net_rx_action() query in dev.c?

[pradeep singh]

Hi All,

I was going through the implementation code for net_rx_action, whcih
is invoked everytime the NET_RX_SOFTIRQ gets to run.

The logic at line number 1916 in file net/core/dev.c is -

while(!list_empty(&queue->poll_list)) {
/*something here*/
if(budget <= 0 || jiffies - start_time > 1) <- line 1916
goto softnet_break;

/*something here*/
}

if budget <= 0 we must go to softnet_break is understandable but why
if budget is > 0 we must check for the time elapsed??
And if the difference is > 1, why we should goto softnet_break?

I didn't quite get it.

Any help?

Thanks
~psr

--
play the game

--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to eca...@nl.linux.org
Please read the FAQ at http://kernelnewbies.org/FAQ

[Ajay Singh (ajaysi)]

> -----Original Message-----
> From: kernelnewb...@nl.linux.org
> [mailto:kernelnewb...@nl.linux.org] On Behalf Of pradeep singh
> Sent: Wednesday, April 25, 2007 3:31 PM
> To: kernel newbies
> Subject: net_rx_action() query in dev.c?
>
> Hi All,
>
> I was going through the implementation code for
> net_rx_action, whcih is invoked everytime the NET_RX_SOFTIRQ
> gets to run.
>
> The logic at line number 1916 in file net/core/dev.c is -
>
> while(!list_empty(&queue->poll_list)) {
> /*something here*/
> if(budget <= 0 || jiffies - start_time > 1) <- line 1916
> goto softnet_break;
>
> /*something here*/
> }
>
> if budget <= 0 we must go to softnet_break is understandable
> but why if budget is > 0 we must check for the time elapsed??
> And if the difference is > 1, why we should goto softnet_break?
>

The loop is getting only '1 jiffy' for processing the packets and the
maximum number of packets it can process is netdev_budget (which is
assigned value of 300).
So if either of the condition is true it breaks the loop and queues the
softirq again.

[pradeep singh]

Ok, got it.
Thanks a lot ajay :-)

~psr