Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 24 Jul 2010 22:42:41 +0300
From:      Alexander Motin <mav@FreeBSD.org>
To:        Doug Barton <dougb@FreeBSD.org>
Cc:        svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org
Subject:   Re: svn commit: r210444 - head/sys/x86/x86
Message-ID:  <4C4B4231.80406@FreeBSD.org>
In-Reply-To: <alpine.BSF.2.00.1007241224380.94574@qbhto.arg>
References:  <201007241049.o6OAnxvo001874@svn.freebsd.org> <alpine.BSF.2.00.1007241224380.94574@qbhto.arg>

next in thread | previous in thread | raw e-mail | index | archive | help
Doug Barton wrote:
> On Sat, 24 Jul 2010, Alexander Motin wrote:
>> Author: mav
>> Date: Sat Jul 24 10:49:59 2010
>> New Revision: 210444
>> URL: http://svn.freebsd.org/changeset/base/210444
>>
>> Log:
>>  Increment td->td_intr_nesting_level for LAPIC timer interrupts. Among
>> other
>>  things it hints SCHED_ULE to run clock swi handlers on their native
>> CPUs,
>>  avoiding many unneeded IPI_PREEMPT calls.
> 
> Will this help SCHED_4BSD at all? I'm finding it to be slightly better
> for interactivity on a loaded system, but frankly at this point they
> both leave a lot to be desired.

At least not directly. SCHED_4BSD doesn't use this variable. But I have
no idea if the problem exist there. Even on SCHED_ULE it happens only in
some cases, if swi thread was accidentally pushed out of it's CPU by
higher priority thread. 4BSD has different migration logic, so behavior
could be different there.

-- 
Alexander Motin



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4C4B4231.80406>