/*
* This function is used to acquire a contested lock.
*/
int
__sysv_umtx_lock(volatile umtx_t *mtx, int timo)
{
int v, errval, ret = 0;
/* contested */
do {
v = *mtx;
if (v == 2 || atomic_cmpset_acq_int(mtx, 1, 2)) {
if (timo == 0)
umtx_sleep(mtx, 2, timo);
else if ( (errval = umtx_sleep(mtx, 2, timo)) > 0) {
if (errval == EAGAIN) {
if (atomic_cmpset_acq_int(mtx, 0, 2))
ret = 0;
else
ret = ETIMEDOUT;
break;
}
}
}
} while (!atomic_cmpset_acq_int(mtx, 0, 2));
return (ret);
}
int
_pthread_once(pthread_once_t *once_control, void (*init_routine) (void))
{
struct pthread *curthread;
int state;
_thr_check_init();
for (;;) {
state = once_control->state;
if (state == ONCE_DONE)
return (0);
if (state == ONCE_NEVER_DONE) {
if (atomic_cmpset_acq_int(&once_control->state, state, ONCE_IN_PROGRESS))
break;
} else if (state == ONCE_IN_PROGRESS) {
if (atomic_cmpset_acq_int(&once_control->state, state, ONCE_WAIT))
_thr_umtx_wait_uint(&once_control->state, ONCE_WAIT, NULL, 0);
} else if (state == ONCE_WAIT) {
_thr_umtx_wait_uint(&once_control->state, state, NULL, 0);
} else
return (EINVAL);
}
curthread = _get_curthread();
THR_CLEANUP_PUSH(curthread, once_cancel_handler, once_control);
init_routine();
THR_CLEANUP_POP(curthread, 0);
if (atomic_cmpset_rel_int(&once_control->state, ONCE_IN_PROGRESS, ONCE_DONE))
return (0);
atomic_store_rel_int(&once_control->state, ONCE_DONE);
_thr_umtx_wake(&once_control->state, INT_MAX, 0);
return (0);
}
int
_pthread_cancel(pthread_t pthread)
{
struct pthread *curthread = tls_get_curthread();
int oldval, newval = 0;
int oldtype;
int ret;
/*
* POSIX says _pthread_cancel should be async cancellation safe,
* so we temporarily disable async cancellation.
*/
_pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, &oldtype);
if ((ret = _thr_ref_add(curthread, pthread, 0)) != 0) {
_pthread_setcanceltype(oldtype, NULL);
return (ret);
}
do {
oldval = pthread->cancelflags;
if (oldval & THR_CANCEL_NEEDED)
break;
newval = oldval | THR_CANCEL_NEEDED;
} while (!atomic_cmpset_acq_int(&pthread->cancelflags, oldval, newval));
if (!(oldval & THR_CANCEL_NEEDED) && SHOULD_ASYNC_CANCEL(newval))
_thr_send_sig(pthread, SIGCANCEL);
_thr_ref_delete(curthread, pthread);
_pthread_setcanceltype(oldtype, NULL);
return (0);
}
static void
def_wlock_acquire(void *lock)
{
Lock *l = (Lock *)lock;
sigset_t tmp_oldsigmask;
for ( ; ; ) {
sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask);
if (atomic_cmpset_acq_int(&l->lock, 0, WAFLAG))
break;
sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL);
}
oldsigmask = tmp_oldsigmask;
}
/*
* This function is used to acquire a contested lock.
*
* A *mtx value of 1 indicates locked normally.
* A *mtx value of 2 indicates locked and contested.
*/
int
__thr_umtx_lock(volatile umtx_t *mtx, int id, int timo)
{
int v;
int errval;
int ret = 0;
int retry = 4;
v = *mtx;
cpu_ccfence();
id &= 0x3FFFFFFF;
for (;;) {
cpu_pause();
if (v == 0) {
if (atomic_fcmpset_int(mtx, &v, id))
break;
continue;
}
if (--retry) {
sched_yield();
v = *mtx;
continue;
}
/*
* Set the waiting bit. If the fcmpset fails v is loaded
* with the current content of the mutex, and if the waiting
* bit is already set, we can also sleep.
*/
if (atomic_fcmpset_int(mtx, &v, v|0x40000000) ||
(v & 0x40000000)) {
if (timo == 0) {
_umtx_sleep_err(mtx, v|0x40000000, timo);
} else if ((errval = _umtx_sleep_err(mtx, v|0x40000000, timo)) > 0) {
if (errval == EAGAIN) {
if (atomic_cmpset_acq_int(mtx, 0, id))
ret = 0;
else
ret = ETIMEDOUT;
break;
}
}
}
retry = 4;
}
return (ret);
}
void
__sysv_umtx_unlock(volatile umtx_t *mtx)
{
int v;
for (;;) {
v = *mtx;
if (atomic_cmpset_acq_int(mtx, v, v-1)) {
if (v != 1) {
*mtx = 0;
umtx_wakeup(mtx, 1);
}
break;
}
}
}
/*
* MP-friendly version of ppsratecheck().
*
* Returns non-negative if we are in the rate, negative otherwise.
* 0 - rate limit not reached.
* -1 - rate limit reached.
* >0 - rate limit was reached before, and was just reset. The return value
* is number of events since last reset.
*/
int64_t
counter_ratecheck(struct counter_rate *cr, int64_t limit)
{
int64_t val;
int now;
val = cr->cr_over;
now = ticks;
if (now - cr->cr_ticks >= hz) {
/*
* Time to clear the structure, we are in the next second.
* First try unlocked read, and then proceed with atomic.
*/
if ((cr->cr_lock == 0) &&
atomic_cmpset_acq_int(&cr->cr_lock, 0, 1)) {
/*
* Check if other thread has just went through the
* reset sequence before us.
*/
if (now - cr->cr_ticks >= hz) {
val = counter_u64_fetch(cr->cr_rate);
counter_u64_zero(cr->cr_rate);
cr->cr_over = 0;
cr->cr_ticks = now;
if (val <= limit)
val = 0;
}
atomic_store_rel_int(&cr->cr_lock, 0);
} else
/*
* We failed to lock, in this case other thread may
* be running counter_u64_zero(), so it is not safe
* to do an update, we skip it.
*/
return (val);
}
counter_u64_add(cr->cr_rate, 1);
if (cr->cr_over != 0)
return (-1);
if (counter_u64_fetch(cr->cr_rate) > limit)
val = cr->cr_over = -1;
return (val);
}
/* _mcount; may be static, inline, etc */
_MCOUNT_DECL(uintfptr_t frompc, uintfptr_t selfpc)
{
#ifdef GUPROF
u_int delta;
#endif
fptrdiff_t frompci;
u_short *frompcindex;
struct tostruct *top, *prevtop;
struct gmonparam *p;
long toindex;
#ifdef _KERNEL
MCOUNT_DECL(s)
#endif
p = &_gmonparam;
#ifndef GUPROF /* XXX */
/*
* check that we are profiling
* and that we aren't recursively invoked.
*/
if (p->state != GMON_PROF_ON)
return;
#endif
#ifdef _KERNEL
MCOUNT_ENTER(s);
#else
if (!atomic_cmpset_acq_int(&p->state, GMON_PROF_ON, GMON_PROF_BUSY))
return;
#endif
frompci = frompc - p->lowpc;
#ifdef _KERNEL
/*
* When we are called from an exception handler, frompci may be
* for a user address. Convert such frompci's to the index of
* user() to merge all user counts.
*/
if (frompci >= p->textsize) {
if (frompci + p->lowpc
>= (uintfptr_t)(VM_MAXUSER_ADDRESS + UPAGES * PAGE_SIZE))
goto done;
frompci = (uintfptr_t)user - p->lowpc;
if (frompci >= p->textsize)
goto done;
}
#endif
#ifdef GUPROF
if (p->state != GMON_PROF_HIRES)
goto skip_guprof_stuff;
/*
* Look at the clock and add the count of clock cycles since the
* clock was last looked at to a counter for frompc. This
* solidifies the count for the function containing frompc and
* effectively starts another clock for the current function.
* The count for the new clock will be solidified when another
* function call is made or the function returns.
*
* We use the usual sampling counters since they can be located
* efficiently. 4-byte counters are usually necessary.
*
* There are many complications for subtracting the profiling
* overheads from the counts for normal functions and adding
* them to the counts for mcount(), mexitcount() and cputime().
* We attempt to handle fractional cycles, but the overheads
* are usually underestimated because they are calibrated for
* a simpler than usual setup.
*/
delta = cputime() - p->mcount_overhead;
p->cputime_overhead_resid += p->cputime_overhead_frac;
p->mcount_overhead_resid += p->mcount_overhead_frac;
if ((int)delta < 0)
*p->mcount_count += delta + p->mcount_overhead
- p->cputime_overhead;
else if (delta != 0) {
if (p->cputime_overhead_resid >= CALIB_SCALE) {
p->cputime_overhead_resid -= CALIB_SCALE;
++*p->cputime_count;
--delta;
}
if (delta != 0) {
if (p->mcount_overhead_resid >= CALIB_SCALE) {
p->mcount_overhead_resid -= CALIB_SCALE;
++*p->mcount_count;
--delta;
}
KCOUNT(p, frompci) += delta;
}
*p->mcount_count += p->mcount_overhead_sub;
}
*p->cputime_count += p->cputime_overhead;
skip_guprof_stuff:
#endif /* GUPROF */
#ifdef _KERNEL
/*
* When we are called from an exception handler, frompc is faked
* to be for where the exception occurred. We've just solidified
* the count for there. Now convert frompci to the index of btrap()
* for trap handlers and bintr() for interrupt handlers to make
* exceptions appear in the call graph as calls from btrap() and
* bintr() instead of calls from all over.
*/
if ((uintfptr_t)selfpc >= (uintfptr_t)btrap
&& (uintfptr_t)selfpc < (uintfptr_t)eintr) {
if ((uintfptr_t)selfpc >= (uintfptr_t)bintr)
frompci = (uintfptr_t)bintr - p->lowpc;
else
frompci = (uintfptr_t)btrap - p->lowpc;
}
#endif
/*
* check that frompc is a reasonable pc value.
* for example: signal catchers get called from the stack,
* not from text space. too bad.
*/
if (frompci >= p->textsize)
goto done;
frompcindex = &p->froms[frompci / (p->hashfraction * sizeof(*p->froms))];
toindex = *frompcindex;
if (toindex == 0) {
/*
* first time traversing this arc
*/
toindex = ++p->tos[0].link;
if (toindex >= p->tolimit)
/* halt further profiling */
goto overflow;
*frompcindex = toindex;
top = &p->tos[toindex];
top->selfpc = selfpc;
top->count = 1;
top->link = 0;
goto done;
}
top = &p->tos[toindex];
if (top->selfpc == selfpc) {
/*
* arc at front of chain; usual case.
*/
top->count++;
goto done;
}
/*
* have to go looking down chain for it.
* top points to what we are looking at,
* prevtop points to previous top.
* we know it is not at the head of the chain.
*/
for (; /* goto done */; ) {
if (top->link == 0) {
/*
* top is end of the chain and none of the chain
* had top->selfpc == selfpc.
* so we allocate a new tostruct
* and link it to the head of the chain.
*/
toindex = ++p->tos[0].link;
if (toindex >= p->tolimit)
goto overflow;
top = &p->tos[toindex];
top->selfpc = selfpc;
top->count = 1;
top->link = *frompcindex;
*frompcindex = toindex;
goto done;
}
/*
* otherwise, check the next arc on the chain.
*/
prevtop = top;
top = &p->tos[top->link];
if (top->selfpc == selfpc) {
/*
* there it is.
* increment its count
* move it to the head of the chain.
*/
top->count++;
toindex = prevtop->link;
prevtop->link = top->link;
top->link = *frompcindex;
*frompcindex = toindex;
goto done;
}
}
done:
#ifdef _KERNEL
MCOUNT_EXIT(s);
#else
atomic_store_rel_int(&p->state, GMON_PROF_ON);
#endif
return;
overflow:
atomic_store_rel_int(&p->state, GMON_PROF_ERROR);
#ifdef _KERNEL
MCOUNT_EXIT(s);
#endif
return;
}
