void added a comment. In D110869#3034616 <https://reviews.llvm.org/D110869#3034616>, @nickdesaulniers wrote:
> We'll probably need to investigate code gen a little. > > A mainline linux kernel defconfig built with `CONFIG_ZERO_CALL_USED_REGS=y` > enabled doesn't boot, for example. I consider that a blocker before landing > this (much-appreciated) feature; marking it as such. (Though it's possible > that there are TUs in the kernel that may need to be built with > `-fzero-call-used-regs=skip` that aren't (yet) failing with GCC). I don't > observe the kernel getting to `start_kernel`, which is the arch-agnostic > entry point of the boot; so there's likely x86 specific C code invoked before > the generic boot code takes over that may not play well with the > instrumentation as implemented. I found a likely issue. In `arch/x86/kernel/e820.c`, the function `cpcompare` returns a value, but we generate `xorq %rax, %rax` before returning. So not great. The issue is telling which registers are "live out" of the function (or exit block) so that we don't accidentally zero them out. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D110869/new/ https://reviews.llvm.org/D110869 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
