On 18.05.2022 16:05, Marek Szyprowski wrote:
Hi Dave,
On 11.05.2022 17:47, Dave Stevenson wrote:
On Wed, 11 May 2022 at 15:58, Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> wrote:
On 05.04.2022 13:43, Dave Stevenson wrote:
On Fri, 18 Mar 2022 at 12:25, Dave Stevenson
<dave.stevenson@xxxxxxxxxxxxxxx> wrote:
On Fri, 4 Mar 2022 at 15:18, Dave Stevenson
<dave.stevenson@xxxxxxxxxxxxxxx> wrote:
Hi All
A gentle ping on this series. Any comments on the approach?
Thanks.
I realise the merge window has just closed and therefore folks have
been busy, but no responses on this after a month?
Do I give up and submit a patch to document that DSI is broken and no one cares?
Thanks for pointing this patchset in the 'drm: bridge: Add Samsung MIPI
DSIM bridge' thread, otherwise I would miss it since I'm not involved
much in the DRM development.
This resolves most of the issues in the Exynos DSI and its recent
conversion to the drm bridge framework. I've added the needed
prepare_upstream_first flags to the panels and everything works fine
without the bridge chain order hacks.
Feel free to add:
Tested-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
Thanks for testing it. I was almost at the stage of abandoning the patch set.
The only remaining thing to resolve is the moment of enabling DSI host.
The proper sequence is:
1. host power on, 2. device power on, 3. host init, 4. device init, 5.
video enable.
#1 is done in dsi's pre_enable, #2 is done in panel's prepare. #3 was so
far done in the first host transfer call, which usually happens in
panel's prepare, then the #4 happens. Then video enable is done in the
enable callbacks.
What's your definition of host power on and host init here? What state
are you defining the DSI interface to be in after each operation?
Well, lets start from the point that I'm not a DSI specialist nor I'm
not the exynos-dsi author. I just played a bit with the code trying to
restore proper driver operation on the various Exynos based boards I have.
By the host/device power on I mean enabling their power regulators. By
host init I mean executing the samsung_dsim_init() function, which
basically sets the lp-11 state if I understand it right.
Jagan wants to move it to the dsi host pre_enable() to let it work with
DSI bridges controlled over different interfaces
(https://lore.kernel.org/all/20220504114021.33265-6-jagan@xxxxxxxxxxxxxxxxxxxx/
).
I think I'm in agreement with Jagan.
As documented in patch 4/4:
+ * A DSI host should keep the PHY powered down until the pre_enable
operation is
+ * called. All lanes are in an undefined idle state up to this point, and it
+ * must not be assumed that it is LP-11.
+ * pre_enable should initialise the PHY, set the data lanes to LP-11, and the
+ * clock lane to either LP-11 or HS depending on the mode_flag
+ * %MIPI_DSI_CLOCK_NON_CONTINUOUS.
Right, this theory makes sense.
However Exynos DSI for some reasons did the host initialization in the
first call of the samsung_dsim_host_transfer().
It was a way to fit exynos DSI order of initialization to the frameworks
present at the time it was written:
exynos_dsi was an drm_encoder, and it was connected to drm_panel (DSI
controlled panel).
1. In exynos_dsi_enable host was powered on then drm_panel_prepare was
called.
2. drm_panel_prepare called .prepare callback of the panel, which
powered on the panel and then requested dsi transfers to initialize panel.
3. 1st dsi transfer performed dsi host init (LP-11), after that all
transfers started by panel performed panel initialization.
4. after that control goes back to exynos_dsi_enable.
5. exynos_dsi_enable starts video transfer and notify panel about it via
drm_panel_enable.
6. .enable callback of the panel starts displaying frames (after some
delay).
This construction allowed to keep proper order of hw initialization:
host power on, panel power on, host init, panel init, start video
transmission, start display frames.
Almost all elements of this sequence are enforced by DSI specification
(at least for devices controlled via dsi).
The only thing which I am not sure is placement of "panel power on". It
does not seem to be a part of DSI spec, but I am not 100% sure.
It could be specific to platform (mis)configuration (regulators, level
shifters, ...), or just dsi host init sequence tries to do too much.
I wonder if dropping checking stop state in exynos_dsi wouldn't solve
the mystery [1], or moving it to 1st transfer, maybe with(or w/o)
subsequent code.
Does IMX have some 'vendor code' of DSI host to compare with?
[1]:
https://elixir.bootlin.com/linux/v5.15.1/source/drivers/gpu/drm/exynos/exynos_drm_dsi.c#L860
Regards
Andrzej
If I moved the host
initialization to pre_enable (before powering the panel on), executing
DSI commands failed (timeout). This issue happens on all boards I have
access (Trats, Trats2, Arndale, TM2e), so this must be an issue with
Exynos DSI host itself not related to particular panel/bridge.
If I call samsung_dsim_init() once again, before issuing the first DSI
command, then everything works fine. I've tried to check which part of
that function is needed to be executed before transferring the commands,
but it turned out that the complete host reset and (re)configuration is
necessary. It looks that the initialization will need to be done twice,
first time in the pre_enable to satisfy Jagan case, then on the first
dsi transfer to make it work with real DSI panels.
Here is a git repo with such change:
https://github.com/mszyprow/linux/tree/v5.18-next-20220511-dsi-rework-v2
This however fails on Exynos with DSI panels, because when dsi's
pre_enable is called, the dsi device is not yet powered.
Sorry, I'm not following what the issue is here? DSI lanes being at
LP-11 when the sink isn't powered, so potential for leakage through
the device?
I also have no idea why sending DSI commands fails if host is
initialized without device being powered up first. Maybe powering it
later causes some glitches on the lines? However it looks doing the
initialization again on the first transfer is enough to fix it.
In which case the device should NOT set pre_enable_upstream first, and
the host gets powered up and down with each host_transfer call the
device makes in pre_enable.
Doing the initialization on each host_transfer also is not an option,
because in such case the panel is not initialized properly. I get no
errors, but also there is no valid display on the panel in such case.
(Whilst I can't claim that I know of every single DSI device, most
datasheets I've encountered have required LP-11 on the lanes before
powering up the device).
I've discussed
this with Andrzej Hajda and we came to the conclusion that this can be
resolved by adding the init() callback to the struct mipi_dsi_host_ops.
Then DSI client (next bridge or panel) would call it after powering self
on, but before sending any DSI commands in its pre_enable/prepare functions.
You may as well have a mipi_dsi_host_ops call to fully control the DSI
host state, and forget about changing the pre_enable/post_disable
order. However it involves even more changes to all the panel and
bridge drivers.
True. Although setting prepare_upstream_first/pre_enable_upstream_first
flags also requires to revisit all the dsi panels and bridges.
It looks that I've focused too much on finding a single place of the dsi
initialization, what resulted in that host_init callback. I can live
without it, doing the initialization twice.
If you've added an init hook, don't you also need a deinint hook to
ensure that the host is restored to the "power on but not inited"
state before device power off? Currently it feels unbalanced, but
largely as I don't know exactly what you're defining that power on
state to be.
So far I had no use case for that deinit hook.
Best regards
