Line data Source code
1 : //==- llvm/CodeGen/DwarfAccelTable.h - Dwarf Accelerator Tables --*- C++ -*-==//
2 : //
3 : // The LLVM Compiler Infrastructure
4 : //
5 : // This file is distributed under the University of Illinois Open Source
6 : // License. See LICENSE.TXT for details.
7 : //
8 : //===----------------------------------------------------------------------===//
9 : //
10 : // This file contains support for writing dwarf accelerator tables.
11 : //
12 : //===----------------------------------------------------------------------===//
13 :
14 : #ifndef LLVM_LIB_CODEGEN_ASMPRINTER_DWARFACCELTABLE_H
15 : #define LLVM_LIB_CODEGEN_ASMPRINTER_DWARFACCELTABLE_H
16 :
17 : #include "llvm/ADT/ArrayRef.h"
18 : #include "llvm/ADT/SmallVector.h"
19 : #include "llvm/ADT/StringMap.h"
20 : #include "llvm/ADT/StringRef.h"
21 : #include "llvm/BinaryFormat/Dwarf.h"
22 : #include "llvm/CodeGen/DIE.h"
23 : #include "llvm/CodeGen/DwarfStringPoolEntry.h"
24 : #include "llvm/MC/MCSymbol.h"
25 : #include "llvm/Support/Allocator.h"
26 : #include "llvm/Support/Debug.h"
27 : #include "llvm/Support/Format.h"
28 : #include "llvm/Support/raw_ostream.h"
29 : #include <cstddef>
30 : #include <cstdint>
31 : #include <vector>
32 :
33 : // The dwarf accelerator tables are an indirect hash table optimized
34 : // for null lookup rather than access to known data. They are output into
35 : // an on-disk format that looks like this:
36 : //
37 : // .-------------.
38 : // | HEADER |
39 : // |-------------|
40 : // | BUCKETS |
41 : // |-------------|
42 : // | HASHES |
43 : // |-------------|
44 : // | OFFSETS |
45 : // |-------------|
46 : // | DATA |
47 : // `-------------'
48 : //
49 : // where the header contains a magic number, version, type of hash function,
50 : // the number of buckets, total number of hashes, and room for a special
51 : // struct of data and the length of that struct.
52 : //
53 : // The buckets contain an index (e.g. 6) into the hashes array. The hashes
54 : // section contains all of the 32-bit hash values in contiguous memory, and
55 : // the offsets contain the offset into the data area for the particular
56 : // hash.
57 : //
58 : // For a lookup example, we could hash a function name and take it modulo the
59 : // number of buckets giving us our bucket. From there we take the bucket value
60 : // as an index into the hashes table and look at each successive hash as long
61 : // as the hash value is still the same modulo result (bucket value) as earlier.
62 : // If we have a match we look at that same entry in the offsets table and
63 : // grab the offset in the data for our final match.
64 :
65 : namespace llvm {
66 :
67 : class AsmPrinter;
68 : class DwarfDebug;
69 :
70 262046 : class DwarfAccelTable {
71 : static uint32_t HashDJB(StringRef Str) {
72 906 : uint32_t h = 5381;
73 7476 : for (unsigned i = 0, e = Str.size(); i != e; ++i)
74 13140 : h = ((h << 5) + h) + Str[i];
75 : return h;
76 : }
77 :
78 : // Helper function to compute the number of buckets needed based on
79 : // the number of unique hashes.
80 : void ComputeBucketCount();
81 :
82 : struct TableHeader {
83 : uint32_t magic = MagicHash; // 'HASH' magic value to allow endian detection
84 : uint16_t version = 1; // Version number.
85 : uint16_t hash_function = dwarf::DW_hash_function_djb;
86 : // The hash function enumeration that was used.
87 : uint32_t bucket_count = 0; // The number of buckets in this hash table.
88 : uint32_t hashes_count = 0; // The total number of unique hash values
89 : // and hash data offsets in this table.
90 : uint32_t header_data_len; // The bytes to skip to get to the hash
91 : // indexes (buckets) for correct alignment.
92 : // Also written to disk is the implementation specific header data.
93 :
94 : static const uint32_t MagicHash = 0x48415348;
95 :
96 65896 : TableHeader(uint32_t data_len) : header_data_len(data_len) {}
97 :
98 : #ifndef NDEBUG
99 : void print(raw_ostream &OS) {
100 : OS << "Magic: " << format("0x%x", magic) << "\n"
101 : << "Version: " << version << "\n"
102 : << "Hash Function: " << hash_function << "\n"
103 : << "Bucket Count: " << bucket_count << "\n"
104 : << "Header Data Length: " << header_data_len << "\n";
105 : }
106 :
107 : void dump() { print(dbgs()); }
108 : #endif
109 : };
110 :
111 : public:
112 : // The HeaderData describes the form of each set of data. In general this
113 : // is as a list of atoms (atom_count) where each atom contains a type
114 : // (AtomType type) of data, and an encoding form (form). In the case of
115 : // data that is referenced via DW_FORM_ref_* the die_offset_base is
116 : // used to describe the offset for all forms in the list of atoms.
117 : // This also serves as a public interface of sorts.
118 : // When written to disk this will have the form:
119 : //
120 : // uint32_t die_offset_base
121 : // uint32_t atom_count
122 : // atom_count Atoms
123 :
124 : // Make these public so that they can be used as a general interface to
125 : // the class.
126 : struct Atom {
127 : uint16_t type; // enum AtomType
128 : uint16_t form; // DWARF DW_FORM_ defines
129 :
130 49422 : constexpr Atom(uint16_t type, uint16_t form) : type(type), form(form) {}
131 :
132 : #ifndef NDEBUG
133 : void print(raw_ostream &OS) {
134 : OS << "Type: " << dwarf::AtomTypeString(type) << "\n"
135 : << "Form: " << dwarf::FormEncodingString(form) << "\n";
136 : }
137 :
138 : void dump() { print(dbgs()); }
139 : #endif
140 : };
141 :
142 : private:
143 131024 : struct TableHeaderData {
144 : uint32_t die_offset_base;
145 : SmallVector<Atom, 3> Atoms;
146 :
147 : TableHeaderData(ArrayRef<Atom> AtomList, uint32_t offset = 0)
148 197687 : : die_offset_base(offset), Atoms(AtomList.begin(), AtomList.end()) {}
149 :
150 : #ifndef NDEBUG
151 : void print(raw_ostream &OS) {
152 : OS << "die_offset_base: " << die_offset_base << "\n";
153 : for (size_t i = 0; i < Atoms.size(); i++)
154 : Atoms[i].print(OS);
155 : }
156 :
157 : void dump() { print(dbgs()); }
158 : #endif
159 : };
160 :
161 : // The data itself consists of a str_offset, a count of the DIEs in the
162 : // hash and the offsets to the DIEs themselves.
163 : // On disk each data section is ended with a 0 KeyType as the end of the
164 : // hash chain.
165 : // On output this looks like:
166 : // uint32_t str_offset
167 : // uint32_t hash_data_count
168 : // HashData[hash_data_count]
169 : public:
170 : struct HashDataContents {
171 : const DIE *Die; // Offsets
172 : char Flags; // Specific flags to output
173 :
174 961 : HashDataContents(const DIE *D, char Flags) : Die(D), Flags(Flags) {}
175 :
176 : #ifndef NDEBUG
177 : void print(raw_ostream &OS) const {
178 : OS << " Offset: " << Die->getOffset() << "\n"
179 : << " Tag: " << dwarf::TagString(Die->getTag()) << "\n"
180 : << " Flags: " << Flags << "\n";
181 : }
182 : #endif
183 : };
184 :
185 : private:
186 : // String Data
187 3624 : struct DataArray {
188 : DwarfStringPoolEntryRef Name;
189 : std::vector<HashDataContents *> Values;
190 : };
191 :
192 : friend struct HashData;
193 :
194 : struct HashData {
195 : StringRef Str;
196 : uint32_t HashValue;
197 : MCSymbol *Sym;
198 : DwarfAccelTable::DataArray &Data; // offsets
199 :
200 : HashData(StringRef S, DwarfAccelTable::DataArray &Data)
201 906 : : Str(S), Data(Data) {
202 906 : HashValue = DwarfAccelTable::HashDJB(S);
203 : }
204 :
205 : #ifndef NDEBUG
206 : void print(raw_ostream &OS) {
207 : OS << "Name: " << Str << "\n";
208 : OS << " Hash Value: " << format("0x%x", HashValue) << "\n";
209 : OS << " Symbol: ";
210 : if (Sym)
211 : OS << *Sym;
212 : else
213 : OS << "<none>";
214 : OS << "\n";
215 : for (HashDataContents *C : Data.Values) {
216 : OS << " Offset: " << C->Die->getOffset() << "\n";
217 : OS << " Tag: " << dwarf::TagString(C->Die->getTag()) << "\n";
218 : OS << " Flags: " << C->Flags << "\n";
219 : }
220 : }
221 :
222 : void dump() { print(dbgs()); }
223 : #endif
224 : };
225 :
226 : // Internal Functions
227 : void EmitHeader(AsmPrinter *);
228 : void EmitBuckets(AsmPrinter *);
229 : void EmitHashes(AsmPrinter *);
230 : void emitOffsets(AsmPrinter *, const MCSymbol *);
231 : void EmitData(AsmPrinter *, DwarfDebug *D);
232 :
233 : // Allocator for HashData and HashDataContents.
234 : BumpPtrAllocator Allocator;
235 :
236 : // Output Variables
237 : TableHeader Header;
238 : TableHeaderData HeaderData;
239 : std::vector<HashData *> Data;
240 :
241 : using StringEntries = StringMap<DataArray, BumpPtrAllocator &>;
242 :
243 : StringEntries Entries;
244 :
245 : // Buckets/Hashes/Offsets
246 : using HashList = std::vector<HashData *>;
247 : using BucketList = std::vector<HashList>;
248 : BucketList Buckets;
249 : HashList Hashes;
250 :
251 : // Public Implementation
252 : public:
253 : DwarfAccelTable(ArrayRef<DwarfAccelTable::Atom>);
254 : DwarfAccelTable(const DwarfAccelTable &) = delete;
255 : DwarfAccelTable &operator=(const DwarfAccelTable &) = delete;
256 :
257 : void AddName(DwarfStringPoolEntryRef Name, const DIE *Die, char Flags = 0);
258 : void FinalizeTable(AsmPrinter *, StringRef);
259 : void emit(AsmPrinter *, const MCSymbol *, DwarfDebug *);
260 : #ifndef NDEBUG
261 : void print(raw_ostream &OS);
262 : void dump() { print(dbgs()); }
263 : #endif
264 : };
265 :
266 : } // end namespace llvm
267 :
268 : #endif // LLVM_LIB_CODEGEN_ASMPRINTER_DWARFACCELTABLE_H
|