From owner-freebsd-arch@FreeBSD.ORG Sat Oct 25 22:47:05 2014 Return-Path: Delivered-To: freebsd-arch@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by hub.freebsd.org (Postfix) with ESMTPS id 88A20374 for ; Sat, 25 Oct 2014 22:47:05 +0000 (UTC) Received: from h2.funkthat.com (gate2.funkthat.com [208.87.223.18]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client CN "funkthat.com", Issuer "funkthat.com" (not verified)) by mx1.freebsd.org (Postfix) with ESMTPS id 549A8601 for ; Sat, 25 Oct 2014 22:47:05 +0000 (UTC) Received: from h2.funkthat.com (localhost [127.0.0.1]) by h2.funkthat.com (8.14.3/8.14.3) with ESMTP id s9PMl2TF033192 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Sat, 25 Oct 2014 15:47:03 -0700 (PDT) (envelope-from jmg@h2.funkthat.com) Received: (from jmg@localhost) by h2.funkthat.com (8.14.3/8.14.3/Submit) id s9PMl2GH033191; Sat, 25 Oct 2014 15:47:02 -0700 (PDT) (envelope-from jmg) Date: Sat, 25 Oct 2014 15:47:02 -0700 From: John-Mark Gurney To: Mateusz Guzik Subject: Re: refcount_release_take_##lock Message-ID: <20141025224702.GY82214@funkthat.com> Mail-Followup-To: Mateusz Guzik , freebsd-arch@freebsd.org References: <20141025184448.GA19066@dft-labs.eu> <20141025190407.GU82214@funkthat.com> <20141025192632.GB19066@dft-labs.eu> <20141025195334.GW82214@funkthat.com> <20141025201240.GC19066@dft-labs.eu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20141025201240.GC19066@dft-labs.eu> User-Agent: Mutt/1.4.2.3i X-Operating-System: FreeBSD 7.2-RELEASE i386 X-PGP-Fingerprint: 54BA 873B 6515 3F10 9E88 9322 9CB1 8F74 6D3F A396 X-Files: The truth is out there X-URL: http://resnet.uoregon.edu/~gurney_j/ X-Resume: http://resnet.uoregon.edu/~gurney_j/resume.html X-TipJar: bitcoin:13Qmb6AeTgQecazTWph4XasEsP7nGRbAPE X-to-the-FBI-CIA-and-NSA: HI! HOW YA DOIN? can i haz chizburger? X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.2.2 (h2.funkthat.com [127.0.0.1]); Sat, 25 Oct 2014 15:47:03 -0700 (PDT) Cc: freebsd-arch@freebsd.org X-BeenThere: freebsd-arch@freebsd.org X-Mailman-Version: 2.1.18-1 Precedence: list List-Id: Discussion related to FreeBSD architecture List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 25 Oct 2014 22:47:05 -0000 Mateusz Guzik wrote this message on Sat, Oct 25, 2014 at 22:12 +0200: > On Sat, Oct 25, 2014 at 12:53:34PM -0700, John-Mark Gurney wrote: > > Mateusz Guzik wrote this message on Sat, Oct 25, 2014 at 21:26 +0200: > > > On Sat, Oct 25, 2014 at 12:04:07PM -0700, John-Mark Gurney wrote: > > > > Mateusz Guzik wrote this message on Sat, Oct 25, 2014 at 20:44 +0200: > > > > > The following idiom is used here and there: > > > > > > > > > > int old; > > > > > old = obj->ref; > > > > > if (old > 1 && atomic_cmpset_int(&obj->ref, old, old -1)) > > > > > return; > > > > > lock(&something); > > > > > if (refcount_release(&obj->ref) == 0) { > > > > > unlock(&something); > > > > > return; > > > > > } > > > > > free up > > > > > unlock(&something); > > > > > > > > > > ========== > > > > > > > > Couldn't this be better written as: > > > > if (__predict_false(refcount_release(&obj->ref) == 0)) { > > > > lock(&something); > > > > if (__predict_true(!obj->ref)) { > > > > free up > > > > } > > > > unlock(&something); > > > > } > > > > > > > > The reason I'm asking is that I changed how IPsec SA ref counting was > > > > handled, and used something similar... > > > > > > > > My code gets rid of a branch, and is better in that it uses refcount > > > > API properly, instead of using atomic_cmpset_int... > > > > > > This is used when given obj is kept on a list and code which traverses > > > it (locked) expects found objects to be valid to ref. > > > > > > If we were to reach count of 0 and then lock, it would be possible that > > > other thread refed + unrefed the object and is now trying to lock as > > > well. > > > > Per the email I wrote to Ian, this "assumption" needs to be well > > documented that though the "list" has a reference, and that this > > reference is not accounted for in the ref count... > > > > And I personally think that it's a bug for the list to not hold it's > > own reference... Yes, then you need to compare for when the ref count > > hits one, and do the lock/dec/free/unlock, but that keeps the refcount > > sane... > > Well, this is for stuff which cleans up after itself. I hope that everything cleans up after itself.. :) otherwise we'd have memory leaks everywhere... > Example usage is with per-uid stats for resource limits. These > automatically free themselves with the last cred with given uid. This example doesn't give me enough information to decide what you mean.. Is there a hash table that we look up these cred structures? or are they referenced from an already existing reference? > This has its own problems (like constant creation and destruction of > stuff for the same cred), but seems ok enough for some cases. > > Otherwise we would have to actively free these structs somehow. I'm still not sure how your example addresses this.. I believe you wrote the data structure case, but it wasn't clear that is what you were doing, and as I said, I think it's a bug to have an implicit ref in such data structures w/o properly documenting them... Part of the reason why we need documentation to make sure people don't make mistakes like these... > > > That could be remedied for type stable object by having a generation > > > counter, but I doubt it's worth it. Not to mention objects we lock here > > > are freeable :) > > > > That's too heavy weight... > > > > > > > I decided to implement it as a common function. > > > > > > > > > > We have only refcount.h and I didn't want to bloat all including code > > > > > with additional definitions and as such I came up with a macro that has > > > > > to be used in .c file and that will define appropriate inline func. > > > > > > > > > > I'm definitely looking for better names for REFCOUNT_RELEASE_TAKE_USE_ > > > > > macro, assuming it has to stay. > > > > > > > > You could shorten it to REFCNT_REL_TAKE_ > > > > > > > > > > All function use full 'refcount_release' and the like, so that would be > > > inconsistent. > > > > > > Losing 'take' may be an option, I don't know. > > > > Yeh, the only advantage is that it only appears once per file used, > > so it's not THAT long... > > > > > > > Comments? > > > > > > > > > > > Will you update the refcount(9) man page w/ documentation before > > > > committing? > > > > > > Sure. > > > > Thanks. -- John-Mark Gurney Voice: +1 415 225 5579 "All that I will do, has been done, All that I have, has not."