[PATCH v4 5/9] drm/xe/xe_survivability: Add support for Runtime survivability mode

[Riana Tauro]

Hi Umesh

On 7/10/2025 10:42 PM, Umesh Nerlige Ramappa wrote:
> On Thu, Jul 10, 2025 at 11:29:44AM +0530, Riana Tauro wrote:
>> Hi Umesh
>>
>> On 7/10/2025 5:14 AM, Umesh Nerlige Ramappa wrote:
>>> On Wed, Jul 09, 2025 at 04:50:17PM +0530, Riana Tauro wrote:
>>>> Certain runtime firmware errors can cause the device to be in a 
>>>> unusable
>>>> state requiring a firmware flash to restore normal operation.
>>>> Runtime Survivability Mode indicates firmware flash is necessary by
>>>> wedging the device and exposing survivability mode sysfs.
>>>>
>>>> The below sysfs is an indication that device is in survivability mode
>>>>
>>>> /sys/bus/pci/devices/<device>/survivability_mode
>>>>
>>>> Signed-off-by: Riana Tauro <riana.tauro at intel.com>
>>>> ---
>>>> drivers/gpu/drm/xe/xe_survivability_mode.c    | 42 ++++++++++++++++++-
>>>> drivers/gpu/drm/xe/xe_survivability_mode.h    |  1 +
>>>> .../gpu/drm/xe/xe_survivability_mode_types.h  |  1 +
>>>> 3 files changed, 43 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/xe/xe_survivability_mode.c b/drivers/ 
>>>> gpu/ drm/xe/xe_survivability_mode.c
>>>> index fefb027b1c84..ca1cfa13525a 100644
>>>> --- a/drivers/gpu/drm/xe/xe_survivability_mode.c
>>>> +++ b/drivers/gpu/drm/xe/xe_survivability_mode.c
>>>> @@ -137,7 +137,8 @@ static ssize_t survivability_mode_show(struct 
>>>> device *dev,
>>>>     struct xe_survivability_info *info = survivability->info;
>>>>     int index = 0, count = 0;
>>>>
>>>> -    count += sysfs_emit_at(buff, count, "Survivability mode type: 
>>>> Boot\n");
>>>> +    count += sysfs_emit_at(buff, count, "Survivability mode type: 
>>>> %s\n",
>>>> +                   survivability->type ? "Runtime" : "Boot");
>>>>
>>>>     if (!check_boot_failure(xe))
>>>>         return count;
>>>> @@ -288,6 +289,45 @@ bool xe_survivability_mode_is_requested(struct 
>>>> xe_device *xe)
>>>>     return check_boot_failure(xe);
>>>> }
>>>>
>>>> +/**
>>>> + * xe_survivability_mode_runtime_enable - Initialize and enable 
>>>> runtime survivability mode
>>>> + * @xe: xe device instance
>>>> + *
>>>> + * Initialize survivability information and enable runtime 
>>>> survivability mode.
>>>> + * Runtime survivability mode is enabled when certain errors cause 
>>>> the device to be
>>>> + * in non-recoverable state. The device is declared wedged with the 
>>>> appropriate
>>>> + * recovery method and survivability mode sysfs exposed to userspace
>>>> + *
>>>> + * Return: 0 if runtime survivability mode is enabled or not 
>>>> requested, negative error
>>>
>>> is the "not requested" still applicable here?
>>
>> Copied it from boot survivability. Not applicable, will remove this
>>
>>>
>>>
>>>> + * code otherwise.
>>>> + */
>>>> +int xe_survivability_mode_runtime_enable(struct xe_device *xe)
>>>> +{
>>>> +    struct xe_survivability *survivability = &xe->survivability;
>>>> +    struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
>>>> +    int ret;
>>>> +
>>>> +    if (!IS_DGFX(xe) || IS_SRIOV_VF(xe) || xe->info.platform < 
>>>> XE_BATTLEMAGE) {
>>>
>>> Do you think this condition can be better handled with a 
>>> has_runtime_survivability for platforms that support it?
>>
>> Was used once so added it here. Can be split out to a different function
> 
> Oh, not a different function. I mean a has_* property. More like entries 
> defined in xe_pci_types.h under struct xe_graphics_desc.

That might be unnecessary for now, since the function says not 
applicable prior to bmg.
If in future, this is not consistent then we could add per pci_desc

Thanks
Riana

> 
> Regards,
> Umesh
> 
>>>
>>>> +        dev_err(&pdev->dev, "Runtime Survivability Mode not 
>>>> supported\n");
>>>> +        return -EINVAL;
>>>> +    }
>>>> +
>>>> +    ret = init_survivability_mode(xe);
>>>> +    if (ret)
>>>> +        return ret;
>>>> +
>>>> +    ret = create_survivability_sysfs(pdev);
>>>> +    if (ret)
>>>> +        dev_err(&pdev->dev, "Failed to create survivability mode 
>>>> sysfs\n");
>>>
>>> You do not return ret in the above if condition. Is that intenational?
>>
>> yeah this is intentional. The device has to be wedged since it is not 
>> usable on such errors even without the sysfs.
>>
>> Thanks
>> Riana
>>
>>>
>>> Regards,
>>> Umesh
>>>
>>>> +
>>>> +    survivability->type = XE_SURVIVABILITY_TYPE_RUNTIME;
>>>> +    xe_device_set_wedged_method(xe, DRM_WEDGE_RECOVERY_VENDOR);
>>>> +    xe_device_declare_wedged(xe);
>>>> +
>>>> +    dev_err(&pdev->dev, "Runtime Survivability mode enabled\n");
>>>> +    return 0;
>>>> +}
>>>> +
>>>> /**
>>>>  * xe_survivability_mode_boot_enable - Initialize and enable boot 
>>>> survivability mode
>>>>  * @xe: xe device instance
>>>> diff --git a/drivers/gpu/drm/xe/xe_survivability_mode.h b/drivers/ 
>>>> gpu/ drm/xe/xe_survivability_mode.h
>>>> index f6ee283ea5e8..1cc94226aa82 100644
>>>> --- a/drivers/gpu/drm/xe/xe_survivability_mode.h
>>>> +++ b/drivers/gpu/drm/xe/xe_survivability_mode.h
>>>> @@ -11,6 +11,7 @@
>>>> struct xe_device;
>>>>
>>>> int xe_survivability_mode_boot_enable(struct xe_device *xe);
>>>> +int xe_survivability_mode_runtime_enable(struct xe_device *xe);
>>>> bool xe_survivability_mode_is_boot_enabled(struct xe_device *xe);
>>>> bool xe_survivability_mode_is_requested(struct xe_device *xe);
>>>>
>>>> diff --git a/drivers/gpu/drm/xe/xe_survivability_mode_types.h b/ 
>>>> drivers/gpu/drm/xe/xe_survivability_mode_types.h
>>>> index 5dce393498da..cd65a5d167c9 100644
>>>> --- a/drivers/gpu/drm/xe/xe_survivability_mode_types.h
>>>> +++ b/drivers/gpu/drm/xe/xe_survivability_mode_types.h
>>>> @@ -11,6 +11,7 @@
>>>>
>>>> enum xe_survivability_type {
>>>>     XE_SURVIVABILITY_TYPE_BOOT,
>>>> +    XE_SURVIVABILITY_TYPE_RUNTIME,
>>>> };
>>>>
>>>> struct xe_survivability_info {
>>>> -- 
>>>> 2.47.1
>>>>
>>
>>