003 File Manager
Current Path:
/usr/src/sys/kern
usr
/
src
/
sys
/
kern
/
📁
..
📄
Make.tags.inc
(2.13 KB)
📄
Makefile
(302 B)
📄
bus_if.m
(26.31 KB)
📄
capabilities.conf
(13.67 KB)
📄
clock_if.m
(1.7 KB)
📄
cpufreq_if.m
(2.27 KB)
📄
device_if.m
(10.41 KB)
📄
firmw.S
(2.15 KB)
📄
genassym.sh
(1.11 KB)
📄
genoffset.c
(1.68 KB)
📄
genoffset.sh
(3.58 KB)
📄
imgact_aout.c
(9.45 KB)
📄
imgact_binmisc.c
(18.64 KB)
📄
imgact_elf.c
(76.32 KB)
📄
imgact_elf32.c
(1.47 KB)
📄
imgact_elf64.c
(1.47 KB)
📄
imgact_shell.c
(8.41 KB)
📄
init_main.c
(24.31 KB)
📄
init_sysent.c
(95.3 KB)
📄
kern_acct.c
(19.03 KB)
📄
kern_alq.c
(24.97 KB)
📄
kern_clock.c
(21.12 KB)
📄
kern_clocksource.c
(23.34 KB)
📄
kern_condvar.c
(11.28 KB)
📄
kern_conf.c
(36.14 KB)
📄
kern_cons.c
(15.75 KB)
📄
kern_context.c
(3.59 KB)
📄
kern_cpu.c
(30.77 KB)
📄
kern_cpuset.c
(59.78 KB)
📄
kern_ctf.c
(8.73 KB)
📄
kern_descrip.c
(112.87 KB)
📄
kern_dtrace.c
(2.94 KB)
📄
kern_dump.c
(8.51 KB)
📄
kern_environment.c
(22.75 KB)
📄
kern_et.c
(7.1 KB)
📄
kern_event.c
(62.49 KB)
📄
kern_exec.c
(46.67 KB)
📄
kern_exit.c
(34.61 KB)
📄
kern_fail.c
(29.32 KB)
📄
kern_ffclock.c
(12.66 KB)
📄
kern_fork.c
(28.29 KB)
📄
kern_hhook.c
(13.58 KB)
📄
kern_idle.c
(2.74 KB)
📄
kern_intr.c
(40.44 KB)
📄
kern_jail.c
(112.67 KB)
📄
kern_kcov.c
(15.32 KB)
📄
kern_khelp.c
(9.45 KB)
📄
kern_kthread.c
(11.8 KB)
📄
kern_ktr.c
(11.93 KB)
📄
kern_ktrace.c
(31.41 KB)
📄
kern_linker.c
(54.3 KB)
📄
kern_lock.c
(46.99 KB)
📄
kern_lockf.c
(64.46 KB)
📄
kern_lockstat.c
(3.8 KB)
📄
kern_loginclass.c
(6.69 KB)
📄
kern_malloc.c
(37.09 KB)
📄
kern_mbuf.c
(43.16 KB)
📄
kern_mib.c
(24.26 KB)
📄
kern_module.c
(11.05 KB)
📄
kern_mtxpool.c
(5.82 KB)
📄
kern_mutex.c
(33.62 KB)
📄
kern_ntptime.c
(32.49 KB)
📄
kern_osd.c
(12.37 KB)
📄
kern_physio.c
(5.74 KB)
📄
kern_pmc.c
(8.89 KB)
📄
kern_poll.c
(15.86 KB)
📄
kern_priv.c
(9.14 KB)
📄
kern_proc.c
(80.01 KB)
📄
kern_procctl.c
(19.48 KB)
📄
kern_prot.c
(57.94 KB)
📄
kern_racct.c
(34.01 KB)
📄
kern_rangelock.c
(8.67 KB)
📄
kern_rctl.c
(53.87 KB)
📄
kern_resource.c
(36.66 KB)
📄
kern_rmlock.c
(28.27 KB)
📄
kern_rwlock.c
(40.72 KB)
📄
kern_sdt.c
(2.05 KB)
📄
kern_sema.c
(4.85 KB)
📄
kern_sendfile.c
(33.97 KB)
📄
kern_sharedpage.c
(10.37 KB)
📄
kern_shutdown.c
(43.34 KB)
📄
kern_sig.c
(101.89 KB)
📄
kern_switch.c
(13.85 KB)
📄
kern_sx.c
(40.27 KB)
📄
kern_synch.c
(18.17 KB)
📄
kern_syscalls.c
(6.74 KB)
📄
kern_sysctl.c
(67.24 KB)
📄
kern_tc.c
(55.73 KB)
📄
kern_thr.c
(14.14 KB)
📄
kern_thread.c
(41.75 KB)
📄
kern_time.c
(40.89 KB)
📄
kern_timeout.c
(43.08 KB)
📄
kern_tslog.c
(3.44 KB)
📄
kern_ubsan.c
(50.74 KB)
📄
kern_umtx.c
(107.14 KB)
📄
kern_uuid.c
(11.68 KB)
📄
kern_xxx.c
(10.44 KB)
📄
ksched.c
(6.56 KB)
📄
link_elf.c
(47.99 KB)
📄
link_elf_obj.c
(44.41 KB)
📄
linker_if.m
(3.96 KB)
📄
makesyscalls.sh
(23.57 KB)
📄
md4c.c
(7.89 KB)
📄
md5c.c
(9.56 KB)
📄
msi_if.m
(2.48 KB)
📄
p1003_1b.c
(8.84 KB)
📄
pic_if.m
(3.9 KB)
📄
posix4_mib.c
(5.59 KB)
📄
sched_4bsd.c
(45.03 KB)
📄
sched_ule.c
(82.65 KB)
📄
serdev_if.m
(3.49 KB)
📄
stack_protector.c
(613 B)
📄
subr_acl_nfs4.c
(37.42 KB)
📄
subr_acl_posix1e.c
(17.71 KB)
📄
subr_atomic64.c
(3.97 KB)
📄
subr_autoconf.c
(7.7 KB)
📄
subr_blist.c
(31.88 KB)
📄
subr_boot.c
(5.8 KB)
📄
subr_bufring.c
(2.21 KB)
📄
subr_bus.c
(145.4 KB)
📄
subr_bus_dma.c
(19.67 KB)
📄
subr_busdma_bufalloc.c
(5.24 KB)
📄
subr_capability.c
(11.93 KB)
📄
subr_clock.c
(10.61 KB)
📄
subr_compressor.c
(13.11 KB)
📄
subr_counter.c
(4.44 KB)
📄
subr_coverage.c
(6.17 KB)
📄
subr_csan.c
(25.39 KB)
📄
subr_devmap.c
(9.8 KB)
📄
subr_devstat.c
(16.21 KB)
📄
subr_disk.c
(8.54 KB)
📄
subr_dummy_vdso_tc.c
(1.7 KB)
📄
subr_early.c
(2.26 KB)
📄
subr_epoch.c
(25.02 KB)
📄
subr_eventhandler.c
(9.17 KB)
📄
subr_fattime.c
(9.98 KB)
📄
subr_filter.c
(12.2 KB)
📄
subr_firmware.c
(13.88 KB)
📄
subr_gtaskqueue.c
(20.19 KB)
📄
subr_hash.c
(4.8 KB)
📄
subr_hints.c
(12.87 KB)
📄
subr_intr.c
(40.61 KB)
📄
subr_kdb.c
(16.13 KB)
📄
subr_kobj.c
(7.1 KB)
📄
subr_lock.c
(18.81 KB)
📄
subr_log.c
(7.64 KB)
📄
subr_mchain.c
(11.06 KB)
📄
subr_module.c
(12.98 KB)
📄
subr_msgbuf.c
(10.6 KB)
📄
subr_param.c
(10.93 KB)
📄
subr_pcpu.c
(10.18 KB)
📄
subr_pctrie.c
(20.99 KB)
📄
subr_physmem.c
(11.52 KB)
📄
subr_pidctrl.c
(5.43 KB)
📄
subr_power.c
(3.13 KB)
📄
subr_prf.c
(27.42 KB)
📄
subr_prng.c
(3.36 KB)
📄
subr_prof.c
(15.43 KB)
📄
subr_rangeset.c
(8.5 KB)
📄
subr_rman.c
(27.61 KB)
📄
subr_rtc.c
(11.42 KB)
📄
subr_sbuf.c
(20.53 KB)
📄
subr_scanf.c
(15.59 KB)
📄
subr_sfbuf.c
(6.17 KB)
📄
subr_sglist.c
(22.83 KB)
📄
subr_sleepqueue.c
(39.43 KB)
📄
subr_smp.c
(31.62 KB)
📄
subr_smr.c
(20.17 KB)
📄
subr_stack.c
(6.47 KB)
📄
subr_stats.c
(103.01 KB)
📄
subr_syscall.c
(7.98 KB)
📄
subr_taskqueue.c
(21.1 KB)
📄
subr_terminal.c
(15.52 KB)
📄
subr_trap.c
(10.87 KB)
📄
subr_turnstile.c
(35.58 KB)
📄
subr_uio.c
(11.38 KB)
📄
subr_unit.c
(22.97 KB)
📄
subr_vmem.c
(43.25 KB)
📄
subr_witness.c
(84.59 KB)
📄
sys_capability.c
(15.06 KB)
📄
sys_eventfd.c
(8.42 KB)
📄
sys_generic.c
(44.22 KB)
📄
sys_getrandom.c
(4.21 KB)
📄
sys_pipe.c
(45.14 KB)
📄
sys_procdesc.c
(14.57 KB)
📄
sys_process.c
(30.73 KB)
📄
sys_socket.c
(20.11 KB)
📄
syscalls.c
(22.73 KB)
📄
syscalls.master
(60.26 KB)
📄
systrace_args.c
(178.49 KB)
📄
sysv_ipc.c
(6.53 KB)
📄
sysv_msg.c
(48.65 KB)
📄
sysv_sem.c
(49.85 KB)
📄
sysv_shm.c
(43.93 KB)
📄
tty.c
(55.14 KB)
📄
tty_compat.c
(11.46 KB)
📄
tty_info.c
(9.93 KB)
📄
tty_inq.c
(12.22 KB)
📄
tty_outq.c
(8.74 KB)
📄
tty_pts.c
(19.74 KB)
📄
tty_tty.c
(2.83 KB)
📄
tty_ttydisc.c
(28.6 KB)
📄
uipc_accf.c
(8.07 KB)
📄
uipc_debug.c
(12.42 KB)
📄
uipc_domain.c
(13.13 KB)
📄
uipc_ktls.c
(55.7 KB)
📄
uipc_mbuf.c
(52.45 KB)
📄
uipc_mbuf2.c
(12.64 KB)
📄
uipc_mbufhash.c
(4.9 KB)
📄
uipc_mqueue.c
(64.64 KB)
📄
uipc_sem.c
(25.18 KB)
📄
uipc_shm.c
(50.47 KB)
📄
uipc_sockbuf.c
(42.9 KB)
📄
uipc_socket.c
(110.61 KB)
📄
uipc_syscalls.c
(35.94 KB)
📄
uipc_usrreq.c
(75.11 KB)
📄
vfs_acl.c
(14.5 KB)
📄
vfs_aio.c
(76.32 KB)
📄
vfs_bio.c
(145.39 KB)
📄
vfs_cache.c
(143.09 KB)
📄
vfs_cluster.c
(28.36 KB)
📄
vfs_default.c
(33.16 KB)
📄
vfs_export.c
(14.55 KB)
📄
vfs_extattr.c
(17.91 KB)
📄
vfs_hash.c
(6 KB)
📄
vfs_init.c
(15.86 KB)
📄
vfs_lookup.c
(45.48 KB)
📄
vfs_mount.c
(62.58 KB)
📄
vfs_mountroot.c
(26.23 KB)
📄
vfs_subr.c
(167.52 KB)
📄
vfs_syscalls.c
(106.86 KB)
📄
vfs_vnops.c
(86.28 KB)
📄
vnode_if.src
(13.66 KB)
Editing: kern_rmlock.c
/*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007 Stephan Uphoff <ups@FreeBSD.org> * 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, 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. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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. */ /* * Machine independent bits of reader/writer lock implementation. */ #include <sys/cdefs.h> __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include <sys/param.h> #include <sys/systm.h> #include <sys/kernel.h> #include <sys/kdb.h> #include <sys/ktr.h> #include <sys/lock.h> #include <sys/mutex.h> #include <sys/proc.h> #include <sys/rmlock.h> #include <sys/sched.h> #include <sys/smp.h> #include <sys/turnstile.h> #include <sys/lock_profile.h> #include <machine/cpu.h> #include <vm/uma.h> #ifdef DDB #include <ddb/ddb.h> #endif /* * A cookie to mark destroyed rmlocks. This is stored in the head of * rm_activeReaders. */ #define RM_DESTROYED ((void *)0xdead) #define rm_destroyed(rm) \ (LIST_FIRST(&(rm)->rm_activeReaders) == RM_DESTROYED) #define RMPF_ONQUEUE 1 #define RMPF_SIGNAL 2 #ifndef INVARIANTS #define _rm_assert(c, what, file, line) #endif static void assert_rm(const struct lock_object *lock, int what); #ifdef DDB static void db_show_rm(const struct lock_object *lock); #endif static void lock_rm(struct lock_object *lock, uintptr_t how); #ifdef KDTRACE_HOOKS static int owner_rm(const struct lock_object *lock, struct thread **owner); #endif static uintptr_t unlock_rm(struct lock_object *lock); struct lock_class lock_class_rm = { .lc_name = "rm", .lc_flags = LC_SLEEPLOCK | LC_RECURSABLE, .lc_assert = assert_rm, #ifdef DDB .lc_ddb_show = db_show_rm, #endif .lc_lock = lock_rm, .lc_unlock = unlock_rm, #ifdef KDTRACE_HOOKS .lc_owner = owner_rm, #endif }; struct lock_class lock_class_rm_sleepable = { .lc_name = "sleepable rm", .lc_flags = LC_SLEEPLOCK | LC_SLEEPABLE | LC_RECURSABLE, .lc_assert = assert_rm, #ifdef DDB .lc_ddb_show = db_show_rm, #endif .lc_lock = lock_rm, .lc_unlock = unlock_rm, #ifdef KDTRACE_HOOKS .lc_owner = owner_rm, #endif }; static void assert_rm(const struct lock_object *lock, int what) { rm_assert((const struct rmlock *)lock, what); } static void lock_rm(struct lock_object *lock, uintptr_t how) { struct rmlock *rm; struct rm_priotracker *tracker; rm = (struct rmlock *)lock; if (how == 0) rm_wlock(rm); else { tracker = (struct rm_priotracker *)how; rm_rlock(rm, tracker); } } static uintptr_t unlock_rm(struct lock_object *lock) { struct thread *td; struct pcpu *pc; struct rmlock *rm; struct rm_queue *queue; struct rm_priotracker *tracker; uintptr_t how; rm = (struct rmlock *)lock; tracker = NULL; how = 0; rm_assert(rm, RA_LOCKED | RA_NOTRECURSED); if (rm_wowned(rm)) rm_wunlock(rm); else { /* * Find the right rm_priotracker structure for curthread. * The guarantee about its uniqueness is given by the fact * we already asserted the lock wasn't recursively acquired. */ critical_enter(); td = curthread; pc = get_pcpu(); for (queue = pc->pc_rm_queue.rmq_next; queue != &pc->pc_rm_queue; queue = queue->rmq_next) { tracker = (struct rm_priotracker *)queue; if ((tracker->rmp_rmlock == rm) && (tracker->rmp_thread == td)) { how = (uintptr_t)tracker; break; } } KASSERT(tracker != NULL, ("rm_priotracker is non-NULL when lock held in read mode")); critical_exit(); rm_runlock(rm, tracker); } return (how); } #ifdef KDTRACE_HOOKS static int owner_rm(const struct lock_object *lock, struct thread **owner) { const struct rmlock *rm; struct lock_class *lc; rm = (const struct rmlock *)lock; lc = LOCK_CLASS(&rm->rm_wlock_object); return (lc->lc_owner(&rm->rm_wlock_object, owner)); } #endif static struct mtx rm_spinlock; MTX_SYSINIT(rm_spinlock, &rm_spinlock, "rm_spinlock", MTX_SPIN); /* * Add or remove tracker from per-cpu list. * * The per-cpu list can be traversed at any time in forward direction from an * interrupt on the *local* cpu. */ static void inline rm_tracker_add(struct pcpu *pc, struct rm_priotracker *tracker) { struct rm_queue *next; /* Initialize all tracker pointers */ tracker->rmp_cpuQueue.rmq_prev = &pc->pc_rm_queue; next = pc->pc_rm_queue.rmq_next; tracker->rmp_cpuQueue.rmq_next = next; /* rmq_prev is not used during froward traversal. */ next->rmq_prev = &tracker->rmp_cpuQueue; /* Update pointer to first element. */ pc->pc_rm_queue.rmq_next = &tracker->rmp_cpuQueue; } /* * Return a count of the number of trackers the thread 'td' already * has on this CPU for the lock 'rm'. */ static int rm_trackers_present(const struct pcpu *pc, const struct rmlock *rm, const struct thread *td) { struct rm_queue *queue; struct rm_priotracker *tracker; int count; count = 0; for (queue = pc->pc_rm_queue.rmq_next; queue != &pc->pc_rm_queue; queue = queue->rmq_next) { tracker = (struct rm_priotracker *)queue; if ((tracker->rmp_rmlock == rm) && (tracker->rmp_thread == td)) count++; } return (count); } static void inline rm_tracker_remove(struct pcpu *pc, struct rm_priotracker *tracker) { struct rm_queue *next, *prev; next = tracker->rmp_cpuQueue.rmq_next; prev = tracker->rmp_cpuQueue.rmq_prev; /* Not used during forward traversal. */ next->rmq_prev = prev; /* Remove from list. */ prev->rmq_next = next; } static void rm_cleanIPI(void *arg) { struct pcpu *pc; struct rmlock *rm = arg; struct rm_priotracker *tracker; struct rm_queue *queue; pc = get_pcpu(); for (queue = pc->pc_rm_queue.rmq_next; queue != &pc->pc_rm_queue; queue = queue->rmq_next) { tracker = (struct rm_priotracker *)queue; if (tracker->rmp_rmlock == rm && tracker->rmp_flags == 0) { tracker->rmp_flags = RMPF_ONQUEUE; mtx_lock_spin(&rm_spinlock); LIST_INSERT_HEAD(&rm->rm_activeReaders, tracker, rmp_qentry); mtx_unlock_spin(&rm_spinlock); } } } void rm_init_flags(struct rmlock *rm, const char *name, int opts) { struct lock_class *lc; int liflags, xflags; liflags = 0; if (!(opts & RM_NOWITNESS)) liflags |= LO_WITNESS; if (opts & RM_RECURSE) liflags |= LO_RECURSABLE; if (opts & RM_NEW) liflags |= LO_NEW; rm->rm_writecpus = all_cpus; LIST_INIT(&rm->rm_activeReaders); if (opts & RM_SLEEPABLE) { liflags |= LO_SLEEPABLE; lc = &lock_class_rm_sleepable; xflags = (opts & RM_NEW ? SX_NEW : 0); sx_init_flags(&rm->rm_lock_sx, "rmlock_sx", xflags | SX_NOWITNESS); } else { lc = &lock_class_rm; xflags = (opts & RM_NEW ? MTX_NEW : 0); mtx_init(&rm->rm_lock_mtx, name, "rmlock_mtx", xflags | MTX_NOWITNESS); } lock_init(&rm->lock_object, lc, name, NULL, liflags); } void rm_init(struct rmlock *rm, const char *name) { rm_init_flags(rm, name, 0); } void rm_destroy(struct rmlock *rm) { rm_assert(rm, RA_UNLOCKED); LIST_FIRST(&rm->rm_activeReaders) = RM_DESTROYED; if (rm->lock_object.lo_flags & LO_SLEEPABLE) sx_destroy(&rm->rm_lock_sx); else mtx_destroy(&rm->rm_lock_mtx); lock_destroy(&rm->lock_object); } int rm_wowned(const struct rmlock *rm) { if (rm->lock_object.lo_flags & LO_SLEEPABLE) return (sx_xlocked(&rm->rm_lock_sx)); else return (mtx_owned(&rm->rm_lock_mtx)); } void rm_sysinit(void *arg) { struct rm_args *args; args = arg; rm_init_flags(args->ra_rm, args->ra_desc, args->ra_flags); } static __noinline int _rm_rlock_hard(struct rmlock *rm, struct rm_priotracker *tracker, int trylock) { struct pcpu *pc; critical_enter(); pc = get_pcpu(); /* Check if we just need to do a proper critical_exit. */ if (!CPU_ISSET(pc->pc_cpuid, &rm->rm_writecpus)) { critical_exit(); return (1); } /* Remove our tracker from the per-cpu list. */ rm_tracker_remove(pc, tracker); /* * Check to see if the IPI granted us the lock after all. The load of * rmp_flags must happen after the tracker is removed from the list. */ atomic_interrupt_fence(); if (tracker->rmp_flags) { /* Just add back tracker - we hold the lock. */ rm_tracker_add(pc, tracker); critical_exit(); return (1); } /* * We allow readers to acquire a lock even if a writer is blocked if * the lock is recursive and the reader already holds the lock. */ if ((rm->lock_object.lo_flags & LO_RECURSABLE) != 0) { /* * Just grant the lock if this thread already has a tracker * for this lock on the per-cpu queue. */ if (rm_trackers_present(pc, rm, curthread) != 0) { mtx_lock_spin(&rm_spinlock); LIST_INSERT_HEAD(&rm->rm_activeReaders, tracker, rmp_qentry); tracker->rmp_flags = RMPF_ONQUEUE; mtx_unlock_spin(&rm_spinlock); rm_tracker_add(pc, tracker); critical_exit(); return (1); } } sched_unpin(); critical_exit(); if (trylock) { if (rm->lock_object.lo_flags & LO_SLEEPABLE) { if (!sx_try_xlock(&rm->rm_lock_sx)) return (0); } else { if (!mtx_trylock(&rm->rm_lock_mtx)) return (0); } } else { if (rm->lock_object.lo_flags & LO_SLEEPABLE) { THREAD_SLEEPING_OK(); sx_xlock(&rm->rm_lock_sx); THREAD_NO_SLEEPING(); } else mtx_lock(&rm->rm_lock_mtx); } critical_enter(); pc = get_pcpu(); CPU_CLR(pc->pc_cpuid, &rm->rm_writecpus); rm_tracker_add(pc, tracker); sched_pin(); critical_exit(); if (rm->lock_object.lo_flags & LO_SLEEPABLE) sx_xunlock(&rm->rm_lock_sx); else mtx_unlock(&rm->rm_lock_mtx); return (1); } int _rm_rlock(struct rmlock *rm, struct rm_priotracker *tracker, int trylock) { struct thread *td = curthread; struct pcpu *pc; if (SCHEDULER_STOPPED()) return (1); tracker->rmp_flags = 0; tracker->rmp_thread = td; tracker->rmp_rmlock = rm; if (rm->lock_object.lo_flags & LO_SLEEPABLE) THREAD_NO_SLEEPING(); td->td_critnest++; /* critical_enter(); */ atomic_interrupt_fence(); pc = cpuid_to_pcpu[td->td_oncpu]; /* pcpu_find(td->td_oncpu); */ rm_tracker_add(pc, tracker); sched_pin(); atomic_interrupt_fence(); td->td_critnest--; /* * Fast path to combine two common conditions into a single * conditional jump. */ if (__predict_true(0 == (td->td_owepreempt | CPU_ISSET(pc->pc_cpuid, &rm->rm_writecpus)))) return (1); /* We do not have a read token and need to acquire one. */ return _rm_rlock_hard(rm, tracker, trylock); } static __noinline void _rm_unlock_hard(struct thread *td,struct rm_priotracker *tracker) { if (td->td_owepreempt) { td->td_critnest++; critical_exit(); } if (!tracker->rmp_flags) return; mtx_lock_spin(&rm_spinlock); LIST_REMOVE(tracker, rmp_qentry); if (tracker->rmp_flags & RMPF_SIGNAL) { struct rmlock *rm; struct turnstile *ts; rm = tracker->rmp_rmlock; turnstile_chain_lock(&rm->lock_object); mtx_unlock_spin(&rm_spinlock); ts = turnstile_lookup(&rm->lock_object); turnstile_signal(ts, TS_EXCLUSIVE_QUEUE); turnstile_unpend(ts); turnstile_chain_unlock(&rm->lock_object); } else mtx_unlock_spin(&rm_spinlock); } void _rm_runlock(struct rmlock *rm, struct rm_priotracker *tracker) { struct pcpu *pc; struct thread *td = tracker->rmp_thread; if (SCHEDULER_STOPPED()) return; td->td_critnest++; /* critical_enter(); */ pc = cpuid_to_pcpu[td->td_oncpu]; /* pcpu_find(td->td_oncpu); */ rm_tracker_remove(pc, tracker); td->td_critnest--; sched_unpin(); if (rm->lock_object.lo_flags & LO_SLEEPABLE) THREAD_SLEEPING_OK(); if (__predict_true(0 == (td->td_owepreempt | tracker->rmp_flags))) return; _rm_unlock_hard(td, tracker); } void _rm_wlock(struct rmlock *rm) { struct rm_priotracker *prio; struct turnstile *ts; cpuset_t readcpus; if (SCHEDULER_STOPPED()) return; if (rm->lock_object.lo_flags & LO_SLEEPABLE) sx_xlock(&rm->rm_lock_sx); else mtx_lock(&rm->rm_lock_mtx); if (CPU_CMP(&rm->rm_writecpus, &all_cpus)) { /* Get all read tokens back */ readcpus = all_cpus; CPU_ANDNOT(&readcpus, &rm->rm_writecpus); rm->rm_writecpus = all_cpus; /* * Assumes rm->rm_writecpus update is visible on other CPUs * before rm_cleanIPI is called. */ #ifdef SMP smp_rendezvous_cpus(readcpus, smp_no_rendezvous_barrier, rm_cleanIPI, smp_no_rendezvous_barrier, rm); #else rm_cleanIPI(rm); #endif mtx_lock_spin(&rm_spinlock); while ((prio = LIST_FIRST(&rm->rm_activeReaders)) != NULL) { ts = turnstile_trywait(&rm->lock_object); prio->rmp_flags = RMPF_ONQUEUE | RMPF_SIGNAL; mtx_unlock_spin(&rm_spinlock); turnstile_wait(ts, prio->rmp_thread, TS_EXCLUSIVE_QUEUE); mtx_lock_spin(&rm_spinlock); } mtx_unlock_spin(&rm_spinlock); } } void _rm_wunlock(struct rmlock *rm) { if (rm->lock_object.lo_flags & LO_SLEEPABLE) sx_xunlock(&rm->rm_lock_sx); else mtx_unlock(&rm->rm_lock_mtx); } #if LOCK_DEBUG > 0 void _rm_wlock_debug(struct rmlock *rm, const char *file, int line) { if (SCHEDULER_STOPPED()) return; KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread), ("rm_wlock() by idle thread %p on rmlock %s @ %s:%d", curthread, rm->lock_object.lo_name, file, line)); KASSERT(!rm_destroyed(rm), ("rm_wlock() of destroyed rmlock @ %s:%d", file, line)); _rm_assert(rm, RA_UNLOCKED, file, line); WITNESS_CHECKORDER(&rm->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL); _rm_wlock(rm); LOCK_LOG_LOCK("RMWLOCK", &rm->lock_object, 0, 0, file, line); WITNESS_LOCK(&rm->lock_object, LOP_EXCLUSIVE, file, line); TD_LOCKS_INC(curthread); } void _rm_wunlock_debug(struct rmlock *rm, const char *file, int line) { if (SCHEDULER_STOPPED()) return; KASSERT(!rm_destroyed(rm), ("rm_wunlock() of destroyed rmlock @ %s:%d", file, line)); _rm_assert(rm, RA_WLOCKED, file, line); WITNESS_UNLOCK(&rm->lock_object, LOP_EXCLUSIVE, file, line); LOCK_LOG_LOCK("RMWUNLOCK", &rm->lock_object, 0, 0, file, line); _rm_wunlock(rm); TD_LOCKS_DEC(curthread); } int _rm_rlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, int trylock, const char *file, int line) { if (SCHEDULER_STOPPED()) return (1); #ifdef INVARIANTS if (!(rm->lock_object.lo_flags & LO_RECURSABLE) && !trylock) { critical_enter(); KASSERT(rm_trackers_present(get_pcpu(), rm, curthread) == 0, ("rm_rlock: recursed on non-recursive rmlock %s @ %s:%d\n", rm->lock_object.lo_name, file, line)); critical_exit(); } #endif KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread), ("rm_rlock() by idle thread %p on rmlock %s @ %s:%d", curthread, rm->lock_object.lo_name, file, line)); KASSERT(!rm_destroyed(rm), ("rm_rlock() of destroyed rmlock @ %s:%d", file, line)); if (!trylock) { KASSERT(!rm_wowned(rm), ("rm_rlock: wlock already held for %s @ %s:%d", rm->lock_object.lo_name, file, line)); WITNESS_CHECKORDER(&rm->lock_object, LOP_NEWORDER | LOP_NOSLEEP, file, line, NULL); } if (_rm_rlock(rm, tracker, trylock)) { if (trylock) LOCK_LOG_TRY("RMRLOCK", &rm->lock_object, 0, 1, file, line); else LOCK_LOG_LOCK("RMRLOCK", &rm->lock_object, 0, 0, file, line); WITNESS_LOCK(&rm->lock_object, LOP_NOSLEEP, file, line); TD_LOCKS_INC(curthread); return (1); } else if (trylock) LOCK_LOG_TRY("RMRLOCK", &rm->lock_object, 0, 0, file, line); return (0); } void _rm_runlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, const char *file, int line) { if (SCHEDULER_STOPPED()) return; KASSERT(!rm_destroyed(rm), ("rm_runlock() of destroyed rmlock @ %s:%d", file, line)); _rm_assert(rm, RA_RLOCKED, file, line); WITNESS_UNLOCK(&rm->lock_object, 0, file, line); LOCK_LOG_LOCK("RMRUNLOCK", &rm->lock_object, 0, 0, file, line); _rm_runlock(rm, tracker); TD_LOCKS_DEC(curthread); } #else /* * Just strip out file and line arguments if no lock debugging is enabled in * the kernel - we are called from a kernel module. */ void _rm_wlock_debug(struct rmlock *rm, const char *file, int line) { _rm_wlock(rm); } void _rm_wunlock_debug(struct rmlock *rm, const char *file, int line) { _rm_wunlock(rm); } int _rm_rlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, int trylock, const char *file, int line) { return _rm_rlock(rm, tracker, trylock); } void _rm_runlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, const char *file, int line) { _rm_runlock(rm, tracker); } #endif #ifdef INVARIANT_SUPPORT #ifndef INVARIANTS #undef _rm_assert #endif /* * Note that this does not need to use witness_assert() for read lock * assertions since an exact count of read locks held by this thread * is computable. */ void _rm_assert(const struct rmlock *rm, int what, const char *file, int line) { int count; if (SCHEDULER_STOPPED()) return; switch (what) { case RA_LOCKED: case RA_LOCKED | RA_RECURSED: case RA_LOCKED | RA_NOTRECURSED: case RA_RLOCKED: case RA_RLOCKED | RA_RECURSED: case RA_RLOCKED | RA_NOTRECURSED: /* * Handle the write-locked case. Unlike other * primitives, writers can never recurse. */ if (rm_wowned(rm)) { if (what & RA_RLOCKED) panic("Lock %s exclusively locked @ %s:%d\n", rm->lock_object.lo_name, file, line); if (what & RA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", rm->lock_object.lo_name, file, line); break; } critical_enter(); count = rm_trackers_present(get_pcpu(), rm, curthread); critical_exit(); if (count == 0) panic("Lock %s not %slocked @ %s:%d\n", rm->lock_object.lo_name, (what & RA_RLOCKED) ? "read " : "", file, line); if (count > 1) { if (what & RA_NOTRECURSED) panic("Lock %s recursed @ %s:%d\n", rm->lock_object.lo_name, file, line); } else if (what & RA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", rm->lock_object.lo_name, file, line); break; case RA_WLOCKED: if (!rm_wowned(rm)) panic("Lock %s not exclusively locked @ %s:%d\n", rm->lock_object.lo_name, file, line); break; case RA_UNLOCKED: if (rm_wowned(rm)) panic("Lock %s exclusively locked @ %s:%d\n", rm->lock_object.lo_name, file, line); critical_enter(); count = rm_trackers_present(get_pcpu(), rm, curthread); critical_exit(); if (count != 0) panic("Lock %s read locked @ %s:%d\n", rm->lock_object.lo_name, file, line); break; default: panic("Unknown rm lock assertion: %d @ %s:%d", what, file, line); } } #endif /* INVARIANT_SUPPORT */ #ifdef DDB static void print_tracker(struct rm_priotracker *tr) { struct thread *td; td = tr->rmp_thread; db_printf(" thread %p (tid %d, pid %d, \"%s\") {", td, td->td_tid, td->td_proc->p_pid, td->td_name); if (tr->rmp_flags & RMPF_ONQUEUE) { db_printf("ONQUEUE"); if (tr->rmp_flags & RMPF_SIGNAL) db_printf(",SIGNAL"); } else db_printf("0"); db_printf("}\n"); } static void db_show_rm(const struct lock_object *lock) { struct rm_priotracker *tr; struct rm_queue *queue; const struct rmlock *rm; struct lock_class *lc; struct pcpu *pc; rm = (const struct rmlock *)lock; db_printf(" writecpus: "); ddb_display_cpuset(__DEQUALIFY(const cpuset_t *, &rm->rm_writecpus)); db_printf("\n"); db_printf(" per-CPU readers:\n"); STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) for (queue = pc->pc_rm_queue.rmq_next; queue != &pc->pc_rm_queue; queue = queue->rmq_next) { tr = (struct rm_priotracker *)queue; if (tr->rmp_rmlock == rm) print_tracker(tr); } db_printf(" active readers:\n"); LIST_FOREACH(tr, &rm->rm_activeReaders, rmp_qentry) print_tracker(tr); lc = LOCK_CLASS(&rm->rm_wlock_object); db_printf("Backing write-lock (%s):\n", lc->lc_name); lc->lc_ddb_show(&rm->rm_wlock_object); } #endif /* * Read-mostly sleepable locks. * * These primitives allow both readers and writers to sleep. However, neither * readers nor writers are tracked and subsequently there is no priority * propagation. * * They are intended to be only used when write-locking is almost never needed * (e.g., they can guard against unloading a kernel module) while read-locking * happens all the time. * * Concurrent writers take turns taking the lock while going off cpu. If this is * of concern for your usecase, this is not the right primitive. * * Neither rms_rlock nor rms_runlock use thread fences. Instead interrupt * fences are inserted to ensure ordering with the code executed in the IPI * handler. * * No attempt is made to track which CPUs read locked at least once, * consequently write locking sends IPIs to all of them. This will become a * problem at some point. The easiest way to lessen it is to provide a bitmap. */ #define RMS_NOOWNER ((void *)0x1) #define RMS_TRANSIENT ((void *)0x2) #define RMS_FLAGMASK 0xf struct rmslock_pcpu { int influx; int readers; }; _Static_assert(sizeof(struct rmslock_pcpu) == 8, "bad size"); /* * Internal routines */ static struct rmslock_pcpu * rms_int_pcpu(struct rmslock *rms) { CRITICAL_ASSERT(curthread); return (zpcpu_get(rms->pcpu)); } static struct rmslock_pcpu * rms_int_remote_pcpu(struct rmslock *rms, int cpu) { return (zpcpu_get_cpu(rms->pcpu, cpu)); } static void rms_int_influx_enter(struct rmslock *rms, struct rmslock_pcpu *pcpu) { CRITICAL_ASSERT(curthread); MPASS(pcpu->influx == 0); pcpu->influx = 1; } static void rms_int_influx_exit(struct rmslock *rms, struct rmslock_pcpu *pcpu) { CRITICAL_ASSERT(curthread); MPASS(pcpu->influx == 1); pcpu->influx = 0; } #ifdef INVARIANTS static void rms_int_debug_readers_inc(struct rmslock *rms) { int old; old = atomic_fetchadd_int(&rms->debug_readers, 1); KASSERT(old >= 0, ("%s: bad readers count %d\n", __func__, old)); } static void rms_int_debug_readers_dec(struct rmslock *rms) { int old; old = atomic_fetchadd_int(&rms->debug_readers, -1); KASSERT(old > 0, ("%s: bad readers count %d\n", __func__, old)); } #else static void rms_int_debug_readers_inc(struct rmslock *rms) { } static void rms_int_debug_readers_dec(struct rmslock *rms) { } #endif static void rms_int_readers_inc(struct rmslock *rms, struct rmslock_pcpu *pcpu) { CRITICAL_ASSERT(curthread); rms_int_debug_readers_inc(rms); pcpu->readers++; } static void rms_int_readers_dec(struct rmslock *rms, struct rmslock_pcpu *pcpu) { CRITICAL_ASSERT(curthread); rms_int_debug_readers_dec(rms); pcpu->readers--; } /* * Public API */ void rms_init(struct rmslock *rms, const char *name) { rms->owner = RMS_NOOWNER; rms->writers = 0; rms->readers = 0; rms->debug_readers = 0; mtx_init(&rms->mtx, name, NULL, MTX_DEF | MTX_NEW); rms->pcpu = uma_zalloc_pcpu(pcpu_zone_8, M_WAITOK | M_ZERO); } void rms_destroy(struct rmslock *rms) { MPASS(rms->writers == 0); MPASS(rms->readers == 0); mtx_destroy(&rms->mtx); uma_zfree_pcpu(pcpu_zone_8, rms->pcpu); } static void __noinline rms_rlock_fallback(struct rmslock *rms) { rms_int_influx_exit(rms, rms_int_pcpu(rms)); critical_exit(); mtx_lock(&rms->mtx); while (rms->writers > 0) msleep(&rms->readers, &rms->mtx, PUSER - 1, mtx_name(&rms->mtx), 0); critical_enter(); rms_int_readers_inc(rms, rms_int_pcpu(rms)); mtx_unlock(&rms->mtx); critical_exit(); } void rms_rlock(struct rmslock *rms) { struct rmslock_pcpu *pcpu; WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__); MPASS(atomic_load_ptr(&rms->owner) != curthread); critical_enter(); pcpu = rms_int_pcpu(rms); rms_int_influx_enter(rms, pcpu); atomic_interrupt_fence(); if (__predict_false(rms->writers > 0)) { rms_rlock_fallback(rms); return; } atomic_interrupt_fence(); rms_int_readers_inc(rms, pcpu); atomic_interrupt_fence(); rms_int_influx_exit(rms, pcpu); critical_exit(); } int rms_try_rlock(struct rmslock *rms) { struct rmslock_pcpu *pcpu; MPASS(atomic_load_ptr(&rms->owner) != curthread); critical_enter(); pcpu = rms_int_pcpu(rms); rms_int_influx_enter(rms, pcpu); atomic_interrupt_fence(); if (__predict_false(rms->writers > 0)) { rms_int_influx_exit(rms, pcpu); critical_exit(); return (0); } atomic_interrupt_fence(); rms_int_readers_inc(rms, pcpu); atomic_interrupt_fence(); rms_int_influx_exit(rms, pcpu); critical_exit(); return (1); } static void __noinline rms_runlock_fallback(struct rmslock *rms) { rms_int_influx_exit(rms, rms_int_pcpu(rms)); critical_exit(); mtx_lock(&rms->mtx); MPASS(rms->writers > 0); MPASS(rms->readers > 0); MPASS(rms->debug_readers == rms->readers); rms_int_debug_readers_dec(rms); rms->readers--; if (rms->readers == 0) wakeup_one(&rms->writers); mtx_unlock(&rms->mtx); } void rms_runlock(struct rmslock *rms) { struct rmslock_pcpu *pcpu; critical_enter(); pcpu = rms_int_pcpu(rms); rms_int_influx_enter(rms, pcpu); atomic_interrupt_fence(); if (__predict_false(rms->writers > 0)) { rms_runlock_fallback(rms); return; } atomic_interrupt_fence(); rms_int_readers_dec(rms, pcpu); atomic_interrupt_fence(); rms_int_influx_exit(rms, pcpu); critical_exit(); } struct rmslock_ipi { struct rmslock *rms; struct smp_rendezvous_cpus_retry_arg srcra; }; static void rms_action_func(void *arg) { struct rmslock_ipi *rmsipi; struct rmslock_pcpu *pcpu; struct rmslock *rms; rmsipi = __containerof(arg, struct rmslock_ipi, srcra); rms = rmsipi->rms; pcpu = rms_int_pcpu(rms); if (pcpu->influx) return; if (pcpu->readers != 0) { atomic_add_int(&rms->readers, pcpu->readers); pcpu->readers = 0; } smp_rendezvous_cpus_done(arg); } static void rms_wait_func(void *arg, int cpu) { struct rmslock_ipi *rmsipi; struct rmslock_pcpu *pcpu; struct rmslock *rms; rmsipi = __containerof(arg, struct rmslock_ipi, srcra); rms = rmsipi->rms; pcpu = rms_int_remote_pcpu(rms, cpu); while (atomic_load_int(&pcpu->influx)) cpu_spinwait(); } #ifdef INVARIANTS static void rms_assert_no_pcpu_readers(struct rmslock *rms) { struct rmslock_pcpu *pcpu; int cpu; CPU_FOREACH(cpu) { pcpu = rms_int_remote_pcpu(rms, cpu); if (pcpu->readers != 0) { panic("%s: got %d readers on cpu %d\n", __func__, pcpu->readers, cpu); } } } #else static void rms_assert_no_pcpu_readers(struct rmslock *rms) { } #endif static void rms_wlock_switch(struct rmslock *rms) { struct rmslock_ipi rmsipi; MPASS(rms->readers == 0); MPASS(rms->writers == 1); rmsipi.rms = rms; smp_rendezvous_cpus_retry(all_cpus, smp_no_rendezvous_barrier, rms_action_func, smp_no_rendezvous_barrier, rms_wait_func, &rmsipi.srcra); } void rms_wlock(struct rmslock *rms) { WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__); MPASS(atomic_load_ptr(&rms->owner) != curthread); mtx_lock(&rms->mtx); rms->writers++; if (rms->writers > 1) { msleep(&rms->owner, &rms->mtx, (PUSER - 1), mtx_name(&rms->mtx), 0); MPASS(rms->readers == 0); KASSERT(rms->owner == RMS_TRANSIENT, ("%s: unexpected owner value %p\n", __func__, rms->owner)); goto out_grab; } KASSERT(rms->owner == RMS_NOOWNER, ("%s: unexpected owner value %p\n", __func__, rms->owner)); rms_wlock_switch(rms); rms_assert_no_pcpu_readers(rms); if (rms->readers > 0) { msleep(&rms->writers, &rms->mtx, (PUSER - 1), mtx_name(&rms->mtx), 0); } out_grab: rms->owner = curthread; rms_assert_no_pcpu_readers(rms); mtx_unlock(&rms->mtx); MPASS(rms->readers == 0); } void rms_wunlock(struct rmslock *rms) { mtx_lock(&rms->mtx); KASSERT(rms->owner == curthread, ("%s: unexpected owner value %p\n", __func__, rms->owner)); MPASS(rms->writers >= 1); MPASS(rms->readers == 0); rms->writers--; if (rms->writers > 0) { wakeup_one(&rms->owner); rms->owner = RMS_TRANSIENT; } else { wakeup(&rms->readers); rms->owner = RMS_NOOWNER; } mtx_unlock(&rms->mtx); } void rms_unlock(struct rmslock *rms) { if (rms_wowned(rms)) rms_wunlock(rms); else rms_runlock(rms); }
Upload File
Create Folder