Re: [PATCH] locking/atomic/x86: Introduce try_cmpxchg64

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, May 13, 2022 at 11:10 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Tue, May 10, 2022 at 05:42:17PM +0200, Uros Bizjak wrote:
>
> For the Changelog I would focus on the 64bit improvement and leave 32bit
> as a side-note.

Thanks, I will rephrase the ChangeLog.

>
> > ---
> >  arch/x86/include/asm/cmpxchg_32.h          | 43 ++++++++++++++++++++++
> >  arch/x86/include/asm/cmpxchg_64.h          |  6 +++
> >  include/linux/atomic/atomic-instrumented.h | 40 +++++++++++++++++++-
> >  scripts/atomic/gen-atomic-instrumented.sh  |  2 +-
> >  4 files changed, 89 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/cmpxchg_32.h b/arch/x86/include/asm/cmpxchg_32.h
> > index 0a7fe0321613..e874ff7f7529 100644
> > --- a/arch/x86/include/asm/cmpxchg_32.h
> > +++ b/arch/x86/include/asm/cmpxchg_32.h
> > @@ -42,6 +42,9 @@ static inline void set_64bit(volatile u64 *ptr, u64 value)
> >  #define arch_cmpxchg64_local(ptr, o, n)                                      \
> >       ((__typeof__(*(ptr)))__cmpxchg64_local((ptr), (unsigned long long)(o), \
> >                                              (unsigned long long)(n)))
> > +#define arch_try_cmpxchg64(ptr, po, n)                                       \
> > +     ((__typeof__(*(ptr)))__try_cmpxchg64((ptr), (unsigned long long *)(po), \
> > +                                          (unsigned long long)(n)))
> >  #endif
> >
> >  static inline u64 __cmpxchg64(volatile u64 *ptr, u64 old, u64 new)
> > @@ -70,6 +73,25 @@ static inline u64 __cmpxchg64_local(volatile u64 *ptr, u64 old, u64 new)
> >       return prev;
> >  }
> >
> > +static inline bool __try_cmpxchg64(volatile u64 *ptr, u64 *pold, u64 new)
> > +{
> > +     bool success;
> > +     u64 prev;
> > +     asm volatile(LOCK_PREFIX "cmpxchg8b %2"
> > +                  CC_SET(z)
> > +                  : CC_OUT(z) (success),
> > +                    "=A" (prev),
> > +                    "+m" (*ptr)
> > +                  : "b" ((u32)new),
> > +                    "c" ((u32)(new >> 32)),
> > +                    "1" (*pold)
> > +                  : "memory");
> > +
> > +     if (unlikely(!success))
> > +             *pold = prev;
>
> I would prefer this be more like the existing try_cmpxchg code,
> perhaps:
>
>         u64 old = *pold;
>
>         asm volatile (LOCK_PREFIX "cmpxchg8b %[ptr]"
>                       CC_SET(z)
>                       : CC_OUT(z) (success),
>                         [ptr] "+m" (*ptr)
>                         "+A" (old)
>                       : "b" ((u32)new)
>                         "c" ((u32)(new >> 32))
>                       : "memory");
>
>         if (unlikely(!success))
>                 *pold = old;
>
> The existing 32bit cmpxchg code is a 'bit' crusty.

I was trying to follow the existing __cmpxchg64 as much as possible,
with the intention of a follow-up patch that would modernize
everything in cmpxchg_32.h. I can surely go the other way and submit
modernized new code.

> > +     return success;
> > +}
> > +
> >  #ifndef CONFIG_X86_CMPXCHG64
> >  /*
> >   * Building a kernel capable running on 80386 and 80486. It may be necessary
> > @@ -108,6 +130,27 @@ static inline u64 __cmpxchg64_local(volatile u64 *ptr, u64 old, u64 new)
> >                      : "memory");                             \
> >       __ret; })
> >
> > +#define arch_try_cmpxchg64(ptr, po, n)                               \
> > +({                                                           \
> > +     bool success;                                           \
> > +     __typeof__(*(ptr)) __prev;                              \
> > +     __typeof__(ptr) _old = (__typeof__(ptr))(po);           \
> > +     __typeof__(*(ptr)) __old = *_old;                       \
> > +     __typeof__(*(ptr)) __new = (n);                         \
> > +     alternative_io(LOCK_PREFIX_HERE                         \
> > +                     "call cmpxchg8b_emu",                   \
> > +                     "lock; cmpxchg8b (%%esi)" ,             \
> > +                    X86_FEATURE_CX8,                         \
> > +                    "=A" (__prev),                           \
> > +                    "S" ((ptr)), "0" (__old),                \
> > +                    "b" ((unsigned int)__new),               \
> > +                    "c" ((unsigned int)(__new>>32))          \
> > +                    : "memory");                             \
> > +     success = (__prev == __old);                            \
> > +     if (unlikely(!success))                                 \
> > +             *_old = __prev;                                 \
> > +     likely(success);                                        \
> > +})
>
> Wouldn't this be better written like the normal fallback wrapper?
>
> static __always_inline bool
> arch_try_cmpxchg64(u64 *v, u64 *old, u64 new)
> {
>         u64 r, o = *old;
>         r = arch_cmpxchg64(v, o, new);
>         if (unlikely(r != o))
>                 *old = r;
>         return likely(r == o);
> }
>
> Less magical, same exact code.

Also, I tried to follow up the existing #defines. Will improve the
code according to your suggestion here.

Thanks,
Uros.



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux