Re: [PATCH] Fix issues with reading rnglists, especially from dwo files, for DWARF v5

[Caroline Tice <cmtice@google.com>]

[-- Attachment #1: Type: text/plain, Size: 3141 bytes --]

On Mon, Jun 1, 2020 at 1:33 PM Tom Tromey <tom@tromey.com> wrote:
>
> >>>>> "Caroline" == Caroline Tice via Gdb-patches <gdb-patches@sourceware.org> writes:
>
> Caroline> This patch fixes these issues.  I verified that it fixes the
> Caroline> issues I saw and that it does not cause any testsuite
> Caroline> regressions (on x86_64 ubuntu linux).
>
> Is there a way to reproduce the failures with the current test suite?
> Say, by running some test using some board file?

I have not been able to reproduce it with the current testsuite.  The
issue is largely related to processing DW_FORM_rmglistx, which I have
not been able to get GCC to produce (it generated a DW_FORM_sec_offset
instead); I get my issue using llvm.  I will attempt to attach a
gzip'ed tarball (if gmail will let me) containing my small test case.
It has 2 .cpp files, 1 .h file, the two clang-generated .dwo files,
and the final binary.  The command I used to compile the sources, and
obtain the .dwo & binary files is:

$ clang++ -gdwarf-5 -O0 -gsplit-dwarf pre-order.cpp
pre-order-common.cpp -o pre-order

To see the first of the issues, you need to enable complaints (change
the value of "stop_whining" in complaints.c to something like 10).
When you do that, rebuild gdb, and load the binary into gdb, it
immediately complains:

$ ~/fsf-gdb.clean.obj/gdb/gdb pre-order
...
Reading symbols from pre-order...
During symbol reading: Invalid .debug_rnglists data (no base address)
During symbol reading: Invalid .debug_rnglists data (no base address)
(gdb) quit

This is because of the incorrect reading of the DW_FORM_rnglistx
index.  GDB then ignores the rnglists altogether because it can't read
them.



>
> Caroline> @@ -793,12 +796,18 @@ struct virtual_v2_dwo_sections
> Caroline>    bfd_size_type loc_offset;
> Caroline>    bfd_size_type loc_size;
>
> Caroline> +  bfd_size_type loclists_offset;
> Caroline> +  bfd_size_type loclists_size;
> Caroline> +
> Caroline>    bfd_size_type macinfo_offset;
> Caroline>    bfd_size_type macinfo_size;
>
> Caroline>    bfd_size_type macro_offset;
> Caroline>    bfd_size_type macro_size;
>
> Caroline> +  bfd_size_type rnglists_offset;
> Caroline> +  bfd_size_type rnglists_size;
>
> These new members don't seem to be used anywhere.

Oh.  Those get used in my next upcoming patch (where I update GDB to
handle DWARFv5 .dwp files).  I can either leave them in this patch, or
remove them from here and put them in the next one, whichever you
prefer.

>
> Caroline> +static struct dwarf2_section_info *cu_debug_rnglist_section (struct
> Caroline> +                                                          dwarf2_cu *cu);
>
> Probably better to split before the "(" and then indent the continuation
> line 2 spaces.
>
> Caroline> -     default:
> Caroline> +        case DW_RLE_startx_endx:
>
> Looks like the indentation here is incorrect.
>
> Caroline> +
> Caroline> +  attr =  die->attr (DW_AT_rnglists_base);
>
> Extra space after the "=".

Thanks for the review!  I will fix the formatting issues, and resubmit
as soon as you indicate whether to leave in or take out the currently
unused fields.

-- Caroline

>
> Tom

[-- Attachment #2: gdb-rnglist-test.tar.gz --]
[-- Type: application/x-gzip, Size: 27275 bytes --]