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. 21 * - for astronomical applications: add a new function to get 22 * non ambiguous timestamps even around leap seconds. This needs 23 * a new timestamp format and a good name. 24 * 25 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 26 * "A Kernel Model for Precision Timekeeping" by Dave Mills 27 * 28 * This program is free software; you can redistribute it and/or 29 * modify it under the terms of the GNU General Public License 30 * as published by the Free Software Foundation; either version 31 * 2 of the License, or (at your option) any later version. 32 */ 33 34 #include <linux/errno.h> 35 #include <linux/export.h> 36 #include <linux/sched.h> 37 #include <linux/kernel.h> 38 #include <linux/param.h> 39 #include <linux/string.h> 40 #include <linux/mm.h> 41 #include <linux/interrupt.h> 42 #include <linux/timex.h> 43 #include <linux/kernel_stat.h> 44 #include <linux/time.h> 45 #include <linux/init.h> 46 #include <linux/profile.h> 47 #include <linux/cpu.h> 48 #include <linux/security.h> 49 #include <linux/percpu.h> 50 #include <linux/rtc.h> 51 #include <linux/jiffies.h> 52 #include <linux/posix-timers.h> 53 #include <linux/irq.h> 54 #include <linux/delay.h> 55 #include <linux/irq_work.h> 56 #include <asm/trace.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 73 /* powerpc clocksource/clockevent code */ 74 75 #include <linux/clockchips.h> 76 #include <linux/timekeeper_internal.h> 77 78 static cycle_t rtc_read(struct clocksource *); 79 static struct clocksource clocksource_rtc = { 80 .name = "rtc", 81 .rating = 400, 82 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 83 .mask = CLOCKSOURCE_MASK(64), 84 .read = rtc_read, 85 }; 86 87 static cycle_t timebase_read(struct clocksource *); 88 static struct clocksource clocksource_timebase = { 89 .name = "timebase", 90 .rating = 400, 91 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 92 .mask = CLOCKSOURCE_MASK(64), 93 .read = timebase_read, 94 }; 95 96 #define DECREMENTER_MAX 0x7fffffff 97 98 static int decrementer_set_next_event(unsigned long evt, 99 struct clock_event_device *dev); 100 static void decrementer_set_mode(enum clock_event_mode mode, 101 struct clock_event_device *dev); 102 103 struct clock_event_device decrementer_clockevent = { 104 .name = "decrementer", 105 .rating = 200, 106 .irq = 0, 107 .set_next_event = decrementer_set_next_event, 108 .set_mode = decrementer_set_mode, 109 .features = CLOCK_EVT_FEAT_ONESHOT, 110 }; 111 EXPORT_SYMBOL(decrementer_clockevent); 112 113 DEFINE_PER_CPU(u64, decrementers_next_tb); 114 static DEFINE_PER_CPU(struct clock_event_device, decrementers); 115 116 #define XSEC_PER_SEC (1024*1024) 117 118 #ifdef CONFIG_PPC64 119 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC) 120 #else 121 /* compute ((xsec << 12) * max) >> 32 */ 122 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max) 123 #endif 124 125 unsigned long tb_ticks_per_jiffy; 126 unsigned long tb_ticks_per_usec = 100; /* sane default */ 127 EXPORT_SYMBOL(tb_ticks_per_usec); 128 unsigned long tb_ticks_per_sec; 129 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */ 130 131 DEFINE_SPINLOCK(rtc_lock); 132 EXPORT_SYMBOL_GPL(rtc_lock); 133 134 static u64 tb_to_ns_scale __read_mostly; 135 static unsigned tb_to_ns_shift __read_mostly; 136 static u64 boot_tb __read_mostly; 137 138 extern struct timezone sys_tz; 139 static long timezone_offset; 140 141 unsigned long ppc_proc_freq; 142 EXPORT_SYMBOL_GPL(ppc_proc_freq); 143 unsigned long ppc_tb_freq; 144 EXPORT_SYMBOL_GPL(ppc_tb_freq); 145 146 #ifdef CONFIG_VIRT_CPU_ACCOUNTING 147 /* 148 * Factors for converting from cputime_t (timebase ticks) to 149 * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds). 150 * These are all stored as 0.64 fixed-point binary fractions. 151 */ 152 u64 __cputime_jiffies_factor; 153 EXPORT_SYMBOL(__cputime_jiffies_factor); 154 u64 __cputime_usec_factor; 155 EXPORT_SYMBOL(__cputime_usec_factor); 156 u64 __cputime_sec_factor; 157 EXPORT_SYMBOL(__cputime_sec_factor); 158 u64 __cputime_clockt_factor; 159 EXPORT_SYMBOL(__cputime_clockt_factor); 160 DEFINE_PER_CPU(unsigned long, cputime_last_delta); 161 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta); 162 163 cputime_t cputime_one_jiffy; 164 165 void (*dtl_consumer)(struct dtl_entry *, u64); 166 167 static void calc_cputime_factors(void) 168 { 169 struct div_result res; 170 171 div128_by_32(HZ, 0, tb_ticks_per_sec, &res); 172 __cputime_jiffies_factor = res.result_low; 173 div128_by_32(1000000, 0, tb_ticks_per_sec, &res); 174 __cputime_usec_factor = res.result_low; 175 div128_by_32(1, 0, tb_ticks_per_sec, &res); 176 __cputime_sec_factor = res.result_low; 177 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res); 178 __cputime_clockt_factor = res.result_low; 179 } 180 181 /* 182 * Read the SPURR on systems that have it, otherwise the PURR, 183 * or if that doesn't exist return the timebase value passed in. 184 */ 185 static u64 read_spurr(u64 tb) 186 { 187 if (cpu_has_feature(CPU_FTR_SPURR)) 188 return mfspr(SPRN_SPURR); 189 if (cpu_has_feature(CPU_FTR_PURR)) 190 return mfspr(SPRN_PURR); 191 return tb; 192 } 193 194 #ifdef CONFIG_PPC_SPLPAR 195 196 /* 197 * Scan the dispatch trace log and count up the stolen time. 198 * Should be called with interrupts disabled. 199 */ 200 static u64 scan_dispatch_log(u64 stop_tb) 201 { 202 u64 i = local_paca->dtl_ridx; 203 struct dtl_entry *dtl = local_paca->dtl_curr; 204 struct dtl_entry *dtl_end = local_paca->dispatch_log_end; 205 struct lppaca *vpa = local_paca->lppaca_ptr; 206 u64 tb_delta; 207 u64 stolen = 0; 208 u64 dtb; 209 210 if (!dtl) 211 return 0; 212 213 if (i == vpa->dtl_idx) 214 return 0; 215 while (i < vpa->dtl_idx) { 216 if (dtl_consumer) 217 dtl_consumer(dtl, i); 218 dtb = dtl->timebase; 219 tb_delta = dtl->enqueue_to_dispatch_time + 220 dtl->ready_to_enqueue_time; 221 barrier(); 222 if (i + N_DISPATCH_LOG < vpa->dtl_idx) { 223 /* buffer has overflowed */ 224 i = vpa->dtl_idx - N_DISPATCH_LOG; 225 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG); 226 continue; 227 } 228 if (dtb > stop_tb) 229 break; 230 stolen += tb_delta; 231 ++i; 232 ++dtl; 233 if (dtl == dtl_end) 234 dtl = local_paca->dispatch_log; 235 } 236 local_paca->dtl_ridx = i; 237 local_paca->dtl_curr = dtl; 238 return stolen; 239 } 240 241 /* 242 * Accumulate stolen time by scanning the dispatch trace log. 243 * Called on entry from user mode. 244 */ 245 void accumulate_stolen_time(void) 246 { 247 u64 sst, ust; 248 249 u8 save_soft_enabled = local_paca->soft_enabled; 250 251 /* We are called early in the exception entry, before 252 * soft/hard_enabled are sync'ed to the expected state 253 * for the exception. We are hard disabled but the PACA 254 * needs to reflect that so various debug stuff doesn't 255 * complain 256 */ 257 local_paca->soft_enabled = 0; 258 259 sst = scan_dispatch_log(local_paca->starttime_user); 260 ust = scan_dispatch_log(local_paca->starttime); 261 local_paca->system_time -= sst; 262 local_paca->user_time -= ust; 263 local_paca->stolen_time += ust + sst; 264 265 local_paca->soft_enabled = save_soft_enabled; 266 } 267 268 static inline u64 calculate_stolen_time(u64 stop_tb) 269 { 270 u64 stolen = 0; 271 272 if (get_paca()->dtl_ridx != get_paca()->lppaca_ptr->dtl_idx) { 273 stolen = scan_dispatch_log(stop_tb); 274 get_paca()->system_time -= stolen; 275 } 276 277 stolen += get_paca()->stolen_time; 278 get_paca()->stolen_time = 0; 279 return stolen; 280 } 281 282 #else /* CONFIG_PPC_SPLPAR */ 283 static inline u64 calculate_stolen_time(u64 stop_tb) 284 { 285 return 0; 286 } 287 288 #endif /* CONFIG_PPC_SPLPAR */ 289 290 /* 291 * Account time for a transition between system, hard irq 292 * or soft irq state. 293 */ 294 static u64 vtime_delta(struct task_struct *tsk, 295 u64 *sys_scaled, u64 *stolen) 296 { 297 u64 now, nowscaled, deltascaled; 298 u64 udelta, delta, user_scaled; 299 300 WARN_ON_ONCE(!irqs_disabled()); 301 302 now = mftb(); 303 nowscaled = read_spurr(now); 304 get_paca()->system_time += now - get_paca()->starttime; 305 get_paca()->starttime = now; 306 deltascaled = nowscaled - get_paca()->startspurr; 307 get_paca()->startspurr = nowscaled; 308 309 *stolen = calculate_stolen_time(now); 310 311 delta = get_paca()->system_time; 312 get_paca()->system_time = 0; 313 udelta = get_paca()->user_time - get_paca()->utime_sspurr; 314 get_paca()->utime_sspurr = get_paca()->user_time; 315 316 /* 317 * Because we don't read the SPURR on every kernel entry/exit, 318 * deltascaled includes both user and system SPURR ticks. 319 * Apportion these ticks to system SPURR ticks and user 320 * SPURR ticks in the same ratio as the system time (delta) 321 * and user time (udelta) values obtained from the timebase 322 * over the same interval. The system ticks get accounted here; 323 * the user ticks get saved up in paca->user_time_scaled to be 324 * used by account_process_tick. 325 */ 326 *sys_scaled = delta; 327 user_scaled = udelta; 328 if (deltascaled != delta + udelta) { 329 if (udelta) { 330 *sys_scaled = deltascaled * delta / (delta + udelta); 331 user_scaled = deltascaled - *sys_scaled; 332 } else { 333 *sys_scaled = deltascaled; 334 } 335 } 336 get_paca()->user_time_scaled += user_scaled; 337 338 return delta; 339 } 340 341 void vtime_account_system(struct task_struct *tsk) 342 { 343 u64 delta, sys_scaled, stolen; 344 345 delta = vtime_delta(tsk, &sys_scaled, &stolen); 346 account_system_time(tsk, 0, delta, sys_scaled); 347 if (stolen) 348 account_steal_time(stolen); 349 } 350 351 void vtime_account_idle(struct task_struct *tsk) 352 { 353 u64 delta, sys_scaled, stolen; 354 355 delta = vtime_delta(tsk, &sys_scaled, &stolen); 356 account_idle_time(delta + stolen); 357 } 358 359 /* 360 * Transfer the user time accumulated in the paca 361 * by the exception entry and exit code to the generic 362 * process user time records. 363 * Must be called with interrupts disabled. 364 * Assumes that vtime_account_system/idle() has been called 365 * recently (i.e. since the last entry from usermode) so that 366 * get_paca()->user_time_scaled is up to date. 367 */ 368 void vtime_account_user(struct task_struct *tsk) 369 { 370 cputime_t utime, utimescaled; 371 372 utime = get_paca()->user_time; 373 utimescaled = get_paca()->user_time_scaled; 374 get_paca()->user_time = 0; 375 get_paca()->user_time_scaled = 0; 376 get_paca()->utime_sspurr = 0; 377 account_user_time(tsk, utime, utimescaled); 378 } 379 380 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING */ 381 #define calc_cputime_factors() 382 #endif 383 384 void __delay(unsigned long loops) 385 { 386 unsigned long start; 387 int diff; 388 389 if (__USE_RTC()) { 390 start = get_rtcl(); 391 do { 392 /* the RTCL register wraps at 1000000000 */ 393 diff = get_rtcl() - start; 394 if (diff < 0) 395 diff += 1000000000; 396 } while (diff < loops); 397 } else { 398 start = get_tbl(); 399 while (get_tbl() - start < loops) 400 HMT_low(); 401 HMT_medium(); 402 } 403 } 404 EXPORT_SYMBOL(__delay); 405 406 void udelay(unsigned long usecs) 407 { 408 __delay(tb_ticks_per_usec * usecs); 409 } 410 EXPORT_SYMBOL(udelay); 411 412 #ifdef CONFIG_SMP 413 unsigned long profile_pc(struct pt_regs *regs) 414 { 415 unsigned long pc = instruction_pointer(regs); 416 417 if (in_lock_functions(pc)) 418 return regs->link; 419 420 return pc; 421 } 422 EXPORT_SYMBOL(profile_pc); 423 #endif 424 425 #ifdef CONFIG_IRQ_WORK 426 427 /* 428 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable... 429 */ 430 #ifdef CONFIG_PPC64 431 static inline unsigned long test_irq_work_pending(void) 432 { 433 unsigned long x; 434 435 asm volatile("lbz %0,%1(13)" 436 : "=r" (x) 437 : "i" (offsetof(struct paca_struct, irq_work_pending))); 438 return x; 439 } 440 441 static inline void set_irq_work_pending_flag(void) 442 { 443 asm volatile("stb %0,%1(13)" : : 444 "r" (1), 445 "i" (offsetof(struct paca_struct, irq_work_pending))); 446 } 447 448 static inline void clear_irq_work_pending(void) 449 { 450 asm volatile("stb %0,%1(13)" : : 451 "r" (0), 452 "i" (offsetof(struct paca_struct, irq_work_pending))); 453 } 454 455 #else /* 32-bit */ 456 457 DEFINE_PER_CPU(u8, irq_work_pending); 458 459 #define set_irq_work_pending_flag() __get_cpu_var(irq_work_pending) = 1 460 #define test_irq_work_pending() __get_cpu_var(irq_work_pending) 461 #define clear_irq_work_pending() __get_cpu_var(irq_work_pending) = 0 462 463 #endif /* 32 vs 64 bit */ 464 465 void arch_irq_work_raise(void) 466 { 467 preempt_disable(); 468 set_irq_work_pending_flag(); 469 set_dec(1); 470 preempt_enable(); 471 } 472 473 #else /* CONFIG_IRQ_WORK */ 474 475 #define test_irq_work_pending() 0 476 #define clear_irq_work_pending() 477 478 #endif /* CONFIG_IRQ_WORK */ 479 480 /* 481 * timer_interrupt - gets called when the decrementer overflows, 482 * with interrupts disabled. 483 */ 484 void timer_interrupt(struct pt_regs * regs) 485 { 486 struct pt_regs *old_regs; 487 u64 *next_tb = &__get_cpu_var(decrementers_next_tb); 488 struct clock_event_device *evt = &__get_cpu_var(decrementers); 489 u64 now; 490 491 /* Ensure a positive value is written to the decrementer, or else 492 * some CPUs will continue to take decrementer exceptions. 493 */ 494 set_dec(DECREMENTER_MAX); 495 496 /* Some implementations of hotplug will get timer interrupts while 497 * offline, just ignore these and we also need to set 498 * decrementers_next_tb as MAX to make sure __check_irq_replay 499 * don't replay timer interrupt when return, otherwise we'll trap 500 * here infinitely :( 501 */ 502 if (!cpu_online(smp_processor_id())) { 503 *next_tb = ~(u64)0; 504 return; 505 } 506 507 /* Conditionally hard-enable interrupts now that the DEC has been 508 * bumped to its maximum value 509 */ 510 may_hard_irq_enable(); 511 512 __get_cpu_var(irq_stat).timer_irqs++; 513 514 #if defined(CONFIG_PPC32) && defined(CONFIG_PMAC) 515 if (atomic_read(&ppc_n_lost_interrupts) != 0) 516 do_IRQ(regs); 517 #endif 518 519 old_regs = set_irq_regs(regs); 520 irq_enter(); 521 522 trace_timer_interrupt_entry(regs); 523 524 if (test_irq_work_pending()) { 525 clear_irq_work_pending(); 526 irq_work_run(); 527 } 528 529 now = get_tb_or_rtc(); 530 if (now >= *next_tb) { 531 *next_tb = ~(u64)0; 532 if (evt->event_handler) 533 evt->event_handler(evt); 534 } else { 535 now = *next_tb - now; 536 if (now <= DECREMENTER_MAX) 537 set_dec((int)now); 538 } 539 540 #ifdef CONFIG_PPC64 541 /* collect purr register values often, for accurate calculations */ 542 if (firmware_has_feature(FW_FEATURE_SPLPAR)) { 543 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array); 544 cu->current_tb = mfspr(SPRN_PURR); 545 } 546 #endif 547 548 trace_timer_interrupt_exit(regs); 549 550 irq_exit(); 551 set_irq_regs(old_regs); 552 } 553 554 /* 555 * Hypervisor decrementer interrupts shouldn't occur but are sometimes 556 * left pending on exit from a KVM guest. We don't need to do anything 557 * to clear them, as they are edge-triggered. 558 */ 559 void hdec_interrupt(struct pt_regs *regs) 560 { 561 } 562 563 #ifdef CONFIG_SUSPEND 564 static void generic_suspend_disable_irqs(void) 565 { 566 /* Disable the decrementer, so that it doesn't interfere 567 * with suspending. 568 */ 569 570 set_dec(DECREMENTER_MAX); 571 local_irq_disable(); 572 set_dec(DECREMENTER_MAX); 573 } 574 575 static void generic_suspend_enable_irqs(void) 576 { 577 local_irq_enable(); 578 } 579 580 /* Overrides the weak version in kernel/power/main.c */ 581 void arch_suspend_disable_irqs(void) 582 { 583 if (ppc_md.suspend_disable_irqs) 584 ppc_md.suspend_disable_irqs(); 585 generic_suspend_disable_irqs(); 586 } 587 588 /* Overrides the weak version in kernel/power/main.c */ 589 void arch_suspend_enable_irqs(void) 590 { 591 generic_suspend_enable_irqs(); 592 if (ppc_md.suspend_enable_irqs) 593 ppc_md.suspend_enable_irqs(); 594 } 595 #endif 596 597 /* 598 * Scheduler clock - returns current time in nanosec units. 599 * 600 * Note: mulhdu(a, b) (multiply high double unsigned) returns 601 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 602 * are 64-bit unsigned numbers. 603 */ 604 unsigned long long sched_clock(void) 605 { 606 if (__USE_RTC()) 607 return get_rtc(); 608 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 609 } 610 611 static int __init get_freq(char *name, int cells, unsigned long *val) 612 { 613 struct device_node *cpu; 614 const unsigned int *fp; 615 int found = 0; 616 617 /* The cpu node should have timebase and clock frequency properties */ 618 cpu = of_find_node_by_type(NULL, "cpu"); 619 620 if (cpu) { 621 fp = of_get_property(cpu, name, NULL); 622 if (fp) { 623 found = 1; 624 *val = of_read_ulong(fp, cells); 625 } 626 627 of_node_put(cpu); 628 } 629 630 return found; 631 } 632 633 /* should become __cpuinit when secondary_cpu_time_init also is */ 634 void start_cpu_decrementer(void) 635 { 636 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 637 /* Clear any pending timer interrupts */ 638 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 639 640 /* Enable decrementer interrupt */ 641 mtspr(SPRN_TCR, TCR_DIE); 642 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */ 643 } 644 645 void __init generic_calibrate_decr(void) 646 { 647 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 648 649 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 650 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 651 652 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 653 "(not found)\n"); 654 } 655 656 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 657 658 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 659 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 660 661 printk(KERN_ERR "WARNING: Estimating processor frequency " 662 "(not found)\n"); 663 } 664 } 665 666 int update_persistent_clock(struct timespec now) 667 { 668 struct rtc_time tm; 669 670 if (!ppc_md.set_rtc_time) 671 return 0; 672 673 to_tm(now.tv_sec + 1 + timezone_offset, &tm); 674 tm.tm_year -= 1900; 675 tm.tm_mon -= 1; 676 677 return ppc_md.set_rtc_time(&tm); 678 } 679 680 static void __read_persistent_clock(struct timespec *ts) 681 { 682 struct rtc_time tm; 683 static int first = 1; 684 685 ts->tv_nsec = 0; 686 /* XXX this is a litle fragile but will work okay in the short term */ 687 if (first) { 688 first = 0; 689 if (ppc_md.time_init) 690 timezone_offset = ppc_md.time_init(); 691 692 /* get_boot_time() isn't guaranteed to be safe to call late */ 693 if (ppc_md.get_boot_time) { 694 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset; 695 return; 696 } 697 } 698 if (!ppc_md.get_rtc_time) { 699 ts->tv_sec = 0; 700 return; 701 } 702 ppc_md.get_rtc_time(&tm); 703 704 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday, 705 tm.tm_hour, tm.tm_min, tm.tm_sec); 706 } 707 708 void read_persistent_clock(struct timespec *ts) 709 { 710 __read_persistent_clock(ts); 711 712 /* Sanitize it in case real time clock is set below EPOCH */ 713 if (ts->tv_sec < 0) { 714 ts->tv_sec = 0; 715 ts->tv_nsec = 0; 716 } 717 718 } 719 720 /* clocksource code */ 721 static cycle_t rtc_read(struct clocksource *cs) 722 { 723 return (cycle_t)get_rtc(); 724 } 725 726 static cycle_t timebase_read(struct clocksource *cs) 727 { 728 return (cycle_t)get_tb(); 729 } 730 731 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm, 732 struct clocksource *clock, u32 mult) 733 { 734 u64 new_tb_to_xs, new_stamp_xsec; 735 u32 frac_sec; 736 737 if (clock != &clocksource_timebase) 738 return; 739 740 /* Make userspace gettimeofday spin until we're done. */ 741 ++vdso_data->tb_update_count; 742 smp_mb(); 743 744 /* 19342813113834067 ~= 2^(20+64) / 1e9 */ 745 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift); 746 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC; 747 do_div(new_stamp_xsec, 1000000000); 748 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC; 749 750 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC); 751 /* this is tv_nsec / 1e9 as a 0.32 fraction */ 752 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32; 753 754 /* 755 * tb_update_count is used to allow the userspace gettimeofday code 756 * to assure itself that it sees a consistent view of the tb_to_xs and 757 * stamp_xsec variables. It reads the tb_update_count, then reads 758 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 759 * the two values of tb_update_count match and are even then the 760 * tb_to_xs and stamp_xsec values are consistent. If not, then it 761 * loops back and reads them again until this criteria is met. 762 * We expect the caller to have done the first increment of 763 * vdso_data->tb_update_count already. 764 */ 765 vdso_data->tb_orig_stamp = clock->cycle_last; 766 vdso_data->stamp_xsec = new_stamp_xsec; 767 vdso_data->tb_to_xs = new_tb_to_xs; 768 vdso_data->wtom_clock_sec = wtm->tv_sec; 769 vdso_data->wtom_clock_nsec = wtm->tv_nsec; 770 vdso_data->stamp_xtime = *wall_time; 771 vdso_data->stamp_sec_fraction = frac_sec; 772 smp_wmb(); 773 ++(vdso_data->tb_update_count); 774 } 775 776 void update_vsyscall_tz(void) 777 { 778 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 779 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 780 } 781 782 static void __init clocksource_init(void) 783 { 784 struct clocksource *clock; 785 786 if (__USE_RTC()) 787 clock = &clocksource_rtc; 788 else 789 clock = &clocksource_timebase; 790 791 if (clocksource_register_hz(clock, tb_ticks_per_sec)) { 792 printk(KERN_ERR "clocksource: %s is already registered\n", 793 clock->name); 794 return; 795 } 796 797 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 798 clock->name, clock->mult, clock->shift); 799 } 800 801 static int decrementer_set_next_event(unsigned long evt, 802 struct clock_event_device *dev) 803 { 804 __get_cpu_var(decrementers_next_tb) = get_tb_or_rtc() + evt; 805 set_dec(evt); 806 return 0; 807 } 808 809 static void decrementer_set_mode(enum clock_event_mode mode, 810 struct clock_event_device *dev) 811 { 812 if (mode != CLOCK_EVT_MODE_ONESHOT) 813 decrementer_set_next_event(DECREMENTER_MAX, dev); 814 } 815 816 static void register_decrementer_clockevent(int cpu) 817 { 818 struct clock_event_device *dec = &per_cpu(decrementers, cpu); 819 820 *dec = decrementer_clockevent; 821 dec->cpumask = cpumask_of(cpu); 822 823 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n", 824 dec->name, dec->mult, dec->shift, cpu); 825 826 clockevents_register_device(dec); 827 } 828 829 static void __init init_decrementer_clockevent(void) 830 { 831 int cpu = smp_processor_id(); 832 833 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4); 834 835 decrementer_clockevent.max_delta_ns = 836 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent); 837 decrementer_clockevent.min_delta_ns = 838 clockevent_delta2ns(2, &decrementer_clockevent); 839 840 register_decrementer_clockevent(cpu); 841 } 842 843 void secondary_cpu_time_init(void) 844 { 845 /* Start the decrementer on CPUs that have manual control 846 * such as BookE 847 */ 848 start_cpu_decrementer(); 849 850 /* FIME: Should make unrelatred change to move snapshot_timebase 851 * call here ! */ 852 register_decrementer_clockevent(smp_processor_id()); 853 } 854 855 /* This function is only called on the boot processor */ 856 void __init time_init(void) 857 { 858 struct div_result res; 859 u64 scale; 860 unsigned shift; 861 862 if (__USE_RTC()) { 863 /* 601 processor: dec counts down by 128 every 128ns */ 864 ppc_tb_freq = 1000000000; 865 } else { 866 /* Normal PowerPC with timebase register */ 867 ppc_md.calibrate_decr(); 868 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 869 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 870 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 871 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 872 } 873 874 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 875 tb_ticks_per_sec = ppc_tb_freq; 876 tb_ticks_per_usec = ppc_tb_freq / 1000000; 877 calc_cputime_factors(); 878 setup_cputime_one_jiffy(); 879 880 /* 881 * Compute scale factor for sched_clock. 882 * The calibrate_decr() function has set tb_ticks_per_sec, 883 * which is the timebase frequency. 884 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 885 * the 128-bit result as a 64.64 fixed-point number. 886 * We then shift that number right until it is less than 1.0, 887 * giving us the scale factor and shift count to use in 888 * sched_clock(). 889 */ 890 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 891 scale = res.result_low; 892 for (shift = 0; res.result_high != 0; ++shift) { 893 scale = (scale >> 1) | (res.result_high << 63); 894 res.result_high >>= 1; 895 } 896 tb_to_ns_scale = scale; 897 tb_to_ns_shift = shift; 898 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 899 boot_tb = get_tb_or_rtc(); 900 901 /* If platform provided a timezone (pmac), we correct the time */ 902 if (timezone_offset) { 903 sys_tz.tz_minuteswest = -timezone_offset / 60; 904 sys_tz.tz_dsttime = 0; 905 } 906 907 vdso_data->tb_update_count = 0; 908 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 909 910 /* Start the decrementer on CPUs that have manual control 911 * such as BookE 912 */ 913 start_cpu_decrementer(); 914 915 /* Register the clocksource */ 916 clocksource_init(); 917 918 init_decrementer_clockevent(); 919 } 920 921 922 #define FEBRUARY 2 923 #define STARTOFTIME 1970 924 #define SECDAY 86400L 925 #define SECYR (SECDAY * 365) 926 #define leapyear(year) ((year) % 4 == 0 && \ 927 ((year) % 100 != 0 || (year) % 400 == 0)) 928 #define days_in_year(a) (leapyear(a) ? 366 : 365) 929 #define days_in_month(a) (month_days[(a) - 1]) 930 931 static int month_days[12] = { 932 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 933 }; 934 935 /* 936 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK) 937 */ 938 void GregorianDay(struct rtc_time * tm) 939 { 940 int leapsToDate; 941 int lastYear; 942 int day; 943 int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 }; 944 945 lastYear = tm->tm_year - 1; 946 947 /* 948 * Number of leap corrections to apply up to end of last year 949 */ 950 leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400; 951 952 /* 953 * This year is a leap year if it is divisible by 4 except when it is 954 * divisible by 100 unless it is divisible by 400 955 * 956 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was 957 */ 958 day = tm->tm_mon > 2 && leapyear(tm->tm_year); 959 960 day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] + 961 tm->tm_mday; 962 963 tm->tm_wday = day % 7; 964 } 965 966 void to_tm(int tim, struct rtc_time * tm) 967 { 968 register int i; 969 register long hms, day; 970 971 day = tim / SECDAY; 972 hms = tim % SECDAY; 973 974 /* Hours, minutes, seconds are easy */ 975 tm->tm_hour = hms / 3600; 976 tm->tm_min = (hms % 3600) / 60; 977 tm->tm_sec = (hms % 3600) % 60; 978 979 /* Number of years in days */ 980 for (i = STARTOFTIME; day >= days_in_year(i); i++) 981 day -= days_in_year(i); 982 tm->tm_year = i; 983 984 /* Number of months in days left */ 985 if (leapyear(tm->tm_year)) 986 days_in_month(FEBRUARY) = 29; 987 for (i = 1; day >= days_in_month(i); i++) 988 day -= days_in_month(i); 989 days_in_month(FEBRUARY) = 28; 990 tm->tm_mon = i; 991 992 /* Days are what is left over (+1) from all that. */ 993 tm->tm_mday = day + 1; 994 995 /* 996 * Determine the day of week 997 */ 998 GregorianDay(tm); 999 } 1000 1001 /* 1002 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1003 * result. 1004 */ 1005 void div128_by_32(u64 dividend_high, u64 dividend_low, 1006 unsigned divisor, struct div_result *dr) 1007 { 1008 unsigned long a, b, c, d; 1009 unsigned long w, x, y, z; 1010 u64 ra, rb, rc; 1011 1012 a = dividend_high >> 32; 1013 b = dividend_high & 0xffffffff; 1014 c = dividend_low >> 32; 1015 d = dividend_low & 0xffffffff; 1016 1017 w = a / divisor; 1018 ra = ((u64)(a - (w * divisor)) << 32) + b; 1019 1020 rb = ((u64) do_div(ra, divisor) << 32) + c; 1021 x = ra; 1022 1023 rc = ((u64) do_div(rb, divisor) << 32) + d; 1024 y = rb; 1025 1026 do_div(rc, divisor); 1027 z = rc; 1028 1029 dr->result_high = ((u64)w << 32) + x; 1030 dr->result_low = ((u64)y << 32) + z; 1031 1032 } 1033 1034 /* We don't need to calibrate delay, we use the CPU timebase for that */ 1035 void calibrate_delay(void) 1036 { 1037 /* Some generic code (such as spinlock debug) use loops_per_jiffy 1038 * as the number of __delay(1) in a jiffy, so make it so 1039 */ 1040 loops_per_jiffy = tb_ticks_per_jiffy; 1041 } 1042 1043 static int __init rtc_init(void) 1044 { 1045 struct platform_device *pdev; 1046 1047 if (!ppc_md.get_rtc_time) 1048 return -ENODEV; 1049 1050 pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0); 1051 if (IS_ERR(pdev)) 1052 return PTR_ERR(pdev); 1053 1054 return 0; 1055 } 1056 1057 module_init(rtc_init); 1058