From mboxrd@z Thu Jan 1 00:00:00 1970 From: Len Brown Subject: Re: Re: [PATCHES] add _CST support Date: 23 Dec 2004 00:30:55 -0500 Message-ID: <1103779855.2577.76.camel@d845pe> References: <20041127215118.GA30309@dominikbrodowski.de> <1103698332.17318.170.camel@d845pe> <1103747852.17331.237.camel@d845pe> <20041223001830.GC8289@dominikbrodowski.de> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20041223001830.GC8289-X3ehHDuj6sIIGcDfoQAp7BvVK+yQ3ZXh@public.gmane.org> Sender: acpi-devel-admin-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org Errors-To: acpi-devel-admin-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , List-Archive: To: Dominik Brodowski Cc: ACPI Developers , Robert Moore , Bruno Ducrot List-Id: linux-acpi@vger.kernel.org On Wed, 2004-12-22 at 19:18, Dominik Brodowski wrote: > On Wed, Dec 22, 2004 at 03:37:32PM -0500, Len Brown wrote: > > I removed the D600 from the port replicator, and now > > it is able to run in C4 w/o any hacks: > Glad to hear that. Was the port replicator causing some sort of > periodic bus master activity which keeps it out of getting to C4 > sleep? Yes, bus-master activity went from always on to sometimes on. Maybe there is a USB hub in the port replicator... > > > So a list of things we need comes to mind: > - wait and see if issues appear with the split-up and _CST addition I'll push this into 2.6.11 as soon as it opens up, so it should get broad exposure quite quickly, and quite early in the release cycle. > - > Enable > C1 on SMP > ... this I'd like to delay a short bit until we make the idle state > algorithm a bit smarter. Bob and Venki prototyped updated algorithms a while back. I'm hopeful that they can port that work to the new code and it can make early 2.6.11 also. > - > max_cstate able to disable C4 w/o disabling C3 > ... see other post, suggesting max_idle_state instead > - > "nocst" flag to compare FADT method w/ _CST > - > bm_check code fix -- currently has no clue of HZ > > a 100HZ kernel gets into C4 less than > > a 1000HZ kernel, likely because of this. > ... Yes, this is a bug IMHO. I think the bug is that our sampling of BM activity ages 10x faster in the 1000HZ case. > Also, if the idle loop hasn't been entered for one > (or two? or hundred?) ticks, I think it should re-start at C1-type > sleep. Agreed. The current promote/demote code looks at idleness when it should look at busy-ness and system load. > - > tick-less idle loop > - _CSD support (depends on _OSC/_PDC framework) We need to handle SMP C-states even if the system doesn't supply us with _CSD. cheers, -Len ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://productguide.itmanagersjournal.com/