[Intel-gfx] [PATCH 11/16] drm/i915: Remove (struct_mutex) locking for busy-ioctl

[Joonas Lahtinen]

On ma, 2016-08-01 at 19:22 +0100, Chris Wilson wrote:
> By applying the same logic as for wait-ioctl, we can query whether a
> request has completed without holding struct_mutex. The biggest impact
> system-wide is removing the flush_active and the contention that causes.
> 
> Testcase: igt/gem_busy
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Akash Goel <akash.goel at intel.com>
> ---
>  drivers/gpu/drm/i915/i915_gem.c | 110 +++++++++++++++++++++++++++++-----------
>  1 file changed, 80 insertions(+), 30 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 43069b05bdd2..f2f70f5ff9f4 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -3721,49 +3721,99 @@ i915_gem_object_ggtt_unpin_view(struct drm_i915_gem_object *obj,
>  	i915_vma_unpin(i915_gem_obj_to_ggtt_view(obj, view));
>  }
>  
> +static __always_inline unsigned
> +__busy_read_flag(const struct drm_i915_gem_request *request)
> +{
> +	return 0x10000 << request->engine->exec_id;
> +}
> +
> +static __always_inline unsigned int
> +__busy_write_flag(const struct drm_i915_gem_request *request)
> +{
> +	return request->engine->exec_id;

Just realized (to my horror) this is not a flag, it's a bare ID, so
better not call the function _flag, but rather _id?

> +}
> +
> +static __always_inline unsigned
> +__busy_flag(const struct i915_gem_active *active,
> +	    unsigned int (*flag)(const struct drm_i915_gem_request *))
> +{
> +	struct drm_i915_gem_request *request;
> +
> +	request = rcu_dereference(active->request);
> +	if (!request || i915_gem_request_completed(request))
> +		return 0;
> +
> +	return flag(request);
> +}
> +
> +static inline unsigned
> +busy_read_flag(const struct i915_gem_active *active)
> +{
> +	return __busy_flag(active, __busy_read_flag);
> +}
> +
> +static inline unsigned
> +busy_write_flag(const struct i915_gem_active *active)
> +{
> +	return __busy_flag(active, __busy_write_flag);
> +}
> +
>  int
>  i915_gem_busy_ioctl(struct drm_device *dev, void *data,
>  		    struct drm_file *file)
>  {
>  	struct drm_i915_gem_busy *args = data;
>  	struct drm_i915_gem_object *obj;
> -	int ret;
> -
> -	ret = i915_mutex_lock_interruptible(dev);
> -	if (ret)
> -		return ret;
> +	unsigned long active;
>  
>  	obj = i915_gem_object_lookup(file, args->handle);
> -	if (!obj) {
> -		ret = -ENOENT;
> -		goto unlock;
> -	}
> +	if (!obj)
> +		return -ENOENT;
>  
> -	/* Count all active objects as busy, even if they are currently not used
> -	 * by the gpu. Users of this interface expect objects to eventually
> -	 * become non-busy without any further actions.
> -	 */
>  	args->busy = 0;
> -	if (i915_gem_object_is_active(obj)) {
> -		struct drm_i915_gem_request *req;
> -		int i;
> +	active = __I915_BO_ACTIVE(obj);
> +	if (active) {
> +		int idx;
>  
> -		for (i = 0; i < I915_NUM_ENGINES; i++) {
> -			req = i915_gem_active_peek(&obj->last_read[i],
> -						   &obj->base.dev->struct_mutex);
> -			if (req)
> -				args->busy |= 1 << (16 + req->engine->exec_id);
> -		}
> -		req = i915_gem_active_peek(&obj->last_write,
> -					   &obj->base.dev->struct_mutex);
> -		if (req)
> -			args->busy |= req->engine->exec_id;
> +		/* Yes, the lookups are intentionally racy.
> +		 *
> +		 * Even though we guard the pointer lookup by RCU, that only
> +		 * guarantees that the pointer and its contents remain
> +		 * dereferencable and does *not* mean that the request we
> +		 * have is the same as the one being tracked by the object.
> +		 *
> +		 * Consider that we lookup the request just as it is being
> +		 * retired and free. We take a local copy of the pointer,

still s/free/freed/

> +		 * but before we add its engine into the busy set, the other
> +		 * thread reallocates it and assigns it to a task on another
> +		 * engine with a fresh and incomplete seqno.
> +		 *
> +		 * So after we lookup the engine's id, we double check that
> +		 * the active request is the same and only then do we add it
> +		 * into the busy set.
> +		 */
> +		rcu_read_lock();
> +
> +		for_each_active(active, idx)
> +			args->busy |= busy_read_flag(&obj->last_read[idx]);

So you mean this is double check against __I915_BO_ACTIVE, right?

We're getting there, though. With above fixed;

Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation