Re: [PATCH v2] gdb: skip objfiles with no BFD in DWARF unwinder

[Simon Marchi <simark@simark.ca>]

On 4/6/22 14:55, Jan Vrany via Gdb-patches wrote:
> Changes since v1:
> 
> * use with_test_prefix as Lancelot suggested
> * rewrote test as Pedro suggested
> * updated commit message
> 
> -- >8 --
> While playing with JIT reader I experienced GDB to crash on null-pointer
> dereference when stepping through non-jitted code.
> 
> The problem was that dwarf2_frame_find_fde () assumed that all objfiles
> have BFD but that's not always true. To address this problem, this
> commit skips such objfiles.
> 
> To test the fix we put breakpoint in jit_function_add (). The JIT reader
> does not know how unwind this function so unwinding eventually falls
> back to DWARF unwinder which in turn iterates over objfiles. Since the
> the code is jitted, it is guaranteed it would eventually process JIT
> objfile.
> ---
>  gdb/dwarf2/frame.c                    |  3 +++
>  gdb/objfiles.h                        |  4 +++-
>  gdb/testsuite/gdb.base/jit-reader.exp | 13 +++++++++++++
>  3 files changed, 19 insertions(+), 1 deletion(-)
> 
> diff --git a/gdb/dwarf2/frame.c b/gdb/dwarf2/frame.c
> index 5878d72f922..514ae8c694f 100644
> --- a/gdb/dwarf2/frame.c
> +++ b/gdb/dwarf2/frame.c
> @@ -1565,6 +1565,9 @@ dwarf2_frame_find_fde (CORE_ADDR *pc, dwarf2_per_objfile **out_per_objfile)
>        CORE_ADDR offset;
>        CORE_ADDR seek_pc;
>  
> +      if (objfile->obfd == nullptr)
> +	continue;
> +
>        comp_unit *unit = find_comp_unit (objfile);
>        if (unit == NULL)
>  	{
> diff --git a/gdb/objfiles.h b/gdb/objfiles.h
> index 8bd76705688..429dea1da4c 100644
> --- a/gdb/objfiles.h
> +++ b/gdb/objfiles.h
> @@ -636,7 +636,9 @@ struct objfile
>    struct compunit_symtab *compunit_symtabs = nullptr;
>  
>    /* The object file's BFD.  Can be null if the objfile contains only
> -     minimal symbols, e.g. the run time common symbols for SunOS4.  */
> +     minimal symbols (e.g. the run time common symbols for SunOS4) or
> +     if the objfile is a dynamic objfile (e.g. created by JIT reader
> +     API).  */
>  
>    bfd *obfd;
>  
> diff --git a/gdb/testsuite/gdb.base/jit-reader.exp b/gdb/testsuite/gdb.base/jit-reader.exp
> index d94360cd7d9..756b400906c 100644
> --- a/gdb/testsuite/gdb.base/jit-reader.exp
> +++ b/gdb/testsuite/gdb.base/jit-reader.exp
> @@ -271,6 +271,19 @@ proc jit_reader_test {} {
>  		 "#1 ${any} in main ${any}" \
>  		]
>      }
> +
> +    with_test_prefix "test dwarf unwinder" {
> +	# Check that the DWARF unwinder does not crash in presence of
> +	# JIT objfiles.
> +	gdb_test "up"
> +	gdb_breakpoint "*function_add" temporary
> +	gdb_test "cont" ".*Temporary breakpoint ${any} in jit_function_add .*"
> +	gdb_test "bt" \
> +	    [multi_line \
> +		 "#0 ${any} in jit_function_add ${any}" \
> +		 "#1 ${any} in main ${any}" \
> +		]
> +    }

This test function is getting a tad long.  In general, I find that it
makes it difficult to know if some part of the test depends on the state
left by a previous portion of the test, if adding something in the
middle will break something later (or silently make a test not test what
it intended to test anymore).  I therefore like to split things up in
small independent functions that each start with a clean GDB.

This still LGTM though, not a blocker, but I thought I'd mention it for
future changes.

Approved-By: Simon Marchi <simon.marchi@efficios.com>

Simon