A good Linux kernel vintage?

[Eli Billauer]

Nadav Har'El wrote:

> Well, basically, all released kernels fit your description. It's not
> like Linus, and a few hundreds of the kernel developers, are going to
> release a version, after a few weeks of intensive testing, which doesn't
> "work". You could be sure that all of them "love multitasking", and
> (of course) "compile a kernel", and also "check their mail" and "watch
> youtube videos". So for all this, *all* kernel versions should be fine.
>   
That should be true in theory. And since Oleg asked about my own 
experience, I'll explain myself a bit.

I started off with the distribution kernel just like anyone else, but 
upgraded to vanilla 2.6.35.4 to solve some driver issues (this is where 
normal people upgrade their entire distro, I know). At this point I 
discovered the disk freeze issue. And yes, I tried every scheduler 
available. This was a bug, which was fixed after it had been *known* for 
several kernel versions (2.6.32 to 2.6.35, IIRC). It's a known secret, 
that a reported bug can survive for quite a while in the kernel, because 
nobody takes the initiative to fix it. This was the case.

I summarized the disk freeze issue in my blog: 
http://billauer.co.il/blog/?p=759

But hey, since the bug was fixed on 2.6.36, I happily jumped on that 
kernel version. When booting, I couldn't login to my computer. As it 
turned out, it was a new kernel bug, which locked up all TTY devices, 
including the login/password entry. The thread I started in LKML about 
this can be seen at https://lkml.org/lkml/2010/10/30/16

As the discussion in the thread reveals, the problem was that the Big 
Kernel Lock had been removed from the TTY drivers, and replaced with an 
all-TTY lock. Unfortunately, the new lock was held during 30-second 
sleeps, which the maintainer didn't consider when making the lock 
reorganization. So anything related to TTY devices, including logins, 
was frozen for long periods of time.

So what should I do now? I could upgrade to the latest kernel again. I 
could also patch the existing one, which has been running smooth for a 
year, except for a couple of issues.

So how can you tell a kernel which doesn't have this kind of bugs? My 
thought was, that since both the complaints and their solution appear in 
LKML, someone who follows this mailing list tightly would know, in 
retrospective, which kernel had major issues, and which one didn't. So I 
hoped someone on this list could just name a kernel version and say 
"this one went through with nothing dramatic".

Since that didn't happen, I'll patch my kernel for the disk bug, try to 
upgrade the ALSA drivers specifically, and hopefully not break something 
on the way.

Thanks all for trying.
   Eli

-- 
Web: http://www.billauer.co.il