Re: ATF testing

[Valery Ushakov <uwe%stderr.spb.ru@localhost>]

On Sat, Feb 24, 2018 at 17:13:09 +0000, Sevan Janiyan wrote:

> I thought I'd start running the test suite on NetBSD/macppc to extend
> the range of testing systems, I'm not fully setup yet but I gave it a
> spin on my 12" PowerBook to see how things go.
> 3 hours later, I have a shiny report which some 5 unexpected failures.
> See:
> https://www.geeklan.co.uk/netbsd-atf/macppc/


* lib/libc/gen/t_siginfo sigbus_adraln

it looks like at least some poweprc cpus allow unaligned accesses and
on those that don't the kernel will take the trap and emulate it.  I
don't see any sysctl that would control this or that would tell
whether unaligned accesses are trapped.

We should probably disable that test.  But see
http://gnats.netbsd.org/49653


* dev/md/t_md basic

The problem is that different cpus have different VM_MAXUSER_ADDRESS,
some have it at ~2GB (shared kernel address space), some at ~4GB
(separate kernel address space), but rump is compiled with "default"
~2GB definition.  So when you run this test on a cpu with ~4GB limit,
the rump sees the malloc'ed address in the upper VA region and that
triggers end >= VM_MAXUSER_ADDRESS check.

-uwe