Re: [PATCH v12 1/1] serial: core: Start managing serial controllers to enable runtime PM

From: Tony Lindgren <tony@atomide.com>
To: John Ogness <john.ogness@linutronix.de>
Cc: "Chen-Yu Tsai" <wenst@chromium.org>,
	"Greg Kroah-Hartman" <gregkh@linuxfoundation.org>,
	"Jiri Slaby" <jirislaby@kernel.org>,
	"Andy Shevchenko" <andriy.shevchenko@intel.com>,
	"Dhruva Gole" <d-gole@ti.com>,
	"Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>,
	"Johan Hovold" <johan@kernel.org>,
	"Sebastian Andrzej Siewior" <bigeasy@linutronix.de>,
	"Vignesh Raghavendra" <vigneshr@ti.com>,
	linux-omap@vger.kernel.org,
	"Andy Shevchenko" <andriy.shevchenko@linux.intel.com>,
	linux-kernel@vger.kernel.org, linux-serial@vger.kernel.org,
	"Nícolas F. R. A. Prado" <nfraprado@collabora.com>,
	"AngeloGioacchino Del Regno"
	<angelogioacchino.delregno@collabora.com>,
	linux-mediatek@lists.infradead.org
Subject: Re: [PATCH v12 1/1] serial: core: Start managing serial controllers to enable runtime PM
Date: Sat, 3 Jun 2023 08:41:39 +0300	[thread overview]
Message-ID: <20230603054139.GR14287@atomide.com> (raw)
In-Reply-To: <87a5xii33r.fsf@jogness.linutronix.de>

Hi,

* John Ogness <john.ogness@linutronix.de> [230602 10:13]:
> Unfortunately next-20230601 also brought in a series that added
> spinlocking to the 8250 driver. That may be the issue here instead.

I think you're off the hook here with the spinlocking changes :)

My guess right now is that 8250_mtk does not want runtime PM resume called
on probe for some reason, and assumes it won't happen until until in
mtk8250_do_pm()?

Looking at the probe, the driver does pm_runtime_enable(), but then calls
mtk8250_runtime_resume() directly. Not sure what the intention here is.
Maybe adding pm_runtime_set_active() in probe might provide more clues.

When we add the new serial_ctrl and serial_port devices, their runtime PM
functions propagate to the parent 8250_mtk device. And then something goes
wrong, well this is my guess on what's going on..

To me it seems the 8250_mtk should just do pm_runtime_resume_and_get()
instead of mtk8250_runtime_resume(), then do pm_runtime_put() at the end
of the probe?

I don't think 8250_mtk needs to do register access before and after the
serial port registration, but if it does, then adding custom read/write
functions can be done that do not rely on initialized port like
serial_out().

Looking at the kernelci.org test boot results for Linux next [0], seems
this issue is somehow 8250_mtk specific. I don't think the rk3399 boot
issue is serial port related.

Regards,

Tony

[0] https://linux.kernelci.org/test/job/next/branch/master/kernel/next-20230602/plan/baseline/