Protect v_usecount with a critical section for now (we depend on the BGL),
[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 *
41a01a4d 26 * $DragonFly: src/sys/kern/lwkt_thread.c,v 1.56 2004/03/01 06:33:17 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
41a01a4d 177lwkt_wait_init(lwkt_wait_t w)
7d0bac62 178{
41a01a4d 179 lwkt_token_init(&w->wa_token);
7d0bac62 180 TAILQ_INIT(&w->wa_waitq);
41a01a4d
MD
181 w->wa_gen = 0;
182 w->wa_count = 0;
7d0bac62
MD
183}
184
185/*
186 * Create a new thread. The thread must be associated with a process context
75cdbe6c
MD
187 * or LWKT start address before it can be scheduled. If the target cpu is
188 * -1 the thread will be created on the current cpu.
0cfcada1
MD
189 *
190 * If you intend to create a thread without a process context this function
191 * does everything except load the startup and switcher function.
7d0bac62
MD
192 */
193thread_t
75cdbe6c 194lwkt_alloc_thread(struct thread *td, int cpu)
7d0bac62 195{
99df837e 196 void *stack;
ef0fdad1 197 int flags = 0;
7d0bac62 198
ef0fdad1 199 if (td == NULL) {
26a0694b 200 crit_enter();
ef0fdad1
MD
201 if (mycpu->gd_tdfreecount > 0) {
202 --mycpu->gd_tdfreecount;
203 td = TAILQ_FIRST(&mycpu->gd_tdfreeq);
d9eea1a5 204 KASSERT(td != NULL && (td->td_flags & TDF_RUNNING) == 0,
ef0fdad1
MD
205 ("lwkt_alloc_thread: unexpected NULL or corrupted td"));
206 TAILQ_REMOVE(&mycpu->gd_tdfreeq, td, td_threadq);
207 crit_exit();
208 stack = td->td_kstack;
209 flags = td->td_flags & (TDF_ALLOCATED_STACK|TDF_ALLOCATED_THREAD);
210 } else {
211 crit_exit();
05220613 212#ifdef _KERNEL
ef0fdad1 213 td = zalloc(thread_zone);
05220613
MD
214#else
215 td = malloc(sizeof(struct thread));
216#endif
ef0fdad1
MD
217 td->td_kstack = NULL;
218 flags |= TDF_ALLOCATED_THREAD;
219 }
220 }
221 if ((stack = td->td_kstack) == NULL) {
05220613
MD
222#ifdef _KERNEL
223 stack = (void *)kmem_alloc(kernel_map, THREAD_STACK);
224#else
fb04f4fd 225 stack = libcaps_alloc_stack(THREAD_STACK);
05220613 226#endif
ef0fdad1 227 flags |= TDF_ALLOCATED_STACK;
99df837e 228 }
75cdbe6c
MD
229 if (cpu < 0)
230 lwkt_init_thread(td, stack, flags, mycpu);
231 else
232 lwkt_init_thread(td, stack, flags, globaldata_find(cpu));
99df837e 233 return(td);
7d0bac62
MD
234}
235
709799ea
MD
236#ifdef _KERNEL
237
7d0bac62
MD
238/*
239 * Initialize a preexisting thread structure. This function is used by
240 * lwkt_alloc_thread() and also used to initialize the per-cpu idlethread.
241 *
f8c3996b
MD
242 * All threads start out in a critical section at a priority of
243 * TDPRI_KERN_DAEMON. Higher level code will modify the priority as
75cdbe6c
MD
244 * appropriate. This function may send an IPI message when the
245 * requested cpu is not the current cpu and consequently gd_tdallq may
246 * not be initialized synchronously from the point of view of the originating
247 * cpu.
248 *
249 * NOTE! we have to be careful in regards to creating threads for other cpus
250 * if SMP has not yet been activated.
7d0bac62 251 */
41a01a4d
MD
252#ifdef SMP
253
75cdbe6c
MD
254static void
255lwkt_init_thread_remote(void *arg)
256{
257 thread_t td = arg;
258
259 TAILQ_INSERT_TAIL(&td->td_gd->gd_tdallq, td, td_allq);
260}
261
41a01a4d
MD
262#endif
263
7d0bac62 264void
26a0694b 265lwkt_init_thread(thread_t td, void *stack, int flags, struct globaldata *gd)
7d0bac62 266{
99df837e
MD
267 bzero(td, sizeof(struct thread));
268 td->td_kstack = stack;
269 td->td_flags |= flags;
26a0694b 270 td->td_gd = gd;
f8c3996b 271 td->td_pri = TDPRI_KERN_DAEMON + TDPRI_CRIT;
c95cd171 272 lwkt_initport(&td->td_msgport, td);
99df837e 273 pmap_init_thread(td);
0f7a3396
MD
274#ifdef SMP
275 if (gd == mycpu) {
75cdbe6c
MD
276 crit_enter();
277 TAILQ_INSERT_TAIL(&gd->gd_tdallq, td, td_allq);
278 crit_exit();
279 } else {
2db3b277 280 lwkt_send_ipiq(gd, lwkt_init_thread_remote, td);
75cdbe6c 281 }
0f7a3396
MD
282#else
283 crit_enter();
284 TAILQ_INSERT_TAIL(&gd->gd_tdallq, td, td_allq);
285 crit_exit();
286#endif
73e4f7b9
MD
287}
288
2d93b37a
MD
289#endif /* _KERNEL */
290
73e4f7b9
MD
291void
292lwkt_set_comm(thread_t td, const char *ctl, ...)
293{
e2565a42 294 __va_list va;
73e4f7b9 295
e2565a42 296 __va_start(va, ctl);
73e4f7b9 297 vsnprintf(td->td_comm, sizeof(td->td_comm), ctl, va);
e2565a42 298 __va_end(va);
7d0bac62
MD
299}
300
99df837e 301void
73e4f7b9 302lwkt_hold(thread_t td)
99df837e 303{
73e4f7b9
MD
304 ++td->td_refs;
305}
306
307void
308lwkt_rele(thread_t td)
309{
310 KKASSERT(td->td_refs > 0);
311 --td->td_refs;
312}
313
c95cd171
MD
314#ifdef _KERNEL
315
73e4f7b9
MD
316void
317lwkt_wait_free(thread_t td)
318{
319 while (td->td_refs)
377d4740 320 tsleep(td, 0, "tdreap", hz);
73e4f7b9
MD
321}
322
c95cd171
MD
323#endif
324
73e4f7b9
MD
325void
326lwkt_free_thread(thread_t td)
327{
328 struct globaldata *gd = mycpu;
329
d9eea1a5 330 KASSERT((td->td_flags & TDF_RUNNING) == 0,
99df837e
MD
331 ("lwkt_free_thread: did not exit! %p", td));
332
333 crit_enter();
73e4f7b9
MD
334 TAILQ_REMOVE(&gd->gd_tdallq, td, td_allq);
335 if (gd->gd_tdfreecount < CACHE_NTHREADS &&
99df837e
MD
336 (td->td_flags & TDF_ALLOCATED_THREAD)
337 ) {
73e4f7b9
MD
338 ++gd->gd_tdfreecount;
339 TAILQ_INSERT_HEAD(&gd->gd_tdfreeq, td, td_threadq);
99df837e
MD
340 crit_exit();
341 } else {
342 crit_exit();
343 if (td->td_kstack && (td->td_flags & TDF_ALLOCATED_STACK)) {
05220613
MD
344#ifdef _KERNEL
345 kmem_free(kernel_map, (vm_offset_t)td->td_kstack, THREAD_STACK);
346#else
fb04f4fd 347 libcaps_free_stack(td->td_kstack, THREAD_STACK);
05220613 348#endif
73e4f7b9 349 /* gd invalid */
99df837e
MD
350 td->td_kstack = NULL;
351 }
05220613
MD
352 if (td->td_flags & TDF_ALLOCATED_THREAD) {
353#ifdef _KERNEL
99df837e 354 zfree(thread_zone, td);
05220613
MD
355#else
356 free(td);
357#endif
358 }
99df837e
MD
359 }
360}
361
362
8ad65e08
MD
363/*
364 * Switch to the next runnable lwkt. If no LWKTs are runnable then
f1d1c3fa
MD
365 * switch to the idlethread. Switching must occur within a critical
366 * section to avoid races with the scheduling queue.
367 *
368 * We always have full control over our cpu's run queue. Other cpus
369 * that wish to manipulate our queue must use the cpu_*msg() calls to
370 * talk to our cpu, so a critical section is all that is needed and
371 * the result is very, very fast thread switching.
372 *
96728c05
MD
373 * The LWKT scheduler uses a fixed priority model and round-robins at
374 * each priority level. User process scheduling is a totally
375 * different beast and LWKT priorities should not be confused with
376 * user process priorities.
f1d1c3fa 377 *
96728c05
MD
378 * The MP lock may be out of sync with the thread's td_mpcount. lwkt_switch()
379 * cleans it up. Note that the td_switch() function cannot do anything that
380 * requires the MP lock since the MP lock will have already been setup for
71ef2f5c
MD
381 * the target thread (not the current thread). It's nice to have a scheduler
382 * that does not need the MP lock to work because it allows us to do some
383 * really cool high-performance MP lock optimizations.
8ad65e08 384 */
96728c05 385
8ad65e08
MD
386void
387lwkt_switch(void)
388{
41a01a4d 389 globaldata_t gd;
f1d1c3fa 390 thread_t td = curthread;
8ad65e08 391 thread_t ntd;
8a8d5d85
MD
392#ifdef SMP
393 int mpheld;
394#endif
8ad65e08 395
46a3f46d
MD
396 /*
397 * Switching from within a 'fast' (non thread switched) interrupt is
398 * illegal.
399 */
400 if (mycpu->gd_intr_nesting_level && panicstr == NULL) {
03aa8d99 401 panic("lwkt_switch: cannot switch from within a fast interrupt, yet\n");
96728c05 402 }
ef0fdad1 403
cb973d15
MD
404 /*
405 * Passive release (used to transition from user to kernel mode
406 * when we block or switch rather then when we enter the kernel).
407 * This function is NOT called if we are switching into a preemption
408 * or returning from a preemption. Typically this causes us to lose
409 * our P_CURPROC designation (if we have one) and become a true LWKT
410 * thread, and may also hand P_CURPROC to another process and schedule
411 * its thread.
412 */
413 if (td->td_release)
414 td->td_release(td);
415
f1d1c3fa 416 crit_enter();
4b5f931b 417 ++switch_count;
8a8d5d85
MD
418
419#ifdef SMP
420 /*
421 * td_mpcount cannot be used to determine if we currently hold the
422 * MP lock because get_mplock() will increment it prior to attempting
71ef2f5c
MD
423 * to get the lock, and switch out if it can't. Our ownership of
424 * the actual lock will remain stable while we are in a critical section
425 * (but, of course, another cpu may own or release the lock so the
426 * actual value of mp_lock is not stable).
8a8d5d85
MD
427 */
428 mpheld = MP_LOCK_HELD();
0f7a3396
MD
429#ifdef INVARIANTS
430 if (td->td_cscount) {
431 printf("Diagnostic: attempt to switch while mastering cpusync: %p\n",
432 td);
433 if (panic_on_cscount)
434 panic("switching while mastering cpusync");
435 }
436#endif
8a8d5d85 437#endif
99df837e
MD
438 if ((ntd = td->td_preempted) != NULL) {
439 /*
440 * We had preempted another thread on this cpu, resume the preempted
26a0694b
MD
441 * thread. This occurs transparently, whether the preempted thread
442 * was scheduled or not (it may have been preempted after descheduling
8a8d5d85
MD
443 * itself).
444 *
445 * We have to setup the MP lock for the original thread after backing
446 * out the adjustment that was made to curthread when the original
447 * was preempted.
99df837e 448 */
26a0694b 449 KKASSERT(ntd->td_flags & TDF_PREEMPT_LOCK);
8a8d5d85 450#ifdef SMP
96728c05
MD
451 if (ntd->td_mpcount && mpheld == 0) {
452 panic("MPLOCK NOT HELD ON RETURN: %p %p %d %d\n",
453 td, ntd, td->td_mpcount, ntd->td_mpcount);
454 }
8a8d5d85
MD
455 if (ntd->td_mpcount) {
456 td->td_mpcount -= ntd->td_mpcount;
457 KKASSERT(td->td_mpcount >= 0);
458 }
459#endif
26a0694b 460 ntd->td_flags |= TDF_PREEMPT_DONE;
8a8d5d85 461 /* YYY release mp lock on switchback if original doesn't need it */
8ad65e08 462 } else {
4b5f931b
MD
463 /*
464 * Priority queue / round-robin at each priority. Note that user
465 * processes run at a fixed, low priority and the user process
466 * scheduler deals with interactions between user processes
467 * by scheduling and descheduling them from the LWKT queue as
468 * necessary.
8a8d5d85
MD
469 *
470 * We have to adjust the MP lock for the target thread. If we
471 * need the MP lock and cannot obtain it we try to locate a
41a01a4d
MD
472 * thread that does not need the MP lock. If we cannot, we spin
473 * instead of HLT.
474 *
475 * A similar issue exists for the tokens held by the target thread.
476 * If we cannot obtain ownership of the tokens we cannot immediately
477 * schedule the thread.
478 */
479
480 /*
481 * We are switching threads. If there are any pending requests for
482 * tokens we can satisfy all of them here.
4b5f931b
MD
483 */
484 gd = mycpu;
41a01a4d
MD
485#ifdef SMP
486 if (gd->gd_tokreqbase)
487 lwkt_drain_token_requests();
488#endif
489
4b5f931b
MD
490again:
491 if (gd->gd_runqmask) {
492 int nq = bsrl(gd->gd_runqmask);
493 if ((ntd = TAILQ_FIRST(&gd->gd_tdrunq[nq])) == NULL) {
494 gd->gd_runqmask &= ~(1 << nq);
495 goto again;
496 }
8a8d5d85 497#ifdef SMP
41a01a4d
MD
498 /*
499 * If the target needs the MP lock and we couldn't get it,
500 * or if the target is holding tokens and we could not
501 * gain ownership of the tokens, continue looking for a
502 * thread to schedule and spin instead of HLT if we can't.
503 */
504 if ((ntd->td_mpcount && mpheld == 0 && !cpu_try_mplock()) ||
505 (ntd->td_toks && lwkt_chktokens(ntd) == 0)
506 ) {
8a8d5d85
MD
507 u_int32_t rqmask = gd->gd_runqmask;
508 while (rqmask) {
509 TAILQ_FOREACH(ntd, &gd->gd_tdrunq[nq], td_threadq) {
41a01a4d
MD
510 if (ntd->td_mpcount && !mpheld && !cpu_try_mplock())
511 continue;
512 mpheld = MP_LOCK_HELD();
513 if (ntd->td_toks && !lwkt_chktokens(ntd))
514 continue;
515 break;
8a8d5d85
MD
516 }
517 if (ntd)
518 break;
519 rqmask &= ~(1 << nq);
520 nq = bsrl(rqmask);
521 }
522 if (ntd == NULL) {
a2a5ad0d
MD
523 ntd = &gd->gd_idlethread;
524 ntd->td_flags |= TDF_IDLE_NOHLT;
8a8d5d85
MD
525 } else {
526 TAILQ_REMOVE(&gd->gd_tdrunq[nq], ntd, td_threadq);
527 TAILQ_INSERT_TAIL(&gd->gd_tdrunq[nq], ntd, td_threadq);
528 }
529 } else {
530 TAILQ_REMOVE(&gd->gd_tdrunq[nq], ntd, td_threadq);
531 TAILQ_INSERT_TAIL(&gd->gd_tdrunq[nq], ntd, td_threadq);
532 }
533#else
4b5f931b
MD
534 TAILQ_REMOVE(&gd->gd_tdrunq[nq], ntd, td_threadq);
535 TAILQ_INSERT_TAIL(&gd->gd_tdrunq[nq], ntd, td_threadq);
8a8d5d85 536#endif
4b5f931b 537 } else {
3c23a41a 538 /*
60f945af
MD
539 * We have nothing to run but only let the idle loop halt
540 * the cpu if there are no pending interrupts.
3c23a41a 541 */
a2a5ad0d 542 ntd = &gd->gd_idlethread;
60f945af 543 if (gd->gd_reqflags & RQF_IDLECHECK_MASK)
3c23a41a 544 ntd->td_flags |= TDF_IDLE_NOHLT;
4b5f931b 545 }
f1d1c3fa 546 }
26a0694b
MD
547 KASSERT(ntd->td_pri >= TDPRI_CRIT,
548 ("priority problem in lwkt_switch %d %d", td->td_pri, ntd->td_pri));
8a8d5d85
MD
549
550 /*
551 * Do the actual switch. If the new target does not need the MP lock
552 * and we are holding it, release the MP lock. If the new target requires
553 * the MP lock we have already acquired it for the target.
554 */
555#ifdef SMP
556 if (ntd->td_mpcount == 0 ) {
557 if (MP_LOCK_HELD())
558 cpu_rel_mplock();
559 } else {
560 ASSERT_MP_LOCK_HELD();
561 }
562#endif
8a8d5d85 563 if (td != ntd) {
f1d1c3fa 564 td->td_switch(ntd);
8a8d5d85 565 }
96728c05 566
f1d1c3fa 567 crit_exit();
8ad65e08
MD
568}
569
cb973d15
MD
570/*
571 * Switch if another thread has a higher priority. Do not switch to other
572 * threads at the same priority.
573 */
574void
575lwkt_maybe_switch()
576{
577 struct globaldata *gd = mycpu;
578 struct thread *td = gd->gd_curthread;
579
580 if ((td->td_pri & TDPRI_MASK) < bsrl(gd->gd_runqmask)) {
581 lwkt_switch();
582 }
583}
584
b68b7282 585/*
96728c05
MD
586 * Request that the target thread preempt the current thread. Preemption
587 * only works under a specific set of conditions:
b68b7282 588 *
96728c05
MD
589 * - We are not preempting ourselves
590 * - The target thread is owned by the current cpu
591 * - We are not currently being preempted
592 * - The target is not currently being preempted
593 * - We are able to satisfy the target's MP lock requirements (if any).
594 *
595 * THE CALLER OF LWKT_PREEMPT() MUST BE IN A CRITICAL SECTION. Typically
596 * this is called via lwkt_schedule() through the td_preemptable callback.
597 * critpri is the managed critical priority that we should ignore in order
598 * to determine whether preemption is possible (aka usually just the crit
599 * priority of lwkt_schedule() itself).
b68b7282 600 *
26a0694b
MD
601 * XXX at the moment we run the target thread in a critical section during
602 * the preemption in order to prevent the target from taking interrupts
603 * that *WE* can't. Preemption is strictly limited to interrupt threads
604 * and interrupt-like threads, outside of a critical section, and the
605 * preempted source thread will be resumed the instant the target blocks
606 * whether or not the source is scheduled (i.e. preemption is supposed to
607 * be as transparent as possible).
4b5f931b 608 *
8a8d5d85
MD
609 * The target thread inherits our MP count (added to its own) for the
610 * duration of the preemption in order to preserve the atomicy of the
96728c05
MD
611 * MP lock during the preemption. Therefore, any preempting targets must be
612 * careful in regards to MP assertions. Note that the MP count may be
71ef2f5c
MD
613 * out of sync with the physical mp_lock, but we do not have to preserve
614 * the original ownership of the lock if it was out of synch (that is, we
615 * can leave it synchronized on return).
b68b7282
MD
616 */
617void
96728c05 618lwkt_preempt(thread_t ntd, int critpri)
b68b7282 619{
46a3f46d
MD
620 struct globaldata *gd = mycpu;
621 thread_t td = gd->gd_curthread;
8a8d5d85
MD
622#ifdef SMP
623 int mpheld;
57c254db 624 int savecnt;
8a8d5d85 625#endif
b68b7282 626
26a0694b 627 /*
96728c05
MD
628 * The caller has put us in a critical section. We can only preempt
629 * if the caller of the caller was not in a critical section (basically
57c254db
MD
630 * a local interrupt), as determined by the 'critpri' parameter. If
631 * we are unable to preempt
96728c05
MD
632 *
633 * YYY The target thread must be in a critical section (else it must
634 * inherit our critical section? I dunno yet).
41a01a4d
MD
635 *
636 * Any tokens held by the target may not be held by thread(s) being
637 * preempted. We take the easy way out and do not preempt if
638 * the target is holding tokens.
26a0694b
MD
639 */
640 KASSERT(ntd->td_pri >= TDPRI_CRIT, ("BADCRIT0 %d", ntd->td_pri));
26a0694b 641
cb973d15 642 need_resched();
57c254db
MD
643 if (!_lwkt_wantresched(ntd, td)) {
644 ++preempt_miss;
645 return;
646 }
96728c05
MD
647 if ((td->td_pri & ~TDPRI_MASK) > critpri) {
648 ++preempt_miss;
649 return;
650 }
651#ifdef SMP
46a3f46d 652 if (ntd->td_gd != gd) {
96728c05
MD
653 ++preempt_miss;
654 return;
655 }
656#endif
41a01a4d
MD
657 /*
658 * Take the easy way out and do not preempt if the target is holding
659 * one or more tokens. We could test whether the thread(s) being
660 * preempted interlock against the target thread's tokens and whether
661 * we can get all the target thread's tokens, but this situation
662 * should not occur very often so its easier to simply not preempt.
663 */
664 if (ntd->td_toks != NULL) {
665 ++preempt_miss;
666 return;
667 }
26a0694b
MD
668 if (td == ntd || ((td->td_flags | ntd->td_flags) & TDF_PREEMPT_LOCK)) {
669 ++preempt_weird;
670 return;
671 }
672 if (ntd->td_preempted) {
4b5f931b 673 ++preempt_hit;
26a0694b 674 return;
b68b7282 675 }
8a8d5d85 676#ifdef SMP
a2a5ad0d
MD
677 /*
678 * note: an interrupt might have occured just as we were transitioning
71ef2f5c
MD
679 * to or from the MP lock. In this case td_mpcount will be pre-disposed
680 * (non-zero) but not actually synchronized with the actual state of the
681 * lock. We can use it to imply an MP lock requirement for the
682 * preemption but we cannot use it to test whether we hold the MP lock
683 * or not.
a2a5ad0d 684 */
96728c05 685 savecnt = td->td_mpcount;
71ef2f5c 686 mpheld = MP_LOCK_HELD();
8a8d5d85
MD
687 ntd->td_mpcount += td->td_mpcount;
688 if (mpheld == 0 && ntd->td_mpcount && !cpu_try_mplock()) {
689 ntd->td_mpcount -= td->td_mpcount;
690 ++preempt_miss;
691 return;
692 }
693#endif
26a0694b
MD
694
695 ++preempt_hit;
696 ntd->td_preempted = td;
697 td->td_flags |= TDF_PREEMPT_LOCK;
698 td->td_switch(ntd);
699 KKASSERT(ntd->td_preempted && (td->td_flags & TDF_PREEMPT_DONE));
96728c05
MD
700#ifdef SMP
701 KKASSERT(savecnt == td->td_mpcount);
71ef2f5c
MD
702 mpheld = MP_LOCK_HELD();
703 if (mpheld && td->td_mpcount == 0)
96728c05 704 cpu_rel_mplock();
71ef2f5c 705 else if (mpheld == 0 && td->td_mpcount)
96728c05
MD
706 panic("lwkt_preempt(): MP lock was not held through");
707#endif
26a0694b
MD
708 ntd->td_preempted = NULL;
709 td->td_flags &= ~(TDF_PREEMPT_LOCK|TDF_PREEMPT_DONE);
b68b7282
MD
710}
711
f1d1c3fa
MD
712/*
713 * Yield our thread while higher priority threads are pending. This is
714 * typically called when we leave a critical section but it can be safely
715 * called while we are in a critical section.
716 *
717 * This function will not generally yield to equal priority threads but it
718 * can occur as a side effect. Note that lwkt_switch() is called from
46a3f46d 719 * inside the critical section to prevent its own crit_exit() from reentering
f1d1c3fa
MD
720 * lwkt_yield_quick().
721 *
235957ed 722 * gd_reqflags indicates that *something* changed, e.g. an interrupt or softint
ef0fdad1
MD
723 * came along but was blocked and made pending.
724 *
f1d1c3fa
MD
725 * (self contained on a per cpu basis)
726 */
727void
728lwkt_yield_quick(void)
729{
7966cb69
MD
730 globaldata_t gd = mycpu;
731 thread_t td = gd->gd_curthread;
ef0fdad1 732
a2a5ad0d 733 /*
235957ed 734 * gd_reqflags is cleared in splz if the cpl is 0. If we were to clear
a2a5ad0d
MD
735 * it with a non-zero cpl then we might not wind up calling splz after
736 * a task switch when the critical section is exited even though the
46a3f46d 737 * new task could accept the interrupt.
a2a5ad0d
MD
738 *
739 * XXX from crit_exit() only called after last crit section is released.
740 * If called directly will run splz() even if in a critical section.
46a3f46d
MD
741 *
742 * td_nest_count prevent deep nesting via splz() or doreti(). Note that
743 * except for this special case, we MUST call splz() here to handle any
744 * pending ints, particularly after we switch, or we might accidently
745 * halt the cpu with interrupts pending.
a2a5ad0d 746 */
46a3f46d 747 if (gd->gd_reqflags && td->td_nest_count < 2)
f1d1c3fa 748 splz();
f1d1c3fa
MD
749
750 /*
751 * YYY enabling will cause wakeup() to task-switch, which really
752 * confused the old 4.x code. This is a good way to simulate
7d0bac62
MD
753 * preemption and MP without actually doing preemption or MP, because a
754 * lot of code assumes that wakeup() does not block.
f1d1c3fa 755 */
46a3f46d
MD
756 if (untimely_switch && td->td_nest_count == 0 &&
757 gd->gd_intr_nesting_level == 0
758 ) {
f1d1c3fa
MD
759 crit_enter();
760 /*
761 * YYY temporary hacks until we disassociate the userland scheduler
762 * from the LWKT scheduler.
763 */
764 if (td->td_flags & TDF_RUNQ) {
765 lwkt_switch(); /* will not reenter yield function */
766 } else {
767 lwkt_schedule_self(); /* make sure we are scheduled */
768 lwkt_switch(); /* will not reenter yield function */
769 lwkt_deschedule_self(); /* make sure we are descheduled */
770 }
7966cb69 771 crit_exit_noyield(td);
f1d1c3fa 772 }
f1d1c3fa
MD
773}
774
8ad65e08 775/*
f1d1c3fa 776 * This implements a normal yield which, unlike _quick, will yield to equal
235957ed 777 * priority threads as well. Note that gd_reqflags tests will be handled by
f1d1c3fa
MD
778 * the crit_exit() call in lwkt_switch().
779 *
780 * (self contained on a per cpu basis)
8ad65e08
MD
781 */
782void
f1d1c3fa 783lwkt_yield(void)
8ad65e08 784{
f1d1c3fa
MD
785 lwkt_schedule_self();
786 lwkt_switch();
787}
788
789/*
790 * Schedule a thread to run. As the current thread we can always safely
791 * schedule ourselves, and a shortcut procedure is provided for that
792 * function.
793 *
794 * (non-blocking, self contained on a per cpu basis)
795 */
796void
797lwkt_schedule_self(void)
798{
799 thread_t td = curthread;
800
41a01a4d 801 crit_enter_quick(td);
f1d1c3fa 802 KASSERT(td->td_wait == NULL, ("lwkt_schedule_self(): td_wait not NULL!"));
41a01a4d 803 KASSERT(td != &td->td_gd->gd_idlethread, ("lwkt_schedule_self(): scheduling gd_idlethread is illegal!"));
f1d1c3fa 804 _lwkt_enqueue(td);
05220613 805#ifdef _KERNEL
26a0694b
MD
806 if (td->td_proc && td->td_proc->p_stat == SSLEEP)
807 panic("SCHED SELF PANIC");
05220613 808#endif
41a01a4d 809 crit_exit_quick(td);
8ad65e08 810}
8ad65e08
MD
811
812/*
f1d1c3fa
MD
813 * Generic schedule. Possibly schedule threads belonging to other cpus and
814 * deal with threads that might be blocked on a wait queue.
815 *
96728c05 816 * YYY this is one of the best places to implement load balancing code.
f1d1c3fa
MD
817 * Load balancing can be accomplished by requesting other sorts of actions
818 * for the thread in question.
8ad65e08
MD
819 */
820void
821lwkt_schedule(thread_t td)
822{
96728c05 823#ifdef INVARIANTS
41a01a4d 824 KASSERT(td != &td->td_gd->gd_idlethread, ("lwkt_schedule(): scheduling gd_idlethread is illegal!"));
26a0694b
MD
825 if ((td->td_flags & TDF_PREEMPT_LOCK) == 0 && td->td_proc
826 && td->td_proc->p_stat == SSLEEP
827 ) {
828 printf("PANIC schedule curtd = %p (%d %d) target %p (%d %d)\n",
829 curthread,
830 curthread->td_proc ? curthread->td_proc->p_pid : -1,
831 curthread->td_proc ? curthread->td_proc->p_stat : -1,
832 td,
833 td->td_proc ? curthread->td_proc->p_pid : -1,
834 td->td_proc ? curthread->td_proc->p_stat : -1
835 );
836 panic("SCHED PANIC");
837 }
96728c05 838#endif
f1d1c3fa
MD
839 crit_enter();
840 if (td == curthread) {
841 _lwkt_enqueue(td);
842 } else {
843 lwkt_wait_t w;
844
845 /*
846 * If the thread is on a wait list we have to send our scheduling
847 * request to the owner of the wait structure. Otherwise we send
848 * the scheduling request to the cpu owning the thread. Races
849 * are ok, the target will forward the message as necessary (the
850 * message may chase the thread around before it finally gets
851 * acted upon).
852 *
853 * (remember, wait structures use stable storage)
854 */
855 if ((w = td->td_wait) != NULL) {
41a01a4d
MD
856 lwkt_tokref wref;
857
858 if (lwkt_trytoken(&wref, &w->wa_token)) {
f1d1c3fa
MD
859 TAILQ_REMOVE(&w->wa_waitq, td, td_threadq);
860 --w->wa_count;
861 td->td_wait = NULL;
0f7a3396
MD
862#ifdef SMP
863 if (td->td_gd == mycpu) {
f1d1c3fa 864 _lwkt_enqueue(td);
0f7a3396 865 if (td->td_preemptable)
96728c05 866 td->td_preemptable(td, TDPRI_CRIT*2); /* YYY +token */
0f7a3396 867 else if (_lwkt_wantresched(td, curthread))
57c254db 868 need_resched();
f1d1c3fa 869 } else {
2db3b277 870 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa 871 }
0f7a3396
MD
872#else
873 _lwkt_enqueue(td);
874 if (td->td_preemptable)
875 td->td_preemptable(td, TDPRI_CRIT*2); /* YYY +token */
876 else if (_lwkt_wantresched(td, curthread))
877 need_resched();
878#endif
41a01a4d 879 lwkt_reltoken(&wref);
f1d1c3fa 880 } else {
96728c05 881 lwkt_send_ipiq(w->wa_token.t_cpu, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa
MD
882 }
883 } else {
884 /*
885 * If the wait structure is NULL and we own the thread, there
886 * is no race (since we are in a critical section). If we
887 * do not own the thread there might be a race but the
888 * target cpu will deal with it.
889 */
0f7a3396
MD
890#ifdef SMP
891 if (td->td_gd == mycpu) {
f1d1c3fa 892 _lwkt_enqueue(td);
57c254db 893 if (td->td_preemptable) {
96728c05 894 td->td_preemptable(td, TDPRI_CRIT);
57c254db
MD
895 } else if (_lwkt_wantresched(td, curthread)) {
896 need_resched();
897 }
f1d1c3fa 898 } else {
2db3b277 899 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa 900 }
0f7a3396
MD
901#else
902 _lwkt_enqueue(td);
903 if (td->td_preemptable) {
904 td->td_preemptable(td, TDPRI_CRIT);
905 } else if (_lwkt_wantresched(td, curthread)) {
906 need_resched();
907 }
908#endif
f1d1c3fa 909 }
8ad65e08 910 }
f1d1c3fa 911 crit_exit();
8ad65e08
MD
912}
913
d9eea1a5
MD
914/*
915 * Managed acquisition. This code assumes that the MP lock is held for
916 * the tdallq operation and that the thread has been descheduled from its
917 * original cpu. We also have to wait for the thread to be entirely switched
918 * out on its original cpu (this is usually fast enough that we never loop)
919 * since the LWKT system does not have to hold the MP lock while switching
920 * and the target may have released it before switching.
921 */
a2a5ad0d
MD
922void
923lwkt_acquire(thread_t td)
924{
925 struct globaldata *gd;
926
927 gd = td->td_gd;
928 KKASSERT((td->td_flags & TDF_RUNQ) == 0);
d9eea1a5
MD
929 while (td->td_flags & TDF_RUNNING) /* XXX spin */
930 ;
a2a5ad0d
MD
931 if (gd != mycpu) {
932 crit_enter();
933 TAILQ_REMOVE(&gd->gd_tdallq, td, td_allq); /* protected by BGL */
934 gd = mycpu;
935 td->td_gd = gd;
a2a5ad0d
MD
936 TAILQ_INSERT_TAIL(&gd->gd_tdallq, td, td_allq); /* protected by BGL */
937 crit_exit();
938 }
939}
940
8ad65e08 941/*
f1d1c3fa
MD
942 * Deschedule a thread.
943 *
944 * (non-blocking, self contained on a per cpu basis)
945 */
946void
947lwkt_deschedule_self(void)
948{
949 thread_t td = curthread;
950
951 crit_enter();
952 KASSERT(td->td_wait == NULL, ("lwkt_schedule_self(): td_wait not NULL!"));
f1d1c3fa
MD
953 _lwkt_dequeue(td);
954 crit_exit();
955}
956
957/*
958 * Generic deschedule. Descheduling threads other then your own should be
959 * done only in carefully controlled circumstances. Descheduling is
960 * asynchronous.
961 *
962 * This function may block if the cpu has run out of messages.
8ad65e08
MD
963 */
964void
965lwkt_deschedule(thread_t td)
966{
f1d1c3fa
MD
967 crit_enter();
968 if (td == curthread) {
969 _lwkt_dequeue(td);
970 } else {
a72187e9 971 if (td->td_gd == mycpu) {
f1d1c3fa
MD
972 _lwkt_dequeue(td);
973 } else {
2db3b277 974 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_deschedule, td);
f1d1c3fa
MD
975 }
976 }
977 crit_exit();
978}
979
4b5f931b
MD
980/*
981 * Set the target thread's priority. This routine does not automatically
982 * switch to a higher priority thread, LWKT threads are not designed for
983 * continuous priority changes. Yield if you want to switch.
984 *
985 * We have to retain the critical section count which uses the high bits
26a0694b
MD
986 * of the td_pri field. The specified priority may also indicate zero or
987 * more critical sections by adding TDPRI_CRIT*N.
4b5f931b
MD
988 */
989void
990lwkt_setpri(thread_t td, int pri)
991{
26a0694b 992 KKASSERT(pri >= 0);
a72187e9 993 KKASSERT(td->td_gd == mycpu);
26a0694b
MD
994 crit_enter();
995 if (td->td_flags & TDF_RUNQ) {
996 _lwkt_dequeue(td);
997 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
998 _lwkt_enqueue(td);
999 } else {
1000 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
1001 }
1002 crit_exit();
1003}
1004
1005void
1006lwkt_setpri_self(int pri)
1007{
1008 thread_t td = curthread;
1009
4b5f931b
MD
1010 KKASSERT(pri >= 0 && pri <= TDPRI_MAX);
1011 crit_enter();
1012 if (td->td_flags & TDF_RUNQ) {
1013 _lwkt_dequeue(td);
1014 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
1015 _lwkt_enqueue(td);
1016 } else {
1017 td->td_pri = (td->td_pri & ~TDPRI_MASK) + pri;
1018 }
1019 crit_exit();
1020}
1021
1022struct proc *
1023lwkt_preempted_proc(void)
1024{
73e4f7b9 1025 thread_t td = curthread;
4b5f931b
MD
1026 while (td->td_preempted)
1027 td = td->td_preempted;
1028 return(td->td_proc);
1029}
1030
f1d1c3fa 1031/*
41a01a4d
MD
1032 * Block on the specified wait queue until signaled. A generation number
1033 * must be supplied to interlock the wait queue. The function will
1034 * return immediately if the generation number does not match the wait
1035 * structure's generation number.
f1d1c3fa
MD
1036 */
1037void
ae8050a4 1038lwkt_block(lwkt_wait_t w, const char *wmesg, int *gen)
f1d1c3fa
MD
1039{
1040 thread_t td = curthread;
41a01a4d 1041 lwkt_tokref ilock;
f1d1c3fa 1042
41a01a4d
MD
1043 lwkt_gettoken(&ilock, &w->wa_token);
1044 crit_enter();
ae8050a4 1045 if (w->wa_gen == *gen) {
f1d1c3fa
MD
1046 _lwkt_dequeue(td);
1047 TAILQ_INSERT_TAIL(&w->wa_waitq, td, td_threadq);
1048 ++w->wa_count;
1049 td->td_wait = w;
ae8050a4 1050 td->td_wmesg = wmesg;
41a01a4d 1051 again:
f1d1c3fa 1052 lwkt_switch();
ece04fd0
MD
1053 if (td->td_wmesg != NULL) {
1054 _lwkt_dequeue(td);
1055 goto again;
1056 }
8ad65e08 1057 }
41a01a4d 1058 crit_exit();
ae8050a4 1059 *gen = w->wa_gen;
41a01a4d 1060 lwkt_reltoken(&ilock);
f1d1c3fa
MD
1061}
1062
1063/*
1064 * Signal a wait queue. We gain ownership of the wait queue in order to
1065 * signal it. Once a thread is removed from the wait queue we have to
1066 * deal with the cpu owning the thread.
1067 *
1068 * Note: alternatively we could message the target cpu owning the wait
1069 * queue. YYY implement as sysctl.
1070 */
1071void
ece04fd0 1072lwkt_signal(lwkt_wait_t w, int count)
f1d1c3fa
MD
1073{
1074 thread_t td;
41a01a4d 1075 lwkt_tokref ilock;
f1d1c3fa 1076
41a01a4d 1077 lwkt_gettoken(&ilock, &w->wa_token);
f1d1c3fa 1078 ++w->wa_gen;
41a01a4d 1079 crit_enter();
ece04fd0
MD
1080 if (count < 0)
1081 count = w->wa_count;
f1d1c3fa
MD
1082 while ((td = TAILQ_FIRST(&w->wa_waitq)) != NULL && count) {
1083 --count;
1084 --w->wa_count;
1085 TAILQ_REMOVE(&w->wa_waitq, td, td_threadq);
1086 td->td_wait = NULL;
ae8050a4 1087 td->td_wmesg = NULL;
a72187e9 1088 if (td->td_gd == mycpu) {
f1d1c3fa
MD
1089 _lwkt_enqueue(td);
1090 } else {
2db3b277 1091 lwkt_send_ipiq(td->td_gd, (ipifunc_t)lwkt_schedule, td);
f1d1c3fa 1092 }
f1d1c3fa 1093 }
41a01a4d
MD
1094 crit_exit();
1095 lwkt_reltoken(&ilock);
f1d1c3fa
MD
1096}
1097
99df837e
MD
1098/*
1099 * Create a kernel process/thread/whatever. It shares it's address space
1100 * with proc0 - ie: kernel only.
1101 *
365fa13f
MD
1102 * NOTE! By default new threads are created with the MP lock held. A
1103 * thread which does not require the MP lock should release it by calling
1104 * rel_mplock() at the start of the new thread.
99df837e
MD
1105 */
1106int
1107lwkt_create(void (*func)(void *), void *arg,
75cdbe6c 1108 struct thread **tdp, thread_t template, int tdflags, int cpu,
ef0fdad1 1109 const char *fmt, ...)
99df837e 1110{
73e4f7b9 1111 thread_t td;
e2565a42 1112 __va_list ap;
99df837e 1113
75cdbe6c 1114 td = lwkt_alloc_thread(template, cpu);
a2a5ad0d
MD
1115 if (tdp)
1116 *tdp = td;
709799ea 1117 cpu_set_thread_handler(td, lwkt_exit, func, arg);
ef0fdad1 1118 td->td_flags |= TDF_VERBOSE | tdflags;
8a8d5d85
MD
1119#ifdef SMP
1120 td->td_mpcount = 1;
1121#endif
99df837e
MD
1122
1123 /*
1124 * Set up arg0 for 'ps' etc
1125 */
e2565a42 1126 __va_start(ap, fmt);
99df837e 1127 vsnprintf(td->td_comm, sizeof(td->td_comm), fmt, ap);
e2565a42 1128 __va_end(ap);
99df837e
MD
1129
1130 /*
1131 * Schedule the thread to run
1132 */
ef0fdad1
MD
1133 if ((td->td_flags & TDF_STOPREQ) == 0)
1134 lwkt_schedule(td);
1135 else
1136 td->td_flags &= ~TDF_STOPREQ;
99df837e
MD
1137 return 0;
1138}
1139
2d93b37a 1140/*
2d93b37a
MD
1141 * kthread_* is specific to the kernel and is not needed by userland.
1142 */
1143#ifdef _KERNEL
1144
99df837e
MD
1145/*
1146 * Destroy an LWKT thread. Warning! This function is not called when
1147 * a process exits, cpu_proc_exit() directly calls cpu_thread_exit() and
1148 * uses a different reaping mechanism.
1149 */
1150void
1151lwkt_exit(void)
1152{
1153 thread_t td = curthread;
1154
1155 if (td->td_flags & TDF_VERBOSE)
1156 printf("kthread %p %s has exited\n", td, td->td_comm);
f6bf3af1 1157 caps_exit(td);
99df837e
MD
1158 crit_enter();
1159 lwkt_deschedule_self();
1160 ++mycpu->gd_tdfreecount;
1161 TAILQ_INSERT_TAIL(&mycpu->gd_tdfreeq, td, td_threadq);
1162 cpu_thread_exit();
1163}
1164
1165/*
1166 * Create a kernel process/thread/whatever. It shares it's address space
ef0fdad1 1167 * with proc0 - ie: kernel only. 5.x compatible.
365fa13f
MD
1168 *
1169 * NOTE! By default kthreads are created with the MP lock held. A
1170 * thread which does not require the MP lock should release it by calling
1171 * rel_mplock() at the start of the new thread.
99df837e
MD
1172 */
1173int
1174kthread_create(void (*func)(void *), void *arg,
1175 struct thread **tdp, const char *fmt, ...)
1176{
73e4f7b9 1177 thread_t td;
e2565a42 1178 __va_list ap;
99df837e 1179
75cdbe6c 1180 td = lwkt_alloc_thread(NULL, -1);
a2a5ad0d
MD
1181 if (tdp)
1182 *tdp = td;
99df837e
MD
1183 cpu_set_thread_handler(td, kthread_exit, func, arg);
1184 td->td_flags |= TDF_VERBOSE;
8a8d5d85
MD
1185#ifdef SMP
1186 td->td_mpcount = 1;
1187#endif
99df837e
MD
1188
1189 /*
1190 * Set up arg0 for 'ps' etc
1191 */
e2565a42 1192 __va_start(ap, fmt);
99df837e 1193 vsnprintf(td->td_comm, sizeof(td->td_comm), fmt, ap);
e2565a42 1194 __va_end(ap);
99df837e
MD
1195
1196 /*
1197 * Schedule the thread to run
1198 */
1199 lwkt_schedule(td);
1200 return 0;
1201}
1202
1203/*
1204 * Destroy an LWKT thread. Warning! This function is not called when
1205 * a process exits, cpu_proc_exit() directly calls cpu_thread_exit() and
1206 * uses a different reaping mechanism.
1207 *
1208 * XXX duplicates lwkt_exit()
1209 */
1210void
1211kthread_exit(void)
1212{
1213 lwkt_exit();
1214}
1215
2d93b37a
MD
1216#endif /* _KERNEL */
1217
1218void
1219crit_panic(void)
1220{
1221 thread_t td = curthread;
1222 int lpri = td->td_pri;
1223
1224 td->td_pri = 0;
1225 panic("td_pri is/would-go negative! %p %d", td, lpri);
1226}
1227