Date: Mon, 14 Dec 2009 18:02:54 +0100 From: Attilio Rao <attilio@freebsd.org> To: John Baldwin <jhb@freebsd.org> Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org Subject: Re: svn commit: r200447 - in head: share/man/man9 sys/kern sys/sys Message-ID: <3bbf2fe10912140902m407fa766q3a5e5bb6993723f9@mail.gmail.com> In-Reply-To: <200912141013.32839.jhb@freebsd.org> References: <200912122131.nBCLV71f064304@svn.freebsd.org> <200912141013.32839.jhb@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
2009/12/14 John Baldwin <jhb@freebsd.org>: > On Saturday 12 December 2009 4:31:07 pm Attilio Rao wrote: >> Author: attilio >> Date: Sat Dec 12 21:31:07 2009 >> New Revision: 200447 >> URL: http://svn.freebsd.org/changeset/base/200447 >> >> Log: >> =C2=A0 In current code, threads performing an interruptible sleep (on bo= th >> =C2=A0 sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr),= will >> =C2=A0 leave the waiters flag on forcing the owner to do a wakeup even w= hen if >> =C2=A0 the waiter queue is empty. >> =C2=A0 That operation may lead to a deadlock in the case of doing a fake= wakeup >> =C2=A0 on the "preferred" (based on the wakeup algorithm) queue while th= e other >> =C2=A0 queue has real waiters on it, because nobody is going to wakeup t= he 2nd >> =C2=A0 queue waiters and they will sleep indefinitively. >> >> =C2=A0 A similar bug, is present, for lockmgr in the case the waiters ar= e >> =C2=A0 sleeping with LK_SLEEPFAIL on. =C2=A0In this case, even if the wa= iters queue >> =C2=A0 is not empty, the waiters won't progress after being awake but th= ey will >> =C2=A0 just fail, still not taking care of the 2nd queue waiters (as ins= tead the >> =C2=A0 lock owned doing the wakeup would expect). >> >> =C2=A0 In order to fix this bug in a cheap way (without adding too much = locking >> =C2=A0 and complicating too much the semantic) add a sleepqueue interfac= e which >> =C2=A0 does report the actual number of waiters on a specified queue of = a >> =C2=A0 waitchannel (sleepq_sleepcnt()) and use it in order to determine = if the >> =C2=A0 exclusive waiters (or shared waiters) are actually present on the= lockmgr >> =C2=A0 (or sx) before to give them precedence in the wakeup algorithm. >> =C2=A0 This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In or= der to >> =C2=A0 cope with it, add the tracking of how many exclusive LK_SLEEPFAIL= waiters >> =C2=A0 a lockmgr has and if all the waiters on the exclusive waiters que= ue are >> =C2=A0 LK_SLEEPFAIL just wake both queues. >> >> =C2=A0 The sleepq_sleepcnt() introduction and ABI breakage require >> =C2=A0 __FreeBSD_version bumping. > > Hmm, do you need an actual count of waiters or would a 'sleepq_empty()' > (similar to turnstile_empty()) method be sufficient? I need the count in order to fix properly LK_SLEEPFAIL case (the idea is: track exclusive waiters with LK_SLEEPFAIL on; if the number is equal to the actual sleepers on the queue then wake up both queues, otherwise nobody is going to take care of the shared waiters queue). Attilio --=20 Peace can only be achieved by understanding - A. Einstein
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3bbf2fe10912140902m407fa766q3a5e5bb6993723f9>