On 14/05/16 06:34, Emilio G. Cota wrote:
> +static inline void qemu_spin_lock(QemuSpin *spin)
> +{
> + while (atomic_test_and_set_acquire(&spin->value)) {
A possible optimization might be using unlikely() here, copmare:
spin.o: file format elf64-littleaarch64
Disassembly of section .text:
0000000000000000 <spin_lock__no_hint>:
0: 52800022 mov w2, #0x1 // #1
4: 885ffc01 ldaxr w1, [x0]
8: 88037c02 stxr w3, w2, [x0]
c: 35ffffc3 cbnz w3, 4 <spin_lock__no_hint+0x4>
10: 340000a1 cbz w1, 24 <spin_lock__no_hint+0x24>
14: b9400001 ldr w1, [x0]
18: 34ffff61 cbz w1, 4 <spin_lock__no_hint+0x4>
1c: d503203f yield
20: 17fffffd b 14 <spin_lock__no_hint+0x14>
24: d65f03c0 ret
0000000000000028 <spin_lock__hint>:
28: 52800022 mov w2, #0x1 // #1
2c: 885ffc01 ldaxr w1, [x0]
30: 88037c02 stxr w3, w2, [x0]
34: 35ffffc3 cbnz w3, 2c <spin_lock__hint+0x4>
38: 35000061 cbnz w1, 44 <spin_lock__hint+0x1c>
3c: d65f03c0 ret
40: d503203f yield
44: b9400001 ldr w1, [x0]
48: 35ffffc1 cbnz w1, 40 <spin_lock__hint+0x18>
4c: 17fffff8 b 2c <spin_lock__hint+0x4>
spin_lock__hint(), the one where unlikely() used, gives a bit more
CPU-pipeline-friendly fast-path.
> + while (atomic_read(&spin->value)) {
> + cpu_relax();
> + }
> + }
> +}
> +
> +static inline int qemu_spin_trylock(QemuSpin *spin)
> +{
> + if (atomic_test_and_set_acquire(&spin->value)) {
> + return -EBUSY;
> + }
> + return 0;
> +}
Here we could also benefit from unlikely(), I think.
Kind regards,
Sergey
