1. Memory Writes = Posted Transactions

• Key behavior:
  ✅ Requester does not wait for a Completion TLP.
  ✅ Transaction is considered "done" on that Link once the packet is transmitted successfully.

• Performance impact:

  • Faster → no round-trip latency.
  • More efficient use of bandwidth → no Completion packets are consuming the Link.
  • Great for bulk data transfer (DMA, frame buffers, etc.).

• Trade-off:
  ❌ Requester gets no direct feedback.

  • If Completer has an error, Requester is not notified.
  • Completer may log the error and optionally send a Message to the Root Complex → then system software can react.

2. Message Writes = Special Case of Posted Transactions

• Purpose:
  Messages are used to communicate events or signals (instead of data movement).
  Examples:

  • Interrupt requests (MSI/MSI-X)
  • Power management events
  • Error reporting

• Routing flexibility:

  • Directed message → routed to a specific Completer.
  • Broadcast message → Root sends to all Endpoints.
  • Implicit Root-directed message → Endpoint sends, automatically routed to Root Complex.

• Design benefit:

  • Eliminates side-band pins that were used in PCI (for interrupts, PME#, SERR#, etc.).
  • Reduces board complexity & pin count.
  • Everything travels as a normal PCIe TLP.