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>
10 #include <asm/irq_remapping.h>
13 #include <asm/mwait.h>
16 #define pr_fmt(fmt) "hpet: " fmt
26 struct clock_event_device evt;
32 unsigned int boot_cfg;
37 unsigned int nr_channels;
38 unsigned int nr_clockevents;
39 unsigned int boot_cfg;
40 struct hpet_channel *channels;
43 #define HPET_MASK CLOCKSOURCE_MASK(32)
45 #define HPET_MIN_CYCLES 128
46 #define HPET_MIN_PROG_DELTA (HPET_MIN_CYCLES + (HPET_MIN_CYCLES >> 1))
49 * HPET address is set in acpi/boot.c, when an ACPI entry exists
51 unsigned long hpet_address;
52 u8 hpet_blockid; /* OS timer block num */
53 bool hpet_msi_disable;
55 #ifdef CONFIG_GENERIC_MSI_IRQ
56 static DEFINE_PER_CPU(struct hpet_channel *, cpu_hpet_channel);
57 static struct irq_domain *hpet_domain;
60 static void __iomem *hpet_virt_address;
62 static struct hpet_base hpet_base;
64 static bool hpet_legacy_int_enabled;
65 static unsigned long hpet_freq;
67 bool boot_hpet_disable;
69 static bool hpet_verbose;
72 struct hpet_channel *clockevent_to_channel(struct clock_event_device *evt)
74 return container_of(evt, struct hpet_channel, evt);
77 inline unsigned int hpet_readl(unsigned int a)
79 return readl(hpet_virt_address + a);
82 static inline void hpet_writel(unsigned int d, unsigned int a)
84 writel(d, hpet_virt_address + a);
87 static inline void hpet_set_mapping(void)
89 hpet_virt_address = ioremap(hpet_address, HPET_MMAP_SIZE);
92 static inline void hpet_clear_mapping(void)
94 iounmap(hpet_virt_address);
95 hpet_virt_address = NULL;
99 * HPET command line enable / disable
101 static int __init hpet_setup(char *str)
104 char *next = strchr(str, ',');
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))
118 __setup("hpet=", hpet_setup);
120 static int __init disable_hpet(char *str)
122 boot_hpet_disable = true;
125 __setup("nohpet", disable_hpet);
127 static inline int is_hpet_capable(void)
129 return !boot_hpet_disable && hpet_address;
133 * is_hpet_enabled - Check whether the legacy HPET timer interrupt is enabled
135 int is_hpet_enabled(void)
137 return is_hpet_capable() && hpet_legacy_int_enabled;
139 EXPORT_SYMBOL_GPL(is_hpet_enabled);
141 static void _hpet_print_config(const char *function, int line)
143 u32 i, id, period, cfg, status, channels, l, h;
145 pr_info("%s(%d):\n", function, line);
147 id = hpet_readl(HPET_ID);
148 period = hpet_readl(HPET_PERIOD);
149 pr_info("ID: 0x%x, PERIOD: 0x%x\n", id, period);
151 cfg = hpet_readl(HPET_CFG);
152 status = hpet_readl(HPET_STATUS);
153 pr_info("CFG: 0x%x, STATUS: 0x%x\n", cfg, status);
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);
159 channels = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
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);
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);
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);
176 #define hpet_print_config() \
179 _hpet_print_config(__func__, __LINE__); \
183 * When the HPET driver (/dev/hpet) is enabled, we need to reserve
184 * timer 0 and timer 1 in case of RTC emulation.
188 static void __init hpet_reserve_platform_timers(void)
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;
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.
203 hd.hd_irq[0] = HPET_LEGACY_8254;
204 hd.hd_irq[1] = HPET_LEGACY_RTC;
206 for (i = 0; i < hpet_base.nr_channels; i++) {
207 struct hpet_channel *hc = hpet_base.channels + i;
210 hd.hd_irq[i] = hc->irq;
213 case HPET_MODE_UNUSED:
214 case HPET_MODE_DEVICE:
215 hc->mode = HPET_MODE_DEVICE;
217 case HPET_MODE_CLOCKEVT:
218 case HPET_MODE_LEGACY:
219 hpet_reserve_timer(&hd, hc->num);
227 static void __init hpet_select_device_channel(void)
231 for (i = 0; i < hpet_base.nr_channels; i++) {
232 struct hpet_channel *hc = hpet_base.channels + i;
234 /* Associate the first unused channel to /dev/hpet */
235 if (hc->mode == HPET_MODE_UNUSED) {
236 hc->mode = HPET_MODE_DEVICE;
243 static inline void hpet_reserve_platform_timers(void) { }
244 static inline void hpet_select_device_channel(void) {}
247 /* Common HPET functions */
248 static void hpet_stop_counter(void)
250 u32 cfg = hpet_readl(HPET_CFG);
252 cfg &= ~HPET_CFG_ENABLE;
253 hpet_writel(cfg, HPET_CFG);
256 static void hpet_reset_counter(void)
258 hpet_writel(0, HPET_COUNTER);
259 hpet_writel(0, HPET_COUNTER + 4);
262 static void hpet_start_counter(void)
264 unsigned int cfg = hpet_readl(HPET_CFG);
266 cfg |= HPET_CFG_ENABLE;
267 hpet_writel(cfg, HPET_CFG);
270 static void hpet_restart_counter(void)
273 hpet_reset_counter();
274 hpet_start_counter();
277 static void hpet_resume_device(void)
282 static void hpet_resume_counter(struct clocksource *cs)
284 hpet_resume_device();
285 hpet_restart_counter();
288 static void hpet_enable_legacy_int(void)
290 unsigned int cfg = hpet_readl(HPET_CFG);
292 cfg |= HPET_CFG_LEGACY;
293 hpet_writel(cfg, HPET_CFG);
294 hpet_legacy_int_enabled = true;
297 static int hpet_clkevt_set_state_periodic(struct clock_event_device *evt)
299 unsigned int channel = clockevent_to_channel(evt)->num;
300 unsigned int cfg, cmp, now;
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 |
311 hpet_writel(cfg, HPET_Tn_CFG(channel));
312 hpet_writel(cmp, HPET_Tn_CMP(channel));
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)
321 hpet_writel((unsigned int)delta, HPET_Tn_CMP(channel));
322 hpet_start_counter();
328 static int hpet_clkevt_set_state_oneshot(struct clock_event_device *evt)
330 unsigned int channel = clockevent_to_channel(evt)->num;
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));
341 static int hpet_clkevt_set_state_shutdown(struct clock_event_device *evt)
343 unsigned int channel = clockevent_to_channel(evt)->num;
346 cfg = hpet_readl(HPET_Tn_CFG(channel));
347 cfg &= ~HPET_TN_ENABLE;
348 hpet_writel(cfg, HPET_Tn_CFG(channel));
353 static int hpet_clkevt_legacy_resume(struct clock_event_device *evt)
355 hpet_enable_legacy_int();
361 hpet_clkevt_set_next_event(unsigned long delta, struct clock_event_device *evt)
363 unsigned int channel = clockevent_to_channel(evt)->num;
367 cnt = hpet_readl(HPET_COUNTER);
369 hpet_writel(cnt, HPET_Tn_CMP(channel));
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.
393 res = (s32)(cnt - hpet_readl(HPET_COUNTER));
395 return res < HPET_MIN_CYCLES ? -ETIME : 0;
398 static void hpet_init_clockevent(struct hpet_channel *hc, unsigned int rating)
400 struct clock_event_device *evt = &hc->evt;
402 evt->rating = rating;
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;
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;
417 static void __init hpet_legacy_clockevent_register(struct hpet_channel *hc)
420 * Start HPET with the boot CPU's cpumask and make it global after
421 * the IO_APIC has been initialized.
423 hc->cpu = boot_cpu_data.cpu_index;
424 strncpy(hc->name, "hpet", sizeof(hc->name));
425 hpet_init_clockevent(hc, 50);
427 hc->evt.tick_resume = hpet_clkevt_legacy_resume;
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:
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:
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!
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...
457 hc->evt.features |= CLOCK_EVT_FEAT_PERIODIC;
458 hc->evt.set_state_periodic = hpet_clkevt_set_state_periodic;
460 /* Start HPET legacy interrupts */
461 hpet_enable_legacy_int();
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");
472 #ifdef CONFIG_GENERIC_MSI_IRQ
473 static void hpet_msi_unmask(struct irq_data *data)
475 struct hpet_channel *hc = irq_data_get_irq_handler_data(data);
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));
483 static void hpet_msi_mask(struct irq_data *data)
485 struct hpet_channel *hc = irq_data_get_irq_handler_data(data);
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));
493 static void hpet_msi_write(struct hpet_channel *hc, struct msi_msg *msg)
495 hpet_writel(msg->data, HPET_Tn_ROUTE(hc->num));
496 hpet_writel(msg->address_lo, HPET_Tn_ROUTE(hc->num) + 4);
499 static void hpet_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
501 hpet_msi_write(irq_data_get_irq_handler_data(data), msg);
504 static struct irq_chip hpet_msi_controller __ro_after_init = {
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,
515 static 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)
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");
526 static void hpet_msi_free(struct irq_domain *domain,
527 struct msi_domain_info *info, unsigned int virq)
529 irq_clear_status_flags(virq, IRQ_MOVE_PCNTXT);
532 static struct msi_domain_ops hpet_msi_domain_ops = {
533 .msi_init = hpet_msi_init,
534 .msi_free = hpet_msi_free,
537 static 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,
543 static struct irq_domain *hpet_create_irq_domain(int hpet_id)
545 struct msi_domain_info *domain_info;
546 struct irq_domain *parent, *d;
547 struct fwnode_handle *fn;
548 struct irq_fwspec fwspec;
550 if (x86_vector_domain == NULL)
553 domain_info = kzalloc(sizeof(*domain_info), GFP_KERNEL);
557 *domain_info = hpet_msi_domain_info;
558 domain_info->data = (void *)(long)hpet_id;
560 fn = irq_domain_alloc_named_id_fwnode(hpet_msi_controller.name,
568 fwspec.param_count = 1;
569 fwspec.param[0] = hpet_id;
571 parent = irq_find_matching_fwspec(&fwspec, DOMAIN_BUS_ANY);
573 irq_domain_free_fwnode(fn);
577 if (parent != x86_vector_domain)
578 hpet_msi_controller.name = "IR-HPET-MSI";
580 d = msi_create_irq_domain(fn, domain_info, parent);
582 irq_domain_free_fwnode(fn);
588 static inline int hpet_dev_id(struct irq_domain *domain)
590 struct msi_domain_info *info = msi_get_domain_info(domain);
592 return (int)(long)info->data;
595 static int hpet_assign_irq(struct irq_domain *domain, struct hpet_channel *hc,
598 struct irq_alloc_info info;
600 init_irq_alloc_info(&info, NULL);
601 info.type = X86_IRQ_ALLOC_TYPE_HPET;
603 info.devid = hpet_dev_id(domain);
604 info.hwirq = dev_num;
606 return irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
609 static int hpet_clkevt_msi_resume(struct clock_event_device *evt)
611 struct hpet_channel *hc = clockevent_to_channel(evt);
612 struct irq_data *data = irq_get_irq_data(hc->irq);
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);
622 static irqreturn_t hpet_msi_interrupt_handler(int irq, void *data)
624 struct hpet_channel *hc = data;
625 struct clock_event_device *evt = &hc->evt;
627 if (!evt->event_handler) {
628 pr_info("Spurious interrupt HPET channel %d\n", hc->num);
632 evt->event_handler(evt);
636 static int hpet_setup_msi_irq(struct hpet_channel *hc)
638 if (request_irq(hc->irq, hpet_msi_interrupt_handler,
639 IRQF_TIMER | IRQF_NOBALANCING,
643 disable_irq(hc->irq);
644 irq_set_affinity(hc->irq, cpumask_of(hc->cpu));
647 pr_debug("%s irq %u for MSI\n", hc->name, hc->irq);
652 /* Invoked from the hotplug callback on @cpu */
653 static void init_one_hpet_msi_clockevent(struct hpet_channel *hc, int cpu)
655 struct clock_event_device *evt = &hc->evt;
658 per_cpu(cpu_hpet_channel, cpu) = hc;
659 hpet_setup_msi_irq(hc);
661 hpet_init_clockevent(hc, 110);
662 evt->tick_resume = hpet_clkevt_msi_resume;
664 clockevents_config_and_register(evt, hpet_freq, HPET_MIN_PROG_DELTA,
668 static struct hpet_channel *hpet_get_unused_clockevent(void)
672 for (i = 0; i < hpet_base.nr_channels; i++) {
673 struct hpet_channel *hc = hpet_base.channels + i;
675 if (hc->mode != HPET_MODE_CLOCKEVT || hc->in_use)
683 static int hpet_cpuhp_online(unsigned int cpu)
685 struct hpet_channel *hc = hpet_get_unused_clockevent();
688 init_one_hpet_msi_clockevent(hc, cpu);
692 static int hpet_cpuhp_dead(unsigned int cpu)
694 struct hpet_channel *hc = per_cpu(cpu_hpet_channel, cpu);
698 free_irq(hc->irq, hc);
700 per_cpu(cpu_hpet_channel, cpu) = NULL;
704 static void __init hpet_select_clockevents(void)
708 hpet_base.nr_clockevents = 0;
710 /* No point if MSI is disabled or CPU has an Always Runing APIC Timer */
711 if (hpet_msi_disable || boot_cpu_has(X86_FEATURE_ARAT))
716 hpet_domain = hpet_create_irq_domain(hpet_blockid);
720 for (i = 0; i < hpet_base.nr_channels; i++) {
721 struct hpet_channel *hc = hpet_base.channels + i;
724 if (hc->mode != HPET_MODE_UNUSED)
727 /* Only consider HPET channel with MSI support */
728 if (!(hc->boot_cfg & HPET_TN_FSB_CAP))
731 sprintf(hc->name, "hpet%d", i);
733 irq = hpet_assign_irq(hpet_domain, hc, hc->num);
738 hc->mode = HPET_MODE_CLOCKEVT;
740 if (++hpet_base.nr_clockevents == num_possible_cpus())
744 pr_info("%d channels of %d reserved for per-cpu timers\n",
745 hpet_base.nr_channels, hpet_base.nr_clockevents);
750 static inline void hpet_select_clockevents(void) { }
752 #define hpet_cpuhp_online NULL
753 #define hpet_cpuhp_dead NULL
758 * Clock source related code
760 #if defined(CONFIG_SMP) && defined(CONFIG_64BIT)
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.
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.
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.
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.
784 arch_spinlock_t lock;
790 static union hpet_lock hpet __cacheline_aligned = {
791 { .lock = __ARCH_SPIN_LOCK_UNLOCKED, },
794 static u64 read_hpet(struct clocksource *cs)
797 union hpet_lock old, new;
799 BUILD_BUG_ON(sizeof(union hpet_lock) != 8);
802 * Read HPET directly if in NMI.
805 return (u64)hpet_readl(HPET_COUNTER);
808 * Read the current state of the lock and HPET value atomically.
810 old.lockval = READ_ONCE(hpet.lockval);
812 if (arch_spin_is_locked(&old.lock))
815 local_irq_save(flags);
816 if (arch_spin_trylock(&hpet.lock)) {
817 new.value = hpet_readl(HPET_COUNTER);
819 * Use WRITE_ONCE() to prevent store tearing.
821 WRITE_ONCE(hpet.value, new.value);
822 arch_spin_unlock(&hpet.lock);
823 local_irq_restore(flags);
824 return (u64)new.value;
826 local_irq_restore(flags);
832 * Wait until the HPET value change or the lock is free to indicate
833 * its value is up-to-date.
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
843 new.lockval = READ_ONCE(hpet.lockval);
844 } while ((new.value == old.value) && arch_spin_is_locked(&new.lock));
846 return (u64)new.value;
852 static u64 read_hpet(struct clocksource *cs)
854 return (u64)hpet_readl(HPET_COUNTER);
858 static struct clocksource clocksource_hpet = {
863 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
864 .resume = hpet_resume_counter,
868 * AMD SB700 based systems with spread spectrum enabled use a SMM based
869 * HPET emulation to provide proper frequency setting.
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.
877 * A counting loop is safe, as the HPET access takes thousands of CPU cycles.
879 * On non-SB700 based machines this check is only done once and has no
882 static bool __init hpet_cfg_working(void)
886 for (i = 0; i < 1000; i++) {
887 if (hpet_readl(HPET_CFG) != 0xFFFFFFFF)
891 pr_warn("Config register invalid. Disabling HPET\n");
895 static bool __init hpet_counting(void)
899 hpet_restart_counter();
901 t1 = hpet_readl(HPET_COUNTER);
905 * We don't know the TSC frequency yet, but waiting for
906 * 200000 TSC cycles is safe:
911 if (t1 != hpet_readl(HPET_COUNTER))
914 } while ((now - start) < 200000UL);
916 pr_warn("Counter not counting. HPET disabled\n");
920 static bool __init mwait_pc10_supported(void)
922 unsigned int eax, ebx, ecx, mwait_substates;
924 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
927 if (!cpu_feature_enabled(X86_FEATURE_MWAIT))
930 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
933 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
935 return (ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) &&
936 (ecx & CPUID5_ECX_INTERRUPT_BREAK) &&
937 (mwait_substates & (0xF << 28));
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:
946 * - Enablement of intel_idle
947 * - Command line arguments which limit intel_idle C-state support
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
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
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.
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.
967 * The probability that this problem is going to be solved in the
968 * forseeable 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.
974 static bool __init hpet_is_pc10_damaged(void)
976 unsigned long long pcfg;
978 /* Check whether PC10 substates are supported */
979 if (!mwait_pc10_supported())
982 /* Check whether PC10 is enabled in PKG C-state limit */
983 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, pcfg);
984 if ((pcfg & 0xF) < 8)
987 if (hpet_force_user) {
988 pr_warn("HPET force enabled via command line, but dysfunctional in PC10.\n");
992 pr_info("HPET dysfunctional in PC10. Force disabled.\n");
993 boot_hpet_disable = true;
998 * hpet_enable - Try to setup the HPET timer. Returns 1 on success.
1000 int __init hpet_enable(void)
1002 u32 hpet_period, cfg, id, irq;
1003 unsigned int i, channels;
1004 struct hpet_channel *hc;
1007 if (!is_hpet_capable())
1010 if (hpet_is_pc10_damaged())
1014 if (!hpet_virt_address)
1017 /* Validate that the config register is working */
1018 if (!hpet_cfg_working())
1022 * Read the period and check for a sane value:
1024 hpet_period = hpet_readl(HPET_PERIOD);
1025 if (hpet_period < HPET_MIN_PERIOD || hpet_period > HPET_MAX_PERIOD)
1028 /* The period is a femtoseconds value. Convert it to a frequency. */
1029 freq = FSEC_PER_SEC;
1030 do_div(freq, hpet_period);
1034 * Read the HPET ID register to retrieve the IRQ routing
1035 * information and the number of channels
1037 id = hpet_readl(HPET_ID);
1038 hpet_print_config();
1040 /* This is the HPET channel number which is zero based */
1041 channels = ((id & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT) + 1;
1044 * The legacy routing mode needs at least two channels, tick timer
1045 * and the rtc emulation channel.
1047 if (IS_ENABLED(CONFIG_HPET_EMULATE_RTC) && channels < 2)
1050 hc = kcalloc(channels, sizeof(*hc), GFP_KERNEL);
1052 pr_warn("Disabling HPET.\n");
1055 hpet_base.channels = hc;
1056 hpet_base.nr_channels = channels;
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);
1064 pr_warn("Global config: Unknown bits %#x\n", cfg);
1066 /* Read, store and sanitize the per channel configuration */
1067 for (i = 0; i < channels; i++, hc++) {
1070 cfg = hpet_readl(HPET_Tn_CFG(i));
1072 irq = (cfg & Tn_INT_ROUTE_CNF_MASK) >> Tn_INT_ROUTE_CNF_SHIFT;
1075 cfg &= ~(HPET_TN_ENABLE | HPET_TN_LEVEL | HPET_TN_FSB);
1076 hpet_writel(cfg, HPET_Tn_CFG(i));
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);
1082 pr_warn("Channel #%u config: Unknown bits %#x\n", i, cfg);
1084 hpet_print_config();
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.
1091 if (!hpet_counting())
1094 clocksource_register_hz(&clocksource_hpet, (u32)hpet_freq);
1096 if (id & HPET_ID_LEGSUP) {
1097 hpet_legacy_clockevent_register(&hpet_base.channels[0]);
1098 hpet_base.channels[0].mode = HPET_MODE_LEGACY;
1099 if (IS_ENABLED(CONFIG_HPET_EMULATE_RTC))
1100 hpet_base.channels[1].mode = HPET_MODE_LEGACY;
1106 kfree(hpet_base.channels);
1107 hpet_base.channels = NULL;
1108 hpet_base.nr_channels = 0;
1109 hpet_clear_mapping();
1115 * The late initialization runs after the PCI quirks have been invoked
1116 * which might have detected a system on which the HPET can be enforced.
1118 * Also, the MSI machinery is not working yet when the HPET is initialized
1121 * If the HPET is enabled, then:
1123 * 1) Reserve one channel for /dev/hpet if CONFIG_HPET=y
1124 * 2) Reserve up to num_possible_cpus() channels as per CPU clockevents
1125 * 3) Setup /dev/hpet if CONFIG_HPET=y
1126 * 4) Register hotplug callbacks when clockevents are available
1128 static __init int hpet_late_init(void)
1132 if (!hpet_address) {
1133 if (!force_hpet_address)
1136 hpet_address = force_hpet_address;
1140 if (!hpet_virt_address)
1143 hpet_select_device_channel();
1144 hpet_select_clockevents();
1145 hpet_reserve_platform_timers();
1146 hpet_print_config();
1148 if (!hpet_base.nr_clockevents)
1151 ret = cpuhp_setup_state(CPUHP_AP_X86_HPET_ONLINE, "x86/hpet:online",
1152 hpet_cpuhp_online, NULL);
1155 ret = cpuhp_setup_state(CPUHP_X86_HPET_DEAD, "x86/hpet:dead", NULL,
1162 cpuhp_remove_state(CPUHP_AP_X86_HPET_ONLINE);
1165 fs_initcall(hpet_late_init);
1167 void hpet_disable(void)
1172 if (!is_hpet_capable() || !hpet_virt_address)
1175 /* Restore boot configuration with the enable bit cleared */
1176 cfg = hpet_base.boot_cfg;
1177 cfg &= ~HPET_CFG_ENABLE;
1178 hpet_writel(cfg, HPET_CFG);
1180 /* Restore the channel boot configuration */
1181 for (i = 0; i < hpet_base.nr_channels; i++)
1182 hpet_writel(hpet_base.channels[i].boot_cfg, HPET_Tn_CFG(i));
1184 /* If the HPET was enabled at boot time, reenable it */
1185 if (hpet_base.boot_cfg & HPET_CFG_ENABLE)
1186 hpet_writel(hpet_base.boot_cfg, HPET_CFG);
1189 #ifdef CONFIG_HPET_EMULATE_RTC
1192 * HPET in LegacyReplacement mode eats up the RTC interrupt line. When HPET
1193 * is enabled, we support RTC interrupt functionality in software.
1195 * RTC has 3 kinds of interrupts:
1197 * 1) Update Interrupt - generate an interrupt, every second, when the
1198 * RTC clock is updated
1199 * 2) Alarm Interrupt - generate an interrupt at a specific time of day
1200 * 3) Periodic Interrupt - generate periodic interrupt, with frequencies
1201 * 2Hz-8192Hz (2Hz-64Hz for non-root user) (all frequencies in powers of 2)
1203 * (1) and (2) above are implemented using polling at a frequency of 64 Hz:
1204 * DEFAULT_RTC_INT_FREQ.
1206 * The exact frequency is a tradeoff between accuracy and interrupt overhead.
1208 * For (3), we use interrupts at 64 Hz, or the user specified periodic frequency,
1211 #include <linux/mc146818rtc.h>
1212 #include <linux/rtc.h>
1214 #define DEFAULT_RTC_INT_FREQ 64
1215 #define DEFAULT_RTC_SHIFT 6
1216 #define RTC_NUM_INTS 1
1218 static unsigned long hpet_rtc_flags;
1219 static int hpet_prev_update_sec;
1220 static struct rtc_time hpet_alarm_time;
1221 static unsigned long hpet_pie_count;
1222 static u32 hpet_t1_cmp;
1223 static u32 hpet_default_delta;
1224 static u32 hpet_pie_delta;
1225 static unsigned long hpet_pie_limit;
1227 static rtc_irq_handler irq_handler;
1230 * Check that the HPET counter c1 is ahead of c2
1232 static inline int hpet_cnt_ahead(u32 c1, u32 c2)
1234 return (s32)(c2 - c1) < 0;
1238 * Registers a IRQ handler.
1240 int hpet_register_irq_handler(rtc_irq_handler handler)
1242 if (!is_hpet_enabled())
1247 irq_handler = handler;
1251 EXPORT_SYMBOL_GPL(hpet_register_irq_handler);
1254 * Deregisters the IRQ handler registered with hpet_register_irq_handler()
1257 void hpet_unregister_irq_handler(rtc_irq_handler handler)
1259 if (!is_hpet_enabled())
1265 EXPORT_SYMBOL_GPL(hpet_unregister_irq_handler);
1268 * Channel 1 for RTC emulation. We use one shot mode, as periodic mode
1269 * is not supported by all HPET implementations for channel 1.
1271 * hpet_rtc_timer_init() is called when the rtc is initialized.
1273 int hpet_rtc_timer_init(void)
1275 unsigned int cfg, cnt, delta;
1276 unsigned long flags;
1278 if (!is_hpet_enabled())
1281 if (!hpet_default_delta) {
1282 struct clock_event_device *evt = &hpet_base.channels[0].evt;
1285 clc = (uint64_t) evt->mult * NSEC_PER_SEC;
1286 clc >>= evt->shift + DEFAULT_RTC_SHIFT;
1287 hpet_default_delta = clc;
1290 if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1291 delta = hpet_default_delta;
1293 delta = hpet_pie_delta;
1295 local_irq_save(flags);
1297 cnt = delta + hpet_readl(HPET_COUNTER);
1298 hpet_writel(cnt, HPET_T1_CMP);
1301 cfg = hpet_readl(HPET_T1_CFG);
1302 cfg &= ~HPET_TN_PERIODIC;
1303 cfg |= HPET_TN_ENABLE | HPET_TN_32BIT;
1304 hpet_writel(cfg, HPET_T1_CFG);
1306 local_irq_restore(flags);
1310 EXPORT_SYMBOL_GPL(hpet_rtc_timer_init);
1312 static void hpet_disable_rtc_channel(void)
1314 u32 cfg = hpet_readl(HPET_T1_CFG);
1316 cfg &= ~HPET_TN_ENABLE;
1317 hpet_writel(cfg, HPET_T1_CFG);
1321 * The functions below are called from rtc driver.
1322 * Return 0 if HPET is not being used.
1323 * Otherwise do the necessary changes and return 1.
1325 int hpet_mask_rtc_irq_bit(unsigned long bit_mask)
1327 if (!is_hpet_enabled())
1330 hpet_rtc_flags &= ~bit_mask;
1331 if (unlikely(!hpet_rtc_flags))
1332 hpet_disable_rtc_channel();
1336 EXPORT_SYMBOL_GPL(hpet_mask_rtc_irq_bit);
1338 int hpet_set_rtc_irq_bit(unsigned long bit_mask)
1340 unsigned long oldbits = hpet_rtc_flags;
1342 if (!is_hpet_enabled())
1345 hpet_rtc_flags |= bit_mask;
1347 if ((bit_mask & RTC_UIE) && !(oldbits & RTC_UIE))
1348 hpet_prev_update_sec = -1;
1351 hpet_rtc_timer_init();
1355 EXPORT_SYMBOL_GPL(hpet_set_rtc_irq_bit);
1357 int hpet_set_alarm_time(unsigned char hrs, unsigned char min, unsigned char sec)
1359 if (!is_hpet_enabled())
1362 hpet_alarm_time.tm_hour = hrs;
1363 hpet_alarm_time.tm_min = min;
1364 hpet_alarm_time.tm_sec = sec;
1368 EXPORT_SYMBOL_GPL(hpet_set_alarm_time);
1370 int hpet_set_periodic_freq(unsigned long freq)
1374 if (!is_hpet_enabled())
1377 if (freq <= DEFAULT_RTC_INT_FREQ) {
1378 hpet_pie_limit = DEFAULT_RTC_INT_FREQ / freq;
1380 struct clock_event_device *evt = &hpet_base.channels[0].evt;
1382 clc = (uint64_t) evt->mult * NSEC_PER_SEC;
1385 hpet_pie_delta = clc;
1391 EXPORT_SYMBOL_GPL(hpet_set_periodic_freq);
1393 int hpet_rtc_dropped_irq(void)
1395 return is_hpet_enabled();
1397 EXPORT_SYMBOL_GPL(hpet_rtc_dropped_irq);
1399 static void hpet_rtc_timer_reinit(void)
1404 if (unlikely(!hpet_rtc_flags))
1405 hpet_disable_rtc_channel();
1407 if (!(hpet_rtc_flags & RTC_PIE) || hpet_pie_limit)
1408 delta = hpet_default_delta;
1410 delta = hpet_pie_delta;
1413 * Increment the comparator value until we are ahead of the
1417 hpet_t1_cmp += delta;
1418 hpet_writel(hpet_t1_cmp, HPET_T1_CMP);
1420 } while (!hpet_cnt_ahead(hpet_t1_cmp, hpet_readl(HPET_COUNTER)));
1423 if (hpet_rtc_flags & RTC_PIE)
1424 hpet_pie_count += lost_ints;
1425 if (printk_ratelimit())
1426 pr_warn("Lost %d RTC interrupts\n", lost_ints);
1430 irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id)
1432 struct rtc_time curr_time;
1433 unsigned long rtc_int_flag = 0;
1435 hpet_rtc_timer_reinit();
1436 memset(&curr_time, 0, sizeof(struct rtc_time));
1438 if (hpet_rtc_flags & (RTC_UIE | RTC_AIE)) {
1439 if (unlikely(mc146818_get_time(&curr_time) < 0)) {
1440 pr_err_ratelimited("unable to read current time from RTC\n");
1445 if (hpet_rtc_flags & RTC_UIE &&
1446 curr_time.tm_sec != hpet_prev_update_sec) {
1447 if (hpet_prev_update_sec >= 0)
1448 rtc_int_flag = RTC_UF;
1449 hpet_prev_update_sec = curr_time.tm_sec;
1452 if (hpet_rtc_flags & RTC_PIE && ++hpet_pie_count >= hpet_pie_limit) {
1453 rtc_int_flag |= RTC_PF;
1457 if (hpet_rtc_flags & RTC_AIE &&
1458 (curr_time.tm_sec == hpet_alarm_time.tm_sec) &&
1459 (curr_time.tm_min == hpet_alarm_time.tm_min) &&
1460 (curr_time.tm_hour == hpet_alarm_time.tm_hour))
1461 rtc_int_flag |= RTC_AF;
1464 rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8));
1466 irq_handler(rtc_int_flag, dev_id);
1470 EXPORT_SYMBOL_GPL(hpet_rtc_interrupt);