diff mbox series

[08/10] drm/msm: async commit support

Message ID 20190829164601.11615-9-robdclark@gmail.com (mailing list archive)
State New, archived
Headers show
Series drm/msm: async commit support (v2) | expand

Commit Message

Rob Clark Aug. 29, 2019, 4:45 p.m. UTC
From: Rob Clark <robdclark@chromium.org>

Now that flush/wait/complete is decoupled from the "synchronous" part of
atomic commit_tail(), add support to defer flush to a timer that expires
shortly before vblank for async commits.  In this way, multiple atomic
commits (for example, cursor updates) can be coalesced into a single
flush at the end of the frame.

v2: don't hold lock over ->wait_flush(), to avoid locking interaction
    that was causing fps drop when combining page flips or non-async
    atomic commits and lots of legacy cursor updates

Signed-off-by: Rob Clark <robdclark@chromium.org>
---
 drivers/gpu/drm/msm/msm_atomic.c | 156 ++++++++++++++++++++++++++++++-
 drivers/gpu/drm/msm/msm_drv.c    |   1 +
 drivers/gpu/drm/msm/msm_drv.h    |   4 +
 drivers/gpu/drm/msm/msm_kms.h    |  50 ++++++++++
 4 files changed, 210 insertions(+), 1 deletion(-)

Comments

Sean Paul Sept. 3, 2019, 8:51 p.m. UTC | #1
On Thu, Aug 29, 2019 at 09:45:16AM -0700, Rob Clark wrote:
> From: Rob Clark <robdclark@chromium.org>
> 
> Now that flush/wait/complete is decoupled from the "synchronous" part of
> atomic commit_tail(), add support to defer flush to a timer that expires
> shortly before vblank for async commits.  In this way, multiple atomic
> commits (for example, cursor updates) can be coalesced into a single
> flush at the end of the frame.
> 
> v2: don't hold lock over ->wait_flush(), to avoid locking interaction
>     that was causing fps drop when combining page flips or non-async
>     atomic commits and lots of legacy cursor updates
> 
> Signed-off-by: Rob Clark <robdclark@chromium.org>

Some nits, with them addressed,

Reviewed-by: Sean Paul <sean@poorly.run>


> ---
>  drivers/gpu/drm/msm/msm_atomic.c | 156 ++++++++++++++++++++++++++++++-
>  drivers/gpu/drm/msm/msm_drv.c    |   1 +
>  drivers/gpu/drm/msm/msm_drv.h    |   4 +
>  drivers/gpu/drm/msm/msm_kms.h    |  50 ++++++++++
>  4 files changed, 210 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/msm/msm_atomic.c b/drivers/gpu/drm/msm/msm_atomic.c
> index 614fb9c5bb58..8f8f74337cb4 100644
> --- a/drivers/gpu/drm/msm/msm_atomic.c
> +++ b/drivers/gpu/drm/msm/msm_atomic.c
> @@ -29,6 +29,95 @@ int msm_atomic_prepare_fb(struct drm_plane *plane,
>  	return msm_framebuffer_prepare(new_state->fb, kms->aspace);
>  }
>  
> +static void msm_atomic_async_commit(struct msm_kms *kms, int crtc_idx)
> +{
> +	unsigned crtc_mask = BIT(crtc_idx);
> +
> +	mutex_lock(&kms->commit_lock);
> +
> +	if (!(kms->pending_crtc_mask & crtc_mask)) {
> +		mutex_unlock(&kms->commit_lock);
> +		return;
> +	}
> +
> +	kms->pending_crtc_mask &= ~crtc_mask;
> +
> +	kms->funcs->enable_commit(kms);
> +
> +	/*
> +	 * Flush hardware updates:
> +	 */
> +	DRM_DEBUG_ATOMIC("triggering async commit\n");
> +	kms->funcs->flush_commit(kms, crtc_mask);
> +	mutex_unlock(&kms->commit_lock);
> +
> +	/*
> +	 * Wait for flush to complete:
> +	 */
> +	kms->funcs->wait_flush(kms, crtc_mask);
> +
> +	mutex_lock(&kms->commit_lock);
> +	kms->funcs->complete_commit(kms, crtc_mask);
> +	mutex_unlock(&kms->commit_lock);
> +	kms->funcs->disable_commit(kms);
> +}
> +
> +static enum hrtimer_restart msm_atomic_pending_timer(struct hrtimer *t)
> +{
> +	struct msm_pending_timer *timer = container_of(t,
> +			struct msm_pending_timer, timer);
> +	struct msm_drm_private *priv = timer->kms->dev->dev_private;
> +

As discussed on IRC, a TODO here explaining that we'd like to remove this worker eventually but there are mutexes further down that need to be removed.

> +	queue_work(priv->wq, &timer->work);
> +
> +	return HRTIMER_NORESTART;
> +}
> +
> +static void msm_atomic_pending_work(struct work_struct *work)
> +{
> +	struct msm_pending_timer *timer = container_of(work,
> +			struct msm_pending_timer, work);
> +
> +	msm_atomic_async_commit(timer->kms, timer->crtc_idx);
> +}
> +
> +void msm_atomic_init_pending_timer(struct msm_pending_timer *timer,
> +		struct msm_kms *kms, int crtc_idx)
> +{
> +	timer->kms = kms;
> +	timer->crtc_idx = crtc_idx;
> +	hrtimer_init(&timer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
> +	timer->timer.function = msm_atomic_pending_timer;
> +	INIT_WORK(&timer->work, msm_atomic_pending_work);
> +}
> +
> +static bool can_do_async(struct drm_atomic_state *state,
> +		struct drm_crtc **async_crtc)
> +{
> +	struct drm_connector_state *connector_state;
> +	struct drm_connector *connector;
> +	struct drm_crtc_state *crtc_state;
> +	struct drm_crtc *crtc;
> +	int i, num_crtcs = 0;
> +
> +	if (!(state->legacy_cursor_update || state->async_update))
> +		return false;
> +
> +	/* any connector change, means slow path: */
> +	for_each_new_connector_in_state(state, connector, connector_state, i)
> +		return false;
> +
> +	for_each_new_crtc_in_state(state, crtc, crtc_state, i) {
> +		if (drm_atomic_crtc_needs_modeset(crtc_state))
> +			return false;
> +		if (++num_crtcs > 1)
> +			return false;
> +		*async_crtc = crtc;
> +	}
> +
> +	return true;
> +}
> +
>  /* Get bitmask of crtcs that will need to be flushed.  The bitmask
>   * can be used with for_each_crtc_mask() iterator, to iterate
>   * effected crtcs without needing to preserve the atomic state.
> @@ -50,9 +139,25 @@ void msm_atomic_commit_tail(struct drm_atomic_state *state)
>  	struct drm_device *dev = state->dev;
>  	struct msm_drm_private *priv = dev->dev_private;
>  	struct msm_kms *kms = priv->kms;
> +	struct drm_crtc *async_crtc = NULL;
>  	unsigned crtc_mask = get_crtc_mask(state);
> +	bool async = kms->funcs->vsync_time &&
> +			can_do_async(state, &async_crtc);
>  
>  	kms->funcs->enable_commit(kms);
> +
> +	/*
> +	 * Ensure any previous (potentially async) commit has
> +	 * completed:
> +	 */
> +	kms->funcs->wait_flush(kms, crtc_mask);
> +
> +	mutex_lock(&kms->commit_lock);
> +
> +	/*
> +	 * Now that there is no in-progress flush is complete,

s/is complete//

> +	 * prepare the current update:
> +	 */
>  	kms->funcs->prepare_commit(kms, state);
>  
>  	/*
> @@ -62,6 +167,49 @@ void msm_atomic_commit_tail(struct drm_atomic_state *state)
>  	drm_atomic_helper_commit_planes(dev, state, 0);
>  	drm_atomic_helper_commit_modeset_enables(dev, state);
>  
> +	if (async) {
> +		struct msm_pending_timer *timer =
> +			&kms->pending_timers[drm_crtc_index(async_crtc)];
> +
> +		/* async updates are limited to single-crtc updates: */
> +		WARN_ON(crtc_mask != drm_crtc_mask(async_crtc));

Put this in can_do_async()?

> +
> +		/*
> +		 * Start timer if we don't already have an update pending
> +		 * on this crtc:
> +		 */
> +		if (!(kms->pending_crtc_mask & crtc_mask)) {
> +			ktime_t vsync_time, wakeup_time;
> +
> +			kms->pending_crtc_mask |= crtc_mask;
> +
> +			vsync_time = kms->funcs->vsync_time(kms, async_crtc);
> +			wakeup_time = ktime_sub(vsync_time, ms_to_ktime(1));
> +
> +			hrtimer_start(&timer->timer, wakeup_time,
> +					HRTIMER_MODE_ABS);
> +		}
> +
> +		kms->funcs->disable_commit(kms);

enable and disable are pretty loaded. How about begin/end instead?

> +		mutex_unlock(&kms->commit_lock);
> +
> +		/*
> +		 * At this point, from drm core's perspective, we
> +		 * are done with the atomic update, so we can just
> +		 * go ahead and signal that it is done:
> +		 */
> +		drm_atomic_helper_commit_hw_done(state);
> +		drm_atomic_helper_cleanup_planes(dev, state);
> +
> +		return;
> +	}
> +
> +	/*
> +	 * If there is any async flush pending on updated crtcs, fold
> +	 * them into the current flush.
> +	 */
> +	kms->pending_crtc_mask &= ~crtc_mask;
> +
>  	/*
>  	 * Flush hardware updates:
>  	 */
> @@ -70,12 +218,18 @@ void msm_atomic_commit_tail(struct drm_atomic_state *state)
>  		kms->funcs->commit(kms, state);
>  	}
>  	kms->funcs->flush_commit(kms, crtc_mask);
> +	mutex_unlock(&kms->commit_lock);
>  
> +	/*
> +	 * Wait for flush to complete:
> +	 */
>  	kms->funcs->wait_flush(kms, crtc_mask);
> +
> +	mutex_lock(&kms->commit_lock);
>  	kms->funcs->complete_commit(kms, crtc_mask);
> +	mutex_unlock(&kms->commit_lock);
>  	kms->funcs->disable_commit(kms);
>  
>  	drm_atomic_helper_commit_hw_done(state);
> -
>  	drm_atomic_helper_cleanup_planes(dev, state);
>  }
> diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
> index 336a6d0a4cd3..65262a993440 100644
> --- a/drivers/gpu/drm/msm/msm_drv.c
> +++ b/drivers/gpu/drm/msm/msm_drv.c
> @@ -532,6 +532,7 @@ static int msm_drm_init(struct device *dev, struct drm_driver *drv)
>  	ddev->mode_config.normalize_zpos = true;
>  
>  	if (kms) {
> +		kms->dev = ddev;
>  		ret = kms->funcs->hw_init(kms);
>  		if (ret) {
>  			DRM_DEV_ERROR(dev, "kms hw init failed: %d\n", ret);
> diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
> index 79d480a7d97d..7d164d5c18b4 100644
> --- a/drivers/gpu/drm/msm/msm_drv.h
> +++ b/drivers/gpu/drm/msm/msm_drv.h
> @@ -222,8 +222,12 @@ struct msm_format {
>  	uint32_t pixel_format;
>  };
>  
> +struct msm_pending_timer;
> +
>  int msm_atomic_prepare_fb(struct drm_plane *plane,
>  			  struct drm_plane_state *new_state);
> +void msm_atomic_init_pending_timer(struct msm_pending_timer *timer,
> +		struct msm_kms *kms, int crtc_idx);
>  void msm_atomic_commit_tail(struct drm_atomic_state *state);
>  struct drm_atomic_state *msm_atomic_state_alloc(struct drm_device *dev);
>  void msm_atomic_state_clear(struct drm_atomic_state *state);
> diff --git a/drivers/gpu/drm/msm/msm_kms.h b/drivers/gpu/drm/msm/msm_kms.h
> index 811f5e2c2405..5eafc9686d29 100644
> --- a/drivers/gpu/drm/msm/msm_kms.h
> +++ b/drivers/gpu/drm/msm/msm_kms.h
> @@ -33,6 +33,20 @@ struct msm_kms_funcs {
>  
>  	/*
>  	 * Atomic commit handling:
> +	 *
> +	 * Note that in the case of async commits, the funcs which take
> +	 * a crtc_mask (ie. ->flush_commit(), and ->complete_commit())
> +	 * might not be evenly balanced with ->prepare_commit(), however
> +	 * each crtc that effected by a ->perpare_commit() (potentially

s/perpare/prepare/

> +	 * multiple times) will eventually (at end of vsync period) be
> +	 * flushed and completed.
> +	 *
> +	 * This has some implications about tracking of cleanup state,
> +	 * for example SMP blocks to release after commit completes.  Ie.
> +	 * cleanup state should be also duplicated in the various
> +	 * duplicate_state() methods, as the current cleanup state at
> +	 * ->complete_commit() time may have accumulated cleanup work
> +	 * from multiple commits.
>  	 */
>  
>  	/**
> @@ -45,6 +59,14 @@ struct msm_kms_funcs {
>  	void (*enable_commit)(struct msm_kms *kms);
>  	void (*disable_commit)(struct msm_kms *kms);
>  
> +	/**
> +	 * If the kms backend supports async commit, it should implement
> +	 * this method to return the time of the next vsync.  This is
> +	 * used to determine a time slightly before vsync, for the async
> +	 * commit timer to run and complete an async commit.
> +	 */
> +	ktime_t (*vsync_time)(struct msm_kms *kms, struct drm_crtc *crtc);
> +
>  	/**
>  	 * Prepare for atomic commit.  This is called after any previous
>  	 * (async or otherwise) commit has completed.
> @@ -109,20 +131,48 @@ struct msm_kms_funcs {
>  #endif
>  };
>  
> +struct msm_kms;
> +
> +/*
> + * A per-crtc timer for pending async atomic flushes.  Scheduled to expire
> + * shortly before vblank to flush pending async updates.
> + */
> +struct msm_pending_timer {
> +	struct hrtimer timer;
> +	struct work_struct work;
> +	struct msm_kms *kms;
> +	unsigned crtc_idx;
> +};
> +
>  struct msm_kms {
>  	const struct msm_kms_funcs *funcs;
> +	struct drm_device *dev;
>  
>  	/* irq number to be passed on to drm_irq_install */
>  	int irq;
>  
>  	/* mapper-id used to request GEM buffer mapped for scanout: */
>  	struct msm_gem_address_space *aspace;
> +
> +	/*
> +	 * For async commit, where ->flush_commit() and later happens
> +	 * from the crtc's pending_timer close to end of the frame:
> +	 */
> +	struct mutex commit_lock;
> +	unsigned pending_crtc_mask;
> +	struct msm_pending_timer pending_timers[MAX_CRTCS];
>  };
>  
>  static inline void msm_kms_init(struct msm_kms *kms,
>  		const struct msm_kms_funcs *funcs)
>  {
> +	unsigned i;
> +
> +	mutex_init(&kms->commit_lock);
>  	kms->funcs = funcs;
> +
> +	for (i = 0; i < ARRAY_SIZE(kms->pending_timers); i++)
> +		msm_atomic_init_pending_timer(&kms->pending_timers[i], kms, i);
>  }
>  
>  struct msm_kms *mdp4_kms_init(struct drm_device *dev);
> -- 
> 2.21.0
>
diff mbox series

Patch

diff --git a/drivers/gpu/drm/msm/msm_atomic.c b/drivers/gpu/drm/msm/msm_atomic.c
index 614fb9c5bb58..8f8f74337cb4 100644
--- a/drivers/gpu/drm/msm/msm_atomic.c
+++ b/drivers/gpu/drm/msm/msm_atomic.c
@@ -29,6 +29,95 @@  int msm_atomic_prepare_fb(struct drm_plane *plane,
 	return msm_framebuffer_prepare(new_state->fb, kms->aspace);
 }
 
+static void msm_atomic_async_commit(struct msm_kms *kms, int crtc_idx)
+{
+	unsigned crtc_mask = BIT(crtc_idx);
+
+	mutex_lock(&kms->commit_lock);
+
+	if (!(kms->pending_crtc_mask & crtc_mask)) {
+		mutex_unlock(&kms->commit_lock);
+		return;
+	}
+
+	kms->pending_crtc_mask &= ~crtc_mask;
+
+	kms->funcs->enable_commit(kms);
+
+	/*
+	 * Flush hardware updates:
+	 */
+	DRM_DEBUG_ATOMIC("triggering async commit\n");
+	kms->funcs->flush_commit(kms, crtc_mask);
+	mutex_unlock(&kms->commit_lock);
+
+	/*
+	 * Wait for flush to complete:
+	 */
+	kms->funcs->wait_flush(kms, crtc_mask);
+
+	mutex_lock(&kms->commit_lock);
+	kms->funcs->complete_commit(kms, crtc_mask);
+	mutex_unlock(&kms->commit_lock);
+	kms->funcs->disable_commit(kms);
+}
+
+static enum hrtimer_restart msm_atomic_pending_timer(struct hrtimer *t)
+{
+	struct msm_pending_timer *timer = container_of(t,
+			struct msm_pending_timer, timer);
+	struct msm_drm_private *priv = timer->kms->dev->dev_private;
+
+	queue_work(priv->wq, &timer->work);
+
+	return HRTIMER_NORESTART;
+}
+
+static void msm_atomic_pending_work(struct work_struct *work)
+{
+	struct msm_pending_timer *timer = container_of(work,
+			struct msm_pending_timer, work);
+
+	msm_atomic_async_commit(timer->kms, timer->crtc_idx);
+}
+
+void msm_atomic_init_pending_timer(struct msm_pending_timer *timer,
+		struct msm_kms *kms, int crtc_idx)
+{
+	timer->kms = kms;
+	timer->crtc_idx = crtc_idx;
+	hrtimer_init(&timer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
+	timer->timer.function = msm_atomic_pending_timer;
+	INIT_WORK(&timer->work, msm_atomic_pending_work);
+}
+
+static bool can_do_async(struct drm_atomic_state *state,
+		struct drm_crtc **async_crtc)
+{
+	struct drm_connector_state *connector_state;
+	struct drm_connector *connector;
+	struct drm_crtc_state *crtc_state;
+	struct drm_crtc *crtc;
+	int i, num_crtcs = 0;
+
+	if (!(state->legacy_cursor_update || state->async_update))
+		return false;
+
+	/* any connector change, means slow path: */
+	for_each_new_connector_in_state(state, connector, connector_state, i)
+		return false;
+
+	for_each_new_crtc_in_state(state, crtc, crtc_state, i) {
+		if (drm_atomic_crtc_needs_modeset(crtc_state))
+			return false;
+		if (++num_crtcs > 1)
+			return false;
+		*async_crtc = crtc;
+	}
+
+	return true;
+}
+
 /* Get bitmask of crtcs that will need to be flushed.  The bitmask
  * can be used with for_each_crtc_mask() iterator, to iterate
  * effected crtcs without needing to preserve the atomic state.
@@ -50,9 +139,25 @@  void msm_atomic_commit_tail(struct drm_atomic_state *state)
 	struct drm_device *dev = state->dev;
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_kms *kms = priv->kms;
+	struct drm_crtc *async_crtc = NULL;
 	unsigned crtc_mask = get_crtc_mask(state);
+	bool async = kms->funcs->vsync_time &&
+			can_do_async(state, &async_crtc);
 
 	kms->funcs->enable_commit(kms);
+
+	/*
+	 * Ensure any previous (potentially async) commit has
+	 * completed:
+	 */
+	kms->funcs->wait_flush(kms, crtc_mask);
+
+	mutex_lock(&kms->commit_lock);
+
+	/*
+	 * Now that there is no in-progress flush is complete,
+	 * prepare the current update:
+	 */
 	kms->funcs->prepare_commit(kms, state);
 
 	/*
@@ -62,6 +167,49 @@  void msm_atomic_commit_tail(struct drm_atomic_state *state)
 	drm_atomic_helper_commit_planes(dev, state, 0);
 	drm_atomic_helper_commit_modeset_enables(dev, state);
 
+	if (async) {
+		struct msm_pending_timer *timer =
+			&kms->pending_timers[drm_crtc_index(async_crtc)];
+
+		/* async updates are limited to single-crtc updates: */
+		WARN_ON(crtc_mask != drm_crtc_mask(async_crtc));
+
+		/*
+		 * Start timer if we don't already have an update pending
+		 * on this crtc:
+		 */
+		if (!(kms->pending_crtc_mask & crtc_mask)) {
+			ktime_t vsync_time, wakeup_time;
+
+			kms->pending_crtc_mask |= crtc_mask;
+
+			vsync_time = kms->funcs->vsync_time(kms, async_crtc);
+			wakeup_time = ktime_sub(vsync_time, ms_to_ktime(1));
+
+			hrtimer_start(&timer->timer, wakeup_time,
+					HRTIMER_MODE_ABS);
+		}
+
+		kms->funcs->disable_commit(kms);
+		mutex_unlock(&kms->commit_lock);
+
+		/*
+		 * At this point, from drm core's perspective, we
+		 * are done with the atomic update, so we can just
+		 * go ahead and signal that it is done:
+		 */
+		drm_atomic_helper_commit_hw_done(state);
+		drm_atomic_helper_cleanup_planes(dev, state);
+
+		return;
+	}
+
+	/*
+	 * If there is any async flush pending on updated crtcs, fold
+	 * them into the current flush.
+	 */
+	kms->pending_crtc_mask &= ~crtc_mask;
+
 	/*
 	 * Flush hardware updates:
 	 */
@@ -70,12 +218,18 @@  void msm_atomic_commit_tail(struct drm_atomic_state *state)
 		kms->funcs->commit(kms, state);
 	}
 	kms->funcs->flush_commit(kms, crtc_mask);
+	mutex_unlock(&kms->commit_lock);
 
+	/*
+	 * Wait for flush to complete:
+	 */
 	kms->funcs->wait_flush(kms, crtc_mask);
+
+	mutex_lock(&kms->commit_lock);
 	kms->funcs->complete_commit(kms, crtc_mask);
+	mutex_unlock(&kms->commit_lock);
 	kms->funcs->disable_commit(kms);
 
 	drm_atomic_helper_commit_hw_done(state);
-
 	drm_atomic_helper_cleanup_planes(dev, state);
 }
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index 336a6d0a4cd3..65262a993440 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -532,6 +532,7 @@  static int msm_drm_init(struct device *dev, struct drm_driver *drv)
 	ddev->mode_config.normalize_zpos = true;
 
 	if (kms) {
+		kms->dev = ddev;
 		ret = kms->funcs->hw_init(kms);
 		if (ret) {
 			DRM_DEV_ERROR(dev, "kms hw init failed: %d\n", ret);
diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index 79d480a7d97d..7d164d5c18b4 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -222,8 +222,12 @@  struct msm_format {
 	uint32_t pixel_format;
 };
 
+struct msm_pending_timer;
+
 int msm_atomic_prepare_fb(struct drm_plane *plane,
 			  struct drm_plane_state *new_state);
+void msm_atomic_init_pending_timer(struct msm_pending_timer *timer,
+		struct msm_kms *kms, int crtc_idx);
 void msm_atomic_commit_tail(struct drm_atomic_state *state);
 struct drm_atomic_state *msm_atomic_state_alloc(struct drm_device *dev);
 void msm_atomic_state_clear(struct drm_atomic_state *state);
diff --git a/drivers/gpu/drm/msm/msm_kms.h b/drivers/gpu/drm/msm/msm_kms.h
index 811f5e2c2405..5eafc9686d29 100644
--- a/drivers/gpu/drm/msm/msm_kms.h
+++ b/drivers/gpu/drm/msm/msm_kms.h
@@ -33,6 +33,20 @@  struct msm_kms_funcs {
 
 	/*
 	 * Atomic commit handling:
+	 *
+	 * Note that in the case of async commits, the funcs which take
+	 * a crtc_mask (ie. ->flush_commit(), and ->complete_commit())
+	 * might not be evenly balanced with ->prepare_commit(), however
+	 * each crtc that effected by a ->perpare_commit() (potentially
+	 * multiple times) will eventually (at end of vsync period) be
+	 * flushed and completed.
+	 *
+	 * This has some implications about tracking of cleanup state,
+	 * for example SMP blocks to release after commit completes.  Ie.
+	 * cleanup state should be also duplicated in the various
+	 * duplicate_state() methods, as the current cleanup state at
+	 * ->complete_commit() time may have accumulated cleanup work
+	 * from multiple commits.
 	 */
 
 	/**
@@ -45,6 +59,14 @@  struct msm_kms_funcs {
 	void (*enable_commit)(struct msm_kms *kms);
 	void (*disable_commit)(struct msm_kms *kms);
 
+	/**
+	 * If the kms backend supports async commit, it should implement
+	 * this method to return the time of the next vsync.  This is
+	 * used to determine a time slightly before vsync, for the async
+	 * commit timer to run and complete an async commit.
+	 */
+	ktime_t (*vsync_time)(struct msm_kms *kms, struct drm_crtc *crtc);
+
 	/**
 	 * Prepare for atomic commit.  This is called after any previous
 	 * (async or otherwise) commit has completed.
@@ -109,20 +131,48 @@  struct msm_kms_funcs {
 #endif
 };
 
+struct msm_kms;
+
+/*
+ * A per-crtc timer for pending async atomic flushes.  Scheduled to expire
+ * shortly before vblank to flush pending async updates.
+ */
+struct msm_pending_timer {
+	struct hrtimer timer;
+	struct work_struct work;
+	struct msm_kms *kms;
+	unsigned crtc_idx;
+};
+
 struct msm_kms {
 	const struct msm_kms_funcs *funcs;
+	struct drm_device *dev;
 
 	/* irq number to be passed on to drm_irq_install */
 	int irq;
 
 	/* mapper-id used to request GEM buffer mapped for scanout: */
 	struct msm_gem_address_space *aspace;
+
+	/*
+	 * For async commit, where ->flush_commit() and later happens
+	 * from the crtc's pending_timer close to end of the frame:
+	 */
+	struct mutex commit_lock;
+	unsigned pending_crtc_mask;
+	struct msm_pending_timer pending_timers[MAX_CRTCS];
 };
 
 static inline void msm_kms_init(struct msm_kms *kms,
 		const struct msm_kms_funcs *funcs)
 {
+	unsigned i;
+
+	mutex_init(&kms->commit_lock);
 	kms->funcs = funcs;
+
+	for (i = 0; i < ARRAY_SIZE(kms->pending_timers); i++)
+		msm_atomic_init_pending_timer(&kms->pending_timers[i], kms, i);
 }
 
 struct msm_kms *mdp4_kms_init(struct drm_device *dev);