>From: Neil Horman [mailto:[EMAIL PROTECTED] >Replying to myself... > I've looked through the driver pretty throughly with regards to my >above >concern, and it appears the driver is reasonably free of netpoll issues at >the >moment, at least as far as what we found in e1000 was concerned. I do
Thanks for reviewing the code.. >however, >see a concern in the use of the in_netpoll flag within the driver. Given >that >the primary registered net_device, and all the dummy net_devices in the >rx_ring >point to the same ixgbe_adapter structure, there can be some level of >confusion >over weather a given rx queue is in netpoll_mode or not. The revised driver I am going to post today will not have fake netdevs... >adapter prforms a netpoll, all the individual rx queues will follow the >in_netpoll path in the receive path (assuming misx interrupts are used). >The >result I think is the potential for a large amount of packet reordering >during a >netpoll operation. Perhaps not a serious problem, but likely worth looking Multiple Rx queues are used in non-NAPI mode only, and all Rx queues use one netdev (which is associated with the adapter struct). Also, the RSS (receive side scaling or rx packet steering) feature is used in multiple rx queues mode. In this mode, HW will always select the same Rx queue (for a flow) and this should prevent any packet reordering issue. >Neil Ayyappan - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
