Re: [Qemu-devel] [PATCH 2/5] tests: more thorough test of ds1338

[Michael Davidsaver <mdavidsaver@gmail.com>]

On 02/18/2018 11:39 PM, Thomas Huth wrote:
> On 19.02.2018 05:03, Michael Davidsaver wrote:
>> Test current time and set+get round trip.
>>
>> The set+get test is repeated 4 times.  These cases are
>> spread across a single day in an attempt to trigger some potential
>> issues regardless of the timezone of the machine running the tests.
>>
>> Signed-off-by: Michael Davidsaver <mdavidsaver@gmail.com>
>> ---
>>  tests/Makefile.include  |   2 +
>>  tests/ds-rtc-i2c-test.c | 193 ++++++++++++++++++++++++++++++++++++++++++++++++
>>  2 files changed, 195 insertions(+)
>>  create mode 100644 tests/ds-rtc-i2c-test.c
> [...]
>>  tests/q35-test$(EXESUF): tests/q35-test.o $(libqos-pc-obj-y)
>> diff --git a/tests/ds-rtc-i2c-test.c b/tests/ds-rtc-i2c-test.c
>> new file mode 100644
>> index 0000000000..464eb08558
>> --- /dev/null
>> +++ b/tests/ds-rtc-i2c-test.c
>> @@ -0,0 +1,193 @@
>> +/* Testing of Dallas/Maxim I2C bus RTC devices
>> + *
>> + * Copyright (c) 2017 Michael Davidsaver
>> + *
>> + * This work is licensed under the terms of the GNU GPL, version 2.  See
>> + * the LICENSE file in the top-level directory.
>> + */
>> +#include <stdio.h>
>> +
>> +#include "qemu/osdep.h"
>> +#include "qemu/bcd.h"
>> +#include "qemu/cutils.h"
>> +#include "qemu/timer.h"
>> +#include "libqtest.h"
>> +#include "libqos/libqos.h"
>> +#include "libqos/i2c.h"
>> +
>> +#define IMX25_I2C_0_BASE 0x43F80000
>> +#define DS1338_ADDR 0x68
>> +
>> +static I2CAdapter *i2c;
>> +static uint8_t addr;
>> +static bool use_century;
>> +
>> +static
>> +time_t rtc_gettime(void)
>> +{
>> +    struct tm parts;
>> +    uint8_t buf[7];
>> +
>> +    buf[0] = 0;
>> +    i2c_send(i2c, addr, buf, 1);
>> +    i2c_recv(i2c, addr, buf, 7);
>> +
>> +    parts.tm_sec = from_bcd(buf[0]);
>> +    parts.tm_min = from_bcd(buf[1]);
>> +    if (buf[2] & 0x40) {
>> +        /* 12 hour */
>> +        /* HOUR register is 1-12. */
>> +        parts.tm_hour = from_bcd(buf[2] & 0x1f);
>> +        g_assert_cmpuint(parts.tm_hour, >=, 1);
>> +        g_assert_cmpuint(parts.tm_hour, <=, 12);
>> +        parts.tm_hour %= 12u; /* wrap 12 -> 0 */
>> +        if (buf[2] & 0x20) {
>> +            parts.tm_hour += 12u;
>> +        }
>> +    } else {
>> +        /* 24 hour */
>> +        parts.tm_hour = from_bcd(buf[2] & 0x3f);
>> +    }
>> +    parts.tm_wday = from_bcd(buf[3]);
>> +    parts.tm_mday = from_bcd(buf[4]);
>> +    parts.tm_mon =  from_bcd((buf[5] & 0x1f) - 1u);
>> +    parts.tm_year = from_bcd(buf[6]);
>> +    if (!use_century || (buf[5] & 0x80)) {
>> +        parts.tm_year += 100u;
>> +    }
>> +
>> +    return mktimegm(&parts);
>> +}
>> +
>> +/* read back and compare with current system time */
>> +static
>> +void test_rtc_current(void)
>> +{
>> +    uint8_t buf;
>> +    time_t expected, actual;
>> +
>> +    /* magic address to zero RTC time offset
>> +     * as tests may be run in any order
>> +     */
>> +    buf = 0xff;
>> +    i2c_send(i2c, addr, &buf, 1);
> 
> That magic (together with patch 1/5) is IMHO a little bit ugly. I've hit
> the same problem with the m48t59 test recently, and I solved it by
> moving the qtest_start() and qtest_end() calls from the main() function
> into the single tests instead, so that each test starts with a clean state:
> 
> https://git.qemu.org/?p=qemu.git;a=commitdiff;h=9c29830c90d82f27f
> 
> Could you maybe try whether that approach works for your test cases
> here, too? Then you could do this without the "0xff" hack here...

Your right, this looks clearer.  I'll try this approach.

>> +
>> +    actual = time(NULL);
>> +    /* new second may start here */
>> +    expected = rtc_gettime();
>> +    g_assert_cmpuint(expected, <=, actual + 1);
>> +    g_assert_cmpuint(expected, >=, actual);
>> +}
>> +
>> +
>> +static uint8_t test_time_24_12am[8] = {
>> +    0, /* address */
>> +    /* Wed, 22 Nov 2017 00:30:53 +0000 */
>> +    0x53,
>> +    0x30,
>> +    0x00, /* 12 AM in 24 hour mode */
>> +    0x03, /* monday is our day 1 */
>> +    0x22,
>> +    0x11 | 0x80,
>> +    0x17,
>> +};
>> +
>> +static uint8_t test_time_24_6am[8] = {
>> +    0, /* address */
>> +    /* Wed, 22 Nov 2017 06:30:53 +0000 */
>> +    0x53,
>> +    0x30,
>> +    0x06, /* 6 AM in 24 hour mode */
>> +    0x03, /* monday is our day 1 */
>> +    0x22,
>> +    0x11 | 0x80,
>> +    0x17,
>> +};
>> +
>> +static uint8_t test_time_24_12pm[8] = {
>> +    0, /* address */
>> +    /* Wed, 22 Nov 2017 12:30:53 +0000 */
>> +    0x53,
>> +    0x30,
>> +    0x12, /* 12 PM in 24 hour mode */
>> +    0x03, /* monday is our day 1 */
>> +    0x22,
>> +    0x11 | 0x80,
>> +    0x17,
>> +};
>> +
>> +static uint8_t test_time_24_6pm[8] = {
>> +    0, /* address */
>> +    /* Wed, 22 Nov 2017 18:30:53 +0000 */
>> +    0x53,
>> +    0x30,
>> +    0x18, /* 6 PM in 24 hour mode */
>> +    0x03, /* monday is our day 1 */
>> +    0x22,
>> +    0x11 | 0x80,
>> +    0x17,
>> +};
>> +
>> +/* write in and read back known time */
>> +static
>> +void test_rtc_set(const void *raw)
>> +{
>> +    const uint8_t *testtime = raw;
>> +    uint8_t buf[7];
>> +    unsigned retry = 2;
>> +
>> +    for (; retry; retry--) {
>> +        i2c_send(i2c, addr, testtime, 8);
>> +        /* new second may start here */
>> +        i2c_send(i2c, addr, testtime, 1);
>> +        i2c_recv(i2c, addr, buf, 7);
>> +
>> +        if (testtime[1] == buf[0]) {
> 
> Please also check the minutes here (reason: see below).
> 
>> +            break;
>> +        }
>> +        /* we raced start of second, retry */
>> +    };
>> +
>> +    g_assert_cmpuint(testtime[1], ==, buf[0]); /* SEC */
>> +    g_assert_cmpuint(testtime[2], ==, buf[1]); /* MIN */
> 
> Could you please wrap the SEC and MIN lines in a "if (!g_test_slow()) {
> ... }" statement? The problem is: The "make check" tests are run as CI
> on a system that is sometimes *very* overloaded. It might happen that
> the test is sometimes interrupted for dozens of seconds, so it might
> fail to properly read the time on a granularity of seconds. With
> !g_test_slow() you can make sure that the check is not done on such
> overloaded CI systems.

Ok I guess.  I certainly don't want to add more false positive test failures.

>> +    g_assert_cmpuint(testtime[3], ==, buf[2]); /* HOUR */
>> +    /* skip comparing Day of Week.  Not handled correctly */
>> +    g_assert_cmpuint(testtime[5], ==, buf[4]); /* DoM */
>> +    if (use_century) {
>> +        g_assert_cmpuint(testtime[6], ==, buf[5]); /* MON+century */
>> +    } else {
>> +        g_assert_cmpuint(testtime[6] & 0x7f, ==, buf[5]); /* MON */
>> +    }
>> +    g_assert_cmpuint(testtime[7], ==, buf[6]); /* YEAR */
>> +
>> +    g_assert_cmpuint(retry, >, 0);
>> +}
>> +
>> +int main(int argc, char *argv[])
>> +{
>> +    int ret;
>> +    const char *arch = qtest_get_arch();
>> +    QTestState *s = NULL;
>> +
>> +    g_test_init(&argc, &argv, NULL);
>> +
>> +    if (strcmp(arch, "arm") == 0) {
>> +        s = qtest_start("-display none -machine imx25-pdk");
> 
> Do you really need the "-display none" parameter here? ... I thought
> that was the default for qtests anyway?

This is a straight copy+paste from the ds1338-test I'm replacing.
I'll remove it.

>> +        i2c = imx_i2c_create(s, IMX25_I2C_0_BASE);
>> +        addr = DS1338_ADDR;
>> +        use_century = false;
>> +
>> +    }
>> +
>> +    qtest_add_data_func("/ds-rtc-i2c/set24_12am", test_time_24_12am, test_rtc_set);
>> +    qtest_add_data_func("/ds-rtc-i2c/set24_6am", test_time_24_6am, test_rtc_set);
>> +    qtest_add_data_func("/ds-rtc-i2c/set24_12pm", test_time_24_12pm, test_rtc_set);
>> +    qtest_add_data_func("/ds-rtc-i2c/set24_6pm", test_time_24_6pm, test_rtc_set);
>> +    qtest_add_func("/ds-rtc-i2c/current", test_rtc_current);
>> +
>> +    ret = g_test_run();
>> +
>> +    qtest_end();
>> +
>> +    return ret;
>> +}
>>
> 
>  Thomas
>