Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753185AbdHIKI6 (ORCPT ); Wed, 9 Aug 2017 06:08:58 -0400 Received: from foss.arm.com ([217.140.101.70]:45284 "EHLO foss.arm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753016AbdHIKI4 (ORCPT ); Wed, 9 Aug 2017 06:08:56 -0400 Cc: Sudeep Holla , ALKML , LKML , DTML , Roy Franz , Harb Abdulhamid , Nishanth Menon , Arnd Bergmann , Loc Ho , Alexey Klimov , Ryan Harkin , Jassi Brar , "Rafael J. Wysocki" , linux-pm@vger.kernel.org Subject: Re: [PATCH v2 18/18] cpufreq: scmi: add support for fast frequency switching To: Viresh Kumar References: <1501857104-11279-1-git-send-email-sudeep.holla@arm.com> <1501857104-11279-19-git-send-email-sudeep.holla@arm.com> <20170809042825.GI28857@vireshk-i7> From: Sudeep Holla Organization: ARM Message-ID: <0f89f543-39a9-30a5-4976-0c447f5fe843@arm.com> Date: Wed, 9 Aug 2017 11:09:02 +0100 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.2.1 MIME-Version: 1.0 In-Reply-To: <20170809042825.GI28857@vireshk-i7> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3372 Lines: 98 On 09/08/17 05:28, Viresh Kumar wrote: > On 04-08-17, 15:31, Sudeep Holla wrote: >> The cpufreq core provides option for drivers to implement fast_switch >> callback which is invoked for frequency switching from interrupt context. >> >> This patch adds support for fast_switch callback in SCMI cpufreq driver >> by making use of polling based SCMI transfer. It also sets the flag >> fast_switch_possible. >> >> Cc: "Rafael J. Wysocki" >> Cc: Viresh Kumar >> Cc: linux-pm@vger.kernel.org >> Signed-off-by: Sudeep Holla >> --- >> drivers/cpufreq/scmi-cpufreq.c | 16 ++++++++++++++++ >> 1 file changed, 16 insertions(+) >> >> diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c >> index 034359cafea5..cb1084cb1ef1 100644 >> --- a/drivers/cpufreq/scmi-cpufreq.c >> +++ b/drivers/cpufreq/scmi-cpufreq.c >> @@ -61,6 +61,19 @@ scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index) >> return perf_ops->freq_set(priv->handle, priv->domain_id, freq, false); >> } >> >> +static unsigned int scmi_cpufreq_fast_switch(struct cpufreq_policy *policy, >> + unsigned int target_freq) >> +{ >> + struct scmi_data *priv = policy->driver_data; >> + struct scmi_perf_ops *perf_ops = priv->handle->perf_ops; >> + >> + if (!perf_ops->freq_set(priv->handle, priv->domain_id, >> + target_freq * 1000, true)) >> + return target_freq; >> + >> + return CPUFREQ_ENTRY_INVALID; >> +} > > This is very much similar to the target routine, perhaps we should write another > local routine which is used by both target and fast switch. > Just one difference, fast switch uses polling based mailbox while target_index uses interrupt based. I thought initially to reuse, but it's comes done to just perf_ops->freq_set, so dropped that idea. > Do we guarantee that the frequency is changed by the time this routine returns? No, firmware may return acknowledging the request not it's completion. > Or we just send a SCMI request and return back ? > Yes, exactly. > If we just send the request and don't wait for freq to get changed, what > protects another scmi_cpufreq_fast_switch() to get called before the first one > is finished? And what will happen on that call ? Firmware needs to serialize or override based on the timing of the two consecutive requests. > >> static int >> scmi_get_sharing_cpus(struct device *cpu_dev, struct cpumask *cpumask) >> { >> @@ -164,6 +177,7 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy) >> >> policy->cpuinfo.transition_latency = latency; >> >> + policy->fast_switch_possible = true; >> return 0; >> >> out_free_cpufreq_table: >> @@ -180,6 +194,7 @@ static int scmi_cpufreq_exit(struct cpufreq_policy *policy) >> { >> struct scmi_data *priv = policy->driver_data; >> >> + policy->fast_switch_possible = false; >> cpufreq_cooling_unregister(priv->cdev); >> dev_pm_opp_free_cpufreq_table(priv->cpu_dev, &policy->freq_table); >> dev_pm_opp_cpumask_remove_table(policy->related_cpus); >> @@ -228,6 +243,7 @@ static struct cpufreq_driver scmi_cpufreq_driver = { >> .init = scmi_cpufreq_init, >> .exit = scmi_cpufreq_exit, >> .ready = scmi_cpufreq_ready, >> + .fast_switch = scmi_cpufreq_fast_switch, > > Maybe add it right after target_index ? > Done -- Regards, Sudeep