Re: [PATCH v6 07/14] drm/bridge: analogix_dp: Add support to get panel from the DP AUX bus

Hi,

On Thu, Jan 23, 2025 at 3:25 AM Dmitry Baryshkov
[off-list ref] wrote:

On Thu, Jan 23, 2025 at 06:07:40PM +0800, Damon Ding wrote:

quoted

The main modification is moving the DP AUX initialization from function
analogix_dp_bind() to analogix_dp_probe(). In order to get the EDID of
eDP panel during probing, it is also needed to advance PM operaions to

nit: s/operaions/operations

quoted

ensure that eDP controller and phy are prepared for AUX transmission.

This doesn't sound right. Per the documentation of
drm_dp_aux::transfer(), the device should power itself up if transfer()
is called when it is powered off. The caller must only ensure that the
panel is on.

Doug, what's your opinion?

I think maybe the CL description is a bit confusing, but looking at
the patch I think that the general idea is correct. drm_dp_aux_init()
is expected to be called in probe() and not in bind(). ...and in order
for it to work then pm_runtime needs to be enabled at probe and not at
bind. So both of those two things that this patch does are (in my
opinion) correct.

quoted

In addtion, add a new function analogix_dp_remove() to ensure symmetry
for PM operations.

Signed-off-by: Damon Ding <redacted>

---

Changes in v4:
- Use done_probing() to call drm_of_find_panel_or_bridge() and
  component_add() when getting panel from the DP AUX bus

Changes in v5:
- Advance PM operations to make eDP AUX work well

Changes in v6:
- Use devm_pm_runtime_enable() instead of devm_add_action_or_reset()
- Add a new function analogix_dp_remove() to ensure symmetry for PM
  operations
---
 .../drm/bridge/analogix/analogix_dp_core.c    | 57 ++++++++++---------
 .../gpu/drm/rockchip/analogix_dp-rockchip.c   |  4 ++
 include/drm/bridge/analogix_dp.h              |  1 +
 3 files changed, 34 insertions(+), 28 deletions(-)

diff --git a/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c b/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c
index 8251adfce2f9..30da8a14361e 100644
--- a/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c
+++ b/drivers/gpu/drm/bridge/analogix/analogix_dp_core.c

@@ -1658,14 +1658,42 @@ analogix_dp_probe(struct device *dev, struct analogix_dp_plat_data *plat_data)
      }
      disable_irq(dp->irq);

not related to your patch, but probably needs to be a prerequisite of
your patch: instead of calling disable_irq() here, you should OR in
"IRQF_NO_AUTOEN" to the "irq_flags" of devm_request_threaded_irq().
That not only closes a potential race condition but also makes all the
error handling much more robust.

quoted

+     dp->aux.name = "DP-AUX";
+     dp->aux.transfer = analogix_dpaux_transfer;
+     dp->aux.dev = dp->dev;
+     drm_dp_aux_init(&dp->aux);

FWIW: I would highly encourage you to (in a separate patch) add
wait_hpd_asserted() support here. It is deprecated to not implement
wait_hpd_asserted(). See the definition of "wait_hpd_asserted" in
"struct drm_dp_aux" if you're going to support eDP panels.

quoted

+     if (IS_ENABLED(CONFIG_PM)) {

Do we really truly need this? Is there anyone actually using this
driver without "CONFIG_PM", or can we just assume CONFIG_PM. For the
most part drivers I've interacted with just assume CONFIG_PM and
they're a lot simpler because of it. If there's truly a need then we
can keep this complexity, but I'd rather wait until there is a user.
Maybe you could add this as a dependency in the Kconfig if needed.

quoted

+             pm_runtime_use_autosuspend(dp->dev);
+             pm_runtime_set_autosuspend_delay(dp->dev, 100);
+             ret = devm_pm_runtime_enable(dp->dev);
+             if (ret)
+                     goto err_disable_pm_runtime;
+     } else {
+             ret = analogix_dp_resume(dp);
+             if (ret)
+                     goto err_disable_clk;

IMO: if analogix_dp_resume() succeeds, use devm_add_action_or_reset()
to have a function call analogix_dp_suspend(). Then you can keep using
"devm" for everything and totally get rid of the need for the
analogix_dp_remove() function.

quoted

+     }
+
      return dp;

+err_disable_pm_runtime:
+     pm_runtime_dont_use_autosuspend(dp->dev);

You don't need to call pm_runtime_dont_use_autosuspend(). If you
enabled pm_runtime with devm_pm_runtime_enable() then it's documented
to handle calling pm_runtime_dont_use_autosuspend() for you. See the
kernel doc comment for devm_pm_runtime_enable(). So you can get rid of
this.

quoted

 err_disable_clk:
      clk_disable_unprepare(dp->clock);
      return ERR_PTR(ret);

Huh? Why would you call "err_disable_clk" here? The only thing that
enables the clock is analogix_dp_resume(), right? There's something
fishy here and it predates your patch. I suspect there were problems
in commit f37952339cc2 ("drm/bridge: analogix_dp: handle clock via
runtime PM"). You should fix that in a separate patch before yours.

quoted

+void analogix_dp_remove(struct analogix_dp_device *dp)
+{
+     if (IS_ENABLED(CONFIG_PM))
+             pm_runtime_dont_use_autosuspend(dp->dev);
+     else
+             analogix_dp_suspend(dp);
+}
+EXPORT_SYMBOL_GPL(analogix_dp_remove);

See above. Proper use of "devm" should mean you don't need a remove() function.