[PowerPC][LLD] Extend R2 save stub to support offsets of more than 26 bits

[stefanp]

Summary

The R2 save stub will now support offsets up to 64 bits.

There are three cases that will be used.

1. The offset fits in 26 bits.

b <26 bit offset>

2. The offset does not fit in 26 bits but fits in 34 bits.

paddi r12, 0, <34 bit offset>, 1
mtctr r12
bctr

3. The offset does not fit in 34 bits. Since this is an R2 save stub we can use

the TOC in R2. We are not loading the offset but the actual address we want to
branch to.

addis r12, r2, <address in TOC lo>
ld r12 <address in TOC hi>(r12)
mtctr r12
bctr

In case 1) the stub is only 8 bytes while in cases 2) and 3) the stub will be
20 bytes.

[MaskRay]

lld/ELF/Thunks.cpp

293

Any reason the alignment is 16?

[stefanp]

lld/ELF/Thunks.cpp

293

I've added a prefixed instruction (or the possibility of getting a prefixed instruction) paddi to the Thunk. The issue with prefixed instructions is that they are not allowed to cross a 64 byte boundary. It's a hardware limitation. If I align to 16 bytes I know that the next 16 bytes cannot cross a 64 byte boundary and so my paddi won't cross a 64 byte boundary.

[MaskRay]

Hope another reviewer can verify this patch.

lld/ELF/Thunks.cpp

312

Add const

[sfertile]

lld/ELF/Thunks.cpp

295

Arm thunks avoid oscillating between short and long thunk bodies. The comment says it can prevent the layout from converging. Should we do the same?

932

I know its a bit of a pain, but I think we should rename the local to aviod having a variable name start with an uppercase letter. Maybe 'offsetFromTOC'?

[MaskRay]

lld/ELF/Thunks.cpp

295

Ah, yes, we should do it.

[NeHuang]

lld/ELF/Thunks.cpp

295

Why do we have isInt<24> instead of isInt<26> here?

[stefanp]

Fixed 24 to 26 bits as the 24 bits was not correct.
Added mayUseShortThunk handling.

[MaskRay]

lld/test/ELF/ppc64-toc-call-to-pcrel-long-jump.s

3

This should use PHDRS now to avoid large executables. See D88037

[stefanp]

Added PHDRS to try to avoid large binary files.

[MaskRay]

lld/ELF/Thunks.cpp

925

const

932

const.

lld/test/ELF/ppc64-toc-call-to-pcrel-long-jump.s

3

There are too many \. split-file probably looks better

[stefanp]

Added const in two places.
Added split-file.

[MaskRay]

Looks great! Worth getting @sfertile's sign-off.

[NeHuang]

LGTM. Only two nits comment.

lld/test/ELF/ppc64-toc-call-to-pcrel-long-jump.s

50

CHECK-LABEL to CHECK

51

CHECK-LABEL to CHECK-NEXT

[sfertile]

2 minor comments but otherwise LGTM.

lld/ELF/Thunks.cpp

933

Can we change the name to avoid the clang tidy warnings?

[NeHuang]

lld/ELF/Thunks.cpp

933

+1

[stefanp]

Fixed last nits.

lld/ELF/Thunks.cpp

 // PPC64 R2 Save Stub
 // When the caller requires a valid R2 TOC pointer but the callee does not
 // require a TOC pointer and the callee cannot guarantee that it doesn't
 // clobber R2 then we need to save R2. This stub:
 // 1) Saves the TOC pointer to the stack.
 // 2) Tail calls the callee.
 class PPC64R2SaveStub final : public Thunk {
 public:
-  PPC64R2SaveStub(Symbol &dest) : Thunk(dest, 0) {}
-  uint32_t size() override { return 8; }
+  PPC64R2SaveStub(Symbol &dest, int64_t addend) : Thunk(dest, addend) {
+    alignment = 16;
+  }
+
+  // To prevent oscillations in layout when moving from short to long thunks
+  // we make sure that once a thunk has been set to long it cannot go back.
+  bool getMayUseShortThunk() {
+    if (!mayUseShortThunk)
+      return false;
+    if (!isInt<26>(computeOffset())) {
+      mayUseShortThunk = false;
+      return false;
+    }
+    return true;
+  }
+  uint32_t size() override { return getMayUseShortThunk() ? 8 : 20; }
   void writeTo(uint8_t *buf) override;
   void addSymbols(ThunkSection &isec) override;
+
+private:
+  // Transitioning from long to short can create layout oscillations in
+  // certain corner cases which would prevent the layout from converging.
+  // This is similar to the handling for ARMThunk.
+  bool mayUseShortThunk = true;
+  int64_t computeOffset() const {
+    return destination.getVA() - (getThunkTargetSym()->getVA() + 4);
+  }
 };
 
 // PPC64 R12 Setup Stub
 // When a caller that does not maintain a toc-pointer performs a local call to
 // a callee which requires a toc-pointer then we need this stub to place the
 // callee's global entry point into r12 without a save of R2.
 class PPC64R12SetupStub final : public Thunk {
 public:

 }
 
 bool PPC64PltCallStub::isCompatibleWith(const InputSection &isec,
                                         const Relocation &rel) const {
   return rel.type == R_PPC64_REL24 || rel.type == R_PPC64_REL14;
 }
 
 void PPC64R2SaveStub::writeTo(uint8_t *buf) {
-  int64_t offset = destination.getVA() - (getThunkTargetSym()->getVA() + 4);
-  // The branch offset needs to fit in 26 bits.
-  if (!isInt<26>(offset))
-    reportRangeError(buf, offset, 26, destination, "R2 save stub offset");
+  const int64_t offset = computeOffset();
   write32(buf + 0, 0xf8410018);                         // std  r2,24(r1)
+  // The branch offset needs to fit in 26 bits.
+  if (getMayUseShortThunk()) {
     write32(buf + 4, 0x48000000 | (offset & 0x03fffffc)); // b    <offset>
+  } else if (isInt<34>(offset)) {
+    uint64_t paddi = PADDI_R12_NO_DISP | (((offset >> 16) & 0x3ffff) << 32) |
+                     (offset & 0xffff);
+    writePrefixedInstruction(buf + 4, paddi); // paddi r12, 0, func@pcrel, 1
+    write32(buf + 12, MTCTR_R12);             // mtctr r12
+    write32(buf + 16, BCTR);                  // bctr
+  } else {
+    in.ppc64LongBranchTarget->addEntry(&destination, addend);
+    int64_t TOCoffset =
+        in.ppc64LongBranchTarget->getEntryVA(&destination, addend) -
+        getPPC64TocBase();
+    writePPC64LoadAndBranch(buf + 4, TOCoffset);
+  }
 }
 
 void PPC64R2SaveStub::addSymbols(ThunkSection &isec) {
   Defined *s = addSymbol(saver.save("__toc_save_" + destination.getName()),
                          STT_FUNC, 0, isec);
   s->needsTocRestore = true;
 }
 
static Thunk *addThunkPPC64(RelType type, Symbol &s, int64_t a) {
   if (s.isInPlt())
     return type == R_PPC64_REL24_NOTOC ? (Thunk *)make<PPC64PCRelPLTStub>(s)
                                        : (Thunk *)make<PPC64PltCallStub>(s);
 
   // This check looks at the st_other bits of the callee. If the value is 1
   // then the callee clobbers the TOC and we need an R2 save stub when RelType
   // is R_PPC64_REL14 or R_PPC64_REL24.
   if ((type == R_PPC64_REL14 || type == R_PPC64_REL24) && (s.stOther >> 5) == 1)
-    return make<PPC64R2SaveStub>(s);
+    return make<PPC64R2SaveStub>(s, a);
 
   if (type == R_PPC64_REL24_NOTOC)
     return (s.stOther >> 5) > 1
                ? (Thunk *)make<PPC64R12SetupStub>(s)
                : (Thunk *)make<PPC64PCRelLongBranchThunk>(s, a);
 
   if (config->picThunk)
     return make<PPC64PILongBranchThunk>(s, a);

lld/test/ELF/ppc64-toc-call-to-pcrel-long-jump.s

 # REQUIRES: ppc
-# RUN: echo 'SECTIONS { \
-# RUN:       .text_callee 0x10010000 : { *(.text_callee) } \
-# RUN:       .text_caller 0x20020000 : { *(.text_caller) } \
+
+# RUN: echo ' \
+# RUN:      PHDRS { \
+# RUN:        callee PT_LOAD FLAGS(0x1 | 0x4); \
+# RUN:        close PT_LOAD FLAGS(0x1 | 0x4); \
+# RUN:        caller PT_LOAD FLAGS(0x1 | 0x4); \
+# RUN:        far PT_LOAD FLAGS(0x1 | 0x4); \
+# RUN:      } \
+# RUN:      SECTIONS { \
+# RUN:       .text_callee 0x10010000 : { *(.text_callee) } :callee \
+# RUN:       .text_caller_close 0x10020000 : { *(.text_caller_close) } :close \
+# RUN:       .text_caller 0x20020000 : { *(.text_caller) } :caller \
+# RUN:       .text_caller_far 0x520020000 : { *(.text_caller_far) } :far \
 # RUN:       }' > %t.script
 
 # RUN: llvm-mc -filetype=obj -triple=powerpc64le %s -o %t.o
-# RUN: not ld.lld -T %t.script %t.o -o /dev/null 2>&1 | FileCheck %s
+# RUN: ld.lld -T %t.script %t.o -o %t
+# RUN: llvm-objdump --mcpu=pwr10 --no-show-raw-insn -d %t | FileCheck %s
+# RUN: llvm-readelf -s %t | FileCheck %s --check-prefix=SYM
 
 # RUN: llvm-mc -filetype=obj -triple=powerpc64 %s -o %t.o
-# RUN: not ld.lld -T %t.script %t.o -o /dev/null 2>&1 | FileCheck %s
-
-# CHECK:      error: R2 save stub offset is out of range: -268501028 is not in [-33554432, 33554431]; references callee
-# CHECK-NEXT: >>> defined in {{.*}}.o
+# RUN: ld.lld -T %t.script %t.o -o %t
+# RUN: llvm-objdump --mcpu=pwr10 --no-show-raw-insn -d %t | FileCheck %s
+# RUN: llvm-readelf -s %t | FileCheck %s --check-prefix=SYM
+
+# SYM:      Symbol table '.symtab' contains 9 entries:
+# SYM:      1: 0000000010010000     0 NOTYPE  LOCAL  DEFAULT [<other: 0x20>]   1 callee
+# SYM-NEXT: 2: 0000000020020008     0 NOTYPE  LOCAL  DEFAULT [<other: 0x60>]   3 caller
+# SYM-NEXT: 3: 0000000010020008     0 NOTYPE  LOCAL  DEFAULT                  2 caller_close
+# SYM-NEXT: 4: 0000000520020008     0 NOTYPE  LOCAL  DEFAULT                  4 caller_far
+# SYM-NEXT: 5: 0000000520028038     0 NOTYPE  LOCAL  HIDDEN                   6 .TOC.
+# SYM-NEXT: 6: 0000000010020020     8 FUNC    LOCAL  DEFAULT                  2 __toc_save_callee
+# SYM-NEXT: 7: 0000000020020020    20 FUNC    LOCAL  DEFAULT                  3 __toc_save_callee
+# SYM-NEXT: 8: 0000000520020020    20 FUNC    LOCAL  DEFAULT                  4 __toc_save_callee
 
-# RUN: ld.lld -T %t.script %t.o -o /dev/null --noinhibit-exec
 
+# CHECK-LABEL: <callee>:
+# CHECK:         blr
 .section .text_callee, "ax", %progbits
 callee:
   .localentry callee, 1
   blr
 
-.section .text_caller, "ax", %progbits
-caller:
+# CHECK-LABEL: <caller_close>:
+# CHECK:         bl 0x10020020
+# CHECK-NEXT:    ld 2, 24(1)
+# CHECK-NEXT:    blr
+# CHECK-LABEL: <__toc_save_callee>:
+# CHECK-LABEL:   std 2, 24(1)
+# CHECK-LABEL:   b 0x10010000
+.section .text_caller_close, "ax", %progbits
+.Lfunc_toc1:
+  .quad .TOC.-.Lfunc_gep1
+caller_close:
 .Lfunc_gep1:
-  addis 2, 12, .TOC.-.Lfunc_gep1@ha
-  addi 2, 2, .TOC.-.Lfunc_gep1@l
+  ld 2, .Lfunc_toc1-.Lfunc_gep1(12)
+  add 2, 2, 12
 .Lfunc_lep1:
   .localentry caller, .Lfunc_lep1-.Lfunc_gep1
-  addis 30, 2, global@toc@ha
-  lwz 3, global@toc@l(30)
   bl callee
   nop
   blr
-global:
-  .long 0
+
+# CHECK-LABEL: <caller>:
+# CHECK:         bl 0x20020020
+# CHECK-NEXT:    ld 2, 24(1)
+# CHECK-NEXT:    blr
+# CHECK-LABEL: <__toc_save_callee>:
+# CHECK:         std 2, 24(1)
+# CHECK-NEXT:    paddi 12, 0, -268501028, 1
+# CHECK-NEXT:    mtctr 12
+# CHECK-NEXT:    bctr
+.section .text_caller, "ax", %progbits
+.Lfunc_toc2:
+  .quad .TOC.-.Lfunc_gep2
+caller:
+.Lfunc_gep2:
+  ld 2, .Lfunc_toc2-.Lfunc_gep2(12)
+  add 2, 2, 12
+.Lfunc_lep2:
+  .localentry caller, .Lfunc_lep2-.Lfunc_gep2
+  bl callee
+  nop
+  blr
+
+# CHECK-LABEL: <caller_far>:
+# CHECK:         ld 2, -8(12)
+# CHECK-NEXT:    add 2, 2, 12
+# CHECK-NEXT:    bl 0x520020020
+# CHECK-NEXT:    ld 2, 24(1)
+# CHECK-NEXT:    blr
+# CHECK-LABEL: <__toc_save_callee>:
+# CHECK:         std 2, 24(1)
+# CHECK-NEXT:    addis 12, 2, 0
+# CHECK-NEXT:    ld 12, -32760(12)
+# CHECK-NEXT:    mtctr 12
+# CHECK-NEXT:    bctr
+.section .text_caller_far, "ax", %progbits
+.Lfunc_toc3:
+  .quad .TOC.-.Lfunc_gep3
+caller_far:
+.Lfunc_gep3:
+  ld 2, .Lfunc_toc3-.Lfunc_gep3(12)
+  add 2, 2, 12
+.Lfunc_lep3:
+  .localentry caller, .Lfunc_lep3-.Lfunc_gep3
+  bl callee
+  nop
+  blr