Running interrupt handlers with interrupts enabled can cause stack
overflows. That has been observed with multiqueue NICs delivering all
their interrupts to a single core. We might band aid that somehow by
checking the interrupt stacks, but the real safe fix is to run the irq
handlers with interrupts disabled.
Drivers for whacky hardware still can reenable them in the handler
itself, if the need arises. (They do already due to lockdep)
The risk of doing this is rather low:
- lockdep already enforces this
- CONFIG_NOHZ has shaken out the drivers which relied on jiffies updates
- time keeping is not longer sensitive to the timer interrupt being delayed
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Miller <davem@davemloft.net>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Linus Torvalds <torvalds@osdl.org>
LKML-Reference: <20100326000405.758579387@linutronix.de>
Now that we enjoy threaded interrupts, we're starting to see irq_chip
implementations (wm831x, pca953x) that make use of threaded interrupts
for the controller, and nested interrupts for the client interrupt. It
all works very well, with one drawback:
Drivers requesting an IRQ must now know whether the handler will
run in a thread context or not, and call request_threaded_irq() or
request_irq() accordingly.
The problem is that the requesting driver sometimes doesn't know
about the nature of the interrupt, specially when the interrupt
controller is a discrete chip (typically a GPIO expander connected
over I2C) that can be connected to a wide variety of otherwise perfectly
supported hardware.
This patch introduces the request_any_context_irq() function that mostly
mimics the usual request_irq(), except that it checks whether the irq
level is configured as nested or not, and calls the right backend.
On success, it also returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.
[ tglx: Made return value an enum, simplified code and made the export
of request_any_context_irq GPL ]
Signed-off-by: Marc Zyngier <maz@misterjones.org>
Cc: <joachim.eastwood@jotron.com>
LKML-Reference: <927ea285bd0c68934ddae1a47e44a9ba@localhost>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
With the earlier logarithmic time accumulation patch, xtime will now
always be within one "tick" of the current time, instead of possibly
half a second off.
This removes the need for the xtime_cache value, which always stored the
time at the last interrupt, so this patch cleans that up removing the
xtime_cache related code.
This patch also addresses an issue with an earlier version of this change,
where xtime_cache was normalizing xtime, which could in some cases be
not valid (ie: tv_nsec == NSEC_PER_SEC). This is fixed by handling
the edge case in update_wall_time().
Signed-off-by: John Stultz <johnstul@us.ibm.com>
Cc: Petr Titěra <P.Titera@century.cz>
LKML-Reference: <1270589451-30773-1-git-send-email-johnstul@us.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
When CONFIG_DEBUG_BLOCK_EXT_DEVT is set we decode the device
improperly by old_decode_dev and it results in an error while
hibernating with s2disk.
All users already pass the new device number, so switch to
new_decode_dev().
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Reported-and-tested-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: "Rafael J. Wysocki" <rjw@sisk.pl>
The comment suggests that this usage is stale. There is no bkl in the
exec path so if there is a race lurking there, the bkl in ptrace is
not going to help in this regard.
Overview of the possibility of "accidental" races this bkl might
protect:
- ptrace_traceme() is protected against task removal and concurrent
read/write on current->ptrace as it locks write tasklist_lock.
- arch_ptrace_attach() is serialized by ptrace_traceme() against
concurrent PTRACE_TRACEME or PTRACE_ATTACH
- ptrace_attach() is protected the same way ptrace_traceme() and
in turn serializes arch_ptrace_attach()
- ptrace_check_attach() does its own well described serializing too.
There is no obvious race here.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Roland McGrath <roland@redhat.com>
- We weren't zeroing p->rss_stat[] at fork()
- Consequently sync_mm_rss() was dereferencing tsk->mm for kernel
threads and was oopsing.
- Make __sync_task_rss_stat() static, too.
Addresses https://bugzilla.kernel.org/show_bug.cgi?id=15648
[akpm@linux-foundation.org: remove the BUG_ON(!mm->rss)]
Reported-by: Troels Liebe Bentsen <tlb@rapanden.dk>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
"Michael S. Tsirkin" <mst@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
genirq: Force MSI irq handlers to run with interrupts disabled
The current version of schedule_hrtimeout() always uses the
monotonic clock. Some system calls such as mq_timedsend()
and mq_timedreceive(), however, require the use of the wall
clock due to the definition of the system call.
This patch provides the infrastructure to use schedule_hrtimeout()
with a CLOCK_REALTIME timer.
Signed-off-by: Carsten Emde <C.Emde@osadl.org>
Tested-by: Pradyumna Sampath <pradysam@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Arjan van de Veen <arjan@infradead.org>
LKML-Reference: <20100402204331.167439615@osadl.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
While HR timers have had the concept of timer slack for quite some time
now, the legacy timers lacked this concept, and had to make do with
round_jiffies() and friends.
Timer slack is important for power management; grouping timers reduces the
number of wakeups which in turn reduces power consumption.
This patch introduces timer slack to the legacy timers using the following
pieces:
* A slack field in the timer struct
* An api (set_timer_slack) that callers can use to set explicit timer slack
* A default slack of 0.4% of the requested delay for callers that do not set
any explicit slack
* Rounding code that is part of mod_timer() that tries to
group timers around jiffies values every 'power of two'
(so quick timers will group around every 2, but longer timers
will group around every 4, 8, 16, 32 etc)
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: johnstul@us.ibm.com
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
taskset on 2.6.34-rc3 fails on one of my ppc64 test boxes with
the following error:
sched_getaffinity(0, 16, 0x10029650030) = -1 EINVAL (Invalid argument)
This box has 128 threads and 16 bytes is enough to cover it.
Commit cd3d8031eb (sched:
sched_getaffinity(): Allow less than NR_CPUS length) is
comparing this 16 bytes agains nr_cpu_ids.
Fix it by comparing nr_cpu_ids to the number of bits in the
cpumask we pass in.
Signed-off-by: Anton Blanchard <anton@samba.org>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Sharyathi Nagesh <sharyath@in.ibm.com>
Cc: Ulrich Drepper <drepper@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Russ Anderson <rja@sgi.com>
Cc: Mike Travis <travis@sgi.com>
LKML-Reference: <20100406070218.GM5594@kryten>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Module refcounting is implemented with a per-cpu counter for speed.
However there is a race when tallying the counter where a reference may
be taken by one CPU and released by another. Reference count summation
may then see the decrement without having seen the previous increment,
leading to lower than expected count. A module which never has its
actual reference drop below 1 may return a reference count of 0 due to
this race.
Module removal generally runs under stop_machine, which prevents this
race causing bugs due to removal of in-use modules. However there are
other real bugs in module.c code and driver code (module_refcount is
exported) where the callers do not run under stop_machine.
Fix this by maintaining running per-cpu counters for the number of
module refcount increments and the number of refcount decrements. The
increments are tallied after the decrements, so any decrement seen will
always have its corresponding increment counted. The final refcount is
the difference of the total increments and decrements, preventing a
low-refcount from being returned.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Locking statistics are implemented using global atomic
variables. This is usually fine unless some path write them very
often.
This is the case for the function and function graph tracers
that disable irqs for each entry saved (except if the function
tracer is in preempt disabled only mode).
And calls to local_irq_save/restore() increment
hardirqs_on_events and hardirqs_off_events stats (or similar
stats for redundant versions).
Incrementing these global vars for each function ends up in too
much cache bouncing if lockstats are enabled.
To solve this, implement the debug_atomic_*() operations using
per cpu vars.
-v2: Use per_cpu() instead of get_cpu_var() to fetch the desired
cpu vars on debug_atomic_read()
-v3: Store the stats in a structure. No need for local_t as we
are NMI/irq safe.
-v4: Fix tons of build errors. I thought I had tested it but I
probably forgot to select the relevant config.
Suggested-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1270505417-8144-1-git-send-regression-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
There have been a number of reports of people seeing the message:
"name_count maxed, losing inode data: dev=00:05, inode=3185"
in dmesg. These usually lead to people reporting problems to the filesystem
group who are in turn clueless what they mean.
Eventually someone finds me and I explain what is going on and that
these come from the audit system. The basics of the problem is that the
audit subsystem never expects a single syscall to 'interact' (for some
wish washy meaning of interact) with more than 20 inodes. But in fact
some operations like loading kernel modules can cause changes to lots of
inodes in debugfs.
There are a couple real fixes being bandied about including removing the
fixed compile time limit of 20 or not auditing changes in debugfs (or
both) but neither are small and obvious so I am not sending them for
immediate inclusion (I hope Al forwards a real solution next devel
window).
In the meantime this patch simply adds 'audit' to the beginning of the
crap message so if a user sees it, they come blame me first and we can
talk about what it means and make sure we understand all of the reasons
it can happen and make sure this gets solved correctly in the long run.
Signed-off-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'slabh' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/misc:
eeepc-wmi: include slab.h
staging/otus: include slab.h from usbdrv.h
percpu: don't implicitly include slab.h from percpu.h
kmemcheck: Fix build errors due to missing slab.h
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
iwlwifi: don't include iwl-dev.h from iwl-devtrace.h
x86: don't include slab.h from arch/x86/include/asm/pgtable_32.h
Fix up trivial conflicts in include/linux/percpu.h due to
is_kernel_percpu_address() having been introduced since the slab.h
cleanup with the percpu_up.c splitup.
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu:
module: add stub for is_module_percpu_address
percpu, module: implement and use is_kernel/module_percpu_address()
module: encapsulate percpu handling better and record percpu_size
Because a local variable is not initialized, I got these
when I did 'cat tracing/trace'. (not trace_pipe):
CPU:0 [LOST 18446744071579453134 EVENTS]
ps-3099 [000] 560.770221: lock_acquire: ffff880030865010 &(&dentry->d_lock)->rlock
CPU:0 [LOST 18446744071579453134 EVENTS]
ps-3099 [000] 560.770221: lock_release: ffff880030865010 &(&dentry->d_lock)->rlock
CPU:0 [LOST 18446612133255294080 EVENTS]
ps-3099 [000] 560.770221: lock_acquire: ffff880030865010 &(&dentry->d_lock)->rlock
CPU:0 [LOST 18446744071579453134 EVENTS]
ps-3099 [000] 560.770222: lock_release: ffff880030865010 &(&dentry->d_lock)->rlock
CPU:0 [LOST 18446744071579453134 EVENTS]
ps-3099 [000] 560.770222: lock_release: ffffffff816cfb98 dcache_lock
See peek_next_entry(), it does not set *lost_events when we 'cat tracing/trace'
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
LKML-Reference: <4BB9A929.2000303@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
perf: Always build the powerpc perf_arch_fetch_caller_regs version
perf: Always build the stub perf_arch_fetch_caller_regs version
perf, probe-finder: Build fix on Debian
perf/scripts: Tuple was set from long in both branches in python_process_event()
perf: Fix 'perf sched record' deadlock
perf, x86: Fix callgraphs of 32-bit processes on 64-bit kernels
perf, x86: Fix AMD hotplug & constraint initialization
x86: Move notify_cpu_starting() callback to a later stage
x86,kgdb: Always initialize the hw breakpoint attribute
perf: Use hot regs with software sched switch/migrate events
perf: Correctly align perf event tracing buffer
* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: set_cpus_allowed_ptr(): Don't use rq->migration_thread after unlock
sched: Fix proc_sched_set_task()
Perf_event does not need seeking, so prevent it in order to
get rid of default_llseek, which uses the BKL.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@elte.hu>
[drop the nonseekable_open, not needed for anon inodes]
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Now that software events use perf_arch_fetch_caller_regs() too, we
need the stub version to be always built in for archs that don't
implement it.
Fixes the following build error in PARISC:
kernel/built-in.o: In function `perf_event_task_sched_out':
(.text.perf_event_task_sched_out+0x54): undefined reference to `perf_arch_fetch_caller_regs'
Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
* 'kgdb-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jwessel/linux-2.6-kgdb:
kgdb: Turn off tracing while in the debugger
kgdb: use atomic_inc and atomic_dec instead of atomic_set
kgdb: eliminate kgdb_wait(), all cpus enter the same way
kgdbts,sh: Add in breakpoint pc offset for superh
kgdb: have ebin2mem call probe_kernel_write once
* 'pm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/suspend-2.6:
Freezer: Fix buggy resume test for tasks frozen with cgroup freezer
Freezer: Only show the state of tasks refusing to freeze
The kernel debugger should turn off kernel tracing any time the
debugger is active and restore it on resume.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Reviewed-by: Steven Rostedt <rostedt@goodmis.org>
Memory barriers should be used for the kgdb cpu synchronization. The
atomic_set() does not imply a memory barrier.
Reported-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
This is a kgdb architectural change to have all the cpus (master or
slave) enter the same function.
A cpu that hits an exception (wants to be the master cpu) will call
kgdb_handle_exception() from the trap handler and then invoke a
kgdb_roundup_cpu() to synchronize the other cpus and bring them into
the kgdb_handle_exception() as well.
A slave cpu will enter kgdb_handle_exception() from the
kgdb_nmicallback() and set the exception state to note that the
processor is a slave.
Previously the salve cpu would have called kgdb_wait(). This change
allows the debug core to change cpus without resuming the system in
order to inspect arch specific cpu information.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Rather than call probe_kernel_write() one byte at a time, process the
whole buffer locally and pass the entire result in one go. This way,
architectures that need to do special handling based on the length can
do so, or we only end up calling memcpy() once.
[sonic.zhang@analog.com: Reported original problem and preliminary patch]
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Sonic Zhang <sonic.zhang@analog.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
In order to reduce the dependency on TASK_WAKING rework the enqueue
interface to support a proper flags field.
Replace the int wakeup, bool head arguments with an int flags argument
and create the following flags:
ENQUEUE_WAKEUP - the enqueue is a wakeup of a sleeping task,
ENQUEUE_WAKING - the enqueue has relative vruntime due to
having sched_class::task_waking() called,
ENQUEUE_HEAD - the waking task should be places on the head
of the priority queue (where appropriate).
For symmetry also convert sched_class::dequeue() to a flags scheme.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The cpuload calculation in calc_load_account_active() assumes
rq->nr_uninterruptible will not change on an offline cpu after
migrate_nr_uninterruptible(). However the recent migrate on wakeup
changes broke that and would result in decrementing the offline cpu's
rq->nr_uninterruptible.
Fix this by accounting the nr_uninterruptible on the waking cpu.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Now that we hold the rq->lock over set_task_cpu() again, we can do
away with most of the TASK_WAKING checks and reduce them again to
set_cpus_allowed_ptr().
Removes some conditionals from scheduling hot-paths.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Oleg Nesterov <oleg@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Oleg noticed a few races with the TASK_WAKING usage on fork.
- since TASK_WAKING is basically a spinlock, it should be IRQ safe
- since we set TASK_WAKING (*) without holding rq->lock it could
be there still is a rq->lock holder, thereby not actually
providing full serialization.
(*) in fact we clear PF_STARTING, which in effect enables TASK_WAKING.
Cure the second issue by not setting TASK_WAKING in sched_fork(), but
only temporarily in wake_up_new_task() while calling select_task_rq().
Cure the first by holding rq->lock around the select_task_rq() call,
this will disable IRQs, this however requires that we push down the
rq->lock release into select_task_rq_fair()'s cgroup stuff.
Because select_task_rq_fair() still needs to drop the rq->lock we
cannot fully get rid of TASK_WAKING.
Reported-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Introduce cpuset_cpus_allowed_fallback() helper to fix the cpuset problems
with select_fallback_rq(). It can be called from any context and can't use
any cpuset locks including task_lock(). It is called when the task doesn't
have online cpus in ->cpus_allowed but ttwu/etc must be able to find a
suitable cpu.
I am not proud of this patch. Everything which needs such a fat comment
can't be good even if correct. But I'd prefer to not change the locking
rules in the code I hardly understand, and in any case I believe this
simple change make the code much more correct compared to deadlocks we
currently have.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100315091027.GA9155@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
_cpu_down() changes the current task's affinity and then recovers it at
the end. The problems are well known: we can't restore old_allowed if it
was bound to the now-dead-cpu, and we can race with the userspace which
can change cpu-affinity during unplug.
_cpu_down() should not play with current->cpus_allowed at all. Instead,
take_cpu_down() can migrate the caller of _cpu_down() after __cpu_disable()
removes the dying cpu from cpu_online_mask.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100315091023.GA9148@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
sched_exec()->select_task_rq() reads/updates ->cpus_allowed lockless.
This can race with other CPUs updating our ->cpus_allowed, and this
looks meaningless to me.
The task is current and running, it must have online cpus in ->cpus_allowed,
the fallback mode is bogus. And, if ->sched_class returns the "wrong" cpu,
this likely means we raced with set_cpus_allowed() which was called
for reason, why should sched_exec() retry and call ->select_task_rq()
again?
Change the code to call sched_class->select_task_rq() directly and do
nothing if the returned cpu is wrong after re-checking under rq->lock.
From now task_struct->cpus_allowed is always stable under TASK_WAKING,
select_fallback_rq() is always called under rq-lock or the caller or
the caller owns TASK_WAKING (select_task_rq).
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100315091019.GA9141@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The previous patch preserved the retry logic, but it looks unneeded.
__migrate_task() can only fail if we raced with migration after we dropped
the lock, but in this case the caller of set_cpus_allowed/etc must initiate
migration itself if ->on_rq == T.
We already fixed p->cpus_allowed, the changes in active/online masks must
be visible to racer, it should migrate the task to online cpu correctly.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100315091014.GA9138@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
move_task_off_dead_cpu()->select_fallback_rq() reads/updates ->cpus_allowed
lockless. We can race with set_cpus_allowed() running in parallel.
Change it to take rq->lock around select_fallback_rq(). Note that it is not
trivial to move this spin_lock() into select_fallback_rq(), we must recheck
the task was not migrated after we take the lock and other callers do not
need this lock.
To avoid the races with other callers of select_fallback_rq() which rely on
TASK_WAKING, we also check p->state != TASK_WAKING and do nothing otherwise.
The owner of TASK_WAKING must update ->cpus_allowed and choose the correct
CPU anyway, and the subsequent __migrate_task() is just meaningless because
p->se.on_rq must be false.
Alternatively, we could change select_task_rq() to take rq->lock right
after it calls sched_class->select_task_rq(), but this looks a bit ugly.
Also, change it to not assume irqs are disabled and absorb __migrate_task_irq().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100315091010.GA9131@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch just states the fact the cpusets/cpuhotplug interaction is
broken and removes the deadlockable code which only pretends to work.
- cpuset_lock() doesn't really work. It is needed for
cpuset_cpus_allowed_locked() but we can't take this lock in
try_to_wake_up()->select_fallback_rq() path.
- cpuset_lock() is deadlockable. Suppose that a task T bound to CPU takes
callback_mutex. If cpu_down(CPU) happens before T drops callback_mutex
stop_machine() preempts T, then migration_call(CPU_DEAD) tries to take
cpuset_lock() and hangs forever because CPU is already dead and thus
T can't be scheduled.
- cpuset_cpus_allowed_locked() is deadlockable too. It takes task_lock()
which is not irq-safe, but try_to_wake_up() can be called from irq.
Kill them, and change select_fallback_rq() to use cpu_possible_mask, like
we currently do without CONFIG_CPUSETS.
Also, with or without this patch, with or without CONFIG_CPUSETS, the
callers of select_fallback_rq() can race with each other or with
set_cpus_allowed() pathes.
The subsequent patches try to to fix these problems.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100315091003.GA9123@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
