Home | History | Annotate | Download | only in sys
      1 /*
      2  * CDDL HEADER START
      3  *
      4  * The contents of this file are subject to the terms of the
      5  * Common Development and Distribution License (the "License").
      6  * You may not use this file except in compliance with the License.
      7  *
      8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9  * or http://www.opensolaris.org/os/licensing.
     10  * See the License for the specific language governing permissions
     11  * and limitations under the License.
     12  *
     13  * When distributing Covered Code, include this CDDL HEADER in each
     14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15  * If applicable, add the following below this CDDL HEADER, with the
     16  * fields enclosed by brackets "[]" replaced with your own identifying
     17  * information: Portions Copyright [yyyy] [name of copyright owner]
     18  *
     19  * CDDL HEADER END
     20  */
     21 /*
     22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
     23  * Use is subject to license terms.
     24  */
     25 
     26 #ifndef _SYS_CPUDRV_H
     27 #define	_SYS_CPUDRV_H
     28 
     29 #include <sys/promif.h>
     30 #include <sys/cpuvar.h>
     31 #include <sys/taskq.h>
     32 
     33 #ifdef	__cplusplus
     34 extern "C" {
     35 #endif
     36 
     37 #ifdef _KERNEL
     38 
     39 /*
     40  * CPU power management data
     41  */
     42 /*
     43  * Data related to a particular speed.
     44  *
     45  * All per speed data nodes for a CPU are linked together using down_spd.
     46  * The link list is ordered with first node containing data for
     47  * normal (maximum) speed. up_spd points to the next speed up. Currently
     48  * all up_spd's point to the normal speed but this can be changed in future.
     49  * quant_cnt is the number of ticks when monitoring system will be called
     50  * next. There are different quant_cnt for different speeds.
     51  *
     52  * Note that 'speed' has different meaning depending upon the platform.
     53  * On SPARC, the speed is really a divisor of the maximum speed (e.g., a speed
     54  * of 2 means that it's 1/2 the maximum speed). On x86, speed is a processor
     55  * frequency.
     56  */
     57 typedef struct cpudrv_pm_spd {
     58 	uint_t			speed;		/* platform dependent notion */
     59 	uint_t			quant_cnt;	/* quantum count in ticks */
     60 	struct cpudrv_pm_spd	*down_spd;	/* ptr to next speed down */
     61 	struct cpudrv_pm_spd	*up_spd;	/* ptr to next speed up */
     62 	uint_t			idle_hwm;	/* down if idle thread >= hwm */
     63 	uint_t			idle_lwm;	/* up if idle thread < lwm */
     64 	uint_t			idle_bhwm_cnt;	/* # of iters idle is < hwm */
     65 	uint_t			idle_blwm_cnt;	/* # of iters idle is < lwm */
     66 	uint_t			user_hwm;	/* up if user thread > hwm */
     67 	int			user_lwm;	/* down if user thread <= lwm */
     68 	int			pm_level;	/* power level for framework */
     69 } cpudrv_pm_spd_t;
     70 
     71 /*
     72  * Power management data
     73  */
     74 typedef struct cpudrv_pm {
     75 	cpudrv_pm_spd_t	*head_spd;	/* ptr to head of speed */
     76 	cpudrv_pm_spd_t	*cur_spd;	/* ptr to current speed */
     77 	cpudrv_pm_spd_t	*targ_spd;	/* target speed when cur_spd */
     78 					/* is unknown (i.e. NULL) */
     79 	uint_t		num_spd;	/* number of speeds */
     80 	hrtime_t	lastquan_mstate[NCMSTATES]; /* last quantum's mstate */
     81 	clock_t		lastquan_lbolt;	/* last quantum's lbolt */
     82 	int		pm_busycnt;	/* pm_busy_component() count  */
     83 	taskq_t		*tq;		/* taskq handler for CPU monitor */
     84 	timeout_id_t	timeout_id;	/* cpudrv_pm_monitor()'s timeout_id */
     85 	int		timeout_count;	/* count dispatched timeouts */
     86 	kmutex_t	timeout_lock;	/* protect timeout_count */
     87 	kcondvar_t	timeout_cv;	/* wait on timeout_count change */
     88 #if defined(__x86)
     89 	kthread_t	*pm_governor_thread; /* governor thread */
     90 #endif
     91 	boolean_t	pm_started;	/* PM really started */
     92 } cpudrv_pm_t;
     93 
     94 /*
     95  * Idle & user threads water marks in percentage
     96  */
     97 #if defined(__x86)
     98 #define	CPUDRV_PM_IDLE_HWM		85	/* idle high water mark */
     99 #define	CPUDRV_PM_IDLE_LWM		70	/* idle low water mark */
    100 #define	CPUDRV_PM_IDLE_BLWM_CNT_MAX	1    /* # of iters idle can be < lwm */
    101 #define	CPUDRV_PM_IDLE_BHWM_CNT_MAX	1    /* # of iters idle can be < hwm */
    102 #else
    103 #define	CPUDRV_PM_IDLE_HWM		98	/* idle high water mark */
    104 #define	CPUDRV_PM_IDLE_LWM		8	/* idle low water mark */
    105 #define	CPUDRV_PM_IDLE_BLWM_CNT_MAX	2    /* # of iters idle can be < lwm */
    106 #define	CPUDRV_PM_IDLE_BHWM_CNT_MAX	2    /* # of iters idle can be < hwm */
    107 #endif
    108 #define	CPUDRV_PM_USER_HWM		20	/* user high water mark */
    109 #define	CPUDRV_PM_IDLE_BUF_ZONE		4    /* buffer zone when going down */
    110 
    111 
    112 /*
    113  * Maximums for creating 'pm-components' property
    114  */
    115 #define	CPUDRV_PM_COMP_MAX_DIG	4	/* max digits in power level */
    116 					/* or divisor */
    117 #define	CPUDRV_PM_COMP_MAX_VAL	9999	/* max value in above digits */
    118 
    119 /*
    120  * Component number for calls to PM framework
    121  */
    122 #define	CPUDRV_PM_COMP_NUM	0	/* first component is 0 */
    123 
    124 /*
    125  * Quantum counts for normal and other clock speeds in terms of ticks.
    126  *
    127  * In determining the quantum count, we need to balance two opposing factors:
    128  *
    129  *	1) Minimal delay when user start using the CPU that is in low
    130  *	power mode -- requires that we monitor more frequently,
    131  *
    132  *	2) Extra code executed because of frequent monitoring -- requires
    133  *	that we monitor less frequently.
    134  *
    135  * We reach a tradeoff between these two requirements by monitoring
    136  * more frequently when we are in low speed mode (CPUDRV_PM_QUANT_CNT_OTHR)
    137  * so we can bring the CPU up without user noticing it. Moreover, at low
    138  * speed we are not using CPU much so extra code execution should be fine.
    139  * Since we are in no hurry to bring CPU down and at normal speed and we
    140  * might really be using the CPU fully, we monitor less frequently
    141  * (CPUDRV_PM_QUANT_CNT_NORMAL).
    142  */
    143 #if defined(__x86)
    144 #define	CPUDRV_PM_QUANT_CNT_NORMAL	(hz * 1)	/* 1 sec */
    145 #else
    146 #define	CPUDRV_PM_QUANT_CNT_NORMAL	(hz * 5)	/* 5 sec */
    147 #endif
    148 #define	CPUDRV_PM_QUANT_CNT_OTHR	(hz * 1)	/* 1 sec */
    149 
    150 /*
    151  * Taskq parameters
    152  */
    153 #define	CPUDRV_PM_TASKQ_THREADS		1    /* # threads to run CPU monitor */
    154 #define	CPUDRV_PM_TASKQ_MIN		2	/* min # of taskq entries */
    155 #define	CPUDRV_PM_TASKQ_MAX		2	/* max # of taskq entries */
    156 
    157 
    158 /*
    159  * Device driver state structure
    160  */
    161 typedef struct cpudrv_devstate {
    162 	dev_info_t	*dip;		/* devinfo handle */
    163 	processorid_t	cpu_id;		/* CPU number for this node */
    164 	cpudrv_pm_t	cpudrv_pm;	/* power management data */
    165 	kmutex_t	lock;		/* protects state struct */
    166 	void		*mach_state; /* machine specific state */
    167 } cpudrv_devstate_t;
    168 
    169 extern void	*cpudrv_state;
    170 
    171 /*
    172  * Debugging definitions
    173  */
    174 #ifdef	DEBUG
    175 #define	D_INIT			0x00000001
    176 #define	D_FINI			0x00000002
    177 #define	D_ATTACH		0x00000004
    178 #define	D_DETACH		0x00000008
    179 #define	D_POWER			0x00000010
    180 #define	D_PM_INIT		0x00000020
    181 #define	D_PM_FREE		0x00000040
    182 #define	D_PM_COMP_CREATE	0x00000080
    183 #define	D_PM_MONITOR		0x00000100
    184 #define	D_PM_MONITOR_VERBOSE	0x00000200
    185 #define	D_PM_MONITOR_DELAY	0x00000400
    186 
    187 extern uint_t	cpudrv_debug;
    188 
    189 #define	_PRINTF prom_printf
    190 #define	DPRINTF(flag, args)	if (cpudrv_debug & flag) _PRINTF args;
    191 #else
    192 #define	DPRINTF(flag, args)
    193 #endif /* DEBUG */
    194 
    195 extern int cpudrv_pm_change_speed(cpudrv_devstate_t *, cpudrv_pm_spd_t *);
    196 extern boolean_t cpudrv_pm_get_cpu_id(dev_info_t *, processorid_t *);
    197 extern boolean_t cpudrv_pm_power_ready(void);
    198 extern boolean_t cpudrv_pm_is_governor_thread(cpudrv_pm_t *);
    199 extern boolean_t cpudrv_mach_pm_init(cpudrv_devstate_t *);
    200 extern void cpudrv_mach_pm_free(cpudrv_devstate_t *);
    201 
    202 #endif /* _KERNEL */
    203 
    204 #ifdef	__cplusplus
    205 }
    206 #endif
    207 
    208 #endif /* _SYS_CPUDRV_H */
    209