Minor change (not visible to end-users): this commit introduces a
liburcu-common.so to hold wfqueue, wfstack and futex fall-back code.
This .so is used by all URCU flavors, as well as liburcu-cds.so.
I thought about it some more, and had discussions with various people,
and there are a few reasons to go for a scheme where rcu read lock
should be taken by the caller, and to pass call_rcu as a parameter to
the data structure init function:
A) The advantage, as Paul E. McKenney pointed out, is that one single .so
is enough to support all RCU flavors. Very convenient for external data
structure containers.
B) It clearly documents where rcu read-side locks are needed, so the user
keep control and in-depth understanding of their read-side locks.
C) When multiple API functions that require RCU read-side lock to be
held (sometimes even the same lock) throughout a sequence of API
calls, we have no choice but to let the caller hold the read-side
lock.
D) Due to support of multiple nesting of rcu read-side lock, any
"improvement" we could get by releasing the read-side lock in
retry loops would vanish in the cases where we are called within
nested C.S..
E) If a library uses synchronize_rcu, this should be clearly documented,
and even frowned upon, because this involves important limitations on
the design of the caller, and important performance hit. There are
usually ways to reach the same result through use of call_rcu, which
should really be used thoroughout these libraries.
F) It clearly documents when a data structure needs to use call_rcu
internally.
G) Some very early benchmark results show that there is indeed not
much performance gain to achieve by inlining call_rcu, even if it is
a version with a cache for the "call_rcu structure" lookup
(per-cpu/per-thread/global). So passing it as a parameter to
the data structure init function should be fine, even in cases
where it is called very often.
H) For use-cases where applications would like to use more than one
RCU flavor concurrently (which is now supported), leaving management
of RCU read-side C.S. to the reader allows the application to take
more than one RCU read-side lock across API calls. It also lets the
application specify its own call_rcu function that could handle more
than one RCU flavor.
So for all these reasons, I reverting back to the API we have in our
last release (0.6.4).
Paolo Bonzini [Wed, 17 Aug 2011 09:42:51 +0000 (05:42 -0400)]
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>
Paolo Bonzini [Wed, 17 Aug 2011 09:31:21 +0000 (05:31 -0400)]
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>
Paolo Bonzini [Tue, 9 Aug 2011 12:37:14 +0000 (08:37 -0400)]
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>
Duncan Sands [Tue, 16 Aug 2011 11:10:01 +0000 (07:10 -0400)]
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.
Paolo Bonzini [Tue, 9 Aug 2011 12:37:14 +0000 (08:37 -0400)]
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>
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.
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.
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
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]
* 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).
./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
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>
projects / urcu.git / log
