[PATCH v4 i-g-t] tests/xe_gt_freq: Avoid RPe usage in subtests
Belgaumkar, Vinay
vinay.belgaumkar at intel.com
Wed Jul 31 15:11:08 UTC 2024
On 7/30/2024 10:07 PM, Riana Tauro wrote:
> Hi Vinay
>
> On 7/31/2024 4:31 AM, Vinay Belgaumkar wrote:
>> We are seeing several instances where the RPe, which can be altered by
>> pcode dynamically, is causing subtests to fail randomly. Instead of
>> relying
>> on it, we can use a mid frequency value for these subtests and avoid
>> these
>> failures.
>>
>> v2: Fix bug in the tolerance function. Remove rpe usage from freq_range*
>> as well (Badal). Fix test documentation to reflect change (Riana).
>>
>> v3: Actual frequency cannot be guaranteed to follow the requested value
>>
>> v4: Handle cases where RPe can be greater than min/cur freq.
>>
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2200
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2196
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2262
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2256
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2203
>> Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2412
> Closes?
>>
>> Cc: Riana Tauro <riana.tauro at intel.com>
>> Cc: Badal Nilawar <badal.nilawar at intel.com>
>> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar at intel.com>
>> ---
>> tests/intel/xe_gt_freq.c | 57 ++++++++++++++++++++++++++++------------
>> 1 file changed, 40 insertions(+), 17 deletions(-)
>>
>> diff --git a/tests/intel/xe_gt_freq.c b/tests/intel/xe_gt_freq.c
>> index 93ebb5ed0..c9d59508b 100644
>> --- a/tests/intel/xe_gt_freq.c
>> +++ b/tests/intel/xe_gt_freq.c
>> @@ -26,6 +26,9 @@
>> #include <sys/time.h>
>> #define MAX_N_EXEC_QUEUES 16
>> +#define GT_FREQUENCY_MULTIPLIER 50
>> +#define GT_FREQUENCY_SCALER 3
>> +#define FREQ_UNIT_MHZ DIV_ROUND_CLOSEST(GT_FREQUENCY_MULTIPLIER,
>> GT_FREQUENCY_SCALER)
>> /*
>> * Too many intermediate components and steps before freq is adjusted
>> @@ -70,9 +73,14 @@ static uint32_t get_freq(int fd, int gt_id, const
>> char *freq_name)
>> return freq;
>> }
>> -static uint32_t rpe(int fd, int gt_id)
>> +static bool within_expected_range(uint32_t freq, uint32_t val)
>> {
>> - return get_freq(fd, gt_id, "rpe");
>> + /*
>> + * GT Frequencies are requested at units of 16.66 Mhz, so allow
>> + * that tolerance.
>> + */
>> + return (freq <= val + FREQ_UNIT_MHZ) &&
>> + (freq >= val - FREQ_UNIT_MHZ);
>> }
>> static uint32_t get_throttle(int fd, int gt_id, const char
>> *throttle_file)
>> @@ -128,6 +136,8 @@ static void test_freq_basic_api(int fd, int gt_id)
>> {
>> uint32_t rpn = get_freq(fd, gt_id, "rpn");
>> uint32_t rp0 = get_freq(fd, gt_id, "rp0");
>> + uint32_t rpmid = (rp0 + rpn) / 2;
>> + uint32_t min_freq, max_freq;
>> /*
>> * Negative bound tests
>> @@ -142,16 +152,19 @@ static void test_freq_basic_api(int fd, int gt_id)
>> /* Assert min requests are respected from rp0 to rpn */
>> igt_assert(set_freq(fd, gt_id, "min", rp0) > 0);
>> igt_assert(get_freq(fd, gt_id, "min") == rp0);
>> - igt_assert(set_freq(fd, gt_id, "min", rpe(fd, gt_id)) > 0);
>> - igt_assert(get_freq(fd, gt_id, "min") == rpe(fd, gt_id));
>> + igt_assert(set_freq(fd, gt_id, "min", rpmid) > 0);
>> + min_freq = get_freq(fd, gt_id, "min");
>> + /* SLPC can set min higher than rpmid - as it follows RPe */
>> + igt_assert(min_freq >= (rpmid - FREQ_UNIT_MHZ));
>> igt_assert(set_freq(fd, gt_id, "min", rpn) > 0);
>> igt_assert(get_freq(fd, gt_id, "min") == rpn);
>> /* Assert max requests are respected from rpn to rp0 */
>> igt_assert(set_freq(fd, gt_id, "max", rpn) > 0);
>> igt_assert(get_freq(fd, gt_id, "max") == rpn);
>> - igt_assert(set_freq(fd, gt_id, "max", rpe(fd, gt_id)) > 0);
>> - igt_assert(get_freq(fd, gt_id, "max") == rpe(fd, gt_id));
>> + igt_assert(set_freq(fd, gt_id, "max", rpmid) > 0);
>> + max_freq = get_freq(fd, gt_id, "max");
>> + igt_assert(within_expected_range(max_freq, rpmid));
>> igt_assert(set_freq(fd, gt_id, "max", rp0) > 0);
>> igt_assert(get_freq(fd, gt_id, "max") == rp0);
>> }
>> @@ -168,13 +181,16 @@ static void test_freq_fixed(int fd, int gt_id,
>> bool gt_idle)
>> {
>> uint32_t rpn = get_freq(fd, gt_id, "rpn");
>> uint32_t rp0 = get_freq(fd, gt_id, "rp0");
>> + uint32_t rpmid = (rp0 + rpn) / 2;
>> + uint32_t cur_freq, act_freq;
>> igt_debug("Starting testing fixed request\n");
>> /*
>> * For Fixed freq we need to set both min and max to the
>> desired value
>> * Then we check if hardware is actually operating at the
>> desired freq
>> - * And let's do this for all the 3 known Render Performance (RP)
>> values.
>> + * And let's do this for all the 2 known Render Performance (RP)
>> values
>> + * RP0 and RPn and something in between.
>> */
>> igt_assert(set_freq(fd, gt_id, "min", rpn) > 0);
>> igt_assert(set_freq(fd, gt_id, "max", rpn) > 0);
>> @@ -190,17 +206,20 @@ static void test_freq_fixed(int fd, int gt_id,
>> bool gt_idle)
>> igt_assert(get_freq(fd, gt_id, "act") == rpn);
>> }
>> - igt_assert(set_freq(fd, gt_id, "min", rpe(fd, gt_id)) > 0);
>> - igt_assert(set_freq(fd, gt_id, "max", rpe(fd, gt_id)) > 0);
>> + igt_assert(set_freq(fd, gt_id, "min", rpmid) > 0);
>> + igt_assert(set_freq(fd, gt_id, "max", rpmid) > 0);
>> usleep(ACT_FREQ_LATENCY_US);
>> - igt_assert(get_freq(fd, gt_id, "cur") == rpe(fd, gt_id));
>> + cur_freq = get_freq(fd, gt_id, "cur");
>> + /* If rpmid is around RPe, we could see SLPC follow it */
>> + igt_assert(cur_freq >= (rpmid - FREQ_UNIT_MHZ));
>> if (gt_idle) {
>> igt_assert_f(igt_wait(xe_is_gt_in_c6(fd, gt_id), 1000, 10),
>> "GT %d should be in C6\n", gt_id);
>> igt_assert(get_freq(fd, gt_id, "act") == 0);
>> } else {
>> - igt_assert(get_freq(fd, gt_id, "act") == rpe(fd, gt_id));
>> + act_freq = get_freq(fd, gt_id, "act");
>> + igt_assert(act_freq <= cur_freq + FREQ_UNIT_MHZ);
> Can actual frequency be below the expected range of requested?
Yes, pcode can throttle it based on thermals.
Thanks,
Vinay.
>
> Thanks,
> Riana
>> }
>> igt_assert(set_freq(fd, gt_id, "min", rp0) > 0);
>> @@ -232,15 +251,17 @@ static void test_freq_fixed(int fd, int gt_id,
>> bool gt_idle)
>> static void test_freq_range(int fd, int gt_id, bool gt_idle)
>> {
>> uint32_t rpn = get_freq(fd, gt_id, "rpn");
>> + uint32_t rp0 = get_freq(fd, gt_id, "rp0");
>> + uint32_t rpmid = (rp0 + rpn) / 2;
>> uint32_t cur, act;
>> igt_debug("Starting testing range request\n");
>> igt_assert(set_freq(fd, gt_id, "min", rpn) > 0);
>> - igt_assert(set_freq(fd, gt_id, "max", rpe(fd, gt_id)) > 0);
>> + igt_assert(set_freq(fd, gt_id, "max", rpmid) > 0);
>> usleep(ACT_FREQ_LATENCY_US);
>> cur = get_freq(fd, gt_id, "cur");
>> - igt_assert(rpn <= cur && cur <= rpe(fd, gt_id));
>> + igt_assert(rpn <= cur && cur <= rpmid + FREQ_UNIT_MHZ);
>> if (gt_idle) {
>> igt_assert_f(igt_wait(xe_is_gt_in_c6(fd, gt_id), 1000, 10),
>> @@ -248,7 +269,7 @@ static void test_freq_range(int fd, int gt_id,
>> bool gt_idle)
>> igt_assert(get_freq(fd, gt_id, "act") == 0);
>> } else {
>> act = get_freq(fd, gt_id, "act");
>> - igt_assert(rpn <= act && act <= rpe(fd, gt_id));
>> + igt_assert((rpn <= act) && (act <= cur + FREQ_UNIT_MHZ));
>> }
>> igt_debug("Finished testing range request\n");
>> @@ -262,17 +283,19 @@ static void test_freq_range(int fd, int gt_id,
>> bool gt_idle)
>> static void test_freq_low_max(int fd, int gt_id)
>> {
>> uint32_t rpn = get_freq(fd, gt_id, "rpn");
>> + uint32_t rp0 = get_freq(fd, gt_id, "rp0");
>> + uint32_t rpmid = (rp0 + rpn) / 2;
>> /*
>> * When max request < min request, max is ignored and min
>> works like
>> * a fixed one. Let's assert this assumption
>> */
>> - igt_assert(set_freq(fd, gt_id, "min", rpe(fd, gt_id)) > 0);
>> + igt_assert(set_freq(fd, gt_id, "min", rpmid) > 0);
>> igt_assert(set_freq(fd, gt_id, "max", rpn) > 0);
>> usleep(ACT_FREQ_LATENCY_US);
>> - /* Refresh value of rpe, pcode could have adjusted it */
>> - igt_assert(get_freq(fd, gt_id, "cur") == rpe(fd, gt_id));
>> + /* Cur freq will follow RPe, which could be higher than min freq */
>> + igt_assert(get_freq(fd, gt_id, "cur") >= (rpmid - FREQ_UNIT_MHZ));
>> }
>> /**
More information about the Intel-gfx
mailing list