[ELF] --gc-sections: retain dependent sections of non-SHF_ALLOC sections

[MaskRay]

Summary

Fix http://lists.llvm.org/pipermail/llvm-dev/2020-October/145908.html

Currently non-SHF_ALLOC SHT_REL[A] (due to --emit-relocs) and SHF_LINK_ORDER are not
marked live.

[MaskRay]

Drop an unneeded change

[psmith]

LGTM. If we are marking a section live it makes sense to mark its dependent sections live.

[grimar]

LGTM too.

lld/ELF/MarkLive.cpp

368

nit: perhaps worth updating this comment.

[MaskRay]

Update comment

lld/ELF/MarkLive.cpp

for (Symbol *sym : symtab->symbols())
       if (auto *s = dyn_cast<SharedSymbol>(sym))
         if (s->isUsedInRegularObj && !s->isWeak())
           s->getFile().isNeeded = true;
     return;
   }
 
   // Otherwise, do mark-sweep GC.
   //
-  // The -gc-sections option works only for SHF_ALLOC sections
-  // (sections that are memory-mapped at runtime). So we can
-  // unconditionally make non-SHF_ALLOC sections alive except
-  // SHF_LINK_ORDER and SHT_REL/SHT_RELA sections.
+  // The -gc-sections option works only for SHF_ALLOC sections (sections that
+  // are memory-mapped at runtime). So we can unconditionally make non-SHF_ALLOC
+  // sections alive except SHF_LINK_ORDER, SHT_REL/SHT_RELA sections, and
+  // sections in a group.
   //
   // Usually, non-SHF_ALLOC sections are not removed even if they are
-  // unreachable through relocations because reachability is not
-  // a good signal whether they are garbage or not (e.g. there is
-  // usually no section referring to a .comment section, but we
-  // want to keep it.).
+  // unreachable through relocations because reachability is not a good signal
+  // whether they are garbage or not (e.g. there is usually no section referring
+  // to a .comment section, but we want to keep it.) When a non-SHF_ALLOC
+  // section is retained, we also retain sections dependent on it.
   //
   // Note on SHF_LINK_ORDER: Such sections contain metadata and they
   // have a reverse dependency on the InputSection they are linked with.
   // We are able to garbage collect them.
   //
   // Note on SHF_REL{,A}: Such sections reach here only when -r
   // or -emit-reloc were given. And they are subject of garbage
   // collection because, if we remove a text section, we also
   // remove its relocation section.
   //
   // Note on nextInSectionGroup: The ELF spec says that group sections are
   // included or omitted as a unit. We take the interpretation that:
   //
   // - Group members (nextInSectionGroup != nullptr) are subject to garbage
   //   collection.
   // - Groups members are retained or discarded as a unit.
   for (InputSectionBase *sec : inputSections) {
     bool isAlloc = (sec->flags & SHF_ALLOC);
     bool isLinkOrder = (sec->flags & SHF_LINK_ORDER);
     bool isRel = (sec->type == SHT_REL || sec->type == SHT_RELA);
 
-    if (!isAlloc && !isLinkOrder && !isRel && !sec->nextInSectionGroup)
+    if (!isAlloc && !isLinkOrder && !isRel && !sec->nextInSectionGroup) {
       sec->markLive();
+      for (InputSection *isec : sec->dependentSections)
+        isec->markLive();
+    }
   }
 
   // Follow the graph to mark all live sections.
   for (unsigned curPart = 1; curPart <= partitions.size(); ++curPart)
     MarkLive<ELFT>(curPart).run();
 
   // If we have multiple partitions, some sections need to live in the main
   // partition even if they were allocated to a loadable partition. Move them

lld/test/ELF/gc-sections-linkorder2.s

 # REQUIRES: x86
 
 # RUN: llvm-mc -filetype=obj -triple=x86_64-pc-linux %s -o %t.o
 # RUN: ld.lld --gc-sections --print-gc-sections %t.o -o /dev/null | count 0
 
 .globl _start
 _start:
 .quad .foo
 
 ## .foo is retained, so sections linking to it are retained as well.
 .section .foo,"a"
 .quad 0
 .section .bar,"ao",@progbits,.foo
 .quad 0
 .section .zed,"ao",@progbits,.foo
 .quad 0
+
+.section .nonalloc
+.quad 0
+
+.section .nonalloc_linkorder,"o",@progbits,.nonalloc
+.quad 0