Date: Mon, 28 Dec 1998 09:45:02 -0500
From: "Kaleb S. KEITHLEY" <kaleb@ics.com>
To: hackers@FreeBSD.ORG
Subject: Re: ld (bfd): wrong function names for ELF shared library DT_{INIT,FINI}
Message-ID: <3687996E.63DECDAD@ics.com>
References: <Pine.BSF.4.01.9812271747560.383-100000@herring.nlsystems.com> <199812280523.VAA08366@vashon.polstra.com>
next in thread | previous in thread | raw e-mail | index | archive | help
John Polstra wrote: > > In article <36869BCF.2F1CF0FB@ics.com>, > Kaleb S. KEITHLEY <kaleb@ics.com> wrote: > > > > > > > > What is named ".init" and ".fini" are the **section** where these > > > > functions are placed in. This are not the function names. In fact, > > > > the function names are not specified at all. _init and _fini come > > > > from folklore (speek, the first ELF implementations). > > > > > > I haven't looked at the i386 code but the alpha _init() and _fini() are > > > already in the correct sections. > > > > On x86 ELF binutils/ld does not create .init or .fini sections. > > Nor does it need to. Nor should it. That's taken care of by the > various crt*.o modules in /usr/lib. That's not what's wanted. If a shared library is created with cc, which pulls in /usr/lib/crti.o, then we get the .init and .fini sections, plus no-op _init() and _fini() functions. That's bad if I've written my own _init() and _fini() functions then I get multiple definitions and the link fails. Those no-op functions don't even have returns. What happens if something tries to call them? Apparently the rtld is smart enough not to call them if they're zero length??? If I create the shared library with ld (and deliberately don't link crti.o), then I don't get .init and .fini sections. Right now it doesn't seem to hurt anything. Perhaps the _init() and _fini functions in crti.S need to be weak symbols. Otherwise if I want my own _init() and _fini() funcions and I'm being pedantic and want them in .init and .fini sections, I have to resort to compile-to-assembler, edit, and assemble in order to accomplish that. > > > In the past, given the lack of tools to tag things as belonging in > > the .init and .fini sections that I have kludged it with assembler > > to create .init() and .fini() functions, and they were called. Given > > that the .init and .fini sections are simply executable code, I > > think the distinction between being sections and being functions is > > too subtle to support your case. > > There is nothing subtle about the difference between sections and > functions. I did not say that the difference between sections and functions was subtle. I said the difference between __these__ two sections and functions was subtle. .init and .fini sections are __special__ sections. The spec says they point to executable code. A function label points to executable code too. As such the distinction between them seems subtle. And as I recall, circa Solaris 2.3, functions named .init() and .fini() in a shared library would be called. I don't have a 2.3 system to retry this, and it doesn't work on 2.4, so it's moot. -- Kaleb To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3687996E.63DECDAD>