Avoid that sparse complains about using 0 as a pointer, about
missing function declarations and also avoid that sparse complains
about endianness.
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Andrew Boyer <andrew.boyer@dell.com>
Cc: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
- Series of qedr fixes
- Series of rxe fixes
- One isolated i40iw fix
- One isolated cma fix
- One isolated cxgb4 fix
-----BEGIN PGP SIGNATURE-----
iQIcBAABAgAGBQJYXAGvAAoJELgmozMOVy/dDukQAMMNarWp0U8KfNYRU5tyCBwd
aIQC1gFT6GUCFys40Z6L84m1D3NpGR+vzVv3grVBeuge73b79zAOHXvVDwJCA+Jl
QQLG3vZ13C3158sLDiK8zL+4Ob5OfOQ5nQ2spvDfJWpye9SD+pWFcrpqvK02ANRN
kFHILk1gROBTNi46yBR5hjWOkw7Bua6XLsPxh6xoaDZ43NL0r0xgm43FTnj/19x3
0zpZYYKP+3C6U7678rqaog9zfXHvadghW5/WBJ/VgfKqEmH89ESx4J2MvbB8DxFD
1tWAOpr5TNY5jnh8mtUsceDjCzQivc/RWqAu05BspEwcavjSLFyRYr1epR0/4oAd
PqLSmfORmhpJ8+5Kmn+chtXo3TT4SYGHIzSUbgbEV/ClwX/7UW+w8mfQZ3buUBq/
cQp/oRnJcsrQIEDFO3AH7P+6Sxy6t3zbSl5oKBUOI1u4RFmC7YBPqo9fQu2Z2mGk
3+AWQaPr7qgEcFzXBgLzvd4LhTYKsvmiNwrcXi9KjjwQjNEVg15qqF2YtmxEUgi9
kh3IOcGan3iSblhV/WLrxcOjlPQrPpBOVnTPhUskFtlsrD+032OxeOBpVoU3nCUt
MjTYWoNTYdw4wHz0w373o0uR4+4nl4a5OmO4Fh6Drmg5hm4Bl9BWy0Kziu93Z1Ay
Z2utZVWLWhBzn8yJujUz
=NW9g
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull rdma fixes from Doug Ledford:
"First round of -rc fixes for 4.10 kernel:
- a series of qedr fixes
- a series of rxe fixes
- one i40iw fix
- one cma fix
- one cxgb4 fix"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma:
IB/rxe: Don't check for null ptr in send()
IB/rxe: Drop future atomic/read packets rather than retrying
IB/rxe: Use BTH_PSN_MASK when ACKing duplicate sends
qedr: Always notify the verb consumer of flushed CQEs
qedr: clear the vendor error field in the work completion
qedr: post_send/recv according to QP state
qedr: ignore inline flag in read verbs
qedr: modify QP state to error when destroying it
qedr: return correct value on modify qp
qedr: return error if destroy CQ failed
qedr: configure the number of CQEs on CQ creation
i40iw: Set 128B as the only supported RQ WQE size
IB/cma: Fix a race condition in iboe_addr_get_sgid()
IB/rxe: Fix a memory leak in rxe_qp_cleanup()
iw_cxgb4: set correct FetchBurstMax for QPs
pkt->qp was already dereferenced earlier in the function.
Fixes Smatch complaint:
drivers/infiniband/sw/rxe/rxe_net.c:458 send()
warn: variable dereferenced before check 'pkt->qp' (see line 441)
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
If the completer is in the middle of a large read operation, one
lost packet can cause havoc. Going to COMPST_ERROR_RETRY will
cause the requester to resend the request. After that, any packet
from the first attempt still in the receive queue will be
interpreted as an error, restarting the error/retry sequence.
The transfer will quickly exhaust its retries.
This behavior is very noticeable when doing 512KB reads on a
QEMU system configured with 1500B MTU.
Also, a resent request here will prompt the responder on the
other side to immediately start resending, but the resent
packets will get stuck in the already-loaded receive queue and
will never be processed.
Rather than erroring out every time an unexpected future packet
arrives, just drop it. Eventually the retry timer will send a
duplicate request; the completer will be able to make progress since
the queue will start relatively empty.
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A socket is associated with every QP by the rxe driver but sock_release()
is never called. Add a call to sock_release() in rxe_qp_cleanup().
Fixes: commit 8700e3e7c48A5 ("Add Soft RoCE driver")
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Moni Shoua <monis@mellanox.com>
Cc: Kamal Heib <kamalh@mellanox.com>
Cc: Amir Vadai <amirv@mellanox.com>
Cc: Haggai Eran <haggaie@mellanox.com>
Cc: <stable@vger.kernel.org>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Get rid of this warning:
drivers/infiniband/sw/rdmavt/cq.c: In function ‘rvt_cq_exit’:
drivers/infiniband/sw/rdmavt/cq.c:542:2: warning: ‘worker’ may be used uninitialized in this function [-Wmaybe-uninitialized]
kthread_destroy_worker(worker);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
by fixing the function to actually work.
Fixes: 6efaf10f16 ("IB/rdmavt: Avoid queuing work into a destroyed cq kthread worker")
Cc: Petr Mladek <pmladek@suse.com>
Cc: Doug Ledford <dledford@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
- Shared mlx5 updates with net stack (will drop out on merge if Dave's
tree has already been merged)
- Driver updates: cxgb4, hfi1, hns-roce, i40iw, mlx4, mlx5, qedr, rxe
- Debug cleanups
- New connection rejection helpers
- SRP updates
- Various misc fixes
- New paravirt driver from vmware
-----BEGIN PGP SIGNATURE-----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=QeGy
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull rdma updates from Doug Ledford:
"This is the complete update for the rdma stack for this release cycle.
Most of it is typical driver and core updates, but there is the
entirely new VMWare pvrdma driver. You may have noticed that there
were changes in DaveM's pull request to the bnxt Ethernet driver to
support a RoCE RDMA driver. The bnxt_re driver was tentatively set to
be pulled in this release cycle, but it simply wasn't ready in time
and was dropped (a few review comments still to address, and some
multi-arch build issues like prefetch() not working across all
arches).
Summary:
- shared mlx5 updates with net stack (will drop out on merge if
Dave's tree has already been merged)
- driver updates: cxgb4, hfi1, hns-roce, i40iw, mlx4, mlx5, qedr, rxe
- debug cleanups
- new connection rejection helpers
- SRP updates
- various misc fixes
- new paravirt driver from vmware"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (210 commits)
IB: Add vmw_pvrdma driver
IB/mlx4: fix improper return value
IB/ocrdma: fix bad initialization
infiniband: nes: return value of skb_linearize should be handled
MAINTAINERS: Update Intel RDMA RNIC driver maintainers
MAINTAINERS: Remove Mitesh Ahuja from emulex maintainers
IB/core: fix unmap_sg argument
qede: fix general protection fault may occur on probe
IB/mthca: Replace pci_pool_alloc by pci_pool_zalloc
mlx5, calc_sq_size(): Make a debug message more informative
mlx5: Remove a set-but-not-used variable
mlx5: Use { } instead of { 0 } to init struct
IB/srp: Make writing the add_target sysfs attr interruptible
IB/srp: Make mapping failures easier to debug
IB/srp: Make login failures easier to debug
IB/srp: Introduce a local variable in srp_add_one()
IB/srp: Fix CONFIG_DYNAMIC_DEBUG=n build
IB/multicast: Check ib_find_pkey() return value
IPoIB: Avoid reading an uninitialized member variable
IB/mad: Fix an array index check
...
rvt_create_qp() creates qp->ip only when a qp creation request comes from
userspace (udata is not NULL). If we exceed the number of available
queue pairs however, the error path always attempts to put a kref to this
structure. If the requestor is inside the kernel, this leads to a crash.
We fix this by checking that qp->ip is not NULL before caling kref_put().
Signed-off-by: Jim Foraker <foraker1@llnl.gov>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Acked-by: Jonathan Toppins <jtoppins@redhat.com>
Acked-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Use the new API to create and destroy the cq kthread worker.
The API hides some implementation details.
In particular, kthread_create_worker() allocates and initializes
struct kthread_worker. It runs the kthread the right way and stores
task_struct into the worker structure. In addition, the *on_cpu()
variant binds the kthread to the given cpu and the related memory
node.
kthread_destroy_worker() flushes all pending works, stops
the kthread and frees the structure.
This patch does not change the existing behavior. Note that we must
use the on_cpu() variant because the function starts the kthread
and it must bind it to the right CPU before waking. The numa node
is associated for given CPU as well.
Signed-off-by: Petr Mladek <pmladek@suse.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The memory barrier is not enough to protect queuing works into
a destroyed cq kthread. Just imagine the following situation:
CPU1 CPU2
rvt_cq_enter()
worker = cq->rdi->worker;
rvt_cq_exit()
rdi->worker = NULL;
smp_wmb();
kthread_flush_worker(worker);
kthread_stop(worker->task);
kfree(worker);
// nothing queued yet =>
// nothing flushed and
// happily stopped and freed
if (likely(worker)) {
// true => read before CPU2 acted
cq->notify = RVT_CQ_NONE;
cq->triggered++;
kthread_queue_work(worker, &cq->comptask);
BANG: worker has been flushed/stopped/freed in the meantime.
This patch solves this by protecting the critical sections by
rdi->n_cqs_lock. It seems that this lock is not much contended
and looks reasonable for this purpose.
One catch is that rvt_cq_enter() might be called from IRQ context.
Therefore we must always take the lock with IRQs disabled to avoid
a possible deadlock.
Signed-off-by: Petr Mladek <pmladek@suse.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add struct ib_udata to the signature of create_ah callback that is
implemented by IB device drivers. This allows HW drivers to return extra
data to the userspace library.
This patch prepares the ground for mlx5 driver to resolve destination
mac address for a given GID and return it to userspace.
This patch was previously submitted by Knut Omang as a part of the
patch set to support Oracle's Infiniband HCA (SIF).
Signed-off-by: Knut Omang <knut.omang@oracle.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Increase limit of max CQE from 8K to 32K to allow demanding
applications to work over SoftRoCE with same configuration
as most RoCEv2 HW vendors have.
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
It might be possible for all of a QP's references to be dropped
while one of that QP's tasklets is running.
For example, the completer might run during QP destroy.
If qp->valid is false, it will drop all of the packets on
the resp_pkts list, potentially removing the last reference.
Then it tries to advance the SQ consumer pointer. If the
SQ's buffer has already been destroyed, the system will
panic.
To be safe, hold a reference on the QP for the duration
of each tasklet.
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The system may crash when a malformed request is received and
the error is detected by the responder.
NodeA: $ ibv_rc_pingpong -g 0 -d rxe0 -i 1 -n 1 -s 50000
NodeB: $ ibv_rc_pingpong -g 0 -d rxe0 -i 1 -n 1 -s 1024 <NodeA_ip>
The responder generates a receive error on node B since the incoming
SEND is oversized. If the client tears down the QP before the responder
or the completer finish running, a page fault may occur.
The fix makes the destroy operation spin until the tasks complete, which
appears to be original intent of the design.
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The udata->inlen error path needs to clean up the ref
added by rxe_alloc().
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Peek at the CQ after arming it so that we can return a hint.
This avoids missed completions due to a race between posting
CQEs and arming the CQ.
For example, CM teardown waits on MAD requests to complete with
ib_cq_poll_work(). Without this fix, the last completion might be
left on the CQ, hanging the kthread doing the teardown.
The console backtraces look like this:
[ 4199.911284] Call Trace:
[ 4199.911401] [<ffffffff9657fe95>] schedule+0x35/0x80
[ 4199.911556] [<ffffffff965830df>] schedule_timeout+0x22f/0x2c0
[ 4199.911727] [<ffffffff9657f7a8>] ? __schedule+0x368/0xa20
[ 4199.911891] [<ffffffff96580903>] wait_for_completion+0xb3/0x130
[ 4199.912067] [<ffffffff960a17e0>] ? wake_up_q+0x70/0x70
[ 4199.912243] [<ffffffffc074a06d>] cm_destroy_id+0x13d/0x450 [ib_cm]
[ 4199.912422] [<ffffffff961615d5>] ? printk+0x57/0x73
[ 4199.912578] [<ffffffffc074a390>] ib_destroy_cm_id+0x10/0x20 [ib_cm]
[ 4199.912759] [<ffffffffc076098c>] rdma_destroy_id+0xac/0x340 [rdma_cm]
[ 4199.912941] [<ffffffffc076f2cc>] 0xffffffffc076f2cc
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The last_psn algorithm fails in the zero-byte case: it calculates
first_psn = N, last_psn = N-1. This makes the operation unretryable since
the res structure will fail the (first_psn <= psn <= last_psn) test in
find_resource().
While here, use BTH_PSN_MASK to mask the calculated last_psn.
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
skb_out is decremented in rxe_skb_tx_dtor(), which is not called in the
loopback() path. Move the increment to the send() path rather than
rxe_xmit_packet().
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Acked-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A client might post a read followed by a send. The partner receives
and acknowledges both transactions, posting an RCQ entry for the
send, but something goes wrong with the read ACK. When the client
retries the read, the partner's responder processes the duplicate
read but incorrectly resets the PSN to the value preceding the
original send. When the duplicate send arrives, the responder cannot
tell that it is a duplicate, so the responder generates a duplicate
RCQ entry, confusing the client.
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Reviewed-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A simple userspace application might poll the CQ, find a completion,
and then attempt to post a new WQE to the SQ. A spurious error can
occur if the userspace application detects a full SQ in the instant
before the kernel is able to advance the SQ consumer pointer.
This is noticeable when using single-entry SQs with ibv_rc_pingpong
if lots of kernel and userspace library debugging is enabled.
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Reviewed-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Avoid smashing the stack when an ICRC error occurs on an IPv6 network.
Signed-off-by: Andrew Boyer <andrew.boyer@dell.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
It makes me nervous when we cast pointer parameters. I would estimate
that around 50% of the time, it indicates a bug. Here the cast is not
needed becaue u32 and and unsigned int are the same thing. Removing the
cast makes the code more robust and future proof in case any of the
types change.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Acked-by: Moni Shoua <monis@mellanox.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
spinlock can be initialized automatically with DEFINE_SPINLOCK()
rather than explicitly calling spin_lock_init().
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Leon Romanosky <leonro@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A race condition fix added an rxe_qp structure to the stack in order
to be able to perform rollback in rxe_requester(), but the structure
is large enough to trigger the warning for possible stack overflow:
drivers/infiniband/sw/rxe/rxe_req.c: In function 'rxe_requester':
drivers/infiniband/sw/rxe/rxe_req.c:757:1: error: the frame size of 2064 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]
This changes the rollback function to only save the psn inside
the qp, which is the only field we access in the rollback_qp
anyway.
Fixes: 3050b99850 ("IB/rxe: Fix race condition between requester and completer")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This is for use by client drivers to drive
send completions into a CQ.
A new exported table allows for the mapping
of ib_wr_opcode into a ib_wc_opcode.
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Some parts of the code don't use the standard driver
wrapper for memory region reference counters. Use the
standard driver wrapper throughout the code.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Some parts of the code don't use the standard release
wrapper rvt_put_qp() for decrementing and testing
the refcount to then try to use a resource.
Replace this code with the standard driver wrapper.
Fixes: Commit 4d6f85c3fa ("IB/rdmavt, IB/qib, IB/hfi1: Use new QP put get routines")
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Split rdmavt traces into separate files to preserve the original
hierarchy since only one trace sub system may now be defined per header
file.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The prints after [k|v][m|z|c]alloc() functions are not needed,
because in case of failure, allocator will print their internal
error prints anyway.
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The method rxe_qp_error() transitions QP to error state
and make sure the QP is drained. It did not though update
the QP state for user's query.
This patch fixes this.
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
RXE resets the send-q only once in rxe_qp_init_req() when
QP is created, but when the QP is reused after QP reset, the send-q
holds previous garbage data.
This garbage data wrongly fails CQEs that otherwise
should have completed successfully.
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
To correctly handle a erroneous WR this fix does the following
1. Make sure the bad WQE causes a user completion event.
2. Call rxe_completer to handle the erred WQE.
Before the fix, when rxe_requester found a bad WQE, it changed its
status to IB_WC_LOC_PROT_ERR and exit with 0 for non RC QPs.
If this was the 1st WQE then there would be no ACK to invoke the
completer and this bad WQE would be stuck in the QP's send-q.
On top of that the requester exiting with 0 caused rxe_do_task to
endlessly invoke rxe_requester, resulting in a soft-lockup attached
below.
In case the WQE was not the 1st and rxe_completer did get a chance to
handle the bad WQE, it did not cause a complete event since the WQE's
IB_SEND_SIGNALED flag was not set.
Setting WQE status to IB_SEND_SIGNALED is subject to IBA spec
version 1.2.1, section 10.7.3.1 Signaled Completions.
NMI watchdog: BUG: soft lockup - CPU#7 stuck for 22s!
[<ffffffffa0590145>] ? rxe_pool_get_index+0x35/0xb0 [rdma_rxe]
[<ffffffffa05952ec>] lookup_mem+0x3c/0xc0 [rdma_rxe]
[<ffffffffa0595534>] copy_data+0x1c4/0x230 [rdma_rxe]
[<ffffffffa058c180>] rxe_requester+0x9d0/0x1100 [rdma_rxe]
[<ffffffff8158e98a>] ? kfree_skbmem+0x5a/0x60
[<ffffffffa05962c9>] rxe_do_task+0x89/0xf0 [rdma_rxe]
[<ffffffffa05963e2>] rxe_run_task+0x12/0x30 [rdma_rxe]
[<ffffffffa059110a>] rxe_post_send+0x41a/0x550 [rdma_rxe]
[<ffffffff811ef922>] ? __kmalloc+0x182/0x200
[<ffffffff816ba512>] ? down_read+0x12/0x40
[<ffffffffa054bd32>] ib_uverbs_post_send+0x532/0x540 [ib_uverbs]
[<ffffffff815f8722>] ? tcp_sendmsg+0x402/0xb80
[<ffffffffa05453dc>] ib_uverbs_write+0x18c/0x3f0 [ib_uverbs]
[<ffffffff81623c2e>] ? inet_recvmsg+0x7e/0xb0
[<ffffffff8158764d>] ? sock_recvmsg+0x3d/0x50
[<ffffffff81215b87>] __vfs_write+0x37/0x140
[<ffffffff81216892>] vfs_write+0xb2/0x1b0
[<ffffffff81217ce5>] SyS_write+0x55/0xc0
[<ffffffff816bc672>] entry_SYSCALL_64_fastpath+0x1a/0xa
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Reviewed-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Profiling shows that the key validation is susceptible
to cache line trading when accessing the lkey table.
Fix by separating out the read mostly fields from the write
fields. In addition the shift amount, which is function
of the lkey table size, is precomputed and stored with the
table pointer. Since both the shift and table pointer
are in the same read mostly cacheline, this saves a cache
line in this hot path.
Reviewed-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The initial code for rdmavt carried with it a restriction that was a
vestige from the qib driver, that to dma map a page it had to be less
than a page size. This is not the case on modern hardware, both qib and
hfi1 will be just fine with unaligned map requests.
This fixes a 4.8 regression where by an IPoIB transfer of > PAGE_SIZE
will hang because the dma map page call always fails. This was
introduced after commit 5faba54695 ("IB/ipoib: Report SG feature
regardless of HW UD CSUM capability") added the capability to use SG by
default. Rather than override this, the HW supports it, so allow SG.
Cc: Stable <stable@vger.kernel.org> # 4.8
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
A good practice is to prefix the names of functions by the name
of the subsystem.
The kthread worker API is a mix of classic kthreads and workqueues. Each
worker has a dedicated kthread. It runs a generic function that process
queued works. It is implemented as part of the kthread subsystem.
This patch renames the existing kthread worker API to use
the corresponding name from the workqueues API prefixed by
kthread_:
__init_kthread_worker() -> __kthread_init_worker()
init_kthread_worker() -> kthread_init_worker()
init_kthread_work() -> kthread_init_work()
insert_kthread_work() -> kthread_insert_work()
queue_kthread_work() -> kthread_queue_work()
flush_kthread_work() -> kthread_flush_work()
flush_kthread_worker() -> kthread_flush_worker()
Note that the names of DEFINE_KTHREAD_WORK*() macros stay
as they are. It is common that the "DEFINE_" prefix has
precedence over the subsystem names.
Note that INIT() macros and init() functions use different
naming scheme. There is no good solution. There are several
reasons for this solution:
+ "init" in the function names stands for the verb "initialize"
aka "initialize worker". While "INIT" in the macro names
stands for the noun "INITIALIZER" aka "worker initializer".
+ INIT() macros are used only in DEFINE() macros
+ init() functions are used close to the other kthread()
functions. It looks much better if all the functions
use the same scheme.
+ There will be also kthread_destroy_worker() that will
be used close to kthread_cancel_work(). It is related
to the init() function. Again it looks better if all
functions use the same naming scheme.
+ there are several precedents for such init() function
names, e.g. amd_iommu_init_device(), free_area_init_node(),
jump_label_init_type(), regmap_init_mmio_clk(),
+ It is not an argument but it was inconsistent even before.
[arnd@arndb.de: fix linux-next merge conflict]
Link: http://lkml.kernel.org/r/20160908135724.1311726-1-arnd@arndb.de
Link: http://lkml.kernel.org/r/1470754545-17632-3-git-send-email-pmladek@suse.com
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Petr Mladek <pmladek@suse.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Borislav Petkov <bp@suse.de>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1. Debugging qp state transitions and qp errors in loopback and
multiple QP tests is difficult without qp numbers in debug logs.
This patch adds qp number to important debug logs.
2. Instead of having rxe: prefix in few logs and not having in
few logs, using uniform module name prefix using pr_fmt macro.
3. Code cleanup for various warnings reported by checkpatch for
incomplete unsigned data type, line over 80 characters, return
statements.
Signed-off-by: Parav Pandit <pandit.parav@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is a problem when CONFIG_RDMA_RXE=y and CONFIG_IPV6=y. This
results in the rdma_rxe initialization occurring before the IPv6
services are ready. This patch delays the initialization of rdma_rxe
until after the IPv6 services are ready. This fix is based on one
proposed by Logan Gunthorpe on a much older code base.
Signed-off-by: Stephen Bates <sbates@raithlin.com>
Reviewed-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch fixes below kernel crash on memory registration for rxe
and other transport drivers which has dma_ops extension.
IB/core invokes ib_map_sg_attrs() in generic manner with dma attributes
which is used by mlx5 and mthca adapters. However in doing so it
ignored honoring dma_ops extension of software based transports for
sg map/unmap operation. This results in calling dma_map_sg_attrs of
hardware virtual device resulting in crash for null reference.
We extend the core to support sg_map/unmap_attrs and transport drivers
to implement those dma_ops callback functions.
Verified usign perftest applications.
BUG: unable to handle kernel NULL pointer dereference at (null)
IP: [<ffffffff81032a75>] check_addr+0x35/0x60
...
Call Trace:
[<ffffffff81032b39>] ? nommu_map_sg+0x99/0xd0
[<ffffffffa02b31c6>] ib_umem_get+0x3d6/0x470 [ib_core]
[<ffffffffa01cc329>] rxe_mem_init_user+0x49/0x270 [rdma_rxe]
[<ffffffffa01c793a>] ? rxe_add_index+0xca/0x100 [rdma_rxe]
[<ffffffffa01c995f>] rxe_reg_user_mr+0x9f/0x130 [rdma_rxe]
[<ffffffffa00419fe>] ib_uverbs_reg_mr+0x14e/0x2c0 [ib_uverbs]
[<ffffffffa003d3ab>] ib_uverbs_write+0x15b/0x3b0 [ib_uverbs]
[<ffffffff811e92a6>] ? mem_cgroup_commit_charge+0x76/0xe0
[<ffffffff811af0a9>] ? page_add_new_anon_rmap+0x89/0xc0
[<ffffffff8117e6c9>] ? lru_cache_add_active_or_unevictable+0x39/0xc0
[<ffffffff811f0da8>] __vfs_write+0x28/0x120
[<ffffffff811f1239>] ? rw_verify_area+0x49/0xb0
[<ffffffff811f1492>] vfs_write+0xb2/0x1b0
[<ffffffff811f27d6>] SyS_write+0x46/0xa0
[<ffffffff814f7d32>] entry_SYSCALL_64_fastpath+0x1a/0xa4
Signed-off-by: Parav Pandit <pandit.parav@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Both prepare4 and prepare6 sets loopback mask in pkt_info structure
instance of skb. The xmit_packet and other requester side functions
use a pkt_info struct from the stack without the proper mask. This
results in sending out the packet to the actual netdev device and
loopback functionality is broken.
Modify prepare() to pass its correctly marked pkt_info struct to
prepare4() and prepare6() instead of them using SKB_TO_PKT(skb) and
getting an incorrectly set mask.
Verified with perftest applications.
Signed-off-by: Parav Pandit <pandit.parav@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch avoids scheduing tasklet for WQE and protocol processing
for user space QP. It performs the task in calling process context.
To improve code readability kernel specific post_send handling moved to
post_send_kernel() function.
Signed-off-by: Parav Pandit <pandit.parav@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Corrected function name in comment from qib_ to rvt_.
Signed-off-by: Parav Pandit <pandit.parav@gmail.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch adds lockdep asserts in key code paths for
insuring lock correctness.
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add an rvt_qp_init() to initialize specific
common fields as the qp is created or reset.
The routine is shared by the rvt_reset_qp() and
the rvt_create_qp().
The intent is that lock dep assertions will only
appear in the rvt_reset_qp().
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The reset calldown is misplaced.
It should only be called in the code that actually
transitions the QP to reset.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The __must_hold() is sufficent to correct the sparse
context imbalance inside a function.
Per Documentation/sparse.txt:
__must_hold - The specified lock is held on function entry and exit.
Fixes: Commit c0a67f6ba3 ("IB/rdmavt: Annotate rvt_reset_qp()")
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This improves readability and hides the reference count
mechanism from the client drivers.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The userspace memory region 'mr' is allocated with kzalloc in
__rvt_alloc_mr however it is incorrectly being freed with vfree in
__rvt_free_mr. Fix this by using kfree to free it.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
rxe_requester() is sending a pkt with rxe_xmit_packet() and
then calls rxe_update() to update the wqe and qp's psn values.
But sometimes the response is received before the requester
had time to update the wqe in which case the completer
acts on errornous wqe values.
This fix updates the wqe and qp before actually sending
the request and rolls back when xmit fails.
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
When handling ack for atomic opcodes like "fetch&add"
or "cmp&swp", the method send_atomic_ack() saves the ack
before sending it, in case it gets lost and never reach the
requester. In which case the method duplicate_request()
will need to find it using the duplicated request.psn.
But send_atomic_ack() used a wrong psn value and thus
the above ack was never found.
This fix uses the ack.psn to locate the ack in case
its needed.
This fix also copies the ack packet to the skb's control buffer
since duplicate_request() will need it when calling rxe_xmit_packet()
Fixes: 8700e3e7c4 ("Soft RoCE driver")
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is skb_clone(skb, GFP_KERNEL) in spinlock context
in rxe_rcv_mcast_pkt().
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Acked-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The unwind logic for creating a user QP has a double vfree
of the non-shared receive queue when handling a "too many qps"
failure.
The code unwinds the mmmap info by decrementing a reference
count which will call rvt_release_mmap_info() which in turn
does the vfree() of the r_rq.wq. The unwind code then does
the same free.
Fix by guarding the vfree() with the same test that is done
in close and only do the vfree() if qp->ip is NULL.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
- hfi1 driver updates
- Fix for max SGEs allowed via RDMA R/W API
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=cQcp
-----END PGP SIGNATURE-----
Merge tag 'for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull second round of rdma updates from Doug Ledford:
"This can be split out into just two categories:
- fixes to the RDMA R/W API in regards to SG list length limits
(about 5 patches)
- fixes/features for the Intel hfi1 driver (everything else)
The hfi1 driver is still being brought to full feature support by
Intel, and they have a lot of people working on it, so that amounts to
almost the entirety of this pull request"
* tag 'for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (84 commits)
IB/hfi1: Add cache evict LRU list
IB/hfi1: Fix memory leak during unexpected shutdown
IB/hfi1: Remove unneeded mm argument in remove function
IB/hfi1: Consistently call ops->remove outside spinlock
IB/hfi1: Use evict mmu rb operation
IB/hfi1: Add evict operation to the mmu rb handler
IB/hfi1: Fix TID caching actions
IB/hfi1: Make the cache handler own its rb tree root
IB/hfi1: Make use of mm consistent
IB/hfi1: Fix user SDMA racy user request claim
IB/hfi1: Fix error condition that needs to clean up
IB/hfi1: Release node on insert failure
IB/hfi1: Validate SDMA user iovector count
IB/hfi1: Validate SDMA user request index
IB/hfi1: Use the same capability state for all shared contexts
IB/hfi1: Prevent null pointer dereference
IB/hfi1: Rename TID mmu_rb_* functions
IB/hfi1: Remove unneeded empty check in hfi1_mmu_rb_unregister()
IB/hfi1: Restructure hfi1_file_open
IB/hfi1: Make iovec loop index easy to understand
...
- Updates/fixes for iw_cxgb4 driver
- Updates/fixes for mlx5 driver
- Add flow steering and RSS API
- Add hardware stats to mlx4 and mlx5 drivers
- Add firmware version API for RDMA driver use
- Add the rxe driver (this is a software RoCE driver that makes any
Ethernet device a RoCE device)
- Fixes for i40iw driver
- Support for send only multicast joins in the cma layer
- Other minor fixes
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=oZOO
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Pull base rdma updates from Doug Ledford:
"Round one of 4.8 code: while this is mostly normal, there is a new
driver in here (the driver was hosted outside the kernel for several
years and is actually a fairly mature and well coded driver). It
amounts to 13,000 of the 16,000 lines of added code in here.
Summary:
- Updates/fixes for iw_cxgb4 driver
- Updates/fixes for mlx5 driver
- Add flow steering and RSS API
- Add hardware stats to mlx4 and mlx5 drivers
- Add firmware version API for RDMA driver use
- Add the rxe driver (this is a software RoCE driver that makes any
Ethernet device a RoCE device)
- Fixes for i40iw driver
- Support for send only multicast joins in the cma layer
- Other minor fixes"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (72 commits)
Soft RoCE driver
IB/core: Support for CMA multicast join flags
IB/sa: Add cached attribute containing SM information to SA port
IB/uverbs: Fix race between uverbs_close and remove_one
IB/mthca: Clean up error unwind flow in mthca_reset()
IB/mthca: NULL arg to pci_dev_put is OK
IB/hfi1: NULL arg to sc_return_credits is OK
IB/mlx4: Add diagnostic hardware counters
net/mlx4: Query performance and diagnostics counters
net/mlx4: Add diagnostic counters capability bit
Use smaller 512 byte messages for portmapper messages
IB/ipoib: Report SG feature regardless of HW UD CSUM capability
IB/mlx4: Don't use GFP_ATOMIC for CQ resize struct
IB/hfi1: Disable by default
IB/rdmavt: Disable by default
IB/mlx5: Fix port counter ID association to QP offset
IB/mlx5: Fix iteration overrun in GSI qps
i40iw: Add NULL check for puda buffer
i40iw: Change dup_ack_thresh to u8
i40iw: Remove unnecessary check for moving CQ head
...
Soft RoCE (RXE) - The software RoCE driver
ib_rxe implements the RDMA transport and registers to the RDMA core
device as a kernel verbs provider. It also implements the packet IO
layer. On the other hand ib_rxe registers to the Linux netdev stack
as a udp encapsulating protocol, in that case RDMA, for sending and
receiving packets over any Ethernet device. This yields a RDMA
transport over the UDP/Ethernet network layer forming a RoCEv2
compatible device.
The configuration procedure of the Soft RoCE drivers requires
binding to any existing Ethernet network device. This is done with
/sys interface.
A userspace Soft RoCE library (librxe) provides user applications
the ability to run with Soft RoCE devices. The use of rxe verbs ins
user space requires the inclusion of librxe as a device specifics
plug-in to libibverbs. librxe is packaged separately.
Architecture:
+-----------------------------------------------------------+
| Application |
+-----------------------------------------------------------+
+-----------------------------------+
| libibverbs |
User +-----------------------------------+
+----------------+ +----------------+
| librxe | | HW RoCE lib |
+----------------+ +----------------+
+---------------------------------------------------------------+
+--------------+ +------------+
| Sockets | | RDMA ULP |
+--------------+ +------------+
+--------------+ +---------------------+
| TCP/IP | | ib_core |
+--------------+ +---------------------+
+------------+ +----------------+
Kernel | ib_rxe | | HW RoCE driver |
+------------+ +----------------+
+------------------------------------+
| NIC driver |
+------------------------------------+
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+-----------------------------------------------------------+
| Application |
+-----------------------------------------------------------+
+-----------------------------------+
| libibverbs |
User +-----------------------------------+
+----------------+ +----------------+
| librxe | | HW RoCE lib |
+----------------+ +----------------+
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+--------------+ +------------+
| Sockets | | RDMA ULP |
+--------------+ +------------+
+--------------+ +---------------------+
| TCP/IP | | ib_core |
+--------------+ +---------------------+
+------------+ +----------------+
Kernel | ib_rxe | | HW RoCE driver |
+------------+ +----------------+
+------------------------------------+
| NIC driver |
+------------------------------------+
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Soft RoCE resources:
[1[ https://github.com/SoftRoCE/librxe-dev librxe - source code in
Github
[2] https://github.com/SoftRoCE/rxe-dev/wiki/rxe-dev:-Home - Soft RoCE
Wiki page
[3] https://github.com/SoftRoCE/librxe-dev - Soft RoCE userspace library
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Reviewed-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
There is a strict policy in the Linux kernel that new drivers must be
disabled by default. Hence leave out the "default m" line from Kconfig.
Fixes: 0194621b22 ("IB/rdmavt: Create module framework and handle driver registration")
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Jubin John <jubin.john@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: <stable@vger.kernel.org> # v4.6+
Signed-off-by: Doug Ledford <dledford@redhat.com>
The use of the specific opcode test is redundant since
all ack entry users correctly manipulate the mr pointer
to selectively trigger the reference clearing.
The overly specific test hinders the use of implementation
specific operations.
The change needs to get rid of the union to insure that
an atomic value is not seen as an MR pointer.
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Hanging has been observed while writing a file over NFSoRDMA. Dmesg on
the server contains messages like these:
[ 931.992501] svcrdma: Error -22 posting RDMA_READ
[ 952.076879] svcrdma: Error -22 posting RDMA_READ
[ 982.154127] svcrdma: Error -22 posting RDMA_READ
[ 1012.235884] svcrdma: Error -22 posting RDMA_READ
[ 1042.319194] svcrdma: Error -22 posting RDMA_READ
Here is why:
With the base memory management extension enabled, FRMR is used instead
of FMR. The xprtrdma server issues each RDMA read request as the following
bundle:
(1)IB_WR_REG_MR, signaled;
(2)IB_WR_RDMA_READ, signaled;
(3)IB_WR_LOCAL_INV, signaled & fencing.
These requests are signaled. In order to generate completion, the fast
register work request is processed by the hfi1 send engine after being
posted to the work queue, and the corresponding lkey is not valid until
the request is processed. However, the rdmavt driver validates lkey when
the RDMA read request is posted and thus it fails immediately with error
-EINVAL (-22).
This patch changes the work flow of local operations (fast register and
local invalidate) so that fast register work requests are always
processed immediately to ensure that the corresponding lkey is valid
when subsequent work requests are posted. Local invalidate requests are
processed immediately if fencing is not required and no previous local
invalidate request is pending.
To allow completion generation for signaled local operations that have
been processed before posting to the work queue, an internal send flag
RVT_SEND_COMPLETION_ONLY is added. The hfi1 send engine checks this flag
and only generates completion for such requests.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This fix allows for support of in-kernel reserved operations
without impacting the ULP user.
The low level driver can register a non-zero value which
will be transparently added to the send queue size and hidden
from the ULP in every respect.
ULP post sends will never see a full queue due to a reserved
post send and reserved operations will never exceed that
registered value.
The s_avail will continue to track the ULP swqe availability
and the difference between the reserved value and the reserved
in use will track reserved availabity.
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Some work requests are local operations, such as IB_WR_REG_MR and
IB_WR_LOCAL_INV. They differ from non-local operations in that:
(1) Local operations can be processed immediately without being posted
to the send queue if neither fencing nor completion generation is needed.
However, to ensure correct ordering, once a local operation is posted to
the work queue due to fencing or completion requiement, all subsequent
local operations must also be posted to the work queue until all the
local operations on the work queue have completed.
(2) Local operations don't send packets over the wire and thus don't
need (and shouldn't update) the packet sequence numbers.
Define a new a flag bit for the post send table to identify local
operations.
Add a new field to the QP structure to track the number of local
operations on the send queue to determine if direct processing of new
local operations should be enabled/disabled.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
In order to support extended memory management, add the mechanism to
invalidate MR keys. This includes a flag "lkey_invalid" in the MR data
structure that is to be checked when validating access to the MR via
the associated key, and two utility functions to perform fast memory
registration and memory key invalidate operations.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This implements the device specific function needed by the verbs
API function ib_map_mr_sg().
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Change rvt_post_one_wr to use the new table mechanism for
post send.
Validate that each low level driver specifies the table.
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add flexibility for driver dependent operations in post send
because different drivers will have differing post send
operation support.
This includes data structure definitions to support a table
driven scheme along with the necessary validation routine
using the new table.
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Jianxin Xiong <jianxin.xiong@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The current drivers return errors from this calldown
wrapped in an ERR_PTR().
The rdmavt code incorrectly tests for NULL.
The code is fixed to use IS_ERR() and change ret according
to the driver return value.
Cc: Stable <stable@vger.kernel.org> # 4.6+
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Since rvt_reset_qp already zero's out qp->s_ack_queue head and tail
pointers, there is no need to zero out qp->s_ack_queue itself.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Correct calculation of the low order bits which should be unset
based on use of qos_shift parameter when assigning QPN.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Functions required for MODIFY_PORT were incorrectly being
required for MODIFY_QP.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
This patch avoids that sparse reports the following warning:
rdmavt/qp.c:507:17: warning: context imbalance in 'rvt_reset_qp' - unexpected unlock
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
rdmavt allows the driver to specify the size of the ack queue, but
only uses it for the modify QP limit testing for setting the atomic
limit value.
The driver dependent size is now used to size the s_ack_queue ring
dynamicially.
Since the driver knows its size, the driver will use its define
for any ring size dependent code.
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Use kzalloc_node instead of kzalloc for rdmavt memory region segment
allocation to optimize for performance on NUMA platforms.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The usage of the various vmalloc APIs do not consistently zero memory
when allocating the swqe. Insure zeroing variants are used.
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The priority of the send engines is higher than the CQ completion
thread potentially causing completions to be starved for very
fast interfaces.
Change the CQ kthread to match the send engine threads to minimize
this delay for ULP completion processing.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The awkward coding for setting the allowed_ops field
was tripping an smatch warning.
This patch uses the more appropriate defines from include/rdma
to avoid the issue.
As part of the patch remove a mask that was duplicated
in rdmavt include files and use that mask as appropriate.
Fixes: 8bea6b1cfe6f ("IB/rdmavt: Add create queue pair functionality")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
rdi->ports has memory allocated in rvt_alloc_device(), but does not get
freed because the hfi1 and qib drivers drivers call ib_dealloc_device()
directly instead of going through rdmavt. Add a rvt_dealloc_device()
that frees rdi->ports and then calls ib_dealloc_device(). Switch hfi1
and qib drivers to calling rvt_dealloc_device() instead of
ib_dealloc_device() directly.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Brian Welty <brian.welty@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
call_send is used to determine whether to send immediately or schedule
a send for later. The current logic in rdmavt is inverted and has a
negative impact on the latency of the hfi1 and qib drivers. Fix this
regression by correctly calling send immediately when call_send is set.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Accordingly IB Spec post WR to receive queue must
complete with error if QP is in Error state.
Please refer to C10-42, C10-97.2.1
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Alex Estrin <alex.estrin@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Tracking user/QP ownership is needed to debug issues with
user ULPs like OpenMPI.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The lkey_table_size driver specific parameter value is used before its
value is sanity checked and restricted to RVT_MAX_LKEY_TABLE_BITS.
This causes a vmalloc allocation failure for large values. Fix this
by moving the value check before the first usage of the value.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The new check routine causes a larger than supported frame size
on s390.
Changing the check routine to noinline fixes the issue.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
The non-rdamvt versions of qib and hfi1 allow for a differing
heuristic to override a schedule progress in favor of a direct
call the the progress routine.
This patch adds that for both drivers and rdmavt.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Remove exported functions which are no longer required as the
functionality has moved into rdmavt. This also requires re-ordering some
of the functions since their prototype no longer appears in a header
file. Rather than add forward declarations it is just cleaner to
re-order some of the functions.
Reviewed-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
While hfi1 and qib were still supporting bits and pieces of core verbs
components there needed to be a way to convey if rdmavt should handle
allocation and initialize of resources like the queue pair table. Now
that all of this is moved into rdmavt there is no need for these flags.
They are no longer used in the drivers.
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
For each verb validate that all requirements for driver callbacks are met.
If a function is called without checking for a valid pointer, it is a
required function. Also document what each callback function does.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Add, remove, and otherwise clean up existing comments that are leftover
from the initial code postings of rdmavt. Many of the comments were added
to provide an idea on the direction we were thinking of going. Now that the
design is solidified make a pass over and clean everything up. Also add
details where lacking.
Ensure all non static functions have nano comments.
Reviewed-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
These trace and error print statements would help in debugging issues which
are caused due to messed up QP ring buffer pointers.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
