[igt-dev] [Intel-gfx] [PATCH i-g-t 6/6] tests/i915_query: Engine queues tests

Lionel Landwerlin lionel.g.landwerlin at intel.com
Fri Jun 8 10:13:14 UTC 2018


On 08/06/18 11:02, Tvrtko Ursulin wrote:
>
> On 06/06/2018 15:33, Lionel Landwerlin wrote:
>> Just a few suggestions below. Otherwise looks good to me.
>>
>> On 06/06/18 13:49, Tvrtko Ursulin wrote:
>>> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>>
>>> Basic tests to cover engine queued/runnable/running metric as reported
>>> by the DRM_I915_QUERY_ENGINE_QUEUES query.
>>>
>>> v2:
>>>   * Update ABI for i915 changes.
>>>   * Use igt_spin_busywait_until_running.
>>>   * Support no hardware contexts.
>>>   * More comments. (Lionel Landwerlin)
>>>   Chris Wilson:
>>>   * Check for sw sync support.
>>>   * Multiple contexts queued test.
>>>   * Simplify context and bb allocation.
>>>   * Fix asserts in the running subtest.
>>>
>>> v3:
>>>   * Update for uAPI change.
>>>
>>> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>> ---
>>>   tests/i915_query.c | 442 
>>> +++++++++++++++++++++++++++++++++++++++++++++
>>>   1 file changed, 442 insertions(+)
>>>
>>> diff --git a/tests/i915_query.c b/tests/i915_query.c
>>> index c7de8cbd8371..588428749c50 100644
>>> --- a/tests/i915_query.c
>>> +++ b/tests/i915_query.c
>>> @@ -22,6 +22,7 @@
>>>    */
>>>   #include "igt.h"
>>> +#include "sw_sync.h"
>>>   #include <limits.h>
>>> @@ -477,8 +478,414 @@ test_query_topology_known_pci_ids(int fd, int 
>>> devid)
>>>       free(topo_info);
>>>   }
>>> +#define DRM_I915_QUERY_ENGINE_QUEUES    2
>>> +
>>> +struct drm_i915_query_engine_queues {
>>> +    /** Engine class as in enum drm_i915_gem_engine_class. */
>>> +    __u16 class;
>>> +
>>> +    /** Engine instance number. */
>>> +    __u16 instance;
>>> +
>>> +    /** Number of requests with unresolved fences and dependencies. */
>>> +    __u32 queued;
>>> +
>>> +    /** Number of ready requests waiting on a slot on GPU. */
>>> +    __u32 runnable;
>>> +
>>> +    /** Number of requests executing on the GPU. */
>>> +    __u32 running;
>>> +
>>> +    __u32 rsvd[6];
>>> +};
>>
>> I suppose this above would be in the updated i915_drm.h.
>> I thought you would have put it there as part of your first commit.
>
> Yes, or I should prefix the copy with local_ as we usually do.

I think the point of copying the headers locally was to get rid of the 
local_ etc...

>
>>> +
>>> +static bool query_engine_queues_supported(int fd)
>>> +{
>>> +    struct drm_i915_query_item item = {
>>> +        .query_id = DRM_I915_QUERY_ENGINE_QUEUES,
>>> +    };
>>> +
>>> +    return __i915_query_items(fd, &item, 1) == 0 && item.length > 0;
>>> +}
>>> +
>>> +static void engine_queues_invalid(int fd)
>>> +{
>>> +    struct drm_i915_query_engine_queues queues;
>>> +    struct drm_i915_query_item item;
>>> +    unsigned int len;
>>> +    unsigned int i;
>>> +
>>> +    /* Flags is MBZ. */
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.flags = 1;
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, -EINVAL);
>>> +
>>> +    /* Length not zero and not greater or equal required size. */
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.length = 1;
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, -EINVAL);
>>> +
>>> +    /* Query correct length. */
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert(item.length >= 0);
>>
>> Could be item.length > 0?
>
> True, should be even. :)
>
>>
>>> +    len = item.length;
>>> +
>>> +    /* Ivalid pointer. */
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.length = len;
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, -EFAULT);
>>
>> Maybe verify ULONG_MAX too?
>
> item.data_ptr = ULONG_MAX? Can do.
>
>>
>>> +
>>> +    /* Reserved fields are MBZ. */
>>> +
>>> +    for (i = 0; i < ARRAY_SIZE(queues.rsvd); i++) {
>>> +        memset(&queues, 0, sizeof(queues));
>>> +        queues.rsvd[i] = 1;
>>> +        memset(&item, 0, sizeof(item));
>>> +        item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +        item.length = len;
>>> +        item.data_ptr = to_user_pointer(&queues);
>>> +        i915_query_items(fd, &item, 1);
>>> +        igt_assert_eq(item.length, -EINVAL);
>>> +    }
>>> +
>>> +    memset(&queues, 0, sizeof(queues));
>>> +    queues.class = -1;
>>
>> class = I915_ENGINE_CLASS_INVALID ?
>
> Yes thanks!
>
>>
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.length = len;
>>> +    item.data_ptr = to_user_pointer(&queues);
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, -ENOENT);
>>> +
>>> +    memset(&queues, 0, sizeof(queues));
>>> +    queues.instance = -1;
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.length = len;
>>> +        item.data_ptr = to_user_pointer(&queues);
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, -ENOENT);
>>> +}
>>> +
>>> +static void engine_queues(int fd, const struct 
>>> intel_execution_engine2 *e)
>>> +{
>>> +    struct drm_i915_query_engine_queues queues;
>>> +    struct drm_i915_query_item item;
>>> +    unsigned int len;
>>> +
>>> +    /* Query required buffer length. */
>>> +    memset(&queues, 0, sizeof(queues));
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.data_ptr = to_user_pointer(&queues);
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert(item.length >= 0);
>>
>> item.length > 0?
>
> Marking on TODO..
>
>>
>>> +    igt_assert(item.length <= sizeof(queues));
>>
>> I guess we can expect item.length == sizeof(queues) here.
>
> Hmm yes, no idea what I was thinking there. :I
>
>>
>>
>>> +    len = item.length;
>>> +
>>> +    /* Check length larger than required works and reports same 
>>> length. */
>>> +    memset(&queues, 0, sizeof(queues));
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.data_ptr = to_user_pointer(&queues);
>>> +    item.length = len + 1;
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, len);
>>> +
>>> +    /* Actual query. */
>>> +    memset(&queues, 0, sizeof(queues));
>>> +    queues.class = e->class;
>>> +    queues.instance = e->instance;
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.data_ptr = to_user_pointer(&queues);
>>> +    item.length = len;
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, len);
>>> +}
>>> +
>>> +static unsigned int e2ring(int gem_fd, const struct 
>>> intel_execution_engine2 *e)
>>> +{
>>> +    return gem_class_instance_to_eb_flags(gem_fd, e->class, 
>>> e->instance);
>>> +}
>>> +
>>> +static void
>>> +__query_queues(int fd, const struct intel_execution_engine2 *e,
>>> +           struct drm_i915_query_engine_queues *queues)
>>> +{
>>> +    struct drm_i915_query_item item;
>>> +
>>> +    memset(queues, 0, sizeof(*queues));
>>> +    queues->class = e->class;
>>> +    queues->instance = e->instance;
>>> +    memset(&item, 0, sizeof(item));
>>> +    item.query_id = DRM_I915_QUERY_ENGINE_QUEUES;
>>> +    item.data_ptr = to_user_pointer(queues);
>>> +    item.length = sizeof(*queues);
>>> +    i915_query_items(fd, &item, 1);
>>> +    igt_assert_eq(item.length, sizeof(*queues));
>>> +}
>>> +
>>> +#define TEST_CONTEXTS (1 << 0)
>>> +
>>> +/*
>>> + * Test that the reported number of queued (not ready for execution 
>>> due fences
>>> + * or dependencies) requests on an engine is correct.
>>> + */
>>> +static void
>>> +engine_queued(int gem_fd, const struct intel_execution_engine2 *e,
>>> +          unsigned int flags)
>>> +{
>>> +    const unsigned long engine = e2ring(gem_fd, e);
>>> +    struct drm_i915_query_engine_queues queues;
>>> +    const uint32_t bbe = MI_BATCH_BUFFER_END;
>>> +    const unsigned int max_rq = 10;
>>> +    uint32_t queued[max_rq + 1];
>>> +    unsigned int n, i;
>>> +    uint32_t bo;
>>> +
>>> +    igt_require_sw_sync();
>>> +    if (flags & TEST_CONTEXTS)
>>> +        gem_require_contexts(gem_fd);
>>> +
>>> +    memset(queued, 0, sizeof(queued));
>>> +
>>> +    bo = gem_create(gem_fd, 4096);
>>> +    gem_write(gem_fd, bo, 4092, &bbe, sizeof(bbe));
>>> +
>>> +     /* Create a specific queue depth of unready requests. */
>>> +    for (n = 0; n <= max_rq; n++) {
>>> +        int fence = -1;
>>> +        IGT_CORK_FENCE(cork);
>>> +
>>> +        gem_quiescent_gpu(gem_fd);
>>> +
>>> +        /* Create a cork so we can create a dependency chain. */
>>> +        if (n)
>>> +            fence = igt_cork_plug(&cork, -1);
>>> +
>>> +        /* Submit n unready requests depending on the cork. */
>>> +        for (i = 0; i < n; i++) {
>>> +            struct drm_i915_gem_exec_object2 obj = { };
>>> +            struct drm_i915_gem_execbuffer2 eb = { };
>>> +
>>> +            obj.handle = bo;
>>> +
>>> +            eb.buffer_count = 1;
>>> +            eb.buffers_ptr = to_user_pointer(&obj);
>>> +
>>> +            eb.flags = engine | I915_EXEC_FENCE_IN;
>>> +
>>> +            /*
>>> +             * In context mode each submission is on a separate
>>> +             * context.
>>> +             */
>>> +            if (flags & TEST_CONTEXTS)
>>> +                eb.rsvd1 = gem_context_create(gem_fd);
>>> +
>>> +            eb.rsvd2 = fence;
>>> +
>>> +            gem_execbuf(gem_fd, &eb);
>>> +
>>> +            if (flags & TEST_CONTEXTS)
>>> +                gem_context_destroy(gem_fd, eb.rsvd1);
>>> +        }
>>> +
>>> +        /* Store reported queue depth to assert against later. */
>>> +        __query_queues(gem_fd, e, &queues);
>>> +        queued[n] = queues.queued;
>>> +        igt_info("n=%u queued=%u\n", n, queued[n]);
>>> +
>>> +        /* Unplug the queue and proceed to the next queue depth. */
>>> +        if (fence >= 0)
>>> +            igt_cork_unplug(&cork);
>>> +
>>> +        gem_sync(gem_fd, bo);
>>> +    }
>>> +
>>> +    gem_close(gem_fd, bo);
>>> +
>>> +    for (i = 0; i <= max_rq; i++)
>>> +        igt_assert_eq(queued[i], i);
>>> +}
>>> +
>>> +static igt_spin_t * __spin_poll(int fd, uint32_t ctx, unsigned long 
>>> flags)
>>> +{
>>> +    if (gem_can_store_dword(fd, flags))
>>> +        return __igt_spin_batch_new_poll(fd, ctx, flags);
>>> +    else
>>> +        return __igt_spin_batch_new(fd, ctx, flags, 0);
>>> +}
>>> +
>>> +static unsigned long __spin_wait(igt_spin_t *spin, unsigned int n)
>>> +{
>>> +    struct timespec ts = { };
>>> +    unsigned long t;
>>> +
>>> +    igt_nsec_elapsed(&ts);
>>> +
>>> +    if (spin->running) {
>>> +        igt_spin_busywait_until_running(spin);
>>> +    } else {
>>> +        igt_debug("__spin_wait - usleep mode\n");
>>> +        usleep(500e3); /* Better than nothing! */
>>> +    }
>>> +
>>> +    t = igt_nsec_elapsed(&ts);
>>> +
>>> +    return spin->running ? t : 500e6 / n;
>>> +}
>>> +
>>> +/*
>>> + * Test that the number of requests ready for execution but waiting 
>>> on space on
>>> + * GPU is correctly reported.
>>> + */
>>> +static void
>>> +engine_runnable(int gem_fd, const struct intel_execution_engine2 *e)
>>> +{
>>> +    const unsigned long engine = e2ring(gem_fd, e);
>>> +    struct drm_i915_query_engine_queues queues;
>>> +    bool contexts = gem_has_contexts(gem_fd);
>>> +    const unsigned int max_rq = 10;
>>> +    igt_spin_t *spin[max_rq + 1];
>>> +    uint32_t runnable[max_rq + 1];
>>> +    uint32_t ctx[max_rq];
>>> +    unsigned int n, i;
>>> +
>>> +    memset(runnable, 0, sizeof(runnable));
>>> +
>>> +    if (contexts) {
>>> +        for (i = 0; i < max_rq; i++)
>>> +            ctx[i] = gem_context_create(gem_fd);
>>> +    }
>>> +
>>> +    /*
>>> +     * Submit different number of requests, potentially against 
>>> different
>>> +     * contexts, in order to provoke engine runnable metric returning
>>> +     * different numbers.
>>> +     */
>>> +    for (n = 0; n <= max_rq; n++) {
>>> +        gem_quiescent_gpu(gem_fd);
>>> +
>>> +        for (i = 0; i < n; i++) {
>>> +            uint32_t ctx_ = contexts ? ctx[i] : 0;
>>> +
>>> +            if (i == 0)
>>> +                spin[i] = __spin_poll(gem_fd, ctx_, engine);
>>> +            else
>>> +                spin[i] = __igt_spin_batch_new(gem_fd, ctx_,
>>> +                                   engine, 0);
>>> +        }
>>> +
>>> +        if (n)
>>> +            usleep(__spin_wait(spin[0], n) * n);
>>> +
>>> +        /* Query and store for later checking. */
>>> +        __query_queues(gem_fd, e, &queues);
>>> +        runnable[n] = queues.runnable;
>>> +        igt_info("n=%u runnable=%u\n", n, runnable[n]);
>>> +
>>> +        for (i = 0; i < n; i++) {
>>> +            igt_spin_batch_end(spin[i]);
>>> +            gem_sync(gem_fd, spin[i]->handle);
>>> +            igt_spin_batch_free(gem_fd, spin[i]);
>>> +        }
>>> +    }
>>> +
>>> +    if (contexts) {
>>> +        for (i = 0; i < max_rq; i++)
>>> +            gem_context_destroy(gem_fd, ctx[i]);
>>> +    }
>>> +
>>> +    /*
>>> +     * Check that the runnable metric is zero when nothing is 
>>> submitted,
>>> +     * and that it is greater than zero on the maximum queue depth.
>>> +     *
>>> +     * We cannot assert the exact value since we do not know how many
>>> +     * requests can the submission backend consume.
>>> +     */
>>> +    igt_assert_eq(runnable[0], 0);
>>> +    igt_assert(runnable[max_rq] > 0);
>>> +
>>> +    /*
>>> +     * We can only test that the runnable metric is growing by one 
>>> if we
>>> +     * have context support.
>>> +     */
>>> +    if (contexts)
>>> +        igt_assert_eq(runnable[max_rq] - runnable[max_rq - 1], 1);
>>> +}
>>> +
>>> +/*
>>> + * Test that the number of requests currently executing on the GPU 
>>> is correctly
>>> + * reported.
>>> + */
>>> +static void
>>> +engine_running(int gem_fd, const struct intel_execution_engine2 *e)
>>> +{
>>> +    const unsigned long engine = e2ring(gem_fd, e);
>>> +    struct drm_i915_query_engine_queues queues;
>>> +    const unsigned int max_rq = 10;
>>> +    igt_spin_t *spin[max_rq + 1];
>>> +    uint32_t running[max_rq + 1];
>>> +    unsigned int n, i;
>>> +
>>> +    memset(running, 0, sizeof(running));
>>> +    memset(spin, 0, sizeof(spin));
>>> +
>>> +    /*
>>> +     * Create various queue depths of requests against the same 
>>> context to
>>> +     * try and get submission backed execute one or more on the GPU.
>>> +     */
>>> +    for (n = 0; n <= max_rq; n++) {
>>> +        gem_quiescent_gpu(gem_fd);
>>> +
>>> +        for (i = 0; i < n; i++) {
>>> +            if (i == 0)
>>> +                spin[i] = __spin_poll(gem_fd, 0, engine);
>>> +            else
>>> +                spin[i] = __igt_spin_batch_new(gem_fd, 0,
>>> +                                   engine, 0);
>>> +        }
>>> +
>>> +        if (n)
>>> +            usleep(__spin_wait(spin[0], n) * n);
>>> +
>>> +        /* Query and store for later checking. */
>>> +        __query_queues(gem_fd, e, &queues);
>>> +        running[n] = queues.running;
>>> +        igt_info("n=%u running=%u\n", n, running[n]);
>>> +
>>> +        for (i = 0; i < n; i++) {
>>> +            igt_spin_batch_end(spin[i]);
>>> +            gem_sync(gem_fd, spin[i]->handle);
>>> +            igt_spin_batch_free(gem_fd, spin[i]);
>>> +        }
>>> +    }
>>> +
>>> +    /*
>>> +     * Check that the running metric is zero when nothing is 
>>> submitted,
>>> +     * one when one request is submitted, and at least one for any 
>>> greater
>>> +     * queue depth.
>>> +     *
>>> +     * We cannot assert the exact value since we do not know how many
>>> +     * requests can the submission backend consume.
>>> +     */
>>> +    igt_assert_eq(running[0], 0);
>>> +    for (i = 1; i <= max_rq; i++)
>>> +        igt_assert(running[i] > 0);
>>> +}
>>> +
>>>   igt_main
>>>   {
>>> +    const struct intel_execution_engine2 *e;
>>>       int fd = -1;
>>>       int devid;
>>> @@ -524,6 +931,41 @@ igt_main
>>>           test_query_topology_known_pci_ids(fd, devid);
>>>       }
>>> +    igt_subtest_group {
>>> +        igt_fixture {
>>> +            igt_require(query_engine_queues_supported(fd));
>>> +        }
>>> +
>>> +        igt_subtest("engine-queues-invalid")
>>> +            engine_queues_invalid(fd);
>>> +
>>> +        __for_each_engine_class_instance(fd, e) {
>>> +            igt_subtest_group {
>>> +                igt_fixture {
>>> +                    gem_require_engine(fd,
>>> +                               e->class,
>>> +                               e->instance);
>>> +                }
>>> +
>>> +                igt_subtest_f("engine-queues-%s", e->name)
>>> +                    engine_queues(fd, e);
>>> +
>>> +                igt_subtest_f("engine-queued-%s", e->name)
>>> +                    engine_queued(fd, e, 0);
>>> +
>>> +                igt_subtest_f("engine-queued-contexts-%s",
>>> +                          e->name)
>>> +                    engine_queued(fd, e, TEST_CONTEXTS);
>>> +
>>> +                igt_subtest_f("engine-runnable-%s", e->name)
>>> +                    engine_runnable(fd, e);
>>> +
>>> +                igt_subtest_f("engine-running-%s", e->name)
>>> +                    engine_running(fd, e);
>>> +            }
>>> +        }
>>> +    }
>>> +
>>>       igt_fixture {
>>>           close(fd);
>>>       }
>
> Thanks for reading it through. I have marked the tweaks for the next 
> resend. Problem with this query is that userspace is AWOL so until it 
> materializes it will be just for reference.
>
> Regards,
>
> Tvrtko
>



More information about the igt-dev mailing list