Re: [Intel-gfx] [PATCH] drm/i915/gem: Take a copy of the engines for context_barrier_task

[Maarten Lankhorst <maarten.lankhorst@linux.intel.com>]

Op 11-03-2020 om 13:49 schreef Chris Wilson:
> When applying the context-barrier, we only care about the current
> engines, as the next set of engines will be naturally after the barrier.
> So we can skip holding the ctx->engines_mutex while constructing the
> request by taking a sneaky reference to the i915_gem_engines instead.
>
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> ---
>  drivers/gpu/drm/i915/gem/i915_gem_context.c | 89 ++++++++++++++-------
>  1 file changed, 58 insertions(+), 31 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> index 50ecc0b2b235..e2357099a9ed 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> @@ -261,6 +261,34 @@ static void free_engines_rcu(struct rcu_head *rcu)
>  	free_engines(engines);
>  }
>  
> +static int engines_notify(struct i915_sw_fence *fence,
> +			  enum i915_sw_fence_notify state)
> +{
> +	struct i915_gem_engines *engines =
> +		container_of(fence, typeof(*engines), fence);
> +
> +	switch (state) {
> +	case FENCE_COMPLETE:
> +		if (!list_empty(&engines->link)) {
> +			struct i915_gem_context *ctx = engines->ctx;
> +			unsigned long flags;
> +
> +			spin_lock_irqsave(&ctx->stale.lock, flags);
> +			list_del(&engines->link);
> +			spin_unlock_irqrestore(&ctx->stale.lock, flags);
> +		}
> +		i915_gem_context_put(engines->ctx);
> +		break;
> +
> +	case FENCE_FREE:
> +		init_rcu_head(&engines->rcu);
> +		call_rcu(&engines->rcu, free_engines_rcu);
> +		break;
> +	}
> +
> +	return NOTIFY_DONE;
> +}
> +
>  static struct i915_gem_engines *default_engines(struct i915_gem_context *ctx)
>  {
>  	const struct intel_gt *gt = &ctx->i915->gt;
> @@ -272,6 +300,8 @@ static struct i915_gem_engines *default_engines(struct i915_gem_context *ctx)
>  	if (!e)
>  		return ERR_PTR(-ENOMEM);
>  
> +	i915_sw_fence_init(&e->fence, engines_notify);
> +
>  	for_each_engine(engine, gt, id) {
>  		struct intel_context *ce;
>  
> @@ -519,41 +549,12 @@ static void kill_context(struct i915_gem_context *ctx)
>  	kill_stale_engines(ctx);
>  }
>  
> -static int engines_notify(struct i915_sw_fence *fence,
> -			  enum i915_sw_fence_notify state)
> -{
> -	struct i915_gem_engines *engines =
> -		container_of(fence, typeof(*engines), fence);
> -
> -	switch (state) {
> -	case FENCE_COMPLETE:
> -		if (!list_empty(&engines->link)) {
> -			struct i915_gem_context *ctx = engines->ctx;
> -			unsigned long flags;
> -
> -			spin_lock_irqsave(&ctx->stale.lock, flags);
> -			list_del(&engines->link);
> -			spin_unlock_irqrestore(&ctx->stale.lock, flags);
> -		}
> -		i915_gem_context_put(engines->ctx);
> -		break;
> -
> -	case FENCE_FREE:
> -		init_rcu_head(&engines->rcu);
> -		call_rcu(&engines->rcu, free_engines_rcu);
> -		break;
> -	}
> -
> -	return NOTIFY_DONE;
> -}
> -
>  static void engines_idle_release(struct i915_gem_context *ctx,
>  				 struct i915_gem_engines *engines)
>  {
>  	struct i915_gem_engines_iter it;
>  	struct intel_context *ce;
>  
> -	i915_sw_fence_init(&engines->fence, engines_notify);
>  	INIT_LIST_HEAD(&engines->link);
>  
>  	engines->ctx = i915_gem_context_get(ctx);
> @@ -1079,6 +1080,30 @@ static void cb_retire(struct i915_active *base)
>  	kfree(cb);
>  }
>  
> +static inline struct i915_gem_engines *
> +__context_engines_await(const struct i915_gem_context *ctx)
> +{
> +	struct i915_gem_engines *engines;
> +
> +	rcu_read_lock();
> +	do {
> +		engines = rcu_dereference(ctx->engines);
> +		if (!engines)
> +			break;
> +
> +		if (!i915_sw_fence_await(&engines->fence))
> +			continue;
> +
> +		if (engines == rcu_access_pointer(ctx->engines))
> +			break;
> +
> +		i915_sw_fence_complete(&engines->fence);
> +	} while(1);
> +	rcu_read_unlock();
> +
> +	return engines;
> +}
> +
>  I915_SELFTEST_DECLARE(static intel_engine_mask_t context_barrier_inject_fault);
>  static int context_barrier_task(struct i915_gem_context *ctx,
>  				intel_engine_mask_t engines,
> @@ -1089,6 +1114,7 @@ static int context_barrier_task(struct i915_gem_context *ctx,
>  {
>  	struct context_barrier_task *cb;
>  	struct i915_gem_engines_iter it;
> +	struct i915_gem_engines *e;
>  	struct intel_context *ce;
>  	int err = 0;
>  
> @@ -1105,7 +1131,8 @@ static int context_barrier_task(struct i915_gem_context *ctx,
>  		return err;
>  	}
>  
> -	for_each_gem_engine(ce, i915_gem_context_lock_engines(ctx), it) {
> +	e = __context_engines_await(ctx);
> +	for_each_gem_engine(ce, e, it) {
>  		struct i915_request *rq;
>  
>  		if (I915_SELFTEST_ONLY(context_barrier_inject_fault &

This doesn't need RCU, but it will work anyway. :)

Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>

> @@ -1136,7 +1163,7 @@ static int context_barrier_task(struct i915_gem_context *ctx,
>  		if (err)
>  			break;
>  	}
> -	i915_gem_context_unlock_engines(ctx);
> +	i915_sw_fence_complete(&e->fence);
>  
>  	cb->task = err ? NULL : task; /* caller needs to unwind instead */
>  	cb->data = data;


_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx