Re: [PATCH V1 07/16] xen/dm: Make x86's DM feature common

[Jan Beulich <jbeulich@xxxxxxxx>]

On 22.09.2020 18:46, Oleksandr wrote:
> 
> On 14.09.20 18:56, Jan Beulich wrote:
> Hi Jan
> 
>> On 10.09.2020 22:22, Oleksandr Tyshchenko wrote:
>>> --- a/xen/include/xen/hypercall.h
>>> +++ b/xen/include/xen/hypercall.h
>>> @@ -150,6 +150,18 @@ do_dm_op(
>>>       unsigned int nr_bufs,
>>>       XEN_GUEST_HANDLE_PARAM(xen_dm_op_buf_t) bufs);
>>>   
>>> +struct dmop_args {
>>> +    domid_t domid;
>>> +    unsigned int nr_bufs;
>>> +    /* Reserve enough buf elements for all current hypercalls. */
>>> +    struct xen_dm_op_buf buf[2];
>>> +};
>>> +
>>> +int arch_dm_op(struct xen_dm_op *op,
>>> +               struct domain *d,
>>> +               const struct dmop_args *op_args,
>>> +               bool *const_op);
>>> +
>>>   #ifdef CONFIG_HYPFS
>>>   extern long
>>>   do_hypfs_op(
>> There are exactly two CUs which need to see these two declarations.
>> Personally I think they should go into a new header, or at least
>> into one that half-way fits (from the pov of its other contents)
>> and doesn't get included by half the code base. But maybe it's
>> just me ...
> 
> I am afraid, I didn't get why this header is not suitable for keeping 
> this stuff...

While I have no major objection against exposing arch_dm_op() to more
than just the relevant CUs, I don't think I'd like to see struct
dmop_args becoming visible to "everyone", and in particular changes
to it causing a re-build of (almost) everything.

Jan