public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Szabolcs Nagy <szabolcs.nagy@arm.com>
To: "H.J. Lu" <hjl.tools@gmail.com>
Cc: Carlos O'Donell <carlos@redhat.com>,
	GNU C Library <libc-alpha@sourceware.org>
Subject: Re: [PATCH 3/3] csu: Move static pie self relocation later [BZ #27072]
Date: Thu, 7 Jan 2021 13:02:18 +0000	[thread overview]
Message-ID: <20210107130217.GN2379@arm.com> (raw)
In-Reply-To: <CAMe9rOrdJ+nZ09BAotKwpyZxY07=iyg9GPd-XmniSphuO+7s7w@mail.gmail.com>

The 01/07/2021 04:51, H.J. Lu wrote:
> On Thu, Jan 7, 2021 at 4:46 AM Carlos O'Donell <carlos@redhat.com> wrote:
> >
> > On 1/7/21 7:36 AM, H.J. Lu via Libc-alpha wrote:
> > > On Thu, Jan 7, 2021 at 3:04 AM Szabolcs Nagy via Libc-alpha
> > > <libc-alpha@sourceware.org> wrote:
> > >>
> > >> On targets where hidden symbol access does not need RELATIVE
> > >> relocs, move the static pie self relocation after tunables and
> > >> cpu features are set up.  This allows processing IRELATIVE
> > >> relocs with correct ifunc dispatch logic.
> > >>
> > >> Unfortunately it is hard to guarantee that there will be no
> > >> dynamic relocations in the early start up code, so this is a
> > >> bit fragile. Ideally the RELATIVE relocs would be processed as
> > >> early as possible and IRELATIVE relocs after cpu features are
> > >> setup, but in glibc it is hard to separate them into two steps.
> > >
> > > x86 linker places IRELATIVE relocations the last:
> > >
> > > https://sourceware.org/bugzilla/show_bug.cgi?id=13302
> > >
> > > Does your linker have this bug fixed?
> >
> > Agreed, this is something I asked about during the design of
> > IFUNCs and I was told by Ulrich and others that IRELATIVE has
> > to be placed in a group and after RELATIVE relocs.
> 
> Not just before RELATIVE.   IRELATIVE should be processed the
> last before all other relocations.   See PR 13302 for other cases.

this is about static pie: sorting relocs does not help, the
problem is not that ifunc resolvers have some relocs in them,
but that the target specific relocation processing is one
monolithic switch in a loop that handles all relocs in one go,
there is no api to request only a subset of relocs to be
processed (either by type or ordering in the list of relocs).

in __libc_start_main we need to do

  do_relative_relocs();
  setup_auxv_tunables_cpu_etc();
  do_irelative_relocs();

which cannot be done without major changes in all targets and
generic elf reloc handling code. so the second best is

  setup_auxv_tunables_cpu_etc(); // avoid relative relocs
  do_all_relocs();

on targets where this can be done (other targets do
not support static pie).

  reply	other threads:[~2021-01-07 13:02 UTC|newest]

Thread overview: 21+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-01-07 11:00 [PATCH 0/3] fix ifunc with static pie " Szabolcs Nagy
2021-01-07 11:01 ` [PATCH 1/3] Make libc symbols hidden in static PIE Szabolcs Nagy
2021-01-07 11:01 ` [PATCH 2/3] [RFC] elf: hack up tunables to avoid RELATIVE relocs Szabolcs Nagy
2021-01-07 13:50   ` Siddhesh Poyarekar
2021-01-07 11:01 ` [PATCH 3/3] csu: Move static pie self relocation later [BZ #27072] Szabolcs Nagy
2021-01-07 12:36   ` H.J. Lu
2021-01-07 12:46     ` Carlos O'Donell
2021-01-07 12:51       ` H.J. Lu
2021-01-07 13:02         ` Szabolcs Nagy [this message]
2021-01-07 14:25           ` Szabolcs Nagy
2021-01-07 14:48             ` Siddhesh Poyarekar
2021-01-07 15:25               ` Szabolcs Nagy
2021-01-07 17:48                 ` H.J. Lu
2021-01-07 21:03   ` H.J. Lu
2021-01-08 10:08     ` Szabolcs Nagy
2021-01-07 12:48 ` [PATCH 0/3] fix ifunc with static pie " H.J. Lu
2021-01-07 12:50   ` Carlos O'Donell
2021-01-07 12:55     ` H.J. Lu
2021-01-07 13:03       ` Szabolcs Nagy
2021-01-07 13:15         ` H.J. Lu
2021-01-07 13:28           ` Szabolcs Nagy

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20210107130217.GN2379@arm.com \
    --to=szabolcs.nagy@arm.com \
    --cc=carlos@redhat.com \
    --cc=hjl.tools@gmail.com \
    --cc=libc-alpha@sourceware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).