Re: [PATCH v2 1/2] drm/amdkfd: Init the base cu processor id

[Felix Kuehling <felix.kuehling@amd.com>]

On 2022-10-26 05:03, Ma Jun wrote:
> Init and save the base cu processor id for later use
>
> Signed-off-by: Ma Jun <Jun.Ma2@amd.com>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 20 +++++++++++++-------
>   drivers/gpu/drm/amd/amdkfd/kfd_priv.h |  3 +++
>   2 files changed, 16 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
> index d25ac9cbe5b2..35d09e29aafb 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
> @@ -41,13 +41,18 @@ static uint32_t gpu_processor_id_low = 0x80001000;
>    *	@total_cu_count - Total CUs present in the GPU including ones
>    *			  masked off
>    */
> -static inline unsigned int get_and_inc_gpu_processor_id(
> -				unsigned int total_cu_count)
> +static inline void assign_gpu_processor_id(struct kfd_dev *kfd)
>   {
> -	int current_id = gpu_processor_id_low;
> +	struct amdgpu_device *adev = kfd->adev;
> +	unsigned int array_count = 0;
> +	unsigned int total_cu_count = 0;
> +
> +	kfd->processor_id_low = gpu_processor_id_low;

This still seems backwards. And I don't think you need it. You can 
access this after the topology has been parsed from the 
kfd_topology_device (dev->node_props.simd_id_base). I see you're using 
it in the next patch in kfd_fill_cache_non_crat_info, which has a 
kfd_topology_device parameter. This should run after parsing the CRAT 
tables, so the info should be available at that time.

Regards,
   Felix


> +
> +	array_count = adev->gfx.config.max_sh_per_se * adev->gfx.config.max_shader_engines;
> +	total_cu_count = array_count * adev->gfx.config.max_cu_per_sh;
>   
>   	gpu_processor_id_low += total_cu_count;
> -	return current_id;
>   }
>   
>   /* Static table to describe GPU Cache information */
> @@ -2223,7 +2228,6 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image,
>   	struct crat_subtype_computeunit *cu;
>   	struct kfd_cu_info cu_info;
>   	int avail_size = *size;
> -	uint32_t total_num_of_cu;
>   	int num_of_cache_entries = 0;
>   	int cache_mem_filled = 0;
>   	uint32_t nid = 0;
> @@ -2275,8 +2279,10 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image,
>   	cu->wave_front_size = cu_info.wave_front_size;
>   	cu->array_count = cu_info.num_shader_arrays_per_engine *
>   		cu_info.num_shader_engines;
> -	total_num_of_cu = (cu->array_count * cu_info.num_cu_per_sh);
> -	cu->processor_id_low = get_and_inc_gpu_processor_id(total_num_of_cu);
> +
> +	assign_gpu_processor_id(kdev);
> +	cu->processor_id_low = kdev->processor_id_low;
> +
>   	cu->num_cu_per_array = cu_info.num_cu_per_sh;
>   	cu->max_slots_scatch_cu = cu_info.max_scratch_slots_per_cu;
>   	cu->num_banks = cu_info.num_shader_engines;
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> index 182eb67edbc5..4c06b233472f 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
> @@ -370,6 +370,9 @@ struct kfd_dev {
>   
>   	/* Track per device allocated watch points. */
>   	uint32_t alloc_watch_ids;
> +
> +	/* cu processor id base */
> +	unsigned int	processor_id_low;
>   };
>   
>   struct kfd_ipc_obj;