Re: [PATCH] powerpc: 405/440/464/476 support and optimizations

["Ryan S. Arnold" <ryan.arnold@gmail.com>]

On Fri, Jan 20, 2012 at 8:24 AM, acrux <acrux_it@libero.it> wrote:
> On Thu, 19 Jan 2012 14:34:50 -0500
> "Carlos O'Donell" <carlos@systemhalted.org> wrote:
>
>> On Wed, Jan 18, 2012 at 3:30 PM, acrux@cruxppc.org
>> <acrux@linuxmail.org> wrote:
>> >
>> > just tried to build glibc-2.13 "--with-cpu=440 --with-fp" on a
>> > Sam440ep[1] (PPC440EP SoC [2]) but it remains stuck in this point:
>> > CPP='gcc -m32 -E -x c-header'
>> >  /home/999/new/work/src/build32/elf/ld.so.1
>> > --library-path
>> > /home/999/new/work/src/build32:/home/999/new/work/src/build32/math:/home/999/new/work/src/build32/elf:/home/999/new/work/src/build32/dlfcn:/home/999/new/work/src/build32/nss:/home/999/new/work/src/build32/nis:/home/999/new/work/src/build32/rt:/home/999/new/work/src/build32/resolv:/home/999/new/work/src/build32/crypt:/home/999/new/work/src/build32/nptl
>> > /home/999/new/work/src/build32/sunrpc/rpcgen -Y ../scripts -c
>> > rpcsvc/bootparam_prot.x -o
>> > /home/999/new/work/src/build32/sunrpc/xbootparam_prot.T
>>
>> This is the first use of the newly build dynamic loader.
>>
>> A failure here means that the dynamic loader has not been correctly
>> compiled.
>>
>> You should debug this to figure out what is going wrong in the loader.
>>
>
> i know, but i've no resource and i guess not enough skill to debug and
> fix it.
>
> As i received a borda with a 440EP SoC, just only for fun, i
> tested a build "--with-cpu=440 --with-fp".
> I'd like to know if somobody really tested these features.
>
>> > Here my config.log: http://cruxppc.org/~acrux/config.log
>> > Instead i successfully built glibc-2.13 without "--with-cpu=440
>> > --with-fp" . I'm using an updated CRUX PPC 2.7 (32bit): gcc-4.5.3,
>> > binutils-2.21.1, glibc-2.12.2
>>
>> Have you tested your compiler? What were the test results?
>>
>> Did you test binutils? What were the test results?
>>
>
> they finished their testsuites with only the well know failures for
> their own releases. Thus i can say they are good.

Integration can often expose problems, but I suspect something else is
going on here...

Does the machine you're building this toolchain on understand the
440fp instruction set?  If not then the loader is likely encountering
a sigill.

The solution to this situation is to enable cross compiling prior to
running configure.  This will prevent the build from attempting to run
any of the code.

echo "cross-compiling=yes" >> configparms

If that's not the problem I would suggest trying to build using 440 without fp.

If that works then I suspect that the problem is related to the string
routine optimizations that one of my guys put in for the 476
processor.  It was requested that we provide it to the entire 4xx
series since the instructions used (allegedly) weren't unique to the
476.

It'd be interesting to run a debugger against the loader at this point
and identify whether you're encountering a sigill or a sigsegv.

Ryan S. Arnold