/*
* Like cv_wait_sig_swap but allows the caller to indicate (with a
* non-NULL sigret) that they will take care of signalling the cv
* after wakeup, if necessary. This is a vile hack that should only
* be used when no other option is available; almost all callers
* should just use cv_wait_sig_swap (which takes care of the cv_signal
* stuff automatically) instead.
*/
int
cv_wait_sig_swap_core(kcondvar_t *cvp, kmutex_t *mp, int *sigret)
{
kthread_t *t = curthread;
proc_t *p = ttoproc(t);
klwp_t *lwp = ttolwp(t);
int rval = 1;
int signalled = 0;
if (panicstr)
return (rval);
/*
* The check for t_intr is to catch an interrupt thread
* that has not yet unpinned the thread underneath.
*/
if (lwp == NULL || t->t_intr) {
cv_wait(cvp, mp);
return (rval);
}
lwp->lwp_asleep = 1;
lwp->lwp_sysabort = 0;
thread_lock(t);
t->t_kpri_req = 0; /* don't need kernel priority */
cv_block_sig(t, (condvar_impl_t *)cvp);
/* I can be swapped now */
curthread->t_schedflag &= ~TS_DONT_SWAP;
thread_unlock_nopreempt(t);
mutex_exit(mp);
if (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t))
setrun(t);
/* ASSERT(no locks are held) */
swtch();
signalled = (t->t_schedflag & TS_SIGNALLED);
t->t_flag &= ~T_WAKEABLE;
/* TS_DONT_SWAP set by disp() */
ASSERT(curthread->t_schedflag & TS_DONT_SWAP);
mutex_enter(mp);
if (ISSIG_PENDING(t, lwp, p)) {
mutex_exit(mp);
if (issig(FORREAL))
rval = 0;
mutex_enter(mp);
}
if (lwp->lwp_sysabort || MUSTRETURN(p, t))
rval = 0;
lwp->lwp_asleep = 0;
lwp->lwp_sysabort = 0;
if (rval == 0) {
if (sigret != NULL)
*sigret = signalled; /* just tell the caller */
else if (signalled)
cv_signal(cvp); /* avoid consuming the cv_signal() */
}
return (rval);
}
int
cv_wait_sig(kcondvar_t *cvp, kmutex_t *mp)
{
kthread_t *t = curthread;
proc_t *p = ttoproc(t);
klwp_t *lwp = ttolwp(t);
int rval = 1;
int signalled = 0;
if (panicstr)
return (rval);
/*
* The check for t_intr is to catch an interrupt thread
* that has not yet unpinned the thread underneath.
*/
if (lwp == NULL || t->t_intr) {
cv_wait(cvp, mp);
return (rval);
}
ASSERT(curthread->t_schedflag & TS_DONT_SWAP);
lwp->lwp_asleep = 1;
lwp->lwp_sysabort = 0;
thread_lock(t);
cv_block_sig(t, (condvar_impl_t *)cvp);
thread_unlock_nopreempt(t);
mutex_exit(mp);
if (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t))
setrun(t);
/* ASSERT(no locks are held) */
swtch();
signalled = (t->t_schedflag & TS_SIGNALLED);
t->t_flag &= ~T_WAKEABLE;
mutex_enter(mp);
if (ISSIG_PENDING(t, lwp, p)) {
mutex_exit(mp);
if (issig(FORREAL))
rval = 0;
mutex_enter(mp);
}
if (lwp->lwp_sysabort || MUSTRETURN(p, t))
rval = 0;
lwp->lwp_asleep = 0;
lwp->lwp_sysabort = 0;
if (rval == 0 && signalled) /* avoid consuming the cv_signal() */
cv_signal(cvp);
return (rval);
}
/*
* Mark the current thread as sleeping on a shuttle object, and
* switch to a new thread.
* No locks other than 'l' should be held at this point.
*/
void
shuttle_swtch(kmutex_t *l)
{
klwp_t *lwp = ttolwp(curthread);
thread_lock(curthread);
disp_lock_enter_high(&shuttle_lock);
lwp->lwp_asleep = 1; /* /proc */
lwp->lwp_sysabort = 0; /* /proc */
lwp->lwp_ru.nvcsw++;
curthread->t_flag |= T_WAKEABLE;
curthread->t_sobj_ops = &shuttle_sobj_ops;
curthread->t_wchan0 = (caddr_t)1;
CL_INACTIVE(curthread);
DTRACE_SCHED(sleep);
THREAD_SLEEP(curthread, &shuttle_lock);
(void) new_mstate(curthread, LMS_SLEEP);
disp_lock_exit_high(&shuttle_lock);
mutex_exit(l);
if (ISSIG(curthread, JUSTLOOKING) || MUSTRETURN(curproc, curthread))
setrun(curthread);
swtch();
/*
* Caller must check for ISSIG/lwp_sysabort conditions
* and clear lwp->lwp_asleep/lwp->lwp_sysabort
*/
}
/*
* Mark the specified thread as once again sleeping on a shuttle object. This
* routine is called to put a server thread -- one that was dequeued but for
* which shuttle_resume() was _not_ called -- back to sleep on a shuttle
* object. Because we don't hit the sched:::wakeup DTrace probe until
* shuttle_resume(), we do _not_ have a sched:::sleep probe here.
*/
void
shuttle_sleep(kthread_t *t)
{
klwp_t *lwp = ttolwp(t);
proc_t *p = ttoproc(t);
thread_lock(t);
disp_lock_enter_high(&shuttle_lock);
if (lwp != NULL) {
lwp->lwp_asleep = 1; /* /proc */
lwp->lwp_sysabort = 0; /* /proc */
lwp->lwp_ru.nvcsw++;
}
t->t_flag |= T_WAKEABLE;
t->t_sobj_ops = &shuttle_sobj_ops;
t->t_wchan0 = (caddr_t)1;
CL_INACTIVE(t);
ASSERT(t->t_mstate == LMS_SLEEP);
THREAD_SLEEP(t, &shuttle_lock);
disp_lock_exit_high(&shuttle_lock);
if (lwp && (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t)))
setrun(t);
}
int
turnstile_block(turnstile_t *ts, int qnum, void *sobj, sobj_ops_t *sobj_ops,
kmutex_t *mp, lwp_timer_t *lwptp)
{
kthread_t *owner;
kthread_t *t = curthread;
proc_t *p = ttoproc(t);
klwp_t *lwp = ttolwp(t);
turnstile_chain_t *tc = &TURNSTILE_CHAIN(sobj);
int error = 0;
int loser = 0;
ASSERT(DISP_LOCK_HELD(&tc->tc_lock));
ASSERT(mp == NULL || IS_UPI(mp));
ASSERT((SOBJ_TYPE(sobj_ops) == SOBJ_USER_PI) ^ (mp == NULL));
thread_lock_high(t);
if (ts == NULL) {
/*
* This is the first thread to block on this sobj.
* Take its attached turnstile and add it to the hash chain.
*/
ts = t->t_ts;
ts->ts_sobj = sobj;
ts->ts_next = tc->tc_first;
tc->tc_first = ts;
ASSERT(ts->ts_waiters == 0);
} else {
/*
* Another thread has already donated its turnstile
* to block on this sobj, so ours isn't needed.
* Stash it on the active turnstile's freelist.
*/
turnstile_t *myts = t->t_ts;
myts->ts_free = ts->ts_free;
ts->ts_free = myts;
t->t_ts = ts;
ASSERT(ts->ts_sobj == sobj);
ASSERT(ts->ts_waiters > 0);
}
/*
* Put the thread to sleep.
*/
ASSERT(t != CPU->cpu_idle_thread);
ASSERT(CPU_ON_INTR(CPU) == 0);
ASSERT(t->t_wchan0 == NULL && t->t_wchan == NULL);
ASSERT(t->t_state == TS_ONPROC);
if (SOBJ_TYPE(sobj_ops) == SOBJ_USER_PI) {
curthread->t_flag |= T_WAKEABLE;
}
CL_SLEEP(t); /* assign kernel priority */
THREAD_SLEEP(t, &tc->tc_lock);
t->t_wchan = sobj;
t->t_sobj_ops = sobj_ops;
DTRACE_SCHED(sleep);
if (lwp != NULL) {
lwp->lwp_ru.nvcsw++;
(void) new_mstate(t, LMS_SLEEP);
if (SOBJ_TYPE(sobj_ops) == SOBJ_USER_PI) {
lwp->lwp_asleep = 1;
lwp->lwp_sysabort = 0;
/*
* make wchan0 non-zero to conform to the rule that
* threads blocking for user-level objects have a
* non-zero wchan0: this prevents spurious wake-ups
* by, for example, /proc.
*/
t->t_wchan0 = (caddr_t)1;
}
}
ts->ts_waiters++;
sleepq_insert(&ts->ts_sleepq[qnum], t);
if (SOBJ_TYPE(sobj_ops) == SOBJ_MUTEX &&
SOBJ_OWNER(sobj_ops, sobj) == NULL)
panic("turnstile_block(%p): unowned mutex", (void *)ts);
/*
* Follow the blocking chain to its end, willing our priority to
* everyone who's in our way.
*/
while (t->t_sobj_ops != NULL &&
(owner = SOBJ_OWNER(t->t_sobj_ops, t->t_wchan)) != NULL) {
if (owner == curthread) {
if (SOBJ_TYPE(sobj_ops) != SOBJ_USER_PI) {
panic("Deadlock: cycle in blocking chain");
}
/*
* If the cycle we've encountered ends in mp,
* then we know it isn't a 'real' cycle because
* we're going to drop mp before we go to sleep.
* Moreover, since we've come full circle we know
* that we must have willed priority to everyone
* in our way. Therefore, we can break out now.
*/
if (t->t_wchan == (void *)mp)
break;
if (loser)
lock_clear(&turnstile_loser_lock);
/*
* For SOBJ_USER_PI, a cycle is an application
* deadlock which needs to be communicated
* back to the application.
*/
thread_unlock_nopreempt(t);
mutex_exit(mp);
setrun(curthread);
swtch(); /* necessary to transition state */
curthread->t_flag &= ~T_WAKEABLE;
if (lwptp->lwpt_id != 0)
(void) lwp_timer_dequeue(lwptp);
setallwatch();
lwp->lwp_asleep = 0;
lwp->lwp_sysabort = 0;
return (EDEADLK);
}
if (!turnstile_interlock(t->t_lockp, &owner->t_lockp)) {
/*
* If we failed to grab the owner's thread lock,
* turnstile_interlock() will have dropped t's
* thread lock, so at this point we don't even know
* that 't' exists anymore. The simplest solution
* is to restart the entire priority inheritance dance
* from the beginning of the blocking chain, since
* we *do* know that 'curthread' still exists.
* Application of priority inheritance is idempotent,
* so it's OK that we're doing it more than once.
* Note also that since we've dropped our thread lock,
* we may already have been woken up; if so, our
* t_sobj_ops will be NULL, the loop will terminate,
* and the call to swtch() will be a no-op. Phew.
*
* There is one further complication: if two (or more)
* threads keep trying to grab the turnstile locks out
* of order and keep losing the race to another thread,
* these "dueling losers" can livelock the system.
* Therefore, once we get into this rare situation,
* we serialize all the losers.
*/
if (loser == 0) {
loser = 1;
lock_set(&turnstile_loser_lock);
}
t = curthread;
thread_lock_high(t);
continue;
}
/*
* We now have the owner's thread lock. If we are traversing
* from non-SOBJ_USER_PI ops to SOBJ_USER_PI ops, then we know
* that we have caught the thread while in the TS_SLEEP state,
* but holding mp. We know that this situation is transient
* (mp will be dropped before the holder actually sleeps on
* the SOBJ_USER_PI sobj), so we will spin waiting for mp to
* be dropped. Then, as in the turnstile_interlock() failure
* case, we will restart the priority inheritance dance.
*/
if (SOBJ_TYPE(t->t_sobj_ops) != SOBJ_USER_PI &&
owner->t_sobj_ops != NULL &&
SOBJ_TYPE(owner->t_sobj_ops) == SOBJ_USER_PI) {
kmutex_t *upi_lock = (kmutex_t *)t->t_wchan;
ASSERT(IS_UPI(upi_lock));
ASSERT(SOBJ_TYPE(t->t_sobj_ops) == SOBJ_MUTEX);
if (t->t_lockp != owner->t_lockp)
thread_unlock_high(owner);
thread_unlock_high(t);
if (loser)
lock_clear(&turnstile_loser_lock);
while (mutex_owner(upi_lock) == owner) {
SMT_PAUSE();
continue;
}
if (loser)
lock_set(&turnstile_loser_lock);
t = curthread;
thread_lock_high(t);
continue;
}
turnstile_pi_inherit(t->t_ts, owner, DISP_PRIO(t));
if (t->t_lockp != owner->t_lockp)
thread_unlock_high(t);
t = owner;
}
if (loser)
lock_clear(&turnstile_loser_lock);
/*
* Note: 't' and 'curthread' were synonymous before the loop above,
* but now they may be different. ('t' is now the last thread in
* the blocking chain.)
*/
if (SOBJ_TYPE(sobj_ops) == SOBJ_USER_PI) {
ushort_t s = curthread->t_oldspl;
int timedwait = 0;
uint_t imm_timeout = 0;
clock_t tim = -1;
thread_unlock_high(t);
if (lwptp->lwpt_id != 0) {
/*
* We enqueued a timeout. If it has already fired,
* lwptp->lwpt_imm_timeout has been set with cas,
* so fetch it with cas.
*/
timedwait = 1;
imm_timeout =
atomic_cas_uint(&lwptp->lwpt_imm_timeout, 0, 0);
}
mutex_exit(mp);
splx(s);
if (ISSIG(curthread, JUSTLOOKING) ||
MUSTRETURN(p, curthread) || imm_timeout)
setrun(curthread);
swtch();
curthread->t_flag &= ~T_WAKEABLE;
if (timedwait)
tim = lwp_timer_dequeue(lwptp);
setallwatch();
if (ISSIG(curthread, FORREAL) || lwp->lwp_sysabort ||
MUSTRETURN(p, curthread))
error = EINTR;
else if (imm_timeout || (timedwait && tim == -1))
error = ETIME;
lwp->lwp_sysabort = 0;
lwp->lwp_asleep = 0;
} else {
thread_unlock_nopreempt(t);
swtch();
}
return (error);
}
/*
* Mark the current thread as sleeping on a shuttle object, and
* resume the specified thread. The 't' thread must be marked as ONPROC.
*
* No locks other than 'l' should be held at this point.
*/
void
shuttle_resume(kthread_t *t, kmutex_t *l)
{
klwp_t *lwp = ttolwp(curthread);
cpu_t *cp;
disp_lock_t *oldtlp;
thread_lock(curthread);
disp_lock_enter_high(&shuttle_lock);
if (lwp != NULL) {
lwp->lwp_asleep = 1; /* /proc */
lwp->lwp_sysabort = 0; /* /proc */
lwp->lwp_ru.nvcsw++;
}
curthread->t_flag |= T_WAKEABLE;
curthread->t_sobj_ops = &shuttle_sobj_ops;
/*
* setting cpu_dispthread before changing thread state
* so that kernel preemption will be deferred to after swtch_to()
*/
cp = CPU;
cp->cpu_dispthread = t;
cp->cpu_dispatch_pri = DISP_PRIO(t);
/*
* Set the wchan0 field so that /proc won't just do a setrun
* on this thread when trying to stop a process. Instead,
* /proc will mark the thread as VSTOPPED similar to threads
* that are blocked on user level condition variables.
*/
curthread->t_wchan0 = (caddr_t)1;
CL_INACTIVE(curthread);
DTRACE_SCHED1(wakeup, kthread_t *, t);
DTRACE_SCHED(sleep);
THREAD_SLEEP(curthread, &shuttle_lock);
disp_lock_exit_high(&shuttle_lock);
/*
* Update ustate records (there is no waitrq obviously)
*/
(void) new_mstate(curthread, LMS_SLEEP);
thread_lock_high(t);
oldtlp = t->t_lockp;
restore_mstate(t);
t->t_flag &= ~T_WAKEABLE;
t->t_wchan0 = NULL;
t->t_sobj_ops = NULL;
/*
* Make sure we end up on the right CPU if we are dealing with bound
* CPU's or processor partitions.
*/
if (t->t_bound_cpu != NULL || t->t_cpupart != cp->cpu_part) {
aston(t);
cp->cpu_runrun = 1;
}
/*
* We re-assign t_disp_queue and t_lockp of 't' here because
* 't' could have been preempted.
*/
if (t->t_disp_queue != cp->cpu_disp) {
t->t_disp_queue = cp->cpu_disp;
thread_onproc(t, cp);
}
/*
* We can't call thread_unlock_high() here because t's thread lock
* could have changed by thread_onproc() call above to point to
* CPU->cpu_thread_lock.
*/
disp_lock_exit_high(oldtlp);
mutex_exit(l);
/*
* Make sure we didn't receive any important events while
* we weren't looking
*/
if (lwp &&
(ISSIG(curthread, JUSTLOOKING) || MUSTRETURN(curproc, curthread)))
setrun(curthread);
swtch_to(t);
/*
* Caller must check for ISSIG/lwp_sysabort conditions
* and clear lwp->lwp_asleep/lwp->lwp_sysabort
*/
}
/*
* similiar to sema_p except that it blocks at an interruptible
* priority. if a signal is present then return 1 otherwise 0.
*/
int
sema_p_sig(ksema_t *sp)
{
kthread_t *t = curthread;
klwp_t *lwp = ttolwp(t);
sema_impl_t *s;
disp_lock_t *sqlp;
if (lwp == NULL) {
sema_p(sp);
return (0);
}
s = (sema_impl_t *)sp;
sqlp = &SQHASH(s)->sq_lock;
disp_lock_enter(sqlp);
ASSERT(s->s_count >= 0);
while (s->s_count == 0) {
proc_t *p = ttoproc(t);
thread_lock_high(t);
t->t_flag |= T_WAKEABLE;
SEMA_BLOCK(s, sqlp);
lwp->lwp_asleep = 1;
lwp->lwp_sysabort = 0;
thread_unlock_nopreempt(t);
if (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t))
setrun(t);
swtch();
t->t_flag &= ~T_WAKEABLE;
if (ISSIG(t, FORREAL) ||
lwp->lwp_sysabort || MUSTRETURN(p, t)) {
kthread_t *sq, *tp;
lwp->lwp_asleep = 0;
lwp->lwp_sysabort = 0;
disp_lock_enter(sqlp);
sq = s->s_slpq;
/*
* in case sema_v and interrupt happen
* at the same time, we need to pass the
* sema_v to the next thread.
*/
if ((sq != NULL) && (s->s_count > 0)) {
tp = sq;
ASSERT(THREAD_LOCK_HELD(tp));
sq = sq->t_link;
tp->t_link = NULL;
DTRACE_SCHED1(wakeup, kthread_t *, tp);
tp->t_sobj_ops = NULL;
tp->t_wchan = NULL;
ASSERT(tp->t_state == TS_SLEEP);
CL_WAKEUP(tp);
s->s_slpq = sq;
disp_lock_exit_high(sqlp);
thread_unlock(tp);
} else {
disp_lock_exit(sqlp);
}
return (1);
}
lwp->lwp_asleep = 0;
disp_lock_enter(sqlp);
}
s->s_count--;
disp_lock_exit(sqlp);
return (0);
}
/*
* Returns:
* Function result in order of presidence:
* 0 if a signal was received
* -1 if timeout occured
* >0 if awakened via cv_signal() or cv_broadcast().
* (returns time remaining)
*
* cv_timedwait_sig() is now part of the DDI.
*/
clock_t
cv_timedwait_sig(kcondvar_t *cvp, kmutex_t *mp, clock_t tim)
{
kthread_t *t = curthread;
proc_t *p = ttoproc(t);
klwp_t *lwp = ttolwp(t);
timeout_id_t id;
clock_t rval = 1;
clock_t timeleft;
int signalled = 0;
if (panicstr)
return (rval);
/*
* If there is no lwp, then we don't need to wait for a signal.
* The check for t_intr is to catch an interrupt thread
* that has not yet unpinned the thread underneath.
*/
if (lwp == NULL || t->t_intr)
return (cv_timedwait(cvp, mp, tim));
/*
* If tim is less than or equal to lbolt, then the timeout
* has already occured. So just check to see if there is a signal
* pending. If so return 0 indicating that there is a signal pending.
* Else return -1 indicating that the timeout occured. No need to
* wait on anything.
*/
timeleft = tim - lbolt;
if (timeleft <= 0) {
lwp->lwp_asleep = 1;
lwp->lwp_sysabort = 0;
rval = -1;
goto out;
}
/*
* Set the timeout and wait.
*/
id = realtime_timeout((void (*)(void *))setrun, t, timeleft);
lwp->lwp_asleep = 1;
lwp->lwp_sysabort = 0;
thread_lock(t);
cv_block_sig(t, (condvar_impl_t *)cvp);
thread_unlock_nopreempt(t);
mutex_exit(mp);
if (ISSIG(t, JUSTLOOKING) || MUSTRETURN(p, t) || (tim - lbolt <= 0))
setrun(t);
/* ASSERT(no locks are held) */
swtch();
signalled = (t->t_schedflag & TS_SIGNALLED);
t->t_flag &= ~T_WAKEABLE;
mutex_enter(mp);
/*
* Untimeout the thread. untimeout() returns -1 if the timeout has
* occured or the time remaining. If the time remaining is zero,
* the timeout has occured between when we were awoken and
* we called untimeout. We will treat this as if the timeout
* has occured and set rval to -1.
*/
rval = untimeout(id);
if (rval <= 0)
rval = -1;
/*
* Check to see if a signal is pending. If so, regardless of whether
* or not we were awoken due to the signal, the signal is now pending
* and a return of 0 has the highest priority.
*/
out:
if (ISSIG_PENDING(t, lwp, p)) {
mutex_exit(mp);
if (issig(FORREAL))
rval = 0;
mutex_enter(mp);
}
if (lwp->lwp_sysabort || MUSTRETURN(p, t))
rval = 0;
lwp->lwp_asleep = 0;
lwp->lwp_sysabort = 0;
if (rval <= 0 && signalled) /* avoid consuming the cv_signal() */
cv_signal(cvp);
return (rval);
}