Re: [PATCH 5/8] KVM: use jump label to optimize checking for HW enabled APIC in APIC_BASE MSR.

[Gleb Natapov <gleb@redhat.com>]

On Sun, Aug 05, 2012 at 05:48:42PM +0300, Avi Kivity wrote:
> On 08/05/2012 05:42 PM, Gleb Natapov wrote:
> > On Sun, Aug 05, 2012 at 05:35:21PM +0300, Avi Kivity wrote:
> >> On 08/05/2012 03:58 PM, Gleb Natapov wrote:
> >> > Usually all APICs are HW enabled so the check can be optimized out.
> >> > 
> >> > Signed-off-by: Gleb Natapov <gleb@redhat.com>
> >> > ---
> >> >  arch/x86/kvm/lapic.c |   29 ++++++++++++++++++++++++++++-
> >> >  arch/x86/kvm/lapic.h |    1 +
> >> >  arch/x86/kvm/x86.c   |    1 +
> >> >  3 files changed, 30 insertions(+), 1 deletion(-)
> >> > 
> >> > diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> >> > index c3f14fe..1aa5528 100644
> >> > --- a/arch/x86/kvm/lapic.c
> >> > +++ b/arch/x86/kvm/lapic.c
> >> > @@ -34,6 +34,7 @@
> >> >  #include <asm/current.h>
> >> >  #include <asm/apicdef.h>
> >> >  #include <linux/atomic.h>
> >> > +#include <linux/jump_label.h>
> >> >  #include "kvm_cache_regs.h"
> >> >  #include "irq.h"
> >> >  #include "trace.h"
> >> > @@ -117,9 +118,13 @@ static inline int __apic_test_and_clear_vector(int vec, void *bitmap)
> >> >  	return __test_and_clear_bit(VEC_POS(vec), (bitmap) + REG_POS(vec));
> >> >  }
> >> >  
> >> > +struct static_key_deferred apic_hw_disabled __read_mostly;
> >> 
> >> On top of file please.  Add all_ to the name to make it clear we're
> >> talking about all apics.
> >> 
> > This is count of disabled apics really. So I think all_apic_hw_disabled
> > is misleading.
> > 
> >> > +
> >> >  static inline int apic_hw_enabled(struct kvm_lapic *apic)
> >> >  {
> >> > -	return (apic)->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
> >> > +	if (static_key_false(&apic_hw_disabled.key))
> >> > +		return apic->vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE;
> >> 
> >> Hm, for the test to be readable, it needs to be
> >> 
> >>    if (static_key_false(&all_apics_hw_enabled))
> >> 
> > Exactly. all_ makes it so because apic_hw_disabled is a counter that
> > counts disabled apics. So may be call it global_hw_disabled_apic_counter?
> > 
> 
> The problem is how static_key_false() is defined.  It returns true if
> the count > 0, opposite from what I'd expect.  So anything with counter
> semantics will be confusing.  I guess we need to pick a neutral name
> (apic_disabled_key or apic_disabled_slowpath or such) to force the
> reader to look at the definitions.
> 
There was a long thread about readability of static_key_true/false and
this is the result of it :( I read it this way: drop static_key_(true|false)
and read the if() to understand its meaning. Now look at static_key_(true|false)
to see what fast path will be in asm code.

--
			Gleb.