tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Kernel preemption

Hubert Feyrer wrote:
On Sat, 26 Apr 2008, Andrew Doran wrote:
- Adds code for the FPU stuff.

I admit I'm too dumb to understand the parts of the patch that covers this, but I'm still curious: IIRC FPU state is saved when entering from userland to kernel, and restored when going back. Also, IIUC no FPU code is used (allowed?) inside the kernel, right?

In what context/scenario does the FPU state need handling, if the above assumptions are correct?

FPU context switching is done when needed. Not every process uses the FPU. On prozess switching, the FPU gets disabled. The next FPU
instruction arises an FPU exception. The kernel then does the FPU
context switching and turns it on again.


 - Hubert

P.S.: On making the kernel preemptible: AWESOME!

That's the start to realtime-handling, not the end. To fullfill realtime
demands, we need to do some other things:

- allow to pin a process to a CPU
- guarantee scheduling time for a process
  at certain timepoints and/or time intervalls
- guarantee access to certain hw resources in a certain time
- guarantee syscall to finish in a certain time
- some more points I can't think of at the moment :)

Home | Main Index | Thread Index | Old Index