Instruction address catching

Instruction address catching

[Fredrik Tolf]

Is it possible, with the current version of gdb, to make it detect every
time an address is written to, and to make it stop only if the address
of the instruction that wrote to that address isn't equal to a specific
value. If it's possible, does it work in a multithreaded application?

Re: Instruction address catching

[Kevin Buettner]

On Oct 5,  1:36am, Fredrik Tolf wrote:

> Is it possible, with the current version of gdb, to make it detect every
> time an address is written to, and to make it stop only if the address
> of the instruction that wrote to that address isn't equal to a specific
> value.

Do it like this:

    (gdb) watch foo
    Hardware watchpoint 1: foo
    (gdb) condition 1 $pc != 0x80484c8

The address to compare $pc against should be the address of the
instruction *after* the one that writes to ``foo''.  It's probably
easier to figure this out by simply setting a watchpoint on foo,
running 'til it's hit, and if that's not the one you're interested
in, add a condition so that it won't stop there again.

> If it's possible, does it work in a multithreaded application?

It should.  If it doesn't, it's a bug.

Kevin

Re: Instruction address catching

[Fredrik Tolf]

On Sat, 2002-10-05 at 02:11, Kevin Buettner wrote:
> On Oct 5,  1:36am, Fredrik Tolf wrote:
> 
> > Is it possible, with the current version of gdb, to make it detect every
> > time an address is written to, and to make it stop only if the address
> > of the instruction that wrote to that address isn't equal to a specific
> > value.
> 
> Do it like this:
> 
>     (gdb) watch foo
>     Hardware watchpoint 1: foo
>     (gdb) condition 1 $pc != 0x80484c8
> 
> The address to compare $pc against should be the address of the
> instruction *after* the one that writes to ``foo''.  It's probably
> easier to figure this out by simply setting a watchpoint on foo,
> running 'til it's hit, and if that's not the one you're interested
> in, add a condition so that it won't stop there again.
> 
> > If it's possible, does it work in a multithreaded application?
> 
> It should.  If it doesn't, it's a bug.
Are you sure? Wouldn't the pc it compares to be that of the selected
thread and not that of the thread that wrote?

> 
> Kevin
> 

Re: Instruction address catching

[Kevin Buettner]

On Oct 5,  2:21am, Fredrik Tolf wrote:

> On Sat, 2002-10-05 at 02:11, Kevin Buettner wrote:
> > On Oct 5,  1:36am, Fredrik Tolf wrote:
> > 
> > > Is it possible, with the current version of gdb, to make it detect every
> > > time an address is written to, and to make it stop only if the address
> > > of the instruction that wrote to that address isn't equal to a specific
> > > value.
> > 
> > Do it like this:
> > 
> >     (gdb) watch foo
> >     Hardware watchpoint 1: foo
> >     (gdb) condition 1 $pc != 0x80484c8
> > 
> > The address to compare $pc against should be the address of the
> > instruction *after* the one that writes to ``foo''.  It's probably
> > easier to figure this out by simply setting a watchpoint on foo,
> > running 'til it's hit, and if that's not the one you're interested
> > in, add a condition so that it won't stop there again.
> > 
> > > If it's possible, does it work in a multithreaded application?
> > 
> > It should.  If it doesn't, it's a bug.
>
> Are you sure? Wouldn't the pc it compares to be that of the selected
> thread and not that of the thread that wrote?

The thread that triggers the watchpoint becomes the selected thread
for the purposes of checking the condition.  (At least, I think that's
the way it works.)  I've just tried it with a simple (threaded) example
and it seems to work as expected...

Kevin

Re: Instruction address catching

[Fredrik Tolf]

On Sat, 2002-10-05 at 02:52, Kevin Buettner wrote:
> On Oct 5,  2:21am, Fredrik Tolf wrote:
> 
> > On Sat, 2002-10-05 at 02:11, Kevin Buettner wrote:
> > > On Oct 5,  1:36am, Fredrik Tolf wrote:
> > > 
> > > > Is it possible, with the current version of gdb, to make it detect every
> > > > time an address is written to, and to make it stop only if the address
> > > > of the instruction that wrote to that address isn't equal to a specific
> > > > value.
> > > 
> > > Do it like this:
> > > 
> > >     (gdb) watch foo
> > >     Hardware watchpoint 1: foo
By the way: I don't seem to be able to set hardware watchpoints... I'm
running i 2.4.19 linux kernel on an AMD Thunderbird, could that have
something to do with it?
> > >     (gdb) condition 1 $pc != 0x80484c8
> > > 
> > > The address to compare $pc against should be the address of the
> > > instruction *after* the one that writes to ``foo''.  It's probably
> > > easier to figure this out by simply setting a watchpoint on foo,
> > > running 'til it's hit, and if that's not the one you're interested
> > > in, add a condition so that it won't stop there again.
> > > 
> > > > If it's possible, does it work in a multithreaded application?
> > > 
> > > It should.  If it doesn't, it's a bug.
> >
> > Are you sure? Wouldn't the pc it compares to be that of the selected
> > thread and not that of the thread that wrote?
> 
> The thread that triggers the watchpoint becomes the selected thread
> for the purposes of checking the condition.  (At least, I think that's
> the way it works.)  I've just tried it with a simple (threaded) example
> and it seems to work as expected...
OK... if you say so... I'll try it! Thanks a thousandfold!
> 
> Kevin
> 

Re: Instruction address catching

[Fredrik Tolf]

On Sat, 2002-10-05 at 03:09, Fredrik Tolf wrote:
> On Sat, 2002-10-05 at 02:52, Kevin Buettner wrote:
> > On Oct 5,  2:21am, Fredrik Tolf wrote:
> > 
> > > On Sat, 2002-10-05 at 02:11, Kevin Buettner wrote:
> > > > On Oct 5,  1:36am, Fredrik Tolf wrote:
> > > > 
> > > > > Is it possible, with the current version of gdb, to make it detect every
> > > > > time an address is written to, and to make it stop only if the address
> > > > > of the instruction that wrote to that address isn't equal to a specific
> > > > > value.
> > > > 
> > > > Do it like this:
> > > > 
> > > >     (gdb) watch foo
> > > >     Hardware watchpoint 1: foo
> By the way: I don't seem to be able to set hardware watchpoints... I'm
> running i 2.4.19 linux kernel on an AMD Thunderbird, could that have
> something to do with it?
Sorry... I was wrong. I made a quick test before which was ultimately
wrong. I can set hardware watchpoints.

> > > >     (gdb) condition 1 $pc != 0x80484c8
> > > > 
> > > > The address to compare $pc against should be the address of the
> > > > instruction *after* the one that writes to ``foo''.  It's probably
> > > > easier to figure this out by simply setting a watchpoint on foo,
> > > > running 'til it's hit, and if that's not the one you're interested
> > > > in, add a condition so that it won't stop there again.
> > > > 
> > > > > If it's possible, does it work in a multithreaded application?
> > > > 
> > > > It should.  If it doesn't, it's a bug.
> > >
> > > Are you sure? Wouldn't the pc it compares to be that of the selected
> > > thread and not that of the thread that wrote?
> > 
> > The thread that triggers the watchpoint becomes the selected thread
> > for the purposes of checking the condition.  (At least, I think that's
> > the way it works.)  I've just tried it with a simple (threaded) example
> > and it seems to work as expected...
> OK... if you say so... I'll try it! Thanks a thousandfold!
The strangest thing happened. I made a small test program to test this.
The program spawns two threads, each of which increments a variable of
its own once a second. Even when I set a watchpoint on any of the
variables (i.e. "watch var1"), the program doesn't halt. I checked
manually, and the vars are increased, but gdb won't stop the program.
What can I possibly have done wrong?
> > 
> > Kevin
> > 
> 
> 

Re: Instruction address catching

[Kevin Buettner]

On Oct 5,  3:37am, Fredrik Tolf wrote:

> The strangest thing happened. I made a small test program to test this.
> The program spawns two threads, each of which increments a variable of
> its own once a second. Even when I set a watchpoint on any of the
> variables (i.e. "watch var1"), the program doesn't halt. I checked
> manually, and the vars are increased, but gdb won't stop the program.
> What can I possibly have done wrong?

Are your variables local to a function or are they globals?

Kevin