<html><body><div style="font-size: 11pt; font-family: arial, helvetica, sans-serif; direction: null; color: #000000;" data-attr="forced_root_block_attrs"> <div> <div data-attr="forced_root_block_attrs"> <p>This series fixes two correctness issues in the cxgbe PMD that can cause<br>traffic loss on Chelsio T6 adapters, especially when rte_flow QUEUE<br>rules concentrate ingress on a small set of RX queues.</p> <p>Patch 1 fixes packed Rx response handling. The current PMD assumes every<br>response descriptor starts a new Free List buffer by requiring<br>F_RSPD_NEWBUF on each response. That assumption does not always hold for<br>packed ingress responses. Under sustained small-packet traffic to a<br>single ingress queue, the FL/IQ state goes out of sync and the affected<br>Rx path stops making forward progress.</p> <p>Patch 2 scopes rte_flow rules to the ingress port they were created on.<br>Chelsio filters are programmed in adapter-wide tables, and the PMD<br>already supports the iport field in the hardware filter spec. However,<br>the flow parser never fills it for normal per-port rules, so a rule<br>created on one port can also match traffic arriving on sibling ports of<br>the same adapter.</p> <p>Both issues reproduce with stock testpmd on T62100-LP-CR. The per-patch<br>commit messages include the details and reproducers.</p> <p>Abdulrahman Alshawi (2):<br>net/cxgbe: fix Rx handling for packed responses<br>net/cxgbe: restrict rte_flow rules to ingress port</p> <p>.mailmap | 1 +<br>drivers/net/cxgbe/base/adapter.h | 1 +<br>drivers/net/cxgbe/cxgbe_flow.c | 19 +++++<br>drivers/net/cxgbe/sge.c | 122 ++++++++++++++++++++++++-------<br>4 files changed, 118 insertions(+), 25 deletions(-)</p> <p>--<br>2.39.5</p> </div> </div> <div id="signature-content-no-signature" data-marker="__SIG_PRE__"></div> </div></body></html>