Skip to content

Commit

Permalink
[pdb] Correctly parse the hash adjusters table from TPI stream.
Browse files Browse the repository at this point in the history
This is not a list of pairs, it is a hash table data structure. We now
correctly parse this out and dump it from llvm-pdbdump.

We still need to understand the conditions that lead to a type
getting an entry in the hash adjuster table.  That will be done
in a followup investigation / patch.

Differential Revision: https://reviews.llvm.org/D29090

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@293090 91177308-0d34-0410-b5e6-96231b3b80d8
  • Loading branch information
Zachary Turner committed Jan 25, 2017
1 parent d1752d7 commit a74ee64
Show file tree
Hide file tree
Showing 4 changed files with 95 additions and 77 deletions.
5 changes: 3 additions & 2 deletions include/llvm/DebugInfo/PDB/Raw/TpiStream.h
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
#include "llvm/DebugInfo/CodeView/TypeRecord.h"
#include "llvm/DebugInfo/MSF/StreamArray.h"
#include "llvm/DebugInfo/PDB/PDBTypes.h"
#include "llvm/DebugInfo/PDB/Raw/HashTable.h"
#include "llvm/DebugInfo/PDB/Raw/RawConstants.h"
#include "llvm/DebugInfo/PDB/Raw/RawTypes.h"
#include "llvm/Support/raw_ostream.h"
Expand Down Expand Up @@ -47,7 +48,7 @@ class TpiStream {
uint32_t NumHashBuckets() const;
msf::FixedStreamArray<support::ulittle32_t> getHashValues() const;
msf::FixedStreamArray<TypeIndexOffset> getTypeIndexOffsets() const;
msf::FixedStreamArray<TypeIndexOffset> getHashAdjustments() const;
HashTable &getHashAdjusters();

iterator_range<codeview::CVTypeArray::Iterator> types(bool *HadError) const;

Expand All @@ -64,7 +65,7 @@ class TpiStream {
std::unique_ptr<msf::ReadableStream> HashStream;
msf::FixedStreamArray<support::ulittle32_t> HashValues;
msf::FixedStreamArray<TypeIndexOffset> TypeIndexOffsets;
msf::FixedStreamArray<TypeIndexOffset> HashAdjustments;
HashTable HashAdjusters;

const TpiStreamHeader *Header;
};
Expand Down
15 changes: 6 additions & 9 deletions lib/DebugInfo/PDB/Raw/TpiStream.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -113,11 +113,11 @@ Error TpiStream::reload() {
if (auto EC = HSR.readArray(TypeIndexOffsets, NumTypeIndexOffsets))
return EC;

HSR.setOffset(Header->HashAdjBuffer.Off);
uint32_t NumHashAdjustments =
Header->HashAdjBuffer.Length / sizeof(TypeIndexOffset);
if (auto EC = HSR.readArray(HashAdjustments, NumHashAdjustments))
return EC;
if (Header->HashAdjBuffer.Length > 0) {
HSR.setOffset(Header->HashAdjBuffer.Off);
if (auto EC = HashAdjusters.load(HSR))
return EC;
}

HashStream = std::move(HS);

Expand Down Expand Up @@ -164,10 +164,7 @@ TpiStream::getTypeIndexOffsets() const {
return TypeIndexOffsets;
}

FixedStreamArray<TypeIndexOffset>
TpiStream::getHashAdjustments() const {
return HashAdjustments;
}
HashTable &TpiStream::getHashAdjusters() { return HashAdjusters; }

iterator_range<CVTypeArray::Iterator>
TpiStream::types(bool *HadError) const {
Expand Down
14 changes: 6 additions & 8 deletions test/DebugInfo/PDB/pdbdump-headers.test
Original file line number Diff line number Diff line change
Expand Up @@ -152,9 +152,11 @@
; EMPTY-NEXT: Number of Hash Buckets: 262143
; EMPTY-NEXT: Hash Key Size: 4
; EMPTY-NEXT: Values: [205956, 163561, 59811, 208239, 16377, 247078, 194342, 254156, 194536, 167492, 185421, 119540, 261871, 198119, 48056, 251486, 134580, 148190, 113636, 53336, 55779, 220695, 198114, 148734, 81128, 60158, 217249, 174209, 159978, 249504, 141941, 238785, 6214, 94935, 151449, 135589, 73373, 96512, 254299, 17744, 239514, 173189, 130544, 204437, 238560, 144673, 115151, 197306, 256035, 101096, 231280, 52156, 48854, 170035, 177041, 102745, 16947, 183703, 98548, 35693, 171328, 203640, 139292, 49018, 43821, 202555, 165040, 215835, 142625, 52534, 44186, 103930, 110942, 17991, 213215]
; EMPTY-NEXT: Type Index Offsets: [{4096, 0}]
; EMPTY-NEXT: Hash Adjustments: []
; EMPTY-NEXT: }
; EMPTY-NEXT: TypeIndexOffsets [
; EMPTY-NEXT: Index: 0x1000, Offset: 0
; EMPTY-NEXT: ]
; EMPTY: Type Info Stream (IPI) {
; EMPTY-NEXT: IPI Version: 20040203
; EMPTY-NEXT: Record count: 15
Expand Down Expand Up @@ -231,13 +233,9 @@
; EMPTY-NEXT: 0000: 42100000 01000000 6C000000 0100F2F1 |B.......l.......|
; EMPTY-NEXT: )
; EMPTY-NEXT: }
; EMPTY: Hash {
; EMPTY-NEXT: Number of Hash Buckets: 262143
; EMPTY-NEXT: Hash Key Size: 4
; EMPTY-NEXT: Values: [7186, 7198, 7180, 7191, 7201, 7241, 7249, 80727, 154177, 75189, 253662, 193467, 222705, 186099, 257108]
; EMPTY-NEXT: Type Index Offsets: [{4096, 0}]
; EMPTY-NEXT: Hash Adjustments: []
; EMPTY-NEXT: }
; EMPTY: TypeIndexOffsets [
; EMPTY-NEXT: Index: 0x1000, Offset: 0
; EMPTY-NEXT: ]
; EMPTY: DBI Stream {
; EMPTY-NEXT: Dbi Version: 19990903
; EMPTY-NEXT: Age: 1
Expand Down
138 changes: 80 additions & 58 deletions tools/llvm-pdbdump/LLVMOutputStyle.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -510,24 +510,27 @@ static void printTypeIndexOffset(raw_ostream &OS,
OS << "{" << TIOff.Type.getIndex() << ", " << TIOff.Offset << "}";
}

static void dumpTpiHash(ScopedPrinter &P, TpiStream &Tpi) {
if (!opts::raw::DumpTpiHash)
return;
DictScope DD(P, "Hash");
P.printNumber("Number of Hash Buckets", Tpi.NumHashBuckets());
P.printNumber("Hash Key Size", Tpi.getHashKeySize());
P.printList("Values", Tpi.getHashValues());
P.printList("Type Index Offsets", Tpi.getTypeIndexOffsets(),
printTypeIndexOffset);
P.printList("Hash Adjustments", Tpi.getHashAdjustments(),
printTypeIndexOffset);
namespace {
class RecordBytesVisitor : public TypeVisitorCallbacks {
public:
explicit RecordBytesVisitor(ScopedPrinter &P) : P(P) {}

Error visitTypeEnd(CVType &Record) override {
P.printBinaryBlock("Bytes", Record.content());
return Error::success();
}

private:
ScopedPrinter &P;
};
}

Error LLVMOutputStyle::dumpTpiStream(uint32_t StreamIdx) {
assert(StreamIdx == StreamTPI || StreamIdx == StreamIPI);

bool DumpRecordBytes = false;
bool DumpRecords = false;
bool DumpTpiHash = false;
StringRef Label;
StringRef VerLabel;
if (StreamIdx == StreamTPI) {
Expand All @@ -537,6 +540,7 @@ Error LLVMOutputStyle::dumpTpiStream(uint32_t StreamIdx) {
}
DumpRecordBytes = opts::raw::DumpTpiRecordBytes;
DumpRecords = opts::raw::DumpTpiRecords;
DumpTpiHash = opts::raw::DumpTpiHash;
Label = "Type Info Stream (TPI)";
VerLabel = "TPI Version";
} else if (StreamIdx == StreamIPI) {
Expand All @@ -549,80 +553,98 @@ Error LLVMOutputStyle::dumpTpiStream(uint32_t StreamIdx) {
Label = "Type Info Stream (IPI)";
VerLabel = "IPI Version";
}
if (!DumpRecordBytes && !DumpRecords && !opts::raw::DumpModuleSyms)
if (!DumpRecordBytes && !DumpRecords && !DumpTpiHash &&
!opts::raw::DumpModuleSyms)
return Error::success();

bool IsSilentDatabaseBuild = !DumpRecordBytes && !DumpRecords && !DumpTpiHash;

auto Tpi = (StreamIdx == StreamTPI) ? File.getPDBTpiStream()
: File.getPDBIpiStream();
if (!Tpi)
return Tpi.takeError();

// Even if the user doesn't want to dump type records, we still need to
// iterate them in order to build the type database. So when they want to
// dump symbols but not types, don't stick a dumper on the end, just build
// the type database.
std::unique_ptr<DictScope> StreamScope;
std::unique_ptr<ListScope> RecordScope;

if (!IsSilentDatabaseBuild) {
StreamScope = llvm::make_unique<DictScope>(P, Label);
P.printNumber(VerLabel, Tpi->getTpiVersion());
P.printNumber("Record count", Tpi->NumTypeRecords());
}

TypeDatabaseVisitor DBV(TypeDB);
CompactTypeDumpVisitor CTDV(TypeDB, &P);
TypeDumpVisitor TDV(TypeDB, &P, false);
RecordBytesVisitor RBV(P);
TypeDeserializer Deserializer;

// We always need to deserialize and add it to the type database. This is
// true if even if we're not dumping anything, because we could need the
// type database for the purposes of dumping symbols.
TypeVisitorCallbackPipeline Pipeline;
Pipeline.addCallbackToPipeline(Deserializer);
Pipeline.addCallbackToPipeline(DBV);

CVTypeVisitor Visitor(Pipeline);

if (DumpRecords || DumpRecordBytes) {
DictScope D(P, Label);

P.printNumber(VerLabel, Tpi->getTpiVersion());
P.printNumber("Record count", Tpi->NumTypeRecords());
ListScope L(P, "Records");

bool HadError = false;
// If we're in dump mode, add a dumper with the appropriate detail level.
if (DumpRecords) {
if (opts::raw::CompactRecords)
Pipeline.addCallbackToPipeline(CTDV);
else
Pipeline.addCallbackToPipeline(TDV);
}
if (DumpRecordBytes)
Pipeline.addCallbackToPipeline(RBV);

for (auto Type : Tpi->types(&HadError)) {
std::unique_ptr<DictScope> Scope;
if (!opts::raw::CompactRecords)
Scope.reset(new DictScope(P, ""));
CVTypeVisitor Visitor(Pipeline);

if (DumpRecords) {
if (auto EC = Visitor.visitTypeRecord(Type))
return EC;
}
if (DumpRecords || DumpRecordBytes)
RecordScope = llvm::make_unique<ListScope>(P, "Records");

if (DumpRecordBytes)
P.printBinaryBlock("Bytes", Type.content());
}
dumpTpiHash(P, *Tpi);
if (HadError)
return make_error<RawError>(raw_error_code::corrupt_file,
"TPI stream contained corrupt record");
{
ListScope L(P, "TypeIndexOffsets");
for (const auto &IO : Tpi->getTypeIndexOffsets()) {
P.printString(formatv("Index: {0:x}, Offset: {1:N}", IO.Type.getIndex(),
(uint32_t)IO.Offset)
.str());
}
}
bool HadError = false;

TypeIndex T(TypeIndex::FirstNonSimpleIndex);
for (auto Type : Tpi->types(&HadError)) {
std::unique_ptr<DictScope> OneRecordScope;

} else if (opts::raw::DumpModuleSyms) {
if ((DumpRecords || DumpRecordBytes) && !opts::raw::CompactRecords)
OneRecordScope = llvm::make_unique<DictScope>(P, "");

bool HadError = false;
for (auto Type : Tpi->types(&HadError)) {
if (auto EC = Visitor.visitTypeRecord(Type))
return EC;
if (auto EC = Visitor.visitTypeRecord(Type))
return EC;
}
if (HadError)
return make_error<RawError>(raw_error_code::corrupt_file,
"TPI stream contained corrupt record");

if (DumpTpiHash) {
DictScope DD(P, "Hash");
P.printNumber("Number of Hash Buckets", Tpi->NumHashBuckets());
P.printNumber("Hash Key Size", Tpi->getHashKeySize());
P.printList("Values", Tpi->getHashValues());

ListScope LHA(P, "Adjusters");
auto ExpectedST = File.getStringTable();
if (!ExpectedST)
return ExpectedST.takeError();
const auto &ST = *ExpectedST;
for (const auto &E : Tpi->getHashAdjusters()) {
DictScope DHA(P);
StringRef Name = ST.getStringForID(E.first);
P.printString("Type", Name);
P.printHex("TI", E.second);
}
}

dumpTpiHash(P, *Tpi);
if (HadError)
return make_error<RawError>(raw_error_code::corrupt_file,
"TPI stream contained corrupt record");
if (!IsSilentDatabaseBuild) {
ListScope L(P, "TypeIndexOffsets");
for (const auto &IO : Tpi->getTypeIndexOffsets()) {
P.printString(formatv("Index: {0:x}, Offset: {1:N}", IO.Type.getIndex(),
(uint32_t)IO.Offset)
.str());
}
}

P.flush();
return Error::success();
}
Expand Down

0 comments on commit a74ee64

Please sign in to comment.