Synthesis object file for #[used] and exported symbols

Author: nbdd0121

compiler/rustc_codegen_ssa/src/back/link.rs

@@ -7,6 +7,7 @@ use rustc_errors::{ErrorGuaranteed, Handler};
 use rustc_fs_util::fix_windows_verbatim_for_gcc;
 use rustc_hir::def_id::CrateNum;
 use rustc_middle::middle::dependency_format::Linkage;
+use rustc_middle::middle::exported_symbols::SymbolExportKind;
 use rustc_session::config::{self, CFGuard, CrateType, DebugInfo, LdImpl, Strip};
 use rustc_session::config::{OutputFilenames, OutputType, PrintRequest, SplitDwarfKind};
 use rustc_session::cstore::DllImport;
@@ -1654,6 +1655,67 @@ fn add_post_link_args(cmd: &mut dyn Linker, sess: &Session, flavor: LinkerFlavor
     }
 }
 
+/// Add a synthetic object file that contains reference to all symbols that we want to expose to
+/// the linker.
+///
+/// Background: we implement rlibs as static library (archives). Linkers treat archives
+/// differently from object files: all object files participate in linking, while archives will
+/// only participate in linking if they can satisfy at least one undefined reference (version
+/// scripts doesn't count). This causes `#[no_mangle]` or `#[used]` items to be ignored by the
+/// linker, and since they never participate in the linking, using `KEEP` in the linker scripts
+/// can't keep them either. This causes #47384.
+///
+/// To keep them around, we could use `--whole-archive` and equivalents to force rlib to
+/// participate in linking like object files, but this proves to be expensive (#93791). Therefore
+/// we instead just introduce an undefined reference to them. This could be done by `-u` command
+/// line option to the linker or `EXTERN(...)` in linker scripts, however they does not only
+/// introduce an undefined reference, but also make them the GC roots, preventing `--gc-sections`
+/// from removing them, and this is especially problematic for embedded programming where every
+/// byte counts.
+///
+/// This method creates a synthetic object file, which contains undefined references to all symbols
+/// that are necessary for the linking. They are only present in symbol table but not actually
+/// used in any sections, so the linker will therefore pick relevant rlibs for linking, but
+/// unused `#[no_mangle]` or `#[used]` can still be discard by GC sections.
+fn add_linked_symbol_object(
+    cmd: &mut dyn Linker,
+    sess: &Session,
+    tmpdir: &Path,
+    symbols: &[(String, SymbolExportKind)],
+) {
+    if symbols.is_empty() {
+        return;
+    }
+
+    let Some(mut file) = super::metadata::create_object_file(sess) else {
+        return;
+    };
+
+    for (sym, kind) in symbols.iter() {
+        file.add_symbol(object::write::Symbol {
+            name: sym.clone().into(),
+            value: 0,
+            size: 0,
+            kind: match kind {
+                SymbolExportKind::Text => object::SymbolKind::Text,
+                SymbolExportKind::Data => object::SymbolKind::Data,
+                SymbolExportKind::Tls => object::SymbolKind::Tls,
+            },
+            scope: object::SymbolScope::Unknown,
+            weak: false,
+            section: object::write::SymbolSection::Undefined,
+            flags: object::SymbolFlags::None,
+        });
+    }
+
+    let path = tmpdir.join("symbols.o");
+    let result = std::fs::write(&path, file.write().unwrap());
+    if let Err(e) = result {
+        sess.fatal(&format!("failed to write {}: {}", path.display(), e));
+    }
+    cmd.add_object(&path);
+}
+
 /// Add object files containing code from the current crate.
 fn add_local_crate_regular_objects(cmd: &mut dyn Linker, codegen_results: &CodegenResults) {
     for obj in codegen_results.modules.iter().filter_map(|m| m.object.as_ref()) {
@@ -1794,6 +1856,13 @@ fn linker_with_args<'a, B: ArchiveBuilder<'a>>(
     // Pre-link CRT objects.
     add_pre_link_objects(cmd, sess, link_output_kind, crt_objects_fallback);
 
+    add_linked_symbol_object(
+        cmd,
+        sess,
+        tmpdir,
+        &codegen_results.crate_info.linked_symbols[&crate_type],
+    );
+
     // Sanitizer libraries.
     add_sanitizer_libraries(sess, crate_type, cmd);
 

compiler/rustc_codegen_ssa/src/back/linker.rs

@@ -12,6 +12,7 @@ use std::{env, mem, str};
 
 use rustc_hir::def_id::{CrateNum, LOCAL_CRATE};
 use rustc_middle::middle::dependency_format::Linkage;
+use rustc_middle::middle::exported_symbols::SymbolExportKind;
 use rustc_middle::ty::TyCtxt;
 use rustc_serialize::{json, Encoder};
 use rustc_session::config::{self, CrateType, DebugInfo, LinkerPluginLto, Lto, OptLevel, Strip};
@@ -1557,6 +1558,51 @@ pub(crate) fn exported_symbols(tcx: TyCtxt<'_>, crate_type: CrateType) -> Vec<St
     symbols
 }
 
+pub(crate) fn linked_symbols(
+    tcx: TyCtxt<'_>,
+    crate_type: CrateType,
+) -> Vec<(String, SymbolExportKind)> {
+    match crate_type {
+        CrateType::Executable | CrateType::Cdylib => (),
+        CrateType::Staticlib | CrateType::ProcMacro | CrateType::Rlib | CrateType::Dylib => {
+            return Vec::new();
+        }
+    }
+
+    let mut symbols = Vec::new();
+
+    let export_threshold = symbol_export::crates_export_threshold(&[crate_type]);
+    for &(symbol, info) in tcx.exported_symbols(LOCAL_CRATE).iter() {
+        if info.level.is_below_threshold(export_threshold) || info.used {
+            symbols.push((
+                symbol_export::symbol_name_for_instance_in_crate(tcx, symbol, LOCAL_CRATE),
+                info.kind,
+            ));
+        }
+    }
+
+    let formats = tcx.dependency_formats(());
+    let deps = formats.iter().find_map(|(t, list)| (*t == crate_type).then_some(list)).unwrap();
+
+    for (index, dep_format) in deps.iter().enumerate() {
+        let cnum = CrateNum::new(index + 1);
+        // For each dependency that we are linking to statically ...
+        if *dep_format == Linkage::Static {
+            // ... we add its symbol list to our export list.
+            for &(symbol, info) in tcx.exported_symbols(cnum).iter() {
+                if info.level.is_below_threshold(export_threshold) || info.used {
+                    symbols.push((
+                        symbol_export::symbol_name_for_instance_in_crate(tcx, symbol, cnum),
+                        info.kind,
+                    ));
+                }
+            }
+        }
+    }
+
+    symbols
+}
+
 /// Much simplified and explicit CLI for the NVPTX linker. The linker operates
 /// with bitcode and uses LLVM backend to generate a PTX assembly.
 pub struct PtxLinker<'a> {

compiler/rustc_codegen_ssa/src/back/metadata.rs

@@ -94,7 +94,7 @@ fn search_for_metadata<'a>(
         .map_err(|e| format!("failed to read {} section in '{}': {}", section, path.display(), e))
 }
 
-fn create_object_file(sess: &Session) -> Option<write::Object<'static>> {
+pub(crate) fn create_object_file(sess: &Session) -> Option<write::Object<'static>> {
     let endianness = match sess.target.options.endian {
         Endian::Little => Endianness::Little,
         Endian::Big => Endianness::Big,

compiler/rustc_codegen_ssa/src/back/symbol_export.rs

@@ -9,7 +9,7 @@ use rustc_hir::Node;
 use rustc_index::vec::IndexVec;
 use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrFlags;
 use rustc_middle::middle::exported_symbols::{
-    metadata_symbol_name, ExportedSymbol, SymbolExportInfo, SymbolExportLevel,
+    metadata_symbol_name, ExportedSymbol, SymbolExportInfo, SymbolExportKind, SymbolExportLevel,
 };
 use rustc_middle::ty::query::{ExternProviders, Providers};
 use rustc_middle::ty::subst::{GenericArgKind, SubstsRef};
@@ -124,21 +124,37 @@ fn reachable_non_generics_provider(tcx: TyCtxt<'_>, cnum: CrateNum) -> DefIdMap<
             } else {
                 symbol_export_level(tcx, def_id.to_def_id())
             };
+            let codegen_attrs = tcx.codegen_fn_attrs(def_id.to_def_id());
             debug!(
                 "EXPORTED SYMBOL (local): {} ({:?})",
                 tcx.symbol_name(Instance::mono(tcx, def_id.to_def_id())),
                 export_level
             );
             (def_id.to_def_id(), SymbolExportInfo {
                 level: export_level,
+                kind: if tcx.is_static(def_id.to_def_id()) {
+                    if codegen_attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL) {
+                        SymbolExportKind::Tls
+                    } else {
+                        SymbolExportKind::Data
+                    }
+                } else {
+                    SymbolExportKind::Text
+                },
+                used: codegen_attrs.flags.contains(CodegenFnAttrFlags::USED)
+                    || codegen_attrs.flags.contains(CodegenFnAttrFlags::USED_LINKER),
             })
         })
         .collect();
 
     if let Some(id) = tcx.proc_macro_decls_static(()) {
         reachable_non_generics.insert(
             id.to_def_id(),
-            SymbolExportInfo { level: SymbolExportLevel::C },
+            SymbolExportInfo {
+                level: SymbolExportLevel::C,
+                kind: SymbolExportKind::Data,
+                used: false,
+            },
         );
     }
 
@@ -180,7 +196,11 @@ fn exported_symbols_provider_local<'tcx>(
 
         symbols.push((
             exported_symbol,
-            SymbolExportInfo { level: SymbolExportLevel::C },
+            SymbolExportInfo {
+                level: SymbolExportLevel::C,
+                kind: SymbolExportKind::Text,
+                used: false,
+            },
         ));
     }
 
@@ -191,7 +211,11 @@ fn exported_symbols_provider_local<'tcx>(
 
             symbols.push((
                 exported_symbol,
-                SymbolExportInfo { level: SymbolExportLevel::Rust },
+                SymbolExportInfo {
+                    level: SymbolExportLevel::Rust,
+                    kind: SymbolExportKind::Text,
+                    used: false,
+                },
             ));
         }
     }
@@ -207,7 +231,11 @@ fn exported_symbols_provider_local<'tcx>(
             let exported_symbol = ExportedSymbol::NoDefId(SymbolName::new(tcx, sym));
             (
                 exported_symbol,
-                SymbolExportInfo { level: SymbolExportLevel::C },
+                SymbolExportInfo {
+                    level: SymbolExportLevel::C,
+                    kind: SymbolExportKind::Data,
+                    used: false,
+                },
             )
         }));
     }
@@ -220,7 +248,11 @@ fn exported_symbols_provider_local<'tcx>(
             let exported_symbol = ExportedSymbol::NoDefId(SymbolName::new(tcx, sym));
             (
                 exported_symbol,
-                SymbolExportInfo { level: SymbolExportLevel::C },
+                SymbolExportInfo {
+                    level: SymbolExportLevel::C,
+                    kind: SymbolExportKind::Data,
+                    used: false,
+                },
             )
         }));
     }
@@ -231,7 +263,11 @@ fn exported_symbols_provider_local<'tcx>(
 
         symbols.push((
             exported_symbol,
-            SymbolExportInfo { level: SymbolExportLevel::Rust },
+            SymbolExportInfo {
+                level: SymbolExportLevel::Rust,
+                kind: SymbolExportKind::Data,
+                used: false,
+            },
         ));
     }
 
@@ -269,6 +305,8 @@ fn exported_symbols_provider_local<'tcx>(
                             symbol,
                             SymbolExportInfo {
                                 level: SymbolExportLevel::Rust,
+                                kind: SymbolExportKind::Text,
+                                used: false,
                             },
                         ));
                     }
@@ -283,6 +321,8 @@ fn exported_symbols_provider_local<'tcx>(
                         ExportedSymbol::DropGlue(ty),
                         SymbolExportInfo {
                             level: SymbolExportLevel::Rust,
+                            kind: SymbolExportKind::Text,
+                            used: false,
                         },
                     ));
                 }

compiler/rustc_codegen_ssa/src/base.rs

@@ -801,6 +801,12 @@ impl CrateInfo {
             .iter()
             .map(|&c| (c, crate::back::linker::exported_symbols(tcx, c)))
             .collect();
+        let linked_symbols = tcx
+            .sess
+            .crate_types()
+            .iter()
+            .map(|&c| (c, crate::back::linker::linked_symbols(tcx, c)))
+            .collect();
         let local_crate_name = tcx.crate_name(LOCAL_CRATE);
         let crate_attrs = tcx.hir().attrs(rustc_hir::CRATE_HIR_ID);
         let subsystem = tcx.sess.first_attr_value_str_by_name(crate_attrs, sym::windows_subsystem);
@@ -834,6 +840,7 @@ impl CrateInfo {
         let mut info = CrateInfo {
             target_cpu,
             exported_symbols,
+            linked_symbols,
             local_crate_name,
             compiler_builtins: None,
             profiler_runtime: None,

compiler/rustc_codegen_ssa/src/lib.rs

@@ -28,6 +28,7 @@ use rustc_hir::def_id::CrateNum;
 use rustc_hir::LangItem;
 use rustc_middle::dep_graph::WorkProduct;
 use rustc_middle::middle::dependency_format::Dependencies;
+use rustc_middle::middle::exported_symbols::SymbolExportKind;
 use rustc_middle::ty::query::{ExternProviders, Providers};
 use rustc_session::config::{CrateType, OutputFilenames, OutputType, RUST_CGU_EXT};
 use rustc_session::cstore::{self, CrateSource};
@@ -140,6 +141,7 @@ impl From<&cstore::NativeLib> for NativeLib {
 pub struct CrateInfo {
     pub target_cpu: String,
     pub exported_symbols: FxHashMap<CrateType, Vec<String>>,
+    pub linked_symbols: FxHashMap<CrateType, Vec<(String, SymbolExportKind)>>,
     pub local_crate_name: Symbol,
     pub compiler_builtins: Option<CrateNum>,
     pub profiler_runtime: Option<CrateNum>,

compiler/rustc_middle/src/middle/exported_symbols.rs

@@ -21,11 +21,21 @@ impl SymbolExportLevel {
     }
 }
 
+/// Kind of exported symbols.
+#[derive(Eq, PartialEq, Debug, Copy, Clone, Encodable, Decodable, HashStable)]
+pub enum SymbolExportKind {
+    Text,
+    Data,
+    Tls,
+}
+
 /// The `SymbolExportInfo` of a symbols specifies symbol-related information
 /// that is relevant to code generation and linking.
 #[derive(Eq, PartialEq, Debug, Copy, Clone, TyEncodable, TyDecodable, HashStable)]
 pub struct SymbolExportInfo {
     pub level: SymbolExportLevel,
+    pub kind: SymbolExportKind,
+    pub used: bool,
 }
 
 #[derive(Eq, PartialEq, Debug, Copy, Clone, TyEncodable, TyDecodable, HashStable)]