From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754703AbZBGXil (ORCPT ); Sat, 7 Feb 2009 18:38:41 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1753866AbZBGXic (ORCPT ); Sat, 7 Feb 2009 18:38:32 -0500 Received: from tomts20-srv.bellnexxia.net ([209.226.175.74]:42031 "EHLO tomts20-srv.bellnexxia.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753776AbZBGXib (ORCPT ); Sat, 7 Feb 2009 18:38:31 -0500 X-IronPort-Anti-Spam-Filtered: true X-IronPort-Anti-Spam-Result: ApsEAA6kjUlMQWt2/2dsb2JhbACBbcpwhBoG Date: Sat, 7 Feb 2009 18:38:27 -0500 From: Mathieu Desnoyers To: "Paul E. McKenney" Cc: ltt-dev@lists.casi.polymtl.ca, linux-kernel@vger.kernel.org, Robert Wisniewski Subject: Re: [RFC git tree] Userspace RCU (urcu) for Linux (repost) Message-ID: <20090207233827.GA3557@Krystal> References: <20090206030543.GB8560@Krystal> <20090206045841.GA12995@Krystal> <20090206130640.GB10918@linux.vnet.ibm.com> <20090206163432.GF10918@linux.vnet.ibm.com> <20090207151028.GA11150@linux.vnet.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Content-Disposition: inline In-Reply-To: <20090207151028.GA11150@linux.vnet.ibm.com> X-Editor: vi X-Info: http://krystal.dyndns.org:8080 X-Operating-System: Linux/2.6.21.3-grsec (i686) X-Uptime: 18:27:34 up 37 days, 23:25, 3 users, load average: 0.84, 0.64, 0.47 User-Agent: Mutt/1.5.18 (2008-05-17) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Paul E. McKenney (paulmck@linux.vnet.ibm.com) wrote: > On Fri, Feb 06, 2009 at 08:34:32AM -0800, 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://lttng.org/userspace-rcu.git > > > > > > > > http://lttng.org/cgi-bin/gitweb.cgi?p=userspace-rcu.git;a=summary > > > > > > Very cool!!! I will take a look! > > > > > > I will also point you at a few that I have put together: > > > > > > git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/perfbook.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. > > > > 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. > > > > Of course, I cannot resist pointing to an old LKML thread: > > > > http://lkml.org/lkml/2001/10/8/189 > > > > 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. > > > > 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). > > > > 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. > > > > 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. > > > > 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. > > > > 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. > > > > 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...) > > > > 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 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.) > > > > 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. ;-) > > OK, here is one sequence of concern... > Let's see.. > o Thread 0 starts rcu_read_lock(), picking up the current > get_urcu_qparity() into the local variable urcu_parity(). > Assume that the value returned is zero. > > o Thread 0 is now preempted. > > o Thread 1 invokes urcu_publish_content(): > > o It substitutes the pointer. > > o It forces all threads to execute a memory barrier > (thread 0 runs just long enough to process its signal > and then is immediately preempted again). > > o It switches the parity, which is now one. > > o It waits for all readers on parity zero, and there are > none, because thread 0 has not yet registered itself. > > o It therefore returns the old pointer. So far, so good. > > o Thread 0 now resumes: > > o It increments its urcu_active_readers[0]. > > o It forces a compiler barrier. > > o It returns zero (why not store this in thread-local > storage rather than returning?). > To support nested rcu_read_locks. (that's the only reason) > o It enters its critical section, obtaining a reference > to the new pointer that thread 1 just published. > > o Thread 1 now again invokes urcu_publish_content(): > > o It substitutes the pointer. > > o It forces all threads to execute a memory barrier, > including thread 0. > > o It switches the parity, which is now zero. > > o It waits for all readers on parity one, and there are > none, because thread 0 has registered itself on parity > zero!!! > > o Thread 1 therefore returns the old pointer. > > o Thread 1 frees the old pointer, which thread 0 is still > using!!! > Ah, yes, you are right. > 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. > This approach seems very interesting. > 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. ;-) Hrm, I'm trying to figure out what kind of memory backend you need to put your counters for each quiescent state period. Is this free-running counter indexing a very large array ? I doubt it does. Then how does it make sure we don't roll back to the old array entries ? This latter solution could break jump-based probing of programs soon-to-be available in gcc. The probes are meant to be of short duration, but the fact is that this design lets the debugger inject code without resorting to a breakpoint, which might therefore break your "short time between instructions" assumption. It's very unlikely, but possible. > > 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. > Yes. It's not because we are not able to create the faulty condition that it will _never_ happen. It must therefore be taken care of. Mathieu > Thanx, Paul > -- Mathieu Desnoyers OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68