NVDRV: Implement sessions and initial implementation of SMMU

src/core/hle/service/hle_ipc.h

@@ -19,6 +19,8 @@
 #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;
 
@@ -41,6 +43,8 @@ class KernelCore;
 class KHandleTable;
 class KProcess;
 class KServerSession;
+template <typename T>
+class KScopedAutoObject;
 class KThread;
 } // namespace Kernel
 
@@ -373,6 +377,10 @@ 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();
     }

src/core/hle/service/nvdrv/core/container.cpp

@@ -2,19 +2,30 @@
 // SPDX-FileCopyrightText: 2022 Skyline Team and Contributors
 // SPDX-License-Identifier: GPL-3.0-or-later
 
+#include <atomic>
+#include <deque>
+#include <mutex>
+
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/service/nvdrv/core/container.h"
 #include "core/hle/service/nvdrv/core/nvmap.h"
 #include "core/hle/service/nvdrv/core/syncpoint_manager.h"
+#include "core/memory.h"
 #include "video_core/host1x/host1x.h"
 
 namespace Service::Nvidia::NvCore {
 
 struct ContainerImpl {
     explicit ContainerImpl(Tegra::Host1x::Host1x& host1x_)
-        : file{host1x_}, manager{host1x_}, device_file_data{} {}
+        : host1x{host1x_}, file{host1x_}, manager{host1x_}, device_file_data{} {}
+    Tegra::Host1x::Host1x& host1x;
     NvMap file;
     SyncpointManager manager;
     Container::Host1xDeviceFileData device_file_data;
+    std::deque<Session> sessions;
+    size_t new_ids{};
+    std::deque<size_t> id_pool;
+    std::mutex session_guard;
 };
 
 Container::Container(Tegra::Host1x::Host1x& host1x_) {
@@ -23,6 +34,37 @@ Container::Container(Tegra::Host1x::Host1x& host1x_) {
 
 Container::~Container() = default;
 
+size_t Container::OpenSession(Kernel::KProcess* process) {
+    std::scoped_lock lk(impl->session_guard);
+    size_t new_id{};
+    auto* memory_interface = &process->GetMemory();
+    auto& smmu = impl->host1x.MemoryManager();
+    auto smmu_id = smmu.RegisterProcess(memory_interface);
+    if (!impl->id_pool.empty()) {
+        new_id = impl->id_pool.front();
+        impl->id_pool.pop_front();
+        impl->sessions[new_id] = Session{new_id, process, smmu_id};
+    } else {
+        impl->sessions.emplace_back(new_id, process, smmu_id);
+        new_id = impl->new_ids++;
+    }
+    LOG_CRITICAL(Debug, "Created Session {}", new_id);
+    return new_id;
+}
+
+void Container::CloseSession(size_t id) {
+    std::scoped_lock lk(impl->session_guard);
+    auto& smmu = impl->host1x.MemoryManager();
+    smmu.UnregisterProcess(impl->sessions[id].smmu_id);
+    impl->id_pool.emplace_front(id);
+    LOG_CRITICAL(Debug, "Closed Session {}", id);
+}
+
+Session* Container::GetSession(size_t id) {
+    std::atomic_thread_fence(std::memory_order_acquire);
+    return &impl->sessions[id];
+}
+
 NvMap& Container::GetNvMapFile() {
     return impl->file;
 }

src/core/hle/service/nvdrv/core/container.h

@@ -10,6 +10,10 @@
 
 #include "core/hle/service/nvdrv/nvdata.h"
 
+namespace Kernel {
+class KProcess;
+}
+
 namespace Tegra::Host1x {
 class Host1x;
 } // namespace Tegra::Host1x
@@ -21,11 +25,22 @@ class SyncpointManager;
 
 struct ContainerImpl;
 
+struct Session {
+    size_t id;
+    Kernel::KProcess* process;
+    size_t smmu_id;
+};
+
 class Container {
 public:
     explicit Container(Tegra::Host1x::Host1x& host1x);
     ~Container();
 
+    size_t OpenSession(Kernel::KProcess* process);
+    void CloseSession(size_t id);
+
+    Session* GetSession(size_t id);
+
     NvMap& GetNvMapFile();
 
     const NvMap& GetNvMapFile() const;

src/core/hle/service/nvdrv/core/nvmap.cpp

@@ -18,8 +18,6 @@ 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;
@@ -79,10 +77,11 @@ void NvMap::UnmapHandle(Handle& handle_description) {
     }
 
     // Free and unmap the handle from the SMMU
-    host1x.MemoryManager().Unmap(static_cast<GPUVAddr>(handle_description.pin_virt_address),
-                                 handle_description.aligned_size);
-    host1x.Allocator().Free(handle_description.pin_virt_address,
-                            static_cast<u32>(handle_description.aligned_size));
+    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;
 }
 
@@ -133,7 +132,32 @@ VAddr NvMap::GetHandleAddress(Handle::Id handle) {
     }
 }
 
-u32 NvMap::PinHandle(NvMap::Handle::Id handle) {
+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) {
     auto handle_description{GetHandle(handle)};
     if (!handle_description) [[unlikely]] {
         return 0;
@@ -157,11 +181,10 @@ u32 NvMap::PinHandle(NvMap::Handle::Id handle) {
         }
 
         // If not then allocate some space and map it
-        u32 address{};
-        auto& smmu_allocator = host1x.Allocator();
-        auto& smmu_memory_manager = host1x.MemoryManager();
-        while ((address = smmu_allocator.Allocate(
-                    static_cast<u32>(handle_description->aligned_size))) == 0) {
+        DAddr address{};
+        auto& smmu = host1x.MemoryManager();
+        while ((address = smmu.AllocatePinned(
+                    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()}) {
@@ -175,9 +198,9 @@ u32 NvMap::PinHandle(NvMap::Handle::Id handle) {
             }
         }
 
-        smmu_memory_manager.Map(static_cast<GPUVAddr>(address), handle_description->address,
-                                handle_description->aligned_size);
-        handle_description->pin_virt_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++;
@@ -236,6 +259,11 @@ std::optional<NvMap::FreeInfo> NvMap::FreeHandle(Handle::Id handle, bool interna
                     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;
             }

src/core/hle/service/nvdrv/core/nvmap.h

@@ -61,8 +61,10 @@ public:
         } flags{};
         static_assert(sizeof(Flags) == sizeof(u32));
 
-        u64 address{}; //!< The memory location in the guest's AS that this handle corresponds to,
-                       //!< this can also be in the nvdrv tmem
+        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
 
@@ -125,7 +127,15 @@ public:
      * number of calls to `UnpinHandle`
      * @return The SMMU virtual address that the handle has been mapped to
      */
-    u32 PinHandle(Handle::Id handle);
+    u32 PinHandle(Handle::Id handle, size_t session_id);
+
+    /**
+     * @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
+     */
+    NvResult AllocateHandle(Handle::Id handle, Handle::Flags pFlags, u32 pAlign, u8 pKind, u64 pAddress, size_t session_id);
 
     /**
      * @brief When this has been called an equal number of times to `PinHandle` for the supplied

src/core/hle/service/nvdrv/devices/nvdevice.h

@@ -62,7 +62,7 @@ public:
      * Called once a device is opened
      * @param fd The device fd
      */
-    virtual void OnOpen(DeviceFD fd) = 0;
+    virtual void OnOpen(size_t session_id, DeviceFD fd) = 0;
 
     /**
      * Called once a device is closed

src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp

@@ -35,7 +35,7 @@ NvResult nvdisp_disp0::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> in
     return NvResult::NotImplemented;
 }
 
-void nvdisp_disp0::OnOpen(DeviceFD fd) {}
+void nvdisp_disp0::OnOpen(size_t session_id, DeviceFD fd) {}
 void nvdisp_disp0::OnClose(DeviceFD fd) {}
 
 void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, android::PixelFormat format, u32 width,

src/core/hle/service/nvdrv/devices/nvdisp_disp0.h

@@ -32,7 +32,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 
     /// Performs a screen flip, drawing the buffer pointed to by the handle.

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp

@@ -86,7 +86,7 @@ NvResult nvhost_as_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> i
     return NvResult::NotImplemented;
 }
 
-void nvhost_as_gpu::OnOpen(DeviceFD fd) {}
+void nvhost_as_gpu::OnOpen(size_t session_id, DeviceFD fd) {}
 void nvhost_as_gpu::OnClose(DeviceFD fd) {}
 
 NvResult nvhost_as_gpu::AllocAsEx(IoctlAllocAsEx& params) {

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h

@@ -55,7 +55,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 
     Kernel::KEvent* QueryEvent(u32 event_id) override;

src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp

@@ -76,7 +76,7 @@ NvResult nvhost_ctrl::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> inp
     return NvResult::NotImplemented;
 }
 
-void nvhost_ctrl::OnOpen(DeviceFD fd) {}
+void nvhost_ctrl::OnOpen(size_t session_id, DeviceFD fd) {}
 
 void nvhost_ctrl::OnClose(DeviceFD fd) {}
 

src/core/hle/service/nvdrv/devices/nvhost_ctrl.h

@@ -32,7 +32,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 
     Kernel::KEvent* QueryEvent(u32 event_id) override;

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp

@@ -82,7 +82,7 @@ NvResult nvhost_ctrl_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8>
     return NvResult::NotImplemented;
 }
 
-void nvhost_ctrl_gpu::OnOpen(DeviceFD fd) {}
+void nvhost_ctrl_gpu::OnOpen(size_t session_id, DeviceFD fd) {}
 void nvhost_ctrl_gpu::OnClose(DeviceFD fd) {}
 
 NvResult nvhost_ctrl_gpu::GetCharacteristics1(IoctlCharacteristics& params) {

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h

@@ -28,7 +28,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 
     Kernel::KEvent* QueryEvent(u32 event_id) override;

src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp

@@ -120,7 +120,7 @@ NvResult nvhost_gpu::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> inpu
     return NvResult::NotImplemented;
 }
 
-void nvhost_gpu::OnOpen(DeviceFD fd) {}
+void nvhost_gpu::OnOpen(size_t session_id, DeviceFD fd) {}
 void nvhost_gpu::OnClose(DeviceFD fd) {}
 
 NvResult nvhost_gpu::SetNVMAPfd(IoctlSetNvmapFD& params) {

src/core/hle/service/nvdrv/devices/nvhost_gpu.h

@@ -47,7 +47,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 
     Kernel::KEvent* QueryEvent(u32 event_id) override;

src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp

@@ -35,7 +35,7 @@ NvResult nvhost_nvdec::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> in
         case 0x7:
             return WrapFixed(this, &nvhost_nvdec::SetSubmitTimeout, input, output);
         case 0x9:
-            return WrapFixedVariable(this, &nvhost_nvdec::MapBuffer, input, output);
+            return WrapFixedVariable(this, &nvhost_nvdec::MapBuffer, input, output, fd);
         case 0xa:
             return WrapFixedVariable(this, &nvhost_nvdec::UnmapBuffer, input, output);
         default:
@@ -68,9 +68,10 @@ NvResult nvhost_nvdec::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> in
     return NvResult::NotImplemented;
 }
 
-void nvhost_nvdec::OnOpen(DeviceFD fd) {
+void nvhost_nvdec::OnOpen(size_t session_id, DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream started");
     system.SetNVDECActive(true);
+    sessions[fd] = session_id;
 }
 
 void nvhost_nvdec::OnClose(DeviceFD fd) {
@@ -81,6 +82,10 @@ void nvhost_nvdec::OnClose(DeviceFD fd) {
         system.GPU().ClearCdmaInstance(iter->second);
     }
     system.SetNVDECActive(false);
+    auto it = sessions.find(fd);
+    if (it != sessions.end()) {
+        sessions.erase(it);
+    }
 }
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_nvdec.h

@@ -20,7 +20,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 };
 

src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp

@@ -133,10 +133,10 @@ NvResult nvhost_nvdec_common::GetWaitbase(IoctlGetWaitbase& params) {
     return NvResult::Success;
 }
 
-NvResult nvhost_nvdec_common::MapBuffer(IoctlMapBuffer& params, std::span<MapBufferEntry> entries) {
+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);
+        entries[i].map_address = nvmap.PinHandle(entries[i].map_handle, sessions[fd]);
     }
 
     return NvResult::Success;

src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h

@@ -5,6 +5,8 @@
 
 #include <deque>
 #include <vector>
+#include <unordered_map>
+
 #include "common/common_types.h"
 #include "common/swap.h"
 #include "core/hle/service/nvdrv/core/syncpoint_manager.h"
@@ -111,7 +113,7 @@ protected:
     NvResult Submit(IoctlSubmit& params, std::span<u8> input, DeviceFD fd);
     NvResult GetSyncpoint(IoctlGetSyncpoint& params);
     NvResult GetWaitbase(IoctlGetWaitbase& params);
-    NvResult MapBuffer(IoctlMapBuffer& params, std::span<MapBufferEntry> entries);
+    NvResult MapBuffer(IoctlMapBuffer& params, std::span<MapBufferEntry> entries, DeviceFD fd);
     NvResult UnmapBuffer(IoctlMapBuffer& params, std::span<MapBufferEntry> entries);
     NvResult SetSubmitTimeout(u32 timeout);
 
@@ -125,6 +127,7 @@ protected:
     NvCore::NvMap& nvmap;
     NvCore::ChannelType channel_type;
     std::array<u32, MaxSyncPoints> device_syncpoints{};
+    std::unordered_map<DeviceFD, size_t> sessions;
 };
 }; // namespace Devices
 } // namespace Service::Nvidia

src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp

@@ -44,7 +44,7 @@ NvResult nvhost_nvjpg::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> in
     return NvResult::NotImplemented;
 }
 
-void nvhost_nvjpg::OnOpen(DeviceFD fd) {}
+void nvhost_nvjpg::OnOpen(size_t session_id, DeviceFD fd) {}
 void nvhost_nvjpg::OnClose(DeviceFD fd) {}
 
 NvResult nvhost_nvjpg::SetNVMAPfd(IoctlSetNvmapFD& params) {

src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h

@@ -22,7 +22,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 
 private:

src/core/hle/service/nvdrv/devices/nvhost_vic.cpp

@@ -33,7 +33,7 @@ NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> inpu
         case 0x3:
             return WrapFixed(this, &nvhost_vic::GetWaitbase, input, output);
         case 0x9:
-            return WrapFixedVariable(this, &nvhost_vic::MapBuffer, input, output);
+            return WrapFixedVariable(this, &nvhost_vic::MapBuffer, input, output, fd);
         case 0xa:
             return WrapFixedVariable(this, &nvhost_vic::UnmapBuffer, input, output);
         default:
@@ -68,7 +68,9 @@ NvResult nvhost_vic::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> inpu
     return NvResult::NotImplemented;
 }
 
-void nvhost_vic::OnOpen(DeviceFD fd) {}
+void nvhost_vic::OnOpen(size_t session_id, DeviceFD fd) {
+        sessions[fd] = session_id;
+}
 
 void nvhost_vic::OnClose(DeviceFD fd) {
     auto& host1x_file = core.Host1xDeviceFile();
@@ -76,6 +78,10 @@ void nvhost_vic::OnClose(DeviceFD fd) {
     if (iter != host1x_file.fd_to_id.end()) {
         system.GPU().ClearCdmaInstance(iter->second);
     }
+    auto it = sessions.find(fd);
+    if (it != sessions.end()) {
+        sessions.erase(it);
+    }
 }
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_vic.h

@@ -19,7 +19,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 };
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvmap.cpp

@@ -36,9 +36,9 @@ NvResult nvmap::Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input,
         case 0x3:
             return WrapFixed(this, &nvmap::IocFromId, input, output);
         case 0x4:
-            return WrapFixed(this, &nvmap::IocAlloc, input, output);
+            return WrapFixed(this, &nvmap::IocAlloc, input, output, fd);
         case 0x5:
-            return WrapFixed(this, &nvmap::IocFree, input, output);
+            return WrapFixed(this, &nvmap::IocFree, input, output, fd);
         case 0x9:
             return WrapFixed(this, &nvmap::IocParam, input, output);
         case 0xe:
@@ -67,8 +67,15 @@ NvResult nvmap::Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, st
     return NvResult::NotImplemented;
 }
 
-void nvmap::OnOpen(DeviceFD fd) {}
-void nvmap::OnClose(DeviceFD fd) {}
+void nvmap::OnOpen(size_t session_id, DeviceFD fd) {
+    sessions[fd] = session_id;
+}
+void nvmap::OnClose(DeviceFD fd) {
+    auto it = sessions.find(fd);
+    if (it != sessions.end()) {
+        sessions.erase(it);
+    }
+}
 
 NvResult nvmap::IocCreate(IocCreateParams& params) {
     LOG_DEBUG(Service_NVDRV, "called, size=0x{:08X}", params.size);
@@ -87,7 +94,7 @@ NvResult nvmap::IocCreate(IocCreateParams& params) {
     return NvResult::Success;
 }
 
-NvResult nvmap::IocAlloc(IocAllocParams& params) {
+NvResult nvmap::IocAlloc(IocAllocParams& params, DeviceFD fd) {
     LOG_DEBUG(Service_NVDRV, "called, addr={:X}", params.address);
 
     if (!params.handle) {
@@ -116,15 +123,15 @@ NvResult nvmap::IocAlloc(IocAllocParams& params) {
         return NvResult::InsufficientMemory;
     }
 
-    const auto result =
-        handle_description->Alloc(params.flags, params.align, params.kind, params.address);
+    const auto result = file.AllocateHandle(params.handle, params.flags, params.align, params.kind,
+                                            params.address, sessions[fd]);
     if (result != NvResult::Success) {
         LOG_CRITICAL(Service_NVDRV, "Object failed to allocate, handle={:08X}", params.handle);
         return result;
     }
     bool is_out_io{};
-    ASSERT(system.ApplicationProcess()
-               ->GetPageTable()
+    auto process = container.GetSession(sessions[fd])->process;
+    ASSERT(process->GetPageTable()
                .LockForMapDeviceAddressSpace(&is_out_io, handle_description->address,
                                              handle_description->size,
                                              Kernel::KMemoryPermission::None, true, false)
@@ -224,7 +231,7 @@ NvResult nvmap::IocParam(IocParamParams& params) {
     return NvResult::Success;
 }
 
-NvResult nvmap::IocFree(IocFreeParams& params) {
+NvResult nvmap::IocFree(IocFreeParams& params, DeviceFD fd) {
     LOG_DEBUG(Service_NVDRV, "called");
 
     if (!params.handle) {
@@ -233,9 +240,9 @@ NvResult nvmap::IocFree(IocFreeParams& params) {
     }
 
     if (auto freeInfo{file.FreeHandle(params.handle, false)}) {
+        auto process = container.GetSession(sessions[fd])->process;
         if (freeInfo->can_unlock) {
-            ASSERT(system.ApplicationProcess()
-                       ->GetPageTable()
+            ASSERT(process->GetPageTable()
                        .UnlockForDeviceAddressSpace(freeInfo->address, freeInfo->size)
                        .IsSuccess());
         }

src/core/hle/service/nvdrv/devices/nvmap.h

@@ -33,7 +33,7 @@ public:
     NvResult Ioctl3(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output,
                     std::span<u8> inline_output) override;
 
-    void OnOpen(DeviceFD fd) override;
+    void OnOpen(size_t session_id, DeviceFD fd) override;
     void OnClose(DeviceFD fd) override;
 
     enum class HandleParameterType : u32_le {
@@ -100,11 +100,11 @@ public:
     static_assert(sizeof(IocGetIdParams) == 8, "IocGetIdParams has wrong size");
 
     NvResult IocCreate(IocCreateParams& params);
-    NvResult IocAlloc(IocAllocParams& params);
+    NvResult IocAlloc(IocAllocParams& params, DeviceFD fd);
     NvResult IocGetId(IocGetIdParams& params);
     NvResult IocFromId(IocFromIdParams& params);
     NvResult IocParam(IocParamParams& params);
-    NvResult IocFree(IocFreeParams& params);
+    NvResult IocFree(IocFreeParams& params, DeviceFD fd);
 
 private:
     /// Id to use for the next handle that is created.
@@ -115,6 +115,7 @@ private:
 
     NvCore::Container& container;
     NvCore::NvMap& file;
+    std::unordered_map<DeviceFD, size_t> sessions;
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/nvdrv.cpp

@@ -45,13 +45,22 @@ void EventInterface::FreeEvent(Kernel::KEvent* event) {
 void LoopProcess(Nvnflinger::Nvnflinger& nvnflinger, Core::System& system) {
     auto server_manager = std::make_unique<ServerManager>(system);
     auto module = std::make_shared<Module>(system);
-    server_manager->RegisterNamedService("nvdrv", std::make_shared<NVDRV>(system, module, "nvdrv"));
-    server_manager->RegisterNamedService("nvdrv:a",
-                                         std::make_shared<NVDRV>(system, module, "nvdrv:a"));
-    server_manager->RegisterNamedService("nvdrv:s",
-                                         std::make_shared<NVDRV>(system, module, "nvdrv:s"));
-    server_manager->RegisterNamedService("nvdrv:t",
-                                         std::make_shared<NVDRV>(system, module, "nvdrv:t"));
+    const auto NvdrvInterfaceFactoryForApplication = [&, module] {
+        return std::make_shared<NVDRV>(system, module, "nvdrv");
+    };
+    const auto NvdrvInterfaceFactoryForApplets = [&, module] {
+        return std::make_shared<NVDRV>(system, module, "nvdrv:a");
+    };
+    const auto NvdrvInterfaceFactoryForSysmodules = [&, module] {
+        return std::make_shared<NVDRV>(system, module, "nvdrv:a");
+    };
+    const auto NvdrvInterfaceFactory = [&, module] {
+        return std::make_shared<NVDRV>(system, module, "nvdrv:t");
+    };
+    server_manager->RegisterNamedService("nvdrv", NvdrvInterfaceFactoryForApplication);
+    server_manager->RegisterNamedService("nvdrv:a", NvdrvInterfaceFactoryForApplets);
+    server_manager->RegisterNamedService("nvdrv:s", NvdrvInterfaceFactoryForSysmodules);
+    server_manager->RegisterNamedService("nvdrv:t", NvdrvInterfaceFactory);
     server_manager->RegisterNamedService("nvmemp", std::make_shared<NVMEMP>(system));
     nvnflinger.SetNVDrvInstance(module);
     ServerManager::RunServer(std::move(server_manager));
@@ -113,7 +122,7 @@ NvResult Module::VerifyFD(DeviceFD fd) const {
     return NvResult::Success;
 }
 
-DeviceFD Module::Open(const std::string& device_name) {
+DeviceFD Module::Open(const std::string& device_name, size_t session_id) {
     auto it = builders.find(device_name);
     if (it == builders.end()) {
         LOG_ERROR(Service_NVDRV, "Trying to open unknown device {}", device_name);
@@ -124,7 +133,7 @@ DeviceFD Module::Open(const std::string& device_name) {
     auto& builder = it->second;
     auto device = builder(fd)->second;
 
-    device->OnOpen(fd);
+    device->OnOpen(session_id, fd);
 
     return fd;
 }

src/core/hle/service/nvdrv/nvdrv.h

@@ -77,7 +77,7 @@ public:
     NvResult VerifyFD(DeviceFD fd) const;
 
     /// Opens a device node and returns a file descriptor to it.
-    DeviceFD Open(const std::string& device_name);
+    DeviceFD Open(const std::string& device_name, size_t session_id);
 
     /// Sends an ioctl command to the specified file descriptor.
     NvResult Ioctl1(DeviceFD fd, Ioctl command, std::span<const u8> input, std::span<u8> output);
@@ -93,6 +93,10 @@ public:
 
     NvResult QueryEvent(DeviceFD fd, u32 event_id, Kernel::KEvent*& event);
 
+    NvCore::Container& GetContainer() {
+        return container;
+    }
+
 private:
     friend class EventInterface;
     friend class Service::Nvnflinger::Nvnflinger;

src/core/hle/service/nvdrv/nvdrv_interface.cpp

@@ -3,14 +3,18 @@
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #include "common/logging/log.h"
+#include "common/scope_exit.h"
 #include "core/core.h"
 #include "core/hle/kernel/k_event.h"
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/k_readable_event.h"
 #include "core/hle/service/ipc_helpers.h"
 #include "core/hle/service/nvdrv/nvdata.h"
 #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) {
@@ -37,7 +41,7 @@ void NVDRV::Open(HLERequestContext& ctx) {
         return;
     }
 
-    DeviceFD fd = nvdrv->Open(device_name);
+    DeviceFD fd = nvdrv->Open(device_name, session_id);
 
     rb.Push<DeviceFD>(fd);
     rb.PushEnum(fd != INVALID_NVDRV_FD ? NvResult::Success : NvResult::FileOperationFailed);
@@ -150,12 +154,29 @@ void NVDRV::Close(HLERequestContext& ctx) {
 
 void NVDRV::Initialize(HLERequestContext& ctx) {
     LOG_WARNING(Service_NVDRV, "(STUBBED) called");
+    IPC::ResponseBuilder rb{ctx, 3};
+    SCOPE_EXIT({
+        rb.Push(ResultSuccess);
+        rb.PushEnum(NvResult::Success);
+    });
 
-    is_initialized = true;
+    if (is_initialized) {
+        // No need to initialize again
+        return;
+    }
 
-    IPC::ResponseBuilder rb{ctx, 3};
-    rb.Push(ResultSuccess);
-    rb.PushEnum(NvResult::Success);
+    IPC::RequestParser rp{ctx};
+    const auto process_handle{ctx.GetCopyHandle(0)};
+    // The transfer memory is lent to nvdrv as a work buffer since nvdrv is
+    // unable to allocate as much memory on its own. For HLE it's unnecessary to handle it
+    [[maybe_unused]] const auto transfer_memory_handle{ctx.GetCopyHandle(1)};
+    [[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*>());
+
+    is_initialized = true;
 }
 
 void NVDRV::QueryEvent(HLERequestContext& ctx) {
@@ -242,6 +263,9 @@ NVDRV::NVDRV(Core::System& system_, std::shared_ptr<Module> nvdrv_, const char*
     RegisterHandlers(functions);
 }
 
-NVDRV::~NVDRV() = default;
+NVDRV::~NVDRV() {
+    auto& container = nvdrv->GetContainer();
+    container.CloseSession(session_id);
+}
 
 } // namespace Service::Nvidia

src/core/hle/service/nvdrv/nvdrv_interface.h

@@ -35,6 +35,7 @@ private:
 
     u64 pid{};
     bool is_initialized{};
+    size_t session_id{};
     Common::ScratchBuffer<u8> output_buffer;
     Common::ScratchBuffer<u8> inline_output_buffer;
 };

src/core/hle/service/nvnflinger/fb_share_buffer_manager.cpp

@@ -87,19 +87,19 @@ Result CreateNvMapHandle(u32* out_nv_map_handle, Nvidia::Devices::nvmap& nvmap,
     R_SUCCEED();
 }
 
-Result FreeNvMapHandle(Nvidia::Devices::nvmap& nvmap, u32 handle) {
+Result FreeNvMapHandle(Nvidia::Devices::nvmap& nvmap, u32 handle, Nvidia::DeviceFD nvmap_fd) {
     // Free the handle.
     Nvidia::Devices::nvmap::IocFreeParams free_params{
         .handle = handle,
     };
-    R_UNLESS(nvmap.IocFree(free_params) == Nvidia::NvResult::Success, VI::ResultOperationFailed);
+    R_UNLESS(nvmap.IocFree(free_params, nvmap_fd) == Nvidia::NvResult::Success, VI::ResultOperationFailed);
 
     // We succeeded.
     R_SUCCEED();
 }
 
 Result AllocNvMapHandle(Nvidia::Devices::nvmap& nvmap, u32 handle, Common::ProcessAddress buffer,
-                        u32 size) {
+                        u32 size, Nvidia::DeviceFD nvmap_fd) {
     // Assign the allocated memory to the handle.
     Nvidia::Devices::nvmap::IocAllocParams alloc_params{
         .handle = handle,
@@ -109,16 +109,15 @@ Result AllocNvMapHandle(Nvidia::Devices::nvmap& nvmap, u32 handle, Common::Proce
         .kind = 0,
         .address = GetInteger(buffer),
     };
-    R_UNLESS(nvmap.IocAlloc(alloc_params) == Nvidia::NvResult::Success, VI::ResultOperationFailed);
+    R_UNLESS(nvmap.IocAlloc(alloc_params, nvmap_fd) == Nvidia::NvResult::Success, VI::ResultOperationFailed);
 
     // We succeeded.
     R_SUCCEED();
 }
 
-Result AllocateHandleForBuffer(u32* out_handle, Nvidia::Module& nvdrv,
+Result AllocateHandleForBuffer(u32* out_handle, Nvidia::Module& nvdrv, Nvidia::DeviceFD nvmap_fd,
                                Common::ProcessAddress buffer, u32 size) {
     // Get the nvmap device.
-    auto nvmap_fd = nvdrv.Open("/dev/nvmap");
     auto nvmap = nvdrv.GetDevice<Nvidia::Devices::nvmap>(nvmap_fd);
     ASSERT(nvmap != nullptr);
 
@@ -127,11 +126,11 @@ Result AllocateHandleForBuffer(u32* out_handle, Nvidia::Module& nvdrv,
 
     // Ensure we maintain a clean state on failure.
     ON_RESULT_FAILURE {
-        ASSERT(R_SUCCEEDED(FreeNvMapHandle(*nvmap, *out_handle)));
+        ASSERT(R_SUCCEEDED(FreeNvMapHandle(*nvmap, *out_handle, nvmap_fd)));
     };
 
     // Assign the allocated memory to the handle.
-    R_RETURN(AllocNvMapHandle(*nvmap, *out_handle, buffer, size));
+    R_RETURN(AllocNvMapHandle(*nvmap, *out_handle, buffer, size, nvmap_fd));
 }
 
 constexpr auto SharedBufferBlockLinearFormat = android::PixelFormat::Rgba8888;
@@ -197,8 +196,12 @@ Result FbShareBufferManager::Initialize(u64* out_buffer_id, u64* out_layer_id, u
                                            std::addressof(m_buffer_page_group), m_system,
                                            SharedBufferSize));
 
+    auto& container = m_nvdrv->GetContainer();
+    m_session_id = container.OpenSession(m_system.ApplicationProcess());
+    m_nvmap_fd = m_nvdrv->Open("/dev/nvmap", m_session_id);
+
     // Create an nvmap handle for the buffer and assign the memory to it.
-    R_TRY(AllocateHandleForBuffer(std::addressof(m_buffer_nvmap_handle), *m_nvdrv, map_address,
+    R_TRY(AllocateHandleForBuffer(std::addressof(m_buffer_nvmap_handle), *m_nvdrv, m_nvmap_fd, map_address,
                                   SharedBufferSize));
 
     // Record the display id.

src/core/hle/service/nvnflinger/fb_share_buffer_manager.h

@@ -6,6 +6,7 @@
 #include "common/math_util.h"
 #include "core/hle/service/nvnflinger/nvnflinger.h"
 #include "core/hle/service/nvnflinger/ui/fence.h"
+#include "core/hle/service/nvdrv/nvdata.h"
 
 namespace Kernel {
 class KPageGroup;
@@ -53,13 +54,15 @@ private:
     u64 m_layer_id = 0;
     u32 m_buffer_nvmap_handle = 0;
     SharedMemoryPoolLayout m_pool_layout = {};
-
+    Nvidia::DeviceFD m_nvmap_fd = {};
+    size_t m_session_id = {};
     std::unique_ptr<Kernel::KPageGroup> m_buffer_page_group;
 
     std::mutex m_guard;
     Core::System& m_system;
     Nvnflinger& m_flinger;
     std::shared_ptr<Nvidia::Module> m_nvdrv;
+
 };
 
 } // namespace Service::Nvnflinger

src/core/hle/service/nvnflinger/nvnflinger.cpp

@@ -126,7 +126,7 @@ void Nvnflinger::ShutdownLayers() {
 
 void Nvnflinger::SetNVDrvInstance(std::shared_ptr<Nvidia::Module> instance) {
     nvdrv = std::move(instance);
-    disp_fd = nvdrv->Open("/dev/nvdisp_disp0");
+    disp_fd = nvdrv->Open("/dev/nvdisp_disp0", 0);
 }
 
 std::optional<u64> Nvnflinger::OpenDisplay(std::string_view name) {

src/video_core/CMakeLists.txt

@@ -71,6 +71,8 @@ add_library(video_core STATIC
     host1x/ffmpeg/ffmpeg.h
     host1x/control.cpp
     host1x/control.h
+    host1x/gpu_device_memory_manager.cpp
+    host1x/gpu_device_memory_manager.h
     host1x/host1x.cpp
     host1x/host1x.h
     host1x/nvdec.cpp

src/video_core/gpu.cpp

@@ -85,7 +85,7 @@ struct GPU::Impl {
     void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) {
         renderer = std::move(renderer_);
         rasterizer = renderer->ReadRasterizer();
-        host1x.MemoryManager().BindRasterizer(rasterizer);
+        host1x.MemoryManager().BindInterface(rasterizer);
     }
 
     /// Flush all current written commands into the host GPU for execution.

src/video_core/host1x/host1x.cpp

@@ -9,8 +9,7 @@ namespace Tegra {
 namespace Host1x {
 
 Host1x::Host1x(Core::System& system_)
-    : system{system_}, syncpoint_manager{}, memory_manager{system, 32, 12},
-      allocator{std::make_unique<Common::FlatAllocator<u32, 0, 32>>(1 << 12)} {}
+    : system{system_}, syncpoint_manager{}, memory_manager(system.DeviceMemory()) {}
 
 } // namespace Host1x
 

src/video_core/host1x/host1x.h

@@ -5,9 +5,8 @@
 
 #include "common/common_types.h"
 
-#include "common/address_space.h"
+#include "video_core/host1x/gpu_device_memory_manager.h"
 #include "video_core/host1x/syncpoint_manager.h"
-#include "video_core/memory_manager.h"
 
 namespace Core {
 class System;
@@ -29,27 +28,18 @@ public:
         return syncpoint_manager;
     }
 
-    Tegra::MemoryManager& MemoryManager() {
+    Tegra::MaxwellDeviceMemoryManager& MemoryManager() {
         return memory_manager;
     }
 
-    const Tegra::MemoryManager& MemoryManager() const {
+    const Tegra::MaxwellDeviceMemoryManager& MemoryManager() const {
         return memory_manager;
     }
 
-    Common::FlatAllocator<u32, 0, 32>& Allocator() {
-        return *allocator;
-    }
-
-    const Common::FlatAllocator<u32, 0, 32>& Allocator() const {
-        return *allocator;
-    }
-
 private:
     Core::System& system;
     SyncpointManager syncpoint_manager;
-    Tegra::MemoryManager memory_manager;
-    std::unique_ptr<Common::FlatAllocator<u32, 0, 32>> allocator;
+    Tegra::MaxwellDeviceMemoryManager memory_manager;
 };
 
 } // namespace Host1x