Fix ADR-0025: IPCQ direction addressing via address-based matching
2-rank bidirectional ring deadlock: when E and W neighbors point to the same peer, sender-coord matching in _handle_meta_arrival / _credit_worker picked the first direction in dict order, landing data in the wrong rx slot relative to what the kernel recv(W) was waiting on. Fix (ADR-0025 D1/D2/D3): - install.reverse_direction: prefer OPPOSITE direction (E↔W, N↔S) when peer has it pointing back to us; fallback to any matching for topologies without opposite convention (tree_binary parent/child). - _handle_meta_arrival: match by token.dst_addr range against each qp's my_rx_base_pa + n_slots × slot_size window (unambiguous). - _credit_worker: match by credit.dst_rx_base_pa == qp.peer.rx_base_pa. - IpcqCreditMetadata: new dst_rx_base_pa field carrying receiver-side rx base; _delayed_credit_send fills it from the consuming qp. Tests (Phase 1 → Phase 2): - test_reverse_direction_opposite_preference_2rank_ring - test_reverse_direction_opposite_preference_4rank_ring_sanity - test_meta_arrival_matches_by_dst_addr_same_peer - test_credit_matches_by_dst_rx_base_pa_same_peer - Existing credit-return test updated with dst_rx_base_pa. 508 tests pass. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -370,11 +370,21 @@ class PeIpcqComponent(ComponentBase):
|
||||
# ── Metadata arrival from PE_DMA (D9) ──
|
||||
|
||||
def _handle_meta_arrival(self, msg: IpcqMetaArrival) -> None:
|
||||
"""Match arrival to the correct direction by dst_addr range (ADR-0025 D2).
|
||||
|
||||
Each direction has a unique rx buffer address range
|
||||
([my_rx_base_pa, my_rx_base_pa + n_slots * slot_size)). The token's
|
||||
dst_addr (set by the sender's IPCQ when computing the peer slot
|
||||
address) falls within exactly one such range. Address-based matching
|
||||
is unambiguous even when multiple directions share the same peer
|
||||
(2-rank bidirectional ring).
|
||||
"""
|
||||
token = msg.token
|
||||
sender_key = (token.src_sip, token.src_cube, token.src_pe)
|
||||
dst_addr = token.dst_addr
|
||||
for d, qp in self._queue_pairs.items():
|
||||
p = qp["peer"]
|
||||
if (p.sip, p.cube, p.pe) == sender_key:
|
||||
base = qp["my_rx_base_pa"]
|
||||
size = qp["n_slots"] * qp["slot_size"]
|
||||
if base <= dst_addr < base + size:
|
||||
qp["peer_head_cache"] = max(qp["peer_head_cache"], token.sender_seq + 1)
|
||||
# Track arrived token for strict-mode peek
|
||||
self._arrived_tokens.setdefault(d, []).append(token)
|
||||
@@ -391,19 +401,22 @@ class PeIpcqComponent(ComponentBase):
|
||||
if not ev.triggered:
|
||||
ev.succeed()
|
||||
return
|
||||
# Unknown sender — silently drop (could log)
|
||||
# Unknown dst_addr — silently drop (could log)
|
||||
|
||||
# ── Credit return (fast path) ──
|
||||
|
||||
def _credit_worker(self, env: simpy.Environment) -> Generator:
|
||||
"""Process IpcqCreditMetadata from credit_inbox."""
|
||||
"""Process IpcqCreditMetadata from credit_inbox.
|
||||
|
||||
Matches credit to the correct direction by `credit.dst_rx_base_pa ==
|
||||
qp.peer.rx_base_pa` (ADR-0025 D3). This is unambiguous even when
|
||||
multiple directions share the same peer (2-rank bidirectional ring).
|
||||
"""
|
||||
assert self._credit_inbox is not None
|
||||
while True:
|
||||
credit: IpcqCreditMetadata = yield self._credit_inbox.get()
|
||||
sender_key = (credit.src_sip, credit.src_cube, credit.src_pe)
|
||||
for d, qp in self._queue_pairs.items():
|
||||
p = qp["peer"]
|
||||
if (p.sip, p.cube, p.pe) == sender_key:
|
||||
if qp["peer"].rx_base_pa == credit.dst_rx_base_pa:
|
||||
qp["peer_tail_cache"] = max(qp["peer_tail_cache"], credit.consumer_seq)
|
||||
# Wake any blocked send on this direction
|
||||
waiters = self._send_waiters.get(d, [])
|
||||
@@ -421,12 +434,19 @@ class PeIpcqComponent(ComponentBase):
|
||||
new_tail: int,
|
||||
) -> Generator:
|
||||
"""Wait bottleneck-BW latency, then put IpcqCreditMetadata into peer
|
||||
credit store (D9 fast path)."""
|
||||
credit store (D9 fast path).
|
||||
|
||||
Carries ``dst_rx_base_pa`` = this PE's my_rx_base_pa for the
|
||||
consumed direction. The peer (original sender) matches this against
|
||||
qp.peer.rx_base_pa to identify the correct qp (ADR-0025 D3).
|
||||
"""
|
||||
latency_ns = self._credit_latency_ns(direction)
|
||||
if latency_ns > 0:
|
||||
yield env.timeout(latency_ns)
|
||||
qp = self._queue_pairs[direction]
|
||||
meta = IpcqCreditMetadata(
|
||||
consumer_seq=new_tail,
|
||||
dst_rx_base_pa=qp["my_rx_base_pa"],
|
||||
src_sip=self._self_sip,
|
||||
src_cube=self._self_cube,
|
||||
src_pe=self._self_pe,
|
||||
|
||||
Reference in New Issue
Block a user