From: Chris Wilson
> Sent: 15 January 2020 20:53
>
> Since we may try and flush the cachelines associated with large buffers
> (an 8K framebuffer is about 128MiB, even before we try HDR), this leads
> to unacceptably long latencies (when using a voluntary CONFIG_PREEMPT).
> If we call cond_resche
> I'll do some measurements later this afternoon.
This is an Ivy bridge cpu, so clflush (not clflushopt).
With a cond_resched for every page I get:
(Note these calls are every 10 seconds)
# tracer: function_graph
#
# CPU DURATION FUNCTION CALLS
# | | |
From: Chris Wilson
> Sent: 16 January 2020 07:40
> Quoting Daniel Vetter (2020-01-16 06:52:42)
> > On Wed, Jan 15, 2020 at 08:52:45PM +, Chris Wilson wrote:
> > > Since we may try and flush the cachelines associated with large buffers
> > > (an 8K framebuffer is about 128MiB, even before we tr
From: Chris Wilson
> Sent: 16 January 2020 12:29
>
> Quoting David Laight (2020-01-16 12:26:45)
> > However there is a call from __i915_gem_objet_set_pages() that
> > is preceded by a lockdep_assert_held() check - so mustn't sleep.
>
> That is a mutex; it's allowed to sleep. The might_sleep() he
From: David Laight
> Sent: 16 January 2020 14:41
> > I'll do some measurements later this afternoon.
>
> This is an Ivy bridge cpu, so clflush (not clflushopt).
> With a cond_resched for every page I get:
> (Note these calls are every 10 seconds)
For comparison some times booted with the orig
Quoting David Laight (2020-01-16 13:58:44)
> From: Chris Wilson
> > Sent: 16 January 2020 12:29
> >
> > Quoting David Laight (2020-01-16 12:26:45)
> > > However there is a call from __i915_gem_objet_set_pages() that
> > > is preceded by a lockdep_assert_held() check - so mustn't sleep.
> >
> > T
Quoting David Laight (2020-01-16 12:26:45)
> However there is a call from __i915_gem_objet_set_pages() that
> is preceded by a lockdep_assert_held() check - so mustn't sleep.
That is a mutex; it's allowed to sleep. The might_sleep() here should
help assuage your fears.
-Chris
_
Quoting Daniel Vetter (2020-01-16 06:52:42)
> On Wed, Jan 15, 2020 at 08:52:45PM +, Chris Wilson wrote:
> > Since we may try and flush the cachelines associated with large buffers
> > (an 8K framebuffer is about 128MiB, even before we try HDR), this leads
> > to unacceptably long latencies (whe
On Wed, Jan 15, 2020 at 08:52:45PM +, Chris Wilson wrote:
> Since we may try and flush the cachelines associated with large buffers
> (an 8K framebuffer is about 128MiB, even before we try HDR), this leads
> to unacceptably long latencies (when using a voluntary CONFIG_PREEMPT).
> If we call co
