1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/toolkit/crashreporter/google-breakpad/src/common/dwarf/dwarf2reader.cc Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,2351 @@
1.4 +// Copyright (c) 2010 Google Inc. All Rights Reserved.
1.5 +//
1.6 +// Redistribution and use in source and binary forms, with or without
1.7 +// modification, are permitted provided that the following conditions are
1.8 +// met:
1.9 +//
1.10 +// * Redistributions of source code must retain the above copyright
1.11 +// notice, this list of conditions and the following disclaimer.
1.12 +// * Redistributions in binary form must reproduce the above
1.13 +// copyright notice, this list of conditions and the following disclaimer
1.14 +// in the documentation and/or other materials provided with the
1.15 +// distribution.
1.16 +// * Neither the name of Google Inc. nor the names of its
1.17 +// contributors may be used to endorse or promote products derived from
1.18 +// this software without specific prior written permission.
1.19 +//
1.20 +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
1.21 +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
1.22 +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
1.23 +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
1.24 +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
1.25 +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
1.26 +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
1.27 +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
1.28 +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
1.29 +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
1.30 +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
1.31 +
1.32 +// CFI reader author: Jim Blandy <jimb@mozilla.com> <jimb@red-bean.com>
1.33 +
1.34 +// Implementation of dwarf2reader::LineInfo, dwarf2reader::CompilationUnit,
1.35 +// and dwarf2reader::CallFrameInfo. See dwarf2reader.h for details.
1.36 +
1.37 +#include "common/dwarf/dwarf2reader.h"
1.38 +
1.39 +#include <assert.h>
1.40 +#include <stdint.h>
1.41 +#include <stdio.h>
1.42 +#include <string.h>
1.43 +
1.44 +#include <map>
1.45 +#include <memory>
1.46 +#include <stack>
1.47 +#include <string>
1.48 +#include <utility>
1.49 +
1.50 +#include "common/dwarf/bytereader-inl.h"
1.51 +#include "common/dwarf/bytereader.h"
1.52 +#include "common/dwarf/line_state_machine.h"
1.53 +#include "common/using_std_string.h"
1.54 +
1.55 +namespace dwarf2reader {
1.56 +
1.57 +CompilationUnit::CompilationUnit(const SectionMap& sections, uint64 offset,
1.58 + ByteReader* reader, Dwarf2Handler* handler)
1.59 + : offset_from_section_start_(offset), reader_(reader),
1.60 + sections_(sections), handler_(handler), abbrevs_(NULL),
1.61 + string_buffer_(NULL), string_buffer_length_(0) {}
1.62 +
1.63 +// Read a DWARF2/3 abbreviation section.
1.64 +// Each abbrev consists of a abbreviation number, a tag, a byte
1.65 +// specifying whether the tag has children, and a list of
1.66 +// attribute/form pairs.
1.67 +// The list of forms is terminated by a 0 for the attribute, and a
1.68 +// zero for the form. The entire abbreviation section is terminated
1.69 +// by a zero for the code.
1.70 +
1.71 +void CompilationUnit::ReadAbbrevs() {
1.72 + if (abbrevs_)
1.73 + return;
1.74 +
1.75 + // First get the debug_abbrev section. ".debug_abbrev" is the name
1.76 + // recommended in the DWARF spec, and used on Linux;
1.77 + // "__debug_abbrev" is the name used in Mac OS X Mach-O files.
1.78 + SectionMap::const_iterator iter = sections_.find(".debug_abbrev");
1.79 + if (iter == sections_.end())
1.80 + iter = sections_.find("__debug_abbrev");
1.81 + assert(iter != sections_.end());
1.82 +
1.83 + abbrevs_ = new std::vector<Abbrev>;
1.84 + abbrevs_->resize(1);
1.85 +
1.86 + // The only way to check whether we are reading over the end of the
1.87 + // buffer would be to first compute the size of the leb128 data by
1.88 + // reading it, then go back and read it again.
1.89 + const char* abbrev_start = iter->second.first +
1.90 + header_.abbrev_offset;
1.91 + const char* abbrevptr = abbrev_start;
1.92 +#ifndef NDEBUG
1.93 + const uint64 abbrev_length = iter->second.second - header_.abbrev_offset;
1.94 +#endif
1.95 +
1.96 + while (1) {
1.97 + CompilationUnit::Abbrev abbrev;
1.98 + size_t len;
1.99 + const uint64 number = reader_->ReadUnsignedLEB128(abbrevptr, &len);
1.100 +
1.101 + if (number == 0)
1.102 + break;
1.103 + abbrev.number = number;
1.104 + abbrevptr += len;
1.105 +
1.106 + assert(abbrevptr < abbrev_start + abbrev_length);
1.107 + const uint64 tag = reader_->ReadUnsignedLEB128(abbrevptr, &len);
1.108 + abbrevptr += len;
1.109 + abbrev.tag = static_cast<enum DwarfTag>(tag);
1.110 +
1.111 + assert(abbrevptr < abbrev_start + abbrev_length);
1.112 + abbrev.has_children = reader_->ReadOneByte(abbrevptr);
1.113 + abbrevptr += 1;
1.114 +
1.115 + assert(abbrevptr < abbrev_start + abbrev_length);
1.116 +
1.117 + while (1) {
1.118 + const uint64 nametemp = reader_->ReadUnsignedLEB128(abbrevptr, &len);
1.119 + abbrevptr += len;
1.120 +
1.121 + assert(abbrevptr < abbrev_start + abbrev_length);
1.122 + const uint64 formtemp = reader_->ReadUnsignedLEB128(abbrevptr, &len);
1.123 + abbrevptr += len;
1.124 + if (nametemp == 0 && formtemp == 0)
1.125 + break;
1.126 +
1.127 + const enum DwarfAttribute name =
1.128 + static_cast<enum DwarfAttribute>(nametemp);
1.129 + const enum DwarfForm form = static_cast<enum DwarfForm>(formtemp);
1.130 + abbrev.attributes.push_back(std::make_pair(name, form));
1.131 + }
1.132 + assert(abbrev.number == abbrevs_->size());
1.133 + abbrevs_->push_back(abbrev);
1.134 + }
1.135 +}
1.136 +
1.137 +// Skips a single DIE's attributes.
1.138 +const char* CompilationUnit::SkipDIE(const char* start,
1.139 + const Abbrev& abbrev) {
1.140 + for (AttributeList::const_iterator i = abbrev.attributes.begin();
1.141 + i != abbrev.attributes.end();
1.142 + i++) {
1.143 + start = SkipAttribute(start, i->second);
1.144 + }
1.145 + return start;
1.146 +}
1.147 +
1.148 +// Skips a single attribute form's data.
1.149 +const char* CompilationUnit::SkipAttribute(const char* start,
1.150 + enum DwarfForm form) {
1.151 + size_t len;
1.152 +
1.153 + switch (form) {
1.154 + case DW_FORM_indirect:
1.155 + form = static_cast<enum DwarfForm>(reader_->ReadUnsignedLEB128(start,
1.156 + &len));
1.157 + start += len;
1.158 + return SkipAttribute(start, form);
1.159 +
1.160 + case DW_FORM_flag_present:
1.161 + return start;
1.162 + case DW_FORM_data1:
1.163 + case DW_FORM_flag:
1.164 + case DW_FORM_ref1:
1.165 + return start + 1;
1.166 + case DW_FORM_ref2:
1.167 + case DW_FORM_data2:
1.168 + return start + 2;
1.169 + case DW_FORM_ref4:
1.170 + case DW_FORM_data4:
1.171 + return start + 4;
1.172 + case DW_FORM_ref8:
1.173 + case DW_FORM_data8:
1.174 + case DW_FORM_ref_sig8:
1.175 + return start + 8;
1.176 + case DW_FORM_string:
1.177 + return start + strlen(start) + 1;
1.178 + case DW_FORM_udata:
1.179 + case DW_FORM_ref_udata:
1.180 + reader_->ReadUnsignedLEB128(start, &len);
1.181 + return start + len;
1.182 +
1.183 + case DW_FORM_sdata:
1.184 + reader_->ReadSignedLEB128(start, &len);
1.185 + return start + len;
1.186 + case DW_FORM_addr:
1.187 + return start + reader_->AddressSize();
1.188 + case DW_FORM_ref_addr:
1.189 + // DWARF2 and 3 differ on whether ref_addr is address size or
1.190 + // offset size.
1.191 + assert(header_.version == 2 || header_.version == 3);
1.192 + if (header_.version == 2) {
1.193 + return start + reader_->AddressSize();
1.194 + } else if (header_.version == 3) {
1.195 + return start + reader_->OffsetSize();
1.196 + }
1.197 +
1.198 + case DW_FORM_block1:
1.199 + return start + 1 + reader_->ReadOneByte(start);
1.200 + case DW_FORM_block2:
1.201 + return start + 2 + reader_->ReadTwoBytes(start);
1.202 + case DW_FORM_block4:
1.203 + return start + 4 + reader_->ReadFourBytes(start);
1.204 + case DW_FORM_block:
1.205 + case DW_FORM_exprloc: {
1.206 + uint64 size = reader_->ReadUnsignedLEB128(start, &len);
1.207 + return start + size + len;
1.208 + }
1.209 + case DW_FORM_strp:
1.210 + case DW_FORM_sec_offset:
1.211 + return start + reader_->OffsetSize();
1.212 + }
1.213 + fprintf(stderr,"Unhandled form type");
1.214 + return NULL;
1.215 +}
1.216 +
1.217 +// Read a DWARF2/3 header.
1.218 +// The header is variable length in DWARF3 (and DWARF2 as extended by
1.219 +// most compilers), and consists of an length field, a version number,
1.220 +// the offset in the .debug_abbrev section for our abbrevs, and an
1.221 +// address size.
1.222 +void CompilationUnit::ReadHeader() {
1.223 + const char* headerptr = buffer_;
1.224 + size_t initial_length_size;
1.225 +
1.226 + assert(headerptr + 4 < buffer_ + buffer_length_);
1.227 + const uint64 initial_length
1.228 + = reader_->ReadInitialLength(headerptr, &initial_length_size);
1.229 + headerptr += initial_length_size;
1.230 + header_.length = initial_length;
1.231 +
1.232 + assert(headerptr + 2 < buffer_ + buffer_length_);
1.233 + header_.version = reader_->ReadTwoBytes(headerptr);
1.234 + headerptr += 2;
1.235 +
1.236 + assert(headerptr + reader_->OffsetSize() < buffer_ + buffer_length_);
1.237 + header_.abbrev_offset = reader_->ReadOffset(headerptr);
1.238 + headerptr += reader_->OffsetSize();
1.239 +
1.240 + assert(headerptr + 1 < buffer_ + buffer_length_);
1.241 + header_.address_size = reader_->ReadOneByte(headerptr);
1.242 + reader_->SetAddressSize(header_.address_size);
1.243 + headerptr += 1;
1.244 +
1.245 + after_header_ = headerptr;
1.246 +
1.247 + // This check ensures that we don't have to do checking during the
1.248 + // reading of DIEs. header_.length does not include the size of the
1.249 + // initial length.
1.250 + assert(buffer_ + initial_length_size + header_.length <=
1.251 + buffer_ + buffer_length_);
1.252 +}
1.253 +
1.254 +uint64 CompilationUnit::Start() {
1.255 + // First get the debug_info section. ".debug_info" is the name
1.256 + // recommended in the DWARF spec, and used on Linux; "__debug_info"
1.257 + // is the name used in Mac OS X Mach-O files.
1.258 + SectionMap::const_iterator iter = sections_.find(".debug_info");
1.259 + if (iter == sections_.end())
1.260 + iter = sections_.find("__debug_info");
1.261 + assert(iter != sections_.end());
1.262 +
1.263 + // Set up our buffer
1.264 + buffer_ = iter->second.first + offset_from_section_start_;
1.265 + buffer_length_ = iter->second.second - offset_from_section_start_;
1.266 +
1.267 + // Read the header
1.268 + ReadHeader();
1.269 +
1.270 + // Figure out the real length from the end of the initial length to
1.271 + // the end of the compilation unit, since that is the value we
1.272 + // return.
1.273 + uint64 ourlength = header_.length;
1.274 + if (reader_->OffsetSize() == 8)
1.275 + ourlength += 12;
1.276 + else
1.277 + ourlength += 4;
1.278 +
1.279 + // See if the user wants this compilation unit, and if not, just return.
1.280 + if (!handler_->StartCompilationUnit(offset_from_section_start_,
1.281 + reader_->AddressSize(),
1.282 + reader_->OffsetSize(),
1.283 + header_.length,
1.284 + header_.version))
1.285 + return ourlength;
1.286 +
1.287 + // Otherwise, continue by reading our abbreviation entries.
1.288 + ReadAbbrevs();
1.289 +
1.290 + // Set the string section if we have one. ".debug_str" is the name
1.291 + // recommended in the DWARF spec, and used on Linux; "__debug_str"
1.292 + // is the name used in Mac OS X Mach-O files.
1.293 + iter = sections_.find(".debug_str");
1.294 + if (iter == sections_.end())
1.295 + iter = sections_.find("__debug_str");
1.296 + if (iter != sections_.end()) {
1.297 + string_buffer_ = iter->second.first;
1.298 + string_buffer_length_ = iter->second.second;
1.299 + }
1.300 +
1.301 + // Now that we have our abbreviations, start processing DIE's.
1.302 + ProcessDIEs();
1.303 +
1.304 + return ourlength;
1.305 +}
1.306 +
1.307 +// If one really wanted, you could merge SkipAttribute and
1.308 +// ProcessAttribute
1.309 +// This is all boring data manipulation and calling of the handler.
1.310 +const char* CompilationUnit::ProcessAttribute(
1.311 + uint64 dieoffset, const char* start, enum DwarfAttribute attr,
1.312 + enum DwarfForm form) {
1.313 + size_t len;
1.314 +
1.315 + switch (form) {
1.316 + // DW_FORM_indirect is never used because it is such a space
1.317 + // waster.
1.318 + case DW_FORM_indirect:
1.319 + form = static_cast<enum DwarfForm>(reader_->ReadUnsignedLEB128(start,
1.320 + &len));
1.321 + start += len;
1.322 + return ProcessAttribute(dieoffset, start, attr, form);
1.323 +
1.324 + case DW_FORM_flag_present:
1.325 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form, 1);
1.326 + return start;
1.327 + case DW_FORM_data1:
1.328 + case DW_FORM_flag:
1.329 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form,
1.330 + reader_->ReadOneByte(start));
1.331 + return start + 1;
1.332 + case DW_FORM_data2:
1.333 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form,
1.334 + reader_->ReadTwoBytes(start));
1.335 + return start + 2;
1.336 + case DW_FORM_data4:
1.337 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form,
1.338 + reader_->ReadFourBytes(start));
1.339 + return start + 4;
1.340 + case DW_FORM_data8:
1.341 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form,
1.342 + reader_->ReadEightBytes(start));
1.343 + return start + 8;
1.344 + case DW_FORM_string: {
1.345 + const char* str = start;
1.346 + handler_->ProcessAttributeString(dieoffset, attr, form,
1.347 + str);
1.348 + return start + strlen(str) + 1;
1.349 + }
1.350 + case DW_FORM_udata:
1.351 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form,
1.352 + reader_->ReadUnsignedLEB128(start,
1.353 + &len));
1.354 + return start + len;
1.355 +
1.356 + case DW_FORM_sdata:
1.357 + handler_->ProcessAttributeSigned(dieoffset, attr, form,
1.358 + reader_->ReadSignedLEB128(start, &len));
1.359 + return start + len;
1.360 + case DW_FORM_addr:
1.361 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form,
1.362 + reader_->ReadAddress(start));
1.363 + return start + reader_->AddressSize();
1.364 + case DW_FORM_sec_offset:
1.365 + handler_->ProcessAttributeUnsigned(dieoffset, attr, form,
1.366 + reader_->ReadOffset(start));
1.367 + return start + reader_->OffsetSize();
1.368 +
1.369 + case DW_FORM_ref1:
1.370 + handler_->ProcessAttributeReference(dieoffset, attr, form,
1.371 + reader_->ReadOneByte(start)
1.372 + + offset_from_section_start_);
1.373 + return start + 1;
1.374 + case DW_FORM_ref2:
1.375 + handler_->ProcessAttributeReference(dieoffset, attr, form,
1.376 + reader_->ReadTwoBytes(start)
1.377 + + offset_from_section_start_);
1.378 + return start + 2;
1.379 + case DW_FORM_ref4:
1.380 + handler_->ProcessAttributeReference(dieoffset, attr, form,
1.381 + reader_->ReadFourBytes(start)
1.382 + + offset_from_section_start_);
1.383 + return start + 4;
1.384 + case DW_FORM_ref8:
1.385 + handler_->ProcessAttributeReference(dieoffset, attr, form,
1.386 + reader_->ReadEightBytes(start)
1.387 + + offset_from_section_start_);
1.388 + return start + 8;
1.389 + case DW_FORM_ref_udata:
1.390 + handler_->ProcessAttributeReference(dieoffset, attr, form,
1.391 + reader_->ReadUnsignedLEB128(start,
1.392 + &len)
1.393 + + offset_from_section_start_);
1.394 + return start + len;
1.395 + case DW_FORM_ref_addr:
1.396 + // DWARF2 and 3 differ on whether ref_addr is address size or
1.397 + // offset size.
1.398 + assert(header_.version == 2 || header_.version == 3);
1.399 + if (header_.version == 2) {
1.400 + handler_->ProcessAttributeReference(dieoffset, attr, form,
1.401 + reader_->ReadAddress(start));
1.402 + return start + reader_->AddressSize();
1.403 + } else if (header_.version == 3) {
1.404 + handler_->ProcessAttributeReference(dieoffset, attr, form,
1.405 + reader_->ReadOffset(start));
1.406 + return start + reader_->OffsetSize();
1.407 + }
1.408 + break;
1.409 + case DW_FORM_ref_sig8:
1.410 + handler_->ProcessAttributeSignature(dieoffset, attr, form,
1.411 + reader_->ReadEightBytes(start));
1.412 + return start + 8;
1.413 +
1.414 + case DW_FORM_block1: {
1.415 + uint64 datalen = reader_->ReadOneByte(start);
1.416 + handler_->ProcessAttributeBuffer(dieoffset, attr, form, start + 1,
1.417 + datalen);
1.418 + return start + 1 + datalen;
1.419 + }
1.420 + case DW_FORM_block2: {
1.421 + uint64 datalen = reader_->ReadTwoBytes(start);
1.422 + handler_->ProcessAttributeBuffer(dieoffset, attr, form, start + 2,
1.423 + datalen);
1.424 + return start + 2 + datalen;
1.425 + }
1.426 + case DW_FORM_block4: {
1.427 + uint64 datalen = reader_->ReadFourBytes(start);
1.428 + handler_->ProcessAttributeBuffer(dieoffset, attr, form, start + 4,
1.429 + datalen);
1.430 + return start + 4 + datalen;
1.431 + }
1.432 + case DW_FORM_block:
1.433 + case DW_FORM_exprloc: {
1.434 + uint64 datalen = reader_->ReadUnsignedLEB128(start, &len);
1.435 + handler_->ProcessAttributeBuffer(dieoffset, attr, form, start + len,
1.436 + datalen);
1.437 + return start + datalen + len;
1.438 + }
1.439 + case DW_FORM_strp: {
1.440 + assert(string_buffer_ != NULL);
1.441 +
1.442 + const uint64 offset = reader_->ReadOffset(start);
1.443 + assert(string_buffer_ + offset < string_buffer_ + string_buffer_length_);
1.444 +
1.445 + const char* str = string_buffer_ + offset;
1.446 + handler_->ProcessAttributeString(dieoffset, attr, form,
1.447 + str);
1.448 + return start + reader_->OffsetSize();
1.449 + }
1.450 + }
1.451 + fprintf(stderr, "Unhandled form type\n");
1.452 + return NULL;
1.453 +}
1.454 +
1.455 +const char* CompilationUnit::ProcessDIE(uint64 dieoffset,
1.456 + const char* start,
1.457 + const Abbrev& abbrev) {
1.458 + for (AttributeList::const_iterator i = abbrev.attributes.begin();
1.459 + i != abbrev.attributes.end();
1.460 + i++) {
1.461 + start = ProcessAttribute(dieoffset, start, i->first, i->second);
1.462 + }
1.463 + return start;
1.464 +}
1.465 +
1.466 +void CompilationUnit::ProcessDIEs() {
1.467 + const char* dieptr = after_header_;
1.468 + size_t len;
1.469 +
1.470 + // lengthstart is the place the length field is based on.
1.471 + // It is the point in the header after the initial length field
1.472 + const char* lengthstart = buffer_;
1.473 +
1.474 + // In 64 bit dwarf, the initial length is 12 bytes, because of the
1.475 + // 0xffffffff at the start.
1.476 + if (reader_->OffsetSize() == 8)
1.477 + lengthstart += 12;
1.478 + else
1.479 + lengthstart += 4;
1.480 +
1.481 + std::stack<uint64> die_stack;
1.482 +
1.483 + while (dieptr < (lengthstart + header_.length)) {
1.484 + // We give the user the absolute offset from the beginning of
1.485 + // debug_info, since they need it to deal with ref_addr forms.
1.486 + uint64 absolute_offset = (dieptr - buffer_) + offset_from_section_start_;
1.487 +
1.488 + uint64 abbrev_num = reader_->ReadUnsignedLEB128(dieptr, &len);
1.489 +
1.490 + dieptr += len;
1.491 +
1.492 + // Abbrev == 0 represents the end of a list of children, or padding
1.493 + // at the end of the compilation unit.
1.494 + if (abbrev_num == 0) {
1.495 + if (die_stack.size() == 0)
1.496 + // If it is padding, then we are done with the compilation unit's DIEs.
1.497 + return;
1.498 + const uint64 offset = die_stack.top();
1.499 + die_stack.pop();
1.500 + handler_->EndDIE(offset);
1.501 + continue;
1.502 + }
1.503 +
1.504 + const Abbrev& abbrev = abbrevs_->at(static_cast<size_t>(abbrev_num));
1.505 + const enum DwarfTag tag = abbrev.tag;
1.506 + if (!handler_->StartDIE(absolute_offset, tag)) {
1.507 + dieptr = SkipDIE(dieptr, abbrev);
1.508 + } else {
1.509 + dieptr = ProcessDIE(absolute_offset, dieptr, abbrev);
1.510 + }
1.511 +
1.512 + if (abbrev.has_children) {
1.513 + die_stack.push(absolute_offset);
1.514 + } else {
1.515 + handler_->EndDIE(absolute_offset);
1.516 + }
1.517 + }
1.518 +}
1.519 +
1.520 +LineInfo::LineInfo(const char* buffer, uint64 buffer_length,
1.521 + ByteReader* reader, LineInfoHandler* handler):
1.522 + handler_(handler), reader_(reader), buffer_(buffer),
1.523 + buffer_length_(buffer_length) {
1.524 + header_.std_opcode_lengths = NULL;
1.525 +}
1.526 +
1.527 +uint64 LineInfo::Start() {
1.528 + ReadHeader();
1.529 + ReadLines();
1.530 + return after_header_ - buffer_;
1.531 +}
1.532 +
1.533 +// The header for a debug_line section is mildly complicated, because
1.534 +// the line info is very tightly encoded.
1.535 +void LineInfo::ReadHeader() {
1.536 + const char* lineptr = buffer_;
1.537 + size_t initial_length_size;
1.538 +
1.539 + const uint64 initial_length
1.540 + = reader_->ReadInitialLength(lineptr, &initial_length_size);
1.541 +
1.542 + lineptr += initial_length_size;
1.543 + header_.total_length = initial_length;
1.544 + assert(buffer_ + initial_length_size + header_.total_length <=
1.545 + buffer_ + buffer_length_);
1.546 +
1.547 + // Address size *must* be set by CU ahead of time.
1.548 + assert(reader_->AddressSize() != 0);
1.549 +
1.550 + header_.version = reader_->ReadTwoBytes(lineptr);
1.551 + lineptr += 2;
1.552 +
1.553 + header_.prologue_length = reader_->ReadOffset(lineptr);
1.554 + lineptr += reader_->OffsetSize();
1.555 +
1.556 + header_.min_insn_length = reader_->ReadOneByte(lineptr);
1.557 + lineptr += 1;
1.558 +
1.559 + header_.default_is_stmt = reader_->ReadOneByte(lineptr);
1.560 + lineptr += 1;
1.561 +
1.562 + header_.line_base = *reinterpret_cast<const int8*>(lineptr);
1.563 + lineptr += 1;
1.564 +
1.565 + header_.line_range = reader_->ReadOneByte(lineptr);
1.566 + lineptr += 1;
1.567 +
1.568 + header_.opcode_base = reader_->ReadOneByte(lineptr);
1.569 + lineptr += 1;
1.570 +
1.571 + header_.std_opcode_lengths = new std::vector<unsigned char>;
1.572 + header_.std_opcode_lengths->resize(header_.opcode_base + 1);
1.573 + (*header_.std_opcode_lengths)[0] = 0;
1.574 + for (int i = 1; i < header_.opcode_base; i++) {
1.575 + (*header_.std_opcode_lengths)[i] = reader_->ReadOneByte(lineptr);
1.576 + lineptr += 1;
1.577 + }
1.578 +
1.579 + // It is legal for the directory entry table to be empty.
1.580 + if (*lineptr) {
1.581 + uint32 dirindex = 1;
1.582 + while (*lineptr) {
1.583 + const char* dirname = lineptr;
1.584 + handler_->DefineDir(dirname, dirindex);
1.585 + lineptr += strlen(dirname) + 1;
1.586 + dirindex++;
1.587 + }
1.588 + }
1.589 + lineptr++;
1.590 +
1.591 + // It is also legal for the file entry table to be empty.
1.592 + if (*lineptr) {
1.593 + uint32 fileindex = 1;
1.594 + size_t len;
1.595 + while (*lineptr) {
1.596 + const char* filename = lineptr;
1.597 + lineptr += strlen(filename) + 1;
1.598 +
1.599 + uint64 dirindex = reader_->ReadUnsignedLEB128(lineptr, &len);
1.600 + lineptr += len;
1.601 +
1.602 + uint64 mod_time = reader_->ReadUnsignedLEB128(lineptr, &len);
1.603 + lineptr += len;
1.604 +
1.605 + uint64 filelength = reader_->ReadUnsignedLEB128(lineptr, &len);
1.606 + lineptr += len;
1.607 + handler_->DefineFile(filename, fileindex, static_cast<uint32>(dirindex),
1.608 + mod_time, filelength);
1.609 + fileindex++;
1.610 + }
1.611 + }
1.612 + lineptr++;
1.613 +
1.614 + after_header_ = lineptr;
1.615 +}
1.616 +
1.617 +/* static */
1.618 +bool LineInfo::ProcessOneOpcode(ByteReader* reader,
1.619 + LineInfoHandler* handler,
1.620 + const struct LineInfoHeader &header,
1.621 + const char* start,
1.622 + struct LineStateMachine* lsm,
1.623 + size_t* len,
1.624 + uintptr pc,
1.625 + bool *lsm_passes_pc) {
1.626 + size_t oplen = 0;
1.627 + size_t templen;
1.628 + uint8 opcode = reader->ReadOneByte(start);
1.629 + oplen++;
1.630 + start++;
1.631 +
1.632 + // If the opcode is great than the opcode_base, it is a special
1.633 + // opcode. Most line programs consist mainly of special opcodes.
1.634 + if (opcode >= header.opcode_base) {
1.635 + opcode -= header.opcode_base;
1.636 + const int64 advance_address = (opcode / header.line_range)
1.637 + * header.min_insn_length;
1.638 + const int32 advance_line = (opcode % header.line_range)
1.639 + + header.line_base;
1.640 +
1.641 + // Check if the lsm passes "pc". If so, mark it as passed.
1.642 + if (lsm_passes_pc &&
1.643 + lsm->address <= pc && pc < lsm->address + advance_address) {
1.644 + *lsm_passes_pc = true;
1.645 + }
1.646 +
1.647 + lsm->address += advance_address;
1.648 + lsm->line_num += advance_line;
1.649 + lsm->basic_block = true;
1.650 + *len = oplen;
1.651 + return true;
1.652 + }
1.653 +
1.654 + // Otherwise, we have the regular opcodes
1.655 + switch (opcode) {
1.656 + case DW_LNS_copy: {
1.657 + lsm->basic_block = false;
1.658 + *len = oplen;
1.659 + return true;
1.660 + }
1.661 +
1.662 + case DW_LNS_advance_pc: {
1.663 + uint64 advance_address = reader->ReadUnsignedLEB128(start, &templen);
1.664 + oplen += templen;
1.665 +
1.666 + // Check if the lsm passes "pc". If so, mark it as passed.
1.667 + if (lsm_passes_pc && lsm->address <= pc &&
1.668 + pc < lsm->address + header.min_insn_length * advance_address) {
1.669 + *lsm_passes_pc = true;
1.670 + }
1.671 +
1.672 + lsm->address += header.min_insn_length * advance_address;
1.673 + }
1.674 + break;
1.675 + case DW_LNS_advance_line: {
1.676 + const int64 advance_line = reader->ReadSignedLEB128(start, &templen);
1.677 + oplen += templen;
1.678 + lsm->line_num += static_cast<int32>(advance_line);
1.679 +
1.680 + // With gcc 4.2.1, we can get the line_no here for the first time
1.681 + // since DW_LNS_advance_line is called after DW_LNE_set_address is
1.682 + // called. So we check if the lsm passes "pc" here, not in
1.683 + // DW_LNE_set_address.
1.684 + if (lsm_passes_pc && lsm->address == pc) {
1.685 + *lsm_passes_pc = true;
1.686 + }
1.687 + }
1.688 + break;
1.689 + case DW_LNS_set_file: {
1.690 + const uint64 fileno = reader->ReadUnsignedLEB128(start, &templen);
1.691 + oplen += templen;
1.692 + lsm->file_num = static_cast<uint32>(fileno);
1.693 + }
1.694 + break;
1.695 + case DW_LNS_set_column: {
1.696 + const uint64 colno = reader->ReadUnsignedLEB128(start, &templen);
1.697 + oplen += templen;
1.698 + lsm->column_num = static_cast<uint32>(colno);
1.699 + }
1.700 + break;
1.701 + case DW_LNS_negate_stmt: {
1.702 + lsm->is_stmt = !lsm->is_stmt;
1.703 + }
1.704 + break;
1.705 + case DW_LNS_set_basic_block: {
1.706 + lsm->basic_block = true;
1.707 + }
1.708 + break;
1.709 + case DW_LNS_fixed_advance_pc: {
1.710 + const uint16 advance_address = reader->ReadTwoBytes(start);
1.711 + oplen += 2;
1.712 +
1.713 + // Check if the lsm passes "pc". If so, mark it as passed.
1.714 + if (lsm_passes_pc &&
1.715 + lsm->address <= pc && pc < lsm->address + advance_address) {
1.716 + *lsm_passes_pc = true;
1.717 + }
1.718 +
1.719 + lsm->address += advance_address;
1.720 + }
1.721 + break;
1.722 + case DW_LNS_const_add_pc: {
1.723 + const int64 advance_address = header.min_insn_length
1.724 + * ((255 - header.opcode_base)
1.725 + / header.line_range);
1.726 +
1.727 + // Check if the lsm passes "pc". If so, mark it as passed.
1.728 + if (lsm_passes_pc &&
1.729 + lsm->address <= pc && pc < lsm->address + advance_address) {
1.730 + *lsm_passes_pc = true;
1.731 + }
1.732 +
1.733 + lsm->address += advance_address;
1.734 + }
1.735 + break;
1.736 + case DW_LNS_extended_op: {
1.737 + const uint64 extended_op_len = reader->ReadUnsignedLEB128(start,
1.738 + &templen);
1.739 + start += templen;
1.740 + oplen += templen + extended_op_len;
1.741 +
1.742 + const uint64 extended_op = reader->ReadOneByte(start);
1.743 + start++;
1.744 +
1.745 + switch (extended_op) {
1.746 + case DW_LNE_end_sequence: {
1.747 + lsm->end_sequence = true;
1.748 + *len = oplen;
1.749 + return true;
1.750 + }
1.751 + break;
1.752 + case DW_LNE_set_address: {
1.753 + // With gcc 4.2.1, we cannot tell the line_no here since
1.754 + // DW_LNE_set_address is called before DW_LNS_advance_line is
1.755 + // called. So we do not check if the lsm passes "pc" here. See
1.756 + // also the comment in DW_LNS_advance_line.
1.757 + uint64 address = reader->ReadAddress(start);
1.758 + lsm->address = address;
1.759 + }
1.760 + break;
1.761 + case DW_LNE_define_file: {
1.762 + const char* filename = start;
1.763 +
1.764 + templen = strlen(filename) + 1;
1.765 + start += templen;
1.766 +
1.767 + uint64 dirindex = reader->ReadUnsignedLEB128(start, &templen);
1.768 + oplen += templen;
1.769 +
1.770 + const uint64 mod_time = reader->ReadUnsignedLEB128(start,
1.771 + &templen);
1.772 + oplen += templen;
1.773 +
1.774 + const uint64 filelength = reader->ReadUnsignedLEB128(start,
1.775 + &templen);
1.776 + oplen += templen;
1.777 +
1.778 + if (handler) {
1.779 + handler->DefineFile(filename, -1, static_cast<uint32>(dirindex),
1.780 + mod_time, filelength);
1.781 + }
1.782 + }
1.783 + break;
1.784 + }
1.785 + }
1.786 + break;
1.787 +
1.788 + default: {
1.789 + // Ignore unknown opcode silently
1.790 + if (header.std_opcode_lengths) {
1.791 + for (int i = 0; i < (*header.std_opcode_lengths)[opcode]; i++) {
1.792 + reader->ReadUnsignedLEB128(start, &templen);
1.793 + start += templen;
1.794 + oplen += templen;
1.795 + }
1.796 + }
1.797 + }
1.798 + break;
1.799 + }
1.800 + *len = oplen;
1.801 + return false;
1.802 +}
1.803 +
1.804 +void LineInfo::ReadLines() {
1.805 + struct LineStateMachine lsm;
1.806 +
1.807 + // lengthstart is the place the length field is based on.
1.808 + // It is the point in the header after the initial length field
1.809 + const char* lengthstart = buffer_;
1.810 +
1.811 + // In 64 bit dwarf, the initial length is 12 bytes, because of the
1.812 + // 0xffffffff at the start.
1.813 + if (reader_->OffsetSize() == 8)
1.814 + lengthstart += 12;
1.815 + else
1.816 + lengthstart += 4;
1.817 +
1.818 + const char* lineptr = after_header_;
1.819 + lsm.Reset(header_.default_is_stmt);
1.820 +
1.821 + // The LineInfoHandler interface expects each line's length along
1.822 + // with its address, but DWARF only provides addresses (sans
1.823 + // length), and an end-of-sequence address; one infers the length
1.824 + // from the next address. So we report a line only when we get the
1.825 + // next line's address, or the end-of-sequence address.
1.826 + bool have_pending_line = false;
1.827 + uint64 pending_address = 0;
1.828 + uint32 pending_file_num = 0, pending_line_num = 0, pending_column_num = 0;
1.829 +
1.830 + while (lineptr < lengthstart + header_.total_length) {
1.831 + size_t oplength;
1.832 + bool add_row = ProcessOneOpcode(reader_, handler_, header_,
1.833 + lineptr, &lsm, &oplength, (uintptr)-1,
1.834 + NULL);
1.835 + if (add_row) {
1.836 + if (have_pending_line)
1.837 + handler_->AddLine(pending_address, lsm.address - pending_address,
1.838 + pending_file_num, pending_line_num,
1.839 + pending_column_num);
1.840 + if (lsm.end_sequence) {
1.841 + lsm.Reset(header_.default_is_stmt);
1.842 + have_pending_line = false;
1.843 + } else {
1.844 + pending_address = lsm.address;
1.845 + pending_file_num = lsm.file_num;
1.846 + pending_line_num = lsm.line_num;
1.847 + pending_column_num = lsm.column_num;
1.848 + have_pending_line = true;
1.849 + }
1.850 + }
1.851 + lineptr += oplength;
1.852 + }
1.853 +
1.854 + after_header_ = lengthstart + header_.total_length;
1.855 +}
1.856 +
1.857 +// A DWARF rule for recovering the address or value of a register, or
1.858 +// computing the canonical frame address. There is one subclass of this for
1.859 +// each '*Rule' member function in CallFrameInfo::Handler.
1.860 +//
1.861 +// It's annoying that we have to handle Rules using pointers (because
1.862 +// the concrete instances can have an arbitrary size). They're small,
1.863 +// so it would be much nicer if we could just handle them by value
1.864 +// instead of fretting about ownership and destruction.
1.865 +//
1.866 +// It seems like all these could simply be instances of std::tr1::bind,
1.867 +// except that we need instances to be EqualityComparable, too.
1.868 +//
1.869 +// This could logically be nested within State, but then the qualified names
1.870 +// get horrendous.
1.871 +class CallFrameInfo::Rule {
1.872 + public:
1.873 + virtual ~Rule() { }
1.874 +
1.875 + // Tell HANDLER that, at ADDRESS in the program, REGISTER can be
1.876 + // recovered using this rule. If REGISTER is kCFARegister, then this rule
1.877 + // describes how to compute the canonical frame address. Return what the
1.878 + // HANDLER member function returned.
1.879 + virtual bool Handle(Handler *handler,
1.880 + uint64 address, int register) const = 0;
1.881 +
1.882 + // Equality on rules. We use these to decide which rules we need
1.883 + // to report after a DW_CFA_restore_state instruction.
1.884 + virtual bool operator==(const Rule &rhs) const = 0;
1.885 +
1.886 + bool operator!=(const Rule &rhs) const { return ! (*this == rhs); }
1.887 +
1.888 + // Return a pointer to a copy of this rule.
1.889 + virtual Rule *Copy() const = 0;
1.890 +
1.891 + // If this is a base+offset rule, change its base register to REG.
1.892 + // Otherwise, do nothing. (Ugly, but required for DW_CFA_def_cfa_register.)
1.893 + virtual void SetBaseRegister(unsigned reg) { }
1.894 +
1.895 + // If this is a base+offset rule, change its offset to OFFSET. Otherwise,
1.896 + // do nothing. (Ugly, but required for DW_CFA_def_cfa_offset.)
1.897 + virtual void SetOffset(long long offset) { }
1.898 +
1.899 + // A RTTI workaround, to make it possible to implement equality
1.900 + // comparisons on classes derived from this one.
1.901 + enum CFIRTag {
1.902 + CFIR_UNDEFINED_RULE,
1.903 + CFIR_SAME_VALUE_RULE,
1.904 + CFIR_OFFSET_RULE,
1.905 + CFIR_VAL_OFFSET_RULE,
1.906 + CFIR_REGISTER_RULE,
1.907 + CFIR_EXPRESSION_RULE,
1.908 + CFIR_VAL_EXPRESSION_RULE
1.909 + };
1.910 +
1.911 + // Produce the tag that identifies the child class of this object.
1.912 + virtual CFIRTag getTag() const = 0;
1.913 +};
1.914 +
1.915 +// Rule: the value the register had in the caller cannot be recovered.
1.916 +class CallFrameInfo::UndefinedRule: public CallFrameInfo::Rule {
1.917 + public:
1.918 + UndefinedRule() { }
1.919 + ~UndefinedRule() { }
1.920 + CFIRTag getTag() const { return CFIR_UNDEFINED_RULE; }
1.921 + bool Handle(Handler *handler, uint64 address, int reg) const {
1.922 + return handler->UndefinedRule(address, reg);
1.923 + }
1.924 + bool operator==(const Rule &rhs) const {
1.925 + if (rhs.getTag() != CFIR_UNDEFINED_RULE) return false;
1.926 + return true;
1.927 + }
1.928 + Rule *Copy() const { return new UndefinedRule(*this); }
1.929 +};
1.930 +
1.931 +// Rule: the register's value is the same as that it had in the caller.
1.932 +class CallFrameInfo::SameValueRule: public CallFrameInfo::Rule {
1.933 + public:
1.934 + SameValueRule() { }
1.935 + ~SameValueRule() { }
1.936 + CFIRTag getTag() const { return CFIR_SAME_VALUE_RULE; }
1.937 + bool Handle(Handler *handler, uint64 address, int reg) const {
1.938 + return handler->SameValueRule(address, reg);
1.939 + }
1.940 + bool operator==(const Rule &rhs) const {
1.941 + if (rhs.getTag() != CFIR_SAME_VALUE_RULE) return false;
1.942 + return true;
1.943 + }
1.944 + Rule *Copy() const { return new SameValueRule(*this); }
1.945 +};
1.946 +
1.947 +// Rule: the register is saved at OFFSET from BASE_REGISTER. BASE_REGISTER
1.948 +// may be CallFrameInfo::Handler::kCFARegister.
1.949 +class CallFrameInfo::OffsetRule: public CallFrameInfo::Rule {
1.950 + public:
1.951 + OffsetRule(int base_register, long offset)
1.952 + : base_register_(base_register), offset_(offset) { }
1.953 + ~OffsetRule() { }
1.954 + CFIRTag getTag() const { return CFIR_OFFSET_RULE; }
1.955 + bool Handle(Handler *handler, uint64 address, int reg) const {
1.956 + return handler->OffsetRule(address, reg, base_register_, offset_);
1.957 + }
1.958 + bool operator==(const Rule &rhs) const {
1.959 + if (rhs.getTag() != CFIR_OFFSET_RULE) return false;
1.960 + const OffsetRule *our_rhs = static_cast<const OffsetRule *>(&rhs);
1.961 + return (base_register_ == our_rhs->base_register_ &&
1.962 + offset_ == our_rhs->offset_);
1.963 + }
1.964 + Rule *Copy() const { return new OffsetRule(*this); }
1.965 + // We don't actually need SetBaseRegister or SetOffset here, since they
1.966 + // are only ever applied to CFA rules, for DW_CFA_def_cfa_offset, and it
1.967 + // doesn't make sense to use OffsetRule for computing the CFA: it
1.968 + // computes the address at which a register is saved, not a value.
1.969 + private:
1.970 + int base_register_;
1.971 + long offset_;
1.972 +};
1.973 +
1.974 +// Rule: the value the register had in the caller is the value of
1.975 +// BASE_REGISTER plus offset. BASE_REGISTER may be
1.976 +// CallFrameInfo::Handler::kCFARegister.
1.977 +class CallFrameInfo::ValOffsetRule: public CallFrameInfo::Rule {
1.978 + public:
1.979 + ValOffsetRule(int base_register, long offset)
1.980 + : base_register_(base_register), offset_(offset) { }
1.981 + ~ValOffsetRule() { }
1.982 + CFIRTag getTag() const { return CFIR_VAL_OFFSET_RULE; }
1.983 + bool Handle(Handler *handler, uint64 address, int reg) const {
1.984 + return handler->ValOffsetRule(address, reg, base_register_, offset_);
1.985 + }
1.986 + bool operator==(const Rule &rhs) const {
1.987 + if (rhs.getTag() != CFIR_VAL_OFFSET_RULE) return false;
1.988 + const ValOffsetRule *our_rhs = static_cast<const ValOffsetRule *>(&rhs);
1.989 + return (base_register_ == our_rhs->base_register_ &&
1.990 + offset_ == our_rhs->offset_);
1.991 + }
1.992 + Rule *Copy() const { return new ValOffsetRule(*this); }
1.993 + void SetBaseRegister(unsigned reg) { base_register_ = reg; }
1.994 + void SetOffset(long long offset) { offset_ = offset; }
1.995 + private:
1.996 + int base_register_;
1.997 + long offset_;
1.998 +};
1.999 +
1.1000 +// Rule: the register has been saved in another register REGISTER_NUMBER_.
1.1001 +class CallFrameInfo::RegisterRule: public CallFrameInfo::Rule {
1.1002 + public:
1.1003 + explicit RegisterRule(int register_number)
1.1004 + : register_number_(register_number) { }
1.1005 + ~RegisterRule() { }
1.1006 + CFIRTag getTag() const { return CFIR_REGISTER_RULE; }
1.1007 + bool Handle(Handler *handler, uint64 address, int reg) const {
1.1008 + return handler->RegisterRule(address, reg, register_number_);
1.1009 + }
1.1010 + bool operator==(const Rule &rhs) const {
1.1011 + if (rhs.getTag() != CFIR_REGISTER_RULE) return false;
1.1012 + const RegisterRule *our_rhs = static_cast<const RegisterRule *>(&rhs);
1.1013 + return (register_number_ == our_rhs->register_number_);
1.1014 + }
1.1015 + Rule *Copy() const { return new RegisterRule(*this); }
1.1016 + private:
1.1017 + int register_number_;
1.1018 +};
1.1019 +
1.1020 +// Rule: EXPRESSION evaluates to the address at which the register is saved.
1.1021 +class CallFrameInfo::ExpressionRule: public CallFrameInfo::Rule {
1.1022 + public:
1.1023 + explicit ExpressionRule(const string &expression)
1.1024 + : expression_(expression) { }
1.1025 + ~ExpressionRule() { }
1.1026 + CFIRTag getTag() const { return CFIR_EXPRESSION_RULE; }
1.1027 + bool Handle(Handler *handler, uint64 address, int reg) const {
1.1028 + return handler->ExpressionRule(address, reg, expression_);
1.1029 + }
1.1030 + bool operator==(const Rule &rhs) const {
1.1031 + if (rhs.getTag() != CFIR_EXPRESSION_RULE) return false;
1.1032 + const ExpressionRule *our_rhs = static_cast<const ExpressionRule *>(&rhs);
1.1033 + return (expression_ == our_rhs->expression_);
1.1034 + }
1.1035 + Rule *Copy() const { return new ExpressionRule(*this); }
1.1036 + private:
1.1037 + string expression_;
1.1038 +};
1.1039 +
1.1040 +// Rule: EXPRESSION evaluates to the address at which the register is saved.
1.1041 +class CallFrameInfo::ValExpressionRule: public CallFrameInfo::Rule {
1.1042 + public:
1.1043 + explicit ValExpressionRule(const string &expression)
1.1044 + : expression_(expression) { }
1.1045 + ~ValExpressionRule() { }
1.1046 + CFIRTag getTag() const { return CFIR_VAL_EXPRESSION_RULE; }
1.1047 + bool Handle(Handler *handler, uint64 address, int reg) const {
1.1048 + return handler->ValExpressionRule(address, reg, expression_);
1.1049 + }
1.1050 + bool operator==(const Rule &rhs) const {
1.1051 + if (rhs.getTag() != CFIR_VAL_EXPRESSION_RULE) return false;
1.1052 + const ValExpressionRule *our_rhs =
1.1053 + static_cast<const ValExpressionRule *>(&rhs);
1.1054 + return (expression_ == our_rhs->expression_);
1.1055 + }
1.1056 + Rule *Copy() const { return new ValExpressionRule(*this); }
1.1057 + private:
1.1058 + string expression_;
1.1059 +};
1.1060 +
1.1061 +// A map from register numbers to rules.
1.1062 +class CallFrameInfo::RuleMap {
1.1063 + public:
1.1064 + RuleMap() : cfa_rule_(NULL) { }
1.1065 + RuleMap(const RuleMap &rhs) : cfa_rule_(NULL) { *this = rhs; }
1.1066 + ~RuleMap() { Clear(); }
1.1067 +
1.1068 + RuleMap &operator=(const RuleMap &rhs);
1.1069 +
1.1070 + // Set the rule for computing the CFA to RULE. Take ownership of RULE.
1.1071 + void SetCFARule(Rule *rule) { delete cfa_rule_; cfa_rule_ = rule; }
1.1072 +
1.1073 + // Return the current CFA rule. Unlike RegisterRule, this RuleMap retains
1.1074 + // ownership of the rule. We use this for DW_CFA_def_cfa_offset and
1.1075 + // DW_CFA_def_cfa_register, and for detecting references to the CFA before
1.1076 + // a rule for it has been established.
1.1077 + Rule *CFARule() const { return cfa_rule_; }
1.1078 +
1.1079 + // Return the rule for REG, or NULL if there is none. The caller takes
1.1080 + // ownership of the result.
1.1081 + Rule *RegisterRule(int reg) const;
1.1082 +
1.1083 + // Set the rule for computing REG to RULE. Take ownership of RULE.
1.1084 + void SetRegisterRule(int reg, Rule *rule);
1.1085 +
1.1086 + // Make all the appropriate calls to HANDLER as if we were changing from
1.1087 + // this RuleMap to NEW_RULES at ADDRESS. We use this to implement
1.1088 + // DW_CFA_restore_state, where lots of rules can change simultaneously.
1.1089 + // Return true if all handlers returned true; otherwise, return false.
1.1090 + bool HandleTransitionTo(Handler *handler, uint64 address,
1.1091 + const RuleMap &new_rules) const;
1.1092 +
1.1093 + private:
1.1094 + // A map from register numbers to Rules.
1.1095 + typedef std::map<int, Rule *> RuleByNumber;
1.1096 +
1.1097 + // Remove all register rules and clear cfa_rule_.
1.1098 + void Clear();
1.1099 +
1.1100 + // The rule for computing the canonical frame address. This RuleMap owns
1.1101 + // this rule.
1.1102 + Rule *cfa_rule_;
1.1103 +
1.1104 + // A map from register numbers to postfix expressions to recover
1.1105 + // their values. This RuleMap owns the Rules the map refers to.
1.1106 + RuleByNumber registers_;
1.1107 +};
1.1108 +
1.1109 +CallFrameInfo::RuleMap &CallFrameInfo::RuleMap::operator=(const RuleMap &rhs) {
1.1110 + Clear();
1.1111 + // Since each map owns the rules it refers to, assignment must copy them.
1.1112 + if (rhs.cfa_rule_) cfa_rule_ = rhs.cfa_rule_->Copy();
1.1113 + for (RuleByNumber::const_iterator it = rhs.registers_.begin();
1.1114 + it != rhs.registers_.end(); it++)
1.1115 + registers_[it->first] = it->second->Copy();
1.1116 + return *this;
1.1117 +}
1.1118 +
1.1119 +CallFrameInfo::Rule *CallFrameInfo::RuleMap::RegisterRule(int reg) const {
1.1120 + assert(reg != Handler::kCFARegister);
1.1121 + RuleByNumber::const_iterator it = registers_.find(reg);
1.1122 + if (it != registers_.end())
1.1123 + return it->second->Copy();
1.1124 + else
1.1125 + return NULL;
1.1126 +}
1.1127 +
1.1128 +void CallFrameInfo::RuleMap::SetRegisterRule(int reg, Rule *rule) {
1.1129 + assert(reg != Handler::kCFARegister);
1.1130 + assert(rule);
1.1131 + Rule **slot = ®isters_[reg];
1.1132 + delete *slot;
1.1133 + *slot = rule;
1.1134 +}
1.1135 +
1.1136 +bool CallFrameInfo::RuleMap::HandleTransitionTo(
1.1137 + Handler *handler,
1.1138 + uint64 address,
1.1139 + const RuleMap &new_rules) const {
1.1140 + // Transition from cfa_rule_ to new_rules.cfa_rule_.
1.1141 + if (cfa_rule_ && new_rules.cfa_rule_) {
1.1142 + if (*cfa_rule_ != *new_rules.cfa_rule_ &&
1.1143 + !new_rules.cfa_rule_->Handle(handler, address,
1.1144 + Handler::kCFARegister))
1.1145 + return false;
1.1146 + } else if (cfa_rule_) {
1.1147 + // this RuleMap has a CFA rule but new_rules doesn't.
1.1148 + // CallFrameInfo::Handler has no way to handle this --- and shouldn't;
1.1149 + // it's garbage input. The instruction interpreter should have
1.1150 + // detected this and warned, so take no action here.
1.1151 + } else if (new_rules.cfa_rule_) {
1.1152 + // This shouldn't be possible: NEW_RULES is some prior state, and
1.1153 + // there's no way to remove entries.
1.1154 + assert(0);
1.1155 + } else {
1.1156 + // Both CFA rules are empty. No action needed.
1.1157 + }
1.1158 +
1.1159 + // Traverse the two maps in order by register number, and report
1.1160 + // whatever differences we find.
1.1161 + RuleByNumber::const_iterator old_it = registers_.begin();
1.1162 + RuleByNumber::const_iterator new_it = new_rules.registers_.begin();
1.1163 + while (old_it != registers_.end() && new_it != new_rules.registers_.end()) {
1.1164 + if (old_it->first < new_it->first) {
1.1165 + // This RuleMap has an entry for old_it->first, but NEW_RULES
1.1166 + // doesn't.
1.1167 + //
1.1168 + // This isn't really the right thing to do, but since CFI generally
1.1169 + // only mentions callee-saves registers, and GCC's convention for
1.1170 + // callee-saves registers is that they are unchanged, it's a good
1.1171 + // approximation.
1.1172 + if (!handler->SameValueRule(address, old_it->first))
1.1173 + return false;
1.1174 + old_it++;
1.1175 + } else if (old_it->first > new_it->first) {
1.1176 + // NEW_RULES has entry for new_it->first, but this RuleMap
1.1177 + // doesn't. This shouldn't be possible: NEW_RULES is some prior
1.1178 + // state, and there's no way to remove entries.
1.1179 + assert(0);
1.1180 + } else {
1.1181 + // Both maps have an entry for this register. Report the new
1.1182 + // rule if it is different.
1.1183 + if (*old_it->second != *new_it->second &&
1.1184 + !new_it->second->Handle(handler, address, new_it->first))
1.1185 + return false;
1.1186 + new_it++, old_it++;
1.1187 + }
1.1188 + }
1.1189 + // Finish off entries from this RuleMap with no counterparts in new_rules.
1.1190 + while (old_it != registers_.end()) {
1.1191 + if (!handler->SameValueRule(address, old_it->first))
1.1192 + return false;
1.1193 + old_it++;
1.1194 + }
1.1195 + // Since we only make transitions from a rule set to some previously
1.1196 + // saved rule set, and we can only add rules to the map, NEW_RULES
1.1197 + // must have fewer rules than *this.
1.1198 + assert(new_it == new_rules.registers_.end());
1.1199 +
1.1200 + return true;
1.1201 +}
1.1202 +
1.1203 +// Remove all register rules and clear cfa_rule_.
1.1204 +void CallFrameInfo::RuleMap::Clear() {
1.1205 + delete cfa_rule_;
1.1206 + cfa_rule_ = NULL;
1.1207 + for (RuleByNumber::iterator it = registers_.begin();
1.1208 + it != registers_.end(); it++)
1.1209 + delete it->second;
1.1210 + registers_.clear();
1.1211 +}
1.1212 +
1.1213 +// The state of the call frame information interpreter as it processes
1.1214 +// instructions from a CIE and FDE.
1.1215 +class CallFrameInfo::State {
1.1216 + public:
1.1217 + // Create a call frame information interpreter state with the given
1.1218 + // reporter, reader, handler, and initial call frame info address.
1.1219 + State(ByteReader *reader, Handler *handler, Reporter *reporter,
1.1220 + uint64 address)
1.1221 + : reader_(reader), handler_(handler), reporter_(reporter),
1.1222 + address_(address), entry_(NULL), cursor_(NULL) { }
1.1223 +
1.1224 + // Interpret instructions from CIE, save the resulting rule set for
1.1225 + // DW_CFA_restore instructions, and return true. On error, report
1.1226 + // the problem to reporter_ and return false.
1.1227 + bool InterpretCIE(const CIE &cie);
1.1228 +
1.1229 + // Interpret instructions from FDE, and return true. On error,
1.1230 + // report the problem to reporter_ and return false.
1.1231 + bool InterpretFDE(const FDE &fde);
1.1232 +
1.1233 + private:
1.1234 + // The operands of a CFI instruction, for ParseOperands.
1.1235 + struct Operands {
1.1236 + unsigned register_number; // A register number.
1.1237 + uint64 offset; // An offset or address.
1.1238 + long signed_offset; // A signed offset.
1.1239 + string expression; // A DWARF expression.
1.1240 + };
1.1241 +
1.1242 + // Parse CFI instruction operands from STATE's instruction stream as
1.1243 + // described by FORMAT. On success, populate OPERANDS with the
1.1244 + // results, and return true. On failure, report the problem and
1.1245 + // return false.
1.1246 + //
1.1247 + // Each character of FORMAT should be one of the following:
1.1248 + //
1.1249 + // 'r' unsigned LEB128 register number (OPERANDS->register_number)
1.1250 + // 'o' unsigned LEB128 offset (OPERANDS->offset)
1.1251 + // 's' signed LEB128 offset (OPERANDS->signed_offset)
1.1252 + // 'a' machine-size address (OPERANDS->offset)
1.1253 + // (If the CIE has a 'z' augmentation string, 'a' uses the
1.1254 + // encoding specified by the 'R' argument.)
1.1255 + // '1' a one-byte offset (OPERANDS->offset)
1.1256 + // '2' a two-byte offset (OPERANDS->offset)
1.1257 + // '4' a four-byte offset (OPERANDS->offset)
1.1258 + // '8' an eight-byte offset (OPERANDS->offset)
1.1259 + // 'e' a DW_FORM_block holding a (OPERANDS->expression)
1.1260 + // DWARF expression
1.1261 + bool ParseOperands(const char *format, Operands *operands);
1.1262 +
1.1263 + // Interpret one CFI instruction from STATE's instruction stream, update
1.1264 + // STATE, report any rule changes to handler_, and return true. On
1.1265 + // failure, report the problem and return false.
1.1266 + bool DoInstruction();
1.1267 +
1.1268 + // The following Do* member functions are subroutines of DoInstruction,
1.1269 + // factoring out the actual work of operations that have several
1.1270 + // different encodings.
1.1271 +
1.1272 + // Set the CFA rule to be the value of BASE_REGISTER plus OFFSET, and
1.1273 + // return true. On failure, report and return false. (Used for
1.1274 + // DW_CFA_def_cfa and DW_CFA_def_cfa_sf.)
1.1275 + bool DoDefCFA(unsigned base_register, long offset);
1.1276 +
1.1277 + // Change the offset of the CFA rule to OFFSET, and return true. On
1.1278 + // failure, report and return false. (Subroutine for
1.1279 + // DW_CFA_def_cfa_offset and DW_CFA_def_cfa_offset_sf.)
1.1280 + bool DoDefCFAOffset(long offset);
1.1281 +
1.1282 + // Specify that REG can be recovered using RULE, and return true. On
1.1283 + // failure, report and return false.
1.1284 + bool DoRule(unsigned reg, Rule *rule);
1.1285 +
1.1286 + // Specify that REG can be found at OFFSET from the CFA, and return true.
1.1287 + // On failure, report and return false. (Subroutine for DW_CFA_offset,
1.1288 + // DW_CFA_offset_extended, and DW_CFA_offset_extended_sf.)
1.1289 + bool DoOffset(unsigned reg, long offset);
1.1290 +
1.1291 + // Specify that the caller's value for REG is the CFA plus OFFSET,
1.1292 + // and return true. On failure, report and return false. (Subroutine
1.1293 + // for DW_CFA_val_offset and DW_CFA_val_offset_sf.)
1.1294 + bool DoValOffset(unsigned reg, long offset);
1.1295 +
1.1296 + // Restore REG to the rule established in the CIE, and return true. On
1.1297 + // failure, report and return false. (Subroutine for DW_CFA_restore and
1.1298 + // DW_CFA_restore_extended.)
1.1299 + bool DoRestore(unsigned reg);
1.1300 +
1.1301 + // Return the section offset of the instruction at cursor. For use
1.1302 + // in error messages.
1.1303 + uint64 CursorOffset() { return entry_->offset + (cursor_ - entry_->start); }
1.1304 +
1.1305 + // Report that entry_ is incomplete, and return false. For brevity.
1.1306 + bool ReportIncomplete() {
1.1307 + reporter_->Incomplete(entry_->offset, entry_->kind);
1.1308 + return false;
1.1309 + }
1.1310 +
1.1311 + // For reading multi-byte values with the appropriate endianness.
1.1312 + ByteReader *reader_;
1.1313 +
1.1314 + // The handler to which we should report the data we find.
1.1315 + Handler *handler_;
1.1316 +
1.1317 + // For reporting problems in the info we're parsing.
1.1318 + Reporter *reporter_;
1.1319 +
1.1320 + // The code address to which the next instruction in the stream applies.
1.1321 + uint64 address_;
1.1322 +
1.1323 + // The entry whose instructions we are currently processing. This is
1.1324 + // first a CIE, and then an FDE.
1.1325 + const Entry *entry_;
1.1326 +
1.1327 + // The next instruction to process.
1.1328 + const char *cursor_;
1.1329 +
1.1330 + // The current set of rules.
1.1331 + RuleMap rules_;
1.1332 +
1.1333 + // The set of rules established by the CIE, used by DW_CFA_restore
1.1334 + // and DW_CFA_restore_extended. We set this after interpreting the
1.1335 + // CIE's instructions.
1.1336 + RuleMap cie_rules_;
1.1337 +
1.1338 + // A stack of saved states, for DW_CFA_remember_state and
1.1339 + // DW_CFA_restore_state.
1.1340 + std::stack<RuleMap> saved_rules_;
1.1341 +};
1.1342 +
1.1343 +bool CallFrameInfo::State::InterpretCIE(const CIE &cie) {
1.1344 + entry_ = &cie;
1.1345 + cursor_ = entry_->instructions;
1.1346 + while (cursor_ < entry_->end)
1.1347 + if (!DoInstruction())
1.1348 + return false;
1.1349 + // Note the rules established by the CIE, for use by DW_CFA_restore
1.1350 + // and DW_CFA_restore_extended.
1.1351 + cie_rules_ = rules_;
1.1352 + return true;
1.1353 +}
1.1354 +
1.1355 +bool CallFrameInfo::State::InterpretFDE(const FDE &fde) {
1.1356 + entry_ = &fde;
1.1357 + cursor_ = entry_->instructions;
1.1358 + while (cursor_ < entry_->end)
1.1359 + if (!DoInstruction())
1.1360 + return false;
1.1361 + return true;
1.1362 +}
1.1363 +
1.1364 +bool CallFrameInfo::State::ParseOperands(const char *format,
1.1365 + Operands *operands) {
1.1366 + size_t len;
1.1367 + const char *operand;
1.1368 +
1.1369 + for (operand = format; *operand; operand++) {
1.1370 + size_t bytes_left = entry_->end - cursor_;
1.1371 + switch (*operand) {
1.1372 + case 'r':
1.1373 + operands->register_number = reader_->ReadUnsignedLEB128(cursor_, &len);
1.1374 + if (len > bytes_left) return ReportIncomplete();
1.1375 + cursor_ += len;
1.1376 + break;
1.1377 +
1.1378 + case 'o':
1.1379 + operands->offset = reader_->ReadUnsignedLEB128(cursor_, &len);
1.1380 + if (len > bytes_left) return ReportIncomplete();
1.1381 + cursor_ += len;
1.1382 + break;
1.1383 +
1.1384 + case 's':
1.1385 + operands->signed_offset = reader_->ReadSignedLEB128(cursor_, &len);
1.1386 + if (len > bytes_left) return ReportIncomplete();
1.1387 + cursor_ += len;
1.1388 + break;
1.1389 +
1.1390 + case 'a':
1.1391 + operands->offset =
1.1392 + reader_->ReadEncodedPointer(cursor_, entry_->cie->pointer_encoding,
1.1393 + &len);
1.1394 + if (len > bytes_left) return ReportIncomplete();
1.1395 + cursor_ += len;
1.1396 + break;
1.1397 +
1.1398 + case '1':
1.1399 + if (1 > bytes_left) return ReportIncomplete();
1.1400 + operands->offset = static_cast<unsigned char>(*cursor_++);
1.1401 + break;
1.1402 +
1.1403 + case '2':
1.1404 + if (2 > bytes_left) return ReportIncomplete();
1.1405 + operands->offset = reader_->ReadTwoBytes(cursor_);
1.1406 + cursor_ += 2;
1.1407 + break;
1.1408 +
1.1409 + case '4':
1.1410 + if (4 > bytes_left) return ReportIncomplete();
1.1411 + operands->offset = reader_->ReadFourBytes(cursor_);
1.1412 + cursor_ += 4;
1.1413 + break;
1.1414 +
1.1415 + case '8':
1.1416 + if (8 > bytes_left) return ReportIncomplete();
1.1417 + operands->offset = reader_->ReadEightBytes(cursor_);
1.1418 + cursor_ += 8;
1.1419 + break;
1.1420 +
1.1421 + case 'e': {
1.1422 + size_t expression_length = reader_->ReadUnsignedLEB128(cursor_, &len);
1.1423 + if (len > bytes_left || expression_length > bytes_left - len)
1.1424 + return ReportIncomplete();
1.1425 + cursor_ += len;
1.1426 + operands->expression = string(cursor_, expression_length);
1.1427 + cursor_ += expression_length;
1.1428 + break;
1.1429 + }
1.1430 +
1.1431 + default:
1.1432 + assert(0);
1.1433 + }
1.1434 + }
1.1435 +
1.1436 + return true;
1.1437 +}
1.1438 +
1.1439 +bool CallFrameInfo::State::DoInstruction() {
1.1440 + CIE *cie = entry_->cie;
1.1441 + Operands ops;
1.1442 +
1.1443 + // Our entry's kind should have been set by now.
1.1444 + assert(entry_->kind != kUnknown);
1.1445 +
1.1446 + // We shouldn't have been invoked unless there were more
1.1447 + // instructions to parse.
1.1448 + assert(cursor_ < entry_->end);
1.1449 +
1.1450 + unsigned opcode = *cursor_++;
1.1451 + if ((opcode & 0xc0) != 0) {
1.1452 + switch (opcode & 0xc0) {
1.1453 + // Advance the address.
1.1454 + case DW_CFA_advance_loc: {
1.1455 + size_t code_offset = opcode & 0x3f;
1.1456 + address_ += code_offset * cie->code_alignment_factor;
1.1457 + break;
1.1458 + }
1.1459 +
1.1460 + // Find a register at an offset from the CFA.
1.1461 + case DW_CFA_offset:
1.1462 + if (!ParseOperands("o", &ops) ||
1.1463 + !DoOffset(opcode & 0x3f, ops.offset * cie->data_alignment_factor))
1.1464 + return false;
1.1465 + break;
1.1466 +
1.1467 + // Restore the rule established for a register by the CIE.
1.1468 + case DW_CFA_restore:
1.1469 + if (!DoRestore(opcode & 0x3f)) return false;
1.1470 + break;
1.1471 +
1.1472 + // The 'if' above should have excluded this possibility.
1.1473 + default:
1.1474 + assert(0);
1.1475 + }
1.1476 +
1.1477 + // Return here, so the big switch below won't be indented.
1.1478 + return true;
1.1479 + }
1.1480 +
1.1481 + switch (opcode) {
1.1482 + // Set the address.
1.1483 + case DW_CFA_set_loc:
1.1484 + if (!ParseOperands("a", &ops)) return false;
1.1485 + address_ = ops.offset;
1.1486 + break;
1.1487 +
1.1488 + // Advance the address.
1.1489 + case DW_CFA_advance_loc1:
1.1490 + if (!ParseOperands("1", &ops)) return false;
1.1491 + address_ += ops.offset * cie->code_alignment_factor;
1.1492 + break;
1.1493 +
1.1494 + // Advance the address.
1.1495 + case DW_CFA_advance_loc2:
1.1496 + if (!ParseOperands("2", &ops)) return false;
1.1497 + address_ += ops.offset * cie->code_alignment_factor;
1.1498 + break;
1.1499 +
1.1500 + // Advance the address.
1.1501 + case DW_CFA_advance_loc4:
1.1502 + if (!ParseOperands("4", &ops)) return false;
1.1503 + address_ += ops.offset * cie->code_alignment_factor;
1.1504 + break;
1.1505 +
1.1506 + // Advance the address.
1.1507 + case DW_CFA_MIPS_advance_loc8:
1.1508 + if (!ParseOperands("8", &ops)) return false;
1.1509 + address_ += ops.offset * cie->code_alignment_factor;
1.1510 + break;
1.1511 +
1.1512 + // Compute the CFA by adding an offset to a register.
1.1513 + case DW_CFA_def_cfa:
1.1514 + if (!ParseOperands("ro", &ops) ||
1.1515 + !DoDefCFA(ops.register_number, ops.offset))
1.1516 + return false;
1.1517 + break;
1.1518 +
1.1519 + // Compute the CFA by adding an offset to a register.
1.1520 + case DW_CFA_def_cfa_sf:
1.1521 + if (!ParseOperands("rs", &ops) ||
1.1522 + !DoDefCFA(ops.register_number,
1.1523 + ops.signed_offset * cie->data_alignment_factor))
1.1524 + return false;
1.1525 + break;
1.1526 +
1.1527 + // Change the base register used to compute the CFA.
1.1528 + case DW_CFA_def_cfa_register: {
1.1529 + Rule *cfa_rule = rules_.CFARule();
1.1530 + if (!cfa_rule) {
1.1531 + reporter_->NoCFARule(entry_->offset, entry_->kind, CursorOffset());
1.1532 + return false;
1.1533 + }
1.1534 + if (!ParseOperands("r", &ops)) return false;
1.1535 + cfa_rule->SetBaseRegister(ops.register_number);
1.1536 + if (!cfa_rule->Handle(handler_, address_,
1.1537 + Handler::kCFARegister))
1.1538 + return false;
1.1539 + break;
1.1540 + }
1.1541 +
1.1542 + // Change the offset used to compute the CFA.
1.1543 + case DW_CFA_def_cfa_offset:
1.1544 + if (!ParseOperands("o", &ops) ||
1.1545 + !DoDefCFAOffset(ops.offset))
1.1546 + return false;
1.1547 + break;
1.1548 +
1.1549 + // Change the offset used to compute the CFA.
1.1550 + case DW_CFA_def_cfa_offset_sf:
1.1551 + if (!ParseOperands("s", &ops) ||
1.1552 + !DoDefCFAOffset(ops.signed_offset * cie->data_alignment_factor))
1.1553 + return false;
1.1554 + break;
1.1555 +
1.1556 + // Specify an expression whose value is the CFA.
1.1557 + case DW_CFA_def_cfa_expression: {
1.1558 + if (!ParseOperands("e", &ops))
1.1559 + return false;
1.1560 + Rule *rule = new ValExpressionRule(ops.expression);
1.1561 + rules_.SetCFARule(rule);
1.1562 + if (!rule->Handle(handler_, address_,
1.1563 + Handler::kCFARegister))
1.1564 + return false;
1.1565 + break;
1.1566 + }
1.1567 +
1.1568 + // The register's value cannot be recovered.
1.1569 + case DW_CFA_undefined: {
1.1570 + if (!ParseOperands("r", &ops) ||
1.1571 + !DoRule(ops.register_number, new UndefinedRule()))
1.1572 + return false;
1.1573 + break;
1.1574 + }
1.1575 +
1.1576 + // The register's value is unchanged from its value in the caller.
1.1577 + case DW_CFA_same_value: {
1.1578 + if (!ParseOperands("r", &ops) ||
1.1579 + !DoRule(ops.register_number, new SameValueRule()))
1.1580 + return false;
1.1581 + break;
1.1582 + }
1.1583 +
1.1584 + // Find a register at an offset from the CFA.
1.1585 + case DW_CFA_offset_extended:
1.1586 + if (!ParseOperands("ro", &ops) ||
1.1587 + !DoOffset(ops.register_number,
1.1588 + ops.offset * cie->data_alignment_factor))
1.1589 + return false;
1.1590 + break;
1.1591 +
1.1592 + // The register is saved at an offset from the CFA.
1.1593 + case DW_CFA_offset_extended_sf:
1.1594 + if (!ParseOperands("rs", &ops) ||
1.1595 + !DoOffset(ops.register_number,
1.1596 + ops.signed_offset * cie->data_alignment_factor))
1.1597 + return false;
1.1598 + break;
1.1599 +
1.1600 + // The register is saved at an offset from the CFA.
1.1601 + case DW_CFA_GNU_negative_offset_extended:
1.1602 + if (!ParseOperands("ro", &ops) ||
1.1603 + !DoOffset(ops.register_number,
1.1604 + -ops.offset * cie->data_alignment_factor))
1.1605 + return false;
1.1606 + break;
1.1607 +
1.1608 + // The register's value is the sum of the CFA plus an offset.
1.1609 + case DW_CFA_val_offset:
1.1610 + if (!ParseOperands("ro", &ops) ||
1.1611 + !DoValOffset(ops.register_number,
1.1612 + ops.offset * cie->data_alignment_factor))
1.1613 + return false;
1.1614 + break;
1.1615 +
1.1616 + // The register's value is the sum of the CFA plus an offset.
1.1617 + case DW_CFA_val_offset_sf:
1.1618 + if (!ParseOperands("rs", &ops) ||
1.1619 + !DoValOffset(ops.register_number,
1.1620 + ops.signed_offset * cie->data_alignment_factor))
1.1621 + return false;
1.1622 + break;
1.1623 +
1.1624 + // The register has been saved in another register.
1.1625 + case DW_CFA_register: {
1.1626 + if (!ParseOperands("ro", &ops) ||
1.1627 + !DoRule(ops.register_number, new RegisterRule(ops.offset)))
1.1628 + return false;
1.1629 + break;
1.1630 + }
1.1631 +
1.1632 + // An expression yields the address at which the register is saved.
1.1633 + case DW_CFA_expression: {
1.1634 + if (!ParseOperands("re", &ops) ||
1.1635 + !DoRule(ops.register_number, new ExpressionRule(ops.expression)))
1.1636 + return false;
1.1637 + break;
1.1638 + }
1.1639 +
1.1640 + // An expression yields the caller's value for the register.
1.1641 + case DW_CFA_val_expression: {
1.1642 + if (!ParseOperands("re", &ops) ||
1.1643 + !DoRule(ops.register_number, new ValExpressionRule(ops.expression)))
1.1644 + return false;
1.1645 + break;
1.1646 + }
1.1647 +
1.1648 + // Restore the rule established for a register by the CIE.
1.1649 + case DW_CFA_restore_extended:
1.1650 + if (!ParseOperands("r", &ops) ||
1.1651 + !DoRestore( ops.register_number))
1.1652 + return false;
1.1653 + break;
1.1654 +
1.1655 + // Save the current set of rules on a stack.
1.1656 + case DW_CFA_remember_state:
1.1657 + saved_rules_.push(rules_);
1.1658 + break;
1.1659 +
1.1660 + // Pop the current set of rules off the stack.
1.1661 + case DW_CFA_restore_state: {
1.1662 + if (saved_rules_.empty()) {
1.1663 + reporter_->EmptyStateStack(entry_->offset, entry_->kind,
1.1664 + CursorOffset());
1.1665 + return false;
1.1666 + }
1.1667 + const RuleMap &new_rules = saved_rules_.top();
1.1668 + if (rules_.CFARule() && !new_rules.CFARule()) {
1.1669 + reporter_->ClearingCFARule(entry_->offset, entry_->kind,
1.1670 + CursorOffset());
1.1671 + return false;
1.1672 + }
1.1673 + rules_.HandleTransitionTo(handler_, address_, new_rules);
1.1674 + rules_ = new_rules;
1.1675 + saved_rules_.pop();
1.1676 + break;
1.1677 + }
1.1678 +
1.1679 + // No operation. (Padding instruction.)
1.1680 + case DW_CFA_nop:
1.1681 + break;
1.1682 +
1.1683 + // A SPARC register window save: Registers 8 through 15 (%o0-%o7)
1.1684 + // are saved in registers 24 through 31 (%i0-%i7), and registers
1.1685 + // 16 through 31 (%l0-%l7 and %i0-%i7) are saved at CFA offsets
1.1686 + // (0-15 * the register size). The register numbers must be
1.1687 + // hard-coded. A GNU extension, and not a pretty one.
1.1688 + case DW_CFA_GNU_window_save: {
1.1689 + // Save %o0-%o7 in %i0-%i7.
1.1690 + for (int i = 8; i < 16; i++)
1.1691 + if (!DoRule(i, new RegisterRule(i + 16)))
1.1692 + return false;
1.1693 + // Save %l0-%l7 and %i0-%i7 at the CFA.
1.1694 + for (int i = 16; i < 32; i++)
1.1695 + // Assume that the byte reader's address size is the same as
1.1696 + // the architecture's register size. !@#%*^ hilarious.
1.1697 + if (!DoRule(i, new OffsetRule(Handler::kCFARegister,
1.1698 + (i - 16) * reader_->AddressSize())))
1.1699 + return false;
1.1700 + break;
1.1701 + }
1.1702 +
1.1703 + // I'm not sure what this is. GDB doesn't use it for unwinding.
1.1704 + case DW_CFA_GNU_args_size:
1.1705 + if (!ParseOperands("o", &ops)) return false;
1.1706 + break;
1.1707 +
1.1708 + // An opcode we don't recognize.
1.1709 + default: {
1.1710 + reporter_->BadInstruction(entry_->offset, entry_->kind, CursorOffset());
1.1711 + return false;
1.1712 + }
1.1713 + }
1.1714 +
1.1715 + return true;
1.1716 +}
1.1717 +
1.1718 +bool CallFrameInfo::State::DoDefCFA(unsigned base_register, long offset) {
1.1719 + Rule *rule = new ValOffsetRule(base_register, offset);
1.1720 + rules_.SetCFARule(rule);
1.1721 + return rule->Handle(handler_, address_,
1.1722 + Handler::kCFARegister);
1.1723 +}
1.1724 +
1.1725 +bool CallFrameInfo::State::DoDefCFAOffset(long offset) {
1.1726 + Rule *cfa_rule = rules_.CFARule();
1.1727 + if (!cfa_rule) {
1.1728 + reporter_->NoCFARule(entry_->offset, entry_->kind, CursorOffset());
1.1729 + return false;
1.1730 + }
1.1731 + cfa_rule->SetOffset(offset);
1.1732 + return cfa_rule->Handle(handler_, address_,
1.1733 + Handler::kCFARegister);
1.1734 +}
1.1735 +
1.1736 +bool CallFrameInfo::State::DoRule(unsigned reg, Rule *rule) {
1.1737 + rules_.SetRegisterRule(reg, rule);
1.1738 + return rule->Handle(handler_, address_, reg);
1.1739 +}
1.1740 +
1.1741 +bool CallFrameInfo::State::DoOffset(unsigned reg, long offset) {
1.1742 + if (!rules_.CFARule()) {
1.1743 + reporter_->NoCFARule(entry_->offset, entry_->kind, CursorOffset());
1.1744 + return false;
1.1745 + }
1.1746 + return DoRule(reg,
1.1747 + new OffsetRule(Handler::kCFARegister, offset));
1.1748 +}
1.1749 +
1.1750 +bool CallFrameInfo::State::DoValOffset(unsigned reg, long offset) {
1.1751 + if (!rules_.CFARule()) {
1.1752 + reporter_->NoCFARule(entry_->offset, entry_->kind, CursorOffset());
1.1753 + return false;
1.1754 + }
1.1755 + return DoRule(reg,
1.1756 + new ValOffsetRule(Handler::kCFARegister, offset));
1.1757 +}
1.1758 +
1.1759 +bool CallFrameInfo::State::DoRestore(unsigned reg) {
1.1760 + // DW_CFA_restore and DW_CFA_restore_extended don't make sense in a CIE.
1.1761 + if (entry_->kind == kCIE) {
1.1762 + reporter_->RestoreInCIE(entry_->offset, CursorOffset());
1.1763 + return false;
1.1764 + }
1.1765 + Rule *rule = cie_rules_.RegisterRule(reg);
1.1766 + if (!rule) {
1.1767 + // This isn't really the right thing to do, but since CFI generally
1.1768 + // only mentions callee-saves registers, and GCC's convention for
1.1769 + // callee-saves registers is that they are unchanged, it's a good
1.1770 + // approximation.
1.1771 + rule = new SameValueRule();
1.1772 + }
1.1773 + return DoRule(reg, rule);
1.1774 +}
1.1775 +
1.1776 +bool CallFrameInfo::ReadEntryPrologue(const char *cursor, Entry *entry) {
1.1777 + const char *buffer_end = buffer_ + buffer_length_;
1.1778 +
1.1779 + // Initialize enough of ENTRY for use in error reporting.
1.1780 + entry->offset = cursor - buffer_;
1.1781 + entry->start = cursor;
1.1782 + entry->kind = kUnknown;
1.1783 + entry->end = NULL;
1.1784 +
1.1785 + // Read the initial length. This sets reader_'s offset size.
1.1786 + size_t length_size;
1.1787 + uint64 length = reader_->ReadInitialLength(cursor, &length_size);
1.1788 + if (length_size > size_t(buffer_end - cursor))
1.1789 + return ReportIncomplete(entry);
1.1790 + cursor += length_size;
1.1791 +
1.1792 + // In a .eh_frame section, a length of zero marks the end of the series
1.1793 + // of entries.
1.1794 + if (length == 0 && eh_frame_) {
1.1795 + entry->kind = kTerminator;
1.1796 + entry->end = cursor;
1.1797 + return true;
1.1798 + }
1.1799 +
1.1800 + // Validate the length.
1.1801 + if (length > size_t(buffer_end - cursor))
1.1802 + return ReportIncomplete(entry);
1.1803 +
1.1804 + // The length is the number of bytes after the initial length field;
1.1805 + // we have that position handy at this point, so compute the end
1.1806 + // now. (If we're parsing 64-bit-offset DWARF on a 32-bit machine,
1.1807 + // and the length didn't fit in a size_t, we would have rejected it
1.1808 + // above.)
1.1809 + entry->end = cursor + length;
1.1810 +
1.1811 + // Parse the next field: either the offset of a CIE or a CIE id.
1.1812 + size_t offset_size = reader_->OffsetSize();
1.1813 + if (offset_size > size_t(entry->end - cursor)) return ReportIncomplete(entry);
1.1814 + entry->id = reader_->ReadOffset(cursor);
1.1815 +
1.1816 + // Don't advance cursor past id field yet; in .eh_frame data we need
1.1817 + // the id's position to compute the section offset of an FDE's CIE.
1.1818 +
1.1819 + // Now we can decide what kind of entry this is.
1.1820 + if (eh_frame_) {
1.1821 + // In .eh_frame data, an ID of zero marks the entry as a CIE, and
1.1822 + // anything else is an offset from the id field of the FDE to the start
1.1823 + // of the CIE.
1.1824 + if (entry->id == 0) {
1.1825 + entry->kind = kCIE;
1.1826 + } else {
1.1827 + entry->kind = kFDE;
1.1828 + // Turn the offset from the id into an offset from the buffer's start.
1.1829 + entry->id = (cursor - buffer_) - entry->id;
1.1830 + }
1.1831 + } else {
1.1832 + // In DWARF CFI data, an ID of ~0 (of the appropriate width, given the
1.1833 + // offset size for the entry) marks the entry as a CIE, and anything
1.1834 + // else is the offset of the CIE from the beginning of the section.
1.1835 + if (offset_size == 4)
1.1836 + entry->kind = (entry->id == 0xffffffff) ? kCIE : kFDE;
1.1837 + else {
1.1838 + assert(offset_size == 8);
1.1839 + entry->kind = (entry->id == 0xffffffffffffffffULL) ? kCIE : kFDE;
1.1840 + }
1.1841 + }
1.1842 +
1.1843 + // Now advance cursor past the id.
1.1844 + cursor += offset_size;
1.1845 +
1.1846 + // The fields specific to this kind of entry start here.
1.1847 + entry->fields = cursor;
1.1848 +
1.1849 + entry->cie = NULL;
1.1850 +
1.1851 + return true;
1.1852 +}
1.1853 +
1.1854 +bool CallFrameInfo::ReadCIEFields(CIE *cie) {
1.1855 + const char *cursor = cie->fields;
1.1856 + size_t len;
1.1857 +
1.1858 + assert(cie->kind == kCIE);
1.1859 +
1.1860 + // Prepare for early exit.
1.1861 + cie->version = 0;
1.1862 + cie->augmentation.clear();
1.1863 + cie->code_alignment_factor = 0;
1.1864 + cie->data_alignment_factor = 0;
1.1865 + cie->return_address_register = 0;
1.1866 + cie->has_z_augmentation = false;
1.1867 + cie->pointer_encoding = DW_EH_PE_absptr;
1.1868 + cie->instructions = 0;
1.1869 +
1.1870 + // Parse the version number.
1.1871 + if (cie->end - cursor < 1)
1.1872 + return ReportIncomplete(cie);
1.1873 + cie->version = reader_->ReadOneByte(cursor);
1.1874 + cursor++;
1.1875 +
1.1876 + // If we don't recognize the version, we can't parse any more fields of the
1.1877 + // CIE. For DWARF CFI, we handle versions 1 through 3 (there was never a
1.1878 + // version 2 of CFI data). For .eh_frame, we handle versions 1 and 3 as well;
1.1879 + // the difference between those versions seems to be the same as for
1.1880 + // .debug_frame.
1.1881 + if (cie->version < 1 || cie->version > 3) {
1.1882 + reporter_->UnrecognizedVersion(cie->offset, cie->version);
1.1883 + return false;
1.1884 + }
1.1885 +
1.1886 + const char *augmentation_start = cursor;
1.1887 + const void *augmentation_end =
1.1888 + memchr(augmentation_start, '\0', cie->end - augmentation_start);
1.1889 + if (! augmentation_end) return ReportIncomplete(cie);
1.1890 + cursor = static_cast<const char *>(augmentation_end);
1.1891 + cie->augmentation = string(augmentation_start,
1.1892 + cursor - augmentation_start);
1.1893 + // Skip the terminating '\0'.
1.1894 + cursor++;
1.1895 +
1.1896 + // Is this CFI augmented?
1.1897 + if (!cie->augmentation.empty()) {
1.1898 + // Is it an augmentation we recognize?
1.1899 + if (cie->augmentation[0] == DW_Z_augmentation_start) {
1.1900 + // Linux C++ ABI 'z' augmentation, used for exception handling data.
1.1901 + cie->has_z_augmentation = true;
1.1902 + } else {
1.1903 + // Not an augmentation we recognize. Augmentations can have arbitrary
1.1904 + // effects on the form of rest of the content, so we have to give up.
1.1905 + reporter_->UnrecognizedAugmentation(cie->offset, cie->augmentation);
1.1906 + return false;
1.1907 + }
1.1908 + }
1.1909 +
1.1910 + // Parse the code alignment factor.
1.1911 + cie->code_alignment_factor = reader_->ReadUnsignedLEB128(cursor, &len);
1.1912 + if (size_t(cie->end - cursor) < len) return ReportIncomplete(cie);
1.1913 + cursor += len;
1.1914 +
1.1915 + // Parse the data alignment factor.
1.1916 + cie->data_alignment_factor = reader_->ReadSignedLEB128(cursor, &len);
1.1917 + if (size_t(cie->end - cursor) < len) return ReportIncomplete(cie);
1.1918 + cursor += len;
1.1919 +
1.1920 + // Parse the return address register. This is a ubyte in version 1, and
1.1921 + // a ULEB128 in version 3.
1.1922 + if (cie->version == 1) {
1.1923 + if (cursor >= cie->end) return ReportIncomplete(cie);
1.1924 + cie->return_address_register = uint8(*cursor++);
1.1925 + } else {
1.1926 + cie->return_address_register = reader_->ReadUnsignedLEB128(cursor, &len);
1.1927 + if (size_t(cie->end - cursor) < len) return ReportIncomplete(cie);
1.1928 + cursor += len;
1.1929 + }
1.1930 +
1.1931 + // If we have a 'z' augmentation string, find the augmentation data and
1.1932 + // use the augmentation string to parse it.
1.1933 + if (cie->has_z_augmentation) {
1.1934 + uint64_t data_size = reader_->ReadUnsignedLEB128(cursor, &len);
1.1935 + if (size_t(cie->end - cursor) < len + data_size)
1.1936 + return ReportIncomplete(cie);
1.1937 + cursor += len;
1.1938 + const char *data = cursor;
1.1939 + cursor += data_size;
1.1940 + const char *data_end = cursor;
1.1941 +
1.1942 + cie->has_z_lsda = false;
1.1943 + cie->has_z_personality = false;
1.1944 + cie->has_z_signal_frame = false;
1.1945 +
1.1946 + // Walk the augmentation string, and extract values from the
1.1947 + // augmentation data as the string directs.
1.1948 + for (size_t i = 1; i < cie->augmentation.size(); i++) {
1.1949 + switch (cie->augmentation[i]) {
1.1950 + case DW_Z_has_LSDA:
1.1951 + // The CIE's augmentation data holds the language-specific data
1.1952 + // area pointer's encoding, and the FDE's augmentation data holds
1.1953 + // the pointer itself.
1.1954 + cie->has_z_lsda = true;
1.1955 + // Fetch the LSDA encoding from the augmentation data.
1.1956 + if (data >= data_end) return ReportIncomplete(cie);
1.1957 + cie->lsda_encoding = DwarfPointerEncoding(*data++);
1.1958 + if (!reader_->ValidEncoding(cie->lsda_encoding)) {
1.1959 + reporter_->InvalidPointerEncoding(cie->offset, cie->lsda_encoding);
1.1960 + return false;
1.1961 + }
1.1962 + // Don't check if the encoding is usable here --- we haven't
1.1963 + // read the FDE's fields yet, so we're not prepared for
1.1964 + // DW_EH_PE_funcrel, although that's a fine encoding for the
1.1965 + // LSDA to use, since it appears in the FDE.
1.1966 + break;
1.1967 +
1.1968 + case DW_Z_has_personality_routine:
1.1969 + // The CIE's augmentation data holds the personality routine
1.1970 + // pointer's encoding, followed by the pointer itself.
1.1971 + cie->has_z_personality = true;
1.1972 + // Fetch the personality routine pointer's encoding from the
1.1973 + // augmentation data.
1.1974 + if (data >= data_end) return ReportIncomplete(cie);
1.1975 + cie->personality_encoding = DwarfPointerEncoding(*data++);
1.1976 + if (!reader_->ValidEncoding(cie->personality_encoding)) {
1.1977 + reporter_->InvalidPointerEncoding(cie->offset,
1.1978 + cie->personality_encoding);
1.1979 + return false;
1.1980 + }
1.1981 + if (!reader_->UsableEncoding(cie->personality_encoding)) {
1.1982 + reporter_->UnusablePointerEncoding(cie->offset,
1.1983 + cie->personality_encoding);
1.1984 + return false;
1.1985 + }
1.1986 + // Fetch the personality routine's pointer itself from the data.
1.1987 + cie->personality_address =
1.1988 + reader_->ReadEncodedPointer(data, cie->personality_encoding,
1.1989 + &len);
1.1990 + if (len > size_t(data_end - data))
1.1991 + return ReportIncomplete(cie);
1.1992 + data += len;
1.1993 + break;
1.1994 +
1.1995 + case DW_Z_has_FDE_address_encoding:
1.1996 + // The CIE's augmentation data holds the pointer encoding to use
1.1997 + // for addresses in the FDE.
1.1998 + if (data >= data_end) return ReportIncomplete(cie);
1.1999 + cie->pointer_encoding = DwarfPointerEncoding(*data++);
1.2000 + if (!reader_->ValidEncoding(cie->pointer_encoding)) {
1.2001 + reporter_->InvalidPointerEncoding(cie->offset,
1.2002 + cie->pointer_encoding);
1.2003 + return false;
1.2004 + }
1.2005 + if (!reader_->UsableEncoding(cie->pointer_encoding)) {
1.2006 + reporter_->UnusablePointerEncoding(cie->offset,
1.2007 + cie->pointer_encoding);
1.2008 + return false;
1.2009 + }
1.2010 + break;
1.2011 +
1.2012 + case DW_Z_is_signal_trampoline:
1.2013 + // Frames using this CIE are signal delivery frames.
1.2014 + cie->has_z_signal_frame = true;
1.2015 + break;
1.2016 +
1.2017 + default:
1.2018 + // An augmentation we don't recognize.
1.2019 + reporter_->UnrecognizedAugmentation(cie->offset, cie->augmentation);
1.2020 + return false;
1.2021 + }
1.2022 + }
1.2023 + }
1.2024 +
1.2025 + // The CIE's instructions start here.
1.2026 + cie->instructions = cursor;
1.2027 +
1.2028 + return true;
1.2029 +}
1.2030 +
1.2031 +bool CallFrameInfo::ReadFDEFields(FDE *fde) {
1.2032 + const char *cursor = fde->fields;
1.2033 + size_t size;
1.2034 +
1.2035 + fde->address = reader_->ReadEncodedPointer(cursor, fde->cie->pointer_encoding,
1.2036 + &size);
1.2037 + if (size > size_t(fde->end - cursor))
1.2038 + return ReportIncomplete(fde);
1.2039 + cursor += size;
1.2040 + reader_->SetFunctionBase(fde->address);
1.2041 +
1.2042 + // For the length, we strip off the upper nybble of the encoding used for
1.2043 + // the starting address.
1.2044 + DwarfPointerEncoding length_encoding =
1.2045 + DwarfPointerEncoding(fde->cie->pointer_encoding & 0x0f);
1.2046 + fde->size = reader_->ReadEncodedPointer(cursor, length_encoding, &size);
1.2047 + if (size > size_t(fde->end - cursor))
1.2048 + return ReportIncomplete(fde);
1.2049 + cursor += size;
1.2050 +
1.2051 + // If the CIE has a 'z' augmentation string, then augmentation data
1.2052 + // appears here.
1.2053 + if (fde->cie->has_z_augmentation) {
1.2054 + uint64_t data_size = reader_->ReadUnsignedLEB128(cursor, &size);
1.2055 + if (size_t(fde->end - cursor) < size + data_size)
1.2056 + return ReportIncomplete(fde);
1.2057 + cursor += size;
1.2058 +
1.2059 + // In the abstract, we should walk the augmentation string, and extract
1.2060 + // items from the FDE's augmentation data as we encounter augmentation
1.2061 + // string characters that specify their presence: the ordering of items
1.2062 + // in the augmentation string determines the arrangement of values in
1.2063 + // the augmentation data.
1.2064 + //
1.2065 + // In practice, there's only ever one value in FDE augmentation data
1.2066 + // that we support --- the LSDA pointer --- and we have to bail if we
1.2067 + // see any unrecognized augmentation string characters. So if there is
1.2068 + // anything here at all, we know what it is, and where it starts.
1.2069 + if (fde->cie->has_z_lsda) {
1.2070 + // Check whether the LSDA's pointer encoding is usable now: only once
1.2071 + // we've parsed the FDE's starting address do we call reader_->
1.2072 + // SetFunctionBase, so that the DW_EH_PE_funcrel encoding becomes
1.2073 + // usable.
1.2074 + if (!reader_->UsableEncoding(fde->cie->lsda_encoding)) {
1.2075 + reporter_->UnusablePointerEncoding(fde->cie->offset,
1.2076 + fde->cie->lsda_encoding);
1.2077 + return false;
1.2078 + }
1.2079 +
1.2080 + fde->lsda_address =
1.2081 + reader_->ReadEncodedPointer(cursor, fde->cie->lsda_encoding, &size);
1.2082 + if (size > data_size)
1.2083 + return ReportIncomplete(fde);
1.2084 + // Ideally, we would also complain here if there were unconsumed
1.2085 + // augmentation data.
1.2086 + }
1.2087 +
1.2088 + cursor += data_size;
1.2089 + }
1.2090 +
1.2091 + // The FDE's instructions start after those.
1.2092 + fde->instructions = cursor;
1.2093 +
1.2094 + return true;
1.2095 +}
1.2096 +
1.2097 +bool CallFrameInfo::Start() {
1.2098 + const char *buffer_end = buffer_ + buffer_length_;
1.2099 + const char *cursor;
1.2100 + bool all_ok = true;
1.2101 + const char *entry_end;
1.2102 + bool ok;
1.2103 +
1.2104 + // Traverse all the entries in buffer_, skipping CIEs and offering
1.2105 + // FDEs to the handler.
1.2106 + for (cursor = buffer_; cursor < buffer_end;
1.2107 + cursor = entry_end, all_ok = all_ok && ok) {
1.2108 + FDE fde;
1.2109 +
1.2110 + // Make it easy to skip this entry with 'continue': assume that
1.2111 + // things are not okay until we've checked all the data, and
1.2112 + // prepare the address of the next entry.
1.2113 + ok = false;
1.2114 +
1.2115 + // Read the entry's prologue.
1.2116 + if (!ReadEntryPrologue(cursor, &fde)) {
1.2117 + if (!fde.end) {
1.2118 + // If we couldn't even figure out this entry's extent, then we
1.2119 + // must stop processing entries altogether.
1.2120 + all_ok = false;
1.2121 + break;
1.2122 + }
1.2123 + entry_end = fde.end;
1.2124 + continue;
1.2125 + }
1.2126 +
1.2127 + // The next iteration picks up after this entry.
1.2128 + entry_end = fde.end;
1.2129 +
1.2130 + // Did we see an .eh_frame terminating mark?
1.2131 + if (fde.kind == kTerminator) {
1.2132 + // If there appears to be more data left in the section after the
1.2133 + // terminating mark, warn the user. But this is just a warning;
1.2134 + // we leave all_ok true.
1.2135 + if (fde.end < buffer_end) reporter_->EarlyEHTerminator(fde.offset);
1.2136 + break;
1.2137 + }
1.2138 +
1.2139 + // In this loop, we skip CIEs. We only parse them fully when we
1.2140 + // parse an FDE that refers to them. This limits our memory
1.2141 + // consumption (beyond the buffer itself) to that needed to
1.2142 + // process the largest single entry.
1.2143 + if (fde.kind != kFDE) {
1.2144 + ok = true;
1.2145 + continue;
1.2146 + }
1.2147 +
1.2148 + // Validate the CIE pointer.
1.2149 + if (fde.id > buffer_length_) {
1.2150 + reporter_->CIEPointerOutOfRange(fde.offset, fde.id);
1.2151 + continue;
1.2152 + }
1.2153 +
1.2154 + CIE cie;
1.2155 +
1.2156 + // Parse this FDE's CIE header.
1.2157 + if (!ReadEntryPrologue(buffer_ + fde.id, &cie))
1.2158 + continue;
1.2159 + // This had better be an actual CIE.
1.2160 + if (cie.kind != kCIE) {
1.2161 + reporter_->BadCIEId(fde.offset, fde.id);
1.2162 + continue;
1.2163 + }
1.2164 + if (!ReadCIEFields(&cie))
1.2165 + continue;
1.2166 +
1.2167 + // We now have the values that govern both the CIE and the FDE.
1.2168 + cie.cie = &cie;
1.2169 + fde.cie = &cie;
1.2170 +
1.2171 + // Parse the FDE's header.
1.2172 + if (!ReadFDEFields(&fde))
1.2173 + continue;
1.2174 +
1.2175 + // Call Entry to ask the consumer if they're interested.
1.2176 + if (!handler_->Entry(fde.offset, fde.address, fde.size,
1.2177 + cie.version, cie.augmentation,
1.2178 + cie.return_address_register)) {
1.2179 + // The handler isn't interested in this entry. That's not an error.
1.2180 + ok = true;
1.2181 + continue;
1.2182 + }
1.2183 +
1.2184 + if (cie.has_z_augmentation) {
1.2185 + // Report the personality routine address, if we have one.
1.2186 + if (cie.has_z_personality) {
1.2187 + if (!handler_
1.2188 + ->PersonalityRoutine(cie.personality_address,
1.2189 + IsIndirectEncoding(cie.personality_encoding)))
1.2190 + continue;
1.2191 + }
1.2192 +
1.2193 + // Report the language-specific data area address, if we have one.
1.2194 + if (cie.has_z_lsda) {
1.2195 + if (!handler_
1.2196 + ->LanguageSpecificDataArea(fde.lsda_address,
1.2197 + IsIndirectEncoding(cie.lsda_encoding)))
1.2198 + continue;
1.2199 + }
1.2200 +
1.2201 + // If this is a signal-handling frame, report that.
1.2202 + if (cie.has_z_signal_frame) {
1.2203 + if (!handler_->SignalHandler())
1.2204 + continue;
1.2205 + }
1.2206 + }
1.2207 +
1.2208 + // Interpret the CIE's instructions, and then the FDE's instructions.
1.2209 + State state(reader_, handler_, reporter_, fde.address);
1.2210 + ok = state.InterpretCIE(cie) && state.InterpretFDE(fde);
1.2211 +
1.2212 + // Tell the ByteReader that the function start address from the
1.2213 + // FDE header is no longer valid.
1.2214 + reader_->ClearFunctionBase();
1.2215 +
1.2216 + // Report the end of the entry.
1.2217 + handler_->End();
1.2218 + }
1.2219 +
1.2220 + return all_ok;
1.2221 +}
1.2222 +
1.2223 +const char *CallFrameInfo::KindName(EntryKind kind) {
1.2224 + if (kind == CallFrameInfo::kUnknown)
1.2225 + return "entry";
1.2226 + else if (kind == CallFrameInfo::kCIE)
1.2227 + return "common information entry";
1.2228 + else if (kind == CallFrameInfo::kFDE)
1.2229 + return "frame description entry";
1.2230 + else {
1.2231 + assert (kind == CallFrameInfo::kTerminator);
1.2232 + return ".eh_frame sequence terminator";
1.2233 + }
1.2234 +}
1.2235 +
1.2236 +bool CallFrameInfo::ReportIncomplete(Entry *entry) {
1.2237 + reporter_->Incomplete(entry->offset, entry->kind);
1.2238 + return false;
1.2239 +}
1.2240 +
1.2241 +void CallFrameInfo::Reporter::Incomplete(uint64 offset,
1.2242 + CallFrameInfo::EntryKind kind) {
1.2243 + fprintf(stderr,
1.2244 + "%s: CFI %s at offset 0x%llx in '%s': entry ends early\n",
1.2245 + filename_.c_str(), CallFrameInfo::KindName(kind), offset,
1.2246 + section_.c_str());
1.2247 +}
1.2248 +
1.2249 +void CallFrameInfo::Reporter::EarlyEHTerminator(uint64 offset) {
1.2250 + fprintf(stderr,
1.2251 + "%s: CFI at offset 0x%llx in '%s': saw end-of-data marker"
1.2252 + " before end of section contents\n",
1.2253 + filename_.c_str(), offset, section_.c_str());
1.2254 +}
1.2255 +
1.2256 +void CallFrameInfo::Reporter::CIEPointerOutOfRange(uint64 offset,
1.2257 + uint64 cie_offset) {
1.2258 + fprintf(stderr,
1.2259 + "%s: CFI frame description entry at offset 0x%llx in '%s':"
1.2260 + " CIE pointer is out of range: 0x%llx\n",
1.2261 + filename_.c_str(), offset, section_.c_str(), cie_offset);
1.2262 +}
1.2263 +
1.2264 +void CallFrameInfo::Reporter::BadCIEId(uint64 offset, uint64 cie_offset) {
1.2265 + fprintf(stderr,
1.2266 + "%s: CFI frame description entry at offset 0x%llx in '%s':"
1.2267 + " CIE pointer does not point to a CIE: 0x%llx\n",
1.2268 + filename_.c_str(), offset, section_.c_str(), cie_offset);
1.2269 +}
1.2270 +
1.2271 +void CallFrameInfo::Reporter::UnrecognizedVersion(uint64 offset, int version) {
1.2272 + fprintf(stderr,
1.2273 + "%s: CFI frame description entry at offset 0x%llx in '%s':"
1.2274 + " CIE specifies unrecognized version: %d\n",
1.2275 + filename_.c_str(), offset, section_.c_str(), version);
1.2276 +}
1.2277 +
1.2278 +void CallFrameInfo::Reporter::UnrecognizedAugmentation(uint64 offset,
1.2279 + const string &aug) {
1.2280 + fprintf(stderr,
1.2281 + "%s: CFI frame description entry at offset 0x%llx in '%s':"
1.2282 + " CIE specifies unrecognized augmentation: '%s'\n",
1.2283 + filename_.c_str(), offset, section_.c_str(), aug.c_str());
1.2284 +}
1.2285 +
1.2286 +void CallFrameInfo::Reporter::InvalidPointerEncoding(uint64 offset,
1.2287 + uint8 encoding) {
1.2288 + fprintf(stderr,
1.2289 + "%s: CFI common information entry at offset 0x%llx in '%s':"
1.2290 + " 'z' augmentation specifies invalid pointer encoding: 0x%02x\n",
1.2291 + filename_.c_str(), offset, section_.c_str(), encoding);
1.2292 +}
1.2293 +
1.2294 +void CallFrameInfo::Reporter::UnusablePointerEncoding(uint64 offset,
1.2295 + uint8 encoding) {
1.2296 + fprintf(stderr,
1.2297 + "%s: CFI common information entry at offset 0x%llx in '%s':"
1.2298 + " 'z' augmentation specifies a pointer encoding for which"
1.2299 + " we have no base address: 0x%02x\n",
1.2300 + filename_.c_str(), offset, section_.c_str(), encoding);
1.2301 +}
1.2302 +
1.2303 +void CallFrameInfo::Reporter::RestoreInCIE(uint64 offset, uint64 insn_offset) {
1.2304 + fprintf(stderr,
1.2305 + "%s: CFI common information entry at offset 0x%llx in '%s':"
1.2306 + " the DW_CFA_restore instruction at offset 0x%llx"
1.2307 + " cannot be used in a common information entry\n",
1.2308 + filename_.c_str(), offset, section_.c_str(), insn_offset);
1.2309 +}
1.2310 +
1.2311 +void CallFrameInfo::Reporter::BadInstruction(uint64 offset,
1.2312 + CallFrameInfo::EntryKind kind,
1.2313 + uint64 insn_offset) {
1.2314 + fprintf(stderr,
1.2315 + "%s: CFI %s at offset 0x%llx in section '%s':"
1.2316 + " the instruction at offset 0x%llx is unrecognized\n",
1.2317 + filename_.c_str(), CallFrameInfo::KindName(kind),
1.2318 + offset, section_.c_str(), insn_offset);
1.2319 +}
1.2320 +
1.2321 +void CallFrameInfo::Reporter::NoCFARule(uint64 offset,
1.2322 + CallFrameInfo::EntryKind kind,
1.2323 + uint64 insn_offset) {
1.2324 + fprintf(stderr,
1.2325 + "%s: CFI %s at offset 0x%llx in section '%s':"
1.2326 + " the instruction at offset 0x%llx assumes that a CFA rule has"
1.2327 + " been set, but none has been set\n",
1.2328 + filename_.c_str(), CallFrameInfo::KindName(kind), offset,
1.2329 + section_.c_str(), insn_offset);
1.2330 +}
1.2331 +
1.2332 +void CallFrameInfo::Reporter::EmptyStateStack(uint64 offset,
1.2333 + CallFrameInfo::EntryKind kind,
1.2334 + uint64 insn_offset) {
1.2335 + fprintf(stderr,
1.2336 + "%s: CFI %s at offset 0x%llx in section '%s':"
1.2337 + " the DW_CFA_restore_state instruction at offset 0x%llx"
1.2338 + " should pop a saved state from the stack, but the stack is empty\n",
1.2339 + filename_.c_str(), CallFrameInfo::KindName(kind), offset,
1.2340 + section_.c_str(), insn_offset);
1.2341 +}
1.2342 +
1.2343 +void CallFrameInfo::Reporter::ClearingCFARule(uint64 offset,
1.2344 + CallFrameInfo::EntryKind kind,
1.2345 + uint64 insn_offset) {
1.2346 + fprintf(stderr,
1.2347 + "%s: CFI %s at offset 0x%llx in section '%s':"
1.2348 + " the DW_CFA_restore_state instruction at offset 0x%llx"
1.2349 + " would clear the CFA rule in effect\n",
1.2350 + filename_.c_str(), CallFrameInfo::KindName(kind), offset,
1.2351 + section_.c_str(), insn_offset);
1.2352 +}
1.2353 +
1.2354 +} // namespace dwarf2reader
