Re: [PATCH v2 8/9] atomic,x86: Alternative atomic_*_overflow() scheme

[Linus Torvalds]

On Fri, Dec 10, 2021 at 8:27 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> Shift the overflow range from [0,INT_MIN] to [-1,INT_MIN], this allows
> optimizing atomic_inc_overflow() to use "jle" to detect increment
> from free-or-negative (with -1 being the new free and it's increment
> being 0 which sets ZF).

Thanks.

However, I think you can simplify this further:

> This then gives the following primitives:
>
> [-1, INT_MIN]                                   [0, INT_MIN]
>
> inc()                                           inc()
>         lock inc %[var]                                 mov       $-1, %[reg]
>         jle     error-free-or-negative                  lock xadd %[reg], %[var]
>                                                         test      %[reg], %[reg]
>                                                         jle       error-zero-or-negative
>
> dec()                                           dec()
>         lock sub $1, %[var]                             lock dec %[var]
>         jc      error-to-free                           jle     error-zero-or-negative
>         jl      error-from-negative
>
> dec_and_test()                                  dec_and_test()
>         lock sub $1, %[var]                             lock dec %[var]
>         jc      do-free                                 jl      error-from-negative
>         jl      error-from-negative                     je      do-free

That "dec()" case could be just

        lock dec %[var]
        js error

because an underflow is an underflow - it doesn't matter if it's a "it
went to free" or "it became some other negative number".

That said - it may not matter - I'm not sure a plain "dec" is even a
valid operation on a ref in the first place. How could you ever
validly decrement a ref without checking for it being the last entry?

So I'm not sure "atomic_dec_overflow()" is even worth having as a
primitive, because I can't see any valid use for it. Is it for some
legacy case?

                  Linus