MSR virtualization

MSR virtualization

[Dong, Eddie]

Avi:
	We (Yunfeng) encountered some warning from KVM in certain
situation like:
"kvm: 9612: cpu0 unhandled wrmsr: 0xc1"
	Further check find that we are doing MSR write virtualization
per a predefined whitelist and give gp fault for others. On the other
hand, Xen just silently return (no gp fault). We may not implement
policy for all MSRs, but not sure if injecting gp fault for them will
cause problem.
	Any comments?
thx,eddie

-------------------------------------------------------------------------
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: MSR virtualization

[Avi Kivity]

Dong, Eddie wrote:
> Avi:
> 	We (Yunfeng) encountered some warning from KVM in certain
> situation like:
> "kvm: 9612: cpu0 unhandled wrmsr: 0xc1"
> 	Further check find that we are doing MSR write virtualization
> per a predefined whitelist and give gp fault for others. On the other
> hand, Xen just silently return (no gp fault). We may not implement
> policy for all MSRs, but not sure if injecting gp fault for them will
> cause problem.
>   

Silently returning is IMO problematic.  If a guest depends on the 
correct behavior of some msr, and we mis-emulate it by ignoring it, then 
we get a guest failure with no message in the kernel to point us in the 
right direction.

In the case of 0xc1, this is the performance counter, likely used for 
the nmi watchdog.  If we ignore it, the guest kernel will just report a 
soft lockup and hang.  So there are only two realistic options left:

- printk() and ignore - that is a mis-emulation of writes to msrs that 
are supposed to #gp
- printk() and #gp - that is a mis-emulation of writes to msrs that must 
be handled.  But there are fewer of these than msrs that need to #gp.

So I think the current behavior is best.  Unhandled msrs are rare, we 
just need to implement them when they happen.


-------------------------------------------------------------------------
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/