Re: [PATCH v2 3/3] btrace, vdso: add vdso target sections

[Pedro Alves <palves at redhat dot com>]

Thanks.

On 05/20/2014 01:00 PM, Markus Metzger wrote:

> @@ -92,6 +92,7 @@ symbol_file_add_from_memory (struct bfd *templ, CORE_ADDR addr,
>    struct section_addr_info *sai;
>    unsigned int i;
>    struct cleanup *cleanup;
> +  struct target_section *sections, *sections_end, *tsec;
>  
>    if (bfd_get_flavour (templ) != bfd_target_elf_flavour)
>      error (_("add-symbol-file-from-memory not supported for this target"));
> @@ -131,6 +132,22 @@ symbol_file_add_from_memory (struct bfd *templ, CORE_ADDR addr,
>  				   from_tty ? SYMFILE_VERBOSE : 0,
>                                     sai, OBJF_SHARED, NULL);
>  
> +  sections = NULL;
> +  sections_end = NULL;
> +  make_cleanup (xfree, sections);

This will always xfree NULL.  You either want:

 make_cleanup (free_current_contents, &sections);

or move the cleanup to within the if/then block.  You also need
to make sure to discard the cleanup on success.

Or, better yet, not install a cleanup at all?  build_section_table
allocates the memory, and then the array is passed directly to
add_target_sections, which can be seen as a transfer of ownership.
If something throws after that call, we don't want the sections be
be simply xfree'd as that'd leave dangling pointers in the target
sections table.

> +
> +  if (build_section_table (nbfd, &sections, &sections_end) == 0)
> +    {
> +      /* Adjust the target section addresses by the load address.  */
> +      for (tsec = sections; tsec != sections_end; ++tsec)
> +	{
> +	  tsec->addr += loadbase;
> +	  tsec->endaddr += loadbase;
> +	}
> +
> +      add_target_sections (&nbfd, sections, sections_end);
> +    }
> +
>    /* This might change our ideas about frames already looked at.  */
>    reinit_frame_cache ();


> +# capture the disassembly of gettimeofday while live debugging
> +set live_gettimeofday [capture_command_output "disassemble gettimeofday" ""]
> +
> +# trace the test code
> +gdb_test_no_output "record btrace"
> +gdb_test "next" "main\.2.*"
> +
> +# capture the disassembly of gettimeofday while replaying
> +gdb_test "record goto begin" "main\.1.*"
> +set replay_gettimeofday [capture_command_output "disassemble gettimeofday" ""]

I think these two capture_command_output calls generate duplicate
gdb.sum output ?  That's why I had with_test_prefix in my suggestion.

-- 
Pedro Alves