[ELF] - Implemented -z max-page-size option.

[grimar]

Summary

-z max-page-size can be used to set the target's page size.
This can be used for some platforms requiring different value
from common one.

[• rafael]

What is using this?

Cheers,
Rafael

[grimar]

In D19600#414041, @rafael wrote:

What is using this?

Cheers,
Rafael

I saw 3 reasons to implement this:

1. If we are going to support linkerscript commands one day like:

. = ALIGN (CONSTANT (MAXPAGESIZE)) - ((CONSTANT (MAXPAGESIZE) - .) & (CONSTANT (MAXPAGESIZE) - 1));
. = DATA_SEGMENT_ALIGN (CONSTANT (MAXPAGESIZE), CONSTANT (COMMONPAGESIZE));
. = DATA_SEGMENT_END (.);

It uses 2 different page sizes: common page size and max page size.
DATA_SEGMENT_ALIGN command can be used to perform optimization - save
a bit of runtime memory, but consume more file space More details are here: https://sourceware.org/ml/binutils/2002-02/msg00265.html
Even if we will not support DATA_SEGMENT_ALIGN (I think it is possible just to ignore such command and its optimization), then we still might want to implement
that optimization for regular use (without LS). And we will need options to control max page size and common page size I guess.

2. It seems that tweaking page size can affect perfomance: ld currently uses 2mb page for x64 (https://www.sourceware.org/ml/binutils/2006-05/msg00279.html).
3. Some platforms requires different page sizes, for example armv6 wants 16kb: https://bugzilla.mozilla.org/show_bug.cgi?id=772886

[• rafael]

I saw 3 reasons to implement this:

1. If we are going to support linkerscript commands one day like:

. = ALIGN (CONSTANT (MAXPAGESIZE)) - ((CONSTANT (MAXPAGESIZE) - .) & (CONSTANT (MAXPAGESIZE) - 1));
. = DATA_SEGMENT_ALIGN (CONSTANT (MAXPAGESIZE), CONSTANT (COMMONPAGESIZE));
. = DATA_SEGMENT_END (.);

It uses 2 different page sizes: common page size and max page size.
DATA_SEGMENT_ALIGN command can be used to perform optimization - save
a bit of runtime memory, but consume more file space More details are here: https://sourceware.org/ml/binutils/2002-02/msg00265.html
Even if we will not support DATA_SEGMENT_ALIGN (I think it is possible just to ignore such command and its optimization), then we still might want to implement
that optimization for regular use (without LS). And we will need options to control max page size and common page size I guess.

All that can be implemented initially with just a single page size.

2. It seems that tweaking page size can affect perfomance: ld currently uses 2mb page for x64 (https://www.sourceware.org/ml/binutils/2006-05/msg00279.html).

I would like to see someone measure the benefit of that in a lld
produced binary before we add this feature.

3. Some platforms requires different page sizes, for example armv6 wants 16kb: https://bugzilla.mozilla.org/show_bug.cgi?id=772886

That can be done directly in the linker.

Cheers,
Rafael

[grimar]

We are not going to support this now.

ELF/Config.h

struct Configuration {
   bool ZOrigin;
   bool ZRelro;
   BuildIdKind BuildId = BuildIdKind::None;
   ELFKind EKind = ELFNoneKind;
   uint16_t EMachine = llvm::ELF::EM_NONE;
   uint64_t EntryAddr = -1;
   unsigned LtoJobs;
   unsigned LtoO;
+  unsigned ZMaxPageSize;
   unsigned Optimize;
 };
 
 // The only instance of Configuration struct.
 extern Configuration *Config;
 
 } // namespace elf
 } // namespace lld
 
 #endif

ELF/Driver.cpp

 
 static bool hasZOption(opt::InputArgList &Args, StringRef Key) {
   for (auto *Arg : Args.filtered(OPT_z))
     if (Key == Arg->getValue())
       return true;
   return false;
 }
 
+static unsigned getZOptionValue(opt::InputArgList &Args, StringRef Key) {
+  for (auto *Arg : Args.filtered(OPT_z)) {
+    StringRef ZExpr = Arg->getValue();
+    if (!ZExpr.startswith(Key))
+      continue;
+
+    ZExpr = ZExpr.drop_front(Key.size());
+    if (ZExpr.empty() || !ZExpr.startswith("=")) {
+      error("invalid format for -z " + Key + ": " + ZExpr);
+      return -1U;
+    }
+
+    StringRef Val = ZExpr.drop_front(1);
+    unsigned Ret = -1;
+    if (Val.getAsInteger(0, Ret))
+      error("-z " + Key + ": number expected, but got " + Val);
+    return Ret;
+  }
+  return -1U;
+}
+
 static void logCommandline(ArrayRef<const char *> Args) {
   if (std::error_code EC = sys::fs::create_directories(
         Config->Reproduce, /*IgnoreExisting=*/false)) {
     error(EC, Config->Reproduce + ": can't create directory");
     return;
   }
 
   SmallString<128> Path;
void LinkerDriver::readConfigs(opt::InputArgList &Args) {
   if (Config->LtoJobs == 0)
     error("number of threads must be > 0");
 
   Config->ZExecStack = hasZOption(Args, "execstack");
   Config->ZNodelete = hasZOption(Args, "nodelete");
   Config->ZNow = hasZOption(Args, "now");
   Config->ZOrigin = hasZOption(Args, "origin");
   Config->ZRelro = !hasZOption(Args, "norelro");
+  Config->ZMaxPageSize = getZOptionValue(Args, "max-page-size");
 
   if (Config->Relocatable)
     Config->StripAll = false;
 
   // --strip-all implies --strip-debug.
   if (Config->StripAll)
     Config->StripDebug = true;
 
template <class ELFT> void LinkerDriver::link(opt::InputArgList &Args) {
 
   // Set either EntryAddr (if S is a number) or EntrySym (otherwise).
   if (!Config->Entry.empty()) {
     StringRef S = Config->Entry;
     if (S.getAsInteger(0, Config->EntryAddr))
       Config->EntrySym = Symtab.addUndefined(S)->Backref;
   }
 
+  // Target page size can be changed by -z max-page-size.
+  if (Config->ZMaxPageSize != -1U) {
+    if (Config->ZMaxPageSize < Target->PageSize ||
+        (Config->ZMaxPageSize % Target->PageSize != 0))
+      error("wrong page size specified: " + Twine(Config->ZMaxPageSize));
+    Target->PageSize = Config->ZMaxPageSize;
+  }
+
   for (std::unique_ptr<InputFile> &F : Files)
     Symtab.addFile(std::move(F));
   if (HasError)
     return; // There were duplicate symbols or incompatible files
 
   Symtab.scanUndefinedFlags();
   Symtab.scanShlibUndefined();
   Symtab.scanDynamicList();

test/ELF/max-page-size.s

+# REQUIRES: x86
+
+# RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t
+
+# RUN: ld.lld -z max-page-size=0x20000 %t -o %t2
+# RUN: llvm-readobj -program-headers %t2 | FileCheck %s
+# CHECK:      ProgramHeaders [
+# CHECK:        ProgramHeader {
+# CHECK:          Type: PT_LOAD
+# CHECK-NEXT:     Offset: 0x0
+# CHECK-NEXT:     VirtualAddress: 0x10000
+# CHECK-NEXT:     PhysicalAddress: 0x10000
+# CHECK-NEXT:     FileSize: 344
+# CHECK-NEXT:     MemSize: 344
+# CHECK-NEXT:     Flags [
+# CHECK-NEXT:       PF_R
+# CHECK-NEXT:     ]
+# CHECK-NEXT:     Alignment: 131072
+# CHECK-NEXT:   }
+# CHECK-NEXT:   ProgramHeader {
+# CHECK-NEXT:     Type: PT_LOAD
+# CHECK-NEXT:     Offset: 0x20000
+# CHECK-NEXT:     VirtualAddress: 0x20000
+# CHECK-NEXT:     PhysicalAddress: 0x20000
+# CHECK-NEXT:     FileSize: 1
+# CHECK-NEXT:     MemSize: 1
+# CHECK-NEXT:     Flags [
+# CHECK-NEXT:       PF_R
+# CHECK-NEXT:       PF_X
+# CHECK-NEXT:     ]
+# CHECK-NEXT:     Alignment: 131072
+# CHECK-NEXT:   }
+# CHECK-NEXT:   ProgramHeader {
+# CHECK-NEXT:     Type: PT_LOAD
+# CHECK-NEXT:     Offset: 0x40000
+# CHECK-NEXT:     VirtualAddress: 0x40000
+# CHECK-NEXT:     PhysicalAddress: 0x40000
+# CHECK-NEXT:     FileSize: 8
+# CHECK-NEXT:     MemSize: 8
+# CHECK-NEXT:     Flags [
+# CHECK-NEXT:       PF_R
+# CHECK-NEXT:       PF_W
+# CHECK-NEXT:     ]
+# CHECK-NEXT:     Alignment: 131072
+# CHECK-NEXT:   }
+
+# RUN: ld.lld -z max-page-size=0x1000 %t -o %t2
+# RUN: llvm-readobj -program-headers %t2 | FileCheck -check-prefix STANDART %s
+# STANDART:      ProgramHeaders [
+# STANDART:        ProgramHeader {
+# STANDART:          Type: PT_LOAD
+# STANDART-NEXT:     Offset: 0x0
+# STANDART-NEXT:     VirtualAddress: 0x10000
+# STANDART-NEXT:     PhysicalAddress: 0x10000
+# STANDART-NEXT:     FileSize: 344
+# STANDART-NEXT:     MemSize: 344
+# STANDART-NEXT:     Flags [
+# STANDART-NEXT:       PF_R
+# STANDART-NEXT:     ]
+# STANDART-NEXT:     Alignment: 4096
+# STANDART-NEXT:   }
+# STANDART-NEXT:   ProgramHeader {
+# STANDART-NEXT:     Type: PT_LOAD
+# STANDART-NEXT:     Offset: 0x1000
+# STANDART-NEXT:     VirtualAddress: 0x11000
+# STANDART-NEXT:     PhysicalAddress: 0x11000
+# STANDART-NEXT:     FileSize: 1
+# STANDART-NEXT:     MemSize: 1
+# STANDART-NEXT:     Flags [
+# STANDART-NEXT:       PF_R
+# STANDART-NEXT:       PF_X
+# STANDART-NEXT:     ]
+# STANDART-NEXT:     Alignment: 4096
+# STANDART-NEXT:   }
+# STANDART-NEXT:   ProgramHeader {
+# STANDART-NEXT:     Type: PT_LOAD
+# STANDART-NEXT:     Offset: 0x2000
+# STANDART-NEXT:     VirtualAddress: 0x12000
+# STANDART-NEXT:     PhysicalAddress: 0x12000
+# STANDART-NEXT:     FileSize: 8
+# STANDART-NEXT:     MemSize: 8
+# STANDART-NEXT:     Flags [
+# STANDART-NEXT:       PF_R
+# STANDART-NEXT:       PF_W
+# STANDART-NEXT:     ]
+# STANDART-NEXT:     Alignment: 4096
+# STANDART-NEXT:   }
+
+# RUN: not ld.lld -z max-page-sizeX %t -o %t2 2>&1 | \
+# RUN:   FileCheck --check-prefix=INVALID1 %s
+# INVALID1: invalid format for -z max-page-size
+
+# RUN: not ld.lld -z max-page-size=X %t -o %t2 2>&1 | \
+# RUN:   FileCheck --check-prefix=INVALID2 %s
+# INVALID2: -z max-page-size: number expected, but got X
+
+# RUN: not ld.lld -z max-page-size=0x1 %t -o %t2 2>&1 | \
+# RUN:   FileCheck --check-prefix=INVALID3 %s
+# INVALID3: wrong page size specified: 1
+
+# RUN: not ld.lld -z max-page-size=0x1000001 %t -o %t2 2>&1 | \
+# RUN:   FileCheck --check-prefix=INVALID4 %s
+# INVALID4: wrong page size specified: 16777217
+
+.globl _start
+_start:
+  nop
+
+.section .a, "aw"
+.quad 0