patch-2.4.0-test11 linux/Documentation/networking/x25-iface.txt
Next file: linux/Documentation/usb/hotplug.txt
Previous file: linux/Documentation/networking/tulip.txt
Back to the patch index
Back to the overall index
- Lines: 63
- Date:
Sat Nov 11 19:02:39 2000
- Orig file:
v2.4.0-test10/linux/Documentation/networking/x25-iface.txt
- Orig date:
Tue Apr 28 14:22:04 1998
diff -u --recursive --new-file v2.4.0-test10/linux/Documentation/networking/x25-iface.txt linux/Documentation/networking/x25-iface.txt
@@ -62,3 +62,62 @@
First Byte = 0x03
LAPB parameters. To be defined.
+
+
+
+Possible Problems
+=================
+
+(Henner Eisen, 2000-10-28)
+
+The X.25 packet layer protocol depends on a reliable datalink service.
+The LAPB protocol provides such reliable service. But this reliability
+is not preserved by the Linux network device driver interface:
+
+- With Linux 2.4.x (and above) SMP kernels, packet ordering is not
+ preserved. Even if a device driver calls netif_rx(skb1) and later
+ netif_rx(skb2), skb2 might be delivered to the network layer
+ earlier that skb1.
+- Data passed upstream by means of netif_rx() might be dropped by the
+ kernel if the backlog queue is congested.
+
+The X.25 packet layer protocol will detect this and reset the virtual
+call in question. But many upper layer protocols are not designed to
+handle such N-Reset events gracefully. And frequent N-Reset events
+will always degrade performance.
+
+Thus, driver authors should make netif_rx() as reliable as possible:
+
+SMP re-ordering will not occur if the driver's interrupt handler is
+always executed on the same CPU. Thus,
+
+- Driver authors should use irq affinity for the interrupt handler.
+
+The probability of packet loss due to backlog congestion can be
+reduced by the following measures or a combination thereof:
+
+(1) Drivers for kernel versions 2.4.x and above should always check the
+ return value of netif_rx(). If it returns NET_RX_DROP, the
+ driver's LAPB protocol must not confirm reception of the frame
+ to the peer.
+ This will reliably suppress packet loss. The LAPB protocol will
+ automatically cause the peer to re-transmit the dropped packet
+ later.
+ The lapb module interface was modified to support this. Its
+ data_indication() method should now transparently pass the
+ netif_rx() return value to the (lapb mopdule) caller.
+(2) Drivers for kernel versions 2.2.x should always check the global
+ variable netdev_dropping when a new frame is received. The driver
+ should only call netif_rx() if netdev_dropping is zero. Otherwise
+ the driver should not confirm delivery of the frame and drop it.
+ Alternatively, the driver can queue the frame internally and call
+ netif_rx() later when netif_dropping is 0 again. In that case, delivery
+ confirmation should also be deferred such that the internal queue
+ cannot grow to much.
+ This will not reliably avoid packet loss, but the probability
+ of packet loss in netif_rx() path will be significantly reduced.
+(3) Additionally, driver authors might consider to support
+ CONFIG_NET_HW_FLOWCONTROL. This allows the driver to be woken up
+ when a previously congested backlog queue becomes empty again.
+ The driver could uses this for flow-controlling the peer by means
+ of the LAPB protocol's flow-control service.
FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen (who was at: slshen@lbl.gov)