On Tue, Apr 17, 2012 at 12:00, Anthony Liguori <anth...@codemonkey.ws> wrote: > On 04/17/2012 02:44 AM, Paolo Bonzini wrote: >> >> Il 17/04/2012 09:33, Peter Maydell ha scritto: >>>> >>>> I think the issue is that all of these machines mandate a -kernel >>>> option. >>>> qtest doesn't care if you pass a -kernel but requiring a kernel in >>>> order to >>>> test a device sucks especially if you don't possess the toolchain to >>>> build >>>> such a kernel. >>> >>> >>> Just testing a device shouldn't require running a particular >>> board model either, of course. >> >> >> qtest isn't so much testing a device. It is really testing a board with >> improved determinism, debuggability and logging. >> >> That said, I think a simpler solution is to set kernel/initrd/dtb to >> "/dev/null" if qtest is enabled. > > > Why not refactor the check to be a QEMUMachine property (requires_kernel). > That way the check can be moved common code and that common code can have a > single check where it sets the parameters to /dev/null if qtest is enabled.
The same problem exists in addition to kernel/initrd/dtb, also with BIOS, other ROMs and blobs. Perhaps this could be handled at lower level instead of higher: make load_elf/load_image fake success when using qtest. > > Regards, > > Anthony Liguori > >> >> Paolo >> >
