When rcu_barrier() calls rcu_rdp_cpu_online() and observes a CPU off
rnp->qsmaskinitnext, it means that all accesses from the offline CPU
preceding the CPUHP_TEARDOWN_CPU are visible to RCU barrier, including
callbacks expiration and counter updates.
However interrupts can still fire after stop_machine() re-enables
interrupts and before rcutree_report_cpu_dead(). The related accesses
happening between CPUHP_TEARDOWN_CPU and rnp->qsmaskinitnext clearing
are _NOT_ guaranteed to be seen by rcu_barrier() without proper
ordering, especially when callbacks are invoked there to the end, making
rcutree_migrate_callback() bypass barrier_lock.
The following theoretical race example can make rcu_barrier() hang:
CPU 0 CPU 1
----- -----
//cpu_down()
smpboot_park_threads()
//ksoftirqd is parked now
<IRQ>
rcu_sched_clock_irq()
invoke_rcu_core()
do_softirq()
rcu_core()
rcu_do_batch()
// callback storm
// rcu_do_batch() returns
// before completing all
// of them
// do_softirq also returns early because of
// timeout. It defers to ksoftirqd but
// it's parked
</IRQ>
stop_machine()
take_cpu_down()
rcu_barrier()
spin_lock(barrier_lock)
// observes rcu_segcblist_n_cbs(&rdp->cblist) != 0
<IRQ>
do_softirq()
rcu_core()
rcu_do_batch()
//completes all pending callbacks
//smp_mb() implied _after_ callback number dec
</IRQ>
rcutree_report_cpu_dead()
rnp->qsmaskinitnext &= ~rdp->grpmask;
rcutree_migrate_callback()
// no callback, early return without locking
// barrier_lock
//observes !rcu_rdp_cpu_online(rdp)
rcu_barrier_entrain()
rcu_segcblist_entrain()
// Observe rcu_segcblist_n_cbs(rsclp) == 0
// because no barrier between reading
// rnp->qsmaskinitnext and rsclp->len
rcu_segcblist_add_len()
smp_mb__before_atomic()
// will now observe the 0 count and empty
// list, but too late, we enqueue regardless
WRITE_ONCE(rsclp->len, rsclp->len + v);
// ignored barrier callback
// rcu barrier stall...
This could be solved with a read memory barrier, enforcing the message
passing between rnp->qsmaskinitnext and rsclp->len, matching the full
memory barrier after rsclp->len addition in rcu_segcblist_add_len()
performed at the end of rcu_do_batch().
However the rcu_barrier() is complicated enough and probably doesn't
need too many more subtleties. CPU down is a slowpath and the
barrier_lock seldom contended. Solve the issue with unconditionally
locking the barrier_lock on rcutree_migrate_callbacks(). This makes sure
that either rcu_barrier() sees the empty queue or its entrained
callback will be migrated.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_sync structure's ->gp_count field is always accessed under the
protection of that same structure's ->rss_lock field, with the exception
of a pair of WARN_ON_ONCE() calls just prior to acquiring that lock in
functions rcu_sync_exit() and rcu_sync_dtor(). These lockless accesses
are unnecessary and impair KCSAN's ability to catch bugs that might be
inserted via other lockless accesses.
This commit therefore moves those WARN_ON_ONCE() calls under the lock.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
If a CPU is running either a userspace application or a guest OS in
nohz_full mode, it is possible for a system call to occur just as an
RCU grace period is starting. If that CPU also has the scheduling-clock
tick enabled for any reason (such as a second runnable task), and if the
system was booted with rcutree.use_softirq=0, then RCU can add insult to
injury by awakening that CPU's rcuc kthread, resulting in yet another
task and yet more OS jitter due to switching to that task, running it,
and switching back.
In addition, in the common case where that system call is not of
excessively long duration, awakening the rcuc task is pointless.
This pointlessness is due to the fact that the CPU will enter an extended
quiescent state upon returning to the userspace application or guest OS.
In this case, the rcuc kthread cannot do anything that the main RCU
grace-period kthread cannot do on its behalf, at least if it is given
a few additional milliseconds (for example, given the time duration
specified by rcutree.jiffies_till_first_fqs, give or take scheduling
delays).
This commit therefore adds a rcutree.nohz_full_patience_delay kernel
boot parameter that specifies the grace period age (in milliseconds,
rounded to jiffies) before which RCU will refrain from awakening the
rcuc kthread. Preliminary experimentation suggests a value of 1000,
that is, one second. Increasing rcutree.nohz_full_patience_delay will
increase grace-period latency and in turn increase memory footprint,
so systems with constrained memory might choose a smaller value.
Systems with less-aggressive OS-jitter requirements might choose the
default value of zero, which keeps the traditional immediate-wakeup
behavior, thus avoiding increases in grace-period latency.
[ paulmck: Apply Leonardo Bras feedback. ]
Link: https://lore.kernel.org/all/20240328171949.743211-1-leobras@redhat.com/
Reported-by: Leonardo Bras <leobras@redhat.com>
Suggested-by: Leonardo Bras <leobras@redhat.com>
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Leonardo Bras <leobras@redhat.com>
A full memory barrier is necessary at the end of the expedited grace
period to order:
1) The grace period completion (pictured by the GP sequence
number) with all preceding accesses. This pairs with rcu_seq_end()
performed by the concurrent kworker.
2) The grace period completion and subsequent post-GP update side
accesses. Pairs again against rcu_seq_end().
This full barrier is already provided by the final sync_exp_work_done()
test, making the subsequent explicit one redundant. Remove it and
improve comments.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
RCU stall printout fetches the EQS state of a CPU with a preceding full
memory barrier. However there is nothing to order this read against at
this debugging stage. It is inherently racy when performed remotely.
Do a plain read instead.
This was the last user of rcu_dynticks_snap().
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
When the boot CPU initializes the per-CPU data on behalf of all possible
CPUs, a sanity check is performed on each of them to make sure none is
initialized in an extended quiescent state.
This check involves a full memory barrier which is useless at this early
boot stage.
Do a plain access instead.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
When the grace period kthread checks the extended quiescent state
counter of a CPU, full ordering is necessary to ensure that either:
* If the GP kthread observes the remote target in an extended quiescent
state, then that target must observe all accesses prior to the current
grace period, including the current grace period sequence number, once
it exits that extended quiescent state.
or:
* If the GP kthread observes the remote target NOT in an extended
quiescent state, then the target further entering in an extended
quiescent state must observe all accesses prior to the current
grace period, including the current grace period sequence number, once
it enters that extended quiescent state.
This ordering is enforced through a full memory barrier placed right
before taking the first EQS snapshot. However this is superfluous
because the snapshot is taken while holding the target's rnp lock which
provides the necessary ordering through its chain of
smp_mb__after_unlock_lock().
Remove the needless explicit barrier before the snapshot and put a
comment about the implicit barrier newly relied upon here.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When the grace period kthread checks the extended quiescent state
counter of a CPU, full ordering is necessary to ensure that either:
* If the GP kthread observes the remote target in an extended quiescent
state, then that target must observe all accesses prior to the current
grace period, including the current grace period sequence number, once
it exits that extended quiescent state.
or:
* If the GP kthread observes the remote target NOT in an extended
quiescent state, then the target further entering in an extended
quiescent state must observe all accesses prior to the current
grace period, including the current grace period sequence number, once
it enters that extended quiescent state.
This ordering is enforced through a full memory barrier placed right
before taking the first EQS snapshot. However this is superfluous
because the snapshot is taken while holding the target's rnp lock which
provides the necessary ordering through its chain of
smp_mb__after_unlock_lock().
Remove the needless explicit barrier before the snapshot and put a
comment about the implicit barrier newly relied upon here.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When the grace period kthread checks the extended quiescent state
counter of a CPU, full ordering is necessary to ensure that either:
* If the GP kthread observes the remote target in an extended quiescent
state, then that target must observe all accesses prior to the current
grace period, including the current grace period sequence number, once
it exits that extended quiescent state. Also the GP kthread must
observe all accesses performed by the target prior it entering in
EQS.
or:
* If the GP kthread observes the remote target NOT in an extended
quiescent state, then the target further entering in an extended
quiescent state must observe all accesses prior to the current
grace period, including the current grace period sequence number, once
it enters that extended quiescent state. Also the GP kthread later
observing that EQS must also observe all accesses performed by the
target prior it entering in EQS.
This ordering is explicitly performed both on the first EQS snapshot
and on the second one as well through the combination of a preceding
full barrier followed by an acquire read. However the second snapshot's
full memory barrier is redundant and not needed to enforce the above
guarantees:
GP kthread Remote target
---- -----
// Access prior GP
WRITE_ONCE(A, 1)
// first snapshot
smp_mb()
x = smp_load_acquire(EQS)
// Access prior GP
WRITE_ONCE(B, 1)
// EQS enter
// implied full barrier by atomic_add_return()
atomic_add_return(RCU_DYNTICKS_IDX, EQS)
// implied full barrier by atomic_add_return()
READ_ONCE(A)
// second snapshot
y = smp_load_acquire(EQS)
z = READ_ONCE(B)
If the GP kthread above fails to observe the remote target in EQS
(x not in EQS), the remote target will observe A == 1 after further
entering in EQS. Then the second snapshot taken by the GP kthread only
need to be an acquire read in order to observe z == 1.
Therefore remove the needless full memory barrier on second snapshot.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds the get_completed_synchronize_srcu() and the
same_state_synchronize_srcu() functions. The first returns a cookie
that is always interpreted as corresponding to an expired grace period.
The second does an equality comparison of a pair of cookies.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Now that we have polled SRCU grace periods, a grace period can be
started by start_poll_synchronize_srcu() as well as call_srcu(),
synchronize_srcu(), and synchronize_srcu_expedited(). This commit
therefore calls out this new start_poll_synchronize_srcu() possibility
in the comment on the WARN_ON().
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
In the synchronize_rcu() common case, we will have less than
SR_MAX_USERS_WAKE_FROM_GP number of users per GP. Waking up the kworker
is pointless just to free the last injected wait head since at that point,
all the users have already been awakened.
Introduce a new counter to track this and prevent the wakeup in the
common case.
[ paulmck: Remove atomic_dec_return_release in cannot-happen state. ]
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Reviewed-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When RCU-TASKS-TRACE pre-gp takes a snapshot of the current task running
on all online CPUs, no explicit ordering synchronizes properly with a
context switch. This lack of ordering can permit the new task to miss
pre-grace-period update-side accesses. The following diagram, courtesy
of Paul, shows the possible bad scenario:
CPU 0 CPU 1
----- -----
// Pre-GP update side access
WRITE_ONCE(*X, 1);
smp_mb();
r0 = rq->curr;
RCU_INIT_POINTER(rq->curr, TASK_B)
spin_unlock(rq)
rcu_read_lock_trace()
r1 = X;
/* ignore TASK_B */
Either r0==TASK_B or r1==1 is needed but neither is guaranteed.
One possible solution to solve this is to wait for an RCU grace period
at the beginning of the RCU-tasks-trace grace period before taking the
current tasks snaphot. However this would introduce large additional
latencies to RCU-tasks-trace grace periods.
Another solution is to lock the target runqueue while taking the current
task snapshot. This ensures that the update side sees the latest context
switch and subsequent context switches will see the pre-grace-period
update side accesses.
This commit therefore adds runqueue locking to cpu_curr_snapshot().
Fixes: e386b67257 ("rcu-tasks: Eliminate RCU Tasks Trace IPIs to online CPUs")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Fix the following 'make W=1' warnings:
WARNING: modpost: missing MODULE_DESCRIPTION() in kernel/rcu/rcutorture.o
WARNING: modpost: missing MODULE_DESCRIPTION() in kernel/rcu/rcuscale.o
WARNING: modpost: missing MODULE_DESCRIPTION() in kernel/rcu/refscale.o
Signed-off-by: Jeff Johnson <quic_jjohnson@quicinc.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
On powerpc systems, spinlock acquisition does not order prior stores
against later loads. This means that this statement:
rfcp->rfc_next = NULL;
Can be reordered to follow this statement:
WRITE_ONCE(*rfcpp, rfcp);
Which is then a data race with rcu_torture_fwd_prog_cr(), specifically,
this statement:
rfcpn = READ_ONCE(rfcp->rfc_next)
KCSAN located this data race, which represents a real failure on powerpc.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Acked-by: Marco Elver <elver@google.com>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: <kasan-dev@googlegroups.com>
This commit allows rcutorture to test double-call_srcu() when the
CONFIG_DEBUG_OBJECTS_RCU_HEAD Kconfig option is enabled. The non-raw
sdp structure's ->spinlock will be acquired in call_srcu(), hence this
commit also removes the current IRQ and preemption disabling so as to
avoid lockdep complaints.
Link: https://lore.kernel.org/all/20240407112714.24460-1-qiang.zhang1211@gmail.com/
Signed-off-by: Zqiang <qiang.zhang1211@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This reverts commit 28319d6dc5. The race
it fixed was subject to conditions that don't exist anymore since:
1612160b91 ("rcu-tasks: Eliminate deadlocks involving do_exit() and RCU tasks")
This latter commit removes the use of SRCU that used to cover the
RCU-tasks blind spot on exit between the tasklist's removal and the
final preemption disabling. The task is now placed instead into a
temporary list inside which voluntary sleeps are accounted as RCU-tasks
quiescent states. This would disarm the deadlock initially reported
against PID namespace exit.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The bypass lock contention mitigation assumes there can be at most
2 contenders on the bypass lock, following this scheme:
1) One kthread takes the bypass lock
2) Another one spins on it and increment the contended counter
3) A third one (a bypass enqueuer) sees the contended counter on and
busy loops waiting on it to decrement.
However this assumption is wrong. There can be only one CPU to find the
lock contended because call_rcu() (the bypass enqueuer) is the only
bypass lock acquire site that may not already hold the NOCB lock
beforehand, all the other sites must first contend on the NOCB lock.
Therefore step 2) is impossible.
The other problem is that the mitigation assumes that contenders all
belong to the same rdp CPU, which is also impossible for a raw spinlock.
In theory the warning could trigger if the enqueuer holds the bypass
lock and another CPU flushes the bypass queue concurrently but this is
prevented from all flush users:
1) NOCB kthreads only flush if they successfully _tried_ to lock the
bypass lock. So no contention management here.
2) Flush on callbacks migration happen remotely when the CPU is offline.
No concurrency against bypass enqueue.
3) Flush on deoffloading happen either locally with IRQs disabled or
remotely when the CPU is not yet online. No concurrency against
bypass enqueue.
4) Flush on barrier entrain happen either locally with IRQs disabled or
remotely when the CPU is offline. No concurrency against
bypass enqueue.
For those reasons, the bypass lock contention mitigation isn't needed
and is even wrong. Remove it but keep the warning reporting a contended
bypass lock on a remote CPU, to keep unexpected contention awareness.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Upon NOCB deoffloading, the rcuo kthread must be forced to sleep
until the corresponding rdp is ever offloaded again. The deoffloader
clears the SEGCBLIST_OFFLOADED flag, wakes up the rcuo kthread which
then notices that change and clears in turn its SEGCBLIST_KTHREAD_CB
flag before going to sleep, until it ever sees the SEGCBLIST_OFFLOADED
flag again, should a re-offloading happen.
Upon NOCB offloading, the rcuo kthread must be forced to wake up and
handle callbacks until the corresponding rdp is ever deoffloaded again.
The offloader sets the SEGCBLIST_OFFLOADED flag, wakes up the rcuo
kthread which then notices that change and sets in turn its
SEGCBLIST_KTHREAD_CB flag before going to check callbacks, until it
ever sees the SEGCBLIST_OFFLOADED flag cleared again, should a
de-offloading happen again.
This is all a crude ad-hoc and error-prone kthread (un-)parking
re-implementation.
Consolidate the behaviour with the appropriate API instead.
[ paulmck: Apply Qiang Zhang feedback provided in Link: below. ]
Link: https://lore.kernel.org/all/20240509074046.15629-1-qiang.zhang1211@gmail.com/
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_gp_slow_register/unregister() is only useful in tests where
torture_type=rcu, so this commit therefore generates ->gp_slow_register()
and ->gp_slow_unregister() function pointers in the rcu_torture_ops
structure, and slows grace periods only when these function pointers
exist.
Signed-off-by: Zqiang <qiang.zhang1211@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
