2 * Time of day based timer functions.
5 * Copyright IBM Corp. 1999, 2008
6 * Author(s): Hartmut Penner (hp@de.ibm.com),
7 * Martin Schwidefsky (schwidefsky@de.ibm.com),
8 * Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com)
10 * Derived from "arch/i386/kernel/time.c"
11 * Copyright (C) 1991, 1992, 1995 Linus Torvalds
14 #define KMSG_COMPONENT "time"
15 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
17 #include <linux/kernel_stat.h>
18 #include <linux/errno.h>
19 #include <linux/export.h>
20 #include <linux/sched.h>
21 #include <linux/sched/clock.h>
22 #include <linux/kernel.h>
23 #include <linux/param.h>
24 #include <linux/string.h>
26 #include <linux/interrupt.h>
27 #include <linux/cpu.h>
28 #include <linux/stop_machine.h>
29 #include <linux/time.h>
30 #include <linux/device.h>
31 #include <linux/delay.h>
32 #include <linux/init.h>
33 #include <linux/smp.h>
34 #include <linux/types.h>
35 #include <linux/profile.h>
36 #include <linux/timex.h>
37 #include <linux/notifier.h>
38 #include <linux/timekeeper_internal.h>
39 #include <linux/clockchips.h>
40 #include <linux/gfp.h>
41 #include <linux/kprobes.h>
42 #include <linux/uaccess.h>
43 #include <asm/facility.h>
44 #include <asm/delay.h>
45 #include <asm/div64.h>
48 #include <asm/irq_regs.h>
49 #include <asm/vtimer.h>
54 u64 sched_clock_base_cc = -1; /* Force to data section. */
55 EXPORT_SYMBOL_GPL(sched_clock_base_cc);
57 static DEFINE_PER_CPU(struct clock_event_device, comparators);
59 ATOMIC_NOTIFIER_HEAD(s390_epoch_delta_notifier);
60 EXPORT_SYMBOL(s390_epoch_delta_notifier);
62 unsigned char ptff_function_mask[16];
64 static unsigned long long lpar_offset;
65 static unsigned long long initial_leap_seconds;
66 static unsigned long long tod_steering_end;
67 static long long tod_steering_delta;
70 * Get time offsets with PTFF
72 void __init time_early_init(void)
77 /* Initialize TOD steering parameters */
78 tod_steering_end = sched_clock_base_cc;
79 vdso_data->ts_end = tod_steering_end;
81 if (!test_facility(28))
84 ptff(&ptff_function_mask, sizeof(ptff_function_mask), PTFF_QAF);
87 if (ptff_query(PTFF_QTO) && ptff(&qto, sizeof(qto), PTFF_QTO) == 0)
88 lpar_offset = qto.tod_epoch_difference;
90 /* get initial leap seconds */
91 if (ptff_query(PTFF_QUI) && ptff(&qui, sizeof(qui), PTFF_QUI) == 0)
92 initial_leap_seconds = (unsigned long long)
93 ((long) qui.old_leap * 4096000000L);
97 * Scheduler clock - returns current time in nanosec units.
99 unsigned long long notrace sched_clock(void)
101 return tod_to_ns(get_tod_clock_monotonic());
103 NOKPROBE_SYMBOL(sched_clock);
106 * Monotonic_clock - returns # of nanoseconds passed since time_init()
108 unsigned long long monotonic_clock(void)
110 return sched_clock();
112 EXPORT_SYMBOL(monotonic_clock);
114 static void tod_to_timeval(__u64 todval, struct timespec64 *xt)
116 unsigned long long sec;
119 do_div(sec, 1000000);
121 todval -= (sec * 1000000) << 12;
122 xt->tv_nsec = ((todval * 1000) >> 12);
125 void clock_comparator_work(void)
127 struct clock_event_device *cd;
129 S390_lowcore.clock_comparator = -1ULL;
130 cd = this_cpu_ptr(&comparators);
131 cd->event_handler(cd);
134 static int s390_next_event(unsigned long delta,
135 struct clock_event_device *evt)
137 S390_lowcore.clock_comparator = get_tod_clock() + delta;
138 set_clock_comparator(S390_lowcore.clock_comparator);
143 * Set up lowcore and control register of the current cpu to
144 * enable TOD clock and clock comparator interrupts.
146 void init_cpu_timer(void)
148 struct clock_event_device *cd;
151 S390_lowcore.clock_comparator = -1ULL;
152 set_clock_comparator(S390_lowcore.clock_comparator);
154 cpu = smp_processor_id();
155 cd = &per_cpu(comparators, cpu);
156 cd->name = "comparator";
157 cd->features = CLOCK_EVT_FEAT_ONESHOT;
160 cd->min_delta_ns = 1;
161 cd->min_delta_ticks = 1;
162 cd->max_delta_ns = LONG_MAX;
163 cd->max_delta_ticks = ULONG_MAX;
165 cd->cpumask = cpumask_of(cpu);
166 cd->set_next_event = s390_next_event;
168 clockevents_register_device(cd);
170 /* Enable clock comparator timer interrupt. */
173 /* Always allow the timing alert external interrupt. */
177 static void clock_comparator_interrupt(struct ext_code ext_code,
178 unsigned int param32,
179 unsigned long param64)
181 inc_irq_stat(IRQEXT_CLK);
182 if (S390_lowcore.clock_comparator == -1ULL)
183 set_clock_comparator(S390_lowcore.clock_comparator);
186 static void stp_timing_alert(struct stp_irq_parm *);
188 static void timing_alert_interrupt(struct ext_code ext_code,
189 unsigned int param32, unsigned long param64)
191 inc_irq_stat(IRQEXT_TLA);
192 if (param32 & 0x00038000)
193 stp_timing_alert((struct stp_irq_parm *) ¶m32);
196 static void stp_reset(void);
198 void read_persistent_clock64(struct timespec64 *ts)
202 clock = get_tod_clock() - initial_leap_seconds;
203 tod_to_timeval(clock - TOD_UNIX_EPOCH, ts);
206 void read_boot_clock64(struct timespec64 *ts)
210 clock = sched_clock_base_cc - initial_leap_seconds;
211 tod_to_timeval(clock - TOD_UNIX_EPOCH, ts);
214 static u64 read_tod_clock(struct clocksource *cs)
216 unsigned long long now, adj;
218 preempt_disable(); /* protect from changes to steering parameters */
219 now = get_tod_clock();
220 adj = tod_steering_end - now;
221 if (unlikely((s64) adj >= 0))
223 * manually steer by 1 cycle every 2^16 cycles. This
224 * corresponds to shifting the tod delta by 15. 1s is
225 * therefore steered in ~9h. The adjust will decrease
226 * over time, until it finally reaches 0.
228 now += (tod_steering_delta < 0) ? (adj >> 15) : -(adj >> 15);
233 static struct clocksource clocksource_tod = {
236 .read = read_tod_clock,
240 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
243 struct clocksource * __init clocksource_default_clock(void)
245 return &clocksource_tod;
248 void update_vsyscall(struct timekeeper *tk)
252 if (tk->tkr_mono.clock != &clocksource_tod)
255 /* Make userspace gettimeofday spin until we're done. */
256 ++vdso_data->tb_update_count;
258 vdso_data->xtime_tod_stamp = tk->tkr_mono.cycle_last;
259 vdso_data->xtime_clock_sec = tk->xtime_sec;
260 vdso_data->xtime_clock_nsec = tk->tkr_mono.xtime_nsec;
261 vdso_data->wtom_clock_sec =
262 tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
263 vdso_data->wtom_clock_nsec = tk->tkr_mono.xtime_nsec +
264 + ((u64) tk->wall_to_monotonic.tv_nsec << tk->tkr_mono.shift);
265 nsecps = (u64) NSEC_PER_SEC << tk->tkr_mono.shift;
266 while (vdso_data->wtom_clock_nsec >= nsecps) {
267 vdso_data->wtom_clock_nsec -= nsecps;
268 vdso_data->wtom_clock_sec++;
271 vdso_data->xtime_coarse_sec = tk->xtime_sec;
272 vdso_data->xtime_coarse_nsec =
273 (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
274 vdso_data->wtom_coarse_sec =
275 vdso_data->xtime_coarse_sec + tk->wall_to_monotonic.tv_sec;
276 vdso_data->wtom_coarse_nsec =
277 vdso_data->xtime_coarse_nsec + tk->wall_to_monotonic.tv_nsec;
278 while (vdso_data->wtom_coarse_nsec >= NSEC_PER_SEC) {
279 vdso_data->wtom_coarse_nsec -= NSEC_PER_SEC;
280 vdso_data->wtom_coarse_sec++;
283 vdso_data->tk_mult = tk->tkr_mono.mult;
284 vdso_data->tk_shift = tk->tkr_mono.shift;
286 ++vdso_data->tb_update_count;
289 extern struct timezone sys_tz;
291 void update_vsyscall_tz(void)
293 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
294 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
298 * Initialize the TOD clock and the CPU timer of
301 void __init time_init(void)
303 /* Reset time synchronization interfaces. */
306 /* request the clock comparator external interrupt */
307 if (register_external_irq(EXT_IRQ_CLK_COMP, clock_comparator_interrupt))
308 panic("Couldn't request external interrupt 0x1004");
310 /* request the timing alert external interrupt */
311 if (register_external_irq(EXT_IRQ_TIMING_ALERT, timing_alert_interrupt))
312 panic("Couldn't request external interrupt 0x1406");
314 if (__clocksource_register(&clocksource_tod) != 0)
315 panic("Could not register TOD clock source");
317 /* Enable TOD clock interrupts on the boot cpu. */
320 /* Enable cpu timer interrupts on the boot cpu. */
324 static DEFINE_PER_CPU(atomic_t, clock_sync_word);
325 static DEFINE_MUTEX(clock_sync_mutex);
326 static unsigned long clock_sync_flags;
328 #define CLOCK_SYNC_HAS_STP 0
329 #define CLOCK_SYNC_STP 1
332 * The get_clock function for the physical clock. It will get the current
333 * TOD clock, subtract the LPAR offset and write the result to *clock.
334 * The function returns 0 if the clock is in sync with the external time
335 * source. If the clock mode is local it will return -EOPNOTSUPP and
336 * -EAGAIN if the clock is not in sync with the external reference.
338 int get_phys_clock(unsigned long long *clock)
341 unsigned int sw0, sw1;
343 sw_ptr = &get_cpu_var(clock_sync_word);
344 sw0 = atomic_read(sw_ptr);
345 *clock = get_tod_clock() - lpar_offset;
346 sw1 = atomic_read(sw_ptr);
347 put_cpu_var(clock_sync_word);
348 if (sw0 == sw1 && (sw0 & 0x80000000U))
349 /* Success: time is in sync. */
351 if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
353 if (!test_bit(CLOCK_SYNC_STP, &clock_sync_flags))
357 EXPORT_SYMBOL(get_phys_clock);
360 * Make get_phys_clock() return -EAGAIN.
362 static void disable_sync_clock(void *dummy)
364 atomic_t *sw_ptr = this_cpu_ptr(&clock_sync_word);
366 * Clear the in-sync bit 2^31. All get_phys_clock calls will
367 * fail until the sync bit is turned back on. In addition
368 * increase the "sequence" counter to avoid the race of an
369 * stp event and the complete recovery against get_phys_clock.
371 atomic_andnot(0x80000000, sw_ptr);
376 * Make get_phys_clock() return 0 again.
377 * Needs to be called from a context disabled for preemption.
379 static void enable_sync_clock(void)
381 atomic_t *sw_ptr = this_cpu_ptr(&clock_sync_word);
382 atomic_or(0x80000000, sw_ptr);
386 * Function to check if the clock is in sync.
388 static inline int check_sync_clock(void)
393 sw_ptr = &get_cpu_var(clock_sync_word);
394 rc = (atomic_read(sw_ptr) & 0x80000000U) != 0;
395 put_cpu_var(clock_sync_word);
400 * Apply clock delta to the global data structures.
401 * This is called once on the CPU that performed the clock sync.
403 static void clock_sync_global(unsigned long long delta)
405 unsigned long now, adj;
408 /* Fixup the monotonic sched clock. */
409 sched_clock_base_cc += delta;
410 /* Adjust TOD steering parameters. */
411 vdso_data->tb_update_count++;
412 now = get_tod_clock();
413 adj = tod_steering_end - now;
414 if (unlikely((s64) adj >= 0))
415 /* Calculate how much of the old adjustment is left. */
416 tod_steering_delta = (tod_steering_delta < 0) ?
417 -(adj >> 15) : (adj >> 15);
418 tod_steering_delta += delta;
419 if ((abs(tod_steering_delta) >> 48) != 0)
420 panic("TOD clock sync offset %lli is too large to drift\n",
422 tod_steering_end = now + (abs(tod_steering_delta) << 15);
423 vdso_data->ts_dir = (tod_steering_delta < 0) ? 0 : 1;
424 vdso_data->ts_end = tod_steering_end;
425 vdso_data->tb_update_count++;
426 /* Update LPAR offset. */
427 if (ptff_query(PTFF_QTO) && ptff(&qto, sizeof(qto), PTFF_QTO) == 0)
428 lpar_offset = qto.tod_epoch_difference;
429 /* Call the TOD clock change notifier. */
430 atomic_notifier_call_chain(&s390_epoch_delta_notifier, 0, &delta);
434 * Apply clock delta to the per-CPU data structures of this CPU.
435 * This is called for each online CPU after the call to clock_sync_global.
437 static void clock_sync_local(unsigned long long delta)
439 /* Add the delta to the clock comparator. */
440 if (S390_lowcore.clock_comparator != -1ULL) {
441 S390_lowcore.clock_comparator += delta;
442 set_clock_comparator(S390_lowcore.clock_comparator);
444 /* Adjust the last_update_clock time-stamp. */
445 S390_lowcore.last_update_clock += delta;
448 /* Single threaded workqueue used for stp sync events */
449 static struct workqueue_struct *time_sync_wq;
451 static void __init time_init_wq(void)
455 time_sync_wq = create_singlethread_workqueue("timesync");
458 struct clock_sync_data {
461 unsigned long long clock_delta;
465 * Server Time Protocol (STP) code.
467 static bool stp_online;
468 static struct stp_sstpi stp_info;
469 static void *stp_page;
471 static void stp_work_fn(struct work_struct *work);
472 static DEFINE_MUTEX(stp_work_mutex);
473 static DECLARE_WORK(stp_work, stp_work_fn);
474 static struct timer_list stp_timer;
476 static int __init early_parse_stp(char *p)
478 return kstrtobool(p, &stp_online);
480 early_param("stp", early_parse_stp);
483 * Reset STP attachment.
485 static void __init stp_reset(void)
489 stp_page = (void *) get_zeroed_page(GFP_ATOMIC);
490 rc = chsc_sstpc(stp_page, STP_OP_CTRL, 0x0000, NULL);
492 set_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags);
493 else if (stp_online) {
494 pr_warn("The real or virtual hardware system does not provide an STP interface\n");
495 free_page((unsigned long) stp_page);
501 static void stp_timeout(unsigned long dummy)
503 queue_work(time_sync_wq, &stp_work);
506 static int __init stp_init(void)
508 if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
510 setup_timer(&stp_timer, stp_timeout, 0UL);
514 queue_work(time_sync_wq, &stp_work);
518 arch_initcall(stp_init);
521 * STP timing alert. There are three causes:
522 * 1) timing status change
523 * 2) link availability change
524 * 3) time control parameter change
525 * In all three cases we are only interested in the clock source state.
526 * If a STP clock source is now available use it.
528 static void stp_timing_alert(struct stp_irq_parm *intparm)
530 if (intparm->tsc || intparm->lac || intparm->tcpc)
531 queue_work(time_sync_wq, &stp_work);
535 * STP sync check machine check. This is called when the timing state
536 * changes from the synchronized state to the unsynchronized state.
537 * After a STP sync check the clock is not in sync. The machine check
538 * is broadcasted to all cpus at the same time.
540 int stp_sync_check(void)
542 disable_sync_clock(NULL);
547 * STP island condition machine check. This is called when an attached
548 * server attempts to communicate over an STP link and the servers
549 * have matching CTN ids and have a valid stratum-1 configuration
550 * but the configurations do not match.
552 int stp_island_check(void)
554 disable_sync_clock(NULL);
558 void stp_queue_work(void)
560 queue_work(time_sync_wq, &stp_work);
563 static int stp_sync_clock(void *data)
565 struct clock_sync_data *sync = data;
566 unsigned long long clock_delta;
571 if (xchg(&first, 1) == 0) {
572 /* Wait until all other cpus entered the sync function. */
573 while (atomic_read(&sync->cpus) != 0)
576 if (stp_info.todoff[0] || stp_info.todoff[1] ||
577 stp_info.todoff[2] || stp_info.todoff[3] ||
579 rc = chsc_sstpc(stp_page, STP_OP_SYNC, 0,
582 sync->clock_delta = clock_delta;
583 clock_sync_global(clock_delta);
584 rc = chsc_sstpi(stp_page, &stp_info,
585 sizeof(struct stp_sstpi));
586 if (rc == 0 && stp_info.tmd != 2)
590 sync->in_sync = rc ? -EAGAIN : 1;
594 atomic_dec(&sync->cpus);
595 /* Wait for in_sync to be set. */
596 while (READ_ONCE(sync->in_sync) == 0)
599 if (sync->in_sync != 1)
600 /* Didn't work. Clear per-cpu in sync bit again. */
601 disable_sync_clock(NULL);
602 /* Apply clock delta to per-CPU fields of this CPU. */
603 clock_sync_local(sync->clock_delta);
609 * STP work. Check for the STP state and take over the clock
610 * synchronization if the STP clock source is usable.
612 static void stp_work_fn(struct work_struct *work)
614 struct clock_sync_data stp_sync;
617 /* prevent multiple execution. */
618 mutex_lock(&stp_work_mutex);
621 chsc_sstpc(stp_page, STP_OP_CTRL, 0x0000, NULL);
622 del_timer_sync(&stp_timer);
626 rc = chsc_sstpc(stp_page, STP_OP_CTRL, 0xb0e0, NULL);
630 rc = chsc_sstpi(stp_page, &stp_info, sizeof(struct stp_sstpi));
631 if (rc || stp_info.c == 0)
634 /* Skip synchronization if the clock is already in sync. */
635 if (check_sync_clock())
638 memset(&stp_sync, 0, sizeof(stp_sync));
640 atomic_set(&stp_sync.cpus, num_online_cpus() - 1);
641 stop_machine_cpuslocked(stp_sync_clock, &stp_sync, cpu_online_mask);
644 if (!check_sync_clock())
646 * There is a usable clock but the synchonization failed.
647 * Retry after a second.
649 mod_timer(&stp_timer, jiffies + HZ);
652 mutex_unlock(&stp_work_mutex);
656 * STP subsys sysfs interface functions
658 static struct bus_type stp_subsys = {
663 static ssize_t stp_ctn_id_show(struct device *dev,
664 struct device_attribute *attr,
669 return sprintf(buf, "%016llx\n",
670 *(unsigned long long *) stp_info.ctnid);
673 static DEVICE_ATTR(ctn_id, 0400, stp_ctn_id_show, NULL);
675 static ssize_t stp_ctn_type_show(struct device *dev,
676 struct device_attribute *attr,
681 return sprintf(buf, "%i\n", stp_info.ctn);
684 static DEVICE_ATTR(ctn_type, 0400, stp_ctn_type_show, NULL);
686 static ssize_t stp_dst_offset_show(struct device *dev,
687 struct device_attribute *attr,
690 if (!stp_online || !(stp_info.vbits & 0x2000))
692 return sprintf(buf, "%i\n", (int)(s16) stp_info.dsto);
695 static DEVICE_ATTR(dst_offset, 0400, stp_dst_offset_show, NULL);
697 static ssize_t stp_leap_seconds_show(struct device *dev,
698 struct device_attribute *attr,
701 if (!stp_online || !(stp_info.vbits & 0x8000))
703 return sprintf(buf, "%i\n", (int)(s16) stp_info.leaps);
706 static DEVICE_ATTR(leap_seconds, 0400, stp_leap_seconds_show, NULL);
708 static ssize_t stp_stratum_show(struct device *dev,
709 struct device_attribute *attr,
714 return sprintf(buf, "%i\n", (int)(s16) stp_info.stratum);
717 static DEVICE_ATTR(stratum, 0400, stp_stratum_show, NULL);
719 static ssize_t stp_time_offset_show(struct device *dev,
720 struct device_attribute *attr,
723 if (!stp_online || !(stp_info.vbits & 0x0800))
725 return sprintf(buf, "%i\n", (int) stp_info.tto);
728 static DEVICE_ATTR(time_offset, 0400, stp_time_offset_show, NULL);
730 static ssize_t stp_time_zone_offset_show(struct device *dev,
731 struct device_attribute *attr,
734 if (!stp_online || !(stp_info.vbits & 0x4000))
736 return sprintf(buf, "%i\n", (int)(s16) stp_info.tzo);
739 static DEVICE_ATTR(time_zone_offset, 0400,
740 stp_time_zone_offset_show, NULL);
742 static ssize_t stp_timing_mode_show(struct device *dev,
743 struct device_attribute *attr,
748 return sprintf(buf, "%i\n", stp_info.tmd);
751 static DEVICE_ATTR(timing_mode, 0400, stp_timing_mode_show, NULL);
753 static ssize_t stp_timing_state_show(struct device *dev,
754 struct device_attribute *attr,
759 return sprintf(buf, "%i\n", stp_info.tst);
762 static DEVICE_ATTR(timing_state, 0400, stp_timing_state_show, NULL);
764 static ssize_t stp_online_show(struct device *dev,
765 struct device_attribute *attr,
768 return sprintf(buf, "%i\n", stp_online);
771 static ssize_t stp_online_store(struct device *dev,
772 struct device_attribute *attr,
773 const char *buf, size_t count)
777 value = simple_strtoul(buf, NULL, 0);
778 if (value != 0 && value != 1)
780 if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
782 mutex_lock(&clock_sync_mutex);
785 set_bit(CLOCK_SYNC_STP, &clock_sync_flags);
787 clear_bit(CLOCK_SYNC_STP, &clock_sync_flags);
788 queue_work(time_sync_wq, &stp_work);
789 mutex_unlock(&clock_sync_mutex);
794 * Can't use DEVICE_ATTR because the attribute should be named
795 * stp/online but dev_attr_online already exists in this file ..
797 static struct device_attribute dev_attr_stp_online = {
798 .attr = { .name = "online", .mode = 0600 },
799 .show = stp_online_show,
800 .store = stp_online_store,
803 static struct device_attribute *stp_attributes[] = {
806 &dev_attr_dst_offset,
807 &dev_attr_leap_seconds,
808 &dev_attr_stp_online,
810 &dev_attr_time_offset,
811 &dev_attr_time_zone_offset,
812 &dev_attr_timing_mode,
813 &dev_attr_timing_state,
817 static int __init stp_init_sysfs(void)
819 struct device_attribute **attr;
822 rc = subsys_system_register(&stp_subsys, NULL);
825 for (attr = stp_attributes; *attr; attr++) {
826 rc = device_create_file(stp_subsys.dev_root, *attr);
832 for (; attr >= stp_attributes; attr--)
833 device_remove_file(stp_subsys.dev_root, *attr);
834 bus_unregister(&stp_subsys);
839 device_initcall(stp_init_sysfs);