Re: PATCH: Don't align text/data/bss sections in x86 assembler

["Jan Beulich" <JBeulich at suse dot com>]

>>> On 03.06.13 at 17:27, "H.J. Lu" <hjl.tools@gmail.com> wrote:
> On Mon, Jun 3, 2013 at 1:25 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>>> On 31.05.13 at 19:32, "H.J. Lu" <hongjiu.lu@intel.com> wrote:
>>> There is no need to align text/data/bss sections for ELF in x86
>>> assembler.  The alignment directive should be used if a symbol needs
>>> a specfic alignment.  I checked in this patch to avoid align
>>> text/data/bss sections in x86 assembler.
>>
>> Are you reasonably sure that this will not result in regressions?
>> While conceptionally I agree with the change (and never
>> understood why the alignment was being enforced), I think such
>> a change should only be done by providing a command line option
>> to suppress the automatic alignment.
>>
> 
> I thought about it and decided against it.  The section
> alignment is very arbitrary:
> 
> 1. Many arches don't do it.
> 2. Only x86 ELF did it.
> 3. It only aligned text/data/bss sections and there are many
> other sections with code and data which aren't aligned.
> 4. Neither gABI nor psABI require it.

These are all valid points, but neglect the (from my pov) most
important aspect here: Assembly code is bound to the arch (i.e.
point 1 isn't really meaningful), often also bound to the object
format (i.e. point 2 isn't very meaningful either). And people
doing simple things with as little code as possible are likely to be
at risk the most (largely voiding 3 and 4): Having got used to
simply write ".data" or ".bss" followed by simple data items would
automatically take care of minimal alignment, I would bet many
aren't even aware of the need to put in some form of ".align".
For text this likely is less of an issue, but you never know.

Jan