freebsd-dev/sys/kern/subr_sleepqueue.c

1370 lines
36 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2004 John Baldwin <jhb@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.
*
* 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.
*/
/*
* Implementation of sleep queues used to hold queue of threads blocked on
* a wait channel. Sleep queues different from turnstiles in that wait
* channels are not owned by anyone, so there is no priority propagation.
* Sleep queues can also provide a timeout and can also be interrupted by
* signals. That said, there are several similarities between the turnstile
* and sleep queue implementations. (Note: turnstiles were implemented
* first.) For example, both use a hash table of the same size where each
* bucket is referred to as a "chain" that contains both a spin lock and
* a linked list of queues. An individual queue is located by using a hash
* to pick a chain, locking the chain, and then walking the chain searching
* for the queue. This means that a wait channel object does not need to
* embed it's queue head just as locks do not embed their turnstile queue
* head. Threads also carry around a sleep queue that they lend to the
* wait channel when blocking. Just as in turnstiles, the queue includes
* a free list of the sleep queues of other threads blocked on the same
* wait channel in the case of multiple waiters.
*
* Some additional functionality provided by sleep queues include the
* ability to set a timeout. The timeout is managed using a per-thread
* callout that resumes a thread if it is asleep. A thread may also
* catch signals while it is asleep (aka an interruptible sleep). The
* signal code uses sleepq_abort() to interrupt a sleeping thread. Finally,
* sleep queues also provide some extra assertions. One is not allowed to
* mix the sleep/wakeup and cv APIs for a given wait channel. Also, one
* must consistently use the same lock to synchronize with a wait channel,
* though this check is currently only a warning for sleep/wakeup due to
* pre-existing abuse of that API. The same lock must also be held when
* awakening threads, though that is currently only enforced for condition
* variables.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_sleepqueue_profiling.h"
#include "opt_ddb.h"
2007-06-12 23:27:31 +00:00
#include "opt_sched.h"
#include "opt_stack.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/lock.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/sbuf.h>
#include <sys/sched.h>
#include <sys/sdt.h>
#include <sys/signalvar.h>
#include <sys/sleepqueue.h>
#include <sys/stack.h>
#include <sys/sysctl.h>
#include <vm/uma.h>
#ifdef DDB
#include <ddb/ddb.h>
#endif
/*
* Constants for the hash table of sleep queue chains.
* SC_TABLESIZE must be a power of two for SC_MASK to work properly.
*/
#define SC_TABLESIZE 256 /* Must be power of 2. */
#define SC_MASK (SC_TABLESIZE - 1)
#define SC_SHIFT 8
#define SC_HASH(wc) ((((uintptr_t)(wc) >> SC_SHIFT) ^ (uintptr_t)(wc)) & \
SC_MASK)
#define SC_LOOKUP(wc) &sleepq_chains[SC_HASH(wc)]
#define NR_SLEEPQS 2
/*
* There two different lists of sleep queues. Both lists are connected
* via the sq_hash entries. The first list is the sleep queue chain list
* that a sleep queue is on when it is attached to a wait channel. The
* second list is the free list hung off of a sleep queue that is attached
* to a wait channel.
*
* Each sleep queue also contains the wait channel it is attached to, the
* list of threads blocked on that wait channel, flags specific to the
* wait channel, and the lock used to synchronize with a wait channel.
* The flags are used to catch mismatches between the various consumers
* of the sleep queue API (e.g. sleep/wakeup and condition variables).
* The lock pointer is only used when invariants are enabled for various
* debugging checks.
*
* Locking key:
* c - sleep queue chain lock
*/
struct sleepqueue {
TAILQ_HEAD(, thread) sq_blocked[NR_SLEEPQS]; /* (c) Blocked threads. */
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
u_int sq_blockedcnt[NR_SLEEPQS]; /* (c) N. of blocked threads. */
LIST_ENTRY(sleepqueue) sq_hash; /* (c) Chain and free list. */
LIST_HEAD(, sleepqueue) sq_free; /* (c) Free queues. */
void *sq_wchan; /* (c) Wait channel. */
int sq_type; /* (c) Queue type. */
#ifdef INVARIANTS
struct lock_object *sq_lock; /* (c) Associated lock. */
#endif
};
struct sleepqueue_chain {
LIST_HEAD(, sleepqueue) sc_queues; /* List of sleep queues. */
struct mtx sc_lock; /* Spin lock for this chain. */
#ifdef SLEEPQUEUE_PROFILING
u_int sc_depth; /* Length of sc_queues. */
u_int sc_max_depth; /* Max length of sc_queues. */
#endif
};
#ifdef SLEEPQUEUE_PROFILING
u_int sleepq_max_depth;
static SYSCTL_NODE(_debug, OID_AUTO, sleepq, CTLFLAG_RD, 0, "sleepq profiling");
static SYSCTL_NODE(_debug_sleepq, OID_AUTO, chains, CTLFLAG_RD, 0,
"sleepq chain stats");
SYSCTL_UINT(_debug_sleepq, OID_AUTO, max_depth, CTLFLAG_RD, &sleepq_max_depth,
0, "maxmimum depth achieved of a single chain");
static void sleepq_profile(const char *wmesg);
static int prof_enabled;
#endif
static struct sleepqueue_chain sleepq_chains[SC_TABLESIZE];
static uma_zone_t sleepq_zone;
/*
* Prototypes for non-exported routines.
*/
static int sleepq_catch_signals(void *wchan, int pri);
static int sleepq_check_signals(void);
static int sleepq_check_timeout(void);
#ifdef INVARIANTS
static void sleepq_dtor(void *mem, int size, void *arg);
#endif
static int sleepq_init(void *mem, int size, int flags);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
static int sleepq_resume_thread(struct sleepqueue *sq, struct thread *td,
int pri);
static void sleepq_switch(void *wchan, int pri);
static void sleepq_timeout(void *arg);
SDT_PROBE_DECLARE(sched, , , sleep);
SDT_PROBE_DECLARE(sched, , , wakeup);
/*
* Initialize SLEEPQUEUE_PROFILING specific sysctl nodes.
* Note that it must happen after sleepinit() has been fully executed, so
* it must happen after SI_SUB_KMEM SYSINIT() subsystem setup.
*/
#ifdef SLEEPQUEUE_PROFILING
static void
init_sleepqueue_profiling(void)
{
char chain_name[10];
struct sysctl_oid *chain_oid;
u_int i;
for (i = 0; i < SC_TABLESIZE; i++) {
snprintf(chain_name, sizeof(chain_name), "%u", i);
chain_oid = SYSCTL_ADD_NODE(NULL,
SYSCTL_STATIC_CHILDREN(_debug_sleepq_chains), OID_AUTO,
chain_name, CTLFLAG_RD, NULL, "sleepq chain stats");
SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
"depth", CTLFLAG_RD, &sleepq_chains[i].sc_depth, 0, NULL);
SYSCTL_ADD_UINT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
"max_depth", CTLFLAG_RD, &sleepq_chains[i].sc_max_depth, 0,
NULL);
}
}
SYSINIT(sleepqueue_profiling, SI_SUB_LOCK, SI_ORDER_ANY,
init_sleepqueue_profiling, NULL);
#endif
/*
* Early initialization of sleep queues that is called from the sleepinit()
* SYSINIT.
*/
void
init_sleepqueues(void)
{
int i;
for (i = 0; i < SC_TABLESIZE; i++) {
LIST_INIT(&sleepq_chains[i].sc_queues);
mtx_init(&sleepq_chains[i].sc_lock, "sleepq chain", NULL,
MTX_SPIN | MTX_RECURSE);
}
sleepq_zone = uma_zcreate("SLEEPQUEUE", sizeof(struct sleepqueue),
#ifdef INVARIANTS
NULL, sleepq_dtor, sleepq_init, NULL, UMA_ALIGN_CACHE, 0);
#else
NULL, NULL, sleepq_init, NULL, UMA_ALIGN_CACHE, 0);
#endif
thread0.td_sleepqueue = sleepq_alloc();
}
/*
* Get a sleep queue for a new thread.
*/
struct sleepqueue *
sleepq_alloc(void)
{
return (uma_zalloc(sleepq_zone, M_WAITOK));
}
/*
* Free a sleep queue when a thread is destroyed.
*/
void
sleepq_free(struct sleepqueue *sq)
{
uma_zfree(sleepq_zone, sq);
}
/*
* Lock the sleep queue chain associated with the specified wait channel.
*/
void
sleepq_lock(void *wchan)
{
struct sleepqueue_chain *sc;
sc = SC_LOOKUP(wchan);
mtx_lock_spin(&sc->sc_lock);
}
/*
* Look up the sleep queue associated with a given wait channel in the hash
* table locking the associated sleep queue chain. If no queue is found in
* the table, NULL is returned.
*/
struct sleepqueue *
sleepq_lookup(void *wchan)
{
struct sleepqueue_chain *sc;
struct sleepqueue *sq;
KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
sc = SC_LOOKUP(wchan);
mtx_assert(&sc->sc_lock, MA_OWNED);
LIST_FOREACH(sq, &sc->sc_queues, sq_hash)
if (sq->sq_wchan == wchan)
return (sq);
return (NULL);
}
/*
* Unlock the sleep queue chain associated with a given wait channel.
*/
void
sleepq_release(void *wchan)
{
struct sleepqueue_chain *sc;
sc = SC_LOOKUP(wchan);
mtx_unlock_spin(&sc->sc_lock);
}
/*
* Places the current thread on the sleep queue for the specified wait
* channel. If INVARIANTS is enabled, then it associates the passed in
* lock with the sleepq to make sure it is held when that sleep queue is
* woken up.
*/
void
sleepq_add(void *wchan, struct lock_object *lock, const char *wmesg, int flags,
int queue)
{
struct sleepqueue_chain *sc;
struct sleepqueue *sq;
struct thread *td;
td = curthread;
sc = SC_LOOKUP(wchan);
mtx_assert(&sc->sc_lock, MA_OWNED);
MPASS(td->td_sleepqueue != NULL);
MPASS(wchan != NULL);
MPASS((queue >= 0) && (queue < NR_SLEEPQS));
/* If this thread is not allowed to sleep, die a horrible death. */
KASSERT(td->td_no_sleeping == 0,
("%s: td %p to sleep on wchan %p with sleeping prohibited",
__func__, td, wchan));
/* Look up the sleep queue associated with the wait channel 'wchan'. */
sq = sleepq_lookup(wchan);
/*
* If the wait channel does not already have a sleep queue, use
* this thread's sleep queue. Otherwise, insert the current thread
* into the sleep queue already in use by this wait channel.
*/
if (sq == NULL) {
#ifdef INVARIANTS
int i;
sq = td->td_sleepqueue;
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
for (i = 0; i < NR_SLEEPQS; i++) {
KASSERT(TAILQ_EMPTY(&sq->sq_blocked[i]),
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
("thread's sleep queue %d is not empty", i));
KASSERT(sq->sq_blockedcnt[i] == 0,
("thread's sleep queue %d count mismatches", i));
}
KASSERT(LIST_EMPTY(&sq->sq_free),
("thread's sleep queue has a non-empty free list"));
KASSERT(sq->sq_wchan == NULL, ("stale sq_wchan pointer"));
sq->sq_lock = lock;
#endif
#ifdef SLEEPQUEUE_PROFILING
sc->sc_depth++;
if (sc->sc_depth > sc->sc_max_depth) {
sc->sc_max_depth = sc->sc_depth;
if (sc->sc_max_depth > sleepq_max_depth)
sleepq_max_depth = sc->sc_max_depth;
}
#endif
sq = td->td_sleepqueue;
LIST_INSERT_HEAD(&sc->sc_queues, sq, sq_hash);
sq->sq_wchan = wchan;
sq->sq_type = flags & SLEEPQ_TYPE;
} else {
MPASS(wchan == sq->sq_wchan);
MPASS(lock == sq->sq_lock);
MPASS((flags & SLEEPQ_TYPE) == sq->sq_type);
LIST_INSERT_HEAD(&sq->sq_free, td->td_sleepqueue, sq_hash);
}
thread_lock(td);
TAILQ_INSERT_TAIL(&sq->sq_blocked[queue], td, td_slpq);
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
sq->sq_blockedcnt[queue]++;
td->td_sleepqueue = NULL;
td->td_sqqueue = queue;
td->td_wchan = wchan;
td->td_wmesg = wmesg;
if (flags & SLEEPQ_INTERRUPTIBLE) {
Now that the return value semantics of cv's for multithreaded processes have been unified with that of msleep(9), further refine the sleepq interface and consolidate some duplicated code: - Move the pre-sleep checks for theaded processes into a thread_sleep_check() function in kern_thread.c. - Move all handling of TDF_SINTR to be internal to subr_sleepqueue.c. Specifically, if a thread is awakened by something other than a signal while checking for signals before going to sleep, clear TDF_SINTR in sleepq_catch_signals(). This removes a sched_lock lock/unlock combo in that edge case during an interruptible sleep. Also, fix sleepq_check_signals() to properly handle the condition if TDF_SINTR is clear rather than requiring the callers of the sleepq API to notice this edge case and call a non-_sig variant of sleepq_wait(). - Clarify the flags arguments to sleepq_add(), sleepq_signal() and sleepq_broadcast() by creating an explicit submask for sleepq types. Also, add an explicit SLEEPQ_MSLEEP type rather than a magic number of 0. Also, add a SLEEPQ_INTERRUPTIBLE flag for use with sleepq_add() and move the setting of TDF_SINTR to sleepq_add() if this flag is set rather than sleepq_catch_signals(). Note that it is the caller's responsibility to ensure that sleepq_catch_signals() is called if and only if this flag is passed to the preceeding sleepq_add(). Note that this also removes a sched_lock lock/unlock pair from sleepq_catch_signals(). It also ensures that for an interruptible sleep, TDF_SINTR is always set when TD_ON_SLEEPQ() is true.
2004-08-19 11:31:42 +00:00
td->td_flags |= TDF_SINTR;
td->td_flags &= ~TDF_SLEEPABORT;
}
thread_unlock(td);
}
/*
* Sets a timeout that will remove the current thread from the specified
* sleep queue after timo ticks if the thread has not already been awakened.
*/
void
sleepq_set_timeout_sbt(void *wchan, sbintime_t sbt, sbintime_t pr,
int flags)
{
struct sleepqueue_chain *sc;
struct thread *td;
sbintime_t pr1;
td = curthread;
sc = SC_LOOKUP(wchan);
mtx_assert(&sc->sc_lock, MA_OWNED);
MPASS(TD_ON_SLEEPQ(td));
MPASS(td->td_sleepqueue == NULL);
MPASS(wchan != NULL);
if (cold && td == &thread0)
panic("timed sleep before timers are working");
KASSERT(td->td_sleeptimo == 0, ("td %d %p td_sleeptimo %jx",
td->td_tid, td, (uintmax_t)td->td_sleeptimo));
thread_lock(td);
callout_when(sbt, pr, flags, &td->td_sleeptimo, &pr1);
thread_unlock(td);
callout_reset_sbt_on(&td->td_slpcallout, td->td_sleeptimo, pr1,
sleepq_timeout, td, PCPU_GET(cpuid), flags | C_PRECALC |
C_DIRECT_EXEC);
}
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
/*
* Return the number of actual sleepers for the specified queue.
*/
u_int
sleepq_sleepcnt(void *wchan, int queue)
{
struct sleepqueue *sq;
KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
MPASS((queue >= 0) && (queue < NR_SLEEPQS));
sq = sleepq_lookup(wchan);
if (sq == NULL)
return (0);
return (sq->sq_blockedcnt[queue]);
}
/*
* Marks the pending sleep of the current thread as interruptible and
* makes an initial check for pending signals before putting a thread
* to sleep. Enters and exits with the thread lock held. Thread lock
* may have transitioned from the sleepq lock to a run lock.
*/
static int
sleepq_catch_signals(void *wchan, int pri)
{
struct sleepqueue_chain *sc;
struct sleepqueue *sq;
struct thread *td;
struct proc *p;
struct sigacts *ps;
int sig, ret;
td = curthread;
p = curproc;
sc = SC_LOOKUP(wchan);
mtx_assert(&sc->sc_lock, MA_OWNED);
MPASS(wchan != NULL);
if ((td->td_pflags & TDP_WAKEUP) != 0) {
td->td_pflags &= ~TDP_WAKEUP;
ret = EINTR;
2010-08-20 23:51:34 +00:00
thread_lock(td);
goto out;
}
/*
* See if there are any pending signals for this thread. If not
* we can switch immediately. Otherwise do the signal processing
* directly.
*/
thread_lock(td);
if ((td->td_flags & (TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK)) == 0) {
sleepq_switch(wchan, pri);
return (0);
}
thread_unlock(td);
mtx_unlock_spin(&sc->sc_lock);
CTR3(KTR_PROC, "sleepq catching signals: thread %p (pid %ld, %s)",
(void *)td, (long)p->p_pid, td->td_name);
PROC_LOCK(p);
ps = p->p_sigacts;
mtx_lock(&ps->ps_mtx);
sig = cursig(td);
if (sig == -1) {
mtx_unlock(&ps->ps_mtx);
KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
KASSERT(TD_SBDRY_INTR(td),
("lost TDF_SERESTART of TDF_SEINTR"));
KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
(TDF_SEINTR | TDF_SERESTART),
("both TDF_SEINTR and TDF_SERESTART"));
ret = TD_SBDRY_ERRNO(td);
} else if (sig == 0) {
mtx_unlock(&ps->ps_mtx);
ret = thread_suspend_check(1);
MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
} else {
if (SIGISMEMBER(ps->ps_sigintr, sig))
ret = EINTR;
else
ret = ERESTART;
mtx_unlock(&ps->ps_mtx);
}
/*
* Lock the per-process spinlock prior to dropping the PROC_LOCK
* to avoid a signal delivery race. PROC_LOCK, PROC_SLOCK, and
* thread_lock() are currently held in tdsendsignal().
*/
PROC_SLOCK(p);
mtx_lock_spin(&sc->sc_lock);
PROC_UNLOCK(p);
thread_lock(td);
PROC_SUNLOCK(p);
if (ret == 0) {
sleepq_switch(wchan, pri);
return (0);
}
out:
2006-02-23 03:42:17 +00:00
/*
* There were pending signals and this thread is still
* on the sleep queue, remove it from the sleep queue.
*/
if (TD_ON_SLEEPQ(td)) {
sq = sleepq_lookup(wchan);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
if (sleepq_resume_thread(sq, td, 0)) {
#ifdef INVARIANTS
/*
* This thread hasn't gone to sleep yet, so it
* should not be swapped out.
*/
panic("not waking up swapper");
#endif
}
}
mtx_unlock_spin(&sc->sc_lock);
MPASS(td->td_lock != &sc->sc_lock);
return (ret);
}
/*
* Switches to another thread if we are still asleep on a sleep queue.
* Returns with thread lock.
*/
static void
sleepq_switch(void *wchan, int pri)
{
struct sleepqueue_chain *sc;
struct sleepqueue *sq;
struct thread *td;
td = curthread;
sc = SC_LOOKUP(wchan);
mtx_assert(&sc->sc_lock, MA_OWNED);
THREAD_LOCK_ASSERT(td, MA_OWNED);
/*
* If we have a sleep queue, then we've already been woken up, so
* just return.
*/
if (td->td_sleepqueue != NULL) {
mtx_unlock_spin(&sc->sc_lock);
return;
}
/*
* If TDF_TIMEOUT is set, then our sleep has been timed out
* already but we are still on the sleep queue, so dequeue the
* thread and return.
*/
if (td->td_flags & TDF_TIMEOUT) {
MPASS(TD_ON_SLEEPQ(td));
sq = sleepq_lookup(wchan);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
if (sleepq_resume_thread(sq, td, 0)) {
#ifdef INVARIANTS
/*
* This thread hasn't gone to sleep yet, so it
* should not be swapped out.
*/
panic("not waking up swapper");
#endif
}
mtx_unlock_spin(&sc->sc_lock);
return;
}
#ifdef SLEEPQUEUE_PROFILING
if (prof_enabled)
sleepq_profile(td->td_wmesg);
#endif
MPASS(td->td_sleepqueue == NULL);
sched_sleep(td, pri);
thread_lock_set(td, &sc->sc_lock);
SDT_PROBE0(sched, , , sleep);
TD_SET_SLEEPING(td);
mi_switch(SW_VOL | SWT_SLEEPQ, NULL);
KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING"));
CTR3(KTR_PROC, "sleepq resume: thread %p (pid %ld, %s)",
(void *)td, (long)td->td_proc->p_pid, (void *)td->td_name);
}
/*
* Check to see if we timed out.
*/
static int
sleepq_check_timeout(void)
{
struct thread *td;
int res;
td = curthread;
THREAD_LOCK_ASSERT(td, MA_OWNED);
/*
* If TDF_TIMEOUT is set, we timed out. But recheck
* td_sleeptimo anyway.
*/
res = 0;
if (td->td_sleeptimo != 0) {
if (td->td_sleeptimo <= sbinuptime())
res = EWOULDBLOCK;
td->td_sleeptimo = 0;
}
if (td->td_flags & TDF_TIMEOUT)
td->td_flags &= ~TDF_TIMEOUT;
else
/*
* We ignore the situation where timeout subsystem was
* unable to stop our callout. The struct thread is
* type-stable, the callout will use the correct
* memory when running. The checks of the
* td_sleeptimo value in this function and in
* sleepq_timeout() ensure that the thread does not
* get spurious wakeups, even if the callout was reset
* or thread reused.
*/
callout_stop(&td->td_slpcallout);
return (res);
Major callout subsystem cleanup and rewrite: - Close a migration race where callout_reset() failed to set the CALLOUT_ACTIVE flag. - Callout callback functions are now allowed to be protected by spinlocks. - Switching the callout CPU number cannot always be done on a per-callout basis. See the updated timeout(9) manual page for more information. - The timeout(9) manual page has been updated to reflect how all the functions inside the callout API are working. The manual page has been made function oriented to make it easier to deduce how each of the functions making up the callout API are working without having to first read the whole manual page. Group all functions into a handful of sections which should give a quick top-level overview when the different functions should be used. - The CALLOUT_SHAREDLOCK flag and its functionality has been removed to reduce the complexity in the callout code and to avoid problems about atomically stopping callouts via callout_stop(). If someone needs it, it can be re-added. From my quick grep there are no CALLOUT_SHAREDLOCK clients in the kernel. - A new callout API function named "callout_drain_async()" has been added. See the updated timeout(9) manual page for a complete description. - Update the callout clients in the "kern/" folder to use the callout API properly, like cv_timedwait(). Previously there was some custom sleepqueue code in the callout subsystem, which has been removed, because we now allow callouts to be protected by spinlocks. This allows us to tear down the callout like done with regular mutexes, and a "td_slpmutex" has been added to "struct thread" to atomically teardown the "td_slpcallout". Further the "TDF_TIMOFAIL" and "SWT_SLEEPQTIMO" states can now be completely removed. Currently they are marked as available and will be cleaned up in a follow up commit. - Bump the __FreeBSD_version to indicate kernel modules need recompilation. - There has been several reports that this patch "seems to squash a serious bug leading to a callout timeout and panic". Kernel build testing: all architectures were built MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D1438 Sponsored by: Mellanox Technologies Reviewed by: jhb, adrian, sbruno and emaste
2015-01-15 15:32:30 +00:00
}
/*
* Check to see if we were awoken by a signal.
*/
static int
sleepq_check_signals(void)
{
struct thread *td;
td = curthread;
THREAD_LOCK_ASSERT(td, MA_OWNED);
/* We are no longer in an interruptible sleep. */
if (td->td_flags & TDF_SINTR)
td->td_flags &= ~TDF_SINTR;
if (td->td_flags & TDF_SLEEPABORT) {
td->td_flags &= ~TDF_SLEEPABORT;
return (td->td_intrval);
}
return (0);
}
/*
* Block the current thread until it is awakened from its sleep queue.
*/
void
sleepq_wait(void *wchan, int pri)
{
struct thread *td;
td = curthread;
MPASS(!(td->td_flags & TDF_SINTR));
thread_lock(td);
sleepq_switch(wchan, pri);
thread_unlock(td);
}
/*
* Block the current thread until it is awakened from its sleep queue
* or it is interrupted by a signal.
*/
int
sleepq_wait_sig(void *wchan, int pri)
{
int rcatch;
int rval;
rcatch = sleepq_catch_signals(wchan, pri);
rval = sleepq_check_signals();
thread_unlock(curthread);
if (rcatch)
return (rcatch);
return (rval);
}
/*
* Block the current thread until it is awakened from its sleep queue
* or it times out while waiting.
*/
int
sleepq_timedwait(void *wchan, int pri)
{
struct thread *td;
int rval;
td = curthread;
MPASS(!(td->td_flags & TDF_SINTR));
thread_lock(td);
sleepq_switch(wchan, pri);
rval = sleepq_check_timeout();
thread_unlock(td);
return (rval);
}
/*
* Block the current thread until it is awakened from its sleep queue,
* it is interrupted by a signal, or it times out waiting to be awakened.
*/
int
sleepq_timedwait_sig(void *wchan, int pri)
{
int rcatch, rvalt, rvals;
rcatch = sleepq_catch_signals(wchan, pri);
rvalt = sleepq_check_timeout();
rvals = sleepq_check_signals();
thread_unlock(curthread);
if (rcatch)
return (rcatch);
if (rvals)
return (rvals);
return (rvalt);
}
/*
* Returns the type of sleepqueue given a waitchannel.
*/
int
sleepq_type(void *wchan)
{
struct sleepqueue *sq;
int type;
MPASS(wchan != NULL);
sleepq_lock(wchan);
sq = sleepq_lookup(wchan);
if (sq == NULL) {
sleepq_release(wchan);
return (-1);
}
type = sq->sq_type;
sleepq_release(wchan);
return (type);
}
/*
* Removes a thread from a sleep queue and makes it
* runnable.
*/
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
static int
sleepq_resume_thread(struct sleepqueue *sq, struct thread *td, int pri)
{
struct sleepqueue_chain *sc;
MPASS(td != NULL);
MPASS(sq->sq_wchan != NULL);
MPASS(td->td_wchan == sq->sq_wchan);
MPASS(td->td_sqqueue < NR_SLEEPQS && td->td_sqqueue >= 0);
THREAD_LOCK_ASSERT(td, MA_OWNED);
sc = SC_LOOKUP(sq->sq_wchan);
mtx_assert(&sc->sc_lock, MA_OWNED);
SDT_PROBE2(sched, , , wakeup, td, td->td_proc);
/* Remove the thread from the queue. */
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
sq->sq_blockedcnt[td->td_sqqueue]--;
TAILQ_REMOVE(&sq->sq_blocked[td->td_sqqueue], td, td_slpq);
/*
* Get a sleep queue for this thread. If this is the last waiter,
* use the queue itself and take it out of the chain, otherwise,
* remove a queue from the free list.
*/
if (LIST_EMPTY(&sq->sq_free)) {
td->td_sleepqueue = sq;
#ifdef INVARIANTS
sq->sq_wchan = NULL;
#endif
#ifdef SLEEPQUEUE_PROFILING
sc->sc_depth--;
#endif
} else
td->td_sleepqueue = LIST_FIRST(&sq->sq_free);
LIST_REMOVE(td->td_sleepqueue, sq_hash);
td->td_wmesg = NULL;
td->td_wchan = NULL;
td->td_flags &= ~TDF_SINTR;
CTR3(KTR_PROC, "sleepq_wakeup: thread %p (pid %ld, %s)",
(void *)td, (long)td->td_proc->p_pid, td->td_name);
/* Adjust priority if requested. */
MPASS(pri == 0 || (pri >= PRI_MIN && pri <= PRI_MAX));
if (pri != 0 && td->td_priority > pri &&
PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
sched_prio(td, pri);
/*
* Note that thread td might not be sleeping if it is running
* sleepq_catch_signals() on another CPU or is blocked on its
* proc lock to check signals. There's no need to mark the
* thread runnable in that case.
*/
if (TD_IS_SLEEPING(td)) {
TD_CLR_SLEEPING(td);
return (setrunnable(td));
}
return (0);
}
#ifdef INVARIANTS
/*
* UMA zone item deallocator.
*/
static void
sleepq_dtor(void *mem, int size, void *arg)
{
struct sleepqueue *sq;
int i;
sq = mem;
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
for (i = 0; i < NR_SLEEPQS; i++) {
MPASS(TAILQ_EMPTY(&sq->sq_blocked[i]));
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
MPASS(sq->sq_blockedcnt[i] == 0);
}
}
#endif
/*
* UMA zone item initializer.
*/
static int
sleepq_init(void *mem, int size, int flags)
{
struct sleepqueue *sq;
int i;
bzero(mem, size);
sq = mem;
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
for (i = 0; i < NR_SLEEPQS; i++) {
TAILQ_INIT(&sq->sq_blocked[i]);
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
sq->sq_blockedcnt[i] = 0;
}
LIST_INIT(&sq->sq_free);
return (0);
}
/*
* Find the highest priority thread sleeping on a wait channel and resume it.
*/
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
int
sleepq_signal(void *wchan, int flags, int pri, int queue)
{
struct sleepqueue *sq;
struct thread *td, *besttd;
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
int wakeup_swapper;
CTR2(KTR_PROC, "sleepq_signal(%p, %d)", wchan, flags);
KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
MPASS((queue >= 0) && (queue < NR_SLEEPQS));
sq = sleepq_lookup(wchan);
if (sq == NULL)
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (0);
Now that the return value semantics of cv's for multithreaded processes have been unified with that of msleep(9), further refine the sleepq interface and consolidate some duplicated code: - Move the pre-sleep checks for theaded processes into a thread_sleep_check() function in kern_thread.c. - Move all handling of TDF_SINTR to be internal to subr_sleepqueue.c. Specifically, if a thread is awakened by something other than a signal while checking for signals before going to sleep, clear TDF_SINTR in sleepq_catch_signals(). This removes a sched_lock lock/unlock combo in that edge case during an interruptible sleep. Also, fix sleepq_check_signals() to properly handle the condition if TDF_SINTR is clear rather than requiring the callers of the sleepq API to notice this edge case and call a non-_sig variant of sleepq_wait(). - Clarify the flags arguments to sleepq_add(), sleepq_signal() and sleepq_broadcast() by creating an explicit submask for sleepq types. Also, add an explicit SLEEPQ_MSLEEP type rather than a magic number of 0. Also, add a SLEEPQ_INTERRUPTIBLE flag for use with sleepq_add() and move the setting of TDF_SINTR to sleepq_add() if this flag is set rather than sleepq_catch_signals(). Note that it is the caller's responsibility to ensure that sleepq_catch_signals() is called if and only if this flag is passed to the preceeding sleepq_add(). Note that this also removes a sched_lock lock/unlock pair from sleepq_catch_signals(). It also ensures that for an interruptible sleep, TDF_SINTR is always set when TD_ON_SLEEPQ() is true.
2004-08-19 11:31:42 +00:00
KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE),
("%s: mismatch between sleep/wakeup and cv_*", __func__));
/*
* Find the highest priority thread on the queue. If there is a
* tie, use the thread that first appears in the queue as it has
* been sleeping the longest since threads are always added to
* the tail of sleep queues.
*/
besttd = TAILQ_FIRST(&sq->sq_blocked[queue]);
TAILQ_FOREACH(td, &sq->sq_blocked[queue], td_slpq) {
if (td->td_priority < besttd->td_priority)
besttd = td;
}
MPASS(besttd != NULL);
thread_lock(besttd);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
wakeup_swapper = sleepq_resume_thread(sq, besttd, pri);
thread_unlock(besttd);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (wakeup_swapper);
}
/*
* Resume all threads sleeping on a specified wait channel.
*/
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
int
sleepq_broadcast(void *wchan, int flags, int pri, int queue)
{
struct sleepqueue *sq;
struct thread *td, *tdn;
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
int wakeup_swapper;
CTR2(KTR_PROC, "sleepq_broadcast(%p, %d)", wchan, flags);
KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
MPASS((queue >= 0) && (queue < NR_SLEEPQS));
sq = sleepq_lookup(wchan);
if (sq == NULL)
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (0);
Now that the return value semantics of cv's for multithreaded processes have been unified with that of msleep(9), further refine the sleepq interface and consolidate some duplicated code: - Move the pre-sleep checks for theaded processes into a thread_sleep_check() function in kern_thread.c. - Move all handling of TDF_SINTR to be internal to subr_sleepqueue.c. Specifically, if a thread is awakened by something other than a signal while checking for signals before going to sleep, clear TDF_SINTR in sleepq_catch_signals(). This removes a sched_lock lock/unlock combo in that edge case during an interruptible sleep. Also, fix sleepq_check_signals() to properly handle the condition if TDF_SINTR is clear rather than requiring the callers of the sleepq API to notice this edge case and call a non-_sig variant of sleepq_wait(). - Clarify the flags arguments to sleepq_add(), sleepq_signal() and sleepq_broadcast() by creating an explicit submask for sleepq types. Also, add an explicit SLEEPQ_MSLEEP type rather than a magic number of 0. Also, add a SLEEPQ_INTERRUPTIBLE flag for use with sleepq_add() and move the setting of TDF_SINTR to sleepq_add() if this flag is set rather than sleepq_catch_signals(). Note that it is the caller's responsibility to ensure that sleepq_catch_signals() is called if and only if this flag is passed to the preceeding sleepq_add(). Note that this also removes a sched_lock lock/unlock pair from sleepq_catch_signals(). It also ensures that for an interruptible sleep, TDF_SINTR is always set when TD_ON_SLEEPQ() is true.
2004-08-19 11:31:42 +00:00
KASSERT(sq->sq_type == (flags & SLEEPQ_TYPE),
("%s: mismatch between sleep/wakeup and cv_*", __func__));
/*
* Resume all blocked threads on the sleep queue. The last thread will
* be given ownership of sq and may re-enqueue itself before
* sleepq_resume_thread() returns, so we must cache the "next" queue
* item at the beginning of the final iteration.
*/
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
wakeup_swapper = 0;
TAILQ_FOREACH_SAFE(td, &sq->sq_blocked[queue], td_slpq, tdn) {
thread_lock(td);
wakeup_swapper |= sleepq_resume_thread(sq, td, pri);
thread_unlock(td);
}
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (wakeup_swapper);
}
/*
* Time sleeping threads out. When the timeout expires, the thread is
* removed from the sleep queue and made runnable if it is still asleep.
*/
static void
sleepq_timeout(void *arg)
{
struct sleepqueue_chain *sc;
struct sleepqueue *sq;
struct thread *td;
void *wchan;
int wakeup_swapper;
td = arg;
wakeup_swapper = 0;
CTR3(KTR_PROC, "sleepq_timeout: thread %p (pid %ld, %s)",
(void *)td, (long)td->td_proc->p_pid, (void *)td->td_name);
Major callout subsystem cleanup and rewrite: - Close a migration race where callout_reset() failed to set the CALLOUT_ACTIVE flag. - Callout callback functions are now allowed to be protected by spinlocks. - Switching the callout CPU number cannot always be done on a per-callout basis. See the updated timeout(9) manual page for more information. - The timeout(9) manual page has been updated to reflect how all the functions inside the callout API are working. The manual page has been made function oriented to make it easier to deduce how each of the functions making up the callout API are working without having to first read the whole manual page. Group all functions into a handful of sections which should give a quick top-level overview when the different functions should be used. - The CALLOUT_SHAREDLOCK flag and its functionality has been removed to reduce the complexity in the callout code and to avoid problems about atomically stopping callouts via callout_stop(). If someone needs it, it can be re-added. From my quick grep there are no CALLOUT_SHAREDLOCK clients in the kernel. - A new callout API function named "callout_drain_async()" has been added. See the updated timeout(9) manual page for a complete description. - Update the callout clients in the "kern/" folder to use the callout API properly, like cv_timedwait(). Previously there was some custom sleepqueue code in the callout subsystem, which has been removed, because we now allow callouts to be protected by spinlocks. This allows us to tear down the callout like done with regular mutexes, and a "td_slpmutex" has been added to "struct thread" to atomically teardown the "td_slpcallout". Further the "TDF_TIMOFAIL" and "SWT_SLEEPQTIMO" states can now be completely removed. Currently they are marked as available and will be cleaned up in a follow up commit. - Bump the __FreeBSD_version to indicate kernel modules need recompilation. - There has been several reports that this patch "seems to squash a serious bug leading to a callout timeout and panic". Kernel build testing: all architectures were built MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D1438 Sponsored by: Mellanox Technologies Reviewed by: jhb, adrian, sbruno and emaste
2015-01-15 15:32:30 +00:00
thread_lock(td);
if (td->td_sleeptimo > sbinuptime() || td->td_sleeptimo == 0) {
/*
* The thread does not want a timeout (yet).
*/
} else if (TD_IS_SLEEPING(td) && TD_ON_SLEEPQ(td)) {
/*
* See if the thread is asleep and get the wait
* channel if it is.
*/
wchan = td->td_wchan;
sc = SC_LOOKUP(wchan);
THREAD_LOCKPTR_ASSERT(td, &sc->sc_lock);
sq = sleepq_lookup(wchan);
MPASS(sq != NULL);
td->td_flags |= TDF_TIMEOUT;
wakeup_swapper = sleepq_resume_thread(sq, td, 0);
} else if (TD_ON_SLEEPQ(td)) {
/*
* If the thread is on the SLEEPQ but isn't sleeping
* yet, it can either be on another CPU in between
* sleepq_add() and one of the sleepq_*wait*()
* routines or it can be in sleepq_catch_signals().
*/
td->td_flags |= TDF_TIMEOUT;
Major callout subsystem cleanup and rewrite: - Close a migration race where callout_reset() failed to set the CALLOUT_ACTIVE flag. - Callout callback functions are now allowed to be protected by spinlocks. - Switching the callout CPU number cannot always be done on a per-callout basis. See the updated timeout(9) manual page for more information. - The timeout(9) manual page has been updated to reflect how all the functions inside the callout API are working. The manual page has been made function oriented to make it easier to deduce how each of the functions making up the callout API are working without having to first read the whole manual page. Group all functions into a handful of sections which should give a quick top-level overview when the different functions should be used. - The CALLOUT_SHAREDLOCK flag and its functionality has been removed to reduce the complexity in the callout code and to avoid problems about atomically stopping callouts via callout_stop(). If someone needs it, it can be re-added. From my quick grep there are no CALLOUT_SHAREDLOCK clients in the kernel. - A new callout API function named "callout_drain_async()" has been added. See the updated timeout(9) manual page for a complete description. - Update the callout clients in the "kern/" folder to use the callout API properly, like cv_timedwait(). Previously there was some custom sleepqueue code in the callout subsystem, which has been removed, because we now allow callouts to be protected by spinlocks. This allows us to tear down the callout like done with regular mutexes, and a "td_slpmutex" has been added to "struct thread" to atomically teardown the "td_slpcallout". Further the "TDF_TIMOFAIL" and "SWT_SLEEPQTIMO" states can now be completely removed. Currently they are marked as available and will be cleaned up in a follow up commit. - Bump the __FreeBSD_version to indicate kernel modules need recompilation. - There has been several reports that this patch "seems to squash a serious bug leading to a callout timeout and panic". Kernel build testing: all architectures were built MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D1438 Sponsored by: Mellanox Technologies Reviewed by: jhb, adrian, sbruno and emaste
2015-01-15 15:32:30 +00:00
}
thread_unlock(td);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
if (wakeup_swapper)
kick_proc0();
}
/*
* Resumes a specific thread from the sleep queue associated with a specific
* wait channel if it is on that queue.
*/
void
sleepq_remove(struct thread *td, void *wchan)
{
struct sleepqueue *sq;
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
int wakeup_swapper;
/*
* Look up the sleep queue for this wait channel, then re-check
* that the thread is asleep on that channel, if it is not, then
* bail.
*/
MPASS(wchan != NULL);
sleepq_lock(wchan);
sq = sleepq_lookup(wchan);
/*
* We can not lock the thread here as it may be sleeping on a
* different sleepq. However, holding the sleepq lock for this
* wchan can guarantee that we do not miss a wakeup for this
* channel. The asserts below will catch any false positives.
*/
if (!TD_ON_SLEEPQ(td) || td->td_wchan != wchan) {
sleepq_release(wchan);
return;
}
/* Thread is asleep on sleep queue sq, so wake it up. */
thread_lock(td);
MPASS(sq != NULL);
MPASS(td->td_wchan == wchan);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
wakeup_swapper = sleepq_resume_thread(sq, td, 0);
thread_unlock(td);
sleepq_release(wchan);
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
if (wakeup_swapper)
kick_proc0();
}
/*
* Abort a thread as if an interrupt had occurred. Only abort
* interruptible waits (unfortunately it isn't safe to abort others).
*/
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
int
sleepq_abort(struct thread *td, int intrval)
{
struct sleepqueue *sq;
void *wchan;
THREAD_LOCK_ASSERT(td, MA_OWNED);
MPASS(TD_ON_SLEEPQ(td));
MPASS(td->td_flags & TDF_SINTR);
MPASS(intrval == EINTR || intrval == ERESTART);
/*
* If the TDF_TIMEOUT flag is set, just leave. A
* timeout is scheduled anyhow.
*/
if (td->td_flags & TDF_TIMEOUT)
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (0);
CTR3(KTR_PROC, "sleepq_abort: thread %p (pid %ld, %s)",
(void *)td, (long)td->td_proc->p_pid, (void *)td->td_name);
td->td_intrval = intrval;
td->td_flags |= TDF_SLEEPABORT;
/*
* If the thread has not slept yet it will find the signal in
* sleepq_catch_signals() and call sleepq_resume_thread. Otherwise
* we have to do it here.
*/
if (!TD_IS_SLEEPING(td))
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (0);
wchan = td->td_wchan;
MPASS(wchan != NULL);
sq = sleepq_lookup(wchan);
MPASS(sq != NULL);
/* Thread is asleep on sleep queue sq, so wake it up. */
If a thread that is swapped out is made runnable, then the setrunnable() routine wakes up proc0 so that proc0 can swap the thread back in. Historically, this has been done by waking up proc0 directly from setrunnable() itself via a wakeup(). When waking up a sleeping thread that was swapped out (the usual case when waking proc0 since only sleeping threads are eligible to be swapped out), this resulted in a bit of recursion (e.g. wakeup() -> setrunnable() -> wakeup()). With sleep queues having separate locks in 6.x and later, this caused a spin lock LOR (sleepq lock -> sched_lock/thread lock -> sleepq lock). An attempt was made to fix this in 7.0 by making the proc0 wakeup use the ithread mechanism for doing the wakeup. However, this required grabbing proc0's thread lock to perform the wakeup. If proc0 was asleep elsewhere in the kernel (e.g. waiting for disk I/O), then this degenerated into the same LOR since the thread lock would be some other sleepq lock. Fix this by deferring the wakeup of the swapper until after the sleepq lock held by the upper layer has been locked. The setrunnable() routine now returns a boolean value to indicate whether or not proc0 needs to be woken up. The end result is that consumers of the sleepq API such as *sleep/wakeup, condition variables, sx locks, and lockmgr, have to wakeup proc0 if they get a non-zero return value from sleepq_abort(), sleepq_broadcast(), or sleepq_signal(). Discussed with: jeff Glanced at by: sam Tested by: Jurgen Weber jurgen - ish com au MFC after: 2 weeks
2008-08-05 20:02:31 +00:00
return (sleepq_resume_thread(sq, td, 0));
}
/*
* Prints the stacks of all threads presently sleeping on wchan/queue to
* the sbuf sb. Sets count_stacks_printed to the number of stacks actually
* printed. Typically, this will equal the number of threads sleeping on the
* queue, but may be less if sb overflowed before all stacks were printed.
*/
#ifdef STACK
int
sleepq_sbuf_print_stacks(struct sbuf *sb, void *wchan, int queue,
int *count_stacks_printed)
{
struct thread *td, *td_next;
struct sleepqueue *sq;
struct stack **st;
struct sbuf **td_infos;
int i, stack_idx, error, stacks_to_allocate;
bool finished, partial_print;
error = 0;
finished = false;
partial_print = false;
KASSERT(wchan != NULL, ("%s: invalid NULL wait channel", __func__));
MPASS((queue >= 0) && (queue < NR_SLEEPQS));
stacks_to_allocate = 10;
for (i = 0; i < 3 && !finished ; i++) {
/* We cannot malloc while holding the queue's spinlock, so
* we do our mallocs now, and hope it is enough. If it
* isn't, we will free these, drop the lock, malloc more,
* and try again, up to a point. After that point we will
* give up and report ENOMEM. We also cannot write to sb
* during this time since the client may have set the
* SBUF_AUTOEXTEND flag on their sbuf, which could cause a
* malloc as we print to it. So we defer actually printing
* to sb until after we drop the spinlock.
*/
/* Where we will store the stacks. */
st = malloc(sizeof(struct stack *) * stacks_to_allocate,
M_TEMP, M_WAITOK);
for (stack_idx = 0; stack_idx < stacks_to_allocate;
stack_idx++)
st[stack_idx] = stack_create();
/* Where we will store the td name, tid, etc. */
td_infos = malloc(sizeof(struct sbuf *) * stacks_to_allocate,
M_TEMP, M_WAITOK);
for (stack_idx = 0; stack_idx < stacks_to_allocate;
stack_idx++)
td_infos[stack_idx] = sbuf_new(NULL, NULL,
MAXCOMLEN + sizeof(struct thread *) * 2 + 40,
SBUF_FIXEDLEN);
sleepq_lock(wchan);
sq = sleepq_lookup(wchan);
if (sq == NULL) {
/* This sleepq does not exist; exit and return ENOENT. */
error = ENOENT;
finished = true;
sleepq_release(wchan);
goto loop_end;
}
stack_idx = 0;
/* Save thread info */
TAILQ_FOREACH_SAFE(td, &sq->sq_blocked[queue], td_slpq,
td_next) {
if (stack_idx >= stacks_to_allocate)
goto loop_end;
/* Note the td_lock is equal to the sleepq_lock here. */
stack_save_td(st[stack_idx], td);
sbuf_printf(td_infos[stack_idx], "%d: %s %p",
td->td_tid, td->td_name, td);
++stack_idx;
}
finished = true;
sleepq_release(wchan);
/* Print the stacks */
for (i = 0; i < stack_idx; i++) {
sbuf_finish(td_infos[i]);
sbuf_printf(sb, "--- thread %s: ---\n", sbuf_data(td_infos[i]));
stack_sbuf_print(sb, st[i]);
sbuf_printf(sb, "\n");
error = sbuf_error(sb);
if (error == 0)
*count_stacks_printed = stack_idx;
}
loop_end:
if (!finished)
sleepq_release(wchan);
for (stack_idx = 0; stack_idx < stacks_to_allocate;
stack_idx++)
stack_destroy(st[stack_idx]);
for (stack_idx = 0; stack_idx < stacks_to_allocate;
stack_idx++)
sbuf_delete(td_infos[stack_idx]);
free(st, M_TEMP);
free(td_infos, M_TEMP);
stacks_to_allocate *= 10;
}
if (!finished && error == 0)
error = ENOMEM;
return (error);
}
#endif
#ifdef SLEEPQUEUE_PROFILING
#define SLEEPQ_PROF_LOCATIONS 1024
#define SLEEPQ_SBUFSIZE 512
struct sleepq_prof {
LIST_ENTRY(sleepq_prof) sp_link;
const char *sp_wmesg;
long sp_count;
};
LIST_HEAD(sqphead, sleepq_prof);
struct sqphead sleepq_prof_free;
struct sqphead sleepq_hash[SC_TABLESIZE];
static struct sleepq_prof sleepq_profent[SLEEPQ_PROF_LOCATIONS];
static struct mtx sleepq_prof_lock;
MTX_SYSINIT(sleepq_prof_lock, &sleepq_prof_lock, "sleepq_prof", MTX_SPIN);
static void
sleepq_profile(const char *wmesg)
{
struct sleepq_prof *sp;
mtx_lock_spin(&sleepq_prof_lock);
if (prof_enabled == 0)
goto unlock;
LIST_FOREACH(sp, &sleepq_hash[SC_HASH(wmesg)], sp_link)
if (sp->sp_wmesg == wmesg)
goto done;
sp = LIST_FIRST(&sleepq_prof_free);
if (sp == NULL)
goto unlock;
sp->sp_wmesg = wmesg;
LIST_REMOVE(sp, sp_link);
LIST_INSERT_HEAD(&sleepq_hash[SC_HASH(wmesg)], sp, sp_link);
done:
sp->sp_count++;
unlock:
mtx_unlock_spin(&sleepq_prof_lock);
return;
}
static void
sleepq_prof_reset(void)
{
struct sleepq_prof *sp;
int enabled;
int i;
mtx_lock_spin(&sleepq_prof_lock);
enabled = prof_enabled;
prof_enabled = 0;
for (i = 0; i < SC_TABLESIZE; i++)
LIST_INIT(&sleepq_hash[i]);
LIST_INIT(&sleepq_prof_free);
for (i = 0; i < SLEEPQ_PROF_LOCATIONS; i++) {
sp = &sleepq_profent[i];
sp->sp_wmesg = NULL;
sp->sp_count = 0;
LIST_INSERT_HEAD(&sleepq_prof_free, sp, sp_link);
}
prof_enabled = enabled;
mtx_unlock_spin(&sleepq_prof_lock);
}
static int
enable_sleepq_prof(SYSCTL_HANDLER_ARGS)
{
int error, v;
v = prof_enabled;
error = sysctl_handle_int(oidp, &v, v, req);
if (error)
return (error);
if (req->newptr == NULL)
return (error);
if (v == prof_enabled)
return (0);
if (v == 1)
sleepq_prof_reset();
mtx_lock_spin(&sleepq_prof_lock);
prof_enabled = !!v;
mtx_unlock_spin(&sleepq_prof_lock);
return (0);
}
static int
reset_sleepq_prof_stats(SYSCTL_HANDLER_ARGS)
{
int error, v;
v = 0;
error = sysctl_handle_int(oidp, &v, 0, req);
if (error)
return (error);
if (req->newptr == NULL)
return (error);
if (v == 0)
return (0);
sleepq_prof_reset();
return (0);
}
static int
dump_sleepq_prof_stats(SYSCTL_HANDLER_ARGS)
{
struct sleepq_prof *sp;
struct sbuf *sb;
int enabled;
int error;
int i;
error = sysctl_wire_old_buffer(req, 0);
if (error != 0)
return (error);
sb = sbuf_new_for_sysctl(NULL, NULL, SLEEPQ_SBUFSIZE, req);
sbuf_printf(sb, "\nwmesg\tcount\n");
enabled = prof_enabled;
mtx_lock_spin(&sleepq_prof_lock);
prof_enabled = 0;
mtx_unlock_spin(&sleepq_prof_lock);
for (i = 0; i < SC_TABLESIZE; i++) {
LIST_FOREACH(sp, &sleepq_hash[i], sp_link) {
sbuf_printf(sb, "%s\t%ld\n",
sp->sp_wmesg, sp->sp_count);
}
}
mtx_lock_spin(&sleepq_prof_lock);
prof_enabled = enabled;
mtx_unlock_spin(&sleepq_prof_lock);
error = sbuf_finish(sb);
sbuf_delete(sb);
return (error);
}
SYSCTL_PROC(_debug_sleepq, OID_AUTO, stats, CTLTYPE_STRING | CTLFLAG_RD,
NULL, 0, dump_sleepq_prof_stats, "A", "Sleepqueue profiling statistics");
SYSCTL_PROC(_debug_sleepq, OID_AUTO, reset, CTLTYPE_INT | CTLFLAG_RW,
NULL, 0, reset_sleepq_prof_stats, "I",
"Reset sleepqueue profiling statistics");
SYSCTL_PROC(_debug_sleepq, OID_AUTO, enable, CTLTYPE_INT | CTLFLAG_RW,
NULL, 0, enable_sleepq_prof, "I", "Enable sleepqueue profiling");
#endif
#ifdef DDB
DB_SHOW_COMMAND(sleepq, db_show_sleepqueue)
{
struct sleepqueue_chain *sc;
struct sleepqueue *sq;
#ifdef INVARIANTS
struct lock_object *lock;
#endif
struct thread *td;
void *wchan;
int i;
if (!have_addr)
return;
/*
* First, see if there is an active sleep queue for the wait channel
* indicated by the address.
*/
wchan = (void *)addr;
sc = SC_LOOKUP(wchan);
LIST_FOREACH(sq, &sc->sc_queues, sq_hash)
if (sq->sq_wchan == wchan)
goto found;
/*
* Second, see if there is an active sleep queue at the address
* indicated.
*/
for (i = 0; i < SC_TABLESIZE; i++)
LIST_FOREACH(sq, &sleepq_chains[i].sc_queues, sq_hash) {
if (sq == (struct sleepqueue *)addr)
goto found;
}
db_printf("Unable to locate a sleep queue via %p\n", (void *)addr);
return;
found:
db_printf("Wait channel: %p\n", sq->sq_wchan);
db_printf("Queue type: %d\n", sq->sq_type);
#ifdef INVARIANTS
if (sq->sq_lock) {
lock = sq->sq_lock;
db_printf("Associated Interlock: %p - (%s) %s\n", lock,
LOCK_CLASS(lock)->lc_name, lock->lo_name);
}
#endif
db_printf("Blocked threads:\n");
for (i = 0; i < NR_SLEEPQS; i++) {
db_printf("\nQueue[%d]:\n", i);
if (TAILQ_EMPTY(&sq->sq_blocked[i]))
db_printf("\tempty\n");
else
TAILQ_FOREACH(td, &sq->sq_blocked[0],
td_slpq) {
db_printf("\t%p (tid %d, pid %d, \"%s\")\n", td,
td->td_tid, td->td_proc->p_pid,
2008-07-28 18:33:43 +00:00
td->td_name);
}
In current code, threads performing an interruptible sleep (on both sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will leave the waiters flag on forcing the owner to do a wakeup even when if the waiter queue is empty. That operation may lead to a deadlock in the case of doing a fake wakeup on the "preferred" (based on the wakeup algorithm) queue while the other queue has real waiters on it, because nobody is going to wakeup the 2nd queue waiters and they will sleep indefinitively. A similar bug, is present, for lockmgr in the case the waiters are sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue is not empty, the waiters won't progress after being awake but they will just fail, still not taking care of the 2nd queue waiters (as instead the lock owned doing the wakeup would expect). In order to fix this bug in a cheap way (without adding too much locking and complicating too much the semantic) add a sleepqueue interface which does report the actual number of waiters on a specified queue of a waitchannel (sleepq_sleepcnt()) and use it in order to determine if the exclusive waiters (or shared waiters) are actually present on the lockmgr (or sx) before to give them precedence in the wakeup algorithm. This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters a lockmgr has and if all the waiters on the exclusive waiters queue are LK_SLEEPFAIL just wake both queues. The sleepq_sleepcnt() introduction and ABI breakage require __FreeBSD_version bumping. Reported by: avg, kib, pho Reviewed by: kib Tested by: pho
2009-12-12 21:31:07 +00:00
db_printf("(expected: %u)\n", sq->sq_blockedcnt[i]);
}
}
/* Alias 'show sleepqueue' to 'show sleepq'. */
DB_SHOW_ALIAS(sleepqueue, db_show_sleepqueue);
#endif