DW_OP_form_tls_address debug info not outputted for aarch64

[pinskia]

Take:

__thread unsigned long rage = 1; 

On x86_64 we get:

        .byte   14
        .quad   rage@DTPOFF

but that is not outputted for aarch64.

Corresponding GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97344
Corresponding binutils bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28351
Coresponding gdb bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27886

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: Andrew Pinski (pinskia)

Details

Take: ``` __thread unsigned long rage = 1; ```

On x86_64 we get:

        .byte   14
        .quad   rage@<!-- -->DTPOFF

but that is not outputted for aarch64.

Corresponding GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97344
Corresponding binutils bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28351
Coresponding gdb bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27886

[llvmbot]

@llvm/issue-subscribers-debuginfo

Author: Andrew Pinski (pinskia)

Details

Take: ``` __thread unsigned long rage = 1; ```

On x86_64 we get:

        .byte   14
        .quad   rage@<!-- -->DTPOFF

but that is not outputted for aarch64.

Corresponding GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97344
Corresponding binutils bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28351
Coresponding gdb bug: https://sourceware.org/bugzilla/show_bug.cgi?id=27886

[xgupta]

Hi @pinskia

What is the use of DW_OP_form_tls_address since AArch64 ELF ABI does not define a static relocation type for TLS offset, so compiler cannot generate valid DW_AT_location content involving DW_OP_form_tls_address for thread local variables like it is doing for X86. I think that is why getDebugThreadLocalSymbol is not extended for Aarch64.

[pinskia]

Hi @pinskia

What is the use of DW_OP_form_tls_address since AArch64 ELF ABI does not define a static relocation type for TLS offset, so compiler cannot generate valid DW_AT_location content involving DW_OP_form_tls_address for thread local variables like it is doing for X86. I think that is why getDebugThreadLocalSymbol is not extended for Aarch64.

Huh?

R_<CLS>_TLS_DTPREL is definitely defined. See https://github.com/ARM-software/abi-aa/blob/main/aaelf64/aaelf64.rst

R_<CLS>_TLS_TPREL is also defined.

[xgupta]

Thanks @pinskia, have misunderstood it. Raised a PR to resolve the issue - #146572. I tested the patch inside Aarch64 ubuntu docker image on macOs with GDB, LLDB seems to have some issue on my side probably because it is threads debugging and docker combo on mac.

[avikivity]

The program in the description can be debugged by gdb, so something changed.

However, the following C++ program cannot:

static thread_local unsigned long v = 1; 

int main() { return v; }

On x86, gdb says:

$ gdb a.out
...
(gdb) start
Temporary breakpoint 1 at 0x40043b: file thread-local-simple.cc, line 4.
Starting program: /home/avi/tests/a.out 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".

Temporary breakpoint 1, main () at thread-local-simple.cc:4
4	int main() { return v; }
(gdb) p v
$1 = 1

While on aarch64

$ gdb a.out
..
(gdb) start
...
Temporary breakpoint 1, main () at thread-local-simple.cc:3
3	int main() { return v; }
(gdb) p v
$1 = <optimized out>