[lld] [lld-macho] Category Merger: add support for addrsig references (PR #90903)
via llvm-commits
llvm-commits at lists.llvm.org
Thu May 2 13:15:49 PDT 2024
https://github.com/alx32 created https://github.com/llvm/llvm-project/pull/90903
When generating categories, clang sometimes will generate references in the `.addrsig` section to the various category data items. Since we may erase such items after merging them, we also need to remove them from the `.addrsig` section - otherwise this will cause runtime asserts with the `.addrsig` section trying to access invalid data.
Implementation wise, we use a hashset to keep track of all erased `InputSection`'s and then go through all `.addrsig` sections and remove references to any erased `InputSection`.
>From 9a44efbc04ec75704ec2206f598f4e3223f5d6f9 Mon Sep 17 00:00:00 2001
From: Alex Borcan <alexborcan at fb.com>
Date: Thu, 2 May 2024 13:10:12 -0700
Subject: [PATCH] [lld-macho] Category Merger: add support for addrsig
references
---
lld/MachO/ObjC.cpp | 91 ++++++++++++++++---
...jc-category-merging-extern-class-minimal.s | 3 +
2 files changed, 83 insertions(+), 11 deletions(-)
diff --git a/lld/MachO/ObjC.cpp b/lld/MachO/ObjC.cpp
index 95fe0c9374f150..caff8f00ff1b8a 100644
--- a/lld/MachO/ObjC.cpp
+++ b/lld/MachO/ObjC.cpp
@@ -21,6 +21,8 @@
#include "llvm/Bitcode/BitcodeReader.h"
#include "llvm/Support/TimeProfiler.h"
+#include <unordered_set>
+
using namespace llvm;
using namespace llvm::MachO;
using namespace lld;
@@ -431,6 +433,7 @@ class ObjcCategoryMerger {
mergeCategoriesIntoSingleCategory(std::vector<InfoInputCategory> &categories);
void eraseISec(ConcatInputSection *isec);
+ void removeRefsToErasedIsecs(std::unordered_set<InputSection *> erasedIsecs);
void eraseMergedCategories();
void generateCatListForNonErasedCategories(
@@ -471,8 +474,11 @@ class ObjcCategoryMerger {
uint32_t offset);
Defined *tryGetDefinedAtIsecOffset(const ConcatInputSection *isec,
uint32_t offset);
+ Defined *tryFindDefinedOnIsec(const InputSection *isec, uint32_t offset);
void tryEraseDefinedAtIsecOffset(const ConcatInputSection *isec,
- uint32_t offset);
+ uint32_t offset,
+ std::unordered_set<InputSection *> &erased);
+ void eraseSymbolAtIsecOffset(ConcatInputSection *isec, uint32_t offset);
// Allocate a null-terminated StringRef backed by generatedSectionData
StringRef newStringData(const char *str);
@@ -533,12 +539,23 @@ void ObjcCategoryMerger::collectSectionWriteInfoFromIsec(
Symbol *
ObjcCategoryMerger::tryGetSymbolAtIsecOffset(const ConcatInputSection *isec,
uint32_t offset) {
+ if (!isec)
+ return nullptr;
const Reloc *reloc = isec->getRelocAt(offset);
if (!reloc)
return nullptr;
- return reloc->referent.get<Symbol *>();
+ Symbol *sym = reloc->referent.get<Symbol *>();
+
+ if (reloc->addend) {
+ assert(isa<Defined>(sym) && "Expected defined for non-zero addend");
+ Defined *definedSym = dyn_cast<Defined>(sym);
+ InputSection *targetIsec = definedSym->isec();
+ sym = tryFindDefinedOnIsec(targetIsec, definedSym->value + reloc->addend);
+ }
+
+ return sym;
}
Defined *
@@ -548,10 +565,20 @@ ObjcCategoryMerger::tryGetDefinedAtIsecOffset(const ConcatInputSection *isec,
return dyn_cast_or_null<Defined>(sym);
}
+Defined *ObjcCategoryMerger::tryFindDefinedOnIsec(const InputSection *isec,
+ uint32_t offset) {
+ for (Defined *sym : isec->symbols)
+ if ((sym->value <= offset) && (sym->value + sym->size > offset))
+ return sym;
+
+ return nullptr;
+}
+
// Given an ConcatInputSection or CStringInputSection and an offset, if there is
// a symbol(Defined) at that offset, then erase the symbol (mark it not live)
void ObjcCategoryMerger::tryEraseDefinedAtIsecOffset(
- const ConcatInputSection *isec, uint32_t offset) {
+ const ConcatInputSection *isec, uint32_t offset,
+ std::unordered_set<InputSection *> &erased) {
const Reloc *reloc = isec->getRelocAt(offset);
if (!reloc)
@@ -561,9 +588,11 @@ void ObjcCategoryMerger::tryEraseDefinedAtIsecOffset(
if (!sym)
return;
- if (auto *cisec = dyn_cast_or_null<ConcatInputSection>(sym->isec()))
+ if (auto *cisec = dyn_cast_or_null<ConcatInputSection>(sym->isec())) {
eraseISec(cisec);
- else if (auto *csisec = dyn_cast_or_null<CStringInputSection>(sym->isec())) {
+ erased.insert(cisec);
+ } else if (auto *csisec =
+ dyn_cast_or_null<CStringInputSection>(sym->isec())) {
uint32_t totalOffset = sym->value + reloc->addend;
StringPiece &piece = csisec->getStringPiece(totalOffset);
piece.live = false;
@@ -1183,26 +1212,66 @@ void ObjcCategoryMerger::eraseMergedCategories() {
// the references to the ones we merged.
generateCatListForNonErasedCategories(catListToErasedOffsets);
+ // We use erasedIsecs below to track erased sections so we can later remove
+ // references to it.
+ std::unordered_set<InputSection *> erasedIsecs;
+ erasedIsecs.reserve(categoryMap.size());
+
// Erase the old method lists & names of the categories that were merged
for (auto &mapEntry : categoryMap) {
for (InfoInputCategory &catInfo : mapEntry.second) {
if (!catInfo.wasMerged)
continue;
+ erasedIsecs.insert(catInfo.catBodyIsec);
+ erasedIsecs.insert(catInfo.catListIsec);
+
eraseISec(catInfo.catBodyIsec);
- tryEraseDefinedAtIsecOffset(catInfo.catBodyIsec, catLayout.nameOffset);
+ tryEraseDefinedAtIsecOffset(catInfo.catBodyIsec, catLayout.nameOffset,
+ erasedIsecs);
tryEraseDefinedAtIsecOffset(catInfo.catBodyIsec,
- catLayout.instanceMethodsOffset);
+ catLayout.instanceMethodsOffset, erasedIsecs);
tryEraseDefinedAtIsecOffset(catInfo.catBodyIsec,
- catLayout.classMethodsOffset);
+ catLayout.classMethodsOffset, erasedIsecs);
tryEraseDefinedAtIsecOffset(catInfo.catBodyIsec,
- catLayout.protocolsOffset);
+ catLayout.protocolsOffset, erasedIsecs);
tryEraseDefinedAtIsecOffset(catInfo.catBodyIsec,
- catLayout.classPropsOffset);
+ catLayout.classPropsOffset, erasedIsecs);
tryEraseDefinedAtIsecOffset(catInfo.catBodyIsec,
- catLayout.instancePropsOffset);
+ catLayout.instancePropsOffset, erasedIsecs);
}
}
+
+ removeRefsToErasedIsecs(erasedIsecs);
+}
+
+// The compiler may generate references to categories inside the addrsig
+// section. This function will erase these references.
+void ObjcCategoryMerger::removeRefsToErasedIsecs(
+ std::unordered_set<InputSection *> erasedIsecs) {
+ for (InputSection *isec : inputSections) {
+ if (isec->getName() != section_names::addrSig)
+ continue;
+
+ auto removeRelocs = [&erasedIsecs](Reloc &r) {
+ ConcatInputSection *isec = nullptr;
+ isec = dyn_cast_or_null<ConcatInputSection>(
+ r.referent.dyn_cast<InputSection *>());
+ if (!isec) {
+ Defined *sym =
+ dyn_cast_or_null<Defined>(r.referent.dyn_cast<Symbol *>());
+ if (sym)
+ isec = dyn_cast<ConcatInputSection>(sym->isec());
+ }
+ if (!isec)
+ return false;
+ return erasedIsecs.count(isec) > 0;
+ };
+
+ isec->relocs.erase(
+ std::remove_if(isec->relocs.begin(), isec->relocs.end(), removeRelocs),
+ isec->relocs.end());
+ }
}
void ObjcCategoryMerger::doMerge() {
diff --git a/lld/test/MachO/objc-category-merging-extern-class-minimal.s b/lld/test/MachO/objc-category-merging-extern-class-minimal.s
index ede7ef5d9c32d4..3ca881c2da156e 100644
--- a/lld/test/MachO/objc-category-merging-extern-class-minimal.s
+++ b/lld/test/MachO/objc-category-merging-extern-class-minimal.s
@@ -153,3 +153,6 @@ L_OBJC_IMAGE_INFO:
.long 0
.long 96
.subsections_via_symbols
+
+.addrsig
+.addrsig_sym __OBJC_$_CATEGORY_MyBaseClass_$_Category01
More information about the llvm-commits
mailing list