summaryrefslogtreecommitdiffstats
path: root/src/video_core/memory_manager.cpp
blob: 11b1d022ad2d599b1912d15a6ec423463fa131c1 (plain) (blame)
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
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
// Copyright 2018 yuzu emulator team
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.

#include "common/alignment.h"
#include "common/assert.h"
#include "common/logging/log.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_base.h"

namespace Tegra {

MemoryManager::MemoryManager() {
    std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
    std::fill(page_table.attributes.begin(), page_table.attributes.end(),
              Common::PageType::Unmapped);
    page_table.Resize(address_space_width);

    // Initialize the map with a single free region covering the entire managed space.
    VirtualMemoryArea initial_vma;
    initial_vma.size = address_space_end;
    vma_map.emplace(initial_vma.base, initial_vma);

    UpdatePageTableForVMA(initial_vma);
}

GPUVAddr MemoryManager::AllocateSpace(u64 size, u64 align) {
    const u64 aligned_size{Common::AlignUp(size, page_size)};
    const GPUVAddr gpu_addr{FindFreeRegion(address_space_base, aligned_size)};

    AllocateMemory(gpu_addr, 0, aligned_size);

    return gpu_addr;
}

GPUVAddr MemoryManager::AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align) {
    const u64 aligned_size{Common::AlignUp(size, page_size)};

    AllocateMemory(gpu_addr, 0, aligned_size);

    return gpu_addr;
}

GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) {
    const u64 aligned_size{Common::AlignUp(size, page_size)};
    const GPUVAddr gpu_addr{FindFreeRegion(address_space_base, aligned_size)};

    MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);

    return gpu_addr;
}

GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size) {
    ASSERT((gpu_addr & page_mask) == 0);

    const u64 aligned_size{Common::AlignUp(size, page_size)};

    MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);

    return gpu_addr;
}

GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) {
    ASSERT((gpu_addr & page_mask) == 0);

    const u64 aligned_size{Common::AlignUp(size, page_size)};
    const CacheAddr cache_addr{ToCacheAddr(GetPointer(gpu_addr))};

    Core::System::GetInstance().Renderer().Rasterizer().FlushAndInvalidateRegion(cache_addr,
                                                                                 aligned_size);
    UnmapRange(gpu_addr, aligned_size);

    return gpu_addr;
}

GPUVAddr MemoryManager::FindFreeRegion(GPUVAddr region_start, u64 size) const {
    // Find the first Free VMA.
    const VMAHandle vma_handle{
        std::find_if(vma_map.begin(), vma_map.end(), [region_start, size](const auto& vma) {
            if (vma.second.type != VirtualMemoryArea::Type::Unmapped) {
                return false;
            }

            const VAddr vma_end{vma.second.base + vma.second.size};
            return vma_end > region_start && vma_end >= region_start + size;
        })};

    if (vma_handle == vma_map.end()) {
        return {};
    }

    return std::max(region_start, vma_handle->second.base);
}

bool MemoryManager::IsAddressValid(GPUVAddr addr) const {
    return (addr >> page_bits) < page_table.pointers.size();
}

std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr) const {
    if (!IsAddressValid(addr)) {
        return {};
    }

    const VAddr cpu_addr{page_table.backing_addr[addr >> page_bits]};
    if (cpu_addr) {
        return cpu_addr + (addr & page_mask);
    }

    return {};
}

template <typename T>
T MemoryManager::Read(GPUVAddr addr) {
    if (!IsAddressValid(addr)) {
        return {};
    }

    const u8* page_pointer{page_table.pointers[addr >> page_bits]};
    if (page_pointer) {
        // NOTE: Avoid adding any extra logic to this fast-path block
        T value;
        std::memcpy(&value, &page_pointer[addr & page_mask], sizeof(T));
        return value;
    }

    switch (page_table.attributes[addr >> page_bits]) {
    case Common::PageType::Unmapped:
        LOG_ERROR(HW_GPU, "Unmapped Read{} @ 0x{:08X}", sizeof(T) * 8, addr);
        return 0;
    case Common::PageType::Memory:
        ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", addr);
        break;
    default:
        UNREACHABLE();
    }
    return {};
}

template <typename T>
void MemoryManager::Write(GPUVAddr addr, T data) {
    if (!IsAddressValid(addr)) {
        return;
    }

    u8* page_pointer{page_table.pointers[addr >> page_bits]};
    if (page_pointer) {
        // NOTE: Avoid adding any extra logic to this fast-path block
        std::memcpy(&page_pointer[addr & page_mask], &data, sizeof(T));
        return;
    }

    switch (page_table.attributes[addr >> page_bits]) {
    case Common::PageType::Unmapped:
        LOG_ERROR(HW_GPU, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
                  static_cast<u32>(data), addr);
        return;
    case Common::PageType::Memory:
        ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", addr);
        break;
    default:
        UNREACHABLE();
    }
}

template u8 MemoryManager::Read<u8>(GPUVAddr addr);
template u16 MemoryManager::Read<u16>(GPUVAddr addr);
template u32 MemoryManager::Read<u32>(GPUVAddr addr);
template u64 MemoryManager::Read<u64>(GPUVAddr addr);
template void MemoryManager::Write<u8>(GPUVAddr addr, u8 data);
template void MemoryManager::Write<u16>(GPUVAddr addr, u16 data);
template void MemoryManager::Write<u32>(GPUVAddr addr, u32 data);
template void MemoryManager::Write<u64>(GPUVAddr addr, u64 data);

u8* MemoryManager::GetPointer(GPUVAddr addr) {
    if (!IsAddressValid(addr)) {
        return {};
    }

    u8* page_pointer{page_table.pointers[addr >> page_bits]};
    if (page_pointer) {
        return page_pointer + (addr & page_mask);
    }

    LOG_ERROR(HW_GPU, "Unknown GetPointer @ 0x{:016X}", addr);
    return {};
}

void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) {
    std::memcpy(dest_buffer, GetPointer(src_addr), size);
}
void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size) {
    std::memcpy(GetPointer(dest_addr), src_buffer, size);
}

void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size) {
    std::memcpy(GetPointer(dest_addr), GetPointer(src_addr), size);
}

void MemoryManager::MapPages(GPUVAddr base, u64 size, u8* memory, Common::PageType type,
                             VAddr backing_addr) {
    LOG_DEBUG(HW_GPU, "Mapping {} onto {:016X}-{:016X}", fmt::ptr(memory), base * page_size,
              (base + size) * page_size);

    const VAddr end{base + size};
    ASSERT_MSG(end <= page_table.pointers.size(), "out of range mapping at {:016X}",
               base + page_table.pointers.size());

    std::fill(page_table.attributes.begin() + base, page_table.attributes.begin() + end, type);

    if (memory == nullptr) {
        std::fill(page_table.pointers.begin() + base, page_table.pointers.begin() + end, memory);
        std::fill(page_table.backing_addr.begin() + base, page_table.backing_addr.begin() + end,
                  backing_addr);
    } else {
        while (base != end) {
            page_table.pointers[base] = memory;
            page_table.backing_addr[base] = backing_addr;

            base += 1;
            memory += page_size;
            backing_addr += page_size;
        }
    }
}

void MemoryManager::MapMemoryRegion(GPUVAddr base, u64 size, u8* target, VAddr backing_addr) {
    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: {:016X}", size);
    ASSERT_MSG((base & page_mask) == 0, "non-page aligned base: {:016X}", base);
    MapPages(base / page_size, size / page_size, target, Common::PageType::Memory, backing_addr);
}

void MemoryManager::UnmapRegion(GPUVAddr base, u64 size) {
    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: {:016X}", size);
    ASSERT_MSG((base & page_mask) == 0, "non-page aligned base: {:016X}", base);
    MapPages(base / page_size, size / page_size, nullptr, Common::PageType::Unmapped);
}

bool VirtualMemoryArea::CanBeMergedWith(const VirtualMemoryArea& next) const {
    ASSERT(base + size == next.base);
    if (type != next.type) {
        return {};
    }
    if (type == VirtualMemoryArea::Type::Allocated && (offset + size != next.offset)) {
        return {};
    }
    if (type == VirtualMemoryArea::Type::Mapped && backing_memory + size != next.backing_memory) {
        return {};
    }
    return true;
}

MemoryManager::VMAHandle MemoryManager::FindVMA(GPUVAddr target) const {
    if (target >= address_space_end) {
        return vma_map.end();
    } else {
        return std::prev(vma_map.upper_bound(target));
    }
}

MemoryManager::VMAIter MemoryManager::Allocate(VMAIter vma_handle) {
    VirtualMemoryArea& vma{vma_handle->second};

    vma.type = VirtualMemoryArea::Type::Allocated;
    vma.backing_addr = 0;
    vma.backing_memory = {};
    UpdatePageTableForVMA(vma);

    return MergeAdjacent(vma_handle);
}

MemoryManager::VMAHandle MemoryManager::AllocateMemory(GPUVAddr target, std::size_t offset,
                                                       u64 size) {

    // This is the appropriately sized VMA that will turn into our allocation.
    VMAIter vma_handle{CarveVMA(target, size)};
    VirtualMemoryArea& vma{vma_handle->second};

    ASSERT(vma.size == size);

    vma.offset = offset;

    return Allocate(vma_handle);
}

MemoryManager::VMAHandle MemoryManager::MapBackingMemory(GPUVAddr target, u8* memory, u64 size,
                                                         VAddr backing_addr) {
    // This is the appropriately sized VMA that will turn into our allocation.
    VMAIter vma_handle{CarveVMA(target, size)};
    VirtualMemoryArea& vma{vma_handle->second};

    ASSERT(vma.size == size);

    vma.type = VirtualMemoryArea::Type::Mapped;
    vma.backing_memory = memory;
    vma.backing_addr = backing_addr;
    UpdatePageTableForVMA(vma);

    return MergeAdjacent(vma_handle);
}

void MemoryManager::UnmapRange(GPUVAddr target, u64 size) {
    VMAIter vma{CarveVMARange(target, size)};
    const VAddr target_end{target + size};
    const VMAIter end{vma_map.end()};

    // The comparison against the end of the range must be done using addresses since VMAs can be
    // merged during this process, causing invalidation of the iterators.
    while (vma != end && vma->second.base < target_end) {
        // Unmapped ranges return to allocated state and can be reused
        // This behavior is used by Super Mario Odyssey, Sonic Forces, and likely other games
        vma = std::next(Allocate(vma));
    }

    ASSERT(FindVMA(target)->second.size >= size);
}

MemoryManager::VMAIter MemoryManager::StripIterConstness(const VMAHandle& iter) {
    // This uses a neat C++ trick to convert a const_iterator to a regular iterator, given
    // non-const access to its container.
    return vma_map.erase(iter, iter); // Erases an empty range of elements
}

MemoryManager::VMAIter MemoryManager::CarveVMA(GPUVAddr base, u64 size) {
    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: 0x{:016X}", size);
    ASSERT_MSG((base & page_mask) == 0, "non-page aligned base: 0x{:016X}", base);

    VMAIter vma_handle{StripIterConstness(FindVMA(base))};
    if (vma_handle == vma_map.end()) {
        // Target address is outside the managed range
        return {};
    }

    const VirtualMemoryArea& vma{vma_handle->second};
    if (vma.type == VirtualMemoryArea::Type::Mapped) {
        // Region is already allocated
        return {};
    }

    const VAddr start_in_vma{base - vma.base};
    const VAddr end_in_vma{start_in_vma + size};

    ASSERT_MSG(end_in_vma <= vma.size, "region size 0x{:016X} is less than required size 0x{:016X}",
               vma.size, end_in_vma);

    if (end_in_vma < vma.size) {
        // Split VMA at the end of the allocated region
        SplitVMA(vma_handle, end_in_vma);
    }
    if (start_in_vma != 0) {
        // Split VMA at the start of the allocated region
        vma_handle = SplitVMA(vma_handle, start_in_vma);
    }

    return vma_handle;
}

MemoryManager::VMAIter MemoryManager::CarveVMARange(GPUVAddr target, u64 size) {
    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: 0x{:016X}", size);
    ASSERT_MSG((target & page_mask) == 0, "non-page aligned base: 0x{:016X}", target);

    const VAddr target_end{target + size};
    ASSERT(target_end >= target);
    ASSERT(size > 0);

    VMAIter begin_vma{StripIterConstness(FindVMA(target))};
    const VMAIter i_end{vma_map.lower_bound(target_end)};
    if (std::any_of(begin_vma, i_end, [](const auto& entry) {
            return entry.second.type == VirtualMemoryArea::Type::Unmapped;
        })) {
        return {};
    }

    if (target != begin_vma->second.base) {
        begin_vma = SplitVMA(begin_vma, target - begin_vma->second.base);
    }

    VMAIter end_vma{StripIterConstness(FindVMA(target_end))};
    if (end_vma != vma_map.end() && target_end != end_vma->second.base) {
        end_vma = SplitVMA(end_vma, target_end - end_vma->second.base);
    }

    return begin_vma;
}

MemoryManager::VMAIter MemoryManager::SplitVMA(VMAIter vma_handle, u64 offset_in_vma) {
    VirtualMemoryArea& old_vma{vma_handle->second};
    VirtualMemoryArea new_vma{old_vma}; // Make a copy of the VMA

    // For now, don't allow no-op VMA splits (trying to split at a boundary) because it's probably
    // a bug. This restriction might be removed later.
    ASSERT(offset_in_vma < old_vma.size);
    ASSERT(offset_in_vma > 0);

    old_vma.size = offset_in_vma;
    new_vma.base += offset_in_vma;
    new_vma.size -= offset_in_vma;

    switch (new_vma.type) {
    case VirtualMemoryArea::Type::Unmapped:
        break;
    case VirtualMemoryArea::Type::Allocated:
        new_vma.offset += offset_in_vma;
        break;
    case VirtualMemoryArea::Type::Mapped:
        new_vma.backing_memory += offset_in_vma;
        break;
    }

    ASSERT(old_vma.CanBeMergedWith(new_vma));

    return vma_map.emplace_hint(std::next(vma_handle), new_vma.base, new_vma);
}

MemoryManager::VMAIter MemoryManager::MergeAdjacent(VMAIter iter) {
    const VMAIter next_vma{std::next(iter)};
    if (next_vma != vma_map.end() && iter->second.CanBeMergedWith(next_vma->second)) {
        iter->second.size += next_vma->second.size;
        vma_map.erase(next_vma);
    }

    if (iter != vma_map.begin()) {
        VMAIter prev_vma{std::prev(iter)};
        if (prev_vma->second.CanBeMergedWith(iter->second)) {
            prev_vma->second.size += iter->second.size;
            vma_map.erase(iter);
            iter = prev_vma;
        }
    }

    return iter;
}

void MemoryManager::UpdatePageTableForVMA(const VirtualMemoryArea& vma) {
    switch (vma.type) {
    case VirtualMemoryArea::Type::Unmapped:
        UnmapRegion(vma.base, vma.size);
        break;
    case VirtualMemoryArea::Type::Allocated:
        MapMemoryRegion(vma.base, vma.size, nullptr, vma.backing_addr);
        break;
    case VirtualMemoryArea::Type::Mapped:
        MapMemoryRegion(vma.base, vma.size, vma.backing_memory, vma.backing_addr);
        break;
    }
}

} // namespace Tegra