Skip to content

scsi: qla2xxx: Fix hang during NVMe session tear down

John Meneghini requested to merge johnmeneghini/centos-stream-9:bz_2053113 into main

Bugzilla: http://bugzilla.redhat.com/2053113

The following hung task call trace was seen:

    [ 1230.183294] INFO: task qla2xxx_wq:523 blocked for more than 120 seconds.
    [ 1230.197749] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
    [ 1230.205585] qla2xxx_wq      D    0   523      2 0x80004000
    [ 1230.205636] Workqueue: qla2xxx_wq qlt_free_session_done [qla2xxx]
    [ 1230.205639] Call Trace:
    [ 1230.208100]  __schedule+0x2c4/0x700
    [ 1230.211607]  schedule+0x38/0xa0
    [ 1230.214769]  schedule_timeout+0x246/0x2f0
    [ 1230.222651]  wait_for_completion+0x97/0x100
    [ 1230.226921]  qlt_free_session_done+0x6a0/0x6f0 [qla2xxx]
    [ 1230.232254]  process_one_work+0x1a7/0x360

...when device side port resets were done.

Abort threads were getting out without processing due to the "deleted"
flag check. The delete thread, meanwhile, could not proceed with a
logout (that would have cleared out pending requests) as the logout IOCB
work was not progressing. It appears like the hung qlt_free_session_done()
thread is causing the ha->wq works on hold. The qlt_free_session_done()
was hung waiting for nvme_fc_unregister_remoteport() + localport_delete cb
to be complete, which would only happen when all I/Os are released.

Fix this by allowing abort to progress until device delete is completely
done. This should make the qlt_free_session_done() proceed without hang and
thus clear up the deadlock.

Link: https://lore.kernel.org/r/20210817051315.2477-5-njavali@marvell.com
Signed-off-by: Arun Easi <aeasi@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
(cherry picked from commit 310e69edfbd57995868a428eeddea09a7b5d2749)

Signed-off-by: John Meneghini jmeneghi@redhat.com

Merge request reports