[PATCH v2] drm: Log recovery method with wedged event

[PATCH v2] drm: Log recovery method with wedged event

[Raag Jadav]

Currently, drm_dev_wedged_event() logs a basic message of device being
wedged, but doesn't print the recovery method used by the caller. Log it
for easier bug report analysis.

While at it, as device wedging is a catastrophic event for the cases that
do need recovery, distinguish them and raise their log level to drm_warn().

Signed-off-by: Raag Jadav <raag.jadav@intel.com>
---
v2: Distinguish between recovery and no recovery cases (Christian, Rodrigo)
---
 drivers/gpu/drm/drm_drv.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
index cb53baa70995..7f8abb72f548 100644
--- a/drivers/gpu/drm/drm_drv.c
+++ b/drivers/gpu/drm/drm_drv.c
@@ -598,8 +598,10 @@ int drm_dev_wedged_event(struct drm_device *dev, unsigned long method,
 		/* Caller is unsure about recovery, do the best we can at this point. */
 		snprintf(event_string, sizeof(event_string), "%s", "WEDGED=unknown");
 
-	drm_info(dev, "device wedged, %s\n", method == DRM_WEDGE_RECOVERY_NONE ?
-		 "but no recovery needed" : "needs recovery");
+	if (method == DRM_WEDGE_RECOVERY_NONE)
+		drm_info(dev, "device wedged, but no recovery needed\n");
+	else
+		drm_warn(dev, "device wedged with method %#lx, needs recovery\n", method);
 
 	if (info && (info->comm[0] != '\0') && (info->pid >= 0)) {
 		snprintf(pid_string, sizeof(pid_string), "PID=%u", info->pid);
-- 
2.43.0

Re: [PATCH v2] drm: Log recovery method with wedged event

[Christian König]

On 10/20/26 04:21, Raag Jadav wrote:
> Currently, drm_dev_wedged_event() logs a basic message of device being
> wedged, but doesn't print the recovery method used by the caller. Log it
> for easier bug report analysis.
> 
> While at it, as device wedging is a catastrophic event for the cases that
> do need recovery, distinguish them and raise their log level to drm_warn().
> 
> Signed-off-by: Raag Jadav <raag.jadav@intel.com>
> ---
> v2: Distinguish between recovery and no recovery cases (Christian, Rodrigo)
> ---
>  drivers/gpu/drm/drm_drv.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
> index cb53baa70995..7f8abb72f548 100644
> --- a/drivers/gpu/drm/drm_drv.c
> +++ b/drivers/gpu/drm/drm_drv.c
> @@ -598,8 +598,10 @@ int drm_dev_wedged_event(struct drm_device *dev, unsigned long method,
>  		/* Caller is unsure about recovery, do the best we can at this point. */
>  		snprintf(event_string, sizeof(event_string), "%s", "WEDGED=unknown");
>  
> -	drm_info(dev, "device wedged, %s\n", method == DRM_WEDGE_RECOVERY_NONE ?
> -		 "but no recovery needed" : "needs recovery");
> +	if (method == DRM_WEDGE_RECOVERY_NONE)
> +		drm_info(dev, "device wedged, but no recovery needed\n");
> +	else
> +		drm_warn(dev, "device wedged with method %#lx, needs recovery\n", method);

That is still not ok.

Again device wedging is a perfectly normal part of operation and clearly doesn't deserves a warning.

As far as I can see userspace can spam the log with warnings by just submitting shaders with an endless loop in it.

Regards,
Christian.

>  
>  	if (info && (info->comm[0] != '\0') && (info->pid >= 0)) {
>  		snprintf(pid_string, sizeof(pid_string), "PID=%u", info->pid);

Re: [PATCH v2] drm: Log recovery method with wedged event

[Rodrigo Vivi]

On Tue, Oct 20, 2026 at 01:13:50PM +0200, Christian König wrote:
> On 10/20/26 04:21, Raag Jadav wrote:
> > Currently, drm_dev_wedged_event() logs a basic message of device being
> > wedged, but doesn't print the recovery method used by the caller. Log it
> > for easier bug report analysis.
> > 
> > While at it, as device wedging is a catastrophic event for the cases that
> > do need recovery, distinguish them and raise their log level to drm_warn().
> > 
> > Signed-off-by: Raag Jadav <raag.jadav@intel.com>
> > ---
> > v2: Distinguish between recovery and no recovery cases (Christian, Rodrigo)
> > ---
> >  drivers/gpu/drm/drm_drv.c | 6 ++++--
> >  1 file changed, 4 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
> > index cb53baa70995..7f8abb72f548 100644
> > --- a/drivers/gpu/drm/drm_drv.c
> > +++ b/drivers/gpu/drm/drm_drv.c
> > @@ -598,8 +598,10 @@ int drm_dev_wedged_event(struct drm_device *dev, unsigned long method,
> >  		/* Caller is unsure about recovery, do the best we can at this point. */
> >  		snprintf(event_string, sizeof(event_string), "%s", "WEDGED=unknown");
> >  
> > -	drm_info(dev, "device wedged, %s\n", method == DRM_WEDGE_RECOVERY_NONE ?
> > -		 "but no recovery needed" : "needs recovery");
> > +	if (method == DRM_WEDGE_RECOVERY_NONE)
> > +		drm_info(dev, "device wedged, but no recovery needed\n");
> > +	else
> > +		drm_warn(dev, "device wedged with method %#lx, needs recovery\n", method);
> 
> That is still not ok.

what I had in mind was a explicit option so the drivers can select if it is
okay or not.

> 
> Again device wedging is a perfectly normal part of operation and clearly doesn't deserves a warning.

sometimes it is not, but I understand your point.

Perhaps we just skip this and in the case where it is not an info + a custom
driver warn is okay since everything will be stopped anyway in this cathastropic
case.

> 
> As far as I can see userspace can spam the log with warnings by just submitting shaders with an endless loop in it.
> 
> Regards,
> Christian.
> 
> >  
> >  	if (info && (info->comm[0] != '\0') && (info->pid >= 0)) {
> >  		snprintf(pid_string, sizeof(pid_string), "PID=%u", info->pid);
>