Return-Path: Received: from ?82.130.80.49? (hg-public-dock-49-dhcp.ethz.ch [82.130.80.49]) by mx.google.com with ESMTPS id 13sm998483fks.27.2009.03.12.06.21.43 (version=SSLv3 cipher=RC4-MD5); Thu, 12 Mar 2009 06:21:44 -0700 (PDT) Message-ID: <49B90C69.60703@gmail.com> Date: Thu, 12 Mar 2009 14:21:45 +0100 From: phcoder User-Agent: Thunderbird 2.0.0.19 (X11/20090105) MIME-Version: 1.0 To: David Miller Subject: Re: ELF bugfixes References: <49B8C686.5060508@gmail.com> <20090312.020757.187232522.davem@davemloft.net> <49B8F067.2040503@gmail.com> <20090312.055819.95768237.davem@davemloft.net> In-Reply-To: <20090312.055819.95768237.davem@davemloft.net> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit David Miller wrote: > From: phcoder > Date: Thu, 12 Mar 2009 12:22:15 +0100 > >> David Miller wrote: >>> From: phcoder >>> Date: Thu, 12 Mar 2009 09:23:34 +0100 >>> >>>> Index: include/grub/elf.h >>>> =================================================================== >>>> --- include/grub/elf.h (revision 2036) >>>> +++ include/grub/elf.h (working copy) >>>> @@ -77,7 +77,7 @@ >>>> Elf32_Half e_shentsize; /* Section header table entry size */ >>>> Elf32_Half e_shnum; /* Section header table entry count */ >>>> Elf32_Half e_shstrndx; /* Section header string table index */ >>>> -} Elf32_Ehdr; >>>> +} __attribute__ ((packed)) Elf32_Ehdr; >>>> >>> There is no reason why you should need the packed attribute here. >>> I can't think of any cpu where this could even remotely be necessary. >>> And if it's not necessary, all it does it emit terribly suboptimal >>> code on RISC cpus. >> If it doesn't make a difference in the structure why does the compiler generate different code? > > Because packed also means that the alignment of the structure > can't be assumed to be naturally word aligned. So the compiler > has to use byte loads and stores to load a word, for example, > on RISC machines. > > This is because the packed attribute influences how the structure is > embedded into other structures. > > This attribute controls two different things, not just local packing. I knew it but normally when you parse files normally offsets aren't guaranteed to be aligned. But now it seems that elf parser is written in a way to guarantee at least some alignments. Then this part of patch probably is to be dropped or changed to proper aligned attribute -- Regards Vladimir 'phcoder' Serbinenko