Re: [PATCH v6 12/16] drm/msm/gem: Validate lazy VM in GEM_NEW

[sashiko-bot@kernel.org]

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

New issues:
- [Medium] The check `!vm` fails to detect shared VMs, breaking the `MSM_BO_NO_SHARE` contract.

Pre-existing issues:
- [Critical] NULL Pointer Dereference when accessing uninitialized `ctx->vm` in `msm_ioctl_gem_submit`.
--

--- Patch [12]: [PATCH v6 12/16] drm/msm/gem: Validate lazy VM in GEM_NEW ---
commit 2c5a1210206f5856b0becc6857b23e80e4e3fe79
Author: Rob Clark <robin.clark@oss.qualcomm.com>

drm/msm/gem: Validate lazy VM in GEM_NEW

This commit adds a check in msm_gem_new_handle() to ensure the VM is
available before creating a BO with MSM_BO_NO_SHARE. It prevents a NULL
pointer dereference that could occur if a BO is created before any BOs have
been mapped.

> diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
> index 12d9b8a9c360a..b9da5b7ce8a31 100644
> --- a/drivers/gpu/drm/msm/msm_gem.c
> +++ b/drivers/gpu/drm/msm/msm_gem.c
> @@ -1139,9 +1139,13 @@ int msm_gem_new_handle(struct drm_device *dev, struct drm_file *file,
>  	int ret;
>  
>  	if (flags & MSM_BO_NO_SHARE) {
> +		struct msm_drm_private *priv = dev->dev_private;
>  		struct msm_context *ctx = file->driver_priv;
>  		struct drm_gpuvm *vm = msm_context_vm(dev, ctx);
>  
> +		if (!priv->gpu || !vm)
> +			return UERR(EINVAL, dev, "not supported with shared VM");
> +

[Severity: Medium]
Does the check for !vm correctly detect if a shared VM is in use?

If a platform does not support private VMs and falls back to the global
VM, msm_context_vm() will return the shared gpu->vm. In this case, !vm
will evaluate to false, allowing the allocation to proceed.

Could this allow the MSM_BO_NO_SHARE flag to succeed while silently sharing
the global reservation object, defeating the purpose of the flag and
creating hidden lock contention?

Would it be better to check if vm == priv->gpu->vm to ensure a private
VM is actually in use?

>  		r_obj = drm_gpuvm_resv_obj(vm);
>  	}
>  

[Severity: Critical]
This is a pre-existing issue, but does msm_ioctl_gem_submit() have a
similar problem with lazy VM initialization?

Looking at msm_ioctl_gem_submit() in drivers/gpu/drm/msm/msm_gem_submit.c:

	if (!gpu)
		return -ENXIO;

	if (args->pad)
		return -EINVAL;

	if (to_msm_vm(ctx->vm)->unusable)
		return UERR(EPIPE, dev, "context is unusable");

If a userspace process calls the DRM_IOCTL_MSM_GEM_SUBMIT ioctl immediately
after opening the DRM device, ctx->vm will be NULL because no previous
operation has initialized it.

Since this code accesses ctx->vm directly instead of using msm_context_vm(),
will this cause a NULL pointer dereference when to_msm_vm() evaluates to NULL
and ->unusable is accessed?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260717204841.26120-1-robin.clark@oss.qualcomm.com?part=12