]> git.baikalelectronics.ru Git - kernel.git/commit
ata: libata-scsi: Fix delayed scsi_rescan_device() execution
authorDamien Le Moal <dlemoal@kernel.org>
Tue, 5 Sep 2023 00:06:23 +0000 (09:06 +0900)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 10 Oct 2023 20:00:35 +0000 (22:00 +0200)
commit2f09a09d73cb65533de7fbea7685f36a78c59d29
tree82eaa59727c110b24488d55c3f7d41c0b8af94ec
parentf2b359e3a4f3ef405840446c7a82cd5d1c3dfd19
ata: libata-scsi: Fix delayed scsi_rescan_device() execution

[ Upstream commit 8b4d9469d0b0e553208ee6f62f2807111fde18b9 ]

Commit 6aa0365a3c85 ("ata: libata-scsi: Avoid deadlock on rescan after
device resume") modified ata_scsi_dev_rescan() to check the scsi device
"is_suspended" power field to ensure that the scsi device associated
with an ATA device is fully resumed when scsi_rescan_device() is
executed. However, this fix is problematic as:
1) It relies on a PM internal field that should not be used without PM
   device locking protection.
2) The check for is_suspended and the call to scsi_rescan_device() are
   not atomic and a suspend PM event may be triggered between them,
   casuing scsi_rescan_device() to be called on a suspended device and
   in that function blocking while holding the scsi device lock. This
   would deadlock a following resume operation.
These problems can trigger PM deadlocks on resume, especially with
resume operations triggered quickly after or during suspend operations.
E.g., a simple bash script like:

for (( i=0; i<10; i++ )); do
echo "+2 > /sys/class/rtc/rtc0/wakealarm
echo mem > /sys/power/state
done

that triggers a resume 2 seconds after starting suspending a system can
quickly lead to a PM deadlock preventing the system from correctly
resuming.

Fix this by replacing the check on is_suspended with a check on the
return value given by scsi_rescan_device() as that function will fail if
called against a suspended device. Also make sure rescan tasks already
scheduled are first cancelled before suspending an ata port.

Fixes: 6aa0365a3c85 ("ata: libata-scsi: Avoid deadlock on rescan after device resume")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/ata/libata-core.c
drivers/ata/libata-scsi.c