Re: struct blkif_request_segment_aligned

["Roger Pau Monné" <roger.pau@citrix.com>]

On 03/02/14 12:36, Jan Beulich wrote:
>>>> On 03.02.14 at 10:40, Ian Campbell <Ian.Campbell@citrix.com> wrote:
>> On Fri, 2014-01-31 at 12:04 +0000, Jan Beulich wrote:
>>> Roger,
>>>
>>> so you introduced this, yet looking in a little closer detail I can't seem
>>> to understand why: struct blkif_request_segment is identical in layout,
>>> the sole difference between the two is that in the new structure the
>>> padding field has a name, whereas in the old one it doesn't.
>>
>> Is this something to do with Linux' use of __attribute__((packed)) once
>> again causing confusion? (I really hope not API deviation...)
> 
> Yes, I think it has to do with Linux'es way of defining these
> structures: My assumption is that the embedded (but not such
> attributed) definition of struct blkif_request_segment inside struct
> struct blkif_request_rw was assumed to also be packed (which it
> isn't, or else upstream Linux front-/backends wouldn't work with
> other back-/frontends), thus apparently making it necessary to
> have an "aligned" (i.e. un-packed) variant thereof.

Yes, this is my fault for wrongly assuming struct blkif_request_segment
inside struct blkif_request_rw was also packed, which it is not (or else
it would break with non Linux backends). Thanks for sending the Xen side
patch, I will take care of the Linux side if it's fine with you.

Roger.