Index: head/sys/kern/kern_kse.c =================================================================== --- head/sys/kern/kern_kse.c (revision 113640) +++ head/sys/kern/kern_kse.c (revision 113641) @@ -1,2082 +1,2085 @@ /* * Copyright (C) 2001 Julian Elischer . * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice(s), this list of conditions and the following disclaimer as * the first lines of this file unmodified other than the possible * addition of one or more copyright notices. * 2. Redistributions in binary form must reproduce the above copyright * notice(s), this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * KSEGRP related storage. */ static uma_zone_t ksegrp_zone; static uma_zone_t kse_zone; static uma_zone_t thread_zone; static uma_zone_t upcall_zone; /* DEBUG ONLY */ SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0, "thread allocation"); static int thread_debug = 0; SYSCTL_INT(_kern_threads, OID_AUTO, debug, CTLFLAG_RW, &thread_debug, 0, "thread debug"); static int max_threads_per_proc = 30; SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW, &max_threads_per_proc, 0, "Limit on threads per proc"); static int max_groups_per_proc = 5; SYSCTL_INT(_kern_threads, OID_AUTO, max_groups_per_proc, CTLFLAG_RW, &max_groups_per_proc, 0, "Limit on thread groups per proc"); static int max_threads_hits; SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD, &max_threads_hits, 0, ""); static int virtual_cpu; #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); TAILQ_HEAD(, kse) zombie_kses = TAILQ_HEAD_INITIALIZER(zombie_kses); TAILQ_HEAD(, ksegrp) zombie_ksegrps = TAILQ_HEAD_INITIALIZER(zombie_ksegrps); TAILQ_HEAD(, kse_upcall) zombie_upcalls = TAILQ_HEAD_INITIALIZER(zombie_upcalls); struct mtx kse_zombie_lock; MTX_SYSINIT(kse_zombie_lock, &kse_zombie_lock, "kse zombie lock", MTX_SPIN); static void kse_purge(struct proc *p, struct thread *td); static void kse_purge_group(struct thread *td); static int thread_update_usr_ticks(struct thread *td, int user); static void thread_alloc_spare(struct thread *td, struct thread *spare); static int sysctl_kse_virtual_cpu(SYSCTL_HANDLER_ARGS) { int error, new_val; int def_val; #ifdef SMP def_val = mp_ncpus; #else def_val = 1; #endif if (virtual_cpu == 0) new_val = def_val; else new_val = virtual_cpu; error = sysctl_handle_int(oidp, &new_val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new_val < 0) return (EINVAL); virtual_cpu = new_val; return (0); } /* DEBUG ONLY */ SYSCTL_PROC(_kern_threads, OID_AUTO, virtual_cpu, CTLTYPE_INT|CTLFLAG_RW, 0, sizeof(virtual_cpu), sysctl_kse_virtual_cpu, "I", "debug virtual cpus"); /* * Prepare a thread for use. */ static void thread_ctor(void *mem, int size, void *arg) { struct thread *td; td = (struct thread *)mem; td->td_state = TDS_INACTIVE; td->td_oncpu = NOCPU; } /* * Reclaim a thread after use. */ static void thread_dtor(void *mem, int size, void *arg) { struct thread *td; td = (struct thread *)mem; #ifdef INVARIANTS /* Verify that this thread is in a safe state to free. */ switch (td->td_state) { case TDS_INHIBITED: case TDS_RUNNING: case TDS_CAN_RUN: case TDS_RUNQ: /* * We must never unlink a thread that is in one of * these states, because it is currently active. */ panic("bad state for thread unlinking"); /* NOTREACHED */ case TDS_INACTIVE: break; default: panic("bad thread state"); /* NOTREACHED */ } #endif } /* * Initialize type-stable parts of a thread (when newly created). */ static void thread_init(void *mem, int size) { struct thread *td; td = (struct thread *)mem; mtx_lock(&Giant); pmap_new_thread(td, 0); mtx_unlock(&Giant); cpu_thread_setup(td); td->td_sched = (struct td_sched *)&td[1]; } /* * Tear down type-stable parts of a thread (just before being discarded). */ static void thread_fini(void *mem, int size) { struct thread *td; td = (struct thread *)mem; pmap_dispose_thread(td); } /* * Initialize type-stable parts of a kse (when newly created). */ static void kse_init(void *mem, int size) { struct kse *ke; ke = (struct kse *)mem; ke->ke_sched = (struct ke_sched *)&ke[1]; } /* * Initialize type-stable parts of a ksegrp (when newly created). */ static void ksegrp_init(void *mem, int size) { struct ksegrp *kg; kg = (struct ksegrp *)mem; kg->kg_sched = (struct kg_sched *)&kg[1]; } /* * KSE is linked into kse group. */ void kse_link(struct kse *ke, struct ksegrp *kg) { struct proc *p = kg->kg_proc; TAILQ_INSERT_HEAD(&kg->kg_kseq, ke, ke_kglist); kg->kg_kses++; ke->ke_state = KES_UNQUEUED; ke->ke_proc = p; ke->ke_ksegrp = kg; ke->ke_thread = NULL; ke->ke_oncpu = NOCPU; ke->ke_flags = 0; } void kse_unlink(struct kse *ke) { struct ksegrp *kg; mtx_assert(&sched_lock, MA_OWNED); kg = ke->ke_ksegrp; TAILQ_REMOVE(&kg->kg_kseq, ke, ke_kglist); if (ke->ke_state == KES_IDLE) { TAILQ_REMOVE(&kg->kg_iq, ke, ke_kgrlist); kg->kg_idle_kses--; } if (--kg->kg_kses == 0) ksegrp_unlink(kg); /* * Aggregate stats from the KSE */ kse_stash(ke); } void ksegrp_link(struct ksegrp *kg, struct proc *p) { TAILQ_INIT(&kg->kg_threads); TAILQ_INIT(&kg->kg_runq); /* links with td_runq */ TAILQ_INIT(&kg->kg_slpq); /* links with td_runq */ TAILQ_INIT(&kg->kg_kseq); /* all kses in ksegrp */ TAILQ_INIT(&kg->kg_iq); /* all idle kses in ksegrp */ TAILQ_INIT(&kg->kg_upcalls); /* all upcall structure in ksegrp */ kg->kg_proc = p; /* * the following counters are in the -zero- section * and may not need clearing */ kg->kg_numthreads = 0; kg->kg_runnable = 0; kg->kg_kses = 0; kg->kg_runq_kses = 0; /* XXXKSE change name */ kg->kg_idle_kses = 0; kg->kg_numupcalls = 0; /* link it in now that it's consistent */ p->p_numksegrps++; TAILQ_INSERT_HEAD(&p->p_ksegrps, kg, kg_ksegrp); } void ksegrp_unlink(struct ksegrp *kg) { struct proc *p; mtx_assert(&sched_lock, MA_OWNED); KASSERT((kg->kg_numthreads == 0), ("ksegrp_unlink: residual threads")); KASSERT((kg->kg_kses == 0), ("ksegrp_unlink: residual kses")); KASSERT((kg->kg_numupcalls == 0), ("ksegrp_unlink: residual upcalls")); p = kg->kg_proc; TAILQ_REMOVE(&p->p_ksegrps, kg, kg_ksegrp); p->p_numksegrps--; /* * Aggregate stats from the KSE */ ksegrp_stash(kg); } struct kse_upcall * upcall_alloc(void) { struct kse_upcall *ku; ku = uma_zalloc(upcall_zone, M_WAITOK); bzero(ku, sizeof(*ku)); return (ku); } void upcall_free(struct kse_upcall *ku) { uma_zfree(upcall_zone, ku); } void upcall_link(struct kse_upcall *ku, struct ksegrp *kg) { mtx_assert(&sched_lock, MA_OWNED); TAILQ_INSERT_TAIL(&kg->kg_upcalls, ku, ku_link); ku->ku_ksegrp = kg; kg->kg_numupcalls++; } void upcall_unlink(struct kse_upcall *ku) { struct ksegrp *kg = ku->ku_ksegrp; mtx_assert(&sched_lock, MA_OWNED); KASSERT(ku->ku_owner == NULL, ("%s: have owner", __func__)); TAILQ_REMOVE(&kg->kg_upcalls, ku, ku_link); kg->kg_numupcalls--; upcall_stash(ku); } void upcall_remove(struct thread *td) { if (td->td_upcall) { td->td_upcall->ku_owner = NULL; upcall_unlink(td->td_upcall); td->td_upcall = 0; } } /* * For a newly created process, * link up all the structures and its initial threads etc. */ void proc_linkup(struct proc *p, struct ksegrp *kg, struct kse *ke, struct thread *td) { TAILQ_INIT(&p->p_ksegrps); /* all ksegrps in proc */ TAILQ_INIT(&p->p_threads); /* all threads in proc */ TAILQ_INIT(&p->p_suspended); /* Threads suspended */ p->p_numksegrps = 0; p->p_numthreads = 0; ksegrp_link(kg, p); kse_link(ke, kg); thread_link(td, kg); } /* struct kse_thr_interrupt_args { struct kse_thr_mailbox * tmbx; }; */ int kse_thr_interrupt(struct thread *td, struct kse_thr_interrupt_args *uap) { struct proc *p; struct thread *td2; p = td->td_proc; if (!(p->p_flag & P_THREADED) || (uap->tmbx == NULL)) return (EINVAL); mtx_lock_spin(&sched_lock); FOREACH_THREAD_IN_PROC(p, td2) { if (td2->td_mailbox == uap->tmbx) { td2->td_flags |= TDF_INTERRUPT; if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR)) { if (td2->td_flags & TDF_CVWAITQ) cv_abort(td2); else abortsleep(td2); } mtx_unlock_spin(&sched_lock); return (0); } } mtx_unlock_spin(&sched_lock); return (ESRCH); } /* struct kse_exit_args { register_t dummy; }; */ int kse_exit(struct thread *td, struct kse_exit_args *uap) { struct proc *p; struct ksegrp *kg; struct kse *ke; p = td->td_proc; /* * Only UTS can call the syscall and current group * should be a threaded group. */ if ((td->td_mailbox != NULL) || (td->td_ksegrp->kg_numupcalls == 0)) return (EINVAL); KASSERT((td->td_upcall != NULL), ("%s: not own an upcall", __func__)); kg = td->td_ksegrp; /* Serialize removing upcall */ PROC_LOCK(p); mtx_lock_spin(&sched_lock); if ((kg->kg_numupcalls == 1) && (kg->kg_numthreads > 1)) { mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (EDEADLK); } ke = td->td_kse; upcall_remove(td); if (p->p_numthreads == 1) { kse_purge(p, td); p->p_flag &= ~P_THREADED; mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); } else { if (kg->kg_numthreads == 1) { /* Shutdown a group */ kse_purge_group(td); ke->ke_flags |= KEF_EXIT; } thread_stopped(p); thread_exit(); /* NOTREACHED */ } return (0); } /* * Either becomes an upcall or waits for an awakening event and * then becomes an upcall. Only error cases return. */ /* struct kse_release_args { struct timespec *timeout; }; */ int kse_release(struct thread *td, struct kse_release_args *uap) { struct proc *p; struct ksegrp *kg; struct timespec ts, ts2, ts3, timeout; struct timeval tv; int error; p = td->td_proc; kg = td->td_ksegrp; /* * Only UTS can call the syscall and current group * should be a threaded group. */ if ((td->td_mailbox != NULL) || (td->td_ksegrp->kg_numupcalls == 0)) return (EINVAL); KASSERT((td->td_upcall != NULL), ("%s: not own an upcall", __func__)); if (uap->timeout != NULL) { if ((error = copyin(uap->timeout, &timeout, sizeof(timeout)))) return (error); getnanouptime(&ts); timespecadd(&ts, &timeout); TIMESPEC_TO_TIMEVAL(&tv, &timeout); } mtx_lock_spin(&sched_lock); /* Change OURSELF to become an upcall. */ td->td_flags = TDF_UPCALLING; #if 0 /* XXX This shouldn't be necessary */ if (p->p_sflag & PS_NEEDSIGCHK) td->td_flags |= TDF_ASTPENDING; #endif mtx_unlock_spin(&sched_lock); PROC_LOCK(p); while ((td->td_upcall->ku_flags & KUF_DOUPCALL) == 0 && (kg->kg_completed == NULL)) { kg->kg_upsleeps++; error = msleep(&kg->kg_completed, &p->p_mtx, PPAUSE|PCATCH, "kse_rel", (uap->timeout ? tvtohz(&tv) : 0)); kg->kg_upsleeps--; PROC_UNLOCK(p); if (uap->timeout == NULL || error != EWOULDBLOCK) return (0); getnanouptime(&ts2); if (timespeccmp(&ts2, &ts, >=)) return (0); ts3 = ts; timespecsub(&ts3, &ts2); TIMESPEC_TO_TIMEVAL(&tv, &ts3); PROC_LOCK(p); } PROC_UNLOCK(p); return (0); } /* struct kse_wakeup_args { struct kse_mailbox *mbx; }; */ int kse_wakeup(struct thread *td, struct kse_wakeup_args *uap) { struct proc *p; struct ksegrp *kg; struct kse_upcall *ku; struct thread *td2; p = td->td_proc; td2 = NULL; ku = NULL; /* KSE-enabled processes only, please. */ if (!(p->p_flag & P_THREADED)) return (EINVAL); PROC_LOCK(p); mtx_lock_spin(&sched_lock); if (uap->mbx) { FOREACH_KSEGRP_IN_PROC(p, kg) { FOREACH_UPCALL_IN_GROUP(kg, ku) { if (ku->ku_mailbox == uap->mbx) break; } if (ku) break; } } else { kg = td->td_ksegrp; if (kg->kg_upsleeps) { wakeup_one(&kg->kg_completed); mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (0); } ku = TAILQ_FIRST(&kg->kg_upcalls); } if (ku) { if ((td2 = ku->ku_owner) == NULL) { panic("%s: no owner", __func__); } else if (TD_ON_SLEEPQ(td2) && (td2->td_wchan == &kg->kg_completed)) { abortsleep(td2); } else { ku->ku_flags |= KUF_DOUPCALL; } mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (0); } mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (ESRCH); } /* * No new KSEG: first call: use current KSE, don't schedule an upcall * All other situations, do allocate max new KSEs and schedule an upcall. */ /* struct kse_create_args { struct kse_mailbox *mbx; int newgroup; }; */ int kse_create(struct thread *td, struct kse_create_args *uap) { struct kse *newke; struct ksegrp *newkg; struct ksegrp *kg; struct proc *p; struct kse_mailbox mbx; struct kse_upcall *newku; int err, ncpus; p = td->td_proc; if ((err = copyin(uap->mbx, &mbx, sizeof(mbx)))) return (err); /* Too bad, why hasn't kernel always a cpu counter !? */ #ifdef SMP ncpus = mp_ncpus; #else ncpus = 1; #endif if (thread_debug && virtual_cpu != 0) ncpus = virtual_cpu; /* Easier to just set it than to test and set */ PROC_LOCK(p); p->p_flag |= P_THREADED; PROC_UNLOCK(p); kg = td->td_ksegrp; if (uap->newgroup) { /* Have race condition but it is cheap */ if (p->p_numksegrps >= max_groups_per_proc) return (EPROCLIM); /* * If we want a new KSEGRP it doesn't matter whether * we have already fired up KSE mode before or not. * We put the process in KSE mode and create a new KSEGRP. */ newkg = ksegrp_alloc(); bzero(&newkg->kg_startzero, RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); bcopy(&kg->kg_startcopy, &newkg->kg_startcopy, RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); mtx_lock_spin(&sched_lock); if (p->p_numksegrps >= max_groups_per_proc) { mtx_unlock_spin(&sched_lock); ksegrp_free(newkg); return (EPROCLIM); } ksegrp_link(newkg, p); mtx_unlock_spin(&sched_lock); } else { newkg = kg; } /* * Creating upcalls more than number of physical cpu does * not help performance. */ if (newkg->kg_numupcalls >= ncpus) return (EPROCLIM); if (newkg->kg_numupcalls == 0) { /* * Initialize KSE group, optimized for MP. * Create KSEs as many as physical cpus, this increases * concurrent even if userland is not MP safe and can only run * on single CPU (for early version of libpthread, it is true). * In ideal world, every physical cpu should execute a thread. * If there is enough KSEs, threads in kernel can be * executed parallel on different cpus with full speed, * Concurrent in kernel shouldn't be restricted by number of * upcalls userland provides. * Adding more upcall structures only increases concurrent * in userland. * Highest performance configuration is: * N kses = N upcalls = N phyiscal cpus */ while (newkg->kg_kses < ncpus) { newke = kse_alloc(); bzero(&newke->ke_startzero, RANGEOF(struct kse, ke_startzero, ke_endzero)); #if 0 mtx_lock_spin(&sched_lock); bcopy(&ke->ke_startcopy, &newke->ke_startcopy, RANGEOF(struct kse, ke_startcopy, ke_endcopy)); mtx_unlock_spin(&sched_lock); #endif mtx_lock_spin(&sched_lock); kse_link(newke, newkg); /* Add engine */ kse_reassign(newke); mtx_unlock_spin(&sched_lock); } } newku = upcall_alloc(); newku->ku_mailbox = uap->mbx; newku->ku_func = mbx.km_func; bcopy(&mbx.km_stack, &newku->ku_stack, sizeof(stack_t)); /* For the first call this may not have been set */ if (td->td_standin == NULL) thread_alloc_spare(td, NULL); mtx_lock_spin(&sched_lock); if (newkg->kg_numupcalls >= ncpus) { mtx_unlock_spin(&sched_lock); upcall_free(newku); return (EPROCLIM); } upcall_link(newku, newkg); if (mbx.km_quantum) newkg->kg_upquantum = max(1, mbx.km_quantum/tick); /* * Each upcall structure has an owner thread, find which * one owns it. */ if (uap->newgroup) { /* * Because new ksegrp hasn't thread, * create an initial upcall thread to own it. */ thread_schedule_upcall(td, newku); } else { /* * If current thread hasn't an upcall structure, * just assign the upcall to it. */ if (td->td_upcall == NULL) { newku->ku_owner = td; td->td_upcall = newku; } else { /* * Create a new upcall thread to own it. */ thread_schedule_upcall(td, newku); } } mtx_unlock_spin(&sched_lock); return (0); } /* * Fill a ucontext_t with a thread's context information. * * This is an analogue to getcontext(3). */ void thread_getcontext(struct thread *td, ucontext_t *uc) { /* * XXX this is declared in a MD include file, i386/include/ucontext.h but * is used in MI code. */ #ifdef __i386__ get_mcontext(td, &uc->uc_mcontext); #endif PROC_LOCK(td->td_proc); uc->uc_sigmask = td->td_sigmask; PROC_UNLOCK(td->td_proc); } /* * Set a thread's context from a ucontext_t. * * This is an analogue to setcontext(3). */ int thread_setcontext(struct thread *td, ucontext_t *uc) { int ret; /* * XXX this is declared in a MD include file, i386/include/ucontext.h but * is used in MI code. */ #ifdef __i386__ ret = set_mcontext(td, &uc->uc_mcontext); #else ret = ENOSYS; #endif if (ret == 0) { SIG_CANTMASK(uc->uc_sigmask); PROC_LOCK(td->td_proc); td->td_sigmask = uc->uc_sigmask; PROC_UNLOCK(td->td_proc); } return (ret); } /* * Initialize global thread allocation resources. */ void threadinit(void) { #ifndef __ia64__ thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), thread_ctor, thread_dtor, thread_init, thread_fini, UMA_ALIGN_CACHE, 0); #else /* * XXX the ia64 kstack allocator is really lame and is at the mercy * of contigmallloc(). This hackery is to pre-construct a whole * pile of thread structures with associated kernel stacks early * in the system startup while contigmalloc() still works. Once we * have them, keep them. Sigh. */ thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), thread_ctor, thread_dtor, thread_init, thread_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE); uma_prealloc(thread_zone, 512); /* XXX arbitary */ #endif ksegrp_zone = uma_zcreate("KSEGRP", sched_sizeof_ksegrp(), NULL, NULL, ksegrp_init, NULL, UMA_ALIGN_CACHE, 0); kse_zone = uma_zcreate("KSE", sched_sizeof_kse(), NULL, NULL, kse_init, NULL, UMA_ALIGN_CACHE, 0); upcall_zone = uma_zcreate("UPCALL", sizeof(struct kse_upcall), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); } /* * Stash an embarasingly extra thread into the zombie thread queue. */ void thread_stash(struct thread *td) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_threads, td, td_runq); mtx_unlock_spin(&kse_zombie_lock); } /* * Stash an embarasingly extra kse into the zombie kse queue. */ void kse_stash(struct kse *ke) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_kses, ke, ke_procq); mtx_unlock_spin(&kse_zombie_lock); } /* * Stash an embarasingly extra upcall into the zombie upcall queue. */ void upcall_stash(struct kse_upcall *ku) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_upcalls, ku, ku_link); mtx_unlock_spin(&kse_zombie_lock); } /* * Stash an embarasingly extra ksegrp into the zombie ksegrp queue. */ void ksegrp_stash(struct ksegrp *kg) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_ksegrps, kg, kg_ksegrp); mtx_unlock_spin(&kse_zombie_lock); } /* * Reap zombie kse resource. */ void thread_reap(void) { struct thread *td_first, *td_next; struct kse *ke_first, *ke_next; struct ksegrp *kg_first, * kg_next; struct kse_upcall *ku_first, *ku_next; /* * Don't even bother to lock if none at this instant, * we really don't care about the next instant.. */ if ((!TAILQ_EMPTY(&zombie_threads)) || (!TAILQ_EMPTY(&zombie_kses)) || (!TAILQ_EMPTY(&zombie_ksegrps)) || (!TAILQ_EMPTY(&zombie_upcalls))) { mtx_lock_spin(&kse_zombie_lock); td_first = TAILQ_FIRST(&zombie_threads); ke_first = TAILQ_FIRST(&zombie_kses); kg_first = TAILQ_FIRST(&zombie_ksegrps); ku_first = TAILQ_FIRST(&zombie_upcalls); if (td_first) TAILQ_INIT(&zombie_threads); if (ke_first) TAILQ_INIT(&zombie_kses); if (kg_first) TAILQ_INIT(&zombie_ksegrps); if (ku_first) TAILQ_INIT(&zombie_upcalls); mtx_unlock_spin(&kse_zombie_lock); while (td_first) { td_next = TAILQ_NEXT(td_first, td_runq); if (td_first->td_ucred) crfree(td_first->td_ucred); thread_free(td_first); td_first = td_next; } while (ke_first) { ke_next = TAILQ_NEXT(ke_first, ke_procq); kse_free(ke_first); ke_first = ke_next; } while (kg_first) { kg_next = TAILQ_NEXT(kg_first, kg_ksegrp); ksegrp_free(kg_first); kg_first = kg_next; } while (ku_first) { ku_next = TAILQ_NEXT(ku_first, ku_link); upcall_free(ku_first); ku_first = ku_next; } } } /* * Allocate a ksegrp. */ struct ksegrp * ksegrp_alloc(void) { return (uma_zalloc(ksegrp_zone, M_WAITOK)); } /* * Allocate a kse. */ struct kse * kse_alloc(void) { return (uma_zalloc(kse_zone, M_WAITOK)); } /* * Allocate a thread. */ struct thread * thread_alloc(void) { thread_reap(); /* check if any zombies to get */ return (uma_zalloc(thread_zone, M_WAITOK)); } /* * Deallocate a ksegrp. */ void ksegrp_free(struct ksegrp *td) { uma_zfree(ksegrp_zone, td); } /* * Deallocate a kse. */ void kse_free(struct kse *td) { uma_zfree(kse_zone, td); } /* * Deallocate a thread. */ void thread_free(struct thread *td) { cpu_thread_clean(td); uma_zfree(thread_zone, td); } /* * Store the thread context in the UTS's mailbox. * then add the mailbox at the head of a list we are building in user space. * The list is anchored in the ksegrp structure. */ int thread_export_context(struct thread *td) { struct proc *p; struct ksegrp *kg; uintptr_t mbx; void *addr; int error,temp; ucontext_t uc; p = td->td_proc; kg = td->td_ksegrp; /* Export the user/machine context. */ addr = (void *)(&td->td_mailbox->tm_context); error = copyin(addr, &uc, sizeof(ucontext_t)); if (error) goto bad; thread_getcontext(td, &uc); error = copyout(&uc, addr, sizeof(ucontext_t)); if (error) goto bad; /* Exports clock ticks in kernel mode */ addr = (caddr_t)(&td->td_mailbox->tm_sticks); temp = fuword(addr) + td->td_usticks; if (suword(addr, temp)) goto bad; /* Get address in latest mbox of list pointer */ addr = (void *)(&td->td_mailbox->tm_next); /* * Put the saved address of the previous first * entry into this one */ for (;;) { mbx = (uintptr_t)kg->kg_completed; if (suword(addr, mbx)) { error = EFAULT; goto bad; } PROC_LOCK(p); if (mbx == (uintptr_t)kg->kg_completed) { kg->kg_completed = td->td_mailbox; /* * The thread context may be taken away by * other upcall threads when we unlock * process lock. it's no longer valid to * use it again in any other places. */ td->td_mailbox = NULL; PROC_UNLOCK(p); break; } PROC_UNLOCK(p); } td->td_usticks = 0; return (0); bad: PROC_LOCK(p); psignal(p, SIGSEGV); PROC_UNLOCK(p); /* The mailbox is bad, don't use it */ td->td_mailbox = NULL; td->td_usticks = 0; return (error); } /* * Take the list of completed mailboxes for this KSEGRP and put them on this * upcall's mailbox as it's the next one going up. */ static int thread_link_mboxes(struct ksegrp *kg, struct kse_upcall *ku) { struct proc *p = kg->kg_proc; void *addr; uintptr_t mbx; addr = (void *)(&ku->ku_mailbox->km_completed); for (;;) { mbx = (uintptr_t)kg->kg_completed; if (suword(addr, mbx)) { PROC_LOCK(p); psignal(p, SIGSEGV); PROC_UNLOCK(p); return (EFAULT); } PROC_LOCK(p); if (mbx == (uintptr_t)kg->kg_completed) { kg->kg_completed = NULL; PROC_UNLOCK(p); break; } PROC_UNLOCK(p); } return (0); } /* * This function should be called at statclock interrupt time */ int thread_statclock(int user) { struct thread *td = curthread; if (td->td_ksegrp->kg_numupcalls == 0) return (-1); if (user) { /* Current always do via ast() */ mtx_lock_spin(&sched_lock); td->td_flags |= (TDF_USTATCLOCK|TDF_ASTPENDING); mtx_unlock_spin(&sched_lock); td->td_uuticks++; } else { if (td->td_mailbox != NULL) td->td_usticks++; else { /* XXXKSE * We will call thread_user_enter() for every * kernel entry in future, so if the thread mailbox * is NULL, it must be a UTS kernel, don't account * clock ticks for it. */ } } return (0); } /* * Export state clock ticks for userland */ static int thread_update_usr_ticks(struct thread *td, int user) { struct proc *p = td->td_proc; struct kse_thr_mailbox *tmbx; struct kse_upcall *ku; struct ksegrp *kg; caddr_t addr; uint uticks; if ((ku = td->td_upcall) == NULL) return (-1); tmbx = (void *)fuword((void *)&ku->ku_mailbox->km_curthread); if ((tmbx == NULL) || (tmbx == (void *)-1)) return (-1); if (user) { uticks = td->td_uuticks; td->td_uuticks = 0; addr = (caddr_t)&tmbx->tm_uticks; } else { uticks = td->td_usticks; td->td_usticks = 0; addr = (caddr_t)&tmbx->tm_sticks; } if (uticks) { if (suword(addr, uticks+fuword(addr))) { PROC_LOCK(p); psignal(p, SIGSEGV); PROC_UNLOCK(p); return (-2); } } kg = td->td_ksegrp; if (kg->kg_upquantum && ticks >= kg->kg_nextupcall) { mtx_lock_spin(&sched_lock); td->td_upcall->ku_flags |= KUF_DOUPCALL; mtx_unlock_spin(&sched_lock); } return (0); } /* * Discard the current thread and exit from its context. * * Because we can't free a thread while we're operating under its context, * push the current thread into our CPU's deadthread holder. This means * we needn't worry about someone else grabbing our context before we * do a cpu_throw(). */ void thread_exit(void) { struct thread *td; struct kse *ke; struct proc *p; struct ksegrp *kg; td = curthread; kg = td->td_ksegrp; p = td->td_proc; ke = td->td_kse; mtx_assert(&sched_lock, MA_OWNED); KASSERT(p != NULL, ("thread exiting without a process")); KASSERT(ke != NULL, ("thread exiting without a kse")); KASSERT(kg != NULL, ("thread exiting without a kse group")); PROC_LOCK_ASSERT(p, MA_OWNED); CTR1(KTR_PROC, "thread_exit: thread %p", td); KASSERT(!mtx_owned(&Giant), ("dying thread owns giant")); if (td->td_standin != NULL) { thread_stash(td->td_standin); td->td_standin = NULL; } cpu_thread_exit(td); /* XXXSMP */ /* * The last thread is left attached to the process * So that the whole bundle gets recycled. Skip * all this stuff. */ if (p->p_numthreads > 1) { - /* - * Unlink this thread from its proc and the kseg. - * In keeping with the other structs we probably should - * have a thread_unlink() that does some of this but it - * would only be called from here (I think) so it would - * be a waste. (might be useful for proc_fini() as well.) - */ - TAILQ_REMOVE(&p->p_threads, td, td_plist); - p->p_numthreads--; - TAILQ_REMOVE(&kg->kg_threads, td, td_kglist); - kg->kg_numthreads--; + thread_unlink(td); if (p->p_maxthrwaits) wakeup(&p->p_numthreads); /* * The test below is NOT true if we are the * sole exiting thread. P_STOPPED_SNGL is unset * in exit1() after it is the only survivor. */ if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { if (p->p_numthreads == p->p_suspcount) { thread_unsuspend_one(p->p_singlethread); } } /* * Because each upcall structure has an owner thread, * owner thread exits only when process is in exiting * state, so upcall to userland is no longer needed, * deleting upcall structure is safe here. * So when all threads in a group is exited, all upcalls * in the group should be automatically freed. */ if (td->td_upcall) upcall_remove(td); ke->ke_state = KES_UNQUEUED; ke->ke_thread = NULL; /* * Decide what to do with the KSE attached to this thread. */ if (ke->ke_flags & KEF_EXIT) kse_unlink(ke); else kse_reassign(ke); PROC_UNLOCK(p); td->td_kse = NULL; td->td_state = TDS_INACTIVE; #if 0 td->td_proc = NULL; #endif td->td_ksegrp = NULL; td->td_last_kse = NULL; PCPU_SET(deadthread, td); } else { PROC_UNLOCK(p); } /* XXX Shouldn't cpu_throw() here. */ mtx_assert(&sched_lock, MA_OWNED); #if defined(__i386__) || defined(__sparc64__) cpu_throw(td, choosethread()); #else cpu_throw(); #endif panic("I'm a teapot!"); /* NOTREACHED */ } /* * Do any thread specific cleanups that may be needed in wait() * called with Giant held, proc and schedlock not held. */ void thread_wait(struct proc *p) { struct thread *td; KASSERT((p->p_numthreads == 1), ("Muliple threads in wait1()")); KASSERT((p->p_numksegrps == 1), ("Muliple ksegrps in wait1()")); FOREACH_THREAD_IN_PROC(p, td) { if (td->td_standin != NULL) { thread_free(td->td_standin); td->td_standin = NULL; } cpu_thread_clean(td); } thread_reap(); /* check for zombie threads etc. */ } /* * Link a thread to a process. * set up anything that needs to be initialized for it to * be used by the process. * * Note that we do not link to the proc's ucred here. * The thread is linked as if running but no KSE assigned. */ void thread_link(struct thread *td, struct ksegrp *kg) { struct proc *p; p = kg->kg_proc; td->td_state = TDS_INACTIVE; td->td_proc = p; td->td_ksegrp = kg; td->td_last_kse = NULL; td->td_flags = 0; td->td_kse = NULL; LIST_INIT(&td->td_contested); callout_init(&td->td_slpcallout, 1); TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist); TAILQ_INSERT_HEAD(&kg->kg_threads, td, td_kglist); p->p_numthreads++; kg->kg_numthreads++; } + +void +thread_unlink(struct thread *td) +{ + struct proc *p = td->td_proc; + struct ksegrp *kg = td->td_ksegrp; + + TAILQ_REMOVE(&p->p_threads, td, td_plist); + p->p_numthreads--; + TAILQ_REMOVE(&kg->kg_threads, td, td_kglist); + kg->kg_numthreads--; + /* could clear a few other things here */ +} /* * Purge a ksegrp resource. When a ksegrp is preparing to * exit, it calls this function. */ void kse_purge_group(struct thread *td) { struct ksegrp *kg; struct kse *ke; kg = td->td_ksegrp; KASSERT(kg->kg_numthreads == 1, ("%s: bad thread number", __func__)); while ((ke = TAILQ_FIRST(&kg->kg_iq)) != NULL) { KASSERT(ke->ke_state == KES_IDLE, ("%s: wrong idle KSE state", __func__)); kse_unlink(ke); } KASSERT((kg->kg_kses == 1), ("%s: ksegrp still has %d KSEs", __func__, kg->kg_kses)); KASSERT((kg->kg_numupcalls == 0), ("%s: ksegrp still has %d upcall datas", __func__, kg->kg_numupcalls)); } /* * Purge a process's KSE resource. When a process is preparing to * exit, it calls kse_purge to release any extra KSE resources in * the process. */ void kse_purge(struct proc *p, struct thread *td) { struct ksegrp *kg; struct kse *ke; KASSERT(p->p_numthreads == 1, ("bad thread number")); mtx_lock_spin(&sched_lock); while ((kg = TAILQ_FIRST(&p->p_ksegrps)) != NULL) { TAILQ_REMOVE(&p->p_ksegrps, kg, kg_ksegrp); p->p_numksegrps--; /* * There is no ownership for KSE, after all threads * in the group exited, it is possible that some KSEs * were left in idle queue, gc them now. */ while ((ke = TAILQ_FIRST(&kg->kg_iq)) != NULL) { KASSERT(ke->ke_state == KES_IDLE, ("%s: wrong idle KSE state", __func__)); TAILQ_REMOVE(&kg->kg_iq, ke, ke_kgrlist); kg->kg_idle_kses--; TAILQ_REMOVE(&kg->kg_kseq, ke, ke_kglist); kg->kg_kses--; kse_stash(ke); } KASSERT(((kg->kg_kses == 0) && (kg != td->td_ksegrp)) || ((kg->kg_kses == 1) && (kg == td->td_ksegrp)), ("ksegrp has wrong kg_kses: %d", kg->kg_kses)); KASSERT((kg->kg_numupcalls == 0), ("%s: ksegrp still has %d upcall datas", __func__, kg->kg_numupcalls)); if (kg != td->td_ksegrp) ksegrp_stash(kg); } TAILQ_INSERT_HEAD(&p->p_ksegrps, td->td_ksegrp, kg_ksegrp); p->p_numksegrps++; mtx_unlock_spin(&sched_lock); } /* * This function is intended to be used to initialize a spare thread * for upcall. Initialize thread's large data area outside sched_lock * for thread_schedule_upcall(). */ void thread_alloc_spare(struct thread *td, struct thread *spare) { if (td->td_standin) return; if (spare == NULL) spare = thread_alloc(); td->td_standin = spare; bzero(&spare->td_startzero, (unsigned)RANGEOF(struct thread, td_startzero, td_endzero)); spare->td_proc = td->td_proc; spare->td_ucred = crhold(td->td_ucred); } /* * Create a thread and schedule it for upcall on the KSE given. * Use our thread's standin so that we don't have to allocate one. */ struct thread * thread_schedule_upcall(struct thread *td, struct kse_upcall *ku) { struct thread *td2; mtx_assert(&sched_lock, MA_OWNED); /* * Schedule an upcall thread on specified kse_upcall, * the kse_upcall must be free. * td must have a spare thread. */ KASSERT(ku->ku_owner == NULL, ("%s: upcall has owner", __func__)); if ((td2 = td->td_standin) != NULL) { td->td_standin = NULL; } else { panic("no reserve thread when scheduling an upcall"); return (NULL); } CTR3(KTR_PROC, "thread_schedule_upcall: thread %p (pid %d, %s)", td2, td->td_proc->p_pid, td->td_proc->p_comm); bcopy(&td->td_startcopy, &td2->td_startcopy, (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); thread_link(td2, ku->ku_ksegrp); /* inherit blocked thread's context */ bcopy(td->td_frame, td2->td_frame, sizeof(struct trapframe)); cpu_set_upcall(td2, td->td_pcb); /* Let the new thread become owner of the upcall */ ku->ku_owner = td2; td2->td_upcall = ku; td2->td_flags = TDF_UPCALLING; #if 0 /* XXX This shouldn't be necessary */ if (td->td_proc->p_sflag & PS_NEEDSIGCHK) td2->td_flags |= TDF_ASTPENDING; #endif td2->td_kse = NULL; td2->td_state = TDS_CAN_RUN; td2->td_inhibitors = 0; setrunqueue(td2); return (td2); /* bogus.. should be a void function */ } void thread_signal_add(struct thread *td, int sig) { struct kse_upcall *ku; struct proc *p; sigset_t ss; int error; PROC_LOCK_ASSERT(td->td_proc, MA_OWNED); td = curthread; ku = td->td_upcall; p = td->td_proc; PROC_UNLOCK(p); error = copyin(&ku->ku_mailbox->km_sigscaught, &ss, sizeof(sigset_t)); if (error) goto error; SIGADDSET(ss, sig); error = copyout(&ss, &ku->ku_mailbox->km_sigscaught, sizeof(sigset_t)); if (error) goto error; PROC_LOCK(p); return; error: PROC_LOCK(p); sigexit(td, SIGILL); } /* * Schedule an upcall to notify a KSE process recieved signals. * */ void thread_signal_upcall(struct thread *td) { mtx_lock_spin(&sched_lock); td->td_flags |= TDF_UPCALLING; mtx_unlock_spin(&sched_lock); return; } void thread_switchout(struct thread *td) { struct kse_upcall *ku; mtx_assert(&sched_lock, MA_OWNED); /* * If the outgoing thread is in threaded group and has never * scheduled an upcall, decide whether this is a short * or long term event and thus whether or not to schedule * an upcall. * If it is a short term event, just suspend it in * a way that takes its KSE with it. * Select the events for which we want to schedule upcalls. * For now it's just sleep. * XXXKSE eventually almost any inhibition could do. */ if (TD_CAN_UNBIND(td) && (td->td_standin) && TD_ON_SLEEPQ(td)) { /* * Release ownership of upcall, and schedule an upcall * thread, this new upcall thread becomes the owner of * the upcall structure. */ ku = td->td_upcall; ku->ku_owner = NULL; td->td_upcall = NULL; td->td_flags &= ~TDF_CAN_UNBIND; thread_schedule_upcall(td, ku); } } /* * Setup done on the thread when it enters the kernel. * XXXKSE Presently only for syscalls but eventually all kernel entries. */ void thread_user_enter(struct proc *p, struct thread *td) { struct ksegrp *kg; struct kse_upcall *ku; kg = td->td_ksegrp; /* * First check that we shouldn't just abort. * But check if we are the single thread first! * XXX p_singlethread not locked, but should be safe. */ if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { PROC_LOCK(p); mtx_lock_spin(&sched_lock); thread_stopped(p); thread_exit(); /* NOTREACHED */ } /* * If we are doing a syscall in a KSE environment, * note where our mailbox is. There is always the * possibility that we could do this lazily (in kse_reassign()), * but for now do it every time. */ kg = td->td_ksegrp; if (kg->kg_numupcalls) { ku = td->td_upcall; KASSERT(ku, ("%s: no upcall owned", __func__)); KASSERT((ku->ku_owner == td), ("%s: wrong owner", __func__)); td->td_mailbox = (void *)fuword((void *)&ku->ku_mailbox->km_curthread); if ((td->td_mailbox == NULL) || (td->td_mailbox == (void *)-1)) { /* Don't schedule upcall when blocked */ td->td_mailbox = NULL; mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_CAN_UNBIND; mtx_unlock_spin(&sched_lock); } else { if (td->td_standin == NULL) thread_alloc_spare(td, NULL); mtx_lock_spin(&sched_lock); td->td_flags |= TDF_CAN_UNBIND; mtx_unlock_spin(&sched_lock); } } } /* * The extra work we go through if we are a threaded process when we * return to userland. * * If we are a KSE process and returning to user mode, check for * extra work to do before we return (e.g. for more syscalls * to complete first). If we were in a critical section, we should * just return to let it finish. Same if we were in the UTS (in * which case the mailbox's context's busy indicator will be set). * The only traps we suport will have set the mailbox. * We will clear it here. */ int thread_userret(struct thread *td, struct trapframe *frame) { int error = 0, upcalls; struct kse_upcall *ku; struct ksegrp *kg, *kg2; struct proc *p; struct timespec ts; p = td->td_proc; kg = td->td_ksegrp; /* Nothing to do with non-threaded group/process */ if (td->td_ksegrp->kg_numupcalls == 0) return (0); /* * Stat clock interrupt hit in userland, it * is returning from interrupt, charge thread's * userland time for UTS. */ if (td->td_flags & TDF_USTATCLOCK) { thread_update_usr_ticks(td, 1); mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_USTATCLOCK; mtx_unlock_spin(&sched_lock); if (kg->kg_completed || (td->td_upcall->ku_flags & KUF_DOUPCALL)) thread_user_enter(p, td); } /* * Optimisation: * This thread has not started any upcall. * If there is no work to report other than ourself, * then it can return direct to userland. */ if (TD_CAN_UNBIND(td)) { mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_CAN_UNBIND; ku = td->td_upcall; if ((td->td_flags & TDF_NEEDSIGCHK) == 0 && (kg->kg_completed == NULL) && (ku->ku_flags & KUF_DOUPCALL) == 0 && (kg->kg_upquantum && ticks >= kg->kg_nextupcall)) { mtx_unlock_spin(&sched_lock); thread_update_usr_ticks(td, 0); nanotime(&ts); error = copyout(&ts, (caddr_t)&ku->ku_mailbox->km_timeofday, sizeof(ts)); td->td_mailbox = 0; if (error) goto out; return (0); } mtx_unlock_spin(&sched_lock); error = thread_export_context(td); if (error) { /* * Failing to do the KSE operation just defaults * back to synchonous operation, so just return from * the syscall. */ return (0); } /* * There is something to report, and we own an upcall * strucuture, we can go to userland. * Turn ourself into an upcall thread. */ mtx_lock_spin(&sched_lock); td->td_flags |= TDF_UPCALLING; mtx_unlock_spin(&sched_lock); } else if (td->td_mailbox) { error = thread_export_context(td); /* possibly upcall with error? */ PROC_LOCK(p); /* * There are upcall threads waiting for * work to do, wake one of them up. * XXXKSE Maybe wake all of them up. */ if (!error && kg->kg_upsleeps) wakeup_one(&kg->kg_completed); mtx_lock_spin(&sched_lock); thread_stopped(p); thread_exit(); /* NOTREACHED */ } KASSERT(TD_CAN_UNBIND(td) == 0, ("can unbind")); if (p->p_numthreads > max_threads_per_proc) { max_threads_hits++; PROC_LOCK(p); while (p->p_numthreads > max_threads_per_proc) { if (P_SHOULDSTOP(p)) break; upcalls = 0; mtx_lock_spin(&sched_lock); FOREACH_KSEGRP_IN_PROC(p, kg2) { if (kg2->kg_numupcalls == 0) upcalls++; else upcalls += kg2->kg_numupcalls; } mtx_unlock_spin(&sched_lock); if (upcalls >= max_threads_per_proc) break; p->p_maxthrwaits++; msleep(&p->p_numthreads, &p->p_mtx, PPAUSE|PCATCH, "maxthreads", NULL); p->p_maxthrwaits--; } PROC_UNLOCK(p); } if (td->td_flags & TDF_UPCALLING) { kg->kg_nextupcall = ticks+kg->kg_upquantum; ku = td->td_upcall; /* * There is no more work to do and we are going to ride * this thread up to userland as an upcall. * Do the last parts of the setup needed for the upcall. */ CTR3(KTR_PROC, "userret: upcall thread %p (pid %d, %s)", td, td->td_proc->p_pid, td->td_proc->p_comm); /* * Set user context to the UTS. * Will use Giant in cpu_thread_clean() because it uses * kmem_free(kernel_map, ...) */ cpu_set_upcall_kse(td, ku); mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_UPCALLING; if (ku->ku_flags & KUF_DOUPCALL) ku->ku_flags &= ~KUF_DOUPCALL; mtx_unlock_spin(&sched_lock); /* * Unhook the list of completed threads. * anything that completes after this gets to * come in next time. * Put the list of completed thread mailboxes on * this KSE's mailbox. */ error = thread_link_mboxes(kg, ku); if (error) goto out; /* * Set state and clear the thread mailbox pointer. * From now on we are just a bound outgoing process. * **Problem** userret is often called several times. * it would be nice if this all happenned only on the first * time through. (the scan for extra work etc.) */ error = suword((caddr_t)&ku->ku_mailbox->km_curthread, 0); if (error) goto out; /* Export current system time */ nanotime(&ts); error = copyout(&ts, (caddr_t)&ku->ku_mailbox->km_timeofday, sizeof(ts)); } out: if (error) { /* * Things are going to be so screwed we should just kill * the process. * how do we do that? */ PROC_LOCK(td->td_proc); psignal(td->td_proc, SIGSEGV); PROC_UNLOCK(td->td_proc); } else { /* * Optimisation: * Ensure that we have a spare thread available, * for when we re-enter the kernel. */ if (td->td_standin == NULL) thread_alloc_spare(td, NULL); } /* * Clear thread mailbox first, then clear system tick count. * The order is important because thread_statclock() use * mailbox pointer to see if it is an userland thread or * an UTS kernel thread. */ td->td_mailbox = NULL; td->td_usticks = 0; return (error); /* go sync */ } /* * Enforce single-threading. * * Returns 1 if the caller must abort (another thread is waiting to * exit the process or similar). Process is locked! * Returns 0 when you are successfully the only thread running. * A process has successfully single threaded in the suspend mode when * There are no threads in user mode. Threads in the kernel must be * allowed to continue until they get to the user boundary. They may even * copy out their return values and data before suspending. They may however be * accellerated in reaching the user boundary as we will wake up * any sleeping threads that are interruptable. (PCATCH). */ int thread_single(int force_exit) { struct thread *td; struct thread *td2; struct proc *p; td = curthread; p = td->td_proc; mtx_assert(&Giant, MA_OWNED); PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT((td != NULL), ("curthread is NULL")); if ((p->p_flag & P_THREADED) == 0 && p->p_numthreads == 1) return (0); /* Is someone already single threading? */ if (p->p_singlethread) return (1); if (force_exit == SINGLE_EXIT) { p->p_flag |= P_SINGLE_EXIT; } else p->p_flag &= ~P_SINGLE_EXIT; p->p_flag |= P_STOPPED_SINGLE; p->p_singlethread = td; /* XXXKSE Which lock protects the below values? */ while ((p->p_numthreads - p->p_suspcount) != 1) { mtx_lock_spin(&sched_lock); FOREACH_THREAD_IN_PROC(p, td2) { if (td2 == td) continue; td->td_flags |= TDF_ASTPENDING; if (TD_IS_INHIBITED(td2)) { if (force_exit == SINGLE_EXIT) { if (TD_IS_SUSPENDED(td2)) { thread_unsuspend_one(td2); } if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR)) { if (td2->td_flags & TDF_CVWAITQ) cv_abort(td2); else abortsleep(td2); } } else { if (TD_IS_SUSPENDED(td2)) continue; /* * maybe other inhibitted states too? * XXXKSE Is it totally safe to * suspend a non-interruptable thread? */ if (td2->td_inhibitors & (TDI_SLEEPING | TDI_SWAPPED)) thread_suspend_one(td2); } } } /* * Maybe we suspended some threads.. was it enough? */ if ((p->p_numthreads - p->p_suspcount) == 1) { mtx_unlock_spin(&sched_lock); break; } /* * Wake us up when everyone else has suspended. * In the mean time we suspend as well. */ thread_suspend_one(td); /* XXX If you recursed this is broken. */ mtx_unlock(&Giant); PROC_UNLOCK(p); p->p_stats->p_ru.ru_nvcsw++; mi_switch(); mtx_unlock_spin(&sched_lock); mtx_lock(&Giant); PROC_LOCK(p); } if (force_exit == SINGLE_EXIT) { if (td->td_upcall) { mtx_lock_spin(&sched_lock); upcall_remove(td); mtx_unlock_spin(&sched_lock); } kse_purge(p, td); } return (0); } /* * Called in from locations that can safely check to see * whether we have to suspend or at least throttle for a * single-thread event (e.g. fork). * * Such locations include userret(). * If the "return_instead" argument is non zero, the thread must be able to * accept 0 (caller may continue), or 1 (caller must abort) as a result. * * The 'return_instead' argument tells the function if it may do a * thread_exit() or suspend, or whether the caller must abort and back * out instead. * * If the thread that set the single_threading request has set the * P_SINGLE_EXIT bit in the process flags then this call will never return * if 'return_instead' is false, but will exit. * * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 *---------------+--------------------+--------------------- * 0 | returns 0 | returns 0 or 1 * | when ST ends | immediatly *---------------+--------------------+--------------------- * 1 | thread exits | returns 1 * | | immediatly * 0 = thread_exit() or suspension ok, * other = return error instead of stopping the thread. * * While a full suspension is under effect, even a single threading * thread would be suspended if it made this call (but it shouldn't). * This call should only be made from places where * thread_exit() would be safe as that may be the outcome unless * return_instead is set. */ int thread_suspend_check(int return_instead) { struct thread *td; struct proc *p; struct ksegrp *kg; td = curthread; p = td->td_proc; kg = td->td_ksegrp; PROC_LOCK_ASSERT(p, MA_OWNED); while (P_SHOULDSTOP(p)) { if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { KASSERT(p->p_singlethread != NULL, ("singlethread not set")); /* * The only suspension in action is a * single-threading. Single threader need not stop. * XXX Should be safe to access unlocked * as it can only be set to be true by us. */ if (p->p_singlethread == td) return (0); /* Exempt from stopping. */ } if (return_instead) return (1); mtx_lock_spin(&sched_lock); thread_stopped(p); /* * If the process is waiting for us to exit, * this thread should just suicide. * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. */ if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { while (mtx_owned(&Giant)) mtx_unlock(&Giant); if (p->p_flag & P_THREADED) thread_exit(); else thr_exit1(); } mtx_assert(&Giant, MA_NOTOWNED); /* * When a thread suspends, it just * moves to the processes's suspend queue * and stays there. */ thread_suspend_one(td); PROC_UNLOCK(p); if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { if (p->p_numthreads == p->p_suspcount) { thread_unsuspend_one(p->p_singlethread); } } p->p_stats->p_ru.ru_nivcsw++; mi_switch(); mtx_unlock_spin(&sched_lock); PROC_LOCK(p); } return (0); } void thread_suspend_one(struct thread *td) { struct proc *p = td->td_proc; mtx_assert(&sched_lock, MA_OWNED); KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); p->p_suspcount++; TD_SET_SUSPENDED(td); TAILQ_INSERT_TAIL(&p->p_suspended, td, td_runq); /* * Hack: If we are suspending but are on the sleep queue * then we are in msleep or the cv equivalent. We * want to look like we have two Inhibitors. * May already be set.. doesn't matter. */ if (TD_ON_SLEEPQ(td)) TD_SET_SLEEPING(td); } void thread_unsuspend_one(struct thread *td) { struct proc *p = td->td_proc; mtx_assert(&sched_lock, MA_OWNED); TAILQ_REMOVE(&p->p_suspended, td, td_runq); TD_CLR_SUSPENDED(td); p->p_suspcount--; setrunnable(td); } /* * Allow all threads blocked by single threading to continue running. */ void thread_unsuspend(struct proc *p) { struct thread *td; mtx_assert(&sched_lock, MA_OWNED); PROC_LOCK_ASSERT(p, MA_OWNED); if (!P_SHOULDSTOP(p)) { while (( td = TAILQ_FIRST(&p->p_suspended))) { thread_unsuspend_one(td); } } else if ((P_SHOULDSTOP(p) == P_STOPPED_SINGLE) && (p->p_numthreads == p->p_suspcount)) { /* * Stopping everything also did the job for the single * threading request. Now we've downgraded to single-threaded, * let it continue. */ thread_unsuspend_one(p->p_singlethread); } } void thread_single_end(void) { struct thread *td; struct proc *p; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); p->p_flag &= ~P_STOPPED_SINGLE; p->p_singlethread = NULL; /* * If there are other threads they mey now run, * unless of course there is a blanket 'stop order' * on the process. The single threader must be allowed * to continue however as this is a bad place to stop. */ if ((p->p_numthreads != 1) && (!P_SHOULDSTOP(p))) { mtx_lock_spin(&sched_lock); while (( td = TAILQ_FIRST(&p->p_suspended))) { thread_unsuspend_one(td); } mtx_unlock_spin(&sched_lock); } } Index: head/sys/kern/kern_thread.c =================================================================== --- head/sys/kern/kern_thread.c (revision 113640) +++ head/sys/kern/kern_thread.c (revision 113641) @@ -1,2082 +1,2085 @@ /* * Copyright (C) 2001 Julian Elischer . * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice(s), this list of conditions and the following disclaimer as * the first lines of this file unmodified other than the possible * addition of one or more copyright notices. * 2. Redistributions in binary form must reproduce the above copyright * notice(s), this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * KSEGRP related storage. */ static uma_zone_t ksegrp_zone; static uma_zone_t kse_zone; static uma_zone_t thread_zone; static uma_zone_t upcall_zone; /* DEBUG ONLY */ SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0, "thread allocation"); static int thread_debug = 0; SYSCTL_INT(_kern_threads, OID_AUTO, debug, CTLFLAG_RW, &thread_debug, 0, "thread debug"); static int max_threads_per_proc = 30; SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW, &max_threads_per_proc, 0, "Limit on threads per proc"); static int max_groups_per_proc = 5; SYSCTL_INT(_kern_threads, OID_AUTO, max_groups_per_proc, CTLFLAG_RW, &max_groups_per_proc, 0, "Limit on thread groups per proc"); static int max_threads_hits; SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD, &max_threads_hits, 0, ""); static int virtual_cpu; #define RANGEOF(type, start, end) (offsetof(type, end) - offsetof(type, start)) TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); TAILQ_HEAD(, kse) zombie_kses = TAILQ_HEAD_INITIALIZER(zombie_kses); TAILQ_HEAD(, ksegrp) zombie_ksegrps = TAILQ_HEAD_INITIALIZER(zombie_ksegrps); TAILQ_HEAD(, kse_upcall) zombie_upcalls = TAILQ_HEAD_INITIALIZER(zombie_upcalls); struct mtx kse_zombie_lock; MTX_SYSINIT(kse_zombie_lock, &kse_zombie_lock, "kse zombie lock", MTX_SPIN); static void kse_purge(struct proc *p, struct thread *td); static void kse_purge_group(struct thread *td); static int thread_update_usr_ticks(struct thread *td, int user); static void thread_alloc_spare(struct thread *td, struct thread *spare); static int sysctl_kse_virtual_cpu(SYSCTL_HANDLER_ARGS) { int error, new_val; int def_val; #ifdef SMP def_val = mp_ncpus; #else def_val = 1; #endif if (virtual_cpu == 0) new_val = def_val; else new_val = virtual_cpu; error = sysctl_handle_int(oidp, &new_val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new_val < 0) return (EINVAL); virtual_cpu = new_val; return (0); } /* DEBUG ONLY */ SYSCTL_PROC(_kern_threads, OID_AUTO, virtual_cpu, CTLTYPE_INT|CTLFLAG_RW, 0, sizeof(virtual_cpu), sysctl_kse_virtual_cpu, "I", "debug virtual cpus"); /* * Prepare a thread for use. */ static void thread_ctor(void *mem, int size, void *arg) { struct thread *td; td = (struct thread *)mem; td->td_state = TDS_INACTIVE; td->td_oncpu = NOCPU; } /* * Reclaim a thread after use. */ static void thread_dtor(void *mem, int size, void *arg) { struct thread *td; td = (struct thread *)mem; #ifdef INVARIANTS /* Verify that this thread is in a safe state to free. */ switch (td->td_state) { case TDS_INHIBITED: case TDS_RUNNING: case TDS_CAN_RUN: case TDS_RUNQ: /* * We must never unlink a thread that is in one of * these states, because it is currently active. */ panic("bad state for thread unlinking"); /* NOTREACHED */ case TDS_INACTIVE: break; default: panic("bad thread state"); /* NOTREACHED */ } #endif } /* * Initialize type-stable parts of a thread (when newly created). */ static void thread_init(void *mem, int size) { struct thread *td; td = (struct thread *)mem; mtx_lock(&Giant); pmap_new_thread(td, 0); mtx_unlock(&Giant); cpu_thread_setup(td); td->td_sched = (struct td_sched *)&td[1]; } /* * Tear down type-stable parts of a thread (just before being discarded). */ static void thread_fini(void *mem, int size) { struct thread *td; td = (struct thread *)mem; pmap_dispose_thread(td); } /* * Initialize type-stable parts of a kse (when newly created). */ static void kse_init(void *mem, int size) { struct kse *ke; ke = (struct kse *)mem; ke->ke_sched = (struct ke_sched *)&ke[1]; } /* * Initialize type-stable parts of a ksegrp (when newly created). */ static void ksegrp_init(void *mem, int size) { struct ksegrp *kg; kg = (struct ksegrp *)mem; kg->kg_sched = (struct kg_sched *)&kg[1]; } /* * KSE is linked into kse group. */ void kse_link(struct kse *ke, struct ksegrp *kg) { struct proc *p = kg->kg_proc; TAILQ_INSERT_HEAD(&kg->kg_kseq, ke, ke_kglist); kg->kg_kses++; ke->ke_state = KES_UNQUEUED; ke->ke_proc = p; ke->ke_ksegrp = kg; ke->ke_thread = NULL; ke->ke_oncpu = NOCPU; ke->ke_flags = 0; } void kse_unlink(struct kse *ke) { struct ksegrp *kg; mtx_assert(&sched_lock, MA_OWNED); kg = ke->ke_ksegrp; TAILQ_REMOVE(&kg->kg_kseq, ke, ke_kglist); if (ke->ke_state == KES_IDLE) { TAILQ_REMOVE(&kg->kg_iq, ke, ke_kgrlist); kg->kg_idle_kses--; } if (--kg->kg_kses == 0) ksegrp_unlink(kg); /* * Aggregate stats from the KSE */ kse_stash(ke); } void ksegrp_link(struct ksegrp *kg, struct proc *p) { TAILQ_INIT(&kg->kg_threads); TAILQ_INIT(&kg->kg_runq); /* links with td_runq */ TAILQ_INIT(&kg->kg_slpq); /* links with td_runq */ TAILQ_INIT(&kg->kg_kseq); /* all kses in ksegrp */ TAILQ_INIT(&kg->kg_iq); /* all idle kses in ksegrp */ TAILQ_INIT(&kg->kg_upcalls); /* all upcall structure in ksegrp */ kg->kg_proc = p; /* * the following counters are in the -zero- section * and may not need clearing */ kg->kg_numthreads = 0; kg->kg_runnable = 0; kg->kg_kses = 0; kg->kg_runq_kses = 0; /* XXXKSE change name */ kg->kg_idle_kses = 0; kg->kg_numupcalls = 0; /* link it in now that it's consistent */ p->p_numksegrps++; TAILQ_INSERT_HEAD(&p->p_ksegrps, kg, kg_ksegrp); } void ksegrp_unlink(struct ksegrp *kg) { struct proc *p; mtx_assert(&sched_lock, MA_OWNED); KASSERT((kg->kg_numthreads == 0), ("ksegrp_unlink: residual threads")); KASSERT((kg->kg_kses == 0), ("ksegrp_unlink: residual kses")); KASSERT((kg->kg_numupcalls == 0), ("ksegrp_unlink: residual upcalls")); p = kg->kg_proc; TAILQ_REMOVE(&p->p_ksegrps, kg, kg_ksegrp); p->p_numksegrps--; /* * Aggregate stats from the KSE */ ksegrp_stash(kg); } struct kse_upcall * upcall_alloc(void) { struct kse_upcall *ku; ku = uma_zalloc(upcall_zone, M_WAITOK); bzero(ku, sizeof(*ku)); return (ku); } void upcall_free(struct kse_upcall *ku) { uma_zfree(upcall_zone, ku); } void upcall_link(struct kse_upcall *ku, struct ksegrp *kg) { mtx_assert(&sched_lock, MA_OWNED); TAILQ_INSERT_TAIL(&kg->kg_upcalls, ku, ku_link); ku->ku_ksegrp = kg; kg->kg_numupcalls++; } void upcall_unlink(struct kse_upcall *ku) { struct ksegrp *kg = ku->ku_ksegrp; mtx_assert(&sched_lock, MA_OWNED); KASSERT(ku->ku_owner == NULL, ("%s: have owner", __func__)); TAILQ_REMOVE(&kg->kg_upcalls, ku, ku_link); kg->kg_numupcalls--; upcall_stash(ku); } void upcall_remove(struct thread *td) { if (td->td_upcall) { td->td_upcall->ku_owner = NULL; upcall_unlink(td->td_upcall); td->td_upcall = 0; } } /* * For a newly created process, * link up all the structures and its initial threads etc. */ void proc_linkup(struct proc *p, struct ksegrp *kg, struct kse *ke, struct thread *td) { TAILQ_INIT(&p->p_ksegrps); /* all ksegrps in proc */ TAILQ_INIT(&p->p_threads); /* all threads in proc */ TAILQ_INIT(&p->p_suspended); /* Threads suspended */ p->p_numksegrps = 0; p->p_numthreads = 0; ksegrp_link(kg, p); kse_link(ke, kg); thread_link(td, kg); } /* struct kse_thr_interrupt_args { struct kse_thr_mailbox * tmbx; }; */ int kse_thr_interrupt(struct thread *td, struct kse_thr_interrupt_args *uap) { struct proc *p; struct thread *td2; p = td->td_proc; if (!(p->p_flag & P_THREADED) || (uap->tmbx == NULL)) return (EINVAL); mtx_lock_spin(&sched_lock); FOREACH_THREAD_IN_PROC(p, td2) { if (td2->td_mailbox == uap->tmbx) { td2->td_flags |= TDF_INTERRUPT; if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR)) { if (td2->td_flags & TDF_CVWAITQ) cv_abort(td2); else abortsleep(td2); } mtx_unlock_spin(&sched_lock); return (0); } } mtx_unlock_spin(&sched_lock); return (ESRCH); } /* struct kse_exit_args { register_t dummy; }; */ int kse_exit(struct thread *td, struct kse_exit_args *uap) { struct proc *p; struct ksegrp *kg; struct kse *ke; p = td->td_proc; /* * Only UTS can call the syscall and current group * should be a threaded group. */ if ((td->td_mailbox != NULL) || (td->td_ksegrp->kg_numupcalls == 0)) return (EINVAL); KASSERT((td->td_upcall != NULL), ("%s: not own an upcall", __func__)); kg = td->td_ksegrp; /* Serialize removing upcall */ PROC_LOCK(p); mtx_lock_spin(&sched_lock); if ((kg->kg_numupcalls == 1) && (kg->kg_numthreads > 1)) { mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (EDEADLK); } ke = td->td_kse; upcall_remove(td); if (p->p_numthreads == 1) { kse_purge(p, td); p->p_flag &= ~P_THREADED; mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); } else { if (kg->kg_numthreads == 1) { /* Shutdown a group */ kse_purge_group(td); ke->ke_flags |= KEF_EXIT; } thread_stopped(p); thread_exit(); /* NOTREACHED */ } return (0); } /* * Either becomes an upcall or waits for an awakening event and * then becomes an upcall. Only error cases return. */ /* struct kse_release_args { struct timespec *timeout; }; */ int kse_release(struct thread *td, struct kse_release_args *uap) { struct proc *p; struct ksegrp *kg; struct timespec ts, ts2, ts3, timeout; struct timeval tv; int error; p = td->td_proc; kg = td->td_ksegrp; /* * Only UTS can call the syscall and current group * should be a threaded group. */ if ((td->td_mailbox != NULL) || (td->td_ksegrp->kg_numupcalls == 0)) return (EINVAL); KASSERT((td->td_upcall != NULL), ("%s: not own an upcall", __func__)); if (uap->timeout != NULL) { if ((error = copyin(uap->timeout, &timeout, sizeof(timeout)))) return (error); getnanouptime(&ts); timespecadd(&ts, &timeout); TIMESPEC_TO_TIMEVAL(&tv, &timeout); } mtx_lock_spin(&sched_lock); /* Change OURSELF to become an upcall. */ td->td_flags = TDF_UPCALLING; #if 0 /* XXX This shouldn't be necessary */ if (p->p_sflag & PS_NEEDSIGCHK) td->td_flags |= TDF_ASTPENDING; #endif mtx_unlock_spin(&sched_lock); PROC_LOCK(p); while ((td->td_upcall->ku_flags & KUF_DOUPCALL) == 0 && (kg->kg_completed == NULL)) { kg->kg_upsleeps++; error = msleep(&kg->kg_completed, &p->p_mtx, PPAUSE|PCATCH, "kse_rel", (uap->timeout ? tvtohz(&tv) : 0)); kg->kg_upsleeps--; PROC_UNLOCK(p); if (uap->timeout == NULL || error != EWOULDBLOCK) return (0); getnanouptime(&ts2); if (timespeccmp(&ts2, &ts, >=)) return (0); ts3 = ts; timespecsub(&ts3, &ts2); TIMESPEC_TO_TIMEVAL(&tv, &ts3); PROC_LOCK(p); } PROC_UNLOCK(p); return (0); } /* struct kse_wakeup_args { struct kse_mailbox *mbx; }; */ int kse_wakeup(struct thread *td, struct kse_wakeup_args *uap) { struct proc *p; struct ksegrp *kg; struct kse_upcall *ku; struct thread *td2; p = td->td_proc; td2 = NULL; ku = NULL; /* KSE-enabled processes only, please. */ if (!(p->p_flag & P_THREADED)) return (EINVAL); PROC_LOCK(p); mtx_lock_spin(&sched_lock); if (uap->mbx) { FOREACH_KSEGRP_IN_PROC(p, kg) { FOREACH_UPCALL_IN_GROUP(kg, ku) { if (ku->ku_mailbox == uap->mbx) break; } if (ku) break; } } else { kg = td->td_ksegrp; if (kg->kg_upsleeps) { wakeup_one(&kg->kg_completed); mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (0); } ku = TAILQ_FIRST(&kg->kg_upcalls); } if (ku) { if ((td2 = ku->ku_owner) == NULL) { panic("%s: no owner", __func__); } else if (TD_ON_SLEEPQ(td2) && (td2->td_wchan == &kg->kg_completed)) { abortsleep(td2); } else { ku->ku_flags |= KUF_DOUPCALL; } mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (0); } mtx_unlock_spin(&sched_lock); PROC_UNLOCK(p); return (ESRCH); } /* * No new KSEG: first call: use current KSE, don't schedule an upcall * All other situations, do allocate max new KSEs and schedule an upcall. */ /* struct kse_create_args { struct kse_mailbox *mbx; int newgroup; }; */ int kse_create(struct thread *td, struct kse_create_args *uap) { struct kse *newke; struct ksegrp *newkg; struct ksegrp *kg; struct proc *p; struct kse_mailbox mbx; struct kse_upcall *newku; int err, ncpus; p = td->td_proc; if ((err = copyin(uap->mbx, &mbx, sizeof(mbx)))) return (err); /* Too bad, why hasn't kernel always a cpu counter !? */ #ifdef SMP ncpus = mp_ncpus; #else ncpus = 1; #endif if (thread_debug && virtual_cpu != 0) ncpus = virtual_cpu; /* Easier to just set it than to test and set */ PROC_LOCK(p); p->p_flag |= P_THREADED; PROC_UNLOCK(p); kg = td->td_ksegrp; if (uap->newgroup) { /* Have race condition but it is cheap */ if (p->p_numksegrps >= max_groups_per_proc) return (EPROCLIM); /* * If we want a new KSEGRP it doesn't matter whether * we have already fired up KSE mode before or not. * We put the process in KSE mode and create a new KSEGRP. */ newkg = ksegrp_alloc(); bzero(&newkg->kg_startzero, RANGEOF(struct ksegrp, kg_startzero, kg_endzero)); bcopy(&kg->kg_startcopy, &newkg->kg_startcopy, RANGEOF(struct ksegrp, kg_startcopy, kg_endcopy)); mtx_lock_spin(&sched_lock); if (p->p_numksegrps >= max_groups_per_proc) { mtx_unlock_spin(&sched_lock); ksegrp_free(newkg); return (EPROCLIM); } ksegrp_link(newkg, p); mtx_unlock_spin(&sched_lock); } else { newkg = kg; } /* * Creating upcalls more than number of physical cpu does * not help performance. */ if (newkg->kg_numupcalls >= ncpus) return (EPROCLIM); if (newkg->kg_numupcalls == 0) { /* * Initialize KSE group, optimized for MP. * Create KSEs as many as physical cpus, this increases * concurrent even if userland is not MP safe and can only run * on single CPU (for early version of libpthread, it is true). * In ideal world, every physical cpu should execute a thread. * If there is enough KSEs, threads in kernel can be * executed parallel on different cpus with full speed, * Concurrent in kernel shouldn't be restricted by number of * upcalls userland provides. * Adding more upcall structures only increases concurrent * in userland. * Highest performance configuration is: * N kses = N upcalls = N phyiscal cpus */ while (newkg->kg_kses < ncpus) { newke = kse_alloc(); bzero(&newke->ke_startzero, RANGEOF(struct kse, ke_startzero, ke_endzero)); #if 0 mtx_lock_spin(&sched_lock); bcopy(&ke->ke_startcopy, &newke->ke_startcopy, RANGEOF(struct kse, ke_startcopy, ke_endcopy)); mtx_unlock_spin(&sched_lock); #endif mtx_lock_spin(&sched_lock); kse_link(newke, newkg); /* Add engine */ kse_reassign(newke); mtx_unlock_spin(&sched_lock); } } newku = upcall_alloc(); newku->ku_mailbox = uap->mbx; newku->ku_func = mbx.km_func; bcopy(&mbx.km_stack, &newku->ku_stack, sizeof(stack_t)); /* For the first call this may not have been set */ if (td->td_standin == NULL) thread_alloc_spare(td, NULL); mtx_lock_spin(&sched_lock); if (newkg->kg_numupcalls >= ncpus) { mtx_unlock_spin(&sched_lock); upcall_free(newku); return (EPROCLIM); } upcall_link(newku, newkg); if (mbx.km_quantum) newkg->kg_upquantum = max(1, mbx.km_quantum/tick); /* * Each upcall structure has an owner thread, find which * one owns it. */ if (uap->newgroup) { /* * Because new ksegrp hasn't thread, * create an initial upcall thread to own it. */ thread_schedule_upcall(td, newku); } else { /* * If current thread hasn't an upcall structure, * just assign the upcall to it. */ if (td->td_upcall == NULL) { newku->ku_owner = td; td->td_upcall = newku; } else { /* * Create a new upcall thread to own it. */ thread_schedule_upcall(td, newku); } } mtx_unlock_spin(&sched_lock); return (0); } /* * Fill a ucontext_t with a thread's context information. * * This is an analogue to getcontext(3). */ void thread_getcontext(struct thread *td, ucontext_t *uc) { /* * XXX this is declared in a MD include file, i386/include/ucontext.h but * is used in MI code. */ #ifdef __i386__ get_mcontext(td, &uc->uc_mcontext); #endif PROC_LOCK(td->td_proc); uc->uc_sigmask = td->td_sigmask; PROC_UNLOCK(td->td_proc); } /* * Set a thread's context from a ucontext_t. * * This is an analogue to setcontext(3). */ int thread_setcontext(struct thread *td, ucontext_t *uc) { int ret; /* * XXX this is declared in a MD include file, i386/include/ucontext.h but * is used in MI code. */ #ifdef __i386__ ret = set_mcontext(td, &uc->uc_mcontext); #else ret = ENOSYS; #endif if (ret == 0) { SIG_CANTMASK(uc->uc_sigmask); PROC_LOCK(td->td_proc); td->td_sigmask = uc->uc_sigmask; PROC_UNLOCK(td->td_proc); } return (ret); } /* * Initialize global thread allocation resources. */ void threadinit(void) { #ifndef __ia64__ thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), thread_ctor, thread_dtor, thread_init, thread_fini, UMA_ALIGN_CACHE, 0); #else /* * XXX the ia64 kstack allocator is really lame and is at the mercy * of contigmallloc(). This hackery is to pre-construct a whole * pile of thread structures with associated kernel stacks early * in the system startup while contigmalloc() still works. Once we * have them, keep them. Sigh. */ thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), thread_ctor, thread_dtor, thread_init, thread_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE); uma_prealloc(thread_zone, 512); /* XXX arbitary */ #endif ksegrp_zone = uma_zcreate("KSEGRP", sched_sizeof_ksegrp(), NULL, NULL, ksegrp_init, NULL, UMA_ALIGN_CACHE, 0); kse_zone = uma_zcreate("KSE", sched_sizeof_kse(), NULL, NULL, kse_init, NULL, UMA_ALIGN_CACHE, 0); upcall_zone = uma_zcreate("UPCALL", sizeof(struct kse_upcall), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); } /* * Stash an embarasingly extra thread into the zombie thread queue. */ void thread_stash(struct thread *td) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_threads, td, td_runq); mtx_unlock_spin(&kse_zombie_lock); } /* * Stash an embarasingly extra kse into the zombie kse queue. */ void kse_stash(struct kse *ke) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_kses, ke, ke_procq); mtx_unlock_spin(&kse_zombie_lock); } /* * Stash an embarasingly extra upcall into the zombie upcall queue. */ void upcall_stash(struct kse_upcall *ku) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_upcalls, ku, ku_link); mtx_unlock_spin(&kse_zombie_lock); } /* * Stash an embarasingly extra ksegrp into the zombie ksegrp queue. */ void ksegrp_stash(struct ksegrp *kg) { mtx_lock_spin(&kse_zombie_lock); TAILQ_INSERT_HEAD(&zombie_ksegrps, kg, kg_ksegrp); mtx_unlock_spin(&kse_zombie_lock); } /* * Reap zombie kse resource. */ void thread_reap(void) { struct thread *td_first, *td_next; struct kse *ke_first, *ke_next; struct ksegrp *kg_first, * kg_next; struct kse_upcall *ku_first, *ku_next; /* * Don't even bother to lock if none at this instant, * we really don't care about the next instant.. */ if ((!TAILQ_EMPTY(&zombie_threads)) || (!TAILQ_EMPTY(&zombie_kses)) || (!TAILQ_EMPTY(&zombie_ksegrps)) || (!TAILQ_EMPTY(&zombie_upcalls))) { mtx_lock_spin(&kse_zombie_lock); td_first = TAILQ_FIRST(&zombie_threads); ke_first = TAILQ_FIRST(&zombie_kses); kg_first = TAILQ_FIRST(&zombie_ksegrps); ku_first = TAILQ_FIRST(&zombie_upcalls); if (td_first) TAILQ_INIT(&zombie_threads); if (ke_first) TAILQ_INIT(&zombie_kses); if (kg_first) TAILQ_INIT(&zombie_ksegrps); if (ku_first) TAILQ_INIT(&zombie_upcalls); mtx_unlock_spin(&kse_zombie_lock); while (td_first) { td_next = TAILQ_NEXT(td_first, td_runq); if (td_first->td_ucred) crfree(td_first->td_ucred); thread_free(td_first); td_first = td_next; } while (ke_first) { ke_next = TAILQ_NEXT(ke_first, ke_procq); kse_free(ke_first); ke_first = ke_next; } while (kg_first) { kg_next = TAILQ_NEXT(kg_first, kg_ksegrp); ksegrp_free(kg_first); kg_first = kg_next; } while (ku_first) { ku_next = TAILQ_NEXT(ku_first, ku_link); upcall_free(ku_first); ku_first = ku_next; } } } /* * Allocate a ksegrp. */ struct ksegrp * ksegrp_alloc(void) { return (uma_zalloc(ksegrp_zone, M_WAITOK)); } /* * Allocate a kse. */ struct kse * kse_alloc(void) { return (uma_zalloc(kse_zone, M_WAITOK)); } /* * Allocate a thread. */ struct thread * thread_alloc(void) { thread_reap(); /* check if any zombies to get */ return (uma_zalloc(thread_zone, M_WAITOK)); } /* * Deallocate a ksegrp. */ void ksegrp_free(struct ksegrp *td) { uma_zfree(ksegrp_zone, td); } /* * Deallocate a kse. */ void kse_free(struct kse *td) { uma_zfree(kse_zone, td); } /* * Deallocate a thread. */ void thread_free(struct thread *td) { cpu_thread_clean(td); uma_zfree(thread_zone, td); } /* * Store the thread context in the UTS's mailbox. * then add the mailbox at the head of a list we are building in user space. * The list is anchored in the ksegrp structure. */ int thread_export_context(struct thread *td) { struct proc *p; struct ksegrp *kg; uintptr_t mbx; void *addr; int error,temp; ucontext_t uc; p = td->td_proc; kg = td->td_ksegrp; /* Export the user/machine context. */ addr = (void *)(&td->td_mailbox->tm_context); error = copyin(addr, &uc, sizeof(ucontext_t)); if (error) goto bad; thread_getcontext(td, &uc); error = copyout(&uc, addr, sizeof(ucontext_t)); if (error) goto bad; /* Exports clock ticks in kernel mode */ addr = (caddr_t)(&td->td_mailbox->tm_sticks); temp = fuword(addr) + td->td_usticks; if (suword(addr, temp)) goto bad; /* Get address in latest mbox of list pointer */ addr = (void *)(&td->td_mailbox->tm_next); /* * Put the saved address of the previous first * entry into this one */ for (;;) { mbx = (uintptr_t)kg->kg_completed; if (suword(addr, mbx)) { error = EFAULT; goto bad; } PROC_LOCK(p); if (mbx == (uintptr_t)kg->kg_completed) { kg->kg_completed = td->td_mailbox; /* * The thread context may be taken away by * other upcall threads when we unlock * process lock. it's no longer valid to * use it again in any other places. */ td->td_mailbox = NULL; PROC_UNLOCK(p); break; } PROC_UNLOCK(p); } td->td_usticks = 0; return (0); bad: PROC_LOCK(p); psignal(p, SIGSEGV); PROC_UNLOCK(p); /* The mailbox is bad, don't use it */ td->td_mailbox = NULL; td->td_usticks = 0; return (error); } /* * Take the list of completed mailboxes for this KSEGRP and put them on this * upcall's mailbox as it's the next one going up. */ static int thread_link_mboxes(struct ksegrp *kg, struct kse_upcall *ku) { struct proc *p = kg->kg_proc; void *addr; uintptr_t mbx; addr = (void *)(&ku->ku_mailbox->km_completed); for (;;) { mbx = (uintptr_t)kg->kg_completed; if (suword(addr, mbx)) { PROC_LOCK(p); psignal(p, SIGSEGV); PROC_UNLOCK(p); return (EFAULT); } PROC_LOCK(p); if (mbx == (uintptr_t)kg->kg_completed) { kg->kg_completed = NULL; PROC_UNLOCK(p); break; } PROC_UNLOCK(p); } return (0); } /* * This function should be called at statclock interrupt time */ int thread_statclock(int user) { struct thread *td = curthread; if (td->td_ksegrp->kg_numupcalls == 0) return (-1); if (user) { /* Current always do via ast() */ mtx_lock_spin(&sched_lock); td->td_flags |= (TDF_USTATCLOCK|TDF_ASTPENDING); mtx_unlock_spin(&sched_lock); td->td_uuticks++; } else { if (td->td_mailbox != NULL) td->td_usticks++; else { /* XXXKSE * We will call thread_user_enter() for every * kernel entry in future, so if the thread mailbox * is NULL, it must be a UTS kernel, don't account * clock ticks for it. */ } } return (0); } /* * Export state clock ticks for userland */ static int thread_update_usr_ticks(struct thread *td, int user) { struct proc *p = td->td_proc; struct kse_thr_mailbox *tmbx; struct kse_upcall *ku; struct ksegrp *kg; caddr_t addr; uint uticks; if ((ku = td->td_upcall) == NULL) return (-1); tmbx = (void *)fuword((void *)&ku->ku_mailbox->km_curthread); if ((tmbx == NULL) || (tmbx == (void *)-1)) return (-1); if (user) { uticks = td->td_uuticks; td->td_uuticks = 0; addr = (caddr_t)&tmbx->tm_uticks; } else { uticks = td->td_usticks; td->td_usticks = 0; addr = (caddr_t)&tmbx->tm_sticks; } if (uticks) { if (suword(addr, uticks+fuword(addr))) { PROC_LOCK(p); psignal(p, SIGSEGV); PROC_UNLOCK(p); return (-2); } } kg = td->td_ksegrp; if (kg->kg_upquantum && ticks >= kg->kg_nextupcall) { mtx_lock_spin(&sched_lock); td->td_upcall->ku_flags |= KUF_DOUPCALL; mtx_unlock_spin(&sched_lock); } return (0); } /* * Discard the current thread and exit from its context. * * Because we can't free a thread while we're operating under its context, * push the current thread into our CPU's deadthread holder. This means * we needn't worry about someone else grabbing our context before we * do a cpu_throw(). */ void thread_exit(void) { struct thread *td; struct kse *ke; struct proc *p; struct ksegrp *kg; td = curthread; kg = td->td_ksegrp; p = td->td_proc; ke = td->td_kse; mtx_assert(&sched_lock, MA_OWNED); KASSERT(p != NULL, ("thread exiting without a process")); KASSERT(ke != NULL, ("thread exiting without a kse")); KASSERT(kg != NULL, ("thread exiting without a kse group")); PROC_LOCK_ASSERT(p, MA_OWNED); CTR1(KTR_PROC, "thread_exit: thread %p", td); KASSERT(!mtx_owned(&Giant), ("dying thread owns giant")); if (td->td_standin != NULL) { thread_stash(td->td_standin); td->td_standin = NULL; } cpu_thread_exit(td); /* XXXSMP */ /* * The last thread is left attached to the process * So that the whole bundle gets recycled. Skip * all this stuff. */ if (p->p_numthreads > 1) { - /* - * Unlink this thread from its proc and the kseg. - * In keeping with the other structs we probably should - * have a thread_unlink() that does some of this but it - * would only be called from here (I think) so it would - * be a waste. (might be useful for proc_fini() as well.) - */ - TAILQ_REMOVE(&p->p_threads, td, td_plist); - p->p_numthreads--; - TAILQ_REMOVE(&kg->kg_threads, td, td_kglist); - kg->kg_numthreads--; + thread_unlink(td); if (p->p_maxthrwaits) wakeup(&p->p_numthreads); /* * The test below is NOT true if we are the * sole exiting thread. P_STOPPED_SNGL is unset * in exit1() after it is the only survivor. */ if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { if (p->p_numthreads == p->p_suspcount) { thread_unsuspend_one(p->p_singlethread); } } /* * Because each upcall structure has an owner thread, * owner thread exits only when process is in exiting * state, so upcall to userland is no longer needed, * deleting upcall structure is safe here. * So when all threads in a group is exited, all upcalls * in the group should be automatically freed. */ if (td->td_upcall) upcall_remove(td); ke->ke_state = KES_UNQUEUED; ke->ke_thread = NULL; /* * Decide what to do with the KSE attached to this thread. */ if (ke->ke_flags & KEF_EXIT) kse_unlink(ke); else kse_reassign(ke); PROC_UNLOCK(p); td->td_kse = NULL; td->td_state = TDS_INACTIVE; #if 0 td->td_proc = NULL; #endif td->td_ksegrp = NULL; td->td_last_kse = NULL; PCPU_SET(deadthread, td); } else { PROC_UNLOCK(p); } /* XXX Shouldn't cpu_throw() here. */ mtx_assert(&sched_lock, MA_OWNED); #if defined(__i386__) || defined(__sparc64__) cpu_throw(td, choosethread()); #else cpu_throw(); #endif panic("I'm a teapot!"); /* NOTREACHED */ } /* * Do any thread specific cleanups that may be needed in wait() * called with Giant held, proc and schedlock not held. */ void thread_wait(struct proc *p) { struct thread *td; KASSERT((p->p_numthreads == 1), ("Muliple threads in wait1()")); KASSERT((p->p_numksegrps == 1), ("Muliple ksegrps in wait1()")); FOREACH_THREAD_IN_PROC(p, td) { if (td->td_standin != NULL) { thread_free(td->td_standin); td->td_standin = NULL; } cpu_thread_clean(td); } thread_reap(); /* check for zombie threads etc. */ } /* * Link a thread to a process. * set up anything that needs to be initialized for it to * be used by the process. * * Note that we do not link to the proc's ucred here. * The thread is linked as if running but no KSE assigned. */ void thread_link(struct thread *td, struct ksegrp *kg) { struct proc *p; p = kg->kg_proc; td->td_state = TDS_INACTIVE; td->td_proc = p; td->td_ksegrp = kg; td->td_last_kse = NULL; td->td_flags = 0; td->td_kse = NULL; LIST_INIT(&td->td_contested); callout_init(&td->td_slpcallout, 1); TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist); TAILQ_INSERT_HEAD(&kg->kg_threads, td, td_kglist); p->p_numthreads++; kg->kg_numthreads++; } + +void +thread_unlink(struct thread *td) +{ + struct proc *p = td->td_proc; + struct ksegrp *kg = td->td_ksegrp; + + TAILQ_REMOVE(&p->p_threads, td, td_plist); + p->p_numthreads--; + TAILQ_REMOVE(&kg->kg_threads, td, td_kglist); + kg->kg_numthreads--; + /* could clear a few other things here */ +} /* * Purge a ksegrp resource. When a ksegrp is preparing to * exit, it calls this function. */ void kse_purge_group(struct thread *td) { struct ksegrp *kg; struct kse *ke; kg = td->td_ksegrp; KASSERT(kg->kg_numthreads == 1, ("%s: bad thread number", __func__)); while ((ke = TAILQ_FIRST(&kg->kg_iq)) != NULL) { KASSERT(ke->ke_state == KES_IDLE, ("%s: wrong idle KSE state", __func__)); kse_unlink(ke); } KASSERT((kg->kg_kses == 1), ("%s: ksegrp still has %d KSEs", __func__, kg->kg_kses)); KASSERT((kg->kg_numupcalls == 0), ("%s: ksegrp still has %d upcall datas", __func__, kg->kg_numupcalls)); } /* * Purge a process's KSE resource. When a process is preparing to * exit, it calls kse_purge to release any extra KSE resources in * the process. */ void kse_purge(struct proc *p, struct thread *td) { struct ksegrp *kg; struct kse *ke; KASSERT(p->p_numthreads == 1, ("bad thread number")); mtx_lock_spin(&sched_lock); while ((kg = TAILQ_FIRST(&p->p_ksegrps)) != NULL) { TAILQ_REMOVE(&p->p_ksegrps, kg, kg_ksegrp); p->p_numksegrps--; /* * There is no ownership for KSE, after all threads * in the group exited, it is possible that some KSEs * were left in idle queue, gc them now. */ while ((ke = TAILQ_FIRST(&kg->kg_iq)) != NULL) { KASSERT(ke->ke_state == KES_IDLE, ("%s: wrong idle KSE state", __func__)); TAILQ_REMOVE(&kg->kg_iq, ke, ke_kgrlist); kg->kg_idle_kses--; TAILQ_REMOVE(&kg->kg_kseq, ke, ke_kglist); kg->kg_kses--; kse_stash(ke); } KASSERT(((kg->kg_kses == 0) && (kg != td->td_ksegrp)) || ((kg->kg_kses == 1) && (kg == td->td_ksegrp)), ("ksegrp has wrong kg_kses: %d", kg->kg_kses)); KASSERT((kg->kg_numupcalls == 0), ("%s: ksegrp still has %d upcall datas", __func__, kg->kg_numupcalls)); if (kg != td->td_ksegrp) ksegrp_stash(kg); } TAILQ_INSERT_HEAD(&p->p_ksegrps, td->td_ksegrp, kg_ksegrp); p->p_numksegrps++; mtx_unlock_spin(&sched_lock); } /* * This function is intended to be used to initialize a spare thread * for upcall. Initialize thread's large data area outside sched_lock * for thread_schedule_upcall(). */ void thread_alloc_spare(struct thread *td, struct thread *spare) { if (td->td_standin) return; if (spare == NULL) spare = thread_alloc(); td->td_standin = spare; bzero(&spare->td_startzero, (unsigned)RANGEOF(struct thread, td_startzero, td_endzero)); spare->td_proc = td->td_proc; spare->td_ucred = crhold(td->td_ucred); } /* * Create a thread and schedule it for upcall on the KSE given. * Use our thread's standin so that we don't have to allocate one. */ struct thread * thread_schedule_upcall(struct thread *td, struct kse_upcall *ku) { struct thread *td2; mtx_assert(&sched_lock, MA_OWNED); /* * Schedule an upcall thread on specified kse_upcall, * the kse_upcall must be free. * td must have a spare thread. */ KASSERT(ku->ku_owner == NULL, ("%s: upcall has owner", __func__)); if ((td2 = td->td_standin) != NULL) { td->td_standin = NULL; } else { panic("no reserve thread when scheduling an upcall"); return (NULL); } CTR3(KTR_PROC, "thread_schedule_upcall: thread %p (pid %d, %s)", td2, td->td_proc->p_pid, td->td_proc->p_comm); bcopy(&td->td_startcopy, &td2->td_startcopy, (unsigned) RANGEOF(struct thread, td_startcopy, td_endcopy)); thread_link(td2, ku->ku_ksegrp); /* inherit blocked thread's context */ bcopy(td->td_frame, td2->td_frame, sizeof(struct trapframe)); cpu_set_upcall(td2, td->td_pcb); /* Let the new thread become owner of the upcall */ ku->ku_owner = td2; td2->td_upcall = ku; td2->td_flags = TDF_UPCALLING; #if 0 /* XXX This shouldn't be necessary */ if (td->td_proc->p_sflag & PS_NEEDSIGCHK) td2->td_flags |= TDF_ASTPENDING; #endif td2->td_kse = NULL; td2->td_state = TDS_CAN_RUN; td2->td_inhibitors = 0; setrunqueue(td2); return (td2); /* bogus.. should be a void function */ } void thread_signal_add(struct thread *td, int sig) { struct kse_upcall *ku; struct proc *p; sigset_t ss; int error; PROC_LOCK_ASSERT(td->td_proc, MA_OWNED); td = curthread; ku = td->td_upcall; p = td->td_proc; PROC_UNLOCK(p); error = copyin(&ku->ku_mailbox->km_sigscaught, &ss, sizeof(sigset_t)); if (error) goto error; SIGADDSET(ss, sig); error = copyout(&ss, &ku->ku_mailbox->km_sigscaught, sizeof(sigset_t)); if (error) goto error; PROC_LOCK(p); return; error: PROC_LOCK(p); sigexit(td, SIGILL); } /* * Schedule an upcall to notify a KSE process recieved signals. * */ void thread_signal_upcall(struct thread *td) { mtx_lock_spin(&sched_lock); td->td_flags |= TDF_UPCALLING; mtx_unlock_spin(&sched_lock); return; } void thread_switchout(struct thread *td) { struct kse_upcall *ku; mtx_assert(&sched_lock, MA_OWNED); /* * If the outgoing thread is in threaded group and has never * scheduled an upcall, decide whether this is a short * or long term event and thus whether or not to schedule * an upcall. * If it is a short term event, just suspend it in * a way that takes its KSE with it. * Select the events for which we want to schedule upcalls. * For now it's just sleep. * XXXKSE eventually almost any inhibition could do. */ if (TD_CAN_UNBIND(td) && (td->td_standin) && TD_ON_SLEEPQ(td)) { /* * Release ownership of upcall, and schedule an upcall * thread, this new upcall thread becomes the owner of * the upcall structure. */ ku = td->td_upcall; ku->ku_owner = NULL; td->td_upcall = NULL; td->td_flags &= ~TDF_CAN_UNBIND; thread_schedule_upcall(td, ku); } } /* * Setup done on the thread when it enters the kernel. * XXXKSE Presently only for syscalls but eventually all kernel entries. */ void thread_user_enter(struct proc *p, struct thread *td) { struct ksegrp *kg; struct kse_upcall *ku; kg = td->td_ksegrp; /* * First check that we shouldn't just abort. * But check if we are the single thread first! * XXX p_singlethread not locked, but should be safe. */ if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { PROC_LOCK(p); mtx_lock_spin(&sched_lock); thread_stopped(p); thread_exit(); /* NOTREACHED */ } /* * If we are doing a syscall in a KSE environment, * note where our mailbox is. There is always the * possibility that we could do this lazily (in kse_reassign()), * but for now do it every time. */ kg = td->td_ksegrp; if (kg->kg_numupcalls) { ku = td->td_upcall; KASSERT(ku, ("%s: no upcall owned", __func__)); KASSERT((ku->ku_owner == td), ("%s: wrong owner", __func__)); td->td_mailbox = (void *)fuword((void *)&ku->ku_mailbox->km_curthread); if ((td->td_mailbox == NULL) || (td->td_mailbox == (void *)-1)) { /* Don't schedule upcall when blocked */ td->td_mailbox = NULL; mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_CAN_UNBIND; mtx_unlock_spin(&sched_lock); } else { if (td->td_standin == NULL) thread_alloc_spare(td, NULL); mtx_lock_spin(&sched_lock); td->td_flags |= TDF_CAN_UNBIND; mtx_unlock_spin(&sched_lock); } } } /* * The extra work we go through if we are a threaded process when we * return to userland. * * If we are a KSE process and returning to user mode, check for * extra work to do before we return (e.g. for more syscalls * to complete first). If we were in a critical section, we should * just return to let it finish. Same if we were in the UTS (in * which case the mailbox's context's busy indicator will be set). * The only traps we suport will have set the mailbox. * We will clear it here. */ int thread_userret(struct thread *td, struct trapframe *frame) { int error = 0, upcalls; struct kse_upcall *ku; struct ksegrp *kg, *kg2; struct proc *p; struct timespec ts; p = td->td_proc; kg = td->td_ksegrp; /* Nothing to do with non-threaded group/process */ if (td->td_ksegrp->kg_numupcalls == 0) return (0); /* * Stat clock interrupt hit in userland, it * is returning from interrupt, charge thread's * userland time for UTS. */ if (td->td_flags & TDF_USTATCLOCK) { thread_update_usr_ticks(td, 1); mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_USTATCLOCK; mtx_unlock_spin(&sched_lock); if (kg->kg_completed || (td->td_upcall->ku_flags & KUF_DOUPCALL)) thread_user_enter(p, td); } /* * Optimisation: * This thread has not started any upcall. * If there is no work to report other than ourself, * then it can return direct to userland. */ if (TD_CAN_UNBIND(td)) { mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_CAN_UNBIND; ku = td->td_upcall; if ((td->td_flags & TDF_NEEDSIGCHK) == 0 && (kg->kg_completed == NULL) && (ku->ku_flags & KUF_DOUPCALL) == 0 && (kg->kg_upquantum && ticks >= kg->kg_nextupcall)) { mtx_unlock_spin(&sched_lock); thread_update_usr_ticks(td, 0); nanotime(&ts); error = copyout(&ts, (caddr_t)&ku->ku_mailbox->km_timeofday, sizeof(ts)); td->td_mailbox = 0; if (error) goto out; return (0); } mtx_unlock_spin(&sched_lock); error = thread_export_context(td); if (error) { /* * Failing to do the KSE operation just defaults * back to synchonous operation, so just return from * the syscall. */ return (0); } /* * There is something to report, and we own an upcall * strucuture, we can go to userland. * Turn ourself into an upcall thread. */ mtx_lock_spin(&sched_lock); td->td_flags |= TDF_UPCALLING; mtx_unlock_spin(&sched_lock); } else if (td->td_mailbox) { error = thread_export_context(td); /* possibly upcall with error? */ PROC_LOCK(p); /* * There are upcall threads waiting for * work to do, wake one of them up. * XXXKSE Maybe wake all of them up. */ if (!error && kg->kg_upsleeps) wakeup_one(&kg->kg_completed); mtx_lock_spin(&sched_lock); thread_stopped(p); thread_exit(); /* NOTREACHED */ } KASSERT(TD_CAN_UNBIND(td) == 0, ("can unbind")); if (p->p_numthreads > max_threads_per_proc) { max_threads_hits++; PROC_LOCK(p); while (p->p_numthreads > max_threads_per_proc) { if (P_SHOULDSTOP(p)) break; upcalls = 0; mtx_lock_spin(&sched_lock); FOREACH_KSEGRP_IN_PROC(p, kg2) { if (kg2->kg_numupcalls == 0) upcalls++; else upcalls += kg2->kg_numupcalls; } mtx_unlock_spin(&sched_lock); if (upcalls >= max_threads_per_proc) break; p->p_maxthrwaits++; msleep(&p->p_numthreads, &p->p_mtx, PPAUSE|PCATCH, "maxthreads", NULL); p->p_maxthrwaits--; } PROC_UNLOCK(p); } if (td->td_flags & TDF_UPCALLING) { kg->kg_nextupcall = ticks+kg->kg_upquantum; ku = td->td_upcall; /* * There is no more work to do and we are going to ride * this thread up to userland as an upcall. * Do the last parts of the setup needed for the upcall. */ CTR3(KTR_PROC, "userret: upcall thread %p (pid %d, %s)", td, td->td_proc->p_pid, td->td_proc->p_comm); /* * Set user context to the UTS. * Will use Giant in cpu_thread_clean() because it uses * kmem_free(kernel_map, ...) */ cpu_set_upcall_kse(td, ku); mtx_lock_spin(&sched_lock); td->td_flags &= ~TDF_UPCALLING; if (ku->ku_flags & KUF_DOUPCALL) ku->ku_flags &= ~KUF_DOUPCALL; mtx_unlock_spin(&sched_lock); /* * Unhook the list of completed threads. * anything that completes after this gets to * come in next time. * Put the list of completed thread mailboxes on * this KSE's mailbox. */ error = thread_link_mboxes(kg, ku); if (error) goto out; /* * Set state and clear the thread mailbox pointer. * From now on we are just a bound outgoing process. * **Problem** userret is often called several times. * it would be nice if this all happenned only on the first * time through. (the scan for extra work etc.) */ error = suword((caddr_t)&ku->ku_mailbox->km_curthread, 0); if (error) goto out; /* Export current system time */ nanotime(&ts); error = copyout(&ts, (caddr_t)&ku->ku_mailbox->km_timeofday, sizeof(ts)); } out: if (error) { /* * Things are going to be so screwed we should just kill * the process. * how do we do that? */ PROC_LOCK(td->td_proc); psignal(td->td_proc, SIGSEGV); PROC_UNLOCK(td->td_proc); } else { /* * Optimisation: * Ensure that we have a spare thread available, * for when we re-enter the kernel. */ if (td->td_standin == NULL) thread_alloc_spare(td, NULL); } /* * Clear thread mailbox first, then clear system tick count. * The order is important because thread_statclock() use * mailbox pointer to see if it is an userland thread or * an UTS kernel thread. */ td->td_mailbox = NULL; td->td_usticks = 0; return (error); /* go sync */ } /* * Enforce single-threading. * * Returns 1 if the caller must abort (another thread is waiting to * exit the process or similar). Process is locked! * Returns 0 when you are successfully the only thread running. * A process has successfully single threaded in the suspend mode when * There are no threads in user mode. Threads in the kernel must be * allowed to continue until they get to the user boundary. They may even * copy out their return values and data before suspending. They may however be * accellerated in reaching the user boundary as we will wake up * any sleeping threads that are interruptable. (PCATCH). */ int thread_single(int force_exit) { struct thread *td; struct thread *td2; struct proc *p; td = curthread; p = td->td_proc; mtx_assert(&Giant, MA_OWNED); PROC_LOCK_ASSERT(p, MA_OWNED); KASSERT((td != NULL), ("curthread is NULL")); if ((p->p_flag & P_THREADED) == 0 && p->p_numthreads == 1) return (0); /* Is someone already single threading? */ if (p->p_singlethread) return (1); if (force_exit == SINGLE_EXIT) { p->p_flag |= P_SINGLE_EXIT; } else p->p_flag &= ~P_SINGLE_EXIT; p->p_flag |= P_STOPPED_SINGLE; p->p_singlethread = td; /* XXXKSE Which lock protects the below values? */ while ((p->p_numthreads - p->p_suspcount) != 1) { mtx_lock_spin(&sched_lock); FOREACH_THREAD_IN_PROC(p, td2) { if (td2 == td) continue; td->td_flags |= TDF_ASTPENDING; if (TD_IS_INHIBITED(td2)) { if (force_exit == SINGLE_EXIT) { if (TD_IS_SUSPENDED(td2)) { thread_unsuspend_one(td2); } if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR)) { if (td2->td_flags & TDF_CVWAITQ) cv_abort(td2); else abortsleep(td2); } } else { if (TD_IS_SUSPENDED(td2)) continue; /* * maybe other inhibitted states too? * XXXKSE Is it totally safe to * suspend a non-interruptable thread? */ if (td2->td_inhibitors & (TDI_SLEEPING | TDI_SWAPPED)) thread_suspend_one(td2); } } } /* * Maybe we suspended some threads.. was it enough? */ if ((p->p_numthreads - p->p_suspcount) == 1) { mtx_unlock_spin(&sched_lock); break; } /* * Wake us up when everyone else has suspended. * In the mean time we suspend as well. */ thread_suspend_one(td); /* XXX If you recursed this is broken. */ mtx_unlock(&Giant); PROC_UNLOCK(p); p->p_stats->p_ru.ru_nvcsw++; mi_switch(); mtx_unlock_spin(&sched_lock); mtx_lock(&Giant); PROC_LOCK(p); } if (force_exit == SINGLE_EXIT) { if (td->td_upcall) { mtx_lock_spin(&sched_lock); upcall_remove(td); mtx_unlock_spin(&sched_lock); } kse_purge(p, td); } return (0); } /* * Called in from locations that can safely check to see * whether we have to suspend or at least throttle for a * single-thread event (e.g. fork). * * Such locations include userret(). * If the "return_instead" argument is non zero, the thread must be able to * accept 0 (caller may continue), or 1 (caller must abort) as a result. * * The 'return_instead' argument tells the function if it may do a * thread_exit() or suspend, or whether the caller must abort and back * out instead. * * If the thread that set the single_threading request has set the * P_SINGLE_EXIT bit in the process flags then this call will never return * if 'return_instead' is false, but will exit. * * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 *---------------+--------------------+--------------------- * 0 | returns 0 | returns 0 or 1 * | when ST ends | immediatly *---------------+--------------------+--------------------- * 1 | thread exits | returns 1 * | | immediatly * 0 = thread_exit() or suspension ok, * other = return error instead of stopping the thread. * * While a full suspension is under effect, even a single threading * thread would be suspended if it made this call (but it shouldn't). * This call should only be made from places where * thread_exit() would be safe as that may be the outcome unless * return_instead is set. */ int thread_suspend_check(int return_instead) { struct thread *td; struct proc *p; struct ksegrp *kg; td = curthread; p = td->td_proc; kg = td->td_ksegrp; PROC_LOCK_ASSERT(p, MA_OWNED); while (P_SHOULDSTOP(p)) { if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { KASSERT(p->p_singlethread != NULL, ("singlethread not set")); /* * The only suspension in action is a * single-threading. Single threader need not stop. * XXX Should be safe to access unlocked * as it can only be set to be true by us. */ if (p->p_singlethread == td) return (0); /* Exempt from stopping. */ } if (return_instead) return (1); mtx_lock_spin(&sched_lock); thread_stopped(p); /* * If the process is waiting for us to exit, * this thread should just suicide. * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. */ if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { while (mtx_owned(&Giant)) mtx_unlock(&Giant); if (p->p_flag & P_THREADED) thread_exit(); else thr_exit1(); } mtx_assert(&Giant, MA_NOTOWNED); /* * When a thread suspends, it just * moves to the processes's suspend queue * and stays there. */ thread_suspend_one(td); PROC_UNLOCK(p); if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { if (p->p_numthreads == p->p_suspcount) { thread_unsuspend_one(p->p_singlethread); } } p->p_stats->p_ru.ru_nivcsw++; mi_switch(); mtx_unlock_spin(&sched_lock); PROC_LOCK(p); } return (0); } void thread_suspend_one(struct thread *td) { struct proc *p = td->td_proc; mtx_assert(&sched_lock, MA_OWNED); KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); p->p_suspcount++; TD_SET_SUSPENDED(td); TAILQ_INSERT_TAIL(&p->p_suspended, td, td_runq); /* * Hack: If we are suspending but are on the sleep queue * then we are in msleep or the cv equivalent. We * want to look like we have two Inhibitors. * May already be set.. doesn't matter. */ if (TD_ON_SLEEPQ(td)) TD_SET_SLEEPING(td); } void thread_unsuspend_one(struct thread *td) { struct proc *p = td->td_proc; mtx_assert(&sched_lock, MA_OWNED); TAILQ_REMOVE(&p->p_suspended, td, td_runq); TD_CLR_SUSPENDED(td); p->p_suspcount--; setrunnable(td); } /* * Allow all threads blocked by single threading to continue running. */ void thread_unsuspend(struct proc *p) { struct thread *td; mtx_assert(&sched_lock, MA_OWNED); PROC_LOCK_ASSERT(p, MA_OWNED); if (!P_SHOULDSTOP(p)) { while (( td = TAILQ_FIRST(&p->p_suspended))) { thread_unsuspend_one(td); } } else if ((P_SHOULDSTOP(p) == P_STOPPED_SINGLE) && (p->p_numthreads == p->p_suspcount)) { /* * Stopping everything also did the job for the single * threading request. Now we've downgraded to single-threaded, * let it continue. */ thread_unsuspend_one(p->p_singlethread); } } void thread_single_end(void) { struct thread *td; struct proc *p; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); p->p_flag &= ~P_STOPPED_SINGLE; p->p_singlethread = NULL; /* * If there are other threads they mey now run, * unless of course there is a blanket 'stop order' * on the process. The single threader must be allowed * to continue however as this is a bad place to stop. */ if ((p->p_numthreads != 1) && (!P_SHOULDSTOP(p))) { mtx_lock_spin(&sched_lock); while (( td = TAILQ_FIRST(&p->p_suspended))) { thread_unsuspend_one(td); } mtx_unlock_spin(&sched_lock); } } Index: head/sys/sys/proc.h =================================================================== --- head/sys/sys/proc.h (revision 113640) +++ head/sys/sys/proc.h (revision 113641) @@ -1,964 +1,965 @@ /*- * Copyright (c) 1986, 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)proc.h 8.15 (Berkeley) 5/19/95 * $FreeBSD$ */ #ifndef _SYS_PROC_H_ #define _SYS_PROC_H_ #include /* For struct callout. */ #include /* For struct klist. */ #ifndef _KERNEL #include #endif #include #include #include #include #include /* XXX */ #include #include #include #include #ifndef _KERNEL #include /* For structs itimerval, timeval. */ #else #include #endif #include #include #include /* Machine-dependent proc substruct. */ /* * One structure allocated per session. * * List of locks * (m) locked by s_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct session { int s_count; /* (m) Ref cnt; pgrps in session. */ struct proc *s_leader; /* (m + e) Session leader. */ struct vnode *s_ttyvp; /* (m) Vnode of controlling tty. */ struct tty *s_ttyp; /* (m) Controlling tty. */ pid_t s_sid; /* (c) Session ID. */ /* (m) Setlogin() name: */ char s_login[roundup(MAXLOGNAME, sizeof(long))]; struct mtx s_mtx; /* Mutex to protect members */ }; /* * One structure allocated per process group. * * List of locks * (m) locked by pg_mtx mtx * (e) locked by proctree_lock sx * (c) const until freeing */ struct pgrp { LIST_ENTRY(pgrp) pg_hash; /* (e) Hash chain. */ LIST_HEAD(, proc) pg_members; /* (m + e) Pointer to pgrp members. */ struct session *pg_session; /* (c) Pointer to session. */ struct sigiolst pg_sigiolst; /* (m) List of sigio sources. */ pid_t pg_id; /* (c) Pgrp id. */ int pg_jobc; /* (m) job cntl proc count */ struct mtx pg_mtx; /* Mutex to protect members */ }; struct procsig { sigset_t ps_sigignore; /* Signals being ignored. */ sigset_t ps_sigcatch; /* Signals being caught by user. */ int ps_flag; struct sigacts *ps_sigacts; /* Signal actions, state. */ int ps_refcnt; }; #define PS_NOCLDWAIT 0x0001 /* No zombies if child dies */ #define PS_NOCLDSTOP 0x0002 /* No SIGCHLD when children stop. */ #define PS_CLDSIGIGN 0x0004 /* The SIGCHLD handler is SIG_IGN. */ /* * pargs, used to hold a copy of the command line, if it had a sane length. */ struct pargs { u_int ar_ref; /* Reference count. */ u_int ar_length; /* Length. */ u_char ar_args[1]; /* Arguments. */ }; /*- * Description of a process. * * This structure contains the information needed to manage a thread of * control, known in UN*X as a process; it has references to substructures * containing descriptions of things that the process uses, but may share * with related processes. The process structure and the substructures * are always addressable except for those marked "(CPU)" below, * which might be addressable only on a processor on which the process * is running. * * Below is a key of locks used to protect each member of struct proc. The * lock is indicated by a reference to a specific character in parens in the * associated comment. * * - not yet protected * a - only touched by curproc or parent during fork/wait * b - created at fork, never changes * (exception aiods switch vmspaces, but they are also * marked 'P_SYSTEM' so hopefully it will be left alone) * c - locked by proc mtx * d - locked by allproc_lock lock * e - locked by proctree_lock lock * f - session mtx * g - process group mtx * h - callout_lock mtx * i - by curproc or the master session mtx * j - locked by sched_lock mtx * k - only accessed by curthread * l - the attaching proc or attaching proc parent * m - Giant * n - not locked, lazy * o - ktrace lock * p - select lock (sellock) * r - p_peers lock * * If the locking key specifies two identifiers (for example, p_pptr) then * either lock is sufficient for read access, but both locks must be held * for write access. */ struct ithd; struct ke_sched; struct kg_sched; struct nlminfo; struct p_sched; struct td_sched; struct trapframe; /* * Here we define the four structures used for process information. * * The first is the thread. It might be though of as a "Kernel * Schedulable Entity Context". * This structure contains all the information as to where a thread of * execution is now, or was when it was suspended, why it was suspended, * and anything else that will be needed to restart it when it is * rescheduled. Always associated with a KSE when running, but can be * reassigned to an equivalent KSE when being restarted for * load balancing. Each of these is associated with a kernel stack * and a pcb. * * It is important to remember that a particular thread structure only * exists as long as the system call or kernel entrance (e.g. by pagefault) * which it is currently executing. It should therefore NEVER be referenced * by pointers in long lived structures that live longer than a single * request. If several threads complete their work at the same time, * they will all rewind their stacks to the user boundary, report their * completion state, and all but one will be freed. That last one will * be kept to provide a kernel stack and pcb for the NEXT syscall or kernel * entrance. (basically to save freeing and then re-allocating it) The KSE * keeps a cached thread available to allow it to quickly * get one when it needs a new one. There is also a system * cache of free threads. Threads have priority and partake in priority * inheritance schemes. */ struct thread; /* * The second structure is the Kernel Schedulable Entity. (KSE) * It represents the ability to take a slot in the scheduler queue. * As long as this is scheduled, it could continue to run any threads that * are assigned to the KSEGRP (see later) until either it runs out * of runnable threads of high enough priority, or CPU. * It runs on one CPU and is assigned a quantum of time. When a thread is * blocked, The KSE continues to run and will search for another thread * in a runnable state amongst those it has. It May decide to return to user * mode with a new 'empty' thread if there are no runnable threads. * Threads are temporarily associated with a KSE for scheduling reasons. */ struct kse; /* * The KSEGRP is allocated resources across a number of CPUs. * (Including a number of CPUxQUANTA. It parcels these QUANTA up among * its KSEs, each of which should be running in a different CPU. * BASE priority and total available quanta are properties of a KSEGRP. * Multiple KSEGRPs in a single process compete against each other * for total quanta in the same way that a forked child competes against * it's parent process. */ struct ksegrp; /* * A process is the owner of all system resources allocated to a task * except CPU quanta. * All KSEGs under one process see, and have the same access to, these * resources (e.g. files, memory, sockets, permissions kqueues). * A process may compete for CPU cycles on the same basis as a * forked process cluster by spawning several KSEGRPs. */ struct proc; /*************** * In pictures: With a single run queue used by all processors: RUNQ: --->KSE---KSE--... SLEEPQ:[]---THREAD---THREAD---THREAD | / []---THREAD KSEG---THREAD--THREAD--THREAD [] []---THREAD---THREAD (processors run THREADs from the KSEG until they are exhausted or the KSEG exhausts its quantum) With PER-CPU run queues: KSEs on the separate run queues directly They would be given priorities calculated from the KSEG. * *****************/ /* * Kernel runnable context (thread). * This is what is put to sleep and reactivated. * The first KSE available in the correct group will run this thread. * If several are available, use the one on the same CPU as last time. * When waiting to be run, threads are hung off the KSEGRP in priority order. * with N runnable and queued KSEs in the KSEGRP, the first N threads * are linked to them. Other threads are not yet assigned. */ struct thread { struct proc *td_proc; /* Associated process. */ struct ksegrp *td_ksegrp; /* Associated KSEG. */ TAILQ_ENTRY(thread) td_plist; /* All threads in this proc */ TAILQ_ENTRY(thread) td_kglist; /* All threads in this ksegrp */ /* The two queues below should someday be merged */ TAILQ_ENTRY(thread) td_slpq; /* (j) Sleep queue. XXXKSE */ TAILQ_ENTRY(thread) td_lockq; /* (j) Lock queue. XXXKSE */ TAILQ_ENTRY(thread) td_runq; /* (j) Run queue(s). XXXKSE */ TAILQ_HEAD(, selinfo) td_selq; /* (p) List of selinfos. */ /* Cleared during fork1() or thread_sched_upcall() */ #define td_startzero td_flags int td_flags; /* (j) TDF_* flags. */ int td_inhibitors; /* (j) Why can not run */ struct kse *td_last_kse; /* (j) Previous value of td_kse */ struct kse *td_kse; /* (j) Current KSE if running. */ int td_dupfd; /* (k) Ret value from fdopen. XXX */ void *td_wchan; /* (j) Sleep address. */ const char *td_wmesg; /* (j) Reason for sleep. */ u_char td_lastcpu; /* (j) Last cpu we were on. */ u_char td_inktr; /* (k) Currently handling a KTR. */ u_char td_inktrace; /* (k) Currently handling a KTRACE. */ u_char td_oncpu; /* (j) Which cpu we are on. */ short td_locks; /* (k) DEBUG: lockmgr count of locks */ struct mtx *td_blocked; /* (j) Mutex process is blocked on. */ struct ithd *td_ithd; /* (b) For interrupt threads only. */ const char *td_lockname; /* (j) Name of lock blocked on. */ LIST_HEAD(, mtx) td_contested; /* (j) Contested locks. */ struct lock_list_entry *td_sleeplocks; /* (k) Held sleep locks. */ int td_intr_nesting_level; /* (k) Interrupt recursion. */ struct kse_thr_mailbox *td_mailbox; /* The userland mailbox address */ struct ucred *td_ucred; /* (k) Reference to credentials. */ void (*td_switchin)(void); /* (k) Switchin special func. */ struct thread *td_standin; /* (?) Use this for an upcall */ u_int td_prticks; /* (?) Profclock hits in sys for user */ struct kse_upcall *td_upcall; /* our upcall structure. */ u_int64_t td_sticks; /* (j) Statclock hits in system mode. */ u_int td_uuticks; /* Statclock hits in user, for UTS */ u_int td_usticks; /* Statclock hits in kernel, for UTS */ u_int td_critnest; /* (k) Critical section nest level. */ sigset_t td_oldsigmask; /* (c) Saved mask from pre sigpause. */ sigset_t td_sigmask; /* (c) Current signal mask. */ sigset_t td_siglist; /* (c) Sigs arrived, not delivered. */ STAILQ_HEAD(, thread) td_umtxq; /* (p) List of threads blocked by us. */ STAILQ_ENTRY(thread) td_umtx; /* (p) Link for when we're blocked. */ #define td_endzero td_base_pri /* Copied during fork1() or thread_sched_upcall() */ #define td_startcopy td_endzero u_char td_base_pri; /* (j) Thread base kernel priority. */ u_char td_priority; /* (j) Thread active priority. */ #define td_endcopy td_pcb /* * fields that must be manually set in fork1() or thread_sched_upcall() * or already have been set in the allocator, contstructor, etc.. */ struct pcb *td_pcb; /* (k) Kernel VA of pcb and kstack. */ enum { TDS_INACTIVE = 0x0, TDS_INHIBITED, TDS_CAN_RUN, TDS_RUNQ, TDS_RUNNING } td_state; register_t td_retval[2]; /* (k) Syscall aux returns. */ struct callout td_slpcallout; /* (h) Callout for sleep. */ struct trapframe *td_frame; /* (k) */ struct vm_object *td_kstack_obj;/* (a) Kstack object. */ vm_offset_t td_kstack; /* Kernel VA of kstack. */ int td_kstack_pages; /* Size of the kstack */ struct vm_object *td_altkstack_obj;/* (a) Alternate kstack object. */ vm_offset_t td_altkstack; /* Kernel VA of alternate kstack. */ int td_altkstack_pages; /* Size of the alternate kstack */ struct mdthread td_md; /* (k) Any machine-dependent fields. */ struct td_sched *td_sched; /* Scheduler specific data */ }; /* flags kept in td_flags */ #define TDF_OLDMASK 0x000001 /* Need to restore mask after suspend. */ #define TDF_INPANIC 0x000002 /* Caused a panic, let it drive crashdump. */ #define TDF_CAN_UNBIND 0x000004 /* Only temporarily bound. */ #define TDF_SINTR 0x000008 /* Sleep is interruptible. */ #define TDF_TIMEOUT 0x000010 /* Timing out during sleep. */ #define TDF_SELECT 0x000040 /* Selecting; wakeup/waiting danger. */ #define TDF_CVWAITQ 0x000080 /* Thread is on a cv_waitq (not slpq). */ #define TDF_UPCALLING 0x000100 /* This thread is doing an upcall. */ #define TDF_ONSLEEPQ 0x000200 /* On the sleep queue. */ #define TDF_INMSLEEP 0x000400 /* Don't recurse in msleep(). */ #define TDF_ASTPENDING 0x000800 /* Thread has some asynchronous events. */ #define TDF_TIMOFAIL 0x001000 /* Timeout from sleep after we were awake. */ #define TDF_INTERRUPT 0x002000 /* Thread is marked as interrupted. */ #define TDF_USTATCLOCK 0x004000 /* Stat clock hits in userland. */ #define TDF_OWEUPC 0x008000 /* Owe thread an addupc() call at next AST. */ #define TDF_NEEDRESCHED 0x010000 /* Thread needs to yield. */ #define TDF_NEEDSIGCHK 0x020000 /* Thread may need signal delivery. */ #define TDF_DEADLKTREAT 0x800000 /* Lock aquisition - deadlock treatment. */ #define TDI_SUSPENDED 0x0001 /* On suspension queue. */ #define TDI_SLEEPING 0x0002 /* Actually asleep! (tricky). */ #define TDI_SWAPPED 0x0004 /* Stack not in mem.. bad juju if run. */ #define TDI_LOCK 0x0008 /* Stopped on a lock. */ #define TDI_IWAIT 0x0010 /* Awaiting interrupt. */ #define TD_CAN_UNBIND(td) \ (((td)->td_flags & TDF_CAN_UNBIND) == TDF_CAN_UNBIND && \ ((td)->td_upcall != NULL)) #define TD_IS_SLEEPING(td) ((td)->td_inhibitors & TDI_SLEEPING) #define TD_ON_SLEEPQ(td) ((td)->td_wchan != NULL) #define TD_IS_SUSPENDED(td) ((td)->td_inhibitors & TDI_SUSPENDED) #define TD_IS_SWAPPED(td) ((td)->td_inhibitors & TDI_SWAPPED) #define TD_ON_LOCK(td) ((td)->td_inhibitors & TDI_LOCK) #define TD_AWAITING_INTR(td) ((td)->td_inhibitors & TDI_IWAIT) #define TD_IS_RUNNING(td) ((td)->td_state == TDS_RUNNING) #define TD_ON_RUNQ(td) ((td)->td_state == TDS_RUNQ) #define TD_CAN_RUN(td) ((td)->td_state == TDS_CAN_RUN) #define TD_IS_INHIBITED(td) ((td)->td_state == TDS_INHIBITED) #define TD_SET_INHIB(td, inhib) do { \ (td)->td_state = TDS_INHIBITED; \ (td)->td_inhibitors |= (inhib); \ } while (0) #define TD_CLR_INHIB(td, inhib) do { \ if (((td)->td_inhibitors & (inhib)) && \ (((td)->td_inhibitors &= ~(inhib)) == 0)) \ (td)->td_state = TDS_CAN_RUN; \ } while (0) #define TD_SET_SLEEPING(td) TD_SET_INHIB((td), TDI_SLEEPING) #define TD_SET_SWAPPED(td) TD_SET_INHIB((td), TDI_SWAPPED) #define TD_SET_LOCK(td) TD_SET_INHIB((td), TDI_LOCK) #define TD_SET_SUSPENDED(td) TD_SET_INHIB((td), TDI_SUSPENDED) #define TD_SET_IWAIT(td) TD_SET_INHIB((td), TDI_IWAIT) #define TD_SET_EXITING(td) TD_SET_INHIB((td), TDI_EXITING) #define TD_CLR_SLEEPING(td) TD_CLR_INHIB((td), TDI_SLEEPING) #define TD_CLR_SWAPPED(td) TD_CLR_INHIB((td), TDI_SWAPPED) #define TD_CLR_LOCK(td) TD_CLR_INHIB((td), TDI_LOCK) #define TD_CLR_SUSPENDED(td) TD_CLR_INHIB((td), TDI_SUSPENDED) #define TD_CLR_IWAIT(td) TD_CLR_INHIB((td), TDI_IWAIT) #define TD_SET_RUNNING(td) do {(td)->td_state = TDS_RUNNING; } while (0) #define TD_SET_RUNQ(td) do {(td)->td_state = TDS_RUNQ; } while (0) #define TD_SET_CAN_RUN(td) do {(td)->td_state = TDS_CAN_RUN; } while (0) #define TD_SET_ON_SLEEPQ(td) do {(td)->td_flags |= TDF_ONSLEEPQ; } while (0) #define TD_CLR_ON_SLEEPQ(td) do { \ (td)->td_flags &= ~TDF_ONSLEEPQ; \ (td)->td_wchan = NULL; \ } while (0) /* * The schedulable entity that can be given a context to run. * A process may have several of these. Probably one per processor * but posibly a few more. In this universe they are grouped * with a KSEG that contains the priority and niceness * for the group. */ struct kse { struct proc *ke_proc; /* Associated process. */ struct ksegrp *ke_ksegrp; /* Associated KSEG. */ TAILQ_ENTRY(kse) ke_kglist; /* Queue of all KSEs in ke_ksegrp. */ TAILQ_ENTRY(kse) ke_kgrlist; /* Queue of all KSEs in this state. */ TAILQ_ENTRY(kse) ke_procq; /* (j) Run queue. */ #define ke_startzero ke_flags int ke_flags; /* (j) KEF_* flags. */ struct thread *ke_thread; /* Active associated thread. */ fixpt_t ke_pctcpu; /* (j) %cpu during p_swtime. */ u_char ke_oncpu; /* (j) Which cpu we are on. */ char ke_rqindex; /* (j) Run queue index. */ enum { KES_UNUSED = 0x0, KES_IDLE, KES_ONRUNQ, KES_UNQUEUED, /* in transit */ KES_THREAD /* slaved to thread state */ } ke_state; /* (j) S* process status. */ #define ke_endzero ke_dummy u_char ke_dummy; struct ke_sched *ke_sched; /* Scheduler specific data */ }; /* flags kept in ke_flags */ #define KEF_IDLEKSE 0x00004 /* A 'Per CPU idle process'.. has one thread */ #define KEF_DIDRUN 0x02000 /* KSE actually ran. */ #define KEF_EXIT 0x04000 /* KSE is being killed. */ /* * The upcall management structure. * The upcall is used when returning to userland. If a thread does not have * an upcall on return to userland the thread exports its context and exits. */ struct kse_upcall { TAILQ_ENTRY(kse_upcall) ku_link; /* List of upcalls in KSEG. */ struct ksegrp *ku_ksegrp; /* Associated KSEG. */ struct thread *ku_owner; /* owning thread */ int ku_flags; /* KUF_* flags. */ struct kse_mailbox *ku_mailbox; /* userland mailbox address. */ stack_t ku_stack; /* userland upcall stack. */ void *ku_func; /* userland upcall function. */ }; #define KUF_DOUPCALL 0x00001 /* Do upcall now, don't wait */ /* * Kernel-scheduled entity group (KSEG). The scheduler considers each KSEG to * be an indivisible unit from a time-sharing perspective, though each KSEG may * contain multiple KSEs. */ struct ksegrp { struct proc *kg_proc; /* Process that contains this KSEG. */ TAILQ_ENTRY(ksegrp) kg_ksegrp; /* Queue of KSEGs in kg_proc. */ TAILQ_HEAD(, kse) kg_kseq; /* (ke_kglist) All KSEs. */ TAILQ_HEAD(, kse) kg_iq; /* (ke_kgrlist) All idle KSEs. */ TAILQ_HEAD(, thread) kg_threads;/* (td_kglist) All threads. */ TAILQ_HEAD(, thread) kg_runq; /* (td_runq) waiting RUNNABLE threads */ TAILQ_HEAD(, thread) kg_slpq; /* (td_runq) NONRUNNABLE threads. */ TAILQ_HEAD(, kse_upcall) kg_upcalls; /* All upcalls in the group */ #define kg_startzero kg_estcpu u_int kg_estcpu; /* Sum of the same field in KSEs. */ u_int kg_slptime; /* (j) How long completely blocked. */ struct thread *kg_last_assigned; /* (j) Last thread assigned to a KSE */ int kg_runnable; /* (j) Num runnable threads on queue. */ int kg_runq_kses; /* (j) Num KSEs on runq. */ int kg_idle_kses; /* (j) Num KSEs on iq */ int kg_numupcalls; /* (j) Num upcalls */ int kg_upsleeps; /* (c) Num threads in kse_release() */ struct kse_thr_mailbox *kg_completed; /* (c) completed thread mboxes */ int kg_nextupcall; /* next upcall time */ int kg_upquantum; /* quantum to schedule an upcall */ #define kg_endzero kg_pri_class #define kg_startcopy kg_endzero u_char kg_pri_class; /* (j) Scheduling class. */ u_char kg_user_pri; /* (j) User pri from estcpu and nice. */ char kg_nice; /* (j?/k?) Process "nice" value. */ #define kg_endcopy kg_numthreads int kg_numthreads; /* (j) Num threads in total */ int kg_kses; /* (j) Num KSEs in group. */ struct kg_sched *kg_sched; /* Scheduler specific data */ }; /* * The old fashionned process. May have multiple threads, KSEGRPs * and KSEs. Starts off with a single embedded KSEGRP, KSE and THREAD. */ struct proc { LIST_ENTRY(proc) p_list; /* (d) List of all processes. */ TAILQ_HEAD(, ksegrp) p_ksegrps; /* (kg_ksegrp) All KSEGs. */ TAILQ_HEAD(, thread) p_threads; /* (td_plist) Threads. (shortcut) */ TAILQ_HEAD(, thread) p_suspended; /* (td_runq) suspended threads */ struct ucred *p_ucred; /* (c) Process owner's identity. */ struct filedesc *p_fd; /* (b) Ptr to open files structure. */ /* Accumulated stats for all KSEs? */ struct pstats *p_stats; /* (b) Accounting/statistics (CPU). */ struct plimit *p_limit; /* (m) Process limits. */ struct vm_object *p_upages_obj; /* (a) Upages object. */ struct procsig *p_procsig; /* (c) Signal actions, state (CPU). */ /*struct ksegrp p_ksegrp; struct kse p_kse; */ /* * The following don't make too much sense.. * See the td_ or ke_ versions of the same flags */ int p_flag; /* (c) P_* flags. */ int p_sflag; /* (j) PS_* flags. */ enum { PRS_NEW = 0, /* In creation */ PRS_NORMAL, /* KSEs can be run */ PRS_ZOMBIE } p_state; /* (j) S* process status. */ pid_t p_pid; /* (b) Process identifier. */ LIST_ENTRY(proc) p_hash; /* (d) Hash chain. */ LIST_ENTRY(proc) p_pglist; /* (g + e) List of processes in pgrp. */ struct proc *p_pptr; /* (c + e) Pointer to parent process. */ LIST_ENTRY(proc) p_sibling; /* (e) List of sibling processes. */ LIST_HEAD(, proc) p_children; /* (e) Pointer to list of children. */ struct mtx p_mtx; /* (k) Lock for this struct. */ /* The following fields are all zeroed upon creation in fork. */ #define p_startzero p_oppid pid_t p_oppid; /* (c + e) Save ppid in ptrace. XXX */ struct vmspace *p_vmspace; /* (b) Address space. */ u_int p_swtime; /* (j) Time swapped in or out. */ struct itimerval p_realtimer; /* (c) Alarm timer. */ struct bintime p_runtime; /* (j) Real time. */ u_int64_t p_uu; /* (j) Previous user time in usec. */ u_int64_t p_su; /* (j) Previous system time in usec. */ u_int64_t p_iu; /* (j) Previous intr time in usec. */ u_int64_t p_uticks; /* (j) Statclock hits in user mode. */ u_int64_t p_sticks; /* (j) Statclock hits in system mode. */ u_int64_t p_iticks; /* (j) Statclock hits in intr. */ int p_profthreads; /* (c) Num threads in addupc_task */ int p_maxthrwaits; /* (c) Max threads num waiters */ int p_traceflag; /* (o) Kernel trace points. */ struct vnode *p_tracevp; /* (c + o) Trace to vnode. */ struct ucred *p_tracecred; /* (o) Credentials to trace with. */ struct vnode *p_textvp; /* (b) Vnode of executable. */ sigset_t p_siglist; /* (c) Sigs not delivered to a td. */ char p_lock; /* (c) Proclock (prevent swap) count. */ struct klist p_klist; /* (c) Knotes attached to this proc. */ struct sigiolst p_sigiolst; /* (c) List of sigio sources. */ int p_sigparent; /* (c) Signal to parent on exit. */ int p_sig; /* (n) For core dump/debugger XXX. */ u_long p_code; /* (n) For core dump/debugger XXX. */ u_int p_stops; /* (c) Stop event bitmask. */ u_int p_stype; /* (c) Stop event type. */ char p_step; /* (c) Process is stopped. */ u_char p_pfsflags; /* (c) Procfs flags. */ struct nlminfo *p_nlminfo; /* (?) Only used by/for lockd. */ void *p_aioinfo; /* (c) ASYNC I/O info. */ struct thread *p_singlethread;/* (j) If single threading this is it */ int p_suspcount; /* (j) # threads in suspended mode */ /* End area that is zeroed on creation. */ #define p_endzero p_sigstk /* The following fields are all copied upon creation in fork. */ #define p_startcopy p_endzero stack_t p_sigstk; /* (c) Stack ptr and on-stack flag. */ u_int p_magic; /* (b) Magic number. */ char p_comm[MAXCOMLEN + 1]; /* (b) Process name. */ struct pgrp *p_pgrp; /* (c + e) Pointer to process group. */ struct sysentvec *p_sysent; /* (b) Syscall dispatch info. */ struct pargs *p_args; /* (c) Process arguments. */ rlim_t p_cpulimit; /* (j) Current CPU limit in seconds. */ /* End area that is copied on creation. */ #define p_endcopy p_xstat u_short p_xstat; /* (c) Exit status; also stop sig. */ int p_numthreads; /* (?) number of threads */ int p_numksegrps; /* (?) number of ksegrps */ struct mdproc p_md; /* (c) Any machine-dependent fields. */ struct callout p_itcallout; /* (h) Interval timer callout. */ struct user *p_uarea; /* (k) Kernel VA of u-area (CPU) */ u_short p_acflag; /* (c) Accounting flags. */ struct rusage *p_ru; /* (a) Exit information. XXX */ struct proc *p_peers; /* (r) */ struct proc *p_leader; /* (b) */ void *p_emuldata; /* (c) Emulator state data. */ struct label p_label; /* process (not subject) MAC label */ struct p_sched *p_sched; /* Scheduler specific data */ }; #define p_rlimit p_limit->pl_rlimit #define p_sigacts p_procsig->ps_sigacts #define p_sigignore p_procsig->ps_sigignore #define p_sigcatch p_procsig->ps_sigcatch #define p_session p_pgrp->pg_session #define p_pgid p_pgrp->pg_id #define NOCPU 0xff /* For when we aren't on a CPU. (SMP) */ /* Status values (p_stat). */ /* These flags are kept in p_flag. */ #define P_ADVLOCK 0x00001 /* Process may hold a POSIX advisory lock. */ #define P_CONTROLT 0x00002 /* Has a controlling terminal. */ #define P_KTHREAD 0x00004 /* Kernel thread. (*)*/ #define P_NOLOAD 0x00008 /* Ignore during load avg calculations. */ #define P_PPWAIT 0x00010 /* Parent is waiting for child to exec/exit. */ #define P_SUGID 0x00100 /* Had set id privileges since last exec. */ #define P_SYSTEM 0x00200 /* System proc: no sigs, stats or swapping. */ #define P_WAITED 0x01000 /* Someone is waiting for us */ #define P_WEXIT 0x02000 /* Working on exiting. */ #define P_EXEC 0x04000 /* Process called exec. */ #define P_THREADED 0x08000 /* Process is using threads. */ #define P_CONTINUED 0x10000 /* Proc has continued from a stopped state. */ #define P_PROTECTED 0x20000 /* Do not kill on memory overcommit. */ /* flags that control how threads may be suspended for some reason */ #define P_STOPPED_SIG 0x20000 /* Stopped due to SIGSTOP/SIGTSTP */ #define P_STOPPED_TRACE 0x40000 /* Stopped because of tracing */ #define P_STOPPED_SINGLE 0x80000 /* Only one thread can continue */ /* (not to user) */ #define P_SINGLE_EXIT 0x00400 /* Threads suspending should exit, */ /* not wait */ #define P_TRACED 0x00800 /* Debugged process being traced. */ #define P_STOPPED (P_STOPPED_SIG|P_STOPPED_SINGLE|P_STOPPED_TRACE) #define P_SHOULDSTOP(p) ((p)->p_flag & P_STOPPED) /* Should be moved to machine-dependent areas. */ #define P_UNUSED100000 0x100000 #define P_COWINPROGRESS 0x400000 /* Snapshot copy-on-write in progress. */ #define P_JAILED 0x1000000 /* Process is in jail. */ #define P_ALTSTACK 0x2000000 /* Have alternate signal stack. */ #define P_INEXEC 0x4000000 /* Process is in execve(). */ /* These flags are kept in p_sflag and are protected with sched_lock. */ #define PS_INMEM 0x00001 /* Loaded into memory. */ #define PS_XCPU 0x00002 /* Exceeded CPU limit. */ #define PS_PROFIL 0x00004 /* Has started profiling. */ #define PS_STOPPROF 0x00008 /* Has thread in requesting to stop prof */ #define PS_ALRMPEND 0x00020 /* Pending SIGVTALRM needs to be posted. */ #define PS_PROFPEND 0x00040 /* Pending SIGPROF needs to be posted. */ #define PS_SWAPINREQ 0x00100 /* Swapin request due to wakeup. */ #define PS_SWAPPING 0x00200 /* Process is being swapped. */ #define PS_SWAPPINGIN 0x04000 /* Swapin in progress. */ #define PS_MACPEND 0x08000 /* Ast()-based MAC event pending. */ /* used only in legacy conversion code */ #define SIDL 1 /* Process being created by fork. */ #define SRUN 2 /* Currently runnable. */ #define SSLEEP 3 /* Sleeping on an address. */ #define SSTOP 4 /* Process debugging or suspension. */ #define SZOMB 5 /* Awaiting collection by parent. */ #define SWAIT 6 /* Waiting for interrupt. */ #define SLOCK 7 /* Blocked on a lock. */ #define P_MAGIC 0xbeefface #ifdef _KERNEL #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_PARGS); MALLOC_DECLARE(M_PGRP); MALLOC_DECLARE(M_SESSION); MALLOC_DECLARE(M_SUBPROC); MALLOC_DECLARE(M_ZOMBIE); #endif #define FOREACH_PROC_IN_SYSTEM(p) \ LIST_FOREACH((p), &allproc, p_list) #define FOREACH_KSEGRP_IN_PROC(p, kg) \ TAILQ_FOREACH((kg), &(p)->p_ksegrps, kg_ksegrp) #define FOREACH_THREAD_IN_GROUP(kg, td) \ TAILQ_FOREACH((td), &(kg)->kg_threads, td_kglist) #define FOREACH_KSE_IN_GROUP(kg, ke) \ TAILQ_FOREACH((ke), &(kg)->kg_kseq, ke_kglist) #define FOREACH_UPCALL_IN_GROUP(kg, ku) \ TAILQ_FOREACH((ku), &(kg)->kg_upcalls, ku_link) #define FOREACH_THREAD_IN_PROC(p, td) \ TAILQ_FOREACH((td), &(p)->p_threads, td_plist) /* XXXKSE the lines below should probably only be used in 1:1 code */ #define FIRST_THREAD_IN_PROC(p) TAILQ_FIRST(&p->p_threads) #define FIRST_KSEGRP_IN_PROC(p) TAILQ_FIRST(&p->p_ksegrps) #define FIRST_KSE_IN_KSEGRP(kg) TAILQ_FIRST(&kg->kg_kseq) #define FIRST_KSE_IN_PROC(p) FIRST_KSE_IN_KSEGRP(FIRST_KSEGRP_IN_PROC(p)) static __inline int sigonstack(size_t sp) { register struct thread *td = curthread; struct proc *p = td->td_proc; return ((p->p_flag & P_ALTSTACK) ? #if defined(COMPAT_43) || defined(COMPAT_SUNOS) ((p->p_sigstk.ss_size == 0) ? (p->p_sigstk.ss_flags & SS_ONSTACK) : ((sp - (size_t)p->p_sigstk.ss_sp) < p->p_sigstk.ss_size)) #else ((sp - (size_t)p->p_sigstk.ss_sp) < p->p_sigstk.ss_size) #endif : 0); } /* * We use process IDs <= PID_MAX; PID_MAX + 1 must also fit in a pid_t, * as it is used to represent "no process group". */ #define PID_MAX 99999 #define NO_PID 100000 #define SESS_LEADER(p) ((p)->p_session->s_leader == (p)) #define SESSHOLD(s) ((s)->s_count++) #define SESSRELE(s) { \ if (--(s)->s_count == 0) \ FREE(s, M_SESSION); \ } #define STOPEVENT(p, e, v) do { \ PROC_LOCK(p); \ _STOPEVENT((p), (e), (v)); \ PROC_UNLOCK(p); \ } while (0) #define _STOPEVENT(p, e, v) do { \ PROC_LOCK_ASSERT(p, MA_OWNED); \ if ((p)->p_stops & (e)) { \ stopevent((p), (e), (v)); \ } \ } while (0) /* Lock and unlock a process. */ #define PROC_LOCK(p) mtx_lock(&(p)->p_mtx) #define PROC_TRYLOCK(p) mtx_trylock(&(p)->p_mtx) #define PROC_UNLOCK(p) mtx_unlock(&(p)->p_mtx) #define PROC_LOCKED(p) mtx_owned(&(p)->p_mtx) #define PROC_LOCK_ASSERT(p, type) mtx_assert(&(p)->p_mtx, (type)) /* Lock and unlock a process group. */ #define PGRP_LOCK(pg) mtx_lock(&(pg)->pg_mtx) #define PGRP_UNLOCK(pg) mtx_unlock(&(pg)->pg_mtx) #define PGRP_LOCKED(pg) mtx_owned(&(pg)->pg_mtx) #define PGRP_LOCK_ASSERT(pg, type) mtx_assert(&(pg)->pg_mtx, (type)) #define PGRP_LOCK_PGSIGNAL(pg) \ do { \ if ((pg) != NULL) \ PGRP_LOCK(pg); \ } while (0); #define PGRP_UNLOCK_PGSIGNAL(pg) \ do { \ if ((pg) != NULL) \ PGRP_UNLOCK(pg); \ } while (0); /* Lock and unlock a session. */ #define SESS_LOCK(s) mtx_lock(&(s)->s_mtx) #define SESS_UNLOCK(s) mtx_unlock(&(s)->s_mtx) #define SESS_LOCKED(s) mtx_owned(&(s)->s_mtx) #define SESS_LOCK_ASSERT(s, type) mtx_assert(&(s)->s_mtx, (type)) /* Hold process U-area in memory, normally for ptrace/procfs work. */ #define PHOLD(p) do { \ PROC_LOCK(p); \ _PHOLD(p); \ PROC_UNLOCK(p); \ } while (0) #define _PHOLD(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ if ((p)->p_lock++ == 0) { \ mtx_lock_spin(&sched_lock); \ faultin((p)); \ mtx_unlock_spin(&sched_lock); \ } \ } while (0) #define PRELE(p) do { \ PROC_LOCK((p)); \ _PRELE((p)); \ PROC_UNLOCK((p)); \ } while (0) #define _PRELE(p) do { \ PROC_LOCK_ASSERT((p), MA_OWNED); \ (--(p)->p_lock); \ } while (0) /* Check whether a thread is safe to be swapped out. */ #define thread_safetoswapout(td) (TD_IS_SLEEPING(td) || TD_IS_SUSPENDED(td)) /* Lock and unlock process arguments. */ #define PARGS_LOCK(p) mtx_lock(&pargs_ref_lock) #define PARGS_UNLOCK(p) mtx_unlock(&pargs_ref_lock) #define PIDHASH(pid) (&pidhashtbl[(pid) & pidhash]) extern LIST_HEAD(pidhashhead, proc) *pidhashtbl; extern u_long pidhash; #define PGRPHASH(pgid) (&pgrphashtbl[(pgid) & pgrphash]) extern LIST_HEAD(pgrphashhead, pgrp) *pgrphashtbl; extern u_long pgrphash; extern struct sx allproc_lock; extern struct sx proctree_lock; extern struct mtx pargs_ref_lock; extern struct mtx ppeers_lock; extern struct proc proc0; /* Process slot for swapper. */ extern struct thread thread0; /* Primary thread in proc0 */ extern struct ksegrp ksegrp0; /* Primary ksegrp in proc0 */ extern struct kse kse0; /* Primary kse in proc0 */ extern struct vmspace vmspace0; /* VM space for proc0. */ extern int hogticks; /* Limit on kernel cpu hogs. */ extern int nprocs, maxproc; /* Current and max number of procs. */ extern int maxprocperuid; /* Max procs per uid. */ extern u_long ps_arg_cache_limit; extern int ps_argsopen; extern int ps_showallprocs; extern int sched_quantum; /* Scheduling quantum in ticks. */ LIST_HEAD(proclist, proc); TAILQ_HEAD(procqueue, proc); TAILQ_HEAD(threadqueue, thread); extern struct proclist allproc; /* List of all processes. */ extern struct proclist zombproc; /* List of zombie processes. */ extern struct proc *initproc, *pageproc; /* Process slots for init, pager. */ extern struct proc *updateproc; /* Process slot for syncer (sic). */ extern struct uma_zone *proc_zone; extern int lastpid; struct proc *pfind(pid_t); /* Find process by id. */ struct pgrp *pgfind(pid_t); /* Find process group by id. */ struct proc *zpfind(pid_t); /* Find zombie process by id. */ void adjustrunqueue(struct thread *, int newpri); void ast(struct trapframe *framep); struct thread *choosethread(void); int cr_cansignal(struct ucred *cred, struct proc *proc, int signum); int enterpgrp(struct proc *p, pid_t pgid, struct pgrp *pgrp, struct session *sess); int enterthispgrp(struct proc *p, struct pgrp *pgrp); void faultin(struct proc *p); void fixjobc(struct proc *p, struct pgrp *pgrp, int entering); int fork1(struct thread *, int, int, struct proc **); void fork_exit(void (*)(void *, struct trapframe *), void *, struct trapframe *); void fork_return(struct thread *, struct trapframe *); int inferior(struct proc *p); int leavepgrp(struct proc *p); void mi_switch(void); int p_candebug(struct thread *td, struct proc *p); int p_cansee(struct thread *td, struct proc *p); int p_cansched(struct thread *td, struct proc *p); int p_cansignal(struct thread *td, struct proc *p, int signum); struct pargs *pargs_alloc(int len); void pargs_drop(struct pargs *pa); void pargs_free(struct pargs *pa); void pargs_hold(struct pargs *pa); void procinit(void); void threadinit(void); void proc_linkup(struct proc *p, struct ksegrp *kg, struct kse *ke, struct thread *td); void proc_reparent(struct proc *child, struct proc *newparent); int securelevel_ge(struct ucred *cr, int level); int securelevel_gt(struct ucred *cr, int level); void setrunnable(struct thread *); void setrunqueue(struct thread *); void setsugid(struct proc *p); void sleepinit(void); void stopevent(struct proc *, u_int, u_int); void cpu_idle(void); #if defined(__i386__) || defined(__sparc64__) void cpu_switch(struct thread *old, struct thread *new); void cpu_throw(struct thread *old, struct thread *new) __dead2; #else void cpu_switch(void); void cpu_throw(void) __dead2; #endif void unsleep(struct thread *); void userret(struct thread *, struct trapframe *, u_int); void cpu_exit(struct thread *); void cpu_sched_exit(struct thread *); void exit1(struct thread *, int) __dead2; void cpu_fork(struct thread *, struct proc *, struct thread *, int); void cpu_set_fork_handler(struct thread *, void (*)(void *), void *); void cpu_wait(struct proc *); /* New in KSE. */ struct ksegrp *ksegrp_alloc(void); void ksegrp_free(struct ksegrp *kg); void ksegrp_stash(struct ksegrp *kg); struct kse *kse_alloc(void); void kse_free(struct kse *ke); void kse_stash(struct kse *ke); void cpu_set_upcall(struct thread *td, void *pcb); void cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku); void cpu_thread_clean(struct thread *); void cpu_thread_exit(struct thread *); void cpu_thread_setup(struct thread *td); void kse_reassign(struct kse *ke); void kse_link(struct kse *ke, struct ksegrp *kg); void kse_unlink(struct kse *ke); void ksegrp_link(struct ksegrp *kg, struct proc *p); void ksegrp_unlink(struct ksegrp *kg); void thread_signal_add(struct thread *td, int sig); void thread_signal_upcall(struct thread *td); struct thread *thread_alloc(void); void thread_exit(void) __dead2; int thread_export_context(struct thread *td); void thread_free(struct thread *td); void thread_getcontext(struct thread *td, ucontext_t *uc); void thread_link(struct thread *td, struct ksegrp *kg); void thread_reap(void); struct thread *thread_schedule_upcall(struct thread *td, struct kse_upcall *ku); int thread_setcontext(struct thread *td, ucontext_t *uc); int thread_single(int how); #define SINGLE_NO_EXIT 0 /* values for 'how' */ #define SINGLE_EXIT 1 void thread_single_end(void); void thread_stash(struct thread *td); int thread_suspend_check(int how); void thread_suspend_one(struct thread *td); +void thread_unlink(struct thread *td); void thread_unsuspend(struct proc *p); void thread_unsuspend_one(struct thread *td); int thread_userret(struct thread *td, struct trapframe *frame); void thread_user_enter(struct proc *p, struct thread *td); void thread_wait(struct proc *p); int thread_statclock(int user); struct kse_upcall *upcall_alloc(void); void upcall_free(struct kse_upcall *ku); void upcall_link(struct kse_upcall *ku, struct ksegrp *kg); void upcall_unlink(struct kse_upcall *ku); void upcall_remove(struct thread *td); void upcall_stash(struct kse_upcall *ke); void thread_sanity_check(struct thread *td, char *); void thread_stopped(struct proc *p); void thread_switchout(struct thread *td); void thr_exit1(void); #endif /* _KERNEL */ #endif /* !_SYS_PROC_H_ */