lkml.org 
[lkml]   [2000]   [Oct]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Weightless process class
On Wed, Oct 04, 2000 at 06:16:57PM -0300, Rik van Riel wrote:
> > > ---
> > > One way would be to set a flag "I'm holding a lock" and when
> > > it releases the lock(s), deschedule it?
> >
> > There is a well-known name for this -- priority inversion.
> >
> > Implement the whole shebang of starvation avoidance tricks,
> > and we can whak the scheduler to group processes into separate
> > subsets, which in current system leads to starvation lockups.
> >
> > A thing for 2.5 ? (With possible backport to 2.4 latter.)
>
> We don't need that.
>
> We just need one boolean per thread ... is it holding a kernel
> lock or not?

The BKL or *any* (kernel) lock ?

For my knowledge there is no limitation on how many
locks a thread can hold. Having a single bool might
not be enough. A counter is better ?

For example: Two filesystems are locking on their respective
superblocks, and then they are locking on some underlying IO
facility which both are sharing.

Should the lock-count be incremented before the lock acquisition
has happened, or only afterwards, that is other story.

> If it is, make sure its scheduling latency isn't too high.

e.g. all processes having *any* locks are raised to the highest
possible class to make sure they are not starved out ?

> If it isn't holding any lock, we can do with it what we want,
> including completely starving the task for several seconds
> (or even minutes) if scheduling latency or VM pressure warrants
> it.

Yes, that is obvious.

> regards,
>
> Rik
> http://www.conectiva.com/ http://www.surriel.com/

/Matti Aarnio
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 12:39    [W:0.079 / U:3.184 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site