Skip to contentSkip to conversation

lib/tests: add KUnit tests for errseq

PATCH Applied Thomas Maarseveen 3 participants 3 messages Aug 12

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lib/tests: add KUnit tests for errseq

The errseq_t infrastructure (lib/errseq.c) underpins writeback error reporting but has no regression tests. Its semantics are subtle enough to have needed fixing before: commit b4678df184b3 ("errseq: Always report a writeback error once") changed how unseen errors reach new samplers.

Add a KUnit suite covering the documented single-threaded semantics:

- a zeroed errseq_t is the "no error yet" epoch
- errors are recorded, overwrite one another, and both ends of the valid errno range round-trip exactly
- an error nobody has seen samples as zero, so a check against a fresh sample still reports it
- errseq_check_and_advance() reports a given error exactly once per cursor and leaves the cursor in place when nothing has changed
- once an error has been seen, a fresh sample is current and a check against it reports nothing
- the same error recorded again after being seen is reported again, even to a cursor that consumed the first occurrence while another cursor marked the repeat as seen
- independent cursors each observe each error

The lockless behaviour of errseq_t under concurrent updates and the WARN path for invalid error values are deliberately out of scope.

Tested with ./tools/testing/kunit/kunit.py run, with a kunitconfig enabling CONFIG_KUNIT=y and CONFIG_ERRSEQ_KUNIT_TEST=y; all 13 tests pass under ARCH=um.

---
Single patch adding the first KUnit suite for lib/errseq.c. (Cover not sent for a single-patch series; trailers below drive the recipients.)

---
 MAINTAINERS              |   1 +
 lib/Kconfig.debug        |  15 +++
 lib/tests/Makefile       |   1 +
 lib/tests/errseq_kunit.c | 237 +++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 254 insertions(+)
MAINTAINERS +1 -0
index 8014b9f8253edf15231df215d1b14063a54a55ec..70de54372c4f8f12b2431e693cb3b0b5ec5fdfc9 100644
@@ -9624,6 +9624,7 @@ M:	Jeff Layton <[email protected]>
 S:	Maintained
 F:	include/linux/errseq.h
 F:	lib/errseq.c
+F:	lib/tests/errseq_kunit.c
 
 ESD CAN NETWORK DRIVERS
 M:	Stefan Mätje <[email protected]>
lib/Kconfig.debug +15 -0
index 1244dcac2294ad99fda37fa6767c9e76f16a4d14..3f1a7103328678926dd82620349957fdede32ee4 100644
@@ -2826,6 +2826,21 @@ config SYSCTL_KUNIT_TEST
 
 	  If unsure, say N.
 
+config ERRSEQ_KUNIT_TEST
+	tristate "KUnit test for errseq" if !KUNIT_ALL_TESTS
+	depends on KUNIT
+	default KUNIT_ALL_TESTS
+	help
+	  This builds the errseq KUnit test suite.
+	  It tests the documented semantics of the errseq_t error-tracking
+	  infrastructure (lib/errseq.c), which underpins writeback error
+	  reporting.
+
+	  For more information on KUnit and unit tests in general please refer
+	  to the KUnit documentation in Documentation/dev-tools/kunit/.
+
+	  If unsure, say N.
+
 config KFIFO_KUNIT_TEST
 	tristate "KUnit Test for the generic kernel FIFO implementation" if !KUNIT_ALL_TESTS
 	depends on KUNIT
lib/tests/Makefile +1 -0
index 4ead57602eac468e1b9b8148593dc8c41ddc5779..a1e69b2c4318ae03f7834ba41414caf986997f02 100644
@@ -13,6 +13,7 @@ obj-$(CONFIG_BLACKHOLE_DEV_KUNIT_TEST) += blackhole_dev_kunit.o
 obj-$(CONFIG_CHECKSUM_KUNIT) += checksum_kunit.o
 obj-$(CONFIG_CMDLINE_KUNIT_TEST) += cmdline_kunit.o
 obj-$(CONFIG_CPUMASK_KUNIT_TEST) += cpumask_kunit.o
+obj-$(CONFIG_ERRSEQ_KUNIT_TEST) += errseq_kunit.o
 obj-$(CONFIG_FFS_KUNIT_TEST) += ffs_kunit.o
 CFLAGS_fortify_kunit.o += $(call cc-disable-warning, unsequenced)
 CFLAGS_fortify_kunit.o += $(call cc-disable-warning, stringop-overread)
lib/tests/errseq_kunit.c new file +237 -0

237 lines not on this page · raw message

---
base-commit: 3d6d817622b0a9721e3cc404df3469171582be13
change-id: 20260812-errseq-kunit-1f1328fbcee1

Best regards,

Signature · 1 line
Thomas Maarseveen <[email protected]>
Acked-by on v1 1/1
#1 lore raw
Quoted 327 lines · On Wed, 2026-08-12 at 20:55 +0200, Thomas Maarseveen via B4 Relay wrote:

From: Thomas Maarseveen <[email protected]>

The errseq_t infrastructure (lib/errseq.c) underpins writeback error reporting but has no regression tests. Its semantics are subtle enough to have needed fixing before: commit b4678df184b3 ("errseq: Always report a writeback error once") changed how unseen errors reach new samplers.

Add a KUnit suite covering the documented single-threaded semantics:

- a zeroed errseq_t is the "no error yet" epoch
- errors are recorded, overwrite one another, and both ends of the valid errno range round-trip exactly
- an error nobody has seen samples as zero, so a check against a fresh sample still reports it
- errseq_check_and_advance() reports a given error exactly once per cursor and leaves the cursor in place when nothing has changed
- once an error has been seen, a fresh sample is current and a check against it reports nothing
- the same error recorded again after being seen is reported again, even to a cursor that consumed the first occurrence while another cursor marked the repeat as seen
- independent cursors each observe each error

The lockless behaviour of errseq_t under concurrent updates and the WARN path for invalid error values are deliberately out of scope.

Tested with ./tools/testing/kunit/kunit.py run, with a kunitconfig enabling CONFIG_KUNIT=y and CONFIG_ERRSEQ_KUNIT_TEST=y; all 13 tests pass under ARCH=um.

---
Single patch adding the first KUnit suite for lib/errseq.c. (Cover not sent for a single-patch series; trailers below drive the recipients.)

---
 MAINTAINERS              |   1 +
 lib/Kconfig.debug        |  15 +++
 lib/tests/Makefile       |   1 +
 lib/tests/errseq_kunit.c | 237 +++++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 254 insertions(+)
MAINTAINERS +1 -0
index 8014b9f8253edf15231df215d1b14063a54a55ec..70de54372c4f8f12b2431e693cb3b0b5ec5fdfc9 100644
@@ -9624,6 +9624,7 @@ M:	Jeff Layton <[email protected]>
 S:	Maintained
 F:	include/linux/errseq.h
 F:	lib/errseq.c
+F:	lib/tests/errseq_kunit.c
 
 ESD CAN NETWORK DRIVERS
 M:	Stefan Mätje <[email protected]>
lib/Kconfig.debug +15 -0
index 1244dcac2294ad99fda37fa6767c9e76f16a4d14..3f1a7103328678926dd82620349957fdede32ee4 100644
@@ -2826,6 +2826,21 @@ config SYSCTL_KUNIT_TEST
 
 	  If unsure, say N.
 
+config ERRSEQ_KUNIT_TEST
+	tristate "KUnit test for errseq" if !KUNIT_ALL_TESTS
+	depends on KUNIT
+	default KUNIT_ALL_TESTS
+	help
+	  This builds the errseq KUnit test suite.
+	  It tests the documented semantics of the errseq_t error-tracking
+	  infrastructure (lib/errseq.c), which underpins writeback error
+	  reporting.
+
+	  For more information on KUnit and unit tests in general please refer
+	  to the KUnit documentation in Documentation/dev-tools/kunit/.
+
+	  If unsure, say N.
+
 config KFIFO_KUNIT_TEST
 	tristate "KUnit Test for the generic kernel FIFO implementation" if !KUNIT_ALL_TESTS
 	depends on KUNIT
lib/tests/Makefile +1 -0
index 4ead57602eac468e1b9b8148593dc8c41ddc5779..a1e69b2c4318ae03f7834ba41414caf986997f02 100644
@@ -13,6 +13,7 @@ obj-$(CONFIG_BLACKHOLE_DEV_KUNIT_TEST) += blackhole_dev_kunit.o
 obj-$(CONFIG_CHECKSUM_KUNIT) += checksum_kunit.o
 obj-$(CONFIG_CMDLINE_KUNIT_TEST) += cmdline_kunit.o
 obj-$(CONFIG_CPUMASK_KUNIT_TEST) += cpumask_kunit.o
+obj-$(CONFIG_ERRSEQ_KUNIT_TEST) += errseq_kunit.o
 obj-$(CONFIG_FFS_KUNIT_TEST) += ffs_kunit.o
 CFLAGS_fortify_kunit.o += $(call cc-disable-warning, unsequenced)
 CFLAGS_fortify_kunit.o += $(call cc-disable-warning, stringop-overread)
lib/tests/errseq_kunit.c new file +227 -0
new file mode 100644
index 0000000000000000000000000000000000000000..8f39ebc4a2488e564efe216883008b0d96a75fe6
@@ -0,0 +1,237 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * KUnit tests for the errseq_t error-tracking infrastructure.
+ *
+ * These exercise the documented single-threaded semantics of the errseq
+ * API (see Documentation/core-api/errseq.rst and lib/errseq.c): error
+ * recording and overwriting, the "seen" handoff between errseq_sample()
+ * and errseq_check_and_advance(), and the re-reporting of an error that
+ * is recorded again after it has been seen.
+ *
+ * The lockless properties of errseq_t under concurrent updates are
+ * outside the scope of these deterministic tests, as is the WARN path
+ * for invalid error values.
+ */
+#include <kunit/test.h>
+
+#include <linux/err.h>
+#include <linux/errno.h>
+#include <linux/errseq.h>
+
+/*
+ * A zeroed errseq_t is the "no error has ever occurred" epoch: it
+ * samples as zero and no check against it reports anything.
+ */
+static void errseq_test_zero_epoch_reports_no_error(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t since = 0;
+
+	KUNIT_EXPECT_EQ(test, errseq_sample(&eseq), 0);
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, 0), 0);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), 0);
+	KUNIT_EXPECT_EQ(test, since, 0);
+}
+
+static void errseq_test_set_records_error(struct kunit *test)
+{
+	errseq_t eseq = 0;
+
+	/* errseq_set() returns the previous value; the epoch is zero. */
+	KUNIT_EXPECT_EQ(test, errseq_set(&eseq, -EIO), 0);
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, 0), -EIO);
+}
+
+/* Any error set always overwrites an existing error. */
+static void errseq_test_set_overwrites_error(struct kunit *test)
+{
+	errseq_t eseq = 0;
+
+	errseq_set(&eseq, -EIO);
+	errseq_set(&eseq, -ENOSPC);
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, 0), -ENOSPC);
+}
+
+/* Both ends of the valid error range are recorded exactly. */
+static void errseq_test_errno_range_extremes(struct kunit *test)
+{
+	errseq_t lo = 0;
+	errseq_t hi = 0;
+
+	errseq_set(&lo, -1);
+	KUNIT_EXPECT_EQ(test, errseq_check(&lo, 0), -1);
+
+	errseq_set(&hi, -MAX_ERRNO);
+	KUNIT_EXPECT_EQ(test, errseq_check(&hi, 0), -MAX_ERRNO);
+}
+
+/*
+ * An error nobody has seen yet samples as zero, so that a check against
+ * the sample still reports it (see commit b4678df184b3 ("errseq: Always
+ * report a writeback error once")).
+ */
+static void errseq_test_sample_of_unseen_error_is_zero(struct kunit *test)
+{
+	errseq_t eseq = 0;
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_sample(&eseq), 0);
+}
+
+static void errseq_test_new_sampler_sees_unseen_error(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t since;
+
+	errseq_set(&eseq, -EIO);
+	since = errseq_sample(&eseq);
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, since), -EIO);
+}
+
+/* A given error is reported exactly once per advancing cursor. */
+static void errseq_test_check_and_advance_reports_once(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t since = errseq_sample(&eseq);
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), 0);
+}
+
+/*
+ * Once an error has been seen, a fresh sample is non-zero and checking
+ * against it reports nothing: handled errors do not reach new samplers.
+ */
+static void errseq_test_sample_after_seen_is_current(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t since = 0;
+	errseq_t sample;
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), -EIO);
+
+	sample = errseq_sample(&eseq);
+	KUNIT_EXPECT_NE(test, sample, 0);
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, sample), 0);
+}
+
+static void errseq_test_new_error_after_advance(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t since = 0;
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), -EIO);
+
+	errseq_set(&eseq, -ENOSPC);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), -ENOSPC);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), 0);
+}
+
+/*
+ * Recording the same error again after it has been seen must bump the
+ * sequence, so cursors that consumed the first occurrence see the
+ * second one too.
+ */
+static void errseq_test_same_error_reported_again_after_seen(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t since = 0;
+	errseq_t seen_cursor;
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), -EIO);
+
+	seen_cursor = since;
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, since), -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), -EIO);
+	/* The repeat must advance the sequence, not just re-toggle "seen". */
+	KUNIT_EXPECT_NE(test, since, seen_cursor);
+}
+
+/*
+ * A cursor that consumed an error must still observe a repeat of that
+ * error even when another cursor has already marked the repeat seen:
+ * recording over a seen value must advance the sequence.
+ */
+static void errseq_test_repeat_error_visible_to_all_cursors(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t cursor_a = 0;
+	errseq_t cursor_b = 0;
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_a), -EIO);
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_b), -EIO);
+
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, cursor_a), -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_a), -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_a), 0);
+}
+
+/* An advance with no new error reports nothing and leaves the cursor put. */
+static void errseq_test_advance_stable_when_unchanged(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t since = 0;
+	errseq_t cursor;
+
+	errseq_set(&eseq, -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), -EIO);
+
+	cursor = since;
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &since), 0);
+	KUNIT_EXPECT_EQ(test, since, cursor);
+}
+
+/*
+ * Cursors are independent: one subscriber consuming an error does not
+ * consume it for another, and each subscriber sees each error once.
+ */
+static void errseq_test_two_subscribers_independent(struct kunit *test)
+{
+	errseq_t eseq = 0;
+	errseq_t cursor_a = errseq_sample(&eseq);
+	errseq_t cursor_b = errseq_sample(&eseq);
+
+	errseq_set(&eseq, -EIO);
+
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_a), -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check(&eseq, cursor_b), -EIO);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_b), -EIO);
+
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_a), 0);
+	KUNIT_EXPECT_EQ(test, errseq_check_and_advance(&eseq, &cursor_b), 0);
+}
+
+static struct kunit_case errseq_test_cases[] = {
+	KUNIT_CASE(errseq_test_zero_epoch_reports_no_error),
+	KUNIT_CASE(errseq_test_set_records_error),
+	KUNIT_CASE(errseq_test_set_overwrites_error),
+	KUNIT_CASE(errseq_test_errno_range_extremes),
+	KUNIT_CASE(errseq_test_sample_of_unseen_error_is_zero),
+	KUNIT_CASE(errseq_test_new_sampler_sees_unseen_error),
+	KUNIT_CASE(errseq_test_check_and_advance_reports_once),
+	KUNIT_CASE(errseq_test_sample_after_seen_is_current),
+	KUNIT_CASE(errseq_test_new_error_after_advance),
+	KUNIT_CASE(errseq_test_same_error_reported_again_after_seen),
+	KUNIT_CASE(errseq_test_repeat_error_visible_to_all_cursors),
+	KUNIT_CASE(errseq_test_advance_stable_when_unchanged),
+	KUNIT_CASE(errseq_test_two_subscribers_independent),
+	{}
+};

lib/tests/errseq_kunit.c line 237

Quoted 10 lines
lib/tests/errseq_kunit.c new file +10 -0
+
+static struct kunit_suite errseq_test_suite = {
+	.name = "errseq",
+	.test_cases = errseq_test_cases,
+};
+
+kunit_test_suite(errseq_test_suite);
+
+MODULE_DESCRIPTION("KUnit tests for the errseq infrastructure");
+MODULE_LICENSE("GPL");

Sounds like a good idea.

maintainer
#2 lore raw
In reply to Thomas Maarseveen
Quoted 7 lines · On Wed, 12 Aug 2026 20:55:33 +0200 Thomas Maarseveen via B4 Relay…

The errseq_t infrastructure (lib/errseq.c) underpins writeback error reporting but has no regression tests. Its semantics are subtle enough to have needed fixing before: commit b4678df184b3 ("errseq: Always report a writeback error once") changed how unseen errors reach new samplers.

Add a KUnit suite covering the documented single-threaded semantics:

Thanks. I'll set this aside for additional consideration after 7.3-rc1.