Date: Thu, 29 Oct 2015 00:54:06 +0100 From: Mark Martinec <Mark.Martinec+freebsd@ijs.si> To: freebsd-hackers@freebsd.org Subject: Re: ABI stability for loadable modules Message-ID: <022117707d6196d7d117db0288033996@mailbox.ijs.si> In-Reply-To: <alpine.GSO.1.10.1510281920020.26829@multics.mit.edu> References: <CAHRLrwNcLPrztuVvhrRj3Ya=Aw-5eDMRGD=L0wwVzfmPgQXwCQ@mail.gmail.com> <alpine.GSO.1.10.1510281920020.26829@multics.mit.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
> Kernel ABI stability is guaranteed(*) within a major release branch. > -Ben > > (*) there are occasional exceptions involving portions of the network > stack, but 10.x is probably better than 9.x in this regard Two ports come to mind which crash the system when built under 10.0 and are attempted to be loaded under 10.2: emulators/virtualbox-ose-kmod x11/nvidia-driver Mark 2015-10-29 00:21, Benjamin Kaduk wrote: > On Wed, 28 Oct 2015, Koert van der Veer wrote: > >> TL;DR: Can I safely load a kernel module compiled against a slightly >> different kernel? >> >> >> I maintain a fleet of FreeBSD machines. We've encountered a problem >> with >> one of the modules (iscsi doesn't allow passwords > 16 chars). I've >> determined that patching the module solves the problem for us, and >> doesn't >> affect any other kernel component. I'd like to distribute the patched >> module and user-space components. My use-case allows me to unload the >> iscsi >> module, but doesn't allow me to reboot the machine. >> >> >> However, we're running an array of different kernels (10.0-RELEASE, >> 10.0-RELEASE-p9, 10.0-STABLE, 10.1-RELEASE, 10.1-RELEASE-p6). I'd like >> to >> minimize the number of distinct binaries being distributed, so >> preferably >> I'd compile only one or two. My preliminary tests on VMs show me that >> this >> does not cause any immediate problems. Is this actually safe, or do I >> need >> to make distinct sets of binaries for each minor version and >> patchlevel of >> kernel in production? > > Kernel ABI stability is guaranteed(*) within a major release branch. > > -Ben > > (*) there are occasional exceptions involving portions of the network > stack, but 10.x is probably better than 9.x in this regard
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?022117707d6196d7d117db0288033996>