Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753905AbcCVRWJ (ORCPT ); Tue, 22 Mar 2016 13:22:09 -0400 Received: from mail-pf0-f175.google.com ([209.85.192.175]:36093 "EHLO mail-pf0-f175.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751493AbcCVRV5 (ORCPT ); Tue, 22 Mar 2016 13:21:57 -0400 Subject: Re: [PATCH v5 3/6] ACPI: parse SPCR and enable matching console To: Aleksey Makarov , Andy Shevchenko References: <1458643595-14719-1-git-send-email-aleksey.makarov@linaro.org> <1458643595-14719-4-git-send-email-aleksey.makarov@linaro.org> <56F16DC2.7030901@hurleysoftware.com> <56F17B0B.306@linaro.org> Cc: Greg Kroah-Hartman , "Rafael J . Wysocki" , Len Brown , "linux-serial@vger.kernel.org" , "linux-acpi@vger.kernel.org" , "linux-kernel@vger.kernel.org" , linux-arm Mailing List , Russell King , Leif Lindholm , Graeme Gregory , Al Stone , Christopher Covington , Yury Norov , "Zheng, Lv" , Jiri Slaby From: Peter Hurley Message-ID: <56F17F2F.7050305@hurleysoftware.com> Date: Tue, 22 Mar 2016 10:21:51 -0700 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.6.0 MIME-Version: 1.0 In-Reply-To: <56F17B0B.306@linaro.org> Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2451 Lines: 74 On 03/22/2016 10:04 AM, Aleksey Makarov wrote: > > > On 03/22/2016 07:07 PM, Peter Hurley wrote: >> On 03/22/2016 04:09 AM, Andy Shevchenko wrote: >>> On Tue, Mar 22, 2016 at 12:46 PM, Aleksey Makarov >>> wrote: > >>>> + sprintf(opts, "%s,%s,0x%llx,%d", uart, iotype, >>>> + table->serial_port.address, baud_rate); >>> >>> You may use snprintf(), though my question here what would happen on >>> 32-bit kernel when you supply 64-bit address as an option? >> >> Yeah this should probably use %pa for the printf specifier. >> >> But note this exposes underlying bug in the earlycon support, because >> that was originally written without 32/64-mixed bitness in mind; ie., >> the address is parsed and handled as unsigned long in most places. > > I don't quite follow this. table->serial_port.address is explicitly u64, > not pointer, so, according to printk-formats.txt %llx is ok here, %pa is wrong. > Am I missing something? No, you're right. There is still the underlying bug I noted; I'll fix that in this release cycle. >>>> /* >>>> - * Just 'earlycon' is a valid param for devicetree earlycons; >>>> - * don't generate a warning from parse_early_params() in that case >>>> + * Just 'earlycon' is a valid param for devicetree and ACPI SPCR >>>> + * earlycons; don't generate a warning from parse_early_params() >>>> + * in that case >>>> */ >>>> - if (!buf || !buf[0]) >>>> - return early_init_dt_scan_chosen_serial(); >>>> + if (!buf || !buf[0]) { >>>> + init_spcr_earlycon(); >>> >>>> + early_init_dt_scan_chosen_serial(); >>>> + return 0; >>> >>> And you hide an error? >> >> Well, this is a little bit tricky because "earlycon" early parameter with >> missing /chosen/stdout-path node is no longer an error, since ACPI may be >> specifying the earlycon instead. > > Agree, but note the email by Rob Herring. The code should be like this: > > if (!buf || !buf[0]) { > if (acpi_disabled) { > return early_init_dt_scan_chosen_serial(); > } else { > init_spcr_earlycon(); > return 0; > } > } Ok. > But that requires to have made ACPI/DT decision at this point. It's an unfortunate command-line option dependency, but I think it's ok for the moment. The same problem would exist if DT could be disabled via the command line. > > Thank you > Aleksey >