On May 19, 2021 / 17:26, Mike Christie wrote:
> This has us use raw_smp_processor_id in iblock's plug_device callout.
> smp_processor_id is not needed here, because we are running from a per CPU
> work item that is also queued to run on a worker thread that is normally
> bound to a specific CPU. If the worker thread did end up switching CPUs
> then it's handled the same way we handle when the work got moved to a
> different CPU's worker thread, where we will just end up sending IO from
> the new CPU.
>
> Fixes: 415ccd9811da ("scsi: target: iblock: Add backend plug/unplug
> callouts")
> Signed-off-by: Mike Christie <michael.christie@xxxxxxxxxx>
Thanks Mike. To confirm this patch, I tried to trigger the BUG message using
block backstore on kernel configured with DEBUG_PREEMPT, but I was not able to
trigger it. Having said that, I think this change prevents the potential BUG
message with the same reasoning as the other fix in target_core_transport [1].
As far as I read through the code, the code should work even when CPU switches.
[1] https://marc.info/?l=linux-scsi&m=162106219905019
--
Best Regards,
Shin'ichiro Kawasaki
