1 /*
2 * Copyright (c) 2018, Google 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 "runtime/handles.inline.hpp"
26 #include "runtime/sharedRuntime.hpp"
27 #include "runtime/threadHeapSampler.hpp"
28
29 // Cheap random number generator
30 uint64_t ThreadHeapSampler::_rnd;
31 // Default is 512kb.
32 int ThreadHeapSampler::_sampling_rate = 512 * 1024;
33 int ThreadHeapSampler::_enabled;
34
35 // Statics for the fast log
36 static const int FastLogNumBits = 10;
37 static const int FastLogMask = (1 << FastLogNumBits) - 1;
38 static double log_table[1<<FastLogNumBits]; // Constant
39 static bool log_table_initialized;
40
41 // Returns the next prng value.
42 // pRNG is: aX+b mod c with a = 0x5DEECE66D, b = 0xB, c = 1<<48
43 // This is the lrand64 generator.
44 static uint64_t next_random(uint64_t rnd) {
45 const uint64_t PrngMult = 0x5DEECE66DLL;
46 const uint64_t PrngAdd = 0xB;
47 const uint64_t PrngModPower = 48;
48 const uint64_t PrngModMask = right_n_bits(PrngModPower);
49 //assert(IS_SAFE_SIZE_MUL(PrngMult, rnd), "Overflow on multiplication.");
50 //assert(IS_SAFE_SIZE_ADD(PrngMult * rnd, PrngAdd), "Overflow on addition.");
51 return (PrngMult * rnd + PrngAdd) & PrngModMask;
52 }
53
54 static double fast_log2(const double & d) {
55 assert(d>0, "bad value passed to assert");
56 uint64_t x = 0;
57 assert(sizeof(d) == sizeof(x),
58 "double and uint64_t do not have the same size");
59 x = *reinterpret_cast<const uint64_t*>(&d);
60 const uint32_t x_high = x >> 32;
61 assert(FastLogNumBits <= 20, "FastLogNumBits should be less than 20.");
62 const uint32_t y = x_high >> (20 - FastLogNumBits) & FastLogMask;
63 const int32_t exponent = ((x_high >> 20) & 0x7FF) - 1023;
64 return exponent + log_table[y];
65 }
66
67 // Generates a geometric variable with the specified mean (512K by default).
68 // This is done by generating a random number between 0 and 1 and applying
69 // the inverse cumulative distribution function for an exponential.
70 // Specifically: Let m be the inverse of the sample rate, then
71 // the probability distribution function is m*exp(-mx) so the CDF is
72 // p = 1 - exp(-mx), so
73 // q = 1 - p = exp(-mx)
74 // log_e(q) = -mx
75 // -log_e(q)/m = x
76 // log_2(q) * (-log_e(2) * 1/m) = x
77 // In the code, q is actually in the range 1 to 2**26, hence the -26 below
78 void ThreadHeapSampler::pick_next_geometric_sample() {
79 _rnd = next_random(_rnd);
80 // Take the top 26 bits as the random number
81 // (This plus a 1<<58 sampling bound gives a max possible step of
82 // 5194297183973780480 bytes. In this case,
83 // for sample_parameter = 1<<19, max possible step is
84 // 9448372 bytes (24 bits).
85 const uint64_t PrngModPower = 48; // Number of bits in prng
86 // The uint32_t cast is to prevent a (hard-to-reproduce) NAN
87 // under piii debug for some binaries.
88 double q = static_cast<uint32_t>(_rnd >> (PrngModPower - 26)) + 1.0;
89 // Put the computed p-value through the CDF of a geometric.
90 // For faster performance (save ~1/20th exec time), replace
91 // min(0.0, FastLog2(q) - 26) by (Fastlog2(q) - 26.000705)
92 // The value 26.000705 is used rather than 26 to compensate
93 // for inaccuracies in FastLog2 which otherwise result in a
94 // negative answer.
95 double log_val = (fast_log2(q) - 26);
96 double result =
97 (0.0 < log_val ? 0.0 : log_val) * (-log(2.0) * (get_sampling_rate())) + 1;
98 assert(result > 0 && result < SIZE_MAX, "Result is not in an acceptable range.");
99 size_t rate = static_cast<size_t>(result);
100 _bytes_until_sample = rate;
101 }
102
103 void ThreadHeapSampler::pick_next_sample(size_t overflowed_bytes) {
104 if (get_sampling_rate() == 1) {
105 _bytes_until_sample = 1;
106 return;
107 }
108
109 pick_next_geometric_sample();
110
111 // Try to correct sample size by removing extra space from last allocation.
112 if (overflowed_bytes > 0 && _bytes_until_sample > overflowed_bytes) {
113 _bytes_until_sample -= overflowed_bytes;
114 }
115 }
116
117 void ThreadHeapSampler::check_for_sampling(HeapWord* ptr, size_t allocation_size, size_t bytes_since_allocation) {
118 oopDesc* oop = reinterpret_cast<oopDesc*>(ptr);
119 size_t total_allocated_bytes = bytes_since_allocation + allocation_size;
120
121 // If not yet time for a sample, skip it.
122 if (total_allocated_bytes < _bytes_until_sample) {
123 _bytes_until_sample -= total_allocated_bytes;
124 return;
125 }
126
127 JvmtiExport::sampled_object_alloc_event_collector(oop);
128
129 size_t overflow_bytes = total_allocated_bytes - _bytes_until_sample;
130 pick_next_sample(overflow_bytes);
131 }
132
133 void ThreadHeapSampler::init_log_table() {
134 MutexLockerEx mu(ThreadHeapSampler_lock, Mutex::_no_safepoint_check_flag);
135
136 if (log_table_initialized) {
137 return;
138 }
139
140 for (int i = 0; i < (1 << FastLogNumBits); i++) {
141 log_table[i] = (log(1.0 + static_cast<double>(i+0.5) / (1 << FastLogNumBits))
142 / log(2.0));
143 }
144
145 log_table_initialized = true;
146 }
147
148 void ThreadHeapSampler::enable() {
149 // Done here to be done when things have settled. This adds a mutex lock but
150 // presumably, users won't be enabling and disabling all the time.
151 init_log_table();
152 OrderAccess::release_store(&_enabled, 1);
153 }
154
155 // Methods used in assertion mode to check if a collector is present or not at
156 // the moment of TLAB sampling, ie a slow allocation path.
157 bool ThreadHeapSampler::sampling_collector_present() const {
158 return _collectors_present > 0;
159 }
160
161 bool ThreadHeapSampler::remove_sampling_collector() {
162 assert(_collectors_present > 0, "Problem with collector counter.");
163 _collectors_present--;
164 return true;
165 }
166
167 bool ThreadHeapSampler::add_sampling_collector() {
168 _collectors_present++;
169 return true;
170 }