On 13/09/2024 13.46, Peter Maydell wrote:
On Fri, 13 Sept 2024 at 12:29, Markus Armbruster <arm...@redhat.com> wrote:
Peter Maydell <peter.mayd...@linaro.org> writes:
Specifically, if you don't disable the error-exit when qtest
is in use, then the generic qom-test tests which say "can we
at least instantiate every machine?" will fail, because they
assume that "qemu-system-foo -machine bar -accel qtest" will
at least start.
It doesn't really seem feasible to me to have qom-test
know about every machine's specific requirements for
how to pass a guest image.
Yes.
The other approach would be to standardize on "every machine
type should happily start up with no warnings even if there
is no guest code specified by the user and it would simply
execute zeroes". We already do this for quite a lot of
boards, including some major ones, so we're certainly not
consistent about trying to diagnose user errors in this area.
IMHO executing zeros is also a bad idea ... most of those boards crash after
a while when the program counter reaches an unmapped memory region.
Maybe we could simply put a "branch to self" instruction on the first
program counter address in case the kernel/firmware cannot be loaded?
Fatal error unless qtest is bad, because we take a different path.
Silently executing zero can be hard for users to diagnose.
Possible compromise: warn unless qtest?
That runs into the "tests that pass and do what they're
supposed to do shouldn't provoke warnings" unofficial
guideline... Some of these qtest_enabled() checks are
exactly to suppress a warning.
FWIW, I like the idea of having a error_report_user() function that is
silent when running with qtest_enabled().
Thomas
