[Intel-gfx] [PATCH 088/190] drm/i915: Move execlists interrupt based submission to a bottom-half

[Chris Wilson]

On Fri, Feb 19, 2016 at 02:10:44PM +0000, Tvrtko Ursulin wrote:
> 
> On 19/02/16 12:29, Chris Wilson wrote:
> >On Fri, Feb 19, 2016 at 12:08:14PM +0000, Tvrtko Ursulin wrote:
> >>
> >>Hi,
> >>
> >>On 11/01/16 10:44, Chris Wilson wrote:
> >>>[  196.988204] clocksource: timekeeping watchdog: Marking clocksource 'tsc' as unstable because the skew is too large:
> >>>[  196.988512] clocksource:                       'refined-jiffies' wd_now: ffff9b48 wd_last: ffff9acb mask: ffffffff
> >>>[  196.988559] clocksource:                       'tsc' cs_now: 4fcfa84354 cs_last: 4f95425e98 mask: ffffffffffffffff
> >>>[  196.992115] clocksource: Switched to clocksource refined-jiffies
> >>>
> >>>Followed by a hard lockup.
> >>
> >>What does the above mean? Irq handler taking too long interferes
> >>with time keeping ?
> >
> >That's exactly what it means, we run for too long in interrupt context
> >(i.e. with interrupts disabled).
> 
> Okay, just please spell it out in the commit.
> 
> >>I like it BTW. Just the commit message needs more work. :)
> >>
> >>How is performance impact with just this patch in isolation? Worth
> >>progressing it on its own?
> >
> >I only looked for regressions, which I didn't find. It fixes a machine
> >freeze/panic, so I wasn't looking for any other reason to justify the
> >patch!
> 
> Then both of the above also need to be documented in the commit message.

Hah, one thing I just rediscovered was that the benchmarks for this kill
the machine without the patch.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre