17. Debugging remote programs

17.1 Connecting to a remote target
17.2 Using the gdbserver program		Using the gdbserver program
17.3 Using the gdbserve.nlm program		Using the gdbserve.nlm program
17.4 Remote configuration
17.5 Implementing a remote stub

17.1 Connecting to a remote target

On the GDB host machine, you will need an unstripped copy of your program, since GDB needs symobl and debugging information. Start up GDB as usual, using the name of the local copy of your program as the first argument.

If you're using a serial line, you may want to give GDB the `--baud' option, or use the set remotebaud command before the target command.

After that, use target remote to establish communications with the target machine. Its argument specifies how to communicate--either via a devicename attached to a direct serial line, or a TCP or UDP port (possibly to a terminal server which in turn has a serial line to the target). For example, to use a serial line connected to the device named `/dev/ttyb':

target remote /dev/ttyb

To use a TCP connection, use an argument of the form host:port or tcp:host:port. For example, to connect to port 2828 on a terminal server named manyfarms:

target remote manyfarms:2828

If your remote target is actually running on the same machine as your debugger session (e.g. a simulator of your target running on the same host), you can omit the hostname. For example, to connect to port 1234 on your local machine:

target remote :1234

Note that the colon is still required here.

To use a UDP connection, use an argument of the form udp:host:port. For example, to connect to UDP port 2828 on a terminal server named manyfarms:

target remote udp:manyfarms:2828

When using a UDP connection for remote debugging, you should keep in mind that the `U' stands for "Unreliable". UDP can silently drop packets on busy or unreliable networks, which will cause havoc with your debugging session.

Now you can use all the usual commands to examine and change data and to step and continue the remote program.

Whenever GDB is waiting for the remote program, if you type the interrupt character (often C-C), GDB attempts to stop the program. This may or may not succeed, depending in part on the hardware and the serial drivers the remote system uses. If you type the interrupt character once again, GDB displays this prompt:

Interrupted while waiting for the program. Give up (and stop debugging it)? (y or n)

If you type y, GDB abandons the remote debugging session. (If you decide you want to try again later, you can use `target remote' again to connect once more.) If you type n, GDB goes back to waiting.

detach

When you have finished debugging the remote program, you can use the detach command to release it from GDB control. Detaching from the target normally resumes its execution, but the results will depend on your particular remote stub. After the detach command, GDB is free to connect to another target.

disconnect

The disconnect command behaves like detach, except that the target is generally not resumed. It will wait for GDB (this instance or another one) to connect and continue debugging. After the disconnect command, GDB is again free to connect to another target.

17.2 Using the gdbserver program

gdbserver is a control program for Unix-like systems, which allows you to connect your program with a remote GDB via target remote---but without linking in the usual debugging stub.

gdbserver is not a complete replacement for the debugging stubs, because it requires essentially the same operating-system facilities that GDB itself does. In fact, a system that can run gdbserver to connect to a remote GDB could also run GDB locally! gdbserver is sometimes useful nevertheless, because it is a much smaller program than GDB itself. It is also easier to port than all of GDB, so you may be able to get started more quickly on a new system by using gdbserver. Finally, if you develop code for real-time systems, you may find that the tradeoffs involved in real-time operation make it more convenient to do as much development work as possible on another system, for example by cross-compiling. You can use gdbserver to make a similar choice for debugging.

GDB and gdbserver communicate via either a serial line or a TCP connection, using the standard GDB remote serial protocol.

On the target machine,

you need to have a copy of the program you want to debug. gdbserver does not need your program's symbol table, so you can strip the program if necessary to save space. GDB on the host system does all the symbol handling.

To use the server, you must tell it how to communicate with GDB; the name of your program; and the arguments for your program. The usual syntax is:

target> gdbserver comm program [ args ... ]

comm is either a device name (to use a serial line) or a TCP hostname and portnumber. For example, to debug Emacs with the argument `foo.txt' and communicate with GDB over the serial port `/dev/com1':

target> gdbserver /dev/com1 emacs foo.txt

gdbserver waits passively for the host GDB to communicate with it.

To use a TCP connection instead of a serial line:

target> gdbserver host:2345 emacs foo.txt

The only difference from the previous example is the first argument, specifying that you are communicating with the host GDB via TCP. The `host:2345' argument means that gdbserver is to expect a TCP connection from machine `host' to local TCP port 2345. (Currently, the `host' part is ignored.) You can choose any number you want for the port number as long as it does not conflict with any TCP ports already in use on the target system (for example, 23 is reserved for telnet).(4) You must use the same port number with the host GDB target remote command.

On some targets, gdbserver can also attach to running programs. This is accomplished via the --attach argument. The syntax is:

target> gdbserver comm --attach pid

pid is the process ID of a currently running process. It isn't necessary to point gdbserver at a binary for the running process.

You can debug processes by name instead of process ID if your target has the pidof utility:

target> gdbserver comm --attach `pidof PROGRAM`

In case more than one copy of PROGRAM is running, or PROGRAM has multiple threads, most versions of pidof support the -s option to only return the first process ID.

On the host machine,

connect to your target (see section Connecting to a remote target). For TCP connections, you must start up gdbserver prior to using the target remote command. Otherwise you may get an error whose text depends on the host system, but which usually looks something like `Connection refused'. You don't need to use the load command in GDB when using gdbserver, since the program is already on the target.

17.3 Using the gdbserve.nlm program

gdbserve.nlm is a control program for NetWare systems, which allows you to connect your program with a remote GDB via target remote.

GDB and gdbserve.nlm communicate via a serial line, using the standard GDB remote serial protocol.

On the target machine,

you need to have a copy of the program you want to debug. gdbserve.nlm does not need your program's symbol table, so you can strip the program if necessary to save space. GDB on the host system does all the symbol handling.

To use the server, you must tell it how to communicate with GDB; the name of your program; and the arguments for your program. The syntax is:

load gdbserve [ BOARD=board ] [ PORT=port ] [ BAUD=baud ] program [ args ... ]

board and port specify the serial line; baud specifies the baud rate used by the connection. port and node default to 0, baud defaults to 9600bps.

For example, to debug Emacs with the argument `foo.txt'and communicate with GDB over serial port number 2 or board 1 using a 19200bps connection:

load gdbserve BOARD=1 PORT=2 BAUD=19200 emacs foo.txt

On the GDB host machine, connect to your target (see section Connecting to a remote target).

17.4 Remote configuration

The following configuration options are available when debugging remote programs:

set remote hardware-watchpoint-limit limit

set remote hardware-breakpoint-limit limit

Restrict GDB to using limit remote hardware breakpoint or watchpoints. A limit of -1, the default, is treated as unlimited.

17.5 Implementing a remote stub

The stub files provided with GDB implement the target side of the communication protocol, and the GDB side is implemented in the GDB source file `remote.c'. Normally, you can simply allow these subroutines to communicate, and ignore the details. (If you're implementing your own stub file, you can still ignore the details: start with one of the existing stub files. `sparc-stub.c' is the best organized, and therefore the easiest to read.)

To debug a program running on another machine (the debugging target machine), you must first arrange for all the usual prerequisites for the program to run by itself. For example, for a C program, you need:

1. A startup routine to set up the C runtime environment; these usually have a name like `crt0'. The startup routine may be supplied by your hardware supplier, or you may have to write your own.

2. A C subroutine library to support your program's subroutine calls, notably managing input and output.

3. A way of getting your program to the other machine--for example, a download program. These are often supplied by the hardware manufacturer, but you may have to write your own from hardware documentation.

The next step is to arrange for your program to use a serial port to communicate with the machine where GDB is running (the host machine). In general terms, the scheme looks like this:

On the host,

GDB already understands how to use this protocol; when everything else is set up, you can simply use the `target remote' command (see section Specifying a Debugging Target).

On the target,

you must link with your program a few special-purpose subroutines that implement the GDB remote serial protocol. The file containing these subroutines is called a debugging stub.

On certain remote targets, you can use an auxiliary program gdbserver instead of linking a stub into your program. See section Using the gdbserver program, for details.

The debugging stub is specific to the architecture of the remote machine; for example, use `sparc-stub.c' to debug programs on SPARC boards.

These working remote stubs are distributed with GDB:

i386-stub.c

For Intel 386 and compatible architectures.

m68k-stub.c

For Motorola 680x0 architectures.

sh-stub.c

For Renesas SH architectures.

sparc-stub.c

For SPARC architectures.

sparcl-stub.c

For Fujitsu SPARCLITE architectures.

The `README' file in the GDB distribution may list other recently added stubs.

17.5.1 What the stub can do for you
17.5.2 What you must do for the stub
17.5.3 Putting it all together

17.5.1 What the stub can do for you

The debugging stub for your architecture supplies these three subroutines:

set_debug_traps

This routine arranges for handle_exception to run when your program stops. You must call this subroutine explicitly near the beginning of your program.

handle_exception

This is the central workhorse, but your program never calls it explicitly--the setup code arranges for handle_exception to run when a trap is triggered.

handle_exception takes control when your program stops during execution (for example, on a breakpoint), and mediates communications with GDB on the host machine. This is where the communications protocol is implemented; handle_exception acts as the GDB representative on the target machine. It begins by sending summary information on the state of your program, then continues to execute, retrieving and transmitting any information GDB needs, until you execute a GDB command that makes your program resume; at that point, handle_exception returns control to your own code on the target machine.

breakpoint

Use this auxiliary subroutine to make your program contain a breakpoint. Depending on the particular situation, this may be the only way for GDB to get control. For instance, if your target machine has some sort of interrupt button, you won't need to call this; pressing the interrupt button transfers control to handle_exception---in effect, to GDB. On some machines, simply receiving characters on the serial port may also trigger a trap; again, in that situation, you don't need to call breakpoint from your own program--simply running `target remote' from the host GDB session gets control.

Call breakpoint if none of these is true, or if you simply want to make certain your program stops at a predetermined point for the start of your debugging session.

17.5.2 What you must do for the stub

The debugging stubs that come with GDB are set up for a particular chip architecture, but they have no information about the rest of your debugging target machine.

First of all you need to tell the stub how to communicate with the serial port.

int getDebugChar()

Write this subroutine to read a single character from the serial port. It may be identical to getchar for your target system; a different name is used to allow you to distinguish the two if you wish.

void putDebugChar(int)

Write this subroutine to write a single character to the serial port. It may be identical to putchar for your target system; a different name is used to allow you to distinguish the two if you wish.

If you want GDB to be able to stop your program while it is running, you need to use an interrupt-driven serial driver, and arrange for it to stop when it receives a ^C (`\003', the control-C character). That is the character which GDB uses to tell the remote system to stop.

Getting the debugging target to return the proper status to GDB probably requires changes to the standard stub; one quick and dirty way is to just execute a breakpoint instruction (the "dirty" part is that GDB reports a SIGTRAP instead of a SIGINT).

Other routines you need to supply are:

void exceptionHandler (int exception_number, void *exception_address)

Write this function to install exception_address in the exception handling tables. You need to do this because the stub does not have any way of knowing what the exception handling tables on your target system are like (for example, the processor's table might be in ROM, containing entries which point to a table in RAM). exception_number is the exception number which should be changed; its meaning is architecture-dependent (for example, different numbers might represent divide by zero, misaligned access, etc). When this exception occurs, control should be transferred directly to exception_address, and the processor state (stack, registers, and so on) should be just as it is when a processor exception occurs. So if you want to use a jump instruction to reach exception_address, it should be a simple jump, not a jump to subroutine.

For the 386, exception_address should be installed as an interrupt gate so that interrupts are masked while the handler runs. The gate should be at privilege level 0 (the most privileged level). The SPARC and 68k stubs are able to mask interrupts themselves without help from exceptionHandler.

void flush_i_cache()

On SPARC and SPARCLITE only, write this subroutine to flush the instruction cache, if any, on your target machine. If there is no instruction cache, this subroutine may be a no-op.

On target machines that have instruction caches, GDB requires this function to make certain that the state of your program is stable.

You must also make sure this library routine is available:

void *memset(void *, int, int)

This is the standard library function memset that sets an area of memory to a known value. If you have one of the free versions of libc.a, memset can be found there; otherwise, you must either obtain it from your hardware manufacturer, or write your own.

If you do not use the GNU C compiler, you may need other standard library subroutines as well; this varies from one stub to another, but in general the stubs are likely to use any of the common library subroutines which gcc generates as inline code.

17.5.3 Putting it all together

In summary, when your program is ready to debug, you must follow these steps.

1. Make sure you have defined the supporting low-level routines (see section What you must do for the stub):

   getDebugChar, putDebugChar, flush_i_cache, memset, exceptionHandler.

2. Insert these lines near the top of your program:

   set_debug_traps(); breakpoint();

3. For the 680x0 stub only, you need to provide a variable called exceptionHook. Normally you just use:

   void (*exceptionHook)() = 0;

   but if before calling set_debug_traps, you set it to point to a function in your program, that function is called when GDB continues after stopping on a trap (for example, bus error). The function indicated by exceptionHook is called with one parameter: an int which is the exception number.

4. Compile and link together: your program, the GDB debugging stub for your target architecture, and the supporting subroutines.

5. Make sure you have a serial connection between your target machine and the GDB host, and identify the serial port on the host.

6. Download your program to your target machine (or get it there by whatever means the manufacturer provides), and start it.

7. Start GDB on the host, and connect to the target (see section Connecting to a remote target).