Make this a transport-level decision instead. TCP and RDMA do want to
abort, but PCIe cannot because these commands may still be receiving DMA
operations from the device.
Change-Id: I305acddc3819c903eb3217e8f710d4216d0b3931
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11509
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Reviewed-by: Michael Haeuptle <michaelhaeuptle@gmail.com>
It simplifies code and removes cast of nvme_qpair
to rdma_qpair
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I363246cf9d8c9cbafd48b26facdb5cc37fdd8e67
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/12701
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
When qpair is attached to a poll group, disconnect
process is async - we are waiting for the DISCONNECTED
event from rdmacm to destroy rdma resources. However
the user (nvme_perf) can destroy qpair immediatelly,
so memory allocated for qpair is freed but rdma
resouces are still allocated. That means that we may
receive rdmacm event (DISCONNECTED) for the destroyed qpair,
that leads to use-after-free.
To fix this problem, add a check for internal qpair state
when qpair is destroyed, if disconnect is not finished, then
we forcefully destroy rdma resources.
Fixes issue #2515
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Reported-by: Or Gerlitz <ogerlitz@nvidia.com>
Change-Id: I7bfa53c9f6fe6ed787323a8941f1f2db17ea0c20
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/12700
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Now controller initialization with RDMA
transport is fully async
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I26e857740d3137d0b0e987facc81fc5f6ef81f2b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10756
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
The RDMA transport can disconnect qpair asynchronously now.
Previously, we tried to release the resource of the qpair after disconnected.
However it did not work because it was done when deleting the qpair.
The admin qpair was not deleted in a ctrlr reset sequence.
This patch tries to satisfy the same aim again but by a different way.
Previously, we released the resource of the qpair before starting
actual disconnection process. This patch release the resource of the
qpair after the qpair is actually disconnected.
The related patches are:
b9518a5540eb09178a59
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: Id6a814895a35b1589b781a91744ef872b42aaa69
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11783
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
The code to handle the lingering qpair when deleting it was really
complicated.
The RDMA transport can connect or disconnect qpair asynchronously.
Then we can include the code to handle the lingering qpair into the
code to disconnect qpair now.
If the disconnected qpair is still busy, defer completion of the
disconnection until qpair becomes idle.
If poll group is not used, we can complete disconnection immediately
because cq is already destroyed.
The related data and unit test cases are not necessary anymore.
So delete them in this patch.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: Ic8f81143fcad0714ac9b7db862313aa8094eeefb
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11778
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Include delayed disconnect/connect retries with finite times into
the state machine of asynchronous qpair connnection.
We do not need to call back to the common transport layer but
we need to do the following, clear rqpair->cq before starting disconnection
if qpair uses poll group, and clear qpair->transport_failure_reason after
disconnected.
Additionally locate the new state STALE_CONN before INITIALIZING
because cq is not ready to use for admin qpair when the state is
STALE_CONN.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: Ibc779a2b772be9506ffd8226d5f64d6d12102ff2
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11690
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
TCP transport already does it but was not documented clearly.
RDMA and PCIe transports follow it and document it clearly.
Then we can check each qpair's state if
spdk_nvme_poll_group_process_completions() returns -ENXIO before
disconnected_qpair_cb() is called.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I2afe920cfd06c374251fccc1c205948fb498dd33
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11328
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Add three states, INITIALIZING, EXITING, and EXITED to the rqpair
state.
Add async parameter to nvme_rdma_ctrlr_create_qpair() and set it
to opts->async_mode for I/O qpair and true for admin qpair.
Replace all nvme_rdma_process_event() calls by
nvme_rdma_process_event_start() calls.
nvme_rdma_ctrlr_connect_qpair() sets rqpair->state to INITIALIZING
when starting to process CM events.
nvme_rdma_ctrlr_connect_qpair_poll() calls
nvme_rdma_process_event_poll() with ctrlr->ctrlr_lock if qpair is
not admin qpair.
nvme_rdma_ctrlr_disconnect_qpair() returns if qpair->async is true
or qpair->poll_group is not NULL before polling CM events, or polls
CM events until completion otherwise. Add comments to clarify why
we do like this.
nvme_rdma_poll_group_process_completions() does not process submission
for any qpair which is still connecting.
Change-Id: Ie04c3408785124f2919eaaba7b2bd68f8da452c9
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11442
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Separate nvme_rdma_process_event() into nvme_rdma_process_event_start()
and nvme_rdma_process_event_poll().
Use nvme_rdma_process_event_start() and nvme_rdma_process_event_poll()
in nvme_rdma_process_event() to ensure compatibility.
Change-Id: Idc960fab2540efec612dcf22f156acabd2e2874e
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10594
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
It will be convenient for the following patches to return
negated errno directly.
Change-Id: Ic80181b2ee449946dd60ad0c97a325fd48b92231
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10990
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
nvme_rdma_poll_events() gets the cm_channel pointer itself.
Before calling nvme_rdma_process_event(), we checks the
rctrlr is valid.
Hence we do not have to pass the cm_channel pointer to
nvme_rdma_process_event() via a parameter.
This simplifies the code and makes the following patches a little easier.
Change-Id: I03f095833469c5b64592264d63a592106d49e13b
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11167
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Replace nvme_fabric_qpair_connect() by nvme_fabric_qpair_connect_async()
and nvme_fabric_qpair_connect_poll().
The following is a detail.
Define state of the nvme_rdma_qpair and each rqpair holds it.
Initialize rqpair->state by INVALID at nvme_rdma_ctrlr_create_qpair().
_nvme_rdma_ctrlr_connect_qpair() sets rqpair->state to
FABRIC_CONNECT_SEND instead of calling nvme_fabric_qpair_connect().
Then the new function nvme_rdma_ctrlr_connect_qpair_poll() calls
nvme_fabric_qpair_connect_async() at FABRIC_CONNECT_SEND and
nvme_fabric_qpair_connect_poll() until it returns 0 at FABRIC_CONNECT_POLL.
nvme_rdma_qpair_process_completions() or
nvme_rdma_poll_group_process_completions() calls
nvme_rdma_ctrlr_connect_qpair_poll() if qpair->state is CONECTING.
This patter follows the TCP transport.
Change-Id: I411f4fa8071cb5ea27581f3820eba9b02c731e4c
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11334
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
responder_resources parameter of rdma cm tells remote
side how many outstaing RDMA_READ of atomic operations
local side can handle.
Previously it was adjusted on queue depth but that was
not correct since these parameters do not depend on
each other. Even with qdepth=1 remote side may send
several RDMA_READ operations per 1 IO request.
With this change we report responder_resources
equal to the maximum supported by RDMA device.
Linux kernel nvme rdma driver reports this value
in the same way.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I77e5c2ead6269da44c32a75a9188429f50d32ae4
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11698
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The feature will be redesigned and restored in the following patches.
For the NVMe bdev module, it can reconnect by itself without relying
on the feature.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I2d9c0437f7ad8412ad8cf40d11e574723b735bee
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11440
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
For nvme_rdma_qpair_process_completions(), consolidate the operations to
call nvme_rdma_fail_qpair() and return -ENXIO into a single place.
Besides, shorten pointer references for
nvme_rdma_qpair_process_completions() and
nvme_rdma_poll_group_process_completions().
These will make the following patches a little easier.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: Iaf72cfca0b5b3ba223d86e267da8069d43a15292
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11439
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
This is a preparation to make nvme_transport_ctrlr_disconnect_qpair()
asynchronous.
For nvme_transport_ctrlr_disconnect_qpair(), factor out operations after
returning from transport's specific ctrlr_disconnect_qpair() into a helper
function nvme_transport_ctrlr_disconnect_qpair_done().
Then move nvme_transport_ctrlr_disconnect_qpair_done() into the end of
the transport specific ctrlr_disconnect_qpair().
Additionally remove the operation to overwrite the qpair state to
DISCONNECTED from nvme_transport_connect_qpair_fail() because
this is duplicated and nvme_transport_ctrlr_disconnect_qpair() is responsible
to make the qpair disconnected even after it completes asynchronously.
Change-Id: I9c8faa7039d306d3e31a8f51826755ce8840a8aa
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10851
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
We want to call disconnected_qpairs_cb only if qpair is actually
disconnected. When we disconnect qpair asynchronously, for qpairs in
the group->disconnected_qpairs list, we want to poll them until
actually disconnected and then call disconnected_qpairs_cb for them.
As a preparation, call disconnected_qpair_cb only for qpairs which is
in the group->disconnected_qpairs list.
For TCP and PCIe transports, disconnecting qpair will continue to be
synchronous for now.
So we change only RDMA transport.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: Ifaf6157e1e02fa13f52a66409c9e60fc814d71dd
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11495
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
According to NVMe over Fabrics spec number of SGLs supported by the
controller is reported in MSDBD. But it is also implicitly limited by
command capsule size (IOCCSZ) since SGL are passed in capsule.
This patch adjusts max_sges to capsule size if required. Adjustment to
MSDBD is also moved to transport layer because it is fabrics specific
parameter and is not valid for PCIe transport.
Signed-off-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Change-Id: I44918eb949345c61242ca50a524d21d04b6ac058
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11669
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The patch
nvme: Set dnr to zero for nvme_qpair_abort_reqs()
1b3172f726
did the change stated in the title.
However,
Revert "nvme/rdma: Correct qpair disconnect process"
c8f986c7ee
destroyed it for RDMA transport.
Additionally, we had still set DNR to 1 in nvme_qpair_init().
This patch fixes both.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: Iee60ac24aa7e04cce0f394014c9d9afc9d2b56ec
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11644
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
SPDK can submit more commands to remote NVMf target than allowed by
negotiated queue size. SPDK submits up to SQSIZE commands, but only
SQSIZE-1 are allowed.
Here is a relevant quote from NVMe over Fabrics rev.1.1a ch.2.4.1
“Submission Queue Flow Control Negotiation”:
If SQ flow control is disabled, then the host should limit the number
of outstanding commands for a queue pair to be less than the size of
the Submission Queue. If the controller detects that the number of
outstanding commands for a queue pair is greater than or equal to the
size of the Submission Queue, then the controller shall:
a) stop processing commands and set the Controller Fatal
Status (CSTS.CFS) bit to ‘1’ (refer to section 10.5 in the NVMe Base
specification); and
b) terminate the NVMe Transport connection and end the association
between the host and the controller.
Signed-off-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Change-Id: Ifbcf5d51911fc4ddcea1f7cde3135571648606f3
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11413
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
According to NVMe over Fabrics specification (rev.1.1a) HSQSIZE sent
in RDMA_CM_REQUEST private data (ch.7.3.6.4) shall be the same as
SQSIZE later sent in Connect command (ch.3.3).
SPDK NVMe RDMA initiator adjusts SQSIZE to CRQSIZE received from
target in RDMA_CM_ACCEPT private data. Target is allowed to send
CRQSIZE < HSQSIZE if RNR retries are used. So, it is possible that
SQSIZE sent by SPDK will be lower than previously sent HSQSIZE. There
are targets validating this match and they reject connection from
SPDK.
Linux kernel NVMe initiator doesn't perform such adjustments and
connects well to such targets.
This patch aligns SPDK behavior with specification and Linux kernel
implementation.
Signed-off-by: Evgeniy Kochetov <evgeniik@nvidia.com>
Change-Id: I01968d1c07d284396fa5939932d85841351d7a45
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11350
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
There are errors occur that uninitialised value created by a stack allocation when running unittest_accel and unittest_nvme_rdma with valgrind.
Change-Id: I4b48b472cc7c189cbcaf8ca772830a23118e7e17
Signed-off-by: Jaylyn Ren <jaylyn.ren@arm.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10559
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Xiaodong Liu <xiaodong.liu@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
We saw this unexpected behavior by the current SPDK master.
Add the check to clarify this behavior occurs only when we use
Soft RoCE.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I3a5eaa9064a0601c65139e7868898545926d0dbf
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/11225
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Dong Yi <dongx.yi@intel.com>
This reverts commit eb09178a59.
Reason for revert:
This caused a degradation for adminq.
For adminq, ctrlr_delete_io_qpair() is not called until ctrlr is destructed.
So necessary delete operations are not done for adminq.
Reverting the patch is practical for now.
Change-Id: Ib55ff81dfe97ee1e2c83876912e851c61f20e354
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10878
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
nvme_poll_group_disconnect_qpair() is called only by a single place now.
We do not need the flag poll_group_disconnect_in_progress any more.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I8f9c0f14baa8fcb9b0637635a5bb3d34a8b11af5
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10673
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
spdk_nvme_poll_group_remove() is available only for disconnected
qpairs now. Hence spdk_nvme_poll_group_remove() does not have to
check if qpair is connected and call nvme_ctrlr_disconnect_qpair().
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I3b05246c4be6adfa3392b8f0e5ecaf274a8a7795
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10846
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Monica Kenguva <monica.kenguva@intel.com>
This is necessary to failover another path when multipath is configured.
Signed-off-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Change-Id: I0b6bcf63501e38f75efb4b0d6bec58abb4b67aef
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10250
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Monica Kenguva <monica.kenguva@intel.com>
We recently improved qpair disconnect process and added assert
if we get a completion without any error when a qpair is disconnected.
However unexpectedly we saw this case very often when we ran the test
test/nvmf/host/multipath.sh for the real hardware in the test pool.
So we remove the assert and change the ERRLOG to INFOLOG.
Fixes one of the issues in #2300
Signed-off-by: Shuhei Matsumoto <shuheimatsumoto@gmail.com>
Change-Id: Iedbf7e0afa5025da6a810043ba95348ba5b856b3
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10901
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
In current implementation RDMA qpair is destroyed right after
disconnect. That is not graceful qpair shutdown process since
there can be requests submitted to HW and we may receive
completions for already destroyed/freed qpair.
To avoid this, only disconnect qpair in ctrlr_disconnect_qpair
transport callback, all other resources will be released in
ctrlr_delete_io_qpair cb.
This patch is useful when nvme poll groups are used since in
that case we use shared CQ, if the disconnected qpair has WRs
submitted to HW then qpair's destruction will be deferred to
poll group.
When nvme poll groups are not used, this patch doesn't change
anything, in that case destruction flow is still ungraceful.
However since CQ is destroyed immediately after qpair,
we shouldn't receive any requests which point to released
resources. A correct solution for non-poll group case
requires async diconnect API which may lead to significant
rework.
There is a bug when Soft Roce is used - we may receive
a completion with "normal" status when qpair is already
disconnected and all nvme requests are aborted. Added
a workaround for it.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I0680d9ef9aaa8737d7a6d1454cd70a384bb8efac
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10327
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Reviewed-by: Shuhei Matsumoto <shuheimatsumoto@gmail.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
spdk_nvme_ctrlr_free_io_qpair can be called when
qpair is already disconnected. In that case qpair's
state is changed to NVME_QPAIR_DESTROYING and
transport's ctrlr_delete_io_qpair callback is
called. RDMA and TCP transports call
nvme_transport_ctrlr_disconnect_qpair in
the callback and since qpair's state is
not DISCONNECTED or DISCONNECTING, qpair
is disconnected for the second time.
If spdk_nvme_ctrlr_free_io_qpair is called
when qpair is in ENABLED state than nothing
changes, qpair will be disconnected before destroy.
PCIE/vfio_user don't implement transport disconnect
callback, so they are not affected.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I23e11856ecafb51669acf4a3118be049c11eecda
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10326
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Dong Yi <dongx.yi@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Add a parameter which determines the owner of the
map - target or initiator. It allows to set different
access flags when creating Memory Regions
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I0016847fe116e193d0954db1c8e65066b4ff82bf
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/10283
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
In some cases a single virtually contriguos memory
buffer can be translated to several chunks of memory.
To make such translation possible, update structure
spdk_memory_domain_translation_result to use a pointer
to iovec.
Add a single iov structure or cases where translation
is always 1:1, it will make easier translation callback
implementation. For RDMA transport translation of address
is always 1:1, so treat iovcnt other than 1 as an
error.
Change-Id: I65605575d43a490490eba72c1eb19f3a09d55ec6
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9779
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Instead of a union with domain type specific
parameters, store an opaque pointer to user
context. Depending on the memory domain type,
this context can be cast to a specific struct,
e.g. to spdk_memory_domain_rdma_ctx for RDMA
memory domains.
This change provides more flexibility to
applications to create and manage custom
memory domains
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Change-Id: Ib0a8297de80773d86edc9849beb4cbc693ef5414
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9778
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Allow to return more than one memory domain.
This change aligns bdev and nvme API and provides
more flexibility for custom transports.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: Ica9b12ad8463c361be6cb62ee2c0513eec0b486d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9546
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This type of errors is not fatal and can be observed when
qpairs are diconnected. The same approach is used in target
side.
Change-Id: Ic3c7b1731c0cbd2e98d776f0f0c5d82464b3d556
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9416
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
When poll_group is used, several qpairs share the same CQ
and it is possible to receive a completion with error (e.g.
IBV_WC_WR_FLUSH_ERR) for already disconnected qpair
That happens due to qpair is destroyed while there are
submitted but not completed send/receive Work Requests
To avoid such situation, we should not detroy ibv
qpair until we reap completions for all submitted
send/receive work requests. That requires some
rework in rdma transport and will be implemented
later
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: Idb6213d45c2a7954b9ab280f5eb5e021be00505f
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9056
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Poll group holds lists of qpairs in different states and
when we got rdma completion with error, we iterate these
lists to find a qpair which qp_num matches. qp_num
is stored inside of ibv_qp which belongs to spdk_rdma_qp
structure. When nvme_rdma_qpair is disconnected, pointer
to spdk_rdma_qp is cleaned but qpair may still exist in
poll group list and when we start searhing for qpair by
qp_num we may dereference NULL pointer.
This patch adds a check that pointer to spdk_rdma_qp
is valid before dereferencing it. To minimize boilerplate code,
wrap all check in macro. Add unit test to verify this fix.
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: I1925f93efb633fd5c176323d3bbd3641a1a632a9
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9050
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
These functions accept extendable structure with IO request options.
The options structure contains a memory domain that can be used to
translate or fetch data, metadata pointer and end-to-end data
protection parameters
Change-Id: I65bfba279904e77539348520c3dfac7aadbe80d9
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6270
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Add a global list of memory domains with reference counter.
Memory domains are used by NVME RDMA qpairs.
Also refactor ibv_resize_cq in nvme_rdma_ut.c to stub
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Change-Id: Ie58b7e99fcb2c57c967f5dee0417e74845d9e2d1
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/8127
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Replaced poll cycle count with a timeout when destroying a qpair that is
part of a poll group. Tracking the time instead of a poll count is more
stable, as the number of poll cycles can vary based on the application's
behavior when destroying a qpair.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I7445bc1b411f2905aab7bf3dc7b2d3344712e1eb
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/9200
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
async_mode option is currently supported in PCIe transport layer
to create io qpair asynchronously. User polls the io_qpair for
completions, after create cq and sq completes in order, pqpair
is set to READY state. I/O submitted before the qpair is ready
is queued internally. Currently other transports only support
synchronous io qpair creation.
Signed-off-by: Monica Kenguva <monica.kenguva@intel.com>
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Change-Id: Ib2f9043872bd5602274e2508cf1fe9ff4211cabb
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/8911
Community-CI: Mellanox Build Bot
Community-CI: Broadcom CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
