Wary: uniprocessor kernel?

Wary is targeting older hardware, most of which are uniprocessor systems. An SMP-enabled (multi-core CPU) kernel does not work on all old uniprocessor CPUs, even when the appropriate kernel boot command to turn off SMP is used -- I was reminded of this recently in a pm from someone.

The thing is, even if a uniprocessor kernel is run on a multi-core CPU system, I don't think that users would notice the difference in speed. Except for some multimedia uses such as playing videos. But even then, playing a DVD or whatever, you are going to get sufficient Frames Per Second running on one core, as evidenced by DVDs that I have happily played on various old systems.

So, I am thinking of releasing the next Wary with a uniprocessor kernel, but perhaps I should "put a toe in the water first" and release it alongside the build with SMP kernel. Users can then compare.

I could even roll the kernel back to the old IDE pre-PATA drivers (yes I know that IDE and PATA mean much the same thing, but we use these terms to designate two different generations of drivers in the kernel. PATA and SATA are the new generation).


Posted on 26 Aug 2011, 8:55


Comments:

Posted on 26 Aug 2011, 15:04 by rodin.s
2.6.32
May be I have one computer on the planet that doesn't like 2.6.32 and 2.6.33. So I cannot fully use latest Wary or Lupu (hangs up when pluging in usb-device). I use Wary-511 with 2.6.31ide, Drake-0.3 or Slacko. So maybe if you are doing a build on alternative kernel let it be not 2.6.32(33). Although I can learn to use Woof and build it myself.


Posted on 26 Aug 2011, 16:54 by BarryK
USB hangs kernel
I have the same problem with my new laptop. The kernel segfaults one out of three times (approx., sometimes it happens first time) that I plug in a USB drive.

I was wondering if I was the only one!

My problem was one reason why I was thinking of compiling a uniprocessor kernel, in case the kernel has acquired a weird timing problem with multi-core handling. Even for the 2.6.33 kernel, configuring for uniprocessor would be going back to older stable code in the kernel source.

...let's try it!



Posted on 26 Aug 2011, 20:47 by BarryK
Uniprocessor kernel
Very interesting. I am running 2.6.32.45 configured as a uniprocessor kernel, and I have been replugging a USB pen drive many times, the kernel won't crash.



Posted on 26 Aug 2011, 21:01 by BarryK
Tickless kernel
There is another possibility. I configured this kernel with 'tickless' disable as well as uniprocessor. I need to do another compile, with SMP kernel but tickless disabled. Will do that now.



Posted on 27 Aug 2011, 3:33 by technosaurus
atoms & smp
the only ones I could see noticing would be those with intel atom or some of the new low power amd chips in the 800Mhz-1.6Ghz range with multiple cores/threads ... but it wouldn't be that noticable to most users (however I would notice it a _lot_ on my box when compiling large programs, but not so much in everyday use ... make j=2 vs j=8)


Posted on 27 Aug 2011, 7:08 by abushcrafter
Please Always Support Multiple Core
If I don't get the new Gecko Edu Book and you drop multiple core support than 1.5years of my life will be wasted :( and I won't have a set-up for my education :(.


Posted on 27 Aug 2011, 8:16 by BarryK
Want to capture crash dump
When the SMP kernel crashes, a dump appears on the screen (when X not running), but the system is then dead and I have no way of saving it for later examination. The screen dump is only showing the latter part of what gets dumped.

The KGDB kernel debugger was merged into the 2.6.26 kernel, known as "KGDB Light", but it needs an rs232 serial port and another computer to log what has happened. My laptop doesn't have a rs232 port.

There was something awhile back, LKCD, Linux Kernel Crash Dump, that looks like what I need, but the latest version of that is for the 2.6.9 kernel.



Posted on 28 Aug 2011, 13:43 by zygo
virtual crash
Barry,

could you boot the iso in a virtual machine that's similar to its physical host?