blob: 1094f9df9dbe05d01a4154e4c326ced6503124b2 [file] [log] [blame]
/*
* Copyright (C) 2010-2012, 2014, 2016 ARM Limited. All rights reserved.
*
* This program is free software and is provided to you under the terms of the GNU General Public License version 2
* as published by the Free Software Foundation, and any use by you of this program is subject to the terms of such GNU licence.
*
* A copy of the licence is included with the program, and can also be obtained from Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*/
#include <linux/mali/mali_utgard.h>
#include "mali_kernel_common.h"
#include "mali_scheduler.h"
#include "mali_dvfs_policy.h"
#include "mali_osk_mali.h"
#include "mali_osk_profiling.h"
#define CLOCK_TUNING_TIME_DEBUG 0
#define MAX_PERFORMANCE_VALUE 256
#define MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(percent) ((int) ((percent)*(MAX_PERFORMANCE_VALUE)/100.0 + 0.5))
/** The max fps the same as display vsync default 60, can set by module insert parameter */
int mali_max_system_fps = 60;
/** A lower limit on their desired FPS default 58, can set by module insert parameter */
int mali_desired_fps = 58;
static int mali_fps_step1 = 0;
static int mali_fps_step2 = 0;
static int clock_step = -1;
static int cur_clk_step = -1;
static struct mali_gpu_clock *gpu_clk = NULL;
/*Function prototype */
static int (*mali_gpu_set_freq)(int) = NULL;
static int (*mali_gpu_get_freq)(void) = NULL;
static mali_bool mali_dvfs_enabled = MALI_FALSE;
#define NUMBER_OF_NANOSECONDS_PER_SECOND 1000000000ULL
static u32 calculate_window_render_fps(u64 time_period)
{
u32 max_window_number;
u64 tmp;
u64 max = time_period;
u32 leading_zeroes;
u32 shift_val;
u32 time_period_shift;
u32 max_window_number_shift;
u32 ret_val;
max_window_number = mali_session_max_window_num();
/* To avoid float division, extend the dividend to ns unit */
tmp = (u64)max_window_number * NUMBER_OF_NANOSECONDS_PER_SECOND;
if (tmp > time_period) {
max = tmp;
}
/*
* We may have 64-bit values, a dividend or a divisor or both
* To avoid dependencies to a 64-bit divider, we shift down the two values
* equally first.
*/
leading_zeroes = _mali_osk_clz((u32)(max >> 32));
shift_val = 32 - leading_zeroes;
time_period_shift = (u32)(time_period >> shift_val);
max_window_number_shift = (u32)(tmp >> shift_val);
ret_val = max_window_number_shift / time_period_shift;
return ret_val;
}
static bool mali_pickup_closest_avail_clock(int target_clock_mhz, mali_bool pick_clock_up)
{
int i = 0;
bool clock_changed = false;
/* Round up the closest available frequency step for target_clock_hz */
for (i = 0; i < gpu_clk->num_of_steps; i++) {
/* Find the first item > target_clock_hz */
if (((int)(gpu_clk->item[i].clock) - target_clock_mhz) > 0) {
break;
}
}
/* If the target clock greater than the maximum clock just pick the maximum one*/
if (i == gpu_clk->num_of_steps) {
i = gpu_clk->num_of_steps - 1;
} else {
if ((!pick_clock_up) && (i > 0)) {
i = i - 1;
}
}
clock_step = i;
if (cur_clk_step != clock_step) {
clock_changed = true;
}
return clock_changed;
}
void mali_dvfs_policy_realize(struct mali_gpu_utilization_data *data, u64 time_period)
{
int under_perform_boundary_value = 0;
int over_perform_boundary_value = 0;
int current_fps = 0;
int current_gpu_util = 0;
bool clock_changed = false;
#if CLOCK_TUNING_TIME_DEBUG
struct timeval start;
struct timeval stop;
unsigned int elapse_time;
do_gettimeofday(&start);
#endif
u32 window_render_fps;
if (NULL == gpu_clk) {
MALI_DEBUG_PRINT(2, ("Enable DVFS but patform doesn't Support freq change. \n"));
return;
}
window_render_fps = calculate_window_render_fps(time_period);
current_fps = window_render_fps;
current_gpu_util = data->utilization_gpu;
/* Get the specific under_perform_boundary_value and over_perform_boundary_value */
if ((mali_desired_fps <= current_fps) && (current_fps < mali_max_system_fps)) {
under_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(90);
over_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(70);
} else if ((mali_fps_step1 <= current_fps) && (current_fps < mali_desired_fps)) {
under_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(55);
over_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(35);
} else if ((mali_fps_step2 <= current_fps) && (current_fps < mali_fps_step1)) {
under_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(70);
over_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(50);
} else {
under_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(55);
over_perform_boundary_value = MALI_PERCENTAGE_TO_UTILIZATION_FRACTION(35);
}
MALI_DEBUG_PRINT(5, ("Using ARM power policy: gpu util = %d \n", current_gpu_util));
MALI_DEBUG_PRINT(5, ("Using ARM power policy: under_perform = %d, over_perform = %d \n", under_perform_boundary_value, over_perform_boundary_value));
MALI_DEBUG_PRINT(5, ("Using ARM power policy: render fps = %d, pressure render fps = %d \n", current_fps, window_render_fps));
/* Get current clock value */
cur_clk_step = mali_gpu_get_freq();
/* Consider offscreen */
if (0 == current_fps) {
/* GP or PP under perform, need to give full power */
if (current_gpu_util > over_perform_boundary_value) {
if (cur_clk_step != gpu_clk->num_of_steps - 1) {
clock_changed = true;
clock_step = gpu_clk->num_of_steps - 1;
}
}
/* If GPU is idle, use lowest power */
if (0 == current_gpu_util) {
if (cur_clk_step != 0) {
clock_changed = true;
clock_step = 0;
}
}
goto real_setting;
}
/* 2. Calculate target clock if the GPU clock can be tuned */
if (-1 != cur_clk_step) {
int target_clk_mhz = -1;
mali_bool pick_clock_up = MALI_TRUE;
if (current_gpu_util > under_perform_boundary_value) {
/* when under perform, need to consider the fps part */
target_clk_mhz = gpu_clk->item[cur_clk_step].clock * current_gpu_util * mali_desired_fps / under_perform_boundary_value / current_fps;
pick_clock_up = MALI_TRUE;
} else if (current_gpu_util < over_perform_boundary_value) {
/* when over perform, did't need to consider fps, system didn't want to reach desired fps */
target_clk_mhz = gpu_clk->item[cur_clk_step].clock * current_gpu_util / under_perform_boundary_value;
pick_clock_up = MALI_FALSE;
}
if (-1 != target_clk_mhz) {
clock_changed = mali_pickup_closest_avail_clock(target_clk_mhz, pick_clock_up);
}
}
real_setting:
if (clock_changed) {
mali_gpu_set_freq(clock_step);
_mali_osk_profiling_add_event(MALI_PROFILING_EVENT_TYPE_SINGLE |
MALI_PROFILING_EVENT_CHANNEL_GPU |
MALI_PROFILING_EVENT_REASON_SINGLE_GPU_FREQ_VOLT_CHANGE,
gpu_clk->item[clock_step].clock,
gpu_clk->item[clock_step].vol / 1000,
0, 0, 0);
}
#if CLOCK_TUNING_TIME_DEBUG
do_gettimeofday(&stop);
elapse_time = timeval_to_ns(&stop) - timeval_to_ns(&start);
MALI_DEBUG_PRINT(2, ("Using ARM power policy: eclapse time = %d\n", elapse_time));
#endif
}
_mali_osk_errcode_t mali_dvfs_policy_init(void)
{
_mali_osk_device_data data;
_mali_osk_errcode_t err = _MALI_OSK_ERR_OK;
if (_MALI_OSK_ERR_OK == _mali_osk_device_data_get(&data)) {
if ((NULL != data.get_clock_info) && (NULL != data.set_freq) && (NULL != data.get_freq)) {
MALI_DEBUG_PRINT(2, ("Mali DVFS init: using arm dvfs policy \n"));
mali_fps_step1 = mali_max_system_fps / 3;
mali_fps_step2 = mali_max_system_fps / 5;
data.get_clock_info(&gpu_clk);
if (gpu_clk != NULL) {
#ifdef DEBUG
int i;
for (i = 0; i < gpu_clk->num_of_steps; i++) {
MALI_DEBUG_PRINT(5, ("mali gpu clock info: step%d clock(%d)Hz,vol(%d) \n",
i, gpu_clk->item[i].clock, gpu_clk->item[i].vol));
}
#endif
} else {
MALI_DEBUG_PRINT(2, ("Mali DVFS init: platform didn't define enough info for ddk to do DVFS \n"));
}
mali_gpu_get_freq = data.get_freq;
mali_gpu_set_freq = data.set_freq;
if ((NULL != gpu_clk) && (gpu_clk->num_of_steps > 0)
&& (NULL != mali_gpu_get_freq) && (NULL != mali_gpu_set_freq)) {
mali_dvfs_enabled = MALI_TRUE;
}
} else {
MALI_DEBUG_PRINT(2, ("Mali DVFS init: platform function callback incomplete, need check mali_gpu_device_data in platform .\n"));
}
} else {
err = _MALI_OSK_ERR_FAULT;
MALI_DEBUG_PRINT(2, ("Mali DVFS init: get platform data error .\n"));
}
return err;
}
/*
* Always give full power when start a new period,
* if mali dvfs enabled, for performance consideration
*/
void mali_dvfs_policy_new_period(void)
{
/* Always give full power when start a new period */
unsigned int cur_clk_step = 0;
cur_clk_step = mali_gpu_get_freq();
if (cur_clk_step != (gpu_clk->num_of_steps - 1)) {
mali_gpu_set_freq(gpu_clk->num_of_steps - 1);
_mali_osk_profiling_add_event(MALI_PROFILING_EVENT_TYPE_SINGLE |
MALI_PROFILING_EVENT_CHANNEL_GPU |
MALI_PROFILING_EVENT_REASON_SINGLE_GPU_FREQ_VOLT_CHANGE, gpu_clk->item[gpu_clk->num_of_steps - 1].clock,
gpu_clk->item[gpu_clk->num_of_steps - 1].vol / 1000, 0, 0, 0);
}
}
mali_bool mali_dvfs_policy_enabled(void)
{
return mali_dvfs_enabled;
}
#if defined(CONFIG_MALI400_PROFILING)
void mali_get_current_gpu_clk_item(struct mali_gpu_clk_item *clk_item)
{
if (mali_platform_device != NULL) {
struct mali_gpu_device_data *device_data = NULL;
device_data = (struct mali_gpu_device_data *)mali_platform_device->dev.platform_data;
if ((NULL != device_data->get_clock_info) && (NULL != device_data->get_freq)) {
int cur_clk_step = device_data->get_freq();
struct mali_gpu_clock *mali_gpu_clk = NULL;
device_data->get_clock_info(&mali_gpu_clk);
clk_item->clock = mali_gpu_clk->item[cur_clk_step].clock;
clk_item->vol = mali_gpu_clk->item[cur_clk_step].vol;
} else {
MALI_DEBUG_PRINT(2, ("Mali GPU Utilization: platform function callback incomplete, need check mali_gpu_device_data in platform .\n"));
}
}
}
#endif