Skip to contentSkip to conversation

drm/virtio: Fix possible memory allocation failure

PATCH In review Triet Hoang 2 participants 3 messages Aug 18

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/virtio: Fix possible memory allocation failure

When kmem_cache_zalloc in virtio_gpu_get_vbuf fails, it will return an error code. But none of its callers checks this error code, and a core dump will take place.

Add __GFP_NOFAIL flag when calling kmem_cache_zalloc to make sure it won't fail

Fixes 6f3d9d0dd335 ("drm/virtio: Add drm_panic support")
---
 drivers/gpu/drm/virtio/virtgpu_vq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/virtio/virtgpu_vq.c +1 -1
index e5e1af8b8e8a..fdf0f3c2d3b7 100644
@@ -93,7 +93,7 @@ virtio_gpu_panic_get_vbuf(struct virtio_gpu_device *vgdev, int size)
 {
 	struct virtio_gpu_vbuffer *vbuf;
 
-	vbuf = kmem_cache_zalloc(vgdev->vbufs, GFP_ATOMIC);
+	vbuf = kmem_cache_zalloc(vgdev->vbufs, GFP_ATOMIC | __GFP_NOFAIL);
 
 	vbuf->buf = (void *)vbuf + sizeof(*vbuf);
 	vbuf->size = size;
Signature · 1 line
2.53.0
#1 lore raw
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

New issues:
- [High] Adding __GFP_NOFAIL to a GFP_ATOMIC allocation is invalid and triggers a kernel warning

Pre-existing issues:
- [High] NULL pointer dereference remains unfixed and slab allocation in panic handler is fundamentally unsafe --

--- Patch [1]: [PATCH] drm/virtio: Fix possible memory allocation failure --- commit 2b29d148bc8be5eb73791a39e29cb1b504f8f37b
Author: Triet Hoang <[email protected]>

drm/virtio: Fix possible memory allocation failure

This patch attempts to fix a possible NULL pointer dereference in virtio_gpu_panic_get_vbuf() by adding the __GFP_NOFAIL flag to the kmem_cache_zalloc() call instead of checking the return value.

drivers/gpu/drm/virtio/virtgpu_vq.c virtio_gpu_panic_get_vbuf(struct virtio_gpu_device *vgdev, int size) · line 96

Quoted 10 lines
drivers/gpu/drm/virtio/virtgpu_vq.c +1 -1
index 905b1f42cd983..be051ac4bd78c 100644
@@ -93,7 +93,7 @@ virtio_gpu_panic_get_vbuf(struct virtio_gpu_device *vgdev, int size)
 {
 	struct virtio_gpu_vbuffer *vbuf;
 
-	vbuf = kmem_cache_zalloc(vgdev->vbufs, GFP_ATOMIC);
+	vbuf = kmem_cache_zalloc(vgdev->vbufs, GFP_ATOMIC | __GFP_NOFAIL);

[Severity: High]
Does this combination of flags work as intended? The page allocator explicitly prohibits combining __GFP_NOFAIL with non-blocking flags like GFP_ATOMIC.

The allocator's slow path checks for !can_direct_reclaim (which is true for GFP_ATOMIC), emits a WARN_ON_ONCE_GFP, and jumps to the failure path. This returns NULL anyway, failing to guarantee allocation success while introducing a new warning during a system panic.

drivers/gpu/drm/virtio/virtgpu_vq.c virtio_gpu_panic_get_vbuf(struct virtio_gpu_device *vgdev, int size) · line 98

Quoted 2 lines
drivers/gpu/drm/virtio/virtgpu_vq.c +0 -0
 	vbuf->buf = (void *)vbuf + sizeof(*vbuf);

[Severity: High]
This isn't a bug introduced by this patch, but since the allocator ignores __GFP_NOFAIL for atomic allocations, can kmem_cache_zalloc() still return NULL here?

If it does, this will dereference NULL and cause a double fault, breaking the panic handler.

Additionally, is it safe to call kmem_cache_zalloc() during a kernel panic? If a stopped CPU was holding a slab lock, wouldn't the panicking CPU spin forever attempting to acquire it, resulting in a hard system hang?

Quoted 1 line
drivers/gpu/drm/virtio/virtgpu_vq.c +0 -0
 	vbuf->size = size;
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1