summaryrefslogtreecommitdiffstats
path: root/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h')
-rw-r--r--src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h163
1 files changed, 86 insertions, 77 deletions
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
index f5fb33ba7..1d27739e2 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
@@ -5,14 +5,19 @@
#pragma once
+#include <bit>
+#include <list>
#include <map>
#include <memory>
+#include <mutex>
#include <optional>
#include <vector>
+#include "common/address_space.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/swap.h"
+#include "core/hle/service/nvdrv/core/nvmap.h"
#include "core/hle/service/nvdrv/devices/nvdevice.h"
namespace Tegra {
@@ -30,17 +35,13 @@ class NvMap;
namespace Service::Nvidia::Devices {
-constexpr u32 DEFAULT_BIG_PAGE_SIZE = 1 << 16;
-constexpr u32 DEFAULT_SMALL_PAGE_SIZE = 1 << 12;
-
-class nvmap;
-
-enum class AddressSpaceFlags : u32 {
- None = 0x0,
- FixedOffset = 0x1,
- Remap = 0x100,
+enum class MappingFlags : u32 {
+ None = 0,
+ Fixed = 1 << 0,
+ Sparse = 1 << 1,
+ Remap = 1 << 8,
};
-DECLARE_ENUM_FLAG_OPERATORS(AddressSpaceFlags);
+DECLARE_ENUM_FLAG_OPERATORS(MappingFlags);
class nvhost_as_gpu final : public nvdevice {
public:
@@ -59,46 +60,15 @@ public:
Kernel::KEvent* QueryEvent(u32 event_id) override;
-private:
- class BufferMap final {
- public:
- constexpr BufferMap() = default;
-
- constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_)
- : start_addr{start_addr_}, end_addr{start_addr_ + size_} {}
-
- constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_, VAddr cpu_addr_,
- bool is_allocated_)
- : start_addr{start_addr_}, end_addr{start_addr_ + size_}, cpu_addr{cpu_addr_},
- is_allocated{is_allocated_} {}
-
- constexpr VAddr StartAddr() const {
- return start_addr;
- }
-
- constexpr VAddr EndAddr() const {
- return end_addr;
- }
-
- constexpr std::size_t Size() const {
- return end_addr - start_addr;
- }
-
- constexpr VAddr CpuAddr() const {
- return cpu_addr;
- }
-
- constexpr bool IsAllocated() const {
- return is_allocated;
- }
-
- private:
- GPUVAddr start_addr{};
- GPUVAddr end_addr{};
- VAddr cpu_addr{};
- bool is_allocated{};
+ struct VaRegion {
+ u64 offset;
+ u32 page_size;
+ u32 _pad0_;
+ u64 pages;
};
+ static_assert(sizeof(VaRegion) == 0x18);
+private:
struct IoctlAllocAsEx {
u32_le flags{}; // usually passes 1
s32_le as_fd{}; // ignored; passes 0
@@ -113,7 +83,7 @@ private:
struct IoctlAllocSpace {
u32_le pages{};
u32_le page_size{};
- AddressSpaceFlags flags{};
+ MappingFlags flags{};
INSERT_PADDING_WORDS(1);
union {
u64_le offset;
@@ -130,19 +100,19 @@ private:
static_assert(sizeof(IoctlFreeSpace) == 16, "IoctlFreeSpace is incorrect size");
struct IoctlRemapEntry {
- u16_le flags{};
- u16_le kind{};
- u32_le nvmap_handle{};
- u32_le map_offset{};
- u32_le offset{};
- u32_le pages{};
+ u16 flags;
+ u16 kind;
+ NvCore::NvMap::Handle::Id handle;
+ u32 handle_offset_big_pages;
+ u32 as_offset_big_pages;
+ u32 big_pages;
};
static_assert(sizeof(IoctlRemapEntry) == 20, "IoctlRemapEntry is incorrect size");
struct IoctlMapBufferEx {
- AddressSpaceFlags flags{}; // bit0: fixed_offset, bit2: cacheable
- u32_le kind{}; // -1 is default
- u32_le nvmap_handle{};
+ MappingFlags flags{}; // bit0: fixed_offset, bit2: cacheable
+ u32_le kind{}; // -1 is default
+ NvCore::NvMap::Handle::Id handle;
u32_le page_size{}; // 0 means don't care
s64_le buffer_offset{};
u64_le mapping_size{};
@@ -160,27 +130,15 @@ private:
};
static_assert(sizeof(IoctlBindChannel) == 4, "IoctlBindChannel is incorrect size");
- struct IoctlVaRegion {
- u64_le offset{};
- u32_le page_size{};
- INSERT_PADDING_WORDS(1);
- u64_le pages{};
- };
- static_assert(sizeof(IoctlVaRegion) == 24, "IoctlVaRegion is incorrect size");
-
struct IoctlGetVaRegions {
u64_le buf_addr{}; // (contained output user ptr on linux, ignored)
u32_le buf_size{}; // forced to 2*sizeof(struct va_region)
u32_le reserved{};
- IoctlVaRegion small{};
- IoctlVaRegion big{};
+ std::array<VaRegion, 2> regions{};
};
- static_assert(sizeof(IoctlGetVaRegions) == 16 + sizeof(IoctlVaRegion) * 2,
+ static_assert(sizeof(IoctlGetVaRegions) == 16 + sizeof(VaRegion) * 2,
"IoctlGetVaRegions is incorrect size");
- s32 channel{};
- u32 big_page_size{DEFAULT_BIG_PAGE_SIZE};
-
NvResult AllocAsEx(const std::vector<u8>& input, std::vector<u8>& output);
NvResult AllocateSpace(const std::vector<u8>& input, std::vector<u8>& output);
NvResult Remap(const std::vector<u8>& input, std::vector<u8>& output);
@@ -189,23 +147,74 @@ private:
NvResult FreeSpace(const std::vector<u8>& input, std::vector<u8>& output);
NvResult BindChannel(const std::vector<u8>& input, std::vector<u8>& output);
+ void GetVARegionsImpl(IoctlGetVaRegions& params);
NvResult GetVARegions(const std::vector<u8>& input, std::vector<u8>& output);
NvResult GetVARegions(const std::vector<u8>& input, std::vector<u8>& output,
std::vector<u8>& inline_output);
- std::optional<BufferMap> FindBufferMap(GPUVAddr gpu_addr) const;
- void AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr, bool is_allocated);
- std::optional<std::size_t> RemoveBufferMap(GPUVAddr gpu_addr);
+ void FreeMappingLocked(u64 offset);
Module& module;
NvCore::Container& container;
NvCore::NvMap& nvmap;
+ struct Mapping {
+ VAddr ptr;
+ u64 offset;
+ u64 size;
+ bool fixed;
+ bool big_page; // Only valid if fixed == false
+ bool sparse_alloc;
+
+ Mapping(VAddr ptr_, u64 offset_, u64 size_, bool fixed_, bool big_page_, bool sparse_alloc_)
+ : ptr(ptr_), offset(offset_), size(size_), fixed(fixed_), big_page(big_page_),
+ sparse_alloc(sparse_alloc_) {}
+ };
+
+ struct Allocation {
+ u64 size;
+ std::list<std::shared_ptr<Mapping>> mappings;
+ u32 page_size;
+ bool sparse;
+ };
+
+ std::map<u64, std::shared_ptr<Mapping>>
+ mapping_map; //!< This maps the base addresses of mapped buffers to their total sizes and
+ //!< mapping type, this is needed as what was originally a single buffer may
+ //!< have been split into multiple GPU side buffers with the remap flag.
+ std::map<u64, Allocation> allocation_map; //!< Holds allocations created by AllocSpace from
+ //!< which fixed buffers can be mapped into
+ std::mutex mutex; //!< Locks all AS operations
+
+ struct VM {
+ static constexpr u32 YUZU_PAGESIZE{0x1000};
+ static constexpr u32 PAGE_SIZE_BITS{std::countr_zero(YUZU_PAGESIZE)};
+
+ static constexpr u32 SUPPORTED_BIG_PAGE_SIZES{0x30000};
+ static constexpr u32 DEFAULT_BIG_PAGE_SIZE{0x20000};
+ u32 big_page_size{DEFAULT_BIG_PAGE_SIZE};
+ u32 big_page_size_bits{std::countr_zero(DEFAULT_BIG_PAGE_SIZE)};
+
+ static constexpr u32 VA_START_SHIFT{10};
+ static constexpr u64 DEFAULT_VA_SPLIT{1ULL << 34};
+ static constexpr u64 DEFAULT_VA_RANGE{1ULL << 37};
+ u64 va_range_start{DEFAULT_BIG_PAGE_SIZE << VA_START_SHIFT};
+ u64 va_range_split{DEFAULT_VA_SPLIT};
+ u64 va_range_end{DEFAULT_VA_RANGE};
+
+ using Allocator = Common::FlatAllocator<u32, 0, 32>;
+
+ std::unique_ptr<Allocator> big_page_allocator;
+ std::shared_ptr<Allocator>
+ small_page_allocator; //! Shared as this is also used by nvhost::GpuChannel
+
+ bool initialised{};
+ } vm;
std::shared_ptr<Tegra::MemoryManager> gmmu;
- // This is expected to be ordered, therefore we must use a map, not unordered_map
- std::map<GPUVAddr, BufferMap> buffer_mappings;
+ // s32 channel{};
+ // u32 big_page_size{VM::DEFAULT_BIG_PAGE_SIZE};
};
} // namespace Service::Nvidia::Devices
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-02 10:13:44 +0200