[libc][RISCV] Add naked attribute to setjmp/longjmp

[ilovepi]

We want to avoid any possibility that the compiler will insert a
prologue/epilogue violating the calling contracts for these special
functions, potentially clobbering registers that must be preserved. To
do that they should be marked naked, as is already the case on ARM.

[llvmbot]

@llvm/pr-subscribers-libc

Author: Paul Kirth (ilovepi)

Changes

We want to avoid any possibility that the compiler will insert a
prologue/epilogue violating the calling contracts for these special
functions, potentially clobbering registers that must be preserved. To
do that they should be marked naked, as is already the case on ARM.

---

Full diff: https://github.com/llvm/llvm-project/pull/100036.diff

2 Files Affected:

• (modified) libc/src/setjmp/riscv/longjmp.cpp (+1)
• (modified) libc/src/setjmp/riscv/setjmp.cpp (+1)

diff --git a/libc/src/setjmp/riscv/longjmp.cpp b/libc/src/setjmp/riscv/longjmp.cpp
index 0f9537ccc4151..b14f636659ac3 100644
--- a/libc/src/setjmp/riscv/longjmp.cpp
+++ b/libc/src/setjmp/riscv/longjmp.cpp
@@ -30,6 +30,7 @@
 
 namespace LIBC_NAMESPACE_DECL {
 
+[[gnu::naked]]
 LLVM_LIBC_FUNCTION(void, longjmp, (__jmp_buf * buf, int val)) {
   LOAD(ra, buf->__pc);
   LOAD(s0, buf->__regs[0]);
diff --git a/libc/src/setjmp/riscv/setjmp.cpp b/libc/src/setjmp/riscv/setjmp.cpp
index 12def578b56f3..92982cc9d74d4 100644
--- a/libc/src/setjmp/riscv/setjmp.cpp
+++ b/libc/src/setjmp/riscv/setjmp.cpp
@@ -29,6 +29,7 @@
 
 namespace LIBC_NAMESPACE_DECL {
 
+[[gnu::naked]]
 LLVM_LIBC_FUNCTION(int, setjmp, (__jmp_buf * buf)) {
   STORE(ra, buf->__pc);
   STORE(s0, buf->__regs[0]);

[mikhailramalho]

LGTM but please wait for @michaelrj-google @lntue or @petrhosek to give the go-ahead.

There were some discussions about these functions in the past so better to double-check with them first.

[ilovepi]

LGTM but please wait for @michaelrj-google @lntue or @petrhosek to give the go-ahead.

There were some discussions about these functions in the past so better to double-check with them first.

Happy to wait, but naked is already on the ARM implementation, and AFAICT is required on any of these functions. TBH, I'd prefer these to be implemented in assembly files, so that we don't have to worry about it doing the right thing.

[lntue]

LGTM but please wait for @michaelrj-google @lntue or @petrhosek to give the go-ahead.
There were some discussions about these functions in the past so better to double-check with them first.

Happy to wait, but naked is already on the ARM implementation, and AFAICT is required on any of these functions. TBH, I'd prefer these to be implemented in assembly files, so that we don't have to worry about it doing the right thing.

From the previous discussion and experiment by @nickdesaulniers , naked attribute + inline asm seems to be the cleanest: #87837 (comment)