Loading...
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 */
37unsigned long hpet_address;
38u8 hpet_blockid; /* OS timer block num */
39u8 hpet_msi_disable;
40
41#ifdef CONFIG_PCI_MSI
42static unsigned long hpet_num_timers;
43#endif
44static void __iomem *hpet_virt_address;
45
46struct 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
55inline struct hpet_dev *EVT_TO_HPET_DEV(struct clock_event_device *evtdev)
56{
57 return container_of(evtdev, struct hpet_dev, evt);
58}
59
60inline unsigned int hpet_readl(unsigned int a)
61{
62 return readl(hpet_virt_address + a);
63}
64
65static 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
74static 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
82static 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 */
91static int boot_hpet_disable;
92int hpet_force_user;
93static int hpet_verbose;
94
95static 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
114static int __init disable_hpet(char *str)
115{
116 boot_hpet_disable = 1;
117 return 1;
118}
119__setup("nohpet", disable_hpet);
120
121static inline int is_hpet_capable(void)
122{
123 return !boot_hpet_disable && hpet_address;
124}
125
126/*
127 * HPET timer interrupt enable / disable
128 */
129static int hpet_legacy_int_enabled;
130
131/**
132 * is_hpet_enabled - check whether the hpet timer interrupt is enabled
133 */
134int is_hpet_enabled(void)
135{
136 return is_hpet_capable() && hpet_legacy_int_enabled;
137}
138EXPORT_SYMBOL_GPL(is_hpet_enabled);
139
140static 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() \
172do { \
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
183static void hpet_reserve_msi_timers(struct hpet_data *hd);
184
185static 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
223static void hpet_reserve_platform_timers(unsigned int id) { }
224#endif
225
226/*
227 * Common hpet info
228 */
229static unsigned long hpet_freq;
230
231static void hpet_legacy_set_mode(enum clock_event_mode mode,
232 struct clock_event_device *evt);
233static int hpet_legacy_next_event(unsigned long delta,
234 struct clock_event_device *evt);
235
236/*
237 * The hpet clock event device
238 */
239static 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
248static 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
255static void hpet_reset_counter(void)
256{
257 hpet_writel(0, HPET_COUNTER);
258 hpet_writel(0, HPET_COUNTER + 4);
259}
260
261static 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
268static void hpet_restart_counter(void)
269{
270 hpet_stop_counter();
271 hpet_reset_counter();
272 hpet_start_counter();
273}
274
275static void hpet_resume_device(void)
276{
277 force_hpet_resume();
278}
279
280static void hpet_resume_counter(struct clocksource *cs)
281{
282 hpet_resume_device();
283 hpet_restart_counter();
284}
285
286static 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
295static 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
311static int hpet_setup_msi_irq(unsigned int irq);
312
313static 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
373static 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
410static 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
416static 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
427static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
428static struct hpet_dev *hpet_devs;
429
430void 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_FSB;
438 hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
439}
440
441void 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_FSB;
449 hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
450}
451
452void 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
458void 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
465static 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
472static 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
479static 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
488static 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
505static 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
520static 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 */
539static 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
576static 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
629static 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
648static 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
668struct hpet_work_struct {
669 struct delayed_work work;
670 struct completion complete;
671};
672
673static 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
688static 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
716static int hpet_setup_msi_irq(unsigned int irq)
717{
718 return 0;
719}
720static void hpet_msi_capability_lookup(unsigned int start_timer)
721{
722 return;
723}
724
725#ifdef CONFIG_HPET
726static void hpet_reserve_msi_timers(struct hpet_data *hd)
727{
728 return;
729}
730#endif
731
732static 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 */
743static cycle_t read_hpet(struct clocksource *cs)
744{
745 return (cycle_t)hpet_readl(HPET_COUNTER);
746}
747
748static 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
760static 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
793static u32 *hpet_boot_cfg;
794
795/**
796 * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
797 */
798int __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
902out_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 */
914static __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}
955fs_initcall(hpet_late_init);
956
957void 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
1009static unsigned long hpet_rtc_flags;
1010static int hpet_prev_update_sec;
1011static struct rtc_time hpet_alarm_time;
1012static unsigned long hpet_pie_count;
1013static u32 hpet_t1_cmp;
1014static u32 hpet_default_delta;
1015static u32 hpet_pie_delta;
1016static unsigned long hpet_pie_limit;
1017
1018static rtc_irq_handler irq_handler;
1019
1020/*
1021 * Check that the hpet counter c1 is ahead of the c2
1022 */
1023static 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 */
1031int 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}
1042EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1043
1044/*
1045 * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1046 * and does cleanup.
1047 */
1048void 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}
1056EXPORT_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 */
1064int 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}
1100EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1101
1102static 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 */
1115int 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}
1126EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1127
1128int 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}
1145EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1146
1147int 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}
1159EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1160
1161int 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}
1179EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1180
1181int hpet_rtc_dropped_irq(void)
1182{
1183 return is_hpet_enabled();
1184}
1185EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1186
1187static 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
1219irqreturn_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}
1256EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1257#endif
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/clockchips.h>
3#include <linux/interrupt.h>
4#include <linux/export.h>
5#include <linux/delay.h>
6#include <linux/hpet.h>
7#include <linux/cpu.h>
8#include <linux/irq.h>
9
10#include <asm/irq_remapping.h>
11#include <asm/hpet.h>
12#include <asm/time.h>
13#include <asm/mwait.h>
14
15#undef pr_fmt
16#define pr_fmt(fmt) "hpet: " fmt
17
18enum hpet_mode {
19 HPET_MODE_UNUSED,
20 HPET_MODE_LEGACY,
21 HPET_MODE_CLOCKEVT,
22 HPET_MODE_DEVICE,
23};
24
25struct hpet_channel {
26 struct clock_event_device evt;
27 unsigned int num;
28 unsigned int cpu;
29 unsigned int irq;
30 unsigned int in_use;
31 enum hpet_mode mode;
32 unsigned int boot_cfg;
33 char name[10];
34};
35
36struct hpet_base {
37 unsigned int nr_channels;
38 unsigned int nr_clockevents;
39 unsigned int boot_cfg;
40 struct hpet_channel *channels;
41};
42
43#define HPET_MASK CLOCKSOURCE_MASK(32)
44
45#define HPET_MIN_CYCLES 128
46#define HPET_MIN_PROG_DELTA (HPET_MIN_CYCLES + (HPET_MIN_CYCLES >> 1))
47
48/*
49 * HPET address is set in acpi/boot.c, when an ACPI entry exists
50 */
51unsigned long hpet_address;
52u8 hpet_blockid; /* OS timer block num */
53bool hpet_msi_disable;
54
55#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_GENERIC_MSI_IRQ)
56static DEFINE_PER_CPU(struct hpet_channel *, cpu_hpet_channel);
57static struct irq_domain *hpet_domain;
58#endif
59
60static void __iomem *hpet_virt_address;
61
62static struct hpet_base hpet_base;
63
64static bool hpet_legacy_int_enabled;
65static unsigned long hpet_freq;
66
67bool boot_hpet_disable;
68bool hpet_force_user;
69static bool hpet_verbose;
70
71static inline
72struct hpet_channel *clockevent_to_channel(struct clock_event_device *evt)
73{
74 return container_of(evt, struct hpet_channel, evt);
75}
76
77inline unsigned int hpet_readl(unsigned int a)
78{
79 return readl(hpet_virt_address + a);
80}
81
82static inline void hpet_writel(unsigned int d, unsigned int a)
83{
84 writel(d, hpet_virt_address + a);
85}
86
87static inline void hpet_set_mapping(void)
88{
89 hpet_virt_address = ioremap(hpet_address, HPET_MMAP_SIZE);
90}
91
92static inline void hpet_clear_mapping(void)
93{
94 iounmap(hpet_virt_address);
95 hpet_virt_address = NULL;
96}
97
98/*
99 * HPET command line enable / disable
100 */
101static int __init hpet_setup(char *str)
102{
103 while (str) {
104 char *next = strchr(str, ',');
105
106 if (next)
107 *next++ = 0;
108 if (!strncmp("disable", str, 7))
109 boot_hpet_disable = true;
110 if (!strncmp("force", str, 5))
111 hpet_force_user = true;
112 if (!strncmp("verbose", str, 7))
113 hpet_verbose = true;
114 str = next;
115 }
116 return 1;
117}
118__setup("hpet=", hpet_setup);
119
120static int __init disable_hpet(char *str)
121{
122 boot_hpet_disable = true;
123 return 1;
124}
125__setup("nohpet", disable_hpet);
126
127static inline int is_hpet_capable(void)
128{
129 return !boot_hpet_disable && hpet_address;
130}
131
132/**
133 * is_hpet_enabled - Check whether the legacy HPET timer interrupt is enabled
134 */
135int is_hpet_enabled(void)
136{
137 return is_hpet_capable() && hpet_legacy_int_enabled;
138}
139EXPORT_SYMBOL_GPL(is_hpet_enabled);
140
141static void _hpet_print_config(const char *function, int line)
142{
143 u32 i, id, period, cfg, status, channels, l, h;
144
145 pr_info("%s(%d):\n", function, line);
146
147 id = hpet_readl(HPET_ID);
148 period = hpet_readl(HPET_PERIOD);
149 pr_info("ID: 0x%x, PERIOD: 0x%x\n", id, period);
150
151 cfg = hpet_readl(HPET_CFG);
152 status = hpet_readl(HPET_STATUS);
153 pr_info("CFG: 0x%x, STATUS: 0x%x\n", cfg, status);
154
155 l = hpet_readl(HPET_COUNTER);
156 h = hpet_readl(HPET_COUNTER+4);
157 pr_info("COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
158
159 channels = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
160
161 for (i = 0; i < channels; i++) {
162 l = hpet_readl(HPET_Tn_CFG(i));
163 h = hpet_readl(HPET_Tn_CFG(i)+4);
164 pr_info("T%d: CFG_l: 0x%x, CFG_h: 0x%x\n", i, l, h);
165
166 l = hpet_readl(HPET_Tn_CMP(i));
167 h = hpet_readl(HPET_Tn_CMP(i)+4);
168 pr_info("T%d: CMP_l: 0x%x, CMP_h: 0x%x\n", i, l, h);
169
170 l = hpet_readl(HPET_Tn_ROUTE(i));
171 h = hpet_readl(HPET_Tn_ROUTE(i)+4);
172 pr_info("T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n", i, l, h);
173 }
174}
175
176#define hpet_print_config() \
177do { \
178 if (hpet_verbose) \
179 _hpet_print_config(__func__, __LINE__); \
180} while (0)
181
182/*
183 * When the HPET driver (/dev/hpet) is enabled, we need to reserve
184 * timer 0 and timer 1 in case of RTC emulation.
185 */
186#ifdef CONFIG_HPET
187
188static void __init hpet_reserve_platform_timers(void)
189{
190 struct hpet_data hd;
191 unsigned int i;
192
193 memset(&hd, 0, sizeof(hd));
194 hd.hd_phys_address = hpet_address;
195 hd.hd_address = hpet_virt_address;
196 hd.hd_nirqs = hpet_base.nr_channels;
197
198 /*
199 * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
200 * is wrong for i8259!) not the output IRQ. Many BIOS writers
201 * don't bother configuring *any* comparator interrupts.
202 */
203 hd.hd_irq[0] = HPET_LEGACY_8254;
204 hd.hd_irq[1] = HPET_LEGACY_RTC;
205
206 for (i = 0; i < hpet_base.nr_channels; i++) {
207 struct hpet_channel *hc = hpet_base.channels + i;
208
209 if (i >= 2)
210 hd.hd_irq[i] = hc->irq;
211
212 switch (hc->mode) {
213 case HPET_MODE_UNUSED:
214 case HPET_MODE_DEVICE:
215 hc->mode = HPET_MODE_DEVICE;
216 break;
217 case HPET_MODE_CLOCKEVT:
218 case HPET_MODE_LEGACY:
219 hpet_reserve_timer(&hd, hc->num);
220 break;
221 }
222 }
223
224 hpet_alloc(&hd);
225}
226
227static void __init hpet_select_device_channel(void)
228{
229 int i;
230
231 for (i = 0; i < hpet_base.nr_channels; i++) {
232 struct hpet_channel *hc = hpet_base.channels + i;
233
234 /* Associate the first unused channel to /dev/hpet */
235 if (hc->mode == HPET_MODE_UNUSED) {
236 hc->mode = HPET_MODE_DEVICE;
237 return;
238 }
239 }
240}
241
242#else
243static inline void hpet_reserve_platform_timers(void) { }
244static inline void hpet_select_device_channel(void) {}
245#endif
246
247/* Common HPET functions */
248static void hpet_stop_counter(void)
249{
250 u32 cfg = hpet_readl(HPET_CFG);
251
252 cfg &= ~HPET_CFG_ENABLE;
253 hpet_writel(cfg, HPET_CFG);
254}
255
256static void hpet_reset_counter(void)
257{
258 hpet_writel(0, HPET_COUNTER);
259 hpet_writel(0, HPET_COUNTER + 4);
260}
261
262static void hpet_start_counter(void)
263{
264 unsigned int cfg = hpet_readl(HPET_CFG);
265
266 cfg |= HPET_CFG_ENABLE;
267 hpet_writel(cfg, HPET_CFG);
268}
269
270static void hpet_restart_counter(void)
271{
272 hpet_stop_counter();
273 hpet_reset_counter();
274 hpet_start_counter();
275}
276
277static void hpet_resume_device(void)
278{
279 force_hpet_resume();
280}
281
282static void hpet_resume_counter(struct clocksource *cs)
283{
284 hpet_resume_device();
285 hpet_restart_counter();
286}
287
288static void hpet_enable_legacy_int(void)
289{
290 unsigned int cfg = hpet_readl(HPET_CFG);
291
292 cfg |= HPET_CFG_LEGACY;
293 hpet_writel(cfg, HPET_CFG);
294 hpet_legacy_int_enabled = true;
295}
296
297static int hpet_clkevt_set_state_periodic(struct clock_event_device *evt)
298{
299 unsigned int channel = clockevent_to_channel(evt)->num;
300 unsigned int cfg, cmp, now;
301 uint64_t delta;
302
303 hpet_stop_counter();
304 delta = ((uint64_t)(NSEC_PER_SEC / HZ)) * evt->mult;
305 delta >>= evt->shift;
306 now = hpet_readl(HPET_COUNTER);
307 cmp = now + (unsigned int)delta;
308 cfg = hpet_readl(HPET_Tn_CFG(channel));
309 cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC | HPET_TN_SETVAL |
310 HPET_TN_32BIT;
311 hpet_writel(cfg, HPET_Tn_CFG(channel));
312 hpet_writel(cmp, HPET_Tn_CMP(channel));
313 udelay(1);
314 /*
315 * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
316 * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
317 * bit is automatically cleared after the first write.
318 * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
319 * Publication # 24674)
320 */
321 hpet_writel((unsigned int)delta, HPET_Tn_CMP(channel));
322 hpet_start_counter();
323 hpet_print_config();
324
325 return 0;
326}
327
328static int hpet_clkevt_set_state_oneshot(struct clock_event_device *evt)
329{
330 unsigned int channel = clockevent_to_channel(evt)->num;
331 unsigned int cfg;
332
333 cfg = hpet_readl(HPET_Tn_CFG(channel));
334 cfg &= ~HPET_TN_PERIODIC;
335 cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
336 hpet_writel(cfg, HPET_Tn_CFG(channel));
337
338 return 0;
339}
340
341static int hpet_clkevt_set_state_shutdown(struct clock_event_device *evt)
342{
343 unsigned int channel = clockevent_to_channel(evt)->num;
344 unsigned int cfg;
345
346 cfg = hpet_readl(HPET_Tn_CFG(channel));
347 cfg &= ~HPET_TN_ENABLE;
348 hpet_writel(cfg, HPET_Tn_CFG(channel));
349
350 return 0;
351}
352
353static int hpet_clkevt_legacy_resume(struct clock_event_device *evt)
354{
355 hpet_enable_legacy_int();
356 hpet_print_config();
357 return 0;
358}
359
360static int
361hpet_clkevt_set_next_event(unsigned long delta, struct clock_event_device *evt)
362{
363 unsigned int channel = clockevent_to_channel(evt)->num;
364 u32 cnt;
365 s32 res;
366
367 cnt = hpet_readl(HPET_COUNTER);
368 cnt += (u32) delta;
369 hpet_writel(cnt, HPET_Tn_CMP(channel));
370
371 /*
372 * HPETs are a complete disaster. The compare register is
373 * based on a equal comparison and neither provides a less
374 * than or equal functionality (which would require to take
375 * the wraparound into account) nor a simple count down event
376 * mode. Further the write to the comparator register is
377 * delayed internally up to two HPET clock cycles in certain
378 * chipsets (ATI, ICH9,10). Some newer AMD chipsets have even
379 * longer delays. We worked around that by reading back the
380 * compare register, but that required another workaround for
381 * ICH9,10 chips where the first readout after write can
382 * return the old stale value. We already had a minimum
383 * programming delta of 5us enforced, but a NMI or SMI hitting
384 * between the counter readout and the comparator write can
385 * move us behind that point easily. Now instead of reading
386 * the compare register back several times, we make the ETIME
387 * decision based on the following: Return ETIME if the
388 * counter value after the write is less than HPET_MIN_CYCLES
389 * away from the event or if the counter is already ahead of
390 * the event. The minimum programming delta for the generic
391 * clockevents code is set to 1.5 * HPET_MIN_CYCLES.
392 */
393 res = (s32)(cnt - hpet_readl(HPET_COUNTER));
394
395 return res < HPET_MIN_CYCLES ? -ETIME : 0;
396}
397
398static void hpet_init_clockevent(struct hpet_channel *hc, unsigned int rating)
399{
400 struct clock_event_device *evt = &hc->evt;
401
402 evt->rating = rating;
403 evt->irq = hc->irq;
404 evt->name = hc->name;
405 evt->cpumask = cpumask_of(hc->cpu);
406 evt->set_state_oneshot = hpet_clkevt_set_state_oneshot;
407 evt->set_next_event = hpet_clkevt_set_next_event;
408 evt->set_state_shutdown = hpet_clkevt_set_state_shutdown;
409
410 evt->features = CLOCK_EVT_FEAT_ONESHOT;
411 if (hc->boot_cfg & HPET_TN_PERIODIC) {
412 evt->features |= CLOCK_EVT_FEAT_PERIODIC;
413 evt->set_state_periodic = hpet_clkevt_set_state_periodic;
414 }
415}
416
417static void __init hpet_legacy_clockevent_register(struct hpet_channel *hc)
418{
419 /*
420 * Start HPET with the boot CPU's cpumask and make it global after
421 * the IO_APIC has been initialized.
422 */
423 hc->cpu = boot_cpu_data.cpu_index;
424 strscpy(hc->name, "hpet", sizeof(hc->name));
425 hpet_init_clockevent(hc, 50);
426
427 hc->evt.tick_resume = hpet_clkevt_legacy_resume;
428
429 /*
430 * Legacy horrors and sins from the past. HPET used periodic mode
431 * unconditionally forever on the legacy channel 0. Removing the
432 * below hack and using the conditional in hpet_init_clockevent()
433 * makes at least Qemu and one hardware machine fail to boot.
434 * There are two issues which cause the boot failure:
435 *
436 * #1 After the timer delivery test in IOAPIC and the IOAPIC setup
437 * the next interrupt is not delivered despite the HPET channel
438 * being programmed correctly. Reprogramming the HPET after
439 * switching to IOAPIC makes it work again. After fixing this,
440 * the next issue surfaces:
441 *
442 * #2 Due to the unconditional periodic mode availability the Local
443 * APIC timer calibration can hijack the global clockevents
444 * event handler without causing damage. Using oneshot at this
445 * stage makes if hang because the HPET does not get
446 * reprogrammed due to the handler hijacking. Duh, stupid me!
447 *
448 * Both issues require major surgery and especially the kick HPET
449 * again after enabling IOAPIC results in really nasty hackery.
450 * This 'assume periodic works' magic has survived since HPET
451 * support got added, so it's questionable whether this should be
452 * fixed. Both Qemu and the failing hardware machine support
453 * periodic mode despite the fact that both don't advertise it in
454 * the configuration register and both need that extra kick after
455 * switching to IOAPIC. Seems to be a feature...
456 */
457 hc->evt.features |= CLOCK_EVT_FEAT_PERIODIC;
458 hc->evt.set_state_periodic = hpet_clkevt_set_state_periodic;
459
460 /* Start HPET legacy interrupts */
461 hpet_enable_legacy_int();
462
463 clockevents_config_and_register(&hc->evt, hpet_freq,
464 HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
465 global_clock_event = &hc->evt;
466 pr_debug("Clockevent registered\n");
467}
468
469/*
470 * HPET MSI Support
471 */
472#if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_GENERIC_MSI_IRQ)
473static void hpet_msi_unmask(struct irq_data *data)
474{
475 struct hpet_channel *hc = irq_data_get_irq_handler_data(data);
476 unsigned int cfg;
477
478 cfg = hpet_readl(HPET_Tn_CFG(hc->num));
479 cfg |= HPET_TN_ENABLE | HPET_TN_FSB;
480 hpet_writel(cfg, HPET_Tn_CFG(hc->num));
481}
482
483static void hpet_msi_mask(struct irq_data *data)
484{
485 struct hpet_channel *hc = irq_data_get_irq_handler_data(data);
486 unsigned int cfg;
487
488 cfg = hpet_readl(HPET_Tn_CFG(hc->num));
489 cfg &= ~(HPET_TN_ENABLE | HPET_TN_FSB);
490 hpet_writel(cfg, HPET_Tn_CFG(hc->num));
491}
492
493static void hpet_msi_write(struct hpet_channel *hc, struct msi_msg *msg)
494{
495 hpet_writel(msg->data, HPET_Tn_ROUTE(hc->num));
496 hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hc->num) + 4);
497}
498
499static void hpet_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
500{
501 hpet_msi_write(irq_data_get_irq_handler_data(data), msg);
502}
503
504static struct irq_chip hpet_msi_controller __ro_after_init = {
505 .name = "HPET-MSI",
506 .irq_unmask = hpet_msi_unmask,
507 .irq_mask = hpet_msi_mask,
508 .irq_ack = irq_chip_ack_parent,
509 .irq_set_affinity = msi_domain_set_affinity,
510 .irq_retrigger = irq_chip_retrigger_hierarchy,
511 .irq_write_msi_msg = hpet_msi_write_msg,
512 .flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_AFFINITY_PRE_STARTUP,
513};
514
515static int hpet_msi_init(struct irq_domain *domain,
516 struct msi_domain_info *info, unsigned int virq,
517 irq_hw_number_t hwirq, msi_alloc_info_t *arg)
518{
519 irq_set_status_flags(virq, IRQ_MOVE_PCNTXT);
520 irq_domain_set_info(domain, virq, arg->hwirq, info->chip, NULL,
521 handle_edge_irq, arg->data, "edge");
522
523 return 0;
524}
525
526static void hpet_msi_free(struct irq_domain *domain,
527 struct msi_domain_info *info, unsigned int virq)
528{
529 irq_clear_status_flags(virq, IRQ_MOVE_PCNTXT);
530}
531
532static struct msi_domain_ops hpet_msi_domain_ops = {
533 .msi_init = hpet_msi_init,
534 .msi_free = hpet_msi_free,
535};
536
537static struct msi_domain_info hpet_msi_domain_info = {
538 .ops = &hpet_msi_domain_ops,
539 .chip = &hpet_msi_controller,
540 .flags = MSI_FLAG_USE_DEF_DOM_OPS,
541};
542
543static struct irq_domain *hpet_create_irq_domain(int hpet_id)
544{
545 struct msi_domain_info *domain_info;
546 struct irq_domain *parent, *d;
547 struct fwnode_handle *fn;
548 struct irq_fwspec fwspec;
549
550 if (x86_vector_domain == NULL)
551 return NULL;
552
553 domain_info = kzalloc(sizeof(*domain_info), GFP_KERNEL);
554 if (!domain_info)
555 return NULL;
556
557 *domain_info = hpet_msi_domain_info;
558 domain_info->data = (void *)(long)hpet_id;
559
560 fn = irq_domain_alloc_named_id_fwnode(hpet_msi_controller.name,
561 hpet_id);
562 if (!fn) {
563 kfree(domain_info);
564 return NULL;
565 }
566
567 fwspec.fwnode = fn;
568 fwspec.param_count = 1;
569 fwspec.param[0] = hpet_id;
570
571 parent = irq_find_matching_fwspec(&fwspec, DOMAIN_BUS_ANY);
572 if (!parent) {
573 irq_domain_free_fwnode(fn);
574 kfree(domain_info);
575 return NULL;
576 }
577 if (parent != x86_vector_domain)
578 hpet_msi_controller.name = "IR-HPET-MSI";
579
580 d = msi_create_irq_domain(fn, domain_info, parent);
581 if (!d) {
582 irq_domain_free_fwnode(fn);
583 kfree(domain_info);
584 }
585 return d;
586}
587
588static inline int hpet_dev_id(struct irq_domain *domain)
589{
590 struct msi_domain_info *info = msi_get_domain_info(domain);
591
592 return (int)(long)info->data;
593}
594
595static int hpet_assign_irq(struct irq_domain *domain, struct hpet_channel *hc,
596 int dev_num)
597{
598 struct irq_alloc_info info;
599
600 init_irq_alloc_info(&info, NULL);
601 info.type = X86_IRQ_ALLOC_TYPE_HPET;
602 info.data = hc;
603 info.devid = hpet_dev_id(domain);
604 info.hwirq = dev_num;
605
606 return irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
607}
608
609static int hpet_clkevt_msi_resume(struct clock_event_device *evt)
610{
611 struct hpet_channel *hc = clockevent_to_channel(evt);
612 struct irq_data *data = irq_get_irq_data(hc->irq);
613 struct msi_msg msg;
614
615 /* Restore the MSI msg and unmask the interrupt */
616 irq_chip_compose_msi_msg(data, &msg);
617 hpet_msi_write(hc, &msg);
618 hpet_msi_unmask(data);
619 return 0;
620}
621
622static irqreturn_t hpet_msi_interrupt_handler(int irq, void *data)
623{
624 struct hpet_channel *hc = data;
625 struct clock_event_device *evt = &hc->evt;
626
627 if (!evt->event_handler) {
628 pr_info("Spurious interrupt HPET channel %d\n", hc->num);
629 return IRQ_HANDLED;
630 }
631
632 evt->event_handler(evt);
633 return IRQ_HANDLED;
634}
635
636static int hpet_setup_msi_irq(struct hpet_channel *hc)
637{
638 if (request_irq(hc->irq, hpet_msi_interrupt_handler,
639 IRQF_TIMER | IRQF_NOBALANCING,
640 hc->name, hc))
641 return -1;
642
643 disable_irq(hc->irq);
644 irq_set_affinity(hc->irq, cpumask_of(hc->cpu));
645 enable_irq(hc->irq);
646
647 pr_debug("%s irq %u for MSI\n", hc->name, hc->irq);
648
649 return 0;
650}
651
652/* Invoked from the hotplug callback on @cpu */
653static void init_one_hpet_msi_clockevent(struct hpet_channel *hc, int cpu)
654{
655 struct clock_event_device *evt = &hc->evt;
656
657 hc->cpu = cpu;
658 per_cpu(cpu_hpet_channel, cpu) = hc;
659 hpet_setup_msi_irq(hc);
660
661 hpet_init_clockevent(hc, 110);
662 evt->tick_resume = hpet_clkevt_msi_resume;
663
664 clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
665 0x7FFFFFFF);
666}
667
668static struct hpet_channel *hpet_get_unused_clockevent(void)
669{
670 int i;
671
672 for (i = 0; i < hpet_base.nr_channels; i++) {
673 struct hpet_channel *hc = hpet_base.channels + i;
674
675 if (hc->mode != HPET_MODE_CLOCKEVT || hc->in_use)
676 continue;
677 hc->in_use = 1;
678 return hc;
679 }
680 return NULL;
681}
682
683static int hpet_cpuhp_online(unsigned int cpu)
684{
685 struct hpet_channel *hc = hpet_get_unused_clockevent();
686
687 if (hc)
688 init_one_hpet_msi_clockevent(hc, cpu);
689 return 0;
690}
691
692static int hpet_cpuhp_dead(unsigned int cpu)
693{
694 struct hpet_channel *hc = per_cpu(cpu_hpet_channel, cpu);
695
696 if (!hc)
697 return 0;
698 free_irq(hc->irq, hc);
699 hc->in_use = 0;
700 per_cpu(cpu_hpet_channel, cpu) = NULL;
701 return 0;
702}
703
704static void __init hpet_select_clockevents(void)
705{
706 unsigned int i;
707
708 hpet_base.nr_clockevents = 0;
709
710 /* No point if MSI is disabled or CPU has an Always Running APIC Timer */
711 if (hpet_msi_disable || boot_cpu_has(X86_FEATURE_ARAT))
712 return;
713
714 hpet_print_config();
715
716 hpet_domain = hpet_create_irq_domain(hpet_blockid);
717 if (!hpet_domain)
718 return;
719
720 for (i = 0; i < hpet_base.nr_channels; i++) {
721 struct hpet_channel *hc = hpet_base.channels + i;
722 int irq;
723
724 if (hc->mode != HPET_MODE_UNUSED)
725 continue;
726
727 /* Only consider HPET channel with MSI support */
728 if (!(hc->boot_cfg & HPET_TN_FSB_CAP))
729 continue;
730
731 sprintf(hc->name, "hpet%d", i);
732
733 irq = hpet_assign_irq(hpet_domain, hc, hc->num);
734 if (irq <= 0)
735 continue;
736
737 hc->irq = irq;
738 hc->mode = HPET_MODE_CLOCKEVT;
739
740 if (++hpet_base.nr_clockevents == num_possible_cpus())
741 break;
742 }
743
744 pr_info("%d channels of %d reserved for per-cpu timers\n",
745 hpet_base.nr_channels, hpet_base.nr_clockevents);
746}
747
748#else
749
750static inline void hpet_select_clockevents(void) { }
751
752#define hpet_cpuhp_online NULL
753#define hpet_cpuhp_dead NULL
754
755#endif
756
757/*
758 * Clock source related code
759 */
760#if defined(CONFIG_SMP) && defined(CONFIG_64BIT)
761/*
762 * Reading the HPET counter is a very slow operation. If a large number of
763 * CPUs are trying to access the HPET counter simultaneously, it can cause
764 * massive delays and slow down system performance dramatically. This may
765 * happen when HPET is the default clock source instead of TSC. For a
766 * really large system with hundreds of CPUs, the slowdown may be so
767 * severe, that it can actually crash the system because of a NMI watchdog
768 * soft lockup, for example.
769 *
770 * If multiple CPUs are trying to access the HPET counter at the same time,
771 * we don't actually need to read the counter multiple times. Instead, the
772 * other CPUs can use the counter value read by the first CPU in the group.
773 *
774 * This special feature is only enabled on x86-64 systems. It is unlikely
775 * that 32-bit x86 systems will have enough CPUs to require this feature
776 * with its associated locking overhead. We also need 64-bit atomic read.
777 *
778 * The lock and the HPET value are stored together and can be read in a
779 * single atomic 64-bit read. It is explicitly assumed that arch_spinlock_t
780 * is 32 bits in size.
781 */
782union hpet_lock {
783 struct {
784 arch_spinlock_t lock;
785 u32 value;
786 };
787 u64 lockval;
788};
789
790static union hpet_lock hpet __cacheline_aligned = {
791 { .lock = __ARCH_SPIN_LOCK_UNLOCKED, },
792};
793
794static u64 read_hpet(struct clocksource *cs)
795{
796 unsigned long flags;
797 union hpet_lock old, new;
798
799 BUILD_BUG_ON(sizeof(union hpet_lock) != 8);
800
801 /*
802 * Read HPET directly if in NMI.
803 */
804 if (in_nmi())
805 return (u64)hpet_readl(HPET_COUNTER);
806
807 /*
808 * Read the current state of the lock and HPET value atomically.
809 */
810 old.lockval = READ_ONCE(hpet.lockval);
811
812 if (arch_spin_is_locked(&old.lock))
813 goto contended;
814
815 local_irq_save(flags);
816 if (arch_spin_trylock(&hpet.lock)) {
817 new.value = hpet_readl(HPET_COUNTER);
818 /*
819 * Use WRITE_ONCE() to prevent store tearing.
820 */
821 WRITE_ONCE(hpet.value, new.value);
822 arch_spin_unlock(&hpet.lock);
823 local_irq_restore(flags);
824 return (u64)new.value;
825 }
826 local_irq_restore(flags);
827
828contended:
829 /*
830 * Contended case
831 * --------------
832 * Wait until the HPET value change or the lock is free to indicate
833 * its value is up-to-date.
834 *
835 * It is possible that old.value has already contained the latest
836 * HPET value while the lock holder was in the process of releasing
837 * the lock. Checking for lock state change will enable us to return
838 * the value immediately instead of waiting for the next HPET reader
839 * to come along.
840 */
841 do {
842 cpu_relax();
843 new.lockval = READ_ONCE(hpet.lockval);
844 } while ((new.value == old.value) && arch_spin_is_locked(&new.lock));
845
846 return (u64)new.value;
847}
848#else
849/*
850 * For UP or 32-bit.
851 */
852static u64 read_hpet(struct clocksource *cs)
853{
854 return (u64)hpet_readl(HPET_COUNTER);
855}
856#endif
857
858static struct clocksource clocksource_hpet = {
859 .name = "hpet",
860 .rating = 250,
861 .read = read_hpet,
862 .mask = HPET_MASK,
863 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
864 .resume = hpet_resume_counter,
865};
866
867/*
868 * AMD SB700 based systems with spread spectrum enabled use a SMM based
869 * HPET emulation to provide proper frequency setting.
870 *
871 * On such systems the SMM code is initialized with the first HPET register
872 * access and takes some time to complete. During this time the config
873 * register reads 0xffffffff. We check for max 1000 loops whether the
874 * config register reads a non-0xffffffff value to make sure that the
875 * HPET is up and running before we proceed any further.
876 *
877 * A counting loop is safe, as the HPET access takes thousands of CPU cycles.
878 *
879 * On non-SB700 based machines this check is only done once and has no
880 * side effects.
881 */
882static bool __init hpet_cfg_working(void)
883{
884 int i;
885
886 for (i = 0; i < 1000; i++) {
887 if (hpet_readl(HPET_CFG) != 0xFFFFFFFF)
888 return true;
889 }
890
891 pr_warn("Config register invalid. Disabling HPET\n");
892 return false;
893}
894
895static bool __init hpet_counting(void)
896{
897 u64 start, now, t1;
898
899 hpet_restart_counter();
900
901 t1 = hpet_readl(HPET_COUNTER);
902 start = rdtsc();
903
904 /*
905 * We don't know the TSC frequency yet, but waiting for
906 * 200000 TSC cycles is safe:
907 * 4 GHz == 50us
908 * 1 GHz == 200us
909 */
910 do {
911 if (t1 != hpet_readl(HPET_COUNTER))
912 return true;
913 now = rdtsc();
914 } while ((now - start) < 200000UL);
915
916 pr_warn("Counter not counting. HPET disabled\n");
917 return false;
918}
919
920static bool __init mwait_pc10_supported(void)
921{
922 unsigned int eax, ebx, ecx, mwait_substates;
923
924 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
925 return false;
926
927 if (!cpu_feature_enabled(X86_FEATURE_MWAIT))
928 return false;
929
930 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
931 return false;
932
933 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
934
935 return (ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) &&
936 (ecx & CPUID5_ECX_INTERRUPT_BREAK) &&
937 (mwait_substates & (0xF << 28));
938}
939
940/*
941 * Check whether the system supports PC10. If so force disable HPET as that
942 * stops counting in PC10. This check is overbroad as it does not take any
943 * of the following into account:
944 *
945 * - ACPI tables
946 * - Enablement of intel_idle
947 * - Command line arguments which limit intel_idle C-state support
948 *
949 * That's perfectly fine. HPET is a piece of hardware designed by committee
950 * and the only reasons why it is still in use on modern systems is the
951 * fact that it is impossible to reliably query TSC and CPU frequency via
952 * CPUID or firmware.
953 *
954 * If HPET is functional it is useful for calibrating TSC, but this can be
955 * done via PMTIMER as well which seems to be the last remaining timer on
956 * X86/INTEL platforms that has not been completely wreckaged by feature
957 * creep.
958 *
959 * In theory HPET support should be removed altogether, but there are older
960 * systems out there which depend on it because TSC and APIC timer are
961 * dysfunctional in deeper C-states.
962 *
963 * It's only 20 years now that hardware people have been asked to provide
964 * reliable and discoverable facilities which can be used for timekeeping
965 * and per CPU timer interrupts.
966 *
967 * The probability that this problem is going to be solved in the
968 * foreseeable future is close to zero, so the kernel has to be cluttered
969 * with heuristics to keep up with the ever growing amount of hardware and
970 * firmware trainwrecks. Hopefully some day hardware people will understand
971 * that the approach of "This can be fixed in software" is not sustainable.
972 * Hope dies last...
973 */
974static bool __init hpet_is_pc10_damaged(void)
975{
976 unsigned long long pcfg;
977
978 /* Check whether PC10 substates are supported */
979 if (!mwait_pc10_supported())
980 return false;
981
982 /* Check whether PC10 is enabled in PKG C-state limit */
983 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, pcfg);
984 if ((pcfg & 0xF) < 8)
985 return false;
986
987 if (hpet_force_user) {
988 pr_warn("HPET force enabled via command line, but dysfunctional in PC10.\n");
989 return false;
990 }
991
992 pr_info("HPET dysfunctional in PC10. Force disabled.\n");
993 boot_hpet_disable = true;
994 return true;
995}
996
997/**
998 * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
999 */
1000int __init hpet_enable(void)
1001{
1002 u32 hpet_period, cfg, id, irq;
1003 unsigned int i, channels;
1004 struct hpet_channel *hc;
1005 u64 freq;
1006
1007 if (!is_hpet_capable())
1008 return 0;
1009
1010 if (hpet_is_pc10_damaged())
1011 return 0;
1012
1013 hpet_set_mapping();
1014 if (!hpet_virt_address)
1015 return 0;
1016
1017 /* Validate that the config register is working */
1018 if (!hpet_cfg_working())
1019 goto out_nohpet;
1020
1021 /*
1022 * Read the period and check for a sane value:
1023 */
1024 hpet_period = hpet_readl(HPET_PERIOD);
1025 if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
1026 goto out_nohpet;
1027
1028 /* The period is a femtoseconds value. Convert it to a frequency. */
1029 freq = FSEC_PER_SEC;
1030 do_div(freq, hpet_period);
1031 hpet_freq = freq;
1032
1033 /*
1034 * Read the HPET ID register to retrieve the IRQ routing
1035 * information and the number of channels
1036 */
1037 id = hpet_readl(HPET_ID);
1038 hpet_print_config();
1039
1040 /* This is the HPET channel number which is zero based */
1041 channels = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
1042
1043 /*
1044 * The legacy routing mode needs at least two channels, tick timer
1045 * and the rtc emulation channel.
1046 */
1047 if (IS_ENABLED(CONFIG_HPET_EMULATE_RTC) && channels < 2)
1048 goto out_nohpet;
1049
1050 hc = kcalloc(channels, sizeof(*hc), GFP_KERNEL);
1051 if (!hc) {
1052 pr_warn("Disabling HPET.\n");
1053 goto out_nohpet;
1054 }
1055 hpet_base.channels = hc;
1056 hpet_base.nr_channels = channels;
1057
1058 /* Read, store and sanitize the global configuration */
1059 cfg = hpet_readl(HPET_CFG);
1060 hpet_base.boot_cfg = cfg;
1061 cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
1062 hpet_writel(cfg, HPET_CFG);
1063 if (cfg)
1064 pr_warn("Global config: Unknown bits %#x\n", cfg);
1065
1066 /* Read, store and sanitize the per channel configuration */
1067 for (i = 0; i < channels; i++, hc++) {
1068 hc->num = i;
1069
1070 cfg = hpet_readl(HPET_Tn_CFG(i));
1071 hc->boot_cfg = cfg;
1072 irq = (cfg & Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
1073 hc->irq = irq;
1074
1075 cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
1076 hpet_writel(cfg, HPET_Tn_CFG(i));
1077
1078 cfg &= ~(HPET_TN_PERIODIC | HPET_TN_PERIODIC_CAP
1079 | HPET_TN_64BIT_CAP | HPET_TN_32BIT | HPET_TN_ROUTE
1080 | HPET_TN_FSB | HPET_TN_FSB_CAP);
1081 if (cfg)
1082 pr_warn("Channel #%u config: Unknown bits %#x\n", i, cfg);
1083 }
1084 hpet_print_config();
1085
1086 /*
1087 * Validate that the counter is counting. This needs to be done
1088 * after sanitizing the config registers to properly deal with
1089 * force enabled HPETs.
1090 */
1091 if (!hpet_counting())
1092 goto out_nohpet;
1093
1094 if (tsc_clocksource_watchdog_disabled())
1095 clocksource_hpet.flags |= CLOCK_SOURCE_MUST_VERIFY;
1096 clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
1097
1098 if (id & HPET_ID_LEGSUP) {
1099 hpet_legacy_clockevent_register(&hpet_base.channels[0]);
1100 hpet_base.channels[0].mode = HPET_MODE_LEGACY;
1101 if (IS_ENABLED(CONFIG_HPET_EMULATE_RTC))
1102 hpet_base.channels[1].mode = HPET_MODE_LEGACY;
1103 return 1;
1104 }
1105 return 0;
1106
1107out_nohpet:
1108 kfree(hpet_base.channels);
1109 hpet_base.channels = NULL;
1110 hpet_base.nr_channels = 0;
1111 hpet_clear_mapping();
1112 hpet_address = 0;
1113 return 0;
1114}
1115
1116/*
1117 * The late initialization runs after the PCI quirks have been invoked
1118 * which might have detected a system on which the HPET can be enforced.
1119 *
1120 * Also, the MSI machinery is not working yet when the HPET is initialized
1121 * early.
1122 *
1123 * If the HPET is enabled, then:
1124 *
1125 * 1) Reserve one channel for /dev/hpet if CONFIG_HPET=y
1126 * 2) Reserve up to num_possible_cpus() channels as per CPU clockevents
1127 * 3) Setup /dev/hpet if CONFIG_HPET=y
1128 * 4) Register hotplug callbacks when clockevents are available
1129 */
1130static __init int hpet_late_init(void)
1131{
1132 int ret;
1133
1134 if (!hpet_address) {
1135 if (!force_hpet_address)
1136 return -ENODEV;
1137
1138 hpet_address = force_hpet_address;
1139 hpet_enable();
1140 }
1141
1142 if (!hpet_virt_address)
1143 return -ENODEV;
1144
1145 hpet_select_device_channel();
1146 hpet_select_clockevents();
1147 hpet_reserve_platform_timers();
1148 hpet_print_config();
1149
1150 if (!hpet_base.nr_clockevents)
1151 return 0;
1152
1153 ret = cpuhp_setup_state(CPUHP_AP_X86_HPET_ONLINE, "x86/hpet:online",
1154 hpet_cpuhp_online, NULL);
1155 if (ret)
1156 return ret;
1157 ret = cpuhp_setup_state(CPUHP_X86_HPET_DEAD, "x86/hpet:dead", NULL,
1158 hpet_cpuhp_dead);
1159 if (ret)
1160 goto err_cpuhp;
1161 return 0;
1162
1163err_cpuhp:
1164 cpuhp_remove_state(CPUHP_AP_X86_HPET_ONLINE);
1165 return ret;
1166}
1167fs_initcall(hpet_late_init);
1168
1169void hpet_disable(void)
1170{
1171 unsigned int i;
1172 u32 cfg;
1173
1174 if (!is_hpet_capable() || !hpet_virt_address)
1175 return;
1176
1177 /* Restore boot configuration with the enable bit cleared */
1178 cfg = hpet_base.boot_cfg;
1179 cfg &= ~HPET_CFG_ENABLE;
1180 hpet_writel(cfg, HPET_CFG);
1181
1182 /* Restore the channel boot configuration */
1183 for (i = 0; i < hpet_base.nr_channels; i++)
1184 hpet_writel(hpet_base.channels[i].boot_cfg, HPET_Tn_CFG(i));
1185
1186 /* If the HPET was enabled at boot time, reenable it */
1187 if (hpet_base.boot_cfg & HPET_CFG_ENABLE)
1188 hpet_writel(hpet_base.boot_cfg, HPET_CFG);
1189}
1190
1191#ifdef CONFIG_HPET_EMULATE_RTC
1192
1193/*
1194 * HPET in LegacyReplacement mode eats up the RTC interrupt line. When HPET
1195 * is enabled, we support RTC interrupt functionality in software.
1196 *
1197 * RTC has 3 kinds of interrupts:
1198 *
1199 * 1) Update Interrupt - generate an interrupt, every second, when the
1200 * RTC clock is updated
1201 * 2) Alarm Interrupt - generate an interrupt at a specific time of day
1202 * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
1203 * 2Hz-8192Hz (2Hz-64Hz for non-root user) (all frequencies in powers of 2)
1204 *
1205 * (1) and (2) above are implemented using polling at a frequency of 64 Hz:
1206 * DEFAULT_RTC_INT_FREQ.
1207 *
1208 * The exact frequency is a tradeoff between accuracy and interrupt overhead.
1209 *
1210 * For (3), we use interrupts at 64 Hz, or the user specified periodic frequency,
1211 * if it's higher.
1212 */
1213#include <linux/mc146818rtc.h>
1214#include <linux/rtc.h>
1215
1216#define DEFAULT_RTC_INT_FREQ 64
1217#define DEFAULT_RTC_SHIFT 6
1218#define RTC_NUM_INTS 1
1219
1220static unsigned long hpet_rtc_flags;
1221static int hpet_prev_update_sec;
1222static struct rtc_time hpet_alarm_time;
1223static unsigned long hpet_pie_count;
1224static u32 hpet_t1_cmp;
1225static u32 hpet_default_delta;
1226static u32 hpet_pie_delta;
1227static unsigned long hpet_pie_limit;
1228
1229static rtc_irq_handler irq_handler;
1230
1231/*
1232 * Check that the HPET counter c1 is ahead of c2
1233 */
1234static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1235{
1236 return (s32)(c2 - c1) < 0;
1237}
1238
1239/*
1240 * Registers a IRQ handler.
1241 */
1242int hpet_register_irq_handler(rtc_irq_handler handler)
1243{
1244 if (!is_hpet_enabled())
1245 return -ENODEV;
1246 if (irq_handler)
1247 return -EBUSY;
1248
1249 irq_handler = handler;
1250
1251 return 0;
1252}
1253EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1254
1255/*
1256 * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1257 * and does cleanup.
1258 */
1259void hpet_unregister_irq_handler(rtc_irq_handler handler)
1260{
1261 if (!is_hpet_enabled())
1262 return;
1263
1264 irq_handler = NULL;
1265 hpet_rtc_flags = 0;
1266}
1267EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1268
1269/*
1270 * Channel 1 for RTC emulation. We use one shot mode, as periodic mode
1271 * is not supported by all HPET implementations for channel 1.
1272 *
1273 * hpet_rtc_timer_init() is called when the rtc is initialized.
1274 */
1275int hpet_rtc_timer_init(void)
1276{
1277 unsigned int cfg, cnt, delta;
1278 unsigned long flags;
1279
1280 if (!is_hpet_enabled())
1281 return 0;
1282
1283 if (!hpet_default_delta) {
1284 struct clock_event_device *evt = &hpet_base.channels[0].evt;
1285 uint64_t clc;
1286
1287 clc = (uint64_t) evt->mult * NSEC_PER_SEC;
1288 clc >>= evt->shift + DEFAULT_RTC_SHIFT;
1289 hpet_default_delta = clc;
1290 }
1291
1292 if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1293 delta = hpet_default_delta;
1294 else
1295 delta = hpet_pie_delta;
1296
1297 local_irq_save(flags);
1298
1299 cnt = delta + hpet_readl(HPET_COUNTER);
1300 hpet_writel(cnt, HPET_T1_CMP);
1301 hpet_t1_cmp = cnt;
1302
1303 cfg = hpet_readl(HPET_T1_CFG);
1304 cfg &= ~HPET_TN_PERIODIC;
1305 cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1306 hpet_writel(cfg, HPET_T1_CFG);
1307
1308 local_irq_restore(flags);
1309
1310 return 1;
1311}
1312EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1313
1314static void hpet_disable_rtc_channel(void)
1315{
1316 u32 cfg = hpet_readl(HPET_T1_CFG);
1317
1318 cfg &= ~HPET_TN_ENABLE;
1319 hpet_writel(cfg, HPET_T1_CFG);
1320}
1321
1322/*
1323 * The functions below are called from rtc driver.
1324 * Return 0 if HPET is not being used.
1325 * Otherwise do the necessary changes and return 1.
1326 */
1327int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1328{
1329 if (!is_hpet_enabled())
1330 return 0;
1331
1332 hpet_rtc_flags &= ~bit_mask;
1333 if (unlikely(!hpet_rtc_flags))
1334 hpet_disable_rtc_channel();
1335
1336 return 1;
1337}
1338EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1339
1340int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1341{
1342 unsigned long oldbits = hpet_rtc_flags;
1343
1344 if (!is_hpet_enabled())
1345 return 0;
1346
1347 hpet_rtc_flags |= bit_mask;
1348
1349 if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1350 hpet_prev_update_sec = -1;
1351
1352 if (!oldbits)
1353 hpet_rtc_timer_init();
1354
1355 return 1;
1356}
1357EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1358
1359int hpet_set_alarm_time(unsigned char hrs, unsigned char min, unsigned char sec)
1360{
1361 if (!is_hpet_enabled())
1362 return 0;
1363
1364 hpet_alarm_time.tm_hour = hrs;
1365 hpet_alarm_time.tm_min = min;
1366 hpet_alarm_time.tm_sec = sec;
1367
1368 return 1;
1369}
1370EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1371
1372int hpet_set_periodic_freq(unsigned long freq)
1373{
1374 uint64_t clc;
1375
1376 if (!is_hpet_enabled())
1377 return 0;
1378
1379 if (freq <= DEFAULT_RTC_INT_FREQ) {
1380 hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1381 } else {
1382 struct clock_event_device *evt = &hpet_base.channels[0].evt;
1383
1384 clc = (uint64_t) evt->mult * NSEC_PER_SEC;
1385 do_div(clc, freq);
1386 clc >>= evt->shift;
1387 hpet_pie_delta = clc;
1388 hpet_pie_limit = 0;
1389 }
1390
1391 return 1;
1392}
1393EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1394
1395int hpet_rtc_dropped_irq(void)
1396{
1397 return is_hpet_enabled();
1398}
1399EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1400
1401static void hpet_rtc_timer_reinit(void)
1402{
1403 unsigned int delta;
1404 int lost_ints = -1;
1405
1406 if (unlikely(!hpet_rtc_flags))
1407 hpet_disable_rtc_channel();
1408
1409 if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1410 delta = hpet_default_delta;
1411 else
1412 delta = hpet_pie_delta;
1413
1414 /*
1415 * Increment the comparator value until we are ahead of the
1416 * current count.
1417 */
1418 do {
1419 hpet_t1_cmp += delta;
1420 hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1421 lost_ints++;
1422 } while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1423
1424 if (lost_ints) {
1425 if (hpet_rtc_flags & RTC_PIE)
1426 hpet_pie_count += lost_ints;
1427 if (printk_ratelimit())
1428 pr_warn("Lost %d RTC interrupts\n", lost_ints);
1429 }
1430}
1431
1432irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1433{
1434 struct rtc_time curr_time;
1435 unsigned long rtc_int_flag = 0;
1436
1437 hpet_rtc_timer_reinit();
1438 memset(&curr_time, 0, sizeof(struct rtc_time));
1439
1440 if (hpet_rtc_flags & (RTC_UIE | RTC_AIE)) {
1441 if (unlikely(mc146818_get_time(&curr_time, 10) < 0)) {
1442 pr_err_ratelimited("unable to read current time from RTC\n");
1443 return IRQ_HANDLED;
1444 }
1445 }
1446
1447 if (hpet_rtc_flags & RTC_UIE &&
1448 curr_time.tm_sec != hpet_prev_update_sec) {
1449 if (hpet_prev_update_sec >= 0)
1450 rtc_int_flag = RTC_UF;
1451 hpet_prev_update_sec = curr_time.tm_sec;
1452 }
1453
1454 if (hpet_rtc_flags & RTC_PIE && ++hpet_pie_count >= hpet_pie_limit) {
1455 rtc_int_flag |= RTC_PF;
1456 hpet_pie_count = 0;
1457 }
1458
1459 if (hpet_rtc_flags & RTC_AIE &&
1460 (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1461 (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1462 (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1463 rtc_int_flag |= RTC_AF;
1464
1465 if (rtc_int_flag) {
1466 rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1467 if (irq_handler)
1468 irq_handler(rtc_int_flag, dev_id);
1469 }
1470 return IRQ_HANDLED;
1471}
1472EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1473#endif