| Message ID | 20240222162712.722914-1-maciej.fijalkowski@intel.com |
|---|---|
| State | RFC |
| Headers | show |
| Series | [RFC,iwl-net] ixgbe: do not account for XDP rings in Tx watchdog | expand |
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c index bd541527c8c7..083f7e75f67c 100644 --- a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c +++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c @@ -7624,13 +7624,6 @@ static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter) return true; } - for (i = 0; i < adapter->num_xdp_queues; i++) { - struct ixgbe_ring *ring = adapter->xdp_ring[i]; - - if (ring->next_to_use != ring->next_to_clean) - return true; - } - return false; } diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c index 59798bc33298..7d617a309967 100644 --- a/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c @@ -409,9 +409,6 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget) break; } - if (!netif_carrier_ok(xdp_ring->netdev)) - break; - if (!xsk_tx_peek_desc(pool, &desc)) break;
Fix issue where ixgbe driver triggers resets in endless manner. A snippet from dmesg: [1879966.583523] ixgbe 0000:86:00.0 enp134s0f0: initiating reset to clear Tx work after link loss [1879968.405694] ixgbe 0000:86:00.0 enp134s0f0: Reset adapter [1879968.925689] ixgbe 0000:86:00.0 enp134s0f0: detected SFP+: 3 [1879969.185504] ixgbe 0000:86:00.0 enp134s0f0: NIC Link is Up 10 Gbps, Flow Control: RX/TX [1879969.592433] ixgbe 0000:86:00.0 enp134s0f0: NIC Link is Down [1879969.592539] ixgbe 0000:86:00.0 enp134s0f0: initiating reset to clear Tx work after link loss [1879971.413253] ixgbe 0000:86:00.0 enp134s0f0: Reset adapter [1879971.901400] ixgbe 0000:86:00.0 enp134s0f0: detected SFP+: 3 [1879972.161224] ixgbe 0000:86:00.0 enp134s0f0: NIC Link is Up 10 Gbps, Flow Control: RX/TX [1879972.567922] ixgbe 0000:86:00.0 enp134s0f0: NIC Link is Down [1879972.568023] ixgbe 0000:86:00.0 enp134s0f0: initiating reset to clear Tx work after link loss [...] This can happen when user launches multiple xdpsock in txonly mode and flap the link few times. Commit c685c69fba71 ("ixgbe: don't do any AF_XDP zero-copy transmit if netif is not OK") tried to address same thing by calling netif_carrier_ok() in ZC Tx path, however as shown above I am still able to trigger reset loop. In watchdog subtask, when bringing link up, messages that are shown in dmesg are: Link is up Link is down <--- at this state, !netif_carrier_ok() xsk managed to produce tx desc, ixgbe_ring_tx_pending() will see ntc != ntu at XDP ring at will schedule reset Link is up I am not sure if sequence of events when bringing interface up is correct. I would expect that link would go just from downed state to up, not up/down/up. For now, do not look at XDP rings within watchdog. It is not common to do so. After that it is no longer needed to call netif_carrier_ok() in ZC Tx. Fixes: 33fdc82f0883 ("ixgbe: add support for XDP_TX action") Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> --- drivers/net/ethernet/intel/ixgbe/ixgbe_main.c | 7 ------- drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 3 --- 2 files changed, 10 deletions(-)