 36.562 MB/s 148.668 MB/s(306.61%) 161.564 MB/s(341.89%)
128 72.961 MB/s 274.913 MB/s(276.80%) 325.363 MB/s(345.94%)
256 144.705 MB/s 512.059 MB/s(253.86%) 633.743 MB/s(337.96%)
512 288.873 MB/s 884.977 MB/s(206.35%) 1250.681 MB/s(332.95%)
1024 574.180 MB/s 1337.736 MB/s(132.98%) 2246.121 MB/s(291.19%)
2048 1095.192 MB/s 1865.952 MB/s( 70.38%) 2057.767 MB/s( 87.89%)
4096 2066.157 MB/s 2380.337 MB/s( 15.21%) 2173.983 MB/s( 5.22%)
8192 3717.198 MB/s 2733.073 MB/s(-26.47%) 3491.223 MB/s( -6.08%)
16384 4742.221 MB/s 2958.693 MB/s(-37.61%) 4637.692 MB/s( -2.20%)
32768 5349.550 MB/s 3061.285 MB/s(-42.77%) 5385.796 MB/s( 0.68%)
65536 5162.919 MB/s 3731.408 MB/s(-27.73%) 5223.890 MB/s( 1.18%)
==== Latency ====
MsgSize(Bytes) SMC-NoCork TCP SMC-AutoCorking
1 10.540 us 11.938 us( 13.26%) 10.573 us( 0.31%)
2 10.996 us 11.992 us( 9.06%) 10.269 us( -6.61%)
4 10.229 us 11.687 us( 14.25%) 10.240 us( 0.11%)
8 10.203 us 11.653 us( 14.21%) 10.402 us( 1.95%)
16 10.530 us 11.313 us( 7.44%) 10.599 us( 0.66%)
32 10.241 us 11.586 us( 13.13%) 10.223 us( -0.18%)
64 10.693 us 11.652 us( 8.97%) 10.251 us( -4.13%)
128 10.597 us 11.579 us( 9.27%) 10.494 us( -0.97%)
256 10.409 us 11.957 us( 14.87%) 10.710 us( 2.89%)
512 11.088 us 12.505 us( 12.78%) 10.547 us( -4.88%)
1024 11.240 us 12.255 us( 9.03%) 10.787 us( -4.03%)
2048 11.485 us 16.970 us( 47.76%) 11.256 us( -1.99%)
4096 12.077 us 13.948 us( 15.49%) 12.230 us( 1.27%)
8192 13.683 us 16.693 us( 22.00%) 13.786 us( 0.75%)
16384 16.470 us 23.615 us( 43.38%) 16.459 us( -0.07%)
32768 22.540 us 40.966 us( 81.75%) 23.284 us( 3.30%)
65536 34.192 us 73.003 us(113.51%) 34.233 us( 0.12%)
With SMC autocorking support, we can archive better throughput
than TCP in most message sizes without any latency trade-off.
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch add sysctl interface to support container environment
for SMC as we talk in the mail list.
Link: https://lore.kernel.org/netdev/20220224020253.GF5443@linux.alibaba.com
Co-developed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Dust Li <dust.li@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This also calls trace_smc_tx_sendmsg() even if data is corked. For ease
of understanding, if statements are not expanded here.
Link: https://lore.kernel.org/all/f4166712-9a1e-51a0-409d-b7df25a66c52@linux.ibm.com/
Fixes: 139653bc66 ("net/smc: Remove corked dealyed work")
Suggested-by: Stefan Raspl <raspl@linux.ibm.com>
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch calls smc_ib_unregister_client() when tcp_register_ulp()
fails, and make sure to clean it up.
Fixes: d7cd421da9 ("net/smc: Introduce TCP ULP support")
Signed-off-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There's a potential leak issue under following execution sequence :
smc_release smc_connect_work
if (sk->sk_state == SMC_INIT)
send_clc_confirim
tcp_abort();
...
sk.sk_state = SMC_ACTIVE
smc_close_active
switch(sk->sk_state) {
...
case SMC_ACTIVE:
smc_close_final()
// then wait peer closed
Unfortunately, tcp_abort() may discard CLC CONFIRM messages that are
still in the tcp send buffer, in which case our connection token cannot
be delivered to the server side, which means that we cannot get a
passive close message at all. Therefore, it is impossible for the to be
disconnected at all.
This patch tries a very simple way to avoid this issue, once the state
has changed to SMC_ACTIVE after tcp_abort(), we can actively abort the
smc connection, considering that the state is SMC_INIT before
tcp_abort(), abandoning the complete disconnection process should not
cause too much problem.
In fact, this problem may exist as long as the CLC CONFIRM message is
not received by the server. Whether a timer should be added after
smc_close_final() needs to be discussed in the future. But even so, this
patch provides a faster release for connection in above case, it should
also be valuable.
Fixes: 39f41f367b ("net/smc: common release code for non-accepted sockets")
Signed-off-by: D. Wythe <alibuda@linux.alibaba.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
smc_pnetid_by_table_ib() uses read_lock() and then it calls smc_pnet_apply_ib()
which, in turn, calls mutex_lock(&smc_ib_devices.mutex).
read_lock() disables preemption. Therefore, the code acquires a mutex while in
atomic context and it leads to a SAC bug.
Fix this bug by replacing the rwlock with a mutex.
Reported-and-tested-by: syzbot+4f322a6d84e991c38775@syzkaller.appspotmail.com
Fixes: 64e28b52c7 ("net/smc: add pnet table namespace support")
Confirmed-by: Tony Lu <tonylu@linux.alibaba.com>
Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com>
Acked-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/20220223100252.22562-1-fmdefrancesco@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
These two error paths need to release_sock(sk) before returning.
Fixes: a6a6fe27ba ("net/smc: Dynamic control handshake limitation by socket options")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: D. Wythe <alibuda@linux.alibaba.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When smc_connect_clc() times out, it will return -EAGAIN(tcp_recvmsg
retuns -EAGAIN while timeout), then this value will passed to the
application, which is quite confusing to the applications, makes
inconsistency with TCP.
From the manual of connect, ETIMEDOUT is more suitable, and this patch
try convert EAGAIN to ETIMEDOUT in that case.
Signed-off-by: D. Wythe <alibuda@linux.alibaba.com>
Reviewed-by: Karsten Graul <kgraul@linux.ibm.com>
Link: https://lore.kernel.org/r/1644913490-21594-1-git-send-email-alibuda@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
