Re: netif_tx_disable and lockless TX

[jamal <hadi@cyberus.ca>]

On Wed, 2006-31-05 at 19:52 +0200, Robert Olsson wrote:
> jamal writes:
> 
>  > Latency-wise: TX completion interrupt provides the best latency.
>  > Processing in the poll() -aka softirq- was almost close to the hardirq
>  > variant. So if you can make things run in a softirq such as transmit
>  > one, then the numbers will likely stay the same.
>  
>  I don't remember we tried tasklet for TX a la Herbert's suggestion but we 
>  used use tasklets for controlling RX processing to avoid hardirq livelock
>  in pre-NAPI versions.
> 

Hrm - it may have been a private thing i did then. I could swear we did
that experiment together ...
Perhaps Herbert's motivation was not really to optimize but rather to
get something unstuck in the transmit path state machine maybe in a
context of netconsole? The conditions for which that tasklet would even
run require a CPU collision to the transmit. Sorry, I didnt quiet follow
the motivation/discussion that ended in that patch.

>  Had variants of tulip driver with both TX cleaning at ->poll and TX
>  cleaning at hardirq and didn't see any performance difference. The 
>  ->poll was much cleaner but we kept Alexey's original work for tulip.
> 

It certainly is cleaner - but i do recall the hardirq variant had better
latency much observable under high packet rates aka small packets. 

>  > Sorry, I havent been following discussions on netchannels[1] so i am not
>  > qualified to comment on the "replacement" part Dave mentioned earlier.
>  > What I can say is the tx processing doesnt have to be part of the NAPI
>  > poll() and still use hardirq.
> 
>  Yes true but I see TX numbers with newer boards (wire rate small pakets)
>  with cleaing in ->poll. Also now linux is very safe in network "overload" 
>  situations. Moving work to hardirq may change that.
> 

Oh, I am not suggesting a change - i am a lot more conservative than
that ;-> these areas are delicate (not code-delicate Acme ;->) but
rather what seems obvious requires a lot of experimental results first.

Robert, your transmit results Intel or AMD based?

cheers,
jamal