[Top] [All Lists]

Re: [PATCH] (4/25) sk98: change #define to typedef

To: Mirko Lindner <mlindner@xxxxxxxxxxxxx>
Subject: Re: [PATCH] (4/25) sk98: change #define to typedef
From: Stephen Hemminger <shemminger@xxxxxxxx>
Date: Tue, 16 Nov 2004 09:12:51 -0800
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, Jeff Garzik <jgarzik@xxxxxxxxx>, Mirko Lindner <demon@xxxxxxxxxxxx>, netdev@xxxxxxxxxxx, Ralph Roesler <rroesler@xxxxxxxxxxxxx>
In-reply-to: <>
Organization: Open Source Development Lab
References: <>
Sender: netdev-bounce@xxxxxxxxxxx
On Tue, 16 Nov 2004 15:56:15 +0000
Mirko Lindner <mlindner@xxxxxxxxxxxxx> wrote:

>  > I don't think the driver is complex enough for all that os-common mess.
>  > If you look at the BSD sk driver it's about a third of the size of the
>  > Linux driver because it doesn't have all this mess.
> Note that the kernel BSD sk driver was _not_ written by SysKonnect and 
> only supports Genesis and Yukon1; linux supports Genesis (single and 
> dual), Yukon1, Yukon Plus, Yukon EC, Yukon FE and Yukon2 (single and 
> dual). The BSD sk driver supports substantially less functionality than 
> the Linux driver. For example, there is no link failover capability in 
> the sk driver; it has no power management; numerous chip changes and 
> minor tweaks have not been added, since many problems have become 
> apparent through our other supported platforms (some 15 operating 
> systems) and have since been fixed in our common modules; sk has no 
> wake-on-LAN and so on. The official SysKonnect BSD driver (yk) also uses 
> the common modules and is not open source.
> However, the real point is that the Linux driver (as indeed _all_ of our 
> drivers) shares common code with all our supported platforms and this 
> necessarily introduces some overhead. But it isn't excessive, and the 
> benefits _far_ outweigh the drawbacks, namely, that we have a very wide 
> range of systems sharing common code and thus the probability of finding 
> and fixing bugs is very substantially higher than if we wrote a 
> dedicated driver for each platform.
> For instance the symbol SK_IOC mentioned in your mail:
>  > -#define SK_IOC char __iomem *
>  > +typedef void __iomem *SK_IOC;
> is used in a large number of OS independent driver files (e.g. skvpd.c).
> The -for instance- file skvpd.c used by our Linux driver is used
> at the same time without any changes by all our other drivers running
> on other operating systems (e.g. Solaris, Windows).
> This is the reason why we need to redefine this symbol to the Linux
> specific implementation (void __iomem).

I understand the benefits of a common driver, but there are costs too.
First, it means the community can't really support or fix the common code
only SysKonnect can. Also, there is functionality in the common part that
really isn't necessary:
        * Vendor specific MIB support -- what tool could/does use that?
        * Board specific bridging support
If the regular Linux infrastructure was used for these, then the
driver would get additional kernel bridge filtering, common management and 
security fixes.

My wish is that the common hardware layer could be used without dragging
in all the MIB and bridging stuff.

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