[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <1182960660.1170.12.camel@pmac.infradead.org>
Date: Wed, 27 Jun 2007 17:11:00 +0100
From: David Woodhouse <dwmw2@...radead.org>
To: Rodolfo Giometti <giometti@...eenne.com>
Cc: linux-kernel@...r.kernel.org,
Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [PATCH] LinuxPPS (with new syscalls API)
On Wed, 2007-06-27 at 14:58 +0200, Rodolfo Giometti wrote:
> On Wed, Jun 27, 2007 at 11:18:30AM +0100, David Woodhouse wrote:
> > On Wed, 2007-06-27 at 12:14 +0200, Rodolfo Giometti wrote:
> > > On Tue, Jun 26, 2007 at 06:38:40PM +0100, David Woodhouse wrote:
> > > >
> > > > 64-bit kernels can run 32-bit userspace programs. But some structures
> > > > come out _differently_ between 32-bit and 64-bit compilation, so the
> > > > system call needs a special 'compat' handler instead of just running the
> > > > normal 64-bit system call.
> > > >
> > > > The 'struct timespec' is one structure which is sometimes different for
> > > > 32-bit vs. 64-bit, so any system call taking a 'struct timespec' must
> > > > have a separate compat_sys_xxxx() to handle that. See something like
> > > > compat_sys_clock_settime() in kernel/compat.c for an example (but don't
> > > > use set_fs() like it does; just see how it handles the compat_timespec).
> > >
> > > Did you mean something like this?
> >
> > How will 64-bit system calls work if you do it like that? You need to
> > provide _both_ sys_time_pps_fetch() and compat_sys_time_pps_fetch().
>
> Sorry, I'm new to this 32/64 bits issues...
>
> Now is it correct?
No, because you're passing a _kernel_ pointer to sys_time_pps_fetch()
where it expects a userspace pointer. Use compat_alloc_user_space() to
find somewhere to put it in user space, instead. Or change your internal
__sys_time_pps_fetch() function to take a number of ticks instead of a
pointer to a timespec, then call that directly with appropriate
arguments, from both the normal and compat syscall routines.
>
> Since I have no way to test this code maybe is better add no function
> at all and simply using a warning message if someone try compiling
> this code with CONFIG_COMPAT enabled...
No. The fact that you cannot test it is no excuse for submitting
something which is _known_ to be broken. Once you have it to the point
where a casual observer can't point out errors, then people can test it
for you.
--
dwmw2
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists