1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
|
// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <atomic>
#include <map>
#include <optional>
#include <vector>
#include "common/common_types.h"
#include "common/multi_level_page_table.h"
#include "common/virtual_buffer.h"
#include "video_core/pte_kind.h"
namespace VideoCore {
class RasterizerInterface;
}
namespace Core {
class DeviceMemory;
namespace Memory {
class Memory;
} // namespace Memory
class System;
} // namespace Core
namespace Tegra {
class MemoryManager final {
public:
explicit MemoryManager(Core::System& system_, u64 address_space_bits_ = 40,
u64 big_page_bits_ = 16, u64 page_bits_ = 12);
~MemoryManager();
size_t GetID() const {
return unique_identifier;
}
/// Binds a renderer to the memory manager.
void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
[[nodiscard]] std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr) const;
[[nodiscard]] std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr, std::size_t size) const;
template <typename T>
[[nodiscard]] T Read(GPUVAddr addr) const;
template <typename T>
void Write(GPUVAddr addr, T data);
[[nodiscard]] u8* GetPointer(GPUVAddr addr);
[[nodiscard]] const u8* GetPointer(GPUVAddr addr) const;
/**
* ReadBlock and WriteBlock are full read and write operations over virtual
* GPU Memory. It's important to use these when GPU memory may not be continuous
* in the Host Memory counterpart. Note: This functions cause Host GPU Memory
* Flushes and Invalidations, respectively to each operation.
*/
void ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const;
void WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size);
void CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size);
/**
* ReadBlockUnsafe and WriteBlockUnsafe are special versions of ReadBlock and
* WriteBlock respectively. In this versions, no flushing or invalidation is actually
* done and their performance is similar to a memcpy. This functions can be used
* on either of this 2 scenarios instead of their safe counterpart:
* - Memory which is sure to never be represented in the Host GPU.
* - Memory Managed by a Cache Manager. Example: Texture Flushing should use
* WriteBlockUnsafe instead of WriteBlock since it shouldn't invalidate the texture
* being flushed.
*/
void ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const;
void WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size);
/**
* Checks if a gpu region can be simply read with a pointer.
*/
[[nodiscard]] bool IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const;
/**
* Checks if a gpu region is mapped by a single range of cpu addresses.
*/
[[nodiscard]] bool IsContinousRange(GPUVAddr gpu_addr, std::size_t size) const;
/**
* Checks if a gpu region is mapped entirely.
*/
[[nodiscard]] bool IsFullyMappedRange(GPUVAddr gpu_addr, std::size_t size) const;
/**
* Returns a vector with all the subranges of cpu addresses mapped beneath.
* if the region is continous, a single pair will be returned. If it's unmapped, an empty vector
* will be returned;
*/
std::vector<std::pair<GPUVAddr, std::size_t>> GetSubmappedRange(GPUVAddr gpu_addr,
std::size_t size) const;
GPUVAddr Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size,
PTEKind kind = PTEKind::INVALID, bool is_big_pages = true);
GPUVAddr MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages = true);
void Unmap(GPUVAddr gpu_addr, std::size_t size);
void FlushRegion(GPUVAddr gpu_addr, size_t size) const;
void InvalidateRegion(GPUVAddr gpu_addr, size_t size) const;
bool IsMemoryDirty(GPUVAddr gpu_addr, size_t size) const;
size_t MaxContinousRange(GPUVAddr gpu_addr, size_t size) const;
bool IsWithinGPUAddressRange(GPUVAddr gpu_addr) const {
return gpu_addr < address_space_size;
}
PTEKind GetPageKind(GPUVAddr gpu_addr) const;
private:
template <bool is_big_pages, typename FuncMapped, typename FuncReserved, typename FuncUnmapped>
inline void MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, FuncMapped&& func_mapped,
FuncReserved&& func_reserved, FuncUnmapped&& func_unmapped) const;
template <bool is_safe>
void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const;
template <bool is_safe>
void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size);
template <bool is_big_page>
[[nodiscard]] std::size_t PageEntryIndex(GPUVAddr gpu_addr) const {
if constexpr (is_big_page) {
return (gpu_addr >> big_page_bits) & big_page_table_mask;
} else {
return (gpu_addr >> page_bits) & page_table_mask;
}
}
inline bool IsBigPageContinous(size_t big_page_index) const;
inline void SetBigPageContinous(size_t big_page_index, bool value);
Core::System& system;
Core::Memory::Memory& memory;
Core::DeviceMemory& device_memory;
const u64 address_space_bits;
const u64 page_bits;
u64 address_space_size;
u64 page_size;
u64 page_mask;
u64 page_table_mask;
static constexpr u64 cpu_page_bits{12};
const u64 big_page_bits;
u64 big_page_size;
u64 big_page_mask;
u64 big_page_table_mask;
VideoCore::RasterizerInterface* rasterizer = nullptr;
enum class EntryType : u64 {
Free = 0,
Reserved = 1,
Mapped = 2,
};
std::vector<u64> entries;
std::vector<u64> big_entries;
template <EntryType entry_type>
GPUVAddr PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size,
PTEKind kind);
template <EntryType entry_type>
GPUVAddr BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size,
PTEKind kind);
template <bool is_big_page>
inline EntryType GetEntry(size_t position) const;
template <bool is_big_page>
inline void SetEntry(size_t position, EntryType entry);
std::vector<std::array<PTEKind, 32>> kinds;
std::vector<std::array<PTEKind, 32>> big_kinds;
template <bool is_big_page>
inline PTEKind GetKind(size_t position) const;
template <bool is_big_page>
inline void SetKind(size_t position, PTEKind kind);
Common::MultiLevelPageTable<u32> page_table;
Common::VirtualBuffer<u32> big_page_table_cpu;
std::vector<u64> big_page_continous;
constexpr static size_t continous_bits = 64;
const size_t unique_identifier;
static std::atomic<size_t> unique_identifier_generator;
};
} // namespace Tegra
|
