When rcu_tasks_trace_postgp() function detects an RCU Tasks Trace
CPU stall, it adds all tasks blocking the current grace period to
a list, invoking get_task_struct() on each to prevent them from
being freed while on the list. It then traverses that list,
printing stall-warning messages for each one that is still blocking
the current grace period and removing it from the list. The list
removal invokes the matching put_task_struct().
This of course means that in the admittedly unlikely event that some
task executes its outermost rcu_read_unlock_trace() in the meantime, it
won't be removed from the list and put_task_struct() won't be executing,
resulting in a task_struct leak. This commit therefore makes the list
removal and put_task_struct() unconditional, stopping the leak.
Note further that this bug can occur only after an RCU Tasks Trace CPU
stall warning, which by default only happens after a grace period has
extended for ten minutes (yes, not a typo, minutes).
Fixes: 4593e772b5 ("rcu-tasks: Add stall warnings for RCU Tasks Trace")
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: <bpf@vger.kernel.org>
Cc: <stable@vger.kernel.org> # 5.7.x
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The more intense grace-period processing resulting from the 50x RCU
Tasks Trace grace-period speedups exposed the following race condition:
o Task A running on CPU 0 executes rcu_read_lock_trace(),
entering a read-side critical section.
o When Task A eventually invokes rcu_read_unlock_trace()
to exit its read-side critical section, this function
notes that the ->trc_reader_special.s flag is zero and
and therefore invoke wil set ->trc_reader_nesting to zero
using WRITE_ONCE(). But before that happens...
o The RCU Tasks Trace grace-period kthread running on some other
CPU interrogates Task A, but this fails because this task is
currently running. This kthread therefore sends an IPI to CPU 0.
o CPU 0 receives the IPI, and thus invokes trc_read_check_handler().
Because Task A has not yet cleared its ->trc_reader_nesting
counter, this function sees that Task A is still within its
read-side critical section. This function therefore sets the
->trc_reader_nesting.b.need_qs flag, AKA the .need_qs flag.
Except that Task A has already checked the .need_qs flag, which
is part of the ->trc_reader_special.s flag. The .need_qs flag
therefore remains set until Task A's next rcu_read_unlock_trace().
o Task A now invokes synchronize_rcu_tasks_trace(), which cannot
start a new grace period until the current grace period completes.
And thus cannot return until after that time.
But Task A's .need_qs flag is still set, which prevents the current
grace period from completing. And because Task A is blocked, it
will never execute rcu_read_unlock_trace() until its call to
synchronize_rcu_tasks_trace() returns.
We are therefore deadlocked.
This race is improbable, but 80 hours of rcutorture made it happen twice.
The race was possible before the grace-period speedup, but roughly 50x
less probable. Several thousand hours of rcutorture would have been
necessary to have a reasonable chance of making this happen before this
50x speedup.
This commit therefore eliminates this deadlock by setting
->trc_reader_nesting to a large negative number before checking the
.need_qs and zeroing (or decrementing with respect to its initial
value) ->trc_reader_nesting. For its part, the IPI handler's
trc_read_check_handler() function adds a check for negative values,
deferring evaluation of the task in this case. Taken together, these
changes avoid this deadlock scenario.
Fixes: 276c410448 ("rcu-tasks: Split ->trc_reader_need_end")
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: <bpf@vger.kernel.org>
Cc: <stable@vger.kernel.org> # 5.7.x
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The various RCU tasks flavors currently wait 100 milliseconds between each
grace period in order to prevent CPU-bound loops and to favor efficiency
over latency. However, RCU Tasks Trace needs to have a grace-period
latency of roughly 25 milliseconds, which is completely infeasible given
the 100-millisecond per-grace-period sleep. This commit therefore reduces
this sleep duration to 5 milliseconds (or one jiffy, whichever is longer)
in kernels built with CONFIG_TASKS_TRACE_RCU_READ_MB=y.
Link: https://lore.kernel.org/bpf/CAADnVQK_AiX+S_L_A4CQWT11XyveppBbQSQgH_qWGyzu_E8Yeg@mail.gmail.com/
Reported-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: <bpf@vger.kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Many workloads are quite sensitive to IPIs, and such workloads should
build kernels with CONFIG_TASKS_TRACE_RCU_READ_MB=y to prevent RCU
Tasks Trace from using them under normal conditions. However, other
workloads are quite happy to permit more IPIs if doing so makes BPF
program updates go faster. This commit therefore sets the default
value for the rcupdate.rcu_task_ipi_delay kernel parameter to zero for
kernels that have been built with CONFIG_TASKS_TRACE_RCU_READ_MB=n,
while retaining the old default of (HZ / 10) for kernels that have
indicated an aversion to IPIs via CONFIG_TASKS_TRACE_RCU_READ_MB=y.
Link: https://lore.kernel.org/bpf/CAADnVQK_AiX+S_L_A4CQWT11XyveppBbQSQgH_qWGyzu_E8Yeg@mail.gmail.com/
Reported-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: <bpf@vger.kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Commit 8344496e8b ("rcu-tasks: Conditionally compile
show_rcu_tasks_gp_kthreads()") introduced conditional
compilation of several functions, but forgot one occurrence of
show_rcu_tasks_classic_gp_kthread() that causes the compiler to warn of
an unused static function. This commit uses "static inline" to avoid
these complaints and possibly also to avoid emitting an actual definition
of this function.
Fixes: 8344496e8b ("rcu-tasks: Conditionally compile show_rcu_tasks_gp_kthreads()")
Cc: <stable@vger.kernel.org> # 5.8.x
Reported-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The RCU Tasks Trace grace periods are too slow, as in 40x slower than
those of RCU Tasks. This is due to my having assumed a one-second grace
period was OK, and thus not having optimized any further. This commit
provides the first step in this optimization process, namely by allowing
the task_list scan backoff interval to be specified on a per-flavor basis,
and then speeding up the scans for RCU Tasks Trace. However, kernels
built with CONFIG_TASKS_TRACE_RCU_READ_MB=y continue to use the old slower
backoff, consistent with that Kconfig option's goal of reducing IPIs.
Link: https://lore.kernel.org/bpf/CAADnVQK_AiX+S_L_A4CQWT11XyveppBbQSQgH_qWGyzu_E8Yeg@mail.gmail.com/
Reported-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: <bpf@vger.kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The n_heavy_reader_attempts, n_heavy_reader_updates, and
n_heavy_reader_ofl_updates variables are not used outside of their
translation unit, so this commit marks them static.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
CPUs can go offline shortly after kfree_call_rcu() has been invoked,
which can leave memory stranded until those CPUs come back online.
This commit therefore drains the kcrp of each CPU, not just the
ones that happen to be online.
Acked-by: Joel Fernandes <joel@joelfernandes.org>
Signed-off-by: Zqiang <qiang.zhang@windriver.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_segcblist_accelerate() function returns true iff it is necessary
to request another grace period. A tracing session showed that this
function unnecessarily requests grace periods.
For example, consider the following sequence of events:
1. Callbacks are queued only on the NEXT segment of CPU A's callback list.
2. CPU A runs RCU_SOFTIRQ, accelerating these callbacks from NEXT to WAIT.
3. Thus rcu_segcblist_accelerate() returns true, requesting grace period N.
4. RCU's grace-period kthread wakes up on CPU B and starts grace period N.
4. CPU A notices the new grace period and invokes RCU_SOFTIRQ.
5. CPU A's RCU_SOFTIRQ again invokes rcu_segcblist_accelerate(), but
there are no new callbacks. However, rcu_segcblist_accelerate()
nevertheless (uselessly) requests a new grace period N+1.
This extra grace period results in additional lock contention and also
additional wakeups, all for no good reason.
This commit therefore adds a check to rcu_segcblist_accelerate() that
prevents the return of true when there are no new callbacks.
This change reduces the number of grace periods (GPs) and wakeups in each
of eleven five-second rcutorture runs as follows:
+----+-------------------+-------------------+
| # | Number of GPs | Number of Wakeups |
+====+=========+=========+=========+=========+
| 1 | With | Without | With | Without |
+----+---------+---------+---------+---------+
| 2 | 75 | 89 | 113 | 119 |
+----+---------+---------+---------+---------+
| 3 | 62 | 91 | 105 | 123 |
+----+---------+---------+---------+---------+
| 4 | 60 | 79 | 98 | 110 |
+----+---------+---------+---------+---------+
| 5 | 63 | 79 | 99 | 112 |
+----+---------+---------+---------+---------+
| 6 | 57 | 89 | 96 | 123 |
+----+---------+---------+---------+---------+
| 7 | 64 | 85 | 97 | 118 |
+----+---------+---------+---------+---------+
| 8 | 58 | 83 | 98 | 113 |
+----+---------+---------+---------+---------+
| 9 | 57 | 77 | 89 | 104 |
+----+---------+---------+---------+---------+
| 10 | 66 | 82 | 98 | 119 |
+----+---------+---------+---------+---------+
| 11 | 52 | 82 | 83 | 117 |
+----+---------+---------+---------+---------+
The reduction in the number of wakeups ranges from 5% to 40%.
Cc: urezki@gmail.com
[ paulmck: Rework commit log and comment. ]
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds an rcutorture.leakpointer module parameter that
intentionally leaks an RCU-protected pointer out of the RCU read-side
critical section and checks to see if the corresponding grace period
has elapsed, emitting a WARN_ON_ONCE() if so. This module parameter can
be used to test facilities like CONFIG_RCU_STRICT_GRACE_PERIOD that end
grace periods quickly.
While in the area, also document rcutorture.irqreader, which was
previously left out.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, registering and unregistering the OOM notifier is done
right before and after the test, respectively. This will not work
well for multi-threaded tests, so this commit hoists this registering
and unregistering up into the rcu_torture_fwd_prog_init() and
rcu_torture_fwd_prog_cleanup() functions.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently in the unlikely event that buf fails to be allocated it
is dereferenced a few times. Use the errexit flag to determine if
buf should be written to to avoid the null pointer dereferences.
Addresses-Coverity: ("Dereference after null check")
Fixes: f518f154ec ("refperf: Dynamically allocate experiment-summary output buffer")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The current rcutorture forward-progress code assumes that it is the
only cause of out-of-memory (OOM) events. For script-based rcutorture
testing, this assumption is in fact correct. However, testing based
on modprobe/rmmod might well encounter external OOM events, which could
happen at any time.
This commit therefore properly synchronizes the interaction between
rcutorture's forward-progress testing and its OOM notifier by adding a
global mutex.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The conversion of rcu_fwds to dynamic allocation failed to actually
allocate the required structure. This commit therefore allocates it,
frees it, and updates rcu_fwds accordingly. While in the area, it
abstracts the cleanup actions into rcu_torture_fwd_prog_cleanup().
Fixes: 5155be9994 ("rcutorture: Dynamically allocate rcu_fwds structure")
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds code to print the grace-period number at the start
of the test along with both the grace-period number and the number of
elapsed grace periods at the end of the test. Note that variants of
RCU)without the notion of a grace-period number (for example, Tiny RCU)
just print zeroes.
[ paulmck: Adjust commit log. ]
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
KCSAN is now in mainline, so this commit removes the stubs for the
data_race(), ASSERT_EXCLUSIVE_WRITER(), and ASSERT_EXCLUSIVE_ACCESS()
macros.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The "cpu" parameter to rcu_report_qs_rdp() is not used, with rdp->cpu
being used instead. Furtheremore, every call to rcu_report_qs_rdp()
invokes it on rdp->cpu. This commit therefore removes this unused "cpu"
parameter and converts a check of rdp->cpu against smp_processor_id()
to a WARN_ON_ONCE().
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The CONFIG_PREEMPT=n instance of rcu_read_unlock is even more
aggressively than that of CONFIG_PREEMPT=y in deferring reporting
quiescent states to the RCU core. This is just what is wanted in normal
use because it reduces overhead, but the resulting delay is not what
is wanted for kernels built with CONFIG_RCU_STRICT_GRACE_PERIOD=y.
This commit therefore adds an rcu_read_unlock_strict() function that
checks for exceptional conditions, and reports the newly started
quiescent state if it is safe to do so, also doing a spin-delay if
requested via rcutree.rcu_unlock_delay. This commit also adds a call
to rcu_read_unlock_strict() from the CONFIG_PREEMPT=n instance of
__rcu_read_unlock().
[ paulmck: Fixed bug located by kernel test robot <lkp@intel.com> ]
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
A kernel built with CONFIG_RCU_STRICT_GRACE_PERIOD=y needs a quiescent
state to appear very shortly after a CPU has noticed a new grace period.
Placing an RCU reader immediately after this point is ineffective because
this normally happens in softirq context, which acts as a big RCU reader.
This commit therefore introduces a new per-CPU work_struct, which is
used at the end of rcu_core() processing to schedule an RCU read-side
critical section from within a clean environment.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The goal of this series is to increase the probability of tools like
KASAN detecting that an RCU-protected pointer was used outside of its
RCU read-side critical section. Thus far, the approach has been to make
grace periods and callback processing happen faster. Another approach
is to delay the pointer leaker. This commit therefore allows a delay
to be applied to exit from RCU read-side critical sections.
This slowdown is specified by a new rcutree.rcu_unlock_delay kernel boot
parameter that specifies this delay in microseconds, defaulting to zero.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, each CPU discovers the end of a given grace period on its
own time, which is again good for efficiency but bad for fast grace
periods, given that it is things like kfree() within the RCU callbacks
that will cause trouble for pointers leaked from RCU read-side critical
sections. This commit therefore uses on_each_cpu() to IPI each CPU
after grace-period cleanup in order to inform each CPU of the end of
the old grace period in a timely manner, but only in kernels build with
CONFIG_RCU_STRICT_GRACE_PERIOD=y.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, each CPU discovers the beginning of a given grace period
on its own time, which is again good for efficiency but bad for fast
grace periods. This commit therefore uses on_each_cpu() to IPI each
CPU after grace-period initialization in order to inform each CPU of
the new grace period in a timely manner, but only in kernels build with
CONFIG_RCU_STRICT_GRACE_PERIOD=y.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
A given CPU normally notes a new grace period during one RCU_SOFTIRQ,
but avoids reporting the corresponding quiescent state until some later
RCU_SOFTIRQ. This leisurly approach improves efficiency by increasing
the number of update requests served by each grace period, but is not
what is needed for kernels built with CONFIG_RCU_STRICT_GRACE_PERIOD=y.
This commit therefore adds a new rcu_strict_gp_check_qs() function
which, in CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels, simply enters and
immediately exist an RCU read-side critical section. If the CPU is
in a quiescent state, the rcu_read_unlock() will attempt to report an
immediate quiescent state. This rcu_strict_gp_check_qs() function is
invoked from note_gp_changes(), so that a CPU just noticing a new grace
period might immediately report a quiescent state for that grace period.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_preempt_deferred_qs_irqrestore() function is invoked at
the end of an RCU read-side critical section (for example, directly
from rcu_read_unlock()) and, if .need_qs is set, invokes rcu_qs() to
report the new quiescent state. This works, except that rcu_qs() only
updates per-CPU state, leaving reporting of the actual quiescent state
to a later call to rcu_report_qs_rdp(), for example from within a later
RCU_SOFTIRQ instance. Although this approach is exactly what you want if
you are more concerned about efficiency than about short grace periods,
in CONFIG_RCU_STRICT_GRACE_PERIOD=y kernels, short grace periods are
the name of the game.
This commit therefore makes rcu_preempt_deferred_qs_irqrestore() directly
invoke rcu_report_qs_rdp() in CONFIG_RCU_STRICT_GRACE_PERIOD=y, thus
shortening grace periods.
Historical note: To the best of my knowledge, causing rcu_read_unlock()
to directly report a quiescent state first appeared in Jim Houston's
and Joe Korty's JRCU. This is the second instance of a Linux-kernel RCU
feature being inspired by JRCU, the first being RCU callback offloading
(as in the RCU_NOCB_CPU Kconfig option).
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The ->rcu_read_unlock_special.b.need_qs field in the task_struct
structure indicates that the RCU core needs a quiscent state from the
corresponding task. The __rcu_read_unlock() function checks this (via
an eventual call to rcu_preempt_deferred_qs_irqrestore()), and if set
reports a quiscent state immediately upon exit from the outermost RCU
read-side critical section.
Currently, this flag is only set when the scheduling-clock interrupt
decides that the current RCU grace period is too old, as in about
one full second too old. But if the kernel has been built with
CONFIG_RCU_STRICT_GRACE_PERIOD=y, we clearly do not want to wait that
long. This commit therefore sets the .need_qs field immediately at the
start of the RCU read-side critical section from within __rcu_read_lock()
in order to unconditionally enlist help from __rcu_read_unlock().
But note the additional check for rcu_state.gp_kthread, which prevents
attempts to awaken RCU's grace-period kthread during early boot before
there is a scheduler. Leaving off this check results in early boot hangs.
So early that there is no console output. Thus, this additional check
fails until such time as RCU's grace-period kthread has been created,
avoiding these empty-console hangs.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The value of DEFAULT_RCU_BLIMIT is normally set to 10, the idea being to
avoid needless response-time degradation due to RCU callback invocation.
However, when CONFIG_RCU_STRICT_GRACE_PERIOD=y it is better to avoid
throttling callback execution in order to better detect pointer
leaks from RCU read-side critical sections. This commit therefore
sets the value of DEFAULT_RCU_BLIMIT to 1000 in kernels built with
CONFIG_RCU_STRICT_GRACE_PERIOD=y.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
If there are idle CPUs, RCU's grace-period kthread will wait several
jiffies before even thinking about polling them. This promotes
efficiency, which is normally a good thing, but when the kernel
has been built with CONFIG_RCU_STRICT_GRACE_PERIOD=y, we care more
about short grace periods. This commit therefore restricts the
default jiffies_till_first_fqs value to zero in kernels built with
CONFIG_RCU_STRICT_GRACE_PERIOD=y, which causes RCU's grace-period kthread
to poll for idle CPUs immediately after starting a grace period.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Because strict RCU grace periods will complete more quickly, they will
experience greater lock contention on each leaf rcu_node structure's
->lock. This commit therefore reduces the leaf fanout in order to reduce
this lock contention.
Note that this also has the effect of reducing the number of CPUs
supported to 16 in the case of CONFIG_RCU_FANOUT_LEAF=2 or 81 in the
case of CONFIG_RCU_FANOUT_LEAF=3. However, greater numbers of CPUs are
probably a bad idea when using CONFIG_RCU_STRICT_GRACE_PERIOD=y. Those
wishing to live dangerously are free to edit their kernel/rcu/Kconfig
files accordingly.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
People running automated tests have asked for a way to make RCU minimize
grace-period duration in order to increase the probability of KASAN
detecting a pointer being improperly leaked from an RCU read-side critical
section, for example, like this:
rcu_read_lock();
p = rcu_dereference(gp);
do_something_with(p); // OK
rcu_read_unlock();
do_something_else_with(p); // BUG!!!
The rcupdate.rcu_expedited boot parameter is a start in this direction,
given that it makes calls to synchronize_rcu() instead invoke the faster
(and more wasteful) synchronize_rcu_expedited(). However, this does
nothing to shorten RCU grace periods that are instead initiated by
call_rcu(), and RCU pointer-leak bugs can involve call_rcu() just as
surely as they can synchronize_rcu().
This commit therefore adds a RCU_STRICT_GRACE_PERIOD Kconfig option
that will be used to shorten normal (non-expedited) RCU grace periods.
This commit also dumps out a message when this option is in effect.
Later commits will actually shorten grace periods.
Reported-by Jann Horn <jannh@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit further avoids conflation of rcuperf with the kernel's perf
feature by renaming kernel/rcu/rcuperf.c to kernel/rcu/rcuscale.c, and
also by similarly renaming the functions and variables inside this file.
This has the side effect of changing the names of the kernel boot
parameters, so kernel-parameters.txt and ver_functions.sh are also
updated. The rcutorture --torture type was also updated from rcuperf
to rcuscale.
[ paulmck: Fix bugs located by Stephen Rothwell. ]
Reported-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The x86/entry work removed all uses of __rcu_is_watching(), therefore
this commit removes it entirely.
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The RCU grace-period kthread's force-quiescent state (FQS) loop should
never see an offline CPU that has not yet reported a quiescent state.
After all, the offline CPU should have reported a quiescent state
during the CPU-offline process, or, failing that, by rcu_gp_init()
if it ran concurrently with either the CPU going offline or the last
task on a leaf rcu_node structure exiting its RCU read-side critical
section while all CPUs corresponding to that structure are offline.
The FQS loop should therefore complain if it does see an offline CPU
that has not yet reported a quiescent state.
And it does, but only once the grace period has been in force for a
full second. This commit therefore makes this warning more aggressive,
so that it will trigger as soon as the condition makes its appearance.
Light testing with TREE03 and hotplug shows no warnings. This commit
also converts the warning to WARN_ON_ONCE() in order to stave off possible
log spam.
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Since at least v4.19, the FQS loop no longer reports quiescent states
for offline CPUs except in emergency situations.
This commit therefore fixes the comment in rcu_gp_init() to match the
current code.
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit increases RCU's ability to defend itself by emitting a warning
if one of the nocb CB kthreads invokes the GP kthread's wait function.
This warning augments a similar check that is carried out at the end
of rcutorture testing and when RCU CPU stall warnings are emitted.
The problem with those checks is that the miscreants have long since
departed and disposed of any and all evidence.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
When the rcu_cpu_started per-CPU variable was added by commit
f64c6013a2 ("rcu/x86: Provide early rcu_cpu_starting() callback"),
there were multiple sets of per-CPU rcu_data structures. Therefore, the
rcu_cpu_started flag was added as a separate per-CPU variable. But now
there is only one set of per-CPU rcu_data structures, so this commit
moves rcu_cpu_started to a new ->cpu_started field in that structure.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Given that sysfs can change the value of rcu_cpu_stall_ftrace_dump at any
time, this commit adds a READ_ONCE() to the accesses to that variable.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Given that sysfs can change the value of rcu_kick_kthreads at any time,
this commit adds a READ_ONCE() to the sole access to that variable.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Given that sysfs can change the value of rcu_resched_ns at any time,
this commit adds a READ_ONCE() to the sole access to that variable.
While in the area, this commit also adds bounds checking, clamping the
value to at least a millisecond, but no longer than a second.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Given that sysfs can change the value of rcu_divisor at any time, this
commit adds a READ_ONCE to the sole access to that variable. While in
the area, this commit also adds bounds checking, clamping the value to
a shift that makes sense for a signed long.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_data structure's ->nocb_timer field is used to defer wakeups of
the corresponding no-CBs CPU's grace-period kthread ("rcuog*"), and that
structure's ->nocb_defer_wakeup field is used to track such deferral.
This means that the show_rcu_nocb_state() printing an error when those
fields are set for a CPU not corresponding to a no-CBs grace-period
kthread is erroneous.
This commit therefore switches the check from ->nocb_timer to
->nocb_bypass_timer and removes the check of ->nocb_defer_wakeup.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Originally, the call to rcu_preempt_blocked_readers_cgp() from
force_qs_rnp() had to be conditioned on CONFIG_PREEMPT_RCU=y, as in
commit a77da14ce9 ("rcu: Yet another fix for preemption and CPU
hotplug"). However, there is now a CONFIG_PREEMPT_RCU=n definition of
rcu_preempt_blocked_readers_cgp() that unconditionally returns zero, so
invoking it is now safe. In addition, the CONFIG_PREEMPT_RCU=n definition
of rcu_initiate_boost() simply releases the rcu_node structure's ->lock,
which is what happens when the "if" condition evaluates to false.
This commit therefore drops the IS_ENABLED(CONFIG_PREEMPT_RCU) check,
so that rcu_initiate_boost() is called only in CONFIG_PREEMPT_RCU=y
kernels when there are readers blocking the current grace period.
This does not change the behavior, but reduces code-reader confusion by
eliminating non-CONFIG_PREEMPT_RCU=y calls to rcu_initiate_boost().
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
On callback overload, it is necessary to quickly detect idle CPUs,
and rcu_gp_fqs_check_wake() checks for this condition. Unfortunately,
the code following the call to this function does not repeat this check,
which means that in reality no actual quiescent-state forcing, instead
only a couple of quick and pointless wakeups at the beginning of the
grace period.
This commit therefore adds a check for the RCU_GP_FLAG_OVLD flag in
the post-wakeup "if" statement in rcu_gp_fqs_loop().
Fixes: 1fca4d12f4 ("rcu: Expedite first two FQS scans under callback-overload conditions")
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
A message of the form "rcu: !!! lDTs ." can be tracked down, but
doing so is not trivial. This commit therefore eases this process by
adding text so that this error message now reads as follows:
"rcu: nocb GP activity on CB-only CPU!!! lDTs ."
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
During acceleration of CB, the rsp's gp_seq is rcu_seq_snap'd. This is
the value used for acceleration - it is the value of gp_seq at which it
is safe the execute all callbacks in the callback list.
The rdp's gp_seq is not very useful for this scenario. Make
rcu_grace_period report the gp_seq_req instead as it allows one to
reason about how the acceleration works.
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit moves the initialization of the CONFIG_PREEMPT=n version of
the rcu_exp_handler() function's rdp and rnp local variables into their
respective declarations to save a couple lines of code.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
KCSAN is now in mainline, so this commit removes the stubs for the
data_race(), ASSERT_EXCLUSIVE_WRITER(), and ASSERT_EXCLUSIVE_ACCESS()
macros.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
KCSAN is now in mainline, so this commit removes the stubs for the
data_race(), ASSERT_EXCLUSIVE_WRITER(), and ASSERT_EXCLUSIVE_ACCESS()
macros.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
KCSAN is now in mainline, so this commit removes the stubs for the
data_race(), ASSERT_EXCLUSIVE_WRITER(), and ASSERT_EXCLUSIVE_ACCESS()
macros.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Patch series "kasan: memorize and print call_rcu stack", v8.
This patchset improves KASAN reports by making them to have call_rcu()
call stack information. It is useful for programmers to solve
use-after-free or double-free memory issue.
The KASAN report was as follows(cleaned up slightly):
BUG: KASAN: use-after-free in kasan_rcu_reclaim+0x58/0x60
Freed by task 0:
kasan_save_stack+0x24/0x50
kasan_set_track+0x24/0x38
kasan_set_free_info+0x18/0x20
__kasan_slab_free+0x10c/0x170
kasan_slab_free+0x10/0x18
kfree+0x98/0x270
kasan_rcu_reclaim+0x1c/0x60
Last call_rcu():
kasan_save_stack+0x24/0x50
kasan_record_aux_stack+0xbc/0xd0
call_rcu+0x8c/0x580
kasan_rcu_uaf+0xf4/0xf8
Generic KASAN will record the last two call_rcu() call stacks and print up
to 2 call_rcu() call stacks in KASAN report. it is only suitable for
generic KASAN.
This feature considers the size of struct kasan_alloc_meta and
kasan_free_meta, we try to optimize the structure layout and size, lets it
get better memory consumption.
[1]https://bugzilla.kernel.org/show_bug.cgi?id=198437
[2]https://groups.google.com/forum/#!searchin/kasan-dev/better$20stack$20traces$20for$20rcu%7Csort:date/kasan-dev/KQsjT_88hDE/7rNUZprRBgAJ
This patch (of 4):
This feature will record the last two call_rcu() call stacks and prints up
to 2 call_rcu() call stacks in KASAN report.
When call_rcu() is called, we store the call_rcu() call stack into slub
alloc meta-data, so that the KASAN report can print rcu stack.
[1]https://bugzilla.kernel.org/show_bug.cgi?id=198437
[2]https://groups.google.com/forum/#!searchin/kasan-dev/better$20stack$20traces$20for$20rcu%7Csort:date/kasan-dev/KQsjT_88hDE/7rNUZprRBgAJ
[walter-zh.wu@mediatek.com: build fix]
Link: http://lkml.kernel.org/r/20200710162401.23816-1-walter-zh.wu@mediatek.com
Suggested-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Walter Wu <walter-zh.wu@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Andrey Konovalov <andreyknvl@google.com>
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthias Brugger <matthias.bgg@gmail.com>
Link: http://lkml.kernel.org/r/20200710162123.23713-1-walter-zh.wu@mediatek.com
Link: http://lkml.kernel.org/r/20200601050847.1096-1-walter-zh.wu@mediatek.com
Link: http://lkml.kernel.org/r/20200601050927.1153-1-walter-zh.wu@mediatek.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
static priority level knowledge from non-scheduler code.
The three APIs for non-scheduler code to set SCHED_FIFO are:
- sched_set_fifo()
- sched_set_fifo_low()
- sched_set_normal()
These are two FIFO priority levels: default (high), and a 'low' priority level,
plus sched_set_normal() to set the policy back to non-SCHED_FIFO.
Since the changes affect a lot of non-scheduler code, we kept this in a separate
tree.
When merging to the latest upstream tree there's a conflict in drivers/spi/spi.c,
which can be resolved via:
sched_set_fifo(ctlr->kworker_task);
Signed-off-by: Ingo Molnar <mingo@kernel.org>
-----BEGIN PGP SIGNATURE-----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=/7JH
-----END PGP SIGNATURE-----
Merge tag 'sched-fifo-2020-08-04' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull sched/fifo updates from Ingo Molnar:
"This adds the sched_set_fifo*() encapsulation APIs to remove static
priority level knowledge from non-scheduler code.
The three APIs for non-scheduler code to set SCHED_FIFO are:
- sched_set_fifo()
- sched_set_fifo_low()
- sched_set_normal()
These are two FIFO priority levels: default (high), and a 'low'
priority level, plus sched_set_normal() to set the policy back to
non-SCHED_FIFO.
Since the changes affect a lot of non-scheduler code, we kept this in
a separate tree"
* tag 'sched-fifo-2020-08-04' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (24 commits)
sched,tracing: Convert to sched_set_fifo()
sched: Remove sched_set_*() return value
sched: Remove sched_setscheduler*() EXPORTs
sched,psi: Convert to sched_set_fifo_low()
sched,rcutorture: Convert to sched_set_fifo_low()
sched,rcuperf: Convert to sched_set_fifo_low()
sched,locktorture: Convert to sched_set_fifo()
sched,irq: Convert to sched_set_fifo()
sched,watchdog: Convert to sched_set_fifo()
sched,serial: Convert to sched_set_fifo()
sched,powerclamp: Convert to sched_set_fifo()
sched,ion: Convert to sched_set_normal()
sched,powercap: Convert to sched_set_fifo*()
sched,spi: Convert to sched_set_fifo*()
sched,mmc: Convert to sched_set_fifo*()
sched,ivtv: Convert to sched_set_fifo*()
sched,drm/scheduler: Convert to sched_set_fifo*()
sched,msm: Convert to sched_set_fifo*()
sched,psci: Convert to sched_set_fifo*()
sched,drbd: Convert to sched_set_fifo*()
...
-----BEGIN PGP SIGNATURE-----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=p6ry
-----END PGP SIGNATURE-----
Merge tag 'core-urgent-2020-07-05' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull rcu fixlet from Thomas Gleixner:
"A single fix for a printk format warning in RCU"
* tag 'core-urgent-2020-07-05' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
rcuperf: Fix printk format warning
RCU is supposed to be watching all non-idle kernel code and also all
softirq handlers. This commit adds some teeth to this statement by
adding a WARN_ON_ONCE().
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Coccinelle reports a warning
WARNING: Assignment of 0/1 to bool variable
The root cause is that the variable lastphase is a bool, but is
initialised with integer 0. This commit therefore replaces the 0 with
a false.
Signed-off-by: Jules Irenge <jbi.octave@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, the rcu_torture_current variable remains non-NULL until after
all readers have stopped. During this time, rcu_torture_stats_print()
will think that the test is still ongoing, which can result in confusing
dmesg output. This commit therefore NULLs rcu_torture_current immediately
after the rcu_torture_writer() kthread has decided to stop, thus informing
rcu_torture_stats_print() much sooner.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Several variants of Linux-kernel RCU interact with task-exit processing,
including preemptible RCU, Tasks RCU, and Tasks Trace RCU. This commit
therefore adds testing of this interaction to rcutorture by adding
rcutorture.read_exit_burst and rcutorture.read_exit_delay kernel-boot
parameters. These kernel parameters control the frequency and spacing
of special read-then-exit kthreads that are spawned.
[ paulmck: Apply feedback from Dan Carpenter's static checker. ]
[ paulmck: Reduce latency to avoid false-positive shutdown hangs. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
SRCU disables interrupts to get a stable per-CPU pointer and then
acquires the spinlock which is in the per-CPU data structure. The
release uses spin_unlock_irqrestore(). While this is correct on a non-RT
kernel, this conflicts with the RT semantics because the spinlock is
converted to a 'sleeping' spinlock. Sleeping locks can obviously not be
acquired with interrupts disabled.
Acquire the per-CPU pointer `ssp->sda' without disabling preemption and
then acquire the spinlock_t of the per-CPU data structure. The lock will
ensure that the data is consistent.
The added call to check_init_srcu_struct() is now needed because a
statically defined srcu_struct may remain uninitialized until this
point and the newly introduced locking operation requires an initialized
spinlock_t.
This change was tested for four hours with 8*SRCU-N and 8*SRCU-P without
causing any warnings.
Cc: Lai Jiangshan <jiangshanlai@gmail.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: rcu@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit further avoids conflation of refperf with the kernel's perf
feature by renaming kernel/rcu/refperf.c to kernel/rcu/refscale.c,
and also by similarly renaming the functions and variables inside
this file. This has the side effect of changing the names of the
kernel boot parameters, so kernel-parameters.txt and ver_functions.sh
are also updated.
The rcutorture --torture type remains refperf, and this will be
addressed in a separate commit.
Reported-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The old Kconfig option name is all too easy to conflate with the
unrelated "perf" feature, so this commit renames RCU_REF_PERF_TEST to
RCU_REF_SCALE_TEST.
Reported-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The synchronize_rcu_tasks_trace() header comment incorrectly claims that
any number of things delimit RCU Tasks Trace read-side critical sections,
when in fact only rcu_read_lock_trace() and rcu_read_unlock_trace() do so.
This commit therefore fixes this comment, and, while in the area, fixes
a typo in the rcu_read_lock_trace() header comment.
Reported-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds testing for RCU Tasks readers to the refperf module.
This also applies to RCU Rude readers, as both flavors have empty
(as in non-existent) read-side markers.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The current units of microseconds are too coarse, so this commit
changes the units to nanoseconds. However, ndelay is used only for the
nanoseconds with udelay being used for whole microseconds. For example,
setting refperf.readdelay=1500 results in a udelay(1) followed by an
ndelay(500).
Suggested-by: Akira Yokosawa <akiyks@gmail.com>
[ paulmck: Abstracted delay per Akira feedback and move from 80 to 100 lines. ]
[ paulmck: Fix names as suggested by kbuild test robot. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
A 64-bit division was introduced in refperf, breaking compilation
on all 32-bit architectures:
kernel/rcu/refperf.o: in function `main_func':
refperf.c:(.text+0x57c): undefined reference to `__aeabi_uldivmod'
Fix this by using div_u64 to mark the expensive operation.
[ paulmck: Update primitive and format per Nathan Chancellor. ]
Fixes: bd5b16d6c88d ("refperf: Allow decimal nanoseconds")
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Valdis Klētnieks <valdis.kletnieks@vt.edu>
Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
With the various measurement optimizations, 10,000 loops normally
suffices. This commit therefore reduces the refperf.loops default value
from 10,000,000 to 10,000.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds a refperf.readdelay module parameter that controls the
duration of each critical section. This parameter allows gathering data
showing how the performance differences between the various primitives
vary with critical-section length.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit moves the reader-launch wait loop from ref_perf_init()
to main_func(), removing one layer of wakeup and allowing slightly
faster system boot.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The experiment-number column is currently labeled "Threads", which is
misleading at best. This commit therefore relabels it as "Runs", and
adjusts the scripts accordingly.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit causes all the readers to start running unmeasured load
until all readers have done at least one such run (thus having warmed
up), then run the measured load, and then run unmeasured load until all
readers have completed their measured load. This approach avoids any
thread running measured load while other readers are idle.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, readers are awakened individually. On most systems, this
results in significant wakeup delay from one reader to the next, which
can result in the first and last reader having sole access to the
synchronization primitive in question. If that synchronization primitive
involves shared memory, those readers will rack up a huge number of
operations in a very short time, causing large perturbations in the
results.
This commit therefore has the readers busy-wait after being awakened,
and uses a new n_started variable to synchronize their start times.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit converts the reader_task structure's "start" field to int
in order to demote a full barrier to an smp_load_acquire() and also to
simplify the code a bit. While in the area, and to enlist the compiler's
help in ensuring that nothing was missed, the field's name was changed
to start_reader.
Also while in the area, change the main_func() store to use
smp_store_release() to further fortify against wait/wake races.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit moves a printk() out of the measurement interval, converts
a atomic_dec()/atomic_read() pair to atomic_dec_and_test(), and adds
a smp_mb__before_atomic() to avoid potential wake/wait hangs. These
changes have the added benefit of reducing the number of loops required
for amortizing loop overhead for CONFIG_PREEMPT=n RCU measurements from
1,000,000 to 10,000. This reduction in turn shortens the test, reducing
the probability of interference.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Because the reset_readers() and process_durations() functions are used
only within kernel/rcu/refperf.c, this commit makes them static.
Reported-by: kbuild test robot <lkp@intel.com>
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, the buffer used to accumulate the thread-summary output is
fixed size, which will cause problems if someone decides to run on a large
number of PCUs. This commit therefore dynamically allocates this buffer.
[ paulmck: Fix memory allocation as suggested by KASAN. ]
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, the buffer used to accumulate the experiment-summary output
is fixed size, which will cause problems if someone decides to run
one hundred experiments. This commit therefore dynamically allocates
this buffer.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The current code uses the number of threads both to limit the number
of threads and to specify the number of experiments, but also varies
the number of threads as the experiments progress. This commit takes
a different approach by adding an refperf.nruns module parameter that
specifies the number of experiments, and furthermore uses the same
number of threads for each experiment.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit converts nreaders to a module parameter, with the default
of -1 specifying the old behavior of using 75% of the readers.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The CONFIG_PREEMPT=n rcu_read_lock()/rcu_read_unlock() pair's overhead,
even including loop overhead, is far less than one nanosecond.
Since logscale plots are not all that happy with zero values, provide
picoseconds as decimals.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Current runs show PREEMPT=n rcu_read_lock()/rcu_read_unlock() pairs
consuming between 20 and 30 nanoseconds, when in fact the actual value is
zero, give or take the barrier() asm's effect on compiler optimizations.
The additional overhead is caused by function calls through pointers
(especially in these days of Spectre mitigations) and perhaps also
needless argument passing, a non-const loop limit, and an upcounting loop.
This commit therefore combines the ->readlock() and ->readunlock()
function pointers into a single ->readsection() function pointer that
takes the loop count as a const parameter and keeps any data passed
from the read-lock to the read-unlock internal to this new function.
These changes reduce the measured overhead of the aforementioned
PREEMPT=n rcu_read_lock()/rcu_read_unlock() pairs from between 20 and
30 nanoseconds to somewhere south of 500 picoseconds.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds an rcuperf module parameter named "holdoff" that
defaults to 10 seconds if refperf is built in and to zero otherwise.
The assumption is that all the CPUs are online by the time that the
modprobe and insmod commands are going to do anything, and that normal
systems will have all the CPUs online within ten seconds.
Larger systems may take many tens of seconds or even minutes to get
to this point, hence this being a module parameter instead of being a
hard-coded constant.
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit adds comments explaining why the readers have otherwise insane
levels of measurement overhead, namely that they are intended as a test
load for update-side performance measurements, not as a straight-up
read-side performance test.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Add a test for comparing the performance of RCU with various read-side
synchronization mechanisms. The test has proved useful for collecting
data and performing these comparisons.
Currently RCU, SRCU, reader-writer lock, reader-writer semaphore and
reference counting can be measured using refperf.perf_type parameter.
Each invocation of the test runs measures performance of a specific
mechanism.
The maximum number of CPUs to concurrently run readers on is chosen by
the test itself and is 75% of the total number of CPUs. So if you had 24
CPUs, the test runs with a maximum of 18 parallel readers.
A number of experiments are conducted, and in each experiment, the
number of readers is increased by 1, upto the 75% of CPUs mark. During
each experiment, all readers execute an empty loop with refperf.loops
iterations and time the total loop duration. This is then averaged.
Example output:
Parameters "refperf.perf_type=srcu refperf.loops=2000000" looks like:
[ 3.347133] srcu-ref-perf:
[ 3.347133] Threads Time(ns)
[ 3.347133] 1 36
[ 3.347133] 2 34
[ 3.347133] 3 34
[ 3.347133] 4 34
[ 3.347133] 5 33
[ 3.347133] 6 33
[ 3.347133] 7 33
[ 3.347133] 8 33
[ 3.347133] 9 33
[ 3.347133] 10 33
[ 3.347133] 11 33
[ 3.347133] 12 33
[ 3.347133] 13 33
[ 3.347133] 14 33
[ 3.347133] 15 32
[ 3.347133] 16 33
[ 3.347133] 17 33
[ 3.347133] 18 34
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
wait_event() already retries if the condition for the wake up is not
satisifed after wake up. Remove them from the rcuperf test.
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit declares trc_n_readers_need_end and trc_wait static and
replaced a "&" with "&&". The "&" happened to work because the values
are bool, but accidents waiting to happen and all that...
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The show_rcu_tasks_gp_kthreads() function is not invoked by Tiny RCU,
but is nevertheless defined in Tiny RCU builds that enable Tasks Trace
RCU. This commit therefore conditionally compiles this function so
that it is defined only in builds that actually use it.
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Although this is in some strict sense unnecessary, it is good to allow
the compiler to compare the function declaration with its definition.
This commit therefore adds a #include of linux/rcupdate_trace.h to
kernel/rcu/update.c.
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The rcu_tasks_postscan() function is not used outside of RCU's tasks.h
file, so this commit makes it be static.
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
This commit converts the long-standing schedule_timeout_interruptible()
and schedule_timeout_uninterruptible() calls used by the various Tasks
RCU's grace-period kthreads to schedule_timeout_idle(). This conversion
avoids polluting the load-average with Tasks-RCU-related sleeping.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Update the kvfree_call_rcu() function with head-less support.
This allows RCU to reclaim objects without an embedded rcu_head.
tree-RCU:
We introduce two chains of arrays to store SLAB-backed and vmalloc
pointers, each. Storage in either of these arrays does not require
embedding an rcu_head within the object.
Maintaining the arrays may become impossible due to high memory
pressure. For such cases there is an emergency path. Objects with
rcu_head inside are just queued on a backup rcu_head list. Later on
that list is drained. As for the head-less variant, as the current
context can sleep, the following emergency measures are applied:
a) Synchronously wait until a grace period has elapsed.
b) Call kvfree().
tiny-RCU:
For double argument calls, there are no new changes in behavior. For
single argument call, kvfree() is directly inlined on the current
stack after a synchronize_rcu() call. Note that for tiny-RCU, any
call to synchronize_rcu() is actually a quiescent state, therefore
it does nothing.
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Co-developed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The following changes are introduced:
1. Rename rcu_invoke_kfree_callback() to rcu_invoke_kvfree_callback(),
as well as the associated trace events, so the rcu_kfree_callback(),
becomes rcu_kvfree_callback(). The reason is to be aligned with kvfree()
notation.
2. Rename __is_kfree_rcu_offset to __is_kvfree_rcu_offset. All RCU
paths use kvfree() now instead of kfree(), thus rename it.
3. Rename kfree_call_rcu() to the kvfree_call_rcu(). The reason is,
it is capable of freeing vmalloc() memory now. Do the same with
__kfree_rcu() macro, it becomes __kvfree_rcu(), the goal is the
same.
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Co-developed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Replace kfree() with kvfree() in rcu_reclaim_tiny().
This makes it possible to release either SLAB or vmalloc
objects after a GP.
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
To do so, we use an array of kvfree_rcu_bulk_data structures.
It consists of two elements:
- index number 0 corresponds to slab pointers.
- index number 1 corresponds to vmalloc pointers.
Keeping vmalloc pointers separated from slab pointers makes
it possible to invoke the right freeing API for the right
kind of pointer.
It also prepares us for future headless support for vmalloc
and SLAB objects. Such objects cannot be queued on a linked
list and are instead directly into an array.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Co-developed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
In order to reduce the dynamic need for pages in kfree_rcu(),
pre-allocate a configurable number of pages per CPU and link
them in a list. When kfree_rcu() reclaims objects, the object's
container page is cached into a list instead of being released
to the low-level page allocator.
Such an approach provides O(1) access to free pages while also
reducing the number of requests to the page allocator. It also
makes the kfree_rcu() code to have free pages available during
a low memory condition.
A read-only sysfs parameter (rcu_min_cached_objs) reflects the
minimum number of allowed cached pages per CPU.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The per-CPU variable is initialized at runtime in
kfree_rcu_batch_init(). This function is invoked before
'rcu_scheduler_active' is set to 'RCU_SCHEDULER_RUNNING'.
After the initialisation, '->initialized' is to true.
The raw_spin_lock is only acquired if '->initialized' is
set to true. The worqueue item is only used if 'rcu_scheduler_active'
set to RCU_SCHEDULER_RUNNING which happens after initialisation.
Use a static initializer for krc.lock and remove the runtime
initialisation of the lock. Since the lock can now be always
acquired, remove the '->initialized' check.
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Introduce helpers to lock and unlock per-cpu "kfree_rcu_cpu"
structures. That will make kfree_call_rcu() more readable
and prevent programming errors.
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
