There are two mechanisms via which M5 detects the OS: 1) the OSABI
field in the header of the ELF file (note that this field is NOT set
by the distributed compilers), or 2) the .note.ABI-tag section. One
could change the first with a hex editor, but I will describe a simple
solution for adding the latter onto an arbitrary executable.
I am assuming that the alpha-crosstool/bin is in your path. I also
assume that m5 is installed into ~/m5. You can fix either of these as
needed, of course.
All we need to do is extract a copy of the .note.ABI-tag section, so
we use objcopy. Fortunately, M5 ships with a working "hello, world",
so we will use that:
alpha-unknown-linux-gnu-objcopy -j .note.ABI-tag ~/m5/tests/test-
progs/hello/bin/alpha/linux/hello abi-tag.o
You can now link that into your program:
gcc -nostdlib nm.c abi-tag.o
Kick it off into M5, and you are greeted with a delightful endless
loop :P
- Clint
On Nov 13, 2008, at 12:07 AM, Shoaib Akram wrote:
> Thanks both. As you guys described, things work fine using x86 gcc
> on an x86 machine. But when I use crosscompiler and take the alpha
> binary to m5, i get, unknown/unsupported OS. Any idea?
>
> ---- Original message ----
>> Date: Wed, 12 Nov 2008 19:35:41 -0500
>> From: Clint Smullen <[EMAIL PROTECTED]>
>> Subject: Re: [m5-users] statictics
>> To: M5 users mailing list <[email protected]>
>>
>> Different platforms choose whether or not to automatically prefix
>> underscores and whether start or _start is the default symbol. On a
>> linux system, I was able to compile and link the following program:
>>
>> void _start() { }
>>
>> Using just:
>>
>> gcc -nostdlib nm.c
>>
>> The resulting program crashes, however, since it does not have what
>> is
>> typically contained within crt0. The important thing to keep in mind,
>> is that start() should never return. Adding a trivial loop proves
>> this, as with the following variant:
>>
>> void start() { while (1) { } }
>>
>> Since you are running it within M5, I would suggest putting a
>> trailing
>> m5_exit to keep it from running of into no-man's land.
>>
>> - Clint
>>
>> On Nov 12, 2008, at 7:18 PM, nathan binkert wrote:
>>
>>> You should be able to just create a function called _start. Does
>>> that
>>> not work? What does nm say when you do that?
>>>
>>> I think the thing you're missing here is that you need to tell the
>>> linker that you're using _start as the entry point for your program.
>>> The kernel doesn't actually look for _start (you should be able to
>>> call it whatever you want). It looks in the elf header for the
>>> address of an entry point. Because you've told it not to use the
>>> stdlib, I'm guessing that the linker might not add the entry point
>>> if
>>> you don't tell it which symbol it is.
>>>
>>> Good luck.
>>>
>>> Nate
>>>
>>> On Wed, Nov 12, 2008 at 4:12 PM, Shoaib Akram <[EMAIL PROTECTED]>
>>> wrote:
>>>> Kind of not related to m5, but do you know how do I define my own
>>>> _start symbol in a c program?
>>>>
>>>> ---- Original message ----
>>>>> Date: Tue, 11 Nov 2008 01:05:35 -0800
>>>>> From: Gabe Black <[EMAIL PROTECTED]>
>>>>> Subject: Re: [m5-users] statictics
>>>>> To: M5 users mailing list <[email protected]>
>>>>>
>>>>> I believe you use the -nostdlib option to leave out all of the
>>>>> standard
>>>>> library stuff, including _start which kicks off all the things Ali
>>>>> mentioned and more and then finally calls main(). In your program
>>>>> you'll
>>>>> then need to define _start yourself. If you're writing your
>>>>> program in
>>>>> assembly, then all you need to do is include a label called _start
>>>>> and
>>>>> put everything you want to run after it. If your program uses any
>>>>> system
>>>>> calls you'll have to implement stubs for those since those are
>>>>> part of
>>>>> the standard library too.
>>>>>
>>>>> Gabe
>>>>>
>>>>> Ali Saidi wrote:
>>>>>> No it's probably from all the libc _start code that is executed.
>>>>>> Even
>>>>>> if main() is a single line there is much more code that is
>>>>>> included in
>>>>>> the binary to setup the environment properly, get the arguments
>>>>>> in
>>>>>> order, see if the terminal is character or line buffered, etc.
>>>>>> You
>>>>>> should be able to convince the linker to not include all of
>>>>>> this -
>>>>>> fno-
>>>>>> builtin, but you'll need to define a _start symbol.
>>>>>>
>>>>>> Ali
>>>>>>
>>>>>> On Nov 10, 2008, at 11:21 PM, Shoaib Akram wrote:
>>>>>>
>>>>>>
>>>>>>> I wrote a simple program with 3 branches and ran it on m5. The
>>>>>>> statstics show much more branches and the number of instructions
>>>>>>> executed are also more than the assembly generated. Is it
>>>>>>> because of
>>>>>>> noise from emulated system calls?
>>>>>>> _______________________________________________
>>>>>>> m5-users mailing list
>>>>>>> [email protected]
>>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> m5-users mailing list
>>>>>> [email protected]
>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> m5-users mailing list
>>>>> [email protected]
>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>> _______________________________________________
>>>> m5-users mailing list
>>>> [email protected]
>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>
>>>>
>>> _______________________________________________
>>> m5-users mailing list
>>> [email protected]
>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>
>> _______________________________________________
>> m5-users mailing list
>> [email protected]
>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
