[PATCH] drm/xe: Decrement client count immediately on file close

[Upadhyay, Tejas]

> -----Original Message-----
> From: Intel-xe <intel-xe-bounces at lists.freedesktop.org> On Behalf Of
> Balasubramani Vivekanandan
> Sent: Wednesday, September 18, 2024 1:42 PM
> To: intel-xe at lists.freedesktop.org
> Cc: Nerlige Ramappa, Umesh <umesh.nerlige.ramappa at intel.com>;
> Vishwanathapura, Niranjana <niranjana.vishwanathapura at intel.com>; De
> Marchi, Lucas <lucas.demarchi at intel.com>; Vivekanandan, Balasubramani
> <balasubramani.vivekanandan at intel.com>
> Subject: [PATCH] drm/xe: Decrement client count immediately on file close
> 
> Decrement the client count immediately on file close. It is not required to be
> deferred to the resource cleanup function. Otherwise there will be a small
> time window, where there will be a non-zero client count even after closing all
> open file handles.
> This affects ccs_mode(xe_compute) igt tests as these tests try to change the
> ccs_mode immediately after closing all file handles, but the driver rejects the
> ccs_mode change request as it sees a non-zero client count.
> 
> Fixes: ce8c161cbad4 ("drm/xe: Add ref counting for xe_file")
> Signed-off-by: Balasubramani Vivekanandan
> <balasubramani.vivekanandan at intel.com>
> ---
>  drivers/gpu/drm/xe/xe_device.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> index 4d3c794f134c..3bccea6212ff 100644
> --- a/drivers/gpu/drm/xe/xe_device.c
> +++ b/drivers/gpu/drm/xe/xe_device.c
> @@ -107,17 +107,12 @@ static int xe_file_open(struct drm_device *dev,
> struct drm_file *file)  static void xe_file_destroy(struct kref *ref)  {
>  	struct xe_file *xef = container_of(ref, struct xe_file, refcount);
> -	struct xe_device *xe = xef->xe;
> 
>  	xa_destroy(&xef->exec_queue.xa);
>  	mutex_destroy(&xef->exec_queue.lock);
>  	xa_destroy(&xef->vm.xa);
>  	mutex_destroy(&xef->vm.lock);
> 
> -	spin_lock(&xe->clients.lock);
> -	xe->clients.count--;
> -	spin_unlock(&xe->clients.lock);
> -
>  	xe_drm_client_put(xef->client);
>  	kfree(xef->process_name);
>  	kfree(xef);
> @@ -178,6 +173,10 @@ static void xe_file_close(struct drm_device *dev,
> struct drm_file *file)
> 
>  	xe_file_put(xef);
> 
> +	spin_lock(&xe->clients.lock);
> +	xe->clients.count--;
> +	spin_unlock(&xe->clients.lock);

The file_close here is sychronus and serialized call with respect to userspace. Any settings done through sysfs post file_close should not required this change as far as I know. Would please explain scenario better?

Tejas
> +
>  	xe_pm_runtime_put(xe);
>  }
> 
> --
> 2.34.1