Upgrade to 4.4.50-rt62
[kvmfornfv.git] / kernel / arch / x86 / kernel / hpet.c
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/irqdomain.h>
16 #include <asm/fixmap.h>
17 #include <asm/hpet.h>
18 #include <asm/time.h>
19
20 #define HPET_MASK                       CLOCKSOURCE_MASK(32)
21
22 /* FSEC = 10^-15
23    NSEC = 10^-9 */
24 #define FSEC_PER_NSEC                   1000000L
25
26 #define HPET_DEV_USED_BIT               2
27 #define HPET_DEV_USED                   (1 << HPET_DEV_USED_BIT)
28 #define HPET_DEV_VALID                  0x8
29 #define HPET_DEV_FSB_CAP                0x1000
30 #define HPET_DEV_PERI_CAP               0x2000
31
32 #define HPET_MIN_CYCLES                 128
33 #define HPET_MIN_PROG_DELTA             (HPET_MIN_CYCLES + (HPET_MIN_CYCLES >> 1))
34
35 /*
36  * HPET address is set in acpi/boot.c, when an ACPI entry exists
37  */
38 unsigned long                           hpet_address;
39 u8                                      hpet_blockid; /* OS timer block num */
40 bool                                    hpet_msi_disable;
41
42 #ifdef CONFIG_PCI_MSI
43 static unsigned int                     hpet_num_timers;
44 #endif
45 static void __iomem                     *hpet_virt_address;
46
47 struct hpet_dev {
48         struct clock_event_device       evt;
49         unsigned int                    num;
50         int                             cpu;
51         unsigned int                    irq;
52         unsigned int                    flags;
53         char                            name[10];
54 };
55
56 inline struct hpet_dev *EVT_TO_HPET_DEV(struct clock_event_device *evtdev)
57 {
58         return container_of(evtdev, struct hpet_dev, evt);
59 }
60
61 inline unsigned int hpet_readl(unsigned int a)
62 {
63         return readl(hpet_virt_address + a);
64 }
65
66 static inline void hpet_writel(unsigned int d, unsigned int a)
67 {
68         writel(d, hpet_virt_address + a);
69 }
70
71 #ifdef CONFIG_X86_64
72 #include <asm/pgtable.h>
73 #endif
74
75 static inline void hpet_set_mapping(void)
76 {
77         hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
78 }
79
80 static inline void hpet_clear_mapping(void)
81 {
82         iounmap(hpet_virt_address);
83         hpet_virt_address = NULL;
84 }
85
86 /*
87  * HPET command line enable / disable
88  */
89 bool boot_hpet_disable;
90 bool hpet_force_user;
91 static bool hpet_verbose;
92
93 static int __init hpet_setup(char *str)
94 {
95         while (str) {
96                 char *next = strchr(str, ',');
97
98                 if (next)
99                         *next++ = 0;
100                 if (!strncmp("disable", str, 7))
101                         boot_hpet_disable = true;
102                 if (!strncmp("force", str, 5))
103                         hpet_force_user = true;
104                 if (!strncmp("verbose", str, 7))
105                         hpet_verbose = true;
106                 str = next;
107         }
108         return 1;
109 }
110 __setup("hpet=", hpet_setup);
111
112 static int __init disable_hpet(char *str)
113 {
114         boot_hpet_disable = true;
115         return 1;
116 }
117 __setup("nohpet", disable_hpet);
118
119 static inline int is_hpet_capable(void)
120 {
121         return !boot_hpet_disable && hpet_address;
122 }
123
124 /*
125  * HPET timer interrupt enable / disable
126  */
127 static bool hpet_legacy_int_enabled;
128
129 /**
130  * is_hpet_enabled - check whether the hpet timer interrupt is enabled
131  */
132 int is_hpet_enabled(void)
133 {
134         return is_hpet_capable() && hpet_legacy_int_enabled;
135 }
136 EXPORT_SYMBOL_GPL(is_hpet_enabled);
137
138 static void _hpet_print_config(const char *function, int line)
139 {
140         u32 i, timers, l, h;
141         printk(KERN_INFO "hpet: %s(%d):\n", function, line);
142         l = hpet_readl(HPET_ID);
143         h = hpet_readl(HPET_PERIOD);
144         timers = ((l & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
145         printk(KERN_INFO "hpet: ID: 0x%x, PERIOD: 0x%x\n", l, h);
146         l = hpet_readl(HPET_CFG);
147         h = hpet_readl(HPET_STATUS);
148         printk(KERN_INFO "hpet: CFG: 0x%x, STATUS: 0x%x\n", l, h);
149         l = hpet_readl(HPET_COUNTER);
150         h = hpet_readl(HPET_COUNTER+4);
151         printk(KERN_INFO "hpet: COUNTER_l: 0x%x, COUNTER_h: 0x%x\n", l, h);
152
153         for (i = 0; i < timers; i++) {
154                 l = hpet_readl(HPET_Tn_CFG(i));
155                 h = hpet_readl(HPET_Tn_CFG(i)+4);
156                 printk(KERN_INFO "hpet: T%d: CFG_l: 0x%x, CFG_h: 0x%x\n",
157                        i, l, h);
158                 l = hpet_readl(HPET_Tn_CMP(i));
159                 h = hpet_readl(HPET_Tn_CMP(i)+4);
160                 printk(KERN_INFO "hpet: T%d: CMP_l: 0x%x, CMP_h: 0x%x\n",
161                        i, l, h);
162                 l = hpet_readl(HPET_Tn_ROUTE(i));
163                 h = hpet_readl(HPET_Tn_ROUTE(i)+4);
164                 printk(KERN_INFO "hpet: T%d ROUTE_l: 0x%x, ROUTE_h: 0x%x\n",
165                        i, l, h);
166         }
167 }
168
169 #define hpet_print_config()                                     \
170 do {                                                            \
171         if (hpet_verbose)                                       \
172                 _hpet_print_config(__func__, __LINE__); \
173 } while (0)
174
175 /*
176  * When the hpet driver (/dev/hpet) is enabled, we need to reserve
177  * timer 0 and timer 1 in case of RTC emulation.
178  */
179 #ifdef CONFIG_HPET
180
181 static void hpet_reserve_msi_timers(struct hpet_data *hd);
182
183 static void hpet_reserve_platform_timers(unsigned int id)
184 {
185         struct hpet __iomem *hpet = hpet_virt_address;
186         struct hpet_timer __iomem *timer = &hpet->hpet_timers[2];
187         unsigned int nrtimers, i;
188         struct hpet_data hd;
189
190         nrtimers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
191
192         memset(&hd, 0, sizeof(hd));
193         hd.hd_phys_address      = hpet_address;
194         hd.hd_address           = hpet;
195         hd.hd_nirqs             = nrtimers;
196         hpet_reserve_timer(&hd, 0);
197
198 #ifdef CONFIG_HPET_EMULATE_RTC
199         hpet_reserve_timer(&hd, 1);
200 #endif
201
202         /*
203          * NOTE that hd_irq[] reflects IOAPIC input pins (LEGACY_8254
204          * is wrong for i8259!) not the output IRQ.  Many BIOS writers
205          * don't bother configuring *any* comparator interrupts.
206          */
207         hd.hd_irq[0] = HPET_LEGACY_8254;
208         hd.hd_irq[1] = HPET_LEGACY_RTC;
209
210         for (i = 2; i < nrtimers; timer++, i++) {
211                 hd.hd_irq[i] = (readl(&timer->hpet_config) &
212                         Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
213         }
214
215         hpet_reserve_msi_timers(&hd);
216
217         hpet_alloc(&hd);
218
219 }
220 #else
221 static void hpet_reserve_platform_timers(unsigned int id) { }
222 #endif
223
224 /*
225  * Common hpet info
226  */
227 static unsigned long hpet_freq;
228
229 static struct clock_event_device hpet_clockevent;
230
231 static void hpet_stop_counter(void)
232 {
233         u32 cfg = hpet_readl(HPET_CFG);
234         cfg &= ~HPET_CFG_ENABLE;
235         hpet_writel(cfg, HPET_CFG);
236 }
237
238 static void hpet_reset_counter(void)
239 {
240         hpet_writel(0, HPET_COUNTER);
241         hpet_writel(0, HPET_COUNTER + 4);
242 }
243
244 static void hpet_start_counter(void)
245 {
246         unsigned int cfg = hpet_readl(HPET_CFG);
247         cfg |= HPET_CFG_ENABLE;
248         hpet_writel(cfg, HPET_CFG);
249 }
250
251 static void hpet_restart_counter(void)
252 {
253         hpet_stop_counter();
254         hpet_reset_counter();
255         hpet_start_counter();
256 }
257
258 static void hpet_resume_device(void)
259 {
260         force_hpet_resume();
261 }
262
263 static void hpet_resume_counter(struct clocksource *cs)
264 {
265         hpet_resume_device();
266         hpet_restart_counter();
267 }
268
269 static void hpet_enable_legacy_int(void)
270 {
271         unsigned int cfg = hpet_readl(HPET_CFG);
272
273         cfg |= HPET_CFG_LEGACY;
274         hpet_writel(cfg, HPET_CFG);
275         hpet_legacy_int_enabled = true;
276 }
277
278 static void hpet_legacy_clockevent_register(void)
279 {
280         /* Start HPET legacy interrupts */
281         hpet_enable_legacy_int();
282
283         /*
284          * Start hpet with the boot cpu mask and make it
285          * global after the IO_APIC has been initialized.
286          */
287         hpet_clockevent.cpumask = cpumask_of(smp_processor_id());
288         clockevents_config_and_register(&hpet_clockevent, hpet_freq,
289                                         HPET_MIN_PROG_DELTA, 0x7FFFFFFF);
290         global_clock_event = &hpet_clockevent;
291         printk(KERN_DEBUG "hpet clockevent registered\n");
292 }
293
294 static int hpet_set_periodic(struct clock_event_device *evt, int timer)
295 {
296         unsigned int cfg, cmp, now;
297         uint64_t delta;
298
299         hpet_stop_counter();
300         delta = ((uint64_t)(NSEC_PER_SEC / HZ)) * evt->mult;
301         delta >>= evt->shift;
302         now = hpet_readl(HPET_COUNTER);
303         cmp = now + (unsigned int)delta;
304         cfg = hpet_readl(HPET_Tn_CFG(timer));
305         cfg |= HPET_TN_ENABLE | HPET_TN_PERIODIC | HPET_TN_SETVAL |
306                HPET_TN_32BIT;
307         hpet_writel(cfg, HPET_Tn_CFG(timer));
308         hpet_writel(cmp, HPET_Tn_CMP(timer));
309         udelay(1);
310         /*
311          * HPET on AMD 81xx needs a second write (with HPET_TN_SETVAL
312          * cleared) to T0_CMP to set the period. The HPET_TN_SETVAL
313          * bit is automatically cleared after the first write.
314          * (See AMD-8111 HyperTransport I/O Hub Data Sheet,
315          * Publication # 24674)
316          */
317         hpet_writel((unsigned int)delta, HPET_Tn_CMP(timer));
318         hpet_start_counter();
319         hpet_print_config();
320
321         return 0;
322 }
323
324 static int hpet_set_oneshot(struct clock_event_device *evt, int timer)
325 {
326         unsigned int cfg;
327
328         cfg = hpet_readl(HPET_Tn_CFG(timer));
329         cfg &= ~HPET_TN_PERIODIC;
330         cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
331         hpet_writel(cfg, HPET_Tn_CFG(timer));
332
333         return 0;
334 }
335
336 static int hpet_shutdown(struct clock_event_device *evt, int timer)
337 {
338         unsigned int cfg;
339
340         cfg = hpet_readl(HPET_Tn_CFG(timer));
341         cfg &= ~HPET_TN_ENABLE;
342         hpet_writel(cfg, HPET_Tn_CFG(timer));
343
344         return 0;
345 }
346
347 static int hpet_resume(struct clock_event_device *evt, int timer)
348 {
349         if (!timer) {
350                 hpet_enable_legacy_int();
351         } else {
352                 struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
353
354                 irq_domain_deactivate_irq(irq_get_irq_data(hdev->irq));
355                 irq_domain_activate_irq(irq_get_irq_data(hdev->irq));
356                 disable_irq(hdev->irq);
357                 irq_set_affinity(hdev->irq, cpumask_of(hdev->cpu));
358                 enable_irq(hdev->irq);
359         }
360         hpet_print_config();
361
362         return 0;
363 }
364
365 static int hpet_next_event(unsigned long delta,
366                            struct clock_event_device *evt, int timer)
367 {
368         u32 cnt;
369         s32 res;
370
371         cnt = hpet_readl(HPET_COUNTER);
372         cnt += (u32) delta;
373         hpet_writel(cnt, HPET_Tn_CMP(timer));
374
375         /*
376          * HPETs are a complete disaster. The compare register is
377          * based on a equal comparison and neither provides a less
378          * than or equal functionality (which would require to take
379          * the wraparound into account) nor a simple count down event
380          * mode. Further the write to the comparator register is
381          * delayed internally up to two HPET clock cycles in certain
382          * chipsets (ATI, ICH9,10). Some newer AMD chipsets have even
383          * longer delays. We worked around that by reading back the
384          * compare register, but that required another workaround for
385          * ICH9,10 chips where the first readout after write can
386          * return the old stale value. We already had a minimum
387          * programming delta of 5us enforced, but a NMI or SMI hitting
388          * between the counter readout and the comparator write can
389          * move us behind that point easily. Now instead of reading
390          * the compare register back several times, we make the ETIME
391          * decision based on the following: Return ETIME if the
392          * counter value after the write is less than HPET_MIN_CYCLES
393          * away from the event or if the counter is already ahead of
394          * the event. The minimum programming delta for the generic
395          * clockevents code is set to 1.5 * HPET_MIN_CYCLES.
396          */
397         res = (s32)(cnt - hpet_readl(HPET_COUNTER));
398
399         return res < HPET_MIN_CYCLES ? -ETIME : 0;
400 }
401
402 static int hpet_legacy_shutdown(struct clock_event_device *evt)
403 {
404         return hpet_shutdown(evt, 0);
405 }
406
407 static int hpet_legacy_set_oneshot(struct clock_event_device *evt)
408 {
409         return hpet_set_oneshot(evt, 0);
410 }
411
412 static int hpet_legacy_set_periodic(struct clock_event_device *evt)
413 {
414         return hpet_set_periodic(evt, 0);
415 }
416
417 static int hpet_legacy_resume(struct clock_event_device *evt)
418 {
419         return hpet_resume(evt, 0);
420 }
421
422 static int hpet_legacy_next_event(unsigned long delta,
423                         struct clock_event_device *evt)
424 {
425         return hpet_next_event(delta, evt, 0);
426 }
427
428 /*
429  * The hpet clock event device
430  */
431 static struct clock_event_device hpet_clockevent = {
432         .name                   = "hpet",
433         .features               = CLOCK_EVT_FEAT_PERIODIC |
434                                   CLOCK_EVT_FEAT_ONESHOT,
435         .set_state_periodic     = hpet_legacy_set_periodic,
436         .set_state_oneshot      = hpet_legacy_set_oneshot,
437         .set_state_shutdown     = hpet_legacy_shutdown,
438         .tick_resume            = hpet_legacy_resume,
439         .set_next_event         = hpet_legacy_next_event,
440         .irq                    = 0,
441         .rating                 = 50,
442 };
443
444 /*
445  * HPET MSI Support
446  */
447 #ifdef CONFIG_PCI_MSI
448
449 static DEFINE_PER_CPU(struct hpet_dev *, cpu_hpet_dev);
450 static struct hpet_dev  *hpet_devs;
451 static struct irq_domain *hpet_domain;
452
453 void hpet_msi_unmask(struct irq_data *data)
454 {
455         struct hpet_dev *hdev = irq_data_get_irq_handler_data(data);
456         unsigned int cfg;
457
458         /* unmask it */
459         cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
460         cfg |= HPET_TN_ENABLE | HPET_TN_FSB;
461         hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
462 }
463
464 void hpet_msi_mask(struct irq_data *data)
465 {
466         struct hpet_dev *hdev = irq_data_get_irq_handler_data(data);
467         unsigned int cfg;
468
469         /* mask it */
470         cfg = hpet_readl(HPET_Tn_CFG(hdev->num));
471         cfg &= ~(HPET_TN_ENABLE | HPET_TN_FSB);
472         hpet_writel(cfg, HPET_Tn_CFG(hdev->num));
473 }
474
475 void hpet_msi_write(struct hpet_dev *hdev, struct msi_msg *msg)
476 {
477         hpet_writel(msg->data, HPET_Tn_ROUTE(hdev->num));
478         hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hdev->num) + 4);
479 }
480
481 void hpet_msi_read(struct hpet_dev *hdev, struct msi_msg *msg)
482 {
483         msg->data = hpet_readl(HPET_Tn_ROUTE(hdev->num));
484         msg->address_lo = hpet_readl(HPET_Tn_ROUTE(hdev->num) + 4);
485         msg->address_hi = 0;
486 }
487
488 static int hpet_msi_shutdown(struct clock_event_device *evt)
489 {
490         struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
491
492         return hpet_shutdown(evt, hdev->num);
493 }
494
495 static int hpet_msi_set_oneshot(struct clock_event_device *evt)
496 {
497         struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
498
499         return hpet_set_oneshot(evt, hdev->num);
500 }
501
502 static int hpet_msi_set_periodic(struct clock_event_device *evt)
503 {
504         struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
505
506         return hpet_set_periodic(evt, hdev->num);
507 }
508
509 static int hpet_msi_resume(struct clock_event_device *evt)
510 {
511         struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
512
513         return hpet_resume(evt, hdev->num);
514 }
515
516 static int hpet_msi_next_event(unsigned long delta,
517                                 struct clock_event_device *evt)
518 {
519         struct hpet_dev *hdev = EVT_TO_HPET_DEV(evt);
520         return hpet_next_event(delta, evt, hdev->num);
521 }
522
523 static irqreturn_t hpet_interrupt_handler(int irq, void *data)
524 {
525         struct hpet_dev *dev = (struct hpet_dev *)data;
526         struct clock_event_device *hevt = &dev->evt;
527
528         if (!hevt->event_handler) {
529                 printk(KERN_INFO "Spurious HPET timer interrupt on HPET timer %d\n",
530                                 dev->num);
531                 return IRQ_HANDLED;
532         }
533
534         hevt->event_handler(hevt);
535         return IRQ_HANDLED;
536 }
537
538 static int hpet_setup_irq(struct hpet_dev *dev)
539 {
540
541         if (request_irq(dev->irq, hpet_interrupt_handler,
542                         IRQF_TIMER | IRQF_NOBALANCING,
543                         dev->name, dev))
544                 return -1;
545
546         disable_irq(dev->irq);
547         irq_set_affinity(dev->irq, cpumask_of(dev->cpu));
548         enable_irq(dev->irq);
549
550         printk(KERN_DEBUG "hpet: %s irq %d for MSI\n",
551                          dev->name, dev->irq);
552
553         return 0;
554 }
555
556 /* This should be called in specific @cpu */
557 static void init_one_hpet_msi_clockevent(struct hpet_dev *hdev, int cpu)
558 {
559         struct clock_event_device *evt = &hdev->evt;
560
561         WARN_ON(cpu != smp_processor_id());
562         if (!(hdev->flags & HPET_DEV_VALID))
563                 return;
564
565         hdev->cpu = cpu;
566         per_cpu(cpu_hpet_dev, cpu) = hdev;
567         evt->name = hdev->name;
568         hpet_setup_irq(hdev);
569         evt->irq = hdev->irq;
570
571         evt->rating = 110;
572         evt->features = CLOCK_EVT_FEAT_ONESHOT;
573         if (hdev->flags & HPET_DEV_PERI_CAP) {
574                 evt->features |= CLOCK_EVT_FEAT_PERIODIC;
575                 evt->set_state_periodic = hpet_msi_set_periodic;
576         }
577
578         evt->set_state_shutdown = hpet_msi_shutdown;
579         evt->set_state_oneshot = hpet_msi_set_oneshot;
580         evt->tick_resume = hpet_msi_resume;
581         evt->set_next_event = hpet_msi_next_event;
582         evt->cpumask = cpumask_of(hdev->cpu);
583
584         clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
585                                         0x7FFFFFFF);
586 }
587
588 #ifdef CONFIG_HPET
589 /* Reserve at least one timer for userspace (/dev/hpet) */
590 #define RESERVE_TIMERS 1
591 #else
592 #define RESERVE_TIMERS 0
593 #endif
594
595 static void hpet_msi_capability_lookup(unsigned int start_timer)
596 {
597         unsigned int id;
598         unsigned int num_timers;
599         unsigned int num_timers_used = 0;
600         int i, irq;
601
602         if (hpet_msi_disable)
603                 return;
604
605         if (boot_cpu_has(X86_FEATURE_ARAT))
606                 return;
607         id = hpet_readl(HPET_ID);
608
609         num_timers = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
610         num_timers++; /* Value read out starts from 0 */
611         hpet_print_config();
612
613         hpet_domain = hpet_create_irq_domain(hpet_blockid);
614         if (!hpet_domain)
615                 return;
616
617         hpet_devs = kzalloc(sizeof(struct hpet_dev) * num_timers, GFP_KERNEL);
618         if (!hpet_devs)
619                 return;
620
621         hpet_num_timers = num_timers;
622
623         for (i = start_timer; i < num_timers - RESERVE_TIMERS; i++) {
624                 struct hpet_dev *hdev = &hpet_devs[num_timers_used];
625                 unsigned int cfg = hpet_readl(HPET_Tn_CFG(i));
626
627                 /* Only consider HPET timer with MSI support */
628                 if (!(cfg & HPET_TN_FSB_CAP))
629                         continue;
630
631                 hdev->flags = 0;
632                 if (cfg & HPET_TN_PERIODIC_CAP)
633                         hdev->flags |= HPET_DEV_PERI_CAP;
634                 sprintf(hdev->name, "hpet%d", i);
635                 hdev->num = i;
636
637                 irq = hpet_assign_irq(hpet_domain, hdev, hdev->num);
638                 if (irq <= 0)
639                         continue;
640
641                 hdev->irq = irq;
642                 hdev->flags |= HPET_DEV_FSB_CAP;
643                 hdev->flags |= HPET_DEV_VALID;
644                 num_timers_used++;
645                 if (num_timers_used == num_possible_cpus())
646                         break;
647         }
648
649         printk(KERN_INFO "HPET: %d timers in total, %d timers will be used for per-cpu timer\n",
650                 num_timers, num_timers_used);
651 }
652
653 #ifdef CONFIG_HPET
654 static void hpet_reserve_msi_timers(struct hpet_data *hd)
655 {
656         int i;
657
658         if (!hpet_devs)
659                 return;
660
661         for (i = 0; i < hpet_num_timers; i++) {
662                 struct hpet_dev *hdev = &hpet_devs[i];
663
664                 if (!(hdev->flags & HPET_DEV_VALID))
665                         continue;
666
667                 hd->hd_irq[hdev->num] = hdev->irq;
668                 hpet_reserve_timer(hd, hdev->num);
669         }
670 }
671 #endif
672
673 static struct hpet_dev *hpet_get_unused_timer(void)
674 {
675         int i;
676
677         if (!hpet_devs)
678                 return NULL;
679
680         for (i = 0; i < hpet_num_timers; i++) {
681                 struct hpet_dev *hdev = &hpet_devs[i];
682
683                 if (!(hdev->flags & HPET_DEV_VALID))
684                         continue;
685                 if (test_and_set_bit(HPET_DEV_USED_BIT,
686                         (unsigned long *)&hdev->flags))
687                         continue;
688                 return hdev;
689         }
690         return NULL;
691 }
692
693 struct hpet_work_struct {
694         struct delayed_work work;
695         struct completion complete;
696 };
697
698 static void hpet_work(struct work_struct *w)
699 {
700         struct hpet_dev *hdev;
701         int cpu = smp_processor_id();
702         struct hpet_work_struct *hpet_work;
703
704         hpet_work = container_of(w, struct hpet_work_struct, work.work);
705
706         hdev = hpet_get_unused_timer();
707         if (hdev)
708                 init_one_hpet_msi_clockevent(hdev, cpu);
709
710         complete(&hpet_work->complete);
711 }
712
713 static int hpet_cpuhp_notify(struct notifier_block *n,
714                 unsigned long action, void *hcpu)
715 {
716         unsigned long cpu = (unsigned long)hcpu;
717         struct hpet_work_struct work;
718         struct hpet_dev *hdev = per_cpu(cpu_hpet_dev, cpu);
719
720         switch (action & 0xf) {
721         case CPU_ONLINE:
722                 INIT_DELAYED_WORK_ONSTACK(&work.work, hpet_work);
723                 init_completion(&work.complete);
724                 /* FIXME: add schedule_work_on() */
725                 schedule_delayed_work_on(cpu, &work.work, 0);
726                 wait_for_completion(&work.complete);
727                 destroy_delayed_work_on_stack(&work.work);
728                 break;
729         case CPU_DEAD:
730                 if (hdev) {
731                         free_irq(hdev->irq, hdev);
732                         hdev->flags &= ~HPET_DEV_USED;
733                         per_cpu(cpu_hpet_dev, cpu) = NULL;
734                 }
735                 break;
736         }
737         return NOTIFY_OK;
738 }
739 #else
740
741 static void hpet_msi_capability_lookup(unsigned int start_timer)
742 {
743         return;
744 }
745
746 #ifdef CONFIG_HPET
747 static void hpet_reserve_msi_timers(struct hpet_data *hd)
748 {
749         return;
750 }
751 #endif
752
753 static int hpet_cpuhp_notify(struct notifier_block *n,
754                 unsigned long action, void *hcpu)
755 {
756         return NOTIFY_OK;
757 }
758
759 #endif
760
761 /*
762  * Clock source related code
763  */
764 static cycle_t read_hpet(struct clocksource *cs)
765 {
766         return (cycle_t)hpet_readl(HPET_COUNTER);
767 }
768
769 static struct clocksource clocksource_hpet = {
770         .name           = "hpet",
771         .rating         = 250,
772         .read           = read_hpet,
773         .mask           = HPET_MASK,
774         .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
775         .resume         = hpet_resume_counter,
776         .archdata       = { .vclock_mode = VCLOCK_HPET },
777 };
778
779 static int hpet_clocksource_register(void)
780 {
781         u64 start, now;
782         cycle_t t1;
783
784         /* Start the counter */
785         hpet_restart_counter();
786
787         /* Verify whether hpet counter works */
788         t1 = hpet_readl(HPET_COUNTER);
789         start = rdtsc();
790
791         /*
792          * We don't know the TSC frequency yet, but waiting for
793          * 200000 TSC cycles is safe:
794          * 4 GHz == 50us
795          * 1 GHz == 200us
796          */
797         do {
798                 rep_nop();
799                 now = rdtsc();
800         } while ((now - start) < 200000UL);
801
802         if (t1 == hpet_readl(HPET_COUNTER)) {
803                 printk(KERN_WARNING
804                        "HPET counter not counting. HPET disabled\n");
805                 return -ENODEV;
806         }
807
808         clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
809         return 0;
810 }
811
812 static u32 *hpet_boot_cfg;
813
814 /**
815  * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
816  */
817 int __init hpet_enable(void)
818 {
819         u32 hpet_period, cfg, id;
820         u64 freq;
821         unsigned int i, last;
822
823         if (!is_hpet_capable())
824                 return 0;
825
826         hpet_set_mapping();
827
828         /*
829          * Read the period and check for a sane value:
830          */
831         hpet_period = hpet_readl(HPET_PERIOD);
832
833         /*
834          * AMD SB700 based systems with spread spectrum enabled use a
835          * SMM based HPET emulation to provide proper frequency
836          * setting. The SMM code is initialized with the first HPET
837          * register access and takes some time to complete. During
838          * this time the config register reads 0xffffffff. We check
839          * for max. 1000 loops whether the config register reads a non
840          * 0xffffffff value to make sure that HPET is up and running
841          * before we go further. A counting loop is safe, as the HPET
842          * access takes thousands of CPU cycles. On non SB700 based
843          * machines this check is only done once and has no side
844          * effects.
845          */
846         for (i = 0; hpet_readl(HPET_CFG) == 0xFFFFFFFF; i++) {
847                 if (i == 1000) {
848                         printk(KERN_WARNING
849                                "HPET config register value = 0xFFFFFFFF. "
850                                "Disabling HPET\n");
851                         goto out_nohpet;
852                 }
853         }
854
855         if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
856                 goto out_nohpet;
857
858         /*
859          * The period is a femto seconds value. Convert it to a
860          * frequency.
861          */
862         freq = FSEC_PER_SEC;
863         do_div(freq, hpet_period);
864         hpet_freq = freq;
865
866         /*
867          * Read the HPET ID register to retrieve the IRQ routing
868          * information and the number of channels
869          */
870         id = hpet_readl(HPET_ID);
871         hpet_print_config();
872
873         last = (id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT;
874
875 #ifdef CONFIG_HPET_EMULATE_RTC
876         /*
877          * The legacy routing mode needs at least two channels, tick timer
878          * and the rtc emulation channel.
879          */
880         if (!last)
881                 goto out_nohpet;
882 #endif
883
884         cfg = hpet_readl(HPET_CFG);
885         hpet_boot_cfg = kmalloc((last + 2) * sizeof(*hpet_boot_cfg),
886                                 GFP_KERNEL);
887         if (hpet_boot_cfg)
888                 *hpet_boot_cfg = cfg;
889         else
890                 pr_warn("HPET initial state will not be saved\n");
891         cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY);
892         hpet_writel(cfg, HPET_CFG);
893         if (cfg)
894                 pr_warn("HPET: Unrecognized bits %#x set in global cfg\n",
895                         cfg);
896
897         for (i = 0; i <= last; ++i) {
898                 cfg = hpet_readl(HPET_Tn_CFG(i));
899                 if (hpet_boot_cfg)
900                         hpet_boot_cfg[i + 1] = cfg;
901                 cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
902                 hpet_writel(cfg, HPET_Tn_CFG(i));
903                 cfg &= ~(HPET_TN_PERIODIC | HPET_TN_PERIODIC_CAP
904                          | HPET_TN_64BIT_CAP | HPET_TN_32BIT | HPET_TN_ROUTE
905                          | HPET_TN_FSB | HPET_TN_FSB_CAP);
906                 if (cfg)
907                         pr_warn("HPET: Unrecognized bits %#x set in cfg#%u\n",
908                                 cfg, i);
909         }
910         hpet_print_config();
911
912         if (hpet_clocksource_register())
913                 goto out_nohpet;
914
915         if (id & HPET_ID_LEGSUP) {
916                 hpet_legacy_clockevent_register();
917                 return 1;
918         }
919         return 0;
920
921 out_nohpet:
922         hpet_clear_mapping();
923         hpet_address = 0;
924         return 0;
925 }
926
927 /*
928  * Needs to be late, as the reserve_timer code calls kalloc !
929  *
930  * Not a problem on i386 as hpet_enable is called from late_time_init,
931  * but on x86_64 it is necessary !
932  */
933 static __init int hpet_late_init(void)
934 {
935         int cpu;
936
937         if (boot_hpet_disable)
938                 return -ENODEV;
939
940         if (!hpet_address) {
941                 if (!force_hpet_address)
942                         return -ENODEV;
943
944                 hpet_address = force_hpet_address;
945                 hpet_enable();
946         }
947
948         if (!hpet_virt_address)
949                 return -ENODEV;
950
951         if (hpet_readl(HPET_ID) & HPET_ID_LEGSUP)
952                 hpet_msi_capability_lookup(2);
953         else
954                 hpet_msi_capability_lookup(0);
955
956         hpet_reserve_platform_timers(hpet_readl(HPET_ID));
957         hpet_print_config();
958
959         if (hpet_msi_disable)
960                 return 0;
961
962         if (boot_cpu_has(X86_FEATURE_ARAT))
963                 return 0;
964
965         cpu_notifier_register_begin();
966         for_each_online_cpu(cpu) {
967                 hpet_cpuhp_notify(NULL, CPU_ONLINE, (void *)(long)cpu);
968         }
969
970         /* This notifier should be called after workqueue is ready */
971         __hotcpu_notifier(hpet_cpuhp_notify, -20);
972         cpu_notifier_register_done();
973
974         return 0;
975 }
976 fs_initcall(hpet_late_init);
977
978 void hpet_disable(void)
979 {
980         if (is_hpet_capable() && hpet_virt_address) {
981                 unsigned int cfg = hpet_readl(HPET_CFG), id, last;
982
983                 if (hpet_boot_cfg)
984                         cfg = *hpet_boot_cfg;
985                 else if (hpet_legacy_int_enabled) {
986                         cfg &= ~HPET_CFG_LEGACY;
987                         hpet_legacy_int_enabled = false;
988                 }
989                 cfg &= ~HPET_CFG_ENABLE;
990                 hpet_writel(cfg, HPET_CFG);
991
992                 if (!hpet_boot_cfg)
993                         return;
994
995                 id = hpet_readl(HPET_ID);
996                 last = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT);
997
998                 for (id = 0; id <= last; ++id)
999                         hpet_writel(hpet_boot_cfg[id + 1], HPET_Tn_CFG(id));
1000
1001                 if (*hpet_boot_cfg & HPET_CFG_ENABLE)
1002                         hpet_writel(*hpet_boot_cfg, HPET_CFG);
1003         }
1004 }
1005
1006 #ifdef CONFIG_HPET_EMULATE_RTC
1007
1008 /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET
1009  * is enabled, we support RTC interrupt functionality in software.
1010  * RTC has 3 kinds of interrupts:
1011  * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock
1012  *    is updated
1013  * 2) Alarm Interrupt - generate an interrupt at a specific time of day
1014  * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
1015  *    2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2)
1016  * (1) and (2) above are implemented using polling at a frequency of
1017  * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt
1018  * overhead. (DEFAULT_RTC_INT_FREQ)
1019  * For (3), we use interrupts at 64Hz or user specified periodic
1020  * frequency, whichever is higher.
1021  */
1022 #include <linux/mc146818rtc.h>
1023 #include <linux/rtc.h>
1024 #include <asm/rtc.h>
1025
1026 #define DEFAULT_RTC_INT_FREQ    64
1027 #define DEFAULT_RTC_SHIFT       6
1028 #define RTC_NUM_INTS            1
1029
1030 static unsigned long hpet_rtc_flags;
1031 static int hpet_prev_update_sec;
1032 static struct rtc_time hpet_alarm_time;
1033 static unsigned long hpet_pie_count;
1034 static u32 hpet_t1_cmp;
1035 static u32 hpet_default_delta;
1036 static u32 hpet_pie_delta;
1037 static unsigned long hpet_pie_limit;
1038
1039 static rtc_irq_handler irq_handler;
1040
1041 /*
1042  * Check that the hpet counter c1 is ahead of the c2
1043  */
1044 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1045 {
1046         return (s32)(c2 - c1) < 0;
1047 }
1048
1049 /*
1050  * Registers a IRQ handler.
1051  */
1052 int hpet_register_irq_handler(rtc_irq_handler handler)
1053 {
1054         if (!is_hpet_enabled())
1055                 return -ENODEV;
1056         if (irq_handler)
1057                 return -EBUSY;
1058
1059         irq_handler = handler;
1060
1061         return 0;
1062 }
1063 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1064
1065 /*
1066  * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1067  * and does cleanup.
1068  */
1069 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1070 {
1071         if (!is_hpet_enabled())
1072                 return;
1073
1074         irq_handler = NULL;
1075         hpet_rtc_flags = 0;
1076 }
1077 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1078
1079 /*
1080  * Timer 1 for RTC emulation. We use one shot mode, as periodic mode
1081  * is not supported by all HPET implementations for timer 1.
1082  *
1083  * hpet_rtc_timer_init() is called when the rtc is initialized.
1084  */
1085 int hpet_rtc_timer_init(void)
1086 {
1087         unsigned int cfg, cnt, delta;
1088         unsigned long flags;
1089
1090         if (!is_hpet_enabled())
1091                 return 0;
1092
1093         if (!hpet_default_delta) {
1094                 uint64_t clc;
1095
1096                 clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1097                 clc >>= hpet_clockevent.shift + DEFAULT_RTC_SHIFT;
1098                 hpet_default_delta = clc;
1099         }
1100
1101         if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1102                 delta = hpet_default_delta;
1103         else
1104                 delta = hpet_pie_delta;
1105
1106         local_irq_save(flags);
1107
1108         cnt = delta + hpet_readl(HPET_COUNTER);
1109         hpet_writel(cnt, HPET_T1_CMP);
1110         hpet_t1_cmp = cnt;
1111
1112         cfg = hpet_readl(HPET_T1_CFG);
1113         cfg &= ~HPET_TN_PERIODIC;
1114         cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1115         hpet_writel(cfg, HPET_T1_CFG);
1116
1117         local_irq_restore(flags);
1118
1119         return 1;
1120 }
1121 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1122
1123 static void hpet_disable_rtc_channel(void)
1124 {
1125         u32 cfg = hpet_readl(HPET_T1_CFG);
1126         cfg &= ~HPET_TN_ENABLE;
1127         hpet_writel(cfg, HPET_T1_CFG);
1128 }
1129
1130 /*
1131  * The functions below are called from rtc driver.
1132  * Return 0 if HPET is not being used.
1133  * Otherwise do the necessary changes and return 1.
1134  */
1135 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1136 {
1137         if (!is_hpet_enabled())
1138                 return 0;
1139
1140         hpet_rtc_flags &= ~bit_mask;
1141         if (unlikely(!hpet_rtc_flags))
1142                 hpet_disable_rtc_channel();
1143
1144         return 1;
1145 }
1146 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1147
1148 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1149 {
1150         unsigned long oldbits = hpet_rtc_flags;
1151
1152         if (!is_hpet_enabled())
1153                 return 0;
1154
1155         hpet_rtc_flags |= bit_mask;
1156
1157         if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1158                 hpet_prev_update_sec = -1;
1159
1160         if (!oldbits)
1161                 hpet_rtc_timer_init();
1162
1163         return 1;
1164 }
1165 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1166
1167 int hpet_set_alarm_time(unsigned char hrs, unsigned char min,
1168                         unsigned char sec)
1169 {
1170         if (!is_hpet_enabled())
1171                 return 0;
1172
1173         hpet_alarm_time.tm_hour = hrs;
1174         hpet_alarm_time.tm_min = min;
1175         hpet_alarm_time.tm_sec = sec;
1176
1177         return 1;
1178 }
1179 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1180
1181 int hpet_set_periodic_freq(unsigned long freq)
1182 {
1183         uint64_t clc;
1184
1185         if (!is_hpet_enabled())
1186                 return 0;
1187
1188         if (freq <= DEFAULT_RTC_INT_FREQ)
1189                 hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1190         else {
1191                 clc = (uint64_t) hpet_clockevent.mult * NSEC_PER_SEC;
1192                 do_div(clc, freq);
1193                 clc >>= hpet_clockevent.shift;
1194                 hpet_pie_delta = clc;
1195                 hpet_pie_limit = 0;
1196         }
1197         return 1;
1198 }
1199 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1200
1201 int hpet_rtc_dropped_irq(void)
1202 {
1203         return is_hpet_enabled();
1204 }
1205 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1206
1207 static void hpet_rtc_timer_reinit(void)
1208 {
1209         unsigned int delta;
1210         int lost_ints = -1;
1211
1212         if (unlikely(!hpet_rtc_flags))
1213                 hpet_disable_rtc_channel();
1214
1215         if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1216                 delta = hpet_default_delta;
1217         else
1218                 delta = hpet_pie_delta;
1219
1220         /*
1221          * Increment the comparator value until we are ahead of the
1222          * current count.
1223          */
1224         do {
1225                 hpet_t1_cmp += delta;
1226                 hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1227                 lost_ints++;
1228         } while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1229
1230         if (lost_ints) {
1231                 if (hpet_rtc_flags & RTC_PIE)
1232                         hpet_pie_count += lost_ints;
1233                 if (printk_ratelimit())
1234                         printk(KERN_WARNING "hpet1: lost %d rtc interrupts\n",
1235                                 lost_ints);
1236         }
1237 }
1238
1239 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1240 {
1241         struct rtc_time curr_time;
1242         unsigned long rtc_int_flag = 0;
1243
1244         hpet_rtc_timer_reinit();
1245         memset(&curr_time, 0, sizeof(struct rtc_time));
1246
1247         if (hpet_rtc_flags & (RTC_UIE | RTC_AIE))
1248                 get_rtc_time(&curr_time);
1249
1250         if (hpet_rtc_flags & RTC_UIE &&
1251             curr_time.tm_sec != hpet_prev_update_sec) {
1252                 if (hpet_prev_update_sec >= 0)
1253                         rtc_int_flag = RTC_UF;
1254                 hpet_prev_update_sec = curr_time.tm_sec;
1255         }
1256
1257         if (hpet_rtc_flags & RTC_PIE &&
1258             ++hpet_pie_count >= hpet_pie_limit) {
1259                 rtc_int_flag |= RTC_PF;
1260                 hpet_pie_count = 0;
1261         }
1262
1263         if (hpet_rtc_flags & RTC_AIE &&
1264             (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1265             (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1266             (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1267                         rtc_int_flag |= RTC_AF;
1268
1269         if (rtc_int_flag) {
1270                 rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1271                 if (irq_handler)
1272                         irq_handler(rtc_int_flag, dev_id);
1273         }
1274         return IRQ_HANDLED;
1275 }
1276 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);
1277 #endif