[PATCH] drm/xe/ufence: ufence can be signaled right after wait_woken

[PATCH] drm/xe/ufence: ufence can be signaled right after wait_woken

[Nirmoy Das]

do_comapre() can return success after wait_woken() which is treated as
-ETIME here.

Fixes: e670f0b4ef24 ("drm/xe/uapi: Return correct error code for xe_wait_user_fence_ioctl")
Cc: <stable@vger.kernel.org> # v6.8+
Cc: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/1630
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
---
 drivers/gpu/drm/xe/xe_wait_user_fence.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_wait_user_fence.c b/drivers/gpu/drm/xe/xe_wait_user_fence.c
index d46fa8374980..d532283d4aa3 100644
--- a/drivers/gpu/drm/xe/xe_wait_user_fence.c
+++ b/drivers/gpu/drm/xe/xe_wait_user_fence.c
@@ -169,7 +169,7 @@ int xe_wait_user_fence_ioctl(struct drm_device *dev, void *data,
 			args->timeout = 0;
 	}
 
-	if (!timeout && !(err < 0))
+	if (!timeout && err < 0)
 		err = -ETIME;
 
 	if (q)
-- 
2.46.0

Re: [PATCH] drm/xe/ufence: ufence can be signaled right after wait_woken

[Nirmoy Das]

On 10/11/2024 3:25 PM, Nirmoy Das wrote:
> do_comapre() can return success after wait_woken() which is treated as
> -ETIME here.

s/after wait_woken()/after timedout wait_woken()

I will resend with that change.

>
> Fixes: e670f0b4ef24 ("drm/xe/uapi: Return correct error code for xe_wait_user_fence_ioctl")
> Cc: <stable@vger.kernel.org> # v6.8+
> Cc: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/1630
> Cc: Matthew Auld <matthew.auld@intel.com>
> Cc: Matthew Brost <matthew.brost@intel.com>
> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_wait_user_fence.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_wait_user_fence.c b/drivers/gpu/drm/xe/xe_wait_user_fence.c
> index d46fa8374980..d532283d4aa3 100644
> --- a/drivers/gpu/drm/xe/xe_wait_user_fence.c
> +++ b/drivers/gpu/drm/xe/xe_wait_user_fence.c
> @@ -169,7 +169,7 @@ int xe_wait_user_fence_ioctl(struct drm_device *dev, void *data,
>  			args->timeout = 0;
>  	}
>  
> -	if (!timeout && !(err < 0))
> +	if (!timeout && err < 0)
>  		err = -ETIME;
>  
>  	if (q)

Re: [PATCH] drm/xe/ufence: ufence can be signaled right after wait_woken

[Matthew Auld]

On 11/10/2024 15:10, Nirmoy Das wrote:
> 
> On 10/11/2024 3:25 PM, Nirmoy Das wrote:
>> do_comapre() can return success after wait_woken() which is treated as
>> -ETIME here.
> 
> s/after wait_woken()/after timedout wait_woken()
> 
> I will resend with that change.
> 
>>
>> Fixes: e670f0b4ef24 ("drm/xe/uapi: Return correct error code for xe_wait_user_fence_ioctl")
>> Cc: <stable@vger.kernel.org> # v6.8+
>> Cc: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/1630
>> Cc: Matthew Auld <matthew.auld@intel.com>
>> Cc: Matthew Brost <matthew.brost@intel.com>
>> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_wait_user_fence.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_wait_user_fence.c b/drivers/gpu/drm/xe/xe_wait_user_fence.c
>> index d46fa8374980..d532283d4aa3 100644
>> --- a/drivers/gpu/drm/xe/xe_wait_user_fence.c
>> +++ b/drivers/gpu/drm/xe/xe_wait_user_fence.c
>> @@ -169,7 +169,7 @@ int xe_wait_user_fence_ioctl(struct drm_device *dev, void *data,
>>   			args->timeout = 0;
>>   	}
>>   
>> -	if (!timeout && !(err < 0))

Since err > 0 is impossible, this could be written as: && err == 0.

So I think this is saying: if we have timedout and err does not already 
have an error set then go ahead and set to -ETIME since we hit the 
timeout. But it might have -EIO or -ERESTARTSYS for example, which 
should then take precedence over -ETIME...

>> +	if (!timeout && err < 0)

...this would then trample the existing err. The err can either be zero 
or an existing error at this point, so I think just remove this entire 
check:

-       if (!timeout && !(err < 0))
-               err = -ETIME;
-

?

>>   		err = -ETIME;
>>   
>>   	if (q)

Re: [PATCH] drm/xe/ufence: ufence can be signaled right after wait_woken

[Nirmoy Das]

On 10/11/2024 5:16 PM, Matthew Auld wrote:
> On 11/10/2024 15:10, Nirmoy Das wrote:
>>
>> On 10/11/2024 3:25 PM, Nirmoy Das wrote:
>>> do_comapre() can return success after wait_woken() which is treated as
>>> -ETIME here.
>>
>> s/after wait_woken()/after timedout wait_woken()
>>
>> I will resend with that change.
>>
>>>
>>> Fixes: e670f0b4ef24 ("drm/xe/uapi: Return correct error code for xe_wait_user_fence_ioctl")
>>> Cc: <stable@vger.kernel.org> # v6.8+
>>> Cc: Bommu Krishnaiah <krishnaiah.bommu@intel.com>
>>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/1630
>>> Cc: Matthew Auld <matthew.auld@intel.com>
>>> Cc: Matthew Brost <matthew.brost@intel.com>
>>> Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
>>> ---
>>>   drivers/gpu/drm/xe/xe_wait_user_fence.c | 2 +-
>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/xe/xe_wait_user_fence.c b/drivers/gpu/drm/xe/xe_wait_user_fence.c
>>> index d46fa8374980..d532283d4aa3 100644
>>> --- a/drivers/gpu/drm/xe/xe_wait_user_fence.c
>>> +++ b/drivers/gpu/drm/xe/xe_wait_user_fence.c
>>> @@ -169,7 +169,7 @@ int xe_wait_user_fence_ioctl(struct drm_device *dev, void *data,
>>>               args->timeout = 0;
>>>       }
>>>   -    if (!timeout && !(err < 0))
>
> Since err > 0 is impossible, this could be written as: && err == 0.
>
> So I think this is saying: if we have timedout and err does not already have an error set then go ahead and set to -ETIME since we hit the timeout.

This is the issue here. This assumption is wrong that if timeout happen then return -ETIME even though the fence is signaled.


> But it might have -EIO or -ERESTARTSYS for example, which should then take precedence over -ETIME...
>
>>> +    if (!timeout && err < 0)
>
> ...this would then trample the existing err. The err can either be zero or an existing error at this point, so I think just remove this entire check:
>
> -       if (!timeout && !(err < 0))
> -               err = -ETIME;
> -
>
> ?


Yes, this works for me. The for loops sets err correctly even when there is real timeout on not-signaled fence.

I will resend a v2.


Regards,

Nirmoy

>
>>>           err = -ETIME;
>>>         if (q)