Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 4 Oct 2007 19:11:03 -0700 (PDT)
From:      Jeff Roberson <jroberson@chesapeake.net>
To:        Daniel Eischen <deischen@freebsd.org>
Cc:        freebsd-current@freebsd.org, "Gelsema, P \(Patrick\) - FreeBSD" <freebsd@superhero.nl>
Subject:   Re: ULE/yielding patch for testing.
Message-ID:  <20071004190854.I912@10.0.0.1>
In-Reply-To: <Pine.GSO.4.64.0710042127170.11107@sea.ntplx.net>
References:  <20071002165007.D587@10.0.0.1> <20071003110727.411aa2de@pleiades.nextvenue.com> <2155.10.202.77.103.1191443576.squirrel@webmail.superhero.nl> <20071004174044.E912@10.0.0.1> <Pine.GSO.4.64.0710042051500.11107@sea.ntplx.net> <20071004182539.H912@10.0.0.1> <Pine.GSO.4.64.0710042127170.11107@sea.ntplx.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 4 Oct 2007, Daniel Eischen wrote:

> On Thu, 4 Oct 2007, Jeff Roberson wrote:
>
>> On Thu, 4 Oct 2007, Daniel Eischen wrote:
>> 
>>> On Thu, 4 Oct 2007, Jeff Roberson wrote:
>>> 
>>>> 
>>>> I believe I have fixed this bug in the enclosed patch.  It is rooted from 
>>>> /usr/src/sys so you should cd there to apply it.
>>> 
>>> This doesn't break realtime threads doing a sched_yield() does
>>> it?  I couldn't easily see how the priority gets set back into
>>> the realtime class range.  But then, maybe I'm a dummy ;-)
>> 
>> Well the historical behavior was for sched_yield() to not adjust 
>> priorities. It just requeues at the back of the queue for that priority. Xu 
>> changed this in 7.0 but he didn't answer my mail as to why.  We have a 
>> yield() call that does drop to the max timeshare priority, however, it 
>> doesn't seem to have a man page.
>> 
>> The code removed was this:
>> 
>> -       if (td->td_pri_class == PRI_TIMESHARE)
>> -               sched_prio(td, PRI_MAX_TIMESHARE);
>> 
>> 
>> So it really only effected timesharing threads.
>
> As I read the change, now it affects real-time (which is the
> desired behavior since it is a POSIX real-time extension).
> But it should have POSIX-defined behavior, which is to requeue
> at the back of the queue for that priority.

It effected real-time before.  There is a mi_switch() after that priority 
adjustment.  That switch re-adds the thread at the back of the queue for 
that priority.  This patch just removes the priority adjustment for 
timeshare threads.

>
> -- 
> DE
> _______________________________________________
> freebsd-current@freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe@freebsd.org"
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20071004190854.I912>