aboutsummaryrefslogtreecommitdiffstats
path: root/lib/ReaderWriter/MachO/GOTPass.cpp
blob: bc66d49eafb94ebdaba244ea317bfb56d7c38189 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
//===- lib/ReaderWriter/MachO/GOTPass.cpp -----------------------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
///
/// \file
/// This linker pass transforms all GOT kind references to real references.
/// That is, in assembly you can write something like:
///     movq foo@GOTPCREL(%rip), %rax
/// which means you want to load a pointer to "foo" out of the GOT (global
/// Offsets Table). In the object file, the Atom containing this instruction
/// has a Reference whose target is an Atom named "foo" and the Reference
/// kind is a GOT load.  The linker needs to instantiate a pointer sized
/// GOT entry.  This is done be creating a GOT Atom to represent that pointer
/// sized data in this pass, and altering the Atom graph so the Reference now
/// points to the GOT Atom entry (corresponding to "foo") and changing the
/// Reference Kind to reflect it is now pointing to a GOT entry (rather
/// then needing a GOT entry).
///
/// There is one optimization the linker can do here.  If the target of the GOT
/// is in the same linkage unit and does not need to be interposable, and
/// the GOT use is just a load (not some other operation), this pass can
/// transform that load into an LEA (add).  This optimizes away one memory load
/// which at runtime that could stall the pipeline.  This optimization only
/// works for architectures in which a (GOT) load instruction can be change to
/// an LEA instruction that is the same size.  The method isGOTAccess() should
/// only return true for "canBypassGOT" if this optimization is supported.
///
//===----------------------------------------------------------------------===//

#include "ArchHandler.h"
#include "File.h"
#include "MachOPasses.h"
#include "lld/Common/LLVM.h"
#include "lld/Core/DefinedAtom.h"
#include "lld/Core/File.h"
#include "lld/Core/Reference.h"
#include "lld/Core/Simple.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/STLExtras.h"

namespace lld {
namespace mach_o {

//
//  GOT Entry Atom created by the GOT pass.
//
class GOTEntryAtom : public SimpleDefinedAtom {
public:
  GOTEntryAtom(const File &file, bool is64, StringRef name)
    : SimpleDefinedAtom(file), _is64(is64), _name(name) { }

  ~GOTEntryAtom() override = default;

  ContentType contentType() const override {
    return DefinedAtom::typeGOT;
  }

  Alignment alignment() const override {
    return _is64 ? 8 : 4;
  }

  uint64_t size() const override {
    return _is64 ? 8 : 4;
  }

  ContentPermissions permissions() const override {
    return DefinedAtom::permRW_;
  }

  ArrayRef<uint8_t> rawContent() const override {
    static const uint8_t zeros[] =
        { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
    return llvm::makeArrayRef(zeros, size());
  }

  StringRef slotName() const {
    return _name;
  }

private:
  const bool _is64;
  StringRef _name;
};

/// Pass for instantiating and optimizing GOT slots.
///
class GOTPass : public Pass {
public:
  GOTPass(const MachOLinkingContext &context)
      : _ctx(context), _archHandler(_ctx.archHandler()),
        _file(*_ctx.make_file<MachOFile>("<mach-o GOT Pass>")) {
    _file.setOrdinal(_ctx.getNextOrdinalAndIncrement());
  }

private:
  llvm::Error perform(SimpleFile &mergedFile) override {
    // Scan all references in all atoms.
    for (const DefinedAtom *atom : mergedFile.defined()) {
      for (const Reference *ref : *atom) {
        // Look at instructions accessing the GOT.
        bool canBypassGOT;
        if (!_archHandler.isGOTAccess(*ref, canBypassGOT))
          continue;
        const Atom *target = ref->target();
        assert(target != nullptr);

        if (!shouldReplaceTargetWithGOTAtom(target, canBypassGOT)) {
          // Update reference kind to reflect that target is a direct accesss.
          _archHandler.updateReferenceToGOT(ref, false);
        } else {
          // Replace the target with a reference to a GOT entry.
          const DefinedAtom *gotEntry = makeGOTEntry(target);
          const_cast<Reference *>(ref)->setTarget(gotEntry);
          // Update reference kind to reflect that target is now a GOT entry.
          _archHandler.updateReferenceToGOT(ref, true);
        }
      }
    }

    // Sort and add all created GOT Atoms to master file
    std::vector<const GOTEntryAtom *> entries;
    entries.reserve(_targetToGOT.size());
    for (auto &it : _targetToGOT)
      entries.push_back(it.second);
    std::sort(entries.begin(), entries.end(),
              [](const GOTEntryAtom *left, const GOTEntryAtom *right) {
      return (left->slotName().compare(right->slotName()) < 0);
    });
    for (const GOTEntryAtom *slot : entries)
      mergedFile.addAtom(*slot);

    return llvm::Error::success();
  }

  bool shouldReplaceTargetWithGOTAtom(const Atom *target, bool canBypassGOT) {
    // Accesses to shared library symbols must go through GOT.
    if (isa<SharedLibraryAtom>(target))
      return true;
    // Accesses to interposable symbols in same linkage unit must also go
    // through GOT.
    const DefinedAtom *defTarget = dyn_cast<DefinedAtom>(target);
    if (defTarget != nullptr &&
        defTarget->interposable() != DefinedAtom::interposeNo) {
      assert(defTarget->scope() != DefinedAtom::scopeTranslationUnit);
      return true;
    }
    // Target does not require indirection.  So, if instruction allows GOT to be
    // by-passed, do that optimization and don't create GOT entry.
    return !canBypassGOT;
  }

  const DefinedAtom *makeGOTEntry(const Atom *target) {
    auto pos = _targetToGOT.find(target);
    if (pos == _targetToGOT.end()) {
      auto *gotEntry = new (_file.allocator())
          GOTEntryAtom(_file, _ctx.is64Bit(), target->name());
      _targetToGOT[target] = gotEntry;
      const ArchHandler::ReferenceInfo &nlInfo = _archHandler.stubInfo().
                                                nonLazyPointerReferenceToBinder;
      gotEntry->addReference(Reference::KindNamespace::mach_o, nlInfo.arch,
                             nlInfo.kind, 0, target, 0);
      return gotEntry;
    }
    return pos->second;
  }

  const MachOLinkingContext &_ctx;
  mach_o::ArchHandler                             &_archHandler;
  MachOFile                                       &_file;
  llvm::DenseMap<const Atom*, const GOTEntryAtom*> _targetToGOT;
};

void addGOTPass(PassManager &pm, const MachOLinkingContext &ctx) {
  assert(ctx.needsGOTPass());
  pm.add(llvm::make_unique<GOTPass>(ctx));
}

} // end namesapce mach_o
} // end namesapce lld