kern_umtx.c revision 140245
1139804Simp/*- 2139013Sdavidxu * Copyright (c) 2004, David Xu <davidxu@freebsd.org> 3112904Sjeff * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org> 4112904Sjeff * All rights reserved. 5112904Sjeff * 6112904Sjeff * Redistribution and use in source and binary forms, with or without 7112904Sjeff * modification, are permitted provided that the following conditions 8112904Sjeff * are met: 9112904Sjeff * 1. Redistributions of source code must retain the above copyright 10112904Sjeff * notice unmodified, this list of conditions, and the following 11112904Sjeff * disclaimer. 12112904Sjeff * 2. Redistributions in binary form must reproduce the above copyright 13112904Sjeff * notice, this list of conditions and the following disclaimer in the 14112904Sjeff * documentation and/or other materials provided with the distribution. 15112904Sjeff * 16112904Sjeff * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17112904Sjeff * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18112904Sjeff * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19112904Sjeff * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20112904Sjeff * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21112904Sjeff * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22112904Sjeff * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23112904Sjeff * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24112904Sjeff * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25112904Sjeff * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26112904Sjeff */ 27112904Sjeff 28116182Sobrien#include <sys/cdefs.h> 29116182Sobrien__FBSDID("$FreeBSD: head/sys/kern/kern_umtx.c 140245 2005-01-14 13:38:15Z davidxu $"); 30116182Sobrien 31112904Sjeff#include <sys/param.h> 32112904Sjeff#include <sys/kernel.h> 33131431Smarcel#include <sys/limits.h> 34112904Sjeff#include <sys/lock.h> 35115765Sjeff#include <sys/malloc.h> 36112904Sjeff#include <sys/mutex.h> 37112904Sjeff#include <sys/proc.h> 38112904Sjeff#include <sys/sysent.h> 39112904Sjeff#include <sys/systm.h> 40112904Sjeff#include <sys/sysproto.h> 41139013Sdavidxu#include <sys/eventhandler.h> 42112904Sjeff#include <sys/thr.h> 43112904Sjeff#include <sys/umtx.h> 44112904Sjeff 45139013Sdavidxu#include <vm/vm.h> 46139013Sdavidxu#include <vm/vm_param.h> 47139013Sdavidxu#include <vm/pmap.h> 48139013Sdavidxu#include <vm/vm_map.h> 49139013Sdavidxu#include <vm/vm_object.h> 50139013Sdavidxu 51139013Sdavidxu#define UMTX_PRIVATE 0 52139013Sdavidxu#define UMTX_SHARED 1 53139013Sdavidxu 54139013Sdavidxu#define UMTX_STATIC_SHARED 55139013Sdavidxu 56139013Sdavidxustruct umtx_key { 57139013Sdavidxu int type; 58139013Sdavidxu union { 59139013Sdavidxu struct { 60139013Sdavidxu vm_object_t object; 61139013Sdavidxu long offset; 62139013Sdavidxu } shared; 63139013Sdavidxu struct { 64139013Sdavidxu struct umtx *umtx; 65139013Sdavidxu long pid; 66139013Sdavidxu } private; 67139013Sdavidxu struct { 68139013Sdavidxu void *ptr; 69139013Sdavidxu long word; 70139013Sdavidxu } both; 71139013Sdavidxu } info; 72139013Sdavidxu}; 73139013Sdavidxu 74115765Sjeffstruct umtx_q { 75115765Sjeff LIST_ENTRY(umtx_q) uq_next; /* Linked list for the hash. */ 76139013Sdavidxu struct umtx_key uq_key; /* Umtx key. */ 77139257Sdavidxu struct thread *uq_thread; /* The thread waits on. */ 78139013Sdavidxu LIST_ENTRY(umtx_q) uq_rqnext; /* Linked list for requeuing. */ 79139013Sdavidxu vm_offset_t uq_addr; /* Umtx's virtual address. */ 80115765Sjeff}; 81115765Sjeff 82115765SjeffLIST_HEAD(umtx_head, umtx_q); 83138224Sdavidxustruct umtxq_chain { 84139013Sdavidxu struct mtx uc_lock; /* Lock for this chain. */ 85139013Sdavidxu struct umtx_head uc_queue; /* List of sleep queues. */ 86139257Sdavidxu#define UCF_BUSY 0x01 87139257Sdavidxu#define UCF_WANT 0x02 88139257Sdavidxu int uc_flags; 89138224Sdavidxu}; 90115765Sjeff 91138224Sdavidxu#define GOLDEN_RATIO_PRIME 2654404609U 92138224Sdavidxu#define UMTX_CHAINS 128 93138224Sdavidxu#define UMTX_SHIFTS (__WORD_BIT - 7) 94115765Sjeff 95138224Sdavidxustatic struct umtxq_chain umtxq_chains[UMTX_CHAINS]; 96138224Sdavidxustatic MALLOC_DEFINE(M_UMTX, "umtx", "UMTX queue memory"); 97115310Sjeff 98139013Sdavidxustatic void umtxq_init_chains(void *); 99139013Sdavidxustatic int umtxq_hash(struct umtx_key *key); 100139013Sdavidxustatic struct mtx *umtxq_mtx(int chain); 101139013Sdavidxustatic void umtxq_lock(struct umtx_key *key); 102139013Sdavidxustatic void umtxq_unlock(struct umtx_key *key); 103139257Sdavidxustatic void umtxq_busy(struct umtx_key *key); 104139257Sdavidxustatic void umtxq_unbusy(struct umtx_key *key); 105139013Sdavidxustatic void umtxq_insert(struct umtx_q *uq); 106139013Sdavidxustatic void umtxq_remove(struct umtx_q *uq); 107139013Sdavidxustatic int umtxq_sleep(struct thread *td, struct umtx_key *key, 108139013Sdavidxu int prio, const char *wmesg, int timo); 109139257Sdavidxustatic int umtxq_count(struct umtx_key *key); 110139257Sdavidxustatic int umtxq_signal(struct umtx_key *key, int nr_wakeup); 111139013Sdavidxu#ifdef UMTX_DYNAMIC_SHARED 112139013Sdavidxustatic void fork_handler(void *arg, struct proc *p1, struct proc *p2, 113139013Sdavidxu int flags); 114139013Sdavidxu#endif 115139013Sdavidxustatic int umtx_key_match(const struct umtx_key *k1, const struct umtx_key *k2); 116139013Sdavidxustatic int umtx_key_get(struct thread *td, struct umtx *umtx, 117139013Sdavidxu struct umtx_key *key); 118139013Sdavidxustatic void umtx_key_release(struct umtx_key *key); 119115310Sjeff 120139013SdavidxuSYSINIT(umtx, SI_SUB_EVENTHANDLER+1, SI_ORDER_MIDDLE, umtxq_init_chains, NULL); 121138224Sdavidxu 122138224Sdavidxustatic void 123139013Sdavidxuumtxq_init_chains(void *arg __unused) 124138224Sdavidxu{ 125138224Sdavidxu int i; 126138224Sdavidxu 127138224Sdavidxu for (i = 0; i < UMTX_CHAINS; ++i) { 128138224Sdavidxu mtx_init(&umtxq_chains[i].uc_lock, "umtxq_lock", NULL, 129138224Sdavidxu MTX_DEF | MTX_DUPOK); 130139013Sdavidxu LIST_INIT(&umtxq_chains[i].uc_queue); 131139257Sdavidxu umtxq_chains[i].uc_flags = 0; 132138224Sdavidxu } 133139013Sdavidxu#ifdef UMTX_DYNAMIC_SHARED 134139013Sdavidxu EVENTHANDLER_REGISTER(process_fork, fork_handler, 0, 10000); 135139013Sdavidxu#endif 136138224Sdavidxu} 137138224Sdavidxu 138138224Sdavidxustatic inline int 139139013Sdavidxuumtxq_hash(struct umtx_key *key) 140138224Sdavidxu{ 141139013Sdavidxu unsigned n = (uintptr_t)key->info.both.ptr + key->info.both.word; 142138224Sdavidxu return (((n * GOLDEN_RATIO_PRIME) >> UMTX_SHIFTS) % UMTX_CHAINS); 143138224Sdavidxu} 144138224Sdavidxu 145139013Sdavidxustatic inline int 146139013Sdavidxuumtx_key_match(const struct umtx_key *k1, const struct umtx_key *k2) 147139013Sdavidxu{ 148139013Sdavidxu return (k1->type == k2->type && 149139013Sdavidxu k1->info.both.ptr == k2->info.both.ptr && 150139013Sdavidxu k1->info.both.word == k2->info.both.word); 151139013Sdavidxu} 152139013Sdavidxu 153139013Sdavidxustatic inline struct mtx * 154139013Sdavidxuumtxq_mtx(int chain) 155139013Sdavidxu{ 156139013Sdavidxu return (&umtxq_chains[chain].uc_lock); 157139013Sdavidxu} 158139013Sdavidxu 159138224Sdavidxustatic inline void 160139257Sdavidxuumtxq_busy(struct umtx_key *key) 161139257Sdavidxu{ 162139257Sdavidxu int chain = umtxq_hash(key); 163139257Sdavidxu 164139257Sdavidxu mtx_assert(umtxq_mtx(chain), MA_OWNED); 165139257Sdavidxu while (umtxq_chains[chain].uc_flags & UCF_BUSY) { 166139257Sdavidxu umtxq_chains[chain].uc_flags |= UCF_WANT; 167139257Sdavidxu msleep(&umtxq_chains[chain], umtxq_mtx(chain), 168139257Sdavidxu curthread->td_priority, "umtxq_busy", 0); 169139257Sdavidxu } 170139257Sdavidxu umtxq_chains[chain].uc_flags |= UCF_BUSY; 171139257Sdavidxu} 172139257Sdavidxu 173139257Sdavidxustatic inline void 174139257Sdavidxuumtxq_unbusy(struct umtx_key *key) 175139257Sdavidxu{ 176139257Sdavidxu int chain = umtxq_hash(key); 177139257Sdavidxu 178139257Sdavidxu mtx_assert(umtxq_mtx(chain), MA_OWNED); 179139258Sdavidxu KASSERT(umtxq_chains[chain].uc_flags & UCF_BUSY, ("not busy")); 180139257Sdavidxu umtxq_chains[chain].uc_flags &= ~UCF_BUSY; 181139257Sdavidxu if (umtxq_chains[chain].uc_flags & UCF_WANT) { 182139257Sdavidxu umtxq_chains[chain].uc_flags &= ~UCF_WANT; 183139257Sdavidxu wakeup(&umtxq_chains[chain]); 184139257Sdavidxu } 185139257Sdavidxu} 186139257Sdavidxu 187139257Sdavidxustatic inline void 188139013Sdavidxuumtxq_lock(struct umtx_key *key) 189138224Sdavidxu{ 190139013Sdavidxu int chain = umtxq_hash(key); 191139013Sdavidxu mtx_lock(umtxq_mtx(chain)); 192138224Sdavidxu} 193138224Sdavidxu 194138225Sdavidxustatic inline void 195139013Sdavidxuumtxq_unlock(struct umtx_key *key) 196138224Sdavidxu{ 197139013Sdavidxu int chain = umtxq_hash(key); 198139013Sdavidxu mtx_unlock(umtxq_mtx(chain)); 199138224Sdavidxu} 200138224Sdavidxu 201139013Sdavidxu/* 202139013Sdavidxu * Insert a thread onto the umtx queue. 203139013Sdavidxu */ 204139013Sdavidxustatic inline void 205139013Sdavidxuumtxq_insert(struct umtx_q *uq) 206115765Sjeff{ 207115765Sjeff struct umtx_head *head; 208139013Sdavidxu int chain = umtxq_hash(&uq->uq_key); 209139013Sdavidxu 210139257Sdavidxu mtx_assert(umtxq_mtx(chain), MA_OWNED); 211139013Sdavidxu head = &umtxq_chains[chain].uc_queue; 212139013Sdavidxu LIST_INSERT_HEAD(head, uq, uq_next); 213139013Sdavidxu uq->uq_thread->td_umtxq = uq; 214139013Sdavidxu mtx_lock_spin(&sched_lock); 215139013Sdavidxu uq->uq_thread->td_flags |= TDF_UMTXQ; 216139013Sdavidxu mtx_unlock_spin(&sched_lock); 217139013Sdavidxu} 218139013Sdavidxu 219139013Sdavidxu/* 220139013Sdavidxu * Remove thread from the umtx queue. 221139013Sdavidxu */ 222139013Sdavidxustatic inline void 223139013Sdavidxuumtxq_remove(struct umtx_q *uq) 224139013Sdavidxu{ 225139257Sdavidxu mtx_assert(umtxq_mtx(umtxq_hash(&uq->uq_key)), MA_OWNED); 226139013Sdavidxu if (uq->uq_thread->td_flags & TDF_UMTXQ) { 227139013Sdavidxu LIST_REMOVE(uq, uq_next); 228139013Sdavidxu uq->uq_thread->td_umtxq = NULL; 229139013Sdavidxu /* turning off TDF_UMTXQ should be the last thing. */ 230139013Sdavidxu mtx_lock_spin(&sched_lock); 231139013Sdavidxu uq->uq_thread->td_flags &= ~TDF_UMTXQ; 232139013Sdavidxu mtx_unlock_spin(&sched_lock); 233139013Sdavidxu } 234139013Sdavidxu} 235139013Sdavidxu 236139013Sdavidxustatic int 237139013Sdavidxuumtxq_count(struct umtx_key *key) 238139013Sdavidxu{ 239115765Sjeff struct umtx_q *uq; 240139013Sdavidxu struct umtx_head *head; 241139013Sdavidxu int chain, count = 0; 242115765Sjeff 243139013Sdavidxu chain = umtxq_hash(key); 244139257Sdavidxu mtx_assert(umtxq_mtx(chain), MA_OWNED); 245139013Sdavidxu head = &umtxq_chains[chain].uc_queue; 246115765Sjeff LIST_FOREACH(uq, head, uq_next) { 247139013Sdavidxu if (umtx_key_match(&uq->uq_key, key)) { 248139013Sdavidxu if (++count > 1) 249139013Sdavidxu break; 250139013Sdavidxu } 251115765Sjeff } 252139013Sdavidxu return (count); 253115765Sjeff} 254115765Sjeff 255139257Sdavidxustatic int 256139257Sdavidxuumtxq_signal(struct umtx_key *key, int n_wake) 257115765Sjeff{ 258139257Sdavidxu struct umtx_q *uq, *next; 259115765Sjeff struct umtx_head *head; 260139013Sdavidxu struct thread *blocked = NULL; 261139257Sdavidxu int chain, ret; 262115765Sjeff 263139257Sdavidxu ret = 0; 264139013Sdavidxu chain = umtxq_hash(key); 265139257Sdavidxu mtx_assert(umtxq_mtx(chain), MA_OWNED); 266139013Sdavidxu head = &umtxq_chains[chain].uc_queue; 267139257Sdavidxu for (uq = LIST_FIRST(head); uq; uq = next) { 268139013Sdavidxu next = LIST_NEXT(uq, uq_next); 269139013Sdavidxu if (umtx_key_match(&uq->uq_key, key)) { 270139013Sdavidxu blocked = uq->uq_thread; 271139013Sdavidxu umtxq_remove(uq); 272139013Sdavidxu wakeup(blocked); 273139257Sdavidxu if (++ret >= n_wake) 274139257Sdavidxu break; 275139013Sdavidxu } 276139013Sdavidxu } 277139257Sdavidxu return (ret); 278138224Sdavidxu} 279138224Sdavidxu 280138224Sdavidxustatic inline int 281139013Sdavidxuumtxq_sleep(struct thread *td, struct umtx_key *key, int priority, 282139013Sdavidxu const char *wmesg, int timo) 283138224Sdavidxu{ 284139013Sdavidxu int chain = umtxq_hash(key); 285139751Sdavidxu int error = msleep(td, umtxq_mtx(chain), priority, wmesg, timo); 286139751Sdavidxu if (error == EWOULDBLOCK) 287139751Sdavidxu error = ETIMEDOUT; 288139751Sdavidxu return (error); 289138224Sdavidxu} 290138224Sdavidxu 291139013Sdavidxustatic int 292139013Sdavidxuumtx_key_get(struct thread *td, struct umtx *umtx, struct umtx_key *key) 293139013Sdavidxu{ 294139013Sdavidxu#if defined(UMTX_DYNAMIC_SHARED) || defined(UMTX_STATIC_SHARED) 295139013Sdavidxu vm_map_t map; 296139013Sdavidxu vm_map_entry_t entry; 297139013Sdavidxu vm_pindex_t pindex; 298139013Sdavidxu vm_prot_t prot; 299139013Sdavidxu boolean_t wired; 300139013Sdavidxu 301139013Sdavidxu map = &td->td_proc->p_vmspace->vm_map; 302139013Sdavidxu if (vm_map_lookup(&map, (vm_offset_t)umtx, VM_PROT_WRITE, 303139013Sdavidxu &entry, &key->info.shared.object, &pindex, &prot, 304139013Sdavidxu &wired) != KERN_SUCCESS) { 305139013Sdavidxu return EFAULT; 306139013Sdavidxu } 307139013Sdavidxu#endif 308139013Sdavidxu 309139013Sdavidxu#if defined(UMTX_DYNAMIC_SHARED) 310139013Sdavidxu key->type = UMTX_SHARED; 311139013Sdavidxu key->info.shared.offset = entry->offset + entry->start - 312139013Sdavidxu (vm_offset_t)umtx; 313139013Sdavidxu /* 314139013Sdavidxu * Add object reference, if we don't do this, a buggy application 315139013Sdavidxu * deallocates the object, the object will be reused by other 316139013Sdavidxu * applications, then unlock will wake wrong thread. 317139013Sdavidxu */ 318139013Sdavidxu vm_object_reference(key->info.shared.object); 319139013Sdavidxu vm_map_lookup_done(map, entry); 320139013Sdavidxu#elif defined(UMTX_STATIC_SHARED) 321139013Sdavidxu if (VM_INHERIT_SHARE == entry->inheritance) { 322139013Sdavidxu key->type = UMTX_SHARED; 323139013Sdavidxu key->info.shared.offset = entry->offset + entry->start - 324139013Sdavidxu (vm_offset_t)umtx; 325139013Sdavidxu vm_object_reference(key->info.shared.object); 326139013Sdavidxu } else { 327139013Sdavidxu key->type = UMTX_PRIVATE; 328139013Sdavidxu key->info.private.umtx = umtx; 329139013Sdavidxu key->info.private.pid = td->td_proc->p_pid; 330139013Sdavidxu } 331139013Sdavidxu vm_map_lookup_done(map, entry); 332139013Sdavidxu#else 333139013Sdavidxu key->type = UMTX_PRIVATE; 334139013Sdavidxu key->info.private.umtx = umtx; 335139013Sdavidxu key->info.private.pid = td->td_proc->p_pid; 336139013Sdavidxu#endif 337139013Sdavidxu return (0); 338139013Sdavidxu} 339139013Sdavidxu 340139013Sdavidxustatic inline void 341139013Sdavidxuumtx_key_release(struct umtx_key *key) 342139013Sdavidxu{ 343139013Sdavidxu if (key->type == UMTX_SHARED) 344139013Sdavidxu vm_object_deallocate(key->info.shared.object); 345139013Sdavidxu} 346139013Sdavidxu 347138224Sdavidxustatic inline int 348139013Sdavidxuumtxq_queue_me(struct thread *td, struct umtx *umtx, struct umtx_q *uq) 349138224Sdavidxu{ 350139013Sdavidxu int error; 351138224Sdavidxu 352139013Sdavidxu if ((error = umtx_key_get(td, umtx, &uq->uq_key)) != 0) 353139013Sdavidxu return (error); 354139013Sdavidxu 355139013Sdavidxu uq->uq_addr = (vm_offset_t)umtx; 356139013Sdavidxu uq->uq_thread = td; 357139013Sdavidxu umtxq_lock(&uq->uq_key); 358139257Sdavidxu /* hmm, for condition variable, we don't need busy flag. */ 359139257Sdavidxu umtxq_busy(&uq->uq_key); 360139013Sdavidxu umtxq_insert(uq); 361139257Sdavidxu umtxq_unbusy(&uq->uq_key); 362139013Sdavidxu umtxq_unlock(&uq->uq_key); 363139013Sdavidxu return (0); 364138224Sdavidxu} 365138224Sdavidxu 366139013Sdavidxu#if defined(UMTX_DYNAMIC_SHARED) 367138224Sdavidxustatic void 368139013Sdavidxufork_handler(void *arg, struct proc *p1, struct proc *p2, int flags) 369138224Sdavidxu{ 370139013Sdavidxu vm_map_t map; 371139013Sdavidxu vm_map_entry_t entry; 372139013Sdavidxu vm_object_t object; 373139013Sdavidxu vm_pindex_t pindex; 374139013Sdavidxu vm_prot_t prot; 375139013Sdavidxu boolean_t wired; 376139013Sdavidxu struct umtx_key key; 377139013Sdavidxu LIST_HEAD(, umtx_q) workq; 378138224Sdavidxu struct umtx_q *uq; 379139013Sdavidxu struct thread *td; 380139013Sdavidxu int onq; 381138224Sdavidxu 382139013Sdavidxu LIST_INIT(&workq); 383139013Sdavidxu 384139013Sdavidxu /* Collect threads waiting on umtxq */ 385139013Sdavidxu PROC_LOCK(p1); 386139013Sdavidxu FOREACH_THREAD_IN_PROC(p1, td) { 387139013Sdavidxu if (td->td_flags & TDF_UMTXQ) { 388139013Sdavidxu uq = td->td_umtxq; 389139013Sdavidxu if (uq) 390139013Sdavidxu LIST_INSERT_HEAD(&workq, uq, uq_rqnext); 391138224Sdavidxu } 392115765Sjeff } 393139013Sdavidxu PROC_UNLOCK(p1); 394139013Sdavidxu 395139013Sdavidxu LIST_FOREACH(uq, &workq, uq_rqnext) { 396139013Sdavidxu map = &p1->p_vmspace->vm_map; 397139013Sdavidxu if (vm_map_lookup(&map, uq->uq_addr, VM_PROT_WRITE, 398139013Sdavidxu &entry, &object, &pindex, &prot, &wired) != KERN_SUCCESS) { 399139013Sdavidxu continue; 400139013Sdavidxu } 401139013Sdavidxu key.type = UMTX_SHARED; 402139013Sdavidxu key.info.shared.object = object; 403139013Sdavidxu key.info.shared.offset = entry->offset + entry->start - 404139013Sdavidxu uq->uq_addr; 405139013Sdavidxu if (umtx_key_match(&key, &uq->uq_key)) { 406139013Sdavidxu vm_map_lookup_done(map, entry); 407139013Sdavidxu continue; 408139013Sdavidxu } 409139013Sdavidxu 410139013Sdavidxu umtxq_lock(&uq->uq_key); 411139257Sdavidxu umtxq_busy(&uq->uq_key); 412139013Sdavidxu if (uq->uq_thread->td_flags & TDF_UMTXQ) { 413139013Sdavidxu umtxq_remove(uq); 414139013Sdavidxu onq = 1; 415139013Sdavidxu } else 416139013Sdavidxu onq = 0; 417139257Sdavidxu umtxq_unbusy(&uq->uq_key); 418139013Sdavidxu umtxq_unlock(&uq->uq_key); 419139013Sdavidxu if (onq) { 420139013Sdavidxu vm_object_deallocate(uq->uq_key.info.shared.object); 421139013Sdavidxu uq->uq_key = key; 422139013Sdavidxu umtxq_lock(&uq->uq_key); 423139257Sdavidxu umtxq_busy(&uq->uq_key); 424139013Sdavidxu umtxq_insert(uq); 425139257Sdavidxu umtxq_unbusy(&uq->uq_key); 426139013Sdavidxu umtxq_unlock(&uq->uq_key); 427139013Sdavidxu vm_object_reference(uq->uq_key.info.shared.object); 428139013Sdavidxu } 429139013Sdavidxu vm_map_lookup_done(map, entry); 430139013Sdavidxu } 431115765Sjeff} 432139013Sdavidxu#endif 433115765Sjeff 434139013Sdavidxustatic int 435139013Sdavidxu_do_lock(struct thread *td, struct umtx *umtx, long id, int timo) 436112904Sjeff{ 437139013Sdavidxu struct umtx_q uq; 438112904Sjeff intptr_t owner; 439112967Sjake intptr_t old; 440138224Sdavidxu int error = 0; 441112904Sjeff 442112904Sjeff /* 443139013Sdavidxu * Care must be exercised when dealing with umtx structure. It 444112904Sjeff * can fault on any access. 445112904Sjeff */ 446112904Sjeff 447112904Sjeff for (;;) { 448112904Sjeff /* 449112904Sjeff * Try the uncontested case. This should be done in userland. 450112904Sjeff */ 451112904Sjeff owner = casuptr((intptr_t *)&umtx->u_owner, 452139013Sdavidxu UMTX_UNOWNED, id); 453112904Sjeff 454138224Sdavidxu /* The acquire succeeded. */ 455138224Sdavidxu if (owner == UMTX_UNOWNED) 456138224Sdavidxu return (0); 457138224Sdavidxu 458115765Sjeff /* The address was invalid. */ 459115765Sjeff if (owner == -1) 460115765Sjeff return (EFAULT); 461115765Sjeff 462115765Sjeff /* If no one owns it but it is contested try to acquire it. */ 463115765Sjeff if (owner == UMTX_CONTESTED) { 464115765Sjeff owner = casuptr((intptr_t *)&umtx->u_owner, 465139013Sdavidxu UMTX_CONTESTED, id | UMTX_CONTESTED); 466115765Sjeff 467138224Sdavidxu if (owner == UMTX_CONTESTED) 468138224Sdavidxu return (0); 469138224Sdavidxu 470115765Sjeff /* The address was invalid. */ 471115765Sjeff if (owner == -1) 472115765Sjeff return (EFAULT); 473115765Sjeff 474115765Sjeff /* If this failed the lock has changed, restart. */ 475115765Sjeff continue; 476112904Sjeff } 477112904Sjeff 478138224Sdavidxu /* 479138224Sdavidxu * If we caught a signal, we have retried and now 480138224Sdavidxu * exit immediately. 481138224Sdavidxu */ 482139013Sdavidxu if (error || (error = umtxq_queue_me(td, umtx, &uq)) != 0) 483138224Sdavidxu return (error); 484112904Sjeff 485112904Sjeff /* 486112904Sjeff * Set the contested bit so that a release in user space 487112904Sjeff * knows to use the system call for unlock. If this fails 488112904Sjeff * either some one else has acquired the lock or it has been 489112904Sjeff * released. 490112904Sjeff */ 491112967Sjake old = casuptr((intptr_t *)&umtx->u_owner, owner, 492112967Sjake owner | UMTX_CONTESTED); 493112904Sjeff 494112904Sjeff /* The address was invalid. */ 495112967Sjake if (old == -1) { 496139013Sdavidxu umtxq_lock(&uq.uq_key); 497139257Sdavidxu umtxq_busy(&uq.uq_key); 498139013Sdavidxu umtxq_remove(&uq); 499139257Sdavidxu umtxq_unbusy(&uq.uq_key); 500139013Sdavidxu umtxq_unlock(&uq.uq_key); 501139013Sdavidxu umtx_key_release(&uq.uq_key); 502115765Sjeff return (EFAULT); 503112904Sjeff } 504112904Sjeff 505112904Sjeff /* 506115765Sjeff * We set the contested bit, sleep. Otherwise the lock changed 507117685Smtm * and we need to retry or we lost a race to the thread 508117685Smtm * unlocking the umtx. 509112904Sjeff */ 510139013Sdavidxu umtxq_lock(&uq.uq_key); 511139013Sdavidxu if (old == owner && (td->td_flags & TDF_UMTXQ)) { 512139013Sdavidxu error = umtxq_sleep(td, &uq.uq_key, 513139257Sdavidxu td->td_priority | PCATCH, 514139013Sdavidxu "umtx", timo); 515138224Sdavidxu } 516139257Sdavidxu umtxq_busy(&uq.uq_key); 517139257Sdavidxu umtxq_remove(&uq); 518139257Sdavidxu umtxq_unbusy(&uq.uq_key); 519139257Sdavidxu umtxq_unlock(&uq.uq_key); 520139013Sdavidxu umtx_key_release(&uq.uq_key); 521112904Sjeff } 522117743Smtm 523117743Smtm return (0); 524112904Sjeff} 525112904Sjeff 526139013Sdavidxustatic int 527139013Sdavidxudo_lock(struct thread *td, struct umtx *umtx, long id, 528140245Sdavidxu struct timespec *timeout) 529112904Sjeff{ 530140245Sdavidxu struct timespec ts, ts2, ts3; 531139013Sdavidxu struct timeval tv; 532140245Sdavidxu int error; 533139013Sdavidxu 534140245Sdavidxu if (timeout == NULL) { 535139013Sdavidxu error = _do_lock(td, umtx, id, 0); 536139013Sdavidxu } else { 537140245Sdavidxu getnanouptime(&ts); 538140245Sdavidxu timespecadd(&ts, timeout); 539140245Sdavidxu TIMESPEC_TO_TIMEVAL(&tv, timeout); 540139013Sdavidxu for (;;) { 541140245Sdavidxu error = _do_lock(td, umtx, id, tvtohz(&tv)); 542140245Sdavidxu if (error != ETIMEDOUT) 543140245Sdavidxu break; 544140245Sdavidxu getnanouptime(&ts2); 545140245Sdavidxu if (timespeccmp(&ts2, &ts, >=)) { 546139751Sdavidxu error = ETIMEDOUT; 547139013Sdavidxu break; 548139013Sdavidxu } 549140245Sdavidxu ts3 = ts; 550140245Sdavidxu timespecsub(&ts3, &ts2); 551140245Sdavidxu TIMESPEC_TO_TIMEVAL(&tv, &ts3); 552139013Sdavidxu } 553139013Sdavidxu } 554139258Sdavidxu /* 555139258Sdavidxu * This lets userland back off critical region if needed. 556139258Sdavidxu */ 557139258Sdavidxu if (error == ERESTART) 558139258Sdavidxu error = EINTR; 559139013Sdavidxu return (error); 560139013Sdavidxu} 561139013Sdavidxu 562139013Sdavidxustatic int 563139013Sdavidxudo_unlock(struct thread *td, struct umtx *umtx, long id) 564139013Sdavidxu{ 565139013Sdavidxu struct umtx_key key; 566112904Sjeff intptr_t owner; 567112967Sjake intptr_t old; 568139257Sdavidxu int error; 569139257Sdavidxu int count; 570112904Sjeff 571112904Sjeff /* 572112904Sjeff * Make sure we own this mtx. 573112904Sjeff * 574112904Sjeff * XXX Need a {fu,su}ptr this is not correct on arch where 575112904Sjeff * sizeof(intptr_t) != sizeof(long). 576112904Sjeff */ 577115765Sjeff if ((owner = fuword(&umtx->u_owner)) == -1) 578115765Sjeff return (EFAULT); 579115765Sjeff 580139013Sdavidxu if ((owner & ~UMTX_CONTESTED) != id) 581115765Sjeff return (EPERM); 582112904Sjeff 583117685Smtm /* We should only ever be in here for contested locks */ 584119836Stjr if ((owner & UMTX_CONTESTED) == 0) 585119836Stjr return (EINVAL); 586112904Sjeff 587139257Sdavidxu if ((error = umtx_key_get(td, umtx, &key)) != 0) 588139257Sdavidxu return (error); 589139257Sdavidxu 590139257Sdavidxu umtxq_lock(&key); 591139257Sdavidxu umtxq_busy(&key); 592139257Sdavidxu count = umtxq_count(&key); 593139257Sdavidxu umtxq_unlock(&key); 594139257Sdavidxu 595117743Smtm /* 596117743Smtm * When unlocking the umtx, it must be marked as unowned if 597117743Smtm * there is zero or one thread only waiting for it. 598117743Smtm * Otherwise, it must be marked as contested. 599117743Smtm */ 600139257Sdavidxu old = casuptr((intptr_t *)&umtx->u_owner, owner, 601139257Sdavidxu count <= 1 ? UMTX_UNOWNED : UMTX_CONTESTED); 602139257Sdavidxu umtxq_lock(&key); 603139257Sdavidxu umtxq_signal(&key, 0); 604139257Sdavidxu umtxq_unbusy(&key); 605139257Sdavidxu umtxq_unlock(&key); 606139257Sdavidxu umtx_key_release(&key); 607115765Sjeff if (old == -1) 608115765Sjeff return (EFAULT); 609138224Sdavidxu if (old != owner) 610138224Sdavidxu return (EINVAL); 611115765Sjeff return (0); 612112904Sjeff} 613139013Sdavidxu 614139013Sdavidxustatic int 615140245Sdavidxudo_wait(struct thread *td, struct umtx *umtx, long id, struct timespec *timeout) 616139013Sdavidxu{ 617139013Sdavidxu struct umtx_q uq; 618140245Sdavidxu struct timespec ts, ts2, ts3; 619139013Sdavidxu struct timeval tv; 620139427Sdavidxu long tmp; 621140245Sdavidxu int error = 0; 622139013Sdavidxu 623139427Sdavidxu if ((error = umtxq_queue_me(td, umtx, &uq)) != 0) 624139013Sdavidxu return (error); 625139427Sdavidxu tmp = fuword(&umtx->u_owner); 626139427Sdavidxu if (tmp != id) { 627139013Sdavidxu umtxq_lock(&uq.uq_key); 628139013Sdavidxu umtxq_remove(&uq); 629139013Sdavidxu umtxq_unlock(&uq.uq_key); 630140245Sdavidxu } else if (timeout == NULL) { 631139013Sdavidxu umtxq_lock(&uq.uq_key); 632139013Sdavidxu if (td->td_flags & TDF_UMTXQ) 633139013Sdavidxu error = umtxq_sleep(td, &uq.uq_key, 634139013Sdavidxu td->td_priority | PCATCH, "ucond", 0); 635139257Sdavidxu if (!(td->td_flags & TDF_UMTXQ)) 636139257Sdavidxu error = 0; 637139257Sdavidxu else 638139257Sdavidxu umtxq_remove(&uq); 639139013Sdavidxu umtxq_unlock(&uq.uq_key); 640139013Sdavidxu } else { 641140245Sdavidxu getnanouptime(&ts); 642140245Sdavidxu timespecadd(&ts, timeout); 643140245Sdavidxu TIMESPEC_TO_TIMEVAL(&tv, timeout); 644139013Sdavidxu for (;;) { 645139257Sdavidxu umtxq_lock(&uq.uq_key); 646140245Sdavidxu if (td->td_flags & TDF_UMTXQ) { 647140245Sdavidxu error = umtxq_sleep(td, &uq.uq_key, 648140245Sdavidxu td->td_priority | PCATCH, 649140245Sdavidxu "ucond", tvtohz(&tv)); 650140245Sdavidxu } 651140245Sdavidxu if (!(td->td_flags & TDF_UMTXQ)) { 652140245Sdavidxu umtxq_unlock(&uq.uq_key); 653140245Sdavidxu goto out; 654140245Sdavidxu } 655140245Sdavidxu umtxq_unlock(&uq.uq_key); 656140245Sdavidxu if (error != ETIMEDOUT) 657140245Sdavidxu break; 658140245Sdavidxu getnanouptime(&ts2); 659140245Sdavidxu if (timespeccmp(&ts2, &ts, >=)) { 660139751Sdavidxu error = ETIMEDOUT; 661139013Sdavidxu break; 662139013Sdavidxu } 663140245Sdavidxu ts3 = ts; 664140245Sdavidxu timespecsub(&ts3, &ts2); 665140245Sdavidxu TIMESPEC_TO_TIMEVAL(&tv, &ts3); 666139013Sdavidxu } 667140245Sdavidxu umtxq_lock(&uq.uq_key); 668140245Sdavidxu umtxq_remove(&uq); 669139257Sdavidxu umtxq_unlock(&uq.uq_key); 670139013Sdavidxu } 671140245Sdavidxuout: 672139013Sdavidxu umtx_key_release(&uq.uq_key); 673139257Sdavidxu if (error == ERESTART) 674139257Sdavidxu error = EINTR; 675139013Sdavidxu return (error); 676139013Sdavidxu} 677139013Sdavidxu 678139013Sdavidxustatic int 679139257Sdavidxudo_wake(struct thread *td, void *uaddr, int n_wake) 680139013Sdavidxu{ 681139013Sdavidxu struct umtx_key key; 682139257Sdavidxu int ret; 683139013Sdavidxu 684139257Sdavidxu if ((ret = umtx_key_get(td, uaddr, &key)) != 0) 685139257Sdavidxu return (ret); 686139258Sdavidxu umtxq_lock(&key); 687139257Sdavidxu ret = umtxq_signal(&key, n_wake); 688139258Sdavidxu umtxq_unlock(&key); 689139257Sdavidxu umtx_key_release(&key); 690139013Sdavidxu return (0); 691139013Sdavidxu} 692139013Sdavidxu 693139013Sdavidxuint 694139013Sdavidxu_umtx_lock(struct thread *td, struct _umtx_lock_args *uap) 695139013Sdavidxu /* struct umtx *umtx */ 696139013Sdavidxu{ 697139013Sdavidxu return _do_lock(td, uap->umtx, td->td_tid, 0); 698139013Sdavidxu} 699139013Sdavidxu 700139013Sdavidxuint 701139013Sdavidxu_umtx_unlock(struct thread *td, struct _umtx_unlock_args *uap) 702139013Sdavidxu /* struct umtx *umtx */ 703139013Sdavidxu{ 704139013Sdavidxu return do_unlock(td, uap->umtx, td->td_tid); 705139013Sdavidxu} 706139013Sdavidxu 707139013Sdavidxuint 708139013Sdavidxu_umtx_op(struct thread *td, struct _umtx_op_args *uap) 709139013Sdavidxu{ 710140245Sdavidxu struct timespec timeout; 711139013Sdavidxu struct timespec *ts; 712139013Sdavidxu int error; 713139013Sdavidxu 714139013Sdavidxu switch(uap->op) { 715139013Sdavidxu case UMTX_OP_LOCK: 716139013Sdavidxu /* Allow a null timespec (wait forever). */ 717139292Sdavidxu if (uap->uaddr2 == NULL) 718139013Sdavidxu ts = NULL; 719139013Sdavidxu else { 720140245Sdavidxu error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 721139013Sdavidxu if (error != 0) 722140102Sdavidxu break; 723140245Sdavidxu if (timeout.tv_nsec >= 1000000000 || 724140245Sdavidxu timeout.tv_nsec < 0) { 725140102Sdavidxu error = EINVAL; 726140102Sdavidxu break; 727140102Sdavidxu } 728140245Sdavidxu ts = &timeout; 729139013Sdavidxu } 730140102Sdavidxu error = do_lock(td, uap->umtx, uap->id, ts); 731140102Sdavidxu break; 732139013Sdavidxu case UMTX_OP_UNLOCK: 733140102Sdavidxu error = do_unlock(td, uap->umtx, uap->id); 734140102Sdavidxu break; 735139427Sdavidxu case UMTX_OP_WAIT: 736139013Sdavidxu /* Allow a null timespec (wait forever). */ 737139292Sdavidxu if (uap->uaddr2 == NULL) 738139013Sdavidxu ts = NULL; 739139013Sdavidxu else { 740140245Sdavidxu error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 741139013Sdavidxu if (error != 0) 742140102Sdavidxu break; 743140245Sdavidxu if (timeout.tv_nsec >= 1000000000 || 744140245Sdavidxu timeout.tv_nsec < 0) { 745140102Sdavidxu error = EINVAL; 746140102Sdavidxu break; 747140102Sdavidxu } 748140245Sdavidxu ts = &timeout; 749139013Sdavidxu } 750140102Sdavidxu error = do_wait(td, uap->umtx, uap->id, ts); 751140102Sdavidxu break; 752139013Sdavidxu case UMTX_OP_WAKE: 753140102Sdavidxu error = do_wake(td, uap->umtx, uap->id); 754140102Sdavidxu break; 755139013Sdavidxu default: 756140102Sdavidxu error = EINVAL; 757140102Sdavidxu break; 758139013Sdavidxu } 759140102Sdavidxu td->td_retval[0] = -error; 760140102Sdavidxu return (0); 761139013Sdavidxu} 762