The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_kse.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (C) 2001 Julian Elischer <julian@freebsd.org>.
    3  *  All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice(s), this list of conditions and the following disclaimer as
   10  *    the first lines of this file unmodified other than the possible
   11  *    addition of one or more copyright notices.
   12  * 2. Redistributions in binary form must reproduce the above copyright
   13  *    notice(s), this list of conditions and the following disclaimer in the
   14  *    documentation and/or other materials provided with the distribution.
   15  *
   16  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
   17  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
   18  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
   19  * DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY
   20  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
   21  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
   22  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
   23  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
   26  * DAMAGE.
   27  */
   28 
   29 #include <sys/cdefs.h>
   30 __FBSDID("$FreeBSD: releng/6.0/sys/kern/kern_kse.c 151012 2005-10-06 18:24:24Z delphij $");
   31 
   32 #include <sys/param.h>
   33 #include <sys/systm.h>
   34 #include <sys/kernel.h>
   35 #include <sys/imgact.h>
   36 #include <sys/lock.h>
   37 #include <sys/mutex.h>
   38 #include <sys/proc.h>
   39 #include <sys/ptrace.h>
   40 #include <sys/smp.h>
   41 #include <sys/syscallsubr.h>
   42 #include <sys/sysproto.h>
   43 #include <sys/sched.h>
   44 #include <sys/signalvar.h>
   45 #include <sys/sleepqueue.h>
   46 #include <sys/kse.h>
   47 #include <sys/ktr.h>
   48 #include <vm/uma.h>
   49 
   50 /*
   51  * KSEGRP related storage.
   52  */
   53 static uma_zone_t upcall_zone;
   54 
   55 /* DEBUG ONLY */
   56 extern int virtual_cpu;
   57 extern int thread_debug;
   58 
   59 extern int max_threads_per_proc;
   60 extern int max_groups_per_proc;
   61 extern int max_threads_hits;
   62 extern struct mtx kse_zombie_lock;
   63 
   64 
   65 TAILQ_HEAD(, kse_upcall) zombie_upcalls =
   66         TAILQ_HEAD_INITIALIZER(zombie_upcalls);
   67 
   68 static int thread_update_usr_ticks(struct thread *td);
   69 static void thread_alloc_spare(struct thread *td);
   70 
   71 struct kse_upcall *
   72 upcall_alloc(void)
   73 {
   74         struct kse_upcall *ku;
   75 
   76         ku = uma_zalloc(upcall_zone, M_WAITOK | M_ZERO);
   77         return (ku);
   78 }
   79 
   80 void
   81 upcall_free(struct kse_upcall *ku)
   82 {
   83 
   84         uma_zfree(upcall_zone, ku);
   85 }
   86 
   87 void
   88 upcall_link(struct kse_upcall *ku, struct ksegrp *kg)
   89 {
   90 
   91         mtx_assert(&sched_lock, MA_OWNED);
   92         TAILQ_INSERT_TAIL(&kg->kg_upcalls, ku, ku_link);
   93         ku->ku_ksegrp = kg;
   94         kg->kg_numupcalls++;
   95 }
   96 
   97 void
   98 upcall_unlink(struct kse_upcall *ku)
   99 {
  100         struct ksegrp *kg = ku->ku_ksegrp;
  101 
  102         mtx_assert(&sched_lock, MA_OWNED);
  103         KASSERT(ku->ku_owner == NULL, ("%s: have owner", __func__));
  104         TAILQ_REMOVE(&kg->kg_upcalls, ku, ku_link);
  105         kg->kg_numupcalls--;
  106         upcall_stash(ku);
  107 }
  108 
  109 void
  110 upcall_remove(struct thread *td)
  111 {
  112 
  113         mtx_assert(&sched_lock, MA_OWNED);
  114         if (td->td_upcall != NULL) {
  115                 td->td_upcall->ku_owner = NULL;
  116                 upcall_unlink(td->td_upcall);
  117                 td->td_upcall = NULL;
  118         }
  119 }
  120 
  121 #ifndef _SYS_SYSPROTO_H_
  122 struct kse_switchin_args {
  123         struct kse_thr_mailbox *tmbx;
  124         int flags;
  125 };
  126 #endif
  127 
  128 int
  129 kse_switchin(struct thread *td, struct kse_switchin_args *uap)
  130 {
  131         struct kse_thr_mailbox tmbx;
  132         struct kse_upcall *ku;
  133         int error;
  134 
  135         if ((ku = td->td_upcall) == NULL || TD_CAN_UNBIND(td))
  136                 return (EINVAL);
  137         error = (uap->tmbx == NULL) ? EINVAL : 0;
  138         if (!error)
  139                 error = copyin(uap->tmbx, &tmbx, sizeof(tmbx));
  140         if (!error && (uap->flags & KSE_SWITCHIN_SETTMBX))
  141                 error = (suword(&ku->ku_mailbox->km_curthread,
  142                          (long)uap->tmbx) != 0 ? EINVAL : 0);
  143         if (!error)
  144                 error = set_mcontext(td, &tmbx.tm_context.uc_mcontext);
  145         if (!error) {
  146                 suword32(&uap->tmbx->tm_lwp, td->td_tid);
  147                 if (uap->flags & KSE_SWITCHIN_SETTMBX) {
  148                         td->td_mailbox = uap->tmbx;
  149                         td->td_pflags |= TDP_CAN_UNBIND;
  150                 }
  151                 if (td->td_proc->p_flag & P_TRACED) {
  152                         if (tmbx.tm_dflags & TMDF_SSTEP)
  153                                 ptrace_single_step(td);
  154                         else
  155                                 ptrace_clear_single_step(td);
  156                         if (tmbx.tm_dflags & TMDF_SUSPEND) {
  157                                 mtx_lock_spin(&sched_lock);
  158                                 /* fuword can block, check again */
  159                                 if (td->td_upcall)
  160                                         ku->ku_flags |= KUF_DOUPCALL;
  161                                 mtx_unlock_spin(&sched_lock);
  162                         }
  163                 }
  164         }
  165         return ((error == 0) ? EJUSTRETURN : error);
  166 }
  167 
  168 /*
  169 struct kse_thr_interrupt_args {
  170         struct kse_thr_mailbox * tmbx;
  171         int cmd;
  172         long data;
  173 };
  174 */
  175 int
  176 kse_thr_interrupt(struct thread *td, struct kse_thr_interrupt_args *uap)
  177 {
  178         struct kse_execve_args args;
  179         struct image_args iargs;
  180         struct proc *p;
  181         struct thread *td2;
  182         struct kse_upcall *ku;
  183         struct kse_thr_mailbox *tmbx;
  184         uint32_t flags;
  185         int error;
  186 
  187         p = td->td_proc;
  188 
  189         if (!(p->p_flag & P_SA))
  190                 return (EINVAL);
  191 
  192         switch (uap->cmd) {
  193         case KSE_INTR_SENDSIG:
  194                 if (uap->data < 0 || uap->data > _SIG_MAXSIG)
  195                         return (EINVAL);
  196         case KSE_INTR_INTERRUPT:
  197         case KSE_INTR_RESTART:
  198                 PROC_LOCK(p);
  199                 mtx_lock_spin(&sched_lock);
  200                 FOREACH_THREAD_IN_PROC(p, td2) {
  201                         if (td2->td_mailbox == uap->tmbx)
  202                                 break;
  203                 }
  204                 if (td2 == NULL) {
  205                         mtx_unlock_spin(&sched_lock);
  206                         PROC_UNLOCK(p);
  207                         return (ESRCH);
  208                 }
  209                 if (uap->cmd == KSE_INTR_SENDSIG) {
  210                         if (uap->data > 0) {
  211                                 td2->td_flags &= ~TDF_INTERRUPT;
  212                                 mtx_unlock_spin(&sched_lock);
  213                                 tdsignal(td2, (int)uap->data, SIGTARGET_TD);
  214                         } else {
  215                                 mtx_unlock_spin(&sched_lock);
  216                         }
  217                 } else {
  218                         td2->td_flags |= TDF_INTERRUPT | TDF_ASTPENDING;
  219                         if (TD_CAN_UNBIND(td2))
  220                                 td2->td_upcall->ku_flags |= KUF_DOUPCALL;
  221                         if (uap->cmd == KSE_INTR_INTERRUPT)
  222                                 td2->td_intrval = EINTR;
  223                         else
  224                                 td2->td_intrval = ERESTART;
  225                         if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR))
  226                                 sleepq_abort(td2);
  227                         mtx_unlock_spin(&sched_lock);
  228                 }
  229                 PROC_UNLOCK(p);
  230                 break;
  231         case KSE_INTR_SIGEXIT:
  232                 if (uap->data < 1 || uap->data > _SIG_MAXSIG)
  233                         return (EINVAL);
  234                 PROC_LOCK(p);
  235                 sigexit(td, (int)uap->data);
  236                 break;
  237 
  238         case KSE_INTR_DBSUSPEND:
  239                 /* this sub-function is only for bound thread */
  240                 if (td->td_pflags & TDP_SA)
  241                         return (EINVAL);
  242                 ku = td->td_upcall;
  243                 tmbx = (void *)fuword((void *)&ku->ku_mailbox->km_curthread);
  244                 if (tmbx == NULL || tmbx == (void *)-1)
  245                         return (EINVAL);
  246                 flags = 0;
  247                 while ((p->p_flag & P_TRACED) && !(p->p_flag & P_SINGLE_EXIT)) {
  248                         flags = fuword32(&tmbx->tm_dflags);
  249                         if (!(flags & TMDF_SUSPEND))
  250                                 break;
  251                         PROC_LOCK(p);
  252                         mtx_lock_spin(&sched_lock);
  253                         thread_stopped(p);
  254                         thread_suspend_one(td);
  255                         PROC_UNLOCK(p);
  256                         mi_switch(SW_VOL, NULL);
  257                         mtx_unlock_spin(&sched_lock);
  258                 }
  259                 return (0);
  260 
  261         case KSE_INTR_EXECVE:
  262                 error = copyin((void *)uap->data, &args, sizeof(args));
  263                 if (error)
  264                         return (error);
  265                 error = exec_copyin_args(&iargs, args.path, UIO_USERSPACE,
  266                     args.argv, args.envp);
  267                 if (error == 0)
  268                         error = kern_execve(td, &iargs, NULL);
  269                 exec_free_args(&iargs);
  270                 if (error == 0) {
  271                         PROC_LOCK(p);
  272                         SIGSETOR(td->td_siglist, args.sigpend);
  273                         PROC_UNLOCK(p);
  274                         kern_sigprocmask(td, SIG_SETMASK, &args.sigmask, NULL,
  275                             0);
  276                 }
  277                 return (error);
  278 
  279         default:
  280                 return (EINVAL);
  281         }
  282         return (0);
  283 }
  284 
  285 /*
  286 struct kse_exit_args {
  287         register_t dummy;
  288 };
  289 */
  290 int
  291 kse_exit(struct thread *td, struct kse_exit_args *uap)
  292 {
  293         struct proc *p;
  294         struct ksegrp *kg;
  295         struct kse_upcall *ku, *ku2;
  296         int    error, count;
  297 
  298         p = td->td_proc;
  299         /* 
  300          * Ensure that this is only called from the UTS
  301          */
  302         if ((ku = td->td_upcall) == NULL || TD_CAN_UNBIND(td))
  303                 return (EINVAL);
  304 
  305         kg = td->td_ksegrp;
  306         count = 0;
  307 
  308         /*
  309          * Calculate the existing non-exiting upcalls in this ksegroup.
  310          * If we are the last upcall but there are still other threads,
  311          * then do not exit. We need the other threads to be able to 
  312          * complete whatever they are doing.
  313          * XXX This relies on the userland knowing what to do if we return.
  314          * It may be a better choice to convert ourselves into a kse_release
  315          * ( or similar) and wait in the kernel to be needed.
  316          */
  317         PROC_LOCK(p);
  318         mtx_lock_spin(&sched_lock);
  319         FOREACH_UPCALL_IN_GROUP(kg, ku2) {
  320                 if (ku2->ku_flags & KUF_EXITING)
  321                         count++;
  322         }
  323         if ((kg->kg_numupcalls - count) == 1 &&
  324             (kg->kg_numthreads > 1)) {
  325                 mtx_unlock_spin(&sched_lock);
  326                 PROC_UNLOCK(p);
  327                 return (EDEADLK);
  328         }
  329         ku->ku_flags |= KUF_EXITING;
  330         mtx_unlock_spin(&sched_lock);
  331         PROC_UNLOCK(p);
  332 
  333         /* 
  334          * Mark the UTS mailbox as having been finished with.
  335          * If that fails then just go for a segfault.
  336          * XXX need to check it that can be deliverred without a mailbox.
  337          */
  338         error = suword32(&ku->ku_mailbox->km_flags, ku->ku_mflags|KMF_DONE);
  339         if (!(td->td_pflags & TDP_SA))
  340                 if (suword32(&td->td_mailbox->tm_lwp, 0))
  341                         error = EFAULT;
  342         PROC_LOCK(p);
  343         if (error)
  344                 psignal(p, SIGSEGV);
  345         mtx_lock_spin(&sched_lock);
  346         upcall_remove(td);
  347         if (p->p_numthreads != 1) {
  348                 /*
  349                  * If we are not the last thread, but we are the last
  350                  * thread in this ksegrp, then by definition this is not
  351                  * the last group and we need to clean it up as well.
  352                  * thread_exit will clean up the kseg as needed.
  353                  */
  354                 thread_stopped(p);
  355                 thread_exit();
  356                 /* NOTREACHED */
  357         }
  358         /*
  359          * This is the last thread. Just return to the user.
  360          * We know that there is only one ksegrp too, as any others
  361          * would have been discarded in previous calls to thread_exit().
  362          * Effectively we have left threading mode..
  363          * The only real thing left to do is ensure that the
  364          * scheduler sets out concurrency back to 1 as that may be a
  365          * resource leak otherwise.
  366          * This is an A[PB]I issue.. what SHOULD we do?
  367          * One possibility is to return to the user. It may not cope well.
  368          * The other possibility would be to let the process exit.
  369          */
  370         thread_unthread(td);
  371         mtx_unlock_spin(&sched_lock);
  372         PROC_UNLOCK(p);
  373 #if 1
  374         return (0);
  375 #else
  376         exit1(td, 0);
  377 #endif
  378 }
  379 
  380 /*
  381  * Either becomes an upcall or waits for an awakening event and
  382  * then becomes an upcall. Only error cases return.
  383  */
  384 /*
  385 struct kse_release_args {
  386         struct timespec *timeout;
  387 };
  388 */
  389 int
  390 kse_release(struct thread *td, struct kse_release_args *uap)
  391 {
  392         struct proc *p;
  393         struct ksegrp *kg;
  394         struct kse_upcall *ku;
  395         struct timespec timeout;
  396         struct timeval tv;
  397         sigset_t sigset;
  398         int error;
  399 
  400         p = td->td_proc;
  401         kg = td->td_ksegrp;
  402         if ((ku = td->td_upcall) == NULL || TD_CAN_UNBIND(td))
  403                 return (EINVAL);
  404         if (uap->timeout != NULL) {
  405                 if ((error = copyin(uap->timeout, &timeout, sizeof(timeout))))
  406                         return (error);
  407                 TIMESPEC_TO_TIMEVAL(&tv, &timeout);
  408         }
  409         if (td->td_pflags & TDP_SA)
  410                 td->td_pflags |= TDP_UPCALLING;
  411         else {
  412                 ku->ku_mflags = fuword32(&ku->ku_mailbox->km_flags);
  413                 if (ku->ku_mflags == -1) {
  414                         PROC_LOCK(p);
  415                         sigexit(td, SIGSEGV);
  416                 }
  417         }
  418         PROC_LOCK(p);
  419         if (ku->ku_mflags & KMF_WAITSIGEVENT) {
  420                 /* UTS wants to wait for signal event */
  421                 if (!(p->p_flag & P_SIGEVENT) &&
  422                     !(ku->ku_flags & KUF_DOUPCALL)) {
  423                         td->td_kflags |= TDK_KSERELSIG;
  424                         error = msleep(&p->p_siglist, &p->p_mtx, PPAUSE|PCATCH,
  425                             "ksesigwait", (uap->timeout ? tvtohz(&tv) : 0));
  426                         td->td_kflags &= ~(TDK_KSERELSIG | TDK_WAKEUP);
  427                 }
  428                 p->p_flag &= ~P_SIGEVENT;
  429                 sigset = p->p_siglist;
  430                 PROC_UNLOCK(p);
  431                 error = copyout(&sigset, &ku->ku_mailbox->km_sigscaught,
  432                     sizeof(sigset));
  433         } else {
  434                 if ((ku->ku_flags & KUF_DOUPCALL) == 0 &&
  435                     ((ku->ku_mflags & KMF_NOCOMPLETED) ||
  436                      (kg->kg_completed == NULL))) {
  437                         kg->kg_upsleeps++;
  438                         td->td_kflags |= TDK_KSEREL;
  439                         error = msleep(&kg->kg_completed, &p->p_mtx,
  440                                 PPAUSE|PCATCH, "kserel",
  441                                 (uap->timeout ? tvtohz(&tv) : 0));
  442                         td->td_kflags &= ~(TDK_KSEREL | TDK_WAKEUP);
  443                         kg->kg_upsleeps--;
  444                 }
  445                 PROC_UNLOCK(p);
  446         }
  447         if (ku->ku_flags & KUF_DOUPCALL) {
  448                 mtx_lock_spin(&sched_lock);
  449                 ku->ku_flags &= ~KUF_DOUPCALL;
  450                 mtx_unlock_spin(&sched_lock);
  451         }
  452         return (0);
  453 }
  454 
  455 /* struct kse_wakeup_args {
  456         struct kse_mailbox *mbx;
  457 }; */
  458 int
  459 kse_wakeup(struct thread *td, struct kse_wakeup_args *uap)
  460 {
  461         struct proc *p;
  462         struct ksegrp *kg;
  463         struct kse_upcall *ku;
  464         struct thread *td2;
  465 
  466         p = td->td_proc;
  467         td2 = NULL;
  468         ku = NULL;
  469         /* KSE-enabled processes only, please. */
  470         if (!(p->p_flag & P_SA))
  471                 return (EINVAL);
  472         PROC_LOCK(p);
  473         mtx_lock_spin(&sched_lock);
  474         if (uap->mbx) {
  475                 FOREACH_KSEGRP_IN_PROC(p, kg) {
  476                         FOREACH_UPCALL_IN_GROUP(kg, ku) {
  477                                 if (ku->ku_mailbox == uap->mbx)
  478                                         break;
  479                         }
  480                         if (ku)
  481                                 break;
  482                 }
  483         } else {
  484                 kg = td->td_ksegrp;
  485                 if (kg->kg_upsleeps) {
  486                         mtx_unlock_spin(&sched_lock);
  487                         wakeup(&kg->kg_completed);
  488                         PROC_UNLOCK(p);
  489                         return (0);
  490                 }
  491                 ku = TAILQ_FIRST(&kg->kg_upcalls);
  492         }
  493         if (ku == NULL) {
  494                 mtx_unlock_spin(&sched_lock);
  495                 PROC_UNLOCK(p);
  496                 return (ESRCH);
  497         }
  498         if ((td2 = ku->ku_owner) == NULL) {
  499                 mtx_unlock_spin(&sched_lock);
  500                 panic("%s: no owner", __func__);
  501         } else if (td2->td_kflags & (TDK_KSEREL | TDK_KSERELSIG)) {
  502                 mtx_unlock_spin(&sched_lock);
  503                 if (!(td2->td_kflags & TDK_WAKEUP)) {
  504                         td2->td_kflags |= TDK_WAKEUP;
  505                         if (td2->td_kflags & TDK_KSEREL)
  506                                 sleepq_remove(td2, &kg->kg_completed);
  507                         else
  508                                 sleepq_remove(td2, &p->p_siglist);
  509                 }
  510         } else {
  511                 ku->ku_flags |= KUF_DOUPCALL;
  512                 mtx_unlock_spin(&sched_lock);
  513         }
  514         PROC_UNLOCK(p);
  515         return (0);
  516 }
  517 
  518 /*
  519  * No new KSEG: first call: use current KSE, don't schedule an upcall
  520  * All other situations, do allocate max new KSEs and schedule an upcall.
  521  *
  522  * XXX should be changed so that 'first' behaviour lasts for as long
  523  * as you have not made a kse in this ksegrp. i.e. as long as we do not have
  524  * a mailbox..
  525  */
  526 /* struct kse_create_args {
  527         struct kse_mailbox *mbx;
  528         int newgroup;
  529 }; */
  530 int
  531 kse_create(struct thread *td, struct kse_create_args *uap)
  532 {
  533         struct ksegrp *newkg;
  534         struct ksegrp *kg;
  535         struct proc *p;
  536         struct kse_mailbox mbx;
  537         struct kse_upcall *newku;
  538         int err, ncpus, sa = 0, first = 0;
  539         struct thread *newtd;
  540 
  541         p = td->td_proc;
  542         kg = td->td_ksegrp;
  543         if ((err = copyin(uap->mbx, &mbx, sizeof(mbx))))
  544                 return (err);
  545 
  546         ncpus = mp_ncpus;
  547         if (virtual_cpu != 0)
  548                 ncpus = virtual_cpu;
  549         /*
  550          * If the new UTS mailbox says that this
  551          * will be a BOUND lwp, then it had better
  552          * have its thread mailbox already there.
  553          * In addition, this ksegrp will be limited to
  554          * a concurrency of 1. There is more on this later.
  555          */
  556         if (mbx.km_flags & KMF_BOUND) {
  557                 if (mbx.km_curthread == NULL) 
  558                         return (EINVAL);
  559                 ncpus = 1;
  560         } else {
  561                 sa = TDP_SA;
  562         }
  563 
  564         PROC_LOCK(p);
  565         /*
  566          * Processes using the other threading model can't
  567          * suddenly start calling this one
  568          */
  569         if ((p->p_flag & (P_SA|P_HADTHREADS)) == P_HADTHREADS) {
  570                 PROC_UNLOCK(p);
  571                 return (EINVAL);
  572         }
  573 
  574         /*
  575          * Limit it to NCPU upcall contexts per ksegrp in any case.
  576          * There is a small race here as we don't hold proclock
  577          * until we inc the ksegrp count, but it's not really a big problem
  578          * if we get one too many, but we save a proc lock.
  579          */
  580         if ((!uap->newgroup) && (kg->kg_numupcalls >= ncpus)) {
  581                 PROC_UNLOCK(p);
  582                 return (EPROCLIM);
  583         }
  584 
  585         if (!(p->p_flag & P_SA)) {
  586                 first = 1;
  587                 p->p_flag |= P_SA|P_HADTHREADS;
  588         }
  589 
  590         PROC_UNLOCK(p);
  591         /*
  592          * Now pay attention!
  593          * If we are going to be bound, then we need to be either
  594          * a new group, or the first call ever. In either
  595          * case we will be creating (or be) the only thread in a group.
  596          * and the concurrency will be set to 1.
  597          * This is not quite right, as we may still make ourself 
  598          * bound after making other ksegrps but it will do for now.
  599          * The library will only try do this much.
  600          */
  601         if (!sa && !(uap->newgroup || first))
  602                 return (EINVAL);
  603 
  604         if (uap->newgroup) {
  605                 newkg = ksegrp_alloc();
  606                 bzero(&newkg->kg_startzero,
  607                     __rangeof(struct ksegrp, kg_startzero, kg_endzero));
  608                 bcopy(&kg->kg_startcopy, &newkg->kg_startcopy,
  609                     __rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
  610                 sched_init_concurrency(newkg);
  611                 PROC_LOCK(p);
  612                 if (p->p_numksegrps >= max_groups_per_proc) {
  613                         PROC_UNLOCK(p);
  614                         ksegrp_free(newkg);
  615                         return (EPROCLIM);
  616                 }
  617                 ksegrp_link(newkg, p);
  618                 mtx_lock_spin(&sched_lock);
  619                 sched_fork_ksegrp(td, newkg);
  620                 mtx_unlock_spin(&sched_lock);
  621                 PROC_UNLOCK(p);
  622         } else {
  623                 /*
  624                  * We want to make a thread in our own ksegrp.
  625                  * If we are just the first call, either kind
  626                  * is ok, but if not then either we must be 
  627                  * already an upcallable thread to make another,
  628                  * or a bound thread to make one of those.
  629                  * Once again, not quite right but good enough for now.. XXXKSE
  630                  */
  631                 if (!first && ((td->td_pflags & TDP_SA) != sa))
  632                         return (EINVAL);
  633 
  634                 newkg = kg;
  635         }
  636 
  637         /* 
  638          * This test is a bit "indirect".
  639          * It might simplify things if we made a direct way of testing
  640          * if a ksegrp has been worked on before.
  641          * In the case of a bound request and the concurrency being set to 
  642          * one, the concurrency will already be 1 so it's just inefficient
  643          * but not dangerous to call this again. XXX
  644          */
  645         if (newkg->kg_numupcalls == 0) {
  646                 /*
  647                  * Initialize KSE group with the appropriate
  648                  * concurrency.
  649                  *
  650                  * For a multiplexed group, create as as much concurrency
  651                  * as the number of physical cpus.
  652                  * This increases concurrency in the kernel even if the
  653                  * userland is not MP safe and can only run on a single CPU.
  654                  * In an ideal world, every physical cpu should execute a
  655                  * thread.  If there is enough concurrency, threads in the
  656                  * kernel can be executed parallel on different cpus at
  657                  * full speed without being restricted by the number of
  658                  * upcalls the userland provides.
  659                  * Adding more upcall structures only increases concurrency
  660                  * in userland.
  661                  *
  662                  * For a bound thread group, because there is only one thread
  663                  * in the group, we only set the concurrency for the group 
  664                  * to 1.  A thread in this kind of group will never schedule
  665                  * an upcall when blocked.  This simulates pthread system
  666                  * scope thread behaviour.
  667                  */
  668                 sched_set_concurrency(newkg, ncpus);
  669         }
  670         /* 
  671          * Even bound LWPs get a mailbox and an upcall to hold it.
  672          */
  673         newku = upcall_alloc();
  674         newku->ku_mailbox = uap->mbx;
  675         newku->ku_func = mbx.km_func;
  676         bcopy(&mbx.km_stack, &newku->ku_stack, sizeof(stack_t));
  677 
  678         /*
  679          * For the first call this may not have been set.
  680          * Of course nor may it actually be needed.
  681          */
  682         if (td->td_standin == NULL)
  683                 thread_alloc_spare(td);
  684 
  685         PROC_LOCK(p);
  686         mtx_lock_spin(&sched_lock);
  687         if (newkg->kg_numupcalls >= ncpus) {
  688                 mtx_unlock_spin(&sched_lock);
  689                 PROC_UNLOCK(p);
  690                 upcall_free(newku);
  691                 return (EPROCLIM);
  692         }
  693 
  694         /*
  695          * If we are the first time, and a normal thread,
  696          * then transfer all the signals back to the 'process'.
  697          * SA threading will make a special thread to handle them.
  698          */
  699         if (first && sa) {
  700                 SIGSETOR(p->p_siglist, td->td_siglist);
  701                 SIGEMPTYSET(td->td_siglist);
  702                 SIGFILLSET(td->td_sigmask);
  703                 SIG_CANTMASK(td->td_sigmask);
  704         }
  705 
  706         /*
  707          * Make the new upcall available to the ksegrp.
  708          * It may or may not use it, but it's available.
  709          */
  710         upcall_link(newku, newkg);
  711         PROC_UNLOCK(p);
  712         if (mbx.km_quantum)
  713                 newkg->kg_upquantum = max(1, mbx.km_quantum / tick);
  714 
  715         /*
  716          * Each upcall structure has an owner thread, find which
  717          * one owns it.
  718          */
  719         if (uap->newgroup) {
  720                 /*
  721                  * Because the new ksegrp hasn't a thread,
  722                  * create an initial upcall thread to own it.
  723                  */
  724                 newtd = thread_schedule_upcall(td, newku);
  725         } else {
  726                 /*
  727                  * If the current thread hasn't an upcall structure,
  728                  * just assign the upcall to it.
  729                  * It'll just return.
  730                  */
  731                 if (td->td_upcall == NULL) {
  732                         newku->ku_owner = td;
  733                         td->td_upcall = newku;
  734                         newtd = td;
  735                 } else {
  736                         /*
  737                          * Create a new upcall thread to own it.
  738                          */
  739                         newtd = thread_schedule_upcall(td, newku);
  740                 }
  741         }
  742         mtx_unlock_spin(&sched_lock);
  743 
  744         /*
  745          * Let the UTS instance know its LWPID.
  746          * It doesn't really care. But the debugger will.
  747          */
  748         suword32(&newku->ku_mailbox->km_lwp, newtd->td_tid);
  749 
  750         /*
  751          * In the same manner, if the UTS has a current user thread, 
  752          * then it is also running on this LWP so set it as well.
  753          * The library could do that of course.. but why not..
  754          */
  755         if (mbx.km_curthread)
  756                 suword32(&mbx.km_curthread->tm_lwp, newtd->td_tid);
  757 
  758         
  759         if (sa) {
  760                 newtd->td_pflags |= TDP_SA;
  761         } else {
  762                 newtd->td_pflags &= ~TDP_SA;
  763 
  764                 /*
  765                  * Since a library will use the mailbox pointer to 
  766                  * identify even a bound thread, and the mailbox pointer
  767                  * will never be allowed to change after this syscall
  768                  * for a bound thread, set it here so the library can
  769                  * find the thread after the syscall returns.
  770                  */
  771                 newtd->td_mailbox = mbx.km_curthread;
  772 
  773                 if (newtd != td) {
  774                         /*
  775                          * If we did create a new thread then
  776                          * make sure it goes to the right place
  777                          * when it starts up, and make sure that it runs 
  778                          * at full speed when it gets there. 
  779                          * thread_schedule_upcall() copies all cpu state
  780                          * to the new thread, so we should clear single step
  781                          * flag here.
  782                          */
  783                         cpu_set_upcall_kse(newtd, newku->ku_func,
  784                                 newku->ku_mailbox, &newku->ku_stack);
  785                         if (p->p_flag & P_TRACED)
  786                                 ptrace_clear_single_step(newtd);
  787                 }
  788         }
  789         
  790         /* 
  791          * If we are starting a new thread, kick it off.
  792          */
  793         if (newtd != td) {
  794                 mtx_lock_spin(&sched_lock);
  795                 setrunqueue(newtd, SRQ_BORING);
  796                 mtx_unlock_spin(&sched_lock);
  797         }
  798         return (0);
  799 }
  800 
  801 /*
  802  * Initialize global thread allocation resources.
  803  */
  804 void
  805 kseinit(void)
  806 {
  807 
  808         upcall_zone = uma_zcreate("UPCALL", sizeof(struct kse_upcall),
  809             NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0);
  810 }
  811 
  812 /*
  813  * Stash an embarasingly extra upcall into the zombie upcall queue.
  814  */
  815 
  816 void
  817 upcall_stash(struct kse_upcall *ku)
  818 {
  819         mtx_lock_spin(&kse_zombie_lock);
  820         TAILQ_INSERT_HEAD(&zombie_upcalls, ku, ku_link);
  821         mtx_unlock_spin(&kse_zombie_lock);
  822 }
  823 
  824 /*
  825  * Reap zombie kse resource.
  826  */
  827 void
  828 kse_GC(void)
  829 {
  830         struct kse_upcall *ku_first, *ku_next;
  831 
  832         /*
  833          * Don't even bother to lock if none at this instant,
  834          * we really don't care about the next instant..
  835          */
  836         if (!TAILQ_EMPTY(&zombie_upcalls)) {
  837                 mtx_lock_spin(&kse_zombie_lock);
  838                 ku_first = TAILQ_FIRST(&zombie_upcalls);
  839                 if (ku_first)
  840                         TAILQ_INIT(&zombie_upcalls);
  841                 mtx_unlock_spin(&kse_zombie_lock);
  842                 while (ku_first) {
  843                         ku_next = TAILQ_NEXT(ku_first, ku_link);
  844                         upcall_free(ku_first);
  845                         ku_first = ku_next;
  846                 }
  847         }
  848 }
  849 
  850 /*
  851  * Store the thread context in the UTS's mailbox.
  852  * then add the mailbox at the head of a list we are building in user space.
  853  * The list is anchored in the ksegrp structure.
  854  */
  855 int
  856 thread_export_context(struct thread *td, int willexit)
  857 {
  858         struct proc *p;
  859         struct ksegrp *kg;
  860         uintptr_t mbx;
  861         void *addr;
  862         int error = 0, sig;
  863         mcontext_t mc;
  864 
  865         p = td->td_proc;
  866         kg = td->td_ksegrp;
  867 
  868         /*
  869          * Post sync signal, or process SIGKILL and SIGSTOP.
  870          * For sync signal, it is only possible when the signal is not
  871          * caught by userland or process is being debugged.
  872          */
  873         PROC_LOCK(p);
  874         if (td->td_flags & TDF_NEEDSIGCHK) {
  875                 mtx_lock_spin(&sched_lock);
  876                 td->td_flags &= ~TDF_NEEDSIGCHK;
  877                 mtx_unlock_spin(&sched_lock);
  878                 mtx_lock(&p->p_sigacts->ps_mtx);
  879                 while ((sig = cursig(td)) != 0)
  880                         postsig(sig);
  881                 mtx_unlock(&p->p_sigacts->ps_mtx);
  882         }
  883         if (willexit)
  884                 SIGFILLSET(td->td_sigmask);
  885         PROC_UNLOCK(p);
  886 
  887         /* Export the user/machine context. */
  888         get_mcontext(td, &mc, 0);
  889         addr = (void *)(&td->td_mailbox->tm_context.uc_mcontext);
  890         error = copyout(&mc, addr, sizeof(mcontext_t));
  891         if (error)
  892                 goto bad;
  893 
  894         addr = (caddr_t)(&td->td_mailbox->tm_lwp);
  895         if (suword32(addr, 0)) {
  896                 error = EFAULT;
  897                 goto bad;
  898         }
  899 
  900         /* Get address in latest mbox of list pointer */
  901         addr = (void *)(&td->td_mailbox->tm_next);
  902         /*
  903          * Put the saved address of the previous first
  904          * entry into this one
  905          */
  906         for (;;) {
  907                 mbx = (uintptr_t)kg->kg_completed;
  908                 if (suword(addr, mbx)) {
  909                         error = EFAULT;
  910                         goto bad;
  911                 }
  912                 PROC_LOCK(p);
  913                 if (mbx == (uintptr_t)kg->kg_completed) {
  914                         kg->kg_completed = td->td_mailbox;
  915                         /*
  916                          * The thread context may be taken away by
  917                          * other upcall threads when we unlock
  918                          * process lock. it's no longer valid to
  919                          * use it again in any other places.
  920                          */
  921                         td->td_mailbox = NULL;
  922                         PROC_UNLOCK(p);
  923                         break;
  924                 }
  925                 PROC_UNLOCK(p);
  926         }
  927         td->td_usticks = 0;
  928         return (0);
  929 
  930 bad:
  931         PROC_LOCK(p);
  932         sigexit(td, SIGILL);
  933         return (error);
  934 }
  935 
  936 /*
  937  * Take the list of completed mailboxes for this KSEGRP and put them on this
  938  * upcall's mailbox as it's the next one going up.
  939  */
  940 static int
  941 thread_link_mboxes(struct ksegrp *kg, struct kse_upcall *ku)
  942 {
  943         struct proc *p = kg->kg_proc;
  944         void *addr;
  945         uintptr_t mbx;
  946 
  947         addr = (void *)(&ku->ku_mailbox->km_completed);
  948         for (;;) {
  949                 mbx = (uintptr_t)kg->kg_completed;
  950                 if (suword(addr, mbx)) {
  951                         PROC_LOCK(p);
  952                         psignal(p, SIGSEGV);
  953                         PROC_UNLOCK(p);
  954                         return (EFAULT);
  955                 }
  956                 PROC_LOCK(p);
  957                 if (mbx == (uintptr_t)kg->kg_completed) {
  958                         kg->kg_completed = NULL;
  959                         PROC_UNLOCK(p);
  960                         break;
  961                 }
  962                 PROC_UNLOCK(p);
  963         }
  964         return (0);
  965 }
  966 
  967 /*
  968  * This function should be called at statclock interrupt time
  969  */
  970 int
  971 thread_statclock(int user)
  972 {
  973         struct thread *td = curthread;
  974 
  975         if (!(td->td_pflags & TDP_SA))
  976                 return (0);
  977         if (user) {
  978                 /* Current always do via ast() */
  979                 mtx_lock_spin(&sched_lock);
  980                 td->td_flags |= TDF_ASTPENDING;
  981                 mtx_unlock_spin(&sched_lock);
  982                 td->td_uuticks++;
  983         } else if (td->td_mailbox != NULL)
  984                 td->td_usticks++;
  985         return (0);
  986 }
  987 
  988 /*
  989  * Export state clock ticks for userland
  990  */
  991 static int
  992 thread_update_usr_ticks(struct thread *td)
  993 {
  994         struct proc *p = td->td_proc;
  995         caddr_t addr;
  996         u_int uticks;
  997 
  998         if (td->td_mailbox == NULL)
  999                 return (-1);
 1000 
 1001         if ((uticks = td->td_uuticks) != 0) {
 1002                 td->td_uuticks = 0;
 1003                 addr = (caddr_t)&td->td_mailbox->tm_uticks;
 1004                 if (suword32(addr, uticks+fuword32(addr)))
 1005                         goto error;
 1006         }
 1007         if ((uticks = td->td_usticks) != 0) {
 1008                 td->td_usticks = 0;
 1009                 addr = (caddr_t)&td->td_mailbox->tm_sticks;
 1010                 if (suword32(addr, uticks+fuword32(addr)))
 1011                         goto error;
 1012         }
 1013         return (0);
 1014 
 1015 error:
 1016         PROC_LOCK(p);
 1017         psignal(p, SIGSEGV);
 1018         PROC_UNLOCK(p);
 1019         return (-2);
 1020 }
 1021 
 1022 /*
 1023  * This function is intended to be used to initialize a spare thread
 1024  * for upcall. Initialize thread's large data area outside sched_lock
 1025  * for thread_schedule_upcall(). The crhold is also here to get it out
 1026  * from the schedlock as it has a mutex op itself.
 1027  * XXX BUG.. we need to get the cr ref after the thread has 
 1028  * checked and chenged its own, not 6 months before...  
 1029  */
 1030 void
 1031 thread_alloc_spare(struct thread *td)
 1032 {
 1033         struct thread *spare;
 1034 
 1035         if (td->td_standin)
 1036                 return;
 1037         spare = thread_alloc();
 1038         td->td_standin = spare;
 1039         bzero(&spare->td_startzero,
 1040             __rangeof(struct thread, td_startzero, td_endzero));
 1041         spare->td_proc = td->td_proc;
 1042         spare->td_ucred = crhold(td->td_ucred);
 1043 }
 1044 
 1045 /*
 1046  * Create a thread and schedule it for upcall on the KSE given.
 1047  * Use our thread's standin so that we don't have to allocate one.
 1048  */
 1049 struct thread *
 1050 thread_schedule_upcall(struct thread *td, struct kse_upcall *ku)
 1051 {
 1052         struct thread *td2;
 1053 
 1054         mtx_assert(&sched_lock, MA_OWNED);
 1055 
 1056         /*
 1057          * Schedule an upcall thread on specified kse_upcall,
 1058          * the kse_upcall must be free.
 1059          * td must have a spare thread.
 1060          */
 1061         KASSERT(ku->ku_owner == NULL, ("%s: upcall has owner", __func__));
 1062         if ((td2 = td->td_standin) != NULL) {
 1063                 td->td_standin = NULL;
 1064         } else {
 1065                 panic("no reserve thread when scheduling an upcall");
 1066                 return (NULL);
 1067         }
 1068         CTR3(KTR_PROC, "thread_schedule_upcall: thread %p (pid %d, %s)",
 1069              td2, td->td_proc->p_pid, td->td_proc->p_comm);
 1070         /*
 1071          * Bzero already done in thread_alloc_spare() because we can't
 1072          * do the crhold here because we are in schedlock already.
 1073          */
 1074         bcopy(&td->td_startcopy, &td2->td_startcopy,
 1075             __rangeof(struct thread, td_startcopy, td_endcopy));
 1076         thread_link(td2, ku->ku_ksegrp);
 1077         /* inherit parts of blocked thread's context as a good template */
 1078         cpu_set_upcall(td2, td);
 1079         /* Let the new thread become owner of the upcall */
 1080         ku->ku_owner   = td2;
 1081         td2->td_upcall = ku;
 1082         td2->td_flags  = 0;
 1083         td2->td_pflags = TDP_SA|TDP_UPCALLING;
 1084         td2->td_state  = TDS_CAN_RUN;
 1085         td2->td_inhibitors = 0;
 1086         SIGFILLSET(td2->td_sigmask);
 1087         SIG_CANTMASK(td2->td_sigmask);
 1088         sched_fork_thread(td, td2);
 1089         return (td2);   /* bogus.. should be a void function */
 1090 }
 1091 
 1092 /*
 1093  * It is only used when thread generated a trap and process is being
 1094  * debugged.
 1095  */
 1096 void
 1097 thread_signal_add(struct thread *td, int sig)
 1098 {
 1099         struct proc *p;
 1100         siginfo_t siginfo;
 1101         struct sigacts *ps;
 1102         int error;
 1103 
 1104         p = td->td_proc;
 1105         PROC_LOCK_ASSERT(p, MA_OWNED);
 1106         ps = p->p_sigacts;
 1107         mtx_assert(&ps->ps_mtx, MA_OWNED);
 1108 
 1109         cpu_thread_siginfo(sig, 0, &siginfo);
 1110         mtx_unlock(&ps->ps_mtx);
 1111         SIGADDSET(td->td_sigmask, sig);
 1112         PROC_UNLOCK(p);
 1113         error = copyout(&siginfo, &td->td_mailbox->tm_syncsig, sizeof(siginfo));
 1114         if (error) {
 1115                 PROC_LOCK(p);
 1116                 sigexit(td, SIGSEGV);
 1117         }
 1118         PROC_LOCK(p);
 1119         mtx_lock(&ps->ps_mtx);
 1120 }
 1121 #include "opt_sched.h"
 1122 struct thread *
 1123 thread_switchout(struct thread *td, int flags, struct thread *nextthread)
 1124 {
 1125         struct kse_upcall *ku;
 1126         struct thread *td2;
 1127 
 1128         mtx_assert(&sched_lock, MA_OWNED);
 1129 
 1130         /*
 1131          * If the outgoing thread is in threaded group and has never
 1132          * scheduled an upcall, decide whether this is a short
 1133          * or long term event and thus whether or not to schedule
 1134          * an upcall.
 1135          * If it is a short term event, just suspend it in
 1136          * a way that takes its KSE with it.
 1137          * Select the events for which we want to schedule upcalls.
 1138          * For now it's just sleep or if thread is suspended but
 1139          * process wide suspending flag is not set (debugger
 1140          * suspends thread).
 1141          * XXXKSE eventually almost any inhibition could do.
 1142          */
 1143         if (TD_CAN_UNBIND(td) && (td->td_standin) &&
 1144             (TD_ON_SLEEPQ(td) || (TD_IS_SUSPENDED(td) &&
 1145              !P_SHOULDSTOP(td->td_proc)))) {
 1146                 /*
 1147                  * Release ownership of upcall, and schedule an upcall
 1148                  * thread, this new upcall thread becomes the owner of
 1149                  * the upcall structure. It will be ahead of us in the
 1150                  * run queue, so as we are stopping, it should either
 1151                  * start up immediatly, or at least before us if
 1152                  * we release our slot.
 1153                  */
 1154                 ku = td->td_upcall;
 1155                 ku->ku_owner = NULL;
 1156                 td->td_upcall = NULL;
 1157                 td->td_pflags &= ~TDP_CAN_UNBIND;
 1158                 td2 = thread_schedule_upcall(td, ku);
 1159                 if (flags & SW_INVOL || nextthread) {
 1160                         setrunqueue(td2, SRQ_YIELDING);
 1161                 } else {
 1162                         /* Keep up with reality.. we have one extra thread 
 1163                          * in the picture.. and it's 'running'.
 1164                          */
 1165                         return td2;
 1166                 }
 1167         }
 1168         return (nextthread);
 1169 }
 1170 
 1171 /*
 1172  * Setup done on the thread when it enters the kernel.
 1173  */
 1174 void
 1175 thread_user_enter(struct thread *td)
 1176 {
 1177         struct proc *p = td->td_proc;
 1178         struct ksegrp *kg;
 1179         struct kse_upcall *ku;
 1180         struct kse_thr_mailbox *tmbx;
 1181         uint32_t flags;
 1182 
 1183         /*
 1184          * First check that we shouldn't just abort. we
 1185          * can suspend it here or just exit.
 1186          */
 1187         if (__predict_false(P_SHOULDSTOP(p))) {
 1188                 PROC_LOCK(p);
 1189                 thread_suspend_check(0);
 1190                 PROC_UNLOCK(p);
 1191         }
 1192 
 1193         if (!(td->td_pflags & TDP_SA))
 1194                 return;
 1195 
 1196         /*
 1197          * If we are doing a syscall in a KSE environment,
 1198          * note where our mailbox is.
 1199          */
 1200 
 1201         kg = td->td_ksegrp;
 1202         ku = td->td_upcall;
 1203 
 1204         KASSERT(ku != NULL, ("no upcall owned"));
 1205         KASSERT(ku->ku_owner == td, ("wrong owner"));
 1206         KASSERT(!TD_CAN_UNBIND(td), ("can unbind"));
 1207 
 1208         if (td->td_standin == NULL)
 1209                 thread_alloc_spare(td);
 1210         ku->ku_mflags = fuword32((void *)&ku->ku_mailbox->km_flags);
 1211         tmbx = (void *)fuword((void *)&ku->ku_mailbox->km_curthread);
 1212         if ((tmbx == NULL) || (tmbx == (void *)-1L) ||
 1213             (ku->ku_mflags & KMF_NOUPCALL)) {
 1214                 td->td_mailbox = NULL;
 1215         } else {
 1216                 flags = fuword32(&tmbx->tm_flags);
 1217                 /*
 1218                  * On some architectures, TP register points to thread
 1219                  * mailbox but not points to kse mailbox, and userland
 1220                  * can not atomically clear km_curthread, but can
 1221                  * use TP register, and set TMF_NOUPCALL in thread
 1222                  * flag to indicate a critical region.
 1223                  */
 1224                 if (flags & TMF_NOUPCALL) {
 1225                         td->td_mailbox = NULL;
 1226                 } else {
 1227                         td->td_mailbox = tmbx;
 1228                         td->td_pflags |= TDP_CAN_UNBIND;
 1229                         if (__predict_false(p->p_flag & P_TRACED)) {
 1230                                 flags = fuword32(&tmbx->tm_dflags);
 1231                                 if (flags & TMDF_SUSPEND) {
 1232                                         mtx_lock_spin(&sched_lock);
 1233                                         /* fuword can block, check again */
 1234                                         if (td->td_upcall)
 1235                                                 ku->ku_flags |= KUF_DOUPCALL;
 1236                                         mtx_unlock_spin(&sched_lock);
 1237                                 }
 1238                         }
 1239                 }
 1240         }
 1241 }
 1242 
 1243 /*
 1244  * The extra work we go through if we are a threaded process when we
 1245  * return to userland.
 1246  *
 1247  * If we are a KSE process and returning to user mode, check for
 1248  * extra work to do before we return (e.g. for more syscalls
 1249  * to complete first).  If we were in a critical section, we should
 1250  * just return to let it finish. Same if we were in the UTS (in
 1251  * which case the mailbox's context's busy indicator will be set).
 1252  * The only traps we suport will have set the mailbox.
 1253  * We will clear it here.
 1254  */
 1255 int
 1256 thread_userret(struct thread *td, struct trapframe *frame)
 1257 {
 1258         struct kse_upcall *ku;
 1259         struct ksegrp *kg, *kg2;
 1260         struct proc *p;
 1261         struct timespec ts;
 1262         int error = 0, upcalls, uts_crit;
 1263 
 1264         /* Nothing to do with bound thread */
 1265         if (!(td->td_pflags & TDP_SA))
 1266                 return (0);
 1267 
 1268         /*
 1269          * Update stat clock count for userland
 1270          */
 1271         if (td->td_mailbox != NULL) {
 1272                 thread_update_usr_ticks(td);
 1273                 uts_crit = 0;
 1274         } else {
 1275                 uts_crit = 1;
 1276         }
 1277 
 1278         p = td->td_proc;
 1279         kg = td->td_ksegrp;
 1280         ku = td->td_upcall;
 1281 
 1282         /*
 1283          * Optimisation:
 1284          * This thread has not started any upcall.
 1285          * If there is no work to report other than ourself,
 1286          * then it can return direct to userland.
 1287          */
 1288         if (TD_CAN_UNBIND(td)) {
 1289                 td->td_pflags &= ~TDP_CAN_UNBIND;
 1290                 if ((td->td_flags & TDF_NEEDSIGCHK) == 0 &&
 1291                     (kg->kg_completed == NULL) &&
 1292                     (ku->ku_flags & KUF_DOUPCALL) == 0 &&
 1293                     (kg->kg_upquantum && ticks < kg->kg_nextupcall)) {
 1294                         nanotime(&ts);
 1295                         error = copyout(&ts,
 1296                                 (caddr_t)&ku->ku_mailbox->km_timeofday,
 1297                                 sizeof(ts));
 1298                         td->td_mailbox = 0;
 1299                         ku->ku_mflags = 0;
 1300                         if (error)
 1301                                 goto out;
 1302                         return (0);
 1303                 }
 1304                 thread_export_context(td, 0);
 1305                 /*
 1306                  * There is something to report, and we own an upcall
 1307                  * structure, we can go to userland.
 1308                  * Turn ourself into an upcall thread.
 1309                  */
 1310                 td->td_pflags |= TDP_UPCALLING;
 1311         } else if (td->td_mailbox && (ku == NULL)) {
 1312                 thread_export_context(td, 1);
 1313                 PROC_LOCK(p);
 1314                 if (kg->kg_upsleeps)
 1315                         wakeup(&kg->kg_completed);
 1316                 WITNESS_WARN(WARN_PANIC, &p->p_mtx.mtx_object,
 1317                     "thread exiting in userret");
 1318                 mtx_lock_spin(&sched_lock);
 1319                 thread_stopped(p);
 1320                 thread_exit();
 1321                 /* NOTREACHED */
 1322         }
 1323 
 1324         KASSERT(ku != NULL, ("upcall is NULL"));
 1325         KASSERT(TD_CAN_UNBIND(td) == 0, ("can unbind"));
 1326 
 1327         if (p->p_numthreads > max_threads_per_proc) {
 1328                 max_threads_hits++;
 1329                 PROC_LOCK(p);
 1330                 mtx_lock_spin(&sched_lock);
 1331                 p->p_maxthrwaits++;
 1332                 while (p->p_numthreads > max_threads_per_proc) {
 1333                         upcalls = 0;
 1334                         FOREACH_KSEGRP_IN_PROC(p, kg2) {
 1335                                 if (kg2->kg_numupcalls == 0)
 1336                                         upcalls++;
 1337                                 else
 1338                                         upcalls += kg2->kg_numupcalls;
 1339                         }
 1340                         if (upcalls >= max_threads_per_proc)
 1341                                 break;
 1342                         mtx_unlock_spin(&sched_lock);
 1343                         if (msleep(&p->p_numthreads, &p->p_mtx, PPAUSE|PCATCH,
 1344                             "maxthreads", hz/10) != EWOULDBLOCK) {
 1345                                 mtx_lock_spin(&sched_lock);
 1346                                 break;
 1347                         } else {
 1348                                 mtx_lock_spin(&sched_lock);
 1349                         }
 1350                 }
 1351                 p->p_maxthrwaits--;
 1352                 mtx_unlock_spin(&sched_lock);
 1353                 PROC_UNLOCK(p);
 1354         }
 1355 
 1356         if (td->td_pflags & TDP_UPCALLING) {
 1357                 uts_crit = 0;
 1358                 kg->kg_nextupcall = ticks + kg->kg_upquantum;
 1359                 /*
 1360                  * There is no more work to do and we are going to ride
 1361                  * this thread up to userland as an upcall.
 1362                  * Do the last parts of the setup needed for the upcall.
 1363                  */
 1364                 CTR3(KTR_PROC, "userret: upcall thread %p (pid %d, %s)",
 1365                     td, td->td_proc->p_pid, td->td_proc->p_comm);
 1366 
 1367                 td->td_pflags &= ~TDP_UPCALLING;
 1368                 if (ku->ku_flags & KUF_DOUPCALL) {
 1369                         mtx_lock_spin(&sched_lock);
 1370                         ku->ku_flags &= ~KUF_DOUPCALL;
 1371                         mtx_unlock_spin(&sched_lock);
 1372                 }
 1373                 /*
 1374                  * Set user context to the UTS
 1375                  */
 1376                 if (!(ku->ku_mflags & KMF_NOUPCALL)) {
 1377                         cpu_set_upcall_kse(td, ku->ku_func, ku->ku_mailbox,
 1378                                 &ku->ku_stack);
 1379                         if (p->p_flag & P_TRACED)
 1380                                 ptrace_clear_single_step(td);
 1381                         error = suword32(&ku->ku_mailbox->km_lwp,
 1382                                         td->td_tid);
 1383                         if (error)
 1384                                 goto out;
 1385                         error = suword(&ku->ku_mailbox->km_curthread, 0);
 1386                         if (error)
 1387                                 goto out;
 1388                 }
 1389 
 1390                 /*
 1391                  * Unhook the list of completed threads.
 1392                  * anything that completes after this gets to
 1393                  * come in next time.
 1394                  * Put the list of completed thread mailboxes on
 1395                  * this KSE's mailbox.
 1396                  */
 1397                 if (!(ku->ku_mflags & KMF_NOCOMPLETED) &&
 1398                     (error = thread_link_mboxes(kg, ku)) != 0)
 1399                         goto out;
 1400         }
 1401         if (!uts_crit) {
 1402                 nanotime(&ts);
 1403                 error = copyout(&ts, &ku->ku_mailbox->km_timeofday, sizeof(ts));
 1404         }
 1405 
 1406 out:
 1407         if (error) {
 1408                 /*
 1409                  * Things are going to be so screwed we should just kill
 1410                  * the process.
 1411                  * how do we do that?
 1412                  */
 1413                 PROC_LOCK(p);
 1414                 psignal(p, SIGSEGV);
 1415                 PROC_UNLOCK(p);
 1416         } else {
 1417                 /*
 1418                  * Optimisation:
 1419                  * Ensure that we have a spare thread available,
 1420                  * for when we re-enter the kernel.
 1421                  */
 1422                 if (td->td_standin == NULL)
 1423                         thread_alloc_spare(td);
 1424         }
 1425 
 1426         ku->ku_mflags = 0;
 1427         td->td_mailbox = NULL;
 1428         td->td_usticks = 0;
 1429         return (error); /* go sync */
 1430 }
 1431 
 1432 /*
 1433  * called after ptrace resumed a process, force all
 1434  * virtual CPUs to schedule upcall for SA process,
 1435  * because debugger may have changed something in userland,
 1436  * we should notice UTS as soon as possible.
 1437  */
 1438 void
 1439 thread_continued(struct proc *p)
 1440 {
 1441         struct ksegrp *kg;
 1442         struct kse_upcall *ku;
 1443         struct thread *td;
 1444 
 1445         PROC_LOCK_ASSERT(p, MA_OWNED);
 1446         KASSERT(P_SHOULDSTOP(p), ("process not stopped"));
 1447 
 1448         if (!(p->p_flag & P_SA))
 1449                 return;
 1450 
 1451         if (p->p_flag & P_TRACED) {
 1452                 FOREACH_KSEGRP_IN_PROC(p, kg) {
 1453                         td = TAILQ_FIRST(&kg->kg_threads);
 1454                         if (td == NULL)
 1455                                 continue;
 1456                         /* not a SA group, nothing to do */
 1457                         if (!(td->td_pflags & TDP_SA))
 1458                                 continue;
 1459                         FOREACH_UPCALL_IN_GROUP(kg, ku) {
 1460                                 mtx_lock_spin(&sched_lock);
 1461                                 ku->ku_flags |= KUF_DOUPCALL;
 1462                                 mtx_unlock_spin(&sched_lock);
 1463                                 wakeup(&kg->kg_completed);
 1464                         }
 1465                 }
 1466         }
 1467 }

Cache object: 4aa3b68f87417ded174f0d84e3c6dcf2


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.