Re: [PATCH 3/3] drm/xe: Memset g2g buffer on resume

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

On Thu, Oct 15, 2026 at 09:54:55AM -0700, Vinay Belgaumkar wrote:
> KMD creates this buffer for use by GuC, it is also responsible for zeroing
> it during resume. The buffer is marked as XE_BO_FLAG_PINNED_NORESTORE,
> so we cannot trust its contents after a suspend cycle. This commit adds
> xe_gt_resume_early() function for handling this and the C6 disable
> during resume.
> 

Fixes?

Also I have question here: Does this mean after d3cold memory does not
init to zero? This was an open question I had which merging
clear-of-free as I didn't know upon d3cold exit if buddy blocks in
cleared list would still be cleared? Any insight here would be helpful
as I think Tejas is picking up the clear-on-free work and this would be
good input to give him.

Matt 

> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_gt.c | 18 ++++++++++++++++++
>  drivers/gpu/drm/xe/xe_gt.h |  1 +
>  drivers/gpu/drm/xe/xe_pm.c |  4 ++--
>  3 files changed, 21 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> index 775c826b68b4..02f5d48b1a65 100644
> --- a/drivers/gpu/drm/xe/xe_gt.c
> +++ b/drivers/gpu/drm/xe/xe_gt.c
> @@ -1069,6 +1069,24 @@ int xe_gt_sanitize_freq(struct xe_gt *gt)
>  	return ret;
>  }
>  
> +/**
> + * xe_gt_resume_early() - Early init for GT resume
> + * @gt: the GT object
> + *
> + * Handles GT related init before GT restart after a suspend cycle.
> + */
> +void xe_gt_resume_early(struct xe_gt *gt)
> +{
> +	struct xe_device *xe = gt_to_xe(gt);
> +	struct xe_guc *guc = &gt->uc.guc;
> +
> +	xe_gt_idle_disable_c6(gt);
> +
> +	/* clear the g2g buffer which is marked as NORESTORE during alloc */
> +	if (guc->g2g.bo && guc->g2g.owned)
> +		xe_map_memset(xe, &guc->g2g.bo->vmap, 0, 0, xe_bo_size(guc->g2g.bo));
> +}
> +
>  int xe_gt_resume(struct xe_gt *gt)
>  {
>  	int err;
> diff --git a/drivers/gpu/drm/xe/xe_gt.h b/drivers/gpu/drm/xe/xe_gt.h
> index 92d934d50244..9b69f36ebc7a 100644
> --- a/drivers/gpu/drm/xe/xe_gt.h
> +++ b/drivers/gpu/drm/xe/xe_gt.h
> @@ -58,6 +58,7 @@ void xe_gt_record_user_engines(struct xe_gt *gt);
>  void xe_gt_suspend_prepare(struct xe_gt *gt);
>  int xe_gt_suspend(struct xe_gt *gt);
>  void xe_gt_shutdown(struct xe_gt *gt);
> +void xe_gt_resume_early(struct xe_gt *gt);
>  int xe_gt_resume(struct xe_gt *gt);
>  void xe_gt_reset_async(struct xe_gt *gt);
>  int xe_gt_runtime_resume(struct xe_gt *gt);
> diff --git a/drivers/gpu/drm/xe/xe_pm.c b/drivers/gpu/drm/xe/xe_pm.c
> index 1f5097875bcd..89d64aa6577d 100644
> --- a/drivers/gpu/drm/xe/xe_pm.c
> +++ b/drivers/gpu/drm/xe/xe_pm.c
> @@ -249,7 +249,7 @@ int xe_pm_resume(struct xe_device *xe)
>  		return err;
>  
>  	for_each_gt(gt, xe, id)
> -		xe_gt_idle_disable_c6(gt);
> +		xe_gt_resume_early(gt);
>  
>  	xe_display_pm_resume_early(xe);
>  
> @@ -703,7 +703,7 @@ int xe_pm_runtime_resume(struct xe_device *xe)
>  			goto out;
>  
>  		for_each_gt(gt, xe, id)
> -			xe_gt_idle_disable_c6(gt);
> +			xe_gt_resume_early(gt);
>  
>  		xe_display_pm_resume_early(xe);
>  
> -- 
> 2.53.0
>