RE: [PATCH 6/9] gdb: Implement 'bt shadow' to print the shadow stack backtrace.

["Schimpe, Christina" <christina.schimpe@intel.com>]

> -----Original Message-----
> From: Guinevere Larsen <guinevere@redhat.com>
> Sent: Dienstag, 4. November 2025 12:54
> To: Schimpe, Christina <christina.schimpe@intel.com>; 'Simon Marchi'
> <simark@simark.ca>; 'Eli Zaretskii' <eliz@gnu.org>
> Cc: 'gdb-patches@sourceware.org' <gdb-patches@sourceware.org>;
> 'thiago.bauermann@linaro.org' <thiago.bauermann@linaro.org>
> Subject: Re: [PATCH 6/9] gdb: Implement 'bt shadow' to print the shadow stack
> backtrace.
> 
> On 11/3/25 4:47 PM, Schimpe, Christina wrote:
> >> -----Original Message-----
> >> From: Guinevere Larsen <guinevere@redhat.com>
> >> Sent: Mittwoch, 29. Oktober 2025 16:28
> >> To: Schimpe, Christina <christina.schimpe@intel.com>; 'Simon Marchi'
> >> <simark@simark.ca>; 'Eli Zaretskii' <eliz@gnu.org>
> >> Cc: 'gdb-patches@sourceware.org' <gdb-patches@sourceware.org>;
> >> 'thiago.bauermann@linaro.org' <thiago.bauermann@linaro.org>
> >> Subject: Re: [PATCH 6/9] gdb: Implement 'bt shadow' to print the
> >> shadow stack backtrace.
> >>
> >> On 10/29/25 12:05 PM, Schimpe, Christina wrote:
> >>> Kindly pinging for feedback on this discussion to clarify if we
> >>> should better use
> >>> - "bt -shadow" (command line option)
> >>> - "bt shadow" (subcommand approach, current implementation).
> >>>
> >>> My personal opinion is still to use the subcommand approach, but, as
> >>> I already said, I think "bt -shadow" would be fine, too.
> >>>
> >>> Please also see this discussion with Thiago:
> >>> https://sourceware.org/pipermail/gdb-patches/2025-
> October/221660.htm
> >>> l
> >> If it makes any difference, I personally prefer the command line
> >> option approach. That's based on looking at how record, record full
> >> and record btrace works, it is much easier to understand the code if
> >> the decision is made on the command's function itself
> > Hi Guinevere,
> >
> > I might be missing something here. Aren’t record, record full, and record
> btrace considered subcommands?
> > If that’s the case, wouldn’t the example of the record command actually
> support the subcommand approach? Or are you pointing to record as a
> negative example?
> 
> I am using it as a negative example, yes, sorry if it was confusing.
> 
> record full and record btrace are subcommands of record, and I found it
> needlessly confusing to figure out how the code decided what to do. I didn't
> even know that commands with subcommands were supported to be quite
> honest, so I expected "full" and "btrace" to be handled like options, and was
> quite confused when that wasn't the case.
> >
> > About your comment:
> > "it is much easier to understand the code if the decision is made on the
> command's function itself"
> > Could you clarify what you mean? Are you referring to making it easier for a
> reviewer or developer to go through my patches?
> >  From my perspective, the most important thing is that the GDB user can find
> the subcommand or command-line option easily and intuitively.
> 
> I mean making it easier to maintain as a GDB developer, yes. However, I do
> think this will also make things more confusing to end-users. Right now,
> backtrace accepts '-' prefixed options and a few non '-' prefixed options for
> backwards compatibility. from the POV of an end-user who doesn't know the
> difference between a command option and a subcommand, you'd be
> introducing a new option in the backwards compatible (but not
> preferred) style. Or worse, they'd think that "backtrace full", "backtrace no-
> filters" and "backtrace hide" were also subcommands but now the list of
> backtrace subcommands would no longer show those,  which makes it sound
> like they stopped being supported or that the help text is incorrect.

This is how "bt shadow" is shown in the help text of "help bt"

~~~
(gdb) help bt
backtrace, where, bt
Print backtrace of all stack frames, or innermost COUNT frames.
Usage: backtrace [OPTION]... [QUALIFIER]... [COUNT | -COUNT]

Options:
  -entry-values no|only|preferred|if-needed|both|compact|default
   [..]
  -hide
    Causes Python frame filter elided frames to not be printed.

For backward compatibility, the following qualifiers are supported:

   full       - same as -full option.
   no-filters - same as -no-filters option.
   hide       - same as -hide.

With a negative COUNT, print outermost -COUNT frames.

List of "backtrace" subcommands:

backtrace shadow -- Print backtrace of all shadow stack frames, or innermost COUNT frames.
~~~

I had assumed this was clear from the help text, but I understand your perspective, too.

Here is the summarized pro/con list:

Command-line option: bt -shadow

Pros:
- Matches the general trend of moving away from subcommands for backtrace. (Eli, Thiago)
- Easier for GDB developers to maintain. (Guinevere)

Cons:
- Existing backtrace options affect output but not the algorithm, unlike shadow stack backtrace.
  This makes -shadow harder to discover as new command type. (Christina)
- Some existing options don’t make sense when combined with -shadow. (Christina, Simon)

Subcommand: bt shadow

Pros:
- Indicates that the output is fundamentally different from the ordinary backtrace,
  even if it looks similar. (Christina)

Cons:
- backtrace is both a prefix and a regular command. Adding subcommands could cause
  conflicts if future options like "backtrace SYMBOL" are introduced. (Simon)
- Confusing for users who don’t understand the difference between options and
  subcommands. (Guinevere)

Since Simon, Thiago, Guinevere, and Eli seem to prefer the command-line option,
I will proceed in that direction for v2 of this series unless I hear otherwise in the
next couple of days.

This is my suggestion for the new help text:

~~~
(gdb) help bt
backtrace, where, bt
Print backtrace of all stack frames, or innermost COUNT frames.
Usage: backtrace [OPTION]... [QUALIFIER]... [COUNT | -COUNT]

Options:
  -entry-values no|only|preferred|if-needed|both|compact|default
   [..]
  -frame-arguments all|scalars|none|presence
    Set printing of non-scalar frame arguments.
  -raw-frame-arguments [on|off]
    [..]
  -frame-info auto|source-line|location|source-and-location
              |location-and-address|short-location
    Set printing of frame information.
  -past-main [on|off]
    [..]
  -past-entry [on|off]
    [..]
  -full
    Print values of local variables.
  -no-filters
    Prohibit frame filters from executing on a backtrace.
  -hide
    Causes Python frame filter elided frames to not be printed.- shadow
   Print shadow stack frames instead of normal frames.  This option may be combined
   with “-frame-info” and implies ‘-no-filters’ and ‘-frame-arguments none’.
- shadow
   Print shadow stack frames instead of normal frames.  This option may be combined
   with “-frame-info” and implies ‘-no-filters’ and ‘-frame-arguments none’.

For backward compatibility, the following qualifiers are supported:

   full       - same as -full option.
   no-filters - same as -no-filters option.
   hide       - same as -hide.

With a negative COUNT, print outermost -COUNT frames.
~~~

Thank you all for the valuable feedback!
Christina
Intel Deutschland GmbH
Registered Address: Dornacher Straße 1, 85622 Feldkirchen, Germany
Tel: +49 89 991 430, www.intel.de
Managing Directors: Harry Demas, Jeffrey Schneiderman, Yin Chong Sorrell
Chairperson of the Supervisory Board: Nicole Lau
Registered Seat: Munich
Commercial Register: Amtsgericht München HRB 186928