test/ublk/ublk.sh 256
will now create 256 ublk devices, to try to test
the ctrlr_cmd queueing logic when the ctrl uring
runs out of sqes.
It's actually difficult to induce the condition,
since the kernel SQPOLL thread can usually keep
up with the SPDK process submitting the control
commands. The best way to induce it is by
*not* stopping the disks at the end of a test
with a lot of disks, and letting ublk_destroy_target
stop all of them at once. Even then, with the
ublk logging enabled, even that extra delay for
each commands opcode is enough to help the
SQPOLL thread to keep up. Commenting out the
first UBLK_DEBUGLOG in ublk_ctrl_cmd() does help
it run out of sqes, at least in my setup.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I8f7a6ca29fd69613d44a89adc7e60563b274d155
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/16458
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
This will help with debugging more complex
ublk configurations - needed especially knowing
that ublk kernel driver is still a bit flaky.
Create a new LOG flag "ublk_io" for the existing
per-IO debug logs, and use the existing "ublk"
flag for ctrl-related debug logs.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ic019c1e837b04dbf5d210c46a98cfbed732278a0
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/16457
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Xiaodong Liu <xiaodong.liu@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
