gdb internal functionality

gdb internal functionality

[raja.saleru]

Hi,

I would like to know more of GDB internals especially the control flow and
functionality for the following gdb commands. For ex. the development
environment include gdb on cygwin environment, which is connected to
OpenOCD, which inturn connects to ARM target. Any information on this
would be very helpful.

1. target remote host:port
2. stepi
3. contrinue
4. break linenum
5. delete bpnum
6. watch expression
7. set/read register value
8. read/write memory
9. quit
10. monitor reset

Thanks in Advance
Raja Saleru

Re: gdb internal functionality

[EBo]

I do not know if the following is helpful, but on of my colleagues just helped
set me up with debugging a custom avr board with JTAG and gdb.

Here are some of the relevant commands and info:

to program and set up the connection:
  avarice --erase --program --file irrigation.elf --mkI \
     --jtag /dev/ttyUSB0 --jtag-bitrate 250KHz :4242

to start gdb:
  avr-gdb --batch --quiet -x test.2.exec test_prog.elf

The test.2.exec file looks like:

  target remote localhost:4242

  break 17
  comm 1
  set adc_data=850
  #printf "adc_data=%d,", adc_data
  continue
  end

  break 28
  comm 2
  printf "adc_data=%d\n", adc_data
  printf "    PASSED!\n"
  quit
  end

  break 85
  comm 3
  printf "adc_data=%d\n", adc_data
  printf "    FAILED!\n"
  quit
  end

  continue

This sets a temp break point to set one of the variables, and the other two
break points are set up to trap the correct and failed tests in a simple
regression test.

Hope that helps, but your millage may vary...

  EBo --


raja.saleru@iap-online.com said:

> Hi,
> 
> I would like to know more of GDB internals especially the control flow and
> functionality for the following gdb commands. For ex. the development
> environment include gdb on cygwin environment, which is connected to
> OpenOCD, which inturn connects to ARM target. Any information on this
> would be very helpful.
> 
> 1. target remote host:port
> 2. stepi
> 3. contrinue
> 4. break linenum
> 5. delete bpnum
> 6. watch expression
> 7. set/read register value
> 8. read/write memory
> 9. quit
> 10. monitor reset
> 
> Thanks in Advance
> Raja Saleru
> 



-- 

Re: gdb internal functionality

[Michael Snyder]

raja.saleru@iap-online.com wrote:
> Hi,
> 
> I would like to know more of GDB internals especially the control flow and
> functionality for the following gdb commands. For ex. the development
> environment include gdb on cygwin environment, which is connected to
> OpenOCD, which inturn connects to ARM target. Any information on this
> would be very helpful.
> 
> 1. target remote host:port

Set a breakpoint at remote_open_1, then look at
your call stack.

> 2. stepi

break at stepi_command.

> 3. contrinue

break at continue_command.

> 4. break linenum

This is complex.  You'll need to look around within
breakpoint.c and, if you want to get down and dirty,
in linespec.c.

> 5. delete bpnum

break at delete_command.

> 6. watch expression

watch_command

> 7. set/read register value

There is no "set register" command.  You can just
use the regular "set" command to assign a value to
a register, eg. "set $pc = 0xfeedface".

The internal code for manipulating and storing
register values is quite complex -- start by looking
at regcache.c.

> 8. read/write memory

Also quite complex.  There is a function pointer
called "to_xfer_partial" (I know, probably seems
non-intuitive).  It lives in a struct called a "target_ops",
which is defined in target.c.

Since you're using "target remote", this pointer will
point to "remote_xfer_partial".  You can start looking
there.

> 9. quit

break at quit_command.

> 10. monitor reset

"monitor" is a prefix command that means "send the
following to my target, without interpreting it".

So "reset" is a command that will be passed to your target.

You can find "monitor_command" in target.c, and for
the remote target, it will be implemented by function
"remote_rcmd" in remote.c

Re: gdb internal functionality

[Jeremy Bennett]

On Fri, 2008-11-07 at 05:21 -0700, EBo wrote:
> > Hi,
> > 
> > I would like to know more of GDB internals especially the control flow and
> > functionality for the following gdb commands. For ex. the development
> > environment include gdb on cygwin environment, which is connected to
> > OpenOCD, which inturn connects to ARM target. Any information on this
> > would be very helpful.
> > 
> > 1. target remote host:port
> > 2. stepi
> > 3. contrinue
> > 4. break linenum
> > 5. delete bpnum
> > 6. watch expression
> > 7. set/read register value
> > 8. read/write memory
> > 9. quit
> > 10. monitor reset
> > 
> > Thanks in Advance
> > Raja Saleru
> > 

Hi Raja,

I assume you've looked at the GDB Internals document (in the
distribution or on the website). You'll also find the GDB IRC (#gdb on
irc.freenode.net) very helpful for specific questions.

Michael Snyder has given you some suggestions for the commands you list.
You may also like to look at the guide to porting GDB I wrote earlier
this year.

        http://www.embecosm.com/download/ean3.html

Chapter 2 includes some example procedural flows, which include some of
the commands you list above.

Hope this helps,


Jeremy

-- 
Tel:      +44 (1202) 416955
Cell:     +44 (7970) 676050
SkypeID: jeremybennett
Email:   jeremy.bennett@embecosm.com
Web:     www.embecosm.com

gdb load command flow

[raja.saleru]

Hi

I would like to know the flow between the gdb and OpenOPCD communication
when the below command is executed.

(gdb) load

I understand till the flow

 command_line_handler -> load_command() -> target_load() -> generic_load()->

from generic load, how it will communicate with gdbserver of OpenOCD ? Is
it through socket write() or through RSP (remtoe serial protocol) ? for
sending the executable segments into the target for loading ?


At the OpenOCD, which function is called for loading ?

Thanks in Advance
Raja Saleru

Re: gdb load command flow

[Ramana Radhakrishnan]

On Fri, Nov 14, 2008 at 8:06 AM,  <raja.saleru@iap-online.com> wrote:
>
> Hi
>
> I would like to know the flow between the gdb and OpenOPCD communication
> when the below command is executed.
>
> (gdb) load
>
> I understand till the flow
>
>  command_line_handler -> load_command() -> target_load() -> generic_load()->
>
> from generic load, how it will communicate with gdbserver of OpenOCD ? Is
> it through socket write() or through RSP (remtoe serial protocol) ? for
> sending the executable segments into the target for loading ?

set debug remote on

help set debug

will help you.


cheers
Ramana

>
>
> At the OpenOCD, which function is called for loading ?
>
> Thanks in Advance
> Raja Saleru
>
>

Re: gdb load command flow

[Jeremy Bennett]

On Fri, 2008-11-14 at 03:06 -0500, raja.saleru@iap-online.com wrote:

> I would like to know the flow between the gdb and OpenOPCD communication
> when the below command is executed.
> 
> (gdb) load
> 
> I understand till the flow
> 
>  command_line_handler -> load_command() -> target_load() -> generic_load()->
> 
> from generic load, how it will communicate with gdbserver of OpenOCD ? Is
> it through socket write() or through RSP (remtoe serial protocol) ? for
> sending the executable segments into the target for loading ?

Hi Raja,

target_load() should eventually call target_xfer_partial(), to transfer
the bytes to the target. For diagrams of the flow for this and other GDB
commands take a look at:

        http://www.embecosm.com/download/ean3.html

The connection to gdbserver will be through RSP. Presumably on the
client GDB session you've connected using something like:

        target remote hostname:port

target_xfer_partial() will attempt to use RSP X packets to write the
data as binary if the communication channel is 8-bit clean. If not it
will use M packets. For a sequence diagram and explanation for this and
other GDB commands using RSP take a look at:

        http://www.embecosm.com/download/ean4.html

I'm not hugely familiar with OpenOCD, but I understand that they have
their own implementation of gdbserver. In which case I would expect that
to be receiving the RSP packets and mapping them into OpenOCD JTAG and
using any responses to generate RSP reply packets as required.

Any feedback on the above two documents much appreciated.

Hope this helps,


Jeremy

-- 
Tel:      +44 (1202) 416955
Cell:     +44 (7970) 676050
SkypeID: jeremybennett
Email:   jeremy.bennett@embecosm.com
Web:     www.embecosm.com

gdb remote serial packet has wrong address

[raja.saleru]

Hi,

I am trying to debug the application running on arm based target,
gdb,OpenOCD.

I am using watch points to debug the application.

In debug mode, the below RSP communication shows that the address for the
packets m, Z2 and Z0 shows wrong. Is it so ?
If yes, what can be the wrong ? Any hint to debug either at gdb side or
OpenOCD side ?

Is there any possibility of addresses can become single degit like this ?



Breakpoint 1, main () at test.c:13
13              int x = 10, y = 2;
(gdb) watch x
Sending packet: $p8#a8...Ack
Packet received: fdffffff
Sending packet: $m1,4#fe...Ack
Packet received: 00000000
Sending packet: $m560,48#a0...Ack
Packet received:
80264402603e0680185c0a00603e0200185c0200603e0400683e0400005e010
0603e0200093c0000493cdc2e185c0b00083a683e0220683e04000844881c0400001fa002001eda0
2
Sending packet: $m9,4#06...Ack
Packet received: 05000000
Sending packet: $m9,4#06...Ack
Packet received: 05000000
Hardware watchpoint 2: x
(gdb) c
Continuing.
Sending packet: $Hc0#db...Ack
Packet received:
Sending packet: $s#73...Ack
Packet received: T05
Sending packet: $p1e#06...Ack
Packet received: 6c050000
Sending packet: $Z0,568,2#b7...Ack
Packet received: OK
Sending packet: $p8#a8...Ack
Packet received: fdffffff
Sending packet: $m560,48#a0...Ack
Packet received:
80264402603e0680185c0a00603e0200185c0200603e0400683e0400005e010
0603e0200093c0000493cdc2e185c0b00083a683e0220683e04000844881c0400001fa002001eda0
2
Sending packet: $m1,4#fe...Ack
Packet received: 00000000
Sending packet: $Z2,1,4#49...Ack
Packet received: E10
Packet Z2 (write-watchpoint) is supported
Sending packet: $Z2,0,4#48...Ack
Packet received: E10
Sending packet: $z2,1,4#69...Ack
Packet received: OK
Sending packet: $z2,0,4#68...Ack
Packet received: OK
Sending packet: $Z0,5,2#49...Ack
Packet received: OK
Warning:
Could not insert hardware watchpoint 2.
Could not insert hardware breakpoints:
You may have requested too many hardware breakpoints/watchpoints.

(gdb)

Thanks in Advance
Raja Saleru

Re: gdb remote serial packet has wrong address

[Michael Snyder]

raja.saleru@iap-online.com wrote:
> Hi,
> 
> I am trying to debug the application running on arm based target,
> gdb,OpenOCD.
> 
> I am using watch points to debug the application.
> 
> In debug mode, the below RSP communication shows that the address for the
> packets m, Z2 and Z0 shows wrong. Is it so ?
> If yes, what can be the wrong ? Any hint to debug either at gdb side or
> OpenOCD side ?
> 
> Is there any possibility of addresses can become single degit like this ?


Could be truncation?  From 64 bit to 32 bit, eg.?

What target architecture does gdb report?


> Breakpoint 1, main () at test.c:13
> 13              int x = 10, y = 2;
> (gdb) watch x
> Sending packet: $p8#a8...Ack
> Packet received: fdffffff
> Sending packet: $m1,4#fe...Ack
> Packet received: 00000000
> Sending packet: $m560,48#a0...Ack
> Packet received:
> 80264402603e0680185c0a00603e0200185c0200603e0400683e0400005e010
> 0603e0200093c0000493cdc2e185c0b00083a683e0220683e04000844881c0400001fa002001eda0
> 2
> Sending packet: $m9,4#06...Ack
> Packet received: 05000000
> Sending packet: $m9,4#06...Ack
> Packet received: 05000000
> Hardware watchpoint 2: x
> (gdb) c
> Continuing.
> Sending packet: $Hc0#db...Ack
> Packet received:
> Sending packet: $s#73...Ack
> Packet received: T05
> Sending packet: $p1e#06...Ack
> Packet received: 6c050000
> Sending packet: $Z0,568,2#b7...Ack
> Packet received: OK
> Sending packet: $p8#a8...Ack
> Packet received: fdffffff
> Sending packet: $m560,48#a0...Ack
> Packet received:
> 80264402603e0680185c0a00603e0200185c0200603e0400683e0400005e010
> 0603e0200093c0000493cdc2e185c0b00083a683e0220683e04000844881c0400001fa002001eda0
> 2
> Sending packet: $m1,4#fe...Ack
> Packet received: 00000000
> Sending packet: $Z2,1,4#49...Ack
> Packet received: E10
> Packet Z2 (write-watchpoint) is supported
> Sending packet: $Z2,0,4#48...Ack
> Packet received: E10
> Sending packet: $z2,1,4#69...Ack
> Packet received: OK
> Sending packet: $z2,0,4#68...Ack
> Packet received: OK
> Sending packet: $Z0,5,2#49...Ack
> Packet received: OK
> Warning:
> Could not insert hardware watchpoint 2.
> Could not insert hardware breakpoints:
> You may have requested too many hardware breakpoints/watchpoints.
> 
> (gdb)
> 
> Thanks in Advance
> Raja Saleru
> 

adding new commandline interface to gdb ?

[raja.saleru]

Hi,

We are connecting gdb running on cygwin host into gdbserver running on
target using the command "target remote host:port". The gdbserver contains
the name of the target, for example

TargetName = "xxx".

We would like to read the name of the target from gdb, using command-line
interface.
Lets say we would like to add the new command

(gdb) target info target-name

This should display the target name "xxx".

Initially before thinking on RSP protocol, we need to register this command.

Please give us some guidelines on how to register or how to proceed to add
new commandline interface to get the target name.

Thanks in advance
Raja Saleru

Re: adding new commandline interface to gdb ?

[teawater]

Read source in
src/gdb/remote.c
void
_initialize_remote (void)

On Tue, Jan 6, 2009 at 18:35, <raja.saleru@iap-online.com> wrote:
>
> Hi,
>
> We are connecting gdb running on cygwin host into gdbserver running on
> target using the command "target remote host:port". The gdbserver contains
> the name of the target, for example
>
> TargetName = "xxx".
>
> We would like to read the name of the target from gdb, using command-line
> interface.
> Lets say we would like to add the new command
>
> (gdb) target info target-name
>
> This should display the target name "xxx".
>
> Initially before thinking on RSP protocol, we need to register this command.
>
> Please give us some guidelines on how to register or how to proceed to add
> new commandline interface to get the target name.
>
> Thanks in advance
> Raja Saleru
>

info watch command

[raja.saleru]

Hi,

I am having query on "info watch command". consider the below sample

(gdb) watch x
Hardware watchpoint 2: x
(gdb) rwatch y
Hardware read watchpoint 3: y
(gdb) awatch c
Hardware access (read/write) watchpoint 4: c

(gdb) info watch
Num Type           Disp Enb Address    What
1   breakpoint     keep y   0x00000542 in main at test.c:10
        breakpoint already hit 1 time
2   hw watchpoint  keep y              x
3   read watchpoint keep y              y
4   acc watchpoint keep y              c

by default, it will be hardware watchpoint. for the "info watch" command,
only for watch it is showing as "hw watchpoint". however why it is not
differentiated for rwatch as "read watchpoint" "hw read watchpoint" ?

similarly for awatch.

For write watchpoint, there is differentiation.

Thanks in advance
Raja Saleru

Re: info watch command

[Daniel Jacobowitz]

On Tue, Jan 20, 2009 at 03:36:22AM -0500, raja.saleru@iap-online.com wrote:
> by default, it will be hardware watchpoint. for the "info watch" command,
> only for watch it is showing as "hw watchpoint". however why it is not
> differentiated for rwatch as "read watchpoint" "hw read watchpoint" ?

Because all read and access watchpoints are hardware; it is not
practical to do either in software.

-- 
Daniel Jacobowitz
CodeSourcery