sched_clock.c 4.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212
  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. #include <linux/sched_clock.h>
  17. struct clock_data {
  18. u64 epoch_ns;
  19. u32 epoch_cyc;
  20. u32 epoch_cyc_copy;
  21. unsigned long rate;
  22. u32 mult;
  23. u32 shift;
  24. bool suspended;
  25. };
  26. static void sched_clock_poll(unsigned long wrap_ticks);
  27. static DEFINE_TIMER(sched_clock_timer, sched_clock_poll, 0, 0);
  28. static int irqtime = -1;
  29. core_param(irqtime, irqtime, int, 0400);
  30. static struct clock_data cd = {
  31. .mult = NSEC_PER_SEC / HZ,
  32. };
  33. static u32 __read_mostly sched_clock_mask = 0xffffffff;
  34. static u32 notrace jiffy_sched_clock_read(void)
  35. {
  36. return (u32)(jiffies - INITIAL_JIFFIES);
  37. }
  38. static u32 __read_mostly (*read_sched_clock)(void) = jiffy_sched_clock_read;
  39. static inline u64 notrace cyc_to_ns(u64 cyc, u32 mult, u32 shift)
  40. {
  41. return (cyc * mult) >> shift;
  42. }
  43. static unsigned long long notrace sched_clock_32(void)
  44. {
  45. u64 epoch_ns;
  46. u32 epoch_cyc;
  47. u32 cyc;
  48. if (cd.suspended)
  49. return cd.epoch_ns;
  50. /*
  51. * Load the epoch_cyc and epoch_ns atomically. We do this by
  52. * ensuring that we always write epoch_cyc, epoch_ns and
  53. * epoch_cyc_copy in strict order, and read them in strict order.
  54. * If epoch_cyc and epoch_cyc_copy are not equal, then we're in
  55. * the middle of an update, and we should repeat the load.
  56. */
  57. do {
  58. epoch_cyc = cd.epoch_cyc;
  59. smp_rmb();
  60. epoch_ns = cd.epoch_ns;
  61. smp_rmb();
  62. } while (epoch_cyc != cd.epoch_cyc_copy);
  63. cyc = read_sched_clock();
  64. cyc = (cyc - epoch_cyc) & sched_clock_mask;
  65. return epoch_ns + cyc_to_ns(cyc, cd.mult, cd.shift);
  66. }
  67. /*
  68. * Atomically update the sched_clock epoch.
  69. */
  70. static void notrace update_sched_clock(void)
  71. {
  72. unsigned long flags;
  73. u32 cyc;
  74. u64 ns;
  75. cyc = read_sched_clock();
  76. ns = cd.epoch_ns +
  77. cyc_to_ns((cyc - cd.epoch_cyc) & sched_clock_mask,
  78. cd.mult, cd.shift);
  79. /*
  80. * Write epoch_cyc and epoch_ns in a way that the update is
  81. * detectable in cyc_to_fixed_sched_clock().
  82. */
  83. raw_local_irq_save(flags);
  84. cd.epoch_cyc_copy = cyc;
  85. smp_wmb();
  86. cd.epoch_ns = ns;
  87. smp_wmb();
  88. cd.epoch_cyc = cyc;
  89. raw_local_irq_restore(flags);
  90. }
  91. static void sched_clock_poll(unsigned long wrap_ticks)
  92. {
  93. mod_timer(&sched_clock_timer, round_jiffies(jiffies + wrap_ticks));
  94. update_sched_clock();
  95. }
  96. void __init setup_sched_clock(u32 (*read)(void), int bits, unsigned long rate)
  97. {
  98. unsigned long r, w;
  99. u64 res, wrap;
  100. char r_unit;
  101. if (cd.rate > rate)
  102. return;
  103. BUG_ON(bits > 32);
  104. WARN_ON(!irqs_disabled());
  105. read_sched_clock = read;
  106. sched_clock_mask = (1ULL << bits) - 1;
  107. cd.rate = rate;
  108. /* calculate the mult/shift to convert counter ticks to ns. */
  109. clocks_calc_mult_shift(&cd.mult, &cd.shift, rate, NSEC_PER_SEC, 0);
  110. r = rate;
  111. if (r >= 4000000) {
  112. r /= 1000000;
  113. r_unit = 'M';
  114. } else if (r >= 1000) {
  115. r /= 1000;
  116. r_unit = 'k';
  117. } else
  118. r_unit = ' ';
  119. /* calculate how many ns until we wrap */
  120. wrap = cyc_to_ns((1ULL << bits) - 1, cd.mult, cd.shift);
  121. do_div(wrap, NSEC_PER_MSEC);
  122. w = wrap;
  123. /* calculate the ns resolution of this counter */
  124. res = cyc_to_ns(1ULL, cd.mult, cd.shift);
  125. pr_info("sched_clock: %u bits at %lu%cHz, resolution %lluns, wraps every %lums\n",
  126. bits, r, r_unit, res, w);
  127. /*
  128. * Start the timer to keep sched_clock() properly updated and
  129. * sets the initial epoch.
  130. */
  131. sched_clock_timer.data = msecs_to_jiffies(w - (w / 10));
  132. update_sched_clock();
  133. /*
  134. * Ensure that sched_clock() starts off at 0ns
  135. */
  136. cd.epoch_ns = 0;
  137. /* Enable IRQ time accounting if we have a fast enough sched_clock */
  138. if (irqtime > 0 || (irqtime == -1 && rate >= 1000000))
  139. enable_sched_clock_irqtime();
  140. pr_debug("Registered %pF as sched_clock source\n", read);
  141. }
  142. unsigned long long __read_mostly (*sched_clock_func)(void) = sched_clock_32;
  143. unsigned long long notrace sched_clock(void)
  144. {
  145. return sched_clock_func();
  146. }
  147. void __init sched_clock_postinit(void)
  148. {
  149. /*
  150. * If no sched_clock function has been provided at that point,
  151. * make it the final one one.
  152. */
  153. if (read_sched_clock == jiffy_sched_clock_read)
  154. setup_sched_clock(jiffy_sched_clock_read, 32, HZ);
  155. sched_clock_poll(sched_clock_timer.data);
  156. }
  157. static int sched_clock_suspend(void)
  158. {
  159. sched_clock_poll(sched_clock_timer.data);
  160. cd.suspended = true;
  161. return 0;
  162. }
  163. static void sched_clock_resume(void)
  164. {
  165. cd.epoch_cyc = read_sched_clock();
  166. cd.epoch_cyc_copy = cd.epoch_cyc;
  167. cd.suspended = false;
  168. }
  169. static struct syscore_ops sched_clock_ops = {
  170. .suspend = sched_clock_suspend,
  171. .resume = sched_clock_resume,
  172. };
  173. static int __init sched_clock_syscore_init(void)
  174. {
  175. register_syscore_ops(&sched_clock_ops);
  176. return 0;
  177. }
  178. device_initcall(sched_clock_syscore_init);