Merge FreeBSD rev. 1.8:
[dragonfly.git] / sys / kern / lwkt_thread.c
CommitLineData
8ad65e08
MD
1/*
2 * Copyright (c) 2003 Matthew Dillon <dillon@backplane.com>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
0f7a3396 26 * $DragonFly: src/sys/kern/lwkt_thread.c,v 1.55 2004/02/17 19:38:49 dillon Exp $
75cdbe6c
MD
27 */
28
29/*
30 * Each cpu in a system has its own self-contained light weight kernel
31 * thread scheduler, which means that generally speaking we only need
32 * to use a critical section to avoid problems. Foreign thread
33 * scheduling is queued via (async) IPIs.
8ad65e08
MD
34 */
35
05220613
MD
36#ifdef _KERNEL
37
8ad65e08
MD
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/kernel.h>
41#include <sys/proc.h>
42#include <sys/rtprio.h>
43#include <sys/queue.h>
f1d1c3fa 44#include <sys/thread2.h>
7d0bac62 45#include <sys/sysctl.h>
99df837e 46#include <sys/kthread.h>
f1d1c3fa 47#include <machine/cpu.h>
99df837e 48#include <sys/lock.h>
f6bf3af1 49#include <sys/caps.h>
f1d1c3fa 50
7d0bac62
MD
51#include <vm/vm.h>
52#include <vm/vm_param.h>
53#include <vm/vm_kern.h>
54#include <vm/vm_object.h>
55#include <vm/vm_page.h>
56#include <vm/vm_map.h>
57#include <vm/vm_pager.h>
58#include <vm/vm_extern.h>
59#include <vm/vm_zone.h>
60
99df837e 61#include <machine/stdarg.h>
57c254db 62#include <machine/ipl.h>
96728c05 63#include <machine/smp.h>
99df837e 64
05220613
MD
65#define THREAD_STACK (UPAGES * PAGE_SIZE)
66
67#else
68
69#include <sys/stdint.h>
fb04f4fd 70#include <libcaps/thread.h>
05220613
MD
71#include <sys/thread.h>
72#include <sys/msgport.h>
73#include <sys/errno.h>
fb04f4fd 74#include <libcaps/globaldata.h>
05220613
MD
75#include <sys/thread2.h>
76#include <sys/msgport2.h>
709799ea 77#include <stdio.h>
05220613 78#include <stdlib.h>
709799ea 79#include <string.h>
c95cd171 80#include <machine/cpufunc.h>
709799ea 81#include <machine/lock.h>
05220613
MD
82
83#endif
84
7d0bac62 85static int untimely_switch = 0;
0f7a3396
MD
86#ifdef INVARIANTS
87static int panic_on_cscount = 0;
88#endif
05220613
MD
89static __int64_t switch_count = 0;
90static __int64_t preempt_hit = 0;
91static __int64_t preempt_miss = 0;
92static __int64_t preempt_weird = 0;
05220613
MD
93
94#ifdef _KERNEL
95
96SYSCTL_INT(_lwkt, OID_AUTO, untimely_switch, CTLFLAG_RW, &untimely_switch, 0, "");
0f7a3396
MD
97#ifdef INVARIANTS
98SYSCTL_INT(_lwkt, OID_AUTO, panic_on_cscount, CTLFLAG_RW, &panic_on_cscount, 0, "");
99#endif
4b5f931b 100SYSCTL_QUAD(_lwkt, OID_AUTO, switch_count, CTLFLAG_RW, &switch_count, 0, "");
4b5f931b 101SYSCTL_QUAD(_lwkt, OID_AUTO, preempt_hit, CTLFLAG_RW, &preempt_hit, 0, "");
4b5f931b 102SYSCTL_QUAD(_lwkt, OID_AUTO, preempt_miss, CTLFLAG_RW, &preempt_miss, 0, "");
26a0694b 103SYSCTL_QUAD(_lwkt, OID_AUTO, preempt_weird, CTLFLAG_RW, &preempt_weird, 0, "");
7d0bac62 104
05220613
MD
105#endif
106
4b5f931b
MD
107/*
108 * These helper procedures handle the runq, they can only be called from
109 * within a critical section.
75cdbe6c
MD
110 *
111 * WARNING! Prior to SMP being brought up it is possible to enqueue and
112 * dequeue threads belonging to other cpus, so be sure to use td->td_gd
113 * instead of 'mycpu' when referencing the globaldata structure. Once
114 * SMP live enqueuing and dequeueing only occurs on the current cpu.
4b5f931b 115 */
f1d1c3fa
MD
116static __inline
117void
118_lwkt_dequeue(thread_t td)
119{
120 if (td->td_flags & TDF_RUNQ) {
4b5f931b 121 int nq = td->td_pri & TDPRI_MASK;
75cdbe6c 122 struct globaldata *gd = td->td_gd;
4b5f931b 123
f1d1c3fa 124 td->td_flags &= ~TDF_RUNQ;
4b5f931b
MD
125 TAILQ_REMOVE(&gd->gd_tdrunq[nq], td, td_threadq);
126 /* runqmask is passively cleaned up by the switcher */
f1d1c3fa
MD
127 }
128}
129
130static __inline
131void
132_lwkt_enqueue(thread_t td)
133{
134 if ((td->td_flags & TDF_RUNQ) == 0) {
4b5f931b 135 int nq = td->td_pri & TDPRI_MASK;
75cdbe6c 136 struct globaldata *gd = td->td_gd;
4b5f931b 137
f1d1c3fa 138 td->td_flags |= TDF_RUNQ;
4b5f931b
MD
139 TAILQ_INSERT_TAIL(&gd->gd_tdrunq[nq], td, td_threadq);
140 gd->gd_runqmask |= 1 << nq;
f1d1c3fa
MD
141 }
142}
8ad65e08 143
57c254db
MD
144static __inline
145int
146_lwkt_wantresched(thread_t ntd, thread_t cur)
147{
148 return((ntd->td_pri & TDPRI_MASK) > (cur->td_pri & TDPRI_MASK));
149}
150
2d93b37a
MD
151#ifdef _KERNEL
152
8ad65e08
MD
153/*
154 * LWKTs operate on a per-cpu basis
155 *
73e4f7b9 156 * WARNING! Called from early boot, 'mycpu' may not work yet.
8ad65e08
MD
157 */
158void
159lwkt_gdinit(struct globaldata *gd)
160{
4b5f931b
MD
161 int i;
162
163 for (i = 0; i < sizeof(gd->gd_tdrunq)/sizeof(gd->gd_tdrunq[0]); ++i)
164 TAILQ_INIT(&gd->gd_tdrunq[i]);
165 gd->gd_runqmask = 0;
73e4f7b9 166 TAILQ_INIT(&gd->gd_tdallq);
8ad65e08
MD
167}
168
2d93b37a
MD
169#endif /* _KERNEL */
170
7d0bac62
MD
171/*
172 * Initialize a thread wait structure prior to first use.
173 *
174 * NOTE! called from low level boot code, we cannot do anything fancy!
175 */
176void
177lwkt_init_wait(lwkt_wait_t w)
178{
179 TAILQ_INIT(&w->wa_waitq);
180}
181
182/*
183 * Create a new thread. The thread must be associated with a process context
75cdbe6c
MD
184 * or LWKT start address before it can be scheduled. If the target cpu is
185 * -1 the thread will be created on the current cpu.
0cfcada1
MD
186 *
187 * If you intend to create a thread without a process context this function
188 * does everything except load the startup and switcher function.
7d0bac62
MD
189 */
190thread_t
75cdbe6c 191lwkt_alloc_thread(struct thread *td, int cpu)
7d0bac62 192{
99df837e 193 void *stack;
ef0fdad1 194 int flags = 0;
7d0bac62 195
ef0fdad1 196 if (td == NULL) {
26a0694b 197 crit_enter();
ef0fdad1
MD
198 if (mycpu->gd_tdfreecount > 0) {
199 --mycpu->gd_tdfreecount;
200 td = TAILQ_FIRST(&mycpu->gd_tdfreeq);
d9eea1a5 201 KASSERT(td != NULL && (td->td_flags & TDF_RUNNING) == 0,
ef0fdad1
MD
202 ("lwkt_alloc_thread: unexpected NULL or corrupted td"));
203 TAILQ_REMOVE(&mycpu->gd_tdfreeq, td, td_threadq);
204 crit_exit();
205 stack = td->td_kstack;
206 flags = td->td_flags & (TDF_ALLOCATED_STACK|TDF_ALLOCATED_THREAD);
207 } else {
208 crit_exit();
05220613 209#ifdef _KERNEL
ef0fdad1 210 td = zalloc(thread_zone);
05220613
MD
211#else
212 td = malloc(sizeof(struct thread));
213#endif
ef0fdad1
MD
214 td->td_kstack = NULL;
215 flags |= TDF_ALLOCATED_THREAD;
216 }
217 }
218 if ((stack = td->td_kstack) == NULL) {
05220613
MD
219#ifdef _KERNEL
220 stack = (void *)kmem_alloc(kernel_map, THREAD_STACK);
221#else
fb04f4fd 222 stack = libcaps_alloc_stack(THREAD_STACK);
05220613 223#endif
ef0fdad1 224 flags |= TDF_ALLOCATED_STACK;
99df837e 225 }
75cdbe6c
MD
226 if (cpu < 0)
227 lwkt_init_thread(td, stack, flags, mycpu);
228 else
229 lwkt_init_thread(td, stack, flags, globaldata_find(cpu));
99df837e 230 return(td);
7d0bac62
MD
231}
232
709799ea
MD
233#ifdef _KERNEL
234
7d0bac62
MD
235/*
236 * Initialize a preexisting thread structure. This function is used by
237 * lwkt_alloc_thread() and also used to initialize the per-cpu idlethread.
238 *
f8c3996b
MD
239 * All threads start out in a critical section at a priority of
240 * TDPRI_KERN_DAEMON. Higher level code will modify the priority as
75cdbe6c
MD
241 * appropriate. This function may send an IPI message when the
242 * requested cpu is not the current cpu and consequently gd_tdallq may
243 * not be initialized synchronously from the point of view of the originating
244 * cpu.
245 *
246 * NOTE! we have to be careful in regards to creating threads for other cpus
247 * if SMP has not yet been activated.
7d0bac62 248 */
75cdbe6c
MD
249static void
250lwkt_init_thread_remote(void *arg)
251{
252 thread_t td = arg;
253
254 TAILQ_INSERT_TAIL(&td->td_gd->gd_tdallq, td, td_allq);
255}
256
7d0bac62 257void
26a0694b 258lwkt_init_thread(thread_t td, void *stack, int flags, struct globaldata *gd)
7d0bac62 259{
99df837e
MD
260 bzero(td, sizeof(struct thread));
261 td->td_kstack = stack;
262 td->td_flags |= flags;
26a0694b 263 td->td_gd = gd;
f8c3996b 264 td->td_pri = TDPRI_KERN_DAEMON + TDPRI_CRIT;
c95cd171 265 lwkt_initport(&td->td_msgport, td);
99df837e 266 pmap_init_thread(td);
0f7a3396
MD
267#ifdef SMP
268 if (gd == mycpu) {
75cdbe6c
MD
269 crit_enter();
270 TAILQ_INSERT_TAIL(&gd->gd_tdallq, td, td_allq);
271 crit_exit();
272 } else {
2db3b277 273 lwkt_send_ipiq(gd, lwkt_init_thread_remote, td);
75cdbe6c 274 }
0f7a3396
MD
275#else
276 crit_enter();
277 TAILQ_INSERT_TAIL(&gd->gd_tdallq, td, td_allq);
278 crit_exit();
279#endif
73e4f7b9
MD
280}
281
2d93b37a
MD
282#endif /* _KERNEL */
283
73e4f7b9
MD
284void
285lwkt_set_comm(thread_t td, const char *ctl, ...)
286{
e2565a42 287 __va_list va;
73e4f7b9 288
e2565a42 289 __va_start(va, ctl);
73e4f7b9 290 vsnprintf(td->td_comm, sizeof(td->td_comm), ctl, va);
e2565a42 291 __va_end(va);
7d0bac62
MD
292}
293
99df837e 294void
73e4f7b9 295lwkt_hold(thread_t td)
99df837e 296{
73e4f7b9
MD
297 ++td->td_refs;
298}
299
300void
301lwkt_rele(thread_t td)
302{
303 KKASSERT(td->td_refs > 0);
304 --td->td_refs;
305}
306
c95cd171
MD
307#ifdef _KERNEL
308
73e4f7b9
MD
309void
310lwkt_wait_free(thread_t td)
311{
312 while (td->td_refs)
377d4740 313 tsleep(td, 0, "tdreap", hz);
73e4f7b9
MD
314}
315
c95cd171
MD
316#endif
317
73e4f7b9
MD
318void
319lwkt_free_thread(thread_t td)
320{
321 struct globaldata *gd = mycpu;
322
d9eea1a5 323 KASSERT((td->td_flags & TDF_RUNNING) == 0,
99df837e
MD
324 ("lwkt_free_thread: did not exit! %p", td));
325
326 crit_enter();
73e4f7b9
MD
327 TAILQ_REMOVE(&gd->gd_tdallq, td, td_allq);
328 if (gd->gd_tdfreecount < CACHE_NTHREADS &&
99df837e
MD
329 (td->td_flags & TDF_ALLOCATED_THREAD)
330 ) {
73e4f7b9
MD
331 ++gd->gd_tdfreecount;
332 TAILQ_INSERT_HEAD(&gd->gd_tdfreeq, td, td_threadq);
99df837e
MD
333 crit_exit();
334 } else {
335 crit_exit();
336 if (td->td_kstack && (td->td_flags & TDF_ALLOCATED_STACK)) {
05220613
MD
337#ifdef _KERNEL
338 kmem_free(kernel_map, (vm_offset_t)td->td_kstack, THREAD_STACK);
339#else
fb04f4fd 340 libcaps_free_stack(td->td_kstack, THREAD_STACK);
05220613 341#endif
73e4f7b9 342 /* gd invalid */
99df837e
MD
343 td->td_kstack = NULL;
344 }
05220613
MD
345 if (td->td_flags & TDF_ALLOCATED_THREAD) {
346#ifdef _KERNEL
99df837e 347 zfree(thread_zone, td);
05220613
MD
348#else
349 free(td);
350#endif
351 }
99df837e
MD
352 }
353}
354
355
8ad65e08
MD
356/*
357 * Switch to the next runnable lwkt. If no LWKTs are runnable then
f1d1c3fa
MD
358 * switch to the idlethread. Switching must occur within a critical
359 * section to avoid races with the scheduling queue.
360 *
361 * We always have full control over our cpu's run queue. Other cpus
362 * that wish to manipulate our queue must use the cpu_*msg() calls to
363 * talk to our cpu, so a critical section is all that is needed and
364 * the result is very, very fast thread switching.
365 *
96728c05
MD
366 * The LWKT scheduler uses a fixed priority model and round-robins at
367 * each priority level. User process scheduling is a totally
368 * different beast and LWKT priorities should not be confused with
369 * user process priorities.
f1d1c3fa 370 *
96728c05
MD
371 * The MP lock may be out of sync with the thread's td_mpcount. lwkt_switch()
372 * cleans it up. Note that the td_switch() function cannot do anything that
373 * requires the MP lock since the MP lock will have already been setup for
71ef2f5c
MD
374 * the target thread (not the current thread). It's nice to have a scheduler
375 * that does not need the MP lock to work because it allows us to do some
376 * really cool high-performance MP lock optimizations.
8ad65e08 377 */
96728c05 378
8ad65e08
MD
379void
380lwkt_switch(void)
381{
4b5f931b 382 struct globaldata *gd;
f1d1c3fa 383 thread_t td = curthread;
8ad65e08 384 thread_t ntd;
8a8d5d85
MD
385#ifdef SMP
386 int mpheld;
387#endif
8ad65e08 388
46a3f46d
MD
389 /*
390 * Switching from within a 'fast' (non thread switched) interrupt is
391 * illegal.
392 */
393 if (mycpu->gd_intr_nesting_level && panicstr == NULL) {
03aa8d99 394 panic("lwkt_switch: cannot switch from within a fast interrupt, yet\n");
96728c05 395 }
ef0fdad1 396
cb973d15
MD
397 /*
398 * Passive release (used to transition from user to kernel mode
399 * when we block or switch rather then when we enter the kernel).
400 * This function is NOT called if we are switching into a preemption
401 * or returning from a preemption. Typically this causes us to lose
402 * our P_CURPROC designation (if we have one) and become a true LWKT
403 * thread, and may also hand P_CURPROC to another process and schedule
404 * its thread.
405 */
406 if (td->td_release)
407 td->td_release(td);
408
f1d1c3fa 409 crit_enter();
4b5f931b 410 ++switch_count;
8a8d5d85
MD
411
412#ifdef SMP
413 /*
414 * td_mpcount cannot be used to determine if we currently hold the
415 * MP lock because get_mplock() will increment it prior to attempting
71ef2f5c
MD
416 * to get the lock, and switch out if it can't. Our ownership of
417 * the actual lock will remain stable while we are in a critical section
418 * (but, of course, another cpu may own or release the lock so the
419 * actual value of mp_lock is not stable).
8a8d5d85
MD
420 */
421 mpheld = MP_LOCK_HELD();
0f7a3396
MD
422#ifdef INVARIANTS
423 if (td->td_cscount) {
424 printf("Diagnostic: attempt to switch while mastering cpusync: %p\n",
425 td);
426 if (panic_on_cscount)
427 panic("switching while mastering cpusync");
428 }
429#endif
8a8d5d85 430#endif
99df837e
MD
431 if ((ntd = td->td_preempted) != NULL) {
432 /*
433 * We had preempted another thread on this cpu, resume the preempted
26a0694b
MD
434 * thread. This occurs transparently, whether the preempted thread
435 * was scheduled or not (it may have been preempted after descheduling
8a8d5d85
MD
436 * itself).
437 *
438 * We have to setup the MP lock for the original thread after backing
439 * out the adjustment that was made to curthread when the original
440 * was preempted.
99df837e 441 */
26a0694b 442 KKASSERT(ntd->td_flags & TDF_PREEMPT_LOCK);
8a8d5d85 443#ifdef SMP
96728c05
MD
444 if (ntd->td_mpcount && mpheld == 0) {
445 panic("MPLOCK NOT HELD ON RETURN: %p %p %d %d\n",
446 td, ntd, td->td_mpcount, ntd->td_mpcount);
447 }
8a8d5d85
MD
448 if (ntd->td_mpcount) {
449 td->td_mpcount -= ntd->td_mpcount;
450 KKASSERT(td->td_mpcount >= 0);
451 }
452#endif
26a0694b 453 ntd->td_flags |= TDF_PREEMPT_DONE;
8a8d5d85 454 /* YYY release mp lock on switchback if original doesn't need it */
8ad65e08 455 } else {
4b5f931b
MD
456 /*
457 * Priority queue / round-robin at each priority. Note that user
458 * processes run at a fixed, low priority and the user process
459 * scheduler deals with interactions between user processes
460 * by scheduling and descheduling them from the LWKT queue as
461 * necessary.
8a8d5d85
MD
462 *
463 * We have to adjust the MP lock for the target thread. If we
464 * need the MP lock and cannot obtain it we try to locate a
465 * thread that does not need the MP lock.
4b5f931b
MD
466 */
467 gd = mycpu;
4b5f931b
MD
468again:
469 if (gd->gd_runqmask) {
470 int nq = bsrl(gd->gd_runqmask);
471 if ((ntd = TAILQ_FIRST(&gd->gd_tdrunq[nq])) == NULL) {
472 gd->gd_runqmask &= ~(1 << nq);
473 goto again;
474 }
8a8d5d85
MD
475#ifdef SMP
476 if (ntd->td_mpcount && mpheld == 0 && !cpu_try_mplock()) {
477 /*
96728c05
MD
478 * Target needs MP lock and we couldn't get it, try
479 * to locate a thread which does not need the MP lock
3c23a41a 480 * to run. If we cannot locate a thread spin in idle.
8a8d5d85
MD
481 */
482 u_int32_t rqmask = gd->gd_runqmask;
483 while (rqmask) {
484 TAILQ_FOREACH(ntd, &gd->gd_tdrunq[nq], td_threadq) {
485 if (ntd->td_mpcount == 0)
486 break;
487 }
488 if (ntd)
489 break;
490 rqmask &= ~(1 << nq);
491 nq = bsrl(rqmask);
492 }
493 if (ntd == NULL) {
a2a5ad0d
MD
494 ntd = &gd->gd_idlethread;
495 ntd->td_flags |= TDF_IDLE_NOHLT;
8a8d5d85
MD
496 } else {
497 TAILQ_REMOVE(&gd->gd_tdrunq[nq], ntd, td_threadq);
498 TAILQ_INSERT_TAIL(&gd->gd_tdrunq[nq], ntd, td_threadq);
499 }
500 } else {
501 TAILQ_REMOVE(&gd->gd_tdrunq[nq], ntd, td_threadq);
502 TAILQ_INSERT_TAIL(&gd->gd_tdrunq[nq], ntd, td_threadq);
503 }
504#else
4b5f931b
MD
505 TAILQ_REMOVE(&gd->gd_tdrunq[nq], ntd, td_threadq);
506 TAILQ_INSERT_TAIL(&gd->gd_tdrunq[nq], ntd, td_threadq);
8a8d5d85 507#endif
4b5f931b 508 } else {
3c23a41a 509 /*
60f945af
MD
510 * We have nothing to run but only let the idle loop halt
511 * the cpu if there are no pending interrupts.
3c23a41a 512 */
a2a5ad0d 513 ntd = &gd->gd_idlethread;
60f945af 514 if (gd->gd_reqflags & RQF_IDLECHECK_MASK)
3c23a41a 515 ntd->td_flags |= TDF_IDLE_NOHLT;
4b5f931b 516 }
f1d1c3fa 517 }
26a0694b
MD
518 KASSERT(ntd->td_pri >= TDPRI_CRIT,
519 ("priority problem in lwkt_switch %d %d", td->td_pri, ntd->td_pri));
8a8d5d85
MD
520
521 /*
522 * Do the actual switch. If the new target does not need the MP lock
523 * and we are holding it, release the MP lock. If the new target requires
524 * the MP lock we have already acquired it for the target.
525 */
526#ifdef SMP
527 if (ntd->td_mpcount == 0 ) {
528 if (MP_LOCK_HELD())
529 cpu_rel_mplock();
530 } else {
531 ASSERT_MP_LOCK_HELD();
532 }
533#endif
8a8d5d85 534 if (td != ntd) {
f1d1c3fa 535 td->td_switch(ntd);
8a8d5d85 536 }
96728c05 537
f1d1c3fa 538 crit_exit();
8ad65e08
MD
539}
540
cb973d15
MD
541/*
542 * Switch if another thread has a higher priority. Do not switch to other
543 * threads at the same priority.
544 */
545void
546lwkt_maybe_switch()
547{
548 struct globaldata *gd = mycpu;
549 struct thread *td = gd->gd_curthread;
550
551 if ((td->td_pri & TDPRI_MASK) < bsrl(gd->gd_runqmask)) {
552 lwkt_switch();
553 }
554}
555
b68b7282 556/*
96728c05
MD
557 * Request that the target thread preempt the current thread. Preemption
558 * only works under a specific set of conditions:
b68b7282 559 *
96728c05
MD
560 * - We are not preempting ourselves
561 * - The target thread is owned by the current cpu
562 * - We are not currently being preempted
563 * - The target is not currently being preempted
564 * - We are able to satisfy the target's MP lock requirements (if any).
565 *
566 * THE CALLER OF LWKT_PREEMPT() MUST BE IN A CRITICAL SECTION. Typically
567 * this is called via lwkt_schedule() through the td_preemptable callback.
568 * critpri is the managed critical priority that we should ignore in order
569 * to determine whether preemption is possible (aka usually just the crit
570 * priority of lwkt_schedule() itself).
b68b7282 571 *
26a0694b
MD
572 * XXX at the moment we run the target thread in a critical section during
573 * the preemption in order to prevent the target from taking interrupts
574 * that *WE* can't. Preemption is strictly limited to interrupt threads
575 * and interrupt-like threads, outside of a critical section, and the
576 * preempted source thread will be resumed the instant the target blocks
577 * whether or not the source is scheduled (i.e. preemption is supposed to
578 * be as transparent as possible).
4b5f931b 579 *
8a8d5d85
MD
580 * The target thread inherits our MP count (added to its own) for the
581 * duration of the preemption in order to preserve the atomicy of the
96728c05
MD
582 * MP lock during the preemption. Therefore, any preempting targets must be
583 * careful in regards to MP assertions. Note that the MP count may be
71ef2f5c
MD
584 * out of sync with the physical mp_lock, but we do not have to preserve
585 * the original ownership of the lock if it was out of synch (that is, we
586 * can leave it synchronized on return).
b68b7282
MD
587 */
588void
96728c05 589lwkt_preempt(thread_t ntd, int critpri)
b68b7282 590{
46a3f46d
MD
591 struct globaldata *gd = mycpu;
592 thread_t td = gd->gd_curthread;
8a8d5d85
MD
593#ifdef SMP
594 int mpheld;
57c254db 595 int savecnt;
8a8d5d85 596#endif
b68b7282 597
26a0694b 598 /*
96728c05
MD
599 * The caller has put us in a critical section. We can only preempt
600 * if the caller of the caller was not in a critical section (basically
57c254db
MD
601 * a local interrupt), as determined by the 'critpri' parameter. If
602 * we are unable to preempt
96728c05
MD
603 *
604 * YYY The target thread must be in a critical section (else it must
605 * inherit our critical section? I dunno yet).
26a0694b
MD
606 */
607 KASSERT(ntd->td_pri >= TDPRI_CRIT, ("BADCRIT0 %d", ntd->td_pri));
26a0694b 608
cb973d15 609 need_resched();
57c254db
MD
610 if (!_lwkt_wantresched(ntd, td)) {
611 ++preempt_miss;
612 return;
613 }
96728c05
MD
614 if ((td->td_pri & ~TDPRI_MASK) > critpri) {
615 ++preempt_miss;
616 return;
617 }
618#ifdef SMP
46a3f46d 619 if (ntd->td_gd != gd) {
96728c05
MD
620 ++preempt_miss;
621 return;
622 }
623#endif
26a0694b
MD
624 if (td == ntd || ((td->td_flags | ntd->td_flags) & TDF_PREEMPT_LOCK)) {
625 ++preempt_weird;
626 return;
627 }
628 if (ntd->td_preempted) {
4b5f931b 629 ++preempt_hit;
26a0694b 630 return;
b68b7282 631 }
8a8d5d85 632#ifdef SMP
a2a5ad0d
MD
633 /*
634 * note: an interrupt might have occured just as we were transitioning
71ef2f5c
MD
635 * to or from the MP lock. In this case td_mpcount will be pre-disposed
636 * (non-zero) but not actually synchronized with the actual state of the
637 * lock. We can use it to imply an MP lock requirement for the
638 * preemption but we cannot use it to test whether we hold the MP lock
639 * or not.
a2a5ad0d 640 */
96728c05 641 savecnt = td->td_mpcount;
71ef2f5c 642 mpheld = MP_LOCK_HELD();
8a8d5d85
MD
643 ntd->td_mpcount += td->td_mpcount;
644 if (mpheld == 0 && ntd->td_mpcount && !cpu_try_mplock()) {
645 ntd->td_mpcount -= td->td_mpcount;
646 ++preempt_miss;
647 return;
648 }
649#endif
26a0694b
MD
650
651 ++preempt_hit;
652 ntd->td_preempted = td;
653 td->td_flags |= TDF_PREEMPT_LOCK;
654 td->td_switch(ntd);
655 KKASSERT(ntd->td_preempted && (td->td_flags & TDF_PREEMPT_DONE));
96728c05
MD
656#ifdef SMP
657 KKASSERT(savecnt == td->td_mpcount);
71ef2f5c
MD
658 mpheld = MP_LOCK_HELD();
659 if (mpheld && td->td_mpcount == 0)
96728c05 660 cpu_rel_mplock();
71ef2f5c 661 else if (mpheld == 0 && td->td_mpcount)
96728c05
MD
662 panic("lwkt_preempt(): MP lock was not held through");
663#endif
26a0694b
MD
664 ntd->td_preempted = NULL;
665 td->td_flags &= ~(TDF_PREEMPT_LOCK|TDF_PREEMPT_DONE);
b68b7282
MD
666}
667
f1d1c3fa
MD
668/*
669 * Yield our thread while higher priority threads are pending. This is
670 * typically called when we leave a critical section but it can be safely
671 * called while we are in a critical section.
672 *
673 * This function will not generally yield to equal priority threads but it
674 * can occur as a side effect. Note that lwkt_switch() is called from
46a3f46d 675 * inside the critical section to prevent its own crit_exit() from reentering
f1d1c3fa
MD
676 * lwkt_yield_quick().
677 *
235957ed 678 * gd_reqflags indicates that *something* changed, e.g. an interrupt or softint
ef0fdad1
MD
679 * came along but was blocked and made pending.
680 *
f1d1c3fa
MD
681 * (self contained on a per cpu basis)
682 */
683void
684lwkt_yield_quick(void)
685{
7966cb69
MD
686 globaldata_t gd = mycpu;
687 thread_t td = gd->gd_curthread;
ef0fdad1 688
a2a5ad0d 689 /*
235957ed 690 * gd_reqflags is cleared in splz if the cpl is 0. If we were to clear
a2a5ad0d
MD
691 * it with a non-zero cpl then we might not wind up calling splz after
692 * a task switch when the critical section is exited even though the
46a3f46d 693 * new task could accept the interrupt.
a2a5ad0d
MD
694 *
695 * XXX from crit_exit() only called after last crit section is released.
696 * If called directly will run splz() even if in a critical section.
46a3f46d
MD
697 *
698 * td_nest_count prevent deep nesting via splz() or doreti(). Note that
699 * except for this special case, we MUST call splz() here to handle any
700 * pending ints, particularly after we switch, or we might accidently
701 * halt the cpu with interrupts pending.
a2a5ad0d 702 */
46a3f46d 703 if (gd->gd_reqflags && td->td_nest_count < 2)
f1d1c3fa 704 splz();
f1d1c3fa
MD
705
706 /*
707 * YYY enabling will cause wakeup() to task-switch, which really
708 * confused the old 4.x code. This is a good way to simulate
7d0bac62
MD
709 * preemption and MP without actually doing preemption or MP, because a
710 * lot of code assumes that wakeup() does not block.
f1d1c3fa 711 */
46a3f46d
MD
712 if (untimely_switch && td->td_nest_count == 0 &&
713 gd->gd_intr_nesting_level == 0
714 ) {
f1d1c3fa
MD
715 crit_enter();
716 /*
717 * YYY temporary hacks until we disassociate the userland scheduler
718 * from the LWKT scheduler.
719 */
720 if (td->td_flags & TDF_RUNQ) {
721 lwkt_switch(); /* will not reenter yield function */
722 } else {
723 lwkt_schedule_self(); /* make sure we are scheduled */
724 lwkt_switch(); /* will not reenter yield function */
725 lwkt_deschedule_self(); /* make sure we are descheduled */
726 }
7966cb69 727 crit_exit_noyield(td);
f1d1c3fa 728 }
f1d1c3fa
MD
729}
730
8ad65e08 731/*
f1d1c3fa 732 * This implements a normal yield which, unlike _quick, will yield to equal
235957ed 733 * priority threads as well. Note that gd_reqflags tests will be handled by
f1d1c3fa
MD
734 * the crit_exit() call in lwkt_switch().
735 *
736 * (self contained on a per cpu basis)
8ad65e08
MD
737 */
738void
f1d1c3fa 739lwkt_yield(void)
8ad65e08 740{
f1d1c3fa
MD
741 lwkt_schedule_self();
742 lwkt_switch();
743}
744
745/*
746 * Schedule a thread to run. As the current thread we can always safely
747 * schedule ourselves, and a shortcut procedure is provided for that
748 * function.
749 *
750 * (non-blocking, self contained on a per cpu basis)
751 */
752void
753lwkt_schedule_self(void)
754{
755 thread_t td = curthread;
756
757 crit_enter();
758 KASSERT(td->td_wait == NULL, ("lwkt_schedule_self(): td_wait not NULL!"));
f1d1c3fa 759 _lwkt_enqueue(td);
05220613 760#ifdef _KERNEL
26a0694b
MD
761 if (td->td_proc && td->td_proc->p_stat == SSLEEP)
762 panic("SCHED SELF PANIC");
05220613 763#endif
f1d1c3fa 764 crit_exit();
8ad65e08 765}
8ad65e08
MD
766
767/*
f1d1c3fa
MD
768 * Generic schedule. Possibly schedule threads belonging to other cpus and
769 * deal with threads that might be blocked on a wait queue.
770 *
96728c05 771 * YYY this is one of the best places to implement load balancing code.
f1d1c3fa
MD
772 * Load balancing can be accomplished by requesting other sorts of actions
773 * for the thread in question.
8ad65e08
MD
774 */
775void
776lwkt_schedule(thread_t td)
777{
96728c05 778#ifdef INVARIANTS
26a0694b
MD
779 if ((td->td_flags & TDF_PREEMPT_LOCK) == 0 && td->td_proc
780 && td->td_proc->p_stat == SSLEEP
781 ) {
782 printf("PANIC schedule curtd = %p (%d %d) target %p (%d %d)\n",
783 curthread,
784 curthread->td_proc ? curthread->td_proc->p_pid : -1,
785 curthread->td_proc ? curthread->td_proc->p_stat : -1,
786 td,
787 td->td_proc ? curthread->td_proc->p_pid : -1,
788 td->td_proc ? curthread->td_proc->p_stat : -1
789 );
790 panic("SCHED PANIC");
791 }
96728c05 792#endif
f1d1c3fa
MD
793 crit_enter();
794 if (td == curthread) {
795 _lwkt_enqueue(td);
796 } else {
797 lwkt_wait_t w;
798
799 /*
800 * If the thread is on a wait list we have to send our scheduling
801 * request to the owner of the wait structure. Otherwise we send
802 * the scheduling request to the cpu owning the thread. Races
803 * are ok, the target will forward the message as necessary (the
804 * message may chase the thread around before it finally gets
805 * acted upon).
806 *
807 * (remember, wait structures use stable storage)
808 */
809 if ((w = td->td_wait) != NULL) {
96728c05 810 if (lwkt_trytoken(&w->wa_token)) {
f1d1c3fa
MD
811 TAILQ_REMOVE(&w->wa_waitq, td, td_threadq);
812 --w->wa_count;
813 td->td_wait = NULL;
0f7a3396
MD
814#ifdef SMP
815 if (td->td_gd == mycpu) {
f1d1c3fa 816 _lwkt_enqueue(td);
0f7a3396 817 if (td->td_preemptable)
96728c05 818 td->td_preemptable(td, TDPRI_CRIT*2); /* YYY +token */
0f7a3396 819 else if (_lwkt_wantresched(td, curthread))
57c254db 820 need_resched();
f1d1c3fa 821 } else {
2db3b277 822 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa 823 }
0f7a3396
MD
824#else
825 _lwkt_enqueue(td);
826 if (td->td_preemptable)
827 td->td_preemptable(td, TDPRI_CRIT*2); /* YYY +token */
828 else if (_lwkt_wantresched(td, curthread))
829 need_resched();
830#endif
96728c05 831 lwkt_reltoken(&w->wa_token);
f1d1c3fa 832 } else {
96728c05 833 lwkt_send_ipiq(w->wa_token.t_cpu, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa
MD
834 }
835 } else {
836 /*
837 * If the wait structure is NULL and we own the thread, there
838 * is no race (since we are in a critical section). If we
839 * do not own the thread there might be a race but the
840 * target cpu will deal with it.
841 */
0f7a3396
MD
842#ifdef SMP
843 if (td->td_gd == mycpu) {
f1d1c3fa 844 _lwkt_enqueue(td);
57c254db 845 if (td->td_preemptable) {
96728c05 846 td->td_preemptable(td, TDPRI_CRIT);
57c254db
MD
847 } else if (_lwkt_wantresched(td, curthread)) {
848 need_resched();
849 }
f1d1c3fa 850 } else {
2db3b277 851 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa 852 }
0f7a3396
MD
853#else
854 _lwkt_enqueue(td);
855 if (td->td_preemptable) {
856 td->td_preemptable(td, TDPRI_CRIT);
857 } else if (_lwkt_wantresched(td, curthread)) {
858 need_resched();
859 }
860#endif
f1d1c3fa 861 }
8ad65e08 862 }
f1d1c3fa 863 crit_exit();
8ad65e08
MD
864}
865
d9eea1a5
MD
866/*
867 * Managed acquisition. This code assumes that the MP lock is held for
868 * the tdallq operation and that the thread has been descheduled from its
869 * original cpu. We also have to wait for the thread to be entirely switched
870 * out on its original cpu (this is usually fast enough that we never loop)
871 * since the LWKT system does not have to hold the MP lock while switching
872 * and the target may have released it before switching.
873 */
a2a5ad0d
MD
874void
875lwkt_acquire(thread_t td)
876{
877 struct globaldata *gd;
878
879 gd = td->td_gd;
880 KKASSERT((td->td_flags & TDF_RUNQ) == 0);
d9eea1a5
MD
881 while (td->td_flags & TDF_RUNNING) /* XXX spin */
882 ;
a2a5ad0d
MD
883 if (gd != mycpu) {
884 crit_enter();
885 TAILQ_REMOVE(&gd->gd_tdallq, td, td_allq); /* protected by BGL */
886 gd = mycpu;
887 td->td_gd = gd;
a2a5ad0d
MD
888 TAILQ_INSERT_TAIL(&gd->gd_tdallq, td, td_allq); /* protected by BGL */
889 crit_exit();
890 }
891}
892
8ad65e08 893/*
f1d1c3fa
MD
894 * Deschedule a thread.
895 *
896 * (non-blocking, self contained on a per cpu basis)
897 */
898void
899lwkt_deschedule_self(void)
900{
901 thread_t td = curthread;
902
903 crit_enter();
904 KASSERT(td->td_wait == NULL, ("lwkt_schedule_self(): td_wait not NULL!"));
f1d1c3fa
MD
905 _lwkt_dequeue(td);
906 crit_exit();
907}
908
909/*
910 * Generic deschedule. Descheduling threads other then your own should be
911 * done only in carefully controlled circumstances. Descheduling is
912 * asynchronous.
913 *
914 * This function may block if the cpu has run out of messages.
8ad65e08
MD
915 */
916void
917lwkt_deschedule(thread_t td)
918{
f1d1c3fa
MD
919 crit_enter();
920 if (td == curthread) {
921 _lwkt_dequeue(td);
922 } else {
a72187e9 923 if (td->td_gd == mycpu) {
f1d1c3fa
MD
924 _lwkt_dequeue(td);
925 } else {
2db3b277 926 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_deschedule, td);
f1d1c3fa
MD
927 }
928 }
929 crit_exit();
930}
931
4b5f931b
MD
932/*
933 * Set the target thread's priority. This routine does not automatically
934 * switch to a higher priority thread, LWKT threads are not designed for
935 * continuous priority changes. Yield if you want to switch.
936 *
937 * We have to retain the critical section count which uses the high bits
26a0694b
MD
938 * of the td_pri field. The specified priority may also indicate zero or
939 * more critical sections by adding TDPRI_CRIT*N.
4b5f931b
MD
940 */
941void
942lwkt_setpri(thread_t td, int pri)
943{
26a0694b 944 KKASSERT(pri >= 0);
a72187e9 945 KKASSERT(td->td_gd == mycpu);
26a0694b
MD
946 crit_enter();
947 if (td->td_flags & TDF_RUNQ) {
948 _lwkt_dequeue(td);
949 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
950 _lwkt_enqueue(td);
951 } else {
952 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
953 }
954 crit_exit();
955}
956
957void
958lwkt_setpri_self(int pri)
959{
960 thread_t td = curthread;
961
4b5f931b
MD
962 KKASSERT(pri >= 0 && pri <= TDPRI_MAX);
963 crit_enter();
964 if (td->td_flags & TDF_RUNQ) {
965 _lwkt_dequeue(td);
966 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
967 _lwkt_enqueue(td);
968 } else {
969 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
970 }
971 crit_exit();
972}
973
974struct proc *
975lwkt_preempted_proc(void)
976{
73e4f7b9 977 thread_t td = curthread;
4b5f931b
MD
978 while (td->td_preempted)
979 td = td->td_preempted;
980 return(td->td_proc);
981}
982
ece04fd0 983#if 0
4b5f931b 984
f1d1c3fa
MD
985/*
986 * This function deschedules the current thread and blocks on the specified
987 * wait queue. We obtain ownership of the wait queue in order to block
988 * on it. A generation number is used to interlock the wait queue in case
989 * it gets signalled while we are blocked waiting on the token.
990 *
991 * Note: alternatively we could dequeue our thread and then message the
992 * target cpu owning the wait queue. YYY implement as sysctl.
993 *
994 * Note: wait queue signals normally ping-pong the cpu as an optimization.
995 */
96728c05 996
f1d1c3fa 997void
ae8050a4 998lwkt_block(lwkt_wait_t w, const char *wmesg, int *gen)
f1d1c3fa
MD
999{
1000 thread_t td = curthread;
f1d1c3fa 1001
f1d1c3fa 1002 lwkt_gettoken(&w->wa_token);
ae8050a4 1003 if (w->wa_gen == *gen) {
f1d1c3fa
MD
1004 _lwkt_dequeue(td);
1005 TAILQ_INSERT_TAIL(&w->wa_waitq, td, td_threadq);
1006 ++w->wa_count;
1007 td->td_wait = w;
ae8050a4 1008 td->td_wmesg = wmesg;
ece04fd0 1009again:
f1d1c3fa 1010 lwkt_switch();
ece04fd0
MD
1011 lwkt_regettoken(&w->wa_token);
1012 if (td->td_wmesg != NULL) {
1013 _lwkt_dequeue(td);
1014 goto again;
1015 }
8ad65e08 1016 }
ae8050a4
MD
1017 /* token might be lost, doesn't matter for gen update */
1018 *gen = w->wa_gen;
f1d1c3fa
MD
1019 lwkt_reltoken(&w->wa_token);
1020}
1021
1022/*
1023 * Signal a wait queue. We gain ownership of the wait queue in order to
1024 * signal it. Once a thread is removed from the wait queue we have to
1025 * deal with the cpu owning the thread.
1026 *
1027 * Note: alternatively we could message the target cpu owning the wait
1028 * queue. YYY implement as sysctl.
1029 */
1030void
ece04fd0 1031lwkt_signal(lwkt_wait_t w, int count)
f1d1c3fa
MD
1032{
1033 thread_t td;
1034 int count;
1035
1036 lwkt_gettoken(&w->wa_token);
1037 ++w->wa_gen;
ece04fd0
MD
1038 if (count < 0)
1039 count = w->wa_count;
f1d1c3fa
MD
1040 while ((td = TAILQ_FIRST(&w->wa_waitq)) != NULL && count) {
1041 --count;
1042 --w->wa_count;
1043 TAILQ_REMOVE(&w->wa_waitq, td, td_threadq);
1044 td->td_wait = NULL;
ae8050a4 1045 td->td_wmesg = NULL;
a72187e9 1046 if (td->td_gd == mycpu) {
f1d1c3fa
MD
1047 _lwkt_enqueue(td);
1048 } else {
2db3b277 1049 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa
MD
1050 }
1051 lwkt_regettoken(&w->wa_token);
1052 }
1053 lwkt_reltoken(&w->wa_token);
1054}
1055
ece04fd0
MD
1056#endif
1057
99df837e
MD
1058/*
1059 * Create a kernel process/thread/whatever. It shares it's address space
1060 * with proc0 - ie: kernel only.
1061 *
365fa13f
MD
1062 * NOTE! By default new threads are created with the MP lock held. A
1063 * thread which does not require the MP lock should release it by calling
1064 * rel_mplock() at the start of the new thread.
99df837e
MD
1065 */
1066int
1067lwkt_create(void (*func)(void *), void *arg,
75cdbe6c 1068 struct thread **tdp, thread_t template, int tdflags, int cpu,
ef0fdad1 1069 const char *fmt, ...)
99df837e 1070{
73e4f7b9 1071 thread_t td;
e2565a42 1072 __va_list ap;
99df837e 1073
75cdbe6c 1074 td = lwkt_alloc_thread(template, cpu);
a2a5ad0d
MD
1075 if (tdp)
1076 *tdp = td;
709799ea 1077 cpu_set_thread_handler(td, lwkt_exit, func, arg);
ef0fdad1 1078 td->td_flags |= TDF_VERBOSE | tdflags;
8a8d5d85
MD
1079#ifdef SMP
1080 td->td_mpcount = 1;
1081#endif
99df837e
MD
1082
1083 /*
1084 * Set up arg0 for 'ps' etc
1085 */
e2565a42 1086 __va_start(ap, fmt);
99df837e 1087 vsnprintf(td->td_comm, sizeof(td->td_comm), fmt, ap);
e2565a42 1088 __va_end(ap);
99df837e
MD
1089
1090 /*
1091 * Schedule the thread to run
1092 */
ef0fdad1
MD
1093 if ((td->td_flags & TDF_STOPREQ) == 0)
1094 lwkt_schedule(td);
1095 else
1096 td->td_flags &= ~TDF_STOPREQ;
99df837e
MD
1097 return 0;
1098}
1099
2d93b37a 1100/*
2d93b37a
MD
1101 * kthread_* is specific to the kernel and is not needed by userland.
1102 */
1103#ifdef _KERNEL
1104
99df837e
MD
1105/*
1106 * Destroy an LWKT thread. Warning! This function is not called when
1107 * a process exits, cpu_proc_exit() directly calls cpu_thread_exit() and
1108 * uses a different reaping mechanism.
1109 */
1110void
1111lwkt_exit(void)
1112{
1113 thread_t td = curthread;
1114
1115 if (td->td_flags & TDF_VERBOSE)
1116 printf("kthread %p %s has exited\n", td, td->td_comm);
f6bf3af1 1117 caps_exit(td);
99df837e
MD
1118 crit_enter();
1119 lwkt_deschedule_self();
1120 ++mycpu->gd_tdfreecount;
1121 TAILQ_INSERT_TAIL(&mycpu->gd_tdfreeq, td, td_threadq);
1122 cpu_thread_exit();
1123}
1124
1125/*
1126 * Create a kernel process/thread/whatever. It shares it's address space
ef0fdad1 1127 * with proc0 - ie: kernel only. 5.x compatible.
365fa13f
MD
1128 *
1129 * NOTE! By default kthreads are created with the MP lock held. A
1130 * thread which does not require the MP lock should release it by calling
1131 * rel_mplock() at the start of the new thread.
99df837e
MD
1132 */
1133int
1134kthread_create(void (*func)(void *), void *arg,
1135 struct thread **tdp, const char *fmt, ...)
1136{
73e4f7b9 1137 thread_t td;
e2565a42 1138 __va_list ap;
99df837e 1139
75cdbe6c 1140 td = lwkt_alloc_thread(NULL, -1);
a2a5ad0d
MD
1141 if (tdp)
1142 *tdp = td;
99df837e
MD
1143 cpu_set_thread_handler(td, kthread_exit, func, arg);
1144 td->td_flags |= TDF_VERBOSE;
8a8d5d85
MD
1145#ifdef SMP
1146 td->td_mpcount = 1;
1147#endif
99df837e
MD
1148
1149 /*
1150 * Set up arg0 for 'ps' etc
1151 */
e2565a42 1152 __va_start(ap, fmt);
99df837e 1153 vsnprintf(td->td_comm, sizeof(td->td_comm), fmt, ap);
e2565a42 1154 __va_end(ap);
99df837e
MD
1155
1156 /*
1157 * Schedule the thread to run
1158 */
1159 lwkt_schedule(td);
1160 return 0;
1161}
1162
1163/*
1164 * Destroy an LWKT thread. Warning! This function is not called when
1165 * a process exits, cpu_proc_exit() directly calls cpu_thread_exit() and
1166 * uses a different reaping mechanism.
1167 *
1168 * XXX duplicates lwkt_exit()
1169 */
1170void
1171kthread_exit(void)
1172{
1173 lwkt_exit();
1174}
1175
2d93b37a
MD
1176#endif /* _KERNEL */
1177
1178void
1179crit_panic(void)
1180{
1181 thread_t td = curthread;
1182 int lpri = td->td_pri;
1183
1184 td->td_pri = 0;
1185 panic("td_pri is/would-go negative! %p %d", td, lpri);
1186}
1187