Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756935AbZFYW0Y (ORCPT ); Thu, 25 Jun 2009 18:26:24 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1756875AbZFYW0G (ORCPT ); Thu, 25 Jun 2009 18:26:06 -0400 Received: from cantor.suse.de ([195.135.220.2]:42690 "EHLO mx1.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754713AbZFYW0D (ORCPT ); Thu, 25 Jun 2009 18:26:03 -0400 From: Thomas Renninger To: Mathieu Desnoyers Subject: Re: [PATCH 1/2] CPUFREQ: Remove unneeded dbs_mutexes from ondemand and conservative governors Date: Fri, 26 Jun 2009 00:26:20 +0200 User-Agent: KMail/1.9.10 Cc: kernel@stable.org, cpufreq@vger.kernel.org, linux-kernel@vger.kernel.org, mingo@elte.hu, rjw@sisk.pl, hidave.darkstar@gmail.com, penberg@cs.helsinki.fi, kernel-testers@vger.kernel.org, davej@redhat.com, Venkatesh Pallipadi References: <20090623193215.GA31374@elte.hu> <20090625142552.GA18000@Krystal> <200906260017.10730.trenn@suse.de> In-Reply-To: <200906260017.10730.trenn@suse.de> MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200906260026.21394.trenn@suse.de> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 13918 Lines: 406 On Friday 26 June 2009 12:17:09 am Thomas Renninger wrote: > On Thursday 25 June 2009 04:25:52 pm Mathieu Desnoyers wrote: > > * Thomas Renninger (trenn@suse.de) wrote: > > > Comment from Venkatesh: > > > ... > > > This mutex is just serializing the changes to those variables. I > > > could't think of any functionality issues of not having the lock as > > > such. > > > > > > -> rip it out. > > > > > > CC: Venkatesh Pallipadi > > > Signed-off-by: Thomas Renninger > > > --- > > > drivers/cpufreq/cpufreq_conservative.c | 61 > > > +++----------------------------- drivers/cpufreq/cpufreq_ondemand.c > > > | 48 +++---------------------- 2 files changed, 10 insertions(+), 99 > > > deletions(-) > > > > > > diff --git a/drivers/cpufreq/cpufreq_conservative.c > > > b/drivers/cpufreq/cpufreq_conservative.c index 7a74d17..6303379 100644 > > > --- a/drivers/cpufreq/cpufreq_conservative.c > > > +++ b/drivers/cpufreq/cpufreq_conservative.c > > > @@ -18,7 +18,6 @@ > > > #include > > > #include > > > #include > > > -#include > > > #include > > > #include > > > #include > > > @@ -84,19 +83,6 @@ static DEFINE_PER_CPU(struct cpu_dbs_info_s, > > > cpu_dbs_info); > > > > > > static unsigned int dbs_enable; /* number of CPUs using this policy */ > > > > > > -/* > > > - * DEADLOCK ALERT! There is a ordering requirement between cpu_hotplug > > > - * lock and dbs_mutex. cpu_hotplug lock should always be held before > > > - * dbs_mutex. If any function that can potentially take cpu_hotplug > > > lock - * (like __cpufreq_driver_target()) is being called with > > > dbs_mutex taken, then - * cpu_hotplug lock should be taken before that. > > > Note that cpu_hotplug lock - * is recursive for the same process. > > > -Venki - * DEADLOCK ALERT! (2) : do_dbs_timer() must not take the > > > dbs_mutex, because it - * would deadlock with > > > cancel_delayed_work_sync(), which is needed for proper - * raceless > > > workqueue teardown. > > > - */ > > > -static DEFINE_MUTEX(dbs_mutex); > > > - > > > static struct workqueue_struct *kconservative_wq; > > > > > > static struct dbs_tuners { > > > @@ -236,10 +222,7 @@ static ssize_t store_sampling_down_factor(struct > > > cpufreq_policy *unused, if (ret != 1 || input > > > > MAX_SAMPLING_DOWN_FACTOR > > > > > > || input < 1) return -EINVAL; > > > > > > - mutex_lock(&dbs_mutex); > > > dbs_tuners_ins.sampling_down_factor = input; > > > - mutex_unlock(&dbs_mutex); > > > - > > > return count; > > > } > > > > > > @@ -253,10 +236,7 @@ static ssize_t store_sampling_rate(struct > > > cpufreq_policy *unused, if (ret != 1) > > > return -EINVAL; > > > > > > - mutex_lock(&dbs_mutex); > > > dbs_tuners_ins.sampling_rate = max(input, minimum_sampling_rate()); > > > - mutex_unlock(&dbs_mutex); > > > - > > > return count; > > > } > > > > > > @@ -267,16 +247,11 @@ static ssize_t store_up_threshold(struct > > > cpufreq_policy *unused, int ret; > > > ret = sscanf(buf, "%u", &input); > > > > > > - mutex_lock(&dbs_mutex); > > > if (ret != 1 || input > 100 || > > > - input <= dbs_tuners_ins.down_threshold) { > > > - mutex_unlock(&dbs_mutex); > > > + input <= dbs_tuners_ins.down_threshold) > > > return -EINVAL; > > > - } > > > > > > dbs_tuners_ins.up_threshold = input; > > > - mutex_unlock(&dbs_mutex); > > > > Here, for instance, there might be a problem if down_threshold is > > changed concurrently with a store_up_threshold() call. See that there is > > a test before the modification, and we need the mutex there for it to be > > consistent. > > Thanks, I was rather quick with the conservative changes..., but > it should still be ok. > > It should be assured that if userspace is doing: > echo x > down_threshold > echo y > up_threshold > that the first one will be served/finished first? > > If userspace is writing different values for each core to the global > conservative/ondemand tunables, or you have concurent userspace tools > trying to configure ondemand/conservative, it's a userspace bug. > It's confusing that ondemand/conservative allows per core reads/writes to > global variables and I hope to be able to provide something to change that > in some days, maybe weeks. > > If you still can think of a possible issue, a userspace scenario would > help. > > > > - > > > return count; > > > } > > > > > > @@ -287,17 +262,12 @@ static ssize_t store_down_threshold(struct > > > cpufreq_policy *unused, int ret; > > > ret = sscanf(buf, "%u", &input); > > > > > > - mutex_lock(&dbs_mutex); > > > /* cannot be lower than 11 otherwise freq will not fall */ > > > if (ret != 1 || input < 11 || input > 100 || > > > - input >= dbs_tuners_ins.up_threshold) { > > > - mutex_unlock(&dbs_mutex); > > > + input >= dbs_tuners_ins.up_threshold) > > > return -EINVAL; > > > - } > > > > > > dbs_tuners_ins.down_threshold = input; > > > - mutex_unlock(&dbs_mutex); > > > - > > > return count; > > > } > > > > > > @@ -316,11 +286,9 @@ static ssize_t store_ignore_nice_load(struct > > > cpufreq_policy *policy, if (input > 1) > > > input = 1; > > > > > > - mutex_lock(&dbs_mutex); > > > - if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */ > > > - mutex_unlock(&dbs_mutex); > > > + if (input == dbs_tuners_ins.ignore_nice) /* nothing to do */ > > > return count; > > > - } > > > + > > > dbs_tuners_ins.ignore_nice = input; > > > > > > /* we need to re-evaluate prev_cpu_idle */ > > > @@ -332,8 +300,6 @@ static ssize_t store_ignore_nice_load(struct > > > cpufreq_policy *policy, if (dbs_tuners_ins.ignore_nice) > > > dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; > > > } > > > - mutex_unlock(&dbs_mutex); > > > - > > > return count; > > > } > > > > > > @@ -352,10 +318,7 @@ static ssize_t store_freq_step(struct > > > cpufreq_policy *policy, > > > > > > /* no need to test here if freq_step is zero as the user might > > > actually * want this, they would be crazy though :) */ > > > - mutex_lock(&dbs_mutex); > > > dbs_tuners_ins.freq_step = input; > > > - mutex_unlock(&dbs_mutex); > > > - > > > return count; > > > } > > > > > > @@ -566,13 +529,9 @@ static int cpufreq_governor_dbs(struct > > > cpufreq_policy *policy, > > > > Hrm, this is where we want the mutexes removed, but I fear this is > > creating a race between sysfs_create_group (sysfs file creation) and > > policy initialization... > > This can be solved by moving this_dbs_info->enable incremenation > after sysfs_create_group. Forget this sentence, don't think about it, it's crap. I better go to bed now... Thomas > But yes, I forgot that in my patch, thanks! > > > I'm not convinced this mutex is not needed. > > I am. Maybe it still takes some more thinking or step by step rework. > Finding an unintrusive, riskless short term solution for .30 is a > challenge, though. > > Thomas > > > Mathieu > > > > > if (this_dbs_info->enable) /* Already enabled */ > > > break; > > > > > > - mutex_lock(&dbs_mutex); > > > - > > > rc = sysfs_create_group(&policy->kobj, &dbs_attr_group); > > > - if (rc) { > > > - mutex_unlock(&dbs_mutex); > > > + if (rc) > > > return rc; > > > - } > > > > > > for_each_cpu(j, policy->cpus) { > > > struct cpu_dbs_info_s *j_dbs_info; > > > @@ -612,13 +571,9 @@ static int cpufreq_governor_dbs(struct > > > cpufreq_policy *policy, CPUFREQ_TRANSITION_NOTIFIER); > > > } > > > dbs_timer_init(this_dbs_info); > > > - > > > - mutex_unlock(&dbs_mutex); > > > - > > > break; > > > > > > case CPUFREQ_GOV_STOP: > > > - mutex_lock(&dbs_mutex); > > > dbs_timer_exit(this_dbs_info); > > > sysfs_remove_group(&policy->kobj, &dbs_attr_group); > > > dbs_enable--; > > > @@ -631,13 +586,9 @@ static int cpufreq_governor_dbs(struct > > > cpufreq_policy *policy, cpufreq_unregister_notifier( > > > &dbs_cpufreq_notifier_block, > > > CPUFREQ_TRANSITION_NOTIFIER); > > > - > > > - mutex_unlock(&dbs_mutex); > > > - > > > break; > > > > > > case CPUFREQ_GOV_LIMITS: > > > - mutex_lock(&dbs_mutex); > > > if (policy->max < this_dbs_info->cur_policy->cur) > > > __cpufreq_driver_target( > > > this_dbs_info->cur_policy, > > > @@ -646,8 +597,6 @@ static int cpufreq_governor_dbs(struct > > > cpufreq_policy *policy, __cpufreq_driver_target( > > > this_dbs_info->cur_policy, > > > policy->min, CPUFREQ_RELATION_L); > > > - mutex_unlock(&dbs_mutex); > > > - > > > break; > > > } > > > return 0; > > > diff --git a/drivers/cpufreq/cpufreq_ondemand.c > > > b/drivers/cpufreq/cpufreq_ondemand.c index e741c33..d080a48 100644 > > > --- a/drivers/cpufreq/cpufreq_ondemand.c > > > +++ b/drivers/cpufreq/cpufreq_ondemand.c > > > @@ -17,7 +17,6 @@ > > > #include > > > #include > > > #include > > > -#include > > > #include > > > #include > > > #include > > > @@ -91,19 +90,6 @@ static DEFINE_PER_CPU(struct cpu_dbs_info_s, > > > cpu_dbs_info); > > > > > > static unsigned int dbs_enable; /* number of CPUs using this policy */ > > > > > > -/* > > > - * DEADLOCK ALERT! There is a ordering requirement between cpu_hotplug > > > - * lock and dbs_mutex. cpu_hotplug lock should always be held before > > > - * dbs_mutex. If any function that can potentially take cpu_hotplug > > > lock - * (like __cpufreq_driver_target()) is being called with > > > dbs_mutex taken, then - * cpu_hotplug lock should be taken before that. > > > Note that cpu_hotplug lock - * is recursive for the same process. > > > -Venki - * DEADLOCK ALERT! (2) : do_dbs_timer() must not take the > > > dbs_mutex, because it - * would deadlock with > > > cancel_delayed_work_sync(), which is needed for proper - * raceless > > > workqueue teardown. > > > - */ > > > -static DEFINE_MUTEX(dbs_mutex); > > > - > > > static struct workqueue_struct *kondemand_wq; > > > > > > static struct dbs_tuners { > > > @@ -269,14 +255,10 @@ static ssize_t store_sampling_rate(struct > > > cpufreq_policy *unused, int ret; > > > ret = sscanf(buf, "%u", &input); > > > > > > - mutex_lock(&dbs_mutex); > > > - if (ret != 1) { > > > - mutex_unlock(&dbs_mutex); > > > + if (ret != 1) > > > return -EINVAL; > > > - } > > > - dbs_tuners_ins.sampling_rate = max(input, minimum_sampling_rate()); > > > - mutex_unlock(&dbs_mutex); > > > > > > + dbs_tuners_ins.sampling_rate = max(input, minimum_sampling_rate()); > > > return count; > > > } > > > > > > @@ -287,16 +269,11 @@ static ssize_t store_up_threshold(struct > > > cpufreq_policy *unused, int ret; > > > ret = sscanf(buf, "%u", &input); > > > > > > - mutex_lock(&dbs_mutex); > > > if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD || > > > - input < MIN_FREQUENCY_UP_THRESHOLD) { > > > - mutex_unlock(&dbs_mutex); > > > + input < MIN_FREQUENCY_UP_THRESHOLD) > > > return -EINVAL; > > > - } > > > > > > dbs_tuners_ins.up_threshold = input; > > > - mutex_unlock(&dbs_mutex); > > > - > > > return count; > > > } > > > > > > @@ -315,11 +292,9 @@ static ssize_t store_ignore_nice_load(struct > > > cpufreq_policy *policy, if (input > 1) > > > input = 1; > > > > > > - mutex_lock(&dbs_mutex); > > > - if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */ > > > - mutex_unlock(&dbs_mutex); > > > + if (input == dbs_tuners_ins.ignore_nice) /* nothing to do */ > > > return count; > > > - } > > > + > > > dbs_tuners_ins.ignore_nice = input; > > > > > > /* we need to re-evaluate prev_cpu_idle */ > > > @@ -332,8 +307,6 @@ static ssize_t store_ignore_nice_load(struct > > > cpufreq_policy *policy, dbs_info->prev_cpu_nice = > > > kstat_cpu(j).cpustat.nice; > > > > > > } > > > - mutex_unlock(&dbs_mutex); > > > - > > > return count; > > > } > > > > > > @@ -350,10 +323,8 @@ static ssize_t store_powersave_bias(struct > > > cpufreq_policy *unused, if (input > 1000) > > > input = 1000; > > > > > > - mutex_lock(&dbs_mutex); > > > dbs_tuners_ins.powersave_bias = input; > > > ondemand_powersave_bias_init(); > > > - mutex_unlock(&dbs_mutex); > > > > > > return count; > > > } > > > @@ -586,13 +557,11 @@ static int cpufreq_governor_dbs(struct > > > cpufreq_policy *policy, if (this_dbs_info->enable) /* Already enabled > > > */ break; > > > > > > - mutex_lock(&dbs_mutex); > > > dbs_enable++; > > > > > > rc = sysfs_create_group(&policy->kobj, &dbs_attr_group); > > > if (rc) { > > > dbs_enable--; > > > - mutex_unlock(&dbs_mutex); > > > return rc; > > > } > > > > > > @@ -627,28 +596,21 @@ static int cpufreq_governor_dbs(struct > > > cpufreq_policy *policy, dbs_tuners_ins.sampling_rate = > > > def_sampling_rate; } > > > dbs_timer_init(this_dbs_info); > > > - > > > - mutex_unlock(&dbs_mutex); > > > break; > > > > > > case CPUFREQ_GOV_STOP: > > > - mutex_lock(&dbs_mutex); > > > dbs_timer_exit(this_dbs_info); > > > sysfs_remove_group(&policy->kobj, &dbs_attr_group); > > > dbs_enable--; > > > - mutex_unlock(&dbs_mutex); > > > - > > > break; > > > > > > case CPUFREQ_GOV_LIMITS: > > > - mutex_lock(&dbs_mutex); > > > if (policy->max < this_dbs_info->cur_policy->cur) > > > __cpufreq_driver_target(this_dbs_info->cur_policy, > > > policy->max, CPUFREQ_RELATION_H); > > > else if (policy->min > this_dbs_info->cur_policy->cur) > > > __cpufreq_driver_target(this_dbs_info->cur_policy, > > > policy->min, CPUFREQ_RELATION_L); > > > - mutex_unlock(&dbs_mutex); > > > break; > > > } > > > return 0; > > > -- > > > 1.6.0.2 -- 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/