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/cpufreq.h>
  43 #include <linux/hpet.h>
  44 #include <asm/apic.h>
  45 #include <asm/hpet.h>
  46
  47 extern void i8254_timer_resume(void);
  48 extern int using_apic_timer;
  49
  50 static char *timename = NULL;
  51
  52 DEFINE_SPINLOCK(rtc_lock);
  53 EXPORT_SYMBOL(rtc_lock);
  54 DEFINE_SPINLOCK(i8253_lock);
  55
  56 volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES;
  57
  58 unsigned long profile_pc(struct pt_regs *regs)
  59 {
  60         unsigned long pc = instruction_pointer(regs);
  61
  62         /* Assume the lock function has either no stack frame or a copy
  63            of eflags from PUSHF
  64            Eflags always has bits 22 and up cleared unlike kernel addresses. */
  65         if (!user_mode(regs) && in_lock_functions(pc)) {
  66                 unsigned long *sp = (unsigned long *)regs->rsp;
  67                 if (sp[0] >> 22)
  68                         return sp[0];
  69                 if (sp[1] >> 22)
  70                         return sp[1];
  71         }
  72         return pc;
  73 }
  74 EXPORT_SYMBOL(profile_pc);
  75
  76 /*
  77  * In order to set the CMOS clock precisely, set_rtc_mmss has to be called 500
  78  * ms after the second nowtime has started, because when nowtime is written
  79  * into the registers of the CMOS clock, it will jump to the next second
  80  * precisely 500 ms later. Check the Motorola MC146818A or Dallas DS12887 data
  81  * sheet for details.
  82  */
  83
  84 static void set_rtc_mmss(unsigned long nowtime)
  85 {
  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         } else {
 126                 BIN_TO_BCD(real_seconds);
 127                 BIN_TO_BCD(real_minutes);
 128                 CMOS_WRITE(real_seconds, RTC_SECONDS);
 129                 CMOS_WRITE(real_minutes, RTC_MINUTES);
 130         }
 131
 132 /*
 133  * The following flags have to be released exactly in this order, otherwise the
 134  * DS12887 (popular MC146818A clone with integrated battery and quartz) will
 135  * not reset the oscillator and will not update precisely 500 ms later. You
 136  * won't find this mentioned in the Dallas Semiconductor data sheets, but who
 137  * believes data sheets anyway ... -- Markus Kuhn
 138  */
 139
 140         CMOS_WRITE(control, RTC_CONTROL);
 141         CMOS_WRITE(freq_select, RTC_FREQ_SELECT);
 142
 143         spin_unlock(&rtc_lock);
 144 }
 145
 146
 147 void main_timer_handler(void)
 148 {
 149         static unsigned long rtc_update = 0;
 150 /*
 151  * Here we are in the timer irq handler. We have irqs locally disabled (so we
 152  * don't need spin_lock_irqsave()) but we don't know if the timer_bh is running
 153  * on the other CPU, so we need a lock. We also need to lock the vsyscall
 154  * variables, because both do_timer() and us change them -arca+vojtech
 155  */
 156
 157         write_seqlock(&xtime_lock);
 158
 159 /*
 160  * Do the timer stuff.
 161  */
 162
 163         do_timer(1);
 164 #ifndef CONFIG_SMP
 165         update_process_times(user_mode(get_irq_regs()));
 166 #endif
 167
 168 /*
 169  * In the SMP case we use the local APIC timer interrupt to do the profiling,
 170  * except when we simulate SMP mode on a uniprocessor system, in that case we
 171  * have to call the local interrupt handler.
 172  */
 173
 174         if (!using_apic_timer)
 175                 smp_local_timer_interrupt();
 176
 177 /*
 178  * If we have an externally synchronized Linux clock, then update CMOS clock
 179  * accordingly every ~11 minutes. set_rtc_mmss() will be called in the jiffy
 180  * closest to exactly 500 ms before the next second. If the update fails, we
 181  * don't care, as it'll be updated on the next turn, and the problem (time way
 182  * off) isn't likely to go away much sooner anyway.
 183  */
 184
 185         if (ntp_synced() && xtime.tv_sec > rtc_update &&
 186                 abs(xtime.tv_nsec - 500000000) <= tick_nsec / 2) {
 187                 set_rtc_mmss(xtime.tv_sec);
 188                 rtc_update = xtime.tv_sec + 660;
 189         }
 190
 191         write_sequnlock(&xtime_lock);
 192 }
 193
 194 static irqreturn_t timer_interrupt(int irq, void *dev_id)
 195 {
 196         if (apic_runs_main_timer > 1)
 197                 return IRQ_HANDLED;
 198         main_timer_handler();
 199         if (using_apic_timer)
 200                 smp_send_timer_broadcast_ipi();
 201         return IRQ_HANDLED;
 202 }
 203
 204 static unsigned long get_cmos_time(void)
 205 {
 206         unsigned int year, mon, day, hour, min, sec;
 207         unsigned long flags;
 208         unsigned century = 0;
 209
 210         spin_lock_irqsave(&rtc_lock, flags);
 211
 212         do {
 213                 sec = CMOS_READ(RTC_SECONDS);
 214                 min = CMOS_READ(RTC_MINUTES);
 215                 hour = CMOS_READ(RTC_HOURS);
 216                 day = CMOS_READ(RTC_DAY_OF_MONTH);
 217                 mon = CMOS_READ(RTC_MONTH);
 218                 year = CMOS_READ(RTC_YEAR);
 219 #ifdef CONFIG_ACPI
 220                 if (acpi_gbl_FADT.header.revision >= FADT2_REVISION_ID &&
 221                                         acpi_gbl_FADT.century)
 222                         century = CMOS_READ(acpi_gbl_FADT.century);
 223 #endif
 224         } while (sec != CMOS_READ(RTC_SECONDS));
 225
 226         spin_unlock_irqrestore(&rtc_lock, flags);
 227
 228         /*
 229          * We know that x86-64 always uses BCD format, no need to check the
 230          * config register.
 231          */
 232
 233         BCD_TO_BIN(sec);
 234         BCD_TO_BIN(min);
 235         BCD_TO_BIN(hour);
 236         BCD_TO_BIN(day);
 237         BCD_TO_BIN(mon);
 238         BCD_TO_BIN(year);
 239
 240         if (century) {
 241                 BCD_TO_BIN(century);
 242                 year += century * 100;
 243                 printk(KERN_INFO "Extended CMOS year: %d\n", century * 100);
 244         } else {
 245                 /*
 246                  * x86-64 systems only exists since 2002.
 247                  * This will work up to Dec 31, 2100
 248                  */
 249                 year += 2000;
 250         }
 251
 252         return mktime(year, mon, day, hour, min, sec);
 253 }
 254
 255
 256 /*
 257  * pit_calibrate_tsc() uses the speaker output (channel 2) of
 258  * the PIT. This is better than using the timer interrupt output,
 259  * because we can read the value of the speaker with just one inb(),
 260  * where we need three i/o operations for the interrupt channel.
 261  * We count how many ticks the TSC does in 50 ms.
 262  */
 263
 264 static unsigned int __init pit_calibrate_tsc(void)
 265 {
 266         unsigned long start, end;
 267         unsigned long flags;
 268
 269         spin_lock_irqsave(&i8253_lock, flags);
 270
 271         outb((inb(0x61) & ~0x02) | 0x01, 0x61);
 272
 273         outb(0xb0, 0x43);
 274         outb((PIT_TICK_RATE / (1000 / 50)) & 0xff, 0x42);
 275         outb((PIT_TICK_RATE / (1000 / 50)) >> 8, 0x42);
 276         start = get_cycles_sync();
 277         while ((inb(0x61) & 0x20) == 0);
 278         end = get_cycles_sync();
 279
 280         spin_unlock_irqrestore(&i8253_lock, flags);
 281
 282         return (end - start) / 50;
 283 }
 284
 285 #define PIT_MODE 0x43
 286 #define PIT_CH0  0x40
 287
 288 static void __init __pit_init(int val, u8 mode)
 289 {
 290         unsigned long flags;
 291
 292         spin_lock_irqsave(&i8253_lock, flags);
 293         outb_p(mode, PIT_MODE);
 294         outb_p(val & 0xff, PIT_CH0);    /* LSB */
 295         outb_p(val >> 8, PIT_CH0);      /* MSB */
 296         spin_unlock_irqrestore(&i8253_lock, flags);
 297 }
 298
 299 void __init pit_init(void)
 300 {
 301         __pit_init(LATCH, 0x34); /* binary, mode 2, LSB/MSB, ch 0 */
 302 }
 303
 304 void __init pit_stop_interrupt(void)
 305 {
 306         __pit_init(0, 0x30); /* mode 0 */
 307 }
 308
 309 void __init stop_timer_interrupt(void)
 310 {
 311         char *name;
 312         if (hpet_address) {
 313                 name = "HPET";
 314                 hpet_timer_stop_set_go(0);
 315         } else {
 316                 name = "PIT";
 317                 pit_stop_interrupt();
 318         }
 319         printk(KERN_INFO "timer: %s interrupt stopped.\n", name);
 320 }
 321
 322 static struct irqaction irq0 = {
 323         timer_interrupt, IRQF_DISABLED, CPU_MASK_NONE, "timer", NULL, NULL
 324 };
 325
 326 void __init time_init(void)
 327 {
 328         if (nohpet)
 329                 hpet_address = 0;
 330         xtime.tv_sec = get_cmos_time();
 331         xtime.tv_nsec = 0;
 332
 333         set_normalized_timespec(&wall_to_monotonic,
 334                                 -xtime.tv_sec, -xtime.tv_nsec);
 335
 336         if (hpet_arch_init())
 337                 hpet_address = 0;
 338
 339         if (hpet_use_timer) {
 340                 /* set tick_nsec to use the proper rate for HPET */
 341                 tick_nsec = TICK_NSEC_HPET;
 342                 cpu_khz = hpet_calibrate_tsc();
 343                 timename = "HPET";
 344         } else {
 345                 pit_init();
 346                 cpu_khz = pit_calibrate_tsc();
 347                 timename = "PIT";
 348         }
 349
 350         if (unsynchronized_tsc())
 351                 mark_tsc_unstable();
 352
 353         if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
 354                 vgetcpu_mode = VGETCPU_RDTSCP;
 355         else
 356                 vgetcpu_mode = VGETCPU_LSL;
 357
 358         set_cyc2ns_scale(cpu_khz);
 359         printk(KERN_INFO "time.c: Detected %d.%03d MHz processor.\n",
 360                 cpu_khz / 1000, cpu_khz % 1000);
 361         setup_irq(0, &irq0);
 362 }
 363
 364
 365 static long clock_cmos_diff;
 366 static unsigned long sleep_start;
 367
 368 /*
 369  * sysfs support for the timer.
 370  */
 371
 372 static int timer_suspend(struct sys_device *dev, pm_message_t state)
 373 {
 374         /*
 375          * Estimate time zone so that set_time can update the clock
 376          */
 377         long cmos_time =  get_cmos_time();
 378
 379         clock_cmos_diff = -cmos_time;
 380         clock_cmos_diff += get_seconds();
 381         sleep_start = cmos_time;
 382         return 0;
 383 }
 384
 385 static int timer_resume(struct sys_device *dev)
 386 {
 387         unsigned long flags;
 388         unsigned long sec;
 389         unsigned long ctime = get_cmos_time();
 390         long sleep_length = (ctime - sleep_start) * HZ;
 391
 392         if (sleep_length < 0) {
 393                 printk(KERN_WARNING "Time skew detected in timer resume!\n");
 394                 /* The time after the resume must not be earlier than the time
 395                  * before the suspend or some nasty things will happen
 396                  */
 397                 sleep_length = 0;
 398                 ctime = sleep_start;
 399         }
 400         if (hpet_address)
 401                 hpet_reenable();
 402         else
 403                 i8254_timer_resume();
 404
 405         sec = ctime + clock_cmos_diff;
 406         write_seqlock_irqsave(&xtime_lock,flags);
 407         xtime.tv_sec = sec;
 408         xtime.tv_nsec = 0;
 409         jiffies += sleep_length;
 410         write_sequnlock_irqrestore(&xtime_lock,flags);
 411         touch_softlockup_watchdog();
 412         return 0;
 413 }
 414
 415 static struct sysdev_class timer_sysclass = {
 416         .resume = timer_resume,
 417         .suspend = timer_suspend,
 418         set_kset_name("timer"),
 419 };
 420
 421 /* XXX this sysfs stuff should probably go elsewhere later -john */
 422 static struct sys_device device_timer = {
 423         .id     = 0,
 424         .cls    = &timer_sysclass,
 425 };
 426
 427 static int time_init_device(void)
 428 {
 429         int error = sysdev_class_register(&timer_sysclass);
 430         if (!error)
 431                 error = sysdev_register(&device_timer);
 432         return error;
 433 }
 434
 435 device_initcall(time_init_device);