| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
The current rcu_batch_end event trace records only the name of the RCU
flavor and the total number of callbacks that remain queued on the
current CPU. This is insufficient for testing and tuning the new
dyntick-idle RCU_FAST_NO_HZ code, so this commit adds idle state along
with whether or not any of the callbacks that were ready to invoke
at the beginning of rcu_do_batch() are still queued.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
No point in having two identical rcu_cpu_stall_suppress declarations,
so remove the more obscure of the two.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The rcu_do_batch() function that invokes callbacks for TREE_RCU and
TREE_PREEMPT_RCU normally throttles callback invocation to avoid degrading
scheduling latency. However, as long as the CPU would otherwise be idle,
there is no downside to continuing to invoke any callbacks that have passed
through their grace periods. In fact, processing such callbacks in a
timely manner has the benefit of increasing the probability that the
CPU can enter the power-saving dyntick-idle mode.
Therefore, this commit allows callback invocation to continue beyond the
preset limit as long as the scheduler does not have some other task to
run and as long as context is that of the idle task or the relevant
RCU kthread.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Because tasks don't nest, the ->dyntick_nesting must always be zero upon
entry to rcu_idle_enter_common(). Therefore, pass "0" rather than the
counter itself.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Because tasks do not nest, rcu_idle_enter() and rcu_idle_exit() do
not need to check for nesting. This commit therefore moves nesting
checks from rcu_idle_enter_common() to rcu_irq_exit() and from
rcu_idle_exit_common() to rcu_irq_enter().
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The current implementation of RCU_FAST_NO_HZ prevents CPUs from entering
dyntick-idle state if they have RCU callbacks pending. Unfortunately,
this has the side-effect of often preventing them from entering this
state, especially if at least one other CPU is not in dyntick-idle state.
However, the resulting per-tick wakeup is wasteful in many cases: if the
CPU has already fully responded to the current RCU grace period, there
will be nothing for it to do until this grace period ends, which will
frequently take several jiffies.
This commit therefore permits a CPU that has done everything that the
current grace period has asked of it (rcu_pending() == 0) even if it
still as RCU callbacks pending. However, such a CPU posts a timer to
wake it up several jiffies later (6 jiffies, based on experience with
grace-period lengths). This wakeup is required to handle situations
that can result in all CPUs being in dyntick-idle mode, thus failing
to ever complete the current grace period. If a CPU wakes up before
the timer goes off, then it cancels that timer, thus avoiding spurious
wakeups.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Fixes and workarounds for a number of issues (for example, that in
df4012edc) make it safe to once again detect dyntick-idle CPUs on the
first pass of force_quiescent_state(), so this commit makes that change.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Assertions in rcu_init_percpu_data() unknowingly relied on outgoing
CPUs being turned off before reaching the idle loop. Unfortunately,
when running under kvm/qemu on x86, CPUs really can get to idle before
begin shut off. These CPUs are then born in dyntick-idle mode from an
RCU perspective, which results in splats in rcu_init_percpu_data() and
in RCU wrongly ignoring those CPUs despite them being active. This in
turn can cause RCU to end grace periods prematurely, potentially freeing
up memory that the newly onlined CPUs were still using. This is most
decidedly not what we need to see in an RCU implementation.
This commit therefore replaces the assertions in rcu_init_percpu_data()
with code that forces RCU's dyntick-idle view of newly onlined CPUs to
match reality.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
With the new implementation of RCU_FAST_NO_HZ, it was possible to hang
RCU grace periods as follows:
o CPU 0 attempts to go idle, cycles several times through the
rcu_prepare_for_idle() loop, then goes dyntick-idle when
RCU needs nothing more from it, while still having at least
on RCU callback pending.
o CPU 1 goes idle with no callbacks.
Both CPUs can then stay in dyntick-idle mode indefinitely, preventing
the RCU grace period from ever completing, possibly hanging the system.
This commit therefore prevents CPUs that have RCU callbacks from entering
dyntick-idle mode. This approach also eliminates the need for the
end-of-grace-period IPIs used previously.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Change from direct comparison of ->pid with zero to is_idle_task().
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
RCU has traditionally relied on idle_cpu() to determine whether a given
CPU is running in the context of an idle task, but commit 908a3283
(Fix idle_cpu()) has invalidated this approach. After commit 908a3283,
idle_cpu() will return true if the current CPU is currently running the
idle task, and will be doing so for the foreseeable future. RCU instead
needs to know whether or not the current CPU is currently running the
idle task, regardless of what the near future might bring.
This commit therefore switches from idle_cpu() to "current->pid != 0".
Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Suggested-by: Carsten Emde <C.Emde@osadl.org>
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Tested-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Currently, RCU does not permit a CPU to enter dyntick-idle mode if that
CPU has any RCU callbacks queued. This means that workloads for which
each CPU wakes up and does some RCU updates every few ticks will never
enter dyntick-idle mode. This can result in significant unnecessary power
consumption, so this patch permits a given to enter dyntick-idle mode if
it has callbacks, but only if that same CPU has completed all current
work for the RCU core. We determine use rcu_pending() to determine
whether a given CPU has completed all current work for the RCU core.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The current code just complains if the current task is not the idle task.
This commit therefore adds printing of the identity of the idle task.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
The trace_rcu_dyntick() trace event did not print both the old and
the new value of the nesting level, and furthermore printed only
the low-order 32 bits of it. This could result in some confusion
when interpreting trace-event dumps, so this commit prints both
the old and the new value, prints the full 64 bits, and also selects
the process-entry/exit increment to print nicely in hexadecimal.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
Report that none of the rcu read lock maps are held while in an RCU
extended quiescent state (the section between rcu_idle_enter()
and rcu_idle_exit()). This helps detect any use of rcu_dereference()
and friends from within the section in idle where RCU is not allowed.
This way we can guarantee an extended quiescent window where the CPU
can be put in dyntick idle mode or can simply aoid to be part of any
global grace period completion while in the idle loop.
Uses of RCU from such mode are totally ignored by RCU, hence the
importance of these checks.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
When setting up an expedited grace period, if there were no readers, the
task will awaken itself. This commit removes this useless self-awakening.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Because rcu_is_cpu_idle() is to be used to check for extended quiescent
states in RCU-preempt read-side critical sections, it cannot assume that
preemption is disabled. And preemption must be disabled when accessing
the dyntick-idle state, because otherwise the following sequence of events
could occur:
1. Task A on CPU 1 enters rcu_is_cpu_idle() and picks up the pointer
to CPU 1's per-CPU variables.
2. Task B preempts Task A and starts running on CPU 1.
3. Task A migrates to CPU 2.
4. Task B blocks, leaving CPU 1 idle.
5. Task A continues execution on CPU 2, accessing CPU 1's dyntick-idle
information using the pointer fetched in step 1 above, and finds
that CPU 1 is idle.
6. Task A therefore incorrectly concludes that it is executing in
an extended quiescent state, possibly issuing a spurious splat.
Therefore, this commit disables preemption within the rcu_is_cpu_idle()
function.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
Earlier versions of RCU used the scheduling-clock tick to detect idleness
by checking for the idle task, but handled idleness differently for
CONFIG_NO_HZ=y. But there are now a number of uses of RCU read-side
critical sections in the idle task, for example, for tracing. A more
fine-grained detection of idleness is therefore required.
This commit presses the old dyntick-idle code into full-time service,
so that rcu_idle_enter(), previously known as rcu_enter_nohz(), is
always invoked at the beginning of an idle loop iteration. Similarly,
rcu_idle_exit(), previously known as rcu_exit_nohz(), is always invoked
at the end of an idle-loop iteration. This allows the idle task to
use RCU everywhere except between consecutive rcu_idle_enter() and
rcu_idle_exit() calls, in turn allowing architecture maintainers to
specify exactly where in the idle loop that RCU may be used.
Because some of the userspace upcall uses can result in what looks
to RCU like half of an interrupt, it is not possible to expect that
the irq_enter() and irq_exit() hooks will give exact counts. This
patch therefore expands the ->dynticks_nesting counter to 64 bits
and uses two separate bitfields to count process/idle transitions
and interrupt entry/exit transitions. It is presumed that userspace
upcalls do not happen in the idle loop or from usermode execution
(though usermode might do a system call that results in an upcall).
The counter is hard-reset on each process/idle transition, which
avoids the interrupt entry/exit error from accumulating. Overflow
is avoided by the 64-bitness of the ->dyntick_nesting counter.
This commit also adds warnings if a non-idle task asks RCU to enter
idle state (and these checks will need some adjustment before applying
Frederic's OS-jitter patches (http://lkml.org/lkml/2011/10/7/246).
In addition, validation of ->dynticks and ->dynticks_nesting is added.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
The ->signaled field was named before complications in the form of
dyntick-idle mode and offlined CPUs. These complications have required
that force_quiescent_state() be implemented as a state machine, instead
of simply unconditionally sending reschedule IPIs. Therefore, this
commit renames ->signaled to ->fqs_state to catch up with the new
force_quiescent_state() reality.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
The changed files were only including linux/module.h for the
EXPORT_SYMBOL infrastructure, and nothing else. Revector them
onto the isolated export header for faster compile times.
Nothing to see here but a whole lot of instances of:
-#include <linux/module.h>
+#include <linux/export.h>
This commit is only changing the kernel dir; next targets
will probably be mm, fs, the arch dirs, etc.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
|
|
It is possible for the CPU that noted the end of the prior grace period
to not need a new one, and therefore to decide to propagate ->completed
throughout the rcu_node tree without starting another grace period.
However, in so doing, it releases the root rcu_node structure's lock,
which can allow some other CPU to start another grace period. The first
CPU will be propagating ->completed in parallel with the second CPU
initializing the rcu_node tree for the new grace period. In theory
this is harmless, but in practice we need to keep things simple.
This commit therefore moves the propagation of ->completed to
rcu_report_qs_rsp(), and refrains from marking the old grace period
as having been completed until it has finished doing this. This
prevents anyone from starting a new grace period concurrently with
marking the old grace period as having been completed.
Of course, the optimization where a CPU needing a new grace period
doesn't bother marking the old one completed is still in effect:
In that case, the marking happens implicitly as part of initializing
the new grace period.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The purpose of rcu_needs_cpu_flush() was to iterate on pushing the
current grace period in order to help the current CPU enter dyntick-idle
mode. However, this can result in failures if the CPU starts entering
dyntick-idle mode, but then backs out. In this case, the call to
rcu_pending() from rcu_needs_cpu_flush() might end up announcing a
non-existing quiescent state.
This commit therefore removes rcu_needs_cpu_flush() in favor of letting
the dyntick-idle machinery at the end of the softirq handler push the
loop along via its call to rcu_pending().
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
RCU boost threads start life at RCU_BOOST_PRIO, while others remain
at RCU_KTHREAD_PRIO. While here, change thread names to match other
kthreads, and adjust rcu_yield() to not override the priority set by
the user. This last change sets the stage for runtime changes to
priority in the -rt tree.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
CPUs set rdp->qs_pending when coming online to resolve races with
grace-period start. However, this means that if RCU is idle, the
just-onlined CPU might needlessly send itself resched IPIs. Adjust
the online-CPU initialization to avoid this, and also to correctly
cause the CPU to respond to the current grace period if needed.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Josh Boyer <jwboyer@redhat.com>
Tested-by: Christian Hoffmann <email@christianhoffmann.info>
|
|
It is possible for an RCU CPU stall to end just as it is detected, in
which case the current code will uselessly dump all CPU's stacks.
This commit therefore checks for this condition and refrains from
sending needless NMIs.
And yes, the stall might also end just after we checked all CPUs and
tasks, but in that case we would at least have given some clue as
to which CPU/task was at fault.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Greater use of RCU during early boot (before the scheduler is operating)
is causing RCU to attempt to start grace periods during that time, which
in turn is resulting in both RCU and the callback functions attempting
to use the scheduler before it is ready.
This commit prevents these problems by prohibiting RCU grace periods
until after the scheduler has spawned the first non-idle task.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
There isn't a whole lot of point in poking the scheduler before there
are other tasks to switch to. This commit therefore adds a check
for rcu_scheduler_fully_active in __rcu_pending() to suppress any
pre-scheduler calls to set_need_resched(). The downside of this approach
is additional runtime overhead in a reasonably hot code path.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The trigger_all_cpu_backtrace() function is a no-op in architectures that
do not define arch_trigger_all_cpu_backtrace. On such architectures, RCU
CPU stall warning messages contain no stack trace information, which makes
debugging quite difficult. This commit therefore substitutes dump_stack()
for architectures that do not define arch_trigger_all_cpu_backtrace,
so that at least the local CPU's stack is dumped as part of the RCU CPU
stall warning message.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
When the ->dynticks field in the rcu_dynticks structure changed to an
atomic_t, its size on 64-bit systems changed from 64 bits to 32 bits.
The local variables in rcu_implicit_dynticks_qs() need to change as
well, hence this commit.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The in_irq() check in rcu_enter_nohz() is redundant because if we really
are in an interrupt, the attempt to re-enter dyntick-idle mode will invoke
rcu_needs_cpu() in any case, which will force the check for RCU callbacks.
So this commit removes the check along with the set_need_resched().
Suggested-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
There is often a delay between the time that a CPU passes through a
quiescent state and the time that this quiescent state is reported to the
RCU core. It is quite possible that the grace period ended before the
quiescent state could be reported, for example, some other CPU might have
deduced that this CPU passed through dyntick-idle mode. It is critically
important that quiescent state be counted only against the grace period
that was in effect at the time that the quiescent state was detected.
Previously, this was handled by recording the number of the last grace
period to complete when passing through a quiescent state. The RCU
core then checks this number against the current value, and rejects
the quiescent state if there is a mismatch. However, one additional
possibility must be accounted for, namely that the quiescent state was
recorded after the prior grace period completed but before the current
grace period started. In this case, the RCU core must reject the
quiescent state, but the recorded number will match. This is handled
when the CPU becomes aware of a new grace period -- at that point,
it invalidates any prior quiescent state.
This works, but is a bit indirect. The new approach records the current
grace period, and the RCU core checks to see (1) that this is still the
current grace period and (2) that this grace period has not yet ended.
This approach simplifies reasoning about correctness, and this commit
changes over to this new approach.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Add trace events to record grace-period start and end, quiescent states,
CPUs noticing grace-period start and end, grace-period initialization,
call_rcu() invocation, tasks blocking in RCU read-side critical sections,
tasks exiting those same critical sections, force_quiescent_state()
detection of dyntick-idle and offline CPUs, CPUs entering and leaving
dyntick-idle mode (except from NMIs), CPUs coming online and going
offline, and CPUs being kicked for staying in dyntick-idle mode for too
long (as in many weeks, even on 32-bit systems).
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
rcu: Add the rcu flavor to callback trace events
The earlier trace events for registering RCU callbacks and for invoking
them did not include the RCU flavor (rcu_bh, rcu_preempt, or rcu_sched).
This commit adds the RCU flavor to those trace events.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
We now have kthreads only for flavors of RCU that support boosting,
so update the now-misleading comments accordingly.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Add a string to the rcu_batch_start() and rcu_batch_end() trace
messages that indicates the RCU type ("rcu_sched", "rcu_bh", or
"rcu_preempt"). The trace messages for the actual invocations
themselves are not marked, as it should be clear from the
rcu_batch_start() and rcu_batch_end() events before and after.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
In order to allow event tracing to distinguish between flavors of
RCU, we need those names in the relevant RCU data structures. TINY_RCU
has avoided them for memory-footprint reasons, so add them only if
CONFIG_RCU_TRACE=y.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This commit adds the trace_rcu_utilization() marker that is to be
used to allow postprocessing scripts compute RCU's CPU utilization,
give or take event-trace overhead. Note that we do not include RCU's
dyntick-idle interface because event tracing requires RCU protection,
which is not available in dyntick-idle mode.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
There was recently some controversy about the overhead of invoking RCU
callbacks. Add TRACE_EVENT()s to obtain fine-grained timings for the
start and stop of a batch of callbacks and also for each callback invoked.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Pull the code that waits for an RCU grace period into a single function,
which is then called by synchronize_rcu() and friends in the case of
TREE_RCU and TREE_PREEMPT_RCU, and from rcu_barrier() and friends in
the case of TINY_RCU and TINY_PREEMPT_RCU.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Under some rare but real combinations of configuration parameters, RCU
callbacks are posted during early boot that use kernel facilities that
are not yet initialized. Therefore, when these callbacks are invoked,
hard hangs and crashes ensue. This commit therefore prevents RCU
callbacks from being invoked until after the scheduler is fully up and
running, as in after multiple tasks have been spawned.
It might well turn out that a better approach is to identify the specific
RCU callbacks that are causing this problem, but that discussion will
wait until such time as someone really needs an RCU callback to be invoked
(as opposed to merely registered) during early boot.
Reported-by: julie Sullivan <kernelmail.jms@gmail.com>
Reported-by: RKK <kulkarni.ravi4@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Tested-by: julie Sullivan <kernelmail.jms@gmail.com>
Tested-by: RKK <kulkarni.ravi4@gmail.com>
|
|
The commit "use softirq instead of kthreads except when RCU_BOOST=y"
just applied #ifdef in place. This commit is a cleanup that moves
the newly #ifdef'ed code to the header file kernel/rcutree_plugin.h.
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This patch #ifdefs RCU kthreads out of the kernel unless RCU_BOOST=y,
thus eliminating context-switch overhead if RCU priority boosting has
not been configured.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Commit a26ac2455ffcf3(rcu: move TREE_RCU from softirq to kthread)
introduced performance regression. In an AIM7 test, this commit degraded
performance by about 40%.
The commit runs rcu callbacks in a kthread instead of softirq. We observed
high rate of context switch which is caused by this. Out test system has
64 CPUs and HZ is 1000, so we saw more than 64k context switch per second
which is caused by RCU's per-CPU kthread. A trace showed that most of
the time the RCU per-CPU kthread doesn't actually handle any callbacks,
but instead just does a very small amount of work handling grace periods.
This means that RCU's per-CPU kthreads are making the scheduler do quite
a bit of work in order to allow a very small amount of RCU-related
processing to be done.
Alex Shi's analysis determined that this slowdown is due to lock
contention within the scheduler. Unfortunately, as Peter Zijlstra points
out, the scheduler's real-time semantics require global action, which
means that this contention is inherent in real-time scheduling. (Yes,
perhaps someone will come up with a workaround -- otherwise, -rt is not
going to do well on large SMP systems -- but this patch will work around
this issue in the meantime. And "the meantime" might well be forever.)
This patch therefore re-introduces softirq processing to RCU, but only
for core RCU work. RCU callbacks are still executed in kthread context,
so that only a small amount of RCU work runs in softirq context in the
common case. This should minimize ksoftirqd execution, allowing us to
skip boosting of ksoftirqd for CONFIG_RCU_BOOST=y kernels.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Tested-by: "Alex,Shi" <alex.shi@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Make the functions creating the kthreads wake them up. Leverage the
fact that the per-node and boost kthreads can run anywhere, thus
dispensing with the need to wake them up once the incoming CPU has
gone fully online.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Daniel J Blueman <daniel.blueman@gmail.com>
|
|
Commit cc3ce5176d83 (rcu: Start RCU kthreads in TASK_INTERRUPTIBLE
state) fudges a sleeping task' state, resulting in the scheduler seeing
a TASK_UNINTERRUPTIBLE task going to sleep, but a TASK_INTERRUPTIBLE
task waking up. The result is unbalanced load calculation.
The problem that patch tried to address is that the RCU threads could
stay in UNINTERRUPTIBLE state for quite a while and triggering the hung
task detector due to on-demand wake-ups.
Cure the problem differently by always giving the tasks at least one
wake-up once the CPU is fully up and running, this will kick them out of
the initial UNINTERRUPTIBLE state and into the regular INTERRUPTIBLE
wait state.
[ The alternative would be teaching kthread_create() to start threads as
INTERRUPTIBLE but that needs a tad more thought. ]
Reported-by: Damien Wyart <damien.wyart@free.fr>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Paul E. McKenney <paul.mckenney@linaro.org>
Link: http://lkml.kernel.org/r/1306755291.1200.2872.camel@twins
Signed-off-by: Ingo Molnar <mingo@elte.hu>
|
|
Upon creation, kthreads are in TASK_UNINTERRUPTIBLE state, which can
result in softlockup warnings. Because some of RCU's kthreads can
legitimately be idle indefinitely, start them in TASK_INTERRUPTIBLE
state in order to avoid those warnings.
Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
|
|
It is not necessary to use waitqueues for the RCU kthreads because
we always know exactly which thread is to be awakened. In addition,
wake_up() only issues an actual wakeup when there is a thread waiting on
the queue, which was why there was an extra explicit wake_up_process()
to get the RCU kthreads started.
Eliminating the waitqueues (and wake_up()) in favor of wake_up_process()
eliminates the need for the initial wake_up_process() and also shrinks
the data structure size a bit. The wakeup logic is placed in a new
rcu_wait() macro.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
|
|
This commit switches manipulations of the rcu_node ->wakemask field
to atomic operations, which allows rcu_cpu_kthread_timer() to avoid
acquiring the rcu_node lock. This should avoid the following lockdep
splat reported by Valdis Kletnieks:
[ 12.872150] usb 1-4: new high speed USB device number 3 using ehci_hcd
[ 12.986667] usb 1-4: New USB device found, idVendor=413c, idProduct=2513
[ 12.986679] usb 1-4: New USB device strings: Mfr=0, Product=0, SerialNumber=0
[ 12.987691] hub 1-4:1.0: USB hub found
[ 12.987877] hub 1-4:1.0: 3 ports detected
[ 12.996372] input: PS/2 Generic Mouse as /devices/platform/i8042/serio1/input/input10
[ 13.071471] udevadm used greatest stack depth: 3984 bytes left
[ 13.172129]
[ 13.172130] =======================================================
[ 13.172425] [ INFO: possible circular locking dependency detected ]
[ 13.172650] 2.6.39-rc6-mmotm0506 #1
[ 13.172773] -------------------------------------------------------
[ 13.172997] blkid/267 is trying to acquire lock:
[ 13.173009] (&p->pi_lock){-.-.-.}, at: [<ffffffff81032d8f>] try_to_wake_up+0x29/0x1aa
[ 13.173009]
[ 13.173009] but task is already holding lock:
[ 13.173009] (rcu_node_level_0){..-...}, at: [<ffffffff810901cc>] rcu_cpu_kthread_timer+0x27/0x58
[ 13.173009]
[ 13.173009] which lock already depends on the new lock.
[ 13.173009]
[ 13.173009]
[ 13.173009] the existing dependency chain (in reverse order) is:
[ 13.173009]
[ 13.173009] -> #2 (rcu_node_level_0){..-...}:
[ 13.173009] [<ffffffff810679b9>] check_prevs_add+0x8b/0x104
[ 13.173009] [<ffffffff81067da1>] validate_chain+0x36f/0x3ab
[ 13.173009] [<ffffffff8106846b>] __lock_acquire+0x369/0x3e2
[ 13.173009] [<ffffffff81068a0f>] lock_acquire+0xfc/0x14c
[ 13.173009] [<ffffffff815697f1>] _raw_spin_lock+0x36/0x45
[ 13.173009] [<ffffffff81090794>] rcu_read_unlock_special+0x8c/0x1d5
[ 13.173009] [<ffffffff8109092c>] __rcu_read_unlock+0x4f/0xd7
[ 13.173009] [<ffffffff81027bd3>] rcu_read_unlock+0x21/0x23
[ 13.173009] [<ffffffff8102cc34>] cpuacct_charge+0x6c/0x75
[ 13.173009] [<ffffffff81030cc6>] update_curr+0x101/0x12e
[ 13.173009] [<ffffffff810311d0>] check_preempt_wakeup+0xf7/0x23b
[ 13.173009] [<ffffffff8102acb3>] check_preempt_curr+0x2b/0x68
[ 13.173009] [<ffffffff81031d40>] ttwu_do_wakeup+0x76/0x128
[ 13.173009] [<ffffffff81031e49>] ttwu_do_activate.constprop.63+0x57/0x5c
[ 13.173009] [<ffffffff81031e96>] scheduler_ipi+0x48/0x5d
[ 13.173009] [<ffffffff810177d5>] smp_reschedule_interrupt+0x16/0x18
[ 13.173009] [<ffffffff815710f3>] reschedule_interrupt+0x13/0x20
[ 13.173009] [<ffffffff810b66d1>] rcu_read_unlock+0x21/0x23
[ 13.173009] [<ffffffff810b739c>] find_get_page+0xa9/0xb9
[ 13.173009] [<ffffffff810b8b48>] filemap_fault+0x6a/0x34d
[ 13.173009] [<ffffffff810d1a25>] __do_fault+0x54/0x3e6
[ 13.173009] [<ffffffff810d447a>] handle_pte_fault+0x12c/0x1ed
[ 13.173009] [<ffffffff810d48f7>] handle_mm_fault+0x1cd/0x1e0
[ 13.173009] [<ffffffff8156cfee>] do_page_fault+0x42d/0x5de
[ 13.173009] [<ffffffff8156a75f>] page_fault+0x1f/0x30
[ 13.173009]
[ 13.173009] -> #1 (&rq->lock){-.-.-.}:
[ 13.173009] [<ffffffff810679b9>] check_prevs_add+0x8b/0x104
[ 13.173009] [<ffffffff81067da1>] validate_chain+0x36f/0x3ab
[ 13.173009] [<ffffffff8106846b>] __lock_acquire+0x369/0x3e2
[ 13.173009] [<ffffffff81068a0f>] lock_acquire+0xfc/0x14c
[ 13.173009] [<ffffffff815697f1>] _raw_spin_lock+0x36/0x45
[ 13.173009] [<ffffffff81027e19>] __task_rq_lock+0x8b/0xd3
[ 13.173009] [<ffffffff81032f7f>] wake_up_new_task+0x41/0x108
[ 13.173009] [<ffffffff810376c3>] do_fork+0x265/0x33f
[ 13.173009] [<ffffffff81007d02>] kernel_thread+0x6b/0x6d
[ 13.173009] [<ffffffff8153a9dd>] rest_init+0x21/0xd2
[ 13.173009] [<ffffffff81b1db4f>] start_kernel+0x3bb/0x3c6
[ 13.173009] [<ffffffff81b1d29f>] x86_64_start_reservations+0xaf/0xb3
[ 13.173009] [<ffffffff81b1d393>] x86_64_start_kernel+0xf0/0xf7
[ 13.173009]
[ 13.173009] -> #0 (&p->pi_lock){-.-.-.}:
[ 13.173009] [<ffffffff81067788>] check_prev_add+0x68/0x20e
[ 13.173009] [<ffffffff810679b9>] check_prevs_add+0x8b/0x104
[ 13.173009] [<ffffffff81067da1>] validate_chain+0x36f/0x3ab
[ 13.173009] [<ffffffff8106846b>] __lock_acquire+0x369/0x3e2
[ 13.173009] [<ffffffff81068a0f>] lock_acquire+0xfc/0x14c
[ 13.173009] [<ffffffff815698ea>] _raw_spin_lock_irqsave+0x44/0x57
[ 13.173009] [<ffffffff81032d8f>] try_to_wake_up+0x29/0x1aa
[ 13.173009] [<ffffffff81032f3c>] wake_up_process+0x10/0x12
[ 13.173009] [<ffffffff810901e9>] rcu_cpu_kthread_timer+0x44/0x58
[ 13.173009] [<ffffffff81045286>] call_timer_fn+0xac/0x1e9
[ 13.173009] [<ffffffff8104556d>] run_timer_softirq+0x1aa/0x1f2
[ 13.173009] [<ffffffff8103e487>] __do_softirq+0x109/0x26a
[ 13.173009] [<ffffffff8157144c>] call_softirq+0x1c/0x30
[ 13.173009] [<ffffffff81003207>] do_softirq+0x44/0xf1
[ 13.173009] [<ffffffff8103e8b9>] irq_exit+0x58/0xc8
[ 13.173009] [<ffffffff81017f5a>] smp_apic_timer_interrupt+0x79/0x87
[ 13.173009] [<ffffffff81570fd3>] apic_timer_interrupt+0x13/0x20
[ 13.173009] [<ffffffff810bd51a>] get_page_from_freelist+0x2aa/0x310
[ 13.173009] [<ffffffff810bdf03>] __alloc_pages_nodemask+0x178/0x243
[ 13.173009] [<ffffffff8101fe2f>] pte_alloc_one+0x1e/0x3a
[ 13.173009] [<ffffffff810d27fe>] __pte_alloc+0x22/0x14b
[ 13.173009] [<ffffffff810d48a8>] handle_mm_fault+0x17e/0x1e0
[ 13.173009] [<ffffffff8156cfee>] do_page_fault+0x42d/0x5de
[ 13.173009] [<ffffffff8156a75f>] page_fault+0x1f/0x30
[ 13.173009]
[ 13.173009] other info that might help us debug this:
[ 13.173009]
[ 13.173009] Chain exists of:
[ 13.173009] &p->pi_lock --> &rq->lock --> rcu_node_level_0
[ 13.173009]
[ 13.173009] Possible unsafe locking scenario:
[ 13.173009]
[ 13.173009] CPU0 CPU1
[ 13.173009] ---- ----
[ 13.173009] lock(rcu_node_level_0);
[ 13.173009] lock(&rq->lock);
[ 13.173009] lock(rcu_node_level_0);
[ 13.173009] lock(&p->pi_lock);
[ 13.173009]
[ 13.173009] *** DEADLOCK ***
[ 13.173009]
[ 13.173009] 3 locks held by blkid/267:
[ 13.173009] #0: (&mm->mmap_sem){++++++}, at: [<ffffffff8156cdb4>] do_page_fault+0x1f3/0x5de
[ 13.173009] #1: (&yield_timer){+.-...}, at: [<ffffffff810451da>] call_timer_fn+0x0/0x1e9
[ 13.173009] #2: (rcu_node_level_0){..-...}, at: [<ffffffff810901cc>] rcu_cpu_kthread_timer+0x27/0x58
[ 13.173009]
[ 13.173009] stack backtrace:
[ 13.173009] Pid: 267, comm: blkid Not tainted 2.6.39-rc6-mmotm0506 #1
[ 13.173009] Call Trace:
[ 13.173009] <IRQ> [<ffffffff8154a529>] print_circular_bug+0xc8/0xd9
[ 13.173009] [<ffffffff81067788>] check_prev_add+0x68/0x20e
[ 13.173009] [<ffffffff8100c861>] ? save_stack_trace+0x28/0x46
[ 13.173009] [<ffffffff810679b9>] check_prevs_add+0x8b/0x104
[ 13.173009] [<ffffffff81067da1>] validate_chain+0x36f/0x3ab
[ 13.173009] [<ffffffff8106846b>] __lock_acquire+0x369/0x3e2
[ 13.173009] [<ffffffff81032d8f>] ? try_to_wake_up+0x29/0x1aa
[ 13.173009] [<ffffffff81068a0f>] lock_acquire+0xfc/0x14c
[ 13.173009] [<ffffffff81032d8f>] ? try_to_wake_up+0x29/0x1aa
[ 13.173009] [<ffffffff810901a5>] ? rcu_check_quiescent_state+0x82/0x82
[ 13.173009] [<ffffffff815698ea>] _raw_spin_lock_irqsave+0x44/0x57
[ 13.173009] [<ffffffff81032d8f>] ? try_to_wake_up+0x29/0x1aa
[ 13.173009] [<ffffffff81032d8f>] try_to_wake_up+0x29/0x1aa
[ 13.173009] [<ffffffff810901a5>] ? rcu_check_quiescent_state+0x82/0x82
[ 13.173009] [<ffffffff81032f3c>] wake_up_process+0x10/0x12
[ 13.173009] [<ffffffff810901e9>] rcu_cpu_kthread_timer+0x44/0x58
[ 13.173009] [<ffffffff810901a5>] ? rcu_check_quiescent_state+0x82/0x82
[ 13.173009] [<ffffffff81045286>] call_timer_fn+0xac/0x1e9
[ 13.173009] [<ffffffff810451da>] ? del_timer+0x75/0x75
[ 13.173009] [<ffffffff810901a5>] ? rcu_check_quiescent_state+0x82/0x82
[ 13.173009] [<ffffffff8104556d>] run_timer_softirq+0x1aa/0x1f2
[ 13.173009] [<ffffffff8103e487>] __do_softirq+0x109/0x26a
[ 13.173009] [<ffffffff8106365f>] ? tick_dev_program_event+0x37/0xf6
[ 13.173009] [<ffffffff810a0e4a>] ? time_hardirqs_off+0x1b/0x2f
[ 13.173009] [<ffffffff8157144c>] call_softirq+0x1c/0x30
[ 13.173009] [<ffffffff81003207>] do_softirq+0x44/0xf1
[ 13.173009] [<ffffffff8103e8b9>] irq_exit+0x58/0xc8
[ 13.173009] [<ffffffff81017f5a>] smp_apic_timer_interrupt+0x79/0x87
[ 13.173009] [<ffffffff81570fd3>] apic_timer_interrupt+0x13/0x20
[ 13.173009] <EOI> [<ffffffff810bd384>] ? get_page_from_freelist+0x114/0x310
[ 13.173009] [<ffffffff810bd51a>] ? get_page_from_freelist+0x2aa/0x310
[ 13.173009] [<ffffffff812220e7>] ? clear_page_c+0x7/0x10
[ 13.173009] [<ffffffff810bd1ef>] ? prep_new_page+0x14c/0x1cd
[ 13.173009] [<ffffffff810bd51a>] get_page_from_freelist+0x2aa/0x310
[ 13.173009] [<ffffffff810bdf03>] __alloc_pages_nodemask+0x178/0x243
[ 13.173009] [<ffffffff810d46b9>] ? __pmd_alloc+0x87/0x99
[ 13.173009] [<ffffffff8101fe2f>] pte_alloc_one+0x1e/0x3a
[ 13.173009] [<ffffffff810d46b9>] ? __pmd_alloc+0x87/0x99
[ 13.173009] [<ffffffff810d27fe>] __pte_alloc+0x22/0x14b
[ 13.173009] [<ffffffff810d48a8>] handle_mm_fault+0x17e/0x1e0
[ 13.173009] [<ffffffff8156cfee>] do_page_fault+0x42d/0x5de
[ 13.173009] [<ffffffff810d915f>] ? sys_brk+0x32/0x10c
[ 13.173009] [<ffffffff810a0e4a>] ? time_hardirqs_off+0x1b/0x2f
[ 13.173009] [<ffffffff81065c4f>] ? trace_hardirqs_off_caller+0x3f/0x9c
[ 13.173009] [<ffffffff812235dd>] ? trace_hardirqs_off_thunk+0x3a/0x3c
[ 13.173009] [<ffffffff8156a75f>] page_fault+0x1f/0x30
[ 14.010075] usb 5-1: new full speed USB device number 2 using uhci_hcd
Reported-by: Valdis Kletnieks <Valdis.Kletnieks@vt.edu>
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-2.6-rcu into core/urgent
|
|
(Note: this was reverted, and is now being re-applied in pieces, with
this being the fifth and final piece. See below for the reason that
it is now felt to be safe to re-apply this.)
Commit d09b62d fixed grace-period synchronization, but left some smp_mb()
invocations in rcu_process_callbacks() that are no longer needed, but
sheer paranoia prevented them from being removed. This commit removes
them and provides a proof of correctness in their absence. It also adds
a memory barrier to rcu_report_qs_rsp() immediately before the update to
rsp->completed in order to handle the theoretical possibility that the
compiler or CPU might move massive quantities of code into a lock-based
critical section. This also proves that the sheer paranoia was not
entirely unjustified, at least from a theoretical point of view.
In addition, the old dyntick-idle synchronization depended on the fact
that grace periods were many milliseconds in duration, so that it could
be assumed that no dyntick-idle CPU could reorder a memory reference
across an entire grace period. Unfortunately for this design, the
addition of expedited grace periods breaks this assumption, which has
the unfortunate side-effect of requiring atomic operations in the
functions that track dyntick-idle state for RCU. (There is some hope
that the algorithms used in user-level RCU might be applied here, but
some work is required to handle the NMIs that user-space applications
can happily ignore. For the short term, better safe than sorry.)
This proof assumes that neither compiler nor CPU will allow a lock
acquisition and release to be reordered, as doing so can result in
deadlock. The proof is as follows:
1. A given CPU declares a quiescent state under the protection of
its leaf rcu_node's lock.
2. If there is more than one level of rcu_node hierarchy, the
last CPU to declare a quiescent state will also acquire the
->lock of the next rcu_node up in the hierarchy, but only
after releasing the lower level's lock. The acquisition of this
lock clearly cannot occur prior to the acquisition of the leaf
node's lock.
3. Step 2 repeats until we reach the root rcu_node structure.
Please note again that only one lock is held at a time through
this process. The acquisition of the root rcu_node's ->lock
must occur after the release of that of the leaf rcu_node.
4. At this point, we set the ->completed field in the rcu_state
structure in rcu_report_qs_rsp(). However, if the rcu_node
hierarchy contains only one rcu_node, then in theory the code
preceding the quiescent state could leak into the critical
section. We therefore precede the update of ->completed with a
memory barrier. All CPUs will therefore agree that any updates
preceding any report of a quiescent state will have happened
before the update of ->completed.
5. Regardless of whether a new grace period is needed, rcu_start_gp()
will propagate the new value of ->completed to all of the leaf
rcu_node structures, under the protection of each rcu_node's ->lock.
If a new grace period is needed immediately, this propagation
will occur in the same critical section that ->completed was
set in, but courtesy of the memory barrier in #4 above, is still
seen to follow any pre-quiescent-state activity.
6. When a given CPU invokes __rcu_process_gp_end(), it becomes
aware of the end of the old grace period and therefore makes
any RCU callbacks that were waiting on that grace period eligible
for invocation.
If this CPU is the same one that detected the end of the grace
period, and if there is but a single rcu_node in the hierarchy,
we will still be in the single critical section. In this case,
the memory barrier in step #4 guarantees that all callbacks will
be seen to execute after each CPU's quiescent state.
On the other hand, if this is a different CPU, it will acquire
the leaf rcu_node's ->lock, and will again be serialized after
each CPU's quiescent state for the old grace period.
On the strength of this proof, this commit therefore removes the memory
barriers from rcu_process_callbacks() and adds one to rcu_report_qs_rsp().
The effect is to reduce the number of memory barriers by one and to
reduce the frequency of execution from about once per scheduling tick
per CPU to once per grace period.
This was reverted do to hangs found during testing by Yinghai Lu and
Ingo Molnar. Frederic Weisbecker supplied Yinghai with tracing that
located the underlying problem, and Frederic also provided the fix.
The underlying problem was that the HARDIRQ_ENTER() macro from
lib/locking-selftest.c invoked irq_enter(), which in turn invokes
rcu_irq_enter(), but HARDIRQ_EXIT() invoked __irq_exit(), which
does not invoke rcu_irq_exit(). This situation resulted in calls
to rcu_irq_enter() that were not balanced by the required calls to
rcu_irq_exit(). Therefore, after these locking selftests completed,
RCU's dyntick-idle nesting count was a large number (for example,
72), which caused RCU to to conclude that the affected CPU was not in
dyntick-idle mode when in fact it was.
RCU would therefore incorrectly wait for this dyntick-idle CPU, resulting
in hangs.
In contrast, with Frederic's patch, which replaces the irq_enter()
in HARDIRQ_ENTER() with an __irq_enter(), these tests don't ever call
either rcu_irq_enter() or rcu_irq_exit(), which works because the CPU
running the test is already marked as not being in dyntick-idle mode.
This means that the rcu_irq_enter() and rcu_irq_exit() calls and RCU
then has no problem working out which CPUs are in dyntick-idle mode and
which are not.
The reason that the imbalance was not noticed before the barrier patch
was applied is that the old implementation of rcu_enter_nohz() ignored
the nesting depth. This could still result in delays, but much shorter
ones. Whenever there was a delay, RCU would IPI the CPU with the
unbalanced nesting level, which would eventually result in rcu_enter_nohz()
being called, which in turn would force RCU to see that the CPU was in
dyntick-idle mode.
The reason that very few people noticed the problem is that the mismatched
irq_enter() vs. __irq_exit() occured only when the kernel was built with
CONFIG_DEBUG_LOCKING_API_SELFTESTS.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
|
|
The old version of rcu_enter_nohz() forced RCU into nohz mode even if
the nesting count was non-zero. This change causes rcu_enter_nohz()
to hold off for non-zero nesting counts.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
