On Sun, Jul 06, 2003 at 12:06:01AM +0300, Mark Veltzer wrote:

> In any case it should be quite easy to port drivers from 2.4 to 2.6 unless 
> the drivers are very complex and large (for instance - porting a small 
> communcation device may be very easy but porting an entire file
> system may be hard). 

This is a good time to mention lwn's driver porting series,
http://lwn.net/Articles/driver-porting/ 

> 2. The reason that driver API is broken is to *** IMPROVE *** the interaction 
> between the kernel and the drivers and make the API more fine grained so that 
> better use of the hardware would be possible and enable more fine grained 
> control of the hardware and better flow of data to and from the user level 
> applications and the hardware component. This situation is *** much better 
> *** than in the closed source world where the API between the kernel and the 
> drivers is rarely broken since the drivers are binary only drivers which are 
> written by the HW maker. The turn around time between changin driver API in 
> windows and getting to the full set of drivers you need is a couple of years 
> at best while in Linux it could be a couple of months. That is part of the 
> reason why linux is moving at such a rapid pace forward and windows has such 
> a hard time with stability.

Hear hear. 

> 3. The availability of the source code for all drivers is a big part of why 
> linux is more stable - this is because people can try and find innovative 
> ways to search for common driver oriented bugs (like the stanford checker for 
> instance). Since most of the code in the OS is drivers (this is true for 
> Linux and Windows) then most of the bugs that you encounter are actually 
> driver oriented bugs and not core system bugs. Software quality in the driver 
> code is what ultimately determines the stability of an operating system and 
> open source does it better.

Again, hear hear. 

> 4. As Muli said - user space should be no problem although I'm pretty sure you 
> will need a new version of glibc if you want to use new syscalls and 
> features. Muli ?!? 

If you want to take advantage of some of the new features, yes, you
need a new glibc and maybe some new user space utilities. Things would
still work flawlessly (just not optimally) with an old glibc. I run
the latest 2.5 kernels (2.5.74, currently) on an otherwise untouched
debian unstable system. 

-- 
Muli Ben-Yehuda
http://www.mulix.org
http://www.livejournal.com/~mulix/

Attachment: pgp00000.pgp
Description: PGP signature

Reply via email to