/******************************************************************************* Copyright (c) 2015-2022 NVidia Corporation Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. *******************************************************************************/ #ifndef __UVM_TEST_IOCTL_H__ #define __UVM_TEST_IOCTL_H__ #include "uvm_types.h" #include "uvm_ioctl.h" #include "nv_uvm_types.h" #define UVM_TEST_SKIP_MIGRATE_VMA UVM_TEST_IOCTL_BASE(103) typedef struct { NvBool skip; // In NV_STATUS rmStatus; // Out } UVM_TEST_SKIP_MIGRATE_VMA_PARAMS; #ifdef __cplusplus extern "C" { #endif // Offset the test ioctl to leave space for the api ones #define UVM_TEST_IOCTL_BASE(i) UVM_IOCTL_BASE(200 + i) #define UVM_TEST_GET_GPU_REF_COUNT UVM_TEST_IOCTL_BASE(0) typedef struct { // In params NvProcessorUuid gpu_uuid; NvU32 swizz_id; // Out params NvU64 ref_count NV_ALIGN_BYTES(8); NV_STATUS rmStatus; } UVM_TEST_GET_GPU_REF_COUNT_PARAMS; #define UVM_TEST_RNG_SANITY UVM_TEST_IOCTL_BASE(1) typedef struct { NV_STATUS rmStatus; } UVM_TEST_RNG_SANITY_PARAMS; #define UVM_TEST_RANGE_TREE_DIRECTED UVM_TEST_IOCTL_BASE(2) typedef struct { NV_STATUS rmStatus; } UVM_TEST_RANGE_TREE_DIRECTED_PARAMS; #define UVM_TEST_RANGE_TREE_RANDOM UVM_TEST_IOCTL_BASE(3) typedef struct { NvU32 seed; // In NvU64 main_iterations NV_ALIGN_BYTES(8); // In NvU32 verbose; // In // Probability (0-100) // // When the test starts up, it adds and splits ranges with high_probability. // Eventually when adds and splits fail too often, they'll invert their // probability to 100 - high_probability. They'll switch back when the tree // becomes too empty. // // This can be < 50, but the test will not be very interesting. NvU32 high_probability; // In // Probability (0-100) // // Every main iteration a group of operations is selected with this // probability. The group consists of either "add/remove" or "split/merge." // This is the chance that the "add/remove" group is selected each // iteration. NvU32 add_remove_shrink_group_probability; // Probability (0-100) // // Probability of picking the shrink operation instead of add/remove if the // add/remove/shrink group of operations is selected. NvU32 shrink_probability; // The number of collision verification checks to make each main iteration NvU32 collision_checks; // In // The number of tree iterator verification checks to make each main // iteration. NvU32 iterator_checks; // In // Highest range value to use NvU64 max_end NV_ALIGN_BYTES(8); // In // Maximum number of range nodes to put in the tree NvU64 max_ranges NV_ALIGN_BYTES(8); // In // Maximum number of range nodes to add or remove at one time NvU64 max_batch_count NV_ALIGN_BYTES(8); // In // add, split, and merge operations all operate on randomly-selected ranges // or nodes. It's possible, sometimes even likely, that the operation cannot // be performed on the selected range or node. // // For example, when a range node is added its range is selected at random // without regard to range nodes already in the tree. If a collision occurs // when the test attempts to add that node to the tree, a new, smaller // random range is selected and the attempt is made again. // // max_attempts is the maximum number of times to keep picking new ranges or // nodes before giving up on the operation. NvU32 max_attempts; // In struct { NvU64 total_adds NV_ALIGN_BYTES(8); NvU64 failed_adds NV_ALIGN_BYTES(8); NvU64 max_attempts_add NV_ALIGN_BYTES(8); NvU64 total_removes NV_ALIGN_BYTES(8); NvU64 total_splits NV_ALIGN_BYTES(8); NvU64 failed_splits NV_ALIGN_BYTES(8); NvU64 max_attempts_split NV_ALIGN_BYTES(8); NvU64 total_merges NV_ALIGN_BYTES(8); NvU64 failed_merges NV_ALIGN_BYTES(8); NvU64 max_attempts_merge NV_ALIGN_BYTES(8); NvU64 total_shrinks NV_ALIGN_BYTES(8); NvU64 failed_shrinks NV_ALIGN_BYTES(8); } stats; // Out NV_STATUS rmStatus; // Out } UVM_TEST_RANGE_TREE_RANDOM_PARAMS; // Keep this in sync with uvm_va_range_type_t in uvm_va_range.h typedef enum { UVM_TEST_VA_RANGE_TYPE_INVALID = 0, UVM_TEST_VA_RANGE_TYPE_MANAGED, UVM_TEST_VA_RANGE_TYPE_EXTERNAL, UVM_TEST_VA_RANGE_TYPE_CHANNEL, UVM_TEST_VA_RANGE_TYPE_SKED_REFLECTED, UVM_TEST_VA_RANGE_TYPE_SEMAPHORE_POOL, UVM_TEST_VA_RANGE_TYPE_MAX } UVM_TEST_VA_RANGE_TYPE; typedef enum { UVM_TEST_RANGE_SUBTYPE_INVALID = 0, UVM_TEST_RANGE_SUBTYPE_UVM, UVM_TEST_RANGE_SUBTYPE_HMM, UVM_TEST_RANGE_SUBTYPE_MAX } UVM_TEST_RANGE_SUBTYPE; // Keep this in sync with uvm_read_duplication_t in uvm_va_range.h typedef enum { UVM_TEST_READ_DUPLICATION_UNSET = 0, UVM_TEST_READ_DUPLICATION_ENABLED, UVM_TEST_READ_DUPLICATION_DISABLED, UVM_TEST_READ_DUPLICATION_MAX } UVM_TEST_READ_DUPLICATION_POLICY; typedef struct { // Note: if this is a zombie or not owned by the calling process, the vma info // will not be filled out and is invalid. NvU64 vma_start NV_ALIGN_BYTES(8); // Out NvU64 vma_end NV_ALIGN_BYTES(8); // Out, inclusive NvBool is_zombie; // Out // Note: if this is a zombie, this field is meaningless. NvBool owned_by_calling_process; // Out NvU32 subtype; // Out (UVM_TEST_RANGE_SUBTYPE) } UVM_TEST_VA_RANGE_INFO_MANAGED; #define UVM_TEST_VA_RANGE_INFO UVM_TEST_IOCTL_BASE(4) typedef struct { NvU64 lookup_address NV_ALIGN_BYTES(8); // In // For HMM ranges va_range_start/end will contain the lookup address but not // neccessarily the maximal range over which the returned policy applies. // For example there could be adjacent ranges with the same policy, implying // the returned range could be as small as a page in the worst case for HMM. NvU64 va_range_start NV_ALIGN_BYTES(8); // Out NvU64 va_range_end NV_ALIGN_BYTES(8); // Out, inclusive NvU32 read_duplication; // Out (UVM_TEST_READ_DUPLICATION_POLICY) NvProcessorUuid preferred_location; // Out NvProcessorUuid accessed_by[UVM_MAX_PROCESSORS]; // Out NvU32 accessed_by_count; // Out NvU32 type; // Out (UVM_TEST_VA_RANGE_TYPE) union { UVM_TEST_VA_RANGE_INFO_MANAGED managed NV_ALIGN_BYTES(8); // Out // More here eventually }; // NV_ERR_INVALID_ADDRESS lookup_address doesn't match a UVM range NV_STATUS rmStatus; // Out } UVM_TEST_VA_RANGE_INFO_PARAMS; #define UVM_TEST_RM_MEM_SANITY UVM_TEST_IOCTL_BASE(5) typedef struct { // Out params NV_STATUS rmStatus; } UVM_TEST_RM_MEM_SANITY_PARAMS; #define UVM_TEST_GPU_SEMAPHORE_SANITY UVM_TEST_IOCTL_BASE(6) typedef struct { // Out params NV_STATUS rmStatus; } UVM_TEST_GPU_SEMAPHORE_SANITY_PARAMS; #define UVM_TEST_PEER_REF_COUNT UVM_TEST_IOCTL_BASE(7) typedef struct { // In params NvProcessorUuid gpu_uuid_1; NvProcessorUuid gpu_uuid_2; // Out params NV_STATUS rmStatus; NvU64 ref_count NV_ALIGN_BYTES(8); } UVM_TEST_PEER_REF_COUNT_PARAMS; // Force an existing UVM range to split. split_address will be the new end of // the existing range. A new range will be created covering // [split_address+1, original end]. // // Error returns: // NV_ERR_INVALID_ADDRESS // - split_address+1 isn't page-aligned // - split_address doesn't match a splittable UVM range // - The range cannot be split at split_address because split_address is // already the end of the range. #define UVM_TEST_VA_RANGE_SPLIT UVM_TEST_IOCTL_BASE(8) typedef struct { NvU64 split_address NV_ALIGN_BYTES(8); // In NV_STATUS rmStatus; // Out } UVM_TEST_VA_RANGE_SPLIT_PARAMS; // Forces the next range split on the range covering lookup_address to fail with // an out-of-memory error. Only the next split will fail. Subsequent ones will // succeed. The split can come from any source, such as vma splitting or // UVM_TEST_VA_RANGE_SPLIT. // // Error returns: // NV_ERR_INVALID_ADDRESS // - lookup_address doesn't match a UVM range #define UVM_TEST_VA_RANGE_INJECT_SPLIT_ERROR UVM_TEST_IOCTL_BASE(9) typedef struct { NvU64 lookup_address NV_ALIGN_BYTES(8); // In NV_STATUS rmStatus; // Out } UVM_TEST_VA_RANGE_INJECT_SPLIT_ERROR_PARAMS; #define UVM_TEST_PAGE_TREE UVM_TEST_IOCTL_BASE(10) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_PAGE_TREE_PARAMS; // Given a VA and a target processor, forcibly set that processor's mapping to // the VA to the given permissions. This may require changing other processors' // mappings. For example, setting an atomic mapping for a given GPU might make // other GPUs' mappings read-only. // // If the mapping changes from invalid to anything else, this call always // attempts to create direct mappings from the given processor to the current // physical memory backing the target address. If a direct mapping cannot be // created, or no physical memory currently backs the VA, // NV_ERR_INVALID_OPERATION is returned. // // uuid is allowed to be NV_PROCESSOR_UUID_CPU_DEFAULT. // // Error returns: // NV_ERR_INVALID_DEVICE // - uuid is an unknown value // - uuid is a GPU that hasn't been registered with this process // // NV_ERR_INVALID_ADDRESS // - VA is unknown to the kernel // - VA isn't aligned to the system page size // // NV_ERR_INVALID_STATE // - A mapping for va can't be accessed because it belongs to another process // // NV_ERR_INVALID_ARGUMENT // - mapping is not a valid enum value // // NV_ERR_INVALID_ACCESS_TYPE // - The mapping permissions aren't logically allowed. For example, // UVM_TEST_PTE_MAPPING_READ_WRITE can't be set on a read-only mapping. // // NV_ERR_INVALID_OPERATION // - mapping is not UVM_TEST_PTE_MAPPING_INVALID, and a direct mapping from the // given processor to the physical memory currently backing VA cannot be // created. #define UVM_TEST_CHANGE_PTE_MAPPING UVM_TEST_IOCTL_BASE(11) typedef enum { UVM_TEST_PTE_MAPPING_INVALID = 0, UVM_TEST_PTE_MAPPING_READ_ONLY, UVM_TEST_PTE_MAPPING_READ_WRITE, UVM_TEST_PTE_MAPPING_READ_WRITE_ATOMIC, UVM_TEST_PTE_MAPPING_MAX } UVM_TEST_PTE_MAPPING; typedef struct { NvProcessorUuid uuid NV_ALIGN_BYTES(8); // In NvU64 va NV_ALIGN_BYTES(8); // In NvU32 mapping; // In (UVM_TEST_PTE_MAPPING) NV_STATUS rmStatus; // Out } UVM_TEST_CHANGE_PTE_MAPPING_PARAMS; #define UVM_TEST_TRACKER_SANITY UVM_TEST_IOCTL_BASE(12) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_TRACKER_SANITY_PARAMS; #define UVM_TEST_PUSH_SANITY UVM_TEST_IOCTL_BASE(13) typedef struct { NvBool skipTimestampTest; // In NV_STATUS rmStatus; // Out } UVM_TEST_PUSH_SANITY_PARAMS; #define UVM_TEST_CHANNEL_SANITY UVM_TEST_IOCTL_BASE(14) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_CHANNEL_SANITY_PARAMS; typedef enum { UVM_TEST_CHANNEL_STRESS_MODE_NOOP_PUSH = 0, UVM_TEST_CHANNEL_STRESS_MODE_UPDATE_CHANNELS, UVM_TEST_CHANNEL_STRESS_MODE_STREAM, } UVM_TEST_CHANNEL_STRESS_MODE; #define UVM_TEST_CHANNEL_STRESS UVM_TEST_IOCTL_BASE(15) typedef struct { NvU32 mode; // In // Number of iterations: // mode == NOOP_PUSH: number of noop pushes // mode == UPDATE_CHANNELS: number of updates // mode == STREAM: number of iterations per stream NvU32 iterations; NvU32 num_streams; // In, used only for mode == UVM_TEST_CHANNEL_STRESS_MODE_STREAM NvU32 seed; // In NvU32 verbose; // In NV_STATUS rmStatus; // Out } UVM_TEST_CHANNEL_STRESS_PARAMS; #define UVM_TEST_CE_SANITY UVM_TEST_IOCTL_BASE(16) typedef struct { NvBool skipTimestampTest; // In NV_STATUS rmStatus; // Out } UVM_TEST_CE_SANITY_PARAMS; #define UVM_TEST_VA_BLOCK_INFO UVM_TEST_IOCTL_BASE(17) // See UVM_VA_BLOCK_SIZE in uvm_va_block.h for an explanation of this number #define UVM_TEST_VA_BLOCK_SIZE (2ull*1024*1024) typedef struct { NvU64 lookup_address NV_ALIGN_BYTES(8); // In NvU64 va_block_start NV_ALIGN_BYTES(8); // Out NvU64 va_block_end NV_ALIGN_BYTES(8); // Out, inclusive // NV_ERR_INVALID_ADDRESS lookup_address doesn't match a UVM range // // NV_ERR_OBJECT_NOT_FOUND lookup_address matched a UVM range on this file // but the corresponding block has not yet been // populated NV_STATUS rmStatus; // Out } UVM_TEST_VA_BLOCK_INFO_PARAMS; #define UVM_TEST_LOCK_SANITY UVM_TEST_IOCTL_BASE(18) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_LOCK_SANITY_PARAMS; #define UVM_TEST_PERF_UTILS_SANITY UVM_TEST_IOCTL_BASE(19) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_PERF_UTILS_SANITY_PARAMS; #define UVM_TEST_KVMALLOC UVM_TEST_IOCTL_BASE(20) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_KVMALLOC_PARAMS; #define UVM_TEST_PMM_QUERY UVM_TEST_IOCTL_BASE(21) typedef enum { // Get the value of valid user allocations as key UVM_TEST_CHUNK_SIZE_GET_USER_SIZE } uvm_test_pmm_query_key_t; typedef struct { // In params NvProcessorUuid gpu_uuid; NvU64 key; // Out params NvU64 value; NV_STATUS rmStatus; } UVM_TEST_PMM_QUERY_PARAMS; #define UVM_TEST_PMM_CHECK_LEAK UVM_TEST_IOCTL_BASE(22) typedef struct { NvProcessorUuid gpu_uuid; // In NvU64 chunk_size; // In NvS64 alloc_limit; // In. Number of chunks to allocate. -1 means unlimited NvU64 allocated; // Out. Number of chunks actually allocated NV_STATUS rmStatus; // Out } UVM_TEST_PMM_CHECK_LEAK_PARAMS; #define UVM_TEST_PERF_EVENTS_SANITY UVM_TEST_IOCTL_BASE(23) typedef struct { // Out params NV_STATUS rmStatus; } UVM_TEST_PERF_EVENTS_SANITY_PARAMS; #define UVM_TEST_PERF_MODULE_SANITY UVM_TEST_IOCTL_BASE(24) typedef struct { // In params NvU64 range_address NV_ALIGN_BYTES(8); NvU32 range_size; // Out params NV_STATUS rmStatus; } UVM_TEST_PERF_MODULE_SANITY_PARAMS; #define UVM_TEST_RANGE_ALLOCATOR_SANITY UVM_TEST_IOCTL_BASE(25) typedef struct { // In params NvU32 verbose; NvU32 seed; NvU32 iters; // Out params NV_STATUS rmStatus; } UVM_TEST_RANGE_ALLOCATOR_SANITY_PARAMS; #define UVM_TEST_GET_RM_PTES UVM_TEST_IOCTL_BASE(26) typedef enum { UVM_TEST_GET_RM_PTES_SINGLE_GPU = 0, UVM_TEST_GET_RM_PTES_MULTI_GPU_SUPPORTED, UVM_TEST_GET_RM_PTES_MULTI_GPU_SLI_SUPPORTED, UVM_TEST_GET_RM_PTES_MULTI_GPU_NOT_SUPPORTED, UVM_TEST_GET_RM_PTES_MAX } UVM_TEST_PTE_RM_PTES_TEST_MODE; typedef struct { // In NvS32 rmCtrlFd; // For future use. (security check) NvHandle hClient; NvHandle hMemory; NvU32 test_mode; // (UVM_TEST_PTE_RM_PTES_TEST_MODE) NvU64 size NV_ALIGN_BYTES(8); NvProcessorUuid gpu_uuid; // Out NV_STATUS rmStatus; } UVM_TEST_GET_RM_PTES_PARAMS; #define UVM_TEST_FAULT_BUFFER_FLUSH UVM_TEST_IOCTL_BASE(27) typedef struct { NvU64 iterations; // In NV_STATUS rmStatus; // Out } UVM_TEST_FAULT_BUFFER_FLUSH_PARAMS; #define UVM_TEST_INJECT_TOOLS_EVENT UVM_TEST_IOCTL_BASE(28) typedef struct { // In params UvmEventEntry entry; // contains only NvUxx types NvU32 count; // Out param NV_STATUS rmStatus; } UVM_TEST_INJECT_TOOLS_EVENT_PARAMS; #define UVM_TEST_INCREMENT_TOOLS_COUNTER UVM_TEST_IOCTL_BASE(29) typedef struct { // In params NvU64 amount NV_ALIGN_BYTES(8); // amount to increment NvU32 counter; // name of counter NvProcessorUuid processor; NvU32 count; // number of times to increment // Out param NV_STATUS rmStatus; } UVM_TEST_INCREMENT_TOOLS_COUNTER_PARAMS; #define UVM_TEST_MEM_SANITY UVM_TEST_IOCTL_BASE(30) typedef struct { // Out params NV_STATUS rmStatus; } UVM_TEST_MEM_SANITY_PARAMS; #define UVM_TEST_MAKE_CHANNEL_STOPS_IMMEDIATE UVM_TEST_IOCTL_BASE(32) typedef struct { // Out params NV_STATUS rmStatus; } UVM_TEST_MAKE_CHANNEL_STOPS_IMMEDIATE_PARAMS; // Inject an error into the VA block covering the lookup_address // // If page_table_allocation_retry_force_count is non-0 then the next count // page table allocations under the VA block will be forced to do // allocation-retry. // // If user_pages_allocation_retry_force_count is non-0 then the next count user // memory allocations under the VA block will be forced to do allocation-retry. // // If cpu_pages_allocation_error_count is not zero, the subsequent operations // that need to allocate CPU pages will fail with NV_ERR_NO_MEMORY for // cpu_pages_allocation_error_count times. If cpu_pages_allocation_error_count // is equal to ~0U, the count is infinite. // // If eviction_failure is NV_TRUE, the next eviction attempt from the VA block // will fail with NV_ERR_NO_MEMORY. // // If populate_failure is NV_TRUE, a retry error will be injected after the next // successful user memory allocation under the VA block but before that // allocation is used by the block. This is similar to // user_pages_allocation_retry_force_count, but the injection point simulates // driver metadata allocation failure. // // Error returns: // NV_ERR_INVALID_ADDRESS // - lookup_address doesn't match a UVM range #define UVM_TEST_VA_BLOCK_INJECT_ERROR UVM_TEST_IOCTL_BASE(33) typedef struct { NvU64 lookup_address NV_ALIGN_BYTES(8); // In NvU32 page_table_allocation_retry_force_count; // In NvU32 user_pages_allocation_retry_force_count; // In NvU32 cpu_chunk_allocation_size_mask; // In NvU32 cpu_pages_allocation_error_count; // In NvBool eviction_error; // In NvBool populate_error; // In NV_STATUS rmStatus; // Out } UVM_TEST_VA_BLOCK_INJECT_ERROR_PARAMS; #define UVM_TEST_PEER_IDENTITY_MAPPINGS UVM_TEST_IOCTL_BASE(34) typedef struct { // In params NvProcessorUuid gpuA; NvProcessorUuid gpuB; // Out param NV_STATUS rmStatus; } UVM_TEST_PEER_IDENTITY_MAPPINGS_PARAMS; #define UVM_TEST_VA_RESIDENCY_INFO UVM_TEST_IOCTL_BASE(35) typedef struct { NvU64 lookup_address NV_ALIGN_BYTES(8); // In // Whether to wait on the block tracker before returning. Fields like // resident_on and mapped_on represent state which will be valid when the // block tracker is complete. If is_async is true, then those fields will // still be filled out as if the tracker is done, but the actual residency // or mapping changes may not have been performed yet. NvBool is_async; // In // Array of processors which have a resident copy of the page containing // lookup_address. NvProcessorUuid resident_on[UVM_MAX_PROCESSORS]; // Out NvU32 resident_on_count; // Out // The size of the physical allocation backing lookup_address. Only the // system-page-sized portion of this allocation which contains // lookup_address is guaranteed to be resident on the corresponding // processor. NvU32 resident_physical_size[UVM_MAX_PROCESSORS]; // Out // The physical address of the physical allocation backing lookup_address. NvU64 resident_physical_address[UVM_MAX_PROCESSORS] NV_ALIGN_BYTES(8); // Out // Array of processors which have a virtual mapping covering lookup_address. NvProcessorUuid mapped_on[UVM_MAX_PROCESSORS]; // Out NvU32 mapping_type[UVM_MAX_PROCESSORS]; // Out NvU64 mapping_physical_address[UVM_MAX_PROCESSORS] NV_ALIGN_BYTES(8); // Out NvU32 mapped_on_count; // Out // The size of the virtual mapping covering lookup_address on each // mapped_on processor. NvU32 page_size[UVM_MAX_PROCESSORS]; // Out // Array of processors which have physical memory populated that would back // lookup_address if it was resident. NvProcessorUuid populated_on[UVM_MAX_PROCESSORS]; // Out NvU32 populated_on_count; // Out NV_STATUS rmStatus; // Out } UVM_TEST_VA_RESIDENCY_INFO_PARAMS; #define UVM_TEST_PMM_ASYNC_ALLOC UVM_TEST_IOCTL_BASE(36) typedef struct { NvProcessorUuid gpu_uuid; // In NvU32 num_chunks; // In NvU32 num_work_iterations; // In NV_STATUS rmStatus; // Out } UVM_TEST_PMM_ASYNC_ALLOC_PARAMS; typedef enum { UVM_TEST_PREFETCH_FILTERING_MODE_FILTER_ALL, // Disable all prefetch faults UVM_TEST_PREFETCH_FILTERING_MODE_FILTER_NONE, // Enable all prefetch faults } UvmTestPrefetchFilteringMode; #define UVM_TEST_SET_PREFETCH_FILTERING UVM_TEST_IOCTL_BASE(37) typedef struct { NvProcessorUuid gpu_uuid; // In NvU32 filtering_mode; // In (UvmTestPrefetchFilteringMode) NV_STATUS rmStatus; // Out } UVM_TEST_SET_PREFETCH_FILTERING_PARAMS; typedef enum { UvmTestPmmSanityModeFull = 1, UvmTestPmmSanityModeBasic = 2, } UvmTestPmmSanityMode; #define UVM_TEST_PMM_SANITY UVM_TEST_IOCTL_BASE(40) typedef struct { // Test mode of type UvmTestPmmSanityMode NvU32 mode; // In NV_STATUS rmStatus; // Out } UVM_TEST_PMM_SANITY_PARAMS; typedef enum { UvmInvalidateTlbMemBarNone = 1, UvmInvalidateTlbMemBarSys = 2, UvmInvalidateTlbMemBarLocal = 3, } UvmInvalidateTlbMembarType; typedef enum { UvmInvalidatePageTableLevelAll = 1, UvmInvalidatePageTableLevelPte = 2, UvmInvalidatePageTableLevelPde0 = 3, UvmInvalidatePageTableLevelPde1 = 4, UvmInvalidatePageTableLevelPde2 = 5, UvmInvalidatePageTableLevelPde3 = 6, UvmInvalidatePageTableLevelPde4 = 7, } UvmInvalidatePageTableLevel; typedef enum { UvmTargetVaModeAll = 1, UvmTargetVaModeTargeted = 2, } UvmTargetVaMode; #define UVM_TEST_INVALIDATE_TLB UVM_TEST_IOCTL_BASE(41) typedef struct { // In params NvProcessorUuid gpu_uuid; NvU64 va NV_ALIGN_BYTES(8); NvU32 target_va_mode; // UvmTargetVaMode NvU32 page_table_level; // UvmInvalidatePageTableLevel NvU32 membar; // UvmInvalidateTlbMembarType NvBool disable_gpc_invalidate; // Out params NV_STATUS rmStatus; } UVM_TEST_INVALIDATE_TLB_PARAMS; #define UVM_TEST_VA_BLOCK UVM_TEST_IOCTL_BASE(42) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_VA_BLOCK_PARAMS; typedef enum { // Default policy based eviction // // Evicts a chunk that the default eviction path would pick. UvmTestEvictModeDefault = 1, // Virtual address based eviction // // Evicts the root chunk that the chunk backing the provided virtual address // belongs to. UvmTestEvictModeVirtual, // Physical address based eviction // // Evicts the root chunk covering the provided physical address. UvmTestEvictModePhysical, } UvmTestEvictMode; // Evict a chunk chosen according to one the test eviction modes specified // above. Eviction may not always be possible, but as long as the arguments are // valid NV_OK will be returned. To check whether eviction happened, the // chunk_was_evicted flag needs to be inspected. #define UVM_TEST_EVICT_CHUNK UVM_TEST_IOCTL_BASE(43) typedef struct { // The GPU to evict from, has to be registered in the VA space. NvProcessorUuid gpu_uuid; // In // UvmTestEvictMode NvU32 eviction_mode; // In // Virtual or physical address if evictionMode is UvmTestEvictModeVirtual or // UvmTestEvictModePhysical. NvU64 address NV_ALIGN_BYTES(8); // In // Flag indicating whether the eviction was performed. NvBool chunk_was_evicted; // Out // Physical address of the evicted root chunk. Notably 0 is a valid physical address. NvU64 evicted_physical_address NV_ALIGN_BYTES(8); // Out // For the virtual eviction mode, returns the size of the chunk that was // backing the virtual address before being evicted. 0 otherwise. NvU64 chunk_size_backing_virtual NV_ALIGN_BYTES(8); // Out NV_STATUS rmStatus; // Out } UVM_TEST_EVICT_CHUNK_PARAMS; typedef enum { // Flush deferred accessed by mappings UvmTestDeferredWorkTypeAcessedByMappings = 1, } UvmTestDeferredWorkType; #define UVM_TEST_FLUSH_DEFERRED_WORK UVM_TEST_IOCTL_BASE(44) typedef struct { // UvmTestDeferredWorkType NvU32 work_type; // In NV_STATUS rmStatus; // Out } UVM_TEST_FLUSH_DEFERRED_WORK_PARAMS; #define UVM_TEST_NV_KTHREAD_Q UVM_TEST_IOCTL_BASE(45) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_NV_KTHREAD_Q_PARAMS; typedef enum { UVM_TEST_PAGE_PREFETCH_POLICY_ENABLE = 0, UVM_TEST_PAGE_PREFETCH_POLICY_DISABLE, UVM_TEST_PAGE_PREFETCH_POLICY_MAX } UVM_TEST_PAGE_PREFETCH_POLICY; #define UVM_TEST_SET_PAGE_PREFETCH_POLICY UVM_TEST_IOCTL_BASE(46) typedef struct { NvU32 policy; // In (UVM_TEST_PAGE_PREFETCH_POLICY) NV_STATUS rmStatus; // Out } UVM_TEST_SET_PAGE_PREFETCH_POLICY_PARAMS; #define UVM_TEST_RANGE_GROUP_TREE UVM_TEST_IOCTL_BASE(47) typedef struct { NvU64 rangeGroupIds[4] NV_ALIGN_BYTES(8); // In NV_STATUS rmStatus; // Out } UVM_TEST_RANGE_GROUP_TREE_PARAMS; #define UVM_TEST_RANGE_GROUP_RANGE_INFO UVM_TEST_IOCTL_BASE(48) typedef struct { NvU64 lookup_address NV_ALIGN_BYTES(8); // In NvU64 range_group_range_start NV_ALIGN_BYTES(8); // Out NvU64 range_group_range_end NV_ALIGN_BYTES(8); // Out, inclusive NvU64 range_group_id NV_ALIGN_BYTES(8); // Out NvU32 range_group_present; // Out NV_STATUS rmStatus; // Out } UVM_TEST_RANGE_GROUP_RANGE_INFO_PARAMS; #define UVM_TEST_RANGE_GROUP_RANGE_COUNT UVM_TEST_IOCTL_BASE(49) typedef struct { NvU64 rangeGroupId NV_ALIGN_BYTES(8); // In NvU64 count NV_ALIGN_BYTES(8); // Out NV_STATUS rmStatus; // Out } UVM_TEST_RANGE_GROUP_RANGE_COUNT_PARAMS; #define UVM_TEST_GET_PREFETCH_FAULTS_REENABLE_LAPSE UVM_TEST_IOCTL_BASE(50) typedef struct { NvU32 reenable_lapse; // Out: Lapse in miliseconds NV_STATUS rmStatus; // Out } UVM_TEST_GET_PREFETCH_FAULTS_REENABLE_LAPSE_PARAMS; #define UVM_TEST_SET_PREFETCH_FAULTS_REENABLE_LAPSE UVM_TEST_IOCTL_BASE(51) typedef struct { NvU32 reenable_lapse; // In: Lapse in miliseconds NV_STATUS rmStatus; // Out } UVM_TEST_SET_PREFETCH_FAULTS_REENABLE_LAPSE_PARAMS; #define UVM_TEST_GET_KERNEL_VIRTUAL_ADDRESS UVM_TEST_IOCTL_BASE(52) typedef struct { NvU64 addr NV_ALIGN_BYTES(8); // Out NV_STATUS rmStatus; // Out } UVM_TEST_GET_KERNEL_VIRTUAL_ADDRESS_PARAMS; // Allocate and free memory directly from PMA with eviction enabled. This allows // to simulate RM-like allocations, but without the RM API lock serializing // everything. #define UVM_TEST_PMA_ALLOC_FREE UVM_TEST_IOCTL_BASE(53) typedef struct { NvProcessorUuid gpu_uuid; // In NvU32 page_size; NvBool contiguous; NvU64 num_pages NV_ALIGN_BYTES(8); // In NvU64 phys_begin NV_ALIGN_BYTES(8); // In NvU64 phys_end NV_ALIGN_BYTES(8); // In NvU32 nap_us_before_free; // In NV_STATUS rmStatus; // Out } UVM_TEST_PMA_ALLOC_FREE_PARAMS; // Allocate and free user memory directly from PMM with eviction enabled. // // Provides a direct way of exercising PMM allocs, eviction and frees of user // memory type. #define UVM_TEST_PMM_ALLOC_FREE_ROOT UVM_TEST_IOCTL_BASE(54) typedef struct { NvProcessorUuid gpu_uuid; // In NvU32 nap_us_before_free; // In NV_STATUS rmStatus; // Out } UVM_TEST_PMM_ALLOC_FREE_ROOT_PARAMS; // Inject a PMA eviction error after the specified number of chunks are // evicted. #define UVM_TEST_PMM_INJECT_PMA_EVICT_ERROR UVM_TEST_IOCTL_BASE(55) typedef struct { NvProcessorUuid gpu_uuid; // In NvU32 error_after_num_chunks; // In NV_STATUS rmStatus; // Out } UVM_TEST_PMM_INJECT_PMA_EVICT_ERROR_PARAMS; // Change configuration of access counters. This call will disable access // counters and reenable them using the new configuration. All previous // notifications will be lost // // The reconfiguration affects all VA spaces that rely on the access // counters information for the same GPU. To avoid conflicting configurations, // only one VA space is allowed to reconfigure the GPU at a time. // // Error returns: // NV_ERR_INVALID_STATE // - The GPU has already been reconfigured in a different VA space #define UVM_TEST_RECONFIGURE_ACCESS_COUNTERS UVM_TEST_IOCTL_BASE(56) typedef struct { NvProcessorUuid gpu_uuid; // In // Type UVM_ACCESS_COUNTER_GRANULARITY from nv_uvm_types.h NvU32 mimc_granularity; // In NvU32 momc_granularity; // In // Type UVM_ACCESS_COUNTER_USE_LIMIT from nv_uvm_types.h NvU32 mimc_use_limit; // In NvU32 momc_use_limit; // In NvU32 threshold; // In NvBool enable_mimc_migrations; // In NvBool enable_momc_migrations; // In NV_STATUS rmStatus; // Out } UVM_TEST_RECONFIGURE_ACCESS_COUNTERS_PARAMS; typedef enum { UVM_TEST_ACCESS_COUNTER_RESET_MODE_ALL = 0, UVM_TEST_ACCESS_COUNTER_RESET_MODE_TARGETED, UVM_TEST_ACCESS_COUNTER_RESET_MODE_MAX } UVM_TEST_ACCESS_COUNTER_RESET_MODE; typedef enum { UVM_TEST_ACCESS_COUNTER_TYPE_MIMC = 0, UVM_TEST_ACCESS_COUNTER_TYPE_MOMC, UVM_TEST_ACCESS_COUNTER_TYPE_MAX } UVM_TEST_ACCESS_COUNTER_TYPE; // Clear the contents of the access counters. This call supports different // modes for targeted/global resets. #define UVM_TEST_RESET_ACCESS_COUNTERS UVM_TEST_IOCTL_BASE(57) typedef struct { NvProcessorUuid gpu_uuid; // In // Type UVM_TEST_ACCESS_COUNTER_RESET_MODE NvU32 mode; // In // Type UVM_TEST_ACCESS_COUNTER_TYPE NvU32 counter_type; // In NvU32 bank; // In NvU32 tag; // In NV_STATUS rmStatus; // Out } UVM_TEST_RESET_ACCESS_COUNTERS_PARAMS; // Do not handle access counter notifications when they arrive. This call is // used to force an overflow of the access counter notification buffer #define UVM_TEST_SET_IGNORE_ACCESS_COUNTERS UVM_TEST_IOCTL_BASE(58) typedef struct { NvProcessorUuid gpu_uuid; // In NvBool ignore; // In NV_STATUS rmStatus; // Out } UVM_TEST_SET_IGNORE_ACCESS_COUNTERS_PARAMS; // Verifies that the given channel is registered under the UVM VA space of // vaSpaceFd. Returns NV_OK if so, NV_ERR_INVALID_CHANNEL if not. #define UVM_TEST_CHECK_CHANNEL_VA_SPACE UVM_TEST_IOCTL_BASE(59) typedef struct { NvProcessorUuid gpu_uuid; // In NvS32 rm_ctrl_fd; // In NvHandle client; // In NvHandle channel; // In NvU32 ve_id; // In NvS32 va_space_fd; // In NV_STATUS rmStatus; // Out } UVM_TEST_CHECK_CHANNEL_VA_SPACE_PARAMS; // // UvmTestEnableNvlinkPeerAccess // #define UVM_TEST_ENABLE_NVLINK_PEER_ACCESS UVM_TEST_IOCTL_BASE(60) typedef struct { NvProcessorUuid gpuUuidA; // IN NvProcessorUuid gpuUuidB; // IN NV_STATUS rmStatus; // OUT } UVM_TEST_ENABLE_NVLINK_PEER_ACCESS_PARAMS; // // UvmTestDisableNvlinkPeerAccess // #define UVM_TEST_DISABLE_NVLINK_PEER_ACCESS UVM_TEST_IOCTL_BASE(61) typedef struct { NvProcessorUuid gpuUuidA; // IN NvProcessorUuid gpuUuidB; // IN NV_STATUS rmStatus; // OUT } UVM_TEST_DISABLE_NVLINK_PEER_ACCESS_PARAMS; typedef enum { UVM_TEST_PAGE_THRASHING_POLICY_ENABLE = 0, UVM_TEST_PAGE_THRASHING_POLICY_DISABLE, UVM_TEST_PAGE_THRASHING_POLICY_MAX } UVM_TEST_PAGE_THRASHING_POLICY; // This ioctl returns the thrashing mitigation parameters on the current VA // space. Note that these values may change after a simulated/emulated GPU is // registered on the VA space. #define UVM_TEST_GET_PAGE_THRASHING_POLICY UVM_TEST_IOCTL_BASE(62) typedef struct { NvU32 policy; // Out (UVM_TEST_PAGE_THRASHING_POLICY) NvU64 nap_ns NV_ALIGN_BYTES(8); // Out NvU64 pin_ns NV_ALIGN_BYTES(8); // Out NvBool map_remote_on_native_atomics_fault; // Out NV_STATUS rmStatus; // Out } UVM_TEST_GET_PAGE_THRASHING_POLICY_PARAMS; #define UVM_TEST_SET_PAGE_THRASHING_POLICY UVM_TEST_IOCTL_BASE(63) typedef struct { NvU32 policy; // In (UVM_TEST_PAGE_THRASHING_POLICY) NvU64 pin_ns NV_ALIGN_BYTES(8); // In NV_STATUS rmStatus; // Out } UVM_TEST_SET_PAGE_THRASHING_POLICY_PARAMS; #define UVM_TEST_PMM_SYSMEM UVM_TEST_IOCTL_BASE(64) typedef struct { NvU64 range_address1 NV_ALIGN_BYTES(8); // In NvU64 range_address2 NV_ALIGN_BYTES(8); // In NV_STATUS rmStatus; // Out } UVM_TEST_PMM_SYSMEM_PARAMS; #define UVM_TEST_PMM_REVERSE_MAP UVM_TEST_IOCTL_BASE(65) typedef struct { NvProcessorUuid gpu_uuid; // In NvU64 range_address1 NV_ALIGN_BYTES(8); // In NvU64 range_address2 NV_ALIGN_BYTES(8); // In NvU64 range_size2 NV_ALIGN_BYTES(8); // In NV_STATUS rmStatus; // Out } UVM_TEST_PMM_REVERSE_MAP_PARAMS; #define UVM_TEST_PMM_INDIRECT_PEERS UVM_TEST_IOCTL_BASE(66) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_PMM_INDIRECT_PEERS_PARAMS; // Calls uvm_va_space_mm_retain on a VA space, operates on the mm, optionally // sleeps for a while, then releases the va_space_mm and returns. The idea is to // simulate retaining a va_space_mm from a thread like the GPU fault handler // which operates outside of the normal context of the VA space. #define UVM_TEST_VA_SPACE_MM_RETAIN UVM_TEST_IOCTL_BASE(67) typedef struct { // The kernel virtual address of the uvm_va_space on which to attempt // retain. This can be obtained via UVM_TEST_GET_KERNEL_VIRTUAL_ADDRESS. // // The reason to use this instead of looking it up from an fd as normal is // to allow testing of calling threads which race with UVM VA space destroy // (file close). We wouldn't be able to test that path if this was an fd. NvU64 va_space_ptr NV_ALIGN_BYTES(8); // In // User virtual address within the va_space_mm. If the va_space_mm is // successfully retained, this address is read once before sleeping and once // after (if sleep_us > 0). NvU64 addr NV_ALIGN_BYTES(8); // In // On success, this contains the value of addr read prior to the sleep. NvU64 val_before NV_ALIGN_BYTES(8); // In // On success, and if sleep_us > 0, this contains the value of addr read // after the sleep. This is invalid if sleep_us == 0. NvU64 val_after NV_ALIGN_BYTES(8); // In // Approximate duration for which to sleep with the va_space_mm retained. NvU64 sleep_us NV_ALIGN_BYTES(8); // In // NV_ERR_MISSING_TABLE_ENTRY va_space_ptr is not a valid VA space // NV_ERR_PAGE_TABLE_NOT_AVAIL Could not retain va_space_mm // (uvm_va_space_mm_retain returned NULL) // NV_ERR_INVALID_ADDRESS addr is invalid in va_space_mm NV_STATUS rmStatus; // Out } UVM_TEST_VA_SPACE_MM_RETAIN_PARAMS; #define UVM_TEST_PMM_CHUNK_WITH_ELEVATED_PAGE UVM_TEST_IOCTL_BASE(69) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_PMM_CHUNK_WITH_ELEVATED_PAGE_PARAMS; #define UVM_TEST_GET_GPU_TIME UVM_TEST_IOCTL_BASE(70) typedef struct { // GPU to query time from. GPU must have been previously registered NvProcessorUuid gpu_uuid; // In NvU64 timestamp_ns NV_ALIGN_BYTES(8); // Out NV_STATUS rmStatus; // Out } UVM_TEST_GET_GPU_TIME_PARAMS; // Check if access counters are enabled upon registration of the given GPU #define UVM_TEST_ACCESS_COUNTERS_ENABLED_BY_DEFAULT UVM_TEST_IOCTL_BASE(71) typedef struct { NvProcessorUuid gpu_uuid; // In NvBool enabled; // Out NV_STATUS rmStatus; // Out } UVM_TEST_ACCESS_COUNTERS_ENABLED_BY_DEFAULT_PARAMS; // Inject an error into the VA space // // If migrate_vma_allocation_fail_nth is greater than 0, the nth page // allocation within migrate_vma will fail. // // If va_block_allocation_fail_nth is greater than 0, the nth call to // uvm_va_block_find_create() will fail with NV_ERR_NO_MEMORY. #define UVM_TEST_VA_SPACE_INJECT_ERROR UVM_TEST_IOCTL_BASE(72) typedef struct { NvU32 migrate_vma_allocation_fail_nth; // In NvU32 va_block_allocation_fail_nth; // In NV_STATUS rmStatus; // Out } UVM_TEST_VA_SPACE_INJECT_ERROR_PARAMS; // Release to PMA all free root chunks #define UVM_TEST_PMM_RELEASE_FREE_ROOT_CHUNKS UVM_TEST_IOCTL_BASE(73) typedef struct { NvProcessorUuid gpu_uuid; // In NV_STATUS rmStatus; // Out } UVM_TEST_PMM_RELEASE_FREE_ROOT_CHUNKS_PARAMS; // Wait until all pending replayable faults have been processed. If there are // still pending packets when timeout_ns is reached, the ioctl returns // NV_ERR_TIMEOUT. // // This function should be called after the kernel producing the faults has been // synchronized. This should ensure that PUT != GET and faults will not be // missed even if the driver has not started to process them, yet. #define UVM_TEST_DRAIN_REPLAYABLE_FAULTS UVM_TEST_IOCTL_BASE(74) typedef struct { NvProcessorUuid gpu_uuid; // In NvU64 timeout_ns; // In NV_STATUS rmStatus; // Out } UVM_TEST_DRAIN_REPLAYABLE_FAULTS_PARAMS; // Get module config PMA batch size in bytes #define UVM_TEST_PMA_GET_BATCH_SIZE UVM_TEST_IOCTL_BASE(75) typedef struct { NvProcessorUuid gpu_uuid; // In NvU64 pma_batch_size; NV_ALIGN_BYTES(8) // Out NV_STATUS rmStatus; // Out } UVM_TEST_PMA_GET_BATCH_SIZE_PARAMS; // Request PMA's global statistics #define UVM_TEST_PMM_QUERY_PMA_STATS UVM_TEST_IOCTL_BASE(76) typedef struct { NvProcessorUuid gpu_uuid; // In UvmPmaStatistics pma_stats; // Out NV_STATUS rmStatus; // Out } UVM_TEST_PMM_QUERY_PMA_STATS_PARAMS; #define UVM_TEST_NUMA_GET_CLOSEST_CPU_NODE_TO_GPU UVM_TEST_IOCTL_BASE(77) typedef struct { NvProcessorUuid gpu_uuid; // In NvHandle client; // In NvHandle smc_part_ref; // In // On kernels with NUMA support, this entry contains the closest CPU NUMA // node to this GPU. Otherwise, the value will be -1. NvS32 node_id; // Out NV_STATUS rmStatus; // Out } UVM_TEST_NUMA_GET_CLOSEST_CPU_NODE_TO_GPU_PARAMS; // Test whether the bottom halves have run on the correct CPUs based on the // NUMA node locality of the GPU. // // Failure is reported if: // 1. The GPU has serviced faults but the mask tracking which CPUs the // bottom half ran on was empty, or // 2. The set of CPUs where the bottom half ran is not a subset of the CPUs // attached to the NUMA node. // // This IOCTL returns NV_OK on success, NV_ERR_INVALID_STATE on failure, or // NV_ERR_NOT_SUPPORTED if UVM thread affinity is not supported. #define UVM_TEST_NUMA_CHECK_AFFINITY UVM_TEST_IOCTL_BASE(78) typedef struct { NvProcessorUuid gpu_uuid; // In NV_STATUS rmStatus; // Out } UVM_TEST_NUMA_CHECK_AFFINITY_PARAMS; #define UVM_TEST_VA_SPACE_ADD_DUMMY_THREAD_CONTEXTS UVM_TEST_IOCTL_BASE(79) typedef struct { // Number of thread contexts to add per thread context table entry NvU32 num_dummy_thread_contexts; // In NV_STATUS rmStatus; // Out } UVM_TEST_VA_SPACE_ADD_DUMMY_THREAD_CONTEXTS_PARAMS; #define UVM_TEST_VA_SPACE_REMOVE_DUMMY_THREAD_CONTEXTS UVM_TEST_IOCTL_BASE(80) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_VA_SPACE_REMOVE_DUMMY_THREAD_CONTEXTS_PARAMS; #define UVM_TEST_THREAD_CONTEXT_SANITY UVM_TEST_IOCTL_BASE(81) typedef struct { // Iterations to run. NvU32 iterations; // In NV_STATUS rmStatus; // Out } UVM_TEST_THREAD_CONTEXT_SANITY_PARAMS; #define UVM_TEST_THREAD_CONTEXT_PERF UVM_TEST_IOCTL_BASE(82) typedef struct { // Iterations to run. NvU32 iterations; // In // Delay, in microseconds, between thread context addition and removal NvU32 delay_us; // In // Median time, in nanoseconds, spent in adding and then deleting a thread // context. NvU64 ns NV_ALIGN_BYTES(8); // Out NV_STATUS rmStatus; // Out } UVM_TEST_THREAD_CONTEXT_PERF_PARAMS; typedef enum { UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE_NONE = 0, // Pageable memory cannot be accessed, but there is an association between // this VA space and its owning process. For example, this enables the GPU // fault handler to establish CPU mappings. UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE_MMU_NOTIFIER, UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE_HMM, UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE_ATS_KERNEL, UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE_ATS_DRIVER, UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE_COUNT } UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE; #define UVM_TEST_GET_PAGEABLE_MEM_ACCESS_TYPE UVM_TEST_IOCTL_BASE(83) typedef struct { // UVM_TEST_PAGEABLE_MEM_ACCESS_TYPE NvU32 type; // Out NV_STATUS rmStatus; // Out } UVM_TEST_GET_PAGEABLE_MEM_ACCESS_TYPE_PARAMS; // Some events, like fault replays, may not immediately show up in the events // queue despite calling UVM_TOOLS_FLUSH_EVENTS since that will only flush // completed events but not pending events. Successful completion of this IOCTL // guarantees that any replays issued on the given GPU prior to the call will // have its event enqueued in all the tools sessions which have replay events // enabled. Also, this IOCTL includes an implicit UVM_TOOLS_FLUSH_EVENTS call. // Hence, this IOCTL is a superset of UVM_TOOLS_FLUSH_EVENTS. Since this call is // more expensive than UVM_TOOLS_FLUSH_EVENTS, callers who don't need the above // mentioned guarantee should consider calling UVM_TOOLS_FLUSH_EVENTS instead. #define UVM_TEST_TOOLS_FLUSH_REPLAY_EVENTS UVM_TEST_IOCTL_BASE(84) typedef struct { NvProcessorUuid gpuUuid; // In NV_STATUS rmStatus; // Out } UVM_TEST_TOOLS_FLUSH_REPLAY_EVENTS_PARAMS; // Many checks are performed when the driver is unloaded. In the event of an // error, a warning message may be printed to the kernel log. In automated // testing, a systematic way to check the state of the driver after it is // unloaded is required for additional test coverage. One userland process may // register to receive the driver state after its unload, since we cannot use // /proc or /sys to retrieve driver-specific information for an unloaded driver. // Any userland process registers the given address (unload_state_buf) with the // UVM driver. On module unload, if an address has been registered, debugging // state is written to that address. The data in the address is valid once // module unload completes. // Error returns: // NV_ERR_IN_USE // - The unload state buffer has already been registered. // NV_ERR_INVALID_ADDRESS // - unload_state_buf is invalid. // - unload_state_buf is not 8-byte aligned. #define UVM_TEST_REGISTER_UNLOAD_STATE_BUFFER UVM_TEST_IOCTL_BASE(85) // Unload debugging states: #define UVM_TEST_UNLOAD_STATE_MEMORY_LEAK ((NvU64)0x1) typedef struct { // unload_state_buf points to a 8-byte buf and must be aligned to 8 bytes. NvU64 unload_state_buf; // In NV_STATUS rmStatus; // Out } UVM_TEST_REGISTER_UNLOAD_STATE_BUFFER_PARAMS; #define UVM_TEST_RB_TREE_DIRECTED UVM_TEST_IOCTL_BASE(86) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_RB_TREE_DIRECTED_PARAMS; #define UVM_TEST_RB_TREE_RANDOM UVM_TEST_IOCTL_BASE(87) typedef struct { NvU64 iterations NV_ALIGN_BYTES(8); // In // Upper key range bound. Randomly generated node keys will not exceed this // value. NvU64 range_max; // In // This parameter is used to control the size of the tree. // The number of nodes in the tree will bounce between 0 and this limit. // See uvm_rb_tree_test.c:rbtt_test_get_random_op() for full description. NvU32 node_limit; // In NvU32 seed; // In NV_STATUS rmStatus; // Out } UVM_TEST_RB_TREE_RANDOM_PARAMS; #define UVM_TEST_HOST_SANITY UVM_TEST_IOCTL_BASE(88) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_HOST_SANITY_PARAMS; // Calls uvm_va_space_mm_or_current_retain() on a VA space, // then releases the va_space_mm and returns. #define UVM_TEST_VA_SPACE_MM_OR_CURRENT_RETAIN UVM_TEST_IOCTL_BASE(89) typedef struct { // User address of a flag to act as a semaphore. If non-NULL, the address // is set to 1 after successful retain but before the sleep. NvU64 retain_done_ptr NV_ALIGN_BYTES(8); // In // Approximate duration for which to sleep with the va_space_mm retained. NvU64 sleep_us NV_ALIGN_BYTES(8); // In // NV_ERR_PAGE_TABLE_NOT_AVAIL Could not retain va_space_mm // (uvm_va_space_mm_or_current_retain returned // NULL) NV_STATUS rmStatus; // Out } UVM_TEST_VA_SPACE_MM_OR_CURRENT_RETAIN_PARAMS; #define UVM_TEST_GET_USER_SPACE_END_ADDRESS UVM_TEST_IOCTL_BASE(90) typedef struct { NvU64 user_space_end_address; // Out NV_STATUS rmStatus; // Out } UVM_TEST_GET_USER_SPACE_END_ADDRESS_PARAMS; #define UVM_TEST_GET_CPU_CHUNK_ALLOC_SIZES UVM_TEST_IOCTL_BASE(91) typedef struct { NvU32 alloc_size_mask; // Out NvU32 rmStatus; // Out } UVM_TEST_GET_CPU_CHUNK_ALLOC_SIZES_PARAMS; // Forces the next range covering the lookup_address to fail in // uvm_va_range_add_gpu_va_space() with an out-of-memory error. Only the next // uvm_va_range_add_gpu_va_space() will fail. Subsequent ones will succeed. // // Error returns: // NV_ERR_INVALID_ADDRESS // - lookup_address doesn't match a UVM range #define UVM_TEST_VA_RANGE_INJECT_ADD_GPU_VA_SPACE_ERROR UVM_TEST_IOCTL_BASE(93) typedef struct { NvU64 lookup_address NV_ALIGN_BYTES(8); // In NV_STATUS rmStatus; // Out } UVM_TEST_VA_RANGE_INJECT_ADD_GPU_VA_SPACE_ERROR_PARAMS; // Forces destroy_gpu_va_space() to delay execution. This provides a high // probability of exercising the race condition between concurrent // UvmRegisterGpuVaSpace() calls on the same {va_space, gpu} pair in the // ATS_KERNEL case. #define UVM_TEST_DESTROY_GPU_VA_SPACE_DELAY UVM_TEST_IOCTL_BASE(94) typedef struct { NvU64 delay_us; // In NV_STATUS rmStatus; // Out } UVM_TEST_DESTROY_GPU_VA_SPACE_DELAY_PARAMS; #define UVM_TEST_SEC2_SANITY UVM_TEST_IOCTL_BASE(95) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_SEC2_SANITY_PARAMS; #define UVM_TEST_CGROUP_ACCOUNTING_SUPPORTED UVM_TEST_IOCTL_BASE(96) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_CGROUP_ACCOUNTING_SUPPORTED_PARAMS; #define UVM_TEST_SPLIT_INVALIDATE_DELAY UVM_TEST_IOCTL_BASE(98) typedef struct { NvU64 delay_us; // In NV_STATUS rmStatus; // Out } UVM_TEST_SPLIT_INVALIDATE_DELAY_PARAMS; // Tests the CSL/SEC2 encryption/decryption methods by doing a secure transfer // of memory from CPU->GPU and a subsequent GPU->CPU transfer. #define UVM_TEST_SEC2_CPU_GPU_ROUNDTRIP UVM_TEST_IOCTL_BASE(99) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_SEC2_CPU_GPU_ROUNDTRIP_PARAMS; #define UVM_TEST_CPU_CHUNK_API UVM_TEST_IOCTL_BASE(100) typedef struct { NV_STATUS rmStatus; // Out } UVM_TEST_CPU_CHUNK_API_PARAMS; #ifdef __cplusplus } #endif #endif // __UVM_TEST_IOCTL_H__