Introduce a new rb_thread_sleep_interruptible that does not
execute interrupts before sleeping. Skipping the interrupt
check before sleep is required for out-of-GVL ruby_waitpid_all
to function properly when setting waitpid_state.ret
Now that ubf_select can be called by the gvl.timer thread
without recursive locking gvl.lock, we can safely use
rb_threadptr_interrupt to deal with waking up sleeping
processes,
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64576 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This simplifies the locking logic somewhat.
While we're at it, designate_timer_thread is worthless in
ubf_select because gvl_acquire_common already guarantees there
is a gvl.timer if gvl->waitq is populated.
In the future (for auto-fiber), this will allow using
th->unblock.func for rb_waitpid callers (via rb_sigchld_handler).
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64575 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We need to ensure Signal.trap handlers can function if the main
thread is sleeping after a subthread has grabbed sigwait_fd,
but later exited.
Consider the following timeline:
main_thread sub-thread
-----------------------------------------
Signal.trap() { ... }
get sigwait_fd
ppoll on sigwait_fd
native_cond_sleep
(via pthread_cond_wait)
ppoll times-out
put sigwait_fd
sub-thread exits
only thread alive
SIGNAL HITS
The problem is pthread_cond_wait cannot return EINTR,
so we can never run the Signal.trap handler. So we
will avoid using native_cond_sleep in the main thread
and always use ppoll to sleep when in the main thread.
This can guarantee the main thread remains aware of
signals; even if it cannot safely read off sigwait_fd
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64538 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Relying on "struct timespec" was too annoying API-wise and
used more stack space. "double" was a bit wacky w.r.t rounding
in the past, so now we'll switch to using a 64-bit type.
Unsigned 64-bit integer is able to give us over nearly 585
years of range with nanoseconds. This range is good enough
for the Linux kernel internal time representation, so it
ought to be good enough for us.
This reduces the stack usage of functions while GVL is held
(and thus subject to marking) on x86-64 Linux (with ppoll):
rb_wait_for_single_fd 120 => 104
do_select 120 => 88
[ruby-core:88582] [Misc #15014]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64533 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Based on r64478, any regular user creating more than 1024 pipes
on Linux will end up with tiny pipes with only a single page
capacity. So avoid wasting user resources and use lighter
eventfd on Linux.
[ruby-core:88563] [Misc #15011]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64527 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
TIMER_THREAD_CREATED_P already checks that pid, and glibc 2.25+
no longer caches getpid(2).
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64524 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
It's possible for the ubf_list_head to be populated with dead
threads at fork or the ubf_list_lock to be held, so reinitialize
both at startup.
And while we're at it, use a static initializer at startup
to save a library call and kill some ifdef.
[ruby-core:88578] [Bug #15013]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64485 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This matches the behavior of old timer thread more closely
and seems to fix [Bug #14999] when limited to a single CPU.
I cannot reproduce the error on a multi-core system unless
I use schedtool to force affinity to a single CPU:
schedtool -a 0x01 -e make test-spec \
MSPECOPT='-R1000 spec/ruby/library/conditionvariable/wait_spec.rb'
While it may be good enough to pass the spec, I don't have
huge degree of confidence in the interrupt handling robustness
under extremely heavy load (these may be ancient bugs, though).
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64467 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This is not called in signal handlers, so there's no reason for
it. glibc 2.25+ no longer caches getpid(), so it will cost a
syscall for those users.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64403 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We do not need to rely on SIGVTALRM for non-sighandler wakeups.
This will reduce spurious wakeups in cases where sigwait_fd
is not grabbed again, soon.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64389 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
For (rare) blocking functions which are not affected by signals,
we need to call the appropriate unblocking function via
`threadptr_trap_interrupt'
While we're at it, handling waitpid/SIGCHLD from gvl.timer isn't
harmful, here.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64388 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This hopefully clarifies the roles of UBF_TIMER and vm->gvl.timer
[ruby-core:88475] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64377 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Do not waste extra memory for each thread, but make
thread_pthread.c easier-to-follow as a result.
[ruby-core:88475] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64375 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
These functions will not be exported outside of thread_pthread.c
and we need to clarify the timer here is used for ubf and not
timeslice.
[ruby-core:88475] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64373 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This data structure has nothing to do with timers or threads.
[ruby-core:88475] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64372 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Hopefully this makes the code easier-to-follow
[ruby-core:88475] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64371 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
I'm not sure what's causing this failure in Solaris and only
on rubyspec, since rb_io_wait_readable is a well-exercised
code path in other places. But maybe using a pthread for
timing (similar to old timer-thread) can solve the issue.
cf. 20180814T042506Z.fail.html.gz
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64357 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
timer_create does not seem to support CLOCK_MONOTONIC on Solaris,
and CLOCK_HIRES seems like it could fail with insufficient permissions:
https://docs.oracle.com/cd/E86824_01/html/E54766/timer-create-3c.html
(Only tested on Linux and FreeBSD)
[ruby-core:88360] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64356 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
It looks like I forgot to account for a situation involving 3
threads.
[ruby-core:88360] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64354 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This closes race condition where GVL is uncontended and a thread
receives a signal immediately before calling the blocking
function when releasing GVL:
1) check interrupts
2) release GVL
3) blocking function
If signal fires after 1) but before 3), that thread may never
wake up if GVL is uncontended
We also need to wakeup the ubf_list unconditionally on
gvl_yield; because two threads can be yielding to each other
while waiting on IO#close while waiting on threads in IO#read or
IO#gets.
[ruby-core:88360] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64353 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This reverts commit 194a6a2c68 (r64203).
Race conditions which caused the original reversion will be fixed
in the subsequent commit.
[ruby-core:88360] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64352 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We don't use pthreads cancellation ourselves and it's painful to
use correctly. Any cancelled threads would break
vm->living_threads, GVL, thread_sync.c, autoload, etc...
So don't bother caring; because we can't stop rogue extensions
from completely breaking the VM in other ways, either.
[ruby-core:88282] [Misc #14962]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
By holding into sigwait_fd until after we acquire GVL, we can
hit the faster native_cond_sleep path instead of ppoll when
another thread wants to start sleeping. ppoll-ing on sigwait_fd
isn't really useful in program where GVL is contended
This also allows reducing vm->gvl.lock mutex contention on
waitpid sleep migrations.
r64170 this patch
vm_thread_condvar1 0.921 1.356
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64193 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
It seems to be unavailable on some platforms including my Android phone.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64166 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We must not allow reentry into ubf_list_head once we delete
ourselves, otherwise we could hang in there forever.
[ruby-core:88218] [Bug #14945]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64134 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Not having contention for GVL could mean everybody else is stuck
in blocking region without GVL, so we kick the ubf list in that
case.
I expect this to fix test_thread_fd_close timeout:
http://ci.rvm.jp/results/trunk-test@ruby-sky3/1173398
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64133 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* thread_pthread.c (USE_NATIVE_SLEEP_COND): revised wrongly removed
line with the ifndef guard.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64131 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Maybe some platforms have strange condition variable implementations
which have a "memory" of which mutexes they're associated with.
In any case, it makes documentation easier even on GNU/Linux and
FreeBSD.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64124 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We need to be able to perform periodic ubf_list wakeups when a
thread is sleeping and waiting on signals.
[ruby-core:88088] [Misc #14937] [Bug #5343]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64115 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
It's possible for another thread to take vm->gvl.lock
during gvl_release at the end of thread_start_func_2
during VM shutdown, at least.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64114 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We can't always designate a timer thread, so any sleepers must
also perform ubf wakeups. Note: a similar change needs to be
made for rb_thread_fd_select and rb_wait_for_single_fd.
[ruby-core:88088] [Misc #14937] [Bug #5343]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64111 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
thread_pthread.c relies on ppoll for rb_sigwait_sleep, so ensure
the compatibility wrapper is available for it.
[Bug #14950]
Reported-by: SHIBATA Hiroshi <hsbt@ruby-lang.org>
Reported-by: Greg L <Greg.mpls@gmail.com>
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64110 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We still need to to designate a timer thread after registering target
thread for the ubf list.
Oops :x
Note: I was never able to reproduce
test/thread/test_queue.rb::test_thr_kill failures on my on
Debian machines.
[ruby-core:88088] [Misc #14937] [Bug #5343]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64108 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
To reduce resource use and reduce CI failure; remove
timer-thread. Single-threaded Ruby processes (including forked
children) will never see extra thread overhead. This prevents
glibc and jemalloc from going into multi-threaded mode and
initializing locks or causing fragmentation via arena explosion.
The GVL is implements its own wait-queue as a ccan/list to
permit controlling wakeup order. Timeslice under contention is
handled by a designated timer thread (similar to choosing a
"patrol_thread" for current deadlock checking).
There is only one self-pipe, now, as wakeups for timeslice are
done independently using condition variables. This reduces FD
pressure slightly.
Signal handling is handled directly by a Ruby Thread (instead
of timer-thread) by exposing signal self-pipe to callers of
rb_thread_fd_select, native_sleep, rb_wait_for_single_fd, etc...
Acquiring, using, and releasing the self-pipe is exposed via 4
new internal functions:
1) rb_sigwait_fd_get - exclusively acquire timer_thread_pipe.normal[0]
2) rb_sigwait_fd_sleep - sleep and wait for signal (and no other FDs)
3) rb_sigwait_fd_put - release acquired result from rb_sigwait_fd_get
4) rb_sigwait_fd_migrate - migrate signal handling to another thread
after calling rb_sigwait_fd_put.
rb_sigwait_fd_migrate is necessary for waitpid callers because
only one thread can wait on self-pipe at a time, otherwise a
deadlock will occur if threads fight over the self-pipe.
TRAP_INTERRUPT_MASK is now set for the main thread directly in
signal handler via rb_thread_wakeup_timer_thread.
Originally, I wanted to use POSIX timers
(timer_create/timer_settime) for this. Unfortunately, this
proved unfeasible as Mutex#sleep resumes on spurious wakeups and
test/thread/test_cv.rb::test_condvar_timed_wait failed. Using
pthread_sigmask to mask out SIGVTALRM fixed that test, but
test/fiddle/test_function.rb::test_nogvl_poll proved there'd be
some unavoidable (and frequent) incompatibilities from that
approach.
Finally, this allows us to drop thread_destruct_lock and
interrupt current ec directly.
We don't need to rely on vm->thread_destruct_lock or a coherent
vm->running_thread on any platform. Separate timer-thread for
time slice and signal handling is relegated to thread_win32.c,
now.
[ruby-core:88088] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64107 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Otherwise, an altstack may live past ObjectSpace destruction
and xfree-ing the altstack will segfault.
[ruby-core:85621] [Feature #14487]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64102 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
There's actually no need to close the pipes used by the
sleepy timer thread before forking, only to stop the timer
thread itself.
Instead, we only close the parent pipes in the child process,
either via close-on-exec flag or when reinitializing the timer
thread.
This change will be necessary when we allow
rb_wait_for_single_fd and rb_thread_fd_select to wait on the
timer_thread_pipe.normal[0] directly and eliminate timer thread.
I don't anticipate compatibility problems with this change
alone.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63960 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
timer-thread will continue to be supported, but future
"timer" implementation may not be a thread.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63948 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
pthread_atfork is not idempotent and repeatedly calling it
causes it to register the same hook repeatedly; leading to
unbound memory growth.
Ruby already has a (confusing-named) internal API for to call
in the forked child process: rb_thread_atfork
Call the MJIT child_after_fork hook inside that to prevent
unbound growth with the following loop:
loop do
RubyVM::MJIT.pause
RubyVM::MJIT.resume
end
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63884 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Reading win32/win32.c waitpid implementation, maybe waitpid(-1, ...)
on that platform will never conflict with mjit use of waitpid.
In any case, I've added WAITPID_USE_SIGCHLD macro to vm_core.h
so it can be easy for Linux/BSD users to test (hopefully!)
win32-compatible code.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63855 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
The change is unstable on Windows. Please re-commit it when it correctly
supports Windows.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63852 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
I love `container_of' for generic data structures, but
in this case it's unnecessary and slightly harder-to-read.
This will make "Timeout in VM" slightly easier-to-read:
https://bugs.ruby-lang.org/issues/14859
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63846 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
