Track whether all relocations in a section are implicit, and don't emit them in that case

[pete]

Summary

Hi Nick

The __eh_frame section contains relocations which can always be implicitly generated.

This patch tracks whether sections have only implicitly relocations and skips emitting them to the object file if that is the case.

The test case here ensures that this is the case for __eh_frame sections.

[pete]

Rebased to trunk

[lhames]

Apart from that, LGTM. Thanks Pete!

lib/ReaderWriter/MachO/MachONormalizedFileBinaryUtils.h

172

Maybe "relocsToDefinedCanBeImplicit"? Either way I think there should be a clear comment on the field in the struct definition below.

lib/ReaderWriter/MachO/MachONormalizedFileFromAtoms.cpp

63

We should have a comment explaining what this field implies when parsing and writing out a section.

[pete]

Thanks Lang. Committed with those fixes in r257099.

lib/ReaderWriter/MachO/MachONormalizedFileBinaryUtils.h

                                           StringRef &segmentName,
                                           StringRef &sectionName,
                                           SectionType &sectionType,
-                                          SectionAttr &sectionAttrs);
+                                          SectionAttr &sectionAttrs,
+                                          bool &allRelocsCanBeImplicit);
 
 } // namespace normalized
 } // namespace mach_o

lib/ReaderWriter/MachO/MachONormalizedFileFromAtoms.cpp

 
 struct SectionInfo {
   SectionInfo(StringRef seg, StringRef sect, SectionType type,
-              const MachOLinkingContext &ctxt, uint32_t attr=0);
+              const MachOLinkingContext &ctxt, uint32_t attr,
+              bool allRelocsCanBeImplicit);
 
   StringRef                 segmentName;
   StringRef                 sectionName;

   uint64_t                  address;
   uint64_t                  size;
   uint16_t                  alignment;
+  bool                      allRelocsCanBeImplicit;
   std::vector<AtomInfo>     atomsAndOffsets;
   uint32_t                  normalizedSectionIndex;
   uint32_t                  finalSectionIndex;
 };
 
 SectionInfo::SectionInfo(StringRef sg, StringRef sct, SectionType t,
-                         const MachOLinkingContext &ctxt, uint32_t attrs)
+                         const MachOLinkingContext &ctxt, uint32_t attrs,
+                         bool allRelocsCanBeImplicit)
  : segmentName(sg), sectionName(sct), type(t), attributes(attrs),
                  address(0), size(0), alignment(1),
+                 allRelocsCanBeImplicit(allRelocsCanBeImplicit),
                  normalizedSectionIndex(0), finalSectionIndex(0) {
   uint16_t align = 1;
   if (ctxt.sectionAligned(segmentName, sectionName, align)) {

   StringRef sectionName;
   SectionType sectionType;
   SectionAttr sectionAttrs;
+  bool allRelocsCanBeImplicit;
 
   // Use same table used by when parsing .o files.
   relocatableSectionInfoForContentType(type, segmentName, sectionName,
-                                       sectionType, sectionAttrs);
+                                       sectionType, sectionAttrs,
+                                       allRelocsCanBeImplicit);
   // If we already have a SectionInfo with this name, re-use it.
   // This can happen if two ContentType map to the same mach-o section.
   for (auto sect : _sectionMap) {

   }
   // Otherwise allocate new SectionInfo object.
   auto *sect = new (_allocator)
-      SectionInfo(segmentName, sectionName, sectionType, _ctx, sectionAttrs);
+      SectionInfo(segmentName, sectionName, sectionType, _ctx, sectionAttrs,
+                  allRelocsCanBeImplicit);
   _sectionInfos.push_back(sect);
   _sectionMap[type] = sect;
   return sect;

     }
     // Otherwise allocate new SectionInfo object.
     auto *sect = new (_allocator) SectionInfo(
-        p.segmentName, p.sectionName, p.sectionType, _ctx, sectionAttrs);
+        p.segmentName, p.sectionName, p.sectionType, _ctx, sectionAttrs,
+        /* allRelocsCanBeImplicit */ false);
     _sectionInfos.push_back(sect);
     _sectionMap[atomType] = sect;
     return sect;

     StringRef segName = customName.slice(0, seperatorIndex);
     StringRef sectName = customName.drop_front(seperatorIndex + 1);
     auto *sect =
-        new (_allocator) SectionInfo(segName, sectName, S_REGULAR, _ctx);
+        new (_allocator) SectionInfo(segName, sectName, S_REGULAR, _ctx,
+                                     0, /* allRelocsCanBeImplicit */ false);
     _customSections.push_back(sect);
     _sectionInfos.push_back(sect);
     return sect;

     for (const AtomInfo &info : si->atomsAndOffsets) {
       const DefinedAtom *atom = info.atom;
       for (const Reference *ref : *atom) {
+        // Skip emitting relocs for sections which are always able to be
+        // implicitly regenerated and where the relocation targets an address
+        // which is defined.
+        if (si->allRelocsCanBeImplicit && isa<DefinedAtom>(ref->target()))
+          continue;
         _archHandler.appendSectionRelocations(*atom, info.offsetInSection, *ref,
                                               symIndexForAtom,
                                               sectIndexForAtom,

lib/ReaderWriter/MachO/MachONormalizedFileToAtoms.cpp

                                           StringRef &segmentName,
                                           StringRef &sectionName,
                                           SectionType &sectionType,
-                                          SectionAttr &sectionAttrs) {
+                                          SectionAttr &sectionAttrs,
+                                          bool &allRelocsCanBeImplicit) {
 
   for (const MachORelocatableSectionToAtomType *p = sectsToAtomType ;
                                  p->atomType != DefinedAtom::typeUnknown; ++p) {

     sectionName = p->sectionName;
     sectionType = p->sectionType;
     sectionAttrs = 0;
+    allRelocsCanBeImplicit = false;
     if (atomType == DefinedAtom::typeCode)
       sectionAttrs = S_ATTR_PURE_INSTRUCTIONS;
+    if (atomType == DefinedAtom::typeCFI)
+      allRelocsCanBeImplicit = true;
     return;
   }
   llvm_unreachable("content type not yet supported");

test/mach-o/do-not-emit-unwind-fde-arm64.yaml

+# RUN: lld -flavor darwin -arch arm64 -r -print_atoms %s -o %t | FileCheck %s
+# RUN: lld -flavor darwin -arch arm64 -r -print_atoms %t -o %t2 | FileCheck %s
+# RUN: llvm-objdump -r -macho %t | FileCheck -check-prefix=CODE %s
+# RUN: llvm-objdump -r -macho %t2 | FileCheck -check-prefix=CODE %s
+
+
+--- !mach-o
+arch:            arm64
+file-type:       MH_OBJECT
+flags:           [ MH_SUBSECTIONS_VIA_SYMBOLS ]
+compat-version:  0.0
+current-version: 0.0
+has-UUID:        false
+OS:              unknown
+sections:        
+  - segment:         __TEXT
+    section:         __text
+    type:            S_REGULAR
+    attributes:      [ S_ATTR_PURE_INSTRUCTIONS, S_ATTR_SOME_INSTRUCTIONS ]
+    alignment:       4
+    address:         0x0000000000000000
+    content:         [ 0xFD, 0x7B, 0xBF, 0xA9, 0xFD, 0x03, 0x00, 0x91, 
+                       0x00, 0x00, 0x00, 0x90, 0x00, 0x00, 0x00, 0x91, 
+                       0x00, 0x00, 0x00, 0x94, 0x00, 0x00, 0x80, 0x52, 
+                       0xFD, 0x7B, 0xC1, 0xA8, 0xC0, 0x03, 0x5F, 0xD6 ]
+    relocations:     
+      - offset:          0x00000010
+        type:            ARM64_RELOC_BRANCH26
+        length:          2
+        pc-rel:          true
+        extern:          true
+        symbol:          9
+      - offset:          0x0000000C
+        type:            ARM64_RELOC_PAGEOFF12
+        length:          2
+        pc-rel:          false
+        extern:          true
+        symbol:          1
+      - offset:          0x00000008
+        type:            ARM64_RELOC_PAGE21
+        length:          2
+        pc-rel:          true
+        extern:          true
+        symbol:          1
+  - segment:         __TEXT
+    section:         __cstring
+    type:            S_CSTRING_LITERALS
+    attributes:      [  ]
+    address:         0x0000000000000020
+    content:         [ 0x48, 0x65, 0x6C, 0x6C, 0x6F, 0x20, 0x77, 0x6F, 
+                       0x72, 0x6C, 0x64, 0x00 ]
+  - segment:         __LD
+    section:         __compact_unwind
+    type:            S_REGULAR
+    attributes:      [  ]
+    alignment:       8
+    address:         0x0000000000000030
+    content:         [ 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 
+                       0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 
+                       0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 
+                       0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 ]
+    relocations:     
+      - offset:          0x00000000
+        type:            ARM64_RELOC_UNSIGNED
+        length:          3
+        pc-rel:          false
+        extern:          false
+        symbol:          1
+  - segment:         __TEXT
+    section:         __eh_frame
+    type:            S_COALESCED
+    attributes:      [  ]
+    alignment:       8
+    address:         0x0000000000000050
+    content:         [ 0x1C, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 
+                       0x01, 0x7A, 0x50, 0x4C, 0x52, 0x00, 0x01, 0x78, 
+                       0x1E, 0x0B, 0x00, 0xED, 0xFF, 0xFF, 0xFF, 0xFF, 
+                       0xFF, 0xFF, 0xFF, 0x00, 0x10, 0x0C, 0x1F, 0x00, 
+                       0x24, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 
+                       0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 
+                       0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 
+                       0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 
+                       0x00, 0x48, 0x0E, 0x10, 0x9E, 0x01, 0x9D, 0x02 ]
+local-symbols:   
+  - name:            ltmp0
+    type:            N_SECT
+    sect:            1
+    value:           0x0000000000000000
+  - name:            L_str
+    type:            N_SECT
+    sect:            2
+    value:           0x0000000000000020
+  - name:            ltmp1
+    type:            N_SECT
+    sect:            2
+    value:           0x0000000000000020
+  - name:            ltmp2
+    type:            N_SECT
+    sect:            3
+    value:           0x0000000000000030
+  - name:            ltmp3
+    type:            N_SECT
+    sect:            4
+    value:           0x0000000000000050
+  - name:            ltmp4
+    type:            N_SECT
+    sect:            4
+    value:           0x0000000000000070
+global-symbols:  
+  - name:            __Z3fooi
+    type:            N_SECT
+    scope:           [ N_EXT ]
+    sect:            1
+    value:           0x0000000000000000
+undefined-symbols: 
+  - name:            __gxx_personality_v0
+    type:            N_UNDF
+    scope:           [ N_EXT ]
+    value:           0x0000000000000000
+  - name:            _bar
+    type:            N_UNDF
+    scope:           [ N_EXT ]
+    value:           0x0000000000000000
+  - name:            _puts
+    type:            N_UNDF
+    scope:           [ N_EXT ]
+    value:           0x0000000000000000
+page-size:       0x00000000
+
+# CHECK: defined-atoms:   
+# CHECK:   - ref-name:        L{{[0-9]*}}
+# CHECK:     scope:           hidden
+# CHECK:     type:            c-string
+# CHECK:     content:         [ 48, 65, 6C, 6C, 6F, 20, 77, 6F, 72, 6C, 64, 00 ]
+# CHECK:     merge:           by-content
+# CHECK:     type:            unwind-cfi
+# CHECK:     content:         [ 1C, 00, 00, 00, 00, 00, 00, 00, 01, 7A, 50, 4C, 
+# CHECK:                        52, 00, 01, 78, 1E, 0B, 00, ED, FF, FF, FF, FF, 
+# CHECK:                        FF, FF, FF, 00, 10, 0C, 1F, 00 ]
+# CHECK:     alignment:       8
+# CHECK:   - ref-name:        L{{[0-9]*}}
+# CHECK:     type:            unwind-cfi
+# CHECK:     content:         [ 24, 00, 00, 00, 24, 00, 00, 00, 00, 00, 00, 00, 
+# CHECK:                        00, 00, 00, 00, 20, 00, 00, 00, 00, 00, 00, 00, 
+# CHECK:                        08, 00, 00, 00, 00, 00, 00, 00, 00, 48, 0E, 10, 
+# CHECK:                        9E, 01, 9D, 02 ]
+# CHECK:     alignment:       8
+# CHECK:   - type:            compact-unwind
+# CHECK:     content:         [ 00, 00, 00, 00, 00, 00, 00, 00, 20, 00, 00, 00, 
+# CHECK:                        00, 00, 00, 03, 00, 00, 00, 00, 00, 00, 00, 00, 
+# CHECK:                        00, 00, 00, 00, 00, 00, 00, 00 ]
+# CHECK:     alignment:       8
+# CHECK:     references:      
+# CHECK:       - kind:            pointer64
+# CHECK:         offset:          0
+# CHECK:         target:          __Z3fooi
+# CHECK:   - name:            __Z3fooi
+# CHECK:     scope:           global
+# CHECK:     content:         [ FD, 7B, BF, A9, FD, 03, 00, 91, 00, 00, 00, 90, 
+# CHECK:                        00, 00, 00, 91, 00, 00, 00, 94, 00, 00, 80, 52, 
+# CHECK:                        FD, 7B, C1, A8, C0, 03, 5F, D6 ]
+# CHECK:     alignment:       4
+# CHECK:     references:      
+# CHECK:       - kind:            page21
+# CHECK:         offset:          8
+# CHECK:         target:          L{{[0-9]*}}
+# CHECK:       - kind:            offset12
+# CHECK:         offset:          12
+# CHECK:         target:          L{{[0-9]*}}
+# CHECK:       - kind:            branch26
+# CHECK:         offset:          16
+# CHECK:         target:          _puts
+
+# Make sure we don't have any relocations in the __eh_frame section
+# CODE-NOT: __eh_frame
 No newline at end of file