Re: [Question] Use of CONFIG_X86_TSC in pkt_sched.h and profile.h

From: kuznet@ms2.inr.ac.ru
Date: Sun Jan 26 2003 - 17:39:26 EST


Hello!

> Alexey, what do you think about this?

Cleanup is really required. Actually, that code in pkt_sched.h looks so funny
exactly because kernel/&arch/ part used to be bizarre.

I can describe what we need exactly:

pkt_sched.h needs maximally cheap clock source, which will fallback
to jiffies when tsc are not available. So, I would expect raw 64bit cycle
counter, when it is available, or jiffies64 when it is not. do_gettimefoday()
cannot be used there. Also, on archs which have 32bit cc sort of alpha,
high 32 bit must be simulated by arch/.

What's about profile facilities, they simply do not make sense when
TSC is not available, so no questions arise at all, provided it compiles.

What's about skb->tstamp, it cannot use the same source as pkt_sched,
unless this clock is synchronized to do_gettimeofday() exactly.
If we want to optimize this, I see only radical solution,
sort of that which I used on i386: namely, getting rid
of SIOCGSTAMP and doing do_gettimeofday() at socket level immediately
before enqueueing to socket queue when SO_TIMESTAMP is enabled.

Alexey
-
To unsubscribe from this list: 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



This archive was generated by hypermail 2b29 : Fri Jan 31 2003 - 22:00:01 EST