On 8/9/2017 12:46 PM, Don Hiatt wrote:
> On 8/8/2017 12:05 PM, Doug Ledford wrote:
>> On Thu, 2017-06-08 at 13:37 -0400, Don Hiatt wrote:
>>> slid field in struct ib_wc is increased to 32 bits.
>>> This enables core components to use larger LIDs if needed.
>>> The user ABI is unchanged and return 16 bit values when queried.
>> Hi Don,
>>
>> My comments here are not really about this particular patch, I just
>> grabbed it from the pile as one to respond to. But my question is
>> this:
>>
>> Multiple of these patch say the user ABI is unchanged and return 16 bit
>> values when queried. What's your plan for a long term solution to
>> this? Eventually, the user ABI will need to be extended to support
>> 32bit LIDs, yes? Or do you plan to try and hide this forever?
>>
>> As for the kernel to user space driver API, I waffled a bit about
>> whether we shouldn't go ahead and return the 32bit value to the driver
>> and let it do the conversion to 16 bit for the user ABI, then decided
>> this method provides a cleaner means of supporting newer kernels with
>> older verbs OPA drivers and so didn't object, but it still leaves the
>> question about the long term plans here.
>>
> Hi Doug,
>
> Our long term plan is to use the new verbs 2 API for extending
> the LID size. We do not plan to keep this hidden forever and plan to
> use the new API once it's available.
Good. That's more or less what I wanted to hear ;-). Thanks.
--
Doug Ledford <dledford@xxxxxxxxxx>
GPG Key ID: B826A3330E572FDD
Key fingerprint = AE6B 1BDA 122B 23B4 265B 1274 B826 A333 0E57 2FDD
Attachment:
signature.asc
Description: OpenPGP digital signature
