On Thu, Sep 12 2013 at 6:55pm -0400,
Mikulas Patocka <mpatocka@xxxxxxxxxx> wrote:
> There's a race condition (it's not serious):
>
> + if (num_bios > ACCESS_ONCE(peak_rq_based_ios))
> + ACCESS_ONCE(peak_rq_based_ios) = num_bios;
>
> You can double-check it using code like this:
> if (num_bios > ACCESS_ONCE(peak_rq_based_ios)) {
> spin_lock_irqsave(&peak_rq_based_ios_lock, flags);
> if (num_bios > peak_rq_based_ios)
> peak_rq_based_ios = num_bios;
> spin_unlock_irqrestore(&peak_rq_based_ios_lock, flags);
> }
Yeah I thought about using double checked locking but didn't want to
incur the extra locking. Figured precise answer for peak_rq_based_ios
wasn't of utmost importance. But I'll think about doing it.
> Maybe - reset the value peak_rq_based_ios if the user clears
> track_peak_rq_based_ios?
Well until patch 7 there isn't a convenient place to trap the change.
And you already took issue with how I did it in that patch.
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
