[PATCH 1/1] drm/amdgpu: unify BO evicting method in amdgpu_ttm

Das, Nirmoy nirmoy.das at amd.com
Thu Oct 7 10:59:24 UTC 2021


On 10/7/2021 12:38 PM, Christian König wrote:
> Am 07.10.21 um 12:00 schrieb Nirmoy Das:
>> Unify BO evicting functionality for possible memory
>> types in amdgpu_ttm.c.
>>
>> Signed-off-by: Nirmoy Das <nirmoy.das at amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c |  8 +++---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  | 30 ++++++++++++++++-----
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 ----------------
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 -
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c     | 30 +++++++++++++++++++++
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h     |  1 +
>>   6 files changed, 58 insertions(+), 35 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
>> index 5497e2d31d1a..164d6a9e9fbb 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
>> @@ -1328,7 +1328,7 @@ static int amdgpu_debugfs_evict_vram(void 
>> *data, u64 *val)
>>           return r;
>>       }
>>   -    *val = amdgpu_bo_evict_vram(adev);
>> +    *val = amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM);
>>         pm_runtime_mark_last_busy(dev->dev);
>>       pm_runtime_put_autosuspend(dev->dev);
>> @@ -1341,17 +1341,15 @@ static int amdgpu_debugfs_evict_gtt(void 
>> *data, u64 *val)
>>   {
>>       struct amdgpu_device *adev = (struct amdgpu_device *)data;
>>       struct drm_device *dev = adev_to_drm(adev);
>> -    struct ttm_resource_manager *man;
>>       int r;
>>         r = pm_runtime_get_sync(dev->dev);
>>       if (r < 0) {
>> -        pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
>> +        pm_runtime_put_autosuspend(dev->dev);
>>           return r;
>>       }
>>   -    man = ttm_manager_type(&adev->mman.bdev, TTM_PL_TT);
>> -    *val = ttm_resource_manager_evict_all(&adev->mman.bdev, man);
>> +    *val = amdgpu_ttm_evict_resources(adev, TTM_PL_TT);
>>         pm_runtime_mark_last_busy(dev->dev);
>>       pm_runtime_put_autosuspend(dev->dev);
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index 57638fe9cfc2..032deca4cea2 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -3880,6 +3880,25 @@ void amdgpu_device_fini_sw(struct 
>> amdgpu_device *adev)
>>     }
>>   +/**
>> + * amdgpu_device_evict_resources - evict device resources
>> + * @adev: amdgpu device object
>> + *
>> + * Evicts all ttm device resources(vram BOs, gart table) from the 
>> lru list
>> + * of the vram memory type. Mainly used for evicting device resources
>> + * at suspend time.
>> + *
>> + */
>> +void amdgpu_device_evict_resources(struct amdgpu_device *adev)
>
> Please add static here, apart from that the patch is Reviewed-by: 
> Christian König <christian.koenig at amd.com>


Thanks, I will add that and push the commit.


Nirmoy

>
> Thanks,
> Christian.
>
>> +{
>> +    /* No need to evict vram on APUs for suspend to ram */
>> +    if (adev->in_s3 && (adev->flags & AMD_IS_APU))
>> +        return;
>> +
>> +    if (amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM))
>> +        DRM_WARN("evicting device resources failed\n");
>> +
>> +}
>>     /*
>>    * Suspend & resume.
>> @@ -3920,17 +3939,16 @@ int amdgpu_device_suspend(struct drm_device 
>> *dev, bool fbcon)
>>       if (!adev->in_s0ix)
>>           amdgpu_amdkfd_suspend(adev, adev->in_runpm);
>>   -    /* evict vram memory */
>> -    amdgpu_bo_evict_vram(adev);
>> +    /* First evict vram memory */
>> +    amdgpu_device_evict_resources(adev);
>>         amdgpu_fence_driver_hw_fini(adev);
>>         amdgpu_device_ip_suspend_phase2(adev);
>> -    /* evict remaining vram memory
>> -     * This second call to evict vram is to evict the gart page table
>> -     * using the CPU.
>> +    /* This second call to evict device resources is to evict
>> +     * the gart page table using the CPU.
>>        */
>> -    amdgpu_bo_evict_vram(adev);
>> +    amdgpu_device_evict_resources(adev);
>>         return 0;
>>   }
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> index 4ec904f36ceb..073ba2af0b9c 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> @@ -1004,29 +1004,6 @@ void amdgpu_bo_unpin(struct amdgpu_bo *bo)
>>       }
>>   }
>>   -/**
>> - * amdgpu_bo_evict_vram - evict VRAM buffers
>> - * @adev: amdgpu device object
>> - *
>> - * Evicts all VRAM buffers on the lru list of the memory type.
>> - * Mainly used for evicting vram at suspend time.
>> - *
>> - * Returns:
>> - * 0 for success or a negative error code on failure.
>> - */
>> -int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
>> -{
>> -    struct ttm_resource_manager *man;
>> -
>> -    if (adev->in_s3 && (adev->flags & AMD_IS_APU)) {
>> -        /* No need to evict vram on APUs for suspend to ram */
>> -        return 0;
>> -    }
>> -
>> -    man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
>> -    return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
>> -}
>> -
>>   static const char *amdgpu_vram_names[] = {
>>       "UNKNOWN",
>>       "GDDR1",
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> index 8ff61bad4138..d787e0e89e0b 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> @@ -305,7 +305,6 @@ int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain);
>>   int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
>>                    u64 min_offset, u64 max_offset);
>>   void amdgpu_bo_unpin(struct amdgpu_bo *bo);
>> -int amdgpu_bo_evict_vram(struct amdgpu_device *adev);
>>   int amdgpu_bo_init(struct amdgpu_device *adev);
>>   void amdgpu_bo_fini(struct amdgpu_device *adev);
>>   int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 
>> tiling_flags);
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> index e2896ac2c9ce..bd5dda8066fa 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> @@ -2034,6 +2034,36 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
>>       return r;
>>   }
>>   +/**
>> + * amdgpu_ttm_evict_resources - evict memory buffers
>> + * @adev: amdgpu device object
>> + * @mem_type: evicted BO's memory type
>> + *
>> + * Evicts all @mem_type buffers on the lru list of the memory type.
>> + *
>> + * Returns:
>> + * 0 for success or a negative error code on failure.
>> + */
>> +int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int 
>> mem_type)
>> +{
>> +    struct ttm_resource_manager *man;
>> +
>> +    switch (mem_type) {
>> +    case TTM_PL_VRAM:
>> +    case TTM_PL_TT:
>> +    case AMDGPU_PL_GWS:
>> +    case AMDGPU_PL_GDS:
>> +    case AMDGPU_PL_OA:
>> +        man = ttm_manager_type(&adev->mman.bdev, mem_type);
>> +        break;
>> +    default:
>> +        DRM_ERROR("Trying to evict invalid memory type\n");
>> +        return -EINVAL;
>> +    }
>> +
>> +    return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
>> +}
>> +
>>   #if defined(CONFIG_DEBUG_FS)
>>     static int amdgpu_mm_vram_table_show(struct seq_file *m, void 
>> *unused)
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> index e69f3e8e06e5..ba5c864b8de1 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>> @@ -190,6 +190,7 @@ bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm);
>>   uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct 
>> ttm_resource *mem);
>>   uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct 
>> ttm_tt *ttm,
>>                    struct ttm_resource *mem);
>> +int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int 
>> mem_type);
>>     void amdgpu_ttm_debugfs_init(struct amdgpu_device *adev);
>


More information about the amd-gfx mailing list