xref: /freebsd/sys/kern/kern_ffclock.c (revision 389e4940069316fe667ffa263fa7d6390d0a960f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 The University of Melbourne
5  * All rights reserved.
6  *
7  * This software was developed by Julien Ridoux at the University of Melbourne
8  * under sponsorship from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_ffclock.h"
36 
37 #include <sys/param.h>
38 #include <sys/bus.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/module.h>
42 #include <sys/mutex.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/sbuf.h>
46 #include <sys/sysent.h>
47 #include <sys/sysproto.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 #include <sys/timeffc.h>
51 
52 #ifdef FFCLOCK
53 
54 FEATURE(ffclock, "Feed-forward clock support");
55 
56 extern struct ffclock_estimate ffclock_estimate;
57 extern struct bintime ffclock_boottime;
58 extern int8_t ffclock_updated;
59 extern struct mtx ffclock_mtx;
60 
61 /*
62  * Feed-forward clock absolute time. This should be the preferred way to read
63  * the feed-forward clock for "wall-clock" type time. The flags allow to compose
64  * various flavours of absolute time (e.g. with or without leap seconds taken
65  * into account). If valid pointers are provided, the ffcounter value and an
66  * upper bound on clock error associated with the bintime are provided.
67  * NOTE: use ffclock_convert_abs() to differ the conversion of a ffcounter value
68  * read earlier.
69  */
70 void
71 ffclock_abstime(ffcounter *ffcount, struct bintime *bt,
72     struct bintime *error_bound, uint32_t flags)
73 {
74 	struct ffclock_estimate cest;
75 	ffcounter ffc;
76 	ffcounter update_ffcount;
77 	ffcounter ffdelta_error;
78 
79 	/* Get counter and corresponding time. */
80 	if ((flags & FFCLOCK_FAST) == FFCLOCK_FAST)
81 		ffclock_last_tick(&ffc, bt, flags);
82 	else {
83 		ffclock_read_counter(&ffc);
84 		ffclock_convert_abs(ffc, bt, flags);
85 	}
86 
87 	/* Current ffclock estimate, use update_ffcount as generation number. */
88 	do {
89 		update_ffcount = ffclock_estimate.update_ffcount;
90 		bcopy(&ffclock_estimate, &cest, sizeof(struct ffclock_estimate));
91 	} while (update_ffcount != ffclock_estimate.update_ffcount);
92 
93 	/*
94 	 * Leap second adjustment. Total as seen by synchronisation algorithm
95 	 * since it started. cest.leapsec_next is the ffcounter prediction of
96 	 * when the next leapsecond occurs.
97 	 */
98 	if ((flags & FFCLOCK_LEAPSEC) == FFCLOCK_LEAPSEC) {
99 		bt->sec -= cest.leapsec_total;
100 		if (ffc > cest.leapsec_next)
101 			bt->sec -= cest.leapsec;
102 	}
103 
104 	/* Boot time adjustment, for uptime/monotonic clocks. */
105 	if ((flags & FFCLOCK_UPTIME) == FFCLOCK_UPTIME) {
106 		bintime_sub(bt, &ffclock_boottime);
107 	}
108 
109 	/* Compute error bound if a valid pointer has been passed. */
110 	if (error_bound) {
111 		ffdelta_error = ffc - cest.update_ffcount;
112 		ffclock_convert_diff(ffdelta_error, error_bound);
113 		/* 18446744073709 = int(2^64/1e12), err_bound_rate in [ps/s] */
114 		bintime_mul(error_bound, cest.errb_rate *
115 		    (uint64_t)18446744073709LL);
116 		/* 18446744073 = int(2^64 / 1e9), since err_abs in [ns] */
117 		bintime_addx(error_bound, cest.errb_abs *
118 		    (uint64_t)18446744073LL);
119 	}
120 
121 	if (ffcount)
122 		*ffcount = ffc;
123 }
124 
125 /*
126  * Feed-forward difference clock. This should be the preferred way to convert a
127  * time interval in ffcounter values into a time interval in seconds. If a valid
128  * pointer is passed, an upper bound on the error in computing the time interval
129  * in seconds is provided.
130  */
131 void
132 ffclock_difftime(ffcounter ffdelta, struct bintime *bt,
133     struct bintime *error_bound)
134 {
135 	ffcounter update_ffcount;
136 	uint32_t err_rate;
137 
138 	ffclock_convert_diff(ffdelta, bt);
139 
140 	if (error_bound) {
141 		do {
142 			update_ffcount = ffclock_estimate.update_ffcount;
143 			err_rate = ffclock_estimate.errb_rate;
144 		} while (update_ffcount != ffclock_estimate.update_ffcount);
145 
146 		ffclock_convert_diff(ffdelta, error_bound);
147 		/* 18446744073709 = int(2^64/1e12), err_bound_rate in [ps/s] */
148 		bintime_mul(error_bound, err_rate * (uint64_t)18446744073709LL);
149 	}
150 }
151 
152 /*
153  * Create a new kern.sysclock sysctl node, which will be home to some generic
154  * sysclock configuration variables. Feed-forward clock specific variables will
155  * live under the ffclock subnode.
156  */
157 
158 SYSCTL_NODE(_kern, OID_AUTO, sysclock, CTLFLAG_RW, 0,
159     "System clock related configuration");
160 SYSCTL_NODE(_kern_sysclock, OID_AUTO, ffclock, CTLFLAG_RW, 0,
161     "Feed-forward clock configuration");
162 
163 static char *sysclocks[] = {"feedback", "feed-forward"};
164 #define	MAX_SYSCLOCK_NAME_LEN 16
165 #define	NUM_SYSCLOCKS nitems(sysclocks)
166 
167 static int ffclock_version = 2;
168 SYSCTL_INT(_kern_sysclock_ffclock, OID_AUTO, version, CTLFLAG_RD,
169     &ffclock_version, 0, "Feed-forward clock kernel version");
170 
171 /* List available sysclocks. */
172 static int
173 sysctl_kern_sysclock_available(SYSCTL_HANDLER_ARGS)
174 {
175 	struct sbuf *s;
176 	int clk, error;
177 
178 	s = sbuf_new_for_sysctl(NULL, NULL,
179 	    MAX_SYSCLOCK_NAME_LEN * NUM_SYSCLOCKS, req);
180 	if (s == NULL)
181 		return (ENOMEM);
182 
183 	for (clk = 0; clk < NUM_SYSCLOCKS; clk++) {
184 		sbuf_cat(s, sysclocks[clk]);
185 		if (clk + 1 < NUM_SYSCLOCKS)
186 			sbuf_cat(s, " ");
187 	}
188 	error = sbuf_finish(s);
189 	sbuf_delete(s);
190 
191 	return (error);
192 }
193 
194 SYSCTL_PROC(_kern_sysclock, OID_AUTO, available, CTLTYPE_STRING | CTLFLAG_RD,
195     0, 0, sysctl_kern_sysclock_available, "A",
196     "List of available system clocks");
197 
198 /*
199  * Return the name of the active system clock if read, or attempt to change
200  * the active system clock to the user specified one if written to. The active
201  * system clock is read when calling any of the [get]{bin,nano,micro}[up]time()
202  * functions.
203  */
204 static int
205 sysctl_kern_sysclock_active(SYSCTL_HANDLER_ARGS)
206 {
207 	char newclock[MAX_SYSCLOCK_NAME_LEN];
208 	int error;
209 	int clk;
210 
211 	/* Return the name of the current active sysclock. */
212 	strlcpy(newclock, sysclocks[sysclock_active], sizeof(newclock));
213 	error = sysctl_handle_string(oidp, newclock, sizeof(newclock), req);
214 
215 	/* Check for error or no change */
216 	if (error != 0 || req->newptr == NULL)
217 		goto done;
218 
219 	/* Change the active sysclock to the user specified one: */
220 	error = EINVAL;
221 	for (clk = 0; clk < NUM_SYSCLOCKS; clk++) {
222 		if (strncmp(newclock, sysclocks[clk],
223 		    MAX_SYSCLOCK_NAME_LEN - 1)) {
224 			continue;
225 		}
226 		sysclock_active = clk;
227 		error = 0;
228 		break;
229 	}
230 done:
231 	return (error);
232 }
233 
234 SYSCTL_PROC(_kern_sysclock, OID_AUTO, active, CTLTYPE_STRING | CTLFLAG_RW,
235     0, 0, sysctl_kern_sysclock_active, "A",
236     "Name of the active system clock which is currently serving time");
237 
238 static int sysctl_kern_ffclock_ffcounter_bypass = 0;
239 SYSCTL_INT(_kern_sysclock_ffclock, OID_AUTO, ffcounter_bypass, CTLFLAG_RW,
240     &sysctl_kern_ffclock_ffcounter_bypass, 0,
241     "Use reliable hardware timecounter as the feed-forward counter");
242 
243 /*
244  * High level functions to access the Feed-Forward Clock.
245  */
246 void
247 ffclock_bintime(struct bintime *bt)
248 {
249 
250 	ffclock_abstime(NULL, bt, NULL, FFCLOCK_LERP | FFCLOCK_LEAPSEC);
251 }
252 
253 void
254 ffclock_nanotime(struct timespec *tsp)
255 {
256 	struct bintime bt;
257 
258 	ffclock_abstime(NULL, &bt, NULL, FFCLOCK_LERP | FFCLOCK_LEAPSEC);
259 	bintime2timespec(&bt, tsp);
260 }
261 
262 void
263 ffclock_microtime(struct timeval *tvp)
264 {
265 	struct bintime bt;
266 
267 	ffclock_abstime(NULL, &bt, NULL, FFCLOCK_LERP | FFCLOCK_LEAPSEC);
268 	bintime2timeval(&bt, tvp);
269 }
270 
271 void
272 ffclock_getbintime(struct bintime *bt)
273 {
274 
275 	ffclock_abstime(NULL, bt, NULL,
276 	    FFCLOCK_LERP | FFCLOCK_LEAPSEC | FFCLOCK_FAST);
277 }
278 
279 void
280 ffclock_getnanotime(struct timespec *tsp)
281 {
282 	struct bintime bt;
283 
284 	ffclock_abstime(NULL, &bt, NULL,
285 	    FFCLOCK_LERP | FFCLOCK_LEAPSEC | FFCLOCK_FAST);
286 	bintime2timespec(&bt, tsp);
287 }
288 
289 void
290 ffclock_getmicrotime(struct timeval *tvp)
291 {
292 	struct bintime bt;
293 
294 	ffclock_abstime(NULL, &bt, NULL,
295 	    FFCLOCK_LERP | FFCLOCK_LEAPSEC | FFCLOCK_FAST);
296 	bintime2timeval(&bt, tvp);
297 }
298 
299 void
300 ffclock_binuptime(struct bintime *bt)
301 {
302 
303 	ffclock_abstime(NULL, bt, NULL, FFCLOCK_LERP | FFCLOCK_UPTIME);
304 }
305 
306 void
307 ffclock_nanouptime(struct timespec *tsp)
308 {
309 	struct bintime bt;
310 
311 	ffclock_abstime(NULL, &bt, NULL, FFCLOCK_LERP | FFCLOCK_UPTIME);
312 	bintime2timespec(&bt, tsp);
313 }
314 
315 void
316 ffclock_microuptime(struct timeval *tvp)
317 {
318 	struct bintime bt;
319 
320 	ffclock_abstime(NULL, &bt, NULL, FFCLOCK_LERP | FFCLOCK_UPTIME);
321 	bintime2timeval(&bt, tvp);
322 }
323 
324 void
325 ffclock_getbinuptime(struct bintime *bt)
326 {
327 
328 	ffclock_abstime(NULL, bt, NULL,
329 	    FFCLOCK_LERP | FFCLOCK_UPTIME | FFCLOCK_FAST);
330 }
331 
332 void
333 ffclock_getnanouptime(struct timespec *tsp)
334 {
335 	struct bintime bt;
336 
337 	ffclock_abstime(NULL, &bt, NULL,
338 	    FFCLOCK_LERP | FFCLOCK_UPTIME | FFCLOCK_FAST);
339 	bintime2timespec(&bt, tsp);
340 }
341 
342 void
343 ffclock_getmicrouptime(struct timeval *tvp)
344 {
345 	struct bintime bt;
346 
347 	ffclock_abstime(NULL, &bt, NULL,
348 	    FFCLOCK_LERP | FFCLOCK_UPTIME | FFCLOCK_FAST);
349 	bintime2timeval(&bt, tvp);
350 }
351 
352 void
353 ffclock_bindifftime(ffcounter ffdelta, struct bintime *bt)
354 {
355 
356 	ffclock_difftime(ffdelta, bt, NULL);
357 }
358 
359 void
360 ffclock_nanodifftime(ffcounter ffdelta, struct timespec *tsp)
361 {
362 	struct bintime bt;
363 
364 	ffclock_difftime(ffdelta, &bt, NULL);
365 	bintime2timespec(&bt, tsp);
366 }
367 
368 void
369 ffclock_microdifftime(ffcounter ffdelta, struct timeval *tvp)
370 {
371 	struct bintime bt;
372 
373 	ffclock_difftime(ffdelta, &bt, NULL);
374 	bintime2timeval(&bt, tvp);
375 }
376 
377 /*
378  * System call allowing userland applications to retrieve the current value of
379  * the Feed-Forward Clock counter.
380  */
381 #ifndef _SYS_SYSPROTO_H_
382 struct ffclock_getcounter_args {
383 	ffcounter *ffcount;
384 };
385 #endif
386 /* ARGSUSED */
387 int
388 sys_ffclock_getcounter(struct thread *td, struct ffclock_getcounter_args *uap)
389 {
390 	ffcounter ffcount;
391 	int error;
392 
393 	ffcount = 0;
394 	ffclock_read_counter(&ffcount);
395 	if (ffcount == 0)
396 		return (EAGAIN);
397 	error = copyout(&ffcount, uap->ffcount, sizeof(ffcounter));
398 
399 	return (error);
400 }
401 
402 /*
403  * System call allowing the synchronisation daemon to push new feed-foward clock
404  * estimates to the kernel. Acquire ffclock_mtx to prevent concurrent updates
405  * and ensure data consistency.
406  * NOTE: ffclock_updated signals the fftimehands that new estimates are
407  * available. The updated estimates are picked up by the fftimehands on next
408  * tick, which could take as long as 1/hz seconds (if ticks are not missed).
409  */
410 #ifndef _SYS_SYSPROTO_H_
411 struct ffclock_setestimate_args {
412 	struct ffclock_estimate *cest;
413 };
414 #endif
415 /* ARGSUSED */
416 int
417 sys_ffclock_setestimate(struct thread *td, struct ffclock_setestimate_args *uap)
418 {
419 	struct ffclock_estimate cest;
420 	int error;
421 
422 	/* Reuse of PRIV_CLOCK_SETTIME. */
423 	if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
424 		return (error);
425 
426 	if ((error = copyin(uap->cest, &cest, sizeof(struct ffclock_estimate)))
427 	    != 0)
428 		return (error);
429 
430 	mtx_lock(&ffclock_mtx);
431 	memcpy(&ffclock_estimate, &cest, sizeof(struct ffclock_estimate));
432 	ffclock_updated++;
433 	mtx_unlock(&ffclock_mtx);
434 	return (error);
435 }
436 
437 /*
438  * System call allowing userland applications to retrieve the clock estimates
439  * stored within the kernel. It is useful to kickstart the synchronisation
440  * daemon with the kernel's knowledge of hardware timecounter.
441  */
442 #ifndef _SYS_SYSPROTO_H_
443 struct ffclock_getestimate_args {
444 	struct ffclock_estimate *cest;
445 };
446 #endif
447 /* ARGSUSED */
448 int
449 sys_ffclock_getestimate(struct thread *td, struct ffclock_getestimate_args *uap)
450 {
451 	struct ffclock_estimate cest;
452 	int error;
453 
454 	mtx_lock(&ffclock_mtx);
455 	memcpy(&cest, &ffclock_estimate, sizeof(struct ffclock_estimate));
456 	mtx_unlock(&ffclock_mtx);
457 	error = copyout(&cest, uap->cest, sizeof(struct ffclock_estimate));
458 	return (error);
459 }
460 
461 #else /* !FFCLOCK */
462 
463 int
464 sys_ffclock_getcounter(struct thread *td, struct ffclock_getcounter_args *uap)
465 {
466 
467 	return (ENOSYS);
468 }
469 
470 int
471 sys_ffclock_setestimate(struct thread *td, struct ffclock_setestimate_args *uap)
472 {
473 
474 	return (ENOSYS);
475 }
476 
477 int
478 sys_ffclock_getestimate(struct thread *td, struct ffclock_getestimate_args *uap)
479 {
480 
481 	return (ENOSYS);
482 }
483 
484 #endif /* FFCLOCK */
485