diff mbox

[v3,2/4] sched_clock: Optimize cache line usage

Message ID 1422644602-11953-3-git-send-email-daniel.thompson@linaro.org
State New
Headers show

Commit Message

Daniel Thompson Jan. 30, 2015, 7:03 p.m. UTC
Currently sched_clock(), a very hot code path, is not optimized to
minimise its cache profile. In particular:

  1. cd is not ____cacheline_aligned,

  2. struct clock_data does not distinguish between hotpath and
     coldpath data, reducing locality of reference in the hotpath,

  3. Some hotpath data is missing from struct clock_data and is marked
     __read_mostly (which more or less guarantees it will not share a
     cache line with cd).

This patch corrects these problems by extracting all hotpath data
into a separate structure and using ____cacheline_aligned to ensure
the hotpath uses a single (64 byte) cache line.

Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
---
 kernel/time/sched_clock.c | 98 ++++++++++++++++++++++++++++++-----------------
 1 file changed, 62 insertions(+), 36 deletions(-)

Comments

Daniel Thompson Feb. 5, 2015, 10:21 a.m. UTC | #1
On 05/02/15 01:14, Stephen Boyd wrote:
> On 01/30, Daniel Thompson wrote:
>> diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c
>> index 3d21a8719444..cb69a47dfee4 100644
>> --- a/kernel/time/sched_clock.c
>> +++ b/kernel/time/sched_clock.c
>> @@ -18,28 +18,44 @@
>>  #include <linux/seqlock.h>
>>  #include <linux/bitops.h>
>>  
>> -struct clock_data {
>> -	ktime_t wrap_kt;
>> +/**
>> + * struct clock_read_data - data required to read from sched_clock
>> + *
> 
> Nitpick: Won't kernel-doc complain that members aren't
> documented?

It does indeed. I'll add descriptions here...


>> + * Care must be taken when updating this structure; it is read by
>> + * some very hot code paths. It occupies <=48 bytes and, when combined
>> + * with the seqcount used to synchronize access, comfortably fits into
>> + * a 64 byte cache line.
>> + */
>> +struct clock_read_data {
>>  	u64 epoch_ns;
>>  	u64 epoch_cyc;
>> -	seqcount_t seq;
>> -	unsigned long rate;
>> +	u64 sched_clock_mask;
>> +	u64 (*read_sched_clock)(void);
>>  	u32 mult;
>>  	u32 shift;
>>  	bool suspended;
>>  };
>>  
>> +/**
>> + * struct clock_data - all data needed for sched_clock (including
>> + *                     registration of a new clock source)
>> + *
> 
> Same comment.

... and here.


>> + * The ordering of this structure has been chosen to optimize cache
>> + * performance. In particular seq and read_data (combined) should fit
>> + * into a single 64 byte cache line.
>> + */
>> +struct clock_data {
>> +	seqcount_t seq;
>> +	struct clock_read_data read_data;
>> +	ktime_t wrap_kt;
>> +	unsigned long rate;
>> +};
>> @@ -60,15 +79,16 @@ unsigned long long notrace sched_clock(void)
>>  {
>>  	u64 cyc, res;
>>  	unsigned long seq;
>> +	struct clock_read_data *rd = &cd.read_data;
>>  
>>  	do {
>>  		seq = raw_read_seqcount_begin(&cd.seq);
>>  
>> -		res = cd.epoch_ns;
>> -		if (!cd.suspended) {
>> -			cyc = read_sched_clock();
>> -			cyc = (cyc - cd.epoch_cyc) & sched_clock_mask;
>> -			res += cyc_to_ns(cyc, cd.mult, cd.shift);
>> +		res = rd->epoch_ns;
>> +		if (!rd->suspended) {
> 
> Should this have likely() treatment? It would be really nice if
> we could use static branches here to avoid any branch penalty at
> all. I guess that would need some sort of special cased
> stop_machine() though. Or I wonder if we could replace
> rd->read_sched_clock() with a dumb function that returns
> cd.epoch_cyc so that the math turns out to be 0?

Great idea.

Making this code branchless with a special function sounds very much
better than using likely().
diff mbox

Patch

diff --git a/kernel/time/sched_clock.c b/kernel/time/sched_clock.c
index 3d21a8719444..cb69a47dfee4 100644
--- a/kernel/time/sched_clock.c
+++ b/kernel/time/sched_clock.c
@@ -18,28 +18,44 @@ 
 #include <linux/seqlock.h>
 #include <linux/bitops.h>
 
-struct clock_data {
-	ktime_t wrap_kt;
+/**
+ * struct clock_read_data - data required to read from sched_clock
+ *
+ * Care must be taken when updating this structure; it is read by
+ * some very hot code paths. It occupies <=48 bytes and, when combined
+ * with the seqcount used to synchronize access, comfortably fits into
+ * a 64 byte cache line.
+ */
+struct clock_read_data {
 	u64 epoch_ns;
 	u64 epoch_cyc;
-	seqcount_t seq;
-	unsigned long rate;
+	u64 sched_clock_mask;
+	u64 (*read_sched_clock)(void);
 	u32 mult;
 	u32 shift;
 	bool suspended;
 };
 
+/**
+ * struct clock_data - all data needed for sched_clock (including
+ *                     registration of a new clock source)
+ *
+ * The ordering of this structure has been chosen to optimize cache
+ * performance. In particular seq and read_data (combined) should fit
+ * into a single 64 byte cache line.
+ */
+struct clock_data {
+	seqcount_t seq;
+	struct clock_read_data read_data;
+	ktime_t wrap_kt;
+	unsigned long rate;
+};
+
 static struct hrtimer sched_clock_timer;
 static int irqtime = -1;
 
 core_param(irqtime, irqtime, int, 0400);
 
-static struct clock_data cd = {
-	.mult	= NSEC_PER_SEC / HZ,
-};
-
-static u64 __read_mostly sched_clock_mask;
-
 static u64 notrace jiffy_sched_clock_read(void)
 {
 	/*
@@ -49,7 +65,10 @@  static u64 notrace jiffy_sched_clock_read(void)
 	return (u64)(jiffies - INITIAL_JIFFIES);
 }
 
-static u64 __read_mostly (*read_sched_clock)(void) = jiffy_sched_clock_read;
+static struct clock_data cd ____cacheline_aligned = {
+	.read_data = { .mult = NSEC_PER_SEC / HZ,
+		       .read_sched_clock = jiffy_sched_clock_read, },
+};
 
 static inline u64 notrace cyc_to_ns(u64 cyc, u32 mult, u32 shift)
 {
@@ -60,15 +79,16 @@  unsigned long long notrace sched_clock(void)
 {
 	u64 cyc, res;
 	unsigned long seq;
+	struct clock_read_data *rd = &cd.read_data;
 
 	do {
 		seq = raw_read_seqcount_begin(&cd.seq);
 
-		res = cd.epoch_ns;
-		if (!cd.suspended) {
-			cyc = read_sched_clock();
-			cyc = (cyc - cd.epoch_cyc) & sched_clock_mask;
-			res += cyc_to_ns(cyc, cd.mult, cd.shift);
+		res = rd->epoch_ns;
+		if (!rd->suspended) {
+			cyc = rd->read_sched_clock();
+			cyc = (cyc - rd->epoch_cyc) & rd->sched_clock_mask;
+			res += cyc_to_ns(cyc, rd->mult, rd->shift);
 		}
 	} while (read_seqcount_retry(&cd.seq, seq));
 
@@ -83,16 +103,17 @@  static void notrace update_sched_clock(void)
 	unsigned long flags;
 	u64 cyc;
 	u64 ns;
+	struct clock_read_data *rd = &cd.read_data;
 
-	cyc = read_sched_clock();
-	ns = cd.epoch_ns +
-		cyc_to_ns((cyc - cd.epoch_cyc) & sched_clock_mask,
-			  cd.mult, cd.shift);
+	cyc = rd->read_sched_clock();
+	ns = rd->epoch_ns +
+	     cyc_to_ns((cyc - rd->epoch_cyc) & rd->sched_clock_mask,
+		       rd->mult, rd->shift);
 
 	raw_local_irq_save(flags);
 	raw_write_seqcount_begin(&cd.seq);
-	cd.epoch_ns = ns;
-	cd.epoch_cyc = cyc;
+	rd->epoch_ns = ns;
+	rd->epoch_cyc = cyc;
 	raw_write_seqcount_end(&cd.seq);
 	raw_local_irq_restore(flags);
 }
@@ -109,9 +130,9 @@  void __init sched_clock_register(u64 (*read)(void), int bits,
 {
 	u64 res, wrap, new_mask, new_epoch, cyc, ns;
 	u32 new_mult, new_shift;
-	ktime_t new_wrap_kt;
 	unsigned long r;
 	char r_unit;
+	struct clock_read_data *rd = &cd.read_data;
 
 	if (cd.rate > rate)
 		return;
@@ -130,17 +151,18 @@  void __init sched_clock_register(u64 (*read)(void), int bits,
 
 	/* update epoch for new counter and update epoch_ns from old counter*/
 	new_epoch = read();
-	cyc = read_sched_clock();
-	ns = cd.epoch_ns + cyc_to_ns((cyc - cd.epoch_cyc) & sched_clock_mask,
-			  cd.mult, cd.shift);
+	cyc = rd->read_sched_clock();
+	ns = rd->epoch_ns +
+	     cyc_to_ns((cyc - rd->epoch_cyc) & rd->sched_clock_mask,
+		       rd->mult, rd->shift);
 
 	raw_write_seqcount_begin(&cd.seq);
-	read_sched_clock = read;
-	sched_clock_mask = new_mask;
-	cd.mult = new_mult;
-	cd.shift = new_shift;
-	cd.epoch_cyc = new_epoch;
-	cd.epoch_ns = ns;
+	rd->read_sched_clock = read;
+	rd->sched_clock_mask = new_mask;
+	rd->mult = new_mult;
+	rd->shift = new_shift;
+	rd->epoch_cyc = new_epoch;
+	rd->epoch_ns = ns;
 	raw_write_seqcount_end(&cd.seq);
 
 	r = rate;
@@ -172,7 +194,7 @@  void __init sched_clock_postinit(void)
 	 * If no sched_clock function has been provided at that point,
 	 * make it the final one one.
 	 */
-	if (read_sched_clock == jiffy_sched_clock_read)
+	if (cd.read_data.read_sched_clock == jiffy_sched_clock_read)
 		sched_clock_register(jiffy_sched_clock_read, BITS_PER_LONG, HZ);
 
 	update_sched_clock();
@@ -188,17 +210,21 @@  void __init sched_clock_postinit(void)
 
 static int sched_clock_suspend(void)
 {
+	struct clock_read_data *rd = &cd.read_data;
+
 	update_sched_clock();
 	hrtimer_cancel(&sched_clock_timer);
-	cd.suspended = true;
+	rd->suspended = true;
 	return 0;
 }
 
 static void sched_clock_resume(void)
 {
-	cd.epoch_cyc = read_sched_clock();
+	struct clock_read_data *rd = &cd.read_data;
+
+	rd->epoch_cyc = rd->read_sched_clock();
 	hrtimer_start(&sched_clock_timer, cd.wrap_kt, HRTIMER_MODE_REL);
-	cd.suspended = false;
+	rd->suspended = false;
 }
 
 static struct syscore_ops sched_clock_ops = {