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 <linux/suspend.h> 59 #include <linux/rtc.h> 60 #include <asm/trace.h> 61 62 #include <asm/io.h> 63 #include <asm/processor.h> 64 #include <asm/nvram.h> 65 #include <asm/cache.h> 66 #include <asm/machdep.h> 67 #include <linux/uaccess.h> 68 #include <asm/time.h> 69 #include <asm/prom.h> 70 #include <asm/irq.h> 71 #include <asm/div64.h> 72 #include <asm/smp.h> 73 #include <asm/vdso_datapage.h> 74 #include <asm/firmware.h> 75 #include <asm/cputime.h> 76 #include <asm/asm-prototypes.h> 77 78 /* powerpc clocksource/clockevent code */ 79 80 #include <linux/clockchips.h> 81 #include <linux/timekeeper_internal.h> 82 83 static u64 rtc_read(struct clocksource *); 84 static struct clocksource clocksource_rtc = { 85 .name = "rtc", 86 .rating = 400, 87 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 88 .mask = CLOCKSOURCE_MASK(64), 89 .read = rtc_read, 90 }; 91 92 static u64 timebase_read(struct clocksource *); 93 static struct clocksource clocksource_timebase = { 94 .name = "timebase", 95 .rating = 400, 96 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 97 .mask = CLOCKSOURCE_MASK(64), 98 .read = timebase_read, 99 }; 100 101 #define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF 102 u64 decrementer_max = DECREMENTER_DEFAULT_MAX; 103 104 static int decrementer_set_next_event(unsigned long evt, 105 struct clock_event_device *dev); 106 static int decrementer_shutdown(struct clock_event_device *evt); 107 108 struct clock_event_device decrementer_clockevent = { 109 .name = "decrementer", 110 .rating = 200, 111 .irq = 0, 112 .set_next_event = decrementer_set_next_event, 113 .set_state_shutdown = decrementer_shutdown, 114 .tick_resume = decrementer_shutdown, 115 .features = CLOCK_EVT_FEAT_ONESHOT | 116 CLOCK_EVT_FEAT_C3STOP, 117 }; 118 EXPORT_SYMBOL(decrementer_clockevent); 119 120 DEFINE_PER_CPU(u64, decrementers_next_tb); 121 static DEFINE_PER_CPU(struct clock_event_device, decrementers); 122 123 #define XSEC_PER_SEC (1024*1024) 124 125 #ifdef CONFIG_PPC64 126 #define SCALE_XSEC(xsec, max) (((xsec) * max) / XSEC_PER_SEC) 127 #else 128 /* compute ((xsec << 12) * max) >> 32 */ 129 #define SCALE_XSEC(xsec, max) mulhwu((xsec) << 12, max) 130 #endif 131 132 unsigned long tb_ticks_per_jiffy; 133 unsigned long tb_ticks_per_usec = 100; /* sane default */ 134 EXPORT_SYMBOL(tb_ticks_per_usec); 135 unsigned long tb_ticks_per_sec; 136 EXPORT_SYMBOL(tb_ticks_per_sec); /* for cputime_t conversions */ 137 138 DEFINE_SPINLOCK(rtc_lock); 139 EXPORT_SYMBOL_GPL(rtc_lock); 140 141 static u64 tb_to_ns_scale __read_mostly; 142 static unsigned tb_to_ns_shift __read_mostly; 143 static u64 boot_tb __read_mostly; 144 145 extern struct timezone sys_tz; 146 static long timezone_offset; 147 148 unsigned long ppc_proc_freq; 149 EXPORT_SYMBOL_GPL(ppc_proc_freq); 150 unsigned long ppc_tb_freq; 151 EXPORT_SYMBOL_GPL(ppc_tb_freq); 152 153 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE 154 /* 155 * Factors for converting from cputime_t (timebase ticks) to 156 * jiffies, microseconds, seconds, and clock_t (1/USER_HZ seconds). 157 * These are all stored as 0.64 fixed-point binary fractions. 158 */ 159 u64 __cputime_jiffies_factor; 160 EXPORT_SYMBOL(__cputime_jiffies_factor); 161 u64 __cputime_usec_factor; 162 EXPORT_SYMBOL(__cputime_usec_factor); 163 u64 __cputime_sec_factor; 164 EXPORT_SYMBOL(__cputime_sec_factor); 165 u64 __cputime_clockt_factor; 166 EXPORT_SYMBOL(__cputime_clockt_factor); 167 168 cputime_t cputime_one_jiffy; 169 170 #ifdef CONFIG_PPC_SPLPAR 171 void (*dtl_consumer)(struct dtl_entry *, u64); 172 #endif 173 174 #ifdef CONFIG_PPC64 175 #define get_accounting(tsk) (&get_paca()->accounting) 176 #else 177 #define get_accounting(tsk) (&task_thread_info(tsk)->accounting) 178 #endif 179 180 static void calc_cputime_factors(void) 181 { 182 struct div_result res; 183 184 div128_by_32(HZ, 0, tb_ticks_per_sec, &res); 185 __cputime_jiffies_factor = res.result_low; 186 div128_by_32(1000000, 0, tb_ticks_per_sec, &res); 187 __cputime_usec_factor = res.result_low; 188 div128_by_32(1, 0, tb_ticks_per_sec, &res); 189 __cputime_sec_factor = res.result_low; 190 div128_by_32(USER_HZ, 0, tb_ticks_per_sec, &res); 191 __cputime_clockt_factor = res.result_low; 192 } 193 194 /* 195 * Read the SPURR on systems that have it, otherwise the PURR, 196 * or if that doesn't exist return the timebase value passed in. 197 */ 198 static unsigned long read_spurr(unsigned long tb) 199 { 200 if (cpu_has_feature(CPU_FTR_SPURR)) 201 return mfspr(SPRN_SPURR); 202 if (cpu_has_feature(CPU_FTR_PURR)) 203 return mfspr(SPRN_PURR); 204 return tb; 205 } 206 207 #ifdef CONFIG_PPC_SPLPAR 208 209 /* 210 * Scan the dispatch trace log and count up the stolen time. 211 * Should be called with interrupts disabled. 212 */ 213 static u64 scan_dispatch_log(u64 stop_tb) 214 { 215 u64 i = local_paca->dtl_ridx; 216 struct dtl_entry *dtl = local_paca->dtl_curr; 217 struct dtl_entry *dtl_end = local_paca->dispatch_log_end; 218 struct lppaca *vpa = local_paca->lppaca_ptr; 219 u64 tb_delta; 220 u64 stolen = 0; 221 u64 dtb; 222 223 if (!dtl) 224 return 0; 225 226 if (i == be64_to_cpu(vpa->dtl_idx)) 227 return 0; 228 while (i < be64_to_cpu(vpa->dtl_idx)) { 229 dtb = be64_to_cpu(dtl->timebase); 230 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) + 231 be32_to_cpu(dtl->ready_to_enqueue_time); 232 barrier(); 233 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) { 234 /* buffer has overflowed */ 235 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG; 236 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG); 237 continue; 238 } 239 if (dtb > stop_tb) 240 break; 241 if (dtl_consumer) 242 dtl_consumer(dtl, i); 243 stolen += tb_delta; 244 ++i; 245 ++dtl; 246 if (dtl == dtl_end) 247 dtl = local_paca->dispatch_log; 248 } 249 local_paca->dtl_ridx = i; 250 local_paca->dtl_curr = dtl; 251 return stolen; 252 } 253 254 /* 255 * Accumulate stolen time by scanning the dispatch trace log. 256 * Called on entry from user mode. 257 */ 258 void accumulate_stolen_time(void) 259 { 260 u64 sst, ust; 261 u8 save_soft_enabled = local_paca->soft_enabled; 262 struct cpu_accounting_data *acct = &local_paca->accounting; 263 264 /* We are called early in the exception entry, before 265 * soft/hard_enabled are sync'ed to the expected state 266 * for the exception. We are hard disabled but the PACA 267 * needs to reflect that so various debug stuff doesn't 268 * complain 269 */ 270 local_paca->soft_enabled = 0; 271 272 sst = scan_dispatch_log(acct->starttime_user); 273 ust = scan_dispatch_log(acct->starttime); 274 acct->stime -= sst; 275 acct->utime -= ust; 276 acct->steal_time += ust + sst; 277 278 local_paca->soft_enabled = save_soft_enabled; 279 } 280 281 static inline u64 calculate_stolen_time(u64 stop_tb) 282 { 283 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) 284 return scan_dispatch_log(stop_tb); 285 286 return 0; 287 } 288 289 #else /* CONFIG_PPC_SPLPAR */ 290 static inline u64 calculate_stolen_time(u64 stop_tb) 291 { 292 return 0; 293 } 294 295 #endif /* CONFIG_PPC_SPLPAR */ 296 297 /* 298 * Account time for a transition between system, hard irq 299 * or soft irq state. 300 */ 301 static unsigned long vtime_delta(struct task_struct *tsk, 302 unsigned long *stime_scaled, 303 unsigned long *steal_time) 304 { 305 unsigned long now, nowscaled, deltascaled; 306 unsigned long stime; 307 unsigned long utime, utime_scaled; 308 struct cpu_accounting_data *acct = get_accounting(tsk); 309 310 WARN_ON_ONCE(!irqs_disabled()); 311 312 now = mftb(); 313 nowscaled = read_spurr(now); 314 stime = now - acct->starttime; 315 acct->starttime = now; 316 deltascaled = nowscaled - acct->startspurr; 317 acct->startspurr = nowscaled; 318 319 *steal_time = calculate_stolen_time(now); 320 321 utime = acct->utime - acct->utime_sspurr; 322 acct->utime_sspurr = acct->utime; 323 324 /* 325 * Because we don't read the SPURR on every kernel entry/exit, 326 * deltascaled includes both user and system SPURR ticks. 327 * Apportion these ticks to system SPURR ticks and user 328 * SPURR ticks in the same ratio as the system time (delta) 329 * and user time (udelta) values obtained from the timebase 330 * over the same interval. The system ticks get accounted here; 331 * the user ticks get saved up in paca->user_time_scaled to be 332 * used by account_process_tick. 333 */ 334 *stime_scaled = stime; 335 utime_scaled = utime; 336 if (deltascaled != stime + utime) { 337 if (utime) { 338 *stime_scaled = deltascaled * stime / (stime + utime); 339 utime_scaled = deltascaled - *stime_scaled; 340 } else { 341 *stime_scaled = deltascaled; 342 } 343 } 344 acct->utime_scaled += utime_scaled; 345 346 return stime; 347 } 348 349 void vtime_account_system(struct task_struct *tsk) 350 { 351 unsigned long stime, stime_scaled, steal_time; 352 struct cpu_accounting_data *acct = get_accounting(tsk); 353 354 stime = vtime_delta(tsk, &stime_scaled, &steal_time); 355 356 stime -= min(stime, steal_time); 357 acct->steal_time += steal_time; 358 359 if ((tsk->flags & PF_VCPU) && !irq_count()) { 360 acct->gtime += stime; 361 acct->utime_scaled += stime_scaled; 362 } else { 363 if (hardirq_count()) 364 acct->hardirq_time += stime; 365 else if (in_serving_softirq()) 366 acct->softirq_time += stime; 367 else 368 acct->stime += stime; 369 370 acct->stime_scaled += stime_scaled; 371 } 372 } 373 EXPORT_SYMBOL_GPL(vtime_account_system); 374 375 void vtime_account_idle(struct task_struct *tsk) 376 { 377 unsigned long stime, stime_scaled, steal_time; 378 struct cpu_accounting_data *acct = get_accounting(tsk); 379 380 stime = vtime_delta(tsk, &stime_scaled, &steal_time); 381 acct->idle_time += stime + steal_time; 382 } 383 384 /* 385 * Account the whole cputime accumulated in the paca 386 * Must be called with interrupts disabled. 387 * Assumes that vtime_account_system/idle() has been called 388 * recently (i.e. since the last entry from usermode) so that 389 * get_paca()->user_time_scaled is up to date. 390 */ 391 void vtime_flush(struct task_struct *tsk) 392 { 393 struct cpu_accounting_data *acct = get_accounting(tsk); 394 395 if (acct->utime) 396 account_user_time(tsk, cputime_to_nsecs(acct->utime)); 397 398 if (acct->utime_scaled) 399 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled); 400 401 if (acct->gtime) 402 account_guest_time(tsk, acct->gtime); 403 404 if (acct->steal_time) 405 account_steal_time(acct->steal_time); 406 407 if (acct->idle_time) 408 account_idle_time(acct->idle_time); 409 410 if (acct->stime) 411 account_system_index_time(tsk, acct->stime, CPUTIME_SYSTEM); 412 413 if (acct->stime_scaled) 414 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled); 415 416 if (acct->hardirq_time) 417 account_system_index_time(tsk, acct->hardirq_time, CPUTIME_IRQ); 418 419 if (acct->softirq_time) 420 account_system_index_time(tsk, acct->softirq_time, CPUTIME_SOFTIRQ); 421 422 acct->utime = 0; 423 acct->utime_scaled = 0; 424 acct->utime_sspurr = 0; 425 acct->gtime = 0; 426 acct->steal_time = 0; 427 acct->idle_time = 0; 428 acct->stime = 0; 429 acct->stime_scaled = 0; 430 acct->hardirq_time = 0; 431 acct->softirq_time = 0; 432 } 433 434 #ifdef CONFIG_PPC32 435 /* 436 * Called from the context switch with interrupts disabled, to charge all 437 * accumulated times to the current process, and to prepare accounting on 438 * the next process. 439 */ 440 void arch_vtime_task_switch(struct task_struct *prev) 441 { 442 struct cpu_accounting_data *acct = get_accounting(current); 443 444 acct->starttime = get_accounting(prev)->starttime; 445 acct->startspurr = get_accounting(prev)->startspurr; 446 } 447 #endif /* CONFIG_PPC32 */ 448 449 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ 450 #define calc_cputime_factors() 451 #endif 452 453 void __delay(unsigned long loops) 454 { 455 unsigned long start; 456 int diff; 457 458 if (__USE_RTC()) { 459 start = get_rtcl(); 460 do { 461 /* the RTCL register wraps at 1000000000 */ 462 diff = get_rtcl() - start; 463 if (diff < 0) 464 diff += 1000000000; 465 } while (diff < loops); 466 } else { 467 start = get_tbl(); 468 while (get_tbl() - start < loops) 469 HMT_low(); 470 HMT_medium(); 471 } 472 } 473 EXPORT_SYMBOL(__delay); 474 475 void udelay(unsigned long usecs) 476 { 477 __delay(tb_ticks_per_usec * usecs); 478 } 479 EXPORT_SYMBOL(udelay); 480 481 #ifdef CONFIG_SMP 482 unsigned long profile_pc(struct pt_regs *regs) 483 { 484 unsigned long pc = instruction_pointer(regs); 485 486 if (in_lock_functions(pc)) 487 return regs->link; 488 489 return pc; 490 } 491 EXPORT_SYMBOL(profile_pc); 492 #endif 493 494 #ifdef CONFIG_IRQ_WORK 495 496 /* 497 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable... 498 */ 499 #ifdef CONFIG_PPC64 500 static inline unsigned long test_irq_work_pending(void) 501 { 502 unsigned long x; 503 504 asm volatile("lbz %0,%1(13)" 505 : "=r" (x) 506 : "i" (offsetof(struct paca_struct, irq_work_pending))); 507 return x; 508 } 509 510 static inline void set_irq_work_pending_flag(void) 511 { 512 asm volatile("stb %0,%1(13)" : : 513 "r" (1), 514 "i" (offsetof(struct paca_struct, irq_work_pending))); 515 } 516 517 static inline void clear_irq_work_pending(void) 518 { 519 asm volatile("stb %0,%1(13)" : : 520 "r" (0), 521 "i" (offsetof(struct paca_struct, irq_work_pending))); 522 } 523 524 #else /* 32-bit */ 525 526 DEFINE_PER_CPU(u8, irq_work_pending); 527 528 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1) 529 #define test_irq_work_pending() __this_cpu_read(irq_work_pending) 530 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0) 531 532 #endif /* 32 vs 64 bit */ 533 534 void arch_irq_work_raise(void) 535 { 536 preempt_disable(); 537 set_irq_work_pending_flag(); 538 set_dec(1); 539 preempt_enable(); 540 } 541 542 #else /* CONFIG_IRQ_WORK */ 543 544 #define test_irq_work_pending() 0 545 #define clear_irq_work_pending() 546 547 #endif /* CONFIG_IRQ_WORK */ 548 549 static void __timer_interrupt(void) 550 { 551 struct pt_regs *regs = get_irq_regs(); 552 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 553 struct clock_event_device *evt = this_cpu_ptr(&decrementers); 554 u64 now; 555 556 trace_timer_interrupt_entry(regs); 557 558 if (test_irq_work_pending()) { 559 clear_irq_work_pending(); 560 irq_work_run(); 561 } 562 563 now = get_tb_or_rtc(); 564 if (now >= *next_tb) { 565 *next_tb = ~(u64)0; 566 if (evt->event_handler) 567 evt->event_handler(evt); 568 __this_cpu_inc(irq_stat.timer_irqs_event); 569 } else { 570 now = *next_tb - now; 571 if (now <= decrementer_max) 572 set_dec(now); 573 /* We may have raced with new irq work */ 574 if (test_irq_work_pending()) 575 set_dec(1); 576 __this_cpu_inc(irq_stat.timer_irqs_others); 577 } 578 579 #ifdef CONFIG_PPC64 580 /* collect purr register values often, for accurate calculations */ 581 if (firmware_has_feature(FW_FEATURE_SPLPAR)) { 582 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array); 583 cu->current_tb = mfspr(SPRN_PURR); 584 } 585 #endif 586 587 trace_timer_interrupt_exit(regs); 588 } 589 590 /* 591 * timer_interrupt - gets called when the decrementer overflows, 592 * with interrupts disabled. 593 */ 594 void timer_interrupt(struct pt_regs * regs) 595 { 596 struct pt_regs *old_regs; 597 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 598 599 /* Ensure a positive value is written to the decrementer, or else 600 * some CPUs will continue to take decrementer exceptions. 601 */ 602 set_dec(decrementer_max); 603 604 /* Some implementations of hotplug will get timer interrupts while 605 * offline, just ignore these and we also need to set 606 * decrementers_next_tb as MAX to make sure __check_irq_replay 607 * don't replay timer interrupt when return, otherwise we'll trap 608 * here infinitely :( 609 */ 610 if (!cpu_online(smp_processor_id())) { 611 *next_tb = ~(u64)0; 612 return; 613 } 614 615 /* Conditionally hard-enable interrupts now that the DEC has been 616 * bumped to its maximum value 617 */ 618 may_hard_irq_enable(); 619 620 621 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC) 622 if (atomic_read(&ppc_n_lost_interrupts) != 0) 623 do_IRQ(regs); 624 #endif 625 626 old_regs = set_irq_regs(regs); 627 irq_enter(); 628 629 __timer_interrupt(); 630 irq_exit(); 631 set_irq_regs(old_regs); 632 } 633 EXPORT_SYMBOL(timer_interrupt); 634 635 /* 636 * Hypervisor decrementer interrupts shouldn't occur but are sometimes 637 * left pending on exit from a KVM guest. We don't need to do anything 638 * to clear them, as they are edge-triggered. 639 */ 640 void hdec_interrupt(struct pt_regs *regs) 641 { 642 } 643 644 #ifdef CONFIG_SUSPEND 645 static void generic_suspend_disable_irqs(void) 646 { 647 /* Disable the decrementer, so that it doesn't interfere 648 * with suspending. 649 */ 650 651 set_dec(decrementer_max); 652 local_irq_disable(); 653 set_dec(decrementer_max); 654 } 655 656 static void generic_suspend_enable_irqs(void) 657 { 658 local_irq_enable(); 659 } 660 661 /* Overrides the weak version in kernel/power/main.c */ 662 void arch_suspend_disable_irqs(void) 663 { 664 if (ppc_md.suspend_disable_irqs) 665 ppc_md.suspend_disable_irqs(); 666 generic_suspend_disable_irqs(); 667 } 668 669 /* Overrides the weak version in kernel/power/main.c */ 670 void arch_suspend_enable_irqs(void) 671 { 672 generic_suspend_enable_irqs(); 673 if (ppc_md.suspend_enable_irqs) 674 ppc_md.suspend_enable_irqs(); 675 } 676 #endif 677 678 unsigned long long tb_to_ns(unsigned long long ticks) 679 { 680 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift; 681 } 682 EXPORT_SYMBOL_GPL(tb_to_ns); 683 684 /* 685 * Scheduler clock - returns current time in nanosec units. 686 * 687 * Note: mulhdu(a, b) (multiply high double unsigned) returns 688 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 689 * are 64-bit unsigned numbers. 690 */ 691 unsigned long long sched_clock(void) 692 { 693 if (__USE_RTC()) 694 return get_rtc(); 695 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 696 } 697 698 699 #ifdef CONFIG_PPC_PSERIES 700 701 /* 702 * Running clock - attempts to give a view of time passing for a virtualised 703 * kernels. 704 * Uses the VTB register if available otherwise a next best guess. 705 */ 706 unsigned long long running_clock(void) 707 { 708 /* 709 * Don't read the VTB as a host since KVM does not switch in host 710 * timebase into the VTB when it takes a guest off the CPU, reading the 711 * VTB would result in reading 'last switched out' guest VTB. 712 * 713 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it 714 * would be unsafe to rely only on the #ifdef above. 715 */ 716 if (firmware_has_feature(FW_FEATURE_LPAR) && 717 cpu_has_feature(CPU_FTR_ARCH_207S)) 718 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 719 720 /* 721 * This is a next best approximation without a VTB. 722 * On a host which is running bare metal there should never be any stolen 723 * time and on a host which doesn't do any virtualisation TB *should* equal 724 * VTB so it makes no difference anyway. 725 */ 726 return local_clock() - cputime_to_nsecs(kcpustat_this_cpu->cpustat[CPUTIME_STEAL]); 727 } 728 #endif 729 730 static int __init get_freq(char *name, int cells, unsigned long *val) 731 { 732 struct device_node *cpu; 733 const __be32 *fp; 734 int found = 0; 735 736 /* The cpu node should have timebase and clock frequency properties */ 737 cpu = of_find_node_by_type(NULL, "cpu"); 738 739 if (cpu) { 740 fp = of_get_property(cpu, name, NULL); 741 if (fp) { 742 found = 1; 743 *val = of_read_ulong(fp, cells); 744 } 745 746 of_node_put(cpu); 747 } 748 749 return found; 750 } 751 752 static void start_cpu_decrementer(void) 753 { 754 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 755 /* Clear any pending timer interrupts */ 756 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 757 758 /* Enable decrementer interrupt */ 759 mtspr(SPRN_TCR, TCR_DIE); 760 #endif /* defined(CONFIG_BOOKE) || defined(CONFIG_40x) */ 761 } 762 763 void __init generic_calibrate_decr(void) 764 { 765 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 766 767 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 768 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 769 770 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 771 "(not found)\n"); 772 } 773 774 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 775 776 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 777 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 778 779 printk(KERN_ERR "WARNING: Estimating processor frequency " 780 "(not found)\n"); 781 } 782 } 783 784 int update_persistent_clock(struct timespec now) 785 { 786 struct rtc_time tm; 787 788 if (!ppc_md.set_rtc_time) 789 return -ENODEV; 790 791 to_tm(now.tv_sec + 1 + timezone_offset, &tm); 792 tm.tm_year -= 1900; 793 tm.tm_mon -= 1; 794 795 return ppc_md.set_rtc_time(&tm); 796 } 797 798 static void __read_persistent_clock(struct timespec *ts) 799 { 800 struct rtc_time tm; 801 static int first = 1; 802 803 ts->tv_nsec = 0; 804 /* XXX this is a litle fragile but will work okay in the short term */ 805 if (first) { 806 first = 0; 807 if (ppc_md.time_init) 808 timezone_offset = ppc_md.time_init(); 809 810 /* get_boot_time() isn't guaranteed to be safe to call late */ 811 if (ppc_md.get_boot_time) { 812 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset; 813 return; 814 } 815 } 816 if (!ppc_md.get_rtc_time) { 817 ts->tv_sec = 0; 818 return; 819 } 820 ppc_md.get_rtc_time(&tm); 821 822 ts->tv_sec = mktime(tm.tm_year+1900, tm.tm_mon+1, tm.tm_mday, 823 tm.tm_hour, tm.tm_min, tm.tm_sec); 824 } 825 826 void read_persistent_clock(struct timespec *ts) 827 { 828 __read_persistent_clock(ts); 829 830 /* Sanitize it in case real time clock is set below EPOCH */ 831 if (ts->tv_sec < 0) { 832 ts->tv_sec = 0; 833 ts->tv_nsec = 0; 834 } 835 836 } 837 838 /* clocksource code */ 839 static u64 rtc_read(struct clocksource *cs) 840 { 841 return (u64)get_rtc(); 842 } 843 844 static u64 timebase_read(struct clocksource *cs) 845 { 846 return (u64)get_tb(); 847 } 848 849 void update_vsyscall_old(struct timespec *wall_time, struct timespec *wtm, 850 struct clocksource *clock, u32 mult, u64 cycle_last) 851 { 852 u64 new_tb_to_xs, new_stamp_xsec; 853 u32 frac_sec; 854 855 if (clock != &clocksource_timebase) 856 return; 857 858 /* Make userspace gettimeofday spin until we're done. */ 859 ++vdso_data->tb_update_count; 860 smp_mb(); 861 862 /* 19342813113834067 ~= 2^(20+64) / 1e9 */ 863 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift); 864 new_stamp_xsec = (u64) wall_time->tv_nsec * XSEC_PER_SEC; 865 do_div(new_stamp_xsec, 1000000000); 866 new_stamp_xsec += (u64) wall_time->tv_sec * XSEC_PER_SEC; 867 868 BUG_ON(wall_time->tv_nsec >= NSEC_PER_SEC); 869 /* this is tv_nsec / 1e9 as a 0.32 fraction */ 870 frac_sec = ((u64) wall_time->tv_nsec * 18446744073ULL) >> 32; 871 872 /* 873 * tb_update_count is used to allow the userspace gettimeofday code 874 * to assure itself that it sees a consistent view of the tb_to_xs and 875 * stamp_xsec variables. It reads the tb_update_count, then reads 876 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 877 * the two values of tb_update_count match and are even then the 878 * tb_to_xs and stamp_xsec values are consistent. If not, then it 879 * loops back and reads them again until this criteria is met. 880 * We expect the caller to have done the first increment of 881 * vdso_data->tb_update_count already. 882 */ 883 vdso_data->tb_orig_stamp = cycle_last; 884 vdso_data->stamp_xsec = new_stamp_xsec; 885 vdso_data->tb_to_xs = new_tb_to_xs; 886 vdso_data->wtom_clock_sec = wtm->tv_sec; 887 vdso_data->wtom_clock_nsec = wtm->tv_nsec; 888 vdso_data->stamp_xtime = *wall_time; 889 vdso_data->stamp_sec_fraction = frac_sec; 890 smp_wmb(); 891 ++(vdso_data->tb_update_count); 892 } 893 894 void update_vsyscall_tz(void) 895 { 896 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 897 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 898 } 899 900 static void __init clocksource_init(void) 901 { 902 struct clocksource *clock; 903 904 if (__USE_RTC()) 905 clock = &clocksource_rtc; 906 else 907 clock = &clocksource_timebase; 908 909 if (clocksource_register_hz(clock, tb_ticks_per_sec)) { 910 printk(KERN_ERR "clocksource: %s is already registered\n", 911 clock->name); 912 return; 913 } 914 915 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 916 clock->name, clock->mult, clock->shift); 917 } 918 919 static int decrementer_set_next_event(unsigned long evt, 920 struct clock_event_device *dev) 921 { 922 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt); 923 set_dec(evt); 924 925 /* We may have raced with new irq work */ 926 if (test_irq_work_pending()) 927 set_dec(1); 928 929 return 0; 930 } 931 932 static int decrementer_shutdown(struct clock_event_device *dev) 933 { 934 decrementer_set_next_event(decrementer_max, dev); 935 return 0; 936 } 937 938 /* Interrupt handler for the timer broadcast IPI */ 939 void tick_broadcast_ipi_handler(void) 940 { 941 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 942 943 *next_tb = get_tb_or_rtc(); 944 __timer_interrupt(); 945 } 946 947 static void register_decrementer_clockevent(int cpu) 948 { 949 struct clock_event_device *dec = &per_cpu(decrementers, cpu); 950 951 *dec = decrementer_clockevent; 952 dec->cpumask = cpumask_of(cpu); 953 954 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n", 955 dec->name, dec->mult, dec->shift, cpu); 956 957 clockevents_register_device(dec); 958 } 959 960 static void enable_large_decrementer(void) 961 { 962 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 963 return; 964 965 if (decrementer_max <= DECREMENTER_DEFAULT_MAX) 966 return; 967 968 /* 969 * If we're running as the hypervisor we need to enable the LD manually 970 * otherwise firmware should have done it for us. 971 */ 972 if (cpu_has_feature(CPU_FTR_HVMODE)) 973 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD); 974 } 975 976 static void __init set_decrementer_max(void) 977 { 978 struct device_node *cpu; 979 u32 bits = 32; 980 981 /* Prior to ISAv3 the decrementer is always 32 bit */ 982 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 983 return; 984 985 cpu = of_find_node_by_type(NULL, "cpu"); 986 987 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) { 988 if (bits > 64 || bits < 32) { 989 pr_warn("time_init: firmware supplied invalid ibm,dec-bits"); 990 bits = 32; 991 } 992 993 /* calculate the signed maximum given this many bits */ 994 decrementer_max = (1ul << (bits - 1)) - 1; 995 } 996 997 of_node_put(cpu); 998 999 pr_info("time_init: %u bit decrementer (max: %llx)\n", 1000 bits, decrementer_max); 1001 } 1002 1003 static void __init init_decrementer_clockevent(void) 1004 { 1005 int cpu = smp_processor_id(); 1006 1007 clockevents_calc_mult_shift(&decrementer_clockevent, ppc_tb_freq, 4); 1008 1009 decrementer_clockevent.max_delta_ns = 1010 clockevent_delta2ns(decrementer_max, &decrementer_clockevent); 1011 decrementer_clockevent.min_delta_ns = 1012 clockevent_delta2ns(2, &decrementer_clockevent); 1013 1014 register_decrementer_clockevent(cpu); 1015 } 1016 1017 void secondary_cpu_time_init(void) 1018 { 1019 /* Enable and test the large decrementer for this cpu */ 1020 enable_large_decrementer(); 1021 1022 /* Start the decrementer on CPUs that have manual control 1023 * such as BookE 1024 */ 1025 start_cpu_decrementer(); 1026 1027 /* FIME: Should make unrelatred change to move snapshot_timebase 1028 * call here ! */ 1029 register_decrementer_clockevent(smp_processor_id()); 1030 } 1031 1032 /* This function is only called on the boot processor */ 1033 void __init time_init(void) 1034 { 1035 struct div_result res; 1036 u64 scale; 1037 unsigned shift; 1038 1039 if (__USE_RTC()) { 1040 /* 601 processor: dec counts down by 128 every 128ns */ 1041 ppc_tb_freq = 1000000000; 1042 } else { 1043 /* Normal PowerPC with timebase register */ 1044 ppc_md.calibrate_decr(); 1045 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 1046 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 1047 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 1048 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 1049 } 1050 1051 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 1052 tb_ticks_per_sec = ppc_tb_freq; 1053 tb_ticks_per_usec = ppc_tb_freq / 1000000; 1054 calc_cputime_factors(); 1055 setup_cputime_one_jiffy(); 1056 1057 /* 1058 * Compute scale factor for sched_clock. 1059 * The calibrate_decr() function has set tb_ticks_per_sec, 1060 * which is the timebase frequency. 1061 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 1062 * the 128-bit result as a 64.64 fixed-point number. 1063 * We then shift that number right until it is less than 1.0, 1064 * giving us the scale factor and shift count to use in 1065 * sched_clock(). 1066 */ 1067 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 1068 scale = res.result_low; 1069 for (shift = 0; res.result_high != 0; ++shift) { 1070 scale = (scale >> 1) | (res.result_high << 63); 1071 res.result_high >>= 1; 1072 } 1073 tb_to_ns_scale = scale; 1074 tb_to_ns_shift = shift; 1075 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 1076 boot_tb = get_tb_or_rtc(); 1077 1078 /* If platform provided a timezone (pmac), we correct the time */ 1079 if (timezone_offset) { 1080 sys_tz.tz_minuteswest = -timezone_offset / 60; 1081 sys_tz.tz_dsttime = 0; 1082 } 1083 1084 vdso_data->tb_update_count = 0; 1085 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 1086 1087 /* initialise and enable the large decrementer (if we have one) */ 1088 set_decrementer_max(); 1089 enable_large_decrementer(); 1090 1091 /* Start the decrementer on CPUs that have manual control 1092 * such as BookE 1093 */ 1094 start_cpu_decrementer(); 1095 1096 /* Register the clocksource */ 1097 clocksource_init(); 1098 1099 init_decrementer_clockevent(); 1100 tick_setup_hrtimer_broadcast(); 1101 1102 #ifdef CONFIG_COMMON_CLK 1103 of_clk_init(NULL); 1104 #endif 1105 } 1106 1107 1108 #define FEBRUARY 2 1109 #define STARTOFTIME 1970 1110 #define SECDAY 86400L 1111 #define SECYR (SECDAY * 365) 1112 #define leapyear(year) ((year) % 4 == 0 && \ 1113 ((year) % 100 != 0 || (year) % 400 == 0)) 1114 #define days_in_year(a) (leapyear(a) ? 366 : 365) 1115 #define days_in_month(a) (month_days[(a) - 1]) 1116 1117 static int month_days[12] = { 1118 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 1119 }; 1120 1121 void to_tm(int tim, struct rtc_time * tm) 1122 { 1123 register int i; 1124 register long hms, day; 1125 1126 day = tim / SECDAY; 1127 hms = tim % SECDAY; 1128 1129 /* Hours, minutes, seconds are easy */ 1130 tm->tm_hour = hms / 3600; 1131 tm->tm_min = (hms % 3600) / 60; 1132 tm->tm_sec = (hms % 3600) % 60; 1133 1134 /* Number of years in days */ 1135 for (i = STARTOFTIME; day >= days_in_year(i); i++) 1136 day -= days_in_year(i); 1137 tm->tm_year = i; 1138 1139 /* Number of months in days left */ 1140 if (leapyear(tm->tm_year)) 1141 days_in_month(FEBRUARY) = 29; 1142 for (i = 1; day >= days_in_month(i); i++) 1143 day -= days_in_month(i); 1144 days_in_month(FEBRUARY) = 28; 1145 tm->tm_mon = i; 1146 1147 /* Days are what is left over (+1) from all that. */ 1148 tm->tm_mday = day + 1; 1149 1150 /* 1151 * No-one uses the day of the week. 1152 */ 1153 tm->tm_wday = -1; 1154 } 1155 EXPORT_SYMBOL(to_tm); 1156 1157 /* 1158 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1159 * result. 1160 */ 1161 void div128_by_32(u64 dividend_high, u64 dividend_low, 1162 unsigned divisor, struct div_result *dr) 1163 { 1164 unsigned long a, b, c, d; 1165 unsigned long w, x, y, z; 1166 u64 ra, rb, rc; 1167 1168 a = dividend_high >> 32; 1169 b = dividend_high & 0xffffffff; 1170 c = dividend_low >> 32; 1171 d = dividend_low & 0xffffffff; 1172 1173 w = a / divisor; 1174 ra = ((u64)(a - (w * divisor)) << 32) + b; 1175 1176 rb = ((u64) do_div(ra, divisor) << 32) + c; 1177 x = ra; 1178 1179 rc = ((u64) do_div(rb, divisor) << 32) + d; 1180 y = rb; 1181 1182 do_div(rc, divisor); 1183 z = rc; 1184 1185 dr->result_high = ((u64)w << 32) + x; 1186 dr->result_low = ((u64)y << 32) + z; 1187 1188 } 1189 1190 /* We don't need to calibrate delay, we use the CPU timebase for that */ 1191 void calibrate_delay(void) 1192 { 1193 /* Some generic code (such as spinlock debug) use loops_per_jiffy 1194 * as the number of __delay(1) in a jiffy, so make it so 1195 */ 1196 loops_per_jiffy = tb_ticks_per_jiffy; 1197 } 1198 1199 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC) 1200 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm) 1201 { 1202 ppc_md.get_rtc_time(tm); 1203 return rtc_valid_tm(tm); 1204 } 1205 1206 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm) 1207 { 1208 if (!ppc_md.set_rtc_time) 1209 return -EOPNOTSUPP; 1210 1211 if (ppc_md.set_rtc_time(tm) < 0) 1212 return -EOPNOTSUPP; 1213 1214 return 0; 1215 } 1216 1217 static const struct rtc_class_ops rtc_generic_ops = { 1218 .read_time = rtc_generic_get_time, 1219 .set_time = rtc_generic_set_time, 1220 }; 1221 1222 static int __init rtc_init(void) 1223 { 1224 struct platform_device *pdev; 1225 1226 if (!ppc_md.get_rtc_time) 1227 return -ENODEV; 1228 1229 pdev = platform_device_register_data(NULL, "rtc-generic", -1, 1230 &rtc_generic_ops, 1231 sizeof(rtc_generic_ops)); 1232 1233 return PTR_ERR_OR_ZERO(pdev); 1234 } 1235 1236 device_initcall(rtc_init); 1237 #endif 1238