[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <DM6PR11MB4657E51AD937BBA5DC2B1FF19BFF9@DM6PR11MB4657.namprd11.prod.outlook.com>
Date: Tue, 10 Jan 2023 10:54:20 +0000
From: "Kubalewski, Arkadiusz" <arkadiusz.kubalewski@...el.com>
To: Jiri Pirko <jiri@...nulli.us>
CC: Jakub Kicinski <kuba@...nel.org>,
Maciek Machnikowski <maciek@...hnikowski.net>,
'Vadim Fedorenko' <vfedorenko@...ek.ru>,
'Jonathan Lemon' <jonathan.lemon@...il.com>,
'Paolo Abeni' <pabeni@...hat.com>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"linux-arm-kernel@...ts.infradead.org"
<linux-arm-kernel@...ts.infradead.org>,
"linux-clk@...r.kernel.org" <linux-clk@...r.kernel.org>
Subject: RE: [RFC PATCH v4 0/4] Create common DPLL/clock configuration API
>From: Jiri Pirko <jiri@...nulli.us>
>Sent: Monday, January 9, 2023 5:30 PM
>
>Mon, Jan 09, 2023 at 03:43:01PM CET, arkadiusz.kubalewski@...el.com wrote:
>>>From: Jiri Pirko <jiri@...nulli.us>
>>>Sent: Monday, December 12, 2022 2:59 PM
>>>
>>>Fri, Dec 09, 2022 at 05:31:04PM CET, kuba@...nel.org wrote:
>>>>On Fri, 9 Dec 2022 15:09:08 +0100 Maciek Machnikowski wrote:
>>>>> On 12/9/2022 12:07 PM, Jiri Pirko wrote:
>>>>> > Looking at the documentation of the chips, they all have mupltiple
>>>DPLLs
>>>>> > on a die. Arkadiusz, in your proposed implementation, do you model
>>>each
>>>>> > DPLL separatelly? If yes, then I understand the urgency of need of a
>>>>> > shared pin. So all DPLLs sharing the pin are part of the same chip?
>>>>> >
>>>>> > Question: can we have an entity, that would be 1:1 mapped to the
>>>actual
>>>>> > device/chip here? Let's call is "a synchronizer". It would contain
>>>>> > multiple DPLLs, user-facing-sources(input_connector),
>>>>> > user-facing-outputs(output_connector), i/o pins.
>>>>> >
>>>>> > An example:
>>>>> > SYNCHRONIZER
>>>>> >
>>>>> >
>>>┌───────────────────────────────────────┐
>>>>> > │
>>>│
>>>>> > │
>>>│
>>>>> > SyncE in connector │ ┌─────────┐
>>>│ SyncE out connector
>>>>> > ┌───┐ │in pin 1 │DPLL_1 │ out pin
>>>1│ ┌───┐
>>>>> > │ ├─────────┼──────────────┤
>>>├──────────────┼────┤ │
>>>>> > │ │ │ │ │
>>>│ │ │
>>>>> > └───┘ │ │ │
>>>│ └───┘
>>>>> > │ │ │
>>>│
>>>>> > │ ┌──┤ │
>>>│
>>>>> > GNSS in connector │ │ └─────────┘
>>>│
>>>>> > ┌───┐ │in pin 2 │ out pin
>>>2│ EXT SMA connector
>>>>> > │ ├─────────┼───────────┘
>>>│ ┌───┐
>>>>> > │ │ │
>>>┌───────────┼────┤ │
>>>>> > └───┘ │ │
>>>│ │ │
>>>>> > │ │
>>>│ └───┘
>>>>> > │ │
>>>│
>>>>> > EXT SMA connector │ │
>>>│
>>>>> > ┌───┐ mux │in pin 3 ┌─────────┐ │
>>>│
>>>>> > │ ├────┬────┼───────────┐ │ │ │
>>>│
>>>>> > │ │ │ │ │ │DPLL_2 │ │
>>>│
>>>>> > └───┘ │ │ │ │ │ │
>>>│
>>>>> > │ │ └──┤ ├──┘
>>>│
>>>>> > │ │ │ │
>>>│
>>>>> > EXT SMA connector │ │ │ │
>>>│
>>>>> > ┌───┐ │ │ │ │
>>>│
>>>>> > │ ├────┘ │ └─────────┘
>>>│
>>>>> > │ │ │
>>>│
>>>>> > └───┘
>>>└───────────────────────────────────────┘
>>>>> >
>>>>> > Do I get that remotelly correct?
>>>>>
>>>>> It looks goot, hence two corrections are needed:
>>>>> - all inputs can go to all DPLLs, and a single source can drive more
>>>>> than one DPLL
>>>>> - The external mux for SMA connector should not be a part of the
>>>>> Synchronizer subsystem - I believe there's already a separate MUX
>>>>> subsystem in the kernel and all external connections should be
>handled
>>>>> by a devtree or a similar concept.
>>>>>
>>>>> The only "muxing" thing that could potentially be modeled is a
>>>>> synchronizer output to synchronizer input relation. Some synchronizers
>>>>> does that internally and can use the output of one DPLL as a source
>for
>>>>> another.
>>>>
>>>>My experience with DT and muxes is rapidly aging, have you worked with
>>>>those recently? From what I remember the muxes were really.. "embedded"
>>>>and static compared to what we want here.
>>>
>>>Why do you think we need something "non-static"? The mux is part of the
>>>board, isn't it? That sounds quite static to me.
>>>
>>>
>>>>
>>>>Using DT may work nicely for defining the topology, but for config we
>>>>still need a different mechanism.
>>>
>>>"config" of what? Each item in topology would be configure according to
>>>the item type, won't it?
>>>
>>>[...]
>>
>>
>>Hi guys,
>>
>>We have been trying to figure out feasibility of new approach proposed on
>our
>>latest meeting - to have a single object which encapsulates multiple
>DPLLs.
>>
>>Please consider following example:
>>
>>Shared common inputs:
>>i0 - GPS / external
>>i1 - SMA1 / external
>>i2 - SMA2 / external
>>i3 - MUX0 / clk recovered from PHY0.X driven by MAC0
>>i4 - MUX1 / clk recovered from PHY1.X driven by MAC1
>>
>>+---------------------------------------------------------+
>>| Channel A / FW0 +---+ |
>>| i0--| | |
>>| +---+ | | |
>>| PHY0.0--| | i1--| D | |
>>| | | | P | |
>>| PHY0.1--| M | i2--| L | +---+ +--------+ |
>>| | U | | L |---| |---| PHY0.0 |--|
>>| PHY0.2--| X |-+---------i3--| 0 | | | +--------+ |
>>| | 0 | |+------+ | |---| M |---| PHY0.1 |--|
>>| ... --| | || MUX1 |-i4--| | | A | +--------+ |
>>| | | |+------+ +---+ | C |---| PHY0.2 |--|
>>| PHY0.7--| | | i0--| | | 0 | +--------+ |
>>| +---+ | | |---| |---| ... |--|
>>| | i1--| D | | | +--------+ |
>>| | | P |---| |---| PHY0.7 |--|
>>| | i2--| L | +---+ +--------+ |
>>| | | L | |
>>| \---------i3--| 1 | |
>>| +------+ | | |
>>| | MUX1 |-i4--| | |
>>| +------+ +---+ |
>>+---------------------------------------------------------+
>>| Channel B / FW1 +---+ |
>>| i0--| | |
>>| | | |
>>| i1--| D | |
>>| +---+ | P | |
>>| PHY1.0--| | i2--| L | +---+ +--------+ |
>>| | | +------+ | L |---| |---| PHY1.0 |--|
>>| PHY1.1--| M | | MUX0 |-i3--| 0 | | | +--------+ |
>>| | U | +------+ | |---| M |---| PHY1.1 |--|
>>| PHY1.2--| X |-+---------i4--| | | A | +--------+ |
>>| | 1 | | +---+ | C |---| PHY1.2 |--|
>>| ... --| | | i0--| | | 1 | +--------+ |
>>| | | | | |---| |---| ... |--|
>>| PHY1.7--| | | i1--| D | | | +--------+ |
>>| +---+ | | P |---| |---| PHY1.7 |--|
>>| | i2--| L | +---+ +--------+ |
>>| |+------+ | L | |
>>| || MUX0 |-i3--| 1 | |
>>| |+------+ | | |
>>| \---------i4--| | |
>>| +---+ |
>>+---------------------------------------------------------+
>
>What is "a channel" here? Are these 2 channels part of the same physival
>chip? Could you add the synchronizer chip/device entities to your drawing?
>
No.
A "Synchronization Channel" on a switch would allow to separate groups
of physical ports. Each channel/group has own "Synchronizer Chip", which is
used to drive PHY clocks of that group.
"Synchronizer chip" would be the 2 DPLLs on old draw, something like this:
+--------------------------------------------------------------+
| Channel A / FW0 +-------------+ +---+ +--------+ |
| i0--|Synchronizer0|---| |---| PHY0.0 |--|
| +---+ | | | | +--------+ |
| PHY0.0--| | i1--| |---| M |---| PHY0.1 |--|
| | | | +-----+ | | A | +--------+ |
| PHY0.1--| M | i2--| |DPLL0| | | C |---| PHY0.2 |--|
| | U | | +-----+ | | 0 | +--------+ |
| PHY0.2--| X |--+---i3--| +-----+ |---| |---| ... |--|
| | 0 | | | |DPLL1| | | | +--------+ |
| ... --| | | /-i4--| +-----+ |---| |---| PHY0.7 |--|
| | | | | +-------------+ +---+ +--------+ |
| PHY0.7--| | | | |
| +---+ | | |
+----------------|-|-------------------------------------------+
| Channel B / FW1| | +-------------+ +---+ +--------+ |
| | | i0--|Synchronizer1|---| |---| PHY1.0 |--|
| +---+ | | | | | | +--------+ |
| PHY1.0--| | | | i1--| |---| M |---| PHY1.1 |--|
| | | | | | +-----+ | | A | +--------+ |
| PHY1.1--| M | | | i2--| |DPLL0| | | C |---| PHY1.2 |--|
| | U | | | | +-----+ | | 1 | +--------+ |
| PHY1.2--| X | \-|-i3--| +-----+ |---| |---| ... |--|
| | 1 | | | |DPLL1| | | | +--------+ |
| ... --| |----+-i4--| +-----+ |---| |---| PHY1.7 |--|
| | | +-------------+ +---+ +--------+ |
| PHY1.7--| | |
| +---+ |
+--------------------------------------------------------------+
Also, please keep in mind that is an example, there could be easily 4
(or more) channels wired similarly.
>
>>
>>This is a simplified network switch board example.
>>It has 2 synchronization channels, where each channel:
>>- provides clk to 8 PHYs driven by separated MAC chips,
>>- controls 2 DPLLs.
>>
>>Basically only given FW has control over its PHYs, so also a control over
>it's
>>MUX inputs.
>>All external sources are shared between the channels.
>>
>>This is why we believe it is not best idea to enclose multiple DPLLs with
>one
>>object:
>>- sources are shared even if DPLLs are not a single synchronizer chip,
>>- control over specific MUX type input shall be controllable from
>different
>>driver/firmware instances.
>>
>>As we know the proposal of having multiple DPLLs in one object was a try
>to
>>simplify currently implemented shared pins. We fully support idea of
>having
>>interfaces as simple as possible, but at the same time they shall be
>flexible
>>enough to serve many use cases.
>>
>>Right now the use case of single "synchronizer chip" is possible (2 DPLLs
>with
>>shared inputs), as well as multiple synchronizer chips with shared inputs.
>>
>>If we would entirely get rid of sharing pins idea and instead allowed only
>to
>>have multiple DPLLs in one object, we would fall back to the problem where
>>change on one input is braking another "synchronizer chip" input.
>>I.e. considering above scheme, user configured both channels to use SMA1
>1MHz.
>>If SMA1 input is changed to 10MHz, all DPLLs are affected, thus all using
>that
>
>You say "SMA1 input *is changed*". Could you add to your drawing:
>1) Who is the one triggering the change.
>2) Entity that manages the SMA input and applies the configuration.
>
A user or some tool, this change requires to switch a frequency on a signal
generator connected to that SMA1. Whatever would make the change is an external
entity here. The draw show connections on board, don't see a point on having a
external signal generator or user connected to the board :)
If something is not clear, we could prepare some different draw, please just
let me know what exactly we want to see. It sound like a sequence diagram?
Thanks!
Arkadiusz
>
>>input shall be notified, as long as that input is shared.
>>For the drivers that have single point of control over dpll, they might
>just
>>skip those requests. But if there are multiple firmware instances
>controlling
>>multiple DPLLs, they would process it independently.
>>
>>Current implementation is the most flexible and least complex for the
>level of
>>flexibility it provides.
>>
>>BR, Happy new year!
>>Arkadiusz
Powered by blists - more mailing lists