[lld] [LLD] Implement --enable-non-contiguous-regions (PR #90007)
Daniel Thornburgh via llvm-commits
llvm-commits at lists.llvm.org
Wed Apr 24 17:17:28 PDT 2024
https://github.com/mysterymath updated https://github.com/llvm/llvm-project/pull/90007
>From 3005ffde1c738d0bb0e5a816af9b7529e52f3ad1 Mon Sep 17 00:00:00 2001
From: Daniel Thornburgh <dthorn at google.com>
Date: Mon, 4 Dec 2023 14:05:37 -0800
Subject: [PATCH] [LLD] Implement --enable-non-contiguous-regions
When enabled, input sections that would otherwise overflow a memory
region are instead spilled to the next matching output section.
This feature parallels the one in GNU LD, but there are some differences
from its documented behavior:
- /DISCARD/ only matches previously-unmatched sections (i.e., the flag
does not affect it).
- If a section fails to fit at any of its matches, the link fails
instead of discarding the section.
- The flag --enable-non-contiguous-regions-warnings is not implemented,
as it exists to warn about such occurrences.
The implementation places stubs at possible spill locations, and
replaces them with the original input section when effecting spills.
Spilling decisions occur after address assignment. Sections are spilled
in reverse order of assignment, with each spill naively decreasing the
size of the affected memory regions. This continues until the memory
regions are brought back under size. Spilling anything causes another
pass of address assignment, and this continues to fixed point.
Spilling after rather than during assignment allows the algorithm to
consider the size effects of unspillable input sections that appear
later in the assignment. Otherwise, such sections (e.g. thunks) may
force an overflow, even if spilling something earlier could have avoided
it.
A few notable feature interactions occur:
- Stubs affect alignment, ONLY_IF_RO, etc, broadly as if a copy of the
input section were actually placed there.
- SHF_MERGE synthetic sections use the spill list of their first
contained input section (the one that gives the section its name).
- ICF occurs oblivious to spill sections; spill lists for merged-away
sections become inert and are removed after assignment.
- SHF_LINK_ORDER and .ARM.exidx are ordered according to the final
section ordering, after all spilling has completed.
- INSERT BEFORE/AFTER and OVERWRITE_SECTIONS are explicitly disallowed.
---
lld/ELF/Config.h | 1 +
lld/ELF/Driver.cpp | 4 +-
lld/ELF/ICF.cpp | 2 +-
lld/ELF/InputSection.cpp | 7 +
lld/ELF/InputSection.h | 24 +-
lld/ELF/LinkerScript.cpp | 183 +++++++++++-
lld/ELF/LinkerScript.h | 16 +-
lld/ELF/Options.td | 3 +
lld/ELF/OutputSections.cpp | 4 +-
lld/ELF/OutputSections.h | 2 +-
lld/ELF/SyntheticSections.cpp | 7 +
lld/ELF/SyntheticSections.h | 4 +
lld/ELF/Writer.cpp | 12 +
lld/docs/ELF/linker_script.rst | 11 +
lld/docs/ReleaseNotes.rst | 11 +
...able-non-contiguous-regions-arm-exidx.test | 54 ++++
.../enable-non-contiguous-regions.test | 279 ++++++++++++++++++
17 files changed, 610 insertions(+), 14 deletions(-)
create mode 100644 lld/test/ELF/linkerscript/enable-non-contiguous-regions-arm-exidx.test
create mode 100644 lld/test/ELF/linkerscript/enable-non-contiguous-regions.test
diff --git a/lld/ELF/Config.h b/lld/ELF/Config.h
index 33bfa42b0fcbf03..bb5d7001f59fbbe 100644
--- a/lld/ELF/Config.h
+++ b/lld/ELF/Config.h
@@ -237,6 +237,7 @@ struct Config {
bool emitLLVM;
bool emitRelocs;
bool enableNewDtags;
+ bool enableNonContiguousRegions;
bool executeOnly;
bool exportDynamic;
bool fixCortexA53Errata843419;
diff --git a/lld/ELF/Driver.cpp b/lld/ELF/Driver.cpp
index a5b47f020f8726b..915184c1aa26320 100644
--- a/lld/ELF/Driver.cpp
+++ b/lld/ELF/Driver.cpp
@@ -1250,6 +1250,8 @@ static void readConfigs(opt::InputArgList &args) {
config->emitRelocs = args.hasArg(OPT_emit_relocs);
config->enableNewDtags =
args.hasFlag(OPT_enable_new_dtags, OPT_disable_new_dtags, true);
+ config->enableNonContiguousRegions =
+ args.hasArg(OPT_enable_non_contiguous_regions);
config->entry = args.getLastArgValue(OPT_entry);
errorHandler().errorHandlingScript =
@@ -3077,7 +3079,7 @@ template <class ELFT> void LinkerDriver::link(opt::InputArgList &args) {
// sectionBases.
for (SectionCommand *cmd : script->sectionCommands)
if (auto *osd = dyn_cast<OutputDesc>(cmd))
- osd->osec.finalizeInputSections();
+ osd->osec.finalizeInputSections(script.get());
}
// Two input sections with different output sections should not be folded.
diff --git a/lld/ELF/ICF.cpp b/lld/ELF/ICF.cpp
index bfc605c793a92ca..9be825ef36ba473 100644
--- a/lld/ELF/ICF.cpp
+++ b/lld/ELF/ICF.cpp
@@ -75,7 +75,7 @@
#include "ICF.h"
#include "Config.h"
#include "InputFiles.h"
-#include "LinkerScript.h"
+#include "InputSection.h"
#include "OutputSections.h"
#include "SymbolTable.h"
#include "Symbols.h"
diff --git a/lld/ELF/InputSection.cpp b/lld/ELF/InputSection.cpp
index fa48552b8f7a12e..c5f050fffdcb3c3 100644
--- a/lld/ELF/InputSection.cpp
+++ b/lld/ELF/InputSection.cpp
@@ -161,6 +161,7 @@ uint64_t SectionBase::getOffset(uint64_t offset) const {
}
case Regular:
case Synthetic:
+ case Spill:
return cast<InputSection>(this)->outSecOff + offset;
case EHFrame: {
// Two code paths may reach here. First, clang_rt.crtbegin.o and GCC
@@ -309,6 +310,12 @@ std::string InputSectionBase::getObjMsg(uint64_t off) const {
.str();
}
+SpillInputSection::SpillInputSection(InputSectionBase *source,
+ InputSectionDescription *isd)
+ : InputSection(source->file, source->flags, source->type, source->addralign,
+ {}, source->name, SectionBase::Spill),
+ isd(isd) {}
+
InputSection InputSection::discarded(nullptr, 0, 0, 0, ArrayRef<uint8_t>(), "");
InputSection::InputSection(InputFile *f, uint64_t flags, uint32_t type,
diff --git a/lld/ELF/InputSection.h b/lld/ELF/InputSection.h
index 1fb7077ca435bdc..3d75a9633a3ccf9 100644
--- a/lld/ELF/InputSection.h
+++ b/lld/ELF/InputSection.h
@@ -48,7 +48,7 @@ template <class ELFT> struct RelsOrRelas {
// sections.
class SectionBase {
public:
- enum Kind { Regular, Synthetic, EHFrame, Merge, Output };
+ enum Kind { Regular, Synthetic, EHFrame, Merge, Output, Spill };
Kind kind() const { return (Kind)sectionKind; }
@@ -382,7 +382,8 @@ class InputSection : public InputSectionBase {
static bool classof(const SectionBase *s) {
return s->kind() == SectionBase::Regular ||
- s->kind() == SectionBase::Synthetic;
+ s->kind() == SectionBase::Synthetic ||
+ s->kind() == SectionBase::Spill;
}
// Write this section to a mmap'ed file, assuming Buf is pointing to
@@ -425,6 +426,25 @@ class InputSection : public InputSectionBase {
template <class ELFT> void copyShtGroup(uint8_t *buf);
};
+// A marker for a potential spill location for another input section. This
+// broadly acts as if it were the original section until address assignment.
+// Then it is either replaced with the real input section or removed.
+class SpillInputSection : public InputSection {
+public:
+ // The containing input section description; used to quickly replace this stub
+ // with the actual section.
+ InputSectionDescription *isd;
+
+ // Next spill location for the same source input section.
+ SpillInputSection *next = nullptr;
+
+ SpillInputSection(InputSectionBase *source, InputSectionDescription *cmd);
+
+ static bool classof(const SectionBase *sec) {
+ return sec->kind() == InputSectionBase::Spill;
+ }
+};
+
static_assert(sizeof(InputSection) <= 160, "InputSection is too big");
class SyntheticSection : public InputSection {
diff --git a/lld/ELF/LinkerScript.cpp b/lld/ELF/LinkerScript.cpp
index f815b3ac6feeda0..8db8e4abc595fb4 100644
--- a/lld/ELF/LinkerScript.cpp
+++ b/lld/ELF/LinkerScript.cpp
@@ -304,6 +304,9 @@ getChangedSymbolAssignment(const SymbolAssignmentMap &oldValues) {
void LinkerScript::processInsertCommands() {
SmallVector<OutputDesc *, 0> moves;
for (const InsertCommand &cmd : insertCommands) {
+ if (config->enableNonContiguousRegions)
+ error("INSERT cannot be used with --enable-non-contiguous-regions");
+
for (StringRef name : cmd.names) {
// If base is empty, it may have been discarded by
// adjustOutputSections(). We do not handle such output sections.
@@ -486,10 +489,12 @@ static void sortInputSections(MutableArrayRef<InputSectionBase *> vec,
// Compute and remember which sections the InputSectionDescription matches.
SmallVector<InputSectionBase *, 0>
LinkerScript::computeInputSections(const InputSectionDescription *cmd,
- ArrayRef<InputSectionBase *> sections) {
+ ArrayRef<InputSectionBase *> sections,
+ const OutputSection &outCmd) {
SmallVector<InputSectionBase *, 0> ret;
SmallVector<size_t, 0> indexes;
DenseSet<size_t> seen;
+ DenseSet<InputSectionBase *> spills;
auto sortByPositionThenCommandLine = [&](size_t begin, size_t end) {
llvm::sort(MutableArrayRef<size_t>(indexes).slice(begin, end - begin));
for (size_t i = begin; i != end; ++i)
@@ -505,12 +510,33 @@ LinkerScript::computeInputSections(const InputSectionDescription *cmd,
size_t sizeBeforeCurrPat = ret.size();
for (size_t i = 0, e = sections.size(); i != e; ++i) {
- // Skip if the section is dead or has been matched by a previous input
- // section description or a previous pattern.
+ // Skip if the section is dead or has been matched by a previous pattern
+ // in this input section description.
InputSectionBase *sec = sections[i];
- if (!sec->isLive() || sec->parent || seen.contains(i))
+ if (!sec->isLive() || seen.contains(i))
continue;
+ if (sec->parent) {
+ // Skip if not allowing multiple matches.
+ if (!config->enableNonContiguousRegions)
+ continue;
+
+ // Disallow spilling into /DISCARD/; special handling would be needed
+ // for this in address assignment, and the semantics are nebulous.
+ if (outCmd.name == "/DISCARD/")
+ continue;
+
+ // Skip if the section's first match was /DISCARD/; such sections are
+ // always discarded.
+ if (sec->parent->name == "/DISCARD/")
+ continue;
+
+ // Skip if the section was already matched by a different input section
+ // description within this output section.
+ if (sec->parent == &outCmd)
+ continue;
+ }
+
// For --emit-relocs we have to ignore entries like
// .rela.dyn : { *(.rela.data) }
// which are common because they are in the default bfd script.
@@ -530,6 +556,8 @@ LinkerScript::computeInputSections(const InputSectionDescription *cmd,
continue;
ret.push_back(sec);
+ if (sec->parent)
+ spills.insert(sec);
indexes.push_back(i);
seen.insert(i);
}
@@ -555,6 +583,28 @@ LinkerScript::computeInputSections(const InputSectionDescription *cmd,
// Matched sections after the last SORT* are sorted by (--sort-alignment,
// input order).
sortByPositionThenCommandLine(sizeAfterPrevSort, ret.size());
+
+ // Replace matches after the first with potential spill sections.
+ if (!spills.empty()) {
+ for (InputSectionBase *&sec : ret) {
+ if (!spills.contains(sec))
+ continue;
+
+ SpillInputSection *sis = make<SpillInputSection>(
+ sec, const_cast<InputSectionDescription *>(cmd));
+
+ // Append the spill input section to the list for the input section,
+ // creating it if necessary.
+ auto res = spillLists.try_emplace(sec, SpillList{sis, sis});
+ if (!res.second) {
+ SpillInputSection *&tail = res.first->second.tail;
+ tail = tail->next = sis;
+ }
+
+ sec = sis;
+ }
+ }
+
return ret;
}
@@ -577,7 +627,7 @@ void LinkerScript::discardSynthetic(OutputSection &outCmd) {
part.armExidx->exidxSections.end());
for (SectionCommand *cmd : outCmd.commands)
if (auto *isd = dyn_cast<InputSectionDescription>(cmd))
- for (InputSectionBase *s : computeInputSections(isd, secs))
+ for (InputSectionBase *s : computeInputSections(isd, secs, outCmd))
discard(*s);
}
}
@@ -588,7 +638,7 @@ LinkerScript::createInputSectionList(OutputSection &outCmd) {
for (SectionCommand *cmd : outCmd.commands) {
if (auto *isd = dyn_cast<InputSectionDescription>(cmd)) {
- isd->sectionBases = computeInputSections(isd, ctx.inputSections);
+ isd->sectionBases = computeInputSections(isd, ctx.inputSections, outCmd);
for (InputSectionBase *s : isd->sectionBases)
s->parent = &outCmd;
ret.insert(ret.end(), isd->sectionBases.begin(), isd->sectionBases.end());
@@ -644,6 +694,9 @@ void LinkerScript::processSectionCommands() {
// Process OVERWRITE_SECTIONS first so that it can overwrite the main script
// or orphans.
+ if (config->enableNonContiguousRegions && !overwriteSections.empty())
+ error("OVERWRITE_SECTIONS cannot be used with "
+ "--enable-non-contiguous-regions");
DenseMap<CachedHashStringRef, OutputDesc *> map;
size_t i = 0;
for (OutputDesc *osd : overwriteSections) {
@@ -911,6 +964,13 @@ void LinkerScript::diagnoseMissingSGSectionAddress() const {
error("no address assigned to the veneers output section " + sec->name);
}
+void LinkerScript::copySpillList(InputSectionBase *dst, InputSectionBase *src) {
+ auto i = spillLists.find(src);
+ if (i == spillLists.end())
+ return;
+ spillLists.try_emplace(dst, i->second);
+}
+
// This function searches for a memory region to place the given output
// section in. If found, a pointer to the appropriate memory region is
// returned in the first member of the pair. Otherwise, a nullptr is returned.
@@ -1066,8 +1126,16 @@ void LinkerScript::assignOffsets(OutputSection *sec) {
// Handle a single input section description command.
// It calculates and assigns the offsets for each section and also
// updates the output section size.
- for (InputSection *isec : cast<InputSectionDescription>(cmd)->sections) {
+
+ DenseSet<InputSection *> spills;
+ auto §ions = cast<InputSectionDescription>(cmd)->sections;
+ for (InputSection *isec : sections) {
assert(isec->getParent() == sec);
+
+ // Skip all possible spills.
+ if (isa<SpillInputSection>(isec))
+ continue;
+
const uint64_t pos = dot;
dot = alignToPowerOf2(dot, isec->addralign);
isec->outSecOff = dot - sec->addr;
@@ -1364,6 +1432,107 @@ const Defined *LinkerScript::assignAddresses() {
return getChangedSymbolAssignment(oldValues);
}
+static bool isRegionOverflowed(MemoryRegion *mr) {
+ if (!mr)
+ return false;
+ return mr->curPos - mr->getOrigin() > mr->getLength();
+}
+
+// Spill input sections in reverse order of address assignment to (potentially)
+// bring memory regions out of overflow. The size savings of a spill can only be
+// estimated, since general linker script arithmetic may occur afterwards.
+// Under-estimates may cause unnecessary spills, but over-estimates can always
+// be corrected on the next pass.
+bool LinkerScript::spillSections() {
+ if (!config->enableNonContiguousRegions)
+ return false;
+
+ bool spilled = false;
+ for (SectionCommand *cmd : reverse(sectionCommands)) {
+ auto *od = dyn_cast<OutputDesc>(cmd);
+ if (!od)
+ continue;
+ OutputSection *osec = &od->osec;
+ if (!osec->size || !osec->memRegion)
+ continue;
+
+ DenseSet<InputSection *> spills;
+ for (SectionCommand *cmd : reverse(osec->commands)) {
+ if (!isRegionOverflowed(osec->memRegion) &&
+ !isRegionOverflowed(osec->lmaRegion))
+ break;
+
+ auto *is = dyn_cast<InputSectionDescription>(cmd);
+ if (!is)
+ continue;
+ for (InputSection *isec : reverse(is->sections)) {
+ // Potential spill locations cannot be spilled.
+ if (isa<SpillInputSection>(isec))
+ continue;
+
+ // Find the next spill location.
+ auto it = spillLists.find(isec);
+ if (it == spillLists.end())
+ continue;
+
+ spilled = true;
+ SpillList &list = it->second;
+
+ SpillInputSection *spill = list.head;
+ if (!spill->next)
+ spillLists.erase(isec);
+ else
+ list.head = spill->next;
+
+ spills.insert(isec);
+
+ // Replace the next spill location with the spilled section and adjust
+ // its properties to match the new location.
+ *llvm::find(spill->isd->sections, spill) = isec;
+ isec->parent = spill->parent;
+ // The alignment of the spill section may have diverged from the
+ // original, but correct assignment requires the spill's alignment,
+ // not the original.
+ isec->addralign = spill->addralign;
+
+ // Record the reduction in overage.
+ osec->memRegion->curPos -= isec->getSize();
+ if (osec->lmaRegion)
+ osec->lmaRegion->curPos -= isec->getSize();
+ if (!isRegionOverflowed(osec->memRegion) &&
+ !isRegionOverflowed(osec->lmaRegion))
+ break;
+ }
+ // Remove any spilled sections.
+ if (!spills.empty())
+ llvm::erase_if(is->sections, [&](InputSection *isec) {
+ return spills.contains(isec);
+ });
+ }
+ }
+
+ return spilled;
+}
+
+// Erase any potential spill sections that were not used.
+void LinkerScript::eraseSpillSections() {
+ if (spillLists.empty())
+ return;
+
+ // Collect the set of input section descriptions that contain potential
+ // spills.
+ DenseSet<InputSectionDescription *> isds;
+ for (const auto &[_, list] : spillLists)
+ for (SpillInputSection *s = list.head; s; s = s->next)
+ isds.insert(s->isd);
+
+ for (InputSectionDescription *isd : isds)
+ llvm::erase_if(isd->sections,
+ [](InputSection *s) { return isa<SpillInputSection>(s); });
+
+ spillLists.clear();
+}
+
// Creates program headers as instructed by PHDRS linker script command.
SmallVector<PhdrEntry *, 0> LinkerScript::createPhdrs() {
SmallVector<PhdrEntry *, 0> ret;
diff --git a/lld/ELF/LinkerScript.h b/lld/ELF/LinkerScript.h
index fa7c6eb9c0d8f79..554df0ea8857123 100644
--- a/lld/ELF/LinkerScript.h
+++ b/lld/ELF/LinkerScript.h
@@ -10,6 +10,7 @@
#define LLD_ELF_LINKER_SCRIPT_H
#include "Config.h"
+#include "InputSection.h"
#include "Writer.h"
#include "lld/Common/LLVM.h"
#include "lld/Common/Strings.h"
@@ -287,7 +288,8 @@ class LinkerScript final {
SmallVector<InputSectionBase *, 0>
computeInputSections(const InputSectionDescription *,
- ArrayRef<InputSectionBase *>);
+ ArrayRef<InputSectionBase *>,
+ const OutputSection &outCmd);
SmallVector<InputSectionBase *, 0> createInputSectionList(OutputSection &cmd);
@@ -312,6 +314,15 @@ class LinkerScript final {
uint64_t dot;
+ // List of potential spill locations (SpillInputSection) for an input
+ // section.
+ struct SpillList {
+ // Never nullptr.
+ SpillInputSection *head;
+ SpillInputSection *tail;
+ };
+ llvm::DenseMap<InputSectionBase *, SpillList> spillLists;
+
public:
OutputDesc *createOutputSection(StringRef name, StringRef location);
OutputDesc *getOrCreateOutputSection(StringRef name);
@@ -325,6 +336,7 @@ class LinkerScript final {
void addOrphanSections();
void diagnoseOrphanHandling() const;
void diagnoseMissingSGSectionAddress() const;
+ void copySpillList(InputSectionBase *dst, InputSectionBase *src);
void adjustOutputSections();
void adjustSectionsAfterSorting();
@@ -333,6 +345,8 @@ class LinkerScript final {
bool shouldKeep(InputSectionBase *s);
const Defined *assignAddresses();
+ bool spillSections();
+ void eraseSpillSections();
void allocateHeaders(SmallVector<PhdrEntry *, 0> &phdrs);
void processSectionCommands();
void processSymbolAssignments();
diff --git a/lld/ELF/Options.td b/lld/ELF/Options.td
index 72eaf157a181cfe..0e5d15124d260bf 100644
--- a/lld/ELF/Options.td
+++ b/lld/ELF/Options.td
@@ -196,6 +196,9 @@ def emit_relocs: F<"emit-relocs">, HelpText<"Generate relocations in output">;
def enable_new_dtags: F<"enable-new-dtags">,
HelpText<"Enable new dynamic tags (default)">;
+def enable_non_contiguous_regions : FF<"enable-non-contiguous-regions">,
+ HelpText<"Spill input sections to later matching output sections to avoid memory region overflow">;
+
def end_group: F<"end-group">,
HelpText<"Ignored for compatibility with GNU unless you pass --warn-backrefs">;
diff --git a/lld/ELF/OutputSections.cpp b/lld/ELF/OutputSections.cpp
index eadab9d745d6870..1a6c5b7a09f4cee 100644
--- a/lld/ELF/OutputSections.cpp
+++ b/lld/ELF/OutputSections.cpp
@@ -186,7 +186,7 @@ static MergeSyntheticSection *createMergeSynthetic(StringRef name,
// new synthetic sections at the location of the first input section
// that it replaces. It then finalizes each synthetic section in order
// to compute an output offset for each piece of each input section.
-void OutputSection::finalizeInputSections() {
+void OutputSection::finalizeInputSections(LinkerScript *script) {
std::vector<MergeSyntheticSection *> mergeSections;
for (SectionCommand *cmd : commands) {
auto *isd = dyn_cast<InputSectionDescription>(cmd);
@@ -226,6 +226,8 @@ void OutputSection::finalizeInputSections() {
i = std::prev(mergeSections.end());
syn->entsize = ms->entsize;
isd->sections.push_back(syn);
+ if (script)
+ script->copySpillList(syn, ms);
}
(*i)->addSection(ms);
}
diff --git a/lld/ELF/OutputSections.h b/lld/ELF/OutputSections.h
index 421a0181feb5dfa..78fede48a23f257 100644
--- a/lld/ELF/OutputSections.h
+++ b/lld/ELF/OutputSections.h
@@ -75,7 +75,7 @@ class OutputSection final : public SectionBase {
void recordSection(InputSectionBase *isec);
void commitSection(InputSection *isec);
- void finalizeInputSections();
+ void finalizeInputSections(LinkerScript *script = nullptr);
// The following members are normally only used in linker scripts.
MemoryRegion *memRegion = nullptr;
diff --git a/lld/ELF/SyntheticSections.cpp b/lld/ELF/SyntheticSections.cpp
index 7b9ada40c0f67b9..298c714adb3b438 100644
--- a/lld/ELF/SyntheticSections.cpp
+++ b/lld/ELF/SyntheticSections.cpp
@@ -4074,6 +4074,13 @@ static bool isDuplicateArmExidxSec(InputSection *prev, InputSection *cur) {
// InputSection with the highest address and any InputSections that have
// mergeable .ARM.exidx table entries are removed from it.
void ARMExidxSyntheticSection::finalizeContents() {
+ // Ensure that any fixed-point iterations after the first see the original set
+ // of sections.
+ if (!originalExecutableSections.empty())
+ executableSections = originalExecutableSections;
+ else if (config->enableNonContiguousRegions)
+ originalExecutableSections = executableSections;
+
// The executableSections and exidxSections that we use to derive the final
// contents of this SyntheticSection are populated before
// processSectionCommands() and ICF. A /DISCARD/ entry in SECTIONS command or
diff --git a/lld/ELF/SyntheticSections.h b/lld/ELF/SyntheticSections.h
index 995fd4b344b07ef..34949025a45f785 100644
--- a/lld/ELF/SyntheticSections.h
+++ b/lld/ELF/SyntheticSections.h
@@ -1255,6 +1255,10 @@ class ARMExidxSyntheticSection : public SyntheticSection {
// either find the .ARM.exidx section or know that we need to generate one.
SmallVector<InputSection *, 0> executableSections;
+ // Value of executableSecitons before finalizeContents(), so that it can be
+ // run repeateadly during fixed point iteration.
+ SmallVector<InputSection *, 0> originalExecutableSections;
+
// The executable InputSection with the highest address to use for the
// sentinel. We store separately from ExecutableSections as merging of
// duplicate entries may mean this InputSection is removed from
diff --git a/lld/ELF/Writer.cpp b/lld/ELF/Writer.cpp
index 240c16a4d8f69b2..4ef027704aa32f0 100644
--- a/lld/ELF/Writer.cpp
+++ b/lld/ELF/Writer.cpp
@@ -1419,6 +1419,8 @@ template <class ELFT> void Writer<ELFT>::finalizeAddressDependentContent() {
for (;;) {
bool changed = target->needsThunks ? tc.createThunks(pass, outputSections)
: target->relaxOnce(pass);
+ bool spilled = script->spillSections();
+ changed |= spilled;
++pass;
// With Thunk Size much smaller than branch range we expect to
@@ -1464,6 +1466,12 @@ template <class ELFT> void Writer<ELFT>::finalizeAddressDependentContent() {
" does not converge");
break;
}
+ } else if (spilled) {
+ // Spilling can change relative section order, so recompute anything that
+ // depends on it.
+ for (Partition &part : partitions)
+ finalizeSynthetic(part.armExidx.get());
+ resolveShfLinkOrder();
}
}
if (!config->relocatable)
@@ -1483,6 +1491,10 @@ template <class ELFT> void Writer<ELFT>::finalizeAddressDependentContent() {
osec->name + " is not a multiple of alignment (" +
Twine(osec->addralign) + ")");
}
+
+ // Sizes are no longer allowed to grow, so all allowable spills have been
+ // taken. Remove any leftover potential spills.
+ script->eraseSpillSections();
}
// If Input Sections have been shrunk (basic block sections) then
diff --git a/lld/docs/ELF/linker_script.rst b/lld/docs/ELF/linker_script.rst
index 3606ef4fe4b8ee1..7a35534be096c26 100644
--- a/lld/docs/ELF/linker_script.rst
+++ b/lld/docs/ELF/linker_script.rst
@@ -197,3 +197,14 @@ the current location to a max-page-size boundary, ensuring that the next
LLD will insert ``.relro_padding`` immediately before the symbol assignment
using ``DATA_SEGMENT_RELRO_END``.
+
+Non-contiguous regions
+~~~~~~~~~~~~~~~~~~~~~~
+
+The flag ``--enable-non-contiguous-regions`` allows input sections to spill to
+later matches rather than causing the link to fail by overflowing a memory
+region. Unlike GNU ld, ``/DISCARD/`` only matches previously-unmatched sections
+(i.e., the flag does not affect it). Also, if a section fails to fit at any of
+its matches, the link fails instead of discarding the section. Accordingly, the
+GNU flag ``--enable-non-contiguous-regions-warnings`` is not implemented, as it
+exists to warn about such occurrences.
diff --git a/lld/docs/ReleaseNotes.rst b/lld/docs/ReleaseNotes.rst
index a7ed49726fd99af..0c558a13c434bd4 100644
--- a/lld/docs/ReleaseNotes.rst
+++ b/lld/docs/ReleaseNotes.rst
@@ -35,6 +35,17 @@ ELF Improvements
* ``--debug-names`` is added to create a merged ``.debug_names`` index
from input ``.debug_names`` sections. Type units are not handled yet.
(`#86508 <https://github.com/llvm/llvm-project/pull/86508>`_)
+* ``--fat-lto-objects`` option is added to support LLVM FatLTO.
+ Without ``--fat-lto-objects``, LLD will link LLVM FatLTO objects using the
+ relocatable object file. (`D146778 <https://reviews.llvm.org/D146778>`_)
+* common-page-size can now be larger than the system page-size.
+ (`#57618 <https://github.com/llvm/llvm-project/issues/57618>`_)
+* ``--enable-non-contiguous-regions`` option allows automatically packing input
+ sections into memory regions by automatically spilling to later matches if a
+ region would overflow. This reduces the toil of manually packing regions
+ (typical for embedded). It also makes full LTO feasible in such cases, since
+ IR merging currently prevents the linker script from referring to input
+ files.
Breaking changes
----------------
diff --git a/lld/test/ELF/linkerscript/enable-non-contiguous-regions-arm-exidx.test b/lld/test/ELF/linkerscript/enable-non-contiguous-regions-arm-exidx.test
new file mode 100644
index 000000000000000..8ead54f7cd8bdb6
--- /dev/null
+++ b/lld/test/ELF/linkerscript/enable-non-contiguous-regions-arm-exidx.test
@@ -0,0 +1,54 @@
+# If a spilling reorders input sections, the .ARM.exidx table must be rebuilt
+# using the new order.
+
+RUN: split-file %s %t
+RUN: llvm-mc -filetype=obj -triple=armv7a-none-linux-gnueabi %t/test.s -o %t/test.o
+RUN: ld.lld -T %t/test.ld %t/test.o -o %t/test --enable-non-contiguous-regions
+RUN: llvm-readobj -x .ARM.exidx %t/test | FileCheck %s
+
+CHECK: 20000000 08849780 1c000000 10849880
+CHECK-NEXT: 18000000 20849980 14000000 01000000
+
+#--- test.ld
+MEMORY {
+ exidx : ORIGIN = 0, LENGTH = 32
+ a : ORIGIN = 32, LENGTH = 4
+ b : ORIGIN = 36, LENGTH = 4
+ c : ORIGIN = 40, LENGTH = 4
+}
+
+SECTIONS {
+ .ARM.exidx : { *(.ARM.exidx) } >exidx
+ .first_chance : { *(.text .text.f2) } >a
+ .text.f1 : { *(.text.f1) } >b
+ .last_chance : { *(.text.f2) } >c
+}
+
+#--- test.s
+ .syntax unified
+ .section .text, "ax",%progbits
+ .globl _start
+_start:
+ .fnstart
+ bx lr
+ .save {r7, lr}
+ .setfp r7, sp, #0
+ .fnend
+
+ .section .text.f1, "ax", %progbits
+ .globl f1
+f1:
+ .fnstart
+ bx lr
+ .save {r8, lr}
+ .setfp r8, sp, #0
+ .fnend
+
+ .section .text.f2, "ax", %progbits
+ .globl f2
+f2:
+ .fnstart
+ bx lr
+ .save {r9, lr}
+ .setfp r9, sp, #0
+ .fnend
diff --git a/lld/test/ELF/linkerscript/enable-non-contiguous-regions.test b/lld/test/ELF/linkerscript/enable-non-contiguous-regions.test
new file mode 100644
index 000000000000000..d59c80b6f9220c4
--- /dev/null
+++ b/lld/test/ELF/linkerscript/enable-non-contiguous-regions.test
@@ -0,0 +1,279 @@
+REQUIRES: x86
+
+RUN: split-file %s %t
+RUN: llvm-mc -n -filetype=obj -triple=x86_64 %t/spill.s -o %t/spill.o
+
+# An input section must spill to a later match if the region of its first match
+# would overflow.
+
+RUN: ld.lld -T %t/spill.ld %t/spill.o -o %t/spill --enable-non-contiguous-regions
+RUN: llvm-readelf -S %t/spill | FileCheck %s -check-prefix=SPILL
+
+SPILL: Name Type Address Off Size
+SPILL: .first_chance PROGBITS 0000000000000000 001000 000001
+SPILL-NEXT: .last_chance PROGBITS 0000000000000002 001002 000002
+
+# A spill off the end must still fails the link.
+
+RUN: not ld.lld -T %t/spill-fail.ld %t/spill.o -o %t/spill-fail --enable-non-contiguous-regions 2>&1 |\
+RUN: FileCheck %s -check-prefix=SPILL-FAIL
+
+SPILL-FAIL: error: section '.last_chance' will not fit in region 'b': overflowed by 2 bytes
+
+# The above spill must still occur if the LMA would overflow, even if the VMA
+# would fit.
+
+RUN: ld.lld -T %t/spill-lma.ld %t/spill.o -o %t/spill-lma --enable-non-contiguous-regions
+RUN: llvm-readelf -S %t/spill-lma | FileCheck %s -check-prefix=SPILL-LMA
+
+SPILL-LMA: Name Type Address Off Size
+SPILL-LMA: .first_chance PROGBITS 0000000000000000 001000 000001
+SPILL-LMA-NEXT: .last_chance PROGBITS 0000000000000003 001003 000002
+
+# A spill must be able to occur to an additional match after the first.
+
+RUN: ld.lld -T %t/spill-later.ld %t/spill.o -o %t/spill-later --enable-non-contiguous-regions
+RUN: llvm-readelf -S %t/spill-later | FileCheck %s -check-prefix=SPILL-LATER
+
+SPILL-LATER: Name Type Address Off Size
+SPILL-LATER: .first_chance PROGBITS 0000000000000000 001000 000001
+SPILL-LATER-NEXT: .second_chance PROGBITS 0000000000000002 001001 000000
+SPILL-LATER-NEXT: .last_chance PROGBITS 0000000000000003 001003 000002
+
+# A later overflow must be able to cause an earlier section to spill.
+
+RUN: ld.lld -T %t/spill-earlier.ld %t/spill.o -o %t/spill-earlier --enable-non-contiguous-regions
+RUN: llvm-readelf -S %t/spill-earlier | FileCheck %s -check-prefix=SPILL-EARLIER
+
+SPILL-EARLIER: Name Type Address Off Size
+SPILL-EARLIER: .first_chance PROGBITS 0000000000000000 001000 000002
+SPILL-EARLIER-NEXT: .last_chance PROGBITS 0000000000000002 001002 000001
+
+# An additional match in /DISCARD/ must have no effect.
+
+RUN: not ld.lld -T %t/no-spill-into-discard.ld %t/spill.o -o %t/no-spill-into-discard --enable-non-contiguous-regions 2>&1 |\
+RUN: FileCheck %s -check-prefix=NO-SPILL-INTO-DISCARD
+
+NO-SPILL-INTO-DISCARD: error: section '.osec' will not fit in region 'a': overflowed by 1 bytes
+
+# An additional match after /DISCARD/ must have no effect.
+
+RUN: ld.lld -T %t/no-spill-from-discard.ld %t/spill.o -o %t/no-spill-from-discard --enable-non-contiguous-regions
+RUN: llvm-readelf -S %t/no-spill-from-discard | FileCheck %s -check-prefix=NO-SPILL-FROM-DISCARD
+
+NO-SPILL-FROM-DISCARD: Name Type Address Off Size
+NO-SPILL-FROM-DISCARD-NOT: .osec
+
+# A spill must use the alignment of the later match.
+
+RUN: ld.lld -T %t/spill-align.ld %t/spill.o -o %t/spill-align --enable-non-contiguous-regions
+RUN: llvm-readelf -S %t/spill-align | FileCheck %s -check-prefix=SPILL-ALIGN
+
+SPILL-ALIGN: Name Type Address Off Size
+SPILL-ALIGN: .first_chance PROGBITS 0000000000000000 000158 000000
+SPILL-ALIGN-NEXT: .last_chance PROGBITS 0000000000000008 001008 00000a
+
+# SHF_MERGEd sections must be spilled according to the matches of the first
+# merged input section (the one giving the resulting section its name).
+
+RUN: llvm-mc -n -filetype=obj -triple=x86_64 %t/merge.s -o %t/merge.o
+RUN: ld.lld -T %t/spill-merge.ld %t/merge.o -o %t/spill-merge --enable-non-contiguous-regions
+RUN: llvm-readelf -S %t/spill-merge | FileCheck %s -check-prefix=SPILL-MERGE
+
+SPILL-MERGE: Name Type Address Off Size
+SPILL-MERGE: .first PROGBITS 0000000000000000 000190 000000
+SPILL-MERGE-NEXT: .second PROGBITS 0000000000000001 001001 000002
+SPILL-MERGE-NEXT: .third PROGBITS 0000000000000003 001003 000000
+
+# An error must be reported for INSERT.
+
+RUN: not ld.lld -T %t/insert.ld %t/spill.o -o %t/insert --enable-non-contiguous-regions 2>&1 |\
+RUN: FileCheck %s -check-prefix=INSERT
+
+INSERT: error: INSERT cannot be used with --enable-non-contiguous-regions
+
+# An error must be reported for OVERWRITE_SECTIONS.
+
+RUN: not ld.lld -T %t/overwrite-sections.ld %t/spill.o -o %t/overwrite-sections --enable-non-contiguous-regions 2>&1 |\
+RUN: FileCheck %s -check-prefix=OVERWRITE_SECTIONS
+
+OVERWRITE_SECTIONS: error: OVERWRITE_SECTIONS cannot be used with --enable-non-contiguous-regions
+
+# SHF_LINK_ORDER must be reordered if spilling changes relative section order.
+
+RUN: llvm-mc -n -filetype=obj -triple=x86_64 %t/link-order.s -o %t/link-order.o
+RUN: ld.lld -T %t/link-order.ld %t/link-order.o -o %t/link-order --enable-non-contiguous-regions
+RUN: llvm-readobj -x .order %t/link-order | FileCheck %s -check-prefix=LINK-ORDER
+
+LINK-ORDER: 0201
+
+#--- spill.s
+.section .one_byte_section,"a", at progbits
+.fill 1
+
+.section .two_byte_section,"a", at progbits
+.fill 2
+
+#--- spill.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 2
+ b : ORIGIN = 2, LENGTH = 2
+}
+
+SECTIONS {
+ .first_chance : { *(.one_byte_section) *(.two_byte_section) } >a
+ .last_chance : { *(.two_byte_section) } >b
+}
+
+#--- spill-fail.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 1
+ b : ORIGIN = 2, LENGTH = 0
+}
+
+SECTIONS {
+ .first_chance : { *(.one_byte_section) *(.two_byte_section) } >a
+ .last_chance : { *(.two_byte_section) } >b
+}
+
+#--- spill-lma.ld
+MEMORY {
+ vma_a : ORIGIN = 0, LENGTH = 3
+ vma_b : ORIGIN = 3, LENGTH = 3
+ lma_a : ORIGIN = 6, LENGTH = 2
+ lma_b : ORIGIN = 8, LENGTH = 2
+}
+
+SECTIONS {
+ .first_chance : { *(.one_byte_section) *(.two_byte_section) } >vma_a AT>lma_a
+ .last_chance : { *(.two_byte_section) } >vma_b AT>lma_b
+}
+
+#--- spill-later.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 2
+ b : ORIGIN = 2, LENGTH = 1
+ c : ORIGIN = 3, LENGTH = 2
+}
+
+SECTIONS {
+ .first_chance : { *(.one_byte_section) *(.two_byte_section) } >a
+ .second_chance : { *(.two_byte_section) } >b
+ .last_chance : { *(.two_byte_section) } >c
+}
+
+#--- spill-earlier.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 2
+ b : ORIGIN = 2, LENGTH = 1
+}
+
+SECTIONS {
+ .first_chance : { *(.one_byte_section) *(.two_byte_section) } >a
+ .last_chance : { *(.one_byte_section) } >b
+}
+
+#--- no-spill-into-discard.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 1
+}
+
+SECTIONS {
+ .osec : { *(.two_byte_section) } >a
+ /DISCARD/ : { *(.one_byte_section) *(.two_byte_section) }
+}
+
+#--- no-spill-from-discard.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 2
+}
+
+SECTIONS {
+ /DISCARD/ : { *(.one_byte_section) *(.two_byte_section) }
+ .osec : { *(.two_byte_section) } >a
+}
+
+#--- spill-align.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 0
+ b : ORIGIN = 2, LENGTH = 16
+}
+
+SECTIONS {
+ .first_chance : SUBALIGN(1) { *(.two_byte_section) } >a
+ .last_chance : SUBALIGN(8) { *(.one_byte_section) *(.two_byte_section) } >b
+}
+
+#--- merge.s
+.section .a,"aM", at progbits,1
+.byte 0x12, 0x34
+
+.section .b,"aM", at progbits,1
+.byte 0x12
+
+#--- spill-merge.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 1
+ b : ORIGIN = 1, LENGTH = 2
+ c : ORIGIN = 3, LENGTH = 2
+}
+
+SECTIONS {
+ .first : { *(.a) *(.b) } >a
+ .second : { *(.a) } >b
+ .third : { *(.b) } >c
+}
+
+#--- insert.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 1
+}
+
+SECTIONS {
+ .a : { *(.two_byte_section) } >a
+}
+
+SECTIONS {
+ .b : { *(.one_byte_section) } >a
+} INSERT AFTER .a;
+
+#--- overwrite-sections.ld
+MEMORY {
+ a : ORIGIN = 0, LENGTH = 1
+}
+
+SECTIONS {
+ .a : { *(.two_byte_section) } >a
+}
+
+OVERWRITE_SECTIONS {
+ .b : { *(.one_byte_section) } >a
+}
+
+#--- link-order.s
+.section .a,"a", at progbits
+.fill 1
+
+.section .b,"a", at progbits
+.fill 1
+
+.section .link_order.a,"ao", at progbits,.a
+.byte 1
+
+.section .link_order.b,"ao", at progbits,.b
+.byte 2
+
+#--- link-order.ld
+MEMORY {
+ order : ORIGIN = 0, LENGTH = 2
+ potential_a : ORIGIN = 2, LENGTH = 0
+ b : ORIGIN = 2, LENGTH = 1
+ actual_a : ORIGIN = 3, LENGTH = 1
+}
+
+SECTIONS {
+ .order : { *(.link_order.*) } > order
+ .potential_a : { *(.a) } >potential_a
+ .b : { *(.b) } >b
+ .actual_a : { *(.a) } >actual_a
+}
More information about the llvm-commits
mailing list