Bug: cpu0 unhandled rdmsr: 0x1900 and inject_general_protection

Bug: cpu0 unhandled rdmsr: 0x1900 and inject_general_protection

[TJ]

Trying to run the latest OLPC XO guest image with kvm-intel on 2.6.22
x86_64 (Ubuntu Gutsy) on a T7200 there are critical problems which don't
occur if kvm-intel is not loaded/used.

The same issue affects kvm-48, kvm-50 and kvm-51.

I don't see these problems with Gutsy or Windows guests. The issues are:

1. Start-up of the image into GRUB is _slow_. The cursor can be seen
scanning as the GRUB menu is drawn. It is *much* faster when kvm-intel
isn't loaded/used.

2. After choosing a boot option the guest freezes and the host kern.log
reports:

kernel: [88380.014893] kvm: 12666: cpu0 unhandled rdmsr: 0x1900
kernel: [88380.014900] inject_general_protection: rip 0xc07211ab

3. Guest locks up at this point with host CPU being driven 100% and requires killing.


I'm guessing they may be related to the OLPC kernel expecting an AMD Geode CPU?

./include/asm-i386/msr.h
#define MSR_GEODE_BUSCONT_CONF0 0x1900

The OLPC disk image and configuration instructions are here:

http://wiki.laptop.org/go/Emulating_the_XO/Quick_Start

The configuration used here is:

vdeq kvm -m 256 -hda olpc-redhat-stream-development-devel_ext3.ovl -soundhw es1370 -net nic,model=rtl8139,macaddr=56:44:45:30:32:31,vlan=0 -net vde,vlan=0

Is this a known issue - I couldn't find anything similar in searches of the mailing-list or bug-tracker ?

Thanks.

TJ.


-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/

Re: Bug: cpu0 unhandled rdmsr: 0x1900 and inject_general_protection

[Avi Kivity]

[-- Attachment #1: Type: text/plain, Size: 1295 bytes --]

TJ wrote:
> Trying to run the latest OLPC XO guest image with kvm-intel on 2.6.22
> x86_64 (Ubuntu Gutsy) on a T7200 there are critical problems which don't
> occur if kvm-intel is not loaded/used.
>
> The same issue affects kvm-48, kvm-50 and kvm-51.
>
> I don't see these problems with Gutsy or Windows guests. The issues are:
>
> 1. Start-up of the image into GRUB is _slow_. The cursor can be seen
> scanning as the GRUB menu is drawn. It is *much* faster when kvm-intel
> isn't loaded/used.
>
> 2. After choosing a boot option the guest freezes and the host kern.log
> reports:
>
> kernel: [88380.014893] kvm: 12666: cpu0 unhandled rdmsr: 0x1900
> kernel: [88380.014900] inject_general_protection: rip 0xc07211ab
>
> 3. Guest locks up at this point with host CPU being driven 100% and requires killing.
>
>
> I'm guessing they may be related to the OLPC kernel expecting an AMD Geode CPU?
>
> ./include/asm-i386/msr.h
> #define MSR_GEODE_BUSCONT_CONF0 0x1900
>
>   

Very likely.  Try the attached patch.  The guest may hang later with 
another bad msr; you can simply add more registers as necessary.

I think we can add an API to do this from userspace, to avoid the need 
to patch the kernel for hundreds of no-op msrs.


-- 
error compiling committee.c: too many arguments to function


[-- Attachment #2: geode.patch --]
[-- Type: text/x-patch, Size: 377 bytes --]

diff --git a/kernel/x86.c b/kernel/x86.c
index e905d46..71307cb 100644
--- a/kernel/x86.c
+++ b/kernel/x86.c
@@ -460,6 +460,7 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
 	case MSR_IA32_UCODE_REV:
 	case MSR_IA32_PERF_STATUS:
 	case MSR_IA32_EBL_CR_POWERON:
+	case MSR_GEODE_BUSCONT_CONF0:
 		/* MTRR registers */
 	case 0xfe:
 	case 0x200 ... 0x2ff:

[-- Attachment #3: Type: text/plain, Size: 314 bytes --]

-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/

[-- Attachment #4: Type: text/plain, Size: 186 bytes --]

_______________________________________________
kvm-devel mailing list
kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
https://lists.sourceforge.net/lists/listinfo/kvm-devel

Re: Bug: cpu0 unhandled rdmsr: 0x1900 and inject_general_protection

[TJ]

Avi, thanks. The patch fixes the issue for the XO image and there are no
more MSR read errors.

I'm still not sure why when starting the image with kvm enabled the
drawing of the GRUB screen is so painfully slow, but without kvm it is
fast.

Any ideas?

TJ.



-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/

Re: Bug: cpu0 unhandled rdmsr: 0x1900 and inject_general_protection

[Avi Kivity]

TJ wrote:
> Avi, thanks. The patch fixes the issue for the XO image and there are no
> more MSR read errors.
>
> I'm still not sure why when starting the image with kvm enabled the
> drawing of the GRUB screen is so painfully slow, but without kvm it is
> fast.
>
>   

Grub probably uses one of the 4-bit vga modes which cannot be direct 
mapped.  This means that every pixel access is a 5-10 microsecond vmexit.

Xen uses some form of mmio buffering to reduce this costs (mmio exits 
under Xen are even more expensive).  Something like this can be added to 
kvm to speed up vga writes.

-- 
error compiling committee.c: too many arguments to function


-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/