urcu.git
15 years agoAdd rcu_assign_pointer
Mathieu Desnoyers [Mon, 9 Feb 2009 05:29:58 +0000 (00:29 -0500)] 
Add rcu_assign_pointer

rcu_assign_pointer has a memory barrier which lets the writer make sure the data
has been properly written to memory before setting the pointer.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoRemove parameter from rcu_read_lock()
Mathieu Desnoyers [Mon, 9 Feb 2009 04:56:15 +0000 (23:56 -0500)] 
Remove parameter from rcu_read_lock()

Also makes the read fast-path twice faster :

7 cycles instead of 14 on a 8-cores x86_64.

Mathieu :

I limited the amount of nested readers to 256. Should be enough and lets us use
testb generically.

Changed the 64-bits code to make it the same as 32-bits. I prefer to have the
exact same behavior on both architectures.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoAdd randomness to yield debug test
Mathieu Desnoyers [Mon, 9 Feb 2009 00:21:34 +0000 (19:21 -0500)] 
Add randomness to yield debug test

Add randomness to allow 1/2 odds that the yield will be done.
This lets some paths go quickly and others not.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoAdd DEBUG_YIELD, add test duration
Mathieu Desnoyers [Mon, 9 Feb 2009 00:01:58 +0000 (19:01 -0500)] 
Add DEBUG_YIELD, add test duration

Add some testing calling the scheduler, and add a duration parameter to
test_urcu.c.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoChange API
Mathieu Desnoyers [Sun, 8 Feb 2009 22:10:22 +0000 (17:10 -0500)] 
Change API

* rcu_read_lock

> A patch below to allow nested rcu_read_lock() while keeping to the Linux
> kernel API, just FYI.  One can argue that the overhead of accessing the
> extra per-thread variables is offset by the fact that there no longer
> needs to be a return value from rcu_read_lock() nor an argument to
> rcu_read_unlock(), but hard to say.
>

I ran your modified version within my benchmarks :

with return value : 14.164 cycles per read
without return value : 16.4017 cycles per read

So we have a 14% performance decrease due to this. We also pollute the
branch prediction buffer and we add a cache access due to the added
variables in the TLS. Returning the value has the clear advantage of
letting the compiler keep it around in registers or on the stack, which
clearly costs less.

So I think the speed factor outweights the visual considerations. Maybe
we could switch to something like :

unsigned int qparity;

urcu_read_lock(&qparity);
...
urcu_read_unlock(&qparity);

That would be a bit like local_irq_save() in the kernel, except that we
could do it in a static inline because we pass the address. I
personnally dislike the local_irq_save() way of hiding the fact that it
writes to the variable in a "clever" macro. I'd really prefer to leave
the " & ".

* rcu_write_lock

Removed.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoAdd timing tests
Mathieu Desnoyers [Sun, 8 Feb 2009 20:42:08 +0000 (15:42 -0500)] 
Add timing tests

initial results :

On a 8-cores x86_64

test_rwlock_timing.c : 4339.07 cycles
test_urcu_timing.c   :   14.16 cycles

Speedup : 306

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoDo two parity flip in the writer to fix race condition
Paul E. McKenney [Sun, 8 Feb 2009 00:00:55 +0000 (19:00 -0500)] 
Do two parity flip in the writer to fix race condition

On Sat, Feb 07, 2009 at 07:10:28AM -0800, Paul E. McKenney wrote:
> So, how to fix?  Here are some approaches:
>
> o     Make urcu_publish_content() do two parity flips rather than one.
>       I use this approach in my rcu_rcpg, rcu_rcpl, and rcu_rcpls
>       algorithms in CodeSamples/defer.
>
> o     Use a single free-running counter, in a manner similar to rcu_nest,
>       as suggested earlier.  This one is interesting, as I rely on a
>       read-side memory barrier to handle the long-preemption case.
>       However, if you believe that any thread that waits several minutes
>       between executing adjacent instructions must have been preempted
>       (which the memory barriers that are required to do a context
>       switch), then a compiler barrier suffices.  ;-)
>
> Of course, the probability of seeing this failure during test is quite
> low, since it is unlikely that thread 0 would run just long enough to
> execute its signal handler.  However, it could happen.  And if you were
> to adapt this algorithm for use in a real-time application, then priority
> boosting could cause this to happen naturally.

And here is a patch, taking the first approach.  It also exposes a
synchronize_rcu() API that is used by the existing urcu_publish_content()
API.  This allows easier handling of structures that are referenced by
more than one pointer.  And should also allow to be more easily plugged
into my rcutorture test.  ;-)

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agotest_urcu.c: use gettid()
Bert Wesarg [Sat, 7 Feb 2009 23:59:29 +0000 (18:59 -0500)] 
test_urcu.c: use gettid()

It's probably better to print the tid for each thread, not the pid.

Signed-off-by: Bert Wesarg <bert.wesarg@googlemail.com>
15 years agoURCU : use pthread_equal()
Bert Wesarg [Fri, 6 Feb 2009 11:45:42 +0000 (06:45 -0500)] 
URCU : use pthread_equal()

But you should use pthread_equal() for you equality test of pthread_t.

Signed-off-by: Bert Wesarg <bert.wesarg@googlemail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoRun longer tests
Mathieu Desnoyers [Fri, 6 Feb 2009 11:13:49 +0000 (06:13 -0500)] 
Run longer tests

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
15 years agoupdate Makefile, -Wall
Mathieu Desnoyers [Fri, 6 Feb 2009 03:25:51 +0000 (22:25 -0500)] 
update Makefile, -Wall

15 years agoremove ugly gcc warning removal ack, simply cast the caller parameter
Mathieu Desnoyers [Fri, 6 Feb 2009 02:47:01 +0000 (21:47 -0500)] 
remove ugly gcc warning removal ack, simply cast the caller parameter

15 years agoadd acknowledgements, fix gcc warnings
Mathieu Desnoyers [Fri, 6 Feb 2009 02:41:04 +0000 (21:41 -0500)] 
add acknowledgements, fix gcc warnings

15 years agofix wait_for_quiescent_state
Mathieu Desnoyers [Fri, 6 Feb 2009 02:14:20 +0000 (21:14 -0500)] 
fix wait_for_quiescent_state

15 years agoadd licensing
Mathieu Desnoyers [Fri, 6 Feb 2009 01:49:36 +0000 (20:49 -0500)] 
add licensing

15 years agomodify test values
Mathieu Desnoyers [Fri, 6 Feb 2009 01:46:14 +0000 (20:46 -0500)] 
modify test values

15 years agoruns
Mathieu Desnoyers [Fri, 6 Feb 2009 01:04:29 +0000 (20:04 -0500)] 
runs

15 years agoruns
Mathieu Desnoyers [Fri, 6 Feb 2009 01:04:03 +0000 (20:04 -0500)] 
runs

15 years agoadd make clean
Mathieu Desnoyers [Fri, 6 Feb 2009 00:42:15 +0000 (19:42 -0500)] 
add make clean

15 years agoruns
Mathieu Desnoyers [Fri, 6 Feb 2009 00:41:34 +0000 (19:41 -0500)] 
runs

15 years agocompile
Mathieu Desnoyers [Fri, 6 Feb 2009 00:16:25 +0000 (19:16 -0500)] 
compile

15 years agoinit version
Mathieu Desnoyers [Fri, 6 Feb 2009 00:06:44 +0000 (19:06 -0500)] 
init version

This page took 0.029669 seconds and 4 git commands to generate.