Re: [Qemu-devel] Default for phys-addr-bits? (was Re: [PATCH 4/5] x86: Allow physical address bits to be set)

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Gerd Hoffmann (kraxel@redhat.com) wrote:
> > So that's mapped at an address beyond host phys-bits.
> > And it hasn't failed/crashed etc - but I guess maybe nothing is using that 2G space?
> 
> root@fedora ~# dmesg | grep Surface
> [    4.830095] [drm] qxl: 2048M of Surface memory size
> 
> qxl bar 4 (64bit) and qxl bar 1 (32bit) are the same thing.  The 64bit
> bar can be alot larger obviously.  The 32bit bar is just an alias for
> the first portion of the 64bit bar.  So I guess qxl just falls back to
> use bar 1 instead of bar 4 because ioremap() on bar 4 fails.

Hmm for me it's saying it mapped 64M on the setup with 64T maxmem and 48bit phys-bits;
even though the bar is showing it as OK; how is the guest ioremap detecting a problem?

> > Obviously 128T is a bit silly for maxmem at the moment, however I was worrying what
> > happens with 36/39/40bit hosts, and it's not unusual to pick a maxmem that's a few TB
> > even if the VMs you're initially creating are only a handful of GB. (oVirt/RHEV seems to use
> > a 4TB default for maxmem).
> 
> Oh, ok.  Should be fixed I guess.
> 
> > Still, this only hits as a problem if you hit the combination of:
> >    a) You use large PCI bars
> 
> ovmf will map all 64bit bars high, even without running out of 32bit
> address space.  And with virtio 1.0 pretty much every virtual machine
> will have 64bit bars.

Hmm OK let me think about this; using current BIOS+old virtio+host phys-bits
on a 36bit host, with maxmem=4T would apparently work because
nothing would actually get mapped that high.

The same with OVMF would work as well, because generally you wouldn't have
any 64bit bars; but then you turn on virtio 1.0 and.. well then what happens?
The guest sees it's 36bit phys-address bits so linux probably drops the
bars associated with the virtio?  Hmm.

With current downstream qemu's 40bit physical address bit you'd
get those bar's mapped; so it might break badly - except if
we can figure out what causes my 2GB qxl bar not to happen
as at the start of this message.

Dave

> cheers,
>   Gerd
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK