arch/x86_64/kernel/time.c

   1 /*
   2  *  linux/arch/x86-64/kernel/time.c
   3  *
   4  *  "High Precision Event Timer" based timekeeping.
   5  *
   6  *  Copyright (c) 1991,1992,1995  Linus Torvalds
   7  *  Copyright (c) 1994  Alan Modra
   8  *  Copyright (c) 1995  Markus Kuhn
   9  *  Copyright (c) 1996  Ingo Molnar
  10  *  Copyright (c) 1998  Andrea Arcangeli
  11  *  Copyright (c) 2002,2006  Vojtech Pavlik
  12  *  Copyright (c) 2003  Andi Kleen
  13  *  RTC support code taken from arch/i386/kernel/timers/time_hpet.c
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/sched.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/init.h>
  20 #include <linux/mc146818rtc.h>
  21 #include <linux/time.h>
  22 #include <linux/ioport.h>
  23 #include <linux/module.h>
  24 #include <linux/device.h>
  25 #include <linux/sysdev.h>
  26 #include <linux/bcd.h>
  27 #include <linux/notifier.h>
  28 #include <linux/cpu.h>
  29 #include <linux/kallsyms.h>
  30 #include <linux/acpi.h>
  31 #ifdef CONFIG_ACPI
  32 #include <acpi/achware.h>       /* for PM timer frequency */
  33 #include <acpi/acpi_bus.h>
  34 #endif
  35 #include <asm/8253pit.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/vsyscall.h>
  38 #include <asm/timex.h>
  39 #include <asm/proto.h>
  40 #include <asm/hpet.h>
  41 #include <asm/sections.h>
  42 #include <linux/hpet.h>
  43 #include <asm/apic.h>
  44 #include <asm/hpet.h>
  45 #include <asm/mpspec.h>
  46 #include <asm/nmi.h>
  47 #include <asm/vgtod.h>
  48
  49 static char *timename = NULL;
  50
  51 DEFINE_SPINLOCK(rtc_lock);
  52 EXPORT_SYMBOL(rtc_lock);
  53 DEFINE_SPINLOCK(i8253_lock);
  54
  55 volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES;
  56
  57 unsigned long profile_pc(struct pt_regs *regs)
  58 {
  59         unsigned long pc = instruction_pointer(regs);
  60
  61         /* Assume the lock function has either no stack frame or a copy
  62            of eflags from PUSHF
  63            Eflags always has bits 22 and up cleared unlike kernel addresses. */
  64         if (!user_mode(regs) && in_lock_functions(pc)) {
  65                 unsigned long *sp = (unsigned long *)regs->rsp;
  66                 if (sp[0] >> 22)
  67                         return sp[0];
  68                 if (sp[1] >> 22)
  69                         return sp[1];
  70         }
  71         return pc;
  72 }
  73 EXPORT_SYMBOL(profile_pc);
  74
  75 /*
  76  * In order to set the CMOS clock precisely, set_rtc_mmss has to be called 500
  77  * ms after the second nowtime has started, because when nowtime is written
  78  * into the registers of the CMOS clock, it will jump to the next second
  79  * precisely 500 ms later. Check the Motorola MC146818A or Dallas DS12887 data
  80  * sheet for details.
  81  */
  82
  83 static int set_rtc_mmss(unsigned long nowtime)
  84 {
  85         int retval = 0;
  86         int real_seconds, real_minutes, cmos_minutes;
  87         unsigned char control, freq_select;
  88
  89 /*
  90  * IRQs are disabled when we're called from the timer interrupt,
  91  * no need for spin_lock_irqsave()
  92  */
  93
  94         spin_lock(&rtc_lock);
  95
  96 /*
  97  * Tell the clock it's being set and stop it.
  98  */
  99
 100         control = CMOS_READ(RTC_CONTROL);
 101         CMOS_WRITE(control | RTC_SET, RTC_CONTROL);
 102
 103         freq_select = CMOS_READ(RTC_FREQ_SELECT);
 104         CMOS_WRITE(freq_select | RTC_DIV_RESET2, RTC_FREQ_SELECT);
 105
 106         cmos_minutes = CMOS_READ(RTC_MINUTES);
 107                 BCD_TO_BIN(cmos_minutes);
 108
 109 /*
 110  * since we're only adjusting minutes and seconds, don't interfere with hour
 111  * overflow. This avoids messing with unknown time zones but requires your RTC
 112  * not to be off by more than 15 minutes. Since we're calling it only when
 113  * our clock is externally synchronized using NTP, this shouldn't be a problem.
 114  */
 115
 116         real_seconds = nowtime % 60;
 117         real_minutes = nowtime / 60;
 118         if (((abs(real_minutes - cmos_minutes) + 15) / 30) & 1)
 119                 real_minutes += 30;             /* correct for half hour time zone */
 120         real_minutes %= 60;
 121
 122         if (abs(real_minutes - cmos_minutes) >= 30) {
 123                 printk(KERN_WARNING "time.c: can't update CMOS clock "
 124                        "from %d to %d\n", cmos_minutes, real_minutes);
 125                 retval = -1;
 126         } else {
 127                 BIN_TO_BCD(real_seconds);
 128                 BIN_TO_BCD(real_minutes);
 129                 CMOS_WRITE(real_seconds, RTC_SECONDS);
 130                 CMOS_WRITE(real_minutes, RTC_MINUTES);
 131         }
 132
 133 /*
 134  * The following flags have to be released exactly in this order, otherwise the
 135  * DS12887 (popular MC146818A clone with integrated battery and quartz) will
 136  * not reset the oscillator and will not update precisely 500 ms later. You
 137  * won't find this mentioned in the Dallas Semiconductor data sheets, but who
 138  * believes data sheets anyway ... -- Markus Kuhn
 139  */
 140
 141         CMOS_WRITE(control, RTC_CONTROL);
 142         CMOS_WRITE(freq_select, RTC_FREQ_SELECT);
 143
 144         spin_unlock(&rtc_lock);
 145
 146         return retval;
 147 }
 148
 149 int update_persistent_clock(struct timespec now)
 150 {
 151         return set_rtc_mmss(now.tv_sec);
 152 }
 153
 154 void main_timer_handler(void)
 155 {
 156 /*
 157  * Here we are in the timer irq handler. We have irqs locally disabled (so we
 158  * don't need spin_lock_irqsave()) but we don't know if the timer_bh is running
 159  * on the other CPU, so we need a lock. We also need to lock the vsyscall
 160  * variables, because both do_timer() and us change them -arca+vojtech
 161  */
 162
 163         write_seqlock(&xtime_lock);
 164
 165 /*
 166  * Do the timer stuff.
 167  */
 168
 169         do_timer(1);
 170 #ifndef CONFIG_SMP
 171         update_process_times(user_mode(get_irq_regs()));
 172 #endif
 173
 174 /*
 175  * In the SMP case we use the local APIC timer interrupt to do the profiling,
 176  * except when we simulate SMP mode on a uniprocessor system, in that case we
 177  * have to call the local interrupt handler.
 178  */
 179
 180         if (!using_apic_timer)
 181                 smp_local_timer_interrupt();
 182
 183         write_sequnlock(&xtime_lock);
 184 }
 185
 186 static irqreturn_t timer_interrupt(int irq, void *dev_id)
 187 {
 188         if (apic_runs_main_timer > 1)
 189                 return IRQ_HANDLED;
 190         main_timer_handler();
 191         if (using_apic_timer)
 192                 smp_send_timer_broadcast_ipi();
 193         return IRQ_HANDLED;
 194 }
 195
 196 unsigned long read_persistent_clock(void)
 197 {
 198         unsigned int year, mon, day, hour, min, sec;
 199         unsigned long flags;
 200         unsigned century = 0;
 201
 202         spin_lock_irqsave(&rtc_lock, flags);
 203
 204         do {
 205                 sec = CMOS_READ(RTC_SECONDS);
 206                 min = CMOS_READ(RTC_MINUTES);
 207                 hour = CMOS_READ(RTC_HOURS);
 208                 day = CMOS_READ(RTC_DAY_OF_MONTH);
 209                 mon = CMOS_READ(RTC_MONTH);
 210                 year = CMOS_READ(RTC_YEAR);
 211 #ifdef CONFIG_ACPI
 212                 if (acpi_gbl_FADT.header.revision >= FADT2_REVISION_ID &&
 213                                         acpi_gbl_FADT.century)
 214                         century = CMOS_READ(acpi_gbl_FADT.century);
 215 #endif
 216         } while (sec != CMOS_READ(RTC_SECONDS));
 217
 218         spin_unlock_irqrestore(&rtc_lock, flags);
 219
 220         /*
 221          * We know that x86-64 always uses BCD format, no need to check the
 222          * config register.
 223          */
 224
 225         BCD_TO_BIN(sec);
 226         BCD_TO_BIN(min);
 227         BCD_TO_BIN(hour);
 228         BCD_TO_BIN(day);
 229         BCD_TO_BIN(mon);
 230         BCD_TO_BIN(year);
 231
 232         if (century) {
 233                 BCD_TO_BIN(century);
 234                 year += century * 100;
 235                 printk(KERN_INFO "Extended CMOS year: %d\n", century * 100);
 236         } else {
 237                 /*
 238                  * x86-64 systems only exists since 2002.
 239                  * This will work up to Dec 31, 2100
 240                  */
 241                 year += 2000;
 242         }
 243
 244         return mktime(year, mon, day, hour, min, sec);
 245 }
 246
 247 /* calibrate_cpu is used on systems with fixed rate TSCs to determine
 248  * processor frequency */
 249 #define TICK_COUNT 100000000
 250 static unsigned int __init tsc_calibrate_cpu_khz(void)
 251 {
 252        int tsc_start, tsc_now;
 253        int i, no_ctr_free;
 254        unsigned long evntsel3 = 0, pmc3 = 0, pmc_now = 0;
 255        unsigned long flags;
 256
 257        for (i = 0; i < 4; i++)
 258                if (avail_to_resrv_perfctr_nmi_bit(i))
 259                        break;
 260        no_ctr_free = (i == 4);
 261        if (no_ctr_free) {
 262                i = 3;
 263                rdmsrl(MSR_K7_EVNTSEL3, evntsel3);
 264                wrmsrl(MSR_K7_EVNTSEL3, 0);
 265                rdmsrl(MSR_K7_PERFCTR3, pmc3);
 266        } else {
 267                reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i);
 268                reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i);
 269        }
 270        local_irq_save(flags);
 271        /* start meauring cycles, incrementing from 0 */
 272        wrmsrl(MSR_K7_PERFCTR0 + i, 0);
 273        wrmsrl(MSR_K7_EVNTSEL0 + i, 1 << 22 | 3 << 16 | 0x76);
 274        rdtscl(tsc_start);
 275        do {
 276                rdmsrl(MSR_K7_PERFCTR0 + i, pmc_now);
 277                tsc_now = get_cycles_sync();
 278        } while ((tsc_now - tsc_start) < TICK_COUNT);
 279
 280        local_irq_restore(flags);
 281        if (no_ctr_free) {
 282                wrmsrl(MSR_K7_EVNTSEL3, 0);
 283                wrmsrl(MSR_K7_PERFCTR3, pmc3);
 284                wrmsrl(MSR_K7_EVNTSEL3, evntsel3);
 285        } else {
 286                release_perfctr_nmi(MSR_K7_PERFCTR0 + i);
 287                release_evntsel_nmi(MSR_K7_EVNTSEL0 + i);
 288        }
 289
 290        return pmc_now * tsc_khz / (tsc_now - tsc_start);
 291 }
 292
 293 /*
 294  * pit_calibrate_tsc() uses the speaker output (channel 2) of
 295  * the PIT. This is better than using the timer interrupt output,
 296  * because we can read the value of the speaker with just one inb(),
 297  * where we need three i/o operations for the interrupt channel.
 298  * We count how many ticks the TSC does in 50 ms.
 299  */
 300
 301 static unsigned int __init pit_calibrate_tsc(void)
 302 {
 303         unsigned long start, end;
 304         unsigned long flags;
 305
 306         spin_lock_irqsave(&i8253_lock, flags);
 307
 308         outb((inb(0x61) & ~0x02) | 0x01, 0x61);
 309
 310         outb(0xb0, 0x43);
 311         outb((PIT_TICK_RATE / (1000 / 50)) & 0xff, 0x42);
 312         outb((PIT_TICK_RATE / (1000 / 50)) >> 8, 0x42);
 313         start = get_cycles_sync();
 314         while ((inb(0x61) & 0x20) == 0);
 315         end = get_cycles_sync();
 316
 317         spin_unlock_irqrestore(&i8253_lock, flags);
 318
 319         return (end - start) / 50;
 320 }
 321
 322 #define PIT_MODE 0x43
 323 #define PIT_CH0  0x40
 324
 325 static void __pit_init(int val, u8 mode)
 326 {
 327         unsigned long flags;
 328
 329         spin_lock_irqsave(&i8253_lock, flags);
 330         outb_p(mode, PIT_MODE);
 331         outb_p(val & 0xff, PIT_CH0);    /* LSB */
 332         outb_p(val >> 8, PIT_CH0);      /* MSB */
 333         spin_unlock_irqrestore(&i8253_lock, flags);
 334 }
 335
 336 void __init pit_init(void)
 337 {
 338         __pit_init(LATCH, 0x34); /* binary, mode 2, LSB/MSB, ch 0 */
 339 }
 340
 341 void pit_stop_interrupt(void)
 342 {
 343         __pit_init(0, 0x30); /* mode 0 */
 344 }
 345
 346 void stop_timer_interrupt(void)
 347 {
 348         char *name;
 349         if (hpet_address) {
 350                 name = "HPET";
 351                 hpet_timer_stop_set_go(0);
 352         } else {
 353                 name = "PIT";
 354                 pit_stop_interrupt();
 355         }
 356         printk(KERN_INFO "timer: %s interrupt stopped.\n", name);
 357 }
 358
 359 static struct irqaction irq0 = {
 360         .handler        = timer_interrupt,
 361         .flags          = IRQF_DISABLED | IRQF_IRQPOLL,
 362         .mask           = CPU_MASK_NONE,
 363         .name           = "timer"
 364 };
 365
 366 void __init time_init(void)
 367 {
 368         if (nohpet)
 369                 hpet_address = 0;
 370
 371         if (hpet_arch_init())
 372                 hpet_address = 0;
 373
 374         if (hpet_use_timer) {
 375                 /* set tick_nsec to use the proper rate for HPET */
 376                 tick_nsec = TICK_NSEC_HPET;
 377                 tsc_khz = hpet_calibrate_tsc();
 378                 timename = "HPET";
 379         } else {
 380                 pit_init();
 381                 tsc_khz = pit_calibrate_tsc();
 382                 timename = "PIT";
 383         }
 384
 385         cpu_khz = tsc_khz;
 386         if (cpu_has(&boot_cpu_data, X86_FEATURE_CONSTANT_TSC) &&
 387                 boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 388                 boot_cpu_data.x86 == 16)
 389                 cpu_khz = tsc_calibrate_cpu_khz();
 390
 391         if (unsynchronized_tsc())
 392                 mark_tsc_unstable("TSCs unsynchronized");
 393
 394         if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
 395                 vgetcpu_mode = VGETCPU_RDTSCP;
 396         else
 397                 vgetcpu_mode = VGETCPU_LSL;
 398
 399         set_cyc2ns_scale(tsc_khz);
 400         printk(KERN_INFO "time.c: Detected %d.%03d MHz processor.\n",
 401                 cpu_khz / 1000, cpu_khz % 1000);
 402         init_tsc_clocksource();
 403
 404         setup_irq(0, &irq0);
 405 }
 406
 407 /*
 408  * sysfs support for the timer.
 409  */
 410
 411 static int timer_suspend(struct sys_device *dev, pm_message_t state)
 412 {
 413         return 0;
 414 }
 415
 416 static int timer_resume(struct sys_device *dev)
 417 {
 418         if (hpet_address)
 419                 hpet_reenable();
 420         else
 421                 i8254_timer_resume();
 422         return 0;
 423 }
 424
 425 static struct sysdev_class timer_sysclass = {
 426         .resume = timer_resume,
 427         .suspend = timer_suspend,
 428         set_kset_name("timer"),
 429 };
 430
 431 /* XXX this sysfs stuff should probably go elsewhere later -john */
 432 static struct sys_device device_timer = {
 433         .id     = 0,
 434         .cls    = &timer_sysclass,
 435 };
 436
 437 static int time_init_device(void)
 438 {
 439         int error = sysdev_class_register(&timer_sysclass);
 440         if (!error)
 441                 error = sysdev_register(&device_timer);
 442         return error;
 443 }
 444
 445 device_initcall(time_init_device);