Re: [Xen-devel] [PATCH v2 for 4.5] arm32: fix build after 063188f4b3

[Tim Deegan <tim@xxxxxxx>]

At 14:13 +0100 on 14 Oct (1413292415), Jan Beulich wrote:
> >>> On 14.10.14 at 15:06, <julien.grall@xxxxxxxxxx> wrote:
> > On 10/14/2014 01:57 PM, Ian Campbell wrote:
> >> On Tue, 2014-10-14 at 13:43 +0100, Julien Grall wrote:
> >>> On 10/14/2014 10:15 AM, Ian Campbell wrote:
> >>>> On Mon, 2014-10-13 at 16:29 +0100, Julien Grall wrote:
> >>>>> +int do_smc(register_t function_id, ...);
> >>>>
> >>>> Are you sure that the variadic function calling convention is the same
> >>>> as for a regular function call? I'm not entirely clear having read
> >>>> AAPCS, it says they are marshalled according to "the standard base". 
> >>>
> >>> All the parameters fits in a register, so the compiler will effectively
> >>> use the first registers to pass arguments.
> >> 
> >> Does it? Even with variadic functions? It's not unheard of for an ABI to
> >> fallback to pushing things onto the stack for such cases, since it works
> >> out far easier in stdargs.h.
> > 
> > You are right, it looks like it's compiler depend how variadic function
> > will be called.
> 
> Now that should never happen - there ought to be an ABI that all
> compilers abide by.

Indeed.  It may depend on compiler _flags_ but it muct be one of the
standard layouts.  By my reading of the AAPCS, calls with up to four
word-sized integer/pointer arguments (even variadic) will use
registers -- "marshalled as for the base standard" just means "not
using any of the optional rules for FP arguments".

All of which reminds me of this (from last year):
https://www.mikeash.com/pyblog/friday-qa-2013-06-28-anatomy-of-a-compiler-bug.html

Tim.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel