hpet.c revision d0fbca8f9304d1760fdc906b35b06e89fbdbb51f
1#include <linux/clocksource.h>
2#include <linux/clockchips.h>
3#include <linux/interrupt.h>
4#include <linux/sysdev.h>
5#include <linux/delay.h>
6#include <linux/errno.h>
7#include <linux/slab.h>
8#include <linux/hpet.h>
9#include <linux/init.h>
10#include <linux/cpu.h>
11#include <linux/pm.h>
12#include <linux/io.h>
13
14#include <asm/fixmap.h>
15#include <asm/i8253.h>
16#include <asm/hpet.h>
17
18#define HPET_MASK			CLOCKSOURCE_MASK(32)
19
20/* FSEC = 10^-15
21   NSEC = 10^-9 */
22#define FSEC_PER_NSEC			1000000L
23
24#define HPET_DEV_USED_BIT		2
25#define HPET_DEV_USED			(1 << HPET_DEV_USED_BIT)
26#define HPET_DEV_VALID			0x8
27#define HPET_DEV_FSB_CAP		0x1000
28#define HPET_DEV_PERI_CAP		0x2000
29
30#define EVT_TO_HPET_DEV(evt) container_of(evt, struct hpet_dev, evt)
31
32/*
33 * HPET address is set in acpi/boot.c, when an ACPI entry exists
34 */
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	 *
403	 * An erratum on some chipsets (ICH9,..), results in
404	 * comparator read immediately following a write returning old
405	 * value. Workaround for this is to read this value second
406	 * time, when first read returns old value.
407	 *
408	 * In fact the write to the comparator register is delayed up
409	 * to two HPET cycles so the workaround we tried to restrict
410	 * the readback to those known to be borked ATI chipsets
411	 * failed miserably. So we give up on optimizations forever
412	 * and penalize all HPET incarnations unconditionally.
413	 */
414	if (unlikely((u32)hpet_readl(HPET_Tn_CMP(timer)) != cnt)) {
415		if (hpet_readl(HPET_Tn_CMP(timer)) != cnt)
416			printk_once(KERN_WARNING
417				"hpet: compare register read back failed.\n");
418	}
419
420	return (s32)(hpet_readl(HPET_COUNTER) - cnt) >= 0 ? -ETIME : 0;
421}
422
423static void hpet_legacy_set_mode(enum clock_event_mode mode,
424			struct clock_event_device *evt)
425{
426	hpet_set_mode(mode, evt, 0);
427}
428
429static int hpet_legacy_next_event(unsigned long delta,
430			struct clock_event_device *evt)
431{
432	return hpet_next_event(delta, evt, 0);
433}
434
435/*
436 * HPET MSI Support
437 */
438#ifdef CONFIG_PCI_MSI
439
440static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
441static struct hpet_dev	*hpet_devs;
442
443void hpet_msi_unmask(struct irq_data *data)
444{
445	struct hpet_dev *hdev = data->handler_data;
446	unsigned int cfg;
447
448	/* unmask it */
449	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
450	cfg |= HPET_TN_FSB;
451	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
452}
453
454void hpet_msi_mask(struct irq_data *data)
455{
456	struct hpet_dev *hdev = data->handler_data;
457	unsigned int cfg;
458
459	/* mask it */
460	cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
461	cfg &= ~HPET_TN_FSB;
462	hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
463}
464
465void hpet_msi_write(struct hpet_dev *hdev, struct msi_msg *msg)
466{
467	hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
468	hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
469}
470
471void hpet_msi_read(struct hpet_dev *hdev, struct msi_msg *msg)
472{
473	msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
474	msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
475	msg->address_hi = 0;
476}
477
478static void hpet_msi_set_mode(enum clock_event_mode mode,
479				struct clock_event_device *evt)
480{
481	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
482	hpet_set_mode(mode, evt, hdev->num);
483}
484
485static int hpet_msi_next_event(unsigned long delta,
486				struct clock_event_device *evt)
487{
488	struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
489	return hpet_next_event(delta, evt, hdev->num);
490}
491
492static int hpet_setup_msi_irq(unsigned int irq)
493{
494	if (arch_setup_hpet_msi(irq, hpet_blockid)) {
495		destroy_irq(irq);
496		return -EINVAL;
497	}
498	return 0;
499}
500
501static int hpet_assign_irq(struct hpet_dev *dev)
502{
503	unsigned int irq;
504
505	irq = create_irq_nr(0, -1);
506	if (!irq)
507		return -EINVAL;
508
509	set_irq_data(irq, dev);
510
511	if (hpet_setup_msi_irq(irq))
512		return -EINVAL;
513
514	dev->irq = irq;
515	return 0;
516}
517
518static irqreturn_t hpet_interrupt_handler(int irq, void *data)
519{
520	struct hpet_dev *dev = (struct hpet_dev *)data;
521	struct clock_event_device *hevt = &dev->evt;
522
523	if (!hevt->event_handler) {
524		printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
525				dev->num);
526		return IRQ_HANDLED;
527	}
528
529	hevt->event_handler(hevt);
530	return IRQ_HANDLED;
531}
532
533static int hpet_setup_irq(struct hpet_dev *dev)
534{
535
536	if (request_irq(dev->irq, hpet_interrupt_handler,
537			IRQF_TIMER | IRQF_DISABLED | IRQF_NOBALANCING,
538			dev->name, dev))
539		return -1;
540
541	disable_irq(dev->irq);
542	irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
543	enable_irq(dev->irq);
544
545	printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
546			 dev->name, dev->irq);
547
548	return 0;
549}
550
551/* This should be called in specific @cpu */
552static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
553{
554	struct clock_event_device *evt = &hdev->evt;
555	uint64_t hpet_freq;
556
557	WARN_ON(cpu != smp_processor_id());
558	if (!(hdev->flags & HPET_DEV_VALID))
559		return;
560
561	if (hpet_setup_msi_irq(hdev->irq))
562		return;
563
564	hdev->cpu = cpu;
565	per_cpu(cpu_hpet_dev, cpu) = hdev;
566	evt->name = hdev->name;
567	hpet_setup_irq(hdev);
568	evt->irq = hdev->irq;
569
570	evt->rating = 110;
571	evt->features = CLOCK_EVT_FEAT_ONESHOT;
572	if (hdev->flags & HPET_DEV_PERI_CAP)
573		evt->features |= CLOCK_EVT_FEAT_PERIODIC;
574
575	evt->set_mode = hpet_msi_set_mode;
576	evt->set_next_event = hpet_msi_next_event;
577	evt->shift = 32;
578
579	/*
580	 * The period is a femto seconds value. We need to calculate the
581	 * scaled math multiplication factor for nanosecond to hpet tick
582	 * conversion.
583	 */
584	hpet_freq = FSEC_PER_SEC;
585	do_div(hpet_freq, hpet_period);
586	evt->mult = div_sc((unsigned long) hpet_freq,
587				      NSEC_PER_SEC, evt->shift);
588	/* Calculate the max delta */
589	evt->max_delta_ns = clockevent_delta2ns(0x7FFFFFFF, evt);
590	/* 5 usec minimum reprogramming delta. */
591	evt->min_delta_ns = 5000;
592
593	evt->cpumask = cpumask_of(hdev->cpu);
594	clockevents_register_device(evt);
595}
596
597#ifdef CONFIG_HPET
598/* Reserve at least one timer for userspace (/dev/hpet) */
599#define RESERVE_TIMERS 1
600#else
601#define RESERVE_TIMERS 0
602#endif
603
604static void hpet_msi_capability_lookup(unsigned int start_timer)
605{
606	unsigned int id;
607	unsigned int num_timers;
608	unsigned int num_timers_used = 0;
609	int i;
610
611	if (hpet_msi_disable)
612		return;
613
614	if (boot_cpu_has(X86_FEATURE_ARAT))
615		return;
616	id = hpet_readl(HPET_ID);
617
618	num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
619	num_timers++; /* Value read out starts from 0 */
620	hpet_print_config();
621
622	hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
623	if (!hpet_devs)
624		return;
625
626	hpet_num_timers = num_timers;
627
628	for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
629		struct hpet_dev *hdev = &hpet_devs[num_timers_used];
630		unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
631
632		/* Only consider HPET timer with MSI support */
633		if (!(cfg & HPET_TN_FSB_CAP))
634			continue;
635
636		hdev->flags = 0;
637		if (cfg & HPET_TN_PERIODIC_CAP)
638			hdev->flags |= HPET_DEV_PERI_CAP;
639		hdev->num = i;
640
641		sprintf(hdev->name, "hpet%d", i);
642		if (hpet_assign_irq(hdev))
643			continue;
644
645		hdev->flags |= HPET_DEV_FSB_CAP;
646		hdev->flags |= HPET_DEV_VALID;
647		num_timers_used++;
648		if (num_timers_used == num_possible_cpus())
649			break;
650	}
651
652	printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
653		num_timers, num_timers_used);
654}
655
656#ifdef CONFIG_HPET
657static void hpet_reserve_msi_timers(struct hpet_data *hd)
658{
659	int i;
660
661	if (!hpet_devs)
662		return;
663
664	for (i = 0; i < hpet_num_timers; i++) {
665		struct hpet_dev *hdev = &hpet_devs[i];
666
667		if (!(hdev->flags & HPET_DEV_VALID))
668			continue;
669
670		hd->hd_irq[hdev->num] = hdev->irq;
671		hpet_reserve_timer(hd, hdev->num);
672	}
673}
674#endif
675
676static struct hpet_dev *hpet_get_unused_timer(void)
677{
678	int i;
679
680	if (!hpet_devs)
681		return NULL;
682
683	for (i = 0; i < hpet_num_timers; i++) {
684		struct hpet_dev *hdev = &hpet_devs[i];
685
686		if (!(hdev->flags & HPET_DEV_VALID))
687			continue;
688		if (test_and_set_bit(HPET_DEV_USED_BIT,
689			(unsigned long *)&hdev->flags))
690			continue;
691		return hdev;
692	}
693	return NULL;
694}
695
696struct hpet_work_struct {
697	struct delayed_work work;
698	struct completion complete;
699};
700
701static void hpet_work(struct work_struct *w)
702{
703	struct hpet_dev *hdev;
704	int cpu = smp_processor_id();
705	struct hpet_work_struct *hpet_work;
706
707	hpet_work = container_of(w, struct hpet_work_struct, work.work);
708
709	hdev = hpet_get_unused_timer();
710	if (hdev)
711		init_one_hpet_msi_clockevent(hdev, cpu);
712
713	complete(&hpet_work->complete);
714}
715
716static int hpet_cpuhp_notify(struct notifier_block *n,
717		unsigned long action, void *hcpu)
718{
719	unsigned long cpu = (unsigned long)hcpu;
720	struct hpet_work_struct work;
721	struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
722
723	switch (action & 0xf) {
724	case CPU_ONLINE:
725		INIT_DELAYED_WORK_ON_STACK(&work.work, hpet_work);
726		init_completion(&work.complete);
727		/* FIXME: add schedule_work_on() */
728		schedule_delayed_work_on(cpu, &work.work, 0);
729		wait_for_completion(&work.complete);
730		destroy_timer_on_stack(&work.work.timer);
731		break;
732	case CPU_DEAD:
733		if (hdev) {
734			free_irq(hdev->irq, hdev);
735			hdev->flags &= ~HPET_DEV_USED;
736			per_cpu(cpu_hpet_dev, cpu) = NULL;
737		}
738		break;
739	}
740	return NOTIFY_OK;
741}
742#else
743
744static int hpet_setup_msi_irq(unsigned int irq)
745{
746	return 0;
747}
748static void hpet_msi_capability_lookup(unsigned int start_timer)
749{
750	return;
751}
752
753#ifdef CONFIG_HPET
754static void hpet_reserve_msi_timers(struct hpet_data *hd)
755{
756	return;
757}
758#endif
759
760static int hpet_cpuhp_notify(struct notifier_block *n,
761		unsigned long action, void *hcpu)
762{
763	return NOTIFY_OK;
764}
765
766#endif
767
768/*
769 * Clock source related code
770 */
771static cycle_t read_hpet(struct clocksource *cs)
772{
773	return (cycle_t)hpet_readl(HPET_COUNTER);
774}
775
776#ifdef CONFIG_X86_64
777static cycle_t __vsyscall_fn vread_hpet(void)
778{
779	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + 0xf0);
780}
781#endif
782
783static struct clocksource clocksource_hpet = {
784	.name		= "hpet",
785	.rating		= 250,
786	.read		= read_hpet,
787	.mask		= HPET_MASK,
788	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
789	.resume		= hpet_resume_counter,
790#ifdef CONFIG_X86_64
791	.vread		= vread_hpet,
792#endif
793};
794
795static int hpet_clocksource_register(void)
796{
797	u64 start, now;
798	u64 hpet_freq;
799	cycle_t t1;
800
801	/* Start the counter */
802	hpet_restart_counter();
803
804	/* Verify whether hpet counter works */
805	t1 = hpet_readl(HPET_COUNTER);
806	rdtscll(start);
807
808	/*
809	 * We don't know the TSC frequency yet, but waiting for
810	 * 200000 TSC cycles is safe:
811	 * 4 GHz == 50us
812	 * 1 GHz == 200us
813	 */
814	do {
815		rep_nop();
816		rdtscll(now);
817	} while ((now - start) < 200000UL);
818
819	if (t1 == hpet_readl(HPET_COUNTER)) {
820		printk(KERN_WARNING
821		       "HPET counter not counting. HPET disabled\n");
822		return -ENODEV;
823	}
824
825	/*
826	 * The definition of mult is (include/linux/clocksource.h)
827	 * mult/2^shift = ns/cyc and hpet_period is in units of fsec/cyc
828	 * so we first need to convert hpet_period to ns/cyc units:
829	 *  mult/2^shift = ns/cyc = hpet_period/10^6
830	 *  mult = (hpet_period * 2^shift)/10^6
831	 *  mult = (hpet_period << shift)/FSEC_PER_NSEC
832	 */
833
834	/* Need to convert hpet_period (fsec/cyc) to cyc/sec:
835	 *
836	 * cyc/sec = FSEC_PER_SEC/hpet_period(fsec/cyc)
837	 * cyc/sec = (FSEC_PER_NSEC * NSEC_PER_SEC)/hpet_period
838	 */
839	hpet_freq = FSEC_PER_SEC;
840	do_div(hpet_freq, hpet_period);
841	clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
842
843	return 0;
844}
845
846/**
847 * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
848 */
849int __init hpet_enable(void)
850{
851	unsigned int id;
852	int i;
853
854	if (!is_hpet_capable())
855		return 0;
856
857	hpet_set_mapping();
858
859	/*
860	 * Read the period and check for a sane value:
861	 */
862	hpet_period = hpet_readl(HPET_PERIOD);
863
864	/*
865	 * AMD SB700 based systems with spread spectrum enabled use a
866	 * SMM based HPET emulation to provide proper frequency
867	 * setting. The SMM code is initialized with the first HPET
868	 * register access and takes some time to complete. During
869	 * this time the config register reads 0xffffffff. We check
870	 * for max. 1000 loops whether the config register reads a non
871	 * 0xffffffff value to make sure that HPET is up and running
872	 * before we go further. A counting loop is safe, as the HPET
873	 * access takes thousands of CPU cycles. On non SB700 based
874	 * machines this check is only done once and has no side
875	 * effects.
876	 */
877	for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
878		if (i == 1000) {
879			printk(KERN_WARNING
880			       "HPET config register value = 0xFFFFFFFF. "
881			       "Disabling HPET\n");
882			goto out_nohpet;
883		}
884	}
885
886	if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
887		goto out_nohpet;
888
889	/*
890	 * Read the HPET ID register to retrieve the IRQ routing
891	 * information and the number of channels
892	 */
893	id = hpet_readl(HPET_ID);
894	hpet_print_config();
895
896#ifdef CONFIG_HPET_EMULATE_RTC
897	/*
898	 * The legacy routing mode needs at least two channels, tick timer
899	 * and the rtc emulation channel.
900	 */
901	if (!(id & HPET_ID_NUMBER))
902		goto out_nohpet;
903#endif
904
905	if (hpet_clocksource_register())
906		goto out_nohpet;
907
908	if (id & HPET_ID_LEGSUP) {
909		hpet_legacy_clockevent_register();
910		return 1;
911	}
912	return 0;
913
914out_nohpet:
915	hpet_clear_mapping();
916	hpet_address = 0;
917	return 0;
918}
919
920/*
921 * Needs to be late, as the reserve_timer code calls kalloc !
922 *
923 * Not a problem on i386 as hpet_enable is called from late_time_init,
924 * but on x86_64 it is necessary !
925 */
926static __init int hpet_late_init(void)
927{
928	int cpu;
929
930	if (boot_hpet_disable)
931		return -ENODEV;
932
933	if (!hpet_address) {
934		if (!force_hpet_address)
935			return -ENODEV;
936
937		hpet_address = force_hpet_address;
938		hpet_enable();
939	}
940
941	if (!hpet_virt_address)
942		return -ENODEV;
943
944	if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
945		hpet_msi_capability_lookup(2);
946	else
947		hpet_msi_capability_lookup(0);
948
949	hpet_reserve_platform_timers(hpet_readl(HPET_ID));
950	hpet_print_config();
951
952	if (hpet_msi_disable)
953		return 0;
954
955	if (boot_cpu_has(X86_FEATURE_ARAT))
956		return 0;
957
958	for_each_online_cpu(cpu) {
959		hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
960	}
961
962	/* This notifier should be called after workqueue is ready */
963	hotcpu_notifier(hpet_cpuhp_notify, -20);
964
965	return 0;
966}
967fs_initcall(hpet_late_init);
968
969void hpet_disable(void)
970{
971	if (is_hpet_capable() && hpet_virt_address) {
972		unsigned int cfg = hpet_readl(HPET_CFG);
973
974		if (hpet_legacy_int_enabled) {
975			cfg &= ~HPET_CFG_LEGACY;
976			hpet_legacy_int_enabled = 0;
977		}
978		cfg &= ~HPET_CFG_ENABLE;
979		hpet_writel(cfg, HPET_CFG);
980	}
981}
982
983#ifdef CONFIG_HPET_EMULATE_RTC
984
985/* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
986 * is enabled, we support RTC interrupt functionality in software.
987 * RTC has 3 kinds of interrupts:
988 * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
989 *    is updated
990 * 2) Alarm Interrupt - generate an interrupt at a specific time of day
991 * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
992 *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
993 * (1) and (2) above are implemented using polling at a frequency of
994 * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
995 * overhead. (DEFAULT_RTC_INT_FREQ)
996 * For (3), we use interrupts at 64Hz or user specified periodic
997 * frequency, whichever is higher.
998 */
999#include <linux/mc146818rtc.h>
1000#include <linux/rtc.h>
1001#include <asm/rtc.h>
1002
1003#define DEFAULT_RTC_INT_FREQ	64
1004#define DEFAULT_RTC_SHIFT	6
1005#define RTC_NUM_INTS		1
1006
1007static unsigned long hpet_rtc_flags;
1008static int hpet_prev_update_sec;
1009static struct rtc_time hpet_alarm_time;
1010static unsigned long hpet_pie_count;
1011static u32 hpet_t1_cmp;
1012static u32 hpet_default_delta;
1013static u32 hpet_pie_delta;
1014static unsigned long hpet_pie_limit;
1015
1016static rtc_irq_handler irq_handler;
1017
1018/*
1019 * Check that the hpet counter c1 is ahead of the c2
1020 */
1021static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1022{
1023	return (s32)(c2 - c1) < 0;
1024}
1025
1026/*
1027 * Registers a IRQ handler.
1028 */
1029int hpet_register_irq_handler(rtc_irq_handler handler)
1030{
1031	if (!is_hpet_enabled())
1032		return -ENODEV;
1033	if (irq_handler)
1034		return -EBUSY;
1035
1036	irq_handler = handler;
1037
1038	return 0;
1039}
1040EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1041
1042/*
1043 * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1044 * and does cleanup.
1045 */
1046void hpet_unregister_irq_handler(rtc_irq_handler handler)
1047{
1048	if (!is_hpet_enabled())
1049		return;
1050
1051	irq_handler = NULL;
1052	hpet_rtc_flags = 0;
1053}
1054EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1055
1056/*
1057 * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1058 * is not supported by all HPET implementations for timer 1.
1059 *
1060 * hpet_rtc_timer_init() is called when the rtc is initialized.
1061 */
1062int hpet_rtc_timer_init(void)
1063{
1064	unsigned int cfg, cnt, delta;
1065	unsigned long flags;
1066
1067	if (!is_hpet_enabled())
1068		return 0;
1069
1070	if (!hpet_default_delta) {
1071		uint64_t clc;
1072
1073		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1074		clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1075		hpet_default_delta = clc;
1076	}
1077
1078	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1079		delta = hpet_default_delta;
1080	else
1081		delta = hpet_pie_delta;
1082
1083	local_irq_save(flags);
1084
1085	cnt = delta + hpet_readl(HPET_COUNTER);
1086	hpet_writel(cnt, HPET_T1_CMP);
1087	hpet_t1_cmp = cnt;
1088
1089	cfg = hpet_readl(HPET_T1_CFG);
1090	cfg &= ~HPET_TN_PERIODIC;
1091	cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1092	hpet_writel(cfg, HPET_T1_CFG);
1093
1094	local_irq_restore(flags);
1095
1096	return 1;
1097}
1098EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1099
1100/*
1101 * The functions below are called from rtc driver.
1102 * Return 0 if HPET is not being used.
1103 * Otherwise do the necessary changes and return 1.
1104 */
1105int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1106{
1107	if (!is_hpet_enabled())
1108		return 0;
1109
1110	hpet_rtc_flags &= ~bit_mask;
1111	return 1;
1112}
1113EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1114
1115int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1116{
1117	unsigned long oldbits = hpet_rtc_flags;
1118
1119	if (!is_hpet_enabled())
1120		return 0;
1121
1122	hpet_rtc_flags |= bit_mask;
1123
1124	if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1125		hpet_prev_update_sec = -1;
1126
1127	if (!oldbits)
1128		hpet_rtc_timer_init();
1129
1130	return 1;
1131}
1132EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1133
1134int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1135			unsigned char sec)
1136{
1137	if (!is_hpet_enabled())
1138		return 0;
1139
1140	hpet_alarm_time.tm_hour = hrs;
1141	hpet_alarm_time.tm_min = min;
1142	hpet_alarm_time.tm_sec = sec;
1143
1144	return 1;
1145}
1146EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1147
1148int hpet_set_periodic_freq(unsigned long freq)
1149{
1150	uint64_t clc;
1151
1152	if (!is_hpet_enabled())
1153		return 0;
1154
1155	if (freq <= DEFAULT_RTC_INT_FREQ)
1156		hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1157	else {
1158		clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1159		do_div(clc, freq);
1160		clc >>= hpet_clockevent.shift;
1161		hpet_pie_delta = clc;
1162		hpet_pie_limit = 0;
1163	}
1164	return 1;
1165}
1166EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1167
1168int hpet_rtc_dropped_irq(void)
1169{
1170	return is_hpet_enabled();
1171}
1172EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1173
1174static void hpet_rtc_timer_reinit(void)
1175{
1176	unsigned int cfg, delta;
1177	int lost_ints = -1;
1178
1179	if (unlikely(!hpet_rtc_flags)) {
1180		cfg = hpet_readl(HPET_T1_CFG);
1181		cfg &= ~HPET_TN_ENABLE;
1182		hpet_writel(cfg, HPET_T1_CFG);
1183		return;
1184	}
1185
1186	if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1187		delta = hpet_default_delta;
1188	else
1189		delta = hpet_pie_delta;
1190
1191	/*
1192	 * Increment the comparator value until we are ahead of the
1193	 * current count.
1194	 */
1195	do {
1196		hpet_t1_cmp += delta;
1197		hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1198		lost_ints++;
1199	} while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1200
1201	if (lost_ints) {
1202		if (hpet_rtc_flags & RTC_PIE)
1203			hpet_pie_count += lost_ints;
1204		if (printk_ratelimit())
1205			printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1206				lost_ints);
1207	}
1208}
1209
1210irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1211{
1212	struct rtc_time curr_time;
1213	unsigned long rtc_int_flag = 0;
1214
1215	hpet_rtc_timer_reinit();
1216	memset(&curr_time, 0, sizeof(struct rtc_time));
1217
1218	if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1219		get_rtc_time(&curr_time);
1220
1221	if (hpet_rtc_flags & RTC_UIE &&
1222	    curr_time.tm_sec != hpet_prev_update_sec) {
1223		if (hpet_prev_update_sec >= 0)
1224			rtc_int_flag = RTC_UF;
1225		hpet_prev_update_sec = curr_time.tm_sec;
1226	}
1227
1228	if (hpet_rtc_flags & RTC_PIE &&
1229	    ++hpet_pie_count >= hpet_pie_limit) {
1230		rtc_int_flag |= RTC_PF;
1231		hpet_pie_count = 0;
1232	}
1233
1234	if (hpet_rtc_flags & RTC_AIE &&
1235	    (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1236	    (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1237	    (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1238			rtc_int_flag |= RTC_AF;
1239
1240	if (rtc_int_flag) {
1241		rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1242		if (irq_handler)
1243			irq_handler(rtc_int_flag, dev_id);
1244	}
1245	return IRQ_HANDLED;
1246}
1247EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1248#endif
1249