RE: GDB solib interface

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

Ok - I am not getting any T stop packets on the gdbserver end.  Is there
something I have to do to get gdb to issue them?

The text you wrote:

Reporting DLL events
--------------------

The `T' stop packet response is extended to take three additional
N:R pairs.  When any of these are encountered, the stopped thread
is assumed to be at a DLL event.  The signal should be SIGTRAP.
There may be multiple load and unload events in the stop packet.

  `load:Name=HEXSTR,TextSeg=ADDR,DataSeg=ADDR[,nop]'

    A new DLL has been loaded; it is described in the stop packet.

    If "nop" is specified, this DLL was already loaded, e.g. by
    an earlier call to dlopen.

  `unload:IDENTIFIER=VALUE[,nop]'

    A DLL has been unloaded.  A single identifier is provided to
    uniquely identify which DLL has been unloaded.  Depending
    on the target, this may be the text or data address; the name
    can be used, but may not be unambiguous.

    If "nop" is specified, this DLL was not unloaded, e.g. because
    another thread still had a handle to it.

  `dll:dll'

    Multiple DLL events have occured.  GDB should query the target for
the
    current state of DLLs.

Re: GDB solib interface

[Daniel Jacobowitz]

On Thu, Jan 04, 2007 at 02:32:37PM -0700, Smith, Stephen (SWCOE) wrote:
> Ok - I am not getting any T stop packets on the gdbserver end.  Is there
> something I have to do to get gdb to issue them?

'T' is something that gdbserver should send to GDB, not the other way
around.

-- 
Daniel Jacobowitz
CodeSourcery

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

So you are saying that after a continue, single step, etc.   If my
libraries have changed,  I send a T packet and then what I normally
would have sent?

sps

-----Original Message-----
From: gdb-owner@sourceware.org [mailto:gdb-owner@sourceware.org] On
Behalf Of Daniel Jacobowitz
Sent: Thursday, January 04, 2007 2:40 PM
To: Smith, Stephen (SWCOE)
Cc: General GDB Mailing List
Subject: Re: GDB solib interface

On Thu, Jan 04, 2007 at 02:32:37PM -0700, Smith, Stephen (SWCOE) wrote:
> Ok - I am not getting any T stop packets on the gdbserver end.  Is
there
> something I have to do to get gdb to issue them?

'T' is something that gdbserver should send to GDB, not the other way
around.

-- 
Daniel Jacobowitz
CodeSourcery

Re: GDB solib interface

[Daniel Jacobowitz]

On Thu, Jan 04, 2007 at 02:46:13PM -0700, Smith, Stephen (SWCOE) wrote:
> So you are saying that after a continue, single step, etc.   If my
> libraries have changed,  I send a T packet and then what I normally
> would have sent?

When you detect a library change, you should stop, and add the
appropriate clause to your 'T' packet response for that stop so that
GDB knows why you stopped.  You should probably be using 'T' already.

-- 
Daniel Jacobowitz
CodeSourcery

Re: GDB solib interface

[Stephen & Linda Smith]

In the patch you sent, did you check to make sure load, unload, or dll 
worked?  Actually I have been stairing at the debugger and can't see 
that any of the following work in remote_await()  [file: remote.c]: 
watch, rwatch, load, unload, or dll.  All of them would be broken 
because of the strncmp() check for awatch which does not check for the 
others.  That would mean that the check for p1==p will fail.

Am I missing something?

Re: GDB solib interface

[Daniel Jacobowitz]

On Thu, Jan 04, 2007 at 08:45:24PM -0700, Stephen & Linda Smith wrote:
> In the patch you sent, did you check to make sure load, unload, or dll 
> worked?  Actually I have been stairing at the debugger and can't see 
> that any of the following work in remote_await()  [file: remote.c]: 
> watch, rwatch, load, unload, or dll.  All of them would be broken 
> because of the strncmp() check for awatch which does not check for the 
> others.  That would mean that the check for p1==p will fail.
> 
> Am I missing something?

The strncmp's the opposite way you probably think it is - I get this
wrong every time I look at that code.  I know for sure that the
load/unload responses do work.

-- 
Daniel Jacobowitz
CodeSourcery

Re: GDB solib interface

[Daniel Jacobowitz]

On Thu, Jan 04, 2007 at 08:54:54PM -0700, Smith, Stephen (SWCOE) wrote:
> Ok,  I have the string 'T05dll:dll'.  Single stepping trough the code, I
> get p1 -> ll:dll and p -> dll:dll.  Therefore I never enter the block.  
> 
> What did I do wrong with the packet?

I said I was certain load and unload worked - I hadn't tested dll.
Looks like the leading 'd' means it needs to be handled just like
awatch above.

-- 
Daniel Jacobowitz
CodeSourcery

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

Ok, I have our gdbserver using the updated solib-remote gdb patch using
the dll stop packet.  You said that you've tested the load/unload packet
so as far as I know the patch has been successfully updated to compile
and link with 6.5 (which was the last release when I started work on
it).  I will make sure that the patch compiles/links against head.
Should I propose the code patch before the documentation patch is
created - so that the code can be critiqued?

I remember that I need to make sure that the files that are to be
changed are formatted properly (Kevin Buettner said to use "indent -fca 
-lc80 -l80").  When I do that the patch grows by about a factor of 10
due to unrelated items being reformatted to match the GNU coding
standard.  Do you want me to run indent before or after the patch is
reviewed?

Also how do we get the published changes to the gdb/gdbserver protocol
approved?

The interface documentation does not need changing unless someone here
doesn't like it.  Then again I've never updated the gdb documentation
and so don't know how to go about it.  Could someone point me in the
right direction?

BTW - I have a copyright assignment on file with FSF and nothing has
changed on this end to negate it.

-----Original Message-----
From: Daniel Jacobowitz [mailto:drow@false.org] 
Sent: Thursday, January 04, 2007 9:01 PM
To: Smith, Stephen (SWCOE)
Cc: Stephen & Linda Smith; gdb@sourceware.org
Subject: Re: GDB solib interface

On Thu, Jan 04, 2007 at 08:54:54PM -0700, Smith, Stephen (SWCOE) wrote:
> Ok,  I have the string 'T05dll:dll'.  Single stepping trough the code,
I
> get p1 -> ll:dll and p -> dll:dll.  Therefore I never enter the block.

> 
> What did I do wrong with the packet?

I said I was certain load and unload worked - I hadn't tested dll.
Looks like the leading 'd' means it needs to be handled just like
awatch above.

-- 
Daniel Jacobowitz
CodeSourcery

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

Weird issue, we have implemented the patch for the remote protocol using
the patch that you posted with some minor tweeks. We are getting ready
to internally release the updated gdbserver and found a problem.

On our platform, it is not possible to hook the function that
dynamically loads the libraries, so I was checking to see if new
libraries were loaded after the processor hit a breakpoint (i.e.
immediatetly following the prepare_resume_reply() for the C, S, c, and s
packets).  Everything works well unless the following two things are
true:
   * there are new libraries which are then reported
   * the program counter stopped at the location 
     which has a GDB applied breakpoint.

In this case, the instruction which is where the breakpoint was inserted
is not executed and in our case a fault occurred.

Since on my platform I won't halt because of a library load, where
should I place this check?

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

We figured out a fix.  

-----Original Message-----
From: Smith, Stephen (SWCOE) 
Sent: Wednesday, February 07, 2007 12:41 PM
To: Smith, Stephen (SWCOE); Daniel Jacobowitz
Cc: gdb@sourceware.org
Subject: RE: GDB solib interface

Weird issue, we have implemented the patch for the remote protocol using
the patch that you posted with some minor tweeks. We are getting ready
to internally release the updated gdbserver and found a problem.

On our platform, it is not possible to hook the function that
dynamically loads the libraries, so I was checking to see if new
libraries were loaded after the processor hit a breakpoint (i.e.
immediatetly following the prepare_resume_reply() for the C, S, c, and s
packets).  Everything works well unless the following two things are
true:
   * there are new libraries which are then reported
   * the program counter stopped at the location 
     which has a GDB applied breakpoint.

In this case, the instruction which is where the breakpoint was inserted
is not executed and in our case a fault occurred.

Since on my platform I won't halt because of a library load, where
should I place this check?

Re: GDB solib interface

[Smith, Stephen (SWCOE)]

I'm sorry - way back when Kevin asked that I investigate this.  So now I
am.  

Anyway - Pretend that my "custom" GDB remote.c has been told that a
library has been loaded - what would I need to write to use the solib
interface?   Another way of asking the question would be,  what would I
need to add to my remote interface file (remote.c) to have it load the
symbols.

I am not looking for a command, per se,  just trying to figure out what
functions/structures etc I need to write to have my "custom" GDB load
the libraries symbols.

-----Original Message-----
From: Daniel Jacobowitz [mailto:drow@false.org] 
Sent: Monday, December 04, 2006 2:58 PM
To: Smith, Stephen (SWCOE)
Cc: General GDB Mailing List
Subject: Re: GDB solib interface

On Mon, Dec 04, 2006 at 02:55:28PM -0700, Smith, Stephen (SWCOE) wrote:
> I am running GDB to debug a remote application.  If I find that a
shared
> library has been loaded, I would like to use the solib interface to
load
> the appropriate shared library symbols.  
> 
> So lets assume that I have the file name of the shared library, that
was
> just loaded, how do I inform the solib subsystem that to load the
> symbols.

There's no command for this.  The solib system is designed for targets
where the target can report that the library was loaded, or where GDB
can otherwise detect it.  If you're willing to have a custom GDB for
your platform, you might be able to make it examine the shared library
table in your target, assuming there is one.

-- 
Daniel Jacobowitz
CodeSourcery

Re: GDB solib interface

[Michael Snyder]

On Mon, 2006-12-04 at 15:12 -0700, Smith, Stephen (SWCOE) wrote:
> I'm sorry - way back when Kevin asked that I investigate this.  So now I
> am.  
> 
> Anyway - Pretend that my "custom" GDB remote.c has been told that a
> library has been loaded - what would I need to write to use the solib
> interface?   Another way of asking the question would be,  what would I
> need to add to my remote interface file (remote.c) to have it load the
> symbols.

Presumably something like solib_add() or add_symbol_file().

Your back-end (eg. remote target) should give you the name
of a shared library (with or without path), and then gdb 
has built-in schemes for finding it.

If your target libraries are different from your host libraries
(which they usually are), you will need to tell GDB where to 
find them (on the host file system), using the commands
    set solib-search-path 
and
    set solib-absolute-prefix

Daniel, I would have assumed that this stuff was 
working with linux-x-linux and gdbserver?


> 
> I am not looking for a command, per se,  just trying to figure out what
> functions/structures etc I need to write to have my "custom" GDB load
> the libraries symbols.
> 
> -----Original Message-----
> From: Daniel Jacobowitz [mailto:drow@false.org] 
> Sent: Monday, December 04, 2006 2:58 PM
> To: Smith, Stephen (SWCOE)
> Cc: General GDB Mailing List
> Subject: Re: GDB solib interface
> 
> On Mon, Dec 04, 2006 at 02:55:28PM -0700, Smith, Stephen (SWCOE) wrote:
> > I am running GDB to debug a remote application.  If I find that a
> shared
> > library has been loaded, I would like to use the solib interface to
> load
> > the appropriate shared library symbols.  
> > 
> > So lets assume that I have the file name of the shared library, that
> was
> > just loaded, how do I inform the solib subsystem that to load the
> > symbols.
> 
> There's no command for this.  The solib system is designed for targets
> where the target can report that the library was loaded, or where GDB
> can otherwise detect it.  If you're willing to have a custom GDB for
> your platform, you might be able to make it examine the shared library
> table in your target, assuming there is one.
> 

Re: GDB solib interface

[Daniel Jacobowitz]

On Mon, Dec 04, 2006 at 02:25:02PM -0800, Michael Snyder wrote:
> Daniel, I would have assumed that this stuff was 
> working with linux-x-linux and gdbserver?

No, solib-svr4 doesn't need any of this because it's just grubbing
around in target memory.

-- 
Daniel Jacobowitz
CodeSourcery

Re: GDB solib interface

[Daniel Jacobowitz]

On Mon, Dec 04, 2006 at 03:12:27PM -0700, Smith, Stephen (SWCOE) wrote:
> 
> I'm sorry - way back when Kevin asked that I investigate this.  So now I
> am.  
> 
> Anyway - Pretend that my "custom" GDB remote.c has been told that a
> library has been loaded - what would I need to write to use the solib
> interface?   Another way of asking the question would be,  what would I
> need to add to my remote interface file (remote.c) to have it load the
> symbols.
> 
> I am not looking for a command, per se,  just trying to figure out what
> functions/structures etc I need to write to have my "custom" GDB load
> the libraries symbols.

Aha.  OK, that's different.

I wrote some code for this purpose, a few months ago, which I meant to
submit.  That was for DLL targets.  I haven't had time yet.  I'm afraid
it's not real easy to untangle from the rest of that project now :-(
If that's what you need, I can try to do it tomorrow.

-- 
Daniel Jacobowitz
CodeSourcery

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

Well in my case the target is elf not pe.  But otherwise it should be
similar.  


-----Original Message-----
From: Daniel Jacobowitz [mailto:drow@false.org] 
Sent: Monday, December 04, 2006 3:29 PM
To: Smith, Stephen (SWCOE)
Cc: General GDB Mailing List
Subject: Re: GDB solib interface

On Mon, Dec 04, 2006 Daniel Jacobowitz wrote:

Aha.  OK, that's different.

I wrote some code for this purpose, a few months ago, which I meant to
submit.  That was for DLL targets.  I haven't had time yet.  I'm afraid
it's not real easy to untangle from the rest of that project now :-(
If that's what you need, I can try to do it tomorrow.

-- 
Daniel Jacobowitz
CodeSourcery

RE: GDB solib interface

[Michael Snyder]

On Mon, 2006-12-04 at 15:48 -0700, Smith, Stephen (SWCOE) wrote:
> Well in my case the target is elf not pe.  But otherwise it should be
> similar.  

Is your target by any chance svr4 compatible?

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

No, the OS is not SVR4 compatible.

sps

-----Original Message-----
From: Michael Snyder [mailto:Michael.Snyder@palmsource.com] 
Sent: Monday, December 04, 2006 4:48 PM
To: Smith, Stephen (SWCOE)
Cc: Daniel Jacobowitz; General GDB Mailing List
Subject: RE: GDB solib interface

On Mon, 2006-12-04 at 15:48 -0700, Smith, Stephen (SWCOE) wrote:
> Well in my case the target is elf not pe.  But otherwise it should be
> similar.  

Is your target by any chance svr4 compatible?

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

Could you try and get it.  I would like to see how to do it.

On Mon, Dec 04, 2006 Daniel Jacobowitz wrote

I wrote some code for this purpose, a few months ago, which I meant to
submit.  That was for DLL targets.  I haven't had time yet.  I'm afraid
it's not real easy to untangle from the rest of that project now :-(
If that's what you need, I can try to do it tomorrow.

Re: GDB solib interface

[Daniel Jacobowitz]

On Mon, Dec 04, 2006 at 05:28:47PM -0500, Daniel Jacobowitz wrote:
> On Mon, Dec 04, 2006 at 03:12:27PM -0700, Smith, Stephen (SWCOE) wrote:
> > 
> > I'm sorry - way back when Kevin asked that I investigate this.  So now I
> > am.  
> > 
> > Anyway - Pretend that my "custom" GDB remote.c has been told that a
> > library has been loaded - what would I need to write to use the solib
> > interface?   Another way of asking the question would be,  what would I
> > need to add to my remote interface file (remote.c) to have it load the
> > symbols.
> > 
> > I am not looking for a command, per se,  just trying to figure out what
> > functions/structures etc I need to write to have my "custom" GDB load
> > the libraries symbols.
> 
> Aha.  OK, that's different.
> 
> I wrote some code for this purpose, a few months ago, which I meant to
> submit.  That was for DLL targets.  I haven't had time yet.  I'm afraid
> it's not real easy to untangle from the rest of that project now :-(

Here you go.  I haven't tested this patch on its own at all, so if it
breaks, you get to keep both pieces.  Add solib-target.o to TDEPFILES
for your target to enable it.

It works by two mechanisms.  You can send "qfDllInfo" to get the entire
list from a remote target, and the remote target can report load /
unload events as special 'T' stop packets.  It should be fairly
apparent from the code.

Some draft docs I wrote at the time:

`qfDllInfo'
`qsDllInfo'

  Obtain a list of currently loaded DLLs from the target.  Like
  `qfThreadInfo', this request works iteratively.  The `qfDllInfo'
  packet is issued first to start at the beginning of the DLL list,
  and the `qsDllInfo' packet is used to fetch additional entries.

  Reply:
  `m' Name=HEXSTR,TextSeg=ADDR,DataSeg=ADDR

    Returns the name (as a hexadecimal-encoded sequence of ASCII
    characters), text load address, and data load address for this DLL.
    Other Name=Value pairs may be added in the future for particular
    targets.

  `m' Name=HEXSTR,TextSeg=ADDR,DataSeg=ADDR;Name=HEXSTR,TextSeg=ADDR,DataSeg=ADDR...

    Multiple DLL entries may be included in the same packet; a semicolon
    marks the start of a new DLL description.  A single DLL
    will never be split across multiple packets.

  `l'

    Marks the end of a list of DLLs.

[Editorial note: maybe use `ml' instead?  Then we can use
`mName=FOO,TextSeg=ADDR,DataSeg=ADDR;l' and save a round trip.  This is not quite
parallel to qfThreadInfo but that should be ok.]

[Editorial note: No support is provided for DLLs with extremely long names
that do not fit in a packet.  A solution for this can be added at a later
date if necessary.]

If any stopped threads had DLL events not yet reported to GDB, either the
effects of those events should not be included in the result of the query,
or the queued DLL events should be discarded before they are reported to
GDB.

[Editorial note: You wrote that every dlopen triggers an event even if the
DLL was already loaded.  Should the stub drop the event in that case?
Should it report it anyway and have duplicates handled on the GDB side?
That would let the long-lost "catch load <libname>" command work again.
If so, then the above paragraph must be deleted and some words about how
GDB should ignore apparent duplicates added.]

Reporting DLL events
--------------------

The `T' stop packet response is extended to take three additional
N:R pairs.  When any of these are encountered, the stopped thread
is assumed to be at a DLL event.  The signal should be SIGTRAP.
There may be multiple load and unload events in the stop packet.

  `load:Name=HEXSTR,TextSeg=ADDR,DataSeg=ADDR[,nop]'

    A new DLL has been loaded; it is described in the stop packet.

    If "nop" is specified, this DLL was already loaded, e.g. by
    an earlier call to dlopen.

  `unload:IDENTIFIER=VALUE[,nop]'

    A DLL has been unloaded.  A single identifier is provided to
    uniquely identify which DLL has been unloaded.  Depending
    on the target, this may be the text or data address; the name
    can be used, but may not be unambiguous.

    If "nop" is specified, this DLL was not unloaded, e.g. because
    another thread still had a handle to it.

  `dll:dll'

    Multiple DLL events have occured.  GDB should query the target for the
    current state of DLLs.

-- 
Daniel Jacobowitz
CodeSourcery

Index: Makefile.in
===================================================================
RCS file: /cvs/src/src/gdb/Makefile.in,v
retrieving revision 1.856
diff -u -p -r1.856 Makefile.in
--- Makefile.in	28 Nov 2006 22:14:31 -0000	1.856
+++ Makefile.in	8 Dec 2006 21:09:03 -0000
@@ -2516,7 +2516,8 @@ remote.o: remote.c $(defs_h) $(gdb_strin
 	$(gdb_stabs_h) $(gdbthread_h) $(remote_h) $(regcache_h) $(value_h) \
 	$(gdb_assert_h) $(event_loop_h) $(event_top_h) $(inf_loop_h) \
 	$(serial_h) $(gdbcore_h) $(remote_fileio_h) $(solib_h) $(observer_h) \
-	$(cli_decode_h) $(cli_setshow_h) $(memory_map_h) $(target_descriptions_h)
+	$(cli_decode_h) $(cli_setshow_h) $(memory_map_h) $(target_descriptions_h) \
+	$(solist_h)
 remote-e7000.o: remote-e7000.c $(defs_h) $(gdbcore_h) $(gdbarch_h) \
 	$(inferior_h) $(target_h) $(value_h) $(command_h) $(gdb_string_h) \
 	$(exceptions_h) $(gdbcmd_h) $(serial_h) $(remote_utils_h) \
@@ -2650,6 +2651,8 @@ solib-svr4.o: solib-svr4.c $(defs_h) $(e
 	$(gdbcore_h) $(target_h) $(inferior_h) $(gdb_assert_h) \
 	$(solist_h) $(solib_h) $(solib_svr4_h) $(bfd_target_h) $(elf_bfd_h) \
 	$(exec_h)
+solib-target.o: solib-target.c $(defs_h) $(solist_h) $(symtab_h) $(symfile_h) \
+	$(target_h) $(gdb_string_h)
 sol-thread.o: sol-thread.c $(defs_h) $(gdbthread_h) $(target_h) \
 	$(inferior_h) $(gdb_stat_h) $(gdbcmd_h) $(gdbcore_h) $(regcache_h) \
 	$(solib_h) $(symfile_h) $(gdb_string_h) $(gregset_h)
Index: infrun.c
===================================================================
RCS file: /cvs/src/src/gdb/infrun.c,v
retrieving revision 1.216
diff -u -p -r1.216 infrun.c
--- infrun.c	18 Oct 2006 16:56:13 -0000	1.216
+++ infrun.c	8 Dec 2006 21:09:04 -0000
@@ -884,6 +884,8 @@ init_wait_for_inferior (void)
   clear_proceed_status ();
 
   stepping_past_singlestep_breakpoint = 0;
+
+  target_last_wait_ptid = minus_one_ptid;
 }
 \f
 /* This enum encodes possible reasons for doing a target_wait, so that
@@ -1262,6 +1264,9 @@ handle_inferior_event (struct execution_
   target_last_wait_ptid = ecs->ptid;
   target_last_waitstatus = *ecs->wp;
 
+  /* Always clear state belonging to the previous time we stopped.  */
+  stop_stack_dummy = 0;
+
   adjust_pc_after_break (ecs);
 
   switch (ecs->infwait_state)
@@ -1322,13 +1327,16 @@ handle_inferior_event (struct execution_
       /* Ignore gracefully during startup of the inferior, as it
          might be the shell which has just loaded some objects,
          otherwise add the symbols for the newly loaded objects.  */
-#ifdef SOLIB_ADD
       if (stop_soon == NO_STOP_QUIETLY)
 	{
+	  int breakpoints_were_inserted;
+
 	  /* Remove breakpoints, SOLIB_ADD might adjust
 	     breakpoint addresses via breakpoint_re_set.  */
+	  breakpoints_were_inserted = breakpoints_inserted;
 	  if (breakpoints_inserted)
 	    remove_breakpoints ();
+	  breakpoints_inserted = 0;
 
 	  /* Check for any newly added shared libraries if we're
 	     supposed to be adding them automatically.  Switch
@@ -1350,17 +1358,52 @@ handle_inferior_event (struct execution_
 	     exec/process stratum, instead relying on the target stack
 	     to propagate relevant changes (stop, section table
 	     changed, ...) up to other layers.  */
+#ifdef SOLIB_ADD
 	  SOLIB_ADD (NULL, 0, &current_target, auto_solib_add);
+#else
+	  solib_add (NULL, 0, &current_target, auto_solib_add);
+#endif
 	  target_terminal_inferior ();
 
+	  /* Try to reenable shared library breakpoints, additional
+	     code segments in shared libraries might be mapped in now. */
+	  re_enable_breakpoints_in_shlibs ();
+
+	  /* If requested, stop when the dynamic linker notifies
+	     gdb of events.  This allows the user to get control
+	     and place breakpoints in initializer routines for
+	     dynamically loaded objects (among other things).  */
+	  if (stop_on_solib_events)
+	    {
+	      stop_stepping (ecs);
+	      return;
+	    }
+
+	  /* NOTE drow/2006-03-14: This might be a good place to check
+	     for "catch load".  */
+
 	  /* Reinsert breakpoints and continue.  */
-	  if (breakpoints_inserted)
-	    insert_breakpoints ();
+	  if (breakpoints_were_inserted)
+	    {
+	      insert_breakpoints ();
+	      breakpoints_inserted = 1;
+	    }
 	}
-#endif
-      resume (0, TARGET_SIGNAL_0);
-      prepare_to_wait (ecs);
-      return;
+
+      /* NOTE drow/2006-03-28: For the reason described before the
+	 previous if statement, GDB used to automatically resume
+	 here.  But that's only true if a shell is running; if
+	 we've just attached to a process, then that's a whole
+	 different case - it might have been stopped at a load
+	 event.  */
+      if (inferior_ignoring_startup_exec_events || stop_soon == NO_STOP_QUIETLY)
+	{
+	  resume (0, TARGET_SIGNAL_0);
+	  prepare_to_wait (ecs);
+	  return;
+	}
+      else
+	break;
 
     case TARGET_WAITKIND_SPURIOUS:
       if (debug_infrun)
@@ -1779,7 +1822,6 @@ handle_inferior_event (struct execution_
   ecs->another_trap = 0;
   bpstat_clear (&stop_bpstat);
   stop_step = 0;
-  stop_stack_dummy = 0;
   stop_print_frame = 1;
   ecs->random_signal = 0;
   stopped_by_random_signal = 0;
Index: remote.c
===================================================================
RCS file: /cvs/src/src/gdb/remote.c,v
retrieving revision 1.241
diff -u -p -r1.241 remote.c
--- remote.c	28 Nov 2006 22:14:31 -0000	1.241
+++ remote.c	8 Dec 2006 21:09:04 -0000
@@ -43,6 +43,7 @@
 #include "gdb_assert.h"
 #include "observer.h"
 #include "solib.h"
+#include "solist.h"
 #include "cli/cli-decode.h"
 #include "cli/cli-setshow.h"
 #include "target-descriptions.h"
@@ -897,6 +898,7 @@ enum {
   PACKET_qGetTLSAddr,
   PACKET_qSupported,
   PACKET_QPassSignals,
+  PACKET_qfDllInfo,
   PACKET_MAX
 };
 
@@ -2076,6 +2078,152 @@ get_offsets (void)
   objfile_relocate (symfile_objfile, offs);
 }
 
+/* Parse a load or unload message for a DLL.  Loads are required to
+   have all fields, unloads must have at least one.  P is not
+   necessarily NUL terminated, but it is always either NUL or semicolon
+   terminated (i.e. *P_END is either NUL or a semicolon) and the containing
+   string is NUL terminated.  */
+
+int
+parse_load_response (const char *p, const char *p_end, int is_load)
+{
+  const char *name_start;
+  char *name;
+  CORE_ADDR text_addr, data_addr;
+  int i;
+
+  name = NULL;
+  text_addr = ~(CORE_ADDR) 0;
+  data_addr = ~(CORE_ADDR) 0;
+
+  if (p < p_end && strncmp (p, "Name=", 5) == 0)
+    {
+      p += 5;
+
+      name_start = p;
+      while (p < p_end && *p != ',')
+	p++;
+
+      if ((p - name_start) % 2 != 0)
+	return -1;
+
+      name = alloca ((p - name_start) / 2 + 1);
+      i = hex2bin (name_start, name, (p - name_start) / 2);
+      name[i] = '\0';
+
+      if (p < p_end)
+	p++;
+    }
+  else if (is_load)
+    return -1;
+
+  if (p < p_end && strncmp (p, "TextSeg=", 8) == 0)
+    {
+      p += 8;
+
+      text_addr = 0;
+      while (p < p_end && *p != ',')
+	text_addr = (text_addr << 4) + fromhex (*p++);
+
+      if (p < p_end)
+	p++;
+    }
+  else if (is_load)
+    return -1;
+
+  if (p < p_end && strncmp (p, "DataSeg=", 8) == 0)
+    {
+      p += 8;
+
+      data_addr = 0;
+      while (p < p_end && *p != ',')
+	data_addr = (data_addr << 4) + fromhex (*p++);
+
+      if (p < p_end)
+	p++;
+    }
+  else if (is_load)
+    return -1;
+
+  if (is_load)
+    current_target_so_ops->add_one_solib (name, text_addr, data_addr);
+  else
+    {
+      if (text_addr == ~(CORE_ADDR) 0
+	  && data_addr == ~(CORE_ADDR) 0
+	  && name == NULL)
+	return -1;
+
+      current_target_so_ops->remove_one_solib (name, text_addr, data_addr);
+    }
+
+  return 0;
+}
+
+/* Query the remote side for loaded solibs.  */
+
+static void
+remote_get_shared_libraries (struct target_ops *ops)
+{
+  struct remote_state *rs = get_remote_state ();
+
+  /* If this target doesn't support remote DLLs, nothing to do.  */
+  if (current_target_so_ops->add_one_solib == NULL)
+    return;
+
+  /* If qfDllInfo is not available, nothing to do.  */
+  if (remote_protocol_packets[PACKET_qfDllInfo].support == PACKET_DISABLE)
+    return;
+
+  putpkt ("qfDllInfo");
+  getpkt (&rs->buf, &rs->buf_size, 0);
+
+  if (packet_ok (rs->buf, &remote_protocol_packets[PACKET_qfDllInfo]) == PACKET_ERROR)
+    {
+      warning (_("Remote failure reply: %s"), rs->buf);
+      return;
+    }
+  else if (remote_protocol_packets[PACKET_qfDllInfo].support == PACKET_DISABLE)
+    /* It wasn't disabled before, but it is now.  */
+    return;
+
+  while (rs->buf[0] == 'm')
+    {
+      char *p = rs->buf + 1;
+
+      while (1)
+	{
+	  char *p_end = p;
+
+	  while (*p_end && *p_end != ';')
+	    p_end++;
+
+	  if (parse_load_response (p, p_end, 1) != 0)
+	    {
+	      warning (_("Malformed response to DLL query, %s"), rs->buf);
+	      return;
+	    }
+
+	  if (*p_end == ';')
+	    p = p_end + 1;
+	  else
+	    break;
+	}
+
+      putpkt ("qsDllInfo");
+      getpkt (&rs->buf, &rs->buf_size, 0);
+    }
+
+  if (strcmp (rs->buf, "l") != 0)
+    {
+      warning (_("Malformed response to DLL query, %s"), rs->buf);
+      return;
+    }
+
+  /* We don't need to call solib_add here, because we're being called from
+     ->current_sos ().  */
+}
+
 /* Stub for catch_exception.  */
 
 static void
@@ -3093,6 +3241,7 @@ remote_wait (ptid_t ptid, struct target_
   struct remote_arch_state *rsa = get_remote_arch_state ();
   ULONGEST thread_num = -1;
   ULONGEST addr;
+  int solibs_changed = 0;
 
   status->kind = TARGET_WAITKIND_EXITED;
   status->value.integer = 0;
@@ -3178,6 +3327,60 @@ Packet: '%s'\n"),
 			p = unpack_varlen_hex (++p1, &addr);
 			remote_watch_data_address = (CORE_ADDR)addr;
 		      }
+		    else if (strncmp (p, "load", p1 - p) == 0)
+		      {
+			p1++;
+			p_temp = p1;
+			while (*p_temp && *p_temp != ';')
+			  p_temp++;
+
+			if (p_temp - p1 > 4
+			    && strncmp (p_temp - 4, ",nop", 4) == 0)
+			  {
+			    /* For now, ignore no-op unload events.  Later,
+			       maybe report them?  */
+			  }
+			else
+			  {
+			    parse_load_response (p1, p_temp, 1);
+
+			    solibs_changed = 1;
+			  }
+
+			p = p_temp;
+		      }
+		    else if (strncmp (p, "unload", p1 - p) == 0)
+		      {
+			p1++;
+			p_temp = p1;
+			while (*p_temp && *p_temp != ';')
+			  p_temp++;
+
+			if (p_temp - p1 > 4
+			    && strncmp (p_temp - 4, ",nop", 4) == 0)
+			  {
+			    /* For now, ignore no-op unload events.  Later,
+			       maybe report them?  */
+			  }
+			else
+			  {
+			    parse_load_response (p1, p_temp, 0);
+
+			    solibs_changed = 1;
+			  }
+
+			p = p_temp;
+		      }
+		    else if (strncmp (p, "dll", p1 - p) == 0)
+		      {
+			p1++;
+			p_temp = p1;
+			while (*p_temp && *p_temp != ';')
+			  p_temp++;
+
+			solibs_changed = -1;
+			p = p_temp;
+		      }
 		    else
  		      {
  			/* Silently skip unknown optional info.  */
@@ -3269,6 +3472,12 @@ Packet: '%s'\n"),
 	}
     }
 got_status:
+  if (solibs_changed == -1)
+    remote_get_shared_libraries (NULL);
+  
+  if (solibs_changed != 0)
+    status->kind = TARGET_WAITKIND_LOADED;
+
   if (thread_num != -1)
     {
       return pid_to_ptid (thread_num);
@@ -6203,6 +6412,7 @@ Specify the serial device it is connecte
   remote_ops.to_flash_erase = remote_flash_erase;
   remote_ops.to_flash_done = remote_flash_done;
   remote_ops.to_read_description = remote_read_description;
+  remote_ops.to_get_shared_libraries = remote_get_shared_libraries;
 }
 
 /* Set up the extended remote vector by making a copy of the standard
@@ -6335,7 +6545,8 @@ Specify the serial device it is connecte
   remote_async_ops.to_memory_map = remote_memory_map;
   remote_async_ops.to_flash_erase = remote_flash_erase;
   remote_async_ops.to_flash_done = remote_flash_done;
-  remote_ops.to_read_description = remote_read_description;
+  remote_async_ops.to_read_description = remote_read_description;
+  remote_async_ops.to_get_shared_libraries = remote_get_shared_libraries;
 }
 
 /* Set up the async extended remote vector by making a copy of the standard
@@ -6601,6 +6812,9 @@ Show the maximum size of the address (in
   add_packet_config_cmd (&remote_protocol_packets[PACKET_qSupported],
 			 "qSupported", "supported-packets", 0);
 
+  add_packet_config_cmd (&remote_protocol_packets[PACKET_qfDllInfo],
+			 "qfDllInfo", "dll-info", 0);
+
   /* Keep the old ``set remote Z-packet ...'' working.  Each individual
      Z sub-packet has its own set and show commands, but users may
      have sets to this variable in their .gdbinit files (or in their
Index: solib-target.c
===================================================================
RCS file: solib-target.c
diff -N solib-target.c
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ solib-target.c	8 Dec 2006 21:09:04 -0000
@@ -0,0 +1,269 @@
+/* Definitions for targets which report shared library events.
+
+   Copyright (C) 2006
+   Free Software Foundation, Inc.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 51 Franklin Street, Fifth Floor,
+   Boston, MA 02110-1301, USA.  */
+
+#include "defs.h"
+#include "solist.h"
+#include "symtab.h"
+#include "symfile.h"
+#include "target.h"
+
+#include "gdb_string.h"
+
+struct lm_info
+{
+  CORE_ADDR textSeg, dataSeg;
+};
+
+static struct so_list *solib_start;
+static int solibs_fetched;
+
+static struct so_list *
+solib_target_current_sos (void)
+{
+  struct so_list *sop;
+  struct so_list *start = NULL;
+  struct so_list *last = NULL;
+
+  /* If we have not asked the target for the list of shared libraries
+     yet, do it now.  */
+  if (!solibs_fetched)
+    {
+      solibs_fetched = 1;
+      target_get_shared_libraries ();
+    }
+
+  for (sop = solib_start; sop; sop = sop->next)
+    {
+      struct so_list *new;
+
+      /* Duplicate the recorded solib.  */
+      new = XZALLOC (struct so_list);
+      strcpy (new->so_name, sop->so_name);
+      strcpy (new->so_original_name, sop->so_original_name);
+      new->lm_info = XMALLOC (struct lm_info);
+      *new->lm_info = *sop->lm_info;
+
+      /* Add it to the list.  */
+      if (!start)
+	last = start = new;
+      else
+	{
+	  last->next = new;
+	  last = new;
+	}
+    }
+
+  return start;
+}
+
+static void
+solib_target_special_symbol_handling (void)
+{
+  /* Nothing needed.  */
+}
+
+static void
+solib_target_solib_create_inferior_hook (void)
+{
+  /* Nothing needed.  */
+}
+
+static void
+solib_target_clear_solib (void)
+{
+  struct so_list *sop, *next;
+
+  for (sop = solib_start; sop; sop = next)
+    {
+      next = sop->next;
+
+      free_so (sop);
+    }
+
+  solib_start = NULL;
+  solibs_fetched = 0;
+}
+
+static void
+solib_target_free_so (struct so_list *so)
+{
+  xfree (so->lm_info);
+}
+
+static void
+solib_target_relocate_section_addresses (struct so_list *so,
+					 struct section_table *sec)
+{
+  int flags = bfd_get_section_flags (sec->bfd, sec->the_bfd_section);
+  CORE_ADDR offset;
+
+  offset = symfile_section_offset_from_segment (sec->bfd, sec->the_bfd_section,
+						so->lm_info->textSeg,
+						so->lm_info->dataSeg);
+
+  sec->addr += offset;
+  sec->endaddr += offset;
+
+  /* If we haven't set these yet, do so now.  If this fails, we may waste some
+     cycles uselessly retrying it, but that is rare and harmless.  */
+  if (so->addr_low == 0 && so->addr_high == 0)
+    {
+      CORE_ADDR text_len, data_len;
+
+      symfile_find_segment_lengths (sec->bfd, &text_len, &data_len);
+
+      if (text_len)
+	{
+	  so->addr_low = so->lm_info->textSeg;
+	  so->addr_high = so->addr_low + text_len;
+	}
+      else if (data_len)
+	{
+	  so->addr_low = so->lm_info->dataSeg;
+	  so->addr_high = so->addr_low + data_len;
+	}
+    }
+}
+
+static int
+solib_target_open_symbol_file_object (void *from_ttyp)
+{
+  /* We can't locate the main symbol file based on the target's
+     knowledge; the user has to specify it.  */
+  return 0;
+}
+
+static int
+solib_target_in_dynsym_resolve_code (CORE_ADDR pc)
+{
+  /* Assume there isn't target dynsym resolution code.  DLL targets
+     generally have only import stubs (which GDB treats as "PLT entries"),
+     and no runtime binding code.  */
+  return 0;
+}
+
+static void
+solib_target_add_one_solib (char *soname, CORE_ADDR textSeg,
+			    CORE_ADDR dataSeg)
+{
+  struct so_list *new_solib, *so;
+
+  /* We should already have queried the target for shared libraries
+     before this point.  If we haven't, we may have just connected;
+     we'll be querying shortly.  */
+  if (!solibs_fetched)
+    return;
+
+  /* Check for duplicates already on the list.  This can happen, for
+     instance, if we are stopped at a DLL load event when we first
+     connect to a remote target: the DLL will already be in the
+     queried list, but also be reported by the initial wait.  */
+  for (so = solib_start; so; so = so->next)
+    if (strcmp (so->so_name, soname) == 0
+	&& so->lm_info->textSeg == textSeg
+	&& so->lm_info->dataSeg == dataSeg)
+      return;
+
+  new_solib = XZALLOC (struct so_list);
+  strncpy (new_solib->so_name, soname, SO_NAME_MAX_PATH_SIZE - 1);
+  new_solib->so_name[SO_NAME_MAX_PATH_SIZE - 1] = '\0';
+  strncpy (new_solib->so_original_name, soname, SO_NAME_MAX_PATH_SIZE - 1);
+  new_solib->so_original_name[SO_NAME_MAX_PATH_SIZE - 1] = '\0';
+
+  new_solib->lm_info = XZALLOC (struct lm_info);
+  new_solib->lm_info->textSeg = textSeg;
+  new_solib->lm_info->dataSeg = dataSeg;
+
+  if (solib_start == NULL)
+    solib_start = new_solib;
+  else
+    {
+      so = solib_start;
+      while (so->next)
+	so = so->next;
+      so->next = new_solib;
+    }
+
+  /* We do not trigger symbol reading here; the target will do it,
+     after all load events have been processed.  */
+}
+
+static void
+solib_target_remove_one_solib (char *soname, CORE_ADDR textSeg,
+			       CORE_ADDR dataSeg)
+{
+  struct so_list **slot, *removed;
+
+  /* We should already have queried the target for shared libraries
+     before this point.  If we haven't, we may have just connected;
+     we'll be querying shortly.  */
+  if (!solibs_fetched)
+    return;
+
+  for (slot = &solib_start; *slot != NULL; slot = &(*slot)->next)
+    {
+      if (textSeg != ~(CORE_ADDR) 0 && textSeg != (*slot)->lm_info->textSeg)
+	continue;
+      if (dataSeg != ~(CORE_ADDR) 0 && dataSeg != (*slot)->lm_info->dataSeg)
+	continue;
+      if (soname != NULL && strcmp (soname, (*slot)->so_name) != 0)
+	continue;
+      break;
+    }
+
+  if (*slot == NULL)
+    return;
+
+  removed = *slot;
+  *slot = removed->next;
+
+  free_so (removed);
+
+  /* We do not trigger symbol unloading here; the target will do it,
+     after all unload events have been processed.  */
+}
+
+static struct target_so_ops solib_target_so_ops;
+
+extern initialize_file_ftype _initialize_solib_target; /* -Wmissing-prototypes */
+
+void
+_initialize_solib_target (void)
+{
+  solib_target_so_ops.relocate_section_addresses
+    = solib_target_relocate_section_addresses;
+  solib_target_so_ops.free_so = solib_target_free_so;
+  solib_target_so_ops.clear_solib = solib_target_clear_solib;
+  solib_target_so_ops.solib_create_inferior_hook
+    = solib_target_solib_create_inferior_hook;
+  solib_target_so_ops.special_symbol_handling
+    = solib_target_special_symbol_handling;
+  solib_target_so_ops.current_sos = solib_target_current_sos;
+  solib_target_so_ops.open_symbol_file_object
+    = solib_target_open_symbol_file_object;
+  solib_target_so_ops.in_dynsym_resolve_code
+    = solib_target_in_dynsym_resolve_code;
+  solib_target_so_ops.add_one_solib = solib_target_add_one_solib;
+  solib_target_so_ops.remove_one_solib = solib_target_remove_one_solib;
+
+  current_target_so_ops = &solib_target_so_ops;
+}
Index: solist.h
===================================================================
RCS file: /cvs/src/src/gdb/solist.h,v
retrieving revision 1.12
diff -u -p -r1.12 solist.h
--- solist.h	17 Dec 2005 22:34:02 -0000	1.12
+++ solist.h	8 Dec 2006 21:09:04 -0000
@@ -104,7 +104,11 @@ struct target_so_ops
        Convenience function for remote debuggers finding host libs.  */
     int (*find_and_open_solib) (char *soname,
         unsigned o_flags, char **temp_pathname);
-    
+
+    void (*add_one_solib) (char *soname, CORE_ADDR textSeg,
+			   CORE_ADDR dataSeg);
+    void (*remove_one_solib) (char *soname, CORE_ADDR textSeg,
+			      CORE_ADDR dataSeg);
   };
 
 /* Free the memory associated with a (so_list *).  */
Index: target.c
===================================================================
RCS file: /cvs/src/src/gdb/target.c,v
retrieving revision 1.131
diff -u -p -r1.131 target.c
--- target.c	5 Dec 2006 20:38:13 -0000	1.131
+++ target.c	8 Dec 2006 21:09:05 -0000
@@ -470,6 +470,7 @@ update_current_target (void)
       /* Do not inherit to_memory_map.  */
       /* Do not inherit to_flash_erase.  */
       /* Do not inherit to_flash_done.  */
+      /* Do not inherit to_get_shared_libraries.  */
     }
 #undef INHERIT
 
@@ -1716,6 +1717,27 @@ target_read_description (struct target_o
 }
 
 /* Look through the list of possible targets for a target that can
+   fetch shared libraries.  */
+
+void
+target_get_shared_libraries (void)
+{
+  struct target_ops *t;
+
+  for (t = current_target.beneath; t != NULL; t = t->beneath)
+    {
+      if (t->to_get_shared_libraries != NULL)
+	{
+	  t->to_get_shared_libraries (t);
+	  break;
+	}
+    }
+
+  if (targetdebug)
+    fprintf_unfiltered (gdb_stdlog, "target_get_shared_libraries ()\n");
+}
+
+/* Look through the list of possible targets for a target that can
    execute a run or attach command without any other data.  This is
    used to locate the default process stratum.
 
Index: target.h
===================================================================
RCS file: /cvs/src/src/gdb/target.h,v
retrieving revision 1.92
diff -u -p -r1.92 target.h
--- target.h	28 Nov 2006 22:10:26 -0000	1.92
+++ target.h	8 Dec 2006 21:09:05 -0000
@@ -497,6 +497,9 @@ struct target_ops
        was available.  */
     const struct target_desc *(*to_read_description) (struct target_ops *ops);
 
+    /* Refresh the list of shared libraries.  */
+    void (*to_get_shared_libraries) (struct target_ops *ops);
+
     int to_magic;
     /* Need sub-structure for target machine related rather than comm related?
      */
@@ -1208,6 +1211,10 @@ extern const struct target_desc *target_
 #define RESUME_EXECD_VFORKING_CHILD_TO_GET_PARENT_VFORK() (0)
 #endif
 
+/* Refresh the list of shared libraries from the target.  */
+
+extern void target_get_shared_libraries (void);
+
 /* Routines for maintenance of the target structures...
 
    add_target:   Add a target to the list of all possible targets.

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

Thanks,  

Now I have a couple of questions.

1)  I take it that there isn't anything special here with regard to DLL
(PE binary files) vs. SO (ELF binraries)

2)  You have HEXSTR in a couple places, since I am not familiar with
this representation, how would foo.dll be coded in HEXSTR format.

I think that the rest is self explanatory.


-----Original Message-----
From: Daniel Jacobowitz [mailto:drow@false.org] 
Sent: Friday, December 08, 2006 2:19 PM
To: Smith, Stephen (SWCOE); General GDB Mailing List
Subject: Re: GDB solib interface

Re: GDB solib interface

[Daniel Jacobowitz]

On Fri, Dec 08, 2006 at 03:23:01PM -0700, Smith, Stephen (SWCOE) wrote:
> Thanks,  
> 
> Now I have a couple of questions.
> 
> 1)  I take it that there isn't anything special here with regard to DLL
> (PE binary files) vs. SO (ELF binraries)

I've only used it with ELF, despite the DLL naming.

> 2)  You have HEXSTR in a couple places, since I am not familiar with
> this representation, how would foo.dll be coded in HEXSTR format.

Just convert the ascii to base 16, two digits per letter.

-- 
Daniel Jacobowitz
CodeSourcery

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

Ok,  I build gdb for to different targets,  i686-pc-elf and 
powerpc-motorola-elf.  My host is i686-pc-cygwin.

So the question becomes how do I decide which file to modify for the
TDEPFILES line?

-----Original Message-----
From: Daniel Jacobowitz [mailto:drow@false.org] 
Sent: Friday, December 08, 2006 2:19 PM
To: Smith, Stephen (SWCOE); General GDB Mailing List
Subject: Re: GDB solib interface

Here you go.  I haven't tested this patch on its own at all, so if it
breaks, you get to keep both pieces.  Add solib-target.o to TDEPFILES
for your target to enable it.

Re: GDB solib interface

[Daniel Jacobowitz]

On Fri, Dec 15, 2006 at 02:12:17PM -0700, Smith, Stephen (SWCOE) wrote:
> Ok,  I build gdb for to different targets,  i686-pc-elf and 
> powerpc-motorola-elf.  My host is i686-pc-cygwin.
> 
> So the question becomes how do I decide which file to modify for the
> TDEPFILES line?

See which one the build is using by checking gdb/configure.tgt.

-- 
Daniel Jacobowitz
CodeSourcery

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

The function solib_target_relocate_section_addresses() refers to
functions and datamembers that no longer exist.  I believe that
borrowing code from the svr4 code is the correct thing to do.

Can anyone confirm that that is the right thing?


-----Original Message-----
From: Daniel Jacobowitz [mailto:drow@false.org] 
Sent: Friday, December 08, 2006 2:19 PM
To: Smith, Stephen (SWCOE); General GDB Mailing List
Subject: Re: GDB solib interface

+static void
+solib_target_free_so (struct so_list *so)
+{
+  xfree (so->lm_info);
+}
+
+static void
+solib_target_relocate_section_addresses (struct so_list *so,
+					 struct section_table *sec)
+{
+  int flags = bfd_get_section_flags (sec->bfd, sec->the_bfd_section);
+  CORE_ADDR offset;
+
+  offset = symfile_section_offset_from_segment (sec->bfd,
sec->the_bfd_section,
+						so->lm_info->textSeg,
+						so->lm_info->dataSeg);
+

Re: GDB solib interface

[Daniel Jacobowitz]

On Sat, Dec 16, 2006 at 05:42:39PM -0700, Smith, Stephen (SWCOE) wrote:
> The function solib_target_relocate_section_addresses() refers to
> functions and datamembers that no longer exist.  I believe that
> borrowing code from the svr4 code is the correct thing to do.
> 
> Can anyone confirm that that is the right thing?

Sorry, it's actually the other way round.  It's not that the things it
refers to no longer exist - they've never existed in the FSF source,
they were part of another patch from the tree I took this from.

I think the equivalent would be initializing so->textsection.

-- 
Daniel Jacobowitz
CodeSourcery

Re: GDB solib interface

[Stephen & Linda Smith]

Maybe I got my answer, but I'm not sure.  Should I lift the code from 
the solib-svr4.c file? 

Daniel Jacobowitz wrote:
> I think the equivalent would be initializing so->textsection.
>
>   

Re: GDB solib interface

[Daniel Jacobowitz]

On Sat, Dec 16, 2006 at 11:41:16PM -0700, Stephen & Linda Smith wrote:
> Maybe I got my answer, but I'm not sure.  Should I lift the code from 
> the solib-svr4.c file? 

I don't know.

-- 
Daniel Jacobowitz
CodeSourcery

Re: GDB solib interface

[Stephen & Linda Smith]

Ok, I modified my gdbserver to conform to the interface and wrote a stub 
main executable and shared library.  The share library symbol file is 
now loaded when gdb connects to the gdbserver, but my guess is that 
solib_target_relocate_section_addresses() needs to do more than just set 
so->textsection. 

The reason is that gdb thinks that the test shared library function is 
somewhere around 0x6* hex.    Specifically,  gdb is reporting that the 
functions reside at the offsets I get when I do an objdump -xt of the 
library. 

Does anyone have an idea of what variables I need to change to make the 
symbol file line up with the load on the remote target. 

I've attached the diff.  The previous one on this list does not compile 
cleanly.  This one does, but has the above mentions problem.

=================== start of diff ==========================
diff -ruN insight-6.5/gdb/Makefile.in insight-6.5-with-patch/gdb/Makefile.in
--- insight-6.5/gdb/Makefile.in    2006-05-12 13:53:15.000000000 -0700
+++ insight-6.5-with-patch/gdb/Makefile.in    2006-12-15 
16:30:06.546875000 -0700
@@ -2467,7 +2467,7 @@
     $(gdb_stabs_h) $(gdbthread_h) $(remote_h) $(regcache_h) $(value_h) \
     $(gdb_assert_h) $(event_loop_h) $(event_top_h) $(inf_loop_h) \
     $(serial_h) $(gdbcore_h) $(remote_fileio_h) $(solib_h) $(observer_h) \
-    $(cli_decode_h) $(cli_setshow_h)
+    $(cli_decode_h) $(cli_setshow_h) $(solist_h)
 remote-e7000.o: remote-e7000.c $(defs_h) $(gdbcore_h) $(gdbarch_h) \
     $(inferior_h) $(target_h) $(value_h) $(command_h) $(gdb_string_h) \
     $(exceptions_h) $(gdbcmd_h) $(serial_h) $(remote_utils_h) \
@@ -2603,6 +2603,8 @@
     $(gdbcore_h) $(target_h) $(inferior_h) $(gdb_assert_h) \
     $(solist_h) $(solib_h) $(solib_svr4_h) $(bfd_target_h) $(elf_bfd_h) \
     $(exec_h)
+solib-target.o: solib-target.c $(defs_h) $(solist_h) $(symtab_h) 
$(symfile_h) \
+    $(target_h) $(gdb_string_h)
 sol-thread.o: sol-thread.c $(defs_h) $(gdbthread_h) $(target_h) \
     $(inferior_h) $(gdb_stat_h) $(gdbcmd_h) $(gdbcore_h) $(regcache_h) \
     $(solib_h) $(symfile_h) $(gdb_string_h) $(gregset_h)
diff -ruN insight-6.5/gdb/config/i386/#i386gnu.mt# 
insight-6.5-with-patch/gdb/config/i386/#i386gnu.mt#
--- insight-6.5/gdb/config/i386/#i386gnu.mt#    1969-12-31 
17:00:00.000000000 -0700
+++ insight-6.5-with-patch/gdb/config/i386/#i386gnu.mt#    2006-12-15 
16:45:03.593750000 -0700
@@ -0,0 +1,2 @@
+# Target: Intel 386 running the GNU Hurd
+TDEPFILES= i386-tdep.o i387-tdep.o i386gnu-tdep.o
diff -ruN insight-6.5/gdb/config/i386/linux.mt 
insight-6.5-with-patch/gdb/config/i386/linux.mt
--- insight-6.5/gdb/config/i386/linux.mt    2004-09-13 
13:55:39.000000000 -0700
+++ insight-6.5-with-patch/gdb/config/i386/linux.mt    2006-12-15 
16:44:16.812500000 -0700
@@ -1,4 +1,4 @@
 # Target: Intel 386 running GNU/Linux
 TDEPFILES= i386-tdep.o i386-linux-tdep.o glibc-tdep.o i387-tdep.o \
-    solib.o solib-svr4.o symfile-mem.o
+    solib.o solib-svr4.o solib-target.o symfile-mem.o
 DEPRECATED_TM_FILE= tm-linux.h
diff -ruN insight-6.5/gdb/config/powerpc/ppc-sim.mt 
insight-6.5-with-patch/gdb/config/powerpc/ppc-sim.mt
--- insight-6.5/gdb/config/powerpc/ppc-sim.mt    2004-11-13 
18:11:07.000000000 -0700
+++ insight-6.5-with-patch/gdb/config/powerpc/ppc-sim.mt    2006-12-16 
14:54:12.906250000 -0700
@@ -1,5 +1,5 @@
 # Target: PowerPC running eabi and including the simulator
-TDEPFILES= rs6000-tdep.o monitor.o dsrec.o ppcbug-rom.o dink32-rom.o 
ocd.o ppc-sysv-tdep.o solib.o solib-svr4.o
+TDEPFILES= rs6000-tdep.o monitor.o dsrec.o ppcbug-rom.o dink32-rom.o 
ocd.o ppc-sysv-tdep.o solib.o solib-svr4.o solib-target.o
 DEPRECATED_TM_FILE= tm-ppc-eabi.h
 
 SIM_OBS = remote-sim.o
diff -ruN insight-6.5/gdb/infrun.c insight-6.5-with-patch/gdb/infrun.c
--- insight-6.5/gdb/infrun.c    2006-06-15 18:14:50.000000000 -0700
+++ insight-6.5-with-patch/gdb/infrun.c    2006-12-15 16:30:06.593750000 
-0700
@@ -877,8 +877,10 @@
   clear_proceed_status ();
 
   stepping_past_singlestep_breakpoint = 0;
+
+  target_last_wait_ptid = minus_one_ptid;
 }
-\f
+
 /* This enum encodes possible reasons for doing a target_wait, so that
    wfi can call target_wait in one place.  (Ultimately the call will be
    moved out of the infinite loop entirely.) */
@@ -1255,6 +1257,9 @@
   target_last_wait_ptid = ecs->ptid;
   target_last_waitstatus = *ecs->wp;
 
+  /* Always clear state belonging to the previous time we stopped.  */ 
+  stop_stack_dummy = 0;
+
   adjust_pc_after_break (ecs);
 
   switch (ecs->infwait_state)
@@ -1315,13 +1320,16 @@
       /* Ignore gracefully during startup of the inferior, as it
          might be the shell which has just loaded some objects,
          otherwise add the symbols for the newly loaded objects.  */
-#ifdef SOLIB_ADD
       if (stop_soon == NO_STOP_QUIETLY)
-    {
+     {
+      int breakpoints_were_inserted;
+
       /* Remove breakpoints, SOLIB_ADD might adjust
          breakpoint addresses via breakpoint_re_set.  */
+      breakpoints_were_inserted = breakpoints_inserted;
       if (breakpoints_inserted)
         remove_breakpoints ();
+      breakpoints_inserted = 0;
 
       /* Check for any newly added shared libraries if we're
          supposed to be adding them automatically.  Switch
@@ -1343,17 +1351,52 @@
          exec/process stratum, instead relying on the target stack
          to propagate relevant changes (stop, section table
          changed, ...) up to other layers.  */
+#ifdef SOLIB_ADD
       SOLIB_ADD (NULL, 0, &current_target, auto_solib_add);
+#else
+      solib_add (NULL, 0, &current_target, auto_solib_add);
+#endif
       target_terminal_inferior ();
 
-      /* Reinsert breakpoints and continue.  */
-      if (breakpoints_inserted)
-        insert_breakpoints ();
+      /* Try to reenable shared library breakpoints, additional
+         code segments in shared libraries might be mapped in now. */
+      re_enable_breakpoints_in_shlibs ();
+
+      /* If requested, stop when the dynamic linker notifies
+         gdb of events.  This allows the user to get control
+         and place breakpoints in initializer routines for
+         dynamically loaded objects (among other things).  */
+      if (stop_on_solib_events)
+        {
+          stop_stepping (ecs);
+          return;
+        }
+
+      /* NOTE drow/2006-03-14: This might be a good place to check
+         for "catch load".  */
+
+      /* Reinsert breakpoints and continue.  */
+      if (breakpoints_were_inserted)
+        {
+          insert_breakpoints ();
+          breakpoints_inserted = 1;
+        }
     }
-#endif
-      resume (0, TARGET_SIGNAL_0);
-      prepare_to_wait (ecs);
-      return;
+
+     /* NOTE drow/2006-03-28: For the reason described before the
+     previous if statement, GDB used to automatically resume
+     here.  But that's only true if a shell is running; if
+     we've just attached to a process, then that's a whole
+     different case - it might have been stopped at a load
+     event.  */
+      if (inferior_ignoring_startup_exec_events || stop_soon == 
NO_STOP_QUIETLY)
+    {
+      resume (0, TARGET_SIGNAL_0);
+      prepare_to_wait (ecs);
+      return;
+    }
+      else
+    break;
 
     case TARGET_WAITKIND_SPURIOUS:
       if (debug_infrun)
@@ -1760,7 +1803,6 @@
   ecs->another_trap = 0;
   bpstat_clear (&stop_bpstat);
   stop_step = 0;
-  stop_stack_dummy = 0;
   stop_print_frame = 1;
   ecs->random_signal = 0;
   stopped_by_random_signal = 0;
diff -ruN insight-6.5/gdb/remote.c insight-6.5-with-patch/gdb/remote.c
--- insight-6.5/gdb/remote.c    2006-05-05 13:08:45.000000000 -0700
+++ insight-6.5-with-patch/gdb/remote.c    2006-12-16 13:11:38.609375000 
-0700
@@ -43,6 +43,7 @@
 #include "gdb_assert.h"
 #include "observer.h"
 #include "solib.h"
+#include "solist.h"
 #include "cli/cli-decode.h"
 #include "cli/cli-setshow.h"
 
@@ -765,6 +766,7 @@
   PACKET_Z4,
   PACKET_qPart_auxv,
   PACKET_qGetTLSAddr,
+  PACKET_qfDllInfo,
   PACKET_MAX
 };
 
@@ -1888,6 +1890,150 @@
   objfile_relocate (symfile_objfile, offs);
 }
 
+/* Parse a load or unload message for a DLL.  Loads are required to
+   have all fields, unloads must have at least one.  P is not
+   necessarily NUL terminated, but it is always either NUL or semicolon
+   terminated (i.e. *P_END is either NUL or a semicolon) and the containing
+   string is NUL terminated.  */
+
+int
+parse_load_response (const char *p, const char *p_end, int is_load) {
+  const char *name_start;
+  char *name;
+  CORE_ADDR text_addr, data_addr;
+  int i;
+
+  name = NULL;
+  text_addr = ~(CORE_ADDR) 0;
+  data_addr = ~(CORE_ADDR) 0;
+
+  if (p < p_end && strncmp (p, "Name=", 5) == 0)
+    {
+      p += 5;
+
+      name_start = p;
+      while (p < p_end && *p != ',')
+    p++;
+
+      if ((p - name_start) % 2 != 0)
+    return -1;
+
+      name = alloca ((p - name_start) / 2 + 1);
+      i = hex2bin (name_start, name, (p - name_start) / 2);
+      name[i] = '\0';
+
+      if (p < p_end)
+    p++;
+    }
+  else if (is_load)
+    return -1;
+
+  if (p < p_end && strncmp (p, "TextSeg=", 8) == 0)
+    {
+      p += 8;
+
+      text_addr = 0;
+      while (p < p_end && *p != ',')
+    text_addr = (text_addr << 4) + fromhex (*p++);
+
+      if (p < p_end)
+    p++;
+    }
+  else if (is_load)
+    return -1;
+
+  if (p < p_end && strncmp (p, "DataSeg=", 8) == 0)
+    {
+      p += 8;
+
+      data_addr = 0;
+      while (p < p_end && *p != ',')
+    data_addr = (data_addr << 4) + fromhex (*p++);
+
+      if (p < p_end)
+    p++;
+    }
+  else if (is_load)
+    return -1;
+
+  if (is_load)
+    current_target_so_ops->add_one_solib (name, text_addr, data_addr);
+  else
+    {
+      if (text_addr == ~(CORE_ADDR) 0
+      && data_addr == ~(CORE_ADDR) 0
+      && name == NULL)
+    return -1;
+
+      current_target_so_ops->remove_one_solib (name, text_addr, data_addr);
+    }
+
+  return 0;
+}
+
+/* Query the remote side for loaded solibs.  */
+
+static void
+remote_get_shared_libraries (struct target_ops *ops) {
+  struct remote_state *rs = get_remote_state ();
+
+  /* If this target doesn't support remote DLLs, nothing to do.  */  if
+ (current_target_so_ops->add_one_solib == NULL)
+    return;
+
+  /* If qfDllInfo is not available, nothing to do.  */  if
+ (remote_protocol_packets[PACKET_qfDllInfo].support == PACKET_DISABLE)
+    return;
+
+  putpkt ("qfDllInfo");
+  getpkt (&rs->buf, &rs->buf_size, 0);
+
+  if (packet_ok (rs->buf, &remote_protocol_packets[PACKET_qfDllInfo]) 
== PACKET_ERROR)
+    {
+      warning (_("Remote failure reply: %s"), rs->buf);
+      return;
+    }
+  else if (remote_protocol_packets[PACKET_qfDllInfo].support == 
PACKET_DISABLE)
+    /* It wasn't disabled before, but it is now.  */
+    return;
+
+  while (rs->buf[0] == 'm')
+    {
+      char *p = rs->buf + 1;
+
+      while (1)
+    {
+      char *p_end = p;
+
+      while (*p_end && *p_end != ';')
+        p_end++;
+
+      if (parse_load_response (p, p_end, 1) != 0)
+        {
+          warning (_("Malformed response to DLL query, %s"), rs->buf);
+          return;
+        }
+
+      if (*p_end == ';')
+        p = p_end + 1;
+      else
+        break;
+    }
+
+      putpkt ("qsDllInfo");
+      getpkt (&rs->buf, &rs->buf_size, 0);
+    }
+
+  if (strcmp (rs->buf, "l") != 0)
+    {
+      warning (_("Malformed response to DLL query, %s"), rs->buf);
+      return;
+    }
+
+  /* We don't need to call solib_add here, because we're being called from
+     ->current_sos ().  */
+}
+
 /* Stub for catch_errors.  */
 
 static int
@@ -2671,6 +2817,7 @@
   char *buf = rs->buf;
   ULONGEST thread_num = -1;
   ULONGEST addr;
+  int solibs_changed = 0;
 
   status->kind = TARGET_WAITKIND_EXITED;
   status->value.integer = 0;
@@ -2754,6 +2901,60 @@
             p = unpack_varlen_hex (++p1, &addr);
             remote_watch_data_address = (CORE_ADDR)addr;
               }
+            else if (strncmp (p, "load", p1 - p) == 0)
+              {
+            p1++;
+            p_temp = p1;
+            while (*p_temp && *p_temp != ';')
+              p_temp++;
+
+            if (p_temp - p1 > 4
+                && strncmp (p_temp - 4, ",nop", 4) == 0)
+              {
+                /* For now, ignore no-op unload events.  Later,
+                   maybe report them?  */
+              }
+            else
+              {
+                parse_load_response (p1, p_temp, 1);
+
+                solibs_changed = 1;
+              }
+
+            p = p_temp;
+              }
+            else if (strncmp (p, "unload", p1 - p) == 0)
+              {
+            p1++;
+            p_temp = p1;
+            while (*p_temp && *p_temp != ';')
+              p_temp++;
+
+            if (p_temp - p1 > 4
+                && strncmp (p_temp - 4, ",nop", 4) == 0)
+              {
+                /* For now, ignore no-op unload events.  Later,
+                   maybe report them?  */
+              }
+            else
+              {
+                parse_load_response (p1, p_temp, 0);
+
+                solibs_changed = 1;
+              }
+
+            p = p_temp;
+              }
+            else if (strncmp (p, "dll", p1 - p) == 0)
+              {
+            p1++;
+            p_temp = p1;
+            while (*p_temp && *p_temp != ';')
+              p_temp++;
+
+            solibs_changed = -1;
+            p = p_temp;
+              }
             else
                {
              /* Silently skip unknown optional info.  */
@@ -2845,6 +3046,12 @@
     }
     }
 got_status:
+  if (solibs_changed == -1)
+    remote_get_shared_libraries (NULL);
+ 
+  if (solibs_changed != 0)
+    status->kind = TARGET_WAITKIND_LOADED;
+
   if (thread_num != -1)
     {
       return pid_to_ptid (thread_num);
@@ -5229,6 +5436,7 @@
 (e.g. /dev/ttyS0, /dev/ttya, COM1, etc.).";
   remote_ops.to_open = remote_open;
   remote_ops.to_close = remote_close;
+  remote_ops.to_get_shared_libraries = remote_get_shared_libraries;
   remote_ops.to_detach = remote_detach;
   remote_ops.to_disconnect = remote_disconnect;
   remote_ops.to_resume = remote_resume;
@@ -5354,6 +5562,7 @@
   remote_async_ops.to_open = remote_async_open;
   remote_async_ops.to_close = remote_close;
   remote_async_ops.to_detach = remote_detach;
+  remote_async_ops.to_get_shared_libraries = remote_get_shared_libraries;
   remote_async_ops.to_disconnect = remote_disconnect;
   remote_async_ops.to_resume = remote_async_resume;
   remote_async_ops.to_wait = remote_async_wait;
@@ -5616,6 +5825,13 @@
              &remote_set_cmdlist, &remote_show_cmdlist,
              0);
 
+  add_packet_config_cmd (&remote_protocol_packets[PACKET_qfDllInfo],
+             "qfDllInfo", "dll-info",
+             set_remote_protocol_packet_cmd,
+             show_remote_protocol_packet_cmd,
+             &remote_set_cmdlist, &remote_show_cmdlist,
+             0);
+
   add_packet_config_cmd (&remote_protocol_packets[PACKET_P],
              "P", "set-register",
              set_remote_protocol_packet_cmd,
diff -ruN insight-6.5/gdb/solib-target.c 
insight-6.5-with-patch/gdb/solib-target.c
--- insight-6.5/gdb/solib-target.c    1969-12-31 17:00:00.000000000 -0700
+++ insight-6.5-with-patch/gdb/solib-target.c    2006-12-29 
12:35:17.453125000 -0700
@@ -0,0 +1,238 @@
+/* Definitions for targets which report shared library events.
+
+   Copyright (C) 2006
+   Free Software Foundation, Inc.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 51 Franklin Street, Fifth Floor,
+   Boston, MA 02110-1301, USA.  */
+
+#include "defs.h"
+#include "solist.h"
+#include "symtab.h"
+#include "symfile.h"
+#include "target.h"
+
+#include "gdb_string.h"
+
+struct lm_info
+{
+  CORE_ADDR textSeg, dataSeg;
+};
+
+static struct so_list *solib_start;
+static int solibs_fetched;
+
+static struct so_list *
+solib_target_current_sos (void)
+{
+  struct so_list *sop;
+  struct so_list *start = NULL;
+  struct so_list *last = NULL;
+
+  /* If we have not asked the target for the list of shared libraries
+     yet, do it now.  */
+  if (!solibs_fetched)
+    {
+      solibs_fetched = 1;
+      target_get_shared_libraries ();
+    }
+
+  for (sop = solib_start; sop; sop = sop->next)
+    {
+      struct so_list *new;
+
+      /* Duplicate the recorded solib.  */
+      new = XZALLOC (struct so_list);
+      strcpy (new->so_name, sop->so_name);
+      strcpy (new->so_original_name, sop->so_original_name);
+      new->lm_info = XMALLOC (struct lm_info);
+      *new->lm_info = *sop->lm_info;
+
+      /* Add it to the list.  */
+      if (!start)
+    last = start = new;
+      else
+    {
+      last->next = new;
+      last = new;
+    }
+    }
+
+  return start;
+}
+
+static void
+solib_target_special_symbol_handling (void) {
+  /* Nothing needed.  */
+}
+
+static void
+solib_target_solib_create_inferior_hook (void) {
+  /* Nothing needed.  */
+}
+
+static void
+solib_target_clear_solib (void)
+{
+  struct so_list *sop, *next;
+
+  for (sop = solib_start; sop; sop = next)
+    {
+      next = sop->next;
+
+      free_so (sop);
+    }
+
+  solib_start = NULL;
+  solibs_fetched = 0;
+}
+
+static void
+solib_target_free_so (struct so_list *so) {
+  xfree (so->lm_info);
+}
+
+static void
+solib_target_relocate_section_addresses (struct so_list *so,
+                     struct section_table *sec)
+{
+  so->textsection = (void *)so->lm_info->textSeg;
+}
+
+static int
+solib_target_open_symbol_file_object (void *from_ttyp) {
+  /* We can't locate the main symbol file based on the target's
+     knowledge; the user has to specify it.  */
+  return 0;
+}
+
+static int
+solib_target_in_dynsym_resolve_code (CORE_ADDR pc) {
+  /* Assume there isn't target dynsym resolution code.  DLL targets
+     generally have only import stubs (which GDB treats as "PLT entries"),
+     and no runtime binding code.  */
+  return 0;
+}
+
+static void
+solib_target_add_one_solib (char *soname, CORE_ADDR textSeg,
+                CORE_ADDR dataSeg)
+{
+  struct so_list *new_solib, *so;
+
+  /* We should already have queried the target for shared libraries
+     before this point.  If we haven't, we may have just connected;
+     we'll be querying shortly.  */
+  if (!solibs_fetched)
+    return;
+
+  /* Check for duplicates already on the list.  This can happen, for
+     instance, if we are stopped at a DLL load event when we first
+     connect to a remote target: the DLL will already be in the
+     queried list, but also be reported by the initial wait.  */
+  for (so = solib_start; so; so = so->next)
+    if (strcmp (so->so_name, soname) == 0
+    && so->lm_info->textSeg == textSeg
+    && so->lm_info->dataSeg == dataSeg)
+      return;
+
+  new_solib = XZALLOC (struct so_list);  strncpy (new_solib->so_name,
+ soname, SO_NAME_MAX_PATH_SIZE - 1); 
+ new_solib->so_name[SO_NAME_MAX_PATH_SIZE - 1] = '\0';  strncpy
+ (new_solib->so_original_name, soname, SO_NAME_MAX_PATH_SIZE - 1); 
+ new_solib->so_original_name[SO_NAME_MAX_PATH_SIZE - 1] = '\0';
+
+  new_solib->lm_info = XZALLOC (struct lm_info); 
+ new_solib->lm_info->textSeg = textSeg;  new_solib->lm_info->dataSeg =
+ dataSeg;
+
+  if (solib_start == NULL)
+    solib_start = new_solib;
+  else
+    {
+      so = solib_start;
+      while (so->next)
+    so = so->next;
+      so->next = new_solib;
+    }
+
+  /* We do not trigger symbol reading here; the target will do it,
+     after all load events have been processed.  */
+}
+
+static void
+solib_target_remove_one_solib (char *soname, CORE_ADDR textSeg,
+                   CORE_ADDR dataSeg)
+{
+  struct so_list **slot, *removed;
+
+  /* We should already have queried the target for shared libraries
+     before this point.  If we haven't, we may have just connected;
+     we'll be querying shortly.  */
+  if (!solibs_fetched)
+    return;
+
+  for (slot = &solib_start; *slot != NULL; slot = &(*slot)->next)
+    {
+      if (textSeg != ~(CORE_ADDR) 0 && textSeg != 
(*slot)->lm_info->textSeg)
+    continue;
+      if (dataSeg != ~(CORE_ADDR) 0 && dataSeg != 
(*slot)->lm_info->dataSeg)
+    continue;
+      if (soname != NULL && strcmp (soname, (*slot)->so_name) != 0)
+    continue;
+      break;
+    }
+
+  if (*slot == NULL)
+    return;
+
+  removed = *slot;
+  *slot = removed->next;
+
+  free_so (removed);
+
+  /* We do not trigger symbol unloading here; the target will do it,
+     after all unload events have been processed.  */
+}
+
+static struct target_so_ops solib_target_so_ops;
+
+extern initialize_file_ftype _initialize_solib_target; /*
+-Wmissing-prototypes */
+
+void
+_initialize_solib_target (void)
+{
+  solib_target_so_ops.relocate_section_addresses
+    = solib_target_relocate_section_addresses;
+  solib_target_so_ops.free_so = solib_target_free_so;
+  solib_target_so_ops.clear_solib = solib_target_clear_solib;
+  solib_target_so_ops.solib_create_inferior_hook
+    = solib_target_solib_create_inferior_hook;
+  solib_target_so_ops.special_symbol_handling
+    = solib_target_special_symbol_handling;
+  solib_target_so_ops.current_sos = solib_target_current_sos;
+  solib_target_so_ops.open_symbol_file_object
+    = solib_target_open_symbol_file_object;
+  solib_target_so_ops.in_dynsym_resolve_code
+    = solib_target_in_dynsym_resolve_code;
+  solib_target_so_ops.add_one_solib = solib_target_add_one_solib;
+  solib_target_so_ops.remove_one_solib = solib_target_remove_one_solib;
+
+  current_target_so_ops = &solib_target_so_ops;
+}
+
diff -ruN insight-6.5/gdb/solib-target.c~ 
insight-6.5-with-patch/gdb/solib-target.c~
--- insight-6.5/gdb/solib-target.c~    1969-12-31 17:00:00.000000000 -0700
+++ insight-6.5-with-patch/gdb/solib-target.c~    2006-12-19 
09:04:15.109375000 -0700
@@ -0,0 +1,236 @@
+/* Definitions for targets which report shared library events.
+
+   Copyright (C) 2006
+   Free Software Foundation, Inc.
+
+   This file is part of GDB.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 2 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 51 Franklin Street, Fifth Floor,
+   Boston, MA 02110-1301, USA.  */
+
+#include "defs.h"
+#include "solist.h"
+#include "symtab.h"
+#include "symfile.h"
+#include "target.h"
+
+#include "gdb_string.h"
+
+struct lm_info
+{
+  CORE_ADDR textSeg, dataSeg;
+};
+
+static struct so_list *solib_start;
+static int solibs_fetched;
+
+static struct so_list *
+solib_target_current_sos (void)
+{
+  struct so_list *sop;
+  struct so_list *start = NULL;
+  struct so_list *last = NULL;
+
+  /* If we have not asked the target for the list of shared libraries
+     yet, do it now.  */
+  if (!solibs_fetched)
+    {
+      solibs_fetched = 1;
+      target_get_shared_libraries ();
+    }
+
+  for (sop = solib_start; sop; sop = sop->next)
+    {
+      struct so_list *new;
+
+      /* Duplicate the recorded solib.  */
+      new = XZALLOC (struct so_list);
+      strcpy (new->so_name, sop->so_name);
+      strcpy (new->so_original_name, sop->so_original_name);
+      new->lm_info = XMALLOC (struct lm_info);
+      *new->lm_info = *sop->lm_info;
+
+      /* Add it to the list.  */
+      if (!start)
+    last = start = new;
+      else
+    {
+      last->next = new;
+      last = new;
+    }
+    }
+
+  return start;
+}
+
+static void
+solib_target_special_symbol_handling (void) {
+  /* Nothing needed.  */
+}
+
+static void
+solib_target_solib_create_inferior_hook (void) {
+  /* Nothing needed.  */
+}
+
+static void
+solib_target_clear_solib (void)
+{
+  struct so_list *sop, *next;
+
+  for (sop = solib_start; sop; sop = next)
+    {
+      next = sop->next;
+
+      free_so (sop);
+    }
+
+  solib_start = NULL;
+  solibs_fetched = 0;
+}
+
+static void
+solib_target_free_so (struct so_list *so) {
+  xfree (so->lm_info);
+}
+
+static void
+solib_target_relocate_section_addresses (struct so_list *so,
+                     struct section_table *sec)
+{
+  so->textsection = (void *)so->lm_info->textSeg;
+}
+
+static int
+solib_target_open_symbol_file_object (void *from_ttyp) {
+  /* We can't locate the main symbol file based on the target's
+     knowledge; the user has to specify it.  */
+  return 0;
+}
+
+static int
+solib_target_in_dynsym_resolve_code (CORE_ADDR pc) {
+  /* Assume there isn't target dynsym resolution code.  DLL targets
+     generally have only import stubs (which GDB treats as "PLT entries"),
+     and no runtime binding code.  */
+  return 0;
+}
+
+static void
+solib_target_add_one_solib (char *soname, CORE_ADDR textSeg,
+                CORE_ADDR dataSeg)
+{
+  struct so_list *new_solib, *so;
+
+  /* We should already have queried the target for shared libraries
+     before this point.  If we haven't, we may have just connected;
+     we'll be querying shortly.  */
+  if (!solibs_fetched)
+    return;
+
+  /* Check for duplicates already on the list.  This can happen, for
+     instance, if we are stopped at a DLL load event when we first
+     connect to a remote target: the DLL will already be in the
+     queried list, but also be reported by the initial wait.  */
+  for (so = solib_start; so; so = so->next)
+    if (strcmp (so->so_name, soname) == 0
+    && so->lm_info->textSeg == textSeg
+    && so->lm_info->dataSeg == dataSeg)
+      return;
+
+  new_solib = XZALLOC (struct so_list);  strncpy (new_solib->so_name,
+ soname, SO_NAME_MAX_PATH_SIZE - 1); 
+ new_solib->so_name[SO_NAME_MAX_PATH_SIZE - 1] = '\0';  strncpy
+ (new_solib->so_original_name, soname, SO_NAME_MAX_PATH_SIZE - 1); 
+ new_solib->so_original_name[SO_NAME_MAX_PATH_SIZE - 1] = '\0';
+
+  new_solib->lm_info = XZALLOC (struct lm_info); 
+ new_solib->lm_info->textSeg = textSeg;  new_solib->lm_info->dataSeg =
+ dataSeg;
+
+  if (solib_start == NULL)
+    solib_start = new_solib;
+  else
+    {
+      so = solib_start;
+      while (so->next)
+    so = so->next;
+      so->next = new_solib;
+    }
+
+  /* We do not trigger symbol reading here; the target will do it,
+     after all load events have been processed.  */ }
+
+static void
+solib_target_remove_one_solib (char *soname, CORE_ADDR textSeg,
+                   CORE_ADDR dataSeg)
+{
+  struct so_list **slot, *removed;
+
+  /* We should already have queried the target for shared libraries
+     before this point.  If we haven't, we may have just connected;
+     we'll be querying shortly.  */
+  if (!solibs_fetched)
+    return;
+
+  for (slot = &solib_start; *slot != NULL; slot = &(*slot)->next)
+    {
+      if (textSeg != ~(CORE_ADDR) 0 && textSeg != 
(*slot)->lm_info->textSeg)
+    continue;
+      if (dataSeg != ~(CORE_ADDR) 0 && dataSeg != 
(*slot)->lm_info->dataSeg)
+    continue;
+      if (soname != NULL && strcmp (soname, (*slot)->so_name) != 0)
+    continue;
+      break;
+    }
+
+  if (*slot == NULL)
+    return;
+
+  removed = *slot;
+  *slot = removed->next;
+
+  free_so (removed);
+
+  /* We do not trigger symbol unloading here; the target will do it,
+     after all unload events have been processed.  */ }
+
+static struct target_so_ops solib_target_so_ops;
+
+extern initialize_file_ftype _initialize_solib_target; /*
+-Wmissing-prototypes */
+
+void
+_initialize_solib_target (void)
+{
+  solib_target_so_ops.relocate_section_addresses
+    = solib_target_relocate_section_addresses;
+  solib_target_so_ops.free_so = solib_target_free_so;
+  solib_target_so_ops.clear_solib = solib_target_clear_solib;
+  solib_target_so_ops.solib_create_inferior_hook
+    = solib_target_solib_create_inferior_hook;
+  solib_target_so_ops.special_symbol_handling
+    = solib_target_special_symbol_handling;
+  solib_target_so_ops.current_sos = solib_target_current_sos;
+  solib_target_so_ops.open_symbol_file_object
+    = solib_target_open_symbol_file_object;
+  solib_target_so_ops.in_dynsym_resolve_code
+    = solib_target_in_dynsym_resolve_code;
+  solib_target_so_ops.add_one_solib = solib_target_add_one_solib;
+  solib_target_so_ops.remove_one_solib = solib_target_remove_one_solib;
+
+  current_target_so_ops = &solib_target_so_ops;
+}
+
diff -ruN insight-6.5/gdb/solist.h insight-6.5-with-patch/gdb/solist.h
--- insight-6.5/gdb/solist.h    2005-12-17 15:34:02.000000000 -0700
+++ insight-6.5-with-patch/gdb/solist.h    2006-12-15 16:30:06.671875000 
-0700
@@ -105,6 +105,10 @@
     int (*find_and_open_solib) (char *soname,
         unsigned o_flags, char **temp_pathname);
    
+    void (*add_one_solib) (char *soname, CORE_ADDR textSeg,
+               CORE_ADDR dataSeg);
+    void (*remove_one_solib) (char *soname, CORE_ADDR textSeg,
+                  CORE_ADDR dataSeg);
   };
 
 /* Free the memory associated with a (so_list *).  */
diff -ruN insight-6.5/gdb/target.c insight-6.5-with-patch/gdb/target.c
--- insight-6.5/gdb/target.c    2006-05-05 13:08:45.000000000 -0700
+++ insight-6.5-with-patch/gdb/target.c    2006-12-16 13:13:21.765625000 
-0700
@@ -456,6 +456,7 @@
       INHERIT (to_make_corefile_notes, t);
       INHERIT (to_get_thread_local_address, t);
       INHERIT (to_magic, t);
+      /* Do not inherit to_get_shared_libraries.  */
     }
 #undef INHERIT
 
@@ -1534,6 +1535,27 @@
 }
 
 /* Look through the list of possible targets for a target that can
+   fetch shared libraries.  */
+
+void
+target_get_shared_libraries (void)
+{
+  struct target_ops *t;
+
+  for (t = current_target.beneath; t != NULL; t = t->beneath)
+    {
+      if (t->to_get_shared_libraries != NULL)
+    {
+      t->to_get_shared_libraries (t);
+      break;
+    }
+    }
+
+  if (targetdebug)
+    fprintf_unfiltered (gdb_stdlog, "target_get_shared_libraries()\n");
+}
+
+/* Look through the list of possible targets for a target that can
    execute a run or attach command without any other data.  This is
    used to locate the default process stratum.
 
diff -ruN insight-6.5/gdb/target.h insight-6.5-with-patch/gdb/target.h
--- insight-6.5/gdb/target.h    2006-05-05 13:08:45.000000000 -0700
+++ insight-6.5-with-patch/gdb/target.h    2006-12-15 16:30:06.703125000 
-0700
@@ -424,6 +424,9 @@
                 gdb_byte *readbuf, const gdb_byte *writebuf,
                 ULONGEST offset, LONGEST len);
 
+    /* Refresh the list of shared libraries.  */
+    void (*to_get_shared_libraries) (struct target_ops *ops);
+
     int to_magic;
     /* Need sub-structure for target machine related rather than comm 
related?
      */
@@ -1092,6 +1095,10 @@
 #define RESUME_EXECD_VFORKING_CHILD_TO_GET_PARENT_VFORK() (0)
 #endif
 
+/* Refresh the list of shared libraries from the target.  */
+
+extern void target_get_shared_libraries (void);
+
 /* Routines for maintenance of the target structures...
 
    add_target:   Add a target to the list of all possible targets.
============================== end of diff ==============================


Daniel Jacobowitz wrote:
> On Sat, Dec 16, 2006 at 05:42:39PM -0700, Smith, Stephen (SWCOE) wrote:
>   
>> The function solib_target_relocate_section_addresses() refers to
>> functions and datamembers that no longer exist.  I believe that
>> borrowing code from the svr4 code is the correct thing to do.
>>
>> Can anyone confirm that that is the right thing?
>>     
>
> Sorry, it's actually the other way round.  It's not that the things it
> refers to no longer exist - they've never existed in the FSF source,
> they were part of another patch from the tree I took this from.
>
> I think the equivalent would be initializing so->textsection.
>
>   

Re: GDB solib interface

[Stephen & Linda Smith]

Hmmm... I'm also finding that the 
solib_target_relocate_section_addresses() function is not being called.

Stephen & Linda Smith wrote:
> Ok, I modified my gdbserver to conform to the interface and wrote a 
> stub main executable and shared library.  The share library symbol 
> file is now loaded when gdb connects to the gdbserver, but my guess is 
> that solib_target_relocate_section_addresses() needs to do more than 
> just set so->textsection.
> The reason is that gdb thinks that the test shared library function is 
> somewhere around 0x6* hex.    Specifically,  gdb is reporting that the 
> functions reside at the offsets I get when I do an objdump -xt of the 
> library.
> Does anyone have an idea of what variables I need to change to make 
> the symbol file line up with the load on the remote target.
> I've attached the diff.  The previous one on this list does not 
> compile cleanly.  This one does, but has the above mentions problem.

RE: GDB solib interface

[Smith, Stephen (SWCOE)]

You say below that if you dll:dll is returned, GDB should query the
target for the state of the dlls.  Does this mean that GDB would
"forget" the dll information that it received previously and then
reissue the qfDllInfo/qsDllInfo queries?


The text you wrote:

Reporting DLL events
--------------------

The `T' stop packet response is extended to take three additional
N:R pairs.  When any of these are encountered, the stopped thread
is assumed to be at a DLL event.  The signal should be SIGTRAP.
There may be multiple load and unload events in the stop packet.

  `load:Name=HEXSTR,TextSeg=ADDR,DataSeg=ADDR[,nop]'

    A new DLL has been loaded; it is described in the stop packet.

    If "nop" is specified, this DLL was already loaded, e.g. by
    an earlier call to dlopen.

  `unload:IDENTIFIER=VALUE[,nop]'

    A DLL has been unloaded.  A single identifier is provided to
    uniquely identify which DLL has been unloaded.  Depending
    on the target, this may be the text or data address; the name
    can be used, but may not be unambiguous.

    If "nop" is specified, this DLL was not unloaded, e.g. because
    another thread still had a handle to it.

  `dll:dll'

    Multiple DLL events have occured.  GDB should query the target for
the
    current state of DLLs.

Re: GDB solib interface

[Daniel Jacobowitz]

On Thu, Jan 04, 2007 at 01:37:35PM -0700, Smith, Stephen (SWCOE) wrote:
> You say below that if you dll:dll is returned, GDB should query the
> target for the state of the dlls.  Does this mean that GDB would
> "forget" the dll information that it received previously and then
> reissue the qfDllInfo/qsDllInfo queries?

That was the plan, but I don't think the patch implemented it.

-- 
Daniel Jacobowitz
CodeSourcery

GDB solib interface

[Smith, Stephen (SWCOE)]

I am running GDB to debug a remote application.  If I find that a shared
library has been loaded, I would like to use the solib interface to load
the appropriate shared library symbols.  

So lets assume that I have the file name of the shared library, that was
just loaded, how do I inform the solib subsystem that to load the
symbols.

Having used the debugger to look at my cross debugger (which is the one
that is doing the remote debugging) I see that both solib-null.c and
solib-svr4.c are both being compiled and linked in.

The debugger is built with --host=i686-pc-cygwin and
--target=powerpc-motorola-elf

Re: GDB solib interface

[Daniel Jacobowitz]

On Mon, Dec 04, 2006 at 02:55:28PM -0700, Smith, Stephen (SWCOE) wrote:
> I am running GDB to debug a remote application.  If I find that a shared
> library has been loaded, I would like to use the solib interface to load
> the appropriate shared library symbols.  
> 
> So lets assume that I have the file name of the shared library, that was
> just loaded, how do I inform the solib subsystem that to load the
> symbols.

There's no command for this.  The solib system is designed for targets
where the target can report that the library was loaded, or where GDB
can otherwise detect it.  If you're willing to have a custom GDB for
your platform, you might be able to make it examine the shared library
table in your target, assuming there is one.

-- 
Daniel Jacobowitz
CodeSourcery