Common kernel image for e300 and e500v2

Common kernel image for e300 and e500v2

[Johns Daniel]

Is it possible -- and prudent -- to use a single kernel binary image
for two similar boards, one based on an e300 core and the other on an
e500v2 core?

I was surprised to see that the e500v2-targeted toolchain did build
the kernel for the e300 board just fine. Don't know whether this will
always be true?

The next logical step was to configure and customize the kernel just
right so that a single binary image will work on both boards. I am not
sure this is possible -- and, even if possible, whether this is
kosher!

Would love to hear the technical nuances involved with this...

-- Johns

Re: Common kernel image for e300 and e500v2

[Scott Wood]

Johns Daniel wrote:
> Is it possible -- and prudent -- to use a single kernel binary image
> for two similar boards, one based on an e300 core and the other on an
> e500v2 core?

No, it is not possible.  They use different MMUs, and the kernel does 
not support choosing between them at runtime.

> I was surprised to see that the e500v2-targeted toolchain did build
> the kernel for the e300 board just fine. Don't know whether this will
> always be true?

"always" is a strong word, but it should generally work.

-Scott

Re: Common kernel image for e300 and e500v2

[Johns Daniel]

[-- Attachment #1: Type: text/plain, Size: 981 bytes --]

Thank you for that piece of info, Scott!

May I expand the question one bit, and ask whether the user-mode binaries
compiled for the e300 will generally work for the e500 -- or, vice versa? Is
one choice safer than the other?

I know for sure that we take a big hit on floating-point ops, but are there
other things I am not considering?

-- Johns

On Fri, Jan 23, 2009 at 10:21 AM, Scott Wood <scottwood@freescale.com>wrote:

> Johns Daniel wrote:
>
>> Is it possible -- and prudent -- to use a single kernel binary image
>> for two similar boards, one based on an e300 core and the other on an
>> e500v2 core?
>>
>
> No, it is not possible.  They use different MMUs, and the kernel does not
> support choosing between them at runtime.
>
>  I was surprised to see that the e500v2-targeted toolchain did build
>> the kernel for the e300 board just fine. Don't know whether this will
>> always be true?
>>
>
> "always" is a strong word, but it should generally work.
>
> -Scott
>
>

[-- Attachment #2: Type: text/html, Size: 1679 bytes --]

Re: Common kernel image for e300 and e500v2

[Scott Wood]

Johns Daniel wrote:
> Thank you for that piece of info, Scott!
> 
> May I expand the question one bit, and ask whether the user-mode 
> binaries compiled for the e300 will generally work for the e500 -- or, 
> vice versa? Is one choice safer than the other?
> 
> I know for sure that we take a big hit on floating-point ops, but are 
> there other things I am not considering?

It should work, though as you note floating point will have to be 
emulated (make sure your kernel supports it); better would be to compile 
as soft-float.  Also avoid string instructions.

-Scott