xref: /freebsd/sys/kern/kern_thread.c (revision 7d1469e555bdce32b3dfc898478ae5564d5072b1)
19454b2d8SWarner Losh /*-
24d846d26SWarner Losh  * SPDX-License-Identifier: BSD-2-Clause
38a36da99SPedro F. Giffuni  *
444990b8cSJulian Elischer  * Copyright (C) 2001 Julian Elischer <julian@freebsd.org>.
544990b8cSJulian Elischer  *  All rights reserved.
644990b8cSJulian Elischer  *
744990b8cSJulian Elischer  * Redistribution and use in source and binary forms, with or without
844990b8cSJulian Elischer  * modification, are permitted provided that the following conditions
944990b8cSJulian Elischer  * are met:
1044990b8cSJulian Elischer  * 1. Redistributions of source code must retain the above copyright
1144990b8cSJulian Elischer  *    notice(s), this list of conditions and the following disclaimer as
1244990b8cSJulian Elischer  *    the first lines of this file unmodified other than the possible
1344990b8cSJulian Elischer  *    addition of one or more copyright notices.
1444990b8cSJulian Elischer  * 2. Redistributions in binary form must reproduce the above copyright
1544990b8cSJulian Elischer  *    notice(s), this list of conditions and the following disclaimer in the
1644990b8cSJulian Elischer  *    documentation and/or other materials provided with the distribution.
1744990b8cSJulian Elischer  *
1844990b8cSJulian Elischer  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
1944990b8cSJulian Elischer  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
2044990b8cSJulian Elischer  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
2144990b8cSJulian Elischer  * DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY
2244990b8cSJulian Elischer  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
2344990b8cSJulian Elischer  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
2444990b8cSJulian Elischer  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
2544990b8cSJulian Elischer  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
2644990b8cSJulian Elischer  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
2744990b8cSJulian Elischer  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
2844990b8cSJulian Elischer  * DAMAGE.
2944990b8cSJulian Elischer  */
3044990b8cSJulian Elischer 
313d06b4b3SAttilio Rao #include "opt_witness.h"
3216d95d4fSJoseph Koshy #include "opt_hwpmc_hooks.h"
333d06b4b3SAttilio Rao 
34677b542eSDavid E. O'Brien #include <sys/cdefs.h>
3544990b8cSJulian Elischer #include <sys/param.h>
3644990b8cSJulian Elischer #include <sys/systm.h>
3744990b8cSJulian Elischer #include <sys/kernel.h>
3844990b8cSJulian Elischer #include <sys/lock.h>
395dda15adSMark Johnston #include <sys/msan.h>
4044990b8cSJulian Elischer #include <sys/mutex.h>
4144990b8cSJulian Elischer #include <sys/proc.h>
4235bb59edSMateusz Guzik #include <sys/bitstring.h>
436febf180SGleb Smirnoff #include <sys/epoch.h>
448f0e9130SKonstantin Belousov #include <sys/rangelock.h>
45e170bfdaSDavid Xu #include <sys/resourcevar.h>
46b3e9e682SRyan Stone #include <sys/sdt.h>
4794e0a4cdSJulian Elischer #include <sys/smp.h>
48de028f5aSJeff Roberson #include <sys/sched.h>
4944f3b092SJohn Baldwin #include <sys/sleepqueue.h>
50ace8398dSJeff Roberson #include <sys/selinfo.h>
51d1e7a4a5SJohn Baldwin #include <sys/syscallsubr.h>
52598f2b81SMateusz Guzik #include <sys/dtrace_bsd.h>
5391d1786fSDmitry Chagin #include <sys/sysent.h>
54961a7b24SJohn Baldwin #include <sys/turnstile.h>
55d116b9f1SMateusz Guzik #include <sys/taskqueue.h>
5644990b8cSJulian Elischer #include <sys/ktr.h>
57cf7d9a8cSDavid Xu #include <sys/rwlock.h>
58af29f399SDmitry Chagin #include <sys/umtxvar.h>
599ed01c32SGleb Smirnoff #include <sys/vmmeter.h>
60d7f687fcSJeff Roberson #include <sys/cpuset.h>
6116d95d4fSJoseph Koshy #ifdef	HWPMC_HOOKS
6216d95d4fSJoseph Koshy #include <sys/pmckern.h>
6316d95d4fSJoseph Koshy #endif
641bd3cf5dSMateusz Guzik #include <sys/priv.h>
6544990b8cSJulian Elischer 
66911b84b0SRobert Watson #include <security/audit/audit.h>
67911b84b0SRobert Watson 
68d116b9f1SMateusz Guzik #include <vm/pmap.h>
6944990b8cSJulian Elischer #include <vm/vm.h>
7049a2507bSAlan Cox #include <vm/vm_extern.h>
7144990b8cSJulian Elischer #include <vm/uma.h>
72d116b9f1SMateusz Guzik #include <vm/vm_phys.h>
73b209f889SRandall Stewart #include <sys/eventhandler.h>
7402fb42b0SPeter Wemm 
75acd9f517SKonstantin Belousov /*
76acd9f517SKonstantin Belousov  * Asserts below verify the stability of struct thread and struct proc
77acd9f517SKonstantin Belousov  * layout, as exposed by KBI to modules.  On head, the KBI is allowed
78acd9f517SKonstantin Belousov  * to drift, change to the structures must be accompanied by the
79acd9f517SKonstantin Belousov  * assert update.
80acd9f517SKonstantin Belousov  *
81acd9f517SKonstantin Belousov  * On the stable branches after KBI freeze, conditions must not be
82acd9f517SKonstantin Belousov  * violated.  Typically new fields are moved to the end of the
83acd9f517SKonstantin Belousov  * structures.
84acd9f517SKonstantin Belousov  */
85acd9f517SKonstantin Belousov #ifdef __amd64__
86a422084aSMark Johnston _Static_assert(offsetof(struct thread, td_flags) == 0x108,
87acd9f517SKonstantin Belousov     "struct thread KBI td_flags");
88c6d31b83SKonstantin Belousov _Static_assert(offsetof(struct thread, td_pflags) == 0x114,
89acd9f517SKonstantin Belousov     "struct thread KBI td_pflags");
907530de77SMateusz Guzik _Static_assert(offsetof(struct thread, td_frame) == 0x4b8,
91acd9f517SKonstantin Belousov     "struct thread KBI td_frame");
92c6d31b83SKonstantin Belousov _Static_assert(offsetof(struct thread, td_emuldata) == 0x6c0,
93acd9f517SKonstantin Belousov     "struct thread KBI td_emuldata");
9485078b85SConrad Meyer _Static_assert(offsetof(struct proc, p_flag) == 0xb8,
95acd9f517SKonstantin Belousov     "struct proc KBI p_flag");
9685078b85SConrad Meyer _Static_assert(offsetof(struct proc, p_pid) == 0xc4,
97acd9f517SKonstantin Belousov     "struct proc KBI p_pid");
98351d5f7fSKonstantin Belousov _Static_assert(offsetof(struct proc, p_filemon) == 0x3c8,
99acd9f517SKonstantin Belousov     "struct proc KBI p_filemon");
100aaa92413SKonstantin Belousov _Static_assert(offsetof(struct proc, p_comm) == 0x3e0,
101acd9f517SKonstantin Belousov     "struct proc KBI p_comm");
10293ca6ff2SKonstantin Belousov _Static_assert(offsetof(struct proc, p_emuldata) == 0x4d0,
103acd9f517SKonstantin Belousov     "struct proc KBI p_emuldata");
104acd9f517SKonstantin Belousov #endif
105acd9f517SKonstantin Belousov #ifdef __i386__
106a422084aSMark Johnston _Static_assert(offsetof(struct thread, td_flags) == 0x9c,
107acd9f517SKonstantin Belousov     "struct thread KBI td_flags");
108c6d31b83SKonstantin Belousov _Static_assert(offsetof(struct thread, td_pflags) == 0xa8,
109acd9f517SKonstantin Belousov     "struct thread KBI td_pflags");
1107530de77SMateusz Guzik _Static_assert(offsetof(struct thread, td_frame) == 0x318,
111acd9f517SKonstantin Belousov     "struct thread KBI td_frame");
1127530de77SMateusz Guzik _Static_assert(offsetof(struct thread, td_emuldata) == 0x35c,
113acd9f517SKonstantin Belousov     "struct thread KBI td_emuldata");
11485078b85SConrad Meyer _Static_assert(offsetof(struct proc, p_flag) == 0x6c,
115acd9f517SKonstantin Belousov     "struct proc KBI p_flag");
11685078b85SConrad Meyer _Static_assert(offsetof(struct proc, p_pid) == 0x78,
117acd9f517SKonstantin Belousov     "struct proc KBI p_pid");
1184d675b80SKonstantin Belousov _Static_assert(offsetof(struct proc, p_filemon) == 0x270,
119acd9f517SKonstantin Belousov     "struct proc KBI p_filemon");
120aaa92413SKonstantin Belousov _Static_assert(offsetof(struct proc, p_comm) == 0x284,
121acd9f517SKonstantin Belousov     "struct proc KBI p_comm");
122aaa92413SKonstantin Belousov _Static_assert(offsetof(struct proc, p_emuldata) == 0x318,
123acd9f517SKonstantin Belousov     "struct proc KBI p_emuldata");
124acd9f517SKonstantin Belousov #endif
125acd9f517SKonstantin Belousov 
126b3e9e682SRyan Stone SDT_PROVIDER_DECLARE(proc);
127d9fae5abSAndriy Gapon SDT_PROBE_DEFINE(proc, , , lwp__exit);
128b3e9e682SRyan Stone 
1298460a577SJohn Birrell /*
1308460a577SJohn Birrell  * thread related storage.
1318460a577SJohn Birrell  */
13244990b8cSJulian Elischer static uma_zone_t thread_zone;
13344990b8cSJulian Elischer 
134d116b9f1SMateusz Guzik struct thread_domain_data {
135d116b9f1SMateusz Guzik 	struct thread	*tdd_zombies;
136d116b9f1SMateusz Guzik 	int		tdd_reapticks;
137d116b9f1SMateusz Guzik } __aligned(CACHE_LINE_SIZE);
138d116b9f1SMateusz Guzik 
139d116b9f1SMateusz Guzik static struct thread_domain_data thread_domain_data[MAXMEMDOM];
140d116b9f1SMateusz Guzik 
141d116b9f1SMateusz Guzik static struct task	thread_reap_task;
142d116b9f1SMateusz Guzik static struct callout  	thread_reap_callout;
14344990b8cSJulian Elischer 
144ff8fbcffSJeff Roberson static void thread_zombie(struct thread *);
145b83e94beSMateusz Guzik static void thread_reap(void);
146d116b9f1SMateusz Guzik static void thread_reap_all(void);
147d116b9f1SMateusz Guzik static void thread_reap_task_cb(void *, int);
148d116b9f1SMateusz Guzik static void thread_reap_callout_cb(void *);
14984cdea97SKonstantin Belousov static int thread_unsuspend_one(struct thread *td, struct proc *p,
15084cdea97SKonstantin Belousov     bool boundary);
151755341dfSMateusz Guzik static void thread_free_batched(struct thread *td);
152ff8fbcffSJeff Roberson 
153d1ca25beSMateusz Guzik static __exclusive_cache_line struct mtx tid_lock;
154934e7e5eSMateusz Guzik static bitstr_t *tid_bitmap;
15535bb59edSMateusz Guzik 
156cf7d9a8cSDavid Xu static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
157cf7d9a8cSDavid Xu 
1581bd3cf5dSMateusz Guzik static int maxthread;
1591bd3cf5dSMateusz Guzik SYSCTL_INT(_kern, OID_AUTO, maxthread, CTLFLAG_RDTUN,
1601bd3cf5dSMateusz Guzik     &maxthread, 0, "Maximum number of threads");
1611bd3cf5dSMateusz Guzik 
16262dbc992SMateusz Guzik static __exclusive_cache_line int nthreads;
1631bd3cf5dSMateusz Guzik 
164aae3547bSMateusz Guzik static LIST_HEAD(tidhashhead, thread) *tidhashtbl;
165aae3547bSMateusz Guzik static u_long	tidhash;
16626007fe3SMateusz Guzik static u_long	tidhashlock;
16726007fe3SMateusz Guzik static struct	rwlock *tidhashtbl_lock;
168aae3547bSMateusz Guzik #define	TIDHASH(tid)		(&tidhashtbl[(tid) & tidhash])
16926007fe3SMateusz Guzik #define	TIDHASHLOCK(tid)	(&tidhashtbl_lock[(tid) & tidhashlock])
170cf7d9a8cSDavid Xu 
1712ca45184SMatt Joras EVENTHANDLER_LIST_DEFINE(thread_ctor);
1722ca45184SMatt Joras EVENTHANDLER_LIST_DEFINE(thread_dtor);
1732ca45184SMatt Joras EVENTHANDLER_LIST_DEFINE(thread_init);
1742ca45184SMatt Joras EVENTHANDLER_LIST_DEFINE(thread_fini);
1752ca45184SMatt Joras 
17662dbc992SMateusz Guzik static bool
177d116b9f1SMateusz Guzik thread_count_inc_try(void)
178ec6ea5e8SDavid Xu {
17962dbc992SMateusz Guzik 	int nthreads_new;
180ec6ea5e8SDavid Xu 
18162dbc992SMateusz Guzik 	nthreads_new = atomic_fetchadd_int(&nthreads, 1) + 1;
18262dbc992SMateusz Guzik 	if (nthreads_new >= maxthread - 100) {
1831bd3cf5dSMateusz Guzik 		if (priv_check_cred(curthread->td_ucred, PRIV_MAXPROC) != 0 ||
18462dbc992SMateusz Guzik 		    nthreads_new >= maxthread) {
18562dbc992SMateusz Guzik 			atomic_subtract_int(&nthreads, 1);
186d116b9f1SMateusz Guzik 			return (false);
187d116b9f1SMateusz Guzik 		}
188d116b9f1SMateusz Guzik 	}
189d116b9f1SMateusz Guzik 	return (true);
190d116b9f1SMateusz Guzik }
191d116b9f1SMateusz Guzik 
192d116b9f1SMateusz Guzik static bool
193d116b9f1SMateusz Guzik thread_count_inc(void)
194d116b9f1SMateusz Guzik {
195d116b9f1SMateusz Guzik 	static struct timeval lastfail;
196d116b9f1SMateusz Guzik 	static int curfail;
197d116b9f1SMateusz Guzik 
198d116b9f1SMateusz Guzik 	thread_reap();
199d116b9f1SMateusz Guzik 	if (thread_count_inc_try()) {
200d116b9f1SMateusz Guzik 		return (true);
201d116b9f1SMateusz Guzik 	}
202d116b9f1SMateusz Guzik 
203d116b9f1SMateusz Guzik 	thread_reap_all();
204d116b9f1SMateusz Guzik 	if (thread_count_inc_try()) {
205d116b9f1SMateusz Guzik 		return (true);
206d116b9f1SMateusz Guzik 	}
207d116b9f1SMateusz Guzik 
2081bd3cf5dSMateusz Guzik 	if (ppsratecheck(&lastfail, &curfail, 1)) {
2091bd3cf5dSMateusz Guzik 		printf("maxthread limit exceeded by uid %u "
2101bd3cf5dSMateusz Guzik 		    "(pid %d); consider increasing kern.maxthread\n",
2111bd3cf5dSMateusz Guzik 		    curthread->td_ucred->cr_ruid, curproc->p_pid);
2121bd3cf5dSMateusz Guzik 	}
21362dbc992SMateusz Guzik 	return (false);
2141bd3cf5dSMateusz Guzik }
2151bd3cf5dSMateusz Guzik 
21662dbc992SMateusz Guzik static void
21762dbc992SMateusz Guzik thread_count_sub(int n)
21862dbc992SMateusz Guzik {
21962dbc992SMateusz Guzik 
22062dbc992SMateusz Guzik 	atomic_subtract_int(&nthreads, n);
22162dbc992SMateusz Guzik }
22262dbc992SMateusz Guzik 
22362dbc992SMateusz Guzik static void
22462dbc992SMateusz Guzik thread_count_dec(void)
22562dbc992SMateusz Guzik {
22662dbc992SMateusz Guzik 
22762dbc992SMateusz Guzik 	thread_count_sub(1);
22862dbc992SMateusz Guzik }
22962dbc992SMateusz Guzik 
23062dbc992SMateusz Guzik static lwpid_t
23162dbc992SMateusz Guzik tid_alloc(void)
23262dbc992SMateusz Guzik {
23362dbc992SMateusz Guzik 	static lwpid_t trytid;
23462dbc992SMateusz Guzik 	lwpid_t tid;
23562dbc992SMateusz Guzik 
23662dbc992SMateusz Guzik 	mtx_lock(&tid_lock);
23735bb59edSMateusz Guzik 	/*
23835bb59edSMateusz Guzik 	 * It is an invariant that the bitmap is big enough to hold maxthread
23935bb59edSMateusz Guzik 	 * IDs. If we got to this point there has to be at least one free.
24035bb59edSMateusz Guzik 	 */
24135bb59edSMateusz Guzik 	if (trytid >= maxthread)
24235bb59edSMateusz Guzik 		trytid = 0;
24335bb59edSMateusz Guzik 	bit_ffc_at(tid_bitmap, trytid, maxthread, &tid);
24435bb59edSMateusz Guzik 	if (tid == -1) {
24535bb59edSMateusz Guzik 		KASSERT(trytid != 0, ("unexpectedly ran out of IDs"));
24635bb59edSMateusz Guzik 		trytid = 0;
24735bb59edSMateusz Guzik 		bit_ffc_at(tid_bitmap, trytid, maxthread, &tid);
24835bb59edSMateusz Guzik 		KASSERT(tid != -1, ("unexpectedly ran out of IDs"));
249ec6ea5e8SDavid Xu 	}
25035bb59edSMateusz Guzik 	bit_set(tid_bitmap, tid);
251934e7e5eSMateusz Guzik 	trytid = tid + 1;
252ec6ea5e8SDavid Xu 	mtx_unlock(&tid_lock);
25335bb59edSMateusz Guzik 	return (tid + NO_PID);
254ec6ea5e8SDavid Xu }
255ec6ea5e8SDavid Xu 
256ec6ea5e8SDavid Xu static void
257755341dfSMateusz Guzik tid_free_locked(lwpid_t rtid)
258ec6ea5e8SDavid Xu {
25935bb59edSMateusz Guzik 	lwpid_t tid;
260ec6ea5e8SDavid Xu 
261755341dfSMateusz Guzik 	mtx_assert(&tid_lock, MA_OWNED);
26235bb59edSMateusz Guzik 	KASSERT(rtid >= NO_PID,
26335bb59edSMateusz Guzik 	    ("%s: invalid tid %d\n", __func__, rtid));
26435bb59edSMateusz Guzik 	tid = rtid - NO_PID;
26535bb59edSMateusz Guzik 	KASSERT(bit_test(tid_bitmap, tid) != 0,
26635bb59edSMateusz Guzik 	    ("thread ID %d not allocated\n", rtid));
26735bb59edSMateusz Guzik 	bit_clear(tid_bitmap, tid);
268755341dfSMateusz Guzik }
269755341dfSMateusz Guzik 
270755341dfSMateusz Guzik static void
271755341dfSMateusz Guzik tid_free(lwpid_t rtid)
272755341dfSMateusz Guzik {
273755341dfSMateusz Guzik 
274755341dfSMateusz Guzik 	mtx_lock(&tid_lock);
275755341dfSMateusz Guzik 	tid_free_locked(rtid);
276755341dfSMateusz Guzik 	mtx_unlock(&tid_lock);
277755341dfSMateusz Guzik }
278755341dfSMateusz Guzik 
279755341dfSMateusz Guzik static void
280755341dfSMateusz Guzik tid_free_batch(lwpid_t *batch, int n)
281755341dfSMateusz Guzik {
282755341dfSMateusz Guzik 	int i;
283755341dfSMateusz Guzik 
284755341dfSMateusz Guzik 	mtx_lock(&tid_lock);
285755341dfSMateusz Guzik 	for (i = 0; i < n; i++) {
286755341dfSMateusz Guzik 		tid_free_locked(batch[i]);
287755341dfSMateusz Guzik 	}
288ec6ea5e8SDavid Xu 	mtx_unlock(&tid_lock);
289ec6ea5e8SDavid Xu }
290ec6ea5e8SDavid Xu 
291fdcac928SMarcel Moolenaar /*
2925ef7b7a0SMateusz Guzik  * Batching for thread reapping.
2935ef7b7a0SMateusz Guzik  */
2945ef7b7a0SMateusz Guzik struct tidbatch {
2955ef7b7a0SMateusz Guzik 	lwpid_t tab[16];
2965ef7b7a0SMateusz Guzik 	int n;
2975ef7b7a0SMateusz Guzik };
2985ef7b7a0SMateusz Guzik 
2995ef7b7a0SMateusz Guzik static void
3005ef7b7a0SMateusz Guzik tidbatch_prep(struct tidbatch *tb)
3015ef7b7a0SMateusz Guzik {
3025ef7b7a0SMateusz Guzik 
3035ef7b7a0SMateusz Guzik 	tb->n = 0;
3045ef7b7a0SMateusz Guzik }
3055ef7b7a0SMateusz Guzik 
3065ef7b7a0SMateusz Guzik static void
3075ef7b7a0SMateusz Guzik tidbatch_add(struct tidbatch *tb, struct thread *td)
3085ef7b7a0SMateusz Guzik {
3095ef7b7a0SMateusz Guzik 
3105ef7b7a0SMateusz Guzik 	KASSERT(tb->n < nitems(tb->tab),
3115ef7b7a0SMateusz Guzik 	    ("%s: count too high %d", __func__, tb->n));
3125ef7b7a0SMateusz Guzik 	tb->tab[tb->n] = td->td_tid;
3135ef7b7a0SMateusz Guzik 	tb->n++;
3145ef7b7a0SMateusz Guzik }
3155ef7b7a0SMateusz Guzik 
3165ef7b7a0SMateusz Guzik static void
3175ef7b7a0SMateusz Guzik tidbatch_process(struct tidbatch *tb)
3185ef7b7a0SMateusz Guzik {
3195ef7b7a0SMateusz Guzik 
3205ef7b7a0SMateusz Guzik 	KASSERT(tb->n <= nitems(tb->tab),
3215ef7b7a0SMateusz Guzik 	    ("%s: count too high %d", __func__, tb->n));
3225ef7b7a0SMateusz Guzik 	if (tb->n == nitems(tb->tab)) {
3235ef7b7a0SMateusz Guzik 		tid_free_batch(tb->tab, tb->n);
3245ef7b7a0SMateusz Guzik 		tb->n = 0;
3255ef7b7a0SMateusz Guzik 	}
3265ef7b7a0SMateusz Guzik }
3275ef7b7a0SMateusz Guzik 
3285ef7b7a0SMateusz Guzik static void
3295ef7b7a0SMateusz Guzik tidbatch_final(struct tidbatch *tb)
3305ef7b7a0SMateusz Guzik {
3315ef7b7a0SMateusz Guzik 
3325ef7b7a0SMateusz Guzik 	KASSERT(tb->n <= nitems(tb->tab),
3335ef7b7a0SMateusz Guzik 	    ("%s: count too high %d", __func__, tb->n));
3345ef7b7a0SMateusz Guzik 	if (tb->n != 0) {
3355ef7b7a0SMateusz Guzik 		tid_free_batch(tb->tab, tb->n);
3365ef7b7a0SMateusz Guzik 	}
3375ef7b7a0SMateusz Guzik }
3385ef7b7a0SMateusz Guzik 
3395ef7b7a0SMateusz Guzik /*
340e0c86f5cSMateusz Guzik  * Batching thread count free, for consistency
341e0c86f5cSMateusz Guzik  */
342e0c86f5cSMateusz Guzik struct tdcountbatch {
343e0c86f5cSMateusz Guzik 	int n;
344e0c86f5cSMateusz Guzik };
345e0c86f5cSMateusz Guzik 
346e0c86f5cSMateusz Guzik static void
347e0c86f5cSMateusz Guzik tdcountbatch_prep(struct tdcountbatch *tb)
348e0c86f5cSMateusz Guzik {
349e0c86f5cSMateusz Guzik 
350e0c86f5cSMateusz Guzik 	tb->n = 0;
351e0c86f5cSMateusz Guzik }
352e0c86f5cSMateusz Guzik 
353e0c86f5cSMateusz Guzik static void
354e0c86f5cSMateusz Guzik tdcountbatch_add(struct tdcountbatch *tb, struct thread *td __unused)
355e0c86f5cSMateusz Guzik {
356e0c86f5cSMateusz Guzik 
357e0c86f5cSMateusz Guzik 	tb->n++;
358e0c86f5cSMateusz Guzik }
359e0c86f5cSMateusz Guzik 
360e0c86f5cSMateusz Guzik static void
361e0c86f5cSMateusz Guzik tdcountbatch_process(struct tdcountbatch *tb)
362e0c86f5cSMateusz Guzik {
363e0c86f5cSMateusz Guzik 
364e0c86f5cSMateusz Guzik 	if (tb->n == 32) {
365e0c86f5cSMateusz Guzik 		thread_count_sub(tb->n);
366e0c86f5cSMateusz Guzik 		tb->n = 0;
367e0c86f5cSMateusz Guzik 	}
368e0c86f5cSMateusz Guzik }
369e0c86f5cSMateusz Guzik 
370e0c86f5cSMateusz Guzik static void
371e0c86f5cSMateusz Guzik tdcountbatch_final(struct tdcountbatch *tb)
372e0c86f5cSMateusz Guzik {
373e0c86f5cSMateusz Guzik 
374e0c86f5cSMateusz Guzik 	if (tb->n != 0) {
375e0c86f5cSMateusz Guzik 		thread_count_sub(tb->n);
376e0c86f5cSMateusz Guzik 	}
377e0c86f5cSMateusz Guzik }
378e0c86f5cSMateusz Guzik 
379e0c86f5cSMateusz Guzik /*
380696058c3SJulian Elischer  * Prepare a thread for use.
38144990b8cSJulian Elischer  */
382b23f72e9SBrian Feldman static int
383b23f72e9SBrian Feldman thread_ctor(void *mem, int size, void *arg, int flags)
38444990b8cSJulian Elischer {
38544990b8cSJulian Elischer 	struct thread	*td;
38644990b8cSJulian Elischer 
38744990b8cSJulian Elischer 	td = (struct thread *)mem;
388fa2528acSAlex Richardson 	TD_SET_STATE(td, TDS_INACTIVE);
38994dd54b9SKonstantin Belousov 	td->td_lastcpu = td->td_oncpu = NOCPU;
3906c27c603SJuli Mallett 
3916c27c603SJuli Mallett 	/*
3926c27c603SJuli Mallett 	 * Note that td_critnest begins life as 1 because the thread is not
3936c27c603SJuli Mallett 	 * running and is thereby implicitly waiting to be on the receiving
394a54e85fdSJeff Roberson 	 * end of a context switch.
3956c27c603SJuli Mallett 	 */
396139b7550SJohn Baldwin 	td->td_critnest = 1;
397acbe332aSDavid Xu 	td->td_lend_user_pri = PRI_MAX;
398911b84b0SRobert Watson #ifdef AUDIT
399911b84b0SRobert Watson 	audit_thread_alloc(td);
400911b84b0SRobert Watson #endif
401598f2b81SMateusz Guzik #ifdef KDTRACE_HOOKS
402598f2b81SMateusz Guzik 	kdtrace_thread_ctor(td);
403598f2b81SMateusz Guzik #endif
404d10183d9SDavid Xu 	umtx_thread_alloc(td);
40519d3e47dSMateusz Guzik 	MPASS(td->td_sel == NULL);
406b23f72e9SBrian Feldman 	return (0);
40744990b8cSJulian Elischer }
40844990b8cSJulian Elischer 
40944990b8cSJulian Elischer /*
41044990b8cSJulian Elischer  * Reclaim a thread after use.
41144990b8cSJulian Elischer  */
41244990b8cSJulian Elischer static void
41344990b8cSJulian Elischer thread_dtor(void *mem, int size, void *arg)
41444990b8cSJulian Elischer {
41544990b8cSJulian Elischer 	struct thread *td;
41644990b8cSJulian Elischer 
41744990b8cSJulian Elischer 	td = (struct thread *)mem;
41844990b8cSJulian Elischer 
41944990b8cSJulian Elischer #ifdef INVARIANTS
42044990b8cSJulian Elischer 	/* Verify that this thread is in a safe state to free. */
421fa2528acSAlex Richardson 	switch (TD_GET_STATE(td)) {
42271fad9fdSJulian Elischer 	case TDS_INHIBITED:
42371fad9fdSJulian Elischer 	case TDS_RUNNING:
42471fad9fdSJulian Elischer 	case TDS_CAN_RUN:
42544990b8cSJulian Elischer 	case TDS_RUNQ:
42644990b8cSJulian Elischer 		/*
42744990b8cSJulian Elischer 		 * We must never unlink a thread that is in one of
42844990b8cSJulian Elischer 		 * these states, because it is currently active.
42944990b8cSJulian Elischer 		 */
43044990b8cSJulian Elischer 		panic("bad state for thread unlinking");
43144990b8cSJulian Elischer 		/* NOTREACHED */
43271fad9fdSJulian Elischer 	case TDS_INACTIVE:
43344990b8cSJulian Elischer 		break;
43444990b8cSJulian Elischer 	default:
43544990b8cSJulian Elischer 		panic("bad thread state");
43644990b8cSJulian Elischer 		/* NOTREACHED */
43744990b8cSJulian Elischer 	}
43844990b8cSJulian Elischer #endif
4396e8525ceSRobert Watson #ifdef AUDIT
4406e8525ceSRobert Watson 	audit_thread_free(td);
4416e8525ceSRobert Watson #endif
442598f2b81SMateusz Guzik #ifdef KDTRACE_HOOKS
443598f2b81SMateusz Guzik 	kdtrace_thread_dtor(td);
444598f2b81SMateusz Guzik #endif
4451ba4a712SPawel Jakub Dawidek 	/* Free all OSD associated to this thread. */
4461ba4a712SPawel Jakub Dawidek 	osd_thread_exit(td);
447c6d31b83SKonstantin Belousov 	ast_kclear(td);
44819d3e47dSMateusz Guzik 	seltdfini(td);
44944990b8cSJulian Elischer }
45044990b8cSJulian Elischer 
45144990b8cSJulian Elischer /*
45244990b8cSJulian Elischer  * Initialize type-stable parts of a thread (when newly created).
45344990b8cSJulian Elischer  */
454b23f72e9SBrian Feldman static int
455b23f72e9SBrian Feldman thread_init(void *mem, int size, int flags)
45644990b8cSJulian Elischer {
45744990b8cSJulian Elischer 	struct thread *td;
45844990b8cSJulian Elischer 
45944990b8cSJulian Elischer 	td = (struct thread *)mem;
460247aba24SMarcel Moolenaar 
461b83e94beSMateusz Guzik 	td->td_allocdomain = vm_phys_domain(vtophys(td));
46244f3b092SJohn Baldwin 	td->td_sleepqueue = sleepq_alloc();
463961a7b24SJohn Baldwin 	td->td_turnstile = turnstile_alloc();
4648f0e9130SKonstantin Belousov 	td->td_rlqe = NULL;
4652ca45184SMatt Joras 	EVENTHANDLER_DIRECT_INVOKE(thread_init, td);
466d10183d9SDavid Xu 	umtx_thread_init(td);
46789b57fcfSKonstantin Belousov 	td->td_kstack = 0;
468ad8b1d85SKonstantin Belousov 	td->td_sel = NULL;
469b23f72e9SBrian Feldman 	return (0);
47044990b8cSJulian Elischer }
47144990b8cSJulian Elischer 
47244990b8cSJulian Elischer /*
47344990b8cSJulian Elischer  * Tear down type-stable parts of a thread (just before being discarded).
47444990b8cSJulian Elischer  */
47544990b8cSJulian Elischer static void
47644990b8cSJulian Elischer thread_fini(void *mem, int size)
47744990b8cSJulian Elischer {
47844990b8cSJulian Elischer 	struct thread *td;
47944990b8cSJulian Elischer 
48044990b8cSJulian Elischer 	td = (struct thread *)mem;
4812ca45184SMatt Joras 	EVENTHANDLER_DIRECT_INVOKE(thread_fini, td);
4828f0e9130SKonstantin Belousov 	rlqentry_free(td->td_rlqe);
483961a7b24SJohn Baldwin 	turnstile_free(td->td_turnstile);
48444f3b092SJohn Baldwin 	sleepq_free(td->td_sleepqueue);
485d10183d9SDavid Xu 	umtx_thread_fini(td);
48619d3e47dSMateusz Guzik 	MPASS(td->td_sel == NULL);
48744990b8cSJulian Elischer }
4885215b187SJeff Roberson 
4895c8329edSJulian Elischer /*
4905215b187SJeff Roberson  * For a newly created process,
4915215b187SJeff Roberson  * link up all the structures and its initial threads etc.
492ed062c8dSJulian Elischer  * called from:
493e7d939bdSMarcel Moolenaar  * {arch}/{arch}/machdep.c   {arch}_init(), init386() etc.
494ed062c8dSJulian Elischer  * proc_dtor() (should go away)
495ed062c8dSJulian Elischer  * proc_init()
4965c8329edSJulian Elischer  */
4975c8329edSJulian Elischer void
49889b57fcfSKonstantin Belousov proc_linkup0(struct proc *p, struct thread *td)
49989b57fcfSKonstantin Belousov {
50089b57fcfSKonstantin Belousov 	TAILQ_INIT(&p->p_threads);	     /* all threads in proc */
50189b57fcfSKonstantin Belousov 	proc_linkup(p, td);
50289b57fcfSKonstantin Belousov }
50389b57fcfSKonstantin Belousov 
50489b57fcfSKonstantin Belousov void
5058460a577SJohn Birrell proc_linkup(struct proc *p, struct thread *td)
5065c8329edSJulian Elischer {
507a54e85fdSJeff Roberson 
5089104847fSDavid Xu 	sigqueue_init(&p->p_sigqueue, p);
509cc29f221SKonstantin Belousov 	p->p_ksi = ksiginfo_alloc(M_WAITOK);
510ebceaf6dSDavid Xu 	if (p->p_ksi != NULL) {
5115c474517SDavid Xu 		/* XXX p_ksi may be null if ksiginfo zone is not ready */
512ebceaf6dSDavid Xu 		p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
513ebceaf6dSDavid Xu 	}
514b2f92ef9SDavid Xu 	LIST_INIT(&p->p_mqnotifier);
5155c8329edSJulian Elischer 	p->p_numthreads = 0;
5168460a577SJohn Birrell 	thread_link(td, p);
5175c8329edSJulian Elischer }
5185c8329edSJulian Elischer 
519c6d31b83SKonstantin Belousov static void
520c6d31b83SKonstantin Belousov ast_suspend(struct thread *td, int tda __unused)
521c6d31b83SKonstantin Belousov {
522c6d31b83SKonstantin Belousov 	struct proc *p;
523c6d31b83SKonstantin Belousov 
524c6d31b83SKonstantin Belousov 	p = td->td_proc;
525c6d31b83SKonstantin Belousov 	/*
526c6d31b83SKonstantin Belousov 	 * We need to check to see if we have to exit or wait due to a
527c6d31b83SKonstantin Belousov 	 * single threading requirement or some other STOP condition.
528c6d31b83SKonstantin Belousov 	 */
529c6d31b83SKonstantin Belousov 	PROC_LOCK(p);
530c6d31b83SKonstantin Belousov 	thread_suspend_check(0);
531c6d31b83SKonstantin Belousov 	PROC_UNLOCK(p);
532c6d31b83SKonstantin Belousov }
533c6d31b83SKonstantin Belousov 
5341bd3cf5dSMateusz Guzik extern int max_threads_per_proc;
5351bd3cf5dSMateusz Guzik 
5365c8329edSJulian Elischer /*
53744990b8cSJulian Elischer  * Initialize global thread allocation resources.
53844990b8cSJulian Elischer  */
53944990b8cSJulian Elischer void
54044990b8cSJulian Elischer threadinit(void)
54144990b8cSJulian Elischer {
54226007fe3SMateusz Guzik 	u_long i;
543cf31cadeSMateusz Guzik 	lwpid_t tid0;
54444990b8cSJulian Elischer 
5451bd3cf5dSMateusz Guzik 	/*
5461bd3cf5dSMateusz Guzik 	 * Place an upper limit on threads which can be allocated.
5471bd3cf5dSMateusz Guzik 	 *
5481bd3cf5dSMateusz Guzik 	 * Note that other factors may make the de facto limit much lower.
5491bd3cf5dSMateusz Guzik 	 *
5501bd3cf5dSMateusz Guzik 	 * Platform limits are somewhat arbitrary but deemed "more than good
5511bd3cf5dSMateusz Guzik 	 * enough" for the foreseable future.
5521bd3cf5dSMateusz Guzik 	 */
5531bd3cf5dSMateusz Guzik 	if (maxthread == 0) {
5541bd3cf5dSMateusz Guzik #ifdef _LP64
5551bd3cf5dSMateusz Guzik 		maxthread = MIN(maxproc * max_threads_per_proc, 1000000);
5561bd3cf5dSMateusz Guzik #else
5571bd3cf5dSMateusz Guzik 		maxthread = MIN(maxproc * max_threads_per_proc, 100000);
5581bd3cf5dSMateusz Guzik #endif
5591bd3cf5dSMateusz Guzik 	}
5601bd3cf5dSMateusz Guzik 
5611ea7a6f8SPoul-Henning Kamp 	mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
56235bb59edSMateusz Guzik 	tid_bitmap = bit_alloc(maxthread, M_TIDHASH, M_WAITOK);
56362dbc992SMateusz Guzik 	/*
56462dbc992SMateusz Guzik 	 * Handle thread0.
56562dbc992SMateusz Guzik 	 */
56662dbc992SMateusz Guzik 	thread_count_inc();
567cf31cadeSMateusz Guzik 	tid0 = tid_alloc();
568cf31cadeSMateusz Guzik 	if (tid0 != THREAD0_TID)
569cf31cadeSMateusz Guzik 		panic("tid0 %d != %d\n", tid0, THREAD0_TID);
5701ea7a6f8SPoul-Henning Kamp 
571*7d1469e5SOlivier Certner 	/*
572*7d1469e5SOlivier Certner 	 * Thread structures are specially aligned so that (at least) the
573*7d1469e5SOlivier Certner 	 * 5 lower bits of a pointer to 'struct thead' must be 0.  These bits
574*7d1469e5SOlivier Certner 	 * are used by synchronization primitives to store flags in pointers to
575*7d1469e5SOlivier Certner 	 * such structures.
576*7d1469e5SOlivier Certner 	 */
577de028f5aSJeff Roberson 	thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
57844990b8cSJulian Elischer 	    thread_ctor, thread_dtor, thread_init, thread_fini,
579*7d1469e5SOlivier Certner 	    UMA_ALIGN_CACHE_AND_MASK(32 - 1), UMA_ZONE_NOFREE);
580cf7d9a8cSDavid Xu 	tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
58126007fe3SMateusz Guzik 	tidhashlock = (tidhash + 1) / 64;
58226007fe3SMateusz Guzik 	if (tidhashlock > 0)
58326007fe3SMateusz Guzik 		tidhashlock--;
58426007fe3SMateusz Guzik 	tidhashtbl_lock = malloc(sizeof(*tidhashtbl_lock) * (tidhashlock + 1),
58526007fe3SMateusz Guzik 	    M_TIDHASH, M_WAITOK | M_ZERO);
58626007fe3SMateusz Guzik 	for (i = 0; i < tidhashlock + 1; i++)
58726007fe3SMateusz Guzik 		rw_init(&tidhashtbl_lock[i], "tidhash");
588d116b9f1SMateusz Guzik 
589d116b9f1SMateusz Guzik 	TASK_INIT(&thread_reap_task, 0, thread_reap_task_cb, NULL);
590d116b9f1SMateusz Guzik 	callout_init(&thread_reap_callout, 1);
591845d7797SKonstantin Belousov 	callout_reset(&thread_reap_callout, 5 * hz,
592845d7797SKonstantin Belousov 	    thread_reap_callout_cb, NULL);
593c6d31b83SKonstantin Belousov 	ast_register(TDA_SUSPEND, ASTR_ASTF_REQUIRED, 0, ast_suspend);
59444990b8cSJulian Elischer }
59544990b8cSJulian Elischer 
59644990b8cSJulian Elischer /*
597ff8fbcffSJeff Roberson  * Place an unused thread on the zombie list.
59844990b8cSJulian Elischer  */
59944990b8cSJulian Elischer void
600ff8fbcffSJeff Roberson thread_zombie(struct thread *td)
60144990b8cSJulian Elischer {
602d116b9f1SMateusz Guzik 	struct thread_domain_data *tdd;
603c5315f51SMateusz Guzik 	struct thread *ztd;
604c5315f51SMateusz Guzik 
605a9568cd2SMateusz Guzik 	tdd = &thread_domain_data[td->td_allocdomain];
606d116b9f1SMateusz Guzik 	ztd = atomic_load_ptr(&tdd->tdd_zombies);
607c5315f51SMateusz Guzik 	for (;;) {
608c5315f51SMateusz Guzik 		td->td_zombie = ztd;
609d116b9f1SMateusz Guzik 		if (atomic_fcmpset_rel_ptr((uintptr_t *)&tdd->tdd_zombies,
610c5315f51SMateusz Guzik 		    (uintptr_t *)&ztd, (uintptr_t)td))
611c5315f51SMateusz Guzik 			break;
612c5315f51SMateusz Guzik 		continue;
613c5315f51SMateusz Guzik 	}
61444990b8cSJulian Elischer }
61544990b8cSJulian Elischer 
6165c8329edSJulian Elischer /*
617ff8fbcffSJeff Roberson  * Release a thread that has exited after cpu_throw().
618ff8fbcffSJeff Roberson  */
619ff8fbcffSJeff Roberson void
620ff8fbcffSJeff Roberson thread_stash(struct thread *td)
621ff8fbcffSJeff Roberson {
622ff8fbcffSJeff Roberson 	atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
623ff8fbcffSJeff Roberson 	thread_zombie(td);
624ff8fbcffSJeff Roberson }
625ff8fbcffSJeff Roberson 
626ff8fbcffSJeff Roberson /*
627d116b9f1SMateusz Guzik  * Reap zombies from passed domain.
62844990b8cSJulian Elischer  */
629d116b9f1SMateusz Guzik static void
630d116b9f1SMateusz Guzik thread_reap_domain(struct thread_domain_data *tdd)
63144990b8cSJulian Elischer {
632c5315f51SMateusz Guzik 	struct thread *itd, *ntd;
6335ef7b7a0SMateusz Guzik 	struct tidbatch tidbatch;
634f34a2f56SMateusz Guzik 	struct credbatch credbatch;
635bbe62559SMateusz Guzik 	struct limbatch limbatch;
636e0c86f5cSMateusz Guzik 	struct tdcountbatch tdcountbatch;
63744990b8cSJulian Elischer 
63844990b8cSJulian Elischer 	/*
639c5315f51SMateusz Guzik 	 * Reading upfront is pessimal if followed by concurrent atomic_swap,
640c5315f51SMateusz Guzik 	 * but most of the time the list is empty.
64144990b8cSJulian Elischer 	 */
642d116b9f1SMateusz Guzik 	if (tdd->tdd_zombies == NULL)
643c5315f51SMateusz Guzik 		return;
644c5315f51SMateusz Guzik 
645d116b9f1SMateusz Guzik 	itd = (struct thread *)atomic_swap_ptr((uintptr_t *)&tdd->tdd_zombies,
646c5315f51SMateusz Guzik 	    (uintptr_t)NULL);
6475ef7b7a0SMateusz Guzik 	if (itd == NULL)
6485ef7b7a0SMateusz Guzik 		return;
6495ef7b7a0SMateusz Guzik 
650d116b9f1SMateusz Guzik 	/*
651d116b9f1SMateusz Guzik 	 * Multiple CPUs can get here, the race is fine as ticks is only
652d116b9f1SMateusz Guzik 	 * advisory.
653d116b9f1SMateusz Guzik 	 */
654d116b9f1SMateusz Guzik 	tdd->tdd_reapticks = ticks;
655d116b9f1SMateusz Guzik 
6565ef7b7a0SMateusz Guzik 	tidbatch_prep(&tidbatch);
657f34a2f56SMateusz Guzik 	credbatch_prep(&credbatch);
658bbe62559SMateusz Guzik 	limbatch_prep(&limbatch);
659e0c86f5cSMateusz Guzik 	tdcountbatch_prep(&tdcountbatch);
660d116b9f1SMateusz Guzik 
661c5315f51SMateusz Guzik 	while (itd != NULL) {
662c5315f51SMateusz Guzik 		ntd = itd->td_zombie;
6635ef7b7a0SMateusz Guzik 		EVENTHANDLER_DIRECT_INVOKE(thread_dtor, itd);
664bbe62559SMateusz Guzik 
6655ef7b7a0SMateusz Guzik 		tidbatch_add(&tidbatch, itd);
666f34a2f56SMateusz Guzik 		credbatch_add(&credbatch, itd);
667bbe62559SMateusz Guzik 		limbatch_add(&limbatch, itd);
668e0c86f5cSMateusz Guzik 		tdcountbatch_add(&tdcountbatch, itd);
669bbe62559SMateusz Guzik 
670755341dfSMateusz Guzik 		thread_free_batched(itd);
671bbe62559SMateusz Guzik 
6725ef7b7a0SMateusz Guzik 		tidbatch_process(&tidbatch);
673f34a2f56SMateusz Guzik 		credbatch_process(&credbatch);
674bbe62559SMateusz Guzik 		limbatch_process(&limbatch);
675e0c86f5cSMateusz Guzik 		tdcountbatch_process(&tdcountbatch);
676bbe62559SMateusz Guzik 
677c5315f51SMateusz Guzik 		itd = ntd;
67844990b8cSJulian Elischer 	}
679755341dfSMateusz Guzik 
6805ef7b7a0SMateusz Guzik 	tidbatch_final(&tidbatch);
681f34a2f56SMateusz Guzik 	credbatch_final(&credbatch);
682bbe62559SMateusz Guzik 	limbatch_final(&limbatch);
683e0c86f5cSMateusz Guzik 	tdcountbatch_final(&tdcountbatch);
684ed062c8dSJulian Elischer }
68544990b8cSJulian Elischer 
6864f0db5e0SJulian Elischer /*
687d116b9f1SMateusz Guzik  * Reap zombies from all domains.
688d116b9f1SMateusz Guzik  */
689d116b9f1SMateusz Guzik static void
690d116b9f1SMateusz Guzik thread_reap_all(void)
691d116b9f1SMateusz Guzik {
692d116b9f1SMateusz Guzik 	struct thread_domain_data *tdd;
693d116b9f1SMateusz Guzik 	int i, domain;
694d116b9f1SMateusz Guzik 
695d116b9f1SMateusz Guzik 	domain = PCPU_GET(domain);
696d116b9f1SMateusz Guzik 	for (i = 0; i < vm_ndomains; i++) {
697d116b9f1SMateusz Guzik 		tdd = &thread_domain_data[(i + domain) % vm_ndomains];
698d116b9f1SMateusz Guzik 		thread_reap_domain(tdd);
699d116b9f1SMateusz Guzik 	}
700d116b9f1SMateusz Guzik }
701d116b9f1SMateusz Guzik 
702d116b9f1SMateusz Guzik /*
703d116b9f1SMateusz Guzik  * Reap zombies from local domain.
704d116b9f1SMateusz Guzik  */
705b83e94beSMateusz Guzik static void
706d116b9f1SMateusz Guzik thread_reap(void)
707d116b9f1SMateusz Guzik {
708d116b9f1SMateusz Guzik 	struct thread_domain_data *tdd;
709d116b9f1SMateusz Guzik 	int domain;
710d116b9f1SMateusz Guzik 
711d116b9f1SMateusz Guzik 	domain = PCPU_GET(domain);
712d116b9f1SMateusz Guzik 	tdd = &thread_domain_data[domain];
713d116b9f1SMateusz Guzik 
714d116b9f1SMateusz Guzik 	thread_reap_domain(tdd);
715d116b9f1SMateusz Guzik }
716d116b9f1SMateusz Guzik 
717d116b9f1SMateusz Guzik static void
718d116b9f1SMateusz Guzik thread_reap_task_cb(void *arg __unused, int pending __unused)
719d116b9f1SMateusz Guzik {
720d116b9f1SMateusz Guzik 
721d116b9f1SMateusz Guzik 	thread_reap_all();
722d116b9f1SMateusz Guzik }
723d116b9f1SMateusz Guzik 
724d116b9f1SMateusz Guzik static void
725d116b9f1SMateusz Guzik thread_reap_callout_cb(void *arg __unused)
726d116b9f1SMateusz Guzik {
727d116b9f1SMateusz Guzik 	struct thread_domain_data *tdd;
728d116b9f1SMateusz Guzik 	int i, cticks, lticks;
729d116b9f1SMateusz Guzik 	bool wantreap;
730d116b9f1SMateusz Guzik 
731d116b9f1SMateusz Guzik 	wantreap = false;
732d116b9f1SMateusz Guzik 	cticks = atomic_load_int(&ticks);
733d116b9f1SMateusz Guzik 	for (i = 0; i < vm_ndomains; i++) {
734d116b9f1SMateusz Guzik 		tdd = &thread_domain_data[i];
735d116b9f1SMateusz Guzik 		lticks = tdd->tdd_reapticks;
736d116b9f1SMateusz Guzik 		if (tdd->tdd_zombies != NULL &&
737d116b9f1SMateusz Guzik 		    (u_int)(cticks - lticks) > 5 * hz) {
738d116b9f1SMateusz Guzik 			wantreap = true;
739d116b9f1SMateusz Guzik 			break;
740d116b9f1SMateusz Guzik 		}
741d116b9f1SMateusz Guzik 	}
742d116b9f1SMateusz Guzik 
743d116b9f1SMateusz Guzik 	if (wantreap)
744d116b9f1SMateusz Guzik 		taskqueue_enqueue(taskqueue_thread, &thread_reap_task);
745845d7797SKonstantin Belousov 	callout_reset(&thread_reap_callout, 5 * hz,
746845d7797SKonstantin Belousov 	    thread_reap_callout_cb, NULL);
747d116b9f1SMateusz Guzik }
748d116b9f1SMateusz Guzik 
749d116b9f1SMateusz Guzik /*
750f62c7e54SKonstantin Belousov  * Calling this function guarantees that any thread that exited before
751f62c7e54SKonstantin Belousov  * the call is reaped when the function returns.  By 'exited' we mean
752f62c7e54SKonstantin Belousov  * a thread removed from the process linkage with thread_unlink().
753f62c7e54SKonstantin Belousov  * Practically this means that caller must lock/unlock corresponding
754f62c7e54SKonstantin Belousov  * process lock before the call, to synchronize with thread_exit().
755f62c7e54SKonstantin Belousov  */
756f62c7e54SKonstantin Belousov void
757f62c7e54SKonstantin Belousov thread_reap_barrier(void)
758f62c7e54SKonstantin Belousov {
759f62c7e54SKonstantin Belousov 	struct task *t;
760f62c7e54SKonstantin Belousov 
761f62c7e54SKonstantin Belousov 	/*
762f62c7e54SKonstantin Belousov 	 * First do context switches to each CPU to ensure that all
763f62c7e54SKonstantin Belousov 	 * PCPU pc_deadthreads are moved to zombie list.
764f62c7e54SKonstantin Belousov 	 */
765f62c7e54SKonstantin Belousov 	quiesce_all_cpus("", PDROP);
766f62c7e54SKonstantin Belousov 
767f62c7e54SKonstantin Belousov 	/*
768f62c7e54SKonstantin Belousov 	 * Second, fire the task in the same thread as normal
769f62c7e54SKonstantin Belousov 	 * thread_reap() is done, to serialize reaping.
770f62c7e54SKonstantin Belousov 	 */
771f62c7e54SKonstantin Belousov 	t = malloc(sizeof(*t), M_TEMP, M_WAITOK);
772f62c7e54SKonstantin Belousov 	TASK_INIT(t, 0, thread_reap_task_cb, t);
773f62c7e54SKonstantin Belousov 	taskqueue_enqueue(taskqueue_thread, t);
774f62c7e54SKonstantin Belousov 	taskqueue_drain(taskqueue_thread, t);
775f62c7e54SKonstantin Belousov 	free(t, M_TEMP);
776f62c7e54SKonstantin Belousov }
777f62c7e54SKonstantin Belousov 
778f62c7e54SKonstantin Belousov /*
77944990b8cSJulian Elischer  * Allocate a thread.
78044990b8cSJulian Elischer  */
78144990b8cSJulian Elischer struct thread *
7828a945d10SKonstantin Belousov thread_alloc(int pages)
78344990b8cSJulian Elischer {
78489b57fcfSKonstantin Belousov 	struct thread *td;
7851bd3cf5dSMateusz Guzik 	lwpid_t tid;
7868460a577SJohn Birrell 
78762dbc992SMateusz Guzik 	if (!thread_count_inc()) {
7881bd3cf5dSMateusz Guzik 		return (NULL);
7891bd3cf5dSMateusz Guzik 	}
7901bd3cf5dSMateusz Guzik 
79162dbc992SMateusz Guzik 	tid = tid_alloc();
7921bd3cf5dSMateusz Guzik 	td = uma_zalloc(thread_zone, M_WAITOK);
79389b57fcfSKonstantin Belousov 	KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
7948a945d10SKonstantin Belousov 	if (!vm_thread_new(td, pages)) {
79589b57fcfSKonstantin Belousov 		uma_zfree(thread_zone, td);
7961bd3cf5dSMateusz Guzik 		tid_free(tid);
79762dbc992SMateusz Guzik 		thread_count_dec();
79889b57fcfSKonstantin Belousov 		return (NULL);
79989b57fcfSKonstantin Belousov 	}
8001bd3cf5dSMateusz Guzik 	td->td_tid = tid;
801f575573cSKonstantin Belousov 	bzero(&td->td_sa.args, sizeof(td->td_sa.args));
8025dda15adSMark Johnston 	kmsan_thread_alloc(td);
8030c3967e7SMarcel Moolenaar 	cpu_thread_alloc(td);
8041bd3cf5dSMateusz Guzik 	EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td);
80589b57fcfSKonstantin Belousov 	return (td);
80644990b8cSJulian Elischer }
80744990b8cSJulian Elischer 
8088a945d10SKonstantin Belousov int
8098a945d10SKonstantin Belousov thread_alloc_stack(struct thread *td, int pages)
8108a945d10SKonstantin Belousov {
8118a945d10SKonstantin Belousov 
8128a945d10SKonstantin Belousov 	KASSERT(td->td_kstack == 0,
8138a945d10SKonstantin Belousov 	    ("thread_alloc_stack called on a thread with kstack"));
8148a945d10SKonstantin Belousov 	if (!vm_thread_new(td, pages))
8158a945d10SKonstantin Belousov 		return (0);
8168a945d10SKonstantin Belousov 	cpu_thread_alloc(td);
8178a945d10SKonstantin Belousov 	return (1);
8188a945d10SKonstantin Belousov }
8194f0db5e0SJulian Elischer 
8204f0db5e0SJulian Elischer /*
82144990b8cSJulian Elischer  * Deallocate a thread.
82244990b8cSJulian Elischer  */
823755341dfSMateusz Guzik static void
824755341dfSMateusz Guzik thread_free_batched(struct thread *td)
82544990b8cSJulian Elischer {
8262e6b8de4SJeff Roberson 
8272e6b8de4SJeff Roberson 	lock_profile_thread_exit(td);
82845aea8deSJeff Roberson 	if (td->td_cpuset)
829d7f687fcSJeff Roberson 		cpuset_rel(td->td_cpuset);
830d7f687fcSJeff Roberson 	td->td_cpuset = NULL;
8310c3967e7SMarcel Moolenaar 	cpu_thread_free(td);
83289b57fcfSKonstantin Belousov 	if (td->td_kstack != 0)
83389b57fcfSKonstantin Belousov 		vm_thread_dispose(td);
8342d19b736SKonstantin Belousov 	callout_drain(&td->td_slpcallout);
835755341dfSMateusz Guzik 	/*
836755341dfSMateusz Guzik 	 * Freeing handled by the caller.
837755341dfSMateusz Guzik 	 */
8381bd3cf5dSMateusz Guzik 	td->td_tid = -1;
8395dda15adSMark Johnston 	kmsan_thread_free(td);
84044990b8cSJulian Elischer 	uma_zfree(thread_zone, td);
84144990b8cSJulian Elischer }
84244990b8cSJulian Elischer 
8434ea6a9a2SMateusz Guzik void
844755341dfSMateusz Guzik thread_free(struct thread *td)
845755341dfSMateusz Guzik {
846755341dfSMateusz Guzik 	lwpid_t tid;
847755341dfSMateusz Guzik 
8485ef7b7a0SMateusz Guzik 	EVENTHANDLER_DIRECT_INVOKE(thread_dtor, td);
849755341dfSMateusz Guzik 	tid = td->td_tid;
850755341dfSMateusz Guzik 	thread_free_batched(td);
851755341dfSMateusz Guzik 	tid_free(tid);
85262dbc992SMateusz Guzik 	thread_count_dec();
853755341dfSMateusz Guzik }
854755341dfSMateusz Guzik 
855755341dfSMateusz Guzik void
8564ea6a9a2SMateusz Guzik thread_cow_get_proc(struct thread *newtd, struct proc *p)
8574ea6a9a2SMateusz Guzik {
8584ea6a9a2SMateusz Guzik 
8594ea6a9a2SMateusz Guzik 	PROC_LOCK_ASSERT(p, MA_OWNED);
8601724c563SMateusz Guzik 	newtd->td_realucred = crcowget(p->p_ucred);
8611724c563SMateusz Guzik 	newtd->td_ucred = newtd->td_realucred;
862f6f6d240SMateusz Guzik 	newtd->td_limit = lim_hold(p->p_limit);
8634ea6a9a2SMateusz Guzik 	newtd->td_cowgen = p->p_cowgen;
8644ea6a9a2SMateusz Guzik }
8654ea6a9a2SMateusz Guzik 
8664ea6a9a2SMateusz Guzik void
8674ea6a9a2SMateusz Guzik thread_cow_get(struct thread *newtd, struct thread *td)
8684ea6a9a2SMateusz Guzik {
8694ea6a9a2SMateusz Guzik 
8701724c563SMateusz Guzik 	MPASS(td->td_realucred == td->td_ucred);
8711724c563SMateusz Guzik 	newtd->td_realucred = crcowget(td->td_realucred);
8721724c563SMateusz Guzik 	newtd->td_ucred = newtd->td_realucred;
873f6f6d240SMateusz Guzik 	newtd->td_limit = lim_hold(td->td_limit);
8744ea6a9a2SMateusz Guzik 	newtd->td_cowgen = td->td_cowgen;
8754ea6a9a2SMateusz Guzik }
8764ea6a9a2SMateusz Guzik 
8774ea6a9a2SMateusz Guzik void
8784ea6a9a2SMateusz Guzik thread_cow_free(struct thread *td)
8794ea6a9a2SMateusz Guzik {
8804ea6a9a2SMateusz Guzik 
8811724c563SMateusz Guzik 	if (td->td_realucred != NULL)
8821724c563SMateusz Guzik 		crcowfree(td);
883cd672ca6SMateusz Guzik 	if (td->td_limit != NULL)
884f6f6d240SMateusz Guzik 		lim_free(td->td_limit);
8854ea6a9a2SMateusz Guzik }
8864ea6a9a2SMateusz Guzik 
8874ea6a9a2SMateusz Guzik void
8884ea6a9a2SMateusz Guzik thread_cow_update(struct thread *td)
8894ea6a9a2SMateusz Guzik {
8904ea6a9a2SMateusz Guzik 	struct proc *p;
891cd672ca6SMateusz Guzik 	struct ucred *oldcred;
892cd672ca6SMateusz Guzik 	struct plimit *oldlimit;
8934ea6a9a2SMateusz Guzik 
8944ea6a9a2SMateusz Guzik 	p = td->td_proc;
8954ea6a9a2SMateusz Guzik 	PROC_LOCK(p);
8961724c563SMateusz Guzik 	oldcred = crcowsync();
8978a0cb04dSMateusz Guzik 	oldlimit = lim_cowsync();
8984ea6a9a2SMateusz Guzik 	td->td_cowgen = p->p_cowgen;
8994ea6a9a2SMateusz Guzik 	PROC_UNLOCK(p);
900cd672ca6SMateusz Guzik 	if (oldcred != NULL)
901cd672ca6SMateusz Guzik 		crfree(oldcred);
902cd672ca6SMateusz Guzik 	if (oldlimit != NULL)
903cd672ca6SMateusz Guzik 		lim_free(oldlimit);
9044ea6a9a2SMateusz Guzik }
9054ea6a9a2SMateusz Guzik 
90632114b63SMateusz Guzik void
90732114b63SMateusz Guzik thread_cow_synced(struct thread *td)
90832114b63SMateusz Guzik {
90932114b63SMateusz Guzik 	struct proc *p;
91032114b63SMateusz Guzik 
91132114b63SMateusz Guzik 	p = td->td_proc;
91232114b63SMateusz Guzik 	PROC_LOCK_ASSERT(p, MA_OWNED);
91332114b63SMateusz Guzik 	MPASS(td->td_cowgen != p->p_cowgen);
91432114b63SMateusz Guzik 	MPASS(td->td_ucred == p->p_ucred);
91532114b63SMateusz Guzik 	MPASS(td->td_limit == p->p_limit);
91632114b63SMateusz Guzik 	td->td_cowgen = p->p_cowgen;
91732114b63SMateusz Guzik }
91832114b63SMateusz Guzik 
91944990b8cSJulian Elischer /*
92044990b8cSJulian Elischer  * Discard the current thread and exit from its context.
92194e0a4cdSJulian Elischer  * Always called with scheduler locked.
92244990b8cSJulian Elischer  *
92344990b8cSJulian Elischer  * Because we can't free a thread while we're operating under its context,
924696058c3SJulian Elischer  * push the current thread into our CPU's deadthread holder. This means
925696058c3SJulian Elischer  * we needn't worry about someone else grabbing our context before we
9266617724cSJeff Roberson  * do a cpu_throw().
92744990b8cSJulian Elischer  */
92844990b8cSJulian Elischer void
92944990b8cSJulian Elischer thread_exit(void)
93044990b8cSJulian Elischer {
9317e3a96eaSJohn Baldwin 	uint64_t runtime, new_switchtime;
93244990b8cSJulian Elischer 	struct thread *td;
9331c4bcd05SJeff Roberson 	struct thread *td2;
93444990b8cSJulian Elischer 	struct proc *p;
9357847a9daSJohn Baldwin 	int wakeup_swapper;
93644990b8cSJulian Elischer 
93744990b8cSJulian Elischer 	td = curthread;
93844990b8cSJulian Elischer 	p = td->td_proc;
93944990b8cSJulian Elischer 
940a54e85fdSJeff Roberson 	PROC_SLOCK_ASSERT(p, MA_OWNED);
941ed062c8dSJulian Elischer 	mtx_assert(&Giant, MA_NOTOWNED);
942a54e85fdSJeff Roberson 
94344990b8cSJulian Elischer 	PROC_LOCK_ASSERT(p, MA_OWNED);
944ed062c8dSJulian Elischer 	KASSERT(p != NULL, ("thread exiting without a process"));
945cc701b73SRobert Watson 	CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
946e01eafefSJulian Elischer 	    (long)p->p_pid, td->td_name);
9476c9271a9SAndriy Gapon 	SDT_PROBE0(proc, , , lwp__exit);
9489104847fSDavid Xu 	KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
949936c24faSMateusz Guzik 	MPASS(td->td_realucred == td->td_ucred);
95044990b8cSJulian Elischer 
951ed062c8dSJulian Elischer 	/*
952ed062c8dSJulian Elischer 	 * drop FPU & debug register state storage, or any other
953ed062c8dSJulian Elischer 	 * architecture specific resources that
954ed062c8dSJulian Elischer 	 * would not be on a new untouched process.
955ed062c8dSJulian Elischer 	 */
956bd07998eSKonstantin Belousov 	cpu_thread_exit(td);
95744990b8cSJulian Elischer 
958ed062c8dSJulian Elischer 	/*
9591faf202eSJulian Elischer 	 * The last thread is left attached to the process
9601faf202eSJulian Elischer 	 * So that the whole bundle gets recycled. Skip
961ed062c8dSJulian Elischer 	 * all this stuff if we never had threads.
962ed062c8dSJulian Elischer 	 * EXIT clears all sign of other threads when
963ed062c8dSJulian Elischer 	 * it goes to single threading, so the last thread always
964ed062c8dSJulian Elischer 	 * takes the short path.
9651faf202eSJulian Elischer 	 */
966ed062c8dSJulian Elischer 	if (p->p_flag & P_HADTHREADS) {
9671faf202eSJulian Elischer 		if (p->p_numthreads > 1) {
968fd229b5bSKonstantin Belousov 			atomic_add_int(&td->td_proc->p_exitthreads, 1);
969d3a0bd78SJulian Elischer 			thread_unlink(td);
9701c4bcd05SJeff Roberson 			td2 = FIRST_THREAD_IN_PROC(p);
9711c4bcd05SJeff Roberson 			sched_exit_thread(td2, td);
972ed062c8dSJulian Elischer 
973ed062c8dSJulian Elischer 			/*
97444990b8cSJulian Elischer 			 * The test below is NOT true if we are the
9759182554aSKonstantin Belousov 			 * sole exiting thread. P_STOPPED_SINGLE is unset
97644990b8cSJulian Elischer 			 * in exit1() after it is the only survivor.
97744990b8cSJulian Elischer 			 */
9781279572aSDavid Xu 			if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
97944990b8cSJulian Elischer 				if (p->p_numthreads == p->p_suspcount) {
980a54e85fdSJeff Roberson 					thread_lock(p->p_singlethread);
9817847a9daSJohn Baldwin 					wakeup_swapper = thread_unsuspend_one(
98284cdea97SKonstantin Belousov 						p->p_singlethread, p, false);
9837847a9daSJohn Baldwin 					if (wakeup_swapper)
9847847a9daSJohn Baldwin 						kick_proc0();
98544990b8cSJulian Elischer 				}
98644990b8cSJulian Elischer 			}
98748bfcdddSJulian Elischer 
988696058c3SJulian Elischer 			PCPU_SET(deadthread, td);
9891faf202eSJulian Elischer 		} else {
990ed062c8dSJulian Elischer 			/*
991ed062c8dSJulian Elischer 			 * The last thread is exiting.. but not through exit()
992ed062c8dSJulian Elischer 			 */
993ed062c8dSJulian Elischer 			panic ("thread_exit: Last thread exiting on its own");
994ed062c8dSJulian Elischer 		}
9951faf202eSJulian Elischer 	}
99616d95d4fSJoseph Koshy #ifdef	HWPMC_HOOKS
99716d95d4fSJoseph Koshy 	/*
99816d95d4fSJoseph Koshy 	 * If this thread is part of a process that is being tracked by hwpmc(4),
99916d95d4fSJoseph Koshy 	 * inform the module of the thread's impending exit.
100016d95d4fSJoseph Koshy 	 */
10016161b98cSMatt Macy 	if (PMC_PROC_IS_USING_PMCS(td->td_proc)) {
100216d95d4fSJoseph Koshy 		PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
10036161b98cSMatt Macy 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT, NULL);
1004ebfaf69cSMatt Macy 	} else if (PMC_SYSTEM_SAMPLING_ACTIVE())
1005ebfaf69cSMatt Macy 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT_LOG, NULL);
100616d95d4fSJoseph Koshy #endif
1007a54e85fdSJeff Roberson 	PROC_UNLOCK(p);
10085c7bebf9SKonstantin Belousov 	PROC_STATLOCK(p);
10095c7bebf9SKonstantin Belousov 	thread_lock(td);
10105c7bebf9SKonstantin Belousov 	PROC_SUNLOCK(p);
10117e3a96eaSJohn Baldwin 
10127e3a96eaSJohn Baldwin 	/* Do the same timestamp bookkeeping that mi_switch() would do. */
10137e3a96eaSJohn Baldwin 	new_switchtime = cpu_ticks();
10147e3a96eaSJohn Baldwin 	runtime = new_switchtime - PCPU_GET(switchtime);
10157e3a96eaSJohn Baldwin 	td->td_runtime += runtime;
10167e3a96eaSJohn Baldwin 	td->td_incruntime += runtime;
10177e3a96eaSJohn Baldwin 	PCPU_SET(switchtime, new_switchtime);
10187e3a96eaSJohn Baldwin 	PCPU_SET(switchticks, ticks);
101983c9dea1SGleb Smirnoff 	VM_CNT_INC(v_swtch);
10207e3a96eaSJohn Baldwin 
10217e3a96eaSJohn Baldwin 	/* Save our resource usage in our process. */
10227e3a96eaSJohn Baldwin 	td->td_ru.ru_nvcsw++;
102361a74c5cSJeff Roberson 	ruxagg_locked(p, td);
10247e3a96eaSJohn Baldwin 	rucollect(&p->p_ru, &td->td_ru);
10255c7bebf9SKonstantin Belousov 	PROC_STATUNLOCK(p);
10267e3a96eaSJohn Baldwin 
1027fa2528acSAlex Richardson 	TD_SET_STATE(td, TDS_INACTIVE);
10283d06b4b3SAttilio Rao #ifdef WITNESS
10293d06b4b3SAttilio Rao 	witness_thread_exit(td);
10303d06b4b3SAttilio Rao #endif
1031732d9528SJulian Elischer 	CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
1032a54e85fdSJeff Roberson 	sched_throw(td);
1033cc66ebe2SPeter Wemm 	panic("I'm a teapot!");
103444990b8cSJulian Elischer 	/* NOTREACHED */
103544990b8cSJulian Elischer }
103644990b8cSJulian Elischer 
103744990b8cSJulian Elischer /*
1038696058c3SJulian Elischer  * Do any thread specific cleanups that may be needed in wait()
103937814395SPeter Wemm  * called with Giant, proc and schedlock not held.
1040696058c3SJulian Elischer  */
1041696058c3SJulian Elischer void
1042696058c3SJulian Elischer thread_wait(struct proc *p)
1043696058c3SJulian Elischer {
1044696058c3SJulian Elischer 	struct thread *td;
1045696058c3SJulian Elischer 
104637814395SPeter Wemm 	mtx_assert(&Giant, MA_NOTOWNED);
1047624bf9e1SKonstantin Belousov 	KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()"));
1048624bf9e1SKonstantin Belousov 	KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking"));
1049ff8fbcffSJeff Roberson 	td = FIRST_THREAD_IN_PROC(p);
1050ff8fbcffSJeff Roberson 	/* Lock the last thread so we spin until it exits cpu_throw(). */
1051ff8fbcffSJeff Roberson 	thread_lock(td);
1052ff8fbcffSJeff Roberson 	thread_unlock(td);
10532e6b8de4SJeff Roberson 	lock_profile_thread_exit(td);
1054d7f687fcSJeff Roberson 	cpuset_rel(td->td_cpuset);
1055d7f687fcSJeff Roberson 	td->td_cpuset = NULL;
1056696058c3SJulian Elischer 	cpu_thread_clean(td);
10574ea6a9a2SMateusz Guzik 	thread_cow_free(td);
10582d19b736SKonstantin Belousov 	callout_drain(&td->td_slpcallout);
1059696058c3SJulian Elischer 	thread_reap();	/* check for zombie threads etc. */
1060696058c3SJulian Elischer }
1061696058c3SJulian Elischer 
1062696058c3SJulian Elischer /*
106344990b8cSJulian Elischer  * Link a thread to a process.
10641faf202eSJulian Elischer  * set up anything that needs to be initialized for it to
10651faf202eSJulian Elischer  * be used by the process.
106644990b8cSJulian Elischer  */
106744990b8cSJulian Elischer void
10688460a577SJohn Birrell thread_link(struct thread *td, struct proc *p)
106944990b8cSJulian Elischer {
107044990b8cSJulian Elischer 
1071a54e85fdSJeff Roberson 	/*
1072a54e85fdSJeff Roberson 	 * XXX This can't be enabled because it's called for proc0 before
1073374ae2a3SJeff Roberson 	 * its lock has been created.
1074374ae2a3SJeff Roberson 	 * PROC_LOCK_ASSERT(p, MA_OWNED);
1075a54e85fdSJeff Roberson 	 */
1076fa2528acSAlex Richardson 	TD_SET_STATE(td, TDS_INACTIVE);
107744990b8cSJulian Elischer 	td->td_proc     = p;
1078b61ce5b0SJeff Roberson 	td->td_flags    = TDF_INMEM;
107944990b8cSJulian Elischer 
10801faf202eSJulian Elischer 	LIST_INIT(&td->td_contested);
1081eea4f254SJeff Roberson 	LIST_INIT(&td->td_lprof[0]);
1082eea4f254SJeff Roberson 	LIST_INIT(&td->td_lprof[1]);
1083f6eccf96SGleb Smirnoff #ifdef EPOCH_TRACE
1084dd902d01SGleb Smirnoff 	SLIST_INIT(&td->td_epochs);
1085f6eccf96SGleb Smirnoff #endif
10869104847fSDavid Xu 	sigqueue_init(&td->td_sigqueue, p);
1087fd90e2edSJung-uk Kim 	callout_init(&td->td_slpcallout, 1);
108866d8df9dSDaniel Eischen 	TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist);
108944990b8cSJulian Elischer 	p->p_numthreads++;
109044990b8cSJulian Elischer }
109144990b8cSJulian Elischer 
1092ed062c8dSJulian Elischer /*
1093ed062c8dSJulian Elischer  * Called from:
1094ed062c8dSJulian Elischer  *  thread_exit()
1095ed062c8dSJulian Elischer  */
1096d3a0bd78SJulian Elischer void
1097d3a0bd78SJulian Elischer thread_unlink(struct thread *td)
1098d3a0bd78SJulian Elischer {
1099d3a0bd78SJulian Elischer 	struct proc *p = td->td_proc;
1100d3a0bd78SJulian Elischer 
1101374ae2a3SJeff Roberson 	PROC_LOCK_ASSERT(p, MA_OWNED);
1102f6eccf96SGleb Smirnoff #ifdef EPOCH_TRACE
1103dd902d01SGleb Smirnoff 	MPASS(SLIST_EMPTY(&td->td_epochs));
1104f6eccf96SGleb Smirnoff #endif
1105dd902d01SGleb Smirnoff 
1106d3a0bd78SJulian Elischer 	TAILQ_REMOVE(&p->p_threads, td, td_plist);
1107d3a0bd78SJulian Elischer 	p->p_numthreads--;
1108d3a0bd78SJulian Elischer 	/* could clear a few other things here */
11098460a577SJohn Birrell 	/* Must  NOT clear links to proc! */
11105c8329edSJulian Elischer }
11115c8329edSJulian Elischer 
111279799053SKonstantin Belousov static int
111379799053SKonstantin Belousov calc_remaining(struct proc *p, int mode)
111479799053SKonstantin Belousov {
111579799053SKonstantin Belousov 	int remaining;
111679799053SKonstantin Belousov 
11177b519077SKonstantin Belousov 	PROC_LOCK_ASSERT(p, MA_OWNED);
11187b519077SKonstantin Belousov 	PROC_SLOCK_ASSERT(p, MA_OWNED);
111979799053SKonstantin Belousov 	if (mode == SINGLE_EXIT)
112079799053SKonstantin Belousov 		remaining = p->p_numthreads;
112179799053SKonstantin Belousov 	else if (mode == SINGLE_BOUNDARY)
112279799053SKonstantin Belousov 		remaining = p->p_numthreads - p->p_boundary_count;
11236ddcc233SKonstantin Belousov 	else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC)
112479799053SKonstantin Belousov 		remaining = p->p_numthreads - p->p_suspcount;
112579799053SKonstantin Belousov 	else
112679799053SKonstantin Belousov 		panic("calc_remaining: wrong mode %d", mode);
112779799053SKonstantin Belousov 	return (remaining);
112879799053SKonstantin Belousov }
112979799053SKonstantin Belousov 
113007a9368aSKonstantin Belousov static int
113107a9368aSKonstantin Belousov remain_for_mode(int mode)
113207a9368aSKonstantin Belousov {
113307a9368aSKonstantin Belousov 
11346ddcc233SKonstantin Belousov 	return (mode == SINGLE_ALLPROC ? 0 : 1);
113507a9368aSKonstantin Belousov }
113607a9368aSKonstantin Belousov 
113707a9368aSKonstantin Belousov static int
113807a9368aSKonstantin Belousov weed_inhib(int mode, struct thread *td2, struct proc *p)
113907a9368aSKonstantin Belousov {
114007a9368aSKonstantin Belousov 	int wakeup_swapper;
114107a9368aSKonstantin Belousov 
114207a9368aSKonstantin Belousov 	PROC_LOCK_ASSERT(p, MA_OWNED);
114307a9368aSKonstantin Belousov 	PROC_SLOCK_ASSERT(p, MA_OWNED);
114407a9368aSKonstantin Belousov 	THREAD_LOCK_ASSERT(td2, MA_OWNED);
114507a9368aSKonstantin Belousov 
114607a9368aSKonstantin Belousov 	wakeup_swapper = 0;
114761a74c5cSJeff Roberson 
114861a74c5cSJeff Roberson 	/*
114961a74c5cSJeff Roberson 	 * Since the thread lock is dropped by the scheduler we have
115061a74c5cSJeff Roberson 	 * to retry to check for races.
115161a74c5cSJeff Roberson 	 */
115261a74c5cSJeff Roberson restart:
115307a9368aSKonstantin Belousov 	switch (mode) {
115407a9368aSKonstantin Belousov 	case SINGLE_EXIT:
115561a74c5cSJeff Roberson 		if (TD_IS_SUSPENDED(td2)) {
115684cdea97SKonstantin Belousov 			wakeup_swapper |= thread_unsuspend_one(td2, p, true);
115761a74c5cSJeff Roberson 			thread_lock(td2);
115861a74c5cSJeff Roberson 			goto restart;
115961a74c5cSJeff Roberson 		}
116061a74c5cSJeff Roberson 		if (TD_CAN_ABORT(td2)) {
116107a9368aSKonstantin Belousov 			wakeup_swapper |= sleepq_abort(td2, EINTR);
116261a74c5cSJeff Roberson 			return (wakeup_swapper);
116361a74c5cSJeff Roberson 		}
116407a9368aSKonstantin Belousov 		break;
116507a9368aSKonstantin Belousov 	case SINGLE_BOUNDARY:
116607a9368aSKonstantin Belousov 	case SINGLE_NO_EXIT:
116761a74c5cSJeff Roberson 		if (TD_IS_SUSPENDED(td2) &&
116861a74c5cSJeff Roberson 		    (td2->td_flags & TDF_BOUNDARY) == 0) {
116984cdea97SKonstantin Belousov 			wakeup_swapper |= thread_unsuspend_one(td2, p, false);
117061a74c5cSJeff Roberson 			thread_lock(td2);
117161a74c5cSJeff Roberson 			goto restart;
117261a74c5cSJeff Roberson 		}
117361a74c5cSJeff Roberson 		if (TD_CAN_ABORT(td2)) {
117407a9368aSKonstantin Belousov 			wakeup_swapper |= sleepq_abort(td2, ERESTART);
117561a74c5cSJeff Roberson 			return (wakeup_swapper);
117661a74c5cSJeff Roberson 		}
1177917dd390SKonstantin Belousov 		break;
11786ddcc233SKonstantin Belousov 	case SINGLE_ALLPROC:
11796ddcc233SKonstantin Belousov 		/*
11806ddcc233SKonstantin Belousov 		 * ALLPROC suspend tries to avoid spurious EINTR for
11816ddcc233SKonstantin Belousov 		 * threads sleeping interruptable, by suspending the
11826ddcc233SKonstantin Belousov 		 * thread directly, similarly to sig_suspend_threads().
1183dd883e9aSKonstantin Belousov 		 * Since such sleep is not neccessary performed at the user
1184dd883e9aSKonstantin Belousov 		 * boundary, TDF_ALLPROCSUSP is used to avoid immediate
1185dd883e9aSKonstantin Belousov 		 * un-suspend.
11866ddcc233SKonstantin Belousov 		 */
11875c274b36SBrooks Davis 		if (TD_IS_SUSPENDED(td2) &&
11885c274b36SBrooks Davis 		    (td2->td_flags & TDF_ALLPROCSUSP) == 0) {
118984cdea97SKonstantin Belousov 			wakeup_swapper |= thread_unsuspend_one(td2, p, false);
119061a74c5cSJeff Roberson 			thread_lock(td2);
119161a74c5cSJeff Roberson 			goto restart;
119261a74c5cSJeff Roberson 		}
119361a74c5cSJeff Roberson 		if (TD_CAN_ABORT(td2)) {
11946ddcc233SKonstantin Belousov 			td2->td_flags |= TDF_ALLPROCSUSP;
11956ddcc233SKonstantin Belousov 			wakeup_swapper |= sleepq_abort(td2, ERESTART);
119661a74c5cSJeff Roberson 			return (wakeup_swapper);
11976ddcc233SKonstantin Belousov 		}
119807a9368aSKonstantin Belousov 		break;
119961a74c5cSJeff Roberson 	default:
120061a74c5cSJeff Roberson 		break;
120107a9368aSKonstantin Belousov 	}
120261a74c5cSJeff Roberson 	thread_unlock(td2);
120307a9368aSKonstantin Belousov 	return (wakeup_swapper);
120407a9368aSKonstantin Belousov }
120507a9368aSKonstantin Belousov 
12065215b187SJeff Roberson /*
120744990b8cSJulian Elischer  * Enforce single-threading.
120844990b8cSJulian Elischer  *
120944990b8cSJulian Elischer  * Returns 1 if the caller must abort (another thread is waiting to
121044990b8cSJulian Elischer  * exit the process or similar). Process is locked!
121144990b8cSJulian Elischer  * Returns 0 when you are successfully the only thread running.
121244990b8cSJulian Elischer  * A process has successfully single threaded in the suspend mode when
121344990b8cSJulian Elischer  * There are no threads in user mode. Threads in the kernel must be
121444990b8cSJulian Elischer  * allowed to continue until they get to the user boundary. They may even
121544990b8cSJulian Elischer  * copy out their return values and data before suspending. They may however be
1216e2668f55SMaxim Konovalov  * accelerated in reaching the user boundary as we will wake up
121744990b8cSJulian Elischer  * any sleeping threads that are interruptable. (PCATCH).
121844990b8cSJulian Elischer  */
121944990b8cSJulian Elischer int
12206ddcc233SKonstantin Belousov thread_single(struct proc *p, int mode)
122144990b8cSJulian Elischer {
122244990b8cSJulian Elischer 	struct thread *td;
122344990b8cSJulian Elischer 	struct thread *td2;
1224da7bbd2cSJohn Baldwin 	int remaining, wakeup_swapper;
122544990b8cSJulian Elischer 
122644990b8cSJulian Elischer 	td = curthread;
12276ddcc233SKonstantin Belousov 	KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
12286ddcc233SKonstantin Belousov 	    mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
12296ddcc233SKonstantin Belousov 	    ("invalid mode %d", mode));
12306ddcc233SKonstantin Belousov 	/*
12316ddcc233SKonstantin Belousov 	 * If allowing non-ALLPROC singlethreading for non-curproc
12326ddcc233SKonstantin Belousov 	 * callers, calc_remaining() and remain_for_mode() should be
12336ddcc233SKonstantin Belousov 	 * adjusted to also account for td->td_proc != p.  For now
12346ddcc233SKonstantin Belousov 	 * this is not implemented because it is not used.
12356ddcc233SKonstantin Belousov 	 */
12366ddcc233SKonstantin Belousov 	KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) ||
12376ddcc233SKonstantin Belousov 	    (mode != SINGLE_ALLPROC && td->td_proc == p),
12386ddcc233SKonstantin Belousov 	    ("mode %d proc %p curproc %p", mode, p, td->td_proc));
123937814395SPeter Wemm 	mtx_assert(&Giant, MA_NOTOWNED);
124044990b8cSJulian Elischer 	PROC_LOCK_ASSERT(p, MA_OWNED);
124144990b8cSJulian Elischer 
1242d7a9e6e7SKonstantin Belousov 	/*
1243d7a9e6e7SKonstantin Belousov 	 * Is someone already single threading?
1244d7a9e6e7SKonstantin Belousov 	 * Or may be singlethreading is not needed at all.
1245d7a9e6e7SKonstantin Belousov 	 */
1246d7a9e6e7SKonstantin Belousov 	if (mode == SINGLE_ALLPROC) {
1247d7a9e6e7SKonstantin Belousov 		while ((p->p_flag & P_STOPPED_SINGLE) != 0) {
1248d7a9e6e7SKonstantin Belousov 			if ((p->p_flag2 & P2_WEXIT) != 0)
1249d7a9e6e7SKonstantin Belousov 				return (1);
1250d7a9e6e7SKonstantin Belousov 			msleep(&p->p_flag, &p->p_mtx, PCATCH, "thrsgl", 0);
1251d7a9e6e7SKonstantin Belousov 		}
1252d7a9e6e7SKonstantin Belousov 	} else if ((p->p_flag & P_HADTHREADS) == 0)
125344990b8cSJulian Elischer 		return (0);
1254906ac69dSDavid Xu 	if (p->p_singlethread != NULL && p->p_singlethread != td)
125544990b8cSJulian Elischer 		return (1);
125644990b8cSJulian Elischer 
1257906ac69dSDavid Xu 	if (mode == SINGLE_EXIT) {
1258906ac69dSDavid Xu 		p->p_flag |= P_SINGLE_EXIT;
1259906ac69dSDavid Xu 		p->p_flag &= ~P_SINGLE_BOUNDARY;
1260906ac69dSDavid Xu 	} else {
1261906ac69dSDavid Xu 		p->p_flag &= ~P_SINGLE_EXIT;
1262906ac69dSDavid Xu 		if (mode == SINGLE_BOUNDARY)
1263906ac69dSDavid Xu 			p->p_flag |= P_SINGLE_BOUNDARY;
1264906ac69dSDavid Xu 		else
1265906ac69dSDavid Xu 			p->p_flag &= ~P_SINGLE_BOUNDARY;
1266906ac69dSDavid Xu 	}
1267f829268bSKonstantin Belousov 	if (mode == SINGLE_ALLPROC)
12686ddcc233SKonstantin Belousov 		p->p_flag |= P_TOTAL_STOP;
12691279572aSDavid Xu 	p->p_flag |= P_STOPPED_SINGLE;
12707b4a950aSDavid Xu 	PROC_SLOCK(p);
1271112afcb2SJohn Baldwin 	p->p_singlethread = td;
127279799053SKonstantin Belousov 	remaining = calc_remaining(p, mode);
127307a9368aSKonstantin Belousov 	while (remaining != remain_for_mode(mode)) {
1274bf1a3220SDavid Xu 		if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
1275bf1a3220SDavid Xu 			goto stopme;
1276da7bbd2cSJohn Baldwin 		wakeup_swapper = 0;
127744990b8cSJulian Elischer 		FOREACH_THREAD_IN_PROC(p, td2) {
127844990b8cSJulian Elischer 			if (td2 == td)
127944990b8cSJulian Elischer 				continue;
1280a54e85fdSJeff Roberson 			thread_lock(td2);
1281c6d31b83SKonstantin Belousov 			ast_sched_locked(td2, TDA_SUSPEND);
12826ddcc233SKonstantin Belousov 			if (TD_IS_INHIBITED(td2)) {
128307a9368aSKonstantin Belousov 				wakeup_swapper |= weed_inhib(mode, td2, p);
1284d8267df7SDavid Xu #ifdef SMP
1285b9009b17SKonstantin Belousov 			} else if (TD_IS_RUNNING(td2)) {
1286d8267df7SDavid Xu 				forward_signal(td2);
128761a74c5cSJeff Roberson 				thread_unlock(td2);
1288d8267df7SDavid Xu #endif
128961a74c5cSJeff Roberson 			} else
1290a54e85fdSJeff Roberson 				thread_unlock(td2);
12919d102777SJulian Elischer 		}
1292da7bbd2cSJohn Baldwin 		if (wakeup_swapper)
1293da7bbd2cSJohn Baldwin 			kick_proc0();
129479799053SKonstantin Belousov 		remaining = calc_remaining(p, mode);
1295ec008e96SDavid Xu 
12969d102777SJulian Elischer 		/*
12979d102777SJulian Elischer 		 * Maybe we suspended some threads.. was it enough?
12989d102777SJulian Elischer 		 */
129907a9368aSKonstantin Belousov 		if (remaining == remain_for_mode(mode))
13009d102777SJulian Elischer 			break;
13019d102777SJulian Elischer 
1302bf1a3220SDavid Xu stopme:
130344990b8cSJulian Elischer 		/*
130444990b8cSJulian Elischer 		 * Wake us up when everyone else has suspended.
1305e3b9bf71SJulian Elischer 		 * In the mean time we suspend as well.
130644990b8cSJulian Elischer 		 */
13076ddcc233SKonstantin Belousov 		thread_suspend_switch(td, p);
130879799053SKonstantin Belousov 		remaining = calc_remaining(p, mode);
130944990b8cSJulian Elischer 	}
1310906ac69dSDavid Xu 	if (mode == SINGLE_EXIT) {
131191599697SJulian Elischer 		/*
13128626a0ddSKonstantin Belousov 		 * Convert the process to an unthreaded process.  The
13138626a0ddSKonstantin Belousov 		 * SINGLE_EXIT is called by exit1() or execve(), in
13148626a0ddSKonstantin Belousov 		 * both cases other threads must be retired.
131591599697SJulian Elischer 		 */
13168626a0ddSKonstantin Belousov 		KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads"));
1317ed062c8dSJulian Elischer 		p->p_singlethread = NULL;
13188626a0ddSKonstantin Belousov 		p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS);
1319fd229b5bSKonstantin Belousov 
1320fd229b5bSKonstantin Belousov 		/*
1321fd229b5bSKonstantin Belousov 		 * Wait for any remaining threads to exit cpu_throw().
1322fd229b5bSKonstantin Belousov 		 */
1323fd229b5bSKonstantin Belousov 		while (p->p_exitthreads != 0) {
1324fd229b5bSKonstantin Belousov 			PROC_SUNLOCK(p);
1325fd229b5bSKonstantin Belousov 			PROC_UNLOCK(p);
1326fd229b5bSKonstantin Belousov 			sched_relinquish(td);
1327fd229b5bSKonstantin Belousov 			PROC_LOCK(p);
1328fd229b5bSKonstantin Belousov 			PROC_SLOCK(p);
1329fd229b5bSKonstantin Belousov 		}
1330ac437c07SKonstantin Belousov 	} else if (mode == SINGLE_BOUNDARY) {
1331ac437c07SKonstantin Belousov 		/*
1332ac437c07SKonstantin Belousov 		 * Wait until all suspended threads are removed from
1333ac437c07SKonstantin Belousov 		 * the processors.  The thread_suspend_check()
1334ac437c07SKonstantin Belousov 		 * increments p_boundary_count while it is still
1335ac437c07SKonstantin Belousov 		 * running, which makes it possible for the execve()
1336ac437c07SKonstantin Belousov 		 * to destroy vmspace while our other threads are
1337ac437c07SKonstantin Belousov 		 * still using the address space.
1338ac437c07SKonstantin Belousov 		 *
1339ac437c07SKonstantin Belousov 		 * We lock the thread, which is only allowed to
1340ac437c07SKonstantin Belousov 		 * succeed after context switch code finished using
1341ac437c07SKonstantin Belousov 		 * the address space.
1342ac437c07SKonstantin Belousov 		 */
1343ac437c07SKonstantin Belousov 		FOREACH_THREAD_IN_PROC(p, td2) {
1344ac437c07SKonstantin Belousov 			if (td2 == td)
1345ac437c07SKonstantin Belousov 				continue;
1346ac437c07SKonstantin Belousov 			thread_lock(td2);
1347ac437c07SKonstantin Belousov 			KASSERT((td2->td_flags & TDF_BOUNDARY) != 0,
1348ac437c07SKonstantin Belousov 			    ("td %p not on boundary", td2));
1349ac437c07SKonstantin Belousov 			KASSERT(TD_IS_SUSPENDED(td2),
1350ac437c07SKonstantin Belousov 			    ("td %p is not suspended", td2));
1351ac437c07SKonstantin Belousov 			thread_unlock(td2);
1352ac437c07SKonstantin Belousov 		}
135391599697SJulian Elischer 	}
13547b4a950aSDavid Xu 	PROC_SUNLOCK(p);
135544990b8cSJulian Elischer 	return (0);
135644990b8cSJulian Elischer }
135744990b8cSJulian Elischer 
13588638fe7bSKonstantin Belousov bool
13598638fe7bSKonstantin Belousov thread_suspend_check_needed(void)
13608638fe7bSKonstantin Belousov {
13618638fe7bSKonstantin Belousov 	struct proc *p;
13628638fe7bSKonstantin Belousov 	struct thread *td;
13638638fe7bSKonstantin Belousov 
13648638fe7bSKonstantin Belousov 	td = curthread;
13658638fe7bSKonstantin Belousov 	p = td->td_proc;
13668638fe7bSKonstantin Belousov 	PROC_LOCK_ASSERT(p, MA_OWNED);
13678638fe7bSKonstantin Belousov 	return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 &&
13688638fe7bSKonstantin Belousov 	    (td->td_dbgflags & TDB_SUSPEND) != 0));
13698638fe7bSKonstantin Belousov }
13708638fe7bSKonstantin Belousov 
137144990b8cSJulian Elischer /*
137244990b8cSJulian Elischer  * Called in from locations that can safely check to see
137344990b8cSJulian Elischer  * whether we have to suspend or at least throttle for a
137444990b8cSJulian Elischer  * single-thread event (e.g. fork).
137544990b8cSJulian Elischer  *
137644990b8cSJulian Elischer  * Such locations include userret().
137744990b8cSJulian Elischer  * If the "return_instead" argument is non zero, the thread must be able to
137844990b8cSJulian Elischer  * accept 0 (caller may continue), or 1 (caller must abort) as a result.
137944990b8cSJulian Elischer  *
138044990b8cSJulian Elischer  * The 'return_instead' argument tells the function if it may do a
138144990b8cSJulian Elischer  * thread_exit() or suspend, or whether the caller must abort and back
138244990b8cSJulian Elischer  * out instead.
138344990b8cSJulian Elischer  *
138444990b8cSJulian Elischer  * If the thread that set the single_threading request has set the
138544990b8cSJulian Elischer  * P_SINGLE_EXIT bit in the process flags then this call will never return
138644990b8cSJulian Elischer  * if 'return_instead' is false, but will exit.
138744990b8cSJulian Elischer  *
138844990b8cSJulian Elischer  * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
138944990b8cSJulian Elischer  *---------------+--------------------+---------------------
139044990b8cSJulian Elischer  *       0       | returns 0          |   returns 0 or 1
1391353374b5SJohn Baldwin  *               | when ST ends       |   immediately
139244990b8cSJulian Elischer  *---------------+--------------------+---------------------
139344990b8cSJulian Elischer  *       1       | thread exits       |   returns 1
1394353374b5SJohn Baldwin  *               |                    |  immediately
139544990b8cSJulian Elischer  * 0 = thread_exit() or suspension ok,
139644990b8cSJulian Elischer  * other = return error instead of stopping the thread.
139744990b8cSJulian Elischer  *
139844990b8cSJulian Elischer  * While a full suspension is under effect, even a single threading
139944990b8cSJulian Elischer  * thread would be suspended if it made this call (but it shouldn't).
140044990b8cSJulian Elischer  * This call should only be made from places where
140144990b8cSJulian Elischer  * thread_exit() would be safe as that may be the outcome unless
140244990b8cSJulian Elischer  * return_instead is set.
140344990b8cSJulian Elischer  */
140444990b8cSJulian Elischer int
140544990b8cSJulian Elischer thread_suspend_check(int return_instead)
140644990b8cSJulian Elischer {
1407ecafb24bSJuli Mallett 	struct thread *td;
1408ecafb24bSJuli Mallett 	struct proc *p;
140946e47c4fSKonstantin Belousov 	int wakeup_swapper;
141044990b8cSJulian Elischer 
141144990b8cSJulian Elischer 	td = curthread;
141244990b8cSJulian Elischer 	p = td->td_proc;
141337814395SPeter Wemm 	mtx_assert(&Giant, MA_NOTOWNED);
141444990b8cSJulian Elischer 	PROC_LOCK_ASSERT(p, MA_OWNED);
14158638fe7bSKonstantin Belousov 	while (thread_suspend_check_needed()) {
14161279572aSDavid Xu 		if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
141744990b8cSJulian Elischer 			KASSERT(p->p_singlethread != NULL,
141844990b8cSJulian Elischer 			    ("singlethread not set"));
141944990b8cSJulian Elischer 			/*
1420e3b9bf71SJulian Elischer 			 * The only suspension in action is a
1421e3b9bf71SJulian Elischer 			 * single-threading. Single threader need not stop.
1422bd07998eSKonstantin Belousov 			 * It is safe to access p->p_singlethread unlocked
1423bd07998eSKonstantin Belousov 			 * because it can only be set to our address by us.
142444990b8cSJulian Elischer 			 */
1425e3b9bf71SJulian Elischer 			if (p->p_singlethread == td)
142644990b8cSJulian Elischer 				return (0);	/* Exempt from stopping. */
142744990b8cSJulian Elischer 		}
142845a4bfa1SDavid Xu 		if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
142994f0972bSDavid Xu 			return (EINTR);
143044990b8cSJulian Elischer 
1431906ac69dSDavid Xu 		/* Should we goto user boundary if we didn't come from there? */
1432906ac69dSDavid Xu 		if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1433906ac69dSDavid Xu 		    (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
143494f0972bSDavid Xu 			return (ERESTART);
1435906ac69dSDavid Xu 
143644990b8cSJulian Elischer 		/*
14373077f938SKonstantin Belousov 		 * Ignore suspend requests if they are deferred.
1438d071a6faSJohn Baldwin 		 */
14393077f938SKonstantin Belousov 		if ((td->td_flags & TDF_SBDRY) != 0) {
1440d071a6faSJohn Baldwin 			KASSERT(return_instead,
1441d071a6faSJohn Baldwin 			    ("TDF_SBDRY set for unsafe thread_suspend_check"));
144246e47c4fSKonstantin Belousov 			KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
144346e47c4fSKonstantin Belousov 			    (TDF_SEINTR | TDF_SERESTART),
144446e47c4fSKonstantin Belousov 			    ("both TDF_SEINTR and TDF_SERESTART"));
144546e47c4fSKonstantin Belousov 			return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0);
1446d071a6faSJohn Baldwin 		}
1447d071a6faSJohn Baldwin 
1448d071a6faSJohn Baldwin 		/*
144944990b8cSJulian Elischer 		 * If the process is waiting for us to exit,
145044990b8cSJulian Elischer 		 * this thread should just suicide.
14511279572aSDavid Xu 		 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
145244990b8cSJulian Elischer 		 */
1453cf7d9a8cSDavid Xu 		if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
1454cf7d9a8cSDavid Xu 			PROC_UNLOCK(p);
145591d1786fSDmitry Chagin 
145691d1786fSDmitry Chagin 			/*
145791d1786fSDmitry Chagin 			 * Allow Linux emulation layer to do some work
145891d1786fSDmitry Chagin 			 * before thread suicide.
145991d1786fSDmitry Chagin 			 */
146091d1786fSDmitry Chagin 			if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
146191d1786fSDmitry Chagin 				(p->p_sysent->sv_thread_detach)(td);
14622a339d9eSKonstantin Belousov 			umtx_thread_exit(td);
1463d1e7a4a5SJohn Baldwin 			kern_thr_exit(td);
1464d1e7a4a5SJohn Baldwin 			panic("stopped thread did not exit");
1465cf7d9a8cSDavid Xu 		}
146621ecd1e9SDavid Xu 
146721ecd1e9SDavid Xu 		PROC_SLOCK(p);
146821ecd1e9SDavid Xu 		thread_stopped(p);
1469a54e85fdSJeff Roberson 		if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
1470a54e85fdSJeff Roberson 			if (p->p_numthreads == p->p_suspcount + 1) {
1471a54e85fdSJeff Roberson 				thread_lock(p->p_singlethread);
147284cdea97SKonstantin Belousov 				wakeup_swapper = thread_unsuspend_one(
147384cdea97SKonstantin Belousov 				    p->p_singlethread, p, false);
14747847a9daSJohn Baldwin 				if (wakeup_swapper)
14757847a9daSJohn Baldwin 					kick_proc0();
1476a54e85fdSJeff Roberson 			}
1477a54e85fdSJeff Roberson 		}
14783f9be10eSDavid Xu 		PROC_UNLOCK(p);
14797b4a950aSDavid Xu 		thread_lock(td);
148044990b8cSJulian Elischer 		/*
148144990b8cSJulian Elischer 		 * When a thread suspends, it just
1482ad1e7d28SJulian Elischer 		 * gets taken off all queues.
148344990b8cSJulian Elischer 		 */
148471fad9fdSJulian Elischer 		thread_suspend_one(td);
1485906ac69dSDavid Xu 		if (return_instead == 0) {
1486906ac69dSDavid Xu 			p->p_boundary_count++;
1487906ac69dSDavid Xu 			td->td_flags |= TDF_BOUNDARY;
1488cf19bf91SJulian Elischer 		}
14897b4a950aSDavid Xu 		PROC_SUNLOCK(p);
1490686bcb5cSJeff Roberson 		mi_switch(SW_INVOL | SWT_SUSPEND);
149144990b8cSJulian Elischer 		PROC_LOCK(p);
149244990b8cSJulian Elischer 	}
149344990b8cSJulian Elischer 	return (0);
149444990b8cSJulian Elischer }
149544990b8cSJulian Elischer 
1496478ca4b0SKonstantin Belousov /*
1497478ca4b0SKonstantin Belousov  * Check for possible stops and suspensions while executing a
1498478ca4b0SKonstantin Belousov  * casueword or similar transiently failing operation.
1499478ca4b0SKonstantin Belousov  *
1500478ca4b0SKonstantin Belousov  * The sleep argument controls whether the function can handle a stop
1501478ca4b0SKonstantin Belousov  * request itself or it should return ERESTART and the request is
1502478ca4b0SKonstantin Belousov  * proceed at the kernel/user boundary in ast.
1503478ca4b0SKonstantin Belousov  *
1504478ca4b0SKonstantin Belousov  * Typically, when retrying due to casueword(9) failure (rv == 1), we
1505478ca4b0SKonstantin Belousov  * should handle the stop requests there, with exception of cases when
1506478ca4b0SKonstantin Belousov  * the thread owns a kernel resource, for instance busied the umtx
1507300b525dSKonstantin Belousov  * key, or when functions return immediately if thread_check_susp()
1508478ca4b0SKonstantin Belousov  * returned non-zero.  On the other hand, retrying the whole lock
1509478ca4b0SKonstantin Belousov  * operation, we better not stop there but delegate the handling to
1510478ca4b0SKonstantin Belousov  * ast.
1511478ca4b0SKonstantin Belousov  *
1512478ca4b0SKonstantin Belousov  * If the request is for thread termination P_SINGLE_EXIT, we cannot
1513478ca4b0SKonstantin Belousov  * handle it at all, and simply return EINTR.
1514478ca4b0SKonstantin Belousov  */
1515478ca4b0SKonstantin Belousov int
1516478ca4b0SKonstantin Belousov thread_check_susp(struct thread *td, bool sleep)
1517478ca4b0SKonstantin Belousov {
1518478ca4b0SKonstantin Belousov 	struct proc *p;
1519478ca4b0SKonstantin Belousov 	int error;
1520478ca4b0SKonstantin Belousov 
1521478ca4b0SKonstantin Belousov 	/*
1522c6d31b83SKonstantin Belousov 	 * The check for TDA_SUSPEND is racy, but it is enough to
1523478ca4b0SKonstantin Belousov 	 * eventually break the lockstep loop.
1524478ca4b0SKonstantin Belousov 	 */
1525c6d31b83SKonstantin Belousov 	if (!td_ast_pending(td, TDA_SUSPEND))
1526478ca4b0SKonstantin Belousov 		return (0);
1527478ca4b0SKonstantin Belousov 	error = 0;
1528478ca4b0SKonstantin Belousov 	p = td->td_proc;
1529478ca4b0SKonstantin Belousov 	PROC_LOCK(p);
1530478ca4b0SKonstantin Belousov 	if (p->p_flag & P_SINGLE_EXIT)
1531478ca4b0SKonstantin Belousov 		error = EINTR;
1532478ca4b0SKonstantin Belousov 	else if (P_SHOULDSTOP(p) ||
1533478ca4b0SKonstantin Belousov 	    ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND)))
1534478ca4b0SKonstantin Belousov 		error = sleep ? thread_suspend_check(0) : ERESTART;
1535478ca4b0SKonstantin Belousov 	PROC_UNLOCK(p);
1536478ca4b0SKonstantin Belousov 	return (error);
1537478ca4b0SKonstantin Belousov }
1538478ca4b0SKonstantin Belousov 
153935c32a76SDavid Xu void
15406ddcc233SKonstantin Belousov thread_suspend_switch(struct thread *td, struct proc *p)
1541a54e85fdSJeff Roberson {
1542a54e85fdSJeff Roberson 
1543a54e85fdSJeff Roberson 	KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1544a54e85fdSJeff Roberson 	PROC_LOCK_ASSERT(p, MA_OWNED);
15457b4a950aSDavid Xu 	PROC_SLOCK_ASSERT(p, MA_OWNED);
1546a54e85fdSJeff Roberson 	/*
1547a54e85fdSJeff Roberson 	 * We implement thread_suspend_one in stages here to avoid
1548a54e85fdSJeff Roberson 	 * dropping the proc lock while the thread lock is owned.
1549a54e85fdSJeff Roberson 	 */
15506ddcc233SKonstantin Belousov 	if (p == td->td_proc) {
1551a54e85fdSJeff Roberson 		thread_stopped(p);
1552a54e85fdSJeff Roberson 		p->p_suspcount++;
15536ddcc233SKonstantin Belousov 	}
15543f9be10eSDavid Xu 	PROC_UNLOCK(p);
15557b4a950aSDavid Xu 	thread_lock(td);
1556c6d31b83SKonstantin Belousov 	ast_unsched_locked(td, TDA_SUSPEND);
1557a54e85fdSJeff Roberson 	TD_SET_SUSPENDED(td);
1558c5aa6b58SJeff Roberson 	sched_sleep(td, 0);
15597b4a950aSDavid Xu 	PROC_SUNLOCK(p);
1560a54e85fdSJeff Roberson 	DROP_GIANT();
1561686bcb5cSJeff Roberson 	mi_switch(SW_VOL | SWT_SUSPEND);
1562a54e85fdSJeff Roberson 	PICKUP_GIANT();
1563a54e85fdSJeff Roberson 	PROC_LOCK(p);
15647b4a950aSDavid Xu 	PROC_SLOCK(p);
1565a54e85fdSJeff Roberson }
1566a54e85fdSJeff Roberson 
1567a54e85fdSJeff Roberson void
156835c32a76SDavid Xu thread_suspend_one(struct thread *td)
156935c32a76SDavid Xu {
15706ddcc233SKonstantin Belousov 	struct proc *p;
157135c32a76SDavid Xu 
15726ddcc233SKonstantin Belousov 	p = td->td_proc;
15737b4a950aSDavid Xu 	PROC_SLOCK_ASSERT(p, MA_OWNED);
1574a54e85fdSJeff Roberson 	THREAD_LOCK_ASSERT(td, MA_OWNED);
1575e574e444SDavid Xu 	KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
157635c32a76SDavid Xu 	p->p_suspcount++;
1577c6d31b83SKonstantin Belousov 	ast_unsched_locked(td, TDA_SUSPEND);
157871fad9fdSJulian Elischer 	TD_SET_SUSPENDED(td);
1579c5aa6b58SJeff Roberson 	sched_sleep(td, 0);
158035c32a76SDavid Xu }
158135c32a76SDavid Xu 
158284cdea97SKonstantin Belousov static int
158384cdea97SKonstantin Belousov thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
158435c32a76SDavid Xu {
158535c32a76SDavid Xu 
1586a54e85fdSJeff Roberson 	THREAD_LOCK_ASSERT(td, MA_OWNED);
1587ad1e7d28SJulian Elischer 	KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
158871fad9fdSJulian Elischer 	TD_CLR_SUSPENDED(td);
15896ddcc233SKonstantin Belousov 	td->td_flags &= ~TDF_ALLPROCSUSP;
15906ddcc233SKonstantin Belousov 	if (td->td_proc == p) {
15916ddcc233SKonstantin Belousov 		PROC_SLOCK_ASSERT(p, MA_OWNED);
159235c32a76SDavid Xu 		p->p_suspcount--;
159384cdea97SKonstantin Belousov 		if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
159484cdea97SKonstantin Belousov 			td->td_flags &= ~TDF_BOUNDARY;
159584cdea97SKonstantin Belousov 			p->p_boundary_count--;
159684cdea97SKonstantin Belousov 		}
15976ddcc233SKonstantin Belousov 	}
159861a74c5cSJeff Roberson 	return (setrunnable(td, 0));
159935c32a76SDavid Xu }
160035c32a76SDavid Xu 
1601af928fdeSKonstantin Belousov void
1602af928fdeSKonstantin Belousov thread_run_flash(struct thread *td)
1603af928fdeSKonstantin Belousov {
1604af928fdeSKonstantin Belousov 	struct proc *p;
1605af928fdeSKonstantin Belousov 
1606af928fdeSKonstantin Belousov 	p = td->td_proc;
1607af928fdeSKonstantin Belousov 	PROC_LOCK_ASSERT(p, MA_OWNED);
1608af928fdeSKonstantin Belousov 
1609af928fdeSKonstantin Belousov 	if (TD_ON_SLEEPQ(td))
1610af928fdeSKonstantin Belousov 		sleepq_remove_nested(td);
1611af928fdeSKonstantin Belousov 	else
1612af928fdeSKonstantin Belousov 		thread_lock(td);
1613af928fdeSKonstantin Belousov 
1614af928fdeSKonstantin Belousov 	THREAD_LOCK_ASSERT(td, MA_OWNED);
1615af928fdeSKonstantin Belousov 	KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
1616af928fdeSKonstantin Belousov 
1617af928fdeSKonstantin Belousov 	TD_CLR_SUSPENDED(td);
1618af928fdeSKonstantin Belousov 	PROC_SLOCK(p);
1619af928fdeSKonstantin Belousov 	MPASS(p->p_suspcount > 0);
1620af928fdeSKonstantin Belousov 	p->p_suspcount--;
1621af928fdeSKonstantin Belousov 	PROC_SUNLOCK(p);
1622af928fdeSKonstantin Belousov 	if (setrunnable(td, 0))
1623af928fdeSKonstantin Belousov 		kick_proc0();
1624af928fdeSKonstantin Belousov }
1625af928fdeSKonstantin Belousov 
162644990b8cSJulian Elischer /*
162744990b8cSJulian Elischer  * Allow all threads blocked by single threading to continue running.
162844990b8cSJulian Elischer  */
162944990b8cSJulian Elischer void
163044990b8cSJulian Elischer thread_unsuspend(struct proc *p)
163144990b8cSJulian Elischer {
163244990b8cSJulian Elischer 	struct thread *td;
16337847a9daSJohn Baldwin 	int wakeup_swapper;
163444990b8cSJulian Elischer 
163544990b8cSJulian Elischer 	PROC_LOCK_ASSERT(p, MA_OWNED);
16367b4a950aSDavid Xu 	PROC_SLOCK_ASSERT(p, MA_OWNED);
16377847a9daSJohn Baldwin 	wakeup_swapper = 0;
163844990b8cSJulian Elischer 	if (!P_SHOULDSTOP(p)) {
1639ad1e7d28SJulian Elischer                 FOREACH_THREAD_IN_PROC(p, td) {
1640a54e85fdSJeff Roberson 			thread_lock(td);
1641f829268bSKonstantin Belousov 			if (TD_IS_SUSPENDED(td))
164284cdea97SKonstantin Belousov 				wakeup_swapper |= thread_unsuspend_one(td, p,
164384cdea97SKonstantin Belousov 				    true);
1644f829268bSKonstantin Belousov 			else
1645a54e85fdSJeff Roberson 				thread_unlock(td);
1646ad1e7d28SJulian Elischer 		}
164784cdea97SKonstantin Belousov 	} else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
164884cdea97SKonstantin Belousov 	    p->p_numthreads == p->p_suspcount) {
164944990b8cSJulian Elischer 		/*
165044990b8cSJulian Elischer 		 * Stopping everything also did the job for the single
165144990b8cSJulian Elischer 		 * threading request. Now we've downgraded to single-threaded,
165244990b8cSJulian Elischer 		 * let it continue.
165344990b8cSJulian Elischer 		 */
16546ddcc233SKonstantin Belousov 		if (p->p_singlethread->td_proc == p) {
1655a54e85fdSJeff Roberson 			thread_lock(p->p_singlethread);
16566ddcc233SKonstantin Belousov 			wakeup_swapper = thread_unsuspend_one(
165784cdea97SKonstantin Belousov 			    p->p_singlethread, p, false);
165844990b8cSJulian Elischer 		}
16596ddcc233SKonstantin Belousov 	}
16607847a9daSJohn Baldwin 	if (wakeup_swapper)
16617847a9daSJohn Baldwin 		kick_proc0();
166244990b8cSJulian Elischer }
166344990b8cSJulian Elischer 
1664ed062c8dSJulian Elischer /*
1665ed062c8dSJulian Elischer  * End the single threading mode..
1666ed062c8dSJulian Elischer  */
166744990b8cSJulian Elischer void
16686ddcc233SKonstantin Belousov thread_single_end(struct proc *p, int mode)
166944990b8cSJulian Elischer {
167044990b8cSJulian Elischer 	struct thread *td;
16717847a9daSJohn Baldwin 	int wakeup_swapper;
167244990b8cSJulian Elischer 
16736ddcc233SKonstantin Belousov 	KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
16746ddcc233SKonstantin Belousov 	    mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
16756ddcc233SKonstantin Belousov 	    ("invalid mode %d", mode));
167644990b8cSJulian Elischer 	PROC_LOCK_ASSERT(p, MA_OWNED);
16776ddcc233SKonstantin Belousov 	KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) ||
16786ddcc233SKonstantin Belousov 	    (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0),
16796ddcc233SKonstantin Belousov 	    ("mode %d does not match P_TOTAL_STOP", mode));
168084cdea97SKonstantin Belousov 	KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread,
168184cdea97SKonstantin Belousov 	    ("thread_single_end from other thread %p %p",
168284cdea97SKonstantin Belousov 	    curthread, p->p_singlethread));
168384cdea97SKonstantin Belousov 	KASSERT(mode != SINGLE_BOUNDARY ||
168484cdea97SKonstantin Belousov 	    (p->p_flag & P_SINGLE_BOUNDARY) != 0,
168584cdea97SKonstantin Belousov 	    ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag));
16866ddcc233SKonstantin Belousov 	p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY |
16876ddcc233SKonstantin Belousov 	    P_TOTAL_STOP);
16887b4a950aSDavid Xu 	PROC_SLOCK(p);
168944990b8cSJulian Elischer 	p->p_singlethread = NULL;
16907847a9daSJohn Baldwin 	wakeup_swapper = 0;
169149539972SJulian Elischer 	/*
16927847a9daSJohn Baldwin 	 * If there are other threads they may now run,
169349539972SJulian Elischer 	 * unless of course there is a blanket 'stop order'
169449539972SJulian Elischer 	 * on the process. The single threader must be allowed
169549539972SJulian Elischer 	 * to continue however as this is a bad place to stop.
169649539972SJulian Elischer 	 */
16976ddcc233SKonstantin Belousov 	if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1698ad1e7d28SJulian Elischer                 FOREACH_THREAD_IN_PROC(p, td) {
1699a54e85fdSJeff Roberson 			thread_lock(td);
1700ad1e7d28SJulian Elischer 			if (TD_IS_SUSPENDED(td)) {
170184cdea97SKonstantin Belousov 				wakeup_swapper |= thread_unsuspend_one(td, p,
17022d5ef216SMark Johnston 				    true);
170361a74c5cSJeff Roberson 			} else
1704a54e85fdSJeff Roberson 				thread_unlock(td);
170549539972SJulian Elischer 		}
1706ad1e7d28SJulian Elischer 	}
170784cdea97SKonstantin Belousov 	KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
170884cdea97SKonstantin Belousov 	    ("inconsistent boundary count %d", p->p_boundary_count));
17097b4a950aSDavid Xu 	PROC_SUNLOCK(p);
17107847a9daSJohn Baldwin 	if (wakeup_swapper)
17117847a9daSJohn Baldwin 		kick_proc0();
1712d7a9e6e7SKonstantin Belousov 	wakeup(&p->p_flag);
171349539972SJulian Elischer }
17144fc21c09SDaniel Eischen 
1715aae3547bSMateusz Guzik /*
1716aae3547bSMateusz Guzik  * Locate a thread by number and return with proc lock held.
1717aae3547bSMateusz Guzik  *
1718aae3547bSMateusz Guzik  * thread exit establishes proc -> tidhash lock ordering, but lookup
1719aae3547bSMateusz Guzik  * takes tidhash first and needs to return locked proc.
1720aae3547bSMateusz Guzik  *
1721aae3547bSMateusz Guzik  * The problem is worked around by relying on type-safety of both
1722aae3547bSMateusz Guzik  * structures and doing the work in 2 steps:
1723aae3547bSMateusz Guzik  * - tidhash-locked lookup which saves both thread and proc pointers
1724aae3547bSMateusz Guzik  * - proc-locked verification that the found thread still matches
1725aae3547bSMateusz Guzik  */
1726aae3547bSMateusz Guzik static bool
1727aae3547bSMateusz Guzik tdfind_hash(lwpid_t tid, pid_t pid, struct proc **pp, struct thread **tdp)
1728cf7d9a8cSDavid Xu {
1729cf7d9a8cSDavid Xu #define RUN_THRESH	16
1730aae3547bSMateusz Guzik 	struct proc *p;
1731cf7d9a8cSDavid Xu 	struct thread *td;
1732aae3547bSMateusz Guzik 	int run;
1733aae3547bSMateusz Guzik 	bool locked;
1734cf7d9a8cSDavid Xu 
1735aae3547bSMateusz Guzik 	run = 0;
173626007fe3SMateusz Guzik 	rw_rlock(TIDHASHLOCK(tid));
1737aae3547bSMateusz Guzik 	locked = true;
1738cf7d9a8cSDavid Xu 	LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1739aae3547bSMateusz Guzik 		if (td->td_tid != tid) {
1740aae3547bSMateusz Guzik 			run++;
1741aae3547bSMateusz Guzik 			continue;
1742cf7d9a8cSDavid Xu 		}
1743aae3547bSMateusz Guzik 		p = td->td_proc;
1744aae3547bSMateusz Guzik 		if (pid != -1 && p->p_pid != pid) {
1745cf7d9a8cSDavid Xu 			td = NULL;
1746cf7d9a8cSDavid Xu 			break;
1747cf7d9a8cSDavid Xu 		}
1748cf7d9a8cSDavid Xu 		if (run > RUN_THRESH) {
174926007fe3SMateusz Guzik 			if (rw_try_upgrade(TIDHASHLOCK(tid))) {
1750cf7d9a8cSDavid Xu 				LIST_REMOVE(td, td_hash);
1751cf7d9a8cSDavid Xu 				LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1752cf7d9a8cSDavid Xu 					td, td_hash);
175326007fe3SMateusz Guzik 				rw_wunlock(TIDHASHLOCK(tid));
1754aae3547bSMateusz Guzik 				locked = false;
1755aae3547bSMateusz Guzik 				break;
1756cf7d9a8cSDavid Xu 			}
1757cf7d9a8cSDavid Xu 		}
1758cf7d9a8cSDavid Xu 		break;
1759cf7d9a8cSDavid Xu 	}
1760aae3547bSMateusz Guzik 	if (locked)
176126007fe3SMateusz Guzik 		rw_runlock(TIDHASHLOCK(tid));
1762aae3547bSMateusz Guzik 	if (td == NULL)
1763aae3547bSMateusz Guzik 		return (false);
1764aae3547bSMateusz Guzik 	*pp = p;
1765aae3547bSMateusz Guzik 	*tdp = td;
1766aae3547bSMateusz Guzik 	return (true);
1767aae3547bSMateusz Guzik }
1768aae3547bSMateusz Guzik 
1769aae3547bSMateusz Guzik struct thread *
1770aae3547bSMateusz Guzik tdfind(lwpid_t tid, pid_t pid)
1771aae3547bSMateusz Guzik {
1772aae3547bSMateusz Guzik 	struct proc *p;
1773aae3547bSMateusz Guzik 	struct thread *td;
1774aae3547bSMateusz Guzik 
1775aae3547bSMateusz Guzik 	td = curthread;
1776aae3547bSMateusz Guzik 	if (td->td_tid == tid) {
1777aae3547bSMateusz Guzik 		if (pid != -1 && td->td_proc->p_pid != pid)
1778aae3547bSMateusz Guzik 			return (NULL);
1779aae3547bSMateusz Guzik 		PROC_LOCK(td->td_proc);
1780cf7d9a8cSDavid Xu 		return (td);
1781cf7d9a8cSDavid Xu 	}
1782cf7d9a8cSDavid Xu 
1783aae3547bSMateusz Guzik 	for (;;) {
1784aae3547bSMateusz Guzik 		if (!tdfind_hash(tid, pid, &p, &td))
1785aae3547bSMateusz Guzik 			return (NULL);
1786aae3547bSMateusz Guzik 		PROC_LOCK(p);
1787aae3547bSMateusz Guzik 		if (td->td_tid != tid) {
1788aae3547bSMateusz Guzik 			PROC_UNLOCK(p);
1789aae3547bSMateusz Guzik 			continue;
1790aae3547bSMateusz Guzik 		}
1791aae3547bSMateusz Guzik 		if (td->td_proc != p) {
1792aae3547bSMateusz Guzik 			PROC_UNLOCK(p);
1793aae3547bSMateusz Guzik 			continue;
1794aae3547bSMateusz Guzik 		}
1795aae3547bSMateusz Guzik 		if (p->p_state == PRS_NEW) {
1796aae3547bSMateusz Guzik 			PROC_UNLOCK(p);
1797aae3547bSMateusz Guzik 			return (NULL);
1798aae3547bSMateusz Guzik 		}
1799aae3547bSMateusz Guzik 		return (td);
1800aae3547bSMateusz Guzik 	}
1801aae3547bSMateusz Guzik }
1802aae3547bSMateusz Guzik 
1803cf7d9a8cSDavid Xu void
1804cf7d9a8cSDavid Xu tidhash_add(struct thread *td)
1805cf7d9a8cSDavid Xu {
180626007fe3SMateusz Guzik 	rw_wlock(TIDHASHLOCK(td->td_tid));
1807cf7d9a8cSDavid Xu 	LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
180826007fe3SMateusz Guzik 	rw_wunlock(TIDHASHLOCK(td->td_tid));
1809cf7d9a8cSDavid Xu }
1810cf7d9a8cSDavid Xu 
1811cf7d9a8cSDavid Xu void
1812cf7d9a8cSDavid Xu tidhash_remove(struct thread *td)
1813cf7d9a8cSDavid Xu {
181426007fe3SMateusz Guzik 
181526007fe3SMateusz Guzik 	rw_wlock(TIDHASHLOCK(td->td_tid));
1816cf7d9a8cSDavid Xu 	LIST_REMOVE(td, td_hash);
181726007fe3SMateusz Guzik 	rw_wunlock(TIDHASHLOCK(td->td_tid));
1818cf7d9a8cSDavid Xu }
1819