kern_lwp.c revision 1.34 1 /* $NetBSD: kern_lwp.c,v 1.34 2006/03/29 23:02:31 cube Exp $ */
2
3 /*-
4 * Copyright (c) 2001 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Nathan J. Williams.
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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: kern_lwp.c,v 1.34 2006/03/29 23:02:31 cube Exp $");
41
42 #include "opt_multiprocessor.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/pool.h>
47 #include <sys/lock.h>
48 #include <sys/proc.h>
49 #include <sys/sa.h>
50 #include <sys/savar.h>
51 #include <sys/types.h>
52 #include <sys/ucontext.h>
53 #include <sys/resourcevar.h>
54 #include <sys/mount.h>
55 #include <sys/syscallargs.h>
56
57 #include <uvm/uvm_extern.h>
58
59 struct lwplist alllwp;
60
61 #define LWP_DEBUG
62
63 #ifdef LWP_DEBUG
64 int lwp_debug = 0;
65 #define DPRINTF(x) if (lwp_debug) printf x
66 #else
67 #define DPRINTF(x)
68 #endif
69 /* ARGSUSED */
70 int
71 sys__lwp_create(struct lwp *l, void *v, register_t *retval)
72 {
73 struct sys__lwp_create_args /* {
74 syscallarg(const ucontext_t *) ucp;
75 syscallarg(u_long) flags;
76 syscallarg(lwpid_t *) new_lwp;
77 } */ *uap = v;
78 struct proc *p = l->l_proc;
79 struct lwp *l2;
80 vaddr_t uaddr;
81 boolean_t inmem;
82 ucontext_t *newuc;
83 int s, error;
84
85 if (p->p_flag & P_SA)
86 return EINVAL;
87
88 newuc = pool_get(&lwp_uc_pool, PR_WAITOK);
89
90 error = copyin(SCARG(uap, ucp), newuc,
91 l->l_proc->p_emul->e_sa->sae_ucsize);
92 if (error)
93 return (error);
94
95 /* XXX check against resource limits */
96
97 inmem = uvm_uarea_alloc(&uaddr);
98 if (__predict_false(uaddr == 0)) {
99 return (ENOMEM);
100 }
101
102 /* XXX flags:
103 * __LWP_ASLWP is probably needed for Solaris compat.
104 */
105
106 newlwp(l, p, uaddr, inmem,
107 SCARG(uap, flags) & LWP_DETACHED,
108 NULL, 0, startlwp, newuc, &l2);
109
110 if ((SCARG(uap, flags) & LWP_SUSPENDED) == 0) {
111 SCHED_LOCK(s);
112 l2->l_stat = LSRUN;
113 setrunqueue(l2);
114 p->p_nrlwps++;
115 SCHED_UNLOCK(s);
116 } else {
117 l2->l_stat = LSSUSPENDED;
118 }
119
120 error = copyout(&l2->l_lid, SCARG(uap, new_lwp),
121 sizeof(l2->l_lid));
122 if (error)
123 return (error);
124
125 return (0);
126 }
127
128
129 int
130 sys__lwp_exit(struct lwp *l, void *v, register_t *retval)
131 {
132
133 lwp_exit(l);
134 /* NOTREACHED */
135 return (0);
136 }
137
138
139 int
140 sys__lwp_self(struct lwp *l, void *v, register_t *retval)
141 {
142
143 *retval = l->l_lid;
144
145 return (0);
146 }
147
148
149 int
150 sys__lwp_getprivate(struct lwp *l, void *v, register_t *retval)
151 {
152
153 *retval = (uintptr_t) l->l_private;
154
155 return (0);
156 }
157
158
159 int
160 sys__lwp_setprivate(struct lwp *l, void *v, register_t *retval)
161 {
162 struct sys__lwp_setprivate_args /* {
163 syscallarg(void *) ptr;
164 } */ *uap = v;
165
166 l->l_private = SCARG(uap, ptr);
167
168 return (0);
169 }
170
171
172 int
173 sys__lwp_suspend(struct lwp *l, void *v, register_t *retval)
174 {
175 struct sys__lwp_suspend_args /* {
176 syscallarg(lwpid_t) target;
177 } */ *uap = v;
178 int target_lid;
179 struct proc *p = l->l_proc;
180 struct lwp *t;
181 struct lwp *t2;
182
183 if (p->p_flag & P_SA)
184 return EINVAL;
185
186 target_lid = SCARG(uap, target);
187
188 LIST_FOREACH(t, &p->p_lwps, l_sibling)
189 if (t->l_lid == target_lid)
190 break;
191
192 if (t == NULL)
193 return (ESRCH);
194
195 if (t == l) {
196 /*
197 * Check for deadlock, which is only possible
198 * when we're suspending ourself.
199 */
200 LIST_FOREACH(t2, &p->p_lwps, l_sibling) {
201 if ((t2 != l) && (t2->l_stat != LSSUSPENDED))
202 break;
203 }
204
205 if (t2 == NULL) /* All other LWPs are suspended */
206 return (EDEADLK);
207 }
208
209 return lwp_suspend(l, t);
210 }
211
212 inline int
213 lwp_suspend(struct lwp *l, struct lwp *t)
214 {
215 struct proc *p = t->l_proc;
216 int s;
217
218 if (t == l) {
219 SCHED_LOCK(s);
220 l->l_stat = LSSUSPENDED;
221 /* XXX NJWLWP check if this makes sense here: */
222 p->p_stats->p_ru.ru_nvcsw++;
223 mi_switch(l, NULL);
224 SCHED_ASSERT_UNLOCKED();
225 splx(s);
226 } else {
227 switch (t->l_stat) {
228 case LSSUSPENDED:
229 return (0); /* _lwp_suspend() is idempotent */
230 case LSRUN:
231 SCHED_LOCK(s);
232 remrunqueue(t);
233 t->l_stat = LSSUSPENDED;
234 p->p_nrlwps--;
235 SCHED_UNLOCK(s);
236 break;
237 case LSSLEEP:
238 t->l_stat = LSSUSPENDED;
239 break;
240 case LSIDL:
241 case LSZOMB:
242 return (EINTR); /* It's what Solaris does..... */
243 case LSSTOP:
244 panic("_lwp_suspend: Stopped LWP in running process!");
245 break;
246 case LSONPROC:
247 /* XXX multiprocessor LWPs? Implement me! */
248 return (EINVAL);
249 }
250 }
251
252 return (0);
253 }
254
255
256 int
257 sys__lwp_continue(struct lwp *l, void *v, register_t *retval)
258 {
259 struct sys__lwp_continue_args /* {
260 syscallarg(lwpid_t) target;
261 } */ *uap = v;
262 int s, target_lid;
263 struct proc *p = l->l_proc;
264 struct lwp *t;
265
266 if (p->p_flag & P_SA)
267 return EINVAL;
268
269 target_lid = SCARG(uap, target);
270
271 LIST_FOREACH(t, &p->p_lwps, l_sibling)
272 if (t->l_lid == target_lid)
273 break;
274
275 if (t == NULL)
276 return (ESRCH);
277
278 SCHED_LOCK(s);
279 lwp_continue(t);
280 SCHED_UNLOCK(s);
281
282 return (0);
283 }
284
285 void
286 lwp_continue(struct lwp *l)
287 {
288
289 DPRINTF(("lwp_continue of %d.%d (%s), state %d, wchan %p\n",
290 l->l_proc->p_pid, l->l_lid, l->l_proc->p_comm, l->l_stat,
291 l->l_wchan));
292
293 if (l->l_stat != LSSUSPENDED)
294 return;
295
296 if (l->l_wchan == 0) {
297 /* LWP was runnable before being suspended. */
298 setrunnable(l);
299 } else {
300 /* LWP was sleeping before being suspended. */
301 l->l_stat = LSSLEEP;
302 }
303 }
304
305 int
306 sys__lwp_wakeup(struct lwp *l, void *v, register_t *retval)
307 {
308 struct sys__lwp_wakeup_args /* {
309 syscallarg(lwpid_t) target;
310 } */ *uap = v;
311 lwpid_t target_lid;
312 struct lwp *t;
313 struct proc *p;
314 int error;
315 int s;
316
317 p = l->l_proc;
318 target_lid = SCARG(uap, target);
319
320 SCHED_LOCK(s);
321
322 LIST_FOREACH(t, &p->p_lwps, l_sibling)
323 if (t->l_lid == target_lid)
324 break;
325
326 if (t == NULL) {
327 error = ESRCH;
328 goto exit;
329 }
330
331 if (t->l_stat != LSSLEEP) {
332 error = ENODEV;
333 goto exit;
334 }
335
336 if ((t->l_flag & L_SINTR) == 0) {
337 error = EBUSY;
338 goto exit;
339 }
340 /*
341 * Tell ltsleep to wakeup.
342 */
343 t->l_flag |= L_CANCELLED;
344
345 setrunnable(t);
346 error = 0;
347 exit:
348 SCHED_UNLOCK(s);
349
350 return error;
351 }
352
353 int
354 sys__lwp_wait(struct lwp *l, void *v, register_t *retval)
355 {
356 struct sys__lwp_wait_args /* {
357 syscallarg(lwpid_t) wait_for;
358 syscallarg(lwpid_t *) departed;
359 } */ *uap = v;
360 int error;
361 lwpid_t dep;
362
363 error = lwp_wait1(l, SCARG(uap, wait_for), &dep, 0);
364 if (error)
365 return (error);
366
367 if (SCARG(uap, departed)) {
368 error = copyout(&dep, SCARG(uap, departed),
369 sizeof(dep));
370 if (error)
371 return (error);
372 }
373
374 return (0);
375 }
376
377
378 int
379 lwp_wait1(struct lwp *l, lwpid_t lid, lwpid_t *departed, int flags)
380 {
381 struct proc *p = l->l_proc;
382 struct lwp *l2, *l3;
383 int nfound, error, wpri;
384 static const char waitstr1[] = "lwpwait";
385 static const char waitstr2[] = "lwpwait2";
386
387 DPRINTF(("lwp_wait1: %d.%d waiting for %d.\n",
388 p->p_pid, l->l_lid, lid));
389
390 if (lid == l->l_lid)
391 return (EDEADLK); /* Waiting for ourselves makes no sense. */
392
393 wpri = PWAIT |
394 ((flags & LWPWAIT_EXITCONTROL) ? PNOEXITERR : PCATCH);
395 loop:
396 nfound = 0;
397 LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
398 if ((l2 == l) || (l2->l_flag & L_DETACHED) ||
399 ((lid != 0) && (lid != l2->l_lid)))
400 continue;
401
402 nfound++;
403 if (l2->l_stat == LSZOMB) {
404 if (departed)
405 *departed = l2->l_lid;
406
407 simple_lock(&p->p_lock);
408 LIST_REMOVE(l2, l_sibling);
409 p->p_nlwps--;
410 p->p_nzlwps--;
411 simple_unlock(&p->p_lock);
412 /* XXX decrement limits */
413
414 pool_put(&lwp_pool, l2);
415
416 return (0);
417 } else if (l2->l_stat == LSSLEEP ||
418 l2->l_stat == LSSUSPENDED) {
419 /* Deadlock checks.
420 * 1. If all other LWPs are waiting for exits
421 * or suspended, we would deadlock.
422 */
423
424 LIST_FOREACH(l3, &p->p_lwps, l_sibling) {
425 if (l3 != l && (l3->l_stat != LSSUSPENDED) &&
426 !(l3->l_stat == LSSLEEP &&
427 l3->l_wchan == (caddr_t) &p->p_nlwps))
428 break;
429 }
430 if (l3 == NULL) /* Everyone else is waiting. */
431 return (EDEADLK);
432
433 /* XXX we'd like to check for a cycle of waiting
434 * LWPs (specific LID waits, not any-LWP waits)
435 * and detect that sort of deadlock, but we don't
436 * have a good place to store the lwp that is
437 * being waited for. wchan is already filled with
438 * &p->p_nlwps, and putting the lwp address in
439 * there for deadlock tracing would require
440 * exiting LWPs to call wakeup on both their
441 * own address and &p->p_nlwps, to get threads
442 * sleeping on any LWP exiting.
443 *
444 * Revisit later. Maybe another auxillary
445 * storage location associated with sleeping
446 * is in order.
447 */
448 }
449 }
450
451 if (nfound == 0)
452 return (ESRCH);
453
454 if ((error = tsleep((caddr_t) &p->p_nlwps, wpri,
455 (lid != 0) ? waitstr1 : waitstr2, 0)) != 0)
456 return (error);
457
458 goto loop;
459 }
460
461
462 int
463 newlwp(struct lwp *l1, struct proc *p2, vaddr_t uaddr, boolean_t inmem,
464 int flags, void *stack, size_t stacksize,
465 void (*func)(void *), void *arg, struct lwp **rnewlwpp)
466 {
467 struct lwp *l2;
468 int s;
469
470 l2 = pool_get(&lwp_pool, PR_WAITOK);
471
472 l2->l_stat = LSIDL;
473 l2->l_forw = l2->l_back = NULL;
474 l2->l_proc = p2;
475
476 memset(&l2->l_startzero, 0,
477 (unsigned) ((caddr_t)&l2->l_endzero -
478 (caddr_t)&l2->l_startzero));
479 memcpy(&l2->l_startcopy, &l1->l_startcopy,
480 (unsigned) ((caddr_t)&l2->l_endcopy -
481 (caddr_t)&l2->l_startcopy));
482
483 #if !defined(MULTIPROCESSOR)
484 /*
485 * In the single-processor case, all processes will always run
486 * on the same CPU. So, initialize the child's CPU to the parent's
487 * now. In the multiprocessor case, the child's CPU will be
488 * initialized in the low-level context switch code when the
489 * process runs.
490 */
491 KASSERT(l1->l_cpu != NULL);
492 l2->l_cpu = l1->l_cpu;
493 #else
494 /*
495 * zero child's CPU pointer so we don't get trash.
496 */
497 l2->l_cpu = NULL;
498 #endif /* ! MULTIPROCESSOR */
499
500 l2->l_flag = inmem ? L_INMEM : 0;
501 l2->l_flag |= (flags & LWP_DETACHED) ? L_DETACHED : 0;
502
503 callout_init(&l2->l_tsleep_ch);
504
505 if (rnewlwpp != NULL)
506 *rnewlwpp = l2;
507
508 l2->l_addr = (struct user *)uaddr;
509 uvm_lwp_fork(l1, l2, stack, stacksize, func,
510 (arg != NULL) ? arg : l2);
511
512 simple_lock(&p2->p_lock);
513 l2->l_lid = ++p2->p_nlwpid;
514 LIST_INSERT_HEAD(&p2->p_lwps, l2, l_sibling);
515 p2->p_nlwps++;
516 simple_unlock(&p2->p_lock);
517
518 /* XXX should be locked differently... */
519 s = proclist_lock_write();
520 LIST_INSERT_HEAD(&alllwp, l2, l_list);
521 proclist_unlock_write(s);
522
523 if (p2->p_emul->e_lwp_fork)
524 (*p2->p_emul->e_lwp_fork)(l1, l2);
525
526 return (0);
527 }
528
529
530 /*
531 * Quit the process. This will call cpu_exit, which will call cpu_switch,
532 * so this can only be used meaningfully if you're willing to switch away.
533 * Calling with l!=curlwp would be weird.
534 */
535 void
536 lwp_exit(struct lwp *l)
537 {
538 struct proc *p = l->l_proc;
539 int s;
540
541 DPRINTF(("lwp_exit: %d.%d exiting.\n", p->p_pid, l->l_lid));
542 DPRINTF((" nlwps: %d nrlwps %d nzlwps: %d\n",
543 p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
544
545 if (p->p_emul->e_lwp_exit)
546 (*p->p_emul->e_lwp_exit)(l);
547
548 /*
549 * If we are the last live LWP in a process, we need to exit
550 * the entire process (if that's not already going on). We do
551 * so with an exit status of zero, because it's a "controlled"
552 * exit, and because that's what Solaris does.
553 */
554 if (((p->p_nlwps - p->p_nzlwps) == 1) && ((p->p_flag & P_WEXIT) == 0)) {
555 DPRINTF(("lwp_exit: %d.%d calling exit1()\n",
556 p->p_pid, l->l_lid));
557 exit1(l, 0);
558 /* NOTREACHED */
559 }
560
561 s = proclist_lock_write();
562 LIST_REMOVE(l, l_list);
563 proclist_unlock_write(s);
564
565 /* Free MD LWP resources */
566 #ifndef __NO_CPU_LWP_FREE
567 cpu_lwp_free(l, 0);
568 #endif
569
570 pmap_deactivate(l);
571
572 if (l->l_flag & L_DETACHED) {
573 simple_lock(&p->p_lock);
574 LIST_REMOVE(l, l_sibling);
575 p->p_nlwps--;
576 simple_unlock(&p->p_lock);
577
578 curlwp = NULL;
579 l->l_proc = NULL;
580 }
581
582 SCHED_LOCK(s);
583 p->p_nrlwps--;
584 l->l_stat = LSDEAD;
585 SCHED_UNLOCK(s);
586
587 /* This LWP no longer needs to hold the kernel lock. */
588 KERNEL_PROC_UNLOCK(l);
589
590 /* cpu_exit() will not return */
591 cpu_exit(l);
592 }
593
594 /*
595 * We are called from cpu_exit() once it is safe to schedule the
596 * dead process's resources to be freed (i.e., once we've switched to
597 * the idle PCB for the current CPU).
598 *
599 * NOTE: One must be careful with locking in this routine. It's
600 * called from a critical section in machine-dependent code, so
601 * we should refrain from changing any interrupt state.
602 */
603 void
604 lwp_exit2(struct lwp *l)
605 {
606 struct proc *p;
607
608 KERNEL_LOCK(LK_EXCLUSIVE);
609 /*
610 * Free the VM resources we're still holding on to.
611 */
612 uvm_lwp_exit(l);
613
614 if (l->l_flag & L_DETACHED) {
615 /* Nobody waits for detached LWPs. */
616 pool_put(&lwp_pool, l);
617 KERNEL_UNLOCK();
618 } else {
619 l->l_stat = LSZOMB;
620 p = l->l_proc;
621 p->p_nzlwps++;
622 KERNEL_UNLOCK();
623 wakeup(&p->p_nlwps);
624 }
625 }
626
627 /*
628 * Pick a LWP to represent the process for those operations which
629 * want information about a "process" that is actually associated
630 * with a LWP.
631 */
632 struct lwp *
633 proc_representative_lwp(struct proc *p)
634 {
635 struct lwp *l, *onproc, *running, *sleeping, *stopped, *suspended;
636 struct lwp *signalled;
637
638 /* Trivial case: only one LWP */
639 if (p->p_nlwps == 1)
640 return (LIST_FIRST(&p->p_lwps));
641
642 switch (p->p_stat) {
643 case SSTOP:
644 case SACTIVE:
645 /* Pick the most live LWP */
646 onproc = running = sleeping = stopped = suspended = NULL;
647 signalled = NULL;
648 LIST_FOREACH(l, &p->p_lwps, l_sibling) {
649 if (l->l_lid == p->p_sigctx.ps_lwp)
650 signalled = l;
651 switch (l->l_stat) {
652 case LSONPROC:
653 onproc = l;
654 break;
655 case LSRUN:
656 running = l;
657 break;
658 case LSSLEEP:
659 sleeping = l;
660 break;
661 case LSSTOP:
662 stopped = l;
663 break;
664 case LSSUSPENDED:
665 suspended = l;
666 break;
667 }
668 }
669 if (signalled)
670 return signalled;
671 if (onproc)
672 return onproc;
673 if (running)
674 return running;
675 if (sleeping)
676 return sleeping;
677 if (stopped)
678 return stopped;
679 if (suspended)
680 return suspended;
681 break;
682 case SZOMB:
683 /* Doesn't really matter... */
684 return (LIST_FIRST(&p->p_lwps));
685 #ifdef DIAGNOSTIC
686 case SIDL:
687 /* We have more than one LWP and we're in SIDL?
688 * How'd that happen?
689 */
690 panic("Too many LWPs (%d) in SIDL process %d (%s)",
691 p->p_nrlwps, p->p_pid, p->p_comm);
692 default:
693 panic("Process %d (%s) in unknown state %d",
694 p->p_pid, p->p_comm, p->p_stat);
695 #endif
696 }
697
698 panic("proc_representative_lwp: couldn't find a lwp for process"
699 " %d (%s)", p->p_pid, p->p_comm);
700 /* NOTREACHED */
701 return NULL;
702 }
703