Re: [PATCH 0/2] PM / runtime: Allow drivers to override runtime PM behaviour on sleep

From: Daniel Vetter <daniel@ffwll.ch>
To: Thierry Reding <thierry.reding@gmail.com>
Cc: "Rafael J . Wysocki" <rjw@rjwysocki.net>,
	Len Brown <len.brown@intel.com>,
	Linux PM list <linux-pm@vger.kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	dri-devel <dri-devel@lists.freedesktop.org>,
	Pavel Machek <pavel@ucw.cz>,
	linux-tegra@vger.kernel.org, Dmitry Osipenko <digetx@gmail.com>
Subject: Re: [PATCH 0/2] PM / runtime: Allow drivers to override runtime PM behaviour on sleep
Date: Thu, 28 Nov 2019 17:47:05 +0100	[thread overview]
Message-ID: <CAKMK7uFEwChTesKdWAOfi3evLO=d0MsPDdK8gizH-jLk4uBdFg@mail.gmail.com> (raw)
In-Reply-To: <20191128160314.2381249-1-thierry.reding@gmail.com>

On Thu, Nov 28, 2019 at 5:03 PM Thierry Reding <thierry.reding@gmail.com> wrote:
>
> From: Thierry Reding <treding@nvidia.com>
>
> This is a result of looking into a more formal way of doing what was
> proposed here:
>
>         http://patchwork.ozlabs.org/patch/1145363/
>
> The Tegra DRM driver is written such that runtime PM controls all
> aspects of bringing up and shutting down the hardware associated with a
> display pipeline. This works very nicely with the DRM/KMS atomic mode-
> setting framework that has very rigorous call sequences. There are also
> suspend/resume helpers for system sleep that are built on top of these
> generic helpers and that cause the same code sequences to be run as if
> users had simply chosen to disable all display pipelines at normal
> runtime.
>
> The current behaviour of the PM core to disallow runtime suspend/resume
> during system sleep gets in the way of this because the devices do not
> in fact runtime suspend/resume during that time. Most of the time this
> causes display outputs to malfunction upon resume.
>
> Now, there are good reasons for preventing runtime suspend during system
> sleep, as given in commit eea3fc0357eb ("PCI / PM: Detect early wakeup
> in pci_pm_prepare()") that originally introduced this mechanism. There
> can, however, also be cases, like the one described above, where it is
> safe to allow this. Add a flag and a set of helpers to set or clear that
> new flag so that drivers that know it will be safe to runtime suspend a
> device at system sleep time can mark the device as such.
>
> If a device has the flag set, the PM core will no longer take a runtime
> PM reference for it, thus allowing the device to runtime suspend at the
> expected time.

What about sprinkling tons of device_links all over this to make sure
system suspend/resume is done in the same order too? Slightly less
neat from a driver pov, but I think that should get the job done.
Maybe could even do a convenience function which converts a dt phandle
(or whatever that was called again) into a device_link?
-Daniel

> Thierry
>
> Thierry Reding (2):
>   PM / runtime: Allow drivers to override runtime PM behaviour on sleep
>   drm/tegra: Allow runtime suspend on system sleep
>
>  drivers/base/power/main.c    |  6 ++++--
>  drivers/base/power/runtime.c | 16 ++++++++++++++++
>  drivers/gpu/drm/tegra/dc.c   |  1 +
>  drivers/gpu/drm/tegra/dsi.c  |  1 +
>  drivers/gpu/drm/tegra/hdmi.c |  1 +
>  drivers/gpu/drm/tegra/hub.c  |  1 +
>  drivers/gpu/drm/tegra/sor.c  |  1 +
>  include/linux/pm.h           |  1 +
>  include/linux/pm_runtime.h   |  2 ++
>  9 files changed, 28 insertions(+), 2 deletions(-)
>
> --
> 2.23.0
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel


-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch