/******************************************************************************* 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. *******************************************************************************/ #ifndef __UVM_PERF_THRASHING_H__ #define __UVM_PERF_THRASHING_H__ #include "uvm_linux.h" #include "uvm_extern_decl.h" #include "uvm_forward_decl.h" #include "uvm_processors.h" #include "uvm_va_block_types.h" typedef enum { // No thrashing detected UVM_PERF_THRASHING_HINT_TYPE_NONE = 0, // Map remotely to avoid future faults (does not help with revocations due // to system-wide atomics) UVM_PERF_THRASHING_HINT_TYPE_PIN = 1, // Throttle execution of the calling processor (this can be implemented by // sleeping or handing other faults) UVM_PERF_THRASHING_HINT_TYPE_THROTTLE = 2, // TODO: Bug 1877578: Implement heuristics-driven read-duplication // Add a thrashing hint type to read-duplicate a page when it is being // accessed read-only from different processors } uvm_perf_thrashing_hint_type_t; typedef struct { uvm_perf_thrashing_hint_type_t type; union { struct { // Map to this processor, which must be accessible, at least, from // the calling processor uvm_processor_id_t residency; // Processors to be mapped, when possible, to the new residency uvm_processor_mask_t processors; } pin; struct { // Absolute timestamp in ns after which the throttled processor is // allowed to start servicing faults on the thrashing page. NvU64 end_time_stamp; } throttle; }; } uvm_perf_thrashing_hint_t; // Obtain a hint to prevent thrashing on the page with given address uvm_perf_thrashing_hint_t uvm_perf_thrashing_get_hint(uvm_va_block_t *va_block, NvU64 address, uvm_processor_id_t requester); // Obtain a pointer to a mask with the processors that are thrashing on the // given page. This function assumes that thrashing has been just reported on // the page. It will fail otherwise. uvm_processor_mask_t *uvm_perf_thrashing_get_thrashing_processors(uvm_va_block_t *va_block, NvU64 address); const uvm_page_mask_t *uvm_perf_thrashing_get_thrashing_pages(uvm_va_block_t *va_block); // Returns true if any page in the block is thrashing, or false otherwise bool uvm_perf_thrashing_is_block_thrashing(uvm_va_block_t *va_block); // Global initialization/cleanup functions NV_STATUS uvm_perf_thrashing_init(void); void uvm_perf_thrashing_exit(void); // Per-GPU initialization/cleanup functions. See comments in // uvm_perf_heuristics.h NV_STATUS uvm_perf_thrashing_add_gpu(uvm_gpu_t *gpu); void uvm_perf_thrashing_remove_gpu(uvm_gpu_t *gpu); // VA space Initialization/cleanup functions. See comments in // uvm_perf_heuristics.h NV_STATUS uvm_perf_thrashing_load(uvm_va_space_t *va_space); NV_STATUS uvm_perf_thrashing_register_gpu(uvm_va_space_t *va_space, uvm_gpu_t *gpu); void uvm_perf_thrashing_stop(uvm_va_space_t *va_space); void uvm_perf_thrashing_unload(uvm_va_space_t *va_space); #endif