[ipxe-devel] eepro100 vs qemu

[Alex Williamson]

On Tue, 2011-03-15 at 02:13 +0000, Michael Brown wrote:
> On Tuesday 15 Mar 2011 01:25:51 Michael Brown wrote:
> > On Monday 14 Mar 2011 19:57:35 Alex Williamson wrote:
> > > Ok, sorry for the distraction, I think I can explain it now.  Last week
> > > I was testing large (8G) guests.  Today I was trying to figure out how
> > > the pcnet32 image ever worked because I couldn't make the old gpxe or
> > > the new ipxe image work.  Eventually I discovered that it worked if I
> > > used a smaller guest memory size (512M).  I then used that to test all
> > > the images under qemu, followed by kvm.  Unfortunately nothing boots
> > > under kvm with a 512M guest.  In fact, I can only boot these ROMs if the
> > > guest has 3.6G or more (note that qemu/kvm currently has a 512MB I/O
> > > hole, so this means there's memory bounding the hole on both sides).
> > >
> > > This was introduced with the commit above, 132c3917.  So we've still got
> > > a blocker for refreshing the current ROMs.  Thanks,
> > 
> > Thanks for this.  I can reproduce a PMM failure using the current qemu git
> > tree with -enable-kvm.
> 
> I think this should be fixed now.  It was a rather nasty bug, in which different 
> rounding granularities resulted in writing over the BIOS's PMM data 
> structures:
> 
>   http://git.ipxe.org/ipxe.git/commitdiff/23daab1
> 
> I have no idea why this shows up only with KVM enabled.  It should affect KVM 
> and non-KVM equally, unless SeaBIOS behaves differently when KVM is enabled.
> 
> Thanks for tracking this down.

Even better, thanks for fixing it!  I'll do some more testing, but I
think that's the last regression.  The only nit I have is that pcnet32
doesn't work with more than 2048MB in the guest (maybe a signed int
somewhere?), but the same seems to be true of the older v0.9.9 gpxe rom
we're currently using, so doesn't seem like it should block us.  Thanks
for all your help,

Alex