Re: Questions about KMS flip

[Christian König]

Am 16.11.21 um 04:27 schrieb Lang Yu:
On Mon, Nov 15, 2021 at 01:04:15PM +0100, Michel DDDnzer wrote:
[SNIP]
Though a single call to dce_v*_0_crtc_do_set_base() will
only pin the BO, I found it will be unpinned in next call to
dce_v*_0_crtc_do_set_base().
Yeah, that's the normal case when the new BO is different from the old one.
To catch the case I described, try something like

diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c 
b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c

index 18a7b3bd633b..5726bd87a355 100644

--- a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c

+++ b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c

@@ -1926,6 +1926,7 @@ static int dce_v11_0_crtc_do_set_base(struct drm_crtc 
*crtc,

                 return r;



         if (!atomic) {

+               WARN_ON_ONCE(target_fb == fb);

                 r = amdgpu_bo_pin(abo, AMDGPU_GEM_DOMAIN_VRAM);

                 if (unlikely(r != 0)) {

                         amdgpu_bo_unreserve(abo);

I did some tests, the warning can be triggered.

pin/unpin operations in *_crtc_do_set_base() and
amdgpu_display_crtc_page_flip_target() are mixed.

Ok sounds like we narrowed down the root cause pretty well.

Question is now how can we fix this? Just not pin the BO when target_fb 
== fb?

Thanks,
Christian.


Regards,
Lang