Re: [RFC PATCH v9 00/10] Create common DPLL configuration API

[Jiri Pirko]

Tue, Jul 11, 2023 at 12:34:11PM CEST, arkadiusz.kubalewski@xxxxxxxxx wrote:
>>From: Jiri Pirko <jiri@xxxxxxxxxxx>
>>Sent: Monday, July 10, 2023 2:10 PM
>>
>>Mon, Jul 10, 2023 at 12:07:30PM CEST, arkadiusz.kubalewski@xxxxxxxxx wrote:
>>>>From: Jiri Pirko <jiri@xxxxxxxxxxx>
>>>>Sent: Wednesday, June 28, 2023 1:16 PM
>>>>Wed, Jun 28, 2023 at 11:15:11AM CEST, arkadiusz.kubalewski@xxxxxxxxx
>>wrote:
>>>>>>From: Jiri Pirko <jiri@xxxxxxxxxxx>
>>>>>>Sent: Tuesday, June 27, 2023 12:18 PM
>>>>>>
>>>>>>Fri, Jun 23, 2023 at 02:38:10PM CEST, arkadiusz.kubalewski@xxxxxxxxx
>>>>>>wrote:
>>>>>>
>>>>>>>v8 -> v9:
>>>>>>
>>>>>>Could you please address all the unresolved issues from v8 and send v10?
>>>>>>I'm not reviewing this one.
>>>>>>
>>>>>>Thanks!
>>>>>
>>>>>Sure, will do, but first missing to-do/discuss list:
>>>>>1) remove mode_set as not used by any driver
>>>
>>>I have implemented in ice (also added back the DPLL_MODE_FREERUN).
>>
>>Uh :/ Why exactly is it needed in this initial submission?
>>
>
>Without mode-set there is no need for device-set at all, right?
>So it is better to implement at least one set command, so we don't
>need remove device-set command entirely.

The enum cmd valu could stay as a placeholder, the rest can go.


>
>>
>>>
>>>>>2) remove "no-added-value" static functions descriptions in
>>>>>   dpll_core/dpll_netlink
>>>
>>>Removed.
>>>
>>>>>3) merge patches [ 03/10, 04/10, 05/10 ] into patches that are compiling
>>>>>   after each patch apply
>>>
>>>Hope Vadim will decide on this, the thing is merging in two patches
>>>doesn't make much sense as there won't be any linking until both patches
>>>are there, so most sense it would be if 3 are merged into one, but
>>>then we will be back to one big blob patch issue.
>>>
>>>>>4) remove function return values descriptions/lists
>>>
>>>Fixed.
>>>
>>>>>5) Fix patch [05/10]:
>>>>>   - status Supported
>>>>>   - additional maintainers
>>>>>   - remove callback:
>>>>>     int (*source_pin_idx_get)(...) from `struct dpll_device_ops`
>>>>>6) Fix patch [08/10]: rethink ice mutex locking scheme
>>>
>>>Fixed.
>>>
>>>>>7) Fix patch [09/10]: multiple comments on
>>>>>https://lore.kernel.org/netdev/ZIQu+%2Fo4J0ZBspVg@nanopsycho/#t
>>>>>8) add PPS DPLL phase offset to the netlink get-device API
>>>>>
>>>
>>>Added few things on this matter
>>>- 1 dpll level attribute:
>>>  - phase-shift - measuring the phase difference between dpll input
>>>    and it's output
>>>- 1 dpll-pin tuple level attribute:
>>>  - pin-phase-adjust - set/get phase adjust of a pin on a dpll
>>>- 2 pin level attributes:
>>>  - pin-phase-adjust-min - provide user with min value that can be set
>>>  - pin-phase-adjust-max - provide user with max value that can be set
>>>- a constant:
>>>  - DPLL_PHASE_SHIFT_DIVIDER similar to DPLL_TEMP_DIVIDER for producing
>>>    fraction value of measured DPLL_A_PHASE_SHIFT
>>
>>Again, why do we need this in this initial submission? Why it can't be a
>>follow-up patchset to extend this? This way we never converge :/
>>Please focus on what we have now and bring it in. Let the extensions to
>>be addressed later on, please.
>>
>
>Well AFAIK, RHEL is doing some monitoring software, so the end-users need this.

They need it for the initial submission? Why? Why can't they wait 1 week
for follow-up patchset?


>
>>
>>
>>>- implemented in dpll netlink and in ice
>>>
>>>>
>>>>You are missing removal of pin->prop.package_label = dev_name(dev); in
>>>>ice.
>>>>
>>>
>>>I didn't touch it, as we still need to discuss it, Jakub didn't respond
>>>on v8 thread.
>>>I don't see why we shall not name it the way. This is most meaningful
>>>label for those pins for the user right now.
>>
>>This is not meaningful, at all. dev_name() changes upon which pci slot
>>you plug the card into. package_label should be an actual label on a
>>silicon package. Why you think this two are related in aby way, makes me
>>really wonder. Could you elaborate the meaningfulness of this?
>>
>
>Without this, from end-user perspective, it would be very confusing.
>As in ice without any label there would 4 pins which differs only with id.

There you go, it does not have any label, yet you are trying hard to
make up some. Does not make sense.


>What names would you suggest?

That is the point I made previously. For synce usecase, the label does
not make sense. There should be no label. You reference the pin by ID
from netdev, that is enough.

I think better to add the check to pin-register so future synce pin
users don't have similar weird ideas. Could you please add this check?

Thanks!



>
>Thank you!
>Arkadiusz
>
>>
>>>
>>>Thank you!
>>>Arkadiusz
>>>
>>>>
>>>>>Thank you!
>>>>>Arkadiusz