[PATCH] tick/nohz: Remove redundant IRQ save/restore

[Usama Arif <usama.arif@linux.dev>]

tick_nohz_update_jiffies() is only reached through tick_irq_enter() from
hard interrupt entry, where local interrupts are already disabled.

tick_do_update_jiffies64() requires interrupts to be disabled. The
local_irq_save()/local_irq_restore() pair around it therefore provides no
additional interrupt exclusion.

Remove it and assert the caller invariant instead.

On x86-64, this avoids the redundant RFLAGS save, CLI, saved-IF test and
conditional branch on the tick-stopped path.

On a 252-CPU host running a Thrift RPC service, tick_irq_enter() ran
165,116 times/s out of 212-244k interrupts/s. Only calls with
TS_FLAG_STOPPED reach the optimized path, so this is an upper bound on
its execution rate.

Signed-off-by: Usama Arif <usama.arif@linux.dev>
---
 kernel/time/tick-sched.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 6c3fea3867139..c8f2c4a503b08 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -738,14 +738,11 @@ bool tick_nohz_tick_stopped_cpu(int cpu)
  */
 static void tick_nohz_update_jiffies(ktime_t now)
 {
-	unsigned long flags;
+	/* Reached only from irq_enter_rcu(), i.e. hard interrupt entry. */
+	lockdep_assert_irqs_disabled();
 
 	__this_cpu_write(tick_cpu_sched.idle_waketime, now);
-
-	local_irq_save(flags);
 	tick_do_update_jiffies64(now);
-	local_irq_restore(flags);
-
 	touch_softlockup_watchdog_sched();
 }
 
-- 
2.53.0-Meta