[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: coturn - bus error on armhf

Hi Michael!

On 12/23/18 8:46 PM, Michael Cree wrote:
>> static inline u64bits _ioa_ntoh64(u64bits v)
>> {
>>         u08bits *src = (u08bits*) &v;
>>         u08bits* dst = src + 7;
>>         while (src < dst) {
>>                 u08bits vdst = *dst;
>>                 *(dst--) = *src;
>>                 *(src++) = vdst;
>>         }
>>         /* OK */
>> #else
>> #endif
>>         return v;
>> }
> It's casting to a (unsigned) char pointer (assuming that is what
> u08bits is typedefed to) and reading and writing through that.
> A char can have any alignment so there is no alignment issue in
> this code.

Hmm, good point.

> The code above is an arsed-about-way of doing byte reversal.
> On a register-based arch the code will force the argument passed
> in a register to be written to the stack, byte order reversed by
> reading and writing bytes, then re-reading from the stack to pass
> back the result in the return register.  That's assuming the
> compiler isn't smart enough to work out what the the hell the
> programmer intends; but if it is it might simplify it down to
> register based logical and shift operations, thereby avoiding any
> memory accesses.
> This suggestion is somewhat concerning:
>> The copying here needs to be done with memcpy().
> Which will fail to reverse the order of bytes and thus break the
> code.  The code above is *not* doing a straight memory copy.

Well, I have to admit I didn't read the code properly and using
memcpy() is my go-to solution for this kind of problem.

>> Just casting pointers
>> and using plain assignment is a guarantee for unaligned access.
> Untrue.  It is not a guarantee for unaligned access.  There are
> other conditions that must be met for the gaurantee, and I don't
> see why the code above---despite being written by a neophyte---
> would cause unaligned accessess.

Ok, "guarantee" is too much of a hard word. I just usually avoid these
types of casts. But I think you are right because the type it's being
cast to is an unsigned char which can have any alignment.

Steve's analysis is probably more accurate than mine then, see [1].


> [1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=916919#5

 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer - glaubitz@debian.org
`. `'   Freie Universitaet Berlin - glaubitz@physik.fu-berlin.de
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Reply to: