[PATCH v3 11/36] hrtimer: Store running timer in hrtimer_clock_base

[Anna-Maria Gleixner <anna-maria@linutronix.de>]

The pointer to the currently running timer is stored in hrtimer_cpu_base
before the base lock is dropped and the callback is invoked.

This results in two levels of indirections and the upcoming support for
softirq based hrtimer requires splitting the "running" storage into soft
and hard irq context expiry.

Storing both in the cpu base would require conditionals in all code paths
accessing that information.

It's possible to have a per clock base sequence count and running pointer
without changing the semantics of the related mechanisms because the timer
base pointer cannot be changed while a timer is running the callback.

Unfortunately this makes cpu_clock base larger than 32 bytes on 32bit
kernels. Instead of having huge gaps due to alignment, remove the alignment
and let the compiler pack cpu base for 32bit. The resulting cache access
patterns are fortunately not really different from the current
behaviour. On 64bit kernels the 64byte alignment stays and the behaviour is
unchanged. This was determined by analyzing the resulting layout and
looking at the number of cache lines involved for the frequently used
clocks.

Signed-off-by: Anna-Maria Gleixner <anna-maria@linutronix.de>
---
 include/linux/hrtimer.h | 20 +++++++++-----------
 kernel/time/hrtimer.c   | 28 +++++++++++++---------------
 2 files changed, 22 insertions(+), 26 deletions(-)

diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index 28f267cf2851..1bae7b9f071d 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -118,9 +118,9 @@ struct hrtimer_sleeper {
 };
 
 #ifdef CONFIG_64BIT
-# define HRTIMER_CLOCK_BASE_ALIGN	64
+# define __hrtimer_clock_base_align	____cacheline_aligned
 #else
-# define HRTIMER_CLOCK_BASE_ALIGN	32
+# define __hrtimer_clock_base_align
 #endif
 
 /**
@@ -129,18 +129,22 @@ struct hrtimer_sleeper {
  * @index:		clock type index for per_cpu support when moving a
  *			timer to a base on another cpu.
  * @clockid:		clock id for per_cpu support
+ * @seq:		seqcount around __run_hrtimer
+ * @running:		pointer to the currently running hrtimer
  * @active:		red black tree root node for the active timers
  * @get_time:		function to retrieve the current time of the clock
  * @offset:		offset of this clock to the monotonic base
  */
 struct hrtimer_clock_base {
 	struct hrtimer_cpu_base	*cpu_base;
-	int			index;
+	unsigned int		index;
 	clockid_t		clockid;
+	seqcount_t		seq;
+	struct hrtimer		*running;
 	struct timerqueue_head	active;
 	ktime_t			(*get_time)(void);
 	ktime_t			offset;
-} __attribute__((__aligned__(HRTIMER_CLOCK_BASE_ALIGN)));
+} __hrtimer_clock_base_align;
 
 enum  hrtimer_base_type {
 	HRTIMER_BASE_MONOTONIC,
@@ -154,8 +158,6 @@ enum  hrtimer_base_type {
  * struct hrtimer_cpu_base - the per cpu clock bases
  * @lock:		lock protecting the base and associated clock bases
  *			and timers
- * @seq:		seqcount around __run_hrtimer
- * @running:		pointer to the currently running hrtimer
  * @cpu:		cpu number
  * @active_bases:	Bitfield to mark bases with active timers
  * @clock_was_set_seq:	Sequence counter of clock was set events
@@ -177,8 +179,6 @@ enum  hrtimer_base_type {
  */
 struct hrtimer_cpu_base {
 	raw_spinlock_t			lock;
-	seqcount_t			seq;
-	struct hrtimer			*running;
 	unsigned int			cpu;
 	unsigned int			active_bases;
 	unsigned int			clock_was_set_seq;
@@ -198,8 +198,6 @@ struct hrtimer_cpu_base {
 
 static inline void hrtimer_set_expires(struct hrtimer *timer, ktime_t time)
 {
-	BUILD_BUG_ON(sizeof(struct hrtimer_clock_base) > HRTIMER_CLOCK_BASE_ALIGN);
-
 	timer->node.expires = time;
 	timer->_softexpires = time;
 }
@@ -424,7 +422,7 @@ static inline int hrtimer_is_queued(struct hrtimer *timer)
  */
 static inline int hrtimer_callback_running(struct hrtimer *timer)
 {
-	return timer->base->cpu_base->running == timer;
+	return timer->base->running == timer;
 }
 
 /* Forward a hrtimer so it expires after now: */
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 326faa5719ff..046b509beed6 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -70,7 +70,6 @@
 DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
 {
 	.lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
-	.seq = SEQCNT_ZERO(hrtimer_bases.seq),
 	.clock_base =
 	{
 		{
@@ -118,7 +117,6 @@ static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
  * timer->base->cpu_base
  */
 static struct hrtimer_cpu_base migration_cpu_base = {
-	.seq = SEQCNT_ZERO(migration_cpu_base),
 	.clock_base = { { .cpu_base = &migration_cpu_base, }, },
 };
 
@@ -1148,19 +1146,19 @@ EXPORT_SYMBOL_GPL(hrtimer_init);
  */
 bool hrtimer_active(const struct hrtimer *timer)
 {
-	struct hrtimer_cpu_base *cpu_base;
+	struct hrtimer_clock_base *base;
 	unsigned int seq;
 
 	do {
-		cpu_base = READ_ONCE(timer->base->cpu_base);
-		seq = raw_read_seqcount_begin(&cpu_base->seq);
+		base = READ_ONCE(timer->base);
+		seq = raw_read_seqcount_begin(&base->seq);
 
 		if (timer->state != HRTIMER_STATE_INACTIVE ||
-		    cpu_base->running == timer)
+		    base->running == timer)
 			return true;
 
-	} while (read_seqcount_retry(&cpu_base->seq, seq) ||
-		 cpu_base != READ_ONCE(timer->base->cpu_base));
+	} while (read_seqcount_retry(&base->seq, seq) ||
+		 base != READ_ONCE(timer->base));
 
 	return false;
 }
@@ -1194,16 +1192,16 @@ static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
 	lockdep_assert_held(&cpu_base->lock);
 
 	debug_deactivate(timer);
-	cpu_base->running = timer;
+	base->running = timer;
 
 	/*
 	 * Separate the ->running assignment from the ->state assignment.
 	 *
 	 * As with a regular write barrier, this ensures the read side in
-	 * hrtimer_active() cannot observe cpu_base->running == NULL &&
+	 * hrtimer_active() cannot observe base->running == NULL &&
 	 * timer->state == INACTIVE.
 	 */
-	raw_write_seqcount_barrier(&cpu_base->seq);
+	raw_write_seqcount_barrier(&base->seq);
 
 	__remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
 	fn = timer->function;
@@ -1244,13 +1242,13 @@ static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
 	 * Separate the ->running assignment from the ->state assignment.
 	 *
 	 * As with a regular write barrier, this ensures the read side in
-	 * hrtimer_active() cannot observe cpu_base->running == NULL &&
+	 * hrtimer_active() cannot observe base->running.timer == NULL &&
 	 * timer->state == INACTIVE.
 	 */
-	raw_write_seqcount_barrier(&cpu_base->seq);
+	raw_write_seqcount_barrier(&base->seq);
 
-	WARN_ON_ONCE(cpu_base->running != timer);
-	cpu_base->running = NULL;
+	WARN_ON_ONCE(base->running != timer);
+	base->running = NULL;
 }
 
 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now)
-- 
2.11.0