Adds perf jitdump support

Patch adds support for the perf jitdump file specification.
With this patch it should be possible to see profile data for code
generated and maped at runtime. Specifically the patch adds support
for the JIT_CODE_LOAD and the JIT_DEBUG_INFO record as described in
the specification. Dumping jitfiles is enabled with the --jitdump
flag. When the -g flag is also used there is an attempt to dump file
and line number information where this option would be most useful
when the WASM file already includes DWARF debug information.

misc/wasmtime-py/src/import.rs

@@ -328,6 +328,7 @@ pub fn into_instance_from_obj(
         &data_initializers,
         signatures.into_boxed_slice(),
         None,
+        None,
         Box::new(import_obj_state),
     )
     .expect("instance"))

src/bin/wasmtime.rs

@@ -62,8 +62,8 @@ including calling the start function if one is present. Additional functions
 given with --invoke are then called.
 
 Usage:
-    wasmtime [-odg] [--enable-simd] [--wasi-c] [--disable-cache | --cache-config=<cache_config_file>] [--preload=<wasm>...] [--env=<env>...] [--dir=<dir>...] [--mapdir=<mapping>...] [--lightbeam | --cranelift] <file> [<arg>...]
-    wasmtime [-odg] [--enable-simd] [--wasi-c] [--disable-cache | --cache-config=<cache_config_file>] [--env=<env>...] [--dir=<dir>...] [--mapdir=<mapping>...] --invoke=<fn> [--lightbeam | --cranelift] <file> [<arg>...]
+    wasmtime [-odg] [--enable-simd] [--jitdump] [--wasi-c] [--disable-cache | --cache-config=<cache_config_file>] [--preload=<wasm>...] [--env=<env>...] [--dir=<dir>...] [--mapdir=<mapping>...] [--lightbeam | --cranelift] <file> [<arg>...]
+    wasmtime [-odg] [--enable-simd] [--jitdump] [--wasi-c] [--disable-cache | --cache-config=<cache_config_file>] [--env=<env>...] [--dir=<dir>...] [--mapdir=<mapping>...] --invoke=<fn> [--lightbeam | --cranelift] <file> [<arg>...]
     wasmtime --create-cache-config [--cache-config=<cache_config_file>]
     wasmtime --help | --version
 
@@ -83,6 +83,7 @@ Options:
     --lightbeam         use Lightbeam for all compilation
     --cranelift         use Cranelift for all compilation
     --enable-simd       enable proposed SIMD instructions
+    --jitdump           generate perf jitdump files for runtime generated code
     --wasi-c            enable the wasi-c implementation of WASI
     --preload=<wasm>    load an additional wasm module before loading the main module
     --env=<env>         pass an environment variable (\"key=value\") to the program
@@ -106,6 +107,7 @@ struct Args {
     flag_enable_simd: bool,
     flag_lightbeam: bool,
     flag_cranelift: bool,
+    flag_jitdump: bool,
     flag_invoke: Option<String>,
     flag_preload: Vec<String>,
     flag_env: Vec<String>,
@@ -267,6 +269,9 @@ fn rmain() -> Result<(), Error> {
         features.simd = true;
     }
 
+    // Enable Jitdump if requested
+    let perf_profile = args.flag_jitdump;
+
     // Enable optimization if requested.
     if args.flag_optimize {
         flag_builder.set("opt_level", "speed")?;
@@ -280,6 +285,7 @@ fn rmain() -> Result<(), Error> {
         features,
         debug_info,
         strategy,
+        perf_profile,
     );
     let engine = HostRef::new(Engine::new(config));
     let store = HostRef::new(Store::new(engine));

tests/instantiate.rs

@@ -28,6 +28,13 @@ fn test_environ_translate() {
     let mut resolver = NullResolver {};
     let mut compiler = Compiler::new(isa, CompilationStrategy::Auto);
     let global_exports = Rc::new(RefCell::new(HashMap::new()));
-    let instance = instantiate(&mut compiler, &data, &mut resolver, global_exports, false);
+    let instance = instantiate(
+        &mut compiler,
+        &data,
+        &mut resolver,
+        global_exports,
+        false,
+        false,
+    );
     assert!(instance.is_ok());
 }

wasmtime-api/src/context.rs

@@ -11,14 +11,21 @@ pub struct Context {
     compiler: Rc<RefCell<Compiler>>,
     features: Features,
     debug_info: bool,
+    perf_profile: bool,
 }
 
 impl Context {
-    pub fn new(compiler: Compiler, features: Features, debug_info: bool) -> Context {
+    pub fn new(
+        compiler: Compiler,
+        features: Features,
+        debug_info: bool,
+        perf_profile: bool,
+    ) -> Context {
         Context {
             compiler: Rc::new(RefCell::new(compiler)),
             features,
             debug_info,
+            perf_profile,
         }
     }
 
@@ -27,14 +34,19 @@ impl Context {
         features: Features,
         debug_info: bool,
         strategy: CompilationStrategy,
+        perf_profile: bool,
     ) -> Context {
-        Context::new(create_compiler(flags, strategy), features, debug_info)
+        Context::new(create_compiler(flags, strategy), features, debug_info, perf_profile)
     }
 
     pub(crate) fn debug_info(&self) -> bool {
         self.debug_info
     }
 
+    pub(crate) fn perf_profile(&self) -> bool {
+        self.perf_profile
+    }
+
     pub(crate) fn compiler(&mut self) -> RefMut<Compiler> {
         self.compiler.borrow_mut()
     }

wasmtime-api/src/instance.rs

@@ -37,13 +37,15 @@ pub fn instantiate_in_context(
 ) -> Result<(InstanceHandle, HashSet<Context>), Error> {
     let mut contexts = HashSet::new();
     let debug_info = context.debug_info();
+    let perf_profile = context.perf_profile();
     let mut resolver = SimpleResolver { imports };
     let instance = instantiate(
         &mut context.compiler(),
         data,
         &mut resolver,
         exports,
         debug_info,
+        perf_profile,
     )?;
     contexts.insert(context);
     Ok((instance, contexts))

wasmtime-api/src/runtime.rs

@@ -24,12 +24,14 @@ pub struct Config {
     features: Features,
     debug_info: bool,
     strategy: CompilationStrategy,
+    perf_profile: bool,
 }
 
 impl Config {
     pub fn default() -> Config {
         Config {
             debug_info: false,
+            perf_profile: false,
             features: Default::default(),
             flags: default_flags(),
             strategy: CompilationStrategy::Auto,
@@ -41,19 +43,25 @@ impl Config {
         features: Features,
         debug_info: bool,
         strategy: CompilationStrategy,
+        perf_profile: bool,
     ) -> Config {
         Config {
             flags,
             features,
             debug_info,
             strategy,
+            perf_profile,
         }
     }
 
     pub(crate) fn debug_info(&self) -> bool {
         self.debug_info
     }
 
+    pub(crate) fn perf_profile(&self) -> bool {
+        self.perf_profile
+    }
+
     pub(crate) fn flags(&self) -> &settings::Flags {
         &self.flags
     }
@@ -107,9 +115,10 @@ impl Store {
         let features = engine.borrow().config().features().clone();
         let debug_info = engine.borrow().config().debug_info();
         let strategy = engine.borrow().config().strategy();
+        let perf_profile = engine.borrow().config().perf_profile();
         Store {
             engine,
-            context: Context::create(flags, features, debug_info, strategy),
+            context: Context::create(flags, features, debug_info, strategy, perf_profile),
             global_exports: Rc::new(RefCell::new(HashMap::new())),
             signature_cache: HashMap::new(),
         }

wasmtime-api/src/trampoline/create_handle.rs

@@ -56,6 +56,7 @@ pub(crate) fn create_handle(
         &data_initializers,
         signatures.into_boxed_slice(),
         None,
+        None,
         state,
     )
     .expect("instance"))

wasmtime-jit/src/code_memory.rs

@@ -5,7 +5,7 @@ use alloc::string::String;
 use alloc::vec::Vec;
 use core::{cmp, mem};
 use region;
-use wasmtime_runtime::{Mmap, VMFunctionBody};
+use wasmtime_runtime::{JitDumpAgent, Mmap, VMFunctionBody};
 
 /// Memory manager for executable code.
 pub struct CodeMemory {
@@ -99,4 +99,21 @@ impl CodeMemory {
         }
         self.published = self.mmaps.len();
     }
+
+    /// Clones jit_dump_agent and calls code for writing the jitdump record for a
+    /// newly loaded module.
+    pub fn perf_module_load(
+        &mut self,
+        module_name: &str,
+        jit_dump_agent: &JitDumpAgent,
+        dbg_image: Option<&[u8]>,
+    ) -> () {
+        for map in &mut self.mmaps {
+            if map.len() > 0 {
+                jit_dump_agent
+                    .clone()
+                    .module_load(module_name, map.as_ptr(), map.len(), dbg_image);
+            }
+        }
+    }
 }

wasmtime-jit/src/compiler.rs

@@ -21,8 +21,8 @@ use wasmtime_environ::{
     Relocations, Traps, Tunables, VMOffsets,
 };
 use wasmtime_runtime::{
-    get_mut_trap_registry, InstantiationError, SignatureRegistry, TrapRegistrationGuard,
-    VMFunctionBody,
+    get_mut_trap_registry, InstantiationError, JitDumpAgent, SignatureRegistry,
+    TrapRegistrationGuard, VMFunctionBody,
 };
 
 /// Select which kind of compilation to use.
@@ -236,6 +236,16 @@ impl Compiler {
         self.code_memory.publish();
     }
 
+    pub(crate) fn perf_module_load(
+        &mut self,
+        module_name: &str,
+        jit_dump_agent: &JitDumpAgent,
+        dbg_image: Option<&[u8]>,
+    ) -> () {
+        self.code_memory
+            .perf_module_load(module_name, jit_dump_agent, dbg_image);
+    }
+
     /// Shared signature registry.
     pub fn signatures(&mut self) -> &mut SignatureRegistry {
         &mut self.signatures

wasmtime-jit/src/context.rs

@@ -77,6 +77,7 @@ pub struct Context {
     compiler: Box<Compiler>,
     global_exports: Rc<RefCell<HashMap<String, Option<wasmtime_runtime::Export>>>>,
     debug_info: bool,
+    perf_profile: bool,
     features: Features,
 }
 
@@ -88,6 +89,7 @@ impl Context {
             compiler,
             global_exports: Rc::new(RefCell::new(HashMap::new())),
             debug_info: false,
+            perf_profile: false,
             features: Default::default(),
         }
     }
@@ -97,6 +99,11 @@ impl Context {
         self.debug_info
     }
 
+    /// Get debug_info settings.
+    pub fn perf_profile(&self) -> bool {
+        self.perf_profile
+    }
+
     /// Set debug_info settings.
     pub fn set_debug_info(&mut self, value: bool) {
         self.debug_info = value;
@@ -127,13 +134,15 @@ impl Context {
     fn instantiate(&mut self, data: &[u8]) -> Result<InstanceHandle, SetupError> {
         self.validate(&data).map_err(SetupError::Validate)?;
         let debug_info = self.debug_info();
+        let perf_profile = self.perf_profile();
 
         instantiate(
             &mut *self.compiler,
             &data,
             &mut self.namespace,
             Rc::clone(&self.global_exports),
             debug_info,
+            perf_profile,
         )
     }
 
@@ -164,13 +173,15 @@ impl Context {
     pub fn compile_module(&mut self, data: &[u8]) -> Result<CompiledModule, SetupError> {
         self.validate(&data).map_err(SetupError::Validate)?;
         let debug_info = self.debug_info();
+        let perf_profile = self.perf_profile();
 
         CompiledModule::new(
             &mut *self.compiler,
             data,
             &mut self.namespace,
             Rc::clone(&self.global_exports),
             debug_info,
+            perf_profile,
         )
     }