/******************************************************************************* Copyright (c) 2016-2019 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. *******************************************************************************/ #include "uvm_test.h" #include "uvm_va_space.h" #include "uvm_mem.h" #include "uvm_push.h" #include "uvm_hal.h" #define MEM_ALLOCATION_SIZE (4 * 1024 * 1024) static NV_STATUS try_peer_access_remote_gpu_memory(uvm_gpu_t *local_gpu, uvm_gpu_t *peer_gpu) { NV_STATUS status = NV_OK; uvm_mem_t *vidmem = NULL; uvm_mem_t *sysmem = NULL; uvm_push_t push; uvm_gpu_address_t local_gpu_sysmem = {0}; uvm_gpu_address_t peer_gpu_sysmem = {0}; uvm_gpu_address_t peer_gpu_vidmem = {0}; void *cpu_va = NULL; volatile NvU32 *cpu_array; NvU32 i; // allocate CPU memory status = uvm_mem_alloc_sysmem_and_map_cpu_kernel(MEM_ALLOCATION_SIZE, current->mm, &sysmem); TEST_CHECK_GOTO(status == NV_OK, cleanup); // get CPU address cpu_va = uvm_mem_get_cpu_addr_kernel(sysmem); TEST_CHECK_GOTO(cpu_va != 0, cleanup); cpu_array = (volatile NvU32 *)cpu_va; // map sysmem to both GPUs status = uvm_mem_map_gpu_kernel(sysmem, local_gpu); TEST_CHECK_GOTO(status == NV_OK, cleanup); status = uvm_mem_map_gpu_kernel(sysmem, peer_gpu); TEST_CHECK_GOTO(status == NV_OK, cleanup); // get local GPU address for the sysmem local_gpu_sysmem = uvm_mem_gpu_address_virtual_kernel(sysmem, local_gpu); TEST_CHECK_GOTO(local_gpu_sysmem.address != 0, cleanup); peer_gpu_sysmem = uvm_mem_gpu_address_virtual_kernel(sysmem, peer_gpu); TEST_CHECK_GOTO(peer_gpu_sysmem.address != 0, cleanup); // allocate vidmem on remote GPU status = uvm_mem_alloc_vidmem(MEM_ALLOCATION_SIZE, peer_gpu, &vidmem); TEST_CHECK_GOTO(status == NV_OK, cleanup); TEST_CHECK_GOTO(IS_ALIGNED(MEM_ALLOCATION_SIZE, vidmem->chunk_size), cleanup); // map onto GPU status = uvm_mem_map_gpu_kernel(vidmem, peer_gpu); TEST_CHECK_GOTO(status == NV_OK, cleanup); // get remote GPU virtual address for its vidmem peer_gpu_vidmem = uvm_mem_gpu_address_virtual_kernel(vidmem, peer_gpu); TEST_CHECK_GOTO(status == NV_OK, cleanup); // initialize memory using CPU for (i = 0; i < MEM_ALLOCATION_SIZE / sizeof(NvU32); i++) cpu_array[i] = i; // copy sysmem to remote GPUs memory status = uvm_push_begin(peer_gpu->channel_manager, UVM_CHANNEL_TYPE_CPU_TO_GPU, &push, "peer identity mapping test initialization"); TEST_CHECK_GOTO(status == NV_OK, cleanup); peer_gpu->parent->ce_hal->memcopy(&push, peer_gpu_vidmem, peer_gpu_sysmem, MEM_ALLOCATION_SIZE); status = uvm_push_end_and_wait(&push); TEST_CHECK_GOTO(status == NV_OK, cleanup); // set the sysmem back to zero memset((void *)cpu_array, '\0', MEM_ALLOCATION_SIZE); // use the peer mapping to copy back to sysmem status = uvm_push_begin(local_gpu->channel_manager, UVM_CHANNEL_TYPE_GPU_TO_GPU, &push, "peer identity mapping test"); TEST_CHECK_GOTO(status == NV_OK, cleanup); for (i = 0; i < MEM_ALLOCATION_SIZE / vidmem->chunk_size; i++) { uvm_gpu_address_t local_gpu_peer = uvm_mem_gpu_address_copy(vidmem, local_gpu, vidmem->chunk_size * i, vidmem->chunk_size); uvm_gpu_address_t local_gpu_sysmem_offset = local_gpu_sysmem; local_gpu_sysmem_offset.address += vidmem->chunk_size * i; local_gpu->parent->ce_hal->memcopy(&push, local_gpu_sysmem_offset, local_gpu_peer, vidmem->chunk_size); } status = uvm_push_end_and_wait(&push); TEST_CHECK_GOTO(status == NV_OK, cleanup); for (i = 0; i < MEM_ALLOCATION_SIZE / sizeof(NvU32); i++) { if (cpu_array[i] != i) { UVM_TEST_PRINT("Expected %u at offset %u but got %u\n", i, i, cpu_array[i]); status = NV_ERR_INVALID_STATE; } } cleanup: uvm_mem_free(vidmem); uvm_mem_free(sysmem); return status; } NV_STATUS uvm_test_peer_identity_mappings(UVM_TEST_PEER_IDENTITY_MAPPINGS_PARAMS *params, struct file *filp) { NV_STATUS status; uvm_gpu_t *gpu_a; uvm_gpu_t *gpu_b; uvm_va_space_t *va_space = uvm_va_space_get(filp); uvm_va_space_down_read(va_space); gpu_a = uvm_va_space_get_gpu_by_uuid(va_space, ¶ms->gpuA); gpu_b = uvm_va_space_get_gpu_by_uuid(va_space, ¶ms->gpuB); if (gpu_a == NULL || gpu_b == NULL) { status = NV_ERR_INVALID_DEVICE; goto done; } if (gpu_a->parent->peer_copy_mode != gpu_b->parent->peer_copy_mode) { status = NV_ERR_INVALID_DEVICE; goto done; } if (gpu_a->parent->peer_copy_mode != UVM_GPU_PEER_COPY_MODE_VIRTUAL) { status = NV_WARN_NOTHING_TO_DO; goto done; } // Indirect peers don't use identity mappings if (!uvm_processor_mask_test(&va_space->can_access[uvm_id_value(gpu_a->id)], gpu_b->id) || uvm_processor_mask_test(&va_space->indirect_peers[uvm_id_value(gpu_a->id)], gpu_b->id)) { status = NV_ERR_INVALID_DEVICE; goto done; } status = try_peer_access_remote_gpu_memory(gpu_a, gpu_b); if (status != NV_OK) goto done; status = try_peer_access_remote_gpu_memory(gpu_b, gpu_a); if (status != NV_OK) goto done; done: uvm_va_space_up_read(va_space); return status; }