Re: [PATCH] drm/amdgpu: add dce6 drm_panic support

[yaolu <yaolu@kylinos.cn>]

在 2024/8/2 17:39, Christian König 写道:
> Am 02.08.24 um 09:17 schrieb Lu Yao:
>> Add support for the drm_panic module, which displays a pretty user
>> friendly message on the screen when a Linux kernel panic occurs.
>>
>> Signed-off-by: Lu Yao <yaolu@kylinos.cn>
>> ---
>> The patch can work properly on the TTY, but after start X, drawn
>> image is messy, it looks like the data isn't linearly arranged.
>> However at this time 'fb->modifier' is 'DRM_FORMAT_MOD_LINEAR'.
>>
>> Another difference I found is:
>>    For TTY, the amdgpu_bo is created with flag
>> 'AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED|AMDGPU_GEM_CREATE_CPU_GTT_USWC|
>> AMDGPU_GEM_CREATE_VRAM_CLEARED|AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS'.
>>    For X, the amdgpu_bo is created with flag
>> 'AMDGPU_GEM_CREATE_NO_CPU_ACCESS|AMDGPU_GEM_CREATE_CPU_GTT_USWC'
>> I try to use same flag for X, it looks like no difference.
>>
>> Can someone provide some insight into this problem or where I am going
>> wrong. Thanks a lot.
>>
>> Test environment: X86 arch + v6.6 kernel + R7340.
>> ---
>>   drivers/gpu/drm/amd/amdgpu/dce_v6_0.c | 32 +++++++++++++++++++++++++++
>>   1 file changed, 32 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c 
>> b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c
>> index 05c0df97f01d..12c3801c264a 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c
>> @@ -28,6 +28,8 @@
>>   #include <drm/drm_modeset_helper.h>
>>   #include <drm/drm_modeset_helper_vtables.h>
>>   #include <drm/drm_vblank.h>
>> +#include <drm/drm_panic.h>
>
>> +#include "../../drm_internal.h"
>
> Well that this file is named "internal" and not in a common include 
> directory is a strong indicator that you should absolutely *not* 
> include it in a driver.
>
Okay, I'll fix it.
>>     #include "amdgpu.h"
>>   #include "amdgpu_pm.h"
>> @@ -2600,6 +2602,35 @@ static const struct drm_crtc_helper_funcs 
>> dce_v6_0_crtc_helper_funcs = {
>>       .get_scanout_position = amdgpu_crtc_get_scanout_position,
>>   };
>>   +static int dce_v6_0_drm_primary_plane_get_scanout_buffer(struct 
>> drm_plane *plane,
>> +                             struct drm_scanout_buffer *sb)
>> +{
>> +    struct drm_framebuffer *fb;
>> +    struct drm_gem_object *obj;
>> +    struct amdgpu_bo *abo;
>> +    int ret = 0;
>> +
>> +    if (!plane->fb || plane->fb->modifier != DRM_FORMAT_MOD_LINEAR)
>> +        return -ENODEV;
>> +
>> +    fb = plane->fb;
>> +    sb->width = fb->width;
>> +    sb->height = fb->height;
>> +    sb->format = fb->format;
>> +    sb->pitch[0] = fb->pitches[0];
>> +
>> +    obj = fb->obj[0];
>> +    abo = gem_to_amdgpu_bo(obj);
>> +    if (!abo || abo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
>> +        return -EINVAL;
>> +
>> +    return drm_gem_vmap(obj, &sb->map[0]);
>
> Yeah that will almost always not work. Most display buffers are tilled 
> and not CPU accessible.
>
> Regards,
> Christian.
>
I did some more tests. After removing the 
'AMDGPU_GEM_CREATE_NO_CPU_ACCESS' judgment here, then starting X, it 
worked well at '1280x960' resolution, but others (e.g. 1920x1080, 
640x480) not.
So, for this problem, it doesn't seem to matter 'Tiled memory' or 'CPU 
can't access'. Or is it just a coincidence ?

Thanks,
Lu Yao
>> +}
>> +
>> +static const struct drm_plane_helper_funcs 
>> dce_v6_0_drm_primary_plane_helper_funcs = {
>> +    .get_scanout_buffer = dce_v6_0_drm_primary_plane_get_scanout_buffer
>> +};
>> +
>>   static int dce_v6_0_crtc_init(struct amdgpu_device *adev, int index)
>>   {
>>       struct amdgpu_crtc *amdgpu_crtc;
>> @@ -2627,6 +2658,7 @@ static int dce_v6_0_crtc_init(struct 
>> amdgpu_device *adev, int index)
>>       amdgpu_crtc->encoder = NULL;
>>       amdgpu_crtc->connector = NULL;
>>       drm_crtc_helper_add(&amdgpu_crtc->base, 
>> &dce_v6_0_crtc_helper_funcs);
>> +    drm_plane_helper_add(amdgpu_crtc->base.primary, 
>> &dce_v6_0_drm_primary_plane_helper_funcs);
>>         return 0;
>>   }
>