From: Rodrigo Alencar Date: Mon, 26 Oct 2026 08:54:06 -0300 Subject: [PATCH v4 2/6] lib: kstrtox: round to nearest, ties to even in kstrto[u]dec64() MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Message-Id: <20261026-e-notation-v4-2-a366a6d14d57@analog.com> References: <20261026-e-notation-v4-0-a366a6d14d57@analog.com> In-Reply-To: <20261026-e-notation-v4-0-a366a6d14d57@analog.com> To: linux-kernel@vger.kernel.org, linux-iio@vger.kernel.org Cc: Andrew Morton , Jonathan Cameron , David Lechner , Andy Shevchenko , Rodrigo Alencar X-Mailer: b4 0.16.0 X-Developer-Signature: v=1; a=ed25519-sha256; t=1793015661; l=8057; i=rodrigo.alencar@analog.com; s=default; h=from:subject:message-id; bh=E4zhgGtHp+WC0bWf/E1TCYalVta/WtXqGC0676rBYxE=; b=PLN0bAlDvEIBwlUtnwDMHqw9+unZgVyPGMtAzwzyKbZ8hC7oUJhRhcDggS/KoDYUQNvleJoS9 S/698onnxwsC1gMZMphQTaGRPF0WHoEjl+IhfJS7pC2BD5tGs3AvzaO X-Developer-Key: i=rodrigo.alencar@analog.com; a=ed25519; pk=ULeHbgU/OYh/PG/4anHDfLgldFItQHAhOktYRVLMFRo= X-Endpoint-Received: by B4 Relay for rodrigo.alencar@analog.com/default with auth_id=561 List-Id: B4 Relay Submissions kstrtoudec64() and kstrtodec64() drop the digits beyond the requested scale, which truncates the result toward zero. The error is up to one unit in the last place and is always in the same direction. This is easy to hit with values produced by userspace floating point formatting. Printing 0.3 with "%.17g" gives "0.29999999999999999", which is parsed as 299999 with a scale of 6 instead of 300000. With E notation a negative exponent can also drop integer digits, so "19e-1" with a scale of 0 gives 1 instead of 2. Round the result to nearest instead, with ties to even. This is the IEEE 754 default rounding mode, as used by strtod() and printf(), and it does not bias results in either direction. The first discarded digit is known from where parsing stopped, so no extra pass over the string is needed: the remaining digits are only checked when the first discarded digit is a 5. Since round-half-even is symmetric, kstrtodec64() keeps rounding the magnitude and negates it. Inputs that fit within the scale are parsed exactly as before. An input that rounds up past the type limit, such as "18446744073.7095516155" with a scale of 9, now returns -ERANGE. Update the existing truncation test cases and add cases for ties and for rounding at the type limits. Signed-off-by: Rodrigo Alencar --- lib/kstrtox.c | 46 ++++++++++++++++++++++++++++++++++++++++------ lib/test-kstrtox.c | 48 +++++++++++++++++++++++++++++++++++------------- 2 files changed, 75 insertions(+), 19 deletions(-) diff --git a/lib/kstrtox.c b/lib/kstrtox.c index 20d94a73f937..47c20079247a 100644 --- a/lib/kstrtox.c +++ b/lib/kstrtox.c @@ -405,9 +405,28 @@ static int skip_digits(const char **s) return *s - start; } +static u64 round_half_even(const char *p, const char *end, u64 res) +{ + if (p == end) + return 0; + + if (*p < '5') + return 0; + if (*p > '5') + return 1; + + /* tie: any nonzero digit after it breaks it, skipping the decimal point */ + while (++p < end) { + if (*p != '0' && *p != '.') + return 1; + } + + return res & 1; +} + static int _kstrtoudec64(const char *s, unsigned int scale, u64 *res) { - const char *s_int, *s_frac; + const char *s_int, *s_frac, *s_end, *s_round; int len_int, len_frac; unsigned int rv; u64 _res = 0; @@ -422,6 +441,7 @@ static int _kstrtoudec64(const char *s, unsigned int scale, u64 *res) s_frac = s; len_frac = skip_digits(&s); + s_end = s; if (!len_int && !len_frac) return -EINVAL; @@ -442,8 +462,16 @@ static int _kstrtoudec64(const char *s, unsigned int scale, u64 *res) return -ERANGE; if (exp < 0) { - len_int = max(len_int + exp, 0); + len_int += exp; exp = 0; + if (len_int < 0) { + s_round = s_end; + len_int = 0; + } else { + s_round = s_int + len_int; + } + } else { + s_round = s_frac; } rv = _parse_integer(s_int, 10, &_res, len_int); @@ -454,6 +482,10 @@ static int _kstrtoudec64(const char *s, unsigned int scale, u64 *res) if (rv & KSTRTOX_OVERFLOW) return -ERANGE; + s_round += rv; + if (check_add_overflow(_res, round_half_even(s_round, s_end, _res), &_res)) + return -ERANGE; + exp -= rv; if (_res && (exp > 19 /* log10(2^64) = 19.26 */ || check_mul_overflow(_res, int_pow(10, exp), &_res))) @@ -473,7 +505,8 @@ static int _kstrtoudec64(const char *s, unsigned int scale, u64 *res) * * For example, a scale of 3 with input "123.45" results in 123450. Note that * trailing zeros in the fractional part input to match the scale are not - * required. Also, digits beyond the specified scale are ignored. + * required. Also, digits beyond the specified scale are rounded to nearest, + * ties to even. * * Return: 0 on success, -ERANGE on overflow and -EINVAL on parsing error. */ @@ -494,9 +527,10 @@ EXPORT_SYMBOL(kstrtoudec64); * @scale: The number of digits to the right of the decimal point. * @res: Where to write the result of the conversion on success. * - * For example, a scale of 4 with input "-3.141592" results in -31415. Note - * that digits beyond the specified scale are ignored. Also, trailing zeros in - * the fractional part input to match the scale are not required. + * For example, a scale of 4 with input "-3.141592" results in -31416. Note + * that digits beyond the specified scale are rounded to nearest, ties to + * even. Also, trailing zeros in the fractional part input to match the scale + * are not required. * * Return: 0 on success, -ERANGE on overflow and -EINVAL on parsing error. */ diff --git a/lib/test-kstrtox.c b/lib/test-kstrtox.c index 9255234752db..1f176b232085 100644 --- a/lib/test-kstrtox.c +++ b/lib/test-kstrtox.c @@ -737,12 +737,23 @@ static void __init test_kstrtoudec64_ok(void) {"0.01", 3, 10}, {"0.1", 9, 100000000ULL}, {"0.01", 9, 10000000}, - /* truncation: more fractional digits than scale */ - {"1.23456", 3, 1234}, - {"3.1415926535", 6, 3141592}, - {"0.999999999", 3, 999}, - {"1.99", 1, 19}, + /* rounding: more fractional digits than scale */ + {"1.23456", 3, 1235}, + {"3.1415926535", 6, 3141593}, + {"0.999999999", 3, 1000}, + {"1.99", 1, 20}, {"1.234", 0, 1}, + /* rounding: ties to even */ + {"0.5", 0, 0}, + {"1.5", 0, 2}, + {"2.5", 0, 2}, + {"2.500000", 0, 2}, + {"2.51", 0, 3}, + {"2.5000001", 0, 3}, + {"0.05", 1, 0}, + {"0.15", 1, 2}, + {"0.25", 1, 2}, + {"99.5", 0, 100}, /* trailing newline */ {"1.5\n", 1, 15}, {"42\n", 3, 42000}, @@ -760,15 +771,16 @@ static void __init test_kstrtoudec64_ok(void) {"18446744073.709551615", 9, ULLONG_MAX}, {"18446744073709.551615", 6, ULLONG_MAX}, {"0.18446744073709551615", 20, ULLONG_MAX}, + {"18446744073.7095516154", 9, ULLONG_MAX}, /* scale > 19: representable when integer part is small */ {"0.00000000000000000001", 20, 1}, {"0.1", 20, 10000000000000000000ULL}, {"0.00000000000000000000001", 23, 1}, - /* truncation with scale > 19 */ + /* rounding with scale > 19 */ {"0.0000000000000000000012345", 23, 123}, - /* truncation with many excess digits */ + /* rounding with many excess digits */ {"0.00000000000000000000000000000000423", 34, 42}, - {"1.99999999999999999999999999999999999", 3, 1999}, + {"1.99999999999999999999999999999999999", 3, 2000}, }; TEST_OK(kstrtoudec64, u64, "%llu", test_udec64_ok); } @@ -795,6 +807,7 @@ static void __init test_kstrtoudec64_fail(void) {"1.5 ", 1}, /* overflow */ {"18446744073710.551615", 6}, + {"18446744073.7095516155", 9}, {"99999999999999999999", 1}, /* overflow with scale > 19 */ {"1.0", 21}, @@ -833,11 +846,16 @@ static void __init test_kstrtodec64_ok(void) {"-1.2", 3, -1200}, {"0.01", 3, 10}, {"-0.01", 3, -10}, - /* truncation */ - {"1.23456", 3, 1234}, - {"-1.23456", 3, -1234}, - {"0.999999999", 3, 999}, - {"-0.999999999", 3, -999}, + /* rounding */ + {"1.23456", 3, 1235}, + {"-1.23456", 3, -1235}, + {"0.999999999", 3, 1000}, + {"-0.999999999", 3, -1000}, + /* rounding: ties to even */ + {"2.5", 0, 2}, + {"-2.5", 0, -2}, + {"-3.5", 0, -4}, + {"-3.141592", 4, -31416}, /* trailing newline */ {"1.5\n", 1, 15}, {"-1.5\n", 1, -15}, @@ -846,6 +864,8 @@ static void __init test_kstrtodec64_ok(void) /* limits */ {"9223372036.854775807", 9, LLONG_MAX}, {"-9223372036.854775808", 9, LLONG_MIN}, + {"9223372036.8547758074", 9, LLONG_MAX}, + {"-9223372036.8547758085", 9, LLONG_MIN}, /* scale > 19 */ {"0.0", 20, 0}, {"-0.0", 20, 0}, @@ -875,6 +895,8 @@ static void __init test_kstrtodec64_fail(void) /* signed overflow */ {"9223372036.854775808", 9}, {"-9223372036.854775809", 9}, + {"9223372036.8547758075", 9}, + {"-9223372036.8547758085001", 9}, {"99999999999999999999", 1}, /* signed overflow with scale > 19 */ {"0.1", 20}, -- 2.43.0