Re: [PATCH] drm/amdkfd: simplify APU VRAM handling

[Felix Kuehling <felix.kuehling@xxxxxxx>]

On 2024-05-24 10:08, Alex Deucher wrote:
> With commit 89773b85599a
> ("drm/amdkfd: Let VRAM allocations go to GTT domain on small APUs")
> big and small APU "VRAM" handling in KFD was unified.  Since AMD_IS_APU
> is set for both big and small APUs, we can simplify the checks in
> the code.
>
> v2: clean up a few more places (Lang)
>
> Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>

This is a lot cleaner, thanks. I was looking for something like this 
when I reviewed the original patch but missed it. I found it now in 
amdgpu_discovery_set_ip_blocks (I think).

Acked-by: Felix Kuehling <felix.kuehling@xxxxxxx>


> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 16 ++++++++--------
>   drivers/gpu/drm/amd/amdkfd/kfd_migrate.c         |  2 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_svm.c             |  6 ++----
>   drivers/gpu/drm/amd/amdkfd/kfd_svm.h             |  1 -
>   4 files changed, 11 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> index 336eb51c4839..3af00b57cd8a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> @@ -196,7 +196,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
>   			return -EINVAL;
>   
>   		vram_size = KFD_XCP_MEMORY_SIZE(adev, xcp_id);
> -		if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) {
> +		if (adev->flags & AMD_IS_APU) {
>   			system_mem_needed = size;
>   			ttm_mem_needed = size;
>   		}
> @@ -233,7 +233,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
>   	if (adev && xcp_id >= 0) {
>   		adev->kfd.vram_used[xcp_id] += vram_needed;
>   		adev->kfd.vram_used_aligned[xcp_id] +=
> -				(adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) ?
> +				(adev->flags & AMD_IS_APU) ?
>   				vram_needed :
>   				ALIGN(vram_needed, VRAM_AVAILABLITY_ALIGN);
>   	}
> @@ -261,7 +261,7 @@ void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
>   
>   		if (adev) {
>   			adev->kfd.vram_used[xcp_id] -= size;
> -			if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) {
> +			if (adev->flags & AMD_IS_APU) {
>   				adev->kfd.vram_used_aligned[xcp_id] -= size;
>   				kfd_mem_limit.system_mem_used -= size;
>   				kfd_mem_limit.ttm_mem_used -= size;
> @@ -894,7 +894,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
>   	 * if peer device has large BAR. In contrast, access over xGMI is
>   	 * allowed for both small and large BAR configurations of peer device
>   	 */
> -	if ((adev != bo_adev && !(adev->gmc.is_app_apu || adev->flags & AMD_IS_APU)) &&
> +	if ((adev != bo_adev && !(adev->flags & AMD_IS_APU)) &&
>   	    ((mem->domain == AMDGPU_GEM_DOMAIN_VRAM) ||
>   	     (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL) ||
>   	     (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) {
> @@ -1682,7 +1682,7 @@ size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev,
>   		- atomic64_read(&adev->vram_pin_size)
>   		- reserved_for_pt;
>   
> -	if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) {
> +	if (adev->flags & AMD_IS_APU) {
>   		system_mem_available = no_system_mem_limit ?
>   					kfd_mem_limit.max_system_mem_limit :
>   					kfd_mem_limit.max_system_mem_limit -
> @@ -1730,7 +1730,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
>   	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
>   		domain = alloc_domain = AMDGPU_GEM_DOMAIN_VRAM;
>   
> -		if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) {
> +		if (adev->flags & AMD_IS_APU) {
>   			domain = AMDGPU_GEM_DOMAIN_GTT;
>   			alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
>   			alloc_flags = 0;
> @@ -1981,7 +1981,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
>   	if (size) {
>   		if (!is_imported &&
>   		   (mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_VRAM ||
> -		   ((adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) &&
> +		   ((adev->flags & AMD_IS_APU) &&
>   		    mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT)))
>   			*size = bo_size;
>   		else
> @@ -2404,7 +2404,7 @@ static int import_obj_create(struct amdgpu_device *adev,
>   	(*mem)->bo = bo;
>   	(*mem)->va = va;
>   	(*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) &&
> -			 !(adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) ?
> +			 !(adev->flags & AMD_IS_APU) ?
>   			 AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT;
>   
>   	(*mem)->mapped_to_gpu_memory = 0;
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> index 4816fcb9803a..8ee3d07ffbdf 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> @@ -1023,7 +1023,7 @@ int kgd2kfd_init_zone_device(struct amdgpu_device *adev)
>   	if (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(9, 0, 1))
>   		return -EINVAL;
>   
> -	if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU)
> +	if (adev->flags & AMD_IS_APU)
>   		return 0;
>   
>   	pgmap = &kfddev->pgmap;
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> index 28c2c1b66226..407636a68814 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> @@ -2634,8 +2634,7 @@ svm_range_best_restore_location(struct svm_range *prange,
>   		return -1;
>   	}
>   
> -	if (node->adev->gmc.is_app_apu ||
> -	    node->adev->flags & AMD_IS_APU)
> +	if (node->adev->flags & AMD_IS_APU)
>   		return 0;
>   
>   	if (prange->preferred_loc == gpuid ||
> @@ -3353,8 +3352,7 @@ svm_range_best_prefetch_location(struct svm_range *prange)
>   		goto out;
>   	}
>   
> -	if (bo_node->adev->gmc.is_app_apu ||
> -	    bo_node->adev->flags & AMD_IS_APU) {
> +	if (bo_node->adev->flags & AMD_IS_APU) {
>   		best_loc = 0;
>   		goto out;
>   	}
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
> index 9c37bd0567ef..70c1776611c4 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
> @@ -201,7 +201,6 @@ void svm_range_list_lock_and_flush_work(struct svm_range_list *svms, struct mm_s
>    * is initialized to not 0 when page migration register device memory.
>    */
>   #define KFD_IS_SVM_API_SUPPORTED(adev) ((adev)->kfd.pgmap.type != 0 ||\
> -					(adev)->gmc.is_app_apu ||\
>   					((adev)->flags & AMD_IS_APU))
>   
>   void svm_range_bo_unref_async(struct svm_range_bo *svm_bo);