1 2Network Devices, the Kernel, and You! 3 4 5Introduction 6============ 7The following is a random collection of documentation regarding 8network devices. 9 10struct net_device allocation rules 11================================== 12Network device structures need to persist even after module is unloaded and 13must be allocated with alloc_netdev_mqs() and friends. 14If device has registered successfully, it will be freed on last use 15by free_netdev(). This is required to handle the pathologic case cleanly 16(example: rmmod mydriver </sys/class/net/myeth/mtu ) 17 18alloc_netdev_mqs()/alloc_netdev() reserve extra space for driver 19private data which gets freed when the network device is freed. If 20separately allocated data is attached to the network device 21(netdev_priv(dev)) then it is up to the module exit handler to free that. 22 23MTU 24=== 25Each network device has a Maximum Transfer Unit. The MTU does not 26include any link layer protocol overhead. Upper layer protocols must 27not pass a socket buffer (skb) to a device to transmit with more data 28than the mtu. The MTU does not include link layer header overhead, so 29for example on Ethernet if the standard MTU is 1500 bytes used, the 30actual skb will contain up to 1514 bytes because of the Ethernet 31header. Devices should allow for the 4 byte VLAN header as well. 32 33Segmentation Offload (GSO, TSO) is an exception to this rule. The 34upper layer protocol may pass a large socket buffer to the device 35transmit routine, and the device will break that up into separate 36packets based on the current MTU. 37 38MTU is symmetrical and applies both to receive and transmit. A device 39must be able to receive at least the maximum size packet allowed by 40the MTU. A network device may use the MTU as mechanism to size receive 41buffers, but the device should allow packets with VLAN header. With 42standard Ethernet mtu of 1500 bytes, the device should allow up to 431518 byte packets (1500 + 14 header + 4 tag). The device may either: 44drop, truncate, or pass up oversize packets, but dropping oversize 45packets is preferred. 46 47 48struct net_device synchronization rules 49======================================= 50ndo_open: 51 Synchronization: rtnl_lock() semaphore. 52 Context: process 53 54ndo_stop: 55 Synchronization: rtnl_lock() semaphore. 56 Context: process 57 Note: netif_running() is guaranteed false 58 59ndo_do_ioctl: 60 Synchronization: rtnl_lock() semaphore. 61 Context: process 62 63ndo_get_stats: 64 Synchronization: dev_base_lock rwlock. 65 Context: nominally process, but don't sleep inside an rwlock 66 67ndo_start_xmit: 68 Synchronization: __netif_tx_lock spinlock. 69 70 When the driver sets NETIF_F_LLTX in dev->features this will be 71 called without holding netif_tx_lock. In this case the driver 72 has to lock by itself when needed. 73 The locking there should also properly protect against 74 set_rx_mode. WARNING: use of NETIF_F_LLTX is deprecated. 75 Don't use it for new drivers. 76 77 Context: Process with BHs disabled or BH (timer), 78 will be called with interrupts disabled by netconsole. 79 80 Return codes: 81 o NETDEV_TX_OK everything ok. 82 o NETDEV_TX_BUSY Cannot transmit packet, try later 83 Usually a bug, means queue start/stop flow control is broken in 84 the driver. Note: the driver must NOT put the skb in its DMA ring. 85 86ndo_tx_timeout: 87 Synchronization: netif_tx_lock spinlock; all TX queues frozen. 88 Context: BHs disabled 89 Notes: netif_queue_stopped() is guaranteed true 90 91ndo_set_rx_mode: 92 Synchronization: netif_addr_lock spinlock. 93 Context: BHs disabled 94 95struct napi_struct synchronization rules 96======================================== 97napi->poll: 98 Synchronization: NAPI_STATE_SCHED bit in napi->state. Device 99 driver's ndo_stop method will invoke napi_disable() on 100 all NAPI instances which will do a sleeping poll on the 101 NAPI_STATE_SCHED napi->state bit, waiting for all pending 102 NAPI activity to cease. 103 Context: softirq 104 will be called with interrupts disabled by netconsole. 105
