Peter Zijlstra: From DOS to kernel hacking

Peter: Around the time Win95 came around, IIRC [if I remember correctly]. I used to do demo coding on DOS, which involved rebooting your machine every time you messed up, and whereas DOS reboots quite quickly, doing the same on Win95 was anything but quick.

A friend of mine introduced me to Unix/Linux at the time, and I started learning all about programming in a real environment. Basically all programming up to that point was in a freestanding environment where you had to poke the hardware to get anything done.

So initially it was the charm of a proper multitasking OS (with memory protection) that got me to use it – not having to reboot your machine every time, and the luxury of being able to run a debugger.

Peter: The kernel ... well that took a seriously long while. The above introduction to Linux was around 95/96 IIRC. My first real kernel patches were about 10 years later.

In those 10 years I learnt a lot about programming. I learnt about Unix system programming, I learnt about C++, multi-threading, database engines, and a whole range of interesting things.

Somewhere along I got a real internet connection and started lurking on mailing lists, including LKML – I must have been reading that on and off for about 5 years by the time I really sat down and wrote some patches.

During that time I might have sent in some trivial build fixes, and I remember finding a priority leak in one of the realtime patches. But I wasn't actively coding on the kernel – I just liked running real exotic stuff, you know Gentoo and building just about everything from CVS.

So what got me started on the kernel ... I can't quite remember how it happened, but I ran into some of Rik's [van Riel] Advanced Page Replacement stuff. I had worked on that problem space earlier while doing database engines, and had recently run into it again at work. So I started reading those papers and some of the proposed kernel patches, and I started to itch.

I dropped basically everything I was working on in my spare time (hacking WindowMaker, writing a C++ ASN.1-DER serialization class, writing a new LDAP server and I'm sure some other projects that are rotting away on a harddrive somewhere :-) and started hacking.

Why ... I'm not sure – it sure got me back to where I started out – crashing machines (and boot times haven't improved over those past 10 years at all).

I think because of the challenge – I knew I could write whatever it was I was coding and this page replacement stuff was a whole new challenge, and TBH [to be honest] the kernel code didn't look too hard at the time (phew how ignorant I was..)

Peter: Some better than others. I think its natural for every kernel hacker to grow a huge pile of discarded patches. Not everything will make it. But don't get discouraged by that, you did get to learn something from doing them.

Mis-steps, feh, still do ;-) Unlike most people seem to think, kernel hackers are human too.

Peter: Just do it – seriously it's that easy. Oh and don't be afraid of criticism, you'll get it anyway – in spades. Criticism is not personal, it's about your patch, there are two things you can do:

• take it and act upon it
• convince the other he's wrong

OK it can get personal, but that is only if you repeatedly fail the above two points.

Peter: I'm not sure. The Kernel Janitors thing doesn't really seem to work out. I think that might be due to two things:

1. we don't have enough simple but interesting things lined up (not saying there are none, but we don't have a ready list). I think a proper challenging project would be much better that moronic code clean ups.
2. the kernel really isn't a place for newbies; now let me explain this before it gets all mis-interpreted :-)
• Things really get a lot easier if you're fairly competent at (Unix) system programming before starting at the kernel.
• Kernel hacking is a solitary business in that you need to do things, nobody is going to do them for you. That is not saying nobody can help you if you have a question. Also, nobody is going to force you to do something – you need to want doing it.

Now, none of this means you can't start hacking the kernel without knowing C or any programming it all, but you'd better be ready for one hell of a ride (Yes, there are people who learnt C from doing kernel stuff, but that is going to take a serious amount of will-power to pull off).

So I guess what I'm saying is that you need to really want to do it. There is no other way to become a kernel hacker than by simply doing it.

Peter: Both; initially it was spare time besides $JOB. But after keeping this up for about a year my wife nudged me to look for a kernel job, since I obviously enjoyed hacking the kernel more than $JOB, and she'd get some of that spare time back ;-)

So I applied for a kernel position at a few of the larger vendors, and Red Hat won the race.

Already having had a year's worth of exposure to kernel code and LKML, certainly helped in getting this amazing opportunity. Have I already mentioned I absolutely love working on the kernel?

So now I get to poke at the kernel all day, every day...

Peter: Current active projects are group scheduling and some -rt work. I should pick up the swap over network code again, and there are some other loose ends.

The future ... well we'll see what happens, loads of interesting stuff to do.