tests/avocado: reverse_debugging drain console to prevent hang
commitcd43f00524070c0267613acc98a153dba0e398d9
authorNicholas Piggin <npiggin@gmail.com>
Thu, 16 Nov 2023 11:53:52 +0000 (16 21:53 +1000)
committerThomas Huth <thuth@redhat.com>
Thu, 16 Nov 2023 13:22:28 +0000 (16 14:22 +0100)
tree9ed93b6c6acae80b8011b133eb17edf45d78d1ca
parent572960cb23217fb23e23e5a0cb315ac263c9d87c
tests/avocado: reverse_debugging drain console to prevent hang

Like replay_linux.py, reverse_debugging.py starts the vm with console
set but does not interact with it (e.g., with wait_for_console_pattern).
In this situation, the console should have a drainer attached so the
socket does not fill. replay_linux.py has a drainer, but it is missing
from reverse_debugging.py.

Per analysis in Link: this can cause the console socket/pipe to fill and
QEMU get stuck in qemu_chr_write_buffer, leading to strange test case
failures (ppc64 fails because it prints a lot to console in early bios).
Attaching a drainer prevents this.

Note, this commit does not fix bugs introduced by the commits referenced
in the first two Fixes: tags, but together those commits conspire to
irritate the problem and cause test case failure, which this commit
fixes.

Link: https://lore.kernel.org/qemu-devel/ZVT-bY9YOr69QTPX@redhat.com/
Fixes: 1d4796cd0083 ("python/machine: use socketpair() for console connections")
Fixes: 761a13b23946 ("tests/avocado: ppc64 reverse debugging tests for pseries and powernv")
Fixes: be52eca30978 ("tests/acceptance: add reverse debugging test")
Tested-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-ID: <20231116115354.228678-1-npiggin@gmail.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
tests/avocado/reverse_debugging.py