Re: [PATCH 7/7] tests/avocado: ppc64 pseries reverse debugging test

[Nicholas Piggin]

On Mon Jun 26, 2023 at 7:34 PM AEST, Nicholas Piggin wrote:
> On Mon Jun 26, 2023 at 5:49 PM AEST, Pavel Dovgalyuk wrote:
> > On 23.06.2023 15:57, Nicholas Piggin wrote:
> > > pseries can run reverse-debugging well enough to pass basic tests.
> > > 
> > > There is strangeness with reverse-continue possibly relating to a bp
> > > being set on the first instruction or on a snapshot, that causes
> > > the PC to be reported on the first instruction rather than last
> > > breakpoint, so a workaround is added for that for now.
> >
> > It means that the test reveals some kind of a bug in PPC debugging 
> > server implementation.
> > In this case it is better to fix that instead of adding workaround.
>
> I agree, and I'm trying to find the cause it hasn't been easy. I
> thought the test was still interesting because it otherwise seems
> to work well, but hopefully I can find the issue before long.

I found the problem after too much staring at record-replay. QEMU works
perfectly, it is the ppc pseries firmware that branches to its own entry
point address within the recorded trace, and that confuses the test
script.

I'm not sure the best way to work around it. Initial skip works okay but
also maybe(?) a good edge case to test a break on the very first
instruction of the trace even if we don't reverse continue to it.

I will send a new series and propose to add the breakpoints before
seeking to the end of the trace, so we will catch a breakpoint being
executed again.

Thanks,
Nick