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 #include <asm/dtl.h> 187 188 /* 189 * Scan the dispatch trace log and count up the stolen time. 190 * Should be called with interrupts disabled. 191 */ 192 static u64 scan_dispatch_log(u64 stop_tb) 193 { 194 u64 i = local_paca->dtl_ridx; 195 struct dtl_entry *dtl = local_paca->dtl_curr; 196 struct dtl_entry *dtl_end = local_paca->dispatch_log_end; 197 struct lppaca *vpa = local_paca->lppaca_ptr; 198 u64 tb_delta; 199 u64 stolen = 0; 200 u64 dtb; 201 202 if (!dtl) 203 return 0; 204 205 if (i == be64_to_cpu(vpa->dtl_idx)) 206 return 0; 207 while (i < be64_to_cpu(vpa->dtl_idx)) { 208 dtb = be64_to_cpu(dtl->timebase); 209 tb_delta = be32_to_cpu(dtl->enqueue_to_dispatch_time) + 210 be32_to_cpu(dtl->ready_to_enqueue_time); 211 barrier(); 212 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) { 213 /* buffer has overflowed */ 214 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG; 215 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG); 216 continue; 217 } 218 if (dtb > stop_tb) 219 break; 220 if (dtl_consumer) 221 dtl_consumer(dtl, i); 222 stolen += tb_delta; 223 ++i; 224 ++dtl; 225 if (dtl == dtl_end) 226 dtl = local_paca->dispatch_log; 227 } 228 local_paca->dtl_ridx = i; 229 local_paca->dtl_curr = dtl; 230 return stolen; 231 } 232 233 /* 234 * Accumulate stolen time by scanning the dispatch trace log. 235 * Called on entry from user mode. 236 */ 237 void notrace accumulate_stolen_time(void) 238 { 239 u64 sst, ust; 240 unsigned long save_irq_soft_mask = irq_soft_mask_return(); 241 struct cpu_accounting_data *acct = &local_paca->accounting; 242 243 /* We are called early in the exception entry, before 244 * soft/hard_enabled are sync'ed to the expected state 245 * for the exception. We are hard disabled but the PACA 246 * needs to reflect that so various debug stuff doesn't 247 * complain 248 */ 249 irq_soft_mask_set(IRQS_DISABLED); 250 251 sst = scan_dispatch_log(acct->starttime_user); 252 ust = scan_dispatch_log(acct->starttime); 253 acct->stime -= sst; 254 acct->utime -= ust; 255 acct->steal_time += ust + sst; 256 257 irq_soft_mask_set(save_irq_soft_mask); 258 } 259 260 static inline u64 calculate_stolen_time(u64 stop_tb) 261 { 262 if (!firmware_has_feature(FW_FEATURE_SPLPAR)) 263 return 0; 264 265 if (get_paca()->dtl_ridx != be64_to_cpu(get_lppaca()->dtl_idx)) 266 return scan_dispatch_log(stop_tb); 267 268 return 0; 269 } 270 271 #else /* CONFIG_PPC_SPLPAR */ 272 static inline u64 calculate_stolen_time(u64 stop_tb) 273 { 274 return 0; 275 } 276 277 #endif /* CONFIG_PPC_SPLPAR */ 278 279 /* 280 * Account time for a transition between system, hard irq 281 * or soft irq state. 282 */ 283 static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct, 284 unsigned long now, unsigned long stime) 285 { 286 unsigned long stime_scaled = 0; 287 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 288 unsigned long nowscaled, deltascaled; 289 unsigned long utime, utime_scaled; 290 291 nowscaled = read_spurr(now); 292 deltascaled = nowscaled - acct->startspurr; 293 acct->startspurr = nowscaled; 294 utime = acct->utime - acct->utime_sspurr; 295 acct->utime_sspurr = acct->utime; 296 297 /* 298 * Because we don't read the SPURR on every kernel entry/exit, 299 * deltascaled includes both user and system SPURR ticks. 300 * Apportion these ticks to system SPURR ticks and user 301 * SPURR ticks in the same ratio as the system time (delta) 302 * and user time (udelta) values obtained from the timebase 303 * over the same interval. The system ticks get accounted here; 304 * the user ticks get saved up in paca->user_time_scaled to be 305 * used by account_process_tick. 306 */ 307 stime_scaled = stime; 308 utime_scaled = utime; 309 if (deltascaled != stime + utime) { 310 if (utime) { 311 stime_scaled = deltascaled * stime / (stime + utime); 312 utime_scaled = deltascaled - stime_scaled; 313 } else { 314 stime_scaled = deltascaled; 315 } 316 } 317 acct->utime_scaled += utime_scaled; 318 #endif 319 320 return stime_scaled; 321 } 322 323 static unsigned long vtime_delta(struct task_struct *tsk, 324 unsigned long *stime_scaled, 325 unsigned long *steal_time) 326 { 327 unsigned long now, stime; 328 struct cpu_accounting_data *acct = get_accounting(tsk); 329 330 WARN_ON_ONCE(!irqs_disabled()); 331 332 now = mftb(); 333 stime = now - acct->starttime; 334 acct->starttime = now; 335 336 *stime_scaled = vtime_delta_scaled(acct, now, stime); 337 338 *steal_time = calculate_stolen_time(now); 339 340 return stime; 341 } 342 343 void vtime_account_kernel(struct task_struct *tsk) 344 { 345 unsigned long stime, stime_scaled, steal_time; 346 struct cpu_accounting_data *acct = get_accounting(tsk); 347 348 stime = vtime_delta(tsk, &stime_scaled, &steal_time); 349 350 stime -= min(stime, steal_time); 351 acct->steal_time += steal_time; 352 353 if ((tsk->flags & PF_VCPU) && !irq_count()) { 354 acct->gtime += stime; 355 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 356 acct->utime_scaled += stime_scaled; 357 #endif 358 } else { 359 if (hardirq_count()) 360 acct->hardirq_time += stime; 361 else if (in_serving_softirq()) 362 acct->softirq_time += stime; 363 else 364 acct->stime += stime; 365 366 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 367 acct->stime_scaled += stime_scaled; 368 #endif 369 } 370 } 371 EXPORT_SYMBOL_GPL(vtime_account_kernel); 372 373 void vtime_account_idle(struct task_struct *tsk) 374 { 375 unsigned long stime, stime_scaled, steal_time; 376 struct cpu_accounting_data *acct = get_accounting(tsk); 377 378 stime = vtime_delta(tsk, &stime_scaled, &steal_time); 379 acct->idle_time += stime + steal_time; 380 } 381 382 static void vtime_flush_scaled(struct task_struct *tsk, 383 struct cpu_accounting_data *acct) 384 { 385 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME 386 if (acct->utime_scaled) 387 tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled); 388 if (acct->stime_scaled) 389 tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled); 390 391 acct->utime_scaled = 0; 392 acct->utime_sspurr = 0; 393 acct->stime_scaled = 0; 394 #endif 395 } 396 397 /* 398 * Account the whole cputime accumulated in the paca 399 * Must be called with interrupts disabled. 400 * Assumes that vtime_account_kernel/idle() has been called 401 * recently (i.e. since the last entry from usermode) so that 402 * get_paca()->user_time_scaled is up to date. 403 */ 404 void vtime_flush(struct task_struct *tsk) 405 { 406 struct cpu_accounting_data *acct = get_accounting(tsk); 407 408 if (acct->utime) 409 account_user_time(tsk, cputime_to_nsecs(acct->utime)); 410 411 if (acct->gtime) 412 account_guest_time(tsk, cputime_to_nsecs(acct->gtime)); 413 414 if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) { 415 account_steal_time(cputime_to_nsecs(acct->steal_time)); 416 acct->steal_time = 0; 417 } 418 419 if (acct->idle_time) 420 account_idle_time(cputime_to_nsecs(acct->idle_time)); 421 422 if (acct->stime) 423 account_system_index_time(tsk, cputime_to_nsecs(acct->stime), 424 CPUTIME_SYSTEM); 425 426 if (acct->hardirq_time) 427 account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time), 428 CPUTIME_IRQ); 429 if (acct->softirq_time) 430 account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time), 431 CPUTIME_SOFTIRQ); 432 433 vtime_flush_scaled(tsk, acct); 434 435 acct->utime = 0; 436 acct->gtime = 0; 437 acct->idle_time = 0; 438 acct->stime = 0; 439 acct->hardirq_time = 0; 440 acct->softirq_time = 0; 441 } 442 443 #else /* ! CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ 444 #define calc_cputime_factors() 445 #endif 446 447 void __delay(unsigned long loops) 448 { 449 unsigned long start; 450 int diff; 451 452 spin_begin(); 453 if (__USE_RTC()) { 454 start = get_rtcl(); 455 do { 456 /* the RTCL register wraps at 1000000000 */ 457 diff = get_rtcl() - start; 458 if (diff < 0) 459 diff += 1000000000; 460 spin_cpu_relax(); 461 } while (diff < loops); 462 } else if (tb_invalid) { 463 /* 464 * TB is in error state and isn't ticking anymore. 465 * HMI handler was unable to recover from TB error. 466 * Return immediately, so that kernel won't get stuck here. 467 */ 468 spin_cpu_relax(); 469 } else { 470 start = get_tbl(); 471 while (get_tbl() - start < loops) 472 spin_cpu_relax(); 473 } 474 spin_end(); 475 } 476 EXPORT_SYMBOL(__delay); 477 478 void udelay(unsigned long usecs) 479 { 480 __delay(tb_ticks_per_usec * usecs); 481 } 482 EXPORT_SYMBOL(udelay); 483 484 #ifdef CONFIG_SMP 485 unsigned long profile_pc(struct pt_regs *regs) 486 { 487 unsigned long pc = instruction_pointer(regs); 488 489 if (in_lock_functions(pc)) 490 return regs->link; 491 492 return pc; 493 } 494 EXPORT_SYMBOL(profile_pc); 495 #endif 496 497 #ifdef CONFIG_IRQ_WORK 498 499 /* 500 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable... 501 */ 502 #ifdef CONFIG_PPC64 503 static inline unsigned long test_irq_work_pending(void) 504 { 505 unsigned long x; 506 507 asm volatile("lbz %0,%1(13)" 508 : "=r" (x) 509 : "i" (offsetof(struct paca_struct, irq_work_pending))); 510 return x; 511 } 512 513 static inline void set_irq_work_pending_flag(void) 514 { 515 asm volatile("stb %0,%1(13)" : : 516 "r" (1), 517 "i" (offsetof(struct paca_struct, irq_work_pending))); 518 } 519 520 static inline void clear_irq_work_pending(void) 521 { 522 asm volatile("stb %0,%1(13)" : : 523 "r" (0), 524 "i" (offsetof(struct paca_struct, irq_work_pending))); 525 } 526 527 #else /* 32-bit */ 528 529 DEFINE_PER_CPU(u8, irq_work_pending); 530 531 #define set_irq_work_pending_flag() __this_cpu_write(irq_work_pending, 1) 532 #define test_irq_work_pending() __this_cpu_read(irq_work_pending) 533 #define clear_irq_work_pending() __this_cpu_write(irq_work_pending, 0) 534 535 #endif /* 32 vs 64 bit */ 536 537 void arch_irq_work_raise(void) 538 { 539 /* 540 * 64-bit code that uses irq soft-mask can just cause an immediate 541 * interrupt here that gets soft masked, if this is called under 542 * local_irq_disable(). It might be possible to prevent that happening 543 * by noticing interrupts are disabled and setting decrementer pending 544 * to be replayed when irqs are enabled. The problem there is that 545 * tracing can call irq_work_raise, including in code that does low 546 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on) 547 * which could get tangled up if we're messing with the same state 548 * here. 549 */ 550 preempt_disable(); 551 set_irq_work_pending_flag(); 552 set_dec(1); 553 preempt_enable(); 554 } 555 556 #else /* CONFIG_IRQ_WORK */ 557 558 #define test_irq_work_pending() 0 559 #define clear_irq_work_pending() 560 561 #endif /* CONFIG_IRQ_WORK */ 562 563 /* 564 * timer_interrupt - gets called when the decrementer overflows, 565 * with interrupts disabled. 566 */ 567 void timer_interrupt(struct pt_regs *regs) 568 { 569 struct clock_event_device *evt = this_cpu_ptr(&decrementers); 570 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 571 struct pt_regs *old_regs; 572 u64 now; 573 574 /* Some implementations of hotplug will get timer interrupts while 575 * offline, just ignore these and we also need to set 576 * decrementers_next_tb as MAX to make sure __check_irq_replay 577 * don't replay timer interrupt when return, otherwise we'll trap 578 * here infinitely :( 579 */ 580 if (unlikely(!cpu_online(smp_processor_id()))) { 581 *next_tb = ~(u64)0; 582 set_dec(decrementer_max); 583 return; 584 } 585 586 /* Ensure a positive value is written to the decrementer, or else 587 * some CPUs will continue to take decrementer exceptions. When the 588 * PPC_WATCHDOG (decrementer based) is configured, keep this at most 589 * 31 bits, which is about 4 seconds on most systems, which gives 590 * the watchdog a chance of catching timer interrupt hard lockups. 591 */ 592 if (IS_ENABLED(CONFIG_PPC_WATCHDOG)) 593 set_dec(0x7fffffff); 594 else 595 set_dec(decrementer_max); 596 597 /* Conditionally hard-enable interrupts now that the DEC has been 598 * bumped to its maximum value 599 */ 600 may_hard_irq_enable(); 601 602 603 #if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC) 604 if (atomic_read(&ppc_n_lost_interrupts) != 0) 605 do_IRQ(regs); 606 #endif 607 608 old_regs = set_irq_regs(regs); 609 irq_enter(); 610 trace_timer_interrupt_entry(regs); 611 612 if (test_irq_work_pending()) { 613 clear_irq_work_pending(); 614 irq_work_run(); 615 } 616 617 now = get_tb_or_rtc(); 618 if (now >= *next_tb) { 619 *next_tb = ~(u64)0; 620 if (evt->event_handler) 621 evt->event_handler(evt); 622 __this_cpu_inc(irq_stat.timer_irqs_event); 623 } else { 624 now = *next_tb - now; 625 if (now <= decrementer_max) 626 set_dec(now); 627 /* We may have raced with new irq work */ 628 if (test_irq_work_pending()) 629 set_dec(1); 630 __this_cpu_inc(irq_stat.timer_irqs_others); 631 } 632 633 trace_timer_interrupt_exit(regs); 634 irq_exit(); 635 set_irq_regs(old_regs); 636 } 637 EXPORT_SYMBOL(timer_interrupt); 638 639 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST 640 void timer_broadcast_interrupt(void) 641 { 642 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb); 643 644 *next_tb = ~(u64)0; 645 tick_receive_broadcast(); 646 __this_cpu_inc(irq_stat.broadcast_irqs_event); 647 } 648 #endif 649 650 #ifdef CONFIG_SUSPEND 651 static void generic_suspend_disable_irqs(void) 652 { 653 /* Disable the decrementer, so that it doesn't interfere 654 * with suspending. 655 */ 656 657 set_dec(decrementer_max); 658 local_irq_disable(); 659 set_dec(decrementer_max); 660 } 661 662 static void generic_suspend_enable_irqs(void) 663 { 664 local_irq_enable(); 665 } 666 667 /* Overrides the weak version in kernel/power/main.c */ 668 void arch_suspend_disable_irqs(void) 669 { 670 if (ppc_md.suspend_disable_irqs) 671 ppc_md.suspend_disable_irqs(); 672 generic_suspend_disable_irqs(); 673 } 674 675 /* Overrides the weak version in kernel/power/main.c */ 676 void arch_suspend_enable_irqs(void) 677 { 678 generic_suspend_enable_irqs(); 679 if (ppc_md.suspend_enable_irqs) 680 ppc_md.suspend_enable_irqs(); 681 } 682 #endif 683 684 unsigned long long tb_to_ns(unsigned long long ticks) 685 { 686 return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift; 687 } 688 EXPORT_SYMBOL_GPL(tb_to_ns); 689 690 /* 691 * Scheduler clock - returns current time in nanosec units. 692 * 693 * Note: mulhdu(a, b) (multiply high double unsigned) returns 694 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b 695 * are 64-bit unsigned numbers. 696 */ 697 notrace unsigned long long sched_clock(void) 698 { 699 if (__USE_RTC()) 700 return get_rtc(); 701 return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 702 } 703 704 705 #ifdef CONFIG_PPC_PSERIES 706 707 /* 708 * Running clock - attempts to give a view of time passing for a virtualised 709 * kernels. 710 * Uses the VTB register if available otherwise a next best guess. 711 */ 712 unsigned long long running_clock(void) 713 { 714 /* 715 * Don't read the VTB as a host since KVM does not switch in host 716 * timebase into the VTB when it takes a guest off the CPU, reading the 717 * VTB would result in reading 'last switched out' guest VTB. 718 * 719 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it 720 * would be unsafe to rely only on the #ifdef above. 721 */ 722 if (firmware_has_feature(FW_FEATURE_LPAR) && 723 cpu_has_feature(CPU_FTR_ARCH_207S)) 724 return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift; 725 726 /* 727 * This is a next best approximation without a VTB. 728 * On a host which is running bare metal there should never be any stolen 729 * time and on a host which doesn't do any virtualisation TB *should* equal 730 * VTB so it makes no difference anyway. 731 */ 732 return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL]; 733 } 734 #endif 735 736 static int __init get_freq(char *name, int cells, unsigned long *val) 737 { 738 struct device_node *cpu; 739 const __be32 *fp; 740 int found = 0; 741 742 /* The cpu node should have timebase and clock frequency properties */ 743 cpu = of_find_node_by_type(NULL, "cpu"); 744 745 if (cpu) { 746 fp = of_get_property(cpu, name, NULL); 747 if (fp) { 748 found = 1; 749 *val = of_read_ulong(fp, cells); 750 } 751 752 of_node_put(cpu); 753 } 754 755 return found; 756 } 757 758 static void start_cpu_decrementer(void) 759 { 760 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x) 761 unsigned int tcr; 762 763 /* Clear any pending timer interrupts */ 764 mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS); 765 766 tcr = mfspr(SPRN_TCR); 767 /* 768 * The watchdog may have already been enabled by u-boot. So leave 769 * TRC[WP] (Watchdog Period) alone. 770 */ 771 tcr &= TCR_WP_MASK; /* Clear all bits except for TCR[WP] */ 772 tcr |= TCR_DIE; /* Enable decrementer */ 773 mtspr(SPRN_TCR, tcr); 774 #endif 775 } 776 777 void __init generic_calibrate_decr(void) 778 { 779 ppc_tb_freq = DEFAULT_TB_FREQ; /* hardcoded default */ 780 781 if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) && 782 !get_freq("timebase-frequency", 1, &ppc_tb_freq)) { 783 784 printk(KERN_ERR "WARNING: Estimating decrementer frequency " 785 "(not found)\n"); 786 } 787 788 ppc_proc_freq = DEFAULT_PROC_FREQ; /* hardcoded default */ 789 790 if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) && 791 !get_freq("clock-frequency", 1, &ppc_proc_freq)) { 792 793 printk(KERN_ERR "WARNING: Estimating processor frequency " 794 "(not found)\n"); 795 } 796 } 797 798 int update_persistent_clock64(struct timespec64 now) 799 { 800 struct rtc_time tm; 801 802 if (!ppc_md.set_rtc_time) 803 return -ENODEV; 804 805 rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm); 806 807 return ppc_md.set_rtc_time(&tm); 808 } 809 810 static void __read_persistent_clock(struct timespec64 *ts) 811 { 812 struct rtc_time tm; 813 static int first = 1; 814 815 ts->tv_nsec = 0; 816 /* XXX this is a litle fragile but will work okay in the short term */ 817 if (first) { 818 first = 0; 819 if (ppc_md.time_init) 820 timezone_offset = ppc_md.time_init(); 821 822 /* get_boot_time() isn't guaranteed to be safe to call late */ 823 if (ppc_md.get_boot_time) { 824 ts->tv_sec = ppc_md.get_boot_time() - timezone_offset; 825 return; 826 } 827 } 828 if (!ppc_md.get_rtc_time) { 829 ts->tv_sec = 0; 830 return; 831 } 832 ppc_md.get_rtc_time(&tm); 833 834 ts->tv_sec = rtc_tm_to_time64(&tm); 835 } 836 837 void read_persistent_clock64(struct timespec64 *ts) 838 { 839 __read_persistent_clock(ts); 840 841 /* Sanitize it in case real time clock is set below EPOCH */ 842 if (ts->tv_sec < 0) { 843 ts->tv_sec = 0; 844 ts->tv_nsec = 0; 845 } 846 847 } 848 849 /* clocksource code */ 850 static notrace u64 rtc_read(struct clocksource *cs) 851 { 852 return (u64)get_rtc(); 853 } 854 855 static notrace u64 timebase_read(struct clocksource *cs) 856 { 857 return (u64)get_tb(); 858 } 859 860 861 void update_vsyscall(struct timekeeper *tk) 862 { 863 struct timespec64 xt; 864 struct clocksource *clock = tk->tkr_mono.clock; 865 u32 mult = tk->tkr_mono.mult; 866 u32 shift = tk->tkr_mono.shift; 867 u64 cycle_last = tk->tkr_mono.cycle_last; 868 u64 new_tb_to_xs, new_stamp_xsec; 869 u64 frac_sec; 870 871 if (clock != &clocksource_timebase) 872 return; 873 874 xt.tv_sec = tk->xtime_sec; 875 xt.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift); 876 877 /* Make userspace gettimeofday spin until we're done. */ 878 ++vdso_data->tb_update_count; 879 smp_mb(); 880 881 /* 882 * This computes ((2^20 / 1e9) * mult) >> shift as a 883 * 0.64 fixed-point fraction. 884 * The computation in the else clause below won't overflow 885 * (as long as the timebase frequency is >= 1.049 MHz) 886 * but loses precision because we lose the low bits of the constant 887 * in the shift. Note that 19342813113834067 ~= 2^(20+64) / 1e9. 888 * For a shift of 24 the error is about 0.5e-9, or about 0.5ns 889 * over a second. (Shift values are usually 22, 23 or 24.) 890 * For high frequency clocks such as the 512MHz timebase clock 891 * on POWER[6789], the mult value is small (e.g. 32768000) 892 * and so we can shift the constant by 16 initially 893 * (295147905179 ~= 2^(20+64-16) / 1e9) and then do the 894 * remaining shifts after the multiplication, which gives a 895 * more accurate result (e.g. with mult = 32768000, shift = 24, 896 * the error is only about 1.2e-12, or 0.7ns over 10 minutes). 897 */ 898 if (mult <= 62500000 && clock->shift >= 16) 899 new_tb_to_xs = ((u64) mult * 295147905179ULL) >> (clock->shift - 16); 900 else 901 new_tb_to_xs = (u64) mult * (19342813113834067ULL >> clock->shift); 902 903 /* 904 * Compute the fractional second in units of 2^-32 seconds. 905 * The fractional second is tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift 906 * in nanoseconds, so multiplying that by 2^32 / 1e9 gives 907 * it in units of 2^-32 seconds. 908 * We assume shift <= 32 because clocks_calc_mult_shift() 909 * generates shift values in the range 0 - 32. 910 */ 911 frac_sec = tk->tkr_mono.xtime_nsec << (32 - shift); 912 do_div(frac_sec, NSEC_PER_SEC); 913 914 /* 915 * Work out new stamp_xsec value for any legacy users of systemcfg. 916 * stamp_xsec is in units of 2^-20 seconds. 917 */ 918 new_stamp_xsec = frac_sec >> 12; 919 new_stamp_xsec += tk->xtime_sec * XSEC_PER_SEC; 920 921 /* 922 * tb_update_count is used to allow the userspace gettimeofday code 923 * to assure itself that it sees a consistent view of the tb_to_xs and 924 * stamp_xsec variables. It reads the tb_update_count, then reads 925 * tb_to_xs and stamp_xsec and then reads tb_update_count again. If 926 * the two values of tb_update_count match and are even then the 927 * tb_to_xs and stamp_xsec values are consistent. If not, then it 928 * loops back and reads them again until this criteria is met. 929 */ 930 vdso_data->tb_orig_stamp = cycle_last; 931 vdso_data->stamp_xsec = new_stamp_xsec; 932 vdso_data->tb_to_xs = new_tb_to_xs; 933 vdso_data->wtom_clock_sec = tk->wall_to_monotonic.tv_sec; 934 vdso_data->wtom_clock_nsec = tk->wall_to_monotonic.tv_nsec; 935 vdso_data->stamp_xtime_sec = xt.tv_sec; 936 vdso_data->stamp_xtime_nsec = xt.tv_nsec; 937 vdso_data->stamp_sec_fraction = frac_sec; 938 vdso_data->hrtimer_res = hrtimer_resolution; 939 smp_wmb(); 940 ++(vdso_data->tb_update_count); 941 } 942 943 void update_vsyscall_tz(void) 944 { 945 vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; 946 vdso_data->tz_dsttime = sys_tz.tz_dsttime; 947 } 948 949 static void __init clocksource_init(void) 950 { 951 struct clocksource *clock; 952 953 if (__USE_RTC()) 954 clock = &clocksource_rtc; 955 else 956 clock = &clocksource_timebase; 957 958 if (clocksource_register_hz(clock, tb_ticks_per_sec)) { 959 printk(KERN_ERR "clocksource: %s is already registered\n", 960 clock->name); 961 return; 962 } 963 964 printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n", 965 clock->name, clock->mult, clock->shift); 966 } 967 968 static int decrementer_set_next_event(unsigned long evt, 969 struct clock_event_device *dev) 970 { 971 __this_cpu_write(decrementers_next_tb, get_tb_or_rtc() + evt); 972 set_dec(evt); 973 974 /* We may have raced with new irq work */ 975 if (test_irq_work_pending()) 976 set_dec(1); 977 978 return 0; 979 } 980 981 static int decrementer_shutdown(struct clock_event_device *dev) 982 { 983 decrementer_set_next_event(decrementer_max, dev); 984 return 0; 985 } 986 987 static void register_decrementer_clockevent(int cpu) 988 { 989 struct clock_event_device *dec = &per_cpu(decrementers, cpu); 990 991 *dec = decrementer_clockevent; 992 dec->cpumask = cpumask_of(cpu); 993 994 clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max); 995 996 printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n", 997 dec->name, dec->mult, dec->shift, cpu); 998 999 /* Set values for KVM, see kvm_emulate_dec() */ 1000 decrementer_clockevent.mult = dec->mult; 1001 decrementer_clockevent.shift = dec->shift; 1002 } 1003 1004 static void enable_large_decrementer(void) 1005 { 1006 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 1007 return; 1008 1009 if (decrementer_max <= DECREMENTER_DEFAULT_MAX) 1010 return; 1011 1012 /* 1013 * If we're running as the hypervisor we need to enable the LD manually 1014 * otherwise firmware should have done it for us. 1015 */ 1016 if (cpu_has_feature(CPU_FTR_HVMODE)) 1017 mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD); 1018 } 1019 1020 static void __init set_decrementer_max(void) 1021 { 1022 struct device_node *cpu; 1023 u32 bits = 32; 1024 1025 /* Prior to ISAv3 the decrementer is always 32 bit */ 1026 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 1027 return; 1028 1029 cpu = of_find_node_by_type(NULL, "cpu"); 1030 1031 if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) { 1032 if (bits > 64 || bits < 32) { 1033 pr_warn("time_init: firmware supplied invalid ibm,dec-bits"); 1034 bits = 32; 1035 } 1036 1037 /* calculate the signed maximum given this many bits */ 1038 decrementer_max = (1ul << (bits - 1)) - 1; 1039 } 1040 1041 of_node_put(cpu); 1042 1043 pr_info("time_init: %u bit decrementer (max: %llx)\n", 1044 bits, decrementer_max); 1045 } 1046 1047 static void __init init_decrementer_clockevent(void) 1048 { 1049 register_decrementer_clockevent(smp_processor_id()); 1050 } 1051 1052 void secondary_cpu_time_init(void) 1053 { 1054 /* Enable and test the large decrementer for this cpu */ 1055 enable_large_decrementer(); 1056 1057 /* Start the decrementer on CPUs that have manual control 1058 * such as BookE 1059 */ 1060 start_cpu_decrementer(); 1061 1062 /* FIME: Should make unrelatred change to move snapshot_timebase 1063 * call here ! */ 1064 register_decrementer_clockevent(smp_processor_id()); 1065 } 1066 1067 /* This function is only called on the boot processor */ 1068 void __init time_init(void) 1069 { 1070 struct div_result res; 1071 u64 scale; 1072 unsigned shift; 1073 1074 if (__USE_RTC()) { 1075 /* 601 processor: dec counts down by 128 every 128ns */ 1076 ppc_tb_freq = 1000000000; 1077 } else { 1078 /* Normal PowerPC with timebase register */ 1079 ppc_md.calibrate_decr(); 1080 printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n", 1081 ppc_tb_freq / 1000000, ppc_tb_freq % 1000000); 1082 printk(KERN_DEBUG "time_init: processor frequency = %lu.%.6lu MHz\n", 1083 ppc_proc_freq / 1000000, ppc_proc_freq % 1000000); 1084 } 1085 1086 tb_ticks_per_jiffy = ppc_tb_freq / HZ; 1087 tb_ticks_per_sec = ppc_tb_freq; 1088 tb_ticks_per_usec = ppc_tb_freq / 1000000; 1089 calc_cputime_factors(); 1090 1091 /* 1092 * Compute scale factor for sched_clock. 1093 * The calibrate_decr() function has set tb_ticks_per_sec, 1094 * which is the timebase frequency. 1095 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret 1096 * the 128-bit result as a 64.64 fixed-point number. 1097 * We then shift that number right until it is less than 1.0, 1098 * giving us the scale factor and shift count to use in 1099 * sched_clock(). 1100 */ 1101 div128_by_32(1000000000, 0, tb_ticks_per_sec, &res); 1102 scale = res.result_low; 1103 for (shift = 0; res.result_high != 0; ++shift) { 1104 scale = (scale >> 1) | (res.result_high << 63); 1105 res.result_high >>= 1; 1106 } 1107 tb_to_ns_scale = scale; 1108 tb_to_ns_shift = shift; 1109 /* Save the current timebase to pretty up CONFIG_PRINTK_TIME */ 1110 boot_tb = get_tb_or_rtc(); 1111 1112 /* If platform provided a timezone (pmac), we correct the time */ 1113 if (timezone_offset) { 1114 sys_tz.tz_minuteswest = -timezone_offset / 60; 1115 sys_tz.tz_dsttime = 0; 1116 } 1117 1118 vdso_data->tb_update_count = 0; 1119 vdso_data->tb_ticks_per_sec = tb_ticks_per_sec; 1120 1121 /* initialise and enable the large decrementer (if we have one) */ 1122 set_decrementer_max(); 1123 enable_large_decrementer(); 1124 1125 /* Start the decrementer on CPUs that have manual control 1126 * such as BookE 1127 */ 1128 start_cpu_decrementer(); 1129 1130 /* Register the clocksource */ 1131 clocksource_init(); 1132 1133 init_decrementer_clockevent(); 1134 tick_setup_hrtimer_broadcast(); 1135 1136 of_clk_init(NULL); 1137 } 1138 1139 /* 1140 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit 1141 * result. 1142 */ 1143 void div128_by_32(u64 dividend_high, u64 dividend_low, 1144 unsigned divisor, struct div_result *dr) 1145 { 1146 unsigned long a, b, c, d; 1147 unsigned long w, x, y, z; 1148 u64 ra, rb, rc; 1149 1150 a = dividend_high >> 32; 1151 b = dividend_high & 0xffffffff; 1152 c = dividend_low >> 32; 1153 d = dividend_low & 0xffffffff; 1154 1155 w = a / divisor; 1156 ra = ((u64)(a - (w * divisor)) << 32) + b; 1157 1158 rb = ((u64) do_div(ra, divisor) << 32) + c; 1159 x = ra; 1160 1161 rc = ((u64) do_div(rb, divisor) << 32) + d; 1162 y = rb; 1163 1164 do_div(rc, divisor); 1165 z = rc; 1166 1167 dr->result_high = ((u64)w << 32) + x; 1168 dr->result_low = ((u64)y << 32) + z; 1169 1170 } 1171 1172 /* We don't need to calibrate delay, we use the CPU timebase for that */ 1173 void calibrate_delay(void) 1174 { 1175 /* Some generic code (such as spinlock debug) use loops_per_jiffy 1176 * as the number of __delay(1) in a jiffy, so make it so 1177 */ 1178 loops_per_jiffy = tb_ticks_per_jiffy; 1179 } 1180 1181 #if IS_ENABLED(CONFIG_RTC_DRV_GENERIC) 1182 static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm) 1183 { 1184 ppc_md.get_rtc_time(tm); 1185 return 0; 1186 } 1187 1188 static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm) 1189 { 1190 if (!ppc_md.set_rtc_time) 1191 return -EOPNOTSUPP; 1192 1193 if (ppc_md.set_rtc_time(tm) < 0) 1194 return -EOPNOTSUPP; 1195 1196 return 0; 1197 } 1198 1199 static const struct rtc_class_ops rtc_generic_ops = { 1200 .read_time = rtc_generic_get_time, 1201 .set_time = rtc_generic_set_time, 1202 }; 1203 1204 static int __init rtc_init(void) 1205 { 1206 struct platform_device *pdev; 1207 1208 if (!ppc_md.get_rtc_time) 1209 return -ENODEV; 1210 1211 pdev = platform_device_register_data(NULL, "rtc-generic", -1, 1212 &rtc_generic_ops, 1213 sizeof(rtc_generic_ops)); 1214 1215 return PTR_ERR_OR_ZERO(pdev); 1216 } 1217 1218 device_initcall(rtc_init); 1219 #endif 1220