Re: xen 4 only seeing one keyboard and mouse

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Sun, Aug 08, 2010 at 09:16:56PM +0100, M A Young wrote:
> On Sun, 8 Aug 2010, M A Young wrote:
> 
> >Unfortunately I don't have a serial console to output to. The best
> >I can do is the dmesg outputs for xen and the kernel with I am
> >attaching (bzipped).
> 
> I found the conring_size setting and got a more complete dmesg log.

<sigh>

I looked through it and also through your previous kernel and Xen
outputs and nothing sparked that 'Aha!' moment.

There are a couple of things we can try:
 - Compare this with the output from Xen 3.4 and see if the IOAPIC lines
   are different. Especially if these:
(XEN) IOAPIC[0]: Set PCI routing entry (2-12 -> 0x78 -> IRQ 12 Mode:0 Active:0)
(XEN) IOAPIC[0]: Set PCI routing entry (2-1 -> 0x28 -> IRQ 1 Mode:0 Active:0

are different. I think that previous to Xen 4, the pv-ops kernel could
not set the IOAPIC entries below pin 16, so you would not see them and
instead it would have these programmed:
(XEN)  01 001 01  0    0    0   0   0    1    1    28
(XEN)  0c 001 01  0    0    0   0   0    1    1    78

Which is OK, as the trigger and polarity look to be correct. 

 - Boot the Xen4, and trigger the IOAPIC debug printout. I think this is
   can be done via "xm send-keys i". Also the 'q' output would be
   usefull (it will tell us which ioports domain 0 has access to - we
   should see dom0 see 0x60 and 0x64), and irq 1, and 12.

 - The /proc/interrupts is interesting. It shows no interrupts on your
   other CPU, but the IOAPIC tells us that the irq's should be blasted
   to both CPUs. So I wonder if it got reprogrammed by the 'irqbalance'
   daemon. One way to find that out is to do the 'xm send-keys i'
   thingie to see if the destination field is the same.

 - We can also compare this to baremetal IOAPIC programming. It should
   be the _same_ as what Xen does. What we can do is provide
   'apic=debug' and that will print out the IOAPIC entries of baremetal
   kernel. The values for irq 1 and 12 ought to be same as what Xen saw
   and programmed it too.