debug: Support big-endian architectures

This fixes some hard-coded assumptions in the debug crate that
the native ELF files being accessed are little-endian; specifically
in create_gdbjit_image as well as in emit_dwarf.

In addition, data in WebAssembly memory always uses little-endian
byte order.  Therefore, if the native architecture is big-endian,
all references to base types need to be marked as little-endian
using the DW_AT_endianity attribute, so that the debugger will
be able to correctly access them.

crates/debug/src/transform/unit.rs

@@ -10,6 +10,7 @@ use anyhow::{Context, Error};
 use gimli::write;
 use gimli::{AttributeValue, DebuggingInformationEntry, Unit};
 use std::collections::HashSet;
+use wasmtime_environ::ir::Endianness;
 use wasmtime_environ::isa::TargetIsa;
 use wasmtime_environ::wasm::DefinedFuncIndex;
 use wasmtime_environ::{CompiledFunctions, ModuleMemoryOffset};
@@ -463,6 +464,19 @@ where
             isa,
         )?;
 
+        // Data in WebAssembly memory always uses little-endian byte order.
+        // If the native architecture is big-endian, we need to mark all
+        // base types used to refer to WebAssembly memory as little-endian
+        // using the DW_AT_endianity attribute, so that the debugger will
+        // be able to correctly access them.
+        if entry.tag() == gimli::DW_TAG_base_type && isa.endianness() == Endianness::Big {
+            let current_scope = comp_unit.get_mut(die_id);
+            current_scope.set(
+                gimli::DW_AT_endianity,
+                write::AttributeValue::Endianity(gimli::DW_END_little),
+            );
+        }
+
         if entry.tag() == gimli::DW_TAG_subprogram && !current_scope_ranges.is_empty() {
             append_vmctx_info(
                 comp_unit,