Instead of rescheduling rxe_requester from rxe_completer() just extend the
duration of rxe_sender() by one pass. Setting run_requester_again forces
rxe_completer() to return 0 which will cause rxe_sender() to be called at
least one more time.
Link: https://lore.kernel.org/r/20240329145513.35381-10-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Now that rxe_completer() is always called serially after rxe_requester()
there is no reason to schedule rxe_completer() from rxe_requester().
Link: https://lore.kernel.org/r/20240329145513.35381-9-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Now that req.task and comp.task are merged it is no longer necessary to
call save_state() before calling rxe_xmit_pkt() and rollback_state() if
rxe_xmit_pkt() fails. This was done originally to prevent races between
rxe_completer() and rxe_requester() which now cannot happen.
Link: https://lore.kernel.org/r/20240329145513.35381-8-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Currently the rxe driver has three work queue tasks per qp. These are the
req.task, comp.task and resp.task which call rxe_requester(),
rxe_completer() and rxe_responder() respectively directly or on work
queues. Each of these subroutines checks to see if there is work to be
performed on the send queue or on the response packet queue or the request
packet queue and will run until there is no work remaining or yield the
cpu and reschedule itself until there is no work remaining.
This commit combines the req.task and comp.task into a single send.task
and renames the resp.task to the recv.task. The combined send.task calls
rxe_requester() and rxe_completer() serially and continues until all work
on both the send queue and the response packet queue are done.
In various benchmarks the performance is either improved or left the
same. At high scale there is a significant reduction in the load on the
cpu.
This is the first step in combining these two tasks. Once they are
serialized cross rescheduling of req.task and comp.task can be more
efficiently handled by just letting the send.task continue to run. This
will be done in the next several patches.
Link: https://lore.kernel.org/r/20240329145513.35381-7-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In rxe_post_send_kernel() if the qp is in the error state after posting
the work requests the rxe_completer() task is scheduled.
But, the only way to move the qp into the error state is to call
rxe_qp_error() which also schedules the rxe_completer() task to drain the
queues. Calling it a second time has no effect. This commit removes the
redundant call.
Link: https://lore.kernel.org/r/20240329145513.35381-6-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
A previous commit incorrectly added an 'if(!err)' before scheduling the
requester task in rxe_post_send_kernel(). But if there were send wrs
successfully added to the send queue before a bad wr they might never get
executed.
This commit fixes this by scheduling the requester task if any wqes were
successfully posted in rxe_post_send_kernel() in rxe_verbs.c.
Link: https://lore.kernel.org/r/20240329145513.35381-5-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Fixes: 5bf944f24129 ("RDMA/rxe: Add error messages")
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In rxe_comp_queue_pkt() an incoming response packet skb is enqueued to the
resp_pkts queue and then a decision is made whether to run the completer
task inline or schedule it. Finally the skb is dereferenced to bump a 'hw'
performance counter. This is wrong because if the completer task is
already running in a separate thread it may have already processed the skb
and freed it which can cause a seg fault. This has been observed
infrequently in testing at high scale.
This patch fixes this by changing the order of enqueuing the packet until
after the counter is accessed.
Link: https://lore.kernel.org/r/20240329145513.35381-4-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Fixes: 0b1e5b99a48b ("IB/rxe: Add port protocol stats")
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
In the function __rxe_add_to_pool, the function xa_alloc_cyclic is
called. The return value of the function xa_alloc_cyclic is as below:
"
Return: 0 if the allocation succeeded without wrapping. 1 if the
allocation succeeded after wrapping, -ENOMEM if memory could not be
allocated or -EBUSY if there are no free entries in @limit.
"
But now the function __rxe_add_to_pool only returns -EINVAL. All the
returned error value should be returned to the caller.
Signed-off-by: Zhu Yanjun <yanjun.zhu@linux.dev>
Link: https://lore.kernel.org/r/20240408142142.792413-1-yanjun.zhu@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
This one is not needed since commit 954afc5a8fd8 ("RDMA/rxe:
Use members of generic struct in rxe_mr").
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20240202124144.16033-1-guoqing.jiang@linux.dev
Reviewed-by: Zhu Yanjun <yanjun.zhu@linux.dev>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
commit 9ac01f434a1e ("RDMA/rxe: Extend dbg log messages to err and info")
newly added this info. But it did only show null device when
the rdma_rxe is being loaded because dev_name(rxe->ib_dev->dev)
has not yet been assigned at the moment:
"(null): rxe_set_mtu: Set mtu to 1024"
Remove it to silent this message, check the mtu from it backend link
instead if needed.
CC: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
Link: https://lore.kernel.org/r/20240109083253.3629967-2-lizhijian@fujitsu.com
Reviewed-by: Zhu Yanjun <yanjun.zhu@linux.dev>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Previously rxe_{dbg,info,err}() macros are appended built-in newline,
but some users will add redundant newline sometimes. So remove the
built-in newline for these macros.
In terms of rxe_{dbg,info,err}_xxx() macros, because they don't have
built-in newline, append newline when using them.
CC: Daisuke Matsuda <matsuda-daisuke@fujitsu.com>
Reviewed-by: Daisuke Matsuda <matsuda-daisuke@fujitsu.com>
Acked-by: Zhu Yanjun <zyjzyj2000@gmail.com>
Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
Link: https://lore.kernel.org/r/20240109083253.3629967-1-lizhijian@fujitsu.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Run test_query_rc_qp against siw failed since siw didn't set qp_state
accordingly. To address it, introduce siw_qp_state_to_ib_qp_state
which convert SIW_QP_STATE_IDLE to IB_QPS_INIT which is similar as
in cxgb4.
rdma-core# ./build/bin/run_tests.py --dev siw0 tests.test_qp.QPTest.test_query_rc_qp -v
test_query_rc_qp (tests.test_qp.QPTest)
Queries an RC QP after creation. Verifies that its properties are as ... FAIL
======================================================================
FAIL: test_query_rc_qp (tests.test_qp.QPTest)
Queries an RC QP after creation. Verifies that its properties are as
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/gjiang/rdma-core/tests/test_qp.py", line 284, in test_query_rc_qp
self.query_qp_common_test(e.IBV_QPT_RC)
File "/home/gjiang/rdma-core/tests/test_qp.py", line 265, in query_qp_common_test
self.verify_qp_attrs(caps, e.IBV_QPS_INIT, qp_init_attr, qp_attr)
File "/home/gjiang/rdma-core/tests/test_qp.py", line 239, in verify_qp_attrs
self.assertEqual(state, attr.qp_state)
AssertionError: <ibv_qp_state.IBV_QPS_INIT: 1> != 0
----------------------------------------------------------------------
Ran 1 test in 0.057s
FAILED (failures=1)
Link: https://lore.kernel.org/r/20231203092655.28102-4-guoqing.jiang@linux.dev
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
There is no siw_sq_work_handler in code, change it with siw_tx_thread
since siw_run_sq -> siw_sq_resume -> siw_qp_sq_process.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-18-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
We can just check max_send_wr here given both max_send_wr and
max_recv_wr are defined as u32 type, and we also need to ensure
num_sqe (derived from max_send_wr) shouldn't be zero.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-16-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Let's move it into siw_sk_assign_cm_upcalls, then we only
need to get sk_callback_lock once.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-14-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
With the initialization of rv and the two added label, we can
simplifiy code a bit.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-13-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Remove the redundant code given they share the same logic.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-8-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Remove the redundate checking since siw_send_terminate check it inside.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-7-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Let's remove it since the failure case only falls through
to the useless label.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-5-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
There are some places share the same logic, factor a common
helper for it.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-3-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Add the wrapper function to get either pbl page or umem page.
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20231113115726.12762-2-guoqing.jiang@linux.dev
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Simplify siw_qp_prepare_tx() by using crypto_shash_digest() instead of
an init+update+final sequence. This should also improve performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20231029045839.154071-1-ebiggers@kernel.org
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Abandon siw private code to pin user pages during user
memory registration, but use ib_umem_get() instead.
This will help maintaining the driver in case of changes
to the memory subsystem.
Signed-off-by: Bernard Metzler <bmt@zurich.ibm.com>
Link: https://lore.kernel.org/r/20231104075643.195186-1-bmt@zurich.ibm.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Prepare for the coming implementation by GCC and Clang of the __counted_by
attribute. Flexible array members annotated with __counted_by can have
their accesses bounds-checked at run-time checking via CONFIG_UBSAN_BOUNDS
(for array indexing) and CONFIG_FORTIFY_SOURCE (for strcpy/memcpy-family
functions).
As found with Coccinelle[1], add __counted_by for struct siw_pbl.
[1] https://github.com/kees/kernel-tools/blob/trunk/coccinelle/examples/counted_by.cocci
Cc: Bernard Metzler <bmt@zurich.ibm.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Leon Romanovsky <leon@kernel.org>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20230929180431.3005464-4-keescook@chromium.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
In case immediate MPA request processing fails, the newly
created endpoint unlinks the listening endpoint and is
ready to be dropped. This special case was not handled
correctly by the code handling the later TCP socket close,
causing a NULL dereference crash in siw_cm_work_handler()
when dereferencing a NULL listener. We now also cancel
the useless MPA timeout, if immediate MPA request
processing fails.
This patch furthermore simplifies MPA processing in general:
Scheduling a useless TCP socket read in sk_data_ready() upcall
is now surpressed, if the socket is already moved out of
TCP_ESTABLISHED state.
Fixes: 6c52fdc244b5 ("rdma/siw: connection management")
Signed-off-by: Bernard Metzler <bmt@zurich.ibm.com>
Link: https://lore.kernel.org/r/20230905145822.446263-1-bmt@zurich.ibm.com
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Documentation/process/license-rules.rst and checkpatch expect the SPDX
identifier syntax for multiple licenses to use capital "OR". Correct it
to keep consistent format and avoid copy-paste issues.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://lore.kernel.org/r/20230823092912.122674-1-krzysztof.kozlowski@linaro.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Many small changes across the subystem, some highlights:
- Usual driver cleanups in qedr, siw, erdma, hfi1, mlx4/5, irdma, mthca,
hns, and bnxt_re
- siw now works over tunnel and other netdevs with a MAC address by
removing assumptions about a MAC/GID from the connection manager
- "Doorbell Pacing" for bnxt_re - this is a best effort scheme to allow
userspace to slow down the doorbell rings if the HW gets full
- irdma egress VLAN priority, better QP/WQ sizing
- rxe bug fixes in queue draining and srq resizing
- Support more ethernet speed options in the core layer
- DMABUF support for bnxt_re
- Multi-stage MTT support for erdma to allow much bigger MR registrations
- A irdma fix with a CVE that came in too late to go to -rc, missing
bounds checking for 0 length MRs
-----BEGIN PGP SIGNATURE-----
iHUEABYIAB0WIQRRRCHOFoQz/8F5bUaFwuHvBreFYQUCZPEqkAAKCRCFwuHvBreF
YZrNAPoCBfU+VjCKNr2yqF7s52os5ZdBV7Uuh4txHcXWW9H7GAD/f19i2u62fzNu
C27jj4cztemMBb8mgwyxPw/wLg7NLwY=
=pC6k
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Pull rdma updates from Jason Gunthorpe:
"Many small changes across the subystem, some highlights:
- Usual driver cleanups in qedr, siw, erdma, hfi1, mlx4/5, irdma,
mthca, hns, and bnxt_re
- siw now works over tunnel and other netdevs with a MAC address by
removing assumptions about a MAC/GID from the connection manager
- "Doorbell Pacing" for bnxt_re - this is a best effort scheme to
allow userspace to slow down the doorbell rings if the HW gets full
- irdma egress VLAN priority, better QP/WQ sizing
- rxe bug fixes in queue draining and srq resizing
- Support more ethernet speed options in the core layer
- DMABUF support for bnxt_re
- Multi-stage MTT support for erdma to allow much bigger MR
registrations
- A irdma fix with a CVE that came in too late to go to -rc, missing
bounds checking for 0 length MRs"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (87 commits)
IB/hfi1: Reduce printing of errors during driver shut down
RDMA/hfi1: Move user SDMA system memory pinning code to its own file
RDMA/hfi1: Use list_for_each_entry() helper
RDMA/mlx5: Fix trailing */ formatting in block comment
RDMA/rxe: Fix redundant break statement in switch-case.
RDMA/efa: Fix wrong resources deallocation order
RDMA/siw: Call llist_reverse_order in siw_run_sq
RDMA/siw: Correct wrong debug message
RDMA/siw: Balance the reference of cep->kref in the error path
Revert "IB/isert: Fix incorrect release of isert connection"
RDMA/bnxt_re: Fix kernel doc errors
RDMA/irdma: Prevent zero-length STAG registration
RDMA/erdma: Implement hierarchical MTT
RDMA/erdma: Refactor the storage structure of MTT entries
RDMA/erdma: Renaming variable names and field names of struct erdma_mem
RDMA/hns: Support hns HW stats
RDMA/hns: Dump whole QP/CQ/MR resource in raw
RDMA/irdma: Add missing kernel-doc in irdma_setup_umode_qp()
RDMA/mlx4: Copy union directly
RDMA/irdma: Drop unused kernel push code
...
We need to print num_sle first then pbl->max_buf per the condition.
Also replace mem->pbl with pbl while at it.
Fixes: 303ae1cdfdf7 ("rdma/siw: application interface")
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20230821133255.31111-3-guoqing.jiang@linux.dev
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
The siw_connect can go to err in below after cep is allocated successfully:
1. If siw_cm_alloc_work returns failure. In this case socket is not
assoicated with cep so siw_cep_put can't be called by siw_socket_disassoc.
We need to call siw_cep_put twice since cep->kref is increased once after
it was initialized.
2. If siw_cm_queue_work can't find a work, which means siw_cep_get is not
called in siw_cm_queue_work, so cep->kref is increased twice by siw_cep_get
and when associate socket with cep after it was initialized. So we need to
call siw_cep_put three times (one in siw_socket_disassoc).
3. siw_send_mpareqrep returns error, this scenario is similar as 2.
So we need to remove one siw_cep_put in the error path.
Fixes: 6c52fdc244b5 ("rdma/siw: connection management")
Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Link: https://lore.kernel.org/r/20230821133255.31111-2-guoqing.jiang@linux.dev
Acked-by: Bernard Metzler <bmt@zurich.ibm.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
If a send packet is dropped by the IP layer in rxe_requester()
the call to rxe_xmit_packet() can fail with err == -EAGAIN.
To recover, the state of the wqe is restored to the state before
the packet was sent so it can be resent. However, the routines
that save and restore the state miss a significnt part of the
variable state in the wqe, the dma struct which is used to process
through the sge table. And, the state is not saved before the packet
is built which modifies the dma struct.
Under heavy stress testing with many QPs on a fast node sending
large messages to a slow node dropped packets are observed and
the resent packets are corrupted because the dma struct was not
restored. This patch fixes this behavior and allows the test cases
to succeed.
Fixes: 3050b9985024 ("IB/rxe: Fix race condition between requester and completer")
Link: https://lore.kernel.org/r/20230721200748.4604-1-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
This patch corrects an error in rxe_modify_srq where if the
caller changes the srq size the actual new value is not returned
to the caller since it may be larger than what is requested.
Additionally it open codes the subroutine rcv_wqe_size() which
adds very little value, and makes some whitespace changes.
Fixes: 8700e3e7c485 ("Soft RoCE driver")
Link: https://lore.kernel.org/r/20230620140142.9452-1-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
This patch:
- Moves code to initialize a qp send work queue to a
subroutine named rxe_init_sq.
- Moves code to initialize a qp recv work queue to a
subroutine named rxe_init_rq.
- Moves initialization of qp request and response packet
queues ahead of work queue initialization so that cleanup
of a qp if it is not fully completed can successfully
attempt to drain the packet queues without a seg fault.
- Makes minor whitespace cleanups.
Fixes: 8700e3e7c485 ("Soft RoCE driver")
Link: https://lore.kernel.org/r/20230620135519.9365-2-rpearsonhpe@gmail.com
Signed-off-by: Bob Pearson <rpearsonhpe@gmail.com>
Acked-by: Zhu Yanjun <zyjzyj2000@gmail.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Immediately removing the siw module after insertion may
crash in siw_stop_tx_thread(), if the according thread did
not yet had a chance to initialize its wait queue and
siw_stop_tx_thread() tries to wakeup that thread. Initializing
the threads state before spwaning it fixes it.
Reported-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Bernard Metzler <bmt@zurich.ibm.com>
Link: https://lore.kernel.org/r/20230728114418.124328-1-bmt@zurich.ibm.com
Tested-by: Guoqing Jiang <guoqing.jiang@linux.dev>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
LOOPBACK and NONE (tunnel) devices have all-zero MAC addresses.
Currently, siw_device_create() falls back to copying the IB device's
name in those cases, because an all-zero MAC address breaks the RDMA
core address resolution mechanism.
However, at the point when siw_device_create() constructs a GID, the
ib_device::name field is uninitialized, leaving the MAC address to
remain in an all-zero state.
Fabricate a random artificial GID for such devices, and ensure this
artificial GID is returned for all device query operations.
Link: https://lore.kernel.org/r/168960673260.3007.12378736853793339110.stgit@manet.1015granger.net
Reported-by: Tom Talpey <tom@talpey.com>
Fixes: a2d36b02c15d ("RDMA/siw: Enable siw on tunnel devices")
Reviewed-by: Bernard Metzler <bmt@zurich.ibm.com>
Reviewed-by: Tom Talpey <tom@talpey.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
