NVIDIA/open-gpu-kernel-modules
1
0
Fork 0
mirror of https://github.com/NVIDIA/open-gpu-kernel-modules.git synced 2026-02-28 19:03:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

555.42.02

Browse Source
This commit is contained in:
Bernhard Stoeckner
2024-05-21 15:11:46 +02:00
parent 083cd9cf17
commit 5a1c474040
955 changed files with 171849 additions and 144768 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
kernel-open/nvidia-uvm/uvm_map_external.c
View File

@@ -1,5 +1,5 @@
/*******************************************************************************
Copyright (c) 2016-2023 NVIDIA Corporation
Copyright (c) 2016-2024 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
@@ -39,6 +39,7 @@
#include "uvm_pte_batch.h"
#include "uvm_tlb_batch.h"
#include "nv_uvm_interface.h"
#include "nv_uvm_types.h"
#include "uvm_pushbuffer.h"
@@ -60,7 +61,7 @@ typedef struct
size_t buffer_size;
// Page size in bytes
NvU32 page_size;
NvU64 page_size;
// Size of a single PTE in bytes
NvU32 pte_size;
@@ -90,7 +91,7 @@ static NV_STATUS uvm_pte_buffer_init(uvm_va_range_t *va_range,
uvm_gpu_t *gpu,
const uvm_map_rm_params_t *map_rm_params,
NvU64 length,
NvU32 page_size,
NvU64 page_size,
uvm_pte_buffer_t *pte_buffer)
{
uvm_gpu_va_space_t *gpu_va_space = uvm_gpu_va_space_get(va_range->va_space, gpu);
@@ -101,11 +102,11 @@ static NV_STATUS uvm_pte_buffer_init(uvm_va_range_t *va_range,
pte_buffer->va_range = va_range;
pte_buffer->gpu = gpu;
pte_buffer->mapping_info.cachingType = map_rm_params->caching_type;
pte_buffer->mapping_info.mappingType = map_rm_params->mapping_type;
pte_buffer->mapping_info.formatType = map_rm_params->format_type;
pte_buffer->mapping_info.elementBits = map_rm_params->element_bits;
pte_buffer->mapping_info.compressionType = map_rm_params->compression_type;
pte_buffer->mapping_info.cachingType = (UvmRmGpuCachingType) map_rm_params->caching_type;
pte_buffer->mapping_info.mappingType = (UvmRmGpuMappingType) map_rm_params->mapping_type;
pte_buffer->mapping_info.formatType = (UvmRmGpuFormatType) map_rm_params->format_type;
pte_buffer->mapping_info.elementBits = (UvmRmGpuFormatElementBits) map_rm_params->element_bits;
pte_buffer->mapping_info.compressionType = (UvmRmGpuCompressionType) map_rm_params->compression_type;
if (va_range->type == UVM_VA_RANGE_TYPE_EXTERNAL)
pte_buffer->mapping_info.mappingPageSize = page_size;
@@ -649,9 +650,7 @@ static NV_STATUS set_ext_gpu_map_location(uvm_ext_gpu_map_t *ext_gpu_map,
return NV_OK;
}
// This is a local or peer allocation, so the owning GPU must have been
// registered.
// This also checks for if EGM owning GPU is registered.
// registered. This also checks for if EGM owning GPU is registered.
owning_gpu = uvm_va_space_get_gpu_by_uuid(va_space, &mem_info->uuid);
if (!owning_gpu)
return NV_ERR_INVALID_DEVICE;
@@ -664,7 +663,6 @@ static NV_STATUS set_ext_gpu_map_location(uvm_ext_gpu_map_t *ext_gpu_map,
// semantics of sysmem allocations.
// Check if peer access for peer memory is enabled.
// This path also handles EGM allocations.
if (owning_gpu != mapping_gpu && (!mem_info->sysmem || mem_info->egm)) {
// TODO: Bug 1757136: In SLI, the returned UUID may be different but a
// local mapping must be used. We need to query SLI groups to know
@@ -855,9 +853,10 @@ static NV_STATUS uvm_map_external_allocation_on_gpu(uvm_va_range_t *va_range,
uvm_ext_gpu_range_tree_t *range_tree = uvm_ext_gpu_range_tree(va_range, mapping_gpu);
UvmGpuMemoryInfo mem_info;
uvm_gpu_va_space_t *gpu_va_space = uvm_gpu_va_space_get(va_space, mapping_gpu);
NvU32 mapping_page_size;
NvU64 mapping_page_size;
NvU64 biggest_mapping_page_size;
NvU64 alignments;
NvU32 smallest_alignment;
NvU64 smallest_alignment;
NV_STATUS status;
uvm_assert_rwsem_locked_read(&va_space->lock);
@@ -946,9 +945,11 @@ static NV_STATUS uvm_map_external_allocation_on_gpu(uvm_va_range_t *va_range,
// Check for the maximum page size for the mapping of vidmem allocations,
// the vMMU segment size may limit the range of page sizes.
biggest_mapping_page_size = uvm_mmu_biggest_page_size_up_to(&gpu_va_space->page_tables,
mapping_gpu->mem_info.max_vidmem_page_size);
if (!ext_gpu_map->is_sysmem && (ext_gpu_map->gpu == ext_gpu_map->owning_gpu) &&
(mapping_page_size > mapping_gpu->mem_info.max_vidmem_page_size))
mapping_page_size = mapping_gpu->mem_info.max_vidmem_page_size;
(mapping_page_size > biggest_mapping_page_size))
mapping_page_size = biggest_mapping_page_size;
mem_info.pageSize = mapping_page_size;
@@ -985,7 +986,7 @@ static NV_STATUS uvm_map_external_allocation(uvm_va_space_t *va_space, UVM_MAP_E
if (uvm_api_range_invalid_4k(params->base, params->length))
return NV_ERR_INVALID_ADDRESS;
if (params->gpuAttributesCount == 0 || params->gpuAttributesCount > UVM_MAX_GPUS_V2)
if (params->gpuAttributesCount == 0 || params->gpuAttributesCount > UVM_MAX_GPUS)
return NV_ERR_INVALID_ARGUMENT;
mapped_gpus = uvm_processor_mask_cache_alloc();