Description of problem: Setting to anything <= 12 causes all network traffic to 'disappear', cannot ping, telnet, etc. Bringing the interface down, then setting to > 12 will correct the issue. Version-Release number of selected component (if applicable): RHEL 5.3 x86_64 (-155 from dzickus tested as well) How reproducible: Always Relevant code: static int bnx2x_set_coalesce(struct net_device *dev, struct ethtool_coalesce *coal) { struct bnx2x *bp = netdev_priv(dev); bp->rx_ticks = (u16) coal->rx_coalesce_usecs; if (bp->rx_ticks > 3000) bp->rx_ticks = 3000; ... if (netif_running(dev)) bnx2x_update_coalesce(bp); and static void bnx2x_update_coalesce(struct bnx2x *bp) { REG_WR8(bp, BAR_USTRORM_INTMEM + USTORM_SB_HC_TIMEOUT_OFFSET(port, sb_id, U_SB_ETH_RX_CQ_INDEX), bp->rx_ticks/12); Usually a code like that 'rx_ticks/12' means the board register accepts multiples of 12, so you can configure 0us, 12us, 24us, 36us, 48us, 60us... Upstream does that too. Broadcom driver does that too. I looks like a hardware limitation, though I don't have any detailed board specification to make sure. However, it should not stop working because someone tried to configure a value less or equal 12us.
Created attachment 350651 [details] patch fixing rx coalesce Hi, I've received a feedback from Broadcom on this. The FW supports coalescing in 12us granularity, and so value of less then 12 should be interpreted as disabling coalescing. Could you give a try on the attached patch? thank you, Flavio
I missed/forgot about that bug. It is duplicate (already solved). *** This bug has been marked as a duplicate of bug 522600 ***