RE: async operation

RE: async operation

[Newman, Mark (N-Superior Technical Resource Inc)]

No response on this - so I am going to change GDB so that it will correctly handle an "interrupt" command to stop the inferior when in async mode.

I will put the final changes in as patches and as a bug report unless someone objects.

                                        Mark Newman

> -----Original Message-----
> From: gdb-owner@sources.redhat.com
> [mailto:gdb-owner@sources.redhat.com]On Behalf Of Newman, Mark
> (N-Superior Technical Resource Inc)
> Sent: Friday, November 14, 2003 2:30 PM
> To: gdb@sources.redhat.com
> Subject: async operation
> 
> 
> 
> Is there anyway to cleanly stop a remote target manually when 
> in async mode.
> 
> I tried using the "stop" command which makes it through the 
> async filtering in top.c - however stop simply says it is not 
> a valid command.  There is logic in here for cleanly stopping 
> when a break occurs (?) but I can't find any to allow the 
> operator to stop the target?
> 
> I add'ed "interrupt" to the filtering but it does not clear 
> the target_executing flag.
> 
>                                                               
>      Mark Newman
> 

Re: async operation

[Andrew Cagney]

> No response on this - so I am going to change GDB so that it will correctly handle an "interrupt" command to stop the inferior when in async mode.

Well you did post on a Saturday ;-)

> I tried using the "stop" command which makes it through the async filtering in top.c - however stop simply says it is not a valid command.  There is logic in here for cleanly stopping when a break occurs (?) but I can't find any to allow the operator to stop the target?

I think the description for "stop" explains why that does nothing:

   if (!dbx_commands)
     stop_command =
       add_cmd ("stop", class_obscure, not_just_help_class_command, 
"There is no
`stop' command, but you can set a hook on `stop'.\n\
This allows you to set a list of commands to be run each time execution\n\
of the program stops.", &cmdlist);

> I add'ed "interrupt" to the filtering but it does not clear the target_executing flag.

You should see "fetch_inferior_event" clear that flag after it's 
received notification that the target really has stopped.

Andrew

Re: async operation

[Mark Newman]

--- Andrew Cagney <cagney@gnu.org> wrote:
> > No response on this - so I am going to change GDB
> so that it will correctly handle an "interrupt"
> command to stop the inferior when in async mode.
> 
> Well you did post on a Saturday ;-)

Sorry - contract is coming to an end - again. 

> 
> > I tried using the "stop" command which makes it
> through the async filtering in top.c - however stop
> simply says it is not a valid command.  There is
> logic in here for cleanly stopping when a break
> occurs (?) but I can't find any to allow the
> operator to stop the target?
> 
> I think the description for "stop" explains why that
> does nothing:
> 
>    if (!dbx_commands)
>      stop_command =
>        add_cmd ("stop", class_obscure,
> not_just_help_class_command, 
> "There is no
> `stop' command, but you can set a hook on `stop'.\n\
> This allows you to set a list of commands to be run
> each time execution\n\
> of the program stops.", &cmdlist);
> 

I don't mean to sound dumb but could you clarify what
"you can set a hook on `stop'" means?

I want to allow an engineer to (for example) be able
to stop a running async process and set additional
tracepoints.  The engineer would then be able to "cont
&" to continue the process.

> > I add'ed "interrupt" to the filtering but it does
> not clear the target_executing flag.
> 
> You should see "fetch_inferior_event" clear that
> flag after it's 
> received notification that the target really has
> stopped.
> 

I see that and am working it now in conjunction with
the interrupt command - unless the stop command(?)
will do what I am looking for.  GDBserver does not
properly respond when in async and an "interrupt" is
issued.


> Andrew
> 
> 

Re: async operation

[Daniel Jacobowitz]

On Mon, Nov 17, 2003 at 09:33:05PM -0800, Mark Newman wrote:
> I don't mean to sound dumb but could you clarify what
> "you can set a hook on `stop'" means?

Search for hooks in the manual.

> I see that and am working it now in conjunction with
> the interrupt command - unless the stop command(?)
> will do what I am looking for.  GDBserver does not
> properly respond when in async and an "interrupt" is
> issued.

What goes wrong?  It should.  Remember, async is an invention of the
client - the remote protocol is the same.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: async operation

[Andrew Cagney]

> could you clarify what
> "you can set a hook on `stop'" means?

This feature:

http://sources.redhat.com/gdb/current/onlinedocs/gdb_21.html#IDX907

provides a mechanism for attaching commands to the target-stopped event.

Andrew

async operation

[Newman, Mark (N-Superior Technical Resource Inc)]

I have made a set of changes that I think will improve the operation in remote async.  Among other things it will allow one to add commands like "interrupt" and "quit" to the repetoire of commands that are legal.  

I made some design decisions that could have gone another way - e.g. use of the flags field and not using a define for async_cmd.  If anyone has strong feelings about this please let me know.

Also I am looking for opinions in general about this.

None of this reflects what Apple has done.  I still am not certain that I can use that code.

The changes are:

For a total of less than 20 lines of change

/////////////////////////////////////////////////////////////////////
cli/cli-decode.h 
Is

#define CMD_DEPRECATED            0x1
#define DEPRECATED_WARN_USER      0x2
#define MALLOCED_REPLACEMENT      0x4

Should be:

#define CMD_DEPRECATED            0x1
#define DEPRECATED_WARN_USER      0x2
#define MALLOCED_REPLACEMENT      0x4
#define ASYNC_OK                  0x08
#define ASYNC_WAIT                0x10

///////////////////////////////////////////////////////////////////////
cli/cli-decode.c
add in the file

struct cmd_list_element *
async_cmd (struct cmd_list_element *cmd, int f)
{

    cmd->flags = f;

  return cmd;
}

///////////////////////////////////////////////////////////////////////
top.c 
Is

      if (event_loop_p && target_can_async_p () && target_executing) {
	if (!(strcmp (c->name, "help") == 0)
	    && !(strcmp (c->name, "pwd") == 0)
	    && !(strcmp (c->name, "show") == 0)
	    && !(strcmp (c->name, "stop") == 0)


Should be

      if (event_loop_p && target_can_async_p () && target_executing) {
	if ((c->flags & ASYNC_OK) != 0)

/////////////////////////////////////////////////////////////////////////////
top.c
Should be  (note add to the end of void execute_command (char *p, int from_tty))
      if (event_loop_p && target_can_async_p () && target_executing) {
	if ((c->flags & ASYNC_WAIT) != 0) {
	   wait_for_inferior();
	  }
      }

////////////////////////////////////////////////////////////////////////////
maint.c
Is
add_prefix_cmd ("show", class_maintenance, maintenance_show_cmd, "\
Show GDB internal variables used by the GDB maintainer.\n\
Configure variables internal to GDB that aid in GDB's maintenance",
		  &maintenance_show_cmdlist, "maintenance show ",
		  0/*allow-unknown*/,
		  &maintenancelist);

Should be
async_cmd (  add_prefix_cmd ("show", class_maintenance, maintenance_show_cmd, "\
Show GDB internal variables used by the GDB maintainer.\n\
Configure variables internal to GDB that aid in GDB's maintenance",
		  &maintenance_show_cmdlist, "maintenance show ",
		  0/*allow-unknown*/,
		  &maintenancelist);
, ASYNC_OK);

////////////////////////////////////////////////////////////////////////////
infrun.c
Is
    stop_command =
      add_cmd ("stop", class_obscure, not_just_help_class_command, "There is no `stop' command, but you can set a hook on `stop'.\n\
This allows you to set a list of commands to be run each time execution\n\
of the program stops.", &cmdlist);

Should be
#include "cli/cli-decode.h"

    stop_command =
      add_cmd ("stop", class_obscure, not_just_help_class_command, "There is no `stop' command, but you can set a hook on `stop'.\n\
This allows you to set a list of commands to be run each time execution\n\
of the program stops.", &cmdlist)
    async_cmd(stop_command , ASYNC_OK);

//////////////////////////////////////////////////////////////////////////////
cli/cli-cmds.c

Is
  c = add_com ("help", class_support, help_command, "Print list of commands.");

Should be

  c = add_com ("help", class_support, help_command, "Print list of commands.");
  async_cmd(c, ASYNC_OK);

//////////////////////////////////////////////////////////////////////////////
cli/cli-cmds.c

Is
  add_com ("pwd", class_files, pwd_command,
	"Print working directory.  This is used for your program as well.");

Should be

  async_cmd(add_com ("pwd", class_files, pwd_command,
	"Print working directory.  This is used for your program as well.")
, ASYNC_OK);

RE: async operation

[Newman, Mark (N-Superior Technical Resource Inc)]

[-- Attachment #1: Type: text/plain, Size: 1962 bytes --]

IMHO async is not an invention of the client but the manner in which gdb
controls the client. ;-)

I am attaching a gdb output with remote_debug set.  In this instance the
sequence

> interrupt
> cont &

worked once but did not work the second time.

However I am going to suspend asking what happened since my version of
GDB has been modified to accept commands on a socket rather than
directly from the keyboard by substituting the FD for the socket for
stdin.  I added (among other things) the following to main:

  	 gdb_stdout = marks_stdout;
	 gdb_stderr = marks_stdout;
       dr_fd = fdopen(dr_in,"rw" );
       instream = dr_fd;
       reinitialize_event_loop ();

where dr_in is a socket.  Among other things I had to add a reinit to
the event loop to update rl_instream.  These changes will not go into
gdb before they are fully discussed on this forum.

It may be my changes that are causing the problem.

                                       Mark Newman

> -----Original Message-----
> From: gdb-owner@sources.redhat.com
> [mailto:gdb-owner@sources.redhat.com]On Behalf Of Daniel Jacobowitz
> Sent: Tuesday, November 18, 2003 2:13 AM
> To: Mark Newman
> Cc: Andrew Cagney; Newman, Mark (N-Superior Technical Resource Inc);
> gdb@sources.redhat.com
> Subject: Re: async operation
> 
> 
> On Mon, Nov 17, 2003 at 09:33:05PM -0800, Mark Newman wrote:
> > I don't mean to sound dumb but could you clarify what
> > "you can set a hook on `stop'" means?
> 
> Search for hooks in the manual.
> 
> > I see that and am working it now in conjunction with
> > the interrupt command - unless the stop command(?)
> > will do what I am looking for.  GDBserver does not
> > properly respond when in async and an "interrupt" is
> > issued.
> 
> What goes wrong?  It should.  Remember, async is an invention of the
> client - the remote protocol is the same.
> 
> -- 
> Daniel Jacobowitz
> MontaVista Software                         Debian GNU/Linux Developer
> 

[-- Attachment #2: temp.txt --]
[-- Type: text/plain, Size: 2769 bytes --]

00000000007f03000000000000ffff000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000801f0000ffffffff

Sending packet: $m804953c,4#6d...Ack
Packet received: 00200000
Sending packet: $mbffff81c,4#93...Ack
Packet received: 0e000000
Sending packet: $mbffff818,4#68...Ack
Packet received: 78563412
Sending packet: $mbffff7c0,50#c2...Ack
Packet received: 0d0000004a6f686e20536d69746800429f6d0040b8240140426f6973652c20496461686f00a000407c280140882e014001000000000000001e5801420100000074f8ffbf7cf8ffbf1db2024234120000

remote_stop called
Packet received: T0205:98f7ffbf;04:bcf5ffbf;08:61e50a42;
Sending packet: $g#67...Ack
Packet received: 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

Program received signal SIGINT, Interrupt.
Sending packet: $M4000acb0,1:55#68...Ack
Packet received: ENN
Sending packet: $M4000acb0,1:55#68...Ack
Packet received: ENN
Sending packet: $mbffff830,4#62...Ack
Packet received: 55320000
Sending packet: $mbffff834,4#66...Ack
Packet received: 55320000
0x080483f7 in main (argc=12885, argv=0x3255) at main.c:52
52      while (j < 1000000) {

Sending packet: $c#63...Ack

remote_stop called

Sending packet: $c#63...Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Sending packet: $c#63...Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Sending packet: $c#63...Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Sending packet: $c#63...Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it
Packet instead of Ack, ignoring it

Re: async operation

[Daniel Jacobowitz]

On Tue, Nov 18, 2003 at 10:49:29AM -0500, Newman, Mark (N-Superior Technical Resource Inc) wrote:
> IMHO async is not an invention of the client but the manner in which gdb
> controls the client. ;-)

GDB is the client.  gdbserver, intuitively, is the server.

> I am attaching a gdb output with remote_debug set.  In this instance the
> sequence

You want to look at "set debug serial 1" for this sort of problem.

> 0x080483f7 in main (argc=12885, argv=0x3255) at main.c:52
> 52      while (j < 1000000) {
> 
> Sending packet: $c#63...Ack
> 
> remote_stop called
> 
> Sending packet: $c#63...Packet instead of Ack, ignoring it

Someone's lost a packet.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

RE: async operation

[Elena Zannoni]

Newman, Mark (N-Superior Technical Resource Inc) writes:
 > IMHO async is not an invention of the client but the manner in which gdb
 > controls the client. ;-)
 > 
 > I am attaching a gdb output with remote_debug set.  In this instance the
 > sequence
 > 
 > > interrupt
 > > cont &
 > 
 > worked once but did not work the second time.

[...]
 > 
 > It may be my changes that are causing the problem.
 
Can you try on a gdb w/o your modifications?


 > 
 > Program received signal SIGINT, Interrupt.
 > Sending packet: $M4000acb0,1:55#68...Ack
 > Packet received: ENN
 > Sending packet: $M4000acb0,1:55#68...Ack
 > Packet received: ENN
 > Sending packet: $mbffff830,4#62...Ack
 > Packet received: 55320000
 > Sending packet: $mbffff834,4#66...Ack
 > Packet received: 55320000
 > 0x080483f7 in main (argc=12885, argv=0x3255) at main.c:52
 > 52      while (j < 1000000) {

I assume you said continue here?

 > 
 > Sending packet: $c#63...Ack
 > 
 > remote_stop called

Hmm do you see any output that says that remote_interrupt has been
called as well? I wonder if the signal handlers are screwed up.

 > 
 > Sending packet: $c#63...Packet instead of Ack, ignoring it

gdb keeps issuing the continue command for some reason. maybe it
hasn't realized that the target is actually running. 

elena

RE: async operation

[Mark Newman]

I did fix this.  If you look at the patch I put in you
will see a reference to ASYNC_WAIT.  When this bit is
set (as it will be for "interrupt" and "quit"} gdb
waits for a response from gdbserver when in async.

This manner of "fixing" the problem should go away
when the select in event_top, etc is combined with the
select fr the remote target.

This set of changes was designed to be under 20 lines.

                           Mark

--- Elena Zannoni <ezannoni@redhat.com> wrote:
> Newman, Mark (N-Superior Technical Resource Inc)
> writes:
>  > IMHO async is not an invention of the client but
> the manner in which gdb
>  > controls the client. ;-)
>  > 
>  > I am attaching a gdb output with remote_debug
> set.  In this instance the
>  > sequence
>  > 
>  > > interrupt
>  > > cont &
>  > 
>  > worked once but did not work the second time.
> 
> [...]
>  > 
>  > It may be my changes that are causing the
> problem.
>  
> Can you try on a gdb w/o your modifications?
> 
> 
>  > 
>  > Program received signal SIGINT, Interrupt.
>  > Sending packet: $M4000acb0,1:55#68...Ack
>  > Packet received: ENN
>  > Sending packet: $M4000acb0,1:55#68...Ack
>  > Packet received: ENN
>  > Sending packet: $mbffff830,4#62...Ack
>  > Packet received: 55320000
>  > Sending packet: $mbffff834,4#66...Ack
>  > Packet received: 55320000
>  > 0x080483f7 in main (argc=12885, argv=0x3255) at
> main.c:52
>  > 52      while (j < 1000000) {
> 
> I assume you said continue here?
> 
>  > 
>  > Sending packet: $c#63...Ack
>  > 
>  > remote_stop called
> 
> Hmm do you see any output that says that
> remote_interrupt has been
> called as well? I wonder if the signal handlers are
> screwed up.
> 
>  > 
>  > Sending packet: $c#63...Packet instead of Ack,
> ignoring it
> 
> gdb keeps issuing the continue command for some
> reason. maybe it
> hasn't realized that the target is actually running.
> 
> 
> elena
> 

async operation

[Newman, Mark (N-Superior Technical Resource Inc)]

Is there anyway to cleanly stop a remote target manually when in async mode.

I tried using the "stop" command which makes it through the async filtering in top.c - however stop simply says it is not a valid command.  There is logic in here for cleanly stopping when a break occurs (?) but I can't find any to allow the operator to stop the target?

I add'ed "interrupt" to the filtering but it does not clear the target_executing flag.

                                                                   Mark Newman