Re: stat64 on mips

[Justin Cormack <justin@specialbusservice.com>]

On Mon, Dec 30, 2013 at 10:13 PM, Rich Felker <dalias@aerifal.cx> wrote:
> On Mon, Dec 30, 2013 at 10:10:59PM +0000, Justin Cormack wrote:
>> On Mon, Dec 30, 2013 at 10:03 PM, Rich Felker <dalias@aerifal.cx> wrote:
>> > On Mon, Dec 30, 2013 at 10:02:19PM +0000, Justin Cormack wrote:
>> >> On Mon, Dec 30, 2013 at 9:29 PM, Rich Felker <dalias@aerifal.cx> wrote:
>> >> > On Mon, Dec 30, 2013 at 06:20:46PM +0000, Justin Cormack wrote:
>> >> >> MIPS is the most confusing architecture, but as far as I can make out,
>> >> >> the definition of struct stat64 that Musl has is probably the right
>> >> >> one for mips n32 but wrong for mips o32 which should be
>> >> >>
>> >> >> struct stat {
>> >> >>   unsigned long   st_dev;
>> >> >>   unsigned long   __st_pad0[3];
>> >> >>   unsigned long long      st_ino;
>> >> >>   mode_t          st_mode;
>> >> >>   nlink_t         st_nlink;
>> >> >>   uid_t           st_uid;
>> >> >>   gid_t           st_gid;
>> >> >>   unsigned long   st_rdev;
>> >> >>   unsigned long   __st_pad1[3];
>> >> >>   long long       st_size;
>> >> >>   time_t          st_atime;
>> >> >>   unsigned long   st_atime_nsec;
>> >> >>   time_t          st_mtime;
>> >> >>   unsigned long   st_mtime_nsec;
>> >> >>   time_t          st_ctime;
>> >> >>   unsigned long   st_ctime_nsec;
>> >> >>   unsigned long   st_blksize;
>> >> >>   unsigned long   __st_pad2;
>> >> >>   long long       st_blocks;
>> >> >> };
>> >> >>
>> >> >> It does appear that the syscalls for the two ABIs differ in this...
>> >> >
>> >> > This structure is identical to the one in musl except that it has
>> >> > 32-bit dev_t plus padding in place of 64-bit dev_t, and the musl
>> >> > version has reserved space at the end. Can you check whether the dev_t
>> >> > issue is actually a problem (it might be, based on endianness, and if
>> >> > so I think it would require ugly fixups in userspace)?
>> >>
>> >> Ah no, my mistake, you are right, the padding seems correct and I was
>> >> getting confused as usual by dev_t. However the 64 bit dev_t is a
>> >> problem on bigendian mips.
>> >>
>> >> (Whats the reason for Musl using 64 bit dev_t? glibc compatibility?)
>> >
>> > And room for expansion, and consistency of the type between archs.
>> > There's no justification for dev_t or similar types to be
>> > arch-specific.
>>
>> But isnt the kernel dev_t 32 bit for all archs?
>
> Yes and no. They have adjacent padding reserved to make it up to
> 128-bit, despite the fact that intmax_t is 64-bit everywhere and thus
> 128-bit types can't really be used. I suspect on big-endian the
> padding is at the other side to allow for this already, but it might
> be misaligned with respect to the 64/128 bit size in musl at present.

I am having toolchain trouble with my machine. Just ignore until I can
replicate after reinstalling with a more recent Linux than the one it
shipped with.

Justin