Move thread_monitor() to monitor.c

pull/506/head
jsteube 8 years ago
parent 1ce58f51d1
commit dcdd0d89cd

@ -0,0 +1,12 @@
/**
* Author......: See docs/credits.txt
* License.....: MIT
*/
#ifndef _DISPATCH_H
#define _DISPATCH_H
void *thread_calc_stdin (void *p);
void *thread_calc (void *p);
#endif // _DISPATCH_H

@ -0,0 +1,11 @@
/**
* Author......: See docs/credits.txt
* License.....: MIT
*/
#ifndef _MONITOR_H
#define _MONITOR_H
void *thread_monitor (void *p);
#endif // _MONITOR_H

@ -163,7 +163,7 @@ LFLAGS_CROSS_WIN := -lpsapi
## Objects
##
OBJS_ALL := affinity attack_mode autotune benchmark bitmap bitops common convert cpu_aes cpu_crc32 cpu_des cpu_md5 cpu_sha1 cpu_sha256 data debugfile dictstat dispatch dynloader ext_ADL ext_nvapi ext_nvml ext_OpenCL ext_xnvctrl filehandling filenames folder hash_management hlfmt hwmon induct interface locking logfile logging loopback memory mpsp opencl outfile_check outfile potfile powertune remove restore rp_cpu rp_kernel_on_cpu runtime shared status stdout terminal thread timer tuningdb usage version weak_hash wordlist
OBJS_ALL := affinity attack_mode autotune benchmark bitmap bitops common convert cpu_aes cpu_crc32 cpu_des cpu_md5 cpu_sha1 cpu_sha256 data debugfile dictstat dispatch dynloader ext_ADL ext_nvapi ext_nvml ext_OpenCL ext_xnvctrl filehandling filenames folder hash_management hlfmt hwmon induct interface locking logfile logging loopback memory monitor mpsp opencl outfile_check outfile potfile powertune remove restore rp_cpu rp_kernel_on_cpu runtime shared status stdout terminal thread timer tuningdb usage version weak_hash wordlist
NATIVE_OBJS := $(foreach OBJ,$(OBJS_ALL),obj/$(OBJ).NATIVE.o)

@ -0,0 +1,536 @@
/**
* Author......: See docs/credits.txt
* License.....: MIT
*/
#include "common.h"
#include "types_int.h"
#include "types.h"
#include "memory.h"
#include "filehandling.h"
#include "interface.h"
#include "timer.h"
#include "logging.h"
#include "ext_OpenCL.h"
#include "ext_ADL.h"
#include "ext_nvapi.h"
#include "ext_nvml.h"
#include "ext_xnvctrl.h"
#include "opencl.h"
#include "thread.h"
#include "rp_cpu.h"
#include "terminal.h"
#include "hwmon.h"
#include "mpsp.h"
#include "restore.h"
#include "outfile.h"
#include "potfile.h"
#include "debugfile.h"
#include "loopback.h"
#include "status.h"
#include "dictstat.h"
#include "wordlist.h"
#include "data.h"
#include "status.h"
#include "shared.h"
#include "dispatch.h"
extern hc_global_data_t data;
extern hc_thread_mutex_t mux_counter;
hc_thread_mutex_t mux_dispatcher;
static void set_kernel_power_final (const u64 kernel_power_final)
{
if (data.quiet == 0)
{
clear_prompt ();
//log_info ("");
log_info ("INFO: approaching final keyspace, workload adjusted");
log_info ("");
send_prompt ();
}
data.kernel_power_final = kernel_power_final;
}
static u32 get_power (hc_device_param_t *device_param)
{
const u64 kernel_power_final = data.kernel_power_final;
if (kernel_power_final)
{
const double device_factor = (double) device_param->hardware_power / data.hardware_power_all;
const u64 words_left_device = (u64) CEIL (kernel_power_final * device_factor);
// work should be at least the hardware power available without any accelerator
const u64 work = MAX (words_left_device, device_param->hardware_power);
return work;
}
return device_param->kernel_power;
}
static uint get_work (hc_device_param_t *device_param, const u64 max)
{
hc_thread_mutex_lock (mux_dispatcher);
const u64 words_cur = data.words_cur;
const u64 words_base = (data.limit == 0) ? data.words_base : MIN (data.limit, data.words_base);
device_param->words_off = words_cur;
const u64 kernel_power_all = data.kernel_power_all;
const u64 words_left = words_base - words_cur;
if (words_left < kernel_power_all)
{
if (data.kernel_power_final == 0)
{
set_kernel_power_final (words_left);
}
}
const u32 kernel_power = get_power (device_param);
uint work = MIN (words_left, kernel_power);
work = MIN (work, max);
data.words_cur += work;
hc_thread_mutex_unlock (mux_dispatcher);
return work;
}
void *thread_calc_stdin (void *p)
{
hc_device_param_t *device_param = (hc_device_param_t *) p;
if (device_param->skipped) return NULL;
hashconfig_t *hashconfig = data.hashconfig;
char *buf = (char *) mymalloc (HCBUFSIZ_LARGE);
const uint attack_kern = data.attack_kern;
while ((data.devices_status != STATUS_CRACKED) && (data.devices_status != STATUS_ABORTED) && (data.devices_status != STATUS_QUIT))
{
hc_thread_mutex_lock (mux_dispatcher);
if (feof (stdin) != 0)
{
hc_thread_mutex_unlock (mux_dispatcher);
break;
}
uint words_cur = 0;
while (words_cur < device_param->kernel_power)
{
char *line_buf = fgets (buf, HCBUFSIZ_LARGE - 1, stdin);
if (line_buf == NULL) break;
uint line_len = in_superchop (line_buf);
line_len = convert_from_hex (line_buf, line_len);
// post-process rule engine
if (run_rule_engine (data.rule_len_l, data.rule_buf_l))
{
char rule_buf_out[BLOCK_SIZE] = { 0 };
int rule_len_out = -1;
if (line_len < BLOCK_SIZE)
{
rule_len_out = _old_apply_rule (data.rule_buf_l, data.rule_len_l, line_buf, line_len, rule_buf_out);
}
if (rule_len_out < 0) continue;
line_buf = rule_buf_out;
line_len = rule_len_out;
}
if (line_len > PW_MAX)
{
continue;
}
// hmm that's always the case, or?
if (attack_kern == ATTACK_KERN_STRAIGHT)
{
if ((line_len < data.pw_min) || (line_len > data.pw_max))
{
hc_thread_mutex_lock (mux_counter);
for (uint salt_pos = 0; salt_pos < data.salts_cnt; salt_pos++)
{
data.words_progress_rejected[salt_pos] += data.kernel_rules_cnt;
}
hc_thread_mutex_unlock (mux_counter);
continue;
}
}
pw_add (device_param, (u8 *) line_buf, line_len);
words_cur++;
if (data.devices_status == STATUS_CRACKED) break;
if (data.devices_status == STATUS_ABORTED) break;
if (data.devices_status == STATUS_QUIT) break;
if (data.devices_status == STATUS_BYPASS) break;
}
hc_thread_mutex_unlock (mux_dispatcher);
if (data.devices_status == STATUS_CRACKED) break;
if (data.devices_status == STATUS_ABORTED) break;
if (data.devices_status == STATUS_QUIT) break;
if (data.devices_status == STATUS_BYPASS) break;
// flush
const uint pws_cnt = device_param->pws_cnt;
if (pws_cnt)
{
run_copy (device_param, hashconfig, pws_cnt);
run_cracker (device_param, hashconfig, pws_cnt);
device_param->pws_cnt = 0;
/*
still required?
if (attack_kern == ATTACK_KERN_STRAIGHT)
{
run_kernel_bzero (device_param, device_param->d_rules_c, device_param->size_rules_c);
}
else if (attack_kern == ATTACK_KERN_COMBI)
{
run_kernel_bzero (device_param, device_param->d_combs_c, device_param->size_combs);
}
*/
}
}
device_param->kernel_accel = 0;
device_param->kernel_loops = 0;
myfree (buf);
return NULL;
}
void *thread_calc (void *p)
{
hc_device_param_t *device_param = (hc_device_param_t *) p;
if (device_param->skipped) return NULL;
hashconfig_t *hashconfig = data.hashconfig;
const uint attack_mode = data.attack_mode;
const uint attack_kern = data.attack_kern;
if (attack_mode == ATTACK_MODE_BF)
{
while ((data.devices_status != STATUS_CRACKED) && (data.devices_status != STATUS_ABORTED) && (data.devices_status != STATUS_QUIT))
{
const uint work = get_work (device_param, -1u);
if (work == 0) break;
const u64 words_off = device_param->words_off;
const u64 words_fin = words_off + work;
const uint pws_cnt = work;
device_param->pws_cnt = pws_cnt;
if (pws_cnt)
{
run_copy (device_param, hashconfig, pws_cnt);
run_cracker (device_param, hashconfig, pws_cnt);
device_param->pws_cnt = 0;
/*
still required?
run_kernel_bzero (device_param, device_param->d_bfs_c, device_param->size_bfs);
*/
}
if (data.devices_status == STATUS_STOP_AT_CHECKPOINT) check_checkpoint ();
if (data.devices_status == STATUS_CRACKED) break;
if (data.devices_status == STATUS_ABORTED) break;
if (data.devices_status == STATUS_QUIT) break;
if (data.devices_status == STATUS_BYPASS) break;
if (data.benchmark == 1) break;
device_param->words_done = words_fin;
}
}
else
{
const uint segment_size = data.segment_size;
char *dictfile = data.dictfile;
if (attack_mode == ATTACK_MODE_COMBI)
{
if (data.combs_mode == COMBINATOR_MODE_BASE_RIGHT)
{
dictfile = data.dictfile2;
}
}
FILE *fd = fopen (dictfile, "rb");
if (fd == NULL)
{
log_error ("ERROR: %s: %s", dictfile, strerror (errno));
return NULL;
}
if (attack_mode == ATTACK_MODE_COMBI)
{
const uint combs_mode = data.combs_mode;
if (combs_mode == COMBINATOR_MODE_BASE_LEFT)
{
const char *dictfilec = data.dictfile2;
FILE *combs_fp = fopen (dictfilec, "rb");
if (combs_fp == NULL)
{
log_error ("ERROR: %s: %s", dictfilec, strerror (errno));
fclose (fd);
return NULL;
}
device_param->combs_fp = combs_fp;
}
else if (combs_mode == COMBINATOR_MODE_BASE_RIGHT)
{
const char *dictfilec = data.dictfile;
FILE *combs_fp = fopen (dictfilec, "rb");
if (combs_fp == NULL)
{
log_error ("ERROR: %s: %s", dictfilec, strerror (errno));
fclose (fd);
return NULL;
}
device_param->combs_fp = combs_fp;
}
}
wl_data_t *wl_data = (wl_data_t *) mymalloc (sizeof (wl_data_t));
wl_data->buf = (char *) mymalloc (segment_size);
wl_data->avail = segment_size;
wl_data->incr = segment_size;
wl_data->cnt = 0;
wl_data->pos = 0;
u64 words_cur = 0;
while ((data.devices_status != STATUS_CRACKED) && (data.devices_status != STATUS_ABORTED) && (data.devices_status != STATUS_QUIT))
{
u64 words_off = 0;
u64 words_fin = 0;
u64 max = -1llu;
while (max)
{
const uint work = get_work (device_param, max);
if (work == 0) break;
max = 0;
words_off = device_param->words_off;
words_fin = words_off + work;
char *line_buf;
uint line_len;
for ( ; words_cur < words_off; words_cur++) get_next_word (wl_data, fd, &line_buf, &line_len);
for ( ; words_cur < words_fin; words_cur++)
{
get_next_word (wl_data, fd, &line_buf, &line_len);
line_len = convert_from_hex (line_buf, line_len);
// post-process rule engine
if (run_rule_engine (data.rule_len_l, data.rule_buf_l))
{
char rule_buf_out[BLOCK_SIZE] = { 0 };
int rule_len_out = -1;
if (line_len < BLOCK_SIZE)
{
rule_len_out = _old_apply_rule (data.rule_buf_l, data.rule_len_l, line_buf, line_len, rule_buf_out);
}
if (rule_len_out < 0) continue;
line_buf = rule_buf_out;
line_len = rule_len_out;
}
if (attack_kern == ATTACK_KERN_STRAIGHT)
{
if ((line_len < data.pw_min) || (line_len > data.pw_max))
{
max++;
hc_thread_mutex_lock (mux_counter);
for (uint salt_pos = 0; salt_pos < data.salts_cnt; salt_pos++)
{
data.words_progress_rejected[salt_pos] += data.kernel_rules_cnt;
}
hc_thread_mutex_unlock (mux_counter);
continue;
}
}
else if (attack_kern == ATTACK_KERN_COMBI)
{
// do not check if minimum restriction is satisfied (line_len >= data.pw_min) here
// since we still need to combine the plains
if (line_len > data.pw_max)
{
max++;
hc_thread_mutex_lock (mux_counter);
for (uint salt_pos = 0; salt_pos < data.salts_cnt; salt_pos++)
{
data.words_progress_rejected[salt_pos] += data.combs_cnt;
}
hc_thread_mutex_unlock (mux_counter);
continue;
}
}
pw_add (device_param, (u8 *) line_buf, line_len);
if (data.devices_status == STATUS_STOP_AT_CHECKPOINT) check_checkpoint ();
if (data.devices_status == STATUS_CRACKED) break;
if (data.devices_status == STATUS_ABORTED) break;
if (data.devices_status == STATUS_QUIT) break;
if (data.devices_status == STATUS_BYPASS) break;
}
if (data.devices_status == STATUS_STOP_AT_CHECKPOINT) check_checkpoint ();
if (data.devices_status == STATUS_CRACKED) break;
if (data.devices_status == STATUS_ABORTED) break;
if (data.devices_status == STATUS_QUIT) break;
if (data.devices_status == STATUS_BYPASS) break;
}
if (data.devices_status == STATUS_STOP_AT_CHECKPOINT) check_checkpoint ();
if (data.devices_status == STATUS_CRACKED) break;
if (data.devices_status == STATUS_ABORTED) break;
if (data.devices_status == STATUS_QUIT) break;
if (data.devices_status == STATUS_BYPASS) break;
//
// flush
//
const uint pws_cnt = device_param->pws_cnt;
if (pws_cnt)
{
run_copy (device_param, hashconfig, pws_cnt);
run_cracker (device_param, hashconfig, pws_cnt);
device_param->pws_cnt = 0;
/*
still required?
if (attack_kern == ATTACK_KERN_STRAIGHT)
{
run_kernel_bzero (device_param, device_param->d_rules_c, device_param->size_rules_c);
}
else if (attack_kern == ATTACK_KERN_COMBI)
{
run_kernel_bzero (device_param, device_param->d_combs_c, device_param->size_combs);
}
*/
}
if (data.devices_status == STATUS_STOP_AT_CHECKPOINT) check_checkpoint ();
if (data.devices_status == STATUS_CRACKED) break;
if (data.devices_status == STATUS_ABORTED) break;
if (data.devices_status == STATUS_QUIT) break;
if (data.devices_status == STATUS_BYPASS) break;
if (words_fin == 0) break;
device_param->words_done = words_fin;
}
if (attack_mode == ATTACK_MODE_COMBI)
{
fclose (device_param->combs_fp);
}
free (wl_data->buf);
free (wl_data);
fclose (fd);
}
device_param->kernel_accel = 0;
device_param->kernel_loops = 0;
return NULL;
}

@ -84,6 +84,7 @@
#include "autotune.h"
#include "induct.h"
#include "dispatch.h"
#include "monitor.h"
extern hc_global_data_t data;
@ -108,362 +109,6 @@ const int comptime = COMPTIME;
#define FORCE 0
static void *thread_monitor (void *p)
{
uint runtime_check = 0;
uint remove_check = 0;
uint status_check = 0;
uint restore_check = 0;
uint restore_left = data.restore_timer;
uint remove_left = data.remove_timer;
uint status_left = data.status_timer;
#if defined (HAVE_HWMON)
uint hwmon_check = 0;
int slowdown_warnings = 0;
// these variables are mainly used for fan control
int *fan_speed_chgd = (int *) mycalloc (data.devices_cnt, sizeof (int));
// temperature controller "loopback" values
int *temp_diff_old = (int *) mycalloc (data.devices_cnt, sizeof (int));
int *temp_diff_sum = (int *) mycalloc (data.devices_cnt, sizeof (int));
int temp_threshold = 1; // degrees celcius
int fan_speed_min = 15; // in percentage
int fan_speed_max = 100;
time_t last_temp_check_time;
#endif // HAVE_HWMON
uint sleep_time = 1;
if (data.runtime)
{
runtime_check = 1;
}
if (data.restore_timer)
{
restore_check = 1;
}
if ((data.remove == 1) && (data.hashlist_mode == HL_MODE_FILE))
{
remove_check = 1;
}
if (data.status == 1)
{
status_check = 1;
}
#if defined (HAVE_HWMON)
if (data.gpu_temp_disable == 0)
{
time (&last_temp_check_time);
hwmon_check = 1;
}
#endif
if ((runtime_check == 0) && (remove_check == 0) && (status_check == 0) && (restore_check == 0))
{
#if defined (HAVE_HWMON)
if (hwmon_check == 0)
#endif
return (p);
}
while (data.shutdown_inner == 0)
{
hc_sleep (sleep_time);
if (data.devices_status != STATUS_RUNNING) continue;
#if defined (HAVE_HWMON)
if (hwmon_check == 1)
{
hc_thread_mutex_lock (mux_hwmon);
for (uint device_id = 0; device_id < data.devices_cnt; device_id++)
{
hc_device_param_t *device_param = &data.devices_param[device_id];
if (device_param->skipped) continue;
if (device_param->device_vendor_id == VENDOR_ID_NV)
{
if (data.hm_nvapi)
{
NV_GPU_PERF_POLICIES_INFO_PARAMS_V1 perfPolicies_info;
NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1 perfPolicies_status;
memset (&perfPolicies_info, 0, sizeof (NV_GPU_PERF_POLICIES_INFO_PARAMS_V1));
memset (&perfPolicies_status, 0, sizeof (NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1));
perfPolicies_info.version = MAKE_NVAPI_VERSION (NV_GPU_PERF_POLICIES_INFO_PARAMS_V1, 1);
perfPolicies_status.version = MAKE_NVAPI_VERSION (NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1, 1);
hm_NvAPI_GPU_GetPerfPoliciesInfo (data.hm_nvapi, data.hm_device[device_id].nvapi, &perfPolicies_info);
perfPolicies_status.info_value = perfPolicies_info.info_value;
hm_NvAPI_GPU_GetPerfPoliciesStatus (data.hm_nvapi, data.hm_device[device_id].nvapi, &perfPolicies_status);
if (perfPolicies_status.throttle & 2)
{
if (slowdown_warnings < 3)
{
if (data.quiet == 0) clear_prompt ();
log_info ("WARNING: Drivers temperature threshold hit on GPU #%d, expect performance to drop...", device_id + 1);
if (slowdown_warnings == 2)
{
log_info ("");
}
if (data.quiet == 0) send_prompt ();
slowdown_warnings++;
}
}
else
{
slowdown_warnings = 0;
}
}
}
}
hc_thread_mutex_unlock (mux_hwmon);
}
if (hwmon_check == 1)
{
hc_thread_mutex_lock (mux_hwmon);
time_t temp_check_time;
time (&temp_check_time);
uint Ta = temp_check_time - last_temp_check_time; // set Ta = sleep_time; is not good enough (see --remove etc)
if (Ta == 0) Ta = 1;
for (uint device_id = 0; device_id < data.devices_cnt; device_id++)
{
hc_device_param_t *device_param = &data.devices_param[device_id];
if (device_param->skipped) continue;
if ((data.devices_param[device_id].device_type & CL_DEVICE_TYPE_GPU) == 0) continue;
const int temperature = hm_get_temperature_with_device_id (device_id);
if (temperature > (int) data.gpu_temp_abort)
{
log_error ("ERROR: Temperature limit on GPU %d reached, aborting...", device_id + 1);
if (data.devices_status != STATUS_QUIT) myabort ();
break;
}
const int gpu_temp_retain = data.gpu_temp_retain;
if (gpu_temp_retain)
{
if (data.hm_device[device_id].fan_set_supported == 1)
{
int temp_cur = temperature;
int temp_diff_new = gpu_temp_retain - temp_cur;
temp_diff_sum[device_id] = temp_diff_sum[device_id] + temp_diff_new;
// calculate Ta value (time difference in seconds between the last check and this check)
last_temp_check_time = temp_check_time;
float Kp = 1.8f;
float Ki = 0.005f;
float Kd = 6;
// PID controller (3-term controller: proportional - Kp, integral - Ki, derivative - Kd)
int fan_diff_required = (int) (Kp * (float)temp_diff_new + Ki * Ta * (float)temp_diff_sum[device_id] + Kd * ((float)(temp_diff_new - temp_diff_old[device_id])) / Ta);
if (abs (fan_diff_required) >= temp_threshold)
{
const int fan_speed_cur = hm_get_fanspeed_with_device_id (device_id);
int fan_speed_level = fan_speed_cur;
if (fan_speed_chgd[device_id] == 0) fan_speed_level = temp_cur;
int fan_speed_new = fan_speed_level - fan_diff_required;
if (fan_speed_new > fan_speed_max) fan_speed_new = fan_speed_max;
if (fan_speed_new < fan_speed_min) fan_speed_new = fan_speed_min;
if (fan_speed_new != fan_speed_cur)
{
int freely_change_fan_speed = (fan_speed_chgd[device_id] == 1);
int fan_speed_must_change = (fan_speed_new > fan_speed_cur);
if ((freely_change_fan_speed == 1) || (fan_speed_must_change == 1))
{
if (device_param->device_vendor_id == VENDOR_ID_AMD)
{
hm_set_fanspeed_with_device_id_adl (device_id, fan_speed_new, 1);
}
else if (device_param->device_vendor_id == VENDOR_ID_NV)
{
#if defined (_WIN)
hm_set_fanspeed_with_device_id_nvapi (device_id, fan_speed_new, 1);
#endif
#if defined (__linux__)
hm_set_fanspeed_with_device_id_xnvctrl (device_id, fan_speed_new);
#endif
}
fan_speed_chgd[device_id] = 1;
}
temp_diff_old[device_id] = temp_diff_new;
}
}
}
}
}
hc_thread_mutex_unlock (mux_hwmon);
}
#endif // HAVE_HWMON
if (restore_check == 1)
{
restore_left--;
if (restore_left == 0)
{
if (data.restore_disable == 0) cycle_restore ();
restore_left = data.restore_timer;
}
}
if ((runtime_check == 1) && (data.runtime_start > 0))
{
double ms_paused = data.ms_paused;
if (data.devices_status == STATUS_PAUSED)
{
double ms_paused_tmp = 0;
hc_timer_get (data.timer_paused, ms_paused_tmp);
ms_paused += ms_paused_tmp;
}
time_t runtime_cur;
time (&runtime_cur);
int runtime_left = data.proc_start + data.runtime + data.prepare_time + (ms_paused / 1000) - runtime_cur;
if (runtime_left <= 0)
{
if (data.benchmark == 0)
{
if (data.quiet == 0) log_info ("\nNOTE: Runtime limit reached, aborting...\n");
}
if (data.devices_status != STATUS_QUIT) myabort ();
}
}
if (remove_check == 1)
{
remove_left--;
if (remove_left == 0)
{
if (data.digests_saved != data.digests_done)
{
data.digests_saved = data.digests_done;
save_hash ();
}
remove_left = data.remove_timer;
}
}
if (status_check == 1)
{
status_left--;
if (status_left == 0)
{
hc_thread_mutex_lock (mux_display);
if (data.quiet == 0) clear_prompt ();
if (data.quiet == 0) log_info ("");
status_display ();
if (data.quiet == 0) log_info ("");
hc_thread_mutex_unlock (mux_display);
status_left = data.status_timer;
}
}
}
#if defined (HAVE_HWMON)
myfree (fan_speed_chgd);
myfree (temp_diff_old);
myfree (temp_diff_sum);
#endif
p = NULL;
return (p);
}
int main (int argc, char **argv)
{
#if defined (_WIN)

@ -0,0 +1,374 @@
/**
* Author......: See docs/credits.txt
* License.....: MIT
*/
#include "common.h"
#include "types_int.h"
#include "types.h"
#include "logging.h"
#include "memory.h"
#include "interface.h"
#include "timer.h"
#include "ext_OpenCL.h"
#include "ext_ADL.h"
#include "ext_nvapi.h"
#include "ext_nvml.h"
#include "ext_xnvctrl.h"
#include "hwmon.h"
#include "mpsp.h"
#include "rp_cpu.h"
#include "restore.h"
#include "opencl.h"
#include "outfile.h"
#include "potfile.h"
#include "debugfile.h"
#include "loopback.h"
#include "data.h"
#include "status.h"
#include "shared.h"
#include "terminal.h"
#include "hwmon.h"
#include "thread.h"
#include "monitor.h"
extern hc_global_data_t data;
extern hc_thread_mutex_t mux_display;
extern hc_thread_mutex_t mux_hwmon;
void *thread_monitor (void *p)
{
uint runtime_check = 0;
uint remove_check = 0;
uint status_check = 0;
uint restore_check = 0;
uint restore_left = data.restore_timer;
uint remove_left = data.remove_timer;
uint status_left = data.status_timer;
#if defined (HAVE_HWMON)
uint hwmon_check = 0;
int slowdown_warnings = 0;
// these variables are mainly used for fan control
int *fan_speed_chgd = (int *) mycalloc (data.devices_cnt, sizeof (int));
// temperature controller "loopback" values
int *temp_diff_old = (int *) mycalloc (data.devices_cnt, sizeof (int));
int *temp_diff_sum = (int *) mycalloc (data.devices_cnt, sizeof (int));
int temp_threshold = 1; // degrees celcius
int fan_speed_min = 15; // in percentage
int fan_speed_max = 100;
time_t last_temp_check_time;
#endif // HAVE_HWMON
uint sleep_time = 1;
if (data.runtime)
{
runtime_check = 1;
}
if (data.restore_timer)
{
restore_check = 1;
}
if ((data.remove == 1) && (data.hashlist_mode == HL_MODE_FILE))
{
remove_check = 1;
}
if (data.status == 1)
{
status_check = 1;
}
#if defined (HAVE_HWMON)
if (data.gpu_temp_disable == 0)
{
time (&last_temp_check_time);
hwmon_check = 1;
}
#endif
if ((runtime_check == 0) && (remove_check == 0) && (status_check == 0) && (restore_check == 0))
{
#if defined (HAVE_HWMON)
if (hwmon_check == 0)
#endif
return (p);
}
while (data.shutdown_inner == 0)
{
hc_sleep (sleep_time);
if (data.devices_status != STATUS_RUNNING) continue;
#if defined (HAVE_HWMON)
if (hwmon_check == 1)
{
hc_thread_mutex_lock (mux_hwmon);
for (uint device_id = 0; device_id < data.devices_cnt; device_id++)
{
hc_device_param_t *device_param = &data.devices_param[device_id];
if (device_param->skipped) continue;
if (device_param->device_vendor_id == VENDOR_ID_NV)
{
if (data.hm_nvapi)
{
NV_GPU_PERF_POLICIES_INFO_PARAMS_V1 perfPolicies_info;
NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1 perfPolicies_status;
memset (&perfPolicies_info, 0, sizeof (NV_GPU_PERF_POLICIES_INFO_PARAMS_V1));
memset (&perfPolicies_status, 0, sizeof (NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1));
perfPolicies_info.version = MAKE_NVAPI_VERSION (NV_GPU_PERF_POLICIES_INFO_PARAMS_V1, 1);
perfPolicies_status.version = MAKE_NVAPI_VERSION (NV_GPU_PERF_POLICIES_STATUS_PARAMS_V1, 1);
hm_NvAPI_GPU_GetPerfPoliciesInfo (data.hm_nvapi, data.hm_device[device_id].nvapi, &perfPolicies_info);
perfPolicies_status.info_value = perfPolicies_info.info_value;
hm_NvAPI_GPU_GetPerfPoliciesStatus (data.hm_nvapi, data.hm_device[device_id].nvapi, &perfPolicies_status);
if (perfPolicies_status.throttle & 2)
{
if (slowdown_warnings < 3)
{
if (data.quiet == 0) clear_prompt ();
log_info ("WARNING: Drivers temperature threshold hit on GPU #%d, expect performance to drop...", device_id + 1);
if (slowdown_warnings == 2)
{
log_info ("");
}
if (data.quiet == 0) send_prompt ();
slowdown_warnings++;
}
}
else
{
slowdown_warnings = 0;
}
}
}
}
hc_thread_mutex_unlock (mux_hwmon);
}
if (hwmon_check == 1)
{
hc_thread_mutex_lock (mux_hwmon);
time_t temp_check_time;
time (&temp_check_time);
uint Ta = temp_check_time - last_temp_check_time; // set Ta = sleep_time; is not good enough (see --remove etc)
if (Ta == 0) Ta = 1;
for (uint device_id = 0; device_id < data.devices_cnt; device_id++)
{
hc_device_param_t *device_param = &data.devices_param[device_id];
if (device_param->skipped) continue;
if ((data.devices_param[device_id].device_type & CL_DEVICE_TYPE_GPU) == 0) continue;
const int temperature = hm_get_temperature_with_device_id (device_id);
if (temperature > (int) data.gpu_temp_abort)
{
log_error ("ERROR: Temperature limit on GPU %d reached, aborting...", device_id + 1);
if (data.devices_status != STATUS_QUIT) myabort ();
break;
}
const int gpu_temp_retain = data.gpu_temp_retain;
if (gpu_temp_retain)
{
if (data.hm_device[device_id].fan_set_supported == 1)
{
int temp_cur = temperature;
int temp_diff_new = gpu_temp_retain - temp_cur;
temp_diff_sum[device_id] = temp_diff_sum[device_id] + temp_diff_new;
// calculate Ta value (time difference in seconds between the last check and this check)
last_temp_check_time = temp_check_time;
float Kp = 1.8f;
float Ki = 0.005f;
float Kd = 6;
// PID controller (3-term controller: proportional - Kp, integral - Ki, derivative - Kd)
int fan_diff_required = (int) (Kp * (float)temp_diff_new + Ki * Ta * (float)temp_diff_sum[device_id] + Kd * ((float)(temp_diff_new - temp_diff_old[device_id])) / Ta);
if (abs (fan_diff_required) >= temp_threshold)
{
const int fan_speed_cur = hm_get_fanspeed_with_device_id (device_id);
int fan_speed_level = fan_speed_cur;
if (fan_speed_chgd[device_id] == 0) fan_speed_level = temp_cur;
int fan_speed_new = fan_speed_level - fan_diff_required;
if (fan_speed_new > fan_speed_max) fan_speed_new = fan_speed_max;
if (fan_speed_new < fan_speed_min) fan_speed_new = fan_speed_min;
if (fan_speed_new != fan_speed_cur)
{
int freely_change_fan_speed = (fan_speed_chgd[device_id] == 1);
int fan_speed_must_change = (fan_speed_new > fan_speed_cur);
if ((freely_change_fan_speed == 1) || (fan_speed_must_change == 1))
{
if (device_param->device_vendor_id == VENDOR_ID_AMD)
{
hm_set_fanspeed_with_device_id_adl (device_id, fan_speed_new, 1);
}
else if (device_param->device_vendor_id == VENDOR_ID_NV)
{
#if defined (_WIN)
hm_set_fanspeed_with_device_id_nvapi (device_id, fan_speed_new, 1);
#endif
#if defined (__linux__)
hm_set_fanspeed_with_device_id_xnvctrl (device_id, fan_speed_new);
#endif
}
fan_speed_chgd[device_id] = 1;
}
temp_diff_old[device_id] = temp_diff_new;
}
}
}
}
}
hc_thread_mutex_unlock (mux_hwmon);
}
#endif // HAVE_HWMON
if (restore_check == 1)
{
restore_left--;
if (restore_left == 0)
{
if (data.restore_disable == 0) cycle_restore ();
restore_left = data.restore_timer;
}
}
if ((runtime_check == 1) && (data.runtime_start > 0))
{
double ms_paused = data.ms_paused;
if (data.devices_status == STATUS_PAUSED)
{
double ms_paused_tmp = 0;
hc_timer_get (data.timer_paused, ms_paused_tmp);
ms_paused += ms_paused_tmp;
}
time_t runtime_cur;
time (&runtime_cur);
int runtime_left = data.proc_start + data.runtime + data.prepare_time + (ms_paused / 1000) - runtime_cur;
if (runtime_left <= 0)
{
if (data.benchmark == 0)
{
if (data.quiet == 0) log_info ("\nNOTE: Runtime limit reached, aborting...\n");
}
if (data.devices_status != STATUS_QUIT) myabort ();
}
}
if (remove_check == 1)
{
remove_left--;
if (remove_left == 0)
{
if (data.digests_saved != data.digests_done)
{
data.digests_saved = data.digests_done;
save_hash ();
}
remove_left = data.remove_timer;
}
}
if (status_check == 1)
{
status_left--;
if (status_left == 0)
{
hc_thread_mutex_lock (mux_display);
if (data.quiet == 0) clear_prompt ();
if (data.quiet == 0) log_info ("");
status_display ();
if (data.quiet == 0) log_info ("");
hc_thread_mutex_unlock (mux_display);
status_left = data.status_timer;
}
}
}
#if defined (HAVE_HWMON)
myfree (fan_speed_chgd);
myfree (temp_diff_old);
myfree (temp_diff_sum);
#endif
p = NULL;
return (p);
}
Loading…
Cancel
Save