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/clockchips.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/irq_work.h> 57 #include <linux/clk-provider.h> 58 #include <asm/trace.h> 59 60 #include <asm/io.h> 61 #include <asm/processor.h> 62 #include <asm/nvram.h> 63 #include <asm/cache.h> 64 #include <asm/machdep.h> 65 #include <asm/uaccess.h> 66 #include <asm/time.h> 67 #include <asm/prom.h> 68 #include <asm/irq.h> 69 #include <asm/div64.h> 70 #include <asm/smp.h> 71 #include <asm/vdso_datapage.h> 72 #include <asm/firmware.h> 73 #include <asm/cputime.h> 74 75 /* powerpc clocksource/clockevent code */ 76 77 #include <linux/clockchips.h> 78 #include <linux/timekeeper_internal.h> 79 80 static cycle_t rtc_read(struct clocksource *); 81 static struct clocksource clocksource_rtc = { 82 .name = "rtc", 83 .rating = 400, 84 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 85 .mask = CLOCKSOURCE_MASK(64), 86 .read = rtc_read, 87 }; 88 89 static cycle_t timebase_read(struct clocksource *); 90 static struct clocksource clocksource_timebase = { 91 .name = "timebase", 92 .rating = 400, 93 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 94 .mask = CLOCKSOURCE_MASK(64), 95 .read = timebase_read, 96 }; 97 98 #define DECREMENTER_MAX 0x7fffffff 99 100 static int decrementer_set_next_event(unsigned long evt, 101 struct clock_event_device *dev); 102 static int decrementer_shutdown(struct clock_event_device *evt); 103 104 struct clock_event_device decrementer_clockevent = { 105 .name = "decrementer", 106 .rating = 200, 107 .irq = 0, 108 .set_next_event = decrementer_set_next_event, 109 .set_state_shutdown = decrementer_shutdown, 110 .tick_resume = decrementer_shutdown, 111 .features = CLOCK_EVT_FEAT_ONESHOT | 112 CLOCK_EVT_FEAT_C3STOP, 113 }; 114 EXPORT_SYMBOL(decrementer_clockevent); 115 116 DEFINE_PER_CPU(u64, decrementers_next_tb); 117 static DEFINE_PER_CPU(struct clock_event_device, decrementers); 118 119 #define XSEC_PER_SEC (1024*1024) 120 121 #ifdef CONFIG_PPC64 122 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC) 123 #else 124 /* compute ((xsec << 12) * max) >> 32 */ 125 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max) 126 #endif 127 128 unsigned long tb_ticks_per_jiffy; 129 unsigned long tb_ticks_per_usec = 100; /* sane default */ 130 EXPORT_SYMBOL(tb_ticks_per_usec); 131 unsigned long tb_ticks_per_sec; 132 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */ 133 134 DEFINE_SPINLOCK(rtc_lock); 135 EXPORT_SYMBOL_GPL(rtc_lock); 136 137 static u64 tb_to_ns_scale __read_mostly; 138 static unsigned tb_to_ns_shift __read_mostly; 139 static u64 boot_tb __read_mostly; 140 141 extern struct timezone sys_tz; 142 static long timezone_offset; 143 144 unsigned long ppc_proc_freq; 145 EXPORT_SYMBOL_GPL(ppc_proc_freq); 146 unsigned long ppc_tb_freq; 147 EXPORT_SYMBOL_GPL(ppc_tb_freq); 148 149 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE 150 /* 151 * Factors for converting from cputime_t (timebase ticks) to 152 * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds). 153 * These are all stored as 0.64 fixed-point binary fractions. 154 */ 155 u64 __cputime_jiffies_factor; 156 EXPORT_SYMBOL(__cputime_jiffies_factor); 157 u64 __cputime_usec_factor; 158 EXPORT_SYMBOL(__cputime_usec_factor); 159 u64 __cputime_sec_factor; 160 EXPORT_SYMBOL(__cputime_sec_factor); 161 u64 __cputime_clockt_factor; 162 EXPORT_SYMBOL(__cputime_clockt_factor); 163 DEFINE_PER_CPU(unsigned long, cputime_last_delta); 164 DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta); 165 166 cputime_t cputime_one_jiffy; 167 168 void (*dtl_consumer)(struct dtl_entry *, u64); 169 170 static void calc_cputime_factors(void) 171 { 172 struct div_result res; 173 174 div128_by_32(HZ, 0, tb_ticks_per_sec, &res); 175 __cputime_jiffies_factor = res.result_low; 176 div128_by_32(1000000, 0, tb_ticks_per_sec, &res); 177 __cputime_usec_factor = res.result_low; 178 div128_by_32(1, 0, tb_ticks_per_sec, &res); 179 __cputime_sec_factor = res.result_low; 180 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res); 181 __cputime_clockt_factor = res.result_low; 182 } 183 184 /* 185 * Read the SPURR on systems that have it, otherwise the PURR, 186 * or if that doesn't exist return the timebase value passed in. 187 */ 188 static u64 read_spurr(u64 tb) 189 { 190 if (cpu_has_feature(CPU_FTR_SPURR)) 191 return mfspr(SPRN_SPURR); 192 if (cpu_has_feature(CPU_FTR_PURR)) 193 return mfspr(SPRN_PURR); 194 return tb; 195 } 196 197 #ifdef CONFIG_PPC_SPLPAR 198 199 /* 200 * Scan the dispatch trace log and count up the stolen time. 201 * Should be called with interrupts disabled. 202 */ 203 static u64 scan_dispatch_log(u64 stop_tb) 204 { 205 u64 i = local_paca->dtl_ridx; 206 struct dtl_entry *dtl = local_paca->dtl_curr; 207 struct dtl_entry *dtl_end = local_paca->dispatch_log_end; 208 struct lppaca *vpa = local_paca->lppaca_ptr; 209 u64 tb_delta; 210 u64 stolen = 0; 211 u64 dtb; 212 213 if (!dtl) 214 return 0; 215 216 if (i == be64_to_cpu(vpa->dtl_idx)) 217 return 0; 218 while (i < be64_to_cpu(vpa->dtl_idx)) { 219 dtb = be64_to_cpu(dtl->timebase); 220 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) + 221 be32_to_cpu(dtl->ready_to_enqueue_time); 222 barrier(); 223 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) { 224 /* buffer has overflowed */ 225 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG; 226 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG); 227 continue; 228 } 229 if (dtb > stop_tb) 230 break; 231 if (dtl_consumer) 232 dtl_consumer(dtl, i); 233 stolen += tb_delta; 234 ++i; 235 ++dtl; 236 if (dtl == dtl_end) 237 dtl = local_paca->dispatch_log; 238 } 239 local_paca->dtl_ridx = i; 240 local_paca->dtl_curr = dtl; 241 return stolen; 242 } 243 244 /* 245 * Accumulate stolen time by scanning the dispatch trace log. 246 * Called on entry from user mode. 247 */ 248 void accumulate_stolen_time(void) 249 { 250 u64 sst, ust; 251 252 u8 save_soft_enabled = local_paca->soft_enabled; 253 254 /* We are called early in the exception entry, before 255 * soft/hard_enabled are sync'ed to the expected state 256 * for the exception. We are hard disabled but the PACA 257 * needs to reflect that so various debug stuff doesn't 258 * complain 259 */ 260 local_paca->soft_enabled = 0; 261 262 sst = scan_dispatch_log(local_paca->starttime_user); 263 ust = scan_dispatch_log(local_paca->starttime); 264 local_paca->system_time -= sst; 265 local_paca->user_time -= ust; 266 local_paca->stolen_time += ust + sst; 267 268 local_paca->soft_enabled = save_soft_enabled; 269 } 270 271 static inline u64 calculate_stolen_time(u64 stop_tb) 272 { 273 u64 stolen = 0; 274 275 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) { 276 stolen = scan_dispatch_log(stop_tb); 277 get_paca()->system_time -= stolen; 278 } 279 280 stolen += get_paca()->stolen_time; 281 get_paca()->stolen_time = 0; 282 return stolen; 283 } 284 285 #else /* CONFIG_PPC_SPLPAR */ 286 static inline u64 calculate_stolen_time(u64 stop_tb) 287 { 288 return 0; 289 } 290 291 #endif /* CONFIG_PPC_SPLPAR */ 292 293 /* 294 * Account time for a transition between system, hard irq 295 * or soft irq state. 296 */ 297 static u64 vtime_delta(struct task_struct *tsk, 298 u64 *sys_scaled, u64 *stolen) 299 { 300 u64 now, nowscaled, deltascaled; 301 u64 udelta, delta, user_scaled; 302 303 WARN_ON_ONCE(!irqs_disabled()); 304 305 now = mftb(); 306 nowscaled = read_spurr(now); 307 get_paca()->system_time += now - get_paca()->starttime; 308 get_paca()->starttime = now; 309 deltascaled = nowscaled - get_paca()->startspurr; 310 get_paca()->startspurr = nowscaled; 311 312 *stolen = calculate_stolen_time(now); 313 314 delta = get_paca()->system_time; 315 get_paca()->system_time = 0; 316 udelta = get_paca()->user_time - get_paca()->utime_sspurr; 317 get_paca()->utime_sspurr = get_paca()->user_time; 318 319 /* 320 * Because we don't read the SPURR on every kernel entry/exit, 321 * deltascaled includes both user and system SPURR ticks. 322 * Apportion these ticks to system SPURR ticks and user 323 * SPURR ticks in the same ratio as the system time (delta) 324 * and user time (udelta) values obtained from the timebase 325 * over the same interval. The system ticks get accounted here; 326 * the user ticks get saved up in paca->user_time_scaled to be 327 * used by account_process_tick. 328 */ 329 *sys_scaled = delta; 330 user_scaled = udelta; 331 if (deltascaled != delta + udelta) { 332 if (udelta) { 333 *sys_scaled = deltascaled * delta / (delta + udelta); 334 user_scaled = deltascaled - *sys_scaled; 335 } else { 336 *sys_scaled = deltascaled; 337 } 338 } 339 get_paca()->user_time_scaled += user_scaled; 340 341 return delta; 342 } 343 344 void vtime_account_system(struct task_struct *tsk) 345 { 346 u64 delta, sys_scaled, stolen; 347 348 delta = vtime_delta(tsk, &sys_scaled, &stolen); 349 account_system_time(tsk, 0, delta, sys_scaled); 350 if (stolen) 351 account_steal_time(stolen); 352 } 353 EXPORT_SYMBOL_GPL(vtime_account_system); 354 355 void vtime_account_idle(struct task_struct *tsk) 356 { 357 u64 delta, sys_scaled, stolen; 358 359 delta = vtime_delta(tsk, &sys_scaled, &stolen); 360 account_idle_time(delta + stolen); 361 } 362 363 /* 364 * Transfer the user time accumulated in the paca 365 * by the exception entry and exit code to the generic 366 * process user time records. 367 * Must be called with interrupts disabled. 368 * Assumes that vtime_account_system/idle() has been called 369 * recently (i.e. since the last entry from usermode) so that 370 * get_paca()->user_time_scaled is up to date. 371 */ 372 void vtime_account_user(struct task_struct *tsk) 373 { 374 cputime_t utime, utimescaled; 375 376 utime = get_paca()->user_time; 377 utimescaled = get_paca()->user_time_scaled; 378 get_paca()->user_time = 0; 379 get_paca()->user_time_scaled = 0; 380 get_paca()->utime_sspurr = 0; 381 account_user_time(tsk, utime, utimescaled); 382 } 383 384 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ 385 #define calc_cputime_factors() 386 #endif 387 388 void __delay(unsigned long loops) 389 { 390 unsigned long start; 391 int diff; 392 393 if (__USE_RTC()) { 394 start = get_rtcl(); 395 do { 396 /* the RTCL register wraps at 1000000000 */ 397 diff = get_rtcl() - start; 398 if (diff < 0) 399 diff += 1000000000; 400 } while (diff < loops); 401 } else { 402 start = get_tbl(); 403 while (get_tbl() - start < loops) 404 HMT_low(); 405 HMT_medium(); 406 } 407 } 408 EXPORT_SYMBOL(__delay); 409 410 void udelay(unsigned long usecs) 411 { 412 __delay(tb_ticks_per_usec * usecs); 413 } 414 EXPORT_SYMBOL(udelay); 415 416 #ifdef CONFIG_SMP 417 unsigned long profile_pc(struct pt_regs *regs) 418 { 419 unsigned long pc = instruction_pointer(regs); 420 421 if (in_lock_functions(pc)) 422 return regs->link; 423 424 return pc; 425 } 426 EXPORT_SYMBOL(profile_pc); 427 #endif 428 429 #ifdef CONFIG_IRQ_WORK 430 431 /* 432 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable... 433 */ 434 #ifdef CONFIG_PPC64 435 static inline unsigned long test_irq_work_pending(void) 436 { 437 unsigned long x; 438 439 asm volatile("lbz %0,%1(13)" 440 : "=r" (x) 441 : "i" (offsetof(struct paca_struct, irq_work_pending))); 442 return x; 443 } 444 445 static inline void set_irq_work_pending_flag(void) 446 { 447 asm volatile("stb %0,%1(13)" : : 448 "r" (1), 449 "i" (offsetof(struct paca_struct, irq_work_pending))); 450 } 451 452 static inline void clear_irq_work_pending(void) 453 { 454 asm volatile("stb %0,%1(13)" : : 455 "r" (0), 456 "i" (offsetof(struct paca_struct, irq_work_pending))); 457 } 458 459 #else /* 32-bit */ 460 461 DEFINE_PER_CPU(u8, irq_work_pending); 462 463 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1) 464 #define test_irq_work_pending() __this_cpu_read(irq_work_pending) 465 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0) 466 467 #endif /* 32 vs 64 bit */ 468 469 void arch_irq_work_raise(void) 470 { 471 preempt_disable(); 472 set_irq_work_pending_flag(); 473 set_dec(1); 474 preempt_enable(); 475 } 476 477 #else /* CONFIG_IRQ_WORK */ 478 479 #define test_irq_work_pending() 0 480 #define clear_irq_work_pending() 481 482 #endif /* CONFIG_IRQ_WORK */ 483 484 static void __timer_interrupt(void) 485 { 486 struct pt_regs *regs = get_irq_regs(); 487 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 488 struct clock_event_device *evt = this_cpu_ptr(&decrementers); 489 u64 now; 490 491 trace_timer_interrupt_entry(regs); 492 493 if (test_irq_work_pending()) { 494 clear_irq_work_pending(); 495 irq_work_run(); 496 } 497 498 now = get_tb_or_rtc(); 499 if (now >= *next_tb) { 500 *next_tb = ~(u64)0; 501 if (evt->event_handler) 502 evt->event_handler(evt); 503 __this_cpu_inc(irq_stat.timer_irqs_event); 504 } else { 505 now = *next_tb - now; 506 if (now <= DECREMENTER_MAX) 507 set_dec((int)now); 508 /* We may have raced with new irq work */ 509 if (test_irq_work_pending()) 510 set_dec(1); 511 __this_cpu_inc(irq_stat.timer_irqs_others); 512 } 513 514 #ifdef CONFIG_PPC64 515 /* collect purr register values often, for accurate calculations */ 516 if (firmware_has_feature(FW_FEATURE_SPLPAR)) { 517 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array); 518 cu->current_tb = mfspr(SPRN_PURR); 519 } 520 #endif 521 522 trace_timer_interrupt_exit(regs); 523 } 524 525 /* 526 * timer_interrupt - gets called when the decrementer overflows, 527 * with interrupts disabled. 528 */ 529 void timer_interrupt(struct pt_regs * regs) 530 { 531 struct pt_regs *old_regs; 532 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 533 534 /* Ensure a positive value is written to the decrementer, or else 535 * some CPUs will continue to take decrementer exceptions. 536 */ 537 set_dec(DECREMENTER_MAX); 538 539 /* Some implementations of hotplug will get timer interrupts while 540 * offline, just ignore these and we also need to set 541 * decrementers_next_tb as MAX to make sure __check_irq_replay 542 * don't replay timer interrupt when return, otherwise we'll trap 543 * here infinitely :( 544 */ 545 if (!cpu_online(smp_processor_id())) { 546 *next_tb = ~(u64)0; 547 return; 548 } 549 550 /* Conditionally hard-enable interrupts now that the DEC has been 551 * bumped to its maximum value 552 */ 553 may_hard_irq_enable(); 554 555 556 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC) 557 if (atomic_read(&ppc_n_lost_interrupts) != 0) 558 do_IRQ(regs); 559 #endif 560 561 old_regs = set_irq_regs(regs); 562 irq_enter(); 563 564 __timer_interrupt(); 565 irq_exit(); 566 set_irq_regs(old_regs); 567 } 568 569 /* 570 * Hypervisor decrementer interrupts shouldn't occur but are sometimes 571 * left pending on exit from a KVM guest. We don't need to do anything 572 * to clear them, as they are edge-triggered. 573 */ 574 void hdec_interrupt(struct pt_regs *regs) 575 { 576 } 577 578 #ifdef CONFIG_SUSPEND 579 static void generic_suspend_disable_irqs(void) 580 { 581 /* Disable the decrementer, so that it doesn't interfere 582 * with suspending. 583 */ 584 585 set_dec(DECREMENTER_MAX); 586 local_irq_disable(); 587 set_dec(DECREMENTER_MAX); 588 } 589 590 static void generic_suspend_enable_irqs(void) 591 { 592 local_irq_enable(); 593 } 594 595 /* Overrides the weak version in kernel/power/main.c */ 596 void arch_suspend_disable_irqs(void) 597 { 598 if (ppc_md.suspend_disable_irqs) 599 ppc_md.suspend_disable_irqs(); 600 generic_suspend_disable_irqs(); 601 } 602 603 /* Overrides the weak version in kernel/power/main.c */ 604 void arch_suspend_enable_irqs(void) 605 { 606 generic_suspend_enable_irqs(); 607 if (ppc_md.suspend_enable_irqs) 608 ppc_md.suspend_enable_irqs(); 609 } 610 #endif 611 612 unsigned long long tb_to_ns(unsigned long long ticks) 613 { 614 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift; 615 } 616 EXPORT_SYMBOL_GPL(tb_to_ns); 617 618 /* 619 * Scheduler clock - returns current time in nanosec units. 620 * 621 * Note: mulhdu(a, b) (multiply high double unsigned) returns 622 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 623 * are 64-bit unsigned numbers. 624 */ 625 unsigned long long sched_clock(void) 626 { 627 if (__USE_RTC()) 628 return get_rtc(); 629 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 630 } 631 632 633 #ifdef CONFIG_PPC_PSERIES 634 635 /* 636 * Running clock - attempts to give a view of time passing for a virtualised 637 * kernels. 638 * Uses the VTB register if available otherwise a next best guess. 639 */ 640 unsigned long long running_clock(void) 641 { 642 /* 643 * Don't read the VTB as a host since KVM does not switch in host 644 * timebase into the VTB when it takes a guest off the CPU, reading the 645 * VTB would result in reading 'last switched out' guest VTB. 646 * 647 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it 648 * would be unsafe to rely only on the #ifdef above. 649 */ 650 if (firmware_has_feature(FW_FEATURE_LPAR) && 651 cpu_has_feature(CPU_FTR_ARCH_207S)) 652 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 653 654 /* 655 * This is a next best approximation without a VTB. 656 * On a host which is running bare metal there should never be any stolen 657 * time and on a host which doesn't do any virtualisation TB *should* equal 658 * VTB so it makes no difference anyway. 659 */ 660 return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]); 661 } 662 #endif 663 664 static int __init get_freq(char *name, int cells, unsigned long *val) 665 { 666 struct device_node *cpu; 667 const __be32 *fp; 668 int found = 0; 669 670 /* The cpu node should have timebase and clock frequency properties */ 671 cpu = of_find_node_by_type(NULL, "cpu"); 672 673 if (cpu) { 674 fp = of_get_property(cpu, name, NULL); 675 if (fp) { 676 found = 1; 677 *val = of_read_ulong(fp, cells); 678 } 679 680 of_node_put(cpu); 681 } 682 683 return found; 684 } 685 686 static void start_cpu_decrementer(void) 687 { 688 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 689 /* Clear any pending timer interrupts */ 690 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 691 692 /* Enable decrementer interrupt */ 693 mtspr(SPRN_TCR, TCR_DIE); 694 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */ 695 } 696 697 void __init generic_calibrate_decr(void) 698 { 699 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 700 701 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 702 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 703 704 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 705 "(not found)\n"); 706 } 707 708 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 709 710 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 711 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 712 713 printk(KERN_ERR "WARNING: Estimating processor frequency " 714 "(not found)\n"); 715 } 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 -ENODEV; 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 static void __read_persistent_clock(struct timespec *ts) 733 { 734 struct rtc_time tm; 735 static int first = 1; 736 737 ts->tv_nsec = 0; 738 /* XXX this is a litle fragile but will work okay in the short term */ 739 if (first) { 740 first = 0; 741 if (ppc_md.time_init) 742 timezone_offset = ppc_md.time_init(); 743 744 /* get_boot_time() isn't guaranteed to be safe to call late */ 745 if (ppc_md.get_boot_time) { 746 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset; 747 return; 748 } 749 } 750 if (!ppc_md.get_rtc_time) { 751 ts->tv_sec = 0; 752 return; 753 } 754 ppc_md.get_rtc_time(&tm); 755 756 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday, 757 tm.tm_hour, tm.tm_min, tm.tm_sec); 758 } 759 760 void read_persistent_clock(struct timespec *ts) 761 { 762 __read_persistent_clock(ts); 763 764 /* Sanitize it in case real time clock is set below EPOCH */ 765 if (ts->tv_sec < 0) { 766 ts->tv_sec = 0; 767 ts->tv_nsec = 0; 768 } 769 770 } 771 772 /* clocksource code */ 773 static cycle_t rtc_read(struct clocksource *cs) 774 { 775 return (cycle_t)get_rtc(); 776 } 777 778 static cycle_t timebase_read(struct clocksource *cs) 779 { 780 return (cycle_t)get_tb(); 781 } 782 783 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm, 784 struct clocksource *clock, u32 mult, cycle_t cycle_last) 785 { 786 u64 new_tb_to_xs, new_stamp_xsec; 787 u32 frac_sec; 788 789 if (clock != &clocksource_timebase) 790 return; 791 792 /* Make userspace gettimeofday spin until we're done. */ 793 ++vdso_data->tb_update_count; 794 smp_mb(); 795 796 /* 19342813113834067 ~= 2^(20+64) / 1e9 */ 797 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift); 798 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC; 799 do_div(new_stamp_xsec, 1000000000); 800 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC; 801 802 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC); 803 /* this is tv_nsec / 1e9 as a 0.32 fraction */ 804 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32; 805 806 /* 807 * tb_update_count is used to allow the userspace gettimeofday code 808 * to assure itself that it sees a consistent view of the tb_to_xs and 809 * stamp_xsec variables. It reads the tb_update_count, then reads 810 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 811 * the two values of tb_update_count match and are even then the 812 * tb_to_xs and stamp_xsec values are consistent. If not, then it 813 * loops back and reads them again until this criteria is met. 814 * We expect the caller to have done the first increment of 815 * vdso_data->tb_update_count already. 816 */ 817 vdso_data->tb_orig_stamp = cycle_last; 818 vdso_data->stamp_xsec = new_stamp_xsec; 819 vdso_data->tb_to_xs = new_tb_to_xs; 820 vdso_data->wtom_clock_sec = wtm->tv_sec; 821 vdso_data->wtom_clock_nsec = wtm->tv_nsec; 822 vdso_data->stamp_xtime = *wall_time; 823 vdso_data->stamp_sec_fraction = frac_sec; 824 smp_wmb(); 825 ++(vdso_data->tb_update_count); 826 } 827 828 void update_vsyscall_tz(void) 829 { 830 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 831 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 832 } 833 834 static void __init clocksource_init(void) 835 { 836 struct clocksource *clock; 837 838 if (__USE_RTC()) 839 clock = &clocksource_rtc; 840 else 841 clock = &clocksource_timebase; 842 843 if (clocksource_register_hz(clock, tb_ticks_per_sec)) { 844 printk(KERN_ERR "clocksource: %s is already registered\n", 845 clock->name); 846 return; 847 } 848 849 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 850 clock->name, clock->mult, clock->shift); 851 } 852 853 static int decrementer_set_next_event(unsigned long evt, 854 struct clock_event_device *dev) 855 { 856 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt); 857 set_dec(evt); 858 859 /* We may have raced with new irq work */ 860 if (test_irq_work_pending()) 861 set_dec(1); 862 863 return 0; 864 } 865 866 static int decrementer_shutdown(struct clock_event_device *dev) 867 { 868 decrementer_set_next_event(DECREMENTER_MAX, dev); 869 return 0; 870 } 871 872 /* Interrupt handler for the timer broadcast IPI */ 873 void tick_broadcast_ipi_handler(void) 874 { 875 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 876 877 *next_tb = get_tb_or_rtc(); 878 __timer_interrupt(); 879 } 880 881 static void register_decrementer_clockevent(int cpu) 882 { 883 struct clock_event_device *dec = &per_cpu(decrementers, cpu); 884 885 *dec = decrementer_clockevent; 886 dec->cpumask = cpumask_of(cpu); 887 888 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n", 889 dec->name, dec->mult, dec->shift, cpu); 890 891 clockevents_register_device(dec); 892 } 893 894 static void __init init_decrementer_clockevent(void) 895 { 896 int cpu = smp_processor_id(); 897 898 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4); 899 900 decrementer_clockevent.max_delta_ns = 901 clockevent_delta2ns(DECREMENTER_MAX, &decrementer_clockevent); 902 decrementer_clockevent.min_delta_ns = 903 clockevent_delta2ns(2, &decrementer_clockevent); 904 905 register_decrementer_clockevent(cpu); 906 } 907 908 void secondary_cpu_time_init(void) 909 { 910 /* Start the decrementer on CPUs that have manual control 911 * such as BookE 912 */ 913 start_cpu_decrementer(); 914 915 /* FIME: Should make unrelatred change to move snapshot_timebase 916 * call here ! */ 917 register_decrementer_clockevent(smp_processor_id()); 918 } 919 920 /* This function is only called on the boot processor */ 921 void __init time_init(void) 922 { 923 struct div_result res; 924 u64 scale; 925 unsigned shift; 926 927 if (__USE_RTC()) { 928 /* 601 processor: dec counts down by 128 every 128ns */ 929 ppc_tb_freq = 1000000000; 930 } else { 931 /* Normal PowerPC with timebase register */ 932 ppc_md.calibrate_decr(); 933 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 934 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 935 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 936 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 937 } 938 939 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 940 tb_ticks_per_sec = ppc_tb_freq; 941 tb_ticks_per_usec = ppc_tb_freq / 1000000; 942 calc_cputime_factors(); 943 setup_cputime_one_jiffy(); 944 945 /* 946 * Compute scale factor for sched_clock. 947 * The calibrate_decr() function has set tb_ticks_per_sec, 948 * which is the timebase frequency. 949 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 950 * the 128-bit result as a 64.64 fixed-point number. 951 * We then shift that number right until it is less than 1.0, 952 * giving us the scale factor and shift count to use in 953 * sched_clock(). 954 */ 955 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 956 scale = res.result_low; 957 for (shift = 0; res.result_high != 0; ++shift) { 958 scale = (scale >> 1) | (res.result_high << 63); 959 res.result_high >>= 1; 960 } 961 tb_to_ns_scale = scale; 962 tb_to_ns_shift = shift; 963 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 964 boot_tb = get_tb_or_rtc(); 965 966 /* If platform provided a timezone (pmac), we correct the time */ 967 if (timezone_offset) { 968 sys_tz.tz_minuteswest = -timezone_offset / 60; 969 sys_tz.tz_dsttime = 0; 970 } 971 972 vdso_data->tb_update_count = 0; 973 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 974 975 /* Start the decrementer on CPUs that have manual control 976 * such as BookE 977 */ 978 start_cpu_decrementer(); 979 980 /* Register the clocksource */ 981 clocksource_init(); 982 983 init_decrementer_clockevent(); 984 tick_setup_hrtimer_broadcast(); 985 986 #ifdef CONFIG_COMMON_CLK 987 of_clk_init(NULL); 988 #endif 989 } 990 991 992 #define FEBRUARY 2 993 #define STARTOFTIME 1970 994 #define SECDAY 86400L 995 #define SECYR (SECDAY * 365) 996 #define leapyear(year) ((year) % 4 == 0 && \ 997 ((year) % 100 != 0 || (year) % 400 == 0)) 998 #define days_in_year(a) (leapyear(a) ? 366 : 365) 999 #define days_in_month(a) (month_days[(a) - 1]) 1000 1001 static int month_days[12] = { 1002 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 1003 }; 1004 1005 /* 1006 * This only works for the Gregorian calendar - i.e. after 1752 (in the UK) 1007 */ 1008 void GregorianDay(struct rtc_time * tm) 1009 { 1010 int leapsToDate; 1011 int lastYear; 1012 int day; 1013 int MonthOffset[] = { 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 }; 1014 1015 lastYear = tm->tm_year - 1; 1016 1017 /* 1018 * Number of leap corrections to apply up to end of last year 1019 */ 1020 leapsToDate = lastYear / 4 - lastYear / 100 + lastYear / 400; 1021 1022 /* 1023 * This year is a leap year if it is divisible by 4 except when it is 1024 * divisible by 100 unless it is divisible by 400 1025 * 1026 * e.g. 1904 was a leap year, 1900 was not, 1996 is, and 2000 was 1027 */ 1028 day = tm->tm_mon > 2 && leapyear(tm->tm_year); 1029 1030 day += lastYear*365 + leapsToDate + MonthOffset[tm->tm_mon-1] + 1031 tm->tm_mday; 1032 1033 tm->tm_wday = day % 7; 1034 } 1035 EXPORT_SYMBOL_GPL(GregorianDay); 1036 1037 void to_tm(int tim, struct rtc_time * tm) 1038 { 1039 register int i; 1040 register long hms, day; 1041 1042 day = tim / SECDAY; 1043 hms = tim % SECDAY; 1044 1045 /* Hours, minutes, seconds are easy */ 1046 tm->tm_hour = hms / 3600; 1047 tm->tm_min = (hms % 3600) / 60; 1048 tm->tm_sec = (hms % 3600) % 60; 1049 1050 /* Number of years in days */ 1051 for (i = STARTOFTIME; day >= days_in_year(i); i++) 1052 day -= days_in_year(i); 1053 tm->tm_year = i; 1054 1055 /* Number of months in days left */ 1056 if (leapyear(tm->tm_year)) 1057 days_in_month(FEBRUARY) = 29; 1058 for (i = 1; day >= days_in_month(i); i++) 1059 day -= days_in_month(i); 1060 days_in_month(FEBRUARY) = 28; 1061 tm->tm_mon = i; 1062 1063 /* Days are what is left over (+1) from all that. */ 1064 tm->tm_mday = day + 1; 1065 1066 /* 1067 * Determine the day of week 1068 */ 1069 GregorianDay(tm); 1070 } 1071 EXPORT_SYMBOL(to_tm); 1072 1073 /* 1074 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1075 * result. 1076 */ 1077 void div128_by_32(u64 dividend_high, u64 dividend_low, 1078 unsigned divisor, struct div_result *dr) 1079 { 1080 unsigned long a, b, c, d; 1081 unsigned long w, x, y, z; 1082 u64 ra, rb, rc; 1083 1084 a = dividend_high >> 32; 1085 b = dividend_high & 0xffffffff; 1086 c = dividend_low >> 32; 1087 d = dividend_low & 0xffffffff; 1088 1089 w = a / divisor; 1090 ra = ((u64)(a - (w * divisor)) << 32) + b; 1091 1092 rb = ((u64) do_div(ra, divisor) << 32) + c; 1093 x = ra; 1094 1095 rc = ((u64) do_div(rb, divisor) << 32) + d; 1096 y = rb; 1097 1098 do_div(rc, divisor); 1099 z = rc; 1100 1101 dr->result_high = ((u64)w << 32) + x; 1102 dr->result_low = ((u64)y << 32) + z; 1103 1104 } 1105 1106 /* We don't need to calibrate delay, we use the CPU timebase for that */ 1107 void calibrate_delay(void) 1108 { 1109 /* Some generic code (such as spinlock debug) use loops_per_jiffy 1110 * as the number of __delay(1) in a jiffy, so make it so 1111 */ 1112 loops_per_jiffy = tb_ticks_per_jiffy; 1113 } 1114 1115 static int __init rtc_init(void) 1116 { 1117 struct platform_device *pdev; 1118 1119 if (!ppc_md.get_rtc_time) 1120 return -ENODEV; 1121 1122 pdev = platform_device_register_simple("rtc-generic", -1, NULL, 0); 1123 1124 return PTR_ERR_OR_ZERO(pdev); 1125 } 1126 1127 device_initcall(rtc_init); 1128