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 };
  28
  29 static void sched_clock_poll(unsigned long wrap_ticks);
  30 static DEFINE_TIMER(sched_clock_timer, sched_clock_poll, 0, 0);
  31 static int irqtime = -1;
  32
  33 core_param(irqtime, irqtime, int, 0400);
  34
  35 static struct clock_data cd = {
  36         .mult   = NSEC_PER_SEC / HZ,
  37 };
  38
  39 static u32 __read_mostly sched_clock_mask = 0xffffffff;
  40
  41 static u32 notrace jiffy_sched_clock_read(void)
  42 {
  43         return (u32)(jiffies - INITIAL_JIFFIES);
  44 }
  45
  46 static u32 __read_mostly (*read_sched_clock)(void) = jiffy_sched_clock_read;
  47
  48 static inline u64 notrace cyc_to_ns(u64 cyc, u32 mult, u32 shift)
  49 {
  50         return (cyc * mult) >> shift;
  51 }
  52
  53 static unsigned long long notrace cyc_to_sched_clock(u32 cyc, u32 mask)
  54 {
  55         u64 epoch_ns;
  56         u32 epoch_cyc;
  57
  58         /*
  59          * Load the epoch_cyc and epoch_ns atomically.  We do this by
  60          * ensuring that we always write epoch_cyc, epoch_ns and
  61          * epoch_cyc_copy in strict order, and read them in strict order.
  62          * If epoch_cyc and epoch_cyc_copy are not equal, then we're in
  63          * the middle of an update, and we should repeat the load.
  64          */
  65         do {
  66                 epoch_cyc = cd.epoch_cyc;
  67                 smp_rmb();
  68                 epoch_ns = cd.epoch_ns;
  69                 smp_rmb();
  70         } while (epoch_cyc != cd.epoch_cyc_copy);
  71
  72         return epoch_ns + cyc_to_ns((cyc - epoch_cyc) & mask, cd.mult, cd.shift);
  73 }
  74
  75 /*
  76  * Atomically update the sched_clock epoch.
  77  */
  78 static void notrace update_sched_clock(void)
  79 {
  80         unsigned long flags;
  81         u32 cyc;
  82         u64 ns;
  83
  84         cyc = read_sched_clock();
  85         ns = cd.epoch_ns +
  86                 cyc_to_ns((cyc - cd.epoch_cyc) & sched_clock_mask,
  87                           cd.mult, cd.shift);
  88         /*
  89          * Write epoch_cyc and epoch_ns in a way that the update is
  90          * detectable in cyc_to_fixed_sched_clock().
  91          */
  92         raw_local_irq_save(flags);
  93         cd.epoch_cyc_copy = cyc;
  94         smp_wmb();
  95         cd.epoch_ns = ns;
  96         smp_wmb();
  97         cd.epoch_cyc = cyc;
  98         raw_local_irq_restore(flags);
  99 }
 100
 101 static void sched_clock_poll(unsigned long wrap_ticks)
 102 {
 103         mod_timer(&sched_clock_timer, round_jiffies(jiffies + wrap_ticks));
 104         update_sched_clock();
 105 }
 106
 107 void __init setup_sched_clock(u32 (*read)(void), int bits, unsigned long rate)
 108 {
 109         unsigned long r, w;
 110         u64 res, wrap;
 111         char r_unit;
 112
 113         if (cd.rate > rate)
 114                 return;
 115
 116         BUG_ON(bits > 32);
 117         WARN_ON(!irqs_disabled());
 118         read_sched_clock = read;
 119         sched_clock_mask = (1 << bits) - 1;
 120         cd.rate = rate;
 121
 122         /* calculate the mult/shift to convert counter ticks to ns. */
 123         clocks_calc_mult_shift(&cd.mult, &cd.shift, rate, NSEC_PER_SEC, 0);
 124
 125         r = rate;
 126         if (r >= 4000000) {
 127                 r /= 1000000;
 128                 r_unit = 'M';
 129         } else if (r >= 1000) {
 130                 r /= 1000;
 131                 r_unit = 'k';
 132         } else
 133                 r_unit = ' ';
 134
 135         /* calculate how many ns until we wrap */
 136         wrap = cyc_to_ns((1ULL << bits) - 1, cd.mult, cd.shift);
 137         do_div(wrap, NSEC_PER_MSEC);
 138         w = wrap;
 139
 140         /* calculate the ns resolution of this counter */
 141         res = cyc_to_ns(1ULL, cd.mult, cd.shift);
 142         pr_info("sched_clock: %u bits at %lu%cHz, resolution %lluns, wraps every %lums\n",
 143                 bits, r, r_unit, res, w);
 144
 145         /*
 146          * Start the timer to keep sched_clock() properly updated and
 147          * sets the initial epoch.
 148          */
 149         sched_clock_timer.data = msecs_to_jiffies(w - (w / 10));
 150         update_sched_clock();
 151
 152         /*
 153          * Ensure that sched_clock() starts off at 0ns
 154          */
 155         cd.epoch_ns = 0;
 156
 157         /* Enable IRQ time accounting if we have a fast enough sched_clock */
 158         if (irqtime > 0 || (irqtime == -1 && rate >= 1000000))
 159                 enable_sched_clock_irqtime();
 160
 161         pr_debug("Registered %pF as sched_clock source\n", read);
 162 }
 163
 164 static unsigned long long notrace sched_clock_32(void)
 165 {
 166         u32 cyc = read_sched_clock();
 167         return cyc_to_sched_clock(cyc, sched_clock_mask);
 168 }
 169
 170 unsigned long long __read_mostly (*sched_clock_func)(void) = sched_clock_32;
 171
 172 unsigned long long notrace sched_clock(void)
 173 {
 174         if (cd.suspended)
 175                 return cd.epoch_ns;
 176
 177         return sched_clock_func();
 178 }
 179
 180 void __init sched_clock_postinit(void)
 181 {
 182         /*
 183          * If no sched_clock function has been provided at that point,
 184          * make it the final one one.
 185          */
 186         if (read_sched_clock == jiffy_sched_clock_read)
 187                 setup_sched_clock(jiffy_sched_clock_read, 32, HZ);
 188
 189         sched_clock_poll(sched_clock_timer.data);
 190 }
 191
 192 static int sched_clock_suspend(void)
 193 {
 194         sched_clock_poll(sched_clock_timer.data);
 195         cd.suspended = true;
 196         return 0;
 197 }
 198
 199 static void sched_clock_resume(void)
 200 {
 201         cd.epoch_cyc = read_sched_clock();
 202         cd.epoch_cyc_copy = cd.epoch_cyc;
 203         cd.suspended = false;
 204 }
 205
 206 static struct syscore_ops sched_clock_ops = {
 207         .suspend = sched_clock_suspend,
 208         .resume = sched_clock_resume,
 209 };
 210
 211 static int __init sched_clock_syscore_init(void)
 212 {
 213         register_syscore_ops(&sched_clock_ops);
 214         return 0;
 215 }
 216 device_initcall(sched_clock_syscore_init);