[Top] [All Lists]

Re: Queue and SMP locking discussion (was Re: 3c59x.c)

To: becker@xxxxxxxxx (Donald Becker)
Subject: Re: Queue and SMP locking discussion (was Re: 3c59x.c)
From: kuznet@xxxxxxxxxxxxx
Date: Fri, 31 Mar 2000 22:46:15 +0400 (MSK DST)
Cc: andrewm@xxxxxxxxxx, hadi@xxxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <Pine.LNX.4.10.10003302258480.2499-100000@xxxxxxxxxxxxx> from "Donald Becker" at Mar 31, 0 10:08:00 am
Sender: owner-netdev@xxxxxxxxxxx

> Acckk!!! I just saw that someone put netif_wake_queue() in the normal path
> of the 3c59x.c Tx routine!  This is BAD.  That is putting an expensive call
> in the critical path, and it's not even the right semantics.

Well, do not stop queue, then there will be no reasons to wake it.
If we stop queue, we have to wake it.

> Most machines could never see a regime where they are overwhelmed by just
> accepting incoming packets.  In the situation where it occurs, usually only
> gigabit cards or multiple 100baseTx connections, there must be
> discard/ignore policy. 

Khm... Please, get some simple benchmark applet sort of netperf
and enjoy with this impossible phenomenon on single 100Mbit interface.
Despite of all the "max job on interrupt" linux-2.2 never leaves irq handler
and does no job in result. BSD (and NT, by a strange reason)
with their silly approach _work_ at any load level, by the way.


<Prev in Thread] Current Thread [Next in Thread>