xref: /openbmc/linux/arch/x86/kernel/hpet.c (revision 609e478b)
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/fixmap.h>
16 #include <asm/hpet.h>
17 #include <asm/time.h>
18 
19 #define HPET_MASK			CLOCKSOURCE_MASK(32)
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 HPET_MIN_CYCLES			128
32 #define HPET_MIN_PROG_DELTA		(HPET_MIN_CYCLES + (HPET_MIN_CYCLES >> 1))
33 
34 /*
35  * HPET address is set in acpi/boot.c, when an ACPI entry exists
36  */
37 unsigned long				hpet_address;
38 u8					hpet_blockid; /* OS timer block num */
39 u8					hpet_msi_disable;
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 struct hpet_dev *EVT_TO_HPET_DEV(struct clock_event_device *evtdev)
56 {
57 	return container_of(evtdev, struct hpet_dev, evt);
58 }
59 
60 inline unsigned int hpet_readl(unsigned int a)
61 {
62 	return readl(hpet_virt_address + a);
63 }
64 
65 static inline void hpet_writel(unsigned int d, unsigned int a)
66 {
67 	writel(d, hpet_virt_address + a);
68 }
69 
70 #ifdef CONFIG_X86_64
71 #include <asm/pgtable.h>
72 #endif
73 
74 static inline void hpet_set_mapping(void)
75 {
76 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
77 }
78 
79 static inline void hpet_clear_mapping(void)
80 {
81 	iounmap(hpet_virt_address);
82 	hpet_virt_address = NULL;
83 }
84 
85 /*
86  * HPET command line enable / disable
87  */
88 int boot_hpet_disable;
89 int hpet_force_user;
90 static int hpet_verbose;
91 
92 static int __init hpet_setup(char *str)
93 {
94 	while (str) {
95 		char *next = strchr(str, ',');
96 
97 		if (next)
98 			*next++ = 0;
99 		if (!strncmp("disable", str, 7))
100 			boot_hpet_disable = 1;
101 		if (!strncmp("force", str, 5))
102 			hpet_force_user = 1;
103 		if (!strncmp("verbose", str, 7))
104 			hpet_verbose = 1;
105 		str = next;
106 	}
107 	return 1;
108 }
109 __setup("hpet=", hpet_setup);
110 
111 static int __init disable_hpet(char *str)
112 {
113 	boot_hpet_disable = 1;
114 	return 1;
115 }
116 __setup("nohpet", disable_hpet);
117 
118 static inline int is_hpet_capable(void)
119 {
120 	return !boot_hpet_disable && hpet_address;
121 }
122 
123 /*
124  * HPET timer interrupt enable / disable
125  */
126 static int hpet_legacy_int_enabled;
127 
128 /**
129  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
130  */
131 int is_hpet_enabled(void)
132 {
133 	return is_hpet_capable() && hpet_legacy_int_enabled;
134 }
135 EXPORT_SYMBOL_GPL(is_hpet_enabled);
136 
137 static void _hpet_print_config(const char *function, int line)
138 {
139 	u32 i, timers, l, h;
140 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
141 	l = hpet_readl(HPET_ID);
142 	h = hpet_readl(HPET_PERIOD);
143 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
144 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
145 	l = hpet_readl(HPET_CFG);
146 	h = hpet_readl(HPET_STATUS);
147 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
148 	l = hpet_readl(HPET_COUNTER);
149 	h = hpet_readl(HPET_COUNTER+4);
150 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
151 
152 	for (i = 0; i < timers; i++) {
153 		l = hpet_readl(HPET_Tn_CFG(i));
154 		h = hpet_readl(HPET_Tn_CFG(i)+4);
155 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
156 		       i, l, h);
157 		l = hpet_readl(HPET_Tn_CMP(i));
158 		h = hpet_readl(HPET_Tn_CMP(i)+4);
159 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
160 		       i, l, h);
161 		l = hpet_readl(HPET_Tn_ROUTE(i));
162 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
163 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
164 		       i, l, h);
165 	}
166 }
167 
168 #define hpet_print_config()					\
169 do {								\
170 	if (hpet_verbose)					\
171 		_hpet_print_config(__FUNCTION__, __LINE__);	\
172 } while (0)
173 
174 /*
175  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
176  * timer 0 and timer 1 in case of RTC emulation.
177  */
178 #ifdef CONFIG_HPET
179 
180 static void hpet_reserve_msi_timers(struct hpet_data *hd);
181 
182 static void hpet_reserve_platform_timers(unsigned int id)
183 {
184 	struct hpet __iomem *hpet = hpet_virt_address;
185 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
186 	unsigned int nrtimers, i;
187 	struct hpet_data hd;
188 
189 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
190 
191 	memset(&hd, 0, sizeof(hd));
192 	hd.hd_phys_address	= hpet_address;
193 	hd.hd_address		= hpet;
194 	hd.hd_nirqs		= nrtimers;
195 	hpet_reserve_timer(&hd, 0);
196 
197 #ifdef CONFIG_HPET_EMULATE_RTC
198 	hpet_reserve_timer(&hd, 1);
199 #endif
200 
201 	/*
202 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
203 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
204 	 * don't bother configuring *any* comparator interrupts.
205 	 */
206 	hd.hd_irq[0] = HPET_LEGACY_8254;
207 	hd.hd_irq[1] = HPET_LEGACY_RTC;
208 
209 	for (i = 2; i < nrtimers; timer++, i++) {
210 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
211 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
212 	}
213 
214 	hpet_reserve_msi_timers(&hd);
215 
216 	hpet_alloc(&hd);
217 
218 }
219 #else
220 static void hpet_reserve_platform_timers(unsigned int id) { }
221 #endif
222 
223 /*
224  * Common hpet info
225  */
226 static unsigned long hpet_freq;
227 
228 static void hpet_legacy_set_mode(enum clock_event_mode mode,
229 			  struct clock_event_device *evt);
230 static int hpet_legacy_next_event(unsigned long delta,
231 			   struct clock_event_device *evt);
232 
233 /*
234  * The hpet clock event device
235  */
236 static struct clock_event_device hpet_clockevent = {
237 	.name		= "hpet",
238 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
239 	.set_mode	= hpet_legacy_set_mode,
240 	.set_next_event = hpet_legacy_next_event,
241 	.irq		= 0,
242 	.rating		= 50,
243 };
244 
245 static void hpet_stop_counter(void)
246 {
247 	unsigned long cfg = hpet_readl(HPET_CFG);
248 	cfg &= ~HPET_CFG_ENABLE;
249 	hpet_writel(cfg, HPET_CFG);
250 }
251 
252 static void hpet_reset_counter(void)
253 {
254 	hpet_writel(0, HPET_COUNTER);
255 	hpet_writel(0, HPET_COUNTER + 4);
256 }
257 
258 static void hpet_start_counter(void)
259 {
260 	unsigned int cfg = hpet_readl(HPET_CFG);
261 	cfg |= HPET_CFG_ENABLE;
262 	hpet_writel(cfg, HPET_CFG);
263 }
264 
265 static void hpet_restart_counter(void)
266 {
267 	hpet_stop_counter();
268 	hpet_reset_counter();
269 	hpet_start_counter();
270 }
271 
272 static void hpet_resume_device(void)
273 {
274 	force_hpet_resume();
275 }
276 
277 static void hpet_resume_counter(struct clocksource *cs)
278 {
279 	hpet_resume_device();
280 	hpet_restart_counter();
281 }
282 
283 static void hpet_enable_legacy_int(void)
284 {
285 	unsigned int cfg = hpet_readl(HPET_CFG);
286 
287 	cfg |= HPET_CFG_LEGACY;
288 	hpet_writel(cfg, HPET_CFG);
289 	hpet_legacy_int_enabled = 1;
290 }
291 
292 static void hpet_legacy_clockevent_register(void)
293 {
294 	/* Start HPET legacy interrupts */
295 	hpet_enable_legacy_int();
296 
297 	/*
298 	 * Start hpet with the boot cpu mask and make it
299 	 * global after the IO_APIC has been initialized.
300 	 */
301 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
302 	clockevents_config_and_register(&hpet_clockevent, hpet_freq,
303 					HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
304 	global_clock_event = &hpet_clockevent;
305 	printk(KERN_DEBUG "hpet clockevent registered\n");
306 }
307 
308 static int hpet_setup_msi_irq(unsigned int irq);
309 
310 static void hpet_set_mode(enum clock_event_mode mode,
311 			  struct clock_event_device *evt, int timer)
312 {
313 	unsigned int cfg, cmp, now;
314 	uint64_t delta;
315 
316 	switch (mode) {
317 	case CLOCK_EVT_MODE_PERIODIC:
318 		hpet_stop_counter();
319 		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
320 		delta >>= evt->shift;
321 		now = hpet_readl(HPET_COUNTER);
322 		cmp = now + (unsigned int) delta;
323 		cfg = hpet_readl(HPET_Tn_CFG(timer));
324 		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
325 		       HPET_TN_SETVAL | HPET_TN_32BIT;
326 		hpet_writel(cfg, HPET_Tn_CFG(timer));
327 		hpet_writel(cmp, HPET_Tn_CMP(timer));
328 		udelay(1);
329 		/*
330 		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
331 		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
332 		 * bit is automatically cleared after the first write.
333 		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
334 		 * Publication # 24674)
335 		 */
336 		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
337 		hpet_start_counter();
338 		hpet_print_config();
339 		break;
340 
341 	case CLOCK_EVT_MODE_ONESHOT:
342 		cfg = hpet_readl(HPET_Tn_CFG(timer));
343 		cfg &= ~HPET_TN_PERIODIC;
344 		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
345 		hpet_writel(cfg, HPET_Tn_CFG(timer));
346 		break;
347 
348 	case CLOCK_EVT_MODE_UNUSED:
349 	case CLOCK_EVT_MODE_SHUTDOWN:
350 		cfg = hpet_readl(HPET_Tn_CFG(timer));
351 		cfg &= ~HPET_TN_ENABLE;
352 		hpet_writel(cfg, HPET_Tn_CFG(timer));
353 		break;
354 
355 	case CLOCK_EVT_MODE_RESUME:
356 		if (timer == 0) {
357 			hpet_enable_legacy_int();
358 		} else {
359 			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
360 			hpet_setup_msi_irq(hdev->irq);
361 			disable_irq(hdev->irq);
362 			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
363 			enable_irq(hdev->irq);
364 		}
365 		hpet_print_config();
366 		break;
367 	}
368 }
369 
370 static int hpet_next_event(unsigned long delta,
371 			   struct clock_event_device *evt, int timer)
372 {
373 	u32 cnt;
374 	s32 res;
375 
376 	cnt = hpet_readl(HPET_COUNTER);
377 	cnt += (u32) delta;
378 	hpet_writel(cnt, HPET_Tn_CMP(timer));
379 
380 	/*
381 	 * HPETs are a complete disaster. The compare register is
382 	 * based on a equal comparison and neither provides a less
383 	 * than or equal functionality (which would require to take
384 	 * the wraparound into account) nor a simple count down event
385 	 * mode. Further the write to the comparator register is
386 	 * delayed internally up to two HPET clock cycles in certain
387 	 * chipsets (ATI, ICH9,10). Some newer AMD chipsets have even
388 	 * longer delays. We worked around that by reading back the
389 	 * compare register, but that required another workaround for
390 	 * ICH9,10 chips where the first readout after write can
391 	 * return the old stale value. We already had a minimum
392 	 * programming delta of 5us enforced, but a NMI or SMI hitting
393 	 * between the counter readout and the comparator write can
394 	 * move us behind that point easily. Now instead of reading
395 	 * the compare register back several times, we make the ETIME
396 	 * decision based on the following: Return ETIME if the
397 	 * counter value after the write is less than HPET_MIN_CYCLES
398 	 * away from the event or if the counter is already ahead of
399 	 * the event. The minimum programming delta for the generic
400 	 * clockevents code is set to 1.5 * HPET_MIN_CYCLES.
401 	 */
402 	res = (s32)(cnt - hpet_readl(HPET_COUNTER));
403 
404 	return res < HPET_MIN_CYCLES ? -ETIME : 0;
405 }
406 
407 static void hpet_legacy_set_mode(enum clock_event_mode mode,
408 			struct clock_event_device *evt)
409 {
410 	hpet_set_mode(mode, evt, 0);
411 }
412 
413 static int hpet_legacy_next_event(unsigned long delta,
414 			struct clock_event_device *evt)
415 {
416 	return hpet_next_event(delta, evt, 0);
417 }
418 
419 /*
420  * HPET MSI Support
421  */
422 #ifdef CONFIG_PCI_MSI
423 
424 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
425 static struct hpet_dev	*hpet_devs;
426 
427 void hpet_msi_unmask(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 int hpet_setup_msi_irq(unsigned int irq)
477 {
478 	if (x86_msi.setup_hpet_msi(irq, hpet_blockid)) {
479 		irq_free_hwirq(irq);
480 		return -EINVAL;
481 	}
482 	return 0;
483 }
484 
485 static int hpet_assign_irq(struct hpet_dev *dev)
486 {
487 	unsigned int irq = irq_alloc_hwirq(-1);
488 
489 	if (!irq)
490 		return -EINVAL;
491 
492 	irq_set_handler_data(irq, dev);
493 
494 	if (hpet_setup_msi_irq(irq))
495 		return -EINVAL;
496 
497 	dev->irq = irq;
498 	return 0;
499 }
500 
501 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
502 {
503 	struct hpet_dev *dev = (struct hpet_dev *)data;
504 	struct clock_event_device *hevt = &dev->evt;
505 
506 	if (!hevt->event_handler) {
507 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
508 				dev->num);
509 		return IRQ_HANDLED;
510 	}
511 
512 	hevt->event_handler(hevt);
513 	return IRQ_HANDLED;
514 }
515 
516 static int hpet_setup_irq(struct hpet_dev *dev)
517 {
518 
519 	if (request_irq(dev->irq, hpet_interrupt_handler,
520 			IRQF_TIMER | IRQF_NOBALANCING,
521 			dev->name, dev))
522 		return -1;
523 
524 	disable_irq(dev->irq);
525 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
526 	enable_irq(dev->irq);
527 
528 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
529 			 dev->name, dev->irq);
530 
531 	return 0;
532 }
533 
534 /* This should be called in specific @cpu */
535 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
536 {
537 	struct clock_event_device *evt = &hdev->evt;
538 
539 	WARN_ON(cpu != smp_processor_id());
540 	if (!(hdev->flags & HPET_DEV_VALID))
541 		return;
542 
543 	if (hpet_setup_msi_irq(hdev->irq))
544 		return;
545 
546 	hdev->cpu = cpu;
547 	per_cpu(cpu_hpet_dev, cpu) = hdev;
548 	evt->name = hdev->name;
549 	hpet_setup_irq(hdev);
550 	evt->irq = hdev->irq;
551 
552 	evt->rating = 110;
553 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
554 	if (hdev->flags & HPET_DEV_PERI_CAP)
555 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
556 
557 	evt->set_mode = hpet_msi_set_mode;
558 	evt->set_next_event = hpet_msi_next_event;
559 	evt->cpumask = cpumask_of(hdev->cpu);
560 
561 	clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
562 					0x7FFFFFFF);
563 }
564 
565 #ifdef CONFIG_HPET
566 /* Reserve at least one timer for userspace (/dev/hpet) */
567 #define RESERVE_TIMERS 1
568 #else
569 #define RESERVE_TIMERS 0
570 #endif
571 
572 static void hpet_msi_capability_lookup(unsigned int start_timer)
573 {
574 	unsigned int id;
575 	unsigned int num_timers;
576 	unsigned int num_timers_used = 0;
577 	int i;
578 
579 	if (hpet_msi_disable)
580 		return;
581 
582 	if (boot_cpu_has(X86_FEATURE_ARAT))
583 		return;
584 	id = hpet_readl(HPET_ID);
585 
586 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
587 	num_timers++; /* Value read out starts from 0 */
588 	hpet_print_config();
589 
590 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
591 	if (!hpet_devs)
592 		return;
593 
594 	hpet_num_timers = num_timers;
595 
596 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
597 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
598 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
599 
600 		/* Only consider HPET timer with MSI support */
601 		if (!(cfg & HPET_TN_FSB_CAP))
602 			continue;
603 
604 		hdev->flags = 0;
605 		if (cfg & HPET_TN_PERIODIC_CAP)
606 			hdev->flags |= HPET_DEV_PERI_CAP;
607 		hdev->num = i;
608 
609 		sprintf(hdev->name, "hpet%d", i);
610 		if (hpet_assign_irq(hdev))
611 			continue;
612 
613 		hdev->flags |= HPET_DEV_FSB_CAP;
614 		hdev->flags |= HPET_DEV_VALID;
615 		num_timers_used++;
616 		if (num_timers_used == num_possible_cpus())
617 			break;
618 	}
619 
620 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
621 		num_timers, num_timers_used);
622 }
623 
624 #ifdef CONFIG_HPET
625 static void hpet_reserve_msi_timers(struct hpet_data *hd)
626 {
627 	int i;
628 
629 	if (!hpet_devs)
630 		return;
631 
632 	for (i = 0; i < hpet_num_timers; i++) {
633 		struct hpet_dev *hdev = &hpet_devs[i];
634 
635 		if (!(hdev->flags & HPET_DEV_VALID))
636 			continue;
637 
638 		hd->hd_irq[hdev->num] = hdev->irq;
639 		hpet_reserve_timer(hd, hdev->num);
640 	}
641 }
642 #endif
643 
644 static struct hpet_dev *hpet_get_unused_timer(void)
645 {
646 	int i;
647 
648 	if (!hpet_devs)
649 		return NULL;
650 
651 	for (i = 0; i < hpet_num_timers; i++) {
652 		struct hpet_dev *hdev = &hpet_devs[i];
653 
654 		if (!(hdev->flags & HPET_DEV_VALID))
655 			continue;
656 		if (test_and_set_bit(HPET_DEV_USED_BIT,
657 			(unsigned long *)&hdev->flags))
658 			continue;
659 		return hdev;
660 	}
661 	return NULL;
662 }
663 
664 struct hpet_work_struct {
665 	struct delayed_work work;
666 	struct completion complete;
667 };
668 
669 static void hpet_work(struct work_struct *w)
670 {
671 	struct hpet_dev *hdev;
672 	int cpu = smp_processor_id();
673 	struct hpet_work_struct *hpet_work;
674 
675 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
676 
677 	hdev = hpet_get_unused_timer();
678 	if (hdev)
679 		init_one_hpet_msi_clockevent(hdev, cpu);
680 
681 	complete(&hpet_work->complete);
682 }
683 
684 static int hpet_cpuhp_notify(struct notifier_block *n,
685 		unsigned long action, void *hcpu)
686 {
687 	unsigned long cpu = (unsigned long)hcpu;
688 	struct hpet_work_struct work;
689 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
690 
691 	switch (action & 0xf) {
692 	case CPU_ONLINE:
693 		INIT_DELAYED_WORK_ONSTACK(&work.work, hpet_work);
694 		init_completion(&work.complete);
695 		/* FIXME: add schedule_work_on() */
696 		schedule_delayed_work_on(cpu, &work.work, 0);
697 		wait_for_completion(&work.complete);
698 		destroy_delayed_work_on_stack(&work.work);
699 		break;
700 	case CPU_DEAD:
701 		if (hdev) {
702 			free_irq(hdev->irq, hdev);
703 			hdev->flags &= ~HPET_DEV_USED;
704 			per_cpu(cpu_hpet_dev, cpu) = NULL;
705 		}
706 		break;
707 	}
708 	return NOTIFY_OK;
709 }
710 #else
711 
712 static int hpet_setup_msi_irq(unsigned int irq)
713 {
714 	return 0;
715 }
716 static void hpet_msi_capability_lookup(unsigned int start_timer)
717 {
718 	return;
719 }
720 
721 #ifdef CONFIG_HPET
722 static void hpet_reserve_msi_timers(struct hpet_data *hd)
723 {
724 	return;
725 }
726 #endif
727 
728 static int hpet_cpuhp_notify(struct notifier_block *n,
729 		unsigned long action, void *hcpu)
730 {
731 	return NOTIFY_OK;
732 }
733 
734 #endif
735 
736 /*
737  * Clock source related code
738  */
739 static cycle_t read_hpet(struct clocksource *cs)
740 {
741 	return (cycle_t)hpet_readl(HPET_COUNTER);
742 }
743 
744 static struct clocksource clocksource_hpet = {
745 	.name		= "hpet",
746 	.rating		= 250,
747 	.read		= read_hpet,
748 	.mask		= HPET_MASK,
749 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
750 	.resume		= hpet_resume_counter,
751 	.archdata	= { .vclock_mode = VCLOCK_HPET },
752 };
753 
754 static int hpet_clocksource_register(void)
755 {
756 	u64 start, now;
757 	cycle_t t1;
758 
759 	/* Start the counter */
760 	hpet_restart_counter();
761 
762 	/* Verify whether hpet counter works */
763 	t1 = hpet_readl(HPET_COUNTER);
764 	rdtscll(start);
765 
766 	/*
767 	 * We don't know the TSC frequency yet, but waiting for
768 	 * 200000 TSC cycles is safe:
769 	 * 4 GHz == 50us
770 	 * 1 GHz == 200us
771 	 */
772 	do {
773 		rep_nop();
774 		rdtscll(now);
775 	} while ((now - start) < 200000UL);
776 
777 	if (t1 == hpet_readl(HPET_COUNTER)) {
778 		printk(KERN_WARNING
779 		       "HPET counter not counting. HPET disabled\n");
780 		return -ENODEV;
781 	}
782 
783 	clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
784 	return 0;
785 }
786 
787 static u32 *hpet_boot_cfg;
788 
789 /**
790  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
791  */
792 int __init hpet_enable(void)
793 {
794 	u32 hpet_period, cfg, id;
795 	u64 freq;
796 	unsigned int i, last;
797 
798 	if (!is_hpet_capable())
799 		return 0;
800 
801 	hpet_set_mapping();
802 
803 	/*
804 	 * Read the period and check for a sane value:
805 	 */
806 	hpet_period = hpet_readl(HPET_PERIOD);
807 
808 	/*
809 	 * AMD SB700 based systems with spread spectrum enabled use a
810 	 * SMM based HPET emulation to provide proper frequency
811 	 * setting. The SMM code is initialized with the first HPET
812 	 * register access and takes some time to complete. During
813 	 * this time the config register reads 0xffffffff. We check
814 	 * for max. 1000 loops whether the config register reads a non
815 	 * 0xffffffff value to make sure that HPET is up and running
816 	 * before we go further. A counting loop is safe, as the HPET
817 	 * access takes thousands of CPU cycles. On non SB700 based
818 	 * machines this check is only done once and has no side
819 	 * effects.
820 	 */
821 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
822 		if (i == 1000) {
823 			printk(KERN_WARNING
824 			       "HPET config register value = 0xFFFFFFFF. "
825 			       "Disabling HPET\n");
826 			goto out_nohpet;
827 		}
828 	}
829 
830 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
831 		goto out_nohpet;
832 
833 	/*
834 	 * The period is a femto seconds value. Convert it to a
835 	 * frequency.
836 	 */
837 	freq = FSEC_PER_SEC;
838 	do_div(freq, hpet_period);
839 	hpet_freq = freq;
840 
841 	/*
842 	 * Read the HPET ID register to retrieve the IRQ routing
843 	 * information and the number of channels
844 	 */
845 	id = hpet_readl(HPET_ID);
846 	hpet_print_config();
847 
848 	last = (id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT;
849 
850 #ifdef CONFIG_HPET_EMULATE_RTC
851 	/*
852 	 * The legacy routing mode needs at least two channels, tick timer
853 	 * and the rtc emulation channel.
854 	 */
855 	if (!last)
856 		goto out_nohpet;
857 #endif
858 
859 	cfg = hpet_readl(HPET_CFG);
860 	hpet_boot_cfg = kmalloc((last + 2) * sizeof(*hpet_boot_cfg),
861 				GFP_KERNEL);
862 	if (hpet_boot_cfg)
863 		*hpet_boot_cfg = cfg;
864 	else
865 		pr_warn("HPET initial state will not be saved\n");
866 	cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
867 	hpet_writel(cfg, HPET_CFG);
868 	if (cfg)
869 		pr_warn("HPET: Unrecognized bits %#x set in global cfg\n",
870 			cfg);
871 
872 	for (i = 0; i <= last; ++i) {
873 		cfg = hpet_readl(HPET_Tn_CFG(i));
874 		if (hpet_boot_cfg)
875 			hpet_boot_cfg[i + 1] = cfg;
876 		cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
877 		hpet_writel(cfg, HPET_Tn_CFG(i));
878 		cfg &= ~(HPET_TN_PERIODIC | HPET_TN_PERIODIC_CAP
879 			 | HPET_TN_64BIT_CAP | HPET_TN_32BIT | HPET_TN_ROUTE
880 			 | HPET_TN_FSB | HPET_TN_FSB_CAP);
881 		if (cfg)
882 			pr_warn("HPET: Unrecognized bits %#x set in cfg#%u\n",
883 				cfg, i);
884 	}
885 	hpet_print_config();
886 
887 	if (hpet_clocksource_register())
888 		goto out_nohpet;
889 
890 	if (id & HPET_ID_LEGSUP) {
891 		hpet_legacy_clockevent_register();
892 		return 1;
893 	}
894 	return 0;
895 
896 out_nohpet:
897 	hpet_clear_mapping();
898 	hpet_address = 0;
899 	return 0;
900 }
901 
902 /*
903  * Needs to be late, as the reserve_timer code calls kalloc !
904  *
905  * Not a problem on i386 as hpet_enable is called from late_time_init,
906  * but on x86_64 it is necessary !
907  */
908 static __init int hpet_late_init(void)
909 {
910 	int cpu;
911 
912 	if (boot_hpet_disable)
913 		return -ENODEV;
914 
915 	if (!hpet_address) {
916 		if (!force_hpet_address)
917 			return -ENODEV;
918 
919 		hpet_address = force_hpet_address;
920 		hpet_enable();
921 	}
922 
923 	if (!hpet_virt_address)
924 		return -ENODEV;
925 
926 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
927 		hpet_msi_capability_lookup(2);
928 	else
929 		hpet_msi_capability_lookup(0);
930 
931 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
932 	hpet_print_config();
933 
934 	if (hpet_msi_disable)
935 		return 0;
936 
937 	if (boot_cpu_has(X86_FEATURE_ARAT))
938 		return 0;
939 
940 	cpu_notifier_register_begin();
941 	for_each_online_cpu(cpu) {
942 		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
943 	}
944 
945 	/* This notifier should be called after workqueue is ready */
946 	__hotcpu_notifier(hpet_cpuhp_notify, -20);
947 	cpu_notifier_register_done();
948 
949 	return 0;
950 }
951 fs_initcall(hpet_late_init);
952 
953 void hpet_disable(void)
954 {
955 	if (is_hpet_capable() && hpet_virt_address) {
956 		unsigned int cfg = hpet_readl(HPET_CFG), id, last;
957 
958 		if (hpet_boot_cfg)
959 			cfg = *hpet_boot_cfg;
960 		else if (hpet_legacy_int_enabled) {
961 			cfg &= ~HPET_CFG_LEGACY;
962 			hpet_legacy_int_enabled = 0;
963 		}
964 		cfg &= ~HPET_CFG_ENABLE;
965 		hpet_writel(cfg, HPET_CFG);
966 
967 		if (!hpet_boot_cfg)
968 			return;
969 
970 		id = hpet_readl(HPET_ID);
971 		last = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
972 
973 		for (id = 0; id <= last; ++id)
974 			hpet_writel(hpet_boot_cfg[id + 1], HPET_Tn_CFG(id));
975 
976 		if (*hpet_boot_cfg & HPET_CFG_ENABLE)
977 			hpet_writel(*hpet_boot_cfg, HPET_CFG);
978 	}
979 }
980 
981 #ifdef CONFIG_HPET_EMULATE_RTC
982 
983 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
984  * is enabled, we support RTC interrupt functionality in software.
985  * RTC has 3 kinds of interrupts:
986  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
987  *    is updated
988  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
989  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
990  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
991  * (1) and (2) above are implemented using polling at a frequency of
992  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
993  * overhead. (DEFAULT_RTC_INT_FREQ)
994  * For (3), we use interrupts at 64Hz or user specified periodic
995  * frequency, whichever is higher.
996  */
997 #include <linux/mc146818rtc.h>
998 #include <linux/rtc.h>
999 #include <asm/rtc.h>
1000 
1001 #define DEFAULT_RTC_INT_FREQ	64
1002 #define DEFAULT_RTC_SHIFT	6
1003 #define RTC_NUM_INTS		1
1004 
1005 static unsigned long hpet_rtc_flags;
1006 static int hpet_prev_update_sec;
1007 static struct rtc_time hpet_alarm_time;
1008 static unsigned long hpet_pie_count;
1009 static u32 hpet_t1_cmp;
1010 static u32 hpet_default_delta;
1011 static u32 hpet_pie_delta;
1012 static unsigned long hpet_pie_limit;
1013 
1014 static rtc_irq_handler irq_handler;
1015 
1016 /*
1017  * Check that the hpet counter c1 is ahead of the c2
1018  */
1019 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1020 {
1021 	return (s32)(c2 - c1) < 0;
1022 }
1023 
1024 /*
1025  * Registers a IRQ handler.
1026  */
1027 int hpet_register_irq_handler(rtc_irq_handler handler)
1028 {
1029 	if (!is_hpet_enabled())
1030 		return -ENODEV;
1031 	if (irq_handler)
1032 		return -EBUSY;
1033 
1034 	irq_handler = handler;
1035 
1036 	return 0;
1037 }
1038 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1039 
1040 /*
1041  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1042  * and does cleanup.
1043  */
1044 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1045 {
1046 	if (!is_hpet_enabled())
1047 		return;
1048 
1049 	irq_handler = NULL;
1050 	hpet_rtc_flags = 0;
1051 }
1052 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1053 
1054 /*
1055  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1056  * is not supported by all HPET implementations for timer 1.
1057  *
1058  * hpet_rtc_timer_init() is called when the rtc is initialized.
1059  */
1060 int hpet_rtc_timer_init(void)
1061 {
1062 	unsigned int cfg, cnt, delta;
1063 	unsigned long flags;
1064 
1065 	if (!is_hpet_enabled())
1066 		return 0;
1067 
1068 	if (!hpet_default_delta) {
1069 		uint64_t clc;
1070 
1071 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1072 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1073 		hpet_default_delta = clc;
1074 	}
1075 
1076 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1077 		delta = hpet_default_delta;
1078 	else
1079 		delta = hpet_pie_delta;
1080 
1081 	local_irq_save(flags);
1082 
1083 	cnt = delta + hpet_readl(HPET_COUNTER);
1084 	hpet_writel(cnt, HPET_T1_CMP);
1085 	hpet_t1_cmp = cnt;
1086 
1087 	cfg = hpet_readl(HPET_T1_CFG);
1088 	cfg &= ~HPET_TN_PERIODIC;
1089 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1090 	hpet_writel(cfg, HPET_T1_CFG);
1091 
1092 	local_irq_restore(flags);
1093 
1094 	return 1;
1095 }
1096 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1097 
1098 static void hpet_disable_rtc_channel(void)
1099 {
1100 	unsigned long cfg;
1101 	cfg = hpet_readl(HPET_T1_CFG);
1102 	cfg &= ~HPET_TN_ENABLE;
1103 	hpet_writel(cfg, HPET_T1_CFG);
1104 }
1105 
1106 /*
1107  * The functions below are called from rtc driver.
1108  * Return 0 if HPET is not being used.
1109  * Otherwise do the necessary changes and return 1.
1110  */
1111 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1112 {
1113 	if (!is_hpet_enabled())
1114 		return 0;
1115 
1116 	hpet_rtc_flags &= ~bit_mask;
1117 	if (unlikely(!hpet_rtc_flags))
1118 		hpet_disable_rtc_channel();
1119 
1120 	return 1;
1121 }
1122 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1123 
1124 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1125 {
1126 	unsigned long oldbits = hpet_rtc_flags;
1127 
1128 	if (!is_hpet_enabled())
1129 		return 0;
1130 
1131 	hpet_rtc_flags |= bit_mask;
1132 
1133 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1134 		hpet_prev_update_sec = -1;
1135 
1136 	if (!oldbits)
1137 		hpet_rtc_timer_init();
1138 
1139 	return 1;
1140 }
1141 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1142 
1143 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1144 			unsigned char sec)
1145 {
1146 	if (!is_hpet_enabled())
1147 		return 0;
1148 
1149 	hpet_alarm_time.tm_hour = hrs;
1150 	hpet_alarm_time.tm_min = min;
1151 	hpet_alarm_time.tm_sec = sec;
1152 
1153 	return 1;
1154 }
1155 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1156 
1157 int hpet_set_periodic_freq(unsigned long freq)
1158 {
1159 	uint64_t clc;
1160 
1161 	if (!is_hpet_enabled())
1162 		return 0;
1163 
1164 	if (freq <= DEFAULT_RTC_INT_FREQ)
1165 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1166 	else {
1167 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1168 		do_div(clc, freq);
1169 		clc >>= hpet_clockevent.shift;
1170 		hpet_pie_delta = clc;
1171 		hpet_pie_limit = 0;
1172 	}
1173 	return 1;
1174 }
1175 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1176 
1177 int hpet_rtc_dropped_irq(void)
1178 {
1179 	return is_hpet_enabled();
1180 }
1181 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1182 
1183 static void hpet_rtc_timer_reinit(void)
1184 {
1185 	unsigned int delta;
1186 	int lost_ints = -1;
1187 
1188 	if (unlikely(!hpet_rtc_flags))
1189 		hpet_disable_rtc_channel();
1190 
1191 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1192 		delta = hpet_default_delta;
1193 	else
1194 		delta = hpet_pie_delta;
1195 
1196 	/*
1197 	 * Increment the comparator value until we are ahead of the
1198 	 * current count.
1199 	 */
1200 	do {
1201 		hpet_t1_cmp += delta;
1202 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1203 		lost_ints++;
1204 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1205 
1206 	if (lost_ints) {
1207 		if (hpet_rtc_flags & RTC_PIE)
1208 			hpet_pie_count += lost_ints;
1209 		if (printk_ratelimit())
1210 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1211 				lost_ints);
1212 	}
1213 }
1214 
1215 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1216 {
1217 	struct rtc_time curr_time;
1218 	unsigned long rtc_int_flag = 0;
1219 
1220 	hpet_rtc_timer_reinit();
1221 	memset(&curr_time, 0, sizeof(struct rtc_time));
1222 
1223 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1224 		get_rtc_time(&curr_time);
1225 
1226 	if (hpet_rtc_flags & RTC_UIE &&
1227 	    curr_time.tm_sec != hpet_prev_update_sec) {
1228 		if (hpet_prev_update_sec >= 0)
1229 			rtc_int_flag = RTC_UF;
1230 		hpet_prev_update_sec = curr_time.tm_sec;
1231 	}
1232 
1233 	if (hpet_rtc_flags & RTC_PIE &&
1234 	    ++hpet_pie_count >= hpet_pie_limit) {
1235 		rtc_int_flag |= RTC_PF;
1236 		hpet_pie_count = 0;
1237 	}
1238 
1239 	if (hpet_rtc_flags & RTC_AIE &&
1240 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1241 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1242 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1243 			rtc_int_flag |= RTC_AF;
1244 
1245 	if (rtc_int_flag) {
1246 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1247 		if (irq_handler)
1248 			irq_handler(rtc_int_flag, dev_id);
1249 	}
1250 	return IRQ_HANDLED;
1251 }
1252 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1253 #endif
1254