驱动篇：inux 电源管理的系统架构和驱动（三）
驱动篇：inux 电源管理的系统架构和驱动（三）
PowerTop
PowerTop 是一款开源的用于进行电量消耗分析和电源管理诊断的工具,其主页位于 Intel 开源技术中心的 https://01.org/powertop/,维护者是 Arjan van de Ven 和 Kristen Accardi 。 PowerTop 可分析系统中软件的功耗,以便找到功耗大户,也可显示系统中不同的 C 状态(与 CPUIdle 驱动对应)和 P 状态(与 CPUFreq 驱动对应)的时间比例,并采用了基于 TAB 的界面风格
Regulator 驱动
Regulator 是 Linux 系统中电源管理的基础设施之一,用于稳压电源的管理,是各种驱动子系统中设置电压的标准接口。前面介绍的CPUFreq 驱动就经常使用它来设定电压
ret = regulator_set_voltage(vddarm, dvfs->vddarm_min, dvfs->vddarm_max);
而 Regulator 则可以管理系统中的供电单元,即稳压器( Low Dropout Regulator , LDO ,即低压差线性稳压器),并提供获取和设置这些供电单元电压的接口。一般在 ARM 电路板上,各个稳压器和设备会形成一个 Regulator 树形结构Linux 的 Regulator 子系统提供如下 API 以用于注册 / 注销一个稳压器:
structregulator_dev* regulator_register(conststructregulator_desc
*regulator_desc, conststructregulator_config *config);
voidregulator_unregister(structregulator_dev *rdev);
regulator_register ()函数的两个参数分别是 regulator_desc 结构体和
regulator_config 结构体的指针。
regulator_desc 结构体是对这个稳压器属性和操作的封装：
struct regulator_desc {
const char *name; /* Regulator 的名字 */
const char *supply_name; /* Regulator Supply 的名字 */
const char *of_match;
const char *regulators_node;
int (*of_parse_cb)(struct device_node *,
const struct regulator_desc *,
struct regulator_config *);
int id;
unsigned int continuous_voltage_range:1;
unsigned n_voltages;
const struct regulator_ops *ops;
int irq;
enum regulator_type type;/* 是电压还是电流Regulator */
struct module *owner;
unsigned int min_uV;/* 在线性映射情况下最低的Selector 的电压 */
unsigned int uV_step;/*在线性映射情况下每步增加/减少的电压*/
unsigned int linear_min_sel;
int fixed_uV;
unsigned int ramp_delay;/* 电压改变后稳定下来所需时间*/
int min_dropout_uV;
const struct regulator_linear_range *linear_ranges;
int n_linear_ranges;
const unsigned int *volt_table;/*基于表映射情况下的电压映射表*/
unsigned int vsel_reg;
unsigned int vsel_mask;
unsigned int csel_reg;
unsigned int csel_mask;
unsigned int apply_reg;
unsigned int apply_bit;
unsigned int enable_reg;
unsigned int enable_mask;
unsigned int enable_val;
unsigned int disable_val;
bool enable_is_inverted;
unsigned int bypass_reg;
unsigned int bypass_mask;
unsigned int bypass_val_on;
unsigned int bypass_val_off;
unsigned int active_discharge_on;
unsigned int active_discharge_off;
unsigned int active_discharge_mask;
unsigned int active_discharge_reg;
unsigned int soft_start_reg;
unsigned int soft_start_mask;
unsigned int soft_start_val_on;
unsigned int pull_down_reg;
unsigned int pull_down_mask;
unsigned int pull_down_val_on;
unsigned int enable_time;
unsigned int off_on_delay;
unsigned int (*of_map_mode)(unsigned int mode);
};
上述结构体中的 regulator_ops 指针 ops 是对这个稳压器硬件操作的封装,其中包含获取、设置电压等的成员函数:
struct regulator_ops {
/* enumerate supported voltages */
int (*list_voltage) (struct regulator_dev *, unsigned selector);
/* get/set regulator voltage */
int (*set_voltage) (struct regulator_dev *, int min_uV, int max_uV,
unsigned *selector);
int (*map_voltage)(struct regulator_dev *, int min_uV, int max_uV);
int (*set_voltage_sel) (struct regulator_dev *, unsigned selector);
int (*get_voltage) (struct regulator_dev *);
int (*get_voltage_sel) (struct regulator_dev *);
/* get/set regulator current */
int (*set_current_limit) (struct regulator_dev *,
int min_uA, int max_uA);
int (*get_current_limit) (struct regulator_dev *);
int (*set_input_current_limit) (struct regulator_dev *, int lim_uA);
int (*set_over_current_protection) (struct regulator_dev *);
int (*set_active_discharge) (struct regulator_dev *, bool enable);
/* enable/disable regulator */
int (*enable) (struct regulator_dev *);
int (*disable) (struct regulator_dev *);
int (*is_enabled) (struct regulator_dev *);
/* get/set regulator operating mode (defined in consumer.h) */
int (*set_mode) (struct regulator_dev *, unsigned int mode);
unsigned int (*get_mode) (struct regulator_dev *);
/* retrieve current error flags on the regulator */
int (*get_error_flags)(struct regulator_dev *, unsigned int *flags);
/* Time taken to enable or set voltage on the regulator */
int (*enable_time) (struct regulator_dev *);
int (*set_ramp_delay) (struct regulator_dev *, int ramp_delay);
int (*set_voltage_time) (struct regulator_dev *, int old_uV,
int new_uV);
int (*set_voltage_time_sel) (struct regulator_dev *,
unsigned int old_selector,
unsigned int new_selector);
int (*set_soft_start) (struct regulator_dev *);
/* report regulator status ... most other accessors report
* control inputs, this reports results of combining inputs
* from Linux (and other sources) with the actual load.
* returns REGULATOR_STATUS_* or negative errno.
*/
int (*get_status)(struct regulator_dev *);
/* get most efficient regulator operating mode for load */
unsigned int (*get_optimum_mode) (struct regulator_dev *, int input_uV,
int output_uV, int load_uA);
/* set the load on the regulator */
int (*set_load)(struct regulator_dev *, int load_uA);
/* control and report on bypass mode */
int (*set_bypass)(struct regulator_dev *dev, bool enable);
int (*get_bypass)(struct regulator_dev *dev, bool *enable);
/* the operations below are for configuration of regulator state when
* its parent PMIC enters a global STANDBY/HIBERNATE state */
/* set regulator suspend voltage */
int (*set_suspend_voltage) (struct regulator_dev *, int uV);
/* enable/disable regulator in suspend state */
int (*set_suspend_enable) (struct regulator_dev *);
int (*set_suspend_disable) (struct regulator_dev *);
/* set regulator suspend operating mode (defined in consumer.h) */
int (*set_suspend_mode) (struct regulator_dev *, unsigned int mode);
int (*set_pull_down) (struct regulator_dev *);
};
在 drivers/regulator 目录下,包含大量的与电源芯片对应的 Regulator 驱动,如 Dialog 的 DA9052 、 Intersil 的 ISL6271A 、 ST-Ericsson 的TPS61050/61052 、 Wolfon 的 WM831x 系列等,它同时提供了一个虚拟的 Regulator 驱动作为参考:
虚拟的 Regulator 驱动
struct regulator_dev *dummy_regulator_rdev;
static struct regulator_init_data dummy_initdata;
static struct regulator_ops dummy_ops;
static struct regulator_desc dummy_desc = {
.name = "regulator-dummy",
.id = -1,
.type = REGULATOR_VOLTAGE,
.owner = THIS_MODULE,
.ops = &dummy_ops,
};
static int __devinit dummy_regulator_probe(struct platform_device *pdev)
{
struct regulator_config config = { };
int ret;
config.dev = &pdev->dev;
config.init_data = &dummy_initdata;
dummy_regulator_rdev = regulator_register(&dummy_desc, &config);
if (IS_ERR(dummy_regulator_rdev)) {
ret = PTR_ERR(dummy_regulator_rdev);
pr_err("Failed to register regulator: %d\n", ret);
return ret;
}
return 0;
}
Linux 的 Regulator 子系统提供消费者( Consumer ) API 以便让其他的驱动获取、设置、关闭和使能稳压器:
struct regulator * regulator_get(structdevice *dev, const char *id);
struct regulator * devm_regulator_get(structdevice *dev, const char *id);
struct regulator * regulator_get_exclusive(structdevice *dev, const char *id);
void regulator_put(structregulator *regulator);
void devm_regulator_put(structregulator *regulator);
int regulator_enable(structregulator *regulator);
int regulator_disable(structregulator *regulator);
int regulator_set_voltage(structregulator *regulator, intmin_uV, intmax_uV);
int regulator_get_voltage(structregulator *regulator);
这些消费者 API 的地位大致与 GPIO 子系统的 gpio_request ()、时钟子系统的 clk_get ()、
dmaengine 子系统的dmaengine_submit ()等相当,属于基础设施
OPP
现今的 SoC 一般包含很多集成组件,在系统运行过程中,并不需要所有的模块都运行于最高频率和最高性能。在SoC 内,某些 domain 可以运行在较低的频率和电压下,而其他 domain 可以运行在较高的频率和电压下,某个domain 所支持的 < 频率,电压 > 对的集合被称为 Operating Performance Point ,缩写为 OPP 。
int opp_add(struct device *dev, unsigned long freq, unsigned long u_volt);
目前, TI OMAP CPUFreq 驱动的底层就使用了 OPP 这种机制来获取 CPU 所支持的频率和电压列表。在开机的过程中, TI OMAP4 芯片会注册针对 CPU 设备的 OPP 表(代码位于 arch/arm/mach-omap2/ 中):
TI OMAP4 CPU 的 OPP 表
static struct omap_opp_def __initdata omap44xx_opp_def_list[] = {
/* MPU OPP1 - OPP50 */
OPP_INITIALIZER("mpu", true, 300000000, OMAP4430_VDD_MPU_OPP50_UV),
/* MPU OPP2 - OPP100 */
OPP_INITIALIZER("mpu", true, 600000000, OMAP4430_VDD_MPU_OPP100_UV),
/* MPU OPP3 - OPP-Turbo */
OPP_INITIALIZER("mpu", true, 800000000, OMAP4430_VDD_MPU_OPPTURBO_UV),
/* MPU OPP4 - OPP-SB */
OPP_INITIALIZER("mpu", true, 1008000000, OMAP4430_VDD_MPU_OPPNITRO_UV),
...
};
/**
* omap4_opp_init() - initialize omap4 opp table 14 */
int __init omap4_opp_init(void)
{
...
r = omap_init_opp_table(omap44xx_opp_def_list,
ARRAY_SIZE(omap44xx_opp_def_list));
return r;
}
device_initcall(omap4_opp_init);
int __init omap_init_opp_table(struct omap_opp_def *opp_def,
u32 opp_def_size)
{
...
/* Lets now register with OPP library */
for (i = 0; i < opp_def_size; i++, opp_def++) {
...
if (!strncmp(opp_def->hwmod_name, "mpu", 3)) {
/*
* All current OMAPs share voltage rail and
* clock source, so CPU0 is used to represent
* the MPU-SS.
*/
dev = get_cpu_device(0);
} ...
r = opp_add(dev, opp_def->freq, opp_def->u_volt);
...
}
return 0;
}
针对与 device 结构体指针 dev 对应的 domain 中增加一个新的 OPP ,参数 freq 和 u_volt 即为该 OPP 对应的频率和电压。
int opp_enable(struct device *dev, unsigned long freq);
int opp_disable(struct device *dev, unsigned long freq);
上述 API 用于使能和禁止某个 OPP ,一旦被禁止,其 available 将成为 false ,之后有设备驱动想设置为这个 OPP 就不再可能了。譬如,当温度超过某个范围后,系统不允许 1GHz 的工作频率,可采用类似下面的代码实现:
if (cur_temp > temp_high_thresh) {
/* Disable 1GHz if it was enabled */
rcu_read_lock();
opp = opp_find_freq_exact(dev, 1000000000, true);
rcu_read_unlock();
/* just error check */
if (!IS_ERR(opp))
ret = opp_disable(dev, 1000000000);
else
goto try_something_else;
}
上述代码中调用的 opp_find_freq_exact ()用于寻找与一个确定频率和 available 匹配的 OPP ,其原型为:
struct opp *opp_find_freq_exact(struct device *dev, unsigned long freq,
bool available);
另外, Linux 还提供两个变体, opp_find_freq_floor ()用于寻找 1 个 OPP ,它的频率向上接近或等于指定的频率;opp_find_freq_ceil ()用于寻找 1 个 OPP ,它的频率向下接近或等于指定的频率,这两个函数的原型为:
struct opp *opp_find_freq_floor(struct device *dev, unsigned long *freq);
struct opp *opp_find_freq_ceil(struct device *dev, unsigned long *freq);
我们可用下面的代码分别寻找 1 个设备的最大和最小工作频率:
freq = ULONG_MAX;
rcu_read_lock();
opp_find_freq_floor(dev, &freq);
rcu_read_unlock();
freq = 0;
rcu_read_lock();
opp_find_freq_ceil(dev, &freq);
rcu_read_unlock();
在频率降低的同时,支撑该频率运行所需的电压也往往可以动态调低;反之,则可能需要调高,下面这两个 API分别用于获取与某 OPP 对应的电压和频率:
unsigned long opp_get_voltage(struct opp *opp);
unsigned long opp_get_freq(struct opp *opp);
举个例子,当某 CPUFreq 驱动想将 CPU 设置为某一频率的时候,它可能会同时设置电压,其代码流程为:
soc_switch_to_freq_voltage(freq){
/* do things */
rcu_read_lock();
opp = opp_find_freq_ceil(dev, &freq);
v = opp_get_voltage(opp);
rcu_read_unlock();
if (v)
regulator_set_voltage(.., v);
/* do other things */
}
如下简单的 API 可用于获取某设备所支持的 OPP 的个数:
int opp_get_opp_count(struct device *dev);
前面提到, TI OMAP CPUFreq 驱动的底层就使用了 OPP 这种机制来获取 CPU 所支持的频率和电压列表。它在omap_init_opp_table ()函数中添加了相应的 OPP ,在 TI OMAP 芯片的 CPUFreq 驱动 drivers/cpufreq/omap-cpufreq.c 中,则借助了快捷函数 opp_init_cpufreq_table ()来根据前面注册的 OPP 建立 CPUFreq 的频率表:
static int __cpuinit omap_cpu_init(struct cpufreq_policy *policy)
{
...
if (!freq_table)
result = opp_init_cpufreq_table(mpu_dev, &freq_table);
...
}
而在 CPUFreq 驱动的目标成员函数 omap_target ()中,则使用与 OPP 相关的 API 来获取频率和电压:
static int omap_target(struct cpufreq_policy *policy,
unsigned int target_freq,
unsigned int relation)
{
...
if (mpu_reg) {
opp = opp_find_freq_ceil(mpu_dev, &freq);
...
volt = opp_get_voltage(opp);
...
}
...
}
drivers/cpufreq/omap-cpufreq.c 相对来说较为规范,它在 < 频率,电压 > 表方面,在底层使用了 OPP ,在设置电压的时候又使用了规范的 Regulator API
比较新的驱动一般不太喜欢直接在代码里面固化 OPP 表,而是喜欢在相应的节点处添加 operating-points 属性,如imx27.dtsi 中的:
cpus {
#size-cells = <0>;
#address-cells = <1>;
cpu: aaa@qq.com0 {
device_type = "cpu";
compatible = "arm,arm926ej-s";
operating-points = <
/* kHz uV */
266000 1300000
399000 1450000
>;
clock-latency = <62500>;
clocks = <&clks IMX27_CLK_CPU_DIV>;
voltage-tolerance = <5>;
};
};
如果 CPUFreq 的变化可以使用非常标准的 regulator 、 clk API ,我们甚至可以直接使用 drivers/cpufreq/cpufreq-dt.c 这个驱动。这样只需要在 CPU 节点上填充好频率电压表,然后在平台代码里面注册 cpufreq-dt 设备就可以了,在arch/arm/mach-imx/imx27-dt.c 、 arch/arm/mach-imx/mach-imx51.c 中可以找到类似的例子:
static void __init imx27_dt_init(void)
{
struct platform_device_info devinfo = { .name = "cpufreq-dt", };
of_platform_populate(NULL, of_default_bus_match_table, NULL, NULL);
platform_device_register_full(&devinfo);
}
PM QoS
Linux 内核的 PM QoS 系统针对内核和应用程序提供了一套接口,通过这个接口,用户可以设定自身对性能的期望。一类是系统级的需求,通过 cpu_dma_latency 、 network_latency 和network_throughput 这些参数来设定;另一类是单个设备可以根据自身的性能需求发起 per-device 的 PM QoS 请求。在内核空间,通过 pm_qos_add_request ()函数可以注册 PM QoS 请求:
void pm_qos_add_request(struct pm_qos_request *req,
int pm_qos_class, s32 value);
通过 pm_qos_update_request ()函数可以更新已注册的 PM QoS 请求:
void pm_qos_update_request(struct pm_qos_request *req,
s32 new_value);
void pm_qos_update_request_timeout(struct pm_qos_request *req, s32 new_value,
unsigned long timeout_us);
通过 pm_qos_remove_request ()函数可以删除已注册的 PM QoS 请求:
void pm_qos_remove_request(struct pm_qos_request *req);
譬如在 drivers/media/platform/via-camera.c 这个摄像头驱动中,当摄像头开启后,通过如下语句可以阻止 CPU 进入C3 级别的深度 Idle :
static int viacam_streamon(struct file *filp, void *priv, enum v4l2_buf_type t)
{
...
pm_qos_add_request(&cam->qos_request, PM_QOS_CPU_DMA_LATENCY, 50);
...
}
这是因为,在 CPUIdle 子系统中,会根据 PM_QOS_CPU_DMA_LATENCY 请求的情况选择合适的 C 状态,如drivers/cpuidle/governors/ladder.c 中的 ladder_select_state ()就会判断目标 C 状态的exit_latency 与 QoS 要求的关系,如代码清单所示：
CPUIdle LADDER governor 对 QoS 的判断
static int ladder_select_state(struct cpuidle_driver *drv,
struct cpuidle_device *dev)
{
...
int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY);
...
/* consider promotion */
if (last_idx < drv->state_count - 1 &&
!drv->states[last_idx + 1].disabled &&
!dev->states_usage[last_idx + 1].disable &&
last_residency > last_state->threshold.promotion_time &&
drv->states[last_idx + 1].exit_latency <= latency_req) {
last_state->stats.promotion_count++;
last_state->stats.demotion_count = 0;
if(last_state->stats.promotion_count>=
last_state->threshold.promotion_count) {
ladder_do_selection(ldev, last_idx, last_idx + 1);
return last_idx + 1;
}
}
...
}
LADDER 在选择是否进入更深层次的 C 状态时,会比较 C 状态的 exit_latency 要小于通过pm_qos_request ( PM_QOS_CPU_DMA_LATENCY )得到的 PM QoS 请求的延迟,drv->states[last_idx + 1].exit_latency <= latency_req)
同样的逻辑也出现于 drivers/cpuidle/governors/menu.c 中,如代码清单的第 18~19 行。
static int menu_select(struct cpuidle_driver *drv, struct cpuidle_device *dev)
{
struct menu_device *data = &__get_cpu_var(menu_devices);
int latency_req = pm_qos_request(PM_QOS_CPU_DMA_LATENCY);
...
/*
* Find the idle state with the lowest power while satisfying
* our constraints.9
*/
for (i = CPUIDLE_DRIVER_STATE_START; i < drv->state_count; i++) {
struct cpuidle_state *s = &drv->states[i];
struct cpuidle_state_usage *su = &dev->states_usage[i];
if (s->disabled || su->disable)
continue;
if (s->target_residency > data->predicted_us)
continue;
if ( s->exit_latency > latency_req)
continue;
if (s->exit_latency * multiplier > data->predicted_us)
continue;
if (s->power_usage < power_usage) {
power_usage = s->power_usage;
data->last_state_idx = i;
data->exit_us = s->exit_latency;
}
}
return data->last_state_idx;
}
还是回到 drivers/media/platform/via-camera.c 中,当摄像头关闭后,它会通过如下语句告知上述代码对PM_QOS_CPU_DMA_LATENCY 的性能要求取消:
static int viacam_streamon(struct file *filp, void *priv, enum v4l2_buf_type t)
{
...
pm_qos_remove_request(&cam->qos_request);
...
}
类似的在设备驱动中申请 QoS 特性的例子还包括 drivers/net/wireless/ipw2x00/ipw2100.c 、 drivers/tty/serial/omap-serial.c 、 drivers/net/ethernet/intel/e1000e/netdev.c 等。
应用程序则可以通过向 /dev/cpu_dma_latency 和 /dev/network_latency 这样的设备节点写入值来发起 QoS 的性能请求。
上一篇： 数控恒压源的分享
下一篇： 《初学者C51自学笔记》之ADC0804