Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1531750 > unrolled thread

[PATCH] drm/vc4: Fix race between page flip completion event and clean-up

Back to article view | Back to linux.kernel

Contents

  [PATCH] drm/vc4: Fix race between page flip completion event and clean-up Eric Anholt <eric@anholt.net> - 2016-11-28 23:40 +0100
    Re: [PATCH] drm/vc4: Fix race between page flip completion event and clean-up Daniel Stone <daniel@fooishbar.org> - 2016-11-29 11:00 +0100

#1531750 — [PATCH] drm/vc4: Fix race between page flip completion event and clean-up

From: Derek Foreman <derekf@osg.samsung.com>

There was a small window where a userspace program could submit
a pageflip after receiving a pageflip completion event yet still
receive EBUSY.

Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
---
 drivers/gpu/drm/vc4/vc4_crtc.c |  8 ++++++++
 drivers/gpu/drm/vc4/vc4_drv.h  |  1 +
 drivers/gpu/drm/vc4/vc4_kms.c  | 33 +++++++++++++++++++++++++--------
 3 files changed, 34 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/vc4/vc4_crtc.c b/drivers/gpu/drm/vc4/vc4_crtc.c
index 7f08d681a74b..82f914af4156 100644
--- a/drivers/gpu/drm/vc4/vc4_crtc.c
+++ b/drivers/gpu/drm/vc4/vc4_crtc.c
@@ -669,6 +669,14 @@ void vc4_disable_vblank(struct drm_device *dev, unsigned int crtc_id)
 	CRTC_WRITE(PV_INTEN, 0);
 }
 
+/* Must be called with the event lock held */
+bool vc4_event_pending(struct drm_crtc *crtc)
+{
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+
+	return !!vc4_crtc->event;
+}
+
 static void vc4_crtc_handle_page_flip(struct vc4_crtc *vc4_crtc)
 {
 	struct drm_crtc *crtc = &vc4_crtc->base;
diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h
index fef172804345..b3064e2e79c1 100644
--- a/drivers/gpu/drm/vc4/vc4_drv.h
+++ b/drivers/gpu/drm/vc4/vc4_drv.h
@@ -442,6 +442,7 @@ int vc4_bo_stats_debugfs(struct seq_file *m, void *arg);
 extern struct platform_driver vc4_crtc_driver;
 int vc4_enable_vblank(struct drm_device *dev, unsigned int crtc_id);
 void vc4_disable_vblank(struct drm_device *dev, unsigned int crtc_id);
+bool vc4_event_pending(struct drm_crtc *crtc);
 int vc4_crtc_debugfs_regs(struct seq_file *m, void *arg);
 int vc4_crtc_get_scanoutpos(struct drm_device *dev, unsigned int crtc_id,
 			    unsigned int flags, int *vpos, int *hpos,
diff --git a/drivers/gpu/drm/vc4/vc4_kms.c b/drivers/gpu/drm/vc4/vc4_kms.c
index c1f65c6c8e60..67af2af70af0 100644
--- a/drivers/gpu/drm/vc4/vc4_kms.c
+++ b/drivers/gpu/drm/vc4/vc4_kms.c
@@ -119,17 +119,34 @@ static int vc4_atomic_commit(struct drm_device *dev,
 
 	/* Make sure that any outstanding modesets have finished. */
 	if (nonblock) {
-		ret = down_trylock(&vc4->async_modeset);
-		if (ret) {
+		struct drm_crtc *crtc;
+		struct drm_crtc_state *crtc_state;
+		unsigned long flags;
+		bool busy = false;
+
+		/*
+		 * If there's an undispatched event to send then we're
+		 * obviously still busy.  If there isn't, then we can
+		 * unconditionally wait for the semaphore because it
+		 * shouldn't be contended (for long).
+		 *
+		 * This is to prevent a race where queuing a new flip
+		 * from userspace immediately on receipt of an event
+		 * beats our clean-up and returns EBUSY.
+		 */
+		spin_lock_irqsave(&dev->event_lock, flags);
+		for_each_crtc_in_state(state, crtc, crtc_state, i)
+			busy |= vc4_event_pending(crtc);
+		spin_unlock_irqrestore(&dev->event_lock, flags);
+		if (busy) {
 			kfree(c);
 			return -EBUSY;
 		}
-	} else {
-		ret = down_interruptible(&vc4->async_modeset);
-		if (ret) {
-			kfree(c);
-			return ret;
-		}
+	}
+	ret = down_interruptible(&vc4->async_modeset);
+	if (ret) {
+		kfree(c);
+		return ret;
 	}
 
 	ret = drm_atomic_helper_prepare_planes(dev, state);
-- 
2.10.2

[toc] | [next] | [standalone]

#1532071

On 28 November 2016 at 22:31, Eric Anholt <eric@anholt.net> wrote:
> There was a small window where a userspace program could submit
> a pageflip after receiving a pageflip completion event yet still
> receive EBUSY.
>
> Signed-off-by: Derek Foreman <derekf@osg.samsung.com>
> Signed-off-by: Eric Anholt <eric@anholt.net>
> Reviewed-by: Eric Anholt <eric@anholt.net>

Reviewed-by: Daniel Stone <daniels@collabora.com>

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web