Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
2f82bde4 | 11 | * Copyright (c) 2002,2006 Vojtech Pavlik |
1da177e4 LT |
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> | |
1da177e4 LT |
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> | |
a670fad0 VP |
27 | #include <linux/notifier.h> |
28 | #include <linux/cpu.h> | |
1da177e4 | 29 | #include <linux/kallsyms.h> |
312df5f1 | 30 | #include <linux/acpi.h> |
8d916406 | 31 | #ifdef CONFIG_ACPI |
312df5f1 | 32 | #include <acpi/achware.h> /* for PM timer frequency */ |
0e5f61b0 | 33 | #include <acpi/acpi_bus.h> |
8d916406 | 34 | #endif |
1da177e4 LT |
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> | |
1da177e4 | 44 | #include <asm/apic.h> |
1da177e4 | 45 | |
1da177e4 LT |
46 | #ifdef CONFIG_CPU_FREQ |
47 | static void cpufreq_delayed_get(void); | |
48 | #endif | |
49 | extern void i8254_timer_resume(void); | |
50 | extern int using_apic_timer; | |
51 | ||
a670fad0 | 52 | static char *timename = NULL; |
e8b91777 | 53 | |
1da177e4 | 54 | DEFINE_SPINLOCK(rtc_lock); |
2ee60e17 | 55 | EXPORT_SYMBOL(rtc_lock); |
1da177e4 LT |
56 | DEFINE_SPINLOCK(i8253_lock); |
57 | ||
73dea47f | 58 | int nohpet __initdata = 0; |
1da177e4 LT |
59 | static int notsc __initdata = 0; |
60 | ||
42211338 VP |
61 | #define USEC_PER_TICK (USEC_PER_SEC / HZ) |
62 | #define NSEC_PER_TICK (NSEC_PER_SEC / HZ) | |
63 | #define FSEC_PER_TICK (FSEC_PER_SEC / HZ) | |
64 | ||
65 | #define NS_SCALE 10 /* 2^10, carefully chosen */ | |
66 | #define US_SCALE 32 /* 2^32, arbitralrily chosen */ | |
67 | ||
1da177e4 | 68 | unsigned int cpu_khz; /* TSC clocks / usec, not used here */ |
2ee60e17 | 69 | EXPORT_SYMBOL(cpu_khz); |
1da177e4 LT |
70 | static unsigned long hpet_period; /* fsecs / HPET clock */ |
71 | unsigned long hpet_tick; /* HPET clocks / interrupt */ | |
33042a9f | 72 | int hpet_use_timer; /* Use counter of hpet for time keeping, otherwise PIT */ |
1da177e4 LT |
73 | unsigned long vxtime_hz = PIT_TICK_RATE; |
74 | int report_lost_ticks; /* command line option */ | |
75 | unsigned long long monotonic_base; | |
76 | ||
77 | struct vxtime_data __vxtime __section_vxtime; /* for vsyscalls */ | |
78 | ||
79 | volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES; | |
1da177e4 LT |
80 | struct timespec __xtime __section_xtime; |
81 | struct timezone __sys_tz __section_sys_tz; | |
82 | ||
1da177e4 LT |
83 | /* |
84 | * do_gettimeoffset() returns microseconds since last timer interrupt was | |
85 | * triggered by hardware. A memory read of HPET is slower than a register read | |
86 | * of TSC, but much more reliable. It's also synchronized to the timer | |
87 | * interrupt. Note that do_gettimeoffset() may return more than hpet_tick, if a | |
88 | * timer interrupt has happened already, but vxtime.trigger wasn't updated yet. | |
89 | * This is not a problem, because jiffies hasn't updated either. They are bound | |
90 | * together by xtime_lock. | |
91 | */ | |
92 | ||
93 | static inline unsigned int do_gettimeoffset_tsc(void) | |
94 | { | |
95 | unsigned long t; | |
96 | unsigned long x; | |
c818a181 | 97 | t = get_cycles_sync(); |
7351c0bf AK |
98 | if (t < vxtime.last_tsc) |
99 | t = vxtime.last_tsc; /* hack */ | |
42211338 | 100 | x = ((t - vxtime.last_tsc) * vxtime.tsc_quot) >> US_SCALE; |
1da177e4 LT |
101 | return x; |
102 | } | |
103 | ||
104 | static inline unsigned int do_gettimeoffset_hpet(void) | |
105 | { | |
a3a00751 | 106 | /* cap counter read to one tick to avoid inconsistencies */ |
107 | unsigned long counter = hpet_readl(HPET_COUNTER) - vxtime.last; | |
42211338 | 108 | return (min(counter,hpet_tick) * vxtime.quot) >> US_SCALE; |
1da177e4 LT |
109 | } |
110 | ||
111 | unsigned int (*do_gettimeoffset)(void) = do_gettimeoffset_tsc; | |
112 | ||
113 | /* | |
114 | * This version of gettimeofday() has microsecond resolution and better than | |
115 | * microsecond precision, as we're using at least a 10 MHz (usually 14.31818 | |
116 | * MHz) HPET timer. | |
117 | */ | |
118 | ||
119 | void do_gettimeofday(struct timeval *tv) | |
120 | { | |
8ef38609 | 121 | unsigned long seq; |
1da177e4 LT |
122 | unsigned int sec, usec; |
123 | ||
124 | do { | |
125 | seq = read_seqbegin(&xtime_lock); | |
126 | ||
127 | sec = xtime.tv_sec; | |
42211338 | 128 | usec = xtime.tv_nsec / NSEC_PER_USEC; |
1da177e4 LT |
129 | |
130 | /* i386 does some correction here to keep the clock | |
131 | monotonous even when ntpd is fixing drift. | |
132 | But they didn't work for me, there is a non monotonic | |
133 | clock anyways with ntp. | |
134 | I dropped all corrections now until a real solution can | |
135 | be found. Note when you fix it here you need to do the same | |
136 | in arch/x86_64/kernel/vsyscall.c and export all needed | |
137 | variables in vmlinux.lds. -AK */ | |
8ef38609 | 138 | usec += do_gettimeoffset(); |
1da177e4 LT |
139 | |
140 | } while (read_seqretry(&xtime_lock, seq)); | |
141 | ||
42211338 VP |
142 | tv->tv_sec = sec + usec / USEC_PER_SEC; |
143 | tv->tv_usec = usec % USEC_PER_SEC; | |
1da177e4 LT |
144 | } |
145 | ||
146 | EXPORT_SYMBOL(do_gettimeofday); | |
147 | ||
148 | /* | |
149 | * settimeofday() first undoes the correction that gettimeofday would do | |
150 | * on the time, and then saves it. This is ugly, but has been like this for | |
151 | * ages already. | |
152 | */ | |
153 | ||
154 | int do_settimeofday(struct timespec *tv) | |
155 | { | |
156 | time_t wtm_sec, sec = tv->tv_sec; | |
157 | long wtm_nsec, nsec = tv->tv_nsec; | |
158 | ||
159 | if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) | |
160 | return -EINVAL; | |
161 | ||
162 | write_seqlock_irq(&xtime_lock); | |
163 | ||
8ef38609 | 164 | nsec -= do_gettimeoffset() * NSEC_PER_USEC; |
1da177e4 LT |
165 | |
166 | wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec); | |
167 | wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec); | |
168 | ||
169 | set_normalized_timespec(&xtime, sec, nsec); | |
170 | set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec); | |
171 | ||
b149ee22 | 172 | ntp_clear(); |
1da177e4 LT |
173 | |
174 | write_sequnlock_irq(&xtime_lock); | |
175 | clock_was_set(); | |
176 | return 0; | |
177 | } | |
178 | ||
179 | EXPORT_SYMBOL(do_settimeofday); | |
180 | ||
181 | unsigned long profile_pc(struct pt_regs *regs) | |
182 | { | |
183 | unsigned long pc = instruction_pointer(regs); | |
184 | ||
31679f38 AK |
185 | /* Assume the lock function has either no stack frame or a copy |
186 | of eflags from PUSHF | |
187 | Eflags always has bits 22 and up cleared unlike kernel addresses. */ | |
d5a26017 | 188 | if (!user_mode(regs) && in_lock_functions(pc)) { |
31679f38 AK |
189 | unsigned long *sp = (unsigned long *)regs->rsp; |
190 | if (sp[0] >> 22) | |
191 | return sp[0]; | |
192 | if (sp[1] >> 22) | |
193 | return sp[1]; | |
1da177e4 LT |
194 | } |
195 | return pc; | |
196 | } | |
197 | EXPORT_SYMBOL(profile_pc); | |
198 | ||
199 | /* | |
200 | * In order to set the CMOS clock precisely, set_rtc_mmss has to be called 500 | |
201 | * ms after the second nowtime has started, because when nowtime is written | |
202 | * into the registers of the CMOS clock, it will jump to the next second | |
203 | * precisely 500 ms later. Check the Motorola MC146818A or Dallas DS12887 data | |
204 | * sheet for details. | |
205 | */ | |
206 | ||
207 | static void set_rtc_mmss(unsigned long nowtime) | |
208 | { | |
209 | int real_seconds, real_minutes, cmos_minutes; | |
210 | unsigned char control, freq_select; | |
211 | ||
212 | /* | |
213 | * IRQs are disabled when we're called from the timer interrupt, | |
214 | * no need for spin_lock_irqsave() | |
215 | */ | |
216 | ||
217 | spin_lock(&rtc_lock); | |
218 | ||
219 | /* | |
220 | * Tell the clock it's being set and stop it. | |
221 | */ | |
222 | ||
223 | control = CMOS_READ(RTC_CONTROL); | |
224 | CMOS_WRITE(control | RTC_SET, RTC_CONTROL); | |
225 | ||
226 | freq_select = CMOS_READ(RTC_FREQ_SELECT); | |
227 | CMOS_WRITE(freq_select | RTC_DIV_RESET2, RTC_FREQ_SELECT); | |
228 | ||
229 | cmos_minutes = CMOS_READ(RTC_MINUTES); | |
230 | BCD_TO_BIN(cmos_minutes); | |
231 | ||
232 | /* | |
233 | * since we're only adjusting minutes and seconds, don't interfere with hour | |
234 | * overflow. This avoids messing with unknown time zones but requires your RTC | |
235 | * not to be off by more than 15 minutes. Since we're calling it only when | |
236 | * our clock is externally synchronized using NTP, this shouldn't be a problem. | |
237 | */ | |
238 | ||
239 | real_seconds = nowtime % 60; | |
240 | real_minutes = nowtime / 60; | |
241 | if (((abs(real_minutes - cmos_minutes) + 15) / 30) & 1) | |
242 | real_minutes += 30; /* correct for half hour time zone */ | |
243 | real_minutes %= 60; | |
244 | ||
1da177e4 LT |
245 | if (abs(real_minutes - cmos_minutes) >= 30) { |
246 | printk(KERN_WARNING "time.c: can't update CMOS clock " | |
247 | "from %d to %d\n", cmos_minutes, real_minutes); | |
28456ede | 248 | } else { |
0b91317e AK |
249 | BIN_TO_BCD(real_seconds); |
250 | BIN_TO_BCD(real_minutes); | |
1da177e4 LT |
251 | CMOS_WRITE(real_seconds, RTC_SECONDS); |
252 | CMOS_WRITE(real_minutes, RTC_MINUTES); | |
253 | } | |
254 | ||
255 | /* | |
256 | * The following flags have to be released exactly in this order, otherwise the | |
257 | * DS12887 (popular MC146818A clone with integrated battery and quartz) will | |
258 | * not reset the oscillator and will not update precisely 500 ms later. You | |
259 | * won't find this mentioned in the Dallas Semiconductor data sheets, but who | |
260 | * believes data sheets anyway ... -- Markus Kuhn | |
261 | */ | |
262 | ||
263 | CMOS_WRITE(control, RTC_CONTROL); | |
264 | CMOS_WRITE(freq_select, RTC_FREQ_SELECT); | |
265 | ||
266 | spin_unlock(&rtc_lock); | |
267 | } | |
268 | ||
269 | ||
270 | /* monotonic_clock(): returns # of nanoseconds passed since time_init() | |
271 | * Note: This function is required to return accurate | |
272 | * time even in the absence of multiple timer ticks. | |
273 | */ | |
cbf9b4bb | 274 | static inline unsigned long long cycles_2_ns(unsigned long long cyc); |
1da177e4 LT |
275 | unsigned long long monotonic_clock(void) |
276 | { | |
277 | unsigned long seq; | |
278 | u32 last_offset, this_offset, offset; | |
279 | unsigned long long base; | |
280 | ||
281 | if (vxtime.mode == VXTIME_HPET) { | |
282 | do { | |
283 | seq = read_seqbegin(&xtime_lock); | |
284 | ||
285 | last_offset = vxtime.last; | |
286 | base = monotonic_base; | |
a3a00751 | 287 | this_offset = hpet_readl(HPET_COUNTER); |
1da177e4 LT |
288 | } while (read_seqretry(&xtime_lock, seq)); |
289 | offset = (this_offset - last_offset); | |
42211338 | 290 | offset *= NSEC_PER_TICK / hpet_tick; |
0b91317e | 291 | } else { |
1da177e4 LT |
292 | do { |
293 | seq = read_seqbegin(&xtime_lock); | |
294 | ||
295 | last_offset = vxtime.last_tsc; | |
296 | base = monotonic_base; | |
297 | } while (read_seqretry(&xtime_lock, seq)); | |
c818a181 | 298 | this_offset = get_cycles_sync(); |
cbf9b4bb | 299 | offset = cycles_2_ns(this_offset - last_offset); |
1da177e4 | 300 | } |
7351c0bf | 301 | return base + offset; |
1da177e4 LT |
302 | } |
303 | EXPORT_SYMBOL(monotonic_clock); | |
304 | ||
7d12e780 | 305 | static noinline void handle_lost_ticks(int lost) |
1da177e4 | 306 | { |
7351c0bf AK |
307 | static long lost_count; |
308 | static int warned; | |
309 | if (report_lost_ticks) { | |
310 | printk(KERN_WARNING "time.c: Lost %d timer tick(s)! ", lost); | |
7d12e780 | 311 | print_symbol("rip %s)\n", get_irq_regs()->rip); |
7351c0bf AK |
312 | } |
313 | ||
314 | if (lost_count == 1000 && !warned) { | |
315 | printk(KERN_WARNING "warning: many lost ticks.\n" | |
316 | KERN_WARNING "Your time source seems to be instable or " | |
1da177e4 | 317 | "some driver is hogging interupts\n"); |
7d12e780 | 318 | print_symbol("rip %s\n", get_irq_regs()->rip); |
7351c0bf AK |
319 | if (vxtime.mode == VXTIME_TSC && vxtime.hpet_address) { |
320 | printk(KERN_WARNING "Falling back to HPET\n"); | |
321 | if (hpet_use_timer) | |
322 | vxtime.last = hpet_readl(HPET_T0_CMP) - | |
323 | hpet_tick; | |
324 | else | |
325 | vxtime.last = hpet_readl(HPET_COUNTER); | |
326 | vxtime.mode = VXTIME_HPET; | |
327 | do_gettimeoffset = do_gettimeoffset_hpet; | |
328 | } | |
329 | /* else should fall back to PIT, but code missing. */ | |
330 | warned = 1; | |
331 | } else | |
332 | lost_count++; | |
1da177e4 LT |
333 | |
334 | #ifdef CONFIG_CPU_FREQ | |
7351c0bf AK |
335 | /* In some cases the CPU can change frequency without us noticing |
336 | Give cpufreq a change to catch up. */ | |
337 | if ((lost_count+1) % 25 == 0) | |
338 | cpufreq_delayed_get(); | |
1da177e4 LT |
339 | #endif |
340 | } | |
341 | ||
7d12e780 | 342 | void main_timer_handler(void) |
1da177e4 LT |
343 | { |
344 | static unsigned long rtc_update = 0; | |
345 | unsigned long tsc; | |
9ede6b09 | 346 | int delay = 0, offset = 0, lost = 0; |
1da177e4 LT |
347 | |
348 | /* | |
349 | * Here we are in the timer irq handler. We have irqs locally disabled (so we | |
350 | * don't need spin_lock_irqsave()) but we don't know if the timer_bh is running | |
351 | * on the other CPU, so we need a lock. We also need to lock the vsyscall | |
352 | * variables, because both do_timer() and us change them -arca+vojtech | |
353 | */ | |
354 | ||
355 | write_seqlock(&xtime_lock); | |
356 | ||
a3a00751 | 357 | if (vxtime.hpet_address) |
358 | offset = hpet_readl(HPET_COUNTER); | |
359 | ||
360 | if (hpet_use_timer) { | |
361 | /* if we're using the hpet timer functionality, | |
362 | * we can more accurately know the counter value | |
363 | * when the timer interrupt occured. | |
364 | */ | |
1da177e4 LT |
365 | offset = hpet_readl(HPET_T0_CMP) - hpet_tick; |
366 | delay = hpet_readl(HPET_COUNTER) - offset; | |
9ede6b09 | 367 | } else if (!pmtmr_ioport) { |
1da177e4 LT |
368 | spin_lock(&i8253_lock); |
369 | outb_p(0x00, 0x43); | |
370 | delay = inb_p(0x40); | |
371 | delay |= inb(0x40) << 8; | |
372 | spin_unlock(&i8253_lock); | |
373 | delay = LATCH - 1 - delay; | |
374 | } | |
375 | ||
c818a181 | 376 | tsc = get_cycles_sync(); |
1da177e4 LT |
377 | |
378 | if (vxtime.mode == VXTIME_HPET) { | |
379 | if (offset - vxtime.last > hpet_tick) { | |
380 | lost = (offset - vxtime.last) / hpet_tick - 1; | |
381 | } | |
382 | ||
383 | monotonic_base += | |
42211338 | 384 | (offset - vxtime.last) * NSEC_PER_TICK / hpet_tick; |
1da177e4 LT |
385 | |
386 | vxtime.last = offset; | |
312df5f1 AK |
387 | #ifdef CONFIG_X86_PM_TIMER |
388 | } else if (vxtime.mode == VXTIME_PMTMR) { | |
389 | lost = pmtimer_mark_offset(); | |
390 | #endif | |
1da177e4 LT |
391 | } else { |
392 | offset = (((tsc - vxtime.last_tsc) * | |
42211338 | 393 | vxtime.tsc_quot) >> US_SCALE) - USEC_PER_TICK; |
1da177e4 LT |
394 | |
395 | if (offset < 0) | |
396 | offset = 0; | |
397 | ||
42211338 VP |
398 | if (offset > USEC_PER_TICK) { |
399 | lost = offset / USEC_PER_TICK; | |
400 | offset %= USEC_PER_TICK; | |
1da177e4 LT |
401 | } |
402 | ||
cbf9b4bb | 403 | monotonic_base += cycles_2_ns(tsc - vxtime.last_tsc); |
1da177e4 LT |
404 | |
405 | vxtime.last_tsc = tsc - vxtime.quot * delay / vxtime.tsc_quot; | |
406 | ||
407 | if ((((tsc - vxtime.last_tsc) * | |
42211338 | 408 | vxtime.tsc_quot) >> US_SCALE) < offset) |
1da177e4 | 409 | vxtime.last_tsc = tsc - |
42211338 | 410 | (((long) offset << US_SCALE) / vxtime.tsc_quot) - 1; |
1da177e4 LT |
411 | } |
412 | ||
3171a030 | 413 | if (lost > 0) |
7d12e780 | 414 | handle_lost_ticks(lost); |
3171a030 AN |
415 | else |
416 | lost = 0; | |
1da177e4 LT |
417 | |
418 | /* | |
419 | * Do the timer stuff. | |
420 | */ | |
421 | ||
3171a030 | 422 | do_timer(lost + 1); |
1da177e4 | 423 | #ifndef CONFIG_SMP |
7d12e780 | 424 | update_process_times(user_mode(get_irq_regs())); |
1da177e4 LT |
425 | #endif |
426 | ||
427 | /* | |
428 | * In the SMP case we use the local APIC timer interrupt to do the profiling, | |
429 | * except when we simulate SMP mode on a uniprocessor system, in that case we | |
430 | * have to call the local interrupt handler. | |
431 | */ | |
432 | ||
1da177e4 | 433 | if (!using_apic_timer) |
7d12e780 | 434 | smp_local_timer_interrupt(); |
1da177e4 LT |
435 | |
436 | /* | |
437 | * If we have an externally synchronized Linux clock, then update CMOS clock | |
438 | * accordingly every ~11 minutes. set_rtc_mmss() will be called in the jiffy | |
439 | * closest to exactly 500 ms before the next second. If the update fails, we | |
440 | * don't care, as it'll be updated on the next turn, and the problem (time way | |
441 | * off) isn't likely to go away much sooner anyway. | |
442 | */ | |
443 | ||
b149ee22 | 444 | if (ntp_synced() && xtime.tv_sec > rtc_update && |
1da177e4 LT |
445 | abs(xtime.tv_nsec - 500000000) <= tick_nsec / 2) { |
446 | set_rtc_mmss(xtime.tv_sec); | |
447 | rtc_update = xtime.tv_sec + 660; | |
448 | } | |
449 | ||
450 | write_sequnlock(&xtime_lock); | |
73dea47f | 451 | } |
1da177e4 | 452 | |
7d12e780 | 453 | static irqreturn_t timer_interrupt(int irq, void *dev_id) |
73dea47f AK |
454 | { |
455 | if (apic_runs_main_timer > 1) | |
456 | return IRQ_HANDLED; | |
7d12e780 | 457 | main_timer_handler(); |
d25bf7e5 VP |
458 | if (using_apic_timer) |
459 | smp_send_timer_broadcast_ipi(); | |
1da177e4 LT |
460 | return IRQ_HANDLED; |
461 | } | |
462 | ||
68ed0040 | 463 | static unsigned int cyc2ns_scale __read_mostly; |
1da177e4 | 464 | |
dacb16b1 | 465 | static inline void set_cyc2ns_scale(unsigned long cpu_khz) |
1da177e4 | 466 | { |
42211338 | 467 | cyc2ns_scale = (NSEC_PER_MSEC << NS_SCALE) / cpu_khz; |
1da177e4 LT |
468 | } |
469 | ||
470 | static inline unsigned long long cycles_2_ns(unsigned long long cyc) | |
471 | { | |
42211338 | 472 | return (cyc * cyc2ns_scale) >> NS_SCALE; |
1da177e4 LT |
473 | } |
474 | ||
475 | unsigned long long sched_clock(void) | |
476 | { | |
477 | unsigned long a = 0; | |
478 | ||
479 | #if 0 | |
480 | /* Don't do a HPET read here. Using TSC always is much faster | |
481 | and HPET may not be mapped yet when the scheduler first runs. | |
482 | Disadvantage is a small drift between CPUs in some configurations, | |
483 | but that should be tolerable. */ | |
484 | if (__vxtime.mode == VXTIME_HPET) | |
42211338 | 485 | return (hpet_readl(HPET_COUNTER) * vxtime.quot) >> US_SCALE; |
1da177e4 LT |
486 | #endif |
487 | ||
488 | /* Could do CPU core sync here. Opteron can execute rdtsc speculatively, | |
489 | which means it is not completely exact and may not be monotonous between | |
490 | CPUs. But the errors should be too small to matter for scheduling | |
491 | purposes. */ | |
492 | ||
493 | rdtscll(a); | |
494 | return cycles_2_ns(a); | |
495 | } | |
496 | ||
bdf2b1c9 | 497 | static unsigned long get_cmos_time(void) |
1da177e4 | 498 | { |
641f71f5 | 499 | unsigned int year, mon, day, hour, min, sec; |
1da177e4 | 500 | unsigned long flags; |
6954bee8 | 501 | unsigned extyear = 0; |
1da177e4 | 502 | |
1da177e4 LT |
503 | spin_lock_irqsave(&rtc_lock, flags); |
504 | ||
641f71f5 MM |
505 | do { |
506 | sec = CMOS_READ(RTC_SECONDS); | |
507 | min = CMOS_READ(RTC_MINUTES); | |
508 | hour = CMOS_READ(RTC_HOURS); | |
509 | day = CMOS_READ(RTC_DAY_OF_MONTH); | |
510 | mon = CMOS_READ(RTC_MONTH); | |
511 | year = CMOS_READ(RTC_YEAR); | |
6954bee8 | 512 | #ifdef CONFIG_ACPI |
641f71f5 MM |
513 | if (acpi_fadt.revision >= FADT2_REVISION_ID && |
514 | acpi_fadt.century) | |
515 | extyear = CMOS_READ(acpi_fadt.century); | |
6954bee8 | 516 | #endif |
641f71f5 | 517 | } while (sec != CMOS_READ(RTC_SECONDS)); |
6954bee8 | 518 | |
1da177e4 LT |
519 | spin_unlock_irqrestore(&rtc_lock, flags); |
520 | ||
0b91317e AK |
521 | /* |
522 | * We know that x86-64 always uses BCD format, no need to check the | |
523 | * config register. | |
7351c0bf | 524 | */ |
1da177e4 | 525 | |
0b91317e AK |
526 | BCD_TO_BIN(sec); |
527 | BCD_TO_BIN(min); | |
528 | BCD_TO_BIN(hour); | |
529 | BCD_TO_BIN(day); | |
530 | BCD_TO_BIN(mon); | |
531 | BCD_TO_BIN(year); | |
1da177e4 | 532 | |
6954bee8 AK |
533 | if (extyear) { |
534 | BCD_TO_BIN(extyear); | |
535 | year += extyear; | |
536 | printk(KERN_INFO "Extended CMOS year: %d\n", extyear); | |
537 | } else { | |
538 | /* | |
539 | * x86-64 systems only exists since 2002. | |
540 | * This will work up to Dec 31, 2100 | |
541 | */ | |
542 | year += 2000; | |
543 | } | |
1da177e4 LT |
544 | |
545 | return mktime(year, mon, day, hour, min, sec); | |
546 | } | |
547 | ||
548 | #ifdef CONFIG_CPU_FREQ | |
549 | ||
550 | /* Frequency scaling support. Adjust the TSC based timer when the cpu frequency | |
551 | changes. | |
552 | ||
553 | RED-PEN: On SMP we assume all CPUs run with the same frequency. It's | |
554 | not that important because current Opteron setups do not support | |
555 | scaling on SMP anyroads. | |
556 | ||
557 | Should fix up last_tsc too. Currently gettimeofday in the | |
558 | first tick after the change will be slightly wrong. */ | |
559 | ||
560 | #include <linux/workqueue.h> | |
561 | ||
562 | static unsigned int cpufreq_delayed_issched = 0; | |
563 | static unsigned int cpufreq_init = 0; | |
564 | static struct work_struct cpufreq_delayed_get_work; | |
565 | ||
566 | static void handle_cpufreq_delayed_get(void *v) | |
567 | { | |
568 | unsigned int cpu; | |
569 | for_each_online_cpu(cpu) { | |
570 | cpufreq_get(cpu); | |
571 | } | |
572 | cpufreq_delayed_issched = 0; | |
573 | } | |
574 | ||
575 | /* if we notice lost ticks, schedule a call to cpufreq_get() as it tries | |
576 | * to verify the CPU frequency the timing core thinks the CPU is running | |
577 | * at is still correct. | |
578 | */ | |
579 | static void cpufreq_delayed_get(void) | |
580 | { | |
581 | static int warned; | |
582 | if (cpufreq_init && !cpufreq_delayed_issched) { | |
583 | cpufreq_delayed_issched = 1; | |
584 | if (!warned) { | |
585 | warned = 1; | |
7351c0bf AK |
586 | printk(KERN_DEBUG |
587 | "Losing some ticks... checking if CPU frequency changed.\n"); | |
1da177e4 LT |
588 | } |
589 | schedule_work(&cpufreq_delayed_get_work); | |
590 | } | |
591 | } | |
592 | ||
593 | static unsigned int ref_freq = 0; | |
594 | static unsigned long loops_per_jiffy_ref = 0; | |
595 | ||
596 | static unsigned long cpu_khz_ref = 0; | |
597 | ||
598 | static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | |
599 | void *data) | |
600 | { | |
601 | struct cpufreq_freqs *freq = data; | |
602 | unsigned long *lpj, dummy; | |
603 | ||
c29601e9 AK |
604 | if (cpu_has(&cpu_data[freq->cpu], X86_FEATURE_CONSTANT_TSC)) |
605 | return 0; | |
606 | ||
1da177e4 LT |
607 | lpj = &dummy; |
608 | if (!(freq->flags & CPUFREQ_CONST_LOOPS)) | |
609 | #ifdef CONFIG_SMP | |
7351c0bf | 610 | lpj = &cpu_data[freq->cpu].loops_per_jiffy; |
1da177e4 | 611 | #else |
7351c0bf | 612 | lpj = &boot_cpu_data.loops_per_jiffy; |
1da177e4 LT |
613 | #endif |
614 | ||
1da177e4 LT |
615 | if (!ref_freq) { |
616 | ref_freq = freq->old; | |
617 | loops_per_jiffy_ref = *lpj; | |
618 | cpu_khz_ref = cpu_khz; | |
619 | } | |
620 | if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) || | |
621 | (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) || | |
622 | (val == CPUFREQ_RESUMECHANGE)) { | |
623 | *lpj = | |
624 | cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new); | |
625 | ||
626 | cpu_khz = cpufreq_scale(cpu_khz_ref, ref_freq, freq->new); | |
627 | if (!(freq->flags & CPUFREQ_CONST_LOOPS)) | |
42211338 | 628 | vxtime.tsc_quot = (USEC_PER_MSEC << US_SCALE) / cpu_khz; |
1da177e4 LT |
629 | } |
630 | ||
dacb16b1 | 631 | set_cyc2ns_scale(cpu_khz_ref); |
1da177e4 LT |
632 | |
633 | return 0; | |
634 | } | |
635 | ||
636 | static struct notifier_block time_cpufreq_notifier_block = { | |
637 | .notifier_call = time_cpufreq_notifier | |
638 | }; | |
639 | ||
640 | static int __init cpufreq_tsc(void) | |
641 | { | |
642 | INIT_WORK(&cpufreq_delayed_get_work, handle_cpufreq_delayed_get, NULL); | |
643 | if (!cpufreq_register_notifier(&time_cpufreq_notifier_block, | |
644 | CPUFREQ_TRANSITION_NOTIFIER)) | |
645 | cpufreq_init = 1; | |
646 | return 0; | |
647 | } | |
648 | ||
649 | core_initcall(cpufreq_tsc); | |
650 | ||
651 | #endif | |
652 | ||
653 | /* | |
654 | * calibrate_tsc() calibrates the processor TSC in a very simple way, comparing | |
655 | * it to the HPET timer of known frequency. | |
656 | */ | |
657 | ||
658 | #define TICK_COUNT 100000000 | |
659 | ||
660 | static unsigned int __init hpet_calibrate_tsc(void) | |
661 | { | |
662 | int tsc_start, hpet_start; | |
663 | int tsc_now, hpet_now; | |
664 | unsigned long flags; | |
665 | ||
666 | local_irq_save(flags); | |
667 | local_irq_disable(); | |
668 | ||
669 | hpet_start = hpet_readl(HPET_COUNTER); | |
670 | rdtscl(tsc_start); | |
671 | ||
672 | do { | |
673 | local_irq_disable(); | |
674 | hpet_now = hpet_readl(HPET_COUNTER); | |
c818a181 | 675 | tsc_now = get_cycles_sync(); |
1da177e4 LT |
676 | local_irq_restore(flags); |
677 | } while ((tsc_now - tsc_start) < TICK_COUNT && | |
678 | (hpet_now - hpet_start) < TICK_COUNT); | |
679 | ||
680 | return (tsc_now - tsc_start) * 1000000000L | |
681 | / ((hpet_now - hpet_start) * hpet_period / 1000); | |
682 | } | |
683 | ||
684 | ||
685 | /* | |
686 | * pit_calibrate_tsc() uses the speaker output (channel 2) of | |
687 | * the PIT. This is better than using the timer interrupt output, | |
688 | * because we can read the value of the speaker with just one inb(), | |
689 | * where we need three i/o operations for the interrupt channel. | |
690 | * We count how many ticks the TSC does in 50 ms. | |
691 | */ | |
692 | ||
693 | static unsigned int __init pit_calibrate_tsc(void) | |
694 | { | |
695 | unsigned long start, end; | |
696 | unsigned long flags; | |
697 | ||
698 | spin_lock_irqsave(&i8253_lock, flags); | |
699 | ||
700 | outb((inb(0x61) & ~0x02) | 0x01, 0x61); | |
701 | ||
702 | outb(0xb0, 0x43); | |
703 | outb((PIT_TICK_RATE / (1000 / 50)) & 0xff, 0x42); | |
704 | outb((PIT_TICK_RATE / (1000 / 50)) >> 8, 0x42); | |
c818a181 | 705 | start = get_cycles_sync(); |
1da177e4 | 706 | while ((inb(0x61) & 0x20) == 0); |
c818a181 | 707 | end = get_cycles_sync(); |
1da177e4 LT |
708 | |
709 | spin_unlock_irqrestore(&i8253_lock, flags); | |
710 | ||
711 | return (end - start) / 50; | |
712 | } | |
713 | ||
714 | #ifdef CONFIG_HPET | |
715 | static __init int late_hpet_init(void) | |
716 | { | |
717 | struct hpet_data hd; | |
718 | unsigned int ntimer; | |
719 | ||
720 | if (!vxtime.hpet_address) | |
3d34ee68 | 721 | return 0; |
1da177e4 LT |
722 | |
723 | memset(&hd, 0, sizeof (hd)); | |
724 | ||
725 | ntimer = hpet_readl(HPET_ID); | |
726 | ntimer = (ntimer & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT; | |
727 | ntimer++; | |
728 | ||
729 | /* | |
730 | * Register with driver. | |
731 | * Timer0 and Timer1 is used by platform. | |
732 | */ | |
733 | hd.hd_phys_address = vxtime.hpet_address; | |
dd42b151 | 734 | hd.hd_address = (void __iomem *)fix_to_virt(FIX_HPET_BASE); |
1da177e4 LT |
735 | hd.hd_nirqs = ntimer; |
736 | hd.hd_flags = HPET_DATA_PLATFORM; | |
737 | hpet_reserve_timer(&hd, 0); | |
738 | #ifdef CONFIG_HPET_EMULATE_RTC | |
739 | hpet_reserve_timer(&hd, 1); | |
740 | #endif | |
741 | hd.hd_irq[0] = HPET_LEGACY_8254; | |
742 | hd.hd_irq[1] = HPET_LEGACY_RTC; | |
743 | if (ntimer > 2) { | |
744 | struct hpet *hpet; | |
745 | struct hpet_timer *timer; | |
746 | int i; | |
747 | ||
748 | hpet = (struct hpet *) fix_to_virt(FIX_HPET_BASE); | |
7351c0bf AK |
749 | timer = &hpet->hpet_timers[2]; |
750 | for (i = 2; i < ntimer; timer++, i++) | |
1da177e4 LT |
751 | hd.hd_irq[i] = (timer->hpet_config & |
752 | Tn_INT_ROUTE_CNF_MASK) >> | |
753 | Tn_INT_ROUTE_CNF_SHIFT; | |
754 | ||
755 | } | |
756 | ||
757 | hpet_alloc(&hd); | |
758 | return 0; | |
759 | } | |
760 | fs_initcall(late_hpet_init); | |
761 | #endif | |
762 | ||
763 | static int hpet_timer_stop_set_go(unsigned long tick) | |
764 | { | |
765 | unsigned int cfg; | |
766 | ||
767 | /* | |
768 | * Stop the timers and reset the main counter. | |
769 | */ | |
770 | ||
771 | cfg = hpet_readl(HPET_CFG); | |
772 | cfg &= ~(HPET_CFG_ENABLE | HPET_CFG_LEGACY); | |
773 | hpet_writel(cfg, HPET_CFG); | |
774 | hpet_writel(0, HPET_COUNTER); | |
775 | hpet_writel(0, HPET_COUNTER + 4); | |
776 | ||
777 | /* | |
778 | * Set up timer 0, as periodic with first interrupt to happen at hpet_tick, | |
779 | * and period also hpet_tick. | |
780 | */ | |
a3a00751 | 781 | if (hpet_use_timer) { |
782 | hpet_writel(HPET_TN_ENABLE | HPET_TN_PERIODIC | HPET_TN_SETVAL | | |
1da177e4 | 783 | HPET_TN_32BIT, HPET_T0_CFG); |
b2df3ddb VP |
784 | hpet_writel(hpet_tick, HPET_T0_CMP); /* next interrupt */ |
785 | hpet_writel(hpet_tick, HPET_T0_CMP); /* period */ | |
a3a00751 | 786 | cfg |= HPET_CFG_LEGACY; |
787 | } | |
1da177e4 LT |
788 | /* |
789 | * Go! | |
790 | */ | |
791 | ||
a3a00751 | 792 | cfg |= HPET_CFG_ENABLE; |
1da177e4 LT |
793 | hpet_writel(cfg, HPET_CFG); |
794 | ||
795 | return 0; | |
796 | } | |
797 | ||
798 | static int hpet_init(void) | |
799 | { | |
800 | unsigned int id; | |
801 | ||
802 | if (!vxtime.hpet_address) | |
803 | return -1; | |
804 | set_fixmap_nocache(FIX_HPET_BASE, vxtime.hpet_address); | |
805 | __set_fixmap(VSYSCALL_HPET, vxtime.hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE); | |
806 | ||
807 | /* | |
808 | * Read the period, compute tick and quotient. | |
809 | */ | |
810 | ||
811 | id = hpet_readl(HPET_ID); | |
812 | ||
a3a00751 | 813 | if (!(id & HPET_ID_VENDOR) || !(id & HPET_ID_NUMBER)) |
1da177e4 LT |
814 | return -1; |
815 | ||
816 | hpet_period = hpet_readl(HPET_PERIOD); | |
817 | if (hpet_period < 100000 || hpet_period > 100000000) | |
818 | return -1; | |
819 | ||
42211338 | 820 | hpet_tick = (FSEC_PER_TICK + hpet_period / 2) / hpet_period; |
1da177e4 | 821 | |
a3a00751 | 822 | hpet_use_timer = (id & HPET_ID_LEGSUP); |
823 | ||
1da177e4 LT |
824 | return hpet_timer_stop_set_go(hpet_tick); |
825 | } | |
826 | ||
827 | static int hpet_reenable(void) | |
828 | { | |
829 | return hpet_timer_stop_set_go(hpet_tick); | |
830 | } | |
831 | ||
73dea47f AK |
832 | #define PIT_MODE 0x43 |
833 | #define PIT_CH0 0x40 | |
834 | ||
835 | static void __init __pit_init(int val, u8 mode) | |
1da177e4 LT |
836 | { |
837 | unsigned long flags; | |
838 | ||
839 | spin_lock_irqsave(&i8253_lock, flags); | |
73dea47f AK |
840 | outb_p(mode, PIT_MODE); |
841 | outb_p(val & 0xff, PIT_CH0); /* LSB */ | |
842 | outb_p(val >> 8, PIT_CH0); /* MSB */ | |
1da177e4 LT |
843 | spin_unlock_irqrestore(&i8253_lock, flags); |
844 | } | |
845 | ||
73dea47f AK |
846 | void __init pit_init(void) |
847 | { | |
848 | __pit_init(LATCH, 0x34); /* binary, mode 2, LSB/MSB, ch 0 */ | |
849 | } | |
850 | ||
851 | void __init pit_stop_interrupt(void) | |
852 | { | |
853 | __pit_init(0, 0x30); /* mode 0 */ | |
854 | } | |
855 | ||
856 | void __init stop_timer_interrupt(void) | |
857 | { | |
858 | char *name; | |
859 | if (vxtime.hpet_address) { | |
860 | name = "HPET"; | |
861 | hpet_timer_stop_set_go(0); | |
862 | } else { | |
863 | name = "PIT"; | |
864 | pit_stop_interrupt(); | |
865 | } | |
866 | printk(KERN_INFO "timer: %s interrupt stopped.\n", name); | |
867 | } | |
868 | ||
1da177e4 LT |
869 | int __init time_setup(char *str) |
870 | { | |
871 | report_lost_ticks = 1; | |
872 | return 1; | |
873 | } | |
874 | ||
875 | static struct irqaction irq0 = { | |
b1e05aa2 | 876 | timer_interrupt, IRQF_DISABLED, CPU_MASK_NONE, "timer", NULL, NULL |
1da177e4 LT |
877 | }; |
878 | ||
a670fad0 VP |
879 | static int __cpuinit |
880 | time_cpu_notifier(struct notifier_block *nb, unsigned long action, void *hcpu) | |
1da177e4 | 881 | { |
a670fad0 | 882 | unsigned cpu = (unsigned long) hcpu; |
c08c8205 VP |
883 | if (action == CPU_ONLINE) |
884 | vsyscall_set_cpu(cpu); | |
a670fad0 VP |
885 | return NOTIFY_DONE; |
886 | } | |
1da177e4 | 887 | |
a670fad0 VP |
888 | void __init time_init(void) |
889 | { | |
1da177e4 LT |
890 | if (nohpet) |
891 | vxtime.hpet_address = 0; | |
892 | ||
893 | xtime.tv_sec = get_cmos_time(); | |
894 | xtime.tv_nsec = 0; | |
895 | ||
896 | set_normalized_timespec(&wall_to_monotonic, | |
897 | -xtime.tv_sec, -xtime.tv_nsec); | |
898 | ||
a3a00751 | 899 | if (!hpet_init()) |
42211338 | 900 | vxtime_hz = (FSEC_PER_SEC + hpet_period / 2) / hpet_period; |
68e18891 AK |
901 | else |
902 | vxtime.hpet_address = 0; | |
a3a00751 | 903 | |
904 | if (hpet_use_timer) { | |
b20367a6 JH |
905 | /* set tick_nsec to use the proper rate for HPET */ |
906 | tick_nsec = TICK_NSEC_HPET; | |
1da177e4 LT |
907 | cpu_khz = hpet_calibrate_tsc(); |
908 | timename = "HPET"; | |
312df5f1 | 909 | #ifdef CONFIG_X86_PM_TIMER |
fd495471 | 910 | } else if (pmtmr_ioport && !vxtime.hpet_address) { |
312df5f1 AK |
911 | vxtime_hz = PM_TIMER_FREQUENCY; |
912 | timename = "PM"; | |
913 | pit_init(); | |
914 | cpu_khz = pit_calibrate_tsc(); | |
915 | #endif | |
1da177e4 LT |
916 | } else { |
917 | pit_init(); | |
918 | cpu_khz = pit_calibrate_tsc(); | |
919 | timename = "PIT"; | |
920 | } | |
921 | ||
e8b91777 | 922 | vxtime.mode = VXTIME_TSC; |
42211338 VP |
923 | vxtime.quot = (USEC_PER_SEC << US_SCALE) / vxtime_hz; |
924 | vxtime.tsc_quot = (USEC_PER_MSEC << US_SCALE) / cpu_khz; | |
c818a181 | 925 | vxtime.last_tsc = get_cycles_sync(); |
dacb16b1 | 926 | set_cyc2ns_scale(cpu_khz); |
1da177e4 | 927 | setup_irq(0, &irq0); |
a670fad0 VP |
928 | hotcpu_notifier(time_cpu_notifier, 0); |
929 | time_cpu_notifier(NULL, CPU_ONLINE, (void *)(long)smp_processor_id()); | |
1da177e4 | 930 | |
a670fad0 VP |
931 | #ifndef CONFIG_SMP |
932 | time_init_gtod(); | |
933 | #endif | |
1da177e4 LT |
934 | } |
935 | ||
312df5f1 AK |
936 | /* |
937 | * Make an educated guess if the TSC is trustworthy and synchronized | |
938 | * over all CPUs. | |
939 | */ | |
396bd50f | 940 | __cpuinit int unsynchronized_tsc(void) |
312df5f1 AK |
941 | { |
942 | #ifdef CONFIG_SMP | |
f8bf3c65 | 943 | if (apic_is_clustered_box()) |
312df5f1 | 944 | return 1; |
312df5f1 | 945 | #endif |
0e5f61b0 AK |
946 | /* Most intel systems have synchronized TSCs except for |
947 | multi node systems */ | |
948 | if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) { | |
949 | #ifdef CONFIG_ACPI | |
950 | /* But TSC doesn't tick in C3 so don't use it there */ | |
951 | if (acpi_fadt.length > 0 && acpi_fadt.plvl3_lat < 100) | |
952 | return 1; | |
953 | #endif | |
954 | return 0; | |
955 | } | |
956 | ||
312df5f1 | 957 | /* Assume multi socket systems are not synchronized */ |
737c5c3b | 958 | return num_present_cpus() > 1; |
312df5f1 AK |
959 | } |
960 | ||
a8ab26fe | 961 | /* |
e8b91777 | 962 | * Decide what mode gettimeofday should use. |
a8ab26fe | 963 | */ |
a670fad0 | 964 | void time_init_gtod(void) |
1da177e4 LT |
965 | { |
966 | char *timetype; | |
967 | ||
312df5f1 | 968 | if (unsynchronized_tsc()) |
1da177e4 | 969 | notsc = 1; |
a670fad0 | 970 | |
c08c8205 VP |
971 | if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP)) |
972 | vgetcpu_mode = VGETCPU_RDTSCP; | |
973 | else | |
974 | vgetcpu_mode = VGETCPU_LSL; | |
975 | ||
1da177e4 | 976 | if (vxtime.hpet_address && notsc) { |
a3a00751 | 977 | timetype = hpet_use_timer ? "HPET" : "PIT/HPET"; |
33042a9f CM |
978 | if (hpet_use_timer) |
979 | vxtime.last = hpet_readl(HPET_T0_CMP) - hpet_tick; | |
980 | else | |
981 | vxtime.last = hpet_readl(HPET_COUNTER); | |
1da177e4 LT |
982 | vxtime.mode = VXTIME_HPET; |
983 | do_gettimeoffset = do_gettimeoffset_hpet; | |
312df5f1 AK |
984 | #ifdef CONFIG_X86_PM_TIMER |
985 | /* Using PM for gettimeofday is quite slow, but we have no other | |
986 | choice because the TSC is too unreliable on some systems. */ | |
987 | } else if (pmtmr_ioport && !vxtime.hpet_address && notsc) { | |
988 | timetype = "PM"; | |
989 | do_gettimeoffset = do_gettimeoffset_pm; | |
990 | vxtime.mode = VXTIME_PMTMR; | |
991 | sysctl_vsyscall = 0; | |
992 | printk(KERN_INFO "Disabling vsyscall due to use of PM timer\n"); | |
993 | #endif | |
1da177e4 | 994 | } else { |
a3a00751 | 995 | timetype = hpet_use_timer ? "HPET/TSC" : "PIT/TSC"; |
1da177e4 LT |
996 | vxtime.mode = VXTIME_TSC; |
997 | } | |
a670fad0 VP |
998 | |
999 | printk(KERN_INFO "time.c: Using %ld.%06ld MHz WALL %s GTOD %s timer.\n", | |
1000 | vxtime_hz / 1000000, vxtime_hz % 1000000, timename, timetype); | |
1001 | printk(KERN_INFO "time.c: Detected %d.%03d MHz processor.\n", | |
1002 | cpu_khz / 1000, cpu_khz % 1000); | |
1003 | vxtime.quot = (USEC_PER_SEC << US_SCALE) / vxtime_hz; | |
1004 | vxtime.tsc_quot = (USEC_PER_MSEC << US_SCALE) / cpu_khz; | |
1005 | vxtime.last_tsc = get_cycles_sync(); | |
1006 | ||
1007 | set_cyc2ns_scale(cpu_khz); | |
1da177e4 LT |
1008 | } |
1009 | ||
1010 | __setup("report_lost_ticks", time_setup); | |
1011 | ||
1012 | static long clock_cmos_diff; | |
1013 | static unsigned long sleep_start; | |
1014 | ||
0b91317e AK |
1015 | /* |
1016 | * sysfs support for the timer. | |
1017 | */ | |
1018 | ||
0b9c33a7 | 1019 | static int timer_suspend(struct sys_device *dev, pm_message_t state) |
1da177e4 LT |
1020 | { |
1021 | /* | |
1022 | * Estimate time zone so that set_time can update the clock | |
1023 | */ | |
1024 | long cmos_time = get_cmos_time(); | |
1025 | ||
1026 | clock_cmos_diff = -cmos_time; | |
1027 | clock_cmos_diff += get_seconds(); | |
1028 | sleep_start = cmos_time; | |
1029 | return 0; | |
1030 | } | |
1031 | ||
1032 | static int timer_resume(struct sys_device *dev) | |
1033 | { | |
1034 | unsigned long flags; | |
1035 | unsigned long sec; | |
1036 | unsigned long ctime = get_cmos_time(); | |
34464a5b | 1037 | long sleep_length = (ctime - sleep_start) * HZ; |
1da177e4 | 1038 | |
34464a5b RW |
1039 | if (sleep_length < 0) { |
1040 | printk(KERN_WARNING "Time skew detected in timer resume!\n"); | |
1041 | /* The time after the resume must not be earlier than the time | |
1042 | * before the suspend or some nasty things will happen | |
1043 | */ | |
1044 | sleep_length = 0; | |
1045 | ctime = sleep_start; | |
1046 | } | |
1da177e4 LT |
1047 | if (vxtime.hpet_address) |
1048 | hpet_reenable(); | |
1049 | else | |
1050 | i8254_timer_resume(); | |
1051 | ||
1052 | sec = ctime + clock_cmos_diff; | |
1053 | write_seqlock_irqsave(&xtime_lock,flags); | |
1054 | xtime.tv_sec = sec; | |
1055 | xtime.tv_nsec = 0; | |
0dd2ea9a SL |
1056 | if (vxtime.mode == VXTIME_HPET) { |
1057 | if (hpet_use_timer) | |
1058 | vxtime.last = hpet_readl(HPET_T0_CMP) - hpet_tick; | |
1059 | else | |
1060 | vxtime.last = hpet_readl(HPET_COUNTER); | |
1061 | #ifdef CONFIG_X86_PM_TIMER | |
1062 | } else if (vxtime.mode == VXTIME_PMTMR) { | |
1063 | pmtimer_resume(); | |
1064 | #endif | |
1065 | } else | |
1066 | vxtime.last_tsc = get_cycles_sync(); | |
1da177e4 LT |
1067 | write_sequnlock_irqrestore(&xtime_lock,flags); |
1068 | jiffies += sleep_length; | |
0dd2ea9a | 1069 | monotonic_base += sleep_length * (NSEC_PER_SEC/HZ); |
8446f1d3 | 1070 | touch_softlockup_watchdog(); |
1da177e4 LT |
1071 | return 0; |
1072 | } | |
1073 | ||
1074 | static struct sysdev_class timer_sysclass = { | |
1075 | .resume = timer_resume, | |
1076 | .suspend = timer_suspend, | |
1077 | set_kset_name("timer"), | |
1078 | }; | |
1079 | ||
1da177e4 LT |
1080 | /* XXX this driverfs stuff should probably go elsewhere later -john */ |
1081 | static struct sys_device device_timer = { | |
1082 | .id = 0, | |
1083 | .cls = &timer_sysclass, | |
1084 | }; | |
1085 | ||
1086 | static int time_init_device(void) | |
1087 | { | |
1088 | int error = sysdev_class_register(&timer_sysclass); | |
1089 | if (!error) | |
1090 | error = sysdev_register(&device_timer); | |
1091 | return error; | |
1092 | } | |
1093 | ||
1094 | device_initcall(time_init_device); | |
1095 | ||
1096 | #ifdef CONFIG_HPET_EMULATE_RTC | |
1097 | /* HPET in LegacyReplacement Mode eats up RTC interrupt line. When, HPET | |
1098 | * is enabled, we support RTC interrupt functionality in software. | |
1099 | * RTC has 3 kinds of interrupts: | |
1100 | * 1) Update Interrupt - generate an interrupt, every sec, when RTC clock | |
1101 | * is updated | |
1102 | * 2) Alarm Interrupt - generate an interrupt at a specific time of day | |
1103 | * 3) Periodic Interrupt - generate periodic interrupt, with frequencies | |
1104 | * 2Hz-8192Hz (2Hz-64Hz for non-root user) (all freqs in powers of 2) | |
1105 | * (1) and (2) above are implemented using polling at a frequency of | |
1106 | * 64 Hz. The exact frequency is a tradeoff between accuracy and interrupt | |
1107 | * overhead. (DEFAULT_RTC_INT_FREQ) | |
1108 | * For (3), we use interrupts at 64Hz or user specified periodic | |
1109 | * frequency, whichever is higher. | |
1110 | */ | |
1111 | #include <linux/rtc.h> | |
1112 | ||
1da177e4 LT |
1113 | #define DEFAULT_RTC_INT_FREQ 64 |
1114 | #define RTC_NUM_INTS 1 | |
1115 | ||
1116 | static unsigned long UIE_on; | |
1117 | static unsigned long prev_update_sec; | |
1118 | ||
1119 | static unsigned long AIE_on; | |
1120 | static struct rtc_time alarm_time; | |
1121 | ||
1122 | static unsigned long PIE_on; | |
1123 | static unsigned long PIE_freq = DEFAULT_RTC_INT_FREQ; | |
1124 | static unsigned long PIE_count; | |
1125 | ||
1126 | static unsigned long hpet_rtc_int_freq; /* RTC interrupt frequency */ | |
7811fb8f | 1127 | static unsigned int hpet_t1_cmp; /* cached comparator register */ |
1da177e4 LT |
1128 | |
1129 | int is_hpet_enabled(void) | |
1130 | { | |
1131 | return vxtime.hpet_address != 0; | |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * Timer 1 for RTC, we do not use periodic interrupt feature, | |
1136 | * even if HPET supports periodic interrupts on Timer 1. | |
1137 | * The reason being, to set up a periodic interrupt in HPET, we need to | |
1138 | * stop the main counter. And if we do that everytime someone diables/enables | |
1139 | * RTC, we will have adverse effect on main kernel timer running on Timer 0. | |
1140 | * So, for the time being, simulate the periodic interrupt in software. | |
1141 | * | |
1142 | * hpet_rtc_timer_init() is called for the first time and during subsequent | |
1143 | * interuppts reinit happens through hpet_rtc_timer_reinit(). | |
1144 | */ | |
1145 | int hpet_rtc_timer_init(void) | |
1146 | { | |
1147 | unsigned int cfg, cnt; | |
1148 | unsigned long flags; | |
1149 | ||
1150 | if (!is_hpet_enabled()) | |
1151 | return 0; | |
1152 | /* | |
1153 | * Set the counter 1 and enable the interrupts. | |
1154 | */ | |
1155 | if (PIE_on && (PIE_freq > DEFAULT_RTC_INT_FREQ)) | |
1156 | hpet_rtc_int_freq = PIE_freq; | |
1157 | else | |
1158 | hpet_rtc_int_freq = DEFAULT_RTC_INT_FREQ; | |
1159 | ||
1160 | local_irq_save(flags); | |
1447c27d | 1161 | |
1da177e4 LT |
1162 | cnt = hpet_readl(HPET_COUNTER); |
1163 | cnt += ((hpet_tick*HZ)/hpet_rtc_int_freq); | |
1164 | hpet_writel(cnt, HPET_T1_CMP); | |
7811fb8f | 1165 | hpet_t1_cmp = cnt; |
1da177e4 LT |
1166 | |
1167 | cfg = hpet_readl(HPET_T1_CFG); | |
5f819949 CL |
1168 | cfg &= ~HPET_TN_PERIODIC; |
1169 | cfg |= HPET_TN_ENABLE | HPET_TN_32BIT; | |
1da177e4 LT |
1170 | hpet_writel(cfg, HPET_T1_CFG); |
1171 | ||
1447c27d CL |
1172 | local_irq_restore(flags); |
1173 | ||
1da177e4 LT |
1174 | return 1; |
1175 | } | |
1176 | ||
1177 | static void hpet_rtc_timer_reinit(void) | |
1178 | { | |
1447c27d | 1179 | unsigned int cfg, cnt, ticks_per_int, lost_ints; |
1da177e4 | 1180 | |
f00c96f3 CL |
1181 | if (unlikely(!(PIE_on | AIE_on | UIE_on))) { |
1182 | cfg = hpet_readl(HPET_T1_CFG); | |
1183 | cfg &= ~HPET_TN_ENABLE; | |
1184 | hpet_writel(cfg, HPET_T1_CFG); | |
1da177e4 | 1185 | return; |
f00c96f3 | 1186 | } |
1da177e4 LT |
1187 | |
1188 | if (PIE_on && (PIE_freq > DEFAULT_RTC_INT_FREQ)) | |
1189 | hpet_rtc_int_freq = PIE_freq; | |
1190 | else | |
1191 | hpet_rtc_int_freq = DEFAULT_RTC_INT_FREQ; | |
1192 | ||
1193 | /* It is more accurate to use the comparator value than current count.*/ | |
1447c27d CL |
1194 | ticks_per_int = hpet_tick * HZ / hpet_rtc_int_freq; |
1195 | hpet_t1_cmp += ticks_per_int; | |
1196 | hpet_writel(hpet_t1_cmp, HPET_T1_CMP); | |
1197 | ||
1198 | /* | |
1199 | * If the interrupt handler was delayed too long, the write above tries | |
1200 | * to schedule the next interrupt in the past and the hardware would | |
1201 | * not interrupt until the counter had wrapped around. | |
1202 | * So we have to check that the comparator wasn't set to a past time. | |
1203 | */ | |
1204 | cnt = hpet_readl(HPET_COUNTER); | |
1205 | if (unlikely((int)(cnt - hpet_t1_cmp) > 0)) { | |
1206 | lost_ints = (cnt - hpet_t1_cmp) / ticks_per_int + 1; | |
1207 | /* Make sure that, even with the time needed to execute | |
1208 | * this code, the next scheduled interrupt has been moved | |
1209 | * back to the future: */ | |
1210 | lost_ints++; | |
1211 | ||
1212 | hpet_t1_cmp += lost_ints * ticks_per_int; | |
1213 | hpet_writel(hpet_t1_cmp, HPET_T1_CMP); | |
1214 | ||
1215 | if (PIE_on) | |
1216 | PIE_count += lost_ints; | |
1217 | ||
1218 | printk(KERN_WARNING "rtc: lost some interrupts at %ldHz.\n", | |
1219 | hpet_rtc_int_freq); | |
1220 | } | |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | /* | |
1224 | * The functions below are called from rtc driver. | |
1225 | * Return 0 if HPET is not being used. | |
1226 | * Otherwise do the necessary changes and return 1. | |
1227 | */ | |
1228 | int hpet_mask_rtc_irq_bit(unsigned long bit_mask) | |
1229 | { | |
1230 | if (!is_hpet_enabled()) | |
1231 | return 0; | |
1232 | ||
1233 | if (bit_mask & RTC_UIE) | |
1234 | UIE_on = 0; | |
1235 | if (bit_mask & RTC_PIE) | |
1236 | PIE_on = 0; | |
1237 | if (bit_mask & RTC_AIE) | |
1238 | AIE_on = 0; | |
1239 | ||
1240 | return 1; | |
1241 | } | |
1242 | ||
1243 | int hpet_set_rtc_irq_bit(unsigned long bit_mask) | |
1244 | { | |
1245 | int timer_init_reqd = 0; | |
1246 | ||
1247 | if (!is_hpet_enabled()) | |
1248 | return 0; | |
1249 | ||
1250 | if (!(PIE_on | AIE_on | UIE_on)) | |
1251 | timer_init_reqd = 1; | |
1252 | ||
1253 | if (bit_mask & RTC_UIE) { | |
1254 | UIE_on = 1; | |
1255 | } | |
1256 | if (bit_mask & RTC_PIE) { | |
1257 | PIE_on = 1; | |
1258 | PIE_count = 0; | |
1259 | } | |
1260 | if (bit_mask & RTC_AIE) { | |
1261 | AIE_on = 1; | |
1262 | } | |
1263 | ||
1264 | if (timer_init_reqd) | |
1265 | hpet_rtc_timer_init(); | |
1266 | ||
1267 | return 1; | |
1268 | } | |
1269 | ||
1270 | int hpet_set_alarm_time(unsigned char hrs, unsigned char min, unsigned char sec) | |
1271 | { | |
1272 | if (!is_hpet_enabled()) | |
1273 | return 0; | |
1274 | ||
1275 | alarm_time.tm_hour = hrs; | |
1276 | alarm_time.tm_min = min; | |
1277 | alarm_time.tm_sec = sec; | |
1278 | ||
1279 | return 1; | |
1280 | } | |
1281 | ||
1282 | int hpet_set_periodic_freq(unsigned long freq) | |
1283 | { | |
1284 | if (!is_hpet_enabled()) | |
1285 | return 0; | |
1286 | ||
1287 | PIE_freq = freq; | |
1288 | PIE_count = 0; | |
1289 | ||
1290 | return 1; | |
1291 | } | |
1292 | ||
1293 | int hpet_rtc_dropped_irq(void) | |
1294 | { | |
1295 | if (!is_hpet_enabled()) | |
1296 | return 0; | |
1297 | ||
1298 | return 1; | |
1299 | } | |
1300 | ||
1301 | irqreturn_t hpet_rtc_interrupt(int irq, void *dev_id, struct pt_regs *regs) | |
1302 | { | |
1303 | struct rtc_time curr_time; | |
1304 | unsigned long rtc_int_flag = 0; | |
1305 | int call_rtc_interrupt = 0; | |
1306 | ||
1307 | hpet_rtc_timer_reinit(); | |
1308 | ||
1309 | if (UIE_on | AIE_on) { | |
1310 | rtc_get_rtc_time(&curr_time); | |
1311 | } | |
1312 | if (UIE_on) { | |
1313 | if (curr_time.tm_sec != prev_update_sec) { | |
1314 | /* Set update int info, call real rtc int routine */ | |
1315 | call_rtc_interrupt = 1; | |
1316 | rtc_int_flag = RTC_UF; | |
1317 | prev_update_sec = curr_time.tm_sec; | |
1318 | } | |
1319 | } | |
1320 | if (PIE_on) { | |
1321 | PIE_count++; | |
1322 | if (PIE_count >= hpet_rtc_int_freq/PIE_freq) { | |
1323 | /* Set periodic int info, call real rtc int routine */ | |
1324 | call_rtc_interrupt = 1; | |
1325 | rtc_int_flag |= RTC_PF; | |
1326 | PIE_count = 0; | |
1327 | } | |
1328 | } | |
1329 | if (AIE_on) { | |
1330 | if ((curr_time.tm_sec == alarm_time.tm_sec) && | |
1331 | (curr_time.tm_min == alarm_time.tm_min) && | |
1332 | (curr_time.tm_hour == alarm_time.tm_hour)) { | |
1333 | /* Set alarm int info, call real rtc int routine */ | |
1334 | call_rtc_interrupt = 1; | |
1335 | rtc_int_flag |= RTC_AF; | |
1336 | } | |
1337 | } | |
1338 | if (call_rtc_interrupt) { | |
1339 | rtc_int_flag |= (RTC_IRQF | (RTC_NUM_INTS << 8)); | |
7d12e780 | 1340 | rtc_interrupt(rtc_int_flag, dev_id); |
1da177e4 LT |
1341 | } |
1342 | return IRQ_HANDLED; | |
1343 | } | |
1344 | #endif | |
1345 | ||
1da177e4 LT |
1346 | static int __init nohpet_setup(char *s) |
1347 | { | |
1348 | nohpet = 1; | |
9b41046c | 1349 | return 1; |
1da177e4 LT |
1350 | } |
1351 | ||
1352 | __setup("nohpet", nohpet_setup); | |
1353 | ||
7fd67843 | 1354 | int __init notsc_setup(char *s) |
1da177e4 LT |
1355 | { |
1356 | notsc = 1; | |
9b41046c | 1357 | return 1; |
1da177e4 LT |
1358 | } |
1359 | ||
1360 | __setup("notsc", notsc_setup); |