xref: /linux/arch/x86/kernel/hpet.c (revision 071bf69a0220253a44acb8b2a27f7a262b9a46bf)
1 #include <linux/clocksource.h>
2 #include <linux/clockchips.h>
3 #include <linux/interrupt.h>
4 #include <linux/export.h>
5 #include <linux/delay.h>
6 #include <linux/errno.h>
7 #include <linux/i8253.h>
8 #include <linux/slab.h>
9 #include <linux/hpet.h>
10 #include <linux/init.h>
11 #include <linux/cpu.h>
12 #include <linux/pm.h>
13 #include <linux/io.h>
14 
15 #include <asm/cpufeature.h>
16 #include <asm/irqdomain.h>
17 #include <asm/fixmap.h>
18 #include <asm/hpet.h>
19 #include <asm/time.h>
20 
21 #define HPET_MASK			CLOCKSOURCE_MASK(32)
22 
23 /* FSEC = 10^-15
24    NSEC = 10^-9 */
25 #define FSEC_PER_NSEC			1000000L
26 
27 #define HPET_DEV_USED_BIT		2
28 #define HPET_DEV_USED			(1 << HPET_DEV_USED_BIT)
29 #define HPET_DEV_VALID			0x8
30 #define HPET_DEV_FSB_CAP		0x1000
31 #define HPET_DEV_PERI_CAP		0x2000
32 
33 #define HPET_MIN_CYCLES			128
34 #define HPET_MIN_PROG_DELTA		(HPET_MIN_CYCLES + (HPET_MIN_CYCLES >> 1))
35 
36 /*
37  * HPET address is set in acpi/boot.c, when an ACPI entry exists
38  */
39 unsigned long				hpet_address;
40 u8					hpet_blockid; /* OS timer block num */
41 bool					hpet_msi_disable;
42 
43 #ifdef CONFIG_PCI_MSI
44 static unsigned int			hpet_num_timers;
45 #endif
46 static void __iomem			*hpet_virt_address;
47 
48 struct hpet_dev {
49 	struct clock_event_device	evt;
50 	unsigned int			num;
51 	int				cpu;
52 	unsigned int			irq;
53 	unsigned int			flags;
54 	char				name[10];
55 };
56 
57 static inline struct hpet_dev *EVT_TO_HPET_DEV(struct clock_event_device *evtdev)
58 {
59 	return container_of(evtdev, struct hpet_dev, evt);
60 }
61 
62 inline unsigned int hpet_readl(unsigned int a)
63 {
64 	return readl(hpet_virt_address + a);
65 }
66 
67 static inline void hpet_writel(unsigned int d, unsigned int a)
68 {
69 	writel(d, hpet_virt_address + a);
70 }
71 
72 #ifdef CONFIG_X86_64
73 #include <asm/pgtable.h>
74 #endif
75 
76 static inline void hpet_set_mapping(void)
77 {
78 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
79 }
80 
81 static inline void hpet_clear_mapping(void)
82 {
83 	iounmap(hpet_virt_address);
84 	hpet_virt_address = NULL;
85 }
86 
87 /*
88  * HPET command line enable / disable
89  */
90 bool boot_hpet_disable;
91 bool hpet_force_user;
92 static bool hpet_verbose;
93 
94 static int __init hpet_setup(char *str)
95 {
96 	while (str) {
97 		char *next = strchr(str, ',');
98 
99 		if (next)
100 			*next++ = 0;
101 		if (!strncmp("disable", str, 7))
102 			boot_hpet_disable = true;
103 		if (!strncmp("force", str, 5))
104 			hpet_force_user = true;
105 		if (!strncmp("verbose", str, 7))
106 			hpet_verbose = true;
107 		str = next;
108 	}
109 	return 1;
110 }
111 __setup("hpet=", hpet_setup);
112 
113 static int __init disable_hpet(char *str)
114 {
115 	boot_hpet_disable = true;
116 	return 1;
117 }
118 __setup("nohpet", disable_hpet);
119 
120 static inline int is_hpet_capable(void)
121 {
122 	return !boot_hpet_disable && hpet_address;
123 }
124 
125 /*
126  * HPET timer interrupt enable / disable
127  */
128 static bool hpet_legacy_int_enabled;
129 
130 /**
131  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
132  */
133 int is_hpet_enabled(void)
134 {
135 	return is_hpet_capable() && hpet_legacy_int_enabled;
136 }
137 EXPORT_SYMBOL_GPL(is_hpet_enabled);
138 
139 static void _hpet_print_config(const char *function, int line)
140 {
141 	u32 i, timers, l, h;
142 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
143 	l = hpet_readl(HPET_ID);
144 	h = hpet_readl(HPET_PERIOD);
145 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
146 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
147 	l = hpet_readl(HPET_CFG);
148 	h = hpet_readl(HPET_STATUS);
149 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
150 	l = hpet_readl(HPET_COUNTER);
151 	h = hpet_readl(HPET_COUNTER+4);
152 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
153 
154 	for (i = 0; i < timers; i++) {
155 		l = hpet_readl(HPET_Tn_CFG(i));
156 		h = hpet_readl(HPET_Tn_CFG(i)+4);
157 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
158 		       i, l, h);
159 		l = hpet_readl(HPET_Tn_CMP(i));
160 		h = hpet_readl(HPET_Tn_CMP(i)+4);
161 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
162 		       i, l, h);
163 		l = hpet_readl(HPET_Tn_ROUTE(i));
164 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
165 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
166 		       i, l, h);
167 	}
168 }
169 
170 #define hpet_print_config()					\
171 do {								\
172 	if (hpet_verbose)					\
173 		_hpet_print_config(__func__, __LINE__);	\
174 } while (0)
175 
176 /*
177  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
178  * timer 0 and timer 1 in case of RTC emulation.
179  */
180 #ifdef CONFIG_HPET
181 
182 static void hpet_reserve_msi_timers(struct hpet_data *hd);
183 
184 static void hpet_reserve_platform_timers(unsigned int id)
185 {
186 	struct hpet __iomem *hpet = hpet_virt_address;
187 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
188 	unsigned int nrtimers, i;
189 	struct hpet_data hd;
190 
191 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
192 
193 	memset(&hd, 0, sizeof(hd));
194 	hd.hd_phys_address	= hpet_address;
195 	hd.hd_address		= hpet;
196 	hd.hd_nirqs		= nrtimers;
197 	hpet_reserve_timer(&hd, 0);
198 
199 #ifdef CONFIG_HPET_EMULATE_RTC
200 	hpet_reserve_timer(&hd, 1);
201 #endif
202 
203 	/*
204 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
205 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
206 	 * don't bother configuring *any* comparator interrupts.
207 	 */
208 	hd.hd_irq[0] = HPET_LEGACY_8254;
209 	hd.hd_irq[1] = HPET_LEGACY_RTC;
210 
211 	for (i = 2; i < nrtimers; timer++, i++) {
212 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
213 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
214 	}
215 
216 	hpet_reserve_msi_timers(&hd);
217 
218 	hpet_alloc(&hd);
219 
220 }
221 #else
222 static void hpet_reserve_platform_timers(unsigned int id) { }
223 #endif
224 
225 /*
226  * Common hpet info
227  */
228 static unsigned long hpet_freq;
229 
230 static struct clock_event_device hpet_clockevent;
231 
232 static void hpet_stop_counter(void)
233 {
234 	u32 cfg = hpet_readl(HPET_CFG);
235 	cfg &= ~HPET_CFG_ENABLE;
236 	hpet_writel(cfg, HPET_CFG);
237 }
238 
239 static void hpet_reset_counter(void)
240 {
241 	hpet_writel(0, HPET_COUNTER);
242 	hpet_writel(0, HPET_COUNTER + 4);
243 }
244 
245 static void hpet_start_counter(void)
246 {
247 	unsigned int cfg = hpet_readl(HPET_CFG);
248 	cfg |= HPET_CFG_ENABLE;
249 	hpet_writel(cfg, HPET_CFG);
250 }
251 
252 static void hpet_restart_counter(void)
253 {
254 	hpet_stop_counter();
255 	hpet_reset_counter();
256 	hpet_start_counter();
257 }
258 
259 static void hpet_resume_device(void)
260 {
261 	force_hpet_resume();
262 }
263 
264 static void hpet_resume_counter(struct clocksource *cs)
265 {
266 	hpet_resume_device();
267 	hpet_restart_counter();
268 }
269 
270 static void hpet_enable_legacy_int(void)
271 {
272 	unsigned int cfg = hpet_readl(HPET_CFG);
273 
274 	cfg |= HPET_CFG_LEGACY;
275 	hpet_writel(cfg, HPET_CFG);
276 	hpet_legacy_int_enabled = true;
277 }
278 
279 static void hpet_legacy_clockevent_register(void)
280 {
281 	/* Start HPET legacy interrupts */
282 	hpet_enable_legacy_int();
283 
284 	/*
285 	 * Start hpet with the boot cpu mask and make it
286 	 * global after the IO_APIC has been initialized.
287 	 */
288 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
289 	clockevents_config_and_register(&hpet_clockevent, hpet_freq,
290 					HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
291 	global_clock_event = &hpet_clockevent;
292 	printk(KERN_DEBUG "hpet clockevent registered\n");
293 }
294 
295 static int hpet_set_periodic(struct clock_event_device *evt, int timer)
296 {
297 	unsigned int cfg, cmp, now;
298 	uint64_t delta;
299 
300 	hpet_stop_counter();
301 	delta = ((uint64_t)(NSEC_PER_SEC / HZ)) * evt->mult;
302 	delta >>= evt->shift;
303 	now = hpet_readl(HPET_COUNTER);
304 	cmp = now + (unsigned int)delta;
305 	cfg = hpet_readl(HPET_Tn_CFG(timer));
306 	cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC | HPET_TN_SETVAL |
307 	       HPET_TN_32BIT;
308 	hpet_writel(cfg, HPET_Tn_CFG(timer));
309 	hpet_writel(cmp, HPET_Tn_CMP(timer));
310 	udelay(1);
311 	/*
312 	 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
313 	 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
314 	 * bit is automatically cleared after the first write.
315 	 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
316 	 * Publication # 24674)
317 	 */
318 	hpet_writel((unsigned int)delta, HPET_Tn_CMP(timer));
319 	hpet_start_counter();
320 	hpet_print_config();
321 
322 	return 0;
323 }
324 
325 static int hpet_set_oneshot(struct clock_event_device *evt, int timer)
326 {
327 	unsigned int cfg;
328 
329 	cfg = hpet_readl(HPET_Tn_CFG(timer));
330 	cfg &= ~HPET_TN_PERIODIC;
331 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
332 	hpet_writel(cfg, HPET_Tn_CFG(timer));
333 
334 	return 0;
335 }
336 
337 static int hpet_shutdown(struct clock_event_device *evt, int timer)
338 {
339 	unsigned int cfg;
340 
341 	cfg = hpet_readl(HPET_Tn_CFG(timer));
342 	cfg &= ~HPET_TN_ENABLE;
343 	hpet_writel(cfg, HPET_Tn_CFG(timer));
344 
345 	return 0;
346 }
347 
348 static int hpet_resume(struct clock_event_device *evt, int timer)
349 {
350 	if (!timer) {
351 		hpet_enable_legacy_int();
352 	} else {
353 		struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
354 
355 		irq_domain_activate_irq(irq_get_irq_data(hdev->irq));
356 		disable_irq(hdev->irq);
357 		irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
358 		enable_irq(hdev->irq);
359 	}
360 	hpet_print_config();
361 
362 	return 0;
363 }
364 
365 static int hpet_next_event(unsigned long delta,
366 			   struct clock_event_device *evt, int timer)
367 {
368 	u32 cnt;
369 	s32 res;
370 
371 	cnt = hpet_readl(HPET_COUNTER);
372 	cnt += (u32) delta;
373 	hpet_writel(cnt, HPET_Tn_CMP(timer));
374 
375 	/*
376 	 * HPETs are a complete disaster. The compare register is
377 	 * based on a equal comparison and neither provides a less
378 	 * than or equal functionality (which would require to take
379 	 * the wraparound into account) nor a simple count down event
380 	 * mode. Further the write to the comparator register is
381 	 * delayed internally up to two HPET clock cycles in certain
382 	 * chipsets (ATI, ICH9,10). Some newer AMD chipsets have even
383 	 * longer delays. We worked around that by reading back the
384 	 * compare register, but that required another workaround for
385 	 * ICH9,10 chips where the first readout after write can
386 	 * return the old stale value. We already had a minimum
387 	 * programming delta of 5us enforced, but a NMI or SMI hitting
388 	 * between the counter readout and the comparator write can
389 	 * move us behind that point easily. Now instead of reading
390 	 * the compare register back several times, we make the ETIME
391 	 * decision based on the following: Return ETIME if the
392 	 * counter value after the write is less than HPET_MIN_CYCLES
393 	 * away from the event or if the counter is already ahead of
394 	 * the event. The minimum programming delta for the generic
395 	 * clockevents code is set to 1.5 * HPET_MIN_CYCLES.
396 	 */
397 	res = (s32)(cnt - hpet_readl(HPET_COUNTER));
398 
399 	return res < HPET_MIN_CYCLES ? -ETIME : 0;
400 }
401 
402 static int hpet_legacy_shutdown(struct clock_event_device *evt)
403 {
404 	return hpet_shutdown(evt, 0);
405 }
406 
407 static int hpet_legacy_set_oneshot(struct clock_event_device *evt)
408 {
409 	return hpet_set_oneshot(evt, 0);
410 }
411 
412 static int hpet_legacy_set_periodic(struct clock_event_device *evt)
413 {
414 	return hpet_set_periodic(evt, 0);
415 }
416 
417 static int hpet_legacy_resume(struct clock_event_device *evt)
418 {
419 	return hpet_resume(evt, 0);
420 }
421 
422 static int hpet_legacy_next_event(unsigned long delta,
423 			struct clock_event_device *evt)
424 {
425 	return hpet_next_event(delta, evt, 0);
426 }
427 
428 /*
429  * The hpet clock event device
430  */
431 static struct clock_event_device hpet_clockevent = {
432 	.name			= "hpet",
433 	.features		= CLOCK_EVT_FEAT_PERIODIC |
434 				  CLOCK_EVT_FEAT_ONESHOT,
435 	.set_state_periodic	= hpet_legacy_set_periodic,
436 	.set_state_oneshot	= hpet_legacy_set_oneshot,
437 	.set_state_shutdown	= hpet_legacy_shutdown,
438 	.tick_resume		= hpet_legacy_resume,
439 	.set_next_event		= hpet_legacy_next_event,
440 	.irq			= 0,
441 	.rating			= 50,
442 };
443 
444 /*
445  * HPET MSI Support
446  */
447 #ifdef CONFIG_PCI_MSI
448 
449 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
450 static struct hpet_dev	*hpet_devs;
451 static struct irq_domain *hpet_domain;
452 
453 void hpet_msi_unmask(struct irq_data *data)
454 {
455 	struct hpet_dev *hdev = irq_data_get_irq_handler_data(data);
456 	unsigned int cfg;
457 
458 	/* unmask it */
459 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
460 	cfg |= HPET_TN_ENABLE | HPET_TN_FSB;
461 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
462 }
463 
464 void hpet_msi_mask(struct irq_data *data)
465 {
466 	struct hpet_dev *hdev = irq_data_get_irq_handler_data(data);
467 	unsigned int cfg;
468 
469 	/* mask it */
470 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
471 	cfg &= ~(HPET_TN_ENABLE | HPET_TN_FSB);
472 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
473 }
474 
475 void hpet_msi_write(struct hpet_dev *hdev, struct msi_msg *msg)
476 {
477 	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
478 	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
479 }
480 
481 void hpet_msi_read(struct hpet_dev *hdev, struct msi_msg *msg)
482 {
483 	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
484 	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
485 	msg->address_hi = 0;
486 }
487 
488 static int hpet_msi_shutdown(struct clock_event_device *evt)
489 {
490 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
491 
492 	return hpet_shutdown(evt, hdev->num);
493 }
494 
495 static int hpet_msi_set_oneshot(struct clock_event_device *evt)
496 {
497 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
498 
499 	return hpet_set_oneshot(evt, hdev->num);
500 }
501 
502 static int hpet_msi_set_periodic(struct clock_event_device *evt)
503 {
504 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
505 
506 	return hpet_set_periodic(evt, hdev->num);
507 }
508 
509 static int hpet_msi_resume(struct clock_event_device *evt)
510 {
511 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
512 
513 	return hpet_resume(evt, hdev->num);
514 }
515 
516 static int hpet_msi_next_event(unsigned long delta,
517 				struct clock_event_device *evt)
518 {
519 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
520 	return hpet_next_event(delta, evt, hdev->num);
521 }
522 
523 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
524 {
525 	struct hpet_dev *dev = (struct hpet_dev *)data;
526 	struct clock_event_device *hevt = &dev->evt;
527 
528 	if (!hevt->event_handler) {
529 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
530 				dev->num);
531 		return IRQ_HANDLED;
532 	}
533 
534 	hevt->event_handler(hevt);
535 	return IRQ_HANDLED;
536 }
537 
538 static int hpet_setup_irq(struct hpet_dev *dev)
539 {
540 
541 	if (request_irq(dev->irq, hpet_interrupt_handler,
542 			IRQF_TIMER | IRQF_NOBALANCING,
543 			dev->name, dev))
544 		return -1;
545 
546 	disable_irq(dev->irq);
547 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
548 	enable_irq(dev->irq);
549 
550 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
551 			 dev->name, dev->irq);
552 
553 	return 0;
554 }
555 
556 /* This should be called in specific @cpu */
557 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
558 {
559 	struct clock_event_device *evt = &hdev->evt;
560 
561 	WARN_ON(cpu != smp_processor_id());
562 	if (!(hdev->flags & HPET_DEV_VALID))
563 		return;
564 
565 	hdev->cpu = cpu;
566 	per_cpu(cpu_hpet_dev, cpu) = hdev;
567 	evt->name = hdev->name;
568 	hpet_setup_irq(hdev);
569 	evt->irq = hdev->irq;
570 
571 	evt->rating = 110;
572 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
573 	if (hdev->flags & HPET_DEV_PERI_CAP) {
574 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
575 		evt->set_state_periodic = hpet_msi_set_periodic;
576 	}
577 
578 	evt->set_state_shutdown = hpet_msi_shutdown;
579 	evt->set_state_oneshot = hpet_msi_set_oneshot;
580 	evt->tick_resume = hpet_msi_resume;
581 	evt->set_next_event = hpet_msi_next_event;
582 	evt->cpumask = cpumask_of(hdev->cpu);
583 
584 	clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
585 					0x7FFFFFFF);
586 }
587 
588 #ifdef CONFIG_HPET
589 /* Reserve at least one timer for userspace (/dev/hpet) */
590 #define RESERVE_TIMERS 1
591 #else
592 #define RESERVE_TIMERS 0
593 #endif
594 
595 static void hpet_msi_capability_lookup(unsigned int start_timer)
596 {
597 	unsigned int id;
598 	unsigned int num_timers;
599 	unsigned int num_timers_used = 0;
600 	int i, irq;
601 
602 	if (hpet_msi_disable)
603 		return;
604 
605 	if (boot_cpu_has(X86_FEATURE_ARAT))
606 		return;
607 	id = hpet_readl(HPET_ID);
608 
609 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
610 	num_timers++; /* Value read out starts from 0 */
611 	hpet_print_config();
612 
613 	hpet_domain = hpet_create_irq_domain(hpet_blockid);
614 	if (!hpet_domain)
615 		return;
616 
617 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
618 	if (!hpet_devs)
619 		return;
620 
621 	hpet_num_timers = num_timers;
622 
623 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
624 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
625 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
626 
627 		/* Only consider HPET timer with MSI support */
628 		if (!(cfg & HPET_TN_FSB_CAP))
629 			continue;
630 
631 		hdev->flags = 0;
632 		if (cfg & HPET_TN_PERIODIC_CAP)
633 			hdev->flags |= HPET_DEV_PERI_CAP;
634 		sprintf(hdev->name, "hpet%d", i);
635 		hdev->num = i;
636 
637 		irq = hpet_assign_irq(hpet_domain, hdev, hdev->num);
638 		if (irq <= 0)
639 			continue;
640 
641 		hdev->irq = irq;
642 		hdev->flags |= HPET_DEV_FSB_CAP;
643 		hdev->flags |= HPET_DEV_VALID;
644 		num_timers_used++;
645 		if (num_timers_used == num_possible_cpus())
646 			break;
647 	}
648 
649 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
650 		num_timers, num_timers_used);
651 }
652 
653 #ifdef CONFIG_HPET
654 static void hpet_reserve_msi_timers(struct hpet_data *hd)
655 {
656 	int i;
657 
658 	if (!hpet_devs)
659 		return;
660 
661 	for (i = 0; i < hpet_num_timers; i++) {
662 		struct hpet_dev *hdev = &hpet_devs[i];
663 
664 		if (!(hdev->flags & HPET_DEV_VALID))
665 			continue;
666 
667 		hd->hd_irq[hdev->num] = hdev->irq;
668 		hpet_reserve_timer(hd, hdev->num);
669 	}
670 }
671 #endif
672 
673 static struct hpet_dev *hpet_get_unused_timer(void)
674 {
675 	int i;
676 
677 	if (!hpet_devs)
678 		return NULL;
679 
680 	for (i = 0; i < hpet_num_timers; i++) {
681 		struct hpet_dev *hdev = &hpet_devs[i];
682 
683 		if (!(hdev->flags & HPET_DEV_VALID))
684 			continue;
685 		if (test_and_set_bit(HPET_DEV_USED_BIT,
686 			(unsigned long *)&hdev->flags))
687 			continue;
688 		return hdev;
689 	}
690 	return NULL;
691 }
692 
693 struct hpet_work_struct {
694 	struct delayed_work work;
695 	struct completion complete;
696 };
697 
698 static void hpet_work(struct work_struct *w)
699 {
700 	struct hpet_dev *hdev;
701 	int cpu = smp_processor_id();
702 	struct hpet_work_struct *hpet_work;
703 
704 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
705 
706 	hdev = hpet_get_unused_timer();
707 	if (hdev)
708 		init_one_hpet_msi_clockevent(hdev, cpu);
709 
710 	complete(&hpet_work->complete);
711 }
712 
713 static int hpet_cpuhp_online(unsigned int cpu)
714 {
715 	struct hpet_work_struct work;
716 
717 	INIT_DELAYED_WORK_ONSTACK(&work.work, hpet_work);
718 	init_completion(&work.complete);
719 	/* FIXME: add schedule_work_on() */
720 	schedule_delayed_work_on(cpu, &work.work, 0);
721 	wait_for_completion(&work.complete);
722 	destroy_delayed_work_on_stack(&work.work);
723 	return 0;
724 }
725 
726 static int hpet_cpuhp_dead(unsigned int cpu)
727 {
728 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
729 
730 	if (!hdev)
731 		return 0;
732 	free_irq(hdev->irq, hdev);
733 	hdev->flags &= ~HPET_DEV_USED;
734 	per_cpu(cpu_hpet_dev, cpu) = NULL;
735 	return 0;
736 }
737 #else
738 
739 static void hpet_msi_capability_lookup(unsigned int start_timer)
740 {
741 	return;
742 }
743 
744 #ifdef CONFIG_HPET
745 static void hpet_reserve_msi_timers(struct hpet_data *hd)
746 {
747 	return;
748 }
749 #endif
750 
751 #define hpet_cpuhp_online	NULL
752 #define hpet_cpuhp_dead		NULL
753 
754 #endif
755 
756 /*
757  * Clock source related code
758  */
759 static cycle_t read_hpet(struct clocksource *cs)
760 {
761 	return (cycle_t)hpet_readl(HPET_COUNTER);
762 }
763 
764 static struct clocksource clocksource_hpet = {
765 	.name		= "hpet",
766 	.rating		= 250,
767 	.read		= read_hpet,
768 	.mask		= HPET_MASK,
769 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
770 	.resume		= hpet_resume_counter,
771 };
772 
773 static int hpet_clocksource_register(void)
774 {
775 	u64 start, now;
776 	cycle_t t1;
777 
778 	/* Start the counter */
779 	hpet_restart_counter();
780 
781 	/* Verify whether hpet counter works */
782 	t1 = hpet_readl(HPET_COUNTER);
783 	start = rdtsc();
784 
785 	/*
786 	 * We don't know the TSC frequency yet, but waiting for
787 	 * 200000 TSC cycles is safe:
788 	 * 4 GHz == 50us
789 	 * 1 GHz == 200us
790 	 */
791 	do {
792 		rep_nop();
793 		now = rdtsc();
794 	} while ((now - start) < 200000UL);
795 
796 	if (t1 == hpet_readl(HPET_COUNTER)) {
797 		printk(KERN_WARNING
798 		       "HPET counter not counting. HPET disabled\n");
799 		return -ENODEV;
800 	}
801 
802 	clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
803 	return 0;
804 }
805 
806 static u32 *hpet_boot_cfg;
807 
808 /**
809  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
810  */
811 int __init hpet_enable(void)
812 {
813 	u32 hpet_period, cfg, id;
814 	u64 freq;
815 	unsigned int i, last;
816 
817 	if (!is_hpet_capable())
818 		return 0;
819 
820 	hpet_set_mapping();
821 
822 	/*
823 	 * Read the period and check for a sane value:
824 	 */
825 	hpet_period = hpet_readl(HPET_PERIOD);
826 
827 	/*
828 	 * AMD SB700 based systems with spread spectrum enabled use a
829 	 * SMM based HPET emulation to provide proper frequency
830 	 * setting. The SMM code is initialized with the first HPET
831 	 * register access and takes some time to complete. During
832 	 * this time the config register reads 0xffffffff. We check
833 	 * for max. 1000 loops whether the config register reads a non
834 	 * 0xffffffff value to make sure that HPET is up and running
835 	 * before we go further. A counting loop is safe, as the HPET
836 	 * access takes thousands of CPU cycles. On non SB700 based
837 	 * machines this check is only done once and has no side
838 	 * effects.
839 	 */
840 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
841 		if (i == 1000) {
842 			printk(KERN_WARNING
843 			       "HPET config register value = 0xFFFFFFFF. "
844 			       "Disabling HPET\n");
845 			goto out_nohpet;
846 		}
847 	}
848 
849 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
850 		goto out_nohpet;
851 
852 	/*
853 	 * The period is a femto seconds value. Convert it to a
854 	 * frequency.
855 	 */
856 	freq = FSEC_PER_SEC;
857 	do_div(freq, hpet_period);
858 	hpet_freq = freq;
859 
860 	/*
861 	 * Read the HPET ID register to retrieve the IRQ routing
862 	 * information and the number of channels
863 	 */
864 	id = hpet_readl(HPET_ID);
865 	hpet_print_config();
866 
867 	last = (id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT;
868 
869 #ifdef CONFIG_HPET_EMULATE_RTC
870 	/*
871 	 * The legacy routing mode needs at least two channels, tick timer
872 	 * and the rtc emulation channel.
873 	 */
874 	if (!last)
875 		goto out_nohpet;
876 #endif
877 
878 	cfg = hpet_readl(HPET_CFG);
879 	hpet_boot_cfg = kmalloc((last + 2) * sizeof(*hpet_boot_cfg),
880 				GFP_KERNEL);
881 	if (hpet_boot_cfg)
882 		*hpet_boot_cfg = cfg;
883 	else
884 		pr_warn("HPET initial state will not be saved\n");
885 	cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
886 	hpet_writel(cfg, HPET_CFG);
887 	if (cfg)
888 		pr_warn("HPET: Unrecognized bits %#x set in global cfg\n",
889 			cfg);
890 
891 	for (i = 0; i <= last; ++i) {
892 		cfg = hpet_readl(HPET_Tn_CFG(i));
893 		if (hpet_boot_cfg)
894 			hpet_boot_cfg[i + 1] = cfg;
895 		cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
896 		hpet_writel(cfg, HPET_Tn_CFG(i));
897 		cfg &= ~(HPET_TN_PERIODIC | HPET_TN_PERIODIC_CAP
898 			 | HPET_TN_64BIT_CAP | HPET_TN_32BIT | HPET_TN_ROUTE
899 			 | HPET_TN_FSB | HPET_TN_FSB_CAP);
900 		if (cfg)
901 			pr_warn("HPET: Unrecognized bits %#x set in cfg#%u\n",
902 				cfg, i);
903 	}
904 	hpet_print_config();
905 
906 	if (hpet_clocksource_register())
907 		goto out_nohpet;
908 
909 	if (id & HPET_ID_LEGSUP) {
910 		hpet_legacy_clockevent_register();
911 		return 1;
912 	}
913 	return 0;
914 
915 out_nohpet:
916 	hpet_clear_mapping();
917 	hpet_address = 0;
918 	return 0;
919 }
920 
921 /*
922  * Needs to be late, as the reserve_timer code calls kalloc !
923  *
924  * Not a problem on i386 as hpet_enable is called from late_time_init,
925  * but on x86_64 it is necessary !
926  */
927 static __init int hpet_late_init(void)
928 {
929 	int ret;
930 
931 	if (boot_hpet_disable)
932 		return -ENODEV;
933 
934 	if (!hpet_address) {
935 		if (!force_hpet_address)
936 			return -ENODEV;
937 
938 		hpet_address = force_hpet_address;
939 		hpet_enable();
940 	}
941 
942 	if (!hpet_virt_address)
943 		return -ENODEV;
944 
945 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
946 		hpet_msi_capability_lookup(2);
947 	else
948 		hpet_msi_capability_lookup(0);
949 
950 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
951 	hpet_print_config();
952 
953 	if (hpet_msi_disable)
954 		return 0;
955 
956 	if (boot_cpu_has(X86_FEATURE_ARAT))
957 		return 0;
958 
959 	/* This notifier should be called after workqueue is ready */
960 	ret = cpuhp_setup_state(CPUHP_AP_X86_HPET_ONLINE, "AP_X86_HPET_ONLINE",
961 				hpet_cpuhp_online, NULL);
962 	if (ret)
963 		return ret;
964 	ret = cpuhp_setup_state(CPUHP_X86_HPET_DEAD, "X86_HPET_DEAD", NULL,
965 				hpet_cpuhp_dead);
966 	if (ret)
967 		goto err_cpuhp;
968 	return 0;
969 
970 err_cpuhp:
971 	cpuhp_remove_state(CPUHP_AP_X86_HPET_ONLINE);
972 	return ret;
973 }
974 fs_initcall(hpet_late_init);
975 
976 void hpet_disable(void)
977 {
978 	if (is_hpet_capable() && hpet_virt_address) {
979 		unsigned int cfg = hpet_readl(HPET_CFG), id, last;
980 
981 		if (hpet_boot_cfg)
982 			cfg = *hpet_boot_cfg;
983 		else if (hpet_legacy_int_enabled) {
984 			cfg &= ~HPET_CFG_LEGACY;
985 			hpet_legacy_int_enabled = false;
986 		}
987 		cfg &= ~HPET_CFG_ENABLE;
988 		hpet_writel(cfg, HPET_CFG);
989 
990 		if (!hpet_boot_cfg)
991 			return;
992 
993 		id = hpet_readl(HPET_ID);
994 		last = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
995 
996 		for (id = 0; id <= last; ++id)
997 			hpet_writel(hpet_boot_cfg[id + 1], HPET_Tn_CFG(id));
998 
999 		if (*hpet_boot_cfg & HPET_CFG_ENABLE)
1000 			hpet_writel(*hpet_boot_cfg, HPET_CFG);
1001 	}
1002 }
1003 
1004 #ifdef CONFIG_HPET_EMULATE_RTC
1005 
1006 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
1007  * is enabled, we support RTC interrupt functionality in software.
1008  * RTC has 3 kinds of interrupts:
1009  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
1010  *    is updated
1011  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
1012  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
1013  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
1014  * (1) and (2) above are implemented using polling at a frequency of
1015  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
1016  * overhead. (DEFAULT_RTC_INT_FREQ)
1017  * For (3), we use interrupts at 64Hz or user specified periodic
1018  * frequency, whichever is higher.
1019  */
1020 #include <linux/mc146818rtc.h>
1021 #include <linux/rtc.h>
1022 
1023 #define DEFAULT_RTC_INT_FREQ	64
1024 #define DEFAULT_RTC_SHIFT	6
1025 #define RTC_NUM_INTS		1
1026 
1027 static unsigned long hpet_rtc_flags;
1028 static int hpet_prev_update_sec;
1029 static struct rtc_time hpet_alarm_time;
1030 static unsigned long hpet_pie_count;
1031 static u32 hpet_t1_cmp;
1032 static u32 hpet_default_delta;
1033 static u32 hpet_pie_delta;
1034 static unsigned long hpet_pie_limit;
1035 
1036 static rtc_irq_handler irq_handler;
1037 
1038 /*
1039  * Check that the hpet counter c1 is ahead of the c2
1040  */
1041 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1042 {
1043 	return (s32)(c2 - c1) < 0;
1044 }
1045 
1046 /*
1047  * Registers a IRQ handler.
1048  */
1049 int hpet_register_irq_handler(rtc_irq_handler handler)
1050 {
1051 	if (!is_hpet_enabled())
1052 		return -ENODEV;
1053 	if (irq_handler)
1054 		return -EBUSY;
1055 
1056 	irq_handler = handler;
1057 
1058 	return 0;
1059 }
1060 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1061 
1062 /*
1063  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1064  * and does cleanup.
1065  */
1066 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1067 {
1068 	if (!is_hpet_enabled())
1069 		return;
1070 
1071 	irq_handler = NULL;
1072 	hpet_rtc_flags = 0;
1073 }
1074 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1075 
1076 /*
1077  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1078  * is not supported by all HPET implementations for timer 1.
1079  *
1080  * hpet_rtc_timer_init() is called when the rtc is initialized.
1081  */
1082 int hpet_rtc_timer_init(void)
1083 {
1084 	unsigned int cfg, cnt, delta;
1085 	unsigned long flags;
1086 
1087 	if (!is_hpet_enabled())
1088 		return 0;
1089 
1090 	if (!hpet_default_delta) {
1091 		uint64_t clc;
1092 
1093 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1094 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1095 		hpet_default_delta = clc;
1096 	}
1097 
1098 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1099 		delta = hpet_default_delta;
1100 	else
1101 		delta = hpet_pie_delta;
1102 
1103 	local_irq_save(flags);
1104 
1105 	cnt = delta + hpet_readl(HPET_COUNTER);
1106 	hpet_writel(cnt, HPET_T1_CMP);
1107 	hpet_t1_cmp = cnt;
1108 
1109 	cfg = hpet_readl(HPET_T1_CFG);
1110 	cfg &= ~HPET_TN_PERIODIC;
1111 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1112 	hpet_writel(cfg, HPET_T1_CFG);
1113 
1114 	local_irq_restore(flags);
1115 
1116 	return 1;
1117 }
1118 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1119 
1120 static void hpet_disable_rtc_channel(void)
1121 {
1122 	u32 cfg = hpet_readl(HPET_T1_CFG);
1123 	cfg &= ~HPET_TN_ENABLE;
1124 	hpet_writel(cfg, HPET_T1_CFG);
1125 }
1126 
1127 /*
1128  * The functions below are called from rtc driver.
1129  * Return 0 if HPET is not being used.
1130  * Otherwise do the necessary changes and return 1.
1131  */
1132 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1133 {
1134 	if (!is_hpet_enabled())
1135 		return 0;
1136 
1137 	hpet_rtc_flags &= ~bit_mask;
1138 	if (unlikely(!hpet_rtc_flags))
1139 		hpet_disable_rtc_channel();
1140 
1141 	return 1;
1142 }
1143 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1144 
1145 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1146 {
1147 	unsigned long oldbits = hpet_rtc_flags;
1148 
1149 	if (!is_hpet_enabled())
1150 		return 0;
1151 
1152 	hpet_rtc_flags |= bit_mask;
1153 
1154 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1155 		hpet_prev_update_sec = -1;
1156 
1157 	if (!oldbits)
1158 		hpet_rtc_timer_init();
1159 
1160 	return 1;
1161 }
1162 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1163 
1164 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1165 			unsigned char sec)
1166 {
1167 	if (!is_hpet_enabled())
1168 		return 0;
1169 
1170 	hpet_alarm_time.tm_hour = hrs;
1171 	hpet_alarm_time.tm_min = min;
1172 	hpet_alarm_time.tm_sec = sec;
1173 
1174 	return 1;
1175 }
1176 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1177 
1178 int hpet_set_periodic_freq(unsigned long freq)
1179 {
1180 	uint64_t clc;
1181 
1182 	if (!is_hpet_enabled())
1183 		return 0;
1184 
1185 	if (freq <= DEFAULT_RTC_INT_FREQ)
1186 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1187 	else {
1188 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1189 		do_div(clc, freq);
1190 		clc >>= hpet_clockevent.shift;
1191 		hpet_pie_delta = clc;
1192 		hpet_pie_limit = 0;
1193 	}
1194 	return 1;
1195 }
1196 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1197 
1198 int hpet_rtc_dropped_irq(void)
1199 {
1200 	return is_hpet_enabled();
1201 }
1202 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1203 
1204 static void hpet_rtc_timer_reinit(void)
1205 {
1206 	unsigned int delta;
1207 	int lost_ints = -1;
1208 
1209 	if (unlikely(!hpet_rtc_flags))
1210 		hpet_disable_rtc_channel();
1211 
1212 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1213 		delta = hpet_default_delta;
1214 	else
1215 		delta = hpet_pie_delta;
1216 
1217 	/*
1218 	 * Increment the comparator value until we are ahead of the
1219 	 * current count.
1220 	 */
1221 	do {
1222 		hpet_t1_cmp += delta;
1223 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1224 		lost_ints++;
1225 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1226 
1227 	if (lost_ints) {
1228 		if (hpet_rtc_flags & RTC_PIE)
1229 			hpet_pie_count += lost_ints;
1230 		if (printk_ratelimit())
1231 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1232 				lost_ints);
1233 	}
1234 }
1235 
1236 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1237 {
1238 	struct rtc_time curr_time;
1239 	unsigned long rtc_int_flag = 0;
1240 
1241 	hpet_rtc_timer_reinit();
1242 	memset(&curr_time, 0, sizeof(struct rtc_time));
1243 
1244 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1245 		mc146818_set_time(&curr_time);
1246 
1247 	if (hpet_rtc_flags & RTC_UIE &&
1248 	    curr_time.tm_sec != hpet_prev_update_sec) {
1249 		if (hpet_prev_update_sec >= 0)
1250 			rtc_int_flag = RTC_UF;
1251 		hpet_prev_update_sec = curr_time.tm_sec;
1252 	}
1253 
1254 	if (hpet_rtc_flags & RTC_PIE &&
1255 	    ++hpet_pie_count >= hpet_pie_limit) {
1256 		rtc_int_flag |= RTC_PF;
1257 		hpet_pie_count = 0;
1258 	}
1259 
1260 	if (hpet_rtc_flags & RTC_AIE &&
1261 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1262 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1263 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1264 			rtc_int_flag |= RTC_AF;
1265 
1266 	if (rtc_int_flag) {
1267 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1268 		if (irq_handler)
1269 			irq_handler(rtc_int_flag, dev_id);
1270 	}
1271 	return IRQ_HANDLED;
1272 }
1273 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1274 #endif
1275