On 2023/7/11 0:46, Shengyu Qu wrote:
> Hello Keith,
>> +
>> +static void vs_plane_atomic_update(struct drm_plane *plane,
>> + struct drm_atomic_state *state)
>> +{
>> + struct drm_plane_state *new_state = drm_atomic_get_new_plane_state(state,
>> + plane);
>> + unsigned char i, num_planes;
>> + struct drm_framebuffer *fb;
>> + struct vs_plane *vs_plane = to_vs_plane(plane);
>> + //struct drm_plane_state *state = plane->state;
>> + struct vs_crtc *vs_crtc = to_vs_crtc(new_state->crtc);
>> + struct vs_plane_state *plane_state = to_vs_plane_state(new_state);
>> + //struct drm_format_name_buf *name = &plane_state->status.format_name;
>> +
>> + if (!new_state->fb || !new_state->crtc)
>> + return;
>> +
>> + fb = new_state->fb;
>> +
>> + num_planes = vs_get_plane_number(fb);
>> +
>> + for (i = 0; i < num_planes; i++) {
>> + struct vs_gem_object *vs_obj;
>> +
>> + vs_obj = vs_fb_get_gem_obj(fb, i);
>> + vs_plane->dma_addr[i] = vs_obj->iova + fb->offsets[i];
>
> There is a dcache flush operation here in downstream driver. Is that a cache coherence problem?
>
> Best regards,
>
> Shengyu
>
>> + }
>> +
>> + plane_state->status.src = drm_plane_state_src(new_state);
>> + plane_state->status.dest = drm_plane_state_dest(new_state);
>> +
>> + vs_plane->funcs->update(vs_crtc->dev, vs_plane, plane, state);
>> +}
>>yes , You can find that the current situation is very professional.
This problem exists at present, but the dma interface is not perfect at now,
and the dma_sync_single_for_device interface needs to be implemented later.
I will consider repairing this problem in the form of patch
