xref: /linux/arch/x86/kernel/hpet.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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 #ifdef CONFIG_X86_64
78 	__set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VVAR_NOCACHE);
79 #endif
80 }
81 
82 static inline void hpet_clear_mapping(void)
83 {
84 	iounmap(hpet_virt_address);
85 	hpet_virt_address = NULL;
86 }
87 
88 /*
89  * HPET command line enable / disable
90  */
91 static int boot_hpet_disable;
92 int hpet_force_user;
93 static int hpet_verbose;
94 
95 static int __init hpet_setup(char *str)
96 {
97 	while (str) {
98 		char *next = strchr(str, ',');
99 
100 		if (next)
101 			*next++ = 0;
102 		if (!strncmp("disable", str, 7))
103 			boot_hpet_disable = 1;
104 		if (!strncmp("force", str, 5))
105 			hpet_force_user = 1;
106 		if (!strncmp("verbose", str, 7))
107 			hpet_verbose = 1;
108 		str = next;
109 	}
110 	return 1;
111 }
112 __setup("hpet=", hpet_setup);
113 
114 static int __init disable_hpet(char *str)
115 {
116 	boot_hpet_disable = 1;
117 	return 1;
118 }
119 __setup("nohpet", disable_hpet);
120 
121 static inline int is_hpet_capable(void)
122 {
123 	return !boot_hpet_disable && hpet_address;
124 }
125 
126 /*
127  * HPET timer interrupt enable / disable
128  */
129 static int hpet_legacy_int_enabled;
130 
131 /**
132  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
133  */
134 int is_hpet_enabled(void)
135 {
136 	return is_hpet_capable() && hpet_legacy_int_enabled;
137 }
138 EXPORT_SYMBOL_GPL(is_hpet_enabled);
139 
140 static void _hpet_print_config(const char *function, int line)
141 {
142 	u32 i, timers, l, h;
143 	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
144 	l = hpet_readl(HPET_ID);
145 	h = hpet_readl(HPET_PERIOD);
146 	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
147 	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
148 	l = hpet_readl(HPET_CFG);
149 	h = hpet_readl(HPET_STATUS);
150 	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
151 	l = hpet_readl(HPET_COUNTER);
152 	h = hpet_readl(HPET_COUNTER+4);
153 	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
154 
155 	for (i = 0; i < timers; i++) {
156 		l = hpet_readl(HPET_Tn_CFG(i));
157 		h = hpet_readl(HPET_Tn_CFG(i)+4);
158 		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
159 		       i, l, h);
160 		l = hpet_readl(HPET_Tn_CMP(i));
161 		h = hpet_readl(HPET_Tn_CMP(i)+4);
162 		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
163 		       i, l, h);
164 		l = hpet_readl(HPET_Tn_ROUTE(i));
165 		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
166 		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
167 		       i, l, h);
168 	}
169 }
170 
171 #define hpet_print_config()					\
172 do {								\
173 	if (hpet_verbose)					\
174 		_hpet_print_config(__FUNCTION__, __LINE__);	\
175 } while (0)
176 
177 /*
178  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
179  * timer 0 and timer 1 in case of RTC emulation.
180  */
181 #ifdef CONFIG_HPET
182 
183 static void hpet_reserve_msi_timers(struct hpet_data *hd);
184 
185 static void hpet_reserve_platform_timers(unsigned int id)
186 {
187 	struct hpet __iomem *hpet = hpet_virt_address;
188 	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
189 	unsigned int nrtimers, i;
190 	struct hpet_data hd;
191 
192 	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
193 
194 	memset(&hd, 0, sizeof(hd));
195 	hd.hd_phys_address	= hpet_address;
196 	hd.hd_address		= hpet;
197 	hd.hd_nirqs		= nrtimers;
198 	hpet_reserve_timer(&hd, 0);
199 
200 #ifdef CONFIG_HPET_EMULATE_RTC
201 	hpet_reserve_timer(&hd, 1);
202 #endif
203 
204 	/*
205 	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
206 	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
207 	 * don't bother configuring *any* comparator interrupts.
208 	 */
209 	hd.hd_irq[0] = HPET_LEGACY_8254;
210 	hd.hd_irq[1] = HPET_LEGACY_RTC;
211 
212 	for (i = 2; i < nrtimers; timer++, i++) {
213 		hd.hd_irq[i] = (readl(&timer->hpet_config) &
214 			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
215 	}
216 
217 	hpet_reserve_msi_timers(&hd);
218 
219 	hpet_alloc(&hd);
220 
221 }
222 #else
223 static void hpet_reserve_platform_timers(unsigned int id) { }
224 #endif
225 
226 /*
227  * Common hpet info
228  */
229 static unsigned long hpet_freq;
230 
231 static void hpet_legacy_set_mode(enum clock_event_mode mode,
232 			  struct clock_event_device *evt);
233 static int hpet_legacy_next_event(unsigned long delta,
234 			   struct clock_event_device *evt);
235 
236 /*
237  * The hpet clock event device
238  */
239 static struct clock_event_device hpet_clockevent = {
240 	.name		= "hpet",
241 	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
242 	.set_mode	= hpet_legacy_set_mode,
243 	.set_next_event = hpet_legacy_next_event,
244 	.irq		= 0,
245 	.rating		= 50,
246 };
247 
248 static void hpet_stop_counter(void)
249 {
250 	unsigned long cfg = hpet_readl(HPET_CFG);
251 	cfg &= ~HPET_CFG_ENABLE;
252 	hpet_writel(cfg, HPET_CFG);
253 }
254 
255 static void hpet_reset_counter(void)
256 {
257 	hpet_writel(0, HPET_COUNTER);
258 	hpet_writel(0, HPET_COUNTER + 4);
259 }
260 
261 static void hpet_start_counter(void)
262 {
263 	unsigned int cfg = hpet_readl(HPET_CFG);
264 	cfg |= HPET_CFG_ENABLE;
265 	hpet_writel(cfg, HPET_CFG);
266 }
267 
268 static void hpet_restart_counter(void)
269 {
270 	hpet_stop_counter();
271 	hpet_reset_counter();
272 	hpet_start_counter();
273 }
274 
275 static void hpet_resume_device(void)
276 {
277 	force_hpet_resume();
278 }
279 
280 static void hpet_resume_counter(struct clocksource *cs)
281 {
282 	hpet_resume_device();
283 	hpet_restart_counter();
284 }
285 
286 static void hpet_enable_legacy_int(void)
287 {
288 	unsigned int cfg = hpet_readl(HPET_CFG);
289 
290 	cfg |= HPET_CFG_LEGACY;
291 	hpet_writel(cfg, HPET_CFG);
292 	hpet_legacy_int_enabled = 1;
293 }
294 
295 static void hpet_legacy_clockevent_register(void)
296 {
297 	/* Start HPET legacy interrupts */
298 	hpet_enable_legacy_int();
299 
300 	/*
301 	 * Start hpet with the boot cpu mask and make it
302 	 * global after the IO_APIC has been initialized.
303 	 */
304 	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
305 	clockevents_config_and_register(&hpet_clockevent, hpet_freq,
306 					HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
307 	global_clock_event = &hpet_clockevent;
308 	printk(KERN_DEBUG "hpet clockevent registered\n");
309 }
310 
311 static int hpet_setup_msi_irq(unsigned int irq);
312 
313 static void hpet_set_mode(enum clock_event_mode mode,
314 			  struct clock_event_device *evt, int timer)
315 {
316 	unsigned int cfg, cmp, now;
317 	uint64_t delta;
318 
319 	switch (mode) {
320 	case CLOCK_EVT_MODE_PERIODIC:
321 		hpet_stop_counter();
322 		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
323 		delta >>= evt->shift;
324 		now = hpet_readl(HPET_COUNTER);
325 		cmp = now + (unsigned int) delta;
326 		cfg = hpet_readl(HPET_Tn_CFG(timer));
327 		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
328 		       HPET_TN_SETVAL | HPET_TN_32BIT;
329 		hpet_writel(cfg, HPET_Tn_CFG(timer));
330 		hpet_writel(cmp, HPET_Tn_CMP(timer));
331 		udelay(1);
332 		/*
333 		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
334 		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
335 		 * bit is automatically cleared after the first write.
336 		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
337 		 * Publication # 24674)
338 		 */
339 		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
340 		hpet_start_counter();
341 		hpet_print_config();
342 		break;
343 
344 	case CLOCK_EVT_MODE_ONESHOT:
345 		cfg = hpet_readl(HPET_Tn_CFG(timer));
346 		cfg &= ~HPET_TN_PERIODIC;
347 		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
348 		hpet_writel(cfg, HPET_Tn_CFG(timer));
349 		break;
350 
351 	case CLOCK_EVT_MODE_UNUSED:
352 	case CLOCK_EVT_MODE_SHUTDOWN:
353 		cfg = hpet_readl(HPET_Tn_CFG(timer));
354 		cfg &= ~HPET_TN_ENABLE;
355 		hpet_writel(cfg, HPET_Tn_CFG(timer));
356 		break;
357 
358 	case CLOCK_EVT_MODE_RESUME:
359 		if (timer == 0) {
360 			hpet_enable_legacy_int();
361 		} else {
362 			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
363 			hpet_setup_msi_irq(hdev->irq);
364 			disable_irq(hdev->irq);
365 			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
366 			enable_irq(hdev->irq);
367 		}
368 		hpet_print_config();
369 		break;
370 	}
371 }
372 
373 static int hpet_next_event(unsigned long delta,
374 			   struct clock_event_device *evt, int timer)
375 {
376 	u32 cnt;
377 	s32 res;
378 
379 	cnt = hpet_readl(HPET_COUNTER);
380 	cnt += (u32) delta;
381 	hpet_writel(cnt, HPET_Tn_CMP(timer));
382 
383 	/*
384 	 * HPETs are a complete disaster. The compare register is
385 	 * based on a equal comparison and neither provides a less
386 	 * than or equal functionality (which would require to take
387 	 * the wraparound into account) nor a simple count down event
388 	 * mode. Further the write to the comparator register is
389 	 * delayed internally up to two HPET clock cycles in certain
390 	 * chipsets (ATI, ICH9,10). Some newer AMD chipsets have even
391 	 * longer delays. We worked around that by reading back the
392 	 * compare register, but that required another workaround for
393 	 * ICH9,10 chips where the first readout after write can
394 	 * return the old stale value. We already had a minimum
395 	 * programming delta of 5us enforced, but a NMI or SMI hitting
396 	 * between the counter readout and the comparator write can
397 	 * move us behind that point easily. Now instead of reading
398 	 * the compare register back several times, we make the ETIME
399 	 * decision based on the following: Return ETIME if the
400 	 * counter value after the write is less than HPET_MIN_CYCLES
401 	 * away from the event or if the counter is already ahead of
402 	 * the event. The minimum programming delta for the generic
403 	 * clockevents code is set to 1.5 * HPET_MIN_CYCLES.
404 	 */
405 	res = (s32)(cnt - hpet_readl(HPET_COUNTER));
406 
407 	return res < HPET_MIN_CYCLES ? -ETIME : 0;
408 }
409 
410 static void hpet_legacy_set_mode(enum clock_event_mode mode,
411 			struct clock_event_device *evt)
412 {
413 	hpet_set_mode(mode, evt, 0);
414 }
415 
416 static int hpet_legacy_next_event(unsigned long delta,
417 			struct clock_event_device *evt)
418 {
419 	return hpet_next_event(delta, evt, 0);
420 }
421 
422 /*
423  * HPET MSI Support
424  */
425 #ifdef CONFIG_PCI_MSI
426 
427 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
428 static struct hpet_dev	*hpet_devs;
429 
430 void hpet_msi_unmask(struct irq_data *data)
431 {
432 	struct hpet_dev *hdev = data->handler_data;
433 	unsigned int cfg;
434 
435 	/* unmask it */
436 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
437 	cfg |= HPET_TN_ENABLE | HPET_TN_FSB;
438 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
439 }
440 
441 void hpet_msi_mask(struct irq_data *data)
442 {
443 	struct hpet_dev *hdev = data->handler_data;
444 	unsigned int cfg;
445 
446 	/* mask it */
447 	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
448 	cfg &= ~(HPET_TN_ENABLE | HPET_TN_FSB);
449 	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
450 }
451 
452 void hpet_msi_write(struct hpet_dev *hdev, struct msi_msg *msg)
453 {
454 	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
455 	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
456 }
457 
458 void hpet_msi_read(struct hpet_dev *hdev, struct msi_msg *msg)
459 {
460 	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
461 	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
462 	msg->address_hi = 0;
463 }
464 
465 static void hpet_msi_set_mode(enum clock_event_mode mode,
466 				struct clock_event_device *evt)
467 {
468 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
469 	hpet_set_mode(mode, evt, hdev->num);
470 }
471 
472 static int hpet_msi_next_event(unsigned long delta,
473 				struct clock_event_device *evt)
474 {
475 	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
476 	return hpet_next_event(delta, evt, hdev->num);
477 }
478 
479 static int hpet_setup_msi_irq(unsigned int irq)
480 {
481 	if (arch_setup_hpet_msi(irq, hpet_blockid)) {
482 		destroy_irq(irq);
483 		return -EINVAL;
484 	}
485 	return 0;
486 }
487 
488 static int hpet_assign_irq(struct hpet_dev *dev)
489 {
490 	unsigned int irq;
491 
492 	irq = create_irq_nr(0, -1);
493 	if (!irq)
494 		return -EINVAL;
495 
496 	irq_set_handler_data(irq, dev);
497 
498 	if (hpet_setup_msi_irq(irq))
499 		return -EINVAL;
500 
501 	dev->irq = irq;
502 	return 0;
503 }
504 
505 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
506 {
507 	struct hpet_dev *dev = (struct hpet_dev *)data;
508 	struct clock_event_device *hevt = &dev->evt;
509 
510 	if (!hevt->event_handler) {
511 		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
512 				dev->num);
513 		return IRQ_HANDLED;
514 	}
515 
516 	hevt->event_handler(hevt);
517 	return IRQ_HANDLED;
518 }
519 
520 static int hpet_setup_irq(struct hpet_dev *dev)
521 {
522 
523 	if (request_irq(dev->irq, hpet_interrupt_handler,
524 			IRQF_TIMER | IRQF_DISABLED | IRQF_NOBALANCING,
525 			dev->name, dev))
526 		return -1;
527 
528 	disable_irq(dev->irq);
529 	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
530 	enable_irq(dev->irq);
531 
532 	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
533 			 dev->name, dev->irq);
534 
535 	return 0;
536 }
537 
538 /* This should be called in specific @cpu */
539 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
540 {
541 	struct clock_event_device *evt = &hdev->evt;
542 
543 	WARN_ON(cpu != smp_processor_id());
544 	if (!(hdev->flags & HPET_DEV_VALID))
545 		return;
546 
547 	if (hpet_setup_msi_irq(hdev->irq))
548 		return;
549 
550 	hdev->cpu = cpu;
551 	per_cpu(cpu_hpet_dev, cpu) = hdev;
552 	evt->name = hdev->name;
553 	hpet_setup_irq(hdev);
554 	evt->irq = hdev->irq;
555 
556 	evt->rating = 110;
557 	evt->features = CLOCK_EVT_FEAT_ONESHOT;
558 	if (hdev->flags & HPET_DEV_PERI_CAP)
559 		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
560 
561 	evt->set_mode = hpet_msi_set_mode;
562 	evt->set_next_event = hpet_msi_next_event;
563 	evt->cpumask = cpumask_of(hdev->cpu);
564 
565 	clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
566 					0x7FFFFFFF);
567 }
568 
569 #ifdef CONFIG_HPET
570 /* Reserve at least one timer for userspace (/dev/hpet) */
571 #define RESERVE_TIMERS 1
572 #else
573 #define RESERVE_TIMERS 0
574 #endif
575 
576 static void hpet_msi_capability_lookup(unsigned int start_timer)
577 {
578 	unsigned int id;
579 	unsigned int num_timers;
580 	unsigned int num_timers_used = 0;
581 	int i;
582 
583 	if (hpet_msi_disable)
584 		return;
585 
586 	if (boot_cpu_has(X86_FEATURE_ARAT))
587 		return;
588 	id = hpet_readl(HPET_ID);
589 
590 	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
591 	num_timers++; /* Value read out starts from 0 */
592 	hpet_print_config();
593 
594 	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
595 	if (!hpet_devs)
596 		return;
597 
598 	hpet_num_timers = num_timers;
599 
600 	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
601 		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
602 		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
603 
604 		/* Only consider HPET timer with MSI support */
605 		if (!(cfg & HPET_TN_FSB_CAP))
606 			continue;
607 
608 		hdev->flags = 0;
609 		if (cfg & HPET_TN_PERIODIC_CAP)
610 			hdev->flags |= HPET_DEV_PERI_CAP;
611 		hdev->num = i;
612 
613 		sprintf(hdev->name, "hpet%d", i);
614 		if (hpet_assign_irq(hdev))
615 			continue;
616 
617 		hdev->flags |= HPET_DEV_FSB_CAP;
618 		hdev->flags |= HPET_DEV_VALID;
619 		num_timers_used++;
620 		if (num_timers_used == num_possible_cpus())
621 			break;
622 	}
623 
624 	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
625 		num_timers, num_timers_used);
626 }
627 
628 #ifdef CONFIG_HPET
629 static void hpet_reserve_msi_timers(struct hpet_data *hd)
630 {
631 	int i;
632 
633 	if (!hpet_devs)
634 		return;
635 
636 	for (i = 0; i < hpet_num_timers; i++) {
637 		struct hpet_dev *hdev = &hpet_devs[i];
638 
639 		if (!(hdev->flags & HPET_DEV_VALID))
640 			continue;
641 
642 		hd->hd_irq[hdev->num] = hdev->irq;
643 		hpet_reserve_timer(hd, hdev->num);
644 	}
645 }
646 #endif
647 
648 static struct hpet_dev *hpet_get_unused_timer(void)
649 {
650 	int i;
651 
652 	if (!hpet_devs)
653 		return NULL;
654 
655 	for (i = 0; i < hpet_num_timers; i++) {
656 		struct hpet_dev *hdev = &hpet_devs[i];
657 
658 		if (!(hdev->flags & HPET_DEV_VALID))
659 			continue;
660 		if (test_and_set_bit(HPET_DEV_USED_BIT,
661 			(unsigned long *)&hdev->flags))
662 			continue;
663 		return hdev;
664 	}
665 	return NULL;
666 }
667 
668 struct hpet_work_struct {
669 	struct delayed_work work;
670 	struct completion complete;
671 };
672 
673 static void hpet_work(struct work_struct *w)
674 {
675 	struct hpet_dev *hdev;
676 	int cpu = smp_processor_id();
677 	struct hpet_work_struct *hpet_work;
678 
679 	hpet_work = container_of(w, struct hpet_work_struct, work.work);
680 
681 	hdev = hpet_get_unused_timer();
682 	if (hdev)
683 		init_one_hpet_msi_clockevent(hdev, cpu);
684 
685 	complete(&hpet_work->complete);
686 }
687 
688 static int hpet_cpuhp_notify(struct notifier_block *n,
689 		unsigned long action, void *hcpu)
690 {
691 	unsigned long cpu = (unsigned long)hcpu;
692 	struct hpet_work_struct work;
693 	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
694 
695 	switch (action & 0xf) {
696 	case CPU_ONLINE:
697 		INIT_DELAYED_WORK_ONSTACK(&work.work, hpet_work);
698 		init_completion(&work.complete);
699 		/* FIXME: add schedule_work_on() */
700 		schedule_delayed_work_on(cpu, &work.work, 0);
701 		wait_for_completion(&work.complete);
702 		destroy_timer_on_stack(&work.work.timer);
703 		break;
704 	case CPU_DEAD:
705 		if (hdev) {
706 			free_irq(hdev->irq, hdev);
707 			hdev->flags &= ~HPET_DEV_USED;
708 			per_cpu(cpu_hpet_dev, cpu) = NULL;
709 		}
710 		break;
711 	}
712 	return NOTIFY_OK;
713 }
714 #else
715 
716 static int hpet_setup_msi_irq(unsigned int irq)
717 {
718 	return 0;
719 }
720 static void hpet_msi_capability_lookup(unsigned int start_timer)
721 {
722 	return;
723 }
724 
725 #ifdef CONFIG_HPET
726 static void hpet_reserve_msi_timers(struct hpet_data *hd)
727 {
728 	return;
729 }
730 #endif
731 
732 static int hpet_cpuhp_notify(struct notifier_block *n,
733 		unsigned long action, void *hcpu)
734 {
735 	return NOTIFY_OK;
736 }
737 
738 #endif
739 
740 /*
741  * Clock source related code
742  */
743 static cycle_t read_hpet(struct clocksource *cs)
744 {
745 	return (cycle_t)hpet_readl(HPET_COUNTER);
746 }
747 
748 static struct clocksource clocksource_hpet = {
749 	.name		= "hpet",
750 	.rating		= 250,
751 	.read		= read_hpet,
752 	.mask		= HPET_MASK,
753 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
754 	.resume		= hpet_resume_counter,
755 #ifdef CONFIG_X86_64
756 	.archdata	= { .vclock_mode = VCLOCK_HPET },
757 #endif
758 };
759 
760 static int hpet_clocksource_register(void)
761 {
762 	u64 start, now;
763 	cycle_t t1;
764 
765 	/* Start the counter */
766 	hpet_restart_counter();
767 
768 	/* Verify whether hpet counter works */
769 	t1 = hpet_readl(HPET_COUNTER);
770 	rdtscll(start);
771 
772 	/*
773 	 * We don't know the TSC frequency yet, but waiting for
774 	 * 200000 TSC cycles is safe:
775 	 * 4 GHz == 50us
776 	 * 1 GHz == 200us
777 	 */
778 	do {
779 		rep_nop();
780 		rdtscll(now);
781 	} while ((now - start) < 200000UL);
782 
783 	if (t1 == hpet_readl(HPET_COUNTER)) {
784 		printk(KERN_WARNING
785 		       "HPET counter not counting. HPET disabled\n");
786 		return -ENODEV;
787 	}
788 
789 	clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
790 	return 0;
791 }
792 
793 static u32 *hpet_boot_cfg;
794 
795 /**
796  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
797  */
798 int __init hpet_enable(void)
799 {
800 	u32 hpet_period, cfg, id;
801 	u64 freq;
802 	unsigned int i, last;
803 
804 	if (!is_hpet_capable())
805 		return 0;
806 
807 	hpet_set_mapping();
808 
809 	/*
810 	 * Read the period and check for a sane value:
811 	 */
812 	hpet_period = hpet_readl(HPET_PERIOD);
813 
814 	/*
815 	 * AMD SB700 based systems with spread spectrum enabled use a
816 	 * SMM based HPET emulation to provide proper frequency
817 	 * setting. The SMM code is initialized with the first HPET
818 	 * register access and takes some time to complete. During
819 	 * this time the config register reads 0xffffffff. We check
820 	 * for max. 1000 loops whether the config register reads a non
821 	 * 0xffffffff value to make sure that HPET is up and running
822 	 * before we go further. A counting loop is safe, as the HPET
823 	 * access takes thousands of CPU cycles. On non SB700 based
824 	 * machines this check is only done once and has no side
825 	 * effects.
826 	 */
827 	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
828 		if (i == 1000) {
829 			printk(KERN_WARNING
830 			       "HPET config register value = 0xFFFFFFFF. "
831 			       "Disabling HPET\n");
832 			goto out_nohpet;
833 		}
834 	}
835 
836 	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
837 		goto out_nohpet;
838 
839 	/*
840 	 * The period is a femto seconds value. Convert it to a
841 	 * frequency.
842 	 */
843 	freq = FSEC_PER_SEC;
844 	do_div(freq, hpet_period);
845 	hpet_freq = freq;
846 
847 	/*
848 	 * Read the HPET ID register to retrieve the IRQ routing
849 	 * information and the number of channels
850 	 */
851 	id = hpet_readl(HPET_ID);
852 	hpet_print_config();
853 
854 	last = (id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT;
855 
856 #ifdef CONFIG_HPET_EMULATE_RTC
857 	/*
858 	 * The legacy routing mode needs at least two channels, tick timer
859 	 * and the rtc emulation channel.
860 	 */
861 	if (!last)
862 		goto out_nohpet;
863 #endif
864 
865 	cfg = hpet_readl(HPET_CFG);
866 	hpet_boot_cfg = kmalloc((last + 2) * sizeof(*hpet_boot_cfg),
867 				GFP_KERNEL);
868 	if (hpet_boot_cfg)
869 		*hpet_boot_cfg = cfg;
870 	else
871 		pr_warn("HPET initial state will not be saved\n");
872 	cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
873 	hpet_writel(cfg, HPET_CFG);
874 	if (cfg)
875 		pr_warn("HPET: Unrecognized bits %#x set in global cfg\n",
876 			cfg);
877 
878 	for (i = 0; i <= last; ++i) {
879 		cfg = hpet_readl(HPET_Tn_CFG(i));
880 		if (hpet_boot_cfg)
881 			hpet_boot_cfg[i + 1] = cfg;
882 		cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
883 		hpet_writel(cfg, HPET_Tn_CFG(i));
884 		cfg &= ~(HPET_TN_PERIODIC | HPET_TN_PERIODIC_CAP
885 			 | HPET_TN_64BIT_CAP | HPET_TN_32BIT | HPET_TN_ROUTE
886 			 | HPET_TN_FSB | HPET_TN_FSB_CAP);
887 		if (cfg)
888 			pr_warn("HPET: Unrecognized bits %#x set in cfg#%u\n",
889 				cfg, i);
890 	}
891 	hpet_print_config();
892 
893 	if (hpet_clocksource_register())
894 		goto out_nohpet;
895 
896 	if (id & HPET_ID_LEGSUP) {
897 		hpet_legacy_clockevent_register();
898 		return 1;
899 	}
900 	return 0;
901 
902 out_nohpet:
903 	hpet_clear_mapping();
904 	hpet_address = 0;
905 	return 0;
906 }
907 
908 /*
909  * Needs to be late, as the reserve_timer code calls kalloc !
910  *
911  * Not a problem on i386 as hpet_enable is called from late_time_init,
912  * but on x86_64 it is necessary !
913  */
914 static __init int hpet_late_init(void)
915 {
916 	int cpu;
917 
918 	if (boot_hpet_disable)
919 		return -ENODEV;
920 
921 	if (!hpet_address) {
922 		if (!force_hpet_address)
923 			return -ENODEV;
924 
925 		hpet_address = force_hpet_address;
926 		hpet_enable();
927 	}
928 
929 	if (!hpet_virt_address)
930 		return -ENODEV;
931 
932 	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
933 		hpet_msi_capability_lookup(2);
934 	else
935 		hpet_msi_capability_lookup(0);
936 
937 	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
938 	hpet_print_config();
939 
940 	if (hpet_msi_disable)
941 		return 0;
942 
943 	if (boot_cpu_has(X86_FEATURE_ARAT))
944 		return 0;
945 
946 	for_each_online_cpu(cpu) {
947 		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
948 	}
949 
950 	/* This notifier should be called after workqueue is ready */
951 	hotcpu_notifier(hpet_cpuhp_notify, -20);
952 
953 	return 0;
954 }
955 fs_initcall(hpet_late_init);
956 
957 void hpet_disable(void)
958 {
959 	if (is_hpet_capable() && hpet_virt_address) {
960 		unsigned int cfg = hpet_readl(HPET_CFG), id, last;
961 
962 		if (hpet_boot_cfg)
963 			cfg = *hpet_boot_cfg;
964 		else if (hpet_legacy_int_enabled) {
965 			cfg &= ~HPET_CFG_LEGACY;
966 			hpet_legacy_int_enabled = 0;
967 		}
968 		cfg &= ~HPET_CFG_ENABLE;
969 		hpet_writel(cfg, HPET_CFG);
970 
971 		if (!hpet_boot_cfg)
972 			return;
973 
974 		id = hpet_readl(HPET_ID);
975 		last = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
976 
977 		for (id = 0; id <= last; ++id)
978 			hpet_writel(hpet_boot_cfg[id + 1], HPET_Tn_CFG(id));
979 
980 		if (*hpet_boot_cfg & HPET_CFG_ENABLE)
981 			hpet_writel(*hpet_boot_cfg, HPET_CFG);
982 	}
983 }
984 
985 #ifdef CONFIG_HPET_EMULATE_RTC
986 
987 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
988  * is enabled, we support RTC interrupt functionality in software.
989  * RTC has 3 kinds of interrupts:
990  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
991  *    is updated
992  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
993  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
994  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
995  * (1) and (2) above are implemented using polling at a frequency of
996  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
997  * overhead. (DEFAULT_RTC_INT_FREQ)
998  * For (3), we use interrupts at 64Hz or user specified periodic
999  * frequency, whichever is higher.
1000  */
1001 #include <linux/mc146818rtc.h>
1002 #include <linux/rtc.h>
1003 #include <asm/rtc.h>
1004 
1005 #define DEFAULT_RTC_INT_FREQ	64
1006 #define DEFAULT_RTC_SHIFT	6
1007 #define RTC_NUM_INTS		1
1008 
1009 static unsigned long hpet_rtc_flags;
1010 static int hpet_prev_update_sec;
1011 static struct rtc_time hpet_alarm_time;
1012 static unsigned long hpet_pie_count;
1013 static u32 hpet_t1_cmp;
1014 static u32 hpet_default_delta;
1015 static u32 hpet_pie_delta;
1016 static unsigned long hpet_pie_limit;
1017 
1018 static rtc_irq_handler irq_handler;
1019 
1020 /*
1021  * Check that the hpet counter c1 is ahead of the c2
1022  */
1023 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1024 {
1025 	return (s32)(c2 - c1) < 0;
1026 }
1027 
1028 /*
1029  * Registers a IRQ handler.
1030  */
1031 int hpet_register_irq_handler(rtc_irq_handler handler)
1032 {
1033 	if (!is_hpet_enabled())
1034 		return -ENODEV;
1035 	if (irq_handler)
1036 		return -EBUSY;
1037 
1038 	irq_handler = handler;
1039 
1040 	return 0;
1041 }
1042 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1043 
1044 /*
1045  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1046  * and does cleanup.
1047  */
1048 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1049 {
1050 	if (!is_hpet_enabled())
1051 		return;
1052 
1053 	irq_handler = NULL;
1054 	hpet_rtc_flags = 0;
1055 }
1056 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1057 
1058 /*
1059  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1060  * is not supported by all HPET implementations for timer 1.
1061  *
1062  * hpet_rtc_timer_init() is called when the rtc is initialized.
1063  */
1064 int hpet_rtc_timer_init(void)
1065 {
1066 	unsigned int cfg, cnt, delta;
1067 	unsigned long flags;
1068 
1069 	if (!is_hpet_enabled())
1070 		return 0;
1071 
1072 	if (!hpet_default_delta) {
1073 		uint64_t clc;
1074 
1075 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1076 		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1077 		hpet_default_delta = clc;
1078 	}
1079 
1080 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1081 		delta = hpet_default_delta;
1082 	else
1083 		delta = hpet_pie_delta;
1084 
1085 	local_irq_save(flags);
1086 
1087 	cnt = delta + hpet_readl(HPET_COUNTER);
1088 	hpet_writel(cnt, HPET_T1_CMP);
1089 	hpet_t1_cmp = cnt;
1090 
1091 	cfg = hpet_readl(HPET_T1_CFG);
1092 	cfg &= ~HPET_TN_PERIODIC;
1093 	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1094 	hpet_writel(cfg, HPET_T1_CFG);
1095 
1096 	local_irq_restore(flags);
1097 
1098 	return 1;
1099 }
1100 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1101 
1102 static void hpet_disable_rtc_channel(void)
1103 {
1104 	unsigned long cfg;
1105 	cfg = hpet_readl(HPET_T1_CFG);
1106 	cfg &= ~HPET_TN_ENABLE;
1107 	hpet_writel(cfg, HPET_T1_CFG);
1108 }
1109 
1110 /*
1111  * The functions below are called from rtc driver.
1112  * Return 0 if HPET is not being used.
1113  * Otherwise do the necessary changes and return 1.
1114  */
1115 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1116 {
1117 	if (!is_hpet_enabled())
1118 		return 0;
1119 
1120 	hpet_rtc_flags &= ~bit_mask;
1121 	if (unlikely(!hpet_rtc_flags))
1122 		hpet_disable_rtc_channel();
1123 
1124 	return 1;
1125 }
1126 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1127 
1128 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1129 {
1130 	unsigned long oldbits = hpet_rtc_flags;
1131 
1132 	if (!is_hpet_enabled())
1133 		return 0;
1134 
1135 	hpet_rtc_flags |= bit_mask;
1136 
1137 	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1138 		hpet_prev_update_sec = -1;
1139 
1140 	if (!oldbits)
1141 		hpet_rtc_timer_init();
1142 
1143 	return 1;
1144 }
1145 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1146 
1147 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1148 			unsigned char sec)
1149 {
1150 	if (!is_hpet_enabled())
1151 		return 0;
1152 
1153 	hpet_alarm_time.tm_hour = hrs;
1154 	hpet_alarm_time.tm_min = min;
1155 	hpet_alarm_time.tm_sec = sec;
1156 
1157 	return 1;
1158 }
1159 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1160 
1161 int hpet_set_periodic_freq(unsigned long freq)
1162 {
1163 	uint64_t clc;
1164 
1165 	if (!is_hpet_enabled())
1166 		return 0;
1167 
1168 	if (freq <= DEFAULT_RTC_INT_FREQ)
1169 		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1170 	else {
1171 		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1172 		do_div(clc, freq);
1173 		clc >>= hpet_clockevent.shift;
1174 		hpet_pie_delta = clc;
1175 		hpet_pie_limit = 0;
1176 	}
1177 	return 1;
1178 }
1179 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1180 
1181 int hpet_rtc_dropped_irq(void)
1182 {
1183 	return is_hpet_enabled();
1184 }
1185 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1186 
1187 static void hpet_rtc_timer_reinit(void)
1188 {
1189 	unsigned int delta;
1190 	int lost_ints = -1;
1191 
1192 	if (unlikely(!hpet_rtc_flags))
1193 		hpet_disable_rtc_channel();
1194 
1195 	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1196 		delta = hpet_default_delta;
1197 	else
1198 		delta = hpet_pie_delta;
1199 
1200 	/*
1201 	 * Increment the comparator value until we are ahead of the
1202 	 * current count.
1203 	 */
1204 	do {
1205 		hpet_t1_cmp += delta;
1206 		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1207 		lost_ints++;
1208 	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1209 
1210 	if (lost_ints) {
1211 		if (hpet_rtc_flags & RTC_PIE)
1212 			hpet_pie_count += lost_ints;
1213 		if (printk_ratelimit())
1214 			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1215 				lost_ints);
1216 	}
1217 }
1218 
1219 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1220 {
1221 	struct rtc_time curr_time;
1222 	unsigned long rtc_int_flag = 0;
1223 
1224 	hpet_rtc_timer_reinit();
1225 	memset(&curr_time, 0, sizeof(struct rtc_time));
1226 
1227 	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1228 		get_rtc_time(&curr_time);
1229 
1230 	if (hpet_rtc_flags & RTC_UIE &&
1231 	    curr_time.tm_sec != hpet_prev_update_sec) {
1232 		if (hpet_prev_update_sec >= 0)
1233 			rtc_int_flag = RTC_UF;
1234 		hpet_prev_update_sec = curr_time.tm_sec;
1235 	}
1236 
1237 	if (hpet_rtc_flags & RTC_PIE &&
1238 	    ++hpet_pie_count >= hpet_pie_limit) {
1239 		rtc_int_flag |= RTC_PF;
1240 		hpet_pie_count = 0;
1241 	}
1242 
1243 	if (hpet_rtc_flags & RTC_AIE &&
1244 	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1245 	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1246 	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1247 			rtc_int_flag |= RTC_AF;
1248 
1249 	if (rtc_int_flag) {
1250 		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1251 		if (irq_handler)
1252 			irq_handler(rtc_int_flag, dev_id);
1253 	}
1254 	return IRQ_HANDLED;
1255 }
1256 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1257 #endif
1258