Date: Wed, 25 Mar 2015 12:27:49 +0100 From: Hans Petter Selasky <hps@selasky.org> To: Gleb Smirnoff <glebius@FreeBSD.org> Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org Subject: Re: svn commit: r280495 - head/sys/kern Message-ID: <55129BB5.4000502@selasky.org> In-Reply-To: <20150325111433.GT64665@FreeBSD.org> References: <201503250855.t2P8tZFp038467@svn.freebsd.org> <20150325105409.GS64665@FreeBSD.org> <551296C6.9070402@selasky.org> <20150325111433.GT64665@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 03/25/15 12:14, Gleb Smirnoff wrote: > On Wed, Mar 25, 2015 at 12:06:46PM +0100, Hans Petter Selasky wrote: > H> On 03/25/15 11:54, Gleb Smirnoff wrote: > H> > On Wed, Mar 25, 2015 at 08:55:35AM +0000, Hans Petter Selasky wrote: > H> > H> Author: hselasky > H> > H> Date: Wed Mar 25 08:55:34 2015 > H> > H> New Revision: 280495 > H> > H> URL: https://svnweb.freebsd.org/changeset/base/280495 > H> > H> > H> > H> Log: > H> > H> Implement a simple OID number garbage collector. Given the increasing > H> > H> number of dynamically created and destroyed SYSCTLs during runtime it > H> > H> is very likely that the current new OID number limit of 0x7fffffff can > H> > H> be reached. Especially if dynamic OID creation and destruction results > H> > H> from automatic tests. Additional changes: > H> > H> > H> > H> - Optimize the typical use case by decrementing the next automatic OID > H> > H> sequence number instead of incrementing it. This saves searching time > H> > H> when inserting new OIDs into a fresh parent OID node. > H> > H> > H> > H> - Add simple check for duplicate non-automatic OID numbers. > H> > > H> > Why didn't you use alloc_unr(9) for that? > H> > > H> > H> Hi Gleb, > H> > H> I thought about using alloc_unr(). The problem is that sysctls have no > H> clear concept of freeing. For example in some existing code sysctl are > H> unregistered and registered again assuming that the oid_number will be > H> preserved. I didn't want touch those parts. Also, hence we are already > H> traversing a list to insert an SYSCTL object in an ordered fashion, the > H> benefit of alloc_unr() is not that big. What do you think? > > Is it possible to split the space into two halves: one for static OIDs and > other for dynamic ones? The latter allocated via alloc_unr? > Yes, it is possible to do. One issue though is that the OID numbers shouldn't be recycled right away to preserve the old behaviour. It is not written anywhere, but I think there is some cleverness into the fact that the newoid value does not repeat instantly, as would happen using alloc_unr() to avoid races accessing OIDs from userspace? Does alloc_unr() support a sort of rotating allocation number mode within the defined range? The OID numbers are only required to be unique per directory level from what I can see. --HPS
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?55129BB5.4000502>