[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [Block issue] VMX status report 13161:a9a43705f26b

That's different from the failures that I'm seeing right now, I think
(although all are variations on 'hanging' during early boot).

I checked in r13159 because without it I was failing
io_apic.c:check_timer(). My guess was that Linux had decided to use HPET to
replace the PIT, and HPET0 always interrupts on GSI 2 in legacy routing
mode. But we were routing PIT IRQ (ISA IRQ 0) to GSI 0 and so it looked to
Linux as though no timer interrupts were happening (it was 'listening' on
GSI 0 rather than GSI 2). This was the root cause of the 'bad delivery mode
7' messages that have been reported -- Linux was eventually programming an
ExtINT delivery mode into the IOAPIC as part of unlock_ExtINT_logic().

It may be that r13159 has introduced some new bug as well as fixing an old
bug. Perhaps I missed an interrupt override (ISA0 -> GSI2) somewhere.

Ah.... Actually various code thinks it 'knows' about IRQ0 and that code
won't know about the reroute to GSI2. For example, we won't be pinning PIT
to CPU0 and various hooks in vpt.c won't be triggering. That certainly needs
fixing... Hopefully that is the last bug. ;-)

 -- Keir

On 28/12/06 9:40 am, "Zhai, Edwin" <edwin.zhai@xxxxxxxxx> wrote:

> keir,
> r13159 caused hvm smp linux guest hang after "ENABLING IO-APIC IRQs", while UP
> is okay.
> seems guest are waiting timer intr.
> any suggestions?
> On Thu, Dec 28, 2006 at 09:29:10AM +0000, Keir Fraser wrote:
>> I've seen all manner of problems booting Linux guests since changesets 13115
>> (add HPET support), 13097 and 13096 (various changes to HVM platform timer
>> support). With luck there have been just one or two issues introduced that
>> are causing all these problems, but we'll have to throw out those patches if
>> the situation doesn't improve in the next week.
>> Xiaowei: I'm pretty sure I can eyeball bugs in the new vpt.c code. For
>> example, how can it be right that last_plt_gtime is incremented by period in
>> some places and period_cycles in others? The two quantities have different
>> units!
>>  -- Keir
>> On 28/12/06 2:50 am, "Yu, Ping Y" <ping.y.yu@xxxxxxxxx> wrote:
>>> The block issue is opened for bug 848, and could anybody take a look at this
>>> issue?
>>> http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=848
>>> Thanks
>>> Ping

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.