[PATCH] drm/amdkfd: Use huge page size to check split svm range alignment

[PATCH] drm/amdkfd: Use huge page size to check split svm range alignment

[Xiaogang.Chen]

From: Xiaogang Chen <xiaogang.chen@amd.com>

Fixes: 7ef6b2d4b7e5 (drm/amdkfd: remap unaligned svm ranges that have split)

When split svm ranges that have been mapped using huge page should use huge
page size(2MB) to check split range alignment, not prange->granularity that
means migration granularity.

Signed-off-by: Xiaogang Chen <xiaogang.chen@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 521c14c7a789..3af85c232659 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -1149,7 +1149,7 @@ svm_range_split_tail(struct svm_range *prange, uint64_t new_last,
 
 	if (!r) {
 		list_add(&tail->list, insert_list);
-		if (!IS_ALIGNED(new_last + 1, 1UL << prange->granularity))
+		if (!IS_ALIGNED(new_last + 1, 512))
 			list_add(&tail->update_list, remap_list);
 	}
 	return r;
@@ -1164,7 +1164,7 @@ svm_range_split_head(struct svm_range *prange, uint64_t new_start,
 
 	if (!r) {
 		list_add(&head->list, insert_list);
-		if (!IS_ALIGNED(new_start, 1UL << prange->granularity))
+		if (!IS_ALIGNED(new_start, 512))
 			list_add(&head->update_list, remap_list);
 	}
 	return r;
-- 
2.34.1

Re: [PATCH] drm/amdkfd: Use huge page size to check split svm range alignment

[Philip Yang]

On 2025-11-10 15:30, Xiaogang.Chen wrote:
> From: Xiaogang Chen <xiaogang.chen@amd.com>
>
> Fixes: 7ef6b2d4b7e5 (drm/amdkfd: remap unaligned svm ranges that have split)
>
> When split svm ranges that have been mapped using huge page should use huge
> page size(2MB) to check split range alignment, not prange->granularity that
> means migration granularity.
>
> Signed-off-by: Xiaogang Chen <xiaogang.chen@amd.com>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> index 521c14c7a789..3af85c232659 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> @@ -1149,7 +1149,7 @@ svm_range_split_tail(struct svm_range *prange, uint64_t new_last,
>   
>   	if (!r) {
>   		list_add(&tail->list, insert_list);
> -		if (!IS_ALIGNED(new_last + 1, 1UL << prange->granularity))
> +		if (!IS_ALIGNED(new_last + 1, 512))
we should check the original prange size is larger than 512 pages
           if (!IS_ALIGNED(new_last + 1, 512) && tail->last - 
prange->start >= 512)
>   			list_add(&tail->update_list, remap_list);
>   	}
>   	return r;
> @@ -1164,7 +1164,7 @@ svm_range_split_head(struct svm_range *prange, uint64_t new_start,
>   
>   	if (!r) {
>   		list_add(&head->list, insert_list);
> -		if (!IS_ALIGNED(new_start, 1UL << prange->granularity))
> +		if (!IS_ALIGNED(new_start, 512))

if (!IS_ALIGNED(new_start, 512) && prange->last - head->start >= 512)

Regards,

Philip

>   			list_add(&head->update_list, remap_list);
>   	}
>   	return r;

Re: [PATCH] drm/amdkfd: Use huge page size to check split svm range alignment

[Chen, Xiaogang]

On 11/11/2025 9:21 AM, Philip Yang wrote:
>
> On 2025-11-10 15:30, Xiaogang.Chen wrote:
>> From: Xiaogang Chen <xiaogang.chen@amd.com>
>>
>> Fixes: 7ef6b2d4b7e5 (drm/amdkfd: remap unaligned svm ranges that have 
>> split)
>>
>> When split svm ranges that have been mapped using huge page should 
>> use huge
>> page size(2MB) to check split range alignment, not 
>> prange->granularity that
>> means migration granularity.
>>
>> Signed-off-by: Xiaogang Chen <xiaogang.chen@amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 4 ++--
>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c 
>> b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
>> index 521c14c7a789..3af85c232659 100644
>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
>> @@ -1149,7 +1149,7 @@ svm_range_split_tail(struct svm_range *prange, 
>> uint64_t new_last,
>>         if (!r) {
>>           list_add(&tail->list, insert_list);
>> -        if (!IS_ALIGNED(new_last + 1, 1UL << prange->granularity))
>> +        if (!IS_ALIGNED(new_last + 1, 512))
> we should check the original prange size is larger than 512 pages
>           if (!IS_ALIGNED(new_last + 1, 512) && tail->last - 
> prange->start >= 512)

ok, but should be following one?

   if (!IS_ALIGNED(tail->start, 512) || (tail->last - tail->start + 1)  
<  512)

>> list_add(&tail->update_list, remap_list);
>>       }
>>       return r;
>> @@ -1164,7 +1164,7 @@ svm_range_split_head(struct svm_range *prange, 
>> uint64_t new_start,
>>         if (!r) {
>>           list_add(&head->list, insert_list);
>> -        if (!IS_ALIGNED(new_start, 1UL << prange->granularity))
>> +        if (!IS_ALIGNED(new_start, 512))
>
> if (!IS_ALIGNED(new_start, 512) && prange->last - head->start >= 512)

same for head case, should use this?

if (!IS_ALIGNED( head->start , 512) || ( head->last - head->start + 1)  
< 512)

Regards

Xiaogang

> Regards,
>
> Philip
>
>> list_add(&head->update_list, remap_list);
>>       }
>>       return r;