blob: d55bd63b65b59b5869acd0e14da01a132b959d96 [file] [log] [blame]
/*
**************************************************************************
* Copyright (c) 2013, 2015-2020 The Linux Foundation. All rights reserved.
* Permission to use, copy, modify, and/or distribute this software for
* any purpose with or without fee is hereby granted, provided that the
* above copyright notice and this permission notice appear in all copies.
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT
* OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
**************************************************************************
*/
/*
* nss_freq.c
* NSS frequency change APIs
*/
#include "nss_stats.h"
#include "nss_tx_rx_common.h"
#include "nss_freq_log.h"
#include "nss_freq_stats.h"
#define NSS_ACK_STARTED 0
#define NSS_ACK_FINISHED 1
#define NSS_FREQ_USG_AVG_FREQUENCY 1000 /* Time in ms over which CPU Usage is averaged */
#define NSS_FREQ_CPU_USAGE_MAX_BOUND 75 /* MAX CPU usage equivalent to running max instructions excluding all the hazards */
#define NSS_FREQ_CPU_USAGE_MAX 100 /* MAX CPU usage equivalent to running max instructions including all the hazards.
This is also the ideal maximum usage value. */
/*
* Spinlock to protect the global data structure nss_freq_cpu_status
*/
DEFINE_SPINLOCK(nss_freq_cpu_usage_lock);
/*
* At any point, this object has the latest data about CPU utilization.
*/
struct nss_freq_cpu_usage nss_freq_cpu_status;
extern struct nss_runtime_sampling nss_runtime_samples;
extern struct workqueue_struct *nss_wq;
extern nss_work_t *nss_work;
/*
* nss_freq_msg_init()
* Initialize the freq message
*/
static void nss_freq_msg_init(struct nss_corefreq_msg *ncm, uint16_t if_num, uint32_t type, uint32_t len,
void *cb, void *app_data)
{
nss_cmn_msg_init(&ncm->cm, if_num, type, len, cb, app_data);
}
/*
* nss_freq_handle_ack()
* Handle the nss ack of frequency change.
*/
static void nss_freq_handle_ack(struct nss_ctx_instance *nss_ctx, struct nss_freq_msg *nfa)
{
if (nfa->ack == NSS_ACK_STARTED) {
/*
* NSS finished start noficiation - HW change clocks and send end notification
*/
nss_info("%px: NSS ACK Received: %d - Change HW CLK/Send Finish to NSS\n", nss_ctx, nfa->ack);
return;
}
if (nfa->ack == NSS_ACK_FINISHED) {
/*
* NSS finished end notification - Done
*/
nss_info("%px: NSS ACK Received: %d - End Notification ACK - Running: %dmhz\n", nss_ctx, nfa->ack, nfa->freq_current);
nss_runtime_samples.freq_scale_ready = 1;
return;
}
nss_info("%px: NSS had an error - Running: %dmhz\n", nss_ctx, nfa->freq_current);
}
/*
* nss_freq_queue_work()
* Queue Work to the NSS Workqueue based on Current index.
*/
static bool nss_freq_queue_work(void)
{
nss_freq_scales_t index = nss_runtime_samples.freq_scale_index;
BUG_ON(!nss_wq);
nss_info("frequency:%d index:%d sample count:%x\n", nss_runtime_samples.freq_scale[index].frequency,
index, nss_runtime_samples.average);
/*
* schedule freq change with autoscale ON
*/
return nss_freq_sched_change(index, true);
}
/*
* nss_freq_get_cpu_usage()
* Returns the CPU usage value in percentage at any instance for a required core. Returns -1 in case of an error.
*
* Calculation frequency is 1 second. Range of usage is 0-100. This API returns -1 if CPU usage is requested for core 1.
* TODO: Extend this API to get CPU usage for core 1.
*/
int8_t nss_freq_get_cpu_usage(uint32_t core_id)
{
int8_t usage;
if (core_id == 0) {
spin_lock_bh(&nss_freq_cpu_usage_lock);
usage = nss_freq_cpu_status.used;
spin_unlock_bh(&nss_freq_cpu_usage_lock);
return usage;
}
nss_warning("CPU usage functionality is not supported for core %u\n", core_id);
return -1;
}
/*
* nss_freq_compute_cpu_usage()
* Computes the CPU utilization and maximum-minumun cpu utilization since boot.
*/
static void nss_freq_compute_cpu_usage(struct nss_ctx_instance *nss_ctx, uint32_t inst_cnt)
{
uint32_t estimated_ins_capacity;
uint8_t actual_usage;
uint8_t usage;
spin_lock_bh(&nss_freq_cpu_usage_lock);
/*
* If actual CPU usage turns up higher than 100, there is something wrong with the received data.
* Upper bound average varies between 80% usage to 100% usage.
*
* TODO: To improve estimation algorithm for calculating how many actual instructions are executed.
*/
actual_usage = (inst_cnt * 100) / nss_freq_cpu_status.max_ins;
if ((actual_usage > NSS_FREQ_CPU_USAGE_MAX) || (actual_usage == 0)) {
spin_unlock_bh(&nss_freq_cpu_usage_lock);
return;
}
/*
* Simpler version of below math: This is calculating the reduced number of maximum instructions
* estimated_ins_capacity = nss_freq_cpu_status.avg_up% of nss_freq_cpu_status.max_ins
* Calculating usage percentage: usage = (inst_cnt/estimated_ins_capacity) * 100
*/
estimated_ins_capacity = ((NSS_FREQ_CPU_USAGE_MAX_BOUND * nss_freq_cpu_status.max_ins) / 100);
if (estimated_ins_capacity == 0) {
spin_unlock_bh(&nss_freq_cpu_usage_lock);
return;
}
usage = (inst_cnt * 100) / estimated_ins_capacity;
/*
* Average the instructions over NSS_FREQ_USG_AVG_FREQUENCY ms
*/
if (nss_freq_cpu_status.avg_ctr == NSS_FREQ_USG_AVG_FREQUENCY) {
nss_freq_cpu_status.used = nss_freq_cpu_status.total / NSS_FREQ_USG_AVG_FREQUENCY;
/*
* Due to our estimation, this could go beyond the end limit of 100%
*/
if (nss_freq_cpu_status.used > NSS_FREQ_CPU_USAGE_MAX) {
nss_freq_cpu_status.used = NSS_FREQ_CPU_USAGE_MAX;
}
/*
* Getting the all time max and min usage
*/
if (nss_freq_cpu_status.used > nss_freq_cpu_status.max) {
nss_freq_cpu_status.max = nss_freq_cpu_status.used;
}
if (nss_freq_cpu_status.used < nss_freq_cpu_status.min) {
nss_freq_cpu_status.min = nss_freq_cpu_status.used;
}
nss_trace("%px: max_instructions:%d cpu_usage:%d max_usage:%d min_usage:%d\n", nss_ctx,
nss_freq_cpu_status.max_ins, nss_freq_cpu_status.used, nss_freq_cpu_status.max, nss_freq_cpu_status.min);
nss_freq_cpu_status.total = 0;
nss_freq_cpu_status.avg_ctr = 0;
}
nss_freq_cpu_status.total += usage;
nss_freq_cpu_status.avg_ctr++;
spin_unlock_bh(&nss_freq_cpu_usage_lock);
}
/*
* nss_freq_scale_frequency()
* Frequency scaling algorithm to scale frequency.
*/
void nss_freq_scale_frequency(struct nss_ctx_instance *nss_ctx, uint32_t inst_cnt)
{
uint32_t b_index;
uint32_t minimum;
uint32_t maximum;
uint32_t index = nss_runtime_samples.freq_scale_index;
/*
* We do not accept any statistics if auto scaling is off,
* we start with a fresh sample set when scaling is
* eventually turned on.
*/
if (!nss_cmd_buf.auto_scale && nss_runtime_samples.initialized) {
return;
}
/*
* Delete Current Index Value, Add New Value, Recalculate new Sum, Shift Index
*/
b_index = nss_runtime_samples.buffer_index;
nss_runtime_samples.sum = nss_runtime_samples.sum - nss_runtime_samples.buffer[b_index];
nss_runtime_samples.buffer[b_index] = inst_cnt;
nss_runtime_samples.sum = nss_runtime_samples.sum + nss_runtime_samples.buffer[b_index];
nss_runtime_samples.buffer_index = (b_index + 1) & NSS_SAMPLE_BUFFER_MASK;
if (nss_runtime_samples.sample_count < NSS_SAMPLE_BUFFER_SIZE) {
nss_runtime_samples.sample_count++;
/*
* Samples Are All Ready, Start Auto Scale
*/
if (nss_runtime_samples.sample_count == NSS_SAMPLE_BUFFER_SIZE ) {
nss_cmd_buf.auto_scale = 1;
nss_runtime_samples.freq_scale_ready = 1;
nss_runtime_samples.initialized = 1;
}
return;
}
nss_runtime_samples.average = nss_runtime_samples.sum / nss_runtime_samples.sample_count;
/*
* Print out statistics every 10 samples
*/
if (nss_runtime_samples.message_rate_limit++ >= NSS_MESSAGE_RATE_LIMIT) {
nss_trace("%px: Running AVG:%x Sample:%x Divider:%d\n", nss_ctx, nss_runtime_samples.average, inst_cnt, nss_runtime_samples.sample_count);
nss_trace("%px: Current Frequency Index:%d\n", nss_ctx, index);
nss_trace("%px: Auto Scale Ready:%d Auto Scale:%d\n", nss_ctx, nss_runtime_samples.freq_scale_ready, nss_cmd_buf.auto_scale);
nss_trace("%px: Current Rate:%x\n", nss_ctx, nss_runtime_samples.average);
nss_runtime_samples.message_rate_limit = 0;
}
/*
* Don't scale if we are not ready or auto scale is disabled.
*/
if ((nss_runtime_samples.freq_scale_ready != 1) || (nss_cmd_buf.auto_scale != 1)) {
return;
}
/*
* Scale Algorithmn
* Algorithmn will limit how fast it will transition each scale, by the number of samples seen.
* If any sample is out of scale during the idle count, the rate_limit will reset to 0.
* Scales are limited to the max number of cpu scales we support.
*/
if (nss_runtime_samples.freq_scale_rate_limit_up++ >= NSS_FREQUENCY_SCALE_RATE_LIMIT_UP) {
maximum = nss_runtime_samples.freq_scale[index].maximum;
if ((nss_runtime_samples.average > maximum) && (index < (NSS_FREQ_MAX_SCALE - 1))) {
nss_runtime_samples.freq_scale_index++;
nss_runtime_samples.freq_scale_ready = 0;
/*
* If fail to increase frequency, decrease index
*/
nss_trace("frequency increase to %d inst:%x > maximum:%x\n", nss_runtime_samples.freq_scale[nss_runtime_samples.freq_scale_index].frequency, inst_cnt, maximum);
if (!nss_freq_queue_work()) {
nss_runtime_samples.freq_scale_index--;
}
}
/*
* Reset the down scale counter based on running average, so can idle properly
*/
if (nss_runtime_samples.average > maximum) {
nss_trace("down scale timeout reset running average:%x\n", nss_runtime_samples.average);
nss_runtime_samples.freq_scale_rate_limit_down = 0;
}
nss_runtime_samples.freq_scale_rate_limit_up = 0;
return;
}
if (nss_runtime_samples.freq_scale_rate_limit_down++ >= NSS_FREQUENCY_SCALE_RATE_LIMIT_DOWN) {
minimum = nss_runtime_samples.freq_scale[index].minimum;
/*
* Check if we need to lower the frequency. For some SoC like IPQ50xx, low frequency
* is not supported. So check if the next lower frequency is configured before shifting down
*/
if ((nss_runtime_samples.average < minimum) && (index > 0) && nss_runtime_samples.freq_scale[index - 1].maximum) {
nss_runtime_samples.freq_scale_index--;
nss_runtime_samples.freq_scale_ready = 0;
/*
* If fail to decrease frequency, increase index
*/
nss_trace("frequency decrease to %d inst:%x < minumum:%x\n", nss_runtime_samples.freq_scale[nss_runtime_samples.freq_scale_index].frequency, nss_runtime_samples.average, minimum);
if (!nss_freq_queue_work()) {
nss_runtime_samples.freq_scale_index++;
}
}
nss_runtime_samples.freq_scale_rate_limit_down = 0;
return;
}
}
/*
* nss_freq_handle_core_stats()
* Handle the core stats.
*/
static void nss_freq_handle_core_stats(struct nss_ctx_instance *nss_ctx, struct nss_core_stats *core_stats)
{
uint32_t inst_cnt = core_stats->inst_cnt_total;
/*
* compute CPU utilization by using the instruction count
*/
nss_freq_compute_cpu_usage(nss_ctx, inst_cnt);
/*
* Perform frequency scaling
*/
nss_freq_scale_frequency(nss_ctx, inst_cnt);
}
/*
* nss_freq_interface_handler()
* Handle NSS -> HLOS messages for Frequency Changes and Statistics.
*/
static void nss_freq_interface_handler(struct nss_ctx_instance *nss_ctx, struct nss_cmn_msg *ncm, __attribute__((unused))void *app_data) {
struct nss_corefreq_msg *ncfm = (struct nss_corefreq_msg *)ncm;
/*
* Trace Messages
*/
nss_freq_log_rx_msg(ncfm);
switch (ncfm->cm.type) {
case COREFREQ_METADATA_TYPE_TX_FREQ_ACK:
nss_freq_handle_ack(nss_ctx, &ncfm->msg.nfc);
break;
case COREFREQ_METADATA_TYPE_TX_CORE_STATS:
nss_freq_handle_core_stats(nss_ctx, &ncfm->msg.ncs);
break;
default:
if (ncm->response != NSS_CMN_RESPONSE_ACK) {
/*
* Check response
*/
nss_info("%px: Received response %d for type %d, interface %d", nss_ctx, ncm->response, ncm->type, ncm->interface);
}
}
}
/*
* nss_freq_change()
* NSS frequency change API.
*/
nss_tx_status_t nss_freq_change(struct nss_ctx_instance *nss_ctx, uint32_t eng, uint32_t stats_enable, uint32_t start_or_end)
{
struct nss_corefreq_msg ncm;
struct nss_freq_msg *nfc;
nss_info("%px: frequency changing to: %d\n", nss_ctx, eng);
/*
* Update the max instruction count for a frequency during down scaling.
* Better to update this as late as possible in the frequency update call.
*/
spin_lock_bh(&nss_freq_cpu_usage_lock);
nss_freq_cpu_status.max_ins = eng / 1000;
spin_unlock_bh(&nss_freq_cpu_usage_lock);
nss_freq_msg_init(&ncm, NSS_COREFREQ_INTERFACE, NSS_TX_METADATA_TYPE_NSS_FREQ_CHANGE,
sizeof(struct nss_freq_msg), NULL, NULL);
nfc = &ncm.msg.nfc;
nfc->frequency = eng;
nfc->start_or_end = start_or_end;
nfc->stats_enable = stats_enable;
return nss_core_send_cmd(nss_ctx, &ncm, sizeof(ncm), NSS_NBUF_PAYLOAD_SIZE);
}
/*
* nss_freq_sched_change()
* Schedule a frequency work.
*/
bool nss_freq_sched_change(nss_freq_scales_t index, bool auto_scale)
{
if (index >= NSS_FREQ_MAX_SCALE) {
nss_info("NSS freq scale beyond limit\n");
return false;
}
nss_work = (nss_work_t *)kmalloc(sizeof(nss_work_t), GFP_ATOMIC);
if (!nss_work) {
nss_info("NSS Freq WQ kmalloc fail");
return false;
}
INIT_WORK((struct work_struct *)nss_work, nss_hal_wq_function);
nss_work->frequency = nss_runtime_samples.freq_scale[index].frequency;
nss_work->stats_enable = auto_scale;
nss_cmd_buf.current_freq = nss_work->frequency;
queue_work(nss_wq, (struct work_struct *)nss_work);
return true;
}
/*
* nss_freq_get_context()
* Get NSS context instance for frequency.
*/
struct nss_ctx_instance *nss_freq_get_context(void)
{
return (struct nss_ctx_instance *)&nss_top_main.nss[nss_top_main.frequency_handler_id];
}
EXPORT_SYMBOL(nss_freq_get_context);
/*
* nss_freq_register_handler()
*/
void nss_freq_register_handler(void)
{
struct nss_ctx_instance *nss_ctx = nss_freq_get_context();
nss_core_register_handler(nss_ctx, NSS_COREFREQ_INTERFACE, nss_freq_interface_handler, NULL);
}
/*
* nss_freq_cpu_usage_init()
* Initialize cpu usage computing.
*
* TODO: Add support to retrieve CPU usage even if frequency scaling is disabled.
*/
void nss_freq_init_cpu_usage(void)
{
nss_freq_cpu_status.used = 0;
nss_freq_cpu_status.max_ins = nss_runtime_samples.freq_scale[nss_runtime_samples.freq_scale_index].frequency / 1000;
nss_freq_cpu_status.total = 0;
nss_freq_cpu_status.max = 0; /* Initial value is 0 to capture the highest most value during the run */
nss_freq_cpu_status.min = NSS_FREQ_CPU_USAGE_MAX; /* Initial value is 100 to capture the lowest most value during the run */
nss_freq_cpu_status.avg_up = NSS_FREQ_CPU_USAGE_MAX_BOUND;
nss_freq_cpu_status.avg_ctr = 0;
nss_freq_stats_dentry_create();
}