On some workloads, (for example when lot of close() syscalls are done), RCU
qlen can be quite large, and RCU heads are no longer in cpu cache when
rcu_do_batch() is called.
This patch adds a prefetch() in rcu_do_batch() to give CPU a hint to bring
back cache lines containing 'struct rcu_head's.
Most list manipulations macros include prefetch(), but not open coded ones
(at least with current C compilers :) )
I got a nice speedup on a trivial benchmark (3.48 us per iteration instead
of 3.95 us on a 1.6 GHz Pentium-M)
while (1) { pipe(p); close(fd[0]); close(fd[1]);}
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Kill a hard-to-calculate 'rsinterval' boot parameter and per-cpu
rcu_data.last_rs_qlen. Instead, it adds adds a flag rcu_ctrlblk.signaled,
which records the fact that one of CPUs has sent a resched IPI since the
last rcu_start_batch().
Roughly speaking, we need two rcu_start_batch()s in order to move callbacks
from ->nxtlist to ->donelist. This means that when ->qlen exceeds qhimark
and continues to grow, we should send a resched IPI, and then do it again
after we gone through a quiescent state.
On the other hand, if it was already sent, we don't need to do it again
when another CPU detects overflow of the queue.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
rcu_do_batch() decrements rdp->qlen with irqs enabled. This is not good,
it can also be modified by call_rcu() from interrupt.
Decrement ->qlen once with irqs disabled, after a main loop.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Few of the callback functions and notifier blocks that are associated with cpu
notifications incorrectly have __devinit and __devinitdata. They should be
__cpuinit and __cpuinitdata instead.
It makes no functional difference but wastes text area when CONFIG_HOTPLUG is
enabled and CONFIG_HOTPLUG_CPU is not.
This patch fixes all those instances.
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Cc: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Locking init improvement:
- introduce and use __SPIN_LOCK_UNLOCKED for array initializations,
to pass in the name string of locks, used by debugging
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
In 2.6.17, there was a problem with cpu_notifiers and XFS. I provided a
band-aid solution to solve that problem. In the process, i undid all the
changes you both were making to ensure that these notifiers were available
only at init time (unless CONFIG_HOTPLUG_CPU is defined).
We deferred the real fix to 2.6.18. Here is a set of patches that fixes the
XFS problem cleanly and makes the cpu notifiers available only at init time
(unless CONFIG_HOTPLUG_CPU is defined).
If CONFIG_HOTPLUG_CPU is defined then cpu notifiers are available at run
time.
This patch reverts the notifier_call changes made in 2.6.17
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Cc: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add operations for the call_rcu_bh() variant of RCU. Also add an
rcu_batches_completed_bh() function, which is needed by rcutorture.
Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Remove synchronize_kernel() (deprecated 2-APR-2005 in
http://lkml.org/lkml/2005/4/3/11) and makes the RCU API inaccessible to
non-GPL Linux kernel modules (as was announced more than one year ago in
http://lkml.org/lkml/2005/4/3/8). Tested on x86 and ppc64.
Signed-off-by: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
With "Paul E. McKenney" <paulmck@us.ibm.com>
Introduce rcu_needs_cpu() interface. This can be used to tell if there
will be a new rcu batch on a cpu soon by looking at the curlist pointer.
This can be used to avoid to enter a tickless idle state where the cpu
would miss that a new batch is ready when rcu_start_batch would be called
on a different cpu.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Few of the notifier_chain_register() callers use __init in the definition
of notifier_call. It is incorrect as the function definition should be
available after the initializations (they do not unregister them during
initializations).
This patch fixes all such usages to _not_ have the notifier_call __init
section.
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Few of the notifier_chain_register() callers use __devinitdata in the
definition of notifier_block data structure. It is incorrect as the
data structure should be available after the initializations (they do
not unregister them during initializations).
This was leading to an oops when notifier_chain_register() call is
invoked for those callback chains after initialization.
This patch fixes all such usages to _not_ have the notifier_block data
structure in the init data section.
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
__rcu_process_callbacks() disables interrupts to protect itself from
call_rcu() which adds new entries to ->nxtlist.
However we can check "->nxtlist != NULL" with interrupts enabled, we can't
get "false positives" because call_rcu() can only change this condition
from 0 to 1.
Tested with rcutorture.ko.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Dipankar Sarma <dipankar@in.ibm.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch makes two needlessly global structs static.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
As the RCU symbols are going to be changed to GPL in the near future,
lets warn users that this is going to happen.
Cc: Paul McKenney <paulmck@us.ibm.com>
Acked-by: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch adds new tunables for RCU queue and finished batches. There are
two types of controls - number of completed RCU updates invoked in a batch
(blimit) and monitoring for high rate of incoming RCUs on a cpu (qhimark,
qlowmark).
By default, the per-cpu batch limit is set to a small value. If the input
RCU rate exceeds the high watermark, we do two things - force quiescent
state on all cpus and set the batch limit of the CPU to INTMAX. Setting
batch limit to INTMAX forces all finished RCUs to be processed in one shot.
If we have more than INTMAX RCUs queued up, then we have bigger problems
anyway. Once the incoming queued RCUs fall below the low watermark, the
batch limit is set to the default.
Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Pointed out by Srivatsa Vaddagiri <vatsa@in.ibm.com>.
rcu_do_batch() stops after processing maxbatch callbacks
on ->donelist leaving rcu_tasklet in TASKLET_STATE_SCHED
state.
If CPU_DEAD event happens remaining ->donelist entries are
lost, rcu_offline_cpu() kills this tasklet.
With this patch ->donelist migrates along with ->curlist
and ->nxtlist to the current cpu.
Compile tested.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch moves rcu_state into the rcu_ctrlblk. I think there
are no reasons why we should have 2 different variables to control
rcu state. Every user of rcu_state has also "rcu_ctrlblk *rcp" in
the parameter list.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
I think it is better to set ->next_pending in the caller, when
it is needed. This saves one parameter, and this coincides with
cpu_quiet() beahaviour, which sets ->completed = ->cur itself.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
__rcu_pending() is rather fat and called twice from rcu_pending().
rcu_pending() has multiple callers, and not that small too.
This patch uninlines both of them.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Use atomic_inc_not_zero for rcu files instead of special case rcuref.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
RCU tasklist_lock and RCU signal handling: send signals RCU-read-locked
instead of tasklist_lock read-locked. This is a scalability improvement on
SMP and a preemption-latency improvement under PREEMPT_RCU.
Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Acked-by: William Irwin <wli@holomorphy.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
____cacheline_maxaligned_in_smp is currently used to align critical structures
and avoid false sharing. It uses per-arch L1_CACHE_SHIFT_MAX and people find
L1_CACHE_SHIFT_MAX useless.
However, we have been using ____cacheline_maxaligned_in_smp to align
structures on the internode cacheline size. As per Andi's suggestion,
following patch kills ____cacheline_maxaligned_in_smp and introduces
INTERNODE_CACHE_SHIFT, which defaults to L1_CACHE_SHIFT for all arches.
Arches needing L3/Internode cacheline alignment can define
INTERNODE_CACHE_SHIFT in the arch asm/cache.h. Patch replaces
____cacheline_maxaligned_in_smp with ____cacheline_internodealigned_in_smp
With this patch, L1_CACHE_SHIFT_MAX can be killed
Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Accessing nohz_cpu_mask before incrementing rcp->cur is racy. It can cause
tickless idle CPUs to be included in rsp->cpumask, which will extend
graceperiods unnecessarily.
Fix this race. It has been tested using extensions to RCU torture module
that forces various CPUs to become idle.
Signed-off-by: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This introduces a new interface - rcu_barrier() which waits until all
the RCUs queued until this call have been completed.
Reiser4 needs this, because we do more than just freeing memory object
in our RCU callback: we also remove it from the list hanging off
super-block. This means, that before freeing reiser4-specific portion
of super-block (during umount) we have to wait until all pending RCU
callbacks are executed.
The only change of reiser4 made to the original patch, is exporting of
rcu_barrier().
Cc: Hans Reiser <reiser@namesys.com>
Cc: Vladimir V. Saveliev <vs@namesys.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch is a rewrite of the one submitted on October 1st, using modules
(http://marc.theaimsgroup.com/?l=linux-kernel&m=112819093522998&w=2).
This rewrite adds a tristate CONFIG_RCU_TORTURE_TEST, which enables an
intense torture test of the RCU infratructure. This is needed due to the
continued changes to the RCU infrastructure to accommodate dynamic ticks,
CPU hotplug, realtime, and so on. Most of the code is in a separate file
that is compiled only if the CONFIG variable is set. Documentation on how
to run the test and interpret the output is also included.
This code has been tested on i386 and ppc64, and an earlier version of the
code has received extensive testing on a number of architectures as part of
the PREEMPT_RT patchset.
Signed-off-by: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This makes call_rcu() keep track of how many events there are on the RCU
list, and cause a reschedule event when the list gets too long.
This helps keep RCU event lists down.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Dipankar made RCU limit the batch size to improve latency, but that
approach is unworkable: it can cause the RCU queues to grow without
bounds, since the batch limiter ended up limiting the callbacks.
So make the limit much higher, and start planning on instead limiting
the batch size by doing RCU callbacks more often if the queue looks like
it might be growing too long.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Adds a set of primitives to do reference counting for objects that are looked
up without locks using RCU.
Signed-off-by: Ravikiran Thirumalai <kiran_th@gmail.com>
Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The synchronize_kernel() primitive is used for quite a few different purposes:
waiting for RCU readers, waiting for NMIs, waiting for interrupts, and so on.
This makes RCU code harder to read, since synchronize_kernel() might or might
not have matching rcu_read_lock()s. This patch creates a new
synchronize_rcu() that is to be used for RCU readers and a new
synchronize_sched() that is used for the rest. These two new primitives
currently have the same implementation, but this is might well change with
additional real-time support. Both new primitives are GPL-only, the old
primitive is deprecated.
Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The gpl exports need to be put back. Moving them to GPL -- but in a
measured manner, as I proposed on this list some months ago -- is fine.
Changing these particular exports precipitously is most definitely -not-
fine. Here is my earlier proposal:
http://marc.theaimsgroup.com/?l=linux-kernel&m=110520930301813&w=2
See below for a patch that puts the exports back, along with an updated
version of my earlier patch that starts the process of moving them to GPL.
I will also be following this message with RFC patches that introduce two
(EXPORT_SYMBOL_GPL) interfaces to replace synchronize_kernel(), which then
becomes deprecated.
Signed-off-by: <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Eric Dumazet