[PATCH 1/2] drm/amdgpu: store userq_managers in a list in adev

[PATCH 1/2] drm/amdgpu: store userq_managers in a list in adev

[Alex Deucher]

So we can iterate across them when we need to manage
all user queues.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h           |  3 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |  3 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c | 15 ++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h |  1 +
 4 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 4f770a362048a..28cfa600b798f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1228,6 +1228,9 @@ struct amdgpu_device {
 	 * in KFD: VRAM or GTT.
 	 */
 	bool                            apu_prefer_gtt;
+
+	struct list_head		userq_mgr_list;
+	struct mutex                    userq_mutex;
 };
 
 static inline uint32_t amdgpu_ip_version(const struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 0396ac30c2a4f..526c5aa32825a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4299,6 +4299,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	mutex_init(&adev->gfx.kfd_sch_mutex);
 	mutex_init(&adev->gfx.workload_profile_mutex);
 	mutex_init(&adev->vcn.workload_profile_mutex);
+	mutex_init(&adev->userq_mutex);
 
 	amdgpu_device_init_apu_flags(adev);
 
@@ -4326,6 +4327,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 
 	INIT_LIST_HEAD(&adev->pm.od_kobj_list);
 
+	INIT_LIST_HEAD(&adev->userq_mgr_list);
+
 	INIT_DELAYED_WORK(&adev->delayed_init_work,
 			  amdgpu_device_delayed_init_work_handler);
 	INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
index a02614cbda36e..b89bfad52abd5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
@@ -365,6 +365,9 @@ amdgpu_userqueue_create(struct drm_file *filp, union drm_amdgpu_userq *args)
 		goto unlock;
 	}
 	args->out.queue_id = qid;
+	mutex_lock(&adev->userq_mutex);
+	list_add(&uq_mgr->list, &adev->userq_mgr_list);
+	mutex_unlock(&adev->userq_mutex);
 
 unlock:
 	mutex_unlock(&uq_mgr->userq_mutex);
@@ -661,14 +664,24 @@ int amdgpu_userq_mgr_init(struct amdgpu_userq_mgr *userq_mgr, struct amdgpu_devi
 
 void amdgpu_userq_mgr_fini(struct amdgpu_userq_mgr *userq_mgr)
 {
-	uint32_t queue_id;
+	struct amdgpu_device *adev = userq_mgr->adev;
 	struct amdgpu_usermode_queue *queue;
+	struct amdgpu_userq_mgr *uqm, *tmp;
+	uint32_t queue_id;
 
 	cancel_delayed_work(&userq_mgr->resume_work);
 
 	mutex_lock(&userq_mgr->userq_mutex);
 	idr_for_each_entry(&userq_mgr->userq_idr, queue, queue_id)
 		amdgpu_userqueue_cleanup(userq_mgr, queue, queue_id);
+	mutex_lock(&adev->userq_mutex);
+	list_for_each_entry_safe(uqm, tmp, &adev->userq_mgr_list, list) {
+		if (uqm == userq_mgr) {
+			list_del(&uqm->list);
+			break;
+		}
+	}
+	mutex_unlock(&adev->userq_mutex);
 	idr_destroy(&userq_mgr->userq_idr);
 	mutex_unlock(&userq_mgr->userq_mutex);
 	mutex_destroy(&userq_mgr->userq_mutex);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
index 0f358f77f2d9b..ec1a4ca6f6321 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
@@ -76,6 +76,7 @@ struct amdgpu_userq_mgr {
 	struct mutex			userq_mutex;
 	struct amdgpu_device		*adev;
 	struct delayed_work		resume_work;
+	struct list_head		list;
 };
 
 struct amdgpu_db_info {
-- 
2.49.0

[PATCH 2/2] drm/amdgpu/userq: prevent runtime pm when userqs are active

[Alex Deucher]

Similar to KFD, prevent runtime pm while user queues are active.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 30 +++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 7d49a0bede1e2..4ff2e423c50cb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -2705,6 +2705,29 @@ static int amdgpu_runtime_idle_check_display(struct device *dev)
 	return 0;
 }
 
+static int amdgpu_runtime_idle_check_userq(struct device *dev)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct drm_device *drm_dev = pci_get_drvdata(pdev);
+	struct amdgpu_device *adev = drm_to_adev(drm_dev);
+	struct amdgpu_usermode_queue *queue;
+	struct amdgpu_userq_mgr *uqm, *tmp;
+	int queue_id;
+	int ret = 0;
+
+	mutex_lock(&adev->userq_mutex);
+	list_for_each_entry_safe(uqm, tmp, &adev->userq_mgr_list, list) {
+		idr_for_each_entry(&uqm->userq_idr, queue, queue_id) {
+			ret = -EBUSY;
+			goto done;
+		}
+	}
+done:
+	mutex_unlock(&adev->userq_mutex);
+
+	return ret;
+}
+
 static int amdgpu_pmops_runtime_suspend(struct device *dev)
 {
 	struct pci_dev *pdev = to_pci_dev(dev);
@@ -2718,6 +2741,9 @@ static int amdgpu_pmops_runtime_suspend(struct device *dev)
 	}
 
 	ret = amdgpu_runtime_idle_check_display(dev);
+	if (ret)
+		return ret;
+	ret = amdgpu_runtime_idle_check_userq(dev);
 	if (ret)
 		return ret;
 
@@ -2841,7 +2867,11 @@ static int amdgpu_pmops_runtime_idle(struct device *dev)
 	}
 
 	ret = amdgpu_runtime_idle_check_display(dev);
+	if (ret)
+		goto done;
 
+	ret = amdgpu_runtime_idle_check_userq(dev);
+done:
 	pm_runtime_mark_last_busy(dev);
 	pm_runtime_autosuspend(dev);
 	return ret;
-- 
2.49.0

Re: [PATCH 1/2] drm/amdgpu: store userq_managers in a list in adev

[Paneer Selvam, Arunpravin]

Hi Alex,

I requested the tester to reproduce the suspend/resume hang issue with 
this patch series.
I will update as soon as I get the results.

Thanks,
Arun.

On 3/20/2025 10:22 PM, Alex Deucher wrote:
> So we can iterate across them when we need to manage
> all user queues.
>
> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu.h           |  3 +++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |  3 +++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c | 15 ++++++++++++++-
>   drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h |  1 +
>   4 files changed, 21 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index 4f770a362048a..28cfa600b798f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -1228,6 +1228,9 @@ struct amdgpu_device {
>   	 * in KFD: VRAM or GTT.
>   	 */
>   	bool                            apu_prefer_gtt;
> +
> +	struct list_head		userq_mgr_list;
> +	struct mutex                    userq_mutex;
>   };
>   
>   static inline uint32_t amdgpu_ip_version(const struct amdgpu_device *adev,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 0396ac30c2a4f..526c5aa32825a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -4299,6 +4299,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>   	mutex_init(&adev->gfx.kfd_sch_mutex);
>   	mutex_init(&adev->gfx.workload_profile_mutex);
>   	mutex_init(&adev->vcn.workload_profile_mutex);
> +	mutex_init(&adev->userq_mutex);
>   
>   	amdgpu_device_init_apu_flags(adev);
>   
> @@ -4326,6 +4327,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>   
>   	INIT_LIST_HEAD(&adev->pm.od_kobj_list);
>   
> +	INIT_LIST_HEAD(&adev->userq_mgr_list);
> +
>   	INIT_DELAYED_WORK(&adev->delayed_init_work,
>   			  amdgpu_device_delayed_init_work_handler);
>   	INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> index a02614cbda36e..b89bfad52abd5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> @@ -365,6 +365,9 @@ amdgpu_userqueue_create(struct drm_file *filp, union drm_amdgpu_userq *args)
>   		goto unlock;
>   	}
>   	args->out.queue_id = qid;
> +	mutex_lock(&adev->userq_mutex);
> +	list_add(&uq_mgr->list, &adev->userq_mgr_list);
> +	mutex_unlock(&adev->userq_mutex);
>   
>   unlock:
>   	mutex_unlock(&uq_mgr->userq_mutex);
> @@ -661,14 +664,24 @@ int amdgpu_userq_mgr_init(struct amdgpu_userq_mgr *userq_mgr, struct amdgpu_devi
>   
>   void amdgpu_userq_mgr_fini(struct amdgpu_userq_mgr *userq_mgr)
>   {
> -	uint32_t queue_id;
> +	struct amdgpu_device *adev = userq_mgr->adev;
>   	struct amdgpu_usermode_queue *queue;
> +	struct amdgpu_userq_mgr *uqm, *tmp;
> +	uint32_t queue_id;
>   
>   	cancel_delayed_work(&userq_mgr->resume_work);
>   
>   	mutex_lock(&userq_mgr->userq_mutex);
>   	idr_for_each_entry(&userq_mgr->userq_idr, queue, queue_id)
>   		amdgpu_userqueue_cleanup(userq_mgr, queue, queue_id);
> +	mutex_lock(&adev->userq_mutex);
> +	list_for_each_entry_safe(uqm, tmp, &adev->userq_mgr_list, list) {
> +		if (uqm == userq_mgr) {
> +			list_del(&uqm->list);
> +			break;
> +		}
> +	}
> +	mutex_unlock(&adev->userq_mutex);
>   	idr_destroy(&userq_mgr->userq_idr);
>   	mutex_unlock(&userq_mgr->userq_mutex);
>   	mutex_destroy(&userq_mgr->userq_mutex);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> index 0f358f77f2d9b..ec1a4ca6f6321 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> @@ -76,6 +76,7 @@ struct amdgpu_userq_mgr {
>   	struct mutex			userq_mutex;
>   	struct amdgpu_device		*adev;
>   	struct delayed_work		resume_work;
> +	struct list_head		list;
>   };
>   
>   struct amdgpu_db_info {

RE: [PATCH 1/2] drm/amdgpu: store userq_managers in a list in adev

[Liang, Prike]

[Public]

Here it may need to test the uq_mgr list to see whether it is a fresh list before adding it to the userq_mgr_list; otherwise, it may add a duplicated uq_mgr list. I have sent a patch for that check.

Regards,
      Prike

> -----Original Message-----
> From: amd-gfx <amd-gfx-bounces@lists.freedesktop.org> On Behalf Of Alex
> Deucher
> Sent: Friday, March 21, 2025 12:53 AM
> To: amd-gfx@lists.freedesktop.org; Paneer Selvam, Arunpravin
> <Arunpravin.PaneerSelvam@amd.com>
> Cc: Deucher, Alexander <Alexander.Deucher@amd.com>
> Subject: [PATCH 1/2] drm/amdgpu: store userq_managers in a list in adev
>
> So we can iterate across them when we need to manage all user queues.
>
> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu.h           |  3 +++
>  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |  3 +++
>  drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c | 15 ++++++++++++++-
> drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h |  1 +
>  4 files changed, 21 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index 4f770a362048a..28cfa600b798f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -1228,6 +1228,9 @@ struct amdgpu_device {
>        * in KFD: VRAM or GTT.
>        */
>       bool                            apu_prefer_gtt;
> +
> +     struct list_head                userq_mgr_list;
> +     struct mutex                    userq_mutex;
>  };
>
>  static inline uint32_t amdgpu_ip_version(const struct amdgpu_device *adev, diff --
> git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 0396ac30c2a4f..526c5aa32825a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -4299,6 +4299,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>       mutex_init(&adev->gfx.kfd_sch_mutex);
>       mutex_init(&adev->gfx.workload_profile_mutex);
>       mutex_init(&adev->vcn.workload_profile_mutex);
> +     mutex_init(&adev->userq_mutex);
>
>       amdgpu_device_init_apu_flags(adev);
>
> @@ -4326,6 +4327,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>
>       INIT_LIST_HEAD(&adev->pm.od_kobj_list);
>
> +     INIT_LIST_HEAD(&adev->userq_mgr_list);
> +
>       INIT_DELAYED_WORK(&adev->delayed_init_work,
>                         amdgpu_device_delayed_init_work_handler);
>       INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> index a02614cbda36e..b89bfad52abd5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> @@ -365,6 +365,9 @@ amdgpu_userqueue_create(struct drm_file *filp, union
> drm_amdgpu_userq *args)
>               goto unlock;
>       }
>       args->out.queue_id = qid;
> +     mutex_lock(&adev->userq_mutex);
> +     list_add(&uq_mgr->list, &adev->userq_mgr_list);
> +     mutex_unlock(&adev->userq_mutex);
>
>  unlock:
>       mutex_unlock(&uq_mgr->userq_mutex);
> @@ -661,14 +664,24 @@ int amdgpu_userq_mgr_init(struct amdgpu_userq_mgr
> *userq_mgr, struct amdgpu_devi
>
>  void amdgpu_userq_mgr_fini(struct amdgpu_userq_mgr *userq_mgr)  {
> -     uint32_t queue_id;
> +     struct amdgpu_device *adev = userq_mgr->adev;
>       struct amdgpu_usermode_queue *queue;
> +     struct amdgpu_userq_mgr *uqm, *tmp;
> +     uint32_t queue_id;
>
>       cancel_delayed_work(&userq_mgr->resume_work);
>
>       mutex_lock(&userq_mgr->userq_mutex);
>       idr_for_each_entry(&userq_mgr->userq_idr, queue, queue_id)
>               amdgpu_userqueue_cleanup(userq_mgr, queue, queue_id);
> +     mutex_lock(&adev->userq_mutex);
> +     list_for_each_entry_safe(uqm, tmp, &adev->userq_mgr_list, list) {
> +             if (uqm == userq_mgr) {
> +                     list_del(&uqm->list);
> +                     break;
> +             }
> +     }
> +     mutex_unlock(&adev->userq_mutex);
>       idr_destroy(&userq_mgr->userq_idr);
>       mutex_unlock(&userq_mgr->userq_mutex);
>       mutex_destroy(&userq_mgr->userq_mutex);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> index 0f358f77f2d9b..ec1a4ca6f6321 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> @@ -76,6 +76,7 @@ struct amdgpu_userq_mgr {
>       struct mutex                    userq_mutex;
>       struct amdgpu_device            *adev;
>       struct delayed_work             resume_work;
> +     struct list_head                list;
>  };
>
>  struct amdgpu_db_info {
> --
> 2.49.0

Re: [PATCH 1/2] drm/amdgpu: store userq_managers in a list in adev

[Alex Deucher]

On Tue, Mar 25, 2025 at 4:23 AM Liang, Prike <Prike.Liang@amd.com> wrote:
>
> [Public]
>
> Here it may need to test the uq_mgr list to see whether it is a fresh list before adding it to the userq_mgr_list; otherwise, it may add a duplicated uq_mgr list. I have sent a patch for that check.

Good catch.  We should just add the mgr once in userq_mgr_init.  Will
send out a new patch set.

Alex


>
> Regards,
>       Prike
>
> > -----Original Message-----
> > From: amd-gfx <amd-gfx-bounces@lists.freedesktop.org> On Behalf Of Alex
> > Deucher
> > Sent: Friday, March 21, 2025 12:53 AM
> > To: amd-gfx@lists.freedesktop.org; Paneer Selvam, Arunpravin
> > <Arunpravin.PaneerSelvam@amd.com>
> > Cc: Deucher, Alexander <Alexander.Deucher@amd.com>
> > Subject: [PATCH 1/2] drm/amdgpu: store userq_managers in a list in adev
> >
> > So we can iterate across them when we need to manage all user queues.
> >
> > Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> > ---
> >  drivers/gpu/drm/amd/amdgpu/amdgpu.h           |  3 +++
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |  3 +++
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c | 15 ++++++++++++++-
> > drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h |  1 +
> >  4 files changed, 21 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > index 4f770a362048a..28cfa600b798f 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > @@ -1228,6 +1228,9 @@ struct amdgpu_device {
> >        * in KFD: VRAM or GTT.
> >        */
> >       bool                            apu_prefer_gtt;
> > +
> > +     struct list_head                userq_mgr_list;
> > +     struct mutex                    userq_mutex;
> >  };
> >
> >  static inline uint32_t amdgpu_ip_version(const struct amdgpu_device *adev, diff --
> > git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > index 0396ac30c2a4f..526c5aa32825a 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> > @@ -4299,6 +4299,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
> >       mutex_init(&adev->gfx.kfd_sch_mutex);
> >       mutex_init(&adev->gfx.workload_profile_mutex);
> >       mutex_init(&adev->vcn.workload_profile_mutex);
> > +     mutex_init(&adev->userq_mutex);
> >
> >       amdgpu_device_init_apu_flags(adev);
> >
> > @@ -4326,6 +4327,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
> >
> >       INIT_LIST_HEAD(&adev->pm.od_kobj_list);
> >
> > +     INIT_LIST_HEAD(&adev->userq_mgr_list);
> > +
> >       INIT_DELAYED_WORK(&adev->delayed_init_work,
> >                         amdgpu_device_delayed_init_work_handler);
> >       INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> > index a02614cbda36e..b89bfad52abd5 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.c
> > @@ -365,6 +365,9 @@ amdgpu_userqueue_create(struct drm_file *filp, union
> > drm_amdgpu_userq *args)
> >               goto unlock;
> >       }
> >       args->out.queue_id = qid;
> > +     mutex_lock(&adev->userq_mutex);
> > +     list_add(&uq_mgr->list, &adev->userq_mgr_list);
> > +     mutex_unlock(&adev->userq_mutex);
> >
> >  unlock:
> >       mutex_unlock(&uq_mgr->userq_mutex);
> > @@ -661,14 +664,24 @@ int amdgpu_userq_mgr_init(struct amdgpu_userq_mgr
> > *userq_mgr, struct amdgpu_devi
> >
> >  void amdgpu_userq_mgr_fini(struct amdgpu_userq_mgr *userq_mgr)  {
> > -     uint32_t queue_id;
> > +     struct amdgpu_device *adev = userq_mgr->adev;
> >       struct amdgpu_usermode_queue *queue;
> > +     struct amdgpu_userq_mgr *uqm, *tmp;
> > +     uint32_t queue_id;
> >
> >       cancel_delayed_work(&userq_mgr->resume_work);
> >
> >       mutex_lock(&userq_mgr->userq_mutex);
> >       idr_for_each_entry(&userq_mgr->userq_idr, queue, queue_id)
> >               amdgpu_userqueue_cleanup(userq_mgr, queue, queue_id);
> > +     mutex_lock(&adev->userq_mutex);
> > +     list_for_each_entry_safe(uqm, tmp, &adev->userq_mgr_list, list) {
> > +             if (uqm == userq_mgr) {
> > +                     list_del(&uqm->list);
> > +                     break;
> > +             }
> > +     }
> > +     mutex_unlock(&adev->userq_mutex);
> >       idr_destroy(&userq_mgr->userq_idr);
> >       mutex_unlock(&userq_mgr->userq_mutex);
> >       mutex_destroy(&userq_mgr->userq_mutex);
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> > index 0f358f77f2d9b..ec1a4ca6f6321 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userqueue.h
> > @@ -76,6 +76,7 @@ struct amdgpu_userq_mgr {
> >       struct mutex                    userq_mutex;
> >       struct amdgpu_device            *adev;
> >       struct delayed_work             resume_work;
> > +     struct list_head                list;
> >  };
> >
> >  struct amdgpu_db_info {
> > --
> > 2.49.0
>

Re: [PATCH 2/2] drm/amdgpu/userq: prevent runtime pm when userqs are active

[Paneer Selvam, Arunpravin]

[-- Attachment #1: Type: text/plain, Size: 2325 bytes --]

Hi Alex,

This patch didn't help to resolve the system hang issue during suspend 
that was introduced
by the series - https://patchwork.freedesktop.org/series/141717/

But as you mentioned, the runtime pm needs this change while user queues 
are running.
Reviewed-by: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>

Regards,
Arun.

On 3/20/2025 10:22 PM, Alex Deucher wrote:
> Similar to KFD, prevent runtime pm while user queues are active.
>
> Signed-off-by: Alex Deucher<alexander.deucher@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 30 +++++++++++++++++++++++++
>   1 file changed, 30 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> index 7d49a0bede1e2..4ff2e423c50cb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> @@ -2705,6 +2705,29 @@ static int amdgpu_runtime_idle_check_display(struct device *dev)
>   	return 0;
>   }
>   
> +static int amdgpu_runtime_idle_check_userq(struct device *dev)
> +{
> +	struct pci_dev *pdev = to_pci_dev(dev);
> +	struct drm_device *drm_dev = pci_get_drvdata(pdev);
> +	struct amdgpu_device *adev = drm_to_adev(drm_dev);
> +	struct amdgpu_usermode_queue *queue;
> +	struct amdgpu_userq_mgr *uqm, *tmp;
> +	int queue_id;
> +	int ret = 0;
> +
> +	mutex_lock(&adev->userq_mutex);
> +	list_for_each_entry_safe(uqm, tmp, &adev->userq_mgr_list, list) {
> +		idr_for_each_entry(&uqm->userq_idr, queue, queue_id) {
> +			ret = -EBUSY;
> +			goto done;
> +		}
> +	}
> +done:
> +	mutex_unlock(&adev->userq_mutex);
> +
> +	return ret;
> +}
> +
>   static int amdgpu_pmops_runtime_suspend(struct device *dev)
>   {
>   	struct pci_dev *pdev = to_pci_dev(dev);
> @@ -2718,6 +2741,9 @@ static int amdgpu_pmops_runtime_suspend(struct device *dev)
>   	}
>   
>   	ret = amdgpu_runtime_idle_check_display(dev);
> +	if (ret)
> +		return ret;
> +	ret = amdgpu_runtime_idle_check_userq(dev);
>   	if (ret)
>   		return ret;
>   
> @@ -2841,7 +2867,11 @@ static int amdgpu_pmops_runtime_idle(struct device *dev)
>   	}
>   
>   	ret = amdgpu_runtime_idle_check_display(dev);
> +	if (ret)
> +		goto done;
>   
> +	ret = amdgpu_runtime_idle_check_userq(dev);
> +done:
>   	pm_runtime_mark_last_busy(dev);
>   	pm_runtime_autosuspend(dev);
>   	return ret;

[-- Attachment #2: Type: text/html, Size: 2925 bytes --]