From: "Belgaumkar, Vinay" <vinay.belgaumkar@intel.com>
To: "Nilawar, Badal" <badal.nilawar@intel.com>,
Riana Tauro <riana.tauro@intel.com>,
<igt-dev@lists.freedesktop.org>
Subject: Re: [igt-dev] [PATCH i-g-t 1/2] RFC tests/xe: Add a test that validates idle residency on exec
Date: Mon, 28 Aug 2023 11:19:55 -0700 [thread overview]
Message-ID: <4bdfc1a2-25d0-1d7f-5007-cc05955a76ed@intel.com> (raw)
In-Reply-To: <b7ded32e-73d3-31b5-2428-33f29ed3a78d@intel.com>
On 8/28/2023 11:18 AM, Nilawar, Badal wrote:
>
>
> On 28-08-2023 22:11, Belgaumkar, Vinay wrote:
>>
>> On 8/11/2023 3:32 AM, Riana Tauro wrote:
>>> Add a test what runs a background load that is
>>> active approximately 1% of the time. Verify that we do enter
>>> GT-C6 the rest of the time and validate idle residency is within
>>> tolerance.
>>>
>>> Signed-off-by: Riana Tauro <riana.tauro@intel.com>
>>> ---
>>> tests/xe/xe_pm_residency.c | 135
>>> ++++++++++++++++++++++++++++++++++++-
>>> 1 file changed, 133 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/tests/xe/xe_pm_residency.c b/tests/xe/xe_pm_residency.c
>>> index 4936de166..abf07c696 100644
>>> --- a/tests/xe/xe_pm_residency.c
>>> +++ b/tests/xe/xe_pm_residency.c
>>> @@ -11,13 +11,18 @@
>>> * Test category: functionality test
>>> */
>>> +#include <time.h>
>>> +
>>> #include "igt.h"
>>> #include "igt_sysfs.h"
>>> +#include "lib/igt_syncobj.h"
>>> +
>>> +#include "xe/xe_ioctl.h"
>>> #include "xe/xe_query.h"
>>> #include "xe/xe_util.h"
>>> -#define SLEEP_DURATION 3000 /* in milliseconds */
>>> +#define SLEEP_DURATION 3 /* in seconds */
>>> const double tolerance = 0.1;
>>> @@ -38,9 +43,98 @@ const double tolerance = 0.1;
>>> * Description: basic residency test to validate idle residency
>>> * measured over a time interval is within the tolerance
>>> * Run type: FULL
>>> + *
>>> + * SUBTEST: idle-residency-on-exec
>>> + * Description: Validate idle residency measured when a background
>>> + * is only active for ~1% of the time
>>> + * Run type: FULL
>>> */
>>> IGT_TEST_DESCRIPTION("Tests for gtidle properties");
>>> +static void exec_load(int fd, struct drm_xe_engine_class_instance
>>> *hwe, unsigned long *done)
>>> +{
>>> + uint32_t bo = 0;
>>> + uint32_t exec_queue, syncobj, vm;
>>> + uint64_t addr = 0x1a0000;
>>> + uint64_t batch_addr, batch_offset, data_addr, data_offset;
>>> + size_t bo_size;
>>> + int b;
>>> + struct {
>>> + uint32_t batch[16];
>>> + uint64_t pad;
>>> + uint32_t data;
>>> + } *data;
>>> +
>>> + struct drm_xe_sync sync = {
>>> + .flags = DRM_XE_SYNC_SYNCOBJ | DRM_XE_SYNC_SIGNAL,
>>> + };
>>> +
>>> + struct drm_xe_exec exec = {
>>> + .num_batch_buffer = 1,
>>> + .num_syncs = 1,
>>> + .syncs = to_user_pointer(&sync),
>>> + };
>>> +
>>> + vm = xe_vm_create(fd, 0, 0);
>>> + exec_queue = xe_exec_queue_create(fd, vm, hwe, 0);
>>> + bo_size = xe_get_default_alignment(fd);
>>> +
>>> + bo = xe_bo_create_flags(fd, vm, bo_size,
>>> + visible_vram_if_possible(fd, hwe->gt_id));
>>> + data = xe_bo_map(fd, bo, bo_size);
>>> + syncobj = syncobj_create(fd, 0);
>>> +
>>> + xe_vm_bind_sync(fd, vm, bo, 0, addr, bo_size);
>>> +
>>> + batch_offset = (char *)&data->batch - (char *)data;
>>> + batch_addr = addr + batch_offset;
>>> + data_offset = (char *)&data->data - (char *)data;
>>> + data_addr = addr + data_offset;
>>> +
>>> + do {
>>> + uint64_t submit, elapsed;
>>> + struct timespec tv = {};
>>> +
>>> + b = 0;
>>> + done[1]++;
>>> + data->batch[b++] = MI_STORE_DWORD_IMM_GEN4;
>>> + data->batch[b++] = data_addr;
>>> + data->batch[b++] = data_addr >> 32;
>>> + data->batch[b++] = done[1];
>>
>> It'll be better to align this batch like the one here -
>> https://patchwork.freedesktop.org/series/122462/
> This WL is not like spinner, it is a pulse after specific sleep
> interval. Is alignment still recommended?
In that case, not required.
Thanks,
Vinay.
>
> Regards,
> Badal
>>
>> Thanks,
>>
>> Vinay.
>>
>>> + data->batch[b++] = MI_BATCH_BUFFER_END;
>>> + igt_assert(b <= ARRAY_SIZE(data->batch));
>>> +
>>> + exec.exec_queue_id = exec_queue;
>>> + exec.address = batch_addr;
>>> + sync.handle = syncobj;
>>> +
>>> + igt_nsec_elapsed(&tv);
>>> + xe_exec(fd, &exec);
>>> + submit = igt_nsec_elapsed(&tv);
>>> +
>>> + igt_assert(syncobj_wait(fd, &syncobj, 1, INT64_MAX, 0, NULL));
>>> + elapsed = igt_nsec_elapsed(&tv);
>>> + igt_assert_eq(data->data, done[1]);
>>> +
>>> + igt_debug("Execution took %.3fms (submit %.1fus, wait
>>> %.1fus)\n",
>>> + 1e-6 * elapsed,
>>> + 1e-3 * submit,
>>> + 1e-3 * (elapsed - submit));
>>> +
>>> + syncobj_reset(fd, &syncobj, 1);
>>> +
>>> + /* Aim for ~1% busy */
>>> + usleep(elapsed / 10);
>>> + } while (!READ_ONCE(*done));
>>> +
>>> + xe_vm_unbind_sync(fd, vm, 0, addr, bo_size);
>>> + syncobj_destroy(fd, syncobj);
>>> + munmap(data, bo_size);
>>> + gem_close(fd, bo);
>>> + xe_exec_queue_destroy(fd, exec_queue);
>>> + xe_vm_destroy(fd, vm);
>>> +}
>>> +
>>> static unsigned int measured_usleep(unsigned int usec)
>>> {
>>> struct timespec ts = { };
>>> @@ -76,7 +170,7 @@ static void test_idle_residency(int fd, int gt)
>>> igt_assert_f(igt_wait(xe_is_gt_in_c6(fd, gt), 1000, 1), "GT
>>> not in C6\n");
>>> residency_start = read_idle_residency(fd, gt);
>>> - elapsed_ms = measured_usleep(SLEEP_DURATION * 1000) / 1000;
>>> + elapsed_ms = measured_usleep(SLEEP_DURATION * USEC_PER_SEC) /
>>> 1000;
>>> residency_end = read_idle_residency(fd, gt);
>>> igt_info("Measured %lums of idle residency in %lums\n",
>>> @@ -85,9 +179,41 @@ static void test_idle_residency(int fd, int gt)
>>> assert_within_epsilon(residency_end - residency_start,
>>> elapsed_ms, tolerance);
>>> }
>>> +static void idle_residency_on_exec(int fd, struct
>>> drm_xe_engine_class_instance *hwe)
>>> +{
>>> + const int tol = 20;
>>> + unsigned long *done;
>>> + unsigned long cycles, elapsed_ms, residency;
>>> +
>>> + done = mmap(0, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
>>> + igt_assert(done != MAP_FAILED);
>>> + memset(done, 0, 4096);
>>> +
>>> + igt_fork(child, 1)
>>> + exec_load(fd, hwe, done);
>>> +
>>> + cycles -= READ_ONCE(done[1]);
>>> + residency -= read_idle_residency(fd, hwe->gt_id);
>>> + elapsed_ms = measured_usleep(SLEEP_DURATION * USEC_PER_SEC) /
>>> 1000;
>>> + residency += read_idle_residency(fd, hwe->gt_id);
>>> + cycles += READ_ONCE(done[1]);
>>> + *done = 1;
>>> +
>>> + igt_waitchildren();
>>> +
>>> + /* At least one wakeup/s needed for a reasonable test */
>>> + igt_assert(cycles >= SLEEP_DURATION);
>>> +
>>> + /* While very nearly busy, expect full GT C6 */
>>> + assert_within_epsilon(residency, elapsed_ms, tol);
>>> +
>>> + munmap(done, 4096);
>>> +}
>>> +
>>> igt_main
>>> {
>>> int fd, gt;
>>> + struct drm_xe_engine_class_instance *hwe;
>>> igt_fixture {
>>> fd = drm_open_driver(DRIVER_XE);
>>> @@ -104,6 +230,11 @@ igt_main
>>> xe_for_each_gt(fd, gt)
>>> test_idle_residency(fd, gt);
>>> + igt_describe("Validate idle residency on exec");
>>> + igt_subtest("idle-residency-on-exec")
>>> + xe_for_each_hw_engine(fd, hwe)
>>> + idle_residency_on_exec(fd, hwe);
>>> +
>>> igt_fixture {
>>> close(fd);
>>> }
next prev parent reply other threads:[~2023-08-28 18:20 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-08-11 10:32 [igt-dev] [PATCH i-g-t 0/2] Idle Residency on Exec Riana Tauro
2023-08-11 10:32 ` [igt-dev] [PATCH i-g-t 1/2] RFC tests/xe: Add a test that validates idle residency on exec Riana Tauro
2023-08-28 16:28 ` Nilawar, Badal
2023-08-30 5:15 ` Riana Tauro
2023-08-28 16:41 ` Belgaumkar, Vinay
2023-08-28 18:18 ` Nilawar, Badal
2023-08-28 18:19 ` Belgaumkar, Vinay [this message]
2023-09-05 10:54 ` [igt-dev] [PATCH i-g-t v2 1/2] " Riana Tauro
2023-09-08 13:22 ` Gupta, Anshuman
2023-09-12 6:09 ` Riana Tauro
2023-09-12 9:50 ` [igt-dev] [PATCH i-g-t v3 " Riana Tauro
2023-09-15 8:51 ` Nilawar, Badal
2023-08-11 10:32 ` [igt-dev] [PATCH i-g-t 2/2] HAX: Add idle-residency-on-exec to xe-fast-feedback.testlist Riana Tauro
2023-08-11 11:18 ` [igt-dev] ○ CI.xeBAT: info for Idle Residency on Exec Patchwork
2023-08-11 11:20 ` [igt-dev] ✓ Fi.CI.BAT: success " Patchwork
2023-08-12 13:55 ` [igt-dev] ✗ Fi.CI.IGT: failure " Patchwork
2023-09-05 14:41 ` [igt-dev] ✓ Fi.CI.BAT: success for Idle Residency on Exec (rev2) Patchwork
2023-09-05 16:53 ` [igt-dev] ✓ Fi.CI.IGT: " Patchwork
2023-09-12 12:48 ` [igt-dev] ✓ Fi.CI.BAT: success for Idle Residency on Exec (rev3) Patchwork
2023-09-12 13:22 ` [igt-dev] ✓ CI.xeBAT: " Patchwork
2023-09-12 13:59 ` [igt-dev] ✗ Fi.CI.IGT: failure " Patchwork
2023-09-18 6:46 ` Gupta, Anshuman
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4bdfc1a2-25d0-1d7f-5007-cc05955a76ed@intel.com \
--to=vinay.belgaumkar@intel.com \
--cc=badal.nilawar@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=riana.tauro@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox