Re: [RFC][PATCH 2/2] efi_pstore: Introducing workqueue updating sysfs entries

[Mike Waychison <mikew@xxxxxxxxxx>]

On Fri, Aug 17, 2012 at 2:15 PM, Seiji Aguchi <seiji.aguchi@xxxxxxx> wrote:
>
>> I'm not a fan of creating a periodic timer that wakes up here to check for an event that should be considered very rare.
>>
>> Can this just become scheduled work?  Scheduling work itself is a very lightweight process and should be relatively safe to do from a
>> pstore write.
>
> I agree that the periodic timer is heavy a bit.
> But I would like to keep a write callback simple as much as possible in panic situation.
> For example, I'm concerned that efi_pstore hangs up due to some spin_locks related workqueue like gcwq->lock.
>
> Also,  a situation which this workqueue is needed is just oops case because, system will be down and users can't access to sysfs files
>  in other cases, panic, reboot and emergency_restart.
>
> So, Can I call schedule_work in oops case only as follows?
>
> efi_pstore_write()
> {
>    <write log to NVRAM>
>
>  if (reason != KMSG_DUMP_OOPS)
>         return;
>
> schedule_work()

You could, but why not always just schedule_work()?  If we are hosed
by broken workqueue/scheduler locking, the user isn't going to see
those files in sysfs either way :)  If you are going to insist that we
shouldn't schedule_work() in the other cases, I'd prefer:

/* The user may want to see an entry for this write in sysfs. */
if (reason == KMSG_DUMP_OOPS)
        schedule_work(...);

>
> }
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/