NVIDIA/open-gpu-kernel-modules
1
0
Fork 0
mirror of https://github.com/NVIDIA/open-gpu-kernel-modules.git synced 2026-03-12 16:49:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

555.42.02

Browse Source 
(cherry picked from commit 5a1c474040)
This commit is contained in:
Bernhard Stoeckner
2024-05-21 15:11:46 +02:00
committed by Gaurav Juvekar
parent caa2dd11a0
commit 3084c04453
1004 changed files with 172522 additions and 150960 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
kernel-open/nvidia-uvm/uvm_va_block.h
View File

@@ -1,5 +1,5 @@
/*******************************************************************************
Copyright (c) 2015-2023 NVIDIA Corporation
Copyright (c) 2015-2024 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
@@ -111,8 +111,6 @@ typedef struct
// Pages that have been evicted to sysmem
uvm_page_mask_t evicted;
NvU64 *cpu_chunks_dma_addrs;
// Array of naturally-aligned chunks. Each chunk has the largest possible
// size which can fit within the block, so they are not uniform size.
//
@@ -2155,8 +2153,7 @@ NV_STATUS uvm_va_block_map_cpu_chunk_on_gpus(uvm_va_block_t *va_block,
// Physically unmap a CPU chunk from all registered GPUs.
// Locking: The va_block lock must be held.
void uvm_va_block_unmap_cpu_chunk_on_gpus(uvm_va_block_t *va_block,
uvm_cpu_chunk_t *chunk,
uvm_page_index_t page_index);
uvm_cpu_chunk_t *chunk);
// Remove any CPU chunks in the given region.
// Locking: The va_block lock must be held.
@@ -2166,19 +2163,19 @@ void uvm_va_block_remove_cpu_chunks(uvm_va_block_t *va_block, uvm_va_block_regio
// specified processor in the block. Returns 0 if the address is not resident on
// the specified processor.
// Locking: The va_block lock must be held.
NvU32 uvm_va_block_get_physical_size(uvm_va_block_t *block,
NvU64 uvm_va_block_get_physical_size(uvm_va_block_t *block,
uvm_processor_id_t processor,
uvm_page_index_t page_index);
// Get CPU page size or 0 if it is not mapped
NvU32 uvm_va_block_page_size_cpu(uvm_va_block_t *va_block,
NvU64 uvm_va_block_page_size_cpu(uvm_va_block_t *va_block,
uvm_page_index_t page_index);
// Get GPU page size or 0 if it is not mapped on the given GPU
NvU32 uvm_va_block_page_size_gpu(uvm_va_block_t *va_block, uvm_gpu_id_t gpu_id, uvm_page_index_t page_index);
NvU64 uvm_va_block_page_size_gpu(uvm_va_block_t *va_block, uvm_gpu_id_t gpu_id, uvm_page_index_t page_index);
// Get page size or 0 if it is not mapped on the given processor
static NvU32 uvm_va_block_page_size_processor(uvm_va_block_t *va_block,
static NvU64 uvm_va_block_page_size_processor(uvm_va_block_t *va_block,
uvm_processor_id_t processor_id,
uvm_page_index_t page_index)
{
@@ -2189,10 +2186,10 @@ static NvU32 uvm_va_block_page_size_processor(uvm_va_block_t *va_block,
}
// Returns the big page size for the GPU VA space of the block
NvU32 uvm_va_block_gpu_big_page_size(uvm_va_block_t *va_block, uvm_gpu_t *gpu);
NvU64 uvm_va_block_gpu_big_page_size(uvm_va_block_t *va_block, uvm_gpu_t *gpu);
// Returns the number of big pages in the VA block for the given size
size_t uvm_va_block_num_big_pages(uvm_va_block_t *va_block, NvU32 big_page_size);
size_t uvm_va_block_num_big_pages(uvm_va_block_t *va_block, NvU64 big_page_size);
// Returns the number of big pages in the VA block for the big page size on the
// given GPU
@@ -2202,29 +2199,29 @@ static size_t uvm_va_block_gpu_num_big_pages(uvm_va_block_t *va_block, uvm_gpu_t
}
// Returns the start address of the given big page index and big page size
NvU64 uvm_va_block_big_page_addr(uvm_va_block_t *va_block, size_t big_page_index, NvU32 big_page_size);
NvU64 uvm_va_block_big_page_addr(uvm_va_block_t *va_block, size_t big_page_index, NvU64 big_page_size);
// Returns the region [start, end] of the given big page index and big page size
uvm_va_block_region_t uvm_va_block_big_page_region(uvm_va_block_t *va_block,
size_t big_page_index,
NvU32 big_page_size);
NvU64 big_page_size);
// Returns the largest sub-region region of [start, end] which can fit big
// pages. If the region cannot fit any big pages, an invalid region (0, 0) is
// returned.
uvm_va_block_region_t uvm_va_block_big_page_region_all(uvm_va_block_t *va_block, NvU32 big_page_size);
uvm_va_block_region_t uvm_va_block_big_page_region_all(uvm_va_block_t *va_block, NvU64 big_page_size);
// Returns the largest sub-region region of 'region' which can fit big pages.
// If the region cannot fit any big pages, an invalid region (0, 0) is returned.
uvm_va_block_region_t uvm_va_block_big_page_region_subset(uvm_va_block_t *va_block,
uvm_va_block_region_t region,
NvU32 big_page_size);
NvU64 big_page_size);
// Returns the big page index (the bit index within
// uvm_va_block_gpu_state_t::big_ptes) corresponding to page_index. If
// page_index cannot be covered by a big PTE due to alignment or block size,
// MAX_BIG_PAGES_PER_UVM_VA_BLOCK is returned.
size_t uvm_va_block_big_page_index(uvm_va_block_t *va_block, uvm_page_index_t page_index, NvU32 big_page_size);
size_t uvm_va_block_big_page_index(uvm_va_block_t *va_block, uvm_page_index_t page_index, NvU64 big_page_size);
// Returns the new residency for a page that faulted or triggered access counter
// notifications. The read_duplicate output parameter indicates if the page