Skip to contentSkip to conversation

drm/exynos: fix a possible null-pointer dereference due to data race in exynos_drm_crtc_atomic_disable()

PATCH Applied Tuo Li 3 participants 6 messages Jun 30, 2023 – Jul 13, 2023

Subsystems (3)
lore
Share & tools
lore

Tuo Li

author
#0 lore raw

PATCH 1/1 drm/exynos: fix a possible null-pointer dereference due to data race in exynos_drm_crtc_atomic_disable()

The variable crtc->state->event is often protected by the lock crtc->dev->event_lock when is accessed. However, it is accessed as a condition of an if statement in exynos_drm_crtc_atomic_disable() without holding the lock:

  if (crtc->state->event && !crtc->state->active)

However, if crtc->state->event is changed to NULL by another thread right after the conditions of the if statement is checked to be true, a null-pointer dereference can occur in drm_crtc_send_vblank_event():

  e->pipe = pipe;

To fix this possible null-pointer dereference caused by data race, the spin lock coverage is extended to protect the if statement as well as the function call to drm_crtc_send_vblank_event().

---
 drivers/gpu/drm/exynos/exynos_drm_crtc.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/gpu/drm/exynos/exynos_drm_crtc.c +2 -3
index 4153f302de7c..d19e796c2061 100644
@@ -39,13 +39,12 @@ static void exynos_drm_crtc_atomic_disable(struct drm_crtc *crtc,
 	if (exynos_crtc->ops->atomic_disable)
 		exynos_crtc->ops->atomic_disable(exynos_crtc);
 
+	spin_lock_irq(&crtc->dev->event_lock);
 	if (crtc->state->event && !crtc->state->active) {
-		spin_lock_irq(&crtc->dev->event_lock);
 		drm_crtc_send_vblank_event(crtc, crtc->state->event);
-		spin_unlock_irq(&crtc->dev->event_lock);
-
 		crtc->state->event = NULL;
 	}
+	spin_unlock_irq(&crtc->dev->event_lock);
 }
 
 static int exynos_crtc_atomic_check(struct drm_crtc *crtc,
Signature · 1 line
2.34.1
#1 lore raw
Quoted 11 lines · Tuo Li

The variable crtc->state->event is often protected by the lock crtc->dev->event_lock when is accessed. However, it is accessed as a condition of an if statement in exynos_drm_crtc_atomic_disable() without holding the lock:

  if (crtc->state->event && !crtc->state->active)

However, if crtc->state->event is changed to NULL by another thread right after the conditions of the if statement is checked to be true, a null-pointer dereference can occur in drm_crtc_send_vblank_event():

Quoted 7 lines
  e->pipe = pipe;

To fix this possible null-pointer dereference caused by data race, the spin lock coverage is extended to protect the if statement as well as the function call to drm_crtc_send_vblank_event().

I cannot find this report. This is an open source work and public collaboration. The "Reported-by" usually means that the issue was reported to us, in some way, usually in public. Can we see the report? Otherwise adding non-public, non-verifiable reports is useless and clutters our report-credit-system.

Best regards,
Krzysztof

author
#2 lore raw

Hello,

Thanks for your reply! The report is publicly available at
https://sites.google.com/view/basscheck/home. And this patch is from the 8th report on this website.

Thanks,
Tuo Li

Quoted 31 lines · Krzysztof Kozlowski
Quoted 7 lines · Tuo Li

The variable crtc->state->event is often protected by the lock crtc->dev->event_lock when is accessed. However, it is accessed as a condition of an if statement in exynos_drm_crtc_atomic_disable() without holding the lock:

  if (crtc->state->event && !crtc->state->active)
Quoted 8 lines

However, if crtc->state->event is changed to NULL by another thread right after the conditions of the if statement is checked to be true, a null-pointer dereference can occur in drm_crtc_send_vblank_event():

  e->pipe = pipe;

To fix this possible null-pointer dereference caused by data race, the spin lock coverage is extended to protect the if statement as well as

the

Quoted 3 lines

function call to drm_crtc_send_vblank_event().

I cannot find this report. This is an open source work and public collaboration. The "Reported-by" usually means that the issue was reported to us, in some way, usually in public. Can we see the report? Otherwise adding non-public, non-verifiable reports is useless and clutters our report-credit-system.

Best regards,
Krzysztof

Reviewed-by on v1 1/1
#4 lore raw
In reply to Tuo Li
Quoted 11 lines · Tuo Li

The variable crtc->state->event is often protected by the lock crtc->dev->event_lock when is accessed. However, it is accessed as a condition of an if statement in exynos_drm_crtc_atomic_disable() without holding the lock:

  if (crtc->state->event && !crtc->state->active)

However, if crtc->state->event is changed to NULL by another thread right after the conditions of the if statement is checked to be true, a null-pointer dereference can occur in drm_crtc_send_vblank_event():

Quoted 8 lines
  e->pipe = pipe;

To fix this possible null-pointer dereference caused by data race, the spin lock coverage is extended to protect the if statement as well as the function call to drm_crtc_send_vblank_event().

Best regards,
Krzysztof

#5 lore raw
In reply to Tuo Li
Quoted 26 lines

-----Original Message-----
From: Tuo Li <[email protected]>
Sent: Friday, June 30, 2023 11:19 AM
To: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Tuo Li <[email protected]>; BassCheck <[email protected]>
Subject: [PATCH] drm/exynos: fix a possible null-pointer dereference due to data race in exynos_drm_crtc_atomic_disable()

The variable crtc->state->event is often protected by the lock crtc->dev->event_lock when is accessed. However, it is accessed as a condition of an if statement in exynos_drm_crtc_atomic_disable() without holding the lock:

  if (crtc->state->event && !crtc->state->active)

However, if crtc->state->event is changed to NULL by another thread right after the conditions of the if statement is checked to be true, a null-pointer dereference can occur in drm_crtc_send_vblank_event():

Quoted 8 lines
  e->pipe = pipe;

To fix this possible null-pointer dereference caused by data race, the spin lock coverage is extended to protect the if statement as well as the function call to drm_crtc_send_vblank_event().

Applied.

Thanks,
Inki Dae

Quoted 28 lines
---
 drivers/gpu/drm/exynos/exynos_drm_crtc.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/gpu/drm/exynos/exynos_drm_crtc.c +2 -3
index 4153f302de7c..d19e796c2061 100644
@@ -39,13 +39,12 @@ static void exynos_drm_crtc_atomic_disable(struct
 drm_crtc *crtc,
 	if (exynos_crtc->ops->atomic_disable)
 		exynos_crtc->ops->atomic_disable(exynos_crtc);
 
+	spin_lock_irq(&crtc->dev->event_lock);
 	if (crtc->state->event && !crtc->state->active) {
-		spin_lock_irq(&crtc->dev->event_lock);
 		drm_crtc_send_vblank_event(crtc, crtc->state->event);
-		spin_unlock_irq(&crtc->dev->event_lock);
-
 		crtc->state->event = NULL;
 	}
+	spin_unlock_irq(&crtc->dev->event_lock);
 }
 

static int exynos_crtc_atomic_check(struct drm_crtc *crtc,
--
2.34.1