2 * Timer device implementation for SGI SN platforms.
4 * This file is subject to the terms and conditions of the GNU General Public
5 * License. See the file "COPYING" in the main directory of this archive
8 * Copyright (c) 2001-2006 Silicon Graphics, Inc. All rights reserved.
10 * This driver exports an API that should be supportable by any HPET or IA-PC
11 * multimedia timer. The code below is currently specific to the SGI Altix
14 * 11/01/01 - jbarnes - initial revision
15 * 9/10/04 - Christoph Lameter - remove interrupt support for kernel inclusion
16 * 10/1/04 - Christoph Lameter - provide posix clock CLOCK_SGI_CYCLE
17 * 10/13/04 - Christoph Lameter, Dimitri Sivanich - provide timer interrupt
18 * support via the posix timer interface
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/ioctl.h>
24 #include <linux/module.h>
25 #include <linux/init.h>
26 #include <linux/errno.h>
29 #include <linux/mmtimer.h>
30 #include <linux/miscdevice.h>
31 #include <linux/posix-timers.h>
32 #include <linux/interrupt.h>
34 #include <asm/uaccess.h>
35 #include <asm/sn/addrs.h>
36 #include <asm/sn/intr.h>
37 #include <asm/sn/shub_mmr.h>
38 #include <asm/sn/nodepda.h>
39 #include <asm/sn/shubio.h>
41 MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
42 MODULE_DESCRIPTION("SGI Altix RTC Timer");
43 MODULE_LICENSE("GPL");
45 /* name of the device, usually in /dev */
46 #define MMTIMER_NAME "mmtimer"
47 #define MMTIMER_DESC "SGI Altix RTC Timer"
48 #define MMTIMER_VERSION "2.1"
50 #define RTC_BITS 55 /* 55 bits for this implementation */
52 extern unsigned long sn_rtc_cycles_per_second;
54 #define RTC_COUNTER_ADDR ((long *)LOCAL_MMR_ADDR(SH_RTC))
56 #define rtc_time() (*RTC_COUNTER_ADDR)
58 static int mmtimer_ioctl(struct inode *inode, struct file *file,
59 unsigned int cmd, unsigned long arg);
60 static int mmtimer_mmap(struct file *file, struct vm_area_struct *vma);
63 * Period in femtoseconds (10^-15 s)
65 static unsigned long mmtimer_femtoperiod = 0;
67 static const struct file_operations mmtimer_fops = {
70 .ioctl = mmtimer_ioctl,
74 * We only have comparison registers RTC1-4 currently available per
75 * node. RTC0 is used by SAL.
77 /* Check for an RTC interrupt pending */
78 static int mmtimer_int_pending(int comparator)
80 if (HUB_L((unsigned long *)LOCAL_MMR_ADDR(SH_EVENT_OCCURRED)) &
81 SH_EVENT_OCCURRED_RTC1_INT_MASK << comparator)
87 /* Clear the RTC interrupt pending bit */
88 static void mmtimer_clr_int_pending(int comparator)
90 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_EVENT_OCCURRED_ALIAS),
91 SH_EVENT_OCCURRED_RTC1_INT_MASK << comparator);
94 /* Setup timer on comparator RTC1 */
95 static void mmtimer_setup_int_0(int cpu, u64 expires)
99 /* Disable interrupt */
100 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_ENABLE), 0UL);
102 /* Initialize comparator value */
103 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPB), -1L);
105 /* Clear pending bit */
106 mmtimer_clr_int_pending(0);
108 val = ((u64)SGI_MMTIMER_VECTOR << SH_RTC1_INT_CONFIG_IDX_SHFT) |
109 ((u64)cpu_physical_id(cpu) <<
110 SH_RTC1_INT_CONFIG_PID_SHFT);
112 /* Set configuration */
113 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_CONFIG), val);
115 /* Enable RTC interrupts */
116 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_ENABLE), 1UL);
118 /* Initialize comparator value */
119 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPB), expires);
124 /* Setup timer on comparator RTC2 */
125 static void mmtimer_setup_int_1(int cpu, u64 expires)
129 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_ENABLE), 0UL);
131 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPC), -1L);
133 mmtimer_clr_int_pending(1);
135 val = ((u64)SGI_MMTIMER_VECTOR << SH_RTC2_INT_CONFIG_IDX_SHFT) |
136 ((u64)cpu_physical_id(cpu) <<
137 SH_RTC2_INT_CONFIG_PID_SHFT);
139 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_CONFIG), val);
141 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_ENABLE), 1UL);
143 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPC), expires);
146 /* Setup timer on comparator RTC3 */
147 static void mmtimer_setup_int_2(int cpu, u64 expires)
151 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_ENABLE), 0UL);
153 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPD), -1L);
155 mmtimer_clr_int_pending(2);
157 val = ((u64)SGI_MMTIMER_VECTOR << SH_RTC3_INT_CONFIG_IDX_SHFT) |
158 ((u64)cpu_physical_id(cpu) <<
159 SH_RTC3_INT_CONFIG_PID_SHFT);
161 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_CONFIG), val);
163 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_ENABLE), 1UL);
165 HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPD), expires);
169 * This function must be called with interrupts disabled and preemption off
170 * in order to insure that the setup succeeds in a deterministic time frame.
171 * It will check if the interrupt setup succeeded.
173 static int mmtimer_setup(int cpu, int comparator, unsigned long expires)
176 switch (comparator) {
178 mmtimer_setup_int_0(cpu, expires);
181 mmtimer_setup_int_1(cpu, expires);
184 mmtimer_setup_int_2(cpu, expires);
187 /* We might've missed our expiration time */
188 if (rtc_time() <= expires)
192 * If an interrupt is already pending then its okay
193 * if not then we failed
195 return mmtimer_int_pending(comparator);
198 static int mmtimer_disable_int(long nasid, int comparator)
200 switch (comparator) {
202 nasid == -1 ? HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_ENABLE),
203 0UL) : REMOTE_HUB_S(nasid, SH_RTC1_INT_ENABLE, 0UL);
206 nasid == -1 ? HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_ENABLE),
207 0UL) : REMOTE_HUB_S(nasid, SH_RTC2_INT_ENABLE, 0UL);
210 nasid == -1 ? HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_ENABLE),
211 0UL) : REMOTE_HUB_S(nasid, SH_RTC3_INT_ENABLE, 0UL);
219 #define COMPARATOR 1 /* The comparator to use */
221 #define TIMER_OFF 0xbadcabLL /* Timer is not setup */
222 #define TIMER_SET 0 /* Comparator is set for this timer */
224 /* There is one of these for each timer */
227 struct k_itimer *timer;
231 struct mmtimer_node {
232 spinlock_t lock ____cacheline_aligned;
233 struct rb_root timer_head;
234 struct rb_node *next;
235 struct tasklet_struct tasklet;
237 static struct mmtimer_node *timers;
241 * Add a new mmtimer struct to the node's mmtimer list.
242 * This function assumes the struct mmtimer_node is locked.
244 static void mmtimer_add_list(struct mmtimer *n)
246 int nodeid = n->timer->it.mmtimer.node;
247 unsigned long expires = n->timer->it.mmtimer.expires;
248 struct rb_node **link = &timers[nodeid].timer_head.rb_node;
249 struct rb_node *parent = NULL;
253 * Find the right place in the rbtree:
257 x = rb_entry(parent, struct mmtimer, list);
259 if (expires < x->timer->it.mmtimer.expires)
260 link = &(*link)->rb_left;
262 link = &(*link)->rb_right;
266 * Insert the timer to the rbtree and check whether it
267 * replaces the first pending timer
269 rb_link_node(&n->list, parent, link);
270 rb_insert_color(&n->list, &timers[nodeid].timer_head);
272 if (!timers[nodeid].next || expires < rb_entry(timers[nodeid].next,
273 struct mmtimer, list)->timer->it.mmtimer.expires)
274 timers[nodeid].next = &n->list;
278 * Set the comparator for the next timer.
279 * This function assumes the struct mmtimer_node is locked.
281 static void mmtimer_set_next_timer(int nodeid)
283 struct mmtimer_node *n = &timers[nodeid];
292 x = rb_entry(n->next, struct mmtimer, list);
294 if (!t->it.mmtimer.incr) {
295 /* Not an interval timer */
296 if (!mmtimer_setup(x->cpu, COMPARATOR,
297 t->it.mmtimer.expires)) {
298 /* Late setup, fire now */
299 tasklet_schedule(&n->tasklet);
306 while (!mmtimer_setup(x->cpu, COMPARATOR, t->it.mmtimer.expires)) {
308 struct rb_node *next;
309 t->it.mmtimer.expires += t->it.mmtimer.incr << o;
310 t->it_overrun += 1 << o;
313 printk(KERN_ALERT "mmtimer: cannot reschedule timer\n");
314 t->it.mmtimer.clock = TIMER_OFF;
315 n->next = rb_next(&x->list);
316 rb_erase(&x->list, &n->timer_head);
321 e = t->it.mmtimer.expires;
322 next = rb_next(&x->list);
327 e1 = rb_entry(next, struct mmtimer, list)->
328 timer->it.mmtimer.expires;
331 rb_erase(&x->list, &n->timer_head);
339 * mmtimer_ioctl - ioctl interface for /dev/mmtimer
340 * @inode: inode of the device
341 * @file: file structure for the device
342 * @cmd: command to execute
343 * @arg: optional argument to command
345 * Executes the command specified by @cmd. Returns 0 for success, < 0 for
350 * %MMTIMER_GETOFFSET - Should return the offset (relative to the start
351 * of the page where the registers are mapped) for the counter in question.
353 * %MMTIMER_GETRES - Returns the resolution of the clock in femto (10^-15)
356 * %MMTIMER_GETFREQ - Copies the frequency of the clock in Hz to the address
359 * %MMTIMER_GETBITS - Returns the number of bits in the clock's counter
361 * %MMTIMER_MMAPAVAIL - Returns 1 if the registers can be mmap'd into userspace
363 * %MMTIMER_GETCOUNTER - Gets the current value in the counter and places it
364 * in the address specified by @arg.
366 static int mmtimer_ioctl(struct inode *inode, struct file *file,
367 unsigned int cmd, unsigned long arg)
372 case MMTIMER_GETOFFSET: /* offset of the counter */
374 * SN RTC registers are on their own 64k page
376 if(PAGE_SIZE <= (1 << 16))
377 ret = (((long)RTC_COUNTER_ADDR) & (PAGE_SIZE-1)) / 8;
382 case MMTIMER_GETRES: /* resolution of the clock in 10^-15 s */
383 if(copy_to_user((unsigned long __user *)arg,
384 &mmtimer_femtoperiod, sizeof(unsigned long)))
388 case MMTIMER_GETFREQ: /* frequency in Hz */
389 if(copy_to_user((unsigned long __user *)arg,
390 &sn_rtc_cycles_per_second,
391 sizeof(unsigned long)))
396 case MMTIMER_GETBITS: /* number of bits in the clock */
400 case MMTIMER_MMAPAVAIL: /* can we mmap the clock into userspace? */
401 ret = (PAGE_SIZE <= (1 << 16)) ? 1 : 0;
404 case MMTIMER_GETCOUNTER:
405 if(copy_to_user((unsigned long __user *)arg,
406 RTC_COUNTER_ADDR, sizeof(unsigned long)))
418 * mmtimer_mmap - maps the clock's registers into userspace
419 * @file: file structure for the device
420 * @vma: VMA to map the registers into
422 * Calls remap_pfn_range() to map the clock's registers into
423 * the calling process' address space.
425 static int mmtimer_mmap(struct file *file, struct vm_area_struct *vma)
427 unsigned long mmtimer_addr;
429 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
432 if (vma->vm_flags & VM_WRITE)
435 if (PAGE_SIZE > (1 << 16))
438 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
440 mmtimer_addr = __pa(RTC_COUNTER_ADDR);
441 mmtimer_addr &= ~(PAGE_SIZE - 1);
442 mmtimer_addr &= 0xfffffffffffffffUL;
444 if (remap_pfn_range(vma, vma->vm_start, mmtimer_addr >> PAGE_SHIFT,
445 PAGE_SIZE, vma->vm_page_prot)) {
446 printk(KERN_ERR "remap_pfn_range failed in mmtimer.c\n");
453 static struct miscdevice mmtimer_miscdev = {
459 static struct timespec sgi_clock_offset;
460 static int sgi_clock_period;
463 * Posix Timer Interface
466 static struct timespec sgi_clock_offset;
467 static int sgi_clock_period;
469 static int sgi_clock_get(clockid_t clockid, struct timespec *tp)
473 nsec = rtc_time() * sgi_clock_period
474 + sgi_clock_offset.tv_nsec;
475 tp->tv_sec = div_long_long_rem(nsec, NSEC_PER_SEC, &tp->tv_nsec)
476 + sgi_clock_offset.tv_sec;
480 static int sgi_clock_set(clockid_t clockid, struct timespec *tp)
486 nsec = rtc_time() * sgi_clock_period;
488 sgi_clock_offset.tv_sec = tp->tv_sec - div_long_long_rem(nsec, NSEC_PER_SEC, &rem);
490 if (rem <= tp->tv_nsec)
491 sgi_clock_offset.tv_nsec = tp->tv_sec - rem;
493 sgi_clock_offset.tv_nsec = tp->tv_sec + NSEC_PER_SEC - rem;
494 sgi_clock_offset.tv_sec--;
500 * mmtimer_interrupt - timer interrupt handler
502 * @dev_id: device the irq came from
504 * Called when one of the comarators matches the counter, This
505 * routine will send signals to processes that have requested
508 * This interrupt is run in an interrupt context
509 * by the SHUB. It is therefore safe to locally access SHub
513 mmtimer_interrupt(int irq, void *dev_id)
515 unsigned long expires = 0;
516 int result = IRQ_NONE;
517 unsigned indx = cpu_to_node(smp_processor_id());
518 struct mmtimer *base;
520 spin_lock(&timers[indx].lock);
521 base = rb_entry(timers[indx].next, struct mmtimer, list);
523 spin_unlock(&timers[indx].lock);
527 if (base->cpu == smp_processor_id()) {
529 expires = base->timer->it.mmtimer.expires;
530 /* expires test won't work with shared irqs */
531 if ((mmtimer_int_pending(COMPARATOR) > 0) ||
532 (expires && (expires <= rtc_time()))) {
533 mmtimer_clr_int_pending(COMPARATOR);
534 tasklet_schedule(&timers[indx].tasklet);
535 result = IRQ_HANDLED;
538 spin_unlock(&timers[indx].lock);
542 static void mmtimer_tasklet(unsigned long data)
545 struct mmtimer_node *mn = &timers[nodeid];
546 struct mmtimer *x = rb_entry(mn->next, struct mmtimer, list);
550 /* Send signal and deal with periodic signals */
551 spin_lock_irqsave(&mn->lock, flags);
555 x = rb_entry(mn->next, struct mmtimer, list);
558 if (t->it.mmtimer.clock == TIMER_OFF)
563 mn->next = rb_next(&x->list);
564 rb_erase(&x->list, &mn->timer_head);
566 if (posix_timer_event(t, 0) != 0)
569 if(t->it.mmtimer.incr) {
570 t->it.mmtimer.expires += t->it.mmtimer.incr;
573 /* Ensure we don't false trigger in mmtimer_interrupt */
574 t->it.mmtimer.clock = TIMER_OFF;
575 t->it.mmtimer.expires = 0;
578 /* Set comparator for next timer, if there is one */
579 mmtimer_set_next_timer(nodeid);
581 t->it_overrun_last = t->it_overrun;
583 spin_unlock_irqrestore(&mn->lock, flags);
586 static int sgi_timer_create(struct k_itimer *timer)
588 /* Insure that a newly created timer is off */
589 timer->it.mmtimer.clock = TIMER_OFF;
593 /* This does not really delete a timer. It just insures
594 * that the timer is not active
596 * Assumption: it_lock is already held with irq's disabled
598 static int sgi_timer_del(struct k_itimer *timr)
600 cnodeid_t nodeid = timr->it.mmtimer.node;
601 unsigned long irqflags;
603 spin_lock_irqsave(&timers[nodeid].lock, irqflags);
604 if (timr->it.mmtimer.clock != TIMER_OFF) {
605 unsigned long expires = timr->it.mmtimer.expires;
606 struct rb_node *n = timers[nodeid].timer_head.rb_node;
607 struct mmtimer *uninitialized_var(t);
610 timr->it.mmtimer.clock = TIMER_OFF;
611 timr->it.mmtimer.expires = 0;
614 t = rb_entry(n, struct mmtimer, list);
615 if (t->timer == timr)
618 if (expires < t->timer->it.mmtimer.expires)
625 spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
629 if (timers[nodeid].next == n) {
630 timers[nodeid].next = rb_next(n);
634 rb_erase(n, &timers[nodeid].timer_head);
638 mmtimer_disable_int(cnodeid_to_nasid(nodeid),
640 mmtimer_set_next_timer(nodeid);
643 spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
647 #define timespec_to_ns(x) ((x).tv_nsec + (x).tv_sec * NSEC_PER_SEC)
648 #define ns_to_timespec(ts, nsec) (ts).tv_sec = div_long_long_rem(nsec, NSEC_PER_SEC, &(ts).tv_nsec)
650 /* Assumption: it_lock is already held with irq's disabled */
651 static void sgi_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting)
654 if (timr->it.mmtimer.clock == TIMER_OFF) {
655 cur_setting->it_interval.tv_nsec = 0;
656 cur_setting->it_interval.tv_sec = 0;
657 cur_setting->it_value.tv_nsec = 0;
658 cur_setting->it_value.tv_sec =0;
662 ns_to_timespec(cur_setting->it_interval, timr->it.mmtimer.incr * sgi_clock_period);
663 ns_to_timespec(cur_setting->it_value, (timr->it.mmtimer.expires - rtc_time())* sgi_clock_period);
668 static int sgi_timer_set(struct k_itimer *timr, int flags,
669 struct itimerspec * new_setting,
670 struct itimerspec * old_setting)
672 unsigned long when, period, irqflags;
675 struct mmtimer *base;
679 sgi_timer_get(timr, old_setting);
682 when = timespec_to_ns(new_setting->it_value);
683 period = timespec_to_ns(new_setting->it_interval);
689 base = kmalloc(sizeof(struct mmtimer), GFP_KERNEL);
693 if (flags & TIMER_ABSTIME) {
698 now = timespec_to_ns(n);
702 /* Fire the timer immediately */
707 * Convert to sgi clock period. Need to keep rtc_time() as near as possible
708 * to getnstimeofday() in order to be as faithful as possible to the time
711 when = (when + sgi_clock_period - 1) / sgi_clock_period + rtc_time();
712 period = (period + sgi_clock_period - 1) / sgi_clock_period;
715 * We are allocating a local SHub comparator. If we would be moved to another
716 * cpu then another SHub may be local to us. Prohibit that by switching off
721 nodeid = cpu_to_node(smp_processor_id());
723 /* Lock the node timer structure */
724 spin_lock_irqsave(&timers[nodeid].lock, irqflags);
727 base->cpu = smp_processor_id();
729 timr->it.mmtimer.clock = TIMER_SET;
730 timr->it.mmtimer.node = nodeid;
731 timr->it.mmtimer.incr = period;
732 timr->it.mmtimer.expires = when;
734 n = timers[nodeid].next;
736 /* Add the new struct mmtimer to node's timer list */
737 mmtimer_add_list(base);
739 if (timers[nodeid].next == n) {
740 /* No need to reprogram comparator for now */
741 spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
746 /* We need to reprogram the comparator */
748 mmtimer_disable_int(cnodeid_to_nasid(nodeid), COMPARATOR);
750 mmtimer_set_next_timer(nodeid);
752 /* Unlock the node timer structure */
753 spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
760 static struct k_clock sgi_clock = {
762 .clock_set = sgi_clock_set,
763 .clock_get = sgi_clock_get,
764 .timer_create = sgi_timer_create,
765 .nsleep = do_posix_clock_nonanosleep,
766 .timer_set = sgi_timer_set,
767 .timer_del = sgi_timer_del,
768 .timer_get = sgi_timer_get
772 * mmtimer_init - device initialization routine
774 * Does initial setup for the mmtimer device.
776 static int __init mmtimer_init(void)
778 cnodeid_t node, maxn = -1;
780 if (!ia64_platform_is("sn2"))
784 * Sanity check the cycles/sec variable
786 if (sn_rtc_cycles_per_second < 100000) {
787 printk(KERN_ERR "%s: unable to determine clock frequency\n",
792 mmtimer_femtoperiod = ((unsigned long)1E15 + sn_rtc_cycles_per_second /
793 2) / sn_rtc_cycles_per_second;
795 if (request_irq(SGI_MMTIMER_VECTOR, mmtimer_interrupt, IRQF_PERCPU, MMTIMER_NAME, NULL)) {
796 printk(KERN_WARNING "%s: unable to allocate interrupt.",
801 if (misc_register(&mmtimer_miscdev)) {
802 printk(KERN_ERR "%s: failed to register device\n",
807 /* Get max numbered node, calculate slots needed */
808 for_each_online_node(node) {
813 /* Allocate list of node ptrs to mmtimer_t's */
814 timers = kzalloc(sizeof(struct mmtimer_node)*maxn, GFP_KERNEL);
815 if (timers == NULL) {
816 printk(KERN_ERR "%s: failed to allocate memory for device\n",
821 /* Initialize struct mmtimer's for each online node */
822 for_each_online_node(node) {
823 spin_lock_init(&timers[node].lock);
824 tasklet_init(&timers[node].tasklet, mmtimer_tasklet,
825 (unsigned long) node);
828 sgi_clock_period = sgi_clock.res = NSEC_PER_SEC / sn_rtc_cycles_per_second;
829 register_posix_clock(CLOCK_SGI_CYCLE, &sgi_clock);
831 printk(KERN_INFO "%s: v%s, %ld MHz\n", MMTIMER_DESC, MMTIMER_VERSION,
832 sn_rtc_cycles_per_second/(unsigned long)1E6);
838 misc_deregister(&mmtimer_miscdev);
840 free_irq(SGI_MMTIMER_VECTOR, NULL);
845 module_init(mmtimer_init);