The test already checked ENOMEM handling, but it only used bdevs that
support chaining (crypto, malloc), so bdev layer didn't need to execute
any accel operations. So, to force bdev layer to do that, a passthru
bdev was added, as it doesn't support chaining.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I322a65ccebb0f144c759692fff285cfd44bbab4b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17766
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
spdk_bdev_queue_io_wait() can only be used when one of bdev submission
functions returns ENOMEM (i.e. there are no more spdk_bdev_ios on that
IO channel). Using it in any other case, e.g. on spdk_accel_append_*()
returning ENOMEM, will most likely result in failure. Therefore, to
avoid that, the IOs are completed with NOMEM status relying on the bdev
layer to retry them.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: Ie0f03496e5d3180c481815b3f1b021e74ae2f46d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17319
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
This actually allows malloc bdev to chain multiple accel operations
together. And, since the last operation will always be a copy, accel
should remove that copy by modifying previous operation's dst/src.
On my system, it improved bdevperf performance (single core, qd=4,
bs=128k, bdev_crypto on top of bdev_malloc, crypto_sw):
randread: 5668M/s -> 8201M/s
randwrite: 5148M/s -> 7856M/s
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I5b9173fa70a42ee56f56c496a34037d46d2f420f
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17202
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This only changes the interface bdev_malloc uses for scheduling the copy
to appends, but it won't chain those copies to an existing sequence, as
bdev_malloc doesn't report support for accel sequences yet. That will
be changed in one of the subsequent patches.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I6db2c79b15cb96a1b07c6cf5514004c76b9d2e92
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17198
Community-CI: Mellanox Build Bot
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
This test sends several read/write requests and verifies the expected
number of accel operations have been executed.
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: Idda46ef00dc5bcc0a176d3dfb39f3f3861964741
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17193
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
