Update gimli version; refactor debug data structures

wasmtime-debug/Cargo.toml

@@ -12,13 +12,12 @@ readme = "README.md"
 edition = "2018"
 
 [dependencies]
-gimli = "0.17.0"
+gimli = "0.19.0"
 wasmparser = { version = "0.32.1" }
 cranelift-codegen = "0.33.0"
 cranelift-entity = "0.33.0"
 cranelift-wasm = "0.33.0"
 faerie = "0.10.1"
-wasmtime-environ = { path = "../wasmtime-environ", default-features = false }
 target-lexicon = { version = "0.4.0", default-features = false }
 failure = { version = "0.1.3", default-features = false }
 failure_derive = { version = "0.1.3", default-features = false }

wasmtime-debug/src/address_transform.rs

@@ -1,11 +1,11 @@
 use crate::read_debuginfo::WasmFileInfo;
+use crate::transform::ModuleAddressMap;
 use cranelift_entity::{EntityRef, PrimaryMap};
 use cranelift_wasm::DefinedFuncIndex;
 use gimli::write;
 use std::collections::BTreeMap;
 use std::ops::Bound::{Included, Unbounded};
 use std::vec::Vec;
-use wasmtime_environ::AddressTransforms;
 
 pub type GeneratedAddress = usize;
 pub type WasmAddress = u64;
@@ -32,7 +32,7 @@ pub struct AddressTransform {
 }
 
 impl AddressTransform {
-    pub fn new(at: &AddressTransforms, wasm_file: &WasmFileInfo) -> Self {
+    pub fn new(at: &ModuleAddressMap, wasm_file: &WasmFileInfo) -> Self {
         let code_section_offset = wasm_file.code_section_offset;
         let function_offsets = &wasm_file.function_offsets_and_sizes;
         let mut lookup = BTreeMap::new();
@@ -50,7 +50,7 @@ impl AddressTransform {
                 (index, ft.body_offset, ft.body_offset),
             );
             let mut fn_map = Vec::new();
-            for t in &ft.locations {
+            for t in &ft.instructions {
                 if t.srcloc.is_default() {
                     // TODO extend some range if possible
                     continue;
@@ -90,7 +90,7 @@ impl AddressTransform {
         }
         let search = self.lookup.range((Unbounded, Included(addr)));
         if let Some((_, value)) = search.last() {
-            return Some(write::Address::Relative {
+            return Some(write::Address::Symbol {
                 symbol: value.0,
                 addend: value.1 as i64,
             });
@@ -106,11 +106,11 @@ impl AddressTransform {
             return None;
         }
         if let (
-            Some(write::Address::Relative {
+            Some(write::Address::Symbol {
                 symbol: s1,
                 addend: a,
             }),
-            Some(write::Address::Relative {
+            Some(write::Address::Symbol {
                 symbol: s2,
                 addend: b,
             }),

wasmtime-debug/src/lib.rs

@@ -4,12 +4,13 @@ use faerie::{Artifact, Decl};
 use failure::Error;
 use target_lexicon::{BinaryFormat, Triple};
 
-pub use crate::read_debuginfo::{read_debuginfo, DebugInfoData};
+pub use crate::read_debuginfo::{read_debuginfo, DebugInfoData, WasmFileInfo};
-pub use crate::transform::transform_dwarf;
+pub use crate::transform::{
+    transform_dwarf, FunctionAddressMap, InstructionAddressMap, ModuleAddressMap, ModuleVmctxInfo,
+    ValueLabelsRanges,
+};
 pub use crate::write_debuginfo::{emit_dwarf, ResolvedSymbol, SymbolResolver};
 
-use wasmtime_environ::AddressTransforms;
-
 mod address_transform;
 mod read_debuginfo;
 mod transform;
@@ -30,11 +31,11 @@ pub fn emit_debugsections(
     obj: &mut Artifact,
     target_config: &TargetFrontendConfig,
     debuginfo_data: &DebugInfoData,
-    at: &AddressTransforms,
+    at: &ModuleAddressMap,
 ) -> Result<(), Error> {
-    let dwarf = transform_dwarf(target_config, debuginfo_data, at)?;
     let resolver = FunctionRelocResolver {};
-    emit_dwarf(obj, dwarf, &resolver);
+    let dwarf = transform_dwarf(target_config, debuginfo_data, at)?;
+    emit_dwarf(obj, dwarf, &resolver)?;
     Ok(())
 }
 
@@ -53,7 +54,7 @@ pub fn emit_debugsections_image(
     triple: Triple,
     target_config: &TargetFrontendConfig,
     debuginfo_data: &DebugInfoData,
-    at: &AddressTransforms,
+    at: &ModuleAddressMap,
     funcs: &Vec<(*const u8, usize)>,
 ) -> Result<Vec<u8>, Error> {
     let ref func_offsets = funcs
@@ -61,8 +62,8 @@ pub fn emit_debugsections_image(
         .map(|(ptr, _)| *ptr as u64)
         .collect::<Vec<u64>>();
     let mut obj = Artifact::new(triple, String::from("module"));
-    let dwarf = transform_dwarf(target_config, debuginfo_data, at)?;
     let resolver = ImageRelocResolver { func_offsets };
+    let dwarf = transform_dwarf(target_config, debuginfo_data, at)?;
 
     // Assuming all functions in the same code block, looking min/max of its range.
     assert!(funcs.len() > 0);
@@ -76,7 +77,7 @@ pub fn emit_debugsections_image(
     let body = unsafe { ::std::slice::from_raw_parts(segment_body.0, segment_body.1) };
     obj.declare_with("all", Decl::function(), body.to_vec())?;
 
-    emit_dwarf(&mut obj, dwarf, &resolver);
+    emit_dwarf(&mut obj, dwarf, &resolver)?;
 
     // LLDB is too "magical" about mach-o, generating elf
     let mut bytes = obj.emit_as(BinaryFormat::Elf)?;

wasmtime-debug/src/transform.rs

@@ -1,7 +1,9 @@
 use crate::address_transform::AddressTransform;
 pub use crate::read_debuginfo::DebugInfoData;
+use cranelift_codegen::ir;
 use cranelift_codegen::isa::TargetFrontendConfig;
-use cranelift_entity::EntityRef;
+use cranelift_entity::{EntityRef, PrimaryMap};
+use cranelift_wasm::DefinedFuncIndex;
 use failure::Error;
 use std::collections::{BTreeMap, HashMap};
 use std::ops::Bound::{Included, Unbounded};
@@ -24,14 +26,45 @@ impl<'input, Endian> Reader for gimli::EndianSlice<'input, Endian> where Endian:
 #[fail(display = "Debug info transform error: {}", _0)]
 pub struct TransformError(&'static str);
 
-pub struct TransformedDwarf {
+/// Single wasm source location to generated address mapping.
-    pub encoding: gimli::Encoding,
+#[derive(Debug, Clone)]
-    pub strings: write::StringTable,
+pub struct InstructionAddressMap {
-    pub units: write::UnitTable,
+    /// Original source location.
-    pub line_strings: write::LineStringTable,
+    pub srcloc: ir::SourceLoc,
-    pub range_lists: write::RangeListTable,
+
+    /// Generated instructions offset.
+    pub code_offset: usize,
+
+    /// Generated instructions length.
+    pub code_len: usize,
 }
 
+/// Function and its instructions addresses mappings.
+#[derive(Debug, Clone)]
+pub struct FunctionAddressMap {
+    /// Instructions maps.
+    /// The array is sorted by the InstructionAddressMap::code_offset field.
+    pub instructions: Vec<InstructionAddressMap>,
+
+    /// Generated function body offset if applicable, otherwise 0.
+    pub body_offset: usize,
+
+    /// Generated function body length.
+    pub body_len: usize,
+}
+
+/// Module functions addresses mappings.
+pub type ModuleAddressMap = PrimaryMap<DefinedFuncIndex, FunctionAddressMap>;
+
+/// Module `vmctx` related info.
+pub struct ModuleVmctxInfo {
+    pub memory_offset: i64,
+    pub stack_slots: PrimaryMap<DefinedFuncIndex, ir::StackSlots>,
+}
+
+/// Value ranges for functions.
+pub type ValueLabelsRanges = PrimaryMap<DefinedFuncIndex, cranelift_codegen::ValueLabelsRanges>;
+
 struct DebugInputContext<'a, R>
 where
     R: Reader,
@@ -84,7 +117,7 @@ where
                 write::AttributeValue::Udata(subprogram_range.unwrap().1)
             }
             AttributeValue::Addr(u) => {
-                let addr = addr_tr.translate(u).unwrap_or(write::Address::Absolute(0));
+                let addr = addr_tr.translate(u).unwrap_or(write::Address::Constant(0));
                 if attr.name() == gimli::DW_AT_low_pc {
                     low_pc = Some((u, addr));
                 }
@@ -370,7 +403,7 @@ where
         for (i, map) in addr_tr.map() {
             let symbol = i.index();
             let base_addr = map.offset;
-            out_program.begin_sequence(Some(write::Address::Relative { symbol, addend: 0 }));
+            out_program.begin_sequence(Some(write::Address::Symbol { symbol, addend: 0 }));
             // TODO track and place function declaration line here
             let mut last_address = None;
             for addr_map in map.addresses.iter() {
@@ -440,9 +473,9 @@ where
     let low_pc = entry.attr_value(gimli::DW_AT_low_pc)?;
     if let Some(AttributeValue::Addr(addr)) = low_pc {
         let transformed = addr_tr.translate(addr);
-        if let Some(write::Address::Relative { symbol, .. }) = transformed {
+        if let Some(write::Address::Symbol { symbol, .. }) = transformed {
             let range = addr_tr.func_range(symbol);
-            let addr = write::Address::Relative {
+            let addr = write::Address::Symbol {
                 symbol,
                 addend: range.0 as i64,
             };
@@ -580,8 +613,8 @@ where
 pub fn transform_dwarf(
     target_config: &TargetFrontendConfig,
     di: &DebugInfoData,
-    at: &wasmtime_environ::AddressTransforms,
+    at: &ModuleAddressMap,
-) -> Result<TransformedDwarf, Error> {
+) -> Result<write::Dwarf, Error> {
     let context = DebugInputContext {
         debug_abbrev: &di.dwarf.debug_abbrev,
         debug_str: &di.dwarf.debug_str,
@@ -605,7 +638,6 @@ pub fn transform_dwarf(
     let mut out_strings = write::StringTable::default();
     let mut out_units = write::UnitTable::default();
 
-    let out_range_lists = write::RangeListTable::default();
     let out_line_strings = write::LineStringTable::default();
 
     let mut iter = di.dwarf.debug_info.units();
@@ -620,22 +652,10 @@ pub fn transform_dwarf(
         )?;
     }
 
-    // let unit_range_list = write::RangeList(Vec::new());
+    Ok(write::Dwarf {
-    // let unit_range_list_id = out_range_lists.add(unit_range_list.clone());
-    // let unit = dwarf.units.get_mut(self.unit_id);
-    // let root = unit.root();
-    // let root = unit.get_mut(root);
-    // root.set(
-    //     gimli::DW_AT_ranges,
-    //     AttributeValue::RangeListRef(unit_range_list_id),
-    // );
-
-    //println!("{:?} \n====\n {:?}", di, at);
-    Ok(TransformedDwarf {
-        encoding: out_encoding,
-        strings: out_strings,
         units: out_units,
+        line_programs: vec![],
         line_strings: out_line_strings,
-        range_lists: out_range_lists,
+        strings: out_strings,
     })
 }

wasmtime-debug/src/write_debuginfo.rs

@@ -1,14 +1,11 @@
-use crate::transform::TransformedDwarf;
+use gimli::write::{Address, Dwarf, EndianVec, Result, Sections, Writer};
-
+use gimli::{RunTimeEndian, SectionId};
-use gimli::write::{
-    Address, DebugAbbrev, DebugInfo, DebugLine, DebugLineStr, DebugRanges, DebugRngLists, DebugStr,
-    EndianVec, Result, SectionId, Sections, Writer,
-};
-use gimli::RunTimeEndian;
 
 use faerie::artifact::{Decl, SectionKind};
 use faerie::*;
+use std::result;
 
+#[derive(Clone)]
 struct DebugReloc {
     offset: u32,
     size: u8,
@@ -16,38 +13,6 @@ struct DebugReloc {
     addend: i64,
 }
 
-macro_rules! decl_section {
-    ($artifact:ident . $section:ident = $name:expr) => {
-        $artifact
-            .declare_with(
-                SectionId::$section.name(),
-                Decl::section(SectionKind::Debug),
-                $name.0.writer.into_vec(),
-            )
-            .unwrap();
-    };
-}
-
-macro_rules! sect_relocs {
-    ($artifact:ident . $section:ident = $name:expr) => {
-        for reloc in $name.0.relocs {
-            $artifact
-                .link_with(
-                    faerie::Link {
-                        from: SectionId::$section.name(),
-                        to: &reloc.name,
-                        at: u64::from(reloc.offset),
-                    },
-                    faerie::Reloc::Debug {
-                        size: reloc.size,
-                        addend: reloc.addend as i32,
-                    },
-                )
-                .expect("faerie relocation error");
-        }
-    };
-}
-
 pub enum ResolvedSymbol {
     PhysicalAddress(u64),
     Reloc { name: String, addend: i64 },
@@ -59,75 +24,48 @@ pub trait SymbolResolver {
 
 pub fn emit_dwarf(
     artifact: &mut Artifact,
-    mut dwarf: TransformedDwarf,
+    mut dwarf: Dwarf,
     symbol_resolver: &SymbolResolver,
-) {
+) -> result::Result<(), failure::Error> {
     let endian = RunTimeEndian::Little;
-    let debug_abbrev = DebugAbbrev::from(WriterRelocate::new(endian, symbol_resolver));
-    let debug_info = DebugInfo::from(WriterRelocate::new(endian, symbol_resolver));
-    let debug_str = DebugStr::from(WriterRelocate::new(endian, symbol_resolver));
-    let debug_line = DebugLine::from(WriterRelocate::new(endian, symbol_resolver));
-    let debug_ranges = DebugRanges::from(WriterRelocate::new(endian, symbol_resolver));
-    let debug_rnglists = DebugRngLists::from(WriterRelocate::new(endian, symbol_resolver));
-    let debug_line_str = DebugLineStr::from(WriterRelocate::new(endian, symbol_resolver));
 
-    let mut sections = Sections {
+    let mut sections = Sections::new(WriterRelocate::new(endian, symbol_resolver));
-        debug_abbrev,
+    dwarf.write(&mut sections)?;
-        debug_info,
+    sections.for_each_mut(|id, s| -> result::Result<(), failure::Error> {
-        debug_line,
+        artifact.declare_with(
-        debug_line_str,
+            id.name(),
-        debug_ranges,
+            Decl::section(SectionKind::Debug),
-        debug_rnglists,
+            s.writer.take(),
-        debug_str,
+        )
-    };
+    })?;
-
+    sections.for_each_mut(|id, s| -> result::Result<(), failure::Error> {
-    let debug_str_offsets = dwarf.strings.write(&mut sections.debug_str).unwrap();
+        for reloc in &s.relocs {
-    let debug_line_str_offsets = dwarf
+            artifact.link_with(
-        .line_strings
+                faerie::Link {
-        .write(&mut sections.debug_line_str)
+                    from: id.name(),
-        .unwrap();
+                    to: &reloc.name,
-    dwarf
+                    at: u64::from(reloc.offset),
-        .units
+                },
-        .write(&mut sections, &debug_line_str_offsets, &debug_str_offsets)
+                faerie::Reloc::Debug {
-        .unwrap();
+                    size: reloc.size,
-
+                    addend: reloc.addend as i32,
-    decl_section!(artifact.DebugAbbrev = sections.debug_abbrev);
+                },
-    decl_section!(artifact.DebugInfo = sections.debug_info);
+            )?;
-    decl_section!(artifact.DebugStr = sections.debug_str);
+        }
-    decl_section!(artifact.DebugLine = sections.debug_line);
+        Ok(())
-
+    })?;
-    let debug_ranges_not_empty = !sections.debug_ranges.0.writer.slice().is_empty();
+    Ok(())
-    if debug_ranges_not_empty {
-        decl_section!(artifact.DebugRanges = sections.debug_ranges);
 }
 
-    let debug_rnglists_not_empty = !sections.debug_rnglists.0.writer.slice().is_empty();
+#[derive(Clone)]
-    if debug_rnglists_not_empty {
+pub struct WriterRelocate<'a> {
-        decl_section!(artifact.DebugRngLists = sections.debug_rnglists);
-    }
-
-    sect_relocs!(artifact.DebugAbbrev = sections.debug_abbrev);
-    sect_relocs!(artifact.DebugInfo = sections.debug_info);
-    sect_relocs!(artifact.DebugStr = sections.debug_str);
-    sect_relocs!(artifact.DebugLine = sections.debug_line);
-
-    if debug_ranges_not_empty {
-        sect_relocs!(artifact.DebugRanges = sections.debug_ranges);
-    }
-
-    if debug_rnglists_not_empty {
-        sect_relocs!(artifact.DebugRngLists = sections.debug_rnglists);
-    }
-}
-
-struct WriterRelocate<'a> {
     relocs: Vec<DebugReloc>,
     writer: EndianVec<RunTimeEndian>,
     symbol_resolver: &'a SymbolResolver,
 }
 
 impl<'a> WriterRelocate<'a> {
-    fn new(endian: RunTimeEndian, symbol_resolver: &'a SymbolResolver) -> Self {
+    pub fn new(endian: RunTimeEndian, symbol_resolver: &'a SymbolResolver) -> Self {
         WriterRelocate {
             relocs: Vec::new(),
             writer: EndianVec::new(endian),
@@ -157,10 +95,10 @@ impl<'a> Writer for WriterRelocate<'a> {
 
     fn write_address(&mut self, address: Address, size: u8) -> Result<()> {
         match address {
-            Address::Absolute(val) => self.write_word(val, size),
+            Address::Constant(val) => self.write_udata(val, size),
-            Address::Relative { symbol, addend } => {
+            Address::Symbol { symbol, addend } => {
                 match self.symbol_resolver.resolve_symbol(symbol, addend as i64) {
-                    ResolvedSymbol::PhysicalAddress(addr) => self.write_word(addr, size),
+                    ResolvedSymbol::PhysicalAddress(addr) => self.write_udata(addr, size),
                     ResolvedSymbol::Reloc { name, addend } => {
                         let offset = self.len() as u64;
                         self.relocs.push(DebugReloc {
@@ -169,7 +107,7 @@ impl<'a> Writer for WriterRelocate<'a> {
                             name,
                             addend,
                         });
-                        self.write_word(addend as u64, size)
+                        self.write_udata(addend as u64, size)
                     }
                 }
             }
@@ -185,7 +123,7 @@ impl<'a> Writer for WriterRelocate<'a> {
             name,
             addend: val as i64,
         });
-        self.write_word(val as u64, size)
+        self.write_udata(val as u64, size)
     }
 
     fn write_offset_at(
@@ -202,6 +140,6 @@ impl<'a> Writer for WriterRelocate<'a> {
             name,
             addend: val as i64,
         });
-        self.write_word_at(offset, val as u64, size)
+        self.write_udata_at(offset, val as u64, size)
     }
 }

wasmtime-environ/Cargo.toml

@@ -20,6 +20,7 @@ failure = { version = "0.1.3", default-features = false }
 failure_derive = { version = "0.1.3", default-features = false }
 indexmap = "1.0.2"
 rayon = "1.0"
+wasmtime-debug = { path = "../wasmtime-debug", default-features = false }
 
 [features]
 default = ["std"]

wasmtime-environ/src/compilation.rs

@@ -8,6 +8,7 @@ use cranelift_entity::PrimaryMap;
 use cranelift_wasm::{DefinedFuncIndex, FuncIndex, WasmError};
 use std::ops::Range;
 use std::vec::Vec;
+use wasmtime_debug::ModuleAddressMap;
 
 /// Compiled machine code: body and jump table offsets.
 #[derive(Debug, Clone)]
@@ -139,35 +140,6 @@ pub enum CompileError {
     Codegen(CodegenError),
 }
 
-/// Single address point transform.
-#[derive(Debug)]
-pub struct InstructionAddressTransform {
-    /// Original source location.
-    pub srcloc: ir::SourceLoc,
-
-    /// Generated instructions offset.
-    pub code_offset: usize,
-
-    /// Generated instructions length.
-    pub code_len: usize,
-}
-
-/// Function and its instructions transforms.
-#[derive(Debug)]
-pub struct FunctionAddressTransform {
-    /// Instructions transforms
-    pub locations: Vec<InstructionAddressTransform>,
-
-    /// Generated function body offset if applicable, otherwise 0.
-    pub body_offset: usize,
-
-    /// Generated function body length.
-    pub body_len: usize,
-}
-
-/// Function AddressTransforms collection.
-pub type AddressTransforms = PrimaryMap<DefinedFuncIndex, FunctionAddressTransform>;
-
 /// An implementation of a compiler from parsed WebAssembly module to native code.
 pub trait Compiler {
     /// Compile a parsed module with the given `TargetIsa`.
@@ -176,5 +148,5 @@ pub trait Compiler {
         function_body_inputs: PrimaryMap<DefinedFuncIndex, FunctionBodyData<'data>>,
         isa: &dyn isa::TargetIsa,
         generate_debug_info: bool,
-    ) -> Result<(Compilation, Relocations, AddressTransforms), CompileError>;
+    ) -> Result<(Compilation, Relocations, ModuleAddressMap), CompileError>;
 }

wasmtime-environ/src/cranelift.rs

@@ -1,8 +1,7 @@
 //! Support for compiling with Cranelift.
 
 use crate::compilation::{
-    AddressTransforms, CodeAndJTOffsets, Compilation, CompileError, FunctionAddressTransform,
+    CodeAndJTOffsets, Compilation, CompileError, Relocation, RelocationTarget, Relocations,
-    InstructionAddressTransform, Relocation, RelocationTarget, Relocations,
 };
 use crate::func_environ::{
     get_func_name, get_imported_memory32_grow_name, get_imported_memory32_size_name,
@@ -19,6 +18,7 @@ use cranelift_entity::PrimaryMap;
 use cranelift_wasm::{DefinedFuncIndex, FuncIndex, FuncTranslator};
 use rayon::prelude::{IntoParallelRefIterator, ParallelIterator};
 use std::vec::Vec;
+use wasmtime_debug::{FunctionAddressMap, InstructionAddressMap, ModuleAddressMap};
 
 /// Implementation of a relocation sink that just saves all the information for later
 pub struct RelocSink {
@@ -89,10 +89,7 @@ impl RelocSink {
     }
 }
 
-fn get_address_transform(
+fn get_address_transform(context: &Context, isa: &isa::TargetIsa) -> Vec<InstructionAddressMap> {
-    context: &Context,
-    isa: &isa::TargetIsa,
-) -> Vec<InstructionAddressTransform> {
     let mut result = Vec::new();
 
     let func = &context.func;
@@ -103,7 +100,7 @@ fn get_address_transform(
     for ebb in ebbs {
         for (offset, inst, size) in func.inst_offsets(ebb, &encinfo) {
             let srcloc = func.srclocs[inst];
-            result.push(InstructionAddressTransform {
+            result.push(InstructionAddressMap {
                 srcloc,
                 code_offset: offset as usize,
                 code_len: size as usize,
@@ -125,7 +122,7 @@ impl crate::compilation::Compiler for Cranelift {
         function_body_inputs: PrimaryMap<DefinedFuncIndex, FunctionBodyData<'data>>,
         isa: &dyn isa::TargetIsa,
         generate_debug_info: bool,
-    ) -> Result<(Compilation, Relocations, AddressTransforms), CompileError> {
+    ) -> Result<(Compilation, Relocations, ModuleAddressMap), CompileError> {
         let mut functions = PrimaryMap::with_capacity(function_body_inputs.len());
         let mut relocations = PrimaryMap::with_capacity(function_body_inputs.len());
         let mut address_transforms = PrimaryMap::with_capacity(function_body_inputs.len());
@@ -162,8 +159,9 @@ impl crate::compilation::Compiler for Cranelift {
                 let address_transform = if generate_debug_info {
                     let body_len = code_buf.len();
                     let at = get_address_transform(&context, isa);
-                    Some(FunctionAddressTransform {
+
-                        locations: at,
+                    Some(FunctionAddressMap {
+                        instructions: at,
                         body_offset: 0,
                         body_len,
                     })

wasmtime-environ/src/lib.rs

@@ -49,8 +49,7 @@ pub mod cranelift;
 pub mod lightbeam;
 
 pub use crate::compilation::{
-    AddressTransforms, Compilation, CompileError, Compiler, InstructionAddressTransform,
+    Compilation, CompileError, Compiler, Relocation, RelocationTarget, Relocations,
-    Relocation, RelocationTarget, Relocations,
 };
 pub use crate::cranelift::Cranelift;
 #[cfg(feature = "lightbeam")]