On 3/20/11, arden jay arden.jay@gmail.com wrote:
Hi Michael,
I still have question. :)
When ARM fetch instruction, it firstly try cache. It then should have cache miss, and forces to reload the instruction from memory?
Why it will have problem while U-boot & Kernel at the same memory location?
If the instruction fetch misses the cache, then it works fine. That's why invalidating solves the problem. The problem is that it hits, but returns the wrong instructions. The code from U-Boot is loaded into the icache when U-Boot is started (before it is relocated) and some of it is still in the icache when we execute linux. So the CPU ends up fetching the wrong instructions.
On ARM if code is stored to RAM, then the modified addresses must be manually invalidated in the instruction cache. The hardware does not do this for us.
Michael