Support for the mno-tls-direct-seg-refs flag

[nruslan]

Summary

Allows to disable direct TLS segment access (%fs or %gs). GCC supports a similar flag, it can be useful in some circumstances, e.g. when a thread context block needs to be updated directly from user space. More info and specific use cases: https://bugs.llvm.org/show_bug.cgi?id=16145

There is another revision for LLVM as well.

Related: D53103

[hans]

Looking good, just a few minor comments.

docs/ClangCommandLineReference.rst

2241 ↗

(On Diff #169063)

This file is automatically generated based on the options .td files, so no need to update it here.

include/clang/Driver/CC1Options.td

195 ↗

(On Diff #169063)

Could mno_tls_direct_seg_refs be used as a cc1 flag instead?

include/clang/Driver/Options.td

2167

Maybe add (default) to the help text to indicate this is the default?

[nruslan]

@hans , @craig.topper : Updated the patch with all requested changes

docs/ClangCommandLineReference.rst

2241 ↗

(On Diff #169063)

ok

include/clang/Driver/CC1Options.td

195 ↗

(On Diff #169063)

done

include/clang/Driver/Options.td

2167

done

[nruslan]

@hans, @craig.topper : If everything is fine, can someone check in the clang and llvm patches on my behalf? I do not have commit access.

[kristina]

@nruslan This patchset is still pending review so be patient, I landed D53103 as it was reviewed and accepted by the code owner, on which this patch depends on. That said it LGTM, it's simple enough as it just forwards the argument to the backend, but still would rather have the X86 code owner sign off on this as well.

[craig.topper]

LGTM with that one style comment fixed.

lib/Driver/ToolChains/Clang.cpp

1743

minor - I think this line should be lined up with the start of the arguments to hasFlags on the previous line.

[kristina]

If the author doesn't mind I can just apply the style fix after patching and then rebuild and run all the relevant tests (or would you prefer to do that?). Seems easier than a new revision for an indentation change on one line.

[nruslan]

In D53102#1268272, @kristina wrote:

If the author doesn't mind I can just apply the style fix after patching and then rebuild and run all the relevant tests (or would you prefer to do that?). Seems easier than a new revision for an indentation change on one line.

@kristina : Thank you! Sure, that is fine. Go ahead.

[kristina]

By the way, out of curiosity is this for anything specific (alternative libc or some user-mode-scheduling implementation)? Not nitpicking, just curious since it's an interesting topic in general and it's frustrating that the architecture is so limited in terms of registers that can be used for TLS/related data unlike newer ARM/ARM64 architectures.

Also FWIW %gs is generally free to use under x86_64 Linux which is where I usually place my thread control blocks which doesn't interfere with libc which uses %fs. (Just started build/test job, waiting on that for now)

[nruslan]

In D53102#1268364, @kristina wrote:

By the way, out of curiosity is this for anything specific (alternative libc or some user-mode-scheduling implementation)? Not nitpicking, just curious since it's an interesting topic in general and it's frustrating that the architecture is so limited in terms of registers that can be used for TLS/related data unlike newer ARM/ARM64 architectures.

Also FWIW %gs is generally free to use under x86_64 Linux which is where I usually place my thread control blocks which doesn't interfere with libc which uses %fs. (Just started build/test job, waiting on that for now)

@kristina : Yes, there are some recent use cases mentioned in the bug report (see the link above). Also, I know that it was used for x86 (32-bit only) Xen guests where it would be recommended to compile an OS with this flag. I also used it in my fork of the NPTL pthread library in VirtuOS to support M:N threading model: http://sigops.org/sosp/sosp13/papers/p116-nikolaev.pdf

%gs is free in x86-64 Linux user space. AFAIK, it is used for per-CPU data in kernel and Win64.

[nruslan]

@kristina : Thank you very much for taking care of the patchsets!

include/clang/Driver/Options.td

def meabi : Separate<["-"], "meabi">, Group<m_Group>, Flags<[CC1Option]>,
   HelpText<"Set EABI type, e.g. 4, 5 or gnu (default depends on triple)">, Values<"default,4,5,gnu">;
 
 def mno_constant_cfstrings : Flag<["-"], "mno-constant-cfstrings">, Group<m_Group>;
 def mno_global_merge : Flag<["-"], "mno-global-merge">, Group<m_Group>, Flags<[CC1Option]>,
   HelpText<"Disable merging of globals">;
 def mno_pascal_strings : Flag<["-"], "mno-pascal-strings">,
   Alias<fno_pascal_strings>;
 def mno_red_zone : Flag<["-"], "mno-red-zone">, Group<m_Group>;
+def mno_tls_direct_seg_refs : Flag<["-"], "mno-tls-direct-seg-refs">, Group<m_Group>, Flags<[CC1Option]>,
+  HelpText<"Disable direct TLS access through segment registers">;
 def mno_relax_all : Flag<["-"], "mno-relax-all">, Group<m_Group>;
 def mno_rtd: Flag<["-"], "mno-rtd">, Group<m_Group>;
 def mno_soft_float : Flag<["-"], "mno-soft-float">, Group<m_Group>;
 def mno_stackrealign : Flag<["-"], "mno-stackrealign">, Group<m_Group>;
 
 def mretpoline : Flag<["-"], "mretpoline">, Group<m_Group>, Flags<[CoreOption,DriverOption]>;
 def mno_retpoline : Flag<["-"], "mno-retpoline">, Group<m_Group>, Flags<[CoreOption,DriverOption]>;
 def mspeculative_load_hardening : Flag<["-"], "mspeculative-load-hardening">,

 
 def mno_warn_nonportable_cfstrings : Flag<["-"], "mno-warn-nonportable-cfstrings">, Group<m_Group>;
 def mno_omit_leaf_frame_pointer : Flag<["-"], "mno-omit-leaf-frame-pointer">, Group<m_Group>;
 def momit_leaf_frame_pointer : Flag<["-"], "momit-leaf-frame-pointer">, Group<m_Group>,
   HelpText<"Omit frame pointer setup for leaf functions">, Flags<[CC1Option]>;
 def moslib_EQ : Joined<["-"], "moslib=">, Group<m_Group>;
 def mpascal_strings : Flag<["-"], "mpascal-strings">, Alias<fpascal_strings>;
 def mred_zone : Flag<["-"], "mred-zone">, Group<m_Group>;
+def mtls_direct_seg_refs : Flag<["-"], "mtls-direct-seg-refs">, Group<m_Group>,
+  HelpText<"Enable direct TLS access through segment registers (default)">;
 def mregparm_EQ : Joined<["-"], "mregparm=">, Group<m_Group>;
 def mrelax_all : Flag<["-"], "mrelax-all">, Group<m_Group>, Flags<[CC1Option,CC1AsOption]>,
   HelpText<"(integrated-as) Relax all machine instructions">;
 def mincremental_linker_compatible : Flag<["-"], "mincremental-linker-compatible">, Group<m_Group>,
   Flags<[CC1Option,CC1AsOption]>,
   HelpText<"(integrated-as) Emit an object file which can be used with an incremental linker">;
 def mno_incremental_linker_compatible : Flag<["-"], "mno-incremental-linker-compatible">, Group<m_Group>,
   HelpText<"(integrated-as) Emit an object file which cannot be used with an incremental linker">;

include/clang/Frontend/CodeGenOptions.def

CODEGENOPT(DisableLLVMPasses , 1, 0) ///< Don't run any LLVM IR passes to get
                                      ///< frontend.
 CODEGENOPT(DisableLifetimeMarkers, 1, 0) ///< Don't emit any lifetime markers
 CODEGENOPT(DisableO0ImplyOptNone , 1, 0) ///< Don't annonate function with optnone at O0
 CODEGENOPT(ExperimentalNewPassManager, 1, 0) ///< Enables the new, experimental
                                              ///< pass manager.
 CODEGENOPT(DebugPassManager, 1, 0) ///< Prints debug information for the new
                                    ///< pass manager.
 CODEGENOPT(DisableRedZone    , 1, 0) ///< Set when -mno-red-zone is enabled.
+CODEGENOPT(IndirectTlsSegRefs, 1, 0) ///< Set when -mno-tls-direct-seg-refs
+                                     ///< is specified.
 CODEGENOPT(DisableTailCalls  , 1, 0) ///< Do not emit tail calls.
 CODEGENOPT(NoEscapingBlockTailCalls, 1, 0) ///< Do not emit tail calls from
                                            ///< escaping blocks.
 CODEGENOPT(EmitDeclMetadata  , 1, 0) ///< Emit special metadata indicating what
                                      ///< Decl* various IR entities came from.
                                      ///< Only useful when running CodeGen as a
                                      ///< subroutine.
 CODEGENOPT(EmitVersionIdentMetadata , 1, 1) ///< Emit compiler version metadata.

lib/CodeGen/CGCall.cpp

if (!HasOptnone) {
     if (CodeGenOpts.OptimizeSize)
       FuncAttrs.addAttribute(llvm::Attribute::OptimizeForSize);
     if (CodeGenOpts.OptimizeSize == 2)
       FuncAttrs.addAttribute(llvm::Attribute::MinSize);
   }
 
   if (CodeGenOpts.DisableRedZone)
     FuncAttrs.addAttribute(llvm::Attribute::NoRedZone);
+  if (CodeGenOpts.IndirectTlsSegRefs)
+    FuncAttrs.addAttribute("indirect-tls-seg-refs");
   if (CodeGenOpts.NoImplicitFloat)
     FuncAttrs.addAttribute(llvm::Attribute::NoImplicitFloat);
 
   if (AttrOnCallSite) {
     // Attributes that should go on the call site only.
     if (!CodeGenOpts.SimplifyLibCalls ||
         CodeGenOpts.isNoBuiltinFunc(Name.data()))
       FuncAttrs.addAttribute(llvm::Attribute::NoBuiltin);

lib/Driver/ToolChains/Clang.cpp

 
 void Clang::AddX86TargetArgs(const ArgList &Args,
                              ArgStringList &CmdArgs) const {
   if (!Args.hasFlag(options::OPT_mred_zone, options::OPT_mno_red_zone, true) ||
       Args.hasArg(options::OPT_mkernel) ||
       Args.hasArg(options::OPT_fapple_kext))
     CmdArgs.push_back("-disable-red-zone");
 
+  if (!Args.hasFlag(options::OPT_mtls_direct_seg_refs,
+      options::OPT_mno_tls_direct_seg_refs, true))
+    CmdArgs.push_back("-mno-tls-direct-seg-refs");
+
   // Default to avoid implicit floating-point for kernel/kext code, but allow
   // that to be overridden with -mno-soft-float.
   bool NoImplicitFloat = (Args.hasArg(options::OPT_mkernel) ||
                           Args.hasArg(options::OPT_fapple_kext));
   if (Arg *A = Args.getLastArg(
           options::OPT_msoft_float, options::OPT_mno_soft_float,
           options::OPT_mimplicit_float, options::OPT_mno_implicit_float)) {
     const Option &O = A->getOption();

lib/Frontend/CompilerInvocation.cpp

static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
   if (const Arg *A =
           Args.getLastArg(OPT_emit_llvm_uselists, OPT_no_emit_llvm_uselists))
     Opts.EmitLLVMUseLists = A->getOption().getID() == OPT_emit_llvm_uselists;
 
   Opts.DisableLLVMPasses = Args.hasArg(OPT_disable_llvm_passes);
   Opts.DisableLifetimeMarkers = Args.hasArg(OPT_disable_lifetimemarkers);
   Opts.DisableO0ImplyOptNone = Args.hasArg(OPT_disable_O0_optnone);
   Opts.DisableRedZone = Args.hasArg(OPT_disable_red_zone);
+  Opts.IndirectTlsSegRefs = Args.hasArg(OPT_mno_tls_direct_seg_refs);
   Opts.ForbidGuardVariables = Args.hasArg(OPT_fforbid_guard_variables);
   Opts.UseRegisterSizedBitfieldAccess = Args.hasArg(
     OPT_fuse_register_sized_bitfield_access);
   Opts.RelaxedAliasing = Args.hasArg(OPT_relaxed_aliasing);
   Opts.StructPathTBAA = !Args.hasArg(OPT_no_struct_path_tbaa);
   Opts.NewStructPathTBAA = !Args.hasArg(OPT_no_struct_path_tbaa) &&
                            Args.hasArg(OPT_new_struct_path_tbaa);
   Opts.FineGrainedBitfieldAccesses =

test/CodeGen/indirect-tls-seg-refs.c

+// RUN: %clang_cc1 %s -emit-llvm -o - -mno-tls-direct-seg-refs | FileCheck %s -check-prefix=NO-TLSDIRECT
+// RUN: %clang_cc1 %s -emit-llvm -o - | FileCheck %s -check-prefix=TLSDIRECT
+
+// NO-TLSDIRECT: attributes #{{[0-9]+}} = {{{.*}} "indirect-tls-seg-refs"
+// TLSDIRECT-NOT: attributes #{{[0-9]+}} = {{{.*}} "indirect-tls-seg-refs"
+
+void test1() {
+}

test/Driver/indirect-tls-seg-refs.c

+// RUN: %clang -### %s 2>&1 | FileCheck %s -check-prefix=TLSDIRECT
+// RUN: %clang -### -mno-tls-direct-seg-refs -mtls-direct-seg-refs %s 2>&1 | FileCheck %s -check-prefix=TLSDIRECT
+// RUN: %clang -### -mtls-direct-seg-refs -mno-tls-direct-seg-refs %s 2>&1 | FileCheck %s -check-prefix=NO-TLSDIRECT
+// REQUIRES: clang-driver
+
+// NO-TLSDIRECT: -mno-tls-direct-seg-refs
+// TLSDIRECT-NOT: -mno-tls-direct-seg-refs