Re: stat64 wrong on sparc64 user

[Luca Bonissi]

On 28/03/23 13:55, Thomas Huth wrote:
> On 28/03/2023 13.48, Luca Bonissi wrote:
> > --- qemu-20230327/linux-user/syscall_defs.h    2023-03-27 
> > 15:41:42.000000000 +0200
> > +++ qemu-20230327/linux-user/syscall_defs.h.new    2023-03-27 
> > 21:43:25.615115126 +0200
> > @@ -1450,7 +1450,7 @@ struct target_stat {
> >       unsigned int    st_dev;
> >       abi_ulong    st_ino;
> >       unsigned int    st_mode;
> > -    unsigned int    st_nlink;
> > +    short int    st_nlink;
> >       unsigned int    st_uid;
>
> That looks wrong at a first glance. IIRC Sparc is a very strictly 
> aligned architecture, so if the previous field "st_mode" was aligned to 
> a 4-byte boundary, the "st_uid" field now would not be aligned 
> anymore... are you sure about this change? Maybe it needs a padding 
> field now?

The padding is automatic (either on Sparc or x86-64): short will be 
aligned to 2-byte boundary, int will be aligned to 4-byte boundary, long 
will be aligned to 8-byte boundary.

E.g.:
st_dev=0x05060708;
st_ino=0x1112131415161718;
st_mode=0x1a1b1c1d;
st_nlink=0x2728;
st_uid=0x2a2b2c2d;
st_gid=0x3a3b3c3d;
st_rdev=0x35363738;
st_size=0x4142434445464748;
st_blksize=0x5152535455565758;

will result (sparc64 - big endian):
00: 05 06 07 08 00 00 00 00
08: 11 12 13 14 15 16 17 18
10: 1A 1B 1C 1D 27 28 00 00
18: 2A 2B 2C 2D 3A 3B 3C 3D
20: 35 36 37 38 00 00 00 00
28: 41 42 43 44 45 46 47 48
30: 00 00 00 00 00 00 00 00
38: 00 00 00 00 00 00 00 00
40: 00 00 00 00 00 00 00 00
48: 51 52 53 54 55 56 57 58
50: 00 00 00 00 00 00 00 00
58: 00 00 00 00 00 00 00 00
60: 00 00 00 00 00 00 00 00

Or on x86-64 (little endian):
00: 08 07 06 05 00 00 00 00
08: 18 17 16 15 14 13 12 11
10: 1D 1C 1B 1A 28 27 00 00
18: 2D 2C 2B 2A 3D 3C 3B 3A
20: 38 37 36 35 00 00 00 00
28: 48 47 46 45 44 43 42 41
30: 00 00 00 00 00 00 00 00
38: 00 00 00 00 00 00 00 00
40: 00 00 00 00 00 00 00 00
48: 58 57 56 55 54 53 52 51
50: 00 00 00 00 00 00 00 00
58: 00 00 00 00 00 00 00 00
60: 00 00 00 00 00 00 00 00

Please note the automatic padding between "st_dev" and "st_ino" (offset 
0x04, 4 bytes), "st_nlink" and "st_uid" (offset 0x16, 2 bytes), 
"st_rdev" and "st_size" (offset 0x24, 4 bytes).

Placing st_nlink as int would result in incorrect big/little endian 
conversion, so it should be set as short. If you like clearer source 
code, you can optionally add padding, but it is not mandatory.

Thanks!
  Luca