Add support for escaped target names in isolate driver (4th try)
[chromium-blink-merge.git] / courgette / encoded_program.cc
blob6d6e416eb82f7f1f4dc9712fcce8585a372702c9
1 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 #include "courgette/encoded_program.h"
7 #include <algorithm>
8 #include <map>
9 #include <string>
10 #include <vector>
12 #include "base/environment.h"
13 #include "base/logging.h"
14 #include "base/memory/scoped_ptr.h"
15 #include "base/numerics/safe_conversions.h"
16 #include "base/numerics/safe_math.h"
17 #include "base/strings/string_number_conversions.h"
18 #include "base/strings/string_util.h"
19 #include "courgette/courgette.h"
20 #include "courgette/disassembler_elf_32_arm.h"
21 #include "courgette/streams.h"
22 #include "courgette/types_elf.h"
24 namespace courgette {
26 // Stream indexes.
27 const int kStreamMisc = 0;
28 const int kStreamOps = 1;
29 const int kStreamBytes = 2;
30 const int kStreamAbs32Indexes = 3;
31 const int kStreamRel32Indexes = 4;
32 const int kStreamAbs32Addresses = 5;
33 const int kStreamRel32Addresses = 6;
34 const int kStreamCopyCounts = 7;
35 const int kStreamOriginAddresses = kStreamMisc;
37 const int kStreamLimit = 9;
39 // Constructor is here rather than in the header. Although the constructor
40 // appears to do nothing it is fact quite large because of the implicit calls to
41 // field constructors. Ditto for the destructor.
42 EncodedProgram::EncodedProgram() : image_base_(0) {}
43 EncodedProgram::~EncodedProgram() {}
45 // Serializes a vector of integral values using Varint32 coding.
46 template<typename V>
47 CheckBool WriteVector(const V& items, SinkStream* buffer) {
48 size_t count = items.size();
49 bool ok = buffer->WriteSizeVarint32(count);
50 for (size_t i = 0; ok && i < count; ++i) {
51 ok = buffer->WriteSizeVarint32(items[i]);
53 return ok;
56 template<typename V>
57 bool ReadVector(V* items, SourceStream* buffer) {
58 uint32 count;
59 if (!buffer->ReadVarint32(&count))
60 return false;
62 items->clear();
64 bool ok = items->reserve(count);
65 for (size_t i = 0; ok && i < count; ++i) {
66 uint32 item;
67 ok = buffer->ReadVarint32(&item);
68 if (ok)
69 ok = items->push_back(static_cast<typename V::value_type>(item));
72 return ok;
75 // Serializes a vector, using delta coding followed by Varint32 coding.
76 template<typename V>
77 CheckBool WriteU32Delta(const V& set, SinkStream* buffer) {
78 size_t count = set.size();
79 bool ok = buffer->WriteSizeVarint32(count);
80 uint32 prev = 0;
81 for (size_t i = 0; ok && i < count; ++i) {
82 uint32 current = set[i];
83 uint32 delta = current - prev;
84 ok = buffer->WriteVarint32(delta);
85 prev = current;
87 return ok;
90 template <typename V>
91 static CheckBool ReadU32Delta(V* set, SourceStream* buffer) {
92 uint32 count;
94 if (!buffer->ReadVarint32(&count))
95 return false;
97 set->clear();
98 bool ok = set->reserve(count);
99 uint32 prev = 0;
101 for (size_t i = 0; ok && i < count; ++i) {
102 uint32 delta;
103 ok = buffer->ReadVarint32(&delta);
104 if (ok) {
105 uint32 current = prev + delta;
106 ok = set->push_back(current);
107 prev = current;
111 return ok;
114 // Write a vector as the byte representation of the contents.
116 // (This only really makes sense for a type T that has sizeof(T)==1, otherwise
117 // serialized representation is not endian-agnostic. But it is useful to keep
118 // the possibility of a greater size for experiments comparing Varint32 encoding
119 // of a vector of larger integrals vs a plain form.)
121 template<typename V>
122 CheckBool WriteVectorU8(const V& items, SinkStream* buffer) {
123 size_t count = items.size();
124 bool ok = buffer->WriteSizeVarint32(count);
125 if (count != 0 && ok) {
126 size_t byte_count = count * sizeof(typename V::value_type);
127 ok = buffer->Write(static_cast<const void*>(&items[0]), byte_count);
129 return ok;
132 template<typename V>
133 bool ReadVectorU8(V* items, SourceStream* buffer) {
134 uint32 count;
135 if (!buffer->ReadVarint32(&count))
136 return false;
138 items->clear();
139 bool ok = items->resize(count, 0);
140 if (ok && count != 0) {
141 size_t byte_count = count * sizeof(typename V::value_type);
142 return buffer->Read(static_cast<void*>(&((*items)[0])), byte_count);
144 return ok;
147 ////////////////////////////////////////////////////////////////////////////////
149 CheckBool EncodedProgram::DefineRel32Label(int index, RVA value) {
150 return DefineLabelCommon(&rel32_rva_, index, value);
153 CheckBool EncodedProgram::DefineAbs32Label(int index, RVA value) {
154 return DefineLabelCommon(&abs32_rva_, index, value);
157 static const RVA kUnassignedRVA = static_cast<RVA>(-1);
159 CheckBool EncodedProgram::DefineLabelCommon(RvaVector* rvas,
160 int index,
161 RVA rva) {
162 bool ok = true;
163 if (static_cast<int>(rvas->size()) <= index)
164 ok = rvas->resize(index + 1, kUnassignedRVA);
166 if (ok) {
167 DCHECK_EQ((*rvas)[index], kUnassignedRVA)
168 << "DefineLabel double assigned " << index;
169 (*rvas)[index] = rva;
172 return ok;
175 void EncodedProgram::EndLabels() {
176 FinishLabelsCommon(&abs32_rva_);
177 FinishLabelsCommon(&rel32_rva_);
180 void EncodedProgram::FinishLabelsCommon(RvaVector* rvas) {
181 // Replace all unassigned slots with the value at the previous index so they
182 // delta-encode to zero. (There might be better values than zero. The way to
183 // get that is have the higher level assembly program assign the unassigned
184 // slots.)
185 RVA previous = 0;
186 size_t size = rvas->size();
187 for (size_t i = 0; i < size; ++i) {
188 if ((*rvas)[i] == kUnassignedRVA)
189 (*rvas)[i] = previous;
190 else
191 previous = (*rvas)[i];
195 CheckBool EncodedProgram::AddOrigin(RVA origin) {
196 return ops_.push_back(ORIGIN) && origins_.push_back(origin);
199 CheckBool EncodedProgram::AddCopy(size_t count, const void* bytes) {
200 const uint8* source = static_cast<const uint8*>(bytes);
202 bool ok = true;
204 // Fold adjacent COPY instructions into one. This nearly halves the size of
205 // an EncodedProgram with only COPY1 instructions since there are approx plain
206 // 16 bytes per reloc. This has a working-set benefit during decompression.
207 // For compression of files with large differences this makes a small (4%)
208 // improvement in size. For files with small differences this degrades the
209 // compressed size by 1.3%
210 if (!ops_.empty()) {
211 if (ops_.back() == COPY1) {
212 ops_.back() = COPY;
213 ok = copy_counts_.push_back(1);
215 if (ok && ops_.back() == COPY) {
216 copy_counts_.back() += count;
217 for (size_t i = 0; ok && i < count; ++i) {
218 ok = copy_bytes_.push_back(source[i]);
220 return ok;
224 if (ok) {
225 if (count == 1) {
226 ok = ops_.push_back(COPY1) && copy_bytes_.push_back(source[0]);
227 } else {
228 ok = ops_.push_back(COPY) && copy_counts_.push_back(count);
229 for (size_t i = 0; ok && i < count; ++i) {
230 ok = copy_bytes_.push_back(source[i]);
235 return ok;
238 CheckBool EncodedProgram::AddAbs32(int label_index) {
239 return ops_.push_back(ABS32) && abs32_ix_.push_back(label_index);
242 CheckBool EncodedProgram::AddAbs64(int label_index) {
243 return ops_.push_back(ABS64) && abs32_ix_.push_back(label_index);
246 CheckBool EncodedProgram::AddRel32(int label_index) {
247 return ops_.push_back(REL32) && rel32_ix_.push_back(label_index);
250 CheckBool EncodedProgram::AddRel32ARM(uint16 op, int label_index) {
251 return ops_.push_back(static_cast<OP>(op)) &&
252 rel32_ix_.push_back(label_index);
255 CheckBool EncodedProgram::AddPeMakeRelocs(ExecutableType kind) {
256 if (kind == EXE_WIN_32_X86)
257 return ops_.push_back(MAKE_PE_RELOCATION_TABLE);
258 return ops_.push_back(MAKE_PE64_RELOCATION_TABLE);
261 CheckBool EncodedProgram::AddElfMakeRelocs() {
262 return ops_.push_back(MAKE_ELF_RELOCATION_TABLE);
265 CheckBool EncodedProgram::AddElfARMMakeRelocs() {
266 return ops_.push_back(MAKE_ELF_ARM_RELOCATION_TABLE);
269 void EncodedProgram::DebuggingSummary() {
270 VLOG(1) << "EncodedProgram Summary"
271 << "\n image base " << image_base_
272 << "\n abs32 rvas " << abs32_rva_.size()
273 << "\n rel32 rvas " << rel32_rva_.size()
274 << "\n ops " << ops_.size()
275 << "\n origins " << origins_.size()
276 << "\n copy_counts " << copy_counts_.size()
277 << "\n copy_bytes " << copy_bytes_.size()
278 << "\n abs32_ix " << abs32_ix_.size()
279 << "\n rel32_ix " << rel32_ix_.size();
282 ////////////////////////////////////////////////////////////////////////////////
284 // For algorithm refinement purposes it is useful to write subsets of the file
285 // format. This gives us the ability to estimate the entropy of the
286 // differential compression of the individual streams, which can provide
287 // invaluable insights. The default, of course, is to include all the streams.
289 enum FieldSelect {
290 INCLUDE_ABS32_ADDRESSES = 0x0001,
291 INCLUDE_REL32_ADDRESSES = 0x0002,
292 INCLUDE_ABS32_INDEXES = 0x0010,
293 INCLUDE_REL32_INDEXES = 0x0020,
294 INCLUDE_OPS = 0x0100,
295 INCLUDE_BYTES = 0x0200,
296 INCLUDE_COPY_COUNTS = 0x0400,
297 INCLUDE_MISC = 0x1000
300 static FieldSelect GetFieldSelect() {
301 // TODO(sra): Use better configuration.
302 scoped_ptr<base::Environment> env(base::Environment::Create());
303 std::string s;
304 env->GetVar("A_FIELDS", &s);
305 uint64 fields;
306 if (!base::StringToUint64(s, &fields))
307 return static_cast<FieldSelect>(~0);
308 return static_cast<FieldSelect>(fields);
311 CheckBool EncodedProgram::WriteTo(SinkStreamSet* streams) {
312 FieldSelect select = GetFieldSelect();
314 // The order of fields must be consistent in WriteTo and ReadFrom, regardless
315 // of the streams used. The code can be configured with all kStreamXXX
316 // constants the same.
318 // If we change the code to pipeline reading with assembly (to avoid temporary
319 // storage vectors by consuming operands directly from the stream) then we
320 // need to read the base address and the random access address tables first,
321 // the rest can be interleaved.
323 if (select & INCLUDE_MISC) {
324 uint32 high = static_cast<uint32>(image_base_ >> 32);
325 uint32 low = static_cast<uint32>(image_base_ & 0xffffffffU);
327 if (!streams->stream(kStreamMisc)->WriteVarint32(high) ||
328 !streams->stream(kStreamMisc)->WriteVarint32(low)) {
329 return false;
333 bool success = true;
335 if (select & INCLUDE_ABS32_ADDRESSES) {
336 success &= WriteU32Delta(abs32_rva_,
337 streams->stream(kStreamAbs32Addresses));
340 if (select & INCLUDE_REL32_ADDRESSES) {
341 success &= WriteU32Delta(rel32_rva_,
342 streams->stream(kStreamRel32Addresses));
345 if (select & INCLUDE_MISC)
346 success &= WriteVector(origins_, streams->stream(kStreamOriginAddresses));
348 if (select & INCLUDE_OPS) {
349 // 5 for length.
350 success &= streams->stream(kStreamOps)->Reserve(ops_.size() + 5);
351 success &= WriteVector(ops_, streams->stream(kStreamOps));
354 if (select & INCLUDE_COPY_COUNTS)
355 success &= WriteVector(copy_counts_, streams->stream(kStreamCopyCounts));
357 if (select & INCLUDE_BYTES)
358 success &= WriteVectorU8(copy_bytes_, streams->stream(kStreamBytes));
360 if (select & INCLUDE_ABS32_INDEXES)
361 success &= WriteVector(abs32_ix_, streams->stream(kStreamAbs32Indexes));
363 if (select & INCLUDE_REL32_INDEXES)
364 success &= WriteVector(rel32_ix_, streams->stream(kStreamRel32Indexes));
366 return success;
369 bool EncodedProgram::ReadFrom(SourceStreamSet* streams) {
370 uint32 high;
371 uint32 low;
373 if (!streams->stream(kStreamMisc)->ReadVarint32(&high) ||
374 !streams->stream(kStreamMisc)->ReadVarint32(&low)) {
375 return false;
377 image_base_ = (static_cast<uint64>(high) << 32) | low;
379 if (!ReadU32Delta(&abs32_rva_, streams->stream(kStreamAbs32Addresses)))
380 return false;
381 if (!ReadU32Delta(&rel32_rva_, streams->stream(kStreamRel32Addresses)))
382 return false;
383 if (!ReadVector(&origins_, streams->stream(kStreamOriginAddresses)))
384 return false;
385 if (!ReadVector(&ops_, streams->stream(kStreamOps)))
386 return false;
387 if (!ReadVector(&copy_counts_, streams->stream(kStreamCopyCounts)))
388 return false;
389 if (!ReadVectorU8(&copy_bytes_, streams->stream(kStreamBytes)))
390 return false;
391 if (!ReadVector(&abs32_ix_, streams->stream(kStreamAbs32Indexes)))
392 return false;
393 if (!ReadVector(&rel32_ix_, streams->stream(kStreamRel32Indexes)))
394 return false;
396 // Check that streams have been completely consumed.
397 for (int i = 0; i < kStreamLimit; ++i) {
398 if (streams->stream(i)->Remaining() > 0)
399 return false;
402 return true;
405 // Safe, non-throwing version of std::vector::at(). Returns 'true' for success,
406 // 'false' for out-of-bounds index error.
407 template<typename V, typename T>
408 bool VectorAt(const V& v, size_t index, T* output) {
409 if (index >= v.size())
410 return false;
411 *output = v[index];
412 return true;
415 CheckBool EncodedProgram::EvaluateRel32ARM(OP op,
416 size_t& ix_rel32_ix,
417 RVA& current_rva,
418 SinkStream* output) {
419 switch (op & 0x0000F000) {
420 case REL32ARM8: {
421 uint32 index;
422 if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
423 return false;
424 ++ix_rel32_ix;
425 RVA rva;
426 if (!VectorAt(rel32_rva_, index, &rva))
427 return false;
428 uint32 decompressed_op;
429 if (!DisassemblerElf32ARM::Decompress(ARM_OFF8,
430 static_cast<uint16>(op),
431 static_cast<uint32>(rva -
432 current_rva),
433 &decompressed_op)) {
434 return false;
436 uint16 op16 = static_cast<uint16>(decompressed_op);
437 if (!output->Write(&op16, 2))
438 return false;
439 current_rva += 2;
440 break;
442 case REL32ARM11: {
443 uint32 index;
444 if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
445 return false;
446 ++ix_rel32_ix;
447 RVA rva;
448 if (!VectorAt(rel32_rva_, index, &rva))
449 return false;
450 uint32 decompressed_op;
451 if (!DisassemblerElf32ARM::Decompress(ARM_OFF11, (uint16) op,
452 (uint32) (rva - current_rva),
453 &decompressed_op)) {
454 return false;
456 uint16 op16 = static_cast<uint16>(decompressed_op);
457 if (!output->Write(&op16, 2))
458 return false;
459 current_rva += 2;
460 break;
462 case REL32ARM24: {
463 uint32 index;
464 if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
465 return false;
466 ++ix_rel32_ix;
467 RVA rva;
468 if (!VectorAt(rel32_rva_, index, &rva))
469 return false;
470 uint32 decompressed_op;
471 if (!DisassemblerElf32ARM::Decompress(ARM_OFF24, (uint16) op,
472 (uint32) (rva - current_rva),
473 &decompressed_op)) {
474 return false;
476 if (!output->Write(&decompressed_op, 4))
477 return false;
478 current_rva += 4;
479 break;
481 case REL32ARM25: {
482 uint32 index;
483 if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
484 return false;
485 ++ix_rel32_ix;
486 RVA rva;
487 if (!VectorAt(rel32_rva_, index, &rva))
488 return false;
489 uint32 decompressed_op;
490 if (!DisassemblerElf32ARM::Decompress(ARM_OFF25, (uint16) op,
491 (uint32) (rva - current_rva),
492 &decompressed_op)) {
493 return false;
495 uint32 words = (decompressed_op << 16) | (decompressed_op >> 16);
496 if (!output->Write(&words, 4))
497 return false;
498 current_rva += 4;
499 break;
501 case REL32ARM21: {
502 uint32 index;
503 if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
504 return false;
505 ++ix_rel32_ix;
506 RVA rva;
507 if (!VectorAt(rel32_rva_, index, &rva))
508 return false;
509 uint32 decompressed_op;
510 if (!DisassemblerElf32ARM::Decompress(ARM_OFF21, (uint16) op,
511 (uint32) (rva - current_rva),
512 &decompressed_op)) {
513 return false;
515 uint32 words = (decompressed_op << 16) | (decompressed_op >> 16);
516 if (!output->Write(&words, 4))
517 return false;
518 current_rva += 4;
519 break;
521 default:
522 return false;
525 return true;
528 CheckBool EncodedProgram::AssembleTo(SinkStream* final_buffer) {
529 // For the most part, the assembly process walks the various tables.
530 // ix_mumble is the index into the mumble table.
531 size_t ix_origins = 0;
532 size_t ix_copy_counts = 0;
533 size_t ix_copy_bytes = 0;
534 size_t ix_abs32_ix = 0;
535 size_t ix_rel32_ix = 0;
537 RVA current_rva = 0;
539 bool pending_pe_relocation_table = false;
540 uint8 pending_pe_relocation_table_type = 0x03; // IMAGE_REL_BASED_HIGHLOW
541 Elf32_Word pending_elf_relocation_table_type = 0;
542 SinkStream bytes_following_relocation_table;
544 SinkStream* output = final_buffer;
546 for (size_t ix_ops = 0; ix_ops < ops_.size(); ++ix_ops) {
547 OP op = ops_[ix_ops];
549 switch (op) {
550 default:
551 if (!EvaluateRel32ARM(op, ix_rel32_ix, current_rva, output))
552 return false;
553 break;
555 case ORIGIN: {
556 RVA section_rva;
557 if (!VectorAt(origins_, ix_origins, &section_rva))
558 return false;
559 ++ix_origins;
560 current_rva = section_rva;
561 break;
564 case COPY: {
565 size_t count;
566 if (!VectorAt(copy_counts_, ix_copy_counts, &count))
567 return false;
568 ++ix_copy_counts;
569 for (size_t i = 0; i < count; ++i) {
570 uint8 b;
571 if (!VectorAt(copy_bytes_, ix_copy_bytes, &b))
572 return false;
573 ++ix_copy_bytes;
574 if (!output->Write(&b, 1))
575 return false;
577 current_rva += static_cast<RVA>(count);
578 break;
581 case COPY1: {
582 uint8 b;
583 if (!VectorAt(copy_bytes_, ix_copy_bytes, &b))
584 return false;
585 ++ix_copy_bytes;
586 if (!output->Write(&b, 1))
587 return false;
588 current_rva += 1;
589 break;
592 case REL32: {
593 uint32 index;
594 if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
595 return false;
596 ++ix_rel32_ix;
597 RVA rva;
598 if (!VectorAt(rel32_rva_, index, &rva))
599 return false;
600 uint32 offset = (rva - (current_rva + 4));
601 if (!output->Write(&offset, 4))
602 return false;
603 current_rva += 4;
604 break;
607 case ABS32:
608 case ABS64: {
609 uint32 index;
610 if (!VectorAt(abs32_ix_, ix_abs32_ix, &index))
611 return false;
612 ++ix_abs32_ix;
613 RVA rva;
614 if (!VectorAt(abs32_rva_, index, &rva))
615 return false;
616 if (op == ABS32) {
617 base::CheckedNumeric<uint32> abs32 = image_base_;
618 abs32 += rva;
619 uint32 safe_abs32 = abs32.ValueOrDie();
620 if (!abs32_relocs_.push_back(current_rva) ||
621 !output->Write(&safe_abs32, 4)) {
622 return false;
624 current_rva += 4;
625 } else {
626 base::CheckedNumeric<uint64> abs64 = image_base_;
627 abs64 += rva;
628 uint64 safe_abs64 = abs64.ValueOrDie();
629 if (!abs32_relocs_.push_back(current_rva) ||
630 !output->Write(&safe_abs64, 8)) {
631 return false;
633 current_rva += 8;
635 break;
638 case MAKE_PE_RELOCATION_TABLE: {
639 // We can see the base relocation anywhere, but we only have the
640 // information to generate it at the very end. So we divert the bytes
641 // we are generating to a temporary stream.
642 if (pending_pe_relocation_table)
643 return false; // Can't have two base relocation tables.
645 pending_pe_relocation_table = true;
646 output = &bytes_following_relocation_table;
647 break;
648 // There is a potential problem *if* the instruction stream contains
649 // some REL32 relocations following the base relocation and in the same
650 // section. We don't know the size of the table, so 'current_rva' will
651 // be wrong, causing REL32 offsets to be miscalculated. This never
652 // happens; the base relocation table is usually in a section of its
653 // own, a data-only section, and following everything else in the
654 // executable except some padding zero bytes. We could fix this by
655 // emitting an ORIGIN after the MAKE_BASE_RELOCATION_TABLE.
658 case MAKE_PE64_RELOCATION_TABLE: {
659 if (pending_pe_relocation_table)
660 return false; // Can't have two base relocation tables.
662 pending_pe_relocation_table = true;
663 pending_pe_relocation_table_type = 0x0A; // IMAGE_REL_BASED_DIR64
664 output = &bytes_following_relocation_table;
665 break;
668 case MAKE_ELF_ARM_RELOCATION_TABLE: {
669 // We can see the base relocation anywhere, but we only have the
670 // information to generate it at the very end. So we divert the bytes
671 // we are generating to a temporary stream.
672 if (pending_elf_relocation_table_type)
673 return false; // Can't have two base relocation tables.
675 pending_elf_relocation_table_type = R_ARM_RELATIVE;
676 output = &bytes_following_relocation_table;
677 break;
680 case MAKE_ELF_RELOCATION_TABLE: {
681 // We can see the base relocation anywhere, but we only have the
682 // information to generate it at the very end. So we divert the bytes
683 // we are generating to a temporary stream.
684 if (pending_elf_relocation_table_type)
685 return false; // Can't have two base relocation tables.
687 pending_elf_relocation_table_type = R_386_RELATIVE;
688 output = &bytes_following_relocation_table;
689 break;
694 if (pending_pe_relocation_table) {
695 if (!GeneratePeRelocations(final_buffer,
696 pending_pe_relocation_table_type) ||
697 !final_buffer->Append(&bytes_following_relocation_table))
698 return false;
701 if (pending_elf_relocation_table_type) {
702 if (!GenerateElfRelocations(pending_elf_relocation_table_type,
703 final_buffer) ||
704 !final_buffer->Append(&bytes_following_relocation_table))
705 return false;
708 // Final verification check: did we consume all lists?
709 if (ix_copy_counts != copy_counts_.size())
710 return false;
711 if (ix_copy_bytes != copy_bytes_.size())
712 return false;
713 if (ix_abs32_ix != abs32_ix_.size())
714 return false;
715 if (ix_rel32_ix != rel32_ix_.size())
716 return false;
718 return true;
721 // RelocBlock has the layout of a block of relocations in the base relocation
722 // table file format.
724 struct RelocBlockPOD {
725 uint32 page_rva;
726 uint32 block_size;
727 uint16 relocs[4096]; // Allow up to one relocation per byte of a 4k page.
730 static_assert(offsetof(RelocBlockPOD, relocs) == 8, "reloc block header size");
732 class RelocBlock {
733 public:
734 RelocBlock() {
735 pod.page_rva = 0xFFFFFFFF;
736 pod.block_size = 8;
739 void Add(uint16 item) {
740 pod.relocs[(pod.block_size-8)/2] = item;
741 pod.block_size += 2;
744 CheckBool Flush(SinkStream* buffer) WARN_UNUSED_RESULT {
745 bool ok = true;
746 if (pod.block_size != 8) {
747 if (pod.block_size % 4 != 0) { // Pad to make size multiple of 4 bytes.
748 Add(0);
750 ok = buffer->Write(&pod, pod.block_size);
751 pod.block_size = 8;
753 return ok;
755 RelocBlockPOD pod;
758 CheckBool EncodedProgram::GeneratePeRelocations(SinkStream* buffer,
759 uint8 type) {
760 std::sort(abs32_relocs_.begin(), abs32_relocs_.end());
762 RelocBlock block;
764 bool ok = true;
765 for (size_t i = 0; ok && i < abs32_relocs_.size(); ++i) {
766 uint32 rva = abs32_relocs_[i];
767 uint32 page_rva = rva & ~0xFFF;
768 if (page_rva != block.pod.page_rva) {
769 ok &= block.Flush(buffer);
770 block.pod.page_rva = page_rva;
772 if (ok)
773 block.Add(((static_cast<uint16>(type)) << 12) | (rva & 0xFFF));
775 ok &= block.Flush(buffer);
776 return ok;
779 CheckBool EncodedProgram::GenerateElfRelocations(Elf32_Word r_info,
780 SinkStream* buffer) {
781 std::sort(abs32_relocs_.begin(), abs32_relocs_.end());
783 Elf32_Rel relocation_block;
785 relocation_block.r_info = r_info;
787 bool ok = true;
788 for (size_t i = 0; ok && i < abs32_relocs_.size(); ++i) {
789 relocation_block.r_offset = abs32_relocs_[i];
790 ok = buffer->Write(&relocation_block, sizeof(Elf32_Rel));
793 return ok;
795 ////////////////////////////////////////////////////////////////////////////////
797 Status WriteEncodedProgram(EncodedProgram* encoded, SinkStreamSet* sink) {
798 if (!encoded->WriteTo(sink))
799 return C_STREAM_ERROR;
800 return C_OK;
803 Status ReadEncodedProgram(SourceStreamSet* streams, EncodedProgram** output) {
804 EncodedProgram* encoded = new EncodedProgram();
805 if (encoded->ReadFrom(streams)) {
806 *output = encoded;
807 return C_OK;
809 delete encoded;
810 return C_DESERIALIZATION_FAILED;
813 Status Assemble(EncodedProgram* encoded, SinkStream* buffer) {
814 bool assembled = encoded->AssembleTo(buffer);
815 if (assembled)
816 return C_OK;
817 return C_ASSEMBLY_FAILED;
820 void DeleteEncodedProgram(EncodedProgram* encoded) {
821 delete encoded;
824 } // namespace courgette