Auto merge of rust-lang#77117 - davidtwco:issue-34651-split-dwarf, r=…

…nagisa

cg_llvm: split dwarf support

cc rust-lang#34651

This PR adds initial support for Split DWARF to rustc, based on the implementation in Clang.

##### Current Status
This PR currently has functioning split-dwarf, running rustc with `-Zsplit-dwarf=split` when compiling a binary will produce a `dwp` alongside the binary, which contains the linked dwarf objects.

```shell-session
$ rustc -Cdebuginfo=2 -Zsplit-dwarf=split -C save-temps ./foo.rs
$ ls foo*
foo
foo.belfx9afw9cmv8.rcgu.dwo
foo.belfx9afw9cmv8.rcgu.o
foo.foo.7rcbfp3g-cgu.0.rcgu.dwo
foo.foo.7rcbfp3g-cgu.0.rcgu.o
foo.foo.7rcbfp3g-cgu.1.rcgu.dwo
foo.foo.7rcbfp3g-cgu.1.rcgu.o
foo.foo.7rcbfp3g-cgu.2.rcgu.dwo
foo.foo.7rcbfp3g-cgu.2.rcgu.o
foo.foo.7rcbfp3g-cgu.3.rcgu.dwo
foo.foo.7rcbfp3g-cgu.3.rcgu.o
foo.foo.7rcbfp3g-cgu.4.rcgu.dwo
foo.foo.7rcbfp3g-cgu.4.rcgu.o
foo.foo.7rcbfp3g-cgu.5.rcgu.dwo
foo.foo.7rcbfp3g-cgu.5.rcgu.o
foo.foo.7rcbfp3g-cgu.6.rcgu.dwo
foo.foo.7rcbfp3g-cgu.6.rcgu.o
foo.foo.7rcbfp3g-cgu.7.rcgu.dwo
foo.foo.7rcbfp3g-cgu.7.rcgu.o
foo.dwp
foo.rs
$ readelf -wi foo.foo.7rcbfp3g-cgu.0.rcgu.o
# ...
  Compilation Unit @ offset 0x90:
   Length:        0x2c (32-bit)
   Version:       4
   Abbrev Offset: 0x5b
   Pointer Size:  8
 <0><9b>: Abbrev Number: 1 (DW_TAG_compile_unit)
    <9c>   DW_AT_stmt_list   : 0xe8
    <a0>   DW_AT_comp_dir    : (indirect string, offset: 0x13b): /home/david/Projects/rust/rust0
    <a4>   DW_AT_GNU_dwo_name: (indirect string, offset: 0x15b): foo.foo.7rcbfp3g-cgu.0.rcgu.dwo
    <a8>   DW_AT_GNU_dwo_id  : 0x357472a2b032d7b9
    <b0>   DW_AT_low_pc      : 0x0
    <b8>   DW_AT_ranges      : 0x40
    <bc>   DW_AT_GNU_addr_base: 0x0
# ...
```

##### To-Do
I've opened this PR as a draft to get feedback and work out how we'd expect rustc to work when Split DWARF is requested. It might be easier to read the PR commit-by-commit.

- [ ] Add error when Split DWARF is requested on platforms where it doesn't make sense.
- [x] Determine whether or not there should be a single `dwo` output from rustc, or one per codegen-unit as exists currently.
- [x] Add tests.
- [x] Fix `single` mode - currently single mode doesn't change the invocation of `addPassesToEmitFile`, which is correct, but it also needs to change the split dwarf path provided to `createCompileUnit` and `createTargetMachine` so that it's just the final binary (currently it is still a non-existent `dwo` file).

r? `@nagisa`
cc `@michaelwoerister` `@eddyb` `@alexcrichton` `@rust-lang/wg-incr-comp`

compiler/rustc_codegen_cranelift/src/driver/aot.rs

@@ -75,6 +75,7 @@ fn emit_module(
             name,
             kind,
             object: Some(tmp_file),
+            dwarf_object: None,
             bytecode: None,
         },
         work_product,
@@ -111,6 +112,7 @@ fn reuse_workproduct_for_cgu(
         name: cgu.name().to_string(),
         kind: ModuleKind::Regular,
         object,
+        dwarf_object: None,
         bytecode: None,
     }
 }
@@ -290,6 +292,7 @@ pub(super) fn run_aot(
             name: metadata_cgu_name,
             kind: ModuleKind::Metadata,
             object: Some(tmp_file),
+            dwarf_object: None,
             bytecode: None,
         })
     } else {

compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -6,7 +6,9 @@ use crate::llvm::{self, build_string, False, True};
 use crate::{LlvmCodegenBackend, ModuleLlvm};
 use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule, ThinShared};
 use rustc_codegen_ssa::back::symbol_export;
-use rustc_codegen_ssa::back::write::{CodegenContext, FatLTOInput, ModuleConfig};
+use rustc_codegen_ssa::back::write::{
+    CodegenContext, FatLTOInput, ModuleConfig, TargetMachineFactoryConfig,
+};
 use rustc_codegen_ssa::traits::*;
 use rustc_codegen_ssa::{looks_like_rust_object_file, ModuleCodegen, ModuleKind};
 use rustc_data_structures::fx::FxHashMap;
@@ -728,20 +730,23 @@ pub unsafe fn optimize_thin_module(
     cgcx: &CodegenContext<LlvmCodegenBackend>,
 ) -> Result<ModuleCodegen<ModuleLlvm>, FatalError> {
     let diag_handler = cgcx.create_diag_handler();
-    let tm = (cgcx.tm_factory.0)().map_err(|e| write::llvm_err(&diag_handler, &e))?;
+
+    let module_name = &thin_module.shared.module_names[thin_module.idx];
+    let split_dwarf_file = cgcx
+        .output_filenames
+        .split_dwarf_filename(cgcx.split_dwarf_kind, Some(module_name.to_str().unwrap()));
+    let tm_factory_config = TargetMachineFactoryConfig { split_dwarf_file };
+    let tm =
+        (cgcx.tm_factory)(tm_factory_config).map_err(|e| write::llvm_err(&diag_handler, &e))?;
 
     // Right now the implementation we've got only works over serialized
     // modules, so we create a fresh new LLVM context and parse the module
     // into that context. One day, however, we may do this for upstream
     // crates but for locally codegened modules we may be able to reuse
     // that LLVM Context and Module.
     let llcx = llvm::LLVMRustContextCreate(cgcx.fewer_names);
-    let llmod_raw = parse_module(
-        llcx,
-        &thin_module.shared.module_names[thin_module.idx],
-        thin_module.data(),
-        &diag_handler,
-    )? as *const _;
+    let llmod_raw =
+        parse_module(llcx, &module_name, thin_module.data(), &diag_handler)? as *const _;
     let module = ModuleCodegen {
         module_llvm: ModuleLlvm { llmod_raw, llcx, tm },
         name: thin_module.name().to_string(),

compiler/rustc_codegen_llvm/src/back/write.rs

@@ -11,7 +11,10 @@ use crate::llvm_util;
 use crate::type_::Type;
 use crate::LlvmCodegenBackend;
 use crate::ModuleLlvm;
-use rustc_codegen_ssa::back::write::{BitcodeSection, CodegenContext, EmitObj, ModuleConfig};
+use rustc_codegen_ssa::back::write::{
+    BitcodeSection, CodegenContext, EmitObj, ModuleConfig, TargetMachineFactoryConfig,
+    TargetMachineFactoryFn,
+};
 use rustc_codegen_ssa::traits::*;
 use rustc_codegen_ssa::{CompiledModule, ModuleCodegen};
 use rustc_data_structures::small_c_str::SmallCStr;
@@ -20,7 +23,9 @@ use rustc_fs_util::{link_or_copy, path_to_c_string};
 use rustc_hir::def_id::LOCAL_CRATE;
 use rustc_middle::bug;
 use rustc_middle::ty::TyCtxt;
-use rustc_session::config::{self, Lto, OutputType, Passes, SanitizerSet, SwitchWithOptPath};
+use rustc_session::config::{
+    self, Lto, OutputType, Passes, SanitizerSet, SplitDwarfKind, SwitchWithOptPath,
+};
 use rustc_session::Session;
 use rustc_span::symbol::sym;
 use rustc_span::InnerSpan;
@@ -49,11 +54,31 @@ pub fn write_output_file(
     pm: &llvm::PassManager<'ll>,
     m: &'ll llvm::Module,
     output: &Path,
+    dwo_output: Option<&Path>,
     file_type: llvm::FileType,
 ) -> Result<(), FatalError> {
     unsafe {
         let output_c = path_to_c_string(output);
-        let result = llvm::LLVMRustWriteOutputFile(target, pm, m, output_c.as_ptr(), file_type);
+        let result = if let Some(dwo_output) = dwo_output {
+            let dwo_output_c = path_to_c_string(dwo_output);
+            llvm::LLVMRustWriteOutputFile(
+                target,
+                pm,
+                m,
+                output_c.as_ptr(),
+                dwo_output_c.as_ptr(),
+                file_type,
+            )
+        } else {
+            llvm::LLVMRustWriteOutputFile(
+                target,
+                pm,
+                m,
+                output_c.as_ptr(),
+                std::ptr::null(),
+                file_type,
+            )
+        };
         result.into_result().map_err(|()| {
             let msg = format!("could not write output to {}", output.display());
             llvm_err(handler, &msg)
@@ -62,12 +87,17 @@ pub fn write_output_file(
 }
 
 pub fn create_informational_target_machine(sess: &Session) -> &'static mut llvm::TargetMachine {
-    target_machine_factory(sess, config::OptLevel::No)()
+    let config = TargetMachineFactoryConfig { split_dwarf_file: None };
+    target_machine_factory(sess, config::OptLevel::No)(config)
         .unwrap_or_else(|err| llvm_err(sess.diagnostic(), &err).raise())
 }
 
-pub fn create_target_machine(tcx: TyCtxt<'_>) -> &'static mut llvm::TargetMachine {
-    target_machine_factory(&tcx.sess, tcx.backend_optimization_level(LOCAL_CRATE))()
+pub fn create_target_machine(tcx: TyCtxt<'_>, mod_name: &str) -> &'static mut llvm::TargetMachine {
+    let split_dwarf_file = tcx
+        .output_filenames(LOCAL_CRATE)
+        .split_dwarf_filename(tcx.sess.opts.debugging_opts.split_dwarf, Some(mod_name));
+    let config = TargetMachineFactoryConfig { split_dwarf_file };
+    target_machine_factory(&tcx.sess, tcx.backend_optimization_level(LOCAL_CRATE))(config)
         .unwrap_or_else(|err| llvm_err(tcx.sess.diagnostic(), &err).raise())
 }
 
@@ -122,7 +152,7 @@ fn to_llvm_code_model(code_model: Option<CodeModel>) -> llvm::CodeModel {
 pub fn target_machine_factory(
     sess: &Session,
     optlvl: config::OptLevel,
-) -> Arc<dyn Fn() -> Result<&'static mut llvm::TargetMachine, String> + Send + Sync> {
+) -> TargetMachineFactoryFn<LlvmCodegenBackend> {
     let reloc_model = to_llvm_relocation_model(sess.relocation_model());
 
     let (opt_level, _) = to_llvm_opt_settings(optlvl);
@@ -163,7 +193,10 @@ pub fn target_machine_factory(
     let use_init_array =
         !sess.opts.debugging_opts.use_ctors_section.unwrap_or(sess.target.use_ctors_section);
 
-    Arc::new(move || {
+    Arc::new(move |config: TargetMachineFactoryConfig| {
+        let split_dwarf_file = config.split_dwarf_file.unwrap_or_default();
+        let split_dwarf_file = CString::new(split_dwarf_file.to_str().unwrap()).unwrap();
+
         let tm = unsafe {
             llvm::LLVMRustCreateTargetMachine(
                 triple.as_ptr(),
@@ -182,6 +215,7 @@ pub fn target_machine_factory(
                 emit_stack_size_section,
                 relax_elf_relocations,
                 use_init_array,
+                split_dwarf_file.as_ptr(),
             )
         };
 
@@ -785,7 +819,15 @@ pub(crate) unsafe fn codegen(
                 llmod
             };
             with_codegen(tm, llmod, config.no_builtins, |cpm| {
-                write_output_file(diag_handler, tm, cpm, llmod, &path, llvm::FileType::AssemblyFile)
+                write_output_file(
+                    diag_handler,
+                    tm,
+                    cpm,
+                    llmod,
+                    &path,
+                    None,
+                    llvm::FileType::AssemblyFile,
+                )
             })?;
         }
 
@@ -794,13 +836,23 @@ pub(crate) unsafe fn codegen(
                 let _timer = cgcx
                     .prof
                     .generic_activity_with_arg("LLVM_module_codegen_emit_obj", &module.name[..]);
+
+                let dwo_out = cgcx.output_filenames.temp_path_dwo(module_name);
+                let dwo_out = match cgcx.split_dwarf_kind {
+                    // Don't change how DWARF is emitted in single mode (or when disabled).
+                    SplitDwarfKind::None | SplitDwarfKind::Single => None,
+                    // Emit (a subset of the) DWARF into a separate file in split mode.
+                    SplitDwarfKind::Split => Some(dwo_out.as_path()),
+                };
+
                 with_codegen(tm, llmod, config.no_builtins, |cpm| {
                     write_output_file(
                         diag_handler,
                         tm,
                         cpm,
                         llmod,
                         &obj_out,
+                        dwo_out,
                         llvm::FileType::ObjectFile,
                     )
                 })?;
@@ -828,6 +880,7 @@ pub(crate) unsafe fn codegen(
 
     Ok(module.into_compiled_module(
         config.emit_obj != EmitObj::None,
+        cgcx.split_dwarf_kind == SplitDwarfKind::Split,
         config.emit_bc,
         &cgcx.output_filenames,
     ))

compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs

@@ -993,9 +993,15 @@ pub fn compile_unit_metadata(
     let producer = format!("clang LLVM ({})", rustc_producer);
 
     let name_in_debuginfo = name_in_debuginfo.to_string_lossy();
-    let work_dir = tcx.sess.working_dir.0.to_string_lossy();
     let flags = "\0";
-    let split_name = "";
+
+    let out_dir = &tcx.output_filenames(LOCAL_CRATE).out_directory;
+    let split_name = tcx
+        .output_filenames(LOCAL_CRATE)
+        .split_dwarf_filename(tcx.sess.opts.debugging_opts.split_dwarf, Some(codegen_unit_name))
+        .unwrap_or_default();
+    let out_dir = out_dir.to_str().unwrap();
+    let split_name = split_name.to_str().unwrap();
 
     // FIXME(#60020):
     //
@@ -1020,8 +1026,8 @@ pub fn compile_unit_metadata(
             debug_context.builder,
             name_in_debuginfo.as_ptr().cast(),
             name_in_debuginfo.len(),
-            work_dir.as_ptr().cast(),
-            work_dir.len(),
+            out_dir.as_ptr().cast(),
+            out_dir.len(),
             llvm::ChecksumKind::None,
             ptr::null(),
             0,
@@ -1039,6 +1045,8 @@ pub fn compile_unit_metadata(
             split_name.as_ptr().cast(),
             split_name.len(),
             kind,
+            0,
+            tcx.sess.opts.debugging_opts.split_dwarf_inlining,
         );
 
         if tcx.sess.opts.debugging_opts.profile {

compiler/rustc_codegen_llvm/src/lib.rs

@@ -19,7 +19,9 @@ use back::write::{create_informational_target_machine, create_target_machine};
 pub use llvm_util::target_features;
 use rustc_ast::expand::allocator::AllocatorKind;
 use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule};
-use rustc_codegen_ssa::back::write::{CodegenContext, FatLTOInput, ModuleConfig};
+use rustc_codegen_ssa::back::write::{
+    CodegenContext, FatLTOInput, ModuleConfig, TargetMachineFactoryConfig, TargetMachineFactoryFn,
+};
 use rustc_codegen_ssa::traits::*;
 use rustc_codegen_ssa::ModuleCodegen;
 use rustc_codegen_ssa::{CodegenResults, CompiledModule};
@@ -34,7 +36,6 @@ use rustc_span::symbol::Symbol;
 
 use std::any::Any;
 use std::ffi::CStr;
-use std::sync::Arc;
 
 mod back {
     pub mod archive;
@@ -109,7 +110,7 @@ impl ExtraBackendMethods for LlvmCodegenBackend {
         &self,
         sess: &Session,
         optlvl: OptLevel,
-    ) -> Arc<dyn Fn() -> Result<&'static mut llvm::TargetMachine, String> + Send + Sync> {
+    ) -> TargetMachineFactoryFn<Self> {
         back::write::target_machine_factory(sess, optlvl)
     }
     fn target_cpu<'b>(&self, sess: &'b Session) -> &'b str {
@@ -331,7 +332,7 @@ impl ModuleLlvm {
         unsafe {
             let llcx = llvm::LLVMRustContextCreate(tcx.sess.fewer_names());
             let llmod_raw = context::create_module(tcx, llcx, mod_name) as *const _;
-            ModuleLlvm { llmod_raw, llcx, tm: create_target_machine(tcx) }
+            ModuleLlvm { llmod_raw, llcx, tm: create_target_machine(tcx, mod_name) }
         }
     }
 
@@ -352,7 +353,13 @@ impl ModuleLlvm {
         unsafe {
             let llcx = llvm::LLVMRustContextCreate(cgcx.fewer_names);
             let llmod_raw = back::lto::parse_module(llcx, name, buffer, handler)?;
-            let tm = match (cgcx.tm_factory.0)() {
+
+            let split_dwarf_file = cgcx
+                .output_filenames
+                .split_dwarf_filename(cgcx.split_dwarf_kind, Some(name.to_str().unwrap()));
+            let tm_factory_config = TargetMachineFactoryConfig { split_dwarf_file };
+
+            let tm = match (cgcx.tm_factory)(tm_factory_config) {
                 Ok(m) => m,
                 Err(e) => {
                     handler.struct_err(&e).emit();

compiler/rustc_codegen_llvm/src/llvm/ffi.rs

@@ -1830,6 +1830,8 @@ extern "C" {
         SplitName: *const c_char,
         SplitNameLen: size_t,
         kind: DebugEmissionKind,
+        DWOId: u64,
+        SplitDebugInlining: bool,
     ) -> &'a DIDescriptor;
 
     pub fn LLVMRustDIBuilderCreateFile(
@@ -2151,6 +2153,7 @@ extern "C" {
         EmitStackSizeSection: bool,
         RelaxELFRelocations: bool,
         UseInitArray: bool,
+        SplitDwarfFile: *const c_char,
     ) -> Option<&'static mut TargetMachine>;
     pub fn LLVMRustDisposeTargetMachine(T: &'static mut TargetMachine);
     pub fn LLVMRustAddBuilderLibraryInfo(
@@ -2179,6 +2182,7 @@ extern "C" {
         PM: &PassManager<'a>,
         M: &'a Module,
         Output: *const c_char,
+        DwoOutput: *const c_char,
         FileType: FileType,
     ) -> LLVMRustResult;
     pub fn LLVMRustOptimizeWithNewPassManager(