Re: [kvm-unit-tests v3 00/13] powerpc: updates, P10, PNV support

["Nicholas Piggin" <npiggin@gmail.com>]

On Tue Mar 28, 2023 at 2:09 AM AEST, Cédric Le Goater wrote:
> On 3/27/23 14:45, Nicholas Piggin wrote:
> > This series is growing a bit I'm sorry. v2 series added extra interrupt
> > vectors support which was actually wrong because interrupt handling
> > code can only cope with 0x100-size vectors and new ones are 0x80 and
> > 0x20. It managed to work because those alias to the 0x100 boundary, but
> > if more than one handler were installed in the same 0x100-aligned
> > block it would crash. So a couple of patches added to cope with that.
> > 
>
> I gave them a try on P9 box

Thanks!

>
> $ ./run_tests.sh
> PASS selftest-setup (2 tests)
> PASS spapr_hcall (9 tests, 1 skipped)
> PASS spapr_vpa (13 tests)
> PASS rtas-get-time-of-day (10 tests)
> PASS rtas-get-time-of-day-base (10 tests)
> PASS rtas-set-time-of-day (5 tests)
> PASS emulator (4 tests)
> PASS h_cede_tm (2 tests)
> FAIL sprs (75 tests, 1 unexpected failures)

Oh you have a SPR failure too? I'll check that on a P9.

> FAIL sprs-migration (75 tests, 5 unexpected failures)
>
> And with TCG:
>
> $ ACCEL=tcg ./run_tests.sh
> PASS selftest-setup (2 tests)
> PASS spapr_hcall (9 tests, 1 skipped)
> FAIL spapr_vpa (13 tests, 1 unexpected failures)
>
> The dispatch count seems bogus after unregister

Yeah, that dispatch count after unregister test may be bogus actually.
PAPR doesn't specify what should happen in that case. It was working
here for me though so interesting it's different for you. I'll
investigate it and maybe just remove that test for now.

>
> PASS rtas-get-time-of-day (10 tests)
> PASS rtas-get-time-of-day-base (10 tests)
> PASS rtas-set-time-of-day (5 tests)
> PASS emulator (4 tests)
> SKIP h_cede_tm (qemu-system-ppc64: TCG cannot support more than 1 thread/core on a pseries machine)
> FAIL sprs (75 tests, 16 unexpected failures)

These should be TCG errors. I have it passing them all with patches
posted to qemu lists. Very simple but effective way to catch a few
classes of errors.

Thanks,
Nick