Analysis of current linux futex code :
--------------------------------------
A central hash table futex_queues[] holds all contexts (futex_q) of waiting
threads.
Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to
perform lookups or insert/deletion of a futex_q.
When a futex_wait() is done, calling thread has to :
1) - Obtain a read lock on mmap_sem to be able to validate the user pointer
(calling find_vma()). This validation tells us if the futex uses
an inode based store (mapped file), or mm based store (anonymous mem)
2) - compute a hash key
3) - Atomic increment of reference counter on an inode or a mm_struct
4) - lock part of futex_queues[] hash table
5) - perform the test on value of futex.
(rollback is value != expected_value, returns EWOULDBLOCK)
(various loops if test triggers mm faults)
6) queue the context into hash table, release the lock got in 4)
7) - release the read_lock on mmap_sem
<block>
8) Eventually unqueue the context (but rarely, as this part may be done
by the futex_wake())
Futexes were designed to improve scalability but current implementation has
various problems :
- Central hashtable :
This means scalability problems if many processes/threads want to use
futexes at the same time.
This means NUMA unbalance because this hashtable is located on one node.
- Using mmap_sem on every futex() syscall :
Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic
ops on mmap_sem, dirtying cache line :
- lot of cache line ping pongs on SMP configurations.
mmap_sem is also extensively used by mm code (page faults, mmap()/munmap())
Highly threaded processes might suffer from mmap_sem contention.
mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded
programs because of contention on the mmap_sem cache line.
- Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter:
It's also a cache line ping pong on SMP. It also increases mmap_sem hold time
because of cache misses.
Most of these scalability problems come from the fact that futexes are in
one global namespace. As we use a central hash table, we must make sure
they are all using the same reference (given by the mm subsystem). We
chose to force all futexes be 'shared'. This has a cost.
But fact is POSIX defined PRIVATE and SHARED, allowing clear separation,
and optimal performance if carefuly implemented. Time has come for linux
to have better threading performance.
The goal is to permit new futex commands to avoid :
- Taking the mmap_sem semaphore, conflicting with other subsystems.
- Modifying a ref_count on mm or an inode, still conflicting with mm or fs.
This is possible because, for one process using PTHREAD_PROCESS_PRIVATE
futexes, we only need to distinguish futexes by their virtual address, no
matter the underlying mm storage is.
If glibc wants to exploit this new infrastructure, it should use new
_PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be
prepared to fallback on old subcommands for old kernels. Using one global
variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK.
PTHREAD_PROCESS_SHARED futexes should still use the old subcommands.
Compatibility with old applications is preserved, they still hit the
scalability problems, but new applications can fly :)
Note : the same SHARED futex (mapped on a file) can be used by old binaries
*and* new binaries, because both binaries will use the old subcommands.
Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic,
as this is the default semantic. Almost all applications should benefit
of this changes (new kernel and updated libc)
Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine)
/* calling futex_wait(addr, value) with value != *addr */
433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes)
424 cycles per futex(FUTEX_WAIT) call (using one futex)
334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes)
334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex)
For reference :
187 cycles per getppid() call
188 cycles per umask() call
181 cycles per ni_syscall() call
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Pierre Peiffer <pierre.peiffer@bull.net>
Cc: "Ulrich Drepper" <drepper@gmail.com>
Cc: "Nick Piggin" <nickpiggin@yahoo.com.au>
Cc: "Ingo Molnar" <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch provides the futex_requeue_pi functionality, which allows some
threads waiting on a normal futex to be requeued on the wait-queue of a
PI-futex.
This provides an optimization, already used for (normal) futexes, to be used
with the PI-futexes.
This optimization is currently used by the glibc in pthread_broadcast, when
using "normal" mutexes. With futex_requeue_pi, it can be used with
PRIO_INHERIT mutexes too.
Signed-off-by: Pierre Peiffer <pierre.peiffer@bull.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch modifies futex_wait() to use an hrtimer + schedule() in place of
schedule_timeout().
schedule_timeout() is tick based, therefore the timeout granularity is the
tick (1 ms, 4 ms or 10 ms depending on HZ). By using a high resolution timer
for timeout wakeup, we can attain a much finer timeout granularity (in the
microsecond range). This parallels what is already done for futex_lock_pi().
The timeout passed to the syscall is no longer converted to jiffies and is
therefore passed to do_futex() and futex_wait() as an absolute ktime_t
therefore keeping nanosecond resolution.
Also this removes the need to pass the nanoseconds timeout part to
futex_lock_pi() in val2.
In futex_wait(), if there is no timeout then a regular schedule() is
performed. Otherwise, an hrtimer is fired before schedule() is called.
[akpm@linux-foundation.org: fix `make headers_check']
Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Signed-off-by: Pierre Peiffer <pierre.peiffer@bull.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Today, all threads waiting for a given futex are woken in FIFO order (first
waiter woken first) instead of priority order.
This patch makes use of plist (pirotity ordered lists) instead of simple list
in futex_hash_bucket.
All non-RT threads are stored with priority MAX_RT_PRIO, causing them to be
woken last, in FIFO order (RT-threads are woken first, in priority order).
Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Signed-off-by: Pierre Peiffer <pierre.peiffer@bull.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Thanks to Jarek Poplawski for the ideas and for spotting the bug in the
initial draft patch.
cancel_rearming_delayed_work() currently has many limitations, because it
requires that dwork always re-arms itself via queue_delayed_work(). So it
hangs forever if dwork doesn't do this, or cancel_rearming_delayed_work/
cancel_delayed_work was already called. It uses flush_workqueue() in a
loop, so it can't be used if workqueue was freezed, and it is potentially
live- lockable on busy system if delay is small.
With this patch cancel_rearming_delayed_work() doesn't make any assumptions
about dwork, it can re-arm itself via queue_delayed_work(), or
queue_work(), or do nothing.
As a "side effect", cancel_work_sync() was changed to handle re-arming works
as well.
Disadvantages:
- this patch adds wmb() to insert_work().
- slowdowns the fast path (when del_timer() succeeds on entry) of
cancel_rearming_delayed_work(), because wait_on_work() is called
unconditionally. In that case, compared to the old version, we are
doing "unneeded" lock/unlock for each online CPU.
On the other hand, this means we don't need to use cancel_work_sync()
after cancel_rearming_delayed_work().
- complicates the code (.text grows by 130 bytes).
[akpm@linux-foundation.org: fix speling]
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: David Chinner <dgc@sgi.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Acked-by: Jarek Poplawski <jarkao2@o2.pl>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We are anyway kthread_stop()ping other per-cpu kernel threads after
move_task_off_dead_cpu(), so we can do it with the stop_machine_run thread
as well.
I just checked with Vatsa if there was any subtle reason why they
had put in the kthread_bind() in cpu.c. Vatsa cannot seem to recollect
any and I can't see any. So let us just remove the kthread_bind.
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently kernel threads use sigprocmask(SIG_BLOCK) to protect against
signals. This doesn't prevent the signal delivery, this only blocks
signal_wake_up(). Every "killall -33 kthreadd" means a "struct siginfo"
leak.
Change kthreadd_setup() to set all handlers to SIG_IGN instead of blocking
them (make a new helper ignore_signals() for that). If the kernel thread
needs some signal, it should use allow_signal() anyway, and in that case it
should not use CLONE_SIGHAND.
Note that we can't change daemonize() (should die!) in the same way,
because it can be used along with CLONE_SIGHAND. This means that
allow_signal() still should unblock the signal to work correctly with
daemonize()ed threads.
However, disallow_signal() doesn't block the signal any longer but ignores
it.
NOTE: with or without this patch the kernel threads are not protected from
handle_stop_signal(), this seems harmless, but not good.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
worker_thread() inherits ignored SIGCHLD and numa_default_policy() from its
parent, kthreadd. No need to setup this again.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
allow_signal(SIGCHLD) does all necessary job, no need to call do_sigaction()
prior to.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When a kernel thread calls daemonize, instead of reparenting the thread to
init reparent the thread to kthreadd next to the threads created by
kthread_create.
This is really just a stop gap until daemonize goes away, but it does
ensure no kernel threads are under init and they are all in one place that
is easy to find.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently there is a circular reference between work queue initialization
and kthread initialization. This prevents the kthread infrastructure from
initializing until after work queues have been initialized.
We want the properties of tasks created with kthread_create to be as close
as possible to the init_task and to not be contaminated by user processes.
The later we start our kthreadd that creates these tasks the harder it is
to avoid contamination from user processes and the more of a mess we have
to clean up because the defaults have changed on us.
So this patch modifies the kthread support to not use work queues but to
instead use a simple list of structures, and to have kthreadd start from
init_task immediately after our kernel thread that execs /sbin/init.
By being a true child of init_task we only have to change those process
settings that we want to have different from init_task, such as our process
name, the cpus that are allowed, blocking all signals and setting SIGCHLD
to SIG_IGN so that all of our children are reaped automatically.
By being a true child of init_task we also naturally get our ppid set to 0
and do not wind up as a child of PID == 1. Ensuring that tasks generated
by kthread_create will not slow down the functioning of the wait family of
functions.
[akpm@linux-foundation.org: use interruptible sleeps]
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
____call_usermodehelper() has no reason for flush_signals(). It is a fresh
forked process which is going to exec a user-space application or exit on
failure.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
flush_work(wq, work) doesn't need the first parameter, we can use cwq->wq
(this was possible from the very beginnig, I missed this). So we can unify
flush_work_keventd and flush_work.
Also, rename flush_work() to cancel_work_sync() and fix all callers.
Perhaps this is not the best name, but "flush_work" is really bad.
(akpm: this is why the earlier patches bypassed maintainers)
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Auke Kok <auke-jan.h.kok@intel.com>,
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We already depend on fact that all sub-threads have ->exit_signal == -1, no
need to set it in zap_other_threads().
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
worker_thread() can miss freeze_process()->signal_wake_up() if it happens
between try_to_freeze() and prepare_to_wait(). We should check freezing()
before entering schedule().
This race was introduced by me in
[PATCH 1/1] workqueue: don't migrate pending works from the dead CPU
Looks like mm/vmscan.c:kswapd() has the same race.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
worker_thread() doesn't need to "Block and flush all signals", this was
already done by its caller, kthread().
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We don't have any users, and it is not so trivial to use NOAUTOREL works
correctly. It is better to simplify API.
Delete NOAUTOREL support and rename work_release to work_clear_pending to
avoid a confusion.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
cancel_rearming_delayed_workqueue(wq, dwork) doesn't need the first
parameter. We don't hang on un-queued dwork any longer, and work->data
doesn't change its type. This means we can always figure out "wq" from
dwork when it is needed.
Remove this parameter, and rename the function to
cancel_rearming_delayed_work(). Re-create an inline "obsolete"
cancel_rearming_delayed_workqueue(wq) which just calls
cancel_rearming_delayed_work().
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Cleanup. A number of per_cpu_ptr(wq->cpu_wq, cpu) users have to check that
cpu is valid for this wq. Make a simple helper.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently typeof(delayed_work->work.data) is
"struct workqueue_struct" when the timer is pending
"struct cpu_workqueue_struct" whe the work is queued
This makes impossible to use flush_fork(delayed_work->work) in addition
to cancel_delayed_work/cancel_rearming_delayed_work, not good.
Change queue_delayed_work/delayed_work_timer_fn to use cwq, not wq. This
complicates (and uglifies) these functions a little bit, but alows us to
use flush_fork(dwork) and imho makes the whole code more consistent.
Also, document the fact that cancel_rearming_delayed_work() doesn't garantee
the completion of work->func() upon return.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
CPU_UP_PREPARE binds cwq->thread to the new CPU. So CPU_UP_CANCELED tries to
wake up the task which is bound to the failed CPU.
With this patch we don't bind cwq->thread until CPU becomes online. The first
wake_up() after kthread_create() is a bit special, make a simple helper for
that.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The only caller of init_workqueues() is do_basic_setup().
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add explicit workqueue_struct->singlethread flag. This lessens .text a
little, but most importantly this allows us to manipulate wq->list without
changine the meaning of is_single_threaded().
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The code like
if (is_single_threaded(wq))
do_something(singlethread_cpu);
else {
for_each_cpu_mask(cpu, cpu_populated_map)
do_something(cpu);
}
looks very annoying. We can add "static cpumask_t cpu_singlethread_map" and
simplify the code. Lessens .text a bit, and imho makes the code more readable.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
cancel_rearming_delayed_workqueue(dwork) will hang forever if dwork was not
scheduled, because in that case cancel_delayed_work()->del_timer_sync() never
returns true.
I don't know if there are any callers which may have problems, but this is not
so convenient, and the fix is very simple.
Q: looks like we don't need "struct workqueue_struct *wq" parameter. If the
timer was aborted successfully, get_wq_data() == wq. Is it worth to add the
new function?
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Because it has no callers.
Actually, I think the whole idea of run_scheduled_work() was not right, not
good to mix "unqueue this work and execute its ->func()" in one function.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently CPU_DEAD uses kthread_stop() to stop cwq->thread and then
transfers cwq->worklist to another CPU. However, it is very unlikely that
worker_thread() will notice kthread_should_stop() before flushing
cwq->worklist. It is only possible if worker_thread() was preempted after
run_workqueue(cwq), a new work_struct was added, and CPU_DEAD happened
before cwq->thread has a chance to run.
This means that take_over_work() mostly adds unneeded complications. Note
also that kthread_stop() is not good per se, wake_up_process() may confuse
work->func() if it sleeps waiting for some event.
Remove take_over_work() and migrate_sequence complications. CPU_DEAD sets
the cwq->should_stop flag (introduced by this patch) and waits for
cwq->thread to flush cwq->worklist and exit. Because the dead CPU is not
on cpu_online_map, no more works can be added to that cwq.
cpu_populated_map was introduced to optimize for_each_possible_cpu(), it is
not strictly needed, and it is more a documentation in fact.
Saves 418 bytes.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pointed out by Srivatsa Vaddagiri.
cleanup_workqueue_thread() sets cwq->thread = NULL and does kthread_stop().
This breaks the "if (cwq->thread == current)" logic in flush_cpu_workqueue()
and leads to deadlock.
Kill the thead first, then clear cwq->thread. workqueue_mutex protects us
from create_workqueue_thread() so we don't need cwq->lock.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Many thanks to Srivatsa Vaddagiri for the helpful discussion and for spotting
the bug in my previous attempt.
work->func() (and thus flush_workqueue()) must not use workqueue_mutex,
this leads to deadlock when CPU_DEAD does kthread_stop(). However without
this mutex held we can't detect CPU_DEAD in progress, which can move pending
works to another CPU while the dead one is not on cpu_online_map.
Change flush_workqueue() to use for_each_possible_cpu(). This means that
flush_cpu_workqueue() may hit CPU which is already dead. However in that
case
!list_empty(&cwq->worklist) || cwq->current_work != NULL
means that CPU_DEAD in progress, it will do kthread_stop() + take_over_work()
so we can proceed and insert a barrier. We hold cwq->lock, so we are safe.
Also, add migrate_sequence incremented by take_over_work() under cwq->lock.
If take_over_work() happened before we checked this CPU, we should see the
new value after spin_unlock().
Further possible changes:
remove CPU_DEAD handling (along with take_over_work, migrate_sequence)
from workqueue.c. CPU_DEAD just sets cwq->please_exit_after_flush flag.
CPU_UP_PREPARE->create_workqueue_thread() clears this flag, and creates
the new thread if cwq->thread == NULL.
This way the workqueue/cpu-hotplug interaction is almost zero, workqueue_mutex
just protects "workqueues" list, CPU_LOCK_ACQUIRE/CPU_LOCK_RELEASE go away.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently ->freezeable is per-cpu, this is wrong. CPU_UP_PREPARE creates
cwq->thread which is not freezeable. Move ->freezeable to workqueue_struct.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This makes cpu hotplug symmetrical: if CPU_UP_PREPARE fails we get
CPU_UP_CANCELED, so we can undo what ever happened on PREPARE. The same
should happen for CPU_DOWN_PREPARE.
[akpm@linux-foundation.org: fix for reduce-size-of-task_struct-on-64-bit-machines]
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Eliminate lock_cpu_hotplug from kernel/sched.c and use sched_hotcpu_mutex
instead to postpone a hotplug event.
In the migration_call hotcpu callback function, take sched_hotcpu_mutex
while handling the event CPU_LOCK_ACQUIRE and release it while handling
CPU_LOCK_RELEASE event.
[akpm@linux-foundation.org: fix deadlock]
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is an attempt to provide an alternate mechanism for postponing
a hotplug event instead of using a global mechanism like lock_cpu_hotplug.
The proposal is to add two new events namely CPU_LOCK_ACQUIRE and
CPU_LOCK_RELEASE. The notification for these two events would be sent
out before and after a cpu_hotplug event respectively.
During the CPU_LOCK_ACQUIRE event, a cpu-hotplug-aware subsystem is
supposed to acquire any per-subsystem hotcpu mutex ( Eg. workqueue_mutex
in kernel/workqueue.c ).
During the CPU_LOCK_RELEASE release event the cpu-hotplug-aware subsystem
is supposed to release the per-subsystem hotcpu mutex.
The reasons for defining new events as opposed to reusing the existing events
like CPU_UP_PREPARE/CPU_UP_FAILED/CPU_ONLINE for locking/unlocking of
per-subsystem hotcpu mutexes are as follow:
- CPU_LOCK_ACQUIRE: All hotcpu mutexes are taken before subsystems
start handling pre-hotplug events like CPU_UP_PREPARE/CPU_DOWN_PREPARE
etc, thus ensuring a clean handling of these events.
- CPU_LOCK_RELEASE: The hotcpu mutexes will be released only after
all subsystems have handled post-hotplug events like CPU_DOWN_FAILED,
CPU_DEAD,CPU_ONLINE etc thereby ensuring that there are no subsequent
clashes amongst the interdependent subsystems after a cpu hotplugs.
This patch also uses __raw_notifier_call chain in _cpu_up to take care
of the dependency between the two consequetive calls to
raw_notifier_call_chain.
[akpm@linux-foundation.org: fix a bug]
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since 2.6.18-something, the community has been bugged by the problem to
provide a clean and a stable mechanism to postpone a cpu-hotplug event as
lock_cpu_hotplug was badly broken.
This is another proposal towards solving that problem. This one is along the
lines of the solution provided in kernel/workqueue.c
Instead of having a global mechanism like lock_cpu_hotplug, we allow the
subsytems to define their own per-subsystem hot cpu mutexes. These would be
taken(released) where ever we are currently calling
lock_cpu_hotplug(unlock_cpu_hotplug).
Also, in the per-subsystem hotcpu callback function,we take this mutex before
we handle any pre-cpu-hotplug events and release it once we finish handling
the post-cpu-hotplug events. A standard means for doing this has been
provided in [PATCH 2/4] and demonstrated in [PATCH 3/4].
The ordering of these per-subsystem mutexes might still prove to be a
problem, but hopefully lockdep should help us get out of that muddle.
The patch set to be applied against linux-2.6.19-rc5 is as follows:
[PATCH 1/4] : Extend notifier_call_chain with an option to specify the
number of notifications to be sent and also count the
number of notifications actually sent.
[PATCH 2/4] : Define events CPU_LOCK_ACQUIRE and CPU_LOCK_RELEASE
and send out notifications for these in _cpu_up and
_cpu_down. This would help us standardise the acquire and
release of the subsystem locks in the hotcpu
callback functions of these subsystems.
[PATCH 3/4] : Eliminate lock_cpu_hotplug from kernel/sched.c.
[PATCH 4/4] : In workqueue_cpu_callback function, acquire(release) the
workqueue_mutex while handling
CPU_LOCK_ACQUIRE(CPU_LOCK_RELEASE).
If the per-subsystem-locking approach survives the test of time, we can expect
a slow phasing out of lock_cpu_hotplug, which has not yet been eliminated in
these patches :)
This patch:
Provide notifier_call_chain with an option to call only a specified number of
notifiers and also record the number of call to notifiers made.
The need for this enhancement was identified in the post entitled
"Slab - Eliminate lock_cpu_hotplug from slab"
(http://lkml.org/lkml/2006/10/28/92) by Ravikiran G Thirumalai and
Andrew Morton.
This patch adds two additional parameters to notifier_call_chain API namely
- int nr_to_calls : Number of notifier_functions to be called.
The don't care value is -1.
- unsigned int *nr_calls : Records the total number of notifier_funtions
called by notifier_call_chain. The don't care
value is NULL.
[michal.k.k.piotrowski@gmail.com: build fix]
Credit: Andrew Morton <akpm@osdl.org>
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Michal Piotrowski <michal.k.k.piotrowski@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
relay doesn't need to use schedule_delayed_work() for waking readers
when a simple timer will do.
Signed-off-by: Tom Zanussi <zanussi@comcast.net>
Cc: Satyam Sharma <satyam.sharma@gmail.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Now when we have ->current_work we can avoid adding a barrier and waiting
for its completition when cwq's queue is empty.
Note: this change is also useful if we change flush_workqueue() to also
check the dead CPUs.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A basic problem with flush_scheduled_work() is that it blocks behind _all_
presently-queued works, rather than just the work whcih the caller wants to
flush. If the caller holds some lock, and if one of the queued work happens
to want that lock as well then accidental deadlocks can occur.
One example of this is the phy layer: it wants to flush work while holding
rtnl_lock(). But if a linkwatch event happens to be queued, the phy code will
deadlock because the linkwatch callback function takes rtnl_lock.
So we implement a new function which will flush a *single* work - just the one
which the caller wants to free up. Thus we avoid the accidental deadlocks
which can arise from unrelated subsystems' callbacks taking shared locks.
flush_work() non-blockingly dequeues the work_struct which we want to kill,
then it waits for its handler to complete on all CPUs.
Add ->current_work to the "struct cpu_workqueue_struct", it points to
currently running "struct work_struct". When flush_work(work) detects
->current_work == work, it inserts a barrier at the _head_ of ->worklist
(and thus right _after_ that work) and waits for completition. This means
that the next work fired on that CPU will be this barrier, or another
barrier queued by concurrent flush_work(), so the caller of flush_work()
will be woken before any "regular" work has a chance to run.
When wait_on_work() unlocks workqueue_mutex (or whatever we choose to protect
against CPU hotplug), CPU may go away. But in that case take_over_work() will
move a barrier we queued to another CPU, it will be fired sometime, and
wait_on_work() will be woken.
Actually, we are doing cleanup_workqueue_thread()->kthread_stop() before
take_over_work(), so cwq->thread should complete its ->worklist (and thus
the barrier), because currently we don't check kthread_should_stop() in
run_workqueue(). But even if we did, everything should be ok.
[akpm@osdl.org: cleanup]
[akpm@osdl.org: add flush_work_keventd() wrapper]
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Remove ->remove_sequence, ->insert_sequence, and ->work_done from struct
cpu_workqueue_struct. To implement flush_workqueue() we can queue a
barrier work on each CPU and wait for its completition.
The barrier is queued under workqueue_mutex to ensure that per cpu
wq->cpu_wq is alive, we drop this mutex before going to sleep. If CPU goes
down while we are waiting for completition, take_over_work() will move the
barrier on another CPU, and the handler will wake up us eventually.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We take workqueue_mutex in there to keep CPU hotplug away. But
preempt_disable() will suffice for that.
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
u64 and s64 are not necessarily 'long long' on some 64-bit
platforms, so explicit the type to kill the compiler warnings.
Also consistently use '%Lu' which is unsigned.
Signed-off-by: David S. Miller <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There's more that need fixing, and fix my own subject spelling error too.
Signed-off-by: Daniel Walker <dwalker@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch moves the sig_kernel_* and related macros from kernel/signal.c
to linux/signal.h, and cleans them up slightly. I need the sig_kernel_*
macros for default signal behavior in the utrace code, and want to avoid
duplication or overhead to share the knowledge.
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Cleanup using simple_read_from_buffer() for /dev/cpuset/tasks and
/proc/config.gz.
Cc: Paul Jackson <pj@sgi.com>
Cc: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A linuxdoc comment had fallen out of date - it refers to an argument which no
longer exists.
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[ With Johannes Berg <johannes@sipsolutions.net> ]
Separate the hibernation (aka suspend to disk code) from the other suspend
code. In particular:
* Remove the definitions related to hibernation from include/linux/pm.h
* Introduce struct hibernation_ops and a new hibernate() function to hibernate
the system, defined in include/linux/suspend.h
* Separate suspend code in kernel/power/main.c from hibernation-related code
in kernel/power/disk.c and kernel/power/user.c (with the help of
hibernation_ops)
* Switch ACPI (the only user of pm_ops.pm_disk_mode) to hibernation_ops
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Greg KH <greg@kroah.com>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Nigel Cunningham <nigel@nigel.suspend2.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Jarek Poplawski