[PATCH v2 13/14] drm/xe/hwmon: Fix mutex destroy

Matthew Auld matthew.auld at intel.com
Fri Feb 7 10:20:18 UTC 2025


On 07/02/2025 03:27, Lucas De Marchi wrote:
> On Thu, Feb 06, 2025 at 03:23:31PM -0800, Lucas De Marchi wrote:
>> Since the mutex is used if userspace kept the fd open, it can't be
>> released together with the dev cleanup and needs to be delayed. Also,

Where is that? I would have thought here that 
devm_hwmon_device_register_with_info() would ensure nothing can reach 
that lock once the physical device is detached, since everything 
contained in xe->hwmon is tied to that, including the lock. It looks a 
little weird but seems like it should be fine?

>> there's no need to add an action just to release the mutex: use the
>> specific drmm_mutex_init() that should be smarter if mutex_destroy() is
>> actually a nop.
>>
>> Cc: Badal Nilawar <badal.nilawar at intel.com>
>> Cc: Karthik Poosa <karthik.poosa at intel.com>
>> Signed-off-by: Lucas De Marchi <lucas.demarchi at intel.com>
> 
> oops... 3a13c2de442d6bfaef9c102cd1092e6cae22b753
> 
> the reasoning is true, but the entire thing is in struct devm allocated.
> Then CI complains:
> 
> <4> [321.873848] WARNING: CPU: 13 PID: 12544 at kernel/locking/mutex- 
> debug.c:114 mutex_destroy+0x5b/0x60
 > >
> Lucas De Marchi
> 
>> ---
>> drivers/gpu/drm/xe/xe_hwmon.c | 10 +---------
>> 1 file changed, 1 insertion(+), 9 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_hwmon.c b/drivers/gpu/drm/xe/ 
>> xe_hwmon.c
>> index 7f327e3342123..0fc91c184570d 100644
>> --- a/drivers/gpu/drm/xe/xe_hwmon.c
>> +++ b/drivers/gpu/drm/xe/xe_hwmon.c
>> @@ -869,13 +869,6 @@ xe_hwmon_get_preregistration_info(struct 
>> xe_device *xe)
>>             xe_hwmon_energy_get(hwmon, channel, &energy);
>> }
>>
>> -static void xe_hwmon_mutex_destroy(void *arg)
>> -{
>> -    struct xe_hwmon *hwmon = arg;
>> -
>> -    mutex_destroy(&hwmon->hwmon_lock);
>> -}
>> -
>> void xe_hwmon_register(struct xe_device *xe)
>> {
>>     struct device *dev = xe->drm.dev;
>> @@ -895,8 +888,7 @@ void xe_hwmon_register(struct xe_device *xe)
>>
>>     xe->hwmon = hwmon;
>>
>> -    mutex_init(&hwmon->hwmon_lock);
>> -    if (devm_add_action_or_reset(dev, xe_hwmon_mutex_destroy, hwmon))
>> +    if (drmm_mutex_init(&xe->drm, &hwmon->hwmon_lock))
>>         return;
>>
>>     /* There's only one instance of hwmon per device */
>> -- 
>> 2.48.1
>>



More information about the Intel-xe mailing list