@@ -55,4 +55,17 @@ config DRM_AMD_SECURE_DISPLAY
Cooperate with specific DMCU FW.
+config AMD_DC_BASICS_KUNIT_TEST
+ bool "Enable KUnit tests for the 'basics' sub-component of DAL" if !KUNIT_ALL_TESTS
+ depends on DRM_AMD_DC && KUNIT
+ default KUNIT_ALL_TESTS
+ help
+ Enables unit tests for the Display Core. Only useful for kernel
+ devs running KUnit.
+
+ 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.
+
endmenu
@@ -43,7 +43,7 @@ endif
#TODO: remove when Timing Sync feature is complete
subdir-ccflags-y += -DBUILD_FEATURE_TIMING_SYNC=0
-DAL_LIBS = amdgpu_dm dc modules/freesync modules/color modules/info_packet modules/power dmub/src
+DAL_LIBS = amdgpu_dm dc modules/freesync modules/color modules/info_packet modules/power dmub/src tests
ifdef CONFIG_DRM_AMD_DC_HDCP
DAL_LIBS += modules/hdcp
new file mode 100644
@@ -0,0 +1,6 @@
+CONFIG_KUNIT=y
+CONFIG_PCI=y
+CONFIG_DRM=y
+CONFIG_DRM_AMDGPU=y
+CONFIG_DRM_AMD_DC=y
+CONFIG_AMD_DC_BASICS_KUNIT_TEST=y
\ No newline at end of file
new file mode 100644
@@ -0,0 +1,12 @@
+# SPDX-License-Identifier: MIT
+#
+# Makefile for the KUnit Tests for DC
+#
+
+ifdef CONFIG_AMD_DC_BASICS_KUNIT_TEST
+ DC_TESTS += dc/basics/fixpt31_32_test.o
+endif
+
+AMD_DAL_DC_TESTS = $(addprefix $(AMDDALPATH)/tests/,$(DC_TESTS))
+
+AMD_DISPLAY_FILES += $(AMD_DAL_DC_TESTS)
new file mode 100644
@@ -0,0 +1,232 @@
+// SPDX-License-Identifier: MIT
+/* Unit tests for display/include/fixed31_32.h and dc/basics/fixpt31_32.c
+ *
+ * Copyright (C) 2022, Tales Aparecida <tales.aparecida@gmail.com>
+ */
+
+#include <kunit/test.h>
+#include "os_types.h"
+#include "fixed31_32.h"
+
+static const struct fixed31_32 dc_fixpt_minus_one = { -0x100000000LL };
+
+/**
+ * dc_fixpt_from_int_test - KUnit test for dc_fixpt_from_int
+ * @test: represents a running instance of a test.
+ */
+static void dc_fixpt_from_int_test(struct kunit *test)
+{
+ struct fixed31_32 res;
+
+ res = dc_fixpt_from_int(0);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_zero.value);
+
+ res = dc_fixpt_from_int(1);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ res = dc_fixpt_from_int(-1);
+ KUNIT_EXPECT_EQ(test, res.value, -dc_fixpt_one.value);
+
+ res = dc_fixpt_from_int(INT_MAX);
+ KUNIT_EXPECT_EQ(test, res.value, 0x7FFFFFFF00000000LL);
+
+ res = dc_fixpt_from_int(INT_MIN);
+ KUNIT_EXPECT_EQ(test, res.value,
+ 0x8000000000000000LL); /* implicit negative signal */
+}
+
+/**
+ * dc_fixpt_from_fraction_test - KUnit test for dc_fixpt_from_fraction
+ * @test: represents a running instance of a test.
+ */
+static void dc_fixpt_from_fraction_test(struct kunit *test)
+{
+ struct fixed31_32 res;
+
+ /* Assert signal works as expected */
+ res = dc_fixpt_from_fraction(1LL, 1LL);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ res = dc_fixpt_from_fraction(-1LL, 1LL);
+ KUNIT_EXPECT_EQ(test, res.value, -dc_fixpt_one.value);
+
+ res = dc_fixpt_from_fraction(1LL, -1LL);
+ KUNIT_EXPECT_EQ(test, res.value, -dc_fixpt_one.value);
+
+ res = dc_fixpt_from_fraction(-1LL, -1LL);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ /* Assert that the greatest parameter values works as expected */
+ res = dc_fixpt_from_fraction(LLONG_MAX, LLONG_MAX);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ res = dc_fixpt_from_fraction(LLONG_MIN, LLONG_MIN);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ /* Edge case using the smallest fraction possible without LSB rounding */
+ res = dc_fixpt_from_fraction(1, 1LL << (FIXED31_32_BITS_PER_FRACTIONAL_PART));
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_epsilon.value);
+
+ /* Edge case using the smallest fraction possible with LSB rounding */
+ res = dc_fixpt_from_fraction(1, 1LL << (FIXED31_32_BITS_PER_FRACTIONAL_PART + 1));
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_epsilon.value);
+
+ /* Assert an nil numerator is a valid input */
+ res = dc_fixpt_from_fraction(0LL, LLONG_MAX);
+ KUNIT_EXPECT_EQ(test, res.value, 0LL);
+
+ /* Edge case using every bit of the decimal part without rounding */
+ res = dc_fixpt_from_fraction(8589934590LL, 8589934592LL);
+ KUNIT_EXPECT_EQ(test, res.value, 0x0FFFFFFFFLL);
+
+ res = dc_fixpt_from_fraction(-8589934590LL, 8589934592LL);
+ KUNIT_EXPECT_EQ(test, res.value, -0x0FFFFFFFFLL);
+
+ /* Edge case using every bit of the decimal part then rounding LSB */
+ res = dc_fixpt_from_fraction(8589934591LL, 8589934592LL);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ res = dc_fixpt_from_fraction(-8589934591LL, 8589934592LL);
+ KUNIT_EXPECT_EQ(test, res.value, -dc_fixpt_one.value);
+ /* A repeating decimal in binary representation that doesn't round up the LSB */
+ res = dc_fixpt_from_fraction(4, 3);
+ KUNIT_EXPECT_EQ(test, res.value, 0x0000000155555555LL);
+
+ res = dc_fixpt_from_fraction(-4, 3);
+ KUNIT_EXPECT_EQ(test, res.value, -0x0000000155555555LL);
+
+ /* A repeating decimal in binary representation that rounds up the LSB */
+ res = dc_fixpt_from_fraction(5, 3);
+ KUNIT_EXPECT_EQ(test, res.value, 0x00000001AAAAAAABLL);
+
+ res = dc_fixpt_from_fraction(-5, 3);
+ KUNIT_EXPECT_EQ(test, res.value, -0x00000001AAAAAAABLL);
+}
+
+/**
+ * dc_fixpt_mul_test - KUnit test for dc_fixpt_mul
+ * @test: represents a running instance of a test.
+ */
+static void dc_fixpt_mul_test(struct kunit *test)
+{
+ struct fixed31_32 res;
+ struct fixed31_32 arg;
+
+ /* Assert signal works as expected */
+ res = dc_fixpt_mul(dc_fixpt_one, dc_fixpt_one);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ res = dc_fixpt_mul(dc_fixpt_minus_one, dc_fixpt_one);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_minus_one.value);
+
+ res = dc_fixpt_mul(dc_fixpt_one, dc_fixpt_minus_one);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_minus_one.value);
+
+ res = dc_fixpt_mul(dc_fixpt_minus_one, dc_fixpt_minus_one);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ /* Assert that the greatest parameter values works as expected */
+ arg.value = LONG_MAX;
+ res = dc_fixpt_mul(arg, dc_fixpt_one);
+ KUNIT_EXPECT_EQ(test, res.value, arg.value);
+
+ arg.value = LONG_MIN;
+ res = dc_fixpt_mul(arg, dc_fixpt_one);
+ KUNIT_EXPECT_EQ(test, res.value, arg.value);
+
+ arg.value = LONG_MAX;
+ res = dc_fixpt_mul(dc_fixpt_one, arg);
+ KUNIT_EXPECT_EQ(test, res.value, arg.value);
+
+ arg.value = LONG_MIN;
+ res = dc_fixpt_mul(dc_fixpt_one, arg);
+ KUNIT_EXPECT_EQ(test, res.value, arg.value);
+
+ /* Assert it doesn't round LSB as expected */
+ arg.value = 0x7FFFFFFF7fffffffLL;
+ res = dc_fixpt_mul(arg, dc_fixpt_epsilon);
+ KUNIT_EXPECT_EQ(test, res.value, 0x000000007FFFFFFF);
+
+ /* Assert it rounds LSB as expected */
+ arg.value = 0x7FFFFFFF80000000LL;
+ res = dc_fixpt_mul(arg, dc_fixpt_epsilon);
+ KUNIT_EXPECT_EQ(test, res.value, 0x0000000080000000);
+}
+
+/**
+ * dc_fixpt_sqr_test - KUnit test for dc_fixpt_sqr
+ * @test: represents a running instance of a test.
+ */
+static void dc_fixpt_sqr_test(struct kunit *test)
+{
+ struct fixed31_32 res;
+ struct fixed31_32 arg;
+
+ arg.value = dc_fixpt_one.value;
+ res = dc_fixpt_sqr(arg);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ arg.value = dc_fixpt_minus_one.value;
+ res = dc_fixpt_sqr(arg);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ arg.value = 0;
+ res = dc_fixpt_sqr(arg);
+ KUNIT_EXPECT_EQ(test, res.value, 0);
+
+ /* Test some recognizable values */
+ arg = dc_fixpt_from_int(100);
+ res = dc_fixpt_sqr(arg);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_from_int(10000).value);
+
+ arg = dc_fixpt_from_fraction(1, 100);
+ res = dc_fixpt_sqr(arg);
+ KUNIT_EXPECT_EQ(test, res.value,
+ dc_fixpt_from_fraction(1, 10000).value);
+
+ /* LSB rounding */
+ arg = dc_fixpt_from_fraction(3, 2);
+ res = dc_fixpt_sqr(arg);
+ KUNIT_EXPECT_EQ(test, res.value,
+ dc_fixpt_from_fraction(9, 4).value + 1LL);
+}
+
+/**
+ * dc_fixpt_recip_test - KUnit test for dc_fixpt_recip
+ * @test: represents a running instance of a test.
+ */
+static void dc_fixpt_recip_test(struct kunit *test)
+{
+ struct fixed31_32 res;
+ struct fixed31_32 arg;
+
+ /* Assert 1/1 works as expected */
+ res = dc_fixpt_recip(dc_fixpt_one);
+ KUNIT_EXPECT_EQ(test, res.value, dc_fixpt_one.value);
+
+ /* Assert smallest parameters work as expected. */
+ arg.value = 3LL;
+ res = dc_fixpt_recip(arg);
+ KUNIT_EXPECT_EQ(test, res.value, 0x5555555555555555LL);
+
+ arg.value = -3LL;
+ res = dc_fixpt_recip(arg);
+ KUNIT_EXPECT_EQ(test, res.value, -0x5555555555555555LL);
+}
+
+static struct kunit_case dc_basics_fixpt31_32_test_cases[] = {
+ KUNIT_CASE(dc_fixpt_from_int_test),
+ KUNIT_CASE(dc_fixpt_from_fraction_test),
+ KUNIT_CASE(dc_fixpt_mul_test),
+ KUNIT_CASE(dc_fixpt_sqr_test),
+ KUNIT_CASE(dc_fixpt_recip_test),
+ {}
+};
+
+static struct kunit_suite dc_basics_fixpt31_32_test_suite = {
+ .name = "dc_basics_fixpt31_32",
+ .test_cases = dc_basics_fixpt31_32_test_cases,
+};
+
+kunit_test_suites(&dc_basics_fixpt31_32_test_suite);
+