netdev
[Top] [All Lists]

Re: [PATCH] allow setting mtu and txqlen via RTM_SETLINK and provide txq

To: Thomas Graf <tgraf@xxxxxxx>
Subject: Re: [PATCH] allow setting mtu and txqlen via RTM_SETLINK and provide txqlen via RTM_GETLINK
From: "David S. Miller" <davem@xxxxxxxxxx>
Date: Tue, 24 Aug 2004 10:32:05 -0700
Cc: herbert@xxxxxxxxxxxxxxxxxxx, kuznet@xxxxxxxxxxxxx, hadi@xxxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <20040824120442.GA21296@postel.suug.ch>
References: <20040824094952.GF21073@postel.suug.ch> <E1BzYMJ-000208-00@gondolin.me.apana.org.au> <20040824120442.GA21296@postel.suug.ch>
Sender: netdev-bounce@xxxxxxxxxxx
On Tue, 24 Aug 2004 14:04:42 +0200
Thomas Graf <tgraf@xxxxxxx> wrote:

> * Herbert Xu <E1BzYMJ-000208-00@xxxxxxxxxxxxxxxxxxxxxxxx> 2004-08-24 20:16
> > Please think about the meaning of the value.  Is anyone going to have a 
> > queue
> > bigger than 2^32?
> > 
> > And if the answer is yes, then please use u64.
> 
> Can you please explain the actual reason for using a fixed width
> type when all existing numeric attributes use arch depedent types?

Because otherwise things will break and explode for 32-bit
binaries running on 64-bit kernels, which is the situation
for the majority of userland on some platforms.

We have a compatability layer to translate 32-bit system
call structures to/from 64-bit data types.  But such
a scheme does not work for things like netlink sockets
where it is the data stream via read/write calls that
contains the structures.

Also, we eventually want to be able to transmit netlink
messages over real networks to remote clients.

Therefore, like any other real network protocol, we should
use fixed sized types in the data.

Stop bickering and use a u32, or I'll make that change for
you :-)

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