diff options
Diffstat (limited to 'arch/x86/kernel/tsc_64.c')
| -rw-r--r-- | arch/x86/kernel/tsc_64.c | 357 |
1 files changed, 0 insertions, 357 deletions
diff --git a/arch/x86/kernel/tsc_64.c b/arch/x86/kernel/tsc_64.c deleted file mode 100644 index 1784b8077a12..000000000000 --- a/arch/x86/kernel/tsc_64.c +++ /dev/null | |||
| @@ -1,357 +0,0 @@ | |||
| 1 | #include <linux/kernel.h> | ||
| 2 | #include <linux/sched.h> | ||
| 3 | #include <linux/interrupt.h> | ||
| 4 | #include <linux/init.h> | ||
| 5 | #include <linux/clocksource.h> | ||
| 6 | #include <linux/time.h> | ||
| 7 | #include <linux/acpi.h> | ||
| 8 | #include <linux/cpufreq.h> | ||
| 9 | #include <linux/acpi_pmtmr.h> | ||
| 10 | |||
| 11 | #include <asm/hpet.h> | ||
| 12 | #include <asm/timex.h> | ||
| 13 | #include <asm/timer.h> | ||
| 14 | #include <asm/vgtod.h> | ||
| 15 | |||
| 16 | static int notsc __initdata = 0; | ||
| 17 | |||
| 18 | unsigned int cpu_khz; /* TSC clocks / usec, not used here */ | ||
| 19 | EXPORT_SYMBOL(cpu_khz); | ||
| 20 | unsigned int tsc_khz; | ||
| 21 | EXPORT_SYMBOL(tsc_khz); | ||
| 22 | |||
| 23 | /* Accelerators for sched_clock() | ||
| 24 | * convert from cycles(64bits) => nanoseconds (64bits) | ||
| 25 | * basic equation: | ||
| 26 | * ns = cycles / (freq / ns_per_sec) | ||
| 27 | * ns = cycles * (ns_per_sec / freq) | ||
| 28 | * ns = cycles * (10^9 / (cpu_khz * 10^3)) | ||
| 29 | * ns = cycles * (10^6 / cpu_khz) | ||
| 30 | * | ||
| 31 | * Then we use scaling math (suggested by george@mvista.com) to get: | ||
| 32 | * ns = cycles * (10^6 * SC / cpu_khz) / SC | ||
| 33 | * ns = cycles * cyc2ns_scale / SC | ||
| 34 | * | ||
| 35 | * And since SC is a constant power of two, we can convert the div | ||
| 36 | * into a shift. | ||
| 37 | * | ||
| 38 | * We can use khz divisor instead of mhz to keep a better precision, since | ||
| 39 | * cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits. | ||
| 40 | * (mathieu.desnoyers@polymtl.ca) | ||
| 41 | * | ||
| 42 | * -johnstul@us.ibm.com "math is hard, lets go shopping!" | ||
| 43 | */ | ||
| 44 | DEFINE_PER_CPU(unsigned long, cyc2ns); | ||
| 45 | |||
| 46 | static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu) | ||
| 47 | { | ||
| 48 | unsigned long long tsc_now, ns_now; | ||
| 49 | unsigned long flags, *scale; | ||
| 50 | |||
| 51 | local_irq_save(flags); | ||
| 52 | sched_clock_idle_sleep_event(); | ||
| 53 | |||
| 54 | scale = &per_cpu(cyc2ns, cpu); | ||
| 55 | |||
| 56 | rdtscll(tsc_now); | ||
| 57 | ns_now = __cycles_2_ns(tsc_now); | ||
| 58 | |||
| 59 | if (cpu_khz) | ||
| 60 | *scale = (NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR)/cpu_khz; | ||
| 61 | |||
| 62 | sched_clock_idle_wakeup_event(0); | ||
| 63 | local_irq_restore(flags); | ||
| 64 | } | ||
| 65 | |||
| 66 | unsigned long long native_sched_clock(void) | ||
| 67 | { | ||
| 68 | unsigned long a = 0; | ||
| 69 | |||
| 70 | /* Could do CPU core sync here. Opteron can execute rdtsc speculatively, | ||
| 71 | * which means it is not completely exact and may not be monotonous | ||
| 72 | * between CPUs. But the errors should be too small to matter for | ||
| 73 | * scheduling purposes. | ||
| 74 | */ | ||
| 75 | |||
| 76 | rdtscll(a); | ||
| 77 | return cycles_2_ns(a); | ||
| 78 | } | ||
| 79 | |||
| 80 | /* We need to define a real function for sched_clock, to override the | ||
| 81 | weak default version */ | ||
| 82 | #ifdef CONFIG_PARAVIRT | ||
| 83 | unsigned long long sched_clock(void) | ||
| 84 | { | ||
| 85 | return paravirt_sched_clock(); | ||
| 86 | } | ||
| 87 | #else | ||
| 88 | unsigned long long | ||
| 89 | sched_clock(void) __attribute__((alias("native_sched_clock"))); | ||
| 90 | #endif | ||
| 91 | |||
| 92 | |||
| 93 | static int tsc_unstable; | ||
| 94 | |||
| 95 | int check_tsc_unstable(void) | ||
| 96 | { | ||
| 97 | return tsc_unstable; | ||
| 98 | } | ||
| 99 | EXPORT_SYMBOL_GPL(check_tsc_unstable); | ||
| 100 | |||
| 101 | #ifdef CONFIG_CPU_FREQ | ||
| 102 | |||
| 103 | /* Frequency scaling support. Adjust the TSC based timer when the cpu frequency | ||
| 104 | * changes. | ||
| 105 | * | ||
| 106 | * RED-PEN: On SMP we assume all CPUs run with the same frequency. It's | ||
| 107 | * not that important because current Opteron setups do not support | ||
| 108 | * scaling on SMP anyroads. | ||
| 109 | * | ||
| 110 | * Should fix up last_tsc too. Currently gettimeofday in the | ||
| 111 | * first tick after the change will be slightly wrong. | ||
| 112 | */ | ||
| 113 | |||
| 114 | static unsigned int ref_freq; | ||
| 115 | static unsigned long loops_per_jiffy_ref; | ||
| 116 | static unsigned long tsc_khz_ref; | ||
| 117 | |||
| 118 | static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | ||
| 119 | void *data) | ||
| 120 | { | ||
| 121 | struct cpufreq_freqs *freq = data; | ||
| 122 | unsigned long *lpj, dummy; | ||
| 123 | |||
| 124 | if (cpu_has(&cpu_data(freq->cpu), X86_FEATURE_CONSTANT_TSC)) | ||
| 125 | return 0; | ||
| 126 | |||
| 127 | lpj = &dummy; | ||
| 128 | if (!(freq->flags & CPUFREQ_CONST_LOOPS)) | ||
| 129 | #ifdef CONFIG_SMP | ||
| 130 | lpj = &cpu_data(freq->cpu).loops_per_jiffy; | ||
| 131 | #else | ||
| 132 | lpj = &boot_cpu_data.loops_per_jiffy; | ||
| 133 | #endif | ||
| 134 | |||
| 135 | if (!ref_freq) { | ||
| 136 | ref_freq = freq->old; | ||
| 137 | loops_per_jiffy_ref = *lpj; | ||
| 138 | tsc_khz_ref = tsc_khz; | ||
| 139 | } | ||
| 140 | if ((val == CPUFREQ_PRECHANGE && freq->old < freq->new) || | ||
| 141 | (val == CPUFREQ_POSTCHANGE && freq->old > freq->new) || | ||
| 142 | (val == CPUFREQ_RESUMECHANGE)) { | ||
| 143 | *lpj = | ||
| 144 | cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new); | ||
| 145 | |||
| 146 | tsc_khz = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new); | ||
| 147 | if (!(freq->flags & CPUFREQ_CONST_LOOPS)) | ||
| 148 | mark_tsc_unstable("cpufreq changes"); | ||
| 149 | } | ||
| 150 | |||
| 151 | set_cyc2ns_scale(tsc_khz_ref, freq->cpu); | ||
| 152 | |||
| 153 | return 0; | ||
| 154 | } | ||
| 155 | |||
| 156 | static struct notifier_block time_cpufreq_notifier_block = { | ||
| 157 | .notifier_call = time_cpufreq_notifier | ||
| 158 | }; | ||
| 159 | |||
| 160 | static int __init cpufreq_tsc(void) | ||
| 161 | { | ||
| 162 | cpufreq_register_notifier(&time_cpufreq_notifier_block, | ||
| 163 | CPUFREQ_TRANSITION_NOTIFIER); | ||
| 164 | return 0; | ||
| 165 | } | ||
| 166 | |||
| 167 | core_initcall(cpufreq_tsc); | ||
| 168 | |||
| 169 | #endif | ||
| 170 | |||
| 171 | #define MAX_RETRIES 5 | ||
| 172 | #define SMI_TRESHOLD 50000 | ||
| 173 | |||
| 174 | /* | ||
| 175 | * Read TSC and the reference counters. Take care of SMI disturbance | ||
| 176 | */ | ||
| 177 | static unsigned long __init tsc_read_refs(unsigned long *pm, | ||
| 178 | unsigned long *hpet) | ||
| 179 | { | ||
| 180 | unsigned long t1, t2; | ||
| 181 | int i; | ||
| 182 | |||
| 183 | for (i = 0; i < MAX_RETRIES; i++) { | ||
| 184 | t1 = get_cycles(); | ||
| 185 | if (hpet) | ||
| 186 | *hpet = hpet_readl(HPET_COUNTER) & 0xFFFFFFFF; | ||
| 187 | else | ||
| 188 | *pm = acpi_pm_read_early(); | ||
| 189 | t2 = get_cycles(); | ||
| 190 | if ((t2 - t1) < SMI_TRESHOLD) | ||
| 191 | return t2; | ||
| 192 | } | ||
| 193 | return ULONG_MAX; | ||
| 194 | } | ||
| 195 | |||
| 196 | /** | ||
| 197 | * tsc_calibrate - calibrate the tsc on boot | ||
| 198 | */ | ||
| 199 | void __init tsc_calibrate(void) | ||
| 200 | { | ||
| 201 | unsigned long flags, tsc1, tsc2, tr1, tr2, pm1, pm2, hpet1, hpet2; | ||
| 202 | int hpet = is_hpet_enabled(), cpu; | ||
| 203 | |||
| 204 | local_irq_save(flags); | ||
| 205 | |||
| 206 | tsc1 = tsc_read_refs(&pm1, hpet ? &hpet1 : NULL); | ||
| 207 | |||
| 208 | outb((inb(0x61) & ~0x02) | 0x01, 0x61); | ||
| 209 | |||
| 210 | outb(0xb0, 0x43); | ||
| 211 | outb((CLOCK_TICK_RATE / (1000 / 50)) & 0xff, 0x42); | ||
| 212 | outb((CLOCK_TICK_RATE / (1000 / 50)) >> 8, 0x42); | ||
| 213 | tr1 = get_cycles(); | ||
| 214 | while ((inb(0x61) & 0x20) == 0); | ||
| 215 | tr2 = get_cycles(); | ||
| 216 | |||
| 217 | tsc2 = tsc_read_refs(&pm2, hpet ? &hpet2 : NULL); | ||
| 218 | |||
| 219 | local_irq_restore(flags); | ||
| 220 | |||
| 221 | /* | ||
| 222 | * Preset the result with the raw and inaccurate PIT | ||
| 223 | * calibration value | ||
| 224 | */ | ||
| 225 | tsc_khz = (tr2 - tr1) / 50; | ||
| 226 | |||
| 227 | /* hpet or pmtimer available ? */ | ||
| 228 | if (!hpet && !pm1 && !pm2) { | ||
| 229 | printk(KERN_INFO "TSC calibrated against PIT\n"); | ||
| 230 | goto out; | ||
| 231 | } | ||
| 232 | |||
| 233 | /* Check, whether the sampling was disturbed by an SMI */ | ||
| 234 | if (tsc1 == ULONG_MAX || tsc2 == ULONG_MAX) { | ||
| 235 | printk(KERN_WARNING "TSC calibration disturbed by SMI, " | ||
| 236 | "using PIT calibration result\n"); | ||
| 237 | goto out; | ||
| 238 | } | ||
| 239 | |||
| 240 | tsc2 = (tsc2 - tsc1) * 1000000L; | ||
| 241 | |||
| 242 | if (hpet) { | ||
| 243 | printk(KERN_INFO "TSC calibrated against HPET\n"); | ||
| 244 | if (hpet2 < hpet1) | ||
| 245 | hpet2 += 0x100000000; | ||
| 246 | hpet2 -= hpet1; | ||
| 247 | tsc1 = (hpet2 * hpet_readl(HPET_PERIOD)) / 1000000; | ||
| 248 | } else { | ||
| 249 | printk(KERN_INFO "TSC calibrated against PM_TIMER\n"); | ||
| 250 | if (pm2 < pm1) | ||
| 251 | pm2 += ACPI_PM_OVRRUN; | ||
| 252 | pm2 -= pm1; | ||
| 253 | tsc1 = (pm2 * 1000000000) / PMTMR_TICKS_PER_SEC; | ||
| 254 | } | ||
| 255 | |||
| 256 | tsc_khz = tsc2 / tsc1; | ||
| 257 | |||
| 258 | out: | ||
| 259 | for_each_possible_cpu(cpu) | ||
| 260 | set_cyc2ns_scale(tsc_khz, cpu); | ||
| 261 | } | ||
| 262 | |||
| 263 | /* | ||
| 264 | * Make an educated guess if the TSC is trustworthy and synchronized | ||
| 265 | * over all CPUs. | ||
| 266 | */ | ||
| 267 | __cpuinit int unsynchronized_tsc(void) | ||
| 268 | { | ||
| 269 | if (tsc_unstable) | ||
| 270 | return 1; | ||
| 271 | |||
| 272 | #ifdef CONFIG_SMP | ||
| 273 | if (apic_is_clustered_box()) | ||
| 274 | return 1; | ||
| 275 | #endif | ||
| 276 | |||
| 277 | if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | ||
| 278 | return 0; | ||
| 279 | |||
| 280 | /* Assume multi socket systems are not synchronized */ | ||
| 281 | return num_present_cpus() > 1; | ||
| 282 | } | ||
| 283 | |||
| 284 | int __init notsc_setup(char *s) | ||
| 285 | { | ||
| 286 | notsc = 1; | ||
| 287 | return 1; | ||
| 288 | } | ||
| 289 | |||
| 290 | __setup("notsc", notsc_setup); | ||
| 291 | |||
| 292 | static struct clocksource clocksource_tsc; | ||
| 293 | |||
| 294 | /* | ||
| 295 | * We compare the TSC to the cycle_last value in the clocksource | ||
| 296 | * structure to avoid a nasty time-warp. This can be observed in a | ||
| 297 | * very small window right after one CPU updated cycle_last under | ||
| 298 | * xtime/vsyscall_gtod lock and the other CPU reads a TSC value which | ||
| 299 | * is smaller than the cycle_last reference value due to a TSC which | ||
| 300 | * is slighty behind. This delta is nowhere else observable, but in | ||
| 301 | * that case it results in a forward time jump in the range of hours | ||
| 302 | * due to the unsigned delta calculation of the time keeping core | ||
| 303 | * code, which is necessary to support wrapping clocksources like pm | ||
| 304 | * timer. | ||
| 305 | */ | ||
| 306 | static cycle_t read_tsc(void) | ||
| 307 | { | ||
| 308 | cycle_t ret = (cycle_t)get_cycles(); | ||
| 309 | |||
| 310 | return ret >= clocksource_tsc.cycle_last ? | ||
| 311 | ret : clocksource_tsc.cycle_last; | ||
| 312 | } | ||
| 313 | |||
| 314 | static cycle_t __vsyscall_fn vread_tsc(void) | ||
| 315 | { | ||
| 316 | cycle_t ret = (cycle_t)vget_cycles(); | ||
| 317 | |||
| 318 | return ret >= __vsyscall_gtod_data.clock.cycle_last ? | ||
| 319 | ret : __vsyscall_gtod_data.clock.cycle_last; | ||
| 320 | } | ||
| 321 | |||
| 322 | static struct clocksource clocksource_tsc = { | ||
| 323 | .name = "tsc", | ||
| 324 | .rating = 300, | ||
| 325 | .read = read_tsc, | ||
| 326 | .mask = CLOCKSOURCE_MASK(64), | ||
| 327 | .shift = 22, | ||
| 328 | .flags = CLOCK_SOURCE_IS_CONTINUOUS | | ||
| 329 | CLOCK_SOURCE_MUST_VERIFY, | ||
| 330 | .vread = vread_tsc, | ||
| 331 | }; | ||
| 332 | |||
| 333 | void mark_tsc_unstable(char *reason) | ||
| 334 | { | ||
| 335 | if (!tsc_unstable) { | ||
| 336 | tsc_unstable = 1; | ||
| 337 | printk("Marking TSC unstable due to %s\n", reason); | ||
| 338 | /* Change only the rating, when not registered */ | ||
| 339 | if (clocksource_tsc.mult) | ||
| 340 | clocksource_change_rating(&clocksource_tsc, 0); | ||
| 341 | else | ||
| 342 | clocksource_tsc.rating = 0; | ||
| 343 | } | ||
| 344 | } | ||
| 345 | EXPORT_SYMBOL_GPL(mark_tsc_unstable); | ||
| 346 | |||
| 347 | void __init init_tsc_clocksource(void) | ||
| 348 | { | ||
| 349 | if (!notsc) { | ||
| 350 | clocksource_tsc.mult = clocksource_khz2mult(tsc_khz, | ||
| 351 | clocksource_tsc.shift); | ||
| 352 | if (check_tsc_unstable()) | ||
| 353 | clocksource_tsc.rating = 0; | ||
| 354 | |||
| 355 | clocksource_register(&clocksource_tsc); | ||
| 356 | } | ||
| 357 | } | ||
