[PATCH 3/3] drm/amdgpu: Fix dmesg warning from xarray

Paneer Selvam, Arunpravin arunpravin.paneerselvam at amd.com
Mon Nov 25 05:35:16 UTC 2024


Hi Christian,


On 11/20/2024 4:06 PM, Christian König wrote:
> Am 20.11.24 um 10:26 schrieb Arunpravin Paneer Selvam:
>> Fix the warning issue generated at lib/xarray.c:1849.
>>
>> v2: keep the error handling same as before when the
>>      xalloc() call fails to store the fence_drv ref.
>>
>> Signed-off-by: Arunpravin Paneer Selvam 
>> <Arunpravin.PaneerSelvam at amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c | 6 ++++--
>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
>> index 76f7babd7a54..bfe7d1d139d1 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
>> @@ -842,8 +842,10 @@ int amdgpu_userq_wait_ioctl(struct drm_device 
>> *dev, void *data,
>>                * have any more space left and crash.
>>                */
>>               if (fence_drv->fence_drv_xa_ptr) {
>> -                r = xa_alloc(fence_drv->fence_drv_xa_ptr, &index, 
>> fence_drv,
>> -                         xa_limit_32b, GFP_KERNEL);
>> +                xa_lock(fence_drv->fence_drv_xa_ptr);
>> +                r = __xa_alloc(fence_drv->fence_drv_xa_ptr, &index, 
>> fence_drv,
>> +                           xa_limit_32b, GFP_KERNEL);
>> +                xa_unlock(fence_drv->fence_drv_xa_ptr);
>
> Why should that fix anything? That code looks identical, just with 
> explicit locking.
The flag XA_FLAGS_TRACK_FREE is cleared for unknown reason and a warning 
is generated when we
run the libdrm fence synchronization test case continuously with just 
xa_alloc function (this function has implicit locking).
The warnings are coming from lib/xarray.c, in __xa_alloc() function, we 
have a WARN_ON_ONCE to check the
XA_FLAGS_TRACK_FREE flag enablement.

This particular warning is not triggered when we do a explicit locking, 
this could be a bug in xa_alloc implementation?
>
> Regards,
> Christian.
>
>>                   if (r)
>>                       goto free_fences;



More information about the amd-gfx mailing list