clockevents_adjust_all_freqs() iterates over all devices in the
clockevent_devices list and adjusts frequencies as appropriate, skipping
over those that have either of CLOCK_EVT_FEAT_DUMMY and
CLOCK_EVT_FEAT_NO_ADJUST set or aren't oneshot capable.
This results in unnecessary memory accesses to these list members.
Avoid this by moving all such devices to the end of the clockevents_devices
list and making clockevents_adjust_all_freqs() return as soon as it
encounters the first of them.
For the list insertion part, introduce the ced_list_add() helper and
use it where appropriate.
Benchmark results:
The following measurements have been carried out on a Raspberry Pi 2B
(armv7, 4 cores, 900MHz). The adjustment process has been driven by
periodically injecting a certain offset via adjtimex(2) approximately
every 3.7h. A 'stress --vm 8 --vm-bytes 32M' was running. The runtime of
clockevents_adjust_all_freqs() has been measured.
- Before this patch:
Mean: 6916.90+-782.63
Quantiles:
0% 25% 50% 75% 100%
3072 6412 6927 7430 10989 (ns)
- After this patch:
Mean: 6505.18+-740.85
Quantiles:
0% 25% 50% 75% 100%
2708 6054 6523 6980 10885 (ns)
Signed-off-by: Nicolai Stange <nicsta...@gmail.com>
---
kernel/time/clockevents.c | 22 +++++++++++++++++++---
1 file changed, 19 insertions(+), 3 deletions(-)
diff --git a/kernel/time/clockevents.c b/kernel/time/clockevents.c
index 6146d2f..038fa82 100644
--- a/kernel/time/clockevents.c
+++ b/kernel/time/clockevents.c
@@ -341,6 +341,22 @@ int clockevents_program_event(struct clock_event_device
*dev, ktime_t expires,
return (rc && force) ? clockevents_program_min_delta(dev) : rc;
}
+static void ced_list_add(struct clock_event_device *dev)
+{
+ /*
+ * Insert all devices which aren't candidates for NTP
+ * frequency adjustments at the end of the list such that
+ * clockevents_adjust_all_freqs() can skip the tail once
+ * encountering the first of them.
+ */
+ if (!(dev->features & CLOCK_EVT_FEAT_ONESHOT) ||
+ (dev->features & CLOCK_EVT_FEAT_DUMMY) ||
+ (dev->features & CLOCK_EVT_FEAT_NO_ADJUST))
+ list_add_tail(&dev->list, &clockevent_devices);
+ else
+ list_add(&dev->list, &clockevent_devices);
+}
+
/*
* Called after a notify add to make devices available which were
* released from the notifier call.
@@ -353,7 +369,7 @@ static void clockevents_notify_released(void)
dev = list_entry(clockevents_released.next,
struct clock_event_device, list);
list_del(&dev->list);
- list_add(&dev->list, &clockevent_devices);
+ ced_list_add(dev);
tick_check_new_device(dev);
}
}
@@ -524,7 +540,7 @@ void clockevents_register_device(struct clock_event_device
*dev)
raw_spin_lock_irqsave(&clockevents_lock, flags);
- list_add(&dev->list, &clockevent_devices);
+ ced_list_add(dev);
tick_check_new_device(dev);
clockevents_notify_released();
@@ -638,7 +654,7 @@ void clockevents_adjust_all_freqs(u32 mult_cs_mono, u32
mult_cs_raw)
if (!(dev->features & CLOCK_EVT_FEAT_ONESHOT) ||
(dev->features & CLOCK_EVT_FEAT_DUMMY) ||
(dev->features & CLOCK_EVT_FEAT_NO_ADJUST))
- continue;
+ break;
/*
* The cached last_mult_adjusted is only valid if
--
2.10.2
