From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-6.8 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_PASS,URIBL_BLOCKED autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 1EF70C0044C for ; Sat, 3 Nov 2018 21:00:47 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id AEF692081F for ; Sat, 3 Nov 2018 21:00:46 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org AEF692081F Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=lucaceresoli.net Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=linux-kernel-owner@vger.kernel.org Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1727098AbeKDGNK (ORCPT ); Sun, 4 Nov 2018 01:13:10 -0500 Received: from srv-hp10-72.netsons.net ([94.141.22.72]:49537 "EHLO srv-hp10-72.netsons.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726622AbeKDGNK (ORCPT ); Sun, 4 Nov 2018 01:13:10 -0500 Received: from [88.147.30.68] (port=45068 helo=[192.168.77.57]) by srv-hp10.netsons.net with esmtpsa (TLSv1.2:ECDHE-RSA-AES128-GCM-SHA256:128) (Exim 4.91) (envelope-from ) id 1gJ322-003e8j-Tm; Sat, 03 Nov 2018 22:00:35 +0100 From: Luca Ceresoli Subject: Re: [PATCH 2/7] dt-bindings: mfd: ds90ux9xx: add description of TI DS90Ux9xx I2C bridge To: Vladimir Zapolskiy , Lee Jones , Rob Herring Cc: Linus Walleij , Marek Vasut , Laurent Pinchart , Wolfram Sang , devicetree@vger.kernel.org, linux-gpio@vger.kernel.org, linux-media@vger.kernel.org, linux-kernel@vger.kernel.org, Vladimir Zapolskiy References: <20181008211205.2900-1-vz@mleia.com> <20181008211205.2900-3-vz@mleia.com> <2c90affe-0972-751d-8312-3d15d130c3fb@mleia.com> Message-ID: <4b235b04-e2ec-3dc4-892a-2ea6c87ff288@lucaceresoli.net> Date: Sat, 3 Nov 2018 22:00:36 +0100 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.2.1 MIME-Version: 1.0 In-Reply-To: <2c90affe-0972-751d-8312-3d15d130c3fb@mleia.com> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 7bit X-AntiAbuse: This header was added to track abuse, please include it with any abuse report X-AntiAbuse: Primary Hostname - srv-hp10.netsons.net X-AntiAbuse: Original Domain - vger.kernel.org X-AntiAbuse: Originator/Caller UID/GID - [47 12] / [47 12] X-AntiAbuse: Sender Address Domain - lucaceresoli.net X-Get-Message-Sender-Via: srv-hp10.netsons.net: authenticated_id: luca+lucaceresoli.net/only user confirmed/virtual account not confirmed X-Authenticated-Sender: srv-hp10.netsons.net: luca@lucaceresoli.net X-Source: X-Source-Args: X-Source-Dir: Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Vladimir, On 31/10/18 21:12, Vladimir Zapolskiy wrote: > Hi Luca, > > thank you for review. > > On 10/30/2018 06:43 PM, Luca Ceresoli wrote: >> Hi Vladimir, >> >> On 08/10/18 23:12, Vladimir Zapolskiy wrote: >>> From: Vladimir Zapolskiy >>> >>> TI DS90Ux9xx de-/serializers are capable to route I2C messages to >>> I2C slave devices connected to a remote de-/serializer in a pair, >>> the change adds description of device tree bindings of the subcontroller >>> to configure and enable this functionality. >>> >>> Signed-off-by: Vladimir Zapolskiy >>> --- >>> .../bindings/mfd/ti,ds90ux9xx-i2c-bridge.txt | 61 +++++++++++++++++++ >>> 1 file changed, 61 insertions(+) >>> create mode 100644 Documentation/devicetree/bindings/mfd/ti,ds90ux9xx-i2c-bridge.txt >>> >>> diff --git a/Documentation/devicetree/bindings/mfd/ti,ds90ux9xx-i2c-bridge.txt b/Documentation/devicetree/bindings/mfd/ti,ds90ux9xx-i2c-bridge.txt >>> new file mode 100644 >>> index 000000000000..4169e382073a >>> --- /dev/null >>> +++ b/Documentation/devicetree/bindings/mfd/ti,ds90ux9xx-i2c-bridge.txt >>> @@ -0,0 +1,61 @@ >>> +TI DS90Ux9xx de-/serializer I2C bridge subcontroller >>> + >>> +Required properties: >>> +- compatible: Must contain a generic "ti,ds90ux9xx-i2c-bridge" value and >>> + may contain one more specific value from the list: >>> + "ti,ds90ux925-i2c-bridge", >>> + "ti,ds90ux926-i2c-bridge", >>> + "ti,ds90ux927-i2c-bridge", >>> + "ti,ds90ux928-i2c-bridge", >>> + "ti,ds90ux940-i2c-bridge". >>> + >>> +Required properties of a de-/serializer device connected to a local I2C bus: >>> +- ti,i2c-bridges: List of phandles to remote de-/serializer devices with >>> + two arguments: id of a local de-/serializer FPD link and an assigned >>> + I2C address of a remote de-/serializer to be accessed on a local >>> + I2C bus. >>> + >>> +Optional properties of a de-/serializer device connected to a local I2C bus: >>> +- ti,i2c-bridge-maps: List of 3-cell values: >>> + - the first argument is id of a local de-/serializer FPD link, >>> + - the second argument is an I2C address of a device connected to >>> + a remote de-/serializer IC, >>> + - the third argument is an I2C address of the remote I2C device >>> + for access on a local I2C bus. >> >> BTW I usually use names "remove slave" address and "alias" for bullets 2 >> and 3. These are the names from the datasheets, and are clearer IMO. >> > > Definitely you are correct, I find that verbose descriptions might be > more appropriate and self-explanatory for anyone, who is not closely familiar > with the IC series. I'll consider to add the names from the datasheets > as well. > >> Now to the big stuff. >> >> I find a static map in the "local" chip DT node is a limit. You might >> have to support multiple models of remote device, where you'll know the >> model only when after it gets connected. Think Beaglebone capes, but >> over FPD-Link 3. This scenario opens several issues, but specifically >> for I2C address mapping I addressed it by adding in the "local" chip's >> DT node a pool of I2C aliases it can use. The DT author is responsible >> to pick addresses that are not used on the same I2C bus, which cannot be >> done at runtime reliably. > > Here I see several important topics raised. > > 1) A static map in the "local" chip DT node is not a limit in sense that > it is optional, so it would be a working model just to omit the property, > however it may (or may not) require another handlers to bridge remote > I2C devices, for instance 'ti,i2c-bridge-pass-all' property, or new > UAPI. Do you mean when the "pass-all" method and the "static map in the local node" method are insufficient, then the solution is to implement a third method that is powerful enough? If that's what you mean, then I think we should rather [try to] implement from the beginning a method that is powerful enough to handle all the cases we can foresee. > 2) About supporting multiple models of remote PCBs in the same dts file, > it might be an excessive complication to predict a proper description > of an unknown in advance complex device, so, a better solution should > be to apply DT overlays in runtime, but at any time the hardware > description and the mapping shall be precisely defined. I agree runtime DT overlays is the correct way of handling multiple remote boards, but not if they contain a ti,i2c-bridge-maps (physical-to-alias I2C address map). Consider the case where we have N different main board (with SoC and "local" [de]serializer) and M peripheral boards (display+deserializer or sensor+serializer). Then we'd need up to N*M DT overlays, each with the alias map for a specific pair. This is because the map maps a physical address that exists on the remote side (peripheral board) to an alias that exists on the local side (main board). To realistically model the hardware, the DT overlay should contain the removable components, including the physical address on the remote side (e.g. touchscreen slave address), but not including the alias on the local side (a slave address that is not used on the SoC i2c bus). > 3) About a pool of vacant I2C addresses, I dislike the idea that there > will be no definite or constant I2C address in runtime for a particular > remote slave device. Indeed that's a little annoyance. But you'll face it only during kernel development and debugging, not at the userspace interface. See below the discussion about i2c_adapter. [...] >> Here's my current draft on a dual/quad port deserializer: >> >> &i2c0 { >> serializer@3d { >> reg = <0x3d>; >> ... >> >> /* Guaranteed not physically present on i2c0 */ >> i2c-alias-pool = /bits/ 8 <0x20 0x21 0x22 0x23 0x24 0x25>; >> >> rxports { >> #address-cells = <1>; >> #size-cells = <0>; >> >> rxport@0 { >> reg = <0>; >> remote-i2c-bus { /* The proxied I2C bus on rxport 0 */ >> #address-cells = <1>; >> #size-cells = <0>; >> >> eeprom@51 { >> reg = <0x51>; >> compatible = "at,24c02"; >> }; >> }; >> >> rxport@1 { >> reg = <1>; >> remote-i2c-bus { /* The proxied I2C bus on rxport 1 */ >> #address-cells = <1>; >> #size-cells = <0>; >> >> eeprom@51 { >> reg = <0x51>; >> compatible = "at,24c02"; >> }; >> }; >> }; >> }; >> }; >> }; >> >> At probe time the serializer driver instantiates one new i2c_adapter for >> each rxport. Any remote device is added (removed) to that adapter, then >> the driver finds an available alias and maps (unmaps) it. The > > I avoid using i2c_adapter object, because then you get a confusing access > to right the same device on two logical I2C buses. This is not the way > how I2C muxes operate or are expected to operate, commonly I2C muxes contain > a protocol to access muxed devices, which are "invisible" on a host bus, > and here a local IC behaves like an I2C device with multiple addresses. Sorry, this whole paragraph is not quite clear to me. Well, except "I avoid using i2c_adapter object", which is very clear and opposite to my approach. :) I'm instantiating an i2c_adapter for each rx port because it's the best model of the real world that I could devise. See the picture: .----------------. | | i2c-5 ,-- touch@50 ----. i2c-0 | adap0----------------+-- eeprom@51 SoC |-----------| atr@3d | ----' | adap1----------------+-- eeprom@51 | | i2c-6 `-- touch@50 `----------------' I called the central object ATR (address translator), which is the i2c remotizer feature in the DS90Ux chipset. Topologically it is somewhat similar to a mux, but: * it also translates addresses (hence the name) * electrically it never attaches the upstream and downstream busses, they are driven independently, with message buffering etc. The touch and eeprom in the example are instantiated on the physical busses where they are wired (i2c-5 or i2c-6) with their physical addresses on those busses. > Note, that following an advice from Wolfram I'm going to send the i2c-bridge > cell driver into inclusion under drivers/i2c/muxes/ , even if the device > driver does not register a mux. > >> transactions are handled in a way similar to i2c-mux, i.e. the ds90* >> i2c_adapter has a master_xfer callback that changes the remote slave >> address to the corresponding alias, then calls parent->algo->master_xfer(). >> >> Note how both eeproms in the example have the same physical address. >> They will be given two different aliases. > > The question is how to determine which runtime assigned address represents > which eeprom of two. The remote/alias scheme I propose makes it transparent. You generally don't care about the assigned alias. To access e.g. each of the eeproms the path is always reliable and represents the physical world: - i2c-0 -> 00-003d -> rxport0 -> ??-0051 - i2c-0 -> 00-003d -> rxport1 -> ??-0051 The alias matters only when debugging the driver or inspecting the physical bus with a protocol analyzer or oscilloscope. Bye, -- Luca