Thank you for your investigation of this. Would you like to help us make sure that GlusterFS works cleanly on 64-bit SPARC?
-JM
----- Original Message -----
> >> I looked at the source code, it seems "dict_get_str (dict,
> >> "volname",
> >> &volname)" did not succeed. Any hint about it?
>
> I found the reason:
> $ vi ./cli/src/cli-rpc-ops.c
>
> ret = dict_allocate_and_serialize (dict,
> &req.dict.dict_val,
> (size_t
> *)&req.dict.dict_len);
>
> size_t is ulong_t, while dict_len's type is uint_t, on 64-bit
> Big-Endian system like SPARC, it causes the wrong result: dict_len
> becomes 0.
> I find that same changes are needed at too many places , so I decided
> to give up building 64-bit SPARC.
>
> Then tried building 32-bit by modifying configure.ac, configure to
> remove "-m64"
> $ CFLAGS="-m32 -g" ./configure
>
> Run gluster volume create ... got bus error again (this is SPARC),
> the
> problem is in hashfn.c
>
> libglusterfs/src$ vi hashfn.c
> change
> #define get16bits(d) (*((const uint16_t *) (d)))
> To:
> static inline uint16_t get16bits(const char *data)
> {
> return ((*data & 0x00ffU)|(*(data+1) & 0xff00U));
> }
>
> Then things look fine.
>
> BTW, the previous " signal ILL" was caused by openssl checking
> processor capabilities, it's not a problem. $ echo "ignore ILL" >
> ~/.dbxrc , solve the debugging issue.
>
> Regards,
> Xinfeng Liu
>
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel@xxxxxxxxxx
> https://lists.nongnu.org/mailman/listinfo/gluster-devel
>
