xref: /openbmc/linux/arch/x86/kernel/hpet.c (revision a09d2831)
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 #ifdef CONFIG_PCI_MSI
38 static unsigned long			hpet_num_timers;
39 #endif
40 static void __iomem			*hpet_virt_address;
41 
42 struct hpet_dev {
43 	struct clock_event_device	evt;
44 	unsigned int			num;
45 	int				cpu;
46 	unsigned int			irq;
47 	unsigned int			flags;
48 	char				name[10];
49 };
50 
51 inline unsigned int hpet_readl(unsigned int a)
52 {
53 	return readl(hpet_virt_address + a);
54 }
55 
56 static inline void hpet_writel(unsigned int d, unsigned int a)
57 {
58 	writel(d, hpet_virt_address + a);
59 }
60 
61 #ifdef CONFIG_X86_64
62 #include <asm/pgtable.h>
63 #endif
64 
65 static inline void hpet_set_mapping(void)
66 {
67 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
68 #ifdef CONFIG_X86_64
69 	__set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
70 #endif
71 }
72 
73 static inline void hpet_clear_mapping(void)
74 {
75 	iounmap(hpet_virt_address);
76 	hpet_virt_address = NULL;
77 }
78 
79 /*
80  * HPET command line enable / disable
81  */
82 static int boot_hpet_disable;
83 int hpet_force_user;
84 static int hpet_verbose;
85 
86 static int __init hpet_setup(char *str)
87 {
88 	if (str) {
89 		if (!strncmp("disable", str, 7))
90 			boot_hpet_disable = 1;
91 		if (!strncmp("force", str, 5))
92 			hpet_force_user = 1;
93 		if (!strncmp("verbose", str, 7))
94 			hpet_verbose = 1;
95 	}
96 	return 1;
97 }
98 __setup("hpet=", hpet_setup);
99 
100 static int __init disable_hpet(char *str)
101 {
102 	boot_hpet_disable = 1;
103 	return 1;
104 }
105 __setup("nohpet", disable_hpet);
106 
107 static inline int is_hpet_capable(void)
108 {
109 	return !boot_hpet_disable && hpet_address;
110 }
111 
112 /*
113  * HPET timer interrupt enable / disable
114  */
115 static int hpet_legacy_int_enabled;
116 
117 /**
118  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
119  */
120 int is_hpet_enabled(void)
121 {
122 	return is_hpet_capable() && hpet_legacy_int_enabled;
123 }
124 EXPORT_SYMBOL_GPL(is_hpet_enabled);
125 
126 static void _hpet_print_config(const char *function, int line)
127 {
128 	u32 i, timers, l, h;
129 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
130 	l = hpet_readl(HPET_ID);
131 	h = hpet_readl(HPET_PERIOD);
132 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
133 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
134 	l = hpet_readl(HPET_CFG);
135 	h = hpet_readl(HPET_STATUS);
136 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
137 	l = hpet_readl(HPET_COUNTER);
138 	h = hpet_readl(HPET_COUNTER+4);
139 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
140 
141 	for (i = 0; i < timers; i++) {
142 		l = hpet_readl(HPET_Tn_CFG(i));
143 		h = hpet_readl(HPET_Tn_CFG(i)+4);
144 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
145 		       i, l, h);
146 		l = hpet_readl(HPET_Tn_CMP(i));
147 		h = hpet_readl(HPET_Tn_CMP(i)+4);
148 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
149 		       i, l, h);
150 		l = hpet_readl(HPET_Tn_ROUTE(i));
151 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
152 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
153 		       i, l, h);
154 	}
155 }
156 
157 #define hpet_print_config()					\
158 do {								\
159 	if (hpet_verbose)					\
160 		_hpet_print_config(__FUNCTION__, __LINE__);	\
161 } while (0)
162 
163 /*
164  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
165  * timer 0 and timer 1 in case of RTC emulation.
166  */
167 #ifdef CONFIG_HPET
168 
169 static void hpet_reserve_msi_timers(struct hpet_data *hd);
170 
171 static void hpet_reserve_platform_timers(unsigned int id)
172 {
173 	struct hpet __iomem *hpet = hpet_virt_address;
174 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
175 	unsigned int nrtimers, i;
176 	struct hpet_data hd;
177 
178 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
179 
180 	memset(&hd, 0, sizeof(hd));
181 	hd.hd_phys_address	= hpet_address;
182 	hd.hd_address		= hpet;
183 	hd.hd_nirqs		= nrtimers;
184 	hpet_reserve_timer(&hd, 0);
185 
186 #ifdef CONFIG_HPET_EMULATE_RTC
187 	hpet_reserve_timer(&hd, 1);
188 #endif
189 
190 	/*
191 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
192 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
193 	 * don't bother configuring *any* comparator interrupts.
194 	 */
195 	hd.hd_irq[0] = HPET_LEGACY_8254;
196 	hd.hd_irq[1] = HPET_LEGACY_RTC;
197 
198 	for (i = 2; i < nrtimers; timer++, i++) {
199 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
200 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
201 	}
202 
203 	hpet_reserve_msi_timers(&hd);
204 
205 	hpet_alloc(&hd);
206 
207 }
208 #else
209 static void hpet_reserve_platform_timers(unsigned int id) { }
210 #endif
211 
212 /*
213  * Common hpet info
214  */
215 static unsigned long hpet_period;
216 
217 static void hpet_legacy_set_mode(enum clock_event_mode mode,
218 			  struct clock_event_device *evt);
219 static int hpet_legacy_next_event(unsigned long delta,
220 			   struct clock_event_device *evt);
221 
222 /*
223  * The hpet clock event device
224  */
225 static struct clock_event_device hpet_clockevent = {
226 	.name		= "hpet",
227 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
228 	.set_mode	= hpet_legacy_set_mode,
229 	.set_next_event = hpet_legacy_next_event,
230 	.shift		= 32,
231 	.irq		= 0,
232 	.rating		= 50,
233 };
234 
235 static void hpet_stop_counter(void)
236 {
237 	unsigned long cfg = hpet_readl(HPET_CFG);
238 	cfg &= ~HPET_CFG_ENABLE;
239 	hpet_writel(cfg, HPET_CFG);
240 }
241 
242 static void hpet_reset_counter(void)
243 {
244 	hpet_writel(0, HPET_COUNTER);
245 	hpet_writel(0, HPET_COUNTER + 4);
246 }
247 
248 static void hpet_start_counter(void)
249 {
250 	unsigned int cfg = hpet_readl(HPET_CFG);
251 	cfg |= HPET_CFG_ENABLE;
252 	hpet_writel(cfg, HPET_CFG);
253 }
254 
255 static void hpet_restart_counter(void)
256 {
257 	hpet_stop_counter();
258 	hpet_reset_counter();
259 	hpet_start_counter();
260 }
261 
262 static void hpet_resume_device(void)
263 {
264 	force_hpet_resume();
265 }
266 
267 static void hpet_resume_counter(void)
268 {
269 	hpet_resume_device();
270 	hpet_restart_counter();
271 }
272 
273 static void hpet_enable_legacy_int(void)
274 {
275 	unsigned int cfg = hpet_readl(HPET_CFG);
276 
277 	cfg |= HPET_CFG_LEGACY;
278 	hpet_writel(cfg, HPET_CFG);
279 	hpet_legacy_int_enabled = 1;
280 }
281 
282 static void hpet_legacy_clockevent_register(void)
283 {
284 	/* Start HPET legacy interrupts */
285 	hpet_enable_legacy_int();
286 
287 	/*
288 	 * The mult factor is defined as (include/linux/clockchips.h)
289 	 *  mult/2^shift = cyc/ns (in contrast to ns/cyc in clocksource.h)
290 	 * hpet_period is in units of femtoseconds (per cycle), so
291 	 *  mult/2^shift = cyc/ns = 10^6/hpet_period
292 	 *  mult = (10^6 * 2^shift)/hpet_period
293 	 *  mult = (FSEC_PER_NSEC << hpet_clockevent.shift)/hpet_period
294 	 */
295 	hpet_clockevent.mult = div_sc((unsigned long) FSEC_PER_NSEC,
296 				      hpet_period, hpet_clockevent.shift);
297 	/* Calculate the min / max delta */
298 	hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF,
299 							   &hpet_clockevent);
300 	/* 5 usec minimum reprogramming delta. */
301 	hpet_clockevent.min_delta_ns = 5000;
302 
303 	/*
304 	 * Start hpet with the boot cpu mask and make it
305 	 * global after the IO_APIC has been initialized.
306 	 */
307 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
308 	clockevents_register_device(&hpet_clockevent);
309 	global_clock_event = &hpet_clockevent;
310 	printk(KERN_DEBUG "hpet clockevent registered\n");
311 }
312 
313 static int hpet_setup_msi_irq(unsigned int irq);
314 
315 static void hpet_set_mode(enum clock_event_mode mode,
316 			  struct clock_event_device *evt, int timer)
317 {
318 	unsigned int cfg, cmp, now;
319 	uint64_t delta;
320 
321 	switch (mode) {
322 	case CLOCK_EVT_MODE_PERIODIC:
323 		hpet_stop_counter();
324 		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
325 		delta >>= evt->shift;
326 		now = hpet_readl(HPET_COUNTER);
327 		cmp = now + (unsigned int) delta;
328 		cfg = hpet_readl(HPET_Tn_CFG(timer));
329 		/* Make sure we use edge triggered interrupts */
330 		cfg &= ~HPET_TN_LEVEL;
331 		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
332 		       HPET_TN_SETVAL | HPET_TN_32BIT;
333 		hpet_writel(cfg, HPET_Tn_CFG(timer));
334 		hpet_writel(cmp, HPET_Tn_CMP(timer));
335 		udelay(1);
336 		/*
337 		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
338 		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
339 		 * bit is automatically cleared after the first write.
340 		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
341 		 * Publication # 24674)
342 		 */
343 		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
344 		hpet_start_counter();
345 		hpet_print_config();
346 		break;
347 
348 	case CLOCK_EVT_MODE_ONESHOT:
349 		cfg = hpet_readl(HPET_Tn_CFG(timer));
350 		cfg &= ~HPET_TN_PERIODIC;
351 		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
352 		hpet_writel(cfg, HPET_Tn_CFG(timer));
353 		break;
354 
355 	case CLOCK_EVT_MODE_UNUSED:
356 	case CLOCK_EVT_MODE_SHUTDOWN:
357 		cfg = hpet_readl(HPET_Tn_CFG(timer));
358 		cfg &= ~HPET_TN_ENABLE;
359 		hpet_writel(cfg, HPET_Tn_CFG(timer));
360 		break;
361 
362 	case CLOCK_EVT_MODE_RESUME:
363 		if (timer == 0) {
364 			hpet_enable_legacy_int();
365 		} else {
366 			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
367 			hpet_setup_msi_irq(hdev->irq);
368 			disable_irq(hdev->irq);
369 			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
370 			enable_irq(hdev->irq);
371 		}
372 		hpet_print_config();
373 		break;
374 	}
375 }
376 
377 static int hpet_next_event(unsigned long delta,
378 			   struct clock_event_device *evt, int timer)
379 {
380 	u32 cnt;
381 
382 	cnt = hpet_readl(HPET_COUNTER);
383 	cnt += (u32) delta;
384 	hpet_writel(cnt, HPET_Tn_CMP(timer));
385 
386 	/*
387 	 * We need to read back the CMP register on certain HPET
388 	 * implementations (ATI chipsets) which seem to delay the
389 	 * transfer of the compare register into the internal compare
390 	 * logic. With small deltas this might actually be too late as
391 	 * the counter could already be higher than the compare value
392 	 * at that point and we would wait for the next hpet interrupt
393 	 * forever. We found out that reading the CMP register back
394 	 * forces the transfer so we can rely on the comparison with
395 	 * the counter register below. If the read back from the
396 	 * compare register does not match the value we programmed
397 	 * then we might have a real hardware problem. We can not do
398 	 * much about it here, but at least alert the user/admin with
399 	 * a prominent warning.
400 	 */
401 	WARN_ONCE(hpet_readl(HPET_Tn_CMP(timer)) != cnt,
402 		  KERN_WARNING "hpet: compare register read back failed.\n");
403 
404 	return (s32)(hpet_readl(HPET_COUNTER) - cnt) >= 0 ? -ETIME : 0;
405 }
406 
407 static void hpet_legacy_set_mode(enum clock_event_mode mode,
408 			struct clock_event_device *evt)
409 {
410 	hpet_set_mode(mode, evt, 0);
411 }
412 
413 static int hpet_legacy_next_event(unsigned long delta,
414 			struct clock_event_device *evt)
415 {
416 	return hpet_next_event(delta, evt, 0);
417 }
418 
419 /*
420  * HPET MSI Support
421  */
422 #ifdef CONFIG_PCI_MSI
423 
424 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
425 static struct hpet_dev	*hpet_devs;
426 
427 void hpet_msi_unmask(unsigned int irq)
428 {
429 	struct hpet_dev *hdev = get_irq_data(irq);
430 	unsigned int cfg;
431 
432 	/* unmask it */
433 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
434 	cfg |= HPET_TN_FSB;
435 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
436 }
437 
438 void hpet_msi_mask(unsigned int irq)
439 {
440 	unsigned int cfg;
441 	struct hpet_dev *hdev = get_irq_data(irq);
442 
443 	/* mask it */
444 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
445 	cfg &= ~HPET_TN_FSB;
446 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
447 }
448 
449 void hpet_msi_write(unsigned int irq, struct msi_msg *msg)
450 {
451 	struct hpet_dev *hdev = get_irq_data(irq);
452 
453 	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
454 	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
455 }
456 
457 void hpet_msi_read(unsigned int irq, struct msi_msg *msg)
458 {
459 	struct hpet_dev *hdev = get_irq_data(irq);
460 
461 	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
462 	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
463 	msg->address_hi = 0;
464 }
465 
466 static void hpet_msi_set_mode(enum clock_event_mode mode,
467 				struct clock_event_device *evt)
468 {
469 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
470 	hpet_set_mode(mode, evt, hdev->num);
471 }
472 
473 static int hpet_msi_next_event(unsigned long delta,
474 				struct clock_event_device *evt)
475 {
476 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
477 	return hpet_next_event(delta, evt, hdev->num);
478 }
479 
480 static int hpet_setup_msi_irq(unsigned int irq)
481 {
482 	if (arch_setup_hpet_msi(irq, hpet_blockid)) {
483 		destroy_irq(irq);
484 		return -EINVAL;
485 	}
486 	return 0;
487 }
488 
489 static int hpet_assign_irq(struct hpet_dev *dev)
490 {
491 	unsigned int irq;
492 
493 	irq = create_irq();
494 	if (!irq)
495 		return -EINVAL;
496 
497 	set_irq_data(irq, dev);
498 
499 	if (hpet_setup_msi_irq(irq))
500 		return -EINVAL;
501 
502 	dev->irq = irq;
503 	return 0;
504 }
505 
506 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
507 {
508 	struct hpet_dev *dev = (struct hpet_dev *)data;
509 	struct clock_event_device *hevt = &dev->evt;
510 
511 	if (!hevt->event_handler) {
512 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
513 				dev->num);
514 		return IRQ_HANDLED;
515 	}
516 
517 	hevt->event_handler(hevt);
518 	return IRQ_HANDLED;
519 }
520 
521 static int hpet_setup_irq(struct hpet_dev *dev)
522 {
523 
524 	if (request_irq(dev->irq, hpet_interrupt_handler,
525 			IRQF_TIMER | IRQF_DISABLED | IRQF_NOBALANCING,
526 			dev->name, dev))
527 		return -1;
528 
529 	disable_irq(dev->irq);
530 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
531 	enable_irq(dev->irq);
532 
533 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
534 			 dev->name, dev->irq);
535 
536 	return 0;
537 }
538 
539 /* This should be called in specific @cpu */
540 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
541 {
542 	struct clock_event_device *evt = &hdev->evt;
543 	uint64_t hpet_freq;
544 
545 	WARN_ON(cpu != smp_processor_id());
546 	if (!(hdev->flags & HPET_DEV_VALID))
547 		return;
548 
549 	if (hpet_setup_msi_irq(hdev->irq))
550 		return;
551 
552 	hdev->cpu = cpu;
553 	per_cpu(cpu_hpet_dev, cpu) = hdev;
554 	evt->name = hdev->name;
555 	hpet_setup_irq(hdev);
556 	evt->irq = hdev->irq;
557 
558 	evt->rating = 110;
559 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
560 	if (hdev->flags & HPET_DEV_PERI_CAP)
561 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
562 
563 	evt->set_mode = hpet_msi_set_mode;
564 	evt->set_next_event = hpet_msi_next_event;
565 	evt->shift = 32;
566 
567 	/*
568 	 * The period is a femto seconds value. We need to calculate the
569 	 * scaled math multiplication factor for nanosecond to hpet tick
570 	 * conversion.
571 	 */
572 	hpet_freq = 1000000000000000ULL;
573 	do_div(hpet_freq, hpet_period);
574 	evt->mult = div_sc((unsigned long) hpet_freq,
575 				      NSEC_PER_SEC, evt->shift);
576 	/* Calculate the max delta */
577 	evt->max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, evt);
578 	/* 5 usec minimum reprogramming delta. */
579 	evt->min_delta_ns = 5000;
580 
581 	evt->cpumask = cpumask_of(hdev->cpu);
582 	clockevents_register_device(evt);
583 }
584 
585 #ifdef CONFIG_HPET
586 /* Reserve at least one timer for userspace (/dev/hpet) */
587 #define RESERVE_TIMERS 1
588 #else
589 #define RESERVE_TIMERS 0
590 #endif
591 
592 static void hpet_msi_capability_lookup(unsigned int start_timer)
593 {
594 	unsigned int id;
595 	unsigned int num_timers;
596 	unsigned int num_timers_used = 0;
597 	int i;
598 
599 	if (boot_cpu_has(X86_FEATURE_ARAT))
600 		return;
601 	id = hpet_readl(HPET_ID);
602 
603 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
604 	num_timers++; /* Value read out starts from 0 */
605 	hpet_print_config();
606 
607 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
608 	if (!hpet_devs)
609 		return;
610 
611 	hpet_num_timers = num_timers;
612 
613 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
614 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
615 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
616 
617 		/* Only consider HPET timer with MSI support */
618 		if (!(cfg & HPET_TN_FSB_CAP))
619 			continue;
620 
621 		hdev->flags = 0;
622 		if (cfg & HPET_TN_PERIODIC_CAP)
623 			hdev->flags |= HPET_DEV_PERI_CAP;
624 		hdev->num = i;
625 
626 		sprintf(hdev->name, "hpet%d", i);
627 		if (hpet_assign_irq(hdev))
628 			continue;
629 
630 		hdev->flags |= HPET_DEV_FSB_CAP;
631 		hdev->flags |= HPET_DEV_VALID;
632 		num_timers_used++;
633 		if (num_timers_used == num_possible_cpus())
634 			break;
635 	}
636 
637 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
638 		num_timers, num_timers_used);
639 }
640 
641 #ifdef CONFIG_HPET
642 static void hpet_reserve_msi_timers(struct hpet_data *hd)
643 {
644 	int i;
645 
646 	if (!hpet_devs)
647 		return;
648 
649 	for (i = 0; i < hpet_num_timers; i++) {
650 		struct hpet_dev *hdev = &hpet_devs[i];
651 
652 		if (!(hdev->flags & HPET_DEV_VALID))
653 			continue;
654 
655 		hd->hd_irq[hdev->num] = hdev->irq;
656 		hpet_reserve_timer(hd, hdev->num);
657 	}
658 }
659 #endif
660 
661 static struct hpet_dev *hpet_get_unused_timer(void)
662 {
663 	int i;
664 
665 	if (!hpet_devs)
666 		return NULL;
667 
668 	for (i = 0; i < hpet_num_timers; i++) {
669 		struct hpet_dev *hdev = &hpet_devs[i];
670 
671 		if (!(hdev->flags & HPET_DEV_VALID))
672 			continue;
673 		if (test_and_set_bit(HPET_DEV_USED_BIT,
674 			(unsigned long *)&hdev->flags))
675 			continue;
676 		return hdev;
677 	}
678 	return NULL;
679 }
680 
681 struct hpet_work_struct {
682 	struct delayed_work work;
683 	struct completion complete;
684 };
685 
686 static void hpet_work(struct work_struct *w)
687 {
688 	struct hpet_dev *hdev;
689 	int cpu = smp_processor_id();
690 	struct hpet_work_struct *hpet_work;
691 
692 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
693 
694 	hdev = hpet_get_unused_timer();
695 	if (hdev)
696 		init_one_hpet_msi_clockevent(hdev, cpu);
697 
698 	complete(&hpet_work->complete);
699 }
700 
701 static int hpet_cpuhp_notify(struct notifier_block *n,
702 		unsigned long action, void *hcpu)
703 {
704 	unsigned long cpu = (unsigned long)hcpu;
705 	struct hpet_work_struct work;
706 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
707 
708 	switch (action & 0xf) {
709 	case CPU_ONLINE:
710 		INIT_DELAYED_WORK_ON_STACK(&work.work, hpet_work);
711 		init_completion(&work.complete);
712 		/* FIXME: add schedule_work_on() */
713 		schedule_delayed_work_on(cpu, &work.work, 0);
714 		wait_for_completion(&work.complete);
715 		destroy_timer_on_stack(&work.work.timer);
716 		break;
717 	case CPU_DEAD:
718 		if (hdev) {
719 			free_irq(hdev->irq, hdev);
720 			hdev->flags &= ~HPET_DEV_USED;
721 			per_cpu(cpu_hpet_dev, cpu) = NULL;
722 		}
723 		break;
724 	}
725 	return NOTIFY_OK;
726 }
727 #else
728 
729 static int hpet_setup_msi_irq(unsigned int irq)
730 {
731 	return 0;
732 }
733 static void hpet_msi_capability_lookup(unsigned int start_timer)
734 {
735 	return;
736 }
737 
738 #ifdef CONFIG_HPET
739 static void hpet_reserve_msi_timers(struct hpet_data *hd)
740 {
741 	return;
742 }
743 #endif
744 
745 static int hpet_cpuhp_notify(struct notifier_block *n,
746 		unsigned long action, void *hcpu)
747 {
748 	return NOTIFY_OK;
749 }
750 
751 #endif
752 
753 /*
754  * Clock source related code
755  */
756 static cycle_t read_hpet(struct clocksource *cs)
757 {
758 	return (cycle_t)hpet_readl(HPET_COUNTER);
759 }
760 
761 #ifdef CONFIG_X86_64
762 static cycle_t __vsyscall_fn vread_hpet(void)
763 {
764 	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + 0xf0);
765 }
766 #endif
767 
768 static struct clocksource clocksource_hpet = {
769 	.name		= "hpet",
770 	.rating		= 250,
771 	.read		= read_hpet,
772 	.mask		= HPET_MASK,
773 	.shift		= HPET_SHIFT,
774 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
775 	.resume		= hpet_resume_counter,
776 #ifdef CONFIG_X86_64
777 	.vread		= vread_hpet,
778 #endif
779 };
780 
781 static int hpet_clocksource_register(void)
782 {
783 	u64 start, now;
784 	cycle_t t1;
785 
786 	/* Start the counter */
787 	hpet_restart_counter();
788 
789 	/* Verify whether hpet counter works */
790 	t1 = hpet_readl(HPET_COUNTER);
791 	rdtscll(start);
792 
793 	/*
794 	 * We don't know the TSC frequency yet, but waiting for
795 	 * 200000 TSC cycles is safe:
796 	 * 4 GHz == 50us
797 	 * 1 GHz == 200us
798 	 */
799 	do {
800 		rep_nop();
801 		rdtscll(now);
802 	} while ((now - start) < 200000UL);
803 
804 	if (t1 == hpet_readl(HPET_COUNTER)) {
805 		printk(KERN_WARNING
806 		       "HPET counter not counting. HPET disabled\n");
807 		return -ENODEV;
808 	}
809 
810 	/*
811 	 * The definition of mult is (include/linux/clocksource.h)
812 	 * mult/2^shift = ns/cyc and hpet_period is in units of fsec/cyc
813 	 * so we first need to convert hpet_period to ns/cyc units:
814 	 *  mult/2^shift = ns/cyc = hpet_period/10^6
815 	 *  mult = (hpet_period * 2^shift)/10^6
816 	 *  mult = (hpet_period << shift)/FSEC_PER_NSEC
817 	 */
818 	clocksource_hpet.mult = div_sc(hpet_period, FSEC_PER_NSEC, HPET_SHIFT);
819 
820 	clocksource_register(&clocksource_hpet);
821 
822 	return 0;
823 }
824 
825 /**
826  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
827  */
828 int __init hpet_enable(void)
829 {
830 	unsigned int id;
831 	int i;
832 
833 	if (!is_hpet_capable())
834 		return 0;
835 
836 	hpet_set_mapping();
837 
838 	/*
839 	 * Read the period and check for a sane value:
840 	 */
841 	hpet_period = hpet_readl(HPET_PERIOD);
842 
843 	/*
844 	 * AMD SB700 based systems with spread spectrum enabled use a
845 	 * SMM based HPET emulation to provide proper frequency
846 	 * setting. The SMM code is initialized with the first HPET
847 	 * register access and takes some time to complete. During
848 	 * this time the config register reads 0xffffffff. We check
849 	 * for max. 1000 loops whether the config register reads a non
850 	 * 0xffffffff value to make sure that HPET is up and running
851 	 * before we go further. A counting loop is safe, as the HPET
852 	 * access takes thousands of CPU cycles. On non SB700 based
853 	 * machines this check is only done once and has no side
854 	 * effects.
855 	 */
856 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
857 		if (i == 1000) {
858 			printk(KERN_WARNING
859 			       "HPET config register value = 0xFFFFFFFF. "
860 			       "Disabling HPET\n");
861 			goto out_nohpet;
862 		}
863 	}
864 
865 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
866 		goto out_nohpet;
867 
868 	/*
869 	 * Read the HPET ID register to retrieve the IRQ routing
870 	 * information and the number of channels
871 	 */
872 	id = hpet_readl(HPET_ID);
873 	hpet_print_config();
874 
875 #ifdef CONFIG_HPET_EMULATE_RTC
876 	/*
877 	 * The legacy routing mode needs at least two channels, tick timer
878 	 * and the rtc emulation channel.
879 	 */
880 	if (!(id & HPET_ID_NUMBER))
881 		goto out_nohpet;
882 #endif
883 
884 	if (hpet_clocksource_register())
885 		goto out_nohpet;
886 
887 	if (id & HPET_ID_LEGSUP) {
888 		hpet_legacy_clockevent_register();
889 		return 1;
890 	}
891 	return 0;
892 
893 out_nohpet:
894 	hpet_clear_mapping();
895 	hpet_address = 0;
896 	return 0;
897 }
898 
899 /*
900  * Needs to be late, as the reserve_timer code calls kalloc !
901  *
902  * Not a problem on i386 as hpet_enable is called from late_time_init,
903  * but on x86_64 it is necessary !
904  */
905 static __init int hpet_late_init(void)
906 {
907 	int cpu;
908 
909 	if (boot_hpet_disable)
910 		return -ENODEV;
911 
912 	if (!hpet_address) {
913 		if (!force_hpet_address)
914 			return -ENODEV;
915 
916 		hpet_address = force_hpet_address;
917 		hpet_enable();
918 	}
919 
920 	if (!hpet_virt_address)
921 		return -ENODEV;
922 
923 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
924 		hpet_msi_capability_lookup(2);
925 	else
926 		hpet_msi_capability_lookup(0);
927 
928 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
929 	hpet_print_config();
930 
931 	if (boot_cpu_has(X86_FEATURE_ARAT))
932 		return 0;
933 
934 	for_each_online_cpu(cpu) {
935 		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
936 	}
937 
938 	/* This notifier should be called after workqueue is ready */
939 	hotcpu_notifier(hpet_cpuhp_notify, -20);
940 
941 	return 0;
942 }
943 fs_initcall(hpet_late_init);
944 
945 void hpet_disable(void)
946 {
947 	if (is_hpet_capable()) {
948 		unsigned int cfg = hpet_readl(HPET_CFG);
949 
950 		if (hpet_legacy_int_enabled) {
951 			cfg &= ~HPET_CFG_LEGACY;
952 			hpet_legacy_int_enabled = 0;
953 		}
954 		cfg &= ~HPET_CFG_ENABLE;
955 		hpet_writel(cfg, HPET_CFG);
956 	}
957 }
958 
959 #ifdef CONFIG_HPET_EMULATE_RTC
960 
961 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
962  * is enabled, we support RTC interrupt functionality in software.
963  * RTC has 3 kinds of interrupts:
964  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
965  *    is updated
966  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
967  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
968  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
969  * (1) and (2) above are implemented using polling at a frequency of
970  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
971  * overhead. (DEFAULT_RTC_INT_FREQ)
972  * For (3), we use interrupts at 64Hz or user specified periodic
973  * frequency, whichever is higher.
974  */
975 #include <linux/mc146818rtc.h>
976 #include <linux/rtc.h>
977 #include <asm/rtc.h>
978 
979 #define DEFAULT_RTC_INT_FREQ	64
980 #define DEFAULT_RTC_SHIFT	6
981 #define RTC_NUM_INTS		1
982 
983 static unsigned long hpet_rtc_flags;
984 static int hpet_prev_update_sec;
985 static struct rtc_time hpet_alarm_time;
986 static unsigned long hpet_pie_count;
987 static u32 hpet_t1_cmp;
988 static u32 hpet_default_delta;
989 static u32 hpet_pie_delta;
990 static unsigned long hpet_pie_limit;
991 
992 static rtc_irq_handler irq_handler;
993 
994 /*
995  * Check that the hpet counter c1 is ahead of the c2
996  */
997 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
998 {
999 	return (s32)(c2 - c1) < 0;
1000 }
1001 
1002 /*
1003  * Registers a IRQ handler.
1004  */
1005 int hpet_register_irq_handler(rtc_irq_handler handler)
1006 {
1007 	if (!is_hpet_enabled())
1008 		return -ENODEV;
1009 	if (irq_handler)
1010 		return -EBUSY;
1011 
1012 	irq_handler = handler;
1013 
1014 	return 0;
1015 }
1016 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1017 
1018 /*
1019  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1020  * and does cleanup.
1021  */
1022 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1023 {
1024 	if (!is_hpet_enabled())
1025 		return;
1026 
1027 	irq_handler = NULL;
1028 	hpet_rtc_flags = 0;
1029 }
1030 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1031 
1032 /*
1033  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1034  * is not supported by all HPET implementations for timer 1.
1035  *
1036  * hpet_rtc_timer_init() is called when the rtc is initialized.
1037  */
1038 int hpet_rtc_timer_init(void)
1039 {
1040 	unsigned int cfg, cnt, delta;
1041 	unsigned long flags;
1042 
1043 	if (!is_hpet_enabled())
1044 		return 0;
1045 
1046 	if (!hpet_default_delta) {
1047 		uint64_t clc;
1048 
1049 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1050 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1051 		hpet_default_delta = clc;
1052 	}
1053 
1054 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1055 		delta = hpet_default_delta;
1056 	else
1057 		delta = hpet_pie_delta;
1058 
1059 	local_irq_save(flags);
1060 
1061 	cnt = delta + hpet_readl(HPET_COUNTER);
1062 	hpet_writel(cnt, HPET_T1_CMP);
1063 	hpet_t1_cmp = cnt;
1064 
1065 	cfg = hpet_readl(HPET_T1_CFG);
1066 	cfg &= ~HPET_TN_PERIODIC;
1067 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1068 	hpet_writel(cfg, HPET_T1_CFG);
1069 
1070 	local_irq_restore(flags);
1071 
1072 	return 1;
1073 }
1074 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1075 
1076 /*
1077  * The functions below are called from rtc driver.
1078  * Return 0 if HPET is not being used.
1079  * Otherwise do the necessary changes and return 1.
1080  */
1081 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1082 {
1083 	if (!is_hpet_enabled())
1084 		return 0;
1085 
1086 	hpet_rtc_flags &= ~bit_mask;
1087 	return 1;
1088 }
1089 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1090 
1091 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1092 {
1093 	unsigned long oldbits = hpet_rtc_flags;
1094 
1095 	if (!is_hpet_enabled())
1096 		return 0;
1097 
1098 	hpet_rtc_flags |= bit_mask;
1099 
1100 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1101 		hpet_prev_update_sec = -1;
1102 
1103 	if (!oldbits)
1104 		hpet_rtc_timer_init();
1105 
1106 	return 1;
1107 }
1108 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1109 
1110 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1111 			unsigned char sec)
1112 {
1113 	if (!is_hpet_enabled())
1114 		return 0;
1115 
1116 	hpet_alarm_time.tm_hour = hrs;
1117 	hpet_alarm_time.tm_min = min;
1118 	hpet_alarm_time.tm_sec = sec;
1119 
1120 	return 1;
1121 }
1122 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1123 
1124 int hpet_set_periodic_freq(unsigned long freq)
1125 {
1126 	uint64_t clc;
1127 
1128 	if (!is_hpet_enabled())
1129 		return 0;
1130 
1131 	if (freq <= DEFAULT_RTC_INT_FREQ)
1132 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1133 	else {
1134 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1135 		do_div(clc, freq);
1136 		clc >>= hpet_clockevent.shift;
1137 		hpet_pie_delta = clc;
1138 	}
1139 	return 1;
1140 }
1141 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1142 
1143 int hpet_rtc_dropped_irq(void)
1144 {
1145 	return is_hpet_enabled();
1146 }
1147 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1148 
1149 static void hpet_rtc_timer_reinit(void)
1150 {
1151 	unsigned int cfg, delta;
1152 	int lost_ints = -1;
1153 
1154 	if (unlikely(!hpet_rtc_flags)) {
1155 		cfg = hpet_readl(HPET_T1_CFG);
1156 		cfg &= ~HPET_TN_ENABLE;
1157 		hpet_writel(cfg, HPET_T1_CFG);
1158 		return;
1159 	}
1160 
1161 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1162 		delta = hpet_default_delta;
1163 	else
1164 		delta = hpet_pie_delta;
1165 
1166 	/*
1167 	 * Increment the comparator value until we are ahead of the
1168 	 * current count.
1169 	 */
1170 	do {
1171 		hpet_t1_cmp += delta;
1172 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1173 		lost_ints++;
1174 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1175 
1176 	if (lost_ints) {
1177 		if (hpet_rtc_flags & RTC_PIE)
1178 			hpet_pie_count += lost_ints;
1179 		if (printk_ratelimit())
1180 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1181 				lost_ints);
1182 	}
1183 }
1184 
1185 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1186 {
1187 	struct rtc_time curr_time;
1188 	unsigned long rtc_int_flag = 0;
1189 
1190 	hpet_rtc_timer_reinit();
1191 	memset(&curr_time, 0, sizeof(struct rtc_time));
1192 
1193 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1194 		get_rtc_time(&curr_time);
1195 
1196 	if (hpet_rtc_flags & RTC_UIE &&
1197 	    curr_time.tm_sec != hpet_prev_update_sec) {
1198 		if (hpet_prev_update_sec >= 0)
1199 			rtc_int_flag = RTC_UF;
1200 		hpet_prev_update_sec = curr_time.tm_sec;
1201 	}
1202 
1203 	if (hpet_rtc_flags & RTC_PIE &&
1204 	    ++hpet_pie_count >= hpet_pie_limit) {
1205 		rtc_int_flag |= RTC_PF;
1206 		hpet_pie_count = 0;
1207 	}
1208 
1209 	if (hpet_rtc_flags & RTC_AIE &&
1210 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1211 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1212 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1213 			rtc_int_flag |= RTC_AF;
1214 
1215 	if (rtc_int_flag) {
1216 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1217 		if (irq_handler)
1218 			irq_handler(rtc_int_flag, dev_id);
1219 	}
1220 	return IRQ_HANDLED;
1221 }
1222 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1223 #endif
1224