Re: [PATCH v2 0/4] CMDQ refinement of Mediatek DRM driver

[Enric Balletbo Serra]

Hi Chun-Kuang,

Missatge de Chun-Kuang Hu <chunkuang.hu@xxxxxxxxxx> del dia dt., 21 de
set. 2021 a les 15:15:
>
> Hi, Enric:
>
> Enric Balletbo Serra <eballetbo@xxxxxxxxx> 於 2021年9月21日 週二 下午4:36寫道：
> >
> > Hi Chun-Kuang,
> >
> > (again without html format, sorry for the noise)
> >
> > Missatge de Chun-Kuang Hu <chunkuang.hu@xxxxxxxxxx> del dia dj., 12
> > d’ag. 2021 a les 2:13:
> > >
> > > Chun-Kuang Hu <chunkuang.hu@xxxxxxxxxx> 於 2021年8月9日 週一 上午7:47寫道：
> > > >
> > > > These refinements include using standard mailbox callback interface,
> > > > timeout detection, and a fixed cmdq_handle.
> > >
> > > For this series, applied to mediatek-drm-next [1].
> > >
> > > [1] https://git.kernel.org/pub/scm/linux/kernel/git/chunkuang.hu/linux.git/log/?h=mediatek-drm-next
> > >
> >
> > These patches seem to break the display on the Acer Chromebook R 13
> > (MT8173) in the current mainline. After running a bisection it pointed
> > me to the following commit
> >
> > commit f4be17cd5b14dd73545b0e014a63ebe9ab5ef837
> > Author: Chun-Kuang Hu <chunkuang.hu@xxxxxxxxxx>
> > Date:   Sun Jul 4 15:36:48 2021 +0800
> >
> >     drm/mediatek: Remove struct cmdq_client
> >
> > Reverting this patch alone is not trivial, so I ended up reverting the
> > full series, and I can confirm that reverting the full series makes
> > the display work again.
>
> I think you could not just revert "drm/mediatek: Remove struct
> cmdq_client", you should also revert the patches after it, such as
>
> "drm/mediatek: Clear pending flag when cmdq packet is done"
> "drm/mediatek: Add cmdq_handle in mtk_crtc"
> "drm/mediatek: Detect CMDQ execution timeout"
>

Yes, in fact I reverted:

9efb16c2fdd6 drm/mediatek: Clear pending flag when cmdq packet is done
bc9241be73d9 drm/mediatek: Add cmdq_handle in mtk_crtc
8cdcb3653424 drm/mediatek: Detect CMDQ execution timeout
f4be17cd5b14 drm/mediatek: Remove struct cmdq_client
c1ec54b7b5af drm/mediatek: Use mailbox rx_callback instead of cmdq_task_cb

Without these patches 5.15-rc2 works again on my platform.

The commit 'c1ec54b7b5af drm/mediatek: Use mailbox rx_callback instead
of cmdq_task_cb' alone introduces lots of warnings in the kernel

WARNING: CPU: 0 PID: 0 at drivers/mailbox/mtk-cmdq-mailbox.c:198
cmdq_task_exec_done+0xb8/0xe0

I think is just a leftover or the mentioned warning, but that confused
me a bit doing the bisection. Then, after commit 'f4be17cd5b14
drm/mediatek: Remove struct cmdq_client' my system simply gets stuck.
For now I don't see any obvious mistake but will dig further.

Can I ask you in which platform did you test? And if you can double
check if your platform is broken too in current mainline?

Thanks,
  Enric

> If "drm/mediatek: Remove struct cmdq_client" is the patch cause
> display abnormal, I think you could compare code w/ and w/o this
> patch. Focus on the value accuracy, such as cmdq_cl and cmdq_chan. And
> focus on the flow accuracy, such as mtk_drm_crtc_update_config() and
> ddp_cmdq_cb(). If this could not find the problem, I think the latest
> way is to break this patch into small patches, changes little in each
> small patches and we could finally find out the problem.
>
> Regards,
> Chun-Kuang.
>
> >
> > Unfortunately, after the merge window, different things broke for this
> > device, and I didn't finish isolating them, and it is not clear to me
> > yet whether the logs I'm getting are useful for this specific issue or
> > not. Basically with this series merged the kernel seems to be stuck,
> > and the display is not working. Latest message is
> >
> > [   12.329173] mtk-iommu 10205000.iommu: Partial TLB flush timed out,
> > falling back to full flush
> >
> > Without the series, the kernel goes far and display works, however
> > there are other issues affecting the cros-ec, but I think that's
> > another issue.
> >
> > I'll try to dig a bit more, but, meanwhile, if you have any idea
> > please let me know.
> >
> > Thanks,
> >  Enric
> >
> >
> > > Regards,
> > > Chun-Kuang.
> > >
> > > >
> > > > Changes in v2:
> > > > 1. Define mtk_drm_cmdq_pkt_create() and mtk_drm_cmdq_pkt_destroy()
> > > >    when CONFIG_MTK_CMDQ is reachable.
> > > >
> > > > Chun-Kuang Hu (4):
> > > >   drm/mediatek: Use mailbox rx_callback instead of cmdq_task_cb
> > > >   drm/mediatek: Remove struct cmdq_client
> > > >   drm/mediatek: Detect CMDQ execution timeout
> > > >   drm/mediatek: Add cmdq_handle in mtk_crtc
> > > >
> > > >  drivers/gpu/drm/mediatek/mtk_drm_crtc.c | 110 ++++++++++++++++++++----
> > > >  1 file changed, 91 insertions(+), 19 deletions(-)
> > > >
> > > > --
> > > > 2.25.1
> > > >