[lkml]   [2009]   [Feb]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [RFC git tree] Userspace RCU (urcu) for Linux (repost)
* Paul E. McKenney ( wrote:
> On Fri, Feb 06, 2009 at 05:06:40AM -0800, Paul E. McKenney wrote:
> > On Thu, Feb 05, 2009 at 11:58:41PM -0500, Mathieu Desnoyers wrote:
> > > (sorry for repost, I got the ltt-dev email wrong in the previous one)
> > >
> > > Hi Paul,
> > >
> > > I figured out I needed some userspace RCU for the userspace tracing part
> > > of LTTng (for quick read access to the control variables) to trace
> > > userspace pthread applications. So I've done a quick-and-dirty userspace
> > > RCU implementation.
> > >
> > > It works so far, but I have not gone through any formal verification
> > > phase. It seems to work on paper, and the tests are also OK (so far),
> > > but I offer no guarantee for this 300-lines-ish 1-day hack. :-) If you
> > > want to comment on it, it would be welcome. It's a userland-only
> > > library. It's also currently x86-only, but only a few basic definitions
> > > must be adapted in urcu.h to port it.
> > >
> > > Here is the link to my git tree :
> > >
> > > git://
> > >
> > >;a=summary
> >
> > Very cool!!! I will take a look!
> >
> > I will also point you at a few that I have put together:
> >
> > git://
> >
> > (In the CodeSamples/defer directory.)
> Interesting approach, using the signal to force memory-barrier execution!
> o One possible optimization would be to avoid sending a signal to
> a blocked thread, as the context switch leading to blocking
> will have implied a memory barrier -- otherwise it would not
> be safe to resume the thread on some other CPU. That said,
> not sure whether checking to see whether a thread is blocked is
> any faster than sending it a signal and forcing it to wake up.

I'm not sure it will be any faster, and it could be racy too. How would
you envision querying the execution state of another thread ?

> Of course, this approach does require that the enclosing
> application be willing to give up a signal. I suspect that most
> applications would be OK with this, though some might not.

If we want to make this transparent to the application, we'll have to
investigate further in sigaction() and signal() library override I

> Of course, I cannot resist pointing to an old LKML thread:
> But I think that the time is now right. ;-)
> o I don't understand the purpose of rcu_write_lock() and
> rcu_write_unlock(). I am concerned that it will lead people
> to decide that a single global lock must protect RCU updates,
> which is of course absolutely not the case. I strongly
> suggest making these internal to the urcu.c file. Yes,
> uses of urcu_publish_content() would then hit two locks (the
> internal-to-urcu.c one and whatever they are using to protect
> their data structure), but let's face it, if you are sending a
> signal to each and every thread, the additional overhead of the
> extra lock is the least of your worries.

Ok, just changed it.

> If you really want to heavily optimize this, I would suggest
> setting up a state machine that permits multiple concurrent
> calls to urcu_publish_content() to share the same set of signal
> invocations. That way, if the caller has partitioned the
> data structure, global locking might be avoided completely
> (or at least greatly restricted in scope).

That brings an interesting question about urcu_publish_content :

void *urcu_publish_content(void **ptr, void *new)
void *oldptr;

oldptr = *ptr;
*ptr = new;


return oldptr;

Given that we take a global lock around the pointer assignment, we can
safely assume, from the caller's perspective, that the update will
happen as an "xchg" operation. So if the caller does not have to copy
the old data, it can simply publish the new data without taking any
lock itself.

So the question that arises if we want to remove global locking is :
should we change this

oldptr = *ptr;
*ptr = new;

for an atomic xchg ?

> Of course, if updates are rare, the optimization would not
> help, but in that case, acquiring two locks would be even less
> of a problem.

I plan updates to be quite rare, but it's always good to foresee how
that kind of infrastructure could be misused. :-)

> o Is urcu_qparity relying on initialization to zero? Or on the
> fact that, for all x, 1-x!=x mod 2^32? Ah, given that this is
> used to index urcu_active_readers[], you must be relying on
> initialization to zero.

Yes, starts at 0.

> o In rcu_read_lock(), why is a non-atomic increment of the
> urcu_active_readers[urcu_parity] element safe? Are you
> relying on the compiler generating an x86 add-to-memory
> instruction?
> Ditto for rcu_read_unlock().
> Ah, never mind!!! I now see the __thread specification,
> and the keeping of references to it in the reader_data list.

Exactly :)

> o Combining the equivalent of rcu_assign_pointer() and
> synchronize_rcu() into urcu_publish_content() is an interesting
> approach. Not yet sure whether or not it is a good idea. I
> guess trying it out on several applications would be the way
> to find out. ;-)
> That said, I suspect that it would be very convenient in a
> number of situations.

I thought so. It seemed to be a natural way to express it to me. Usage
will tell.

> o It would be good to avoid having to pass the return value
> of rcu_read_lock() into rcu_read_unlock(). It should be
> possible to avoid this via counter value tricks, though this
> would add a bit more code in rcu_read_lock() on 32-bit machines.
> (64-bit machines don't have to worry about counter overflow.)
> See the recently updated version of CodeSamples/defer/rcu_nest.[ch]
> in the aforementioned git archive for a way to do this.
> (And perhaps I should apply this change to SRCU...)

See my other mail about this.

> o Your test looks a bit strange, not sure why you test all the
> different variables. It would be nice to take a test duration
> as an argument and run the test for that time.

I made a smaller version which only reads a single variable. I agree
that the initial test was a bit strange on that aspect.

I'll do a version which takes a duration as parameter.

> I killed the test after better part of an hour on my laptop,
> will retry on a larger machine (after noting the 18 threads
> created!). (And yes, I first tried Power, which objected
> strenously to the "mfence" and "lock; incl" instructions,
> so getting an x86 machine to try on.)

That should be easy enough to fix. A bit of primitive cut'n'paste would

> Again, looks interesting! Looks plausible, although I have not 100%
> convinced myself that it is perfectly bug-free. But I do maintain
> a healthy skepticism of purported RCU algorithms, especially ones that
> I have written. ;-)

That's always good. I also tend to always be very skeptical about what I
write and review.

Thanks for the thorough review.


> Thanx, Paul

Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68

 \ /
  Last update: 2009-02-08 23:47    [W:0.178 / U:9.464 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site