Merge branch 'master' into lfqueue-dev

urcu-qsbr: avoid useless futex wakeups and burning CPU for long grace periods

I noticed that urcu makes exactly _one_ attempt at using futexes
to avoid busy looping on synchronize_rcu.  The attached patch instead
switches from busy waiting to futexes after RCU_QS_ACTIVE_ATTEMPTS.
To limit the amount of system calls, reading threads remember whether
they already had a quiescent state in this grace period; if so they were
already removed from the list, and can avoid signaling the futex.

Performance measured with rcutorture (nreaders: 10, nupdaters: 1,
duration: 10, median of nine runs):

     RCU_QS_ACTIVE_ATTEMPTS == 100, no patch         n_updates = 292
     RCU_QS_ACTIVE_ATTEMPTS == 1, no patch           n_updates = 290
     RCU_QS_ACTIVE_ATTEMPTS == 100, with patch       n_updates = 408
     RCU_QS_ACTIVE_ATTEMPTS == 1, with patch         n_updates = 404

(the first two cases are obviously the same; the only change is
when the futex is used, but over many calls there is no difference).

This patch matches the update to the Promela model.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

api: reimplement BUILD_BUG_ON in compiler.h

Even though it's just two lines of code, I'm reimplementing it
cleanly out of paranoia.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

test api cleanup: remove unused primitives

[ Mathieu: the rationale for this is that we can always add back that
  code if every needed. Removing leftover GPLv2 test code is an
  incentive to create the appropriate library-wide LGPL/MIT-style
  abstractions.]

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Merge branch 'master' into lfqueue-dev

put thread offline while waiting for the init flag

Otherwise, the call_rcu thread might end up in a synchronize_rcu that
never ends.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu: move private definitions to .c file

The reader does not and should not know about RCU_QS_ACTIVE_ATTEMPTS
and KICK_READER_LOOPS.  Move the definition to urcu.c.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-bp: move private definitions to .c file

The reader does not and should not know about RCU_QS_ACTIVE_ATTEMPTS.
Move the definition to urcu-bp.c.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-qsbr: move private definitions to .c file

The reader does not and should not know about RCU_QS_ACTIVE_ATTEMPTS.
Move the definition to urcu-qsbr.c.  Also remove KICK_READER_LOOPS.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rcutorture: fix rcutorture-qsbr

rcutorture-qsbr is not marking quiescent states, so it will always
have just one update in the whole test.  Fix it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

wfqueue: fix type-incorrect assignment

The "old_tail = q->tail, q->tail = node" assignment in wfqueue
is not type safe; q->tail is a pointer to pointer to node and the
correct value to assign is &node->next.  While the arithmetic is
the same, it is better to be tidy.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix tests Makefile EXTRA_DIST to use api.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix choice of default flavour in urcu/map/urcu.h

Hi, I noticed in the 0.64 release (and git too) that if a flavour is not
specified explicitly then RCU_MB is chosen in urcu/map/urcu.h, while the
docs say and the Makefile expects RCU_MEMBARRIER.  Note that the header
file urcu/static/urcu.h has similar logic but uses RCU_MEMBARRIER for
the default.

Before this patch:

$ nm *.a | grep rcu_init
00000000000003c0 T rcu_init_mb
00000000000003c0 T rcu_init_mb
0000000000000000 T rcu_init_sig

After this patch:

$ nm *.a | grep rcu_init
00000000000003c0 T rcu_init_memb
00000000000003c0 T rcu_init_mb
0000000000000000 T rcu_init_sig

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

api: remove list/hlist

Remove the parts that actually caused compilation to fail.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

api: remove arch-specific files

No change since they were a subset of api_gcc.h.  api_gcc.h is
renamed to api.h, and configure need not create a symlink anymore.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

api: make api_gcc.h a superset of the other headers

After this patch the only differences are:

- for_each_tid is now defined in api_x86.h

- HAVE_CPU_SET_T and HAVE_SCHED_SETAFFINITY are now defined in api_ppc.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

tests api: remove atomics

The code is #if 0'd already, remove it.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

put thread offline while waiting for the init flag

Otherwise, the call_rcu thread might end up in a synchronize_rcu that
never ends.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu: move private definitions to .c file

The reader does not and should not know about RCU_QS_ACTIVE_ATTEMPTS
and KICK_READER_LOOPS.  Move the definition to urcu.c.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-bp: move private definitions to .c file

The reader does not and should not know about RCU_QS_ACTIVE_ATTEMPTS.
Move the definition to urcu-bp.c.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-qsbr: move private definitions to .c file

The reader does not and should not know about RCU_QS_ACTIVE_ATTEMPTS.
Move the definition to urcu-qsbr.c.  Also remove KICK_READER_LOOPS.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rcutorture: fix rcutorture-qsbr

rcutorture-qsbr is not marking quiescent states, so it will always
have just one update in the whole test.  Fix it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

wfqueue: fix type-incorrect assignment

The "old_tail = q->tail, q->tail = node" assignment in wfqueue
is not type safe; q->tail is a pointer to pointer to node and the
correct value to assign is &node->next.  While the arithmetic is
the same, it is better to be tidy.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Merge branch 'master' into lfqueue-dev

Use caa_ prefix for min() and max()

Defining min/max macros clashes with C++.

Reported-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Version 0.6.4, update changelog

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Merge branch 'master' into lfqueue-dev

Fix ARM uatomic: missing return in uatomic and/or

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rculfqueue: provide locklessness by allowing multiple dummy nodes

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rcu lfqueue: make dequeue lockless by helping out other dequeuers

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rculfqueue: Document dummy pointer value access

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rculfqueue: Keep a reference to the current dummy node rather than using low bit

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rculfqueue: only one dummy node is needed

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rculfstack: update comment

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

RCU lf queue: Add assert for memory allocation

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rculfstack: header implementation should be defined as static inline

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

wfstack: header implementation should be defined as static inline

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

RCU lfqueue: Now works without reference counting (API change)

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

rculfstack: header implementation should be defined as static inline

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

wfstack: header implementation should be defined as static inline

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu tests: hold mutex across use of custom allocator

A thread preempted for a long period of time could race, when scheduled
again, with another thread that would have been allocating/freeing
entries (thus wrapping-around the available buffer), which would trigger
this race only when overcommitting the number of threads compared to the
number of available CPUs.

Taking the mutex across alloc and free to fix this.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-bp: mremap wrapper fix

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-bp: add mremap wrapper for non-linux systems

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-bp: don't copy old region upon mremap

mremap keeps the same virtual pages for the old/new mappings. So
explicitly copying from the old mapping is not needed, and probably
buggy, since the old mapping might have been unmapped.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

api.h: fix type warning

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

non-linux system membarrier fallback update

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

pthread_self should be cast to unsigned long

Used as part of the random seed along with time().

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add MAP_ANONYMOUS mapping to MAP_ANON for BSD build

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

tests: only include syscall.h on linux systems

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-bp, urcu-qsbr: remove unneeded syscall.h include

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu-defer: remove unneeded syscall.h include

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

call_rcu: per_cpu_call_rcu_data should be non-const

On FreeBSD:

In file included from urcu.c:438:
urcu-call-rcu-impl.h: In function 'get_cpu_call_rcu_data_mb':
urcu-call-rcu-impl.h:325: warning: return discards qualifiers from pointer target type

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update README info about gcc compability

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix urcu-call-rcu-impl.h missing call_rcu_wait for non-linux systems

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

compat_futex_async: fix missing return

compat_futex.c: In function `compat_futex_async':
compat_futex.c:114: warning: control reaches end of non-void function

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Check for __linux__ before including syscall.h

Reported-by: Jean-François Brousseau <jfb@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove unused local variables from compat_futex_async

Reported-by: Jean-François Brousseau <jfb@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove unneeded syscall.h include from urcu-call-rcu-impl.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add ChangeLog file to Makefile.am

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add changelog, update version to 0.6.3

uatomic: fix warning about unused variable

Fix:

compat_arch_x86.c: In function '_compat_uatomic_set':
compat_arch_x86.c:104:16: warning: variable 'result' set but not used [-Wunused-but-set-variable]

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

uatomic: fix i386 support

Fix:

* Incorrect prototype for uatomic_and and uatomic_or in i386
  compatibility code.
* Missing $(COMPAT) code inclusion in wfq/lfq tests.
* Silence gcc warnings about compat code (branch volountarily causing a
  linker error, which can never return).

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update version to 0.6.2

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

uatomic generic: map uatomic_and/or to correct symbols

Leading underscore was missing.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix generic atomic ops and/or: add missing return

Fixes build on powerpc:

./urcu/uatomic/generic.h: In function '_uatomic_and':
./urcu/uatomic/generic.h:310:2: warning: 'return' with a value, in
function returning void
./urcu/uatomic/generic.h: In function '_uatomic_or':
./urcu/uatomic/generic.h:374:2: warning: 'return' with a value, in
function returning void

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix test warning

Patch from:
https://build.opensuse.org/package/view_file?file=liburcu0-cleanup.patch&package=liburcu0&project=devel%3Atools%3Alttng&srcmd5=fa3ab9fae0401b4c0f73a9cf474eea16

fixes a warning in the liburcu tests.

Reported-by: Yannick Brosseau <yannick.brosseau@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add missing headers into tarball, version 0.6.1

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update version to 0.6.0

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Futex: turn "int" into "int32_t" for portability

Even though int is 32-bit on all architectures supported by liburcu so
far, make it future-proof by uint a int32_t, which enforces the same
type width used by the system call in the kernel.

Using int32_t and not uint32_t to make comparison with 0 more
straightforward.

Reported-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add back uatomic_arch.h for slow deprecation process

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add back urcu-futex.h and urcu_ref.h for slow deprecation

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add ldconfig step to README

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add a pkg-config file for the new liburcu-cds

Signed-off-by: Yannick Brosseau <yannick.brosseau@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Consolidate _LIBADD and _DEPENDENCIES in Makefile.am

Signed-off-by: Alexandre Montplaisir <alexandre.montplaisir@polymtl.ca>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add urcu/cds.h placeholder for all CDS headers

Using CDS headers becomes as simple as:

 #include <urcu/cds.h>

and link with

 -lurcu-cds

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Move the version-info to the AM_LDFLAGS variable

Since all the lib builded in this directory use the same version
we can put this command in the global variable

Signed-off-by: Yannick Brosseau <yannick.brosseau@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Installed headers should only include other headers with < >

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Rename urcu/urcu-futex.h to urcu/futex.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Rename urcu/urcu_ref.h to urcu/ref.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Headers: move uatomic_*.h to urcu/uatomic/*.h, rename uatomic_arch.h to uatomic.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Headers: move arch_*.h to urcu/arch/*.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Headers: move *-static.h headers to urcu/static/

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Install urcu/map/*.h into system

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Headers: move *-map.h headers to urcu/map/

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add library version information

Following the guidelines from libtool
(http://www.gnu.org/software/libtool/manual/html_node/Updating-version-info.htm)
this patch add version information to the distributed libraries.
For the next release, the version will be 1:0:0.
It will need to be updated before each release.

Signed-off-by: Yannick Brosseau <yannick.brosseau@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

lfqueue: update comments, cleanup

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

lfqueue: don't use defer_rcu anymore

lfqueue now uses call_rcu instead.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Combine stack and queues info liburcu-cds.so

Pretty much all libraries use a single .so for their data types. Do the
same here.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

cds containers: lfqueue and lfstack: don't depend on a particular rcu flavor

Remove rcu_read lock/unlock from the code, require the caller to ensure
protection. First move towards a single .so for all data containers.

This changes the lfqueue API, which is not very much used yet AFAIK.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add missing pkgconfig files to Makefile.am

Signed-off-by: Yannick Brosseau <yannick.brosseau@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

arm: remove useless declarations

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

call_rcu: add back delay in list non-empty case

commit bc94ca9bada25f7403e3e859caa241146ae8e338 changed the !RT behavior
slightly: when the list is not empty, it does not wait for a delay
anymore. Add this delay back, to ensure we don't flood the system with
frequent synchronize_rcu() calls, which would slow down readers.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

call_rcu: handle retry without wait correctly

The wait scheme has an implementation problem: if the list is not empty
when the !RT scheme checks for it, it will restart the loop and
decrement the futex (again) without calling call_rcu_wait() (which would
wait until it is set back to 0). So in this case, we can end up
decrementing "futex" to values well below -1.

Fix this by moving the decrement before the loop, and duplicate it after
return from call_rcu_wait() + poll() delay. Also move the "set futex to
0 upon stopping" outside of the loop: this is the only way the loop can
be stopped anyway.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu: fix dependencies on libwfqueue.la

Without this patch, "make -jN" usually fails for N as low as 4.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

urcu_ref: do error checking on urcu_ref_put

Underflow is what you really want to trap.  When you reach urcu_ref_get,
it is already too late.

[Edit: whitespaces -> tab cleanup]

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

use generic-size macros for common implementation of atomic ops

The definition of _uatomic_cmpxchg is different in x86 and other
architectures.  For x86 it is a 4-argument macro, for other
architectures it is a 3-argument function.  This patch makes it easier
to implement atomic operations incrementally (first as a generic version
and then in machine-specific code), which aids testing and
bisectability.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

call_rcu: drop mutex

The mutex is being used only to protect OR accesses to the flags.
Just use atomic operations for that.

[ Edit: this also fixes busy-looping on flags that were previously read
without volatile access, which could lead to never-ending loop given the
appropriate set of compiler optimisations. ]

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

uatomic: add uatomic_and

Only x86 support is arch-specific for now, the rest uses either cmpxchg
or gcc __sync_ builtins.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

uatomic: add uatomic_or

For now, only the version returning void is used.  Also, I am not
providing PPC versions because I would be unable to test it right
now.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add missing -lwfqueue to add rcu flavors

Otherwise, linking with the flavor .so leads to:

/usr/local/lib/liburcu-bp.so: undefined reference to `cds_wfq_enqueue'
/usr/local/lib/liburcu-bp.so: undefined reference to `cds_wfq_init'
/usr/local/lib/liburcu-bp.so: undefined reference to `cds_wfq_node_init'

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>