| 1 | /* |
| 2 | * Copyright (c) 2013, 2018, Oracle and/or its affiliates. All rights reserved. |
| 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 4 | * |
| 5 | * This code is free software; you can redistribute it and/or modify it |
| 6 | * under the terms of the GNU General Public License version 2 only, as |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
| 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 12 | * version 2 for more details (a copy is included in the LICENSE file that |
| 13 | * accompanied this code). |
| 14 | * |
| 15 | * You should have received a copy of the GNU General Public License version |
| 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
| 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
| 18 | * |
| 19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
| 20 | * or visit www.oracle.com if you need additional information or have any |
| 21 | * questions. |
| 22 | * |
| 23 | */ |
| 24 | |
| 25 | #include "precompiled.hpp" |
| 26 | #include "rdtsc_x86.hpp" |
| 27 | #include "runtime/thread.inline.hpp" |
| 28 | #include "vm_version_ext_x86.hpp" |
| 29 | |
| 30 | // The following header contains the implementations of rdtsc() |
| 31 | #include OS_CPU_HEADER_INLINE(os) |
| 32 | |
| 33 | static jlong _epoch = 0; |
| 34 | static bool rdtsc_elapsed_counter_enabled = false; |
| 35 | static jlong tsc_frequency = 0; |
| 36 | |
| 37 | static jlong set_epoch() { |
| 38 | assert(0 == _epoch, "invariant" ); |
| 39 | _epoch = os::rdtsc(); |
| 40 | return _epoch; |
| 41 | } |
| 42 | |
| 43 | // Base loop to estimate ticks frequency for tsc counter from user mode. |
| 44 | // Volatiles and sleep() are used to prevent compiler from applying optimizations. |
| 45 | static void do_time_measurements(volatile jlong& time_base, |
| 46 | volatile jlong& time_fast, |
| 47 | volatile jlong& time_base_elapsed, |
| 48 | volatile jlong& time_fast_elapsed) { |
| 49 | static const unsigned int FT_SLEEP_MILLISECS = 1; |
| 50 | const unsigned int loopcount = 3; |
| 51 | |
| 52 | volatile jlong start = 0; |
| 53 | volatile jlong fstart = 0; |
| 54 | volatile jlong end = 0; |
| 55 | volatile jlong fend = 0; |
| 56 | |
| 57 | // Figure out the difference between rdtsc and os provided timer. |
| 58 | // base algorithm adopted from JRockit. |
| 59 | for (unsigned int times = 0; times < loopcount; times++) { |
| 60 | start = os::elapsed_counter(); |
| 61 | OrderAccess::fence(); |
| 62 | fstart = os::rdtsc(); |
| 63 | |
| 64 | // use sleep to prevent compiler from optimizing |
| 65 | os::sleep(Thread::current(), FT_SLEEP_MILLISECS, true); |
| 66 | |
| 67 | end = os::elapsed_counter(); |
| 68 | OrderAccess::fence(); |
| 69 | fend = os::rdtsc(); |
| 70 | |
| 71 | time_base += end - start; |
| 72 | time_fast += fend - fstart; |
| 73 | |
| 74 | // basis for calculating the os tick start |
| 75 | // to fast time tick start offset |
| 76 | time_base_elapsed += end; |
| 77 | time_fast_elapsed += (fend - _epoch); |
| 78 | } |
| 79 | |
| 80 | time_base /= loopcount; |
| 81 | time_fast /= loopcount; |
| 82 | time_base_elapsed /= loopcount; |
| 83 | time_fast_elapsed /= loopcount; |
| 84 | } |
| 85 | |
| 86 | static jlong initialize_frequency() { |
| 87 | assert(0 == tsc_frequency, "invariant" ); |
| 88 | assert(0 == _epoch, "invariant" ); |
| 89 | const jlong initial_counter = set_epoch(); |
| 90 | if (initial_counter == 0) { |
| 91 | return 0; |
| 92 | } |
| 93 | // os time frequency |
| 94 | static double os_freq = (double)os::elapsed_frequency(); |
| 95 | assert(os_freq > 0, "os_elapsed frequency corruption!" ); |
| 96 | |
| 97 | double tsc_freq = .0; |
| 98 | double os_to_tsc_conv_factor = 1.0; |
| 99 | |
| 100 | // if platform supports invariant tsc, |
| 101 | // apply higher resolution and granularity for conversion calculations |
| 102 | if (VM_Version_Ext::supports_tscinv_ext()) { |
| 103 | // for invariant tsc platforms, take the maximum qualified cpu frequency |
| 104 | tsc_freq = (double)VM_Version_Ext::maximum_qualified_cpu_frequency(); |
| 105 | os_to_tsc_conv_factor = tsc_freq / os_freq; |
| 106 | } else { |
| 107 | // use measurements to estimate |
| 108 | // a conversion factor and the tsc frequency |
| 109 | |
| 110 | volatile jlong time_base = 0; |
| 111 | volatile jlong time_fast = 0; |
| 112 | volatile jlong time_base_elapsed = 0; |
| 113 | volatile jlong time_fast_elapsed = 0; |
| 114 | |
| 115 | // do measurements to get base data |
| 116 | // on os timer and fast ticks tsc time relation. |
| 117 | do_time_measurements(time_base, time_fast, time_base_elapsed, time_fast_elapsed); |
| 118 | |
| 119 | // if invalid measurements, cannot proceed |
| 120 | if (time_fast == 0 || time_base == 0) { |
| 121 | return 0; |
| 122 | } |
| 123 | |
| 124 | os_to_tsc_conv_factor = (double)time_fast / (double)time_base; |
| 125 | if (os_to_tsc_conv_factor > 1) { |
| 126 | // estimate on tsc counter frequency |
| 127 | tsc_freq = os_to_tsc_conv_factor * os_freq; |
| 128 | } |
| 129 | } |
| 130 | |
| 131 | if ((tsc_freq < 0) || (tsc_freq > 0 && tsc_freq <= os_freq) || (os_to_tsc_conv_factor <= 1)) { |
| 132 | // safer to run with normal os time |
| 133 | tsc_freq = .0; |
| 134 | } |
| 135 | |
| 136 | // frequency of the tsc_counter |
| 137 | return (jlong)tsc_freq; |
| 138 | } |
| 139 | |
| 140 | static bool initialize_elapsed_counter() { |
| 141 | tsc_frequency = initialize_frequency(); |
| 142 | return tsc_frequency != 0 && _epoch != 0; |
| 143 | } |
| 144 | |
| 145 | static bool ergonomics() { |
| 146 | const bool invtsc_support = Rdtsc::is_supported(); |
| 147 | if (FLAG_IS_DEFAULT(UseFastUnorderedTimeStamps) && invtsc_support) { |
| 148 | FLAG_SET_ERGO(UseFastUnorderedTimeStamps, true); |
| 149 | } |
| 150 | |
| 151 | bool ft_enabled = UseFastUnorderedTimeStamps && invtsc_support; |
| 152 | |
| 153 | if (!ft_enabled) { |
| 154 | if (UseFastUnorderedTimeStamps && VM_Version::supports_tsc()) { |
| 155 | warning("\nThe hardware does not support invariant tsc (INVTSC) register and/or cannot guarantee tsc synchronization between sockets at startup.\n" \ |
| 156 | "Values returned via rdtsc() are not guaranteed to be accurate, esp. when comparing values from cross sockets reads. Enabling UseFastUnorderedTimeStamps on non-invariant tsc hardware should be considered experimental.\n" ); |
| 157 | ft_enabled = true; |
| 158 | } |
| 159 | } |
| 160 | |
| 161 | if (!ft_enabled) { |
| 162 | // Warn if unable to support command-line flag |
| 163 | if (UseFastUnorderedTimeStamps && !VM_Version::supports_tsc()) { |
| 164 | warning("Ignoring UseFastUnorderedTimeStamps, hardware does not support normal tsc" ); |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | return ft_enabled; |
| 169 | } |
| 170 | |
| 171 | bool Rdtsc::is_supported() { |
| 172 | return VM_Version_Ext::supports_tscinv_ext(); |
| 173 | } |
| 174 | |
| 175 | bool Rdtsc::is_elapsed_counter_enabled() { |
| 176 | return rdtsc_elapsed_counter_enabled; |
| 177 | } |
| 178 | |
| 179 | jlong Rdtsc::frequency() { |
| 180 | return tsc_frequency; |
| 181 | } |
| 182 | |
| 183 | jlong Rdtsc::elapsed_counter() { |
| 184 | return os::rdtsc() - _epoch; |
| 185 | } |
| 186 | |
| 187 | jlong Rdtsc::epoch() { |
| 188 | return _epoch; |
| 189 | } |
| 190 | |
| 191 | jlong Rdtsc::raw() { |
| 192 | return os::rdtsc(); |
| 193 | } |
| 194 | |
| 195 | bool Rdtsc::initialize() { |
| 196 | static bool initialized = false; |
| 197 | if (!initialized) { |
| 198 | assert(!rdtsc_elapsed_counter_enabled, "invariant" ); |
| 199 | VM_Version_Ext::initialize(); |
| 200 | assert(0 == tsc_frequency, "invariant" ); |
| 201 | assert(0 == _epoch, "invariant" ); |
| 202 | bool result = initialize_elapsed_counter(); // init hw |
| 203 | if (result) { |
| 204 | result = ergonomics(); // check logical state |
| 205 | } |
| 206 | rdtsc_elapsed_counter_enabled = result; |
| 207 | initialized = true; |
| 208 | } |
| 209 | return rdtsc_elapsed_counter_enabled; |
| 210 | } |
| 211 | |