llvm dev - Apr 2019 - Symbols in 'llvm.used' stripped by --gc-sections

Hi all,
  The documentation of `@llvm.used` says:

"If a symbol appears in the @llvm.used list, then the compiler, assembler,
and linker are required to treat the symbol as if there is a reference to
the symbol that it cannot see (which is why they have to be named)."

We've always understood this as: the symbol will survive into the final
executable, also when it is in its own section and with `--gc-sections`.
But I think that's not true.
My understanding now is that `@llvm.used` is completely ignored by the
linker when creating a final whole-program binary. That is, the symbol is
emitted by the compiler into the object file, so not stripped e.g. by DCE.
But when the linker combines multiple object files into the final
executable, the symbol will be removed by `--gc-sections` when the linker
sees no reference to it.
It appears that the only way to keep an unreferenced symbol while linking
with `--gc-sections` (compiling with `-f{function,data}-sections`) is to
put it in a segment that is kept as specified by the linkerscript (`KEEP`).

I hope someone can tell me that my current understanding is correct, and
that we have to use another way to keep a symbol until the very end (tips
welcome, without modifying the system default linker script...).

Thanks!
  Johan
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190401/20f6b89c/attachment.html>

Unfortunately the linker part of @llvm.used is not implemented for ELF. One
way that it could be implemented is to emit a zero-length .init_array
section with an R_*_NONE relocation pointing to the symbol.

If you just want a quick workaround, you could use the -u linker flag to
keep the symbol alive.

Peter

On Mon, Apr 1, 2019 at 10:12 AM Johan Engelen via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hi all,
>   The documentation of `@llvm.used` says:
>
> "If a symbol appears in the @llvm.used list, then the compiler,
> assembler, and linker are required to treat the symbol as if there is a
> reference to the symbol that it cannot see (which is why they have to be
> named)."
>
> We've always understood this as: the symbol will survive into the final
> executable, also when it is in its own section and with `--gc-sections`.
> But I think that's not true.
> My understanding now is that `@llvm.used` is completely ignored by the
> linker when creating a final whole-program binary. That is, the symbol is
> emitted by the compiler into the object file, so not stripped e.g. by DCE.
> But when the linker combines multiple object files into the final
> executable, the symbol will be removed by `--gc-sections` when the linker
> sees no reference to it.
> It appears that the only way to keep an unreferenced symbol while linking
> with `--gc-sections` (compiling with `-f{function,data}-sections`) is to
> put it in a segment that is kept as specified by the linkerscript (`KEEP`).
>
> I hope someone can tell me that my current understanding is correct, and
> that we have to use another way to keep a symbol until the very end (tips
> welcome, without modifying the system default linker script...).
>
> Thanks!
>   Johan
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>

-- 
-- 
Peter
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190401/897defa9/attachment.html>

The MachO implementation of the llvm.used list attaches a NoDeadStrip flag to
each of the symbols on the llvm.used list, the MachO linker is aware of the
NoDeadStrip flag and will retain symbols that are marked with the flag in the
linked output file.

As noted below, the LLVM back end for ELF does not currently communicate
anything to the object file about retaining a symbol in the linked output.

The armclang compiler generates a special tag symbol/label along with the
definition of the function or object that it wants to retain. Its linker knows
to scan for these special symbols and “retain” the sections that they are
defined in.

One could also envision defining a special symbol to communicate additional
information about a symbol in the object file. Again, the linker would need to
comprehend the intended semantics of the special symbol and whatever value is
attached to it to represent the additional information about a symbol.

~ Todd Snider

From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of Peter
Collingbourne via llvm-dev
Sent: Monday, April 1, 2019 12:19 PM
To: Johan Engelen
Cc: llvm-dev
Subject: [EXTERNAL] Re: [llvm-dev] Symbols in 'llvm.used' stripped by
--gc-sections

Unfortunately the linker part of @llvm.used is not implemented for ELF. One way
that it could be implemented is to emit a zero-length .init_array section with
an R_*_NONE relocation pointing to the symbol.

If you just want a quick workaround, you could use the -u linker flag to keep
the symbol alive.

Peter

On Mon, Apr 1, 2019 at 10:12 AM Johan Engelen via llvm-dev <llvm-dev at
lists.llvm.org<mailto:llvm-dev at lists.llvm.org>> wrote:
Hi all,
  The documentation of `@llvm.used` says:

"If a symbol appears in the @llvm.used list, then the compiler, assembler,
and linker are required to treat the symbol as if there is a reference to the
symbol that it cannot see (which is why they have to be named)."

We've always understood this as: the symbol will survive into the final
executable, also when it is in its own section and with `--gc-sections`. But I
think that's not true.
My understanding now is that `@llvm.used` is completely ignored by the linker
when creating a final whole-program binary. That is, the symbol is emitted by
the compiler into the object file, so not stripped e.g. by DCE. But when the
linker combines multiple object files into the final executable, the symbol will
be removed by `--gc-sections` when the linker sees no reference to it.
It appears that the only way to keep an unreferenced symbol while linking with
`--gc-sections` (compiling with `-f{function,data}-sections`) is to put it in a
segment that is kept as specified by the linkerscript (`KEEP`).

I hope someone can tell me that my current understanding is correct, and that we
have to use another way to keep a symbol until the very end (tips welcome,
without modifying the system default linker script...).

Thanks!
  Johan

_______________________________________________
LLVM Developers mailing list
llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev


--
--
Peter
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190401/0c70146f/attachment.html>