Re: [PATCH i-g-t v6 6/9] tests/intel/xe_madvise: Add dontneed-before-exec subtest

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

On 06-04-2026 22:18, Sharma, Nishit wrote:
>
> On 3/25/2026 6:14 PM, Arvind Yadav wrote:
>> This test validates GPU execution behavior when a data BO is purged
>> before submission. The test creates a batch that writes to a data BO,
>> purges the data BO (while keeping the batch BO valid to avoid GPU
>> reset), then submits for execution. With VM_CREATE_FLAG_SCRATCH_PAGE,
>> the GPU write may succeed by landing on scratch memory instead of the
>> purged BO, demonstrating graceful handling of purged memory during
>> GPU operations.
>>
>> v4:
>>    - Added proper resource cleanup before calling igt_skip(). (Nishit)
>>
>> Cc: Nishit Sharma <nishit.sharma@intel.com>
>> Cc: Matthew Brost <matthew.brost@intel.com>
>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
>> Reviewed-by: Pravalika Gurram <pravalika.gurram@intel.com>
>> Signed-off-by: Arvind Yadav <arvind.yadav@intel.com>
>> ---
>>   tests/intel/xe_madvise.c | 148 +++++++++++++++++++++++++++++++++++++++
>>   1 file changed, 148 insertions(+)
>>
>> diff --git a/tests/intel/xe_madvise.c b/tests/intel/xe_madvise.c
>> index 9a157de1d..d126db9ed 100644
>> --- a/tests/intel/xe_madvise.c
>> +++ b/tests/intel/xe_madvise.c
>> @@ -27,7 +27,11 @@ static bool xe_has_purgeable_support(int fd)
>>     /* Purgeable test constants */
>>   #define PURGEABLE_ADDR        0x1a0000
>> +#define PURGEABLE_BATCH_ADDR    0x3c0000
>>   #define PURGEABLE_BO_SIZE    4096
>> +#define PURGEABLE_FENCE_VAL    0xbeef
>> +#define PURGEABLE_TEST_PATTERN    0xc0ffee
>> +#define PURGEABLE_DEAD_PATTERN    0xdead
>>     /**
>>    * trigger_memory_pressure - Fill VRAM/RAM + 50% to force purgeable 
>> reclaim
>> @@ -206,6 +210,62 @@ static void test_purged_mmap_blocked(int fd)
>>       xe_vm_destroy(fd, vm);
>>   }
>>   +/**
>> + * purgeable_setup_batch_and_data - Setup VM with batch and data BOs 
>> for GPU exec
>> + * @fd: DRM file descriptor
>> + * @vm: Output VM handle
>> + * @bind_engine: Output bind engine handle
>> + * @batch_bo: Output batch BO handle
>> + * @data_bo: Output data BO handle
>> + * @batch: Output batch buffer pointer
>> + * @data: Output data buffer pointer
>> + * @batch_addr: Batch virtual address
>> + * @data_addr: Data virtual address
>> + * @batch_size: Batch buffer size
>> + * @data_size: Data buffer size
>> + *
>> + * Helper to create VM, bind engine, batch and data BOs, and bind them.
>> + */
>> +static void purgeable_setup_batch_and_data(int fd, uint32_t *vm,
>> +                       uint32_t *bind_engine,
>> +                       uint32_t *batch_bo,
>> +                       uint32_t *data_bo,
>> +                       uint32_t **batch,
>> +                       uint32_t **data,
>> +                       uint64_t batch_addr,
>> +                       uint64_t data_addr,
>> +                       size_t batch_size,
>> +                       size_t data_size)
>> +{
>> +    struct drm_xe_sync sync = {
>> +        .type = DRM_XE_SYNC_TYPE_USER_FENCE,
>> +        .flags = DRM_XE_SYNC_FLAG_SIGNAL,
>> +        .timeline_value = PURGEABLE_FENCE_VAL,
>> +    };
>> +    uint64_t vm_sync = 0;
>> +
>> +    *vm = xe_vm_create(fd, DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE, 0);
>> +    *bind_engine = xe_bind_exec_queue_create(fd, *vm, 0);
>> +
>> +    /* Create and bind batch BO */
>> +    *batch_bo = xe_bo_create(fd, *vm, batch_size, 
>> vram_if_possible(fd, 0),
>> +                 DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
>> +    *batch = xe_bo_map(fd, *batch_bo, batch_size);
>> +
>> +    sync.addr = to_user_pointer(&vm_sync);
>> +    xe_vm_bind_async(fd, *vm, *bind_engine, *batch_bo, 0, 
>> batch_addr, batch_size, &sync, 1);
>> +    xe_wait_ufence(fd, &vm_sync, PURGEABLE_FENCE_VAL, 0, NSEC_PER_SEC);
>> +
>> +    /* Create and bind data BO */
>> +    *data_bo = xe_bo_create(fd, *vm, data_size, vram_if_possible(fd, 
>> 0),
>> +                DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
>> +    *data = xe_bo_map(fd, *data_bo, data_size);
>> +
>> +    vm_sync = 0;
>> +    xe_vm_bind_async(fd, *vm, *bind_engine, *data_bo, 0, data_addr, 
>> data_size, &sync, 1);
>> +    xe_wait_ufence(fd, &vm_sync, PURGEABLE_FENCE_VAL, 0, NSEC_PER_SEC);
>> +}
>> +
>>   /**
>>    * SUBTEST: dontneed-before-mmap
>>    * Description: Mark BO as DONTNEED before mmap, verify mmap() 
>> fails with -EBUSY
>> @@ -302,6 +362,88 @@ static void test_dontneed_after_mmap(int fd)
>>       xe_vm_destroy(fd, vm);
>>   }
>>   +/**
>> + * SUBTEST: dontneed-before-exec
>> + * Description: Mark BO as DONTNEED before GPU exec, verify GPU 
>> behavior with SCRATCH_PAGE
>> + * Test category: functionality test
>> + */
>> +static void test_dontneed_before_exec(int fd, struct 
>> drm_xe_engine_class_instance *hwe)
>> +{
>> +    uint32_t vm, exec_queue, bo, batch_bo, bind_engine;
>> +    uint64_t data_addr = PURGEABLE_ADDR;
>> +    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[1] = {
>> +        { .type = DRM_XE_SYNC_TYPE_USER_FENCE,
>> +          .flags = DRM_XE_SYNC_FLAG_SIGNAL,
>> +          .timeline_value = PURGEABLE_FENCE_VAL },
>> +    };
>> +    struct drm_xe_exec exec = {
>> +        .num_batch_buffer = 1,
>> +        .num_syncs = 1,
>> +        .syncs = to_user_pointer(sync),
>> +    };
>> +    uint32_t *data, *batch;
>> +    uint64_t vm_sync = 0;
>> +    int b, ret;
>> +
>> +    purgeable_setup_batch_and_data(fd, &vm, &bind_engine, &batch_bo,
>> +                       &bo, &batch, &data, batch_addr,
>> +                       data_addr, batch_size, data_size);
>> +
>> +    /* Prepare batch */
>> +    b = 0;
>> +    batch[b++] = MI_STORE_DWORD_IMM_GEN4;
>> +    batch[b++] = data_addr;
>> +    batch[b++] = data_addr >> 32;
>> +    batch[b++] = PURGEABLE_DEAD_PATTERN;
>> +    batch[b++] = MI_BATCH_BUFFER_END;
>> +
>> +    /* Phase 1: Purge data BO, batch BO still valid */
>> +    if (!purgeable_mark_and_verify_purged(fd, vm, data_addr, 
>> data_size)) {
> Same comment as in patch-4/9.


Fixed — purgeable_mark_and_verify_purged() now handles retained == 0 
from DONTNEED as already-purged success

>> +        munmap(data, data_size);
>> +        munmap(batch, batch_size);
>> +        gem_close(fd, bo);
>> +        gem_close(fd, batch_bo);
>> +        xe_exec_queue_destroy(fd, bind_engine);
>> +        xe_vm_destroy(fd, vm);
>> +        igt_skip("Unable to induce purge on this platform/config");
>> +    }
>> +
>> +    exec_queue = xe_exec_queue_create(fd, vm, hwe, 0);
>> +    exec.exec_queue_id = exec_queue;
>> +    exec.address = batch_addr;
>> +
>> +    vm_sync = 0;
>> +    sync[0].addr = to_user_pointer(&vm_sync);
>> +
>> +    /*
>> +     * VM has SCRATCH_PAGE — exec may succeed with the GPU write
>> +     * landing on scratch instead of the purged data BO.
>> +     */
>> +    ret = __xe_exec(fd, &exec);
> You can use xe_exec(fd, &exec) which is internally calling __xe_exec() 
> and after use xe_wait_fence()


We can't use xe_exec() here,  the data BO has been purged, so exec may 
legitimately fail. xe_exec() asserts success internally, which would 
cause a false test failure. We need __xe_exec() to handle both outcomes: 
success (GPU write lands on scratch page) or failure (exec rejected for 
purged BO).


Thanks,
Arvind

>> +    if (ret == 0) {
>> +        int64_t timeout = NSEC_PER_SEC;
>> +
>> +        __xe_wait_ufence(fd, &vm_sync, PURGEABLE_FENCE_VAL,
>> +                 exec_queue, &timeout);
>> +    }
>> +
>> +    /*
>> +     * Don't purge the batch BO — GPU would fetch zeroed scratch
>> +     * instructions and trigger an engine reset.
>> +     */
>> +
>> +    munmap(data, data_size);
>> +    munmap(batch, batch_size);
>> +    gem_close(fd, bo);
>> +    gem_close(fd, batch_bo);
>> +    xe_exec_queue_destroy(fd, bind_engine);
>> +    xe_exec_queue_destroy(fd, exec_queue);
>> +    xe_vm_destroy(fd, vm);
>> +}
>> +
>>   int igt_main()
>>   {
>>       struct drm_xe_engine_class_instance *hwe;
>> @@ -332,6 +474,12 @@ int igt_main()
>>               break;
>>           }
>>   +    igt_subtest("dontneed-before-exec")
>> +        xe_for_each_engine(fd, hwe) {
>> +            test_dontneed_before_exec(fd, hwe);
>> +            break;
>> +        }
>> +
>>       igt_fixture() {
>>           xe_device_put(fd);
>>           drm_close_driver(fd);