From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1423532AbdD2UDF (ORCPT ); Sat, 29 Apr 2017 16:03:05 -0400 Received: from mail-pg0-f43.google.com ([74.125.83.43]:32823 "EHLO mail-pg0-f43.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1036093AbdD2UC4 (ORCPT ); Sat, 29 Apr 2017 16:02:56 -0400 Date: Sat, 29 Apr 2017 13:02:52 -0700 From: Bjorn Andersson To: Rob Herring Cc: Andy Gross , David Brown , Frank Rowand , Mark Rutland , linux-arm-msm@vger.kernel.org, linux-soc@vger.kernel.org, devicetree@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [RFC 1/3] dt-binding: soc: qcom: Add binding for RFSA Message-ID: <20170429200252.GS15143@minitux> References: <20170422173519.5782-1-bjorn.andersson@linaro.org> <20170428174239.x62opv5vzx2ce2wt@rob-hp-laptop> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20170428174239.x62opv5vzx2ce2wt@rob-hp-laptop> User-Agent: Mutt/1.7.2 (2016-11-26) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri 28 Apr 10:42 PDT 2017, Rob Herring wrote: > On Sat, Apr 22, 2017 at 10:35:17AM -0700, Bjorn Andersson wrote: > > This adds the binding for describing shared memory buffers for > > implementing the remote filesystem protocol. > > > > Signed-off-by: Bjorn Andersson > > --- > > > > My initial attempt was to mimic the ramoops of just adding the compatible to > > the reserved-memory node, but I have not been able to figure out a sane way of > > getting hold of the base address in the case that the memory region is > > described my a "size" only (done on some platforms). > > I prefer the ramoops way. > Me too :) > > The problem is that we create the reserved_mem objects (and remove the > > memblocks) while we're still operating on the flattened representation, so > > without a phandle it doesn't seem like we have anything to perform the > > comparison with later on. > > I'm not sure I follow. > In my first attempt I extended of_platform_default_populate_init() to also probe my platform_driver and like ramoops acquired the values of reg and memremap() these. This works fine. But for some platforms the memory-region doesn't need a fixed location, it's just required to be a consecutive chunk of physical ram. So I replace the "reg = <>" with a "size = <>", this causes __reserved_mem_alloc_size() to carve out a chunk of memory somewhere and update the associated reserved_mem entry. The reserved_mem will be populated with the phandle from the [flattened] of_node. Later my platform_device is instantiated and I need to figure out the base address that was picked earlier - so that I know which region to memremap(). But as my "rfsa-node" stands on its own it will not have any "phandle", so the reserved_mem->phandle will remain 0. Further more I only have the unflattened of_node - so I can't just compare the address with the flattened version previously used. So the problem at hand is how to match my pdev->dev.of_node to an entry in the reserved_mem array (in of_reserved_mem.c). > > > > .../devicetree/bindings/soc/qcom/qcom,rfsa.txt | 43 ++++++++++++++++++++++ > > 1 file changed, 43 insertions(+) > > create mode 100644 Documentation/devicetree/bindings/soc/qcom/qcom,rfsa.txt > > > > diff --git a/Documentation/devicetree/bindings/soc/qcom/qcom,rfsa.txt b/Documentation/devicetree/bindings/soc/qcom/qcom,rfsa.txt > > new file mode 100644 > > index 000000000000..b4de0de74e46 > > --- /dev/null > > +++ b/Documentation/devicetree/bindings/soc/qcom/qcom,rfsa.txt > > @@ -0,0 +1,43 @@ > > +Qualcomm Remote File System Access binding > > + > > +This binding describes the Qualcomm RFSA, which serves the purpose of managing > > +the shared memory region used for remote processors to access block device data > > +using the Remote Filesystem protocol. > > + > > +- compatible: > > + Usage: required > > + Value type: > > + Definition: must be: > > + "qcom,rfsa" > > No versioning? > The binding is used to describe a chunk of memory and an associated client-id of this memory, so I'm not sure if we need a version. > > + > > +- memory-region: > > + Usage: required > > + Value type: > > + Definition: handle to memory reservation the associated rfsa region. > > + > > +- qcom,client-id: > > + Usage: required > > + Value type: > > + Definition: identifier of the client to use this region for buffers. > > What determines these numbers? > The bigger picture of this is that the remote processors (e.g. modem) needs access to block storage, so it sends request to a service on the system with storage access (i.e the application processor) which will read and write from the file system, storing blocks of data in the rfsa-memory. So the client-id is the hard coded identifier of each such remote system requesting IO - each one with its own memory carveout. In later platforms we need to configure the SMMU for each remote in order to give them access to these cerveouts, so I don't see that its possible to mash them together in one chunk and handle the client-id thing only in the application. > > + > > += EXAMPLE > > +The following example shows the RFSA setup for APQ8016, with the RFSA region > > +for the Hexagon DSP (id #1) located at 0x86700000. > > + > > + reserved-memory { > > + #address-cells = <2>; > > + #size-cells = <2>; > > + ranges; > > + > > + rmtfs: rmtfs@86700000 { > > I think you should have a compatible here even with the 2 node approach. > I'm hoping we can figure out a way to fix the Linux implementation so that we can describe it fully in one node. > > + reg = <0x0 0x86700000 0x0 0xe0000>; > > + no-map; > > + }; > > + }; > > + > > + hexagon-rfsa { > > + compatible = "qcom,rfsa"; > > + memory-region = <&rmtfs>; > > + > > + qcom,client-id = <1>; > > + }; > > -- > > 2.12.0 > > Regards, Bjorn