1 | #include <freebsd/machine/rtems-bsd-config.h> |
---|
2 | |
---|
3 | /*- |
---|
4 | * ---------------------------------------------------------------------------- |
---|
5 | * "THE BEER-WARE LICENSE" (Revision 42): |
---|
6 | * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you |
---|
7 | * can do whatever you want with this stuff. If we meet some day, and you think |
---|
8 | * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp |
---|
9 | * ---------------------------------------------------------------------------- |
---|
10 | */ |
---|
11 | |
---|
12 | #include <freebsd/sys/cdefs.h> |
---|
13 | __FBSDID("$FreeBSD$"); |
---|
14 | |
---|
15 | #include <freebsd/local/opt_ntp.h> |
---|
16 | |
---|
17 | #include <freebsd/sys/param.h> |
---|
18 | #include <freebsd/sys/kernel.h> |
---|
19 | #include <freebsd/sys/sysctl.h> |
---|
20 | #include <freebsd/sys/syslog.h> |
---|
21 | #include <freebsd/sys/systm.h> |
---|
22 | #include <freebsd/sys/timepps.h> |
---|
23 | #include <freebsd/sys/timetc.h> |
---|
24 | #include <freebsd/sys/timex.h> |
---|
25 | |
---|
26 | /* |
---|
27 | * A large step happens on boot. This constant detects such steps. |
---|
28 | * It is relatively small so that ntp_update_second gets called enough |
---|
29 | * in the typical 'missed a couple of seconds' case, but doesn't loop |
---|
30 | * forever when the time step is large. |
---|
31 | */ |
---|
32 | #define LARGE_STEP 200 |
---|
33 | |
---|
34 | /* |
---|
35 | * Implement a dummy timecounter which we can use until we get a real one |
---|
36 | * in the air. This allows the console and other early stuff to use |
---|
37 | * time services. |
---|
38 | */ |
---|
39 | |
---|
40 | static u_int |
---|
41 | dummy_get_timecount(struct timecounter *tc) |
---|
42 | { |
---|
43 | static u_int now; |
---|
44 | |
---|
45 | return (++now); |
---|
46 | } |
---|
47 | |
---|
48 | static struct timecounter dummy_timecounter = { |
---|
49 | dummy_get_timecount, 0, ~0u, 1000000, "dummy", -1000000 |
---|
50 | }; |
---|
51 | |
---|
52 | struct timehands { |
---|
53 | /* These fields must be initialized by the driver. */ |
---|
54 | struct timecounter *th_counter; |
---|
55 | int64_t th_adjustment; |
---|
56 | u_int64_t th_scale; |
---|
57 | u_int th_offset_count; |
---|
58 | struct bintime th_offset; |
---|
59 | struct timeval th_microtime; |
---|
60 | struct timespec th_nanotime; |
---|
61 | /* Fields not to be copied in tc_windup start with th_generation. */ |
---|
62 | volatile u_int th_generation; |
---|
63 | struct timehands *th_next; |
---|
64 | }; |
---|
65 | |
---|
66 | static struct timehands th0; |
---|
67 | static struct timehands th9 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th0}; |
---|
68 | static struct timehands th8 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th9}; |
---|
69 | static struct timehands th7 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th8}; |
---|
70 | static struct timehands th6 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th7}; |
---|
71 | static struct timehands th5 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th6}; |
---|
72 | static struct timehands th4 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th5}; |
---|
73 | static struct timehands th3 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th4}; |
---|
74 | static struct timehands th2 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th3}; |
---|
75 | static struct timehands th1 = { NULL, 0, 0, 0, {0, 0}, {0, 0}, {0, 0}, 0, &th2}; |
---|
76 | static struct timehands th0 = { |
---|
77 | &dummy_timecounter, |
---|
78 | 0, |
---|
79 | (uint64_t)-1 / 1000000, |
---|
80 | 0, |
---|
81 | {1, 0}, |
---|
82 | {0, 0}, |
---|
83 | {0, 0}, |
---|
84 | 1, |
---|
85 | &th1 |
---|
86 | }; |
---|
87 | |
---|
88 | static struct timehands *volatile timehands = &th0; |
---|
89 | struct timecounter *timecounter = &dummy_timecounter; |
---|
90 | static struct timecounter *timecounters = &dummy_timecounter; |
---|
91 | |
---|
92 | time_t time_second = 1; |
---|
93 | time_t time_uptime = 1; |
---|
94 | |
---|
95 | static struct bintime boottimebin; |
---|
96 | struct timeval boottime; |
---|
97 | static int sysctl_kern_boottime(SYSCTL_HANDLER_ARGS); |
---|
98 | SYSCTL_PROC(_kern, KERN_BOOTTIME, boottime, CTLTYPE_STRUCT|CTLFLAG_RD, |
---|
99 | NULL, 0, sysctl_kern_boottime, "S,timeval", "System boottime"); |
---|
100 | |
---|
101 | SYSCTL_NODE(_kern, OID_AUTO, timecounter, CTLFLAG_RW, 0, ""); |
---|
102 | SYSCTL_NODE(_kern_timecounter, OID_AUTO, tc, CTLFLAG_RW, 0, ""); |
---|
103 | |
---|
104 | static int timestepwarnings; |
---|
105 | SYSCTL_INT(_kern_timecounter, OID_AUTO, stepwarnings, CTLFLAG_RW, |
---|
106 | ×tepwarnings, 0, "Log time steps"); |
---|
107 | |
---|
108 | static void tc_windup(void); |
---|
109 | static void cpu_tick_calibrate(int); |
---|
110 | |
---|
111 | static int |
---|
112 | sysctl_kern_boottime(SYSCTL_HANDLER_ARGS) |
---|
113 | { |
---|
114 | #ifdef SCTL_MASK32 |
---|
115 | int tv[2]; |
---|
116 | |
---|
117 | if (req->flags & SCTL_MASK32) { |
---|
118 | tv[0] = boottime.tv_sec; |
---|
119 | tv[1] = boottime.tv_usec; |
---|
120 | return SYSCTL_OUT(req, tv, sizeof(tv)); |
---|
121 | } else |
---|
122 | #endif |
---|
123 | return SYSCTL_OUT(req, &boottime, sizeof(boottime)); |
---|
124 | } |
---|
125 | |
---|
126 | static int |
---|
127 | sysctl_kern_timecounter_get(SYSCTL_HANDLER_ARGS) |
---|
128 | { |
---|
129 | u_int ncount; |
---|
130 | struct timecounter *tc = arg1; |
---|
131 | |
---|
132 | ncount = tc->tc_get_timecount(tc); |
---|
133 | return sysctl_handle_int(oidp, &ncount, 0, req); |
---|
134 | } |
---|
135 | |
---|
136 | static int |
---|
137 | sysctl_kern_timecounter_freq(SYSCTL_HANDLER_ARGS) |
---|
138 | { |
---|
139 | u_int64_t freq; |
---|
140 | struct timecounter *tc = arg1; |
---|
141 | |
---|
142 | freq = tc->tc_frequency; |
---|
143 | return sysctl_handle_quad(oidp, &freq, 0, req); |
---|
144 | } |
---|
145 | |
---|
146 | /* |
---|
147 | * Return the difference between the timehands' counter value now and what |
---|
148 | * was when we copied it to the timehands' offset_count. |
---|
149 | */ |
---|
150 | static __inline u_int |
---|
151 | tc_delta(struct timehands *th) |
---|
152 | { |
---|
153 | struct timecounter *tc; |
---|
154 | |
---|
155 | tc = th->th_counter; |
---|
156 | return ((tc->tc_get_timecount(tc) - th->th_offset_count) & |
---|
157 | tc->tc_counter_mask); |
---|
158 | } |
---|
159 | |
---|
160 | /* |
---|
161 | * Functions for reading the time. We have to loop until we are sure that |
---|
162 | * the timehands that we operated on was not updated under our feet. See |
---|
163 | * the comment in <sys/time.h> for a description of these 12 functions. |
---|
164 | */ |
---|
165 | |
---|
166 | void |
---|
167 | binuptime(struct bintime *bt) |
---|
168 | { |
---|
169 | struct timehands *th; |
---|
170 | u_int gen; |
---|
171 | |
---|
172 | do { |
---|
173 | th = timehands; |
---|
174 | gen = th->th_generation; |
---|
175 | *bt = th->th_offset; |
---|
176 | bintime_addx(bt, th->th_scale * tc_delta(th)); |
---|
177 | } while (gen == 0 || gen != th->th_generation); |
---|
178 | } |
---|
179 | |
---|
180 | void |
---|
181 | nanouptime(struct timespec *tsp) |
---|
182 | { |
---|
183 | struct bintime bt; |
---|
184 | |
---|
185 | binuptime(&bt); |
---|
186 | bintime2timespec(&bt, tsp); |
---|
187 | } |
---|
188 | |
---|
189 | void |
---|
190 | microuptime(struct timeval *tvp) |
---|
191 | { |
---|
192 | struct bintime bt; |
---|
193 | |
---|
194 | binuptime(&bt); |
---|
195 | bintime2timeval(&bt, tvp); |
---|
196 | } |
---|
197 | |
---|
198 | void |
---|
199 | bintime(struct bintime *bt) |
---|
200 | { |
---|
201 | |
---|
202 | binuptime(bt); |
---|
203 | bintime_add(bt, &boottimebin); |
---|
204 | } |
---|
205 | |
---|
206 | void |
---|
207 | nanotime(struct timespec *tsp) |
---|
208 | { |
---|
209 | struct bintime bt; |
---|
210 | |
---|
211 | bintime(&bt); |
---|
212 | bintime2timespec(&bt, tsp); |
---|
213 | } |
---|
214 | |
---|
215 | void |
---|
216 | microtime(struct timeval *tvp) |
---|
217 | { |
---|
218 | struct bintime bt; |
---|
219 | |
---|
220 | bintime(&bt); |
---|
221 | bintime2timeval(&bt, tvp); |
---|
222 | } |
---|
223 | |
---|
224 | void |
---|
225 | getbinuptime(struct bintime *bt) |
---|
226 | { |
---|
227 | struct timehands *th; |
---|
228 | u_int gen; |
---|
229 | |
---|
230 | do { |
---|
231 | th = timehands; |
---|
232 | gen = th->th_generation; |
---|
233 | *bt = th->th_offset; |
---|
234 | } while (gen == 0 || gen != th->th_generation); |
---|
235 | } |
---|
236 | |
---|
237 | void |
---|
238 | getnanouptime(struct timespec *tsp) |
---|
239 | { |
---|
240 | struct timehands *th; |
---|
241 | u_int gen; |
---|
242 | |
---|
243 | do { |
---|
244 | th = timehands; |
---|
245 | gen = th->th_generation; |
---|
246 | bintime2timespec(&th->th_offset, tsp); |
---|
247 | } while (gen == 0 || gen != th->th_generation); |
---|
248 | } |
---|
249 | |
---|
250 | void |
---|
251 | getmicrouptime(struct timeval *tvp) |
---|
252 | { |
---|
253 | struct timehands *th; |
---|
254 | u_int gen; |
---|
255 | |
---|
256 | do { |
---|
257 | th = timehands; |
---|
258 | gen = th->th_generation; |
---|
259 | bintime2timeval(&th->th_offset, tvp); |
---|
260 | } while (gen == 0 || gen != th->th_generation); |
---|
261 | } |
---|
262 | |
---|
263 | void |
---|
264 | getbintime(struct bintime *bt) |
---|
265 | { |
---|
266 | struct timehands *th; |
---|
267 | u_int gen; |
---|
268 | |
---|
269 | do { |
---|
270 | th = timehands; |
---|
271 | gen = th->th_generation; |
---|
272 | *bt = th->th_offset; |
---|
273 | } while (gen == 0 || gen != th->th_generation); |
---|
274 | bintime_add(bt, &boottimebin); |
---|
275 | } |
---|
276 | |
---|
277 | void |
---|
278 | getnanotime(struct timespec *tsp) |
---|
279 | { |
---|
280 | struct timehands *th; |
---|
281 | u_int gen; |
---|
282 | |
---|
283 | do { |
---|
284 | th = timehands; |
---|
285 | gen = th->th_generation; |
---|
286 | *tsp = th->th_nanotime; |
---|
287 | } while (gen == 0 || gen != th->th_generation); |
---|
288 | } |
---|
289 | |
---|
290 | void |
---|
291 | getmicrotime(struct timeval *tvp) |
---|
292 | { |
---|
293 | struct timehands *th; |
---|
294 | u_int gen; |
---|
295 | |
---|
296 | do { |
---|
297 | th = timehands; |
---|
298 | gen = th->th_generation; |
---|
299 | *tvp = th->th_microtime; |
---|
300 | } while (gen == 0 || gen != th->th_generation); |
---|
301 | } |
---|
302 | |
---|
303 | /* |
---|
304 | * Initialize a new timecounter and possibly use it. |
---|
305 | */ |
---|
306 | void |
---|
307 | tc_init(struct timecounter *tc) |
---|
308 | { |
---|
309 | u_int u; |
---|
310 | struct sysctl_oid *tc_root; |
---|
311 | |
---|
312 | u = tc->tc_frequency / tc->tc_counter_mask; |
---|
313 | /* XXX: We need some margin here, 10% is a guess */ |
---|
314 | u *= 11; |
---|
315 | u /= 10; |
---|
316 | if (u > hz && tc->tc_quality >= 0) { |
---|
317 | tc->tc_quality = -2000; |
---|
318 | if (bootverbose) { |
---|
319 | printf("Timecounter \"%s\" frequency %ju Hz", |
---|
320 | tc->tc_name, (uintmax_t)tc->tc_frequency); |
---|
321 | printf(" -- Insufficient hz, needs at least %u\n", u); |
---|
322 | } |
---|
323 | } else if (tc->tc_quality >= 0 || bootverbose) { |
---|
324 | printf("Timecounter \"%s\" frequency %ju Hz quality %d\n", |
---|
325 | tc->tc_name, (uintmax_t)tc->tc_frequency, |
---|
326 | tc->tc_quality); |
---|
327 | } |
---|
328 | |
---|
329 | tc->tc_next = timecounters; |
---|
330 | timecounters = tc; |
---|
331 | /* |
---|
332 | * Set up sysctl tree for this counter. |
---|
333 | */ |
---|
334 | tc_root = SYSCTL_ADD_NODE(NULL, |
---|
335 | SYSCTL_STATIC_CHILDREN(_kern_timecounter_tc), OID_AUTO, tc->tc_name, |
---|
336 | CTLFLAG_RW, 0, "timecounter description"); |
---|
337 | SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(tc_root), OID_AUTO, |
---|
338 | "mask", CTLFLAG_RD, &(tc->tc_counter_mask), 0, |
---|
339 | "mask for implemented bits"); |
---|
340 | SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(tc_root), OID_AUTO, |
---|
341 | "counter", CTLTYPE_UINT | CTLFLAG_RD, tc, sizeof(*tc), |
---|
342 | sysctl_kern_timecounter_get, "IU", "current timecounter value"); |
---|
343 | SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(tc_root), OID_AUTO, |
---|
344 | "frequency", CTLTYPE_QUAD | CTLFLAG_RD, tc, sizeof(*tc), |
---|
345 | sysctl_kern_timecounter_freq, "QU", "timecounter frequency"); |
---|
346 | SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(tc_root), OID_AUTO, |
---|
347 | "quality", CTLFLAG_RD, &(tc->tc_quality), 0, |
---|
348 | "goodness of time counter"); |
---|
349 | /* |
---|
350 | * Never automatically use a timecounter with negative quality. |
---|
351 | * Even though we run on the dummy counter, switching here may be |
---|
352 | * worse since this timecounter may not be monotonous. |
---|
353 | */ |
---|
354 | if (tc->tc_quality < 0) |
---|
355 | return; |
---|
356 | if (tc->tc_quality < timecounter->tc_quality) |
---|
357 | return; |
---|
358 | if (tc->tc_quality == timecounter->tc_quality && |
---|
359 | tc->tc_frequency < timecounter->tc_frequency) |
---|
360 | return; |
---|
361 | (void)tc->tc_get_timecount(tc); |
---|
362 | (void)tc->tc_get_timecount(tc); |
---|
363 | timecounter = tc; |
---|
364 | } |
---|
365 | |
---|
366 | /* Report the frequency of the current timecounter. */ |
---|
367 | u_int64_t |
---|
368 | tc_getfrequency(void) |
---|
369 | { |
---|
370 | |
---|
371 | return (timehands->th_counter->tc_frequency); |
---|
372 | } |
---|
373 | |
---|
374 | /* |
---|
375 | * Step our concept of UTC. This is done by modifying our estimate of |
---|
376 | * when we booted. |
---|
377 | * XXX: not locked. |
---|
378 | */ |
---|
379 | void |
---|
380 | tc_setclock(struct timespec *ts) |
---|
381 | { |
---|
382 | struct timespec tbef, taft; |
---|
383 | struct bintime bt, bt2; |
---|
384 | |
---|
385 | cpu_tick_calibrate(1); |
---|
386 | nanotime(&tbef); |
---|
387 | timespec2bintime(ts, &bt); |
---|
388 | binuptime(&bt2); |
---|
389 | bintime_sub(&bt, &bt2); |
---|
390 | bintime_add(&bt2, &boottimebin); |
---|
391 | boottimebin = bt; |
---|
392 | bintime2timeval(&bt, &boottime); |
---|
393 | |
---|
394 | /* XXX fiddle all the little crinkly bits around the fiords... */ |
---|
395 | tc_windup(); |
---|
396 | nanotime(&taft); |
---|
397 | if (timestepwarnings) { |
---|
398 | log(LOG_INFO, |
---|
399 | "Time stepped from %jd.%09ld to %jd.%09ld (%jd.%09ld)\n", |
---|
400 | (intmax_t)tbef.tv_sec, tbef.tv_nsec, |
---|
401 | (intmax_t)taft.tv_sec, taft.tv_nsec, |
---|
402 | (intmax_t)ts->tv_sec, ts->tv_nsec); |
---|
403 | } |
---|
404 | cpu_tick_calibrate(1); |
---|
405 | } |
---|
406 | |
---|
407 | /* |
---|
408 | * Initialize the next struct timehands in the ring and make |
---|
409 | * it the active timehands. Along the way we might switch to a different |
---|
410 | * timecounter and/or do seconds processing in NTP. Slightly magic. |
---|
411 | */ |
---|
412 | static void |
---|
413 | tc_windup(void) |
---|
414 | { |
---|
415 | struct bintime bt; |
---|
416 | struct timehands *th, *tho; |
---|
417 | u_int64_t scale; |
---|
418 | u_int delta, ncount, ogen; |
---|
419 | int i; |
---|
420 | time_t t; |
---|
421 | |
---|
422 | /* |
---|
423 | * Make the next timehands a copy of the current one, but do not |
---|
424 | * overwrite the generation or next pointer. While we update |
---|
425 | * the contents, the generation must be zero. |
---|
426 | */ |
---|
427 | tho = timehands; |
---|
428 | th = tho->th_next; |
---|
429 | ogen = th->th_generation; |
---|
430 | th->th_generation = 0; |
---|
431 | bcopy(tho, th, offsetof(struct timehands, th_generation)); |
---|
432 | |
---|
433 | /* |
---|
434 | * Capture a timecounter delta on the current timecounter and if |
---|
435 | * changing timecounters, a counter value from the new timecounter. |
---|
436 | * Update the offset fields accordingly. |
---|
437 | */ |
---|
438 | delta = tc_delta(th); |
---|
439 | if (th->th_counter != timecounter) |
---|
440 | ncount = timecounter->tc_get_timecount(timecounter); |
---|
441 | else |
---|
442 | ncount = 0; |
---|
443 | th->th_offset_count += delta; |
---|
444 | th->th_offset_count &= th->th_counter->tc_counter_mask; |
---|
445 | while (delta > th->th_counter->tc_frequency) { |
---|
446 | /* Eat complete unadjusted seconds. */ |
---|
447 | delta -= th->th_counter->tc_frequency; |
---|
448 | th->th_offset.sec++; |
---|
449 | } |
---|
450 | if ((delta > th->th_counter->tc_frequency / 2) && |
---|
451 | (th->th_scale * delta < ((uint64_t)1 << 63))) { |
---|
452 | /* The product th_scale * delta just barely overflows. */ |
---|
453 | th->th_offset.sec++; |
---|
454 | } |
---|
455 | bintime_addx(&th->th_offset, th->th_scale * delta); |
---|
456 | |
---|
457 | /* |
---|
458 | * Hardware latching timecounters may not generate interrupts on |
---|
459 | * PPS events, so instead we poll them. There is a finite risk that |
---|
460 | * the hardware might capture a count which is later than the one we |
---|
461 | * got above, and therefore possibly in the next NTP second which might |
---|
462 | * have a different rate than the current NTP second. It doesn't |
---|
463 | * matter in practice. |
---|
464 | */ |
---|
465 | if (tho->th_counter->tc_poll_pps) |
---|
466 | tho->th_counter->tc_poll_pps(tho->th_counter); |
---|
467 | |
---|
468 | /* |
---|
469 | * Deal with NTP second processing. The for loop normally |
---|
470 | * iterates at most once, but in extreme situations it might |
---|
471 | * keep NTP sane if timeouts are not run for several seconds. |
---|
472 | * At boot, the time step can be large when the TOD hardware |
---|
473 | * has been read, so on really large steps, we call |
---|
474 | * ntp_update_second only twice. We need to call it twice in |
---|
475 | * case we missed a leap second. |
---|
476 | */ |
---|
477 | bt = th->th_offset; |
---|
478 | bintime_add(&bt, &boottimebin); |
---|
479 | i = bt.sec - tho->th_microtime.tv_sec; |
---|
480 | if (i > LARGE_STEP) |
---|
481 | i = 2; |
---|
482 | for (; i > 0; i--) { |
---|
483 | t = bt.sec; |
---|
484 | ntp_update_second(&th->th_adjustment, &bt.sec); |
---|
485 | if (bt.sec != t) |
---|
486 | boottimebin.sec += bt.sec - t; |
---|
487 | } |
---|
488 | /* Update the UTC timestamps used by the get*() functions. */ |
---|
489 | /* XXX shouldn't do this here. Should force non-`get' versions. */ |
---|
490 | bintime2timeval(&bt, &th->th_microtime); |
---|
491 | bintime2timespec(&bt, &th->th_nanotime); |
---|
492 | |
---|
493 | /* Now is a good time to change timecounters. */ |
---|
494 | if (th->th_counter != timecounter) { |
---|
495 | th->th_counter = timecounter; |
---|
496 | th->th_offset_count = ncount; |
---|
497 | } |
---|
498 | |
---|
499 | /*- |
---|
500 | * Recalculate the scaling factor. We want the number of 1/2^64 |
---|
501 | * fractions of a second per period of the hardware counter, taking |
---|
502 | * into account the th_adjustment factor which the NTP PLL/adjtime(2) |
---|
503 | * processing provides us with. |
---|
504 | * |
---|
505 | * The th_adjustment is nanoseconds per second with 32 bit binary |
---|
506 | * fraction and we want 64 bit binary fraction of second: |
---|
507 | * |
---|
508 | * x = a * 2^32 / 10^9 = a * 4.294967296 |
---|
509 | * |
---|
510 | * The range of th_adjustment is +/- 5000PPM so inside a 64bit int |
---|
511 | * we can only multiply by about 850 without overflowing, that |
---|
512 | * leaves no suitably precise fractions for multiply before divide. |
---|
513 | * |
---|
514 | * Divide before multiply with a fraction of 2199/512 results in a |
---|
515 | * systematic undercompensation of 10PPM of th_adjustment. On a |
---|
516 | * 5000PPM adjustment this is a 0.05PPM error. This is acceptable. |
---|
517 | * |
---|
518 | * We happily sacrifice the lowest of the 64 bits of our result |
---|
519 | * to the goddess of code clarity. |
---|
520 | * |
---|
521 | */ |
---|
522 | scale = (u_int64_t)1 << 63; |
---|
523 | scale += (th->th_adjustment / 1024) * 2199; |
---|
524 | scale /= th->th_counter->tc_frequency; |
---|
525 | th->th_scale = scale * 2; |
---|
526 | |
---|
527 | /* |
---|
528 | * Now that the struct timehands is again consistent, set the new |
---|
529 | * generation number, making sure to not make it zero. |
---|
530 | */ |
---|
531 | if (++ogen == 0) |
---|
532 | ogen = 1; |
---|
533 | th->th_generation = ogen; |
---|
534 | |
---|
535 | /* Go live with the new struct timehands. */ |
---|
536 | time_second = th->th_microtime.tv_sec; |
---|
537 | time_uptime = th->th_offset.sec; |
---|
538 | timehands = th; |
---|
539 | } |
---|
540 | |
---|
541 | /* Report or change the active timecounter hardware. */ |
---|
542 | static int |
---|
543 | sysctl_kern_timecounter_hardware(SYSCTL_HANDLER_ARGS) |
---|
544 | { |
---|
545 | char newname[32]; |
---|
546 | struct timecounter *newtc, *tc; |
---|
547 | int error; |
---|
548 | |
---|
549 | tc = timecounter; |
---|
550 | strlcpy(newname, tc->tc_name, sizeof(newname)); |
---|
551 | |
---|
552 | error = sysctl_handle_string(oidp, &newname[0], sizeof(newname), req); |
---|
553 | if (error != 0 || req->newptr == NULL || |
---|
554 | strcmp(newname, tc->tc_name) == 0) |
---|
555 | return (error); |
---|
556 | for (newtc = timecounters; newtc != NULL; newtc = newtc->tc_next) { |
---|
557 | if (strcmp(newname, newtc->tc_name) != 0) |
---|
558 | continue; |
---|
559 | |
---|
560 | /* Warm up new timecounter. */ |
---|
561 | (void)newtc->tc_get_timecount(newtc); |
---|
562 | (void)newtc->tc_get_timecount(newtc); |
---|
563 | |
---|
564 | timecounter = newtc; |
---|
565 | return (0); |
---|
566 | } |
---|
567 | return (EINVAL); |
---|
568 | } |
---|
569 | |
---|
570 | SYSCTL_PROC(_kern_timecounter, OID_AUTO, hardware, CTLTYPE_STRING | CTLFLAG_RW, |
---|
571 | 0, 0, sysctl_kern_timecounter_hardware, "A", |
---|
572 | "Timecounter hardware selected"); |
---|
573 | |
---|
574 | |
---|
575 | /* Report or change the active timecounter hardware. */ |
---|
576 | static int |
---|
577 | sysctl_kern_timecounter_choice(SYSCTL_HANDLER_ARGS) |
---|
578 | { |
---|
579 | char buf[32], *spc; |
---|
580 | struct timecounter *tc; |
---|
581 | int error; |
---|
582 | |
---|
583 | spc = ""; |
---|
584 | error = 0; |
---|
585 | for (tc = timecounters; error == 0 && tc != NULL; tc = tc->tc_next) { |
---|
586 | sprintf(buf, "%s%s(%d)", |
---|
587 | spc, tc->tc_name, tc->tc_quality); |
---|
588 | error = SYSCTL_OUT(req, buf, strlen(buf)); |
---|
589 | spc = " "; |
---|
590 | } |
---|
591 | return (error); |
---|
592 | } |
---|
593 | |
---|
594 | SYSCTL_PROC(_kern_timecounter, OID_AUTO, choice, CTLTYPE_STRING | CTLFLAG_RD, |
---|
595 | 0, 0, sysctl_kern_timecounter_choice, "A", "Timecounter hardware detected"); |
---|
596 | |
---|
597 | /* |
---|
598 | * RFC 2783 PPS-API implementation. |
---|
599 | */ |
---|
600 | |
---|
601 | int |
---|
602 | pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps) |
---|
603 | { |
---|
604 | pps_params_t *app; |
---|
605 | struct pps_fetch_args *fapi; |
---|
606 | #ifdef PPS_SYNC |
---|
607 | struct pps_kcbind_args *kapi; |
---|
608 | #endif |
---|
609 | |
---|
610 | KASSERT(pps != NULL, ("NULL pps pointer in pps_ioctl")); |
---|
611 | switch (cmd) { |
---|
612 | case PPS_IOC_CREATE: |
---|
613 | return (0); |
---|
614 | case PPS_IOC_DESTROY: |
---|
615 | return (0); |
---|
616 | case PPS_IOC_SETPARAMS: |
---|
617 | app = (pps_params_t *)data; |
---|
618 | if (app->mode & ~pps->ppscap) |
---|
619 | return (EINVAL); |
---|
620 | pps->ppsparam = *app; |
---|
621 | return (0); |
---|
622 | case PPS_IOC_GETPARAMS: |
---|
623 | app = (pps_params_t *)data; |
---|
624 | *app = pps->ppsparam; |
---|
625 | app->api_version = PPS_API_VERS_1; |
---|
626 | return (0); |
---|
627 | case PPS_IOC_GETCAP: |
---|
628 | *(int*)data = pps->ppscap; |
---|
629 | return (0); |
---|
630 | case PPS_IOC_FETCH: |
---|
631 | fapi = (struct pps_fetch_args *)data; |
---|
632 | if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC) |
---|
633 | return (EINVAL); |
---|
634 | if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec) |
---|
635 | return (EOPNOTSUPP); |
---|
636 | pps->ppsinfo.current_mode = pps->ppsparam.mode; |
---|
637 | fapi->pps_info_buf = pps->ppsinfo; |
---|
638 | return (0); |
---|
639 | case PPS_IOC_KCBIND: |
---|
640 | #ifdef PPS_SYNC |
---|
641 | kapi = (struct pps_kcbind_args *)data; |
---|
642 | /* XXX Only root should be able to do this */ |
---|
643 | if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC) |
---|
644 | return (EINVAL); |
---|
645 | if (kapi->kernel_consumer != PPS_KC_HARDPPS) |
---|
646 | return (EINVAL); |
---|
647 | if (kapi->edge & ~pps->ppscap) |
---|
648 | return (EINVAL); |
---|
649 | pps->kcmode = kapi->edge; |
---|
650 | return (0); |
---|
651 | #else |
---|
652 | return (EOPNOTSUPP); |
---|
653 | #endif |
---|
654 | default: |
---|
655 | return (ENOIOCTL); |
---|
656 | } |
---|
657 | } |
---|
658 | |
---|
659 | void |
---|
660 | pps_init(struct pps_state *pps) |
---|
661 | { |
---|
662 | pps->ppscap |= PPS_TSFMT_TSPEC; |
---|
663 | if (pps->ppscap & PPS_CAPTUREASSERT) |
---|
664 | pps->ppscap |= PPS_OFFSETASSERT; |
---|
665 | if (pps->ppscap & PPS_CAPTURECLEAR) |
---|
666 | pps->ppscap |= PPS_OFFSETCLEAR; |
---|
667 | } |
---|
668 | |
---|
669 | void |
---|
670 | pps_capture(struct pps_state *pps) |
---|
671 | { |
---|
672 | struct timehands *th; |
---|
673 | |
---|
674 | KASSERT(pps != NULL, ("NULL pps pointer in pps_capture")); |
---|
675 | th = timehands; |
---|
676 | pps->capgen = th->th_generation; |
---|
677 | pps->capth = th; |
---|
678 | pps->capcount = th->th_counter->tc_get_timecount(th->th_counter); |
---|
679 | if (pps->capgen != th->th_generation) |
---|
680 | pps->capgen = 0; |
---|
681 | } |
---|
682 | |
---|
683 | void |
---|
684 | pps_event(struct pps_state *pps, int event) |
---|
685 | { |
---|
686 | struct bintime bt; |
---|
687 | struct timespec ts, *tsp, *osp; |
---|
688 | u_int tcount, *pcount; |
---|
689 | int foff, fhard; |
---|
690 | pps_seq_t *pseq; |
---|
691 | |
---|
692 | KASSERT(pps != NULL, ("NULL pps pointer in pps_event")); |
---|
693 | /* If the timecounter was wound up underneath us, bail out. */ |
---|
694 | if (pps->capgen == 0 || pps->capgen != pps->capth->th_generation) |
---|
695 | return; |
---|
696 | |
---|
697 | /* Things would be easier with arrays. */ |
---|
698 | if (event == PPS_CAPTUREASSERT) { |
---|
699 | tsp = &pps->ppsinfo.assert_timestamp; |
---|
700 | osp = &pps->ppsparam.assert_offset; |
---|
701 | foff = pps->ppsparam.mode & PPS_OFFSETASSERT; |
---|
702 | fhard = pps->kcmode & PPS_CAPTUREASSERT; |
---|
703 | pcount = &pps->ppscount[0]; |
---|
704 | pseq = &pps->ppsinfo.assert_sequence; |
---|
705 | } else { |
---|
706 | tsp = &pps->ppsinfo.clear_timestamp; |
---|
707 | osp = &pps->ppsparam.clear_offset; |
---|
708 | foff = pps->ppsparam.mode & PPS_OFFSETCLEAR; |
---|
709 | fhard = pps->kcmode & PPS_CAPTURECLEAR; |
---|
710 | pcount = &pps->ppscount[1]; |
---|
711 | pseq = &pps->ppsinfo.clear_sequence; |
---|
712 | } |
---|
713 | |
---|
714 | /* |
---|
715 | * If the timecounter changed, we cannot compare the count values, so |
---|
716 | * we have to drop the rest of the PPS-stuff until the next event. |
---|
717 | */ |
---|
718 | if (pps->ppstc != pps->capth->th_counter) { |
---|
719 | pps->ppstc = pps->capth->th_counter; |
---|
720 | *pcount = pps->capcount; |
---|
721 | pps->ppscount[2] = pps->capcount; |
---|
722 | return; |
---|
723 | } |
---|
724 | |
---|
725 | /* Convert the count to a timespec. */ |
---|
726 | tcount = pps->capcount - pps->capth->th_offset_count; |
---|
727 | tcount &= pps->capth->th_counter->tc_counter_mask; |
---|
728 | bt = pps->capth->th_offset; |
---|
729 | bintime_addx(&bt, pps->capth->th_scale * tcount); |
---|
730 | bintime_add(&bt, &boottimebin); |
---|
731 | bintime2timespec(&bt, &ts); |
---|
732 | |
---|
733 | /* If the timecounter was wound up underneath us, bail out. */ |
---|
734 | if (pps->capgen != pps->capth->th_generation) |
---|
735 | return; |
---|
736 | |
---|
737 | *pcount = pps->capcount; |
---|
738 | (*pseq)++; |
---|
739 | *tsp = ts; |
---|
740 | |
---|
741 | if (foff) { |
---|
742 | timespecadd(tsp, osp); |
---|
743 | if (tsp->tv_nsec < 0) { |
---|
744 | tsp->tv_nsec += 1000000000; |
---|
745 | tsp->tv_sec -= 1; |
---|
746 | } |
---|
747 | } |
---|
748 | #ifdef PPS_SYNC |
---|
749 | if (fhard) { |
---|
750 | u_int64_t scale; |
---|
751 | |
---|
752 | /* |
---|
753 | * Feed the NTP PLL/FLL. |
---|
754 | * The FLL wants to know how many (hardware) nanoseconds |
---|
755 | * elapsed since the previous event. |
---|
756 | */ |
---|
757 | tcount = pps->capcount - pps->ppscount[2]; |
---|
758 | pps->ppscount[2] = pps->capcount; |
---|
759 | tcount &= pps->capth->th_counter->tc_counter_mask; |
---|
760 | scale = (u_int64_t)1 << 63; |
---|
761 | scale /= pps->capth->th_counter->tc_frequency; |
---|
762 | scale *= 2; |
---|
763 | bt.sec = 0; |
---|
764 | bt.frac = 0; |
---|
765 | bintime_addx(&bt, scale * tcount); |
---|
766 | bintime2timespec(&bt, &ts); |
---|
767 | hardpps(tsp, ts.tv_nsec + 1000000000 * ts.tv_sec); |
---|
768 | } |
---|
769 | #endif |
---|
770 | } |
---|
771 | |
---|
772 | /* |
---|
773 | * Timecounters need to be updated every so often to prevent the hardware |
---|
774 | * counter from overflowing. Updating also recalculates the cached values |
---|
775 | * used by the get*() family of functions, so their precision depends on |
---|
776 | * the update frequency. |
---|
777 | */ |
---|
778 | |
---|
779 | static int tc_tick; |
---|
780 | SYSCTL_INT(_kern_timecounter, OID_AUTO, tick, CTLFLAG_RD, &tc_tick, 0, |
---|
781 | "Approximate number of hardclock ticks in a millisecond"); |
---|
782 | |
---|
783 | void |
---|
784 | tc_ticktock(void) |
---|
785 | { |
---|
786 | static int count; |
---|
787 | static time_t last_calib; |
---|
788 | |
---|
789 | if (++count < tc_tick) |
---|
790 | return; |
---|
791 | count = 0; |
---|
792 | tc_windup(); |
---|
793 | if (time_uptime != last_calib && !(time_uptime & 0xf)) { |
---|
794 | cpu_tick_calibrate(0); |
---|
795 | last_calib = time_uptime; |
---|
796 | } |
---|
797 | } |
---|
798 | |
---|
799 | static void |
---|
800 | inittimecounter(void *dummy) |
---|
801 | { |
---|
802 | u_int p; |
---|
803 | |
---|
804 | /* |
---|
805 | * Set the initial timeout to |
---|
806 | * max(1, <approx. number of hardclock ticks in a millisecond>). |
---|
807 | * People should probably not use the sysctl to set the timeout |
---|
808 | * to smaller than its inital value, since that value is the |
---|
809 | * smallest reasonable one. If they want better timestamps they |
---|
810 | * should use the non-"get"* functions. |
---|
811 | */ |
---|
812 | if (hz > 1000) |
---|
813 | tc_tick = (hz + 500) / 1000; |
---|
814 | else |
---|
815 | tc_tick = 1; |
---|
816 | p = (tc_tick * 1000000) / hz; |
---|
817 | printf("Timecounters tick every %d.%03u msec\n", p / 1000, p % 1000); |
---|
818 | |
---|
819 | /* warm up new timecounter (again) and get rolling. */ |
---|
820 | (void)timecounter->tc_get_timecount(timecounter); |
---|
821 | (void)timecounter->tc_get_timecount(timecounter); |
---|
822 | } |
---|
823 | |
---|
824 | SYSINIT(timecounter, SI_SUB_CLOCKS, SI_ORDER_SECOND, inittimecounter, NULL); |
---|
825 | |
---|
826 | /* Cpu tick handling -------------------------------------------------*/ |
---|
827 | |
---|
828 | static int cpu_tick_variable; |
---|
829 | static uint64_t cpu_tick_frequency; |
---|
830 | |
---|
831 | static uint64_t |
---|
832 | tc_cpu_ticks(void) |
---|
833 | { |
---|
834 | static uint64_t base; |
---|
835 | static unsigned last; |
---|
836 | unsigned u; |
---|
837 | struct timecounter *tc; |
---|
838 | |
---|
839 | tc = timehands->th_counter; |
---|
840 | u = tc->tc_get_timecount(tc) & tc->tc_counter_mask; |
---|
841 | if (u < last) |
---|
842 | base += (uint64_t)tc->tc_counter_mask + 1; |
---|
843 | last = u; |
---|
844 | return (u + base); |
---|
845 | } |
---|
846 | |
---|
847 | /* |
---|
848 | * This function gets called every 16 seconds on only one designated |
---|
849 | * CPU in the system from hardclock() via tc_ticktock(). |
---|
850 | * |
---|
851 | * Whenever the real time clock is stepped we get called with reset=1 |
---|
852 | * to make sure we handle suspend/resume and similar events correctly. |
---|
853 | */ |
---|
854 | |
---|
855 | static void |
---|
856 | cpu_tick_calibrate(int reset) |
---|
857 | { |
---|
858 | static uint64_t c_last; |
---|
859 | uint64_t c_this, c_delta; |
---|
860 | static struct bintime t_last; |
---|
861 | struct bintime t_this, t_delta; |
---|
862 | uint32_t divi; |
---|
863 | |
---|
864 | if (reset) { |
---|
865 | /* The clock was stepped, abort & reset */ |
---|
866 | t_last.sec = 0; |
---|
867 | return; |
---|
868 | } |
---|
869 | |
---|
870 | /* we don't calibrate fixed rate cputicks */ |
---|
871 | if (!cpu_tick_variable) |
---|
872 | return; |
---|
873 | |
---|
874 | getbinuptime(&t_this); |
---|
875 | c_this = cpu_ticks(); |
---|
876 | if (t_last.sec != 0) { |
---|
877 | c_delta = c_this - c_last; |
---|
878 | t_delta = t_this; |
---|
879 | bintime_sub(&t_delta, &t_last); |
---|
880 | /* |
---|
881 | * Validate that 16 +/- 1/256 seconds passed. |
---|
882 | * After division by 16 this gives us a precision of |
---|
883 | * roughly 250PPM which is sufficient |
---|
884 | */ |
---|
885 | if (t_delta.sec > 16 || ( |
---|
886 | t_delta.sec == 16 && t_delta.frac >= (0x01LL << 56))) { |
---|
887 | /* too long */ |
---|
888 | if (bootverbose) |
---|
889 | printf("t_delta %ju.%016jx too long\n", |
---|
890 | (uintmax_t)t_delta.sec, |
---|
891 | (uintmax_t)t_delta.frac); |
---|
892 | } else if (t_delta.sec < 15 || |
---|
893 | (t_delta.sec == 15 && t_delta.frac <= (0xffLL << 56))) { |
---|
894 | /* too short */ |
---|
895 | if (bootverbose) |
---|
896 | printf("t_delta %ju.%016jx too short\n", |
---|
897 | (uintmax_t)t_delta.sec, |
---|
898 | (uintmax_t)t_delta.frac); |
---|
899 | } else { |
---|
900 | /* just right */ |
---|
901 | /* |
---|
902 | * Headroom: |
---|
903 | * 2^(64-20) / 16[s] = |
---|
904 | * 2^(44) / 16[s] = |
---|
905 | * 17.592.186.044.416 / 16 = |
---|
906 | * 1.099.511.627.776 [Hz] |
---|
907 | */ |
---|
908 | divi = t_delta.sec << 20; |
---|
909 | divi |= t_delta.frac >> (64 - 20); |
---|
910 | c_delta <<= 20; |
---|
911 | c_delta /= divi; |
---|
912 | if (c_delta > cpu_tick_frequency) { |
---|
913 | if (0 && bootverbose) |
---|
914 | printf("cpu_tick increased to %ju Hz\n", |
---|
915 | c_delta); |
---|
916 | cpu_tick_frequency = c_delta; |
---|
917 | } |
---|
918 | } |
---|
919 | } |
---|
920 | c_last = c_this; |
---|
921 | t_last = t_this; |
---|
922 | } |
---|
923 | |
---|
924 | void |
---|
925 | set_cputicker(cpu_tick_f *func, uint64_t freq, unsigned var) |
---|
926 | { |
---|
927 | |
---|
928 | if (func == NULL) { |
---|
929 | cpu_ticks = tc_cpu_ticks; |
---|
930 | } else { |
---|
931 | cpu_tick_frequency = freq; |
---|
932 | cpu_tick_variable = var; |
---|
933 | cpu_ticks = func; |
---|
934 | } |
---|
935 | } |
---|
936 | |
---|
937 | uint64_t |
---|
938 | cpu_tickrate(void) |
---|
939 | { |
---|
940 | |
---|
941 | if (cpu_ticks == tc_cpu_ticks) |
---|
942 | return (tc_getfrequency()); |
---|
943 | return (cpu_tick_frequency); |
---|
944 | } |
---|
945 | |
---|
946 | /* |
---|
947 | * We need to be slightly careful converting cputicks to microseconds. |
---|
948 | * There is plenty of margin in 64 bits of microseconds (half a million |
---|
949 | * years) and in 64 bits at 4 GHz (146 years), but if we do a multiply |
---|
950 | * before divide conversion (to retain precision) we find that the |
---|
951 | * margin shrinks to 1.5 hours (one millionth of 146y). |
---|
952 | * With a three prong approach we never lose significant bits, no |
---|
953 | * matter what the cputick rate and length of timeinterval is. |
---|
954 | */ |
---|
955 | |
---|
956 | uint64_t |
---|
957 | cputick2usec(uint64_t tick) |
---|
958 | { |
---|
959 | |
---|
960 | if (tick > 18446744073709551LL) /* floor(2^64 / 1000) */ |
---|
961 | return (tick / (cpu_tickrate() / 1000000LL)); |
---|
962 | else if (tick > 18446744073709LL) /* floor(2^64 / 1000000) */ |
---|
963 | return ((tick * 1000LL) / (cpu_tickrate() / 1000LL)); |
---|
964 | else |
---|
965 | return ((tick * 1000000LL) / cpu_tickrate()); |
---|
966 | } |
---|
967 | |
---|
968 | cpu_tick_f *cpu_ticks = tc_cpu_ticks; |
---|