linux-kernel - Re: [PATCH v2] init/gcov: allow CONFIG_CONSTRUCTORS on UML to fix module gcov

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <2a46ca787df9a44c8b4fbc17ab6b69247ab38400.camel@sipsolutions.net>
Date:   Mon, 10 May 2021 13:37:53 +0200
From:   Johannes Berg <johannes@...solutions.net>
To:     Lambert <lambertdev@...com>,
        Peter Oberparleiter <oberpar@...ux.ibm.com>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        linux-arch@...r.kernel.org, linux-kernel@...r.kernel.org,
        Arnd Bergmann <arnd@...db.de>, Jessica Yu <jeyu@...nel.org>,
        linux-um@...ts.infradead.org, "Lambert." <lambertdev@...mail.com>
Subject: Re: [PATCH v2] init/gcov: allow CONFIG_CONSTRUCTORS on UML to fix
 module gcov

Hi,

> Hi Johannes and Peter, sorry to bother but I have one question 
> on this change. The do_ctors() won’t be executed for UML 
> because  *the constructors have already been called for ELF*. 
> 
> *__ctors_start*  and  *__ctors_end* symbols. See link:
> https://elixir.bootlin.com/linux/v5.12.2/source/include/asm-generic/vmlinux.lds.h#L676
> 
> In my environment, UML+GCC 10, I can't find __gcov_init executed 
> before kernel starts. So I did some trace and found glibc
> __libc_csu_init 
> will only execute constructors between *__init_array_start*and
> *__init_array_end*.  
> Which means if do_ctors() is not executed for UML, no elsewhere will 
> the constructors be executed.
> 
> Shall we remove the *!defined(CONFIG_UML)* for GCC, or I just missed 
> some steps to make the GCOV work for UML? 

No, that doesn't seem like the right solution.

Perhaps then with that toolchain (or configuration thereof) we need to
provide __init_array_start/end labels?

Or ... maybe that actually just needs to be removed, so that the
toolchain gets to choose?

Hmm. Pretty sure it worked for me, I think also with gcc 10, but not
sure exactly where I tested.

johannes