[llvm-objdump] Add symbol 'O' for object data

[Higuoxing]

Summary

When dumping elf object's symbol table info, gnu-objdump will give:

SYMBOL TABLE:
0000000000001004 l     F .text	0000000000000000 lfoo
0000000000001008 l     O .text	0000000000000000 lbar
0000000000001004 g     F .text	0000000000000000 foo
0000000000001008 g     O .text	0000000000000000 bar

llvm-objdump will give:

0000000000001004 l     F .text		 00000000 lfoo
0000000000001008 l       .text		 00000000 lbar
0000000000001004 g     F .text		 00000000 foo
0000000000001008 g       .text		 00000000 bar

To pass llvm-objdump tests, I have to change some tests for Mach-O and Wasm which I am not so familiar with ... Is this ok to apply these changes? Because gnu-objdump display different info on Mach-O, compared with llvm-objdump

[MaskRay]

test/tools/llvm-objdump/common-symbol-elf.test

3

GNU objdump does not print the g flag because in binutils-gdb/bfd/syms.c:bfd_print_symbol_vandf, a common symbol is neither BSF_GLOBAL nor BSF_LOCAL (though a common symbol is usually STB_GLOBAL)... I consider that a quirk that we do not necessarily replicate the behavior. Also see some discussion in https://reviews.llvm.org/D53782

[Higuoxing]

test/tools/llvm-objdump/common-symbol-elf.test

3

Thanks a lot : )

[kristina]

LGTM with minor nitpick, wouldn't using CHECK-NEXT in the Mach-O test make more sense like in the other tests?

Also, do you need it committed?

[Higuoxing]

In D54380#1294616, @kristina wrote:

LGTM with minor nitpick, wouldn't using CHECK-NEXT in the Mach-O test make more sense like in the other tests?

Thanks a lot, I will update it later.

Also, do you need it committed?

Yes, please

[Higuoxing]

Update test cases

[kristina]

In D54380#1294618, @Higuoxing wrote:

In D54380#1294616, @kristina wrote:

LGTM with minor nitpick, wouldn't using CHECK-NEXT in the Mach-O test make more sense like in the other tests?

Thanks a lot, I will update it later.

Also, do you need it committed?

Yes, please

On it, moment.

[kristina]

This is causing some LLVM and LLD tests to fail on buildbots, since they rely on the old format.

I'll try to correct them myself, the culprits are:

Failing Tests (3):
    LLVM :: MC/ELF/common-redeclare.s
    LLVM :: Object/objdump-symbol-table.test
    lld :: ELF/relocation-common.s

[kristina]

rL346611 for LLVM testsuite, and rLLD346612 for LLD testsuite should fix tests affected by this. Will close when/if llvm-clang-lld-x86_64-scei-ps4-ubuntu-fast is happy.

[kristina]

rLLD346613 fixes MIPS LLD tests. Looking at build console, there is a bunch of other target specific tests that need updates after this.

[kristina]

About to fail again:

• Hexagon: llvm/test/MC/Hexagon/lcomm.s, MC/Hexagon/common-redeclare.s
• Wasm: llvm/test/MC/WebAssembly/weak-alias.ll

Updating those two and that should be the last of it.

[kristina]

Last fixes in rL346614 and rL346617. Big sorry for the buildbot storm, I should have reverted it after the first set of failures and dissected it after, or better yet grepped for objdump usage by other targets when reviewing.

[kristina]

Grouped and linked all related commits. Closing.

[Higuoxing]

In D54380#1294716, @kristina wrote:

Grouped and linked all related commits. Closing.

Sorry for that, I should have run 'test all' locally ... Sorry for bothering you too much! Thanks a lot!

[kristina]

In D54380#1294974, @Higuoxing wrote:

In D54380#1294716, @kristina wrote:

Grouped and linked all related commits. Closing.

Sorry for that, I should have run 'test all' locally ... Sorry for bothering you too much! Thanks a lot!

Don't worry, it's not your fault, there were a lot of targets that used llvm-objdump in their tests
that are unlikely to be built by most locally (ie. Hexagon). I guess at the core a lesson for the future
is to grep around for uses of said tool in tests since without the target enabled, you will not be
eligible to run the test and while the majority of the tests work on X86 (which is usually enabled by
default for almost everyone), target specific tests often require said target backend being enabled.

test/tools/llvm-objdump/WebAssembly/symbol-table.test

 RUN: llvm-objdump -t %p/../Inputs/trivial.obj.wasm | FileCheck %s
 
 CHECK:      SYMBOL TABLE:
 CHECK-NEXT: 00000002 g     F CODE main
-CHECK-NEXT: 00000000 l       DATA .L.str
+CHECK-NEXT: 00000000 l     O DATA .L.str
 CHECK-NEXT: 00000000 g     F *UND*  puts
 CHECK-NEXT: 00000003 l     F CODE .LSomeOtherFunction_bitcast
 CHECK-NEXT: 00000000 g     F *UND*  SomeOtherFunction
-CHECK-NEXT: 00000010 g       DATA var
+CHECK-NEXT: 00000010 g     O DATA var

test/tools/llvm-objdump/X86/macho-symbol-table.test

 RUN: llvm-objdump -macho -t %p/Inputs/hello.obj.macho-x86_64 | FileCheck %s
 
 CHECK: SYMBOL TABLE:
-CHECK: 000000000000003b l       __TEXT,__cstring  L_.str
-CHECK: 0000000000000068 l       __TEXT,__eh_frame EH_frame0
+CHECK: 000000000000003b l     O __TEXT,__cstring  L_.str
+CHECK: 0000000000000068 l     O __TEXT,__eh_frame EH_frame0
 CHECK: 0000000000000000 g     F __TEXT,__text         _main
-CHECK: 0000000000000080 g       __TEXT,__eh_frame _main.eh
+CHECK: 0000000000000080 g     O __TEXT,__eh_frame _main.eh
 CHECK: 0000000000000000         *UND*                 _printf

test/tools/llvm-objdump/common-symbol-elf.test

 // RUN: llvm-objdump -t %p/Inputs/common-symbol-elf | FileCheck %s
 
-CHECK: 00000400 g       *COM*  00000008 common_symbol
+CHECK: 00000400 g     O *COM*  00000008 common_symbol

test/tools/llvm-objdump/symbol-table-elf.test

 # RUN: yaml2obj %s > %t
 # RUN: llvm-objdump --syms %t | FileCheck %s
 # RUN: llvm-objdump -t     %t | FileCheck %s
 
 # CHECK:      SYMBOL TABLE:
 # CHECK-NEXT: 0000000000000000         *UND*     00000000
 # CHECK-NEXT: 0000000000001004 l     F .text     00000000 lfoo
-# CHECK-NEXT: 0000000000001008 l       .text     00000000 lbar
+# CHECK-NEXT: 0000000000001008 l     O .text     00000000 lbar
 # CHECK-NEXT: 0000000000001004 g     F .text     00000000 foo
-# CHECK-NEXT: 0000000000001008 g       .text     00000000 bar
+# CHECK-NEXT: 0000000000001008 g     O .text     00000000 bar
 
 !ELF
 FileHeader:
   Class:           ELFCLASS64
   Data:            ELFDATA2LSB
   Type:            ET_EXEC
   Machine:         EM_X86_64
 Sections:

tools/llvm-objdump/llvm-objdump.cpp

if (Type != SymbolRef::ST_Unknown)
       GlobLoc = Global ? 'g' : 'l';
     char Debug = (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File)
                  ? 'd' : ' ';
     char FileFunc = ' ';
     if (Type == SymbolRef::ST_File)
       FileFunc = 'f';
     else if (Type == SymbolRef::ST_Function)
       FileFunc = 'F';
+    else if (Type == SymbolRef::ST_Data)
+      FileFunc = 'O';
 
     const char *Fmt = o->getBytesInAddress() > 4 ? "%016" PRIx64 :
                                                    "%08" PRIx64;
 
     outs() << format(Fmt, Address) << " "
            << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' '
            << (Weak ? 'w' : ' ') // Weak?
            << ' ' // Constructor. Not supported yet.