void
forward_signal(struct thread *td)
{
int id;
/*
* signotify() has already set TDF_ASTPENDING and TDF_NEEDSIGCHECK on
* this thread, so all we need to do is poke it if it is currently
* executing so that it executes ast().
*/
THREAD_LOCK_ASSERT(td, MA_OWNED);
KASSERT(TD_IS_RUNNING(td),
("forward_signal: thread is not TDS_RUNNING"));
CTR1(KTR_SMP, "forward_signal(%p)", td->td_proc);
if (!smp_started || cold || panicstr)
return;
if (!forward_signal_enabled)
return;
/* No need to IPI ourself. */
if (td == curthread)
return;
id = td->td_oncpu;
if (id == NOCPU)
return;
ipi_cpu(id, IPI_AST);
}
/*
* General sleep call. Suspends the current thread until a wakeup is
* performed on the specified identifier. The thread will then be made
* runnable with the specified priority. Sleeps at most sbt units of time
* (0 means no timeout). If pri includes the PCATCH flag, let signals
* interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
* awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
* signal becomes pending, ERESTART is returned if the current system
* call should be restarted if possible, and EINTR is returned if the system
* call should be interrupted by the signal (return EINTR).
*
* The lock argument is unlocked before the caller is suspended, and
* re-locked before _sleep() returns. If priority includes the PDROP
* flag the lock is not re-locked before returning.
*/
int
_sleep(void *ident, struct lock_object *lock, int priority,
const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
{
struct thread *td;
struct proc *p;
struct lock_class *class;
uintptr_t lock_state;
int catch, pri, rval, sleepq_flags;
WITNESS_SAVE_DECL(lock_witness);
td = curthread;
p = td->td_proc;
#ifdef KTRACE
if (KTRPOINT(td, KTR_CSW))
ktrcsw(1, 0, wmesg);
#endif
WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
"Sleeping on \"%s\"", wmesg);
KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
("sleeping without a lock"));
KASSERT(p != NULL, ("msleep1"));
KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
if (priority & PDROP)
KASSERT(lock != NULL && lock != &Giant.lock_object,
("PDROP requires a non-Giant lock"));
if (lock != NULL)
class = LOCK_CLASS(lock);
else
void
stack_save_td(struct stack *st, struct thread *td)
{
if (TD_IS_SWAPPED(td))
panic("stack_save_td: swapped");
if (TD_IS_RUNNING(td))
panic("stack_save_td: running");
stack_capture(st, (struct frame *)(td->td_pcb->pcb_sp + SPOFF));
}
void
stack_save_td(struct stack *st, struct thread *td)
{
if (TD_IS_SWAPPED(td))
panic("stack_save_td: swapped");
if (TD_IS_RUNNING(td))
panic("stack_save_td: running");
stack_zero(st);
}
void
stack_save_td(struct stack *st, struct thread *td)
{
u_int32_t *frame;
if (TD_IS_SWAPPED(td))
panic("stack_save_td: swapped");
if (TD_IS_RUNNING(td))
panic("stack_save_td: running");
frame = (u_int32_t *)td->td_pcb->un_32.pcb32_r11;
stack_capture(st, frame);
}
void
stack_save_td(struct stack *st, struct thread *td)
{
vm_offset_t frame;
if (TD_IS_SWAPPED(td))
panic("stack_save_td: swapped");
if (TD_IS_RUNNING(td))
panic("stack_save_td: running");
frame = td->td_pcb->pcb_sp;
stack_capture(st, frame);
}
void
stack_save_td(struct stack *st, struct thread *td)
{
u_register_t pc, sp;
if (TD_IS_SWAPPED(td))
panic("stack_save_td: swapped");
if (TD_IS_RUNNING(td))
panic("stack_save_td: running");
pc = td->td_pcb->pcb_regs.pc;
sp = td->td_pcb->pcb_regs.sp;
stack_capture(st, pc, sp);
}
static int
proc_sum(struct proc *p, fixpt_t *estcpup)
{
struct thread *td;
int estcpu;
int val;
val = 0;
estcpu = 0;
FOREACH_THREAD_IN_PROC(p, td) {
thread_lock(td);
if (TD_ON_RUNQ(td) ||
TD_IS_RUNNING(td))
val = 1;
estcpu += sched_pctcpu(td);
thread_unlock(td);
}
void
stack_save_td(struct stack *st, struct thread *td)
{
u_int32_t *frame;
if (TD_IS_SWAPPED(td))
panic("stack_save_td: swapped");
if (TD_IS_RUNNING(td))
panic("stack_save_td: running");
/*
* This register, the frame pointer, is incorrect for the ARM EABI
* as it doesn't have a frame pointer, however it's value is not used
* when building for EABI.
*/
frame = (u_int32_t *)td->td_pcb->un_32.pcb32_r11;
stack_zero(st);
stack_capture(st, frame);
}
/*
* This function is called when we are unable to obtain a write lock on the
* first try. This means that at least one other thread holds either a
* read or write lock.
*/
void
__rw_wlock_hard(volatile uintptr_t *c, uintptr_t tid, const char *file,
int line)
{
struct rwlock *rw;
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
uintptr_t v, x;
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
#ifdef KDTRACE_HOOKS
uintptr_t state;
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
if (rw_wlocked(rw)) {
KASSERT(rw->lock_object.lo_flags & LO_RECURSABLE,
("%s: recursing but non-recursive rw %s @ %s:%d\n",
__func__, rw->lock_object.lo_name, file, line));
rw->rw_recurse++;
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p recursing", __func__, rw);
return;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__,
rw->lock_object.lo_name, (void *)rw->rw_lock, file, line);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&rw->lock_object);
state = rw->rw_lock;
#endif
while (!_rw_write_lock(rw, tid)) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&rw->lock_object,
&contested, &waittime);
#ifdef ADAPTIVE_RWLOCKS
/*
* If the lock is write locked and the owner is
* running on another CPU, spin until the owner stops
* running or the state of the lock changes.
*/
v = rw->rw_lock;
owner = (struct thread *)RW_OWNER(v);
if (!(v & RW_LOCK_READ) && TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK, "%s: spinning on %p held by %p",
__func__, rw, owner);
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
while ((struct thread*)RW_OWNER(rw->rw_lock) == owner &&
TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
continue;
}
if ((v & RW_LOCK_READ) && RW_READERS(v) &&
spintries < rowner_retries) {
if (!(v & RW_LOCK_WRITE_SPINNER)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_SPINNER)) {
continue;
}
}
spintries++;
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
for (i = 0; i < rowner_loops; i++) {
if ((rw->rw_lock & RW_LOCK_WRITE_SPINNER) == 0)
break;
cpu_spinwait();
}
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
#ifdef KDTRACE_HOOKS
spin_cnt += rowner_loops - i;
#endif
if (i != rowner_loops)
continue;
}
#endif
ts = turnstile_trywait(&rw->lock_object);
v = rw->rw_lock;
#ifdef ADAPTIVE_RWLOCKS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if (!(v & RW_LOCK_READ)) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* Check for the waiters flags about this rwlock.
* If the lock was released, without maintain any pending
* waiters queue, simply try to acquire it.
* If a pending waiters queue is present, claim the lock
* ownership and maintain the pending queue.
*/
x = v & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER);
if ((v & ~x) == RW_UNLOCKED) {
x &= ~RW_LOCK_WRITE_SPINNER;
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v, tid | x)) {
if (x)
turnstile_claim(ts);
else
turnstile_cancel(ts);
break;
}
turnstile_cancel(ts);
continue;
}
/*
* If the RW_LOCK_WRITE_WAITERS flag isn't set, then try to
* set it. If we fail to set it, then loop back and try
* again.
*/
if (!(v & RW_LOCK_WRITE_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set write waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the write waiters
* flag is set, so we must block on the turnstile.
*/
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__,
rw);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&rw->lock_object);
#endif
turnstile_wait(ts, rw_owner(rw), TS_EXCLUSIVE_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&rw->lock_object);
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
#ifdef ADAPTIVE_RWLOCKS
spintries = 0;
#endif
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&rw->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(rw__block, rw, sleep_time,
LOCKSTAT_WRITER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
/* Record only the loops spinning and not sleeping. */
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD4(rw__spin, rw, all_time - sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
#endif
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested,
waittime, file, line, LOCKSTAT_WRITER);
}
void
__rw_rlock(volatile uintptr_t *c, const char *file, int line)
{
struct rwlock *rw;
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
uintptr_t v;
#ifdef KDTRACE_HOOKS
uintptr_t state;
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
("rw_rlock() by idle thread %p on rwlock %s @ %s:%d",
curthread, rw->lock_object.lo_name, file, line));
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_rlock() of destroyed rwlock @ %s:%d", file, line));
KASSERT(rw_wowner(rw) != curthread,
("rw_rlock: wlock already held for %s @ %s:%d",
rw->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER, file, line, NULL);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&rw->lock_object);
state = rw->rw_lock;
#endif
for (;;) {
/*
* Handle the easy case. If no other thread has a write
* lock, then try to bump up the count of read locks. Note
* that we have to preserve the current state of the
* RW_LOCK_WRITE_WAITERS flag. If we fail to acquire a
* read lock, then rw_lock must have changed, so restart
* the loop. Note that this handles the case of a
* completely unlocked rwlock since such a lock is encoded
* as a read lock with no waiters.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
/*
* The RW_LOCK_READ_WAITERS flag should only be set
* if the lock has been unlocked and write waiters
* were present.
*/
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v,
v + RW_ONE_READER)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR4(KTR_LOCK,
"%s: %p succeed %p -> %p", __func__,
rw, (void *)v,
(void *)(v + RW_ONE_READER));
break;
}
continue;
}
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&rw->lock_object,
&contested, &waittime);
#ifdef ADAPTIVE_RWLOCKS
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, rw, owner);
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname(curthread), "spinning",
"lockname:\"%s\"", rw->lock_object.lo_name);
while ((struct thread*)RW_OWNER(rw->rw_lock) ==
owner && TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname(curthread), "running");
continue;
}
} else if (spintries < rowner_retries) {
spintries++;
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
for (i = 0; i < rowner_loops; i++) {
v = rw->rw_lock;
if ((v & RW_LOCK_READ) == 0 || RW_CAN_READ(v))
break;
cpu_spinwait();
}
#ifdef KDTRACE_HOOKS
spin_cnt += rowner_loops - i;
#endif
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
if (i != rowner_loops)
continue;
}
#endif
/*
* Okay, now it's the hard case. Some other thread already
* has a write lock or there are write waiters present,
* acquire the turnstile lock so we can begin the process
* of blocking.
*/
ts = turnstile_trywait(&rw->lock_object);
/*
* The lock might have been released while we spun, so
* recheck its state and restart the loop if needed.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
turnstile_cancel(ts);
continue;
}
#ifdef ADAPTIVE_RWLOCKS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* The lock is held in write mode or it already has waiters.
*/
MPASS(!RW_CAN_READ(v));
/*
* If the RW_LOCK_READ_WAITERS flag is already set, then
* we can go ahead and block. If it is not set then try
* to set it. If we fail to set it drop the turnstile
* lock and restart the loop.
*/
if (!(v & RW_LOCK_READ_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_READ_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set read waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the read waiters
* flag is set, so we must block on the turnstile.
*/
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__,
rw);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&rw->lock_object);
#endif
turnstile_wait(ts, rw_owner(rw), TS_SHARED_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&rw->lock_object);
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&rw->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(rw__block, rw, sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
/* Record only the loops spinning and not sleeping. */
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD4(rw__spin, rw, all_time - sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
#endif
/*
* TODO: acquire "owner of record" here. Here be turnstile dragons
* however. turnstiles don't like owners changing between calls to
* turnstile_wait() currently.
*/
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested,
waittime, file, line, LOCKSTAT_READER);
LOCK_LOG_LOCK("RLOCK", &rw->lock_object, 0, 0, file, line);
WITNESS_LOCK(&rw->lock_object, 0, file, line);
curthread->td_locks++;
curthread->td_rw_rlocks++;
}
/*
* This function represents the so-called 'hard case' for sx_slock
* operation. All 'easy case' failures are redirected to this. Note
* that ideally this would be a static function, but it needs to be
* accessible from at least sx.h.
*/
int
_sx_slock_hard(struct sx *sx, int opts, const char *file, int line)
{
GIANT_DECLARE;
#ifdef ADAPTIVE_SX
volatile struct thread *owner;
#endif
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
uintptr_t x;
int error = 0;
#ifdef KDTRACE_HOOKS
uintptr_t state;
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return (0);
#ifdef KDTRACE_HOOKS
state = sx->sx_lock;
all_time -= lockstat_nsecs(&sx->lock_object);
#endif
/*
* As with rwlocks, we don't make any attempt to try to block
* shared locks once there is an exclusive waiter.
*/
for (;;) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
x = sx->sx_lock;
/*
* If no other thread has an exclusive lock then try to bump up
* the count of sharers. Since we have to preserve the state
* of SX_LOCK_EXCLUSIVE_WAITERS, if we fail to acquire the
* shared lock loop back and retry.
*/
if (x & SX_LOCK_SHARED) {
MPASS(!(x & SX_LOCK_SHARED_WAITERS));
if (atomic_cmpset_acq_ptr(&sx->sx_lock, x,
x + SX_ONE_SHARER)) {
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR4(KTR_LOCK,
"%s: %p succeed %p -> %p", __func__,
sx, (void *)x,
(void *)(x + SX_ONE_SHARER));
break;
}
continue;
}
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&sx->lock_object, &contested,
&waittime);
#ifdef ADAPTIVE_SX
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
*/
if ((sx->lock_object.lo_flags & SX_NOADAPTIVE) == 0) {
x = SX_OWNER(x);
owner = (struct thread *)x;
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, sx, owner);
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname(curthread), "spinning",
"lockname:\"%s\"", sx->lock_object.lo_name);
GIANT_SAVE();
while (SX_OWNER(sx->sx_lock) == x &&
TD_IS_RUNNING(owner)) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
cpu_spinwait();
}
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname(curthread), "running");
continue;
}
}
#endif
/*
* Some other thread already has an exclusive lock, so
* start the process of blocking.
*/
sleepq_lock(&sx->lock_object);
x = sx->sx_lock;
/*
* The lock could have been released while we spun.
* In this case loop back and retry.
*/
if (x & SX_LOCK_SHARED) {
sleepq_release(&sx->lock_object);
continue;
}
#ifdef ADAPTIVE_SX
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
*/
if (!(x & SX_LOCK_SHARED) &&
(sx->lock_object.lo_flags & SX_NOADAPTIVE) == 0) {
owner = (struct thread *)SX_OWNER(x);
if (TD_IS_RUNNING(owner)) {
sleepq_release(&sx->lock_object);
continue;
}
}
#endif
/*
* Try to set the SX_LOCK_SHARED_WAITERS flag. If we
* fail to set it drop the sleep queue lock and loop
* back.
*/
if (!(x & SX_LOCK_SHARED_WAITERS)) {
if (!atomic_cmpset_ptr(&sx->sx_lock, x,
x | SX_LOCK_SHARED_WAITERS)) {
sleepq_release(&sx->lock_object);
continue;
}
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set shared waiters flag",
__func__, sx);
}
/*
* Since we have been unable to acquire the shared lock,
* we have to sleep.
*/
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on sleep queue",
__func__, sx);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&sx->lock_object);
#endif
GIANT_SAVE();
sleepq_add(&sx->lock_object, NULL, sx->lock_object.lo_name,
SLEEPQ_SX | ((opts & SX_INTERRUPTIBLE) ?
SLEEPQ_INTERRUPTIBLE : 0), SQ_SHARED_QUEUE);
if (!(opts & SX_INTERRUPTIBLE))
sleepq_wait(&sx->lock_object, 0);
else
error = sleepq_wait_sig(&sx->lock_object, 0);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&sx->lock_object);
sleep_cnt++;
#endif
if (error) {
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK,
"%s: interruptible sleep by %p suspended by signal",
__func__, sx);
break;
}
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from sleep queue",
__func__, sx);
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&sx->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(sx__block, sx, sleep_time,
LOCKSTAT_READER, (state & SX_LOCK_SHARED) == 0,
(state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD4(sx__spin, sx, all_time - sleep_time,
LOCKSTAT_READER, (state & SX_LOCK_SHARED) == 0,
(state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
#endif
if (error == 0)
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx,
contested, waittime, file, line, LOCKSTAT_READER);
GIANT_RESTORE();
return (error);
}
/*
* __mtx_lock_sleep: the tougher part of acquiring an MTX_DEF lock.
*
* We call this if the lock is either contested (i.e. we need to go to
* sleep waiting for it), or if we need to recurse on it.
*/
void
__mtx_lock_sleep(volatile uintptr_t *c, uintptr_t tid, int opts,
const char *file, int line)
{
struct mtx *m;
struct turnstile *ts;
uintptr_t v;
#ifdef ADAPTIVE_MUTEXES
volatile struct thread *owner;
#endif
#ifdef KTR
int cont_logged = 0;
#endif
#ifdef LOCK_PROFILING
int contested = 0;
uint64_t waittime = 0;
#endif
#if defined(ADAPTIVE_MUTEXES) || defined(KDTRACE_HOOKS)
struct lock_delay_arg lda;
#endif
#ifdef KDTRACE_HOOKS
u_int sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
#if defined(ADAPTIVE_MUTEXES)
lock_delay_arg_init(&lda, &mtx_delay);
#elif defined(KDTRACE_HOOKS)
lock_delay_arg_init(&lda, NULL);
#endif
m = mtxlock2mtx(c);
if (mtx_owned(m)) {
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0 ||
(opts & MTX_RECURSE) != 0,
("_mtx_lock_sleep: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
opts &= ~MTX_RECURSE;
m->mtx_recurse++;
atomic_set_ptr(&m->mtx_lock, MTX_RECURSED);
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR1(KTR_LOCK, "_mtx_lock_sleep: %p recursing", m);
return;
}
opts &= ~MTX_RECURSE;
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&m->lock_object,
&contested, &waittime);
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR4(KTR_LOCK,
"_mtx_lock_sleep: %s contested (lock=%p) at %s:%d",
m->lock_object.lo_name, (void *)m->mtx_lock, file, line);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&m->lock_object);
#endif
for (;;) {
if (m->mtx_lock == MTX_UNOWNED && _mtx_obtain_lock(m, tid))
break;
#ifdef KDTRACE_HOOKS
lda.spin_cnt++;
#endif
#ifdef ADAPTIVE_MUTEXES
/*
* If the owner is running on another CPU, spin until the
* owner stops running or the state of the lock changes.
*/
v = m->mtx_lock;
if (v != MTX_UNOWNED) {
owner = (struct thread *)(v & ~MTX_FLAGMASK);
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&m->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, m, owner);
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname((struct thread *)tid),
"spinning", "lockname:\"%s\"",
m->lock_object.lo_name);
while (mtx_owner(m) == owner &&
TD_IS_RUNNING(owner))
lock_delay(&lda);
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname((struct thread *)tid),
"running");
continue;
}
}
#endif
ts = turnstile_trywait(&m->lock_object);
v = m->mtx_lock;
/*
* Check if the lock has been released while spinning for
* the turnstile chain lock.
*/
if (v == MTX_UNOWNED) {
turnstile_cancel(ts);
continue;
}
#ifdef ADAPTIVE_MUTEXES
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
owner = (struct thread *)(v & ~MTX_FLAGMASK);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
#endif
/*
* If the mutex isn't already contested and a failure occurs
* setting the contested bit, the mutex was either released
* or the state of the MTX_RECURSED bit changed.
*/
if ((v & MTX_CONTESTED) == 0 &&
!atomic_cmpset_ptr(&m->mtx_lock, v, v | MTX_CONTESTED)) {
turnstile_cancel(ts);
continue;
}
/*
* We definitely must sleep for this lock.
*/
mtx_assert(m, MA_NOTOWNED);
#ifdef KTR
if (!cont_logged) {
CTR6(KTR_CONTENTION,
"contention: %p at %s:%d wants %s, taken by %s:%d",
(void *)tid, file, line, m->lock_object.lo_name,
WITNESS_FILE(&m->lock_object),
WITNESS_LINE(&m->lock_object));
cont_logged = 1;
}
#endif
/*
* Block on the turnstile.
*/
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&m->lock_object);
#endif
turnstile_wait(ts, mtx_owner(m), TS_EXCLUSIVE_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&m->lock_object);
sleep_cnt++;
#endif
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&m->lock_object);
#endif
#ifdef KTR
if (cont_logged) {
CTR4(KTR_CONTENTION,
"contention end: %s acquired by %p at %s:%d",
m->lock_object.lo_name, (void *)tid, file, line);
}
#endif
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(adaptive__acquire, m, contested,
waittime, file, line);
#ifdef KDTRACE_HOOKS
if (sleep_time)
LOCKSTAT_RECORD1(adaptive__block, m, sleep_time);
/*
* Only record the loops spinning and not sleeping.
*/
if (lda.spin_cnt > sleep_cnt)
LOCKSTAT_RECORD1(adaptive__spin, m, all_time - sleep_time);
#endif
}
/*
* Original vm_pageout_oom, will be called if LRU pageout_oom will fail
*/
static void
original_vm_pageout_oom(int shortage)
{
struct proc *p, *bigproc;
vm_offset_t size, bigsize;
struct thread *td;
struct vmspace *vm;
/*
* We keep the process bigproc locked once we find it to keep anyone
* from messing with it; however, there is a possibility of
* deadlock if process B is bigproc and one of it's child processes
* attempts to propagate a signal to B while we are waiting for A's
* lock while walking this list. To avoid this, we don't block on
* the process lock but just skip a process if it is already locked.
*/
bigproc = NULL;
bigsize = 0;
sx_slock(&allproc_lock);
FOREACH_PROC_IN_SYSTEM(p) {
int breakout;
if (PROC_TRYLOCK(p) == 0)
continue;
/*
* If this is a system, protected or killed process, skip it.
*/
if (p->p_state != PRS_NORMAL ||
(p->p_flag & (P_INEXEC | P_PROTECTED | P_SYSTEM)) ||
(p->p_pid == 1) || P_KILLED(p) ||
((p->p_pid < 48) && (swap_pager_avail != 0))) {
PROC_UNLOCK(p);
continue;
}
/*
* If the process is in a non-running type state,
* don't touch it. Check all the threads individually.
*/
breakout = 0;
FOREACH_THREAD_IN_PROC(p, td) {
thread_lock(td);
if (!TD_ON_RUNQ(td) &&
!TD_IS_RUNNING(td) &&
!TD_IS_SLEEPING(td)) {
thread_unlock(td);
breakout = 1;
break;
}
thread_unlock(td);
}
if (breakout) {
PROC_UNLOCK(p);
continue;
}
/*
* get the process size
*/
vm = vmspace_acquire_ref(p);
if (vm == NULL) {
PROC_UNLOCK(p);
continue;
}
if (!vm_map_trylock_read(&vm->vm_map)) {
vmspace_free(vm);
PROC_UNLOCK(p);
continue;
}
size = vmspace_swap_count(vm);
vm_map_unlock_read(&vm->vm_map);
if (shortage == VM_OOM_MEM)
size += vmspace_resident_count(vm);
vmspace_free(vm);
/*
* if the this process is bigger than the biggest one
* remember it.
*/
if (size > bigsize) {
if (bigproc != NULL)
PROC_UNLOCK(bigproc);
bigproc = p;
bigsize = size;
} else
PROC_UNLOCK(p);
}
/*
* Read proc's from memory file into buffer bp, which has space to hold
* at most maxcnt procs.
*/
static int
kvm_proclist(kvm_t *kd, int what, int arg, struct proc *p,
struct kinfo_proc *bp, int maxcnt)
{
int cnt = 0;
struct kinfo_proc kinfo_proc, *kp;
struct pgrp pgrp;
struct session sess;
struct cdev t_cdev;
struct tty tty;
struct vmspace vmspace;
struct sigacts sigacts;
#if 0
struct pstats pstats;
#endif
struct ucred ucred;
struct prison pr;
struct thread mtd;
struct proc proc;
struct proc pproc;
struct sysentvec sysent;
char svname[KI_EMULNAMELEN];
kp = &kinfo_proc;
kp->ki_structsize = sizeof(kinfo_proc);
/*
* Loop on the processes. this is completely broken because we need to be
* able to loop on the threads and merge the ones that are the same process some how.
*/
for (; cnt < maxcnt && p != NULL; p = LIST_NEXT(&proc, p_list)) {
memset(kp, 0, sizeof *kp);
if (KREAD(kd, (u_long)p, &proc)) {
_kvm_err(kd, kd->program, "can't read proc at %p", p);
return (-1);
}
if (proc.p_state == PRS_NEW)
continue;
if (proc.p_state != PRS_ZOMBIE) {
if (KREAD(kd, (u_long)TAILQ_FIRST(&proc.p_threads),
&mtd)) {
_kvm_err(kd, kd->program,
"can't read thread at %p",
TAILQ_FIRST(&proc.p_threads));
return (-1);
}
}
if (KREAD(kd, (u_long)proc.p_ucred, &ucred) == 0) {
kp->ki_ruid = ucred.cr_ruid;
kp->ki_svuid = ucred.cr_svuid;
kp->ki_rgid = ucred.cr_rgid;
kp->ki_svgid = ucred.cr_svgid;
kp->ki_cr_flags = ucred.cr_flags;
if (ucred.cr_ngroups > KI_NGROUPS) {
kp->ki_ngroups = KI_NGROUPS;
kp->ki_cr_flags |= KI_CRF_GRP_OVERFLOW;
} else
kp->ki_ngroups = ucred.cr_ngroups;
kvm_read(kd, (u_long)ucred.cr_groups, kp->ki_groups,
kp->ki_ngroups * sizeof(gid_t));
kp->ki_uid = ucred.cr_uid;
if (ucred.cr_prison != NULL) {
if (KREAD(kd, (u_long)ucred.cr_prison, &pr)) {
_kvm_err(kd, kd->program,
"can't read prison at %p",
ucred.cr_prison);
return (-1);
}
kp->ki_jid = pr.pr_id;
}
}
switch(what & ~KERN_PROC_INC_THREAD) {
case KERN_PROC_GID:
if (kp->ki_groups[0] != (gid_t)arg)
continue;
break;
case KERN_PROC_PID:
if (proc.p_pid != (pid_t)arg)
continue;
break;
case KERN_PROC_RGID:
if (kp->ki_rgid != (gid_t)arg)
continue;
break;
case KERN_PROC_UID:
if (kp->ki_uid != (uid_t)arg)
continue;
break;
case KERN_PROC_RUID:
if (kp->ki_ruid != (uid_t)arg)
continue;
break;
}
/*
* We're going to add another proc to the set. If this
* will overflow the buffer, assume the reason is because
* nprocs (or the proc list) is corrupt and declare an error.
*/
if (cnt >= maxcnt) {
_kvm_err(kd, kd->program, "nprocs corrupt");
return (-1);
}
/*
* gather kinfo_proc
*/
kp->ki_paddr = p;
kp->ki_addr = 0; /* XXX uarea */
/* kp->ki_kstack = proc.p_thread.td_kstack; XXXKSE */
kp->ki_args = proc.p_args;
kp->ki_tracep = proc.p_tracevp;
kp->ki_textvp = proc.p_textvp;
kp->ki_fd = proc.p_fd;
kp->ki_vmspace = proc.p_vmspace;
if (proc.p_sigacts != NULL) {
if (KREAD(kd, (u_long)proc.p_sigacts, &sigacts)) {
_kvm_err(kd, kd->program,
"can't read sigacts at %p", proc.p_sigacts);
return (-1);
}
kp->ki_sigignore = sigacts.ps_sigignore;
kp->ki_sigcatch = sigacts.ps_sigcatch;
}
#if 0
if ((proc.p_flag & P_INMEM) && proc.p_stats != NULL) {
if (KREAD(kd, (u_long)proc.p_stats, &pstats)) {
_kvm_err(kd, kd->program,
"can't read stats at %x", proc.p_stats);
return (-1);
}
kp->ki_start = pstats.p_start;
/*
* XXX: The times here are probably zero and need
* to be calculated from the raw data in p_rux and
* p_crux.
*/
kp->ki_rusage = pstats.p_ru;
kp->ki_childstime = pstats.p_cru.ru_stime;
kp->ki_childutime = pstats.p_cru.ru_utime;
/* Some callers want child-times in a single value */
timeradd(&kp->ki_childstime, &kp->ki_childutime,
&kp->ki_childtime);
}
#endif
if (proc.p_oppid)
kp->ki_ppid = proc.p_oppid;
else if (proc.p_pptr) {
if (KREAD(kd, (u_long)proc.p_pptr, &pproc)) {
_kvm_err(kd, kd->program,
"can't read pproc at %p", proc.p_pptr);
return (-1);
}
kp->ki_ppid = pproc.p_pid;
} else
kp->ki_ppid = 0;
if (proc.p_pgrp == NULL)
goto nopgrp;
if (KREAD(kd, (u_long)proc.p_pgrp, &pgrp)) {
_kvm_err(kd, kd->program, "can't read pgrp at %p",
proc.p_pgrp);
return (-1);
}
kp->ki_pgid = pgrp.pg_id;
kp->ki_jobc = pgrp.pg_jobc;
if (KREAD(kd, (u_long)pgrp.pg_session, &sess)) {
_kvm_err(kd, kd->program, "can't read session at %p",
pgrp.pg_session);
return (-1);
}
kp->ki_sid = sess.s_sid;
(void)memcpy(kp->ki_login, sess.s_login,
sizeof(kp->ki_login));
kp->ki_kiflag = sess.s_ttyvp ? KI_CTTY : 0;
if (sess.s_leader == p)
kp->ki_kiflag |= KI_SLEADER;
if ((proc.p_flag & P_CONTROLT) && sess.s_ttyp != NULL) {
if (KREAD(kd, (u_long)sess.s_ttyp, &tty)) {
_kvm_err(kd, kd->program,
"can't read tty at %p", sess.s_ttyp);
return (-1);
}
if (tty.t_dev != NULL) {
if (KREAD(kd, (u_long)tty.t_dev, &t_cdev)) {
_kvm_err(kd, kd->program,
"can't read cdev at %p",
tty.t_dev);
return (-1);
}
#if 0
kp->ki_tdev = t_cdev.si_udev;
#else
kp->ki_tdev = NODEV;
#endif
}
if (tty.t_pgrp != NULL) {
if (KREAD(kd, (u_long)tty.t_pgrp, &pgrp)) {
_kvm_err(kd, kd->program,
"can't read tpgrp at %p",
tty.t_pgrp);
return (-1);
}
kp->ki_tpgid = pgrp.pg_id;
} else
kp->ki_tpgid = -1;
if (tty.t_session != NULL) {
if (KREAD(kd, (u_long)tty.t_session, &sess)) {
_kvm_err(kd, kd->program,
"can't read session at %p",
tty.t_session);
return (-1);
}
kp->ki_tsid = sess.s_sid;
}
} else {
nopgrp:
kp->ki_tdev = NODEV;
}
if ((proc.p_state != PRS_ZOMBIE) && mtd.td_wmesg)
(void)kvm_read(kd, (u_long)mtd.td_wmesg,
kp->ki_wmesg, WMESGLEN);
(void)kvm_read(kd, (u_long)proc.p_vmspace,
(char *)&vmspace, sizeof(vmspace));
kp->ki_size = vmspace.vm_map.size;
/*
* Approximate the kernel's method of calculating
* this field.
*/
#define pmap_resident_count(pm) ((pm)->pm_stats.resident_count)
kp->ki_rssize = pmap_resident_count(&vmspace.vm_pmap);
kp->ki_swrss = vmspace.vm_swrss;
kp->ki_tsize = vmspace.vm_tsize;
kp->ki_dsize = vmspace.vm_dsize;
kp->ki_ssize = vmspace.vm_ssize;
switch (what & ~KERN_PROC_INC_THREAD) {
case KERN_PROC_PGRP:
if (kp->ki_pgid != (pid_t)arg)
continue;
break;
case KERN_PROC_SESSION:
if (kp->ki_sid != (pid_t)arg)
continue;
break;
case KERN_PROC_TTY:
if ((proc.p_flag & P_CONTROLT) == 0 ||
kp->ki_tdev != (dev_t)arg)
continue;
break;
}
if (proc.p_comm[0] != 0)
strlcpy(kp->ki_comm, proc.p_comm, MAXCOMLEN);
(void)kvm_read(kd, (u_long)proc.p_sysent, (char *)&sysent,
sizeof(sysent));
(void)kvm_read(kd, (u_long)sysent.sv_name, (char *)&svname,
sizeof(svname));
if (svname[0] != 0)
strlcpy(kp->ki_emul, svname, KI_EMULNAMELEN);
if ((proc.p_state != PRS_ZOMBIE) &&
(mtd.td_blocked != 0)) {
kp->ki_kiflag |= KI_LOCKBLOCK;
if (mtd.td_lockname)
(void)kvm_read(kd,
(u_long)mtd.td_lockname,
kp->ki_lockname, LOCKNAMELEN);
kp->ki_lockname[LOCKNAMELEN] = 0;
}
kp->ki_runtime = cputick2usec(proc.p_rux.rux_runtime);
kp->ki_pid = proc.p_pid;
kp->ki_siglist = proc.p_siglist;
SIGSETOR(kp->ki_siglist, mtd.td_siglist);
kp->ki_sigmask = mtd.td_sigmask;
kp->ki_xstat = KW_EXITCODE(proc.p_xexit, proc.p_xsig);
kp->ki_acflag = proc.p_acflag;
kp->ki_lock = proc.p_lock;
if (proc.p_state != PRS_ZOMBIE) {
kp->ki_swtime = (ticks - proc.p_swtick) / hz;
kp->ki_flag = proc.p_flag;
kp->ki_sflag = 0;
kp->ki_nice = proc.p_nice;
kp->ki_traceflag = proc.p_traceflag;
if (proc.p_state == PRS_NORMAL) {
if (TD_ON_RUNQ(&mtd) ||
TD_CAN_RUN(&mtd) ||
TD_IS_RUNNING(&mtd)) {
kp->ki_stat = SRUN;
} else if (mtd.td_state ==
TDS_INHIBITED) {
if (P_SHOULDSTOP(&proc)) {
kp->ki_stat = SSTOP;
} else if (
TD_IS_SLEEPING(&mtd)) {
kp->ki_stat = SSLEEP;
} else if (TD_ON_LOCK(&mtd)) {
kp->ki_stat = SLOCK;
} else {
kp->ki_stat = SWAIT;
}
}
} else {
kp->ki_stat = SIDL;
}
/* Stuff from the thread */
kp->ki_pri.pri_level = mtd.td_priority;
kp->ki_pri.pri_native = mtd.td_base_pri;
kp->ki_lastcpu = mtd.td_lastcpu;
kp->ki_wchan = mtd.td_wchan;
kp->ki_oncpu = mtd.td_oncpu;
if (mtd.td_name[0] != '\0')
strlcpy(kp->ki_tdname, mtd.td_name, sizeof(kp->ki_tdname));
kp->ki_pctcpu = 0;
kp->ki_rqindex = 0;
/*
* Note: legacy fields; wraps at NO_CPU_OLD or the
* old max CPU value as appropriate
*/
if (mtd.td_lastcpu == NOCPU)
kp->ki_lastcpu_old = NOCPU_OLD;
else if (mtd.td_lastcpu > MAXCPU_OLD)
kp->ki_lastcpu_old = MAXCPU_OLD;
else
kp->ki_lastcpu_old = mtd.td_lastcpu;
if (mtd.td_oncpu == NOCPU)
kp->ki_oncpu_old = NOCPU_OLD;
else if (mtd.td_oncpu > MAXCPU_OLD)
kp->ki_oncpu_old = MAXCPU_OLD;
else
kp->ki_oncpu_old = mtd.td_oncpu;
} else {
kp->ki_stat = SZOMB;
}
kp->ki_tdev_freebsd11 = kp->ki_tdev; /* truncate */
bcopy(&kinfo_proc, bp, sizeof(kinfo_proc));
++bp;
++cnt;
}
return (cnt);
}
/*
* _mtx_lock_sleep: the tougher part of acquiring an MTX_DEF lock.
*
* We call this if the lock is either contested (i.e. we need to go to
* sleep waiting for it), or if we need to recurse on it.
*/
void
_mtx_lock_sleep(struct mtx *m, uintptr_t tid, int opts, const char *file,
int line)
{
struct turnstile *ts;
uintptr_t v;
#ifdef ADAPTIVE_MUTEXES
volatile struct thread *owner;
#endif
#ifdef KTR
int cont_logged = 0;
#endif
#ifdef LOCK_PROFILING
int contested = 0;
uint64_t waittime = 0;
#endif
#ifdef KDTRACE_HOOKS
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
#endif
if (mtx_owned(m)) {
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
("_mtx_lock_sleep: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
m->mtx_recurse++;
atomic_set_ptr(&m->mtx_lock, MTX_RECURSED);
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR1(KTR_LOCK, "_mtx_lock_sleep: %p recursing", m);
return;
}
lock_profile_obtain_lock_failed(&m->lock_object,
&contested, &waittime);
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR4(KTR_LOCK,
"_mtx_lock_sleep: %s contested (lock=%p) at %s:%d",
m->lock_object.lo_name, (void *)m->mtx_lock, file, line);
while (!_obtain_lock(m, tid)) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
#ifdef ADAPTIVE_MUTEXES
/*
* If the owner is running on another CPU, spin until the
* owner stops running or the state of the lock changes.
*/
v = m->mtx_lock;
if (v != MTX_UNOWNED) {
owner = (struct thread *)(v & ~MTX_FLAGMASK);
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&m->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, m, owner);
while (mtx_owner(m) == owner &&
TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
continue;
}
}
#endif
ts = turnstile_trywait(&m->lock_object);
v = m->mtx_lock;
/*
* Check if the lock has been released while spinning for
* the turnstile chain lock.
*/
if (v == MTX_UNOWNED) {
turnstile_cancel(ts);
continue;
}
#ifdef ADAPTIVE_MUTEXES
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
owner = (struct thread *)(v & ~MTX_FLAGMASK);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
#endif
/*
* If the mutex isn't already contested and a failure occurs
* setting the contested bit, the mutex was either released
* or the state of the MTX_RECURSED bit changed.
*/
if ((v & MTX_CONTESTED) == 0 &&
!atomic_cmpset_ptr(&m->mtx_lock, v, v | MTX_CONTESTED)) {
turnstile_cancel(ts);
continue;
}
/*
* We definitely must sleep for this lock.
*/
mtx_assert(m, MA_NOTOWNED);
#ifdef KTR
if (!cont_logged) {
CTR6(KTR_CONTENTION,
"contention: %p at %s:%d wants %s, taken by %s:%d",
(void *)tid, file, line, m->lock_object.lo_name,
WITNESS_FILE(&m->lock_object),
WITNESS_LINE(&m->lock_object));
cont_logged = 1;
}
#endif
/*
* Block on the turnstile.
*/
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs();
#endif
turnstile_wait(ts, mtx_owner(m), TS_EXCLUSIVE_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs();
sleep_cnt++;
#endif
}
#ifdef KTR
if (cont_logged) {
CTR4(KTR_CONTENTION,
"contention end: %s acquired by %p at %s:%d",
m->lock_object.lo_name, (void *)tid, file, line);
}
#endif
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_LOCK_ACQUIRE, m, contested,
waittime, file, line);
#ifdef KDTRACE_HOOKS
if (sleep_time)
LOCKSTAT_RECORD1(LS_MTX_LOCK_BLOCK, m, sleep_time);
/*
* Only record the loops spinning and not sleeping.
*/
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD1(LS_MTX_LOCK_SPIN, m, (spin_cnt - sleep_cnt));
#endif
}
/* ARGSUSED */
static void
schedcpu(void)
{
register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
struct thread *td;
struct proc *p;
struct td_sched *ts;
int awake;
sx_slock(&allproc_lock);
FOREACH_PROC_IN_SYSTEM(p) {
PROC_LOCK(p);
if (p->p_state == PRS_NEW) {
PROC_UNLOCK(p);
continue;
}
FOREACH_THREAD_IN_PROC(p, td) {
awake = 0;
thread_lock(td);
ts = td->td_sched;
/*
* Increment sleep time (if sleeping). We
* ignore overflow, as above.
*/
/*
* The td_sched slptimes are not touched in wakeup
* because the thread may not HAVE everything in
* memory? XXX I think this is out of date.
*/
if (TD_ON_RUNQ(td)) {
awake = 1;
td->td_flags &= ~TDF_DIDRUN;
} else if (TD_IS_RUNNING(td)) {
awake = 1;
/* Do not clear TDF_DIDRUN */
} else if (td->td_flags & TDF_DIDRUN) {
awake = 1;
td->td_flags &= ~TDF_DIDRUN;
}
/*
* ts_pctcpu is only for ps and ttyinfo().
*/
ts->ts_pctcpu = (ts->ts_pctcpu * ccpu) >> FSHIFT;
/*
* If the td_sched has been idle the entire second,
* stop recalculating its priority until
* it wakes up.
*/
if (ts->ts_cpticks != 0) {
#if (FSHIFT >= CCPU_SHIFT)
ts->ts_pctcpu += (realstathz == 100)
? ((fixpt_t) ts->ts_cpticks) <<
(FSHIFT - CCPU_SHIFT) :
100 * (((fixpt_t) ts->ts_cpticks)
<< (FSHIFT - CCPU_SHIFT)) / realstathz;
#else
ts->ts_pctcpu += ((FSCALE - ccpu) *
(ts->ts_cpticks *
FSCALE / realstathz)) >> FSHIFT;
#endif
ts->ts_cpticks = 0;
}
/*
* If there are ANY running threads in this process,
* then don't count it as sleeping.
* XXX: this is broken.
*/
if (awake) {
if (ts->ts_slptime > 1) {
/*
* In an ideal world, this should not
* happen, because whoever woke us
* up from the long sleep should have
* unwound the slptime and reset our
* priority before we run at the stale
* priority. Should KASSERT at some
* point when all the cases are fixed.
*/
updatepri(td);
}
ts->ts_slptime = 0;
} else
ts->ts_slptime++;
if (ts->ts_slptime > 1) {
thread_unlock(td);
continue;
}
td->td_estcpu = decay_cpu(loadfac, td->td_estcpu);
resetpriority(td);
resetpriority_thread(td);
thread_unlock(td);
}
/*
* This function represents the so-called 'hard case' for sx_xlock
* operation. All 'easy case' failures are redirected to this. Note
* that ideally this would be a static function, but it needs to be
* accessible from at least sx.h.
*/
int
_sx_xlock_hard(struct sx *sx, uintptr_t tid, int opts, const char *file,
int line)
{
GIANT_DECLARE;
#ifdef ADAPTIVE_SX
volatile struct thread *owner;
u_int i, spintries = 0;
#endif
uintptr_t x;
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
int error = 0;
#ifdef KDTRACE_HOOKS
uintptr_t state;
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return (0);
/* If we already hold an exclusive lock, then recurse. */
if (sx_xlocked(sx)) {
KASSERT((sx->lock_object.lo_flags & LO_RECURSABLE) != 0,
("_sx_xlock_hard: recursed on non-recursive sx %s @ %s:%d\n",
sx->lock_object.lo_name, file, line));
sx->sx_recurse++;
atomic_set_ptr(&sx->sx_lock, SX_LOCK_RECURSED);
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p recursing", __func__, sx);
return (0);
}
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__,
sx->lock_object.lo_name, (void *)sx->sx_lock, file, line);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&sx->lock_object);
state = sx->sx_lock;
#endif
while (!atomic_cmpset_acq_ptr(&sx->sx_lock, SX_LOCK_UNLOCKED, tid)) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&sx->lock_object, &contested,
&waittime);
#ifdef ADAPTIVE_SX
/*
* If the lock is write locked and the owner is
* running on another CPU, spin until the owner stops
* running or the state of the lock changes.
*/
x = sx->sx_lock;
if ((sx->lock_object.lo_flags & SX_NOADAPTIVE) == 0) {
if ((x & SX_LOCK_SHARED) == 0) {
x = SX_OWNER(x);
owner = (struct thread *)x;
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, sx, owner);
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname(curthread), "spinning",
"lockname:\"%s\"",
sx->lock_object.lo_name);
GIANT_SAVE();
while (SX_OWNER(sx->sx_lock) == x &&
TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname(curthread), "running");
continue;
}
} else if (SX_SHARERS(x) && spintries < asx_retries) {
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname(curthread), "spinning",
"lockname:\"%s\"", sx->lock_object.lo_name);
GIANT_SAVE();
spintries++;
for (i = 0; i < asx_loops; i++) {
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR4(KTR_LOCK,
"%s: shared spinning on %p with %u and %u",
__func__, sx, spintries, i);
x = sx->sx_lock;
if ((x & SX_LOCK_SHARED) == 0 ||
SX_SHARERS(x) == 0)
break;
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname(curthread), "running");
if (i != asx_loops)
continue;
}
}
#endif
sleepq_lock(&sx->lock_object);
x = sx->sx_lock;
/*
* If the lock was released while spinning on the
* sleep queue chain lock, try again.
*/
if (x == SX_LOCK_UNLOCKED) {
sleepq_release(&sx->lock_object);
continue;
}
#ifdef ADAPTIVE_SX
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the sleep queue
* chain lock. If so, drop the sleep queue lock and try
* again.
*/
if (!(x & SX_LOCK_SHARED) &&
(sx->lock_object.lo_flags & SX_NOADAPTIVE) == 0) {
owner = (struct thread *)SX_OWNER(x);
if (TD_IS_RUNNING(owner)) {
sleepq_release(&sx->lock_object);
continue;
}
}
#endif
/*
* If an exclusive lock was released with both shared
* and exclusive waiters and a shared waiter hasn't
* woken up and acquired the lock yet, sx_lock will be
* set to SX_LOCK_UNLOCKED | SX_LOCK_EXCLUSIVE_WAITERS.
* If we see that value, try to acquire it once. Note
* that we have to preserve SX_LOCK_EXCLUSIVE_WAITERS
* as there are other exclusive waiters still. If we
* fail, restart the loop.
*/
if (x == (SX_LOCK_UNLOCKED | SX_LOCK_EXCLUSIVE_WAITERS)) {
if (atomic_cmpset_acq_ptr(&sx->sx_lock,
SX_LOCK_UNLOCKED | SX_LOCK_EXCLUSIVE_WAITERS,
tid | SX_LOCK_EXCLUSIVE_WAITERS)) {
sleepq_release(&sx->lock_object);
CTR2(KTR_LOCK, "%s: %p claimed by new writer",
__func__, sx);
break;
}
sleepq_release(&sx->lock_object);
continue;
}
/*
* Try to set the SX_LOCK_EXCLUSIVE_WAITERS. If we fail,
* than loop back and retry.
*/
if (!(x & SX_LOCK_EXCLUSIVE_WAITERS)) {
if (!atomic_cmpset_ptr(&sx->sx_lock, x,
x | SX_LOCK_EXCLUSIVE_WAITERS)) {
sleepq_release(&sx->lock_object);
continue;
}
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set excl waiters flag",
__func__, sx);
}
/*
* Since we have been unable to acquire the exclusive
* lock and the exclusive waiters flag is set, we have
* to sleep.
*/
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on sleep queue",
__func__, sx);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&sx->lock_object);
#endif
GIANT_SAVE();
sleepq_add(&sx->lock_object, NULL, sx->lock_object.lo_name,
SLEEPQ_SX | ((opts & SX_INTERRUPTIBLE) ?
SLEEPQ_INTERRUPTIBLE : 0), SQ_EXCLUSIVE_QUEUE);
if (!(opts & SX_INTERRUPTIBLE))
sleepq_wait(&sx->lock_object, 0);
else
error = sleepq_wait_sig(&sx->lock_object, 0);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&sx->lock_object);
sleep_cnt++;
#endif
if (error) {
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK,
"%s: interruptible sleep by %p suspended by signal",
__func__, sx);
break;
}
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from sleep queue",
__func__, sx);
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&sx->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(sx__block, sx, sleep_time,
LOCKSTAT_WRITER, (state & SX_LOCK_SHARED) == 0,
(state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD4(sx__spin, sx, all_time - sleep_time,
LOCKSTAT_WRITER, (state & SX_LOCK_SHARED) == 0,
(state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
#endif
if (!error)
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx,
contested, waittime, file, line, LOCKSTAT_WRITER);
GIANT_RESTORE();
return (error);
}
