hangcheck-timer.c 6.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204
  1. /*
  2. * hangcheck-timer.c
  3. *
  4. * Driver for a little io fencing timer.
  5. *
  6. * Copyright (C) 2002, 2003 Oracle. All rights reserved.
  7. *
  8. * Author: Joel Becker <joel.becker@oracle.com>
  9. *
  10. * This program is free software; you can redistribute it and/or
  11. * modify it under the terms of the GNU General Public
  12. * License version 2 as published by the Free Software Foundation.
  13. *
  14. * This program is distributed in the hope that it will be useful,
  15. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  17. * General Public License for more details.
  18. *
  19. * You should have received a copy of the GNU General Public
  20. * License along with this program; if not, write to the
  21. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  22. * Boston, MA 021110-1307, USA.
  23. */
  24. /*
  25. * The hangcheck-timer driver uses the TSC to catch delays that
  26. * jiffies does not notice. A timer is set. When the timer fires, it
  27. * checks whether it was delayed and if that delay exceeds a given
  28. * margin of error. The hangcheck_tick module paramter takes the timer
  29. * duration in seconds. The hangcheck_margin parameter defines the
  30. * margin of error, in seconds. The defaults are 60 seconds for the
  31. * timer and 180 seconds for the margin of error. IOW, a timer is set
  32. * for 60 seconds. When the timer fires, the callback checks the
  33. * actual duration that the timer waited. If the duration exceeds the
  34. * alloted time and margin (here 60 + 180, or 240 seconds), the machine
  35. * is restarted. A healthy machine will have the duration match the
  36. * expected timeout very closely.
  37. */
  38. #include <linux/module.h>
  39. #include <linux/moduleparam.h>
  40. #include <linux/types.h>
  41. #include <linux/kernel.h>
  42. #include <linux/fs.h>
  43. #include <linux/mm.h>
  44. #include <linux/reboot.h>
  45. #include <linux/smp_lock.h>
  46. #include <linux/init.h>
  47. #include <linux/delay.h>
  48. #include <asm/uaccess.h>
  49. #include <linux/sysrq.h>
  50. #define VERSION_STR "0.9.0"
  51. #define DEFAULT_IOFENCE_MARGIN 60 /* Default fudge factor, in seconds */
  52. #define DEFAULT_IOFENCE_TICK 180 /* Default timer timeout, in seconds */
  53. static int hangcheck_tick = DEFAULT_IOFENCE_TICK;
  54. static int hangcheck_margin = DEFAULT_IOFENCE_MARGIN;
  55. static int hangcheck_reboot; /* Defaults to not reboot */
  56. static int hangcheck_dump_tasks; /* Defaults to not dumping SysRQ T */
  57. /* options - modular */
  58. module_param(hangcheck_tick, int, 0);
  59. MODULE_PARM_DESC(hangcheck_tick, "Timer delay.");
  60. module_param(hangcheck_margin, int, 0);
  61. MODULE_PARM_DESC(hangcheck_margin, "If the hangcheck timer has been delayed more than hangcheck_margin seconds, the driver will fire.");
  62. module_param(hangcheck_reboot, int, 0);
  63. MODULE_PARM_DESC(hangcheck_reboot, "If nonzero, the machine will reboot when the timer margin is exceeded.");
  64. module_param(hangcheck_dump_tasks, int, 0);
  65. MODULE_PARM_DESC(hangcheck_dump_tasks, "If nonzero, the machine will dump the system task state when the timer margin is exceeded.");
  66. MODULE_AUTHOR("Oracle");
  67. MODULE_DESCRIPTION("Hangcheck-timer detects when the system has gone out to lunch past a certain margin.");
  68. MODULE_LICENSE("GPL");
  69. MODULE_VERSION(VERSION_STR);
  70. /* options - nonmodular */
  71. #ifndef MODULE
  72. static int __init hangcheck_parse_tick(char *str)
  73. {
  74. int par;
  75. if (get_option(&str,&par))
  76. hangcheck_tick = par;
  77. return 1;
  78. }
  79. static int __init hangcheck_parse_margin(char *str)
  80. {
  81. int par;
  82. if (get_option(&str,&par))
  83. hangcheck_margin = par;
  84. return 1;
  85. }
  86. static int __init hangcheck_parse_reboot(char *str)
  87. {
  88. int par;
  89. if (get_option(&str,&par))
  90. hangcheck_reboot = par;
  91. return 1;
  92. }
  93. static int __init hangcheck_parse_dump_tasks(char *str)
  94. {
  95. int par;
  96. if (get_option(&str,&par))
  97. hangcheck_dump_tasks = par;
  98. return 1;
  99. }
  100. __setup("hcheck_tick", hangcheck_parse_tick);
  101. __setup("hcheck_margin", hangcheck_parse_margin);
  102. __setup("hcheck_reboot", hangcheck_parse_reboot);
  103. __setup("hcheck_dump_tasks", hangcheck_parse_dump_tasks);
  104. #endif /* not MODULE */
  105. #if defined(CONFIG_S390)
  106. # define HAVE_MONOTONIC
  107. # define TIMER_FREQ 1000000000ULL
  108. #elif defined(CONFIG_IA64)
  109. # define TIMER_FREQ ((unsigned long long)local_cpu_data->itc_freq)
  110. #else
  111. # define TIMER_FREQ (HZ*loops_per_jiffy)
  112. #endif
  113. #ifdef HAVE_MONOTONIC
  114. extern unsigned long long monotonic_clock(void);
  115. #else
  116. static inline unsigned long long monotonic_clock(void)
  117. {
  118. return get_cycles();
  119. }
  120. #endif /* HAVE_MONOTONIC */
  121. /* Last time scheduled */
  122. static unsigned long long hangcheck_tsc, hangcheck_tsc_margin;
  123. static void hangcheck_fire(unsigned long);
  124. static DEFINE_TIMER(hangcheck_ticktock, hangcheck_fire, 0, 0);
  125. static void hangcheck_fire(unsigned long data)
  126. {
  127. unsigned long long cur_tsc, tsc_diff;
  128. cur_tsc = monotonic_clock();
  129. if (cur_tsc > hangcheck_tsc)
  130. tsc_diff = cur_tsc - hangcheck_tsc;
  131. else
  132. tsc_diff = (cur_tsc + (~0ULL - hangcheck_tsc)); /* or something */
  133. if (tsc_diff > hangcheck_tsc_margin) {
  134. if (hangcheck_dump_tasks) {
  135. printk(KERN_CRIT "Hangcheck: Task state:\n");
  136. #ifdef CONFIG_MAGIC_SYSRQ
  137. handle_sysrq('t', NULL);
  138. #endif /* CONFIG_MAGIC_SYSRQ */
  139. }
  140. if (hangcheck_reboot) {
  141. printk(KERN_CRIT "Hangcheck: hangcheck is restarting the machine.\n");
  142. emergency_restart();
  143. } else {
  144. printk(KERN_CRIT "Hangcheck: hangcheck value past margin!\n");
  145. }
  146. }
  147. mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
  148. hangcheck_tsc = monotonic_clock();
  149. }
  150. static int __init hangcheck_init(void)
  151. {
  152. printk("Hangcheck: starting hangcheck timer %s (tick is %d seconds, margin is %d seconds).\n",
  153. VERSION_STR, hangcheck_tick, hangcheck_margin);
  154. #if defined (HAVE_MONOTONIC)
  155. printk("Hangcheck: Using monotonic_clock().\n");
  156. #else
  157. printk("Hangcheck: Using get_cycles().\n");
  158. #endif /* HAVE_MONOTONIC */
  159. hangcheck_tsc_margin =
  160. (unsigned long long)(hangcheck_margin + hangcheck_tick);
  161. hangcheck_tsc_margin *= (unsigned long long)TIMER_FREQ;
  162. hangcheck_tsc = monotonic_clock();
  163. mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
  164. return 0;
  165. }
  166. static void __exit hangcheck_exit(void)
  167. {
  168. del_timer_sync(&hangcheck_ticktock);
  169. printk("Hangcheck: Stopped hangcheck timer.\n");
  170. }
  171. module_init(hangcheck_init);
  172. module_exit(hangcheck_exit);