| Age | Commit message (Collapse) | Author |
|---|
|
The rcu_kick_nohz_cpu() function is no longer used, and the functionality
it used to provide is now provided by a call to resched_cpu() in the
force-quiescent-state function rcu_implicit_dynticks_qs(). This commit
therefore removes rcu_kick_nohz_cpu().
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
CPUs are expected to report quiescent states when coming online and
when going offline, and grace-period initialization is supposed to
handle any race conditions where a CPU's ->qsmask bit is set just after
it goes offline. This commit adds diagnostics for the case where an
offline CPU nevertheless has a grace period waiting on it.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Grace-period initialization first processes any recent CPU-hotplug
operations, and then initializes state for the new grace period. These
two phases of initialization are currently not distinguished in debug
prints, but the distinction is valuable in a number of debug situations.
This commit therefore introduces two new values for ->gp_state,
RCU_GP_ONOFF and RCU_GP_INIT, in order to make this distinction.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Interactions between CPU-hotplug operations and grace-period
initialization can result in dump_blkd_tasks(). One of the first
debugging actions in this case is to search back in dmesg to work
out which of the affected rcu_node structure's CPUs are online and to
determine the last CPU-hotplug operation affecting any of those CPUs.
This can be laborious and error-prone, especially when console output
is lost.
This commit therefore causes dump_blkd_tasks() to dump the state of
the affected rcu_node structure's CPUs and the last grace period during
which the last offline and online operation affected each of these CPUs.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Now that quiescent states for newly offlined CPUs are reported either
when that CPU goes offline or at the end of grace-period initialization,
the CPU-hotplug failsafe in the force-quiescent-state code path is no
longer needed.
This commit therefore removes this failsafe.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Without special fail-safe quiescent-state-propagation checks, grace-period
hangs can result from the following scenario:
1. CPU 1 goes offline.
2. Because CPU 1 is the only CPU in the system blocking the current
grace period, the grace period ends as soon as
rcu_cleanup_dying_idle_cpu()'s call to rcu_report_qs_rnp()
returns.
3. At this point, the leaf rcu_node structure's ->lock is no longer
held: rcu_report_qs_rnp() has released it, as it must in order
to awaken the RCU grace-period kthread.
4. At this point, that same leaf rcu_node structure's ->qsmaskinitnext
field still records CPU 1 as being online. This is absolutely
necessary because the scheduler uses RCU (in this case on the
wake-up path while awakening RCU's grace-period kthread), and
->qsmaskinitnext contains RCU's idea as to which CPUs are online.
Therefore, invoking rcu_report_qs_rnp() after clearing CPU 1's
bit from ->qsmaskinitnext would result in a lockdep-RCU splat
due to RCU being used from an offline CPU.
5. RCU's grace-period kthread awakens, sees that the old grace period
has completed and that a new one is needed. It therefore starts
a new grace period, but because CPU 1's leaf rcu_node structure's
->qsmaskinitnext field still shows CPU 1 as being online, this new
grace period is initialized to wait for a quiescent state from the
now-offline CPU 1.
6. Without the fail-safe force-quiescent-state checks, there would
be no quiescent state from the now-offline CPU 1, which would
eventually result in RCU CPU stall warnings and memory exhaustion.
It would be good to get rid of the special fail-safe quiescent-state
propagation checks, and thus it would be good to fix things so that
the above scenario cannot happen. This commit therefore adds a new
->ofl_lock to the rcu_state structure. This lock is held by rcu_gp_init()
across the applying of buffered online and offline operations to the
rcu_node tree, and it is also held by rcu_cleanup_dying_idle_cpu()
when buffering a new offline operation. This prevents rcu_gp_init()
from acquiring the leaf rcu_node structure's lock during the interval
between when rcu_cleanup_dying_idle_cpu() invokes rcu_report_qs_rnp(),
which releases ->lock and the re-acquisition of that same lock.
This in turn prevents the failure scenario outlined above, and will
hopefully eventually allow removal of the offline-CPU checks from the
force-quiescent-state code path.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Now that everything has been converted to use ->gp_seq instead of
->gpnum and ->completed, this commit removes ->gpnum and ->completed.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
One problem with the ->need_future_gp[] array is that the grace-period
assignment of each element changes as the grace periods complete.
This means that it is necessary to hold a lock when checking this
array to learn if a given grace period has already been requested.
This increase lock contention, which is the opposite of helpful.
This commit therefore replaces the ->need_future_gp[] with a single
->gp_seq_needed value and keeps it updated in the rcu_data structure.
This will enable reliable lockless checking of whether or not a given
grace period has already been requested.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This commit converts the grace-period request code paths from ->completed
and ->gpnum to ->gp_seq. The need_future_gp_element() macro encapsulates
the shift operation required to use ->gp_seq as an index to the
->need_future_gp[] array. The rcu_cbs_completed() function is removed
in favor of the rcu_seq_snap() function. The rcu_start_this_gp()
gets some temporary consistency checks and uses rcu_seq_done(),
rcu_seq_current(), rcu_seq_state(), and rcu_gp_in_progress() in place
of the earlier open-coded comparisons of ->gpnum and ->completed.
The rcu_future_gp_cleanup() function replaces use of ->completed
with ->gp_seq. The rcu_accelerate_cbs() function replaces a call to
rcu_cbs_completed() with one to rcu_seq_snap(). The rcu_advance_cbs()
function replaces an access to >completed with one to ->gp_seq and adds
some temporary warnings. The rcu_nocb_wait_gp() function replaces a
call to rcu_cbs_completed() with one to rcu_seq_snap() and an open-coded
comparison with rcu_seq_done().
The temporary warnings will be removed when the various ->gpnum and
->completed fields are removed. Their purpose is to locate code who
might still be using ->gpnum and ->completed. (Much easier that way
than trying to trace down the causes of too-short grace periods and
grace-period hangs!)
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This commit switches the interrupt-disabled detection mechanism to
->gp_seq. This mechanism is used as part of RCU CPU stall warnings,
and detects cases where the stall is due to a CPU having interrupts
disabled.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This commit adds grace-period sequence numbers (->gp_seq) to the
rcu_state, rcu_node, and rcu_data structures, and updates them.
It also checks for consistency between rsp->gpnum and rsp->gp_seq.
These ->gp_seq counters will eventually replace the existing ->gpnum
and ->completed counters, allowing a single memory access to determine
whether or not a grace period is in progress and if so, which one.
This in turn will enable changes that will reduce ->lock contention on
the leaf rcu_node structures.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This commit causes a splat if RCU is idle and a request for a new grace
period is ignored for more than one second. This splat normally indicates
that some code path asked for a new grace period, but failed to wake up
the RCU grace-period kthread.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
[ paulmck: Fix bug located by Dan Carpenter and his static checker. ]
[ paulmck: Fix self-deadlock bug located 0day test robot. ]
[ paulmck: Disable unless CONFIG_PROVE_RCU=y. ]
|
|
The WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp()) in
rcu_gp_cleanup() triggers (inexplicably, of course) every so often.
This commit therefore extracts more information.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
and 'torture.2018.05.15a' into HEAD
exp.2018.05.15a: Parallelize expedited grace-period initialization.
fixes.2018.05.15a: Miscellaneous fixes.
lock.2018.05.15a: Decrease lock contention on root rcu_node structure,
which is a step towards merging RCU flavors.
torture.2018.05.15a: Torture-test updates.
|
|
There is no longer any need for ->need_future_gp[] to count the number of
requests for future grace periods, so this commit converts the additions
to assignments to "true" and reduces the size of each element to one byte.
While we are in the area, fix an obsolete comment.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
Currently, the rcu_future_needs_gp() function checks only the current
element of the ->need_future_gps[] array, which might miss elements that
were offset from the expected element, for example, due to races with
the start or the end of a grace period. This commit therefore makes
rcu_future_needs_gp() use the need_any_future_gp() macro to check all
of the elements of this array.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
The rcu_cbs_completed() function provides the value of ->completed
at which new callbacks can safely be invoked. This is recorded in
two-element ->need_future_gp[] arrays in the rcu_node structure, and
the elements of these arrays corresponding to the just-completed grace
period are zeroed at the end of that grace period. However, the
rcu_cbs_completed() function can return the current ->completed value
plus either one or two, so it is possible for the corresponding
->need_future_gp[] entry to be cleared just after it was set, thus
losing a request for a future grace period.
This commit avoids this race by expanding ->need_future_gp[] to four
elements.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
Currently, rcu_gp_cleanup() scans the rcu_node tree in order to reset
state to reflect the end of the grace period. It also checks to see
whether a new grace period is needed, but in a number of cases, rather
than directly cause the new grace period to be immediately started, it
instead leaves the grace-period-needed state where various fail-safes
can find it. This works fine, but results in higher contention on the
root rcu_node structure's ->lock, which is undesirable, and contention
on that lock has recently become noticeable.
This commit therefore makes rcu_gp_cleanup() immediately start a new
grace period if there is any need for one.
It is quite possible that it will later be necessary to throttle the
grace-period rate, but that can be dealt with when and if.
Reported-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
Accessors for the ->need_future_gp[] array are currently open-coded,
which makes them difficult to change. To improve maintainability, this
commit adds need_future_gp_mask() to compute the indexing mask from the
array size, need_future_gp_element() to access the element corresponding
to the specified grace-period number, and need_any_future_gp() to
determine if any future grace period is needed. This commit also applies
need_future_gp_element() to existing open-coded single-element accesses.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
Because rcu_eqs_special_set() is declared only in internal header
kernel/rcu/tree.h and stubbed in include/linux/rcutiny.h, it is
inaccessible outside of the RCU implementation. This patch therefore
moves the rcu_eqs_special_set() declaration to include/linux/rcutree.h,
which allows it to be used in non-rcu kernel code.
Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
Commit ae91aa0adb14 ("rcu: Remove debugfs tracing") removed the
RCU debugfs tracing code, but did not remove the no-longer used
->exp_workdone{0,1,2,3} fields in the srcu_data structure. This commit
therefore removes these fields along with the code that uselessly
updates them.
Signed-off-by: Byungchul Park <byungchul.park@lge.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
The rcu_preempt_do_callbacks() function was introduced in commit
09223371dea(rcu: Use softirq to address performance regression), where it
was necessary to handle kernel builds both containing and not containing
RCU-preempt. Since then, various changes (most notably f8b7fc6b51
("rcu: use softirq instead of kthreads except when RCU_BOOST=y")) have
resulted in this function being invoked only from rcu_kthread_do_work(),
which is present only in kernels containing RCU-preempt, which in turn
means that the rcu_preempt_do_callbacks() function is no longer needed.
This commit therefore inlines rcu_preempt_do_callbacks() into its
sole remaining caller and also removes the rcu_state_p and rcu_data_p
indirection for added clarity.
Signed-off-by: Byungchul Park <byungchul.park@lge.com>
Reviewed-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
[ paulmck: Remove the rcu_state_p and rcu_data_p indirection. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
The latency of RCU expedited grace periods grows with increasing numbers
of CPUs, eventually failing to be all that expedited. Much of the growth
in latency is in the initialization phase, so this commit uses workqueues
to carry out this initialization concurrently on a rcu_node-by-rcu_node
basis.
This change makes use of a new rcu_par_gp_wq because flushing a work
item from another work item running from the same workqueue can result
in deadlock.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Nicholas Piggin <npiggin@gmail.com>
|
|
RCU's nxttail has been optimized to be a rcu_segcblist, which is
a multi-tailed linked list with macros defined for the indexes for
each tail. The indexes have been defined in linux/rcu_segcblist.h,
so this commit removes the redundant definitions in kernel/rcu/tree.h.
Signed-off-by: Liu Changcheng <changcheng.liu@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The debugfs interface displayed statistics on RCU-pending checks but
this interface has since been removed. This commit therefore removes the
no-longer-used rcu_state structure's ->n_force_qs_lh and ->n_force_qs_ngp
fields along with their updates. (Though the ->n_force_qs_ngp field
was actually not used at all, embarrassingly enough.)
If this information proves necessary in the future, the corresponding
event traces will be added.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The debugfs interface displayed statistics on RCU-pending checks
but this interface has since been removed. This commit therefore
removes the no-longer-used rcu_data structure's ->n_rcu_pending,
->n_rp_core_needs_qs, ->n_rp_report_qs, ->n_rp_cb_ready,
->n_rp_cpu_needs_gp, ->n_rp_gp_completed, ->n_rp_gp_started,
->n_rp_nocb_defer_wakeup, and ->n_rp_need_nothing fields along with
their updates.
If this information proves necessary in the future, the corresponding
event traces will be added.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The debugfs interface displayed statistics on RCU callback invocation but
this interface has since been removed. This commit therefore removes the
no-longer-used rcu_data structure's ->n_cbs_invoked and ->n_nocbs_invoked
fields along with their updates.
If this information proves necessary in the future, the corresponding
event traces will be added.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The debugfs interface displayed statistics on RCU priority boosting,
but this interface has since been removed. This commit therefore
removes the no-longer-used rcu_data structure's ->n_tasks_boosted,
->n_exp_boosts, and ->n_exp_boosts and their updates.
If this information proves necessary in the future, the corresponding
event traces will be added.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Because the ->dynticks_nesting field now only contains the process-based
nesting level instead of a value encoding both the process nesting level
and the irq "nesting" level, we no longer need a long long, even on
32-bit systems. This commit therefore changes both the ->dynticks_nesting
and ->dynticks_nmi_nesting fields to long.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Now that ->dynticks_nesting counts only process-level dyntick-idle
entry and exit, there is no need for the elaborate segmented counter
with its guard fields and overflow checking. This commit therefore
makes ->dynticks_nesting be a simple counter.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
One common question upon seeing an RCU CPU stall warning is "did
the stalled CPUs have interrupts disabled?" However, the current
stall warnings are silent on this point. This commit therefore
uses irq_work to check whether stalled CPUs still respond to IPIs,
and flags this state in the RCU CPU stall warning console messages.
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
'hotplug.2017.07.25b', 'misc.2017.08.17a', 'spin_unlock_wait_no.2017.08.17a', 'srcu.2017.07.27c' and 'torture.2017.07.24c' into HEAD
doc.2017.08.17a: Documentation updates.
fixes.2017.08.17a: RCU fixes.
hotplug.2017.07.25b: CPU-hotplug updates.
misc.2017.08.17a: Miscellaneous fixes outside of RCU (give or take conflicts).
spin_unlock_wait_no.2017.08.17a: Remove spin_unlock_wait().
srcu.2017.07.27c: SRCU updates.
torture.2017.07.24c: Torture-test updates.
|
|
Given that the rcu_state structure's >orphan_pend and ->orphan_done
fields are used only during migration of callbacks from the recently
offlined CPU to a surviving CPU, if rcu_send_cbs_to_orphanage() and
rcu_adopt_orphan_cbs() are combined, these fields can become local
variables in the combined function. This commit therefore combines
rcu_send_cbs_to_orphanage() and rcu_adopt_orphan_cbs() into a new
rcu_segcblist_merge() function and removes the ->orphan_pend and
->orphan_done fields.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The ->orphan_lock is acquired and released only within the
rcu_migrate_callbacks() function, which now acquires the root rcu_node
structure's ->lock. This commit therefore eliminates the ->orphan_lock
in favor of the root rcu_node structure's ->lock.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
RCU's CPU-hotplug callback-migration code first moves the outgoing
CPU's callbacks to ->orphan_done and ->orphan_pend, and only then
moves them to the NOCB callback list. This commit avoids the
extra step (and simplifies the code) by moving the callbacks directly
from the outgoing CPU's callback list to the NOCB callback list.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The rcu_node structure's ->n_cbs_orphaned and ->n_cbs_adopted fields
are updated, but never read. This commit therefore removes them.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The handling of RCU's no-CBs CPUs has a maintenance headache, namely
that if call_rcu() is invoked with interrupts disabled, the rcuo kthread
wakeup must be defered to a point where we can be sure that scheduler
locks are not held. Of course, there are a lot of code paths leading
from an interrupts-disabled invocation of call_rcu(), and missing any
one of these can result in excessive callback-invocation latency, and
potentially even system hangs.
This commit therefore uses a timer to guarantee that the wakeup will
eventually occur. If one of the deferred-wakeup points kicks in, then
the timer is simply cancelled.
This commit also fixes up an incomplete removal of commits that were
intended to plug remaining exit paths, which should have the added
benefit of reducing the overhead of RCU's context-switch hooks. In
addition, it simplifies leader-to-follower callback-list handoff by
introducing locking. The call_rcu()-to-leader handoff continues to
use atomic operations in order to maintain good real-time latency for
common-case use of call_rcu().
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
[ paulmck: Dan Carpenter fix for mod_timer() usage bug found by smatch. ]
|
|
RCU's debugfs tracing used to be the only reasonable low-level debug
information available, but ftrace and event tracing has since surpassed
the RCU debugfs level of usefulness. This commit therefore removes
RCU's debugfs tracing.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The NO_HZ_FULL_SYSIDLE full-system-idle capability was added in 2013
by commit 0edd1b1784cb ("nohz_full: Add full-system-idle state machine"),
but has not been used. This commit therefore removes it.
If it turns out to be needed later, this commit can always be reverted.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
This commit moves the now-generic rnp->lock wrapper macros from
kernel/rcu/tree.h to kernel/rcu/rcu.h, thus allowing SRCU to use them.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Use of smp_mb__after_unlock_lock() would allow SRCU to omit a full
memory barrier during callback execution, so this commit converts
raw_spin_lock_rcu_node() from inline functions to type-generic macros
to allow them to handle locks in srcu_node structures as well as
rcu_node structures.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
The RCU_NOGP_WAKE_NOT, RCU_NOGP_WAKE, and RCU_NOGP_WAKE_FORCE flags
are used to mediate wakeups for the no-CBs CPU kthreads. The "NOGP"
really doesn't make any sense, so this commit does s/NOGP/NOCB/.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Although preemptible RCU allows its read-side critical sections to be
preempted, general blocking is forbidden. The reason for this is that
excessive preemption times can be handled by CONFIG_RCU_BOOST=y, but a
voluntarily blocked task doesn't care how high you boost its priority.
Because preemptible RCU is a global mechanism, one ill-behaved reader
hurts everyone. Hence the prohibition against general blocking in
RCU-preempt read-side critical sections. Preemption yes, blocking no.
This commit enforces this prohibition.
There is a special exception for the -rt patchset (which they kindly
volunteered to implement): It is OK to block (as opposed to merely being
preempted) within an RCU-preempt read-side critical section, but only if
the blocking is subject to priority inheritance. This exception permits
CONFIG_RCU_BOOST=y to get -rt RCU readers out of trouble.
Why doesn't this exception also apply to mainline's rt_mutex? Because
of the possibility that someone does general blocking while holding
an rt_mutex. Yes, the priority boosting will affect the rt_mutex,
but it won't help with the task doing general blocking while holding
that rt_mutex.
Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Linus noticed that the <linux/rcu_segcblist.h> has huge inline functions
which should not be inline at all.
As a first step in cleaning this up, move them all to kernel/rcu/ and
only keep an absolute minimum of data type defines in the header:
before: -rw-r--r-- 1 mingo mingo 22284 May 2 10:25 include/linux/rcu_segcblist.h
after: -rw-r--r-- 1 mingo mingo 3180 May 2 10:22 include/linux/rcu_segcblist.h
More can be done, such as uninlining the large functions, which inlining
is unjustified even if it's an RCU internal matter.
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2],
however, there are workloads that could result in a high volume of
concurrent invocations of call_srcu(), which with current SRCU would
result in excessive lock contention on the srcu_struct structure's
->queue_lock, which protects SRCU's callback lists. This commit therefore
moves SRCU to per-CPU callback lists, thus greatly reducing contention.
Because a given SRCU instance no longer has a single centralized callback
list, starting grace periods and invoking callbacks are both more complex
than in the single-list Classic SRCU implementation. Starting grace
periods and handling callbacks are now handled using an srcu_node tree
that is in some ways similar to the rcu_node trees used by RCU-bh,
RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is
controlled by exactly the same Kconfig options and boot parameters that
control the shape of the rcu_node tree).
In addition, the old per-CPU srcu_array structure is now named srcu_data
and contains an rcu_segcblist structure named ->srcu_cblist for its
callbacks (and a spinlock to protect this). The srcu_struct gets
an srcu_gp_seq that is used to associate callback segments with the
corresponding completion-time grace-period number. These completion-time
grace-period numbers are propagated up the srcu_node tree so that the
grace-period workqueue handler can determine whether additional grace
periods are needed on the one hand and where to look for callbacks that
are ready to be invoked.
The srcu_barrier() function must now wait on all instances of the per-CPU
->srcu_cblist. Because each ->srcu_cblist is protected by ->lock,
srcu_barrier() can remotely add the needed callbacks. In theory,
it could also remotely start grace periods, but in practice doing so
is complex and racy. And interestingly enough, it is never necessary
for srcu_barrier() to start a grace period because srcu_barrier() only
enqueues a callback when a callback is already present--and it turns out
that a grace period has to have already been started for this pre-existing
callback. Furthermore, it is only the callback that srcu_barrier()
needs to wait on, not any particular grace period. Therefore, a new
rcu_segcblist_entrain() function enqueues the srcu_barrier() function's
callback into the same segment occupied by the last pre-existing callback
in the list. The special case where all the pre-existing callbacks are
on a different list (because they are in the process of being invoked)
is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL
segment, relying on the done-callbacks check that takes place after all
callbacks are inovked.
Note that the readers use the same algorithm as before. Note that there
is a separate srcu_idx that tells the readers what counter to increment.
This unfortunately cannot be combined with srcu_gp_seq because they
need to be incremented at different times.
This commit introduces some ugly #ifdefs in rcutorture. These will go
away when I feel good enough about Tree SRCU to ditch Classic SRCU.
Some crude performance comparisons, courtesy of a quickly hacked rcuperf
asynchronous-grace-period capability:
Callback Queuing Overhead
-------------------------
# CPUS Classic SRCU Tree SRCU
------ ------------ ---------
2 0.349 us 0.342 us
16 31.66 us 0.4 us
41 --------- 0.417 us
The times are the 90th percentiles, a statistic that was chosen to reject
the overheads of the occasional srcu_barrier() call needed to avoid OOMing
the test machine. The rcuperf test hangs when running Classic SRCU at 41
CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code
and that statistics, this is a convincing demonstration of Tree SRCU's
performance and scalability advantages.
[1] https://lwn.net/Articles/309030/
[2] https://patchwork.kernel.org/patch/5108281/
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
[ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
|
|
This commit moves rcu_for_each_node_breadth_first(),
rcu_for_each_nonleaf_node_breadth_first(), and
rcu_for_each_leaf_node() from kernel/rcu/tree.h to
kernel/rcu/rcu.h so that SRCU can access them.
This commit is code-movement only.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This commit moves the C preprocessor code that defines the default shape
of the rcu_node combining tree to a new include/linux/rcu_node_tree.h
file as a first step towards enabling SRCU to create its own combining
tree, which in turn enables SRCU to implement per-CPU callback handling,
thus avoiding contention on the lock currently guarding the single list
of callbacks. Note that users of SRCU still need to know the size of
the srcu_struct structure, hence include/linux rather than kernel/rcu.
This commit is code-movement only.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
This commit switches SRCU from custom-built callback queues to the new
rcu_segcblist structure. This change associates grace-period sequence
numbers with groups of callbacks, which will be needed for efficient
processing of per-CPU callbacks.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
RCU has only one multi-tail callback list, which is implemented via
the nxtlist, nxttail, nxtcompleted, qlen_lazy, and qlen fields in the
rcu_data structure, and whose operations are open-code throughout the
Tree RCU implementation. This has been more or less OK in the past,
but upcoming callback-list optimizations in SRCU could really use
a multi-tail callback list there as well.
This commit therefore abstracts the multi-tail callback list handling
into a new kernel/rcu/rcu_segcblist.h file, and uses this new API.
The simple head-and-tail pointer callback list is also abstracted and
applied everywhere except for the NOCB callback-offload lists. (Yes,
the plan is to apply them there as well, but this commit is already
bigger than would be good.)
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|
|
If the RCU_EXPERT Kconfig option is not set (the default), then the
RCU_FANOUT_LEAF Kconfig option will not be defined, which will cause
the leaf-level rcu_node tree fanout to default to 32 on 32-bit systems
and 64 on 64-bit systems. This can result in excessive lock contention.
This commit therefore changes the computation of the leaf-level rcu_node
tree fanout so that the result will be 16 unless an explicit Kconfig or
kernel-boot setting says otherwise.
Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
|