Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 29 Nov 2011 14:18:14 -0500
From:      Adam K Kirchhoff <akirchhoff135014@comcast.net>
To:        Jung-uk Kim <jkim@freebsd.org>
Cc:        freebsd-x11@freebsd.org
Subject:   Re: suggested xorg-compatible video HW for FreeBSD/amd64 ?
Message-ID:  <4ED52FF6.9070104@comcast.net>
In-Reply-To: <201111291412.28576.jkim@FreeBSD.org>
References:  <20111128092008.GA58668@onelab2.iet.unipi.it> <4ED52241.5040104@comcast.net> <4ED52648.9020908@comcast.net> <201111291412.28576.jkim@FreeBSD.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On 11/29/11 14:12, Jung-uk Kim wrote:
> On Tuesday 29 November 2011 01:36 pm, Adam K Kirchhoff wrote:
>> On 11/29/11 13:19, Adam K Kirchhoff wrote:
>>> On 11/29/11 13:04, Jung-uk Kim wrote:
>>>> I believe major hurdle is porting TTM but the future of this API
>>>> is not so bright.  In fact, X.org ATI driver uses GEM API now
>>>> and it is internally mapped to TTM calls by Linux DRM (aka
>>>> "GEM-ified TTM manager").  Unfortunately, as always, I don't see
>>>> clear plans from Linux/X.org developers.  I can only guess few
>>>> possibilities.
>>>>
>>>> 1. Linux/X.org folks drop GEM-ified TTM and use native GEM
>>>> calls. 2. Linux/X.org folks drop GEM-ified TTM and use native
>>>> TTM calls. 3. Linux/X.org folks re-invent new wheels (again).
>>>> 4. No change.
>>>>
>>>> My guess is #1 is most likely scenario in the near future.  Even
>>>> if Linux/X.org folks don't do it, we may be able to implement it
>>>> without TTM because X.org ATI driver uses GEM API already and we
>>>> do not have AMD/ATI Catalyst driver for FreeBSD anyway.  So, I
>>>> guess we have two choices ATM:
>>>>
>>>> 1. Fully porting TTM, GEM-ified TTM, and KMS.
>>>> 2. Replacing GEM-ified TTM with GEM and porting KMS.
>>>>
>>>> BTW, I am not volunteering. ;-)
>>>>
>>>> Jung-uk Kim
>>>> _______________________
>>> Every conversation I've had with the radeon driver developers on
>>> the matter, even quite recently, has led me to believe that TTM
>>> will not be going away.  GEM is only appropriate for IGP GPUs.
>>> Unless that changes within GEM, I do believe TTM will be used
>>> internally on the radeon DRM indefinitely.
>>>
>>> If I had to guess, I'd say that anyone on the FreeBSD side
>>> deciding to get rid of TTM and use GEM only GEM for radeons would
>>> eventually come to the same conclusion as the developers who have
>>> been working with radeon hardware for years :-)
>> Correction.  GEM seems to be focused on Intel IGP GPUs.  According
>> to one of the radeon developers on #radeon on freenode, even radeon
>> IGP GPUs need something like TTM.
>>
>> Honestly, I'm wondering how you came to the conclusion that future
>> of TTM is not so bright... ? :-)
> I know it was developed for Intel IGPs (by anholt@, former FreeBSD DRM
> maintainer).  However, I was under (wrong?) impression that
> xf86-video-ati is using GEM API because it wasn't absolutely
> necessary.  AFAIK, Nouveau folks also did something similar,
> OpenChrome guys are doing the same thing, etc.  If TTM is really
> necessary, why "GEM on TTM" hack in the first place?

It is my understanding that to simplify have one unified API for 
interacting with the DRM code, the radeon developers (and others) 
agreed/decided to use the GEM userspace API, even though the internals 
(for radeon DRM) require functionality provided by TTM.

For what it's worth, the radeon developer I just spoke to even said in 
order to remove TTM from the equation, "something" would have to be 
recoded to do "partly what ttm does".

Adam



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4ED52FF6.9070104>