Skip to contentSkip to conversation

timer: Use static_branch_likely() for timers_nohz_active

PATCH In review Jisheng Zhang 2 participants 3 messages Jun 24, 2021 – Aug 20, 2021 High-resolution timers, timer wheel, clockevents

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 timer: Use static_branch_likely() for timers_nohz_active

The static key timers_nohz_active is likely to be true, so use static_branch_likely() to reflect this fact.

Per include/linux/jump_label.h:

* type\branch| likely (1) | unlikely (0)
* -----------+-----------------------+------------------

 *            |                       |
 *  true (1)  |	   ...		      |	   ...
 *            |    NOP		      |	   JMP L
 *            |    <br-stmts>	      |	1: ...
 *            |	L: ...		      |
 *            |			      |
 *            |			      |	L: <br-stmts>
 *            |			      |	   jmp 1b
 *            |                       |

* -----------+-----------------------+------------------

This could improve the finally generated code a bit for the most likely scenario, I.E save two "jmp" instructions.

---
Since v1:
- update the commit msg to avoid confusing and add some descriptions about where's the improvement from.

 kernel/time/timer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/time/timer.c +1 -1
index d111adf4a0cb..40d4fe7fa207 100644
@@ -265,7 +265,7 @@ int timer_migration_handler(struct ctl_table *table, int write,
 
 static inline bool is_timers_nohz_active(void)
 {
-	return static_branch_unlikely(&timers_nohz_active);
+	return static_branch_likely(&timers_nohz_active);
 }
 #else
 static inline bool is_timers_nohz_active(void) { return false; }
Signature · 1 line
2.31.0
#1 lore raw

Jisheng,

Quoted 2 lines · Jisheng Zhang

The static key timers_nohz_active is likely to be true, so use static_branch_likely() to reflect this fact.

you still lack any justification for your statement that NOHZ active is likely.

It might be likely for your system, but is this true in general?

Thanks,

tglx

#2 lore raw
Quoted 3 lines · On Tue, 10 Aug 2021 18:00:16 +0200 Thomas Gleixner wrote:

Jisheng,

Hi Thomas,

Quoted 8 lines · Jisheng Zhang
Quoted 3 lines

The static key timers_nohz_active is likely to be true, so use static_branch_likely() to reflect this fact.

you still lack any justification for your statement that NOHZ active is likely.

This patch only affects NO_HZ_COMMON=y code path which is enabled if either NO_HZ_IDLE or NO_HZ_FULL = y.
Per my understanding, when the system has oneshot tick device, NOHZ active is likely true. After going through drivers/clocksource/*, I found there are about 86 drivers, 69(occupies 80%) of which support CLOCK_EVT_FEAT_ONESHOT. Then I browsed apic, hpet and xen timer source code in x86 directory, they all support oneshot feature.

I'm not sure whether I misunderstood the code or not.

Thanks

Quoted 3 lines

It might be likely for your system, but is this true in general?