|
1 // -*- mode: C++ -*- |
|
2 |
|
3 // Copyright (c) 2010, Google Inc. |
|
4 // All rights reserved. |
|
5 // |
|
6 // Redistribution and use in source and binary forms, with or without |
|
7 // modification, are permitted provided that the following conditions are |
|
8 // met: |
|
9 // |
|
10 // * Redistributions of source code must retain the above copyright |
|
11 // notice, this list of conditions and the following disclaimer. |
|
12 // * Redistributions in binary form must reproduce the above |
|
13 // copyright notice, this list of conditions and the following disclaimer |
|
14 // in the documentation and/or other materials provided with the |
|
15 // distribution. |
|
16 // * Neither the name of Google Inc. nor the names of its |
|
17 // contributors may be used to endorse or promote products derived from |
|
18 // this software without specific prior written permission. |
|
19 // |
|
20 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
|
21 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
|
22 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
|
23 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
|
24 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
|
25 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
|
26 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
|
27 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
|
28 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
|
29 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
|
30 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
|
31 |
|
32 // Original author: Jim Blandy <jimb@mozilla.com> <jimb@red-bean.com> |
|
33 |
|
34 // cfi_assembler.h: Define CFISection, a class for creating properly |
|
35 // (and improperly) formatted DWARF CFI data for unit tests. |
|
36 |
|
37 #ifndef PROCESSOR_CFI_ASSEMBLER_H_ |
|
38 #define PROCESSOR_CFI_ASSEMBLER_H_ |
|
39 |
|
40 #include <string> |
|
41 |
|
42 #include "common/dwarf/dwarf2enums.h" |
|
43 #include "common/test_assembler.h" |
|
44 #include "common/using_std_string.h" |
|
45 #include "google_breakpad/common/breakpad_types.h" |
|
46 |
|
47 namespace google_breakpad { |
|
48 |
|
49 using dwarf2reader::DwarfPointerEncoding; |
|
50 using google_breakpad::test_assembler::Endianness; |
|
51 using google_breakpad::test_assembler::Label; |
|
52 using google_breakpad::test_assembler::Section; |
|
53 |
|
54 class CFISection: public Section { |
|
55 public: |
|
56 |
|
57 // CFI augmentation strings beginning with 'z', defined by the |
|
58 // Linux/IA-64 C++ ABI, can specify interesting encodings for |
|
59 // addresses appearing in FDE headers and call frame instructions (and |
|
60 // for additional fields whose presence the augmentation string |
|
61 // specifies). In particular, pointers can be specified to be relative |
|
62 // to various base address: the start of the .text section, the |
|
63 // location holding the address itself, and so on. These allow the |
|
64 // frame data to be position-independent even when they live in |
|
65 // write-protected pages. These variants are specified at the |
|
66 // following two URLs: |
|
67 // |
|
68 // http://refspecs.linux-foundation.org/LSB_4.0.0/LSB-Core-generic/LSB-Core-generic/dwarfext.html |
|
69 // http://refspecs.linux-foundation.org/LSB_4.0.0/LSB-Core-generic/LSB-Core-generic/ehframechpt.html |
|
70 // |
|
71 // CFISection leaves the production of well-formed 'z'-augmented CIEs and |
|
72 // FDEs to the user, but does provide EncodedPointer, to emit |
|
73 // properly-encoded addresses for a given pointer encoding. |
|
74 // EncodedPointer uses an instance of this structure to find the base |
|
75 // addresses it should use; you can establish a default for all encoded |
|
76 // pointers appended to this section with SetEncodedPointerBases. |
|
77 struct EncodedPointerBases { |
|
78 EncodedPointerBases() : cfi(), text(), data() { } |
|
79 |
|
80 // The starting address of this CFI section in memory, for |
|
81 // DW_EH_PE_pcrel. DW_EH_PE_pcrel pointers may only be used in data |
|
82 // that has is loaded into the program's address space. |
|
83 uint64_t cfi; |
|
84 |
|
85 // The starting address of this file's .text section, for DW_EH_PE_textrel. |
|
86 uint64_t text; |
|
87 |
|
88 // The starting address of this file's .got or .eh_frame_hdr section, |
|
89 // for DW_EH_PE_datarel. |
|
90 uint64_t data; |
|
91 }; |
|
92 |
|
93 // Create a CFISection whose endianness is ENDIANNESS, and where |
|
94 // machine addresses are ADDRESS_SIZE bytes long. If EH_FRAME is |
|
95 // true, use the .eh_frame format, as described by the Linux |
|
96 // Standards Base Core Specification, instead of the DWARF CFI |
|
97 // format. |
|
98 CFISection(Endianness endianness, size_t address_size, |
|
99 bool eh_frame = false) |
|
100 : Section(endianness), address_size_(address_size), eh_frame_(eh_frame), |
|
101 pointer_encoding_(dwarf2reader::DW_EH_PE_absptr), |
|
102 encoded_pointer_bases_(), entry_length_(NULL), in_fde_(false) { |
|
103 // The 'start', 'Here', and 'Mark' members of a CFISection all refer |
|
104 // to section offsets. |
|
105 start() = 0; |
|
106 } |
|
107 |
|
108 // Return this CFISection's address size. |
|
109 size_t AddressSize() const { return address_size_; } |
|
110 |
|
111 // Return true if this CFISection uses the .eh_frame format, or |
|
112 // false if it contains ordinary DWARF CFI data. |
|
113 bool ContainsEHFrame() const { return eh_frame_; } |
|
114 |
|
115 // Use ENCODING for pointers in calls to FDEHeader and EncodedPointer. |
|
116 void SetPointerEncoding(DwarfPointerEncoding encoding) { |
|
117 pointer_encoding_ = encoding; |
|
118 } |
|
119 |
|
120 // Use the addresses in BASES as the base addresses for encoded |
|
121 // pointers in subsequent calls to FDEHeader or EncodedPointer. |
|
122 // This function makes a copy of BASES. |
|
123 void SetEncodedPointerBases(const EncodedPointerBases &bases) { |
|
124 encoded_pointer_bases_ = bases; |
|
125 } |
|
126 |
|
127 // Append a Common Information Entry header to this section with the |
|
128 // given values. If dwarf64 is true, use the 64-bit DWARF initial |
|
129 // length format for the CIE's initial length. Return a reference to |
|
130 // this section. You should call FinishEntry after writing the last |
|
131 // instruction for the CIE. |
|
132 // |
|
133 // Before calling this function, you will typically want to use Mark |
|
134 // or Here to make a label to pass to FDEHeader that refers to this |
|
135 // CIE's position in the section. |
|
136 CFISection &CIEHeader(uint64_t code_alignment_factor, |
|
137 int data_alignment_factor, |
|
138 unsigned return_address_register, |
|
139 uint8_t version = 3, |
|
140 const string &augmentation = "", |
|
141 bool dwarf64 = false); |
|
142 |
|
143 // Append a Frame Description Entry header to this section with the |
|
144 // given values. If dwarf64 is true, use the 64-bit DWARF initial |
|
145 // length format for the CIE's initial length. Return a reference to |
|
146 // this section. You should call FinishEntry after writing the last |
|
147 // instruction for the CIE. |
|
148 // |
|
149 // This function doesn't support entries that are longer than |
|
150 // 0xffffff00 bytes. (The "initial length" is always a 32-bit |
|
151 // value.) Nor does it support .debug_frame sections longer than |
|
152 // 0xffffff00 bytes. |
|
153 CFISection &FDEHeader(Label cie_pointer, |
|
154 uint64_t initial_location, |
|
155 uint64_t address_range, |
|
156 bool dwarf64 = false); |
|
157 |
|
158 // Note the current position as the end of the last CIE or FDE we |
|
159 // started, after padding with DW_CFA_nops for alignment. This |
|
160 // defines the label representing the entry's length, cited in the |
|
161 // entry's header. Return a reference to this section. |
|
162 CFISection &FinishEntry(); |
|
163 |
|
164 // Append the contents of BLOCK as a DW_FORM_block value: an |
|
165 // unsigned LEB128 length, followed by that many bytes of data. |
|
166 CFISection &Block(const string &block) { |
|
167 ULEB128(block.size()); |
|
168 Append(block); |
|
169 return *this; |
|
170 } |
|
171 |
|
172 // Append ADDRESS to this section, in the appropriate size and |
|
173 // endianness. Return a reference to this section. |
|
174 CFISection &Address(uint64_t address) { |
|
175 Section::Append(endianness(), address_size_, address); |
|
176 return *this; |
|
177 } |
|
178 CFISection &Address(Label address) { |
|
179 Section::Append(endianness(), address_size_, address); |
|
180 return *this; |
|
181 } |
|
182 |
|
183 // Append ADDRESS to this section, using ENCODING and BASES. ENCODING |
|
184 // defaults to this section's default encoding, established by |
|
185 // SetPointerEncoding. BASES defaults to this section's bases, set by |
|
186 // SetEncodedPointerBases. If the DW_EH_PE_indirect bit is set in the |
|
187 // encoding, assume that ADDRESS is where the true address is stored. |
|
188 // Return a reference to this section. |
|
189 // |
|
190 // (C++ doesn't let me use default arguments here, because I want to |
|
191 // refer to members of *this in the default argument expression.) |
|
192 CFISection &EncodedPointer(uint64_t address) { |
|
193 return EncodedPointer(address, pointer_encoding_, encoded_pointer_bases_); |
|
194 } |
|
195 CFISection &EncodedPointer(uint64_t address, DwarfPointerEncoding encoding) { |
|
196 return EncodedPointer(address, encoding, encoded_pointer_bases_); |
|
197 } |
|
198 CFISection &EncodedPointer(uint64_t address, DwarfPointerEncoding encoding, |
|
199 const EncodedPointerBases &bases); |
|
200 |
|
201 // Restate some member functions, to keep chaining working nicely. |
|
202 CFISection &Mark(Label *label) { Section::Mark(label); return *this; } |
|
203 CFISection &D8(uint8_t v) { Section::D8(v); return *this; } |
|
204 CFISection &D16(uint16_t v) { Section::D16(v); return *this; } |
|
205 CFISection &D16(Label v) { Section::D16(v); return *this; } |
|
206 CFISection &D32(uint32_t v) { Section::D32(v); return *this; } |
|
207 CFISection &D32(const Label &v) { Section::D32(v); return *this; } |
|
208 CFISection &D64(uint64_t v) { Section::D64(v); return *this; } |
|
209 CFISection &D64(const Label &v) { Section::D64(v); return *this; } |
|
210 CFISection &LEB128(long long v) { Section::LEB128(v); return *this; } |
|
211 CFISection &ULEB128(uint64_t v) { Section::ULEB128(v); return *this; } |
|
212 |
|
213 private: |
|
214 // A length value that we've appended to the section, but is not yet |
|
215 // known. LENGTH is the appended value; START is a label referring |
|
216 // to the start of the data whose length was cited. |
|
217 struct PendingLength { |
|
218 Label length; |
|
219 Label start; |
|
220 }; |
|
221 |
|
222 // Constants used in CFI/.eh_frame data: |
|
223 |
|
224 // If the first four bytes of an "initial length" are this constant, then |
|
225 // the data uses the 64-bit DWARF format, and the length itself is the |
|
226 // subsequent eight bytes. |
|
227 static const uint32_t kDwarf64InitialLengthMarker = 0xffffffffU; |
|
228 |
|
229 // The CIE identifier for 32- and 64-bit DWARF CFI and .eh_frame data. |
|
230 static const uint32_t kDwarf32CIEIdentifier = ~(uint32_t)0; |
|
231 static const uint64_t kDwarf64CIEIdentifier = ~(uint64_t)0; |
|
232 static const uint32_t kEHFrame32CIEIdentifier = 0; |
|
233 static const uint64_t kEHFrame64CIEIdentifier = 0; |
|
234 |
|
235 // The size of a machine address for the data in this section. |
|
236 size_t address_size_; |
|
237 |
|
238 // If true, we are generating a Linux .eh_frame section, instead of |
|
239 // a standard DWARF .debug_frame section. |
|
240 bool eh_frame_; |
|
241 |
|
242 // The encoding to use for FDE pointers. |
|
243 DwarfPointerEncoding pointer_encoding_; |
|
244 |
|
245 // The base addresses to use when emitting encoded pointers. |
|
246 EncodedPointerBases encoded_pointer_bases_; |
|
247 |
|
248 // The length value for the current entry. |
|
249 // |
|
250 // Oddly, this must be dynamically allocated. Labels never get new |
|
251 // values; they only acquire constraints on the value they already |
|
252 // have, or assert if you assign them something incompatible. So |
|
253 // each header needs truly fresh Label objects to cite in their |
|
254 // headers and track their positions. The alternative is explicit |
|
255 // destructor invocation and a placement new. Ick. |
|
256 PendingLength *entry_length_; |
|
257 |
|
258 // True if we are currently emitting an FDE --- that is, we have |
|
259 // called FDEHeader but have not yet called FinishEntry. |
|
260 bool in_fde_; |
|
261 |
|
262 // If in_fde_ is true, this is its starting address. We use this for |
|
263 // emitting DW_EH_PE_funcrel pointers. |
|
264 uint64_t fde_start_address_; |
|
265 }; |
|
266 |
|
267 } // namespace google_breakpad |
|
268 |
|
269 #endif // PROCESSOR_CFI_ASSEMBLER_H_ |