Re: [PATCH v1 6/6] tests/acceptance: pick a random gdb port for reverse debugging

[Alex Bennée]

Philippe Mathieu-Daudé <f4bug@amsat.org> writes:

> On 10/22/20 7:20 AM, Thomas Huth wrote:
>> On 21/10/2020 18.31, Alex Bennée wrote:
>>> Currently the test randomly fails if you are using a shared machine
>>> due to contention on the well known port 1234. We can ameliorate this
>>> a bit by picking a random non-ephemeral port although it doesn't
>>> totally avoid the problem. While we could use a totally unique socket
>>> address for debugging it's impossible to probe for gdb support of the
>>> feature which makes this a sub-optimal but less fiddly option.
>>>
>>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>>> ---
>>>   tests/acceptance/reverse_debugging.py | 6 ++++--
>>>   1 file changed, 4 insertions(+), 2 deletions(-)
>> 
>> Certainly better than before!
>
> I'd prefer another chardev that tcp, but as you said this is
> already an improvement, so:

We've supported sockets gdb and softmmu emulation for some time:

  -gdb unix:path=gdb.sock,server

the trouble is detecting if the host installed gdb is going to connect
before we try and fail after launching the VM. I think we might get away
with a version probe:

  > luispm: I want to know ahead of time for my scripts if gdb can do a
      "target remote gdb.sock"
  <luispm> ajb-linaro, I don't think so. My guess is that GDB will
      always attempt to stablish a connection if the socket is valid.
  <luispm> ajb-linaro, Can you check the validity of the file before
      invoking GDB?
  <luispm> ajb-linaro, No concept of "is this particular remote target
      available?".
  > luispm: it's not that - I know the socket will exist but the older
      gdb just bombs out trying to read it.
  <luispm> ajb-linaro, Not good.
  <luispm> ajb-linaro, So is this a matter of an older GDB that doesn't
      support using socket files and a newer one that does?
  > luispm: I thought I might probe "help target remote" text but it's
      unchanged between versions
  > luispm: yes
  <luispm> ajb-linaro, I think the code is probably hidden within the
      "target remote" implementation.
  > luispm: and most distro gdb's don't at the moment
  > luispm: if I could work out the version it was added that might help
  <luispm> ajb-linaro, I see some bits of it were reverted at some
      point.
  <luispm> ajb-linaro, Let me check.
  <luispm> ajb-linaro, It looks like GDB 8.3 was the first stable to get
      it.
  > luispm: thanks - I'll see if I can script that up
  <luispm> ajb-linaro, Looks like they initially went with an explicit
      prefix of "unix:" before the socket. But then dropped that in
      favor of autodetecting the socket file.
  <luispm> ajb-linaro, The only thing I get for a GDB that doesn't
      support socket connections if
      "/run/user/1000/at-spi2-QTZBS0/socket: No such device or address."
  <luispm> *is*
  <luispm> This is 8.1 in Ubuntu 18.04.
  <luispm> master GDB says "Remote communication error.  Target
      disconnected.: Connection reset by peer."

>
> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
>
>> 
>> Reviewed-by: Thomas Huth <thuth@redhat.com>
>> 


-- 
Alex Bennée