1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Common time routines among all ppc machines. 4 * 5 * Written by Cort Dougan (cort@cs.nmt.edu) to merge 6 * Paul Mackerras' version and mine for PReP and Pmac. 7 * MPC8xx/MBX changes by Dan Malek (dmalek@jlc.net). 8 * Converted for 64-bit by Mike Corrigan (mikejc@us.ibm.com) 9 * 10 * First round of bugfixes by Gabriel Paubert (paubert@iram.es) 11 * to make clock more stable (2.4.0-test5). The only thing 12 * that this code assumes is that the timebases have been synchronized 13 * by firmware on SMP and are never stopped (never do sleep 14 * on SMP then, nap and doze are OK). 15 * 16 * Speeded up do_gettimeofday by getting rid of references to 17 * xtime (which required locks for consistency). (mikejc@us.ibm.com) 18 * 19 * TODO (not necessarily in this file): 20 * - improve precision and reproducibility of timebase frequency 21 * measurement at boot time. 22 * - for astronomical applications: add a new function to get 23 * non ambiguous timestamps even around leap seconds. This needs 24 * a new timestamp format and a good name. 25 * 26 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 27 * "A Kernel Model for Precision Timekeeping" by Dave Mills 28 */ 29 30 #include <linux/errno.h> 31 #include <linux/export.h> 32 #include <linux/sched.h> 33 #include <linux/sched/clock.h> 34 #include <linux/kernel.h> 35 #include <linux/param.h> 36 #include <linux/string.h> 37 #include <linux/mm.h> 38 #include <linux/interrupt.h> 39 #include <linux/timex.h> 40 #include <linux/kernel_stat.h> 41 #include <linux/time.h> 42 #include <linux/init.h> 43 #include <linux/profile.h> 44 #include <linux/cpu.h> 45 #include <linux/security.h> 46 #include <linux/percpu.h> 47 #include <linux/rtc.h> 48 #include <linux/jiffies.h> 49 #include <linux/posix-timers.h> 50 #include <linux/irq.h> 51 #include <linux/delay.h> 52 #include <linux/irq_work.h> 53 #include <linux/of_clk.h> 54 #include <linux/suspend.h> 55 #include <linux/sched/cputime.h> 56 #include <linux/processor.h> 57 #include <asm/trace.h> 58 59 #include <asm/io.h> 60 #include <asm/nvram.h> 61 #include <asm/cache.h> 62 #include <asm/machdep.h> 63 #include <linux/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/asm-prototypes.h> 72 73 /* powerpc clocksource/clockevent code */ 74 75 #include <linux/clockchips.h> 76 #include <linux/timekeeper_internal.h> 77 78 static u64 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 u64 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_DEFAULT_MAX 0x7FFFFFFF 97 u64 decrementer_max = DECREMENTER_DEFAULT_MAX; 98 99 static int decrementer_set_next_event(unsigned long evt, 100 struct clock_event_device *dev); 101 static int decrementer_shutdown(struct clock_event_device *evt); 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_state_oneshot_stopped = decrementer_shutdown, 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 bool tb_invalid; 150 151 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE 152 /* 153 * Factor for converting from cputime_t (timebase ticks) to 154 * microseconds. This is stored as 0.64 fixed-point binary fraction. 155 */ 156 u64 __cputime_usec_factor; 157 EXPORT_SYMBOL(__cputime_usec_factor); 158 159 #ifdef CONFIG_PPC_SPLPAR 160 void (*dtl_consumer)(struct dtl_entry *, u64); 161 #endif 162 163 static void calc_cputime_factors(void) 164 { 165 struct div_result res; 166 167 div128_by_32(1000000, 0, tb_ticks_per_sec, &res); 168 __cputime_usec_factor = res.result_low; 169 } 170 171 /* 172 * Read the SPURR on systems that have it, otherwise the PURR, 173 * or if that doesn't exist return the timebase value passed in. 174 */ 175 static inline unsigned long read_spurr(unsigned long tb) 176 { 177 if (cpu_has_feature(CPU_FTR_SPURR)) 178 return mfspr(SPRN_SPURR); 179 if (cpu_has_feature(CPU_FTR_PURR)) 180 return mfspr(SPRN_PURR); 181 return tb; 182 } 183 184 #ifdef CONFIG_PPC_SPLPAR 185 186 /* 187 * Scan the dispatch trace log and count up the stolen time. 188 * Should be called with interrupts disabled. 189 */ 190 static u64 scan_dispatch_log(u64 stop_tb) 191 { 192 u64 i = local_paca->dtl_ridx; 193 struct dtl_entry *dtl = local_paca->dtl_curr; 194 struct dtl_entry *dtl_end = local_paca->dispatch_log_end; 195 struct lppaca *vpa = local_paca->lppaca_ptr; 196 u64 tb_delta; 197 u64 stolen = 0; 198 u64 dtb; 199 200 if (!dtl) 201 return 0; 202 203 if (i == be64_to_cpu(vpa->dtl_idx)) 204 return 0; 205 while (i < be64_to_cpu(vpa->dtl_idx)) { 206 dtb = be64_to_cpu(dtl->timebase); 207 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) + 208 be32_to_cpu(dtl->ready_to_enqueue_time); 209 barrier(); 210 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) { 211 /* buffer has overflowed */ 212 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG; 213 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG); 214 continue; 215 } 216 if (dtb > stop_tb) 217 break; 218 if (dtl_consumer) 219 dtl_consumer(dtl, i); 220 stolen += tb_delta; 221 ++i; 222 ++dtl; 223 if (dtl == dtl_end) 224 dtl = local_paca->dispatch_log; 225 } 226 local_paca->dtl_ridx = i; 227 local_paca->dtl_curr = dtl; 228 return stolen; 229 } 230 231 /* 232 * Accumulate stolen time by scanning the dispatch trace log. 233 * Called on entry from user mode. 234 */ 235 void notrace accumulate_stolen_time(void) 236 { 237 u64 sst, ust; 238 unsigned long save_irq_soft_mask = irq_soft_mask_return(); 239 struct cpu_accounting_data *acct = &local_paca->accounting; 240 241 /* We are called early in the exception entry, before 242 * soft/hard_enabled are sync'ed to the expected state 243 * for the exception. We are hard disabled but the PACA 244 * needs to reflect that so various debug stuff doesn't 245 * complain 246 */ 247 irq_soft_mask_set(IRQS_DISABLED); 248 249 sst = scan_dispatch_log(acct->starttime_user); 250 ust = scan_dispatch_log(acct->starttime); 251 acct->stime -= sst; 252 acct->utime -= ust; 253 acct->steal_time += ust + sst; 254 255 irq_soft_mask_set(save_irq_soft_mask); 256 } 257 258 static inline u64 calculate_stolen_time(u64 stop_tb) 259 { 260 if (!firmware_has_feature(FW_FEATURE_SPLPAR)) 261 return 0; 262 263 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) 264 return scan_dispatch_log(stop_tb); 265 266 return 0; 267 } 268 269 #else /* CONFIG_PPC_SPLPAR */ 270 static inline u64 calculate_stolen_time(u64 stop_tb) 271 { 272 return 0; 273 } 274 275 #endif /* CONFIG_PPC_SPLPAR */ 276 277 /* 278 * Account time for a transition between system, hard irq 279 * or soft irq state. 280 */ 281 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct, 282 unsigned long now, unsigned long stime) 283 { 284 unsigned long stime_scaled = 0; 285 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 286 unsigned long nowscaled, deltascaled; 287 unsigned long utime, utime_scaled; 288 289 nowscaled = read_spurr(now); 290 deltascaled = nowscaled - acct->startspurr; 291 acct->startspurr = nowscaled; 292 utime = acct->utime - acct->utime_sspurr; 293 acct->utime_sspurr = acct->utime; 294 295 /* 296 * Because we don't read the SPURR on every kernel entry/exit, 297 * deltascaled includes both user and system SPURR ticks. 298 * Apportion these ticks to system SPURR ticks and user 299 * SPURR ticks in the same ratio as the system time (delta) 300 * and user time (udelta) values obtained from the timebase 301 * over the same interval. The system ticks get accounted here; 302 * the user ticks get saved up in paca->user_time_scaled to be 303 * used by account_process_tick. 304 */ 305 stime_scaled = stime; 306 utime_scaled = utime; 307 if (deltascaled != stime + utime) { 308 if (utime) { 309 stime_scaled = deltascaled * stime / (stime + utime); 310 utime_scaled = deltascaled - stime_scaled; 311 } else { 312 stime_scaled = deltascaled; 313 } 314 } 315 acct->utime_scaled += utime_scaled; 316 #endif 317 318 return stime_scaled; 319 } 320 321 static unsigned long vtime_delta(struct task_struct *tsk, 322 unsigned long *stime_scaled, 323 unsigned long *steal_time) 324 { 325 unsigned long now, stime; 326 struct cpu_accounting_data *acct = get_accounting(tsk); 327 328 WARN_ON_ONCE(!irqs_disabled()); 329 330 now = mftb(); 331 stime = now - acct->starttime; 332 acct->starttime = now; 333 334 *stime_scaled = vtime_delta_scaled(acct, now, stime); 335 336 *steal_time = calculate_stolen_time(now); 337 338 return stime; 339 } 340 341 void vtime_account_kernel(struct task_struct *tsk) 342 { 343 unsigned long stime, stime_scaled, steal_time; 344 struct cpu_accounting_data *acct = get_accounting(tsk); 345 346 stime = vtime_delta(tsk, &stime_scaled, &steal_time); 347 348 stime -= min(stime, steal_time); 349 acct->steal_time += steal_time; 350 351 if ((tsk->flags & PF_VCPU) && !irq_count()) { 352 acct->gtime += stime; 353 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 354 acct->utime_scaled += stime_scaled; 355 #endif 356 } else { 357 if (hardirq_count()) 358 acct->hardirq_time += stime; 359 else if (in_serving_softirq()) 360 acct->softirq_time += stime; 361 else 362 acct->stime += stime; 363 364 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 365 acct->stime_scaled += stime_scaled; 366 #endif 367 } 368 } 369 EXPORT_SYMBOL_GPL(vtime_account_kernel); 370 371 void vtime_account_idle(struct task_struct *tsk) 372 { 373 unsigned long stime, stime_scaled, steal_time; 374 struct cpu_accounting_data *acct = get_accounting(tsk); 375 376 stime = vtime_delta(tsk, &stime_scaled, &steal_time); 377 acct->idle_time += stime + steal_time; 378 } 379 380 static void vtime_flush_scaled(struct task_struct *tsk, 381 struct cpu_accounting_data *acct) 382 { 383 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 384 if (acct->utime_scaled) 385 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled); 386 if (acct->stime_scaled) 387 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled); 388 389 acct->utime_scaled = 0; 390 acct->utime_sspurr = 0; 391 acct->stime_scaled = 0; 392 #endif 393 } 394 395 /* 396 * Account the whole cputime accumulated in the paca 397 * Must be called with interrupts disabled. 398 * Assumes that vtime_account_kernel/idle() has been called 399 * recently (i.e. since the last entry from usermode) so that 400 * get_paca()->user_time_scaled is up to date. 401 */ 402 void vtime_flush(struct task_struct *tsk) 403 { 404 struct cpu_accounting_data *acct = get_accounting(tsk); 405 406 if (acct->utime) 407 account_user_time(tsk, cputime_to_nsecs(acct->utime)); 408 409 if (acct->gtime) 410 account_guest_time(tsk, cputime_to_nsecs(acct->gtime)); 411 412 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) { 413 account_steal_time(cputime_to_nsecs(acct->steal_time)); 414 acct->steal_time = 0; 415 } 416 417 if (acct->idle_time) 418 account_idle_time(cputime_to_nsecs(acct->idle_time)); 419 420 if (acct->stime) 421 account_system_index_time(tsk, cputime_to_nsecs(acct->stime), 422 CPUTIME_SYSTEM); 423 424 if (acct->hardirq_time) 425 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time), 426 CPUTIME_IRQ); 427 if (acct->softirq_time) 428 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time), 429 CPUTIME_SOFTIRQ); 430 431 vtime_flush_scaled(tsk, acct); 432 433 acct->utime = 0; 434 acct->gtime = 0; 435 acct->idle_time = 0; 436 acct->stime = 0; 437 acct->hardirq_time = 0; 438 acct->softirq_time = 0; 439 } 440 441 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ 442 #define calc_cputime_factors() 443 #endif 444 445 void __delay(unsigned long loops) 446 { 447 unsigned long start; 448 int diff; 449 450 spin_begin(); 451 if (__USE_RTC()) { 452 start = get_rtcl(); 453 do { 454 /* the RTCL register wraps at 1000000000 */ 455 diff = get_rtcl() - start; 456 if (diff < 0) 457 diff += 1000000000; 458 spin_cpu_relax(); 459 } while (diff < loops); 460 } else if (tb_invalid) { 461 /* 462 * TB is in error state and isn't ticking anymore. 463 * HMI handler was unable to recover from TB error. 464 * Return immediately, so that kernel won't get stuck here. 465 */ 466 spin_cpu_relax(); 467 } else { 468 start = get_tbl(); 469 while (get_tbl() - start < loops) 470 spin_cpu_relax(); 471 } 472 spin_end(); 473 } 474 EXPORT_SYMBOL(__delay); 475 476 void udelay(unsigned long usecs) 477 { 478 __delay(tb_ticks_per_usec * usecs); 479 } 480 EXPORT_SYMBOL(udelay); 481 482 #ifdef CONFIG_SMP 483 unsigned long profile_pc(struct pt_regs *regs) 484 { 485 unsigned long pc = instruction_pointer(regs); 486 487 if (in_lock_functions(pc)) 488 return regs->link; 489 490 return pc; 491 } 492 EXPORT_SYMBOL(profile_pc); 493 #endif 494 495 #ifdef CONFIG_IRQ_WORK 496 497 /* 498 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable... 499 */ 500 #ifdef CONFIG_PPC64 501 static inline unsigned long test_irq_work_pending(void) 502 { 503 unsigned long x; 504 505 asm volatile("lbz %0,%1(13)" 506 : "=r" (x) 507 : "i" (offsetof(struct paca_struct, irq_work_pending))); 508 return x; 509 } 510 511 static inline void set_irq_work_pending_flag(void) 512 { 513 asm volatile("stb %0,%1(13)" : : 514 "r" (1), 515 "i" (offsetof(struct paca_struct, irq_work_pending))); 516 } 517 518 static inline void clear_irq_work_pending(void) 519 { 520 asm volatile("stb %0,%1(13)" : : 521 "r" (0), 522 "i" (offsetof(struct paca_struct, irq_work_pending))); 523 } 524 525 #else /* 32-bit */ 526 527 DEFINE_PER_CPU(u8, irq_work_pending); 528 529 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1) 530 #define test_irq_work_pending() __this_cpu_read(irq_work_pending) 531 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0) 532 533 #endif /* 32 vs 64 bit */ 534 535 void arch_irq_work_raise(void) 536 { 537 /* 538 * 64-bit code that uses irq soft-mask can just cause an immediate 539 * interrupt here that gets soft masked, if this is called under 540 * local_irq_disable(). It might be possible to prevent that happening 541 * by noticing interrupts are disabled and setting decrementer pending 542 * to be replayed when irqs are enabled. The problem there is that 543 * tracing can call irq_work_raise, including in code that does low 544 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on) 545 * which could get tangled up if we're messing with the same state 546 * here. 547 */ 548 preempt_disable(); 549 set_irq_work_pending_flag(); 550 set_dec(1); 551 preempt_enable(); 552 } 553 554 #else /* CONFIG_IRQ_WORK */ 555 556 #define test_irq_work_pending() 0 557 #define clear_irq_work_pending() 558 559 #endif /* CONFIG_IRQ_WORK */ 560 561 /* 562 * timer_interrupt - gets called when the decrementer overflows, 563 * with interrupts disabled. 564 */ 565 void timer_interrupt(struct pt_regs *regs) 566 { 567 struct clock_event_device *evt = this_cpu_ptr(&decrementers); 568 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 569 struct pt_regs *old_regs; 570 u64 now; 571 572 /* Some implementations of hotplug will get timer interrupts while 573 * offline, just ignore these and we also need to set 574 * decrementers_next_tb as MAX to make sure __check_irq_replay 575 * don't replay timer interrupt when return, otherwise we'll trap 576 * here infinitely :( 577 */ 578 if (unlikely(!cpu_online(smp_processor_id()))) { 579 *next_tb = ~(u64)0; 580 set_dec(decrementer_max); 581 return; 582 } 583 584 /* Ensure a positive value is written to the decrementer, or else 585 * some CPUs will continue to take decrementer exceptions. When the 586 * PPC_WATCHDOG (decrementer based) is configured, keep this at most 587 * 31 bits, which is about 4 seconds on most systems, which gives 588 * the watchdog a chance of catching timer interrupt hard lockups. 589 */ 590 if (IS_ENABLED(CONFIG_PPC_WATCHDOG)) 591 set_dec(0x7fffffff); 592 else 593 set_dec(decrementer_max); 594 595 /* Conditionally hard-enable interrupts now that the DEC has been 596 * bumped to its maximum value 597 */ 598 may_hard_irq_enable(); 599 600 601 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC) 602 if (atomic_read(&ppc_n_lost_interrupts) != 0) 603 do_IRQ(regs); 604 #endif 605 606 old_regs = set_irq_regs(regs); 607 irq_enter(); 608 trace_timer_interrupt_entry(regs); 609 610 if (test_irq_work_pending()) { 611 clear_irq_work_pending(); 612 irq_work_run(); 613 } 614 615 now = get_tb_or_rtc(); 616 if (now >= *next_tb) { 617 *next_tb = ~(u64)0; 618 if (evt->event_handler) 619 evt->event_handler(evt); 620 __this_cpu_inc(irq_stat.timer_irqs_event); 621 } else { 622 now = *next_tb - now; 623 if (now <= decrementer_max) 624 set_dec(now); 625 /* We may have raced with new irq work */ 626 if (test_irq_work_pending()) 627 set_dec(1); 628 __this_cpu_inc(irq_stat.timer_irqs_others); 629 } 630 631 trace_timer_interrupt_exit(regs); 632 irq_exit(); 633 set_irq_regs(old_regs); 634 } 635 EXPORT_SYMBOL(timer_interrupt); 636 637 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST 638 void timer_broadcast_interrupt(void) 639 { 640 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 641 642 *next_tb = ~(u64)0; 643 tick_receive_broadcast(); 644 __this_cpu_inc(irq_stat.broadcast_irqs_event); 645 } 646 #endif 647 648 #ifdef CONFIG_SUSPEND 649 static void generic_suspend_disable_irqs(void) 650 { 651 /* Disable the decrementer, so that it doesn't interfere 652 * with suspending. 653 */ 654 655 set_dec(decrementer_max); 656 local_irq_disable(); 657 set_dec(decrementer_max); 658 } 659 660 static void generic_suspend_enable_irqs(void) 661 { 662 local_irq_enable(); 663 } 664 665 /* Overrides the weak version in kernel/power/main.c */ 666 void arch_suspend_disable_irqs(void) 667 { 668 if (ppc_md.suspend_disable_irqs) 669 ppc_md.suspend_disable_irqs(); 670 generic_suspend_disable_irqs(); 671 } 672 673 /* Overrides the weak version in kernel/power/main.c */ 674 void arch_suspend_enable_irqs(void) 675 { 676 generic_suspend_enable_irqs(); 677 if (ppc_md.suspend_enable_irqs) 678 ppc_md.suspend_enable_irqs(); 679 } 680 #endif 681 682 unsigned long long tb_to_ns(unsigned long long ticks) 683 { 684 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift; 685 } 686 EXPORT_SYMBOL_GPL(tb_to_ns); 687 688 /* 689 * Scheduler clock - returns current time in nanosec units. 690 * 691 * Note: mulhdu(a, b) (multiply high double unsigned) returns 692 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 693 * are 64-bit unsigned numbers. 694 */ 695 notrace unsigned long long sched_clock(void) 696 { 697 if (__USE_RTC()) 698 return get_rtc(); 699 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 700 } 701 702 703 #ifdef CONFIG_PPC_PSERIES 704 705 /* 706 * Running clock - attempts to give a view of time passing for a virtualised 707 * kernels. 708 * Uses the VTB register if available otherwise a next best guess. 709 */ 710 unsigned long long running_clock(void) 711 { 712 /* 713 * Don't read the VTB as a host since KVM does not switch in host 714 * timebase into the VTB when it takes a guest off the CPU, reading the 715 * VTB would result in reading 'last switched out' guest VTB. 716 * 717 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it 718 * would be unsafe to rely only on the #ifdef above. 719 */ 720 if (firmware_has_feature(FW_FEATURE_LPAR) && 721 cpu_has_feature(CPU_FTR_ARCH_207S)) 722 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 723 724 /* 725 * This is a next best approximation without a VTB. 726 * On a host which is running bare metal there should never be any stolen 727 * time and on a host which doesn't do any virtualisation TB *should* equal 728 * VTB so it makes no difference anyway. 729 */ 730 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL]; 731 } 732 #endif 733 734 static int __init get_freq(char *name, int cells, unsigned long *val) 735 { 736 struct device_node *cpu; 737 const __be32 *fp; 738 int found = 0; 739 740 /* The cpu node should have timebase and clock frequency properties */ 741 cpu = of_find_node_by_type(NULL, "cpu"); 742 743 if (cpu) { 744 fp = of_get_property(cpu, name, NULL); 745 if (fp) { 746 found = 1; 747 *val = of_read_ulong(fp, cells); 748 } 749 750 of_node_put(cpu); 751 } 752 753 return found; 754 } 755 756 static void start_cpu_decrementer(void) 757 { 758 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 759 unsigned int tcr; 760 761 /* Clear any pending timer interrupts */ 762 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 763 764 tcr = mfspr(SPRN_TCR); 765 /* 766 * The watchdog may have already been enabled by u-boot. So leave 767 * TRC[WP] (Watchdog Period) alone. 768 */ 769 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */ 770 tcr |= TCR_DIE; /* Enable decrementer */ 771 mtspr(SPRN_TCR, tcr); 772 #endif 773 } 774 775 void __init generic_calibrate_decr(void) 776 { 777 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 778 779 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 780 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 781 782 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 783 "(not found)\n"); 784 } 785 786 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 787 788 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 789 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 790 791 printk(KERN_ERR "WARNING: Estimating processor frequency " 792 "(not found)\n"); 793 } 794 } 795 796 int update_persistent_clock64(struct timespec64 now) 797 { 798 struct rtc_time tm; 799 800 if (!ppc_md.set_rtc_time) 801 return -ENODEV; 802 803 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm); 804 805 return ppc_md.set_rtc_time(&tm); 806 } 807 808 static void __read_persistent_clock(struct timespec64 *ts) 809 { 810 struct rtc_time tm; 811 static int first = 1; 812 813 ts->tv_nsec = 0; 814 /* XXX this is a litle fragile but will work okay in the short term */ 815 if (first) { 816 first = 0; 817 if (ppc_md.time_init) 818 timezone_offset = ppc_md.time_init(); 819 820 /* get_boot_time() isn't guaranteed to be safe to call late */ 821 if (ppc_md.get_boot_time) { 822 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset; 823 return; 824 } 825 } 826 if (!ppc_md.get_rtc_time) { 827 ts->tv_sec = 0; 828 return; 829 } 830 ppc_md.get_rtc_time(&tm); 831 832 ts->tv_sec = rtc_tm_to_time64(&tm); 833 } 834 835 void read_persistent_clock64(struct timespec64 *ts) 836 { 837 __read_persistent_clock(ts); 838 839 /* Sanitize it in case real time clock is set below EPOCH */ 840 if (ts->tv_sec < 0) { 841 ts->tv_sec = 0; 842 ts->tv_nsec = 0; 843 } 844 845 } 846 847 /* clocksource code */ 848 static notrace u64 rtc_read(struct clocksource *cs) 849 { 850 return (u64)get_rtc(); 851 } 852 853 static notrace u64 timebase_read(struct clocksource *cs) 854 { 855 return (u64)get_tb(); 856 } 857 858 859 void update_vsyscall(struct timekeeper *tk) 860 { 861 struct timespec64 xt; 862 struct clocksource *clock = tk->tkr_mono.clock; 863 u32 mult = tk->tkr_mono.mult; 864 u32 shift = tk->tkr_mono.shift; 865 u64 cycle_last = tk->tkr_mono.cycle_last; 866 u64 new_tb_to_xs, new_stamp_xsec; 867 u64 frac_sec; 868 869 if (clock != &clocksource_timebase) 870 return; 871 872 xt.tv_sec = tk->xtime_sec; 873 xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); 874 875 /* Make userspace gettimeofday spin until we're done. */ 876 ++vdso_data->tb_update_count; 877 smp_mb(); 878 879 /* 880 * This computes ((2^20 / 1e9) * mult) >> shift as a 881 * 0.64 fixed-point fraction. 882 * The computation in the else clause below won't overflow 883 * (as long as the timebase frequency is >= 1.049 MHz) 884 * but loses precision because we lose the low bits of the constant 885 * in the shift. Note that 19342813113834067 ~= 2^(20+64) / 1e9. 886 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns 887 * over a second. (Shift values are usually 22, 23 or 24.) 888 * For high frequency clocks such as the 512MHz timebase clock 889 * on POWER[6789], the mult value is small (e.g. 32768000) 890 * and so we can shift the constant by 16 initially 891 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the 892 * remaining shifts after the multiplication, which gives a 893 * more accurate result (e.g. with mult = 32768000, shift = 24, 894 * the error is only about 1.2e-12, or 0.7ns over 10 minutes). 895 */ 896 if (mult <= 62500000 && clock->shift >= 16) 897 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16); 898 else 899 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift); 900 901 /* 902 * Compute the fractional second in units of 2^-32 seconds. 903 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift 904 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives 905 * it in units of 2^-32 seconds. 906 * We assume shift <= 32 because clocks_calc_mult_shift() 907 * generates shift values in the range 0 - 32. 908 */ 909 frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift); 910 do_div(frac_sec, NSEC_PER_SEC); 911 912 /* 913 * Work out new stamp_xsec value for any legacy users of systemcfg. 914 * stamp_xsec is in units of 2^-20 seconds. 915 */ 916 new_stamp_xsec = frac_sec >> 12; 917 new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC; 918 919 /* 920 * tb_update_count is used to allow the userspace gettimeofday code 921 * to assure itself that it sees a consistent view of the tb_to_xs and 922 * stamp_xsec variables. It reads the tb_update_count, then reads 923 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 924 * the two values of tb_update_count match and are even then the 925 * tb_to_xs and stamp_xsec values are consistent. If not, then it 926 * loops back and reads them again until this criteria is met. 927 */ 928 vdso_data->tb_orig_stamp = cycle_last; 929 vdso_data->stamp_xsec = new_stamp_xsec; 930 vdso_data->tb_to_xs = new_tb_to_xs; 931 vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec; 932 vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec; 933 vdso_data->stamp_xtime_sec = xt.tv_sec; 934 vdso_data->stamp_xtime_nsec = xt.tv_nsec; 935 vdso_data->stamp_sec_fraction = frac_sec; 936 vdso_data->hrtimer_res = hrtimer_resolution; 937 smp_wmb(); 938 ++(vdso_data->tb_update_count); 939 } 940 941 void update_vsyscall_tz(void) 942 { 943 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 944 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 945 } 946 947 static void __init clocksource_init(void) 948 { 949 struct clocksource *clock; 950 951 if (__USE_RTC()) 952 clock = &clocksource_rtc; 953 else 954 clock = &clocksource_timebase; 955 956 if (clocksource_register_hz(clock, tb_ticks_per_sec)) { 957 printk(KERN_ERR "clocksource: %s is already registered\n", 958 clock->name); 959 return; 960 } 961 962 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 963 clock->name, clock->mult, clock->shift); 964 } 965 966 static int decrementer_set_next_event(unsigned long evt, 967 struct clock_event_device *dev) 968 { 969 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt); 970 set_dec(evt); 971 972 /* We may have raced with new irq work */ 973 if (test_irq_work_pending()) 974 set_dec(1); 975 976 return 0; 977 } 978 979 static int decrementer_shutdown(struct clock_event_device *dev) 980 { 981 decrementer_set_next_event(decrementer_max, dev); 982 return 0; 983 } 984 985 static void register_decrementer_clockevent(int cpu) 986 { 987 struct clock_event_device *dec = &per_cpu(decrementers, cpu); 988 989 *dec = decrementer_clockevent; 990 dec->cpumask = cpumask_of(cpu); 991 992 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max); 993 994 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n", 995 dec->name, dec->mult, dec->shift, cpu); 996 997 /* Set values for KVM, see kvm_emulate_dec() */ 998 decrementer_clockevent.mult = dec->mult; 999 decrementer_clockevent.shift = dec->shift; 1000 } 1001 1002 static void enable_large_decrementer(void) 1003 { 1004 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 1005 return; 1006 1007 if (decrementer_max <= DECREMENTER_DEFAULT_MAX) 1008 return; 1009 1010 /* 1011 * If we're running as the hypervisor we need to enable the LD manually 1012 * otherwise firmware should have done it for us. 1013 */ 1014 if (cpu_has_feature(CPU_FTR_HVMODE)) 1015 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD); 1016 } 1017 1018 static void __init set_decrementer_max(void) 1019 { 1020 struct device_node *cpu; 1021 u32 bits = 32; 1022 1023 /* Prior to ISAv3 the decrementer is always 32 bit */ 1024 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 1025 return; 1026 1027 cpu = of_find_node_by_type(NULL, "cpu"); 1028 1029 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) { 1030 if (bits > 64 || bits < 32) { 1031 pr_warn("time_init: firmware supplied invalid ibm,dec-bits"); 1032 bits = 32; 1033 } 1034 1035 /* calculate the signed maximum given this many bits */ 1036 decrementer_max = (1ul << (bits - 1)) - 1; 1037 } 1038 1039 of_node_put(cpu); 1040 1041 pr_info("time_init: %u bit decrementer (max: %llx)\n", 1042 bits, decrementer_max); 1043 } 1044 1045 static void __init init_decrementer_clockevent(void) 1046 { 1047 register_decrementer_clockevent(smp_processor_id()); 1048 } 1049 1050 void secondary_cpu_time_init(void) 1051 { 1052 /* Enable and test the large decrementer for this cpu */ 1053 enable_large_decrementer(); 1054 1055 /* Start the decrementer on CPUs that have manual control 1056 * such as BookE 1057 */ 1058 start_cpu_decrementer(); 1059 1060 /* FIME: Should make unrelatred change to move snapshot_timebase 1061 * call here ! */ 1062 register_decrementer_clockevent(smp_processor_id()); 1063 } 1064 1065 /* This function is only called on the boot processor */ 1066 void __init time_init(void) 1067 { 1068 struct div_result res; 1069 u64 scale; 1070 unsigned shift; 1071 1072 if (__USE_RTC()) { 1073 /* 601 processor: dec counts down by 128 every 128ns */ 1074 ppc_tb_freq = 1000000000; 1075 } else { 1076 /* Normal PowerPC with timebase register */ 1077 ppc_md.calibrate_decr(); 1078 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 1079 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 1080 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 1081 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 1082 } 1083 1084 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 1085 tb_ticks_per_sec = ppc_tb_freq; 1086 tb_ticks_per_usec = ppc_tb_freq / 1000000; 1087 calc_cputime_factors(); 1088 1089 /* 1090 * Compute scale factor for sched_clock. 1091 * The calibrate_decr() function has set tb_ticks_per_sec, 1092 * which is the timebase frequency. 1093 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 1094 * the 128-bit result as a 64.64 fixed-point number. 1095 * We then shift that number right until it is less than 1.0, 1096 * giving us the scale factor and shift count to use in 1097 * sched_clock(). 1098 */ 1099 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 1100 scale = res.result_low; 1101 for (shift = 0; res.result_high != 0; ++shift) { 1102 scale = (scale >> 1) | (res.result_high << 63); 1103 res.result_high >>= 1; 1104 } 1105 tb_to_ns_scale = scale; 1106 tb_to_ns_shift = shift; 1107 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 1108 boot_tb = get_tb_or_rtc(); 1109 1110 /* If platform provided a timezone (pmac), we correct the time */ 1111 if (timezone_offset) { 1112 sys_tz.tz_minuteswest = -timezone_offset / 60; 1113 sys_tz.tz_dsttime = 0; 1114 } 1115 1116 vdso_data->tb_update_count = 0; 1117 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 1118 1119 /* initialise and enable the large decrementer (if we have one) */ 1120 set_decrementer_max(); 1121 enable_large_decrementer(); 1122 1123 /* Start the decrementer on CPUs that have manual control 1124 * such as BookE 1125 */ 1126 start_cpu_decrementer(); 1127 1128 /* Register the clocksource */ 1129 clocksource_init(); 1130 1131 init_decrementer_clockevent(); 1132 tick_setup_hrtimer_broadcast(); 1133 1134 of_clk_init(NULL); 1135 } 1136 1137 /* 1138 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1139 * result. 1140 */ 1141 void div128_by_32(u64 dividend_high, u64 dividend_low, 1142 unsigned divisor, struct div_result *dr) 1143 { 1144 unsigned long a, b, c, d; 1145 unsigned long w, x, y, z; 1146 u64 ra, rb, rc; 1147 1148 a = dividend_high >> 32; 1149 b = dividend_high & 0xffffffff; 1150 c = dividend_low >> 32; 1151 d = dividend_low & 0xffffffff; 1152 1153 w = a / divisor; 1154 ra = ((u64)(a - (w * divisor)) << 32) + b; 1155 1156 rb = ((u64) do_div(ra, divisor) << 32) + c; 1157 x = ra; 1158 1159 rc = ((u64) do_div(rb, divisor) << 32) + d; 1160 y = rb; 1161 1162 do_div(rc, divisor); 1163 z = rc; 1164 1165 dr->result_high = ((u64)w << 32) + x; 1166 dr->result_low = ((u64)y << 32) + z; 1167 1168 } 1169 1170 /* We don't need to calibrate delay, we use the CPU timebase for that */ 1171 void calibrate_delay(void) 1172 { 1173 /* Some generic code (such as spinlock debug) use loops_per_jiffy 1174 * as the number of __delay(1) in a jiffy, so make it so 1175 */ 1176 loops_per_jiffy = tb_ticks_per_jiffy; 1177 } 1178 1179 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC) 1180 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm) 1181 { 1182 ppc_md.get_rtc_time(tm); 1183 return 0; 1184 } 1185 1186 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm) 1187 { 1188 if (!ppc_md.set_rtc_time) 1189 return -EOPNOTSUPP; 1190 1191 if (ppc_md.set_rtc_time(tm) < 0) 1192 return -EOPNOTSUPP; 1193 1194 return 0; 1195 } 1196 1197 static const struct rtc_class_ops rtc_generic_ops = { 1198 .read_time = rtc_generic_get_time, 1199 .set_time = rtc_generic_set_time, 1200 }; 1201 1202 static int __init rtc_init(void) 1203 { 1204 struct platform_device *pdev; 1205 1206 if (!ppc_md.get_rtc_time) 1207 return -ENODEV; 1208 1209 pdev = platform_device_register_data(NULL, "rtc-generic", -1, 1210 &rtc_generic_ops, 1211 sizeof(rtc_generic_ops)); 1212 1213 return PTR_ERR_OR_ZERO(pdev); 1214 } 1215 1216 device_initcall(rtc_init); 1217 #endif 1218