public inbox for intel-gfx@lists.freedesktop.org
 help / color / mirror / Atom feed
From: Matthew Brost <matthew.brost@intel.com>
To: John Harrison <john.c.harrison@intel.com>
Cc: intel-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org
Subject: Re: [Intel-gfx] [PATCH 4/7] drm/i915/guc: Don't hog IRQs when destroying contexts
Date: Fri, 10 Dec 2021 17:10:50 -0800	[thread overview]
Message-ID: <20211211011049.GA8660@jons-linux-dev-box> (raw)
In-Reply-To: <ec870417-3894-0bb2-6561-722b8345be6f@intel.com>

On Fri, Dec 10, 2021 at 05:07:12PM -0800, John Harrison wrote:
> On 12/10/2021 16:56, Matthew Brost wrote:
> > From: John Harrison <John.C.Harrison@Intel.com>
> > 
> > While attempting to debug a CT deadlock issue in various CI failures
> > (most easily reproduced with gem_ctx_create/basic-files), I was seeing
> > CPU deadlock errors being reported. This were because the context
> > destroy loop was blocking waiting on H2G space from inside an IRQ
> > spinlock. There was deadlock as such, it's just that the H2G queue was
> There was *no* deadlock as such
> 

Let's fix this up when applying the series.

With that:
Reviewed-by: Matthew Brost <matthew.brost@intel.com>

> John.
> 
> > full of context destroy commands and GuC was taking a long time to
> > process them. However, the kernel was seeing the large amount of time
> > spent inside the IRQ lock as a dead CPU. Various Bad Things(tm) would
> > then happen (heartbeat failures, CT deadlock errors, outstanding H2G
> > WARNs, etc.).
> > 
> > Re-working the loop to only acquire the spinlock around the list
> > management (which is all it is meant to protect) rather than the
> > entire destroy operation seems to fix all the above issues.
> > 
> > Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
> > ---
> >   .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 45 ++++++++++++-------
> >   1 file changed, 28 insertions(+), 17 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> > index 36c2965db49b..96fcf869e3ff 100644
> > --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> > +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
> > @@ -2644,7 +2644,6 @@ static inline void guc_lrc_desc_unpin(struct intel_context *ce)
> >   	unsigned long flags;
> >   	bool disabled;
> > -	lockdep_assert_held(&guc->submission_state.lock);
> >   	GEM_BUG_ON(!intel_gt_pm_is_awake(gt));
> >   	GEM_BUG_ON(!lrc_desc_registered(guc, ce->guc_id.id));
> >   	GEM_BUG_ON(ce != __get_context(guc, ce->guc_id.id));
> > @@ -2660,7 +2659,7 @@ static inline void guc_lrc_desc_unpin(struct intel_context *ce)
> >   	}
> >   	spin_unlock_irqrestore(&ce->guc_state.lock, flags);
> >   	if (unlikely(disabled)) {
> > -		__release_guc_id(guc, ce);
> > +		release_guc_id(guc, ce);
> >   		__guc_context_destroy(ce);
> >   		return;
> >   	}
> > @@ -2694,36 +2693,48 @@ static void __guc_context_destroy(struct intel_context *ce)
> >   static void guc_flush_destroyed_contexts(struct intel_guc *guc)
> >   {
> > -	struct intel_context *ce, *cn;
> > +	struct intel_context *ce;
> >   	unsigned long flags;
> >   	GEM_BUG_ON(!submission_disabled(guc) &&
> >   		   guc_submission_initialized(guc));
> > -	spin_lock_irqsave(&guc->submission_state.lock, flags);
> > -	list_for_each_entry_safe(ce, cn,
> > -				 &guc->submission_state.destroyed_contexts,
> > -				 destroyed_link) {
> > -		list_del_init(&ce->destroyed_link);
> > -		__release_guc_id(guc, ce);
> > +	while (!list_empty(&guc->submission_state.destroyed_contexts)) {
> > +		spin_lock_irqsave(&guc->submission_state.lock, flags);
> > +		ce = list_first_entry_or_null(&guc->submission_state.destroyed_contexts,
> > +					      struct intel_context,
> > +					      destroyed_link);
> > +		if (ce)
> > +			list_del_init(&ce->destroyed_link);
> > +		spin_unlock_irqrestore(&guc->submission_state.lock, flags);
> > +
> > +		if (!ce)
> > +			break;
> > +
> > +		release_guc_id(guc, ce);
> >   		__guc_context_destroy(ce);
> >   	}
> > -	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
> >   }
> >   static void deregister_destroyed_contexts(struct intel_guc *guc)
> >   {
> > -	struct intel_context *ce, *cn;
> > +	struct intel_context *ce;
> >   	unsigned long flags;
> > -	spin_lock_irqsave(&guc->submission_state.lock, flags);
> > -	list_for_each_entry_safe(ce, cn,
> > -				 &guc->submission_state.destroyed_contexts,
> > -				 destroyed_link) {
> > -		list_del_init(&ce->destroyed_link);
> > +	while (!list_empty(&guc->submission_state.destroyed_contexts)) {
> > +		spin_lock_irqsave(&guc->submission_state.lock, flags);
> > +		ce = list_first_entry_or_null(&guc->submission_state.destroyed_contexts,
> > +					      struct intel_context,
> > +					      destroyed_link);
> > +		if (ce)
> > +			list_del_init(&ce->destroyed_link);
> > +		spin_unlock_irqrestore(&guc->submission_state.lock, flags);
> > +
> > +		if (!ce)
> > +			break;
> > +
> >   		guc_lrc_desc_unpin(ce);
> >   	}
> > -	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
> >   }
> >   static void destroyed_worker_func(struct work_struct *w)
> 

  reply	other threads:[~2021-12-11  1:16 UTC|newest]

Thread overview: 27+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-12-11  0:56 [Intel-gfx] [PATCH 0/7] Fix stealing guc_ids + test Matthew Brost
2021-12-11  0:56 ` [Intel-gfx] [PATCH 1/7] drm/i915/guc: Use correct context lock when callig clr_context_registered Matthew Brost
2021-12-11  0:56 ` [Intel-gfx] [PATCH 2/7] drm/i915/guc: Only assign guc_id.id when stealing guc_id Matthew Brost
2021-12-11  1:08   ` John Harrison
2021-12-11  0:56 ` [Intel-gfx] [PATCH 3/7] drm/i915/guc: Remove racey GEM_BUG_ON Matthew Brost
2021-12-11  0:56 ` [Intel-gfx] [PATCH 4/7] drm/i915/guc: Don't hog IRQs when destroying contexts Matthew Brost
2021-12-11  1:07   ` John Harrison
2021-12-11  1:10     ` Matthew Brost [this message]
2021-12-11  0:56 ` [Intel-gfx] [PATCH 5/7] drm/i915/guc: Add extra debug on CT deadlock Matthew Brost
2021-12-11  1:43   ` John Harrison
2021-12-11  1:45     ` John Harrison
2021-12-11  3:24       ` Matthew Brost
2021-12-11  0:56 ` [Intel-gfx] [PATCH 6/7] drm/i915/guc: Kick G2H tasklet if no credits Matthew Brost
2021-12-11  0:56 ` [Intel-gfx] [PATCH 7/7] drm/i915/guc: Selftest for stealing of guc ids Matthew Brost
2021-12-11  1:33   ` John Harrison
2021-12-11  3:31     ` Matthew Brost
2021-12-11  3:32     ` Matthew Brost
2021-12-11  2:28 ` [Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for Fix stealing guc_ids + test Patchwork
2021-12-11  2:30 ` [Intel-gfx] ✗ Fi.CI.SPARSE: " Patchwork
2021-12-11  2:58 ` [Intel-gfx] ✓ Fi.CI.BAT: success " Patchwork
2021-12-12  1:32 ` [Intel-gfx] ✓ Fi.CI.IGT: " Patchwork
  -- strict thread matches above, loose matches on Subject: below --
2021-12-11 17:35 [Intel-gfx] [PATCH 0/7] " Matthew Brost
2021-12-11 17:35 ` [Intel-gfx] [PATCH 4/7] drm/i915/guc: Don't hog IRQs when destroying contexts Matthew Brost
2021-12-14 17:04 [Intel-gfx] [PATCH 0/7] Fix stealing guc_ids + test Matthew Brost
2021-12-14 17:04 ` [Intel-gfx] [PATCH 4/7] drm/i915/guc: Don't hog IRQs when destroying contexts Matthew Brost
2021-12-17 11:06   ` Tvrtko Ursulin
2021-12-17 11:14     ` Tvrtko Ursulin
2021-12-22 16:25       ` Tvrtko Ursulin
2021-12-22 20:38         ` Matthew Brost

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20211211011049.GA8660@jons-linux-dev-box \
    --to=matthew.brost@intel.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=john.c.harrison@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox