1 /* 2 * Common time routines among all ppc machines. 3 * 4 * Written by Cort Dougan (cort@cs.nmt.edu) to merge 5 * Paul Mackerras' version and mine for PReP and Pmac. 6 * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net). 7 * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com) 8 * 9 * First round of bugfixes by Gabriel Paubert (paubert@iram.es) 10 * to make clock more stable (2.4.0-test5). The only thing 11 * that this code assumes is that the timebases have been synchronized 12 * by firmware on SMP and are never stopped (never do sleep 13 * on SMP then, nap and doze are OK). 14 * 15 * Speeded up do_gettimeofday by getting rid of references to 16 * xtime (which required locks for consistency). (mikejc@us.ibm.com) 17 * 18 * TODO (not necessarily in this file): 19 * - improve precision and reproducibility of timebase frequency 20 * measurement at boot time. (for iSeries, we calibrate the timebase 21 * against the Titan chip's clock.) 22 * - for astronomical applications: add a new function to get 23 * non ambiguous timestamps even around leap seconds. This needs 24 * a new timestamp format and a good name. 25 * 26 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 27 * "A Kernel Model for Precision Timekeeping" by Dave Mills 28 * 29 * This program is free software; you can redistribute it and/or 30 * modify it under the terms of the GNU General Public License 31 * as published by the Free Software Foundation; either version 32 * 2 of the License, or (at your option) any later version. 33 */ 34 35 #include <linux/errno.h> 36 #include <linux/module.h> 37 #include <linux/sched.h> 38 #include <linux/kernel.h> 39 #include <linux/param.h> 40 #include <linux/string.h> 41 #include <linux/mm.h> 42 #include <linux/interrupt.h> 43 #include <linux/timex.h> 44 #include <linux/kernel_stat.h> 45 #include <linux/time.h> 46 #include <linux/init.h> 47 #include <linux/profile.h> 48 #include <linux/cpu.h> 49 #include <linux/security.h> 50 #include <linux/percpu.h> 51 #include <linux/rtc.h> 52 #include <linux/jiffies.h> 53 #include <linux/posix-timers.h> 54 #include <linux/irq.h> 55 56 #include <asm/io.h> 57 #include <asm/processor.h> 58 #include <asm/nvram.h> 59 #include <asm/cache.h> 60 #include <asm/machdep.h> 61 #include <asm/uaccess.h> 62 #include <asm/time.h> 63 #include <asm/prom.h> 64 #include <asm/irq.h> 65 #include <asm/div64.h> 66 #include <asm/smp.h> 67 #include <asm/vdso_datapage.h> 68 #include <asm/firmware.h> 69 #ifdef CONFIG_PPC_ISERIES 70 #include <asm/iseries/it_lp_queue.h> 71 #include <asm/iseries/hv_call_xm.h> 72 #endif 73 74 /* powerpc clocksource/clockevent code */ 75 76 #include <linux/clockchips.h> 77 #include <linux/clocksource.h> 78 79 static cycle_t rtc_read(void); 80 static struct clocksource clocksource_rtc = { 81 .name = "rtc", 82 .rating = 400, 83 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 84 .mask = CLOCKSOURCE_MASK(64), 85 .shift = 22, 86 .mult = 0, /* To be filled in */ 87 .read = rtc_read, 88 }; 89 90 static cycle_t timebase_read(void); 91 static struct clocksource clocksource_timebase = { 92 .name = "timebase", 93 .rating = 400, 94 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 95 .mask = CLOCKSOURCE_MASK(64), 96 .shift = 22, 97 .mult = 0, /* To be filled in */ 98 .read = timebase_read, 99 }; 100 101 #define DECREMENTER_MAX 0x7fffffff 102 103 static int decrementer_set_next_event(unsigned long evt, 104 struct clock_event_device *dev); 105 static void decrementer_set_mode(enum clock_event_mode mode, 106 struct clock_event_device *dev); 107 108 static struct clock_event_device decrementer_clockevent = { 109 .name = "decrementer", 110 .rating = 200, 111 .shift = 32, 112 .mult = 0, /* To be filled in */ 113 .irq = 0, 114 .set_next_event = decrementer_set_next_event, 115 .set_mode = decrementer_set_mode, 116 .features = CLOCK_EVT_FEAT_ONESHOT, 117 }; 118 119 static DEFINE_PER_CPU(struct clock_event_device, decrementers); 120 void init_decrementer_clockevent(void); 121 static DEFINE_PER_CPU(u64, decrementer_next_tb); 122 123 #ifdef CONFIG_PPC_ISERIES 124 static unsigned long __initdata iSeries_recal_titan; 125 static signed long __initdata iSeries_recal_tb; 126 127 /* Forward declaration is only needed for iSereis compiles */ 128 void __init clocksource_init(void); 129 #endif 130 131 #define XSEC_PER_SEC (1024*1024) 132 133 #ifdef CONFIG_PPC64 134 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC) 135 #else 136 /* compute ((xsec << 12) * max) >> 32 */ 137 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max) 138 #endif 139 140 unsigned long tb_ticks_per_jiffy; 141 unsigned long tb_ticks_per_usec = 100; /* sane default */ 142 EXPORT_SYMBOL(tb_ticks_per_usec); 143 unsigned long tb_ticks_per_sec; 144 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */ 145 u64 tb_to_xs; 146 unsigned tb_to_us; 147 148 #define TICKLEN_SCALE TICK_LENGTH_SHIFT 149 u64 last_tick_len; /* units are ns / 2^TICKLEN_SCALE */ 150 u64 ticklen_to_xs; /* 0.64 fraction */ 151 152 /* If last_tick_len corresponds to about 1/HZ seconds, then 153 last_tick_len << TICKLEN_SHIFT will be about 2^63. */ 154 #define TICKLEN_SHIFT (63 - 30 - TICKLEN_SCALE + SHIFT_HZ) 155 156 DEFINE_SPINLOCK(rtc_lock); 157 EXPORT_SYMBOL_GPL(rtc_lock); 158 159 static u64 tb_to_ns_scale __read_mostly; 160 static unsigned tb_to_ns_shift __read_mostly; 161 static unsigned long boot_tb __read_mostly; 162 163 struct gettimeofday_struct do_gtod; 164 165 extern struct timezone sys_tz; 166 static long timezone_offset; 167 168 unsigned long ppc_proc_freq; 169 EXPORT_SYMBOL(ppc_proc_freq); 170 unsigned long ppc_tb_freq; 171 172 static u64 tb_last_jiffy __cacheline_aligned_in_smp; 173 static DEFINE_PER_CPU(u64, last_jiffy); 174 175 #ifdef CONFIG_VIRT_CPU_ACCOUNTING 176 /* 177 * Factors for converting from cputime_t (timebase ticks) to 178 * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds). 179 * These are all stored as 0.64 fixed-point binary fractions. 180 */ 181 u64 __cputime_jiffies_factor; 182 EXPORT_SYMBOL(__cputime_jiffies_factor); 183 u64 __cputime_msec_factor; 184 EXPORT_SYMBOL(__cputime_msec_factor); 185 u64 __cputime_sec_factor; 186 EXPORT_SYMBOL(__cputime_sec_factor); 187 u64 __cputime_clockt_factor; 188 EXPORT_SYMBOL(__cputime_clockt_factor); 189 190 static void calc_cputime_factors(void) 191 { 192 struct div_result res; 193 194 div128_by_32(HZ, 0, tb_ticks_per_sec, &res); 195 __cputime_jiffies_factor = res.result_low; 196 div128_by_32(1000, 0, tb_ticks_per_sec, &res); 197 __cputime_msec_factor = res.result_low; 198 div128_by_32(1, 0, tb_ticks_per_sec, &res); 199 __cputime_sec_factor = res.result_low; 200 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res); 201 __cputime_clockt_factor = res.result_low; 202 } 203 204 /* 205 * Read the PURR on systems that have it, otherwise the timebase. 206 */ 207 static u64 read_purr(void) 208 { 209 if (cpu_has_feature(CPU_FTR_PURR)) 210 return mfspr(SPRN_PURR); 211 return mftb(); 212 } 213 214 /* 215 * Account time for a transition between system, hard irq 216 * or soft irq state. 217 */ 218 void account_system_vtime(struct task_struct *tsk) 219 { 220 u64 now, delta; 221 unsigned long flags; 222 223 local_irq_save(flags); 224 now = read_purr(); 225 delta = now - get_paca()->startpurr; 226 get_paca()->startpurr = now; 227 if (!in_interrupt()) { 228 delta += get_paca()->system_time; 229 get_paca()->system_time = 0; 230 } 231 account_system_time(tsk, 0, delta); 232 local_irq_restore(flags); 233 } 234 235 /* 236 * Transfer the user and system times accumulated in the paca 237 * by the exception entry and exit code to the generic process 238 * user and system time records. 239 * Must be called with interrupts disabled. 240 */ 241 void account_process_vtime(struct task_struct *tsk) 242 { 243 cputime_t utime; 244 245 utime = get_paca()->user_time; 246 get_paca()->user_time = 0; 247 account_user_time(tsk, utime); 248 } 249 250 static void account_process_time(struct pt_regs *regs) 251 { 252 int cpu = smp_processor_id(); 253 254 account_process_vtime(current); 255 run_local_timers(); 256 if (rcu_pending(cpu)) 257 rcu_check_callbacks(cpu, user_mode(regs)); 258 scheduler_tick(); 259 run_posix_cpu_timers(current); 260 } 261 262 /* 263 * Stuff for accounting stolen time. 264 */ 265 struct cpu_purr_data { 266 int initialized; /* thread is running */ 267 u64 tb; /* last TB value read */ 268 u64 purr; /* last PURR value read */ 269 }; 270 271 /* 272 * Each entry in the cpu_purr_data array is manipulated only by its 273 * "owner" cpu -- usually in the timer interrupt but also occasionally 274 * in process context for cpu online. As long as cpus do not touch 275 * each others' cpu_purr_data, disabling local interrupts is 276 * sufficient to serialize accesses. 277 */ 278 static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data); 279 280 static void snapshot_tb_and_purr(void *data) 281 { 282 unsigned long flags; 283 struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data); 284 285 local_irq_save(flags); 286 p->tb = get_tb_or_rtc(); 287 p->purr = mfspr(SPRN_PURR); 288 wmb(); 289 p->initialized = 1; 290 local_irq_restore(flags); 291 } 292 293 /* 294 * Called during boot when all cpus have come up. 295 */ 296 void snapshot_timebases(void) 297 { 298 if (!cpu_has_feature(CPU_FTR_PURR)) 299 return; 300 on_each_cpu(snapshot_tb_and_purr, NULL, 0, 1); 301 } 302 303 /* 304 * Must be called with interrupts disabled. 305 */ 306 void calculate_steal_time(void) 307 { 308 u64 tb, purr; 309 s64 stolen; 310 struct cpu_purr_data *pme; 311 312 if (!cpu_has_feature(CPU_FTR_PURR)) 313 return; 314 pme = &per_cpu(cpu_purr_data, smp_processor_id()); 315 if (!pme->initialized) 316 return; /* this can happen in early boot */ 317 tb = mftb(); 318 purr = mfspr(SPRN_PURR); 319 stolen = (tb - pme->tb) - (purr - pme->purr); 320 if (stolen > 0) 321 account_steal_time(current, stolen); 322 pme->tb = tb; 323 pme->purr = purr; 324 } 325 326 #ifdef CONFIG_PPC_SPLPAR 327 /* 328 * Must be called before the cpu is added to the online map when 329 * a cpu is being brought up at runtime. 330 */ 331 static void snapshot_purr(void) 332 { 333 struct cpu_purr_data *pme; 334 unsigned long flags; 335 336 if (!cpu_has_feature(CPU_FTR_PURR)) 337 return; 338 local_irq_save(flags); 339 pme = &per_cpu(cpu_purr_data, smp_processor_id()); 340 pme->tb = mftb(); 341 pme->purr = mfspr(SPRN_PURR); 342 pme->initialized = 1; 343 local_irq_restore(flags); 344 } 345 346 #endif /* CONFIG_PPC_SPLPAR */ 347 348 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */ 349 #define calc_cputime_factors() 350 #define account_process_time(regs) update_process_times(user_mode(regs)) 351 #define calculate_steal_time() do { } while (0) 352 #endif 353 354 #if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR)) 355 #define snapshot_purr() do { } while (0) 356 #endif 357 358 /* 359 * Called when a cpu comes up after the system has finished booting, 360 * i.e. as a result of a hotplug cpu action. 361 */ 362 void snapshot_timebase(void) 363 { 364 __get_cpu_var(last_jiffy) = get_tb_or_rtc(); 365 snapshot_purr(); 366 } 367 368 void __delay(unsigned long loops) 369 { 370 unsigned long start; 371 int diff; 372 373 if (__USE_RTC()) { 374 start = get_rtcl(); 375 do { 376 /* the RTCL register wraps at 1000000000 */ 377 diff = get_rtcl() - start; 378 if (diff < 0) 379 diff += 1000000000; 380 } while (diff < loops); 381 } else { 382 start = get_tbl(); 383 while (get_tbl() - start < loops) 384 HMT_low(); 385 HMT_medium(); 386 } 387 } 388 EXPORT_SYMBOL(__delay); 389 390 void udelay(unsigned long usecs) 391 { 392 __delay(tb_ticks_per_usec * usecs); 393 } 394 EXPORT_SYMBOL(udelay); 395 396 397 /* 398 * There are two copies of tb_to_xs and stamp_xsec so that no 399 * lock is needed to access and use these values in 400 * do_gettimeofday. We alternate the copies and as long as a 401 * reasonable time elapses between changes, there will never 402 * be inconsistent values. ntpd has a minimum of one minute 403 * between updates. 404 */ 405 static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec, 406 u64 new_tb_to_xs) 407 { 408 unsigned temp_idx; 409 struct gettimeofday_vars *temp_varp; 410 411 temp_idx = (do_gtod.var_idx == 0); 412 temp_varp = &do_gtod.vars[temp_idx]; 413 414 temp_varp->tb_to_xs = new_tb_to_xs; 415 temp_varp->tb_orig_stamp = new_tb_stamp; 416 temp_varp->stamp_xsec = new_stamp_xsec; 417 smp_mb(); 418 do_gtod.varp = temp_varp; 419 do_gtod.var_idx = temp_idx; 420 421 /* 422 * tb_update_count is used to allow the userspace gettimeofday code 423 * to assure itself that it sees a consistent view of the tb_to_xs and 424 * stamp_xsec variables. It reads the tb_update_count, then reads 425 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 426 * the two values of tb_update_count match and are even then the 427 * tb_to_xs and stamp_xsec values are consistent. If not, then it 428 * loops back and reads them again until this criteria is met. 429 * We expect the caller to have done the first increment of 430 * vdso_data->tb_update_count already. 431 */ 432 vdso_data->tb_orig_stamp = new_tb_stamp; 433 vdso_data->stamp_xsec = new_stamp_xsec; 434 vdso_data->tb_to_xs = new_tb_to_xs; 435 vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec; 436 vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec; 437 smp_wmb(); 438 ++(vdso_data->tb_update_count); 439 } 440 441 #ifdef CONFIG_SMP 442 unsigned long profile_pc(struct pt_regs *regs) 443 { 444 unsigned long pc = instruction_pointer(regs); 445 446 if (in_lock_functions(pc)) 447 return regs->link; 448 449 return pc; 450 } 451 EXPORT_SYMBOL(profile_pc); 452 #endif 453 454 #ifdef CONFIG_PPC_ISERIES 455 456 /* 457 * This function recalibrates the timebase based on the 49-bit time-of-day 458 * value in the Titan chip. The Titan is much more accurate than the value 459 * returned by the service processor for the timebase frequency. 460 */ 461 462 static int __init iSeries_tb_recal(void) 463 { 464 struct div_result divres; 465 unsigned long titan, tb; 466 467 /* Make sure we only run on iSeries */ 468 if (!firmware_has_feature(FW_FEATURE_ISERIES)) 469 return -ENODEV; 470 471 tb = get_tb(); 472 titan = HvCallXm_loadTod(); 473 if ( iSeries_recal_titan ) { 474 unsigned long tb_ticks = tb - iSeries_recal_tb; 475 unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12; 476 unsigned long new_tb_ticks_per_sec = (tb_ticks * USEC_PER_SEC)/titan_usec; 477 unsigned long new_tb_ticks_per_jiffy = (new_tb_ticks_per_sec+(HZ/2))/HZ; 478 long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy; 479 char sign = '+'; 480 /* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */ 481 new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ; 482 483 if ( tick_diff < 0 ) { 484 tick_diff = -tick_diff; 485 sign = '-'; 486 } 487 if ( tick_diff ) { 488 if ( tick_diff < tb_ticks_per_jiffy/25 ) { 489 printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n", 490 new_tb_ticks_per_jiffy, sign, tick_diff ); 491 tb_ticks_per_jiffy = new_tb_ticks_per_jiffy; 492 tb_ticks_per_sec = new_tb_ticks_per_sec; 493 calc_cputime_factors(); 494 div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres ); 495 do_gtod.tb_ticks_per_sec = tb_ticks_per_sec; 496 tb_to_xs = divres.result_low; 497 do_gtod.varp->tb_to_xs = tb_to_xs; 498 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 499 vdso_data->tb_to_xs = tb_to_xs; 500 } 501 else { 502 printk( "Titan recalibrate: FAILED (difference > 4 percent)\n" 503 " new tb_ticks_per_jiffy = %lu\n" 504 " old tb_ticks_per_jiffy = %lu\n", 505 new_tb_ticks_per_jiffy, tb_ticks_per_jiffy ); 506 } 507 } 508 } 509 iSeries_recal_titan = titan; 510 iSeries_recal_tb = tb; 511 512 /* Called here as now we know accurate values for the timebase */ 513 clocksource_init(); 514 return 0; 515 } 516 late_initcall(iSeries_tb_recal); 517 518 /* Called from platform early init */ 519 void __init iSeries_time_init_early(void) 520 { 521 iSeries_recal_tb = get_tb(); 522 iSeries_recal_titan = HvCallXm_loadTod(); 523 } 524 #endif /* CONFIG_PPC_ISERIES */ 525 526 /* 527 * For iSeries shared processors, we have to let the hypervisor 528 * set the hardware decrementer. We set a virtual decrementer 529 * in the lppaca and call the hypervisor if the virtual 530 * decrementer is less than the current value in the hardware 531 * decrementer. (almost always the new decrementer value will 532 * be greater than the current hardware decementer so the hypervisor 533 * call will not be needed) 534 */ 535 536 /* 537 * timer_interrupt - gets called when the decrementer overflows, 538 * with interrupts disabled. 539 */ 540 void timer_interrupt(struct pt_regs * regs) 541 { 542 struct pt_regs *old_regs; 543 int cpu = smp_processor_id(); 544 struct clock_event_device *evt = &per_cpu(decrementers, cpu); 545 u64 now; 546 547 /* Ensure a positive value is written to the decrementer, or else 548 * some CPUs will continuue to take decrementer exceptions */ 549 set_dec(DECREMENTER_MAX); 550 551 #ifdef CONFIG_PPC32 552 if (atomic_read(&ppc_n_lost_interrupts) != 0) 553 do_IRQ(regs); 554 #endif 555 556 now = get_tb_or_rtc(); 557 if (now < per_cpu(decrementer_next_tb, cpu)) { 558 /* not time for this event yet */ 559 now = per_cpu(decrementer_next_tb, cpu) - now; 560 if (now <= DECREMENTER_MAX) 561 set_dec((unsigned int)now - 1); 562 return; 563 } 564 old_regs = set_irq_regs(regs); 565 irq_enter(); 566 567 calculate_steal_time(); 568 569 #ifdef CONFIG_PPC_ISERIES 570 if (firmware_has_feature(FW_FEATURE_ISERIES)) 571 get_lppaca()->int_dword.fields.decr_int = 0; 572 #endif 573 574 /* 575 * We cannot disable the decrementer, so in the period 576 * between this cpu's being marked offline in cpu_online_map 577 * and calling stop-self, it is taking timer interrupts. 578 * Avoid calling into the scheduler rebalancing code if this 579 * is the case. 580 */ 581 if (!cpu_is_offline(cpu)) 582 account_process_time(regs); 583 584 if (evt->event_handler) 585 evt->event_handler(evt); 586 else 587 evt->set_next_event(DECREMENTER_MAX, evt); 588 589 #ifdef CONFIG_PPC_ISERIES 590 if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending()) 591 process_hvlpevents(); 592 #endif 593 594 #ifdef CONFIG_PPC64 595 /* collect purr register values often, for accurate calculations */ 596 if (firmware_has_feature(FW_FEATURE_SPLPAR)) { 597 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array); 598 cu->current_tb = mfspr(SPRN_PURR); 599 } 600 #endif 601 602 irq_exit(); 603 set_irq_regs(old_regs); 604 } 605 606 void wakeup_decrementer(void) 607 { 608 unsigned long ticks; 609 610 /* 611 * The timebase gets saved on sleep and restored on wakeup, 612 * so all we need to do is to reset the decrementer. 613 */ 614 ticks = tb_ticks_since(__get_cpu_var(last_jiffy)); 615 if (ticks < tb_ticks_per_jiffy) 616 ticks = tb_ticks_per_jiffy - ticks; 617 else 618 ticks = 1; 619 set_dec(ticks); 620 } 621 622 #ifdef CONFIG_SMP 623 void __init smp_space_timers(unsigned int max_cpus) 624 { 625 int i; 626 u64 previous_tb = per_cpu(last_jiffy, boot_cpuid); 627 628 /* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */ 629 previous_tb -= tb_ticks_per_jiffy; 630 631 for_each_possible_cpu(i) { 632 if (i == boot_cpuid) 633 continue; 634 per_cpu(last_jiffy, i) = previous_tb; 635 } 636 } 637 #endif 638 639 /* 640 * Scheduler clock - returns current time in nanosec units. 641 * 642 * Note: mulhdu(a, b) (multiply high double unsigned) returns 643 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 644 * are 64-bit unsigned numbers. 645 */ 646 unsigned long long sched_clock(void) 647 { 648 if (__USE_RTC()) 649 return get_rtc(); 650 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 651 } 652 653 static int __init get_freq(char *name, int cells, unsigned long *val) 654 { 655 struct device_node *cpu; 656 const unsigned int *fp; 657 int found = 0; 658 659 /* The cpu node should have timebase and clock frequency properties */ 660 cpu = of_find_node_by_type(NULL, "cpu"); 661 662 if (cpu) { 663 fp = of_get_property(cpu, name, NULL); 664 if (fp) { 665 found = 1; 666 *val = of_read_ulong(fp, cells); 667 } 668 669 of_node_put(cpu); 670 } 671 672 return found; 673 } 674 675 void __init generic_calibrate_decr(void) 676 { 677 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 678 679 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 680 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 681 682 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 683 "(not found)\n"); 684 } 685 686 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 687 688 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 689 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 690 691 printk(KERN_ERR "WARNING: Estimating processor frequency " 692 "(not found)\n"); 693 } 694 695 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 696 /* Set the time base to zero */ 697 mtspr(SPRN_TBWL, 0); 698 mtspr(SPRN_TBWU, 0); 699 700 /* Clear any pending timer interrupts */ 701 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 702 703 /* Enable decrementer interrupt */ 704 mtspr(SPRN_TCR, TCR_DIE); 705 #endif 706 } 707 708 int update_persistent_clock(struct timespec now) 709 { 710 struct rtc_time tm; 711 712 if (!ppc_md.set_rtc_time) 713 return 0; 714 715 to_tm(now.tv_sec + 1 + timezone_offset, &tm); 716 tm.tm_year -= 1900; 717 tm.tm_mon -= 1; 718 719 return ppc_md.set_rtc_time(&tm); 720 } 721 722 unsigned long read_persistent_clock(void) 723 { 724 struct rtc_time tm; 725 static int first = 1; 726 727 /* XXX this is a litle fragile but will work okay in the short term */ 728 if (first) { 729 first = 0; 730 if (ppc_md.time_init) 731 timezone_offset = ppc_md.time_init(); 732 733 /* get_boot_time() isn't guaranteed to be safe to call late */ 734 if (ppc_md.get_boot_time) 735 return ppc_md.get_boot_time() -timezone_offset; 736 } 737 if (!ppc_md.get_rtc_time) 738 return 0; 739 ppc_md.get_rtc_time(&tm); 740 return mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday, 741 tm.tm_hour, tm.tm_min, tm.tm_sec); 742 } 743 744 /* clocksource code */ 745 static cycle_t rtc_read(void) 746 { 747 return (cycle_t)get_rtc(); 748 } 749 750 static cycle_t timebase_read(void) 751 { 752 return (cycle_t)get_tb(); 753 } 754 755 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock) 756 { 757 u64 t2x, stamp_xsec; 758 759 if (clock != &clocksource_timebase) 760 return; 761 762 /* Make userspace gettimeofday spin until we're done. */ 763 ++vdso_data->tb_update_count; 764 smp_mb(); 765 766 /* XXX this assumes clock->shift == 22 */ 767 /* 4611686018 ~= 2^(20+64-22) / 1e9 */ 768 t2x = (u64) clock->mult * 4611686018ULL; 769 stamp_xsec = (u64) xtime.tv_nsec * XSEC_PER_SEC; 770 do_div(stamp_xsec, 1000000000); 771 stamp_xsec += (u64) xtime.tv_sec * XSEC_PER_SEC; 772 update_gtod(clock->cycle_last, stamp_xsec, t2x); 773 } 774 775 void update_vsyscall_tz(void) 776 { 777 /* Make userspace gettimeofday spin until we're done. */ 778 ++vdso_data->tb_update_count; 779 smp_mb(); 780 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 781 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 782 smp_mb(); 783 ++vdso_data->tb_update_count; 784 } 785 786 void __init clocksource_init(void) 787 { 788 struct clocksource *clock; 789 790 if (__USE_RTC()) 791 clock = &clocksource_rtc; 792 else 793 clock = &clocksource_timebase; 794 795 clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift); 796 797 if (clocksource_register(clock)) { 798 printk(KERN_ERR "clocksource: %s is already registered\n", 799 clock->name); 800 return; 801 } 802 803 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 804 clock->name, clock->mult, clock->shift); 805 } 806 807 static int decrementer_set_next_event(unsigned long evt, 808 struct clock_event_device *dev) 809 { 810 __get_cpu_var(decrementer_next_tb) = get_tb_or_rtc() + evt; 811 /* The decrementer interrupts on the 0 -> -1 transition */ 812 if (evt) 813 --evt; 814 set_dec(evt); 815 return 0; 816 } 817 818 static void decrementer_set_mode(enum clock_event_mode mode, 819 struct clock_event_device *dev) 820 { 821 if (mode != CLOCK_EVT_MODE_ONESHOT) 822 decrementer_set_next_event(DECREMENTER_MAX, dev); 823 } 824 825 static void register_decrementer_clockevent(int cpu) 826 { 827 struct clock_event_device *dec = &per_cpu(decrementers, cpu); 828 829 *dec = decrementer_clockevent; 830 dec->cpumask = cpumask_of_cpu(cpu); 831 832 printk(KERN_ERR "clockevent: %s mult[%lx] shift[%d] cpu[%d]\n", 833 dec->name, dec->mult, dec->shift, cpu); 834 835 clockevents_register_device(dec); 836 } 837 838 void init_decrementer_clockevent(void) 839 { 840 int cpu = smp_processor_id(); 841 842 decrementer_clockevent.mult = div_sc(ppc_tb_freq, NSEC_PER_SEC, 843 decrementer_clockevent.shift); 844 decrementer_clockevent.max_delta_ns = 845 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent); 846 decrementer_clockevent.min_delta_ns = 1000; 847 848 register_decrementer_clockevent(cpu); 849 } 850 851 void secondary_cpu_time_init(void) 852 { 853 /* FIME: Should make unrelatred change to move snapshot_timebase 854 * call here ! */ 855 register_decrementer_clockevent(smp_processor_id()); 856 } 857 858 /* This function is only called on the boot processor */ 859 void __init time_init(void) 860 { 861 unsigned long flags; 862 struct div_result res; 863 u64 scale, x; 864 unsigned shift; 865 866 if (__USE_RTC()) { 867 /* 601 processor: dec counts down by 128 every 128ns */ 868 ppc_tb_freq = 1000000000; 869 tb_last_jiffy = get_rtcl(); 870 } else { 871 /* Normal PowerPC with timebase register */ 872 ppc_md.calibrate_decr(); 873 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 874 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 875 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 876 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 877 tb_last_jiffy = get_tb(); 878 } 879 880 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 881 tb_ticks_per_sec = ppc_tb_freq; 882 tb_ticks_per_usec = ppc_tb_freq / 1000000; 883 tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000); 884 calc_cputime_factors(); 885 886 /* 887 * Calculate the length of each tick in ns. It will not be 888 * exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ. 889 * We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq, 890 * rounded up. 891 */ 892 x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1; 893 do_div(x, ppc_tb_freq); 894 tick_nsec = x; 895 last_tick_len = x << TICKLEN_SCALE; 896 897 /* 898 * Compute ticklen_to_xs, which is a factor which gets multiplied 899 * by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value. 900 * It is computed as: 901 * ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9) 902 * where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT 903 * which turns out to be N = 51 - SHIFT_HZ. 904 * This gives the result as a 0.64 fixed-point fraction. 905 * That value is reduced by an offset amounting to 1 xsec per 906 * 2^31 timebase ticks to avoid problems with time going backwards 907 * by 1 xsec when we do timer_recalc_offset due to losing the 908 * fractional xsec. That offset is equal to ppc_tb_freq/2^51 909 * since there are 2^20 xsec in a second. 910 */ 911 div128_by_32((1ULL << 51) - ppc_tb_freq, 0, 912 tb_ticks_per_jiffy << SHIFT_HZ, &res); 913 div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res); 914 ticklen_to_xs = res.result_low; 915 916 /* Compute tb_to_xs from tick_nsec */ 917 tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs); 918 919 /* 920 * Compute scale factor for sched_clock. 921 * The calibrate_decr() function has set tb_ticks_per_sec, 922 * which is the timebase frequency. 923 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 924 * the 128-bit result as a 64.64 fixed-point number. 925 * We then shift that number right until it is less than 1.0, 926 * giving us the scale factor and shift count to use in 927 * sched_clock(). 928 */ 929 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 930 scale = res.result_low; 931 for (shift = 0; res.result_high != 0; ++shift) { 932 scale = (scale >> 1) | (res.result_high << 63); 933 res.result_high >>= 1; 934 } 935 tb_to_ns_scale = scale; 936 tb_to_ns_shift = shift; 937 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 938 boot_tb = get_tb_or_rtc(); 939 940 write_seqlock_irqsave(&xtime_lock, flags); 941 942 /* If platform provided a timezone (pmac), we correct the time */ 943 if (timezone_offset) { 944 sys_tz.tz_minuteswest = -timezone_offset / 60; 945 sys_tz.tz_dsttime = 0; 946 } 947 948 do_gtod.varp = &do_gtod.vars[0]; 949 do_gtod.var_idx = 0; 950 do_gtod.varp->tb_orig_stamp = tb_last_jiffy; 951 __get_cpu_var(last_jiffy) = tb_last_jiffy; 952 do_gtod.varp->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC; 953 do_gtod.tb_ticks_per_sec = tb_ticks_per_sec; 954 do_gtod.varp->tb_to_xs = tb_to_xs; 955 do_gtod.tb_to_us = tb_to_us; 956 957 vdso_data->tb_orig_stamp = tb_last_jiffy; 958 vdso_data->tb_update_count = 0; 959 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 960 vdso_data->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC; 961 vdso_data->tb_to_xs = tb_to_xs; 962 963 time_freq = 0; 964 965 write_sequnlock_irqrestore(&xtime_lock, flags); 966 967 /* Register the clocksource, if we're not running on iSeries */ 968 if (!firmware_has_feature(FW_FEATURE_ISERIES)) 969 clocksource_init(); 970 971 init_decrementer_clockevent(); 972 } 973 974 975 #define FEBRUARY 2 976 #define STARTOFTIME 1970 977 #define SECDAY 86400L 978 #define SECYR (SECDAY * 365) 979 #define leapyear(year) ((year) % 4 == 0 && \ 980 ((year) % 100 != 0 || (year) % 400 == 0)) 981 #define days_in_year(a) (leapyear(a) ? 366 : 365) 982 #define days_in_month(a) (month_days[(a) - 1]) 983 984 static int month_days[12] = { 985 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 986 }; 987 988 /* 989 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK) 990 */ 991 void GregorianDay(struct rtc_time * tm) 992 { 993 int leapsToDate; 994 int lastYear; 995 int day; 996 int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 }; 997 998 lastYear = tm->tm_year - 1; 999 1000 /* 1001 * Number of leap corrections to apply up to end of last year 1002 */ 1003 leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400; 1004 1005 /* 1006 * This year is a leap year if it is divisible by 4 except when it is 1007 * divisible by 100 unless it is divisible by 400 1008 * 1009 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was 1010 */ 1011 day = tm->tm_mon > 2 && leapyear(tm->tm_year); 1012 1013 day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] + 1014 tm->tm_mday; 1015 1016 tm->tm_wday = day % 7; 1017 } 1018 1019 void to_tm(int tim, struct rtc_time * tm) 1020 { 1021 register int i; 1022 register long hms, day; 1023 1024 day = tim / SECDAY; 1025 hms = tim % SECDAY; 1026 1027 /* Hours, minutes, seconds are easy */ 1028 tm->tm_hour = hms / 3600; 1029 tm->tm_min = (hms % 3600) / 60; 1030 tm->tm_sec = (hms % 3600) % 60; 1031 1032 /* Number of years in days */ 1033 for (i = STARTOFTIME; day >= days_in_year(i); i++) 1034 day -= days_in_year(i); 1035 tm->tm_year = i; 1036 1037 /* Number of months in days left */ 1038 if (leapyear(tm->tm_year)) 1039 days_in_month(FEBRUARY) = 29; 1040 for (i = 1; day >= days_in_month(i); i++) 1041 day -= days_in_month(i); 1042 days_in_month(FEBRUARY) = 28; 1043 tm->tm_mon = i; 1044 1045 /* Days are what is left over (+1) from all that. */ 1046 tm->tm_mday = day + 1; 1047 1048 /* 1049 * Determine the day of week 1050 */ 1051 GregorianDay(tm); 1052 } 1053 1054 /* Auxiliary function to compute scaling factors */ 1055 /* Actually the choice of a timebase running at 1/4 the of the bus 1056 * frequency giving resolution of a few tens of nanoseconds is quite nice. 1057 * It makes this computation very precise (27-28 bits typically) which 1058 * is optimistic considering the stability of most processor clock 1059 * oscillators and the precision with which the timebase frequency 1060 * is measured but does not harm. 1061 */ 1062 unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale) 1063 { 1064 unsigned mlt=0, tmp, err; 1065 /* No concern for performance, it's done once: use a stupid 1066 * but safe and compact method to find the multiplier. 1067 */ 1068 1069 for (tmp = 1U<<31; tmp != 0; tmp >>= 1) { 1070 if (mulhwu(inscale, mlt|tmp) < outscale) 1071 mlt |= tmp; 1072 } 1073 1074 /* We might still be off by 1 for the best approximation. 1075 * A side effect of this is that if outscale is too large 1076 * the returned value will be zero. 1077 * Many corner cases have been checked and seem to work, 1078 * some might have been forgotten in the test however. 1079 */ 1080 1081 err = inscale * (mlt+1); 1082 if (err <= inscale/2) 1083 mlt++; 1084 return mlt; 1085 } 1086 1087 /* 1088 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1089 * result. 1090 */ 1091 void div128_by_32(u64 dividend_high, u64 dividend_low, 1092 unsigned divisor, struct div_result *dr) 1093 { 1094 unsigned long a, b, c, d; 1095 unsigned long w, x, y, z; 1096 u64 ra, rb, rc; 1097 1098 a = dividend_high >> 32; 1099 b = dividend_high & 0xffffffff; 1100 c = dividend_low >> 32; 1101 d = dividend_low & 0xffffffff; 1102 1103 w = a / divisor; 1104 ra = ((u64)(a - (w * divisor)) << 32) + b; 1105 1106 rb = ((u64) do_div(ra, divisor) << 32) + c; 1107 x = ra; 1108 1109 rc = ((u64) do_div(rb, divisor) << 32) + d; 1110 y = rb; 1111 1112 do_div(rc, divisor); 1113 z = rc; 1114 1115 dr->result_high = ((u64)w << 32) + x; 1116 dr->result_low = ((u64)y << 32) + z; 1117 1118 } 1119