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>
Implement torture testing for the "sched" variant of RCU, which uses
preempt_disable, preempt_enable, and synchronize_sched.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
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>
Use the newly-generic synchronous deferred free function to implement torture
testing for rcu_bh using synchronize_rcu_bh rather than the asynchronous
call_rcu_bh.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
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>
Use the newly-generic synchronous deferred free function to implement torture
testing for RCU using synchronize_rcu rather than the asynchronous call_rcu.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
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>
Make srcu_torture_deferred_free use cur_ops->sync() so it will work for any
implementation. Move and rename it in preparation for use in the ops of other
implementations.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
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>
rcutorture currently has one writer and an arbitrary number of readers. To
better exercise some of the code paths in RCU implementations, add fake
writer threads which call the synchronize function for the RCU variant in a
loop, with a delay between calls to arrange for different numbers of
writers running in parallel.
[bunk@stusta.de: cleanup]
Acked-by: Paul McKenney <paulmck@us.ibm.com>
Cc: Dipkanar Sarma <dipankar@in.ibm.com>
Signed-off-by: Josh Triplett <josh@freedesktop.org>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
rcu_random uses a counter rrs_count to occasionally mix data from
get_random_bytes into the state of its pseudorandom generator. However,
the rrs_counter gets declared as an unsigned long, and rcu_random checks
for --rrs_count < 0, so this code will never mix any real random data into
the state, and will thus always return the same sequence of random numbers.
Also, change the return value of rcu_random from long to unsigned long, to
avoid potential issues caused by the use of the % operator, which can
return negative values for negative left operands.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
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_torture_init kmallocs the array of reader threads, then creates each
one with kthread_run, cleaning up with rcu_torture_cleanup if this fails.
rcu_torture_cleanup calls kthread_stop on any non-NULL pointer in the
array; however, any readers after the one that failed to start up will have
invalid pointers, not null pointers. Avoid this by using kzalloc instead.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
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>
The comment for rcutorture's torture_type parameter only lists the RCU
variants rcu and srcu, but not rcu_bh; add rcu_bh to the list.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
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>
Currently the init_srcu_struct() routine has no way to report out-of-memory
errors. This patch (as761) makes it return -ENOMEM when the per-cpu data
allocation fails.
The patch also makes srcu_init_notifier_head() report a BUG if a notifier
head can't be initialized. Perhaps it should return -ENOMEM instead, but
in the most likely cases where this might occur I don't think any recovery
is possible. Notifier chains generally are not created dynamically.
[akpm@osdl.org: avoid statement-with-side-effect in macro]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
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>
This patch (as751) adds a new type of notifier chain, based on the SRCU
(Sleepable Read-Copy Update) primitives recently added to the kernel. An
SRCU notifier chain is much like a blocking notifier chain, in that it must
be called in process context and its callout routines are allowed to sleep.
The difference is that the chain's links are protected by the SRCU
mechanism rather than by an rw-semaphore, so calling the chain has
extremely low overhead: no memory barriers and no cache-line bouncing. On
the other hand, unregistering from the chain is expensive and the chain
head requires special runtime initialization (plus cleanup if it is to be
deallocated).
SRCU notifiers are appropriate for notifiers that will be called very
frequently and for which unregistration occurs very seldom. The proposed
"task notifier" scheme qualifies, as may some of the network notifiers.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Acked-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Adds SRCU operations to rcutorture and updates rcutorture documentation.
Also increases the stress imposed by the rcutorture test.
[bunk@stusta.de: make needlessly global code static]
Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Cc: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Updated patch adding a variant of RCU that permits sleeping in read-side
critical sections. SRCU is as follows:
o Each use of SRCU creates its own srcu_struct, and each
srcu_struct has its own set of grace periods. This is
critical, as it prevents one subsystem with a blocking
reader from holding up SRCU grace periods for other
subsystems.
o The SRCU primitives (srcu_read_lock(), srcu_read_unlock(),
and synchronize_srcu()) all take a pointer to a srcu_struct.
o The SRCU primitives must be called from process context.
o srcu_read_lock() returns an int that must be passed to
the matching srcu_read_unlock(). Realtime RCU avoids the
need for this by storing the state in the task struct,
but SRCU needs to allow a given code path to pass through
multiple SRCU domains -- storing state in the task struct
would therefore require either arbitrary space in the
task struct or arbitrary limits on SRCU nesting. So I
kicked the state-storage problem up to the caller.
Of course, it is not permitted to call synchronize_srcu()
while in an SRCU read-side critical section.
o There is no call_srcu(). It would not be hard to implement
one, but it seems like too easy a way to OOM the system.
(Hey, we have enough trouble with call_rcu(), which does
-not- permit readers to sleep!!!) So, if you want it,
please tell me why...
[josht@us.ibm.com: sparse notation]
Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Josh Triplett <josh@freedesktop.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Currently msi.c is doing sanity checks that make certain before an irq is
destroyed it has no more users.
By adding irq_has_action I can perform the test is a generic way, instead of
relying on a msi specific data structure.
By performing the core check in dynamic_irq_cleanup I ensure every user of
dynamic irqs has a test present and we don't free resources that are in use.
In msi.c this allows me to kill the attrib.state member of msi_desc and all of
the assciated code to maintain it.
To keep from freeing data structures when irq cleanup code is called to soon
changing dyanamic_irq_cleanup is insufficient because there are msi specific
data structures that are also not safe to free.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Greg KH <greg@kroah.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
With the msi support comes a new concept in irq handling, irqs that are
created dynamically at run time.
Currently the msi code allocates irqs backwards. First it allocates a
platform dependent routing value for an interrupt the ``vector'' and then it
figures out from the vector which irq you are on.
This msi backwards allocator suffers from two basic problems. The allocator
suffers because it is trying to do something that is architecture specific in
a generic way making it brittle, inflexible, and tied to tightly to the
architecture implementation. The alloctor also suffers from it's very
backwards nature as it has tied things together that should have no
dependencies.
To solve the basic dynamic irq allocation problem two new architecture
specific functions are added: create_irq and destroy_irq.
create_irq takes no input and returns an unused irq number, that won't be
reused until it is returned to the free poll with destroy_irq. The irq then
can be used for any purpose although the only initial consumer is the msi
code.
destroy_irq takes an irq number allocated with create_irq and returns it to
the free pool.
Making this functionality per architecture increases the simplicity of the irq
allocation code and increases it's flexibility.
dynamic_irq_init() and dynamic_irq_cleanup() are added to automate the
irq_desc initializtion that should happen for dynamic irqs.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rajesh Shah <rajesh.shah@intel.com>
Cc: Andi Kleen <ak@muc.de>
Cc: "Protasevich, Natalie" <Natalie.Protasevich@UNISYS.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Currently move_native_irq disables and renables the irq we are migrating to
ensure we don't take that irq when we are actually doing the migration
operation. Disabling the irq needs to happen but sometimes doing the work is
move_native_irq is too late.
On x86 with ioapics the irq move sequences needs to be:
edge_triggered:
mask irq.
move irq.
unmask irq.
ack irq.
level_triggered:
mask irq.
ack irq.
move irq.
unmask irq.
We can easily perform the edge triggered sequence, with the current defintion
of move_native_irq. However the level triggered case does not map well. For
that I have added move_masked_irq, to allow me to disable the irqs around both
the ack and the move.
Q: Why have we not seen this problem earlier?
A: The only symptom I have been able to reproduce is that if we change
the vector before acknowleding an irq the wrong irq is acknowledged.
Since we currently are not reprogramming the irq vector during
migration no problems show up.
We have to mask the irq before we acknowledge the irq or else we could
hit a window where an irq is asserted just before we acknowledge it.
Edge triggered irqs do not have this problem because acknowledgements
do not propogate in the same way.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rajesh Shah <rajesh.shah@intel.com>
Cc: Andi Kleen <ak@muc.de>
Cc: "Protasevich, Natalie" <Natalie.Protasevich@UNISYS.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The primary aim of this patchset is to remove maintenances problems caused by
the irq infrastructure. The two big issues I address are an artificially
small cap on the number of irqs, and that MSI assumes vector == irq. My
primary focus is on x86_64 but I have touched other architectures where
necessary to keep them from breaking.
- To increase the number of irqs I modify the code to look at the (cpu,
vector) pair instead of just looking at the vector.
With a large number of irqs available systems with a large irq count no
longer need to compress their irq numbers to fit. Removing a lot of brittle
special cases.
For acpi guys the result is that irq == gsi.
- Addressing the fact that MSI assumes irq == vector takes a few more
patches. But suffice it to say when I am done none of the generic irq code
even knows what a vector is.
In quick testing on a large Unisys x86_64 machine we stumbled over at least
one driver that assumed that NR_IRQS could always fit into an 8 bit number.
This driver is clearly buggy today. But this has become a class of bugs that
it is now much easier to hit.
This patch:
This is a minor space optimization. In practice I don't think this has any
affect because of our alignment constraints and the other fields but there is
not point in chewing up an uncessary word and since we already read the flag
field this should improve the cache hit ratio of the irq handler.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rajesh Shah <rajesh.shah@intel.com>
Cc: Andi Kleen <ak@muc.de>
Cc: "Protasevich, Natalie" <Natalie.Protasevich@UNISYS.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The comment for the nreaders parameter of rcutorture gives the default as
4*ncpus, but the value actually defaults to 2*ncpus; fix the comment.
Signed-off-by: Josh Triplett <josh@freedesktop.org>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Many files include the filename at the beginning, serveral used a wrong one.
Signed-off-by: Uwe Zeisberger <Uwe_Zeisberger@digi.com>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
When the per cpu sched domains are build then they also need to be placed
on the node where the cpu resides otherwise we will have frequent off node
accesses which will slow down the system.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Up to now sched group's cpu_power for each sched domain is initialized
independently. This made the setup code ugly as the new sched domains are
getting added.
Make the sched group cpu_power setup code generic, by using domain child
field and new domain flag in sched_domain. For most of the sched
domains(except NUMA), sched group's cpu_power is now computed generically
using the domain properties of itself and of the child domain.
sched groups in NUMA domains are setup little differently and hence they
don't use this generic mechanism.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Introduce the child field in sched_domain struct and use it in
sched_balance_self().
We will also use this field in cleaning up the sched group cpu_power
setup(done in a different patch) code.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
