calcload.c

Simple Operating Systems （简称SOS）是一个可以运行在X86平台上（包括QEMU

/* Copyright (C) 2004 David Decotigny

   This program is free software; you can redistribute it and/or
   modify it under the terms of the GNU General Public License
   as published by the Free Software Foundation; either version 2
   of the License, or (at your option) any later version.
   
   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
   
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
   USA. 
*/

#include <hwcore/irq.h>
#include <sos/kmalloc.h>
#include <sos/assert.h>
#include <sos/calcload.h>


/**
 * Multiplicative factor to display digits after the decimal dot. The
 * higher the value, the higher the precision, but the higher the risk
 * that the value you get is incorrect (integer overflow).
 *
 * The CPU ratios will be correctly displayed as long as:
 *    2^32 > (900 * HZ * 100 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR)
 * The "900" above means 900s because the load is computed over 15mn (900s).
 * HZ is the frequency of the timer tick because the load is updated
 * at each timer tick.
 * The "100" above is the multiplication factor to get the ratio value
 * between 0 and 100 (instead of 0-1).
 *
 * The maximum CPU sustainable load that will be correctly displayed
 * is given by the formula:
 *    (2^32 - 1) / (900 * HZ * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR)
 * With HZ=100, these maximum sustainable loads are respectively
 * 47.721, 477.21 and 4772.1 with
 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR being respectively 1000, 100
 * and 10.
 *
 * Hence, among these formulaes, the most limitative one is that
 * concerning the CPU ratios (because of the "100" factor). Actually,
 * for HZ=100, the only correct value is 10.
 */
#define SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR 10 /* 1/10 resolution */


/**
 * To compute the load, at each clock tick we store the number of
 * threads ready in kernel/user mode, and the kind of the thread that
 * is executing (user or kernel mode): this is stored in
 * current_load_entry. We then compute the sum of these numbers over 3
 * periods of time: 1 minute, 5 minutes, 15 minutes. This is the role
 * of the sliding windows data structures. A "sliding window" is only
 * the synthetic sum of these figures, not a real sliding window of
 * load_entries. At each timer tick and everytime the load is updated,
 * the computations are in O(1).
 *
 * All the sliding windows share the main "recorded_load" array of
 * load_entries for that; its role is to store the last 15mns of load
 * data, which encompasses the data for the 1mn, 5mn and 15mn sliding
 * windows.
 */

/* Max number of seconds that we record (ie number of entries in
   recorded_loads) */
#define NB_SECS 900


/* An entry in the longest sliding window */
struct load_entry
{
  sos_ui32_t nb_ticks;

  sos_ui32_t nb_user_running;
  sos_ui32_t nb_kernel_running;

  sos_ui32_t nb_user_ready;
  sos_ui32_t nb_kernel_ready;
};

struct load_entry current_load_entry;


/* The longest sliding window */
struct recorded_loads
{
  sos_ui32_t most_recent;
  sos_ui32_t max_entries;

  struct load_entry *load_entries;
};

#define LOAD_GET_ENTRY(loads,age) \
  (&((loads).load_entries[( (loads).max_entries + (loads).most_recent - (age))\
                          % ((loads).max_entries)]))

/* A sliding window, we manage one for each time interval */
struct sliding_window
{
  sos_ui32_t max_entries;
  sos_ui32_t nb_entries;

  sos_ui32_t sigma_nb_ticks;
  sos_ui32_t sigma_nb_user_running;
  sos_ui32_t sigma_nb_kernel_running;
  sos_ui32_t sigma_nb_user_ready;
  sos_ui32_t sigma_nb_kernel_ready;
};


/* The main sliding window */
static struct recorded_loads recorded_loads;

/* The sliding windows for 3 tims intervals: 1min, 5min, 15min */
static struct sliding_window load_1mn, load_5mn, load_15mn;

/* Forward declaration */
static struct sos_timeout_action calcload_timeout;
static void calcload_routine(struct sos_timeout_action *a);


static void _reinit_load_subsystem()
{
  memset(& recorded_loads, 0x0, sizeof(recorded_loads));
  memset(& current_load_entry, 0x0, sizeof(struct load_entry));
  memset(& load_1mn, 0x0, sizeof(load_1mn));
  memset(& load_5mn, 0x0, sizeof(load_5mn));
  memset(& load_15mn, 0x0, sizeof(load_15mn));
}


sos_ret_t sos_load_subsystem_setup(void)
{
  struct sos_time period;
  _reinit_load_subsystem();

  if (recorded_loads.load_entries)
    sos_kfree((sos_vaddr_t) recorded_loads.load_entries);
  _reinit_load_subsystem();

  /* Allocate 900 entries to store 15mn of data (because 15minutes =
     900s) */
  recorded_loads.max_entries = NB_SECS;
  recorded_loads.load_entries
    = (struct load_entry*) sos_kmalloc(NB_SECS * sizeof(struct load_entry),
				       0);
  if (! recorded_loads.load_entries)
    {
      return -SOS_ENOMEM;
    }

  /* Compute the number of entries in each sliding window */
  load_1mn.max_entries  = 60;
  load_5mn.max_entries  = 300;
  load_15mn.max_entries = 900;

  /* Program the load computation action */
  sos_time_init_action(& calcload_timeout);
  period.sec = 1; period.nanosec = 0;
  return sos_time_register_action_relative(& calcload_timeout,
					   & period,
					   calcload_routine,
					   NULL);
}


/* Shift the given sliding window to record the current_load_entry */
static void update_sliding_window(struct sliding_window *w)
{
  /*
   * Compute the value of the sum over the sliding window
   */

  /* Take the new value into account */
  w->sigma_nb_ticks          += current_load_entry.nb_ticks;
  w->sigma_nb_user_running   += current_load_entry.nb_user_running;
  w->sigma_nb_kernel_running += current_load_entry.nb_kernel_running;
  w->sigma_nb_user_ready     += current_load_entry.nb_user_ready;
  w->sigma_nb_kernel_ready   += current_load_entry.nb_kernel_ready;

  /* Remove the oldest entry, if it is going to be popped out of the
     sliding window */
  if (w->nb_entries < w->max_entries)
    {
      w->nb_entries ++;
    }
  else
    {
      struct load_entry * oldest_entry;
      oldest_entry = LOAD_GET_ENTRY(recorded_loads, w->nb_entries - 1);
      w->sigma_nb_ticks          -= oldest_entry->nb_ticks;
      w->sigma_nb_user_running   -= oldest_entry->nb_user_running;
      w->sigma_nb_kernel_running -= oldest_entry->nb_kernel_running;
      w->sigma_nb_user_ready     -= oldest_entry->nb_user_ready;
      w->sigma_nb_kernel_ready   -= oldest_entry->nb_kernel_ready;
    }
}


/* The timeout action responsible for computing the CPU load */
static void calcload_routine(struct sos_timeout_action *a)
{
  struct load_entry * new_head;
  struct sos_time delay;

  if (! recorded_loads.load_entries)
    return;

  /* Update the sliding windows */
  update_sliding_window(& load_1mn);
  update_sliding_window(& load_5mn);
  update_sliding_window(& load_15mn);

  /* Move the head of the list forward */
  recorded_loads.most_recent
    = (recorded_loads.most_recent + 1) % recorded_loads.max_entries;

  /* Update the new head */
  new_head = & recorded_loads.load_entries[recorded_loads.most_recent];
  memcpy(new_head, & current_load_entry, sizeof(current_load_entry));

  /* Reset the current load entry */
  memset(& current_load_entry, 0x0, sizeof(current_load_entry));

  /* Program next occurence of the action */
  delay.sec = 1;
  delay.nanosec = 0;
  sos_time_register_action_relative(a, & delay, calcload_routine, NULL);
}


sos_ret_t sos_load_do_timer_tick(sos_bool_t cur_is_user,
				 sos_ui32_t nb_user_ready,
				 sos_ui32_t nb_kernel_ready)
{
  sos_ui32_t flags;

  sos_disable_IRQs(flags);
  current_load_entry.nb_ticks ++;
  current_load_entry.nb_user_ready += nb_user_ready;
  current_load_entry.nb_kernel_ready += nb_kernel_ready;
  if (cur_is_user)
    current_load_entry.nb_user_running ++;
  else
    current_load_entry.nb_kernel_running ++;
  sos_restore_IRQs(flags);

  return SOS_OK;
}


void sos_load_to_string(char dest[11], sos_ui32_t load_value)
{
  sos_bool_t print0 = FALSE;
  sos_ui32_t d;

#define PUTCH(c) ({ *dest = (c); dest ++; })

  for (d = 1000000000UL ; d > 0 ; d /= 10)
    {
      sos_ui32_t digit = (load_value / d) % 10;

      if (digit > 0)
	{
	  PUTCH(digit + '0');
	  print0 = TRUE;
	}
      else if (print0)
	PUTCH('0');
      
      if (d == SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR)
	{
	  if (! print0)
	    PUTCH('0');

	  PUTCH('.');
	  print0 = TRUE;
	}
    }
  *dest = '\0';
}


void sos_load_get_uload(sos_ui32_t * _load_1mn,
			sos_ui32_t * _load_5mn,
			sos_ui32_t * _load_15mn)
{
  sos_ui32_t flags;

  if (load_1mn.sigma_nb_ticks < 1)
    return;

  sos_disable_IRQs(flags);
  *_load_1mn  = ( load_1mn.sigma_nb_user_ready
		  + load_1mn.sigma_nb_user_running)
                * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_1mn.sigma_nb_ticks;
  *_load_5mn  = ( load_5mn.sigma_nb_user_ready
		  + load_5mn.sigma_nb_user_running)
                * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_5mn.sigma_nb_ticks;
  *_load_15mn  = ( load_15mn.sigma_nb_user_ready
		   + load_15mn.sigma_nb_user_running)
                 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                 / load_15mn.sigma_nb_ticks;
  sos_restore_IRQs(flags);
}


void sos_load_get_sload(sos_ui32_t * _load_1mn,
			sos_ui32_t * _load_5mn,
			sos_ui32_t * _load_15mn)
{
  sos_ui32_t flags;

  if (load_1mn.sigma_nb_ticks < 1)
    return;

  /* The "IDLE" thread is always either ready or running by definition */
  SOS_ASSERT_FATAL(load_1mn.sigma_nb_kernel_ready
		   + load_1mn.sigma_nb_kernel_running
		   >= load_1mn.sigma_nb_ticks);

  /* Remove the IDLE thread from the load calculation */
  sos_disable_IRQs(flags);
  *_load_1mn  = ( load_1mn.sigma_nb_kernel_ready
		  + load_1mn.sigma_nb_kernel_running
		  - load_1mn.sigma_nb_ticks)
                * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_1mn.sigma_nb_ticks;
  *_load_5mn  = ( load_5mn.sigma_nb_kernel_ready
		  + load_5mn.sigma_nb_kernel_running
		  - load_5mn.sigma_nb_ticks)
                * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_5mn.sigma_nb_ticks;
  *_load_15mn  = ( load_15mn.sigma_nb_kernel_ready
		   + load_15mn.sigma_nb_kernel_running
		   - load_15mn.sigma_nb_ticks)
                 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                 / load_15mn.sigma_nb_ticks;
  sos_restore_IRQs(flags);
}


void sos_load_get_uratio(sos_ui32_t * _load_1mn,
			 sos_ui32_t * _load_5mn,
			 sos_ui32_t * _load_15mn)
{
  sos_ui32_t flags;

  if (load_1mn.sigma_nb_ticks < 1)
    return;

  sos_disable_IRQs(flags);
  *_load_1mn  = load_1mn.sigma_nb_user_running
                * 100 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_1mn.sigma_nb_ticks;
  *_load_5mn  = load_5mn.sigma_nb_user_running
                * 100 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_5mn.sigma_nb_ticks;
  *_load_15mn  = load_15mn.sigma_nb_user_running
                 * 100 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                 / load_15mn.sigma_nb_ticks;
  sos_restore_IRQs(flags);
}


void sos_load_get_sratio(sos_ui32_t * _load_1mn,
			 sos_ui32_t * _load_5mn,
			 sos_ui32_t * _load_15mn)
{
  sos_ui32_t flags;

  if (load_1mn.sigma_nb_ticks < 1)
    return;

  /* Don't remove the CPU occupation ration of the IDLE thread
     here... */
  sos_disable_IRQs(flags);
  *_load_1mn  = load_1mn.sigma_nb_kernel_running
                * 100 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_1mn.sigma_nb_ticks;
  *_load_5mn  = load_5mn.sigma_nb_kernel_running
                * 100 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                / load_5mn.sigma_nb_ticks;
  *_load_15mn  = load_15mn.sigma_nb_kernel_running
                 * 100 * SOS_LOAD_DISPLAY_MULTIPLICATION_FACTOR
                 / load_15mn.sigma_nb_ticks;
  sos_restore_IRQs(flags);
}