Re: [patch] (2nd try) add epoll compat code to kernel/compat.c ...

[James Bottomley <James.Bottomley@SteelEye.com>]

On Tue, 2007-02-13 at 15:35 +1100, Stephen Rothwell wrote:
> OK, I have thought about this some more and I *think* the only
> architecture that needs compat_sys_epoll_ctl or compat_sys_epoll_wait is
> ia64 where the 64 bit version of struct epoll_event is different from the
> 32 bit version.  On x86_64, the struct is explictly packed (so it is the
> same as the 32 bit version) and on all the other 64 bit architectures the
> alignment of the u64 is the same as the equivalent 32 bit version.
> 
> Since ia64 already has its own version of these two, we only have to
> worry about epoll_pwait and then the struct epoll_event is only a problem
> for ia64.
> 
> Am I right?  (I have cc'd linux-arch for guidance.)

Not for parisc at the instruction level.  In narrow mode (32 bit mode),
a u64 load has to be done by two 32 bit loads which gives it a 4 byte
alignment requirement.  In wide mode (64 bit mode) the 64 bit load
instruction explicitly requires 8 byte alignment, so our u64 alignment
requirements are different.  However, this is from the machine code
point of view.  I can't say that gcc doesn't enforce an artificial 8
byte alignment of u64 in narrow mode, so I'll defer to the gcc experts
on that one.

James


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/