Re: [PATCH] xenbus: Add support for xenbus backend in stub domain

[Daniel De Graaf <dgdegra@tycho.nsa.gov>]

On 01/13/2012 03:20 AM, Jan Beulich wrote:
>>>> On 13.01.12 at 00:36, Daniel De Graaf <dgdegra@tycho.nsa.gov> wrote:
>> --- a/include/xen/xenbus_dev.h
>> +++ b/include/xen/xenbus_dev.h
>> @@ -38,4 +38,18 @@
>>  #define IOCTL_XENBUS_BACKEND_EVTCHN			\
>>  	_IOC(_IOC_NONE, 'B', 0, 0)
>>  
>> +#define IOCTL_XENBUS_ALLOC				\
>> +	_IOC(_IOC_NONE, 'B', 1, sizeof(struct ioctl_xenbus_alloc))
>> +struct ioctl_xenbus_alloc {
>> +	/* IN */
>> +	/* The domain ID (must exist) for xenstore */
>> +	uint16_t dom;
>> +	uint16_t pad;
>> +	/* OUT */
>> +	/* The port allocated for xenbus communication */
>> +	uint32_t port;
>> +	/* Always set to GNTTAB_RESERVED_XENSTORE */
>> +	uint32_t grant_ref;
>> +};
> 
> As said in my reply to the previous patch version - if the functionality
> differs, the number *and* name should be different from the legacy
> implementation's. Otherwise, how should compatible user space code
> be written?
> 
> Jan
> 

This implementation has the same functionality as the legacy ioctl,
although it has a different number and is performed on a different file,
so it is already impossible to make automatically compatible userspace
code. The structure name was changed to match other ioctl arguments, but
the contents should be the same as in the legacy ioctl. If changing what
file the ioctl is performed on justifies changing the ioctl name, then I
would prefer the simpler interface where the domain ID is passed in as the
ioctl parameter instead of a structure that only has one useful output.