On 02/09/2022 19.16, Alexander Bulekov wrote:
On 220902 1851, Juan Quintela wrote:
Hi
For a long, long time I have had local hacks on my tree to be able to
run "make tests" when I have a minimal configure guest. This is a
first try to upstream some of it.
- by default we always setup -display none (it already was the
default, but some places added it anyways)
- by default we always setup -net none. Not clear what was the
default, but no tests use the default net, so it is safe change and
now it is explicit.
- by default we always setup -vga none. This is a complete difference
can of worms. Every tests that use vga already set vga correctly,
so this is quite obvious, right? Now they are acpi tables. They
are a mess. And basically this means remove a device for each one
of them. Why going through all the trouble? Because while I am
develping, I normall compile out vga.
- Fix several error strings that were set with copy paste.
- replication test requires CONFIG_REPLICATION.
- test-crypto-secret requires CONFIG_SECRET_KEYRING.
Please review. Except for the acpi changes (that I hope I have done
right following the instructions) the rest is quite obvious.
I think this might break some of the fuzz regression tests, because they
have "baked-in" PCI configuration commands with hard-coded PCI
addresses, which will shift around if some device is removed (e.g. with
-net none). Probably the fix is to add addr=... to the -device parameter
in the fuzz tests to keep the PCI address stable.
-Alex
The patches to default to -net none and -vga none are a good idea, but I
agree with Alexander - this needs some careful examination of the fuzz tests
first to see whether the BARs are changed here or not.
Thomas
