hpet.c revision b4a5e8a1deca7e61ebaffb37344766b0f0e9f327
1#include <linux/clocksource.h>
2#include <linux/clockchips.h>
3#include <linux/interrupt.h>
4#include <linux/sysdev.h>
5#include <linux/delay.h>
6#include <linux/errno.h>
7#include <linux/hpet.h>
8#include <linux/init.h>
9#include <linux/cpu.h>
10#include <linux/pm.h>
11#include <linux/io.h>
12
13#include <asm/fixmap.h>
14#include <asm/i8253.h>
15#include <asm/hpet.h>
16
17#define HPET_MASK			CLOCKSOURCE_MASK(32)
18#define HPET_SHIFT			22
19
20/* FSEC = 10^-15
21   NSEC = 10^-9 */
22#define FSEC_PER_NSEC			1000000L
23
24#define HPET_DEV_USED_BIT		2
25#define HPET_DEV_USED			(1 << HPET_DEV_USED_BIT)
26#define HPET_DEV_VALID			0x8
27#define HPET_DEV_FSB_CAP		0x1000
28#define HPET_DEV_PERI_CAP		0x2000
29
30#define EVT_TO_HPET_DEV(evt) container_of(evt, struct hpet_dev, evt)
31
32/*
33 * HPET address is set in acpi/boot.c, when an ACPI entry exists
34 */
35unsigned long				hpet_address;
36u8					hpet_blockid; /* OS timer block num */
37u8					hpet_msi_disable;
38
39#ifdef CONFIG_PCI_MSI
40static unsigned long			hpet_num_timers;
41#endif
42static void __iomem			*hpet_virt_address;
43
44struct hpet_dev {
45	struct clock_event_device	evt;
46	unsigned int			num;
47	int				cpu;
48	unsigned int			irq;
49	unsigned int			flags;
50	char				name[10];
51};
52
53inline unsigned int hpet_readl(unsigned int a)
54{
55	return readl(hpet_virt_address + a);
56}
57
58static inline void hpet_writel(unsigned int d, unsigned int a)
59{
60	writel(d, hpet_virt_address + a);
61}
62
63#ifdef CONFIG_X86_64
64#include <asm/pgtable.h>
65#endif
66
67static inline void hpet_set_mapping(void)
68{
69	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
70#ifdef CONFIG_X86_64
71	__set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
72#endif
73}
74
75static inline void hpet_clear_mapping(void)
76{
77	iounmap(hpet_virt_address);
78	hpet_virt_address = NULL;
79}
80
81/*
82 * HPET command line enable / disable
83 */
84static int boot_hpet_disable;
85int hpet_force_user;
86static int hpet_verbose;
87
88static int __init hpet_setup(char *str)
89{
90	if (str) {
91		if (!strncmp("disable", str, 7))
92			boot_hpet_disable = 1;
93		if (!strncmp("force", str, 5))
94			hpet_force_user = 1;
95		if (!strncmp("verbose", str, 7))
96			hpet_verbose = 1;
97	}
98	return 1;
99}
100__setup("hpet=", hpet_setup);
101
102static int __init disable_hpet(char *str)
103{
104	boot_hpet_disable = 1;
105	return 1;
106}
107__setup("nohpet", disable_hpet);
108
109static inline int is_hpet_capable(void)
110{
111	return !boot_hpet_disable && hpet_address;
112}
113
114/*
115 * HPET timer interrupt enable / disable
116 */
117static int hpet_legacy_int_enabled;
118
119/**
120 * is_hpet_enabled - check whether the hpet timer interrupt is enabled
121 */
122int is_hpet_enabled(void)
123{
124	return is_hpet_capable() && hpet_legacy_int_enabled;
125}
126EXPORT_SYMBOL_GPL(is_hpet_enabled);
127
128static void _hpet_print_config(const char *function, int line)
129{
130	u32 i, timers, l, h;
131	printk(KERN_INFO "hpet: %s(%d):\n", function, line);
132	l = hpet_readl(HPET_ID);
133	h = hpet_readl(HPET_PERIOD);
134	timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
135	printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
136	l = hpet_readl(HPET_CFG);
137	h = hpet_readl(HPET_STATUS);
138	printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
139	l = hpet_readl(HPET_COUNTER);
140	h = hpet_readl(HPET_COUNTER+4);
141	printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
142
143	for (i = 0; i < timers; i++) {
144		l = hpet_readl(HPET_Tn_CFG(i));
145		h = hpet_readl(HPET_Tn_CFG(i)+4);
146		printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
147		       i, l, h);
148		l = hpet_readl(HPET_Tn_CMP(i));
149		h = hpet_readl(HPET_Tn_CMP(i)+4);
150		printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
151		       i, l, h);
152		l = hpet_readl(HPET_Tn_ROUTE(i));
153		h = hpet_readl(HPET_Tn_ROUTE(i)+4);
154		printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
155		       i, l, h);
156	}
157}
158
159#define hpet_print_config()					\
160do {								\
161	if (hpet_verbose)					\
162		_hpet_print_config(__FUNCTION__, __LINE__);	\
163} while (0)
164
165/*
166 * When the hpet driver (/dev/hpet) is enabled, we need to reserve
167 * timer 0 and timer 1 in case of RTC emulation.
168 */
169#ifdef CONFIG_HPET
170
171static void hpet_reserve_msi_timers(struct hpet_data *hd);
172
173static void hpet_reserve_platform_timers(unsigned int id)
174{
175	struct hpet __iomem *hpet = hpet_virt_address;
176	struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
177	unsigned int nrtimers, i;
178	struct hpet_data hd;
179
180	nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
181
182	memset(&hd, 0, sizeof(hd));
183	hd.hd_phys_address	= hpet_address;
184	hd.hd_address		= hpet;
185	hd.hd_nirqs		= nrtimers;
186	hpet_reserve_timer(&hd, 0);
187
188#ifdef CONFIG_HPET_EMULATE_RTC
189	hpet_reserve_timer(&hd, 1);
190#endif
191
192	/*
193	 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
194	 * is wrong for i8259!) not the output IRQ.  Many BIOS writers
195	 * don't bother configuring *any* comparator interrupts.
196	 */
197	hd.hd_irq[0] = HPET_LEGACY_8254;
198	hd.hd_irq[1] = HPET_LEGACY_RTC;
199
200	for (i = 2; i < nrtimers; timer++, i++) {
201		hd.hd_irq[i] = (readl(&timer->hpet_config) &
202			Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
203	}
204
205	hpet_reserve_msi_timers(&hd);
206
207	hpet_alloc(&hd);
208
209}
210#else
211static void hpet_reserve_platform_timers(unsigned int id) { }
212#endif
213
214/*
215 * Common hpet info
216 */
217static unsigned long hpet_period;
218
219static void hpet_legacy_set_mode(enum clock_event_mode mode,
220			  struct clock_event_device *evt);
221static int hpet_legacy_next_event(unsigned long delta,
222			   struct clock_event_device *evt);
223
224/*
225 * The hpet clock event device
226 */
227static struct clock_event_device hpet_clockevent = {
228	.name		= "hpet",
229	.features	= CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
230	.set_mode	= hpet_legacy_set_mode,
231	.set_next_event = hpet_legacy_next_event,
232	.shift		= 32,
233	.irq		= 0,
234	.rating		= 50,
235};
236
237static void hpet_stop_counter(void)
238{
239	unsigned long cfg = hpet_readl(HPET_CFG);
240	cfg &= ~HPET_CFG_ENABLE;
241	hpet_writel(cfg, HPET_CFG);
242}
243
244static void hpet_reset_counter(void)
245{
246	hpet_writel(0, HPET_COUNTER);
247	hpet_writel(0, HPET_COUNTER + 4);
248}
249
250static void hpet_start_counter(void)
251{
252	unsigned int cfg = hpet_readl(HPET_CFG);
253	cfg |= HPET_CFG_ENABLE;
254	hpet_writel(cfg, HPET_CFG);
255}
256
257static void hpet_restart_counter(void)
258{
259	hpet_stop_counter();
260	hpet_reset_counter();
261	hpet_start_counter();
262}
263
264static void hpet_resume_device(void)
265{
266	force_hpet_resume();
267}
268
269static void hpet_resume_counter(struct clocksource *cs)
270{
271	hpet_resume_device();
272	hpet_restart_counter();
273}
274
275static void hpet_enable_legacy_int(void)
276{
277	unsigned int cfg = hpet_readl(HPET_CFG);
278
279	cfg |= HPET_CFG_LEGACY;
280	hpet_writel(cfg, HPET_CFG);
281	hpet_legacy_int_enabled = 1;
282}
283
284static void hpet_legacy_clockevent_register(void)
285{
286	/* Start HPET legacy interrupts */
287	hpet_enable_legacy_int();
288
289	/*
290	 * The mult factor is defined as (include/linux/clockchips.h)
291	 *  mult/2^shift = cyc/ns (in contrast to ns/cyc in clocksource.h)
292	 * hpet_period is in units of femtoseconds (per cycle), so
293	 *  mult/2^shift = cyc/ns = 10^6/hpet_period
294	 *  mult = (10^6 * 2^shift)/hpet_period
295	 *  mult = (FSEC_PER_NSEC << hpet_clockevent.shift)/hpet_period
296	 */
297	hpet_clockevent.mult = div_sc((unsigned long) FSEC_PER_NSEC,
298				      hpet_period, hpet_clockevent.shift);
299	/* Calculate the min / max delta */
300	hpet_clockevent.max_delta_ns = clockevent_delta2ns(0x7FFFFFFF,
301							   &hpet_clockevent);
302	/* 5 usec minimum reprogramming delta. */
303	hpet_clockevent.min_delta_ns = 5000;
304
305	/*
306	 * Start hpet with the boot cpu mask and make it
307	 * global after the IO_APIC has been initialized.
308	 */
309	hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
310	clockevents_register_device(&hpet_clockevent);
311	global_clock_event = &hpet_clockevent;
312	printk(KERN_DEBUG "hpet clockevent registered\n");
313}
314
315static int hpet_setup_msi_irq(unsigned int irq);
316
317static void hpet_set_mode(enum clock_event_mode mode,
318			  struct clock_event_device *evt, int timer)
319{
320	unsigned int cfg, cmp, now;
321	uint64_t delta;
322
323	switch (mode) {
324	case CLOCK_EVT_MODE_PERIODIC:
325		hpet_stop_counter();
326		delta = ((uint64_t)(NSEC_PER_SEC/HZ)) * evt->mult;
327		delta >>= evt->shift;
328		now = hpet_readl(HPET_COUNTER);
329		cmp = now + (unsigned int) delta;
330		cfg = hpet_readl(HPET_Tn_CFG(timer));
331		/* Make sure we use edge triggered interrupts */
332		cfg &= ~HPET_TN_LEVEL;
333		cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC |
334		       HPET_TN_SETVAL | HPET_TN_32BIT;
335		hpet_writel(cfg, HPET_Tn_CFG(timer));
336		hpet_writel(cmp, HPET_Tn_CMP(timer));
337		udelay(1);
338		/*
339		 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
340		 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
341		 * bit is automatically cleared after the first write.
342		 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
343		 * Publication # 24674)
344		 */
345		hpet_writel((unsigned int) delta, HPET_Tn_CMP(timer));
346		hpet_start_counter();
347		hpet_print_config();
348		break;
349
350	case CLOCK_EVT_MODE_ONESHOT:
351		cfg = hpet_readl(HPET_Tn_CFG(timer));
352		cfg &= ~HPET_TN_PERIODIC;
353		cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
354		hpet_writel(cfg, HPET_Tn_CFG(timer));
355		break;
356
357	case CLOCK_EVT_MODE_UNUSED:
358	case CLOCK_EVT_MODE_SHUTDOWN:
359		cfg = hpet_readl(HPET_Tn_CFG(timer));
360		cfg &= ~HPET_TN_ENABLE;
361		hpet_writel(cfg, HPET_Tn_CFG(timer));
362		break;
363
364	case CLOCK_EVT_MODE_RESUME:
365		if (timer == 0) {
366			hpet_enable_legacy_int();
367		} else {
368			struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
369			hpet_setup_msi_irq(hdev->irq);
370			disable_irq(hdev->irq);
371			irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
372			enable_irq(hdev->irq);
373		}
374		hpet_print_config();
375		break;
376	}
377}
378
379static int hpet_next_event(unsigned long delta,
380			   struct clock_event_device *evt, int timer)
381{
382	u32 cnt;
383
384	cnt = hpet_readl(HPET_COUNTER);
385	cnt += (u32) delta;
386	hpet_writel(cnt, HPET_Tn_CMP(timer));
387
388	/*
389	 * We need to read back the CMP register on certain HPET
390	 * implementations (ATI chipsets) which seem to delay the
391	 * transfer of the compare register into the internal compare
392	 * logic. With small deltas this might actually be too late as
393	 * the counter could already be higher than the compare value
394	 * at that point and we would wait for the next hpet interrupt
395	 * forever. We found out that reading the CMP register back
396	 * forces the transfer so we can rely on the comparison with
397	 * the counter register below. If the read back from the
398	 * compare register does not match the value we programmed
399	 * then we might have a real hardware problem. We can not do
400	 * much about it here, but at least alert the user/admin with
401	 * a prominent warning.
402	 * An erratum on some chipsets (ICH9,..), results in comparator read
403	 * immediately following a write returning old value. Workaround
404	 * for this is to read this value second time, when first
405	 * read returns old value.
406	 */
407	if (unlikely((u32)hpet_readl(HPET_Tn_CMP(timer)) != cnt)) {
408		WARN_ONCE(hpet_readl(HPET_Tn_CMP(timer)) != cnt,
409		  KERN_WARNING "hpet: compare register read back failed.\n");
410	}
411
412	return (s32)(hpet_readl(HPET_COUNTER) - cnt) >= 0 ? -ETIME : 0;
413}
414
415static void hpet_legacy_set_mode(enum clock_event_mode mode,
416			struct clock_event_device *evt)
417{
418	hpet_set_mode(mode, evt, 0);
419}
420
421static int hpet_legacy_next_event(unsigned long delta,
422			struct clock_event_device *evt)
423{
424	return hpet_next_event(delta, evt, 0);
425}
426
427/*
428 * HPET MSI Support
429 */
430#ifdef CONFIG_PCI_MSI
431
432static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
433static struct hpet_dev	*hpet_devs;
434
435void hpet_msi_unmask(unsigned int irq)
436{
437	struct hpet_dev *hdev = get_irq_data(irq);
438	unsigned int cfg;
439
440	/* unmask it */
441	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
442	cfg |= HPET_TN_FSB;
443	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
444}
445
446void hpet_msi_mask(unsigned int irq)
447{
448	unsigned int cfg;
449	struct hpet_dev *hdev = get_irq_data(irq);
450
451	/* mask it */
452	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
453	cfg &= ~HPET_TN_FSB;
454	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
455}
456
457void hpet_msi_write(unsigned int irq, struct msi_msg *msg)
458{
459	struct hpet_dev *hdev = get_irq_data(irq);
460
461	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
462	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
463}
464
465void hpet_msi_read(unsigned int irq, struct msi_msg *msg)
466{
467	struct hpet_dev *hdev = get_irq_data(irq);
468
469	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
470	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
471	msg->address_hi = 0;
472}
473
474static void hpet_msi_set_mode(enum clock_event_mode mode,
475				struct clock_event_device *evt)
476{
477	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
478	hpet_set_mode(mode, evt, hdev->num);
479}
480
481static int hpet_msi_next_event(unsigned long delta,
482				struct clock_event_device *evt)
483{
484	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
485	return hpet_next_event(delta, evt, hdev->num);
486}
487
488static int hpet_setup_msi_irq(unsigned int irq)
489{
490	if (arch_setup_hpet_msi(irq, hpet_blockid)) {
491		destroy_irq(irq);
492		return -EINVAL;
493	}
494	return 0;
495}
496
497static int hpet_assign_irq(struct hpet_dev *dev)
498{
499	unsigned int irq;
500
501	irq = create_irq();
502	if (!irq)
503		return -EINVAL;
504
505	set_irq_data(irq, dev);
506
507	if (hpet_setup_msi_irq(irq))
508		return -EINVAL;
509
510	dev->irq = irq;
511	return 0;
512}
513
514static irqreturn_t hpet_interrupt_handler(int irq, void *data)
515{
516	struct hpet_dev *dev = (struct hpet_dev *)data;
517	struct clock_event_device *hevt = &dev->evt;
518
519	if (!hevt->event_handler) {
520		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
521				dev->num);
522		return IRQ_HANDLED;
523	}
524
525	hevt->event_handler(hevt);
526	return IRQ_HANDLED;
527}
528
529static int hpet_setup_irq(struct hpet_dev *dev)
530{
531
532	if (request_irq(dev->irq, hpet_interrupt_handler,
533			IRQF_TIMER | IRQF_DISABLED | IRQF_NOBALANCING,
534			dev->name, dev))
535		return -1;
536
537	disable_irq(dev->irq);
538	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
539	enable_irq(dev->irq);
540
541	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
542			 dev->name, dev->irq);
543
544	return 0;
545}
546
547/* This should be called in specific @cpu */
548static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
549{
550	struct clock_event_device *evt = &hdev->evt;
551	uint64_t hpet_freq;
552
553	WARN_ON(cpu != smp_processor_id());
554	if (!(hdev->flags & HPET_DEV_VALID))
555		return;
556
557	if (hpet_setup_msi_irq(hdev->irq))
558		return;
559
560	hdev->cpu = cpu;
561	per_cpu(cpu_hpet_dev, cpu) = hdev;
562	evt->name = hdev->name;
563	hpet_setup_irq(hdev);
564	evt->irq = hdev->irq;
565
566	evt->rating = 110;
567	evt->features = CLOCK_EVT_FEAT_ONESHOT;
568	if (hdev->flags & HPET_DEV_PERI_CAP)
569		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
570
571	evt->set_mode = hpet_msi_set_mode;
572	evt->set_next_event = hpet_msi_next_event;
573	evt->shift = 32;
574
575	/*
576	 * The period is a femto seconds value. We need to calculate the
577	 * scaled math multiplication factor for nanosecond to hpet tick
578	 * conversion.
579	 */
580	hpet_freq = 1000000000000000ULL;
581	do_div(hpet_freq, hpet_period);
582	evt->mult = div_sc((unsigned long) hpet_freq,
583				      NSEC_PER_SEC, evt->shift);
584	/* Calculate the max delta */
585	evt->max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, evt);
586	/* 5 usec minimum reprogramming delta. */
587	evt->min_delta_ns = 5000;
588
589	evt->cpumask = cpumask_of(hdev->cpu);
590	clockevents_register_device(evt);
591}
592
593#ifdef CONFIG_HPET
594/* Reserve at least one timer for userspace (/dev/hpet) */
595#define RESERVE_TIMERS 1
596#else
597#define RESERVE_TIMERS 0
598#endif
599
600static void hpet_msi_capability_lookup(unsigned int start_timer)
601{
602	unsigned int id;
603	unsigned int num_timers;
604	unsigned int num_timers_used = 0;
605	int i;
606
607	if (hpet_msi_disable)
608		return;
609
610	if (boot_cpu_has(X86_FEATURE_ARAT))
611		return;
612	id = hpet_readl(HPET_ID);
613
614	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
615	num_timers++; /* Value read out starts from 0 */
616	hpet_print_config();
617
618	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
619	if (!hpet_devs)
620		return;
621
622	hpet_num_timers = num_timers;
623
624	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
625		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
626		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
627
628		/* Only consider HPET timer with MSI support */
629		if (!(cfg & HPET_TN_FSB_CAP))
630			continue;
631
632		hdev->flags = 0;
633		if (cfg & HPET_TN_PERIODIC_CAP)
634			hdev->flags |= HPET_DEV_PERI_CAP;
635		hdev->num = i;
636
637		sprintf(hdev->name, "hpet%d", i);
638		if (hpet_assign_irq(hdev))
639			continue;
640
641		hdev->flags |= HPET_DEV_FSB_CAP;
642		hdev->flags |= HPET_DEV_VALID;
643		num_timers_used++;
644		if (num_timers_used == num_possible_cpus())
645			break;
646	}
647
648	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
649		num_timers, num_timers_used);
650}
651
652#ifdef CONFIG_HPET
653static void hpet_reserve_msi_timers(struct hpet_data *hd)
654{
655	int i;
656
657	if (!hpet_devs)
658		return;
659
660	for (i = 0; i < hpet_num_timers; i++) {
661		struct hpet_dev *hdev = &hpet_devs[i];
662
663		if (!(hdev->flags & HPET_DEV_VALID))
664			continue;
665
666		hd->hd_irq[hdev->num] = hdev->irq;
667		hpet_reserve_timer(hd, hdev->num);
668	}
669}
670#endif
671
672static struct hpet_dev *hpet_get_unused_timer(void)
673{
674	int i;
675
676	if (!hpet_devs)
677		return NULL;
678
679	for (i = 0; i < hpet_num_timers; i++) {
680		struct hpet_dev *hdev = &hpet_devs[i];
681
682		if (!(hdev->flags & HPET_DEV_VALID))
683			continue;
684		if (test_and_set_bit(HPET_DEV_USED_BIT,
685			(unsigned long *)&hdev->flags))
686			continue;
687		return hdev;
688	}
689	return NULL;
690}
691
692struct hpet_work_struct {
693	struct delayed_work work;
694	struct completion complete;
695};
696
697static void hpet_work(struct work_struct *w)
698{
699	struct hpet_dev *hdev;
700	int cpu = smp_processor_id();
701	struct hpet_work_struct *hpet_work;
702
703	hpet_work = container_of(w, struct hpet_work_struct, work.work);
704
705	hdev = hpet_get_unused_timer();
706	if (hdev)
707		init_one_hpet_msi_clockevent(hdev, cpu);
708
709	complete(&hpet_work->complete);
710}
711
712static int hpet_cpuhp_notify(struct notifier_block *n,
713		unsigned long action, void *hcpu)
714{
715	unsigned long cpu = (unsigned long)hcpu;
716	struct hpet_work_struct work;
717	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
718
719	switch (action & 0xf) {
720	case CPU_ONLINE:
721		INIT_DELAYED_WORK_ON_STACK(&work.work, hpet_work);
722		init_completion(&work.complete);
723		/* FIXME: add schedule_work_on() */
724		schedule_delayed_work_on(cpu, &work.work, 0);
725		wait_for_completion(&work.complete);
726		destroy_timer_on_stack(&work.work.timer);
727		break;
728	case CPU_DEAD:
729		if (hdev) {
730			free_irq(hdev->irq, hdev);
731			hdev->flags &= ~HPET_DEV_USED;
732			per_cpu(cpu_hpet_dev, cpu) = NULL;
733		}
734		break;
735	}
736	return NOTIFY_OK;
737}
738#else
739
740static int hpet_setup_msi_irq(unsigned int irq)
741{
742	return 0;
743}
744static void hpet_msi_capability_lookup(unsigned int start_timer)
745{
746	return;
747}
748
749#ifdef CONFIG_HPET
750static void hpet_reserve_msi_timers(struct hpet_data *hd)
751{
752	return;
753}
754#endif
755
756static int hpet_cpuhp_notify(struct notifier_block *n,
757		unsigned long action, void *hcpu)
758{
759	return NOTIFY_OK;
760}
761
762#endif
763
764/*
765 * Clock source related code
766 */
767static cycle_t read_hpet(struct clocksource *cs)
768{
769	return (cycle_t)hpet_readl(HPET_COUNTER);
770}
771
772#ifdef CONFIG_X86_64
773static cycle_t __vsyscall_fn vread_hpet(void)
774{
775	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + 0xf0);
776}
777#endif
778
779static struct clocksource clocksource_hpet = {
780	.name		= "hpet",
781	.rating		= 250,
782	.read		= read_hpet,
783	.mask		= HPET_MASK,
784	.shift		= HPET_SHIFT,
785	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
786	.resume		= hpet_resume_counter,
787#ifdef CONFIG_X86_64
788	.vread		= vread_hpet,
789#endif
790};
791
792static int hpet_clocksource_register(void)
793{
794	u64 start, now;
795	cycle_t t1;
796
797	/* Start the counter */
798	hpet_restart_counter();
799
800	/* Verify whether hpet counter works */
801	t1 = hpet_readl(HPET_COUNTER);
802	rdtscll(start);
803
804	/*
805	 * We don't know the TSC frequency yet, but waiting for
806	 * 200000 TSC cycles is safe:
807	 * 4 GHz == 50us
808	 * 1 GHz == 200us
809	 */
810	do {
811		rep_nop();
812		rdtscll(now);
813	} while ((now - start) < 200000UL);
814
815	if (t1 == hpet_readl(HPET_COUNTER)) {
816		printk(KERN_WARNING
817		       "HPET counter not counting. HPET disabled\n");
818		return -ENODEV;
819	}
820
821	/*
822	 * The definition of mult is (include/linux/clocksource.h)
823	 * mult/2^shift = ns/cyc and hpet_period is in units of fsec/cyc
824	 * so we first need to convert hpet_period to ns/cyc units:
825	 *  mult/2^shift = ns/cyc = hpet_period/10^6
826	 *  mult = (hpet_period * 2^shift)/10^6
827	 *  mult = (hpet_period << shift)/FSEC_PER_NSEC
828	 */
829	clocksource_hpet.mult = div_sc(hpet_period, FSEC_PER_NSEC, HPET_SHIFT);
830
831	clocksource_register(&clocksource_hpet);
832
833	return 0;
834}
835
836/**
837 * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
838 */
839int __init hpet_enable(void)
840{
841	unsigned int id;
842	int i;
843
844	if (!is_hpet_capable())
845		return 0;
846
847	hpet_set_mapping();
848
849	/*
850	 * Read the period and check for a sane value:
851	 */
852	hpet_period = hpet_readl(HPET_PERIOD);
853
854	/*
855	 * AMD SB700 based systems with spread spectrum enabled use a
856	 * SMM based HPET emulation to provide proper frequency
857	 * setting. The SMM code is initialized with the first HPET
858	 * register access and takes some time to complete. During
859	 * this time the config register reads 0xffffffff. We check
860	 * for max. 1000 loops whether the config register reads a non
861	 * 0xffffffff value to make sure that HPET is up and running
862	 * before we go further. A counting loop is safe, as the HPET
863	 * access takes thousands of CPU cycles. On non SB700 based
864	 * machines this check is only done once and has no side
865	 * effects.
866	 */
867	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
868		if (i == 1000) {
869			printk(KERN_WARNING
870			       "HPET config register value = 0xFFFFFFFF. "
871			       "Disabling HPET\n");
872			goto out_nohpet;
873		}
874	}
875
876	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
877		goto out_nohpet;
878
879	/*
880	 * Read the HPET ID register to retrieve the IRQ routing
881	 * information and the number of channels
882	 */
883	id = hpet_readl(HPET_ID);
884	hpet_print_config();
885
886#ifdef CONFIG_HPET_EMULATE_RTC
887	/*
888	 * The legacy routing mode needs at least two channels, tick timer
889	 * and the rtc emulation channel.
890	 */
891	if (!(id & HPET_ID_NUMBER))
892		goto out_nohpet;
893#endif
894
895	if (hpet_clocksource_register())
896		goto out_nohpet;
897
898	if (id & HPET_ID_LEGSUP) {
899		hpet_legacy_clockevent_register();
900		return 1;
901	}
902	return 0;
903
904out_nohpet:
905	hpet_clear_mapping();
906	hpet_address = 0;
907	return 0;
908}
909
910/*
911 * Needs to be late, as the reserve_timer code calls kalloc !
912 *
913 * Not a problem on i386 as hpet_enable is called from late_time_init,
914 * but on x86_64 it is necessary !
915 */
916static __init int hpet_late_init(void)
917{
918	int cpu;
919
920	if (boot_hpet_disable)
921		return -ENODEV;
922
923	if (!hpet_address) {
924		if (!force_hpet_address)
925			return -ENODEV;
926
927		hpet_address = force_hpet_address;
928		hpet_enable();
929	}
930
931	if (!hpet_virt_address)
932		return -ENODEV;
933
934	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
935		hpet_msi_capability_lookup(2);
936	else
937		hpet_msi_capability_lookup(0);
938
939	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
940	hpet_print_config();
941
942	if (hpet_msi_disable)
943		return 0;
944
945	if (boot_cpu_has(X86_FEATURE_ARAT))
946		return 0;
947
948	for_each_online_cpu(cpu) {
949		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
950	}
951
952	/* This notifier should be called after workqueue is ready */
953	hotcpu_notifier(hpet_cpuhp_notify, -20);
954
955	return 0;
956}
957fs_initcall(hpet_late_init);
958
959void hpet_disable(void)
960{
961	if (is_hpet_capable()) {
962		unsigned int cfg = hpet_readl(HPET_CFG);
963
964		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}
972
973#ifdef CONFIG_HPET_EMULATE_RTC
974
975/* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
976 * is enabled, we support RTC interrupt functionality in software.
977 * RTC has 3 kinds of interrupts:
978 * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
979 *    is updated
980 * 2) Alarm Interrupt - generate an interrupt at a specific time of day
981 * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
982 *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
983 * (1) and (2) above are implemented using polling at a frequency of
984 * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
985 * overhead. (DEFAULT_RTC_INT_FREQ)
986 * For (3), we use interrupts at 64Hz or user specified periodic
987 * frequency, whichever is higher.
988 */
989#include <linux/mc146818rtc.h>
990#include <linux/rtc.h>
991#include <asm/rtc.h>
992
993#define DEFAULT_RTC_INT_FREQ	64
994#define DEFAULT_RTC_SHIFT	6
995#define RTC_NUM_INTS		1
996
997static unsigned long hpet_rtc_flags;
998static int hpet_prev_update_sec;
999static struct rtc_time hpet_alarm_time;
1000static unsigned long hpet_pie_count;
1001static u32 hpet_t1_cmp;
1002static u32 hpet_default_delta;
1003static u32 hpet_pie_delta;
1004static unsigned long hpet_pie_limit;
1005
1006static rtc_irq_handler irq_handler;
1007
1008/*
1009 * Check that the hpet counter c1 is ahead of the c2
1010 */
1011static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1012{
1013	return (s32)(c2 - c1) < 0;
1014}
1015
1016/*
1017 * Registers a IRQ handler.
1018 */
1019int hpet_register_irq_handler(rtc_irq_handler handler)
1020{
1021	if (!is_hpet_enabled())
1022		return -ENODEV;
1023	if (irq_handler)
1024		return -EBUSY;
1025
1026	irq_handler = handler;
1027
1028	return 0;
1029}
1030EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1031
1032/*
1033 * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1034 * and does cleanup.
1035 */
1036void hpet_unregister_irq_handler(rtc_irq_handler handler)
1037{
1038	if (!is_hpet_enabled())
1039		return;
1040
1041	irq_handler = NULL;
1042	hpet_rtc_flags = 0;
1043}
1044EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1045
1046/*
1047 * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1048 * is not supported by all HPET implementations for timer 1.
1049 *
1050 * hpet_rtc_timer_init() is called when the rtc is initialized.
1051 */
1052int hpet_rtc_timer_init(void)
1053{
1054	unsigned int cfg, cnt, delta;
1055	unsigned long flags;
1056
1057	if (!is_hpet_enabled())
1058		return 0;
1059
1060	if (!hpet_default_delta) {
1061		uint64_t clc;
1062
1063		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1064		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1065		hpet_default_delta = clc;
1066	}
1067
1068	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1069		delta = hpet_default_delta;
1070	else
1071		delta = hpet_pie_delta;
1072
1073	local_irq_save(flags);
1074
1075	cnt = delta + hpet_readl(HPET_COUNTER);
1076	hpet_writel(cnt, HPET_T1_CMP);
1077	hpet_t1_cmp = cnt;
1078
1079	cfg = hpet_readl(HPET_T1_CFG);
1080	cfg &= ~HPET_TN_PERIODIC;
1081	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1082	hpet_writel(cfg, HPET_T1_CFG);
1083
1084	local_irq_restore(flags);
1085
1086	return 1;
1087}
1088EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1089
1090/*
1091 * The functions below are called from rtc driver.
1092 * Return 0 if HPET is not being used.
1093 * Otherwise do the necessary changes and return 1.
1094 */
1095int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1096{
1097	if (!is_hpet_enabled())
1098		return 0;
1099
1100	hpet_rtc_flags &= ~bit_mask;
1101	return 1;
1102}
1103EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1104
1105int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1106{
1107	unsigned long oldbits = hpet_rtc_flags;
1108
1109	if (!is_hpet_enabled())
1110		return 0;
1111
1112	hpet_rtc_flags |= bit_mask;
1113
1114	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1115		hpet_prev_update_sec = -1;
1116
1117	if (!oldbits)
1118		hpet_rtc_timer_init();
1119
1120	return 1;
1121}
1122EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1123
1124int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1125			unsigned char sec)
1126{
1127	if (!is_hpet_enabled())
1128		return 0;
1129
1130	hpet_alarm_time.tm_hour = hrs;
1131	hpet_alarm_time.tm_min = min;
1132	hpet_alarm_time.tm_sec = sec;
1133
1134	return 1;
1135}
1136EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1137
1138int hpet_set_periodic_freq(unsigned long freq)
1139{
1140	uint64_t clc;
1141
1142	if (!is_hpet_enabled())
1143		return 0;
1144
1145	if (freq <= DEFAULT_RTC_INT_FREQ)
1146		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1147	else {
1148		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1149		do_div(clc, freq);
1150		clc >>= hpet_clockevent.shift;
1151		hpet_pie_delta = clc;
1152		hpet_pie_limit = 0;
1153	}
1154	return 1;
1155}
1156EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1157
1158int hpet_rtc_dropped_irq(void)
1159{
1160	return is_hpet_enabled();
1161}
1162EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1163
1164static void hpet_rtc_timer_reinit(void)
1165{
1166	unsigned int cfg, delta;
1167	int lost_ints = -1;
1168
1169	if (unlikely(!hpet_rtc_flags)) {
1170		cfg = hpet_readl(HPET_T1_CFG);
1171		cfg &= ~HPET_TN_ENABLE;
1172		hpet_writel(cfg, HPET_T1_CFG);
1173		return;
1174	}
1175
1176	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1177		delta = hpet_default_delta;
1178	else
1179		delta = hpet_pie_delta;
1180
1181	/*
1182	 * Increment the comparator value until we are ahead of the
1183	 * current count.
1184	 */
1185	do {
1186		hpet_t1_cmp += delta;
1187		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1188		lost_ints++;
1189	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1190
1191	if (lost_ints) {
1192		if (hpet_rtc_flags & RTC_PIE)
1193			hpet_pie_count += lost_ints;
1194		if (printk_ratelimit())
1195			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1196				lost_ints);
1197	}
1198}
1199
1200irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1201{
1202	struct rtc_time curr_time;
1203	unsigned long rtc_int_flag = 0;
1204
1205	hpet_rtc_timer_reinit();
1206	memset(&curr_time, 0, sizeof(struct rtc_time));
1207
1208	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1209		get_rtc_time(&curr_time);
1210
1211	if (hpet_rtc_flags & RTC_UIE &&
1212	    curr_time.tm_sec != hpet_prev_update_sec) {
1213		if (hpet_prev_update_sec >= 0)
1214			rtc_int_flag = RTC_UF;
1215		hpet_prev_update_sec = curr_time.tm_sec;
1216	}
1217
1218	if (hpet_rtc_flags & RTC_PIE &&
1219	    ++hpet_pie_count >= hpet_pie_limit) {
1220		rtc_int_flag |= RTC_PF;
1221		hpet_pie_count = 0;
1222	}
1223
1224	if (hpet_rtc_flags & RTC_AIE &&
1225	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1226	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1227	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1228			rtc_int_flag |= RTC_AF;
1229
1230	if (rtc_int_flag) {
1231		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1232		if (irq_handler)
1233			irq_handler(rtc_int_flag, dev_id);
1234	}
1235	return IRQ_HANDLED;
1236}
1237EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1238#endif
1239