is vmx_vcpu_run deprecated in kvm-77?

is vmx_vcpu_run deprecated in kvm-77?

[Steven Stovall]

I am porting an app from kvm-33 to kvm-77 (the app is also known to work on kvm-60 i believe). But so far I havent seen a KVM_EXIT_IO, only KVM_EXIT_INTRs. This means the app never changes state. Also, I never see vmx_vcpu_run called -- does x86/x86.c:__vcpu_run deprecate it?

steven

Re: is vmx_vcpu_run deprecated in kvm-77?

[Avi Kivity]

Steven Stovall wrote:
> I am porting an app from kvm-33 to kvm-77 (the app is also known to work on kvm-60 i believe). But so far I havent seen a KVM_EXIT_IO, only KVM_EXIT_INTRs. This means the app never changes state. Also, I never see vmx_vcpu_run called -- does x86/x86.c:__vcpu_run deprecate it?
>   

vmx_vcpu_run() is still central to running guests.  What do you mean 
you're not seeing KVM_EXIT_IO?  are you using kvm-userspace or your own 
userspace?

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

Re: is vmx_vcpu_run deprecated in kvm-77?

[Steven Stovall]

I don't mean to be obtuse, but are you saying kvm_main_loop_wait() *should* dequeue the pending signal or *does* dequeue the pending signal, just not for "host userspace"?

Steven

----- Original Message -----
From: "Avi Kivity" <avi@redhat.com>
To: "Steven Stovall" <sstovall@neuraliq.com>
Sent: Thursday, January 15, 2009 1:46:30 PM GMT -08:00 US/Canada Pacific
Subject: Re: is vmx_vcpu_run deprecated in kvm-77?

Please keep the conversation on the kvm list.

Steven Stovall wrote:
> Avi,
>
> Not sure what you mean by "kvm-userspace", 

The code in kvm-userspace.git, namely libkvm and qemu.

> but the app uses qemu and libkvm ioctls to interface with kvm.

So this is a host-side app, right, not running in the guest?  Please 
describe it.

>  The problem seems to be that for some reason the call to signal_pending() from __vcpu_run is always true and never lets vmx_vcpu_run() be called (i never see the msrs set either). The only exit_reasons seen are KVM_EXIT_INTR. I've been looking at this hard for 2 weeks now, and its driving me crazy. Not to mention my boss.
>   

I don't care much about your boss, but you should have posted earlier.  
What happens is that a signal is pending, and host userspace doesn't 
dequeue it, so it remains pending.  This is handled by kvm_main_loop_wait().

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.