Re: mainthread.cxx - unaligned stack possible

[Gary Thomas <gary at mlbassoc dot com>]

On Sun, 2003-03-09 at 10:42, Larice Robert wrote:
> > Scott Furman wrote:
> > > 
> > > 
> > > eCos provides the stack for main() when building with libc startup 
> > > support and CYGSEM_LIBC_MAIN_STACK_FROM_SYSTEM is defined.  However, the 
> > > stack is declared as an array of  cyg_uint8, so there is the possibility 
> > > that it is unaligned.
> > 
> > Hmm.. I thought the linker was meant to do this for statics/globals. Oh 
> > well, let's err on the side of caution...
> > 
> >  > The attached trivial fix ensures that the stack
> > > is at least octlet-aligned.  (Does any architecture require an alignment 
> > > stricter than 8 bytes ?)
> > 
> > Some can. I've checked in a slightly better version (attached).
> > 
> > Jifl
> 
> Hello Jonathan Larmour,
> 
>   obviously the linker does not what you thought.
>   i've a small test case here:
> 
> 

But, you didn't add the "alignment magic" that Jonathan did.
Try it like this:

#define CYGBLD_ATTRIB_ALIGN(__align__) __attribute__((aligned(__align__)))

char obj0[3][5] CYGBLD_ATTRIB_ALIGN(8);
char obj1[3][5] CYGBLD_ATTRIB_ALIGN(16);

etc.

> file alignment.c :
> 
> char   obj0[3][5];
> char   obj1[3][5];
> char   obj2[3][5];
> char   obj3[3][5];
> 
> main() {;}
> 
> 
> file Makefile :
> 
> prefix=/u2/larice/tools-sh-elf-v2/bin/sh-elf-
> 
> all :  
>         $(prefix)gcc -ml -m3 -o alignment alignment.c
>         $(prefix)nm alignment | grep obj
> 
> 
> produces the following output :
> 
> 00002b37 B _obj0
> 00002b46 B _obj1
> 00002b28 B _obj2
> 00002b55 B _obj3
> 
> obviously obj0 and obj3 are NOT 32 bit aligned.
> 
> 
> several month ago i've sent a patch to fix this for 200 or so global
> arrays occuring in the ecos source code. 
>   (http://sources.redhat.com/ml/ecos-patches/2002-06/msg00021.html)
> all of them 2 dimensional char arrays, which are used for
> stack purpose or for other things requiring some alignment.
> 
> would you please consider to apply this patch, or something else to
> fix this ? this is a really uggly problem of ecos and would well be
> repaired in the upcomming 2.0 .
> 
> Thank you,
>   Robert Larice

-- 
------------------------------------------------------------
Gary Thomas                 |
MLB Associates              |  Consulting for the
+1 (970) 229-1963           |    Embedded world
http://www.mlbassoc.com/    |
email: <gary at mlbassoc dot com>  |
gpg: http://www.chez-thomas.org/gary/gpg_key.asc
------------------------------------------------------------