This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: RFC: Add PT_GNU_PROPERTY to cover .note.gnu.property section
- From: Mark Wielaard <mark at klomp dot org>
- To: Cary Coutant <ccoutant at gmail dot com>
- Cc: "H.J. Lu" <hjl dot tools at gmail dot com>, Florian Weimer <fweimer at redhat dot com>, Binutils <binutils at sourceware dot org>, GNU C Library <libc-alpha at sourceware dot org>, gnu-gabi at sourceware dot org, x86-64-abi <x86-64-abi at googlegroups dot com>
- Date: Wed, 12 Dec 2018 11:44:49 +0100
- Subject: Re: RFC: Add PT_GNU_PROPERTY to cover .note.gnu.property section
- References: <87k1kyhbki.fsf@oldenburg.str.redhat.com> <CAMe9rOq6AgykBthApfHsFgbTHM9LjocDP-DNjB=Dht5uOagT7g@mail.gmail.com> <CAJimCsEDCNWBZsMEMeco0Vkj03iXML08KpLvAxC3DtyzXq8abA@mail.gmail.com> <CAMe9rOqHfqfbu_V1bPgykyN352btS8eZpRm-HAwJDfTf_Fiy-Q@mail.gmail.com> <CAMe9rOp7C_HS4cg-GcHH6NhT5oP-BxzE9emc6+Szx_-mAvxONw@mail.gmail.com> <CAMe9rOq4viqCYVXuVbyzTY0R0XVxzjc_tis9Uxw9OmgZRYcOjA@mail.gmail.com> <CAMe9rOp724pNQhndd8_-bVOS=42=0a4Ajsmgi2LUzA1P2BeWvw@mail.gmail.com> <87zhtcz7cg.fsf@oldenburg2.str.redhat.com> <20181211131933.GA9599@gmail.com> <CAJimCsGvEf5PF81n_-=M78UX81UY_stg_6_wMmYsM_kczOh3-w@mail.gmail.com>
On Tue, Dec 11, 2018 at 09:58:30PM -0800, Cary Coutant wrote:
> > > I would strongly recommended to try harder to get consensus here. I do
> > > not want to be a in a situation were we revise ABI again two years from
> > > now.
> >
> > That is my hope. PT_GNU_PROPERTY is our consensus so far.
>
> As you might expect, I support this new program header. Ideally, I'd
> have liked to replace the input SHT_NOTE sections with
> SHT_GNU_PROPERTY sections and dispense with all the note section
> overhead, but I'll take this as a compromise.
Why can't we switch to SHT_GNU_PROPERTY? My fear with combining
PT_GNU_PROPERTY with SHT_NOTE is that it will be even more confusing
for tools. You will get some allocated SHT_NOTEs in a PT_NOTE segment
and others in this new PT_GNU_PROPERTY segment (or worse, you get
multiple segments with different types covering the same ranges).
Also I thought there was still a question whether any or all
newly proposed property features and flags are actually needed
as loadable segments. There is a clear overlap with the GNU
Attributes (which are non-loadable). I would like to see consensus
first on the new property format/flags and which are and which
aren't needed as loadable properties at runtime.
Cheers,
Mark