Because that RCU grace-period initialization need no longer exclude
CPU-hotplug operations, this commit eliminates the ->onoff_mutex and
its uses.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Races between CPU hotplug and grace periods can be difficult to resolve,
so the ->onoff_mutex is used to exclude the two events. Unfortunately,
this means that it is impossible for an outgoing CPU to perform the
last bits of its offlining from its last pass through the idle loop,
because sleeplocks cannot be acquired in that context.
This commit avoids these problems by buffering online and offline events
in a new ->qsmaskinitnext field in the leaf rcu_node structures. When a
grace period starts, the events accumulated in this mask are applied to
the ->qsmaskinit field, and, if needed, up the rcu_node tree. The special
case of all CPUs corresponding to a given leaf rcu_node structure being
offline while there are still elements in that structure's ->blkd_tasks
list is handled using a new ->wait_blkd_tasks field. In this case,
propagating the offline bits up the tree is deferred until the beginning
of the grace period after all of the tasks have exited their RCU read-side
critical sections and removed themselves from the list, at which point
the ->wait_blkd_tasks flag is cleared. If one of that leaf rcu_node
structure's CPUs comes back online before the list empties, then the
->wait_blkd_tasks flag is simply cleared.
This of course means that RCU's notion of which CPUs are offline can be
out of date. This is OK because RCU need only wait on CPUs that were
online at the time that the grace period started. In addition, RCU's
force-quiescent-state actions will handle the case where a CPU goes
offline after the grace period starts.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Although cond_resched_rcu_qs() only applies to TASKS_RCU, it is used
in places where it would be useful for it to apply to the normal RCU
flavors, rcu_preempt, rcu_sched, and rcu_bh. This is especially the
case for workloads that aggressively overload the system, particularly
those that generate large numbers of RCU updates on systems running
NO_HZ_FULL CPUs. This commit therefore communicates quiescent states
from cond_resched_rcu_qs() to the normal RCU flavors.
Note that it is unfortunately necessary to leave the old ->passed_quiesce
mechanism in place to allow quiescent states that apply to only one
flavor to be recorded. (Yes, we could decrement ->rcu_qs_ctr_snap in
that case, but that is not so good for debugging of RCU internals.)
In addition, if one of the RCU flavor's grace period has stalled, this
will invoke rcu_momentary_dyntick_idle(), resulting in a heavy-weight
quiescent state visible from other CPUs.
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
[ paulmck: Merge commit from Sasha Levin fixing a bug where __this_cpu()
was used in preemptible code. ]
Long ago, the various ->completed fields were of type long, but now are
unsigned long due to signed-integer-overflow concerns. However, the
various _batches_completed() functions remained of type long, even though
their only purpose in life is to return the corresponding ->completed
field. This patch cleans this up by changing these functions' return
types to unsigned long.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Subtle race conditions can result if a CPU stays in dyntick-idle mode
long enough for the ->gpnum and ->completed fields to wrap. For
example, consider the following sequence of events:
o CPU 1 encounters a quiescent state while waiting for grace period
5 to complete, but then enters dyntick-idle mode.
o While CPU 1 is in dyntick-idle mode, the grace-period counters
wrap around so that the grace period number is now 4.
o Just as CPU 1 exits dyntick-idle mode, grace period 4 completes
and grace period 5 begins.
o The quiescent state that CPU 1 passed through during the old
grace period 5 looks like it applies to the new grace period
5. Therefore, the new grace period 5 completes without CPU 1
having passed through a quiescent state.
This could clearly be a fatal surprise to any long-running RCU read-side
critical section that happened to be running on CPU 1 at the time. At one
time, this was not a problem, given that it takes significant time for
the grace-period counters to overflow even on 32-bit systems. However,
with the advent of NO_HZ_FULL and SMP embedded systems, arbitrarily long
idle periods are now becoming quite feasible. It is therefore time to
close this race.
This commit therefore avoids this race condition by having the
quiescent-state forcing code detect when a CPU is falling too far
behind, and setting a new rcu_data field ->gpwrap when this happens.
Whenever this new ->gpwrap field is set, the CPU's ->gpnum and ->completed
fields are known to be untrustworthy, and can be ignored, along with
any associated quiescent states.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The current RCU CPU stall warning code will print "Stall ended before
state dump start" any time that the stall-warning code is triggered on
a CPU that has already reported a quiescent state for the current grace
period and if all quiescent states have been reported for the current
grace period. However, a true stall can result in these symptoms, for
example, by preventing RCU's grace-period kthreads from ever running
This commit therefore checks for this condition, reporting the end of
the stall only if one of the grace-period counters has actually advanced.
Otherwise, it reports the last time that the grace-period kthread made
meaningful progress. (In normal situations, the grace-period kthread
should make meaningful progress at least every jiffies_till_next_fqs
jiffies.)
Reported-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Miroslav Benes <mbenes@suse.cz>
One way that an RCU CPU stall warning can happen is if the grace-period
kthread is not allowed to execute. One proxy for this kthread's
forward progress is the number of force-quiescent-state (fqs) scans.
This commit therefore adds the number of fqs scans to the RCU CPU stall
warning printouts when CONFIG_RCU_CPU_STALL_INFO=y.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The patch dfeb9765ce ("Allow post-unlock reference for rt_mutex")
ensured rcu-boost safe even the rt_mutex has post-unlock reference.
But rt_mutex allowing post-unlock reference is definitely a bug and it was
fixed by the commit 27e35715df ("rtmutex: Plug slow unlock race").
This fix made the previous patch (dfeb9765ce) useless.
And even worse, the priority-inversion introduced by the the previous
patch still exists.
rcu_read_unlock_special() {
rt_mutex_unlock(&rnp->boost_mtx);
/* Priority-Inversion:
* the current task had been deboosted and preempted as a low
* priority task immediately, it could wait long before reschedule in,
* and the rcu-booster also waits on this low priority task and sleeps.
* This priority-inversion makes rcu-booster can't work
* as expected.
*/
complete(&rnp->boost_completion);
}
Just revert the patch to avoid it.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
When the last CPU associated with a given leaf rcu_node structure
goes offline, something must be done about the tasks queued on that
rcu_node structure. Each of these tasks has been preempted on one of
the leaf rcu_node structure's CPUs while in an RCU read-side critical
section that it have not yet exited. Handling these tasks is the job of
rcu_preempt_offline_tasks(), which migrates them from the leaf rcu_node
structure to the root rcu_node structure.
Unfortunately, this migration has to be done one task at a time because
each tasks allegiance must be shifted from the original leaf rcu_node to
the root, so that future attempts to deal with these tasks will acquire
the root rcu_node structure's ->lock rather than that of the leaf.
Worse yet, this migration must be done with interrupts disabled, which
is not so good for realtime response, especially given that there is
no bound on the number of tasks on a given rcu_node structure's list.
(OK, OK, there is a bound, it is just that it is unreasonably large,
especially on 64-bit systems.) This was not considered a problem back
when rcu_preempt_offline_tasks() was first written because realtime
systems were assumed not to do CPU-hotplug operations while real-time
applications were running. This assumption has proved of dubious validity
given that people are starting to run multiple realtime applications
on a single SMP system and that it is common practice to offline then
online a CPU before starting its real-time application in order to clear
extraneous processing off of that CPU. So we now need CPU hotplug
operations to avoid undue latencies.
This commit therefore avoids migrating these tasks, instead letting
them be dequeued one by one from the original leaf rcu_node structure
by rcu_read_unlock_special(). This means that the clearing of bits
from the upper-level rcu_node structures must be deferred until the
last such task has been dequeued, because otherwise subsequent grace
periods won't wait on them. This commit has the beneficial side effect
of simplifying the CPU-hotplug code for TREE_PREEMPT_RCU, especially in
CONFIG_RCU_BOOST builds.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
This commit abstracts rcu_cleanup_dead_rnp() from rcu_cleanup_dead_cpu()
in preparation for the rework of RCU priority boosting. This new function
will be invoked from rcu_read_unlock_special() in the reworked scheme,
which is why rcu_cleanup_dead_rnp() assumes that the leaf rcu_node
structure's ->qsmaskinit field has already been updated.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The 48a7639ce8 ("rcu: Make callers awaken grace-period kthread")
removed the irq_work_queue(), so the TREE_RCU doesn't need
irq work any more. This commit therefore updates RCU's Kconfig and
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The rcu_barrier() no-callbacks check for no-CBs CPUs has race conditions.
It checks a given CPU's lists of callbacks, and if all three no-CBs lists
are empty, ignores that CPU. However, these three lists could potentially
be empty even when callbacks are present if the check executed just as
the callbacks were being moved from one list to another. It turns out
that recent versions of rcutorture can spot this race.
This commit plugs this hole by consolidating the per-list counts of
no-CBs callbacks into a single count, which is incremented before
the corresponding callback is posted and after it is invoked. Then
rcu_barrier() checks this single count to reliably determine whether
the corresponding CPU has no-CBs callbacks.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The "cpu" argument to rcu_cleanup_after_idle() is always the current
CPU, so drop it. This moves the smp_processor_id() from the caller to
rcu_cleanup_after_idle(), saving argument-passing overhead. Again,
the anticipated cross-CPU uses of these functions has been replaced
by NO_HZ_FULL.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Pranith Kumar <bobby.prani@gmail.com>
The "cpu" argument to rcu_prepare_for_idle() is always the current
CPU, so drop it. This in turn allows two of the uses of "cpu" in
this function to be replaced with a this_cpu_ptr() and the third by
smp_processor_id(), replacing that of the call to rcu_prepare_for_idle().
Again, the anticipated cross-CPU uses of these functions has been replaced
by NO_HZ_FULL.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Pranith Kumar <bobby.prani@gmail.com>
The "cpu" argument to rcu_note_context_switch() is always the current
CPU, so drop it. This in turn allows the "cpu" argument to
rcu_preempt_note_context_switch() to be removed, which allows the sole
use of "cpu" in both functions to be replaced with a this_cpu_ptr().
Again, the anticipated cross-CPU uses of these functions has been
replaced by NO_HZ_FULL.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Pranith Kumar <bobby.prani@gmail.com>
Because rcu_preempt_check_callbacks()'s argument is guaranteed to
always be the current CPU, drop the argument and replace per_cpu()
with __this_cpu_read().
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Pranith Kumar <bobby.prani@gmail.com>
For some functions in kernel/rcu/tree* the rdtp parameter is always
this_cpu_ptr(rdtp). Remove the parameter if constant and calculate the
pointer in function.
This will have the advantage that it is obvious that the address are
all per cpu offsets and thus it will enable the use of this_cpu_ops in
the future.
Signed-off-by: Christoph Lameter <cl@linux.com>
[ paulmck: Forward-ported to rcu/dev, whitespace adjustment. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Pranith Kumar <bobby.prani@gmail.com>
PREEMPT_RCU and TREE_PREEMPT_RCU serve the same function after
TINY_PREEMPT_RCU has been removed. This patch removes TREE_PREEMPT_RCU
and uses PREEMPT_RCU config option in its place.
Signed-off-by: Pranith Kumar <bobby.prani@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Commit 35ce7f29a4 (rcu: Create rcuo kthreads only for onlined CPUs)
avoids creating rcuo kthreads for CPUs that never come online. This
fixes a bug in many instances of firmware: Instead of lying about their
age, these systems instead lie about the number of CPUs that they have.
Before commit 35ce7f29a4, this could result in huge numbers of useless
rcuo kthreads being created.
It appears that experience indicates that I should have told the
people suffering from this problem to fix their broken firmware, but
I instead produced what turned out to be a partial fix. The missing
piece supplied by this commit makes sure that rcu_barrier() knows not to
post callbacks for no-CBs CPUs that have not yet come online, because
otherwise rcu_barrier() will hang on systems having firmware that lies
about the number of CPUs.
It is tempting to simply have rcu_barrier() refuse to post a callback on
any no-CBs CPU that does not have an rcuo kthread. This unfortunately
does not work because rcu_barrier() is required to wait for all pending
callbacks. It is therefore required to wait even for those callbacks
that cannot possibly be invoked. Even if doing so hangs the system.
Given that posting a callback to a no-CBs CPU that does not yet have an
rcuo kthread can hang rcu_barrier(), It is tempting to report an error
in this case. Unfortunately, this will result in false positives at
boot time, when it is perfectly legal to post callbacks to the boot CPU
before the scheduler has started, in other words, before it is legal
to invoke rcu_barrier().
So this commit instead has rcu_barrier() avoid posting callbacks to
CPUs having neither rcuo kthread nor pending callbacks, and has it
complain bitterly if it finds CPUs having no rcuo kthread but some
pending callbacks. And when rcu_barrier() does find CPUs having no rcuo
kthread but pending callbacks, as noted earlier, it has no choice but
to hang indefinitely.
Reported-by: Yanko Kaneti <yaneti@declera.com>
Reported-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Reported-by: Meelis Roos <mroos@linux.ee>
Reported-by: Eric B Munson <emunson@akamai.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Eric B Munson <emunson@akamai.com>
Tested-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Tested-by: Yanko Kaneti <yaneti@declera.com>
Tested-by: Kevin Fenzi <kevin@scrye.com>
Tested-by: Meelis Roos <mroos@linux.ee>
RCU currently uses for_each_possible_cpu() to spawn rcuo kthreads,
which can result in more rcuo kthreads than one would expect, for
example, derRichard reported 64 CPUs worth of rcuo kthreads on an
8-CPU image. This commit therefore creates rcuo kthreads only for
those CPUs that actually come online.
This was reported by derRichard on the OFTC IRC network.
Reported-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Tested-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Currently, RCU spawns kthreads from several different early_initcall()
functions. Although this has served RCU well for quite some time,
as more kthreads are added a more deterministic approach is required.
This commit therefore causes all of RCU's early-boot kthreads to be
spawned from a single early_initcall() function.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Tested-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Currently TASKS_RCU would ignore a CPU running a task in nohz_full=
usermode execution. There would be neither a context switch nor a
scheduling-clock interrupt to tell TASKS_RCU that the task in question
had passed through a quiescent state. The grace period would therefore
extend indefinitely. This commit therefore makes RCU's dyntick-idle
subsystem record the task_struct structure of the task that is running
in dyntick-idle mode on each CPU. The TASKS_RCU grace period can
then access this information and record a quiescent state on
behalf of any CPU running in dyntick-idle usermode.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Commit 96d3fd0d31 (rcu: Break call_rcu() deadlock involving scheduler
and perf) covered the case where __call_rcu_nocb_enqueue() needs to wake
the rcuo kthread due to the queue being initially empty, but did not
do anything for the case where the queue was overflowing. This commit
therefore also defers wakeup for the overflow case.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The nocb callbacks generated before the nocb kthreads are spawned are
enqueued in the nocb queue for later processing. Commit fbce7497ee ("rcu:
Parallelize and economize NOCB kthread wakeups") introduced nocb leader kthreads
which checked the nocb_leader_wake flag to see if there were any such pending
callbacks. A case was reported in which newly spawned leader kthreads were not
processing the pending callbacks as this flag was not set, which led to a boot
hang.
The following commit ensures that the newly spawned nocb kthreads process the
pending callbacks by allowing the kthreads to run immediately after spawning
instead of waiting. This is done by inverting the logic of nocb_leader_wake
tests to nocb_leader_sleep which allows us to use the default initialization of
this flag to 0 to let the kthreads run.
Reported-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Pranith Kumar <bobby.prani@gmail.com>
Link: http://www.spinics.net/lists/kernel/msg1802899.html
[ paulmck: Backported to v3.17-rc2. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Amit Shah <amit.shah@redhat.com>
RCU priority boosting currently checks for boosting via a pointer in
task_struct. However, this is not needed: As Oleg noted, if the
rt_mutex is placed in the rcu_node instead of on the booster's stack,
the boostee can simply check it see if it owns the lock. This commit
makes this change, shrinking task_struct by one pointer and the kernel
by thirteen lines.
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
The current approach to RCU priority boosting uses an rt_mutex strictly
for its priority-boosting side effects. The rt_mutex_init_proxy_locked()
function is used by the booster to initialize the lock as held by the
boostee. The booster then uses rt_mutex_lock() to acquire this rt_mutex,
which priority-boosts the boostee. When the boostee reaches the end
of its outermost RCU read-side critical section, it checks a field in
its task structure to see whether it has been boosted, and, if so, uses
rt_mutex_unlock() to release the rt_mutex. The booster can then go on
to boost the next task that is blocking the current RCU grace period.
But reasonable implementations of rt_mutex_unlock() might result in the
boostee referencing the rt_mutex's data after releasing it. But the
booster might have re-initialized the rt_mutex between the time that the
boostee released it and the time that it later referenced it. This is
clearly asking for trouble, so this commit introduces a completion that
forces the booster to wait until the boostee has completely finished with
the rt_mutex, thus avoiding the case where the booster is re-initializing
the rt_mutex before the last boostee's last reference to that rt_mutex.
This of course does introduce some overhead, but the priority-boosting
code paths are miles from any possible fastpath, and the overhead of
executing the completion will normally be quite small compared to the
overhead of priority boosting and deboosting, so this should be OK.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
An 80-CPU system with a context-switch-heavy workload can require so
many NOCB kthread wakeups that the RCU grace-period kthreads spend several
tens of percent of a CPU just awakening things. This clearly will not
scale well: If you add enough CPUs, the RCU grace-period kthreads would
get behind, increasing grace-period latency.
To avoid this problem, this commit divides the NOCB kthreads into leaders
and followers, where the grace-period kthreads awaken the leaders each of
whom in turn awakens its followers. By default, the number of groups of
kthreads is the square root of the number of CPUs, but this default may
be overridden using the rcutree.rcu_nocb_leader_stride boot parameter.
This reduces the number of wakeups done per grace period by the RCU
grace-period kthread by the square root of the number of CPUs, but of
course by shifting those wakeups to the leaders. In addition, because
the leaders do grace periods on behalf of their respective followers,
the number of wakeups of the followers decreases by up to a factor of two.
Instead of being awakened once when new callbacks arrive and again
at the end of the grace period, the followers are awakened only at
the end of the grace period.
For a numerical example, in a 4096-CPU system, the grace-period kthread
would awaken 64 leaders, each of which would awaken its 63 followers
at the end of the grace period. This compares favorably with the 79
wakeups for the grace-period kthread on an 80-CPU system.
Reported-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Commit ac1bea8578 (Make cond_resched() report RCU quiescent states)
fixed a problem where a CPU looping in the kernel with but one runnable
task would give RCU CPU stall warnings, even if the in-kernel loop
contained cond_resched() calls. Unfortunately, in so doing, it introduced
performance regressions in Anton Blanchard's will-it-scale "open1" test.
The problem appears to be not so much the increased cond_resched() path
length as an increase in the rate at which grace periods complete, which
increased per-update grace-period overhead.
This commit takes a different approach to fixing this bug, mainly by
moving the RCU-visible quiescent state from cond_resched() to
rcu_note_context_switch(), and by further reducing the check to a
simple non-zero test of a single per-CPU variable. However, this
approach requires that the force-quiescent-state processing send
resched IPIs to the offending CPUs. These will be sent only once
the grace period has reached an age specified by the boot/sysfs
parameter rcutree.jiffies_till_sched_qs, or once the grace period
reaches an age halfway to the point at which RCU CPU stall warnings
will be emitted, whichever comes first.
Reported-by: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Lameter <cl@gentwo.org>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
[ paulmck: Made rcu_momentary_dyntick_idle() as suggested by the
ktest build robot. Also fixed smp_mb() comment as noted by
Oleg Nesterov. ]
Merge with e552592e (Reduce overhead of cond_resched() checks for RCU)
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
This commit allows rcutorture to print additional state for the
RCU grace-period kthreads in cases where RCU seems reluctant to
start a new grace period.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
The rcu_start_gp_advanced() function currently uses irq_work_queue()
to defer wakeups of the RCU grace-period kthread. This deferring
is necessary to avoid RCU-scheduler deadlocks involving the rcu_node
structure's lock, meaning that RCU cannot call any of the scheduler's
wake-up functions while holding one of these locks.
Unfortunately, the second and subsequent calls to irq_work_queue() are
ignored, and the first call will be ignored (aside from queuing the work
item) if the scheduler-clock tick is turned off. This is OK for many
uses, especially those where irq_work_queue() is called from an interrupt
or softirq handler, because in those cases the scheduler-clock-tick state
will be re-evaluated, which will turn the scheduler-clock tick back on.
On the next tick, any deferred work will then be processed.
However, this strategy does not always work for RCU, which can be invoked
at process level from idle CPUs. In this case, the tick might never
be turned back on, indefinitely defering a grace-period start request.
Note that the RCU CPU stall detector cannot see this condition, because
there is no RCU grace period in progress. Therefore, we can (and do!)
see long tens-of-seconds stalls in grace-period handling. In theory,
we could see a full grace-period hang, but rcutorture testing to date
has seen only the tens-of-seconds stalls. Event tracing demonstrates
that irq_work_queue() is being called repeatedly to no effect during
these stalls: The "newreq" event appears repeatedly from a task that is
not one of the grace-period kthreads.
In theory, irq_work_queue() might be fixed to avoid this sort of issue,
but RCU's requirements are unusual and it is quite straightforward to pass
wake-up responsibility up through RCU's call chain, so that the wakeup
happens when the offending locks are released.
This commit therefore makes this change. The rcu_start_gp_advanced(),
rcu_start_future_gp(), rcu_accelerate_cbs(), rcu_advance_cbs(),
__note_gp_changes(), and rcu_start_gp() functions now return a boolean
which indicates when a wake-up is needed. A new rcu_gp_kthread_wake()
does the wakeup when it is necessary and safe to do so: No self-wakes,
no wake-ups if the ->gp_flags field indicates there is no need (as in
someone else did the wake-up before we got around to it), and no wake-ups
before the grace-period kthread has been created.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
The ->preemptible field in rcu_data is only initialized in the function
rcu_init_percpu_data(), and never used. This commit therefore removes
this field.
Signed-off-by: Iulia Manda <iulia.manda21@gmail.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
In the old days, the only source of requests for future grace periods
was NOCB CPUs. This has changed: CPUs routinely post requests for
future grace periods in order to promote power efficiency and reduce
OS jitter with minimal impact on grace-period latency. This commit
therefore updates cpu_needs_another_gp() to invoke rcu_future_needs_gp()
instead of rcu_nocb_needs_gp(). The latter is no longer used, so is
now removed. This commit also adds tracing for the irq_work_queue()
wakeup case.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
All of the RCU source files have the usual GPL header, which contains a
long-obsolete postal address for FSF. To avoid the need to track the
FSF office's movements, this commit substitutes the URL where GPL may
be found.
Reported-by: Greg KH <gregkh@linuxfoundation.org>
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Whenever a CPU receives a scheduling-clock interrupt, RCU checks to see
if the RCU core needs anything from this CPU. If so, RCU raises
RCU_SOFTIRQ to carry out any needed processing.
This approach has worked well historically, but it is undesirable on
NO_HZ_FULL CPUs. Such CPUs are expected to spend almost all of their time
in userspace, so that scheduling-clock interrupts can be disabled while
there is only one runnable task on the CPU in question. Unfortunately,
raising any softirq has the potential to wake up ksoftirqd, which would
provide the second runnable task on that CPU, preventing disabling of
scheduling-clock interrupts.
What is needed instead is for RCU to leave NO_HZ_FULL CPUs alone,
relying on the grace-period kthreads' quiescent-state forcing to
do any needed RCU work on behalf of those CPUs.
This commit therefore refrains from raising RCU_SOFTIRQ on any
NO_HZ_FULL CPUs during any grace periods that have been in effect
for less than one second. The one-second limit handles the case
where an inappropriate workload is running on a NO_HZ_FULL CPU
that features lots of scheduling-clock interrupts, but no idle
or userspace time.
Reported-by: Mike Galbraith <bitbucket@online.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Mike Galbraith <bitbucket@online.de>
Toasted-by: Frederic Weisbecker <fweisbec@gmail.com>
Dave Jones got the following lockdep splat:
> ======================================================
> [ INFO: possible circular locking dependency detected ]
> 3.12.0-rc3+ #92 Not tainted
> -------------------------------------------------------
> trinity-child2/15191 is trying to acquire lock:
> (&rdp->nocb_wq){......}, at: [<ffffffff8108ff43>] __wake_up+0x23/0x50
>
> but task is already holding lock:
> (&ctx->lock){-.-...}, at: [<ffffffff81154c19>] perf_event_exit_task+0x109/0x230
>
> which lock already depends on the new lock.
>
>
> the existing dependency chain (in reverse order) is:
>
> -> #3 (&ctx->lock){-.-...}:
> [<ffffffff810cc243>] lock_acquire+0x93/0x200
> [<ffffffff81733f90>] _raw_spin_lock+0x40/0x80
> [<ffffffff811500ff>] __perf_event_task_sched_out+0x2df/0x5e0
> [<ffffffff81091b83>] perf_event_task_sched_out+0x93/0xa0
> [<ffffffff81732052>] __schedule+0x1d2/0xa20
> [<ffffffff81732f30>] preempt_schedule_irq+0x50/0xb0
> [<ffffffff817352b6>] retint_kernel+0x26/0x30
> [<ffffffff813eed04>] tty_flip_buffer_push+0x34/0x50
> [<ffffffff813f0504>] pty_write+0x54/0x60
> [<ffffffff813e900d>] n_tty_write+0x32d/0x4e0
> [<ffffffff813e5838>] tty_write+0x158/0x2d0
> [<ffffffff811c4850>] vfs_write+0xc0/0x1f0
> [<ffffffff811c52cc>] SyS_write+0x4c/0xa0
> [<ffffffff8173d4e4>] tracesys+0xdd/0xe2
>
> -> #2 (&rq->lock){-.-.-.}:
> [<ffffffff810cc243>] lock_acquire+0x93/0x200
> [<ffffffff81733f90>] _raw_spin_lock+0x40/0x80
> [<ffffffff810980b2>] wake_up_new_task+0xc2/0x2e0
> [<ffffffff81054336>] do_fork+0x126/0x460
> [<ffffffff81054696>] kernel_thread+0x26/0x30
> [<ffffffff8171ff93>] rest_init+0x23/0x140
> [<ffffffff81ee1e4b>] start_kernel+0x3f6/0x403
> [<ffffffff81ee1571>] x86_64_start_reservations+0x2a/0x2c
> [<ffffffff81ee1664>] x86_64_start_kernel+0xf1/0xf4
>
> -> #1 (&p->pi_lock){-.-.-.}:
> [<ffffffff810cc243>] lock_acquire+0x93/0x200
> [<ffffffff8173419b>] _raw_spin_lock_irqsave+0x4b/0x90
> [<ffffffff810979d1>] try_to_wake_up+0x31/0x350
> [<ffffffff81097d62>] default_wake_function+0x12/0x20
> [<ffffffff81084af8>] autoremove_wake_function+0x18/0x40
> [<ffffffff8108ea38>] __wake_up_common+0x58/0x90
> [<ffffffff8108ff59>] __wake_up+0x39/0x50
> [<ffffffff8110d4f8>] __call_rcu_nocb_enqueue+0xa8/0xc0
> [<ffffffff81111450>] __call_rcu+0x140/0x820
> [<ffffffff81111b8d>] call_rcu+0x1d/0x20
> [<ffffffff81093697>] cpu_attach_domain+0x287/0x360
> [<ffffffff81099d7e>] build_sched_domains+0xe5e/0x10a0
> [<ffffffff81efa7fc>] sched_init_smp+0x3b7/0x47a
> [<ffffffff81ee1f4e>] kernel_init_freeable+0xf6/0x202
> [<ffffffff817200be>] kernel_init+0xe/0x190
> [<ffffffff8173d22c>] ret_from_fork+0x7c/0xb0
>
> -> #0 (&rdp->nocb_wq){......}:
> [<ffffffff810cb7ca>] __lock_acquire+0x191a/0x1be0
> [<ffffffff810cc243>] lock_acquire+0x93/0x200
> [<ffffffff8173419b>] _raw_spin_lock_irqsave+0x4b/0x90
> [<ffffffff8108ff43>] __wake_up+0x23/0x50
> [<ffffffff8110d4f8>] __call_rcu_nocb_enqueue+0xa8/0xc0
> [<ffffffff81111450>] __call_rcu+0x140/0x820
> [<ffffffff81111bb0>] kfree_call_rcu+0x20/0x30
> [<ffffffff81149abf>] put_ctx+0x4f/0x70
> [<ffffffff81154c3e>] perf_event_exit_task+0x12e/0x230
> [<ffffffff81056b8d>] do_exit+0x30d/0xcc0
> [<ffffffff8105893c>] do_group_exit+0x4c/0xc0
> [<ffffffff810589c4>] SyS_exit_group+0x14/0x20
> [<ffffffff8173d4e4>] tracesys+0xdd/0xe2
>
> other info that might help us debug this:
>
> Chain exists of:
> &rdp->nocb_wq --> &rq->lock --> &ctx->lock
>
> Possible unsafe locking scenario:
>
> CPU0 CPU1
> ---- ----
> lock(&ctx->lock);
> lock(&rq->lock);
> lock(&ctx->lock);
> lock(&rdp->nocb_wq);
>
> *** DEADLOCK ***
>
> 1 lock held by trinity-child2/15191:
> #0: (&ctx->lock){-.-...}, at: [<ffffffff81154c19>] perf_event_exit_task+0x109/0x230
>
> stack backtrace:
> CPU: 2 PID: 15191 Comm: trinity-child2 Not tainted 3.12.0-rc3+ #92
> ffffffff82565b70 ffff880070c2dbf8 ffffffff8172a363 ffffffff824edf40
> ffff880070c2dc38 ffffffff81726741 ffff880070c2dc90 ffff88022383b1c0
> ffff88022383aac0 0000000000000000 ffff88022383b188 ffff88022383b1c0
> Call Trace:
> [<ffffffff8172a363>] dump_stack+0x4e/0x82
> [<ffffffff81726741>] print_circular_bug+0x200/0x20f
> [<ffffffff810cb7ca>] __lock_acquire+0x191a/0x1be0
> [<ffffffff810c6439>] ? get_lock_stats+0x19/0x60
> [<ffffffff8100b2f4>] ? native_sched_clock+0x24/0x80
> [<ffffffff810cc243>] lock_acquire+0x93/0x200
> [<ffffffff8108ff43>] ? __wake_up+0x23/0x50
> [<ffffffff8173419b>] _raw_spin_lock_irqsave+0x4b/0x90
> [<ffffffff8108ff43>] ? __wake_up+0x23/0x50
> [<ffffffff8108ff43>] __wake_up+0x23/0x50
> [<ffffffff8110d4f8>] __call_rcu_nocb_enqueue+0xa8/0xc0
> [<ffffffff81111450>] __call_rcu+0x140/0x820
> [<ffffffff8109bc8f>] ? local_clock+0x3f/0x50
> [<ffffffff81111bb0>] kfree_call_rcu+0x20/0x30
> [<ffffffff81149abf>] put_ctx+0x4f/0x70
> [<ffffffff81154c3e>] perf_event_exit_task+0x12e/0x230
> [<ffffffff81056b8d>] do_exit+0x30d/0xcc0
> [<ffffffff810c9af5>] ? trace_hardirqs_on_caller+0x115/0x1e0
> [<ffffffff810c9bcd>] ? trace_hardirqs_on+0xd/0x10
> [<ffffffff8105893c>] do_group_exit+0x4c/0xc0
> [<ffffffff810589c4>] SyS_exit_group+0x14/0x20
> [<ffffffff8173d4e4>] tracesys+0xdd/0xe2
The underlying problem is that perf is invoking call_rcu() with the
scheduler locks held, but in NOCB mode, call_rcu() will with high
probability invoke the scheduler -- which just might want to use its
locks. The reason that call_rcu() needs to invoke the scheduler is
to wake up the corresponding rcuo callback-offload kthread, which
does the job of starting up a grace period and invoking the callbacks
afterwards.
One solution (championed on a related problem by Lai Jiangshan) is to
simply defer the wakeup to some point where scheduler locks are no longer
held. Since we don't want to unnecessarily incur the cost of such
deferral, the task before us is threefold:
1. Determine when it is likely that a relevant scheduler lock is held.
2. Defer the wakeup in such cases.
3. Ensure that all deferred wakeups eventually happen, preferably
sooner rather than later.
We use irqs_disabled_flags() as a proxy for relevant scheduler locks
being held. This works because the relevant locks are always acquired
with interrupts disabled. We may defer more often than needed, but that
is at least safe.
The wakeup deferral is tracked via a new field in the per-CPU and
per-RCU-flavor rcu_data structure, namely ->nocb_defer_wakeup.
This flag is checked by the RCU core processing. The __rcu_pending()
function now checks this flag, which causes rcu_check_callbacks()
to initiate RCU core processing at each scheduling-clock interrupt
where this flag is set. Of course this is not sufficient because
scheduling-clock interrupts are often turned off (the things we used to
be able to count on!). So the flags are also checked on entry to any
state that RCU considers to be idle, which includes both NO_HZ_IDLE idle
state and NO_HZ_FULL user-mode-execution state.
This approach should allow call_rcu() to be invoked regardless of what
locks you might be holding, the key word being "should".
Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
When an RCU CPU stall warning occurs, the CPU invokes resched_cpu() on
itself. This can help move the grace period forward in some situations,
but it would be even better to do this -before- the RCU CPU stall warning.
This commit therefore causes resched_cpu() to be called every five jiffies
once the system is halfway to an RCU CPU stall warning.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
