[PPC64] Add offset to local entry point when calling functions without plt

[syzaara]

Summary

PPC64 V2 ABI describes two entry points to a function. The global entry point sets up the TOC base pointer. When calling a local function, the call should branch to the local entry point rather than the global entry point. Section 3.4.1 describes using the 3 most significant bits of the st_other field to find out how many instructions there are between the local and global entry point. This patch adds the correct offset required to branch to the local entry point of a function.

[ruiu]

lld/ELF/Arch/PPC64.cpp

25–26 ↗

(On Diff #142804)

Can you please expand the comment to include the explanation that you wrote in the commit message? I found that's very useful.

35 ↗

(On Diff #142804)

nit: no else after return

lld/ELF/InputSection.cpp

591–603

Perhaps, it is better to inline the function here? I think that getLocalEntryOffset is called only once here and short enough to be inlined.

[syzaara]

@ruiu Hi Rui, is this okay to approve now?

[ruiu]

Apologies, I missed this patch. LGTM.

[grimar]

ELF/InputSection.cpp

598 ↗

(On Diff #144351)

This produces a warning for me now:
InputSection.cpp(598): warning C4334: '<<': result of 32-bit shift implicitly converted to 64 bits (was 64-bit shift intended?)

[syzaara]

ELF/InputSection.cpp

598 ↗

(On Diff #144351)

Should be fixed with https://reviews.llvm.org/rL331181

[grimar]

ELF/InputSection.cpp

598 ↗

(On Diff #144351)

It`s fixed for me, thanks!

lld/ELF/InputSection.cpp

if (Out::Opd) {
       // If this is a local call, and we currently have the address of a
       // function-descriptor, get the underlying code address instead.
       uint64_t OpdStart = Out::Opd->Addr;
       uint64_t OpdEnd = OpdStart + Out::Opd->Size;
       bool InOpd = OpdStart <= SymVA && SymVA < OpdEnd;
       if (InOpd)
         SymVA = read64be(&Out::OpdBuf[SymVA - OpdStart]);
     }
+
+    // PPC64 V2 ABI describes two entry points to a function. The global entry
+    // point sets up the TOC base pointer. When calling a local function, the
+    // call should branch to the local entry point rather than the global entry
+    // point. Section 3.4.1 describes using the 3 most significant bits of the
+    // st_other field to find out how many instructions there are between the
+    // local and global entry point.
+    uint8_t StOther = (Sym.StOther >> 5) & 7;
+    if (StOther == 0 || StOther == 1)
       return SymVA - P;
+
+    return SymVA - P + (1 << StOther);
   }
   case R_PPC_TOC:
     return getPPC64TocBase() + A;
   case R_RELAX_GOT_PC:
     return Sym.getVA(A) - P;
   case R_RELAX_TLS_GD_TO_LE:
   case R_RELAX_TLS_IE_TO_LE:
   case R_RELAX_TLS_LD_TO_LE:
   case R_TLS:

lld/test/ELF/Inputs/ppc64-func-global-entry.s

+  .text
+  .abiversion 2
+  .globl  foo_external_diff       # -- Begin function foo_external_diff
+  .p2align  4
+  .type foo_external_diff,@function
+foo_external_diff:                      # @foo_external_diff
+.Lfunc_begin0:
+.Lfunc_gep0:
+  addis 2, 12, .TOC.-.Lfunc_gep0@ha
+  addi 2, 2, .TOC.-.Lfunc_gep0@l
+.Lfunc_lep0:
+  .localentry foo_external_diff, .Lfunc_lep0-.Lfunc_gep0
+# %bb.0:                                # %entry
+  addis 5, 2, .LC0@toc@ha
+  add 3, 4, 3
+  ld 5, .LC0@toc@l(5)
+  lwz 5, 0(5)
+  add 3, 3, 5
+  extsw 3, 3
+  blr
+  .long 0
+  .quad 0
+.Lfunc_end0:
+  .size foo_external_diff, .Lfunc_end0-.Lfunc_begin0
+                                        # -- End function
+  .section  .toc,"aw",@progbits
+.LC0:
+  .tc glob2[TC],glob2
+  .type glob2,@object           # @glob2
+  .data
+  .globl  glob2
+  .p2align  2
+glob2:
+  .long 10                      # 0xa
+  .size glob2, 4

lld/test/ELF/Inputs/ppc64-func-local-entry.s

+  .text
+  .abiversion 2
+  .globl  foo_external_same       # -- Begin function foo_external_same
+  .p2align  4
+  .type foo_external_same,@function
+foo_external_same:                      # @foo_external_same
+.Lfunc_begin0:
+# %bb.0:                                # %entry
+  add 3, 4, 3
+  extsw 3, 3
+  blr
+  .long 0
+  .quad 0
+.Lfunc_end0:
+  .size foo_external_same, .Lfunc_end0-.Lfunc_begin0
+                                        # -- End function

lld/test/ELF/ppc64-func-entry-points.s

+// REQUIRES: ppc
+// RUN: llvm-mc -filetype=obj -triple=powerpc64le-unknown-linux %s -o %t.o
+// RUN: llvm-mc -filetype=obj -triple=powerpc64le-unknown-linux %p/Inputs/ppc64-func-global-entry.s -o %t2.o
+// RUN: llvm-mc -filetype=obj -triple=powerpc64le-unknown-linux %p/Inputs/ppc64-func-local-entry.s -o %t3.o
+// RUN: ld.lld -dynamic-linker /lib64/ld64.so.2 %t.o %t2.o %t3.o -o %t
+// RUN: llvm-objdump -d %t | FileCheck %s
+  .text
+  .abiversion 2
+  .globl  _start                    # -- Begin function _start
+  .p2align  4
+  .type _start,@function
+_start:                                   # @_start
+.Lfunc_begin0:
+.Lfunc_gep0:
+  addis 2, 12, .TOC.-.Lfunc_gep0@ha
+  addi 2, 2, .TOC.-.Lfunc_gep0@l
+.Lfunc_lep0:
+  .localentry _start, .Lfunc_lep0-.Lfunc_gep0
+# %bb.0:                                # %entry
+  mflr 0
+  std 0, 16(1)
+  stdu 1, -48(1)
+  li 3, 1
+  li 4, 1
+  std 30, 32(1)                   # 8-byte Folded Spill
+  bl foo_external_same
+  nop
+  mr 30, 3
+  li 3, 2
+  li 4, 2
+  bl foo_external_diff
+  nop
+  addis 4, 2, .LC0@toc@ha
+  add 3, 3, 30
+  ld 30, 32(1)                    # 8-byte Folded Reload
+  ld 4, .LC0@toc@l(4)
+  lwz 4, 0(4)
+  add 3, 3, 4
+  extsw 3, 3
+  addi 1, 1, 48
+  ld 0, 16(1)
+  li 0, 1
+  sc
+  .long 0
+  .quad 0
+.Lfunc_end0:
+  .size _start, .Lfunc_end0-.Lfunc_begin0
+                                        # -- End function
+  .section  .toc,"aw",@progbits
+.LC0:
+  .tc glob[TC],glob
+  .type glob,@object            # @glob
+  .data
+  .globl  glob
+  .p2align  2
+glob:
+  .long 10                      # 0xa
+  .size glob, 4
+
+# Check that foo_external_diff has a global entry point and we branch to
+# foo_external_diff+8. Also check that foo_external_same has no global entry
+# point and we branch to start of foo_external_same.
+
+// CHECK: _start:
+// CHECK: 10010020:       91 00 00 48     bl .+144
+// CHECK: 10010034:       55 00 00 48     bl .+84
+// CHECK: foo_external_diff:
+// CHECK-NEXT: 10010080:       02 00 4c 3c     addis 2, 12, 2
+// CHECK-NEXT: 10010084:       80 7f 42 38     addi 2, 2, 32640
+// CHECK-NEXT: 10010088:       ff ff a2 3c     addis 5, 2, -1
+// CHECK: foo_external_same:
+// CHECK-NEXT: 100100b0:       14 1a 64 7c     add 3, 4, 3