Date: Wed, 11 Mar 1998 08:27:13 -0700 From: "Justin T. Gibbs" <gibbs@plutotech.com> To: Eivind Eklund <eivind@yes.no> Cc: "Justin T. Gibbs" <gibbs@plutotech.com>, hackers@FreeBSD.ORG Subject: Re: userconfig data -> linker set -> ELF segment Message-ID: <199803111530.IAA13420@pluto.plutotech.com> In-Reply-To: Your message of "Wed, 11 Mar 1998 16:22:28 %2B0100." <19980311162228.43166@follo.net>
next in thread | previous in thread | raw e-mail | index | archive | help
>> > You're making things too difficult. 8) >> >> Exactly. Don't use a linker set at all. Convert all drivers to >> LKMs, load they "probe" section of all LKMs, have an entry point >> in the probe section register the driver with userconfig. > >This is not feasible at present. There are several reasons: > * The PCI LKM code in -current doesn't work Probably because of pciconf being broken. If SE approves, this can be easily remedied by taking the pciconf code from CAM. > * Loading 'just the probe code' isn't possible using a.out > (AFAIK) > * Depending on device drivers being LKMs lowers reliability (N > files that can fail, instead of just having a single kernel) Then provide a mechanism to have the driver entry point also called when linked statically. In a complete LKM solution, you should be able to load LKMs from a floppy for recovery or to add vendor supplied, binary only, modules during install. > * The probe is too late - userconfig (presently, at least) run > before anything is probed - to be able to stop harmful > probes etc. Which is one of the many flaws in our configuration scheme. It shouldn't be simply "probe" and "attach", but rather "register configuration items", run userconfig, probe, run userconfig again to set per instance config settings, perform attach. > * It is more work than I can chew, and thus won't be done as > part of the minor set of changes I wanted to do. I certainly agree, I was just stating how I believe it should really be done. >I was only planning to make userconfig data a linker set, and change >config to (also) scan sys/conf/*/files.extra and >sys/conf/*/options.extra, as an enabling technology for externally >developed kernel parts. > >> Linker sets are a pain. > >Why more so than LKMs? As far as I can tell, LKMs are presently more >of a pain than linker sets (though LKMs is also a more powerful >enabling technology). Because linker sets provide yet another stumbling block to providing good LKM support and I feel that real LKM support should be our goal. >Eivind. -- Justin To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199803111530.IAA13420>