Re: 64bit pointer

Re: 64bit pointer

[freindlyuser]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Thu, 03 Sep 2009 12:33:25 +0000 Jan Kratochvil
<jan.kratochvil@redhat.com> wrote:
>On Thu, 03 Sep 2009 14:21:41 +0200, freindlyuser@hushmail.com
>wrote:
>> (gdb) x/s *($rdi+0x8)
>> 0x4210a9b7:      <Address 0x4210a9b7 out of bounds>
>
>This syntax is not much recommended, it means the same as:
>
>(gdb) x/s *(int *) ($rdi+0x8)
>0x4210a9b7:      <Address 0x4210a9b7 out of bounds>
>
>On 64bit arch sizeof (int) == 4 but sizeof (void *) == 8 so you
>will not fetch
>the whole address.
>
>
>> (gdb) x/x $rdi+0x8
>> 0x3a9b4210b7a4: 0x00003a9b4210a9b7
>> (gdb) x/s 0x00003a9b4210a9b7
>> 0x3a9b4210a9b7:  "The string it points to"
>
>Therefore you want one indirection there.
>
>> Should this be happening?
>
>Yes.  Until GDB forbids dereferencing numeric arguments as `int *'
>which IMHO
>is more confusing than convenient.  [Would a patch be approved?]
>
>
>> Is there a work around where I can read the data in
>> 0x00003a9b4210a9b7 without having to manually copy and paste
>(ie:
>> in the commands that are executed on a breakpoint).
>
>This way it should work:
>
>(gdb) x/s *(void **) ($rdi+0x8)
>OR
>(gdb) p *(char **) ($rdi+0x8)
>
>
>Regards,
>Jan

The work around works perfectly.

Thank you.
-----BEGIN PGP SIGNATURE-----
Charset: UTF8
Version: Hush 3.0
Note: This signature can be verified at https://www.hushtools.com/verify

wpwEAQMCAAYFAkqfxDYACgkQWX1/rrecYT2g+gP+J6na4+MXovA7SJRD1oYPKqbZ9Pu1
Ik4vBR5BFt6RLV0zrYXsAnEq1xOXnac1q3r9XNUA0R1jmKtEFYvJjH44KpdzwIwBBRSS
7M7qeaHZZhDEgfJOFUgWS5KUJgjwRo4Rm2lGLNgmaftQVBrgE5HUQxVicwtdwIO9qyhU
i/n0Oas=
=+lL9
-----END PGP SIGNATURE-----

64bit pointer

[freindlyuser]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I didn't know if this was a bug or if I was just doing something
wrong. Basically when I try to use * to get the value pointed to by
something it gives me 32 bits of the pointer which is actually a
64bit address.


(gdb) x/s *($rdi+0x8)
0x4210a9b7:      <Address 0x4210a9b7 out of bounds>
(gdb) x/x $rdi+0x8
0x3a9b4210b7a4: 0x00003a9b4210a9b7
(gdb) x/s 0x00003a9b4210a9b7
0x3a9b4210a9b7:  "The string it points to"

In the above I wanted to read the string pointed to by the pointer
at $rdi+0x8

Should this be happening?
Should I be using a different command or something different to the
asterisk?
Is there a work around where I can read the data in
0x00003a9b4210a9b7 without having to manually copy and paste (ie:
in the commands that are executed on a breakpoint).

Thank you.
-----BEGIN PGP SIGNATURE-----
Charset: UTF8
Version: Hush 3.0
Note: This signature can be verified at https://www.hushtools.com/verify

wpwEAQMCAAYFAkqftNQACgkQWX1/rrecYT2pBgP/RVbdltSzvOihdHAKfpweEjXQ7U1T
f2xyz8uRhyPZQ2n1w2nA/IyU79LyB3sIUcoZ+CkxX0VPh34W8Kw00XfWYq85C3dSW9vm
ibLw6AIE5CVcvz39wX/68s4qEYS/BdpetltS3sEeKswV7ylCjNajtFwu2h5WIxsVhdop
kkXuBfA=
=l5hi
-----END PGP SIGNATURE-----

Re: 64bit pointer

[Jan Kratochvil]

On Thu, 03 Sep 2009 14:21:41 +0200, freindlyuser@hushmail.com wrote:
> (gdb) x/s *($rdi+0x8)
> 0x4210a9b7:      <Address 0x4210a9b7 out of bounds>

This syntax is not much recommended, it means the same as:

(gdb) x/s *(int *) ($rdi+0x8)
0x4210a9b7:      <Address 0x4210a9b7 out of bounds>

On 64bit arch sizeof (int) == 4 but sizeof (void *) == 8 so you will not fetch
the whole address.


> (gdb) x/x $rdi+0x8
> 0x3a9b4210b7a4: 0x00003a9b4210a9b7
> (gdb) x/s 0x00003a9b4210a9b7
> 0x3a9b4210a9b7:  "The string it points to"

Therefore you want one indirection there.

> Should this be happening?

Yes.  Until GDB forbids dereferencing numeric arguments as `int *' which IMHO
is more confusing than convenient.  [Would a patch be approved?]


> Is there a work around where I can read the data in
> 0x00003a9b4210a9b7 without having to manually copy and paste (ie:
> in the commands that are executed on a breakpoint).

This way it should work:

(gdb) x/s *(void **) ($rdi+0x8)
OR
(gdb) p *(char **) ($rdi+0x8)


Regards,
Jan

Re: 64bit pointer

[Tom Tromey]

>>>>> "Jan" == Jan Kratochvil <jan.kratochvil@redhat.com> writes:

Jan> Yes.  Until GDB forbids dereferencing numeric arguments as `int *'
Jan> which IMHO is more confusing than convenient.  [Would a patch be
Jan> approved?]

This behavior does seem strange to me.

GDB as a project seems to be somewhat conservative about changing things
like this, though.  I don't think I could predict whether anybody would
object, you pretty much have to submit a patch and see who speaks up.

Tom

Re: 64bit pointer

[Andreas Schwab]

freindlyuser@hushmail.com writes:

> I didn't know if this was a bug or if I was just doing something
> wrong. Basically when I try to use * to get the value pointed to by
> something it gives me 32 bits of the pointer which is actually a
> 64bit address.
>
>
> (gdb) x/s *($rdi+0x8)
> 0x4210a9b7:      <Address 0x4210a9b7 out of bounds>
> (gdb) x/x $rdi+0x8
> 0x3a9b4210b7a4: 0x00003a9b4210a9b7
> (gdb) x/s 0x00003a9b4210a9b7
> 0x3a9b4210a9b7:  "The string it points to"
>
> In the above I wanted to read the string pointed to by the pointer
> at $rdi+0x8
>
> Should this be happening?

Since the type of $rdi is not a pointer, gdb is just being helpful and
implicitly converts the value to (int *) before applying the indirection
operator.

> Should I be using a different command or something different to the
> asterisk?
> Is there a work around where I can read the data in
> 0x00003a9b4210a9b7 without having to manually copy and paste (ie:
> in the commands that are executed on a breakpoint).

There are several options:

- Add a cast, {type} is short for *(type *):

(gdb) x/s {char *}($rdi+8)

- Use the $__ convenience variable after examining the address:

(gdb) x/x $rdi+8
(gdb) x/s $__

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."