Specifically, sleepq_broadcast() uses td_slpq for its private pending
queue of threads that it is going to wake up after it takes them off the
sleep queue. The problem is that if one of the threads is actually not
asleep yet, then we can end up with td_slpq being corrupted and/or the
thread being made runnable at the wrong time resulting in the td_sleepqueue
== NULL assertion failures occasionally reported under heavy load.
The fix is to stop being so fancy and ditch the whole pending queue bit.
Instead, sleepq_remove_thread() and sleepq_resume_thread() were merged
into one function that requires the caller to hold sched_lock. This
fixes several places that unlocked sched_lock only to call a function
that then locked sched_lock, so even though sched_lock is now held
slightly longer, removing the extra lock acquires (1 pair instead of 3
in some cases) probably makes it an overall win if you don't include the
fact that it closes a race. This is definitely a 5.4 candidate.
PR: kern/79693
Submitted by: Steven Sears stevenjsears at yahoo dot com
MFC after: 4 days
priority. The sleep queues don't get updated when the priority of
threads changes, so sleepq_signal() might not always wakeup the
highest priority thread. Updating the queues when thread priorities
change cannot be easily done due to lock orders, so instead we do an
O(n) walk of the queue for a sleepq_signal() operation instead of O(1).
On the other hand, adding a thread to a sleep queue now goes from O(n)
to O(1) so it ends up as an even tradeoff. The correctness here with
regards to priorities is actually fairly important. msleep() gives
interactive threads their priority "boost" after they are placed on the
queue, but before this fix that "boost" wasn't used to determine the
highest priority thread that sleepq_signal() awoke.
- Fix up some comments.
Inspired by: ups, bde
- Add a new _lock() call to each API that locks the associated chain lock
for a lock_object pointer or wait channel. The _lookup() functions now
require that the chain lock be locked via _lock() when they are called.
- Change sleepq_add(), turnstile_wait() and turnstile_claim() to lookup
the associated queue structure internally via _lookup() rather than
accepting a pointer from the caller. For turnstiles, this means that
the actual lookup of the turnstile in the hash table is only done when
the thread actually blocks rather than being done on each loop iteration
in _mtx_lock_sleep(). For sleep queues, this means that sleepq_lookup()
is no longer used outside of the sleep queue code except to implement an
assertion in cv_destroy().
- Change sleepq_broadcast() and sleepq_signal() to require that the chain
lock is already required. For condition variables, this lets the
cv_broadcast() and cv_signal() functions lock the sleep queue chain lock
while testing the waiters count. This means that the waiters count
internal to condition variables is no longer protected by the interlock
mutex and cv_broadcast() and cv_signal() now no longer require that the
interlock be held when they are called. This lets consumers of condition
variables drop the lock before waking other threads which can result in
fewer context switches.
MFC after: 1 month
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.
switch to. If a non-NULL thread pointer is passed in, then the CPU will
switch to that thread directly rather than calling choosethread() to pick
a thread to choose to.
- Make sched_switch() aware of idle threads and know to do
TD_SET_CAN_RUN() instead of sticking them on the run queue rather than
requiring all callers of mi_switch() to know to do this if they can be
called from an idlethread.
- Move constants for arguments to mi_switch() and thread_single() out of
the middle of the function prototypes and up above into their own
section.
hash tables used in the sleep queue and turnstile code. Each option adds
a sysctl tree under debug containing the maximum depth of any bucket in
the hash table as well as a separate node for each bucket (or chain)
containing the current depth and maximum depth for that bucket.
is "void *" (it isn't) or that the default promotion of pid_t is int.
Instead, assume that casting "struct foo *" to "void *" and printing the
result with %p is useful, and that all pid_t's are representable as longs.
Fixed some minor style bugs (mainly spelling errors in comments).
removes a specific thread from a sleep queue. sleepq_resume_thread()
resumes scheduling of a thread that has been previously removed from a
sleep queue.
- sleepq_catch_signals() just removes a thread from the queue it was just
added to when a pending signal is found.
- sleepq_signal() and sleepq_broadcast() remove threads from a queue,
drop the queue lock, and then resume all the previously removed threads.
This doesn't completely fix the sched_lock <-> sleepq chain LOR, but it
makes it a little better as we no longer call setrunnble() with a sleep
queue lock held meaning if setrunnable() tries to wakeup the swapper we
don't try to lock two sleep queue chains at the same time.
condition where kse_wakeup() doesn't yet see them in (interruptible)
sleep queues. Also add an upcall check to sleepqueue_catch_signals()
suggested by jhb.
This commit should fix recent mysql hangs.
Reviewed by: jhb, davidxu
Mysql'd by: Robin P. Blanchard <robin.blanchard at gactr uga edu>
might be enqueued on a sleep queue but not be asleep when the timeout fires
if it is blocked on a lock trying to check for pending signals before going
to sleep. In the case of fixing up the TDF_TIMEOUT race, however, the
thread must be marked asleep.
Reported by: kan (the bogus one)
to queue threads sleeping on a wait channel similar to how turnstiles are
used to queue threads waiting for a lock. This subsystem will be used as
the backend for sleep/wakeup and condition variables initially. Eventually
it will also be used to replace the ithread-specific iwait thread
inhibitor.
Sleep queues are also not locked by sched_lock, so this splits sched_lock
up a bit further increasing concurrency within the scheduler. Sleep queues
also natively support timeouts on sleeps and interruptible sleeps allowing
for the reduction of a lot of duplicated code between the sleep/wakeup and
condition variable implementations. For more details on the sleep queue
implementation, check the comments in sys/sleepqueue.h and
kern/subr_sleepqueue.c.
