[llvm] [BOLT] Detect Linux kernel based on ELF program headers (PR #80086)
Maksim Panchenko via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 30 16:53:15 PST 2024
https://github.com/maksfb created https://github.com/llvm/llvm-project/pull/80086
Check if program header addresses fall into the kernel space to detect a Linux kernel binary on x86-64.
Delete opts::LinuxKernelMode and use BinaryContext::IsLinuxKernel instead.
>From 8bbda67556e13c7c0465defc80d9100a28c5dd00 Mon Sep 17 00:00:00 2001
From: Maksim Panchenko <maks at fb.com>
Date: Tue, 30 Jan 2024 16:22:55 -0800
Subject: [PATCH] [BOLT] Detect Linux kernel based on ELF program headers
Check if program header addresses fall into the kernel space to detect
a Linux kernel binary on x86-64.
Delete opts::LinuxKernelMode and use BinaryContext::IsLinuxKernel
instead.
---
bolt/include/bolt/Core/BinaryContext.h | 3 +++
bolt/include/bolt/Utils/CommandLineOpts.h | 1 -
bolt/lib/Profile/DataAggregator.cpp | 6 +++---
bolt/lib/Rewrite/LinuxKernelRewriter.cpp | 9 ++++-----
bolt/lib/Rewrite/RewriteInstance.cpp | 16 ++++++++++------
bolt/lib/Utils/CommandLineOpts.cpp | 1 -
bolt/test/X86/linux-orc.s | 10 ++--------
7 files changed, 22 insertions(+), 24 deletions(-)
diff --git a/bolt/include/bolt/Core/BinaryContext.h b/bolt/include/bolt/Core/BinaryContext.h
index f0e7a8272ad0e..65a3442695703 100644
--- a/bolt/include/bolt/Core/BinaryContext.h
+++ b/bolt/include/bolt/Core/BinaryContext.h
@@ -602,6 +602,9 @@ class BinaryContext {
std::unique_ptr<MCAsmBackend> MAB;
+ /// Indicates if the binary is Linux kernel.
+ bool IsLinuxKernel{false};
+
/// Indicates if relocations are available for usage.
bool HasRelocations{false};
diff --git a/bolt/include/bolt/Utils/CommandLineOpts.h b/bolt/include/bolt/Utils/CommandLineOpts.h
index 7b654f19f6d45..30e8bd777b3ca 100644
--- a/bolt/include/bolt/Utils/CommandLineOpts.h
+++ b/bolt/include/bolt/Utils/CommandLineOpts.h
@@ -18,7 +18,6 @@
namespace opts {
extern bool HeatmapMode;
-extern bool LinuxKernelMode;
extern llvm::cl::OptionCategory BoltCategory;
extern llvm::cl::OptionCategory BoltDiffCategory;
diff --git a/bolt/lib/Profile/DataAggregator.cpp b/bolt/lib/Profile/DataAggregator.cpp
index be1e348b338f0..6a64bcde911e6 100644
--- a/bolt/lib/Profile/DataAggregator.cpp
+++ b/bolt/lib/Profile/DataAggregator.cpp
@@ -524,7 +524,7 @@ Error DataAggregator::preprocessProfile(BinaryContext &BC) {
ErrorCallback(ReturnCode, ErrBuf);
};
- if (opts::LinuxKernelMode) {
+ if (BC.IsLinuxKernel) {
// Current MMap parsing logic does not work with linux kernel.
// MMap entries for linux kernel uses PERF_RECORD_MMAP
// format instead of typical PERF_RECORD_MMAP2 format.
@@ -1056,7 +1056,7 @@ ErrorOr<DataAggregator::PerfBranchSample> DataAggregator::parseBranchSample() {
if (std::error_code EC = PIDRes.getError())
return EC;
auto MMapInfoIter = BinaryMMapInfo.find(*PIDRes);
- if (!opts::LinuxKernelMode && MMapInfoIter == BinaryMMapInfo.end()) {
+ if (!BC->IsLinuxKernel && MMapInfoIter == BinaryMMapInfo.end()) {
consumeRestOfLine();
return make_error_code(errc::no_such_process);
}
@@ -1277,7 +1277,7 @@ std::error_code DataAggregator::printLBRHeatMap() {
NamedRegionTimer T("parseBranch", "Parsing branch events", TimerGroupName,
TimerGroupDesc, opts::TimeAggregator);
- if (opts::LinuxKernelMode) {
+ if (BC->IsLinuxKernel) {
opts::HeatmapMaxAddress = 0xffffffffffffffff;
opts::HeatmapMinAddress = KernelBaseAddr;
}
diff --git a/bolt/lib/Rewrite/LinuxKernelRewriter.cpp b/bolt/lib/Rewrite/LinuxKernelRewriter.cpp
index d74177c1c4614..c8674d6b837ad 100644
--- a/bolt/lib/Rewrite/LinuxKernelRewriter.cpp
+++ b/bolt/lib/Rewrite/LinuxKernelRewriter.cpp
@@ -207,9 +207,7 @@ void LinuxKernelRewriter::insertLKMarker(uint64_t PC, uint64_t SectionOffset,
}
void LinuxKernelRewriter::processLKSections() {
- assert(opts::LinuxKernelMode &&
- "process Linux Kernel special sections and their relocations only in "
- "linux kernel mode.\n");
+ assert(BC.IsLinuxKernel && "Linux kernel binary expected.");
processLKExTable();
processLKPCIFixup();
@@ -290,8 +288,9 @@ void LinuxKernelRewriter::processLKExTable() {
void LinuxKernelRewriter::processLKPCIFixup() {
ErrorOr<BinarySection &> SectionOrError =
BC.getUniqueSectionByName(".pci_fixup");
- assert(SectionOrError &&
- ".pci_fixup section not found in Linux Kernel binary");
+ if (!SectionOrError)
+ return;
+
const uint64_t SectionSize = SectionOrError->getSize();
const uint64_t SectionAddress = SectionOrError->getAddress();
assert((SectionSize % 16) == 0 && ".pci_fixup size is not a multiple of 16");
diff --git a/bolt/lib/Rewrite/RewriteInstance.cpp b/bolt/lib/Rewrite/RewriteInstance.cpp
index d24bd18c08409..0645308afafd5 100644
--- a/bolt/lib/Rewrite/RewriteInstance.cpp
+++ b/bolt/lib/Rewrite/RewriteInstance.cpp
@@ -512,6 +512,9 @@ Error RewriteInstance::discoverStorage() {
Phdr.p_offset,
Phdr.p_filesz,
Phdr.p_align};
+ if (BC->TheTriple->getArch() == llvm::Triple::x86_64 &&
+ Phdr.p_vaddr >= 0xFFFF'FFFF'8000'0000)
+ BC->IsLinuxKernel = true;
break;
case ELF::PT_INTERP:
BC->HasInterpHeader = true;
@@ -519,6 +522,9 @@ Error RewriteInstance::discoverStorage() {
}
}
+ if (BC->IsLinuxKernel)
+ outs() << "BOLT-INFO: Linux kernel binary detected\n";
+
for (const SectionRef &Section : InputFile->sections()) {
Expected<StringRef> SectionNameOrErr = Section.getName();
if (Error E = SectionNameOrErr.takeError())
@@ -562,7 +568,7 @@ Error RewriteInstance::discoverStorage() {
if (opts::Hugify && !BC->HasFixedLoadAddress)
NextAvailableAddress += BC->PageAlign;
- if (!opts::UseGnuStack) {
+ if (!opts::UseGnuStack && !BC->IsLinuxKernel) {
// This is where the black magic happens. Creating PHDR table in a segment
// other than that containing ELF header is tricky. Some loaders and/or
// parts of loaders will apply e_phoff from ELF header assuming both are in
@@ -751,7 +757,7 @@ Error RewriteInstance::run() {
if (opts::Instrument && !BC->IsStaticExecutable)
updateRtFiniReloc();
- if (opts::LinuxKernelMode) {
+ if (BC->IsLinuxKernel) {
errs() << "BOLT-WARNING: not writing the output file for Linux Kernel\n";
return Error::success();
} else if (opts::OutputFilename == "/dev/null") {
@@ -1284,7 +1290,7 @@ void RewriteInstance::discoverFileObjects() {
}
}
- if (!opts::LinuxKernelMode) {
+ if (!BC->IsLinuxKernel) {
// Read all relocations now that we have binary functions mapped.
processRelocations();
}
@@ -1813,8 +1819,6 @@ Error RewriteInstance::readSpecialSections() {
<< "\n");
if (isDebugSection(SectionName))
HasDebugInfo = true;
- if (isKSymtabSection(SectionName))
- opts::LinuxKernelMode = true;
}
// Set IsRelro section attribute based on PT_GNU_RELRO segment.
@@ -3037,7 +3041,7 @@ void RewriteInstance::preprocessProfileData() {
}
void RewriteInstance::initializeMetadataManager() {
- if (opts::LinuxKernelMode)
+ if (BC->IsLinuxKernel)
MetadataManager.registerRewriter(createLinuxKernelRewriter(*BC));
MetadataManager.registerRewriter(createPseudoProbeRewriter(*BC));
diff --git a/bolt/lib/Utils/CommandLineOpts.cpp b/bolt/lib/Utils/CommandLineOpts.cpp
index a1df5de262340..e910fa4f86722 100644
--- a/bolt/lib/Utils/CommandLineOpts.cpp
+++ b/bolt/lib/Utils/CommandLineOpts.cpp
@@ -29,7 +29,6 @@ const char *BoltRevision =
namespace opts {
bool HeatmapMode = false;
-bool LinuxKernelMode = false;
cl::OptionCategory BoltCategory("BOLT generic options");
cl::OptionCategory BoltDiffCategory("BOLTDIFF generic options");
diff --git a/bolt/test/X86/linux-orc.s b/bolt/test/X86/linux-orc.s
index bb3ab10754544..3d8a3d77ec2ea 100644
--- a/bolt/test/X86/linux-orc.s
+++ b/bolt/test/X86/linux-orc.s
@@ -3,11 +3,12 @@
## Check that BOLT correctly reads ORC unwind information used by Linux kernel.
# RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown %s -o %t.o
-# RUN: %clang %cflags %t.o -o %t.exe
+# RUN: %clang %cflags -nostdlib %t.o -o %t.exe -Wl,--image-base=0xffffffff80000000,--no-dynamic-linker,--no-eh-frame-hdr
# RUN: llvm-bolt %t.exe --print-normalized --dump-orc --print-orc -o %t.out \
# RUN: |& FileCheck %s
+# CHECK: BOLT-INFO: Linux kernel binary detected
# CHECK: BOLT-INFO: ORC unwind information:
# CHECK-NEXT: {sp: 8, bp: 0, info: 0x5}: _start
# CHECK-NEXT: {sp: 0, bp: 0, info: 0x0}: _start
@@ -22,18 +23,15 @@
.globl _start
.type _start, %function
_start:
- .cfi_startproc
call foo
# CHECK: callq foo # ORC: {sp: 8, bp: 0, info: 0x5}
ret
- .cfi_endproc
.size _start, .-_start
.globl foo
.type foo, %function
foo:
- .cfi_startproc
push %rbp
# CHECK: pushq %rbp # ORC: {sp: 8, bp: 0, info: 0x5}
.L1:
@@ -45,16 +43,13 @@ foo:
.L3:
ret
# CHECK: retq # ORC: {sp: 8, bp: 0, info: 0x5}
- .cfi_endproc
.size foo, .-foo
bar:
- .cfi_startproc
ret
## Same ORC info propagated from foo above.
# CHECK: retq # ORC: {sp: 8, bp: 0, info: 0x5}
.L4:
- .cfi_endproc
.size bar, .-bar
.section .orc_unwind,"a", at progbits
@@ -131,4 +126,3 @@ bar:
## Fake Linux Kernel sections.
.section __ksymtab,"a", at progbits
.section __ksymtab_gpl,"a", at progbits
- .section .pci_fixup,"a", at progbits
More information about the llvm-commits
mailing list