1# SPDX-License-Identifier: GPL-2.0-only 2 3config HAVE_ARCH_KCSAN 4 bool 5 6config HAVE_KCSAN_COMPILER 7 def_bool (CC_IS_CLANG && $(cc-option,-fsanitize=thread -mllvm -tsan-distinguish-volatile=1)) || \ 8 (CC_IS_GCC && $(cc-option,-fsanitize=thread --param tsan-distinguish-volatile=1)) 9 help 10 For the list of compilers that support KCSAN, please see 11 <file:Documentation/dev-tools/kcsan.rst>. 12 13config KCSAN_KCOV_BROKEN 14 def_bool KCOV && CC_HAS_SANCOV_TRACE_PC 15 depends on CC_IS_CLANG 16 depends on !$(cc-option,-Werror=unused-command-line-argument -fsanitize=thread -fsanitize-coverage=trace-pc) 17 help 18 Some versions of clang support either KCSAN and KCOV but not the 19 combination of the two. 20 See https://bugs.llvm.org/show_bug.cgi?id=45831 for the status 21 in newer releases. 22 23menuconfig KCSAN 24 bool "KCSAN: dynamic data race detector" 25 depends on HAVE_ARCH_KCSAN && HAVE_KCSAN_COMPILER 26 depends on DEBUG_KERNEL && !KASAN 27 depends on !KCSAN_KCOV_BROKEN 28 select STACKTRACE 29 help 30 The Kernel Concurrency Sanitizer (KCSAN) is a dynamic 31 data-race detector that relies on compile-time instrumentation. 32 KCSAN uses a watchpoint-based sampling approach to detect races. 33 34 While KCSAN's primary purpose is to detect data races, it 35 also provides assertions to check data access constraints. 36 These assertions can expose bugs that do not manifest as 37 data races. 38 39 See <file:Documentation/dev-tools/kcsan.rst> for more details. 40 41if KCSAN 42 43config CC_HAS_TSAN_COMPOUND_READ_BEFORE_WRITE 44 def_bool (CC_IS_CLANG && $(cc-option,-fsanitize=thread -mllvm -tsan-compound-read-before-write=1)) || \ 45 (CC_IS_GCC && $(cc-option,-fsanitize=thread --param tsan-compound-read-before-write=1)) 46 help 47 The compiler instruments plain compound read-write operations 48 differently (++, --, +=, -=, |=, &=, etc.), which allows KCSAN to 49 distinguish them from other plain accesses. This is currently 50 supported by Clang 12 or later. 51 52config KCSAN_VERBOSE 53 bool "Show verbose reports with more information about system state" 54 depends on PROVE_LOCKING 55 help 56 If enabled, reports show more information about the system state that 57 may help better analyze and debug races. This includes held locks and 58 IRQ trace events. 59 60 While this option should generally be benign, we call into more 61 external functions on report generation; if a race report is 62 generated from any one of them, system stability may suffer due to 63 deadlocks or recursion. If in doubt, say N. 64 65config KCSAN_SELFTEST 66 bool "Perform short selftests on boot" 67 default y 68 help 69 Run KCSAN selftests on boot. On test failure, causes the kernel to 70 panic. Recommended to be enabled, ensuring critical functionality 71 works as intended. 72 73config KCSAN_KUNIT_TEST 74 tristate "KCSAN test for integrated runtime behaviour" if !KUNIT_ALL_TESTS 75 default KUNIT_ALL_TESTS 76 depends on TRACEPOINTS && KUNIT 77 select TORTURE_TEST 78 help 79 KCSAN test focusing on behaviour of the integrated runtime. Tests 80 various race scenarios, and verifies the reports generated to 81 console. Makes use of KUnit for test organization, and the Torture 82 framework for test thread control. 83 84 Each test case may run at least up to KCSAN_REPORT_ONCE_IN_MS 85 milliseconds. Test run duration may be optimized by building the 86 kernel and KCSAN test with KCSAN_REPORT_ONCE_IN_MS set to a lower 87 than default value. 88 89 Say Y here if you want the test to be built into the kernel and run 90 during boot; say M if you want the test to build as a module; say N 91 if you are unsure. 92 93config KCSAN_EARLY_ENABLE 94 bool "Early enable during boot" 95 default y 96 help 97 If KCSAN should be enabled globally as soon as possible. KCSAN can 98 later be enabled/disabled via debugfs. 99 100config KCSAN_NUM_WATCHPOINTS 101 int "Number of available watchpoints" 102 default 64 103 help 104 Total number of available watchpoints. An address range maps into a 105 specific watchpoint slot as specified in kernel/kcsan/encoding.h. 106 Although larger number of watchpoints may not be usable due to 107 limited number of CPUs, a larger value helps to improve performance 108 due to reducing cache-line contention. The chosen default is a 109 conservative value; we should almost never observe "no_capacity" 110 events (see /sys/kernel/debug/kcsan). 111 112config KCSAN_UDELAY_TASK 113 int "Delay in microseconds (for tasks)" 114 default 80 115 help 116 For tasks, the microsecond delay after setting up a watchpoint. 117 118config KCSAN_UDELAY_INTERRUPT 119 int "Delay in microseconds (for interrupts)" 120 default 20 121 help 122 For interrupts, the microsecond delay after setting up a watchpoint. 123 Interrupts have tighter latency requirements, and their delay should 124 be lower than for tasks. 125 126config KCSAN_DELAY_RANDOMIZE 127 bool "Randomize above delays" 128 default y 129 help 130 If delays should be randomized, where the maximum is KCSAN_UDELAY_*. 131 If false, the chosen delays are always the KCSAN_UDELAY_* values 132 as defined above. 133 134config KCSAN_SKIP_WATCH 135 int "Skip instructions before setting up watchpoint" 136 default 4000 137 help 138 The number of per-CPU memory operations to skip, before another 139 watchpoint is set up, i.e. one in KCSAN_WATCH_SKIP per-CPU 140 memory operations are used to set up a watchpoint. A smaller value 141 results in more aggressive race detection, whereas a larger value 142 improves system performance at the cost of missing some races. 143 144config KCSAN_SKIP_WATCH_RANDOMIZE 145 bool "Randomize watchpoint instruction skip count" 146 default y 147 help 148 If instruction skip count should be randomized, where the maximum is 149 KCSAN_WATCH_SKIP. If false, the chosen value is always 150 KCSAN_WATCH_SKIP. 151 152config KCSAN_INTERRUPT_WATCHER 153 bool "Interruptible watchers" if !KCSAN_STRICT 154 default KCSAN_STRICT 155 help 156 If enabled, a task that set up a watchpoint may be interrupted while 157 delayed. This option will allow KCSAN to detect races between 158 interrupted tasks and other threads of execution on the same CPU. 159 160 Currently disabled by default, because not all safe per-CPU access 161 primitives and patterns may be accounted for, and therefore could 162 result in false positives. 163 164config KCSAN_REPORT_ONCE_IN_MS 165 int "Duration in milliseconds, in which any given race is only reported once" 166 default 3000 167 help 168 Any given race is only reported once in the defined time window. 169 Different races may still generate reports within a duration that is 170 smaller than the duration defined here. This allows rate limiting 171 reporting to avoid flooding the console with reports. Setting this 172 to 0 disables rate limiting. 173 174# The main purpose of the below options is to control reported data races, and 175# are not expected to be switched frequently by non-testers or at runtime. 176# The defaults are chosen to be conservative, and can miss certain bugs. 177 178config KCSAN_REPORT_RACE_UNKNOWN_ORIGIN 179 bool "Report races of unknown origin" 180 default y 181 help 182 If KCSAN should report races where only one access is known, and the 183 conflicting access is of unknown origin. This type of race is 184 reported if it was only possible to infer a race due to a data value 185 change while an access is being delayed on a watchpoint. 186 187config KCSAN_STRICT 188 bool "Strict data-race checking" 189 help 190 KCSAN will report data races with the strictest possible rules, which 191 closely aligns with the rules defined by the Linux-kernel memory 192 consistency model (LKMM). 193 194config KCSAN_WEAK_MEMORY 195 bool "Enable weak memory modeling to detect missing memory barriers" 196 default y 197 depends on KCSAN_STRICT 198 # We can either let objtool nop __tsan_func_{entry,exit}() and builtin 199 # atomics instrumentation in .noinstr.text, or use a compiler that can 200 # implement __no_kcsan to really remove all instrumentation. 201 depends on STACK_VALIDATION || CC_IS_GCC || CLANG_VERSION >= 140000 202 help 203 Enable support for modeling a subset of weak memory, which allows 204 detecting a subset of data races due to missing memory barriers. 205 206 Depends on KCSAN_STRICT, because the options strenghtening certain 207 plain accesses by default (depending on !KCSAN_STRICT) reduce the 208 ability to detect any data races invoving reordered accesses, in 209 particular reordered writes. 210 211 Weak memory modeling relies on additional instrumentation and may 212 affect performance. 213 214config KCSAN_REPORT_VALUE_CHANGE_ONLY 215 bool "Only report races where watcher observed a data value change" 216 default y 217 depends on !KCSAN_STRICT 218 help 219 If enabled and a conflicting write is observed via a watchpoint, but 220 the data value of the memory location was observed to remain 221 unchanged, do not report the data race. 222 223config KCSAN_ASSUME_PLAIN_WRITES_ATOMIC 224 bool "Assume that plain aligned writes up to word size are atomic" 225 default y 226 depends on !KCSAN_STRICT 227 help 228 Assume that plain aligned writes up to word size are atomic by 229 default, and also not subject to other unsafe compiler optimizations 230 resulting in data races. This will cause KCSAN to not report data 231 races due to conflicts where the only plain accesses are aligned 232 writes up to word size: conflicts between marked reads and plain 233 aligned writes up to word size will not be reported as data races; 234 notice that data races between two conflicting plain aligned writes 235 will also not be reported. 236 237config KCSAN_IGNORE_ATOMICS 238 bool "Do not instrument marked atomic accesses" 239 depends on !KCSAN_STRICT 240 help 241 Never instrument marked atomic accesses. This option can be used for 242 additional filtering. Conflicting marked atomic reads and plain 243 writes will never be reported as a data race, however, will cause 244 plain reads and marked writes to result in "unknown origin" reports. 245 If combined with CONFIG_KCSAN_REPORT_RACE_UNKNOWN_ORIGIN=n, data 246 races where at least one access is marked atomic will never be 247 reported. 248 249 Similar to KCSAN_ASSUME_PLAIN_WRITES_ATOMIC, but including unaligned 250 accesses, conflicting marked atomic reads and plain writes will not 251 be reported as data races; however, unlike that option, data races 252 due to two conflicting plain writes will be reported (aligned and 253 unaligned, if CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC=n). 254 255config KCSAN_PERMISSIVE 256 bool "Enable all additional permissive rules" 257 depends on KCSAN_REPORT_VALUE_CHANGE_ONLY 258 help 259 Enable additional permissive rules to ignore certain classes of data 260 races (also see kernel/kcsan/permissive.h). None of the permissive 261 rules imply that such data races are generally safe, but can be used 262 to further reduce reported data races due to data-racy patterns 263 common across the kernel. 264 265endif # KCSAN 266