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-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
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/7.3/sys/kern/kern_kse.c 192493 2009-05-20 22:30:57Z kmacy $");
   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/syslog.h>
   47 #include <sys/kse.h>
   48 #include <sys/ktr.h>
   49 #include <vm/uma.h>
   50 
   51 #ifdef KSE
   52 static uma_zone_t upcall_zone;
   53 
   54 /* DEBUG ONLY */
   55 extern int virtual_cpu;
   56 extern int thread_debug;
   57 
   58 extern int max_threads_per_proc;
   59 extern int max_groups_per_proc;
   60 extern int max_threads_hits;
   61 extern struct mtx kse_lock;
   62 
   63 
   64 TAILQ_HEAD(, kse_upcall) zombie_upcalls =
   65         TAILQ_HEAD_INITIALIZER(zombie_upcalls);
   66 
   67 static int thread_update_usr_ticks(struct thread *td);
   68 static int thread_alloc_spare(struct thread *td);
   69 static struct thread *thread_schedule_upcall(struct thread *td, struct kse_upcall *ku);
   70 static struct kse_upcall *upcall_alloc(void);
   71 
   72 
   73 struct mtx kse_lock;
   74 MTX_SYSINIT(kse_lock, &kse_lock, "kse lock", MTX_SPIN);
   75 
   76 struct kse_upcall *
   77 upcall_alloc(void)
   78 {
   79         struct kse_upcall *ku;
   80 
   81         ku = uma_zalloc(upcall_zone, M_WAITOK | M_ZERO);
   82         return (ku);
   83 }
   84 
   85 void
   86 upcall_reap(void)
   87 {
   88         TAILQ_HEAD(, kse_upcall) zupcalls;
   89         struct kse_upcall *ku_item, *ku_tmp;
   90 
   91         TAILQ_INIT(&zupcalls);
   92         mtx_lock_spin(&kse_lock);
   93         if (!TAILQ_EMPTY(&zombie_upcalls)) {
   94                 TAILQ_CONCAT(&zupcalls, &zombie_upcalls, ku_link);
   95                 TAILQ_INIT(&zombie_upcalls);
   96         }
   97         mtx_unlock_spin(&kse_lock);
   98         TAILQ_FOREACH_SAFE(ku_item, &zupcalls, ku_link, ku_tmp)
   99                 uma_zfree(upcall_zone, ku_item);
  100 }
  101 
  102 void
  103 upcall_remove(struct thread *td)
  104 {
  105 
  106         PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED);
  107         THREAD_LOCK_ASSERT(td, MA_OWNED);
  108         if (td->td_upcall != NULL) {
  109                 /*
  110                 * If we are not a bound thread then decrement the count of
  111                 * possible upcall sources
  112                 */
  113                 if (td->td_pflags & TDP_SA) 
  114                         td->td_proc->p_numupcalls--;
  115                 mtx_lock_spin(&kse_lock);
  116                 td->td_upcall->ku_owner = NULL;
  117                 TAILQ_REMOVE(&td->td_upcall->ku_proc->p_upcalls, td->td_upcall,
  118                     ku_link);
  119                 TAILQ_INSERT_HEAD(&zombie_upcalls, td->td_upcall, ku_link);
  120                 mtx_unlock_spin(&kse_lock);
  121                 td->td_upcall = NULL;
  122         }
  123 }
  124 #endif
  125 
  126 #ifndef _SYS_SYSPROTO_H_
  127 struct kse_switchin_args {
  128         struct kse_thr_mailbox *tmbx;
  129         int flags;
  130 };
  131 #endif
  132 
  133 #ifdef KSE
  134 void
  135 kse_unlink(struct thread *td)
  136 {
  137         mtx_lock_spin(&kse_lock);
  138         thread_unlink(td);
  139         mtx_unlock_spin(&kse_lock);
  140         upcall_remove(td);
  141 }
  142 #endif
  143 
  144 int
  145 kse_switchin(struct thread *td, struct kse_switchin_args *uap)
  146 {
  147 #ifdef KSE
  148         struct kse_thr_mailbox tmbx, *tmbxp;
  149         struct kse_upcall *ku;
  150         int error, flags;
  151 
  152         /*
  153          * Put the arguments in local variables, to allow uap to
  154          * point into the trapframe. We clobber the trapframe as
  155          * part of setting a new context.
  156          */
  157         tmbxp = uap->tmbx;
  158         flags = uap->flags;
  159 
  160         thread_lock(td);
  161         if ((ku = td->td_upcall) == NULL || TD_CAN_UNBIND(td)) {
  162                 thread_unlock(td);
  163                 return (EINVAL);
  164         }
  165         thread_unlock(td);
  166         error = (tmbxp == NULL) ? EINVAL : 0;
  167         if (!error)
  168                 error = copyin(tmbxp, &tmbx, sizeof(tmbx));
  169         if (!error && (flags & KSE_SWITCHIN_SETTMBX))
  170                 error = (suword(&ku->ku_mailbox->km_curthread,
  171                          (long)tmbxp) != 0 ? EINVAL : 0);
  172         if (!error)
  173                 error = set_mcontext(td, &tmbx.tm_context.uc_mcontext);
  174         if (!error) {
  175                 suword32(&tmbxp->tm_lwp, td->td_tid);
  176                 if (flags & KSE_SWITCHIN_SETTMBX) {
  177                         td->td_mailbox = tmbxp;
  178                         td->td_pflags |= TDP_CAN_UNBIND;
  179                 }
  180                 PROC_LOCK(td->td_proc);
  181                 if (td->td_proc->p_flag & P_TRACED) {
  182                         _PHOLD(td->td_proc);
  183                         if (tmbx.tm_dflags & TMDF_SSTEP)
  184                                 ptrace_single_step(td);
  185                         else
  186                                 ptrace_clear_single_step(td);
  187                         if (tmbx.tm_dflags & TMDF_SUSPEND) {
  188                                 thread_lock(td);
  189                                 /* fuword can block, check again */
  190                                 if (td->td_upcall)
  191                                         ku->ku_flags |= KUF_DOUPCALL;
  192                                 thread_unlock(td);
  193                         }
  194                         _PRELE(td->td_proc);
  195                 }
  196                 PROC_UNLOCK(td->td_proc);
  197         }
  198         return ((error == 0) ? EJUSTRETURN : error);
  199 #else /* !KSE */
  200         return (EOPNOTSUPP);
  201 #endif
  202 }
  203 
  204 /*
  205 struct kse_thr_interrupt_args {
  206         struct kse_thr_mailbox * tmbx;
  207         int cmd;
  208         long data;
  209 };
  210 */
  211 int
  212 kse_thr_interrupt(struct thread *td, struct kse_thr_interrupt_args *uap)
  213 {
  214 #ifdef KSE
  215         struct kse_execve_args args;
  216         struct image_args iargs;
  217         struct proc *p;
  218         struct thread *td2;
  219         struct kse_upcall *ku;
  220         struct kse_thr_mailbox *tmbx;
  221         uint32_t flags;
  222         int error;
  223 
  224         p = td->td_proc;
  225 
  226         PROC_LOCK(p);
  227         if (!(p->p_flag & P_SA)) {
  228                 PROC_UNLOCK(p);
  229                 return (EINVAL);
  230         }
  231         PROC_UNLOCK(p);
  232 
  233         switch (uap->cmd) {
  234         case KSE_INTR_SENDSIG:
  235                 if (uap->data < 0 || uap->data > _SIG_MAXSIG)
  236                         return (EINVAL);
  237         case KSE_INTR_INTERRUPT:
  238         case KSE_INTR_RESTART:
  239                 PROC_LOCK(p);
  240                 PROC_SLOCK(p);
  241                 FOREACH_THREAD_IN_PROC(p, td2) {
  242                         if (td2->td_mailbox == uap->tmbx)
  243                                 break;
  244                 }
  245                 if (td2 == NULL) {
  246                         PROC_SUNLOCK(p);
  247                         PROC_UNLOCK(p);
  248                         return (ESRCH);
  249                 }
  250                 thread_lock(td2);
  251                 PROC_SUNLOCK(p);
  252                 if (uap->cmd == KSE_INTR_SENDSIG) {
  253                         if (uap->data > 0) {
  254                                 td2->td_flags &= ~TDF_INTERRUPT;
  255                                 thread_unlock(td2);
  256                                 tdsignal(p, td2, (int)uap->data, NULL);
  257                         } else {
  258                                 thread_unlock(td2);
  259                         }
  260                 } else {
  261                         td2->td_flags |= TDF_INTERRUPT | TDF_ASTPENDING;
  262                         if (TD_CAN_UNBIND(td2))
  263                                 td2->td_upcall->ku_flags |= KUF_DOUPCALL;
  264                         if (uap->cmd == KSE_INTR_INTERRUPT)
  265                                 td2->td_intrval = EINTR;
  266                         else
  267                                 td2->td_intrval = ERESTART;
  268                         if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR))
  269                                 sleepq_abort(td2, td2->td_intrval);
  270                         thread_unlock(td2);
  271                 }
  272                 PROC_UNLOCK(p);
  273                 break;
  274         case KSE_INTR_SIGEXIT:
  275                 if (uap->data < 1 || uap->data > _SIG_MAXSIG)
  276                         return (EINVAL);
  277                 PROC_LOCK(p);
  278                 sigexit(td, (int)uap->data);
  279                 break;
  280 
  281         case KSE_INTR_DBSUSPEND:
  282                 /* this sub-function is only for bound thread */
  283                 if (td->td_pflags & TDP_SA)
  284                         return (EINVAL);
  285                 thread_lock(td);
  286                 ku = td->td_upcall;
  287                 thread_unlock(td);
  288                 tmbx = (void *)fuword((void *)&ku->ku_mailbox->km_curthread);
  289                 if (tmbx == NULL || tmbx == (void *)-1)
  290                         return (EINVAL);
  291                 flags = 0;
  292                 PROC_LOCK(p);
  293                 while ((p->p_flag & P_TRACED) && !(p->p_flag & P_SINGLE_EXIT)) {
  294                         flags = fuword32(&tmbx->tm_dflags);
  295                         if (!(flags & TMDF_SUSPEND))
  296                                 break;
  297                         PROC_SLOCK(p);
  298                         thread_stopped(p);
  299                         PROC_UNLOCK(p);
  300                         thread_lock(td);
  301                         thread_suspend_one(td);
  302                         PROC_SUNLOCK(p);
  303                         mi_switch(SW_VOL, NULL);
  304                         thread_unlock(td);
  305                         PROC_LOCK(p);
  306                 }
  307                 PROC_UNLOCK(p);
  308                 return (0);
  309 
  310         case KSE_INTR_EXECVE:
  311                 error = copyin((void *)uap->data, &args, sizeof(args));
  312                 if (error)
  313                         return (error);
  314                 error = exec_copyin_args(&iargs, args.path, UIO_USERSPACE,
  315                     args.argv, args.envp);
  316                 if (error == 0)
  317                         error = kern_execve(td, &iargs, NULL);
  318                 if (error == 0) {
  319                         PROC_LOCK(p);
  320                         SIGSETOR(td->td_siglist, args.sigpend);
  321                         PROC_UNLOCK(p);
  322                         kern_sigprocmask(td, SIG_SETMASK, &args.sigmask, NULL,
  323                             0);
  324                 }
  325                 return (error);
  326 
  327         default:
  328                 return (EINVAL);
  329         }
  330         return (0);
  331 #else /* !KSE */
  332         return (EOPNOTSUPP);
  333 #endif
  334 }
  335 
  336 /*
  337 struct kse_exit_args {
  338         register_t dummy;
  339 };
  340 */
  341 int
  342 kse_exit(struct thread *td, struct kse_exit_args *uap)
  343 {
  344 #ifdef KSE
  345         struct proc *p;
  346         struct kse_upcall *ku, *ku2;
  347         int    error, count;
  348 
  349         p = td->td_proc;
  350         /* 
  351          * Ensure that this is only called from the UTS
  352          */
  353         thread_lock(td);
  354         if ((ku = td->td_upcall) == NULL || TD_CAN_UNBIND(td)) {
  355                 thread_unlock(td);
  356                 return (EINVAL);
  357         }
  358         thread_unlock(td);
  359 
  360         /*
  361          * Calculate the existing non-exiting upcalls in this process.
  362          * If we are the last upcall but there are still other threads,
  363          * then do not exit. We need the other threads to be able to 
  364          * complete whatever they are doing.
  365          * XXX This relies on the userland knowing what to do if we return.
  366          * It may be a better choice to convert ourselves into a kse_release
  367          * ( or similar) and wait in the kernel to be needed.
  368          * XXX Where are those other threads? I suppose they are waiting in
  369          * the kernel. We should wait for them all at the user boundary after
  370          * turning into an exit.
  371          */
  372         count = 0;
  373         PROC_LOCK(p);
  374         PROC_SLOCK(p);
  375         FOREACH_UPCALL_IN_PROC(p, ku2) {
  376                 if ((ku2->ku_flags & KUF_EXITING) == 0)
  377                         count++;
  378         }
  379         if (count == 1 && (p->p_numthreads > 1)) {
  380                 PROC_SUNLOCK(p);
  381                 PROC_UNLOCK(p);
  382                 return (EDEADLK);
  383         }
  384         ku->ku_flags |= KUF_EXITING;
  385         PROC_SUNLOCK(p);
  386         PROC_UNLOCK(p);
  387 
  388         /* 
  389          * Mark the UTS mailbox as having been finished with.
  390          * If that fails then just go for a segfault.
  391          * XXX need to check it that can be deliverred without a mailbox.
  392          */
  393         error = suword32(&ku->ku_mailbox->km_flags, ku->ku_mflags|KMF_DONE);
  394         if (!(td->td_pflags & TDP_SA))
  395                 if (suword32(&td->td_mailbox->tm_lwp, 0))
  396                         error = EFAULT;
  397         PROC_LOCK(p);
  398         if (error)
  399                 psignal(p, SIGSEGV);
  400         sigqueue_flush(&td->td_sigqueue);
  401         PROC_SLOCK(p);
  402         thread_lock(td);
  403         upcall_remove(td);
  404         thread_unlock(td);
  405         if (p->p_numthreads != 1) {
  406                 thread_stopped(p);
  407                 thread_exit();
  408                 /* NOTREACHED */
  409         }
  410         /*
  411          * This is the last thread. Just return to the user.
  412          * Effectively we have left threading mode..
  413          * The only real thing left to do is ensure that the
  414          * scheduler sets out concurrency back to 1 as that may be a
  415          * resource leak otherwise.
  416          * This is an A[PB]I issue.. what SHOULD we do?
  417          * One possibility is to return to the user. It may not cope well.
  418          * The other possibility would be to let the process exit.
  419          */
  420         thread_unthread(td);
  421         PROC_SUNLOCK(p);
  422         PROC_UNLOCK(p);
  423 #if 0
  424         return (0);
  425 #else
  426         printf("kse_exit: called on last thread. Calling exit1()");
  427         exit1(td, 0);
  428 #endif
  429 #else /* !KSE */
  430         return (EOPNOTSUPP);
  431 #endif
  432 }
  433 
  434 /*
  435  * Either becomes an upcall or waits for an awakening event and
  436  * then becomes an upcall. Only error cases return.
  437  */
  438 /*
  439 struct kse_release_args {
  440         struct timespec *timeout;
  441 };
  442 */
  443 int
  444 kse_release(struct thread *td, struct kse_release_args *uap)
  445 {
  446 #ifdef KSE
  447         struct proc *p;
  448         struct kse_upcall *ku;
  449         struct timespec timeout;
  450         struct timeval tv;
  451         sigset_t sigset;
  452         int error;
  453 
  454         p = td->td_proc;
  455         thread_lock(td);
  456         if ((ku = td->td_upcall) == NULL || TD_CAN_UNBIND(td)) {
  457                 thread_unlock(td);
  458                 printf("kse_release: called outside of threading. exiting");
  459                 exit1(td, 0);
  460         }
  461         thread_unlock(td);
  462         if (uap->timeout != NULL) {
  463                 if ((error = copyin(uap->timeout, &timeout, sizeof(timeout))))
  464                         return (error);
  465                 TIMESPEC_TO_TIMEVAL(&tv, &timeout);
  466         }
  467         if (td->td_pflags & TDP_SA)
  468                 td->td_pflags |= TDP_UPCALLING;
  469         else {
  470                 ku->ku_mflags = fuword32(&ku->ku_mailbox->km_flags);
  471                 if (ku->ku_mflags == -1) {
  472                         PROC_LOCK(p);
  473                         sigexit(td, SIGSEGV);
  474                 }
  475         }
  476         PROC_LOCK(p);
  477         if (ku->ku_mflags & KMF_WAITSIGEVENT) {
  478                 /* UTS wants to wait for signal event */
  479                 if (!(p->p_flag & P_SIGEVENT) &&
  480                     !(ku->ku_flags & KUF_DOUPCALL)) {
  481                         td->td_kflags |= TDK_KSERELSIG;
  482                         error = msleep(&p->p_siglist, &p->p_mtx, PPAUSE|PCATCH,
  483                             "ksesigwait", (uap->timeout ? tvtohz(&tv) : 0));
  484                         td->td_kflags &= ~(TDK_KSERELSIG | TDK_WAKEUP);
  485                 }
  486                 p->p_flag &= ~P_SIGEVENT;
  487                 sigset = p->p_siglist;
  488                 PROC_UNLOCK(p);
  489                 error = copyout(&sigset, &ku->ku_mailbox->km_sigscaught,
  490                     sizeof(sigset));
  491         } else {
  492                 if ((ku->ku_flags & KUF_DOUPCALL) == 0 &&
  493                     ((ku->ku_mflags & KMF_NOCOMPLETED) ||
  494                      (p->p_completed == NULL))) {
  495                         p->p_upsleeps++;
  496                         td->td_kflags |= TDK_KSEREL;
  497                         error = msleep(&p->p_completed, &p->p_mtx,
  498                                 PPAUSE|PCATCH, "kserel",
  499                                 (uap->timeout ? tvtohz(&tv) : 0));
  500                         td->td_kflags &= ~(TDK_KSEREL | TDK_WAKEUP);
  501                         p->p_upsleeps--;
  502                 }
  503                 PROC_UNLOCK(p);
  504         }
  505         if (ku->ku_flags & KUF_DOUPCALL) {
  506                 PROC_SLOCK(p);
  507                 ku->ku_flags &= ~KUF_DOUPCALL;
  508                 PROC_SUNLOCK(p);
  509         }
  510         return (0);
  511 #else /* !KSE */
  512         return (EOPNOTSUPP);
  513 #endif
  514 }
  515 
  516 /* struct kse_wakeup_args {
  517         struct kse_mailbox *mbx;
  518 }; */
  519 int
  520 kse_wakeup(struct thread *td, struct kse_wakeup_args *uap)
  521 {
  522 #ifdef KSE
  523         struct proc *p;
  524         struct kse_upcall *ku;
  525         struct thread *td2;
  526 
  527         p = td->td_proc;
  528         td2 = NULL;
  529         ku = NULL;
  530         /* KSE-enabled processes only, please. */
  531         PROC_LOCK(p);
  532         if (!(p->p_flag & P_SA)) {
  533                 PROC_UNLOCK(p);
  534                 return (EINVAL);
  535         }
  536         PROC_SLOCK(p);
  537         if (uap->mbx) {
  538                 FOREACH_UPCALL_IN_PROC(p, ku) {
  539                         if (ku->ku_mailbox == uap->mbx)
  540                                 break;
  541                 }
  542         } else {
  543                 if (p->p_upsleeps) {
  544                         PROC_SUNLOCK(p);
  545                         wakeup(&p->p_completed);
  546                         PROC_UNLOCK(p);
  547                         return (0);
  548                 }
  549                 ku = TAILQ_FIRST(&p->p_upcalls);
  550         }
  551         if (ku == NULL) {
  552                 PROC_SUNLOCK(p);
  553                 PROC_UNLOCK(p);
  554                 return (ESRCH);
  555         }
  556         mtx_lock_spin(&kse_lock);
  557         if ((td2 = ku->ku_owner) == NULL) {
  558                 mtx_unlock_spin(&kse_lock);
  559                 PROC_SUNLOCK(p);
  560                 PROC_UNLOCK(p);
  561                 panic("%s: no owner", __func__);
  562         } else if (td2->td_kflags & (TDK_KSEREL | TDK_KSERELSIG)) {
  563                 mtx_unlock_spin(&kse_lock);
  564                 if (!(td2->td_kflags & TDK_WAKEUP)) {
  565                         td2->td_kflags |= TDK_WAKEUP;
  566                         if (td2->td_kflags & TDK_KSEREL)
  567                                 sleepq_remove(td2, &p->p_completed);
  568                         else
  569                                 sleepq_remove(td2, &p->p_siglist);
  570                 }
  571         } else {
  572                 ku->ku_flags |= KUF_DOUPCALL;
  573                 mtx_unlock_spin(&kse_lock);
  574         }
  575         PROC_SUNLOCK(p);
  576         PROC_UNLOCK(p);
  577         return (0);
  578 #else /* !KSE */
  579         return (EOPNOTSUPP);
  580 #endif
  581 }
  582 
  583 /*
  584  * newgroup == 0: first call: use current KSE, don't schedule an upcall
  585  * All other situations, do allocate max new KSEs and schedule an upcall.
  586  *
  587  * XXX should be changed so that 'first' behaviour lasts for as long
  588  * as you have not made a thread in this proc. i.e. as long as we do not have
  589  * a mailbox..
  590  */
  591 /* struct kse_create_args {
  592         struct kse_mailbox *mbx;
  593         int newgroup;
  594 }; */
  595 int
  596 kse_create(struct thread *td, struct kse_create_args *uap)
  597 {
  598 #ifdef KSE
  599         struct proc *p;
  600         struct kse_mailbox mbx;
  601         struct kse_upcall *newku;
  602         int err, ncpus, sa = 0, first = 0;
  603         struct thread *newtd;
  604 
  605         p = td->td_proc;
  606 
  607         /*
  608          * Processes using the other threading model can't
  609          * suddenly start calling this one
  610          * XXX  maybe...
  611          */
  612         PROC_LOCK(p);
  613         if ((p->p_flag & (P_SA|P_HADTHREADS)) == P_HADTHREADS) {
  614                 PROC_UNLOCK(p);
  615                 return (EINVAL);
  616         }
  617         if (!(p->p_flag & P_SA)) {
  618                 first = 1;
  619                 p->p_flag |= P_SA|P_HADTHREADS;
  620         }
  621         PROC_UNLOCK(p);
  622 
  623         if ((err = copyin(uap->mbx, &mbx, sizeof(mbx))))
  624                 return (err);
  625 
  626         ncpus = mp_ncpus;
  627         if (virtual_cpu != 0)
  628                 ncpus = virtual_cpu;
  629         /*
  630          * If the new UTS mailbox says that this
  631          * will be a BOUND lwp, then it had better
  632          * have its thread mailbox already there.
  633          */
  634         if ((mbx.km_flags & KMF_BOUND) || uap->newgroup) {
  635                 /* It's a bound thread (1:1) */
  636                 if (mbx.km_curthread == NULL) 
  637                         return (EINVAL);
  638                 ncpus = 1;
  639                 if (!(uap->newgroup || first))
  640                         return (EINVAL);
  641         } else {
  642                 /* It's an upcall capable thread */
  643                 sa = TDP_SA;
  644                 PROC_LOCK(p);
  645                 /*
  646                  * Limit it to NCPU upcall contexts per proc in any case.
  647                  * numupcalls will soon be numkse or something
  648                  * as it will represent the number of 
  649                  * non-bound upcalls available.  (i.e. ones that can 
  650                  * actually call up).
  651                  */
  652                 if (p->p_numupcalls >= ncpus) {
  653                         PROC_UNLOCK(p);
  654                         return (EPROCLIM);
  655                 }
  656                 p->p_numupcalls++;
  657                 PROC_UNLOCK(p);
  658         }
  659 
  660         /*
  661          * For the first call this may not have been set.
  662          * Of course nor may it actually be needed.
  663          * thread_schedule_upcall() will look for it.
  664          */
  665         if (td->td_standin == NULL) {
  666                 if (!thread_alloc_spare(td))
  667                         return (ENOMEM);
  668         }
  669 
  670         /* 
  671          * Even bound LWPs get a mailbox and an upcall to hold it.
  672          * XXX This should change.
  673          */
  674         newku = upcall_alloc();
  675         newku->ku_mailbox = uap->mbx;
  676         newku->ku_func = mbx.km_func;
  677         bcopy(&mbx.km_stack, &newku->ku_stack, sizeof(stack_t));
  678 
  679         PROC_LOCK(p);
  680         PROC_SLOCK(p);
  681         /*
  682          * If we are the first time, and a normal thread,
  683          * then transfer all the signals back to the 'process'.
  684          * SA threading will make a special thread to handle them.
  685          */
  686         if (first) {
  687                 sigqueue_move_set(&td->td_sigqueue, &p->p_sigqueue, 
  688                         &td->td_sigqueue.sq_signals);
  689                 SIGFILLSET(td->td_sigmask);
  690                 SIG_CANTMASK(td->td_sigmask);
  691         }
  692 
  693         /*
  694          * Make the new upcall available to the process.
  695          * It may or may not use it, but it's available.
  696          */
  697         TAILQ_INSERT_TAIL(&p->p_upcalls, newku, ku_link);
  698         newku->ku_proc = p;
  699         PROC_UNLOCK(p);
  700         if (mbx.km_quantum)
  701 /* XXX should this be in the thread? */
  702                 p->p_upquantum = max(1, mbx.km_quantum / tick);
  703 
  704         /*
  705          * Each upcall structure has an owner thread, find which
  706          * one owns it.
  707          */
  708         thread_lock(td);
  709         mtx_lock_spin(&kse_lock);
  710         if (uap->newgroup) {
  711                 /*
  712                  * The newgroup parameter now means
  713                  * "bound, non SA, system scope"
  714                  * It is only used for the interrupt thread at the
  715                  * moment I think.. (or system scope threads dopey).
  716                  * We'll rename it later.
  717                  */
  718                 newtd = thread_schedule_upcall(td, newku);
  719         } else {
  720                 /*
  721                  * If the current thread hasn't an upcall structure,
  722                  * just assign the upcall to it.
  723                  * It'll just return.
  724                  */
  725                 if (td->td_upcall == NULL) {
  726                         newku->ku_owner = td;
  727                         td->td_upcall = newku;
  728                         newtd = td;
  729                 } else {
  730                         /*
  731                          * Create a new upcall thread to own it.
  732                          */
  733                         newtd = thread_schedule_upcall(td, newku);
  734                 }
  735         }
  736         mtx_unlock_spin(&kse_lock);
  737         thread_unlock(td);
  738         PROC_SUNLOCK(p);
  739 
  740         /*
  741          * Let the UTS instance know its LWPID.
  742          * It doesn't really care. But the debugger will.
  743          * XXX warning.. remember that this moves.
  744          */
  745         suword32(&newku->ku_mailbox->km_lwp, newtd->td_tid);
  746 
  747         /*
  748          * In the same manner, if the UTS has a current user thread, 
  749          * then it is also running on this LWP so set it as well.
  750          * The library could do that of course.. but why not..
  751          * XXX I'm not sure this can ever happen but ...
  752          * XXX does the UTS ever set this in the mailbox before calling this?
  753          */
  754         if (mbx.km_curthread)
  755                 suword32(&mbx.km_curthread->tm_lwp, newtd->td_tid);
  756         
  757         if (sa) {
  758                 newtd->td_pflags |= TDP_SA;
  759                 /* 
  760                  * If we are starting a new thread, kick it off.
  761                  */
  762                 if (newtd != td) {
  763                         thread_lock(newtd);
  764                         sched_add(newtd, SRQ_BORING);
  765                         thread_unlock(newtd);
  766                 }
  767         } else {
  768                 newtd->td_pflags &= ~TDP_SA;
  769 
  770                 /*
  771                  * Since a library will use the mailbox pointer to 
  772                  * identify even a bound thread, and the mailbox pointer
  773                  * will never be allowed to change after this syscall
  774                  * for a bound thread, set it here so the library can
  775                  * find the thread after the syscall returns.
  776                  */
  777                 newtd->td_mailbox = mbx.km_curthread;
  778 
  779                 if (newtd != td) {
  780                         /*
  781                          * If we did create a new thread then
  782                          * make sure it goes to the right place
  783                          * when it starts up, and make sure that it runs 
  784                          * at full speed when it gets there. 
  785                          * thread_schedule_upcall() copies all cpu state
  786                          * to the new thread, so we should clear single step
  787                          * flag here.
  788                          */
  789                         cpu_set_upcall_kse(newtd, newku->ku_func,
  790                                 newku->ku_mailbox, &newku->ku_stack);
  791                         PROC_LOCK(p);
  792                         if (p->p_flag & P_TRACED) {
  793                                 _PHOLD(p);
  794                                 ptrace_clear_single_step(newtd);
  795                                 _PRELE(p);
  796                         }
  797                         PROC_UNLOCK(p);
  798                         thread_lock(newtd);
  799                         sched_add(newtd, SRQ_BORING);
  800                         thread_unlock(newtd);
  801                 }
  802         }
  803         return (0);
  804 #else /* !KSE */
  805         return (EOPNOTSUPP);
  806 #endif
  807 }
  808 
  809 #ifdef KSE
  810 /*
  811  * Initialize global thread allocation resources.
  812  */
  813 void
  814 kseinit(void)
  815 {
  816 
  817         upcall_zone = uma_zcreate("UPCALL", sizeof(struct kse_upcall),
  818             NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0);
  819 }
  820 
  821 /*
  822  * Store the thread context in the UTS's mailbox.
  823  * then add the mailbox at the head of a list we are building in user space.
  824  * The list is anchored in the proc structure.
  825  */
  826 int
  827 thread_export_context(struct thread *td, int willexit)
  828 {
  829         struct proc *p;
  830         uintptr_t mbx;
  831         void *addr;
  832         int error = 0, sig;
  833         mcontext_t mc;
  834 
  835         p = td->td_proc;
  836 
  837         /*
  838          * Post sync signal, or process SIGKILL and SIGSTOP.
  839          * For sync signal, it is only possible when the signal is not
  840          * caught by userland or process is being debugged.
  841          */
  842         PROC_LOCK(p);
  843         if (td->td_flags & TDF_NEEDSIGCHK) {
  844                 thread_lock(td);
  845                 td->td_flags &= ~TDF_NEEDSIGCHK;
  846                 thread_unlock(td);
  847                 mtx_lock(&p->p_sigacts->ps_mtx);
  848                 while ((sig = cursig(td)) != 0)
  849                         postsig(sig);
  850                 mtx_unlock(&p->p_sigacts->ps_mtx);
  851         }
  852         if (willexit)
  853                 SIGFILLSET(td->td_sigmask);
  854         PROC_UNLOCK(p);
  855 
  856         /* Export the user/machine context. */
  857         get_mcontext(td, &mc, 0);
  858         addr = (void *)(&td->td_mailbox->tm_context.uc_mcontext);
  859         error = copyout(&mc, addr, sizeof(mcontext_t));
  860         if (error)
  861                 goto bad;
  862 
  863         addr = (caddr_t)(&td->td_mailbox->tm_lwp);
  864         if (suword32(addr, 0)) {
  865                 error = EFAULT;
  866                 goto bad;
  867         }
  868 
  869         /* Get address in latest mbox of list pointer */
  870         addr = (void *)(&td->td_mailbox->tm_next);
  871         /*
  872          * Put the saved address of the previous first
  873          * entry into this one
  874          */
  875         for (;;) {
  876                 mbx = (uintptr_t)p->p_completed;
  877                 if (suword(addr, mbx)) {
  878                         error = EFAULT;
  879                         goto bad;
  880                 }
  881                 PROC_LOCK(p);
  882                 if (mbx == (uintptr_t)p->p_completed) {
  883                         thread_lock(td);
  884                         p->p_completed = td->td_mailbox;
  885                         /*
  886                          * The thread context may be taken away by
  887                          * other upcall threads when we unlock
  888                          * process lock. it's no longer valid to
  889                          * use it again in any other places.
  890                          */
  891                         td->td_mailbox = NULL;
  892                         thread_unlock(td);
  893                         PROC_UNLOCK(p);
  894                         break;
  895                 }
  896                 PROC_UNLOCK(p);
  897         }
  898         td->td_usticks = 0;
  899         return (0);
  900 
  901 bad:
  902         PROC_LOCK(p);
  903         sigexit(td, SIGILL);
  904         return (error);
  905 }
  906 
  907 /*
  908  * Take the list of completed mailboxes for this Process and put them on this
  909  * upcall's mailbox as it's the next one going up.
  910  */
  911 static int
  912 thread_link_mboxes(struct proc *p, struct kse_upcall *ku)
  913 {
  914         void *addr;
  915         uintptr_t mbx;
  916 
  917         addr = (void *)(&ku->ku_mailbox->km_completed);
  918         for (;;) {
  919                 mbx = (uintptr_t)p->p_completed;
  920                 if (suword(addr, mbx)) {
  921                         PROC_LOCK(p);
  922                         psignal(p, SIGSEGV);
  923                         PROC_UNLOCK(p);
  924                         return (EFAULT);
  925                 }
  926                 PROC_LOCK(p);
  927                 if (mbx == (uintptr_t)p->p_completed) {
  928                         p->p_completed = NULL;
  929                         PROC_UNLOCK(p);
  930                         break;
  931                 }
  932                 PROC_UNLOCK(p);
  933         }
  934         return (0);
  935 }
  936 
  937 /*
  938  * This function should be called at statclock interrupt time
  939  */
  940 int
  941 thread_statclock(int user)
  942 {
  943         struct thread *td = curthread;
  944 
  945         if (!(td->td_pflags & TDP_SA))
  946                 return (0);
  947         if (user) {
  948                 /* Current always do via ast() */
  949                 thread_lock(td);
  950                 td->td_flags |= TDF_ASTPENDING;
  951                 thread_unlock(td);
  952                 td->td_uuticks++;
  953         } else if (td->td_mailbox != NULL)
  954                 td->td_usticks++;
  955         return (0);
  956 }
  957 
  958 /*
  959  * Export state clock ticks for userland
  960  */
  961 static int
  962 thread_update_usr_ticks(struct thread *td)
  963 {
  964         struct proc *p = td->td_proc;
  965         caddr_t addr;
  966         u_int uticks;
  967 
  968         thread_lock(td);
  969         if (td->td_mailbox == NULL) {
  970                 thread_unlock(td);
  971                 return (-1);
  972         }
  973         thread_unlock(td);
  974 
  975         if ((uticks = td->td_uuticks) != 0) {
  976                 td->td_uuticks = 0;
  977                 addr = (caddr_t)&td->td_mailbox->tm_uticks;
  978                 if (suword32(addr, uticks+fuword32(addr)))
  979                         goto error;
  980         }
  981         if ((uticks = td->td_usticks) != 0) {
  982                 td->td_usticks = 0;
  983                 addr = (caddr_t)&td->td_mailbox->tm_sticks;
  984                 if (suword32(addr, uticks+fuword32(addr)))
  985                         goto error;
  986         }
  987         return (0);
  988 
  989 error:
  990         PROC_LOCK(p);
  991         psignal(p, SIGSEGV);
  992         PROC_UNLOCK(p);
  993         return (-2);
  994 }
  995 
  996 /*
  997  * This function is intended to be used to initialize a spare thread
  998  * for upcall. Initialize thread's large data area outside the thread lock
  999  * for thread_schedule_upcall(). The crhold is also here to get it out
 1000  * from the schedlock as it has a mutex op itself.
 1001  * XXX BUG.. we need to get the cr ref after the thread has 
 1002  * checked and chenged its own, not 6 months before...  
 1003  */
 1004 int
 1005 thread_alloc_spare(struct thread *td)
 1006 {
 1007         struct thread *spare;
 1008 
 1009         if (td->td_standin)
 1010                 return (1);
 1011         spare = thread_alloc();
 1012         if (spare == NULL)
 1013                 return (0);
 1014         td->td_standin = spare;
 1015         bzero(&spare->td_startzero,
 1016             __rangeof(struct thread, td_startzero, td_endzero));
 1017         bzero(&spare->td_osd, sizeof(struct osd));
 1018         spare->td_proc = td->td_proc;
 1019         spare->td_ucred = crhold(td->td_ucred);
 1020         spare->td_flags = TDF_INMEM;
 1021         return (1);
 1022 }
 1023 
 1024 /*
 1025  * Create a thread and schedule it for upcall on the KSE given.
 1026  * Use our thread's standin so that we don't have to allocate one.
 1027  */
 1028 struct thread *
 1029 thread_schedule_upcall(struct thread *td, struct kse_upcall *ku)
 1030 {
 1031         struct thread *td2;
 1032 
 1033         THREAD_LOCK_ASSERT(td, MA_OWNED);
 1034         mtx_assert(&kse_lock, MA_OWNED);
 1035         /*
 1036          * Schedule an upcall thread on specified kse_upcall,
 1037          * the kse_upcall must be free.
 1038          * td must have a spare thread.
 1039          */
 1040         KASSERT(ku->ku_owner == NULL, ("%s: upcall has owner", __func__));
 1041         if ((td2 = td->td_standin) != NULL) {
 1042                 td->td_standin = NULL;
 1043         } else {
 1044                 panic("no reserve thread when scheduling an upcall");
 1045                 return (NULL);
 1046         }
 1047         CTR3(KTR_PROC, "thread_schedule_upcall: thread %p (pid %d, %s)",
 1048              td2, td->td_proc->p_pid, td->td_proc->p_comm);
 1049         /*
 1050          * Bzero already done in thread_alloc_spare() because we can't
 1051          * do the crhold here because we are in schedlock already.
 1052          */
 1053         bcopy(&td->td_startcopy, &td2->td_startcopy,
 1054             __rangeof(struct thread, td_startcopy, td_endcopy));
 1055         sched_fork_thread(td, td2);
 1056         thread_link(td2, ku->ku_proc);
 1057         /* inherit parts of blocked thread's context as a good template */
 1058         cpu_set_upcall(td2, td);
 1059         /* Let the new thread become owner of the upcall */
 1060         ku->ku_owner   = td2;
 1061         td2->td_upcall = ku;
 1062         td2->td_pflags = TDP_SA|TDP_UPCALLING;
 1063         td2->td_state  = TDS_CAN_RUN;
 1064         td2->td_inhibitors = 0;
 1065         SIGFILLSET(td2->td_sigmask);
 1066         SIG_CANTMASK(td2->td_sigmask);
 1067         return (td2);   /* bogus.. should be a void function */
 1068 }
 1069 
 1070 /*
 1071  * It is only used when thread generated a trap and process is being
 1072  * debugged.
 1073  */
 1074 void
 1075 thread_signal_add(struct thread *td, ksiginfo_t *ksi)
 1076 {
 1077         struct proc *p;
 1078         struct sigacts *ps;
 1079         int error;
 1080 
 1081         p = td->td_proc;
 1082         PROC_LOCK_ASSERT(p, MA_OWNED);
 1083         ps = p->p_sigacts;
 1084         mtx_assert(&ps->ps_mtx, MA_OWNED);
 1085 
 1086         mtx_unlock(&ps->ps_mtx);
 1087         SIGADDSET(td->td_sigmask, ksi->ksi_signo);
 1088         PROC_UNLOCK(p);
 1089         error = copyout(&ksi->ksi_info, &td->td_mailbox->tm_syncsig,
 1090                         sizeof(siginfo_t));
 1091         if (error) {
 1092                 PROC_LOCK(p);
 1093                 sigexit(td, SIGSEGV);
 1094         }
 1095         PROC_LOCK(p);
 1096         mtx_lock(&ps->ps_mtx);
 1097 }
 1098 #include "opt_sched.h"
 1099 struct thread *
 1100 thread_switchout(struct thread *td, int flags, struct thread *nextthread)
 1101 {
 1102         struct kse_upcall *ku;
 1103         struct thread *td2;
 1104 
 1105         THREAD_LOCK_ASSERT(td, MA_OWNED);
 1106 
 1107         /*
 1108          * If the outgoing thread is in threaded group and has never
 1109          * scheduled an upcall, decide whether this is a short
 1110          * or long term event and thus whether or not to schedule
 1111          * an upcall.
 1112          * If it is a short term event, just suspend it in
 1113          * a way that takes its KSE with it.
 1114          * Select the events for which we want to schedule upcalls.
 1115          * For now it's just sleep or if thread is suspended but
 1116          * process wide suspending flag is not set (debugger
 1117          * suspends thread).
 1118          * XXXKSE eventually almost any inhibition could do.
 1119          */
 1120         if (TD_CAN_UNBIND(td) && (td->td_standin) &&
 1121             (TD_ON_SLEEPQ(td) || (TD_IS_SUSPENDED(td) &&
 1122              !P_SHOULDSTOP(td->td_proc)))) {
 1123                 /*
 1124                  * Release ownership of upcall, and schedule an upcall
 1125                  * thread, this new upcall thread becomes the owner of
 1126                  * the upcall structure. It will be ahead of us in the
 1127                  * run queue, so as we are stopping, it should either
 1128                  * start up immediatly, or at least before us if
 1129                  * we release our slot.
 1130                  */
 1131                 mtx_lock_spin(&kse_lock);
 1132                 ku = td->td_upcall;
 1133                 ku->ku_owner = NULL;
 1134                 td->td_upcall = NULL;
 1135                 td->td_pflags &= ~TDP_CAN_UNBIND;
 1136                 td2 = thread_schedule_upcall(td, ku);
 1137                 mtx_unlock_spin(&kse_lock);
 1138                 if (flags & SW_INVOL || nextthread) {
 1139                         thread_lock(td2);
 1140                         sched_add(td2, SRQ_YIELDING);
 1141                         thread_unlock(td2);
 1142                 } else {
 1143                         /* Keep up with reality.. we have one extra thread 
 1144                          * in the picture.. and it's 'running'.
 1145                          */
 1146                         return td2;
 1147                 }
 1148         }
 1149         return (nextthread);
 1150 }
 1151 
 1152 /*
 1153  * Setup done on the thread when it enters the kernel.
 1154  */
 1155 void
 1156 thread_user_enter(struct thread *td)
 1157 {
 1158         struct proc *p = td->td_proc;
 1159         struct kse_upcall *ku;
 1160         struct kse_thr_mailbox *tmbx;
 1161         uint32_t flags;
 1162 
 1163         /*
 1164          * First check that we shouldn't just abort. we
 1165          * can suspend it here or just exit.
 1166          */
 1167         if (__predict_false(P_SHOULDSTOP(p))) {
 1168                 PROC_LOCK(p);
 1169                 thread_suspend_check(0);
 1170                 PROC_UNLOCK(p);
 1171         }
 1172 
 1173         if (!(td->td_pflags & TDP_SA))
 1174                 return;
 1175 
 1176         /*
 1177          * If we are doing a syscall in a KSE environment,
 1178          * note where our mailbox is.
 1179          */
 1180 
 1181         thread_lock(td);
 1182         ku = td->td_upcall;
 1183         thread_unlock(td);
 1184 
 1185         KASSERT(ku != NULL, ("no upcall owned"));
 1186         KASSERT(ku->ku_owner == td, ("wrong owner"));
 1187         KASSERT(!TD_CAN_UNBIND(td), ("can unbind"));
 1188 
 1189         if (td->td_standin == NULL) {
 1190                 if (!thread_alloc_spare(td)) {
 1191                         PROC_LOCK(p);
 1192                         if (kern_logsigexit)
 1193                                 log(LOG_INFO,
 1194                                     "pid %d (%s), uid %d: thread_alloc_spare failed\n",
 1195                                     p->p_pid, p->p_comm,
 1196                                     td->td_ucred ? td->td_ucred->cr_uid : -1);
 1197                         sigexit(td, SIGSEGV);   /* XXX ? */
 1198                         /* panic("thread_user_enter: thread_alloc_spare failed"); */
 1199                 }
 1200         }
 1201         ku->ku_mflags = fuword32((void *)&ku->ku_mailbox->km_flags);
 1202         tmbx = (void *)fuword((void *)&ku->ku_mailbox->km_curthread);
 1203         if ((tmbx == NULL) || (tmbx == (void *)-1L) ||
 1204             (ku->ku_mflags & KMF_NOUPCALL)) {
 1205                 td->td_mailbox = NULL;
 1206         } else {
 1207                 flags = fuword32(&tmbx->tm_flags);
 1208                 /*
 1209                  * On some architectures, TP register points to thread
 1210                  * mailbox but not points to kse mailbox, and userland
 1211                  * can not atomically clear km_curthread, but can
 1212                  * use TP register, and set TMF_NOUPCALL in thread
 1213                  * flag to indicate a critical region.
 1214                  */
 1215                 if (flags & TMF_NOUPCALL) {
 1216                         td->td_mailbox = NULL;
 1217                 } else {
 1218                         td->td_mailbox = tmbx;
 1219                         td->td_pflags |= TDP_CAN_UNBIND;
 1220                         PROC_LOCK(p);
 1221                         if (__predict_false(p->p_flag & P_TRACED)) {
 1222                                 flags = fuword32(&tmbx->tm_dflags);
 1223                                 if (flags & TMDF_SUSPEND) {
 1224                                         thread_lock(td);
 1225                                         /* fuword can block, check again */
 1226                                         if (td->td_upcall)
 1227                                                 ku->ku_flags |= KUF_DOUPCALL;
 1228                                         thread_unlock(td);
 1229                                 }
 1230                         }
 1231                         PROC_UNLOCK(p);
 1232                 }
 1233         }
 1234 }
 1235 
 1236 /*
 1237  * The extra work we go through if we are a threaded process when we
 1238  * return to userland.
 1239  *
 1240  * If we are a KSE process and returning to user mode, check for
 1241  * extra work to do before we return (e.g. for more syscalls
 1242  * to complete first).  If we were in a critical section, we should
 1243  * just return to let it finish. Same if we were in the UTS (in
 1244  * which case the mailbox's context's busy indicator will be set).
 1245  * The only traps we suport will have set the mailbox.
 1246  * We will clear it here.
 1247  */
 1248 int
 1249 thread_userret(struct thread *td, struct trapframe *frame)
 1250 {
 1251         struct kse_upcall *ku;
 1252         struct proc *p;
 1253         struct timespec ts;
 1254         int error = 0, uts_crit;
 1255 
 1256         /* Nothing to do with bound thread */
 1257         if (!(td->td_pflags & TDP_SA))
 1258                 return (0);
 1259 
 1260         /*
 1261          * Update stat clock count for userland
 1262          */
 1263         if (td->td_mailbox != NULL) {
 1264                 thread_update_usr_ticks(td);
 1265                 uts_crit = 0;
 1266         } else {
 1267                 uts_crit = 1;
 1268         }
 1269 
 1270         p = td->td_proc;
 1271         thread_lock(td);
 1272         ku = td->td_upcall;
 1273 
 1274         /*
 1275          * Optimisation:
 1276          * This thread has not started any upcall.
 1277          * If there is no work to report other than ourself,
 1278          * then it can return direct to userland.
 1279          */
 1280         if (TD_CAN_UNBIND(td)) {
 1281                 thread_unlock(td);
 1282                 td->td_pflags &= ~TDP_CAN_UNBIND;
 1283                 if ((td->td_flags & TDF_NEEDSIGCHK) == 0 &&
 1284                     (p->p_completed == NULL) &&
 1285                     (ku->ku_flags & KUF_DOUPCALL) == 0 &&
 1286                     (p->p_upquantum && ticks < p->p_nextupcall)) {
 1287                         nanotime(&ts);
 1288                         error = copyout(&ts,
 1289                                 (caddr_t)&ku->ku_mailbox->km_timeofday,
 1290                                 sizeof(ts));
 1291                         td->td_mailbox = 0;
 1292                         ku->ku_mflags = 0;
 1293                         if (error)
 1294                                 goto out;
 1295                         return (0);
 1296                 }
 1297                 thread_export_context(td, 0);
 1298                 /*
 1299                  * There is something to report, and we own an upcall
 1300                  * structure, we can go to userland.
 1301                  * Turn ourself into an upcall thread.
 1302                  */
 1303                 td->td_pflags |= TDP_UPCALLING;
 1304         } else if (td->td_mailbox && (ku == NULL)) {
 1305                 thread_unlock(td);
 1306                 thread_export_context(td, 1);
 1307                 PROC_LOCK(p);
 1308                 if (p->p_upsleeps)
 1309                         wakeup(&p->p_completed);
 1310                 WITNESS_WARN(WARN_PANIC, &p->p_mtx.lock_object,
 1311                     "thread exiting in userret");
 1312                 sigqueue_flush(&td->td_sigqueue);
 1313                 PROC_SLOCK(p);
 1314                 thread_stopped(p);
 1315                 thread_exit();
 1316                 /* NOTREACHED */
 1317         } else
 1318                 thread_unlock(td);
 1319 
 1320         KASSERT(ku != NULL, ("upcall is NULL"));
 1321         KASSERT(TD_CAN_UNBIND(td) == 0, ("can unbind"));
 1322 
 1323         PROC_LOCK(p);
 1324         PROC_SLOCK(p);
 1325         if (p->p_numthreads > max_threads_per_proc) {
 1326                 max_threads_hits++;
 1327                 while (p->p_numthreads > max_threads_per_proc) {
 1328                         if (p->p_numupcalls >= max_threads_per_proc)
 1329                                 break;
 1330                         PROC_SUNLOCK(p);
 1331                         if (msleep(&p->p_numthreads, &p->p_mtx, PPAUSE|PCATCH,
 1332                             "maxthreads", hz/10) != EWOULDBLOCK) {
 1333                                 PROC_SLOCK(p);
 1334                                 break;
 1335                         } else
 1336                                 PROC_SLOCK(p);
 1337                 }
 1338         }
 1339         PROC_SUNLOCK(p);
 1340         PROC_UNLOCK(p);
 1341 
 1342         if (td->td_pflags & TDP_UPCALLING) {
 1343                 uts_crit = 0;
 1344                 p->p_nextupcall = ticks + p->p_upquantum;
 1345                 /*
 1346                  * There is no more work to do and we are going to ride
 1347                  * this thread up to userland as an upcall.
 1348                  * Do the last parts of the setup needed for the upcall.
 1349                  */
 1350                 CTR3(KTR_PROC, "userret: upcall thread %p (pid %d, %s)",
 1351                     td, td->td_proc->p_pid, td->td_proc->p_comm);
 1352 
 1353                 td->td_pflags &= ~TDP_UPCALLING;
 1354                 if (ku->ku_flags & KUF_DOUPCALL) {
 1355                         PROC_SLOCK(p);
 1356                         ku->ku_flags &= ~KUF_DOUPCALL;
 1357                         PROC_SUNLOCK(p);
 1358                 }
 1359                 /*
 1360                  * Set user context to the UTS
 1361                  */
 1362                 if (!(ku->ku_mflags & KMF_NOUPCALL)) {
 1363                         cpu_set_upcall_kse(td, ku->ku_func, ku->ku_mailbox,
 1364                                 &ku->ku_stack);
 1365                         PROC_LOCK(p);
 1366                         if (p->p_flag & P_TRACED) {
 1367                                 _PHOLD(p);
 1368                                 ptrace_clear_single_step(td);
 1369                                 _PRELE(p);
 1370                         }
 1371                         PROC_UNLOCK(p);
 1372                         error = suword32(&ku->ku_mailbox->km_lwp,
 1373                                         td->td_tid);
 1374                         if (error)
 1375                                 goto out;
 1376                         error = suword(&ku->ku_mailbox->km_curthread, 0);
 1377                         if (error)
 1378                                 goto out;
 1379                 }
 1380 
 1381                 /*
 1382                  * Unhook the list of completed threads.
 1383                  * anything that completes after this gets to
 1384                  * come in next time.
 1385                  * Put the list of completed thread mailboxes on
 1386                  * this KSE's mailbox.
 1387                  */
 1388                 if (!(ku->ku_mflags & KMF_NOCOMPLETED) &&
 1389                     (error = thread_link_mboxes(p, ku)) != 0)
 1390                         goto out;
 1391         }
 1392         if (!uts_crit) {
 1393                 nanotime(&ts);
 1394                 error = copyout(&ts, &ku->ku_mailbox->km_timeofday, sizeof(ts));
 1395         }
 1396 
 1397 out:
 1398         if (error) {
 1399                 /*
 1400                  * Things are going to be so screwed we should just kill
 1401                  * the process.
 1402                  * how do we do that?
 1403                  */
 1404                 PROC_LOCK(p);
 1405                 psignal(p, SIGSEGV);
 1406                 PROC_UNLOCK(p);
 1407         } else {
 1408                 /*
 1409                  * Optimisation:
 1410                  * Ensure that we have a spare thread available,
 1411                  * for when we re-enter the kernel.
 1412                  */
 1413                 if (td->td_standin == NULL)
 1414                         thread_alloc_spare(td); /* XXX care of failure ? */
 1415         }
 1416 
 1417         ku->ku_mflags = 0;
 1418         td->td_mailbox = NULL;
 1419         td->td_usticks = 0;
 1420         return (error); /* go sync */
 1421 }
 1422 
 1423 /*
 1424  * called after ptrace resumed a process, force all
 1425  * virtual CPUs to schedule upcall for SA process,
 1426  * because debugger may have changed something in userland,
 1427  * we should notice UTS as soon as possible.
 1428  */
 1429 void
 1430 thread_continued(struct proc *p)
 1431 {
 1432         struct kse_upcall *ku;
 1433         struct thread *td;
 1434 
 1435         PROC_LOCK_ASSERT(p, MA_OWNED);
 1436         KASSERT(P_SHOULDSTOP(p), ("process not stopped"));
 1437 
 1438         if (!(p->p_flag & P_SA))
 1439                 return;
 1440 
 1441         if (p->p_flag & P_TRACED) {
 1442                 td = TAILQ_FIRST(&p->p_threads);
 1443                 if (td && (td->td_pflags & TDP_SA)) {
 1444                         FOREACH_UPCALL_IN_PROC(p, ku) {
 1445                                 PROC_SLOCK(p);
 1446                                 ku->ku_flags |= KUF_DOUPCALL;
 1447                                 PROC_SUNLOCK(p);
 1448                                 wakeup(&p->p_completed);
 1449                         }
 1450                 }
 1451         }
 1452 }
 1453 #endif

Cache object: 87c0d4eaa2500e95caf087c37a90e6da


[ 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.