From owner-freebsd-hackers  Tue Nov 21 05:36:31 1995
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id FAA20778
          for hackers-outgoing; Tue, 21 Nov 1995 05:36:31 -0800
Received: from fun.inria.fr (fun.inria.fr [138.96.24.57])
          by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id FAA20768
          for <hackers@freebsd.org>; Tue, 21 Nov 1995 05:36:26 -0800
Received: by fun.inria.fr (8.6.12/8.6.12) id OAA18209; Tue, 21 Nov 1995 14:29:48 +0100
Message-Id: <199511211329.OAA18209@fun.inria.fr>
X-Mailer: exmh version 1.6.2 7/18/95
To: hackers@freebsd.org
Subject: Re: Notion of time in kernel + malloc + new comment
In-reply-to: your message of Mon, 20 Nov 1995 20:35:33 +0100.
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Tue, 21 Nov 1995 14:29:47 +0100
From: Andres Vega Garcia <Andres.Vega_Garcia@sophia.inria.fr>
Sender: owner-hackers@freebsd.org
Precedence: bulk

: Andres Vega Garcia <Andres.Vega_Garcia@sophia.inria.fr> wrote:
 
	OK, thank you for all your answers.
 
> I 
>need to have a notion of time (seconds (or tens of seconds) elapsed 
>since ...).

	I'm keeping the solution:

	#include <sys/kernel.h>

 	int s = splclock();
 	long  cur_time = time.tv_sec;
 	splx(s);

>
>	I have realized that in order to do malloc I need the "kernel" call 
>to malloc. As this is related with the ethernet drivers (and ppp), I 
>decided to use:

	For malloc, I do the way if_attach does (M_WAITOK):

	malloc(sizeof(fs_ctx_t),M_DEVBUF,M_WAITOK) for the per interface 
structures.

and 
	malloc(sizeof(fs_ses_t),M_DEVBUF,M_NOWAIT) for the dynamic 
srtuctures.

	=> The new comment:

	By now I'm replacing (some) calls to IF_ENQUEUE & IF_DEQUEUE, but I 
do that for every driver I want to modify. It would be very nice just 
to modify IF_ENQUEUE and IF_DEQUEUE and let the new policy take 
effect automatically in every driver, but:

	* I need to retrive a context information, which now, I get by 
indexing with `if_unit' and then calling the new function (e.g. 
fs_enqueue) with the pointer to the structure (the context) (as well 
as the old parameters).

	* I need to distinguish the different sessions (for IP, source and 
destination address and port), then, I need a lookup with the 
sockaddr structure. By now, that has to be done for every driver too.

	Do you see a way of not having to modify every driver?

	Does modifying IF_ENQUEUE and IF_DEQUEUE can have side effects for 
other code? Remember with fair queueing I want to give every session 
the same bandwidth, but what happens with drivers which control a 
device, where the notion of several conexions passing trough doesn't 
exist?

	Thank you in advance.

Andres

------------------------------
INRIA Sophia-Antipolis, France
      avega@fun.inria.fr