[PATCH 01/11] drm/amdgpu: try allocating VRAM as power of two

[Christian König]

Am 11.09.2018 um 11:50 schrieb Huang Rui:
> On Sun, Sep 09, 2018 at 08:03:29PM +0200, Christian König wrote:
>> Try to allocate VRAM in power of two sizes and only fallback to vram
>> split sizes if that fails.
>>
>> Signed-off-by: Christian König <christian.koenig at amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 52 +++++++++++++++++++++-------
>>   1 file changed, 40 insertions(+), 12 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>> index 9cfa8a9ada92..3f9d5d00c9b3 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c
>> @@ -124,6 +124,28 @@ u64 amdgpu_vram_mgr_bo_visible_size(struct amdgpu_bo *bo)
>>   	return usage;
>>   }
>>   
>> +/**
>> + * amdgpu_vram_mgr_virt_start - update virtual start address
>> + *
>> + * @mem: ttm_mem_reg to update
>> + * @node: just allocated node
>> + *
>> + * Calculate a virtual BO start address to easily check if everything is CPU
>> + * accessible.
>> + */
>> +static void amdgpu_vram_mgr_virt_start(struct ttm_mem_reg *mem,
>> +				       struct drm_mm_node *node)
>> +{
>> +	unsigned long start;
>> +
>> +	start = node->start + node->size;
>> +	if (start > mem->num_pages)
>> +		start -= mem->num_pages;
>> +	else
>> +		start = 0;
>> +	mem->start = max(mem->start, start);
>> +}
>> +
> How about move this function into ttm (ttm_bo_manager.c) then export the
> symbol, it looks more like a ttm helper function.

No, that is a completely amdgpu specific hack.

We just provide a virtual start address so that the page fault function 
can quickly check if the BO is in CPU visible VRAM or not.

Regards,
Christian.

>
> Thanks,
> Ray
>
>>   /**
>>    * amdgpu_vram_mgr_new - allocate new ranges
>>    *
>> @@ -176,10 +198,25 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
>>   	pages_left = mem->num_pages;
>>   
>>   	spin_lock(&mgr->lock);
>> -	for (i = 0; i < num_nodes; ++i) {
>> +	for (i = 0; pages_left >= pages_per_node; ++i) {
>> +		unsigned long pages = rounddown_pow_of_two(pages_left);
>> +
>> +		r = drm_mm_insert_node_in_range(mm, &nodes[i], pages,
>> +						pages_per_node, 0,
>> +						place->fpfn, lpfn,
>> +						mode);
>> +		if (unlikely(r))
>> +			break;
>> +
>> +		usage += nodes[i].size << PAGE_SHIFT;
>> +		vis_usage += amdgpu_vram_mgr_vis_size(adev, &nodes[i]);
>> +		amdgpu_vram_mgr_virt_start(mem, &nodes[i]);
>> +		pages_left -= pages;
>> +	}
>> +
>> +	for (; pages_left; ++i) {
>>   		unsigned long pages = min(pages_left, pages_per_node);
>>   		uint32_t alignment = mem->page_alignment;
>> -		unsigned long start;
>>   
>>   		if (pages == pages_per_node)
>>   			alignment = pages_per_node;
>> @@ -193,16 +230,7 @@ static int amdgpu_vram_mgr_new(struct ttm_mem_type_manager *man,
>>   
>>   		usage += nodes[i].size << PAGE_SHIFT;
>>   		vis_usage += amdgpu_vram_mgr_vis_size(adev, &nodes[i]);
>> -
>> -		/* Calculate a virtual BO start address to easily check if
>> -		 * everything is CPU accessible.
>> -		 */
>> -		start = nodes[i].start + nodes[i].size;
>> -		if (start > mem->num_pages)
>> -			start -= mem->num_pages;
>> -		else
>> -			start = 0;
>> -		mem->start = max(mem->start, start);
>> +		amdgpu_vram_mgr_virt_start(mem, &nodes[i]);
>>   		pages_left -= pages;
>>   	}
>>   	spin_unlock(&mgr->lock);
>> -- 
>> 2.14.1
>>
>> _______________________________________________
>> amd-gfx mailing list
>> amd-gfx at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx