Merge remote-tracking branch 'upstream/master' into int-flags

4 files changed, 153 insertions, 85 deletions

diff --git a/src/core/hle/service/nvflinger/buffer_queue.cpp b/src/core/hle/service/nvflinger/buffer_queue.cpp
index 637b310d7..5578181a4 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.cpp
+++ b/src/core/hle/service/nvflinger/buffer_queue.cpp
@@ -22,127 +22,169 @@ BufferQueue::BufferQueue(Kernel::KernelCore& kernel, u32 id, u64 layer_id)
 BufferQueue::~BufferQueue() = default;
 
 void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer) {
+    ASSERT(slot < buffer_slots);
     LOG_WARNING(Service, "Adding graphics buffer {}", slot);
 
-    free_buffers.push_back(slot);
-    queue.push_back({
+    {
+        std::unique_lock lock{free_buffers_mutex};
+        free_buffers.push_back(slot);
+    }
+    free_buffers_condition.notify_one();
+
+    buffers[slot] = {
         .slot = slot,
         .status = Buffer::Status::Free,
         .igbp_buffer = igbp_buffer,
-    });
+        .transform = {},
+        .crop_rect = {},
+        .swap_interval = 0,
+        .multi_fence = {},
+    };
 
     buffer_wait_event.writable->Signal();
 }
 
 std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> BufferQueue::DequeueBuffer(u32 width,
                                                                                        u32 height) {
+    // Wait for first request before trying to dequeue
+    {
+        std::unique_lock lock{free_buffers_mutex};
+        free_buffers_condition.wait(lock, [this] { return !free_buffers.empty() || !is_connect; });
+    }
 
-    if (free_buffers.empty()) {
+    if (!is_connect) {
+        // Buffer was disconnected while the thread was blocked, this is most likely due to
+        // emulation being stopped
         return std::nullopt;
     }
 
+    std::unique_lock lock{free_buffers_mutex};
+
     auto f_itr = free_buffers.begin();
-    auto itr = queue.end();
+    auto slot = buffers.size();
 
     while (f_itr != free_buffers.end()) {
-        auto slot = *f_itr;
-        itr = std::find_if(queue.begin(), queue.end(), [&](const Buffer& buffer) {
-            // Only consider free buffers. Buffers become free once again after they've been
-            // Acquired and Released by the compositor, see the NVFlinger::Compose method.
-            if (buffer.status != Buffer::Status::Free) {
-                return false;
-            }
-
-            if (buffer.slot != slot) {
-                return false;
-            }
-
-            // Make sure that the parameters match.
-            return buffer.igbp_buffer.width == width && buffer.igbp_buffer.height == height;
-        });
-
-        if (itr != queue.end()) {
+        const Buffer& buffer = buffers[*f_itr];
+        if (buffer.status == Buffer::Status::Free && buffer.igbp_buffer.width == width &&
+            buffer.igbp_buffer.height == height) {
+            slot = *f_itr;
             free_buffers.erase(f_itr);
             break;
         }
         ++f_itr;
     }
-
-    if (itr == queue.end()) {
+    if (slot == buffers.size()) {
         return std::nullopt;
     }
-
-    itr->status = Buffer::Status::Dequeued;
-    return {{itr->slot, &itr->multi_fence}};
+    buffers[slot].status = Buffer::Status::Dequeued;
+    return {{buffers[slot].slot, &buffers[slot].multi_fence}};
 }
 
 const IGBPBuffer& BufferQueue::RequestBuffer(u32 slot) const {
-    auto itr = std::find_if(queue.begin(), queue.end(),
-                            [&](const Buffer& buffer) { return buffer.slot == slot; });
-    ASSERT(itr != queue.end());
-    ASSERT(itr->status == Buffer::Status::Dequeued);
-    return itr->igbp_buffer;
+    ASSERT(slot < buffers.size());
+    ASSERT(buffers[slot].status == Buffer::Status::Dequeued);
+    ASSERT(buffers[slot].slot == slot);
+
+    return buffers[slot].igbp_buffer;
 }
 
 void BufferQueue::QueueBuffer(u32 slot, BufferTransformFlags transform,
                               const Common::Rectangle<int>& crop_rect, u32 swap_interval,
                               Service::Nvidia::MultiFence& multi_fence) {
-    auto itr = std::find_if(queue.begin(), queue.end(),
-                            [&](const Buffer& buffer) { return buffer.slot == slot; });
-    ASSERT(itr != queue.end());
-    ASSERT(itr->status == Buffer::Status::Dequeued);
-    itr->status = Buffer::Status::Queued;
-    itr->transform = transform;
-    itr->crop_rect = crop_rect;
-    itr->swap_interval = swap_interval;
-    itr->multi_fence = multi_fence;
+    ASSERT(slot < buffers.size());
+    ASSERT(buffers[slot].status == Buffer::Status::Dequeued);
+    ASSERT(buffers[slot].slot == slot);
+
+    buffers[slot].status = Buffer::Status::Queued;
+    buffers[slot].transform = transform;
+    buffers[slot].crop_rect = crop_rect;
+    buffers[slot].swap_interval = swap_interval;
+    buffers[slot].multi_fence = multi_fence;
+    std::unique_lock lock{queue_sequence_mutex};
     queue_sequence.push_back(slot);
 }
 
+void BufferQueue::CancelBuffer(u32 slot, const Service::Nvidia::MultiFence& multi_fence) {
+    ASSERT(slot < buffers.size());
+    ASSERT(buffers[slot].status != Buffer::Status::Free);
+    ASSERT(buffers[slot].slot == slot);
+
+    buffers[slot].status = Buffer::Status::Free;
+    buffers[slot].multi_fence = multi_fence;
+    buffers[slot].swap_interval = 0;
+
+    {
+        std::unique_lock lock{free_buffers_mutex};
+        free_buffers.push_back(slot);
+    }
+    free_buffers_condition.notify_one();
+
+    buffer_wait_event.writable->Signal();
+}
+
 std::optional<std::reference_wrapper<const BufferQueue::Buffer>> BufferQueue::AcquireBuffer() {
-    auto itr = queue.end();
+    std::unique_lock lock{queue_sequence_mutex};
+    std::size_t buffer_slot = buffers.size();
     // Iterate to find a queued buffer matching the requested slot.
-    while (itr == queue.end() && !queue_sequence.empty()) {
-        const u32 slot = queue_sequence.front();
-        itr = std::find_if(queue.begin(), queue.end(), [&slot](const Buffer& buffer) {
-            return buffer.status == Buffer::Status::Queued && buffer.slot == slot;
-        });
+    while (buffer_slot == buffers.size() && !queue_sequence.empty()) {
+        const auto slot = static_cast<std::size_t>(queue_sequence.front());
+        ASSERT(slot < buffers.size());
+        if (buffers[slot].status == Buffer::Status::Queued) {
+            ASSERT(buffers[slot].slot == slot);
+            buffer_slot = slot;
+        }
         queue_sequence.pop_front();
     }
-    if (itr == queue.end()) {
+    if (buffer_slot == buffers.size()) {
         return std::nullopt;
     }
-    itr->status = Buffer::Status::Acquired;
-    return *itr;
+    buffers[buffer_slot].status = Buffer::Status::Acquired;
+    return {{buffers[buffer_slot]}};
 }
 
 void BufferQueue::ReleaseBuffer(u32 slot) {
-    auto itr = std::find_if(queue.begin(), queue.end(),
-                            [&](const Buffer& buffer) { return buffer.slot == slot; });
-    ASSERT(itr != queue.end());
-    ASSERT(itr->status == Buffer::Status::Acquired);
-    itr->status = Buffer::Status::Free;
-    free_buffers.push_back(slot);
+    ASSERT(slot < buffers.size());
+    ASSERT(buffers[slot].status == Buffer::Status::Acquired);
+    ASSERT(buffers[slot].slot == slot);
+
+    buffers[slot].status = Buffer::Status::Free;
+    {
+        std::unique_lock lock{free_buffers_mutex};
+        free_buffers.push_back(slot);
+    }
+    free_buffers_condition.notify_one();
 
     buffer_wait_event.writable->Signal();
 }
 
-void BufferQueue::Disconnect() {
-    queue.clear();
+void BufferQueue::Connect() {
+    std::unique_lock lock{queue_sequence_mutex};
     queue_sequence.clear();
-    id = 1;
-    layer_id = 1;
+    is_connect = true;
+}
+
+void BufferQueue::Disconnect() {
+    buffers.fill({});
+    {
+        std::unique_lock lock{queue_sequence_mutex};
+        queue_sequence.clear();
+    }
+    buffer_wait_event.writable->Signal();
+    is_connect = false;
+    free_buffers_condition.notify_one();
 }
 
 u32 BufferQueue::Query(QueryType type) {
-    LOG_WARNING(Service, "(STUBBED) called type={}", static_cast<u32>(type));
+    LOG_WARNING(Service, "(STUBBED) called type={}", type);
 
     switch (type) {
     case QueryType::NativeWindowFormat:
         return static_cast<u32>(PixelFormat::RGBA8888);
+    case QueryType::NativeWindowWidth:
+    case QueryType::NativeWindowHeight:
+        break;
     }
-
-    UNIMPLEMENTED();
+    UNIMPLEMENTED_MSG("Unimplemented query type={}", type);
     return 0;
 }
 
diff --git a/src/core/hle/service/nvflinger/buffer_queue.h b/src/core/hle/service/nvflinger/buffer_queue.h
index 8a837e5aa..ad7469277 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.h
+++ b/src/core/hle/service/nvflinger/buffer_queue.h
@@ -4,7 +4,9 @@
 
 #pragma once
 
+#include <condition_variable>
 #include <list>
+#include <mutex>
 #include <optional>
 #include <vector>
 
@@ -21,6 +23,7 @@ class KernelCore;
 
 namespace Service::NVFlinger {
 
+constexpr u32 buffer_slots = 0x40;
 struct IGBPBuffer {
     u32_le magic;
     u32_le width;
@@ -95,8 +98,10 @@ public:
     void QueueBuffer(u32 slot, BufferTransformFlags transform,
                      const Common::Rectangle<int>& crop_rect, u32 swap_interval,
                      Service::Nvidia::MultiFence& multi_fence);
+    void CancelBuffer(u32 slot, const Service::Nvidia::MultiFence& multi_fence);
     std::optional<std::reference_wrapper<const Buffer>> AcquireBuffer();
     void ReleaseBuffer(u32 slot);
+    void Connect();
     void Disconnect();
     u32 Query(QueryType type);
 
@@ -104,18 +109,30 @@ public:
         return id;
     }
 
+    bool IsConnected() const {
+        return is_connect;
+    }
+
     std::shared_ptr<Kernel::WritableEvent> GetWritableBufferWaitEvent() const;
 
     std::shared_ptr<Kernel::ReadableEvent> GetBufferWaitEvent() const;
 
 private:
-    u32 id;
-    u64 layer_id;
+    BufferQueue(const BufferQueue&) = delete;
+
+    u32 id{};
+    u64 layer_id{};
+    std::atomic_bool is_connect{};
 
     std::list<u32> free_buffers;
-    std::vector<Buffer> queue;
+    std::array<Buffer, buffer_slots> buffers;
     std::list<u32> queue_sequence;
     Kernel::EventPair buffer_wait_event;
+
+    std::mutex free_buffers_mutex;
+    std::condition_variable free_buffers_condition;
+
+    std::mutex queue_sequence_mutex;
 };
 
 } // namespace Service::NVFlinger
diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp
index c64673dba..4b3581949 100644
--- a/src/core/hle/service/nvflinger/nvflinger.cpp
+++ b/src/core/hle/service/nvflinger/nvflinger.cpp
@@ -88,6 +88,10 @@ NVFlinger::NVFlinger(Core::System& system) : system(system) {
 }
 
 NVFlinger::~NVFlinger() {
+    for (auto& buffer_queue : buffer_queues) {
+        buffer_queue->Disconnect();
+    }
+
     if (system.IsMulticore()) {
         is_running = false;
         wait_event->Set();
@@ -104,6 +108,8 @@ void NVFlinger::SetNVDrvInstance(std::shared_ptr<Nvidia::Module> instance) {
 }
 
 std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) {
+    const auto guard = Lock();
+
     LOG_DEBUG(Service, "Opening \"{}\" display", name);
 
     // TODO(Subv): Currently we only support the Default display.
@@ -121,6 +127,7 @@ std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) {
 }
 
 std::optional<u64> NVFlinger::CreateLayer(u64 display_id) {
+    const auto guard = Lock();
     auto* const display = FindDisplay(display_id);
 
     if (display == nullptr) {
@@ -129,18 +136,22 @@ std::optional<u64> NVFlinger::CreateLayer(u64 display_id) {
 
     const u64 layer_id = next_layer_id++;
     const u32 buffer_queue_id = next_buffer_queue_id++;
-    buffer_queues.emplace_back(system.Kernel(), buffer_queue_id, layer_id);
-    display->CreateLayer(layer_id, buffer_queues.back());
+    buffer_queues.emplace_back(
+        std::make_unique<BufferQueue>(system.Kernel(), buffer_queue_id, layer_id));
+    display->CreateLayer(layer_id, *buffer_queues.back());
     return layer_id;
 }
 
 void NVFlinger::CloseLayer(u64 layer_id) {
+    const auto guard = Lock();
+
     for (auto& display : displays) {
         display.CloseLayer(layer_id);
     }
 }
 
 std::optional<u32> NVFlinger::FindBufferQueueId(u64 display_id, u64 layer_id) const {
+    const auto guard = Lock();
     const auto* const layer = FindLayer(display_id, layer_id);
 
     if (layer == nullptr) {
@@ -151,6 +162,7 @@ std::optional<u32> NVFlinger::FindBufferQueueId(u64 display_id, u64 layer_id) co
 }
 
 std::shared_ptr<Kernel::ReadableEvent> NVFlinger::FindVsyncEvent(u64 display_id) const {
+    const auto guard = Lock();
     auto* const display = FindDisplay(display_id);
 
     if (display == nullptr) {
@@ -160,20 +172,16 @@ std::shared_ptr<Kernel::ReadableEvent> NVFlinger::FindVsyncEvent(u64 display_id)
     return display->GetVSyncEvent();
 }
 
-BufferQueue& NVFlinger::FindBufferQueue(u32 id) {
+BufferQueue* NVFlinger::FindBufferQueue(u32 id) {
+    const auto guard = Lock();
     const auto itr = std::find_if(buffer_queues.begin(), buffer_queues.end(),
-                                  [id](const auto& queue) { return queue.GetId() == id; });
-
-    ASSERT(itr != buffer_queues.end());
-    return *itr;
-}
+                                  [id](const auto& queue) { return queue->GetId() == id; });
 
-const BufferQueue& NVFlinger::FindBufferQueue(u32 id) const {
-    const auto itr = std::find_if(buffer_queues.begin(), buffer_queues.end(),
-                                  [id](const auto& queue) { return queue.GetId() == id; });
+    if (itr == buffer_queues.end()) {
+        return nullptr;
+    }
 
-    ASSERT(itr != buffer_queues.end());
-    return *itr;
+    return itr->get();
 }
 
 VI::Display* NVFlinger::FindDisplay(u64 display_id) {
@@ -242,6 +250,10 @@ void NVFlinger::Compose() {
 
         const auto& igbp_buffer = buffer->get().igbp_buffer;
 
+        if (!system.IsPoweredOn()) {
+            return; // We are likely shutting down
+        }
+
         auto& gpu = system.GPU();
         const auto& multi_fence = buffer->get().multi_fence;
         guard->unlock();
diff --git a/src/core/hle/service/nvflinger/nvflinger.h b/src/core/hle/service/nvflinger/nvflinger.h
index 1ebe949c0..c6765259f 100644
--- a/src/core/hle/service/nvflinger/nvflinger.h
+++ b/src/core/hle/service/nvflinger/nvflinger.h
@@ -75,10 +75,7 @@ public:
     [[nodiscard]] std::shared_ptr<Kernel::ReadableEvent> FindVsyncEvent(u64 display_id) const;
 
     /// Obtains a buffer queue identified by the ID.
-    [[nodiscard]] BufferQueue& FindBufferQueue(u32 id);
-
-    /// Obtains a buffer queue identified by the ID.
-    [[nodiscard]] const BufferQueue& FindBufferQueue(u32 id) const;
+    [[nodiscard]] BufferQueue* FindBufferQueue(u32 id);
 
     /// Performs a composition request to the emulated nvidia GPU and triggers the vsync events when
     /// finished.
@@ -86,11 +83,11 @@ public:
 
     [[nodiscard]] s64 GetNextTicks() const;
 
+private:
     [[nodiscard]] std::unique_lock<std::mutex> Lock() const {
         return std::unique_lock{*guard};
     }
 
-private:
     /// Finds the display identified by the specified ID.
     [[nodiscard]] VI::Display* FindDisplay(u64 display_id);
 
@@ -110,7 +107,7 @@ private:
     std::shared_ptr<Nvidia::Module> nvdrv;
 
     std::vector<VI::Display> displays;
-    std::vector<BufferQueue> buffer_queues;
+    std::vector<std::unique_ptr<BufferQueue>> buffer_queues;
 
     /// Id to use for the next layer that is created, this counter is shared among all displays.
     u64 next_layer_id = 1;