Re: NetBSD/hpcmips userland crashes (Re: IBM Workpad z50 dmesg collection)

[David Brownlee <abs%absd.org@localhost>]

On Sun, 26 Mar 2023 at 02:46, B 9 <hackerb9%gmail.com@localhost> wrote:
>
> On Fri, Mar 24, 2023 at 2:28 PM David Brownlee abs%absd.org@localhost wrote:
>>
>> On Fri, 24 Mar 2023 at 14:50, Michael <allstarzero%gmail.com@localhost> wrote:
>> >
>> > Hi! I have a MobilePro 780 and a MobilePro 790 I can test on.
>>
>>  that could definitely be two useful data points :)
>
> I am dusting off my Mobile Pro 800 which still happily runs NetBSD-1.6. I’d been thinking I’d upgrade it before finding someone who would want it as a retro-computer. Unfortunately, trying to install NetBSD-9.3 onto a new compact flash card has been impossible because all the sysinst tools keep dumping core, almost randomly. (The ping program is an exception as it always dies with a floating point exception when I hit ^C.)
>
> For example, if I start up vi and hit ^Z to suspend the process, most of the time it works, but often vi gets killed.
>
> I noticed that the problem seems to occur when programs were sent a benign signal, but they would die claiming some other signal had been received. I wrote a simple script, itsatrap, that shows which signal was received. It always reported the signal correctly, when it could catch it, but after acting on it, sometimes its parent process would die with a random signal (“Unknown signal number”, “Abort”, “Illegal instruction”, “CPU over time limit”). For example: If I run chroot /targetdisk then run itsatrap & and send it a signal via kill -1 %1, the chroot process could die with SIGABRT.

That is consistent with what others are seeing on netbsd-8 and later
-netbsd-7 seems to be the last version which runs well on some hpcmips
devices.

We (mainly Alex) are trying to track down when the issue started by
starting with a netbsd-7 install and then testing different kernels
built from source between netbsd-7 and netbsd-8.

So far we've narrowed it down to between 2016-07-01 and 2016-08-01,
during which there were a few potential candidates for "breaking
change". Most likely of them would be the switch of mips to the
generic pmap code on July 11th -
https://github.com/NetBSD/src/commit/f694fbd0571344422f046e53019ab383c37b8816

I have a script building the hpcmips toolchain and GENERIC kernel for
each day between 2016-07-01 and 2016-08-01 in a netbsd-8 chroot and
uploading to sync.absd.org, it's up to 2016-07-25 now, and should be
completed in a few more hours.

If anyone following along would like to help testing :)

- Confirm netbsd-7 install works on your device (install and boot up
multiuser is a good test)
- Download and test boot 2016-07-11 kernel from
http://sync.absd.org/hpcmips/ (with the existing netbsd-7 install)
- If it works, try 2016-07-12, and similar up to 2016-08-01 to see if
you get a failure (*)
- If it fails, try 2016-07-10, and similar back to 2016-07-01 to see
if you get a success (*)

*: Rather than trying each kernel in the sequence, use a binary chop -
pick the one in the middle of the range, and based on whether that
works pick the one in the middle of the top or bottom range and
repeat. Make sure you keep a note of each tested kernel and the
result. For reference, look up "git bisect" :)

David