Re: [PATCH v2] drm/xe/xe_pci_error: Wait for pcode init post SBR

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] drm/xe: xe_wedge_declare() skipped in xe_pci_error_slot_reset()
--

commit d17c5826371a0496850ef94bd4ecf05d8bbcfe46
Author: Riana Tauro <riana.tauro@intel.com>

drm/xe/xe_pci_error: Wait for pcode init post SBR

This patch waits for Pcode initialization to complete after a Secondary Bus
Reset (SBR) in xe_pci_error_slot_reset() and enters runtime survivability
mode if initialization fails.

> diff --git a/drivers/gpu/drm/xe/xe_pci_error.c b/drivers/gpu/drm/xe/xe_pci_error.c
> index 9bf0da3a3d152..0675b5a8f6437 100644
> --- a/drivers/gpu/drm/xe/xe_pci_error.c
> +++ b/drivers/gpu/drm/xe/xe_pci_error.c

[ ... ]

> @@ -112,6 +113,12 @@ static pci_ers_result_t xe_pci_error_slot_reset(struct pci_dev *pdev)
>  		return PCI_ERS_RESULT_DISCONNECT;
>  	}
>  
> +	err = xe_pcode_ready(xe, true);
> +	if (err) {
> +		xe_survivability_mode_runtime_enable(xe);
> +		return PCI_ERS_RESULT_DISCONNECT;
> +	}

[Severity: Medium]
Prior to xe_pci_error_slot_reset() running, prepare_device_for_reset() is
called during PCIe error detection to block userspace access while the reset
is in progress:

drivers/gpu/drm/xe/xe_pci_error.c:prepare_device_for_reset() {
	...
	if (!atomic_xchg(&xe->wedged.flag, 1))
		xe_pm_runtime_get_noresume(xe);
	...
}

When xe_pcode_ready() fails in xe_pci_error_slot_reset(), the driver has not
yet re-probed the device, so xe_survivability_mode_runtime_enable() is called
on the existing xe_device instance where xe->wedged.flag is still 1:

xe_pci_error_slot_reset()
  -> xe_survivability_mode_runtime_enable()
       -> xe_wedge_declare(xe, DRM_WEDGE_RECOVERY_VENDOR)

Because xe->wedged.flag is already set, xe_wedge_declare() returns early:

drivers/gpu/drm/xe/xe_wedge.c:xe_wedge_declare() {
	...
	if (atomic_xchg(&xe->wedged.flag, 1))
		return;

	xe->needs_flr_on_fini = true;
	xe_pm_runtime_get_noresume(xe);
	...
	if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
		xe->wedged.method = DRM_WEDGE_RECOVERY_NONE;
	else
		xe->wedged.method = method;

	schedule_work(&xe->wedged.work);
}

Will this cause xe_wedge_declare() to skip setting xe->wedged.method to
DRM_WEDGE_RECOVERY_VENDOR, skip setting xe->needs_flr_on_fini, and skip
scheduling xe->wedged.work to notify userspace via the DRM wedged uevent?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261028072803.937955-2-riana.tauro@intel.com?part=1