Re: [PATCH 16/17] arm: asm/syscall.h (unfinished)

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Fri, Apr 24, 2009 at 05:15:03PM -0700, Roland McGrath wrote:
> The syscall_get_nr() function here is not really right.  I don't
> know enough about ARM to finish it correctly.  It needs to figure
> out if the blocked user task is really in the kernel for a system
> call and return -1 if not.

That bit is rather hard - we maintain no global state as to whether a
task is in a syscall or not.  We also do not maintain a global view
of which syscall number is being executed.  The kernel just hasn't
required either of these things before.

> I also did not try to handle all the different ABI variants, which I
> don't really understand.

There are two places that the syscall number comes from: EABI and thumb
both use R7.  OABI puts the value in the instruction itself.

In short, I don't know what to do about this either.  I don't think
there's a quick and simple answer.

Your implementation of syscall_get_arguments() looks wrong - for OABI
it makes sense because there is no padding in the allocation of registers.
However, for EABI, there is padding, so 64-bit values always come in
using an even+odd register number.  Short of maintaining some sort of
table describing the argument placements for every kernel syscall (eww)
I'm not sure how this could be fixed.

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