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/kernel.h>
22 #include <linux/param.h>
23 #include <linux/string.h>
25 #include <linux/interrupt.h>
26 #include <linux/cpu.h>
27 #include <linux/stop_machine.h>
28 #include <linux/time.h>
29 #include <linux/device.h>
30 #include <linux/delay.h>
31 #include <linux/init.h>
32 #include <linux/smp.h>
33 #include <linux/types.h>
34 #include <linux/profile.h>
35 #include <linux/timex.h>
36 #include <linux/notifier.h>
37 #include <linux/timekeeper_internal.h>
38 #include <linux/clockchips.h>
39 #include <linux/gfp.h>
40 #include <linux/kprobes.h>
41 #include <linux/uaccess.h>
42 #include <asm/facility.h>
43 #include <asm/delay.h>
44 #include <asm/div64.h>
47 #include <asm/irq_regs.h>
48 #include <asm/vtimer.h>
53 u64 sched_clock_base_cc = -1; /* Force to data section. */
54 EXPORT_SYMBOL_GPL(sched_clock_base_cc);
56 static DEFINE_PER_CPU(struct clock_event_device, comparators);
58 ATOMIC_NOTIFIER_HEAD(s390_epoch_delta_notifier);
59 EXPORT_SYMBOL(s390_epoch_delta_notifier);
61 unsigned char ptff_function_mask[16];
63 static unsigned long long lpar_offset;
64 static unsigned long long initial_leap_seconds;
65 static unsigned long long tod_steering_end;
66 static long long tod_steering_delta;
69 * Get time offsets with PTFF
71 void __init time_early_init(void)
76 /* Initialize TOD steering parameters */
77 tod_steering_end = sched_clock_base_cc;
78 vdso_data->ts_end = tod_steering_end;
80 if (!test_facility(28))
83 ptff(&ptff_function_mask, sizeof(ptff_function_mask), PTFF_QAF);
86 if (ptff_query(PTFF_QTO) && ptff(&qto, sizeof(qto), PTFF_QTO) == 0)
87 lpar_offset = qto.tod_epoch_difference;
89 /* get initial leap seconds */
90 if (ptff_query(PTFF_QUI) && ptff(&qui, sizeof(qui), PTFF_QUI) == 0)
91 initial_leap_seconds = (unsigned long long)
92 ((long) qui.old_leap * 4096000000L);
96 * Scheduler clock - returns current time in nanosec units.
98 unsigned long long notrace sched_clock(void)
100 return tod_to_ns(get_tod_clock_monotonic());
102 NOKPROBE_SYMBOL(sched_clock);
105 * Monotonic_clock - returns # of nanoseconds passed since time_init()
107 unsigned long long monotonic_clock(void)
109 return sched_clock();
111 EXPORT_SYMBOL(monotonic_clock);
113 static void tod_to_timeval(__u64 todval, struct timespec64 *xt)
115 unsigned long long sec;
118 do_div(sec, 1000000);
120 todval -= (sec * 1000000) << 12;
121 xt->tv_nsec = ((todval * 1000) >> 12);
124 void clock_comparator_work(void)
126 struct clock_event_device *cd;
128 S390_lowcore.clock_comparator = -1ULL;
129 cd = this_cpu_ptr(&comparators);
130 cd->event_handler(cd);
133 static int s390_next_event(unsigned long delta,
134 struct clock_event_device *evt)
136 S390_lowcore.clock_comparator = get_tod_clock() + delta;
137 set_clock_comparator(S390_lowcore.clock_comparator);
142 * Set up lowcore and control register of the current cpu to
143 * enable TOD clock and clock comparator interrupts.
145 void init_cpu_timer(void)
147 struct clock_event_device *cd;
150 S390_lowcore.clock_comparator = -1ULL;
151 set_clock_comparator(S390_lowcore.clock_comparator);
153 cpu = smp_processor_id();
154 cd = &per_cpu(comparators, cpu);
155 cd->name = "comparator";
156 cd->features = CLOCK_EVT_FEAT_ONESHOT;
159 cd->min_delta_ns = 1;
160 cd->max_delta_ns = LONG_MAX;
162 cd->cpumask = cpumask_of(cpu);
163 cd->set_next_event = s390_next_event;
165 clockevents_register_device(cd);
167 /* Enable clock comparator timer interrupt. */
170 /* Always allow the timing alert external interrupt. */
174 static void clock_comparator_interrupt(struct ext_code ext_code,
175 unsigned int param32,
176 unsigned long param64)
178 inc_irq_stat(IRQEXT_CLK);
179 if (S390_lowcore.clock_comparator == -1ULL)
180 set_clock_comparator(S390_lowcore.clock_comparator);
183 static void stp_timing_alert(struct stp_irq_parm *);
185 static void timing_alert_interrupt(struct ext_code ext_code,
186 unsigned int param32, unsigned long param64)
188 inc_irq_stat(IRQEXT_TLA);
189 if (param32 & 0x00038000)
190 stp_timing_alert((struct stp_irq_parm *) ¶m32);
193 static void stp_reset(void);
195 void read_persistent_clock64(struct timespec64 *ts)
199 clock = get_tod_clock() - initial_leap_seconds;
200 tod_to_timeval(clock - TOD_UNIX_EPOCH, ts);
203 void read_boot_clock64(struct timespec64 *ts)
207 clock = sched_clock_base_cc - initial_leap_seconds;
208 tod_to_timeval(clock - TOD_UNIX_EPOCH, ts);
211 static u64 read_tod_clock(struct clocksource *cs)
213 unsigned long long now, adj;
215 preempt_disable(); /* protect from changes to steering parameters */
216 now = get_tod_clock();
217 adj = tod_steering_end - now;
218 if (unlikely((s64) adj >= 0))
220 * manually steer by 1 cycle every 2^16 cycles. This
221 * corresponds to shifting the tod delta by 15. 1s is
222 * therefore steered in ~9h. The adjust will decrease
223 * over time, until it finally reaches 0.
225 now += (tod_steering_delta < 0) ? (adj >> 15) : -(adj >> 15);
230 static struct clocksource clocksource_tod = {
233 .read = read_tod_clock,
237 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
240 struct clocksource * __init clocksource_default_clock(void)
242 return &clocksource_tod;
245 void update_vsyscall(struct timekeeper *tk)
249 if (tk->tkr_mono.clock != &clocksource_tod)
252 /* Make userspace gettimeofday spin until we're done. */
253 ++vdso_data->tb_update_count;
255 vdso_data->xtime_tod_stamp = tk->tkr_mono.cycle_last;
256 vdso_data->xtime_clock_sec = tk->xtime_sec;
257 vdso_data->xtime_clock_nsec = tk->tkr_mono.xtime_nsec;
258 vdso_data->wtom_clock_sec =
259 tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
260 vdso_data->wtom_clock_nsec = tk->tkr_mono.xtime_nsec +
261 + ((u64) tk->wall_to_monotonic.tv_nsec << tk->tkr_mono.shift);
262 nsecps = (u64) NSEC_PER_SEC << tk->tkr_mono.shift;
263 while (vdso_data->wtom_clock_nsec >= nsecps) {
264 vdso_data->wtom_clock_nsec -= nsecps;
265 vdso_data->wtom_clock_sec++;
268 vdso_data->xtime_coarse_sec = tk->xtime_sec;
269 vdso_data->xtime_coarse_nsec =
270 (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
271 vdso_data->wtom_coarse_sec =
272 vdso_data->xtime_coarse_sec + tk->wall_to_monotonic.tv_sec;
273 vdso_data->wtom_coarse_nsec =
274 vdso_data->xtime_coarse_nsec + tk->wall_to_monotonic.tv_nsec;
275 while (vdso_data->wtom_coarse_nsec >= NSEC_PER_SEC) {
276 vdso_data->wtom_coarse_nsec -= NSEC_PER_SEC;
277 vdso_data->wtom_coarse_sec++;
280 vdso_data->tk_mult = tk->tkr_mono.mult;
281 vdso_data->tk_shift = tk->tkr_mono.shift;
283 ++vdso_data->tb_update_count;
286 extern struct timezone sys_tz;
288 void update_vsyscall_tz(void)
290 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest;
291 vdso_data->tz_dsttime = sys_tz.tz_dsttime;
295 * Initialize the TOD clock and the CPU timer of
298 void __init time_init(void)
300 /* Reset time synchronization interfaces. */
303 /* request the clock comparator external interrupt */
304 if (register_external_irq(EXT_IRQ_CLK_COMP, clock_comparator_interrupt))
305 panic("Couldn't request external interrupt 0x1004");
307 /* request the timing alert external interrupt */
308 if (register_external_irq(EXT_IRQ_TIMING_ALERT, timing_alert_interrupt))
309 panic("Couldn't request external interrupt 0x1406");
311 if (__clocksource_register(&clocksource_tod) != 0)
312 panic("Could not register TOD clock source");
314 /* Enable TOD clock interrupts on the boot cpu. */
317 /* Enable cpu timer interrupts on the boot cpu. */
321 static DEFINE_PER_CPU(atomic_t, clock_sync_word);
322 static DEFINE_MUTEX(clock_sync_mutex);
323 static unsigned long clock_sync_flags;
325 #define CLOCK_SYNC_HAS_STP 0
326 #define CLOCK_SYNC_STP 1
329 * The get_clock function for the physical clock. It will get the current
330 * TOD clock, subtract the LPAR offset and write the result to *clock.
331 * The function returns 0 if the clock is in sync with the external time
332 * source. If the clock mode is local it will return -EOPNOTSUPP and
333 * -EAGAIN if the clock is not in sync with the external reference.
335 int get_phys_clock(unsigned long long *clock)
338 unsigned int sw0, sw1;
340 sw_ptr = &get_cpu_var(clock_sync_word);
341 sw0 = atomic_read(sw_ptr);
342 *clock = get_tod_clock() - lpar_offset;
343 sw1 = atomic_read(sw_ptr);
344 put_cpu_var(clock_sync_word);
345 if (sw0 == sw1 && (sw0 & 0x80000000U))
346 /* Success: time is in sync. */
348 if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
350 if (!test_bit(CLOCK_SYNC_STP, &clock_sync_flags))
354 EXPORT_SYMBOL(get_phys_clock);
357 * Make get_phys_clock() return -EAGAIN.
359 static void disable_sync_clock(void *dummy)
361 atomic_t *sw_ptr = this_cpu_ptr(&clock_sync_word);
363 * Clear the in-sync bit 2^31. All get_phys_clock calls will
364 * fail until the sync bit is turned back on. In addition
365 * increase the "sequence" counter to avoid the race of an
366 * stp event and the complete recovery against get_phys_clock.
368 atomic_andnot(0x80000000, sw_ptr);
373 * Make get_phys_clock() return 0 again.
374 * Needs to be called from a context disabled for preemption.
376 static void enable_sync_clock(void)
378 atomic_t *sw_ptr = this_cpu_ptr(&clock_sync_word);
379 atomic_or(0x80000000, sw_ptr);
383 * Function to check if the clock is in sync.
385 static inline int check_sync_clock(void)
390 sw_ptr = &get_cpu_var(clock_sync_word);
391 rc = (atomic_read(sw_ptr) & 0x80000000U) != 0;
392 put_cpu_var(clock_sync_word);
397 * Apply clock delta to the global data structures.
398 * This is called once on the CPU that performed the clock sync.
400 static void clock_sync_global(unsigned long long delta)
402 unsigned long now, adj;
405 /* Fixup the monotonic sched clock. */
406 sched_clock_base_cc += delta;
407 /* Adjust TOD steering parameters. */
408 vdso_data->tb_update_count++;
409 now = get_tod_clock();
410 adj = tod_steering_end - now;
411 if (unlikely((s64) adj >= 0))
412 /* Calculate how much of the old adjustment is left. */
413 tod_steering_delta = (tod_steering_delta < 0) ?
414 -(adj >> 15) : (adj >> 15);
415 tod_steering_delta += delta;
416 if ((abs(tod_steering_delta) >> 48) != 0)
417 panic("TOD clock sync offset %lli is too large to drift\n",
419 tod_steering_end = now + (abs(tod_steering_delta) << 15);
420 vdso_data->ts_dir = (tod_steering_delta < 0) ? 0 : 1;
421 vdso_data->ts_end = tod_steering_end;
422 vdso_data->tb_update_count++;
423 /* Update LPAR offset. */
424 if (ptff_query(PTFF_QTO) && ptff(&qto, sizeof(qto), PTFF_QTO) == 0)
425 lpar_offset = qto.tod_epoch_difference;
426 /* Call the TOD clock change notifier. */
427 atomic_notifier_call_chain(&s390_epoch_delta_notifier, 0, &delta);
431 * Apply clock delta to the per-CPU data structures of this CPU.
432 * This is called for each online CPU after the call to clock_sync_global.
434 static void clock_sync_local(unsigned long long delta)
436 /* Add the delta to the clock comparator. */
437 if (S390_lowcore.clock_comparator != -1ULL) {
438 S390_lowcore.clock_comparator += delta;
439 set_clock_comparator(S390_lowcore.clock_comparator);
441 /* Adjust the last_update_clock time-stamp. */
442 S390_lowcore.last_update_clock += delta;
445 /* Single threaded workqueue used for stp sync events */
446 static struct workqueue_struct *time_sync_wq;
448 static void __init time_init_wq(void)
452 time_sync_wq = create_singlethread_workqueue("timesync");
455 struct clock_sync_data {
458 unsigned long long clock_delta;
462 * Server Time Protocol (STP) code.
464 static bool stp_online;
465 static struct stp_sstpi stp_info;
466 static void *stp_page;
468 static void stp_work_fn(struct work_struct *work);
469 static DEFINE_MUTEX(stp_work_mutex);
470 static DECLARE_WORK(stp_work, stp_work_fn);
471 static struct timer_list stp_timer;
473 static int __init early_parse_stp(char *p)
475 return kstrtobool(p, &stp_online);
477 early_param("stp", early_parse_stp);
480 * Reset STP attachment.
482 static void __init stp_reset(void)
486 stp_page = (void *) get_zeroed_page(GFP_ATOMIC);
487 rc = chsc_sstpc(stp_page, STP_OP_CTRL, 0x0000, NULL);
489 set_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags);
490 else if (stp_online) {
491 pr_warn("The real or virtual hardware system does not provide an STP interface\n");
492 free_page((unsigned long) stp_page);
498 static void stp_timeout(unsigned long dummy)
500 queue_work(time_sync_wq, &stp_work);
503 static int __init stp_init(void)
505 if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
507 setup_timer(&stp_timer, stp_timeout, 0UL);
511 queue_work(time_sync_wq, &stp_work);
515 arch_initcall(stp_init);
518 * STP timing alert. There are three causes:
519 * 1) timing status change
520 * 2) link availability change
521 * 3) time control parameter change
522 * In all three cases we are only interested in the clock source state.
523 * If a STP clock source is now available use it.
525 static void stp_timing_alert(struct stp_irq_parm *intparm)
527 if (intparm->tsc || intparm->lac || intparm->tcpc)
528 queue_work(time_sync_wq, &stp_work);
532 * STP sync check machine check. This is called when the timing state
533 * changes from the synchronized state to the unsynchronized state.
534 * After a STP sync check the clock is not in sync. The machine check
535 * is broadcasted to all cpus at the same time.
537 int stp_sync_check(void)
539 disable_sync_clock(NULL);
544 * STP island condition machine check. This is called when an attached
545 * server attempts to communicate over an STP link and the servers
546 * have matching CTN ids and have a valid stratum-1 configuration
547 * but the configurations do not match.
549 int stp_island_check(void)
551 disable_sync_clock(NULL);
555 void stp_queue_work(void)
557 queue_work(time_sync_wq, &stp_work);
560 static int stp_sync_clock(void *data)
562 struct clock_sync_data *sync = data;
563 unsigned long long clock_delta;
568 if (xchg(&first, 1) == 0) {
569 /* Wait until all other cpus entered the sync function. */
570 while (atomic_read(&sync->cpus) != 0)
573 if (stp_info.todoff[0] || stp_info.todoff[1] ||
574 stp_info.todoff[2] || stp_info.todoff[3] ||
576 rc = chsc_sstpc(stp_page, STP_OP_SYNC, 0,
579 sync->clock_delta = clock_delta;
580 clock_sync_global(clock_delta);
581 rc = chsc_sstpi(stp_page, &stp_info,
582 sizeof(struct stp_sstpi));
583 if (rc == 0 && stp_info.tmd != 2)
587 sync->in_sync = rc ? -EAGAIN : 1;
591 atomic_dec(&sync->cpus);
592 /* Wait for in_sync to be set. */
593 while (READ_ONCE(sync->in_sync) == 0)
596 if (sync->in_sync != 1)
597 /* Didn't work. Clear per-cpu in sync bit again. */
598 disable_sync_clock(NULL);
599 /* Apply clock delta to per-CPU fields of this CPU. */
600 clock_sync_local(sync->clock_delta);
606 * STP work. Check for the STP state and take over the clock
607 * synchronization if the STP clock source is usable.
609 static void stp_work_fn(struct work_struct *work)
611 struct clock_sync_data stp_sync;
614 /* prevent multiple execution. */
615 mutex_lock(&stp_work_mutex);
618 chsc_sstpc(stp_page, STP_OP_CTRL, 0x0000, NULL);
619 del_timer_sync(&stp_timer);
623 rc = chsc_sstpc(stp_page, STP_OP_CTRL, 0xb0e0, NULL);
627 rc = chsc_sstpi(stp_page, &stp_info, sizeof(struct stp_sstpi));
628 if (rc || stp_info.c == 0)
631 /* Skip synchronization if the clock is already in sync. */
632 if (check_sync_clock())
635 memset(&stp_sync, 0, sizeof(stp_sync));
637 atomic_set(&stp_sync.cpus, num_online_cpus() - 1);
638 stop_machine(stp_sync_clock, &stp_sync, cpu_online_mask);
641 if (!check_sync_clock())
643 * There is a usable clock but the synchonization failed.
644 * Retry after a second.
646 mod_timer(&stp_timer, jiffies + HZ);
649 mutex_unlock(&stp_work_mutex);
653 * STP subsys sysfs interface functions
655 static struct bus_type stp_subsys = {
660 static ssize_t stp_ctn_id_show(struct device *dev,
661 struct device_attribute *attr,
666 return sprintf(buf, "%016llx\n",
667 *(unsigned long long *) stp_info.ctnid);
670 static DEVICE_ATTR(ctn_id, 0400, stp_ctn_id_show, NULL);
672 static ssize_t stp_ctn_type_show(struct device *dev,
673 struct device_attribute *attr,
678 return sprintf(buf, "%i\n", stp_info.ctn);
681 static DEVICE_ATTR(ctn_type, 0400, stp_ctn_type_show, NULL);
683 static ssize_t stp_dst_offset_show(struct device *dev,
684 struct device_attribute *attr,
687 if (!stp_online || !(stp_info.vbits & 0x2000))
689 return sprintf(buf, "%i\n", (int)(s16) stp_info.dsto);
692 static DEVICE_ATTR(dst_offset, 0400, stp_dst_offset_show, NULL);
694 static ssize_t stp_leap_seconds_show(struct device *dev,
695 struct device_attribute *attr,
698 if (!stp_online || !(stp_info.vbits & 0x8000))
700 return sprintf(buf, "%i\n", (int)(s16) stp_info.leaps);
703 static DEVICE_ATTR(leap_seconds, 0400, stp_leap_seconds_show, NULL);
705 static ssize_t stp_stratum_show(struct device *dev,
706 struct device_attribute *attr,
711 return sprintf(buf, "%i\n", (int)(s16) stp_info.stratum);
714 static DEVICE_ATTR(stratum, 0400, stp_stratum_show, NULL);
716 static ssize_t stp_time_offset_show(struct device *dev,
717 struct device_attribute *attr,
720 if (!stp_online || !(stp_info.vbits & 0x0800))
722 return sprintf(buf, "%i\n", (int) stp_info.tto);
725 static DEVICE_ATTR(time_offset, 0400, stp_time_offset_show, NULL);
727 static ssize_t stp_time_zone_offset_show(struct device *dev,
728 struct device_attribute *attr,
731 if (!stp_online || !(stp_info.vbits & 0x4000))
733 return sprintf(buf, "%i\n", (int)(s16) stp_info.tzo);
736 static DEVICE_ATTR(time_zone_offset, 0400,
737 stp_time_zone_offset_show, NULL);
739 static ssize_t stp_timing_mode_show(struct device *dev,
740 struct device_attribute *attr,
745 return sprintf(buf, "%i\n", stp_info.tmd);
748 static DEVICE_ATTR(timing_mode, 0400, stp_timing_mode_show, NULL);
750 static ssize_t stp_timing_state_show(struct device *dev,
751 struct device_attribute *attr,
756 return sprintf(buf, "%i\n", stp_info.tst);
759 static DEVICE_ATTR(timing_state, 0400, stp_timing_state_show, NULL);
761 static ssize_t stp_online_show(struct device *dev,
762 struct device_attribute *attr,
765 return sprintf(buf, "%i\n", stp_online);
768 static ssize_t stp_online_store(struct device *dev,
769 struct device_attribute *attr,
770 const char *buf, size_t count)
774 value = simple_strtoul(buf, NULL, 0);
775 if (value != 0 && value != 1)
777 if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
779 mutex_lock(&clock_sync_mutex);
782 set_bit(CLOCK_SYNC_STP, &clock_sync_flags);
784 clear_bit(CLOCK_SYNC_STP, &clock_sync_flags);
785 queue_work(time_sync_wq, &stp_work);
786 mutex_unlock(&clock_sync_mutex);
791 * Can't use DEVICE_ATTR because the attribute should be named
792 * stp/online but dev_attr_online already exists in this file ..
794 static struct device_attribute dev_attr_stp_online = {
795 .attr = { .name = "online", .mode = 0600 },
796 .show = stp_online_show,
797 .store = stp_online_store,
800 static struct device_attribute *stp_attributes[] = {
803 &dev_attr_dst_offset,
804 &dev_attr_leap_seconds,
805 &dev_attr_stp_online,
807 &dev_attr_time_offset,
808 &dev_attr_time_zone_offset,
809 &dev_attr_timing_mode,
810 &dev_attr_timing_state,
814 static int __init stp_init_sysfs(void)
816 struct device_attribute **attr;
819 rc = subsys_system_register(&stp_subsys, NULL);
822 for (attr = stp_attributes; *attr; attr++) {
823 rc = device_create_file(stp_subsys.dev_root, *attr);
829 for (; attr >= stp_attributes; attr--)
830 device_remove_file(stp_subsys.dev_root, *attr);
831 bus_unregister(&stp_subsys);
836 device_initcall(stp_init_sysfs);