The-Homebrew-Cloud/skyline
2
0
Fork 0
mirror of https://github.com/skyline-emu/skyline.git synced 2024-12-23 03:51:48 +01:00
Code Issues Packages Projects Releases Wiki Activity

Add locking to GPU VMM and fix a few codestyle issues

Browse Source 
As VMM can be accessed by nvdrv and the GPFIFO thread at the same
time locking is needed to prevent races.
This commit is contained in:
Billy Laws 2021-01-21 21:20:00 +00:00 committed by ◱ Mark
parent c1aec00ed1
commit c489da610e
2 changed files with 32 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
app/src/main/cpp/skyline/gpu/memory_manager.cpp
View File

@ -6,17 +6,17 @@
namespace skyline::gpu::vmm {
MemoryManager::MemoryManager(const DeviceState &state) : state(state) {
constexpr u64 GpuAddressSpaceSize{1ul << 40}; //!< The size of the GPU address space
constexpr u64 GpuAddressSpaceBase{0x100000}; //!< The base of the GPU address space - must be non-zero
constexpr u64 gpuAddressSpaceSize{1UL << 40}; //!< The size of the GPU address space
constexpr u64 gpuAddressSpaceBase{0x100000}; //!< The base of the GPU address space - must be non-zero
// Create the initial chunk that will be split to create new chunks
ChunkDescriptor baseChunk(GpuAddressSpaceBase, GpuAddressSpaceSize, 0, ChunkState::Unmapped);
ChunkDescriptor baseChunk(gpuAddressSpaceBase, gpuAddressSpaceSize, nullptr, ChunkState::Unmapped);
chunks.push_back(baseChunk);
}
std::optional<ChunkDescriptor> MemoryManager::FindChunk(ChunkState state, u64 size, u64 alignment) {
auto chunk{std::find_if(chunks.begin(), chunks.end(), [state, size, alignment](const ChunkDescriptor &chunk) -> bool {
return (alignment ? util::IsAligned(chunk.virtAddr, alignment) : true) && chunk.size > size && chunk.state == state;
std::optional<ChunkDescriptor> MemoryManager::FindChunk(ChunkState desiredState, u64 size, u64 alignment) {
auto chunk{std::find_if(chunks.begin(), chunks.end(), [desiredState, size, alignment](const ChunkDescriptor &chunk) -> bool {
return (alignment ? util::IsAligned(chunk.virtAddr, alignment) : true) && chunk.size > size && chunk.state == desiredState;
})};
if (chunk != chunks.end())
@ -41,7 +41,7 @@ namespace skyline::gpu::vmm {
}
if (extension)
chunks.insert(std::next(chunk), ChunkDescriptor(newChunk.virtAddr + newChunk.size, extension, (oldChunk.state == ChunkState::Mapped) ? (oldChunk.cpuPtr + newSize + newChunk.size) : 0, oldChunk.state));
chunks.insert(std::next(chunk), ChunkDescriptor(newChunk.virtAddr + newChunk.size, extension, (oldChunk.state == ChunkState::Mapped) ? (oldChunk.cpuPtr + newSize + newChunk.size) : nullptr, oldChunk.state));
return newChunk.virtAddr;
} else if (chunk->virtAddr + chunk->size > newChunk.virtAddr) {
@ -83,6 +83,8 @@ namespace skyline::gpu::vmm {
u64 MemoryManager::ReserveSpace(u64 size, u64 alignment) {
size = util::AlignUp(size, constant::GpuPageSize);
std::unique_lock lock(vmmMutex);
auto newChunk{FindChunk(ChunkState::Unmapped, size, alignment)};
if (!newChunk)
return 0;
@ -100,11 +102,14 @@ namespace skyline::gpu::vmm {
size = util::AlignUp(size, constant::GpuPageSize);
std::unique_lock lock(vmmMutex);
return InsertChunk(ChunkDescriptor(virtAddr, size, nullptr, ChunkState::Reserved));
}
u64 MemoryManager::MapAllocate(u8 *cpuPtr, u64 size) {
size = util::AlignUp(size, constant::GpuPageSize);
std::unique_lock lock(vmmMutex);
auto mappedChunk{FindChunk(ChunkState::Unmapped, size)};
if (!mappedChunk)
return 0;
@ -123,6 +128,7 @@ namespace skyline::gpu::vmm {
size = util::AlignUp(size, constant::GpuPageSize);
std::unique_lock lock(vmmMutex);
return InsertChunk(ChunkDescriptor(virtAddr, size, cpuPtr, ChunkState::Mapped));
}
@ -131,7 +137,8 @@ namespace skyline::gpu::vmm {
return false;
try {
InsertChunk(ChunkDescriptor(virtAddr, size, 0, ChunkState::Unmapped));
std::unique_lock lock(vmmMutex);
InsertChunk(ChunkDescriptor(virtAddr, size, nullptr, ChunkState::Unmapped));
} catch (const std::exception &e) {
return false;
}
@ -139,7 +146,9 @@ namespace skyline::gpu::vmm {
return true;
}
void MemoryManager::Read(u8 *destination, u64 virtAddr, u64 size) const {
void MemoryManager::Read(u8 *destination, u64 virtAddr, u64 size) {
std::shared_lock lock(vmmMutex);
auto chunk{std::upper_bound(chunks.begin(), chunks.end(), virtAddr, [](const u64 address, const ChunkDescriptor &chunk) -> bool {
return address < chunk.virtAddr;
})};
@ -169,7 +178,9 @@ namespace skyline::gpu::vmm {
}
}
void MemoryManager::Write(u8 *source, u64 virtAddr, u64 size) const {
void MemoryManager::Write(u8 *source, u64 virtAddr, u64 size) {
std::shared_lock lock(vmmMutex);
auto chunk{std::upper_bound(chunks.begin(), chunks.end(), virtAddr, [](const u64 address, const ChunkDescriptor &chunk) -> bool {
return address < chunk.virtAddr;
})};

19
app/src/main/cpp/skyline/gpu/memory_manager.h
View File

@ -40,18 +40,21 @@ namespace skyline {
private:
const DeviceState &state;
std::vector<ChunkDescriptor> chunks;
std::shared_mutex vmmMutex;
/**
* @brief Finds a chunk in the virtual address space that is larger than meets the given requirements
* @param state The state of the chunk to find
* @note vmmMutex MUST be locked when calling this
* @param desiredState The state of the chunk to find
* @param size The minimum size of the chunk to find
* @param alignment The minimum alignment of the chunk to find
* @return The first applicable chunk
*/
std::optional<ChunkDescriptor> FindChunk(ChunkState state, u64 size, u64 alignment = 0);
std::optional<ChunkDescriptor> FindChunk(ChunkState desiredState, u64 size, u64 alignment = 0);
/**
* @brief Inserts a chunk into the chunk list, resizing and splitting as necessary
* @note vmmMutex MUST be locked when calling this
* @param newChunk The chunk to insert
* @return The base virtual address of the inserted chunk
*/
@ -99,13 +102,13 @@ namespace skyline {
*/
bool Unmap(u64 virtAddr, u64 size);
void Read(u8 *destination, u64 virtAddr, u64 size) const;
void Read(u8 *destination, u64 virtAddr, u64 size);
/**
* @brief Reads in a span from a region of the virtual address space
*/
template<typename T>
void Read(span <T> destination, u64 virtAddr) const {
void Read(span <T> destination, u64 virtAddr) {
Read(reinterpret_cast<u8 *>(destination.data()), virtAddr, destination.size_bytes());
}
@ -114,19 +117,19 @@ namespace skyline {
* @tparam T The type of object to return
*/
template<typename T>
T Read(u64 virtAddr) const {
T Read(u64 virtAddr) {
T obj;
Read(reinterpret_cast<u8 *>(&obj), virtAddr, sizeof(T));
return obj;
}
void Write(u8 *source, u64 virtAddr, u64 size) const;
void Write(u8 *source, u64 virtAddr, u64 size);
/**
* @brief Writes out a span to a region of the virtual address space
*/
template<typename T>
void Write(span <T> source, u64 virtAddr) const {
void Write(span <T> source, u64 virtAddr) {
Write(reinterpret_cast<u8 *>(source.data()), virtAddr, source.size_bytes());
}
@ -134,7 +137,7 @@ namespace skyline {
* @brief Reads in an object from a region of the virtual address space
*/
template<typename T>
void Write(T source, u64 virtAddr) const {
void Write(T source, u64 virtAddr) {
Write(reinterpret_cast<u8 *>(&source), virtAddr, sizeof(T));
}
};