Re: [PATCHv7 3/8] kvm_para: guest side for eoi avoidance

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Jun 18, 2012 at 06:01:59PM +0300, Avi Kivity wrote:
> On 06/18/2012 05:50 PM, Michael S. Tsirkin wrote:
> > 
> >> >  
> >> > +/* size alignment is implied but just to make it explicit. */
> >> > +static DEFINE_PER_CPU(unsigned long, kvm_apic_eoi) __aligned(2) =
> >> > +	KVM_PV_EOI_DISABLED;
> >> 
> >> You're actually breaking the alignment.  ulong has 8 byte alignment
> >> sometimes and you can make it cross cache boundary this way.
> > 
> > No, if you look at the definition of __aligned
> > you will see that it limits the alignment from below.
> > Compiler still applies the natural size alignment.
> > You are not the first to get confused. So I wonder: is it better
> > to add a comment or simply remove __aligned here.
> 
> Both.

Will do.

> >> >  
> >> > +	if (kvm_para_has_feature(KVM_FEATURE_PV_EOI)) {
> >> > +		__get_cpu_var(kvm_apic_eoi) = 0;
> >> > +		wrmsrl(MSR_KVM_PV_EOI_EN, __pa(&__get_cpu_var(kvm_apic_eoi)) |
> >> > +		       KVM_MSR_ENABLED);
> >> 
> >> Bad formatting.
> > 
> > I guess temporary will make it prettier.
> > 	unsigned long pa;
> > 	__get_cpu_var(kvm_apic_eoi) = 0;
> > 	pa = __pa(&__get_cpu_var(kvm_apic_eoi)) | KVM_MSR_ENABLED;
> > 	wrmsrl(MSR_KVM_PV_EOI_EN, pa);
> 
> That, or
> 
> +		wrmsrl(MSR_KVM_PV_EOI_EN,
> +                      __pa(&__get_cpu_var(kvm_apic_eoi)) | _ENABLED);
> 
> You have an argument split over two lines with no helpful indentation to
> show this.
> 
> >> 
> >> 
> >> Please check that the kexec path also disables pveoi.
> > 
> > The chunk in kvm_pv_guest_cpu_reboot does this, doesn't it?
> 
> Dunno, does it?

I thought it absolutely does but now I noticed this:

Without CONFIG_KEXEC_JUMP, it calls kernel_restart_prepare
which invokes notifiers. So fine.
But with CONFIG_KEXEC_JUMP it does suspend which is way more complex -
it stops all other cpus so we are fine but still not sure about the last
one.

Any idea?

How does it work for e.g. ASYNC_PF?

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