Re: APU1 bricked on stable/14 - solved

On Sun, Oct 6, 2024, 1:14 PM Stefan Hegnauer <stefan.hegnauer@gmx.ch> wrote:

> On 06.10.2024 15:47, Warner Losh wrote:
>
>
>
> On Sun, Oct 6, 2024 at 6:35 AM Stefan Hegnauer <stefan.hegnauer@gmx.ch>
> wrote:
>
>> I have a few pc-engines APU1 appliances running in headless mode under
>> Nanobsd. Maintenance is by means of  direct COM port connection.
>> After a recent update a few weeks back I was not able to connect by COM
>> port anymore - console output and input went away after booting and
>> before single- or multi-user mode would start. Even re-flashing the
>> SDcard with a fresh image did not help.
>>
>> After some longish trials and errors it turned out that both
>> - commit 74b9fc7a 'amd64 GENERIC: Switch uart hints from "isa" to
>> "acpi"' as well as
>> - commit 4ba4cfaf 'acpi: Narrow workaround for broken interrupt settings
>> on x86'
>> broke things for me. Restoring hints and setting
>> hw.acpi.override_isa_irq_polarity=1 in /boot/loader.conf.local restored
>> working order.
>>
>> I agree that APU1 is EOL, however I would have expected an entry to
>> UPDATING for such a POLA violation.
>>
>
> Likely, but really really old gear like this is going to hit edge cases
> that
> developers haven't seen. The hint thing wasn't though to actually
> negatively
> affect any deployed hardware since it dealt with details that changed 15 or
> 20 years ago...
>
> Looks like the APU1 used coreboot which at the time was trailing adaptation
> of ACPI, so it makes sense... I knew that Soekris boxes had issues, but
> they
> are another 5 or 10 years older than the APUs and mine sadly isn't
> operational.
>
> So I can write a better UPDATING entry, can you share with me the dmesg
> from both APU1 and APU2?
>
> Warner
>
>
>> Note that pc-engines APU2 models are not affected as the BIOS ACPI
>> tables contain correct UART descriptions.
>>
>> - Stefan
>>
>> Warner: thanks for the quick reply. Not so really, really old in my view
> - the BIOS is from ~August 2022 (APU1). And yes, it uses coreboot:
>     PC Engines apu1
>     coreboot build 20220822
>     BIOS version v4.17.0.3
>     SeaBIOS (version rel-1.16.0.1-0-g77603a32)
>

Yea. ACPI started being reliable in the early 2000s, +/-. Nearly everything
sonce 2010 has provided it....  Here's clearly an exception that needs
special care so we can make newer hw more reliable... My comments are more
to explain the regression not to try to assign blame on the hardware..

From verbose boot, dmesg -a:
> - APU1, original/faulty from today:
> https://hegnauer.selfhost.eu/web/computing/apu1c4_3mdeb_v4_17_0_3_std.txt
> - APU1, fixed in loader.conf.local:
> https://hegnauer.selfhost.eu/web/computing/apu1c4_3mdeb_v4_17_0_3_fixed.txt
> - APU2 (has no problems):
> https://hegnauer.selfhost.eu/web/computing/apu2d4_3mdeb_v4_19_0_1.txt
>

Thanks. I think between the two I know what to write.

Warner


> Let me know if you need additional information.
>
> - Stefan
>