Re: [PATCH] x86/hyperv: Use atomic_try_cmpxchg() to micro-optimize hv_nmi_unknown()

[Uros Bizjak]

On Wed, Nov 15, 2023 at 6:19 PM Michael Kelley <mhklinux@xxxxxxxxxxx> wrote:
>
> From: Uros Bizjak <ubizjak@xxxxxxxxx> Sent: Tuesday, November 14, 2023 8:59 AM
> >
> > Use atomic_try_cmpxchg() instead of atomic_cmpxchg(*ptr, old, new) == old
> > in hv_nmi_unknown(). On x86 the CMPXCHG instruction returns success in
> > the ZF flag, so this change saves a compare after CMPXCHG. The generated
> > asm code improves from:
> >
> >   3e: 65 8b 15 00 00 00 00    mov    %gs:0x0(%rip),%edx
> >   45: b8 ff ff ff ff          mov    $0xffffffff,%eax
> >   4a: f0 0f b1 15 00 00 00    lock cmpxchg %edx,0x0(%rip)
> >   51: 00
> >   52: 83 f8 ff                cmp    $0xffffffff,%eax
> >   55: 0f 95 c0                setne  %al
> >
> > to:
> >
> >   3e: 65 8b 15 00 00 00 00    mov    %gs:0x0(%rip),%edx
> >   45: b8 ff ff ff ff          mov    $0xffffffff,%eax
> >   4a: f0 0f b1 15 00 00 00    lock cmpxchg %edx,0x0(%rip)
> >   51: 00
> >   52: 0f 95 c0                setne  %al
> >
> > No functional change intended.
> >
> > Cc: "K. Y. Srinivasan" <kys@xxxxxxxxxxxxx>
> > Cc: Haiyang Zhang <haiyangz@xxxxxxxxxxxxx>
> > Cc: Wei Liu <wei.liu@xxxxxxxxxx>
> > Cc: Dexuan Cui <decui@xxxxxxxxxxxxx>
> > Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> > Cc: Borislav Petkov <bp@xxxxxxxxx>
> > Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> > Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
> > Signed-off-by: Uros Bizjak <ubizjak@xxxxxxxxx>
> > ---
> >  arch/x86/kernel/cpu/mshyperv.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/kernel/cpu/mshyperv.c
> > b/arch/x86/kernel/cpu/mshyperv.c index e6bba12c759c..01fa06dd06b6
> > 100644
> > --- a/arch/x86/kernel/cpu/mshyperv.c
> > +++ b/arch/x86/kernel/cpu/mshyperv.c
> > @@ -262,11 +262,14 @@ static uint32_t  __init ms_hyperv_platform(void)
> > static int hv_nmi_unknown(unsigned int val, struct pt_regs *regs)  {
> >       static atomic_t nmi_cpu = ATOMIC_INIT(-1);
> > +     unsigned int old_cpu, this_cpu;
> >
> >       if (!unknown_nmi_panic)
> >               return NMI_DONE;
> >
> > -     if (atomic_cmpxchg(&nmi_cpu, -1, raw_smp_processor_id()) != -1)
> > +     old_cpu = -1;
> > +     this_cpu = raw_smp_processor_id();
> > +     if (!atomic_try_cmpxchg(&nmi_cpu, &old_cpu, this_cpu))
> >               return NMI_HANDLED;
> >
> >       return NMI_DONE;
> > --
> > 2.41.0
>
> The change looks correct to me.  But is there any motivation other
> than saving 3 bytes of generated code?  This is not a performance
> sensitive path.  And the change adds 3 lines of source code.  So
> I wonder if the change is worth the churn.

Yes, I was trying to make the function more easy to understand and
similar to nmi_panic() from kernel/panic.c. I had also the idea of
using CPU_INVALID #define instead of -1, but IMO, the above works as
well.

> In any case,
>
> Reviewed-by: Michael Kelley <mhklinux@xxxxxxxxxxx>

Thanks,
Uros.