diff mbox series

[2/5] drm/i915: Allow asynchronous waits on the i915_active barriers

Message ID 20200406091254.17675-2-chris@chris-wilson.co.uk (mailing list archive)
State New, archived
Headers show
Series [1/5] drm/i915: Make exclusive awaits on i915_active optional | expand

Commit Message

Chris Wilson April 6, 2020, 9:12 a.m. UTC 
Allow the caller to also wait upon the barriers stored in i915_active.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_active.c | 60 ++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_active.h |  1 +
 2 files changed, 61 insertions(+)

Comments

Tvrtko Ursulin April 6, 2020, 12:06 p.m. UTC | #1 
On 06/04/2020 10:12, Chris Wilson wrote:
> Allow the caller to also wait upon the barriers stored in i915_active.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>   drivers/gpu/drm/i915/i915_active.c | 60 ++++++++++++++++++++++++++++++
>   drivers/gpu/drm/i915/i915_active.h |  1 +
>   2 files changed, 61 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
> index d5e24be759f7..048ab9edd2c2 100644
> --- a/drivers/gpu/drm/i915/i915_active.c
> +++ b/drivers/gpu/drm/i915/i915_active.c
> @@ -542,6 +542,55 @@ static int __await_active(struct i915_active_fence *active,
>   	return 0;
>   }
>   
> +struct wait_barrier {
> +	struct wait_queue_entry base;
> +	struct i915_active *ref;
> +};
> +
> +static int
> +barrier_wake(wait_queue_entry_t *wq, unsigned int mode, int flags, void *key)
> +{
> +	struct wait_barrier *wb = container_of(wq, typeof(*wb), base);
> +
> +	if (i915_active_is_idle(wb->ref)) { /* shared waitqueue, must check! */

Who shares it?

> +		list_del(&wq->entry);
> +		i915_sw_fence_complete(wq->private);
> +		kfree(wq);
> +	}
> +
> +	return 0;
> +}
> +
> +static int __await_barrier(struct i915_active *ref, struct i915_sw_fence *fence)
> +{
> +	struct wait_barrier *wb;
> +
> +	wb = kmalloc(sizeof(*wb), GFP_KERNEL);
> +	if (unlikely(!wb))
> +		return -ENOMEM;
> +
> +	if (!i915_active_acquire_if_busy(ref)) {
> +		kfree(wb);
> +		return 0;
> +	}
> +
> +	if (!i915_sw_fence_await(fence)) {
> +		kfree(wb);
> +		i915_active_release(ref);
> +		return -EINVAL;
> +	}
> +
> +	wb->base.flags = 0;
> +	wb->base.func = barrier_wake;
> +	wb->base.private = fence;
> +	wb->ref = ref;
> +
> +	add_wait_queue(__var_waitqueue(ref), &wb->base);
> +
> +	i915_active_release(ref);
> +	return 0;
> +}
> +
>   static int await_active(struct i915_active *ref,
>   			unsigned int flags,
>   			int (*fn)(void *arg, struct dma_fence *fence),
> @@ -570,6 +619,16 @@ static int await_active(struct i915_active *ref,
>   			return err;
>   	}
>   
> +	if (flags & I915_ACTIVE_AWAIT_BARRIER) {
> +		err = flush_lazy_signals(ref);
> +		if (err)
> +			return err;
> +
> +		err = __await_barrier(ref, arg);
> +		if (err)
> +			return err;
>

Could have a single set of active_acquire_if_busy/release over the 
previous and this new block. Not sure if that would help with any 
atomicity concerns, or if there are such.

  +	}
> +
>   	return 0;
>   }
>   
> @@ -582,6 +641,7 @@ int i915_request_await_active(struct i915_request *rq,
>   			      struct i915_active *ref,
>   			      unsigned int flags)
>   {
> +	GEM_BUG_ON(flags & I915_ACTIVE_AWAIT_BARRIER);

Why is this an error?

>   	return await_active(ref, flags, rq_await_fence, rq);
>   }
>   
> diff --git a/drivers/gpu/drm/i915/i915_active.h b/drivers/gpu/drm/i915/i915_active.h
> index ffafaa78c494..cf4058150966 100644
> --- a/drivers/gpu/drm/i915/i915_active.h
> +++ b/drivers/gpu/drm/i915/i915_active.h
> @@ -195,6 +195,7 @@ int i915_request_await_active(struct i915_request *rq,
>   			      unsigned int flags);
>   #define I915_ACTIVE_AWAIT_EXCL BIT(0)
>   #define I915_ACTIVE_AWAIT_ACTIVE BIT(1)
> +#define I915_ACTIVE_AWAIT_BARRIER BIT(2)
>   
>   int i915_active_acquire(struct i915_active *ref);
>   bool i915_active_acquire_if_busy(struct i915_active *ref);
> 

Regards,

Tvrtko
Chris Wilson April 6, 2020, 1:09 p.m. UTC | #2 
Quoting Tvrtko Ursulin (2020-04-06 13:06:03)
> 
> On 06/04/2020 10:12, Chris Wilson wrote:
> > Allow the caller to also wait upon the barriers stored in i915_active.
> > 
> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > ---
> >   drivers/gpu/drm/i915/i915_active.c | 60 ++++++++++++++++++++++++++++++
> >   drivers/gpu/drm/i915/i915_active.h |  1 +
> >   2 files changed, 61 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
> > index d5e24be759f7..048ab9edd2c2 100644
> > --- a/drivers/gpu/drm/i915/i915_active.c
> > +++ b/drivers/gpu/drm/i915/i915_active.c
> > @@ -542,6 +542,55 @@ static int __await_active(struct i915_active_fence *active,
> >       return 0;
> >   }
> >   
> > +struct wait_barrier {
> > +     struct wait_queue_entry base;
> > +     struct i915_active *ref;
> > +};
> > +
> > +static int
> > +barrier_wake(wait_queue_entry_t *wq, unsigned int mode, int flags, void *key)
> > +{
> > +     struct wait_barrier *wb = container_of(wq, typeof(*wb), base);
> > +
> > +     if (i915_active_is_idle(wb->ref)) { /* shared waitqueue, must check! */
> 
> Who shares it?

__var_waitqueue(ref) => uses a one of a set of global workqueues based
off hash(ref)

Or we add a wait_queue_head_t to active, but we would still need to
recheck as it may be reused as we are signaled.

> > +     if (flags & I915_ACTIVE_AWAIT_BARRIER) {
> > +             err = flush_lazy_signals(ref);
> > +             if (err)
> > +                     return err;
> > +
> > +             err = __await_barrier(ref, arg);
> > +             if (err)
> > +                     return err;
> >
> 
> Could have a single set of active_acquire_if_busy/release over the 
> previous and this new block. Not sure if that would help with any 
> atomicity concerns, or if there are such.

It would not affect correctness, it will just depend on taste.

>   +     }
> > +
> >       return 0;
> >   }
> >   
> > @@ -582,6 +641,7 @@ int i915_request_await_active(struct i915_request *rq,
> >                             struct i915_active *ref,
> >                             unsigned int flags)
> >   {
> > +     GEM_BUG_ON(flags & I915_ACTIVE_AWAIT_BARRIER);
> 
> Why is this an error?

Because I'm being lazy and not hooking up the correct signaling path.

Instead of signaling arg == fence, we would need &request->submit. Just
messy on how to pass down the details.

Maybe 
	return await_active(ref, flags, rq_await_fence, rq, &rq->submit);
and
	return await_active(ref, flags, sw_await_fence, fence, fence);

That seems better than I was expecting.
-Chris
Chris Wilson April 6, 2020, 1:18 p.m. UTC | #3 
Quoting Chris Wilson (2020-04-06 14:09:44)
> Quoting Tvrtko Ursulin (2020-04-06 13:06:03)
> > 
> > On 06/04/2020 10:12, Chris Wilson wrote:
> > > Allow the caller to also wait upon the barriers stored in i915_active.
> > > 
> > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> > > ---
> > >   drivers/gpu/drm/i915/i915_active.c | 60 ++++++++++++++++++++++++++++++
> > >   drivers/gpu/drm/i915/i915_active.h |  1 +
> > >   2 files changed, 61 insertions(+)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
> > > index d5e24be759f7..048ab9edd2c2 100644
> > > --- a/drivers/gpu/drm/i915/i915_active.c
> > > +++ b/drivers/gpu/drm/i915/i915_active.c
> > > @@ -542,6 +542,55 @@ static int __await_active(struct i915_active_fence *active,
> > >       return 0;
> > >   }
> > >   
> > > +struct wait_barrier {
> > > +     struct wait_queue_entry base;
> > > +     struct i915_active *ref;
> > > +};
> > > +
> > > +static int
> > > +barrier_wake(wait_queue_entry_t *wq, unsigned int mode, int flags, void *key)
> > > +{
> > > +     struct wait_barrier *wb = container_of(wq, typeof(*wb), base);
> > > +
> > > +     if (i915_active_is_idle(wb->ref)) { /* shared waitqueue, must check! */
> > 
> > Who shares it?
> 
> __var_waitqueue(ref) => uses a one of a set of global workqueues based
> off hash(ref)
> 
> Or we add a wait_queue_head_t to active, but we would still need to
> recheck as it may be reused as we are signaled.
> 
> > > +     if (flags & I915_ACTIVE_AWAIT_BARRIER) {
> > > +             err = flush_lazy_signals(ref);
> > > +             if (err)
> > > +                     return err;
> > > +
> > > +             err = __await_barrier(ref, arg);
> > > +             if (err)
> > > +                     return err;
> > >
> > 
> > Could have a single set of active_acquire_if_busy/release over the 
> > previous and this new block. Not sure if that would help with any 
> > atomicity concerns, or if there are such.
> 
> It would not affect correctness, it will just depend on taste.

Actually, flush_lazy_signals needs to be inside the active-ref, so we
should rearrange the acquires.
-Chris
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
index d5e24be759f7..048ab9edd2c2 100644
--- a/drivers/gpu/drm/i915/i915_active.c
+++ b/drivers/gpu/drm/i915/i915_active.c
@@ -542,6 +542,55 @@  static int __await_active(struct i915_active_fence *active,
 	return 0;
 }
 
+struct wait_barrier {
+	struct wait_queue_entry base;
+	struct i915_active *ref;
+};
+
+static int
+barrier_wake(wait_queue_entry_t *wq, unsigned int mode, int flags, void *key)
+{
+	struct wait_barrier *wb = container_of(wq, typeof(*wb), base);
+
+	if (i915_active_is_idle(wb->ref)) { /* shared waitqueue, must check! */
+		list_del(&wq->entry);
+		i915_sw_fence_complete(wq->private);
+		kfree(wq);
+	}
+
+	return 0;
+}
+
+static int __await_barrier(struct i915_active *ref, struct i915_sw_fence *fence)
+{
+	struct wait_barrier *wb;
+
+	wb = kmalloc(sizeof(*wb), GFP_KERNEL);
+	if (unlikely(!wb))
+		return -ENOMEM;
+
+	if (!i915_active_acquire_if_busy(ref)) {
+		kfree(wb);
+		return 0;
+	}
+
+	if (!i915_sw_fence_await(fence)) {
+		kfree(wb);
+		i915_active_release(ref);
+		return -EINVAL;
+	}
+
+	wb->base.flags = 0;
+	wb->base.func = barrier_wake;
+	wb->base.private = fence;
+	wb->ref = ref;
+
+	add_wait_queue(__var_waitqueue(ref), &wb->base);
+
+	i915_active_release(ref);
+	return 0;
+}
+
 static int await_active(struct i915_active *ref,
 			unsigned int flags,
 			int (*fn)(void *arg, struct dma_fence *fence),
@@ -570,6 +619,16 @@  static int await_active(struct i915_active *ref,
 			return err;
 	}
 
+	if (flags & I915_ACTIVE_AWAIT_BARRIER) {
+		err = flush_lazy_signals(ref);
+		if (err)
+			return err;
+
+		err = __await_barrier(ref, arg);
+		if (err)
+			return err;
+	}
+
 	return 0;
 }
 
@@ -582,6 +641,7 @@  int i915_request_await_active(struct i915_request *rq,
 			      struct i915_active *ref,
 			      unsigned int flags)
 {
+	GEM_BUG_ON(flags & I915_ACTIVE_AWAIT_BARRIER);
 	return await_active(ref, flags, rq_await_fence, rq);
 }
 
diff --git a/drivers/gpu/drm/i915/i915_active.h b/drivers/gpu/drm/i915/i915_active.h
index ffafaa78c494..cf4058150966 100644
--- a/drivers/gpu/drm/i915/i915_active.h
+++ b/drivers/gpu/drm/i915/i915_active.h
@@ -195,6 +195,7 @@  int i915_request_await_active(struct i915_request *rq,
 			      unsigned int flags);
 #define I915_ACTIVE_AWAIT_EXCL BIT(0)
 #define I915_ACTIVE_AWAIT_ACTIVE BIT(1)
+#define I915_ACTIVE_AWAIT_BARRIER BIT(2)
 
 int i915_active_acquire(struct i915_active *ref);
 bool i915_active_acquire_if_busy(struct i915_active *ref);