On 11.02.2019 16:40, Andrea Bolognani wrote: > On Mon, 2019-02-11 at 08:14 +0000, Nikolay Shirokovskiy wrote: >> On 08.02.2019 17:34, Andrea Bolognani wrote: >>> Even with that fixed, while your code prevents a <console> element >>> associated to the isa-debugcon to be automatically created, it >>> doesn't prevent something like >>> >>> <console type='pty'/> >>> <serial type='file'> >>> <source path='...'/> >>> <target type='isa-serial'> >>> <model name='isa-debugcon'/> >>> </target> >>> </serial> >>> >>> to result in the same problematic configuration, while the user >>> clearly wanted to have both a regular serial console *and* the >>> isa-debugcon. >> >> Yeah I noticed that too but I though this is like case of usb-serial >> for example. We do not add missing console in that case but allow >> existing console to be alias of usb-serial. > > Hm, that looks like a bug. I think we should be consistent about > this: either the aliasing between <serial/> and <console/> works > both ways, or it should not happen at all IMHO. > >> Can usb-serial actually be console? > > No idea :) > >>>> + <serial type='pipe'> >>>> + <source path='/tmp/debugcon'/> >>>> + <target type='isa-serial' port='0'> >>>> + <model name='isa-debugcon'/> >>>> + </target> >>>> + <address type='isa' iobase='0x402'/> >>>> + </serial> >>> >>> So IIUC iobase=0x402 is the de-facto standard for isa-debugcon, >>> right? >>> >>> Assuming that's indeed the case, I would expect libvirt to fill in >>> that value automatically unless the user has provided an iobase >>> explicitly themselves. >> >> I wonder then why qemu uses a different value - 0xe9? > > If that's the default value for isa-debugcon.iobase as far as QEMU > is concerned, then we should use that value in libvirt too. But I > thought you said 0x402 was the default? > AFAIK different firmwares have diferent iobase: OMVF - 0x402 Seabios - 0x402 IPXE - 0xe9 So QEMU just takes iobase of IPXE. May be should just make iobase mandatory instead of taking value of arbitrary firmware? Nikolay
