inline void acquire_lock(){
int delay;
int cond;
delay = 1;
TAS(lock,cond);
while(cond == locked){
bassume(cond == locked);
pause(delay);
if(delay*2 > delay)
delay *= 2;
TAS(lock,cond);
}
bassume(!(cond == locked));
assert(cond != lock);
}
void
ilock(Lock *l)
{
Proc *up = externup();
Mpl pl;
uintptr_t pc;
uint64_t t0;
pc = getcallerpc();
lockstats.locks++;
pl = splhi();
if(TAS(&l->key) != 0){
cycles(&t0);
lockstats.glare++;
/*
* Cannot also check l->pc, l->m, or l->isilock here
* because they might just not be set yet, or
* (for pc and m) the lock might have just been unlocked.
*/
for(;;){
lockstats.inglare++;
splx(pl);
while(l->key)
;
pl = splhi();
if(TAS(&l->key) == 0){
if(l != &waitstatslk)
addwaitstat(pc, t0, WSlock);
goto acquire;
}
}
}
acquire:
machp()->ilockdepth++;
if(up)
up->lastilock = l;
l->pl = pl;
l->_pc = pc;
l->p = up;
l->isilock = 1;
l->m = machp();
if(LOCKCYCLES)
cycles(&l->lockcycles);
}
int
canlock(Lock *l)
{
Proc *up = externup();
if(up)
ainc(&up->nlocks);
if(TAS(&l->key)){
if(up)
adec(&up->nlocks);
return 0;
}
if(up)
up->lastlock = l;
l->_pc = getcallerpc();
l->p = up;
l->isilock = 0;
if(LOCKCYCLES)
cycles(&l->lockcycles);
return 1;
}
int
lock(Lock *l)
{
Proc *up = externup();
int i;
uintptr_t pc;
uint64_t t0;
pc = getcallerpc();
lockstats.locks++;
if(up)
ainc(&up->nlocks); /* prevent being scheded */
if(TAS(&l->key) == 0){
if(up)
up->lastlock = l;
l->_pc = pc;
l->p = up;
l->isilock = 0;
if(LOCKCYCLES)
cycles(&l->lockcycles);
return 0;
}
if(up)
adec(&up->nlocks);
cycles(&t0);
lockstats.glare++;
for(;;){
lockstats.inglare++;
i = 0;
while(l->key){
if(sys->nmach < 2 && up && up->edf && (up->edf->flags & Admitted)){
/*
* Priority inversion, yield on a uniprocessor; on a
* multiprocessor, the other processor will unlock
*/
print("inversion %#p pc %#p proc %d held by pc %#p proc %d\n",
l, pc, up ? up->pid : 0, l->_pc, l->p ? l->p->pid : 0);
up->edf->d = todget(nil); /* yield to process with lock */
}
if(i++ > 100000000){
i = 0;
lockloop(l, pc);
}
}
if(up)
ainc(&up->nlocks);
if(TAS(&l->key) == 0){
if(up)
up->lastlock = l;
l->_pc = pc;
l->p = up;
l->isilock = 0;
if(LOCKCYCLES)
cycles(&l->lockcycles);
if(l != &waitstatslk)
addwaitstat(pc, t0, WSlock);
return 1;
}
if(up)
adec(&up->nlocks);
}
}
void lock() {
while(TAS(1)) { }
}
/*
* s_lock(lock) - platform-independent portion of waiting for a spinlock.
*/
void
s_lock(volatile slock_t *lock, const char *file, int line TSRMLS_DC)
{
/*
* We loop tightly for awhile, then delay using pg_usleep() and try again.
* Preferably, "awhile" should be a small multiple of the maximum time we
* expect a spinlock to be held. 100 iterations seems about right as an
* initial guess. However, on a uniprocessor the loop is a waste of
* cycles, while in a multi-CPU scenario it's usually better to spin a bit
* longer than to call the kernel, so we try to adapt the spin loop count
* depending on whether we seem to be in a uniprocessor or multiprocessor.
*
* Note: you might think MIN_SPINS_PER_DELAY should be just 1, but you'd
* be wrong; there are platforms where that can result in a "stuck
* spinlock" failure. This has been seen particularly on Alphas; it seems
* that the first TAS after returning from kernel space will always fail
* on that hardware.
*
* Once we do decide to block, we use randomly increasing pg_usleep()
* delays. The first delay is 1 msec, then the delay randomly increases to
* about one second, after which we reset to 1 msec and start again. The
* idea here is that in the presence of heavy contention we need to
* increase the delay, else the spinlock holder may never get to run and
* release the lock. (Consider situation where spinlock holder has been
* nice'd down in priority by the scheduler --- it will not get scheduled
* until all would-be acquirers are sleeping, so if we always use a 1-msec
* sleep, there is a real possibility of starvation.) But we can't just
* clamp the delay to an upper bound, else it would take a long time to
* make a reasonable number of tries.
*
* We time out and declare error after NUM_DELAYS delays (thus, exactly
* that many tries). With the given settings, this will usually take 2 or
* so minutes. It seems better to fix the total number of tries (and thus
* the probability of unintended failure) than to fix the total time
* spent.
*
* The pg_usleep() delays are measured in milliseconds because 1 msec is a
* common resolution limit at the OS level for newer platforms. On older
* platforms the resolution limit is usually 10 msec, in which case the
* total delay before timeout will be a bit more.
*/
#define MIN_SPINS_PER_DELAY 10
#define MAX_SPINS_PER_DELAY 1000
#define NUM_DELAYS 1000
#define MIN_DELAY_MSEC 1
#define MAX_DELAY_MSEC 1000
int spins = 0;
int delays = 0;
int cur_delay = 0;
while (TAS(lock))
{
/* CPU-specific delay each time through the loop */
SPIN_DELAY();
/* Block the process every spins_per_delay tries */
if (++spins >= spins_per_delay)
{
if (++delays > NUM_DELAYS)
s_lock_stuck(lock, file, line TSRMLS_CC);
if (cur_delay == 0) /* first time to delay? */
cur_delay = MIN_DELAY_MSEC;
pg_usleep(cur_delay * 1000L);
#if defined(S_LOCK_TEST)
fprintf(stdout, "*");
fflush(stdout);
#endif
/* increase delay by a random fraction between 1X and 2X */
cur_delay += (int) (cur_delay *
((double) rand() / (double) MAX_RANDOM_VALUE) + 0.5);
/* wrap back to minimum delay when max is exceeded */
if (cur_delay > MAX_DELAY_MSEC)
cur_delay = MIN_DELAY_MSEC;
spins = 0;
}
}
/*
* If we were able to acquire the lock without delaying, it's a good
* indication we are in a multiprocessor. If we had to delay, it's a sign
* (but not a sure thing) that we are in a uniprocessor. Hence, we
* decrement spins_per_delay slowly when we had to delay, and increase it
* rapidly when we didn't. It's expected that spins_per_delay will
* converge to the minimum value on a uniprocessor and to the maximum
* value on a multiprocessor.
*
* Note: spins_per_delay is local within our current process. We want to
* average these observations across multiple backends, since it's
* relatively rare for this function to even get entered, and so a single
* backend might not live long enough to converge on a good value. That
* is handled by the two routines below.
*/
if (cur_delay == 0)
{
/* we never had to delay */
if (spins_per_delay < MAX_SPINS_PER_DELAY)
spins_per_delay = Min(spins_per_delay + 100, MAX_SPINS_PER_DELAY);
}
else
{
if (spins_per_delay > MIN_SPINS_PER_DELAY)
spins_per_delay = Max(spins_per_delay - 1, MIN_SPINS_PER_DELAY);
}
}
bool
pg_atomic_test_set_flag_impl(volatile pg_atomic_flag *ptr)
{
return TAS((slock_t *) &ptr->sema);
}
/*
****************************************************************
* Inicializa a rede ou obtém um EP *
****************************************************************
*/
void
itnopen (dev_t dev, int oflag)
{
ITSCB *ip = &itscb;
KFILE *fp = u.u_fileptr;
IT_MINOR minor = MINOR (dev);
/*
* Verifica se a a rede está inicializada
*/
if (TAS (&ip->it_init_lock) >= 0)
{
if (minor == DAEMON)
{
if (superuser () < 0)
return;
init_it_block ();
ether_init ();
raw_ep_free_init ();
udp_ep_free_init ();
tcp_ep_free_init ();
ip->it_N_BLOCK = scb.y_n_itblock;
if (ip->it_WND_SZ == 0)
ip->it_WND_SZ = WND_SZ;
if (ip->it_GOOD_WND == 0)
ip->it_GOOD_WND = GOOD_WND;
if (ip->it_ALPHA == 0)
ip->it_ALPHA = DEF_ALPHA;
if (ip->it_BETA == 0)
ip->it_BETA = DEF_BETA;
if (ip->it_SRTT == 0)
ip->it_SRTT = INIT_SRTT;
if (ip->it_N_TRANS == 0)
ip->it_N_TRANS = DEF_N_TRANS;
if (ip->it_WAIT == 0)
ip->it_WAIT = DEF_WAIT;
if (ip->it_SILENCE == 0)
ip->it_SILENCE = DEF_SILENCE;
if (ip->it_MAX_SGSZ == 0)
ip->it_MAX_SGSZ = MAX_SGSZ;
SEMAINIT (&ip->it_block_sema, ip->it_N_BLOCK, 0 /* sem histórico */);
/*** ip->it_gateway = 1; ***/
ip->it_pipe_mode = 1;
}
else
{
CLEAR (&ip->it_init_lock);
u.u_error = TBADNET;
}
return;
}
/*
* Pequena consistência
*/
if ((oflag & O_RW) != O_RW)
{ u.u_error = EINVAL; return; }
/*
* Obtém e inicializa um "endpoint"
*/
switch (minor)
{
/*
* O "in_daemon": "open"s seguintes não fazem efeito
*/
case DAEMON:
return;
/*
* Protocolo RAW
*/
case RAW:
{
RAW_EP *rp;
if (fp->f_union != KF_NULL)
{ u.u_error = EBADF; return; }
if ((rp = get_raw_ep ()) == NO_RAW_EP)
{ u.u_error = EAGAIN; return; }
/*** SPINLOCK (&rp->rp_inq_lock); ***/
rp->rp_state = S_UNBOUND;
fp->f_union = KF_ITNET;
fp->f_endpoint = rp;
/*** SPINFREE (&rp->rp_inq_lock); ***/
return;
}
/*
* Protocolo UDP
*/
case UDP:
{
UDP_EP *up;
if (fp->f_union != KF_NULL)
{ u.u_error = EBADF; return; }
if ((up = get_udp_ep ()) == NO_UDP_EP)
{ u.u_error = EAGAIN; return; }
/*** SPINLOCK (&up->up_inq_lock); ***/
up->up_state = S_UNBOUND;
fp->f_union = KF_ITNET;
fp->f_endpoint = up;
/*** SPINFREE (&up->up_inq_lock); ***/
return;
}
/*
* Protocolo TCP
*/
case TCP:
{
TCP_EP *tp;
if (fp->f_union != KF_NULL)
{ u.u_error = EBADF; return; }
if ((tp = get_tcp_ep ()) == NO_TCP_EP)
{ u.u_error = EAGAIN; return; }
/*** SLEEPLOCK (&tp->tp_lock, PITNETOUT); ***/
tp->tp_state = S_UNBOUND;
tp->tp_SRTT = ip->it_SRTT;
tp->tp_max_seg_sz = ip->it_MAX_SGSZ;
tp->tp_good_wnd = ip->it_GOOD_WND;
tp->tp_max_wait = ip->it_WAIT;
tp->tp_max_silence = ip->it_SILENCE;
tp->tp_rnd_in.tp_rnd_sz = ip->it_WND_SZ;
tp->tp_rnd_out.tp_rnd_sz = ip->it_WND_SZ;
tp->tp_last_rcv_time = time;
fp->f_union = KF_ITNET;
fp->f_endpoint = tp;
/*** SLEEPFREE (&tp->tp_lock); ***/
return;
}
} /* end switch */
u.u_error = ENXIO;
return;
} /* end itnopen */
