Re: [1/3] drm/xe: Toggle the D3Cold enable flag at runtime resume

["Belgaumkar, Vinay" <vinay.belgaumkar@intel.com>]

On 10/19/2026 3:08 AM, Matthew Auld wrote:
> On 16/10/2026 19:35, Belgaumkar, Vinay wrote:
>>
>> On 10/16/2026 11:23 AM, Poosa, Karthik wrote:
>>>
>>> On 15-10-2026 22:24, Vinay Belgaumkar wrote:
>>>> We enable this flag during runtime suspend,
>>> i think there should be Fixes tag for this.
>>>
>>> Fixes: f026520367be ("drm/xe: Only set PCI d3cold_allowed when we 
>>> are really allowing.")
>> ok.
>>>
>>>> need to set it back to
>>>> false early during resume to prevent any race with pci driver.
>>> What is race that seen without this ?
>>
>> While D0 is being programmed in xe_pci_runtime_resume(), pci driver's 
>> internal auto suspend timer could expire and, since this d3cold flag 
>> is still true, pci driver could try to put vsp into D3. This is probably 
>
> I think needs more info here? We are in the RPM resume callback at 
> this stage (RESUMING), so surely we are holding the required 
> usage_count refs etc here, so how can it enter D3 here?

That is correct. I don't think the upper bridge/root port can go into D3 
as the pci driver will hold a ref until the endpoint is active. I think 
this mainly resolves more of a book-keeping issue. While the device is 
in d3, querying the d3cold_allowed status via sysfs can show "true" if 
we don't have this fix. So, more of a code consistency change. I will 
update the patch desc with this and remove the part about any possible 
race with the pci driver.

Thanks,

Vinay.

>
>
>> why we toggle this flag as the very first thing in the S3 resume 
>> function xe_pci_resume().
>>
>> Thanks,
>>
>> Vinay.
>>
>>>>
>>>> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
>>>> ---
>>>>   drivers/gpu/drm/xe/xe_pci.c | 3 +++
>>>>   1 file changed, 3 insertions(+)
>>>>
>>>> diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
>>>> index 57c6f778ca24..2269e51868c1 100644
>>>> --- a/drivers/gpu/drm/xe/xe_pci.c
>>>> +++ b/drivers/gpu/drm/xe/xe_pci.c
>>>> @@ -1450,6 +1450,9 @@ static int xe_pci_runtime_resume(struct 
>>>> device *dev)
>>>>       struct xe_device *xe = pdev_to_xe_device(pdev);
>>>>       int err;
>>>>   +    if (xe->d3cold.allowed)
>>>> +        d3cold_toggle(pdev, D3COLD_DISABLE);
>>>> +
>>>>       err = pci_set_power_state(pdev, PCI_D0);
>>>>       if (err)
>>>>           return err;
>