Re: Problems interrupting remote target on powerpc

On Fri, 28 Apr 2023, 9:41 PM Luis Machado, <[email protected]> wrote:

> On 4/28/23 10:38, Chris Packham wrote:
> >
> >
> > On Fri, 28 Apr 2023, 9:19 PM Luis Machado, <[email protected]
> <mailto:[email protected]>> wrote:
> >
> >     On 4/28/23 10:14, Chris Packham via Gdb wrote:
> >      > On Thu, 27 Apr 2023, 4:55 PM Chris Packham, <
> [email protected] <mailto:[email protected]>> wrote:
> >      >
> >      >> Hi GDB,
> >      >>
> >      >> I've had a few users report to me issues with interrupting a
> running
> >      >> process after continuing when attached to a remote gdbserver
> running
> >      >> on a powerpc target. Everything seems to work properly on arm and
> >      >> aarch64 so there might be some powerpc or big-endian specific
> issue
> >      >> lurking.
> >      >>
> >      >> The gdbserver version we're currently using is from gdb-11.2
> built
> >      >> from source and most users have gdb-multiarch 12.1 from ubuntu
> 22.04
> >      >> (some might still be using gdb-multiarch 10.2 from a PPA).
> >      >>
> >      >> Does this ring any bells for anyone? I'm going to try and get
> hold of
> >      >> a powerpc target to test with to see if I can reproduce it for
> myself.
> >      >>
> >      >> Thanks,
> >      >> Chris
> >      >>
> >      >
> >      > Did some digging of my own I can report that indeed things aren't
> working
> >      > well for powerpc. I tried updating gdbserver to 12.1 and 13.1.
> 12.1 was
> >      > much the same as 11.2. 13.1 reported that it was unable to send
> sigkill
> >      > when processing the interrupt command. 13.1 also seems not to be
> able to
> >      > respond to the 'info os processes' command from the gdb client
> >      >
> >      >>
> >
> >     What sort of behavior/output do you see?
> >
> >
> > I'll capture some proper output when I'm back in the office but
> basically once I 'continue' I can't get the gdb prompt back with Ctrl+C.
> >
>
> Is it a tight loop by any change? Or is gdb trapped in a long sequence of
> instruction-steps?
>

It's a daemon using a g_main_loop() not sure if that counts as a tight
loop. GDB shouldn't be doing any stepping. I'll also give a process that
runs and finishes when I get a chance to see if there are issues with that.

>