Hi,
I still have a problem with the above mentioned subject, and am looking
for help how to debug this.
The setup and symptom:
- WLAN (802.11g) device in ad-hoc mode
- udp multicast stream sent out over WLAN device
- queue suddenly stops (no packets transmitted, and the socket's
queue as seen with "cat /proc/net/udp" is non-empty
- sending a single packet out (!) over the same interface (e.g.
an ARP packet by trying to ping a non-existent host on the subnet)
immediately resumes normal operation
Now my problem:
- I have only seen this problem with the device (driver) in question
- how can device queue be the problem if another packet _sent_ over
the same device resumes queue operation? I added printk's in the
driver to track the state of netif_{stop,wake}_queue, and even when
the (in-kernel) queue was stopped, the last printk was "netif_wake_queue".
Unfortunately, the driver is not (yet, they promised to release it) GPL,
so here is the structure of the transmit and irq path:
dev_eth_transmit(){
...
spin_lock_irqsave();
[queue the packet]
if(queue full){
netif_stop_queue(nwdev);
tx_full_flag = 1;
}
[trigger the hardware to send queued packet]
spin_lock_irqrestore();
}
dev_interrupt(){
spin_lock();
[handle interrupt sources]
if(tx_full_flag){
if(queue not full){
netif_wake_queue(nwdev);
tx_full_flag = 0;
}
}
spin_unlock();
}
So as far as I can see, there are no race condidtions.
Any hints on what could cause this, or how what I could do to further
track this down?
Regards,
Wolfgang
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
