[Bug 20509] New: gdb proxy mode: random reply to Offsets query?

http://bugs.winehq.org/show_bug.cgi?id=20509

Summary: gdb proxy mode: random reply to Offsets query?
           Product: Wine
           Version: 1.1.32
          Platform: PC
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: minor
          Priority: P2
         Component: winedbg
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

0) In gdb proxy mode the replies to "query packets" are constructed in
programs/winedbg/gdbproxy.c:packet_query(). The reply to qOffsets is based on
gdbctx->wine_segs (that reply has the form: Text=[...];Data=[...];Bss=[...]).

1) As far as I can see gdbctx->wine_segs is never properly set, so the reply
seems to be (basically) random.

2) If the above is correct, the code needs a patch. I have no idea to what
value(s) gdbctx->wine_segs should be set, so I can't draft that patch.

[Bug 20509] gdb proxy mode: random reply to Offsets query?

http://bugs.winehq.org/show_bug.cgi?id=20509

--- Comment #1 from Paul Bolle   2009-10-29 16:24:15 ---
(In reply to comment #0)
> - As far as I can see gdbctx->wine_segs is never properly set, so the reply
>
> seems to be (basically) random.
> - If the above is correct, the code needs a patch. I have no idea to what
>
> value(s) gdbctx->wine_segs should be set, so I can't draft that patch.

A simple printf() at the top of gdb_remote() trivially proves that wine_segs
will not be changed after gdbctx is defined (and still holds random data once
it is used).

Someone with a suggestion for a solution?

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

Paul Bolle  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|gdb proxy mode: random      |gdb proxy mode: random
                   |reply to Offsets query?     |reply to Offsets query

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

Eric Pouech  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #2 from Eric Pouech   2009-11-01 05:24:23 ---
first of all wine_segs[0..2] must be initialized to 0 in gdb_init_context

it would be rather straightforward to get a correct value for wine_segs[0] (aka
code) as it's the base address of the main module, which we could get in
gdb_startup

however, wine_segs[1] (aka data) would be more difficult to get
note, it seems that when Data=XXX, BSS=XXX is ignored by gdb

moreover, latest gdb version also use CodeSeg, DataSeg (instead of Code, Data)
in order to provide linear addresses instead of offsets for relocation, which
will be handier in our case

HTH
A+

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

--- Comment #3 from Paul Bolle   2009-11-01 06:49:00 ---
(In reply to comment #2)
> first of all wine_segs[0..2] must be initialized to 0 in gdb_init_context

0) Only if they're not set to something meaningful before they're used.
1) The code will currently return an error packet if all three are zero.
> it would be rather straightforward to get a correct value for wine_segs[0] (aka
> code) as it's the base address of the main module,

2) Wine's main module?
> which we could get in gdb_startup
>
> however, wine_segs[1] (aka data) would be more difficult to get
> note, it seems that when Data=XXX, BSS=XXX is ignored by gdb

3) According to gdb's documentation. gdb's code requires Bss=[...], even though
its value must be equal to Data=[...]. Rather silly.
> moreover, latest gdb version also use CodeSeg, DataSeg (instead of Code, Data)
> in order to provide linear addresses instead of offsets for relocation, which
> will be handier in our case

4) How can CodeSeg and DataSeg be determined?

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

--- Comment #4 from Eric Pouech   2009-11-02 13:56:52 ---
(In reply to comment #3)
> - The code will currently return an error packet if all three are zero.

yes, but that what you'd expect if the main exec is not relocated.
> - Wine's main module?

yes
> - According to gdb's documentation. gdb's code requires Bss=[...], even though
>
> its value must be equal to Data=[...]. Rather silly.

agreed
> - How can CodeSeg and DataSeg be determined?

look at dbg_get_debuggee_info in programs/winedbg/winedbg.c, it'll return a
module info for the wine loader...

A+

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

--- Comment #5 from Paul Bolle   2009-11-04 08:08:57 ---
(In reply to comment #4)
> (In reply to comment #3)
> > - The code will currently return an error packet if all three are zero.
>
> yes, but that what you'd expect if the main exec is not relocated.

As far as I can see (untested, however) that will generate a gdb warning:
  if (buf[0] == 'E')
    {
      warning (_("Remote failure reply: %s"), buf);
      return;
    }

So to me it looks better to return three zero offsets explicitly (or return
nothing, which gdb seems to handle juts fine, silently).

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

--- Comment #6 from Paul Bolle  2010-08-28 14:52:42 CDT ---
(In reply to comment #5)
> So to me it looks better to return three zero offsets explicitly (or return
> nothing, which gdb seems to handle just fine, silently).

This is precisely what happens since commit
a116f78dedaf05a8b3f44bf5d56a5b76faaccacc ("winedbg: In proxy mode, correctly
init the wine_segs fields."). Thanks.

I guess we can close this bug. If someone can think of a valid reason to return
more meaningful values to an Offsets query they could open a new report. (I
can't think of a reason right now. Actually, I can't even remember how I first
noticed this bug. And it's been less than a year ...)

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

Eric Pouech  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED

--- Comment #7 from Eric Pouech  2011-04-24 07:43:02 CDT ---
init of segs fixes the behavior

[Bug 20509] gdb proxy mode: random reply to Offsets query

http://bugs.winehq.org/show_bug.cgi?id=20509

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #8 from Alexandre Julliard  2011-04-29 13:16:19 CDT ---
Closing bugs fixed in 1.3.19.