Re: mvsas regression since 3.5

On Sun, 2012-12-23 at 16:26 +0800, Drunkard Zhang wrote:

quoted

Particularly, can you first change

  #define bit(n) ((u64)1 << n)

in drivers/scsi/mvsas/mv_sas.h back to

  #define bit(n) ((u32)1 << n)

and see if it works for you?  Thanks.

Thanks for reply.

This commit just applied in v3.8-rc1, so can't be it. I'm testing
related commits too...

There are only three changes to mvsas between 3.5 and 3.7:

95ab000 [SCSI] mvsas: Fix oops when ata commond timeout.
cca8501 [SCSI] mvsas: remove unused variable in mvs_task_exec()
f0bf750 [SCSI] libsas: trim sas_task of slow path infrastructure

And of those, I can only see f0bf750 being significant, if you could
revert it?

After that, it's probably a bug introduced into libsas, which has a bit
more of a complicated history from 3.5 to 3.7:

8d8e7d1 [SCSI] libsas, ipr: cleanup ata_host flags initialization via ata_host_init
303694e [SCSI] libsas: suspend / resume support
2955b47 [SCSI] async: introduce 'async_domain' type
f0bf750 [SCSI] libsas: trim sas_task of slow path infrastructure
a494fd5 [SCSI] libsas: drop sata port multiplier infrastructure
b17caa1 [SCSI] libsas: fix sas_discover_devices return code handling
26f2f19 [SCSI] libsas: continue revalidation
b2311a2 [SCSI] libsas: sas_rediscover_dev did not look at the SMP exec status.
e7db822 [SCSI] libsas: use ->lldd_I_T_nexus_reset for ->eh_bus_reset_handler
9524c68 [SCSI] libsas: add sas_eh_abort_handler
5db45bd [SCSI] libsas: enforce eh strategy handlers only in eh context
36fed49 [SCSI] libsas: cleanup spurious calls to scsi_schedule_eh
e4a9c37 [SCSI] libata, libsas: introduce sched_eh and end_eh port ops

So I'd try a bisection rather than trying to revert individual commits
in there.

James