From 868bdde25b030e0b71a29a5dbc04a891036e702e Mon Sep 17 00:00:00 2001 From: Sebastian Poeplau Date: Fri, 1 Aug 2025 12:39:50 +0200 Subject: [PATCH] Preserve the .debug_gdb_scripts section Make sure that compiler and linker don't optimize the section's contents away by adding the global holding the data to "llvm.used". The volatile load in the main shim is retained because "llvm.used", which translates to SHF_GNU_RETAIN on ELF targets, requires a reasonably recent linker; emitting the volatile load ensures compatibility with older linkers, at least when libstd is used. Pretty printers in dylib dependencies are now emitted by the main crate instead of the dylib; apart from matching how rlibs are handled, this approach has the advantage that `omit_gdb_pretty_printer_section` keeps working with dylib dependencies. --- compiler/rustc_codegen_gcc/src/debuginfo.rs | 3 +- compiler/rustc_codegen_llvm/src/base.rs | 12 +++--- .../rustc_codegen_llvm/src/debuginfo/gdb.rs | 30 ++++++++------- .../rustc_codegen_llvm/src/debuginfo/mod.rs | 38 +++++++++++-------- compiler/rustc_codegen_ssa/src/base.rs | 10 +---- .../rustc_codegen_ssa/src/traits/debuginfo.rs | 2 +- tests/codegen-llvm/gdb_debug_script_load.rs | 4 +- tests/run-make/symbols-all-mangled/rmake.rs | 35 ++++++++++------- 8 files changed, 73 insertions(+), 61 deletions(-) diff --git a/compiler/rustc_codegen_gcc/src/debuginfo.rs b/compiler/rustc_codegen_gcc/src/debuginfo.rs index 4c8585192a1b..4c0b64395232 100644 --- a/compiler/rustc_codegen_gcc/src/debuginfo.rs +++ b/compiler/rustc_codegen_gcc/src/debuginfo.rs @@ -254,7 +254,8 @@ fn extend_scope_to_file( // TODO(antoyo): implement. } - fn debuginfo_finalize(&self) { + fn debuginfo_finalize(&mut self) { + // TODO: emit section `.debug_gdb_scripts`. self.context.set_debug_info(true) } diff --git a/compiler/rustc_codegen_llvm/src/base.rs b/compiler/rustc_codegen_llvm/src/base.rs index 5dda836988c8..d7da03bf490f 100644 --- a/compiler/rustc_codegen_llvm/src/base.rs +++ b/compiler/rustc_codegen_llvm/src/base.rs @@ -109,11 +109,16 @@ fn module_codegen(tcx: TyCtxt<'_>, cgu_name: Symbol) -> ModuleCodegen, cgu_name: Symbol) -> ModuleCodegen( cx: &CodegenCx<'ll, '_>, ) -> &'ll Value { - let c_section_var_name = c"__rustc_debug_gdb_scripts_section__"; + let c_section_var_name = CString::new(format!( + "__rustc_debug_gdb_scripts_section_{}_{:08x}", + cx.tcx.crate_name(LOCAL_CRATE), + cx.tcx.stable_crate_id(LOCAL_CRATE), + )) + .unwrap(); let section_var_name = c_section_var_name.to_str().unwrap(); let section_var = unsafe { llvm::LLVMGetNamedGlobal(cx.llmod, c_section_var_name.as_ptr()) }; @@ -84,17 +91,10 @@ pub(crate) fn get_or_insert_gdb_debug_scripts_section_global<'ll>( } pub(crate) fn needs_gdb_debug_scripts_section(cx: &CodegenCx<'_, '_>) -> bool { - // To ensure the section `__rustc_debug_gdb_scripts_section__` will not create - // ODR violations at link time, this section will not be emitted for rlibs since - // each rlib could produce a different set of visualizers that would be embedded - // in the `.debug_gdb_scripts` section. For that reason, we make sure that the - // section is only emitted for leaf crates. + // We collect pretty printers transitively for all crates, so we make sure + // that the section is only emitted for leaf crates. let embed_visualizers = cx.tcx.crate_types().iter().any(|&crate_type| match crate_type { - CrateType::Executable - | CrateType::Dylib - | CrateType::Cdylib - | CrateType::Staticlib - | CrateType::Sdylib => { + CrateType::Executable | CrateType::Cdylib | CrateType::Staticlib | CrateType::Sdylib => { // These are crate types for which we will embed pretty printers since they // are treated as leaf crates. true @@ -105,9 +105,11 @@ pub(crate) fn needs_gdb_debug_scripts_section(cx: &CodegenCx<'_, '_>) -> bool { // want to slow down the common case. false } - CrateType::Rlib => { - // As per the above description, embedding pretty printers for rlibs could - // lead to ODR violations so we skip this crate type as well. + CrateType::Rlib | CrateType::Dylib => { + // Don't embed pretty printers for these crate types; the compiler + // can see the `#[debug_visualizer]` attributes when using the + // library, and emitting `.debug_gdb_scripts` regardless would + // break `#![omit_gdb_pretty_printer_section]`. false } }); diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs b/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs index 6cbf2dbf7d3f..c911435967c3 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/mod.rs @@ -30,7 +30,7 @@ use self::metadata::{UNKNOWN_COLUMN_NUMBER, UNKNOWN_LINE_NUMBER, file_metadata, type_di_node}; use self::namespace::mangled_name_of_instance; -use self::utils::{DIB, create_DIArray, is_node_local_to_unit}; +use self::utils::{DIB, create_DIArray, debug_context, is_node_local_to_unit}; use crate::builder::Builder; use crate::common::{AsCCharPtr, CodegenCx}; use crate::llvm; @@ -131,20 +131,28 @@ pub(crate) fn finalize(&self, sess: &Session) { } /// Creates any deferred debug metadata nodes -pub(crate) fn finalize(cx: &CodegenCx<'_, '_>) { - if let Some(dbg_cx) = &cx.dbg_cx { - debug!("finalize"); - - if gdb::needs_gdb_debug_scripts_section(cx) { - // Add a .debug_gdb_scripts section to this compile-unit. This will - // cause GDB to try and load the gdb_load_rust_pretty_printers.py file, - // which activates the Rust pretty printers for binary this section is - // contained in. - gdb::get_or_insert_gdb_debug_scripts_section_global(cx); - } - - dbg_cx.finalize(cx.sess()); +pub(crate) fn finalize(cx: &mut CodegenCx<'_, '_>) { + if cx.dbg_cx.is_none() { + return; } + + debug!("finalize"); + + if gdb::needs_gdb_debug_scripts_section(cx) { + // Add a .debug_gdb_scripts section to this compile-unit. This will + // cause GDB to try and load the gdb_load_rust_pretty_printers.py file, + // which activates the Rust pretty printers for binary this section is + // contained in. + let section_var = gdb::get_or_insert_gdb_debug_scripts_section_global(cx); + + // Make sure that the linker doesn't optimize the global away. Adding + // it to `llvm.used` has the advantage that it works even in no_std + // binaries, where we don't have a main shim and thus don't emit a + // volatile load to preserve the global. + cx.add_used_global(section_var); + } + + debug_context(cx).finalize(cx.sess()); } impl<'ll> Builder<'_, 'll, '_> { @@ -614,7 +622,7 @@ fn extend_scope_to_file( metadata::extend_scope_to_file(self, scope_metadata, file) } - fn debuginfo_finalize(&self) { + fn debuginfo_finalize(&mut self) { finalize(self) } diff --git a/compiler/rustc_codegen_ssa/src/base.rs b/compiler/rustc_codegen_ssa/src/base.rs index b4556ced0b3f..eb21468650c7 100644 --- a/compiler/rustc_codegen_ssa/src/base.rs +++ b/compiler/rustc_codegen_ssa/src/base.rs @@ -609,15 +609,7 @@ pub fn collect_debugger_visualizers_transitive( ) -> BTreeSet { tcx.debugger_visualizers(LOCAL_CRATE) .iter() - .chain( - tcx.crates(()) - .iter() - .filter(|&cnum| { - let used_crate_source = tcx.used_crate_source(*cnum); - used_crate_source.rlib.is_some() || used_crate_source.rmeta.is_some() - }) - .flat_map(|&cnum| tcx.debugger_visualizers(cnum)), - ) + .chain(tcx.crates(()).iter().flat_map(|&cnum| tcx.debugger_visualizers(cnum))) .filter(|visualizer| visualizer.visualizer_type == visualizer_type) .cloned() .collect::>() diff --git a/compiler/rustc_codegen_ssa/src/traits/debuginfo.rs b/compiler/rustc_codegen_ssa/src/traits/debuginfo.rs index b9d4950e0ad3..30a3bd5abe49 100644 --- a/compiler/rustc_codegen_ssa/src/traits/debuginfo.rs +++ b/compiler/rustc_codegen_ssa/src/traits/debuginfo.rs @@ -50,7 +50,7 @@ fn extend_scope_to_file( scope_metadata: Self::DIScope, file: &SourceFile, ) -> Self::DIScope; - fn debuginfo_finalize(&self); + fn debuginfo_finalize(&mut self); // FIXME(eddyb) find a common convention for all of the debuginfo-related // names (choose between `dbg`, `debug`, `debuginfo`, `debug_info` etc.). diff --git a/tests/codegen-llvm/gdb_debug_script_load.rs b/tests/codegen-llvm/gdb_debug_script_load.rs index 3e92eba10b12..90f2be41cf2a 100644 --- a/tests/codegen-llvm/gdb_debug_script_load.rs +++ b/tests/codegen-llvm/gdb_debug_script_load.rs @@ -9,6 +9,8 @@ #![feature(lang_items)] #![no_std] +// CHECK: @llvm.used = {{.+}} @__rustc_debug_gdb_scripts_section + #[panic_handler] fn panic_handler(_: &core::panic::PanicInfo) -> ! { loop {} @@ -22,7 +24,7 @@ extern "C" fn rust_eh_personality() { // Needs rustc to generate `main` as that's where the magic load is inserted. // IOW, we cannot write this test with `#![no_main]`. // CHECK-LABEL: @main -// CHECK: load volatile i8, {{.+}} @__rustc_debug_gdb_scripts_section__ +// CHECK: load volatile i8, {{.+}} @__rustc_debug_gdb_scripts_section #[lang = "start"] fn lang_start( diff --git a/tests/run-make/symbols-all-mangled/rmake.rs b/tests/run-make/symbols-all-mangled/rmake.rs index 2cf579758002..e30bef985807 100644 --- a/tests/run-make/symbols-all-mangled/rmake.rs +++ b/tests/run-make/symbols-all-mangled/rmake.rs @@ -35,13 +35,7 @@ fn symbols_check_archive(path: &str) { continue; // All compiler-builtins symbols must remain unmangled } - if name.contains("rust_eh_personality") { - continue; // Unfortunately LLVM doesn't allow us to mangle this symbol - } - - if name.contains(".llvm.") { - // Starting in LLVM 21 we get various implementation-detail functions which - // contain .llvm. that are not a problem. + if symbol_ok_everywhere(name) { continue; } @@ -71,13 +65,7 @@ fn symbols_check(path: &str) { continue; } - if name.contains("rust_eh_personality") { - continue; // Unfortunately LLVM doesn't allow us to mangle this symbol - } - - if name.contains(".llvm.") { - // Starting in LLVM 21 we get various implementation-detail functions which - // contain .llvm. that are not a problem. + if symbol_ok_everywhere(name) { continue; } @@ -88,3 +76,22 @@ fn symbols_check(path: &str) { fn strip_underscore_if_apple(symbol: &str) -> &str { if cfg!(target_vendor = "apple") { symbol.strip_prefix("_").unwrap() } else { symbol } } + +fn symbol_ok_everywhere(name: &str) -> bool { + if name.contains("rust_eh_personality") { + return true; // Unfortunately LLVM doesn't allow us to mangle this symbol + } + + if name.contains(".llvm.") { + // Starting in LLVM 21 we get various implementation-detail functions which + // contain .llvm. that are not a problem. + return true; + } + + if name.starts_with("__rustc_debug_gdb_scripts_section") { + // These symbols are fine; they're made unique by the crate ID. + return true; + } + + return false; +}