Dear Joakim Tjernlund,
In message OF287114D8.013B6E51-ONC125781C.00667DCD-C125781C.00673260@transmode.se you wrote:
You don't need to make the MMU trick work on all boards, just your board (or cpu family), because it wouldn't be imposing anything on the rest of the system. Do we actually have someone who needs this feature on a board without a suitable MMU, large lockable cache or other SRAM, hardware bank switching, or other mechanism that can be confined to early low-level board/cpu-specific code?
Wolfgang seems to think so. As I read his reply he wants a solution for all
Please be aware that I don't think anything at all. I just comment :-)
I'm not in your position, where I am focussing on "how can I implement this". I'm looking at the code from the outside, and I ask what does it give to me, what of the things I'd like to have does it not give to me, and at which price does it come.
boards which I don't think is possible. I do think my approach comes closest though. I did try BATs but I didn't get very far.
If we are talking about _all_ boards we have to keep a wider view. For example, on MPC8xx there are not BATs. Not to mention other architectures.
Actually I was not asking for support for all boards, not even for all boards of a specific architecture, or a specific CPU. You submitted this patch, and I learned that the code, as is, is only useful for a single board, which appears to be maintained in an out-of-tree port. For all in-tree boards the code, as is, is broken and unusable without further rework. Needless to repeat that it is completely untested for mainline.
So we have a patch that promises a feature, which cannot be used by any of the mainline boards, but it makes the code more complex for zero benefit.
It's a nice and appreciated RFC patch or even example implementation, but I fail to see arguments why we should add this to mainline.
Best regards,
Wolfgang Denk