On 04/01/2014 10:33 AM, Dongsheng Wang wrote:
From: Wang Dongsheng <dongsheng.w...@freescale.com>
Add cpuidle support for e500 family, using cpuidle framework to
manage various low power modes. The new implementation will remain
compatible with original idle method.
I have done test about power consumption and latency. Cpuidle framework
will make CPU response time faster than original method, but power
consumption is higher than original method.
Power consumption:
The original method, power consumption is 10.51202 (W).
The cpuidle framework, power consumption is 10.5311 (W).
Latency:
The original method, avg latency is 6782 (us).
The cpuidle framework, avg latency is 6482 (us).
Initially, this supports PW10, PW20 and subsequent patches will support
DOZE/NAP and PH10, PH20.
Signed-off-by: Wang Dongsheng <dongsheng.w...@freescale.com>
diff --git a/arch/powerpc/include/asm/machdep.h
b/arch/powerpc/include/asm/machdep.h
index 5b6c03f..9301420 100644
--- a/arch/powerpc/include/asm/machdep.h
+++ b/arch/powerpc/include/asm/machdep.h
@@ -294,6 +294,15 @@ extern void power7_idle(void);
extern void ppc6xx_idle(void);
extern void book3e_idle(void);
+static inline void cpuidle_wait(void)
+{
+#ifdef CONFIG_PPC64
+ book3e_idle();
+#else
+ e500_idle();
+#endif
+}
+
/*
* ppc_md contains a copy of the machine description structure for the
* current platform. machine_id contains the initial address where the
diff --git a/arch/powerpc/kernel/sysfs.c b/arch/powerpc/kernel/sysfs.c
index 97e1dc9..edd193f 100644
--- a/arch/powerpc/kernel/sysfs.c
+++ b/arch/powerpc/kernel/sysfs.c
@@ -190,6 +190,9 @@ static ssize_t show_pw20_wait_time(struct device *dev,
return sprintf(buf, "%llu\n", time > 0 ? time : 0);
}
+#ifdef CONFIG_CPU_IDLE_E500
+u32 cpuidle_entry_bit;
+#endif
static void set_pw20_wait_entry_bit(void *val)
{
u32 *value = val;
@@ -204,7 +207,11 @@ static void set_pw20_wait_entry_bit(void *val)
/* set count */
pw20_idle |= ((MAX_BIT - *value) << PWRMGTCR0_PW20_ENT_SHIFT);
+#ifdef CONFIG_CPU_IDLE_E500
+ cpuidle_entry_bit = *value;
+#else
mtspr(SPRN_PWRMGTCR0, pw20_idle);
+#endif
}
static ssize_t store_pw20_wait_time(struct device *dev,
diff --git a/drivers/cpuidle/Kconfig.powerpc b/drivers/cpuidle/Kconfig.powerpc
index 66c3a09..0949dbf 100644
--- a/drivers/cpuidle/Kconfig.powerpc
+++ b/drivers/cpuidle/Kconfig.powerpc
@@ -18,3 +18,10 @@ config POWERNV_CPUIDLE
help
Select this option to enable processor idle state management
through cpuidle subsystem.
+
+config CPU_IDLE_E500
+ bool "CPU Idle Driver for E500 family processors"
+ depends on CPU_IDLE
+ depends on FSL_SOC_BOOKE
+ help
+ Select this to enable cpuidle on e500 family processors.
diff --git a/drivers/cpuidle/Makefile b/drivers/cpuidle/Makefile
index f71ae1b..7e6adea 100644
--- a/drivers/cpuidle/Makefile
+++ b/drivers/cpuidle/Makefile
@@ -18,3 +18,4 @@ obj-$(CONFIG_ARM_AT91_CPUIDLE) += cpuidle-at91.o
# POWERPC drivers
obj-$(CONFIG_PSERIES_CPUIDLE) += cpuidle-pseries.o
obj-$(CONFIG_POWERNV_CPUIDLE) += cpuidle-powernv.o
+obj-$(CONFIG_CPU_IDLE_E500) += cpuidle-e500.o
diff --git a/drivers/cpuidle/cpuidle-e500.c b/drivers/cpuidle/cpuidle-e500.c
new file mode 100644
index 0000000..ddc0def
--- /dev/null
+++ b/drivers/cpuidle/cpuidle-e500.c
@@ -0,0 +1,194 @@
+/*
+ * CPU Idle driver for Freescale PowerPC e500 family processors.
+ *
+ * Copyright 2014 Freescale Semiconductor, Inc.
+ *
+ * Author: Dongsheng Wang <dongsheng.w...@freescale.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/cpu.h>
+#include <linux/cpuidle.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/notifier.h>
+
+#include <asm/cputable.h>
+#include <asm/machdep.h>
+#include <asm/mpc85xx.h>
+
+static unsigned int max_idle_state;
+static struct cpuidle_state *cpuidle_state_table;
+
+struct cpuidle_driver e500_idle_driver = {
+ .name = "e500_idle",
+ .owner = THIS_MODULE,
+};
+
+static void e500_cpuidle(void)
+{
+ if (cpuidle_idle_call())
+ cpuidle_wait();
+}
Nope, that has been changed. No more call to cpuidle_idle_call in a driver.
+
+static int pw10_enter(struct cpuidle_device *dev,
+ struct cpuidle_driver *drv, int index)
+{
+ cpuidle_wait();
+ return index;
+}
+
+#define MAX_BIT 63
+#define MIN_BIT 1
+extern u32 cpuidle_entry_bit;
+static int pw20_enter(struct cpuidle_device *dev,
+ struct cpuidle_driver *drv, int index)
+{
+ u32 pw20_idle;
+ u32 entry_bit;
+ pw20_idle = mfspr(SPRN_PWRMGTCR0);
+ if ((pw20_idle & PWRMGTCR0_PW20_ENT) != PWRMGTCR0_PW20_ENT) {
+ pw20_idle &= ~PWRMGTCR0_PW20_ENT;
+ entry_bit = MAX_BIT - cpuidle_entry_bit;
+ pw20_idle |= (entry_bit << PWRMGTCR0_PW20_ENT_SHIFT);
+ mtspr(SPRN_PWRMGTCR0, pw20_idle);
+ }
+
+ cpuidle_wait();
+
+ pw20_idle &= ~PWRMGTCR0_PW20_ENT;
+ pw20_idle |= (MIN_BIT << PWRMGTCR0_PW20_ENT_SHIFT);
+ mtspr(SPRN_PWRMGTCR0, pw20_idle);
+
+ return index;
+}
Is it possible to give some comments and encapsulate the code with
explicit function names to be implemented in an arch specific directory
file (eg. pm.c) and export these functions in a linux/ header ? We try
to prevent to include asm if possible.
+
+static struct cpuidle_state pw_idle_states[] = {
+ {
+ .name = "pw10",
+ .desc = "pw10",
+ .flags = CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 0,
+ .target_residency = 0,
+ .enter = &pw10_enter
+ },
+
+ {
+ .name = "pw20",
+ .desc = "pw20-core-idle",
+ .flags = CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 1,
+ .target_residency = 50,
+ .enter = &pw20_enter
+ },
+};
No need to define this intermediate structure here, you can directly
initialize the cpuidle_driver:
struct cpuidle_driver e500_idle_driver = {
.name = "e500_idle",
.owner = THIS_MODULE,
.states = {
.name = "pw10",
.desc = "pw10",
.flags = CPUIDLE_FLAG_TIME_VALID,
.target_residency = 0,
.enter = &pw10_enter,
},
....
.state_count = 2,
};
Then in the init function you initialize the state_count consequently:
if (PVR_VER(cur_cpu_spec->pvr_value) != PVR_VER_E6500)
drv->state_count = 1;
Then you can kill:
max_idle_state, cpuidle_state_table, e500_idle_state_probe and
pw_idle_states.
+
+static int cpu_hotplug_notify(struct notifier_block *n,
+ unsigned long action, void *hcpu)
+{
+ unsigned long hotcpu = (unsigned long)hcpu;
+ struct cpuidle_device *dev =
+ per_cpu_ptr(cpuidle_devices, hotcpu);
+
+ if (dev && cpuidle_get_driver()) {
+ switch (action) {
+ case CPU_ONLINE:
+ case CPU_ONLINE_FROZEN:
+ cpuidle_pause_and_lock();
+ cpuidle_enable_device(dev);
+ cpuidle_resume_and_unlock();
+ break;
+
+ case CPU_DEAD:
+ case CPU_DEAD_FROZEN:
+ cpuidle_pause_and_lock();
+ cpuidle_disable_device(dev);
+ cpuidle_resume_and_unlock();
+ break;
+
+ default:
+ return NOTIFY_DONE;
+ }
+ }
+
+ return NOTIFY_OK;
+}
+
+static struct notifier_block cpu_hotplug_notifier = {
+ .notifier_call = cpu_hotplug_notify,
+};
Can you explain why this is needed ?
+static void e500_cpuidle_driver_init(void)
+{
+ int idle_state;
+ struct cpuidle_driver *drv = &e500_idle_driver;
Pass the cpuidle_driver as parameter to the function.
+
+ drv->state_count = 0;
+
+ for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
+ if (!cpuidle_state_table[idle_state].enter)
+ break;
+
+ drv->states[drv->state_count] = cpuidle_state_table[idle_state];
+ drv->state_count++;
+ }
This code should disappear.
As this function will just initialize state_count, you can move it in
caller and kill this function.
+}
+
+static int e500_idle_state_probe(void)
+{
+ if (cpuidle_disable != IDLE_NO_OVERRIDE)
+ return -ENODEV;
+
+ cpuidle_state_table = pw_idle_states;
+ max_idle_state = ARRAY_SIZE(pw_idle_states);
+
+ /* Disable PW20 feature for e500mc, e5500 */
+ if (PVR_VER(cur_cpu_spec->pvr_value) != PVR_VER_E6500)
+ cpuidle_state_table[1].enter = NULL;
+
+ if (!cpuidle_state_table || !max_idle_state)
+ return -ENODEV;
+
+ return 0;
+}
This code should disappear.
+static void replace_orig_idle(void *dummy)
+{
+ return;
+}
+
+static int __init e500_idle_init(void)
+{
+ struct cpuidle_driver *drv = &e500_idle_driver;
+ int err;
+
+ if (e500_idle_state_probe())
+ return -ENODEV;
+
+ e500_cpuidle_driver_init();
+ if (!drv->state_count)
+ return -ENODEV;
No need of this check, because:
1. you know how you initialized the driver (1 or 2 states)
2. this is already by the cpuidle framework
+
+ err = cpuidle_register(drv, NULL);
+ if (err) {
+ pr_err("Register e500 family cpuidle driver failed.\n");
extra carriage return.
+
+ return err;
+ }
+
+ err = register_cpu_notifier(&cpu_hotplug_notifier);
+ if (err)
+ pr_warn("Cpuidle driver: register cpu notifier failed.\n");
+
+ /* Replace the original way of idle after cpuidle registered. */
+ ppc_md.power_save = e500_cpuidle;
+ on_each_cpu(replace_orig_idle, NULL, 1);
Why ?
+ pr_info("e500_idle_driver registered.\n");
+
+ return 0;
+}
+late_initcall(e500_idle_init);
Thanks
-- Daniel
--
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs
Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev
