An atomicity violation occurs when the vc4_crtc_send_vblank function executes simultaneously with modifications to crtc->state->event. Consider a scenario where crtc->state->event is non-null, allowing it to pass the validity check. However, at the same time, crtc->state->event might be set to null. In this case, the validity check in vc4_crtc_send_vblank might act on the old crtc->state->event (before locking), allowing invalid values to pass the validity check, which could lead to a null pointer dereference.
In the drm_device structure, it is mentioned: "@event_lock: Protects @vblank_event_list and event delivery in general." I believe that the validity check and the subsequent null assignment operation are part of the event delivery process, and all of these should be protected by the event_lock. If there is no lock protection before the validity check, it is possible for a null crtc->state->event to be passed into the drm_crtc_send_vblank_event() function, leading to a null pointer dereference error.
We have observed its callers and found that they are from the drm_crtc_helper_funcs driver interface. We believe that functions within driver interfaces can be concurrent, potentially causing a data race on crtc->state->event.
To address this issue, it is recommended to include the validity check of crtc->state and crtc->state->event within the locking section of the function. This modification ensures that the values of crtc->state->event and crtc->state do not change during the validation process, maintaining their valid conditions.
This possible bug is found by an experimental static analysis tool developed by our team. This tool analyzes the locking APIs to extract function pairs that can be concurrently executed, and then analyzes the instructions in the paired functions to identify possible concurrency bugs including data races and atomicity violations.
Fixes: 68e4a69aec4d ("drm/vc4: crtc: Create vblank reporting function") Cc: stable@vger.kernel.org Signed-off-by: Qiu-ji Chen chenqiuji666@gmail.com --- V2: The description of the patch has been modified to make it clearer. Thanks to Simona Vetter for suggesting this improvement. --- drivers/gpu/drm/vc4/vc4_crtc.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/vc4/vc4_crtc.c b/drivers/gpu/drm/vc4/vc4_crtc.c index 8b5a7e5eb146..98885f519827 100644 --- a/drivers/gpu/drm/vc4/vc4_crtc.c +++ b/drivers/gpu/drm/vc4/vc4_crtc.c @@ -575,10 +575,12 @@ void vc4_crtc_send_vblank(struct drm_crtc *crtc) struct drm_device *dev = crtc->dev; unsigned long flags;
- if (!crtc->state || !crtc->state->event) + spin_lock_irqsave(&dev->event_lock, flags); + if (!crtc->state || !crtc->state->event) { + spin_unlock_irqrestore(&dev->event_lock, flags); return; + }
- spin_lock_irqsave(&dev->event_lock, flags); drm_crtc_send_vblank_event(crtc, crtc->state->event); crtc->state->event = NULL; spin_unlock_irqrestore(&dev->event_lock, flags);