On Thursday 27 September 2018 05:01 AM, Tony Lindgren wrote:

* Grygorii Strashko [off-list ref] [180926 21:41]:

quoted

On 09/26/2018 11:23 AM, Tony Lindgren wrote:

quoted

+static int __maybe_unused omap_i2c_suspend(struct device *dev)
+{
+	struct omap_i2c_dev *ddata = dev_get_drvdata(dev);
+	int error;
+
+	/* Is device still enabled because of autosuspend? */
+	if (ddata->is_suspended)
+		return 0;

Sry, but you can't do this. There is no sync between suspend and PM runtime.

quoted

+
+	/* Paired with call in omap_i2c_resume() */
+	error = pm_runtime_put_sync_suspend(dev);

This is nop!!! More over, in general you can't predict how many times pm_runtime_get was called and
what's the current value of usage_count.

Hmm yeah that's a good point.

quoted

To make things work the pm_runtime_force_xx() have to be used, or
like with omap_device, platform/bus code have to handle device state
at suspend_no_irq stage.

OK. So looks like the other i2c bus drivers have already solved it
with simply SET_NOIRQ_SYSTEM_SLEEP_PM_OPS which is along the lines
you're suggesting.

The following works for me, does it look OK to you?

This works pretty well with DS0 on am437x-gp-evm. Tested rtc wakeup,
wakeup using uart and touch. All of the work well.

https://pastebin.ubuntu.com/p/RJhJJKhJHX/

Also tested on beaglebone black DS0 works fine. Although there seems to
be an additional warning which was not seen before:

"l4-wkup-clkctrl:0008:0: failed to disable"

log: https://pastebin.ubuntu.com/p/Ns2WRdVkXS/

Also the above warning is present without/with this patch on [0].
No such warnings on am4 though.

[0]
https://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap.git/log/?h=omap-for-v4.20/dt-ti-sysc-tmp-testing

Regards,
Keerthy