Date: Sat, 9 Nov 2002 15:17:51 -0500 (EST) From: Jeff Roberson <jroberson@chesapeake.net> To: Terry Lambert <tlambert2@mindspring.com> Cc: Andrea Campi <andrea@webcom.it>, <current@FreeBSD.ORG> Subject: Re: Deadlock during buildworld Message-ID: <20021109151628.A97372-100000@mail.chesapeake.net> In-Reply-To: <3DCCF856.66707AA7@mindspring.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 9 Nov 2002, Terry Lambert wrote: > Jeff Roberson wrote: > > [ ... patch ... looks right to me, but I wonder if it opens a race > window in the "vput() blocks" case ...either way, it's no > worse than the code it replaces... ] > > > I'll look into it some more, but it looks like someone is holding the exec > > map while they are trying to lock that vnode. That doesn't seem correct, > > but this patch will fix the problem if that is the case. It's too bad we > > dont have witness on lockmgr.. > > It's too bad that lockmgr() is such a hulking behemoth; it really > should not be being used for this sort of thing (IMO). Every time > someone complained about it in the past (including me, FWIW), the > answer was that its use was a stopgap, and that once the code was > working, the locks would be replaced with lighter-weight code which > was not yet written. Well, now it's written, and it has WITNESS, as > a side benefit, but the lockmgr() calls are unforntunately still > being put in. 8-(. > As it stands, sx locks do not support all of the semantics that are required by the current lockmgr() users. For example, recursive exclusive acquires, interlock, and exclusive upgrades are not supported. Cheers, Jeff To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021109151628.A97372-100000>