Re: [Qemu-devel] [PATCH] Fix conversions from pointer to int and vice versa

[Stefan Weil <weil@mail.berlios.de>]

Am 24.02.2011 11:11, schrieb Markus Armbruster:
> Kevin Wolf <kwolf@redhat.com> writes:
>> Am 24.02.2011 08:21, schrieb Markus Armbruster:
>>> Stefan Weil <weil@mail.berlios.de> writes:
>>>> Here the int values fds[0], sigfd, s, sock and fd are converted
>>>> to void pointers which are later converted back to an int value.
>>>>
>>>> These conversions should always use intptr_t instead of unsigned long.
>>>>
>>>> They are needed for environments where sizeof(long) != sizeof(void *).
>>> To be precise: when you want to cast a pointer to a signed integer type
>>> and back without loss, intptr_t is the signed integer type to use.
>>>
>>> But here we're dealing with the opposite case: cast int to pointer and
>>> back.
>>>> Signed-off-by: Stefan Weil <weil@mail.berlios.de>
>>>> ---
>>>> cpus.c | 8 ++++----
>>>> migration-tcp.c | 4 ++--
>>>> migration-unix.c | 4 ++--
>>>> qemu-char.c | 4 ++--
>>>> 4 files changed, 10 insertions(+), 10 deletions(-)
>>>>
>>>> diff --git a/cpus.c b/cpus.c
>>>> index 0f33945..3c4e1b8 100644
>>>> --- a/cpus.c
>>>> +++ b/cpus.c
>>>> @@ -267,7 +267,7 @@ static void qemu_event_increment(void)
>>>>
>>>> static void qemu_event_read(void *opaque)
>>>> {
>>>> - int fd = (unsigned long)opaque;
>>>> + int fd = (intptr_t)opaque;
>>>> ssize_t len;
>>>> char buffer[512];
>>> Why can't you cast straight to int?
>> You would get warnings about a pointer being cast to an integer of
>> different size
> Fair enough. Stop reading here unless you like language-lawyering ;)
>> (the behaviour is undefined if the integer is too small).
>
> Correct (I looked it up). The detour via intptr_t makes it
> implementation-defined.
>> I think you might also get a warning for the opposite direction.
>
> Implementation-defined.
>
> The standard defines semantics of valid void * -> intptr_t, uintptr_t ->
> void *: you get your original pointer back ("will compare equal").
>
> The standard is silent on converting integer type to pointer type and
> back. Doesn't matter. No sane implementation screws that up.

That's correct. int or long to pointer and back normally works.

But the compiler does not know whether the two conversions are ordered
integer to pointer - pointer to integer or
pointer to integer - integer to pointer.

Here is a short example using int instead of long,
so it will show the warnings on any linux host:

int ptr2int(void *ptr)
{
return (int)ptr;
}

void *int2ptr(int i)
{
return (void *)i;
}

gcc -Wall -c intptr.c
intptr.c: In function ‘ptr2int’:
intptr.c:3: warning: cast from pointer to integer of different size
intptr.c: In function ‘int2ptr’:
intptr.c:8: warning: cast to pointer from integer of different size

The same kind of warnings occur with the current qemu code when
I cross compile using Debian's amd64-mingw32msvc-gcc.

So the patch is needed for w64. For all other currently known
host architectures, it is not needed, but nevertheless it will
make the intention of the code clearer (as was pointed out in
an earlier mail on this subject).

Please apply the patch to qemu master.
If needed, the patch's subject can be modified
(w64: Fix conversions from pointer to int and vice versa)

Thanks,
Stefan