Re: [Xen-devel] [PATCH 3/5] xen/x86: Cleanup use of __attribute__((packed))

[Keir Fraser <keir@xxxxxxx>]

On Thu, Mar 13, 2014 at 11:13 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:

>>> On 13.03.14 at 12:00, Andrew Cooper <andrew.cooper3@xxxxxxxxxx> wrote:
> On 13/03/14 08:32, Jan Beulich wrote:

>> But then again I can't see why the attribute would be needed here
>> in the first place.
>
> As indicated, I was overly hesitant with bitfields, but perhaps overly.

Since that's the only thing we care of, perhaps add a
BUILD_BUG_ON(sizeof(union segment_attributes) != 2) somewhere
where it doesn't conflict with the tools/tests/ use, but the type is in
scope.

I don't mind having __packed attribute used in x86_emulate.h, and then have tools/tests/ have a proper wrapper round that header file, as it already has a wrapper round the .c file.

Really most of the suggestions here are acceptable to me however.

 -- Keir

 

> Alright - All hardware interfaces can keep their redundant __packed,
> which also serves partly as documentation.
>
> One interesting question George raises is whether the trace record
> format should be count as well?  It is just a software interface, but
> mixing __packed on different record structures seems a little inconsistent.

And I agree with that - it is an interface specification (even if
scattered around awkwardly) after all.

Jan

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