Re: [PATCH v2] Add preadv and pwritev system calls.

[Heiko Carstens <heiko.carstens@de.ibm.com>]

On Fri, Dec 12, 2008 at 08:29:29AM -0700, Matthew Wilcox wrote:
> On Fri, Dec 12, 2008 at 03:00:40PM +0100, Gerd Hoffmann wrote:
> > The patch sports the actual system call implementation and the windup in
> > the x86 system call tables.  Other archs are TBD.
> 
> > +asmlinkage ssize_t sys_preadv(unsigned long fd, const struct iovec __user *vec,
> > +                              unsigned long vlen, loff_t pos)
> > +asmlinkage ssize_t sys_pwritev(unsigned long fd, const struct iovec __user *vec,
> > +                               unsigned long vlen, loff_t pos)
> 
> Are these prototypes required?  MIPS and PARISC will need wrappers to
> fix them if they are.  These two architectures have an ABI which
> requires 64-bit arguments to be passed in aligned pairs of registers,
> but glibc doesn't know that (and given the existence of syscall(3),
> can't do much about it even if it knew), so some of the arguments end up
> in the wrong registers.
> 
> Things will go much better if we can prototype these as:
> 
> asmlinkage ssize_t sys_preadv(unsigned int fd, const struct iovec __user *vec,
> 				loff_t pos, unsigned long vlen);
> asmlinkage ssize_t sys_pwritev(unsigned int fd, const struct iovec __user *vec,
> 				loff_t pos, unsigned long vlen);
> 
> That way 'pos' ends up split between arg2 and arg3 and vlen ends up in
> arg4 instead of having vlen in arg2 and pos in arg3 and arg4 which then
> have to be munged to be in arg4 and arg5 by a compat wrapper.
> 
> I seem to recall the s390 folks having some concerns with this kind of
> thing too, but I forget what they are, so I'll let them weigh in on
> this.

I think a lot of stuff that needs to be known for new system calls was
written up with these two postings:

http://marc.info/?l=linux-arch&m=118277150812137&w=2
http://lkml.org/lkml/2007/8/1/354