StagingBuff: support both Mlock & IOUring pin; Use in IoUAssmEngn

We use io_uring_register_buffers() for IoUringAssemblyEngine instead
of using mlock(). This __appears__ to have reduced CPU utilization on
the Dell laptop. Could also be that we recently upgraded total RAM
from 8GiB to 32GiB.

include/user/stagingBuffer.h

@@ -15,6 +15,14 @@ namespace stim_buff {
 // Forward declaration
 class FrameAssemblyDesc;
 
+} // namespace stim_buff
+} // namespace smo
+
+struct io_uring;
+
+namespace smo {
+namespace stim_buff {
+
 /**
  * StagingBuffer manages a large buffer to guide io_uring in assembling some
  * number of Livox Avia pcloud UDP dgrams into a single stim frame.
@@ -28,6 +36,13 @@ class FrameAssemblyDesc;
 class StagingBuffer
 {
 public:
+	enum class PinningMechanism
+	{
+		NONE,
+		MLOCK,
+		IO_URING
+	};
+
 	class IOEngineConstraints
 	{
 	public:
@@ -83,7 +98,7 @@ public:
 		const IOEngineConstraints& inputEngineConstraints,
 		const IOEngineConstraints& outputEngineConstraints,
 		size_t nSlots);
-	~StagingBuffer() = default;
+	~StagingBuffer();
 
 	// Non-copyable, movable
 	StagingBuffer(const StagingBuffer&) = delete;
@@ -91,6 +106,50 @@ public:
 	StagingBuffer(StagingBuffer&&) = default;
 	StagingBuffer& operator=(StagingBuffer&&) = default;
 
+	class Pinner
+	{
+	public:
+		Pinner(const Pinner&) = delete;
+		Pinner& operator=(const Pinner&) = delete;
+		Pinner(Pinner&&) = delete;
+		Pinner& operator=(Pinner&&) = delete;
+
+	protected:
+		explicit Pinner(StagingBuffer& parent_);
+		~Pinner() = default;
+
+		StagingBuffer& parent;
+	};
+
+	class MlockPinner
+	: public Pinner
+	{
+	public:
+		explicit MlockPinner(StagingBuffer& parent);
+		~MlockPinner();
+
+		MlockPinner(const MlockPinner&) = delete;
+		MlockPinner& operator=(const MlockPinner&) = delete;
+		MlockPinner(MlockPinner&&) = delete;
+		MlockPinner& operator=(MlockPinner&&) = delete;
+	};
+
+	class IoUringPinner
+	: public Pinner
+	{
+	public:
+		IoUringPinner(StagingBuffer& parent, struct io_uring* ring);
+		~IoUringPinner();
+
+		IoUringPinner(const IoUringPinner&) = delete;
+		IoUringPinner& operator=(const IoUringPinner&) = delete;
+		IoUringPinner(IoUringPinner&&) = delete;
+		IoUringPinner& operator=(IoUringPinner&&) = delete;
+
+	private:
+		struct io_uring* ring;
+	};
+
 public:
 	/**	EXPLANATION:
 	 * Returns an input-engine-agnostic descriptor describing per-frame packet
@@ -104,6 +163,9 @@ public:
 	void startAssembly() { assemblingFlag.store(true); }
 	void stopAssembly() { assemblingFlag.store(false); }
 
+	std::unique_ptr<MlockPinner> makeMlockPinner();
+	std::unique_ptr<IoUringPinner> makeIoUringPinner(struct io_uring* ring);
+
 	/**	EXPLANATION:
 	 * Returns an iovec for io_uring registration.
 	 * The buffer is mmap()-allocated and suitable for IORING_REGISTER_BUFFERS.
@@ -144,6 +206,7 @@ public:
 
 private:
 	void computeSlotStrideAndBufferSize();
+	void assertUnpinnedAndMarkPinned(PinningMechanism mechanism);
 	static size_t calculateFirstSlotOffsetAndValidate(
 		uint8_t* buffer,
 		size_t bufferNBytes,
@@ -163,7 +226,6 @@ private:
 		{
 			if (ptr != nullptr && size > 0)
 			{
-				munlock(ptr, size);
 				munmap(ptr, size);
 			}
 		}
@@ -173,14 +235,14 @@ private:
 	// Using unique_ptr<uint8_t, MmapDeleter> instead of array syntax
 	// since we have a custom deleter that knows the size
 	std::unique_ptr<uint8_t, MmapDeleter> buffer;
-	size_t bufferNBytes;
+	size_t bufferNBytes = 0;
 
 	// Layout/invariants
-	size_t nSlots;
+	size_t nSlots = 0;
 
 public:
-	size_t slotStrideNBytes;
-	size_t firstSlotOffsetNBytes;  // offset from buffer start to first slot
+	size_t slotStrideNBytes = 0;
+	size_t firstSlotOffsetNBytes = 0;  // offset from buffer start to first slot
 
 private:
 	IOEngineConstraints inputConstraints;
@@ -189,8 +251,13 @@ private:
 	mutable std::shared_ptr<FrameAssemblyDesc> frameDesc;
 
 	// Current state
-	std::atomic<size_t> currentNBytes;
-	std::atomic<bool> assemblingFlag;
+	std::atomic<size_t> currentNBytes{0};
+	std::atomic<bool> assemblingFlag{false};
+	bool currentlyPinned = false;
+	PinningMechanism currentPinningMechanism = PinningMechanism::NONE;
+
+	friend class MlockPinner;
+	friend class IoUringPinner;
 };
 
 } // namespace stim_buff