Wed, 31 Dec 2014 06:09:35 +0100
Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.
michael@0 | 1 | /* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ |
michael@0 | 2 | /* vim: set ts=8 sts=2 et sw=2 tw=80: */ |
michael@0 | 3 | /* This Source Code Form is subject to the terms of the Mozilla Public |
michael@0 | 4 | * License, v. 2.0. If a copy of the MPL was not distributed with this |
michael@0 | 5 | * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ |
michael@0 | 6 | |
michael@0 | 7 | #ifndef LulMainInt_h |
michael@0 | 8 | #define LulMainInt_h |
michael@0 | 9 | |
michael@0 | 10 | #include "LulPlatformMacros.h" |
michael@0 | 11 | |
michael@0 | 12 | #include <vector> |
michael@0 | 13 | |
michael@0 | 14 | #include "mozilla/Assertions.h" |
michael@0 | 15 | |
michael@0 | 16 | // This file is provides internal interface inside LUL. If you are an |
michael@0 | 17 | // end-user of LUL, do not include it in your code. The end-user |
michael@0 | 18 | // interface is in LulMain.h. |
michael@0 | 19 | |
michael@0 | 20 | |
michael@0 | 21 | namespace lul { |
michael@0 | 22 | |
michael@0 | 23 | //////////////////////////////////////////////////////////////// |
michael@0 | 24 | // DW_REG_ constants // |
michael@0 | 25 | //////////////////////////////////////////////////////////////// |
michael@0 | 26 | |
michael@0 | 27 | // These are the Dwarf CFI register numbers, as (presumably) defined |
michael@0 | 28 | // in the ELF ABI supplements for each architecture. |
michael@0 | 29 | |
michael@0 | 30 | enum DW_REG_NUMBER { |
michael@0 | 31 | // No real register has this number. It's convenient to be able to |
michael@0 | 32 | // treat the CFA (Canonical Frame Address) as "just another |
michael@0 | 33 | // register", though. |
michael@0 | 34 | DW_REG_CFA = -1, |
michael@0 | 35 | #if defined(LUL_ARCH_arm) |
michael@0 | 36 | // ARM registers |
michael@0 | 37 | DW_REG_ARM_R7 = 7, |
michael@0 | 38 | DW_REG_ARM_R11 = 11, |
michael@0 | 39 | DW_REG_ARM_R12 = 12, |
michael@0 | 40 | DW_REG_ARM_R13 = 13, |
michael@0 | 41 | DW_REG_ARM_R14 = 14, |
michael@0 | 42 | DW_REG_ARM_R15 = 15, |
michael@0 | 43 | #elif defined(LUL_ARCH_x64) |
michael@0 | 44 | // Because the X86 (32 bit) and AMD64 (64 bit) summarisers are |
michael@0 | 45 | // combined, a merged set of register constants is needed. |
michael@0 | 46 | DW_REG_INTEL_XBP = 6, |
michael@0 | 47 | DW_REG_INTEL_XSP = 7, |
michael@0 | 48 | DW_REG_INTEL_XIP = 16, |
michael@0 | 49 | #elif defined(LUL_ARCH_x86) |
michael@0 | 50 | DW_REG_INTEL_XBP = 5, |
michael@0 | 51 | DW_REG_INTEL_XSP = 4, |
michael@0 | 52 | DW_REG_INTEL_XIP = 8, |
michael@0 | 53 | #else |
michael@0 | 54 | # error "Unknown arch" |
michael@0 | 55 | #endif |
michael@0 | 56 | }; |
michael@0 | 57 | |
michael@0 | 58 | |
michael@0 | 59 | //////////////////////////////////////////////////////////////// |
michael@0 | 60 | // LExpr // |
michael@0 | 61 | //////////////////////////////////////////////////////////////// |
michael@0 | 62 | |
michael@0 | 63 | // An expression -- very primitive. Denotes either "register + |
michael@0 | 64 | // offset" or a dereferenced version of the same. So as to allow |
michael@0 | 65 | // convenient handling of Dwarf-derived unwind info, the register may |
michael@0 | 66 | // also denote the CFA. A large number of these need to be stored, so |
michael@0 | 67 | // we ensure it fits into 8 bytes. See comment below on RuleSet to |
michael@0 | 68 | // see how expressions fit into the bigger picture. |
michael@0 | 69 | |
michael@0 | 70 | struct LExpr { |
michael@0 | 71 | // Denotes an expression with no value. |
michael@0 | 72 | LExpr() |
michael@0 | 73 | : mHow(UNKNOWN) |
michael@0 | 74 | , mReg(0) |
michael@0 | 75 | , mOffset(0) |
michael@0 | 76 | {} |
michael@0 | 77 | |
michael@0 | 78 | // Denotes any expressible expression. |
michael@0 | 79 | LExpr(uint8_t how, int16_t reg, int32_t offset) |
michael@0 | 80 | : mHow(how) |
michael@0 | 81 | , mReg(reg) |
michael@0 | 82 | , mOffset(offset) |
michael@0 | 83 | {} |
michael@0 | 84 | |
michael@0 | 85 | // Change the offset for an expression that references memory. |
michael@0 | 86 | LExpr add_delta(long delta) |
michael@0 | 87 | { |
michael@0 | 88 | MOZ_ASSERT(mHow == NODEREF); |
michael@0 | 89 | // If this is a non-debug build and the above assertion would have |
michael@0 | 90 | // failed, at least return LExpr() so that the machinery that uses |
michael@0 | 91 | // the resulting expression fails in a repeatable way. |
michael@0 | 92 | return (mHow == NODEREF) ? LExpr(mHow, mReg, mOffset+delta) |
michael@0 | 93 | : LExpr(); // Gone bad |
michael@0 | 94 | } |
michael@0 | 95 | |
michael@0 | 96 | // Dereference an expression that denotes a memory address. |
michael@0 | 97 | LExpr deref() |
michael@0 | 98 | { |
michael@0 | 99 | MOZ_ASSERT(mHow == NODEREF); |
michael@0 | 100 | // Same rationale as for add_delta(). |
michael@0 | 101 | return (mHow == NODEREF) ? LExpr(DEREF, mReg, mOffset) |
michael@0 | 102 | : LExpr(); // Gone bad |
michael@0 | 103 | } |
michael@0 | 104 | |
michael@0 | 105 | // Representation of expressions. If |mReg| is DW_REG_CFA (-1) then |
michael@0 | 106 | // it denotes the CFA. All other allowed values for |mReg| are |
michael@0 | 107 | // nonnegative and are DW_REG_ values. |
michael@0 | 108 | |
michael@0 | 109 | enum { UNKNOWN=0, // This LExpr denotes no value. |
michael@0 | 110 | NODEREF, // Value is (mReg + mOffset). |
michael@0 | 111 | DEREF }; // Value is *(mReg + mOffset). |
michael@0 | 112 | |
michael@0 | 113 | uint8_t mHow; // UNKNOWN, NODEREF or DEREF |
michael@0 | 114 | int16_t mReg; // A DW_REG_ value |
michael@0 | 115 | int32_t mOffset; // 32-bit signed offset should be more than enough. |
michael@0 | 116 | }; |
michael@0 | 117 | |
michael@0 | 118 | static_assert(sizeof(LExpr) <= 8, "LExpr size changed unexpectedly"); |
michael@0 | 119 | |
michael@0 | 120 | |
michael@0 | 121 | //////////////////////////////////////////////////////////////// |
michael@0 | 122 | // RuleSet // |
michael@0 | 123 | //////////////////////////////////////////////////////////////// |
michael@0 | 124 | |
michael@0 | 125 | // This is platform-dependent. For some address range, describes how |
michael@0 | 126 | // to recover the CFA and then how to recover the registers for the |
michael@0 | 127 | // previous frame. |
michael@0 | 128 | // |
michael@0 | 129 | // The set of LExprs contained in a given RuleSet describe a DAG which |
michael@0 | 130 | // says how to compute the caller's registers ("new registers") from |
michael@0 | 131 | // the callee's registers ("old registers"). The DAG can contain a |
michael@0 | 132 | // single internal node, which is the value of the CFA for the callee. |
michael@0 | 133 | // It would be possible to construct a DAG that omits the CFA, but |
michael@0 | 134 | // including it makes the summarisers simpler, and the Dwarf CFI spec |
michael@0 | 135 | // has the CFA as a central concept. |
michael@0 | 136 | // |
michael@0 | 137 | // For this to make sense, |mCfaExpr| can't have |
michael@0 | 138 | // |mReg| == DW_REG_CFA since we have no previous value for the CFA. |
michael@0 | 139 | // All of the other |Expr| fields can -- and usually do -- specify |
michael@0 | 140 | // |mReg| == DW_REG_CFA. |
michael@0 | 141 | // |
michael@0 | 142 | // With that in place, the unwind algorithm proceeds as follows. |
michael@0 | 143 | // |
michael@0 | 144 | // (0) Initially: we have values for the old registers, and a memory |
michael@0 | 145 | // image. |
michael@0 | 146 | // |
michael@0 | 147 | // (1) Compute the CFA by evaluating |mCfaExpr|. Add the computed |
michael@0 | 148 | // value to the set of "old registers". |
michael@0 | 149 | // |
michael@0 | 150 | // (2) Compute values for the registers by evaluating all of the other |
michael@0 | 151 | // |Expr| fields in the RuleSet. These can depend on both the old |
michael@0 | 152 | // register values and the just-computed CFA. |
michael@0 | 153 | // |
michael@0 | 154 | // If we are unwinding without computing a CFA, perhaps because the |
michael@0 | 155 | // RuleSets are derived from EXIDX instead of Dwarf, then |
michael@0 | 156 | // |mCfaExpr.mHow| will be LExpr::UNKNOWN, so the computed value will |
michael@0 | 157 | // be invalid -- that is, TaggedUWord() -- and so any attempt to use |
michael@0 | 158 | // that will result in the same value. But that's OK because the |
michael@0 | 159 | // RuleSet would make no sense if depended on the CFA but specified no |
michael@0 | 160 | // way to compute it. |
michael@0 | 161 | // |
michael@0 | 162 | // A RuleSet is not allowed to cover zero address range. Having zero |
michael@0 | 163 | // length would break binary searching in SecMaps and PriMaps. |
michael@0 | 164 | |
michael@0 | 165 | class RuleSet { |
michael@0 | 166 | public: |
michael@0 | 167 | RuleSet(); |
michael@0 | 168 | void Print(void(*aLog)(const char*)); |
michael@0 | 169 | |
michael@0 | 170 | // Find the LExpr* for a given DW_REG_ value in this class. |
michael@0 | 171 | LExpr* ExprForRegno(DW_REG_NUMBER aRegno); |
michael@0 | 172 | |
michael@0 | 173 | uintptr_t mAddr; |
michael@0 | 174 | uintptr_t mLen; |
michael@0 | 175 | // How to compute the CFA. |
michael@0 | 176 | LExpr mCfaExpr; |
michael@0 | 177 | // How to compute caller register values. These may reference the |
michael@0 | 178 | // value defined by |mCfaExpr|. |
michael@0 | 179 | #if defined(LUL_ARCH_x64) || defined(LUL_ARCH_x86) |
michael@0 | 180 | LExpr mXipExpr; // return address |
michael@0 | 181 | LExpr mXspExpr; |
michael@0 | 182 | LExpr mXbpExpr; |
michael@0 | 183 | #elif defined(LUL_ARCH_arm) |
michael@0 | 184 | LExpr mR15expr; // return address |
michael@0 | 185 | LExpr mR14expr; |
michael@0 | 186 | LExpr mR13expr; |
michael@0 | 187 | LExpr mR12expr; |
michael@0 | 188 | LExpr mR11expr; |
michael@0 | 189 | LExpr mR7expr; |
michael@0 | 190 | #else |
michael@0 | 191 | # error "Unknown arch" |
michael@0 | 192 | #endif |
michael@0 | 193 | }; |
michael@0 | 194 | |
michael@0 | 195 | |
michael@0 | 196 | //////////////////////////////////////////////////////////////// |
michael@0 | 197 | // SecMap // |
michael@0 | 198 | //////////////////////////////////////////////////////////////// |
michael@0 | 199 | |
michael@0 | 200 | // A SecMap may have zero address range, temporarily, whilst RuleSets |
michael@0 | 201 | // are being added to it. But adding a zero-range SecMap to a PriMap |
michael@0 | 202 | // will make it impossible to maintain the total order of the PriMap |
michael@0 | 203 | // entries, and so that can't be allowed to happen. |
michael@0 | 204 | |
michael@0 | 205 | class SecMap { |
michael@0 | 206 | public: |
michael@0 | 207 | // These summarise the contained mRuleSets, in that they give |
michael@0 | 208 | // exactly the lowest and highest addresses that any of the entries |
michael@0 | 209 | // in this SecMap cover. Hence invariants: |
michael@0 | 210 | // |
michael@0 | 211 | // mRuleSets is nonempty |
michael@0 | 212 | // <=> mSummaryMinAddr <= mSummaryMaxAddr |
michael@0 | 213 | // && mSummaryMinAddr == mRuleSets[0].mAddr |
michael@0 | 214 | // && mSummaryMaxAddr == mRuleSets[#rulesets-1].mAddr |
michael@0 | 215 | // + mRuleSets[#rulesets-1].mLen - 1; |
michael@0 | 216 | // |
michael@0 | 217 | // This requires that no RuleSet has zero length. |
michael@0 | 218 | // |
michael@0 | 219 | // mRuleSets is empty |
michael@0 | 220 | // <=> mSummaryMinAddr > mSummaryMaxAddr |
michael@0 | 221 | // |
michael@0 | 222 | // This doesn't constrain mSummaryMinAddr and mSummaryMaxAddr uniquely, |
michael@0 | 223 | // so let's use mSummaryMinAddr == 1 and mSummaryMaxAddr == 0 to denote |
michael@0 | 224 | // this case. |
michael@0 | 225 | |
michael@0 | 226 | SecMap(void(*aLog)(const char*)); |
michael@0 | 227 | ~SecMap(); |
michael@0 | 228 | |
michael@0 | 229 | // Binary search mRuleSets to find one that brackets |ia|, or nullptr |
michael@0 | 230 | // if none is found. It's not allowable to do this until PrepareRuleSets |
michael@0 | 231 | // has been called first. |
michael@0 | 232 | RuleSet* FindRuleSet(uintptr_t ia); |
michael@0 | 233 | |
michael@0 | 234 | // Add a RuleSet to the collection. The rule is copied in. Calling |
michael@0 | 235 | // this makes the map non-searchable. |
michael@0 | 236 | void AddRuleSet(RuleSet* rs); |
michael@0 | 237 | |
michael@0 | 238 | // Prepare the map for searching. Also, remove any rules for code |
michael@0 | 239 | // address ranges which don't fall inside [start, +len). |len| may |
michael@0 | 240 | // not be zero. |
michael@0 | 241 | void PrepareRuleSets(uintptr_t start, size_t len); |
michael@0 | 242 | |
michael@0 | 243 | bool IsEmpty(); |
michael@0 | 244 | |
michael@0 | 245 | size_t Size() { return mRuleSets.size(); } |
michael@0 | 246 | |
michael@0 | 247 | // The min and max addresses of the addresses in the contained |
michael@0 | 248 | // RuleSets. See comment above for invariants. |
michael@0 | 249 | uintptr_t mSummaryMinAddr; |
michael@0 | 250 | uintptr_t mSummaryMaxAddr; |
michael@0 | 251 | |
michael@0 | 252 | private: |
michael@0 | 253 | // False whilst adding entries; true once it is safe to call FindRuleSet. |
michael@0 | 254 | // Transition (false->true) is caused by calling PrepareRuleSets(). |
michael@0 | 255 | bool mUsable; |
michael@0 | 256 | |
michael@0 | 257 | // A vector of RuleSets, sorted, nonoverlapping (post Prepare()). |
michael@0 | 258 | std::vector<RuleSet> mRuleSets; |
michael@0 | 259 | |
michael@0 | 260 | // A logging sink, for debugging. |
michael@0 | 261 | void (*mLog)(const char*); |
michael@0 | 262 | }; |
michael@0 | 263 | |
michael@0 | 264 | } // namespace lul |
michael@0 | 265 | |
michael@0 | 266 | #endif // ndef LulMainInt_h |