[Intel-gfx] [PATCH 02/39] drm/i915: Refine i915_reset.lock_map

[Mika Kuoppala]

Chris Wilson <chris at chris-wilson.co.uk> writes:

> We already use a mutex to serialise i915_reset() and wedging, so all we
> need it to link that into i915_request_wait() and we have our lock cycle
> detection.
>
> v2.5: Take error mutex for selftests
>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/gt/intel_reset.c            |  6 ++----
>  drivers/gpu/drm/i915/i915_drv.h                  |  8 --------
>  drivers/gpu/drm/i915/i915_gem.c                  |  3 ---
>  drivers/gpu/drm/i915/i915_request.c              | 12 ++++++++++--
>  drivers/gpu/drm/i915/selftests/mock_gem_device.c |  2 --
>  5 files changed, 12 insertions(+), 19 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_reset.c b/drivers/gpu/drm/i915/gt/intel_reset.c
> index 8ba7af8b7ced..41a294f5cc19 100644
> --- a/drivers/gpu/drm/i915/gt/intel_reset.c
> +++ b/drivers/gpu/drm/i915/gt/intel_reset.c
> @@ -978,7 +978,7 @@ void i915_reset(struct drm_i915_private *i915,
>  
>  	might_sleep();
>  	GEM_BUG_ON(!test_bit(I915_RESET_BACKOFF, &error->flags));
> -	lock_map_acquire(&i915->gt.reset_lockmap);
> +	mutex_lock(&error->wedge_mutex);
>  
>  	/* Clear any previous failed attempts at recovery. Time to try again. */
>  	if (!__i915_gem_unset_wedged(i915))
> @@ -1031,7 +1031,7 @@ void i915_reset(struct drm_i915_private *i915,
>  finish:
>  	reset_finish(i915);
>  unlock:
> -	lock_map_release(&i915->gt.reset_lockmap);
> +	mutex_unlock(&error->wedge_mutex);
>  	return;
>  
>  taint:
> @@ -1147,9 +1147,7 @@ static void i915_reset_device(struct drm_i915_private *i915,
>  		/* Flush everyone using a resource about to be clobbered */
>  		synchronize_srcu_expedited(&error->reset_backoff_srcu);
>  
> -		mutex_lock(&error->wedge_mutex);
>  		i915_reset(i915, engine_mask, reason);
> -		mutex_unlock(&error->wedge_mutex);
>  
>  		intel_finish_reset(i915);
>  	}
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 90d94d904e65..3683ef6d4c28 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1901,14 +1901,6 @@ struct drm_i915_private {
>  		ktime_t last_init_time;
>  
>  		struct i915_vma *scratch;
> -
> -		/*
> -		 * We must never wait on the GPU while holding a lock as we
> -		 * may need to perform a GPU reset. So while we don't need to
> -		 * serialise wait/reset with an explicit lock, we do want
> -		 * lockdep to detect potential dependency cycles.
> -		 */
> -		struct lockdep_map reset_lockmap;
>  	} gt;
>  
>  	struct {
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 4bbded4aa936..7232361973fd 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1746,7 +1746,6 @@ static void i915_gem_init__mm(struct drm_i915_private *i915)
>  
>  int i915_gem_init_early(struct drm_i915_private *dev_priv)
>  {
> -	static struct lock_class_key reset_key;
>  	int err;
>  
>  	intel_gt_pm_init(dev_priv);
> @@ -1754,8 +1753,6 @@ int i915_gem_init_early(struct drm_i915_private *dev_priv)
>  	INIT_LIST_HEAD(&dev_priv->gt.active_rings);
>  	INIT_LIST_HEAD(&dev_priv->gt.closed_vma);
>  	spin_lock_init(&dev_priv->gt.closed_lock);
> -	lockdep_init_map(&dev_priv->gt.reset_lockmap,
> -			 "i915.reset", &reset_key, 0);
>  
>  	i915_gem_init__mm(dev_priv);
>  	i915_gem_init__pm(dev_priv);
> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
> index 1cbc3ef4fc27..5311286578b7 100644
> --- a/drivers/gpu/drm/i915/i915_request.c
> +++ b/drivers/gpu/drm/i915/i915_request.c
> @@ -1444,7 +1444,15 @@ long i915_request_wait(struct i915_request *rq,
>  		return -ETIME;
>  
>  	trace_i915_request_wait_begin(rq, flags);
> -	lock_map_acquire(&rq->i915->gt.reset_lockmap);
> +
> +	/*
> +	 * We must never wait on the GPU while holding a lock as we
> +	 * may need to perform a GPU reset. So while we don't need to
> +	 * serialise wait/reset with an explicit lock, we do want
> +	 * lockdep to detect potential dependency cycles.
> +	 */
> +	mutex_acquire(&rq->i915->gpu_error.wedge_mutex.dep_map,
> +		      0, 0, _THIS_IP_);

Seems to translate to exclusive lock with full checking.

There was ofcourse a slight possibilty that previous reviewer did
read all the lockdep.h. Looked at the wedge mutex and connected
the dots. Well, it is obvious now.

Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>

>  
>  	/*
>  	 * Optimistic spin before touching IRQs.
> @@ -1518,7 +1526,7 @@ long i915_request_wait(struct i915_request *rq,
>  	dma_fence_remove_callback(&rq->fence, &wait.cb);
>  
>  out:
> -	lock_map_release(&rq->i915->gt.reset_lockmap);
> +	mutex_release(&rq->i915->gpu_error.wedge_mutex.dep_map, 0, _THIS_IP_);
>  	trace_i915_request_wait_end(rq);
>  	return timeout;
>  }
> diff --git a/drivers/gpu/drm/i915/selftests/mock_gem_device.c b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
> index 1e9ffced78c1..b7f3fbb4ae89 100644
> --- a/drivers/gpu/drm/i915/selftests/mock_gem_device.c
> +++ b/drivers/gpu/drm/i915/selftests/mock_gem_device.c
> @@ -130,7 +130,6 @@ static struct dev_pm_domain pm_domain = {
>  
>  struct drm_i915_private *mock_gem_device(void)
>  {
> -	static struct lock_class_key reset_key;
>  	struct drm_i915_private *i915;
>  	struct pci_dev *pdev;
>  	int err;
> @@ -205,7 +204,6 @@ struct drm_i915_private *mock_gem_device(void)
>  	INIT_LIST_HEAD(&i915->gt.active_rings);
>  	INIT_LIST_HEAD(&i915->gt.closed_vma);
>  	spin_lock_init(&i915->gt.closed_lock);
> -	lockdep_init_map(&i915->gt.reset_lockmap, "i915.reset", &reset_key, 0);
>  
>  	mutex_lock(&i915->drm.struct_mutex);
>  
> -- 
> 2.20.1