mirror of https://git.suyu.dev/suyu/suyu
Merge pull request #5953 from bunnei/memory-refactor-1
Kernel Rework: Memory updates and refactoring (Part 1)merge-requests/60/head
bunnei
GitHub
commit
09f7c355c6
@ -0,0 +1,250 @@
|
||||
// Copyright 2021 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <array>
|
||||
|
||||
#include "common/alignment.h"
|
||||
#include "common/common_types.h"
|
||||
|
||||
namespace Common {
|
||||
|
||||
// Implementation of TinyMT (mersenne twister RNG).
|
||||
// Like Nintendo, we will use the sample parameters.
|
||||
class TinyMT {
|
||||
public:
|
||||
static constexpr std::size_t NumStateWords = 4;
|
||||
|
||||
struct State {
|
||||
std::array<u32, NumStateWords> data{};
|
||||
};
|
||||
|
||||
private:
|
||||
static constexpr u32 ParamMat1 = 0x8F7011EE;
|
||||
static constexpr u32 ParamMat2 = 0xFC78FF1F;
|
||||
static constexpr u32 ParamTmat = 0x3793FDFF;
|
||||
|
||||
static constexpr u32 ParamMult = 0x6C078965;
|
||||
static constexpr u32 ParamPlus = 0x0019660D;
|
||||
static constexpr u32 ParamXor = 0x5D588B65;
|
||||
|
||||
static constexpr u32 TopBitmask = 0x7FFFFFFF;
|
||||
|
||||
static constexpr int MinimumInitIterations = 8;
|
||||
static constexpr int NumDiscardedInitOutputs = 8;
|
||||
|
||||
static constexpr u32 XorByShifted27(u32 value) {
|
||||
return value ^ (value >> 27);
|
||||
}
|
||||
|
||||
static constexpr u32 XorByShifted30(u32 value) {
|
||||
return value ^ (value >> 30);
|
||||
}
|
||||
|
||||
private:
|
||||
State state{};
|
||||
|
||||
private:
|
||||
// Internal API.
|
||||
void FinalizeInitialization() {
|
||||
const u32 state0 = this->state.data[0] & TopBitmask;
|
||||
const u32 state1 = this->state.data[1];
|
||||
const u32 state2 = this->state.data[2];
|
||||
const u32 state3 = this->state.data[3];
|
||||
|
||||
if (state0 == 0 && state1 == 0 && state2 == 0 && state3 == 0) {
|
||||
this->state.data[0] = 'T';
|
||||
this->state.data[1] = 'I';
|
||||
this->state.data[2] = 'N';
|
||||
this->state.data[3] = 'Y';
|
||||
}
|
||||
|
||||
for (int i = 0; i < NumDiscardedInitOutputs; i++) {
|
||||
this->GenerateRandomU32();
|
||||
}
|
||||
}
|
||||
|
||||
u32 GenerateRandomU24() {
|
||||
return (this->GenerateRandomU32() >> 8);
|
||||
}
|
||||
|
||||
static void GenerateInitialValuePlus(TinyMT::State* state, int index, u32 value) {
|
||||
u32& state0 = state->data[(index + 0) % NumStateWords];
|
||||
u32& state1 = state->data[(index + 1) % NumStateWords];
|
||||
u32& state2 = state->data[(index + 2) % NumStateWords];
|
||||
u32& state3 = state->data[(index + 3) % NumStateWords];
|
||||
|
||||
const u32 x = XorByShifted27(state0 ^ state1 ^ state3) * ParamPlus;
|
||||
const u32 y = x + index + value;
|
||||
|
||||
state0 = y;
|
||||
state1 += x;
|
||||
state2 += y;
|
||||
}
|
||||
|
||||
static void GenerateInitialValueXor(TinyMT::State* state, int index) {
|
||||
u32& state0 = state->data[(index + 0) % NumStateWords];
|
||||
u32& state1 = state->data[(index + 1) % NumStateWords];
|
||||
u32& state2 = state->data[(index + 2) % NumStateWords];
|
||||
u32& state3 = state->data[(index + 3) % NumStateWords];
|
||||
|
||||
const u32 x = XorByShifted27(state0 + state1 + state3) * ParamXor;
|
||||
const u32 y = x - index;
|
||||
|
||||
state0 = y;
|
||||
state1 ^= x;
|
||||
state2 ^= y;
|
||||
}
|
||||
|
||||
public:
|
||||
constexpr TinyMT() = default;
|
||||
|
||||
// Public API.
|
||||
|
||||
// Initialization.
|
||||
void Initialize(u32 seed) {
|
||||
this->state.data[0] = seed;
|
||||
this->state.data[1] = ParamMat1;
|
||||
this->state.data[2] = ParamMat2;
|
||||
this->state.data[3] = ParamTmat;
|
||||
|
||||
for (int i = 1; i < MinimumInitIterations; i++) {
|
||||
const u32 mixed = XorByShifted30(this->state.data[(i - 1) % NumStateWords]);
|
||||
this->state.data[i % NumStateWords] ^= mixed * ParamMult + i;
|
||||
}
|
||||
|
||||
this->FinalizeInitialization();
|
||||
}
|
||||
|
||||
void Initialize(const u32* seed, int seed_count) {
|
||||
this->state.data[0] = 0;
|
||||
this->state.data[1] = ParamMat1;
|
||||
this->state.data[2] = ParamMat2;
|
||||
this->state.data[3] = ParamTmat;
|
||||
|
||||
{
|
||||
const int num_init_iterations = std::max(seed_count + 1, MinimumInitIterations) - 1;
|
||||
|
||||
GenerateInitialValuePlus(&this->state, 0, seed_count);
|
||||
|
||||
for (int i = 0; i < num_init_iterations; i++) {
|
||||
GenerateInitialValuePlus(&this->state, (i + 1) % NumStateWords,
|
||||
(i < seed_count) ? seed[i] : 0);
|
||||
}
|
||||
|
||||
for (int i = 0; i < static_cast<int>(NumStateWords); i++) {
|
||||
GenerateInitialValueXor(&this->state,
|
||||
(i + 1 + num_init_iterations) % NumStateWords);
|
||||
}
|
||||
}
|
||||
|
||||
this->FinalizeInitialization();
|
||||
}
|
||||
|
||||
// State management.
|
||||
void GetState(TinyMT::State& out) const {
|
||||
out.data = this->state.data;
|
||||
}
|
||||
|
||||
void SetState(const TinyMT::State& state_) {
|
||||
this->state.data = state_.data;
|
||||
}
|
||||
|
||||
// Random generation.
|
||||
void GenerateRandomBytes(void* dst, std::size_t size) {
|
||||
const uintptr_t start = reinterpret_cast<uintptr_t>(dst);
|
||||
const uintptr_t end = start + size;
|
||||
const uintptr_t aligned_start = Common::AlignUp(start, 4);
|
||||
const uintptr_t aligned_end = Common::AlignDown(end, 4);
|
||||
|
||||
// Make sure we're aligned.
|
||||
if (start < aligned_start) {
|
||||
const u32 rnd = this->GenerateRandomU32();
|
||||
std::memcpy(dst, &rnd, aligned_start - start);
|
||||
}
|
||||
|
||||
// Write as many aligned u32s as we can.
|
||||
{
|
||||
u32* cur_dst = reinterpret_cast<u32*>(aligned_start);
|
||||
u32* const end_dst = reinterpret_cast<u32*>(aligned_end);
|
||||
|
||||
while (cur_dst < end_dst) {
|
||||
*(cur_dst++) = this->GenerateRandomU32();
|
||||
}
|
||||
}
|
||||
|
||||
// Handle any leftover unaligned data.
|
||||
if (aligned_end < end) {
|
||||
const u32 rnd = this->GenerateRandomU32();
|
||||
std::memcpy(reinterpret_cast<void*>(aligned_end), &rnd, end - aligned_end);
|
||||
}
|
||||
}
|
||||
|
||||
u32 GenerateRandomU32() {
|
||||
// Advance state.
|
||||
const u32 x0 =
|
||||
(this->state.data[0] & TopBitmask) ^ this->state.data[1] ^ this->state.data[2];
|
||||
const u32 y0 = this->state.data[3];
|
||||
const u32 x1 = x0 ^ (x0 << 1);
|
||||
const u32 y1 = y0 ^ (y0 >> 1) ^ x1;
|
||||
|
||||
const u32 state0 = this->state.data[1];
|
||||
u32 state1 = this->state.data[2];
|
||||
u32 state2 = x1 ^ (y1 << 10);
|
||||
const u32 state3 = y1;
|
||||
|
||||
if ((y1 & 1) != 0) {
|
||||
state1 ^= ParamMat1;
|
||||
state2 ^= ParamMat2;
|
||||
}
|
||||
|
||||
this->state.data[0] = state0;
|
||||
this->state.data[1] = state1;
|
||||
this->state.data[2] = state2;
|
||||
this->state.data[3] = state3;
|
||||
|
||||
// Temper.
|
||||
const u32 t1 = state0 + (state2 >> 8);
|
||||
u32 t0 = state3 ^ t1;
|
||||
|
||||
if ((t1 & 1) != 0) {
|
||||
t0 ^= ParamTmat;
|
||||
}
|
||||
|
||||
return t0;
|
||||
}
|
||||
|
||||
u64 GenerateRandomU64() {
|
||||
const u32 lo = this->GenerateRandomU32();
|
||||
const u32 hi = this->GenerateRandomU32();
|
||||
return (u64{hi} << 32) | u64{lo};
|
||||
}
|
||||
|
||||
float GenerateRandomF32() {
|
||||
// Floats have 24 bits of mantissa.
|
||||
constexpr u32 MantissaBits = 24;
|
||||
return static_cast<float>(GenerateRandomU24()) * (1.0f / (1U << MantissaBits));
|
||||
}
|
||||
|
||||
double GenerateRandomF64() {
|
||||
// Doubles have 53 bits of mantissa.
|
||||
// The smart way to generate 53 bits of random would be to use 32 bits
|
||||
// from the first rnd32() call, and then 21 from the second.
|
||||
// Nintendo does not. They use (32 - 5) = 27 bits from the first rnd32()
|
||||
// call, and (32 - 6) bits from the second. We'll do what they do, but
|
||||
// There's not a clear reason why.
|
||||
constexpr u32 MantissaBits = 53;
|
||||
constexpr u32 Shift1st = (64 - MantissaBits) / 2;
|
||||
constexpr u32 Shift2nd = (64 - MantissaBits) - Shift1st;
|
||||
|
||||
const u32 first = (this->GenerateRandomU32() >> Shift1st);
|
||||
const u32 second = (this->GenerateRandomU32() >> Shift2nd);
|
||||
|
||||
return (1.0 * first * (u64{1} << (32 - Shift2nd)) + second) *
|
||||
(1.0 / (u64{1} << MantissaBits));
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace Common
|
||||
@ -0,0 +1,279 @@
|
||||
// Copyright 2021 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <array>
|
||||
#include <bit>
|
||||
|
||||
#include "common/alignment.h"
|
||||
#include "common/assert.h"
|
||||
#include "common/bit_util.h"
|
||||
#include "common/common_types.h"
|
||||
#include "common/tiny_mt.h"
|
||||
#include "core/hle/kernel/k_system_control.h"
|
||||
|
||||
namespace Kernel {
|
||||
|
||||
class KPageBitmap {
|
||||
private:
|
||||
class RandomBitGenerator {
|
||||
private:
|
||||
Common::TinyMT rng{};
|
||||
u32 entropy{};
|
||||
u32 bits_available{};
|
||||
|
||||
private:
|
||||
void RefreshEntropy() {
|
||||
entropy = rng.GenerateRandomU32();
|
||||
bits_available = static_cast<u32>(Common::BitSize<decltype(entropy)>());
|
||||
}
|
||||
|
||||
bool GenerateRandomBit() {
|
||||
if (bits_available == 0) {
|
||||
this->RefreshEntropy();
|
||||
}
|
||||
|
||||
const bool rnd_bit = (entropy & 1) != 0;
|
||||
entropy >>= 1;
|
||||
--bits_available;
|
||||
return rnd_bit;
|
||||
}
|
||||
|
||||
public:
|
||||
RandomBitGenerator() {
|
||||
rng.Initialize(static_cast<u32>(KSystemControl::GenerateRandomU64()));
|
||||
}
|
||||
|
||||
std::size_t SelectRandomBit(u64 bitmap) {
|
||||
u64 selected = 0;
|
||||
|
||||
u64 cur_num_bits = Common::BitSize<decltype(bitmap)>() / 2;
|
||||
u64 cur_mask = (1ULL << cur_num_bits) - 1;
|
||||
|
||||
while (cur_num_bits) {
|
||||
const u64 low = (bitmap >> 0) & cur_mask;
|
||||
const u64 high = (bitmap >> cur_num_bits) & cur_mask;
|
||||
|
||||
bool choose_low;
|
||||
if (high == 0) {
|
||||
// If only low val is set, choose low.
|
||||
choose_low = true;
|
||||
} else if (low == 0) {
|
||||
// If only high val is set, choose high.
|
||||
choose_low = false;
|
||||
} else {
|
||||
// If both are set, choose random.
|
||||
choose_low = this->GenerateRandomBit();
|
||||
}
|
||||
|
||||
// If we chose low, proceed with low.
|
||||
if (choose_low) {
|
||||
bitmap = low;
|
||||
selected += 0;
|
||||
} else {
|
||||
bitmap = high;
|
||||
selected += cur_num_bits;
|
||||
}
|
||||
|
||||
// Proceed.
|
||||
cur_num_bits /= 2;
|
||||
cur_mask >>= cur_num_bits;
|
||||
}
|
||||
|
||||
return selected;
|
||||
}
|
||||
};
|
||||
|
||||
public:
|
||||
static constexpr std::size_t MaxDepth = 4;
|
||||
|
||||
private:
|
||||
std::array<u64*, MaxDepth> bit_storages{};
|
||||
RandomBitGenerator rng{};
|
||||
std::size_t num_bits{};
|
||||
std::size_t used_depths{};
|
||||
|
||||
public:
|
||||
KPageBitmap() = default;
|
||||
|
||||
constexpr std::size_t GetNumBits() const {
|
||||
return num_bits;
|
||||
}
|
||||
constexpr s32 GetHighestDepthIndex() const {
|
||||
return static_cast<s32>(used_depths) - 1;
|
||||
}
|
||||
|
||||
u64* Initialize(u64* storage, std::size_t size) {
|
||||
// Initially, everything is un-set.
|
||||
num_bits = 0;
|
||||
|
||||
// Calculate the needed bitmap depth.
|
||||
used_depths = static_cast<std::size_t>(GetRequiredDepth(size));
|
||||
ASSERT(used_depths <= MaxDepth);
|
||||
|
||||
// Set the bitmap pointers.
|
||||
for (s32 depth = this->GetHighestDepthIndex(); depth >= 0; depth--) {
|
||||
bit_storages[depth] = storage;
|
||||
size = Common::AlignUp(size, Common::BitSize<u64>()) / Common::BitSize<u64>();
|
||||
storage += size;
|
||||
}
|
||||
|
||||
return storage;
|
||||
}
|
||||
|
||||
s64 FindFreeBlock(bool random) {
|
||||
uintptr_t offset = 0;
|
||||
s32 depth = 0;
|
||||
|
||||
if (random) {
|
||||
do {
|
||||
const u64 v = bit_storages[depth][offset];
|
||||
if (v == 0) {
|
||||
// If depth is bigger than zero, then a previous level indicated a block was
|
||||
// free.
|
||||
ASSERT(depth == 0);
|
||||
return -1;
|
||||
}
|
||||
offset = offset * Common::BitSize<u64>() + rng.SelectRandomBit(v);
|
||||
++depth;
|
||||
} while (depth < static_cast<s32>(used_depths));
|
||||
} else {
|
||||
do {
|
||||
const u64 v = bit_storages[depth][offset];
|
||||
if (v == 0) {
|
||||
// If depth is bigger than zero, then a previous level indicated a block was
|
||||
// free.
|
||||
ASSERT(depth == 0);
|
||||
return -1;
|
||||
}
|
||||
offset = offset * Common::BitSize<u64>() + std::countr_zero(v);
|
||||
++depth;
|
||||
} while (depth < static_cast<s32>(used_depths));
|
||||
}
|
||||
|
||||
return static_cast<s64>(offset);
|
||||
}
|
||||
|
||||
void SetBit(std::size_t offset) {
|
||||
this->SetBit(this->GetHighestDepthIndex(), offset);
|
||||
num_bits++;
|
||||
}
|
||||
|
||||
void ClearBit(std::size_t offset) {
|
||||
this->ClearBit(this->GetHighestDepthIndex(), offset);
|
||||
num_bits--;
|
||||
}
|
||||
|
||||
bool ClearRange(std::size_t offset, std::size_t count) {
|
||||
s32 depth = this->GetHighestDepthIndex();
|
||||
u64* bits = bit_storages[depth];
|
||||
std::size_t bit_ind = offset / Common::BitSize<u64>();
|
||||
if (count < Common::BitSize<u64>()) {
|
||||
const std::size_t shift = offset % Common::BitSize<u64>();
|
||||
ASSERT(shift + count <= Common::BitSize<u64>());
|
||||
// Check that all the bits are set.
|
||||
const u64 mask = ((u64(1) << count) - 1) << shift;
|
||||
u64 v = bits[bit_ind];
|
||||
if ((v & mask) != mask) {
|
||||
return false;
|
||||
}
|
||||
|
||||
// Clear the bits.
|
||||
v &= ~mask;
|
||||
bits[bit_ind] = v;
|
||||
if (v == 0) {
|
||||
this->ClearBit(depth - 1, bit_ind);
|
||||
}
|
||||
} else {
|
||||
ASSERT(offset % Common::BitSize<u64>() == 0);
|
||||
ASSERT(count % Common::BitSize<u64>() == 0);
|
||||
// Check that all the bits are set.
|
||||
std::size_t remaining = count;
|
||||
std::size_t i = 0;
|
||||
do {
|
||||
if (bits[bit_ind + i++] != ~u64(0)) {
|
||||
return false;
|
||||
}
|
||||
remaining -= Common::BitSize<u64>();
|
||||
} while (remaining > 0);
|
||||
|
||||
// Clear the bits.
|
||||
remaining = count;
|
||||
i = 0;
|
||||
do {
|
||||
bits[bit_ind + i] = 0;
|
||||
this->ClearBit(depth - 1, bit_ind + i);
|
||||
i++;
|
||||
remaining -= Common::BitSize<u64>();
|
||||
} while (remaining > 0);
|
||||
}
|
||||
|
||||
num_bits -= count;
|
||||
return true;
|
||||
}
|
||||
|
||||
private:
|
||||
void SetBit(s32 depth, std::size_t offset) {
|
||||
while (depth >= 0) {
|
||||
std::size_t ind = offset / Common::BitSize<u64>();
|
||||
std::size_t which = offset % Common::BitSize<u64>();
|
||||
const u64 mask = u64(1) << which;
|
||||
|
||||
u64* bit = std::addressof(bit_storages[depth][ind]);
|
||||
u64 v = *bit;
|
||||
ASSERT((v & mask) == 0);
|
||||
*bit = v | mask;
|
||||
if (v) {
|
||||
break;
|
||||
}
|
||||
offset = ind;
|
||||
depth--;
|
||||
}
|
||||
}
|
||||
|
||||
void ClearBit(s32 depth, std::size_t offset) {
|
||||
while (depth >= 0) {
|
||||
std::size_t ind = offset / Common::BitSize<u64>();
|
||||
std::size_t which = offset % Common::BitSize<u64>();
|
||||
const u64 mask = u64(1) << which;
|
||||
|
||||
u64* bit = std::addressof(bit_storages[depth][ind]);
|
||||
u64 v = *bit;
|
||||
ASSERT((v & mask) != 0);
|
||||
v &= ~mask;
|
||||
*bit = v;
|
||||
if (v) {
|
||||
break;
|
||||
}
|
||||
offset = ind;
|
||||
depth--;
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
static constexpr s32 GetRequiredDepth(std::size_t region_size) {
|
||||
s32 depth = 0;
|
||||
while (true) {
|
||||
region_size /= Common::BitSize<u64>();
|
||||
depth++;
|
||||
if (region_size == 0) {
|
||||
return depth;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
public:
|
||||
static constexpr std::size_t CalculateManagementOverheadSize(std::size_t region_size) {
|
||||
std::size_t overhead_bits = 0;
|
||||
for (s32 depth = GetRequiredDepth(region_size) - 1; depth >= 0; depth--) {
|
||||
region_size =
|
||||
Common::AlignUp(region_size, Common::BitSize<u64>()) / Common::BitSize<u64>();
|
||||
overhead_bits += region_size;
|
||||
}
|
||||
return overhead_bits * sizeof(u64);
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace Kernel
|
||||
@ -0,0 +1,193 @@
|
||||
// Copyright 2020 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <array>
|
||||
#include <bit>
|
||||
#include <vector>
|
||||
|
||||
#include "common/alignment.h"
|
||||
#include "common/assert.h"
|
||||
#include "common/common_funcs.h"
|
||||
#include "common/common_types.h"
|
||||
#include "core/hle/kernel/k_page_bitmap.h"
|
||||
#include "core/hle/kernel/memory_types.h"
|
||||
|
||||
namespace Kernel {
|
||||
|
||||
class KPageHeap final : NonCopyable {
|
||||
public:
|
||||
static constexpr s32 GetAlignedBlockIndex(std::size_t num_pages, std::size_t align_pages) {
|
||||
const auto target_pages{std::max(num_pages, align_pages)};
|
||||
for (std::size_t i = 0; i < NumMemoryBlockPageShifts; i++) {
|
||||
if (target_pages <=
|
||||
(static_cast<std::size_t>(1) << MemoryBlockPageShifts[i]) / PageSize) {
|
||||
return static_cast<s32>(i);
|
||||
}
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
static constexpr s32 GetBlockIndex(std::size_t num_pages) {
|
||||
for (s32 i{static_cast<s32>(NumMemoryBlockPageShifts) - 1}; i >= 0; i--) {
|
||||
if (num_pages >= (static_cast<std::size_t>(1) << MemoryBlockPageShifts[i]) / PageSize) {
|
||||
return i;
|
||||
}
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
static constexpr std::size_t GetBlockSize(std::size_t index) {
|
||||
return static_cast<std::size_t>(1) << MemoryBlockPageShifts[index];
|
||||
}
|
||||
|
||||
static constexpr std::size_t GetBlockNumPages(std::size_t index) {
|
||||
return GetBlockSize(index) / PageSize;
|
||||
}
|
||||
|
||||
private:
|
||||
static constexpr std::size_t NumMemoryBlockPageShifts{7};
|
||||
static constexpr std::array<std::size_t, NumMemoryBlockPageShifts> MemoryBlockPageShifts{
|
||||
0xC, 0x10, 0x15, 0x16, 0x19, 0x1D, 0x1E,
|
||||
};
|
||||
|
||||
class Block final : NonCopyable {
|
||||
private:
|
||||
KPageBitmap bitmap;
|
||||
VAddr heap_address{};
|
||||
uintptr_t end_offset{};
|
||||
std::size_t block_shift{};
|
||||
std::size_t next_block_shift{};
|
||||
|
||||
public:
|
||||
Block() = default;
|
||||
|
||||
constexpr std::size_t GetShift() const {
|
||||
return block_shift;
|
||||
}
|
||||
constexpr std::size_t GetNextShift() const {
|
||||
return next_block_shift;
|
||||
}
|
||||
constexpr std::size_t GetSize() const {
|
||||
return static_cast<std::size_t>(1) << GetShift();
|
||||
}
|
||||
constexpr std::size_t GetNumPages() const {
|
||||
return GetSize() / PageSize;
|
||||
}
|
||||
constexpr std::size_t GetNumFreeBlocks() const {
|
||||
return bitmap.GetNumBits();
|
||||
}
|
||||
constexpr std::size_t GetNumFreePages() const {
|
||||
return GetNumFreeBlocks() * GetNumPages();
|
||||
}
|
||||
|
||||
u64* Initialize(VAddr addr, std::size_t size, std::size_t bs, std::size_t nbs,
|
||||
u64* bit_storage) {
|
||||
// Set shifts
|
||||
block_shift = bs;
|
||||
next_block_shift = nbs;
|
||||
|
||||
// Align up the address
|
||||
VAddr end{addr + size};
|
||||
const auto align{(next_block_shift != 0) ? (1ULL << next_block_shift)
|
||||
: (1ULL << block_shift)};
|
||||
addr = Common::AlignDown((addr), align);
|
||||
end = Common::AlignUp((end), align);
|
||||
|
||||
heap_address = addr;
|
||||
end_offset = (end - addr) / (1ULL << block_shift);
|
||||
return bitmap.Initialize(bit_storage, end_offset);
|
||||
}
|
||||
|
||||
VAddr PushBlock(VAddr address) {
|
||||
// Set the bit for the free block
|
||||
std::size_t offset{(address - heap_address) >> GetShift()};
|
||||
bitmap.SetBit(offset);
|
||||
|
||||
// If we have a next shift, try to clear the blocks below and return the address
|
||||
if (GetNextShift()) {
|
||||
const auto diff{1ULL << (GetNextShift() - GetShift())};
|
||||
offset = Common::AlignDown(offset, diff);
|
||||
if (bitmap.ClearRange(offset, diff)) {
|
||||
return heap_address + (offset << GetShift());
|
||||
}
|
||||
}
|
||||
|
||||
// We couldn't coalesce, or we're already as big as possible
|
||||
return 0;
|
||||
}
|
||||
|
||||
VAddr PopBlock(bool random) {
|
||||
// Find a free block
|
||||
const s64 soffset{bitmap.FindFreeBlock(random)};
|
||||
if (soffset < 0) {
|
||||
return 0;
|
||||
}
|
||||
const auto offset{static_cast<std::size_t>(soffset)};
|
||||
|
||||
// Update our tracking and return it
|
||||
bitmap.ClearBit(offset);
|
||||
return heap_address + (offset << GetShift());
|
||||
}
|
||||
|
||||
public:
|
||||
static constexpr std::size_t CalculateManagementOverheadSize(std::size_t region_size,
|
||||
std::size_t cur_block_shift,
|
||||
std::size_t next_block_shift) {
|
||||
const auto cur_block_size{(1ULL << cur_block_shift)};
|
||||
const auto next_block_size{(1ULL << next_block_shift)};
|
||||
const auto align{(next_block_shift != 0) ? next_block_size : cur_block_size};
|
||||
return KPageBitmap::CalculateManagementOverheadSize(
|
||||
(align * 2 + Common::AlignUp(region_size, align)) / cur_block_size);
|
||||
}
|
||||
};
|
||||
|
||||
public:
|
||||
KPageHeap() = default;
|
||||
|
||||
constexpr VAddr GetAddress() const {
|
||||
return heap_address;
|
||||
}
|
||||
constexpr std::size_t GetSize() const {
|
||||
return heap_size;
|
||||
}
|
||||
constexpr VAddr GetEndAddress() const {
|
||||
return GetAddress() + GetSize();
|
||||
}
|
||||
constexpr std::size_t GetPageOffset(VAddr block) const {
|
||||
return (block - GetAddress()) / PageSize;
|
||||
}
|
||||
|
||||
void Initialize(VAddr heap_address, std::size_t heap_size, std::size_t metadata_size);
|
||||
VAddr AllocateBlock(s32 index, bool random);
|
||||
void Free(VAddr addr, std::size_t num_pages);
|
||||
|
||||
void UpdateUsedSize() {
|
||||
used_size = heap_size - (GetNumFreePages() * PageSize);
|
||||
}
|
||||
|
||||
static std::size_t CalculateManagementOverheadSize(std::size_t region_size);
|
||||
|
||||
private:
|
||||
constexpr std::size_t GetNumFreePages() const {
|
||||
std::size_t num_free{};
|
||||
|
||||
for (const auto& block : blocks) {
|
||||
num_free += block.GetNumFreePages();
|
||||
}
|
||||
|
||||
return num_free;
|
||||
}
|
||||
|
||||
void FreeBlock(VAddr block, s32 index);
|
||||
|
||||
VAddr heap_address{};
|
||||
std::size_t heap_size{};
|
||||
std::size_t used_size{};
|
||||
std::array<Block, NumMemoryBlockPageShifts> blocks{};
|
||||
std::vector<u64> metadata;
|
||||
};
|
||||
|
||||
} // namespace Kernel
|
||||
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,54 @@
|
||||
// Copyright 2021 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#include "core/hle/kernel/k_spin_lock.h"
|
||||
|
||||
#if _MSC_VER
|
||||
#include <intrin.h>
|
||||
#if _M_AMD64
|
||||
#define __x86_64__ 1
|
||||
#endif
|
||||
#if _M_ARM64
|
||||
#define __aarch64__ 1
|
||||
#endif
|
||||
#else
|
||||
#if __x86_64__
|
||||
#include <xmmintrin.h>
|
||||
#endif
|
||||
#endif
|
||||
|
||||
namespace {
|
||||
|
||||
void ThreadPause() {
|
||||
#if __x86_64__
|
||||
_mm_pause();
|
||||
#elif __aarch64__ && _MSC_VER
|
||||
__yield();
|
||||
#elif __aarch64__
|
||||
asm("yield");
|
||||
#endif
|
||||
}
|
||||
|
||||
} // namespace
|
||||
|
||||
namespace Kernel {
|
||||
|
||||
void KSpinLock::Lock() {
|
||||
while (lck.test_and_set(std::memory_order_acquire)) {
|
||||
ThreadPause();
|
||||
}
|
||||
}
|
||||
|
||||
void KSpinLock::Unlock() {
|
||||
lck.clear(std::memory_order_release);
|
||||
}
|
||||
|
||||
bool KSpinLock::TryLock() {
|
||||
if (lck.test_and_set(std::memory_order_acquire)) {
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
} // namespace Kernel
|
||||
@ -0,0 +1,33 @@
|
||||
// Copyright 2021 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <atomic>
|
||||
|
||||
#include "core/hle/kernel/k_scoped_lock.h"
|
||||
|
||||
namespace Kernel {
|
||||
|
||||
class KSpinLock {
|
||||
public:
|
||||
KSpinLock() = default;
|
||||
|
||||
KSpinLock(const KSpinLock&) = delete;
|
||||
KSpinLock& operator=(const KSpinLock&) = delete;
|
||||
|
||||
KSpinLock(KSpinLock&&) = delete;
|
||||
KSpinLock& operator=(KSpinLock&&) = delete;
|
||||
|
||||
void Lock();
|
||||
void Unlock();
|
||||
[[nodiscard]] bool TryLock();
|
||||
|
||||
private:
|
||||
std::atomic_flag lck = ATOMIC_FLAG_INIT;
|
||||
};
|
||||
|
||||
using KScopedSpinLock = KScopedLock<KSpinLock>;
|
||||
|
||||
} // namespace Kernel
|
||||
@ -0,0 +1,19 @@
|
||||
// Copyright 2021 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "common/common_types.h"
|
||||
|
||||
namespace Kernel {
|
||||
|
||||
class KSystemControl {
|
||||
public:
|
||||
KSystemControl() = default;
|
||||
|
||||
static u64 GenerateRandomRange(u64 min, u64 max);
|
||||
static u64 GenerateRandomU64();
|
||||
};
|
||||
|
||||
} // namespace Kernel
|
||||
@ -1,370 +0,0 @@
|
||||
// Copyright 2020 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
// This file references various implementation details from Atmosphere, an open-source firmware for
|
||||
// the Nintendo Switch. Copyright 2018-2020 Atmosphere-NX.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <array>
|
||||
#include <bit>
|
||||
#include <vector>
|
||||
|
||||
#include "common/alignment.h"
|
||||
#include "common/assert.h"
|
||||
#include "common/common_funcs.h"
|
||||
#include "common/common_types.h"
|
||||
#include "core/hle/kernel/memory/memory_types.h"
|
||||
|
||||
namespace Kernel::Memory {
|
||||
|
||||
class PageHeap final : NonCopyable {
|
||||
public:
|
||||
static constexpr s32 GetAlignedBlockIndex(std::size_t num_pages, std::size_t align_pages) {
|
||||
const auto target_pages{std::max(num_pages, align_pages)};
|
||||
for (std::size_t i = 0; i < NumMemoryBlockPageShifts; i++) {
|
||||
if (target_pages <=
|
||||
(static_cast<std::size_t>(1) << MemoryBlockPageShifts[i]) / PageSize) {
|
||||
return static_cast<s32>(i);
|
||||
}
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
static constexpr s32 GetBlockIndex(std::size_t num_pages) {
|
||||
for (s32 i{static_cast<s32>(NumMemoryBlockPageShifts) - 1}; i >= 0; i--) {
|
||||
if (num_pages >= (static_cast<std::size_t>(1) << MemoryBlockPageShifts[i]) / PageSize) {
|
||||
return i;
|
||||
}
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
static constexpr std::size_t GetBlockSize(std::size_t index) {
|
||||
return static_cast<std::size_t>(1) << MemoryBlockPageShifts[index];
|
||||
}
|
||||
|
||||
static constexpr std::size_t GetBlockNumPages(std::size_t index) {
|
||||
return GetBlockSize(index) / PageSize;
|
||||
}
|
||||
|
||||
private:
|
||||
static constexpr std::size_t NumMemoryBlockPageShifts{7};
|
||||
static constexpr std::array<std::size_t, NumMemoryBlockPageShifts> MemoryBlockPageShifts{
|
||||
0xC, 0x10, 0x15, 0x16, 0x19, 0x1D, 0x1E,
|
||||
};
|
||||
|
||||
class Block final : NonCopyable {
|
||||
private:
|
||||
class Bitmap final : NonCopyable {
|
||||
public:
|
||||
static constexpr std::size_t MaxDepth{4};
|
||||
|
||||
private:
|
||||
std::array<u64*, MaxDepth> bit_storages{};
|
||||
std::size_t num_bits{};
|
||||
std::size_t used_depths{};
|
||||
|
||||
public:
|
||||
constexpr Bitmap() = default;
|
||||
|
||||
constexpr std::size_t GetNumBits() const {
|
||||
return num_bits;
|
||||
}
|
||||
constexpr s32 GetHighestDepthIndex() const {
|
||||
return static_cast<s32>(used_depths) - 1;
|
||||
}
|
||||
|
||||
constexpr u64* Initialize(u64* storage, std::size_t size) {
|
||||
//* Initially, everything is un-set
|
||||
num_bits = 0;
|
||||
|
||||
// Calculate the needed bitmap depth
|
||||
used_depths = static_cast<std::size_t>(GetRequiredDepth(size));
|
||||
ASSERT(used_depths <= MaxDepth);
|
||||
|
||||
// Set the bitmap pointers
|
||||
for (s32 depth{GetHighestDepthIndex()}; depth >= 0; depth--) {
|
||||
bit_storages[depth] = storage;
|
||||
size = Common::AlignUp(size, 64) / 64;
|
||||
storage += size;
|
||||
}
|
||||
|
||||
return storage;
|
||||
}
|
||||
|
||||
s64 FindFreeBlock() const {
|
||||
uintptr_t offset{};
|
||||
s32 depth{};
|
||||
|
||||
do {
|
||||
const u64 v{bit_storages[depth][offset]};
|
||||
if (v == 0) {
|
||||
// Non-zero depth indicates that a previous level had a free block
|
||||
ASSERT(depth == 0);
|
||||
return -1;
|
||||
}
|
||||
offset = offset * 64 + static_cast<u32>(std::countr_zero(v));
|
||||
++depth;
|
||||
} while (depth < static_cast<s32>(used_depths));
|
||||
|
||||
return static_cast<s64>(offset);
|
||||
}
|
||||
|
||||
constexpr void SetBit(std::size_t offset) {
|
||||
SetBit(GetHighestDepthIndex(), offset);
|
||||
num_bits++;
|
||||
}
|
||||
|
||||
constexpr void ClearBit(std::size_t offset) {
|
||||
ClearBit(GetHighestDepthIndex(), offset);
|
||||
num_bits--;
|
||||
}
|
||||
|
||||
constexpr bool ClearRange(std::size_t offset, std::size_t count) {
|
||||
const s32 depth{GetHighestDepthIndex()};
|
||||
const auto bit_ind{offset / 64};
|
||||
u64* bits{bit_storages[depth]};
|
||||
if (count < 64) {
|
||||
const auto shift{offset % 64};
|
||||
ASSERT(shift + count <= 64);
|
||||
// Check that all the bits are set
|
||||
const u64 mask{((1ULL << count) - 1) << shift};
|
||||
u64 v{bits[bit_ind]};
|
||||
if ((v & mask) != mask) {
|
||||
return false;
|
||||
}
|
||||
|
||||
// Clear the bits
|
||||
v &= ~mask;
|
||||
bits[bit_ind] = v;
|
||||
if (v == 0) {
|
||||
ClearBit(depth - 1, bit_ind);
|
||||
}
|
||||
} else {
|
||||
ASSERT(offset % 64 == 0);
|
||||
ASSERT(count % 64 == 0);
|
||||
// Check that all the bits are set
|
||||
std::size_t remaining{count};
|
||||
std::size_t i = 0;
|
||||
do {
|
||||
if (bits[bit_ind + i++] != ~u64(0)) {
|
||||
return false;
|
||||
}
|
||||
remaining -= 64;
|
||||
} while (remaining > 0);
|
||||
|
||||
// Clear the bits
|
||||
remaining = count;
|
||||
i = 0;
|
||||
do {
|
||||
bits[bit_ind + i] = 0;
|
||||
ClearBit(depth - 1, bit_ind + i);
|
||||
i++;
|
||||
remaining -= 64;
|
||||
} while (remaining > 0);
|
||||
}
|
||||
|
||||
num_bits -= count;
|
||||
return true;
|
||||
}
|
||||
|
||||
private:
|
||||
constexpr void SetBit(s32 depth, std::size_t offset) {
|
||||
while (depth >= 0) {
|
||||
const auto ind{offset / 64};
|
||||
const auto which{offset % 64};
|
||||
const u64 mask{1ULL << which};
|
||||
|
||||
u64* bit{std::addressof(bit_storages[depth][ind])};
|
||||
const u64 v{*bit};
|
||||
ASSERT((v & mask) == 0);
|
||||
*bit = v | mask;
|
||||
if (v) {
|
||||
break;
|
||||
}
|
||||
offset = ind;
|
||||
depth--;
|
||||
}
|
||||
}
|
||||
|
||||
constexpr void ClearBit(s32 depth, std::size_t offset) {
|
||||
while (depth >= 0) {
|
||||
const auto ind{offset / 64};
|
||||
const auto which{offset % 64};
|
||||
const u64 mask{1ULL << which};
|
||||
|
||||
u64* bit{std::addressof(bit_storages[depth][ind])};
|
||||
u64 v{*bit};
|
||||
ASSERT((v & mask) != 0);
|
||||
v &= ~mask;
|
||||
*bit = v;
|
||||
if (v) {
|
||||
break;
|
||||
}
|
||||
offset = ind;
|
||||
depth--;
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
static constexpr s32 GetRequiredDepth(std::size_t region_size) {
|
||||
s32 depth = 0;
|
||||
while (true) {
|
||||
region_size /= 64;
|
||||
depth++;
|
||||
if (region_size == 0) {
|
||||
return depth;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
public:
|
||||
static constexpr std::size_t CalculateMetadataOverheadSize(std::size_t region_size) {
|
||||
std::size_t overhead_bits = 0;
|
||||
for (s32 depth{GetRequiredDepth(region_size) - 1}; depth >= 0; depth--) {
|
||||
region_size = Common::AlignUp(region_size, 64) / 64;
|
||||
overhead_bits += region_size;
|
||||
}
|
||||
return overhead_bits * sizeof(u64);
|
||||
}
|
||||
};
|
||||
|
||||
private:
|
||||
Bitmap bitmap;
|
||||
VAddr heap_address{};
|
||||
uintptr_t end_offset{};
|
||||
std::size_t block_shift{};
|
||||
std::size_t next_block_shift{};
|
||||
|
||||
public:
|
||||
constexpr Block() = default;
|
||||
|
||||
constexpr std::size_t GetShift() const {
|
||||
return block_shift;
|
||||
}
|
||||
constexpr std::size_t GetNextShift() const {
|
||||
return next_block_shift;
|
||||
}
|
||||
constexpr std::size_t GetSize() const {
|
||||
return static_cast<std::size_t>(1) << GetShift();
|
||||
}
|
||||
constexpr std::size_t GetNumPages() const {
|
||||
return GetSize() / PageSize;
|
||||
}
|
||||
constexpr std::size_t GetNumFreeBlocks() const {
|
||||
return bitmap.GetNumBits();
|
||||
}
|
||||
constexpr std::size_t GetNumFreePages() const {
|
||||
return GetNumFreeBlocks() * GetNumPages();
|
||||
}
|
||||
|
||||
constexpr u64* Initialize(VAddr addr, std::size_t size, std::size_t bs, std::size_t nbs,
|
||||
u64* bit_storage) {
|
||||
// Set shifts
|
||||
block_shift = bs;
|
||||
next_block_shift = nbs;
|
||||
|
||||
// Align up the address
|
||||
VAddr end{addr + size};
|
||||
const auto align{(next_block_shift != 0) ? (1ULL << next_block_shift)
|
||||
: (1ULL << block_shift)};
|
||||
addr = Common::AlignDown((addr), align);
|
||||
end = Common::AlignUp((end), align);
|
||||
|
||||
heap_address = addr;
|
||||
end_offset = (end - addr) / (1ULL << block_shift);
|
||||
return bitmap.Initialize(bit_storage, end_offset);
|
||||
}
|
||||
|
||||
constexpr VAddr PushBlock(VAddr address) {
|
||||
// Set the bit for the free block
|
||||
std::size_t offset{(address - heap_address) >> GetShift()};
|
||||
bitmap.SetBit(offset);
|
||||
|
||||
// If we have a next shift, try to clear the blocks below and return the address
|
||||
if (GetNextShift()) {
|
||||
const auto diff{1ULL << (GetNextShift() - GetShift())};
|
||||
offset = Common::AlignDown(offset, diff);
|
||||
if (bitmap.ClearRange(offset, diff)) {
|
||||
return heap_address + (offset << GetShift());
|
||||
}
|
||||
}
|
||||
|
||||
// We couldn't coalesce, or we're already as big as possible
|
||||
return 0;
|
||||
}
|
||||
|
||||
VAddr PopBlock() {
|
||||
// Find a free block
|
||||
const s64 soffset{bitmap.FindFreeBlock()};
|
||||
if (soffset < 0) {
|
||||
return 0;
|
||||
}
|
||||
const auto offset{static_cast<std::size_t>(soffset)};
|
||||
|
||||
// Update our tracking and return it
|
||||
bitmap.ClearBit(offset);
|
||||
return heap_address + (offset << GetShift());
|
||||
}
|
||||
|
||||
public:
|
||||
static constexpr std::size_t CalculateMetadataOverheadSize(std::size_t region_size,
|
||||
std::size_t cur_block_shift,
|
||||
std::size_t next_block_shift) {
|
||||
const auto cur_block_size{(1ULL << cur_block_shift)};
|
||||
const auto next_block_size{(1ULL << next_block_shift)};
|
||||
const auto align{(next_block_shift != 0) ? next_block_size : cur_block_size};
|
||||
return Bitmap::CalculateMetadataOverheadSize(
|
||||
(align * 2 + Common::AlignUp(region_size, align)) / cur_block_size);
|
||||
}
|
||||
};
|
||||
|
||||
public:
|
||||
PageHeap() = default;
|
||||
|
||||
constexpr VAddr GetAddress() const {
|
||||
return heap_address;
|
||||
}
|
||||
constexpr std::size_t GetSize() const {
|
||||
return heap_size;
|
||||
}
|
||||
constexpr VAddr GetEndAddress() const {
|
||||
return GetAddress() + GetSize();
|
||||
}
|
||||
constexpr std::size_t GetPageOffset(VAddr block) const {
|
||||
return (block - GetAddress()) / PageSize;
|
||||
}
|
||||
|
||||
void Initialize(VAddr heap_address, std::size_t heap_size, std::size_t metadata_size);
|
||||
VAddr AllocateBlock(s32 index);
|
||||
void Free(VAddr addr, std::size_t num_pages);
|
||||
|
||||
void UpdateUsedSize() {
|
||||
used_size = heap_size - (GetNumFreePages() * PageSize);
|
||||
}
|
||||
|
||||
static std::size_t CalculateMetadataOverheadSize(std::size_t region_size);
|
||||
|
||||
private:
|
||||
constexpr std::size_t GetNumFreePages() const {
|
||||
std::size_t num_free{};
|
||||
|
||||
for (const auto& block : blocks) {
|
||||
num_free += block.GetNumFreePages();
|
||||
}
|
||||
|
||||
return num_free;
|
||||
}
|
||||
|
||||
void FreeBlock(VAddr block, s32 index);
|
||||
|
||||
VAddr heap_address{};
|
||||
std::size_t heap_size{};
|
||||
std::size_t used_size{};
|
||||
std::array<Block, NumMemoryBlockPageShifts> blocks{};
|
||||
std::vector<u64> metadata;
|
||||
};
|
||||
|
||||
} // namespace Kernel::Memory
|
||||
@ -1,13 +0,0 @@
|
||||
// Copyright 2020 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "common/common_types.h"
|
||||
|
||||
namespace Kernel::Memory::SystemControl {
|
||||
|
||||
u64 GenerateRandomRange(u64 min, u64 max);
|
||||
|
||||
} // namespace Kernel::Memory::SystemControl
|
||||
Loading…
Reference in New Issue