Re: [PATCH] Fix stepping in rtld without debug symbol

[Kevin Buettner <kevinb@redhat.com>]

On Fri, 4 Nov 2022 18:26:34 +0000
Lancelot SIX <lancelot.six@amd.com> wrote:

> Commit be6276e0aed "Allow debugging of runtime loader / dynamic linker"
> introduced a small regression when stepping into the runtime loader /
> dynamic linker from function we do not have debug information for.  This
> is reported in PR/29747.
> 
> This can be shown by the following example (given by Simon Marchi in
> buzilla bug report):
> 
>     $ cat test.c
>     #include <stdio.h>
> 
>     int main()
>     {
>       printf("Hi\n");
>       return 0;
>     }
>     $ gcc test.c -O0 -o test
>     $ ./gdb -q -nx --data-directory=data-directory test -ex start -ex s
>     Reading symbols from test...
>     (No debugging symbols found in test)
>     Temporary breakpoint 1 at 0x1151
>     Starting program: .../binutils-gdb/gdb/test
>     [Thread debugging using libthread_db enabled]
>     Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
> 
>     Temporary breakpoint 1, 0x0000555555555151 in main ()
>     Single stepping until exit from function main,
>     which has no line number information.
>     /home/smarchi/src/binutils-gdb/gdb/infrun.c:6960:64: runtime error: member call on null pointer of type 'struct symbol'
> 
>     The crash happens here:
> 
>     #0  __sanitizer::Die () at ../../../../src/libsanitizer/sanitizer_common/sanitizer_termination.cpp:50
>     #1  0x00007ffff5dd7128 in __ubsan::__ubsan_handle_type_mismatch_v1_abort (Data=<optimized out>, Pointer=<optimized out>) at ../../../../src/libsanitizer/ubsan/ubsan_handlers.cpp:148
>     #2  0x000055556183e1a7 in process_event_stop_test (ecs=0x7fffffffccd0) at .../binutils-gdb/gdb/infrun.c:6960
>     #3  0x0000555561838ea4 in handle_signal_stop (ecs=0x7fffffffccd0) at .../binutils-gdb/gdb/infrun.c:6615
>     #4  0x000055556182f77b in handle_inferior_event (ecs=0x7fffffffccd0) at .../binutils-gdb/gdb/infrun.c:5866
> 
> When evaluating:
> 
>     6956   if (execution_direction != EXEC_REVERSE
>     6957       && ecs->event_thread->control.step_over_calls == STEP_OVER_UNDEBUGGABLE
>     6958       && in_solib_dynsym_resolve_code (ecs->event_thread->stop_pc ())
>     6959       && !in_solib_dynsym_resolve_code (
>     6961          ecs->event_thread->control.step_start_function->value_block ()
>     6962              ->entry_pc ()))
> 
> we dereference, ecs->event_thread->control.step_start_function which is
> nullptr.
> 
> This patch changes this condition so it evaluates to true if
> ecs->event_thread->control.step_start_function is nullptr since this
> matches the behaviour before be6276e0aed.

Thanks for fixing this!

I see that Bruno has asked for the test case to be tweaked and your
reply that you've fixed it in your local sources.  That being the
case...

Approved-By: Kevin Buettner <kevinb@redhat.com>

Kevin