Date: Sat, 22 Apr 2017 08:40:58 +0300 From: Konstantin Belousov <kostikbel@gmail.com> To: John Baldwin <jhb@freebsd.org> Cc: Eric Badger <badger@freebsd.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-11@freebsd.org Subject: Re: svn commit: r315949 - in stable: 10/sys/kern 10/sys/sys 10/tests/sys/kern 11/sys/kern 11/sys/sys 11/tests/sys/kern Message-ID: <20170422054058.GM1788@kib.kiev.ua> In-Reply-To: <6768216.Fy5yUeFf9F@ralph.baldwin.cx> References: <201703251333.v2PDXOmS004369@repo.freebsd.org> <6768216.Fy5yUeFf9F@ralph.baldwin.cx>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Apr 21, 2017 at 03:11:00PM -0700, John Baldwin wrote: > So this MFC happens to break the KBI of 'struct thread' because > 'struct sigqueue' is embedded in 'struct thread' (rather than > a pointer). This apparently results in panics if you use an > nvidia-driver package built on 11.0-release on 11.0-stable, and > that's the type of KBI we try to preserve. (We tend to only add new > fields to the end of 'struct thread' and 'struct proc' in stable for > example.) This one is a bit ugly to deal with though. I wonder if > for the stable branches only if we could move the sq_ptrace out of > the sigqueue struct and add it as a new field to the end of 'struct > thread' explicitly? It would probably be pretty hacky, but it would be > good to fix this before 11.1 so that the KBI wouldn't be broken for a > user going from 11.0 to 11.1. It might be easier to repurpose existing td_sigqueue as padding with corrected size, and place real sigqueue at the end of the structure. add
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20170422054058.GM1788>