From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933233AbcKWSWv (ORCPT ); Wed, 23 Nov 2016 13:22:51 -0500 Received: from mail.kernel.org ([198.145.29.136]:58660 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933122AbcKWSWs (ORCPT ); Wed, 23 Nov 2016 13:22:48 -0500 Date: Wed, 23 Nov 2016 12:22:43 -0600 From: Bjorn Helgaas To: Tomasz Nowicki Cc: arnd@arndb.de, will.deacon@arm.com, catalin.marinas@arm.com, rafael@kernel.org, hanjun.guo@linaro.org, Lorenzo.Pieralisi@arm.com, okaya@codeaurora.org, jchandra@broadcom.com, robert.richter@caviumnetworks.com, mw@semihalf.com, Liviu.Dudau@arm.com, ddaney@caviumnetworks.com, wangyijing@huawei.com, Suravee.Suthikulpanit@amd.com, msalter@redhat.com, linux-pci@vger.kernel.org, linux-arm-kernel@lists.infradead.org, linux-acpi@vger.kernel.org, linux-kernel@vger.kernel.org, linaro-acpi@lists.linaro.org, jcm@redhat.com, andrea.gallo@linaro.org, dhdang@apm.com, jeremy.linton@arm.com, liudongdong3@huawei.com, cov@codeaurora.org Subject: Re: [PATCH V9 11/11] ARM64/PCI: Support for ACPI based PCI host controller Message-ID: <20161123182243.GF16033@bhelgaas-glaptop.roam.corp.google.com> References: <1465588519-11334-1-git-send-email-tn@semihalf.com> <1465588519-11334-12-git-send-email-tn@semihalf.com> <20161122231321.GA20246@bhelgaas-glaptop.roam.corp.google.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Nov 23, 2016 at 12:21:03PM +0100, Tomasz Nowicki wrote: > Hi Bjorn, > > On 23.11.2016 00:13, Bjorn Helgaas wrote: > >Hi Tomasz, > > > >On Fri, Jun 10, 2016 at 09:55:19PM +0200, Tomasz Nowicki wrote: > >>Implement pci_acpi_scan_root and other arch-specific call so that ARM64 > >>can start using ACPI to setup and enumerate PCI buses. > >> > >>Prior to buses enumeration the pci_acpi_scan_root() implementation looks > >>for configuration space start address (obtained through ACPI _CBA method or > >>MCFG interface). If succeed, it uses ECAM library to create new mapping. > >>Then it attaches generic ECAM ops (pci_generic_ecam_ops) which are used > >>for accessing configuration space later on. > >>... > > > >>+static struct acpi_pci_root_ops acpi_pci_root_ops = { > >>+ .release_info = pci_acpi_generic_release_info, > >>+}; > >>+ > >>+/* Interface called from ACPI code to setup PCI host controller */ > >> struct pci_bus *pci_acpi_scan_root(struct acpi_pci_root *root) > >> { > >>- /* TODO: Should be revisited when implementing PCI on ACPI */ > >>- return NULL; > >>+ int node = acpi_get_node(root->device->handle); > >>+ struct acpi_pci_generic_root_info *ri; > >>+ struct pci_bus *bus, *child; > >>+ > >>+ ri = kzalloc_node(sizeof(*ri), GFP_KERNEL, node); > >>+ if (!ri) > >>+ return NULL; > >>+ > >>+ ri->cfg = pci_acpi_setup_ecam_mapping(root); > >>+ if (!ri->cfg) { > >>+ kfree(ri); > >>+ return NULL; > >>+ } > >>+ > >>+ acpi_pci_root_ops.pci_ops = &ri->cfg->ops->pci_ops; > > > >This has already been merged, but this isn't right, is it? We're > >writing a host controller-specific pointer into the single system-wide > >acpi_pci_root_ops, then passing it on to acpi_pci_root_create(). > > > >Today, I think ri->cfg->ops->pci_ops is always &pci_generic_ecam_ops, > >from this path: > > > > ri->cfg = pci_acpi_setup_ecam_mapping > > cfg = pci_ecam_create(..., &pci_generic_ecam_ops) > > cfg = kzalloc(...) > > cfg->ops = ops # &pci_generic_ecam_ops > > > >But we're about to merge the ECAM quirks series, which will mean it > >may not be &pci_generic_ecam_ops. Even apart from the ECAM quirks, we > >should avoid this pattern of putting device-specific info in a single > >shared structure because it's too difficult to verify that it's > >correct. > > > > Well spotted. I agree, we need to fix this. How about this: > diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c > index fb439c7..31c0e1c 100644 > --- a/arch/arm64/kernel/pci.c > +++ b/arch/arm64/kernel/pci.c > @@ -152,33 +152,35 @@ static void > pci_acpi_generic_release_info(struct acpi_pci_root_info *ci) > > ri = container_of(ci, struct acpi_pci_generic_root_info, common); > pci_ecam_free(ri->cfg); > + kfree(ci->ops); > kfree(ri); > } > > -static struct acpi_pci_root_ops acpi_pci_root_ops = { > - .release_info = pci_acpi_generic_release_info, > -}; > - > /* Interface called from ACPI code to setup PCI host controller */ > struct pci_bus *pci_acpi_scan_root(struct acpi_pci_root *root) > { > int node = acpi_get_node(root->device->handle); > struct acpi_pci_generic_root_info *ri; > struct pci_bus *bus, *child; > + struct acpi_pci_root_ops *root_ops; > > ri = kzalloc_node(sizeof(*ri), GFP_KERNEL, node); > if (!ri) > return NULL; > > + root_ops = kzalloc_node(sizeof(*root_ops), GFP_KERNEL, node); > + if (!root_ops) > + return NULL; > + > ri->cfg = pci_acpi_setup_ecam_mapping(root); > if (!ri->cfg) { > kfree(ri); > + kfree(root_ops); > return NULL; > } > > - acpi_pci_root_ops.pci_ops = &ri->cfg->ops->pci_ops; > - bus = acpi_pci_root_create(root, &acpi_pci_root_ops, &ri->common, > - ri->cfg); > + root_ops->release_info = pci_acpi_generic_release_info; > + root_ops->pci_ops = &ri->cfg->ops->pci_ops; > + bus = acpi_pci_root_create(root, root_ops, &ri->common, ri->cfg); > if (!bus) > return NULL; > > Of course, this should be the part of ECAM quirks core patches. > > The other option we have is to remove "struct pci_ops *pci_ops;" > from acpi_pci_root_ops structure and pass struct pci_ops as an extra > argument to acpi_pci_root_create(). What do you think? I think your patch above is fine and avoids the need to change the x86 and ia64 code. Would you mind packaging this up with a changelog and signed-off-by? I can take care of putting it in the ECAM series. Thanks, Bjorn