omap-i2c: runtime pm issue during suspend to ram

[Thomas Richard <thomas.richard@bootlin.com>]

Hello,

I have a gpio expander (pca953x driver) connected to an i2c controller
managed by the omap-i2c driver.
And I have some issues with pm_runtime_force_suspend/resume during
suspend to ram.
For some reasons, related to hardware design, I need to access to this
gpio expander during suspend_noirq and resume_noirq. So I had to move
the suspend/resume of the pca953x to suspend_noirq/resume_noirq.

The i2c controller is autosuspended when I start the suspend sequence.
In suspend_noirq, I access to one gpio of the expander, so rpm_resume is
called to resume the i2c controller.
And rpm_resume returns an error because disable_depth > 0 [1]. In
suspend_noirq, runtime pm is disabled (disable_depth is incremented when
runtime pm is disabled [2]). So the expander is not reachable, and the
access fails.

[1]
https://elixir.bootlin.com/linux/v6.7-rc6/source/drivers/base/power/runtime.c#L773
[2]
https://elixir.bootlin.com/linux/v6.7-rc6/source/drivers/base/power/runtime.c#L1474

The suspend_noirq of the gpio expander don't do i2c access, so no
problem for pca953x suspend.
The pm_runtime_force_suspend (suspend_noirq [3]) of the i2c controller
does nothing as the device is already suspended [4].

[3]
https://elixir.bootlin.com/linux/v6.7-rc6/source/drivers/i2c/busses/i2c-omap.c#L1579
[4]
https://elixir.bootlin.com/linux/v6.7-rc6/source/drivers/base/power/runtime.c#L1878

Then during the pm_runtime_force_resume (resume_noirq [3]) the i2c
controller is not resumed because needs_for_resume is equal to 0 [5].
The needs_for_resume flag is set in pm_runtime_force_suspend [6] but we
don't reach this point, because the device is already suspended [4].

[5]
https://elixir.bootlin.com/linux/v6.7-rc6/source/drivers/base/power/runtime.c#L1929
[6]
https://elixir.bootlin.com/linux/v6.7-rc6/source/drivers/base/power/runtime.c#L1900

Then the resume_noirq of the pca953x driver is called, consequently
rpm_resume is called to resume the i2c controller. But it is never
resumed because disable_depth > 0 [7] (runtime pm is still disabled in
resume_noirq). So the resume_noirq fails.

[7]
https://elixir.bootlin.com/linux/v6.7-rc6/source/drivers/base/power/runtime.c#L773

I found a workaround which is to resume the controller and disable
runtime pm during suspend, then runtime pm is enabled during resume.
But there is probably a better solution to fix this issue.

Best Regards,

Thomas Richard


diff --git a/drivers/i2c/busses/i2c-omap.c b/drivers/i2c/busses/i2c-omap.c
index 42165ef57946..fe79b27b46fd 100644
--- a/drivers/i2c/busses/i2c-omap.c
+++ b/drivers/i2c/busses/i2c-omap.c
@@ -1575,9 +1575,24 @@ static int __maybe_unused
omap_i2c_runtime_resume(struct device *dev)
        return 0;
 }

+static int omap_i2c_suspend(struct device *dev)
+{
+       pm_runtime_get_sync(dev);
+       pm_runtime_disable(dev);
+       return 0;
+}
+
+static int omap_i2c_resume(struct device *dev)
+{
+       pm_runtime_enable(dev);
+       pm_runtime_put_autosuspend(dev);
+       return 0;
+}
+
 static const struct dev_pm_ops omap_i2c_pm_ops = {
        SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
                                      pm_runtime_force_resume)
+       SET_SYSTEM_SLEEP_PM_OPS(omap_i2c_suspend, omap_i2c_resume)
        SET_RUNTIME_PM_OPS(omap_i2c_runtime_suspend,
                           omap_i2c_runtime_resume, NULL)
 };