Re: [PATCH v2] drm/xe: Add wa_14028434957 to Graphics Version 35.11

["Gupta, Varun" <varun.gupta@intel.com>]

On 12-Oct-26 10:45 AM, Tauro, Riana wrote:
>
> On 12-10-2026 10:42, Upadhyay, Tejas wrote:
>>
>>> -----Original Message-----
>>> From: Gupta, Varun <varun.gupta@intel.com>
>>> Sent: 09 October 2026 13:22
>>> To: intel-xe@lists.freedesktop.org
>>> Cc: Roper, Matthew D <matthew.d.roper@intel.com>; Upadhyay, Tejas
>>> <tejas.upadhyay@intel.com>; Vivi, Rodrigo <rodrigo.vivi@intel.com>
>>> Subject: [PATCH v2] drm/xe: Add wa_14028434957 to Graphics Version
>>> 35.11
>
> This is already sent long back
>
> Intel Xe - Patchwork
>
> <https://patchwork.freedesktop.org/series/173001/>Why are we 
> refloating this ?
>
> Thanks
> Riana
>
Hi Riana,

I was not aware of this series, I was only tracking the workaround ID.
Since your series is already underway, I will drop this patch and wait 
for your
implementation to merge, and if explicit tracking for the WA is still 
required,
I can send a follow-up patch to add the WA tag on top of your changes.

Thanks,
Varun
>>>
>>> Prevent the SNIB AER latch from swallowing back-to-back 
>>> uncorrectable non-
>>> fatal errors.
>>>
>>> Firmware bypasses the faulty AER path by routing these errors 
>>> directly via
>>> MSI. This patch intercepts the MSI and reuses the existing GFSP error
>>> containment logic.
>>>
>>> Because GFSP mailbox queries can sleep, handling is deferred to a 
>>> workqueue
>>> (nonfatal_msi_work) to ensure hardirq safety. If a reset is 
>>> required, the device
>>> is wedged (a stopgap until VR driver quiescing support allows for an 
>>> actual
>>> Secondary Bus Reset).
>>>
>>> v2:
>>>   - Return 'int' from xe_hw_error_init() to propagate 
>>> failures.(Sashiko)
>>>   - Use devm_add_action_or_reset() for teardown to prevent MMIO faults
>>>     during PCI unbind.(Sashiko)
>>>   - Add xe_pm_runtime guard in the worker thread.(Sashiko)
>>>
>>> Assisted-by: LLM
>>> Signed-off-by: Varun Gupta <varun.gupta@intel.com>
>>> ---
>>>   drivers/gpu/drm/xe/xe_device_wa_oob.rules |  2 +
>>>   drivers/gpu/drm/xe/xe_drm_ras_types.h     |  5 +++
>>>   drivers/gpu/drm/xe/xe_hw_error.c          | 52 
>>> ++++++++++++++++++++++-
>>>   drivers/gpu/drm/xe/xe_hw_error.h          |  2 +-
>>>   drivers/gpu/drm/xe/xe_irq.c               |  4 +-
>>>   5 files changed, 61 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/xe/xe_device_wa_oob.rules
>>> b/drivers/gpu/drm/xe/xe_device_wa_oob.rules
>>> index d8dc41851425..29a7d96a5666 100644
>>> --- a/drivers/gpu/drm/xe/xe_device_wa_oob.rules
>>> +++ b/drivers/gpu/drm/xe/xe_device_wa_oob.rules
>>> @@ -7,3 +7,5 @@
>>>   14026633728    PLATFORM(CRESCENTISLAND)
>>>   14026746987    PLATFORM(CRESCENTISLAND)
>>>   14026779378    PLATFORM(CRESCENTISLAND)
>>> +14028434957    PLATFORM(CRESCENTISLAND)
>>> +
>>> diff --git a/drivers/gpu/drm/xe/xe_drm_ras_types.h
>>> b/drivers/gpu/drm/xe/xe_drm_ras_types.h
>>> index 0be218ba2db7..7a35390a9ffd 100644
>>> --- a/drivers/gpu/drm/xe/xe_drm_ras_types.h
>>> +++ b/drivers/gpu/drm/xe/xe_drm_ras_types.h
>>> @@ -9,6 +9,8 @@
>>>   #include <linux/atomic.h>
>>>   #include <drm/xe_drm.h>
>>>
>>> +#include <linux/workqueue.h>
>>> +
>>>   struct drm_ras_node;
>>>
>>>   /* Error categories reported by hardware */ @@ -46,6 +48,9 @@ struct
>>> xe_drm_ras {
>>>
>>>       /** @disable_vram_page_offline: cached configfs policy, immutable
>>> after init */
>>>       bool disable_vram_page_offline;
>>> +
>>> +    /** @nonfatal_msi_work: Worker to process uncorrectable non-fatal
>>> errors */
>>> +    struct work_struct nonfatal_msi_work;
>>>   };
>>>
>>>   #endif
>>> diff --git a/drivers/gpu/drm/xe/xe_hw_error.c
>>> b/drivers/gpu/drm/xe/xe_hw_error.c
>>> index 5f2abc9485ff..bde707a3fb64 100644
>>> --- a/drivers/gpu/drm/xe/xe_hw_error.c
>>> +++ b/drivers/gpu/drm/xe/xe_hw_error.c
>>> @@ -5,6 +5,8 @@
>>>
>>>   #include <linux/bitmap.h>
>>>
>>> +#include <generated/xe_device_wa_oob.h>
>>> +
>>>   #include "regs/xe_gsc_regs.h"
>>>   #include "regs/xe_hw_error_regs.h"
>>>   #include "regs/xe_irq_regs.h"
>>> @@ -14,7 +16,10 @@
>>>   #include "xe_drm_ras.h"
>>>   #include "xe_hw_error.h"
>>>   #include "xe_mmio.h"
>>> +#include "xe_pm.h"
>>> +#include "xe_ras.h"
>>>   #include "xe_survivability_mode.h"
>>> +#include "xe_wa.h"
>>>
>>>   #define GT_HW_ERROR_MAX_ERR_BITS        16
>>>   #define HEC_UNCORR_FW_ERR_BITS            4
>>> @@ -173,6 +178,29 @@ static void csc_hw_error_work(struct work_struct
>>> *work)
>>>       xe_survivability_mode_runtime_enable(xe);
>>>   }
>>>
>>> +static void nonfatal_msi_work(struct work_struct *work) {
>>> +    struct xe_drm_ras *ras = container_of(work, typeof(*ras),
>>> nonfatal_msi_work);
>>> +    struct xe_device *xe = container_of(ras, typeof(*xe), ras);
>>> +
>>> +    guard(xe_pm_runtime)(xe);
>>> +    enum xe_ras_recovery_action action = xe_ras_process_errors(xe);
>>> +
>>> +    /*
>>> +     * TODO: Request an SBR instead of wedging, once VR driver
>>> +     * quiescing support is in place.
>>> +     */
>>> +    if (action != XE_RAS_RECOVERY_ACTION_RECOVERED)
>>> +        xe_device_declare_wedged(xe);
>>> +}
>>> +
>>> +static void hw_error_fini(void *arg)
>>> +{
>>> +    struct xe_device *xe = arg;
>>> +
>>> +    cancel_work_sync(&xe->ras.nonfatal_msi_work);
>>> +}
>>> +
>>>   static void csc_hw_error_handler(struct xe_tile *tile, const enum
>>> hardware_error hw_err)  {
>>>       const enum drm_xe_ras_error_severity severity =
>>> hw_err_to_severity(hw_err); @@ -436,6 +464,15 @@ static void
>>> hw_error_source_handler(struct xe_tile *tile, const enum hardware_er
>>>        * here for those platforms.
>>>        */
>>>       if (xe->info.has_sysctrl) {
>>> +        /*
>>> +         * Defer non-fatal MSI handling to process context because the
>>> GFSP
>>> +         * query can sleep.
>>> +         */
>>> +        if (hw_err == HARDWARE_ERROR_NONFATAL &&
>>> XE_DEVICE_WA(xe, 14028434957)) {
>>> +            schedule_work(&xe->ras.nonfatal_msi_work);
>>> +            return;
>>> +        }
>>> +
>>>           drm_err_ratelimited(&xe->drm, HW_ERR "Invalid error
>>> routing\n");
>>>           return;
>>>       }
>>> @@ -541,15 +578,26 @@ static void process_hw_errors(struct xe_device
>>> *xe)
>>>    *
>>>    * Initialize and check for errors that occurred during boot
>>>    * prior to driver load
>>> + *
>>> + * Return: 0 on success, negative error code on failure.
>>>    */
>>> -void xe_hw_error_init(struct xe_device *xe)
>>> +int xe_hw_error_init(struct xe_device *xe)
>>>   {
>>>       struct xe_tile *tile = xe_device_get_root_tile(xe);
>>> +    int err;
>>> +
>>> +    INIT_WORK(&xe->ras.nonfatal_msi_work, nonfatal_msi_work);
>>> +
>>> +    err = devm_add_action_or_reset(xe->drm.dev, hw_error_fini, xe);
>>> +    if (err)
>>> +        return err;
>>>
>>>       if (!IS_DGFX(xe) || IS_SRIOV_VF(xe))
>>> -        return;
>>> +        return 0;
>> Should this not be first check in this init? Everything that 
>> follows.  With that addressed, LGTM,
>> Reviewed-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
>>
>> @Tauro, Riana do you see any concerns? I think for now it is planned 
>> to handle this way!
>>
>> Tejas
>>> INIT_WORK(&tile->csc_hw_error_work, csc_hw_error_work);
>>>
>>>       process_hw_errors(xe);
>>> +
>>> +    return 0;
>>>   }
>>> diff --git a/drivers/gpu/drm/xe/xe_hw_error.h
>>> b/drivers/gpu/drm/xe/xe_hw_error.h
>>> index 5e3a11424108..a9af4c947d77 100644
>>> --- a/drivers/gpu/drm/xe/xe_hw_error.h
>>> +++ b/drivers/gpu/drm/xe/xe_hw_error.h
>>> @@ -11,5 +11,5 @@ struct xe_tile;
>>>   struct xe_device;
>>>
>>>   void xe_hw_error_irq_handler(struct xe_tile *tile, const u32 
>>> master_ctl); -void
>>> xe_hw_error_init(struct xe_device *xe);
>>> +int xe_hw_error_init(struct xe_device *xe);
>>>   #endif
>>> diff --git a/drivers/gpu/drm/xe/xe_irq.c 
>>> b/drivers/gpu/drm/xe/xe_irq.c index
>>> 9e49e2241da4..5ea50d203db1 100644
>>> --- a/drivers/gpu/drm/xe/xe_irq.c
>>> +++ b/drivers/gpu/drm/xe/xe_irq.c
>>> @@ -811,7 +811,9 @@ int xe_irq_install(struct xe_device *xe)
>>>       int nvec = 1;
>>>       int err;
>>>
>>> -    xe_hw_error_init(xe);
>>> +    err = xe_hw_error_init(xe);
>>> +    if (err)
>>> +        return err;
>>>
>>>       xe_irq_reset(xe);
>>>
>>> -- 
>>> 2.43.0