Re: 9908859acaa9 cpuidle/menu: add per CPU PM QoS resume latency consideration

From: Alex Shi
Date: Wed Feb 22 2017 - 09:53:44 EST


cc Rafael.


On 02/22/2017 09:12 PM, Peter Zijlstra wrote:
> On Wed, Feb 22, 2017 at 01:56:37PM +0100, Mike Galbraith wrote:
>> Hi,
>>
>> Do we really need a spinlock for that in the idle loop?
>
> Urgh, that's broken on RT, you cannot schedule the idle loop.
>
> Also, yeah, reading a s32 should not need no locking, but there's a
> bunch of pointer chases in between :/

Do you mean s/should not/should/ ? :)

Yes, the dev_pm_qos_read_value() using a power.lock, that is right for normal device.
But as to this cpu here, the lock isn't necessary.

Hi Rafael,
Is this fix ok?

===========
diff --git a/drivers/cpuidle/governors/menu.c b/drivers/cpuidle/governors/menu.c
index 8d6d25c..957c56d 100644
--- a/drivers/cpuidle/governors/menu.c
+++ b/drivers/cpuidle/governors/menu.c
@@ -273,6 +273,14 @@ static unsigned int get_typical_interval(struct menu_device *data)
goto again;
}

+int read_this_cpu_resume_latency(int cpu)
+{
+ struct device *dev = get_cpu_device(cpu);
+
+ return IS_ERR_OR_NULL(dev->power.qos) ?
+ 0 : pm_qos_read_value(&dev->power.qos->resume_latency);
+}
+
/**
* menu_select - selects the next idle state to enter
* @drv: cpuidle driver containing state data
@@ -281,13 +289,12 @@ static unsigned int get_typical_interval(struct menu_device *data)
static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
{
struct menu_device *data = this_cpu_ptr(&menu_devices);
- struct device *device = get_cpu_device(dev->cpu);
int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY);
int i;
unsigned int interactivity_req;
unsigned int expected_interval;
unsigned long nr_iowaiters, cpu_load;
- int resume_latency = dev_pm_qos_read_value(device);
+ int resume_latency = read_this_cpu_resume_latency(dev->cpu);

if (data->needs_update) {
menu_update(drv, dev);