From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx1.redhat.com (mx1.redhat.com [209.132.183.28]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ml01.01.org (Postfix) with ESMTPS id 0C66221BC6A24 for ; Fri, 31 Mar 2017 02:47:02 -0700 (PDT) Received: from smtp.corp.redhat.com (int-mx01.intmail.prod.int.phx2.redhat.com [10.5.11.11]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 6968D72875; Fri, 31 Mar 2017 09:47:01 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mx1.redhat.com 6968D72875 Authentication-Results: ext-mx09.extmail.prod.ext.phx2.redhat.com; dmarc=none (p=none dis=none) header.from=redhat.com Authentication-Results: ext-mx09.extmail.prod.ext.phx2.redhat.com; spf=pass smtp.mailfrom=lersek@redhat.com DKIM-Filter: OpenDKIM Filter v2.11.0 mx1.redhat.com 6968D72875 Received: from lacos-laptop-7.usersys.redhat.com (ovpn-116-31.phx2.redhat.com [10.3.116.31]) by smtp.corp.redhat.com (Postfix) with ESMTP id 2687718A6F; Fri, 31 Mar 2017 09:46:59 +0000 (UTC) To: Ard Biesheuvel References: <20170329134833.12956-1-ard.biesheuvel@linaro.org> <54d3f0bc-47e5-760d-8222-ad5325b01722@redhat.com> Cc: "edk2-devel@lists.01.org" , Leif Lindholm , Ryan Harkin From: Laszlo Ersek Message-ID: Date: Fri, 31 Mar 2017 11:46:58 +0200 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.8.0 MIME-Version: 1.0 In-Reply-To: X-Scanned-By: MIMEDefang 2.79 on 10.5.11.11 X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.38]); Fri, 31 Mar 2017 09:47:01 +0000 (UTC) Subject: Re: [PATCH 0/6] Embedded|ArmPlatformPkg: spring cleaning + DtPlatformDxe switch X-BeenThere: edk2-devel@lists.01.org X-Mailman-Version: 2.1.22 Precedence: list List-Id: EDK II Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 31 Mar 2017 09:47:02 -0000 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit On 03/31/17 11:22, Ard Biesheuvel wrote: > On 30 March 2017 at 19:28, Laszlo Ersek wrote: > [...] >> >> What do you think of the following proposal instead (I have no strong >> feelings about this, just picking your mind): >> >> (1) In commit 779cc439e881 ("EmbeddedPkg: add DT platform driver to >> select between DT and ACPI", 2017-03-27), a driver was added that, based >> on an HII checkbox, >> >> - either produces the platform-has-ACPI NULL protocol, >> >> - or immediately installs the DTB, found in any FV section. >> >> (Glossing over the details here, such as, if there is no DTB embedded in >> the firmware image, we always go with ACPI etc.) >> >> I reviewed that patch. I slightly disliked that in the DT case, we >> immediately installed the DT as a sysconfig table, but I figured, given >> that this driver actually "owned" the DTB, it was okay. Therefore, I >> didn't suggest producing the platform-has-DeviceTree NULL protocol, and >> then acting upon that protocol within the exact same driver (i.e., to >> install the DTB as a sysconfig table in a protocol notify). >> >> (2) I feel that, with this set, the DTB ownership is changing. I think >> the following restructuring would be an improvement: >> >> - DtPlatformDxe should only concern itself with translating the HII >> checkbox to the appropriate NULL protocol, namely platform-has-ACPI >> versus platform-has-DeviceTree. A corresponding rename for the driver >> might be in order too. >> >> - The owner of the (multiple possible) DTBs is now ArmVExpressDxe (or >> any similar driver included in any given platform DSC). IMO, this is the >> driver to look up the DTB within the firmware image, based on the >> platform type determination that it already performs. Then, >> ArmVExpressDxe should install the selected DTB as a sysconfig table, >> specifically in a protocol notify callback for platform-has-DeviceTree. >> > > Well, what I would like to avoid is having awareness of DT in more > places than necessary. The nice thing of having a DtPlatformDxe driver > that takes care of it all is that removing the module (and the > PlatformHasAcpi NULL library class resolution) makes a platform > completely ACPI only But ArmVExpressDxe remains DT-aware anyway, because it selects the right DT to install, no? In this patch set, patch #6 does not set an abstract "platform identifier" PCD, what it sets is called PcdDtPlatformDefaultDtbSectionIndex which has "DTB" in the name. And, if you remove the DtPlatformDxe driver only (but not ArmVExpressDxe), for permanent ACPI adoption, then PcdDtPlatformDefaultDtbSectionIndex will become useless (and maybe ArmVExpressDxe will become wholly useless too). So I think you'd always remove more than just DtPlatformDxe and the PlatformHasAcpi plugin lib. > >> (3) Here's an excerpt from the message of commit 65a69b214840, >> "EmbeddedPkg: introduce EDKII Platform Has Device Tree GUID", 2017-03-17: >> >>> In the DXE phase, the protocol is meant to be consumed by the platform >>> driver that >>> - owns the Device Tree description of the hardware, and >>> - is responsible for installing it as a system configuration table. >> >> I think that the above description matches the current situation 1:1, >> and that the proposed driver structuring would keep the responsibilities >> better separated. >> >> Plus, you could eliminate the butt-ugly BEFORE depex :) >> >> The "EmbeddedPkg/Drivers/DtPlatformDxe" driver is not yet used in any >> upstream edk2 platform DSC, so I think it's the appropriate time to set >> the responsibilities right. I haven't looked at your series >> >> [Linaro-uefi] [PATCH 0/6] EDK2 spring cleaning -- OpenPlatformPkg >> edition >> https://lists.linaro.org/pipermail/linaro-uefi/2017-March/004196.html >> >> yet, but there you call both patch sets inter-dependent, so I guess this >> is the one we should be discussing first. >> >> (4) I'm missing a whole lot of details about ArmVExpressDxe, so the >> above might not be feasible or desirable. For example, while it seems to >> identify and expose the DTB to install -- very elaborately, as you say >> --, ultimately it only sets PcdFdtDevicePaths. >> >> The PCD is then consumed by "EmbeddedPkg/Drivers/FdtPlatformDxe". As far >> as I can see, this driver reads the DTB from the EFI system partition, >> as directed by the PCD? >> >> In this patch set, you don't seem to touch FdtPlatformDxe, so I think >> that you are replacing all of the FdtPlatformDxe functionality by >> embedding the DTBs in the FV image. >> >> In that case, my proposal above should not conflict with (or require >> updates for) FdtPlatformDxe either. >> >> (5) Anyway, I just wanted to float the idea. What do you think of it? >> > > I want to get rid of FdtPlatformDxe. It uses things like semihosting > device paths to pull .dtb files that are known by name from various > places. While this is nice for debugging in theory, nobody actually > uses it. On it does not belong on a production system at all. Also, > for a laugh, please check out > 7aec2926b926ad90d09fb026af0ee04c4c831237, specifically the hunk that > adds gEmbeddedTokenSpaceGuid.PcdFdtDevicePaths. Right. My suggestion seems compatible with killing FdtPlatformDxe for good though. > I agree about the butt-uglines of BEFORE depexes, so what I would > prefer is an alternative way to allow DtPlatformDxe to 'own' DT > installation entirely. What do you have in mind? The below, or something in addition to that? > To enforce the ordering, I could also update > DtPlatformDxe to load the correct FV section at EndOfDxe. Would that > be acceptable to you? I think it remains a bit ugly that DT selection and DT installation happen in two different drivers. Again, if you rip DtPlatformDxe out of a platform at some point -- because the platform becomes ACPI only -- there won't be any reason to preserve the other driver that does the DT selection (or minimally, the DT-selection logic in the other driver). This seems to imply that DT selection and DT installation belong in the same place. Also, you know my opinion about doing stuff in "late" callbacks :) Nonetheless, I think an EndOfDxe callback would be a significant improvement over the BEFORE depex. So, feel free to pick whichever method you deem best, I'm ready to ack either; I just wanted to raise some ideas. I don't intend to obsess about them forever :) Thanks! Laszlo