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 #include <linux/delay.h> 56 #include <linux/perf_event.h> 57 58 #include <asm/io.h> 59 #include <asm/processor.h> 60 #include <asm/nvram.h> 61 #include <asm/cache.h> 62 #include <asm/machdep.h> 63 #include <asm/uaccess.h> 64 #include <asm/time.h> 65 #include <asm/prom.h> 66 #include <asm/irq.h> 67 #include <asm/div64.h> 68 #include <asm/smp.h> 69 #include <asm/vdso_datapage.h> 70 #include <asm/firmware.h> 71 #include <asm/cputime.h> 72 #ifdef CONFIG_PPC_ISERIES 73 #include <asm/iseries/it_lp_queue.h> 74 #include <asm/iseries/hv_call_xm.h> 75 #endif 76 77 /* powerpc clocksource/clockevent code */ 78 79 #include <linux/clockchips.h> 80 #include <linux/clocksource.h> 81 82 static cycle_t rtc_read(struct clocksource *); 83 static struct clocksource clocksource_rtc = { 84 .name = "rtc", 85 .rating = 400, 86 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 87 .mask = CLOCKSOURCE_MASK(64), 88 .shift = 22, 89 .mult = 0, /* To be filled in */ 90 .read = rtc_read, 91 }; 92 93 static cycle_t timebase_read(struct clocksource *); 94 static struct clocksource clocksource_timebase = { 95 .name = "timebase", 96 .rating = 400, 97 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 98 .mask = CLOCKSOURCE_MASK(64), 99 .shift = 22, 100 .mult = 0, /* To be filled in */ 101 .read = timebase_read, 102 }; 103 104 #define DECREMENTER_MAX 0x7fffffff 105 106 static int decrementer_set_next_event(unsigned long evt, 107 struct clock_event_device *dev); 108 static void decrementer_set_mode(enum clock_event_mode mode, 109 struct clock_event_device *dev); 110 111 static struct clock_event_device decrementer_clockevent = { 112 .name = "decrementer", 113 .rating = 200, 114 .shift = 0, /* To be filled in */ 115 .mult = 0, /* To be filled in */ 116 .irq = 0, 117 .set_next_event = decrementer_set_next_event, 118 .set_mode = decrementer_set_mode, 119 .features = CLOCK_EVT_FEAT_ONESHOT, 120 }; 121 122 struct decrementer_clock { 123 struct clock_event_device event; 124 u64 next_tb; 125 }; 126 127 static DEFINE_PER_CPU(struct decrementer_clock, decrementers); 128 129 #ifdef CONFIG_PPC_ISERIES 130 static unsigned long __initdata iSeries_recal_titan; 131 static signed long __initdata iSeries_recal_tb; 132 133 /* Forward declaration is only needed for iSereis compiles */ 134 static void __init clocksource_init(void); 135 #endif 136 137 #define XSEC_PER_SEC (1024*1024) 138 139 #ifdef CONFIG_PPC64 140 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC) 141 #else 142 /* compute ((xsec << 12) * max) >> 32 */ 143 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max) 144 #endif 145 146 unsigned long tb_ticks_per_jiffy; 147 unsigned long tb_ticks_per_usec = 100; /* sane default */ 148 EXPORT_SYMBOL(tb_ticks_per_usec); 149 unsigned long tb_ticks_per_sec; 150 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */ 151 u64 tb_to_xs; 152 unsigned tb_to_us; 153 154 #define TICKLEN_SCALE NTP_SCALE_SHIFT 155 static u64 last_tick_len; /* units are ns / 2^TICKLEN_SCALE */ 156 static u64 ticklen_to_xs; /* 0.64 fraction */ 157 158 /* If last_tick_len corresponds to about 1/HZ seconds, then 159 last_tick_len << TICKLEN_SHIFT will be about 2^63. */ 160 #define TICKLEN_SHIFT (63 - 30 - TICKLEN_SCALE + SHIFT_HZ) 161 162 DEFINE_SPINLOCK(rtc_lock); 163 EXPORT_SYMBOL_GPL(rtc_lock); 164 165 static u64 tb_to_ns_scale __read_mostly; 166 static unsigned tb_to_ns_shift __read_mostly; 167 static unsigned long boot_tb __read_mostly; 168 169 extern struct timezone sys_tz; 170 static long timezone_offset; 171 172 unsigned long ppc_proc_freq; 173 EXPORT_SYMBOL(ppc_proc_freq); 174 unsigned long ppc_tb_freq; 175 176 static u64 tb_last_jiffy __cacheline_aligned_in_smp; 177 static DEFINE_PER_CPU(u64, last_jiffy); 178 179 #ifdef CONFIG_VIRT_CPU_ACCOUNTING 180 /* 181 * Factors for converting from cputime_t (timebase ticks) to 182 * jiffies, milliseconds, seconds, and clock_t (1/USER_HZ seconds). 183 * These are all stored as 0.64 fixed-point binary fractions. 184 */ 185 u64 __cputime_jiffies_factor; 186 EXPORT_SYMBOL(__cputime_jiffies_factor); 187 u64 __cputime_msec_factor; 188 EXPORT_SYMBOL(__cputime_msec_factor); 189 u64 __cputime_sec_factor; 190 EXPORT_SYMBOL(__cputime_sec_factor); 191 u64 __cputime_clockt_factor; 192 EXPORT_SYMBOL(__cputime_clockt_factor); 193 DEFINE_PER_CPU(unsigned long, cputime_last_delta); 194 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta); 195 196 static void calc_cputime_factors(void) 197 { 198 struct div_result res; 199 200 div128_by_32(HZ, 0, tb_ticks_per_sec, &res); 201 __cputime_jiffies_factor = res.result_low; 202 div128_by_32(1000, 0, tb_ticks_per_sec, &res); 203 __cputime_msec_factor = res.result_low; 204 div128_by_32(1, 0, tb_ticks_per_sec, &res); 205 __cputime_sec_factor = res.result_low; 206 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res); 207 __cputime_clockt_factor = res.result_low; 208 } 209 210 /* 211 * Read the PURR on systems that have it, otherwise the timebase. 212 */ 213 static u64 read_purr(void) 214 { 215 if (cpu_has_feature(CPU_FTR_PURR)) 216 return mfspr(SPRN_PURR); 217 return mftb(); 218 } 219 220 /* 221 * Read the SPURR on systems that have it, otherwise the purr 222 */ 223 static u64 read_spurr(u64 purr) 224 { 225 /* 226 * cpus without PURR won't have a SPURR 227 * We already know the former when we use this, so tell gcc 228 */ 229 if (cpu_has_feature(CPU_FTR_PURR) && cpu_has_feature(CPU_FTR_SPURR)) 230 return mfspr(SPRN_SPURR); 231 return purr; 232 } 233 234 /* 235 * Account time for a transition between system, hard irq 236 * or soft irq state. 237 */ 238 void account_system_vtime(struct task_struct *tsk) 239 { 240 u64 now, nowscaled, delta, deltascaled, sys_time; 241 unsigned long flags; 242 243 local_irq_save(flags); 244 now = read_purr(); 245 nowscaled = read_spurr(now); 246 delta = now - get_paca()->startpurr; 247 deltascaled = nowscaled - get_paca()->startspurr; 248 get_paca()->startpurr = now; 249 get_paca()->startspurr = nowscaled; 250 if (!in_interrupt()) { 251 /* deltascaled includes both user and system time. 252 * Hence scale it based on the purr ratio to estimate 253 * the system time */ 254 sys_time = get_paca()->system_time; 255 if (get_paca()->user_time) 256 deltascaled = deltascaled * sys_time / 257 (sys_time + get_paca()->user_time); 258 delta += sys_time; 259 get_paca()->system_time = 0; 260 } 261 if (in_irq() || idle_task(smp_processor_id()) != tsk) 262 account_system_time(tsk, 0, delta, deltascaled); 263 else 264 account_idle_time(delta); 265 per_cpu(cputime_last_delta, smp_processor_id()) = delta; 266 per_cpu(cputime_scaled_last_delta, smp_processor_id()) = deltascaled; 267 local_irq_restore(flags); 268 } 269 270 /* 271 * Transfer the user and system times accumulated in the paca 272 * by the exception entry and exit code to the generic process 273 * user and system time records. 274 * Must be called with interrupts disabled. 275 */ 276 void account_process_tick(struct task_struct *tsk, int user_tick) 277 { 278 cputime_t utime, utimescaled; 279 280 utime = get_paca()->user_time; 281 get_paca()->user_time = 0; 282 utimescaled = cputime_to_scaled(utime); 283 account_user_time(tsk, utime, 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, 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 if (idle_task(smp_processor_id()) != current) 345 account_steal_time(stolen); 346 else 347 account_idle_time(stolen); 348 } 349 pme->tb = tb; 350 pme->purr = purr; 351 } 352 353 #ifdef CONFIG_PPC_SPLPAR 354 /* 355 * Must be called before the cpu is added to the online map when 356 * a cpu is being brought up at runtime. 357 */ 358 static void snapshot_purr(void) 359 { 360 struct cpu_purr_data *pme; 361 unsigned long flags; 362 363 if (!cpu_has_feature(CPU_FTR_PURR)) 364 return; 365 local_irq_save(flags); 366 pme = &__get_cpu_var(cpu_purr_data); 367 pme->tb = mftb(); 368 pme->purr = mfspr(SPRN_PURR); 369 pme->initialized = 1; 370 local_irq_restore(flags); 371 } 372 373 #endif /* CONFIG_PPC_SPLPAR */ 374 375 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */ 376 #define calc_cputime_factors() 377 #define calculate_steal_time() do { } while (0) 378 #endif 379 380 #if !(defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR)) 381 #define snapshot_purr() do { } while (0) 382 #endif 383 384 /* 385 * Called when a cpu comes up after the system has finished booting, 386 * i.e. as a result of a hotplug cpu action. 387 */ 388 void snapshot_timebase(void) 389 { 390 __get_cpu_var(last_jiffy) = get_tb_or_rtc(); 391 snapshot_purr(); 392 } 393 394 void __delay(unsigned long loops) 395 { 396 unsigned long start; 397 int diff; 398 399 if (__USE_RTC()) { 400 start = get_rtcl(); 401 do { 402 /* the RTCL register wraps at 1000000000 */ 403 diff = get_rtcl() - start; 404 if (diff < 0) 405 diff += 1000000000; 406 } while (diff < loops); 407 } else { 408 start = get_tbl(); 409 while (get_tbl() - start < loops) 410 HMT_low(); 411 HMT_medium(); 412 } 413 } 414 EXPORT_SYMBOL(__delay); 415 416 void udelay(unsigned long usecs) 417 { 418 __delay(tb_ticks_per_usec * usecs); 419 } 420 EXPORT_SYMBOL(udelay); 421 422 static inline void update_gtod(u64 new_tb_stamp, u64 new_stamp_xsec, 423 u64 new_tb_to_xs) 424 { 425 /* 426 * tb_update_count is used to allow the userspace gettimeofday code 427 * to assure itself that it sees a consistent view of the tb_to_xs and 428 * stamp_xsec variables. It reads the tb_update_count, then reads 429 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 430 * the two values of tb_update_count match and are even then the 431 * tb_to_xs and stamp_xsec values are consistent. If not, then it 432 * loops back and reads them again until this criteria is met. 433 * We expect the caller to have done the first increment of 434 * vdso_data->tb_update_count already. 435 */ 436 vdso_data->tb_orig_stamp = new_tb_stamp; 437 vdso_data->stamp_xsec = new_stamp_xsec; 438 vdso_data->tb_to_xs = new_tb_to_xs; 439 vdso_data->wtom_clock_sec = wall_to_monotonic.tv_sec; 440 vdso_data->wtom_clock_nsec = wall_to_monotonic.tv_nsec; 441 vdso_data->stamp_xtime = xtime; 442 smp_wmb(); 443 ++(vdso_data->tb_update_count); 444 } 445 446 #ifdef CONFIG_SMP 447 unsigned long profile_pc(struct pt_regs *regs) 448 { 449 unsigned long pc = instruction_pointer(regs); 450 451 if (in_lock_functions(pc)) 452 return regs->link; 453 454 return pc; 455 } 456 EXPORT_SYMBOL(profile_pc); 457 #endif 458 459 #ifdef CONFIG_PPC_ISERIES 460 461 /* 462 * This function recalibrates the timebase based on the 49-bit time-of-day 463 * value in the Titan chip. The Titan is much more accurate than the value 464 * returned by the service processor for the timebase frequency. 465 */ 466 467 static int __init iSeries_tb_recal(void) 468 { 469 struct div_result divres; 470 unsigned long titan, tb; 471 472 /* Make sure we only run on iSeries */ 473 if (!firmware_has_feature(FW_FEATURE_ISERIES)) 474 return -ENODEV; 475 476 tb = get_tb(); 477 titan = HvCallXm_loadTod(); 478 if ( iSeries_recal_titan ) { 479 unsigned long tb_ticks = tb - iSeries_recal_tb; 480 unsigned long titan_usec = (titan - iSeries_recal_titan) >> 12; 481 unsigned long new_tb_ticks_per_sec = (tb_ticks * USEC_PER_SEC)/titan_usec; 482 unsigned long new_tb_ticks_per_jiffy = 483 DIV_ROUND_CLOSEST(new_tb_ticks_per_sec, HZ); 484 long tick_diff = new_tb_ticks_per_jiffy - tb_ticks_per_jiffy; 485 char sign = '+'; 486 /* make sure tb_ticks_per_sec and tb_ticks_per_jiffy are consistent */ 487 new_tb_ticks_per_sec = new_tb_ticks_per_jiffy * HZ; 488 489 if ( tick_diff < 0 ) { 490 tick_diff = -tick_diff; 491 sign = '-'; 492 } 493 if ( tick_diff ) { 494 if ( tick_diff < tb_ticks_per_jiffy/25 ) { 495 printk( "Titan recalibrate: new tb_ticks_per_jiffy = %lu (%c%ld)\n", 496 new_tb_ticks_per_jiffy, sign, tick_diff ); 497 tb_ticks_per_jiffy = new_tb_ticks_per_jiffy; 498 tb_ticks_per_sec = new_tb_ticks_per_sec; 499 calc_cputime_factors(); 500 div128_by_32( XSEC_PER_SEC, 0, tb_ticks_per_sec, &divres ); 501 tb_to_xs = divres.result_low; 502 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 503 vdso_data->tb_to_xs = tb_to_xs; 504 } 505 else { 506 printk( "Titan recalibrate: FAILED (difference > 4 percent)\n" 507 " new tb_ticks_per_jiffy = %lu\n" 508 " old tb_ticks_per_jiffy = %lu\n", 509 new_tb_ticks_per_jiffy, tb_ticks_per_jiffy ); 510 } 511 } 512 } 513 iSeries_recal_titan = titan; 514 iSeries_recal_tb = tb; 515 516 /* Called here as now we know accurate values for the timebase */ 517 clocksource_init(); 518 return 0; 519 } 520 late_initcall(iSeries_tb_recal); 521 522 /* Called from platform early init */ 523 void __init iSeries_time_init_early(void) 524 { 525 iSeries_recal_tb = get_tb(); 526 iSeries_recal_titan = HvCallXm_loadTod(); 527 } 528 #endif /* CONFIG_PPC_ISERIES */ 529 530 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_PPC32) 531 DEFINE_PER_CPU(u8, perf_event_pending); 532 533 void set_perf_event_pending(void) 534 { 535 get_cpu_var(perf_event_pending) = 1; 536 set_dec(1); 537 put_cpu_var(perf_event_pending); 538 } 539 540 #define test_perf_event_pending() __get_cpu_var(perf_event_pending) 541 #define clear_perf_event_pending() __get_cpu_var(perf_event_pending) = 0 542 543 #else /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */ 544 545 #define test_perf_event_pending() 0 546 #define clear_perf_event_pending() 547 548 #endif /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */ 549 550 /* 551 * For iSeries shared processors, we have to let the hypervisor 552 * set the hardware decrementer. We set a virtual decrementer 553 * in the lppaca and call the hypervisor if the virtual 554 * decrementer is less than the current value in the hardware 555 * decrementer. (almost always the new decrementer value will 556 * be greater than the current hardware decementer so the hypervisor 557 * call will not be needed) 558 */ 559 560 /* 561 * timer_interrupt - gets called when the decrementer overflows, 562 * with interrupts disabled. 563 */ 564 void timer_interrupt(struct pt_regs * regs) 565 { 566 struct pt_regs *old_regs; 567 struct decrementer_clock *decrementer = &__get_cpu_var(decrementers); 568 struct clock_event_device *evt = &decrementer->event; 569 u64 now; 570 571 /* Ensure a positive value is written to the decrementer, or else 572 * some CPUs will continuue to take decrementer exceptions */ 573 set_dec(DECREMENTER_MAX); 574 575 #ifdef CONFIG_PPC32 576 if (test_perf_event_pending()) { 577 clear_perf_event_pending(); 578 perf_event_do_pending(); 579 } 580 if (atomic_read(&ppc_n_lost_interrupts) != 0) 581 do_IRQ(regs); 582 #endif 583 584 now = get_tb_or_rtc(); 585 if (now < decrementer->next_tb) { 586 /* not time for this event yet */ 587 now = decrementer->next_tb - now; 588 if (now <= DECREMENTER_MAX) 589 set_dec((int)now); 590 return; 591 } 592 old_regs = set_irq_regs(regs); 593 irq_enter(); 594 595 calculate_steal_time(); 596 597 #ifdef CONFIG_PPC_ISERIES 598 if (firmware_has_feature(FW_FEATURE_ISERIES)) 599 get_lppaca()->int_dword.fields.decr_int = 0; 600 #endif 601 602 if (evt->event_handler) 603 evt->event_handler(evt); 604 605 #ifdef CONFIG_PPC_ISERIES 606 if (firmware_has_feature(FW_FEATURE_ISERIES) && hvlpevent_is_pending()) 607 process_hvlpevents(); 608 #endif 609 610 #ifdef CONFIG_PPC64 611 /* collect purr register values often, for accurate calculations */ 612 if (firmware_has_feature(FW_FEATURE_SPLPAR)) { 613 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array); 614 cu->current_tb = mfspr(SPRN_PURR); 615 } 616 #endif 617 618 irq_exit(); 619 set_irq_regs(old_regs); 620 } 621 622 void wakeup_decrementer(void) 623 { 624 unsigned long ticks; 625 626 /* 627 * The timebase gets saved on sleep and restored on wakeup, 628 * so all we need to do is to reset the decrementer. 629 */ 630 ticks = tb_ticks_since(__get_cpu_var(last_jiffy)); 631 if (ticks < tb_ticks_per_jiffy) 632 ticks = tb_ticks_per_jiffy - ticks; 633 else 634 ticks = 1; 635 set_dec(ticks); 636 } 637 638 #ifdef CONFIG_SUSPEND 639 void generic_suspend_disable_irqs(void) 640 { 641 preempt_disable(); 642 643 /* Disable the decrementer, so that it doesn't interfere 644 * with suspending. 645 */ 646 647 set_dec(0x7fffffff); 648 local_irq_disable(); 649 set_dec(0x7fffffff); 650 } 651 652 void generic_suspend_enable_irqs(void) 653 { 654 wakeup_decrementer(); 655 656 local_irq_enable(); 657 preempt_enable(); 658 } 659 660 /* Overrides the weak version in kernel/power/main.c */ 661 void arch_suspend_disable_irqs(void) 662 { 663 if (ppc_md.suspend_disable_irqs) 664 ppc_md.suspend_disable_irqs(); 665 generic_suspend_disable_irqs(); 666 } 667 668 /* Overrides the weak version in kernel/power/main.c */ 669 void arch_suspend_enable_irqs(void) 670 { 671 generic_suspend_enable_irqs(); 672 if (ppc_md.suspend_enable_irqs) 673 ppc_md.suspend_enable_irqs(); 674 } 675 #endif 676 677 #ifdef CONFIG_SMP 678 void __init smp_space_timers(unsigned int max_cpus) 679 { 680 int i; 681 u64 previous_tb = per_cpu(last_jiffy, boot_cpuid); 682 683 /* make sure tb > per_cpu(last_jiffy, cpu) for all cpus always */ 684 previous_tb -= tb_ticks_per_jiffy; 685 686 for_each_possible_cpu(i) { 687 if (i == boot_cpuid) 688 continue; 689 per_cpu(last_jiffy, i) = previous_tb; 690 } 691 } 692 #endif 693 694 /* 695 * Scheduler clock - returns current time in nanosec units. 696 * 697 * Note: mulhdu(a, b) (multiply high double unsigned) returns 698 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 699 * are 64-bit unsigned numbers. 700 */ 701 unsigned long long sched_clock(void) 702 { 703 if (__USE_RTC()) 704 return get_rtc(); 705 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 706 } 707 708 static int __init get_freq(char *name, int cells, unsigned long *val) 709 { 710 struct device_node *cpu; 711 const unsigned int *fp; 712 int found = 0; 713 714 /* The cpu node should have timebase and clock frequency properties */ 715 cpu = of_find_node_by_type(NULL, "cpu"); 716 717 if (cpu) { 718 fp = of_get_property(cpu, name, NULL); 719 if (fp) { 720 found = 1; 721 *val = of_read_ulong(fp, cells); 722 } 723 724 of_node_put(cpu); 725 } 726 727 return found; 728 } 729 730 /* should become __cpuinit when secondary_cpu_time_init also is */ 731 void start_cpu_decrementer(void) 732 { 733 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 734 /* Clear any pending timer interrupts */ 735 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 736 737 /* Enable decrementer interrupt */ 738 mtspr(SPRN_TCR, TCR_DIE); 739 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */ 740 } 741 742 void __init generic_calibrate_decr(void) 743 { 744 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 745 746 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 747 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 748 749 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 750 "(not found)\n"); 751 } 752 753 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 754 755 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 756 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 757 758 printk(KERN_ERR "WARNING: Estimating processor frequency " 759 "(not found)\n"); 760 } 761 } 762 763 int update_persistent_clock(struct timespec now) 764 { 765 struct rtc_time tm; 766 767 if (!ppc_md.set_rtc_time) 768 return 0; 769 770 to_tm(now.tv_sec + 1 + timezone_offset, &tm); 771 tm.tm_year -= 1900; 772 tm.tm_mon -= 1; 773 774 return ppc_md.set_rtc_time(&tm); 775 } 776 777 void read_persistent_clock(struct timespec *ts) 778 { 779 struct rtc_time tm; 780 static int first = 1; 781 782 ts->tv_nsec = 0; 783 /* XXX this is a litle fragile but will work okay in the short term */ 784 if (first) { 785 first = 0; 786 if (ppc_md.time_init) 787 timezone_offset = ppc_md.time_init(); 788 789 /* get_boot_time() isn't guaranteed to be safe to call late */ 790 if (ppc_md.get_boot_time) { 791 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset; 792 return; 793 } 794 } 795 if (!ppc_md.get_rtc_time) { 796 ts->tv_sec = 0; 797 return; 798 } 799 ppc_md.get_rtc_time(&tm); 800 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday, 801 tm.tm_hour, tm.tm_min, tm.tm_sec); 802 } 803 804 /* clocksource code */ 805 static cycle_t rtc_read(struct clocksource *cs) 806 { 807 return (cycle_t)get_rtc(); 808 } 809 810 static cycle_t timebase_read(struct clocksource *cs) 811 { 812 return (cycle_t)get_tb(); 813 } 814 815 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock) 816 { 817 u64 t2x, stamp_xsec; 818 819 if (clock != &clocksource_timebase) 820 return; 821 822 /* Make userspace gettimeofday spin until we're done. */ 823 ++vdso_data->tb_update_count; 824 smp_mb(); 825 826 /* XXX this assumes clock->shift == 22 */ 827 /* 4611686018 ~= 2^(20+64-22) / 1e9 */ 828 t2x = (u64) clock->mult * 4611686018ULL; 829 stamp_xsec = (u64) xtime.tv_nsec * XSEC_PER_SEC; 830 do_div(stamp_xsec, 1000000000); 831 stamp_xsec += (u64) xtime.tv_sec * XSEC_PER_SEC; 832 update_gtod(clock->cycle_last, stamp_xsec, t2x); 833 } 834 835 void update_vsyscall_tz(void) 836 { 837 /* Make userspace gettimeofday spin until we're done. */ 838 ++vdso_data->tb_update_count; 839 smp_mb(); 840 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 841 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 842 smp_mb(); 843 ++vdso_data->tb_update_count; 844 } 845 846 static void __init clocksource_init(void) 847 { 848 struct clocksource *clock; 849 850 if (__USE_RTC()) 851 clock = &clocksource_rtc; 852 else 853 clock = &clocksource_timebase; 854 855 clock->mult = clocksource_hz2mult(tb_ticks_per_sec, clock->shift); 856 857 if (clocksource_register(clock)) { 858 printk(KERN_ERR "clocksource: %s is already registered\n", 859 clock->name); 860 return; 861 } 862 863 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 864 clock->name, clock->mult, clock->shift); 865 } 866 867 static int decrementer_set_next_event(unsigned long evt, 868 struct clock_event_device *dev) 869 { 870 __get_cpu_var(decrementers).next_tb = get_tb_or_rtc() + evt; 871 set_dec(evt); 872 return 0; 873 } 874 875 static void decrementer_set_mode(enum clock_event_mode mode, 876 struct clock_event_device *dev) 877 { 878 if (mode != CLOCK_EVT_MODE_ONESHOT) 879 decrementer_set_next_event(DECREMENTER_MAX, dev); 880 } 881 882 static void __init setup_clockevent_multiplier(unsigned long hz) 883 { 884 u64 mult, shift = 32; 885 886 while (1) { 887 mult = div_sc(hz, NSEC_PER_SEC, shift); 888 if (mult && (mult >> 32UL) == 0UL) 889 break; 890 891 shift--; 892 } 893 894 decrementer_clockevent.shift = shift; 895 decrementer_clockevent.mult = mult; 896 } 897 898 static void register_decrementer_clockevent(int cpu) 899 { 900 struct clock_event_device *dec = &per_cpu(decrementers, cpu).event; 901 902 *dec = decrementer_clockevent; 903 dec->cpumask = cpumask_of(cpu); 904 905 printk(KERN_DEBUG "clockevent: %s mult[%lx] shift[%d] cpu[%d]\n", 906 dec->name, dec->mult, dec->shift, cpu); 907 908 clockevents_register_device(dec); 909 } 910 911 static void __init init_decrementer_clockevent(void) 912 { 913 int cpu = smp_processor_id(); 914 915 setup_clockevent_multiplier(ppc_tb_freq); 916 decrementer_clockevent.max_delta_ns = 917 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent); 918 decrementer_clockevent.min_delta_ns = 919 clockevent_delta2ns(2, &decrementer_clockevent); 920 921 register_decrementer_clockevent(cpu); 922 } 923 924 void secondary_cpu_time_init(void) 925 { 926 /* Start the decrementer on CPUs that have manual control 927 * such as BookE 928 */ 929 start_cpu_decrementer(); 930 931 /* FIME: Should make unrelatred change to move snapshot_timebase 932 * call here ! */ 933 register_decrementer_clockevent(smp_processor_id()); 934 } 935 936 /* This function is only called on the boot processor */ 937 void __init time_init(void) 938 { 939 unsigned long flags; 940 struct div_result res; 941 u64 scale, x; 942 unsigned shift; 943 944 if (__USE_RTC()) { 945 /* 601 processor: dec counts down by 128 every 128ns */ 946 ppc_tb_freq = 1000000000; 947 tb_last_jiffy = get_rtcl(); 948 } else { 949 /* Normal PowerPC with timebase register */ 950 ppc_md.calibrate_decr(); 951 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 952 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 953 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 954 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 955 tb_last_jiffy = get_tb(); 956 } 957 958 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 959 tb_ticks_per_sec = ppc_tb_freq; 960 tb_ticks_per_usec = ppc_tb_freq / 1000000; 961 tb_to_us = mulhwu_scale_factor(ppc_tb_freq, 1000000); 962 calc_cputime_factors(); 963 964 /* 965 * Calculate the length of each tick in ns. It will not be 966 * exactly 1e9/HZ unless ppc_tb_freq is divisible by HZ. 967 * We compute 1e9 * tb_ticks_per_jiffy / ppc_tb_freq, 968 * rounded up. 969 */ 970 x = (u64) NSEC_PER_SEC * tb_ticks_per_jiffy + ppc_tb_freq - 1; 971 do_div(x, ppc_tb_freq); 972 tick_nsec = x; 973 last_tick_len = x << TICKLEN_SCALE; 974 975 /* 976 * Compute ticklen_to_xs, which is a factor which gets multiplied 977 * by (last_tick_len << TICKLEN_SHIFT) to get a tb_to_xs value. 978 * It is computed as: 979 * ticklen_to_xs = 2^N / (tb_ticks_per_jiffy * 1e9) 980 * where N = 64 + 20 - TICKLEN_SCALE - TICKLEN_SHIFT 981 * which turns out to be N = 51 - SHIFT_HZ. 982 * This gives the result as a 0.64 fixed-point fraction. 983 * That value is reduced by an offset amounting to 1 xsec per 984 * 2^31 timebase ticks to avoid problems with time going backwards 985 * by 1 xsec when we do timer_recalc_offset due to losing the 986 * fractional xsec. That offset is equal to ppc_tb_freq/2^51 987 * since there are 2^20 xsec in a second. 988 */ 989 div128_by_32((1ULL << 51) - ppc_tb_freq, 0, 990 tb_ticks_per_jiffy << SHIFT_HZ, &res); 991 div128_by_32(res.result_high, res.result_low, NSEC_PER_SEC, &res); 992 ticklen_to_xs = res.result_low; 993 994 /* Compute tb_to_xs from tick_nsec */ 995 tb_to_xs = mulhdu(last_tick_len << TICKLEN_SHIFT, ticklen_to_xs); 996 997 /* 998 * Compute scale factor for sched_clock. 999 * The calibrate_decr() function has set tb_ticks_per_sec, 1000 * which is the timebase frequency. 1001 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 1002 * the 128-bit result as a 64.64 fixed-point number. 1003 * We then shift that number right until it is less than 1.0, 1004 * giving us the scale factor and shift count to use in 1005 * sched_clock(). 1006 */ 1007 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 1008 scale = res.result_low; 1009 for (shift = 0; res.result_high != 0; ++shift) { 1010 scale = (scale >> 1) | (res.result_high << 63); 1011 res.result_high >>= 1; 1012 } 1013 tb_to_ns_scale = scale; 1014 tb_to_ns_shift = shift; 1015 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 1016 boot_tb = get_tb_or_rtc(); 1017 1018 write_seqlock_irqsave(&xtime_lock, flags); 1019 1020 /* If platform provided a timezone (pmac), we correct the time */ 1021 if (timezone_offset) { 1022 sys_tz.tz_minuteswest = -timezone_offset / 60; 1023 sys_tz.tz_dsttime = 0; 1024 } 1025 1026 vdso_data->tb_orig_stamp = tb_last_jiffy; 1027 vdso_data->tb_update_count = 0; 1028 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 1029 vdso_data->stamp_xsec = (u64) xtime.tv_sec * XSEC_PER_SEC; 1030 vdso_data->tb_to_xs = tb_to_xs; 1031 1032 write_sequnlock_irqrestore(&xtime_lock, flags); 1033 1034 /* Start the decrementer on CPUs that have manual control 1035 * such as BookE 1036 */ 1037 start_cpu_decrementer(); 1038 1039 /* Register the clocksource, if we're not running on iSeries */ 1040 if (!firmware_has_feature(FW_FEATURE_ISERIES)) 1041 clocksource_init(); 1042 1043 init_decrementer_clockevent(); 1044 } 1045 1046 1047 #define FEBRUARY 2 1048 #define STARTOFTIME 1970 1049 #define SECDAY 86400L 1050 #define SECYR (SECDAY * 365) 1051 #define leapyear(year) ((year) % 4 == 0 && \ 1052 ((year) % 100 != 0 || (year) % 400 == 0)) 1053 #define days_in_year(a) (leapyear(a) ? 366 : 365) 1054 #define days_in_month(a) (month_days[(a) - 1]) 1055 1056 static int month_days[12] = { 1057 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 1058 }; 1059 1060 /* 1061 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK) 1062 */ 1063 void GregorianDay(struct rtc_time * tm) 1064 { 1065 int leapsToDate; 1066 int lastYear; 1067 int day; 1068 int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 }; 1069 1070 lastYear = tm->tm_year - 1; 1071 1072 /* 1073 * Number of leap corrections to apply up to end of last year 1074 */ 1075 leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400; 1076 1077 /* 1078 * This year is a leap year if it is divisible by 4 except when it is 1079 * divisible by 100 unless it is divisible by 400 1080 * 1081 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was 1082 */ 1083 day = tm->tm_mon > 2 && leapyear(tm->tm_year); 1084 1085 day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] + 1086 tm->tm_mday; 1087 1088 tm->tm_wday = day % 7; 1089 } 1090 1091 void to_tm(int tim, struct rtc_time * tm) 1092 { 1093 register int i; 1094 register long hms, day; 1095 1096 day = tim / SECDAY; 1097 hms = tim % SECDAY; 1098 1099 /* Hours, minutes, seconds are easy */ 1100 tm->tm_hour = hms / 3600; 1101 tm->tm_min = (hms % 3600) / 60; 1102 tm->tm_sec = (hms % 3600) % 60; 1103 1104 /* Number of years in days */ 1105 for (i = STARTOFTIME; day >= days_in_year(i); i++) 1106 day -= days_in_year(i); 1107 tm->tm_year = i; 1108 1109 /* Number of months in days left */ 1110 if (leapyear(tm->tm_year)) 1111 days_in_month(FEBRUARY) = 29; 1112 for (i = 1; day >= days_in_month(i); i++) 1113 day -= days_in_month(i); 1114 days_in_month(FEBRUARY) = 28; 1115 tm->tm_mon = i; 1116 1117 /* Days are what is left over (+1) from all that. */ 1118 tm->tm_mday = day + 1; 1119 1120 /* 1121 * Determine the day of week 1122 */ 1123 GregorianDay(tm); 1124 } 1125 1126 /* Auxiliary function to compute scaling factors */ 1127 /* Actually the choice of a timebase running at 1/4 the of the bus 1128 * frequency giving resolution of a few tens of nanoseconds is quite nice. 1129 * It makes this computation very precise (27-28 bits typically) which 1130 * is optimistic considering the stability of most processor clock 1131 * oscillators and the precision with which the timebase frequency 1132 * is measured but does not harm. 1133 */ 1134 unsigned mulhwu_scale_factor(unsigned inscale, unsigned outscale) 1135 { 1136 unsigned mlt=0, tmp, err; 1137 /* No concern for performance, it's done once: use a stupid 1138 * but safe and compact method to find the multiplier. 1139 */ 1140 1141 for (tmp = 1U<<31; tmp != 0; tmp >>= 1) { 1142 if (mulhwu(inscale, mlt|tmp) < outscale) 1143 mlt |= tmp; 1144 } 1145 1146 /* We might still be off by 1 for the best approximation. 1147 * A side effect of this is that if outscale is too large 1148 * the returned value will be zero. 1149 * Many corner cases have been checked and seem to work, 1150 * some might have been forgotten in the test however. 1151 */ 1152 1153 err = inscale * (mlt+1); 1154 if (err <= inscale/2) 1155 mlt++; 1156 return mlt; 1157 } 1158 1159 /* 1160 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1161 * result. 1162 */ 1163 void div128_by_32(u64 dividend_high, u64 dividend_low, 1164 unsigned divisor, struct div_result *dr) 1165 { 1166 unsigned long a, b, c, d; 1167 unsigned long w, x, y, z; 1168 u64 ra, rb, rc; 1169 1170 a = dividend_high >> 32; 1171 b = dividend_high & 0xffffffff; 1172 c = dividend_low >> 32; 1173 d = dividend_low & 0xffffffff; 1174 1175 w = a / divisor; 1176 ra = ((u64)(a - (w * divisor)) << 32) + b; 1177 1178 rb = ((u64) do_div(ra, divisor) << 32) + c; 1179 x = ra; 1180 1181 rc = ((u64) do_div(rb, divisor) << 32) + d; 1182 y = rb; 1183 1184 do_div(rc, divisor); 1185 z = rc; 1186 1187 dr->result_high = ((u64)w << 32) + x; 1188 dr->result_low = ((u64)y << 32) + z; 1189 1190 } 1191 1192 /* We don't need to calibrate delay, we use the CPU timebase for that */ 1193 void calibrate_delay(void) 1194 { 1195 /* Some generic code (such as spinlock debug) use loops_per_jiffy 1196 * as the number of __delay(1) in a jiffy, so make it so 1197 */ 1198 loops_per_jiffy = tb_ticks_per_jiffy; 1199 } 1200 1201 static int __init rtc_init(void) 1202 { 1203 struct platform_device *pdev; 1204 1205 if (!ppc_md.get_rtc_time) 1206 return -ENODEV; 1207 1208 pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0); 1209 if (IS_ERR(pdev)) 1210 return PTR_ERR(pdev); 1211 1212 return 0; 1213 } 1214 1215 module_init(rtc_init); 1216