Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932314AbbGUBOh (ORCPT ); Mon, 20 Jul 2015 21:14:37 -0400 Received: from mail-wg0-f43.google.com ([74.125.82.43]:34218 "EHLO mail-wg0-f43.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755995AbbGUBOf (ORCPT ); Mon, 20 Jul 2015 21:14:35 -0400 MIME-Version: 1.0 In-Reply-To: References: <20150718163149.GP7557@n2100.arm.linux.org.uk> <20150720103623.GQ7557@n2100.arm.linux.org.uk> Date: Tue, 21 Jul 2015 03:14:34 +0200 X-Google-Sender-Auth: i7Wkr55NIwuEDrGtrAZieKwDdl0 Message-ID: Subject: Re: [PATCH] cpufreq: Avoid double addition/removal of sysfs links From: "Rafael J. Wysocki" To: Russell King - ARM Linux , Viresh Kumar Cc: Rafael Wysocki , Lists linaro-kernel , "linux-pm@vger.kernel.org" , open list Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5400 Lines: 106 On Tue, Jul 21, 2015 at 2:47 AM, Rafael J. Wysocki wrote: > Hi Russell, > > On Mon, Jul 20, 2015 at 12:36 PM, Russell King - ARM Linux > wrote: >> On Mon, Jul 20, 2015 at 03:17:10PM +0530, Viresh Kumar wrote: >>> Consider a dual core (0/1) system with two CPUs: >>> - sharing clock/voltage rails and hence cpufreq-policy >>> - CPU1 is offline while the cpufreq driver is registered >>> >>> - cpufreq_add_dev() is called from subsys callback for CPU0 and we >>> create the policy for the CPUs and create link for CPU1. >>> - cpufreq_add_dev() is called from subsys callback for CPU1, we find >>> that the cpu is offline and we try to create a sysfs link for CPU1. >>> - This results in double addition of the sysfs link and we get this: >>> >>> WARNING: CPU: 0 PID: 1 at fs/sysfs/dir.c:31 sysfs_warn_dup+0x60/0x7c() >>> sysfs: cannot create duplicate filename '/devices/system/cpu/cpu1/cpufreq' >>> Modules linked in: >>> CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.2.0-rc2+ #1704 >>> Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree) >>> Backtrace: >>> [] (dump_backtrace) from [] (show_stack+0x18/0x1c) >>> r6:c01a1f30 r5:0000001f r4:00000000 r3:00000000 >>> [] (show_stack) from [] (dump_stack+0x7c/0x98) >>> [] (dump_stack) from [] (warn_slowpath_common+0x80/0xbc) >>> r4:d74abbd0 r3:d74c0000 >>> [] (warn_slowpath_common) from [] (warn_slowpath_fmt+0x38/0x40) >>> r8:ffffffef r7:00000000 r6:d75a8960 r5:c0993280 r4:d6b4d000 >>> [] (warn_slowpath_fmt) from [] (sysfs_warn_dup+0x60/0x7c) >>> r3:d6b4dfe7 r2:c0930750 >>> [] (sysfs_warn_dup) from [] (sysfs_do_create_link_sd+0xb8/0xc0) >>> r6:d75a8960 r5:c0993280 r4:d00aba20 >>> [] (sysfs_do_create_link_sd) from [] (sysfs_create_link+0x2c/0x3c) >>> r10:00000001 r8:c14db3c8 r7:d7b89010 r6:c0ae7c60 r5:d7b89010 r4:d00d1200 >>> [] (sysfs_create_link) from [] (add_cpu_dev_symlink+0x34/0x5c) >>> [] (add_cpu_dev_symlink) from [] (cpufreq_add_dev+0x674/0x794) >>> r5:00000001 r4:00000000 >>> [] (cpufreq_add_dev) from [] (subsys_interface_register+0x8c/0xd0) >>> r10:00000003 r9:d7bb01f0 r8:c14db3c8 r7:00106738 r6:c0ae7c60 r5:c0acbd08 >>> r4:c0ae7e20 >>> [] (subsys_interface_register) from [] (cpufreq_register_driver+0x104/0x1f4) >>> >>> >>> The check for offline-cpu in cpufreq_add_dev() is present to ensure that >>> link gets added for the CPUs, that weren't physically present earlier >>> and we missed the case where a CPU is offline while registering the >>> driver. >>> >>> Fix this by keeping track of CPUs for which link is already created, and >>> avoiding duplicate sysfs entries. >> >> Why do we try to create the symlink for CPU devices which we haven't >> "detected" yet (iow, we haven't had cpufreq_add_dev() called for)? >> Surely we are guaranteed to have cpufreq_add_dev() called for every >> CPU which exists in sysfs? So why not _only_ create the sysfs symlinks >> when cpufreq_add_dev() is notified that a CPU subsys interface is >> present? > > That's something I've overlooked. > > Yes, we should be doing exactly that. > >> Sure, if the policy changes, we need to do maintanence on these symlinks, >> but I see only one path down into cpufreq_add_dev_symlink(), which is: >> >> cpufreq_add_dev() -> cpufreq_add_dev_interface() -> >> cpufreq_add_dev_symlink() >> >> In other words, only when we see a new CPU interface appears, not when >> the policy changes. If the set of related CPUs is policy independent, >> why is this information carried in the cpufreq_policy struct? > > It is not policy-dependent, but the way that information is gathered > is not exactly straightforward. It generally depends on what the > platform firmware tells us about the topology. > >> If it is policy dependent, then I see no code which handles the effect >> of a policy change where the policy->related_cpus is different. To me, >> that sounds like a rather huge design hole. >> >> Things get worse. Reading drivers/base/cpu.c, CPU interface nodes are >> only ever created - they're created for the set of _possible_ CPUs in >> the system, not those which are possible and present, and there is no >> unregister_cpu() API, only a register_cpu() API. > > There is unregister_cpu() API too, but it is called from > arch_unregister_cpu(). And it calls device_unregister() and all of > the appropriate things happen AFAICS. Eventually, > cpufreq_remove_dev() is called from that path. That said, cpu_present_mask may only be updated after calling arch_unregister_cpu(), so checking it in cpufreq_remove_dev() doesn't really help. It looks like using cpufreq_remove_dev() as the subsys ->remove_dev callback is a mistake as it cannot really tell the difference between that code path and the CPU offline one. Thanks, Rafael -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/