Subject: Re: Current status of rt2800usb and staging/rt2870
From: Dan Williams <dcbw@redhat.com>
To: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Cc: Ivo van Doorn <ivdoorn@gmail.com>,
	Ozan =?UTF-8?Q?=C3=87a=C4=9Flayan?= <ozan@pardus.org.tr>,
	linux-wireless@vger.kernel.org,
	linux-kernel <linux-kernel@vger.kernel.org>
In-Reply-To: <200910132121.25743.bzolnier@gmail.com>
References: <4AD46380.9020308@pardus.org.tr>
	 <200910131927.24136.bzolnier@gmail.com>
	 <200910132017.47748.IvDoorn@gmail.com>
	 <200910132121.25743.bzolnier@gmail.com>
Content-Type: text/plain
Date: Tue, 13 Oct 2009 13:15:40 -0700
Message-Id: <1255464940.10798.10.camel@localhost.localdomain>
Mime-Version: 1.0
Sender: linux-wireless-owner@vger.kernel.org

On Tue, 2009-10-13 at 21:21 +0200, Bartlomiej Zolnierkiewicz wrote:
> On Tuesday 13 October 2009 20:17:47 Ivo van Doorn wrote:
> > On Tuesday 13 October 2009, Bartlomiej Zolnierkiewicz wrote:
> > > On Tuesday 13 October 2009 17:52:52 Ivo van Doorn wrote:
> > > 
> > > > I am going to merge rt2800pci soon as well and after that I'll send a patch
> > > 
> > > Great, I've been waiting for this for a long time.
> > > 
> > > > to remove all Ralink drivers from the staging tree.
> > > 
> > > Till new drivers obtain support for all hardware currently covered by
> > > the unified staging drivers the latter shouldn't be removed as they
> > > serve well their role as a temporary solution allowing real Linux users
> > > to user their real hardware and as a reference material for developers
> > > willing to work on adding the missing bits to new drivers.
> > 
> > The rt2800pci and rt2800usb drivers cover support for rt2860/rt2870/rt3070 devices,
> > which are present as individual drivers in the staging tree. So far it only managed
> 
> There are no longer individual drivers.
> 
> Instead there is one shared source code and two device drivers:
> 
> - rt2860 covering RT2860 and RT3090 devices
> 
> - rt2870 covering RT2870 and RT3070 devices
> 
> In comparison rt2800usb currently lacks RT3070 support and rt2800pci's
> RT3090 support is basic at best (not to mention that it didn't work for
> RT2860 last time that I tried)..
> 
> > to confuse developers which wanted to work on Ralink support, so I haven't found
> 
> I bet they are truly grateful for being shown "the one and only way"..

There *is* only one way: mac80211.  If the driver is softmac, but does
not use mac82011, then it *is* the wrong way.  We will not have multiple
802.11 stacks in the mainstream kernel.  If a softmac driver uses
mac80211, hey, that's great!  Lets add it!

If it doesn't, then that driver needs to be updated to use mac80211.
Period.

> > a benefit for having those drivers in the staging tree yet. But I am sure there are plenty
> 
> I was able to use my Eee 901 successfully with them for a last year and I'm
> pretty sure that I'm not the only.  I understand my sin now -- I should have
> had help in fixing the proper and clean code before using my hardware.. [*]

That's nice, and everyone appreciates the work that you've done.  Nobody
can tell you what to work on, but it would be nice to get more people
working on the *mac80211* versions of the various drivers.  Imagine how
much better rt2x00 would be if all the effort that had gone into cleanup
up the ralink vendor drivers had instead gone into making rt2x00 work on
the new hardware.  Then you'd have cfg80211 support, background
scanning, etc all for free with no effort on your part.

Maybe it would have taken 6 months longer to get good support for your
hardware but at least the work would have a future.  But now we're stuck
in a situation where either:

1) the vendor driver never gets converted to cfg80211/mac80211, and
stays in staging forever, or finally gets booted out of staging because
nobody is working on converting it to mac80211.  mac80211/cfg80211 is a
hard requirement for being an accepted wireless driver.  And drivers do
have a shelf-life in staging.

2) the vendor driver starts getting converted to use mac80211.  But I
would argue that a better use of *anyone's* time is to make rt2x00
better, instead of porting the ralink vendor drivers to mac80211.  It
would probably take less time to fix up rt2x00 for new hardware than to
port the ralink vendor drivers to mac80211.

> > of people who like the idea of having the original Ralink drivers inside the staging tree,
> > so they can continue debugging that, so I won't talk about the staging downsides further.
> 
> Well, it would be much more productive if people concentrate on improving
> _their_ projects and looking at downsides of _their_ code instead of going
> around and looking at _obvious_ downsides of other people's projects.
> 
> 
> [*] [Slightly off-topic but it shows the same problem with the approach]
> 
> I now also see why some distributions keep pushing some known non-working
> things on their users (Fedora maintainers -- I'm talking to you), the best
> example is PulseAudio -- it simply doesn't work reliably even on modern
> hardware and/or using all scheduler tricks (it seems that making Linux into
> full RT OS is a prerequisite for fixing latency issues observed)..

Pulseaudio works for many, many people.  The most productive use of your
time here is to file a bug report with the specific details of your
audio chipset, let upstream diagnose whether it's a Pulseaudio-specific
problem, a codec driver problem, or an ALSA problem.

You can wait until the cows come home to deploy new technology, and then
of course you'll still have to fix up all the bugs because you waited
too long and nobody was testing it, or you can deploy new technology
that works for most people and fix those bugs much earlier.  Release
early, release often.  The tricky thing is how early.

Dan