For panic=unwind on Wasm targets, define __cpp_exception tag

Since llvm/llvm-project 159143, llvm no longer weak links the __cpp_exception tag into
each object that uses it. They are now defined in compiler-rt. Rust doesn't seem to
get them from compiler-rt so llvm decides they need to be imported. This adds them to
libunwind.

Author: hoodmane

library/unwind/src/lib.rs

@@ -7,7 +7,7 @@
 #![cfg_attr(not(target_env = "msvc"), feature(libc))]
 #![cfg_attr(
     all(target_family = "wasm", any(not(target_os = "emscripten"), emscripten_wasm_eh)),
-    feature(link_llvm_intrinsics, simd_wasm64)
+    feature(link_llvm_intrinsics, simd_wasm64, asm_experimental_arch)
 )]
 #![allow(internal_features)]
 #![deny(unsafe_op_in_unsafe_fn)]

library/unwind/src/wasm.rs

@@ -2,6 +2,30 @@
 
 #![allow(nonstandard_style)]
 
+// Define the __cpp_exception tag that LLVM's wasm exception handling requires.
+// In particular it is required to use either of:
+//  1. the wasm_throw llvm intrinsic, or
+//  2. the Rust try intrinsic.
+//
+// This must be provided since LLVM commit
+// aee99e8015daa9f53ab1fd4e5b24cc4c694bdc4a which changed the tag from being
+// weakly defined in each object file to being an external reference that must
+// be linked from somewhere.
+//
+// We only define this for wasm32-unknown-unknown because on Emscripten/WASI
+// targets, this symbol should be defined by the external toolchain. In
+// particular, defining this on Emscripten would break Emscripten dynamic
+// libraries.
+#[cfg(all(target_os = "unknown", panic = "unwind"))]
+core::arch::global_asm!(
+    ".globl __cpp_exception",
+    #[cfg(target_pointer_width = "64")]
+    ".tagtype __cpp_exception i64",
+    #[cfg(target_pointer_width = "32")]
+    ".tagtype __cpp_exception i32",
+    "__cpp_exception:",
+);
+
 #[repr(C)]
 #[derive(Debug, Copy, Clone, PartialEq)]
 pub enum _Unwind_Reason_Code {

src/doc/rustc/src/platform-support/wasm32-unknown-unknown.md

@@ -267,3 +267,18 @@ the meantime using `-Cpanic=unwind` will require using [`-Zbuild-std`] and
 passing the appropriate flags to rustc.
 
 [`-Zbuild-std`]: ../../cargo/reference/unstable.html#build-std
+
+### The exception tag for panics
+
+Rust panics are currently implemented as a specific class of C++ exceptions.
+This is because llvm only supports throwing and catching the C++ exception tag
+from `wasm_throw` intrinsic and the lowering for the catchpads emitted by the
+Rust try intrinsic.
+
+In particular, llvm throw and catch blocks expect a `WebAssembly.Tag` symbol
+called `__cpp_exception`. If it is not defined somewhere, llvm will generate an
+Emscripten style import from `env.__cpp_exception`. We don't want this, so we
+define the symbol in `libunwind` but only for wasm32-unknown-unknown. WASI
+doesn't currently support unwinding at all, and the Emscripten linker provides
+the tag in an appropriate manner depending on what sort of binary is being
+linked.