Add initial support for DWARF Fission (#8055)

* add cloning for String attributes

* use into_owned instead of to_vec to avoid a clone if possible.

* runs, but does not substitute

* show vars from c program, start cleanup

* tiday

* resolve conflicts

* remove WASI folder

* Add module_builder

add dwarf_package to state for cache

* move dwarf loading to module_environ.rs

pass the dwarf as binary as low as module_environ.rs

* pass dwarf package rather than add to debug_info

* tidy option/result nested if

* revert some toml and whitespace.

* add features cranelift,winch to module_builder and compute_artifacts

remove some `use`s

* address some feedback

remove unused 'use's

* address some feedback

remove unused 'use's

* move wat feature condition to cover whole method.

* More feedback

Another try at wat feature move

* Another try at wat feature move

* change gimli exemption version

add typed-arena exemption

* add None for c-api

* move `use` to #cfg

* fix another config build

* revert unwanted code deletion

* move inner function closer to use

* revert extra param to Module::new

* workaround object crate bug.

* add missing parameter

* add missing parameter

* Merge remote-tracking branch 'origin/main' into dwarf-att-string

# Conflicts:
#	crates/wasmtime/src/engine.rs
#	crates/wasmtime/src/runtime/module.rs
#	src/common.rs

* remove moduke

* use common gimli version of 28.1

* remove wasm feature, revert gimli version

* remove use of object for wasm dwarf

* remove NativeFile workaround, add feature for dwp loading

* sync winch signature

* revert bench api change

* add dwarf for no cache feature

* put back merge loss of module kind

* remove param from docs

* add dwarf fission lldb test

* simplify and include test source

* clang-format

* address feedback, remove packages

add docs
simplify return type

* remove Default use on ModuleTypesBuilder

* Remove an `unwrap()` and use `if let` instead

* Use `&[u8]` instead of `&Vec<u8>`

* Remove an `unwrap()` and return `None` instead

* Clean up some code in `transform_dwarf`

* Clean up some code in `replace_unit_from_split_dwarf`

* Clean up some code in `split_unit`

* Minor refactorings and documentation in `CodeBuilder`

* Restrict visibility of `dwarf_package_binary`

* Revert supply-chain folder changes

* Fix compile error on nightly

* prtest:full

* prtest:full

* prtest:full

* prtest:full

* prtest:full

* prtest:full

* prtest:full

* prtest:full

* use lldb 15

* prtest:full

* prtest:full

* load dwp when loading wasm bytes with path

* correct source file name

* remove debug

---------

Co-authored-by: Alex Crichton <alex@alexcrichton.com>

.github/workflows/main.yml

@@ -635,11 +635,15 @@ jobs:
 
     # Test debug (DWARF) related functionality.
     - run: |
-        sudo apt-get update && sudo apt-get install -y gdb lldb llvm
+        sudo apt-get update && sudo apt-get install -y gdb lldb-15 llvm
+        # woraround for https://bugs.launchpad.net/ubuntu/+source/llvm-defaults/+bug/1972855
+        sudo mkdir -p /usr/lib/local/lib/python3.10/dist-packages/lldb
+        sudo ln -s /usr/lib/llvm-15/lib/python3.10/dist-packages/lldb/* /usr/lib/python3/dist-packages/lldb/
         cargo test test_debug_dwarf -- --ignored --test-threads 1
       if: matrix.os == 'ubuntu-latest' && matrix.target == ''&& needs.determine.outputs.run-full
       env:
         RUST_BACKTRACE: 1
+        LLDB: lldb-15 # override default version, 14
 
     # NB: the test job here is explicitly lacking in cancellation of this run if
     # something goes wrong. These take the longest anyway and otherwise if

crates/cranelift/src/compiler.rs

@@ -536,6 +536,8 @@ impl wasmtime_environ::Compiler for Compiler {
         obj: &mut Object<'_>,
         translation: &ModuleTranslation<'_>,
         funcs: &PrimaryMap<DefinedFuncIndex, (SymbolId, &(dyn Any + Send))>,
+        dwarf_package_bytes: Option<&[u8]>,
+        tunables: &Tunables,
     ) -> Result<()> {
         let ofs = VMOffsets::new(
             self.isa
@@ -578,6 +580,8 @@ impl wasmtime_environ::Compiler for Compiler {
             &translation.debuginfo,
             &functions_info,
             &memory_offset,
+            dwarf_package_bytes,
+            tunables,
         )
         .with_context(|| "failed to emit DWARF debug information")?;
 

crates/cranelift/src/debug/gc.rs

@@ -67,12 +67,13 @@ impl Dependencies {
 
 pub fn build_dependencies<R: Reader<Offset = usize>>(
     dwarf: &read::Dwarf<R>,
+    dwp: &Option<read::DwarfPackage<R>>,
     at: &AddressTransform,
 ) -> read::Result<Dependencies> {
     let mut deps = Dependencies::new();
     let mut units = dwarf.units();
     while let Some(unit) = units.next()? {
-        build_unit_dependencies(unit, dwarf, at, &mut deps)?;
+        build_unit_dependencies(unit, dwarf, dwp, at, &mut deps)?;
     }
     Ok(deps)
 }
@@ -80,6 +81,7 @@ pub fn build_dependencies<R: Reader<Offset = usize>>(
 fn build_unit_dependencies<R: Reader<Offset = usize>>(
     header: read::UnitHeader<R>,
     dwarf: &read::Dwarf<R>,
+    dwp: &Option<read::DwarfPackage<R>>,
     at: &AddressTransform,
     deps: &mut Dependencies,
 ) -> read::Result<()> {
@@ -87,6 +89,17 @@ fn build_unit_dependencies<R: Reader<Offset = usize>>(
     let mut tree = unit.entries_tree(None)?;
     let root = tree.root()?;
     build_die_dependencies(root, dwarf, &unit, at, deps)?;
+
+    if let Some(dwarf_package) = dwp {
+        if let Some(dwo_id) = unit.dwo_id {
+            if let Some(cu) = dwarf_package.find_cu(dwo_id, dwarf)? {
+                if let Some(unit_header) = cu.debug_info.units().next()? {
+                    build_unit_dependencies(unit_header, &cu, &None, at, deps)?;
+                }
+            }
+        }
+    }
+
     Ok(())
 }
 

crates/cranelift/src/debug/transform/attr.rs

@@ -5,10 +5,7 @@ use super::refs::{PendingDebugInfoRefs, PendingUnitRefs};
 use super::{DebugInputContext, Reader, TransformError};
 use anyhow::{bail, Error};
 use cranelift_codegen::isa::TargetIsa;
-use gimli::{
+use gimli::{write, AttributeValue, DebugLineOffset, DebuggingInformationEntry, Unit};
-    write, AttributeValue, DebugLineOffset, DebugLineStr, DebugStr, DebugStrOffsets,
-    DebuggingInformationEntry, Unit,
-};
 
 #[derive(Debug)]
 pub(crate) enum FileAttributeContext<'a> {
@@ -124,17 +121,11 @@ where
                     return Err(TransformError("unexpected file index attribute").into());
                 }
             }
-            AttributeValue::DebugStrRef(str_offset) => {
+            AttributeValue::DebugStrRef(_) | AttributeValue::DebugStrOffsetsIndex(_) => {
-                let s = context.debug_str.get_str(str_offset)?.to_slice()?.to_vec();
+                let s = dwarf
-                write::AttributeValue::StringRef(out_strings.add(s))
+                    .attr_string(unit, attr.value().clone())?
-            }
+                    .to_string_lossy()?
-            AttributeValue::DebugStrOffsetsIndex(i) => {
+                    .into_owned();
-                let str_offset = context.debug_str_offsets.get_str_offset(
-                    gimli::Format::Dwarf32,
-                    unit.str_offsets_base,
-                    i,
-                )?;
-                let s = context.debug_str.get_str(str_offset)?.to_slice()?.to_vec();
                 write::AttributeValue::StringRef(out_strings.add(s))
             }
             AttributeValue::RangeListsRef(r) => {
@@ -308,38 +299,22 @@ pub(crate) fn clone_attr_string<R>(
     attr_value: &AttributeValue<R>,
     form: gimli::DwForm,
     unit: &Unit<R, R::Offset>,
-    debug_str: &DebugStr<R>,
+    dwarf: &gimli::Dwarf<R>,
-    debug_str_offsets: &DebugStrOffsets<R>,
-    debug_line_str: &DebugLineStr<R>,
     out_strings: &mut write::StringTable,
 ) -> Result<write::LineString, Error>
 where
     R: Reader,
 {
-    let content = match attr_value {
+    let content = dwarf
-        AttributeValue::DebugStrRef(str_offset) => {
+        .attr_string(unit, attr_value.clone())?
-            debug_str.get_str(*str_offset)?.to_slice()?.to_vec()
+        .to_string_lossy()?
-        }
+        .into_owned();
-        AttributeValue::DebugStrOffsetsIndex(i) => {
-            let str_offset = debug_str_offsets.get_str_offset(
-                gimli::Format::Dwarf32,
-                unit.str_offsets_base,
-                *i,
-            )?;
-            debug_str.get_str(str_offset)?.to_slice()?.to_vec()
-        }
-        AttributeValue::DebugLineStrRef(str_offset) => {
-            debug_line_str.get_str(*str_offset)?.to_slice()?.to_vec()
-        }
-        AttributeValue::String(b) => b.to_slice()?.to_vec(),
-        v => bail!("Unexpected attribute value: {:?}", v),
-    };
     Ok(match form {
         gimli::DW_FORM_strp => {
             let id = out_strings.add(content);
             write::LineString::StringRef(id)
         }
-        gimli::DW_FORM_string => write::LineString::String(content),
+        gimli::DW_FORM_string => write::LineString::String(content.into()),
         _ => bail!("DW_FORM_line_strp or other not supported"),
     })
 }

crates/cranelift/src/debug/transform/line_program.rs

@@ -3,7 +3,7 @@ use super::attr::clone_attr_string;
 use super::{Reader, TransformError};
 use anyhow::{Context, Error};
 use gimli::{
-    write, DebugLine, DebugLineOffset, DebugLineStr, DebugStr, DebugStrOffsets,
+    write, AttributeValue::DebugLineRef, DebugLine, DebugLineOffset, DebugStr,
     DebuggingInformationEntry, LineEncoding, Unit,
 };
 use wasmtime_environ::{DefinedFuncIndex, EntityRef};
@@ -40,48 +40,57 @@ enum ReadLineProgramState {
 }
 
 pub(crate) fn clone_line_program<R>(
+    dwarf: &gimli::Dwarf<R>,
+    skeleton_dwarf: &gimli::Dwarf<R>,
     unit: &Unit<R, R::Offset>,
     root: &DebuggingInformationEntry<R>,
+    skeleton_die: Option<&DebuggingInformationEntry<R>>,
     addr_tr: &AddressTransform,
     out_encoding: gimli::Encoding,
     debug_str: &DebugStr<R>,
-    debug_str_offsets: &DebugStrOffsets<R>,
-    debug_line_str: &DebugLineStr<R>,
     debug_line: &DebugLine<R>,
     out_strings: &mut write::StringTable,
 ) -> Result<(write::LineProgram, DebugLineOffset, Vec<write::FileId>, u64), Error>
 where
     R: Reader,
 {
-    let offset = match root.attr_value(gimli::DW_AT_stmt_list)? {
+    // Where are the "location" attributes
-        Some(gimli::AttributeValue::DebugLineRef(offset)) => offset,
+    let (location_die, location_dwarf) = match skeleton_die {
+        Some(die) => (die, skeleton_dwarf),
+        _ => (root, dwarf),
+    };
+
+    let offset = match location_die.attr_value(gimli::DW_AT_stmt_list)? {
+        Some(DebugLineRef(offset)) => offset,
+        Some(gimli::AttributeValue::SecOffset(offset)) => DebugLineOffset(offset),
         _ => {
             return Err(TransformError("Debug line offset is not found").into());
         }
     };
-    let comp_dir = root.attr_value(gimli::DW_AT_comp_dir)?;
+    let comp_dir = location_die.attr_value(gimli::DW_AT_comp_dir)?;
     let comp_name = root.attr_value(gimli::DW_AT_name)?;
     let out_comp_dir = match &comp_dir {
         Some(comp_dir) => Some(clone_attr_string(
             comp_dir,
             gimli::DW_FORM_strp,
             unit,
-            debug_str,
+            location_dwarf,
-            debug_str_offsets,
-            debug_line_str,
             out_strings,
         )?),
         None => None,
     };
-    let out_comp_name = clone_attr_string(
+    let out_comp_name = match comp_name {
-        comp_name.as_ref().context("missing DW_AT_name attribute")?,
+        Some(_) => clone_attr_string(
-        gimli::DW_FORM_strp,
+            comp_name
-        unit,
+                .as_ref()
-        debug_str,
+                .context("failed to read DW_AT_name attribute")?,
-        debug_str_offsets,
+            gimli::DW_FORM_strp,
-        debug_line_str,
+            unit,
-        out_strings,
+            dwarf,
-    )?;
+            out_strings,
+        )?,
+        _ => gimli::write::LineString::String("missing DW_AT_name attribute".into()),
+    };
 
     let program = debug_line.program(
         offset,
@@ -114,9 +123,7 @@ where
                 dir_attr,
                 gimli::DW_FORM_string,
                 unit,
-                debug_str,
+                location_dwarf,
-                debug_str_offsets,
-                debug_line_str,
                 out_strings,
             )?);
             dirs.push(dir_id);
@@ -132,9 +139,7 @@ where
                     &file_entry.path_name(),
                     gimli::DW_FORM_string,
                     unit,
-                    debug_str,
+                    location_dwarf,
-                    debug_str_offsets,
-                    debug_line_str,
                     out_strings,
                 )?,
                 dir_id,

crates/cranelift/src/debug/transform/mod.rs

@@ -7,12 +7,12 @@ use crate::CompiledFunctionsMetadata;
 use anyhow::Error;
 use cranelift_codegen::isa::TargetIsa;
 use gimli::{
-    write, DebugAddr, DebugLine, DebugLineStr, DebugStr, DebugStrOffsets, LocationLists,
+    write, DebugAddr, DebugLine, DebugStr, Dwarf, DwarfPackage, LittleEndian, LocationLists,
-    RangeLists, UnitSectionOffset,
+    RangeLists, Section, Unit, UnitSectionOffset,
 };
-use std::collections::HashSet;
+use std::{collections::HashSet, fmt::Debug};
 use thiserror::Error;
-use wasmtime_environ::DebugInfoData;
+use wasmtime_environ::{DebugInfoData, ModuleTranslation, Tunables};
 
 pub use address_transform::AddressTransform;
 
@@ -26,9 +26,12 @@ mod simulate;
 mod unit;
 mod utils;
 
-pub(crate) trait Reader: gimli::Reader<Offset = usize> {}
+pub(crate) trait Reader: gimli::Reader<Offset = usize> + Send + Sync {}
 
-impl<'input, Endian> Reader for gimli::EndianSlice<'input, Endian> where Endian: gimli::Endianity {}
+impl<'input, Endian> Reader for gimli::EndianSlice<'input, Endian> where
+    Endian: gimli::Endianity + Send + Sync
+{
+}
 
 #[derive(Error, Debug)]
 #[error("Debug info transform error: {0}")]
@@ -39,8 +42,6 @@ where
     R: Reader,
 {
     debug_str: &'a DebugStr<R>,
-    debug_str_offsets: &'a DebugStrOffsets<R>,
-    debug_line_str: &'a DebugLineStr<R>,
     debug_line: &'a DebugLine<R>,
     debug_addr: &'a DebugAddr<R>,
     rnglists: &'a RangeLists<R>,
@@ -48,19 +49,113 @@ where
     reachable: &'a HashSet<UnitSectionOffset>,
 }
 
-pub fn transform_dwarf(
+fn load_dwp<'data>(
+    translation: ModuleTranslation<'data>,
+    buffer: &'data [u8],
+) -> anyhow::Result<DwarfPackage<gimli::EndianSlice<'data, gimli::LittleEndian>>> {
+    let endian_slice = gimli::EndianSlice::new(buffer, LittleEndian);
+
+    let dwarf_package = DwarfPackage::load(
+        |id| -> anyhow::Result<_> {
+            let slice = match id {
+                gimli::SectionId::DebugAbbrev => {
+                    translation.debuginfo.dwarf.debug_abbrev.reader().slice()
+                }
+                gimli::SectionId::DebugInfo => {
+                    translation.debuginfo.dwarf.debug_info.reader().slice()
+                }
+                gimli::SectionId::DebugLine => {
+                    translation.debuginfo.dwarf.debug_line.reader().slice()
+                }
+                gimli::SectionId::DebugStr => {
+                    translation.debuginfo.dwarf.debug_str.reader().slice()
+                }
+                gimli::SectionId::DebugStrOffsets => translation
+                    .debuginfo
+                    .dwarf
+                    .debug_str_offsets
+                    .reader()
+                    .slice(),
+                gimli::SectionId::DebugLoc => translation.debuginfo.debug_loc.reader().slice(),
+                gimli::SectionId::DebugLocLists => {
+                    translation.debuginfo.debug_loclists.reader().slice()
+                }
+                gimli::SectionId::DebugRngLists => {
+                    translation.debuginfo.debug_rnglists.reader().slice()
+                }
+                gimli::SectionId::DebugTypes => {
+                    translation.debuginfo.dwarf.debug_types.reader().slice()
+                }
+                gimli::SectionId::DebugCuIndex => {
+                    translation.debuginfo.debug_cu_index.reader().slice()
+                }
+                gimli::SectionId::DebugTuIndex => {
+                    translation.debuginfo.debug_tu_index.reader().slice()
+                }
+                _ => &buffer,
+            };
+
+            Ok(gimli::EndianSlice::new(slice, gimli::LittleEndian))
+        },
+        endian_slice,
+    )?;
+
+    Ok(dwarf_package)
+}
+
+/// Attempts to load a DWARF package using the passed bytes.
+fn read_dwarf_package_from_bytes<'data>(
+    dwp_bytes: &'data [u8],
+    buffer: &'data [u8],
+    tunables: &Tunables,
+) -> Option<DwarfPackage<gimli::EndianSlice<'data, gimli::LittleEndian>>> {
+    let mut validator = wasmparser::Validator::new();
+    let parser = wasmparser::Parser::new(0);
+    let mut types = wasmtime_environ::ModuleTypesBuilder::new(&validator);
+    let translation =
+        match wasmtime_environ::ModuleEnvironment::new(tunables, &mut validator, &mut types)
+            .translate(parser, dwp_bytes)
+        {
+            Ok(translation) => translation,
+            Err(e) => {
+                log::warn!("failed to parse wasm dwarf package: {e:?}");
+                return None;
+            }
+        };
+
+    match load_dwp(translation, buffer) {
+        Ok(package) => Some(package),
+        Err(err) => {
+            log::warn!("Failed to load Dwarf package {}", err);
+            None
+        }
+    }
+}
+
+pub fn transform_dwarf<'data>(
     isa: &dyn TargetIsa,
     di: &DebugInfoData,
     funcs: &CompiledFunctionsMetadata,
     memory_offset: &ModuleMemoryOffset,
+    dwarf_package_bytes: Option<&[u8]>,
+    tunables: &Tunables,
 ) -> Result<write::Dwarf, Error> {
     let addr_tr = AddressTransform::new(funcs, &di.wasm_file);
-    let reachable = build_dependencies(&di.dwarf, &addr_tr)?.get_reachable();
+
+    let buffer = Vec::new();
+
+    let dwarf_package = dwarf_package_bytes
+        .map(
+            |bytes| -> Option<DwarfPackage<gimli::EndianSlice<'_, gimli::LittleEndian>>> {
+                read_dwarf_package_from_bytes(bytes, &buffer, tunables)
+            },
+        )
+        .flatten();
+
+    let reachable = build_dependencies(&di.dwarf, &dwarf_package, &addr_tr)?.get_reachable();
 
     let context = DebugInputContext {
         debug_str: &di.dwarf.debug_str,
-        debug_str_offsets: &di.dwarf.debug_str_offsets,
-        debug_line_str: &di.dwarf.debug_line_str,
         debug_line: &di.dwarf.debug_line,
         debug_addr: &di.dwarf.debug_addr,
         rnglists: &di.dwarf.ranges,
@@ -84,11 +179,29 @@ pub fn transform_dwarf(
 
     let mut translated = HashSet::new();
     let mut iter = di.dwarf.debug_info.units();
+
     while let Some(header) = iter.next().unwrap_or(None) {
         let unit = di.dwarf.unit(header)?;
+
+        let mut resolved_unit = None;
+        let mut split_dwarf = None;
+
+        if let gimli::UnitType::Skeleton(_dwo_id) = unit.header.type_() {
+            if let Some(dwarf_package) = &dwarf_package {
+                if let Some((fused, fused_dwarf)) =
+                    replace_unit_from_split_dwarf(&unit, dwarf_package, &di.dwarf)
+                {
+                    resolved_unit = Some(fused);
+                    split_dwarf = Some(fused_dwarf);
+                }
+            }
+        }
+
         if let Some((id, ref_map, pending_refs)) = clone_unit(
             &di.dwarf,
-            unit,
+            &unit,
+            resolved_unit.as_ref(),
+            split_dwarf.as_ref(),
             &context,
             &addr_tr,
             funcs,
@@ -124,3 +237,20 @@ pub fn transform_dwarf(
         strings: out_strings,
     })
 }
+
+fn replace_unit_from_split_dwarf<'a>(
+    unit: &'a Unit<gimli::EndianSlice<'a, gimli::LittleEndian>, usize>,
+    dwp: &DwarfPackage<gimli::EndianSlice<'a, gimli::LittleEndian>>,
+    parent: &Dwarf<gimli::EndianSlice<'a, gimli::LittleEndian>>,
+) -> Option<(
+    Unit<gimli::EndianSlice<'a, gimli::LittleEndian>, usize>,
+    Dwarf<gimli::EndianSlice<'a, gimli::LittleEndian>>,
+)> {
+    let dwo_id = unit.dwo_id?;
+    let split_unit_dwarf = dwp.find_cu(dwo_id, parent).ok()??;
+    let unit_header = split_unit_dwarf.debug_info.units().next().ok()??;
+    Some((
+        split_unit_dwarf.unit(unit_header).unwrap(),
+        split_unit_dwarf,
+    ))
+}

crates/cranelift/src/debug/transform/unit.rs

@@ -5,7 +5,7 @@ use super::line_program::clone_line_program;
 use super::range_info_builder::RangeInfoBuilder;
 use super::refs::{PendingDebugInfoRefs, PendingUnitRefs, UnitRefsMap};
 use super::utils::{add_internal_types, append_vmctx_info, get_function_frame_info};
-use super::{DebugInputContext, Reader, TransformError};
+use super::{DebugInputContext, Reader};
 use crate::debug::ModuleMemoryOffset;
 use crate::CompiledFunctionsMetadata;
 use anyhow::{Context, Error};
@@ -256,7 +256,9 @@ fn is_dead_code<R: Reader>(entry: &DebuggingInformationEntry<R>) -> bool {
 
 pub(crate) fn clone_unit<'a, R>(
     dwarf: &gimli::Dwarf<R>,
-    unit: Unit<R, R::Offset>,
+    unit: &Unit<R, R::Offset>,
+    split_unit: Option<&Unit<R, R::Offset>>,
+    split_dwarf: Option<&gimli::Dwarf<R>>,
     context: &DebugInputContext<R>,
     addr_tr: &'a AddressTransform,
     funcs: &'a CompiledFunctionsMetadata,
@@ -275,20 +277,35 @@ where
     let mut pending_di_refs = PendingDebugInfoRefs::new();
     let mut stack = Vec::new();
 
+    let mut program_unit = unit;
+    let mut skeleton_die = None;
+
+    // Get entries in outer scope to avoid borrowing on short lived temporary.
+    let mut skeleton_entries = unit.entries();
+    if let Some(unit) = split_unit {
+        program_unit = unit;
+
+        // From the spec, a skeleton unit has no children so we can assume the first, and only, entry is the DW_TAG_skeleton_unit (https://dwarfstd.org/doc/DWARF5.pdf).
+        if let Some(die_tuple) = skeleton_entries.next_dfs()? {
+            skeleton_die = Some(die_tuple.1);
+        }
+    }
+
     // Iterate over all of this compilation unit's entries.
-    let mut entries = unit.entries();
+    let mut entries = program_unit.entries();
     let (mut comp_unit, unit_id, file_map, file_index_base, cu_low_pc, wp_die_id, vmctx_die_id) =
         if let Some((depth_delta, entry)) = entries.next_dfs()? {
             assert_eq!(depth_delta, 0);
             let (out_line_program, debug_line_offset, file_map, file_index_base) =
                 clone_line_program(
-                    &unit,
+                    split_dwarf.unwrap_or(dwarf),
+                    dwarf,
+                    program_unit,
                     entry,
+                    skeleton_die,
                     addr_tr,
                     out_encoding,
                     context.debug_str,
-                    context.debug_str_offsets,
-                    context.debug_line_str,
                     context.debug_line,
                     out_strings,
                 )?;
@@ -299,23 +316,11 @@ where
 
                 let root_id = comp_unit.root();
                 die_ref_map.insert(entry.offset(), root_id);
-
+                let cu_low_pc = unit.low_pc;
-                let cu_low_pc = if let Some(AttributeValue::Addr(addr)) =
-                    entry.attr_value(gimli::DW_AT_low_pc)?
-                {
-                    addr
-                } else if let Some(AttributeValue::DebugAddrIndex(i)) =
-                    entry.attr_value(gimli::DW_AT_low_pc)?
-                {
-                    context.debug_addr.get_address(4, unit.addr_base, i)?
-                } else {
-                    // FIXME? return Err(TransformError("No low_pc for unit header").into());
-                    0
-                };
 
                 clone_die_attributes(
-                    dwarf,
+                    split_dwarf.unwrap_or(dwarf),
-                    &unit,
+                    &program_unit,
                     entry,
                     context,
                     addr_tr,
@@ -346,7 +351,9 @@ where
                     vmctx_die_id,
                 )
             } else {
-                return Err(TransformError("Unexpected unit header").into());
+                // Can happen when the DWARF is split and we dont have the package/dwo files.
+                // This is a better user experience than errorring.
+                return Ok(None); // empty:
             }
         } else {
             return Ok(None); // empty
@@ -469,7 +476,7 @@ where
         die_ref_map.insert(entry.offset(), die_id);
 
         clone_die_attributes(
-            dwarf,
+            split_dwarf.unwrap_or(dwarf),
             &unit,
             entry,
             context,

crates/cranelift/src/debug/write_debuginfo.rs

@@ -179,8 +179,17 @@ pub fn emit_dwarf<'a>(
     debuginfo_data: &DebugInfoData,
     funcs: &CompiledFunctionsMetadata,
     memory_offset: &ModuleMemoryOffset,
+    dwarf_package_bytes: Option<&[u8]>,
+    tunables: &wasmtime_environ::Tunables,
 ) -> anyhow::Result<Vec<DwarfSection>> {
-    let dwarf = transform_dwarf(isa, debuginfo_data, funcs, memory_offset)?;
+    let dwarf = transform_dwarf(
+        isa,
+        debuginfo_data,
+        funcs,
+        memory_offset,
+        dwarf_package_bytes,
+        tunables,
+    )?;
     let frame_table = create_frame_table(isa, funcs);
     let sections = emit_dwarf_sections(isa, dwarf, frame_table)?;
     Ok(sections)

crates/environ/src/compile/mod.rs

@@ -377,6 +377,8 @@ pub trait Compiler: Send + Sync {
         obj: &mut Object<'_>,
         translation: &ModuleTranslation<'_>,
         funcs: &PrimaryMap<DefinedFuncIndex, (SymbolId, &(dyn Any + Send))>,
+        dwarf_package_bytes: Option<&[u8]>,
+        tunables: &Tunables,
     ) -> Result<()>;
 
     /// Creates a new System V Common Information Entry for the ISA.

crates/environ/src/compile/module_artifacts.rs

@@ -2,8 +2,8 @@
 //! with `bincode` as part of a module's compilation process.
 
 use crate::{
-    obj, CompiledFunctionInfo, CompiledModuleInfo, DefinedFuncIndex, FunctionLoc, FunctionName,
+    obj, CompiledFunctionInfo, CompiledModuleInfo, DebugInfoData, DefinedFuncIndex, FunctionLoc,
-    MemoryInitialization, Metadata, ModuleTranslation, PrimaryMap, Tunables,
+    FunctionName, MemoryInitialization, Metadata, ModuleTranslation, PrimaryMap, Tunables,
 };
 use anyhow::{bail, Result};
 use object::write::{Object, SectionId, StandardSegment, WritableBuffer};
@@ -57,6 +57,31 @@ impl<'a> ObjectBuilder<'a> {
         }
     }
 
+    /// Insert the wasm raw wasm-based debuginfo into the output.
+    /// Note that this is distinct from the native debuginfo
+    /// possibly generated by the native compiler, hence these sections
+    /// getting wasm-specific names.
+    pub fn push_debuginfo(
+        &mut self,
+        dwarf: &mut Vec<(u8, Range<u64>)>,
+        debuginfo: &DebugInfoData<'_>,
+    ) {
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_abbrev);
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_addr);
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_aranges);
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_info);
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_line);
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_line_str);
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_str);
+        self.push_debug(dwarf, &debuginfo.dwarf.debug_str_offsets);
+        self.push_debug(dwarf, &debuginfo.debug_ranges);
+        self.push_debug(dwarf, &debuginfo.debug_rnglists);
+        self.push_debug(dwarf, &debuginfo.debug_cu_index);
+
+        // Sort this for binary-search-lookup later in `symbolize_context`.
+        dwarf.sort_by_key(|(id, _)| *id);
+    }
+
     /// Completes compilation of the `translation` specified, inserting
     /// everything necessary into the `Object` being built.
     ///
@@ -188,19 +213,8 @@ impl<'a> ObjectBuilder<'a> {
         // getting wasm-specific names.
         let mut dwarf = Vec::new();
         if self.tunables.parse_wasm_debuginfo {
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_abbrev);
+            self.push_debuginfo(&mut dwarf, &debuginfo);
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_addr);
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_aranges);
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_info);
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_line);
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_line_str);
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_str);
-            self.push_debug(&mut dwarf, &debuginfo.dwarf.debug_str_offsets);
-            self.push_debug(&mut dwarf, &debuginfo.debug_ranges);
-            self.push_debug(&mut dwarf, &debuginfo.debug_rnglists);
         }
-        // Sort this for binary-search-lookup later in `symbolize_context`.
-        dwarf.sort_by_key(|(id, _)| *id);
 
         Ok(CompiledModuleInfo {
             module,

crates/environ/src/module_environ.rs

@@ -122,10 +122,12 @@ pub struct DebugInfoData<'a> {
     pub dwarf: Dwarf<'a>,
     pub name_section: NameSection<'a>,
     pub wasm_file: WasmFileInfo,
-    debug_loc: gimli::DebugLoc<Reader<'a>>,
+    pub debug_loc: gimli::DebugLoc<Reader<'a>>,
-    debug_loclists: gimli::DebugLocLists<Reader<'a>>,
+    pub debug_loclists: gimli::DebugLocLists<Reader<'a>>,
     pub debug_ranges: gimli::DebugRanges<Reader<'a>>,
     pub debug_rnglists: gimli::DebugRngLists<Reader<'a>>,
+    pub debug_cu_index: gimli::DebugCuIndex<Reader<'a>>,
+    pub debug_tu_index: gimli::DebugTuIndex<Reader<'a>>,
 }
 
 #[allow(missing_docs)]
@@ -782,7 +784,7 @@ and for re-adding support for interface types you can see this issue:
     }
 
     fn register_dwarf_section(&mut self, section: &CustomSectionReader<'data>) {
-        let name = section.name();
+        let name = section.name().trim_end_matches(".dwo");
         if !name.starts_with(".debug_") {
             return;
         }
@@ -800,7 +802,9 @@ and for re-adding support for interface types you can see this issue:
             // `gimli::Dwarf` fields.
             ".debug_abbrev" => dwarf.debug_abbrev = gimli::DebugAbbrev::new(data, endian),
             ".debug_addr" => dwarf.debug_addr = gimli::DebugAddr::from(slice),
-            ".debug_info" => dwarf.debug_info = gimli::DebugInfo::new(data, endian),
+            ".debug_info" => {
+                dwarf.debug_info = gimli::DebugInfo::new(data, endian);
+            }
             ".debug_line" => dwarf.debug_line = gimli::DebugLine::new(data, endian),
             ".debug_line_str" => dwarf.debug_line_str = gimli::DebugLineStr::from(slice),
             ".debug_str" => dwarf.debug_str = gimli::DebugStr::new(data, endian),
@@ -818,9 +822,12 @@ and for re-adding support for interface types you can see this issue:
             ".debug_ranges" => info.debug_ranges = gimli::DebugRanges::new(data, endian),
             ".debug_rnglists" => info.debug_rnglists = gimli::DebugRngLists::new(data, endian),
 
+            // DWARF package fields
+            ".debug_cu_index" => info.debug_cu_index = gimli::DebugCuIndex::new(data, endian),
+            ".debug_tu_index" => info.debug_tu_index = gimli::DebugTuIndex::new(data, endian),
+
             // We don't use these at the moment.
             ".debug_aranges" | ".debug_pubnames" | ".debug_pubtypes" => return,
-
             other => {
                 log::warn!("unknown debug section `{}`", other);
                 return;

crates/wasmtime/src/compile.rs

@@ -29,6 +29,7 @@ use std::{
     collections::{btree_map, BTreeMap, BTreeSet, HashMap, HashSet},
     mem,
 };
+
 #[cfg(feature = "component-model")]
 use wasmtime_environ::component::Translator;
 use wasmtime_environ::{
@@ -61,6 +62,7 @@ pub use self::runtime::finish_object;
 pub(crate) fn build_artifacts<T: FinishedObject>(
     engine: &Engine,
     wasm: &[u8],
+    dwarf_package: Option<&[u8]>,
 ) -> Result<(T, Option<(CompiledModuleInfo, ModuleTypes)>)> {
     let tunables = engine.tunables();
 
@@ -101,6 +103,7 @@ pub(crate) fn build_artifacts<T: FinishedObject>(
         engine,
         compiled_funcs,
         std::iter::once(translation).collect(),
+        dwarf_package,
     )?;
 
     let info = compilation_artifacts.unwrap_as_module_info();
@@ -118,9 +121,10 @@ pub(crate) fn build_artifacts<T: FinishedObject>(
 /// The output artifact here is the serialized object file contained within
 /// an owned mmap along with metadata about the compilation itself.
 #[cfg(feature = "component-model")]
-pub(crate) fn build_component_artifacts<T: FinishedObject>(
+pub(crate) fn build_component_artifacts<'a, T: FinishedObject>(
     engine: &Engine,
     binary: &[u8],
+    _dwarf_package: Option<&[u8]>,
 ) -> Result<(T, Option<wasmtime_environ::component::ComponentArtifacts>)> {
     use wasmtime_environ::component::{
         CompiledComponentInfo, ComponentArtifacts, ComponentTypesBuilder,
@@ -159,6 +163,7 @@ pub(crate) fn build_component_artifacts<T: FinishedObject>(
         engine,
         compiled_funcs,
         module_translations,
+        None, // TODO: Support dwarf packages for components.
     )?;
     let (types, ty) = types.finish(
         &compilation_artifacts.modules,
@@ -649,6 +654,7 @@ impl FunctionIndices {
         engine: &'a Engine,
         compiled_funcs: Vec<(String, Box<dyn Any + Send>)>,
         translations: PrimaryMap<StaticModuleIndex, ModuleTranslation<'_>>,
+        dwarf_package_bytes: Option<&[u8]>,
     ) -> Result<(wasmtime_environ::ObjectBuilder<'a>, Artifacts)> {
         // Append all the functions to the ELF file.
         //
@@ -713,7 +719,13 @@ impl FunctionIndices {
                     })
                     .collect();
                 if !funcs.is_empty() {
-                    compiler.append_dwarf(&mut obj, translation, &funcs)?;
+                    compiler.append_dwarf(
+                        &mut obj,
+                        translation,
+                        &funcs,
+                        dwarf_package_bytes,
+                        tunables,
+                    )?;
                 }
             }
         }

crates/wasmtime/src/compile/code_builder.rs

@@ -40,6 +40,8 @@ pub struct CodeBuilder<'a> {
     pub(super) engine: &'a Engine,
     wasm: Option<Cow<'a, [u8]>>,
     wasm_path: Option<Cow<'a, Path>>,
+    dwarf_package: Option<Cow<'a, [u8]>>,
+    dwarf_package_path: Option<Cow<'a, Path>>,
     wat: bool,
 }
 
@@ -51,6 +53,8 @@ impl<'a> CodeBuilder<'a> {
             engine,
             wasm: None,
             wasm_path: None,
+            dwarf_package: None,
+            dwarf_package_path: None,
             wat: cfg!(feature = "wat"),
         }
     }
@@ -76,6 +80,11 @@ impl<'a> CodeBuilder<'a> {
         }
         self.wasm = Some(wasm_bytes.into());
         self.wasm_path = wasm_path.map(|p| p.into());
+
+        if self.wasm_path.is_some() {
+            self.dwarf_package_from_wasm_path()?;
+        }
+
         Ok(self)
     }
 
@@ -104,6 +113,10 @@ impl<'a> CodeBuilder<'a> {
     /// either as a WebAssembly binary or as a WebAssembly text file. The
     /// contents are inspected to do this, the file extension is not consulted.
     ///
+    /// A DWARF package file will be probed using the root of `file` and with a
+    /// `.dwp` extension.  If found, it will be loaded and DWARF fusion
+    /// performed.
+    ///
     /// # Errors
     ///
     /// If wasm bytes have already been configured via a call to this method or
@@ -111,6 +124,9 @@ impl<'a> CodeBuilder<'a> {
     ///
     /// If `file` can't be read or an error happens reading it then that will
     /// also be returned.
+    ///
+    /// If DWARF fusion is performed and the DWARF packaged file cannot be read
+    /// then an error will be returned.
     pub fn wasm_file(&mut self, file: &'a Path) -> Result<&mut Self> {
         if self.wasm.is_some() {
             bail!("cannot call `wasm` or `wasm_file` twice");
@@ -119,6 +135,8 @@ impl<'a> CodeBuilder<'a> {
             .with_context(|| format!("failed to read input file: {}", file.display()))?;
         self.wasm = Some(wasm.into());
         self.wasm_path = Some(file.into());
+        self.dwarf_package_from_wasm_path()?;
+
         Ok(self)
     }
 
@@ -139,6 +157,61 @@ impl<'a> CodeBuilder<'a> {
         Ok((&wasm[..]).into())
     }
 
+    /// Explicitly specify DWARF `.dwp` path.
+    ///
+    /// # Errors
+    ///
+    /// This method will return an error if the `.dwp` file has already been set
+    /// through [`CodeBuilder::dwarf_package`] or auto-detection in
+    /// [`CodeBuilder::wasm_file`].
+    ///
+    /// This method will also return an error if `file` cannot be read.
+    pub fn dwarf_package_file(&mut self, file: &Path) -> Result<&mut Self> {
+        if self.dwarf_package.is_some() {
+            bail!("cannot call `dwarf_package` or `dwarf_package_file` twice");
+        }
+
+        let dwarf_package = std::fs::read(file)
+            .with_context(|| format!("failed to read dwarf input file: {}", file.display()))?;
+        self.dwarf_package_path = Some(Cow::Owned(file.to_owned()));
+        self.dwarf_package = Some(dwarf_package.into());
+
+        Ok(self)
+    }
+
+    fn dwarf_package_from_wasm_path(&mut self) -> Result<&mut Self> {
+        let dwarf_package_path_buf = self.wasm_path.as_ref().unwrap().with_extension("dwp");
+        if dwarf_package_path_buf.exists() {
+            return self.dwarf_package_file(dwarf_package_path_buf.as_path());
+        }
+
+        Ok(self)
+    }
+
+    /// Gets the DWARF package.
+    pub(super) fn dwarf_package_binary(&self) -> Option<&[u8]> {
+        return self.dwarf_package.as_deref();
+    }
+
+    /// Set the DWARF package binary.
+    ///
+    /// Initializes `dwarf_package` from `dwp_bytes` in preparation for
+    /// DWARF fusion. Allows the DWARF package to be supplied as a byte array
+    /// when the file probing performed in `wasm_file` is not appropriate.
+    ///
+    /// # Errors
+    ///
+    /// Returns an error if the `*.dwp` file is already set via auto-probing in
+    /// [`CodeBuilder::wasm_file`] or explicitly via
+    /// [`CodeBuilder::dwarf_package_file`].
+    pub fn dwarf_package(&mut self, dwp_bytes: &'a [u8]) -> Result<&mut Self> {
+        if self.dwarf_package.is_some() {
+            bail!("cannot call `dwarf_package` or `dwarf_package_file` twice");
+        }
+        self.dwarf_package = Some(dwp_bytes.into());
+        Ok(self)
+    }
+
     /// Finishes this compilation and produces a serialized list of bytes.
     ///
     /// This method requires that either [`CodeBuilder::wasm`] or
@@ -157,7 +230,8 @@ impl<'a> CodeBuilder<'a> {
     /// compilation-related error is encountered.
     pub fn compile_module_serialized(&self) -> Result<Vec<u8>> {
         let wasm = self.wasm_binary()?;
-        let (v, _) = super::build_artifacts(self.engine, &wasm)?;
+        let dwarf_package = self.dwarf_package_binary();
+        let (v, _) = super::build_artifacts(self.engine, &wasm, dwarf_package.as_deref())?;
         Ok(v)
     }
 
@@ -168,7 +242,7 @@ impl<'a> CodeBuilder<'a> {
     #[cfg_attr(docsrs, doc(cfg(feature = "component-model")))]
     pub fn compile_component_serialized(&self) -> Result<Vec<u8>> {
         let bytes = self.wasm_binary()?;
-        let (v, _) = super::build_component_artifacts(self.engine, &bytes)?;
+        let (v, _) = super::build_component_artifacts(self.engine, &bytes, None)?;
         Ok(v)
     }
 }

crates/wasmtime/src/compile/runtime.rs

@@ -11,9 +11,10 @@ use wasmtime_runtime::MmapVec;
 impl<'a> CodeBuilder<'a> {
     fn compile_cached<T>(
         &self,
-        build_artifacts: fn(&Engine, &[u8]) -> Result<(MmapVecWrapper, Option<T>)>,
+        build_artifacts: fn(&Engine, &[u8], Option<&[u8]>) -> Result<(MmapVecWrapper, Option<T>)>,
     ) -> Result<(Arc<CodeMemory>, Option<T>)> {
         let wasm = self.wasm_binary()?;
+        let dwarf_package = self.dwarf_package_binary();
 
         self.engine
             .check_compatible_with_native_host()
@@ -24,6 +25,7 @@ impl<'a> CodeBuilder<'a> {
             let state = (
                 HashedEngineCompileEnv(self.engine),
                 &wasm,
+                &dwarf_package,
                 // Don't hash this as it's just its own "pure" function pointer.
                 NotHashed(build_artifacts),
             );
@@ -32,15 +34,18 @@ impl<'a> CodeBuilder<'a> {
                     .get_data_raw(
                         &state,
                         // Cache miss, compute the actual artifacts
-                        |(engine, wasm, build_artifacts)| -> Result<_> {
+                        |(engine, wasm, dwarf_package, build_artifacts)| -> Result<_> {
-                            let (mmap, info) = (build_artifacts.0)(engine.0, wasm)?;
+                            let (mmap, info) =
+                                (build_artifacts.0)(engine.0, wasm, dwarf_package.as_deref())?;
                             let code = publish_mmap(mmap.0)?;
                             Ok((code, info))
                         },
                         // Implementation of how to serialize artifacts
-                        |(_engine, _wasm, _), (code, _info_and_types)| Some(code.mmap().to_vec()),
+                        |(_engine, _wasm, _, _), (code, _info_and_types)| {
+                            Some(code.mmap().to_vec())
+                        },
                         // Cache hit, deserialize the provided artifacts
-                        |(engine, wasm, _), serialized_bytes| {
+                        |(engine, wasm, _, _), serialized_bytes| {
                             let kind = if wasmparser::Parser::is_component(&wasm) {
                                 ObjectKind::Component
                             } else {
@@ -55,7 +60,8 @@ impl<'a> CodeBuilder<'a> {
 
         #[cfg(not(feature = "cache"))]
         {
-            let (mmap, info_and_types) = build_artifacts(self.engine, &wasm)?;
+            let (mmap, info_and_types) =
+                build_artifacts(self.engine, &wasm, dwarf_package.as_deref())?;
             let code = publish_mmap(mmap.0)?;
             return Ok((code, info_and_types));
         }

crates/wasmtime/src/runtime/module.rs

@@ -23,7 +23,6 @@ use wasmtime_runtime::{
     CompiledModuleId, MemoryImage, MmapVec, ModuleMemoryImages, VMArrayCallFunction,
     VMNativeCallFunction, VMWasmCallFunction,
 };
-
 mod registry;
 
 pub use registry::{

crates/winch/src/compiler.rs

@@ -233,6 +233,8 @@ impl wasmtime_environ::Compiler for Compiler {
         _obj: &mut Object<'_>,
         _translation: &ModuleTranslation<'_>,
         _funcs: &PrimaryMap<DefinedFuncIndex, (SymbolId, &(dyn Any + Send))>,
+        _dwarf_package: Option<&[u8]>,
+        _tunables: &wasmtime_environ::Tunables,
     ) -> Result<()> {
         todo!()
     }

tests/all/debug/lldb.rs

@@ -247,3 +247,51 @@ check: exited with status
     )?;
     Ok(())
 }
+
+/* C program used for this test, dwarf_fission.c, compiled with `emcc dwarf_fission.c -o dwarf_fission.wasm -gsplit-dwarf -gdwarf-5 -gpubnames -sWASM_BIGINT`:
+#include <stdio.h>
+
+int main()
+{
+    int i = 1;
+    i++;
+    return i - 2;
+}
+ */
+#[test]
+#[ignore]
+#[cfg(all(
+    any(target_os = "linux", target_os = "macos"),
+    target_pointer_width = "64"
+))]
+pub fn test_debug_dwarf5_fission_lldb() -> Result<()> {
+    let output = lldb_with_script(
+        &[
+            "-Ccache=n",
+            "-Ddebug-info",
+            "tests/all/debug/testsuite/dwarf_fission.wasm",
+        ],
+        r#"breakpoint set --file dwarf_fission.c --line 6
+r
+fr v
+s
+fr v
+c"#,
+    )?;
+
+    check_lldb_output(
+        &output,
+        r#"
+check: Breakpoint 1: no locations (pending)
+check: Unable to resolve breakpoint to any actual locations.
+check: 1 location added to breakpoint 1
+check: stop reason = breakpoint 1.1
+check: i = 1
+check: stop reason = step in
+check: i = 2
+check: resuming
+check: exited with status = 0
+"#,
+    )?;
+    Ok(())
+}

tests/all/debug/testsuite/dwarf_fission.c

@@ -0,0 +1,7 @@
+#include <stdio.h>
+
+int main() {
+  int i = 1;
+  i++;
+  return i - 2;
+}