Support for `Recover` method on bridged endpoints

[jk-ozlabs]

Currently, a Recover invocation will attempt a Get Endpoint ID (to check endpoint connectivity), but will do so using a physical addressing:

	/*
	 * Test if we still have connectivity to the endpoint. If we do, we will get a
	 * response reporting the current EID. This is the test recommended by 8.17.6
	 * of DSP0236 v1.3.1.
	 */
	rc = query_get_endpoint_id(ctx, &peer->phys, &peer->recovery.eid,
				   &peer->recovery.endpoint_type,
				   &peer->recovery.medium_spec, /*peer=*/NULL);

The NULL peer causes the physical addressing there.

Since we cannot target bridged endpoints with a phys address, we will need to use vanilla EID addressing for those.

[faizana-nvidia]

Hello @jk-ozlabs

I too was looking for this, if you still haven't picked it up as your current working item, I'll be happy to push a change.

[jk-ozlabs]

@faizana-nvidia sure, if you want to look at this!

I'd suggest working out a plan here before writing the implementation though, there are some design points for the recovery flow that we would need to sort out too. Mainly:

• we'll probably want to check connectivity to the bridge too
• we'll need to handle cases where the EID has been reassigned by the downstream bus owner (ie, has a new UUID), by generating a remove & re-add.

[faizana-nvidia]

great! let me update here with the flow

[faizana-nvidia]

Downstream Endpoint Recovery Flow

method_endpoint_recover (D-Bus entry point) — unchanged for both peer types

flowchart TD
    A([method_endpoint_recover called]) --> B{peer->degraded?}
    B -- yes --> Z([no-op, return OK])
    B -- no --> C["set recovery.npolls, recovery.delay\nschedule peer_endpoint_recover timer\npeer->degraded = true\nemit Connectivity changed"]
    C --> D([return — common for direct and downstream peers])

Loading

---

peer_endpoint_recover (timer callback) — downstream branch inserted after physical probe

flowchart TD
    TIMER([timer fires: peer_endpoint_recover]) --> DEC[Decrement recovery.npolls]

    DEC --> PHYS["[EXISTING — unchanged]\nPhysical probe\nquery_get_endpoint_id\n  dest=peer->phys, peer=NULL\n  → for direct: tests the device\n  → for downstream: tests the bridge\nstores result in recovery.eid"]

    PHYS --> PHYSOK{responded?}
    PHYSOK -- no --> RESCHED

    PHYSOK -- yes --> BRIDGE["[NEW]\nfind_bridge_for_peer:\nscan peers for p where\n  p.pool_size > 0\n  && match_phys(p.phys, peer.phys)\n  && p.pool_start <= peer.eid\n       < p.pool_start + p.pool_size"]

    BRIDGE --> ISBRIDGE{downstream\npeer?}

    %% ── DOWNSTREAM BRANCH (new) ──────────────────────────────
    ISBRIDGE -- "yes\n(bridge found)" --> EIDRP["[NEW — downstream only]\nEID-routed probe\nquery_get_endpoint_id\n  dest=peer->phys, peer=peer\n  → routed through bridge to device\noverwrites recovery.eid with\ndownstream device's EID"]

    EIDRP --> EIDOK{responded?}
    EIDOK -- no --> RESCHED

    EIDOK -- yes --> UUIDQ["[NEW — downstream only]\nUUID via EID routing\nquery_get_peer_uuid_by_eid\n  → gets downstream device UUID\n  not the bridge UUID"]

    UUIDQ --> UUIDOK{UUID query\nsucceeded?}
    UUIDOK -- no --> RECLAIM

    UUIDOK -- yes --> UUIDMATCH{UUID matches\npeer->uuid?}

    UUIDMATCH -- "yes — same device" --> EIDMATCH{recovery.eid\n== peer->eid?}
    EIDMATCH -- yes --> SUCCESS
    EIDMATCH -- "no — EID drifted\nbridge is bus owner,\ncannot send Set Endpoint ID" --> RECLAIM

    UUIDMATCH -- "no or nil UUID\nnew device at this EID" --> READD["[NEW — downstream only]\nBridge re-assigned EID to new device\n\nsave phys / eid / net\nremove_peer old  → emits EndpointRemoved\nadd_peer eid, net, allow_bridged=true\nsetup_added_peer new  → emits EndpointAdded"]
    READD --> DONE([return])

    %% ── DIRECT ENDPOINT BRANCH (existing, unchanged) ─────────
    ISBRIDGE -- "no\n(direct endpoint)" --> DIRECT["[EXISTING — unchanged]\nif recovery.eid != peer->eid:\n  query_get_peer_uuid_by_phys\n  UUID mismatch → remove_peer\n                  + endpoint_assign_eid\n  UUID match    → endpoint_send_set_endpoint_id\n                  + change_peer_eid"]
    DIRECT --> SUCCESS

    %% ── SHARED SUCCESS PATH (existing, unchanged) ────────────
    SUCCESS(["[EXISTING — unchanged]\npeer->degraded = false\nemit Connectivity changed\nunref recovery.source\nzero recovery.delay / npolls"])

    %% ── SHARED RESCHEDULE / RECLAIM (existing, unchanged) ────
    RESCHED{recovery.npolls > 0?}
    RESCHED -- yes --> RETRY[reschedule timer\nat recovery.delay]
    RESCHED -- no --> RECLAIM
    RECLAIM([reclaim: remove_peer])

Loading

---

[faizana-nvidia]

Updating the flow here, will create PR for the same if this flow looks okay.