xref: /linux/arch/x86/kernel/hpet.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
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 
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 u8					hpet_readback_cmp;
39 
40 #ifdef CONFIG_PCI_MSI
41 static unsigned long			hpet_num_timers;
42 #endif
43 static void __iomem			*hpet_virt_address;
44 
45 struct hpet_dev {
46 	struct clock_event_device	evt;
47 	unsigned int			num;
48 	int				cpu;
49 	unsigned int			irq;
50 	unsigned int			flags;
51 	char				name[10];
52 };
53 
54 inline unsigned int hpet_readl(unsigned int a)
55 {
56 	return readl(hpet_virt_address + a);
57 }
58 
59 static inline void hpet_writel(unsigned int d, unsigned int a)
60 {
61 	writel(d, hpet_virt_address + a);
62 }
63 
64 #ifdef CONFIG_X86_64
65 #include <asm/pgtable.h>
66 #endif
67 
68 static inline void hpet_set_mapping(void)
69 {
70 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
71 #ifdef CONFIG_X86_64
72 	__set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
73 #endif
74 }
75 
76 static inline void hpet_clear_mapping(void)
77 {
78 	iounmap(hpet_virt_address);
79 	hpet_virt_address = NULL;
80 }
81 
82 /*
83  * HPET command line enable / disable
84  */
85 static int boot_hpet_disable;
86 int hpet_force_user;
87 static int hpet_verbose;
88 
89 static int __init hpet_setup(char *str)
90 {
91 	if (str) {
92 		if (!strncmp("disable", str, 7))
93 			boot_hpet_disable = 1;
94 		if (!strncmp("force", str, 5))
95 			hpet_force_user = 1;
96 		if (!strncmp("verbose", str, 7))
97 			hpet_verbose = 1;
98 	}
99 	return 1;
100 }
101 __setup("hpet=", hpet_setup);
102 
103 static int __init disable_hpet(char *str)
104 {
105 	boot_hpet_disable = 1;
106 	return 1;
107 }
108 __setup("nohpet", disable_hpet);
109 
110 static inline int is_hpet_capable(void)
111 {
112 	return !boot_hpet_disable && hpet_address;
113 }
114 
115 /*
116  * HPET timer interrupt enable / disable
117  */
118 static int hpet_legacy_int_enabled;
119 
120 /**
121  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
122  */
123 int is_hpet_enabled(void)
124 {
125 	return is_hpet_capable() && hpet_legacy_int_enabled;
126 }
127 EXPORT_SYMBOL_GPL(is_hpet_enabled);
128 
129 static void _hpet_print_config(const char *function, int line)
130 {
131 	u32 i, timers, l, h;
132 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
133 	l = hpet_readl(HPET_ID);
134 	h = hpet_readl(HPET_PERIOD);
135 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
136 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
137 	l = hpet_readl(HPET_CFG);
138 	h = hpet_readl(HPET_STATUS);
139 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
140 	l = hpet_readl(HPET_COUNTER);
141 	h = hpet_readl(HPET_COUNTER+4);
142 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
143 
144 	for (i = 0; i < timers; i++) {
145 		l = hpet_readl(HPET_Tn_CFG(i));
146 		h = hpet_readl(HPET_Tn_CFG(i)+4);
147 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
148 		       i, l, h);
149 		l = hpet_readl(HPET_Tn_CMP(i));
150 		h = hpet_readl(HPET_Tn_CMP(i)+4);
151 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
152 		       i, l, h);
153 		l = hpet_readl(HPET_Tn_ROUTE(i));
154 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
155 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
156 		       i, l, h);
157 	}
158 }
159 
160 #define hpet_print_config()					\
161 do {								\
162 	if (hpet_verbose)					\
163 		_hpet_print_config(__FUNCTION__, __LINE__);	\
164 } while (0)
165 
166 /*
167  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
168  * timer 0 and timer 1 in case of RTC emulation.
169  */
170 #ifdef CONFIG_HPET
171 
172 static void hpet_reserve_msi_timers(struct hpet_data *hd);
173 
174 static void hpet_reserve_platform_timers(unsigned int id)
175 {
176 	struct hpet __iomem *hpet = hpet_virt_address;
177 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
178 	unsigned int nrtimers, i;
179 	struct hpet_data hd;
180 
181 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
182 
183 	memset(&hd, 0, sizeof(hd));
184 	hd.hd_phys_address	= hpet_address;
185 	hd.hd_address		= hpet;
186 	hd.hd_nirqs		= nrtimers;
187 	hpet_reserve_timer(&hd, 0);
188 
189 #ifdef CONFIG_HPET_EMULATE_RTC
190 	hpet_reserve_timer(&hd, 1);
191 #endif
192 
193 	/*
194 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
195 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
196 	 * don't bother configuring *any* comparator interrupts.
197 	 */
198 	hd.hd_irq[0] = HPET_LEGACY_8254;
199 	hd.hd_irq[1] = HPET_LEGACY_RTC;
200 
201 	for (i = 2; i < nrtimers; timer++, i++) {
202 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
203 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
204 	}
205 
206 	hpet_reserve_msi_timers(&hd);
207 
208 	hpet_alloc(&hd);
209 
210 }
211 #else
212 static void hpet_reserve_platform_timers(unsigned int id) { }
213 #endif
214 
215 /*
216  * Common hpet info
217  */
218 static unsigned long hpet_period;
219 
220 static void hpet_legacy_set_mode(enum clock_event_mode mode,
221 			  struct clock_event_device *evt);
222 static int hpet_legacy_next_event(unsigned long delta,
223 			   struct clock_event_device *evt);
224 
225 /*
226  * The hpet clock event device
227  */
228 static struct clock_event_device hpet_clockevent = {
229 	.name		= "hpet",
230 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
231 	.set_mode	= hpet_legacy_set_mode,
232 	.set_next_event = hpet_legacy_next_event,
233 	.shift		= 32,
234 	.irq		= 0,
235 	.rating		= 50,
236 };
237 
238 static void hpet_stop_counter(void)
239 {
240 	unsigned long cfg = hpet_readl(HPET_CFG);
241 	cfg &= ~HPET_CFG_ENABLE;
242 	hpet_writel(cfg, HPET_CFG);
243 }
244 
245 static void hpet_reset_counter(void)
246 {
247 	hpet_writel(0, HPET_COUNTER);
248 	hpet_writel(0, HPET_COUNTER + 4);
249 }
250 
251 static void hpet_start_counter(void)
252 {
253 	unsigned int cfg = hpet_readl(HPET_CFG);
254 	cfg |= HPET_CFG_ENABLE;
255 	hpet_writel(cfg, HPET_CFG);
256 }
257 
258 static void hpet_restart_counter(void)
259 {
260 	hpet_stop_counter();
261 	hpet_reset_counter();
262 	hpet_start_counter();
263 }
264 
265 static void hpet_resume_device(void)
266 {
267 	force_hpet_resume();
268 }
269 
270 static void hpet_resume_counter(struct clocksource *cs)
271 {
272 	hpet_resume_device();
273 	hpet_restart_counter();
274 }
275 
276 static void hpet_enable_legacy_int(void)
277 {
278 	unsigned int cfg = hpet_readl(HPET_CFG);
279 
280 	cfg |= HPET_CFG_LEGACY;
281 	hpet_writel(cfg, HPET_CFG);
282 	hpet_legacy_int_enabled = 1;
283 }
284 
285 static void hpet_legacy_clockevent_register(void)
286 {
287 	/* Start HPET legacy interrupts */
288 	hpet_enable_legacy_int();
289 
290 	/*
291 	 * The mult factor is defined as (include/linux/clockchips.h)
292 	 *  mult/2^shift = cyc/ns (in contrast to ns/cyc in clocksource.h)
293 	 * hpet_period is in units of femtoseconds (per cycle), so
294 	 *  mult/2^shift = cyc/ns = 10^6/hpet_period
295 	 *  mult = (10^6 * 2^shift)/hpet_period
296 	 *  mult = (FSEC_PER_NSEC << hpet_clockevent.shift)/hpet_period
297 	 */
298 	hpet_clockevent.mult = div_sc((unsigned long) FSEC_PER_NSEC,
299 				      hpet_period, hpet_clockevent.shift);
300 	/* Calculate the min / max delta */
301 	hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF,
302 							   &hpet_clockevent);
303 	/* 5 usec minimum reprogramming delta. */
304 	hpet_clockevent.min_delta_ns = 5000;
305 
306 	/*
307 	 * Start hpet with the boot cpu mask and make it
308 	 * global after the IO_APIC has been initialized.
309 	 */
310 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
311 	clockevents_register_device(&hpet_clockevent);
312 	global_clock_event = &hpet_clockevent;
313 	printk(KERN_DEBUG "hpet clockevent registered\n");
314 }
315 
316 static int hpet_setup_msi_irq(unsigned int irq);
317 
318 static void hpet_set_mode(enum clock_event_mode mode,
319 			  struct clock_event_device *evt, int timer)
320 {
321 	unsigned int cfg, cmp, now;
322 	uint64_t delta;
323 
324 	switch (mode) {
325 	case CLOCK_EVT_MODE_PERIODIC:
326 		hpet_stop_counter();
327 		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
328 		delta >>= evt->shift;
329 		now = hpet_readl(HPET_COUNTER);
330 		cmp = now + (unsigned int) delta;
331 		cfg = hpet_readl(HPET_Tn_CFG(timer));
332 		/* Make sure we use edge triggered interrupts */
333 		cfg &= ~HPET_TN_LEVEL;
334 		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
335 		       HPET_TN_SETVAL | HPET_TN_32BIT;
336 		hpet_writel(cfg, HPET_Tn_CFG(timer));
337 		hpet_writel(cmp, HPET_Tn_CMP(timer));
338 		udelay(1);
339 		/*
340 		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
341 		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
342 		 * bit is automatically cleared after the first write.
343 		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
344 		 * Publication # 24674)
345 		 */
346 		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
347 		hpet_start_counter();
348 		hpet_print_config();
349 		break;
350 
351 	case CLOCK_EVT_MODE_ONESHOT:
352 		cfg = hpet_readl(HPET_Tn_CFG(timer));
353 		cfg &= ~HPET_TN_PERIODIC;
354 		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
355 		hpet_writel(cfg, HPET_Tn_CFG(timer));
356 		break;
357 
358 	case CLOCK_EVT_MODE_UNUSED:
359 	case CLOCK_EVT_MODE_SHUTDOWN:
360 		cfg = hpet_readl(HPET_Tn_CFG(timer));
361 		cfg &= ~HPET_TN_ENABLE;
362 		hpet_writel(cfg, HPET_Tn_CFG(timer));
363 		break;
364 
365 	case CLOCK_EVT_MODE_RESUME:
366 		if (timer == 0) {
367 			hpet_enable_legacy_int();
368 		} else {
369 			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
370 			hpet_setup_msi_irq(hdev->irq);
371 			disable_irq(hdev->irq);
372 			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
373 			enable_irq(hdev->irq);
374 		}
375 		hpet_print_config();
376 		break;
377 	}
378 }
379 
380 static int hpet_next_event(unsigned long delta,
381 			   struct clock_event_device *evt, int timer)
382 {
383 	u32 cnt;
384 
385 	cnt = hpet_readl(HPET_COUNTER);
386 	cnt += (u32) delta;
387 	hpet_writel(cnt, HPET_Tn_CMP(timer));
388 
389 	/*
390 	 * We need to read back the CMP register on certain HPET
391 	 * implementations (ATI chipsets) which seem to delay the
392 	 * transfer of the compare register into the internal compare
393 	 * logic. With small deltas this might actually be too late as
394 	 * the counter could already be higher than the compare value
395 	 * at that point and we would wait for the next hpet interrupt
396 	 * forever. We found out that reading the CMP register back
397 	 * forces the transfer so we can rely on the comparison with
398 	 * the counter register below.
399 	 *
400 	 * That works fine on those ATI chipsets, but on newer Intel
401 	 * chipsets (ICH9...) this triggers due to an erratum: Reading
402 	 * the comparator immediately following a write is returning
403 	 * the old value.
404 	 *
405 	 * We restrict the read back to the affected ATI chipsets (set
406 	 * by quirks) and also run it with hpet=verbose for debugging
407 	 * purposes.
408 	 */
409 	if (hpet_readback_cmp || hpet_verbose) {
410 		u32 cmp = hpet_readl(HPET_Tn_CMP(timer));
411 
412 		if (cmp != cnt)
413 			printk_once(KERN_WARNING
414 			    "hpet: compare register read back failed.\n");
415 	}
416 
417 	return (s32)(hpet_readl(HPET_COUNTER) - cnt) >= 0 ? -ETIME : 0;
418 }
419 
420 static void hpet_legacy_set_mode(enum clock_event_mode mode,
421 			struct clock_event_device *evt)
422 {
423 	hpet_set_mode(mode, evt, 0);
424 }
425 
426 static int hpet_legacy_next_event(unsigned long delta,
427 			struct clock_event_device *evt)
428 {
429 	return hpet_next_event(delta, evt, 0);
430 }
431 
432 /*
433  * HPET MSI Support
434  */
435 #ifdef CONFIG_PCI_MSI
436 
437 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
438 static struct hpet_dev	*hpet_devs;
439 
440 void hpet_msi_unmask(unsigned int irq)
441 {
442 	struct hpet_dev *hdev = get_irq_data(irq);
443 	unsigned int cfg;
444 
445 	/* unmask 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_mask(unsigned int irq)
452 {
453 	unsigned int cfg;
454 	struct hpet_dev *hdev = get_irq_data(irq);
455 
456 	/* mask it */
457 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
458 	cfg &= ~HPET_TN_FSB;
459 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
460 }
461 
462 void hpet_msi_write(unsigned int irq, struct msi_msg *msg)
463 {
464 	struct hpet_dev *hdev = get_irq_data(irq);
465 
466 	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
467 	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
468 }
469 
470 void hpet_msi_read(unsigned int irq, struct msi_msg *msg)
471 {
472 	struct hpet_dev *hdev = get_irq_data(irq);
473 
474 	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
475 	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
476 	msg->address_hi = 0;
477 }
478 
479 static void hpet_msi_set_mode(enum clock_event_mode mode,
480 				struct clock_event_device *evt)
481 {
482 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
483 	hpet_set_mode(mode, evt, hdev->num);
484 }
485 
486 static int hpet_msi_next_event(unsigned long delta,
487 				struct clock_event_device *evt)
488 {
489 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
490 	return hpet_next_event(delta, evt, hdev->num);
491 }
492 
493 static int hpet_setup_msi_irq(unsigned int irq)
494 {
495 	if (arch_setup_hpet_msi(irq, hpet_blockid)) {
496 		destroy_irq(irq);
497 		return -EINVAL;
498 	}
499 	return 0;
500 }
501 
502 static int hpet_assign_irq(struct hpet_dev *dev)
503 {
504 	unsigned int irq;
505 
506 	irq = create_irq();
507 	if (!irq)
508 		return -EINVAL;
509 
510 	set_irq_data(irq, dev);
511 
512 	if (hpet_setup_msi_irq(irq))
513 		return -EINVAL;
514 
515 	dev->irq = irq;
516 	return 0;
517 }
518 
519 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
520 {
521 	struct hpet_dev *dev = (struct hpet_dev *)data;
522 	struct clock_event_device *hevt = &dev->evt;
523 
524 	if (!hevt->event_handler) {
525 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
526 				dev->num);
527 		return IRQ_HANDLED;
528 	}
529 
530 	hevt->event_handler(hevt);
531 	return IRQ_HANDLED;
532 }
533 
534 static int hpet_setup_irq(struct hpet_dev *dev)
535 {
536 
537 	if (request_irq(dev->irq, hpet_interrupt_handler,
538 			IRQF_TIMER | IRQF_DISABLED | IRQF_NOBALANCING,
539 			dev->name, dev))
540 		return -1;
541 
542 	disable_irq(dev->irq);
543 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
544 	enable_irq(dev->irq);
545 
546 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
547 			 dev->name, dev->irq);
548 
549 	return 0;
550 }
551 
552 /* This should be called in specific @cpu */
553 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
554 {
555 	struct clock_event_device *evt = &hdev->evt;
556 	uint64_t hpet_freq;
557 
558 	WARN_ON(cpu != smp_processor_id());
559 	if (!(hdev->flags & HPET_DEV_VALID))
560 		return;
561 
562 	if (hpet_setup_msi_irq(hdev->irq))
563 		return;
564 
565 	hdev->cpu = cpu;
566 	per_cpu(cpu_hpet_dev, cpu) = hdev;
567 	evt->name = hdev->name;
568 	hpet_setup_irq(hdev);
569 	evt->irq = hdev->irq;
570 
571 	evt->rating = 110;
572 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
573 	if (hdev->flags & HPET_DEV_PERI_CAP)
574 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
575 
576 	evt->set_mode = hpet_msi_set_mode;
577 	evt->set_next_event = hpet_msi_next_event;
578 	evt->shift = 32;
579 
580 	/*
581 	 * The period is a femto seconds value. We need to calculate the
582 	 * scaled math multiplication factor for nanosecond to hpet tick
583 	 * conversion.
584 	 */
585 	hpet_freq = FSEC_PER_SEC;
586 	do_div(hpet_freq, hpet_period);
587 	evt->mult = div_sc((unsigned long) hpet_freq,
588 				      NSEC_PER_SEC, evt->shift);
589 	/* Calculate the max delta */
590 	evt->max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, evt);
591 	/* 5 usec minimum reprogramming delta. */
592 	evt->min_delta_ns = 5000;
593 
594 	evt->cpumask = cpumask_of(hdev->cpu);
595 	clockevents_register_device(evt);
596 }
597 
598 #ifdef CONFIG_HPET
599 /* Reserve at least one timer for userspace (/dev/hpet) */
600 #define RESERVE_TIMERS 1
601 #else
602 #define RESERVE_TIMERS 0
603 #endif
604 
605 static void hpet_msi_capability_lookup(unsigned int start_timer)
606 {
607 	unsigned int id;
608 	unsigned int num_timers;
609 	unsigned int num_timers_used = 0;
610 	int i;
611 
612 	if (hpet_msi_disable)
613 		return;
614 
615 	if (boot_cpu_has(X86_FEATURE_ARAT))
616 		return;
617 	id = hpet_readl(HPET_ID);
618 
619 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
620 	num_timers++; /* Value read out starts from 0 */
621 	hpet_print_config();
622 
623 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
624 	if (!hpet_devs)
625 		return;
626 
627 	hpet_num_timers = num_timers;
628 
629 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
630 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
631 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
632 
633 		/* Only consider HPET timer with MSI support */
634 		if (!(cfg & HPET_TN_FSB_CAP))
635 			continue;
636 
637 		hdev->flags = 0;
638 		if (cfg & HPET_TN_PERIODIC_CAP)
639 			hdev->flags |= HPET_DEV_PERI_CAP;
640 		hdev->num = i;
641 
642 		sprintf(hdev->name, "hpet%d", i);
643 		if (hpet_assign_irq(hdev))
644 			continue;
645 
646 		hdev->flags |= HPET_DEV_FSB_CAP;
647 		hdev->flags |= HPET_DEV_VALID;
648 		num_timers_used++;
649 		if (num_timers_used == num_possible_cpus())
650 			break;
651 	}
652 
653 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
654 		num_timers, num_timers_used);
655 }
656 
657 #ifdef CONFIG_HPET
658 static void hpet_reserve_msi_timers(struct hpet_data *hd)
659 {
660 	int i;
661 
662 	if (!hpet_devs)
663 		return;
664 
665 	for (i = 0; i < hpet_num_timers; i++) {
666 		struct hpet_dev *hdev = &hpet_devs[i];
667 
668 		if (!(hdev->flags & HPET_DEV_VALID))
669 			continue;
670 
671 		hd->hd_irq[hdev->num] = hdev->irq;
672 		hpet_reserve_timer(hd, hdev->num);
673 	}
674 }
675 #endif
676 
677 static struct hpet_dev *hpet_get_unused_timer(void)
678 {
679 	int i;
680 
681 	if (!hpet_devs)
682 		return NULL;
683 
684 	for (i = 0; i < hpet_num_timers; i++) {
685 		struct hpet_dev *hdev = &hpet_devs[i];
686 
687 		if (!(hdev->flags & HPET_DEV_VALID))
688 			continue;
689 		if (test_and_set_bit(HPET_DEV_USED_BIT,
690 			(unsigned long *)&hdev->flags))
691 			continue;
692 		return hdev;
693 	}
694 	return NULL;
695 }
696 
697 struct hpet_work_struct {
698 	struct delayed_work work;
699 	struct completion complete;
700 };
701 
702 static void hpet_work(struct work_struct *w)
703 {
704 	struct hpet_dev *hdev;
705 	int cpu = smp_processor_id();
706 	struct hpet_work_struct *hpet_work;
707 
708 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
709 
710 	hdev = hpet_get_unused_timer();
711 	if (hdev)
712 		init_one_hpet_msi_clockevent(hdev, cpu);
713 
714 	complete(&hpet_work->complete);
715 }
716 
717 static int hpet_cpuhp_notify(struct notifier_block *n,
718 		unsigned long action, void *hcpu)
719 {
720 	unsigned long cpu = (unsigned long)hcpu;
721 	struct hpet_work_struct work;
722 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
723 
724 	switch (action & 0xf) {
725 	case CPU_ONLINE:
726 		INIT_DELAYED_WORK_ON_STACK(&work.work, hpet_work);
727 		init_completion(&work.complete);
728 		/* FIXME: add schedule_work_on() */
729 		schedule_delayed_work_on(cpu, &work.work, 0);
730 		wait_for_completion(&work.complete);
731 		destroy_timer_on_stack(&work.work.timer);
732 		break;
733 	case CPU_DEAD:
734 		if (hdev) {
735 			free_irq(hdev->irq, hdev);
736 			hdev->flags &= ~HPET_DEV_USED;
737 			per_cpu(cpu_hpet_dev, cpu) = NULL;
738 		}
739 		break;
740 	}
741 	return NOTIFY_OK;
742 }
743 #else
744 
745 static int hpet_setup_msi_irq(unsigned int irq)
746 {
747 	return 0;
748 }
749 static void hpet_msi_capability_lookup(unsigned int start_timer)
750 {
751 	return;
752 }
753 
754 #ifdef CONFIG_HPET
755 static void hpet_reserve_msi_timers(struct hpet_data *hd)
756 {
757 	return;
758 }
759 #endif
760 
761 static int hpet_cpuhp_notify(struct notifier_block *n,
762 		unsigned long action, void *hcpu)
763 {
764 	return NOTIFY_OK;
765 }
766 
767 #endif
768 
769 /*
770  * Clock source related code
771  */
772 static cycle_t read_hpet(struct clocksource *cs)
773 {
774 	return (cycle_t)hpet_readl(HPET_COUNTER);
775 }
776 
777 #ifdef CONFIG_X86_64
778 static cycle_t __vsyscall_fn vread_hpet(void)
779 {
780 	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + 0xf0);
781 }
782 #endif
783 
784 static struct clocksource clocksource_hpet = {
785 	.name		= "hpet",
786 	.rating		= 250,
787 	.read		= read_hpet,
788 	.mask		= HPET_MASK,
789 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
790 	.resume		= hpet_resume_counter,
791 #ifdef CONFIG_X86_64
792 	.vread		= vread_hpet,
793 #endif
794 };
795 
796 static int hpet_clocksource_register(void)
797 {
798 	u64 start, now;
799 	u64 hpet_freq;
800 	cycle_t t1;
801 
802 	/* Start the counter */
803 	hpet_restart_counter();
804 
805 	/* Verify whether hpet counter works */
806 	t1 = hpet_readl(HPET_COUNTER);
807 	rdtscll(start);
808 
809 	/*
810 	 * We don't know the TSC frequency yet, but waiting for
811 	 * 200000 TSC cycles is safe:
812 	 * 4 GHz == 50us
813 	 * 1 GHz == 200us
814 	 */
815 	do {
816 		rep_nop();
817 		rdtscll(now);
818 	} while ((now - start) < 200000UL);
819 
820 	if (t1 == hpet_readl(HPET_COUNTER)) {
821 		printk(KERN_WARNING
822 		       "HPET counter not counting. HPET disabled\n");
823 		return -ENODEV;
824 	}
825 
826 	/*
827 	 * The definition of mult is (include/linux/clocksource.h)
828 	 * mult/2^shift = ns/cyc and hpet_period is in units of fsec/cyc
829 	 * so we first need to convert hpet_period to ns/cyc units:
830 	 *  mult/2^shift = ns/cyc = hpet_period/10^6
831 	 *  mult = (hpet_period * 2^shift)/10^6
832 	 *  mult = (hpet_period << shift)/FSEC_PER_NSEC
833 	 */
834 
835 	/* Need to convert hpet_period (fsec/cyc) to cyc/sec:
836 	 *
837 	 * cyc/sec = FSEC_PER_SEC/hpet_period(fsec/cyc)
838 	 * cyc/sec = (FSEC_PER_NSEC * NSEC_PER_SEC)/hpet_period
839 	 */
840 	hpet_freq = FSEC_PER_SEC;
841 	do_div(hpet_freq, hpet_period);
842 	clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
843 
844 	return 0;
845 }
846 
847 /**
848  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
849  */
850 int __init hpet_enable(void)
851 {
852 	unsigned int id;
853 	int i;
854 
855 	if (!is_hpet_capable())
856 		return 0;
857 
858 	hpet_set_mapping();
859 
860 	/*
861 	 * Read the period and check for a sane value:
862 	 */
863 	hpet_period = hpet_readl(HPET_PERIOD);
864 
865 	/*
866 	 * AMD SB700 based systems with spread spectrum enabled use a
867 	 * SMM based HPET emulation to provide proper frequency
868 	 * setting. The SMM code is initialized with the first HPET
869 	 * register access and takes some time to complete. During
870 	 * this time the config register reads 0xffffffff. We check
871 	 * for max. 1000 loops whether the config register reads a non
872 	 * 0xffffffff value to make sure that HPET is up and running
873 	 * before we go further. A counting loop is safe, as the HPET
874 	 * access takes thousands of CPU cycles. On non SB700 based
875 	 * machines this check is only done once and has no side
876 	 * effects.
877 	 */
878 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
879 		if (i == 1000) {
880 			printk(KERN_WARNING
881 			       "HPET config register value = 0xFFFFFFFF. "
882 			       "Disabling HPET\n");
883 			goto out_nohpet;
884 		}
885 	}
886 
887 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
888 		goto out_nohpet;
889 
890 	/*
891 	 * Read the HPET ID register to retrieve the IRQ routing
892 	 * information and the number of channels
893 	 */
894 	id = hpet_readl(HPET_ID);
895 	hpet_print_config();
896 
897 #ifdef CONFIG_HPET_EMULATE_RTC
898 	/*
899 	 * The legacy routing mode needs at least two channels, tick timer
900 	 * and the rtc emulation channel.
901 	 */
902 	if (!(id & HPET_ID_NUMBER))
903 		goto out_nohpet;
904 #endif
905 
906 	if (hpet_clocksource_register())
907 		goto out_nohpet;
908 
909 	if (id & HPET_ID_LEGSUP) {
910 		hpet_legacy_clockevent_register();
911 		return 1;
912 	}
913 	return 0;
914 
915 out_nohpet:
916 	hpet_clear_mapping();
917 	hpet_address = 0;
918 	return 0;
919 }
920 
921 /*
922  * Needs to be late, as the reserve_timer code calls kalloc !
923  *
924  * Not a problem on i386 as hpet_enable is called from late_time_init,
925  * but on x86_64 it is necessary !
926  */
927 static __init int hpet_late_init(void)
928 {
929 	int cpu;
930 
931 	if (boot_hpet_disable)
932 		return -ENODEV;
933 
934 	if (!hpet_address) {
935 		if (!force_hpet_address)
936 			return -ENODEV;
937 
938 		hpet_address = force_hpet_address;
939 		hpet_enable();
940 	}
941 
942 	if (!hpet_virt_address)
943 		return -ENODEV;
944 
945 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
946 		hpet_msi_capability_lookup(2);
947 	else
948 		hpet_msi_capability_lookup(0);
949 
950 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
951 	hpet_print_config();
952 
953 	if (hpet_msi_disable)
954 		return 0;
955 
956 	if (boot_cpu_has(X86_FEATURE_ARAT))
957 		return 0;
958 
959 	for_each_online_cpu(cpu) {
960 		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
961 	}
962 
963 	/* This notifier should be called after workqueue is ready */
964 	hotcpu_notifier(hpet_cpuhp_notify, -20);
965 
966 	return 0;
967 }
968 fs_initcall(hpet_late_init);
969 
970 void hpet_disable(void)
971 {
972 	if (is_hpet_capable() && hpet_virt_address) {
973 		unsigned int cfg = hpet_readl(HPET_CFG);
974 
975 		if (hpet_legacy_int_enabled) {
976 			cfg &= ~HPET_CFG_LEGACY;
977 			hpet_legacy_int_enabled = 0;
978 		}
979 		cfg &= ~HPET_CFG_ENABLE;
980 		hpet_writel(cfg, HPET_CFG);
981 	}
982 }
983 
984 #ifdef CONFIG_HPET_EMULATE_RTC
985 
986 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
987  * is enabled, we support RTC interrupt functionality in software.
988  * RTC has 3 kinds of interrupts:
989  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
990  *    is updated
991  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
992  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
993  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
994  * (1) and (2) above are implemented using polling at a frequency of
995  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
996  * overhead. (DEFAULT_RTC_INT_FREQ)
997  * For (3), we use interrupts at 64Hz or user specified periodic
998  * frequency, whichever is higher.
999  */
1000 #include <linux/mc146818rtc.h>
1001 #include <linux/rtc.h>
1002 #include <asm/rtc.h>
1003 
1004 #define DEFAULT_RTC_INT_FREQ	64
1005 #define DEFAULT_RTC_SHIFT	6
1006 #define RTC_NUM_INTS		1
1007 
1008 static unsigned long hpet_rtc_flags;
1009 static int hpet_prev_update_sec;
1010 static struct rtc_time hpet_alarm_time;
1011 static unsigned long hpet_pie_count;
1012 static u32 hpet_t1_cmp;
1013 static u32 hpet_default_delta;
1014 static u32 hpet_pie_delta;
1015 static unsigned long hpet_pie_limit;
1016 
1017 static rtc_irq_handler irq_handler;
1018 
1019 /*
1020  * Check that the hpet counter c1 is ahead of the c2
1021  */
1022 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1023 {
1024 	return (s32)(c2 - c1) < 0;
1025 }
1026 
1027 /*
1028  * Registers a IRQ handler.
1029  */
1030 int hpet_register_irq_handler(rtc_irq_handler handler)
1031 {
1032 	if (!is_hpet_enabled())
1033 		return -ENODEV;
1034 	if (irq_handler)
1035 		return -EBUSY;
1036 
1037 	irq_handler = handler;
1038 
1039 	return 0;
1040 }
1041 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1042 
1043 /*
1044  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1045  * and does cleanup.
1046  */
1047 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1048 {
1049 	if (!is_hpet_enabled())
1050 		return;
1051 
1052 	irq_handler = NULL;
1053 	hpet_rtc_flags = 0;
1054 }
1055 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1056 
1057 /*
1058  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1059  * is not supported by all HPET implementations for timer 1.
1060  *
1061  * hpet_rtc_timer_init() is called when the rtc is initialized.
1062  */
1063 int hpet_rtc_timer_init(void)
1064 {
1065 	unsigned int cfg, cnt, delta;
1066 	unsigned long flags;
1067 
1068 	if (!is_hpet_enabled())
1069 		return 0;
1070 
1071 	if (!hpet_default_delta) {
1072 		uint64_t clc;
1073 
1074 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1075 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1076 		hpet_default_delta = clc;
1077 	}
1078 
1079 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1080 		delta = hpet_default_delta;
1081 	else
1082 		delta = hpet_pie_delta;
1083 
1084 	local_irq_save(flags);
1085 
1086 	cnt = delta + hpet_readl(HPET_COUNTER);
1087 	hpet_writel(cnt, HPET_T1_CMP);
1088 	hpet_t1_cmp = cnt;
1089 
1090 	cfg = hpet_readl(HPET_T1_CFG);
1091 	cfg &= ~HPET_TN_PERIODIC;
1092 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1093 	hpet_writel(cfg, HPET_T1_CFG);
1094 
1095 	local_irq_restore(flags);
1096 
1097 	return 1;
1098 }
1099 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1100 
1101 /*
1102  * The functions below are called from rtc driver.
1103  * Return 0 if HPET is not being used.
1104  * Otherwise do the necessary changes and return 1.
1105  */
1106 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1107 {
1108 	if (!is_hpet_enabled())
1109 		return 0;
1110 
1111 	hpet_rtc_flags &= ~bit_mask;
1112 	return 1;
1113 }
1114 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1115 
1116 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1117 {
1118 	unsigned long oldbits = hpet_rtc_flags;
1119 
1120 	if (!is_hpet_enabled())
1121 		return 0;
1122 
1123 	hpet_rtc_flags |= bit_mask;
1124 
1125 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1126 		hpet_prev_update_sec = -1;
1127 
1128 	if (!oldbits)
1129 		hpet_rtc_timer_init();
1130 
1131 	return 1;
1132 }
1133 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1134 
1135 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1136 			unsigned char sec)
1137 {
1138 	if (!is_hpet_enabled())
1139 		return 0;
1140 
1141 	hpet_alarm_time.tm_hour = hrs;
1142 	hpet_alarm_time.tm_min = min;
1143 	hpet_alarm_time.tm_sec = sec;
1144 
1145 	return 1;
1146 }
1147 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1148 
1149 int hpet_set_periodic_freq(unsigned long freq)
1150 {
1151 	uint64_t clc;
1152 
1153 	if (!is_hpet_enabled())
1154 		return 0;
1155 
1156 	if (freq <= DEFAULT_RTC_INT_FREQ)
1157 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1158 	else {
1159 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1160 		do_div(clc, freq);
1161 		clc >>= hpet_clockevent.shift;
1162 		hpet_pie_delta = clc;
1163 		hpet_pie_limit = 0;
1164 	}
1165 	return 1;
1166 }
1167 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1168 
1169 int hpet_rtc_dropped_irq(void)
1170 {
1171 	return is_hpet_enabled();
1172 }
1173 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1174 
1175 static void hpet_rtc_timer_reinit(void)
1176 {
1177 	unsigned int cfg, delta;
1178 	int lost_ints = -1;
1179 
1180 	if (unlikely(!hpet_rtc_flags)) {
1181 		cfg = hpet_readl(HPET_T1_CFG);
1182 		cfg &= ~HPET_TN_ENABLE;
1183 		hpet_writel(cfg, HPET_T1_CFG);
1184 		return;
1185 	}
1186 
1187 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1188 		delta = hpet_default_delta;
1189 	else
1190 		delta = hpet_pie_delta;
1191 
1192 	/*
1193 	 * Increment the comparator value until we are ahead of the
1194 	 * current count.
1195 	 */
1196 	do {
1197 		hpet_t1_cmp += delta;
1198 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1199 		lost_ints++;
1200 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1201 
1202 	if (lost_ints) {
1203 		if (hpet_rtc_flags & RTC_PIE)
1204 			hpet_pie_count += lost_ints;
1205 		if (printk_ratelimit())
1206 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1207 				lost_ints);
1208 	}
1209 }
1210 
1211 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1212 {
1213 	struct rtc_time curr_time;
1214 	unsigned long rtc_int_flag = 0;
1215 
1216 	hpet_rtc_timer_reinit();
1217 	memset(&curr_time, 0, sizeof(struct rtc_time));
1218 
1219 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1220 		get_rtc_time(&curr_time);
1221 
1222 	if (hpet_rtc_flags & RTC_UIE &&
1223 	    curr_time.tm_sec != hpet_prev_update_sec) {
1224 		if (hpet_prev_update_sec >= 0)
1225 			rtc_int_flag = RTC_UF;
1226 		hpet_prev_update_sec = curr_time.tm_sec;
1227 	}
1228 
1229 	if (hpet_rtc_flags & RTC_PIE &&
1230 	    ++hpet_pie_count >= hpet_pie_limit) {
1231 		rtc_int_flag |= RTC_PF;
1232 		hpet_pie_count = 0;
1233 	}
1234 
1235 	if (hpet_rtc_flags & RTC_AIE &&
1236 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1237 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1238 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1239 			rtc_int_flag |= RTC_AF;
1240 
1241 	if (rtc_int_flag) {
1242 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1243 		if (irq_handler)
1244 			irq_handler(rtc_int_flag, dev_id);
1245 	}
1246 	return IRQ_HANDLED;
1247 }
1248 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1249 #endif
1250