Re: [PATCH v2 04/15] drm/xe: Protect device I/O with DRM device SRCU

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] drm/xe: hardware forcewake drop outside IO protection in send_tlb_inval_ggtt()
--

commit e9f023cf84d9700d7a5ded726d95cff381241557
Author: Arvind Yadav <arvind.yadav@intel.com>

drm/xe: Protect device I/O with DRM device SRCU

This commit introduces wrappers around DRM device SRCU to protect device
I/O operations. It blocks new access during PCI recovery or when the
device is wedged, and drains existing hardware users to safely isolate
the device.

> diff --git a/drivers/gpu/drm/xe/xe_guc_tlb_inval.c b/drivers/gpu/drm/xe/xe_guc_tlb_inval.c
> --- a/drivers/gpu/drm/xe/xe_guc_tlb_inval.c
> +++ b/drivers/gpu/drm/xe/xe_guc_tlb_inval.c

[ ... ]

> @@ -86,8 +84,13 @@ static int send_tlb_inval_ggtt(struct xe_tlb_inval *tlb_inval, u32 seqno)
>  	} else if (xe_device_uc_enabled(xe)) {
>  		struct xe_mmio *mmio = &gt->mmio;
>  
> -		if (IS_SRIOV_VF(xe))
> +		if (xe_device_io_get(xe, &io_idx))
> +			return -ECANCELED;
> +
> +		if (IS_SRIOV_VF(xe)) {
> +			xe_device_io_put(io_idx);
>  			return -ECANCELED;
> +		}
>  
>  		CLASS(xe_force_wake, fw_ref)(gt_to_fw(gt), XE_FW_GT);
>  		if (xe->info.platform == XE_PVC || GRAPHICS_VER(xe) >= 20) {
> @@ -99,6 +102,8 @@ static int send_tlb_inval_ggtt(struct xe_tlb_inval *tlb_inval, u32 seqno)
>  			xe_mmio_write32(mmio, GUC_TLB_INV_CR,
>  					GUC_TLB_INV_CR_INVALIDATE);
>  		}
> +
> +		xe_device_io_put(io_idx);

[Severity: High]
Does this code drop the I/O protection before the scoped forcewake destructor
runs?

The CLASS(xe_force_wake, fw_ref) macro creates a scoped guard whose destructor
executes when leaving the block. Because xe_device_io_put(io_idx) is called
explicitly right before the closing brace, the I/O protection is released
before the forcewake destructor runs.

If PCI error recovery or wedge isolation disables the device immediately after
xe_device_io_put() returns, the automatically executing fw_ref destructor will
perform MMIO writes on a disabled device. This could trigger a PCI Advanced
Error Reporting (AER) fault and crash the system.

>  	}
>  
>  	return -ECANCELED;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260922101721.1583542-1-arvind.yadav@intel.com?part=4