[V2] MC: make section classification a bit more thorough

[compnerd]

Summary

This does *NOT* change the emitted section flags in any way. This only
impacts the internal classification of sections.

Extend the section classification in LLVM for ELF targets. This has one
important change: we now classify sections as text by default rather
than readonly. This matches the behaviour for GAS better.

Ensure that any section that has a writable attribute set is not treated
as readonly. We also special case any section named .debug_ which is
reserved for DWARF as metadata. In the case none of the attributes are
set (or because no attributes were provided), consult the section name
for classification. We match the well known names and classify the
section accordingly. Any remaining section is now classified as text.

This change allows us to classify sections in the MC layer more
precisely which is needed for subsequent changes for handling target
specific behaviour.

Re-apply the change that was reverted with additional changes to
classify section prefixes appropriately and differentiate the TLS
sections, addressing the FIXME and post-commit review comments by
@MaskRay.

[MaskRay]

llvm/lib/MC/MCContext.cpp

567

I think !(... & ...) is more common

576

Drop all gnu.linkonce and llvm.linkonce. linkonce is used before COMDAT era (<2000) and llvm never generates it.

580

.data.rel.ro.* is possible.

583

Instead of a StringSwitch, consider

if (Name.consumes_front(".tbss") && (Name.empty() || Name[0] == '.'))
  ...

so that the many prefixes don't have to be repeated for Case/StartsWith.

592

Is default SectionKind::getText() instead of SectionKind::getReadOnly() desired? I expect that text is covered by a previous if: Flags & ELF::SHF_EXECINSTR

[MaskRay]

This has one important change: we now classify sections as text by default rather than readonly. This matches the behaviour for GAS better.

This is somewhat surprising to me. The default section (before a .section directive) in gas is .text but that does not indicate other sections?

[compnerd]

llvm/lib/MC/MCContext.cpp

592

This is the core of the change itself - and what is needed for subsequent changes.

[MaskRay]

llvm/lib/MC/MCContext.cpp

587

As discussed, we need a comment that we choose SectionKind::getText() because of an unknown Kind (like a forward declaration) and this is used by R_RISCV_{ADD,SUB}*

[compnerd]

llvm/lib/MC/MCContext.cpp

587

Added such a comment. Happy to adjust the comment if you feel it's unclear or overly verbose.

llvm/lib/MC/MCContext.cpp

MCSectionELF *MCContext::getELFSection(const Twine &Section, unsigned Type,
 
   StringRef CachedName = Entry.first.SectionName;
 
   SectionKind Kind;
   if (Flags & ELF::SHF_ARM_PURECODE)
     Kind = SectionKind::getExecuteOnly();
   else if (Flags & ELF::SHF_EXECINSTR)
     Kind = SectionKind::getText();
-  else
+  else if (~Flags & ELF::SHF_WRITE)
     Kind = SectionKind::getReadOnly();
+  else if (Flags & ELF::SHF_TLS)
+    Kind = (Type & ELF::SHT_NOBITS) ? SectionKind::getThreadBSS()
+                                    : SectionKind::getThreadData();
+  else
+    // Default to `SectionKind::getText()`. This is the default for gas as
+    // well. The condition that falls into this case is where we do not have any
+    // section flags and must infer a classification rather than where we have
+    // section flags (i.e. this is not that SHF_EXECINSTR is unset bur rather it
+    // is unknown).
+    Kind = llvm::StringSwitch<SectionKind>(CachedName)
+               .Case(".bss", SectionKind::getBSS())
+               .StartsWith(".bss.", SectionKind::getBSS())
+               .StartsWith(".gnu.linkonce.b.", SectionKind::getBSS())
+               .StartsWith(".llvm.linkonce.b.", SectionKind::getBSS())
+               .Case(".data", SectionKind::getData())
+               .Case(".data1", SectionKind::getData())
+               .Case(".data.rel.ro", SectionKind::getReadOnlyWithRel())
+               .Case(".rodata", SectionKind::getReadOnly())
+               .Case(".rodata1", SectionKind::getReadOnly())
+               .Case(".tbss", SectionKind::getThreadBSS())
+               .StartsWith(".tbss.", SectionKind::getThreadData())
+               .StartsWith(".gnu.linkonce.tb.", SectionKind::getThreadData())
+               .StartsWith(".llvm.linkonce.tb.", SectionKind::getThreadData())
+               .Case(".tdata", SectionKind::getThreadData())
+               .StartsWith(".tdata.", SectionKind::getThreadData())
+               .StartsWith(".gnu.linkonce.td.", SectionKind::getThreadData())
+               .StartsWith(".llvm.linkonce.td.", SectionKind::getThreadData())
+               .StartsWith(".debug_", SectionKind::getMetadata())
+               .Default(SectionKind::getText());
 
   MCSectionELF *Result =
       createELFSectionImpl(CachedName, Type, Flags, Kind, EntrySize, GroupSym,
                            IsComdat, UniqueID, LinkedToSym);
   Entry.second = Result;
 
   recordELFMergeableSectionInfo(Result->getName(), Result->getFlags(),
                                 Result->getUniqueID(), Result->getEntrySize());

llvm/test/MC/ELF/section-classification.ll

+; We run this test explicitly converting from `.ll` to `.s` to `.o` to test the
+; classification of the section and the incorrect handling for the padding flag.
+
+; RUN: llc -filetype asm -o - %s | llvm-mc -triple x86_64-unknown-linux-gnu -filetype obj -o /dev/null -
+; REQUIRES: asserts
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.C = type { i32 }
+
+$_ZZ6getRefvE8instance = comdat any
+
+@_ZZ6getRefvE8instance = linkonce_odr dso_local global %struct.C zeroinitializer, comdat, align 4
[e-mail] = appending global [1 x ptr] [ptr @_ZZ6getRefvE8instance], section "llvm.metadata"