Le Mon, Jan 19, 2026 at 02:02:22PM +0100, Peter Zijlstra a écrit :

On Fri, Jan 16, 2026 at 03:51:56PM +0100, Frederic Weisbecker wrote:

quoted

diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 8ddf74e705d3..f1d07a0276a5 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c

@@ -780,7 +780,7 @@ static u64 get_cpu_sleep_time_us(struct tick_sched *ts, ktime_t *sleeptime,
 	ktime_t now, idle;
 	unsigned int seq;
 
-	if (!tick_nohz_active)
+	if (!tick_nohz_active || vtime_generic_enabled_cpu(cpu))
 		return -1;
 
 	now = ktime_get();

Is this not broken? IIUC this means that you can no longer use
get_cpu_{idle,iowait}_time_us() the moment you have context tracking
enabled.

It is supported again in patch 13/15. And it's not exactly breaking
bisection in the meantime because the sole user is cpufreq and cpufreq
shouldn't be relevant with nohz_full.

Ok a few subsystem rely on the resulting cpufreq API get_cpu_idle_time():

- the legacy drivers/macintosh/rack-meter.c
- drivers/scsi/lpfc/lpfc_init.c

But cpufreq provides a low-resolution version in the worst case for nohz_full
(again until 13/15).

Hmm, but you're right this is confusing. I think I should be able to fix that
in this patch.

Thanks.

-- 
Frederic Weisbecker
SUSE Labs