mips uasm代码有什么优势

Question

In the kernel's TLB processing part of the mips architecture, there are some uasm code, which is another way to generate assembly instead of using gcc.

I don't know why not write inline assembly instead of uasm. Why? What are the advantages of uasm?

Answer 1

In a TLB miss handler, I think you need total control over exactly what instructions are used. (To make sure you only access memory through known-present TLB entries). Using inline asm for just a few special instructions, and letting GCC spill local variables to the stack (if compiled without optimization), wouldn't work.

The TLB-miss handler entry point is also not a function that should return with jr $ra .

So your best bet is to write the whole handler in pure asm instead of trying to make sure the compiler emits a "safe" implementation.

Using an asm("") statement at global scope inside a C source file would work but makes the asm more ugly, and makes it harder for asm to take advantage of the C preprocessor.