1 # SPDX-License-Identifier: GPL-2.0-only
6 config HAVE_KCSAN_COMPILER
7 def_bool (CC_IS_CLANG && $(cc-option,-fsanitize=thread -mllvm -tsan-distinguish-volatile=1)) || \
8 (CC_IS_GCC && $(cc-option,-fsanitize=thread --param tsan-distinguish-volatile=1))
10 For the list of compilers that support KCSAN, please see
11 <file:Documentation/dev-tools/kcsan.rst>.
13 config KCSAN_KCOV_BROKEN
14 def_bool KCOV && CC_HAS_SANCOV_TRACE_PC
15 depends on CC_IS_CLANG
16 depends on !$(cc-option,-Werror=unused-command-line-argument -fsanitize=thread -fsanitize-coverage=trace-pc)
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
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
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.
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
39 See <file:Documentation/dev-tools/kcsan.rst> for more details.
43 # Compiler capabilities that should not fail the test if they are unavailable.
44 config CC_HAS_TSAN_COMPOUND_READ_BEFORE_WRITE
45 def_bool (CC_IS_CLANG && $(cc-option,-fsanitize=thread -mllvm -tsan-compound-read-before-write=1)) || \
46 (CC_IS_GCC && $(cc-option,-fsanitize=thread --param tsan-compound-read-before-write=1))
49 bool "Show verbose reports with more information about system state"
50 depends on PROVE_LOCKING
52 If enabled, reports show more information about the system state that
53 may help better analyze and debug races. This includes held locks and
56 While this option should generally be benign, we call into more
57 external functions on report generation; if a race report is
58 generated from any one of them, system stability may suffer due to
59 deadlocks or recursion. If in doubt, say N.
62 bool "Debugging of KCSAN internals"
65 bool "Perform short selftests on boot"
68 Run KCSAN selftests on boot. On test failure, causes the kernel to
69 panic. Recommended to be enabled, ensuring critical functionality
72 config KCSAN_KUNIT_TEST
73 tristate "KCSAN test for integrated runtime behaviour" if !KUNIT_ALL_TESTS
74 default KUNIT_ALL_TESTS
75 depends on TRACEPOINTS && KUNIT
78 KCSAN test focusing on behaviour of the integrated runtime. Tests
79 various race scenarios, and verifies the reports generated to
80 console. Makes use of KUnit for test organization, and the Torture
81 framework for test thread control.
83 Each test case may run at least up to KCSAN_REPORT_ONCE_IN_MS
84 milliseconds. Test run duration may be optimized by building the
85 kernel and KCSAN test with KCSAN_REPORT_ONCE_IN_MS set to a lower
88 Say Y here if you want the test to be built into the kernel and run
89 during boot; say M if you want the test to build as a module; say N
92 config KCSAN_EARLY_ENABLE
93 bool "Early enable during boot"
96 If KCSAN should be enabled globally as soon as possible. KCSAN can
97 later be enabled/disabled via debugfs.
99 config KCSAN_NUM_WATCHPOINTS
100 int "Number of available watchpoints"
103 Total number of available watchpoints. An address range maps into a
104 specific watchpoint slot as specified in kernel/kcsan/encoding.h.
105 Although larger number of watchpoints may not be usable due to
106 limited number of CPUs, a larger value helps to improve performance
107 due to reducing cache-line contention. The chosen default is a
108 conservative value; we should almost never observe "no_capacity"
109 events (see /sys/kernel/debug/kcsan).
111 config KCSAN_UDELAY_TASK
112 int "Delay in microseconds (for tasks)"
115 For tasks, the microsecond delay after setting up a watchpoint.
117 config KCSAN_UDELAY_INTERRUPT
118 int "Delay in microseconds (for interrupts)"
121 For interrupts, the microsecond delay after setting up a watchpoint.
122 Interrupts have tighter latency requirements, and their delay should
123 be lower than for tasks.
125 config KCSAN_DELAY_RANDOMIZE
126 bool "Randomize above delays"
129 If delays should be randomized, where the maximum is KCSAN_UDELAY_*.
130 If false, the chosen delays are always the KCSAN_UDELAY_* values
133 config KCSAN_SKIP_WATCH
134 int "Skip instructions before setting up watchpoint"
137 The number of per-CPU memory operations to skip, before another
138 watchpoint is set up, i.e. one in KCSAN_WATCH_SKIP per-CPU
139 memory operations are used to set up a watchpoint. A smaller value
140 results in more aggressive race detection, whereas a larger value
141 improves system performance at the cost of missing some races.
143 config KCSAN_SKIP_WATCH_RANDOMIZE
144 bool "Randomize watchpoint instruction skip count"
147 If instruction skip count should be randomized, where the maximum is
148 KCSAN_WATCH_SKIP. If false, the chosen value is always
151 config KCSAN_INTERRUPT_WATCHER
152 bool "Interruptible watchers"
154 If enabled, a task that set up a watchpoint may be interrupted while
155 delayed. This option will allow KCSAN to detect races between
156 interrupted tasks and other threads of execution on the same CPU.
158 Currently disabled by default, because not all safe per-CPU access
159 primitives and patterns may be accounted for, and therefore could
160 result in false positives.
162 config KCSAN_REPORT_ONCE_IN_MS
163 int "Duration in milliseconds, in which any given race is only reported once"
166 Any given race is only reported once in the defined time window.
167 Different races may still generate reports within a duration that is
168 smaller than the duration defined here. This allows rate limiting
169 reporting to avoid flooding the console with reports. Setting this
170 to 0 disables rate limiting.
172 # The main purpose of the below options is to control reported data races (e.g.
173 # in fuzzer configs), and are not expected to be switched frequently by other
174 # users. We could turn some of them into boot parameters, but given they should
175 # not be switched normally, let's keep them here to simplify configuration.
177 # The defaults below are chosen to be very conservative, and may miss certain
180 config KCSAN_REPORT_RACE_UNKNOWN_ORIGIN
181 bool "Report races of unknown origin"
184 If KCSAN should report races where only one access is known, and the
185 conflicting access is of unknown origin. This type of race is
186 reported if it was only possible to infer a race due to a data value
187 change while an access is being delayed on a watchpoint.
189 config KCSAN_REPORT_VALUE_CHANGE_ONLY
190 bool "Only report races where watcher observed a data value change"
193 If enabled and a conflicting write is observed via a watchpoint, but
194 the data value of the memory location was observed to remain
195 unchanged, do not report the data race.
197 config KCSAN_ASSUME_PLAIN_WRITES_ATOMIC
198 bool "Assume that plain aligned writes up to word size are atomic"
201 Assume that plain aligned writes up to word size are atomic by
202 default, and also not subject to other unsafe compiler optimizations
203 resulting in data races. This will cause KCSAN to not report data
204 races due to conflicts where the only plain accesses are aligned
205 writes up to word size: conflicts between marked reads and plain
206 aligned writes up to word size will not be reported as data races;
207 notice that data races between two conflicting plain aligned writes
208 will also not be reported.
210 config KCSAN_IGNORE_ATOMICS
211 bool "Do not instrument marked atomic accesses"
213 Never instrument marked atomic accesses. This option can be used for
214 additional filtering. Conflicting marked atomic reads and plain
215 writes will never be reported as a data race, however, will cause
216 plain reads and marked writes to result in "unknown origin" reports.
217 If combined with CONFIG_KCSAN_REPORT_RACE_UNKNOWN_ORIGIN=n, data
218 races where at least one access is marked atomic will never be
221 Similar to KCSAN_ASSUME_PLAIN_WRITES_ATOMIC, but including unaligned
222 accesses, conflicting marked atomic reads and plain writes will not
223 be reported as data races; however, unlike that option, data races
224 due to two conflicting plain writes will be reported (aligned and
225 unaligned, if CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC=n).