Date: Fri, 6 Apr 2012 11:38:07 +0530 From: "Jayachandran C." <c.jayachandran@gmail.com> To: Stanislav Sedov <stas@freebsd.org> Cc: Doug Barton <dougb@freebsd.org>, mips@freebsd.org, current@freebsd.org, Juli Mallett <jmallett@freebsd.org>, Garrett Cooper <yanegomi@gmail.com>, FreeBSD Tinderbox <tinderbox@freebsd.org> Subject: Re: There is a known problem with MIPS tinderbox. Message-ID: <CA%2B7sy7BwFQxUaHJsom36xGJXO9rd71%2BuERFcxcR4dUGkbwgNjA@mail.gmail.com> In-Reply-To: <20120405142423.90d48b27.stas@FreeBSD.org> References: <CACVs6=_kNJQZLouh6b%2BW9Ktj-CUFyeg7MVhas1qzoLGymJL7Kg@mail.gmail.com> <20120405142423.90d48b27.stas@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Apr 6, 2012 at 2:54 AM, Stanislav Sedov <stas@freebsd.org> wrote: > On Thu, 5 Apr 2012 12:39:00 -0700 > Juli Mallett <jmallett@FreeBSD.org> mentioned: > >> On Thu, Apr 5, 2012 at 12:28, Doug Barton <dougb@freebsd.org> wrote: >> > But as always, if someone comes up with an actual problem related to t= he >> > [BIND] update I'm happy to address it. >> >> The MIPS tinderbox build is currently broken. =A0This is related to a >> binutils bug for which a workaround was committed to head. =A0For some >> reason, the workaround seems to be sufficient for universe done >> independently, but not the tinderbox. >> >> There is a fix for binutils which is known, but the fix was made after >> the GPLv3 switch. =A0HJ Lu, who made the change, has been contacted >> about making the change available under GPLv2, but so far has not >> replied. =A0Independently, jchandra@ is investigating the problem in >> binutils and attempting to come up with a fix from scratch. >> > > What I still cannot understand is that why my workaround doesn't work > on tinderbox and for some people builds. =A0It should disable linking > agains libasn1 for libkafs5 completely on mips which is the source of > the problem in libkafs5. The issue is that two fields in the libkafs5 dynamic symbol table is un-initialized, whether the build crashes seem to depend on the value in fields, if you are lucky the field does not have values that crash the build. [ I'm still looking at this, here's the progress so far] The asn1 library has a export map containing 'global: *', this exports two symbols _fdata and _ftext versioned. When libkafs5 is linked, these symbols confuse the bfd code and the entries corresponding to theses (index 13, and 16) are left un-initialized. One workaround I see is to change the export 'global: *' in kerberos5/lib/libasn1 to the actual list of exported symbols. JC.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CA%2B7sy7BwFQxUaHJsom36xGJXO9rd71%2BuERFcxcR4dUGkbwgNjA>