[PATCH v2] drm/amdgpu: Fix VRAM BO swap issue

[PATCH v2] drm/amdgpu: Fix VRAM BO swap issue

[Arunpravin Paneer Selvam]

DRM buddy manager allocates the contiguous memory requests in
a single block or multiple blocks. So for the ttm move operation
(incase of low vram memory) we should consider all the blocks to
compute the total memory size which compared with the struct
ttm_resource num_pages in order to verify that the blocks are
contiguous for the eviction process.

v2: Added a Fixes tag

Fixes: c9cad937c0c5 ("drm/amdgpu: add drm buddy support to amdgpu")
Signed-off-by: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index b1c455329023..b1223c8e30c9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -426,6 +426,7 @@ static bool amdgpu_mem_visible(struct amdgpu_device *adev,
 {
 	uint64_t mem_size = (u64)mem->num_pages << PAGE_SHIFT;
 	struct amdgpu_res_cursor cursor;
+	u64 start, size, total_size = 0;
 
 	if (mem->mem_type == TTM_PL_SYSTEM ||
 	    mem->mem_type == TTM_PL_TT)
@@ -435,8 +436,23 @@ static bool amdgpu_mem_visible(struct amdgpu_device *adev,
 
 	amdgpu_res_first(mem, 0, mem_size, &cursor);
 
-	/* ttm_resource_ioremap only supports contiguous memory */
-	if (cursor.size != mem_size)
+	do {
+		start = cursor.start;
+		size = cursor.size;
+
+		total_size += size;
+
+		amdgpu_res_next(&cursor, cursor.size);
+
+		if (!cursor.remaining)
+			break;
+
+		/* ttm_resource_ioremap only supports contiguous memory */
+		if (start + size != cursor.start)
+			return false;
+	} while (1);
+
+	if (total_size != mem_size)
 		return false;
 
 	return cursor.start + cursor.size <= adev->gmc.visible_vram_size;
-- 
2.25.1

Re: [PATCH v2] drm/amdgpu: Fix VRAM BO swap issue

[Christian König]

Am 22.09.22 um 17:26 schrieb Arunpravin Paneer Selvam:
> DRM buddy manager allocates the contiguous memory requests in
> a single block or multiple blocks. So for the ttm move operation
> (incase of low vram memory) we should consider all the blocks to
> compute the total memory size which compared with the struct
> ttm_resource num_pages in order to verify that the blocks are
> contiguous for the eviction process.
>
> v2: Added a Fixes tag
>
> Fixes: c9cad937c0c5 ("drm/amdgpu: add drm buddy support to amdgpu")
> Signed-off-by: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 20 ++++++++++++++++++--
>   1 file changed, 18 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index b1c455329023..b1223c8e30c9 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -426,6 +426,7 @@ static bool amdgpu_mem_visible(struct amdgpu_device *adev,
>   {
>   	uint64_t mem_size = (u64)mem->num_pages << PAGE_SHIFT;
>   	struct amdgpu_res_cursor cursor;
> +	u64 start, size, total_size = 0;
>   
>   	if (mem->mem_type == TTM_PL_SYSTEM ||
>   	    mem->mem_type == TTM_PL_TT)
> @@ -435,8 +436,23 @@ static bool amdgpu_mem_visible(struct amdgpu_device *adev,
>   
>   	amdgpu_res_first(mem, 0, mem_size, &cursor);
>   
> -	/* ttm_resource_ioremap only supports contiguous memory */
> -	if (cursor.size != mem_size)
> +	do {
> +		start = cursor.start;
> +		size = cursor.size;
> +
> +		total_size += size;
> +
> +		amdgpu_res_next(&cursor, cursor.size);
> +
> +		if (!cursor.remaining)
> +			break;
> +
> +		/* ttm_resource_ioremap only supports contiguous memory */
> +		if (start + size != cursor.start)
> +			return false;
> +	} while (1);
> +
> +	if (total_size != mem_size)
>   		return false;

I would completely drop this extra check.

>   	return cursor.start + cursor.size <= adev->gmc.visible_vram_size;

Instead of this you should be able to do all of this in one go.

Something like this here should work:

amdgpu_res_first(...
end = cursor.start + cursor.size;
do (
     amdgpu_res_next(....
     if (end != cursor.start)
         return false;
     end = cursor.start + cursor.size;
} while (cursor.remaining);
return end <= visible_vram_size;

Saves a bit of extra calculations and variables.

Regards,
Christian.

Re: [PATCH v2] drm/amdgpu: Fix VRAM BO swap issue

[Christian König]

Am 22.09.22 um 19:14 schrieb Christian König:
>
>
> Am 22.09.22 um 17:26 schrieb Arunpravin Paneer Selvam:
>> DRM buddy manager allocates the contiguous memory requests in
>> a single block or multiple blocks. So for the ttm move operation
>> (incase of low vram memory) we should consider all the blocks to
>> compute the total memory size which compared with the struct
>> ttm_resource num_pages in order to verify that the blocks are
>> contiguous for the eviction process.
>>
>> v2: Added a Fixes tag
>>
>> Fixes: c9cad937c0c5 ("drm/amdgpu: add drm buddy support to amdgpu")
>> Signed-off-by: Arunpravin Paneer Selvam 
>> <Arunpravin.PaneerSelvam@amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 20 ++++++++++++++++++--
>>   1 file changed, 18 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> index b1c455329023..b1223c8e30c9 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> @@ -426,6 +426,7 @@ static bool amdgpu_mem_visible(struct 
>> amdgpu_device *adev,
>>   {
>>       uint64_t mem_size = (u64)mem->num_pages << PAGE_SHIFT;
>>       struct amdgpu_res_cursor cursor;
>> +    u64 start, size, total_size = 0;
>>         if (mem->mem_type == TTM_PL_SYSTEM ||
>>           mem->mem_type == TTM_PL_TT)
>> @@ -435,8 +436,23 @@ static bool amdgpu_mem_visible(struct 
>> amdgpu_device *adev,
>>         amdgpu_res_first(mem, 0, mem_size, &cursor);
>>   -    /* ttm_resource_ioremap only supports contiguous memory */
>> -    if (cursor.size != mem_size)
>> +    do {
>> +        start = cursor.start;
>> +        size = cursor.size;
>> +
>> +        total_size += size;
>> +
>> +        amdgpu_res_next(&cursor, cursor.size);
>> +
>> +        if (!cursor.remaining)
>> +            break;
>> +
>> +        /* ttm_resource_ioremap only supports contiguous memory */
>> +        if (start + size != cursor.start)
>> +            return false;
>> +    } while (1);
>> +
>> +    if (total_size != mem_size)
>>           return false;
>
> I would completely drop this extra check.
>
>>       return cursor.start + cursor.size <= adev->gmc.visible_vram_size;
>
> Instead of this you should be able to do all of this in one go.
>
> Something like this here should work:
>
> amdgpu_res_first(...
> end = cursor.start + cursor.size;
> do (
>     amdgpu_res_next(....
>     if (end != cursor.start)
>         return false;
>     end = cursor.start + cursor.size;
> } while (cursor.remaining);

My fault, this should have been

while (cursor.remaining) {
...
}

Regards,
Christian.

>
> return end <= visible_vram_size;
>
> Saves a bit of extra calculations and variables.
>
> Regards,
> Christian.