RT tasks do not have any timerslack, as this induces jitter. By that, the timer slack is already ignored in the nanosleep family and schedule_hrtimeout_range() (fixed in 0c52310f2600).
The hrtimer_start_range_ns function is indirectly used by glibc-2.33+ for timed waits on condition variables. These are sometimes used in RT applications for realtime queue processing. At least on the combination of kernel 5.10 and glibc-2.31, the timed wait on condition variables in rt tasks was precise (no slack), however glibc-2.33 changed the internal wait implementation, exposing the kernel bug.
This patch makes the timer slack consistent across all hrtimer programming code, by ignoring the timerslack for rt tasks also in the last remaining location in hrtimer_start_range_ns().
Similar to 0c52310f2600, this fix should be backported as well.
Cc: stable@vger.kernel.org Signed-off-by: Felix Moessbauer felix.moessbauer@siemens.com --- kernel/time/hrtimer.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c index 2b1469f61d9c..1b26e095114d 100644 --- a/kernel/time/hrtimer.c +++ b/kernel/time/hrtimer.c @@ -1274,7 +1274,7 @@ static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, * hrtimer_start_range_ns - (re)start an hrtimer * @timer: the timer to be added * @tim: expiry time - * @delta_ns: "slack" range for the timer + * @delta_ns: "slack" range for the timer for SCHED_OTHER tasks * @mode: timer mode: absolute (HRTIMER_MODE_ABS) or * relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED); * softirq based mode is considered for debug purpose only! @@ -1299,6 +1299,10 @@ void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
base = lock_hrtimer_base(timer, &flags);
+ /* rt-tasks do not have a timer slack for obvious reasons */ + if (rt_task(current)) + delta_ns = 0; + if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base)) hrtimer_reprogram(timer, true);
On Mon, Aug 05 2024 at 14:41, Felix Moessbauer wrote:
RT tasks do not have any timerslack, as this induces jitter. By that, the timer slack is already ignored in the nanosleep family and schedule_hrtimeout_range() (fixed in 0c52310f2600).
The hrtimer_start_range_ns function is indirectly used by glibc-2.33+ for timed waits on condition variables. These are sometimes used in RT applications for realtime queue processing. At least on the combination of kernel 5.10 and glibc-2.31, the timed wait on condition variables in rt tasks was precise (no slack), however glibc-2.33 changed the internal wait implementation, exposing the kernel bug.
That's hardly a bug. It's an oversight.
This patch makes the timer slack consistent across all hrtimer
"This patch" ....
programming code, by ignoring the timerslack for rt tasks also in the last remaining location in hrtimer_start_range_ns().
Similar to 0c52310f2600, this fix should be backported as well.
This is not part of the change log.
Cc: stable@vger.kernel.org Signed-off-by: Felix Moessbauer felix.moessbauer@siemens.com
kernel/time/hrtimer.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c index 2b1469f61d9c..1b26e095114d 100644 --- a/kernel/time/hrtimer.c +++ b/kernel/time/hrtimer.c @@ -1274,7 +1274,7 @@ static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
- hrtimer_start_range_ns - (re)start an hrtimer
- @timer: the timer to be added
- @tim: expiry time
- @delta_ns: "slack" range for the timer
- @delta_ns: "slack" range for the timer for SCHED_OTHER tasks
- @mode: timer mode: absolute (HRTIMER_MODE_ABS) or
relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);softirq based mode is considered for debug purpose only!@@ -1299,6 +1299,10 @@ void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, base = lock_hrtimer_base(timer, &flags);
- /* rt-tasks do not have a timer slack for obvious reasons */
- if (rt_task(current))
delta_ns = 0;
task_is_realtime()
please
Thanks,
tglx
linux-stable-mirror@lists.linaro.org
-
Felix Moessbauer -
Thomas Gleixner