xref: /linux/arch/x86/kernel/hpet.c (revision 8b1935e6a36b0967efc593d67ed3aebbfbc1f5b1)
1 #include <linux/clocksource.h>
2 #include <linux/clockchips.h>
3 #include <linux/interrupt.h>
4 #include <linux/sysdev.h>
5 #include <linux/delay.h>
6 #include <linux/errno.h>
7 #include <linux/hpet.h>
8 #include <linux/init.h>
9 #include <linux/cpu.h>
10 #include <linux/pm.h>
11 #include <linux/io.h>
12 
13 #include <asm/fixmap.h>
14 #include <asm/i8253.h>
15 #include <asm/hpet.h>
16 
17 #define HPET_MASK			CLOCKSOURCE_MASK(32)
18 #define HPET_SHIFT			22
19 
20 /* FSEC = 10^-15
21    NSEC = 10^-9 */
22 #define FSEC_PER_NSEC			1000000L
23 
24 #define HPET_DEV_USED_BIT		2
25 #define HPET_DEV_USED			(1 << HPET_DEV_USED_BIT)
26 #define HPET_DEV_VALID			0x8
27 #define HPET_DEV_FSB_CAP		0x1000
28 #define HPET_DEV_PERI_CAP		0x2000
29 
30 #define EVT_TO_HPET_DEV(evt) container_of(evt, struct hpet_dev, evt)
31 
32 /*
33  * HPET address is set in acpi/boot.c, when an ACPI entry exists
34  */
35 unsigned long				hpet_address;
36 u8					hpet_blockid; /* OS timer block num */
37 u8					hpet_msi_disable;
38 
39 #ifdef CONFIG_PCI_MSI
40 static unsigned long			hpet_num_timers;
41 #endif
42 static void __iomem			*hpet_virt_address;
43 
44 struct hpet_dev {
45 	struct clock_event_device	evt;
46 	unsigned int			num;
47 	int				cpu;
48 	unsigned int			irq;
49 	unsigned int			flags;
50 	char				name[10];
51 };
52 
53 inline unsigned int hpet_readl(unsigned int a)
54 {
55 	return readl(hpet_virt_address + a);
56 }
57 
58 static inline void hpet_writel(unsigned int d, unsigned int a)
59 {
60 	writel(d, hpet_virt_address + a);
61 }
62 
63 #ifdef CONFIG_X86_64
64 #include <asm/pgtable.h>
65 #endif
66 
67 static inline void hpet_set_mapping(void)
68 {
69 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
70 #ifdef CONFIG_X86_64
71 	__set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
72 #endif
73 }
74 
75 static inline void hpet_clear_mapping(void)
76 {
77 	iounmap(hpet_virt_address);
78 	hpet_virt_address = NULL;
79 }
80 
81 /*
82  * HPET command line enable / disable
83  */
84 static int boot_hpet_disable;
85 int hpet_force_user;
86 static int hpet_verbose;
87 
88 static int __init hpet_setup(char *str)
89 {
90 	if (str) {
91 		if (!strncmp("disable", str, 7))
92 			boot_hpet_disable = 1;
93 		if (!strncmp("force", str, 5))
94 			hpet_force_user = 1;
95 		if (!strncmp("verbose", str, 7))
96 			hpet_verbose = 1;
97 	}
98 	return 1;
99 }
100 __setup("hpet=", hpet_setup);
101 
102 static int __init disable_hpet(char *str)
103 {
104 	boot_hpet_disable = 1;
105 	return 1;
106 }
107 __setup("nohpet", disable_hpet);
108 
109 static inline int is_hpet_capable(void)
110 {
111 	return !boot_hpet_disable && hpet_address;
112 }
113 
114 /*
115  * HPET timer interrupt enable / disable
116  */
117 static int hpet_legacy_int_enabled;
118 
119 /**
120  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
121  */
122 int is_hpet_enabled(void)
123 {
124 	return is_hpet_capable() && hpet_legacy_int_enabled;
125 }
126 EXPORT_SYMBOL_GPL(is_hpet_enabled);
127 
128 static void _hpet_print_config(const char *function, int line)
129 {
130 	u32 i, timers, l, h;
131 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
132 	l = hpet_readl(HPET_ID);
133 	h = hpet_readl(HPET_PERIOD);
134 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
135 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
136 	l = hpet_readl(HPET_CFG);
137 	h = hpet_readl(HPET_STATUS);
138 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
139 	l = hpet_readl(HPET_COUNTER);
140 	h = hpet_readl(HPET_COUNTER+4);
141 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
142 
143 	for (i = 0; i < timers; i++) {
144 		l = hpet_readl(HPET_Tn_CFG(i));
145 		h = hpet_readl(HPET_Tn_CFG(i)+4);
146 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
147 		       i, l, h);
148 		l = hpet_readl(HPET_Tn_CMP(i));
149 		h = hpet_readl(HPET_Tn_CMP(i)+4);
150 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
151 		       i, l, h);
152 		l = hpet_readl(HPET_Tn_ROUTE(i));
153 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
154 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
155 		       i, l, h);
156 	}
157 }
158 
159 #define hpet_print_config()					\
160 do {								\
161 	if (hpet_verbose)					\
162 		_hpet_print_config(__FUNCTION__, __LINE__);	\
163 } while (0)
164 
165 /*
166  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
167  * timer 0 and timer 1 in case of RTC emulation.
168  */
169 #ifdef CONFIG_HPET
170 
171 static void hpet_reserve_msi_timers(struct hpet_data *hd);
172 
173 static void hpet_reserve_platform_timers(unsigned int id)
174 {
175 	struct hpet __iomem *hpet = hpet_virt_address;
176 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
177 	unsigned int nrtimers, i;
178 	struct hpet_data hd;
179 
180 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
181 
182 	memset(&hd, 0, sizeof(hd));
183 	hd.hd_phys_address	= hpet_address;
184 	hd.hd_address		= hpet;
185 	hd.hd_nirqs		= nrtimers;
186 	hpet_reserve_timer(&hd, 0);
187 
188 #ifdef CONFIG_HPET_EMULATE_RTC
189 	hpet_reserve_timer(&hd, 1);
190 #endif
191 
192 	/*
193 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
194 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
195 	 * don't bother configuring *any* comparator interrupts.
196 	 */
197 	hd.hd_irq[0] = HPET_LEGACY_8254;
198 	hd.hd_irq[1] = HPET_LEGACY_RTC;
199 
200 	for (i = 2; i < nrtimers; timer++, i++) {
201 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
202 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
203 	}
204 
205 	hpet_reserve_msi_timers(&hd);
206 
207 	hpet_alloc(&hd);
208 
209 }
210 #else
211 static void hpet_reserve_platform_timers(unsigned int id) { }
212 #endif
213 
214 /*
215  * Common hpet info
216  */
217 static unsigned long hpet_period;
218 
219 static void hpet_legacy_set_mode(enum clock_event_mode mode,
220 			  struct clock_event_device *evt);
221 static int hpet_legacy_next_event(unsigned long delta,
222 			   struct clock_event_device *evt);
223 
224 /*
225  * The hpet clock event device
226  */
227 static struct clock_event_device hpet_clockevent = {
228 	.name		= "hpet",
229 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
230 	.set_mode	= hpet_legacy_set_mode,
231 	.set_next_event = hpet_legacy_next_event,
232 	.shift		= 32,
233 	.irq		= 0,
234 	.rating		= 50,
235 };
236 
237 static void hpet_stop_counter(void)
238 {
239 	unsigned long cfg = hpet_readl(HPET_CFG);
240 	cfg &= ~HPET_CFG_ENABLE;
241 	hpet_writel(cfg, HPET_CFG);
242 }
243 
244 static void hpet_reset_counter(void)
245 {
246 	hpet_writel(0, HPET_COUNTER);
247 	hpet_writel(0, HPET_COUNTER + 4);
248 }
249 
250 static void hpet_start_counter(void)
251 {
252 	unsigned int cfg = hpet_readl(HPET_CFG);
253 	cfg |= HPET_CFG_ENABLE;
254 	hpet_writel(cfg, HPET_CFG);
255 }
256 
257 static void hpet_restart_counter(void)
258 {
259 	hpet_stop_counter();
260 	hpet_reset_counter();
261 	hpet_start_counter();
262 }
263 
264 static void hpet_resume_device(void)
265 {
266 	force_hpet_resume();
267 }
268 
269 static void hpet_resume_counter(struct clocksource *cs)
270 {
271 	hpet_resume_device();
272 	hpet_restart_counter();
273 }
274 
275 static void hpet_enable_legacy_int(void)
276 {
277 	unsigned int cfg = hpet_readl(HPET_CFG);
278 
279 	cfg |= HPET_CFG_LEGACY;
280 	hpet_writel(cfg, HPET_CFG);
281 	hpet_legacy_int_enabled = 1;
282 }
283 
284 static void hpet_legacy_clockevent_register(void)
285 {
286 	/* Start HPET legacy interrupts */
287 	hpet_enable_legacy_int();
288 
289 	/*
290 	 * The mult factor is defined as (include/linux/clockchips.h)
291 	 *  mult/2^shift = cyc/ns (in contrast to ns/cyc in clocksource.h)
292 	 * hpet_period is in units of femtoseconds (per cycle), so
293 	 *  mult/2^shift = cyc/ns = 10^6/hpet_period
294 	 *  mult = (10^6 * 2^shift)/hpet_period
295 	 *  mult = (FSEC_PER_NSEC << hpet_clockevent.shift)/hpet_period
296 	 */
297 	hpet_clockevent.mult = div_sc((unsigned long) FSEC_PER_NSEC,
298 				      hpet_period, hpet_clockevent.shift);
299 	/* Calculate the min / max delta */
300 	hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF,
301 							   &hpet_clockevent);
302 	/* 5 usec minimum reprogramming delta. */
303 	hpet_clockevent.min_delta_ns = 5000;
304 
305 	/*
306 	 * Start hpet with the boot cpu mask and make it
307 	 * global after the IO_APIC has been initialized.
308 	 */
309 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
310 	clockevents_register_device(&hpet_clockevent);
311 	global_clock_event = &hpet_clockevent;
312 	printk(KERN_DEBUG "hpet clockevent registered\n");
313 }
314 
315 static int hpet_setup_msi_irq(unsigned int irq);
316 
317 static void hpet_set_mode(enum clock_event_mode mode,
318 			  struct clock_event_device *evt, int timer)
319 {
320 	unsigned int cfg, cmp, now;
321 	uint64_t delta;
322 
323 	switch (mode) {
324 	case CLOCK_EVT_MODE_PERIODIC:
325 		hpet_stop_counter();
326 		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
327 		delta >>= evt->shift;
328 		now = hpet_readl(HPET_COUNTER);
329 		cmp = now + (unsigned int) delta;
330 		cfg = hpet_readl(HPET_Tn_CFG(timer));
331 		/* Make sure we use edge triggered interrupts */
332 		cfg &= ~HPET_TN_LEVEL;
333 		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
334 		       HPET_TN_SETVAL | HPET_TN_32BIT;
335 		hpet_writel(cfg, HPET_Tn_CFG(timer));
336 		hpet_writel(cmp, HPET_Tn_CMP(timer));
337 		udelay(1);
338 		/*
339 		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
340 		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
341 		 * bit is automatically cleared after the first write.
342 		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
343 		 * Publication # 24674)
344 		 */
345 		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
346 		hpet_start_counter();
347 		hpet_print_config();
348 		break;
349 
350 	case CLOCK_EVT_MODE_ONESHOT:
351 		cfg = hpet_readl(HPET_Tn_CFG(timer));
352 		cfg &= ~HPET_TN_PERIODIC;
353 		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
354 		hpet_writel(cfg, HPET_Tn_CFG(timer));
355 		break;
356 
357 	case CLOCK_EVT_MODE_UNUSED:
358 	case CLOCK_EVT_MODE_SHUTDOWN:
359 		cfg = hpet_readl(HPET_Tn_CFG(timer));
360 		cfg &= ~HPET_TN_ENABLE;
361 		hpet_writel(cfg, HPET_Tn_CFG(timer));
362 		break;
363 
364 	case CLOCK_EVT_MODE_RESUME:
365 		if (timer == 0) {
366 			hpet_enable_legacy_int();
367 		} else {
368 			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
369 			hpet_setup_msi_irq(hdev->irq);
370 			disable_irq(hdev->irq);
371 			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
372 			enable_irq(hdev->irq);
373 		}
374 		hpet_print_config();
375 		break;
376 	}
377 }
378 
379 static int hpet_next_event(unsigned long delta,
380 			   struct clock_event_device *evt, int timer)
381 {
382 	u32 cnt;
383 
384 	cnt = hpet_readl(HPET_COUNTER);
385 	cnt += (u32) delta;
386 	hpet_writel(cnt, HPET_Tn_CMP(timer));
387 
388 	/*
389 	 * We need to read back the CMP register on certain HPET
390 	 * implementations (ATI chipsets) which seem to delay the
391 	 * transfer of the compare register into the internal compare
392 	 * logic. With small deltas this might actually be too late as
393 	 * the counter could already be higher than the compare value
394 	 * at that point and we would wait for the next hpet interrupt
395 	 * forever. We found out that reading the CMP register back
396 	 * forces the transfer so we can rely on the comparison with
397 	 * the counter register below. If the read back from the
398 	 * compare register does not match the value we programmed
399 	 * then we might have a real hardware problem. We can not do
400 	 * much about it here, but at least alert the user/admin with
401 	 * a prominent warning.
402 	 */
403 	WARN_ONCE(hpet_readl(HPET_Tn_CMP(timer)) != cnt,
404 		  KERN_WARNING "hpet: compare register read back failed.\n");
405 
406 	return (s32)(hpet_readl(HPET_COUNTER) - cnt) >= 0 ? -ETIME : 0;
407 }
408 
409 static void hpet_legacy_set_mode(enum clock_event_mode mode,
410 			struct clock_event_device *evt)
411 {
412 	hpet_set_mode(mode, evt, 0);
413 }
414 
415 static int hpet_legacy_next_event(unsigned long delta,
416 			struct clock_event_device *evt)
417 {
418 	return hpet_next_event(delta, evt, 0);
419 }
420 
421 /*
422  * HPET MSI Support
423  */
424 #ifdef CONFIG_PCI_MSI
425 
426 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
427 static struct hpet_dev	*hpet_devs;
428 
429 void hpet_msi_unmask(unsigned int irq)
430 {
431 	struct hpet_dev *hdev = get_irq_data(irq);
432 	unsigned int cfg;
433 
434 	/* unmask it */
435 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
436 	cfg |= HPET_TN_FSB;
437 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
438 }
439 
440 void hpet_msi_mask(unsigned int irq)
441 {
442 	unsigned int cfg;
443 	struct hpet_dev *hdev = get_irq_data(irq);
444 
445 	/* mask it */
446 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
447 	cfg &= ~HPET_TN_FSB;
448 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
449 }
450 
451 void hpet_msi_write(unsigned int irq, struct msi_msg *msg)
452 {
453 	struct hpet_dev *hdev = get_irq_data(irq);
454 
455 	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
456 	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
457 }
458 
459 void hpet_msi_read(unsigned int irq, struct msi_msg *msg)
460 {
461 	struct hpet_dev *hdev = get_irq_data(irq);
462 
463 	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
464 	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
465 	msg->address_hi = 0;
466 }
467 
468 static void hpet_msi_set_mode(enum clock_event_mode mode,
469 				struct clock_event_device *evt)
470 {
471 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
472 	hpet_set_mode(mode, evt, hdev->num);
473 }
474 
475 static int hpet_msi_next_event(unsigned long delta,
476 				struct clock_event_device *evt)
477 {
478 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
479 	return hpet_next_event(delta, evt, hdev->num);
480 }
481 
482 static int hpet_setup_msi_irq(unsigned int irq)
483 {
484 	if (arch_setup_hpet_msi(irq, hpet_blockid)) {
485 		destroy_irq(irq);
486 		return -EINVAL;
487 	}
488 	return 0;
489 }
490 
491 static int hpet_assign_irq(struct hpet_dev *dev)
492 {
493 	unsigned int irq;
494 
495 	irq = create_irq();
496 	if (!irq)
497 		return -EINVAL;
498 
499 	set_irq_data(irq, dev);
500 
501 	if (hpet_setup_msi_irq(irq))
502 		return -EINVAL;
503 
504 	dev->irq = irq;
505 	return 0;
506 }
507 
508 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
509 {
510 	struct hpet_dev *dev = (struct hpet_dev *)data;
511 	struct clock_event_device *hevt = &dev->evt;
512 
513 	if (!hevt->event_handler) {
514 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
515 				dev->num);
516 		return IRQ_HANDLED;
517 	}
518 
519 	hevt->event_handler(hevt);
520 	return IRQ_HANDLED;
521 }
522 
523 static int hpet_setup_irq(struct hpet_dev *dev)
524 {
525 
526 	if (request_irq(dev->irq, hpet_interrupt_handler,
527 			IRQF_TIMER | IRQF_DISABLED | IRQF_NOBALANCING,
528 			dev->name, dev))
529 		return -1;
530 
531 	disable_irq(dev->irq);
532 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
533 	enable_irq(dev->irq);
534 
535 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
536 			 dev->name, dev->irq);
537 
538 	return 0;
539 }
540 
541 /* This should be called in specific @cpu */
542 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
543 {
544 	struct clock_event_device *evt = &hdev->evt;
545 	uint64_t hpet_freq;
546 
547 	WARN_ON(cpu != smp_processor_id());
548 	if (!(hdev->flags & HPET_DEV_VALID))
549 		return;
550 
551 	if (hpet_setup_msi_irq(hdev->irq))
552 		return;
553 
554 	hdev->cpu = cpu;
555 	per_cpu(cpu_hpet_dev, cpu) = hdev;
556 	evt->name = hdev->name;
557 	hpet_setup_irq(hdev);
558 	evt->irq = hdev->irq;
559 
560 	evt->rating = 110;
561 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
562 	if (hdev->flags & HPET_DEV_PERI_CAP)
563 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
564 
565 	evt->set_mode = hpet_msi_set_mode;
566 	evt->set_next_event = hpet_msi_next_event;
567 	evt->shift = 32;
568 
569 	/*
570 	 * The period is a femto seconds value. We need to calculate the
571 	 * scaled math multiplication factor for nanosecond to hpet tick
572 	 * conversion.
573 	 */
574 	hpet_freq = 1000000000000000ULL;
575 	do_div(hpet_freq, hpet_period);
576 	evt->mult = div_sc((unsigned long) hpet_freq,
577 				      NSEC_PER_SEC, evt->shift);
578 	/* Calculate the max delta */
579 	evt->max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, evt);
580 	/* 5 usec minimum reprogramming delta. */
581 	evt->min_delta_ns = 5000;
582 
583 	evt->cpumask = cpumask_of(hdev->cpu);
584 	clockevents_register_device(evt);
585 }
586 
587 #ifdef CONFIG_HPET
588 /* Reserve at least one timer for userspace (/dev/hpet) */
589 #define RESERVE_TIMERS 1
590 #else
591 #define RESERVE_TIMERS 0
592 #endif
593 
594 static void hpet_msi_capability_lookup(unsigned int start_timer)
595 {
596 	unsigned int id;
597 	unsigned int num_timers;
598 	unsigned int num_timers_used = 0;
599 	int i;
600 
601 	if (hpet_msi_disable)
602 		return;
603 
604 	if (boot_cpu_has(X86_FEATURE_ARAT))
605 		return;
606 	id = hpet_readl(HPET_ID);
607 
608 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
609 	num_timers++; /* Value read out starts from 0 */
610 	hpet_print_config();
611 
612 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
613 	if (!hpet_devs)
614 		return;
615 
616 	hpet_num_timers = num_timers;
617 
618 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
619 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
620 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
621 
622 		/* Only consider HPET timer with MSI support */
623 		if (!(cfg & HPET_TN_FSB_CAP))
624 			continue;
625 
626 		hdev->flags = 0;
627 		if (cfg & HPET_TN_PERIODIC_CAP)
628 			hdev->flags |= HPET_DEV_PERI_CAP;
629 		hdev->num = i;
630 
631 		sprintf(hdev->name, "hpet%d", i);
632 		if (hpet_assign_irq(hdev))
633 			continue;
634 
635 		hdev->flags |= HPET_DEV_FSB_CAP;
636 		hdev->flags |= HPET_DEV_VALID;
637 		num_timers_used++;
638 		if (num_timers_used == num_possible_cpus())
639 			break;
640 	}
641 
642 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
643 		num_timers, num_timers_used);
644 }
645 
646 #ifdef CONFIG_HPET
647 static void hpet_reserve_msi_timers(struct hpet_data *hd)
648 {
649 	int i;
650 
651 	if (!hpet_devs)
652 		return;
653 
654 	for (i = 0; i < hpet_num_timers; i++) {
655 		struct hpet_dev *hdev = &hpet_devs[i];
656 
657 		if (!(hdev->flags & HPET_DEV_VALID))
658 			continue;
659 
660 		hd->hd_irq[hdev->num] = hdev->irq;
661 		hpet_reserve_timer(hd, hdev->num);
662 	}
663 }
664 #endif
665 
666 static struct hpet_dev *hpet_get_unused_timer(void)
667 {
668 	int i;
669 
670 	if (!hpet_devs)
671 		return NULL;
672 
673 	for (i = 0; i < hpet_num_timers; i++) {
674 		struct hpet_dev *hdev = &hpet_devs[i];
675 
676 		if (!(hdev->flags & HPET_DEV_VALID))
677 			continue;
678 		if (test_and_set_bit(HPET_DEV_USED_BIT,
679 			(unsigned long *)&hdev->flags))
680 			continue;
681 		return hdev;
682 	}
683 	return NULL;
684 }
685 
686 struct hpet_work_struct {
687 	struct delayed_work work;
688 	struct completion complete;
689 };
690 
691 static void hpet_work(struct work_struct *w)
692 {
693 	struct hpet_dev *hdev;
694 	int cpu = smp_processor_id();
695 	struct hpet_work_struct *hpet_work;
696 
697 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
698 
699 	hdev = hpet_get_unused_timer();
700 	if (hdev)
701 		init_one_hpet_msi_clockevent(hdev, cpu);
702 
703 	complete(&hpet_work->complete);
704 }
705 
706 static int hpet_cpuhp_notify(struct notifier_block *n,
707 		unsigned long action, void *hcpu)
708 {
709 	unsigned long cpu = (unsigned long)hcpu;
710 	struct hpet_work_struct work;
711 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
712 
713 	switch (action & 0xf) {
714 	case CPU_ONLINE:
715 		INIT_DELAYED_WORK_ON_STACK(&work.work, hpet_work);
716 		init_completion(&work.complete);
717 		/* FIXME: add schedule_work_on() */
718 		schedule_delayed_work_on(cpu, &work.work, 0);
719 		wait_for_completion(&work.complete);
720 		destroy_timer_on_stack(&work.work.timer);
721 		break;
722 	case CPU_DEAD:
723 		if (hdev) {
724 			free_irq(hdev->irq, hdev);
725 			hdev->flags &= ~HPET_DEV_USED;
726 			per_cpu(cpu_hpet_dev, cpu) = NULL;
727 		}
728 		break;
729 	}
730 	return NOTIFY_OK;
731 }
732 #else
733 
734 static int hpet_setup_msi_irq(unsigned int irq)
735 {
736 	return 0;
737 }
738 static void hpet_msi_capability_lookup(unsigned int start_timer)
739 {
740 	return;
741 }
742 
743 #ifdef CONFIG_HPET
744 static void hpet_reserve_msi_timers(struct hpet_data *hd)
745 {
746 	return;
747 }
748 #endif
749 
750 static int hpet_cpuhp_notify(struct notifier_block *n,
751 		unsigned long action, void *hcpu)
752 {
753 	return NOTIFY_OK;
754 }
755 
756 #endif
757 
758 /*
759  * Clock source related code
760  */
761 static cycle_t read_hpet(struct clocksource *cs)
762 {
763 	return (cycle_t)hpet_readl(HPET_COUNTER);
764 }
765 
766 #ifdef CONFIG_X86_64
767 static cycle_t __vsyscall_fn vread_hpet(void)
768 {
769 	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + 0xf0);
770 }
771 #endif
772 
773 static struct clocksource clocksource_hpet = {
774 	.name		= "hpet",
775 	.rating		= 250,
776 	.read		= read_hpet,
777 	.mask		= HPET_MASK,
778 	.shift		= HPET_SHIFT,
779 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
780 	.resume		= hpet_resume_counter,
781 #ifdef CONFIG_X86_64
782 	.vread		= vread_hpet,
783 #endif
784 };
785 
786 static int hpet_clocksource_register(void)
787 {
788 	u64 start, now;
789 	cycle_t t1;
790 
791 	/* Start the counter */
792 	hpet_restart_counter();
793 
794 	/* Verify whether hpet counter works */
795 	t1 = hpet_readl(HPET_COUNTER);
796 	rdtscll(start);
797 
798 	/*
799 	 * We don't know the TSC frequency yet, but waiting for
800 	 * 200000 TSC cycles is safe:
801 	 * 4 GHz == 50us
802 	 * 1 GHz == 200us
803 	 */
804 	do {
805 		rep_nop();
806 		rdtscll(now);
807 	} while ((now - start) < 200000UL);
808 
809 	if (t1 == hpet_readl(HPET_COUNTER)) {
810 		printk(KERN_WARNING
811 		       "HPET counter not counting. HPET disabled\n");
812 		return -ENODEV;
813 	}
814 
815 	/*
816 	 * The definition of mult is (include/linux/clocksource.h)
817 	 * mult/2^shift = ns/cyc and hpet_period is in units of fsec/cyc
818 	 * so we first need to convert hpet_period to ns/cyc units:
819 	 *  mult/2^shift = ns/cyc = hpet_period/10^6
820 	 *  mult = (hpet_period * 2^shift)/10^6
821 	 *  mult = (hpet_period << shift)/FSEC_PER_NSEC
822 	 */
823 	clocksource_hpet.mult = div_sc(hpet_period, FSEC_PER_NSEC, HPET_SHIFT);
824 
825 	clocksource_register(&clocksource_hpet);
826 
827 	return 0;
828 }
829 
830 /**
831  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
832  */
833 int __init hpet_enable(void)
834 {
835 	unsigned int id;
836 	int i;
837 
838 	if (!is_hpet_capable())
839 		return 0;
840 
841 	hpet_set_mapping();
842 
843 	/*
844 	 * Read the period and check for a sane value:
845 	 */
846 	hpet_period = hpet_readl(HPET_PERIOD);
847 
848 	/*
849 	 * AMD SB700 based systems with spread spectrum enabled use a
850 	 * SMM based HPET emulation to provide proper frequency
851 	 * setting. The SMM code is initialized with the first HPET
852 	 * register access and takes some time to complete. During
853 	 * this time the config register reads 0xffffffff. We check
854 	 * for max. 1000 loops whether the config register reads a non
855 	 * 0xffffffff value to make sure that HPET is up and running
856 	 * before we go further. A counting loop is safe, as the HPET
857 	 * access takes thousands of CPU cycles. On non SB700 based
858 	 * machines this check is only done once and has no side
859 	 * effects.
860 	 */
861 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
862 		if (i == 1000) {
863 			printk(KERN_WARNING
864 			       "HPET config register value = 0xFFFFFFFF. "
865 			       "Disabling HPET\n");
866 			goto out_nohpet;
867 		}
868 	}
869 
870 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
871 		goto out_nohpet;
872 
873 	/*
874 	 * Read the HPET ID register to retrieve the IRQ routing
875 	 * information and the number of channels
876 	 */
877 	id = hpet_readl(HPET_ID);
878 	hpet_print_config();
879 
880 #ifdef CONFIG_HPET_EMULATE_RTC
881 	/*
882 	 * The legacy routing mode needs at least two channels, tick timer
883 	 * and the rtc emulation channel.
884 	 */
885 	if (!(id & HPET_ID_NUMBER))
886 		goto out_nohpet;
887 #endif
888 
889 	if (hpet_clocksource_register())
890 		goto out_nohpet;
891 
892 	if (id & HPET_ID_LEGSUP) {
893 		hpet_legacy_clockevent_register();
894 		return 1;
895 	}
896 	return 0;
897 
898 out_nohpet:
899 	hpet_clear_mapping();
900 	hpet_address = 0;
901 	return 0;
902 }
903 
904 /*
905  * Needs to be late, as the reserve_timer code calls kalloc !
906  *
907  * Not a problem on i386 as hpet_enable is called from late_time_init,
908  * but on x86_64 it is necessary !
909  */
910 static __init int hpet_late_init(void)
911 {
912 	int cpu;
913 
914 	if (boot_hpet_disable)
915 		return -ENODEV;
916 
917 	if (!hpet_address) {
918 		if (!force_hpet_address)
919 			return -ENODEV;
920 
921 		hpet_address = force_hpet_address;
922 		hpet_enable();
923 	}
924 
925 	if (!hpet_virt_address)
926 		return -ENODEV;
927 
928 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
929 		hpet_msi_capability_lookup(2);
930 	else
931 		hpet_msi_capability_lookup(0);
932 
933 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
934 	hpet_print_config();
935 
936 	if (hpet_msi_disable)
937 		return 0;
938 
939 	if (boot_cpu_has(X86_FEATURE_ARAT))
940 		return 0;
941 
942 	for_each_online_cpu(cpu) {
943 		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
944 	}
945 
946 	/* This notifier should be called after workqueue is ready */
947 	hotcpu_notifier(hpet_cpuhp_notify, -20);
948 
949 	return 0;
950 }
951 fs_initcall(hpet_late_init);
952 
953 void hpet_disable(void)
954 {
955 	if (is_hpet_capable()) {
956 		unsigned int cfg = hpet_readl(HPET_CFG);
957 
958 		if (hpet_legacy_int_enabled) {
959 			cfg &= ~HPET_CFG_LEGACY;
960 			hpet_legacy_int_enabled = 0;
961 		}
962 		cfg &= ~HPET_CFG_ENABLE;
963 		hpet_writel(cfg, HPET_CFG);
964 	}
965 }
966 
967 #ifdef CONFIG_HPET_EMULATE_RTC
968 
969 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
970  * is enabled, we support RTC interrupt functionality in software.
971  * RTC has 3 kinds of interrupts:
972  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
973  *    is updated
974  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
975  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
976  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
977  * (1) and (2) above are implemented using polling at a frequency of
978  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
979  * overhead. (DEFAULT_RTC_INT_FREQ)
980  * For (3), we use interrupts at 64Hz or user specified periodic
981  * frequency, whichever is higher.
982  */
983 #include <linux/mc146818rtc.h>
984 #include <linux/rtc.h>
985 #include <asm/rtc.h>
986 
987 #define DEFAULT_RTC_INT_FREQ	64
988 #define DEFAULT_RTC_SHIFT	6
989 #define RTC_NUM_INTS		1
990 
991 static unsigned long hpet_rtc_flags;
992 static int hpet_prev_update_sec;
993 static struct rtc_time hpet_alarm_time;
994 static unsigned long hpet_pie_count;
995 static u32 hpet_t1_cmp;
996 static u32 hpet_default_delta;
997 static u32 hpet_pie_delta;
998 static unsigned long hpet_pie_limit;
999 
1000 static rtc_irq_handler irq_handler;
1001 
1002 /*
1003  * Check that the hpet counter c1 is ahead of the c2
1004  */
1005 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1006 {
1007 	return (s32)(c2 - c1) < 0;
1008 }
1009 
1010 /*
1011  * Registers a IRQ handler.
1012  */
1013 int hpet_register_irq_handler(rtc_irq_handler handler)
1014 {
1015 	if (!is_hpet_enabled())
1016 		return -ENODEV;
1017 	if (irq_handler)
1018 		return -EBUSY;
1019 
1020 	irq_handler = handler;
1021 
1022 	return 0;
1023 }
1024 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1025 
1026 /*
1027  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1028  * and does cleanup.
1029  */
1030 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1031 {
1032 	if (!is_hpet_enabled())
1033 		return;
1034 
1035 	irq_handler = NULL;
1036 	hpet_rtc_flags = 0;
1037 }
1038 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1039 
1040 /*
1041  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1042  * is not supported by all HPET implementations for timer 1.
1043  *
1044  * hpet_rtc_timer_init() is called when the rtc is initialized.
1045  */
1046 int hpet_rtc_timer_init(void)
1047 {
1048 	unsigned int cfg, cnt, delta;
1049 	unsigned long flags;
1050 
1051 	if (!is_hpet_enabled())
1052 		return 0;
1053 
1054 	if (!hpet_default_delta) {
1055 		uint64_t clc;
1056 
1057 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1058 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1059 		hpet_default_delta = clc;
1060 	}
1061 
1062 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1063 		delta = hpet_default_delta;
1064 	else
1065 		delta = hpet_pie_delta;
1066 
1067 	local_irq_save(flags);
1068 
1069 	cnt = delta + hpet_readl(HPET_COUNTER);
1070 	hpet_writel(cnt, HPET_T1_CMP);
1071 	hpet_t1_cmp = cnt;
1072 
1073 	cfg = hpet_readl(HPET_T1_CFG);
1074 	cfg &= ~HPET_TN_PERIODIC;
1075 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1076 	hpet_writel(cfg, HPET_T1_CFG);
1077 
1078 	local_irq_restore(flags);
1079 
1080 	return 1;
1081 }
1082 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1083 
1084 /*
1085  * The functions below are called from rtc driver.
1086  * Return 0 if HPET is not being used.
1087  * Otherwise do the necessary changes and return 1.
1088  */
1089 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1090 {
1091 	if (!is_hpet_enabled())
1092 		return 0;
1093 
1094 	hpet_rtc_flags &= ~bit_mask;
1095 	return 1;
1096 }
1097 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1098 
1099 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1100 {
1101 	unsigned long oldbits = hpet_rtc_flags;
1102 
1103 	if (!is_hpet_enabled())
1104 		return 0;
1105 
1106 	hpet_rtc_flags |= bit_mask;
1107 
1108 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1109 		hpet_prev_update_sec = -1;
1110 
1111 	if (!oldbits)
1112 		hpet_rtc_timer_init();
1113 
1114 	return 1;
1115 }
1116 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1117 
1118 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1119 			unsigned char sec)
1120 {
1121 	if (!is_hpet_enabled())
1122 		return 0;
1123 
1124 	hpet_alarm_time.tm_hour = hrs;
1125 	hpet_alarm_time.tm_min = min;
1126 	hpet_alarm_time.tm_sec = sec;
1127 
1128 	return 1;
1129 }
1130 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1131 
1132 int hpet_set_periodic_freq(unsigned long freq)
1133 {
1134 	uint64_t clc;
1135 
1136 	if (!is_hpet_enabled())
1137 		return 0;
1138 
1139 	if (freq <= DEFAULT_RTC_INT_FREQ)
1140 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1141 	else {
1142 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1143 		do_div(clc, freq);
1144 		clc >>= hpet_clockevent.shift;
1145 		hpet_pie_delta = clc;
1146 	}
1147 	return 1;
1148 }
1149 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1150 
1151 int hpet_rtc_dropped_irq(void)
1152 {
1153 	return is_hpet_enabled();
1154 }
1155 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1156 
1157 static void hpet_rtc_timer_reinit(void)
1158 {
1159 	unsigned int cfg, delta;
1160 	int lost_ints = -1;
1161 
1162 	if (unlikely(!hpet_rtc_flags)) {
1163 		cfg = hpet_readl(HPET_T1_CFG);
1164 		cfg &= ~HPET_TN_ENABLE;
1165 		hpet_writel(cfg, HPET_T1_CFG);
1166 		return;
1167 	}
1168 
1169 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1170 		delta = hpet_default_delta;
1171 	else
1172 		delta = hpet_pie_delta;
1173 
1174 	/*
1175 	 * Increment the comparator value until we are ahead of the
1176 	 * current count.
1177 	 */
1178 	do {
1179 		hpet_t1_cmp += delta;
1180 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1181 		lost_ints++;
1182 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1183 
1184 	if (lost_ints) {
1185 		if (hpet_rtc_flags & RTC_PIE)
1186 			hpet_pie_count += lost_ints;
1187 		if (printk_ratelimit())
1188 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1189 				lost_ints);
1190 	}
1191 }
1192 
1193 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1194 {
1195 	struct rtc_time curr_time;
1196 	unsigned long rtc_int_flag = 0;
1197 
1198 	hpet_rtc_timer_reinit();
1199 	memset(&curr_time, 0, sizeof(struct rtc_time));
1200 
1201 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1202 		get_rtc_time(&curr_time);
1203 
1204 	if (hpet_rtc_flags & RTC_UIE &&
1205 	    curr_time.tm_sec != hpet_prev_update_sec) {
1206 		if (hpet_prev_update_sec >= 0)
1207 			rtc_int_flag = RTC_UF;
1208 		hpet_prev_update_sec = curr_time.tm_sec;
1209 	}
1210 
1211 	if (hpet_rtc_flags & RTC_PIE &&
1212 	    ++hpet_pie_count >= hpet_pie_limit) {
1213 		rtc_int_flag |= RTC_PF;
1214 		hpet_pie_count = 0;
1215 	}
1216 
1217 	if (hpet_rtc_flags & RTC_AIE &&
1218 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1219 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1220 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1221 			rtc_int_flag |= RTC_AF;
1222 
1223 	if (rtc_int_flag) {
1224 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1225 		if (irq_handler)
1226 			irq_handler(rtc_int_flag, dev_id);
1227 	}
1228 	return IRQ_HANDLED;
1229 }
1230 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1231 #endif
1232