Add a compilation section to disable address maps (#3598)

* Add a compilation section to disable address maps

This commit adds a new `Config::generate_address_map` compilation
setting which is used to disable emission of the `.wasmtime.addrmap`
section of compiled artifacts. This section is currently around the size
of the entire `.text` section itself unfortunately and for size reasons
may wish to be omitted. Functionality-wise all that is lost is knowing
the precise wasm module offset address of a faulting instruction or in a
backtrace of instructions. This also means that if the module has DWARF
debugging information available with it Wasmtime isn't able to produce a
filename and line number in the backtrace.

This option remains enabled by default. This option may not be needed in
the future with #3547 perhaps, but in the meantime it seems reasonable
enough to support a configuration mode where the section is entirely
omitted if the smallest module possible is desired.

* Fix some CI issues

* Update tests/all/traps.rs

Co-authored-by: Nick Fitzgerald <fitzgen@gmail.com>

* Do less work in compilation for address maps

But only when disabled

Co-authored-by: Nick Fitzgerald <fitzgen@gmail.com>

crates/c-api/src/trap.rs

@@ -157,7 +157,9 @@ pub extern "C" fn wasmtime_frame_module_name(frame: &wasm_frame_t) -> Option<&wa
 
 #[no_mangle]
 pub extern "C" fn wasm_frame_func_offset(frame: &wasm_frame_t) -> usize {
-    frame.trap.trace()[frame.idx].func_offset()
+    frame.trap.trace()[frame.idx]
+        .func_offset()
+        .unwrap_or(usize::MAX)
 }
 
 #[no_mangle]
@@ -167,7 +169,9 @@ pub extern "C" fn wasm_frame_instance(_arg1: *const wasm_frame_t) -> *mut wasm_i
 
 #[no_mangle]
 pub extern "C" fn wasm_frame_module_offset(frame: &wasm_frame_t) -> usize {
-    frame.trap.trace()[frame.idx].module_offset()
+    frame.trap.trace()[frame.idx]
+        .module_offset()
+        .unwrap_or(usize::MAX)
 }
 
 #[no_mangle]

crates/cranelift/src/compiler.rs

@@ -63,6 +63,7 @@ impl Compiler {
         context: &Context,
         data: &FunctionBodyData<'_>,
         body_len: u32,
+        tunables: &Tunables,
     ) -> FunctionAddressMap {
         // Generate artificial srcloc for function start/end to identify boundary
         // within module.
@@ -75,17 +76,21 @@ impl Compiler {
 
         // New-style backend: we have a `MachCompileResult` that will give us `MachSrcLoc` mapping
         // tuples.
-        let instructions = collect_address_maps(
-            body_len,
-            context
-                .mach_compile_result
-                .as_ref()
-                .unwrap()
-                .buffer
-                .get_srclocs_sorted()
-                .into_iter()
-                .map(|&MachSrcLoc { start, end, loc }| (loc, start, (end - start))),
-        );
+        let instructions = if tunables.generate_address_map {
+            collect_address_maps(
+                body_len,
+                context
+                    .mach_compile_result
+                    .as_ref()
+                    .unwrap()
+                    .buffer
+                    .get_srclocs_sorted()
+                    .into_iter()
+                    .map(|&MachSrcLoc { start, end, loc }| (loc, start, (end - start))),
+            )
+        } else {
+            Vec::new()
+        };
 
         FunctionAddressMap {
             instructions: instructions.into(),
@@ -179,7 +184,7 @@ impl wasmtime_environ::Compiler for Compiler {
             .map_err(|error| CompileError::Codegen(pretty_error(&context.func, error)))?;
 
         let address_transform =
-            self.get_function_address_map(&context, &input, code_buf.len() as u32);
+            self.get_function_address_map(&context, &input, code_buf.len() as u32, tunables);
 
         let ranges = if tunables.generate_native_debuginfo {
             Some(
@@ -221,7 +226,7 @@ impl wasmtime_environ::Compiler for Compiler {
         translation: &ModuleTranslation,
         types: &TypeTables,
         funcs: PrimaryMap<DefinedFuncIndex, Box<dyn Any + Send>>,
-        emit_dwarf: bool,
+        tunables: &Tunables,
         obj: &mut Object<'static>,
     ) -> Result<(PrimaryMap<DefinedFuncIndex, FunctionInfo>, Vec<Trampoline>)> {
         let funcs: crate::CompiledFunctions = funcs
@@ -244,7 +249,9 @@ impl wasmtime_environ::Compiler for Compiler {
         let mut func_starts = Vec::with_capacity(funcs.len());
         for (i, func) in funcs.iter() {
             let range = builder.func(i, func);
-            addrs.push(range.clone(), &func.address_map.instructions);
+            if tunables.generate_address_map {
+                addrs.push(range.clone(), &func.address_map.instructions);
+            }
             traps.push(range.clone(), &func.traps);
             func_starts.push(range.start);
             if self.linkopts.padding_between_functions > 0 {
@@ -266,7 +273,7 @@ impl wasmtime_environ::Compiler for Compiler {
 
         builder.unwind_info();
 
-        if emit_dwarf && funcs.len() > 0 {
+        if tunables.generate_native_debuginfo && funcs.len() > 0 {
             let ofs = VMOffsets::new(
                 self.isa
                     .triple()
@@ -297,7 +304,10 @@ impl wasmtime_environ::Compiler for Compiler {
         }
 
         builder.finish()?;
-        addrs.append_to(obj);
+
+        if tunables.generate_address_map {
+            addrs.append_to(obj);
+        }
         traps.append_to(obj);
 
         Ok((

crates/environ/src/compilation.rs

@@ -171,7 +171,7 @@ pub trait Compiler: Send + Sync {
         module: &ModuleTranslation,
         types: &TypeTables,
         funcs: PrimaryMap<DefinedFuncIndex, Box<dyn Any + Send>>,
-        emit_dwarf: bool,
+        tunables: &Tunables,
         obj: &mut Object<'static>,
     ) -> Result<(PrimaryMap<DefinedFuncIndex, FunctionInfo>, Vec<Trampoline>)>;
 

crates/environ/src/tunables.rs

@@ -42,6 +42,10 @@ pub struct Tunables {
     /// Whether or not linear memory allocations will have a guard region at the
     /// beginning of the allocation in addition to the end.
     pub guard_before_linear_memory: bool,
+
+    /// Indicates whether an address map from compiled native code back to wasm
+    /// offsets in the original file is generated.
+    pub generate_address_map: bool,
 }
 
 impl Default for Tunables {
@@ -86,6 +90,7 @@ impl Default for Tunables {
             consume_fuel: false,
             static_memory_bound_is_maximum: false,
             guard_before_linear_memory: true,
+            generate_address_map: true,
         }
     }
 }

crates/jit/src/instantiate.rs

@@ -297,16 +297,21 @@ impl CompiledModule {
         };
 
         let mut ret = Self {
-            meta: info.meta,
             module: Arc::new(info.module),
             funcs: info.funcs,
             trampolines: info.trampolines,
             wasm_data: subslice_range(section(ELF_WASM_DATA)?, code.mmap),
-            address_map_data: subslice_range(section(ELF_WASMTIME_ADDRMAP)?, code.mmap),
+            address_map_data: code
+                .obj
+                .section_by_name(ELF_WASMTIME_ADDRMAP)
+                .and_then(|s| s.data().ok())
+                .map(|slice| subslice_range(slice, code.mmap))
+                .unwrap_or(0..0),
             trap_data: subslice_range(section(ELF_WASMTIME_TRAPS)?, code.mmap),
             code: subslice_range(code.text, code.mmap),
             dbg_jit_registration: None,
             code_memory,
+            meta: info.meta,
         };
         ret.register_debug_and_profiling(profiler)?;
 
@@ -500,6 +505,15 @@ impl CompiledModule {
     pub fn has_unparsed_debuginfo(&self) -> bool {
         self.meta.has_unparsed_debuginfo
     }
+
+    /// Indicates whether this module came with n address map such that lookups
+    /// via `wasmtime_environ::lookup_file_pos` will succeed.
+    ///
+    /// If this function returns `false` then `lookup_file_pos` will always
+    /// return `None`.
+    pub fn has_address_map(&self) -> bool {
+        !self.address_map_data().is_empty()
+    }
 }
 
 type Addr2LineContext<'a> = addr2line::Context<gimli::EndianSlice<'a, gimli::LittleEndian>>;

crates/wasmtime/src/config.rs

@@ -1052,6 +1052,20 @@ impl Config {
         self
     }
 
+    /// Configures whether compiled artifacts will contain information to map
+    /// native program addresses back to the original wasm module.
+    ///
+    /// This configuration option is `true` by default and, if enables,
+    /// generates the appropriate tables in compiled modules to map from native
+    /// address back to wasm source addresses. This is used for displaying wasm
+    /// program counters in backtraces as well as generating filenames/line
+    /// numbers if so configured as well (and the original wasm module has DWARF
+    /// debugging information present).
+    pub fn generate_address_map(&mut self, generate: bool) -> &mut Self {
+        self.tunables.generate_address_map = generate;
+        self
+    }
+
     pub(crate) fn build_allocator(&self) -> Result<Box<dyn InstanceAllocator>> {
         #[cfg(feature = "async")]
         let stack_size = self.async_stack_size;

crates/wasmtime/src/engine.rs

@@ -50,7 +50,7 @@ impl Engine {
         // Ensure that wasmtime_runtime's signal handlers are configured. This
         // is the per-program initialization required for handling traps, such
         // as configuring signals, vectored exception handlers, etc.
-        wasmtime_runtime::init_traps(crate::module::GlobalModuleRegistry::is_wasm_pc);
+        wasmtime_runtime::init_traps(crate::module::GlobalModuleRegistry::is_wasm_trap_pc);
         debug_builtins::ensure_exported();
 
         let registry = SignatureRegistry::new();

crates/wasmtime/src/module.rs

@@ -398,13 +398,10 @@ impl Module {
                 .collect();
 
             let mut obj = engine.compiler().object()?;
-            let (funcs, trampolines) = engine.compiler().emit_obj(
-                &translation,
-                &types,
-                funcs,
-                tunables.generate_native_debuginfo,
-                &mut obj,
-            )?;
+            let (funcs, trampolines) =
+                engine
+                    .compiler()
+                    .emit_obj(&translation, &types, funcs, tunables, &mut obj)?;
 
             // If configured, attempt to use paged memory initialization
             // instead of the default mode of memory initialization

crates/wasmtime/src/module/registry.rs

@@ -172,13 +172,12 @@ pub struct GlobalModuleRegistry(BTreeMap<usize, GlobalRegisteredModule>);
 impl GlobalModuleRegistry {
     /// Returns whether the `pc`, according to globally registered information,
     /// is a wasm trap or not.
-    pub(crate) fn is_wasm_pc(pc: usize) -> bool {
+    pub(crate) fn is_wasm_trap_pc(pc: usize) -> bool {
         let modules = GLOBAL_MODULES.read().unwrap();
 
         match modules.module(pc) {
             Some((entry, text_offset)) => {
-                wasmtime_environ::lookup_file_pos(entry.module.address_map_data(), text_offset)
-                    .is_some()
+                wasmtime_environ::lookup_trap_code(entry.module.trap_data(), text_offset).is_some()
             }
             None => false,
         }
@@ -275,14 +274,15 @@ impl GlobalRegisteredModule {
         // the function, because otherwise something is buggy along the way and
         // not accounting for all the instructions. This isn't super critical
         // though so we can omit this check in release mode.
+        //
+        // Note that if the module doesn't even have an address map due to
+        // compilation settings then it's expected that `instr` is `None`.
         debug_assert!(
-            instr.is_some(),
+            instr.is_some() || !self.module.has_address_map(),
             "failed to find instruction for {:#x}",
             text_offset
         );
 
-        let instr = instr.unwrap_or(info.start_srcloc);
-
         // Use our wasm-relative pc to symbolize this frame. If there's a
         // symbolication context (dwarf debug info) available then we can try to
         // look this up there.
@@ -294,7 +294,7 @@ impl GlobalRegisteredModule {
         let mut symbols = Vec::new();
 
         if let Some(s) = &self.module.symbolize_context().ok().and_then(|c| c) {
-            if let Some(offset) = instr.file_offset() {
+            if let Some(offset) = instr.and_then(|i| i.file_offset()) {
                 let to_lookup = u64::from(offset) - s.code_section_offset();
                 if let Ok(mut frames) = s.addr2line().find_frames(to_lookup) {
                     while let Ok(Some(frame)) = frames.next() {
@@ -344,7 +344,7 @@ pub struct FrameInfo {
     func_index: u32,
     func_name: Option<String>,
     func_start: FilePos,
-    instr: FilePos,
+    instr: Option<FilePos>,
     symbols: Vec<FrameSymbol>,
 }
 
@@ -393,8 +393,14 @@ impl FrameInfo {
     ///
     /// The offset here is the offset from the beginning of the original wasm
     /// module to the instruction that this frame points to.
-    pub fn module_offset(&self) -> usize {
-        self.instr.file_offset().unwrap_or(u32::MAX) as usize
+    ///
+    /// Note that `None` may be returned if the original module was not
+    /// compiled with mapping information to yield this information. This is
+    /// controlled by the
+    /// [`Config::generate_address_map`](crate::Config::generate_address_map)
+    /// configuration option.
+    pub fn module_offset(&self) -> Option<usize> {
+        Some(self.instr?.file_offset()? as usize)
     }
 
     /// Returns the offset from the original wasm module's function to this
@@ -403,11 +409,15 @@ impl FrameInfo {
     /// The offset here is the offset from the beginning of the defining
     /// function of this frame (within the wasm module) to the instruction this
     /// frame points to.
-    pub fn func_offset(&self) -> usize {
-        match self.instr.file_offset() {
-            Some(i) => (i - self.func_start.file_offset().unwrap()) as usize,
-            None => u32::MAX as usize,
-        }
+    ///
+    /// Note that `None` may be returned if the original module was not
+    /// compiled with mapping information to yield this information. This is
+    /// controlled by the
+    /// [`Config::generate_address_map`](crate::Config::generate_address_map)
+    /// configuration option.
+    pub fn func_offset(&self) -> Option<usize> {
+        let instr_offset = self.instr?.file_offset()?;
+        Some((instr_offset - self.func_start.file_offset()?) as usize)
     }
 
     /// Returns the debug symbols found, if any, for this function frame.

crates/wasmtime/src/module/serialization.rs

@@ -601,6 +601,12 @@ impl<'a> SerializedModule<'a> {
 
             // This doesn't affect compilation, it's just a runtime setting.
             dynamic_memory_growth_reserve: _,
+
+            // This does technically affect compilation but modules with/without
+            // trap information can be loaded into engines with the opposite
+            // setting just fine (it's just a section in the compiled file and
+            // whether it's present or not)
+            generate_address_map: _,
         } = self.metadata.tunables;
 
         Self::check_int(

crates/wasmtime/src/trap.rs

@@ -321,7 +321,11 @@ impl fmt::Display for Trap {
         writeln!(f, "\nwasm backtrace:")?;
         for (i, frame) in self.trace().iter().enumerate() {
             let name = frame.module_name().unwrap_or("<unknown>");
-            write!(f, "  {:>3}: {:#6x} - ", i, frame.module_offset())?;
+            write!(f, "  {:>3}: ", i)?;
+
+            if let Some(offset) = frame.module_offset() {
+                write!(f, "{:#6x} - ", offset)?;
+            }
 
             let demangle =
                 |f: &mut fmt::Formatter<'_>, name: &str| match rustc_demangle::try_demangle(name) {

tests/all/traps.rs

@@ -54,13 +54,13 @@ fn test_trap_trace() -> Result<()> {
     assert_eq!(trace[0].module_name().unwrap(), "hello_mod");
     assert_eq!(trace[0].func_index(), 1);
     assert_eq!(trace[0].func_name(), Some("hello"));
-    assert_eq!(trace[0].func_offset(), 1);
-    assert_eq!(trace[0].module_offset(), 0x26);
+    assert_eq!(trace[0].func_offset(), Some(1));
+    assert_eq!(trace[0].module_offset(), Some(0x26));
     assert_eq!(trace[1].module_name().unwrap(), "hello_mod");
     assert_eq!(trace[1].func_index(), 0);
     assert_eq!(trace[1].func_name(), None);
-    assert_eq!(trace[1].func_offset(), 1);
-    assert_eq!(trace[1].module_offset(), 0x21);
+    assert_eq!(trace[1].func_offset(), Some(1));
+    assert_eq!(trace[1].module_offset(), Some(0x21));
     assert!(
         e.to_string().contains("unreachable"),
         "wrong message: {}",
@@ -637,3 +637,37 @@ fn multithreaded_traps() -> Result<()> {
 
     Ok(())
 }
+
+#[test]
+#[cfg_attr(all(target_os = "macos", target_arch = "aarch64"), ignore)] // TODO #2808 system libunwind is broken on aarch64
+fn traps_without_address_map() -> Result<()> {
+    let mut config = Config::new();
+    config.generate_address_map(false);
+    let engine = Engine::new(&config)?;
+    let mut store = Store::new(&engine, ());
+    let wat = r#"
+        (module $hello_mod
+            (func (export "run") (call $hello))
+            (func $hello (unreachable))
+        )
+    "#;
+
+    let module = Module::new(store.engine(), wat)?;
+    let instance = Instance::new(&mut store, &module, &[])?;
+    let run_func = instance.get_typed_func::<(), (), _>(&mut store, "run")?;
+
+    let e = run_func
+        .call(&mut store, ())
+        .err()
+        .expect("error calling function");
+
+    let trace = e.trace();
+    assert_eq!(trace.len(), 2);
+    assert_eq!(trace[0].func_name(), Some("hello"));
+    assert_eq!(trace[0].func_index(), 1);
+    assert_eq!(trace[0].module_offset(), None);
+    assert_eq!(trace[1].func_name(), None);
+    assert_eq!(trace[1].func_index(), 0);
+    assert_eq!(trace[1].module_offset(), None);
+    Ok(())
+}