[LLD][PowerPC] Add support for R_PPC64_TPREL34 used in TLS Local Exec

[stefanp]

Summary

Add Thread Local Storage Local Exec support to LLD. This is to support PC Relative addressing of Local Exec.
The patch teaches LLD to handle:

paddi r9, r13, x1@tprel

The relocation is:

R_PPC_TPREL34

[NeHuang]

lld/test/ELF/ppc64-tls-pcrel-le.s

4

Do we need the address set up for this case? Seems we are not checking/verifying the offset using in the paddi to access x, y and z

55

what is the test purpose of adding symbol y, it seems identical as the case for symbol x?

[stefanp]

Removed the linker script that was not needed.
Added a quick summary of the test case.

lld/test/ELF/ppc64-tls-pcrel-le.s

4

That is a good point. All of the paddi instructions are not PC Relative in this case, they are all TP relative. Therefore the position of the paddi doesn't matter. I'll remove the sections.

55

I know that x and y are pretty much the same thing. The reason why I added the y was because I wanted a variable that was not just at TP - 0x7000. I wanted to check that we got the correct offset based on the size of x. Technically z could also serve this purpose but I would prefer to keep all three variables to keep what each one is testing separate.

[NeHuang]

lld/test/ELF/ppc64-tls-pcrel-le.s

18

SYMBOL-NEXT

19

SYMBOL-NEXT

21

<LocalExecAddrX>: and same for the CHECK-LABEL cases below

[NeHuang]

LGTM. I only have some comments on the nits.

[MaskRay]

[LLD][PowerPC] Add support for TLS Local Exec

Local Exec support exists, so this subject may be a bit confusing. This patch is about R_PPC64_TPREL34.

lld/test/ELF/ppc64-tls-pcrel-le.s

78

You can remove .size directives since they are not significant

[stefanp]

Fixed a couple of things on the test case.
Added the function name inside <>:.
Removed the .size directives.
Added SYMBOL-NEXT: where possible.

[MaskRay]

Reading...

[MaskRay]

I need to apply your patch with curl -L 'https://reviews.llvm.org/D86608?download=1' | patch -p0 (for most other reviews I use -p1. arc diff uploaded patch can be applied with -p1

https://llvm.org/docs/GettingStarted.html#sending-patches

[MaskRay]

lld/test/ELF/ppc64-tls-pcrel-le.s

21

Consider simplifying the test

# CHECK-LABEL: <LocalExecAddr>:
# CHECK-NEXT:    paddi 3, 13, -28672, 0
# CHECK-NEXT:    paddi 3, 13, -28668, 0
# CHECK-NEXT:    

LocalExecAddr:
  paddi 3, 13, x@TPREL, 0
  paddi 3, 13, y@TPREL, 0
  paddi 3, 13, z@TPREL+12, 0

# CHECK-LABEL: <LocalExecVal>:
# CHECK-NEXT:    ...
# CHECK-NEXT:    ...

LocalExecVal:
  paddi 3, 13, x@TPREL, 0
  lwz 3, 0(3)
  paddi 3, 13, y@TPREL, 0
  lwz 3, 0(3)
  paddi 3, 13, z@TPREL+12, 0
  lwz 3, 0(3)

[stefanp]

Thank you for the reviews!

I'm sorry about not consistently using arc. I sometimes just create a patch and upload that. I'll try to use arc more consistently for Phabricator.

I've updated the test case to simplify it a little.
I'll wait until D83404 is in and then I'll commit this patch.

lld/ELF/Arch/PPC64.cpp

RelExpr PPC64::getRelExpr(RelType type, const Symbol &s,
   case R_PPC64_TPREL16_LO:
   case R_PPC64_TPREL16_HI:
   case R_PPC64_TPREL16_DS:
   case R_PPC64_TPREL16_LO_DS:
   case R_PPC64_TPREL16_HIGHER:
   case R_PPC64_TPREL16_HIGHERA:
   case R_PPC64_TPREL16_HIGHEST:
   case R_PPC64_TPREL16_HIGHESTA:
+  case R_PPC64_TPREL34:
     return R_TLS;
   case R_PPC64_DTPREL16:
   case R_PPC64_DTPREL16_DS:
   case R_PPC64_DTPREL16_HA:
   case R_PPC64_DTPREL16_HI:
   case R_PPC64_DTPREL16_HIGHER:
   case R_PPC64_DTPREL16_HIGHERA:
   case R_PPC64_DTPREL16_HIGHEST:
case R_PPC64_PCREL34: {
     const uint64_t fullMask = 0x0003ffff0000ffff;
     checkInt(loc, val, 34, rel);
 
     uint64_t instr = readPrefixedInstruction(loc) & ~fullMask;
     writePrefixedInstruction(loc, instr | ((val & si0Mask) << 16) |
                              (val & si1Mask));
     break;
   }
-  case R_PPC64_GOT_PCREL34: {
+  case R_PPC64_GOT_PCREL34:
+  case R_PPC64_TPREL34: {
     const uint64_t si0Mask = 0x00000003ffff0000;
     const uint64_t si1Mask = 0x000000000000ffff;
     const uint64_t fullMask = 0x0003ffff0000ffff;
     checkInt(loc, val, 34, rel);
 
     uint64_t instr = readPrefixedInstruction(loc) & ~fullMask;
     writePrefixedInstruction(loc, instr | ((val & si0Mask) << 16) |
                              (val & si1Mask));

lld/test/ELF/ppc64-tls-pcrel-le.s

+# REQUIRES: ppc
+# RUN: llvm-mc -filetype=obj -triple=powerpc64le %s -o %t.o
+# RUN: ld.lld %t.o -o %t
+# RUN: llvm-readelf -s %t | FileCheck %s --check-prefix=SYMBOL
+# RUN: llvm-objdump -d --no-show-raw-insn --mcpu=pwr10 %t | FileCheck %s
+
+# RUN: llvm-mc -filetype=obj -triple=powerpc64 %s -o %t.o
+# RUN: ld.lld %t.o -o %t
+# RUN: llvm-readelf -s %t | FileCheck %s --check-prefix=SYMBOL
+# RUN: llvm-objdump -d --no-show-raw-insn --mcpu=pwr10 %t | FileCheck %s
+
+## This test checks the LLD implementation of the Local Exec TLS model
+## when using prefixed instructions like paddi.
+
+# SYMBOL:      Symbol table '.symtab' contains 6 entries:
+# SYMBOL:      3: 0000000000000000     0 TLS     LOCAL DEFAULT     2 x
+# SYMBOL-NEXT: 4: 0000000000000004     0 TLS     LOCAL DEFAULT     2 y
+# SYMBOL-NEXT: 5: 0000000000000008     0 TLS     LOCAL DEFAULT     2 z
+
+# CHECK-LABEL: <LocalExecAddr>:
+# CHECK:       paddi 3, 13, -28672, 0
+# CHECK-NEXT:  paddi 3, 13, -28668, 0
+# CHECK-NEXT:  paddi 3, 13, -28652, 0
+# CHECK-NEXT:  blr
+
+# CHECK-LABEL: <LocalExecVal>:
+# CHECK:       paddi 3, 13, -28672, 0
+# CHECK-NEXT:  lwz 3, 0(3)
+# CHECK-NEXT:  paddi 3, 13, -28668, 0
+# CHECK-NEXT:  lwz 3, 0(3)
+# CHECK-NEXT:  paddi 3, 13, -28652, 0
+# CHECK-NEXT:  lwz 3, 0(3)
+# CHECK-NEXT:  blr
+
+LocalExecAddr:
+  paddi 3, 13, x@TPREL, 0
+  paddi 3, 13, y@TPREL, 0
+  paddi 3, 13, z@TPREL+12, 0
+  blr
+
+LocalExecVal:
+  paddi 3, 13, x@TPREL, 0
+  lwz 3, 0(3)
+  paddi 3, 13, y@TPREL, 0
+  lwz 3, 0(3)
+  paddi 3, 13, z@TPREL+12, 0
+  lwz 3, 0(3)
+  blr
+
+.section .tbss, "awT", @nobits
+x:
+  .long 0
+y:
+  .long 0
+z:
+  .space  20