SMMU: Initial adaptation to video_core.

10 files changed, 94 insertions, 156 deletions

diff --git a/src/core/hle/service/hle_ipc.cpp b/src/core/hle/service/hle_ipc.cpp
index 3f38ceb03..9f6274c7d 100644
--- a/src/core/hle/service/hle_ipc.cpp
+++ b/src/core/hle/service/hle_ipc.cpp
@@ -22,19 +22,7 @@
 #include "core/hle/service/hle_ipc.h"
 #include "core/hle/service/ipc_helpers.h"
 #include "core/memory.h"
-
-namespace {
-static thread_local std::array read_buffer_data_a{
-    Common::ScratchBuffer<u8>(),
-    Common::ScratchBuffer<u8>(),
-    Common::ScratchBuffer<u8>(),
-};
-static thread_local std::array read_buffer_data_x{
-    Common::ScratchBuffer<u8>(),
-    Common::ScratchBuffer<u8>(),
-    Common::ScratchBuffer<u8>(),
-};
-} // Anonymous namespace
+#include "core/guest_memory.h"
 
 namespace Service {
 
@@ -343,48 +331,27 @@ std::vector<u8> HLERequestContext::ReadBufferCopy(std::size_t buffer_index) cons
 }
 
 std::span<const u8> HLERequestContext::ReadBufferA(std::size_t buffer_index) const {
-    static thread_local std::array read_buffer_a{
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-    };
+    Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::UnsafeRead> gm(memory, 0, 0);
 
     ASSERT_OR_EXECUTE_MSG(
         BufferDescriptorA().size() > buffer_index, { return {}; },
         "BufferDescriptorA invalid buffer_index {}", buffer_index);
-    auto& read_buffer = read_buffer_a[buffer_index];
-    return read_buffer.Read(BufferDescriptorA()[buffer_index].Address(),
-                            BufferDescriptorA()[buffer_index].Size(),
-                            &read_buffer_data_a[buffer_index]);
+    return gm.Read(BufferDescriptorA()[buffer_index].Address(),
+                   BufferDescriptorA()[buffer_index].Size(), &read_buffer_data_a[buffer_index]);
 }
 
 std::span<const u8> HLERequestContext::ReadBufferX(std::size_t buffer_index) const {
-    static thread_local std::array read_buffer_x{
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-    };
+    Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::UnsafeRead> gm(memory, 0, 0);
 
     ASSERT_OR_EXECUTE_MSG(
         BufferDescriptorX().size() > buffer_index, { return {}; },
         "BufferDescriptorX invalid buffer_index {}", buffer_index);
-    auto& read_buffer = read_buffer_x[buffer_index];
-    return read_buffer.Read(BufferDescriptorX()[buffer_index].Address(),
-                            BufferDescriptorX()[buffer_index].Size(),
-                            &read_buffer_data_x[buffer_index]);
+    return gm.Read(BufferDescriptorX()[buffer_index].Address(),
+                   BufferDescriptorX()[buffer_index].Size(), &read_buffer_data_x[buffer_index]);
 }
 
 std::span<const u8> HLERequestContext::ReadBuffer(std::size_t buffer_index) const {
-    static thread_local std::array read_buffer_a{
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-    };
-    static thread_local std::array read_buffer_x{
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-        Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead>(memory, 0, 0),
-    };
+    Core::Memory::CpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::UnsafeRead> gm(memory, 0, 0);
 
     const bool is_buffer_a{BufferDescriptorA().size() > buffer_index &&
                            BufferDescriptorA()[buffer_index].Size()};
@@ -401,18 +368,14 @@ std::span<const u8> HLERequestContext::ReadBuffer(std::size_t buffer_index) cons
         ASSERT_OR_EXECUTE_MSG(
             BufferDescriptorA().size() > buffer_index, { return {}; },
             "BufferDescriptorA invalid buffer_index {}", buffer_index);
-        auto& read_buffer = read_buffer_a[buffer_index];
-        return read_buffer.Read(BufferDescriptorA()[buffer_index].Address(),
-                                BufferDescriptorA()[buffer_index].Size(),
-                                &read_buffer_data_a[buffer_index]);
+        return gm.Read(BufferDescriptorA()[buffer_index].Address(),
+                       BufferDescriptorA()[buffer_index].Size(), &read_buffer_data_a[buffer_index]);
     } else {
         ASSERT_OR_EXECUTE_MSG(
             BufferDescriptorX().size() > buffer_index, { return {}; },
             "BufferDescriptorX invalid buffer_index {}", buffer_index);
-        auto& read_buffer = read_buffer_x[buffer_index];
-        return read_buffer.Read(BufferDescriptorX()[buffer_index].Address(),
-                                BufferDescriptorX()[buffer_index].Size(),
-                                &read_buffer_data_x[buffer_index]);
+        return gm.Read(BufferDescriptorX()[buffer_index].Address(),
+                       BufferDescriptorX()[buffer_index].Size(), &read_buffer_data_x[buffer_index]);
     }
 }
 
diff --git a/src/core/hle/service/hle_ipc.h b/src/core/hle/service/hle_ipc.h
index d550a11b7..8329d7265 100644
--- a/src/core/hle/service/hle_ipc.h
+++ b/src/core/hle/service/hle_ipc.h
@@ -19,8 +19,6 @@
 #include "core/hle/ipc.h"
 #include "core/hle/kernel/k_handle_table.h"
 #include "core/hle/kernel/svc_common.h"
-#include "core/hle/kernel/k_auto_object.h"
-#include "core/hle/kernel/k_handle_table.h"
 
 union Result;
 
@@ -377,10 +375,6 @@ public:
         return nullptr;
     }
 
-    Kernel::KScopedAutoObject<Kernel::KAutoObject> GetObjectFromHandle(u32 handle) {
-        return GetClientHandleTable().GetObjectForIpc(handle, thread);
-    }
-
     [[nodiscard]] std::shared_ptr<SessionRequestManager> GetManager() const {
         return manager.lock();
     }
@@ -432,6 +426,9 @@ private:
 
     Kernel::KernelCore& kernel;
     Core::Memory::Memory& memory;
+
+    mutable std::array<Common::ScratchBuffer<u8>, 3> read_buffer_data_a{};
+    mutable std::array<Common::ScratchBuffer<u8>, 3> read_buffer_data_x{};
 };
 
 } // namespace Service
diff --git a/src/core/hle/service/nvdrv/core/nvmap.cpp b/src/core/hle/service/nvdrv/core/nvmap.cpp
index fd6c9aa0c..7879c6f04 100644
--- a/src/core/hle/service/nvdrv/core/nvmap.cpp
+++ b/src/core/hle/service/nvdrv/core/nvmap.cpp
@@ -2,6 +2,8 @@
 // SPDX-FileCopyrightText: 2022 Skyline Team and Contributors
 // SPDX-License-Identifier: GPL-3.0-or-later
 
+#include <functional>
+
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
@@ -18,6 +20,7 @@ NvMap::Handle::Handle(u64 size_, Id id_)
 }
 
 NvResult NvMap::Handle::Alloc(Flags pFlags, u32 pAlign, u8 pKind, u64 pAddress) {
+    std::scoped_lock lock(mutex);
     // Handles cannot be allocated twice
     if (allocated) {
         return NvResult::AccessDenied;
@@ -78,11 +81,9 @@ void NvMap::UnmapHandle(Handle& handle_description) {
 
     // Free and unmap the handle from the SMMU
     auto& smmu = host1x.MemoryManager();
-    smmu.Unmap(static_cast<DAddr>(handle_description.pin_virt_address),
-               handle_description.aligned_size);
-    smmu.Free(handle_description.pin_virt_address,
-              static_cast<size_t>(handle_description.aligned_size));
-    handle_description.pin_virt_address = 0;
+    smmu.Unmap(handle_description.d_address, handle_description.aligned_size);
+    smmu.Free(handle_description.d_address, static_cast<size_t>(handle_description.aligned_size));
+    handle_description.d_address = 0;
 }
 
 bool NvMap::TryRemoveHandle(const Handle& handle_description) {
@@ -123,41 +124,16 @@ std::shared_ptr<NvMap::Handle> NvMap::GetHandle(Handle::Id handle) {
     }
 }
 
-VAddr NvMap::GetHandleAddress(Handle::Id handle) {
+DAddr NvMap::GetHandleAddress(Handle::Id handle) {
     std::scoped_lock lock(handles_lock);
     try {
-        return handles.at(handle)->address;
+        return handles.at(handle)->d_address;
     } catch (std::out_of_range&) {
         return 0;
     }
 }
 
-NvResult NvMap::AllocateHandle(Handle::Id handle, Handle::Flags pFlags, u32 pAlign, u8 pKind, u64 pAddress, size_t session_id) {
-    auto handle_description{GetHandle(handle)};
-    if (!handle_description) [[unlikely]] {
-        return NvResult::BadParameter;
-    }
-
-    if (handle_description->allocated) [[unlikely]] {
-        return NvResult::InsufficientMemory;
-    }
-
-    std::scoped_lock lock(handle_description->mutex);
-    NvResult result = handle_description->Alloc(pFlags, pAlign, pKind, pAddress);
-    if (result != NvResult::Success) {
-        return result;
-    }
-    auto& smmu = host1x.MemoryManager();
-    size_t total_size = static_cast<size_t>(handle_description->aligned_size);
-    handle_description->d_address = smmu.Allocate(total_size);
-    if (handle_description->d_address == 0) {
-        return NvResult::InsufficientMemory;
-    }
-    smmu.Map(handle_description->d_address, handle_description->address, total_size, session_id);
-    return NvResult::Success;
-}
-
-u32 NvMap::PinHandle(NvMap::Handle::Id handle, size_t session_id) {
+DAddr NvMap::PinHandle(NvMap::Handle::Id handle, size_t session_id, bool low_area_pin) {
     auto handle_description{GetHandle(handle)};
     if (!handle_description) [[unlikely]] {
         return 0;
@@ -176,35 +152,38 @@ u32 NvMap::PinHandle(NvMap::Handle::Id handle, size_t session_id) {
                 handle_description->unmap_queue_entry.reset();
 
                 handle_description->pins++;
-                return handle_description->pin_virt_address;
+                return handle_description->d_address;
             }
         }
 
+        using namespace std::placeholders;
         // If not then allocate some space and map it
         DAddr address{};
         auto& smmu = host1x.MemoryManager();
-        while ((address = smmu.AllocatePinned(
-                    static_cast<size_t>(handle_description->aligned_size))) == 0) {
+        auto allocate = std::bind(&Tegra::MaxwellDeviceMemoryManager::Allocate, &smmu, _1);
+                         //: std::bind(&Tegra::MaxwellDeviceMemoryManager::Allocate, &smmu, _1);
+        while ((address = allocate(static_cast<size_t>(handle_description->aligned_size))) == 0) {
             // Free handles until the allocation succeeds
             std::scoped_lock queueLock(unmap_queue_lock);
             if (auto freeHandleDesc{unmap_queue.front()}) {
                 // Handles in the unmap queue are guaranteed not to be pinned so don't bother
                 // checking if they are before unmapping
                 std::scoped_lock freeLock(freeHandleDesc->mutex);
-                if (handle_description->pin_virt_address)
+                if (handle_description->d_address)
                     UnmapHandle(*freeHandleDesc);
             } else {
                 LOG_CRITICAL(Service_NVDRV, "Ran out of SMMU address space!");
             }
         }
 
+        handle_description->d_address = address;
+
         smmu.Map(address, handle_description->address, handle_description->aligned_size,
                  session_id);
-        handle_description->pin_virt_address = static_cast<u32>(address);
     }
 
     handle_description->pins++;
-    return handle_description->pin_virt_address;
+    return handle_description->d_address;
 }
 
 void NvMap::UnpinHandle(Handle::Id handle) {
@@ -255,15 +234,10 @@ std::optional<NvMap::FreeInfo> NvMap::FreeHandle(Handle::Id handle, bool interna
                 LOG_WARNING(Service_NVDRV, "User duplicate count imbalance detected!");
             } else if (handle_description->dupes == 0) {
                 // Force unmap the handle
-                if (handle_description->pin_virt_address) {
+                if (handle_description->d_address) {
                     std::scoped_lock queueLock(unmap_queue_lock);
                     UnmapHandle(*handle_description);
                 }
-                if (handle_description->allocated) {
-                    auto& smmu = host1x.MemoryManager();
-                    smmu.Free(handle_description->d_address, handle_description->aligned_size);
-                    smmu.Unmap(handle_description->d_address, handle_description->aligned_size);
-                }
 
                 handle_description->pins = 0;
             }
diff --git a/src/core/hle/service/nvdrv/core/nvmap.h b/src/core/hle/service/nvdrv/core/nvmap.h
index 7c3110d91..e9e9e8b5b 100644
--- a/src/core/hle/service/nvdrv/core/nvmap.h
+++ b/src/core/hle/service/nvdrv/core/nvmap.h
@@ -48,7 +48,7 @@ public:
         using Id = u32;
         Id id; //!< A globally unique identifier for this handle
 
-        s32 pins{};
+        s64 pins{};
         u32 pin_virt_address{};
         std::optional<typename std::list<std::shared_ptr<Handle>>::iterator> unmap_queue_entry{};
 
@@ -63,15 +63,14 @@ public:
 
         VAddr address{};   //!< The memory location in the guest's AS that this handle corresponds to,
                            //!< this can also be in the nvdrv tmem
-        DAddr d_address{}; //!< The memory location in the device's AS that this handle corresponds to,
-                           //!< this can also be in the nvdrv tmem
         bool is_shared_mem_mapped{}; //!< If this nvmap has been mapped with the MapSharedMem IPC
                                      //!< call
 
         u8 kind{};        //!< Used for memory compression
         bool allocated{}; //!< If the handle has been allocated with `Alloc`
 
-        u64 dma_map_addr{}; //! remove me after implementing pinning.
+        DAddr d_address{}; //!< The memory location in the device's AS that this handle corresponds to,
+                           //!< this can also be in the nvdrv tmem
 
         Handle(u64 size, Id id);
 
@@ -119,15 +118,7 @@ public:
 
     std::shared_ptr<Handle> GetHandle(Handle::Id handle);
 
-    VAddr GetHandleAddress(Handle::Id handle);
-
-    /**
-     * @brief Maps a handle into the SMMU address space
-     * @note This operation is refcounted, the number of calls to this must eventually match the
-     * number of calls to `UnpinHandle`
-     * @return The SMMU virtual address that the handle has been mapped to
-     */
-    u32 PinHandle(Handle::Id handle, size_t session_id);
+    DAddr GetHandleAddress(Handle::Id handle);
 
     /**
      * @brief Maps a handle into the SMMU address space
@@ -135,7 +126,7 @@ public:
      * number of calls to `UnpinHandle`
      * @return The SMMU virtual address that the handle has been mapped to
      */
-    NvResult AllocateHandle(Handle::Id handle, Handle::Flags pFlags, u32 pAlign, u8 pKind, u64 pAddress, size_t session_id);
+    DAddr PinHandle(Handle::Id handle, size_t session_id, bool low_area_pin);
 
     /**
      * @brief When this has been called an equal number of times to `PinHandle` for the supplied
diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
index 0ff41c6b2..f1404b9da 100644
--- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp
@@ -42,7 +42,7 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, android::PixelFormat form
                         u32 height, u32 stride, android::BufferTransformFlags transform,
                         const Common::Rectangle<int>& crop_rect,
                         std::array<Service::Nvidia::NvFence, 4>& fences, u32 num_fences) {
-    const VAddr addr = nvmap.GetHandleAddress(buffer_handle);
+    const DAddr addr = nvmap.GetHandleAddress(buffer_handle);
     LOG_TRACE(Service,
               "Drawing from address {:X} offset {:08X} Width {} Height {} Stride {} Format {}",
               addr, offset, width, height, stride, format);
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
index c92a7b2f6..8bc10eac2 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp
@@ -40,15 +40,15 @@ NvResult nvhost_as_gpu::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> i
         case 0x3:
             return WrapFixed(this, &nvhost_as_gpu::FreeSpace, input, output);
         case 0x5:
-            return WrapFixed(this, &nvhost_as_gpu::UnmapBuffer, input, output);
+            return WrapFixed(this, &nvhost_as_gpu::UnmapBuffer, input, output, fd);
         case 0x6:
-            return WrapFixed(this, &nvhost_as_gpu::MapBufferEx, input, output);
+            return WrapFixed(this, &nvhost_as_gpu::MapBufferEx, input, output, fd);
         case 0x8:
             return WrapFixed(this, &nvhost_as_gpu::GetVARegions1, input, output);
         case 0x9:
             return WrapFixed(this, &nvhost_as_gpu::AllocAsEx, input, output);
         case 0x14:
-            return WrapVariable(this, &nvhost_as_gpu::Remap, input, output);
+            return WrapVariable(this, &nvhost_as_gpu::Remap, input, output, fd);
         default:
             break;
         }
@@ -86,8 +86,15 @@ NvResult nvhost_as_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> i
     return NvResult::NotImplemented;
 }
 
-void nvhost_as_gpu::OnOpen(size_t session_id, DeviceFD fd) {}
-void nvhost_as_gpu::OnClose(DeviceFD fd) {}
+void nvhost_as_gpu::OnOpen(size_t session_id, DeviceFD fd) {
+    sessions[fd] = session_id;
+}
+void nvhost_as_gpu::OnClose(DeviceFD fd) {
+    auto it = sessions.find(fd);
+    if (it != sessions.end()) {
+        sessions.erase(it);
+    }
+}
 
 NvResult nvhost_as_gpu::AllocAsEx(IoctlAllocAsEx& params) {
     LOG_DEBUG(Service_NVDRV, "called, big_page_size=0x{:X}", params.big_page_size);
@@ -206,6 +213,8 @@ void nvhost_as_gpu::FreeMappingLocked(u64 offset) {
                        static_cast<u32>(aligned_size >> page_size_bits));
     }
 
+    nvmap.UnpinHandle(mapping->handle);
+
     // Sparse mappings shouldn't be fully unmapped, just returned to their sparse state
     // Only FreeSpace can unmap them fully
     if (mapping->sparse_alloc) {
@@ -259,7 +268,7 @@ NvResult nvhost_as_gpu::FreeSpace(IoctlFreeSpace& params) {
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::Remap(std::span<IoctlRemapEntry> entries) {
+NvResult nvhost_as_gpu::Remap(std::span<IoctlRemapEntry> entries, DeviceFD fd) {
     LOG_DEBUG(Service_NVDRV, "called, num_entries=0x{:X}", entries.size());
 
     if (!vm.initialised) {
@@ -293,19 +302,19 @@ NvResult nvhost_as_gpu::Remap(std::span<IoctlRemapEntry> entries) {
                 return NvResult::BadValue;
             }
 
-            VAddr cpu_address{static_cast<VAddr>(
-                handle->address +
-                (static_cast<u64>(entry.handle_offset_big_pages) << vm.big_page_size_bits))};
+            DAddr base = nvmap.PinHandle(entry.handle, sessions[fd], false);
+            DAddr device_address{static_cast<DAddr>(
+                base + (static_cast<u64>(entry.handle_offset_big_pages) << vm.big_page_size_bits))};
 
-            gmmu->Map(virtual_address, cpu_address, size, static_cast<Tegra::PTEKind>(entry.kind),
-                      use_big_pages);
+            gmmu->Map(virtual_address, device_address, size,
+                      static_cast<Tegra::PTEKind>(entry.kind), use_big_pages);
         }
     }
 
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::MapBufferEx(IoctlMapBufferEx& params) {
+NvResult nvhost_as_gpu::MapBufferEx(IoctlMapBufferEx& params, DeviceFD fd) {
     LOG_DEBUG(Service_NVDRV,
               "called, flags={:X}, nvmap_handle={:X}, buffer_offset={}, mapping_size={}"
               ", offset={}",
@@ -331,9 +340,9 @@ NvResult nvhost_as_gpu::MapBufferEx(IoctlMapBufferEx& params) {
             }
 
             u64 gpu_address{static_cast<u64>(params.offset + params.buffer_offset)};
-            VAddr cpu_address{mapping->ptr + params.buffer_offset};
+            VAddr device_address{mapping->ptr + params.buffer_offset};
 
-            gmmu->Map(gpu_address, cpu_address, params.mapping_size,
+            gmmu->Map(gpu_address, device_address, params.mapping_size,
                       static_cast<Tegra::PTEKind>(params.kind), mapping->big_page);
 
             return NvResult::Success;
@@ -349,7 +358,8 @@ NvResult nvhost_as_gpu::MapBufferEx(IoctlMapBufferEx& params) {
         return NvResult::BadValue;
     }
 
-    VAddr cpu_address{static_cast<VAddr>(handle->address + params.buffer_offset)};
+    DAddr device_address{static_cast<DAddr>(nvmap.PinHandle(params.handle, sessions[fd], false) +
+                                            params.buffer_offset)};
     u64 size{params.mapping_size ? params.mapping_size : handle->orig_size};
 
     bool big_page{[&]() {
@@ -373,15 +383,14 @@ NvResult nvhost_as_gpu::MapBufferEx(IoctlMapBufferEx& params) {
         }
 
         const bool use_big_pages = alloc->second.big_pages && big_page;
-        gmmu->Map(params.offset, cpu_address, size, static_cast<Tegra::PTEKind>(params.kind),
+        gmmu->Map(params.offset, device_address, size, static_cast<Tegra::PTEKind>(params.kind),
                   use_big_pages);
 
-        auto mapping{std::make_shared<Mapping>(cpu_address, params.offset, size, true,
-                                               use_big_pages, alloc->second.sparse)};
+        auto mapping{std::make_shared<Mapping>(params.handle, device_address, params.offset, size,
+                                               true, use_big_pages, alloc->second.sparse)};
         alloc->second.mappings.push_back(mapping);
         mapping_map[params.offset] = mapping;
     } else {
-
         auto& allocator{big_page ? *vm.big_page_allocator : *vm.small_page_allocator};
         u32 page_size{big_page ? vm.big_page_size : VM::YUZU_PAGESIZE};
         u32 page_size_bits{big_page ? vm.big_page_size_bits : VM::PAGE_SIZE_BITS};
@@ -394,18 +403,18 @@ NvResult nvhost_as_gpu::MapBufferEx(IoctlMapBufferEx& params) {
             return NvResult::InsufficientMemory;
         }
 
-        gmmu->Map(params.offset, cpu_address, Common::AlignUp(size, page_size),
+        gmmu->Map(params.offset, device_address, Common::AlignUp(size, page_size),
                   static_cast<Tegra::PTEKind>(params.kind), big_page);
 
-        auto mapping{
-            std::make_shared<Mapping>(cpu_address, params.offset, size, false, big_page, false)};
+        auto mapping{std::make_shared<Mapping>(params.handle, device_address, params.offset, size,
+                                               false, big_page, false)};
         mapping_map[params.offset] = mapping;
     }
 
     return NvResult::Success;
 }
 
-NvResult nvhost_as_gpu::UnmapBuffer(IoctlUnmapBuffer& params) {
+NvResult nvhost_as_gpu::UnmapBuffer(IoctlUnmapBuffer& params, DeviceFD fd) {
     LOG_DEBUG(Service_NVDRV, "called, offset=0x{:X}", params.offset);
 
     std::scoped_lock lock(mutex);
@@ -433,6 +442,8 @@ NvResult nvhost_as_gpu::UnmapBuffer(IoctlUnmapBuffer& params) {
             gmmu->Unmap(params.offset, mapping->size);
         }
 
+        nvmap.UnpinHandle(mapping->handle);
+
         mapping_map.erase(params.offset);
     } catch (const std::out_of_range&) {
         LOG_WARNING(Service_NVDRV, "Couldn't find region to unmap at 0x{:X}", params.offset);
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 0dd279f88..4b28f5078 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
+++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h
@@ -141,9 +141,9 @@ private:
 
     NvResult AllocAsEx(IoctlAllocAsEx& params);
     NvResult AllocateSpace(IoctlAllocSpace& params);
-    NvResult Remap(std::span<IoctlRemapEntry> params);
-    NvResult MapBufferEx(IoctlMapBufferEx& params);
-    NvResult UnmapBuffer(IoctlUnmapBuffer& params);
+    NvResult Remap(std::span<IoctlRemapEntry> params, DeviceFD fd);
+    NvResult MapBufferEx(IoctlMapBufferEx& params, DeviceFD fd);
+    NvResult UnmapBuffer(IoctlUnmapBuffer& params, DeviceFD fd);
     NvResult FreeSpace(IoctlFreeSpace& params);
     NvResult BindChannel(IoctlBindChannel& params);
 
@@ -159,16 +159,18 @@ private:
     NvCore::NvMap& nvmap;
 
     struct Mapping {
-        VAddr ptr;
+        NvCore::NvMap::Handle::Id handle;
+        DAddr 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_) {}
+        Mapping(NvCore::NvMap::Handle::Id handle_, DAddr ptr_, u64 offset_, u64 size_, bool fixed_,
+                bool big_page_, bool sparse_alloc_)
+            : handle(handle_), ptr(ptr_), offset(offset_), size(size_), fixed(fixed_),
+              big_page(big_page_), sparse_alloc(sparse_alloc_) {}
     };
 
     struct Allocation {
@@ -212,9 +214,7 @@ private:
         bool initialised{};
     } vm;
     std::shared_ptr<Tegra::MemoryManager> gmmu;
-
-    // s32 channel{};
-    // u32 big_page_size{VM::DEFAULT_BIG_PAGE_SIZE};
+    std::unordered_map<DeviceFD, size_t> sessions;
 };
 
 } // namespace Service::Nvidia::Devices
diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
index 9ab0ae4d8..78bc5f3c4 100644
--- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp
@@ -95,6 +95,9 @@ NvResult nvhost_nvdec_common::Submit(IoctlSubmit& params, std::span<u8> data, De
     offset += SliceVectors(data, fence_thresholds, params.fence_count, offset);
 
     auto& gpu = system.GPU();
+    //auto& device_memory = system.Host1x().MemoryManager();
+    auto* session = core.GetSession(sessions[fd]);
+
     if (gpu.UseNvdec()) {
         for (std::size_t i = 0; i < syncpt_increments.size(); i++) {
             const SyncptIncr& syncpt_incr = syncpt_increments[i];
@@ -106,7 +109,7 @@ NvResult nvhost_nvdec_common::Submit(IoctlSubmit& params, std::span<u8> data, De
         const auto object = nvmap.GetHandle(cmd_buffer.memory_id);
         ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;);
         Tegra::ChCommandHeaderList cmdlist(cmd_buffer.word_count);
-        system.ApplicationMemory().ReadBlock(object->address + cmd_buffer.offset, cmdlist.data(),
+        session->process->GetMemory().ReadBlock(object->address + cmd_buffer.offset, cmdlist.data(),
                                              cmdlist.size() * sizeof(u32));
         gpu.PushCommandBuffer(core.Host1xDeviceFile().fd_to_id[fd], cmdlist);
     }
@@ -136,7 +139,8 @@ NvResult nvhost_nvdec_common::GetWaitbase(IoctlGetWaitbase& params) {
 NvResult nvhost_nvdec_common::MapBuffer(IoctlMapBuffer& params, std::span<MapBufferEntry> entries, DeviceFD fd) {
     const size_t num_entries = std::min(params.num_entries, static_cast<u32>(entries.size()));
     for (size_t i = 0; i < num_entries; i++) {
-        entries[i].map_address = nvmap.PinHandle(entries[i].map_handle, sessions[fd]);
+        DAddr pin_address = nvmap.PinHandle(entries[i].map_handle, sessions[fd], true);
+        entries[i].map_address = static_cast<u32>(pin_address);
     }
 
     return NvResult::Success;
diff --git a/src/core/hle/service/nvdrv/devices/nvmap.cpp b/src/core/hle/service/nvdrv/devices/nvmap.cpp
index 2b107f009..7765ca1be 100644
--- a/src/core/hle/service/nvdrv/devices/nvmap.cpp
+++ b/src/core/hle/service/nvdrv/devices/nvmap.cpp
@@ -123,8 +123,8 @@ NvResult nvmap::IocAlloc(IocAllocParams& params, DeviceFD fd) {
         return NvResult::InsufficientMemory;
     }
 
-    const auto result = file.AllocateHandle(params.handle, params.flags, params.align, params.kind,
-                                            params.address, sessions[fd]);
+    const auto result =
+        handle_description->Alloc(params.flags, params.align, params.kind, params.address);
     if (result != NvResult::Success) {
         LOG_CRITICAL(Service_NVDRV, "Object failed to allocate, handle={:08X}", params.handle);
         return result;
diff --git a/src/core/hle/service/nvdrv/nvdrv_interface.cpp b/src/core/hle/service/nvdrv/nvdrv_interface.cpp
index 492ad849a..6e4825313 100644
--- a/src/core/hle/service/nvdrv/nvdrv_interface.cpp
+++ b/src/core/hle/service/nvdrv/nvdrv_interface.cpp
@@ -13,8 +13,6 @@
 #include "core/hle/service/nvdrv/nvdrv.h"
 #include "core/hle/service/nvdrv/nvdrv_interface.h"
 
-#pragma optimize("", off)
-
 namespace Service::Nvidia {
 
 void NVDRV::Open(HLERequestContext& ctx) {
@@ -173,8 +171,8 @@ void NVDRV::Initialize(HLERequestContext& ctx) {
     [[maybe_unused]] const auto transfer_memory_size = rp.Pop<u32>();
 
     auto& container = nvdrv->GetContainer();
-    auto process = ctx.GetObjectFromHandle(process_handle);
-    session_id = container.OpenSession(process->DynamicCast<Kernel::KProcess*>());
+    auto process = ctx.GetObjectFromHandle<Kernel::KProcess>(process_handle);
+    session_id = container.OpenSession(process.GetPointerUnsafe());
 
     is_initialized = true;
 }