Thank you for your contribution! Sashiko AI review found 1 potential issue(s) 
to consider:

Pre-existing issues:
- [High] Kernel panics in common KMS teardown path due to unconditional 
dereference of partially-initialized structures on early probe failures.
--

--- Patch [1]: [PATCH] drm/msm: don't tear down KMS twice when KMS init fails 
---
commit dccb630956dd40369b4b75f78b147548c8331788
Author: Dmitry Baryshkov <[email protected]>

drm/msm: don't tear down KMS twice when KMS init fails

This commit removes the local KMS teardown from the mdp4 and mdp5 error
paths. It instead relies on the common msm_drm_kms_uninit() path to handle
cleanup on failure, preventing a use-after-free crash.

> diff --git a/drivers/gpu/drm/msm/disp/mdp4/mdp4_kms.c 
> b/drivers/gpu/drm/msm/disp/mdp4/mdp4_kms.c
> --- a/drivers/gpu/drm/msm/disp/mdp4/mdp4_kms.c
> +++ b/drivers/gpu/drm/msm/disp/mdp4/mdp4_kms.c
> @@ -398,7 +398,7 @@ static int mdp4_kms_init(struct drm_device *dev)
>       ret = mdp_kms_init(&mdp4_kms->base, &kms_funcs);
>       if (ret) {
>               DRM_DEV_ERROR(dev->dev, "failed to init kms\n");
> -             goto fail;
> +             return ret;
>       }

[Severity: High]
This is a pre-existing issue, but by delegating all cleanup to the common
path on early failures, do we risk kernel panics from dereferencing
uninitialized pointers?

If mdp_kms_init() fails here, msm_drm_kms_uninit() will be called to clean up.
However, if kms->wq was never allocated, won't flush_workqueue(kms->wq)
trigger a panic inside __flush_workqueue() when it attempts to dereference
the workqueue?

Similarly, if initialization fails before kms->funcs is populated,
could msm_irq_uninstall() trigger a NULL pointer dereference when executing
kms->funcs->irq_uninstall(kms)?

Additionally, when kms->funcs->destroy(kms) eventually calls mdp4_destroy(),
would this unconditionally dereference mdp4_kms->dev->dev, where
mdp4_kms->dev remains NULL because mdp_kms_init() failed?

-- 
Sashiko AI review ยท 
https://sashiko.dev/#/patchset/[email protected]?part=1

Reply via email to