Re: [PATCH i-g-t v6 9/9] tests/intel/xe_madvise: Add per-vma-protection subtest

["Yadav, Arvind" <arvind.yadav@intel.com>]

On 07-04-2026 13:01, Sharma, Nishit wrote:
>
> On 3/25/2026 6:14 PM, Arvind Yadav wrote:
>> This test validates that a WILLNEED VMA protects a shared BO from being
>> purged even when other VMAs are marked DONTNEED. The test creates a BO
>> shared across two VMs, marks VMA1 as DONTNEED while keeping VMA2 as
>> WILLNEED, then triggers memory pressure. The BO should survive and GPU
>> execution should succeed. After marking both VMAs as DONTNEED and
>> triggering pressure again, the BO should be purged, demonstrating that
>> all VMAs must be DONTNEED for the BO to be purgeable.
>>
>> v4:
>>    - Added syncobj_wait() after the second exec. (Nishit)
>>
>> v6:
>>    - Move resource cleanup before igt_skip() to avoid leaking VM and BO
>>      handles on platforms where memory pressure cannot be induced; 
>> replace
>>      igt_assert_eq(retained, 0) with a graceful skip. (Nishit)
>>
>> Cc: Matthew Brost <matthew.brost@intel.com>
>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>> Cc: Nishit Sharma <nishit.sharma@intel.com>
>> Cc: Pravalika Gurram <pravalika.gurram@intel.com>
>> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
>> ---
>>   tests/intel/xe_madvise.c | 127 +++++++++++++++++++++++++++++++++++++++
>>   1 file changed, 127 insertions(+)
>>
>> diff --git a/tests/intel/xe_madvise.c b/tests/intel/xe_madvise.c
>> index 6c0a563d8..2375f5475 100644
>> --- a/tests/intel/xe_madvise.c
>> +++ b/tests/intel/xe_madvise.c
>> @@ -654,6 +654,127 @@ static void test_per_vma_tracking(int fd, 
>> struct drm_xe_engine_class_instance *h
>>     }
>>   +/**
>> + * SUBTEST: per-vma-protection
>> + * Description: WILLNEED VMA protects BO from purging; both DONTNEED 
>> makes BO purgeable
>> + * Test category: functionality test
>> + */
>> +static void test_per_vma_protection(int fd, struct 
>> drm_xe_engine_class_instance *hwe)
>> +{
>> +    uint32_t vm1, vm2, exec_queue, bo, batch_bo, bind_engine;
>> +    uint64_t data_addr1 = PURGEABLE_ADDR;
>> +    uint64_t data_addr2 = PURGEABLE_ADDR2;
>> +    uint64_t batch_addr = PURGEABLE_BATCH_ADDR;
>> +    size_t data_size = PURGEABLE_BO_SIZE;
>> +    size_t batch_size = PURGEABLE_BO_SIZE;
>> +    struct drm_xe_sync sync[2] = {
>> +        { .type = DRM_XE_SYNC_TYPE_USER_FENCE,
>> +          .flags = DRM_XE_SYNC_FLAG_SIGNAL,
>> +          .timeline_value = PURGEABLE_FENCE_VAL },
>> +        { .type = DRM_XE_SYNC_TYPE_SYNCOBJ,
>> +          .flags = DRM_XE_SYNC_FLAG_SIGNAL },
>> +    };
>> +    struct drm_xe_exec exec = {
>> +        .num_batch_buffer = 1,
>> +        .num_syncs = 1,
>> +        .syncs = to_user_pointer(&sync[1]),
>> +    };
>> +    uint32_t *data, *batch;
>> +    uint64_t vm_sync = 0;
>> +    uint32_t retained, syncobj;
>> +    int b, ret;
>> +
>> +    /* Create two VMs and bind shared data BO */
>> +    data = purgeable_setup_two_vms_shared_bo(fd, &vm1, &vm2, &bo,
>> +                         data_addr1, data_addr2,
>> +                         data_size, true);
>> +    memset(data, 0, data_size);
>> +    bind_engine = xe_bind_exec_queue_create(fd, vm2, 0);
>> +
>> +    /* Create and bind batch BO in VM2 */
>> +    batch_bo = xe_bo_create(fd, vm2, batch_size, 
>> vram_if_possible(fd, 0),
>> +                DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
>> +    batch = xe_bo_map(fd, batch_bo, batch_size);
>> +    igt_assert(batch != MAP_FAILED);
>> +
>> +    sync[0].addr = to_user_pointer(&vm_sync);
>> +    vm_sync = 0;
>> +    xe_vm_bind_async(fd, vm2, bind_engine, batch_bo, 0, batch_addr, 
>> batch_size, sync, 1);
>> +    xe_wait_ufence(fd, &vm_sync, PURGEABLE_FENCE_VAL, 0, NSEC_PER_SEC);
>> +
>> +    /* Mark VMA1 as DONTNEED, VMA2 stays WILLNEED */
>> +    retained = xe_vm_madvise_purgeable(fd, vm1, data_addr1, data_size,
>> +                       DRM_XE_VMA_PURGEABLE_STATE_DONTNEED);
>> +    igt_assert_eq(retained, 1);
>> +
>> +    /* Trigger pressure - BO should survive (VMA2 protects) */
>> +    trigger_memory_pressure(fd, vm1);
>> +
>> +    retained = xe_vm_madvise_purgeable(fd, vm2, data_addr2, data_size,
>> +                       DRM_XE_VMA_PURGEABLE_STATE_WILLNEED);
>> +    igt_assert_eq(retained, 1);
>> +
>> +    /* GPU workload - should succeed */
>> +    b = 0;
>> +    batch[b++] = MI_STORE_DWORD_IMM_GEN4;
>> +    batch[b++] = data_addr2;
>> +    batch[b++] = data_addr2 >> 32;
>> +    batch[b++] = PURGEABLE_TEST_PATTERN;
>> +    batch[b++] = MI_BATCH_BUFFER_END;
>> +
>> +    syncobj = syncobj_create(fd, 0);
>> +    sync[1].handle = syncobj;
>> +    exec_queue = xe_exec_queue_create(fd, vm2, hwe, 0);
>> +    exec.exec_queue_id = exec_queue;
>> +    exec.address = batch_addr;
>> +
>> +    ret = __xe_exec(fd, &exec);
>> +    igt_assert_eq(ret, 0);
>> +    igt_assert(syncobj_wait(fd, &syncobj, 1, INT64_MAX, 0, NULL));
>> +
>> +    munmap(data, data_size);
>> +    data = xe_bo_map(fd, bo, data_size);
>> +    igt_assert(data != MAP_FAILED);
>> +    igt_assert_eq(data[0], PURGEABLE_TEST_PATTERN);
>> +
>> +    /* Mark both VMAs DONTNEED */
>> +    retained = xe_vm_madvise_purgeable(fd, vm2, data_addr2, data_size,
>> +                       DRM_XE_VMA_PURGEABLE_STATE_DONTNEED);
>> +    igt_assert_eq(retained, 1);
>> +
>> +    /* Trigger pressure - BO should be purged */
>> +    trigger_memory_pressure(fd, vm1);
>> +
>> +    retained = xe_vm_madvise_purgeable(fd, vm2, data_addr2, data_size,
>> +                       DRM_XE_VMA_PURGEABLE_STATE_WILLNEED);
>> +
>> +    if (retained != 0)
>> +        goto out;
>> +
>> +    /* GPU workload - should fail or succeed with NULL rebind */
>> +    batch[3] = PURGEABLE_DEAD_PATTERN;
>> +
>> +    ret = __xe_exec(fd, &exec);
>> +    if (ret == 0) {
>> +        /* Exec succeeded, wait for completion before cleanup */
>> +        syncobj_wait(fd, &syncobj, 1, INT64_MAX, 0, NULL);
>> +    }
>> +
>> +out:
>> +    munmap(data, data_size);
>> +    munmap(batch, batch_size);
>> +    gem_close(fd, bo);
>> +    gem_close(fd, batch_bo);
>> +    syncobj_destroy(fd, syncobj);
>> +    xe_exec_queue_destroy(fd, bind_engine);
>> +    xe_exec_queue_destroy(fd, exec_queue);
>> +    xe_vm_destroy(fd, vm1);
>> +    xe_vm_destroy(fd, vm2);
>> +
>> +    if (retained != 0)
> This condition is not required here, already you have checked above 
> and then jump in goto: 


This also handles the normal case. Without any jumps in the working 
flow, we check after cleanup whether it was purged, and if not, we skip it


Thanks,
Arvind

>> +        igt_skip("Unable to induce purge on this platform/config");
>> +}
>> +
>>   int igt_main()
>>   {
>>       struct drm_xe_engine_class_instance *hwe;
>> @@ -702,6 +823,12 @@ int igt_main()
>>               break;
>>           }
>>   +    igt_subtest("per-vma-protection")
>> +        xe_for_each_engine(fd, hwe) {
>> +            test_per_vma_protection(fd, hwe);
>> +            break;
>> +        }
>> +
>>       igt_fixture() {
>>           xe_device_put(fd);
>>           drm_close_driver(fd);
>
> with above change LGTM:
>
> Reviewed-by: Nishit Sharma <nishit.sharma@intel.com>
>