On Fri, Nov 22, 2019 at 10:36:06PM +0100, Martin Wilck wrote:
> On Fri, 2019-11-22 at 10:14 +0100, Hannes Reinecke wrote:
> > This patchset has the nice benefit that it has fixed the crashes on
> > rmmod we had been seeing.
>
> Well, I investigated two distinct crash-at-rmmod cases, and one was
> already fixed by the earlier commit f45bca8c5052 ("scsi: qla2xxx: Fix
> double scsi_done for abort path"), whereas the other is still present,
> even after applying this series.
>
> Not to say the series is bad - we just shouldn't raise expectations
> too high.
>
Hi Martin,
This patch series only fixes a crash when there's active I/O and ACL of
the initiator is getting deleted. The issue can be reproduced quite
easily:
1. Configure a target with 1 LUN and 1 ACL (and 1 Mapped LUN inside)
2. Run I/O from initiator
3. Delete ACL while running the I/O
The crash happens ~30s after the ACL is deleted when the initiator
starts sending ABORT TASK TMF to abort timed out I/O. It might happen at
rmmod time but that's just coincidence of ABORT TASK being processed. It
might not happen if a rig shuts off in less than 30 seconds.
Thanks,
Roman
