xref: /linux/arch/x86/kernel/hpet.c (revision a13d7201d7deedcbb6ac6efa94a1a7d34d3d79ec)
1 #include <linux/clocksource.h>
2 #include <linux/clockchips.h>
3 #include <linux/interrupt.h>
4 #include <linux/export.h>
5 #include <linux/delay.h>
6 #include <linux/errno.h>
7 #include <linux/i8253.h>
8 #include <linux/slab.h>
9 #include <linux/hpet.h>
10 #include <linux/init.h>
11 #include <linux/cpu.h>
12 #include <linux/pm.h>
13 #include <linux/io.h>
14 
15 #include <asm/irqdomain.h>
16 #include <asm/fixmap.h>
17 #include <asm/hpet.h>
18 #include <asm/time.h>
19 
20 #define HPET_MASK			CLOCKSOURCE_MASK(32)
21 
22 /* FSEC = 10^-15
23    NSEC = 10^-9 */
24 #define FSEC_PER_NSEC			1000000L
25 
26 #define HPET_DEV_USED_BIT		2
27 #define HPET_DEV_USED			(1 << HPET_DEV_USED_BIT)
28 #define HPET_DEV_VALID			0x8
29 #define HPET_DEV_FSB_CAP		0x1000
30 #define HPET_DEV_PERI_CAP		0x2000
31 
32 #define HPET_MIN_CYCLES			128
33 #define HPET_MIN_PROG_DELTA		(HPET_MIN_CYCLES + (HPET_MIN_CYCLES >> 1))
34 
35 /*
36  * HPET address is set in acpi/boot.c, when an ACPI entry exists
37  */
38 unsigned long				hpet_address;
39 u8					hpet_blockid; /* OS timer block num */
40 u8					hpet_msi_disable;
41 
42 #ifdef CONFIG_PCI_MSI
43 static unsigned long			hpet_num_timers;
44 #endif
45 static void __iomem			*hpet_virt_address;
46 
47 struct hpet_dev {
48 	struct clock_event_device	evt;
49 	unsigned int			num;
50 	int				cpu;
51 	unsigned int			irq;
52 	unsigned int			flags;
53 	char				name[10];
54 };
55 
56 inline struct hpet_dev *EVT_TO_HPET_DEV(struct clock_event_device *evtdev)
57 {
58 	return container_of(evtdev, struct hpet_dev, evt);
59 }
60 
61 inline unsigned int hpet_readl(unsigned int a)
62 {
63 	return readl(hpet_virt_address + a);
64 }
65 
66 static inline void hpet_writel(unsigned int d, unsigned int a)
67 {
68 	writel(d, hpet_virt_address + a);
69 }
70 
71 #ifdef CONFIG_X86_64
72 #include <asm/pgtable.h>
73 #endif
74 
75 static inline void hpet_set_mapping(void)
76 {
77 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
78 }
79 
80 static inline void hpet_clear_mapping(void)
81 {
82 	iounmap(hpet_virt_address);
83 	hpet_virt_address = NULL;
84 }
85 
86 /*
87  * HPET command line enable / disable
88  */
89 int boot_hpet_disable;
90 int hpet_force_user;
91 static int hpet_verbose;
92 
93 static int __init hpet_setup(char *str)
94 {
95 	while (str) {
96 		char *next = strchr(str, ',');
97 
98 		if (next)
99 			*next++ = 0;
100 		if (!strncmp("disable", str, 7))
101 			boot_hpet_disable = 1;
102 		if (!strncmp("force", str, 5))
103 			hpet_force_user = 1;
104 		if (!strncmp("verbose", str, 7))
105 			hpet_verbose = 1;
106 		str = next;
107 	}
108 	return 1;
109 }
110 __setup("hpet=", hpet_setup);
111 
112 static int __init disable_hpet(char *str)
113 {
114 	boot_hpet_disable = 1;
115 	return 1;
116 }
117 __setup("nohpet", disable_hpet);
118 
119 static inline int is_hpet_capable(void)
120 {
121 	return !boot_hpet_disable && hpet_address;
122 }
123 
124 /*
125  * HPET timer interrupt enable / disable
126  */
127 static int hpet_legacy_int_enabled;
128 
129 /**
130  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
131  */
132 int is_hpet_enabled(void)
133 {
134 	return is_hpet_capable() && hpet_legacy_int_enabled;
135 }
136 EXPORT_SYMBOL_GPL(is_hpet_enabled);
137 
138 static void _hpet_print_config(const char *function, int line)
139 {
140 	u32 i, timers, l, h;
141 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
142 	l = hpet_readl(HPET_ID);
143 	h = hpet_readl(HPET_PERIOD);
144 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
145 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
146 	l = hpet_readl(HPET_CFG);
147 	h = hpet_readl(HPET_STATUS);
148 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
149 	l = hpet_readl(HPET_COUNTER);
150 	h = hpet_readl(HPET_COUNTER+4);
151 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
152 
153 	for (i = 0; i < timers; i++) {
154 		l = hpet_readl(HPET_Tn_CFG(i));
155 		h = hpet_readl(HPET_Tn_CFG(i)+4);
156 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
157 		       i, l, h);
158 		l = hpet_readl(HPET_Tn_CMP(i));
159 		h = hpet_readl(HPET_Tn_CMP(i)+4);
160 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
161 		       i, l, h);
162 		l = hpet_readl(HPET_Tn_ROUTE(i));
163 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
164 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
165 		       i, l, h);
166 	}
167 }
168 
169 #define hpet_print_config()					\
170 do {								\
171 	if (hpet_verbose)					\
172 		_hpet_print_config(__func__, __LINE__);	\
173 } while (0)
174 
175 /*
176  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
177  * timer 0 and timer 1 in case of RTC emulation.
178  */
179 #ifdef CONFIG_HPET
180 
181 static void hpet_reserve_msi_timers(struct hpet_data *hd);
182 
183 static void hpet_reserve_platform_timers(unsigned int id)
184 {
185 	struct hpet __iomem *hpet = hpet_virt_address;
186 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
187 	unsigned int nrtimers, i;
188 	struct hpet_data hd;
189 
190 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
191 
192 	memset(&hd, 0, sizeof(hd));
193 	hd.hd_phys_address	= hpet_address;
194 	hd.hd_address		= hpet;
195 	hd.hd_nirqs		= nrtimers;
196 	hpet_reserve_timer(&hd, 0);
197 
198 #ifdef CONFIG_HPET_EMULATE_RTC
199 	hpet_reserve_timer(&hd, 1);
200 #endif
201 
202 	/*
203 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
204 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
205 	 * don't bother configuring *any* comparator interrupts.
206 	 */
207 	hd.hd_irq[0] = HPET_LEGACY_8254;
208 	hd.hd_irq[1] = HPET_LEGACY_RTC;
209 
210 	for (i = 2; i < nrtimers; timer++, i++) {
211 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
212 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
213 	}
214 
215 	hpet_reserve_msi_timers(&hd);
216 
217 	hpet_alloc(&hd);
218 
219 }
220 #else
221 static void hpet_reserve_platform_timers(unsigned int id) { }
222 #endif
223 
224 /*
225  * Common hpet info
226  */
227 static unsigned long hpet_freq;
228 
229 static void hpet_legacy_set_mode(enum clock_event_mode mode,
230 			  struct clock_event_device *evt);
231 static int hpet_legacy_next_event(unsigned long delta,
232 			   struct clock_event_device *evt);
233 
234 /*
235  * The hpet clock event device
236  */
237 static struct clock_event_device hpet_clockevent = {
238 	.name		= "hpet",
239 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
240 	.set_mode	= hpet_legacy_set_mode,
241 	.set_next_event = hpet_legacy_next_event,
242 	.irq		= 0,
243 	.rating		= 50,
244 };
245 
246 static void hpet_stop_counter(void)
247 {
248 	unsigned long cfg = hpet_readl(HPET_CFG);
249 	cfg &= ~HPET_CFG_ENABLE;
250 	hpet_writel(cfg, HPET_CFG);
251 }
252 
253 static void hpet_reset_counter(void)
254 {
255 	hpet_writel(0, HPET_COUNTER);
256 	hpet_writel(0, HPET_COUNTER + 4);
257 }
258 
259 static void hpet_start_counter(void)
260 {
261 	unsigned int cfg = hpet_readl(HPET_CFG);
262 	cfg |= HPET_CFG_ENABLE;
263 	hpet_writel(cfg, HPET_CFG);
264 }
265 
266 static void hpet_restart_counter(void)
267 {
268 	hpet_stop_counter();
269 	hpet_reset_counter();
270 	hpet_start_counter();
271 }
272 
273 static void hpet_resume_device(void)
274 {
275 	force_hpet_resume();
276 }
277 
278 static void hpet_resume_counter(struct clocksource *cs)
279 {
280 	hpet_resume_device();
281 	hpet_restart_counter();
282 }
283 
284 static void hpet_enable_legacy_int(void)
285 {
286 	unsigned int cfg = hpet_readl(HPET_CFG);
287 
288 	cfg |= HPET_CFG_LEGACY;
289 	hpet_writel(cfg, HPET_CFG);
290 	hpet_legacy_int_enabled = 1;
291 }
292 
293 static void hpet_legacy_clockevent_register(void)
294 {
295 	/* Start HPET legacy interrupts */
296 	hpet_enable_legacy_int();
297 
298 	/*
299 	 * Start hpet with the boot cpu mask and make it
300 	 * global after the IO_APIC has been initialized.
301 	 */
302 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
303 	clockevents_config_and_register(&hpet_clockevent, hpet_freq,
304 					HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
305 	global_clock_event = &hpet_clockevent;
306 	printk(KERN_DEBUG "hpet clockevent registered\n");
307 }
308 
309 static void hpet_set_mode(enum clock_event_mode mode,
310 			  struct clock_event_device *evt, int timer)
311 {
312 	unsigned int cfg, cmp, now;
313 	uint64_t delta;
314 
315 	switch (mode) {
316 	case CLOCK_EVT_MODE_PERIODIC:
317 		hpet_stop_counter();
318 		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
319 		delta >>= evt->shift;
320 		now = hpet_readl(HPET_COUNTER);
321 		cmp = now + (unsigned int) delta;
322 		cfg = hpet_readl(HPET_Tn_CFG(timer));
323 		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
324 		       HPET_TN_SETVAL | HPET_TN_32BIT;
325 		hpet_writel(cfg, HPET_Tn_CFG(timer));
326 		hpet_writel(cmp, HPET_Tn_CMP(timer));
327 		udelay(1);
328 		/*
329 		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
330 		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
331 		 * bit is automatically cleared after the first write.
332 		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
333 		 * Publication # 24674)
334 		 */
335 		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
336 		hpet_start_counter();
337 		hpet_print_config();
338 		break;
339 
340 	case CLOCK_EVT_MODE_ONESHOT:
341 		cfg = hpet_readl(HPET_Tn_CFG(timer));
342 		cfg &= ~HPET_TN_PERIODIC;
343 		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
344 		hpet_writel(cfg, HPET_Tn_CFG(timer));
345 		break;
346 
347 	case CLOCK_EVT_MODE_UNUSED:
348 	case CLOCK_EVT_MODE_SHUTDOWN:
349 		cfg = hpet_readl(HPET_Tn_CFG(timer));
350 		cfg &= ~HPET_TN_ENABLE;
351 		hpet_writel(cfg, HPET_Tn_CFG(timer));
352 		break;
353 
354 	case CLOCK_EVT_MODE_RESUME:
355 		if (timer == 0) {
356 			hpet_enable_legacy_int();
357 		} else {
358 			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
359 			irq_domain_activate_irq(irq_get_irq_data(hdev->irq));
360 			disable_irq(hdev->irq);
361 			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
362 			enable_irq(hdev->irq);
363 		}
364 		hpet_print_config();
365 		break;
366 	}
367 }
368 
369 static int hpet_next_event(unsigned long delta,
370 			   struct clock_event_device *evt, int timer)
371 {
372 	u32 cnt;
373 	s32 res;
374 
375 	cnt = hpet_readl(HPET_COUNTER);
376 	cnt += (u32) delta;
377 	hpet_writel(cnt, HPET_Tn_CMP(timer));
378 
379 	/*
380 	 * HPETs are a complete disaster. The compare register is
381 	 * based on a equal comparison and neither provides a less
382 	 * than or equal functionality (which would require to take
383 	 * the wraparound into account) nor a simple count down event
384 	 * mode. Further the write to the comparator register is
385 	 * delayed internally up to two HPET clock cycles in certain
386 	 * chipsets (ATI, ICH9,10). Some newer AMD chipsets have even
387 	 * longer delays. We worked around that by reading back the
388 	 * compare register, but that required another workaround for
389 	 * ICH9,10 chips where the first readout after write can
390 	 * return the old stale value. We already had a minimum
391 	 * programming delta of 5us enforced, but a NMI or SMI hitting
392 	 * between the counter readout and the comparator write can
393 	 * move us behind that point easily. Now instead of reading
394 	 * the compare register back several times, we make the ETIME
395 	 * decision based on the following: Return ETIME if the
396 	 * counter value after the write is less than HPET_MIN_CYCLES
397 	 * away from the event or if the counter is already ahead of
398 	 * the event. The minimum programming delta for the generic
399 	 * clockevents code is set to 1.5 * HPET_MIN_CYCLES.
400 	 */
401 	res = (s32)(cnt - hpet_readl(HPET_COUNTER));
402 
403 	return res < HPET_MIN_CYCLES ? -ETIME : 0;
404 }
405 
406 static void hpet_legacy_set_mode(enum clock_event_mode mode,
407 			struct clock_event_device *evt)
408 {
409 	hpet_set_mode(mode, evt, 0);
410 }
411 
412 static int hpet_legacy_next_event(unsigned long delta,
413 			struct clock_event_device *evt)
414 {
415 	return hpet_next_event(delta, evt, 0);
416 }
417 
418 /*
419  * HPET MSI Support
420  */
421 #ifdef CONFIG_PCI_MSI
422 
423 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
424 static struct hpet_dev	*hpet_devs;
425 static struct irq_domain *hpet_domain;
426 
427 void hpet_msi_unmask(struct irq_data *data)
428 {
429 	struct hpet_dev *hdev = data->handler_data;
430 	unsigned int cfg;
431 
432 	/* unmask it */
433 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
434 	cfg |= HPET_TN_ENABLE | HPET_TN_FSB;
435 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
436 }
437 
438 void hpet_msi_mask(struct irq_data *data)
439 {
440 	struct hpet_dev *hdev = data->handler_data;
441 	unsigned int cfg;
442 
443 	/* mask it */
444 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
445 	cfg &= ~(HPET_TN_ENABLE | HPET_TN_FSB);
446 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
447 }
448 
449 void hpet_msi_write(struct hpet_dev *hdev, struct msi_msg *msg)
450 {
451 	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
452 	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
453 }
454 
455 void hpet_msi_read(struct hpet_dev *hdev, struct msi_msg *msg)
456 {
457 	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
458 	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
459 	msg->address_hi = 0;
460 }
461 
462 static void hpet_msi_set_mode(enum clock_event_mode mode,
463 				struct clock_event_device *evt)
464 {
465 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
466 	hpet_set_mode(mode, evt, hdev->num);
467 }
468 
469 static int hpet_msi_next_event(unsigned long delta,
470 				struct clock_event_device *evt)
471 {
472 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
473 	return hpet_next_event(delta, evt, hdev->num);
474 }
475 
476 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
477 {
478 	struct hpet_dev *dev = (struct hpet_dev *)data;
479 	struct clock_event_device *hevt = &dev->evt;
480 
481 	if (!hevt->event_handler) {
482 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
483 				dev->num);
484 		return IRQ_HANDLED;
485 	}
486 
487 	hevt->event_handler(hevt);
488 	return IRQ_HANDLED;
489 }
490 
491 static int hpet_setup_irq(struct hpet_dev *dev)
492 {
493 
494 	if (request_irq(dev->irq, hpet_interrupt_handler,
495 			IRQF_TIMER | IRQF_NOBALANCING,
496 			dev->name, dev))
497 		return -1;
498 
499 	disable_irq(dev->irq);
500 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
501 	enable_irq(dev->irq);
502 
503 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
504 			 dev->name, dev->irq);
505 
506 	return 0;
507 }
508 
509 /* This should be called in specific @cpu */
510 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
511 {
512 	struct clock_event_device *evt = &hdev->evt;
513 
514 	WARN_ON(cpu != smp_processor_id());
515 	if (!(hdev->flags & HPET_DEV_VALID))
516 		return;
517 
518 	hdev->cpu = cpu;
519 	per_cpu(cpu_hpet_dev, cpu) = hdev;
520 	evt->name = hdev->name;
521 	hpet_setup_irq(hdev);
522 	evt->irq = hdev->irq;
523 
524 	evt->rating = 110;
525 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
526 	if (hdev->flags & HPET_DEV_PERI_CAP)
527 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
528 
529 	evt->set_mode = hpet_msi_set_mode;
530 	evt->set_next_event = hpet_msi_next_event;
531 	evt->cpumask = cpumask_of(hdev->cpu);
532 
533 	clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
534 					0x7FFFFFFF);
535 }
536 
537 #ifdef CONFIG_HPET
538 /* Reserve at least one timer for userspace (/dev/hpet) */
539 #define RESERVE_TIMERS 1
540 #else
541 #define RESERVE_TIMERS 0
542 #endif
543 
544 static void hpet_msi_capability_lookup(unsigned int start_timer)
545 {
546 	unsigned int id;
547 	unsigned int num_timers;
548 	unsigned int num_timers_used = 0;
549 	int i, irq;
550 
551 	if (hpet_msi_disable)
552 		return;
553 
554 	if (boot_cpu_has(X86_FEATURE_ARAT))
555 		return;
556 	id = hpet_readl(HPET_ID);
557 
558 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
559 	num_timers++; /* Value read out starts from 0 */
560 	hpet_print_config();
561 
562 	hpet_domain = hpet_create_irq_domain(hpet_blockid);
563 	if (!hpet_domain)
564 		return;
565 
566 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
567 	if (!hpet_devs)
568 		return;
569 
570 	hpet_num_timers = num_timers;
571 
572 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
573 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
574 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
575 
576 		/* Only consider HPET timer with MSI support */
577 		if (!(cfg & HPET_TN_FSB_CAP))
578 			continue;
579 
580 		hdev->flags = 0;
581 		if (cfg & HPET_TN_PERIODIC_CAP)
582 			hdev->flags |= HPET_DEV_PERI_CAP;
583 		sprintf(hdev->name, "hpet%d", i);
584 		hdev->num = i;
585 
586 		irq = hpet_assign_irq(hpet_domain, hdev, hdev->num);
587 		if (irq <= 0)
588 			continue;
589 
590 		hdev->irq = irq;
591 		hdev->flags |= HPET_DEV_FSB_CAP;
592 		hdev->flags |= HPET_DEV_VALID;
593 		num_timers_used++;
594 		if (num_timers_used == num_possible_cpus())
595 			break;
596 	}
597 
598 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
599 		num_timers, num_timers_used);
600 }
601 
602 #ifdef CONFIG_HPET
603 static void hpet_reserve_msi_timers(struct hpet_data *hd)
604 {
605 	int i;
606 
607 	if (!hpet_devs)
608 		return;
609 
610 	for (i = 0; i < hpet_num_timers; i++) {
611 		struct hpet_dev *hdev = &hpet_devs[i];
612 
613 		if (!(hdev->flags & HPET_DEV_VALID))
614 			continue;
615 
616 		hd->hd_irq[hdev->num] = hdev->irq;
617 		hpet_reserve_timer(hd, hdev->num);
618 	}
619 }
620 #endif
621 
622 static struct hpet_dev *hpet_get_unused_timer(void)
623 {
624 	int i;
625 
626 	if (!hpet_devs)
627 		return NULL;
628 
629 	for (i = 0; i < hpet_num_timers; i++) {
630 		struct hpet_dev *hdev = &hpet_devs[i];
631 
632 		if (!(hdev->flags & HPET_DEV_VALID))
633 			continue;
634 		if (test_and_set_bit(HPET_DEV_USED_BIT,
635 			(unsigned long *)&hdev->flags))
636 			continue;
637 		return hdev;
638 	}
639 	return NULL;
640 }
641 
642 struct hpet_work_struct {
643 	struct delayed_work work;
644 	struct completion complete;
645 };
646 
647 static void hpet_work(struct work_struct *w)
648 {
649 	struct hpet_dev *hdev;
650 	int cpu = smp_processor_id();
651 	struct hpet_work_struct *hpet_work;
652 
653 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
654 
655 	hdev = hpet_get_unused_timer();
656 	if (hdev)
657 		init_one_hpet_msi_clockevent(hdev, cpu);
658 
659 	complete(&hpet_work->complete);
660 }
661 
662 static int hpet_cpuhp_notify(struct notifier_block *n,
663 		unsigned long action, void *hcpu)
664 {
665 	unsigned long cpu = (unsigned long)hcpu;
666 	struct hpet_work_struct work;
667 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
668 
669 	switch (action & 0xf) {
670 	case CPU_ONLINE:
671 		INIT_DELAYED_WORK_ONSTACK(&work.work, hpet_work);
672 		init_completion(&work.complete);
673 		/* FIXME: add schedule_work_on() */
674 		schedule_delayed_work_on(cpu, &work.work, 0);
675 		wait_for_completion(&work.complete);
676 		destroy_delayed_work_on_stack(&work.work);
677 		break;
678 	case CPU_DEAD:
679 		if (hdev) {
680 			free_irq(hdev->irq, hdev);
681 			hdev->flags &= ~HPET_DEV_USED;
682 			per_cpu(cpu_hpet_dev, cpu) = NULL;
683 		}
684 		break;
685 	}
686 	return NOTIFY_OK;
687 }
688 #else
689 
690 static void hpet_msi_capability_lookup(unsigned int start_timer)
691 {
692 	return;
693 }
694 
695 #ifdef CONFIG_HPET
696 static void hpet_reserve_msi_timers(struct hpet_data *hd)
697 {
698 	return;
699 }
700 #endif
701 
702 static int hpet_cpuhp_notify(struct notifier_block *n,
703 		unsigned long action, void *hcpu)
704 {
705 	return NOTIFY_OK;
706 }
707 
708 #endif
709 
710 /*
711  * Clock source related code
712  */
713 static cycle_t read_hpet(struct clocksource *cs)
714 {
715 	return (cycle_t)hpet_readl(HPET_COUNTER);
716 }
717 
718 static struct clocksource clocksource_hpet = {
719 	.name		= "hpet",
720 	.rating		= 250,
721 	.read		= read_hpet,
722 	.mask		= HPET_MASK,
723 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
724 	.resume		= hpet_resume_counter,
725 	.archdata	= { .vclock_mode = VCLOCK_HPET },
726 };
727 
728 static int hpet_clocksource_register(void)
729 {
730 	u64 start, now;
731 	cycle_t t1;
732 
733 	/* Start the counter */
734 	hpet_restart_counter();
735 
736 	/* Verify whether hpet counter works */
737 	t1 = hpet_readl(HPET_COUNTER);
738 	rdtscll(start);
739 
740 	/*
741 	 * We don't know the TSC frequency yet, but waiting for
742 	 * 200000 TSC cycles is safe:
743 	 * 4 GHz == 50us
744 	 * 1 GHz == 200us
745 	 */
746 	do {
747 		rep_nop();
748 		rdtscll(now);
749 	} while ((now - start) < 200000UL);
750 
751 	if (t1 == hpet_readl(HPET_COUNTER)) {
752 		printk(KERN_WARNING
753 		       "HPET counter not counting. HPET disabled\n");
754 		return -ENODEV;
755 	}
756 
757 	clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
758 	return 0;
759 }
760 
761 static u32 *hpet_boot_cfg;
762 
763 /**
764  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
765  */
766 int __init hpet_enable(void)
767 {
768 	u32 hpet_period, cfg, id;
769 	u64 freq;
770 	unsigned int i, last;
771 
772 	if (!is_hpet_capable())
773 		return 0;
774 
775 	hpet_set_mapping();
776 
777 	/*
778 	 * Read the period and check for a sane value:
779 	 */
780 	hpet_period = hpet_readl(HPET_PERIOD);
781 
782 	/*
783 	 * AMD SB700 based systems with spread spectrum enabled use a
784 	 * SMM based HPET emulation to provide proper frequency
785 	 * setting. The SMM code is initialized with the first HPET
786 	 * register access and takes some time to complete. During
787 	 * this time the config register reads 0xffffffff. We check
788 	 * for max. 1000 loops whether the config register reads a non
789 	 * 0xffffffff value to make sure that HPET is up and running
790 	 * before we go further. A counting loop is safe, as the HPET
791 	 * access takes thousands of CPU cycles. On non SB700 based
792 	 * machines this check is only done once and has no side
793 	 * effects.
794 	 */
795 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
796 		if (i == 1000) {
797 			printk(KERN_WARNING
798 			       "HPET config register value = 0xFFFFFFFF. "
799 			       "Disabling HPET\n");
800 			goto out_nohpet;
801 		}
802 	}
803 
804 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
805 		goto out_nohpet;
806 
807 	/*
808 	 * The period is a femto seconds value. Convert it to a
809 	 * frequency.
810 	 */
811 	freq = FSEC_PER_SEC;
812 	do_div(freq, hpet_period);
813 	hpet_freq = freq;
814 
815 	/*
816 	 * Read the HPET ID register to retrieve the IRQ routing
817 	 * information and the number of channels
818 	 */
819 	id = hpet_readl(HPET_ID);
820 	hpet_print_config();
821 
822 	last = (id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT;
823 
824 #ifdef CONFIG_HPET_EMULATE_RTC
825 	/*
826 	 * The legacy routing mode needs at least two channels, tick timer
827 	 * and the rtc emulation channel.
828 	 */
829 	if (!last)
830 		goto out_nohpet;
831 #endif
832 
833 	cfg = hpet_readl(HPET_CFG);
834 	hpet_boot_cfg = kmalloc((last + 2) * sizeof(*hpet_boot_cfg),
835 				GFP_KERNEL);
836 	if (hpet_boot_cfg)
837 		*hpet_boot_cfg = cfg;
838 	else
839 		pr_warn("HPET initial state will not be saved\n");
840 	cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
841 	hpet_writel(cfg, HPET_CFG);
842 	if (cfg)
843 		pr_warn("HPET: Unrecognized bits %#x set in global cfg\n",
844 			cfg);
845 
846 	for (i = 0; i <= last; ++i) {
847 		cfg = hpet_readl(HPET_Tn_CFG(i));
848 		if (hpet_boot_cfg)
849 			hpet_boot_cfg[i + 1] = cfg;
850 		cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
851 		hpet_writel(cfg, HPET_Tn_CFG(i));
852 		cfg &= ~(HPET_TN_PERIODIC | HPET_TN_PERIODIC_CAP
853 			 | HPET_TN_64BIT_CAP | HPET_TN_32BIT | HPET_TN_ROUTE
854 			 | HPET_TN_FSB | HPET_TN_FSB_CAP);
855 		if (cfg)
856 			pr_warn("HPET: Unrecognized bits %#x set in cfg#%u\n",
857 				cfg, i);
858 	}
859 	hpet_print_config();
860 
861 	if (hpet_clocksource_register())
862 		goto out_nohpet;
863 
864 	if (id & HPET_ID_LEGSUP) {
865 		hpet_legacy_clockevent_register();
866 		return 1;
867 	}
868 	return 0;
869 
870 out_nohpet:
871 	hpet_clear_mapping();
872 	hpet_address = 0;
873 	return 0;
874 }
875 
876 /*
877  * Needs to be late, as the reserve_timer code calls kalloc !
878  *
879  * Not a problem on i386 as hpet_enable is called from late_time_init,
880  * but on x86_64 it is necessary !
881  */
882 static __init int hpet_late_init(void)
883 {
884 	int cpu;
885 
886 	if (boot_hpet_disable)
887 		return -ENODEV;
888 
889 	if (!hpet_address) {
890 		if (!force_hpet_address)
891 			return -ENODEV;
892 
893 		hpet_address = force_hpet_address;
894 		hpet_enable();
895 	}
896 
897 	if (!hpet_virt_address)
898 		return -ENODEV;
899 
900 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
901 		hpet_msi_capability_lookup(2);
902 	else
903 		hpet_msi_capability_lookup(0);
904 
905 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
906 	hpet_print_config();
907 
908 	if (hpet_msi_disable)
909 		return 0;
910 
911 	if (boot_cpu_has(X86_FEATURE_ARAT))
912 		return 0;
913 
914 	cpu_notifier_register_begin();
915 	for_each_online_cpu(cpu) {
916 		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
917 	}
918 
919 	/* This notifier should be called after workqueue is ready */
920 	__hotcpu_notifier(hpet_cpuhp_notify, -20);
921 	cpu_notifier_register_done();
922 
923 	return 0;
924 }
925 fs_initcall(hpet_late_init);
926 
927 void hpet_disable(void)
928 {
929 	if (is_hpet_capable() && hpet_virt_address) {
930 		unsigned int cfg = hpet_readl(HPET_CFG), id, last;
931 
932 		if (hpet_boot_cfg)
933 			cfg = *hpet_boot_cfg;
934 		else if (hpet_legacy_int_enabled) {
935 			cfg &= ~HPET_CFG_LEGACY;
936 			hpet_legacy_int_enabled = 0;
937 		}
938 		cfg &= ~HPET_CFG_ENABLE;
939 		hpet_writel(cfg, HPET_CFG);
940 
941 		if (!hpet_boot_cfg)
942 			return;
943 
944 		id = hpet_readl(HPET_ID);
945 		last = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
946 
947 		for (id = 0; id <= last; ++id)
948 			hpet_writel(hpet_boot_cfg[id + 1], HPET_Tn_CFG(id));
949 
950 		if (*hpet_boot_cfg & HPET_CFG_ENABLE)
951 			hpet_writel(*hpet_boot_cfg, HPET_CFG);
952 	}
953 }
954 
955 #ifdef CONFIG_HPET_EMULATE_RTC
956 
957 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
958  * is enabled, we support RTC interrupt functionality in software.
959  * RTC has 3 kinds of interrupts:
960  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
961  *    is updated
962  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
963  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
964  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
965  * (1) and (2) above are implemented using polling at a frequency of
966  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
967  * overhead. (DEFAULT_RTC_INT_FREQ)
968  * For (3), we use interrupts at 64Hz or user specified periodic
969  * frequency, whichever is higher.
970  */
971 #include <linux/mc146818rtc.h>
972 #include <linux/rtc.h>
973 #include <asm/rtc.h>
974 
975 #define DEFAULT_RTC_INT_FREQ	64
976 #define DEFAULT_RTC_SHIFT	6
977 #define RTC_NUM_INTS		1
978 
979 static unsigned long hpet_rtc_flags;
980 static int hpet_prev_update_sec;
981 static struct rtc_time hpet_alarm_time;
982 static unsigned long hpet_pie_count;
983 static u32 hpet_t1_cmp;
984 static u32 hpet_default_delta;
985 static u32 hpet_pie_delta;
986 static unsigned long hpet_pie_limit;
987 
988 static rtc_irq_handler irq_handler;
989 
990 /*
991  * Check that the hpet counter c1 is ahead of the c2
992  */
993 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
994 {
995 	return (s32)(c2 - c1) < 0;
996 }
997 
998 /*
999  * Registers a IRQ handler.
1000  */
1001 int hpet_register_irq_handler(rtc_irq_handler handler)
1002 {
1003 	if (!is_hpet_enabled())
1004 		return -ENODEV;
1005 	if (irq_handler)
1006 		return -EBUSY;
1007 
1008 	irq_handler = handler;
1009 
1010 	return 0;
1011 }
1012 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1013 
1014 /*
1015  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1016  * and does cleanup.
1017  */
1018 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1019 {
1020 	if (!is_hpet_enabled())
1021 		return;
1022 
1023 	irq_handler = NULL;
1024 	hpet_rtc_flags = 0;
1025 }
1026 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1027 
1028 /*
1029  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1030  * is not supported by all HPET implementations for timer 1.
1031  *
1032  * hpet_rtc_timer_init() is called when the rtc is initialized.
1033  */
1034 int hpet_rtc_timer_init(void)
1035 {
1036 	unsigned int cfg, cnt, delta;
1037 	unsigned long flags;
1038 
1039 	if (!is_hpet_enabled())
1040 		return 0;
1041 
1042 	if (!hpet_default_delta) {
1043 		uint64_t clc;
1044 
1045 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1046 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1047 		hpet_default_delta = clc;
1048 	}
1049 
1050 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1051 		delta = hpet_default_delta;
1052 	else
1053 		delta = hpet_pie_delta;
1054 
1055 	local_irq_save(flags);
1056 
1057 	cnt = delta + hpet_readl(HPET_COUNTER);
1058 	hpet_writel(cnt, HPET_T1_CMP);
1059 	hpet_t1_cmp = cnt;
1060 
1061 	cfg = hpet_readl(HPET_T1_CFG);
1062 	cfg &= ~HPET_TN_PERIODIC;
1063 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1064 	hpet_writel(cfg, HPET_T1_CFG);
1065 
1066 	local_irq_restore(flags);
1067 
1068 	return 1;
1069 }
1070 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1071 
1072 static void hpet_disable_rtc_channel(void)
1073 {
1074 	unsigned long cfg;
1075 	cfg = hpet_readl(HPET_T1_CFG);
1076 	cfg &= ~HPET_TN_ENABLE;
1077 	hpet_writel(cfg, HPET_T1_CFG);
1078 }
1079 
1080 /*
1081  * The functions below are called from rtc driver.
1082  * Return 0 if HPET is not being used.
1083  * Otherwise do the necessary changes and return 1.
1084  */
1085 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1086 {
1087 	if (!is_hpet_enabled())
1088 		return 0;
1089 
1090 	hpet_rtc_flags &= ~bit_mask;
1091 	if (unlikely(!hpet_rtc_flags))
1092 		hpet_disable_rtc_channel();
1093 
1094 	return 1;
1095 }
1096 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1097 
1098 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1099 {
1100 	unsigned long oldbits = hpet_rtc_flags;
1101 
1102 	if (!is_hpet_enabled())
1103 		return 0;
1104 
1105 	hpet_rtc_flags |= bit_mask;
1106 
1107 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1108 		hpet_prev_update_sec = -1;
1109 
1110 	if (!oldbits)
1111 		hpet_rtc_timer_init();
1112 
1113 	return 1;
1114 }
1115 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1116 
1117 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1118 			unsigned char sec)
1119 {
1120 	if (!is_hpet_enabled())
1121 		return 0;
1122 
1123 	hpet_alarm_time.tm_hour = hrs;
1124 	hpet_alarm_time.tm_min = min;
1125 	hpet_alarm_time.tm_sec = sec;
1126 
1127 	return 1;
1128 }
1129 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1130 
1131 int hpet_set_periodic_freq(unsigned long freq)
1132 {
1133 	uint64_t clc;
1134 
1135 	if (!is_hpet_enabled())
1136 		return 0;
1137 
1138 	if (freq <= DEFAULT_RTC_INT_FREQ)
1139 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1140 	else {
1141 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1142 		do_div(clc, freq);
1143 		clc >>= hpet_clockevent.shift;
1144 		hpet_pie_delta = clc;
1145 		hpet_pie_limit = 0;
1146 	}
1147 	return 1;
1148 }
1149 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1150 
1151 int hpet_rtc_dropped_irq(void)
1152 {
1153 	return is_hpet_enabled();
1154 }
1155 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1156 
1157 static void hpet_rtc_timer_reinit(void)
1158 {
1159 	unsigned int delta;
1160 	int lost_ints = -1;
1161 
1162 	if (unlikely(!hpet_rtc_flags))
1163 		hpet_disable_rtc_channel();
1164 
1165 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1166 		delta = hpet_default_delta;
1167 	else
1168 		delta = hpet_pie_delta;
1169 
1170 	/*
1171 	 * Increment the comparator value until we are ahead of the
1172 	 * current count.
1173 	 */
1174 	do {
1175 		hpet_t1_cmp += delta;
1176 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1177 		lost_ints++;
1178 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1179 
1180 	if (lost_ints) {
1181 		if (hpet_rtc_flags & RTC_PIE)
1182 			hpet_pie_count += lost_ints;
1183 		if (printk_ratelimit())
1184 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1185 				lost_ints);
1186 	}
1187 }
1188 
1189 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1190 {
1191 	struct rtc_time curr_time;
1192 	unsigned long rtc_int_flag = 0;
1193 
1194 	hpet_rtc_timer_reinit();
1195 	memset(&curr_time, 0, sizeof(struct rtc_time));
1196 
1197 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1198 		get_rtc_time(&curr_time);
1199 
1200 	if (hpet_rtc_flags & RTC_UIE &&
1201 	    curr_time.tm_sec != hpet_prev_update_sec) {
1202 		if (hpet_prev_update_sec >= 0)
1203 			rtc_int_flag = RTC_UF;
1204 		hpet_prev_update_sec = curr_time.tm_sec;
1205 	}
1206 
1207 	if (hpet_rtc_flags & RTC_PIE &&
1208 	    ++hpet_pie_count >= hpet_pie_limit) {
1209 		rtc_int_flag |= RTC_PF;
1210 		hpet_pie_count = 0;
1211 	}
1212 
1213 	if (hpet_rtc_flags & RTC_AIE &&
1214 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1215 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1216 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1217 			rtc_int_flag |= RTC_AF;
1218 
1219 	if (rtc_int_flag) {
1220 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1221 		if (irq_handler)
1222 			irq_handler(rtc_int_flag, dev_id);
1223 	}
1224 	return IRQ_HANDLED;
1225 }
1226 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1227 #endif
1228