On Wed, Mar 14, 2018 at 6:57 PM, Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxxxxxxx> wrote: > On 14/03/2018 08:31, Patchwork wrote: > > Pushed it, thanks for the review! > > (And I forgot to copy Arnd on the patch..) > > So Arnd, sorry, I forgot Reported-by does not add Cc from git send-email. > https://patchwork.freedesktop.org/series/39939/ is my version of the fix for > this. (Now merged.) Hopefully it works for your randconfigs as well and > thanks for sending a patch in the first place! The patch looks good to me, thanks for the follow-up Acked-by: Arnd Bergmann <arnd@xxxxxxxx> I didn't test it, but you'll hear from me if it breaks again. One comment about the use of spin_lock_irqsave(), you wrote: "Slight penalty we now pay is an additional irqsave spin lock/unlock cycle on the event enable path. But since enable is not a fast path, that is preferrable to the alternative solution which was doing MMIO under irqsave spinlock." While I don't know about the exact cost on x86, on many architectures spin_lock_irqsave()/spin_unlock_irqrestore() is much more expensive than a plain spin_lock()/spin_unlock() or spin_lock_irq()/spin_unlock_irq() pair that doesn't have to store the disabled-state. I see you already removed the inner irqsave()/irqrestore() pair that is now useless (I failed to notice that in my original patch), but in my experience, it's usually possible to do the same for many others as well after proving that a function is always called with IRQs enabled or always disabled. Arnd _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
