summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--Documentation/dev-tools/kcsan.rst12
-rw-r--r--kernel/kcsan/atomic.h23
-rw-r--r--kernel/kcsan/core.c77
-rw-r--r--kernel/kcsan/kcsan_test.c32
-rw-r--r--kernel/kcsan/permissive.h94
-rw-r--r--lib/Kconfig.kcsan42
-rw-r--r--tools/memory-model/Documentation/access-marking.txt151
7 files changed, 352 insertions, 79 deletions
diff --git a/Documentation/dev-tools/kcsan.rst b/Documentation/dev-tools/kcsan.rst
index 6a600cf8430b..7db43c7c09b8 100644
--- a/Documentation/dev-tools/kcsan.rst
+++ b/Documentation/dev-tools/kcsan.rst
@@ -127,6 +127,18 @@ Kconfig options:
causes KCSAN to not report data races due to conflicts where the only plain
accesses are aligned writes up to word size.
+* ``CONFIG_KCSAN_PERMISSIVE``: Enable additional permissive rules to ignore
+ certain classes of common data races. Unlike the above, the rules are more
+ complex involving value-change patterns, access type, and address. This
+ option depends on ``CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY=y``. For details
+ please see the ``kernel/kcsan/permissive.h``. Testers and maintainers that
+ only focus on reports from specific subsystems and not the whole kernel are
+ recommended to disable this option.
+
+To use the strictest possible rules, select ``CONFIG_KCSAN_STRICT=y``, which
+configures KCSAN to follow the Linux-kernel memory consistency model (LKMM) as
+closely as possible.
+
DebugFS interface
~~~~~~~~~~~~~~~~~
diff --git a/kernel/kcsan/atomic.h b/kernel/kcsan/atomic.h
deleted file mode 100644
index 530ae1bda8e7..000000000000
--- a/kernel/kcsan/atomic.h
+++ /dev/null
@@ -1,23 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-/*
- * Rules for implicitly atomic memory accesses.
- *
- * Copyright (C) 2019, Google LLC.
- */
-
-#ifndef _KERNEL_KCSAN_ATOMIC_H
-#define _KERNEL_KCSAN_ATOMIC_H
-
-#include <linux/types.h>
-
-/*
- * Special rules for certain memory where concurrent conflicting accesses are
- * common, however, the current convention is to not mark them; returns true if
- * access to @ptr should be considered atomic. Called from slow-path.
- */
-static bool kcsan_is_atomic_special(const volatile void *ptr)
-{
- return false;
-}
-
-#endif /* _KERNEL_KCSAN_ATOMIC_H */
diff --git a/kernel/kcsan/core.c b/kernel/kcsan/core.c
index 26709ea65c71..76e67d1e02d4 100644
--- a/kernel/kcsan/core.c
+++ b/kernel/kcsan/core.c
@@ -20,9 +20,9 @@
#include <linux/sched.h>
#include <linux/uaccess.h>
-#include "atomic.h"
#include "encoding.h"
#include "kcsan.h"
+#include "permissive.h"
static bool kcsan_early_enable = IS_ENABLED(CONFIG_KCSAN_EARLY_ENABLE);
unsigned int kcsan_udelay_task = CONFIG_KCSAN_UDELAY_TASK;
@@ -301,9 +301,9 @@ static inline void reset_kcsan_skip(void)
this_cpu_write(kcsan_skip, skip_count);
}
-static __always_inline bool kcsan_is_enabled(void)
+static __always_inline bool kcsan_is_enabled(struct kcsan_ctx *ctx)
{
- return READ_ONCE(kcsan_enabled) && get_ctx()->disable_count == 0;
+ return READ_ONCE(kcsan_enabled) && !ctx->disable_count;
}
/* Introduce delay depending on context and configuration. */
@@ -353,10 +353,18 @@ static noinline void kcsan_found_watchpoint(const volatile void *ptr,
atomic_long_t *watchpoint,
long encoded_watchpoint)
{
+ const bool is_assert = (type & KCSAN_ACCESS_ASSERT) != 0;
+ struct kcsan_ctx *ctx = get_ctx();
unsigned long flags;
bool consumed;
- if (!kcsan_is_enabled())
+ /*
+ * We know a watchpoint exists. Let's try to keep the race-window
+ * between here and finally consuming the watchpoint below as small as
+ * possible -- avoid unneccessarily complex code until consumed.
+ */
+
+ if (!kcsan_is_enabled(ctx))
return;
/*
@@ -364,14 +372,22 @@ static noinline void kcsan_found_watchpoint(const volatile void *ptr,
* reporting a race where e.g. the writer set up the watchpoint, but the
* reader has access_mask!=0, we have to ignore the found watchpoint.
*/
- if (get_ctx()->access_mask != 0)
+ if (ctx->access_mask)
return;
/*
- * Consume the watchpoint as soon as possible, to minimize the chances
- * of !consumed. Consuming the watchpoint must always be guarded by
- * kcsan_is_enabled() check, as otherwise we might erroneously
- * triggering reports when disabled.
+ * If the other thread does not want to ignore the access, and there was
+ * a value change as a result of this thread's operation, we will still
+ * generate a report of unknown origin.
+ *
+ * Use CONFIG_KCSAN_REPORT_RACE_UNKNOWN_ORIGIN=n to filter.
+ */
+ if (!is_assert && kcsan_ignore_address(ptr))
+ return;
+
+ /*
+ * Consuming the watchpoint must be guarded by kcsan_is_enabled() to
+ * avoid erroneously triggering reports if the context is disabled.
*/
consumed = try_consume_watchpoint(watchpoint, encoded_watchpoint);
@@ -391,7 +407,7 @@ static noinline void kcsan_found_watchpoint(const volatile void *ptr,
atomic_long_inc(&kcsan_counters[KCSAN_COUNTER_REPORT_RACES]);
}
- if ((type & KCSAN_ACCESS_ASSERT) != 0)
+ if (is_assert)
atomic_long_inc(&kcsan_counters[KCSAN_COUNTER_ASSERT_FAILURES]);
else
atomic_long_inc(&kcsan_counters[KCSAN_COUNTER_DATA_RACES]);
@@ -409,6 +425,7 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
unsigned long access_mask;
enum kcsan_value_change value_change = KCSAN_VALUE_CHANGE_MAYBE;
unsigned long ua_flags = user_access_save();
+ struct kcsan_ctx *ctx = get_ctx();
unsigned long irq_flags = 0;
/*
@@ -417,16 +434,14 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
*/
reset_kcsan_skip();
- if (!kcsan_is_enabled())
+ if (!kcsan_is_enabled(ctx))
goto out;
/*
- * Special atomic rules: unlikely to be true, so we check them here in
- * the slow-path, and not in the fast-path in is_atomic(). Call after
- * kcsan_is_enabled(), as we may access memory that is not yet
- * initialized during early boot.
+ * Check to-ignore addresses after kcsan_is_enabled(), as we may access
+ * memory that is not yet initialized during early boot.
*/
- if (!is_assert && kcsan_is_atomic_special(ptr))
+ if (!is_assert && kcsan_ignore_address(ptr))
goto out;
if (!check_encodable((unsigned long)ptr, size)) {
@@ -479,15 +494,6 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
break; /* ignore; we do not diff the values */
}
- if (IS_ENABLED(CONFIG_KCSAN_DEBUG)) {
- kcsan_disable_current();
- pr_err("watching %s, size: %zu, addr: %px [slot: %d, encoded: %lx]\n",
- is_write ? "write" : "read", size, ptr,
- watchpoint_slot((unsigned long)ptr),
- encode_watchpoint((unsigned long)ptr, size, is_write));
- kcsan_enable_current();
- }
-
/*
* Delay this thread, to increase probability of observing a racy
* conflicting access.
@@ -498,7 +504,7 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
* Re-read value, and check if it is as expected; if not, we infer a
* racy access.
*/
- access_mask = get_ctx()->access_mask;
+ access_mask = ctx->access_mask;
new = 0;
switch (size) {
case 1:
@@ -521,8 +527,14 @@ kcsan_setup_watchpoint(const volatile void *ptr, size_t size, int type)
if (access_mask)
diff &= access_mask;
- /* Were we able to observe a value-change? */
- if (diff != 0)
+ /*
+ * Check if we observed a value change.
+ *
+ * Also check if the data race should be ignored (the rules depend on
+ * non-zero diff); if it is to be ignored, the below rules for
+ * KCSAN_VALUE_CHANGE_MAYBE apply.
+ */
+ if (diff && !kcsan_ignore_data_race(size, type, old, new, diff))
value_change = KCSAN_VALUE_CHANGE_TRUE;
/* Check if this access raced with another. */
@@ -644,6 +656,15 @@ void __init kcsan_init(void)
pr_info("enabled early\n");
WRITE_ONCE(kcsan_enabled, true);
}
+
+ if (IS_ENABLED(CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY) ||
+ IS_ENABLED(CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC) ||
+ IS_ENABLED(CONFIG_KCSAN_PERMISSIVE) ||
+ IS_ENABLED(CONFIG_KCSAN_IGNORE_ATOMICS)) {
+ pr_warn("non-strict mode configured - use CONFIG_KCSAN_STRICT=y to see all data races\n");
+ } else {
+ pr_info("strict mode configured\n");
+ }
}
/* === Exported interface =================================================== */
diff --git a/kernel/kcsan/kcsan_test.c b/kernel/kcsan/kcsan_test.c
index 8bcffbdef3d3..dc55fd5a36fc 100644
--- a/kernel/kcsan/kcsan_test.c
+++ b/kernel/kcsan/kcsan_test.c
@@ -414,6 +414,14 @@ static noinline void test_kernel_atomic_builtins(void)
__atomic_load_n(&test_var, __ATOMIC_RELAXED);
}
+static noinline void test_kernel_xor_1bit(void)
+{
+ /* Do not report data races between the read-writes. */
+ kcsan_nestable_atomic_begin();
+ test_var ^= 0x10000;
+ kcsan_nestable_atomic_end();
+}
+
/* ===== Test cases ===== */
/* Simple test with normal data race. */
@@ -952,6 +960,29 @@ static void test_atomic_builtins(struct kunit *test)
KUNIT_EXPECT_FALSE(test, match_never);
}
+__no_kcsan
+static void test_1bit_value_change(struct kunit *test)
+{
+ const struct expect_report expect = {
+ .access = {
+ { test_kernel_read, &test_var, sizeof(test_var), 0 },
+ { test_kernel_xor_1bit, &test_var, sizeof(test_var), __KCSAN_ACCESS_RW(KCSAN_ACCESS_WRITE) },
+ },
+ };
+ bool match = false;
+
+ begin_test_checks(test_kernel_read, test_kernel_xor_1bit);
+ do {
+ match = IS_ENABLED(CONFIG_KCSAN_PERMISSIVE)
+ ? report_available()
+ : report_matches(&expect);
+ } while (!end_test_checks(match));
+ if (IS_ENABLED(CONFIG_KCSAN_PERMISSIVE))
+ KUNIT_EXPECT_FALSE(test, match);
+ else
+ KUNIT_EXPECT_TRUE(test, match);
+}
+
/*
* Generate thread counts for all test cases. Values generated are in interval
* [2, 5] followed by exponentially increasing thread counts from 8 to 32.
@@ -1024,6 +1055,7 @@ static struct kunit_case kcsan_test_cases[] = {
KCSAN_KUNIT_CASE(test_jiffies_noreport),
KCSAN_KUNIT_CASE(test_seqlock_noreport),
KCSAN_KUNIT_CASE(test_atomic_builtins),
+ KCSAN_KUNIT_CASE(test_1bit_value_change),
{},
};
diff --git a/kernel/kcsan/permissive.h b/kernel/kcsan/permissive.h
new file mode 100644
index 000000000000..2c01fe4a59ee
--- /dev/null
+++ b/kernel/kcsan/permissive.h
@@ -0,0 +1,94 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Special rules for ignoring entire classes of data-racy memory accesses. None
+ * of the rules here imply that such data races are generally safe!
+ *
+ * All rules in this file can be configured via CONFIG_KCSAN_PERMISSIVE. Keep
+ * them separate from core code to make it easier to audit.
+ *
+ * Copyright (C) 2019, Google LLC.
+ */
+
+#ifndef _KERNEL_KCSAN_PERMISSIVE_H
+#define _KERNEL_KCSAN_PERMISSIVE_H
+
+#include <linux/bitops.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+
+/*
+ * Access ignore rules based on address.
+ */
+static __always_inline bool kcsan_ignore_address(const volatile void *ptr)
+{
+ if (!IS_ENABLED(CONFIG_KCSAN_PERMISSIVE))
+ return false;
+
+ /*
+ * Data-racy bitops on current->flags are too common, ignore completely
+ * for now.
+ */
+ return ptr == &current->flags;
+}
+
+/*
+ * Data race ignore rules based on access type and value change patterns.
+ */
+static bool
+kcsan_ignore_data_race(size_t size, int type, u64 old, u64 new, u64 diff)
+{
+ if (!IS_ENABLED(CONFIG_KCSAN_PERMISSIVE))
+ return false;
+
+ /*
+ * Rules here are only for plain read accesses, so that we still report
+ * data races between plain read-write accesses.
+ */
+ if (type || size > sizeof(long))
+ return false;
+
+ /*
+ * A common pattern is checking/setting just 1 bit in a variable; for
+ * example:
+ *
+ * if (flags & SOME_FLAG) { ... }
+ *
+ * and elsewhere flags is updated concurrently:
+ *
+ * flags |= SOME_OTHER_FLAG; // just 1 bit
+ *
+ * While it is still recommended that such accesses be marked
+ * appropriately, in many cases these types of data races are so common
+ * that marking them all is often unrealistic and left to maintainer
+ * preference.
+ *
+ * The assumption in all cases is that with all known compiler
+ * optimizations (including those that tear accesses), because no more
+ * than 1 bit changed, the plain accesses are safe despite the presence
+ * of data races.
+ *
+ * The rules here will ignore the data races if we observe no more than
+ * 1 bit changed.
+ *
+ * Of course many operations can effecively change just 1 bit, but the
+ * general assuption that data races involving 1-bit changes can be
+ * tolerated still applies.
+ *
+ * And in case a true bug is missed, the bug likely manifests as a
+ * reportable data race elsewhere.
+ */
+ if (hweight64(diff) == 1) {
+ /*
+ * Exception: Report data races where the values look like
+ * ordinary booleans (one of them was 0 and the 0th bit was
+ * changed) More often than not, they come with interesting
+ * memory ordering requirements, so let's report them.
+ */
+ if (!((!old || !new) && diff == 1))
+ return true;
+ }
+
+ return false;
+}
+
+#endif /* _KERNEL_KCSAN_PERMISSIVE_H */
diff --git a/lib/Kconfig.kcsan b/lib/Kconfig.kcsan
index 0440f373248e..e0a93ffdef30 100644
--- a/lib/Kconfig.kcsan
+++ b/lib/Kconfig.kcsan
@@ -40,10 +40,14 @@ menuconfig KCSAN
if KCSAN
-# Compiler capabilities that should not fail the test if they are unavailable.
config CC_HAS_TSAN_COMPOUND_READ_BEFORE_WRITE
def_bool (CC_IS_CLANG && $(cc-option,-fsanitize=thread -mllvm -tsan-compound-read-before-write=1)) || \
(CC_IS_GCC && $(cc-option,-fsanitize=thread --param tsan-compound-read-before-write=1))
+ help
+ The compiler instruments plain compound read-write operations
+ differently (++, --, +=, -=, |=, &=, etc.), which allows KCSAN to
+ distinguish them from other plain accesses. This is currently
+ supported by Clang 12 or later.
config KCSAN_VERBOSE
bool "Show verbose reports with more information about system state"
@@ -58,9 +62,6 @@ config KCSAN_VERBOSE
generated from any one of them, system stability may suffer due to
deadlocks or recursion. If in doubt, say N.
-config KCSAN_DEBUG
- bool "Debugging of KCSAN internals"
-
config KCSAN_SELFTEST
bool "Perform short selftests on boot"
default y
@@ -149,7 +150,8 @@ config KCSAN_SKIP_WATCH_RANDOMIZE
KCSAN_WATCH_SKIP.
config KCSAN_INTERRUPT_WATCHER
- bool "Interruptible watchers"
+ bool "Interruptible watchers" if !KCSAN_STRICT
+ default KCSAN_STRICT
help
If enabled, a task that set up a watchpoint may be interrupted while
delayed. This option will allow KCSAN to detect races between
@@ -169,13 +171,9 @@ config KCSAN_REPORT_ONCE_IN_MS
reporting to avoid flooding the console with reports. Setting this
to 0 disables rate limiting.
-# The main purpose of the below options is to control reported data races (e.g.
-# in fuzzer configs), and are not expected to be switched frequently by other
-# users. We could turn some of them into boot parameters, but given they should
-# not be switched normally, let's keep them here to simplify configuration.
-#
-# The defaults below are chosen to be very conservative, and may miss certain
-# bugs.
+# The main purpose of the below options is to control reported data races, and
+# are not expected to be switched frequently by non-testers or at runtime.
+# The defaults are chosen to be conservative, and can miss certain bugs.
config KCSAN_REPORT_RACE_UNKNOWN_ORIGIN
bool "Report races of unknown origin"
@@ -186,9 +184,17 @@ config KCSAN_REPORT_RACE_UNKNOWN_ORIGIN
reported if it was only possible to infer a race due to a data value
change while an access is being delayed on a watchpoint.
+config KCSAN_STRICT
+ bool "Strict data-race checking"
+ help
+ KCSAN will report data races with the strictest possible rules, which
+ closely aligns with the rules defined by the Linux-kernel memory
+ consistency model (LKMM).
+
config KCSAN_REPORT_VALUE_CHANGE_ONLY
bool "Only report races where watcher observed a data value change"
default y
+ depends on !KCSAN_STRICT
help
If enabled and a conflicting write is observed via a watchpoint, but
the data value of the memory location was observed to remain
@@ -197,6 +203,7 @@ config KCSAN_REPORT_VALUE_CHANGE_ONLY
config KCSAN_ASSUME_PLAIN_WRITES_ATOMIC
bool "Assume that plain aligned writes up to word size are atomic"
default y
+ depends on !KCSAN_STRICT
help
Assume that plain aligned writes up to word size are atomic by
default, and also not subject to other unsafe compiler optimizations
@@ -209,6 +216,7 @@ config KCSAN_ASSUME_PLAIN_WRITES_ATOMIC
config KCSAN_IGNORE_ATOMICS
bool "Do not instrument marked atomic accesses"
+ depends on !KCSAN_STRICT
help
Never instrument marked atomic accesses. This option can be used for
additional filtering. Conflicting marked atomic reads and plain
@@ -224,4 +232,14 @@ config KCSAN_IGNORE_ATOMICS
due to two conflicting plain writes will be reported (aligned and
unaligned, if CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC=n).
+config KCSAN_PERMISSIVE
+ bool "Enable all additional permissive rules"
+ depends on KCSAN_REPORT_VALUE_CHANGE_ONLY
+ help
+ Enable additional permissive rules to ignore certain classes of data
+ races (also see kernel/kcsan/permissive.h). None of the permissive
+ rules imply that such data races are generally safe, but can be used
+ to further reduce reported data races due to data-racy patterns
+ common across the kernel.
+
endif # KCSAN
diff --git a/tools/memory-model/Documentation/access-marking.txt b/tools/memory-model/Documentation/access-marking.txt
index 1ab189f51f55..65778222183e 100644
--- a/tools/memory-model/Documentation/access-marking.txt
+++ b/tools/memory-model/Documentation/access-marking.txt
@@ -37,7 +37,9 @@ compiler's use of code-motion and common-subexpression optimizations.
Therefore, if a given access is involved in an intentional data race,
using READ_ONCE() for loads and WRITE_ONCE() for stores is usually
preferable to data_race(), which in turn is usually preferable to plain
-C-language accesses.
+C-language accesses. It is permissible to combine #2 and #3, for example,
+data_race(READ_ONCE(a)), which will both restrict compiler optimizations
+and disable KCSAN diagnostics.
KCSAN will complain about many types of data races involving plain
C-language accesses, but marking all accesses involved in a given data
@@ -86,6 +88,10 @@ that fail to exclude the updates. In this case, it is important to use
data_race() for the diagnostic reads because otherwise KCSAN would give
false-positive warnings about these diagnostic reads.
+If it is necessary to both restrict compiler optimizations and disable
+KCSAN diagnostics, use both data_race() and READ_ONCE(), for example,
+data_race(READ_ONCE(a)).
+
In theory, plain C-language loads can also be used for this use case.
However, in practice this will have the disadvantage of causing KCSAN
to generate false positives because KCSAN will have no way of knowing
@@ -126,6 +132,11 @@ consistent errors, which in turn are quite capable of breaking heuristics.
Therefore use of data_race() should be limited to cases where some other
code (such as a barrier() call) will force the occasional reload.
+Note that this use case requires that the heuristic be able to handle
+any possible error. In contrast, if the heuristics might be fatally
+confused by one or more of the possible erroneous values, use READ_ONCE()
+instead of data_race().
+
In theory, plain C-language loads can also be used for this use case.
However, in practice this will have the disadvantage of causing KCSAN
to generate false positives because KCSAN will have no way of knowing
@@ -259,9 +270,9 @@ diagnostic purposes. The code might look as follows:
return ret;
}
- int read_foo_diagnostic(void)
+ void read_foo_diagnostic(void)
{
- return data_race(foo);
+ pr_info("Current value of foo: %d\n", data_race(foo));
}
The reader-writer lock prevents the compiler from introducing concurrency
@@ -274,19 +285,34 @@ tells KCSAN that data races are expected, and should be silently
ignored. This data_race() also tells the human reading the code that
read_foo_diagnostic() might sometimes return a bogus value.
-However, please note that your kernel must be built with
-CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC=n in order for KCSAN to
-detect a buggy lockless write. If you need KCSAN to detect such a
-write even if that write did not change the value of foo, you also
-need CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY=n. If you need KCSAN to
-detect such a write happening in an interrupt handler running on the
-same CPU doing the legitimate lock-protected write, you also need
-CONFIG_KCSAN_INTERRUPT_WATCHER=y. With some or all of these Kconfig
-options set properly, KCSAN can be quite helpful, although it is not
-necessarily a full replacement for hardware watchpoints. On the other
-hand, neither are hardware watchpoints a full replacement for KCSAN
-because it is not always easy to tell hardware watchpoint to conditionally
-trap on accesses.
+If it is necessary to suppress compiler optimization and also detect
+buggy lockless writes, read_foo_diagnostic() can be updated as follows:
+
+ void read_foo_diagnostic(void)
+ {
+ pr_info("Current value of foo: %d\n", data_race(READ_ONCE(foo)));
+ }
+
+Alternatively, given that KCSAN is to ignore all accesses in this function,
+this function can be marked __no_kcsan and the data_race() can be dropped:
+
+ void __no_kcsan read_foo_diagnostic(void)
+ {
+ pr_info("Current value of foo: %d\n", READ_ONCE(foo));
+ }
+
+However, in order for KCSAN to detect buggy lockless writes, your kernel
+must be built with CONFIG_KCSAN_ASSUME_PLAIN_WRITES_ATOMIC=n. If you
+need KCSAN to detect such a write even if that write did not change
+the value of foo, you also need CONFIG_KCSAN_REPORT_VALUE_CHANGE_ONLY=n.
+If you need KCSAN to detect such a write happening in an interrupt handler
+running on the same CPU doing the legitimate lock-protected write, you
+also need CONFIG_KCSAN_INTERRUPT_WATCHER=y. With some or all of these
+Kconfig options set properly, KCSAN can be quite helpful, although
+it is not necessarily a full replacement for hardware watchpoints.
+On the other hand, neither are hardware watchpoints a full replacement
+for KCSAN because it is not always easy to tell hardware watchpoint to
+conditionally trap on accesses.
Lock-Protected Writes With Lockless Reads
@@ -319,6 +345,99 @@ of the ASSERT_EXCLUSIVE_WRITER() is to allow KCSAN to check for a buggy
concurrent lockless write.
+Lock-Protected Writes With Heuristic Lockless Reads
+---------------------------------------------------
+
+For another example, suppose that the code can normally make use of
+a per-data-structure lock, but there are times when a global lock
+is required. These times are indicated via a global flag. The code
+might look as follows, and is based loosely on nf_conntrack_lock(),
+nf_conntrack_all_lock(), and nf_conntrack_all_unlock():
+
+ bool global_flag;
+ DEFINE_SPINLOCK(global_lock);
+ struct foo {
+ spinlock_t f_lock;
+ int f_data;
+ };
+
+ /* All foo structures are in the following array. */
+ int nfoo;
+ struct foo *foo_array;
+
+ void do_something_locked(struct foo *fp)
+ {
+ /* This works even if data_race() returns nonsense. */
+ if (!data_race(global_flag)) {
+ spin_lock(&fp->f_lock);
+ if (!smp_load_acquire(&global_flag)) {
+ do_something(fp);
+ spin_unlock(&fp->f_lock);
+ return;
+ }
+ spin_unlock(&fp->f_lock);
+ }
+ spin_lock(&global_lock);
+ /* global_lock held, thus global flag cannot be set. */
+ spin_lock(&fp->f_lock);
+ spin_unlock(&global_lock);
+ /*
+ * global_flag might be set here, but begin_global()
+ * will wait for ->f_lock to be released.
+ */
+ do_something(fp);
+ spin_unlock(&fp->f_lock);
+ }
+
+ void begin_global(void)
+ {
+ int i;
+
+ spin_lock(&global_lock);
+ WRITE_ONCE(global_flag, true);
+ for (i = 0; i < nfoo; i++) {
+ /*
+ * Wait for pre-existing local locks. One at
+ * a time to avoid lockdep limitations.
+ */
+ spin_lock(&fp->f_lock);
+ spin_unlock(&fp->f_lock);
+ }
+ }
+
+ void end_global(void)
+ {
+ smp_store_release(&global_flag, false);
+ spin_unlock(&global_lock);
+ }
+
+All code paths leading from the do_something_locked() function's first
+read from global_flag acquire a lock, so endless load fusing cannot
+happen.
+
+If the value read from global_flag is true, then global_flag is
+rechecked while holding ->f_lock, which, if global_flag is now false,
+prevents begin_global() from completing. It is therefore safe to invoke
+do_something().
+
+Otherwise, if either value read from global_flag is true, then after
+global_lock is acquired global_flag must be false. The acquisition of
+->f_lock will prevent any call to begin_global() from returning, which
+means that it is safe to release global_lock and invoke do_something().
+
+For this to work, only those foo structures in foo_array[] may be passed
+to do_something_locked(). The reason for this is that the synchronization
+with begin_global() relies on momentarily holding the lock of each and
+every foo structure.
+
+The smp_load_acquire() and smp_store_release() are required because
+changes to a foo structure between calls to begin_global() and
+end_global() are carried out without holding that structure's ->f_lock.
+The smp_load_acquire() and smp_store_release() ensure that the next
+invocation of do_something() from do_something_locked() will see those
+changes.
+
+
Lockless Reads and Writes
-------------------------