[PATCH] mvebu: 5.10: fix DVFS caused random boot crashes

[Robert Marko]

On Tue, May 18, 2021 at 8:36 PM Hauke Mehrtens <hauke at hauke-m.de> wrote:
>
> On 5/18/21 7:03 PM, Robert Marko wrote:
> > 5.10.37 introduced a lot of DVFS changes for Armada 37xx from 5.13 kernel.
> >
> > Unfortunately commit:
> > https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/drivers/cpufreq/armada-37xx-cpufreq.c?h=v5.10.37&id=a13b110e7c9e0dc2edcc7a19d4255fc88abd83cc
> >
> > This patch actually corrects the things so that 1 or 1.2GHz models would actually get scaled to their native frequency.
> >
> > However, due to a AVS setting voltages too low this will cause random crashes on 1.2GHz models.
> >
> > So, until a new safe for everybody voltage is agreed on
> > lets revert the patch.
> >
> > Fixes: d337731 ("kernel: bump 5.10 to 5.10.37")
> > Signed-off-by: Robert Marko <robert.marko at sartura.hr>
>
> This commit was also backported to v5.4.119 and is now in OpenWrt 21.02
> branch, we should probably also revbert it there.

Ok, I can make a patch for that as well as part of v2.
>
> @Robert Are you working with upstream Linux to fix this problem upstream
> too? For now this revert should be fine.

Yes, I was talking to Pali who did most of the CPUFreq changes to see
what can be done.
Essentially, it seems like Marvell is not cooperating at all, their
docs state incorrect recommended voltage, etc.
I was only able to get my board stable under 30m stress test with 1.2V
which is listed as max voltage but in the boards
OTP 1.26V is hardcoded as the starting voltage, so it's currently a mess.

Robert
>
> hauke



-- 
Robert Marko
Staff Embedded Linux Engineer
Sartura Ltd.
Lendavska ulica 16a
10000 Zagreb, Croatia
Email: robert.marko at sartura.hr
Web: www.sartura.hr