drivers/char/mmtimer.c

   1 /*
   2  * Timer device implementation for SGI SN platforms.
   3  *
   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
   6  * for more details.
   7  *
   8  * Copyright (c) 2001-2006 Silicon Graphics, Inc.  All rights reserved.
   9  *
  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
  12  * SHub RTC, however.
  13  *
  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
  19  */
  20
  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>
  27 #include <linux/mm.h>
  28 #include <linux/fs.h>
  29 #include <linux/mmtimer.h>
  30 #include <linux/miscdevice.h>
  31 #include <linux/posix-timers.h>
  32 #include <linux/interrupt.h>
  33 #include <linux/time.h>
  34 #include <linux/math64.h>
  35 #include <linux/smp_lock.h>
  36
  37 #include <asm/uaccess.h>
  38 #include <asm/sn/addrs.h>
  39 #include <asm/sn/intr.h>
  40 #include <asm/sn/shub_mmr.h>
  41 #include <asm/sn/nodepda.h>
  42 #include <asm/sn/shubio.h>
  43
  44 MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
  45 MODULE_DESCRIPTION("SGI Altix RTC Timer");
  46 MODULE_LICENSE("GPL");
  47
  48 /* name of the device, usually in /dev */
  49 #define MMTIMER_NAME "mmtimer"
  50 #define MMTIMER_DESC "SGI Altix RTC Timer"
  51 #define MMTIMER_VERSION "2.1"
  52
  53 #define RTC_BITS 55 /* 55 bits for this implementation */
  54
  55 extern unsigned long sn_rtc_cycles_per_second;
  56
  57 #define RTC_COUNTER_ADDR        ((long *)LOCAL_MMR_ADDR(SH_RTC))
  58
  59 #define rtc_time()              (*RTC_COUNTER_ADDR)
  60
  61 static long mmtimer_ioctl(struct file *file, unsigned int cmd,
  62                                                 unsigned long arg);
  63 static int mmtimer_mmap(struct file *file, struct vm_area_struct *vma);
  64
  65 /*
  66  * Period in femtoseconds (10^-15 s)
  67  */
  68 static unsigned long mmtimer_femtoperiod = 0;
  69
  70 static const struct file_operations mmtimer_fops = {
  71         .owner = THIS_MODULE,
  72         .mmap = mmtimer_mmap,
  73         .unlocked_ioctl = mmtimer_ioctl,
  74 };
  75
  76 /*
  77  * We only have comparison registers RTC1-4 currently available per
  78  * node.  RTC0 is used by SAL.
  79  */
  80 /* Check for an RTC interrupt pending */
  81 static int mmtimer_int_pending(int comparator)
  82 {
  83         if (HUB_L((unsigned long *)LOCAL_MMR_ADDR(SH_EVENT_OCCURRED)) &
  84                         SH_EVENT_OCCURRED_RTC1_INT_MASK << comparator)
  85                 return 1;
  86         else
  87                 return 0;
  88 }
  89
  90 /* Clear the RTC interrupt pending bit */
  91 static void mmtimer_clr_int_pending(int comparator)
  92 {
  93         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_EVENT_OCCURRED_ALIAS),
  94                 SH_EVENT_OCCURRED_RTC1_INT_MASK << comparator);
  95 }
  96
  97 /* Setup timer on comparator RTC1 */
  98 static void mmtimer_setup_int_0(int cpu, u64 expires)
  99 {
 100         u64 val;
 101
 102         /* Disable interrupt */
 103         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_ENABLE), 0UL);
 104
 105         /* Initialize comparator value */
 106         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPB), -1L);
 107
 108         /* Clear pending bit */
 109         mmtimer_clr_int_pending(0);
 110
 111         val = ((u64)SGI_MMTIMER_VECTOR << SH_RTC1_INT_CONFIG_IDX_SHFT) |
 112                 ((u64)cpu_physical_id(cpu) <<
 113                         SH_RTC1_INT_CONFIG_PID_SHFT);
 114
 115         /* Set configuration */
 116         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_CONFIG), val);
 117
 118         /* Enable RTC interrupts */
 119         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_ENABLE), 1UL);
 120
 121         /* Initialize comparator value */
 122         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPB), expires);
 123
 124
 125 }
 126
 127 /* Setup timer on comparator RTC2 */
 128 static void mmtimer_setup_int_1(int cpu, u64 expires)
 129 {
 130         u64 val;
 131
 132         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_ENABLE), 0UL);
 133
 134         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPC), -1L);
 135
 136         mmtimer_clr_int_pending(1);
 137
 138         val = ((u64)SGI_MMTIMER_VECTOR << SH_RTC2_INT_CONFIG_IDX_SHFT) |
 139                 ((u64)cpu_physical_id(cpu) <<
 140                         SH_RTC2_INT_CONFIG_PID_SHFT);
 141
 142         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_CONFIG), val);
 143
 144         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_ENABLE), 1UL);
 145
 146         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPC), expires);
 147 }
 148
 149 /* Setup timer on comparator RTC3 */
 150 static void mmtimer_setup_int_2(int cpu, u64 expires)
 151 {
 152         u64 val;
 153
 154         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_ENABLE), 0UL);
 155
 156         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPD), -1L);
 157
 158         mmtimer_clr_int_pending(2);
 159
 160         val = ((u64)SGI_MMTIMER_VECTOR << SH_RTC3_INT_CONFIG_IDX_SHFT) |
 161                 ((u64)cpu_physical_id(cpu) <<
 162                         SH_RTC3_INT_CONFIG_PID_SHFT);
 163
 164         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_CONFIG), val);
 165
 166         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_ENABLE), 1UL);
 167
 168         HUB_S((u64 *)LOCAL_MMR_ADDR(SH_INT_CMPD), expires);
 169 }
 170
 171 /*
 172  * This function must be called with interrupts disabled and preemption off
 173  * in order to insure that the setup succeeds in a deterministic time frame.
 174  * It will check if the interrupt setup succeeded.
 175  */
 176 static int mmtimer_setup(int cpu, int comparator, unsigned long expires)
 177 {
 178
 179         switch (comparator) {
 180         case 0:
 181                 mmtimer_setup_int_0(cpu, expires);
 182                 break;
 183         case 1:
 184                 mmtimer_setup_int_1(cpu, expires);
 185                 break;
 186         case 2:
 187                 mmtimer_setup_int_2(cpu, expires);
 188                 break;
 189         }
 190         /* We might've missed our expiration time */
 191         if (rtc_time() <= expires)
 192                 return 1;
 193
 194         /*
 195          * If an interrupt is already pending then its okay
 196          * if not then we failed
 197          */
 198         return mmtimer_int_pending(comparator);
 199 }
 200
 201 static int mmtimer_disable_int(long nasid, int comparator)
 202 {
 203         switch (comparator) {
 204         case 0:
 205                 nasid == -1 ? HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC1_INT_ENABLE),
 206                         0UL) : REMOTE_HUB_S(nasid, SH_RTC1_INT_ENABLE, 0UL);
 207                 break;
 208         case 1:
 209                 nasid == -1 ? HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC2_INT_ENABLE),
 210                         0UL) : REMOTE_HUB_S(nasid, SH_RTC2_INT_ENABLE, 0UL);
 211                 break;
 212         case 2:
 213                 nasid == -1 ? HUB_S((u64 *)LOCAL_MMR_ADDR(SH_RTC3_INT_ENABLE),
 214                         0UL) : REMOTE_HUB_S(nasid, SH_RTC3_INT_ENABLE, 0UL);
 215                 break;
 216         default:
 217                 return -EFAULT;
 218         }
 219         return 0;
 220 }
 221
 222 #define COMPARATOR      1               /* The comparator to use */
 223
 224 #define TIMER_OFF       0xbadcabLL      /* Timer is not setup */
 225 #define TIMER_SET       0               /* Comparator is set for this timer */
 226
 227 /* There is one of these for each timer */
 228 struct mmtimer {
 229         struct rb_node list;
 230         struct k_itimer *timer;
 231         int cpu;
 232 };
 233
 234 struct mmtimer_node {
 235         spinlock_t lock ____cacheline_aligned;
 236         struct rb_root timer_head;
 237         struct rb_node *next;
 238         struct tasklet_struct tasklet;
 239 };
 240 static struct mmtimer_node *timers;
 241
 242
 243 /*
 244  * Add a new mmtimer struct to the node's mmtimer list.
 245  * This function assumes the struct mmtimer_node is locked.
 246  */
 247 static void mmtimer_add_list(struct mmtimer *n)
 248 {
 249         int nodeid = n->timer->it.mmtimer.node;
 250         unsigned long expires = n->timer->it.mmtimer.expires;
 251         struct rb_node **link = &timers[nodeid].timer_head.rb_node;
 252         struct rb_node *parent = NULL;
 253         struct mmtimer *x;
 254
 255         /*
 256          * Find the right place in the rbtree:
 257          */
 258         while (*link) {
 259                 parent = *link;
 260                 x = rb_entry(parent, struct mmtimer, list);
 261
 262                 if (expires < x->timer->it.mmtimer.expires)
 263                         link = &(*link)->rb_left;
 264                 else
 265                         link = &(*link)->rb_right;
 266         }
 267
 268         /*
 269          * Insert the timer to the rbtree and check whether it
 270          * replaces the first pending timer
 271          */
 272         rb_link_node(&n->list, parent, link);
 273         rb_insert_color(&n->list, &timers[nodeid].timer_head);
 274
 275         if (!timers[nodeid].next || expires < rb_entry(timers[nodeid].next,
 276                         struct mmtimer, list)->timer->it.mmtimer.expires)
 277                 timers[nodeid].next = &n->list;
 278 }
 279
 280 /*
 281  * Set the comparator for the next timer.
 282  * This function assumes the struct mmtimer_node is locked.
 283  */
 284 static void mmtimer_set_next_timer(int nodeid)
 285 {
 286         struct mmtimer_node *n = &timers[nodeid];
 287         struct mmtimer *x;
 288         struct k_itimer *t;
 289         int o;
 290
 291 restart:
 292         if (n->next == NULL)
 293                 return;
 294
 295         x = rb_entry(n->next, struct mmtimer, list);
 296         t = x->timer;
 297         if (!t->it.mmtimer.incr) {
 298                 /* Not an interval timer */
 299                 if (!mmtimer_setup(x->cpu, COMPARATOR,
 300                                         t->it.mmtimer.expires)) {
 301                         /* Late setup, fire now */
 302                         tasklet_schedule(&n->tasklet);
 303                 }
 304                 return;
 305         }
 306
 307         /* Interval timer */
 308         o = 0;
 309         while (!mmtimer_setup(x->cpu, COMPARATOR, t->it.mmtimer.expires)) {
 310                 unsigned long e, e1;
 311                 struct rb_node *next;
 312                 t->it.mmtimer.expires += t->it.mmtimer.incr << o;
 313                 t->it_overrun += 1 << o;
 314                 o++;
 315                 if (o > 20) {
 316                         printk(KERN_ALERT "mmtimer: cannot reschedule timer\n");
 317                         t->it.mmtimer.clock = TIMER_OFF;
 318                         n->next = rb_next(&x->list);
 319                         rb_erase(&x->list, &n->timer_head);
 320                         kfree(x);
 321                         goto restart;
 322                 }
 323
 324                 e = t->it.mmtimer.expires;
 325                 next = rb_next(&x->list);
 326
 327                 if (next == NULL)
 328                         continue;
 329
 330                 e1 = rb_entry(next, struct mmtimer, list)->
 331                         timer->it.mmtimer.expires;
 332                 if (e > e1) {
 333                         n->next = next;
 334                         rb_erase(&x->list, &n->timer_head);
 335                         mmtimer_add_list(x);
 336                         goto restart;
 337                 }
 338         }
 339 }
 340
 341 /**
 342  * mmtimer_ioctl - ioctl interface for /dev/mmtimer
 343  * @file: file structure for the device
 344  * @cmd: command to execute
 345  * @arg: optional argument to command
 346  *
 347  * Executes the command specified by @cmd.  Returns 0 for success, < 0 for
 348  * failure.
 349  *
 350  * Valid commands:
 351  *
 352  * %MMTIMER_GETOFFSET - Should return the offset (relative to the start
 353  * of the page where the registers are mapped) for the counter in question.
 354  *
 355  * %MMTIMER_GETRES - Returns the resolution of the clock in femto (10^-15)
 356  * seconds
 357  *
 358  * %MMTIMER_GETFREQ - Copies the frequency of the clock in Hz to the address
 359  * specified by @arg
 360  *
 361  * %MMTIMER_GETBITS - Returns the number of bits in the clock's counter
 362  *
 363  * %MMTIMER_MMAPAVAIL - Returns 1 if the registers can be mmap'd into userspace
 364  *
 365  * %MMTIMER_GETCOUNTER - Gets the current value in the counter and places it
 366  * in the address specified by @arg.
 367  */
 368 static long mmtimer_ioctl(struct file *file, unsigned int cmd,
 369                                                 unsigned long arg)
 370 {
 371         int ret = 0;
 372
 373         lock_kernel();
 374
 375         switch (cmd) {
 376         case MMTIMER_GETOFFSET: /* offset of the counter */
 377                 /*
 378                  * SN RTC registers are on their own 64k page
 379                  */
 380                 if(PAGE_SIZE <= (1 << 16))
 381                         ret = (((long)RTC_COUNTER_ADDR) & (PAGE_SIZE-1)) / 8;
 382                 else
 383                         ret = -ENOSYS;
 384                 break;
 385
 386         case MMTIMER_GETRES: /* resolution of the clock in 10^-15 s */
 387                 if(copy_to_user((unsigned long __user *)arg,
 388                                 &mmtimer_femtoperiod, sizeof(unsigned long)))
 389                         ret = -EFAULT;
 390                 break;
 391
 392         case MMTIMER_GETFREQ: /* frequency in Hz */
 393                 if(copy_to_user((unsigned long __user *)arg,
 394                                 &sn_rtc_cycles_per_second,
 395                                 sizeof(unsigned long)))
 396                         ret = -EFAULT;
 397                 break;
 398
 399         case MMTIMER_GETBITS: /* number of bits in the clock */
 400                 ret = RTC_BITS;
 401                 break;
 402
 403         case MMTIMER_MMAPAVAIL: /* can we mmap the clock into userspace? */
 404                 ret = (PAGE_SIZE <= (1 << 16)) ? 1 : 0;
 405                 break;
 406
 407         case MMTIMER_GETCOUNTER:
 408                 if(copy_to_user((unsigned long __user *)arg,
 409                                 RTC_COUNTER_ADDR, sizeof(unsigned long)))
 410                         ret = -EFAULT;
 411                 break;
 412         default:
 413                 ret = -ENOTTY;
 414                 break;
 415         }
 416         unlock_kernel();
 417         return ret;
 418 }
 419
 420 /**
 421  * mmtimer_mmap - maps the clock's registers into userspace
 422  * @file: file structure for the device
 423  * @vma: VMA to map the registers into
 424  *
 425  * Calls remap_pfn_range() to map the clock's registers into
 426  * the calling process' address space.
 427  */
 428 static int mmtimer_mmap(struct file *file, struct vm_area_struct *vma)
 429 {
 430         unsigned long mmtimer_addr;
 431
 432         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 433                 return -EINVAL;
 434
 435         if (vma->vm_flags & VM_WRITE)
 436                 return -EPERM;
 437
 438         if (PAGE_SIZE > (1 << 16))
 439                 return -ENOSYS;
 440
 441         vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 442
 443         mmtimer_addr = __pa(RTC_COUNTER_ADDR);
 444         mmtimer_addr &= ~(PAGE_SIZE - 1);
 445         mmtimer_addr &= 0xfffffffffffffffUL;
 446
 447         if (remap_pfn_range(vma, vma->vm_start, mmtimer_addr >> PAGE_SHIFT,
 448                                         PAGE_SIZE, vma->vm_page_prot)) {
 449                 printk(KERN_ERR "remap_pfn_range failed in mmtimer.c\n");
 450                 return -EAGAIN;
 451         }
 452
 453         return 0;
 454 }
 455
 456 static struct miscdevice mmtimer_miscdev = {
 457         SGI_MMTIMER,
 458         MMTIMER_NAME,
 459         &mmtimer_fops
 460 };
 461
 462 static struct timespec sgi_clock_offset;
 463 static int sgi_clock_period;
 464
 465 /*
 466  * Posix Timer Interface
 467  */
 468
 469 static struct timespec sgi_clock_offset;
 470 static int sgi_clock_period;
 471
 472 static int sgi_clock_get(clockid_t clockid, struct timespec *tp)
 473 {
 474         u64 nsec;
 475
 476         nsec = rtc_time() * sgi_clock_period
 477                         + sgi_clock_offset.tv_nsec;
 478         *tp = ns_to_timespec(nsec);
 479         tp->tv_sec += sgi_clock_offset.tv_sec;
 480         return 0;
 481 };
 482
 483 static int sgi_clock_set(clockid_t clockid, struct timespec *tp)
 484 {
 485
 486         u64 nsec;
 487         u32 rem;
 488
 489         nsec = rtc_time() * sgi_clock_period;
 490
 491         sgi_clock_offset.tv_sec = tp->tv_sec - div_u64_rem(nsec, NSEC_PER_SEC, &rem);
 492
 493         if (rem <= tp->tv_nsec)
 494                 sgi_clock_offset.tv_nsec = tp->tv_sec - rem;
 495         else {
 496                 sgi_clock_offset.tv_nsec = tp->tv_sec + NSEC_PER_SEC - rem;
 497                 sgi_clock_offset.tv_sec--;
 498         }
 499         return 0;
 500 }
 501
 502 /**
 503  * mmtimer_interrupt - timer interrupt handler
 504  * @irq: irq received
 505  * @dev_id: device the irq came from
 506  *
 507  * Called when one of the comarators matches the counter, This
 508  * routine will send signals to processes that have requested
 509  * them.
 510  *
 511  * This interrupt is run in an interrupt context
 512  * by the SHUB. It is therefore safe to locally access SHub
 513  * registers.
 514  */
 515 static irqreturn_t
 516 mmtimer_interrupt(int irq, void *dev_id)
 517 {
 518         unsigned long expires = 0;
 519         int result = IRQ_NONE;
 520         unsigned indx = cpu_to_node(smp_processor_id());
 521         struct mmtimer *base;
 522
 523         spin_lock(&timers[indx].lock);
 524         base = rb_entry(timers[indx].next, struct mmtimer, list);
 525         if (base == NULL) {
 526                 spin_unlock(&timers[indx].lock);
 527                 return result;
 528         }
 529
 530         if (base->cpu == smp_processor_id()) {
 531                 if (base->timer)
 532                         expires = base->timer->it.mmtimer.expires;
 533                 /* expires test won't work with shared irqs */
 534                 if ((mmtimer_int_pending(COMPARATOR) > 0) ||
 535                         (expires && (expires <= rtc_time()))) {
 536                         mmtimer_clr_int_pending(COMPARATOR);
 537                         tasklet_schedule(&timers[indx].tasklet);
 538                         result = IRQ_HANDLED;
 539                 }
 540         }
 541         spin_unlock(&timers[indx].lock);
 542         return result;
 543 }
 544
 545 static void mmtimer_tasklet(unsigned long data)
 546 {
 547         int nodeid = data;
 548         struct mmtimer_node *mn = &timers[nodeid];
 549         struct mmtimer *x = rb_entry(mn->next, struct mmtimer, list);
 550         struct k_itimer *t;
 551         unsigned long flags;
 552
 553         /* Send signal and deal with periodic signals */
 554         spin_lock_irqsave(&mn->lock, flags);
 555         if (!mn->next)
 556                 goto out;
 557
 558         x = rb_entry(mn->next, struct mmtimer, list);
 559         t = x->timer;
 560
 561         if (t->it.mmtimer.clock == TIMER_OFF)
 562                 goto out;
 563
 564         t->it_overrun = 0;
 565
 566         mn->next = rb_next(&x->list);
 567         rb_erase(&x->list, &mn->timer_head);
 568
 569         if (posix_timer_event(t, 0) != 0)
 570                 t->it_overrun++;
 571
 572         if(t->it.mmtimer.incr) {
 573                 t->it.mmtimer.expires += t->it.mmtimer.incr;
 574                 mmtimer_add_list(x);
 575         } else {
 576                 /* Ensure we don't false trigger in mmtimer_interrupt */
 577                 t->it.mmtimer.clock = TIMER_OFF;
 578                 t->it.mmtimer.expires = 0;
 579                 kfree(x);
 580         }
 581         /* Set comparator for next timer, if there is one */
 582         mmtimer_set_next_timer(nodeid);
 583
 584         t->it_overrun_last = t->it_overrun;
 585 out:
 586         spin_unlock_irqrestore(&mn->lock, flags);
 587 }
 588
 589 static int sgi_timer_create(struct k_itimer *timer)
 590 {
 591         /* Insure that a newly created timer is off */
 592         timer->it.mmtimer.clock = TIMER_OFF;
 593         return 0;
 594 }
 595
 596 /* This does not really delete a timer. It just insures
 597  * that the timer is not active
 598  *
 599  * Assumption: it_lock is already held with irq's disabled
 600  */
 601 static int sgi_timer_del(struct k_itimer *timr)
 602 {
 603         cnodeid_t nodeid = timr->it.mmtimer.node;
 604         unsigned long irqflags;
 605
 606         spin_lock_irqsave(&timers[nodeid].lock, irqflags);
 607         if (timr->it.mmtimer.clock != TIMER_OFF) {
 608                 unsigned long expires = timr->it.mmtimer.expires;
 609                 struct rb_node *n = timers[nodeid].timer_head.rb_node;
 610                 struct mmtimer *uninitialized_var(t);
 611                 int r = 0;
 612
 613                 timr->it.mmtimer.clock = TIMER_OFF;
 614                 timr->it.mmtimer.expires = 0;
 615
 616                 while (n) {
 617                         t = rb_entry(n, struct mmtimer, list);
 618                         if (t->timer == timr)
 619                                 break;
 620
 621                         if (expires < t->timer->it.mmtimer.expires)
 622                                 n = n->rb_left;
 623                         else
 624                                 n = n->rb_right;
 625                 }
 626
 627                 if (!n) {
 628                         spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
 629                         return 0;
 630                 }
 631
 632                 if (timers[nodeid].next == n) {
 633                         timers[nodeid].next = rb_next(n);
 634                         r = 1;
 635                 }
 636
 637                 rb_erase(n, &timers[nodeid].timer_head);
 638                 kfree(t);
 639
 640                 if (r) {
 641                         mmtimer_disable_int(cnodeid_to_nasid(nodeid),
 642                                 COMPARATOR);
 643                         mmtimer_set_next_timer(nodeid);
 644                 }
 645         }
 646         spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
 647         return 0;
 648 }
 649
 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)
 652 {
 653
 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;
 659                 return;
 660         }
 661
 662         cur_setting->it_interval = ns_to_timespec(timr->it.mmtimer.incr * sgi_clock_period);
 663         cur_setting->it_value = ns_to_timespec((timr->it.mmtimer.expires - rtc_time()) * sgi_clock_period);
 664 }
 665
 666
 667 static int sgi_timer_set(struct k_itimer *timr, int flags,
 668         struct itimerspec * new_setting,
 669         struct itimerspec * old_setting)
 670 {
 671         unsigned long when, period, irqflags;
 672         int err = 0;
 673         cnodeid_t nodeid;
 674         struct mmtimer *base;
 675         struct rb_node *n;
 676
 677         if (old_setting)
 678                 sgi_timer_get(timr, old_setting);
 679
 680         sgi_timer_del(timr);
 681         when = timespec_to_ns(&new_setting->it_value);
 682         period = timespec_to_ns(&new_setting->it_interval);
 683
 684         if (when == 0)
 685                 /* Clear timer */
 686                 return 0;
 687
 688         base = kmalloc(sizeof(struct mmtimer), GFP_KERNEL);
 689         if (base == NULL)
 690                 return -ENOMEM;
 691
 692         if (flags & TIMER_ABSTIME) {
 693                 struct timespec n;
 694                 unsigned long now;
 695
 696                 getnstimeofday(&n);
 697                 now = timespec_to_ns(&n);
 698                 if (when > now)
 699                         when -= now;
 700                 else
 701                         /* Fire the timer immediately */
 702                         when = 0;
 703         }
 704
 705         /*
 706          * Convert to sgi clock period. Need to keep rtc_time() as near as possible
 707          * to getnstimeofday() in order to be as faithful as possible to the time
 708          * specified.
 709          */
 710         when = (when + sgi_clock_period - 1) / sgi_clock_period + rtc_time();
 711         period = (period + sgi_clock_period - 1)  / sgi_clock_period;
 712
 713         /*
 714          * We are allocating a local SHub comparator. If we would be moved to another
 715          * cpu then another SHub may be local to us. Prohibit that by switching off
 716          * preemption.
 717          */
 718         preempt_disable();
 719
 720         nodeid =  cpu_to_node(smp_processor_id());
 721
 722         /* Lock the node timer structure */
 723         spin_lock_irqsave(&timers[nodeid].lock, irqflags);
 724
 725         base->timer = timr;
 726         base->cpu = smp_processor_id();
 727
 728         timr->it.mmtimer.clock = TIMER_SET;
 729         timr->it.mmtimer.node = nodeid;
 730         timr->it.mmtimer.incr = period;
 731         timr->it.mmtimer.expires = when;
 732
 733         n = timers[nodeid].next;
 734
 735         /* Add the new struct mmtimer to node's timer list */
 736         mmtimer_add_list(base);
 737
 738         if (timers[nodeid].next == n) {
 739                 /* No need to reprogram comparator for now */
 740                 spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
 741                 preempt_enable();
 742                 return err;
 743         }
 744
 745         /* We need to reprogram the comparator */
 746         if (n)
 747                 mmtimer_disable_int(cnodeid_to_nasid(nodeid), COMPARATOR);
 748
 749         mmtimer_set_next_timer(nodeid);
 750
 751         /* Unlock the node timer structure */
 752         spin_unlock_irqrestore(&timers[nodeid].lock, irqflags);
 753
 754         preempt_enable();
 755
 756         return err;
 757 }
 758
 759 static struct k_clock sgi_clock = {
 760         .res = 0,
 761         .clock_set = sgi_clock_set,
 762         .clock_get = sgi_clock_get,
 763         .timer_create = sgi_timer_create,
 764         .nsleep = do_posix_clock_nonanosleep,
 765         .timer_set = sgi_timer_set,
 766         .timer_del = sgi_timer_del,
 767         .timer_get = sgi_timer_get
 768 };
 769
 770 /**
 771  * mmtimer_init - device initialization routine
 772  *
 773  * Does initial setup for the mmtimer device.
 774  */
 775 static int __init mmtimer_init(void)
 776 {
 777         cnodeid_t node, maxn = -1;
 778
 779         if (!ia64_platform_is("sn2"))
 780                 return 0;
 781
 782         /*
 783          * Sanity check the cycles/sec variable
 784          */
 785         if (sn_rtc_cycles_per_second < 100000) {
 786                 printk(KERN_ERR "%s: unable to determine clock frequency\n",
 787                        MMTIMER_NAME);
 788                 goto out1;
 789         }
 790
 791         mmtimer_femtoperiod = ((unsigned long)1E15 + sn_rtc_cycles_per_second /
 792                                2) / sn_rtc_cycles_per_second;
 793
 794         if (request_irq(SGI_MMTIMER_VECTOR, mmtimer_interrupt, IRQF_PERCPU, MMTIMER_NAME, NULL)) {
 795                 printk(KERN_WARNING "%s: unable to allocate interrupt.",
 796                         MMTIMER_NAME);
 797                 goto out1;
 798         }
 799
 800         if (misc_register(&mmtimer_miscdev)) {
 801                 printk(KERN_ERR "%s: failed to register device\n",
 802                        MMTIMER_NAME);
 803                 goto out2;
 804         }
 805
 806         /* Get max numbered node, calculate slots needed */
 807         for_each_online_node(node) {
 808                 maxn = node;
 809         }
 810         maxn++;
 811
 812         /* Allocate list of node ptrs to mmtimer_t's */
 813         timers = kzalloc(sizeof(struct mmtimer_node)*maxn, GFP_KERNEL);
 814         if (timers == NULL) {
 815                 printk(KERN_ERR "%s: failed to allocate memory for device\n",
 816                                 MMTIMER_NAME);
 817                 goto out3;
 818         }
 819
 820         /* Initialize struct mmtimer's for each online node */
 821         for_each_online_node(node) {
 822                 spin_lock_init(&timers[node].lock);
 823                 tasklet_init(&timers[node].tasklet, mmtimer_tasklet,
 824                         (unsigned long) node);
 825         }
 826
 827         sgi_clock_period = sgi_clock.res = NSEC_PER_SEC / sn_rtc_cycles_per_second;
 828         register_posix_clock(CLOCK_SGI_CYCLE, &sgi_clock);
 829
 830         printk(KERN_INFO "%s: v%s, %ld MHz\n", MMTIMER_DESC, MMTIMER_VERSION,
 831                sn_rtc_cycles_per_second/(unsigned long)1E6);
 832
 833         return 0;
 834
 835 out3:
 836         kfree(timers);
 837         misc_deregister(&mmtimer_miscdev);
 838 out2:
 839         free_irq(SGI_MMTIMER_VECTOR, NULL);
 840 out1:
 841         return -1;
 842 }
 843
 844 module_init(mmtimer_init);