Hi Vivek,
Sorry, I forgot to tell that we are using MPC8349E processor with a "e300 core". And it is clearly mentioned in the e300 core datasheet that, it supports "True little-endian mode for compatibility with other true little-endian devices".
Data sheets need to be read in the wider context of knowing how the devices are 'typically used'. Just because a core can operate in little-endian, doesn't mean you should :)
So, according to data sheet I have done following changes in U-Boot to configure it in true little endian mode
The convention for this processor, and pretty much all other PowerPCs is to operate in big-endian mode.
We want to configure MPC8349E in little endian as some other PCI devices(DSPs) are operating in little endian.
Any help/suggestion will be appreciated.
The PowerPC should be operated in big-endian mode, so that you have the wider support of the community. There is absolutely no reason to operate the processor in little-endian mode just because you want to talk to little-endian processors over the little-endian PCI bus.
The overhead of any required byte-swapping functions is negligible.
Regardless of what mode your PowerPC host operates versus your DSPs, you will have to use a communications protocol that enforces data alignment and format, eg. a C-structure on the DSP will more than likely not have the same layout as the same C-structure compiled on the PowerPC. There is the issue of endianness, but also the issue of alignment and packing. As such you would write an encode/serialization function to send data, and a decode/deserialization function to decode data.
There are standardized machine independent binary formats already defined, eg. XDR as used by NFS, and CDR as used by CORBA. Application code written using this technique is transparent to the endianness of the underlying processor.
I use little-endian x86 hosts to communicate with about 20 little-endian DSPs, or big-endian PowerPCs (MPC8349EAs). The devices communicate using CDR. I use the ACE C++ library and its CDR encode/decode functions on the x86 and PowerPCs, while on the DSPs, I use C-structures and pack/unpack them to match the equivalent CDR encoded functions. The overhead of data conversion is insignificant.
I hope that given this insight you will understand why you are recommended to operate the processor in its 'standard' big-endian mode.
Cheers, Dave