netdev
[Top] [All Lists]

Re: [PATCH 2.6]: Make packet scheduler clock source configurable

To: "David S. Miller" <davem@xxxxxxxxxx>
Subject: Re: [PATCH 2.6]: Make packet scheduler clock source configurable
From: Patrick McHardy <kaber@xxxxxxxxx>
Date: Fri, 23 Jul 2004 01:36:22 +0200
Cc: shemminger@xxxxxxxx, netdev@xxxxxxxxxxx, devik@xxxxxx
In-reply-to: <20040721143110.4ab944bf.davem@redhat.com>
References: <40F34740.5040100@trash.net> <1107.63.170.215.71.1089689716.squirrel@www.osdl.org> <20040712205037.573411c0.davem@redhat.com> <40F4862D.3070802@trash.net> <40F4AC8B.40706@trash.net> <20040721143110.4ab944bf.davem@redhat.com>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.6) Gecko/20040413 Debian/1.6-5
David S. Miller wrote:
This looks great.

As you mention some platforms return zero, for example sparc32,
for get_cycles().

I suggest we just expand the dependency list for NET_SCH_CLK_TSC
to include SPARC64 PPC64 and perhaps some other easy to verify
as having a working get_cycles() implementation.  I believe that
as long as it increments at some rate >= jiffies, the psched
calibration will get things into a working state.

It needs to increment at slightly above 1Mhz, otherwise delay will be zero after this division and everything will fall apart: delay /= rdelay.

A lot of patches have been posted in this area and I'm losing
track of what to apply first etc.  Can you repost your work
one change at a time?  Thanks.

The following two patches remove some dead timer code and change PSCHED_GET_TIME to use get_cycles. I'm going to send the configurable clock-source patch tomorrow after checking the arches.

Regards
Patrick


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