linux-kernel - [PATCH] sparse LLVM backend (was Re: [rfc] built-in native compiler for Linux?)

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: <49F27032.7050703@garzik.org>
Date:	Fri, 24 Apr 2009 22:06:42 -0400
From:	Jeff Garzik <jeff@...zik.org>
To:	LKML <linux-kernel@...r.kernel.org>, linux-sparse@...r.kernel.org
CC:	Ingo Molnar <mingo@...e.hu>, Al Viro <viro@...IV.linux.org.uk>,
	Linus Torvalds <torvalds@...ux-foundation.org>
Subject: [PATCH] sparse LLVM backend (was Re: [rfc] built-in native compiler
 for Linux?)

(resent -- the mail monster apparently ate the first one)

Just to make this crazy thread even crazier, I have successfully gotten 
sparse to create a valid x86-64 object file from C source code (...which 
I then executed).

The attached patch adds a code generator backend to sparse, that 
generates LLVM bitcode.  This bitcode can then be converted to ELF .o 
files using standard tools -- see instructions in the header of s2l.c.

Do I expect this to be used in the kernel?  Not really.  I'm just doing 
this for fun, to add a compiler backend to sparse.  It was pretty easy 
to get integer and logical operations going.

The work is based on compile-i386; as it turns out, LLVM's bitcode can 
be made type-aware, so I wanted to see how things would turn out 
generating code at a higher level.  Given the higher-level constructs, I 
didn't feel the need to bother with linearized form.

Caveats...

- don't stray too far from simple, integer types
- bitfields caused explosions, but you knew that anyway

Cheers all,

     Jeff

View attachment "patch" of type "text/plain" (51365 bytes)