1 The Kernel Concurrency Sanitizer (KCSAN)
2 ========================================
4 The Kernel Concurrency Sanitizer (KCSAN) is a dynamic race detector, which
5 relies on compile-time instrumentation, and uses a watchpoint-based sampling
6 approach to detect races. KCSAN's primary purpose is to detect `data races`_.
11 KCSAN requires Clang version 11 or later.
13 To enable KCSAN configure the kernel with::
17 KCSAN provides several other configuration options to customize behaviour (see
18 the respective help text in ``lib/Kconfig.kcsan`` for more info).
23 A typical data race report looks like this::
25 ==================================================================
26 BUG: KCSAN: data-race in generic_permission / kernfs_refresh_inode
28 write to 0xffff8fee4c40700c of 4 bytes by task 175 on cpu 4:
29 kernfs_refresh_inode+0x70/0x170
30 kernfs_iop_permission+0x4f/0x90
31 inode_permission+0x190/0x200
32 link_path_walk.part.0+0x503/0x8e0
33 path_lookupat.isra.0+0x69/0x4d0
34 filename_lookup+0x136/0x280
35 user_path_at_empty+0x47/0x60
37 __do_sys_newlstat+0x50/0xb0
38 __x64_sys_newlstat+0x37/0x50
39 do_syscall_64+0x85/0x260
40 entry_SYSCALL_64_after_hwframe+0x44/0xa9
42 read to 0xffff8fee4c40700c of 4 bytes by task 166 on cpu 6:
43 generic_permission+0x5b/0x2a0
44 kernfs_iop_permission+0x66/0x90
45 inode_permission+0x190/0x200
46 link_path_walk.part.0+0x503/0x8e0
47 path_lookupat.isra.0+0x69/0x4d0
48 filename_lookup+0x136/0x280
49 user_path_at_empty+0x47/0x60
50 do_faccessat+0x11a/0x390
51 __x64_sys_access+0x3c/0x50
52 do_syscall_64+0x85/0x260
53 entry_SYSCALL_64_after_hwframe+0x44/0xa9
55 Reported by Kernel Concurrency Sanitizer on:
56 CPU: 6 PID: 166 Comm: systemd-journal Not tainted 5.3.0-rc7+ #1
57 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
58 ==================================================================
60 The header of the report provides a short summary of the functions involved in
61 the race. It is followed by the access types and stack traces of the 2 threads
62 involved in the data race.
64 The other less common type of data race report looks like this::
66 ==================================================================
67 BUG: KCSAN: data-race in e1000_clean_rx_irq+0x551/0xb10
69 race at unknown origin, with read to 0xffff933db8a2ae6c of 1 bytes by interrupt on cpu 0:
70 e1000_clean_rx_irq+0x551/0xb10
71 e1000_clean+0x533/0xda0
72 net_rx_action+0x329/0x900
73 __do_softirq+0xdb/0x2db
76 ret_from_intr+0x0/0x18
77 default_idle+0x3f/0x220
78 arch_cpu_idle+0x21/0x30
80 cpu_startup_entry+0x14/0x20
82 arch_call_rest_init+0x13/0x2b
83 start_kernel+0x6db/0x700
85 Reported by Kernel Concurrency Sanitizer on:
86 CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.3.0-rc7+ #2
87 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
88 ==================================================================
90 This report is generated where it was not possible to determine the other
91 racing thread, but a race was inferred due to the data value of the watched
92 memory location having changed. These can occur either due to missing
93 instrumentation or e.g. DMA accesses. These reports will only be generated if
94 ``CONFIG_KCSAN_REPORT_RACE_UNKNOWN_ORIGIN=y`` (selected by default).
99 It may be desirable to disable data race detection for specific accesses,
100 functions, compilation units, or entire subsystems. For static blacklisting,
101 the below options are available:
103 * KCSAN understands the ``data_race(expr)`` annotation, which tells KCSAN that
104 any data races due to accesses in ``expr`` should be ignored and resulting
105 behaviour when encountering a data race is deemed safe.
107 * Disabling data race detection for entire functions can be accomplished by
108 using the function attribute ``__no_kcsan``::
114 To dynamically limit for which functions to generate reports, see the
115 `DebugFS interface`_ blacklist/whitelist feature.
117 For ``__always_inline`` functions, replace ``__always_inline`` with
118 ``__no_kcsan_or_inline`` (which implies ``__always_inline``)::
120 static __no_kcsan_or_inline void foo(void) {
123 * To disable data race detection for a particular compilation unit, add to the
126 KCSAN_SANITIZE_file.o := n
128 * To disable data race detection for all compilation units listed in a
129 ``Makefile``, add to the respective ``Makefile``::
133 Furthermore, it is possible to tell KCSAN to show or hide entire classes of
134 data races, depending on preferences. These can be changed via the following
137 * ``CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY``: If enabled and a conflicting write
138 is observed via a watchpoint, but the data value of the memory location was
139 observed to remain unchanged, do not report the data race.
141 * ``CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC``: Assume that plain aligned writes
142 up to word size are atomic by default. Assumes that such writes are not
143 subject to unsafe compiler optimizations resulting in data races. The option
144 causes KCSAN to not report data races due to conflicts where the only plain
145 accesses are aligned writes up to word size.
150 The file ``/sys/kernel/debug/kcsan`` provides the following interface:
152 * Reading ``/sys/kernel/debug/kcsan`` returns various runtime statistics.
154 * Writing ``on`` or ``off`` to ``/sys/kernel/debug/kcsan`` allows turning KCSAN
155 on or off, respectively.
157 * Writing ``!some_func_name`` to ``/sys/kernel/debug/kcsan`` adds
158 ``some_func_name`` to the report filter list, which (by default) blacklists
159 reporting data races where either one of the top stackframes are a function
162 * Writing either ``blacklist`` or ``whitelist`` to ``/sys/kernel/debug/kcsan``
163 changes the report filtering behaviour. For example, the blacklist feature
164 can be used to silence frequently occurring data races; the whitelist feature
165 can help with reproduction and testing of fixes.
170 Core parameters that affect KCSAN's overall performance and bug detection
171 ability are exposed as kernel command-line arguments whose defaults can also be
172 changed via the corresponding Kconfig options.
174 * ``kcsan.skip_watch`` (``CONFIG_KCSAN_SKIP_WATCH``): Number of per-CPU memory
175 operations to skip, before another watchpoint is set up. Setting up
176 watchpoints more frequently will result in the likelihood of races to be
177 observed to increase. This parameter has the most significant impact on
178 overall system performance and race detection ability.
180 * ``kcsan.udelay_task`` (``CONFIG_KCSAN_UDELAY_TASK``): For tasks, the
181 microsecond delay to stall execution after a watchpoint has been set up.
182 Larger values result in the window in which we may observe a race to
185 * ``kcsan.udelay_interrupt`` (``CONFIG_KCSAN_UDELAY_INTERRUPT``): For
186 interrupts, the microsecond delay to stall execution after a watchpoint has
187 been set up. Interrupts have tighter latency requirements, and their delay
188 should generally be smaller than the one chosen for tasks.
190 They may be tweaked at runtime via ``/sys/module/kcsan/parameters/``.
195 In an execution, two memory accesses form a *data race* if they *conflict*,
196 they happen concurrently in different threads, and at least one of them is a
197 *plain access*; they *conflict* if both access the same memory location, and at
198 least one is a write. For a more thorough discussion and definition, see `"Plain
199 Accesses and Data Races" in the LKMM`_.
201 .. _"Plain Accesses and Data Races" in the LKMM: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/memory-model/Documentation/explanation.txt#n1922
203 Relationship with the Linux-Kernel Memory Consistency Model (LKMM)
204 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
206 The LKMM defines the propagation and ordering rules of various memory
207 operations, which gives developers the ability to reason about concurrent code.
208 Ultimately this allows to determine the possible executions of concurrent code,
209 and if that code is free from data races.
211 KCSAN is aware of *marked atomic operations* (``READ_ONCE``, ``WRITE_ONCE``,
212 ``atomic_*``, etc.), but is oblivious of any ordering guarantees and simply
213 assumes that memory barriers are placed correctly. In other words, KCSAN
214 assumes that as long as a plain access is not observed to race with another
215 conflicting access, memory operations are correctly ordered.
217 This means that KCSAN will not report *potential* data races due to missing
218 memory ordering. Developers should therefore carefully consider the required
219 memory ordering requirements that remain unchecked. If, however, missing
220 memory ordering (that is observable with a particular compiler and
221 architecture) leads to an observable data race (e.g. entering a critical
222 section erroneously), KCSAN would report the resulting data race.
224 Race Detection Beyond Data Races
225 --------------------------------
227 For code with complex concurrency design, race-condition bugs may not always
228 manifest as data races. Race conditions occur if concurrently executing
229 operations result in unexpected system behaviour. On the other hand, data races
230 are defined at the C-language level. The following macros can be used to check
231 properties of concurrent code where bugs would not manifest as data races.
233 .. kernel-doc:: include/linux/kcsan-checks.h
234 :functions: ASSERT_EXCLUSIVE_WRITER ASSERT_EXCLUSIVE_WRITER_SCOPED
235 ASSERT_EXCLUSIVE_ACCESS ASSERT_EXCLUSIVE_ACCESS_SCOPED
236 ASSERT_EXCLUSIVE_BITS
238 Implementation Details
239 ----------------------
241 KCSAN relies on observing that two accesses happen concurrently. Crucially, we
242 want to (a) increase the chances of observing races (especially for races that
243 manifest rarely), and (b) be able to actually observe them. We can accomplish
244 (a) by injecting various delays, and (b) by using address watchpoints (or
247 If we deliberately stall a memory access, while we have a watchpoint for its
248 address set up, and then observe the watchpoint to fire, two accesses to the
249 same address just raced. Using hardware watchpoints, this is the approach taken
251 <http://usenix.org/legacy/events/osdi10/tech/full_papers/Erickson.pdf>`_.
252 Unlike DataCollider, KCSAN does not use hardware watchpoints, but instead
253 relies on compiler instrumentation and "soft watchpoints".
255 In KCSAN, watchpoints are implemented using an efficient encoding that stores
256 access type, size, and address in a long; the benefits of using "soft
257 watchpoints" are portability and greater flexibility. KCSAN then relies on the
258 compiler instrumenting plain accesses. For each instrumented plain access:
260 1. Check if a matching watchpoint exists; if yes, and at least one access is a
261 write, then we encountered a racing access.
263 2. Periodically, if no matching watchpoint exists, set up a watchpoint and
264 stall for a small randomized delay.
266 3. Also check the data value before the delay, and re-check the data value
267 after delay; if the values mismatch, we infer a race of unknown origin.
269 To detect data races between plain and marked accesses, KCSAN also annotates
270 marked accesses, but only to check if a watchpoint exists; i.e. KCSAN never
271 sets up a watchpoint on marked accesses. By never setting up watchpoints for
272 marked operations, if all accesses to a variable that is accessed concurrently
273 are properly marked, KCSAN will never trigger a watchpoint and therefore never
279 1. **Memory Overhead:** The overall memory overhead is only a few MiB
280 depending on configuration. The current implementation uses a small array of
281 longs to encode watchpoint information, which is negligible.
283 2. **Performance Overhead:** KCSAN's runtime aims to be minimal, using an
284 efficient watchpoint encoding that does not require acquiring any shared
285 locks in the fast-path. For kernel boot on a system with 8 CPUs:
287 - 5.0x slow-down with the default KCSAN config;
288 - 2.8x slow-down from runtime fast-path overhead only (set very large
289 ``KCSAN_SKIP_WATCH`` and unset ``KCSAN_SKIP_WATCH_RANDOMIZE``).
291 3. **Annotation Overheads:** Minimal annotations are required outside the KCSAN
292 runtime. As a result, maintenance overheads are minimal as the kernel
295 4. **Detects Racy Writes from Devices:** Due to checking data values upon
296 setting up watchpoints, racy writes from devices can also be detected.
298 5. **Memory Ordering:** KCSAN is *not* explicitly aware of the LKMM's ordering
299 rules; this may result in missed data races (false negatives).
301 6. **Analysis Accuracy:** For observed executions, due to using a sampling
302 strategy, the analysis is *unsound* (false negatives possible), but aims to
303 be complete (no false positives).
305 Alternatives Considered
306 -----------------------
308 An alternative data race detection approach for the kernel can be found in the
309 `Kernel Thread Sanitizer (KTSAN) <https://github.com/google/ktsan/wiki>`_.
310 KTSAN is a happens-before data race detector, which explicitly establishes the
311 happens-before order between memory operations, which can then be used to
312 determine data races as defined in `Data Races`_.
314 To build a correct happens-before relation, KTSAN must be aware of all ordering
315 rules of the LKMM and synchronization primitives. Unfortunately, any omission
316 leads to large numbers of false positives, which is especially detrimental in
317 the context of the kernel which includes numerous custom synchronization
318 mechanisms. To track the happens-before relation, KTSAN's implementation
319 requires metadata for each memory location (shadow memory), which for each page
320 corresponds to 4 pages of shadow memory, and can translate into overhead of
321 tens of GiB on a large system.