linux-kernel - Re: [PATCH] arm64: dts: meson: remove CPU opps below 1GHz for G12B/SM1

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <7ha6exm8ep.fsf@baylibre.com>
Date:   Fri, 11 Feb 2022 12:54:06 -0800
From:   Kevin Hilman <khilman@...libre.com>
To:     Christian Hewitt <christianshewitt@...il.com>
Cc:     Rob Herring <robh+dt@...nel.org>,
        Mark Rutland <mark.rutland@....com>,
        Neil Armstrong <narmstrong@...libre.com>,
        devicetree <devicetree@...r.kernel.org>,
        linux-arm-kernel@...ts.infradead.org,
        linux-amlogic@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] arm64: dts: meson: remove CPU opps below 1GHz for G12B/SM1

Christian Hewitt <christianshewitt@...il.com> writes:

>> On 10 Feb 2022, at 5:31 am, Kevin Hilman <khilman@...libre.com> wrote:
>> 
>> Christian Hewitt <christianshewitt@...il.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.
>> 
>> Just curious: what CPUfreq governor do you use by default for the
>> LibreELEC kernel?
>
> LE uses ondemand. One of the original clues on the problem us that the
> issue isn’t seen in some of the retro-gaming forks on LE's codebase
> which use the performance governor (and overclocks, etc.)

OK, thanks.  And does LE ever do full system suspend/resume?  Are things
stable for you across multiple suspend/resume cycles on G12B or SM1
devices?

I'm seeing hat with either powersave or performance, repeated
suspend/resume is stable, but with ondemand or schedultil it's not, even
with $SUBJECT patch applied.

If you have some time to test, seeing how long this loop[1] runs with
ondemand vs performance or powersave would be instructive.

Even more interesting...  if I set the governor to performance, but set
the suspend OPP to 1GHz[2] (which is what it would be for the powersave
governor), it is also unstable.  This suggests (to me) that any sort of
OPP change during the suspend/resume process is going to be unstable.
Now the challenge is to understand why so we can avoid it.

Thanks,

Kevin

[1]
while true;  do
    echo "=== SUSPEND ==="
    cat /sys/devices/system/cpu/cpufreq/policy0/scaling_governor
    cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq
    cat /sys/devices/system/cpu/cpufreq/policy2/scaling_governor
    cat /sys/devices/system/cpu/cpufreq/policy2/scaling_cur_freq
    echo rtcwake -d rtc0 -m mem -s4
    echo "=== RESUME ==="
    sleep 4
done

[2]
diff --git a/arch/arm64/boot/dts/amlogic/meson-g12b-s922x.dtsi b/arch/arm64/boot/dts/amlogic/meson-g12b-s922x.dtsi
index 1e5d0ee5d541..37da8be85288 100644
--- a/arch/arm64/boot/dts/amlogic/meson-g12b-s922x.dtsi
+++ b/arch/arm64/boot/dts/amlogic/meson-g12b-s922x.dtsi
@@ -14,6 +14,7 @@ cpu_opp_table_0: opp-table-0 {
 		opp-100000000 {
 			opp-hz = /bits/ 64 <100000000>;
 			opp-microvolt = <731000>;
+ 		        opp-suspend;
 		};
 
 		opp-250000000 {
@@ -79,6 +80,7 @@ cpub_opp_table_1: opp-table-1 {
 		opp-100000000 {
 			opp-hz = /bits/ 64 <100000000>;
 			opp-microvolt = <751000>;
+ 		        opp-suspend;
 		};
 
 		opp-250000000 {