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