[Qemu-devel] AMD 100% CPU bug; patch available?

[Qemu-devel] AMD 100% CPU bug; patch available?

[David Barrett]

Can anybody confirm that qemu works on an AMD processor with a Linux 
host and Linux guest (ideally Ubuntu 8.04 in both cases) without using 
100% CPU?  (qemu process on the host uses 100% CPU despite all processes 
on the guest being idle.)

Or is there a known bug that qemu with a Ubuntu guest always uses 100% 
CPU on AMD Ubuntu hosts?  If so, is there a patch or workaround for it?

I don't know this for certain, but it's the only pattern I can see: I've 
run qemu on a bunch of laptops and servers over the past 12 months -- 
often with the exact same image and exact same commands -- and some 
hosts see the guest VM use limited CPU, and others a solid 100%.  This 
seems to be the case whether or not -kernel-kqemu is used.

At first I thought it was an Ubuntu/Fedora Core issue, as I was running 
Ubuntu on laptops and FC4 on servers (and I only saw this problem on 
servers).  But I just got a new Ubuntu server and the problem remains.

Now I realize there's another commonality: my servers have been 
AMD-based, whereas laptops were Intel based.  With this theory I looked 
through the user forum and saw a strong correlation between reports of 
100% CPU and AMD processors.

Is this old news or a debunked theory?

Is there some tool like CPUIdle I should be running?  Any suggestions 
for how to start debugging the issue?

Thanks!

-david

Re: [Qemu-devel] AMD 100% CPU bug; patch available?

[Rick Vernam]

I'm not so sure there is such a bug - I've been running qemu on AMD 3700 
laptop for years w/o that, or much of any other issue (except once upon a 
time an ACPI issue w/ windows guest causing host cpu to peg 100%).
I run gentoo host + various windows/linux guests.

points of interest may (or may not) be details such as host kernel 
configuration?  qemu version?  qemu options used?  built yourself or 
pre-packaged?
I don't claim to be able to help you solve the problem, but there's not much 
here to work with...

On Tuesday 12 August 2008 11:36:08 am David Barrett wrote:
> Can anybody confirm that qemu works on an AMD processor with a Linux
> host and Linux guest (ideally Ubuntu 8.04 in both cases) without using
> 100% CPU?  (qemu process on the host uses 100% CPU despite all processes
> on the guest being idle.)
>
> Or is there a known bug that qemu with a Ubuntu guest always uses 100%
> CPU on AMD Ubuntu hosts?  If so, is there a patch or workaround for it?
>
> I don't know this for certain, but it's the only pattern I can see: I've
> run qemu on a bunch of laptops and servers over the past 12 months --
> often with the exact same image and exact same commands -- and some
> hosts see the guest VM use limited CPU, and others a solid 100%.  This
> seems to be the case whether or not -kernel-kqemu is used.
>
> At first I thought it was an Ubuntu/Fedora Core issue, as I was running
> Ubuntu on laptops and FC4 on servers (and I only saw this problem on
> servers).  But I just got a new Ubuntu server and the problem remains.
>
> Now I realize there's another commonality: my servers have been
> AMD-based, whereas laptops were Intel based.  With this theory I looked
> through the user forum and saw a strong correlation between reports of
> 100% CPU and AMD processors.
>
> Is this old news or a debunked theory?
>
> Is there some tool like CPUIdle I should be running?  Any suggestions
> for how to start debugging the issue?
>
> Thanks!
>
> -david

Re: [Qemu-devel] AMD 100% CPU bug; patch available?

[Anthony Liguori]

David Barrett wrote:
> Can anybody confirm that qemu works on an AMD processor with a Linux 
> host and Linux guest (ideally Ubuntu 8.04 in both cases) without using 
> 100% CPU?  (qemu process on the host uses 100% CPU despite all 
> processes on the guest being idle.)
>
> Or is there a known bug that qemu with a Ubuntu guest always uses 100% 
> CPU on AMD Ubuntu hosts?  If so, is there a patch or workaround for it?
>
> I don't know this for certain, but it's the only pattern I can see: 
> I've run qemu on a bunch of laptops and servers over the past 12 
> months -- often with the exact same image and exact same commands -- 
> and some hosts see the guest VM use limited CPU, and others a solid 
> 100%.  This seems to be the case whether or not -kernel-kqemu is used.

Do you still see this if you use -no-kqemu?

Regards,

Anthony Liguori

Re: [Qemu-devel] AMD 100% CPU bug; patch available?

[David Barrett]

Aha, figured it out.  The tip was in dmesg on the guest:

> DMI not present or invalid.
> ACPI: RSDP signature @ 0xC00FB010 checksum 0
> ACPI: RSDP 000FB010, 0014 (r0 QEMU  )
> ACPI: RSDT 07FF0000, 002C (r1 QEMU   QEMURSDT        1 QEMU      
>   1)
> ACPI: FACP 07FF002C, 0074 (r1 QEMU   QEMUFACP        1 QEMU      
>   1)
> ACPI: DSDT 07FF0100, 0832 (r1   BXPC   BXDSDT        1 INTL 20061
> 109)
> ACPI: FACS 07FF00C0, 0040
> ACPI: APIC 07FF0938, 0040 (r1 QEMU   QEMUAPIC        1 QEMU      
> ACPI: no DMI BIOS year, acpi=force is required to enable ACPI
> ACPI: Disabling ACPI support

Adding acpi=force to the kernel boot line fixed it up.

(Though strangely, even after adding acpi=force to the kernel boot line, 
it still dumps this exact same message -- odd.  At least the problem 
went away.)

Thanks for the help!

-david

Rick Vernam wrote:
> I'm not so sure there is such a bug - I've been running qemu on AMD 3700 
> laptop for years w/o that, or much of any other issue (except once upon a 
> time an ACPI issue w/ windows guest causing host cpu to peg 100%).
> I run gentoo host + various windows/linux guests.
> 
> points of interest may (or may not) be details such as host kernel 
> configuration?  qemu version?  qemu options used?  built yourself or 
> pre-packaged?
> I don't claim to be able to help you solve the problem, but there's not much 
> here to work with...
> 
> On Tuesday 12 August 2008 11:36:08 am David Barrett wrote:
>> Can anybody confirm that qemu works on an AMD processor with a Linux
>> host and Linux guest (ideally Ubuntu 8.04 in both cases) without using
>> 100% CPU?  (qemu process on the host uses 100% CPU despite all processes
>> on the guest being idle.)
>>
>> Or is there a known bug that qemu with a Ubuntu guest always uses 100%
>> CPU on AMD Ubuntu hosts?  If so, is there a patch or workaround for it?
>>
>> I don't know this for certain, but it's the only pattern I can see: I've
>> run qemu on a bunch of laptops and servers over the past 12 months --
>> often with the exact same image and exact same commands -- and some
>> hosts see the guest VM use limited CPU, and others a solid 100%.  This
>> seems to be the case whether or not -kernel-kqemu is used.
>>
>> At first I thought it was an Ubuntu/Fedora Core issue, as I was running
>> Ubuntu on laptops and FC4 on servers (and I only saw this problem on
>> servers).  But I just got a new Ubuntu server and the problem remains.
>>
>> Now I realize there's another commonality: my servers have been
>> AMD-based, whereas laptops were Intel based.  With this theory I looked
>> through the user forum and saw a strong correlation between reports of
>> 100% CPU and AMD processors.
>>
>> Is this old news or a debunked theory?
>>
>> Is there some tool like CPUIdle I should be running?  Any suggestions
>> for how to start debugging the issue?
>>
>> Thanks!
>>
>> -david
> 
> 
> 
>