linux-kernel - [Patch] idle governor: Avoid lock acquisition to read pm

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <1297300864.2645.128.camel@schen9-DESK>
Date:	Wed, 09 Feb 2011 17:21:04 -0800
From:	Tim Chen <tim.c.chen@...ux.intel.com>
To:	"Rafael J. Wysocki" <rjw@...k.pl>,
	mark gross <markgross@...gnar.org>,
	James Bottomley <James.Bottomley@...e.de>,
	David Alan Gilbert <linux@...blig.org>
Cc:	linux-kernel@...r.kernel.org, Len <len.brown@...el.com>,
	Andi Kleen <ak@...ux.intel.com>,
	Arjan van de Ven <arjan@...ux.intel.com>
Subject: [Patch] idle governor: Avoid lock acquisition to read pm_qos
 before entering idle

I noticed that before entering idle state, the menu idle governor will
look up the current pm_qos value according to the list of qos request
received.  This look up currently needs the acquisition of a lock to go
down a list of qos requests to find the qos value, slowing down the
entrance into idle state due to contention by multiple cpus to traverse
this list.  The contention is severe when there are a lot of cpus waking
and going into idle.  For example, for a simple workload that has 32
pair of processes ping ponging messages to each other, where 64 cores
cores are active in test system, I see the following profile:

-     37.82%          swapper  [kernel.kallsyms]          [k]
_raw_spin_lock_irqsave
   - _raw_spin_lock_irqsave
      - 95.65% pm_qos_request             
           menu_select                                             
           cpuidle_idle_call                      
         - cpu_idle                                                     
              99.98% start_secondary

Perhaps a better approach will be to cache the updated pm_qos value so
reading it does not require lock acquisition as in the patch below.   

Tim

Signed-off-by: Tim Chen <tim.c.chen@...ux.intel.com>
diff --git a/include/linux/pm_qos_params.h b/include/linux/pm_qos_params.h
index 77cbddb..a7d87f9 100644
--- a/include/linux/pm_qos_params.h
+++ b/include/linux/pm_qos_params.h
@@ -16,6 +16,10 @@
 #define PM_QOS_NUM_CLASSES 4
 #define PM_QOS_DEFAULT_VALUE -1
 
+#define PM_QOS_CPU_DMA_LAT_DEFAULT_VALUE	(2000 * USEC_PER_SEC)
+#define PM_QOS_NETWORK_LAT_DEFAULT_VALUE	(2000 * USEC_PER_SEC)
+#define PM_QOS_NETWORK_THROUGHPUT_DEFAULT_VALUE	0
+
 struct pm_qos_request_list {
 	struct plist_node list;
 	int pm_qos_class;
diff --git a/kernel/pm_qos_params.c b/kernel/pm_qos_params.c
index aeaa7f8..b6310d1 100644
--- a/kernel/pm_qos_params.c
+++ b/kernel/pm_qos_params.c
@@ -58,6 +58,7 @@ struct pm_qos_object {
 	struct blocking_notifier_head *notifiers;
 	struct miscdevice pm_qos_power_miscdev;
 	char *name;
+	s32 value;
 	s32 default_value;
 	enum pm_qos_type type;
 };
@@ -70,7 +71,8 @@ static struct pm_qos_object cpu_dma_pm_qos = {
 	.requests = PLIST_HEAD_INIT(cpu_dma_pm_qos.requests, pm_qos_lock),
 	.notifiers = &cpu_dma_lat_notifier,
 	.name = "cpu_dma_latency",
-	.default_value = 2000 * USEC_PER_SEC,
+	.value = PM_QOS_CPU_DMA_LAT_DEFAULT_VALUE,
+	.default_value = PM_QOS_CPU_DMA_LAT_DEFAULT_VALUE,
 	.type = PM_QOS_MIN,
 };
 
@@ -79,7 +81,8 @@ static struct pm_qos_object network_lat_pm_qos = {
 	.requests = PLIST_HEAD_INIT(network_lat_pm_qos.requests, pm_qos_lock),
 	.notifiers = &network_lat_notifier,
 	.name = "network_latency",
-	.default_value = 2000 * USEC_PER_SEC,
+	.value = PM_QOS_NETWORK_LAT_DEFAULT_VALUE,
+	.default_value = PM_QOS_NETWORK_LAT_DEFAULT_VALUE,
 	.type = PM_QOS_MIN
 };
 
@@ -89,7 +92,8 @@ static struct pm_qos_object network_throughput_pm_qos = {
 	.requests = PLIST_HEAD_INIT(network_throughput_pm_qos.requests, pm_qos_lock),
 	.notifiers = &network_throughput_notifier,
 	.name = "network_throughput",
-	.default_value = 0,
+	.value = PM_QOS_NETWORK_THROUGHPUT_DEFAULT_VALUE,
+	.default_value = PM_QOS_NETWORK_THROUGHPUT_DEFAULT_VALUE,
 	.type = PM_QOS_MAX,
 };
 
@@ -132,6 +136,16 @@ static inline int pm_qos_get_value(struct pm_qos_object *o)
 	}
 }
 
+static inline s32 pm_qos_read_value(struct pm_qos_object *o)
+{
+	return o->value;
+}
+
+static inline int pm_qos_set_value(struct pm_qos_object *o, s32 value)
+{
+	o->value = value;
+}
+
 static void update_target(struct pm_qos_object *o, struct plist_node *node,
 			  int del, int value)
 {
@@ -156,6 +170,7 @@ static void update_target(struct pm_qos_object *o, struct plist_node *node,
 		plist_add(node, &o->requests);
 	}
 	curr_value = pm_qos_get_value(o);
+	pm_qos_set_value(o, curr_value);
 	spin_unlock_irqrestore(&pm_qos_lock, flags);
 
 	if (prev_value != curr_value)
@@ -190,18 +205,11 @@ static int find_pm_qos_object_by_minor(int minor)
  * pm_qos_request - returns current system wide qos expectation
  * @pm_qos_class: identification of which qos value is requested
  *
- * This function returns the current target value in an atomic manner.
+ * This function returns the current target value.
  */
 int pm_qos_request(int pm_qos_class)
 {
-	unsigned long flags;
-	int value;
-
-	spin_lock_irqsave(&pm_qos_lock, flags);
-	value = pm_qos_get_value(pm_qos_array[pm_qos_class]);
-	spin_unlock_irqrestore(&pm_qos_lock, flags);
-
-	return value;
+	return pm_qos_read_value(pm_qos_array[pm_qos_class]);
 }
 EXPORT_SYMBOL_GPL(pm_qos_request);
 


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/