[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAG-rBijSASfbfWQNarjGqj2UxQDOSdwM-qj5YA5A9ur=DNJf-g@mail.gmail.com>
Date: Mon, 31 Oct 2022 21:38:42 -0400
From: Sven van Ashbrook <svenva@...omium.org>
To: "Limonciello, Mario" <Mario.Limonciello@....com>
Cc: Rajneesh Bhardwaj <irenic.rajneesh@...il.com>,
Hans de Goede <hdegoede@...hat.com>,
LKML <linux-kernel@...r.kernel.org>,
"S-k, Shyam-sundar" <Shyam-sundar.S-k@....com>,
"rrangel@...omium.org" <rrangel@...omium.org>,
"platform-driver-x86@...r.kernel.org"
<platform-driver-x86@...r.kernel.org>,
Rajneesh Bhardwaj <rajneesh.bhardwaj@...el.com>,
Rafael J Wysocki <rjw@...ysocki.net>,
Rajat Jain <rajatja@...gle.com>,
David E Box <david.e.box@...el.com>,
Mark Gross <markgross@...nel.org>
Subject: Re: [PATCH v1] platform/x86: intel_pmc_core: promote S0ix failure
warn() to WARN()
On Mon, Oct 31, 2022 at 3:39 PM Limonciello, Mario
<Mario.Limonciello@....com> wrote:
>
> Just thinking about it a little bit more, it could be a lot nicer to have something like:
>
> /sys/power/suspend_stats/last_hw_deepest_state
While I agree that reporting through a framework is generally better
than getting infrastructure to grep for specific strings, I believe
that a simple sysfs file is probably too simplistic.
1. We need more sophisticated reporting than just last_hw_deepest_state:
- sometimes the system enters the deep state we want, yet after a
while moves back up and gets "stuck" in an intermediate state (below
S0). Or, the system enters the deep state we want, but moves back to
S0 after a time without apparent reason. These platform-dependent
failures are not so easily describable in a generic framework.
- ChromeOS in particular has multiple independent S0ix / S3 / s2idle
failure report sources. We have the kernel warning above; also our
Embedded Controller monitors suspend failure cases which the simple
kernel warning cannot catch, reported through a separate WARN_ONCE().
2. A simple sysfs file will need to be polled by the infrastructure
after every suspend; it would be preferable to have some signal or
callback which the infrastructure could register itself with.
The generic infrastructure to support this sounds like quite a bit of
work, and for what gain? Compared to simply matching a log string and
sending the whole dmesg if there's a match.
Is the light worth the candle?
Sven
Powered by blists - more mailing lists