Now that the RCU flavors have been consolidated, there is one common
function for checking to see if an expedited RCU grace period has
completed, namely sync_rcu_preempt_exp_done(). Because this function is
no longer specific to RCU-preempt, this commit removes the "_preempt" from
its name. This commit also changes sync_rcu_preempt_exp_done_unlocked()
to sync_rcu_exp_done_unlocked() for the same reason.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The current expedited RCU grace-period code expects that a task
requesting an expedited grace period cannot awaken until that grace
period has reached the wakeup phase. However, it is possible for a long
preemption to result in the waiting task never sleeping. For example,
consider the following sequence of events:
1. Task A starts an expedited grace period by invoking
synchronize_rcu_expedited(). It proceeds normally up to the
wait_event() near the end of that function, and is then preempted
(or interrupted or whatever).
2. The expedited grace period completes, and a kworker task starts
the awaken phase, having incremented the counter and acquired
the rcu_state structure's .exp_wake_mutex. This kworker task
is then preempted or interrupted or whatever.
3. Task A resumes and enters wait_event(), which notes that the
expedited grace period has completed, and thus doesn't sleep.
4. Task B starts an expedited grace period exactly as did Task A,
complete with the preemption (or whatever delay) just before
the call to wait_event().
5. The expedited grace period completes, and another kworker
task starts the awaken phase, having incremented the counter.
However, it blocks when attempting to acquire the rcu_state
structure's .exp_wake_mutex because step 2's kworker task has
not yet released it.
6. Steps 4 and 5 repeat, resulting in overflow of the rcu_node
structure's ->exp_wq[] array.
In theory, this is harmless. Tasks waiting on the various ->exp_wq[]
array will just be spuriously awakened, but they will just sleep again
on noting that the rcu_state structure's ->expedited_sequence value has
not advanced far enough.
In practice, this wastes CPU time and is an accident waiting to happen.
This commit therefore moves the rcu_exp_gp_seq_end() call that officially
ends the expedited grace period (along with associate tracing) until
after the ->exp_wake_mutex has been acquired. This prevents Task A from
awakening prematurely, thus preventing more than one expedited grace
period from being in flight during a previous expedited grace period's
wakeup phase.
Fixes: 3b5f668e71 ("rcu: Overlap wakeups with next expedited grace period")
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>
[ paulmck: Added updated comment. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Tasks waiting within exp_funnel_lock() for an expedited grace period to
elapse can be starved due to the following sequence of events:
1. Tasks A and B both attempt to start an expedited grace
period at about the same time. This grace period will have
completed when the lower four bits of the rcu_state structure's
->expedited_sequence field are 0b'0100', for example, when the
initial value of this counter is zero. Task A wins, and thus
does the actual work of starting the grace period, including
acquiring the rcu_state structure's .exp_mutex and sets the
counter to 0b'0001'.
2. Because task B lost the race to start the grace period, it
waits on ->expedited_sequence to reach 0b'0100' inside of
exp_funnel_lock(). This task therefore blocks on the rcu_node
structure's ->exp_wq[1] field, keeping in mind that the
end-of-grace-period value of ->expedited_sequence (0b'0100')
is shifted down two bits before indexing the ->exp_wq[] field.
3. Task C attempts to start another expedited grace period,
but blocks on ->exp_mutex, which is still held by Task A.
4. The aforementioned expedited grace period completes, so that
->expedited_sequence now has the value 0b'0100'. A kworker task
therefore acquires the rcu_state structure's ->exp_wake_mutex
and starts awakening any tasks waiting for this grace period.
5. One of the first tasks awakened happens to be Task A. Task A
therefore releases the rcu_state structure's ->exp_mutex,
which allows Task C to start the next expedited grace period,
which causes the lower four bits of the rcu_state structure's
->expedited_sequence field to become 0b'0101'.
6. Task C's expedited grace period completes, so that the lower four
bits of the rcu_state structure's ->expedited_sequence field now
become 0b'1000'.
7. The kworker task from step 4 above continues its wakeups.
Unfortunately, the wake_up_all() refetches the rcu_state
structure's .expedited_sequence field:
wake_up_all(&rnp->exp_wq[rcu_seq_ctr(rcu_state.expedited_sequence) & 0x3]);
This results in the wakeup being applied to the rcu_node
structure's ->exp_wq[2] field, which is unfortunate given that
Task B is instead waiting on ->exp_wq[1].
On a busy system, no harm is done (or at least no permanent harm is done).
Some later expedited grace period will redo the wakeup. But on a quiet
system, such as many embedded systems, it might be a good long time before
there was another expedited grace period. On such embedded systems,
this situation could therefore result in a system hang.
This issue manifested as DPM device timeout during suspend (which
usually qualifies as a quiet time) due to a SCSI device being stuck in
_synchronize_rcu_expedited(), with the following stack trace:
schedule()
synchronize_rcu_expedited()
synchronize_rcu()
scsi_device_quiesce()
scsi_bus_suspend()
dpm_run_callback()
__device_suspend()
This commit therefore prevents such delays, timeouts, and hangs by
making rcu_exp_wait_wake() use its "s" argument consistently instead of
refetching from rcu_state.expedited_sequence.
Fixes: 3b5f668e71 ("rcu: Overlap wakeups with next expedited grace period")
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The code in sync_rcu_exp_select_node_cpus() calculates the current
CPU's mask within its rcu_node structure's bitmasks, but this has
already been computed in the ->grpmask field of that CPU's rcu_data
structure. This commit therefore just uses this ->grpmask field.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This fixes a data-race where `atomic_t dynticks` is copied by value. The
copy is performed non-atomically, resulting in a data-race if `dynticks`
is updated concurrently.
This data-race was found with KCSAN:
==================================================================
BUG: KCSAN: data-race in dyntick_save_progress_counter / rcu_irq_enter
write to 0xffff989dbdbe98e0 of 4 bytes by task 10 on cpu 3:
atomic_add_return include/asm-generic/atomic-instrumented.h:78 [inline]
rcu_dynticks_snap kernel/rcu/tree.c:310 [inline]
dyntick_save_progress_counter+0x43/0x1b0 kernel/rcu/tree.c:984
force_qs_rnp+0x183/0x200 kernel/rcu/tree.c:2286
rcu_gp_fqs kernel/rcu/tree.c:1601 [inline]
rcu_gp_fqs_loop+0x71/0x880 kernel/rcu/tree.c:1653
rcu_gp_kthread+0x22c/0x3b0 kernel/rcu/tree.c:1799
kthread+0x1b5/0x200 kernel/kthread.c:255
<snip>
read to 0xffff989dbdbe98e0 of 4 bytes by task 154 on cpu 7:
rcu_nmi_enter_common kernel/rcu/tree.c:828 [inline]
rcu_irq_enter+0xda/0x240 kernel/rcu/tree.c:870
irq_enter+0x5/0x50 kernel/softirq.c:347
<snip>
Reported by Kernel Concurrency Sanitizer on:
CPU: 7 PID: 154 Comm: kworker/7:1H Not tainted 5.3.0+ #5
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
Workqueue: kblockd blk_mq_run_work_fn
==================================================================
Signed-off-by: Marco Elver <elver@google.com>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: rcu@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The "mask_ofl_ipi" is used to track which CPUs get IPIed, however
in the IPI sending loop, "mask_ofl_ipi" along with another variable
"mask_ofl_test" might also get modified to record which CPUs' quiesent
states must be reported by the sync_rcu_exp_select_node_cpus() at
the end of sync_rcu_exp_select_node_cpus(). This overlap of roles
can be confusing, so this patch cleans things a little by using
"mask_ofl_ipi" solely for determining which CPUs must be IPIed and
"mask_ofl_test" for solely determining on behalf of which CPUs
sync_rcu_exp_select_node_cpus() must report a quiscent state.
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Acked-by: Marco Elver <elver@google.com>
The rcu_node structure's ->expmask field is accessed locklessly when
starting a new expedited grace period and when reporting an expedited
RCU CPU stall warning. This commit therefore handles the former by
taking a snapshot of ->expmask while the lock is held and the latter
by applying READ_ONCE() to lockless reads and WRITE_ONCE() to the
corresponding updates.
Link: https://lore.kernel.org/lkml/CANpmjNNmSOagbTpffHr4=Yedckx9Rm2NuGqC9UqE+AOz5f1-ZQ@mail.gmail.com
Reported-by: syzbot+134336b86f728d6e55a0@syzkaller.appspotmail.com
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Acked-by: Marco Elver <elver@google.com>
New tools bring new warnings, and with v5.3 comes:
kernel/rcu/srcutree.c: warning: 'levelspread[<U aa0>]' may be used uninitialized in this function [-Wuninitialized]: => 121:34
This commit suppresses this warning by initializing the full array
to INT_MIN, which will result in failures should any out-of-bounds
references appear.
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
We never set this to false. This probably doesn't affect most people's
runtime because GCC will automatically initialize it to false at certain
common optimization levels. But that behavior is related to a bug in
GCC and obviously should not be relied on.
Fixes: 5d6742b377 ("rcu/nocb: Use rcu_segcblist for no-CBs CPUs")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The RCU-specific resched_cpu() function sends a resched IPI to the
specified CPU, which can be used to force the tick on for a given
nohz_full CPU. This is needed when this nohz_full CPU is looping in the
kernel while blocking the current grace period. However, for the tick
to actually be forced on in all cases, that CPU's rcu_data structure's
->rcu_urgent_qs flag must be set beforehand. This commit therefore
causes rcu_implicit_dynticks_qs() to set this flag prior to invoking
resched_cpu() on a holdout nohz_full CPU.
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
None of rcu_segcblist_set_len(), rcu_segcblist_add_len(), or
rcu_segcblist_xchg_len() are used outside of kernel/rcu/rcu_segcblist.c.
This commit therefore makes them static.
Fixes: eda669a6a2 ("rcu/nocb: Atomic ->len field in rcu_segcblist structure")
Signed-off-by: kbuild test robot <lkp@intel.com>
[ paulmck: "Fixes:" updated per Stephen Rothwell feedback. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
If a nohz_full CPU is idle or executing in userspace, it makes good sense
to keep it out of RCU core processing. After all, the RCU grace-period
kthread can see its quiescent states and all of its callbacks are
offloaded, so there is nothing for RCU core processing to do.
However, if a nohz_full CPU is executing in kernel space, the RCU
grace-period kthread cannot do anything for it, so such a CPU must report
its own quiescent states. This commit therefore makes nohz_full CPUs
skip RCU core processing only if the scheduler-clock interrupt caught
them in idle or in userspace.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Commit 671a63517c ("rcu: Avoid unnecessary softirq when system
is idle") fixed a bug that could result in an indefinite number of
unnecessary invocations of the RCU_SOFTIRQ handler at the trailing edge
of a scheduler-clock interrupt. However, the fix introduced off-CPU
stores to ->core_needs_qs. These writes did not conflict with the
on-CPU stores because the CPU's leaf rcu_node structure's ->lock was
held across all such stores. However, the loads from ->core_needs_qs
were not promoted to READ_ONCE() and, worse yet, the code loading from
->core_needs_qs was written assuming that it was only ever updated by
the corresponding CPU. So operation has been robust, but only by luck.
This situation is therefore an accident waiting to happen.
This commit therefore takes a different approach. Instead of clearing
->core_needs_qs from the grace-period kthread's force-quiescent-state
processing, it modifies the rcu_pending() function to suppress the
rcu_sched_clock_irq() function's call to invoke_rcu_core() if there is no
grace period in progress. This avoids the infinite needless RCU_SOFTIRQ
handlers while still keeping all accesses to ->core_needs_qs local to
the corresponding CPU.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
In some cases, tracing shows that need_heavy_qs is still set even though
urgent_qs was cleared upon reporting of a quiescent state. One such
case is when the softirq reports that a CPU has passed quiescent state.
Commit 671a63517c ("rcu: Avoid unnecessary softirq when system is
idle") fixed a bug where core_needs_qs was not being cleared. In order
to avoid running into similar situations with the urgent-grace-period
flags, this commit causes rcu_disable_urgency_upon_qs(), previously
rcu_disable_tick_upon_qs(), to clear the urgency hints, ->rcu_urgent_qs
and ->rcu_need_heavy_qs. Note that it is possible for CPUs to go
offline with these urgency hints still set. This is handled because
rcu_disable_urgency_upon_qs() is also invoked during the online process.
Because these hints can be cleared both by the corresponding CPU and by
the grace-period kthread, this commit also adds a number of READ_ONCE()
and WRITE_ONCE() calls.
Tested overnight with rcutorture running for 60 minutes on all
configurations of RCU.
Signed-off-by: "Joel Fernandes (Google)" <joel@joelfernandes.org>
[ paulmck: Clear urgency flags in rcu_disable_urgency_upon_qs(). ]
[ paulmck: Remove ->core_needs_qs from the set cleared at quiescent state. ]
[ paulmck: Make rcu_disable_urgency_upon_qs static per kbuild test robot. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
There is interrupt-exit code that forces on the tick for nohz_full CPUs
failing to respond to the current grace period in a timely fashion.
However, this code must compare ->dynticks_nmi_nesting to the value 2
in the interrupt-exit fastpath. This commit therefore moves this code
to the interrupt-entry fastpath, where a lighter-weight comparison to
zero may be used.
Reported-by: Joel Fernandes <joel@joelfernandes.org>
[ paulmck: Apply Joel Fernandes TICK_DEP_MASK_RCU->TICK_DEP_BIT_RCU fix. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
CPUs running for long time periods in the kernel in nohz_full mode
might leave the scheduling-clock interrupt disabled for then full
duration of their in-kernel execution. This can (among other things)
delay grace periods. This commit therefore forces the tick back on
for any nohz_full CPU that is failing to pass through a quiescent state
upon return from interrupt, which the resched_cpu() will induce.
Reported-by: Joel Fernandes <joel@joelfernandes.org>
[ paulmck: Clear ->rcu_forced_tick as reported by Joel Fernandes testing. ]
[ paulmck: Apply Joel Fernandes TICK_DEP_MASK_RCU->TICK_DEP_BIT_RCU fix. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_torture_fwd_prog_nr() tests the ability of RCU to tolerate
in-kernel busy loops. It invokes rcu_torture_fwd_prog_cond_resched()
within its delay loop, which, in PREEMPT && NO_HZ_FULL kernels results
in the occasional direct call to schedule(). Now, this direct call to
schedule() is appropriate for call_rcu() flood testing, in which either
the kernel should restrain itself or userspace transitions will supply
the needed restraint. But in pure in-kernel loops, the occasional
cond_resched() should do the job.
This commit therefore makes rcu_torture_fwd_prog_nr() use cond_resched()
instead of rcu_torture_fwd_prog_cond_resched() in order to increase the
brutality of this aspect of rcutorture testing.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, each of six different types of failure triggers a
single WARN_ON_ONCE(), and it is then necessary to stare at the
rcu_torture_stats(), Reader Pipe, and Reader Batch lines looking for
inappropriately non-zero values. This can be annoying and error-prone,
so this commit provides a separate WARN_ON_ONCE() for each of the
six error conditions and adds short comments to each to ease error
identification.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The variable rcu_perf_writer_state is declared and initialized,
but is never actually referenced. Remove it to clean code.
Signed-off-by: Ethan Hansen <1ethanhansen@gmail.com>
[ paulmck: Also removed unused macros assigned to that variable. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The function rcutorture_record_progress() is declared in rcu.h, but is
never used. This commit therefore removes rcutorture_record_progress()
to clean code.
Signed-off-by: Ethan Hansen <1ethanhansen@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
During an actual call_rcu() flood, there would be frequent trips to
userspace (in-kernel call_rcu() floods must be otherwise housebroken).
Userspace execution on nohz_full CPUs implies an RCU dyntick idle/not-idle
transition pair, so this commit adds emulation of that pair.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
CPU-hotplug removal operations run the multi_cpu_stop() function, which
relies on the scheduler to gain control from whatever is running on the
various online CPUs, including any nohz_full CPUs running long loops in
kernel-mode code. Lack of the scheduler-clock interrupt on such CPUs
can delay multi_cpu_stop() for several minutes and can also result in
RCU CPU stall warnings. This commit therefore causes CPU-hotplug removal
operations to enable the scheduler-clock interrupt on all online CPUs.
[ paulmck: Apply Joel Fernandes TICK_DEP_MASK_RCU->TICK_DEP_BIT_RCU fix. ]
[ paulmck: Apply simplifications suggested by Frederic Weisbecker. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When multi_cpu_stop() loops waiting for other tasks, it can trigger an RCU
CPU stall warning. This can be misleading because what is instead needed
is information on whatever task is blocking multi_cpu_stop(). This commit
therefore inserts an RCU quiescent state into the multi_cpu_stop()
function's waitloop.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Readers and callback flooders in the rcutorture stress-test suite run for
extended time periods by design. They do take pains to relinquish the
CPU from time to time, but in some cases this relies on the scheduler
being active, which in turn relies on the scheduler-clock interrupt
firing from time to time.
This commit therefore forces scheduling-clock interrupts within
these loops. While in the area, this commit also prevents
rcu_torture_reader()'s occasional timed sleeps from delaying shutdown.
[ paulmck: Apply Joel Fernandes TICK_DEP_MASK_RCU->TICK_DEP_BIT_RCU fix. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Callback invocation can run for a significant time period, and within
CONFIG_NO_HZ_FULL=y kernels, this period will be devoid of scheduler-clock
interrupts. In-kernel execution without such interrupts can cause all
manner of malfunction, with RCU CPU stall warnings being but one result.
This commit therefore forces scheduling-clock interrupts on whenever more
than a few RCU callbacks are invoked. Because offloaded callback invocation
can be preempted, this forcing is withdrawn on each context switch. This
in turn requires that the loop invoking RCU callbacks reiterate the forcing
periodically.
[ paulmck: Apply Joel Fernandes TICK_DEP_MASK_RCU->TICK_DEP_BIT_RCU fix. ]
[ paulmck: Remove NO_HZ_FULL check per Frederic Weisbecker feedback. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Pull scheduler updates from Ingo Molnar:
- MAINTAINERS: Add Mark Rutland as perf submaintainer, Juri Lelli and
Vincent Guittot as scheduler submaintainers. Add Dietmar Eggemann,
Steven Rostedt, Ben Segall and Mel Gorman as scheduler reviewers.
As perf and the scheduler is getting bigger and more complex,
document the status quo of current responsibilities and interests,
and spread the review pain^H^H^H^H fun via an increase in the Cc:
linecount generated by scripts/get_maintainer.pl. :-)
- Add another series of patches that brings the -rt (PREEMPT_RT) tree
closer to mainline: split the monolithic CONFIG_PREEMPT dependencies
into a new CONFIG_PREEMPTION category that will allow the eventual
introduction of CONFIG_PREEMPT_RT. Still a few more hundred patches
to go though.
- Extend the CPU cgroup controller with uclamp.min and uclamp.max to
allow the finer shaping of CPU bandwidth usage.
- Micro-optimize energy-aware wake-ups from O(CPUS^2) to O(CPUS).
- Improve the behavior of high CPU count, high thread count
applications running under cpu.cfs_quota_us constraints.
- Improve balancing with SCHED_IDLE (SCHED_BATCH) tasks present.
- Improve CPU isolation housekeeping CPU allocation NUMA locality.
- Fix deadline scheduler bandwidth calculations and logic when cpusets
rebuilds the topology, or when it gets deadline-throttled while it's
being offlined.
- Convert the cpuset_mutex to percpu_rwsem, to allow it to be used from
setscheduler() system calls without creating global serialization.
Add new synchronization between cpuset topology-changing events and
the deadline acceptance tests in setscheduler(), which were broken
before.
- Rework the active_mm state machine to be less confusing and more
optimal.
- Rework (simplify) the pick_next_task() slowpath.
- Improve load-balancing on AMD EPYC systems.
- ... and misc cleanups, smaller fixes and improvements - please see
the Git log for more details.
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (53 commits)
sched/psi: Correct overly pessimistic size calculation
sched/fair: Speed-up energy-aware wake-ups
sched/uclamp: Always use 'enum uclamp_id' for clamp_id values
sched/uclamp: Update CPU's refcount on TG's clamp changes
sched/uclamp: Use TG's clamps to restrict TASK's clamps
sched/uclamp: Propagate system defaults to the root group
sched/uclamp: Propagate parent clamps
sched/uclamp: Extend CPU's cgroup controller
sched/topology: Improve load balancing on AMD EPYC systems
arch, ia64: Make NUMA select SMP
sched, perf: MAINTAINERS update, add submaintainers and reviewers
sched/fair: Use rq_lock/unlock in online_fair_sched_group
cpufreq: schedutil: fix equation in comment
sched: Rework pick_next_task() slow-path
sched: Allow put_prev_task() to drop rq->lock
sched/fair: Expose newidle_balance()
sched: Add task_struct pointer to sched_class::set_curr_task
sched: Rework CPU hotplug task selection
sched/{rt,deadline}: Fix set_next_task vs pick_next_task
sched: Fix kerneldoc comment for ia64_set_curr_task
...
Bimodal behavior of rcu_do_batch() is not really suited to Google
applications like gfe servers.
When a process with millions of sockets exits, closing all files
queues two rcu callbacks per socket.
This eventually reaches the point where RCU enters an emergency
mode, where rcu_do_batch() do not return until whole queue is flushed.
Each rcu callback lasts at least 70 nsec, so with millions of
elements, we easily spend more than 100 msec without rescheduling.
Goal of this patch is to avoid the infamous message like following
"need_resched set for > 51999388 ns (52 ticks) without schedule"
We dynamically adjust the number of elements we process, instead
of 10 / INFINITE choices, we use a floor of ~1 % of current entries.
If the number is above 1000, we switch to a time based limit of 3 msec
per batch, adjustable with /sys/module/rcutree/parameters/rcu_resched_ns
Signed-off-by: Eric Dumazet <edumazet@google.com>
[ paulmck: Forward-port and remove debug statements. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
When under overload conditions, __call_rcu_nocb_wake() will wake the
no-CBs GP kthread any time the no-CBs CB kthread is asleep or there
are no ready-to-invoke callbacks, but only after a timer delay. If the
no-CBs GP kthread has a ->nocb_bypass_timer pending, the deferred wakeup
from __call_rcu_nocb_wake() is redundant. This commit therefore makes
__call_rcu_nocb_wake() avoid posting the redundant deferred wakeup if
->nocb_bypass_timer is pending. This requires adding a bit of ordering
of timer actions.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Currently, __call_rcu_nocb_wake() advances callbacks each time that it
detects excessive numbers of callbacks, though only if it succeeds in
conditionally acquiring its leaf rcu_node structure's ->lock. Despite
the conditional acquisition of ->lock, this does increase contention.
This commit therefore avoids advancing callbacks unless there are
callbacks in ->cblist whose grace period has completed and advancing
has not yet been done during this jiffy.
Note that this decision does not take the presence of new callbacks
into account. That is because on this code path, there will always be
at least one new callback, namely the one we just enqueued.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Currently, nocb_cb_wait() advances callbacks on each pass through its
loop, though only if it succeeds in conditionally acquiring its leaf
rcu_node structure's ->lock. Despite the conditional acquisition of
->lock, this does increase contention. This commit therefore avoids
advancing callbacks unless there are callbacks in ->cblist whose grace
period has completed.
Note that nocb_cb_wait() doesn't worry about callbacks that have not
yet been assigned a grace period. The idea is that the only reason for
nocb_cb_wait() to advance callbacks is to allow it to continue invoking
callbacks. Time will tell whether this is the correct choice.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
The rcutree_migrate_callbacks() invokes rcu_advance_cbs() on both the
offlined CPU's ->cblist and that of the surviving CPU, then merges
them. However, after the merge, and of the offlined CPU's callbacks
that were not ready to be invoked will no longer be associated with a
grace-period number. This commit therefore invokes rcu_advance_cbs()
one more time on the merged ->cblist in order to assign a grace-period
number to these callbacks.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
When callbacks are in full flow, the common case is waiting for a
grace period, and this grace period will normally take a few jiffies to
complete. It therefore isn't all that helpful for __call_rcu_nocb_wake()
to do a synchronous wakeup in this case. This commit therefore turns this
into a timer-based deferred wakeup of the no-CBs grace-period kthread.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
This commit causes locking, sleeping, and callback state to be printed
for no-CBs CPUs when the rcutorture writer is delayed sufficiently for
rcutorture to complain.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Use of the rcu_data structure's segmented ->cblist for no-CBs CPUs
takes advantage of unrelated grace periods, thus reducing the memory
footprint in the face of floods of call_rcu() invocations. However,
the ->cblist field is a more-complex rcu_segcblist structure which must
be protected via locking. Even though there are only three entities
which can acquire this lock (the CPU invoking call_rcu(), the no-CBs
grace-period kthread, and the no-CBs callbacks kthread), the contention
on this lock is excessive under heavy stress.
This commit therefore greatly reduces contention by provisioning
an rcu_cblist structure field named ->nocb_bypass within the
rcu_data structure. Each no-CBs CPU is permitted only a limited
number of enqueues onto the ->cblist per jiffy, controlled by a new
nocb_nobypass_lim_per_jiffy kernel boot parameter that defaults to
about 16 enqueues per millisecond (16 * 1000 / HZ). When that limit is
exceeded, the CPU instead enqueues onto the new ->nocb_bypass.
The ->nocb_bypass is flushed into the ->cblist every jiffy or when
the number of callbacks on ->nocb_bypass exceeds qhimark, whichever
happens first. During call_rcu() floods, this flushing is carried out
by the CPU during the course of its call_rcu() invocations. However,
a CPU could simply stop invoking call_rcu() at any time. The no-CBs
grace-period kthread therefore carries out less-aggressive flushing
(every few jiffies or when the number of callbacks on ->nocb_bypass
exceeds (2 * qhimark), whichever comes first). This means that the
no-CBs grace-period kthread cannot be permitted to do unbounded waits
while there are callbacks on ->nocb_bypass. A ->nocb_bypass_timer is
used to provide the needed wakeups.
[ paulmck: Apply Coverity feedback reported by Colin Ian King. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Upcoming ->nocb_lock contention-reduction work requires that the
rcu_segcblist structure's ->len field be concurrently manipulated,
but only if there are no-CBs CPUs in the kernel. This commit
therefore makes this ->len field be an atomic_long_t, but only
in CONFIG_RCU_NOCB_CPU=y kernels.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
When there are excessive numbers of callbacks, and when either the
corresponding no-CBs callback kthread is asleep or there is no more
ready-to-invoke callbacks, and when least one callback is pending,
__call_rcu_nocb_wake() will advance the callbacks, but refrain from
awakening the corresponding no-CBs grace-period kthread. However,
because rcu_advance_cbs_nowake() is used, it is possible (if a bit
unlikely) that the needed advancement could not happen due to a grace
period not being in progress. Plus there will always be at least one
pending callback due to one having just now been enqueued.
This commit therefore attempts to advance callbacks and awakens the
no-CBs grace-period kthread when there are excessive numbers of callbacks
posted and when the no-CBs callback kthread is not in a position to do
anything helpful.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
The sleep/wakeup of the no-CBs grace-period kthreads is synchronized
using the ->nocb_lock of the first CPU corresponding to that kthread.
This commit provides a separate ->nocb_gp_lock for this purpose, thus
reducing contention on ->nocb_lock.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Currently, nocb_cb_wait() unconditionally acquires the leaf rcu_node
->lock to advance callbacks when done invoking the previous batch.
It does this while holding ->nocb_lock, which means that contention on
the leaf rcu_node ->lock visits itself on the ->nocb_lock. This commit
therefore makes this lock acquisition conditional, forgoing callback
advancement when the leaf rcu_node ->lock is not immediately available.
(In this case, the no-CBs grace-period kthread will eventually do any
needed callback advancement.)
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Currently, __call_rcu_nocb_wake() conditionally acquires the leaf rcu_node
structure's ->lock, and only afterwards does rcu_advance_cbs_nowake()
check to see if it is possible to advance callbacks without potentially
needing to awaken the grace-period kthread. Given that the no-awaken
check can be done locklessly, this commit reverses the order, so that
rcu_advance_cbs_nowake() is invoked without holding the leaf rcu_node
structure's ->lock and rcu_advance_cbs_nowake() checks the grace-period
state before conditionally acquiring that lock, thus reducing the number
of needless acquistions of the leaf rcu_node structure's ->lock.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Currently, when the square root of the number of CPUs is rounded down
by int_sqrt(), this round-down is applied to the number of callback
kthreads per grace-period kthreads. This makes almost no difference
for large systems, but results in oddities such as three no-CBs
grace-period kthreads for a five-CPU system, which is a bit excessive.
This commit therefore causes the round-down to apply to the number of
no-CBs grace-period kthreads, so that systems with from four to eight
CPUs have only two no-CBs grace period kthreads.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
A given rcu_data structure's ->nocb_lock can be acquired very frequently
by the corresponding CPU and occasionally by the corresponding no-CBs
grace-period and callbacks kthreads. In particular, these two kthreads
will have frequent gaps between ->nocb_lock acquisitions that are roughly
a grace period in duration. This means that any excessive ->nocb_lock
contention will be due to the CPU's acquisitions, and this in turn
enables a very naive contention-avoidance strategy to be quite effective.
This commit therefore modifies rcu_nocb_lock() to first
attempt a raw_spin_trylock(), and to atomically increment a
separate ->nocb_lock_contended across a raw_spin_lock(). This new
->nocb_lock_contended field is checked in __call_rcu_nocb_wake() when
interrupts are enabled, with a spin-wait for contending acquisitions
to complete, thus allowing the kthreads a chance to acquire the lock.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Currently, the code provides an extra wakeup for the no-CBs grace-period
kthread if one of its CPUs is generating excessive numbers of callbacks.
But satisfying though it is to wake something up when things are going
south, unless the thing being awakened can actually help solve the
problem, that extra wakeup does nothing but consume additional CPU time,
which is exactly what you don't want during a call_rcu() flood.
This commit therefore avoids doing anything if the corresponding
no-CBs callback kthread is going full tilt. Otherwise, if advancing
callbacks immediately might help and if the leaf rcu_node structure's
lock is immediately available, this commit invokes a new variant of
rcu_advance_cbs() that advances callbacks only if doing so won't require
awakening the grace-period kthread (not to be confused with any of the
no-CBs grace-period kthreads).
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
It might be hard to imagine having more than two billion callbacks
queued on a single CPU's ->cblist, but someone will do it sometime.
This commit therefore makes __call_rcu_nocb_wake() handle this situation
by upgrading local variable "len" from "int" to "long".
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Currently, wake_nocb_gp_defer() simply stores whatever waketype was
passed in, which can result in a RCU_NOCB_WAKE_FORCE being downgraded
to RCU_NOCB_WAKE, which could in turn delay callback processing.
This commit therefore adds a check so that wake_nocb_gp_defer() only
updates ->nocb_defer_wakeup when the update increases the forcefulness,
thus avoiding downgrades.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
The __call_rcu_nocb_wake() function and its predecessors set
->qlen_last_fqs_check to zero for the first callback and to LONG_MAX / 2
for forced reawakenings. The former can result in a too-quick reawakening
when there are many callbacks ready to invoke and the latter prevents a
second reawakening. This commit therefore sets ->qlen_last_fqs_check
to the current number of callbacks in both cases. While in the area,
this commit also moves both assignments under ->nocb_lock.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Historically, no-CBs CPUs allowed the scheduler-clock tick to be
unconditionally disabled on any transition to idle or nohz_full userspace
execution (see the rcu_needs_cpu() implementations). Unfortunately,
the checks used by rcu_needs_cpu() are defeated now that no-CBs CPUs
use ->cblist, which might make users of battery-powered devices rather
unhappy. This commit therefore adds explicit rcu_segcblist_is_offloaded()
checks to return to the historical energy-efficient semantics.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Some compilers complain that wait_gp_seq might be used uninitialized
in nocb_gp_wait(). This cannot actually happen because when wait_gp_seq
is uninitialized, needwait_gp must be false, which prevents wait_gp_seq
from being used. But this analysis is apparently beyond some compilers,
so this commit adds a bogus initialization of wait_gp_seq for the sole
purpose of suppressing the false-positive warning.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
