drm/adp: Free FE IRQ on bind failure
Subsystems (3)
PATCH 1/1 drm/adp: Free FE IRQ on bind failure
adp_drm_bind() requests the front-end IRQ before registering the DRM device. If drm_dev_register() fails, the function returns immediately and leaves the IRQ handler registered.
Release the IRQ on the drm_dev_register() error path. The normal unbind path still frees the IRQ after successful registration.
--- drivers/gpu/drm/adp/adp_drv.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/gpu/drm/adp/adp_drv.c +5 -1
@@ -535,9 +535,13 @@ static int adp_drm_bind(struct device *dev) err =drm_dev_register (&adp->drm,0 );if (err) -return err; + goto err_free_irq;return 0 ; + +err_free_irq: +free_irq (adp->fe_irq, adp); +return err; }static void adp_drm_unbind (struct device *dev)
Signature · 1 line
2.47.1
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:
Pre-existing issues:
- [High] Calling drmm_* initializations inside the component bind callback leads to memory leaks and DRM structure corruption if bind fails and is retried.
- [High] Calling dma_alloc_coherent() with GFP_KERNEL in an atomic context.
- [High] Missing NULL check for the return value of dma_alloc_coherent().
- [High] adp_drm_gem_dumb_create modifies height after validation, causing a 32-bit integer overflow and undersized buffer allocation. --
commit 8a1bec26bfe0bb63ece93f656fc4dc72b509f897
Author: Myeonghun Pak <[email protected]>
drm/adp: Free FE IRQ on bind failure
adp_drm_bind() requests the front-end IRQ before registering the DRM device. If drm_dev_register() fails, the function returns immediately and leaves the IRQ handler registered.
Quoted 7 lines
drivers/gpu/drm/adp/adp_drv.c +0
-0@@ -533,9 +533,13 @@ static int adp_drm_bind(struct device *dev) err =drm_dev_register (&adp->drm,0 );
drivers/gpu/drm/adp/adp_drv.c static int adp_drm_bind(struct device *dev) · line 543
Quoted 10 lines
drivers/gpu/drm/adp/adp_drv.c +5
-1if (err) -return err; + goto err_free_irq;return 0 ; + +err_free_irq: +free_irq (adp->fe_irq, adp); +return err; }
[Severity: High]
This is a pre-existing issue, but does calling drmm_ initializations inside the component bind callback lead to memory leaks if bind fails here?
The drm_device is allocated in probe with devm_, so it persists across bind retries:
drivers/gpu/drm/adp/adp_drv.c:adp_probe() {
...
adp = devm_drm_dev_alloc(&pdev->dev, &adp_driver, struct adp_drv_private, drm);
...
}But adp_drm_bind() calls adp_setup_mode_config(), which initializes drmm_ resources:
drivers/gpu/drm/adp/adp_drv.c:adp_drm_bind() {
...
err = adp_setup_mode_config(adp);
...
}If a later step in bind fails (such as drm_dev_register()) and bind is retried, will re-running drmm_mode_config_init() on the same drm_device re-initialize dev->mode_config mutexes and IDRs without destroying the old ones, leading to memory leaks and corruption?
[Severity: High]
This is a pre-existing issue, but is it safe to use GFP_KERNEL in the atomic commit tail path?
drivers/gpu/drm/adp/adp_drv.c:adp_crtc_atomic_flush() {
...
if (new_size != 0) {
adp->mask_buf = dma_alloc_coherent(crtc->dev->dev, new_size,
&adp->mask_iova, GFP_KERNEL);
memset(adp->mask_buf, 0xFF, new_size);
writel(adp->mask_iova, adp->be + ADBE_MASK_BUF);
}
...
}Since adp_crtc_atomic_flush() is part of the atomic commit pipeline, it runs in an atomic context where sleeping is forbidden.
Does calling dma_alloc_coherent() with GFP_KERNEL here allow the function to sleep, causing kernel warnings or deadlocks?
[Severity: High]
This is a pre-existing issue, but is there a missing NULL check for the return value of dma_alloc_coherent() here?
drivers/gpu/drm/adp/adp_drv.c:adp_crtc_atomic_flush() {
...
adp->mask_buf = dma_alloc_coherent(crtc->dev->dev, new_size,
&adp->mask_iova, GFP_KERNEL);
memset(adp->mask_buf, 0xFF, new_size);
...
}If memory is exhausted and dma_alloc_coherent() returns NULL, will the immediate memset() cause a NULL pointer dereference?
[Severity: High]
This is a pre-existing issue, but does aligning args->height after validation cause an integer overflow?
drivers/gpu/drm/adp/adp_drv.c:adp_drm_gem_dumb_create() {
...
args->height = ALIGN(args->height, 64);
args->size = args->pitch * args->height;
return drm_gem_dma_dumb_create_internal(file_priv, drm, args);
}If userspace requests dimensions that pass the core drm_mode_create_dumb() checks but overflow when args->height is increased, will the 32-bit multiplication in drm_gem_dma_dumb_create_internal() wrap around?
Could this result in an undersized DMA buffer allocation that allows out-of-bounds memory accesses when userspace maps and writes to it?
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1