arch/arm/kernel/sched_clock.c

   1 /*
   2  * sched_clock.c: support for extending counters to full 64-bit ns counter
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  */
   8 #include <linux/clocksource.h>
   9 #include <linux/init.h>
  10 #include <linux/jiffies.h>
  11 #include <linux/kernel.h>
  12 #include <linux/moduleparam.h>
  13 #include <linux/sched.h>
  14 #include <linux/syscore_ops.h>
  15 #include <linux/timer.h>
  16
  17 #include <asm/sched_clock.h>
  18
  19 struct clock_data {
  20         u64 epoch_ns;
  21         u32 epoch_cyc;
  22         u32 epoch_cyc_copy;
  23         unsigned long rate;
  24         u32 mult;
  25         u32 shift;
  26         bool suspended;
  27         bool needs_suspend;
  28 };
  29
  30 static void sched_clock_poll(unsigned long wrap_ticks);
  31 static DEFINE_TIMER(sched_clock_timer, sched_clock_poll, 0, 0);
  32 static int irqtime = -1;
  33
  34 core_param(irqtime, irqtime, int, 0400);
  35
  36 static struct clock_data cd = {
  37         .mult   = NSEC_PER_SEC / HZ,
  38 };
  39
  40 static u32 __read_mostly sched_clock_mask = 0xffffffff;
  41
  42 static u32 notrace jiffy_sched_clock_read(void)
  43 {
  44         return (u32)(jiffies - INITIAL_JIFFIES);
  45 }
  46
  47 static u32 __read_mostly (*read_sched_clock)(void) = jiffy_sched_clock_read;
  48
  49 static inline u64 notrace cyc_to_ns(u64 cyc, u32 mult, u32 shift)
  50 {
  51         return (cyc * mult) >> shift;
  52 }
  53
  54 static unsigned long long notrace sched_clock_32(void)
  55 {
  56         u64 epoch_ns;
  57         u32 epoch_cyc;
  58         u32 cyc;
  59
  60         if (cd.suspended)
  61                 return cd.epoch_ns;
  62
  63         /*
  64          * Load the epoch_cyc and epoch_ns atomically.  We do this by
  65          * ensuring that we always write epoch_cyc, epoch_ns and
  66          * epoch_cyc_copy in strict order, and read them in strict order.
  67          * If epoch_cyc and epoch_cyc_copy are not equal, then we're in
  68          * the middle of an update, and we should repeat the load.
  69          */
  70         do {
  71                 epoch_cyc = cd.epoch_cyc;
  72                 smp_rmb();
  73                 epoch_ns = cd.epoch_ns;
  74                 smp_rmb();
  75         } while (epoch_cyc != cd.epoch_cyc_copy);
  76
  77         cyc = read_sched_clock();
  78         cyc = (cyc - epoch_cyc) & sched_clock_mask;
  79         return epoch_ns + cyc_to_ns(cyc, cd.mult, cd.shift);
  80 }
  81
  82 /*
  83  * Atomically update the sched_clock epoch.
  84  */
  85 static void notrace update_sched_clock(void)
  86 {
  87         unsigned long flags;
  88         u32 cyc;
  89         u64 ns;
  90
  91         cyc = read_sched_clock();
  92         ns = cd.epoch_ns +
  93                 cyc_to_ns((cyc - cd.epoch_cyc) & sched_clock_mask,
  94                           cd.mult, cd.shift);
  95         /*
  96          * Write epoch_cyc and epoch_ns in a way that the update is
  97          * detectable in cyc_to_fixed_sched_clock().
  98          */
  99         raw_local_irq_save(flags);
 100         cd.epoch_cyc_copy = cyc;
 101         smp_wmb();
 102         cd.epoch_ns = ns;
 103         smp_wmb();
 104         cd.epoch_cyc = cyc;
 105         raw_local_irq_restore(flags);
 106 }
 107
 108 static void sched_clock_poll(unsigned long wrap_ticks)
 109 {
 110         mod_timer(&sched_clock_timer, round_jiffies(jiffies + wrap_ticks));
 111         update_sched_clock();
 112 }
 113
 114 void __init setup_sched_clock(u32 (*read)(void), int bits, unsigned long rate)
 115 {
 116         unsigned long r, w;
 117         u64 res, wrap;
 118         char r_unit;
 119
 120         if (cd.rate > rate)
 121                 return;
 122
 123         BUG_ON(bits > 32);
 124         WARN_ON(!irqs_disabled());
 125         read_sched_clock = read;
 126         sched_clock_mask = (1 << bits) - 1;
 127         cd.rate = rate;
 128
 129         /* calculate the mult/shift to convert counter ticks to ns. */
 130         clocks_calc_mult_shift(&cd.mult, &cd.shift, rate, NSEC_PER_SEC, 0);
 131
 132         r = rate;
 133         if (r >= 4000000) {
 134                 r /= 1000000;
 135                 r_unit = 'M';
 136         } else if (r >= 1000) {
 137                 r /= 1000;
 138                 r_unit = 'k';
 139         } else
 140                 r_unit = ' ';
 141
 142         /* calculate how many ns until we wrap */
 143         wrap = cyc_to_ns((1ULL << bits) - 1, cd.mult, cd.shift);
 144         do_div(wrap, NSEC_PER_MSEC);
 145         w = wrap;
 146
 147         /* calculate the ns resolution of this counter */
 148         res = cyc_to_ns(1ULL, cd.mult, cd.shift);
 149         pr_info("sched_clock: %u bits at %lu%cHz, resolution %lluns, wraps every %lums\n",
 150                 bits, r, r_unit, res, w);
 151
 152         /*
 153          * Start the timer to keep sched_clock() properly updated and
 154          * sets the initial epoch.
 155          */
 156         sched_clock_timer.data = msecs_to_jiffies(w - (w / 10));
 157         update_sched_clock();
 158
 159         /*
 160          * Ensure that sched_clock() starts off at 0ns
 161          */
 162         cd.epoch_ns = 0;
 163
 164         /* Enable IRQ time accounting if we have a fast enough sched_clock */
 165         if (irqtime > 0 || (irqtime == -1 && rate >= 1000000))
 166                 enable_sched_clock_irqtime();
 167
 168         pr_debug("Registered %pF as sched_clock source\n", read);
 169 }
 170
 171 unsigned long long __read_mostly (*sched_clock_func)(void) = sched_clock_32;
 172
 173 unsigned long long notrace sched_clock(void)
 174 {
 175         return sched_clock_func();
 176 }
 177
 178 void __init sched_clock_postinit(void)
 179 {
 180         /*
 181          * If no sched_clock function has been provided at that point,
 182          * make it the final one one.
 183          */
 184         if (read_sched_clock == jiffy_sched_clock_read)
 185                 setup_sched_clock(jiffy_sched_clock_read, 32, HZ);
 186
 187         sched_clock_poll(sched_clock_timer.data);
 188 }
 189
 190 static int sched_clock_suspend(void)
 191 {
 192         sched_clock_poll(sched_clock_timer.data);
 193         cd.suspended = true;
 194         return 0;
 195 }
 196
 197 static void sched_clock_resume(void)
 198 {
 199         cd.epoch_cyc = read_sched_clock();
 200         cd.epoch_cyc_copy = cd.epoch_cyc;
 201         cd.suspended = false;
 202 }
 203
 204 static struct syscore_ops sched_clock_ops = {
 205         .suspend = sched_clock_suspend,
 206         .resume = sched_clock_resume,
 207 };
 208
 209 static int __init sched_clock_syscore_init(void)
 210 {
 211         register_syscore_ops(&sched_clock_ops);
 212         return 0;
 213 }
 214 device_initcall(sched_clock_syscore_init);