Re: [PATCH] drm/xe/guc: Register CT fini action before mutex init

[Matthew Brost <matthew.brost@intel.com>]

On Tue, Nov 04, 2025 at 10:16:14PM +0000, Shuicheng Lin wrote:
> xe_guc_ct_init_noalloc() allocates the CT workqueue and other helpers
> before it tries to initialize ct->lock. If drmm_mutex_init() fails
> we currently bail out without releasing those resources because the
> guc_ct_fini() hasn’t been registered yet. Move the drmm_add_action_or_reset()
> call ahead of the mutex setup so that the devres unwinder tears everything
> down on the error path. This is safe because guc_ct_fini() never
> touches ct->lock.

destroy_workqueue in guc_ct_fini can flush the workqueue which does take
ct lock. So I think the correct flow here is at the top of
xe_guc_ct_init_noalloc do the drmm_mutex_init, then initialize all CT
state, then register guc_ct_fini.

Matt 

> 
> Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
> Signed-off-by: Shuicheng Lin <shuicheng.lin@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_guc_ct.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c
> index 671d69958ce7..2680e9096c15 100644
> --- a/drivers/gpu/drm/xe/xe_guc_ct.c
> +++ b/drivers/gpu/drm/xe/xe_guc_ct.c
> @@ -240,16 +240,16 @@ int xe_guc_ct_init_noalloc(struct xe_guc_ct *ct)
>  	init_waitqueue_head(&ct->wq);
>  	init_waitqueue_head(&ct->g2h_fence_wq);
>  
> -	err = drmm_mutex_init(&xe->drm, &ct->lock);
> +	err = drmm_add_action_or_reset(&xe->drm, guc_ct_fini, ct);
>  	if (err)
>  		return err;
>  
> -	primelockdep(ct);
> -
> -	err = drmm_add_action_or_reset(&xe->drm, guc_ct_fini, ct);
> +	err = drmm_mutex_init(&xe->drm, &ct->lock);
>  	if (err)
>  		return err;
>  
> +	primelockdep(ct);
> +
>  	xe_gt_assert(gt, ct->state == XE_GUC_CT_STATE_NOT_INITIALIZED);
>  	ct->state = XE_GUC_CT_STATE_DISABLED;
>  	return 0;
> -- 
> 2.49.0
>