On Tue, Sep 3, 2013 at 2:34 PM, Linus Torvalds
<[email protected]> wrote:
>
> I'll try to hack that up too, but it's looking like it really is just
> the "lock xadd", not the memory dependency chain..
Yeah, no difference: Better code generation with my quick hack for a
percpu spinlock:
│ ffffffff81078e70 <lg_local_lock>:
0.59 │ push %rbp
0.25 │ mov %rsp,%rbp
0.07 │ mov $0x100,%eax
97.55 │ lock xadd %ax,%gs:(%rdi)
0.01 │ movzbl %ah,%edx
│ cmp %al,%dl
0.68 │ ↓ je 29
│ nop
│20: pause
│ mov %gs:(%rdi),%al
│ cmp %dl,%al
│ ↑ jne 20
│29: pop %rbp
0.84 │ ← retq
but the actual cost is pretty much the same:
6.81% lg_local_lock
so it doesn't seem to be some odd weakness of the microarchitecture.
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
