MCJIT support for non-function sections.

Move to a by-section allocation and relocation scheme. This allows
better support for sections which do not contain externally visible
symbols.

Flesh out the relocation address vs. local storage address separation a
bit more as well. Remote process JITs use this to tell the relocation
resolution code where the code will live when it executes.

The startFunctionBody/endFunctionBody interfaces to the JIT and the
memory manager are deprecated. They'll stick around for as long as the
old JIT does, but the MCJIT doesn't use them anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@148258 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Jim Grosbach
2012-01-16 22:26:39 +00:00
parent 27bf56056b
commit 61425c0a7f
10 changed files with 492 additions and 259 deletions

View File

@@ -16,6 +16,7 @@
#include "llvm/ExecutionEngine/RuntimeDyld.h"
#include "llvm/Object/MachOObject.h"
#include "llvm/ADT/IndexedMap.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/Twine.h"
#include "llvm/ADT/SmallVector.h"
@@ -40,17 +41,18 @@ protected:
// The MemoryManager to load objects into.
RTDyldMemoryManager *MemMgr;
// FIXME: This all assumes we're dealing with external symbols for anything
// explicitly referenced. I.e., we can index by name and things
// will work out. In practice, this may not be the case, so we
// should find a way to effectively generalize.
// For each function, we have a MemoryBlock of it's instruction data.
StringMap<sys::MemoryBlock> Functions;
// For each section, we have a MemoryBlock of it's data.
// Indexed by SectionID.
SmallVector<sys::MemoryBlock, 32> Sections;
// For each section, the address it will be considered to live at for
// relocations. The same as the pointer the above memory block for hosted
// JITs. Indexed by SectionID.
SmallVector<uint64_t, 32> SectionLoadAddress;
// Master symbol table. As modules are loaded and external symbols are
// resolved, their addresses are stored here.
StringMap<uint8_t*> SymbolTable;
// resolved, their addresses are stored here as a SectionID/Offset pair.
typedef std::pair<unsigned, uint64_t> SymbolLoc;
StringMap<SymbolLoc> SymbolTable;
bool HasError;
std::string ErrorStr;
@@ -62,6 +64,9 @@ protected:
return true;
}
uint8_t *getSectionAddress(unsigned SectionID) {
return (uint8_t*)Sections[SectionID].base();
}
void extractFunction(StringRef Name, uint8_t *StartAddress,
uint8_t *EndAddress);
@@ -75,12 +80,15 @@ public:
void *getSymbolAddress(StringRef Name) {
// FIXME: Just look up as a function for now. Overly simple of course.
// Work in progress.
return SymbolTable.lookup(Name);
if (SymbolTable.find(Name) == SymbolTable.end())
return 0;
SymbolLoc Loc = SymbolTable.lookup(Name);
return getSectionAddress(Loc.first) + Loc.second;
}
void resolveRelocations();
virtual void resolveRelocations();
virtual void reassignSymbolAddress(StringRef Name, uint8_t *Addr) = 0;
virtual void reassignSectionAddress(unsigned SectionID, uint64_t Addr) = 0;
// Is the linker in an error state?
bool hasError() { return HasError; }
@@ -128,6 +136,8 @@ class RuntimeDyldELF : public RuntimeDyldImpl {
StringMap<RelocationList> Relocations;
unsigned Arch;
void resolveRelocations();
void resolveX86_64Relocation(StringRef Name,
uint8_t *Addr,
const RelocationEntry &RE);
@@ -150,6 +160,7 @@ public:
bool loadObject(MemoryBuffer *InputBuffer);
void reassignSymbolAddress(StringRef Name, uint8_t *Addr);
void reassignSectionAddress(unsigned SectionID, uint64_t Addr);
bool isCompatibleFormat(const MemoryBuffer *InputBuffer) const;
};
@@ -160,30 +171,34 @@ class RuntimeDyldMachO : public RuntimeDyldImpl {
// For each symbol, keep a list of relocations based on it. Anytime
// its address is reassigned (the JIT re-compiled the function, e.g.),
// the relocations get re-resolved.
// The symbol (or section) the relocation is sourced from is the Key
// in the relocation list where it's stored.
struct RelocationEntry {
std::string Target; // Object this relocation is contained in.
uint64_t Offset; // Offset into the object for the relocation.
unsigned SectionID; // Section the relocation is contained in.
uint64_t Offset; // Offset into the section for the relocation.
uint32_t Data; // Second word of the raw macho relocation entry.
int64_t Addend; // Addend encoded in the instruction itself, if any.
bool isResolved; // Has this relocation been resolved previously?
int64_t Addend; // Addend encoded in the instruction itself, if any,
// plus the offset into the source section for
// the symbol once the relocation is resolvable.
RelocationEntry(StringRef t, uint64_t offset, uint32_t data, int64_t addend)
: Target(t), Offset(offset), Data(data), Addend(addend),
isResolved(false) {}
RelocationEntry(unsigned id, uint64_t offset, uint32_t data, int64_t addend)
: SectionID(id), Offset(offset), Data(data), Addend(addend) {}
};
typedef SmallVector<RelocationEntry, 4> RelocationList;
StringMap<RelocationList> Relocations;
// Relocations to sections already loaded. Indexed by SectionID which is the
// source of the address. The target where the address will be writen is
// SectionID/Offset in the relocation itself.
IndexedMap<RelocationList> Relocations;
// Relocations to symbols that are not yet resolved. Must be external
// relocations by definition. Indexed by symbol name.
StringMap<RelocationList> UnresolvedRelocations;
// FIXME: Also keep a map of all the relocations contained in an object. Use
// this to dynamically answer whether all of the relocations in it have
// been resolved or not.
bool resolveRelocation(uint8_t *Address, uint8_t *Value, bool isPCRel,
unsigned Type, unsigned Size);
bool resolveRelocation(uint8_t *Address, uint64_t Value, bool isPCRel,
unsigned Type, unsigned Size, int64_t Addend);
bool resolveX86_64Relocation(uintptr_t Address, uintptr_t Value, bool isPCRel,
unsigned Type, unsigned Size);
unsigned Type, unsigned Size, int64_t Addend);
bool resolveARMRelocation(uintptr_t Address, uintptr_t Value, bool isPCRel,
unsigned Type, unsigned Size);
unsigned Type, unsigned Size, int64_t Addend);
bool loadSegment32(const MachOObject *Obj,
const MachOObject::LoadCommandInfo *SegmentLCI,
@@ -191,13 +206,23 @@ class RuntimeDyldMachO : public RuntimeDyldImpl {
bool loadSegment64(const MachOObject *Obj,
const MachOObject::LoadCommandInfo *SegmentLCI,
const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC);
bool processSymbols32(const MachOObject *Obj,
SmallVectorImpl<unsigned> &SectionMap,
SmallVectorImpl<StringRef> &SymbolNames,
const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC);
bool processSymbols64(const MachOObject *Obj,
SmallVectorImpl<unsigned> &SectionMap,
SmallVectorImpl<StringRef> &SymbolNames,
const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC);
void resolveSymbol(StringRef Name);
public:
RuntimeDyldMachO(RTDyldMemoryManager *mm) : RuntimeDyldImpl(mm) {}
bool loadObject(MemoryBuffer *InputBuffer);
void reassignSymbolAddress(StringRef Name, uint8_t *Addr);
void reassignSectionAddress(unsigned SectionID, uint64_t Addr);
static bool isKnownFormat(const MemoryBuffer *InputBuffer);