init/calibrate.c

   1 /* calibrate.c: default delay calibration
   2  *
   3  * Excised from init/main.c
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/jiffies.h>
   8 #include <linux/delay.h>
   9 #include <linux/init.h>
  10 #include <linux/timex.h>
  11 #include <linux/smp.h>
  12 #include <linux/percpu.h>
  13
  14 unsigned long lpj_fine;
  15 unsigned long preset_lpj;
  16 static int __init lpj_setup(char *str)
  17 {
  18         preset_lpj = simple_strtoul(str,NULL,0);
  19         return 1;
  20 }
  21
  22 __setup("lpj=", lpj_setup);
  23
  24 #ifdef ARCH_HAS_READ_CURRENT_TIMER
  25
  26 /* This routine uses the read_current_timer() routine and gets the
  27  * loops per jiffy directly, instead of guessing it using delay().
  28  * Also, this code tries to handle non-maskable asynchronous events
  29  * (like SMIs)
  30  */
  31 #define DELAY_CALIBRATION_TICKS                 ((HZ < 100) ? 1 : (HZ/100))
  32 #define MAX_DIRECT_CALIBRATION_RETRIES          5
  33
  34 static unsigned long __cpuinit calibrate_delay_direct(void)
  35 {
  36         unsigned long pre_start, start, post_start;
  37         unsigned long pre_end, end, post_end;
  38         unsigned long start_jiffies;
  39         unsigned long timer_rate_min, timer_rate_max;
  40         unsigned long good_timer_sum = 0;
  41         unsigned long good_timer_count = 0;
  42         unsigned long measured_times[MAX_DIRECT_CALIBRATION_RETRIES];
  43         int max = -1; /* index of measured_times with max/min values or not set */
  44         int min = -1;
  45         int i;
  46
  47         if (read_current_timer(&pre_start) < 0 )
  48                 return 0;
  49
  50         /*
  51          * A simple loop like
  52          *      while ( jiffies < start_jiffies+1)
  53          *              start = read_current_timer();
  54          * will not do. As we don't really know whether jiffy switch
  55          * happened first or timer_value was read first. And some asynchronous
  56          * event can happen between these two events introducing errors in lpj.
  57          *
  58          * So, we do
  59          * 1. pre_start <- When we are sure that jiffy switch hasn't happened
  60          * 2. check jiffy switch
  61          * 3. start <- timer value before or after jiffy switch
  62          * 4. post_start <- When we are sure that jiffy switch has happened
  63          *
  64          * Note, we don't know anything about order of 2 and 3.
  65          * Now, by looking at post_start and pre_start difference, we can
  66          * check whether any asynchronous event happened or not
  67          */
  68
  69         for (i = 0; i < MAX_DIRECT_CALIBRATION_RETRIES; i++) {
  70                 pre_start = 0;
  71                 read_current_timer(&start);
  72                 start_jiffies = jiffies;
  73                 while (time_before_eq(jiffies, start_jiffies + 1)) {
  74                         pre_start = start;
  75                         read_current_timer(&start);
  76                 }
  77                 read_current_timer(&post_start);
  78
  79                 pre_end = 0;
  80                 end = post_start;
  81                 while (time_before_eq(jiffies, start_jiffies + 1 +
  82                                                DELAY_CALIBRATION_TICKS)) {
  83                         pre_end = end;
  84                         read_current_timer(&end);
  85                 }
  86                 read_current_timer(&post_end);
  87
  88                 timer_rate_max = (post_end - pre_start) /
  89                                         DELAY_CALIBRATION_TICKS;
  90                 timer_rate_min = (pre_end - post_start) /
  91                                         DELAY_CALIBRATION_TICKS;
  92
  93                 /*
  94                  * If the upper limit and lower limit of the timer_rate is
  95                  * >= 12.5% apart, redo calibration.
  96                  */
  97                 if (start >= post_end)
  98                         printk(KERN_NOTICE "calibrate_delay_direct() ignoring "
  99                                         "timer_rate as we had a TSC wrap around"
 100                                         " start=%lu >=post_end=%lu\n",
 101                                 start, post_end);
 102                 if (start < post_end && pre_start != 0 && pre_end != 0 &&
 103                     (timer_rate_max - timer_rate_min) < (timer_rate_max >> 3)) {
 104                         good_timer_count++;
 105                         good_timer_sum += timer_rate_max;
 106                         measured_times[i] = timer_rate_max;
 107                         if (max < 0 || timer_rate_max > measured_times[max])
 108                                 max = i;
 109                         if (min < 0 || timer_rate_max < measured_times[min])
 110                                 min = i;
 111                 } else
 112                         measured_times[i] = 0;
 113
 114         }
 115
 116         /*
 117          * Find the maximum & minimum - if they differ too much throw out the
 118          * one with the largest difference from the mean and try again...
 119          */
 120         while (good_timer_count > 1) {
 121                 unsigned long estimate;
 122                 unsigned long maxdiff;
 123
 124                 /* compute the estimate */
 125                 estimate = (good_timer_sum/good_timer_count);
 126                 maxdiff = estimate >> 3;
 127
 128                 /* if range is within 12% let's take it */
 129                 if ((measured_times[max] - measured_times[min]) < maxdiff)
 130                         return estimate;
 131
 132                 /* ok - drop the worse value and try again... */
 133                 good_timer_sum = 0;
 134                 good_timer_count = 0;
 135                 if ((measured_times[max] - estimate) <
 136                                 (estimate - measured_times[min])) {
 137                         printk(KERN_NOTICE "calibrate_delay_direct() dropping "
 138                                         "min bogoMips estimate %d = %lu\n",
 139                                 min, measured_times[min]);
 140                         measured_times[min] = 0;
 141                         min = max;
 142                 } else {
 143                         printk(KERN_NOTICE "calibrate_delay_direct() dropping "
 144                                         "max bogoMips estimate %d = %lu\n",
 145                                 max, measured_times[max]);
 146                         measured_times[max] = 0;
 147                         max = min;
 148                 }
 149
 150                 for (i = 0; i < MAX_DIRECT_CALIBRATION_RETRIES; i++) {
 151                         if (measured_times[i] == 0)
 152                                 continue;
 153                         good_timer_count++;
 154                         good_timer_sum += measured_times[i];
 155                         if (measured_times[i] < measured_times[min])
 156                                 min = i;
 157                         if (measured_times[i] > measured_times[max])
 158                                 max = i;
 159                 }
 160
 161         }
 162
 163         printk(KERN_NOTICE "calibrate_delay_direct() failed to get a good "
 164                "estimate for loops_per_jiffy.\nProbably due to long platform "
 165                 "interrupts. Consider using \"lpj=\" boot option.\n");
 166         return 0;
 167 }
 168 #else
 169 static unsigned long __cpuinit calibrate_delay_direct(void) {return 0;}
 170 #endif
 171
 172 /*
 173  * This is the number of bits of precision for the loops_per_jiffy.  Each
 174  * time we refine our estimate after the first takes 1.5/HZ seconds, so try
 175  * to start with a good estimate.
 176  * For the boot cpu we can skip the delay calibration and assign it a value
 177  * calculated based on the timer frequency.
 178  * For the rest of the CPUs we cannot assume that the timer frequency is same as
 179  * the cpu frequency, hence do the calibration for those.
 180  */
 181 #define LPS_PREC 8
 182
 183 static unsigned long __cpuinit calibrate_delay_converge(void)
 184 {
 185         /* First stage - slowly accelerate to find initial bounds */
 186         unsigned long lpj, lpj_base, ticks, loopadd, loopadd_base, chop_limit;
 187         int trials = 0, band = 0, trial_in_band = 0;
 188
 189         lpj = (1<<12);
 190
 191         /* wait for "start of" clock tick */
 192         ticks = jiffies;
 193         while (ticks == jiffies)
 194                 ; /* nothing */
 195         /* Go .. */
 196         ticks = jiffies;
 197         do {
 198                 if (++trial_in_band == (1<<band)) {
 199                         ++band;
 200                         trial_in_band = 0;
 201                 }
 202                 __delay(lpj * band);
 203                 trials += band;
 204         } while (ticks == jiffies);
 205         /*
 206          * We overshot, so retreat to a clear underestimate. Then estimate
 207          * the largest likely undershoot. This defines our chop bounds.
 208          */
 209         trials -= band;
 210         loopadd_base = lpj * band;
 211         lpj_base = lpj * trials;
 212
 213 recalibrate:
 214         lpj = lpj_base;
 215         loopadd = loopadd_base;
 216
 217         /*
 218          * Do a binary approximation to get lpj set to
 219          * equal one clock (up to LPS_PREC bits)
 220          */
 221         chop_limit = lpj >> LPS_PREC;
 222         while (loopadd > chop_limit) {
 223                 lpj += loopadd;
 224                 ticks = jiffies;
 225                 while (ticks == jiffies)
 226                         ; /* nothing */
 227                 ticks = jiffies;
 228                 __delay(lpj);
 229                 if (jiffies != ticks)   /* longer than 1 tick */
 230                         lpj -= loopadd;
 231                 loopadd >>= 1;
 232         }
 233         /*
 234          * If we incremented every single time possible, presume we've
 235          * massively underestimated initially, and retry with a higher
 236          * start, and larger range. (Only seen on x86_64, due to SMIs)
 237          */
 238         if (lpj + loopadd * 2 == lpj_base + loopadd_base * 2) {
 239                 lpj_base = lpj;
 240                 loopadd_base <<= 2;
 241                 goto recalibrate;
 242         }
 243
 244         return lpj;
 245 }
 246
 247 static DEFINE_PER_CPU(unsigned long, cpu_loops_per_jiffy) = { 0 };
 248
 249 void __cpuinit calibrate_delay(void)
 250 {
 251         unsigned long lpj;
 252         static bool printed;
 253         int this_cpu = smp_processor_id();
 254
 255         if (per_cpu(cpu_loops_per_jiffy, this_cpu)) {
 256                 lpj = per_cpu(cpu_loops_per_jiffy, this_cpu);
 257                 pr_info("Calibrating delay loop (skipped) "
 258                                 "already calibrated this CPU");
 259         } else if (preset_lpj) {
 260                 lpj = preset_lpj;
 261                 if (!printed)
 262                         pr_info("Calibrating delay loop (skipped) "
 263                                 "preset value.. ");
 264         } else if ((!printed) && lpj_fine) {
 265                 lpj = lpj_fine;
 266                 pr_info("Calibrating delay loop (skipped), "
 267                         "value calculated using timer frequency.. ");
 268         } else if ((lpj = calibrate_delay_direct()) != 0) {
 269                 if (!printed)
 270                         pr_info("Calibrating delay using timer "
 271                                 "specific routine.. ");
 272         } else {
 273                 if (!printed)
 274                         pr_info("Calibrating delay loop... ");
 275                 lpj = calibrate_delay_converge();
 276         }
 277         per_cpu(cpu_loops_per_jiffy, this_cpu) = lpj;
 278         if (!printed)
 279                 pr_cont("%lu.%02lu BogoMIPS (lpj=%lu)\n",
 280                         lpj/(500000/HZ),
 281                         (lpj/(5000/HZ)) % 100, lpj);
 282
 283         loops_per_jiffy = lpj;
 284         printed = true;
 285 }