Android 229

cmif_serialization: enforce const for references

README.md

@@ -2,9 +2,11 @@
 |----|----|----|----|----|
 | [12461](https://github.com/yuzu-emu/yuzu-android//pull/12461) | [`4c08a0e6d`](https://github.com/yuzu-emu/yuzu-android//pull/12461/files) | Rework Nvdec and VIC to fix out-of-order videos, and speed up decoding. | [Kelebek1](https://github.com/Kelebek1/) | Yes |
 | [12749](https://github.com/yuzu-emu/yuzu-android//pull/12749) | [`aad4b0d6f`](https://github.com/yuzu-emu/yuzu-android//pull/12749/files) | general: workarounds for SMMU syncing issues | [liamwhite](https://github.com/liamwhite/) | Yes |
-| [12756](https://github.com/yuzu-emu/yuzu-android//pull/12756) | [`b19285d4f`](https://github.com/yuzu-emu/yuzu-android//pull/12756/files) | general: applet multiprocess | [liamwhite](https://github.com/liamwhite/) | Yes |
-| [12892](https://github.com/yuzu-emu/yuzu-android//pull/12892) | [`78f72b3bf`](https://github.com/yuzu-emu/yuzu-android//pull/12892/files) | cmif_serialization: enforce const for references | [liamwhite](https://github.com/liamwhite/) | Yes |
+| [12756](https://github.com/yuzu-emu/yuzu-android//pull/12756) | [`cd3de0848`](https://github.com/yuzu-emu/yuzu-android//pull/12756/files) | general: applet multiprocess | [liamwhite](https://github.com/liamwhite/) | Yes |
+| [12873](https://github.com/yuzu-emu/yuzu-android//pull/12873) | [`a8488952f`](https://github.com/yuzu-emu/yuzu-android//pull/12873/files) | GPU: Implement channel scheduling. | [FernandoS27](https://github.com/FernandoS27/) | Yes |
 | [12903](https://github.com/yuzu-emu/yuzu-android//pull/12903) | [`5be8121af`](https://github.com/yuzu-emu/yuzu-android//pull/12903/files) | shader_recompiler: use only ConstOffset for OpImageFetch | [liamwhite](https://github.com/liamwhite/) | Yes |
+| [12905](https://github.com/yuzu-emu/yuzu-android//pull/12905) | [`5eb5c9675`](https://github.com/yuzu-emu/yuzu-android//pull/12905/files) | nvnflinger: release buffers before presentation sleep | [liamwhite](https://github.com/liamwhite/) | Yes |
+| [12915](https://github.com/yuzu-emu/yuzu-android//pull/12915) | [`504abbd6e`](https://github.com/yuzu-emu/yuzu-android//pull/12915/files) | dmnt: cheats: Update cheat vm to latest version | [german77](https://github.com/german77/) | Yes |
 
 
 End of merge log. You can find the original README.md below the break.

src/core/hle/service/am/library_applet_creator.cpp

@@ -118,23 +118,17 @@ std::shared_ptr<ILibraryAppletAccessor> CreateGuestApplet(Core::System& system,
     switch (mode) {
     case LibraryAppletMode::AllForeground:
     case LibraryAppletMode::NoUi:
-        applet->focus_state = FocusState::InFocus;
-        applet->hid_registration.EnableAppletToGetInput(true);
-        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::ChangeIntoForeground);
-        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::FocusStateChanged);
-        break;
-    case LibraryAppletMode::AllForegroundInitiallyHidden:
-        applet->system_buffer_manager.SetWindowVisibility(false);
-        applet->focus_state = FocusState::NotInFocus;
-        applet->hid_registration.EnableAppletToGetInput(false);
-        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::FocusStateChanged);
-        break;
     case LibraryAppletMode::PartialForeground:
     case LibraryAppletMode::PartialForegroundIndirectDisplay:
-    default:
-        applet->focus_state = FocusState::Background;
         applet->hid_registration.EnableAppletToGetInput(true);
-        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::FocusStateChanged);
+        applet->focus_state = FocusState::InFocus;
+        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::ChangeIntoForeground);
+        break;
+    case LibraryAppletMode::AllForegroundInitiallyHidden:
+        applet->hid_registration.EnableAppletToGetInput(false);
+        applet->focus_state = FocusState::NotInFocus;
+        applet->system_buffer_manager.SetWindowVisibility(false);
+        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::ChangeIntoBackground);
         break;
     }
 

src/core/hle/service/am/system_buffer_manager.cpp

@@ -38,7 +38,8 @@ bool SystemBufferManager::Initialize(Nvnflinger::Nvnflinger* nvnflinger, Kernel:
     }
 
     Nvnflinger::LayerBlending blending = Nvnflinger::LayerBlending::None;
-    if (mode == LibraryAppletMode::PartialForeground) {
+    if (mode == LibraryAppletMode::PartialForeground ||
+        mode == LibraryAppletMode::PartialForegroundIndirectDisplay) {
         blending = Nvnflinger::LayerBlending::Coverage;
     }
 

src/core/hle/service/am/window_controller.cpp

@@ -62,12 +62,12 @@ void IWindowController::SetAppletWindowVisibility(HLERequestContext& ctx) {
     applet->hid_registration.EnableAppletToGetInput(visible);
 
     if (visible) {
-        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::ChangeIntoForeground);
         applet->focus_state = FocusState::InFocus;
+        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::ChangeIntoForeground);
     } else {
         applet->focus_state = FocusState::NotInFocus;
+        applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::ChangeIntoBackground);
     }
-    applet->message_queue.PushMessage(AppletMessageQueue::AppletMessage::FocusStateChanged);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(ResultSuccess);

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

@@ -7,7 +7,6 @@
 #include "core/hle/service/nvdrv/devices/nvdisp_disp0.h"
 #include "core/hle/service/nvnflinger/buffer_item.h"
 #include "core/hle/service/nvnflinger/buffer_item_consumer.h"
-#include "core/hle/service/nvnflinger/buffer_queue_producer.h"
 #include "core/hle/service/nvnflinger/hardware_composer.h"
 #include "core/hle/service/nvnflinger/hwc_layer.h"
 #include "core/hle/service/nvnflinger/ui/graphic_buffer.h"
@@ -46,31 +45,9 @@ HardwareComposer::HardwareComposer() = default;
 HardwareComposer::~HardwareComposer() = default;
 
 u32 HardwareComposer::ComposeLocked(f32* out_speed_scale, VI::Display& display,
-                                    Nvidia::Devices::nvdisp_disp0& nvdisp, u32 frame_advance) {
+                                    Nvidia::Devices::nvdisp_disp0& nvdisp) {
     boost::container::small_vector<HwcLayer, 2> composition_stack;
 
-    m_frame_number += frame_advance;
-
-    // Release any necessary framebuffers.
-    for (auto& [layer_id, framebuffer] : m_framebuffers) {
-        if (framebuffer.release_frame_number > m_frame_number) {
-            // Not yet ready to release this framebuffer.
-            continue;
-        }
-
-        if (!framebuffer.is_acquired) {
-            // Already released.
-            continue;
-        }
-
-        if (auto* layer = display.FindLayer(layer_id); layer != nullptr) {
-            // TODO: support release fence
-            // This is needed to prevent screen tearing
-            layer->GetConsumer().ReleaseBuffer(framebuffer.item, android::Fence::NoFence());
-            framebuffer.is_acquired = false;
-        }
-    }
-
     // Set default speed limit to 100%.
     *out_speed_scale = 1.0f;
 
@@ -143,7 +120,30 @@ u32 HardwareComposer::ComposeLocked(f32* out_speed_scale, VI::Display& display,
     MicroProfileFlip();
 
     // Advance by at least one frame.
-    return swap_interval.value_or(1);
+    const u32 frame_advance = swap_interval.value_or(1);
+    m_frame_number += frame_advance;
+
+    // Release any necessary framebuffers.
+    for (auto& [layer_id, framebuffer] : m_framebuffers) {
+        if (framebuffer.release_frame_number > m_frame_number) {
+            // Not yet ready to release this framebuffer.
+            continue;
+        }
+
+        if (!framebuffer.is_acquired) {
+            // Already released.
+            continue;
+        }
+
+        if (auto* layer = display.FindLayer(layer_id); layer != nullptr) {
+            // TODO: support release fence
+            // This is needed to prevent screen tearing
+            layer->GetConsumer().ReleaseBuffer(framebuffer.item, android::Fence::NoFence());
+            framebuffer.is_acquired = false;
+        }
+    }
+
+    return frame_advance;
 }
 
 void HardwareComposer::RemoveLayerLocked(VI::Display& display, LayerId layer_id) {

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

@@ -27,7 +27,7 @@ public:
     ~HardwareComposer();
 
     u32 ComposeLocked(f32* out_speed_scale, VI::Display& display,
-                      Nvidia::Devices::nvdisp_disp0& nvdisp, u32 frame_advance);
+                      Nvidia::Devices::nvdisp_disp0& nvdisp);
     void RemoveLayerLocked(VI::Display& display, LayerId layer_id);
 
 private:

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

@@ -292,8 +292,7 @@ void Nvnflinger::Compose() {
         auto nvdisp = nvdrv->GetDevice<Nvidia::Devices::nvdisp_disp0>(disp_fd);
         ASSERT(nvdisp);
 
-        swap_interval = display.GetComposer().ComposeLocked(&compose_speed_scale, display, *nvdisp,
-                                                            swap_interval);
+        swap_interval = display.GetComposer().ComposeLocked(&compose_speed_scale, display, *nvdisp);
     }
 }
 

src/core/memory/cheat_engine.cpp

@@ -9,6 +9,7 @@
 #include "core/core_timing.h"
 #include "core/hle/kernel/k_page_table.h"
 #include "core/hle/kernel/k_process.h"
+#include "core/hle/kernel/k_process_page_table.h"
 #include "core/hle/service/hid/hid_server.h"
 #include "core/hle/service/sm/sm.h"
 #include "core/memory.h"
@@ -85,8 +86,12 @@ VAddr StandardVmCallbacks::SanitizeAddress(VAddr in) const {
     if ((in < metadata.main_nso_extents.base ||
          in >= metadata.main_nso_extents.base + metadata.main_nso_extents.size) &&
         (in < metadata.heap_extents.base ||
-         in >= metadata.heap_extents.base + metadata.heap_extents.size)) {
-        LOG_ERROR(CheatEngine,
+         in >= metadata.heap_extents.base + metadata.heap_extents.size) &&
+        (in < metadata.alias_extents.base ||
+         in >= metadata.heap_extents.base + metadata.alias_extents.size) &&
+        (in < metadata.aslr_extents.base ||
+         in >= metadata.heap_extents.base + metadata.aslr_extents.size)) {
+        LOG_DEBUG(CheatEngine,
                   "Cheat attempting to access memory at invalid address={:016X}, if this "
                   "persists, "
                   "the cheat may be incorrect. However, this may be normal early in execution if "
@@ -211,16 +216,14 @@ void CheatEngine::Initialize() {
         .base = GetInteger(page_table.GetHeapRegionStart()),
         .size = page_table.GetHeapRegionSize(),
     };
-
-    metadata.address_space_extents = {
-        .base = GetInteger(page_table.GetAddressSpaceStart()),
-        .size = page_table.GetAddressSpaceSize(),
-    };
-
-    metadata.alias_extents = {
+    metadata.aslr_extents = {
         .base = GetInteger(page_table.GetAliasCodeRegionStart()),
         .size = page_table.GetAliasCodeRegionSize(),
     };
+    metadata.alias_extents = {
+        .base = GetInteger(page_table.GetAliasRegionStart()),
+        .size = page_table.GetAliasRegionSize(),
+    };
 
     is_pending_reload.exchange(true);
 }

src/core/memory/cheat_engine.h

@@ -37,7 +37,7 @@ private:
     VAddr SanitizeAddress(VAddr address) const;
 
     const CheatProcessMetadata& metadata;
-    System& system;
+    Core::System& system;
 };
 
 // Intermediary class that parses a text file or other disk format for storing cheats into a

src/core/memory/dmnt_cheat_types.h

@@ -18,7 +18,7 @@ struct CheatProcessMetadata {
     MemoryRegionExtents main_nso_extents{};
     MemoryRegionExtents heap_extents{};
     MemoryRegionExtents alias_extents{};
-    MemoryRegionExtents address_space_extents{};
+    MemoryRegionExtents aslr_extents{};
     std::array<u8, 0x20> main_nso_build_id{};
 };
 

src/core/memory/dmnt_cheat_vm.cpp

@@ -322,8 +322,9 @@ bool DmntCheatVm::DecodeNextOpcode(CheatVmOpcode& out) {
     } break;
     case CheatVmOpcodeType::EndConditionalBlock: {
         // 20000000
-        // There's actually nothing left to process here!
-        opcode.opcode = EndConditionalOpcode{};
+        opcode.opcode = EndConditionalOpcode{
+            .is_else = ((first_dword >> 24) & 0xf) == 1,
+        };
     } break;
     case CheatVmOpcodeType::ControlLoop: {
         // 300R0000 VVVVVVVV
@@ -555,6 +556,18 @@ bool DmntCheatVm::DecodeNextOpcode(CheatVmOpcode& out) {
             .idx = first_dword & 0xF,
         };
     } break;
+    case CheatVmOpcodeType::PauseProcess: {
+        /* FF0????? */
+        /* FF0 = opcode 0xFF0 */
+        /* Pauses the current process. */
+        opcode.opcode = PauseProcessOpcode{};
+    } break;
+    case CheatVmOpcodeType::ResumeProcess: {
+        /* FF0????? */
+        /* FF0 = opcode 0xFF0 */
+        /* Pauses the current process. */
+        opcode.opcode = ResumeProcessOpcode{};
+    } break;
     case CheatVmOpcodeType::DebugLog: {
         // FFFTIX##
         // FFFTI0Ma aaaaaaaa
@@ -621,7 +634,7 @@ bool DmntCheatVm::DecodeNextOpcode(CheatVmOpcode& out) {
     return valid;
 }
 
-void DmntCheatVm::SkipConditionalBlock() {
+void DmntCheatVm::SkipConditionalBlock(bool is_if) {
     if (condition_depth > 0) {
         // We want to continue until we're out of the current block.
         const std::size_t desired_depth = condition_depth - 1;
@@ -637,8 +650,12 @@ void DmntCheatVm::SkipConditionalBlock() {
             // We also support nesting of conditional blocks, and Gateway does not.
             if (skip_opcode.begin_conditional_block) {
                 condition_depth++;
-            } else if (std::holds_alternative<EndConditionalOpcode>(skip_opcode.opcode)) {
-                condition_depth--;
+            } else if (auto end_cond = std::get_if<EndConditionalOpcode>(&skip_opcode.opcode)) {
+                if (!end_cond->is_else) {
+                    condition_depth--;
+                } else if (is_if && condition_depth - 1 == desired_depth) {
+                    break;
+                }
             }
         }
     } else {
@@ -675,6 +692,10 @@ u64 DmntCheatVm::GetCheatProcessAddress(const CheatProcessMetadata& metadata,
         return metadata.main_nso_extents.base + rel_address;
     case MemoryAccessType::Heap:
         return metadata.heap_extents.base + rel_address;
+    case MemoryAccessType::Alias:
+        return metadata.alias_extents.base + rel_address;
+    case MemoryAccessType::Aslr:
+        return metadata.aslr_extents.base + rel_address;
     }
 }
 
@@ -682,7 +703,6 @@ void DmntCheatVm::ResetState() {
     registers.fill(0);
     saved_values.fill(0);
     loop_tops.fill(0);
-    static_registers.fill(0);
     instruction_ptr = 0;
     condition_depth = 0;
     decode_success = true;
@@ -794,13 +814,18 @@ void DmntCheatVm::Execute(const CheatProcessMetadata& metadata) {
             }
             // Skip conditional block if condition not met.
             if (!cond_met) {
-                SkipConditionalBlock();
+                SkipConditionalBlock(true);
             }
-        } else if (std::holds_alternative<EndConditionalOpcode>(cur_opcode.opcode)) {
-            // Decrement the condition depth.
-            // We will assume, graciously, that mismatched conditional block ends are a nop.
-            if (condition_depth > 0) {
-                condition_depth--;
+        } else if (auto end_cond = std::get_if<EndConditionalOpcode>(&cur_opcode.opcode)) {
+            if (end_cond->is_else) {
+                /* Skip to the end of the conditional block. */
+                this->SkipConditionalBlock(false);
+            } else {
+                /* Decrement the condition depth. */
+                /* We will assume, graciously, that mismatched conditional block ends are a nop. */
+                if (condition_depth > 0) {
+                    condition_depth--;
+                }
             }
         } else if (auto ctrl_loop = std::get_if<ControlLoopOpcode>(&cur_opcode.opcode)) {
             if (ctrl_loop->start_loop) {
@@ -908,7 +933,7 @@ void DmntCheatVm::Execute(const CheatProcessMetadata& metadata) {
             // Check for keypress.
             if ((begin_keypress_cond->key_mask & kDown) != begin_keypress_cond->key_mask) {
                 // Keys not pressed. Skip conditional block.
-                SkipConditionalBlock();
+                SkipConditionalBlock(true);
             }
         } else if (auto perform_math_reg =
                        std::get_if<PerformArithmeticRegisterOpcode>(&cur_opcode.opcode)) {
@@ -1116,7 +1141,7 @@ void DmntCheatVm::Execute(const CheatProcessMetadata& metadata) {
 
             // Skip conditional block if condition not met.
             if (!cond_met) {
-                SkipConditionalBlock();
+                SkipConditionalBlock(true);
             }
         } else if (auto save_restore_reg =
                        std::get_if<SaveRestoreRegisterOpcode>(&cur_opcode.opcode)) {
@@ -1178,6 +1203,10 @@ void DmntCheatVm::Execute(const CheatProcessMetadata& metadata) {
                 // Store a register to a static register.
                 static_registers[rw_static_reg->static_idx] = registers[rw_static_reg->idx];
             }
+        } else if (std::holds_alternative<PauseProcessOpcode>(cur_opcode.opcode)) {
+            // TODO: Pause cheat process
+        } else if (std::holds_alternative<ResumeProcessOpcode>(cur_opcode.opcode)) {
+            // TODO: Resume cheat process
         } else if (auto debug_log = std::get_if<DebugLogOpcode>(&cur_opcode.opcode)) {
             // Read value from memory.
             u64 log_value = 0;

src/core/memory/dmnt_cheat_vm.h

@@ -42,12 +42,16 @@ enum class CheatVmOpcodeType : u32 {
     DoubleExtendedWidth = 0xF0,
 
     // Double-extended width opcodes.
+    PauseProcess = 0xFF0,
+    ResumeProcess = 0xFF1,
     DebugLog = 0xFFF,
 };
 
 enum class MemoryAccessType : u32 {
     MainNso = 0,
     Heap = 1,
+    Alias = 2,
+    Aslr = 3,
 };
 
 enum class ConditionalComparisonType : u32 {
@@ -131,7 +135,9 @@ struct BeginConditionalOpcode {
     VmInt value{};
 };
 
-struct EndConditionalOpcode {};
+struct EndConditionalOpcode {
+    bool is_else;
+};
 
 struct ControlLoopOpcode {
     bool start_loop{};
@@ -222,6 +228,10 @@ struct ReadWriteStaticRegisterOpcode {
     u32 idx{};
 };
 
+struct PauseProcessOpcode {};
+
+struct ResumeProcessOpcode {};
+
 struct DebugLogOpcode {
     u32 bit_width{};
     u32 log_id{};
@@ -244,8 +254,8 @@ struct CheatVmOpcode {
                  PerformArithmeticStaticOpcode, BeginKeypressConditionalOpcode,
                  PerformArithmeticRegisterOpcode, StoreRegisterToAddressOpcode,
                  BeginRegisterConditionalOpcode, SaveRestoreRegisterOpcode,
-                 SaveRestoreRegisterMaskOpcode, ReadWriteStaticRegisterOpcode, DebugLogOpcode,
-                 UnrecognizedInstruction>
+                 SaveRestoreRegisterMaskOpcode, ReadWriteStaticRegisterOpcode, PauseProcessOpcode,
+                 ResumeProcessOpcode, DebugLogOpcode, UnrecognizedInstruction>
         opcode{};
 };
 
@@ -296,7 +306,7 @@ private:
     std::array<std::size_t, NumRegisters> loop_tops{};
 
     bool DecodeNextOpcode(CheatVmOpcode& out);
-    void SkipConditionalBlock();
+    void SkipConditionalBlock(bool is_if);
     void ResetState();
 
     // For implementing the DebugLog opcode.

src/video_core/control/scheduler.cpp

@@ -13,20 +13,101 @@ Scheduler::Scheduler(GPU& gpu_) : gpu{gpu_} {}
 
 Scheduler::~Scheduler() = default;
 
+void Scheduler::Init() {
+    master_control = Common::Fiber::ThreadToFiber();
+}
+
+void Scheduler::Resume() {
+    bool nothing_pending;
+    do {
+        nothing_pending = true;
+        current_fifo = nullptr;
+        {
+            std::unique_lock lk(scheduling_guard);
+            size_t num_iters = gpfifos.size();
+            for (size_t i = 0; i < num_iters; i++) {
+                size_t current_id = (current_fifo_rotation_id + i) % gpfifos.size();
+                auto& fifo = gpfifos[current_id];
+                if (!fifo.is_active) {
+                    continue;
+                }
+                std::scoped_lock lk2(fifo.guard);
+                if (!fifo.pending_work.empty() || fifo.working.load(std::memory_order_acquire)) {
+                    current_fifo = &fifo;
+                    current_fifo_rotation_id = current_id;
+                    nothing_pending = false;
+                    break;
+                }
+            }
+        }
+        if (current_fifo) {
+            Common::Fiber::YieldTo(master_control, *current_fifo->context);
+            current_fifo = nullptr;
+        }
+    } while (!nothing_pending);
+}
+
+void Scheduler::Yield() {
+    ASSERT(current_fifo != nullptr);
+    Common::Fiber::YieldTo(current_fifo->context, *master_control);
+}
+
 void Scheduler::Push(s32 channel, CommandList&& entries) {
     std::unique_lock lk(scheduling_guard);
-    auto it = channels.find(channel);
-    ASSERT(it != channels.end());
-    auto channel_state = it->second;
-    gpu.BindChannel(channel_state->bind_id);
-    channel_state->dma_pusher->Push(std::move(entries));
-    channel_state->dma_pusher->DispatchCalls();
+    auto it = channel_gpfifo_ids.find(channel);
+    ASSERT(it != channel_gpfifo_ids.end());
+    auto gpfifo_id = it->second;
+    auto& fifo = gpfifos[gpfifo_id];
+    {
+        std::scoped_lock lk2(fifo.guard);
+        fifo.pending_work.emplace_back(std::move(entries));
+    }
+}
+
+void Scheduler::ChannelLoop(size_t gpfifo_id, s32 channel_id) {
+    gpu.BindChannel(channel_id);
+    auto& fifo = gpfifos[gpfifo_id];
+    while (true) {
+        auto* channel_state = channels[channel_id].get();
+        fifo.guard.lock();
+        while (!fifo.pending_work.empty()) {
+            {
+
+                fifo.working.store(true, std::memory_order_release);
+                CommandList&& entries = std::move(fifo.pending_work.front());
+                channel_state->dma_pusher->Push(std::move(entries));
+                fifo.pending_work.pop_front();
+            }
+            fifo.guard.unlock();
+            channel_state->dma_pusher->DispatchCalls();
+            fifo.guard.lock();
+        }
+        fifo.working.store(false, std::memory_order_relaxed);
+        fifo.guard.unlock();
+        Common::Fiber::YieldTo(fifo.context, *master_control);
+        gpu.BindChannel(channel_id);
+    }
 }
 
 void Scheduler::DeclareChannel(std::shared_ptr<ChannelState> new_channel) {
     s32 channel = new_channel->bind_id;
     std::unique_lock lk(scheduling_guard);
     channels.emplace(channel, new_channel);
+    size_t new_fifo_id;
+    if (!free_fifos.empty()) {
+        new_fifo_id = free_fifos.front();
+        free_fifos.pop_front();
+    } else {
+        new_fifo_id = gpfifos.size();
+        gpfifos.emplace_back();
+    }
+    auto& new_fifo = gpfifos[new_fifo_id];
+    channel_gpfifo_ids[channel] = new_fifo_id;
+    new_fifo.is_active = true;
+    new_fifo.bind_id = channel;
+    new_fifo.pending_work.clear();
+    std::function<void()> callback = std::bind(&Scheduler::ChannelLoop, this, new_fifo_id, channel);
+    new_fifo.context = std::make_shared<Common::Fiber>(std::move(callback));
 }
 
 } // namespace Tegra::Control

src/video_core/control/scheduler.h

@@ -3,10 +3,13 @@
 
 #pragma once
 
+#include <atomic>
+#include <deque>
 #include <memory>
 #include <mutex>
 #include <unordered_map>
 
+#include "common/fiber.h"
 #include "video_core/dma_pusher.h"
 
 namespace Tegra {
@@ -22,14 +25,36 @@ public:
     explicit Scheduler(GPU& gpu_);
     ~Scheduler();
 
+    void Init();
+
+    void Resume();
+
+    void Yield();
+
     void Push(s32 channel, CommandList&& entries);
 
     void DeclareChannel(std::shared_ptr<ChannelState> new_channel);
 
 private:
+    void ChannelLoop(size_t gpfifo_id, s32 channel_id);
+
     std::unordered_map<s32, std::shared_ptr<ChannelState>> channels;
+    std::unordered_map<s32, size_t> channel_gpfifo_ids;
     std::mutex scheduling_guard;
+    std::shared_ptr<Common::Fiber> master_control;
+    struct GPFifoContext {
+        bool is_active;
+        std::shared_ptr<Common::Fiber> context;
+        std::deque<CommandList> pending_work;
+        std::atomic<bool> working{};
+        std::mutex guard;
+        s32 bind_id;
+    };
+    std::deque<GPFifoContext> gpfifos;
+    std::deque<size_t> free_fifos;
     GPU& gpu;
+    size_t current_fifo_rotation_id{};
+    GPFifoContext* current_fifo{};
 };
 
 } // namespace Control

src/video_core/engines/puller.cpp

@@ -6,6 +6,7 @@
 #include "common/settings.h"
 #include "core/core.h"
 #include "video_core/control/channel_state.h"
+#include "video_core/control/scheduler.h"
 #include "video_core/dma_pusher.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/kepler_compute.h"
@@ -14,6 +15,8 @@
 #include "video_core/engines/maxwell_dma.h"
 #include "video_core/engines/puller.h"
 #include "video_core/gpu.h"
+#include "video_core/host1x/host1x.h"
+#include "video_core/host1x/syncpoint_manager.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
 
@@ -60,11 +63,14 @@ void Puller::ProcessBindMethod(const MethodCall& method_call) {
 }
 
 void Puller::ProcessFenceActionMethod() {
+    auto& syncpoint_manager = gpu.Host1x().GetSyncpointManager();
     switch (regs.fence_action.op) {
     case Puller::FenceOperation::Acquire:
-        // UNIMPLEMENTED_MSG("Channel Scheduling pending.");
-        // WaitFence(regs.fence_action.syncpoint_id, regs.fence_value);
-        rasterizer->ReleaseFences();
+        while (regs.fence_value >
+               syncpoint_manager.GetGuestSyncpointValue(regs.fence_action.syncpoint_id)) {
+            rasterizer->ReleaseFences();
+            gpu.Scheduler().Yield();
+        }
         break;
     case Puller::FenceOperation::Increment:
         rasterizer->SignalSyncPoint(regs.fence_action.syncpoint_id);

src/video_core/gpu.cpp

@@ -387,6 +387,14 @@ std::shared_ptr<Control::ChannelState> GPU::AllocateChannel() {
     return impl->AllocateChannel();
 }
 
+Tegra::Control::Scheduler& GPU::Scheduler() {
+    return *impl->scheduler;
+}
+
+const Tegra::Control::Scheduler& GPU::Scheduler() const {
+    return *impl->scheduler;
+}
+
 void GPU::InitChannel(Control::ChannelState& to_init) {
     impl->InitChannel(to_init);
 }

src/video_core/gpu.h

@@ -124,7 +124,8 @@ class KeplerCompute;
 
 namespace Control {
 struct ChannelState;
-}
+class Scheduler;
+} // namespace Control
 
 namespace Host1x {
 class Host1x;
@@ -204,6 +205,12 @@ public:
     /// Returns a const reference to the shader notifier.
     [[nodiscard]] const VideoCore::ShaderNotify& ShaderNotify() const;
 
+    /// Returns GPU Channel Scheduler.
+    [[nodiscard]] Tegra::Control::Scheduler& Scheduler();
+
+    /// Returns GPU Channel Scheduler.
+    [[nodiscard]] const Tegra::Control::Scheduler& Scheduler() const;
+
     [[nodiscard]] u64 GetTicks() const;
 
     [[nodiscard]] bool IsAsync() const;

src/video_core/gpu_thread.cpp

@@ -34,13 +34,15 @@ static void RunThread(std::stop_token stop_token, Core::System& system,
 
     CommandDataContainer next;
 
+    scheduler.Init();
+
     while (!stop_token.stop_requested()) {
         state.queue.PopWait(next, stop_token);
         if (stop_token.stop_requested()) {
             break;
         }
-        if (auto* submit_list = std::get_if<SubmitListCommand>(&next.data)) {
-            scheduler.Push(submit_list->channel, std::move(submit_list->entries));
+        if (std::holds_alternative<SubmitListCommand>(next.data)) {
+            scheduler.Resume();
         } else if (std::holds_alternative<GPUTickCommand>(next.data)) {
             system.GPU().TickWork();
         } else if (const auto* flush = std::get_if<FlushRegionCommand>(&next.data)) {
@@ -67,14 +69,16 @@ ThreadManager::~ThreadManager() = default;
 
 void ThreadManager::StartThread(VideoCore::RendererBase& renderer,
                                 Core::Frontend::GraphicsContext& context,
-                                Tegra::Control::Scheduler& scheduler) {
+                                Tegra::Control::Scheduler& scheduler_) {
     rasterizer = renderer.ReadRasterizer();
+    scheduler = &scheduler_;
     thread = std::jthread(RunThread, std::ref(system), std::ref(renderer), std::ref(context),
-                          std::ref(scheduler), std::ref(state));
+                          std::ref(scheduler_), std::ref(state));
 }
 
 void ThreadManager::SubmitList(s32 channel, Tegra::CommandList&& entries) {
-    PushCommand(SubmitListCommand(channel, std::move(entries)));
+    scheduler->Push(channel, std::move(entries));
+    PushCommand(SubmitListCommand());
 }
 
 void ThreadManager::FlushRegion(DAddr addr, u64 size) {

src/video_core/gpu_thread.h

@@ -36,13 +36,7 @@ class RendererBase;
 namespace VideoCommon::GPUThread {
 
 /// Command to signal to the GPU thread that a command list is ready for processing
-struct SubmitListCommand final {
-    explicit SubmitListCommand(s32 channel_, Tegra::CommandList&& entries_)
-        : channel{channel_}, entries{std::move(entries_)} {}
-
-    s32 channel;
-    Tegra::CommandList entries;
-};
+struct SubmitListCommand final {};
 
 /// Command to signal to the GPU thread to flush a region
 struct FlushRegionCommand final {
@@ -124,6 +118,7 @@ public:
 private:
     /// Pushes a command to be executed by the GPU thread
     u64 PushCommand(CommandData&& command_data, bool block = false);
+    Tegra::Control::Scheduler* scheduler;
 
     Core::System& system;
     const bool is_async;