Paul E. McKenney [Wed, 9 Mar 2011 02:51:54 +0000 (21:51 -0500)]
Provide cleanup interfaces for per-CPU and per-thread call_rcu threads
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Wed, 9 Mar 2011 02:50:26 +0000 (21:50 -0500)]
Maintain list of struct call_rcu_data to keep valgrind happy.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Wed, 9 Mar 2011 02:48:49 +0000 (21:48 -0500)]
Add call_rcu() interface
Adds call_rcu(), with RCU threads to invoke the callbacks. By default,
there will be one such RCU thread per process, created the first time
that call_rcu() is invoked. On systems supporting sched_getcpu(), it
is possible to create one RCU thread per CPU by calling
create_all_cpu_call_rcu_data().
This version includes feedback from Mathieu Desnoyers.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Wed, 9 Mar 2011 02:37:43 +0000 (21:37 -0500)]
Fix ppc asm syntax error.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 4 Mar 2011 17:33:20 +0000 (12:33 -0500)]
update version to 0.5.4
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 4 Mar 2011 17:25:08 +0000 (12:25 -0500)]
urcu-bp: Update fork() handling
Introduce
extern void rcu_bp_before_fork(void);
extern void rcu_bp_after_fork_parent(void);
extern void rcu_bp_after_fork_child(void);
to handle fork gracefully. These disable signals and hold the registry mutex
across forks.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 3 Mar 2011 21:08:21 +0000 (16:08 -0500)]
Version 0.5.3
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 3 Mar 2011 16:49:06 +0000 (11:49 -0500)]
Update README for urcu-bp fork handling
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 3 Mar 2011 15:05:48 +0000 (10:05 -0500)]
README: describe fork() interaction
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 3 Mar 2011 02:19:00 +0000 (21:19 -0500)]
urcu-bp: no "need_mb" in distrust signal config
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Jason Wessel [Mon, 28 Feb 2011 16:36:42 +0000 (11:36 -0500)]
Update the README to change armv7l to arm
ARM v5 and ARM v6 are known to work using compiler barriers.
[ edit: add a missing comma while we are there ]
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Mon, 28 Feb 2011 16:35:09 +0000 (11:35 -0500)]
ARM: use dmb for rmb and wmb in addition to mb
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Jason Wessel [Mon, 28 Feb 2011 16:34:10 +0000 (11:34 -0500)]
Add a test for the dmb instruction on ARM
The dmb instruction is used to implement a memory barrier on modern
ARM hardware. The older generations of processors which are included
in single processor systems can get by on the generic implementation
of the cmm_mb(). This will allow the user space rcu to work on many
other ARM systems.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Jason Wessel [Mon, 28 Feb 2011 16:33:42 +0000 (11:33 -0500)]
hange the arm headers to remove all the references to v7
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Jason Wessel [Mon, 28 Feb 2011 16:33:19 +0000 (11:33 -0500)]
Make armv7l headers generic for the arm architecture
This commit simply moves the files named *armv7l* to *arm*
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Mon, 21 Feb 2011 21:17:44 +0000 (16:17 -0500)]
URCU list: remove leftover namespace clash (list_t type removed)
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 31 Dec 2010 17:54:52 +0000 (12:54 -0500)]
urcu_defer: update test programs (build fix)
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 31 Dec 2010 17:53:34 +0000 (12:53 -0500)]
rcu_defer test: handle error value
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 31 Dec 2010 17:50:59 +0000 (12:50 -0500)]
urcu-defer: handle malloc error value
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 19 Nov 2010 03:20:31 +0000 (22:20 -0500)]
Version 0.5.2
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 19 Nov 2010 03:18:42 +0000 (22:18 -0500)]
Fix get_cycles -> caa_get_cycles
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 19 Nov 2010 02:51:16 +0000 (21:51 -0500)]
Version 0.5.1
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 19 Nov 2010 02:47:14 +0000 (21:47 -0500)]
ACCESS_ONCE should also have CMM_ prefix
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 19 Nov 2010 02:44:59 +0000 (21:44 -0500)]
LOAD_SHARED and STORE_SHARED should have CMM_ prefix
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 19 Nov 2010 02:38:17 +0000 (21:38 -0500)]
Fix CAA_CAA -> CAA
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 19 Nov 2010 02:35:41 +0000 (21:35 -0500)]
Fix missing BITS_PER_LONG -> CAA_BITS_PER_LONG rename
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 18 Nov 2010 20:56:47 +0000 (15:56 -0500)]
update version to 0.5.0
David Goulet [Thu, 18 Nov 2010 20:48:01 +0000 (15:48 -0500)]
Rename all data structure with prefix cds_
This is the third and last patch of the namespace refactoring.
The prefix cds_ stands for Concurrent Data Structure and is
use for queue, list, hlist and stack.
For RCU protected data strucutre, the prefix is also added but
the suffix _rcu is appended. This suffix indicate and RCU protection.
Signed-off-by: David Goulet <david.goulet@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
David Goulet [Thu, 18 Nov 2010 20:46:38 +0000 (15:46 -0500)]
Rename all arch primitives with prefix caa_
This is the second patch for reducing namespace pollution.
The caa_ prefix stands for Concurrent Architecture Abstraction.
Again, suggested by Mathieu Desnoyers and Paul E. Mckenney.
Every define, macro and function specific to the architecture
abstraction of liburcu is modified with that prefix
Signed-off-by: David Goulet <david.goulet@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
David Goulet [Thu, 18 Nov 2010 20:45:20 +0000 (15:45 -0500)]
Rename all memory primitives with prefix cmm_
In order to not pollute the userspace namespace for application
using liburcu or any lib/apps linked with urcu, this patch if the
first of three major refactor for naming convention.
The cmm_ prefix is a short name for Concurrent Memory Model and was
suggested by Mathieu Desnoyers and Paul E. Mckenney. Every memory
primitives such as mb, wmb, rmb, and so on are renamed.
Signed-off-by: David Goulet <david.goulet@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
David Goulet [Mon, 15 Nov 2010 19:12:00 +0000 (14:12 -0500)]
Add missing test binaries to .gitignore
Signed-off-by: David Goulet <david.goulet@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
David Goulet [Sat, 13 Nov 2010 14:30:04 +0000 (09:30 -0500)]
Add config.h and stamp-h2 to gitignore
Signed-off-by: David Goulet <david.goulet@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sat, 6 Nov 2010 19:09:29 +0000 (15:09 -0400)]
Add compiler barriers to read-side for mutex ordering
Ensure that mutex acquisition/release is not reordered wrt rcu read lock/unlock.
No impact in practice because pthreads lock/unlock primitives act as memory
barriers. But we should provide the same guarantees we are expecting from other
locking primitives, hence this modification.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 24 Sep 2010 20:48:19 +0000 (16:48 -0400)]
Fix missing urcu_ref.h in Makefile.am
The 0.4.7 tarball is missing 'urcu/urcu_ref.h', causing the following
error:
CC rculfqueue.lo
In file included from rculfqueue.c:26:
urcu/rculfqueue.h:26:22: error: urcu_ref.h: No such file or directory
In file included from rculfqueue.c:26:
urcu/rculfqueue.h:45: error: field ‘ref’ has incomplete type
In file included from rculfqueue.c:27:
urcu/rculfqueue-static.h:29:27: error: urcu/urcu_ref.h: No such file or directory
Reported-by: Jon Bernard <jbernard@debian.org>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 22 Sep 2010 20:08:47 +0000 (16:08 -0400)]
update version to 0.4.7
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 25 Aug 2010 15:16:34 +0000 (11:16 -0400)]
urcu lfstack: pop needs rcu read lock
The pop side needs to hold the RCU read lock to deal with ABA problem.
The push side does not need to hold the RCU read lock because even if the next
item is changed to the same value initially read, the stack will still be
consistent.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 25 Aug 2010 14:14:19 +0000 (10:14 -0400)]
lfstack: simplification
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 22 Aug 2010 14:21:19 +0000 (10:21 -0400)]
rculfstack: LGPL-ize
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 22 Aug 2010 13:21:58 +0000 (09:21 -0400)]
rculfqueue: LGPL-ize
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 22 Aug 2010 13:08:47 +0000 (09:08 -0400)]
add missing comment to wfstack.c
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 22 Aug 2010 12:52:13 +0000 (08:52 -0400)]
wfstack: LGPL-ize
Create wrapper symbols for LGPL linking.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 22 Aug 2010 12:44:41 +0000 (08:44 -0400)]
wfqueue: provide API for caller-locked dequeue
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 22 Aug 2010 12:34:03 +0000 (08:34 -0400)]
LGPL-ize wfqueue
Provide the appropriate wrappers to ensure that non-LGPL applications can use
the half-blocking wait-free queue.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 18 Aug 2010 14:55:55 +0000 (10:55 -0400)]
WF queue cleanup
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 18 Aug 2010 14:55:14 +0000 (10:55 -0400)]
Wait-free stack: rcu-less implementation
Don't have to hold a rcu read lock anymore. pop side is blocking and holds a
mutex.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 18 Aug 2010 14:18:58 +0000 (10:18 -0400)]
Wait-free queue: wait-free enqueue, blocking dequeue without RCU dependency
Implementation better suited for supporting call_rcu().
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 18 Aug 2010 13:16:08 +0000 (09:16 -0400)]
RCU lf queue/stack: update comments
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 22:21:33 +0000 (18:21 -0400)]
ifdef C, wait-free stack adaptative pop wait
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 22:21:10 +0000 (18:21 -0400)]
urcu static coding style fix (whitespace)
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 21:58:50 +0000 (17:58 -0400)]
Coding style fix
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 21:12:49 +0000 (17:12 -0400)]
RCU wait-free stack: add _blocking in pop() primitive name
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 21:12:13 +0000 (17:12 -0400)]
RCU lock-free queue comment updates
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 21:11:30 +0000 (17:11 -0400)]
RCU wait-free enqueue/blocking dequeue queue
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 20:31:47 +0000 (16:31 -0400)]
Add ifdef cond around headers
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 15:59:55 +0000 (11:59 -0400)]
RCU lock-free queue: don't hold RCU read lock across retry
Be nicer to grace periods.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 15:57:30 +0000 (11:57 -0400)]
RCU lock-free stack: don't hold read lock across retry
Be nicer to grace-periods.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 15:53:30 +0000 (11:53 -0400)]
Wait-free push/Blocking pop stack: Update following Paul's review
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 15:53:16 +0000 (11:53 -0400)]
Add lock-free queue comments from Paul's review
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 13 Jul 2010 00:54:52 +0000 (20:54 -0400)]
Add wait-free push/blocking pop stack
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Mon, 12 Jul 2010 16:29:11 +0000 (12:29 -0400)]
Lock-free stack: add comments
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Mon, 12 Jul 2010 16:16:30 +0000 (12:16 -0400)]
Add lock-free RCU queue and stack
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 18 Jun 2010 16:47:46 +0000 (12:47 -0400)]
Update version to 0.4.6
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 18 Jun 2010 16:47:24 +0000 (12:47 -0400)]
Update README for ARM support
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 18 Jun 2010 16:38:39 +0000 (12:38 -0400)]
Update version to 0.4.5
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 18 Jun 2010 16:35:26 +0000 (12:35 -0400)]
Update email address from @polymtl.ca to @efficios.com
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Fri, 18 Jun 2010 16:26:20 +0000 (12:26 -0400)]
Force build failure on unknown architectures
Create urcu/arch_unknown.h and urcu/uatomic_arch_unknown.h, which
contain only #error statements and explanatory comments. This forces
build failures on unrecognized architectures in preference to trying
to guess at what operations might be safe on such architectures.
One other semi-feasible alternative is to use hashed arrays of locks
that are acquired with signals disabled. However, this seems a bit
too ornate, especially for architectures for which the gcc __sync_
primitives work correctly.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Fri, 18 Jun 2010 16:25:08 +0000 (12:25 -0400)]
Add native ARM port for armv7l
Add native support for armv7l. Other variants of ARM will likely require
separate ports. The gcc __sync_ approach works back to at least 2.6.15
kernels.
As suggested by Paolo Bonzini, defer the definition of CACHE_LINE_SIZE
to arch_generic.h.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Fri, 18 Jun 2010 16:06:19 +0000 (12:06 -0400)]
Add ia64 architecture based on gcc primitives
This of course assumes that ia64 gcc correctly implements the various
__sync_ primitives. ;-)
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Fri, 18 Jun 2010 16:05:12 +0000 (12:05 -0400)]
Add header files supporting gcc __sync_ primitives
Add a urcu/arch_gcc.h and urcu/uatomic_arch_gcc.h whose primitives are
based on the gcc __sync_ primitives. This should be usable for all
systems that have correctly implemented __sync_ primitives, which sadly
does not include all combinations of systems and compilers. In addition,
specific systems may gain higher performance with hand-coded primitives.
Nevertheless, this is nice for getting a new architecture up and running
quickly.
As suggested by Paolo Bonzini, defer the definition of mb() to the common
arch_generic.h file, and also defer the uatomic*() primitives to
uatomic_generic.h.
[Mathieu : fixed arch_gcc.h header which contained incorrect arch_unknown.h in
header.]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 13 Jun 2010 23:55:41 +0000 (19:55 -0400)]
Add test cycles per loop
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 13 Jun 2010 22:44:04 +0000 (18:44 -0400)]
Add uatomic test to make check
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paul E. McKenney [Sun, 13 Jun 2010 21:30:04 +0000 (17:30 -0400)]
Remove unused sync_core() definitions
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 20 May 2010 22:12:11 +0000 (18:12 -0400)]
Update version to 0.4.4
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 20 May 2010 22:09:20 +0000 (18:09 -0400)]
Fix powerpc uatomic 4-byte cmpxchg (cmpd -> cmpw)
Necessary to make it work on 32-bit only powerpc architectures. Was using the
64-bit comparison rather than the 32-bit comparison for 4-byte cmpxchg.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Fri, 14 May 2010 00:15:10 +0000 (20:15 -0400)]
urcu-qsbr: use same ongoing helper for 32 and 64 bits
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Giuseppe Scrivano [Tue, 20 Apr 2010 13:08:16 +0000 (09:08 -0400)]
"make check" runs the test suite
a trivial patch to enable the Makefile "check" rule.
Signed-off-by: Giuseppe Scrivano <gscrivano@gnu.org>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 14 Apr 2010 16:56:53 +0000 (12:56 -0400)]
Simplify qsbr online condition
Alan Stern wrote:
> return v && (v - rcu_gp_ctr > ULONG_MAX / 2);
>
> Wouldn't it be better to write:
>
> return v != 0 && v != rcu_gp_ctr;
>
> or something equivalent? The point is, it's foolish to return 0 when
> rcu_gp_ctr has wrapped around to below v. This can happen only when
> rcu_thread_online() is delayed (as by preemption), in which case we
> know that there's a good chance the algorithm will fail. My version of
> the test eliminates the possibility of this failure, although it
> doesn't eliminate the possibility that rcu_gp_ctr has wrapped around so
> far that it is equal to v.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 14 Apr 2010 16:37:15 +0000 (12:37 -0400)]
Cleanup: head -> node field name
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 17 Mar 2010 13:31:26 +0000 (09:31 -0400)]
Update urcu and urcu-qsbr, make compiler barriers explicit
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 14 Mar 2010 19:33:51 +0000 (15:33 -0400)]
urcu: Add extra "engineering safety factor" memory barrier in update_counter_and_wait()
Even though the memory barriers within update_counter_and_wait() are proven not
to be needed in urcu-mb/signal/qsbr implementations, we leave them in place as
an engineering safety factor. Basically, we've proven they are not required
(formally for urcu-mb and urcu-signal by model checking, less formally for
urcu-qsbr by looking at the execution order of concurrent synchronize_rcu() and
RCU read-sides with out-of-order load/stores). However, given that on the
overall performance impact of synchronize_rcu(), these memory barriers do not
add a significant overhead, let's leave them in place with a comment stating
that they are not required.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 14 Mar 2010 14:27:23 +0000 (10:27 -0400)]
urcu-qsbr: document compiler barrier
This compiler barrier is, at the implementation level, already provided by the
volatile accesses for STORE/LOAD_SHARED. However, given that these primitives
does not provide this guarantee at the high-level (and might use lighter gcc
primitives eventually), we need to add an explicit barrier().
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Sun, 7 Mar 2010 02:07:18 +0000 (21:07 -0500)]
update version to 0.4.3
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Jon Bernard [Sun, 7 Mar 2010 02:06:36 +0000 (21:06 -0500)]
Fix typo in urcu/uatomic_generic.h
Signed-off-by: Jon Bernard <jbernard@debian.org>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 4 Mar 2010 19:57:57 +0000 (14:57 -0500)]
Update version to 0.4.2
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Thu, 4 Mar 2010 14:54:51 +0000 (09:54 -0500)]
Update README
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Wed, 3 Mar 2010 14:08:30 +0000 (09:08 -0500)]
urcu-bp: reset ctr to 0 upon garbage collection
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 2 Mar 2010 16:41:01 +0000 (11:41 -0500)]
compiler.h: add () in ACCESS_ONCE()
Protect x in ACCESS_ONCE macro.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 2 Mar 2010 16:36:49 +0000 (11:36 -0500)]
urcu-defer: remove "ratelimit" leftover
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 2 Mar 2010 16:18:30 +0000 (11:18 -0500)]
Update readme: ia64 and alpha support
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Tue, 2 Mar 2010 16:08:19 +0000 (11:08 -0500)]
Install missing arch_generic.h
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Mon, 1 Mar 2010 21:42:15 +0000 (16:42 -0500)]
urcu: fix multiple register/unregister assertion failure
Caused by the fact that we check for the whole gp count (which includes the last
grace period phase bit) rather than just checking if the nesting count mask is
zero.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Mathieu Desnoyers [Mon, 1 Mar 2010 21:41:51 +0000 (16:41 -0500)]
test cases: add multiple register/unregister tests
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 19:12:21 +0000 (14:12 -0500)]
centralize definition of BITS_PER_LONG
Use __SIZEOF_LONG__, defined by GCC 4.3 or later, or _LP64, defined
by GCC 3.4 or later.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 19:06:01 +0000 (14:06 -0500)]
avoid multiple evaluation of STORE_SHARED argument
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 19:04:11 +0000 (14:04 -0500)]
support compiling on unknown architectures
The new defaults mean that, for example, ia64 support comes for free.
[edit by Mathieu Desnoyers]
defaults -> generic
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 19:02:55 +0000 (14:02 -0500)]
add Alpha support
And you already know what needs to be special-cased for Alpha...
[edit by Mathieu Desnoyers]
defaults -> generic
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 19:00:22 +0000 (14:00 -0500)]
move whether atomic byte/short exists to uatomic_arch_*.h
And add more generic implementations to uatomic_defaults.h.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 18:58:45 +0000 (13:58 -0500)]
use uatomic_generic.h for common fallback implementations
And now, really remove the code for S390/SPARC, using the fallback
implementation of uatomic_generic.h
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 18:57:00 +0000 (13:57 -0500)]
add uatomic_generic.h, use it for common definitions
uatomic_generic.h can be included by uatomic_arch_*.h to provide useful
default definitions. uatomic_arch_*.h can define whatever builtins
it wants to override, then uatomic_generic.h will provide what is not
already defined, as follows:
- uatomic_cmpxchg will use __sync_val_compare_and_swap builtins;
- uatomic_add_return will use __sync_fetch_and_add if uatomic_arch_*.h
did not provide a definition of uatomic_cmpxchg. If it did, we assume
__sync builtins are buggy or otherwise undesirable on this platform,
so uatomic_generic.h will implement uatomic_add_return in terms of
uatomic_cmpxchg too.
- uatomic_xchg is like uatomic_add_return. However, since GCC does
not provide an adequate builtin, it needs to use a compare-and-swap
loop using __sync_bool_compare_and_swap if uatomic_cmpxchg is not
provided.
- uatomic_sub_return/uatomic_add/uatomic_sub will be implemented
in terms of uatomic_add_return;
- uatomic_inc/uatomic_dec will be implemented in terms of uatomic_add.
After this patch, uatomic_generic.h is already used for the latter two
categories. The next patch will use uatomic_generic.h whenever there is
no assembly code involved, or otherwise it makes no difference to use
uatomic_generic.h. We keep custom per-arch macros to provide support for
compilers back to early GCC 3.x versions; however future ports may
not have this limitation.
The hunk in tests/test_uatomic.c is only needed for bisectability
and will be removed later.
[edit by Mathieu Desnoyers]
Fixed Makefile.am conflict.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Paolo Bonzini [Mon, 1 Mar 2010 18:53:51 +0000 (13:53 -0500)]
remove compat_uatomic_cmpxchg #define from non-x86
This is unused everywhere except on x86.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
This page took 0.041641 seconds and 4 git commands to generate.