hrtimer: Store running timer in hrtimer_clock_base
authorAnna-Maria Gleixner <anna-maria@linutronix.de>
Thu, 21 Dec 2017 10:41:40 +0000 (11:41 +0100)
committerIngo Molnar <mingo@kernel.org>
Tue, 16 Jan 2018 01:35:46 +0000 (02:35 +0100)
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 32-bit
kernels. Instead of having huge gaps due to alignment, remove the alignment
and let the compiler pack CPU base for 32-bit kernels. The resulting cache access
patterns are fortunately not really different from the current
behaviour. On 64-bit kernels the 64-byte 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>
Cc: Christoph Hellwig <hch@lst.de>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: keescook@chromium.org
Link: http://lkml.kernel.org/r/20171221104205.7269-12-anna-maria@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
include/linux/hrtimer.h
kernel/time/hrtimer.c

index 28f267c..1bae7b9 100644 (file)
@@ -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: */
index cfcf8de..e56805f 100644 (file)
@@ -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)