[PATCH] drm/ast: Fix start address computation

[PATCH] drm/ast: Fix start address computation

[Jocelyn Falempe]

During the driver conversion to shmem, there is a missing page shift when
computing the start address in GPU memory.
This leads to graphic garbage when connecting to the remote BMC, depending
on the PCI start address.

Tested on a sr645 affected by this bug.

Fixes: f2fa5a99ca81 ("drm/ast: Convert ast to SHMEM")
Signed-off-by: Jocelyn Falempe <jfalempe@redhat.com>
---
 drivers/gpu/drm/ast/ast_mode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/ast/ast_mode.c b/drivers/gpu/drm/ast/ast_mode.c
index c7443317c747..d75e4a7611b3 100644
--- a/drivers/gpu/drm/ast/ast_mode.c
+++ b/drivers/gpu/drm/ast/ast_mode.c
@@ -714,7 +714,7 @@ static int ast_primary_plane_init(struct ast_private *ast)
 	struct ast_plane *ast_primary_plane = &ast->primary_plane;
 	struct drm_plane *primary_plane = &ast_primary_plane->base;
 	void __iomem *vaddr = ast->vram;
-	u64 offset = ast->vram_base;
+	u64 offset = ast->vram_base << PAGE_SHIFT;
 	unsigned long cursor_size = roundup(AST_HWC_SIZE + AST_HWC_SIGNATURE_SIZE, PAGE_SIZE);
 	unsigned long size = ast->vram_fb_available - cursor_size;
 	int ret;
-- 
2.39.1

Re: [PATCH] drm/ast: Fix start address computation

[Thomas Zimmermann]

[-- Attachment #1.1: Type: text/plain, Size: 1837 bytes --]

Hi

Am 07.02.23 um 11:53 schrieb Jocelyn Falempe:
> During the driver conversion to shmem, there is a missing page shift when
> computing the start address in GPU memory.
> This leads to graphic garbage when connecting to the remote BMC, depending
> on the PCI start address.
> 
> Tested on a sr645 affected by this bug.
> 
> Fixes: f2fa5a99ca81 ("drm/ast: Convert ast to SHMEM")
> Signed-off-by: Jocelyn Falempe <jfalempe@redhat.com>
> ---
>   drivers/gpu/drm/ast/ast_mode.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/ast/ast_mode.c b/drivers/gpu/drm/ast/ast_mode.c
> index c7443317c747..d75e4a7611b3 100644
> --- a/drivers/gpu/drm/ast/ast_mode.c
> +++ b/drivers/gpu/drm/ast/ast_mode.c
> @@ -714,7 +714,7 @@ static int ast_primary_plane_init(struct ast_private *ast)
>   	struct ast_plane *ast_primary_plane = &ast->primary_plane;
>   	struct drm_plane *primary_plane = &ast_primary_plane->base;
>   	void __iomem *vaddr = ast->vram;
> -	u64 offset = ast->vram_base;
> +	u64 offset = ast->vram_base << PAGE_SHIFT;

That can't be the problem, I think. vram_base is already an address. 
[1] In the old code, drm_gem_vram_pg_offset() returned a page index. So 
shifting it to the left here is most likely incorrect.

Can you verify that the new offset is the same as the old one?

Best regards
Thomas

[1] 
https://elixir.bootlin.com/linux/v6.2-rc7/source/drivers/gpu/drm/ast/ast_mm.c#L96

>   	unsigned long cursor_size = roundup(AST_HWC_SIZE + AST_HWC_SIGNATURE_SIZE, PAGE_SIZE);
>   	unsigned long size = ast->vram_fb_available - cursor_size;
>   	int ret;

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Ivo Totev

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: [PATCH] drm/ast: Fix start address computation

[Jocelyn Falempe]

On 07/02/2023 12:17, Thomas Zimmermann wrote:
> Hi
> 
> Am 07.02.23 um 11:53 schrieb Jocelyn Falempe:
>> During the driver conversion to shmem, there is a missing page shift when
>> computing the start address in GPU memory.
>> This leads to graphic garbage when connecting to the remote BMC, 
>> depending
>> on the PCI start address.
>>
>> Tested on a sr645 affected by this bug.
>>
>> Fixes: f2fa5a99ca81 ("drm/ast: Convert ast to SHMEM")
>> Signed-off-by: Jocelyn Falempe <jfalempe@redhat.com>
>> ---
>>   drivers/gpu/drm/ast/ast_mode.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/ast/ast_mode.c 
>> b/drivers/gpu/drm/ast/ast_mode.c
>> index c7443317c747..d75e4a7611b3 100644
>> --- a/drivers/gpu/drm/ast/ast_mode.c
>> +++ b/drivers/gpu/drm/ast/ast_mode.c
>> @@ -714,7 +714,7 @@ static int ast_primary_plane_init(struct 
>> ast_private *ast)
>>       struct ast_plane *ast_primary_plane = &ast->primary_plane;
>>       struct drm_plane *primary_plane = &ast_primary_plane->base;
>>       void __iomem *vaddr = ast->vram;
>> -    u64 offset = ast->vram_base;
>> +    u64 offset = ast->vram_base << PAGE_SHIFT;
> 
> That can't be the problem, I think. vram_base is already an address. [1] 
> In the old code, drm_gem_vram_pg_offset() returned a page index. So 
> shifting it to the left here is most likely incorrect.
> 
> Can you verify that the new offset is the same as the old one?

Yes, I was also a bit unsure about this fix.

In my case, with the old code:

PCI base address is 0xce000000

with old code:
start address (which is 24bits) was 0

with shmem code:
start address is 0x800000

So the additional page shift makes it 0, which fixes the display.

I was wondering if this start address is not an offset in the GPU 
memory, and in this case the primary plane offset should always be 0 ?

Best regards,

> 
> Best regards
> Thomas
> 
> [1] 
> https://elixir.bootlin.com/linux/v6.2-rc7/source/drivers/gpu/drm/ast/ast_mm.c#L96
> 
>>       unsigned long cursor_size = roundup(AST_HWC_SIZE + 
>> AST_HWC_SIGNATURE_SIZE, PAGE_SIZE);
>>       unsigned long size = ast->vram_fb_available - cursor_size;
>>       int ret;
> 

Re: [PATCH] drm/ast: Fix start address computation

[Jocelyn Falempe]

On 07/02/2023 12:36, Jocelyn Falempe wrote:
> On 07/02/2023 12:17, Thomas Zimmermann wrote:
>> Hi
>>
> 
> I was wondering if this start address is not an offset in the GPU 
> memory, and in this case the primary plane offset should always be 0 ?

I think it's the case, so I will send a v2 shortly.

> 
> Best regards,
> 
>>
>> Best regards
>> Thomas
>>
>> [1] 
>> https://elixir.bootlin.com/linux/v6.2-rc7/source/drivers/gpu/drm/ast/ast_mm.c#L96
>>
>>>       unsigned long cursor_size = roundup(AST_HWC_SIZE + 
>>> AST_HWC_SIGNATURE_SIZE, PAGE_SIZE);
>>>       unsigned long size = ast->vram_fb_available - cursor_size;
>>>       int ret;
>>
>