time.c

   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/sched/cputime.h>
  35#include <linux/kernel.h>
  36#include <linux/param.h>
  37#include <linux/string.h>
  38#include <linux/mm.h>
  39#include <linux/interrupt.h>
  40#include <linux/timex.h>
  41#include <linux/kernel_stat.h>
  42#include <linux/time.h>
  43#include <linux/init.h>
  44#include <linux/profile.h>
  45#include <linux/cpu.h>
  46#include <linux/security.h>
  47#include <linux/percpu.h>
  48#include <linux/rtc.h>
  49#include <linux/jiffies.h>
  50#include <linux/posix-timers.h>
  51#include <linux/irq.h>
  52#include <linux/delay.h>
  53#include <linux/irq_work.h>
  54#include <linux/of_clk.h>
  55#include <linux/suspend.h>
  56#include <linux/processor.h>
  57#include <linux/mc146818rtc.h>
  58#include <linux/platform_device.h>
  59
  60#include <asm/trace.h>
  61#include <asm/interrupt.h>
  62#include <asm/io.h>
  63#include <asm/nvram.h>
  64#include <asm/cache.h>
  65#include <asm/machdep.h>
  66#include <linux/uaccess.h>
  67#include <asm/time.h>
  68#include <asm/irq.h>
  69#include <asm/div64.h>
  70#include <asm/smp.h>
  71#include <asm/vdso_datapage.h>
  72#include <asm/firmware.h>
  73#include <asm/mce.h>
  74#include <asm/systemcfg.h>
  75
  76/* powerpc clocksource/clockevent code */
  77
  78#include <linux/clockchips.h>
  79
  80static u64 timebase_read(struct clocksource *);
  81static struct clocksource clocksource_timebase = {
  82	.name         = "timebase",
  83	.rating       = 400,
  84	.flags        = CLOCK_SOURCE_IS_CONTINUOUS,
  85	.mask         = CLOCKSOURCE_MASK(64),
  86	.read         = timebase_read,
  87	.vdso_clock_mode	= VDSO_CLOCKMODE_ARCHTIMER,
  88};
  89
  90#define DECREMENTER_DEFAULT_MAX 0x7FFFFFFF
  91u64 decrementer_max = DECREMENTER_DEFAULT_MAX;
  92EXPORT_SYMBOL_GPL(decrementer_max); /* for KVM HDEC */
  93
  94static int decrementer_set_next_event(unsigned long evt,
  95				      struct clock_event_device *dev);
  96static int decrementer_shutdown(struct clock_event_device *evt);
  97
  98struct clock_event_device decrementer_clockevent = {
  99	.name			= "decrementer",
 100	.rating			= 200,
 101	.irq			= 0,
 102	.set_next_event		= decrementer_set_next_event,
 103	.set_state_oneshot_stopped = decrementer_shutdown,
 104	.set_state_shutdown	= decrementer_shutdown,
 105	.tick_resume		= decrementer_shutdown,
 106	.features		= CLOCK_EVT_FEAT_ONESHOT |
 107				  CLOCK_EVT_FEAT_C3STOP,
 108};
 109EXPORT_SYMBOL(decrementer_clockevent);
 110
 111/*
 112 * This always puts next_tb beyond now, so the clock event will never fire
 113 * with the usual comparison, no need for a separate test for stopped.
 114 */
 115#define DEC_CLOCKEVENT_STOPPED ~0ULL
 116DEFINE_PER_CPU(u64, decrementers_next_tb) = DEC_CLOCKEVENT_STOPPED;
 117EXPORT_SYMBOL_GPL(decrementers_next_tb);
 118static DEFINE_PER_CPU(struct clock_event_device, decrementers);
 119
 120#define XSEC_PER_SEC (1024*1024)
 121
 122#ifdef CONFIG_PPC64
 123#define SCALE_XSEC(xsec, max)	(((xsec) * max) / XSEC_PER_SEC)
 124#else
 125/* compute ((xsec << 12) * max) >> 32 */
 126#define SCALE_XSEC(xsec, max)	mulhwu((xsec) << 12, max)
 127#endif
 128
 129unsigned long tb_ticks_per_jiffy;
 130unsigned long tb_ticks_per_usec = 100; /* sane default */
 131EXPORT_SYMBOL(tb_ticks_per_usec);
 132unsigned long tb_ticks_per_sec;
 133EXPORT_SYMBOL(tb_ticks_per_sec);	/* for cputime conversions */
 134
 135DEFINE_SPINLOCK(rtc_lock);
 136EXPORT_SYMBOL_GPL(rtc_lock);
 137
 138static u64 tb_to_ns_scale __read_mostly;
 139static unsigned tb_to_ns_shift __read_mostly;
 140static u64 boot_tb __ro_after_init;
 141
 142extern struct timezone sys_tz;
 143static long timezone_offset;
 144
 145unsigned long ppc_proc_freq;
 146EXPORT_SYMBOL_GPL(ppc_proc_freq);
 147unsigned long ppc_tb_freq;
 148EXPORT_SYMBOL_GPL(ppc_tb_freq);
 149
 150bool tb_invalid;
 151
 152#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 153/*
 154 * Read the SPURR on systems that have it, otherwise the PURR,
 155 * or if that doesn't exist return the timebase value passed in.
 156 */
 157static inline unsigned long read_spurr(unsigned long tb)
 158{
 159	if (cpu_has_feature(CPU_FTR_SPURR))
 160		return mfspr(SPRN_SPURR);
 161	if (cpu_has_feature(CPU_FTR_PURR))
 162		return mfspr(SPRN_PURR);
 163	return tb;
 164}
 165
 166/*
 167 * Account time for a transition between system, hard irq
 168 * or soft irq state.
 169 */
 170static unsigned long vtime_delta_scaled(struct cpu_accounting_data *acct,
 171					unsigned long now, unsigned long stime)
 172{
 173	unsigned long stime_scaled = 0;
 174#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 175	unsigned long nowscaled, deltascaled;
 176	unsigned long utime, utime_scaled;
 177
 178	nowscaled = read_spurr(now);
 179	deltascaled = nowscaled - acct->startspurr;
 180	acct->startspurr = nowscaled;
 181	utime = acct->utime - acct->utime_sspurr;
 182	acct->utime_sspurr = acct->utime;
 183
 184	/*
 185	 * Because we don't read the SPURR on every kernel entry/exit,
 186	 * deltascaled includes both user and system SPURR ticks.
 187	 * Apportion these ticks to system SPURR ticks and user
 188	 * SPURR ticks in the same ratio as the system time (delta)
 189	 * and user time (udelta) values obtained from the timebase
 190	 * over the same interval.  The system ticks get accounted here;
 191	 * the user ticks get saved up in paca->user_time_scaled to be
 192	 * used by account_process_tick.
 193	 */
 194	stime_scaled = stime;
 195	utime_scaled = utime;
 196	if (deltascaled != stime + utime) {
 197		if (utime) {
 198			stime_scaled = deltascaled * stime / (stime + utime);
 199			utime_scaled = deltascaled - stime_scaled;
 200		} else {
 201			stime_scaled = deltascaled;
 202		}
 203	}
 204	acct->utime_scaled += utime_scaled;
 205#endif
 206
 207	return stime_scaled;
 208}
 209
 210static unsigned long vtime_delta(struct cpu_accounting_data *acct,
 211				 unsigned long *stime_scaled,
 212				 unsigned long *steal_time)
 213{
 214	unsigned long now, stime;
 215
 216	WARN_ON_ONCE(!irqs_disabled());
 217
 218	now = mftb();
 219	stime = now - acct->starttime;
 220	acct->starttime = now;
 221
 222	*stime_scaled = vtime_delta_scaled(acct, now, stime);
 223
 224	if (IS_ENABLED(CONFIG_PPC_SPLPAR) &&
 225			firmware_has_feature(FW_FEATURE_SPLPAR))
 226		*steal_time = pseries_calculate_stolen_time(now);
 227	else
 228		*steal_time = 0;
 229
 230	return stime;
 231}
 232
 233static void vtime_delta_kernel(struct cpu_accounting_data *acct,
 234			       unsigned long *stime, unsigned long *stime_scaled)
 235{
 236	unsigned long steal_time;
 237
 238	*stime = vtime_delta(acct, stime_scaled, &steal_time);
 239	*stime -= min(*stime, steal_time);
 240	acct->steal_time += steal_time;
 241}
 242
 243void vtime_account_kernel(struct task_struct *tsk)
 244{
 245	struct cpu_accounting_data *acct = get_accounting(tsk);
 246	unsigned long stime, stime_scaled;
 247
 248	vtime_delta_kernel(acct, &stime, &stime_scaled);
 249
 250	if (tsk->flags & PF_VCPU) {
 251		acct->gtime += stime;
 252#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 253		acct->utime_scaled += stime_scaled;
 254#endif
 255	} else {
 256		acct->stime += stime;
 257#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 258		acct->stime_scaled += stime_scaled;
 259#endif
 260	}
 261}
 262EXPORT_SYMBOL_GPL(vtime_account_kernel);
 263
 264void vtime_account_idle(struct task_struct *tsk)
 265{
 266	unsigned long stime, stime_scaled, steal_time;
 267	struct cpu_accounting_data *acct = get_accounting(tsk);
 268
 269	stime = vtime_delta(acct, &stime_scaled, &steal_time);
 270	acct->idle_time += stime + steal_time;
 271}
 272
 273static void vtime_account_irq_field(struct cpu_accounting_data *acct,
 274				    unsigned long *field)
 275{
 276	unsigned long stime, stime_scaled;
 277
 278	vtime_delta_kernel(acct, &stime, &stime_scaled);
 279	*field += stime;
 280#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 281	acct->stime_scaled += stime_scaled;
 282#endif
 283}
 284
 285void vtime_account_softirq(struct task_struct *tsk)
 286{
 287	struct cpu_accounting_data *acct = get_accounting(tsk);
 288	vtime_account_irq_field(acct, &acct->softirq_time);
 289}
 290
 291void vtime_account_hardirq(struct task_struct *tsk)
 292{
 293	struct cpu_accounting_data *acct = get_accounting(tsk);
 294	vtime_account_irq_field(acct, &acct->hardirq_time);
 295}
 296
 297static void vtime_flush_scaled(struct task_struct *tsk,
 298			       struct cpu_accounting_data *acct)
 299{
 300#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 301	if (acct->utime_scaled)
 302		tsk->utimescaled += cputime_to_nsecs(acct->utime_scaled);
 303	if (acct->stime_scaled)
 304		tsk->stimescaled += cputime_to_nsecs(acct->stime_scaled);
 305
 306	acct->utime_scaled = 0;
 307	acct->utime_sspurr = 0;
 308	acct->stime_scaled = 0;
 309#endif
 310}
 311
 312/*
 313 * Account the whole cputime accumulated in the paca
 314 * Must be called with interrupts disabled.
 315 * Assumes that vtime_account_kernel/idle() has been called
 316 * recently (i.e. since the last entry from usermode) so that
 317 * get_paca()->user_time_scaled is up to date.
 318 */
 319void vtime_flush(struct task_struct *tsk)
 320{
 321	struct cpu_accounting_data *acct = get_accounting(tsk);
 322
 323	if (acct->utime)
 324		account_user_time(tsk, cputime_to_nsecs(acct->utime));
 325
 326	if (acct->gtime)
 327		account_guest_time(tsk, cputime_to_nsecs(acct->gtime));
 328
 329	if (IS_ENABLED(CONFIG_PPC_SPLPAR) && acct->steal_time) {
 330		account_steal_time(cputime_to_nsecs(acct->steal_time));
 331		acct->steal_time = 0;
 332	}
 333
 334	if (acct->idle_time)
 335		account_idle_time(cputime_to_nsecs(acct->idle_time));
 336
 337	if (acct->stime)
 338		account_system_index_time(tsk, cputime_to_nsecs(acct->stime),
 339					  CPUTIME_SYSTEM);
 340
 341	if (acct->hardirq_time)
 342		account_system_index_time(tsk, cputime_to_nsecs(acct->hardirq_time),
 343					  CPUTIME_IRQ);
 344	if (acct->softirq_time)
 345		account_system_index_time(tsk, cputime_to_nsecs(acct->softirq_time),
 346					  CPUTIME_SOFTIRQ);
 347
 348	vtime_flush_scaled(tsk, acct);
 349
 350	acct->utime = 0;
 351	acct->gtime = 0;
 352	acct->idle_time = 0;
 353	acct->stime = 0;
 354	acct->hardirq_time = 0;
 355	acct->softirq_time = 0;
 356}
 357
 358/*
 359 * Called from the context switch with interrupts disabled, to charge all
 360 * accumulated times to the current process, and to prepare accounting on
 361 * the next process.
 362 */
 363void vtime_task_switch(struct task_struct *prev)
 364{
 365	if (is_idle_task(prev))
 366		vtime_account_idle(prev);
 367	else
 368		vtime_account_kernel(prev);
 369
 370	vtime_flush(prev);
 371
 372	if (!IS_ENABLED(CONFIG_PPC64)) {
 373		struct cpu_accounting_data *acct = get_accounting(current);
 374		struct cpu_accounting_data *acct0 = get_accounting(prev);
 375
 376		acct->starttime = acct0->starttime;
 377	}
 378}
 379#endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 380
 381void __no_kcsan __delay(unsigned long loops)
 382{
 383	unsigned long start;
 384
 385	spin_begin();
 386	if (tb_invalid) {
 387		/*
 388		 * TB is in error state and isn't ticking anymore.
 389		 * HMI handler was unable to recover from TB error.
 390		 * Return immediately, so that kernel won't get stuck here.
 391		 */
 392		spin_cpu_relax();
 393	} else {
 394		start = mftb();
 395		while (mftb() - start < loops)
 396			spin_cpu_relax();
 397	}
 398	spin_end();
 399}
 400EXPORT_SYMBOL(__delay);
 401
 402void __no_kcsan udelay(unsigned long usecs)
 403{
 404	__delay(tb_ticks_per_usec * usecs);
 405}
 406EXPORT_SYMBOL(udelay);
 407
 408#ifdef CONFIG_SMP
 409unsigned long profile_pc(struct pt_regs *regs)
 410{
 411	unsigned long pc = instruction_pointer(regs);
 412
 413	if (in_lock_functions(pc))
 414		return regs->link;
 415
 416	return pc;
 417}
 418EXPORT_SYMBOL(profile_pc);
 419#endif
 420
 421#ifdef CONFIG_IRQ_WORK
 422
 423/*
 424 * 64-bit uses a byte in the PACA, 32-bit uses a per-cpu variable...
 425 */
 426#ifdef CONFIG_PPC64
 427static inline unsigned long test_irq_work_pending(void)
 428{
 429	unsigned long x;
 430
 431	asm volatile("lbz %0,%1(13)"
 432		: "=r" (x)
 433		: "i" (offsetof(struct paca_struct, irq_work_pending)));
 434	return x;
 435}
 436
 437static inline void set_irq_work_pending_flag(void)
 438{
 439	asm volatile("stb %0,%1(13)" : :
 440		"r" (1),
 441		"i" (offsetof(struct paca_struct, irq_work_pending)));
 442}
 443
 444static inline void clear_irq_work_pending(void)
 445{
 446	asm volatile("stb %0,%1(13)" : :
 447		"r" (0),
 448		"i" (offsetof(struct paca_struct, irq_work_pending)));
 449}
 450
 451#else /* 32-bit */
 452
 453DEFINE_PER_CPU(u8, irq_work_pending);
 454
 455#define set_irq_work_pending_flag()	__this_cpu_write(irq_work_pending, 1)
 456#define test_irq_work_pending()		__this_cpu_read(irq_work_pending)
 457#define clear_irq_work_pending()	__this_cpu_write(irq_work_pending, 0)
 458
 459#endif /* 32 vs 64 bit */
 460
 461void arch_irq_work_raise(void)
 462{
 463	/*
 464	 * 64-bit code that uses irq soft-mask can just cause an immediate
 465	 * interrupt here that gets soft masked, if this is called under
 466	 * local_irq_disable(). It might be possible to prevent that happening
 467	 * by noticing interrupts are disabled and setting decrementer pending
 468	 * to be replayed when irqs are enabled. The problem there is that
 469	 * tracing can call irq_work_raise, including in code that does low
 470	 * level manipulations of irq soft-mask state (e.g., trace_hardirqs_on)
 471	 * which could get tangled up if we're messing with the same state
 472	 * here.
 473	 */
 474	preempt_disable();
 475	set_irq_work_pending_flag();
 476	set_dec(1);
 477	preempt_enable();
 478}
 479
 480static void set_dec_or_work(u64 val)
 481{
 482	set_dec(val);
 483	/* We may have raced with new irq work */
 484	if (unlikely(test_irq_work_pending()))
 485		set_dec(1);
 486}
 487
 488#else  /* CONFIG_IRQ_WORK */
 489
 490#define test_irq_work_pending()	0
 491#define clear_irq_work_pending()
 492
 493static void set_dec_or_work(u64 val)
 494{
 495	set_dec(val);
 496}
 497#endif /* CONFIG_IRQ_WORK */
 498
 499#ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 500void timer_rearm_host_dec(u64 now)
 501{
 502	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
 503
 504	WARN_ON_ONCE(!arch_irqs_disabled());
 505	WARN_ON_ONCE(mfmsr() & MSR_EE);
 506
 507	if (now >= *next_tb) {
 508		local_paca->irq_happened |= PACA_IRQ_DEC;
 509	} else {
 510		now = *next_tb - now;
 511		if (now > decrementer_max)
 512			now = decrementer_max;
 513		set_dec_or_work(now);
 514	}
 515}
 516EXPORT_SYMBOL_GPL(timer_rearm_host_dec);
 517#endif
 518
 519/*
 520 * timer_interrupt - gets called when the decrementer overflows,
 521 * with interrupts disabled.
 522 */
 523DEFINE_INTERRUPT_HANDLER_ASYNC(timer_interrupt)
 524{
 525	struct clock_event_device *evt = this_cpu_ptr(&decrementers);
 526	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
 527	struct pt_regs *old_regs;
 528	u64 now;
 529
 530	/*
 531	 * Some implementations of hotplug will get timer interrupts while
 532	 * offline, just ignore these.
 533	 */
 534	if (unlikely(!cpu_online(smp_processor_id()))) {
 535		set_dec(decrementer_max);
 536		return;
 537	}
 538
 539	/* Conditionally hard-enable interrupts. */
 540	if (should_hard_irq_enable(regs)) {
 541		/*
 542		 * Ensure a positive value is written to the decrementer, or
 543		 * else some CPUs will continue to take decrementer exceptions.
 544		 * When the PPC_WATCHDOG (decrementer based) is configured,
 545		 * keep this at most 31 bits, which is about 4 seconds on most
 546		 * systems, which gives the watchdog a chance of catching timer
 547		 * interrupt hard lockups.
 548		 */
 549		if (IS_ENABLED(CONFIG_PPC_WATCHDOG))
 550			set_dec(0x7fffffff);
 551		else
 552			set_dec(decrementer_max);
 553
 554		do_hard_irq_enable();
 555	}
 556
 557#if defined(CONFIG_PPC32) && defined(CONFIG_PPC_PMAC)
 558	if (atomic_read(&ppc_n_lost_interrupts) != 0)
 559		__do_IRQ(regs);
 560#endif
 561
 562	old_regs = set_irq_regs(regs);
 563
 564	trace_timer_interrupt_entry(regs);
 565
 566	if (test_irq_work_pending()) {
 567		clear_irq_work_pending();
 568		mce_run_irq_context_handlers();
 569		irq_work_run();
 570	}
 571
 572	now = get_tb();
 573	if (now >= *next_tb) {
 574		evt->event_handler(evt);
 575		__this_cpu_inc(irq_stat.timer_irqs_event);
 576	} else {
 577		now = *next_tb - now;
 578		if (now > decrementer_max)
 579			now = decrementer_max;
 580		set_dec_or_work(now);
 581		__this_cpu_inc(irq_stat.timer_irqs_others);
 582	}
 583
 584	trace_timer_interrupt_exit(regs);
 585
 586	set_irq_regs(old_regs);
 587}
 588EXPORT_SYMBOL(timer_interrupt);
 589
 590#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 591void timer_broadcast_interrupt(void)
 592{
 593	tick_receive_broadcast();
 594	__this_cpu_inc(irq_stat.broadcast_irqs_event);
 595}
 596#endif
 597
 598#ifdef CONFIG_SUSPEND
 599/* Overrides the weak version in kernel/power/main.c */
 600void arch_suspend_disable_irqs(void)
 601{
 602	if (ppc_md.suspend_disable_irqs)
 603		ppc_md.suspend_disable_irqs();
 604
 605	/* Disable the decrementer, so that it doesn't interfere
 606	 * with suspending.
 607	 */
 608
 609	set_dec(decrementer_max);
 610	local_irq_disable();
 611	set_dec(decrementer_max);
 612}
 613
 614/* Overrides the weak version in kernel/power/main.c */
 615void arch_suspend_enable_irqs(void)
 616{
 617	local_irq_enable();
 618
 619	if (ppc_md.suspend_enable_irqs)
 620		ppc_md.suspend_enable_irqs();
 621}
 622#endif
 623
 624unsigned long long tb_to_ns(unsigned long long ticks)
 625{
 626	return mulhdu(ticks, tb_to_ns_scale) << tb_to_ns_shift;
 627}
 628EXPORT_SYMBOL_GPL(tb_to_ns);
 629
 630/*
 631 * Scheduler clock - returns current time in nanosec units.
 632 *
 633 * Note: mulhdu(a, b) (multiply high double unsigned) returns
 634 * the high 64 bits of a * b, i.e. (a * b) >> 64, where a and b
 635 * are 64-bit unsigned numbers.
 636 */
 637notrace unsigned long long sched_clock(void)
 638{
 639	return mulhdu(get_tb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
 640}
 641
 642#ifdef CONFIG_PPC_SPLPAR
 643u64 get_boot_tb(void)
 644{
 645	return boot_tb;
 646}
 647#endif
 648
 649#ifdef CONFIG_PPC_PSERIES
 650
 651/*
 652 * Running clock - attempts to give a view of time passing for a virtualised
 653 * kernels.
 654 * Uses the VTB register if available otherwise a next best guess.
 655 */
 656unsigned long long running_clock(void)
 657{
 658	/*
 659	 * Don't read the VTB as a host since KVM does not switch in host
 660	 * timebase into the VTB when it takes a guest off the CPU, reading the
 661	 * VTB would result in reading 'last switched out' guest VTB.
 662	 *
 663	 * Host kernels are often compiled with CONFIG_PPC_PSERIES checked, it
 664	 * would be unsafe to rely only on the #ifdef above.
 665	 */
 666	if (firmware_has_feature(FW_FEATURE_LPAR) &&
 667	    cpu_has_feature(CPU_FTR_ARCH_207S))
 668		return mulhdu(get_vtb() - boot_tb, tb_to_ns_scale) << tb_to_ns_shift;
 669
 670	/*
 671	 * This is a next best approximation without a VTB.
 672	 * On a host which is running bare metal there should never be any stolen
 673	 * time and on a host which doesn't do any virtualisation TB *should* equal
 674	 * VTB so it makes no difference anyway.
 675	 */
 676	return local_clock() - kcpustat_this_cpu->cpustat[CPUTIME_STEAL];
 677}
 678#endif
 679
 680static int __init get_freq(char *name, int cells, unsigned long *val)
 681{
 682	struct device_node *cpu;
 683	const __be32 *fp;
 684	int found = 0;
 685
 686	/* The cpu node should have timebase and clock frequency properties */
 687	cpu = of_find_node_by_type(NULL, "cpu");
 688
 689	if (cpu) {
 690		fp = of_get_property(cpu, name, NULL);
 691		if (fp) {
 692			found = 1;
 693			*val = of_read_ulong(fp, cells);
 694		}
 695
 696		of_node_put(cpu);
 697	}
 698
 699	return found;
 700}
 701
 702static void start_cpu_decrementer(void)
 703{
 704#ifdef CONFIG_BOOKE
 705	unsigned int tcr;
 706
 707	/* Clear any pending timer interrupts */
 708	mtspr(SPRN_TSR, TSR_ENW | TSR_WIS | TSR_DIS | TSR_FIS);
 709
 710	tcr = mfspr(SPRN_TCR);
 711	/*
 712	 * The watchdog may have already been enabled by u-boot. So leave
 713	 * TRC[WP] (Watchdog Period) alone.
 714	 */
 715	tcr &= TCR_WP_MASK;	/* Clear all bits except for TCR[WP] */
 716	tcr |= TCR_DIE;		/* Enable decrementer */
 717	mtspr(SPRN_TCR, tcr);
 718#endif
 719}
 720
 721void __init generic_calibrate_decr(void)
 722{
 723	ppc_tb_freq = DEFAULT_TB_FREQ;		/* hardcoded default */
 724
 725	if (!get_freq("ibm,extended-timebase-frequency", 2, &ppc_tb_freq) &&
 726	    !get_freq("timebase-frequency", 1, &ppc_tb_freq)) {
 727
 728		printk(KERN_ERR "WARNING: Estimating decrementer frequency "
 729				"(not found)\n");
 730	}
 731
 732	ppc_proc_freq = DEFAULT_PROC_FREQ;	/* hardcoded default */
 733
 734	if (!get_freq("ibm,extended-clock-frequency", 2, &ppc_proc_freq) &&
 735	    !get_freq("clock-frequency", 1, &ppc_proc_freq)) {
 736
 737		printk(KERN_ERR "WARNING: Estimating processor frequency "
 738				"(not found)\n");
 739	}
 740}
 741
 742int update_persistent_clock64(struct timespec64 now)
 743{
 744	struct rtc_time tm;
 745
 746	if (!ppc_md.set_rtc_time)
 747		return -ENODEV;
 748
 749	rtc_time64_to_tm(now.tv_sec + 1 + timezone_offset, &tm);
 750
 751	return ppc_md.set_rtc_time(&tm);
 752}
 753
 754static void __read_persistent_clock(struct timespec64 *ts)
 755{
 756	struct rtc_time tm;
 757	static int first = 1;
 758
 759	ts->tv_nsec = 0;
 760	/* XXX this is a little fragile but will work okay in the short term */
 761	if (first) {
 762		first = 0;
 763		if (ppc_md.time_init)
 764			timezone_offset = ppc_md.time_init();
 765
 766		/* get_boot_time() isn't guaranteed to be safe to call late */
 767		if (ppc_md.get_boot_time) {
 768			ts->tv_sec = ppc_md.get_boot_time() - timezone_offset;
 769			return;
 770		}
 771	}
 772	if (!ppc_md.get_rtc_time) {
 773		ts->tv_sec = 0;
 774		return;
 775	}
 776	ppc_md.get_rtc_time(&tm);
 777
 778	ts->tv_sec = rtc_tm_to_time64(&tm);
 779}
 780
 781void read_persistent_clock64(struct timespec64 *ts)
 782{
 783	__read_persistent_clock(ts);
 784
 785	/* Sanitize it in case real time clock is set below EPOCH */
 786	if (ts->tv_sec < 0) {
 787		ts->tv_sec = 0;
 788		ts->tv_nsec = 0;
 789	}
 790		
 791}
 792
 793/* clocksource code */
 794static notrace u64 timebase_read(struct clocksource *cs)
 795{
 796	return (u64)get_tb();
 797}
 798
 799static void __init clocksource_init(void)
 800{
 801	struct clocksource *clock = &clocksource_timebase;
 802
 803	if (clocksource_register_hz(clock, tb_ticks_per_sec)) {
 804		printk(KERN_ERR "clocksource: %s is already registered\n",
 805		       clock->name);
 806		return;
 807	}
 808
 809	printk(KERN_INFO "clocksource: %s mult[%x] shift[%d] registered\n",
 810	       clock->name, clock->mult, clock->shift);
 811}
 812
 813static int decrementer_set_next_event(unsigned long evt,
 814				      struct clock_event_device *dev)
 815{
 816	__this_cpu_write(decrementers_next_tb, get_tb() + evt);
 817	set_dec_or_work(evt);
 818
 819	return 0;
 820}
 821
 822static int decrementer_shutdown(struct clock_event_device *dev)
 823{
 824	__this_cpu_write(decrementers_next_tb, DEC_CLOCKEVENT_STOPPED);
 825	set_dec_or_work(decrementer_max);
 826
 827	return 0;
 828}
 829
 830static void register_decrementer_clockevent(int cpu)
 831{
 832	struct clock_event_device *dec = &per_cpu(decrementers, cpu);
 833
 834	*dec = decrementer_clockevent;
 835	dec->cpumask = cpumask_of(cpu);
 836
 837	clockevents_config_and_register(dec, ppc_tb_freq, 2, decrementer_max);
 838
 839	printk_once(KERN_DEBUG "clockevent: %s mult[%x] shift[%d] cpu[%d]\n",
 840		    dec->name, dec->mult, dec->shift, cpu);
 841
 842	/* Set values for KVM, see kvm_emulate_dec() */
 843	decrementer_clockevent.mult = dec->mult;
 844	decrementer_clockevent.shift = dec->shift;
 845}
 846
 847static void enable_large_decrementer(void)
 848{
 849	if (!cpu_has_feature(CPU_FTR_ARCH_300))
 850		return;
 851
 852	if (decrementer_max <= DECREMENTER_DEFAULT_MAX)
 853		return;
 854
 855	/*
 856	 * If we're running as the hypervisor we need to enable the LD manually
 857	 * otherwise firmware should have done it for us.
 858	 */
 859	if (cpu_has_feature(CPU_FTR_HVMODE))
 860		mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_LD);
 861}
 862
 863static void __init set_decrementer_max(void)
 864{
 865	struct device_node *cpu;
 866	u32 bits = 32;
 867
 868	/* Prior to ISAv3 the decrementer is always 32 bit */
 869	if (!cpu_has_feature(CPU_FTR_ARCH_300))
 870		return;
 871
 872	cpu = of_find_node_by_type(NULL, "cpu");
 873
 874	if (of_property_read_u32(cpu, "ibm,dec-bits", &bits) == 0) {
 875		if (bits > 64 || bits < 32) {
 876			pr_warn("time_init: firmware supplied invalid ibm,dec-bits");
 877			bits = 32;
 878		}
 879
 880		/* calculate the signed maximum given this many bits */
 881		decrementer_max = (1ul << (bits - 1)) - 1;
 882	}
 883
 884	of_node_put(cpu);
 885
 886	pr_info("time_init: %u bit decrementer (max: %llx)\n",
 887		bits, decrementer_max);
 888}
 889
 890static void __init init_decrementer_clockevent(void)
 891{
 892	register_decrementer_clockevent(smp_processor_id());
 893}
 894
 895void secondary_cpu_time_init(void)
 896{
 897	/* Enable and test the large decrementer for this cpu */
 898	enable_large_decrementer();
 899
 900	/* Start the decrementer on CPUs that have manual control
 901	 * such as BookE
 902	 */
 903	start_cpu_decrementer();
 904
 905	/* FIME: Should make unrelated change to move snapshot_timebase
 906	 * call here ! */
 907	register_decrementer_clockevent(smp_processor_id());
 908}
 909
 910/*
 911 * Divide a 128-bit dividend by a 32-bit divisor, leaving a 128 bit
 912 * result.
 913 */
 914static __init void div128_by_32(u64 dividend_high, u64 dividend_low,
 915				unsigned int divisor, struct div_result *dr)
 916{
 917	unsigned long a, b, c, d;
 918	unsigned long w, x, y, z;
 919	u64 ra, rb, rc;
 920
 921	a = dividend_high >> 32;
 922	b = dividend_high & 0xffffffff;
 923	c = dividend_low >> 32;
 924	d = dividend_low & 0xffffffff;
 925
 926	w = a / divisor;
 927	ra = ((u64)(a - (w * divisor)) << 32) + b;
 928
 929	rb = ((u64)do_div(ra, divisor) << 32) + c;
 930	x = ra;
 931
 932	rc = ((u64)do_div(rb, divisor) << 32) + d;
 933	y = rb;
 934
 935	do_div(rc, divisor);
 936	z = rc;
 937
 938	dr->result_high = ((u64)w << 32) + x;
 939	dr->result_low  = ((u64)y << 32) + z;
 940}
 941
 942/* This function is only called on the boot processor */
 943void __init time_init(void)
 944{
 945	struct div_result res;
 946	u64 scale;
 947	unsigned shift;
 948
 949	/* Normal PowerPC with timebase register */
 950	if (ppc_md.calibrate_decr)
 951		ppc_md.calibrate_decr();
 952	else
 953		generic_calibrate_decr();
 954
 955	printk(KERN_DEBUG "time_init: decrementer frequency = %lu.%.6lu MHz\n",
 956	       ppc_tb_freq / 1000000, ppc_tb_freq % 1000000);
 957	printk(KERN_DEBUG "time_init: processor frequency   = %lu.%.6lu MHz\n",
 958	       ppc_proc_freq / 1000000, ppc_proc_freq % 1000000);
 959
 960	tb_ticks_per_jiffy = ppc_tb_freq / HZ;
 961	tb_ticks_per_sec = ppc_tb_freq;
 962	tb_ticks_per_usec = ppc_tb_freq / 1000000;
 963
 964	/*
 965	 * Compute scale factor for sched_clock.
 966	 * The calibrate_decr() function has set tb_ticks_per_sec,
 967	 * which is the timebase frequency.
 968	 * We compute 1e9 * 2^64 / tb_ticks_per_sec and interpret
 969	 * the 128-bit result as a 64.64 fixed-point number.
 970	 * We then shift that number right until it is less than 1.0,
 971	 * giving us the scale factor and shift count to use in
 972	 * sched_clock().
 973	 */
 974	div128_by_32(1000000000, 0, tb_ticks_per_sec, &res);
 975	scale = res.result_low;
 976	for (shift = 0; res.result_high != 0; ++shift) {
 977		scale = (scale >> 1) | (res.result_high << 63);
 978		res.result_high >>= 1;
 979	}
 980	tb_to_ns_scale = scale;
 981	tb_to_ns_shift = shift;
 982	/* Save the current timebase to pretty up CONFIG_PRINTK_TIME */
 983	boot_tb = get_tb();
 984
 985	/* If platform provided a timezone (pmac), we correct the time */
 986	if (timezone_offset) {
 987		sys_tz.tz_minuteswest = -timezone_offset / 60;
 988		sys_tz.tz_dsttime = 0;
 989	}
 990
 991	vdso_k_arch_data->tb_ticks_per_sec = tb_ticks_per_sec;
 992#ifdef CONFIG_PPC64_PROC_SYSTEMCFG
 993	systemcfg->tb_ticks_per_sec = tb_ticks_per_sec;
 994#endif
 995
 996	/* initialise and enable the large decrementer (if we have one) */
 997	set_decrementer_max();
 998	enable_large_decrementer();
 999
1000	/* Start the decrementer on CPUs that have manual control
1001	 * such as BookE
1002	 */
1003	start_cpu_decrementer();
1004
1005	/* Register the clocksource */
1006	clocksource_init();
1007
1008	init_decrementer_clockevent();
1009	tick_setup_hrtimer_broadcast();
1010
1011	of_clk_init(NULL);
1012	enable_sched_clock_irqtime();
1013}
1014
1015/* We don't need to calibrate delay, we use the CPU timebase for that */
1016void calibrate_delay(void)
1017{
1018	/* Some generic code (such as spinlock debug) use loops_per_jiffy
1019	 * as the number of __delay(1) in a jiffy, so make it so
1020	 */
1021	loops_per_jiffy = tb_ticks_per_jiffy;
1022}
1023
1024#if IS_ENABLED(CONFIG_RTC_DRV_GENERIC)
1025static int rtc_generic_get_time(struct device *dev, struct rtc_time *tm)
1026{
1027	ppc_md.get_rtc_time(tm);
1028	return 0;
1029}
1030
1031static int rtc_generic_set_time(struct device *dev, struct rtc_time *tm)
1032{
1033	if (!ppc_md.set_rtc_time)
1034		return -EOPNOTSUPP;
1035
1036	if (ppc_md.set_rtc_time(tm) < 0)
1037		return -EOPNOTSUPP;
1038
1039	return 0;
1040}
1041
1042static const struct rtc_class_ops rtc_generic_ops = {
1043	.read_time = rtc_generic_get_time,
1044	.set_time = rtc_generic_set_time,
1045};
1046
1047static int __init rtc_init(void)
1048{
1049	struct platform_device *pdev;
1050
1051	if (!ppc_md.get_rtc_time)
1052		return -ENODEV;
1053
1054	pdev = platform_device_register_data(NULL, "rtc-generic", -1,
1055					     &rtc_generic_ops,
1056					     sizeof(rtc_generic_ops));
1057
1058	return PTR_ERR_OR_ZERO(pdev);
1059}
1060
1061device_initcall(rtc_init);
1062#endif