xref: /openbmc/linux/arch/s390/kernel/time.c (revision dfe94d40)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *    Time of day based timer functions.
4  *
5  *  S390 version
6  *    Copyright IBM Corp. 1999, 2008
7  *    Author(s): Hartmut Penner (hp@de.ibm.com),
8  *               Martin Schwidefsky (schwidefsky@de.ibm.com),
9  *               Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com)
10  *
11  *  Derived from "arch/i386/kernel/time.c"
12  *    Copyright (C) 1991, 1992, 1995  Linus Torvalds
13  */
14 
15 #define KMSG_COMPONENT "time"
16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
17 
18 #include <linux/kernel_stat.h>
19 #include <linux/errno.h>
20 #include <linux/export.h>
21 #include <linux/sched.h>
22 #include <linux/sched/clock.h>
23 #include <linux/kernel.h>
24 #include <linux/param.h>
25 #include <linux/string.h>
26 #include <linux/mm.h>
27 #include <linux/interrupt.h>
28 #include <linux/cpu.h>
29 #include <linux/stop_machine.h>
30 #include <linux/time.h>
31 #include <linux/device.h>
32 #include <linux/delay.h>
33 #include <linux/init.h>
34 #include <linux/smp.h>
35 #include <linux/types.h>
36 #include <linux/profile.h>
37 #include <linux/timex.h>
38 #include <linux/notifier.h>
39 #include <linux/timekeeper_internal.h>
40 #include <linux/clockchips.h>
41 #include <linux/gfp.h>
42 #include <linux/kprobes.h>
43 #include <linux/uaccess.h>
44 #include <vdso/vsyscall.h>
45 #include <vdso/clocksource.h>
46 #include <vdso/helpers.h>
47 #include <asm/facility.h>
48 #include <asm/delay.h>
49 #include <asm/div64.h>
50 #include <asm/vdso.h>
51 #include <asm/irq.h>
52 #include <asm/irq_regs.h>
53 #include <asm/vtimer.h>
54 #include <asm/stp.h>
55 #include <asm/cio.h>
56 #include "entry.h"
57 
58 unsigned char tod_clock_base[16] __aligned(8) = {
59 	/* Force to data section. */
60 	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
61 	0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
62 };
63 EXPORT_SYMBOL_GPL(tod_clock_base);
64 
65 u64 clock_comparator_max = -1ULL;
66 EXPORT_SYMBOL_GPL(clock_comparator_max);
67 
68 static DEFINE_PER_CPU(struct clock_event_device, comparators);
69 
70 ATOMIC_NOTIFIER_HEAD(s390_epoch_delta_notifier);
71 EXPORT_SYMBOL(s390_epoch_delta_notifier);
72 
73 unsigned char ptff_function_mask[16];
74 
75 static unsigned long long lpar_offset;
76 static unsigned long long initial_leap_seconds;
77 static unsigned long long tod_steering_end;
78 static long long tod_steering_delta;
79 
80 /*
81  * Get time offsets with PTFF
82  */
83 void __init time_early_init(void)
84 {
85 	struct ptff_qto qto;
86 	struct ptff_qui qui;
87 
88 	/* Initialize TOD steering parameters */
89 	tod_steering_end = *(unsigned long long *) &tod_clock_base[1];
90 	vdso_data->arch_data.tod_steering_end = tod_steering_end;
91 
92 	if (!test_facility(28))
93 		return;
94 
95 	ptff(&ptff_function_mask, sizeof(ptff_function_mask), PTFF_QAF);
96 
97 	/* get LPAR offset */
98 	if (ptff_query(PTFF_QTO) && ptff(&qto, sizeof(qto), PTFF_QTO) == 0)
99 		lpar_offset = qto.tod_epoch_difference;
100 
101 	/* get initial leap seconds */
102 	if (ptff_query(PTFF_QUI) && ptff(&qui, sizeof(qui), PTFF_QUI) == 0)
103 		initial_leap_seconds = (unsigned long long)
104 			((long) qui.old_leap * 4096000000L);
105 }
106 
107 /*
108  * Scheduler clock - returns current time in nanosec units.
109  */
110 unsigned long long notrace sched_clock(void)
111 {
112 	return tod_to_ns(get_tod_clock_monotonic());
113 }
114 NOKPROBE_SYMBOL(sched_clock);
115 
116 static void ext_to_timespec64(unsigned char *clk, struct timespec64 *xt)
117 {
118 	unsigned long long high, low, rem, sec, nsec;
119 
120 	/* Split extendnd TOD clock to micro-seconds and sub-micro-seconds */
121 	high = (*(unsigned long long *) clk) >> 4;
122 	low = (*(unsigned long long *)&clk[7]) << 4;
123 	/* Calculate seconds and nano-seconds */
124 	sec = high;
125 	rem = do_div(sec, 1000000);
126 	nsec = (((low >> 32) + (rem << 32)) * 1000) >> 32;
127 
128 	xt->tv_sec = sec;
129 	xt->tv_nsec = nsec;
130 }
131 
132 void clock_comparator_work(void)
133 {
134 	struct clock_event_device *cd;
135 
136 	S390_lowcore.clock_comparator = clock_comparator_max;
137 	cd = this_cpu_ptr(&comparators);
138 	cd->event_handler(cd);
139 }
140 
141 static int s390_next_event(unsigned long delta,
142 			   struct clock_event_device *evt)
143 {
144 	S390_lowcore.clock_comparator = get_tod_clock() + delta;
145 	set_clock_comparator(S390_lowcore.clock_comparator);
146 	return 0;
147 }
148 
149 /*
150  * Set up lowcore and control register of the current cpu to
151  * enable TOD clock and clock comparator interrupts.
152  */
153 void init_cpu_timer(void)
154 {
155 	struct clock_event_device *cd;
156 	int cpu;
157 
158 	S390_lowcore.clock_comparator = clock_comparator_max;
159 	set_clock_comparator(S390_lowcore.clock_comparator);
160 
161 	cpu = smp_processor_id();
162 	cd = &per_cpu(comparators, cpu);
163 	cd->name		= "comparator";
164 	cd->features		= CLOCK_EVT_FEAT_ONESHOT;
165 	cd->mult		= 16777;
166 	cd->shift		= 12;
167 	cd->min_delta_ns	= 1;
168 	cd->min_delta_ticks	= 1;
169 	cd->max_delta_ns	= LONG_MAX;
170 	cd->max_delta_ticks	= ULONG_MAX;
171 	cd->rating		= 400;
172 	cd->cpumask		= cpumask_of(cpu);
173 	cd->set_next_event	= s390_next_event;
174 
175 	clockevents_register_device(cd);
176 
177 	/* Enable clock comparator timer interrupt. */
178 	__ctl_set_bit(0,11);
179 
180 	/* Always allow the timing alert external interrupt. */
181 	__ctl_set_bit(0, 4);
182 }
183 
184 static void clock_comparator_interrupt(struct ext_code ext_code,
185 				       unsigned int param32,
186 				       unsigned long param64)
187 {
188 	inc_irq_stat(IRQEXT_CLK);
189 	if (S390_lowcore.clock_comparator == clock_comparator_max)
190 		set_clock_comparator(S390_lowcore.clock_comparator);
191 }
192 
193 static void stp_timing_alert(struct stp_irq_parm *);
194 
195 static void timing_alert_interrupt(struct ext_code ext_code,
196 				   unsigned int param32, unsigned long param64)
197 {
198 	inc_irq_stat(IRQEXT_TLA);
199 	if (param32 & 0x00038000)
200 		stp_timing_alert((struct stp_irq_parm *) &param32);
201 }
202 
203 static void stp_reset(void);
204 
205 void read_persistent_clock64(struct timespec64 *ts)
206 {
207 	unsigned char clk[STORE_CLOCK_EXT_SIZE];
208 	__u64 delta;
209 
210 	delta = initial_leap_seconds + TOD_UNIX_EPOCH;
211 	get_tod_clock_ext(clk);
212 	*(__u64 *) &clk[1] -= delta;
213 	if (*(__u64 *) &clk[1] > delta)
214 		clk[0]--;
215 	ext_to_timespec64(clk, ts);
216 }
217 
218 void __init read_persistent_wall_and_boot_offset(struct timespec64 *wall_time,
219 						 struct timespec64 *boot_offset)
220 {
221 	unsigned char clk[STORE_CLOCK_EXT_SIZE];
222 	struct timespec64 boot_time;
223 	__u64 delta;
224 
225 	delta = initial_leap_seconds + TOD_UNIX_EPOCH;
226 	memcpy(clk, tod_clock_base, STORE_CLOCK_EXT_SIZE);
227 	*(__u64 *)&clk[1] -= delta;
228 	if (*(__u64 *)&clk[1] > delta)
229 		clk[0]--;
230 	ext_to_timespec64(clk, &boot_time);
231 
232 	read_persistent_clock64(wall_time);
233 	*boot_offset = timespec64_sub(*wall_time, boot_time);
234 }
235 
236 static u64 read_tod_clock(struct clocksource *cs)
237 {
238 	unsigned long long now, adj;
239 
240 	preempt_disable(); /* protect from changes to steering parameters */
241 	now = get_tod_clock();
242 	adj = tod_steering_end - now;
243 	if (unlikely((s64) adj > 0))
244 		/*
245 		 * manually steer by 1 cycle every 2^16 cycles. This
246 		 * corresponds to shifting the tod delta by 15. 1s is
247 		 * therefore steered in ~9h. The adjust will decrease
248 		 * over time, until it finally reaches 0.
249 		 */
250 		now += (tod_steering_delta < 0) ? (adj >> 15) : -(adj >> 15);
251 	preempt_enable();
252 	return now;
253 }
254 
255 static struct clocksource clocksource_tod = {
256 	.name		= "tod",
257 	.rating		= 400,
258 	.read		= read_tod_clock,
259 	.mask		= CLOCKSOURCE_MASK(64),
260 	.mult		= 1000,
261 	.shift		= 12,
262 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
263 	.vdso_clock_mode = VDSO_CLOCKMODE_TOD,
264 };
265 
266 struct clocksource * __init clocksource_default_clock(void)
267 {
268 	return &clocksource_tod;
269 }
270 
271 /*
272  * Initialize the TOD clock and the CPU timer of
273  * the boot cpu.
274  */
275 void __init time_init(void)
276 {
277 	/* Reset time synchronization interfaces. */
278 	stp_reset();
279 
280 	/* request the clock comparator external interrupt */
281 	if (register_external_irq(EXT_IRQ_CLK_COMP, clock_comparator_interrupt))
282 		panic("Couldn't request external interrupt 0x1004");
283 
284 	/* request the timing alert external interrupt */
285 	if (register_external_irq(EXT_IRQ_TIMING_ALERT, timing_alert_interrupt))
286 		panic("Couldn't request external interrupt 0x1406");
287 
288 	if (__clocksource_register(&clocksource_tod) != 0)
289 		panic("Could not register TOD clock source");
290 
291 	/* Enable TOD clock interrupts on the boot cpu. */
292 	init_cpu_timer();
293 
294 	/* Enable cpu timer interrupts on the boot cpu. */
295 	vtime_init();
296 }
297 
298 static DEFINE_PER_CPU(atomic_t, clock_sync_word);
299 static DEFINE_MUTEX(stp_mutex);
300 static unsigned long clock_sync_flags;
301 
302 #define CLOCK_SYNC_HAS_STP		0
303 #define CLOCK_SYNC_STP			1
304 #define CLOCK_SYNC_STPINFO_VALID	2
305 
306 /*
307  * The get_clock function for the physical clock. It will get the current
308  * TOD clock, subtract the LPAR offset and write the result to *clock.
309  * The function returns 0 if the clock is in sync with the external time
310  * source. If the clock mode is local it will return -EOPNOTSUPP and
311  * -EAGAIN if the clock is not in sync with the external reference.
312  */
313 int get_phys_clock(unsigned long *clock)
314 {
315 	atomic_t *sw_ptr;
316 	unsigned int sw0, sw1;
317 
318 	sw_ptr = &get_cpu_var(clock_sync_word);
319 	sw0 = atomic_read(sw_ptr);
320 	*clock = get_tod_clock() - lpar_offset;
321 	sw1 = atomic_read(sw_ptr);
322 	put_cpu_var(clock_sync_word);
323 	if (sw0 == sw1 && (sw0 & 0x80000000U))
324 		/* Success: time is in sync. */
325 		return 0;
326 	if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
327 		return -EOPNOTSUPP;
328 	if (!test_bit(CLOCK_SYNC_STP, &clock_sync_flags))
329 		return -EACCES;
330 	return -EAGAIN;
331 }
332 EXPORT_SYMBOL(get_phys_clock);
333 
334 /*
335  * Make get_phys_clock() return -EAGAIN.
336  */
337 static void disable_sync_clock(void *dummy)
338 {
339 	atomic_t *sw_ptr = this_cpu_ptr(&clock_sync_word);
340 	/*
341 	 * Clear the in-sync bit 2^31. All get_phys_clock calls will
342 	 * fail until the sync bit is turned back on. In addition
343 	 * increase the "sequence" counter to avoid the race of an
344 	 * stp event and the complete recovery against get_phys_clock.
345 	 */
346 	atomic_andnot(0x80000000, sw_ptr);
347 	atomic_inc(sw_ptr);
348 }
349 
350 /*
351  * Make get_phys_clock() return 0 again.
352  * Needs to be called from a context disabled for preemption.
353  */
354 static void enable_sync_clock(void)
355 {
356 	atomic_t *sw_ptr = this_cpu_ptr(&clock_sync_word);
357 	atomic_or(0x80000000, sw_ptr);
358 }
359 
360 /*
361  * Function to check if the clock is in sync.
362  */
363 static inline int check_sync_clock(void)
364 {
365 	atomic_t *sw_ptr;
366 	int rc;
367 
368 	sw_ptr = &get_cpu_var(clock_sync_word);
369 	rc = (atomic_read(sw_ptr) & 0x80000000U) != 0;
370 	put_cpu_var(clock_sync_word);
371 	return rc;
372 }
373 
374 /*
375  * Apply clock delta to the global data structures.
376  * This is called once on the CPU that performed the clock sync.
377  */
378 static void clock_sync_global(unsigned long long delta)
379 {
380 	unsigned long now, adj;
381 	struct ptff_qto qto;
382 
383 	/* Fixup the monotonic sched clock. */
384 	*(unsigned long long *) &tod_clock_base[1] += delta;
385 	if (*(unsigned long long *) &tod_clock_base[1] < delta)
386 		/* Epoch overflow */
387 		tod_clock_base[0]++;
388 	/* Adjust TOD steering parameters. */
389 	now = get_tod_clock();
390 	adj = tod_steering_end - now;
391 	if (unlikely((s64) adj >= 0))
392 		/* Calculate how much of the old adjustment is left. */
393 		tod_steering_delta = (tod_steering_delta < 0) ?
394 			-(adj >> 15) : (adj >> 15);
395 	tod_steering_delta += delta;
396 	if ((abs(tod_steering_delta) >> 48) != 0)
397 		panic("TOD clock sync offset %lli is too large to drift\n",
398 		      tod_steering_delta);
399 	tod_steering_end = now + (abs(tod_steering_delta) << 15);
400 	vdso_data->arch_data.tod_steering_end = tod_steering_end;
401 
402 	/* Update LPAR offset. */
403 	if (ptff_query(PTFF_QTO) && ptff(&qto, sizeof(qto), PTFF_QTO) == 0)
404 		lpar_offset = qto.tod_epoch_difference;
405 	/* Call the TOD clock change notifier. */
406 	atomic_notifier_call_chain(&s390_epoch_delta_notifier, 0, &delta);
407 }
408 
409 /*
410  * Apply clock delta to the per-CPU data structures of this CPU.
411  * This is called for each online CPU after the call to clock_sync_global.
412  */
413 static void clock_sync_local(unsigned long long delta)
414 {
415 	/* Add the delta to the clock comparator. */
416 	if (S390_lowcore.clock_comparator != clock_comparator_max) {
417 		S390_lowcore.clock_comparator += delta;
418 		set_clock_comparator(S390_lowcore.clock_comparator);
419 	}
420 	/* Adjust the last_update_clock time-stamp. */
421 	S390_lowcore.last_update_clock += delta;
422 }
423 
424 /* Single threaded workqueue used for stp sync events */
425 static struct workqueue_struct *time_sync_wq;
426 
427 static void __init time_init_wq(void)
428 {
429 	if (time_sync_wq)
430 		return;
431 	time_sync_wq = create_singlethread_workqueue("timesync");
432 }
433 
434 struct clock_sync_data {
435 	atomic_t cpus;
436 	int in_sync;
437 	unsigned long long clock_delta;
438 };
439 
440 /*
441  * Server Time Protocol (STP) code.
442  */
443 static bool stp_online;
444 static struct stp_sstpi stp_info;
445 static void *stp_page;
446 
447 static void stp_work_fn(struct work_struct *work);
448 static DECLARE_WORK(stp_work, stp_work_fn);
449 static struct timer_list stp_timer;
450 
451 static int __init early_parse_stp(char *p)
452 {
453 	return kstrtobool(p, &stp_online);
454 }
455 early_param("stp", early_parse_stp);
456 
457 /*
458  * Reset STP attachment.
459  */
460 static void __init stp_reset(void)
461 {
462 	int rc;
463 
464 	stp_page = (void *) get_zeroed_page(GFP_ATOMIC);
465 	rc = chsc_sstpc(stp_page, STP_OP_CTRL, 0x0000, NULL);
466 	if (rc == 0)
467 		set_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags);
468 	else if (stp_online) {
469 		pr_warn("The real or virtual hardware system does not provide an STP interface\n");
470 		free_page((unsigned long) stp_page);
471 		stp_page = NULL;
472 		stp_online = false;
473 	}
474 }
475 
476 static void stp_timeout(struct timer_list *unused)
477 {
478 	queue_work(time_sync_wq, &stp_work);
479 }
480 
481 static int __init stp_init(void)
482 {
483 	if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
484 		return 0;
485 	timer_setup(&stp_timer, stp_timeout, 0);
486 	time_init_wq();
487 	if (!stp_online)
488 		return 0;
489 	queue_work(time_sync_wq, &stp_work);
490 	return 0;
491 }
492 
493 arch_initcall(stp_init);
494 
495 /*
496  * STP timing alert. There are three causes:
497  * 1) timing status change
498  * 2) link availability change
499  * 3) time control parameter change
500  * In all three cases we are only interested in the clock source state.
501  * If a STP clock source is now available use it.
502  */
503 static void stp_timing_alert(struct stp_irq_parm *intparm)
504 {
505 	if (intparm->tsc || intparm->lac || intparm->tcpc)
506 		queue_work(time_sync_wq, &stp_work);
507 }
508 
509 /*
510  * STP sync check machine check. This is called when the timing state
511  * changes from the synchronized state to the unsynchronized state.
512  * After a STP sync check the clock is not in sync. The machine check
513  * is broadcasted to all cpus at the same time.
514  */
515 int stp_sync_check(void)
516 {
517 	disable_sync_clock(NULL);
518 	return 1;
519 }
520 
521 /*
522  * STP island condition machine check. This is called when an attached
523  * server  attempts to communicate over an STP link and the servers
524  * have matching CTN ids and have a valid stratum-1 configuration
525  * but the configurations do not match.
526  */
527 int stp_island_check(void)
528 {
529 	disable_sync_clock(NULL);
530 	return 1;
531 }
532 
533 void stp_queue_work(void)
534 {
535 	queue_work(time_sync_wq, &stp_work);
536 }
537 
538 static int __store_stpinfo(void)
539 {
540 	int rc = chsc_sstpi(stp_page, &stp_info, sizeof(struct stp_sstpi));
541 
542 	if (rc)
543 		clear_bit(CLOCK_SYNC_STPINFO_VALID, &clock_sync_flags);
544 	else
545 		set_bit(CLOCK_SYNC_STPINFO_VALID, &clock_sync_flags);
546 	return rc;
547 }
548 
549 static int stpinfo_valid(void)
550 {
551 	return stp_online && test_bit(CLOCK_SYNC_STPINFO_VALID, &clock_sync_flags);
552 }
553 
554 static int stp_sync_clock(void *data)
555 {
556 	struct clock_sync_data *sync = data;
557 	unsigned long long clock_delta, flags;
558 	static int first;
559 	int rc;
560 
561 	enable_sync_clock();
562 	if (xchg(&first, 1) == 0) {
563 		/* Wait until all other cpus entered the sync function. */
564 		while (atomic_read(&sync->cpus) != 0)
565 			cpu_relax();
566 		rc = 0;
567 		if (stp_info.todoff[0] || stp_info.todoff[1] ||
568 		    stp_info.todoff[2] || stp_info.todoff[3] ||
569 		    stp_info.tmd != 2) {
570 			flags = vdso_update_begin();
571 			rc = chsc_sstpc(stp_page, STP_OP_SYNC, 0,
572 					&clock_delta);
573 			if (rc == 0) {
574 				sync->clock_delta = clock_delta;
575 				clock_sync_global(clock_delta);
576 				rc = __store_stpinfo();
577 				if (rc == 0 && stp_info.tmd != 2)
578 					rc = -EAGAIN;
579 			}
580 			vdso_update_end(flags);
581 		}
582 		sync->in_sync = rc ? -EAGAIN : 1;
583 		xchg(&first, 0);
584 	} else {
585 		/* Slave */
586 		atomic_dec(&sync->cpus);
587 		/* Wait for in_sync to be set. */
588 		while (READ_ONCE(sync->in_sync) == 0)
589 			__udelay(1);
590 	}
591 	if (sync->in_sync != 1)
592 		/* Didn't work. Clear per-cpu in sync bit again. */
593 		disable_sync_clock(NULL);
594 	/* Apply clock delta to per-CPU fields of this CPU. */
595 	clock_sync_local(sync->clock_delta);
596 
597 	return 0;
598 }
599 
600 static int stp_clear_leap(void)
601 {
602 	struct __kernel_timex txc;
603 	int ret;
604 
605 	memset(&txc, 0, sizeof(txc));
606 
607 	ret = do_adjtimex(&txc);
608 	if (ret < 0)
609 		return ret;
610 
611 	txc.modes = ADJ_STATUS;
612 	txc.status &= ~(STA_INS|STA_DEL);
613 	return do_adjtimex(&txc);
614 }
615 
616 static void stp_check_leap(void)
617 {
618 	struct stp_stzi stzi;
619 	struct stp_lsoib *lsoib = &stzi.lsoib;
620 	struct __kernel_timex txc;
621 	int64_t timediff;
622 	int leapdiff, ret;
623 
624 	if (!stp_info.lu || !check_sync_clock()) {
625 		/*
626 		 * Either a scheduled leap second was removed by the operator,
627 		 * or STP is out of sync. In both cases, clear the leap second
628 		 * kernel flags.
629 		 */
630 		if (stp_clear_leap() < 0)
631 			pr_err("failed to clear leap second flags\n");
632 		return;
633 	}
634 
635 	if (chsc_stzi(stp_page, &stzi, sizeof(stzi))) {
636 		pr_err("stzi failed\n");
637 		return;
638 	}
639 
640 	timediff = tod_to_ns(lsoib->nlsout - get_tod_clock()) / NSEC_PER_SEC;
641 	leapdiff = lsoib->nlso - lsoib->also;
642 
643 	if (leapdiff != 1 && leapdiff != -1) {
644 		pr_err("Cannot schedule %d leap seconds\n", leapdiff);
645 		return;
646 	}
647 
648 	if (timediff < 0) {
649 		if (stp_clear_leap() < 0)
650 			pr_err("failed to clear leap second flags\n");
651 	} else if (timediff < 7200) {
652 		memset(&txc, 0, sizeof(txc));
653 		ret = do_adjtimex(&txc);
654 		if (ret < 0)
655 			return;
656 
657 		txc.modes = ADJ_STATUS;
658 		if (leapdiff > 0)
659 			txc.status |= STA_INS;
660 		else
661 			txc.status |= STA_DEL;
662 		ret = do_adjtimex(&txc);
663 		if (ret < 0)
664 			pr_err("failed to set leap second flags\n");
665 		/* arm Timer to clear leap second flags */
666 		mod_timer(&stp_timer, jiffies + msecs_to_jiffies(14400 * MSEC_PER_SEC));
667 	} else {
668 		/* The day the leap second is scheduled for hasn't been reached. Retry
669 		 * in one hour.
670 		 */
671 		mod_timer(&stp_timer, jiffies + msecs_to_jiffies(3600 * MSEC_PER_SEC));
672 	}
673 }
674 
675 /*
676  * STP work. Check for the STP state and take over the clock
677  * synchronization if the STP clock source is usable.
678  */
679 static void stp_work_fn(struct work_struct *work)
680 {
681 	struct clock_sync_data stp_sync;
682 	int rc;
683 
684 	/* prevent multiple execution. */
685 	mutex_lock(&stp_mutex);
686 
687 	if (!stp_online) {
688 		chsc_sstpc(stp_page, STP_OP_CTRL, 0x0000, NULL);
689 		del_timer_sync(&stp_timer);
690 		goto out_unlock;
691 	}
692 
693 	rc = chsc_sstpc(stp_page, STP_OP_CTRL, 0xf0e0, NULL);
694 	if (rc)
695 		goto out_unlock;
696 
697 	rc = __store_stpinfo();
698 	if (rc || stp_info.c == 0)
699 		goto out_unlock;
700 
701 	/* Skip synchronization if the clock is already in sync. */
702 	if (!check_sync_clock()) {
703 		memset(&stp_sync, 0, sizeof(stp_sync));
704 		cpus_read_lock();
705 		atomic_set(&stp_sync.cpus, num_online_cpus() - 1);
706 		stop_machine_cpuslocked(stp_sync_clock, &stp_sync, cpu_online_mask);
707 		cpus_read_unlock();
708 	}
709 
710 	if (!check_sync_clock())
711 		/*
712 		 * There is a usable clock but the synchonization failed.
713 		 * Retry after a second.
714 		 */
715 		mod_timer(&stp_timer, jiffies + msecs_to_jiffies(MSEC_PER_SEC));
716 	else if (stp_info.lu)
717 		stp_check_leap();
718 
719 out_unlock:
720 	mutex_unlock(&stp_mutex);
721 }
722 
723 /*
724  * STP subsys sysfs interface functions
725  */
726 static struct bus_type stp_subsys = {
727 	.name		= "stp",
728 	.dev_name	= "stp",
729 };
730 
731 static ssize_t ctn_id_show(struct device *dev,
732 				struct device_attribute *attr,
733 				char *buf)
734 {
735 	ssize_t ret = -ENODATA;
736 
737 	mutex_lock(&stp_mutex);
738 	if (stpinfo_valid())
739 		ret = sprintf(buf, "%016llx\n",
740 			      *(unsigned long long *) stp_info.ctnid);
741 	mutex_unlock(&stp_mutex);
742 	return ret;
743 }
744 
745 static DEVICE_ATTR_RO(ctn_id);
746 
747 static ssize_t ctn_type_show(struct device *dev,
748 				struct device_attribute *attr,
749 				char *buf)
750 {
751 	ssize_t ret = -ENODATA;
752 
753 	mutex_lock(&stp_mutex);
754 	if (stpinfo_valid())
755 		ret = sprintf(buf, "%i\n", stp_info.ctn);
756 	mutex_unlock(&stp_mutex);
757 	return ret;
758 }
759 
760 static DEVICE_ATTR_RO(ctn_type);
761 
762 static ssize_t dst_offset_show(struct device *dev,
763 				   struct device_attribute *attr,
764 				   char *buf)
765 {
766 	ssize_t ret = -ENODATA;
767 
768 	mutex_lock(&stp_mutex);
769 	if (stpinfo_valid() && (stp_info.vbits & 0x2000))
770 		ret = sprintf(buf, "%i\n", (int)(s16) stp_info.dsto);
771 	mutex_unlock(&stp_mutex);
772 	return ret;
773 }
774 
775 static DEVICE_ATTR_RO(dst_offset);
776 
777 static ssize_t leap_seconds_show(struct device *dev,
778 					struct device_attribute *attr,
779 					char *buf)
780 {
781 	ssize_t ret = -ENODATA;
782 
783 	mutex_lock(&stp_mutex);
784 	if (stpinfo_valid() && (stp_info.vbits & 0x8000))
785 		ret = sprintf(buf, "%i\n", (int)(s16) stp_info.leaps);
786 	mutex_unlock(&stp_mutex);
787 	return ret;
788 }
789 
790 static DEVICE_ATTR_RO(leap_seconds);
791 
792 static ssize_t leap_seconds_scheduled_show(struct device *dev,
793 						struct device_attribute *attr,
794 						char *buf)
795 {
796 	struct stp_stzi stzi;
797 	ssize_t ret;
798 
799 	mutex_lock(&stp_mutex);
800 	if (!stpinfo_valid() || !(stp_info.vbits & 0x8000) || !stp_info.lu) {
801 		mutex_unlock(&stp_mutex);
802 		return -ENODATA;
803 	}
804 
805 	ret = chsc_stzi(stp_page, &stzi, sizeof(stzi));
806 	mutex_unlock(&stp_mutex);
807 	if (ret < 0)
808 		return ret;
809 
810 	if (!stzi.lsoib.p)
811 		return sprintf(buf, "0,0\n");
812 
813 	return sprintf(buf, "%llu,%d\n",
814 		       tod_to_ns(stzi.lsoib.nlsout - TOD_UNIX_EPOCH) / NSEC_PER_SEC,
815 		       stzi.lsoib.nlso - stzi.lsoib.also);
816 }
817 
818 static DEVICE_ATTR_RO(leap_seconds_scheduled);
819 
820 static ssize_t stratum_show(struct device *dev,
821 				struct device_attribute *attr,
822 				char *buf)
823 {
824 	ssize_t ret = -ENODATA;
825 
826 	mutex_lock(&stp_mutex);
827 	if (stpinfo_valid())
828 		ret = sprintf(buf, "%i\n", (int)(s16) stp_info.stratum);
829 	mutex_unlock(&stp_mutex);
830 	return ret;
831 }
832 
833 static DEVICE_ATTR_RO(stratum);
834 
835 static ssize_t time_offset_show(struct device *dev,
836 				struct device_attribute *attr,
837 				char *buf)
838 {
839 	ssize_t ret = -ENODATA;
840 
841 	mutex_lock(&stp_mutex);
842 	if (stpinfo_valid() && (stp_info.vbits & 0x0800))
843 		ret = sprintf(buf, "%i\n", (int) stp_info.tto);
844 	mutex_unlock(&stp_mutex);
845 	return ret;
846 }
847 
848 static DEVICE_ATTR_RO(time_offset);
849 
850 static ssize_t time_zone_offset_show(struct device *dev,
851 				struct device_attribute *attr,
852 				char *buf)
853 {
854 	ssize_t ret = -ENODATA;
855 
856 	mutex_lock(&stp_mutex);
857 	if (stpinfo_valid() && (stp_info.vbits & 0x4000))
858 		ret = sprintf(buf, "%i\n", (int)(s16) stp_info.tzo);
859 	mutex_unlock(&stp_mutex);
860 	return ret;
861 }
862 
863 static DEVICE_ATTR_RO(time_zone_offset);
864 
865 static ssize_t timing_mode_show(struct device *dev,
866 				struct device_attribute *attr,
867 				char *buf)
868 {
869 	ssize_t ret = -ENODATA;
870 
871 	mutex_lock(&stp_mutex);
872 	if (stpinfo_valid())
873 		ret = sprintf(buf, "%i\n", stp_info.tmd);
874 	mutex_unlock(&stp_mutex);
875 	return ret;
876 }
877 
878 static DEVICE_ATTR_RO(timing_mode);
879 
880 static ssize_t timing_state_show(struct device *dev,
881 				struct device_attribute *attr,
882 				char *buf)
883 {
884 	ssize_t ret = -ENODATA;
885 
886 	mutex_lock(&stp_mutex);
887 	if (stpinfo_valid())
888 		ret = sprintf(buf, "%i\n", stp_info.tst);
889 	mutex_unlock(&stp_mutex);
890 	return ret;
891 }
892 
893 static DEVICE_ATTR_RO(timing_state);
894 
895 static ssize_t online_show(struct device *dev,
896 				struct device_attribute *attr,
897 				char *buf)
898 {
899 	return sprintf(buf, "%i\n", stp_online);
900 }
901 
902 static ssize_t online_store(struct device *dev,
903 				struct device_attribute *attr,
904 				const char *buf, size_t count)
905 {
906 	unsigned int value;
907 
908 	value = simple_strtoul(buf, NULL, 0);
909 	if (value != 0 && value != 1)
910 		return -EINVAL;
911 	if (!test_bit(CLOCK_SYNC_HAS_STP, &clock_sync_flags))
912 		return -EOPNOTSUPP;
913 	mutex_lock(&stp_mutex);
914 	stp_online = value;
915 	if (stp_online)
916 		set_bit(CLOCK_SYNC_STP, &clock_sync_flags);
917 	else
918 		clear_bit(CLOCK_SYNC_STP, &clock_sync_flags);
919 	queue_work(time_sync_wq, &stp_work);
920 	mutex_unlock(&stp_mutex);
921 	return count;
922 }
923 
924 /*
925  * Can't use DEVICE_ATTR because the attribute should be named
926  * stp/online but dev_attr_online already exists in this file ..
927  */
928 static DEVICE_ATTR_RW(online);
929 
930 static struct attribute *stp_dev_attrs[] = {
931 	&dev_attr_ctn_id.attr,
932 	&dev_attr_ctn_type.attr,
933 	&dev_attr_dst_offset.attr,
934 	&dev_attr_leap_seconds.attr,
935 	&dev_attr_online.attr,
936 	&dev_attr_leap_seconds_scheduled.attr,
937 	&dev_attr_stratum.attr,
938 	&dev_attr_time_offset.attr,
939 	&dev_attr_time_zone_offset.attr,
940 	&dev_attr_timing_mode.attr,
941 	&dev_attr_timing_state.attr,
942 	NULL
943 };
944 ATTRIBUTE_GROUPS(stp_dev);
945 
946 static int __init stp_init_sysfs(void)
947 {
948 	return subsys_system_register(&stp_subsys, stp_dev_groups);
949 }
950 
951 device_initcall(stp_init_sysfs);
952