[Public]
Hi Christian,
Thanks for your feedback.
Originally, drm_sched_fini is part of amdgpu_fence_driver_hw_fini, I did not move it.
Former patch " cd87a6dcf6af drm/amdgpu: adjust fence driver enable sequence " has dropped amdgpu_fence_driver_suspend, and called amdgpu_fence_driver_hw_fini instead in function amdgpu_device_suspend. I checked the code difference between amdgpu_fence_driver_hw_fini and amdgpu_device_suspend, they are almost the same except drm_sched_fini part, so I think we can leave it as it is, while skipping the execution of drm_sched_fini in suspend/resume case.
Regards,
Guchun
-----Original Message-----
From: Koenig, Christian <Christian.Koenig@xxxxxxx>
Sent: Thursday, July 29, 2021 7:11 PM
To: Chen, Guchun <Guchun.Chen@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx; Gao, Likun <Likun.Gao@xxxxxxx>; Zhang, Hawking <Hawking.Zhang@xxxxxxx>; Deucher, Alexander <Alexander.Deucher@xxxxxxx>
Subject: Re: [PATCH] drm/amdgpu: avoid over-handle of fence driver fini in s3 test
Am 29.07.21 um 12:49 schrieb Guchun Chen:
> In amdgpu_fence_driver_hw_fini, no need to call drm_sched_fini to stop
> scheduler in s3 test, otherwise, fence errors will occur after resume.
> So introduce a new parameter to distingiush the case in this API.
NAK, the problem is rather that drm_sched_fini() is part of the sw shutdown and should never be called during hw_fini.
Christian.
>
> Fixes: cd87a6dcf6af drm/amdgpu: adjust fence driver enable sequence
> Signed-off-by: Guchun Chen <guchun.chen@xxxxxxx>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++--
> drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 8 +++++---
> drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 2 +-
> 3 files changed, 8 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index b1d2dc39e8be..aaff8ebbb7dc 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -3844,7 +3844,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)
> else
> drm_atomic_helper_shutdown(adev_to_drm(adev));
> }
> - amdgpu_fence_driver_hw_fini(adev);
> + amdgpu_fence_driver_hw_fini(adev, false);
>
> if (adev->pm_sysfs_en)
> amdgpu_pm_sysfs_fini(adev);
> @@ -3941,7 +3941,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool fbcon)
> /* evict vram memory */
> amdgpu_bo_evict_vram(adev);
>
> - amdgpu_fence_driver_hw_fini(adev);
> + amdgpu_fence_driver_hw_fini(adev, adev->in_suspend);
>
> amdgpu_device_ip_suspend_phase2(adev);
> /* evict remaining vram memory
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> index 49c5c7331c53..7e6a73c2919d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
> @@ -515,14 +515,15 @@ int amdgpu_fence_driver_init(struct amdgpu_device *adev)
> }
>
> /**
> - * amdgpu_fence_driver_fini - tear down the fence driver
> + * amdgpu_fence_driver_hw_fini - tear down the fence driver
> * for all possible rings.
> *
> * @adev: amdgpu device pointer
> + * @in_reset: indicator to distingiush device removal case or s3 case
> *
> * Tear down the fence driver for all possible rings (all asics).
> */
> -void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
> +void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev, bool
> +in_reset)
> {
> int i, r;
>
> @@ -531,8 +532,9 @@ void amdgpu_fence_driver_hw_fini(struct
> amdgpu_device *adev)
>
> if (!ring || !ring->fence_drv.initialized)
> continue;
> - if (!ring->no_scheduler)
> + if (!ring->no_scheduler && !in_reset)
> drm_sched_fini(&ring->sched);
> +
> /* You can't wait for HW to signal if it's gone */
> if (!drm_dev_is_unplugged(&adev->ddev))
> r = amdgpu_fence_wait_empty(ring); diff --git
> a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> index 27adffa7658d..42cbecfc26a3 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
> @@ -115,7 +115,7 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
> int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
> struct amdgpu_irq_src *irq_src,
> unsigned irq_type);
> -void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev);
> +void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev, bool
> +in_reset);
> void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev);
> void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev);
> int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence
> **fence,
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
