From mboxrd@z Thu Jan 1 00:00:00 1970 From: Hongbo Zhang Subject: Re: [RFC PATCH] thermal: add generic cpu hotplug cooling device Date: Mon, 23 Sep 2013 19:17:31 +0800 Message-ID: <5240234B.4070800@freescale.com> References: <1379715336-22620-1-git-send-email-zoran.markovic@linaro.org> Mime-Version: 1.0 Content-Type: text/plain; charset="ISO-8859-1"; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1379715336-22620-1-git-send-email-zoran.markovic@linaro.org> Sender: linux-doc-owner@vger.kernel.org To: Zoran Markovic Cc: linux-kernel@vger.kernel.org, linux-doc@vger.kernel.org, linux-pm@vger.kernel.org, Zhang Rui , Eduardo Valentin , Rob Landley , Amit Daniel Kachhap , Andrew Morton , Durgadoss R , Christian Daudt , James King List-Id: linux-pm@vger.kernel.org On 09/21/2013 06:15 AM, Zoran Markovic wrote: > This patch implements a generic CPU hotplug cooling device. The > implementation scales down the number of running CPUs when temperature > increases through a thermal trip point and prevents booting CPUs > until thermal conditions are restored. Upon restoration, the action > of starting up a CPU is left to another entity (e.g. CPU offline > governor, for which a patch is in the works). > > In the past two years, ARM considerably reduced the time required for > CPUs to boot and shutdown; this time is now measured in microseconds. > This patch is predominantly intended for ARM big.LITTLE architectures > where big cores are expected to have a much bigger impact on thermal > budget than little cores, resulting in fast temperature ramps to a trip > point, i.e. thermal runaways. Switching off the big core(s) may be one > of the recovery mechanisms to restore system temperature, but the actual > strategy is left to the thermal governor. > > The assumption is that CPU shutdown/startup is a rare event, so no > attempt was made to make the code atomic, i.e. the code evidently races > with CPU hotplug driver. The set_cur_state() function offlines CPUs > iteratively one at a time, checking the cooling state before each CPU > shutdown. A hotplug notifier callback validates any CPU boot requests > against current cooling state and approves/denies accordingly. This > mechanism guarantees that the desired cooling state could be reached in a > maximum of d-c iterations, where d and c are the "desired" and "current" > cooling states expressed in the number of offline CPUs. > > Credits to Amit Daniel Kachhap for initial attempt to upstream this feature. > > Cc: Zhang Rui > Cc: Eduardo Valentin > Cc: Rob Landley > Cc: Amit Daniel Kachhap > Cc: Andrew Morton > Cc: Durgadoss R > Cc: Christian Daudt > Cc: James King > Signed-off-by: Zoran Markovic > --- > Documentation/thermal/cpu-cooling-api.txt | 17 ++ > drivers/thermal/Kconfig | 10 + > drivers/thermal/Makefile | 1 + > drivers/thermal/cpu_hotplug.c | 362 +++++++++++++++++++++++++++++ > include/linux/cpuhp_cooling.h | 57 +++++ > 5 files changed, 447 insertions(+) > create mode 100644 drivers/thermal/cpu_hotplug.c > create mode 100644 include/linux/cpuhp_cooling.h Only form my point of view: I like the name of cpu_hotplug_cooling.c and cpu_hotplug_cooling.h we already have a cpu_cooling.c, that isn't so exact either because we have more than one method to cool a CPU, these c and h files should be renamed to cpu_freq_cooling.c and cpu_freq_cooling.h later. By the way, some servers with tens of CPUs may need this patch too.