Add Position Independent Pages (PIP) relocation model

[rinon]

Summary

The PIP relocation model allows the dynamic loader to load a specially marked
PIP code segments at random addresses, independent of other segments. Code in a
PIP segment must refer to addresses outside of its own segment by indirection
through a Page Offset Table (POT). The address of the POT must be loaded into a
platform-specific register before branching to a PIP segment. For non-PIP
segments, the PIP relocation model is identical to the PIC_ model.

This patch set (D37580, D37581, D37582, D37583, D37584, D37585, D37586, D37587) is a first draft of the pagerando implementation described in http://lists.llvm.org/pipermail/llvm-dev/2017-June/113794.html.

[rinon]

This patch set (D37580, D37581, D37582, D37583, D37584, D37585, D37586, D37587) is a first draft of the pagerando implementation described in http://lists.llvm.org/pipermail/llvm-dev/2017-June/113794.html.

I'm not entirely sure that a new relocation model is the best way to approach this. Pagerando address references are similar to a couple of other ARM-specific reloc models (i.e. ROPI) and are definitely not PC-relative PIC, so I decided to start by specifying it as a new reloc model. However, pagerando modules are a mix of PIC and PIP models depending on the function, so a new reloc model may be too broad. Alternatives are just marking the function and teaching the backend how to lower references for functions marked for pagerando. Feedback on whether to use the new reloc model or a different approach would be appreciated.

[rinon]

Rebase

[peter.smith]

I've got a few comments on some of the later patches, as this is the first in the series I just thought I'd mention that, as far as I know, adding llvm-commits as a subscriber after the initial creation is not sufficient for the mailing list to get any updates. I think the usual advice is to abandon the existing reviews and start new ones with llvm-commits subscribed. I'd also suggest adding a link to your llvm-dev post http://lists.llvm.org/pipermail/llvm-dev/2017-June/113794.html to at least one of the reviews as it isn't possible to understand without it. Some of the descriptions of later reviews are quite terse, especially for the backend changes that could do with some more description on why you've made the changes that you would like to make.

I've no specific comments on this patch.

[rinon]

Thanks for taking a look at this!

It looks like everything but the first revision got sent to llvm-commits (http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20171030/498777.html). Since I've rebased all the patches, the diffs have also been sent to the list. If this is insufficient I'm happy to abandon and reopen revisions; I just don't want to make the conversations harder to follow on phabricator since there are already comments.

Good pooint, I'll link to the RFC to the summary, since the first comment I wrote on each patch will get hidden after a long discussion:

In D37580#863793, @rinon wrote:

This patch set (D37580, D37581, D37582, D37583, D37584, D37585, D37586, D37587) is a first draft of the pagerando implementation described in http://lists.llvm.org/pipermail/llvm-dev/2017-June/113794.html.

[rinon]

Rebase

[rinon]

Rebase

[rinon]

Rebase

[rinon]

Rebase

[rinon]

Rebase

include/llvm/CodeGen/CommandFlags.h

cl::values(
                    "Fully relocatable, position independent code"),
         clEnumValN(Reloc::DynamicNoPIC, "dynamic-no-pic",
                    "Relocatable external references, non-relocatable code"),
         clEnumValN(Reloc::ROPI, "ropi",
                    "Code and read-only data relocatable, accessed PC-relative"),
         clEnumValN(Reloc::RWPI, "rwpi",
                    "Read-write data relocatable, accessed relative to static base"),
         clEnumValN(Reloc::ROPI_RWPI, "ropi-rwpi",
-                   "Combination of ropi and rwpi")));
+                   "Combination of ropi and rwpi"),
+        clEnumValN(Reloc::PIP, "pip",
+                   "Position independent pages for Pagerando")));
 
 static inline Optional<Reloc::Model> getRelocModel() {
   if (RelocModel.getNumOccurrences()) {
     Reloc::Model R = RelocModel;
     return R;
   }
   return None;
 }

include/llvm/Support/CodeGen.h

 
 #ifndef LLVM_SUPPORT_CODEGEN_H
 #define LLVM_SUPPORT_CODEGEN_H
 
 namespace llvm {
 
   // Relocation model types.
   namespace Reloc {
-  enum Model { Static, PIC_, DynamicNoPIC, ROPI, RWPI, ROPI_RWPI };
+  enum Model { Static, PIC_, DynamicNoPIC, ROPI, RWPI, ROPI_RWPI, PIP };
   }
 
   // Code model types.
   namespace CodeModel {
     // Sync changes with CodeGenCWrappers.h.
   enum Model { Small, Kernel, Medium, Large };
   }
 

include/llvm/Target/TargetMachine.h

public:
   }
 
   /// Return true if functions should be emitted into their own section,
   /// corresponding to -ffunction-sections.
   bool getFunctionSections() const {
     return Options.FunctionSections;
   }
 
+  /// Return true if Pagerando is enabled on this target.
+  bool isPagerando() const { return getRelocationModel() == Reloc::PIP; }
+
   /// \brief Get a \c TargetIRAnalysis appropriate for the target.
   ///
   /// This is used to construct the new pass manager's target IR analysis pass,
   /// set up appropriately for this target machine. Even the old pass manager
   /// uses this to answer queries about the IR.
   virtual TargetIRAnalysis getTargetIRAnalysis();
 
   /// Allow the target to modify the pass manager, e.g. by calling

lib/Target/TargetMachine.cpp

 TargetMachine::~TargetMachine() {
   delete AsmInfo;
   delete MRI;
   delete MII;
   delete STI;
 }
 
 bool TargetMachine::isPositionIndependent() const {
-  return getRelocationModel() == Reloc::PIC_;
+  return getRelocationModel() == Reloc::PIC_ ||
+         getRelocationModel() == Reloc::PIP;
 }
 
 /// \brief Reset the target options based on the function's attributes.
 // FIXME: This function needs to go away for a number of reasons:
 // a) global state on the TargetMachine is terrible in general,
 // b) these target options should be passed only on the function
 //    and not on the TargetMachine (via TargetOptions) at all.
 void TargetMachine::resetTargetOptions(const Function &F) const {
bool TargetMachine::shouldAssumeDSOLocal(const Module &M,
 
   // ELF supports preemption of other symbols.
   return false;
 }
 
 TLSModel::Model TargetMachine::getTLSModel(const GlobalValue *GV) const {
   bool IsPIE = GV->getParent()->getPIELevel() != PIELevel::Default;
   Reloc::Model RM = getRelocationModel();
-  bool IsSharedLibrary = RM == Reloc::PIC_ && !IsPIE;
+  bool IsSharedLibrary = (RM == Reloc::PIC_ || RM == Reloc::PIP) && !IsPIE;
   bool IsLocal = shouldAssumeDSOLocal(*GV->getParent(), GV);
 
   TLSModel::Model Model;
   if (IsSharedLibrary) {
     if (IsLocal)
       Model = TLSModel::LocalDynamic;
     else
       Model = TLSModel::GeneralDynamic;

tools/gold/gold-plugin.cpp

if (opt.startswith("mcpu=")) {
               .getAsInteger(10, ParallelCodeGenParallelismLevel))
         message(LDPL_FATAL, "Invalid codegen partition level: %s", opt_ + 5);
     } else if (opt == "disable-verify") {
       DisableVerify = true;
     } else if (opt.startswith("sample-profile=")) {
       sample_profile= opt.substr(strlen("sample-profile="));
     } else if (opt == "new-pass-manager") {
       new_pass_manager = true;
+    } else if (opt == "pagerando") {
+      RelocationModel = Reloc::PIP;
     } else {
       // Save this option to pass to the code generator.
       // ParseCommandLineOptions() expects argv[0] to be program name. Lazily
       // add that.
       if (extra.empty())
         extra.push_back("LLVMgold");
 
       extra.push_back(opt_);
case LDPT_OUTPUT_NAME:
       break;
     case LDPT_LINKER_OUTPUT:
       switch (tv->tv_u.tv_val) {
       case LDPO_REL: // .o
         IsExecutable = false;
         break;
       case LDPO_DYN: // .so
         IsExecutable = false;
-        RelocationModel = Reloc::PIC_;
+        RelocationModel = RelocationModel ? RelocationModel : Reloc::PIC_;
         break;
       case LDPO_PIE: // position independent executable
         IsExecutable = true;
-        RelocationModel = Reloc::PIC_;
+        RelocationModel = RelocationModel ? RelocationModel : Reloc::PIC_;
         break;
       case LDPO_EXEC: // .exe
         IsExecutable = true;
-        RelocationModel = Reloc::Static;
+        RelocationModel = RelocationModel ? RelocationModel : Reloc::Static;
         break;
       default:
         message(LDPL_ERROR, "Unknown output file type %d", tv->tv_u.tv_val);
         return LDPS_ERR;
       }
       break;
     case LDPT_OPTION:
       options::process_plugin_option(tv->tv_u.tv_string);