Re: [Intel-gfx] [PATCH] drm/i915: Be more gentle with exiting non-persistent context

[Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxxxxxxx>]

On 30/04/2021 12:48, Daniel Vetter wrote:
> On Thu, Apr 29, 2021 at 10:46:40AM +0100, Tvrtko Ursulin wrote:
> > From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
> >
> > When a non-persistent context exits we currently mark it as banned in
> > order to trigger fast termination of any outstanding GPU jobs it may have
> > left running.
> >
> > In doing so we apply a very strict 1ms limit in which the left over job
> > has to preempt before we issues an engine resets.
> >
> > Some workloads are not able to cleanly preempt in that time window and it
> > can be argued that it would instead be better to give them a bit more
> > grace since avoiding engine resets is generally preferrable.
>
> Can you pls explain here why this is preferrable?

I think there's always the risk of an innocent request getting axed with 
preempt to busy being very asynchronous and also engine reset can 
sometimes fail as well.

> > To achieve this the patch splits handling of banned contexts from simply
> > exited non-persistent ones and then applies different timeouts for both
> > and also extends the criteria which determines if a request should be
> > scheduled back in after preemption or not.
> >
> > 15ms preempt timeout grace is given to exited non-persistent contexts
> > which have been empirically tested to satisfy customers requirements
> > and still provides reasonably quick cleanup post exit.
>
> Same here, a bit more detail on what exactly was the problem to be fixed
> is needed.

It is a bit multi-faceted. Start with how in some cultures errors 
messages are much bigger error flags than in others and much more 
difficult to hand-wave "oh that's not a problem really". The the 
previous considerations about why not avoid engine reset if we can. Add 
on top how non-persistent context exiting is not really an error worthy 
of a reset, *if* it exits cleanly reasonably quickly.

You could say make it clean up for itself before it exits, not a kernel 
problem. But on the balance of everything, to me it sounds saleable to 
just give it some longer time compared to banned contexts, which are the 
unquestionably naughty/dangerous ones. Also, how fast non-persistent 
contexts will be cleaned up hasn't been defined really. As long as 15ms 
is an order of magnitude, plus some, shorter than the normal preempt 
timeout I think it is fine.

Regards,

Tvrtko

P.S. Otherwise I plan to respin v2 with consolidated CONTEXT_SCHEDULABLE 
flag so fast paths do not have to do too many individual checks.

> -Daniel
>
> > Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
> > Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> > Cc: Zhen Han <zhen.han@xxxxxxxxx>
> > ---
> >   drivers/gpu/drm/i915/gem/i915_gem_context.c     | 15 +++++++++------
> >   drivers/gpu/drm/i915/gt/intel_context.h         | 17 +++++++++++++++++
> >   drivers/gpu/drm/i915/gt/intel_context_types.h   |  1 +
> >   .../drm/i915/gt/intel_execlists_submission.c    | 12 ++++++++++--
> >   drivers/gpu/drm/i915/i915_request.c             |  2 +-
> >   5 files changed, 38 insertions(+), 9 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> > index fd8ee52e17a4..5a6eba1232cd 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> > @@ -426,7 +426,8 @@ static struct intel_engine_cs *active_engine(struct intel_context *ce)
> >   	return engine;
> >   }
> >   
> > -static void kill_engines(struct i915_gem_engines *engines, bool ban)
> > +static void
> > +kill_engines(struct i915_gem_engines *engines, bool ban, bool persistent)
> >   {
> >   	struct i915_gem_engines_iter it;
> >   	struct intel_context *ce;
> > @@ -443,6 +444,8 @@ static void kill_engines(struct i915_gem_engines *engines, bool ban)
> >   
> >   		if (ban && intel_context_set_banned(ce))
> >   			continue;
> > +		else if (!persistent && intel_context_set_non_persistent(ce))
> > +			continue;
> >   
> >   		/*
> >   		 * Check the current active state of this context; if we
> > @@ -454,7 +457,7 @@ static void kill_engines(struct i915_gem_engines *engines, bool ban)
> >   		engine = active_engine(ce);
> >   
> >   		/* First attempt to gracefully cancel the context */
> > -		if (engine && !__cancel_engine(engine) && ban)
> > +		if (engine && !__cancel_engine(engine) && (ban || !persistent))
> >   			/*
> >   			 * If we are unable to send a preemptive pulse to bump
> >   			 * the context from the GPU, we have to resort to a full
> > @@ -466,8 +469,6 @@ static void kill_engines(struct i915_gem_engines *engines, bool ban)
> >   
> >   static void kill_context(struct i915_gem_context *ctx)
> >   {
> > -	bool ban = (!i915_gem_context_is_persistent(ctx) ||
> > -		    !ctx->i915->params.enable_hangcheck);
> >   	struct i915_gem_engines *pos, *next;
> >   
> >   	spin_lock_irq(&ctx->stale.lock);
> > @@ -480,7 +481,8 @@ static void kill_context(struct i915_gem_context *ctx)
> >   
> >   		spin_unlock_irq(&ctx->stale.lock);
> >   
> > -		kill_engines(pos, ban);
> > +		kill_engines(pos, !ctx->i915->params.enable_hangcheck,
> > +			     i915_gem_context_is_persistent(ctx));
> >   
> >   		spin_lock_irq(&ctx->stale.lock);
> >   		GEM_BUG_ON(i915_sw_fence_signaled(&pos->fence));
> > @@ -526,7 +528,8 @@ static void engines_idle_release(struct i915_gem_context *ctx,
> >   
> >   kill:
> >   	if (list_empty(&engines->link)) /* raced, already closed */
> > -		kill_engines(engines, true);
> > +		kill_engines(engines, true,
> > +			     i915_gem_context_is_persistent(ctx));
> >   
> >   	i915_sw_fence_commit(&engines->fence);
> >   }
> > diff --git a/drivers/gpu/drm/i915/gt/intel_context.h b/drivers/gpu/drm/i915/gt/intel_context.h
> > index f83a73a2b39f..b0e812b8ce39 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_context.h
> > +++ b/drivers/gpu/drm/i915/gt/intel_context.h
> > @@ -220,6 +220,23 @@ static inline bool intel_context_set_banned(struct intel_context *ce)
> >   	return test_and_set_bit(CONTEXT_BANNED, &ce->flags);
> >   }
> >   
> > +static inline bool intel_context_is_non_persistent(const struct intel_context *ce)
> > +{
> > +	return test_bit(CONTEXT_NON_PERSISTENT, &ce->flags);
> > +}
> > +
> > +static inline bool intel_context_set_non_persistent(struct intel_context *ce)
> > +{
> > +	return test_and_set_bit(CONTEXT_NON_PERSISTENT, &ce->flags);
> > +}
> > +
> > +static inline bool intel_context_is_schedulable(const struct intel_context *ce)
> > +{
> > +	return !intel_context_is_banned(ce) &&
> > +	       !(intel_context_is_closed(ce) &&
> > +	         intel_context_is_non_persistent(ce));
> > +}
> > +
> >   static inline bool
> >   intel_context_force_single_submission(const struct intel_context *ce)
> >   {
> > diff --git a/drivers/gpu/drm/i915/gt/intel_context_types.h b/drivers/gpu/drm/i915/gt/intel_context_types.h
> > index ed8c447a7346..aa949c539e93 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_context_types.h
> > +++ b/drivers/gpu/drm/i915/gt/intel_context_types.h
> > @@ -95,6 +95,7 @@ struct intel_context {
> >   #define CONTEXT_BANNED			6
> >   #define CONTEXT_FORCE_SINGLE_SUBMISSION	7
> >   #define CONTEXT_NOPREEMPT		8
> > +#define CONTEXT_NON_PERSISTENT		9 /* Only if also closed. */
> >   
> >   	struct {
> >   		u64 timeout_us;
> > diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> > index de124870af44..5ad7272fbbc4 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> > +++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
> > @@ -478,7 +478,7 @@ __execlists_schedule_in(struct i915_request *rq)
> >   		     !intel_engine_has_heartbeat(engine)))
> >   		intel_context_set_banned(ce);
> >   
> > -	if (unlikely(intel_context_is_banned(ce) || bad_request(rq)))
> > +	if (unlikely(!intel_context_is_schedulable(ce) || bad_request(rq)))
> >   		reset_active(rq, engine);
> >   
> >   	if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
> > @@ -1204,12 +1204,20 @@ static void record_preemption(struct intel_engine_execlists *execlists)
> >   static unsigned long active_preempt_timeout(struct intel_engine_cs *engine,
> >   					    const struct i915_request *rq)
> >   {
> > +	struct intel_context *ce;
> > +
> >   	if (!rq)
> >   		return 0;
> >   
> > +	ce = rq->context;
> > +
> >   	/* Force a fast reset for terminated contexts (ignoring sysfs!) */
> > -	if (unlikely(intel_context_is_banned(rq->context) || bad_request(rq)))
> > +	if (unlikely(intel_context_is_banned(ce) || bad_request(rq)))
> >   		return 1;
> > +	/* Longer grace for closed non-persistent contexts to avoid resets. */
> > +	else if (unlikely(intel_context_is_closed(ce) &&
> > +			  intel_context_is_non_persistent(ce)))
> > +		return 15;
> >   
> >   	return READ_ONCE(engine->props.preempt_timeout_ms);
> >   }
> > diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> > index bec9c3652188..bbac87535923 100644
> > --- a/drivers/gpu/drm/i915/i915_request.c
> > +++ b/drivers/gpu/drm/i915/i915_request.c
> > @@ -660,7 +660,7 @@ bool __i915_request_submit(struct i915_request *request)
> >   		goto active;
> >   	}
> >   
> > -	if (unlikely(intel_context_is_banned(request->context)))
> > +	if (unlikely(!intel_context_is_schedulable(request->context)))
> >   		i915_request_set_error_once(request, -EIO);
> >   
> >   	if (unlikely(fatal_error(request->fence.error)))
> > --
> > 2.30.2
> >
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
> > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel