Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
From:   Kevin Hilman <khilman@baylibre.com>
To:     Christian Hewitt <christianshewitt@gmail.com>,
        Rob Herring <robh+dt@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Neil Armstrong <narmstrong@baylibre.com>,
        devicetree@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-amlogic@lists.infradead.org, linux-kernel@vger.kernel.org
Cc:     Christian Hewitt <christianshewitt@gmail.com>
Subject: Re: [PATCH] arm64: dts: meson: remove CPU opps below 1GHz for G12B/SM1
In-Reply-To: <20220209135535.29547-1-christianshewitt@gmail.com>
References: <20220209135535.29547-1-christianshewitt@gmail.com>
Date:   Wed, 09 Feb 2022 12:46:21 -0800
Message-ID: <7ha6ezojj6.fsf@baylibre.com>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

Christian Hewitt <christianshewitt@gmail.com> writes:

> Amlogic G12B and SM1 devices experience CPU stalls and random board
> wedges when the system idles and CPU cores clock down to lower opp
> points. Recent vendor kernels include a change to remove 100-250MHz
> (with no explanation) [0] but other downstream sources also remove
> the 500/667MHz points (also with no explanation). Unless 100-667Mhz
> opps are removed or the CPU governor forced to performance, stalls
> are observed, so let's remove them an improve stability/uptime.
>
> [0] https://github.com/khadas/linux/commit/20e237a4fe9f0302370e24950cb1416e038eee03

hehe, not a very helpful changelog in that khadas kernel commit :(

> Signed-off-by: Christian Hewitt <christianshewitt@gmail.com>
> ---
> Numerous people have experienced this issue and I have tested with
> only the low opp-points removed and numerous voltage tweaks: but it
> makes no difference. With the opp points present an Odroid N2 or
> Khadas VIM3 reliably drop off my network after being left idling
> overnight with UART showing a CPU stall splat. With the opp points
> removed I see weeks of uninterupted uptime. It's beyond my skills
> to research what the cause of the stalls might be, but if anyone
> ever figures it out we can always restore things. NB: This issue
> is not too widely reported in forums, but that's largely because
> most of the Amlogic supporting distros have been including this
> change picked from my kernel patchset for some time.

Very interesting.  I've also noticed instability across suspend resume
on VIM3/VIM3L and only got as far in debugging to noticing it was
DVFS/OPP related, but didn't get much further yet.  I'll give this a try
to see if it helps.

Thanks for finding & posting!

Kevin