xref: /linux/arch/x86/kernel/hpet.c (revision f8324e20f8289dffc646d64366332e05eaacab25)
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/slab.h>
8 #include <linux/hpet.h>
9 #include <linux/init.h>
10 #include <linux/cpu.h>
11 #include <linux/pm.h>
12 #include <linux/io.h>
13 
14 #include <asm/fixmap.h>
15 #include <asm/i8253.h>
16 #include <asm/hpet.h>
17 
18 #define HPET_MASK			CLOCKSOURCE_MASK(32)
19 #define HPET_SHIFT			22
20 
21 /* FSEC = 10^-15
22    NSEC = 10^-9 */
23 #define FSEC_PER_NSEC			1000000L
24 
25 #define HPET_DEV_USED_BIT		2
26 #define HPET_DEV_USED			(1 << HPET_DEV_USED_BIT)
27 #define HPET_DEV_VALID			0x8
28 #define HPET_DEV_FSB_CAP		0x1000
29 #define HPET_DEV_PERI_CAP		0x2000
30 
31 #define EVT_TO_HPET_DEV(evt) container_of(evt, struct hpet_dev, evt)
32 
33 /*
34  * HPET address is set in acpi/boot.c, when an ACPI entry exists
35  */
36 unsigned long				hpet_address;
37 u8					hpet_blockid; /* OS timer block num */
38 u8					hpet_msi_disable;
39 u8					hpet_readback_cmp;
40 
41 #ifdef CONFIG_PCI_MSI
42 static unsigned long			hpet_num_timers;
43 #endif
44 static void __iomem			*hpet_virt_address;
45 
46 struct hpet_dev {
47 	struct clock_event_device	evt;
48 	unsigned int			num;
49 	int				cpu;
50 	unsigned int			irq;
51 	unsigned int			flags;
52 	char				name[10];
53 };
54 
55 inline unsigned int hpet_readl(unsigned int a)
56 {
57 	return readl(hpet_virt_address + a);
58 }
59 
60 static inline void hpet_writel(unsigned int d, unsigned int a)
61 {
62 	writel(d, hpet_virt_address + a);
63 }
64 
65 #ifdef CONFIG_X86_64
66 #include <asm/pgtable.h>
67 #endif
68 
69 static inline void hpet_set_mapping(void)
70 {
71 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
72 #ifdef CONFIG_X86_64
73 	__set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
74 #endif
75 }
76 
77 static inline void hpet_clear_mapping(void)
78 {
79 	iounmap(hpet_virt_address);
80 	hpet_virt_address = NULL;
81 }
82 
83 /*
84  * HPET command line enable / disable
85  */
86 static int boot_hpet_disable;
87 int hpet_force_user;
88 static int hpet_verbose;
89 
90 static int __init hpet_setup(char *str)
91 {
92 	if (str) {
93 		if (!strncmp("disable", str, 7))
94 			boot_hpet_disable = 1;
95 		if (!strncmp("force", str, 5))
96 			hpet_force_user = 1;
97 		if (!strncmp("verbose", str, 7))
98 			hpet_verbose = 1;
99 	}
100 	return 1;
101 }
102 __setup("hpet=", hpet_setup);
103 
104 static int __init disable_hpet(char *str)
105 {
106 	boot_hpet_disable = 1;
107 	return 1;
108 }
109 __setup("nohpet", disable_hpet);
110 
111 static inline int is_hpet_capable(void)
112 {
113 	return !boot_hpet_disable && hpet_address;
114 }
115 
116 /*
117  * HPET timer interrupt enable / disable
118  */
119 static int hpet_legacy_int_enabled;
120 
121 /**
122  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
123  */
124 int is_hpet_enabled(void)
125 {
126 	return is_hpet_capable() && hpet_legacy_int_enabled;
127 }
128 EXPORT_SYMBOL_GPL(is_hpet_enabled);
129 
130 static void _hpet_print_config(const char *function, int line)
131 {
132 	u32 i, timers, l, h;
133 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
134 	l = hpet_readl(HPET_ID);
135 	h = hpet_readl(HPET_PERIOD);
136 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
137 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
138 	l = hpet_readl(HPET_CFG);
139 	h = hpet_readl(HPET_STATUS);
140 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
141 	l = hpet_readl(HPET_COUNTER);
142 	h = hpet_readl(HPET_COUNTER+4);
143 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
144 
145 	for (i = 0; i < timers; i++) {
146 		l = hpet_readl(HPET_Tn_CFG(i));
147 		h = hpet_readl(HPET_Tn_CFG(i)+4);
148 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
149 		       i, l, h);
150 		l = hpet_readl(HPET_Tn_CMP(i));
151 		h = hpet_readl(HPET_Tn_CMP(i)+4);
152 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
153 		       i, l, h);
154 		l = hpet_readl(HPET_Tn_ROUTE(i));
155 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
156 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
157 		       i, l, h);
158 	}
159 }
160 
161 #define hpet_print_config()					\
162 do {								\
163 	if (hpet_verbose)					\
164 		_hpet_print_config(__FUNCTION__, __LINE__);	\
165 } while (0)
166 
167 /*
168  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
169  * timer 0 and timer 1 in case of RTC emulation.
170  */
171 #ifdef CONFIG_HPET
172 
173 static void hpet_reserve_msi_timers(struct hpet_data *hd);
174 
175 static void hpet_reserve_platform_timers(unsigned int id)
176 {
177 	struct hpet __iomem *hpet = hpet_virt_address;
178 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
179 	unsigned int nrtimers, i;
180 	struct hpet_data hd;
181 
182 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
183 
184 	memset(&hd, 0, sizeof(hd));
185 	hd.hd_phys_address	= hpet_address;
186 	hd.hd_address		= hpet;
187 	hd.hd_nirqs		= nrtimers;
188 	hpet_reserve_timer(&hd, 0);
189 
190 #ifdef CONFIG_HPET_EMULATE_RTC
191 	hpet_reserve_timer(&hd, 1);
192 #endif
193 
194 	/*
195 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
196 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
197 	 * don't bother configuring *any* comparator interrupts.
198 	 */
199 	hd.hd_irq[0] = HPET_LEGACY_8254;
200 	hd.hd_irq[1] = HPET_LEGACY_RTC;
201 
202 	for (i = 2; i < nrtimers; timer++, i++) {
203 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
204 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
205 	}
206 
207 	hpet_reserve_msi_timers(&hd);
208 
209 	hpet_alloc(&hd);
210 
211 }
212 #else
213 static void hpet_reserve_platform_timers(unsigned int id) { }
214 #endif
215 
216 /*
217  * Common hpet info
218  */
219 static unsigned long hpet_period;
220 
221 static void hpet_legacy_set_mode(enum clock_event_mode mode,
222 			  struct clock_event_device *evt);
223 static int hpet_legacy_next_event(unsigned long delta,
224 			   struct clock_event_device *evt);
225 
226 /*
227  * The hpet clock event device
228  */
229 static struct clock_event_device hpet_clockevent = {
230 	.name		= "hpet",
231 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
232 	.set_mode	= hpet_legacy_set_mode,
233 	.set_next_event = hpet_legacy_next_event,
234 	.shift		= 32,
235 	.irq		= 0,
236 	.rating		= 50,
237 };
238 
239 static void hpet_stop_counter(void)
240 {
241 	unsigned long cfg = hpet_readl(HPET_CFG);
242 	cfg &= ~HPET_CFG_ENABLE;
243 	hpet_writel(cfg, HPET_CFG);
244 }
245 
246 static void hpet_reset_counter(void)
247 {
248 	hpet_writel(0, HPET_COUNTER);
249 	hpet_writel(0, HPET_COUNTER + 4);
250 }
251 
252 static void hpet_start_counter(void)
253 {
254 	unsigned int cfg = hpet_readl(HPET_CFG);
255 	cfg |= HPET_CFG_ENABLE;
256 	hpet_writel(cfg, HPET_CFG);
257 }
258 
259 static void hpet_restart_counter(void)
260 {
261 	hpet_stop_counter();
262 	hpet_reset_counter();
263 	hpet_start_counter();
264 }
265 
266 static void hpet_resume_device(void)
267 {
268 	force_hpet_resume();
269 }
270 
271 static void hpet_resume_counter(struct clocksource *cs)
272 {
273 	hpet_resume_device();
274 	hpet_restart_counter();
275 }
276 
277 static void hpet_enable_legacy_int(void)
278 {
279 	unsigned int cfg = hpet_readl(HPET_CFG);
280 
281 	cfg |= HPET_CFG_LEGACY;
282 	hpet_writel(cfg, HPET_CFG);
283 	hpet_legacy_int_enabled = 1;
284 }
285 
286 static void hpet_legacy_clockevent_register(void)
287 {
288 	/* Start HPET legacy interrupts */
289 	hpet_enable_legacy_int();
290 
291 	/*
292 	 * The mult factor is defined as (include/linux/clockchips.h)
293 	 *  mult/2^shift = cyc/ns (in contrast to ns/cyc in clocksource.h)
294 	 * hpet_period is in units of femtoseconds (per cycle), so
295 	 *  mult/2^shift = cyc/ns = 10^6/hpet_period
296 	 *  mult = (10^6 * 2^shift)/hpet_period
297 	 *  mult = (FSEC_PER_NSEC << hpet_clockevent.shift)/hpet_period
298 	 */
299 	hpet_clockevent.mult = div_sc((unsigned long) FSEC_PER_NSEC,
300 				      hpet_period, hpet_clockevent.shift);
301 	/* Calculate the min / max delta */
302 	hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF,
303 							   &hpet_clockevent);
304 	/* 5 usec minimum reprogramming delta. */
305 	hpet_clockevent.min_delta_ns = 5000;
306 
307 	/*
308 	 * Start hpet with the boot cpu mask and make it
309 	 * global after the IO_APIC has been initialized.
310 	 */
311 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
312 	clockevents_register_device(&hpet_clockevent);
313 	global_clock_event = &hpet_clockevent;
314 	printk(KERN_DEBUG "hpet clockevent registered\n");
315 }
316 
317 static int hpet_setup_msi_irq(unsigned int irq);
318 
319 static void hpet_set_mode(enum clock_event_mode mode,
320 			  struct clock_event_device *evt, int timer)
321 {
322 	unsigned int cfg, cmp, now;
323 	uint64_t delta;
324 
325 	switch (mode) {
326 	case CLOCK_EVT_MODE_PERIODIC:
327 		hpet_stop_counter();
328 		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
329 		delta >>= evt->shift;
330 		now = hpet_readl(HPET_COUNTER);
331 		cmp = now + (unsigned int) delta;
332 		cfg = hpet_readl(HPET_Tn_CFG(timer));
333 		/* Make sure we use edge triggered interrupts */
334 		cfg &= ~HPET_TN_LEVEL;
335 		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
336 		       HPET_TN_SETVAL | HPET_TN_32BIT;
337 		hpet_writel(cfg, HPET_Tn_CFG(timer));
338 		hpet_writel(cmp, HPET_Tn_CMP(timer));
339 		udelay(1);
340 		/*
341 		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
342 		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
343 		 * bit is automatically cleared after the first write.
344 		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
345 		 * Publication # 24674)
346 		 */
347 		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
348 		hpet_start_counter();
349 		hpet_print_config();
350 		break;
351 
352 	case CLOCK_EVT_MODE_ONESHOT:
353 		cfg = hpet_readl(HPET_Tn_CFG(timer));
354 		cfg &= ~HPET_TN_PERIODIC;
355 		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
356 		hpet_writel(cfg, HPET_Tn_CFG(timer));
357 		break;
358 
359 	case CLOCK_EVT_MODE_UNUSED:
360 	case CLOCK_EVT_MODE_SHUTDOWN:
361 		cfg = hpet_readl(HPET_Tn_CFG(timer));
362 		cfg &= ~HPET_TN_ENABLE;
363 		hpet_writel(cfg, HPET_Tn_CFG(timer));
364 		break;
365 
366 	case CLOCK_EVT_MODE_RESUME:
367 		if (timer == 0) {
368 			hpet_enable_legacy_int();
369 		} else {
370 			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
371 			hpet_setup_msi_irq(hdev->irq);
372 			disable_irq(hdev->irq);
373 			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
374 			enable_irq(hdev->irq);
375 		}
376 		hpet_print_config();
377 		break;
378 	}
379 }
380 
381 static int hpet_next_event(unsigned long delta,
382 			   struct clock_event_device *evt, int timer)
383 {
384 	u32 cnt;
385 
386 	cnt = hpet_readl(HPET_COUNTER);
387 	cnt += (u32) delta;
388 	hpet_writel(cnt, HPET_Tn_CMP(timer));
389 
390 	/*
391 	 * We need to read back the CMP register on certain HPET
392 	 * implementations (ATI chipsets) which seem to delay the
393 	 * transfer of the compare register into the internal compare
394 	 * logic. With small deltas this might actually be too late as
395 	 * the counter could already be higher than the compare value
396 	 * at that point and we would wait for the next hpet interrupt
397 	 * forever. We found out that reading the CMP register back
398 	 * forces the transfer so we can rely on the comparison with
399 	 * the counter register below.
400 	 *
401 	 * That works fine on those ATI chipsets, but on newer Intel
402 	 * chipsets (ICH9...) this triggers due to an erratum: Reading
403 	 * the comparator immediately following a write is returning
404 	 * the old value.
405 	 *
406 	 * We restrict the read back to the affected ATI chipsets (set
407 	 * by quirks) and also run it with hpet=verbose for debugging
408 	 * purposes.
409 	 */
410 	if (hpet_readback_cmp || hpet_verbose) {
411 		u32 cmp = hpet_readl(HPET_Tn_CMP(timer));
412 
413 		if (cmp != cnt)
414 			printk_once(KERN_WARNING
415 			    "hpet: compare register read back failed.\n");
416 	}
417 
418 	return (s32)(hpet_readl(HPET_COUNTER) - cnt) >= 0 ? -ETIME : 0;
419 }
420 
421 static void hpet_legacy_set_mode(enum clock_event_mode mode,
422 			struct clock_event_device *evt)
423 {
424 	hpet_set_mode(mode, evt, 0);
425 }
426 
427 static int hpet_legacy_next_event(unsigned long delta,
428 			struct clock_event_device *evt)
429 {
430 	return hpet_next_event(delta, evt, 0);
431 }
432 
433 /*
434  * HPET MSI Support
435  */
436 #ifdef CONFIG_PCI_MSI
437 
438 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
439 static struct hpet_dev	*hpet_devs;
440 
441 void hpet_msi_unmask(unsigned int irq)
442 {
443 	struct hpet_dev *hdev = get_irq_data(irq);
444 	unsigned int cfg;
445 
446 	/* unmask it */
447 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
448 	cfg |= HPET_TN_FSB;
449 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
450 }
451 
452 void hpet_msi_mask(unsigned int irq)
453 {
454 	unsigned int cfg;
455 	struct hpet_dev *hdev = get_irq_data(irq);
456 
457 	/* mask it */
458 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
459 	cfg &= ~HPET_TN_FSB;
460 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
461 }
462 
463 void hpet_msi_write(unsigned int irq, struct msi_msg *msg)
464 {
465 	struct hpet_dev *hdev = get_irq_data(irq);
466 
467 	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
468 	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
469 }
470 
471 void hpet_msi_read(unsigned int irq, struct msi_msg *msg)
472 {
473 	struct hpet_dev *hdev = get_irq_data(irq);
474 
475 	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
476 	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
477 	msg->address_hi = 0;
478 }
479 
480 static void hpet_msi_set_mode(enum clock_event_mode mode,
481 				struct clock_event_device *evt)
482 {
483 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
484 	hpet_set_mode(mode, evt, hdev->num);
485 }
486 
487 static int hpet_msi_next_event(unsigned long delta,
488 				struct clock_event_device *evt)
489 {
490 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
491 	return hpet_next_event(delta, evt, hdev->num);
492 }
493 
494 static int hpet_setup_msi_irq(unsigned int irq)
495 {
496 	if (arch_setup_hpet_msi(irq, hpet_blockid)) {
497 		destroy_irq(irq);
498 		return -EINVAL;
499 	}
500 	return 0;
501 }
502 
503 static int hpet_assign_irq(struct hpet_dev *dev)
504 {
505 	unsigned int irq;
506 
507 	irq = create_irq();
508 	if (!irq)
509 		return -EINVAL;
510 
511 	set_irq_data(irq, dev);
512 
513 	if (hpet_setup_msi_irq(irq))
514 		return -EINVAL;
515 
516 	dev->irq = irq;
517 	return 0;
518 }
519 
520 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
521 {
522 	struct hpet_dev *dev = (struct hpet_dev *)data;
523 	struct clock_event_device *hevt = &dev->evt;
524 
525 	if (!hevt->event_handler) {
526 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
527 				dev->num);
528 		return IRQ_HANDLED;
529 	}
530 
531 	hevt->event_handler(hevt);
532 	return IRQ_HANDLED;
533 }
534 
535 static int hpet_setup_irq(struct hpet_dev *dev)
536 {
537 
538 	if (request_irq(dev->irq, hpet_interrupt_handler,
539 			IRQF_TIMER | IRQF_DISABLED | IRQF_NOBALANCING,
540 			dev->name, dev))
541 		return -1;
542 
543 	disable_irq(dev->irq);
544 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
545 	enable_irq(dev->irq);
546 
547 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
548 			 dev->name, dev->irq);
549 
550 	return 0;
551 }
552 
553 /* This should be called in specific @cpu */
554 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
555 {
556 	struct clock_event_device *evt = &hdev->evt;
557 	uint64_t hpet_freq;
558 
559 	WARN_ON(cpu != smp_processor_id());
560 	if (!(hdev->flags & HPET_DEV_VALID))
561 		return;
562 
563 	if (hpet_setup_msi_irq(hdev->irq))
564 		return;
565 
566 	hdev->cpu = cpu;
567 	per_cpu(cpu_hpet_dev, cpu) = hdev;
568 	evt->name = hdev->name;
569 	hpet_setup_irq(hdev);
570 	evt->irq = hdev->irq;
571 
572 	evt->rating = 110;
573 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
574 	if (hdev->flags & HPET_DEV_PERI_CAP)
575 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
576 
577 	evt->set_mode = hpet_msi_set_mode;
578 	evt->set_next_event = hpet_msi_next_event;
579 	evt->shift = 32;
580 
581 	/*
582 	 * The period is a femto seconds value. We need to calculate the
583 	 * scaled math multiplication factor for nanosecond to hpet tick
584 	 * conversion.
585 	 */
586 	hpet_freq = 1000000000000000ULL;
587 	do_div(hpet_freq, hpet_period);
588 	evt->mult = div_sc((unsigned long) hpet_freq,
589 				      NSEC_PER_SEC, evt->shift);
590 	/* Calculate the max delta */
591 	evt->max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, evt);
592 	/* 5 usec minimum reprogramming delta. */
593 	evt->min_delta_ns = 5000;
594 
595 	evt->cpumask = cpumask_of(hdev->cpu);
596 	clockevents_register_device(evt);
597 }
598 
599 #ifdef CONFIG_HPET
600 /* Reserve at least one timer for userspace (/dev/hpet) */
601 #define RESERVE_TIMERS 1
602 #else
603 #define RESERVE_TIMERS 0
604 #endif
605 
606 static void hpet_msi_capability_lookup(unsigned int start_timer)
607 {
608 	unsigned int id;
609 	unsigned int num_timers;
610 	unsigned int num_timers_used = 0;
611 	int i;
612 
613 	if (hpet_msi_disable)
614 		return;
615 
616 	if (boot_cpu_has(X86_FEATURE_ARAT))
617 		return;
618 	id = hpet_readl(HPET_ID);
619 
620 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
621 	num_timers++; /* Value read out starts from 0 */
622 	hpet_print_config();
623 
624 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
625 	if (!hpet_devs)
626 		return;
627 
628 	hpet_num_timers = num_timers;
629 
630 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
631 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
632 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
633 
634 		/* Only consider HPET timer with MSI support */
635 		if (!(cfg & HPET_TN_FSB_CAP))
636 			continue;
637 
638 		hdev->flags = 0;
639 		if (cfg & HPET_TN_PERIODIC_CAP)
640 			hdev->flags |= HPET_DEV_PERI_CAP;
641 		hdev->num = i;
642 
643 		sprintf(hdev->name, "hpet%d", i);
644 		if (hpet_assign_irq(hdev))
645 			continue;
646 
647 		hdev->flags |= HPET_DEV_FSB_CAP;
648 		hdev->flags |= HPET_DEV_VALID;
649 		num_timers_used++;
650 		if (num_timers_used == num_possible_cpus())
651 			break;
652 	}
653 
654 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
655 		num_timers, num_timers_used);
656 }
657 
658 #ifdef CONFIG_HPET
659 static void hpet_reserve_msi_timers(struct hpet_data *hd)
660 {
661 	int i;
662 
663 	if (!hpet_devs)
664 		return;
665 
666 	for (i = 0; i < hpet_num_timers; i++) {
667 		struct hpet_dev *hdev = &hpet_devs[i];
668 
669 		if (!(hdev->flags & HPET_DEV_VALID))
670 			continue;
671 
672 		hd->hd_irq[hdev->num] = hdev->irq;
673 		hpet_reserve_timer(hd, hdev->num);
674 	}
675 }
676 #endif
677 
678 static struct hpet_dev *hpet_get_unused_timer(void)
679 {
680 	int i;
681 
682 	if (!hpet_devs)
683 		return NULL;
684 
685 	for (i = 0; i < hpet_num_timers; i++) {
686 		struct hpet_dev *hdev = &hpet_devs[i];
687 
688 		if (!(hdev->flags & HPET_DEV_VALID))
689 			continue;
690 		if (test_and_set_bit(HPET_DEV_USED_BIT,
691 			(unsigned long *)&hdev->flags))
692 			continue;
693 		return hdev;
694 	}
695 	return NULL;
696 }
697 
698 struct hpet_work_struct {
699 	struct delayed_work work;
700 	struct completion complete;
701 };
702 
703 static void hpet_work(struct work_struct *w)
704 {
705 	struct hpet_dev *hdev;
706 	int cpu = smp_processor_id();
707 	struct hpet_work_struct *hpet_work;
708 
709 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
710 
711 	hdev = hpet_get_unused_timer();
712 	if (hdev)
713 		init_one_hpet_msi_clockevent(hdev, cpu);
714 
715 	complete(&hpet_work->complete);
716 }
717 
718 static int hpet_cpuhp_notify(struct notifier_block *n,
719 		unsigned long action, void *hcpu)
720 {
721 	unsigned long cpu = (unsigned long)hcpu;
722 	struct hpet_work_struct work;
723 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
724 
725 	switch (action & 0xf) {
726 	case CPU_ONLINE:
727 		INIT_DELAYED_WORK_ON_STACK(&work.work, hpet_work);
728 		init_completion(&work.complete);
729 		/* FIXME: add schedule_work_on() */
730 		schedule_delayed_work_on(cpu, &work.work, 0);
731 		wait_for_completion(&work.complete);
732 		destroy_timer_on_stack(&work.work.timer);
733 		break;
734 	case CPU_DEAD:
735 		if (hdev) {
736 			free_irq(hdev->irq, hdev);
737 			hdev->flags &= ~HPET_DEV_USED;
738 			per_cpu(cpu_hpet_dev, cpu) = NULL;
739 		}
740 		break;
741 	}
742 	return NOTIFY_OK;
743 }
744 #else
745 
746 static int hpet_setup_msi_irq(unsigned int irq)
747 {
748 	return 0;
749 }
750 static void hpet_msi_capability_lookup(unsigned int start_timer)
751 {
752 	return;
753 }
754 
755 #ifdef CONFIG_HPET
756 static void hpet_reserve_msi_timers(struct hpet_data *hd)
757 {
758 	return;
759 }
760 #endif
761 
762 static int hpet_cpuhp_notify(struct notifier_block *n,
763 		unsigned long action, void *hcpu)
764 {
765 	return NOTIFY_OK;
766 }
767 
768 #endif
769 
770 /*
771  * Clock source related code
772  */
773 static cycle_t read_hpet(struct clocksource *cs)
774 {
775 	return (cycle_t)hpet_readl(HPET_COUNTER);
776 }
777 
778 #ifdef CONFIG_X86_64
779 static cycle_t __vsyscall_fn vread_hpet(void)
780 {
781 	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + 0xf0);
782 }
783 #endif
784 
785 static struct clocksource clocksource_hpet = {
786 	.name		= "hpet",
787 	.rating		= 250,
788 	.read		= read_hpet,
789 	.mask		= HPET_MASK,
790 	.shift		= HPET_SHIFT,
791 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
792 	.resume		= hpet_resume_counter,
793 #ifdef CONFIG_X86_64
794 	.vread		= vread_hpet,
795 #endif
796 };
797 
798 static int hpet_clocksource_register(void)
799 {
800 	u64 start, now;
801 	cycle_t t1;
802 
803 	/* Start the counter */
804 	hpet_restart_counter();
805 
806 	/* Verify whether hpet counter works */
807 	t1 = hpet_readl(HPET_COUNTER);
808 	rdtscll(start);
809 
810 	/*
811 	 * We don't know the TSC frequency yet, but waiting for
812 	 * 200000 TSC cycles is safe:
813 	 * 4 GHz == 50us
814 	 * 1 GHz == 200us
815 	 */
816 	do {
817 		rep_nop();
818 		rdtscll(now);
819 	} while ((now - start) < 200000UL);
820 
821 	if (t1 == hpet_readl(HPET_COUNTER)) {
822 		printk(KERN_WARNING
823 		       "HPET counter not counting. HPET disabled\n");
824 		return -ENODEV;
825 	}
826 
827 	/*
828 	 * The definition of mult is (include/linux/clocksource.h)
829 	 * mult/2^shift = ns/cyc and hpet_period is in units of fsec/cyc
830 	 * so we first need to convert hpet_period to ns/cyc units:
831 	 *  mult/2^shift = ns/cyc = hpet_period/10^6
832 	 *  mult = (hpet_period * 2^shift)/10^6
833 	 *  mult = (hpet_period << shift)/FSEC_PER_NSEC
834 	 */
835 	clocksource_hpet.mult = div_sc(hpet_period, FSEC_PER_NSEC, HPET_SHIFT);
836 
837 	clocksource_register(&clocksource_hpet);
838 
839 	return 0;
840 }
841 
842 /**
843  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
844  */
845 int __init hpet_enable(void)
846 {
847 	unsigned int id;
848 	int i;
849 
850 	if (!is_hpet_capable())
851 		return 0;
852 
853 	hpet_set_mapping();
854 
855 	/*
856 	 * Read the period and check for a sane value:
857 	 */
858 	hpet_period = hpet_readl(HPET_PERIOD);
859 
860 	/*
861 	 * AMD SB700 based systems with spread spectrum enabled use a
862 	 * SMM based HPET emulation to provide proper frequency
863 	 * setting. The SMM code is initialized with the first HPET
864 	 * register access and takes some time to complete. During
865 	 * this time the config register reads 0xffffffff. We check
866 	 * for max. 1000 loops whether the config register reads a non
867 	 * 0xffffffff value to make sure that HPET is up and running
868 	 * before we go further. A counting loop is safe, as the HPET
869 	 * access takes thousands of CPU cycles. On non SB700 based
870 	 * machines this check is only done once and has no side
871 	 * effects.
872 	 */
873 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
874 		if (i == 1000) {
875 			printk(KERN_WARNING
876 			       "HPET config register value = 0xFFFFFFFF. "
877 			       "Disabling HPET\n");
878 			goto out_nohpet;
879 		}
880 	}
881 
882 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
883 		goto out_nohpet;
884 
885 	/*
886 	 * Read the HPET ID register to retrieve the IRQ routing
887 	 * information and the number of channels
888 	 */
889 	id = hpet_readl(HPET_ID);
890 	hpet_print_config();
891 
892 #ifdef CONFIG_HPET_EMULATE_RTC
893 	/*
894 	 * The legacy routing mode needs at least two channels, tick timer
895 	 * and the rtc emulation channel.
896 	 */
897 	if (!(id & HPET_ID_NUMBER))
898 		goto out_nohpet;
899 #endif
900 
901 	if (hpet_clocksource_register())
902 		goto out_nohpet;
903 
904 	if (id & HPET_ID_LEGSUP) {
905 		hpet_legacy_clockevent_register();
906 		return 1;
907 	}
908 	return 0;
909 
910 out_nohpet:
911 	hpet_clear_mapping();
912 	hpet_address = 0;
913 	return 0;
914 }
915 
916 /*
917  * Needs to be late, as the reserve_timer code calls kalloc !
918  *
919  * Not a problem on i386 as hpet_enable is called from late_time_init,
920  * but on x86_64 it is necessary !
921  */
922 static __init int hpet_late_init(void)
923 {
924 	int cpu;
925 
926 	if (boot_hpet_disable)
927 		return -ENODEV;
928 
929 	if (!hpet_address) {
930 		if (!force_hpet_address)
931 			return -ENODEV;
932 
933 		hpet_address = force_hpet_address;
934 		hpet_enable();
935 	}
936 
937 	if (!hpet_virt_address)
938 		return -ENODEV;
939 
940 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
941 		hpet_msi_capability_lookup(2);
942 	else
943 		hpet_msi_capability_lookup(0);
944 
945 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
946 	hpet_print_config();
947 
948 	if (hpet_msi_disable)
949 		return 0;
950 
951 	if (boot_cpu_has(X86_FEATURE_ARAT))
952 		return 0;
953 
954 	for_each_online_cpu(cpu) {
955 		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
956 	}
957 
958 	/* This notifier should be called after workqueue is ready */
959 	hotcpu_notifier(hpet_cpuhp_notify, -20);
960 
961 	return 0;
962 }
963 fs_initcall(hpet_late_init);
964 
965 void hpet_disable(void)
966 {
967 	if (is_hpet_capable()) {
968 		unsigned int cfg = hpet_readl(HPET_CFG);
969 
970 		if (hpet_legacy_int_enabled) {
971 			cfg &= ~HPET_CFG_LEGACY;
972 			hpet_legacy_int_enabled = 0;
973 		}
974 		cfg &= ~HPET_CFG_ENABLE;
975 		hpet_writel(cfg, HPET_CFG);
976 	}
977 }
978 
979 #ifdef CONFIG_HPET_EMULATE_RTC
980 
981 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
982  * is enabled, we support RTC interrupt functionality in software.
983  * RTC has 3 kinds of interrupts:
984  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
985  *    is updated
986  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
987  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
988  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
989  * (1) and (2) above are implemented using polling at a frequency of
990  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
991  * overhead. (DEFAULT_RTC_INT_FREQ)
992  * For (3), we use interrupts at 64Hz or user specified periodic
993  * frequency, whichever is higher.
994  */
995 #include <linux/mc146818rtc.h>
996 #include <linux/rtc.h>
997 #include <asm/rtc.h>
998 
999 #define DEFAULT_RTC_INT_FREQ	64
1000 #define DEFAULT_RTC_SHIFT	6
1001 #define RTC_NUM_INTS		1
1002 
1003 static unsigned long hpet_rtc_flags;
1004 static int hpet_prev_update_sec;
1005 static struct rtc_time hpet_alarm_time;
1006 static unsigned long hpet_pie_count;
1007 static u32 hpet_t1_cmp;
1008 static u32 hpet_default_delta;
1009 static u32 hpet_pie_delta;
1010 static unsigned long hpet_pie_limit;
1011 
1012 static rtc_irq_handler irq_handler;
1013 
1014 /*
1015  * Check that the hpet counter c1 is ahead of the c2
1016  */
1017 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1018 {
1019 	return (s32)(c2 - c1) < 0;
1020 }
1021 
1022 /*
1023  * Registers a IRQ handler.
1024  */
1025 int hpet_register_irq_handler(rtc_irq_handler handler)
1026 {
1027 	if (!is_hpet_enabled())
1028 		return -ENODEV;
1029 	if (irq_handler)
1030 		return -EBUSY;
1031 
1032 	irq_handler = handler;
1033 
1034 	return 0;
1035 }
1036 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1037 
1038 /*
1039  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1040  * and does cleanup.
1041  */
1042 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1043 {
1044 	if (!is_hpet_enabled())
1045 		return;
1046 
1047 	irq_handler = NULL;
1048 	hpet_rtc_flags = 0;
1049 }
1050 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1051 
1052 /*
1053  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1054  * is not supported by all HPET implementations for timer 1.
1055  *
1056  * hpet_rtc_timer_init() is called when the rtc is initialized.
1057  */
1058 int hpet_rtc_timer_init(void)
1059 {
1060 	unsigned int cfg, cnt, delta;
1061 	unsigned long flags;
1062 
1063 	if (!is_hpet_enabled())
1064 		return 0;
1065 
1066 	if (!hpet_default_delta) {
1067 		uint64_t clc;
1068 
1069 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1070 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1071 		hpet_default_delta = clc;
1072 	}
1073 
1074 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1075 		delta = hpet_default_delta;
1076 	else
1077 		delta = hpet_pie_delta;
1078 
1079 	local_irq_save(flags);
1080 
1081 	cnt = delta + hpet_readl(HPET_COUNTER);
1082 	hpet_writel(cnt, HPET_T1_CMP);
1083 	hpet_t1_cmp = cnt;
1084 
1085 	cfg = hpet_readl(HPET_T1_CFG);
1086 	cfg &= ~HPET_TN_PERIODIC;
1087 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1088 	hpet_writel(cfg, HPET_T1_CFG);
1089 
1090 	local_irq_restore(flags);
1091 
1092 	return 1;
1093 }
1094 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1095 
1096 /*
1097  * The functions below are called from rtc driver.
1098  * Return 0 if HPET is not being used.
1099  * Otherwise do the necessary changes and return 1.
1100  */
1101 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1102 {
1103 	if (!is_hpet_enabled())
1104 		return 0;
1105 
1106 	hpet_rtc_flags &= ~bit_mask;
1107 	return 1;
1108 }
1109 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1110 
1111 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1112 {
1113 	unsigned long oldbits = hpet_rtc_flags;
1114 
1115 	if (!is_hpet_enabled())
1116 		return 0;
1117 
1118 	hpet_rtc_flags |= bit_mask;
1119 
1120 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1121 		hpet_prev_update_sec = -1;
1122 
1123 	if (!oldbits)
1124 		hpet_rtc_timer_init();
1125 
1126 	return 1;
1127 }
1128 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1129 
1130 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1131 			unsigned char sec)
1132 {
1133 	if (!is_hpet_enabled())
1134 		return 0;
1135 
1136 	hpet_alarm_time.tm_hour = hrs;
1137 	hpet_alarm_time.tm_min = min;
1138 	hpet_alarm_time.tm_sec = sec;
1139 
1140 	return 1;
1141 }
1142 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1143 
1144 int hpet_set_periodic_freq(unsigned long freq)
1145 {
1146 	uint64_t clc;
1147 
1148 	if (!is_hpet_enabled())
1149 		return 0;
1150 
1151 	if (freq <= DEFAULT_RTC_INT_FREQ)
1152 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1153 	else {
1154 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1155 		do_div(clc, freq);
1156 		clc >>= hpet_clockevent.shift;
1157 		hpet_pie_delta = clc;
1158 		hpet_pie_limit = 0;
1159 	}
1160 	return 1;
1161 }
1162 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1163 
1164 int hpet_rtc_dropped_irq(void)
1165 {
1166 	return is_hpet_enabled();
1167 }
1168 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1169 
1170 static void hpet_rtc_timer_reinit(void)
1171 {
1172 	unsigned int cfg, delta;
1173 	int lost_ints = -1;
1174 
1175 	if (unlikely(!hpet_rtc_flags)) {
1176 		cfg = hpet_readl(HPET_T1_CFG);
1177 		cfg &= ~HPET_TN_ENABLE;
1178 		hpet_writel(cfg, HPET_T1_CFG);
1179 		return;
1180 	}
1181 
1182 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1183 		delta = hpet_default_delta;
1184 	else
1185 		delta = hpet_pie_delta;
1186 
1187 	/*
1188 	 * Increment the comparator value until we are ahead of the
1189 	 * current count.
1190 	 */
1191 	do {
1192 		hpet_t1_cmp += delta;
1193 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1194 		lost_ints++;
1195 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1196 
1197 	if (lost_ints) {
1198 		if (hpet_rtc_flags & RTC_PIE)
1199 			hpet_pie_count += lost_ints;
1200 		if (printk_ratelimit())
1201 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1202 				lost_ints);
1203 	}
1204 }
1205 
1206 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1207 {
1208 	struct rtc_time curr_time;
1209 	unsigned long rtc_int_flag = 0;
1210 
1211 	hpet_rtc_timer_reinit();
1212 	memset(&curr_time, 0, sizeof(struct rtc_time));
1213 
1214 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1215 		get_rtc_time(&curr_time);
1216 
1217 	if (hpet_rtc_flags & RTC_UIE &&
1218 	    curr_time.tm_sec != hpet_prev_update_sec) {
1219 		if (hpet_prev_update_sec >= 0)
1220 			rtc_int_flag = RTC_UF;
1221 		hpet_prev_update_sec = curr_time.tm_sec;
1222 	}
1223 
1224 	if (hpet_rtc_flags & RTC_PIE &&
1225 	    ++hpet_pie_count >= hpet_pie_limit) {
1226 		rtc_int_flag |= RTC_PF;
1227 		hpet_pie_count = 0;
1228 	}
1229 
1230 	if (hpet_rtc_flags & RTC_AIE &&
1231 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1232 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1233 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1234 			rtc_int_flag |= RTC_AF;
1235 
1236 	if (rtc_int_flag) {
1237 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1238 		if (irq_handler)
1239 			irq_handler(rtc_int_flag, dev_id);
1240 	}
1241 	return IRQ_HANDLED;
1242 }
1243 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1244 #endif
1245