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 = 16, 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 struct decrementer_clock { 120 struct clock_event_device event; 121 u64 next_tb; 122 }; 123 124 static DEFINE_PER_CPU(struct decrementer_clock, decrementers); 125 126 #ifdef CONFIG_PPC_ISERIES 127 static unsigned long __initdata iSeries_recal_titan; 128 static signed long __initdata iSeries_recal_tb; 129 130 /* Forward declaration is only needed for iSereis compiles */ 131 void __init clocksource_init(void); 132 #endif 133 134 #define XSEC_PER_SEC (1024*1024) 135 136 #ifdef CONFIG_PPC64 137 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC) 138 #else 139 /* compute ((xsec << 12) * max) >> 32 */ 140 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max) 141 #endif 142 143 unsigned long tb_ticks_per_jiffy; 144 unsigned long tb_ticks_per_usec = 100; /* sane default */ 145 EXPORT_SYMBOL(tb_ticks_per_usec); 146 unsigned long tb_ticks_per_sec; 147 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */ 148 u64 tb_to_xs; 149 unsigned tb_to_us; 150 151 #define TICKLEN_SCALE TICK_LENGTH_SHIFT 152 u64 last_tick_len; /* units are ns / 2^TICKLEN_SCALE */ 153 u64 ticklen_to_xs; /* 0.64 fraction */ 154 155 /* If last_tick_len corresponds to about 1/HZ seconds, then 156 last_tick_len << TICKLEN_SHIFT will be about 2^63. */ 157 #define TICKLEN_SHIFT (63 - 30 - TICKLEN_SCALE + SHIFT_HZ) 158 159 DEFINE_SPINLOCK(rtc_lock); 160 EXPORT_SYMBOL_GPL(rtc_lock); 161 162 static u64 tb_to_ns_scale __read_mostly; 163 static unsigned tb_to_ns_shift __read_mostly; 164 static unsigned long boot_tb __read_mostly; 165 166 struct gettimeofday_struct do_gtod; 167 168 extern struct timezone sys_tz; 169 static long timezone_offset; 170 171 unsigned long ppc_proc_freq; 172 EXPORT_SYMBOL(ppc_proc_freq); 173 unsigned long ppc_tb_freq; 174 175 static u64 tb_last_jiffy __cacheline_aligned_in_smp; 176 static DEFINE_PER_CPU(u64, last_jiffy); 177 178 #ifdef CONFIG_VIRT_CPU_ACCOUNTING 179 /* 180 * Factors for converting from cputime_t (timebase ticks) to 181 * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds). 182 * These are all stored as 0.64 fixed-point binary fractions. 183 */ 184 u64 __cputime_jiffies_factor; 185 EXPORT_SYMBOL(__cputime_jiffies_factor); 186 u64 __cputime_msec_factor; 187 EXPORT_SYMBOL(__cputime_msec_factor); 188 u64 __cputime_sec_factor; 189 EXPORT_SYMBOL(__cputime_sec_factor); 190 u64 __cputime_clockt_factor; 191 EXPORT_SYMBOL(__cputime_clockt_factor); 192 193 static void calc_cputime_factors(void) 194 { 195 struct div_result res; 196 197 div128_by_32(HZ, 0, tb_ticks_per_sec, &res); 198 __cputime_jiffies_factor = res.result_low; 199 div128_by_32(1000, 0, tb_ticks_per_sec, &res); 200 __cputime_msec_factor = res.result_low; 201 div128_by_32(1, 0, tb_ticks_per_sec, &res); 202 __cputime_sec_factor = res.result_low; 203 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res); 204 __cputime_clockt_factor = res.result_low; 205 } 206 207 /* 208 * Read the PURR on systems that have it, otherwise the timebase. 209 */ 210 static u64 read_purr(void) 211 { 212 if (cpu_has_feature(CPU_FTR_PURR)) 213 return mfspr(SPRN_PURR); 214 return mftb(); 215 } 216 217 /* 218 * Read the SPURR on systems that have it, otherwise the purr 219 */ 220 static u64 read_spurr(u64 purr) 221 { 222 /* 223 * cpus without PURR won't have a SPURR 224 * We already know the former when we use this, so tell gcc 225 */ 226 if (cpu_has_feature(CPU_FTR_PURR) && cpu_has_feature(CPU_FTR_SPURR)) 227 return mfspr(SPRN_SPURR); 228 return purr; 229 } 230 231 /* 232 * Account time for a transition between system, hard irq 233 * or soft irq state. 234 */ 235 void account_system_vtime(struct task_struct *tsk) 236 { 237 u64 now, nowscaled, delta, deltascaled, sys_time; 238 unsigned long flags; 239 240 local_irq_save(flags); 241 now = read_purr(); 242 nowscaled = read_spurr(now); 243 delta = now - get_paca()->startpurr; 244 deltascaled = nowscaled - get_paca()->startspurr; 245 get_paca()->startpurr = now; 246 get_paca()->startspurr = nowscaled; 247 if (!in_interrupt()) { 248 /* deltascaled includes both user and system time. 249 * Hence scale it based on the purr ratio to estimate 250 * the system time */ 251 sys_time = get_paca()->system_time; 252 if (get_paca()->user_time) 253 deltascaled = deltascaled * sys_time / 254 (sys_time + get_paca()->user_time); 255 delta += sys_time; 256 get_paca()->system_time = 0; 257 } 258 account_system_time(tsk, 0, delta); 259 account_system_time_scaled(tsk, deltascaled); 260 get_paca()->purrdelta = delta; 261 get_paca()->spurrdelta = deltascaled; 262 local_irq_restore(flags); 263 } 264 265 /* 266 * Transfer the user and system times accumulated in the paca 267 * by the exception entry and exit code to the generic process 268 * user and system time records. 269 * Must be called with interrupts disabled. 270 */ 271 void account_process_tick(struct task_struct *tsk, int user_tick) 272 { 273 cputime_t utime, utimescaled; 274 275 utime = get_paca()->user_time; 276 get_paca()->user_time = 0; 277 account_user_time(tsk, utime); 278 279 /* Estimate the scaled utime by scaling the real utime based 280 * on the last spurr to purr ratio */ 281 utimescaled = utime * get_paca()->spurrdelta / get_paca()->purrdelta; 282 get_paca()->spurrdelta = get_paca()->purrdelta = 0; 283 account_user_time_scaled(tsk, utimescaled); 284 } 285 286 /* 287 * Stuff for accounting stolen time. 288 */ 289 struct cpu_purr_data { 290 int initialized; /* thread is running */ 291 u64 tb; /* last TB value read */ 292 u64 purr; /* last PURR value read */ 293 u64 spurr; /* last SPURR value read */ 294 }; 295 296 /* 297 * Each entry in the cpu_purr_data array is manipulated only by its 298 * "owner" cpu -- usually in the timer interrupt but also occasionally 299 * in process context for cpu online. As long as cpus do not touch 300 * each others' cpu_purr_data, disabling local interrupts is 301 * sufficient to serialize accesses. 302 */ 303 static DEFINE_PER_CPU(struct cpu_purr_data, cpu_purr_data); 304 305 static void snapshot_tb_and_purr(void *data) 306 { 307 unsigned long flags; 308 struct cpu_purr_data *p = &__get_cpu_var(cpu_purr_data); 309 310 local_irq_save(flags); 311 p->tb = get_tb_or_rtc(); 312 p->purr = mfspr(SPRN_PURR); 313 wmb(); 314 p->initialized = 1; 315 local_irq_restore(flags); 316 } 317 318 /* 319 * Called during boot when all cpus have come up. 320 */ 321 void snapshot_timebases(void) 322 { 323 if (!cpu_has_feature(CPU_FTR_PURR)) 324 return; 325 on_each_cpu(snapshot_tb_and_purr, NULL, 0, 1); 326 } 327 328 /* 329 * Must be called with interrupts disabled. 330 */ 331 void calculate_steal_time(void) 332 { 333 u64 tb, purr; 334 s64 stolen; 335 struct cpu_purr_data *pme; 336 337 pme = &__get_cpu_var(cpu_purr_data); 338 if (!pme->initialized) 339 return; /* !CPU_FTR_PURR or early in early boot */ 340 tb = mftb(); 341 purr = mfspr(SPRN_PURR); 342 stolen = (tb - pme->tb) - (purr - pme->purr); 343 if (stolen > 0) 344 account_steal_time(current, stolen); 345 pme->tb = tb; 346 pme->purr = purr; 347 } 348 349 #ifdef CONFIG_PPC_SPLPAR 350 /* 351 * Must be called before the cpu is added to the online map when 352 * a cpu is being brought up at runtime. 353 */ 354 static void snapshot_purr(void) 355 { 356 struct cpu_purr_data *pme; 357 unsigned long flags; 358 359 if (!cpu_has_feature(CPU_FTR_PURR)) 360 return; 361 local_irq_save(flags); 362 pme = &__get_cpu_var(cpu_purr_data); 363 pme->tb = mftb(); 364 pme->purr = mfspr(SPRN_PURR); 365 pme->initialized = 1; 366 local_irq_restore(flags); 367 } 368 369 #endif /* CONFIG_PPC_SPLPAR */ 370 371 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */ 372 #define calc_cputime_factors() 373 #define calculate_steal_time() do { } while (0) 374 #endif 375 376 #if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR)) 377 #define snapshot_purr() do { } while (0) 378 #endif 379 380 /* 381 * Called when a cpu comes up after the system has finished booting, 382 * i.e. as a result of a hotplug cpu action. 383 */ 384 void snapshot_timebase(void) 385 { 386 __get_cpu_var(last_jiffy) = get_tb_or_rtc(); 387 snapshot_purr(); 388 } 389 390 void __delay(unsigned long loops) 391 { 392 unsigned long start; 393 int diff; 394 395 if (__USE_RTC()) { 396 start = get_rtcl(); 397 do { 398 /* the RTCL register wraps at 1000000000 */ 399 diff = get_rtcl() - start; 400 if (diff < 0) 401 diff += 1000000000; 402 } while (diff < loops); 403 } else { 404 start = get_tbl(); 405 while (get_tbl() - start < loops) 406 HMT_low(); 407 HMT_medium(); 408 } 409 } 410 EXPORT_SYMBOL(__delay); 411 412 void udelay(unsigned long usecs) 413 { 414 __delay(tb_ticks_per_usec * usecs); 415 } 416 EXPORT_SYMBOL(udelay); 417 418 419 /* 420 * There are two copies of tb_to_xs and stamp_xsec so that no 421 * lock is needed to access and use these values in 422 * do_gettimeofday. We alternate the copies and as long as a 423 * reasonable time elapses between changes, there will never 424 * be inconsistent values. ntpd has a minimum of one minute 425 * between updates. 426 */ 427 static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec, 428 u64 new_tb_to_xs) 429 { 430 unsigned temp_idx; 431 struct gettimeofday_vars *temp_varp; 432 433 temp_idx = (do_gtod.var_idx == 0); 434 temp_varp = &do_gtod.vars[temp_idx]; 435 436 temp_varp->tb_to_xs = new_tb_to_xs; 437 temp_varp->tb_orig_stamp = new_tb_stamp; 438 temp_varp->stamp_xsec = new_stamp_xsec; 439 smp_mb(); 440 do_gtod.varp = temp_varp; 441 do_gtod.var_idx = temp_idx; 442 443 /* 444 * tb_update_count is used to allow the userspace gettimeofday code 445 * to assure itself that it sees a consistent view of the tb_to_xs and 446 * stamp_xsec variables. It reads the tb_update_count, then reads 447 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 448 * the two values of tb_update_count match and are even then the 449 * tb_to_xs and stamp_xsec values are consistent. If not, then it 450 * loops back and reads them again until this criteria is met. 451 * We expect the caller to have done the first increment of 452 * vdso_data->tb_update_count already. 453 */ 454 vdso_data->tb_orig_stamp = new_tb_stamp; 455 vdso_data->stamp_xsec = new_stamp_xsec; 456 vdso_data->tb_to_xs = new_tb_to_xs; 457 vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec; 458 vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec; 459 smp_wmb(); 460 ++(vdso_data->tb_update_count); 461 } 462 463 #ifdef CONFIG_SMP 464 unsigned long profile_pc(struct pt_regs *regs) 465 { 466 unsigned long pc = instruction_pointer(regs); 467 468 if (in_lock_functions(pc)) 469 return regs->link; 470 471 return pc; 472 } 473 EXPORT_SYMBOL(profile_pc); 474 #endif 475 476 #ifdef CONFIG_PPC_ISERIES 477 478 /* 479 * This function recalibrates the timebase based on the 49-bit time-of-day 480 * value in the Titan chip. The Titan is much more accurate than the value 481 * returned by the service processor for the timebase frequency. 482 */ 483 484 static int __init iSeries_tb_recal(void) 485 { 486 struct div_result divres; 487 unsigned long titan, tb; 488 489 /* Make sure we only run on iSeries */ 490 if (!firmware_has_feature(FW_FEATURE_ISERIES)) 491 return -ENODEV; 492 493 tb = get_tb(); 494 titan = HvCallXm_loadTod(); 495 if ( iSeries_recal_titan ) { 496 unsigned long tb_ticks = tb - iSeries_recal_tb; 497 unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12; 498 unsigned long new_tb_ticks_per_sec = (tb_ticks * USEC_PER_SEC)/titan_usec; 499 unsigned long new_tb_ticks_per_jiffy = (new_tb_ticks_per_sec+(HZ/2))/HZ; 500 long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy; 501 char sign = '+'; 502 /* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */ 503 new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ; 504 505 if ( tick_diff < 0 ) { 506 tick_diff = -tick_diff; 507 sign = '-'; 508 } 509 if ( tick_diff ) { 510 if ( tick_diff < tb_ticks_per_jiffy/25 ) { 511 printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n", 512 new_tb_ticks_per_jiffy, sign, tick_diff ); 513 tb_ticks_per_jiffy = new_tb_ticks_per_jiffy; 514 tb_ticks_per_sec = new_tb_ticks_per_sec; 515 calc_cputime_factors(); 516 div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres ); 517 do_gtod.tb_ticks_per_sec = tb_ticks_per_sec; 518 tb_to_xs = divres.result_low; 519 do_gtod.varp->tb_to_xs = tb_to_xs; 520 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 521 vdso_data->tb_to_xs = tb_to_xs; 522 } 523 else { 524 printk( "Titan recalibrate: FAILED (difference > 4 percent)\n" 525 " new tb_ticks_per_jiffy = %lu\n" 526 " old tb_ticks_per_jiffy = %lu\n", 527 new_tb_ticks_per_jiffy, tb_ticks_per_jiffy ); 528 } 529 } 530 } 531 iSeries_recal_titan = titan; 532 iSeries_recal_tb = tb; 533 534 /* Called here as now we know accurate values for the timebase */ 535 clocksource_init(); 536 return 0; 537 } 538 late_initcall(iSeries_tb_recal); 539 540 /* Called from platform early init */ 541 void __init iSeries_time_init_early(void) 542 { 543 iSeries_recal_tb = get_tb(); 544 iSeries_recal_titan = HvCallXm_loadTod(); 545 } 546 #endif /* CONFIG_PPC_ISERIES */ 547 548 /* 549 * For iSeries shared processors, we have to let the hypervisor 550 * set the hardware decrementer. We set a virtual decrementer 551 * in the lppaca and call the hypervisor if the virtual 552 * decrementer is less than the current value in the hardware 553 * decrementer. (almost always the new decrementer value will 554 * be greater than the current hardware decementer so the hypervisor 555 * call will not be needed) 556 */ 557 558 /* 559 * timer_interrupt - gets called when the decrementer overflows, 560 * with interrupts disabled. 561 */ 562 void timer_interrupt(struct pt_regs * regs) 563 { 564 struct pt_regs *old_regs; 565 struct decrementer_clock *decrementer = &__get_cpu_var(decrementers); 566 struct clock_event_device *evt = &decrementer->event; 567 u64 now; 568 569 /* Ensure a positive value is written to the decrementer, or else 570 * some CPUs will continuue to take decrementer exceptions */ 571 set_dec(DECREMENTER_MAX); 572 573 #ifdef CONFIG_PPC32 574 if (atomic_read(&ppc_n_lost_interrupts) != 0) 575 do_IRQ(regs); 576 #endif 577 578 now = get_tb_or_rtc(); 579 if (now < decrementer->next_tb) { 580 /* not time for this event yet */ 581 now = decrementer->next_tb - now; 582 if (now <= DECREMENTER_MAX) 583 set_dec((int)now); 584 return; 585 } 586 old_regs = set_irq_regs(regs); 587 irq_enter(); 588 589 calculate_steal_time(); 590 591 #ifdef CONFIG_PPC_ISERIES 592 if (firmware_has_feature(FW_FEATURE_ISERIES)) 593 get_lppaca()->int_dword.fields.decr_int = 0; 594 #endif 595 596 if (evt->event_handler) 597 evt->event_handler(evt); 598 599 #ifdef CONFIG_PPC_ISERIES 600 if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending()) 601 process_hvlpevents(); 602 #endif 603 604 #ifdef CONFIG_PPC64 605 /* collect purr register values often, for accurate calculations */ 606 if (firmware_has_feature(FW_FEATURE_SPLPAR)) { 607 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array); 608 cu->current_tb = mfspr(SPRN_PURR); 609 } 610 #endif 611 612 irq_exit(); 613 set_irq_regs(old_regs); 614 } 615 616 void wakeup_decrementer(void) 617 { 618 unsigned long ticks; 619 620 /* 621 * The timebase gets saved on sleep and restored on wakeup, 622 * so all we need to do is to reset the decrementer. 623 */ 624 ticks = tb_ticks_since(__get_cpu_var(last_jiffy)); 625 if (ticks < tb_ticks_per_jiffy) 626 ticks = tb_ticks_per_jiffy - ticks; 627 else 628 ticks = 1; 629 set_dec(ticks); 630 } 631 632 #ifdef CONFIG_SMP 633 void __init smp_space_timers(unsigned int max_cpus) 634 { 635 int i; 636 u64 previous_tb = per_cpu(last_jiffy, boot_cpuid); 637 638 /* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */ 639 previous_tb -= tb_ticks_per_jiffy; 640 641 for_each_possible_cpu(i) { 642 if (i == boot_cpuid) 643 continue; 644 per_cpu(last_jiffy, i) = previous_tb; 645 } 646 } 647 #endif 648 649 /* 650 * Scheduler clock - returns current time in nanosec units. 651 * 652 * Note: mulhdu(a, b) (multiply high double unsigned) returns 653 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 654 * are 64-bit unsigned numbers. 655 */ 656 unsigned long long sched_clock(void) 657 { 658 if (__USE_RTC()) 659 return get_rtc(); 660 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 661 } 662 663 static int __init get_freq(char *name, int cells, unsigned long *val) 664 { 665 struct device_node *cpu; 666 const unsigned int *fp; 667 int found = 0; 668 669 /* The cpu node should have timebase and clock frequency properties */ 670 cpu = of_find_node_by_type(NULL, "cpu"); 671 672 if (cpu) { 673 fp = of_get_property(cpu, name, NULL); 674 if (fp) { 675 found = 1; 676 *val = of_read_ulong(fp, cells); 677 } 678 679 of_node_put(cpu); 680 } 681 682 return found; 683 } 684 685 void __init generic_calibrate_decr(void) 686 { 687 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 688 689 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 690 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 691 692 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 693 "(not found)\n"); 694 } 695 696 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 697 698 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 699 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 700 701 printk(KERN_ERR "WARNING: Estimating processor frequency " 702 "(not found)\n"); 703 } 704 705 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 706 /* Set the time base to zero */ 707 mtspr(SPRN_TBWL, 0); 708 mtspr(SPRN_TBWU, 0); 709 710 /* Clear any pending timer interrupts */ 711 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 712 713 /* Enable decrementer interrupt */ 714 mtspr(SPRN_TCR, TCR_DIE); 715 #endif 716 } 717 718 int update_persistent_clock(struct timespec now) 719 { 720 struct rtc_time tm; 721 722 if (!ppc_md.set_rtc_time) 723 return 0; 724 725 to_tm(now.tv_sec + 1 + timezone_offset, &tm); 726 tm.tm_year -= 1900; 727 tm.tm_mon -= 1; 728 729 return ppc_md.set_rtc_time(&tm); 730 } 731 732 unsigned long read_persistent_clock(void) 733 { 734 struct rtc_time tm; 735 static int first = 1; 736 737 /* XXX this is a litle fragile but will work okay in the short term */ 738 if (first) { 739 first = 0; 740 if (ppc_md.time_init) 741 timezone_offset = ppc_md.time_init(); 742 743 /* get_boot_time() isn't guaranteed to be safe to call late */ 744 if (ppc_md.get_boot_time) 745 return ppc_md.get_boot_time() -timezone_offset; 746 } 747 if (!ppc_md.get_rtc_time) 748 return 0; 749 ppc_md.get_rtc_time(&tm); 750 return mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday, 751 tm.tm_hour, tm.tm_min, tm.tm_sec); 752 } 753 754 /* clocksource code */ 755 static cycle_t rtc_read(void) 756 { 757 return (cycle_t)get_rtc(); 758 } 759 760 static cycle_t timebase_read(void) 761 { 762 return (cycle_t)get_tb(); 763 } 764 765 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock) 766 { 767 u64 t2x, stamp_xsec; 768 769 if (clock != &clocksource_timebase) 770 return; 771 772 /* Make userspace gettimeofday spin until we're done. */ 773 ++vdso_data->tb_update_count; 774 smp_mb(); 775 776 /* XXX this assumes clock->shift == 22 */ 777 /* 4611686018 ~= 2^(20+64-22) / 1e9 */ 778 t2x = (u64) clock->mult * 4611686018ULL; 779 stamp_xsec = (u64) xtime.tv_nsec * XSEC_PER_SEC; 780 do_div(stamp_xsec, 1000000000); 781 stamp_xsec += (u64) xtime.tv_sec * XSEC_PER_SEC; 782 update_gtod(clock->cycle_last, stamp_xsec, t2x); 783 } 784 785 void update_vsyscall_tz(void) 786 { 787 /* Make userspace gettimeofday spin until we're done. */ 788 ++vdso_data->tb_update_count; 789 smp_mb(); 790 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 791 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 792 smp_mb(); 793 ++vdso_data->tb_update_count; 794 } 795 796 void __init clocksource_init(void) 797 { 798 struct clocksource *clock; 799 800 if (__USE_RTC()) 801 clock = &clocksource_rtc; 802 else 803 clock = &clocksource_timebase; 804 805 clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift); 806 807 if (clocksource_register(clock)) { 808 printk(KERN_ERR "clocksource: %s is already registered\n", 809 clock->name); 810 return; 811 } 812 813 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 814 clock->name, clock->mult, clock->shift); 815 } 816 817 static int decrementer_set_next_event(unsigned long evt, 818 struct clock_event_device *dev) 819 { 820 __get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt; 821 set_dec(evt); 822 return 0; 823 } 824 825 static void decrementer_set_mode(enum clock_event_mode mode, 826 struct clock_event_device *dev) 827 { 828 if (mode != CLOCK_EVT_MODE_ONESHOT) 829 decrementer_set_next_event(DECREMENTER_MAX, dev); 830 } 831 832 static void register_decrementer_clockevent(int cpu) 833 { 834 struct clock_event_device *dec = &per_cpu(decrementers, cpu).event; 835 836 *dec = decrementer_clockevent; 837 dec->cpumask = cpumask_of_cpu(cpu); 838 839 printk(KERN_DEBUG "clockevent: %s mult[%lx] shift[%d] cpu[%d]\n", 840 dec->name, dec->mult, dec->shift, cpu); 841 842 clockevents_register_device(dec); 843 } 844 845 static void __init init_decrementer_clockevent(void) 846 { 847 int cpu = smp_processor_id(); 848 849 decrementer_clockevent.mult = div_sc(ppc_tb_freq, NSEC_PER_SEC, 850 decrementer_clockevent.shift); 851 decrementer_clockevent.max_delta_ns = 852 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent); 853 decrementer_clockevent.min_delta_ns = 854 clockevent_delta2ns(2, &decrementer_clockevent); 855 856 register_decrementer_clockevent(cpu); 857 } 858 859 void secondary_cpu_time_init(void) 860 { 861 /* FIME: Should make unrelatred change to move snapshot_timebase 862 * call here ! */ 863 register_decrementer_clockevent(smp_processor_id()); 864 } 865 866 /* This function is only called on the boot processor */ 867 void __init time_init(void) 868 { 869 unsigned long flags; 870 struct div_result res; 871 u64 scale, x; 872 unsigned shift; 873 874 if (__USE_RTC()) { 875 /* 601 processor: dec counts down by 128 every 128ns */ 876 ppc_tb_freq = 1000000000; 877 tb_last_jiffy = get_rtcl(); 878 } else { 879 /* Normal PowerPC with timebase register */ 880 ppc_md.calibrate_decr(); 881 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 882 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 883 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 884 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 885 tb_last_jiffy = get_tb(); 886 } 887 888 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 889 tb_ticks_per_sec = ppc_tb_freq; 890 tb_ticks_per_usec = ppc_tb_freq / 1000000; 891 tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000); 892 calc_cputime_factors(); 893 894 /* 895 * Calculate the length of each tick in ns. It will not be 896 * exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ. 897 * We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq, 898 * rounded up. 899 */ 900 x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1; 901 do_div(x, ppc_tb_freq); 902 tick_nsec = x; 903 last_tick_len = x << TICKLEN_SCALE; 904 905 /* 906 * Compute ticklen_to_xs, which is a factor which gets multiplied 907 * by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value. 908 * It is computed as: 909 * ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9) 910 * where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT 911 * which turns out to be N = 51 - SHIFT_HZ. 912 * This gives the result as a 0.64 fixed-point fraction. 913 * That value is reduced by an offset amounting to 1 xsec per 914 * 2^31 timebase ticks to avoid problems with time going backwards 915 * by 1 xsec when we do timer_recalc_offset due to losing the 916 * fractional xsec. That offset is equal to ppc_tb_freq/2^51 917 * since there are 2^20 xsec in a second. 918 */ 919 div128_by_32((1ULL << 51) - ppc_tb_freq, 0, 920 tb_ticks_per_jiffy << SHIFT_HZ, &res); 921 div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res); 922 ticklen_to_xs = res.result_low; 923 924 /* Compute tb_to_xs from tick_nsec */ 925 tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs); 926 927 /* 928 * Compute scale factor for sched_clock. 929 * The calibrate_decr() function has set tb_ticks_per_sec, 930 * which is the timebase frequency. 931 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 932 * the 128-bit result as a 64.64 fixed-point number. 933 * We then shift that number right until it is less than 1.0, 934 * giving us the scale factor and shift count to use in 935 * sched_clock(). 936 */ 937 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 938 scale = res.result_low; 939 for (shift = 0; res.result_high != 0; ++shift) { 940 scale = (scale >> 1) | (res.result_high << 63); 941 res.result_high >>= 1; 942 } 943 tb_to_ns_scale = scale; 944 tb_to_ns_shift = shift; 945 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 946 boot_tb = get_tb_or_rtc(); 947 948 write_seqlock_irqsave(&xtime_lock, flags); 949 950 /* If platform provided a timezone (pmac), we correct the time */ 951 if (timezone_offset) { 952 sys_tz.tz_minuteswest = -timezone_offset / 60; 953 sys_tz.tz_dsttime = 0; 954 } 955 956 do_gtod.varp = &do_gtod.vars[0]; 957 do_gtod.var_idx = 0; 958 do_gtod.varp->tb_orig_stamp = tb_last_jiffy; 959 __get_cpu_var(last_jiffy) = tb_last_jiffy; 960 do_gtod.varp->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC; 961 do_gtod.tb_ticks_per_sec = tb_ticks_per_sec; 962 do_gtod.varp->tb_to_xs = tb_to_xs; 963 do_gtod.tb_to_us = tb_to_us; 964 965 vdso_data->tb_orig_stamp = tb_last_jiffy; 966 vdso_data->tb_update_count = 0; 967 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 968 vdso_data->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC; 969 vdso_data->tb_to_xs = tb_to_xs; 970 971 time_freq = 0; 972 973 write_sequnlock_irqrestore(&xtime_lock, flags); 974 975 /* Register the clocksource, if we're not running on iSeries */ 976 if (!firmware_has_feature(FW_FEATURE_ISERIES)) 977 clocksource_init(); 978 979 init_decrementer_clockevent(); 980 } 981 982 983 #define FEBRUARY 2 984 #define STARTOFTIME 1970 985 #define SECDAY 86400L 986 #define SECYR (SECDAY * 365) 987 #define leapyear(year) ((year) % 4 == 0 && \ 988 ((year) % 100 != 0 || (year) % 400 == 0)) 989 #define days_in_year(a) (leapyear(a) ? 366 : 365) 990 #define days_in_month(a) (month_days[(a) - 1]) 991 992 static int month_days[12] = { 993 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 994 }; 995 996 /* 997 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK) 998 */ 999 void GregorianDay(struct rtc_time * tm) 1000 { 1001 int leapsToDate; 1002 int lastYear; 1003 int day; 1004 int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 }; 1005 1006 lastYear = tm->tm_year - 1; 1007 1008 /* 1009 * Number of leap corrections to apply up to end of last year 1010 */ 1011 leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400; 1012 1013 /* 1014 * This year is a leap year if it is divisible by 4 except when it is 1015 * divisible by 100 unless it is divisible by 400 1016 * 1017 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was 1018 */ 1019 day = tm->tm_mon > 2 && leapyear(tm->tm_year); 1020 1021 day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] + 1022 tm->tm_mday; 1023 1024 tm->tm_wday = day % 7; 1025 } 1026 1027 void to_tm(int tim, struct rtc_time * tm) 1028 { 1029 register int i; 1030 register long hms, day; 1031 1032 day = tim / SECDAY; 1033 hms = tim % SECDAY; 1034 1035 /* Hours, minutes, seconds are easy */ 1036 tm->tm_hour = hms / 3600; 1037 tm->tm_min = (hms % 3600) / 60; 1038 tm->tm_sec = (hms % 3600) % 60; 1039 1040 /* Number of years in days */ 1041 for (i = STARTOFTIME; day >= days_in_year(i); i++) 1042 day -= days_in_year(i); 1043 tm->tm_year = i; 1044 1045 /* Number of months in days left */ 1046 if (leapyear(tm->tm_year)) 1047 days_in_month(FEBRUARY) = 29; 1048 for (i = 1; day >= days_in_month(i); i++) 1049 day -= days_in_month(i); 1050 days_in_month(FEBRUARY) = 28; 1051 tm->tm_mon = i; 1052 1053 /* Days are what is left over (+1) from all that. */ 1054 tm->tm_mday = day + 1; 1055 1056 /* 1057 * Determine the day of week 1058 */ 1059 GregorianDay(tm); 1060 } 1061 1062 /* Auxiliary function to compute scaling factors */ 1063 /* Actually the choice of a timebase running at 1/4 the of the bus 1064 * frequency giving resolution of a few tens of nanoseconds is quite nice. 1065 * It makes this computation very precise (27-28 bits typically) which 1066 * is optimistic considering the stability of most processor clock 1067 * oscillators and the precision with which the timebase frequency 1068 * is measured but does not harm. 1069 */ 1070 unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale) 1071 { 1072 unsigned mlt=0, tmp, err; 1073 /* No concern for performance, it's done once: use a stupid 1074 * but safe and compact method to find the multiplier. 1075 */ 1076 1077 for (tmp = 1U<<31; tmp != 0; tmp >>= 1) { 1078 if (mulhwu(inscale, mlt|tmp) < outscale) 1079 mlt |= tmp; 1080 } 1081 1082 /* We might still be off by 1 for the best approximation. 1083 * A side effect of this is that if outscale is too large 1084 * the returned value will be zero. 1085 * Many corner cases have been checked and seem to work, 1086 * some might have been forgotten in the test however. 1087 */ 1088 1089 err = inscale * (mlt+1); 1090 if (err <= inscale/2) 1091 mlt++; 1092 return mlt; 1093 } 1094 1095 /* 1096 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1097 * result. 1098 */ 1099 void div128_by_32(u64 dividend_high, u64 dividend_low, 1100 unsigned divisor, struct div_result *dr) 1101 { 1102 unsigned long a, b, c, d; 1103 unsigned long w, x, y, z; 1104 u64 ra, rb, rc; 1105 1106 a = dividend_high >> 32; 1107 b = dividend_high & 0xffffffff; 1108 c = dividend_low >> 32; 1109 d = dividend_low & 0xffffffff; 1110 1111 w = a / divisor; 1112 ra = ((u64)(a - (w * divisor)) << 32) + b; 1113 1114 rb = ((u64) do_div(ra, divisor) << 32) + c; 1115 x = ra; 1116 1117 rc = ((u64) do_div(rb, divisor) << 32) + d; 1118 y = rb; 1119 1120 do_div(rc, divisor); 1121 z = rc; 1122 1123 dr->result_high = ((u64)w << 32) + x; 1124 dr->result_low = ((u64)y << 32) + z; 1125 1126 } 1127