Contents of /alx-src/tags/kernel26-2.6.12-alx-r9/kernel/posix-timers.c
Parent Directory | Revision Log
Revision 630 -
(show annotations)
(download)
Wed Mar 4 11:03:09 2009 UTC (15 years, 6 months ago) by niro
File MIME type: text/plain
File size: 45992 byte(s)
Wed Mar 4 11:03:09 2009 UTC (15 years, 6 months ago) by niro
File MIME type: text/plain
File size: 45992 byte(s)
Tag kernel26-2.6.12-alx-r9
1 | /* |
2 | * linux/kernel/posix_timers.c |
3 | * |
4 | * |
5 | * 2002-10-15 Posix Clocks & timers |
6 | * by George Anzinger george@mvista.com |
7 | * |
8 | * Copyright (C) 2002 2003 by MontaVista Software. |
9 | * |
10 | * 2004-06-01 Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug. |
11 | * Copyright (C) 2004 Boris Hu |
12 | * |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by |
15 | * the Free Software Foundation; either version 2 of the License, or (at |
16 | * your option) any later version. |
17 | * |
18 | * This program is distributed in the hope that it will be useful, but |
19 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
21 | * General Public License for more details. |
22 | |
23 | * You should have received a copy of the GNU General Public License |
24 | * along with this program; if not, write to the Free Software |
25 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
26 | * |
27 | * MontaVista Software | 1237 East Arques Avenue | Sunnyvale | CA 94085 | USA |
28 | */ |
29 | |
30 | /* These are all the functions necessary to implement |
31 | * POSIX clocks & timers |
32 | */ |
33 | #include <linux/mm.h> |
34 | #include <linux/smp_lock.h> |
35 | #include <linux/interrupt.h> |
36 | #include <linux/slab.h> |
37 | #include <linux/time.h> |
38 | |
39 | #include <asm/uaccess.h> |
40 | #include <asm/semaphore.h> |
41 | #include <linux/list.h> |
42 | #include <linux/init.h> |
43 | #include <linux/compiler.h> |
44 | #include <linux/idr.h> |
45 | #include <linux/posix-timers.h> |
46 | #include <linux/syscalls.h> |
47 | #include <linux/wait.h> |
48 | #include <linux/workqueue.h> |
49 | #include <linux/module.h> |
50 | |
51 | #ifndef div_long_long_rem |
52 | #include <asm/div64.h> |
53 | |
54 | #define div_long_long_rem(dividend,divisor,remainder) ({ \ |
55 | u64 result = dividend; \ |
56 | *remainder = do_div(result,divisor); \ |
57 | result; }) |
58 | |
59 | #endif |
60 | #define CLOCK_REALTIME_RES TICK_NSEC /* In nano seconds. */ |
61 | |
62 | static inline u64 mpy_l_X_l_ll(unsigned long mpy1,unsigned long mpy2) |
63 | { |
64 | return (u64)mpy1 * mpy2; |
65 | } |
66 | /* |
67 | * Management arrays for POSIX timers. Timers are kept in slab memory |
68 | * Timer ids are allocated by an external routine that keeps track of the |
69 | * id and the timer. The external interface is: |
70 | * |
71 | * void *idr_find(struct idr *idp, int id); to find timer_id <id> |
72 | * int idr_get_new(struct idr *idp, void *ptr); to get a new id and |
73 | * related it to <ptr> |
74 | * void idr_remove(struct idr *idp, int id); to release <id> |
75 | * void idr_init(struct idr *idp); to initialize <idp> |
76 | * which we supply. |
77 | * The idr_get_new *may* call slab for more memory so it must not be |
78 | * called under a spin lock. Likewise idr_remore may release memory |
79 | * (but it may be ok to do this under a lock...). |
80 | * idr_find is just a memory look up and is quite fast. A -1 return |
81 | * indicates that the requested id does not exist. |
82 | */ |
83 | |
84 | /* |
85 | * Lets keep our timers in a slab cache :-) |
86 | */ |
87 | static kmem_cache_t *posix_timers_cache; |
88 | static struct idr posix_timers_id; |
89 | static DEFINE_SPINLOCK(idr_lock); |
90 | |
91 | /* |
92 | * Just because the timer is not in the timer list does NOT mean it is |
93 | * inactive. It could be in the "fire" routine getting a new expire time. |
94 | */ |
95 | #define TIMER_INACTIVE 1 |
96 | |
97 | #ifdef CONFIG_SMP |
98 | # define timer_active(tmr) \ |
99 | ((tmr)->it.real.timer.entry.prev != (void *)TIMER_INACTIVE) |
100 | # define set_timer_inactive(tmr) \ |
101 | do { \ |
102 | (tmr)->it.real.timer.entry.prev = (void *)TIMER_INACTIVE; \ |
103 | } while (0) |
104 | #else |
105 | # define timer_active(tmr) BARFY // error to use outside of SMP |
106 | # define set_timer_inactive(tmr) do { } while (0) |
107 | #endif |
108 | /* |
109 | * we assume that the new SIGEV_THREAD_ID shares no bits with the other |
110 | * SIGEV values. Here we put out an error if this assumption fails. |
111 | */ |
112 | #if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \ |
113 | ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD)) |
114 | #error "SIGEV_THREAD_ID must not share bit with other SIGEV values!" |
115 | #endif |
116 | |
117 | |
118 | /* |
119 | * The timer ID is turned into a timer address by idr_find(). |
120 | * Verifying a valid ID consists of: |
121 | * |
122 | * a) checking that idr_find() returns other than -1. |
123 | * b) checking that the timer id matches the one in the timer itself. |
124 | * c) that the timer owner is in the callers thread group. |
125 | */ |
126 | |
127 | /* |
128 | * CLOCKs: The POSIX standard calls for a couple of clocks and allows us |
129 | * to implement others. This structure defines the various |
130 | * clocks and allows the possibility of adding others. We |
131 | * provide an interface to add clocks to the table and expect |
132 | * the "arch" code to add at least one clock that is high |
133 | * resolution. Here we define the standard CLOCK_REALTIME as a |
134 | * 1/HZ resolution clock. |
135 | * |
136 | * RESOLUTION: Clock resolution is used to round up timer and interval |
137 | * times, NOT to report clock times, which are reported with as |
138 | * much resolution as the system can muster. In some cases this |
139 | * resolution may depend on the underlying clock hardware and |
140 | * may not be quantifiable until run time, and only then is the |
141 | * necessary code is written. The standard says we should say |
142 | * something about this issue in the documentation... |
143 | * |
144 | * FUNCTIONS: The CLOCKs structure defines possible functions to handle |
145 | * various clock functions. For clocks that use the standard |
146 | * system timer code these entries should be NULL. This will |
147 | * allow dispatch without the overhead of indirect function |
148 | * calls. CLOCKS that depend on other sources (e.g. WWV or GPS) |
149 | * must supply functions here, even if the function just returns |
150 | * ENOSYS. The standard POSIX timer management code assumes the |
151 | * following: 1.) The k_itimer struct (sched.h) is used for the |
152 | * timer. 2.) The list, it_lock, it_clock, it_id and it_process |
153 | * fields are not modified by timer code. |
154 | * |
155 | * At this time all functions EXCEPT clock_nanosleep can be |
156 | * redirected by the CLOCKS structure. Clock_nanosleep is in |
157 | * there, but the code ignores it. |
158 | * |
159 | * Permissions: It is assumed that the clock_settime() function defined |
160 | * for each clock will take care of permission checks. Some |
161 | * clocks may be set able by any user (i.e. local process |
162 | * clocks) others not. Currently the only set able clock we |
163 | * have is CLOCK_REALTIME and its high res counter part, both of |
164 | * which we beg off on and pass to do_sys_settimeofday(). |
165 | */ |
166 | |
167 | static struct k_clock posix_clocks[MAX_CLOCKS]; |
168 | /* |
169 | * We only have one real clock that can be set so we need only one abs list, |
170 | * even if we should want to have several clocks with differing resolutions. |
171 | */ |
172 | static struct k_clock_abs abs_list = {.list = LIST_HEAD_INIT(abs_list.list), |
173 | .lock = SPIN_LOCK_UNLOCKED}; |
174 | |
175 | static void posix_timer_fn(unsigned long); |
176 | static u64 do_posix_clock_monotonic_gettime_parts( |
177 | struct timespec *tp, struct timespec *mo); |
178 | int do_posix_clock_monotonic_gettime(struct timespec *tp); |
179 | static int do_posix_clock_monotonic_get(clockid_t, struct timespec *tp); |
180 | |
181 | static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags); |
182 | |
183 | static inline void unlock_timer(struct k_itimer *timr, unsigned long flags) |
184 | { |
185 | spin_unlock_irqrestore(&timr->it_lock, flags); |
186 | } |
187 | |
188 | /* |
189 | * Call the k_clock hook function if non-null, or the default function. |
190 | */ |
191 | #define CLOCK_DISPATCH(clock, call, arglist) \ |
192 | ((clock) < 0 ? posix_cpu_##call arglist : \ |
193 | (posix_clocks[clock].call != NULL \ |
194 | ? (*posix_clocks[clock].call) arglist : common_##call arglist)) |
195 | |
196 | /* |
197 | * Default clock hook functions when the struct k_clock passed |
198 | * to register_posix_clock leaves a function pointer null. |
199 | * |
200 | * The function common_CALL is the default implementation for |
201 | * the function pointer CALL in struct k_clock. |
202 | */ |
203 | |
204 | static inline int common_clock_getres(clockid_t which_clock, |
205 | struct timespec *tp) |
206 | { |
207 | tp->tv_sec = 0; |
208 | tp->tv_nsec = posix_clocks[which_clock].res; |
209 | return 0; |
210 | } |
211 | |
212 | static inline int common_clock_get(clockid_t which_clock, struct timespec *tp) |
213 | { |
214 | getnstimeofday(tp); |
215 | return 0; |
216 | } |
217 | |
218 | static inline int common_clock_set(clockid_t which_clock, struct timespec *tp) |
219 | { |
220 | return do_sys_settimeofday(tp, NULL); |
221 | } |
222 | |
223 | static inline int common_timer_create(struct k_itimer *new_timer) |
224 | { |
225 | INIT_LIST_HEAD(&new_timer->it.real.abs_timer_entry); |
226 | init_timer(&new_timer->it.real.timer); |
227 | new_timer->it.real.timer.data = (unsigned long) new_timer; |
228 | new_timer->it.real.timer.function = posix_timer_fn; |
229 | set_timer_inactive(new_timer); |
230 | return 0; |
231 | } |
232 | |
233 | /* |
234 | * These ones are defined below. |
235 | */ |
236 | static int common_nsleep(clockid_t, int flags, struct timespec *t); |
237 | static void common_timer_get(struct k_itimer *, struct itimerspec *); |
238 | static int common_timer_set(struct k_itimer *, int, |
239 | struct itimerspec *, struct itimerspec *); |
240 | static int common_timer_del(struct k_itimer *timer); |
241 | |
242 | /* |
243 | * Return nonzero iff we know a priori this clockid_t value is bogus. |
244 | */ |
245 | static inline int invalid_clockid(clockid_t which_clock) |
246 | { |
247 | if (which_clock < 0) /* CPU clock, posix_cpu_* will check it */ |
248 | return 0; |
249 | if ((unsigned) which_clock >= MAX_CLOCKS) |
250 | return 1; |
251 | if (posix_clocks[which_clock].clock_getres != NULL) |
252 | return 0; |
253 | #ifndef CLOCK_DISPATCH_DIRECT |
254 | if (posix_clocks[which_clock].res != 0) |
255 | return 0; |
256 | #endif |
257 | return 1; |
258 | } |
259 | |
260 | |
261 | /* |
262 | * Initialize everything, well, just everything in Posix clocks/timers ;) |
263 | */ |
264 | static __init int init_posix_timers(void) |
265 | { |
266 | struct k_clock clock_realtime = {.res = CLOCK_REALTIME_RES, |
267 | .abs_struct = &abs_list |
268 | }; |
269 | struct k_clock clock_monotonic = {.res = CLOCK_REALTIME_RES, |
270 | .abs_struct = NULL, |
271 | .clock_get = do_posix_clock_monotonic_get, |
272 | .clock_set = do_posix_clock_nosettime |
273 | }; |
274 | |
275 | register_posix_clock(CLOCK_REALTIME, &clock_realtime); |
276 | register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic); |
277 | |
278 | posix_timers_cache = kmem_cache_create("posix_timers_cache", |
279 | sizeof (struct k_itimer), 0, 0, NULL, NULL); |
280 | idr_init(&posix_timers_id); |
281 | return 0; |
282 | } |
283 | |
284 | __initcall(init_posix_timers); |
285 | |
286 | static void tstojiffie(struct timespec *tp, int res, u64 *jiff) |
287 | { |
288 | long sec = tp->tv_sec; |
289 | long nsec = tp->tv_nsec + res - 1; |
290 | |
291 | if (nsec > NSEC_PER_SEC) { |
292 | sec++; |
293 | nsec -= NSEC_PER_SEC; |
294 | } |
295 | |
296 | /* |
297 | * The scaling constants are defined in <linux/time.h> |
298 | * The difference between there and here is that we do the |
299 | * res rounding and compute a 64-bit result (well so does that |
300 | * but it then throws away the high bits). |
301 | */ |
302 | *jiff = (mpy_l_X_l_ll(sec, SEC_CONVERSION) + |
303 | (mpy_l_X_l_ll(nsec, NSEC_CONVERSION) >> |
304 | (NSEC_JIFFIE_SC - SEC_JIFFIE_SC))) >> SEC_JIFFIE_SC; |
305 | } |
306 | |
307 | /* |
308 | * This function adjusts the timer as needed as a result of the clock |
309 | * being set. It should only be called for absolute timers, and then |
310 | * under the abs_list lock. It computes the time difference and sets |
311 | * the new jiffies value in the timer. It also updates the timers |
312 | * reference wall_to_monotonic value. It is complicated by the fact |
313 | * that tstojiffies() only handles positive times and it needs to work |
314 | * with both positive and negative times. Also, for negative offsets, |
315 | * we need to defeat the res round up. |
316 | * |
317 | * Return is true if there is a new time, else false. |
318 | */ |
319 | static long add_clockset_delta(struct k_itimer *timr, |
320 | struct timespec *new_wall_to) |
321 | { |
322 | struct timespec delta; |
323 | int sign = 0; |
324 | u64 exp; |
325 | |
326 | set_normalized_timespec(&delta, |
327 | new_wall_to->tv_sec - |
328 | timr->it.real.wall_to_prev.tv_sec, |
329 | new_wall_to->tv_nsec - |
330 | timr->it.real.wall_to_prev.tv_nsec); |
331 | if (likely(!(delta.tv_sec | delta.tv_nsec))) |
332 | return 0; |
333 | if (delta.tv_sec < 0) { |
334 | set_normalized_timespec(&delta, |
335 | -delta.tv_sec, |
336 | 1 - delta.tv_nsec - |
337 | posix_clocks[timr->it_clock].res); |
338 | sign++; |
339 | } |
340 | tstojiffie(&delta, posix_clocks[timr->it_clock].res, &exp); |
341 | timr->it.real.wall_to_prev = *new_wall_to; |
342 | timr->it.real.timer.expires += (sign ? -exp : exp); |
343 | return 1; |
344 | } |
345 | |
346 | static void remove_from_abslist(struct k_itimer *timr) |
347 | { |
348 | if (!list_empty(&timr->it.real.abs_timer_entry)) { |
349 | spin_lock(&abs_list.lock); |
350 | list_del_init(&timr->it.real.abs_timer_entry); |
351 | spin_unlock(&abs_list.lock); |
352 | } |
353 | } |
354 | |
355 | static void schedule_next_timer(struct k_itimer *timr) |
356 | { |
357 | struct timespec new_wall_to; |
358 | struct now_struct now; |
359 | unsigned long seq; |
360 | |
361 | /* |
362 | * Set up the timer for the next interval (if there is one). |
363 | * Note: this code uses the abs_timer_lock to protect |
364 | * it.real.wall_to_prev and must hold it until exp is set, not exactly |
365 | * obvious... |
366 | |
367 | * This function is used for CLOCK_REALTIME* and |
368 | * CLOCK_MONOTONIC* timers. If we ever want to handle other |
369 | * CLOCKs, the calling code (do_schedule_next_timer) would need |
370 | * to pull the "clock" info from the timer and dispatch the |
371 | * "other" CLOCKs "next timer" code (which, I suppose should |
372 | * also be added to the k_clock structure). |
373 | */ |
374 | if (!timr->it.real.incr) |
375 | return; |
376 | |
377 | do { |
378 | seq = read_seqbegin(&xtime_lock); |
379 | new_wall_to = wall_to_monotonic; |
380 | posix_get_now(&now); |
381 | } while (read_seqretry(&xtime_lock, seq)); |
382 | |
383 | if (!list_empty(&timr->it.real.abs_timer_entry)) { |
384 | spin_lock(&abs_list.lock); |
385 | add_clockset_delta(timr, &new_wall_to); |
386 | |
387 | posix_bump_timer(timr, now); |
388 | |
389 | spin_unlock(&abs_list.lock); |
390 | } else { |
391 | posix_bump_timer(timr, now); |
392 | } |
393 | timr->it_overrun_last = timr->it_overrun; |
394 | timr->it_overrun = -1; |
395 | ++timr->it_requeue_pending; |
396 | add_timer(&timr->it.real.timer); |
397 | } |
398 | |
399 | /* |
400 | * This function is exported for use by the signal deliver code. It is |
401 | * called just prior to the info block being released and passes that |
402 | * block to us. It's function is to update the overrun entry AND to |
403 | * restart the timer. It should only be called if the timer is to be |
404 | * restarted (i.e. we have flagged this in the sys_private entry of the |
405 | * info block). |
406 | * |
407 | * To protect aginst the timer going away while the interrupt is queued, |
408 | * we require that the it_requeue_pending flag be set. |
409 | */ |
410 | void do_schedule_next_timer(struct siginfo *info) |
411 | { |
412 | struct k_itimer *timr; |
413 | unsigned long flags; |
414 | |
415 | timr = lock_timer(info->si_tid, &flags); |
416 | |
417 | if (!timr || timr->it_requeue_pending != info->si_sys_private) |
418 | goto exit; |
419 | |
420 | if (timr->it_clock < 0) /* CPU clock */ |
421 | posix_cpu_timer_schedule(timr); |
422 | else |
423 | schedule_next_timer(timr); |
424 | info->si_overrun = timr->it_overrun_last; |
425 | exit: |
426 | if (timr) |
427 | unlock_timer(timr, flags); |
428 | } |
429 | |
430 | int posix_timer_event(struct k_itimer *timr,int si_private) |
431 | { |
432 | memset(&timr->sigq->info, 0, sizeof(siginfo_t)); |
433 | timr->sigq->info.si_sys_private = si_private; |
434 | /* |
435 | * Send signal to the process that owns this timer. |
436 | |
437 | * This code assumes that all the possible abs_lists share the |
438 | * same lock (there is only one list at this time). If this is |
439 | * not the case, the CLOCK info would need to be used to find |
440 | * the proper abs list lock. |
441 | */ |
442 | |
443 | timr->sigq->info.si_signo = timr->it_sigev_signo; |
444 | timr->sigq->info.si_errno = 0; |
445 | timr->sigq->info.si_code = SI_TIMER; |
446 | timr->sigq->info.si_tid = timr->it_id; |
447 | timr->sigq->info.si_value = timr->it_sigev_value; |
448 | if (timr->it_sigev_notify & SIGEV_THREAD_ID) { |
449 | if (unlikely(timr->it_process->flags & PF_EXITING)) { |
450 | timr->it_sigev_notify = SIGEV_SIGNAL; |
451 | put_task_struct(timr->it_process); |
452 | timr->it_process = timr->it_process->group_leader; |
453 | goto group; |
454 | } |
455 | return send_sigqueue(timr->it_sigev_signo, timr->sigq, |
456 | timr->it_process); |
457 | } |
458 | else { |
459 | group: |
460 | return send_group_sigqueue(timr->it_sigev_signo, timr->sigq, |
461 | timr->it_process); |
462 | } |
463 | } |
464 | EXPORT_SYMBOL_GPL(posix_timer_event); |
465 | |
466 | /* |
467 | * This function gets called when a POSIX.1b interval timer expires. It |
468 | * is used as a callback from the kernel internal timer. The |
469 | * run_timer_list code ALWAYS calls with interrupts on. |
470 | |
471 | * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers. |
472 | */ |
473 | static void posix_timer_fn(unsigned long __data) |
474 | { |
475 | struct k_itimer *timr = (struct k_itimer *) __data; |
476 | unsigned long flags; |
477 | unsigned long seq; |
478 | struct timespec delta, new_wall_to; |
479 | u64 exp = 0; |
480 | int do_notify = 1; |
481 | |
482 | spin_lock_irqsave(&timr->it_lock, flags); |
483 | set_timer_inactive(timr); |
484 | if (!list_empty(&timr->it.real.abs_timer_entry)) { |
485 | spin_lock(&abs_list.lock); |
486 | do { |
487 | seq = read_seqbegin(&xtime_lock); |
488 | new_wall_to = wall_to_monotonic; |
489 | } while (read_seqretry(&xtime_lock, seq)); |
490 | set_normalized_timespec(&delta, |
491 | new_wall_to.tv_sec - |
492 | timr->it.real.wall_to_prev.tv_sec, |
493 | new_wall_to.tv_nsec - |
494 | timr->it.real.wall_to_prev.tv_nsec); |
495 | if (likely((delta.tv_sec | delta.tv_nsec ) == 0)) { |
496 | /* do nothing, timer is on time */ |
497 | } else if (delta.tv_sec < 0) { |
498 | /* do nothing, timer is already late */ |
499 | } else { |
500 | /* timer is early due to a clock set */ |
501 | tstojiffie(&delta, |
502 | posix_clocks[timr->it_clock].res, |
503 | &exp); |
504 | timr->it.real.wall_to_prev = new_wall_to; |
505 | timr->it.real.timer.expires += exp; |
506 | add_timer(&timr->it.real.timer); |
507 | do_notify = 0; |
508 | } |
509 | spin_unlock(&abs_list.lock); |
510 | |
511 | } |
512 | if (do_notify) { |
513 | int si_private=0; |
514 | |
515 | if (timr->it.real.incr) |
516 | si_private = ++timr->it_requeue_pending; |
517 | else { |
518 | remove_from_abslist(timr); |
519 | } |
520 | |
521 | if (posix_timer_event(timr, si_private)) |
522 | /* |
523 | * signal was not sent because of sig_ignor |
524 | * we will not get a call back to restart it AND |
525 | * it should be restarted. |
526 | */ |
527 | schedule_next_timer(timr); |
528 | } |
529 | unlock_timer(timr, flags); /* hold thru abs lock to keep irq off */ |
530 | } |
531 | |
532 | |
533 | static inline struct task_struct * good_sigevent(sigevent_t * event) |
534 | { |
535 | struct task_struct *rtn = current->group_leader; |
536 | |
537 | if ((event->sigev_notify & SIGEV_THREAD_ID ) && |
538 | (!(rtn = find_task_by_pid(event->sigev_notify_thread_id)) || |
539 | rtn->tgid != current->tgid || |
540 | (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL)) |
541 | return NULL; |
542 | |
543 | if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) && |
544 | ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX))) |
545 | return NULL; |
546 | |
547 | return rtn; |
548 | } |
549 | |
550 | void register_posix_clock(clockid_t clock_id, struct k_clock *new_clock) |
551 | { |
552 | if ((unsigned) clock_id >= MAX_CLOCKS) { |
553 | printk("POSIX clock register failed for clock_id %d\n", |
554 | clock_id); |
555 | return; |
556 | } |
557 | |
558 | posix_clocks[clock_id] = *new_clock; |
559 | } |
560 | EXPORT_SYMBOL_GPL(register_posix_clock); |
561 | |
562 | static struct k_itimer * alloc_posix_timer(void) |
563 | { |
564 | struct k_itimer *tmr; |
565 | tmr = kmem_cache_alloc(posix_timers_cache, GFP_KERNEL); |
566 | if (!tmr) |
567 | return tmr; |
568 | memset(tmr, 0, sizeof (struct k_itimer)); |
569 | if (unlikely(!(tmr->sigq = sigqueue_alloc()))) { |
570 | kmem_cache_free(posix_timers_cache, tmr); |
571 | tmr = NULL; |
572 | } |
573 | return tmr; |
574 | } |
575 | |
576 | #define IT_ID_SET 1 |
577 | #define IT_ID_NOT_SET 0 |
578 | static void release_posix_timer(struct k_itimer *tmr, int it_id_set) |
579 | { |
580 | if (it_id_set) { |
581 | unsigned long flags; |
582 | spin_lock_irqsave(&idr_lock, flags); |
583 | idr_remove(&posix_timers_id, tmr->it_id); |
584 | spin_unlock_irqrestore(&idr_lock, flags); |
585 | } |
586 | sigqueue_free(tmr->sigq); |
587 | if (unlikely(tmr->it_process) && |
588 | tmr->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) |
589 | put_task_struct(tmr->it_process); |
590 | kmem_cache_free(posix_timers_cache, tmr); |
591 | } |
592 | |
593 | /* Create a POSIX.1b interval timer. */ |
594 | |
595 | asmlinkage long |
596 | sys_timer_create(clockid_t which_clock, |
597 | struct sigevent __user *timer_event_spec, |
598 | timer_t __user * created_timer_id) |
599 | { |
600 | int error = 0; |
601 | struct k_itimer *new_timer = NULL; |
602 | int new_timer_id; |
603 | struct task_struct *process = NULL; |
604 | unsigned long flags; |
605 | sigevent_t event; |
606 | int it_id_set = IT_ID_NOT_SET; |
607 | |
608 | if (invalid_clockid(which_clock)) |
609 | return -EINVAL; |
610 | |
611 | new_timer = alloc_posix_timer(); |
612 | if (unlikely(!new_timer)) |
613 | return -EAGAIN; |
614 | |
615 | spin_lock_init(&new_timer->it_lock); |
616 | retry: |
617 | if (unlikely(!idr_pre_get(&posix_timers_id, GFP_KERNEL))) { |
618 | error = -EAGAIN; |
619 | goto out; |
620 | } |
621 | spin_lock_irq(&idr_lock); |
622 | error = idr_get_new(&posix_timers_id, |
623 | (void *) new_timer, |
624 | &new_timer_id); |
625 | spin_unlock_irq(&idr_lock); |
626 | if (error == -EAGAIN) |
627 | goto retry; |
628 | else if (error) { |
629 | /* |
630 | * Wierd looking, but we return EAGAIN if the IDR is |
631 | * full (proper POSIX return value for this) |
632 | */ |
633 | error = -EAGAIN; |
634 | goto out; |
635 | } |
636 | |
637 | it_id_set = IT_ID_SET; |
638 | new_timer->it_id = (timer_t) new_timer_id; |
639 | new_timer->it_clock = which_clock; |
640 | new_timer->it_overrun = -1; |
641 | error = CLOCK_DISPATCH(which_clock, timer_create, (new_timer)); |
642 | if (error) |
643 | goto out; |
644 | |
645 | /* |
646 | * return the timer_id now. The next step is hard to |
647 | * back out if there is an error. |
648 | */ |
649 | if (copy_to_user(created_timer_id, |
650 | &new_timer_id, sizeof (new_timer_id))) { |
651 | error = -EFAULT; |
652 | goto out; |
653 | } |
654 | if (timer_event_spec) { |
655 | if (copy_from_user(&event, timer_event_spec, sizeof (event))) { |
656 | error = -EFAULT; |
657 | goto out; |
658 | } |
659 | new_timer->it_sigev_notify = event.sigev_notify; |
660 | new_timer->it_sigev_signo = event.sigev_signo; |
661 | new_timer->it_sigev_value = event.sigev_value; |
662 | |
663 | read_lock(&tasklist_lock); |
664 | if ((process = good_sigevent(&event))) { |
665 | /* |
666 | * We may be setting up this process for another |
667 | * thread. It may be exiting. To catch this |
668 | * case the we check the PF_EXITING flag. If |
669 | * the flag is not set, the siglock will catch |
670 | * him before it is too late (in exit_itimers). |
671 | * |
672 | * The exec case is a bit more invloved but easy |
673 | * to code. If the process is in our thread |
674 | * group (and it must be or we would not allow |
675 | * it here) and is doing an exec, it will cause |
676 | * us to be killed. In this case it will wait |
677 | * for us to die which means we can finish this |
678 | * linkage with our last gasp. I.e. no code :) |
679 | */ |
680 | spin_lock_irqsave(&process->sighand->siglock, flags); |
681 | if (!(process->flags & PF_EXITING)) { |
682 | new_timer->it_process = process; |
683 | list_add(&new_timer->list, |
684 | &process->signal->posix_timers); |
685 | spin_unlock_irqrestore(&process->sighand->siglock, flags); |
686 | if (new_timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) |
687 | get_task_struct(process); |
688 | } else { |
689 | spin_unlock_irqrestore(&process->sighand->siglock, flags); |
690 | process = NULL; |
691 | } |
692 | } |
693 | read_unlock(&tasklist_lock); |
694 | if (!process) { |
695 | error = -EINVAL; |
696 | goto out; |
697 | } |
698 | } else { |
699 | new_timer->it_sigev_notify = SIGEV_SIGNAL; |
700 | new_timer->it_sigev_signo = SIGALRM; |
701 | new_timer->it_sigev_value.sival_int = new_timer->it_id; |
702 | process = current->group_leader; |
703 | spin_lock_irqsave(&process->sighand->siglock, flags); |
704 | new_timer->it_process = process; |
705 | list_add(&new_timer->list, &process->signal->posix_timers); |
706 | spin_unlock_irqrestore(&process->sighand->siglock, flags); |
707 | } |
708 | |
709 | /* |
710 | * In the case of the timer belonging to another task, after |
711 | * the task is unlocked, the timer is owned by the other task |
712 | * and may cease to exist at any time. Don't use or modify |
713 | * new_timer after the unlock call. |
714 | */ |
715 | |
716 | out: |
717 | if (error) |
718 | release_posix_timer(new_timer, it_id_set); |
719 | |
720 | return error; |
721 | } |
722 | |
723 | /* |
724 | * good_timespec |
725 | * |
726 | * This function checks the elements of a timespec structure. |
727 | * |
728 | * Arguments: |
729 | * ts : Pointer to the timespec structure to check |
730 | * |
731 | * Return value: |
732 | * If a NULL pointer was passed in, or the tv_nsec field was less than 0 |
733 | * or greater than NSEC_PER_SEC, or the tv_sec field was less than 0, |
734 | * this function returns 0. Otherwise it returns 1. |
735 | */ |
736 | static int good_timespec(const struct timespec *ts) |
737 | { |
738 | if ((!ts) || (ts->tv_sec < 0) || |
739 | ((unsigned) ts->tv_nsec >= NSEC_PER_SEC)) |
740 | return 0; |
741 | return 1; |
742 | } |
743 | |
744 | /* |
745 | * Locking issues: We need to protect the result of the id look up until |
746 | * we get the timer locked down so it is not deleted under us. The |
747 | * removal is done under the idr spinlock so we use that here to bridge |
748 | * the find to the timer lock. To avoid a dead lock, the timer id MUST |
749 | * be release with out holding the timer lock. |
750 | */ |
751 | static struct k_itimer * lock_timer(timer_t timer_id, unsigned long *flags) |
752 | { |
753 | struct k_itimer *timr; |
754 | /* |
755 | * Watch out here. We do a irqsave on the idr_lock and pass the |
756 | * flags part over to the timer lock. Must not let interrupts in |
757 | * while we are moving the lock. |
758 | */ |
759 | |
760 | spin_lock_irqsave(&idr_lock, *flags); |
761 | timr = (struct k_itimer *) idr_find(&posix_timers_id, (int) timer_id); |
762 | if (timr) { |
763 | spin_lock(&timr->it_lock); |
764 | spin_unlock(&idr_lock); |
765 | |
766 | if ((timr->it_id != timer_id) || !(timr->it_process) || |
767 | timr->it_process->tgid != current->tgid) { |
768 | unlock_timer(timr, *flags); |
769 | timr = NULL; |
770 | } |
771 | } else |
772 | spin_unlock_irqrestore(&idr_lock, *flags); |
773 | |
774 | return timr; |
775 | } |
776 | |
777 | /* |
778 | * Get the time remaining on a POSIX.1b interval timer. This function |
779 | * is ALWAYS called with spin_lock_irq on the timer, thus it must not |
780 | * mess with irq. |
781 | * |
782 | * We have a couple of messes to clean up here. First there is the case |
783 | * of a timer that has a requeue pending. These timers should appear to |
784 | * be in the timer list with an expiry as if we were to requeue them |
785 | * now. |
786 | * |
787 | * The second issue is the SIGEV_NONE timer which may be active but is |
788 | * not really ever put in the timer list (to save system resources). |
789 | * This timer may be expired, and if so, we will do it here. Otherwise |
790 | * it is the same as a requeue pending timer WRT to what we should |
791 | * report. |
792 | */ |
793 | static void |
794 | common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting) |
795 | { |
796 | unsigned long expires; |
797 | struct now_struct now; |
798 | |
799 | do |
800 | expires = timr->it.real.timer.expires; |
801 | while ((volatile long) (timr->it.real.timer.expires) != expires); |
802 | |
803 | posix_get_now(&now); |
804 | |
805 | if (expires && |
806 | ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) && |
807 | !timr->it.real.incr && |
808 | posix_time_before(&timr->it.real.timer, &now)) |
809 | timr->it.real.timer.expires = expires = 0; |
810 | if (expires) { |
811 | if (timr->it_requeue_pending & REQUEUE_PENDING || |
812 | (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
813 | posix_bump_timer(timr, now); |
814 | expires = timr->it.real.timer.expires; |
815 | } |
816 | else |
817 | if (!timer_pending(&timr->it.real.timer)) |
818 | expires = 0; |
819 | if (expires) |
820 | expires -= now.jiffies; |
821 | } |
822 | jiffies_to_timespec(expires, &cur_setting->it_value); |
823 | jiffies_to_timespec(timr->it.real.incr, &cur_setting->it_interval); |
824 | |
825 | if (cur_setting->it_value.tv_sec < 0) { |
826 | cur_setting->it_value.tv_nsec = 1; |
827 | cur_setting->it_value.tv_sec = 0; |
828 | } |
829 | } |
830 | |
831 | /* Get the time remaining on a POSIX.1b interval timer. */ |
832 | asmlinkage long |
833 | sys_timer_gettime(timer_t timer_id, struct itimerspec __user *setting) |
834 | { |
835 | struct k_itimer *timr; |
836 | struct itimerspec cur_setting; |
837 | unsigned long flags; |
838 | |
839 | timr = lock_timer(timer_id, &flags); |
840 | if (!timr) |
841 | return -EINVAL; |
842 | |
843 | CLOCK_DISPATCH(timr->it_clock, timer_get, (timr, &cur_setting)); |
844 | |
845 | unlock_timer(timr, flags); |
846 | |
847 | if (copy_to_user(setting, &cur_setting, sizeof (cur_setting))) |
848 | return -EFAULT; |
849 | |
850 | return 0; |
851 | } |
852 | /* |
853 | * Get the number of overruns of a POSIX.1b interval timer. This is to |
854 | * be the overrun of the timer last delivered. At the same time we are |
855 | * accumulating overruns on the next timer. The overrun is frozen when |
856 | * the signal is delivered, either at the notify time (if the info block |
857 | * is not queued) or at the actual delivery time (as we are informed by |
858 | * the call back to do_schedule_next_timer(). So all we need to do is |
859 | * to pick up the frozen overrun. |
860 | */ |
861 | |
862 | asmlinkage long |
863 | sys_timer_getoverrun(timer_t timer_id) |
864 | { |
865 | struct k_itimer *timr; |
866 | int overrun; |
867 | long flags; |
868 | |
869 | timr = lock_timer(timer_id, &flags); |
870 | if (!timr) |
871 | return -EINVAL; |
872 | |
873 | overrun = timr->it_overrun_last; |
874 | unlock_timer(timr, flags); |
875 | |
876 | return overrun; |
877 | } |
878 | /* |
879 | * Adjust for absolute time |
880 | * |
881 | * If absolute time is given and it is not CLOCK_MONOTONIC, we need to |
882 | * adjust for the offset between the timer clock (CLOCK_MONOTONIC) and |
883 | * what ever clock he is using. |
884 | * |
885 | * If it is relative time, we need to add the current (CLOCK_MONOTONIC) |
886 | * time to it to get the proper time for the timer. |
887 | */ |
888 | static int adjust_abs_time(struct k_clock *clock, struct timespec *tp, |
889 | int abs, u64 *exp, struct timespec *wall_to) |
890 | { |
891 | struct timespec now; |
892 | struct timespec oc = *tp; |
893 | u64 jiffies_64_f; |
894 | int rtn =0; |
895 | |
896 | if (abs) { |
897 | /* |
898 | * The mask pick up the 4 basic clocks |
899 | */ |
900 | if (!((clock - &posix_clocks[0]) & ~CLOCKS_MASK)) { |
901 | jiffies_64_f = do_posix_clock_monotonic_gettime_parts( |
902 | &now, wall_to); |
903 | /* |
904 | * If we are doing a MONOTONIC clock |
905 | */ |
906 | if((clock - &posix_clocks[0]) & CLOCKS_MONO){ |
907 | now.tv_sec += wall_to->tv_sec; |
908 | now.tv_nsec += wall_to->tv_nsec; |
909 | } |
910 | } else { |
911 | /* |
912 | * Not one of the basic clocks |
913 | */ |
914 | clock->clock_get(clock - posix_clocks, &now); |
915 | jiffies_64_f = get_jiffies_64(); |
916 | } |
917 | /* |
918 | * Take away now to get delta |
919 | */ |
920 | oc.tv_sec -= now.tv_sec; |
921 | oc.tv_nsec -= now.tv_nsec; |
922 | /* |
923 | * Normalize... |
924 | */ |
925 | while ((oc.tv_nsec - NSEC_PER_SEC) >= 0) { |
926 | oc.tv_nsec -= NSEC_PER_SEC; |
927 | oc.tv_sec++; |
928 | } |
929 | while ((oc.tv_nsec) < 0) { |
930 | oc.tv_nsec += NSEC_PER_SEC; |
931 | oc.tv_sec--; |
932 | } |
933 | }else{ |
934 | jiffies_64_f = get_jiffies_64(); |
935 | } |
936 | /* |
937 | * Check if the requested time is prior to now (if so set now) |
938 | */ |
939 | if (oc.tv_sec < 0) |
940 | oc.tv_sec = oc.tv_nsec = 0; |
941 | |
942 | if (oc.tv_sec | oc.tv_nsec) |
943 | set_normalized_timespec(&oc, oc.tv_sec, |
944 | oc.tv_nsec + clock->res); |
945 | tstojiffie(&oc, clock->res, exp); |
946 | |
947 | /* |
948 | * Check if the requested time is more than the timer code |
949 | * can handle (if so we error out but return the value too). |
950 | */ |
951 | if (*exp > ((u64)MAX_JIFFY_OFFSET)) |
952 | /* |
953 | * This is a considered response, not exactly in |
954 | * line with the standard (in fact it is silent on |
955 | * possible overflows). We assume such a large |
956 | * value is ALMOST always a programming error and |
957 | * try not to compound it by setting a really dumb |
958 | * value. |
959 | */ |
960 | rtn = -EINVAL; |
961 | /* |
962 | * return the actual jiffies expire time, full 64 bits |
963 | */ |
964 | *exp += jiffies_64_f; |
965 | return rtn; |
966 | } |
967 | |
968 | /* Set a POSIX.1b interval timer. */ |
969 | /* timr->it_lock is taken. */ |
970 | static inline int |
971 | common_timer_set(struct k_itimer *timr, int flags, |
972 | struct itimerspec *new_setting, struct itimerspec *old_setting) |
973 | { |
974 | struct k_clock *clock = &posix_clocks[timr->it_clock]; |
975 | u64 expire_64; |
976 | |
977 | if (old_setting) |
978 | common_timer_get(timr, old_setting); |
979 | |
980 | /* disable the timer */ |
981 | timr->it.real.incr = 0; |
982 | /* |
983 | * careful here. If smp we could be in the "fire" routine which will |
984 | * be spinning as we hold the lock. But this is ONLY an SMP issue. |
985 | */ |
986 | #ifdef CONFIG_SMP |
987 | if (timer_active(timr) && !del_timer(&timr->it.real.timer)) |
988 | /* |
989 | * It can only be active if on an other cpu. Since |
990 | * we have cleared the interval stuff above, it should |
991 | * clear once we release the spin lock. Of course once |
992 | * we do that anything could happen, including the |
993 | * complete melt down of the timer. So return with |
994 | * a "retry" exit status. |
995 | */ |
996 | return TIMER_RETRY; |
997 | |
998 | set_timer_inactive(timr); |
999 | #else |
1000 | del_timer(&timr->it.real.timer); |
1001 | #endif |
1002 | remove_from_abslist(timr); |
1003 | |
1004 | timr->it_requeue_pending = (timr->it_requeue_pending + 2) & |
1005 | ~REQUEUE_PENDING; |
1006 | timr->it_overrun_last = 0; |
1007 | timr->it_overrun = -1; |
1008 | /* |
1009 | *switch off the timer when it_value is zero |
1010 | */ |
1011 | if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) { |
1012 | timr->it.real.timer.expires = 0; |
1013 | return 0; |
1014 | } |
1015 | |
1016 | if (adjust_abs_time(clock, |
1017 | &new_setting->it_value, flags & TIMER_ABSTIME, |
1018 | &expire_64, &(timr->it.real.wall_to_prev))) { |
1019 | return -EINVAL; |
1020 | } |
1021 | timr->it.real.timer.expires = (unsigned long)expire_64; |
1022 | tstojiffie(&new_setting->it_interval, clock->res, &expire_64); |
1023 | timr->it.real.incr = (unsigned long)expire_64; |
1024 | |
1025 | /* |
1026 | * We do not even queue SIGEV_NONE timers! But we do put them |
1027 | * in the abs list so we can do that right. |
1028 | */ |
1029 | if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)) |
1030 | add_timer(&timr->it.real.timer); |
1031 | |
1032 | if (flags & TIMER_ABSTIME && clock->abs_struct) { |
1033 | spin_lock(&clock->abs_struct->lock); |
1034 | list_add_tail(&(timr->it.real.abs_timer_entry), |
1035 | &(clock->abs_struct->list)); |
1036 | spin_unlock(&clock->abs_struct->lock); |
1037 | } |
1038 | return 0; |
1039 | } |
1040 | |
1041 | /* Set a POSIX.1b interval timer */ |
1042 | asmlinkage long |
1043 | sys_timer_settime(timer_t timer_id, int flags, |
1044 | const struct itimerspec __user *new_setting, |
1045 | struct itimerspec __user *old_setting) |
1046 | { |
1047 | struct k_itimer *timr; |
1048 | struct itimerspec new_spec, old_spec; |
1049 | int error = 0; |
1050 | long flag; |
1051 | struct itimerspec *rtn = old_setting ? &old_spec : NULL; |
1052 | |
1053 | if (!new_setting) |
1054 | return -EINVAL; |
1055 | |
1056 | if (copy_from_user(&new_spec, new_setting, sizeof (new_spec))) |
1057 | return -EFAULT; |
1058 | |
1059 | if ((!good_timespec(&new_spec.it_interval)) || |
1060 | (!good_timespec(&new_spec.it_value))) |
1061 | return -EINVAL; |
1062 | retry: |
1063 | timr = lock_timer(timer_id, &flag); |
1064 | if (!timr) |
1065 | return -EINVAL; |
1066 | |
1067 | error = CLOCK_DISPATCH(timr->it_clock, timer_set, |
1068 | (timr, flags, &new_spec, rtn)); |
1069 | |
1070 | unlock_timer(timr, flag); |
1071 | if (error == TIMER_RETRY) { |
1072 | rtn = NULL; // We already got the old time... |
1073 | goto retry; |
1074 | } |
1075 | |
1076 | if (old_setting && !error && copy_to_user(old_setting, |
1077 | &old_spec, sizeof (old_spec))) |
1078 | error = -EFAULT; |
1079 | |
1080 | return error; |
1081 | } |
1082 | |
1083 | static inline int common_timer_del(struct k_itimer *timer) |
1084 | { |
1085 | timer->it.real.incr = 0; |
1086 | #ifdef CONFIG_SMP |
1087 | if (timer_active(timer) && !del_timer(&timer->it.real.timer)) |
1088 | /* |
1089 | * It can only be active if on an other cpu. Since |
1090 | * we have cleared the interval stuff above, it should |
1091 | * clear once we release the spin lock. Of course once |
1092 | * we do that anything could happen, including the |
1093 | * complete melt down of the timer. So return with |
1094 | * a "retry" exit status. |
1095 | */ |
1096 | return TIMER_RETRY; |
1097 | #else |
1098 | del_timer(&timer->it.real.timer); |
1099 | #endif |
1100 | remove_from_abslist(timer); |
1101 | |
1102 | return 0; |
1103 | } |
1104 | |
1105 | static inline int timer_delete_hook(struct k_itimer *timer) |
1106 | { |
1107 | return CLOCK_DISPATCH(timer->it_clock, timer_del, (timer)); |
1108 | } |
1109 | |
1110 | /* Delete a POSIX.1b interval timer. */ |
1111 | asmlinkage long |
1112 | sys_timer_delete(timer_t timer_id) |
1113 | { |
1114 | struct k_itimer *timer; |
1115 | long flags; |
1116 | |
1117 | #ifdef CONFIG_SMP |
1118 | int error; |
1119 | retry_delete: |
1120 | #endif |
1121 | timer = lock_timer(timer_id, &flags); |
1122 | if (!timer) |
1123 | return -EINVAL; |
1124 | |
1125 | #ifdef CONFIG_SMP |
1126 | error = timer_delete_hook(timer); |
1127 | |
1128 | if (error == TIMER_RETRY) { |
1129 | unlock_timer(timer, flags); |
1130 | goto retry_delete; |
1131 | } |
1132 | #else |
1133 | timer_delete_hook(timer); |
1134 | #endif |
1135 | spin_lock(¤t->sighand->siglock); |
1136 | list_del(&timer->list); |
1137 | spin_unlock(¤t->sighand->siglock); |
1138 | /* |
1139 | * This keeps any tasks waiting on the spin lock from thinking |
1140 | * they got something (see the lock code above). |
1141 | */ |
1142 | if (timer->it_process) { |
1143 | if (timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) |
1144 | put_task_struct(timer->it_process); |
1145 | timer->it_process = NULL; |
1146 | } |
1147 | unlock_timer(timer, flags); |
1148 | release_posix_timer(timer, IT_ID_SET); |
1149 | return 0; |
1150 | } |
1151 | /* |
1152 | * return timer owned by the process, used by exit_itimers |
1153 | */ |
1154 | static inline void itimer_delete(struct k_itimer *timer) |
1155 | { |
1156 | unsigned long flags; |
1157 | |
1158 | #ifdef CONFIG_SMP |
1159 | int error; |
1160 | retry_delete: |
1161 | #endif |
1162 | spin_lock_irqsave(&timer->it_lock, flags); |
1163 | |
1164 | #ifdef CONFIG_SMP |
1165 | error = timer_delete_hook(timer); |
1166 | |
1167 | if (error == TIMER_RETRY) { |
1168 | unlock_timer(timer, flags); |
1169 | goto retry_delete; |
1170 | } |
1171 | #else |
1172 | timer_delete_hook(timer); |
1173 | #endif |
1174 | list_del(&timer->list); |
1175 | /* |
1176 | * This keeps any tasks waiting on the spin lock from thinking |
1177 | * they got something (see the lock code above). |
1178 | */ |
1179 | if (timer->it_process) { |
1180 | if (timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID)) |
1181 | put_task_struct(timer->it_process); |
1182 | timer->it_process = NULL; |
1183 | } |
1184 | unlock_timer(timer, flags); |
1185 | release_posix_timer(timer, IT_ID_SET); |
1186 | } |
1187 | |
1188 | /* |
1189 | * This is called by __exit_signal, only when there are no more |
1190 | * references to the shared signal_struct. |
1191 | */ |
1192 | void exit_itimers(struct signal_struct *sig) |
1193 | { |
1194 | struct k_itimer *tmr; |
1195 | |
1196 | while (!list_empty(&sig->posix_timers)) { |
1197 | tmr = list_entry(sig->posix_timers.next, struct k_itimer, list); |
1198 | itimer_delete(tmr); |
1199 | } |
1200 | del_timer_sync(&sig->real_timer); |
1201 | } |
1202 | |
1203 | /* |
1204 | * And now for the "clock" calls |
1205 | * |
1206 | * These functions are called both from timer functions (with the timer |
1207 | * spin_lock_irq() held and from clock calls with no locking. They must |
1208 | * use the save flags versions of locks. |
1209 | */ |
1210 | |
1211 | /* |
1212 | * We do ticks here to avoid the irq lock ( they take sooo long). |
1213 | * The seqlock is great here. Since we a reader, we don't really care |
1214 | * if we are interrupted since we don't take lock that will stall us or |
1215 | * any other cpu. Voila, no irq lock is needed. |
1216 | * |
1217 | */ |
1218 | |
1219 | static u64 do_posix_clock_monotonic_gettime_parts( |
1220 | struct timespec *tp, struct timespec *mo) |
1221 | { |
1222 | u64 jiff; |
1223 | unsigned int seq; |
1224 | |
1225 | do { |
1226 | seq = read_seqbegin(&xtime_lock); |
1227 | getnstimeofday(tp); |
1228 | *mo = wall_to_monotonic; |
1229 | jiff = jiffies_64; |
1230 | |
1231 | } while(read_seqretry(&xtime_lock, seq)); |
1232 | |
1233 | return jiff; |
1234 | } |
1235 | |
1236 | static int do_posix_clock_monotonic_get(clockid_t clock, struct timespec *tp) |
1237 | { |
1238 | struct timespec wall_to_mono; |
1239 | |
1240 | do_posix_clock_monotonic_gettime_parts(tp, &wall_to_mono); |
1241 | |
1242 | tp->tv_sec += wall_to_mono.tv_sec; |
1243 | tp->tv_nsec += wall_to_mono.tv_nsec; |
1244 | |
1245 | if ((tp->tv_nsec - NSEC_PER_SEC) > 0) { |
1246 | tp->tv_nsec -= NSEC_PER_SEC; |
1247 | tp->tv_sec++; |
1248 | } |
1249 | return 0; |
1250 | } |
1251 | |
1252 | int do_posix_clock_monotonic_gettime(struct timespec *tp) |
1253 | { |
1254 | return do_posix_clock_monotonic_get(CLOCK_MONOTONIC, tp); |
1255 | } |
1256 | |
1257 | int do_posix_clock_nosettime(clockid_t clockid, struct timespec *tp) |
1258 | { |
1259 | return -EINVAL; |
1260 | } |
1261 | EXPORT_SYMBOL_GPL(do_posix_clock_nosettime); |
1262 | |
1263 | int do_posix_clock_notimer_create(struct k_itimer *timer) |
1264 | { |
1265 | return -EINVAL; |
1266 | } |
1267 | EXPORT_SYMBOL_GPL(do_posix_clock_notimer_create); |
1268 | |
1269 | int do_posix_clock_nonanosleep(clockid_t clock, int flags, struct timespec *t) |
1270 | { |
1271 | #ifndef ENOTSUP |
1272 | return -EOPNOTSUPP; /* aka ENOTSUP in userland for POSIX */ |
1273 | #else /* parisc does define it separately. */ |
1274 | return -ENOTSUP; |
1275 | #endif |
1276 | } |
1277 | EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep); |
1278 | |
1279 | asmlinkage long |
1280 | sys_clock_settime(clockid_t which_clock, const struct timespec __user *tp) |
1281 | { |
1282 | struct timespec new_tp; |
1283 | |
1284 | if (invalid_clockid(which_clock)) |
1285 | return -EINVAL; |
1286 | if (copy_from_user(&new_tp, tp, sizeof (*tp))) |
1287 | return -EFAULT; |
1288 | |
1289 | return CLOCK_DISPATCH(which_clock, clock_set, (which_clock, &new_tp)); |
1290 | } |
1291 | |
1292 | asmlinkage long |
1293 | sys_clock_gettime(clockid_t which_clock, struct timespec __user *tp) |
1294 | { |
1295 | struct timespec kernel_tp; |
1296 | int error; |
1297 | |
1298 | if (invalid_clockid(which_clock)) |
1299 | return -EINVAL; |
1300 | error = CLOCK_DISPATCH(which_clock, clock_get, |
1301 | (which_clock, &kernel_tp)); |
1302 | if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp))) |
1303 | error = -EFAULT; |
1304 | |
1305 | return error; |
1306 | |
1307 | } |
1308 | |
1309 | asmlinkage long |
1310 | sys_clock_getres(clockid_t which_clock, struct timespec __user *tp) |
1311 | { |
1312 | struct timespec rtn_tp; |
1313 | int error; |
1314 | |
1315 | if (invalid_clockid(which_clock)) |
1316 | return -EINVAL; |
1317 | |
1318 | error = CLOCK_DISPATCH(which_clock, clock_getres, |
1319 | (which_clock, &rtn_tp)); |
1320 | |
1321 | if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) { |
1322 | error = -EFAULT; |
1323 | } |
1324 | |
1325 | return error; |
1326 | } |
1327 | |
1328 | static void nanosleep_wake_up(unsigned long __data) |
1329 | { |
1330 | struct task_struct *p = (struct task_struct *) __data; |
1331 | |
1332 | wake_up_process(p); |
1333 | } |
1334 | |
1335 | /* |
1336 | * The standard says that an absolute nanosleep call MUST wake up at |
1337 | * the requested time in spite of clock settings. Here is what we do: |
1338 | * For each nanosleep call that needs it (only absolute and not on |
1339 | * CLOCK_MONOTONIC* (as it can not be set)) we thread a little structure |
1340 | * into the "nanosleep_abs_list". All we need is the task_struct pointer. |
1341 | * When ever the clock is set we just wake up all those tasks. The rest |
1342 | * is done by the while loop in clock_nanosleep(). |
1343 | * |
1344 | * On locking, clock_was_set() is called from update_wall_clock which |
1345 | * holds (or has held for it) a write_lock_irq( xtime_lock) and is |
1346 | * called from the timer bh code. Thus we need the irq save locks. |
1347 | * |
1348 | * Also, on the call from update_wall_clock, that is done as part of a |
1349 | * softirq thing. We don't want to delay the system that much (possibly |
1350 | * long list of timers to fix), so we defer that work to keventd. |
1351 | */ |
1352 | |
1353 | static DECLARE_WAIT_QUEUE_HEAD(nanosleep_abs_wqueue); |
1354 | static DECLARE_WORK(clock_was_set_work, (void(*)(void*))clock_was_set, NULL); |
1355 | |
1356 | static DECLARE_MUTEX(clock_was_set_lock); |
1357 | |
1358 | void clock_was_set(void) |
1359 | { |
1360 | struct k_itimer *timr; |
1361 | struct timespec new_wall_to; |
1362 | LIST_HEAD(cws_list); |
1363 | unsigned long seq; |
1364 | |
1365 | |
1366 | if (unlikely(in_interrupt())) { |
1367 | schedule_work(&clock_was_set_work); |
1368 | return; |
1369 | } |
1370 | wake_up_all(&nanosleep_abs_wqueue); |
1371 | |
1372 | /* |
1373 | * Check if there exist TIMER_ABSTIME timers to correct. |
1374 | * |
1375 | * Notes on locking: This code is run in task context with irq |
1376 | * on. We CAN be interrupted! All other usage of the abs list |
1377 | * lock is under the timer lock which holds the irq lock as |
1378 | * well. We REALLY don't want to scan the whole list with the |
1379 | * interrupt system off, AND we would like a sequence lock on |
1380 | * this code as well. Since we assume that the clock will not |
1381 | * be set often, it seems ok to take and release the irq lock |
1382 | * for each timer. In fact add_timer will do this, so this is |
1383 | * not an issue. So we know when we are done, we will move the |
1384 | * whole list to a new location. Then as we process each entry, |
1385 | * we will move it to the actual list again. This way, when our |
1386 | * copy is empty, we are done. We are not all that concerned |
1387 | * about preemption so we will use a semaphore lock to protect |
1388 | * aginst reentry. This way we will not stall another |
1389 | * processor. It is possible that this may delay some timers |
1390 | * that should have expired, given the new clock, but even this |
1391 | * will be minimal as we will always update to the current time, |
1392 | * even if it was set by a task that is waiting for entry to |
1393 | * this code. Timers that expire too early will be caught by |
1394 | * the expire code and restarted. |
1395 | |
1396 | * Absolute timers that repeat are left in the abs list while |
1397 | * waiting for the task to pick up the signal. This means we |
1398 | * may find timers that are not in the "add_timer" list, but are |
1399 | * in the abs list. We do the same thing for these, save |
1400 | * putting them back in the "add_timer" list. (Note, these are |
1401 | * left in the abs list mainly to indicate that they are |
1402 | * ABSOLUTE timers, a fact that is used by the re-arm code, and |
1403 | * for which we have no other flag.) |
1404 | |
1405 | */ |
1406 | |
1407 | down(&clock_was_set_lock); |
1408 | spin_lock_irq(&abs_list.lock); |
1409 | list_splice_init(&abs_list.list, &cws_list); |
1410 | spin_unlock_irq(&abs_list.lock); |
1411 | do { |
1412 | do { |
1413 | seq = read_seqbegin(&xtime_lock); |
1414 | new_wall_to = wall_to_monotonic; |
1415 | } while (read_seqretry(&xtime_lock, seq)); |
1416 | |
1417 | spin_lock_irq(&abs_list.lock); |
1418 | if (list_empty(&cws_list)) { |
1419 | spin_unlock_irq(&abs_list.lock); |
1420 | break; |
1421 | } |
1422 | timr = list_entry(cws_list.next, struct k_itimer, |
1423 | it.real.abs_timer_entry); |
1424 | |
1425 | list_del_init(&timr->it.real.abs_timer_entry); |
1426 | if (add_clockset_delta(timr, &new_wall_to) && |
1427 | del_timer(&timr->it.real.timer)) /* timer run yet? */ |
1428 | add_timer(&timr->it.real.timer); |
1429 | list_add(&timr->it.real.abs_timer_entry, &abs_list.list); |
1430 | spin_unlock_irq(&abs_list.lock); |
1431 | } while (1); |
1432 | |
1433 | up(&clock_was_set_lock); |
1434 | } |
1435 | |
1436 | long clock_nanosleep_restart(struct restart_block *restart_block); |
1437 | |
1438 | asmlinkage long |
1439 | sys_clock_nanosleep(clockid_t which_clock, int flags, |
1440 | const struct timespec __user *rqtp, |
1441 | struct timespec __user *rmtp) |
1442 | { |
1443 | struct timespec t; |
1444 | struct restart_block *restart_block = |
1445 | &(current_thread_info()->restart_block); |
1446 | int ret; |
1447 | |
1448 | if (invalid_clockid(which_clock)) |
1449 | return -EINVAL; |
1450 | |
1451 | if (copy_from_user(&t, rqtp, sizeof (struct timespec))) |
1452 | return -EFAULT; |
1453 | |
1454 | if ((unsigned) t.tv_nsec >= NSEC_PER_SEC || t.tv_sec < 0) |
1455 | return -EINVAL; |
1456 | |
1457 | /* |
1458 | * Do this here as nsleep function does not have the real address. |
1459 | */ |
1460 | restart_block->arg1 = (unsigned long)rmtp; |
1461 | |
1462 | ret = CLOCK_DISPATCH(which_clock, nsleep, (which_clock, flags, &t)); |
1463 | |
1464 | if ((ret == -ERESTART_RESTARTBLOCK) && rmtp && |
1465 | copy_to_user(rmtp, &t, sizeof (t))) |
1466 | return -EFAULT; |
1467 | return ret; |
1468 | } |
1469 | |
1470 | |
1471 | static int common_nsleep(clockid_t which_clock, |
1472 | int flags, struct timespec *tsave) |
1473 | { |
1474 | struct timespec t, dum; |
1475 | struct timer_list new_timer; |
1476 | DECLARE_WAITQUEUE(abs_wqueue, current); |
1477 | u64 rq_time = (u64)0; |
1478 | s64 left; |
1479 | int abs; |
1480 | struct restart_block *restart_block = |
1481 | ¤t_thread_info()->restart_block; |
1482 | |
1483 | abs_wqueue.flags = 0; |
1484 | init_timer(&new_timer); |
1485 | new_timer.expires = 0; |
1486 | new_timer.data = (unsigned long) current; |
1487 | new_timer.function = nanosleep_wake_up; |
1488 | abs = flags & TIMER_ABSTIME; |
1489 | |
1490 | if (restart_block->fn == clock_nanosleep_restart) { |
1491 | /* |
1492 | * Interrupted by a non-delivered signal, pick up remaining |
1493 | * time and continue. Remaining time is in arg2 & 3. |
1494 | */ |
1495 | restart_block->fn = do_no_restart_syscall; |
1496 | |
1497 | rq_time = restart_block->arg3; |
1498 | rq_time = (rq_time << 32) + restart_block->arg2; |
1499 | if (!rq_time) |
1500 | return -EINTR; |
1501 | left = rq_time - get_jiffies_64(); |
1502 | if (left <= (s64)0) |
1503 | return 0; /* Already passed */ |
1504 | } |
1505 | |
1506 | if (abs && (posix_clocks[which_clock].clock_get != |
1507 | posix_clocks[CLOCK_MONOTONIC].clock_get)) |
1508 | add_wait_queue(&nanosleep_abs_wqueue, &abs_wqueue); |
1509 | |
1510 | do { |
1511 | t = *tsave; |
1512 | if (abs || !rq_time) { |
1513 | adjust_abs_time(&posix_clocks[which_clock], &t, abs, |
1514 | &rq_time, &dum); |
1515 | } |
1516 | |
1517 | left = rq_time - get_jiffies_64(); |
1518 | if (left >= (s64)MAX_JIFFY_OFFSET) |
1519 | left = (s64)MAX_JIFFY_OFFSET; |
1520 | if (left < (s64)0) |
1521 | break; |
1522 | |
1523 | new_timer.expires = jiffies + left; |
1524 | __set_current_state(TASK_INTERRUPTIBLE); |
1525 | add_timer(&new_timer); |
1526 | |
1527 | schedule(); |
1528 | |
1529 | del_timer_sync(&new_timer); |
1530 | left = rq_time - get_jiffies_64(); |
1531 | } while (left > (s64)0 && !test_thread_flag(TIF_SIGPENDING)); |
1532 | |
1533 | if (abs_wqueue.task_list.next) |
1534 | finish_wait(&nanosleep_abs_wqueue, &abs_wqueue); |
1535 | |
1536 | if (left > (s64)0) { |
1537 | |
1538 | /* |
1539 | * Always restart abs calls from scratch to pick up any |
1540 | * clock shifting that happened while we are away. |
1541 | */ |
1542 | if (abs) |
1543 | return -ERESTARTNOHAND; |
1544 | |
1545 | left *= TICK_NSEC; |
1546 | tsave->tv_sec = div_long_long_rem(left, |
1547 | NSEC_PER_SEC, |
1548 | &tsave->tv_nsec); |
1549 | /* |
1550 | * Restart works by saving the time remaing in |
1551 | * arg2 & 3 (it is 64-bits of jiffies). The other |
1552 | * info we need is the clock_id (saved in arg0). |
1553 | * The sys_call interface needs the users |
1554 | * timespec return address which _it_ saves in arg1. |
1555 | * Since we have cast the nanosleep call to a clock_nanosleep |
1556 | * both can be restarted with the same code. |
1557 | */ |
1558 | restart_block->fn = clock_nanosleep_restart; |
1559 | restart_block->arg0 = which_clock; |
1560 | /* |
1561 | * Caller sets arg1 |
1562 | */ |
1563 | restart_block->arg2 = rq_time & 0xffffffffLL; |
1564 | restart_block->arg3 = rq_time >> 32; |
1565 | |
1566 | return -ERESTART_RESTARTBLOCK; |
1567 | } |
1568 | |
1569 | return 0; |
1570 | } |
1571 | /* |
1572 | * This will restart clock_nanosleep. |
1573 | */ |
1574 | long |
1575 | clock_nanosleep_restart(struct restart_block *restart_block) |
1576 | { |
1577 | struct timespec t; |
1578 | int ret = common_nsleep(restart_block->arg0, 0, &t); |
1579 | |
1580 | if ((ret == -ERESTART_RESTARTBLOCK) && restart_block->arg1 && |
1581 | copy_to_user((struct timespec __user *)(restart_block->arg1), &t, |
1582 | sizeof (t))) |
1583 | return -EFAULT; |
1584 | return ret; |
1585 | } |