[PATCH v2] drm/amdgpu: Fix incorrect VRAM GART mappings on non-4K page size systems

[PATCH v2] drm/amdgpu: Fix incorrect VRAM GART mappings on non-4K page size systems

[Donet Tom]

When mapping VRAM pages into the GART page table,
amdgpu_gart_map_vram_range() assumes that the system page size is the
same as the GPU page size.

On systems with non-4K page sizes, multiple GPU pages can exist within
a single CPU page. As a result, the mappings are created incorrectly
because fewer page table entries are programmed than required.

Fix this by programming the mappings correctly for non-4K page size
systems.

Fixes: 237d623ae659 ("drm/amdgpu/gart: Add helper to bind VRAM pages (v2)")
Signed-off-by: Donet Tom <donettom@linux.ibm.com>
---
v1 -> v2
- Addressed Felix's comment by updating the definition of
  amdgpu_gart_map_vram_range() to support non-4K page sizes.

v1 - https://lore.kernel.org/all/20260522112838.1311531-1-donettom@linux.ibm.com/
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index b6f849d51c2e..66a15ee13b57 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -394,7 +394,8 @@ void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
 				uint64_t start_page, uint64_t num_pages,
 				uint64_t flags, void *dst)
 {
-	u32 i, idx;
+	u32 i, j, t = 0, idx;
+	u64 page_base;
 
 	/* The SYSTEM flag indicates the pages aren't in VRAM. */
 	WARN_ON_ONCE(flags & AMDGPU_PTE_SYSTEM);
@@ -402,9 +403,12 @@ void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
 	if (!drm_dev_enter(adev_to_drm(adev), &idx))
 		return;
 
-	for (i = 0; i < num_pages; ++i) {
-		amdgpu_gmc_set_pte_pde(adev, dst,
-			start_page + i, pa + AMDGPU_GPU_PAGE_SIZE * i, flags);
+	page_base = pa;
+	for (i = 0; i < num_pages; i++) {
+		for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) {
+			amdgpu_gmc_set_pte_pde(adev, dst, start_page + t, page_base, flags);
+			page_base += AMDGPU_GPU_PAGE_SIZE;
+		}
 	}
 
 	drm_dev_exit(idx);
-- 
2.52.0

Re: [PATCH v2] drm/amdgpu: Fix incorrect VRAM GART mappings on non-4K page size systems

[Christian König]

On 5/27/26 13:17, Donet Tom wrote:
> When mapping VRAM pages into the GART page table,
> amdgpu_gart_map_vram_range() assumes that the system page size is the
> same as the GPU page size.
> 
> On systems with non-4K page sizes, multiple GPU pages can exist within
> a single CPU page. As a result, the mappings are created incorrectly
> because fewer page table entries are programmed than required.
> 
> Fix this by programming the mappings correctly for non-4K page size
> systems.
> 
> Fixes: 237d623ae659 ("drm/amdgpu/gart: Add helper to bind VRAM pages (v2)")
> Signed-off-by: Donet Tom <donettom@linux.ibm.com>
> ---
> v1 -> v2
> - Addressed Felix's comment by updating the definition of
>   amdgpu_gart_map_vram_range() to support non-4K page sizes.
> 
> v1 - https://lore.kernel.org/all/20260522112838.1311531-1-donettom@linux.ibm.com/
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> index b6f849d51c2e..66a15ee13b57 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
> @@ -394,7 +394,8 @@ void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
>  				uint64_t start_page, uint64_t num_pages,
>  				uint64_t flags, void *dst)
>  {
> -	u32 i, idx;
> +	u32 i, j, t = 0, idx;
> +	u64 page_base;
>  
>  	/* The SYSTEM flag indicates the pages aren't in VRAM. */
>  	WARN_ON_ONCE(flags & AMDGPU_PTE_SYSTEM);
> @@ -402,9 +403,12 @@ void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
>  	if (!drm_dev_enter(adev_to_drm(adev), &idx))
>  		return;
>  
> -	for (i = 0; i < num_pages; ++i) {
> -		amdgpu_gmc_set_pte_pde(adev, dst,
> -			start_page + i, pa + AMDGPU_GPU_PAGE_SIZE * i, flags);
> +	page_base = pa;
> +	for (i = 0; i < num_pages; i++) {

Please make that for (i = 0, t = 0; .... and remove the initialization of t during deceleration.

I was about to complain that t isn't initialized because I missed the line above.

Apart from that Reviewed-by: Christian König <christian.koenig@amd.com>.

Regards,
Christian

> +		for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) {
> +			amdgpu_gmc_set_pte_pde(adev, dst, start_page + t, page_base, flags);
> +			page_base += AMDGPU_GPU_PAGE_SIZE;
> +		}
>  	}
>  
>  	drm_dev_exit(idx);

Re: [PATCH v2] drm/amdgpu: Fix incorrect VRAM GART mappings on non-4K page size systems

[Donet Tom]

On 5/27/26 5:24 PM, Christian König wrote:
>
> On 5/27/26 13:17, Donet Tom wrote:
>> When mapping VRAM pages into the GART page table,
>> amdgpu_gart_map_vram_range() assumes that the system page size is the
>> same as the GPU page size.
>>
>> On systems with non-4K page sizes, multiple GPU pages can exist within
>> a single CPU page. As a result, the mappings are created incorrectly
>> because fewer page table entries are programmed than required.
>>
>> Fix this by programming the mappings correctly for non-4K page size
>> systems.
>>
>> Fixes: 237d623ae659 ("drm/amdgpu/gart: Add helper to bind VRAM pages (v2)")
>> Signed-off-by: Donet Tom <donettom@linux.ibm.com>
>> ---
>> v1 -> v2
>> - Addressed Felix's comment by updating the definition of
>>    amdgpu_gart_map_vram_range() to support non-4K page sizes.
>>
>> v1 - https://lore.kernel.org/all/20260522112838.1311531-1-donettom@linux.ibm.com/
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 12 ++++++++----
>>   1 file changed, 8 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
>> index b6f849d51c2e..66a15ee13b57 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
>> @@ -394,7 +394,8 @@ void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
>>   				uint64_t start_page, uint64_t num_pages,
>>   				uint64_t flags, void *dst)
>>   {
>> -	u32 i, idx;
>> +	u32 i, j, t = 0, idx;
>> +	u64 page_base;
>>   
>>   	/* The SYSTEM flag indicates the pages aren't in VRAM. */
>>   	WARN_ON_ONCE(flags & AMDGPU_PTE_SYSTEM);
>> @@ -402,9 +403,12 @@ void amdgpu_gart_map_vram_range(struct amdgpu_device *adev, uint64_t pa,
>>   	if (!drm_dev_enter(adev_to_drm(adev), &idx))
>>   		return;
>>   
>> -	for (i = 0; i < num_pages; ++i) {
>> -		amdgpu_gmc_set_pte_pde(adev, dst,
>> -			start_page + i, pa + AMDGPU_GPU_PAGE_SIZE * i, flags);
>> +	page_base = pa;
>> +	for (i = 0; i < num_pages; i++) {
> Please make that for (i = 0, t = 0; .... and remove the initialization of t during deceleration.


Thank you Christian

I will send a v3 with this chanage

-Donet

>
> I was about to complain that t isn't initialized because I missed the line above.
>
> Apart from that Reviewed-by: Christian König <christian.koenig@amd.com>.
>
> Regards,
> Christian
>
>> +		for (j = 0; j < AMDGPU_GPU_PAGES_IN_CPU_PAGE; j++, t++) {
>> +			amdgpu_gmc_set_pte_pde(adev, dst, start_page + t, page_base, flags);
>> +			page_base += AMDGPU_GPU_PAGE_SIZE;
>> +		}
>>   	}
>>   
>>   	drm_dev_exit(idx);