258 lines · plain
1# SPDX-License-Identifier: GPL-2.0-only2 3config HAVE_ARCH_KCSAN4 bool5 6config HAVE_KCSAN_COMPILER7 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 help10 For the list of compilers that support KCSAN, please see11 <file:Documentation/dev-tools/kcsan.rst>.12 13menuconfig KCSAN14 bool "KCSAN: dynamic data race detector"15 depends on HAVE_ARCH_KCSAN && HAVE_KCSAN_COMPILER16 depends on DEBUG_KERNEL && !KASAN17 select CONSTRUCTORS18 select STACKTRACE19 help20 The Kernel Concurrency Sanitizer (KCSAN) is a dynamic21 data-race detector that relies on compile-time instrumentation.22 KCSAN uses a watchpoint-based sampling approach to detect races.23 24 While KCSAN's primary purpose is to detect data races, it25 also provides assertions to check data access constraints.26 These assertions can expose bugs that do not manifest as27 data races.28 29 See <file:Documentation/dev-tools/kcsan.rst> for more details.30 31if KCSAN32 33config CC_HAS_TSAN_COMPOUND_READ_BEFORE_WRITE34 def_bool (CC_IS_CLANG && $(cc-option,-fsanitize=thread -mllvm -tsan-compound-read-before-write=1)) || \35 (CC_IS_GCC && $(cc-option,-fsanitize=thread --param tsan-compound-read-before-write=1))36 help37 The compiler instruments plain compound read-write operations38 differently (++, --, +=, -=, |=, &=, etc.), which allows KCSAN to39 distinguish them from other plain accesses. This is currently40 supported by Clang 12 or later.41 42config KCSAN_VERBOSE43 bool "Show verbose reports with more information about system state"44 depends on PROVE_LOCKING45 help46 If enabled, reports show more information about the system state that47 may help better analyze and debug races. This includes held locks and48 IRQ trace events.49 50 While this option should generally be benign, we call into more51 external functions on report generation; if a race report is52 generated from any one of them, system stability may suffer due to53 deadlocks or recursion. If in doubt, say N.54 55config KCSAN_SELFTEST56 bool "Perform short selftests on boot"57 default y58 help59 Run KCSAN selftests on boot. On test failure, causes the kernel to60 panic. Recommended to be enabled, ensuring critical functionality61 works as intended.62 63config KCSAN_KUNIT_TEST64 tristate "KCSAN test for integrated runtime behaviour" if !KUNIT_ALL_TESTS65 default KUNIT_ALL_TESTS66 depends on TRACEPOINTS && KUNIT67 select TORTURE_TEST68 help69 KCSAN test focusing on behaviour of the integrated runtime. Tests70 various race scenarios, and verifies the reports generated to71 console. Makes use of KUnit for test organization, and the Torture72 framework for test thread control.73 74 Each test case may run at least up to KCSAN_REPORT_ONCE_IN_MS75 milliseconds. Test run duration may be optimized by building the76 kernel and KCSAN test with KCSAN_REPORT_ONCE_IN_MS set to a lower77 than default value.78 79 Say Y here if you want the test to be built into the kernel and run80 during boot; say M if you want the test to build as a module; say N81 if you are unsure.82 83config KCSAN_EARLY_ENABLE84 bool "Early enable during boot"85 default y86 help87 If KCSAN should be enabled globally as soon as possible. KCSAN can88 later be enabled/disabled via debugfs.89 90config KCSAN_NUM_WATCHPOINTS91 int "Number of available watchpoints"92 default 6493 help94 Total number of available watchpoints. An address range maps into a95 specific watchpoint slot as specified in kernel/kcsan/encoding.h.96 Although larger number of watchpoints may not be usable due to97 limited number of CPUs, a larger value helps to improve performance98 due to reducing cache-line contention. The chosen default is a99 conservative value; we should almost never observe "no_capacity"100 events (see /sys/kernel/debug/kcsan).101 102config KCSAN_UDELAY_TASK103 int "Delay in microseconds (for tasks)"104 default 80105 help106 For tasks, the microsecond delay after setting up a watchpoint.107 108config KCSAN_UDELAY_INTERRUPT109 int "Delay in microseconds (for interrupts)"110 default 20111 help112 For interrupts, the microsecond delay after setting up a watchpoint.113 Interrupts have tighter latency requirements, and their delay should114 be lower than for tasks.115 116config KCSAN_DELAY_RANDOMIZE117 bool "Randomize above delays"118 default y119 help120 If delays should be randomized, where the maximum is KCSAN_UDELAY_*.121 If false, the chosen delays are always the KCSAN_UDELAY_* values122 as defined above.123 124config KCSAN_SKIP_WATCH125 int "Skip instructions before setting up watchpoint"126 default 4000127 help128 The number of per-CPU memory operations to skip, before another129 watchpoint is set up, i.e. one in KCSAN_SKIP_WATCH per-CPU130 memory operations are used to set up a watchpoint. A smaller value131 results in more aggressive race detection, whereas a larger value132 improves system performance at the cost of missing some races.133 134config KCSAN_SKIP_WATCH_RANDOMIZE135 bool "Randomize watchpoint instruction skip count"136 default y137 help138 If instruction skip count should be randomized, where the maximum is139 KCSAN_SKIP_WATCH. If false, the chosen value is always140 KCSAN_SKIP_WATCH.141 142config KCSAN_INTERRUPT_WATCHER143 bool "Interruptible watchers" if !KCSAN_STRICT144 default KCSAN_STRICT145 help146 If enabled, a task that set up a watchpoint may be interrupted while147 delayed. This option will allow KCSAN to detect races between148 interrupted tasks and other threads of execution on the same CPU.149 150 Currently disabled by default, because not all safe per-CPU access151 primitives and patterns may be accounted for, and therefore could152 result in false positives.153 154config KCSAN_REPORT_ONCE_IN_MS155 int "Duration in milliseconds, in which any given race is only reported once"156 default 3000157 help158 Any given race is only reported once in the defined time window.159 Different races may still generate reports within a duration that is160 smaller than the duration defined here. This allows rate limiting161 reporting to avoid flooding the console with reports. Setting this162 to 0 disables rate limiting.163 164# The main purpose of the below options is to control reported data races, and165# are not expected to be switched frequently by non-testers or at runtime.166# The defaults are chosen to be conservative, and can miss certain bugs.167 168config KCSAN_REPORT_RACE_UNKNOWN_ORIGIN169 bool "Report races of unknown origin"170 default y171 help172 If KCSAN should report races where only one access is known, and the173 conflicting access is of unknown origin. This type of race is174 reported if it was only possible to infer a race due to a data value175 change while an access is being delayed on a watchpoint.176 177config KCSAN_STRICT178 bool "Strict data-race checking"179 help180 KCSAN will report data races with the strictest possible rules, which181 closely aligns with the rules defined by the Linux-kernel memory182 consistency model (LKMM).183 184config KCSAN_WEAK_MEMORY185 bool "Enable weak memory modeling to detect missing memory barriers"186 default y187 depends on KCSAN_STRICT188 # We can either let objtool nop __tsan_func_{entry,exit}() and builtin189 # atomics instrumentation in .noinstr.text, or use a compiler that can190 # implement __no_kcsan to really remove all instrumentation.191 depends on !ARCH_WANTS_NO_INSTR || HAVE_NOINSTR_HACK || \192 CC_IS_GCC || CLANG_VERSION >= 140000193 select OBJTOOL if HAVE_NOINSTR_HACK194 help195 Enable support for modeling a subset of weak memory, which allows196 detecting a subset of data races due to missing memory barriers.197 198 Depends on KCSAN_STRICT, because the options strengthening certain199 plain accesses by default (depending on !KCSAN_STRICT) reduce the200 ability to detect any data races invoving reordered accesses, in201 particular reordered writes.202 203 Weak memory modeling relies on additional instrumentation and may204 affect performance.205 206config KCSAN_REPORT_VALUE_CHANGE_ONLY207 bool "Only report races where watcher observed a data value change"208 default y209 depends on !KCSAN_STRICT210 help211 If enabled and a conflicting write is observed via a watchpoint, but212 the data value of the memory location was observed to remain213 unchanged, do not report the data race.214 215config KCSAN_ASSUME_PLAIN_WRITES_ATOMIC216 bool "Assume that plain aligned writes up to word size are atomic"217 default y218 depends on !KCSAN_STRICT219 help220 Assume that plain aligned writes up to word size are atomic by221 default, and also not subject to other unsafe compiler optimizations222 resulting in data races. This will cause KCSAN to not report data223 races due to conflicts where the only plain accesses are aligned224 writes up to word size: conflicts between marked reads and plain225 aligned writes up to word size will not be reported as data races;226 notice that data races between two conflicting plain aligned writes227 will also not be reported.228 229config KCSAN_IGNORE_ATOMICS230 bool "Do not instrument marked atomic accesses"231 depends on !KCSAN_STRICT232 help233 Never instrument marked atomic accesses. This option can be used for234 additional filtering. Conflicting marked atomic reads and plain235 writes will never be reported as a data race, however, will cause236 plain reads and marked writes to result in "unknown origin" reports.237 If combined with CONFIG_KCSAN_REPORT_RACE_UNKNOWN_ORIGIN=n, data238 races where at least one access is marked atomic will never be239 reported.240 241 Similar to KCSAN_ASSUME_PLAIN_WRITES_ATOMIC, but including unaligned242 accesses, conflicting marked atomic reads and plain writes will not243 be reported as data races; however, unlike that option, data races244 due to two conflicting plain writes will be reported (aligned and245 unaligned, if CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC=n).246 247config KCSAN_PERMISSIVE248 bool "Enable all additional permissive rules"249 depends on KCSAN_REPORT_VALUE_CHANGE_ONLY250 help251 Enable additional permissive rules to ignore certain classes of data252 races (also see kernel/kcsan/permissive.h). None of the permissive253 rules imply that such data races are generally safe, but can be used254 to further reduce reported data races due to data-racy patterns255 common across the kernel.256 257endif # KCSAN258