ARM64 backend, part 9 / 11: wasmtime support.

This commit adds a few odds and ends required to build wasmtime on ARM64
with the new backend. In particular, it adds:

- Support for the `Arm64Call` relocation type.
- Support for fetching the trap PC when a signal is received.
- A hook for `SIGTRAP`, which is sent by the `brk` opcode (in contrast to
  x86's `SIGILL`).

With the patch sequence up to and including this patch applied,
`wasmtime` can now compile and successfully execute code on arm64. Not
all tests pass yet, but basic Wasm/WASI tests work correctly.

crates/jit/src/link.rs

@@ -2,7 +2,7 @@
 
 use crate::Compilation;
 use cranelift_codegen::binemit::Reloc;
-use std::ptr::write_unaligned;
+use std::ptr::{read_unaligned, write_unaligned};
 use wasmtime_environ::{Module, Relocation, RelocationTarget};
 use wasmtime_runtime::libcalls;
 use wasmtime_runtime::VMFunctionBody;
@@ -101,6 +101,23 @@ fn apply_reloc(
         Reloc::X86PCRelRodata4 => {
             // ignore
         }
+        Reloc::Arm64Call => unsafe {
+            let reloc_address = body.add(r.offset as usize) as usize;
+            let reloc_addend = r.addend as isize;
+            let reloc_delta = (target_func_address as u64).wrapping_sub(reloc_address as u64);
+            // TODO: come up with a PLT-like solution for longer calls. We can't extend the
+            // code segment at this point, but we could conservatively allocate space at the
+            // end of the function during codegen, a fixed amount per call, to allow for
+            // potential branch islands.
+            assert!((reloc_delta as i64) < (1 << 27));
+            assert!((reloc_delta as i64) >= -(1 << 27));
+            let reloc_delta = reloc_delta as u32;
+            let reloc_delta = reloc_delta.wrapping_add(reloc_addend as u32);
+            let delta_bits = reloc_delta >> 2;
+            let insn = read_unaligned(reloc_address as *const u32);
+            let new_insn = (insn & 0xfc00_0000) | (delta_bits & 0x03ff_ffff);
+            write_unaligned(reloc_address as *mut u32, new_insn);
+        },
         _ => panic!("unsupported reloc kind"),
     }
 }
@@ -108,13 +125,10 @@ fn apply_reloc(
 // A declaration for the stack probe function in Rust's standard library, for
 // catching callstack overflow.
 cfg_if::cfg_if! {
-    if #[cfg(any(
+    if #[cfg(all(
-        target_arch="aarch64",
-        all(
             target_os = "windows",
             target_env = "msvc",
             target_pointer_width = "64"
-        )
             ))] {
         extern "C" {
             pub fn __chkstk();
@@ -132,6 +146,8 @@ cfg_if::cfg_if! {
         extern "C" {
             pub fn __rust_probestack();
         }
-        static PROBESTACK: unsafe extern "C" fn() = __rust_probestack;
+        static PROBESTACK: unsafe extern "C" fn() = empty_probestack;
     }
 }
+
+extern "C" fn empty_probestack() {}

crates/runtime/src/helpers.c

@@ -26,3 +26,12 @@ void* GetPcFromUContext(ucontext_t *cx) {
   return (void*) cx->uc_mcontext->__ss.__rip;
 }
 #endif
+
+#if defined(__linux__) && defined(__aarch64__)
+#include <sys/ucontext.h>
+
+void* GetPcFromUContext(ucontext_t *cx) {
+    return (void*) cx->uc_mcontext.pc;
+}
+
+#endif  // __linux__ && __aarch64__

crates/runtime/src/traphandlers.rs

@@ -31,6 +31,7 @@ cfg_if::cfg_if! {
         static mut PREV_SIGBUS: MaybeUninit<libc::sigaction> = MaybeUninit::uninit();
         static mut PREV_SIGILL: MaybeUninit<libc::sigaction> = MaybeUninit::uninit();
         static mut PREV_SIGFPE: MaybeUninit<libc::sigaction> = MaybeUninit::uninit();
+        static mut PREV_SIGTRAP: MaybeUninit<libc::sigaction> = MaybeUninit::uninit();
 
         unsafe fn platform_init() {
             let register = |slot: &mut MaybeUninit<libc::sigaction>, signal: i32| {
@@ -70,6 +71,9 @@ cfg_if::cfg_if! {
                 register(&mut PREV_SIGFPE, libc::SIGFPE);
             }
 
+            // on ARM64, we use `brk` to report traps, which generates SIGTRAP.
+            register(&mut PREV_SIGTRAP, libc::SIGTRAP);
+
             // On ARM, handle Unaligned Accesses.
             // On Darwin, guard page accesses are raised as SIGBUS.
             if cfg!(target_arch = "arm") || cfg!(target_os = "macos") {
@@ -87,6 +91,7 @@ cfg_if::cfg_if! {
                 libc::SIGBUS => &PREV_SIGBUS,
                 libc::SIGFPE => &PREV_SIGFPE,
                 libc::SIGILL => &PREV_SIGILL,
+                libc::SIGTRAP => &PREV_SIGTRAP,
                 _ => panic!("unknown signal: {}", signum),
             };
             let handled = tls::with(|info| {
@@ -158,6 +163,12 @@ cfg_if::cfg_if! {
                 if #[cfg(all(target_os = "linux", target_arch = "x86_64"))] {
                     let cx = &*(cx as *const libc::ucontext_t);
                     cx.uc_mcontext.gregs[libc::REG_RIP as usize] as *const u8
+                } else if #[cfg(all(target_os = "linux", target_arch = "aarch64"))] {
+                    // libc doesn't seem to support Linux/aarch64 at the moment?
+                    extern "C" {
+                        fn GetPcFromUContext(cx: *mut libc::c_void) -> *const u8;
+                    }
+                    GetPcFromUContext(cx)
                 } else if #[cfg(target_os = "macos")] {
                     // FIXME(rust-lang/libc#1702) - once that lands and is
                     // released we should inline the definition here

tests/custom_signal_handler.rs

@@ -122,7 +122,7 @@ mod tests {
                 .downcast::<Trap>()?;
             assert!(
                 trap.message()
-                    .starts_with("wasm trap: out of bounds memory access"),
+                    .starts_with("wasm trap: out of bounds"),
                 "bad trap message: {:?}",
                 trap.message()
             );
@@ -149,7 +149,7 @@ mod tests {
                 .downcast::<Trap>()?;
             assert!(trap
                 .message()
-                .starts_with("wasm trap: out of bounds memory access"));
+                .starts_with("wasm trap: out of bounds"));
         }
         Ok(())
     }