[LLD][PowerPC][test] Update thunk range error report for PPC64PCRelLongBranchThunk

[NeHuang]

Summary

Update thunk range error report for PPC64PCRelLongBranchThunk after re-basing the branch with https://reviews.llvm.org/D87486
Add a range error test for PPC64R12SetupStub

[MaskRay]

lld/ELF/Thunks.cpp

913

We have an existing diagnostic like

error: relocation R_AARCH64_ABS32 cannot be used against symbol hidden; recompile with -fPIC

[MaskRay]

lld/test/ELF/ppc64-pcrel-call-to-extern-error.s

1 ↗

(On Diff #290739)

Delete linux

9 ↗

(On Diff #290739)

If a linked .so is used by another link, add -soname to avoid a gotcha

See commit c2a5459d52b94a50581dec7a542ef00915b8d2b2

[NeHuang]

lld/ELF/Thunks.cpp

913

Just want to clarify, are you suggesting to follow the same error message print format as error: relocation R_AARCH64_ABS32 cannot be used against symbol hidden; recompile with -fPIC ?

lld/test/ELF/ppc64-pcrel-call-to-extern-error.s

1 ↗

(On Diff #290739)

We want to keep the linux to run the test on ppc-linux only since writing to an allocated file will cause time error for this case on freebsd bots. Will add some comments in the tests

9 ↗

(On Diff #290739)

Will do.

[MaskRay]

lld/ELF/Thunks.cpp

913

I mean to make the diagnostic closer with existing styles if possible.

For this particular one, the second sentence is probably "; recompile with -mcmodel=large"

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

34

.size is insignificant and can be deleted.

[MaskRay]

Created D87486 to improve the error messages.

[NeHuang]

Addressed review comments to

• Change the error message
• Change the LIT test accordingly

[NeHuang]

lld/ELF/Thunks.cpp

913

Thanks. Based on Sean's comment in https://reviews.llvm.org/D86706, I would change the second sentence to "; recompile using the large code model".

[NeHuang]

Rebased with ToT to integrate with the changes in https://reviews.llvm.org/D87486

[NeHuang]

Clang-format

[MaskRay]

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

2

Delete system-linux

I think this test is portable and runs on other systems.

27

Consider deleting unneeded setup instructions

[nemanjai]

This LGTM but of course, wait to hear from @MaskRay because he had test case concerns.

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

2

It appears that the file systems on FreeBSD and Linux behave differently. On Linux, writing the large file (8GB in this case) happens in a fraction of a second which suggests that it is not actually writing the bytes to storage. However, on FreeBSD this takes a significant amount of time - perhaps it is actually writing the bytes to storage. So we limit these test cases that produce large files to Linux to avoid this problem.

The issue can be avoided on FreeBSD by writing out to /dev/null or some other special file because that won't mmap - it creates a memory buffer. However, that causes a different problem on systems that limit virtual memory - the OS won't allow the process to allocate a memory buffer that size.

So the easiest solution we found was to write to a real file and limit this to Linux.

[NeHuang]

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

27

I could remove them but it would be better to keep them because the asm doesn't really make sense for a toc-based callee without the toc setup. Of course, I can remove them if you insist.

[NeHuang]

Ping.

[sfertile]

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

2

Thanks for the thorough explanation.
Really minor nit: can we have the comment that explains this right after the requires directive instead of after the linker script?

[MaskRay]

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

2

In this case, PHDRS and :segment should be specified to assign sections to different text segments.

[NeHuang]

Move up the comment in the test case.

[sfertile]

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

2

This is outside my wheelhouse, so I have to ask for clarification: The test can use PHDRS command to create multiple small load segments, then in the SECTIONS command we would have 2 lines like:

.output_section_name  { input_sections_to_match } : load_segment_name

The first to match callers section, then increase the address sufficiently far away and have a second similar line for the callees section?

[MaskRay]

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

2

See D88037 for an example

[sfertile]

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

2

Nice, thanks!

[NeHuang]

Thanks @MaskRay and @sfertile Test cases have been updated with PHDRS to assign sections to different text segments.

[MaskRay]

Thanks!

lld/ELF/Thunks.cpp

void PPC64R2SaveStub::addSymbols(ThunkSection &isec) {
   Defined *s = addSymbol(saver.save("__toc_save_" + destination.getName()),
                          STT_FUNC, 0, isec);
   s->needsTocRestore = true;
 }
 
 void PPC64R12SetupStub::writeTo(uint8_t *buf) {
   int64_t offset = destination.getVA() - getThunkTargetSym()->getVA();
   if (!isInt<34>(offset))
     reportRangeError(buf, offset, 34, destination, "R12 setup stub offset");
   uint64_t paddi = PADDI_R12_NO_DISP | (((offset >> 16) & 0x3ffff) << 32) |
                    (offset & 0xffff);
 
   writePrefixedInstruction(buf + 0, paddi); // paddi r12, 0, func@pcrel, 1
   write32(buf + 8, MTCTR_R12);              // mtctr r12
   write32(buf + 12, BCTR);                  // bctr
 }
 

 bool PPC64LongBranchThunk::isCompatibleWith(const InputSection &isec,
                                             const Relocation &rel) const {
   return rel.type == R_PPC64_REL24 || rel.type == R_PPC64_REL14;
 }
 
 void PPC64PCRelLongBranchThunk::writeTo(uint8_t *buf) {
   int64_t offset = destination.getVA() - getThunkTargetSym()->getVA();
   if (!isInt<34>(offset))
-    fatal("offset overflow 34 bits, please compile using the large code model");
+    reportRangeError(buf, offset, 34, destination,
+                     "PC-relative long branch stub offset");
   uint64_t paddi = PADDI_R12_NO_DISP | (((offset >> 16) & 0x3ffff) << 32) |
                    (offset & 0xffff);
 
   writePrefixedInstruction(buf + 0, paddi); // paddi r12, 0, func@pcrel, 1
   write32(buf + 8, MTCTR_R12);              // mtctr r12
   write32(buf + 12, BCTR);                  // bctr
 }
 

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

+# REQUIRES: ppc
+# RUN: split-file %s %t
+# RUN: llvm-mc -filetype=obj -triple=ppc64le %t/asm -o %t.o
+# RUN: not ld.lld --script %t/lds %t.o -o %t1 2>&1 | FileCheck %s
+# RUN: ld.lld --script %t/lds %t.o -o %t1 --noinhibit-exec
+# RUN: rm %t.o %t1
+
+# RUN: llvm-mc -filetype=obj -triple=ppc64 %t/asm -o %t.o
+# RUN: not ld.lld --script %t/lds %t.o -o %t1 2>&1 | FileCheck %s
+# RUN: ld.lld --script %t/lds %t.o -o %t1 --noinhibit-exec
+# RUN: rm %t.o %t1
+
+# CHECK: error: R12 setup stub offset is out of range: 8589934592 is not in [-8589934592, 8589934591]; references callee
+# CHECK-NEXT: >>> defined in {{.*}}.o
+
+//--- asm
+.section .text_high, "ax", %progbits
+callee:
+  .Lfunc_gep1:
+  addis 2, 12, .TOC.-.Lfunc_gep1@ha
+  addi 2, 2, .TOC.-.Lfunc_gep1@l
+  .Lfunc_lep1:
+  .localentry callee, .Lfunc_lep1-.Lfunc_gep1
+  addis 4, 2, global@toc@ha
+  lwz 4, global@toc@l(4)
+  blr
+
+.section .text_low, "ax", %progbits
+caller:
+  .localentry caller, 1
+  bl callee@notoc
+  blr
+global:
+  .long 0
+
+//--- lds
+PHDRS {
+  low PT_LOAD FLAGS(0x1 | 0x4);
+  high PT_LOAD FLAGS(0x1 | 0x4);
+}
+SECTIONS {
+  .text_low 0x2000 : { *(.text_low) } :low
+  .text_high 0x200002010 : { *(.text_high) } :high
+}

lld/test/ELF/ppc64-pcrel-long-branch-error.s

-# REQUIRES: ppc, system-linux
-# RUN: echo 'SECTIONS { \
-# RUN:       .text_low 0x2000: { *(.text_low) } \
-# RUN:       .text_high 0x200002010 : { *(.text_high) } \
-# RUN:       }' > %t.script
+# REQUIRES: ppc
+# RUN: split-file %s %t
+# RUN: llvm-mc -filetype=obj -triple=ppc64le %t/asm -o %t.o
+# RUN: not ld.lld --script %t/lds %t.o -o %t1 2>&1 | FileCheck %s
+# RUN: ld.lld --script %t/lds %t.o -o %t1 --noinhibit-exec
+# RUN: rm %t.o %t1
+# RUN: llvm-mc -filetype=obj -triple=ppc64le -defsym HIDDEN=1 %t/asm -o %t.o
+# RUN: not ld.lld -shared --script %t/lds %t.o -o %t1 2>&1 | FileCheck %s
+# RUN: ld.lld -shared --script %t/lds %t.o -o %t1 --noinhibit-exec
+# RUN: rm %t.o %t1
+
+# RUN: llvm-mc -filetype=obj -triple=ppc64 %t/asm -o %t.o
+# RUN: not ld.lld --script %t/lds %t.o -o %t1 2>&1 | FileCheck %s
+# RUN: ld.lld --script %t/lds %t.o -o %t1 --noinhibit-exec
+# RUN: rm %t.o %t1
+# RUN: llvm-mc -filetype=obj -triple=ppc64 -defsym HIDDEN=1 %t/asm -o %t.o
+# RUN: not ld.lld -shared --script %t/lds %t.o -o %t1 2>&1 | FileCheck %s
+# RUN: ld.lld -shared --script %t/lds %t.o -o %t1 --noinhibit-exec
+# RUN: rm %t.o %t1
 
-## In this test, we do not use -o /dev/null like other similar cases do since
-## it will fail in some enviroments with out-of-memory errors associated with
-## buffering the output in memeory. The test is enabled for ppc linux only since
-## writing to an allocated file will cause time out error for this case on freebsd.
-
-# RUN: llvm-mc -filetype=obj -triple=ppc64le %s -o %t.o
-# RUN: not ld.lld -T %t.script %t.o -o %t 2>&1 | FileCheck %s
-# RUN: llvm-mc -filetype=obj -triple=ppc64le -defsym HIDDEN=1 %s -o %t.o
-# RUN: not ld.lld -shared -T %t.script %t.o -o %t 2>&1 | FileCheck %s
-
-# RUN: llvm-mc -filetype=obj -triple=ppc64 %s -o %t.o
-# RUN: not ld.lld -T %t.script %t.o -o %t 2>&1 | FileCheck %s
-# RUN: llvm-mc -filetype=obj -triple=ppc64 -defsym HIDDEN=1 %s -o %t.o
-# RUN: not ld.lld -shared -T %t.script %t.o -o %t 2>&1 | FileCheck %s
-
-# CHECK: error: offset overflow 34 bits, please compile using the large code model
+# CHECK: error: PC-relative long branch stub offset is out of range: 8589934592 is not in [-8589934592, 8589934591]; references high
+# CHECK-NEXT: >>> defined in {{.*}}.o
 
+//--- asm
 .section .text_low, "ax", %progbits
 .globl _start
 _start:
   bl high@notoc
   blr
 
 .section .text_high, "ax", %progbits
 .ifdef HIDDEN
 .hidden high
 .endif
 .globl high
 high:
   blr
+
+//--- lds
+PHDRS {
+  low PT_LOAD FLAGS(0x1 | 0x4);
+  high PT_LOAD FLAGS(0x1 | 0x4);
+}
+SECTIONS {
+  .text_low 0x2000 : { *(.text_low) } :low
+  .text_high 0x200002010 : { *(.text_high) } :high
+}