diff options
Diffstat (limited to 'kernel/timer.c')
-rw-r--r-- | kernel/timer.c | 459 |
1 files changed, 1 insertions, 458 deletions
diff --git a/kernel/timer.c b/kernel/timer.c index dbe966feff2f..ba41af2bb6cc 100644 --- a/kernel/timer.c +++ b/kernel/timer.c | |||
@@ -1,7 +1,7 @@ | |||
1 | /* | 1 | /* |
2 | * linux/kernel/timer.c | 2 | * linux/kernel/timer.c |
3 | * | 3 | * |
4 | * Kernel internal timers, kernel timekeeping, basic process system calls | 4 | * Kernel internal timers, basic process system calls |
5 | * | 5 | * |
6 | * Copyright (C) 1991, 1992 Linus Torvalds | 6 | * Copyright (C) 1991, 1992 Linus Torvalds |
7 | * | 7 | * |
@@ -794,455 +794,6 @@ unsigned long next_timer_interrupt(void) | |||
794 | 794 | ||
795 | #endif | 795 | #endif |
796 | 796 | ||
797 | /******************************************************************/ | ||
798 | |||
799 | /* | ||
800 | * The current time | ||
801 | * wall_to_monotonic is what we need to add to xtime (or xtime corrected | ||
802 | * for sub jiffie times) to get to monotonic time. Monotonic is pegged | ||
803 | * at zero at system boot time, so wall_to_monotonic will be negative, | ||
804 | * however, we will ALWAYS keep the tv_nsec part positive so we can use | ||
805 | * the usual normalization. | ||
806 | */ | ||
807 | struct timespec xtime __attribute__ ((aligned (16))); | ||
808 | struct timespec wall_to_monotonic __attribute__ ((aligned (16))); | ||
809 | |||
810 | EXPORT_SYMBOL(xtime); | ||
811 | |||
812 | |||
813 | /* XXX - all of this timekeeping code should be later moved to time.c */ | ||
814 | #include <linux/clocksource.h> | ||
815 | static struct clocksource *clock; /* pointer to current clocksource */ | ||
816 | |||
817 | #ifdef CONFIG_GENERIC_TIME | ||
818 | /** | ||
819 | * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook | ||
820 | * | ||
821 | * private function, must hold xtime_lock lock when being | ||
822 | * called. Returns the number of nanoseconds since the | ||
823 | * last call to update_wall_time() (adjusted by NTP scaling) | ||
824 | */ | ||
825 | static inline s64 __get_nsec_offset(void) | ||
826 | { | ||
827 | cycle_t cycle_now, cycle_delta; | ||
828 | s64 ns_offset; | ||
829 | |||
830 | /* read clocksource: */ | ||
831 | cycle_now = clocksource_read(clock); | ||
832 | |||
833 | /* calculate the delta since the last update_wall_time: */ | ||
834 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | ||
835 | |||
836 | /* convert to nanoseconds: */ | ||
837 | ns_offset = cyc2ns(clock, cycle_delta); | ||
838 | |||
839 | return ns_offset; | ||
840 | } | ||
841 | |||
842 | /** | ||
843 | * __get_realtime_clock_ts - Returns the time of day in a timespec | ||
844 | * @ts: pointer to the timespec to be set | ||
845 | * | ||
846 | * Returns the time of day in a timespec. Used by | ||
847 | * do_gettimeofday() and get_realtime_clock_ts(). | ||
848 | */ | ||
849 | static inline void __get_realtime_clock_ts(struct timespec *ts) | ||
850 | { | ||
851 | unsigned long seq; | ||
852 | s64 nsecs; | ||
853 | |||
854 | do { | ||
855 | seq = read_seqbegin(&xtime_lock); | ||
856 | |||
857 | *ts = xtime; | ||
858 | nsecs = __get_nsec_offset(); | ||
859 | |||
860 | } while (read_seqretry(&xtime_lock, seq)); | ||
861 | |||
862 | timespec_add_ns(ts, nsecs); | ||
863 | } | ||
864 | |||
865 | /** | ||
866 | * getnstimeofday - Returns the time of day in a timespec | ||
867 | * @ts: pointer to the timespec to be set | ||
868 | * | ||
869 | * Returns the time of day in a timespec. | ||
870 | */ | ||
871 | void getnstimeofday(struct timespec *ts) | ||
872 | { | ||
873 | __get_realtime_clock_ts(ts); | ||
874 | } | ||
875 | |||
876 | EXPORT_SYMBOL(getnstimeofday); | ||
877 | |||
878 | /** | ||
879 | * do_gettimeofday - Returns the time of day in a timeval | ||
880 | * @tv: pointer to the timeval to be set | ||
881 | * | ||
882 | * NOTE: Users should be converted to using get_realtime_clock_ts() | ||
883 | */ | ||
884 | void do_gettimeofday(struct timeval *tv) | ||
885 | { | ||
886 | struct timespec now; | ||
887 | |||
888 | __get_realtime_clock_ts(&now); | ||
889 | tv->tv_sec = now.tv_sec; | ||
890 | tv->tv_usec = now.tv_nsec/1000; | ||
891 | } | ||
892 | |||
893 | EXPORT_SYMBOL(do_gettimeofday); | ||
894 | /** | ||
895 | * do_settimeofday - Sets the time of day | ||
896 | * @tv: pointer to the timespec variable containing the new time | ||
897 | * | ||
898 | * Sets the time of day to the new time and update NTP and notify hrtimers | ||
899 | */ | ||
900 | int do_settimeofday(struct timespec *tv) | ||
901 | { | ||
902 | unsigned long flags; | ||
903 | time_t wtm_sec, sec = tv->tv_sec; | ||
904 | long wtm_nsec, nsec = tv->tv_nsec; | ||
905 | |||
906 | if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) | ||
907 | return -EINVAL; | ||
908 | |||
909 | write_seqlock_irqsave(&xtime_lock, flags); | ||
910 | |||
911 | nsec -= __get_nsec_offset(); | ||
912 | |||
913 | wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec); | ||
914 | wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec); | ||
915 | |||
916 | set_normalized_timespec(&xtime, sec, nsec); | ||
917 | set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec); | ||
918 | |||
919 | clock->error = 0; | ||
920 | ntp_clear(); | ||
921 | |||
922 | update_vsyscall(&xtime, clock); | ||
923 | |||
924 | write_sequnlock_irqrestore(&xtime_lock, flags); | ||
925 | |||
926 | /* signal hrtimers about time change */ | ||
927 | clock_was_set(); | ||
928 | |||
929 | return 0; | ||
930 | } | ||
931 | |||
932 | EXPORT_SYMBOL(do_settimeofday); | ||
933 | |||
934 | /** | ||
935 | * change_clocksource - Swaps clocksources if a new one is available | ||
936 | * | ||
937 | * Accumulates current time interval and initializes new clocksource | ||
938 | */ | ||
939 | static void change_clocksource(void) | ||
940 | { | ||
941 | struct clocksource *new; | ||
942 | cycle_t now; | ||
943 | u64 nsec; | ||
944 | |||
945 | new = clocksource_get_next(); | ||
946 | |||
947 | if (clock == new) | ||
948 | return; | ||
949 | |||
950 | now = clocksource_read(new); | ||
951 | nsec = __get_nsec_offset(); | ||
952 | timespec_add_ns(&xtime, nsec); | ||
953 | |||
954 | clock = new; | ||
955 | clock->cycle_last = now; | ||
956 | |||
957 | clock->error = 0; | ||
958 | clock->xtime_nsec = 0; | ||
959 | clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH); | ||
960 | |||
961 | tick_clock_notify(); | ||
962 | |||
963 | printk(KERN_INFO "Time: %s clocksource has been installed.\n", | ||
964 | clock->name); | ||
965 | } | ||
966 | #else | ||
967 | static inline void change_clocksource(void) { } | ||
968 | #endif | ||
969 | |||
970 | /** | ||
971 | * timekeeping_is_continuous - check to see if timekeeping is free running | ||
972 | */ | ||
973 | int timekeeping_is_continuous(void) | ||
974 | { | ||
975 | unsigned long seq; | ||
976 | int ret; | ||
977 | |||
978 | do { | ||
979 | seq = read_seqbegin(&xtime_lock); | ||
980 | |||
981 | ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; | ||
982 | |||
983 | } while (read_seqretry(&xtime_lock, seq)); | ||
984 | |||
985 | return ret; | ||
986 | } | ||
987 | |||
988 | /** | ||
989 | * read_persistent_clock - Return time in seconds from the persistent clock. | ||
990 | * | ||
991 | * Weak dummy function for arches that do not yet support it. | ||
992 | * Returns seconds from epoch using the battery backed persistent clock. | ||
993 | * Returns zero if unsupported. | ||
994 | * | ||
995 | * XXX - Do be sure to remove it once all arches implement it. | ||
996 | */ | ||
997 | unsigned long __attribute__((weak)) read_persistent_clock(void) | ||
998 | { | ||
999 | return 0; | ||
1000 | } | ||
1001 | |||
1002 | /* | ||
1003 | * timekeeping_init - Initializes the clocksource and common timekeeping values | ||
1004 | */ | ||
1005 | void __init timekeeping_init(void) | ||
1006 | { | ||
1007 | unsigned long flags; | ||
1008 | unsigned long sec = read_persistent_clock(); | ||
1009 | |||
1010 | write_seqlock_irqsave(&xtime_lock, flags); | ||
1011 | |||
1012 | ntp_clear(); | ||
1013 | |||
1014 | clock = clocksource_get_next(); | ||
1015 | clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH); | ||
1016 | clock->cycle_last = clocksource_read(clock); | ||
1017 | |||
1018 | xtime.tv_sec = sec; | ||
1019 | xtime.tv_nsec = 0; | ||
1020 | set_normalized_timespec(&wall_to_monotonic, | ||
1021 | -xtime.tv_sec, -xtime.tv_nsec); | ||
1022 | |||
1023 | write_sequnlock_irqrestore(&xtime_lock, flags); | ||
1024 | } | ||
1025 | |||
1026 | /* flag for if timekeeping is suspended */ | ||
1027 | static int timekeeping_suspended; | ||
1028 | /* time in seconds when suspend began */ | ||
1029 | static unsigned long timekeeping_suspend_time; | ||
1030 | |||
1031 | /** | ||
1032 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | ||
1033 | * @dev: unused | ||
1034 | * | ||
1035 | * This is for the generic clocksource timekeeping. | ||
1036 | * xtime/wall_to_monotonic/jiffies/etc are | ||
1037 | * still managed by arch specific suspend/resume code. | ||
1038 | */ | ||
1039 | static int timekeeping_resume(struct sys_device *dev) | ||
1040 | { | ||
1041 | unsigned long flags; | ||
1042 | unsigned long now = read_persistent_clock(); | ||
1043 | |||
1044 | write_seqlock_irqsave(&xtime_lock, flags); | ||
1045 | |||
1046 | if (now && (now > timekeeping_suspend_time)) { | ||
1047 | unsigned long sleep_length = now - timekeeping_suspend_time; | ||
1048 | |||
1049 | xtime.tv_sec += sleep_length; | ||
1050 | wall_to_monotonic.tv_sec -= sleep_length; | ||
1051 | } | ||
1052 | /* re-base the last cycle value */ | ||
1053 | clock->cycle_last = clocksource_read(clock); | ||
1054 | clock->error = 0; | ||
1055 | timekeeping_suspended = 0; | ||
1056 | write_sequnlock_irqrestore(&xtime_lock, flags); | ||
1057 | |||
1058 | touch_softlockup_watchdog(); | ||
1059 | |||
1060 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | ||
1061 | |||
1062 | /* Resume hrtimers */ | ||
1063 | hres_timers_resume(); | ||
1064 | |||
1065 | return 0; | ||
1066 | } | ||
1067 | |||
1068 | static int timekeeping_suspend(struct sys_device *dev, pm_message_t state) | ||
1069 | { | ||
1070 | unsigned long flags; | ||
1071 | |||
1072 | write_seqlock_irqsave(&xtime_lock, flags); | ||
1073 | timekeeping_suspended = 1; | ||
1074 | timekeeping_suspend_time = read_persistent_clock(); | ||
1075 | write_sequnlock_irqrestore(&xtime_lock, flags); | ||
1076 | |||
1077 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | ||
1078 | |||
1079 | return 0; | ||
1080 | } | ||
1081 | |||
1082 | /* sysfs resume/suspend bits for timekeeping */ | ||
1083 | static struct sysdev_class timekeeping_sysclass = { | ||
1084 | .resume = timekeeping_resume, | ||
1085 | .suspend = timekeeping_suspend, | ||
1086 | set_kset_name("timekeeping"), | ||
1087 | }; | ||
1088 | |||
1089 | static struct sys_device device_timer = { | ||
1090 | .id = 0, | ||
1091 | .cls = &timekeeping_sysclass, | ||
1092 | }; | ||
1093 | |||
1094 | static int __init timekeeping_init_device(void) | ||
1095 | { | ||
1096 | int error = sysdev_class_register(&timekeeping_sysclass); | ||
1097 | if (!error) | ||
1098 | error = sysdev_register(&device_timer); | ||
1099 | return error; | ||
1100 | } | ||
1101 | |||
1102 | device_initcall(timekeeping_init_device); | ||
1103 | |||
1104 | /* | ||
1105 | * If the error is already larger, we look ahead even further | ||
1106 | * to compensate for late or lost adjustments. | ||
1107 | */ | ||
1108 | static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, | ||
1109 | s64 *offset) | ||
1110 | { | ||
1111 | s64 tick_error, i; | ||
1112 | u32 look_ahead, adj; | ||
1113 | s32 error2, mult; | ||
1114 | |||
1115 | /* | ||
1116 | * Use the current error value to determine how much to look ahead. | ||
1117 | * The larger the error the slower we adjust for it to avoid problems | ||
1118 | * with losing too many ticks, otherwise we would overadjust and | ||
1119 | * produce an even larger error. The smaller the adjustment the | ||
1120 | * faster we try to adjust for it, as lost ticks can do less harm | ||
1121 | * here. This is tuned so that an error of about 1 msec is adusted | ||
1122 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). | ||
1123 | */ | ||
1124 | error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ); | ||
1125 | error2 = abs(error2); | ||
1126 | for (look_ahead = 0; error2 > 0; look_ahead++) | ||
1127 | error2 >>= 2; | ||
1128 | |||
1129 | /* | ||
1130 | * Now calculate the error in (1 << look_ahead) ticks, but first | ||
1131 | * remove the single look ahead already included in the error. | ||
1132 | */ | ||
1133 | tick_error = current_tick_length() >> | ||
1134 | (TICK_LENGTH_SHIFT - clock->shift + 1); | ||
1135 | tick_error -= clock->xtime_interval >> 1; | ||
1136 | error = ((error - tick_error) >> look_ahead) + tick_error; | ||
1137 | |||
1138 | /* Finally calculate the adjustment shift value. */ | ||
1139 | i = *interval; | ||
1140 | mult = 1; | ||
1141 | if (error < 0) { | ||
1142 | error = -error; | ||
1143 | *interval = -*interval; | ||
1144 | *offset = -*offset; | ||
1145 | mult = -1; | ||
1146 | } | ||
1147 | for (adj = 0; error > i; adj++) | ||
1148 | error >>= 1; | ||
1149 | |||
1150 | *interval <<= adj; | ||
1151 | *offset <<= adj; | ||
1152 | return mult << adj; | ||
1153 | } | ||
1154 | |||
1155 | /* | ||
1156 | * Adjust the multiplier to reduce the error value, | ||
1157 | * this is optimized for the most common adjustments of -1,0,1, | ||
1158 | * for other values we can do a bit more work. | ||
1159 | */ | ||
1160 | static void clocksource_adjust(struct clocksource *clock, s64 offset) | ||
1161 | { | ||
1162 | s64 error, interval = clock->cycle_interval; | ||
1163 | int adj; | ||
1164 | |||
1165 | error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1); | ||
1166 | if (error > interval) { | ||
1167 | error >>= 2; | ||
1168 | if (likely(error <= interval)) | ||
1169 | adj = 1; | ||
1170 | else | ||
1171 | adj = clocksource_bigadjust(error, &interval, &offset); | ||
1172 | } else if (error < -interval) { | ||
1173 | error >>= 2; | ||
1174 | if (likely(error >= -interval)) { | ||
1175 | adj = -1; | ||
1176 | interval = -interval; | ||
1177 | offset = -offset; | ||
1178 | } else | ||
1179 | adj = clocksource_bigadjust(error, &interval, &offset); | ||
1180 | } else | ||
1181 | return; | ||
1182 | |||
1183 | clock->mult += adj; | ||
1184 | clock->xtime_interval += interval; | ||
1185 | clock->xtime_nsec -= offset; | ||
1186 | clock->error -= (interval - offset) << | ||
1187 | (TICK_LENGTH_SHIFT - clock->shift); | ||
1188 | } | ||
1189 | |||
1190 | /** | ||
1191 | * update_wall_time - Uses the current clocksource to increment the wall time | ||
1192 | * | ||
1193 | * Called from the timer interrupt, must hold a write on xtime_lock. | ||
1194 | */ | ||
1195 | static void update_wall_time(void) | ||
1196 | { | ||
1197 | cycle_t offset; | ||
1198 | |||
1199 | /* Make sure we're fully resumed: */ | ||
1200 | if (unlikely(timekeeping_suspended)) | ||
1201 | return; | ||
1202 | |||
1203 | #ifdef CONFIG_GENERIC_TIME | ||
1204 | offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask; | ||
1205 | #else | ||
1206 | offset = clock->cycle_interval; | ||
1207 | #endif | ||
1208 | clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift; | ||
1209 | |||
1210 | /* normally this loop will run just once, however in the | ||
1211 | * case of lost or late ticks, it will accumulate correctly. | ||
1212 | */ | ||
1213 | while (offset >= clock->cycle_interval) { | ||
1214 | /* accumulate one interval */ | ||
1215 | clock->xtime_nsec += clock->xtime_interval; | ||
1216 | clock->cycle_last += clock->cycle_interval; | ||
1217 | offset -= clock->cycle_interval; | ||
1218 | |||
1219 | if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) { | ||
1220 | clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift; | ||
1221 | xtime.tv_sec++; | ||
1222 | second_overflow(); | ||
1223 | } | ||
1224 | |||
1225 | /* interpolator bits */ | ||
1226 | time_interpolator_update(clock->xtime_interval | ||
1227 | >> clock->shift); | ||
1228 | |||
1229 | /* accumulate error between NTP and clock interval */ | ||
1230 | clock->error += current_tick_length(); | ||
1231 | clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift); | ||
1232 | } | ||
1233 | |||
1234 | /* correct the clock when NTP error is too big */ | ||
1235 | clocksource_adjust(clock, offset); | ||
1236 | |||
1237 | /* store full nanoseconds into xtime */ | ||
1238 | xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift; | ||
1239 | clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift; | ||
1240 | |||
1241 | /* check to see if there is a new clocksource to use */ | ||
1242 | change_clocksource(); | ||
1243 | update_vsyscall(&xtime, clock); | ||
1244 | } | ||
1245 | |||
1246 | /* | 797 | /* |
1247 | * Called from the timer interrupt handler to charge one tick to the current | 798 | * Called from the timer interrupt handler to charge one tick to the current |
1248 | * process. user_tick is 1 if the tick is user time, 0 for system. | 799 | * process. user_tick is 1 if the tick is user time, 0 for system. |
@@ -1306,14 +857,6 @@ static inline void calc_load(unsigned long ticks) | |||
1306 | } | 857 | } |
1307 | 858 | ||
1308 | /* | 859 | /* |
1309 | * This read-write spinlock protects us from races in SMP while | ||
1310 | * playing with xtime and avenrun. | ||
1311 | */ | ||
1312 | __attribute__((weak)) __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); | ||
1313 | |||
1314 | EXPORT_SYMBOL(xtime_lock); | ||
1315 | |||
1316 | /* | ||
1317 | * This function runs timers and the timer-tq in bottom half context. | 860 | * This function runs timers and the timer-tq in bottom half context. |
1318 | */ | 861 | */ |
1319 | static void run_timer_softirq(struct softirq_action *h) | 862 | static void run_timer_softirq(struct softirq_action *h) |