FUTEX_CMP_REQUEUE and 6-argument syscall

FUTEX_CMP_REQUEUE and 6-argument syscall

[Pete Zaitcev]

Hi, Martin,

the following changeset adds an extra argument for sys_futex:
 http://linux.bkbits.net:8080/linux-2.5/gnupatch@40bca55f9uSk3jznb3RD7ltnQvJKbA

So, now we have a syscall with 6 arguments. Do you have any suggestions
how to handle it on s390? The main decision would be if we stick to
the C calling convention and put a pointer to the list into %r7, or
if we just put the argument into %r7 (thus precluding syscalls with
7 arguments for sure, but saving a get_user somewhere). After that
we need to change arch/s390/kernel/entry.S:trace_svc, and the 31 bit
compat wrapper.

I'm curious if any other architecture was caught out by this.

-- Pete

Re: FUTEX_CMP_REQUEUE and 6-argument syscall

[Ralf Baechle]

On Thu, Jun 03, 2004 at 11:02:21AM -0700, Pete Zaitcev wrote:

> So, now we have a syscall with 6 arguments. Do you have any suggestions
> how to handle it on s390? The main decision would be if we stick to
> the C calling convention and put a pointer to the list into %r7, or
> if we just put the argument into %r7 (thus precluding syscalls with
> 7 arguments for sure, but saving a get_user somewhere). After that
> we need to change arch/s390/kernel/entry.S:trace_svc, and the 31 bit
> compat wrapper.
> 
> I'm curious if any other architecture was caught out by this.

6 is the limit on o32 MIPS where we pass 4 arguments in registers and two
on the stack but it would be easy to extend that.  On 64-bit kernels we
only use upto 8 register arguments but no stack arguments ever so I hope
we don't exceed that limit ever ...

  Ralf

Re: FUTEX_CMP_REQUEUE and 6-argument syscall

[David S. Miller]

On Thu, 3 Jun 2004 20:13:45 +0200
Ralf Baechle <ralf@linux-mips.org> wrote:

> > I'm curious if any other architecture was caught out by this.
> 
> 6 is the limit on o32 MIPS where we pass 4 arguments in registers and two
> on the stack but it would be easy to extend that.  On 64-bit kernels we
> only use upto 8 register arguments but no stack arguments ever so I hope
> we don't exceed that limit ever ...

6 is the limit on Sparc/Sparc64 as well, registers %o0 - %o5 carry
arguments.

Re: FUTEX_CMP_REQUEUE and 6-argument syscall

[Pete Zaitcev]

On Thu, 3 Jun 2004 14:26:30 -0700
"David S. Miller" <davem@redhat.com> wrote:

> > 6 is the limit on o32 MIPS where we pass 4 arguments in registers and two
> > on the stack but it would be easy to extend that.  On 64-bit kernels we
> > only use upto 8 register arguments but no stack arguments ever so I hope
> > we don't exceed that limit ever ...
> 
> 6 is the limit on Sparc/Sparc64 as well, registers %o0 - %o5 carry
> arguments.

It seems that bumping the limit just to 6 would bring s390 in line
with sparcs, this is why I even considered using %r7 to pass the
argument itself instead of a pointer to stack. It's such a bother
to have things in memory.

-- Pete

P.S. Ralf, et. al.: do not forget to keep me on the cc: list, please.

Re: FUTEX_CMP_REQUEUE and 6-argument syscall

[Paul Mackerras]

Pete Zaitcev writes:

> I'm curious if any other architecture was caught out by this.

Six is currently the limit on ppc and ppc64.  It would be possible to
increase it to 7 fairly easily or 8 with a bit more trouble.  Beyond
that we would start getting arguments on the stack.

Paul.

Re: FUTEX_CMP_REQUEUE and 6-argument syscall

[Russell King]

On Fri, Jun 04, 2004 at 10:33:44AM +1000, Paul Mackerras wrote:
> Pete Zaitcev writes:
> 
> > I'm curious if any other architecture was caught out by this.
> 
> Six is currently the limit on ppc and ppc64.  It would be possible to
> increase it to 7 fairly easily or 8 with a bit more trouble.  Beyond
> that we would start getting arguments on the stack.

Six is also the current limit on ARM, and we have the possibility
to go to 7 with some more stack mangling.  However, 8 is more
problematical - I suspect at that point we'll be in the arguments
on user stack territory.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core