Re: [PATCH 2/3] x86: Pass in pt_regs pointer for syscalls that need it

[Tejun Heo <tj@kernel.org>]

Hello, Brian.

Brian Gerst wrote:
>> Here and at other places where the function takes more than one
>> arguments, wouldn't it be better to just take *regs and use other
>> parameters from regs?  That way we won't have to worry about gcc
>> corrupting register frame at all and I think it's cleaner that way.
> 
> Expanding the parameters is good documentation.

Copying from ptregs to appropriately named local variable would
provide at least similar level of documentation but I don't think this
is a big deal one way or the other.

> If there is a risk of tail-call optimization causing the register
> corruption, then asmlinkage_protect() should be used.  The problem
> isn't limited to just the syscalls that take pt_regs.  It's just
> getting the args out of the pt_regs struct was an easy hack to get
> around it.

If pt_regs is being passed with regparm(1) and no other parameter is
specified, it's a proper solution as we can guarantee that callee
can't corrupt (or discard changes to) the register frame no matter
what gcc does.

> I checked the disassembly of these functions and didn't see this
> happen on gcc 4.3.0.

Well, tracking down why run_init_process() is returning 0 with
-fstack-protector wasn't much of fun.  These breakages are very subtle
and if we're gonna pass in pointer to pt_regs anyway and thus can
guarantee such breakage can't happen at no additional cost, I think we
should do that even if it means slightly more argument fetching in a
few places.

Thanks.

-- 
tejun