This patch fixes a bug located by Vegard Nossum with the aid of
kmemcheck, updated based on review comments from Nick Piggin,
Ingo Molnar, and Andrew Morton. And cleans up the variable-name
and function-name language. ;-)
The boot CPU runs in the context of its idle thread during boot-up.
During this time, idle_cpu(0) will always return nonzero, which will
fool Classic and Hierarchical RCU into deciding that a large chunk of
the boot-up sequence is a big long quiescent state. This in turn causes
RCU to prematurely end grace periods during this time.
This patch changes the rcutree.c and rcuclassic.c rcu_check_callbacks()
function to ignore the idle task as a quiescent state until the
system has started up the scheduler in rest_init(), introducing a
new non-API function rcu_idle_now_means_idle() to inform RCU of this
transition. RCU maintains an internal rcu_idle_cpu_truthful variable
to track this state, which is then used by rcu_check_callback() to
determine if it should believe idle_cpu().
Because this patch has the effect of disallowing RCU grace periods
during long stretches of the boot-up sequence, this patch also introduces
Josh Triplett's UP-only optimization that makes synchronize_rcu() be a
no-op if num_online_cpus() returns 1. This allows boot-time code that
calls synchronize_rcu() to proceed normally. Note, however, that RCU
callbacks registered by call_rcu() will likely queue up until later in
the boot sequence. Although rcuclassic and rcutree can also use this
same optimization after boot completes, rcupreempt must restrict its
use of this optimization to the portion of the boot sequence before the
scheduler starts up, given that an rcupreempt RCU read-side critical
section may be preeempted.
In addition, this patch takes Nick Piggin's suggestion to make the
system_state global variable be __read_mostly.
Changes since v4:
o Changes the name of the introduced function and variable to
be less emotional. ;-)
Changes since v3:
o WARN_ON(nr_context_switches() > 0) to verify that RCU
switches out of boot-time mode before the first context
switch, as suggested by Nick Piggin.
Changes since v2:
o Created rcu_blocking_is_gp() internal-to-RCU API that
determines whether a call to synchronize_rcu() is itself
a grace period.
o The definition of rcu_blocking_is_gp() for rcuclassic and
rcutree checks to see if but a single CPU is online.
o The definition of rcu_blocking_is_gp() for rcupreempt
checks to see both if but a single CPU is online and if
the system is still in early boot.
This allows rcupreempt to again work correctly if running
on a single CPU after booting is complete.
o Added check to rcupreempt's synchronize_sched() for there
being but one online CPU.
Tested all three variants both SMP and !SMP, booted fine, passed a short
rcutorture test on both x86 and Power.
Located-by: Vegard Nossum <vegard.nossum@gmail.com>
Tested-by: Vegard Nossum <vegard.nossum@gmail.com>
Tested-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: use new cpumask API.
rcu_ctrlblk contains a cpumask, and it's highly optimized so I don't want
a cpumask_var_t (ie. a pointer) for the CONFIG_CPUMASK_OFFSTACK case. It
could use a dangling bitmap, and be allocated in __rcu_init to save memory,
but for the moment we use a bitmap.
(Eventually 'struct cpumask' will be undefined for CONFIG_CPUMASK_OFFSTACK,
so we use a bitmap here to show we really mean it).
We remove on-stack cpumasks, using cpumask_var_t for
rcu_torture_shuffle_tasks() and for_each_cpu_and in force_quiescent_state().
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Impact: increase timeout of debug check feature
Increase RCU stall period timeouts to reduce the likelyhood of
false positives.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch adds stalled-CPU detection to Classic RCU. This capability
is enabled by a new config variable CONFIG_RCU_CPU_STALL_DETECTOR, which
defaults disabled.
This is a debugging feature to detect infinite loops in kernel code, not
something that non-kernel-hackers would be expected to care about.
This feature can detect looping CPUs in !PREEMPT builds and looping CPUs
with preemption disabled in PREEMPT builds. This is essentially a port of
this functionality from the treercu patch, replacing the stall debug patch
that is already in tip/core/rcu (commit 67182ae1c4).
The changes from the patch in tip/core/rcu include making the config
variable name match that in treercu, changing from seconds to jiffies to
avoid spurious warnings, and printing a boot message when this feature
is enabled.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
this is a diagnostic patch for Classic RCU.
The approach is to record a timestamp at the beginning
of the grace period (in rcu_start_batch()), then have
rcu_check_callbacks() complain if:
1. it is running on a CPU that has holding up grace periods for
a long time (say one second). This will identify the culprit
assuming that the culprit has not disabled hardware irqs,
instruction execution, or some such.
2. it is running on a CPU that is not holding up grace periods,
but grace periods have been held up for an even longer time
(say two seconds).
It is enabled via the default-off CONFIG_DEBUG_RCU_STALL kernel parameter.
Rather than exponential backoff, it backs off to once per 30 seconds.
My feeling upon thinking on it was that if you have stalled RCU grace
periods for that long, a few extra printk() messages are probably the
least of your worries...
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Cc: David Witbrodt <dawitbro@sbcglobal.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
On Fri, 2008-08-01 at 16:26 -0700, Linus Torvalds wrote:
> On Fri, 1 Aug 2008, David Miller wrote:
> >
> > Taking more than a few locks of the same class at once is bad
> > news and it's better to find an alternative method.
>
> It's not always wrong.
>
> If you can guarantee that anybody that takes more than one lock of a
> particular class will always take a single top-level lock _first_, then
> that's all good. You can obviously screw up and take the same lock _twice_
> (which will deadlock), but at least you cannot get into ABBA situations.
>
> So maybe the right thing to do is to just teach lockdep about "lock
> protection locks". That would have solved the multi-queue issues for
> networking too - all the actual network drivers would still have taken
> just their single queue lock, but the one case that needs to take all of
> them would have taken a separate top-level lock first.
>
> Never mind that the multi-queue locks were always taken in the same order:
> it's never wrong to just have some top-level serialization, and anybody
> who needs to take <n> locks might as well do <n+1>, because they sure as
> hell aren't going to be on _any_ fastpaths.
>
> So the simplest solution really sounds like just teaching lockdep about
> that one special case. It's not "nesting" exactly, although it's obviously
> related to it.
Do as Linus suggested. The lock protection lock is called nest_lock.
Note that we still have the MAX_LOCK_DEPTH (48) limit to consider, so anything
that spills that it still up shit creek.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This is v2, it's a little deference from v1 that I
had send to lkml.
use ACCESS_ONCE
use rcu_batch_after/rcu_batch_before for batch # comparison.
rcutorture test result:
(hotplugs: do cpu-online/offline once per second)
No CONFIG_NO_HZ: OK, 12hours
No CONFIG_NO_HZ, hotplugs: OK, 12hours
CONFIG_NO_HZ=y: OK, 24hours
CONFIG_NO_HZ=y, hotplugs: Failed.
(Failed also without my patch applied, exactly the same bug occurred,
http://lkml.org/lkml/2008/7/3/24)
v1's email thread:
http://lkml.org/lkml/2008/6/2/539
v1's description:
The code/algorithm of the implement of current callbacks-processing
is very efficient and technical. But when I studied it and I found
a disadvantage:
In multi-CPU systems, when a new RCU callback is being
queued(call_rcu[_bh]), this callback will be invoked after the grace
period for the batch with batch number = rcp->cur+2 has completed
very very likely in current implement. Actually, this callback can be
invoked after the grace period for the batch with
batch number = rcp->cur+1 has completed. The delay of invocation means
that latency of synchronize_rcu() is extended. But more important thing
is that the callbacks usually free memory, and these works are delayed
too! it's necessary for reclaimer to free memory as soon as
possible when left memory is few.
A very simple way can solve this problem:
a field(struct rcu_head::batch) is added to record the batch number for
the RCU callback. And when a new RCU callback is being queued, we
determine the batch number for this callback(head->batch = rcp->cur+1)
and we move this callback to rdp->donelist if we find
that head->batch <= rcp->completed when we process callbacks.
This simple way reduces the wait time for invocation a lot. (about
2.5Grace Period -> 1.5Grace Period in average in multi-CPU systems)
This is my algorithm. But I do not add any field for struct rcu_head
in my implement. We just need to memorize the last 2 batches and
their batch number, because these 2 batches include all entries that
for whom the grace period hasn't completed. So we use a special
linked-list rather than add a field.
Please see the comment of struct rcu_data.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
use a batch number(rcp->pending) instead of a flag(rcp->next_pending)
rcu_start_batch() need to change this flag, so mb()s is needed
for memory-access safe.
but(after this patch applied) rcu_start_batch() do not change
this batch number(rcp->pending), rcp->pending is managed by
__rcu_process_callbacks only, and troublesome mb()s are eliminated.
And codes look simpler and clearer.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Fourth cut of patch to provide the call_rcu_sched(). This is again to
synchronize_sched() as call_rcu() is to synchronize_rcu().
Should be fine for experimental and -rt use, but not ready for inclusion.
With some luck, I will be able to tell Andrew to come out of hiding on
the next round.
Passes multi-day rcutorture sessions with concurrent CPU hotplugging.
Fixes since the first version include a bug that could result in
indefinite blocking (spotted by Gautham Shenoy), better resiliency
against CPU-hotplug operations, and other minor fixes.
Fixes since the second version include reworking grace-period detection
to avoid deadlocks that could happen when running concurrently with
CPU hotplug, adding Mathieu's fix to avoid the softlockup messages,
as well as Mathieu's fix to allow use earlier in boot.
Fixes since the third version include a wrong-CPU bug spotted by
Andrew, getting rid of the obsolete synchronize_kernel API that somehow
snuck back in, merging spin_unlock() and local_irq_restore() in a
few places, commenting the code that checks for quiescent states based
on interrupting from user-mode execution or the idle loop, removing
some inline attributes, and some code-style changes.
Known/suspected shortcomings:
o I still do not entirely trust the sleep/wakeup logic. Next step
will be to use a private snapshot of the CPU online mask in
rcu_sched_grace_period() -- if the CPU wasn't there at the start
of the grace period, we don't need to hear from it. And the
bit about accounting for changes in online CPUs inside of
rcu_sched_grace_period() is ugly anyway.
o It might be good for rcu_sched_grace_period() to invoke
resched_cpu() when a given CPU wasn't responding quickly,
but resched_cpu() is declared static...
This patch also fixes a long-standing bug in the earlier preemptable-RCU
implementation of synchronize_rcu() that could result in loss of
concurrent external changes to a task's CPU affinity mask. I still cannot
remember who reported this...
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Remove the "#ifdef __KERNEL__" tests from unexported header files in
linux/include whose entire contents are wrapped in that preprocessor
test.
Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The PREEMPT-RCU can get stuck if a CPU goes idle and NO_HZ is set. The
idle CPU will not progress the RCU through its grace period and a
synchronize_rcu my get stuck. Without this patch I have a box that will
not boot when PREEMPT_RCU and NO_HZ are set. That same box boots fine
with this patch.
This patch comes from the -rt kernel where it has been tested for
several months.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch implements a new version of RCU which allows its read-side
critical sections to be preempted. It uses a set of counter pairs
to keep track of the read-side critical sections and flips them
when all tasks exit read-side critical section. The details
of this implementation can be found in this paper -
http://www.rdrop.com/users/paulmck/RCU/OLSrtRCU.2006.08.11a.pdf
and the article-
http://lwn.net/Articles/253651/
This patch was developed as a part of the -rt kernel development and
meant to provide better latencies when read-side critical sections of
RCU don't disable preemption. As a consequence of keeping track of RCU
readers, the readers have a slight overhead (optimizations in the paper).
This implementation co-exists with the "classic" RCU implementations
and can be switched to at compiler.
Also includes RCU tracing summarized in debugfs.
[ akpm@linux-foundation.org: build fixes on non-preempt architectures ]
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Reviewed-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch re-organizes the RCU code to enable multiple implementations
of RCU. Users of RCU continues to include rcupdate.h and the
RCU interfaces remain the same. This is in preparation for
subsequently merging the preemptible RCU implementation.
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Paul E. McKenney