valgrind-users

[Valgrind-users] Catching changes to a memory location

[Alex B. <ker...@be...>]

Hi,

I'm currently chasing down a bug in my application which looks like a
pointer being corrupted after being cleared.

The default Valgrind run dutifully shows the SEGV caused by following
the corrupted pointer but I'd
like a way of seeing all the places this pointer is touched. So far
adding fprintf's to the code hasn't helped which leads
me to believe its not any of the could that *should* be touching this
location. I've tried tracking it down with
gdb awatch statements but it's proving more than a little problamatic
(and very manual).

Is there any way I could mark an interesting location in memory with
the client mechanism and then get
Valgrind to show me a backtrace everytime the location was touched? If
not it would be a really useful tool :-)

-- 
Alex, homepage: http://www.bennee.com/~alex/
CV: http://www.bennee.com/~alex/cv.php

Re: [Valgrind-users] Catching changes to a memory location

[Dan K. <da...@ke...>]

On Thu, May 21, 2009 at 7:22 AM, Alex Bennee <ker...@be...> wrote:
> Is there any way I could mark an interesting location in memory with
> the client mechanism and then get
> Valgrind to show me a backtrace everytime the location was touched? If
> not it would be a really useful tool :-)

That would be cool.  That said, have you tried a watchpoint in gdb?
http://stackoverflow.com/questions/58851/can-i-set-a-breakpoint-on-memory-access-in-gdb
- Dan

Re: [Valgrind-users] Catching changes to a memory location

[Alex B. <ker...@be...>]

2009/5/21 Dan Kegel <da...@ke...>:
> On Thu, May 21, 2009 at 7:22 AM, Alex Bennee <ker...@be...> wrote:
>> Is there any way I could mark an interesting location in memory with
>> the client mechanism and then get
>> Valgrind to show me a backtrace everytime the location was touched? If
>> not it would be a really useful tool :-)
>
> That would be cool.  That said, have you tried a watchpoint in gdb?
> http://stackoverflow.com/questions/58851/can-i-set-a-breakpoint-on-memory-access-in-gdb

Yep I had been trying that. The main problem was not being able to do
ith progamatically:

b 869 if awatch (void *) *&n->timer

Would of been useful. But conditinal breakpoints aren't really
designed for setting newer breakpoints.

In the end I did solve it with a client request:

	  n->timer = NULL;
          fprintf(stderr,"  2:checking n=%p n->timer=%p\n", n, n->timer);
          VALGRIND_MAKE_MEM_NOACCESS(&n->timer, sizeof(n->timer));

And almost imeadiatly Valgrind showed me an invalid write:

==31858== Invalid write of size 4
==31858==    at 0x413B1D: function (snmp_interface.c:2699)

Which lead to:
    n->vc_qam[us]++;

Which funnily enough was an incorrectly sized array just before n->timer

Time spent reading Valgrind's fine manual: 10 minutes, Time spent
hacking in the macro and running Valgrind 5 minutes, Time I *would
have* saved if I had done this earlier: 4 hours ;-)

-- 
Alex, homepage: http://www.bennee.com/~alex/
CV: http://www.bennee.com/~alex/cv.php

Re: [Valgrind-users] Catching changes to a memory location

[Colin M. <col...@pi...>]

Alex Bennee wrote:
> Yep I had been trying that. The main problem was not being able to do
> ith progamatically:
>
> b 869 if awatch (void *) *&n->timer
>
> Would of been useful. But conditinal breakpoints aren't really
> designed for setting newer breakpoints.
>
>
>   

In GDB you can use the "comm(and)" command to run an GDB command when a 
break point is hit, for example,
if you want to watch the first time the function "goo" is entered after 
"foo" has been entered, do this

b foo
b goo
disable 2
comm 1
enable 2
cont
<blank line>

run


This sets breakpoint 1 on "foo", breakpoint 2 on "goo".
It then disables breakpoint 2.
breakpoint 1, when hit, runs
"enable 2" and "cont"
to re-enable breakpoint 2, and then continue the debuggee.

The should work if  breakpoint 1  is conditional.

HTH,
Colin S. Miller

Re: [Valgrind-users] Catching changes to a memory location

[Julian S. <js...@ac...>]

> Is there any way I could mark an interesting location in memory with
> the client mechanism and then get
> Valgrind to show me a backtrace everytime the location was touched? If
> not it would be a really useful tool :-)

yeh

  #include <memcheck/memcheck.h>

  VALGRIND_MAKE_MEM_UNADDRESSABLE(address, size);

See memcheck/tests/clientperm.c for an example.

J

Re: [Valgrind-users] Catching changes to a memory location

[Alex B. <ker...@be...>]

2009/5/21 Julian Seward <js...@ac...>:
>
>> Is there any way I could mark an interesting location in memory with
>> the client mechanism and then get
>> Valgrind to show me a backtrace everytime the location was touched? If
>> not it would be a really useful tool :-)
>
> yeh
>
>  #include <memcheck/memcheck.h>
>
>  VALGRIND_MAKE_MEM_UNADDRESSABLE(address, size);

Yeah I found VALGRIND_MAKE_MEM_NOACCESS in the end. I was going word
blind reading section 3.1
and wondering why I thought there were more client requests.

It might be an idea to generate an index of all Client requests (and
parameters) and the end of the manual. Is
it some sort of docbook generated thing?

-- 
Alex, homepage: http://www.bennee.com/~alex/
CV: http://www.bennee.com/~alex/cv.php