Skip to main content
GLink failures usually fall into one of four areas: the transport never becomes available, version negotiation fails, the channel never reaches connected state, or data ownership is not returned after a transfer or restart.

​
Check the transport state first

The transport must progress from down through negotiation to opened before channels can communicate. A failed or permanently down transport points to the transport plug-in, its physical link, or its dependency initialization rather than to the client channel. The transport plug-in is responsible for:
  • Reading from and writing to the physical transport
  • Reporting available read data and write space
  • Translating core commands and payloads
  • Closing the transport cleanly
The shared-memory native transport additionally depends on shared memory and interrupts in both directions. Other plug-ins have their own transport-specific dependencies.

​
Check channel state and callbacks

A channel is not ready merely because the open call returned a handle. Confirm that:
  1. The requested edge, transport, and channel name match the remote endpoint.
  2. A connected notification is received.
  3. The receiver queues an intent large enough for the expected packet.
  4. A receive notification is invoked.
  5. The receiver calls the completion function.
  6. The sender receives a transmit-done notification.
A missing transmit-done notification often means that the remote side has not consumed the receive buffer, while a missing receive notification usually means that no matching intent was available or the packet was rejected before delivery.

​
Enable logging

The GLink core and loopback server expose logging control. The supported logging modes are:
  • Informational – Default informational messages
  • Debug – Verbose protocol and state debugging
  • Performance – Performance-only logging
The modes can be combined. Performance mode suppresses other logging and keeps only messages emitted through performance logging macros.

​
Diagnose negotiation failures

Both endpoints independently start version negotiation. This is not a race: the transport is not opened until both sequences complete, and both sides converge on a compatible version and feature set. Check for:
  • No version acknowledgement
  • A selected version of zero (no compatible version found)
  • Features that are not accepted by the peer
  • Negotiation beginning before the transport can reliably exchange commands
  • A transport configuration that advertises incorrect supported versions
A version mismatch should select the lower mutually supported version. An optional feature should be enabled only when both sides advertise it.

​
Diagnose buffer ownership

Track the private token passed at each stage:
  1. Queue a receive intent.
  2. Receive notification with the data.
  3. Call the completion function.
  4. Receive transmit-done notification.
If a channel closes while packets or intents are pending, the client must handle abort callbacks for both receive-intent and transmit-packet private data. Releasing callback-owned memory before the abort notification can cause use-after-free or permanently leaked buffer ownership.

​
Subsystem restart symptoms

A restart can affect every channel connected to an edge. Confirm that the restart path:
  • Identifies the affected subsystem and its GLink edge
  • Notifies dependent endpoints
  • Closes or aborts affected channels
  • Returns receive-intent and transmit private data
  • Waits for cleanup before restart completion
  • Reopens channels after the remote endpoint is available again
Do not treat a transport reconnect as proof that all channels recovered; channel state and buffer ownership must be checked separately.