It ensures error-free delivery of TLPs (Transaction Layer Packets) across every link, using automatic retransmission when errors are detected.
👉 Think of it like a “guaranteed delivery receipt” system at the Data Link Layer.

🧩 How It Works (Step by Step)

1. Sending a TLP
   • Every outgoing TLP gets:
     • A Sequence Number (unique ID for ordering and tracking)
     • An LCRC (Link CRC, error detection checksum)
   • The transmitter also stores a copy of the TLP in its Replay Buffer until it knows the receiver got it safely.

2. Receiver Checks It

   • The receiver checks:

     • LCRC → verifies integrity
     • Sequence Number → checks order

   • Then one of two things happens:

     • ✅ No error → Receiver sends an Ack DLLP with the last good sequence number.
     • ❌ Error detected → Receiver drops the TLP and sends a Nak DLLP.
       

3. Transmitter Responds
   • If Ack received → Transmitter flushes the TLP (and any older ones) from the Replay Buffer.
   • If Nak received → Transmitter replays all unacknowledged TLPs from its Replay Buffer.

Since most errors are transient, simply replaying usually fixes the problem.

🔁 Example Walkthrough (Memory Read across a Switch)

• Step 1: Requester sends MRd TLP → Switch checks it, sends Ack back → Requester flushes replay buffer.
• Step 2: Switch forwards TLP → Completer checks it, sends Ack back → Switch flushes buffer.
• Step 3: Completer generates CplD TLP → Switch checks it, sends Ack back → Completer flushes buffer.
• Step 4: Switch forwards CplD TLP → Requester checks it, sends Ack back → Switch flushes buffer → Requester passes data to core logic.

At every hop (Requester ↔ Switch ↔ Completer), Ack/Nak ensures reliability.