Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /* calibrate.c: default delay calibration
3 : *
4 : * Excised from init/main.c
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : */
7 :
8 : #include <linux/jiffies.h>
9 : #include <linux/delay.h>
10 : #include <linux/init.h>
11 : #include <linux/timex.h>
12 : #include <linux/smp.h>
13 : #include <linux/percpu.h>
14 :
15 : unsigned long lpj_fine;
16 : unsigned long preset_lpj;
17 0 : static int __init lpj_setup(char *str)
18 : {
19 0 : preset_lpj = simple_strtoul(str,NULL,0);
20 0 : return 1;
21 : }
22 :
23 : __setup("lpj=", lpj_setup);
24 :
25 : #ifdef ARCH_HAS_READ_CURRENT_TIMER
26 :
27 : /* This routine uses the read_current_timer() routine and gets the
28 : * loops per jiffy directly, instead of guessing it using delay().
29 : * Also, this code tries to handle non-maskable asynchronous events
30 : * (like SMIs)
31 : */
32 : #define DELAY_CALIBRATION_TICKS ((HZ < 100) ? 1 : (HZ/100))
33 : #define MAX_DIRECT_CALIBRATION_RETRIES 5
34 :
35 : static unsigned long calibrate_delay_direct(void)
36 : {
37 : unsigned long pre_start, start, post_start;
38 : unsigned long pre_end, end, post_end;
39 : unsigned long start_jiffies;
40 : unsigned long timer_rate_min, timer_rate_max;
41 : unsigned long good_timer_sum = 0;
42 : unsigned long good_timer_count = 0;
43 : unsigned long measured_times[MAX_DIRECT_CALIBRATION_RETRIES];
44 : int max = -1; /* index of measured_times with max/min values or not set */
45 : int min = -1;
46 : int i;
47 :
48 : if (read_current_timer(&pre_start) < 0 )
49 : return 0;
50 :
51 : /*
52 : * A simple loop like
53 : * while ( jiffies < start_jiffies+1)
54 : * start = read_current_timer();
55 : * will not do. As we don't really know whether jiffy switch
56 : * happened first or timer_value was read first. And some asynchronous
57 : * event can happen between these two events introducing errors in lpj.
58 : *
59 : * So, we do
60 : * 1. pre_start <- When we are sure that jiffy switch hasn't happened
61 : * 2. check jiffy switch
62 : * 3. start <- timer value before or after jiffy switch
63 : * 4. post_start <- When we are sure that jiffy switch has happened
64 : *
65 : * Note, we don't know anything about order of 2 and 3.
66 : * Now, by looking at post_start and pre_start difference, we can
67 : * check whether any asynchronous event happened or not
68 : */
69 :
70 : for (i = 0; i < MAX_DIRECT_CALIBRATION_RETRIES; i++) {
71 : pre_start = 0;
72 : read_current_timer(&start);
73 : start_jiffies = jiffies;
74 : while (time_before_eq(jiffies, start_jiffies + 1)) {
75 : pre_start = start;
76 : read_current_timer(&start);
77 : }
78 : read_current_timer(&post_start);
79 :
80 : pre_end = 0;
81 : end = post_start;
82 : while (time_before_eq(jiffies, start_jiffies + 1 +
83 : DELAY_CALIBRATION_TICKS)) {
84 : pre_end = end;
85 : read_current_timer(&end);
86 : }
87 : read_current_timer(&post_end);
88 :
89 : timer_rate_max = (post_end - pre_start) /
90 : DELAY_CALIBRATION_TICKS;
91 : timer_rate_min = (pre_end - post_start) /
92 : DELAY_CALIBRATION_TICKS;
93 :
94 : /*
95 : * If the upper limit and lower limit of the timer_rate is
96 : * >= 12.5% apart, redo calibration.
97 : */
98 : if (start >= post_end)
99 : printk(KERN_NOTICE "calibrate_delay_direct() ignoring "
100 : "timer_rate as we had a TSC wrap around"
101 : " start=%lu >=post_end=%lu\n",
102 : start, post_end);
103 : if (start < post_end && pre_start != 0 && pre_end != 0 &&
104 : (timer_rate_max - timer_rate_min) < (timer_rate_max >> 3)) {
105 : good_timer_count++;
106 : good_timer_sum += timer_rate_max;
107 : measured_times[i] = timer_rate_max;
108 : if (max < 0 || timer_rate_max > measured_times[max])
109 : max = i;
110 : if (min < 0 || timer_rate_max < measured_times[min])
111 : min = i;
112 : } else
113 : measured_times[i] = 0;
114 :
115 : }
116 :
117 : /*
118 : * Find the maximum & minimum - if they differ too much throw out the
119 : * one with the largest difference from the mean and try again...
120 : */
121 : while (good_timer_count > 1) {
122 : unsigned long estimate;
123 : unsigned long maxdiff;
124 :
125 : /* compute the estimate */
126 : estimate = (good_timer_sum/good_timer_count);
127 : maxdiff = estimate >> 3;
128 :
129 : /* if range is within 12% let's take it */
130 : if ((measured_times[max] - measured_times[min]) < maxdiff)
131 : return estimate;
132 :
133 : /* ok - drop the worse value and try again... */
134 : good_timer_sum = 0;
135 : good_timer_count = 0;
136 : if ((measured_times[max] - estimate) <
137 : (estimate - measured_times[min])) {
138 : printk(KERN_NOTICE "calibrate_delay_direct() dropping "
139 : "min bogoMips estimate %d = %lu\n",
140 : min, measured_times[min]);
141 : measured_times[min] = 0;
142 : min = max;
143 : } else {
144 : printk(KERN_NOTICE "calibrate_delay_direct() dropping "
145 : "max bogoMips estimate %d = %lu\n",
146 : max, measured_times[max]);
147 : measured_times[max] = 0;
148 : max = min;
149 : }
150 :
151 : for (i = 0; i < MAX_DIRECT_CALIBRATION_RETRIES; i++) {
152 : if (measured_times[i] == 0)
153 : continue;
154 : good_timer_count++;
155 : good_timer_sum += measured_times[i];
156 : if (measured_times[i] < measured_times[min])
157 : min = i;
158 : if (measured_times[i] > measured_times[max])
159 : max = i;
160 : }
161 :
162 : }
163 :
164 : printk(KERN_NOTICE "calibrate_delay_direct() failed to get a good "
165 : "estimate for loops_per_jiffy.\nProbably due to long platform "
166 : "interrupts. Consider using \"lpj=\" boot option.\n");
167 : return 0;
168 : }
169 : #else
170 : static unsigned long calibrate_delay_direct(void)
171 : {
172 : return 0;
173 : }
174 : #endif
175 :
176 : /*
177 : * This is the number of bits of precision for the loops_per_jiffy. Each
178 : * time we refine our estimate after the first takes 1.5/HZ seconds, so try
179 : * to start with a good estimate.
180 : * For the boot cpu we can skip the delay calibration and assign it a value
181 : * calculated based on the timer frequency.
182 : * For the rest of the CPUs we cannot assume that the timer frequency is same as
183 : * the cpu frequency, hence do the calibration for those.
184 : */
185 : #define LPS_PREC 8
186 :
187 1 : static unsigned long calibrate_delay_converge(void)
188 : {
189 : /* First stage - slowly accelerate to find initial bounds */
190 : unsigned long lpj, lpj_base, ticks, loopadd, loopadd_base, chop_limit;
191 1 : int trials = 0, band = 0, trial_in_band = 0;
192 :
193 1 : lpj = (1<<12);
194 :
195 : /* wait for "start of" clock tick */
196 1 : ticks = jiffies;
197 1 : while (ticks == jiffies)
198 : ; /* nothing */
199 : /* Go .. */
200 1 : ticks = jiffies;
201 : do {
202 1110 : if (++trial_in_band == (1<<band)) {
203 10 : ++band;
204 10 : trial_in_band = 0;
205 : }
206 1110 : __delay(lpj * band);
207 1110 : trials += band;
208 1110 : } while (ticks == jiffies);
209 : /*
210 : * We overshot, so retreat to a clear underestimate. Then estimate
211 : * the largest likely undershoot. This defines our chop bounds.
212 : */
213 1 : trials -= band;
214 1 : loopadd_base = lpj * band;
215 1 : lpj_base = lpj * trials;
216 :
217 : recalibrate:
218 2 : lpj = lpj_base;
219 2 : loopadd = loopadd_base;
220 :
221 : /*
222 : * Do a binary approximation to get lpj set to
223 : * equal one clock (up to LPS_PREC bits)
224 : */
225 2 : chop_limit = lpj >> LPS_PREC;
226 5 : while (loopadd > chop_limit) {
227 1 : lpj += loopadd;
228 1 : ticks = jiffies;
229 1 : while (ticks == jiffies)
230 : ; /* nothing */
231 1 : ticks = jiffies;
232 1 : __delay(lpj);
233 1 : if (jiffies != ticks) /* longer than 1 tick */
234 1 : lpj -= loopadd;
235 1 : loopadd >>= 1;
236 : }
237 : /*
238 : * If we incremented every single time possible, presume we've
239 : * massively underestimated initially, and retry with a higher
240 : * start, and larger range. (Only seen on x86_64, due to SMIs)
241 : */
242 2 : if (lpj + loopadd * 2 == lpj_base + loopadd_base * 2) {
243 1 : lpj_base = lpj;
244 1 : loopadd_base <<= 2;
245 1 : goto recalibrate;
246 : }
247 :
248 1 : return lpj;
249 : }
250 :
251 : static DEFINE_PER_CPU(unsigned long, cpu_loops_per_jiffy) = { 0 };
252 :
253 : /*
254 : * Check if cpu calibration delay is already known. For example,
255 : * some processors with multi-core sockets may have all cores
256 : * with the same calibration delay.
257 : *
258 : * Architectures should override this function if a faster calibration
259 : * method is available.
260 : */
261 1 : unsigned long __attribute__((weak)) calibrate_delay_is_known(void)
262 : {
263 1 : return 0;
264 : }
265 :
266 : /*
267 : * Indicate the cpu delay calibration is done. This can be used by
268 : * architectures to stop accepting delay timer registrations after this point.
269 : */
270 :
271 1 : void __attribute__((weak)) calibration_delay_done(void)
272 : {
273 1 : }
274 :
275 1 : void calibrate_delay(void)
276 : {
277 : unsigned long lpj;
278 : static bool printed;
279 1 : int this_cpu = smp_processor_id();
280 :
281 1 : if (per_cpu(cpu_loops_per_jiffy, this_cpu)) {
282 0 : lpj = per_cpu(cpu_loops_per_jiffy, this_cpu);
283 0 : if (!printed)
284 0 : pr_info("Calibrating delay loop (skipped) "
285 : "already calibrated this CPU");
286 1 : } else if (preset_lpj) {
287 0 : lpj = preset_lpj;
288 0 : if (!printed)
289 0 : pr_info("Calibrating delay loop (skipped) "
290 : "preset value.. ");
291 1 : } else if ((!printed) && lpj_fine) {
292 0 : lpj = lpj_fine;
293 0 : pr_info("Calibrating delay loop (skipped), "
294 : "value calculated using timer frequency.. ");
295 1 : } else if ((lpj = calibrate_delay_is_known())) {
296 : ;
297 1 : } else if ((lpj = calibrate_delay_direct()) != 0) {
298 : if (!printed)
299 : pr_info("Calibrating delay using timer "
300 : "specific routine.. ");
301 : } else {
302 1 : if (!printed)
303 1 : pr_info("Calibrating delay loop... ");
304 1 : lpj = calibrate_delay_converge();
305 : }
306 1 : per_cpu(cpu_loops_per_jiffy, this_cpu) = lpj;
307 1 : if (!printed)
308 1 : pr_cont("%lu.%02lu BogoMIPS (lpj=%lu)\n",
309 : lpj/(500000/HZ),
310 : (lpj/(5000/HZ)) % 100, lpj);
311 :
312 1 : loops_per_jiffy = lpj;
313 1 : printed = true;
314 :
315 1 : calibration_delay_done();
316 1 : }
|
