Merging SSB upstream

Merging SSB upstream

[Michael Buesch]

So, now that mac80211 is merged upstream, I think it's
time to merge SSB and the b44-ssb port upstream.
Note that bcm43xx-mac80211 is _not_ ready for upstream, yet.

What do you think? I'd like to merge ssb as-is, although
the embedded-device parts are not quite finished, yet.
But they don't interfere with the non-embedded parts used
by b44 and the bcm43xx PCI cards.
So we _could_ remove the ssb-mips code, but I don't like to
do that for better maintainability. It doesn't hurt anyone IMO.

Some opinions?

-- 
Greetings Michael.

Re: Merging SSB upstream

[John W. Linville]

On Sun, May 06, 2007 at 03:03:17AM +0200, Michael Buesch wrote:
> So, now that mac80211 is merged upstream, I think it's
> time to merge SSB and the b44-ssb port upstream.
> Note that bcm43xx-mac80211 is _not_ ready for upstream, yet.
 
ACK, unfortunately.

> What do you think? I'd like to merge ssb as-is, although
> the embedded-device parts are not quite finished, yet.
> But they don't interfere with the non-embedded parts used
> by b44 and the bcm43xx PCI cards.

How much testing have you (and others) done w/ b44?  I had to remove
the b44 ssb changes from fedora because a) users reported problems;
and b) I was more worried about wireless than b44+ssb. (sorry!)

So, has anyone been using b44 in -mm?

> So we _could_ remove the ssb-mips code, but I don't like to
> do that for better maintainability. It doesn't hurt anyone IMO.

I guess I don't see a problem w/ merging the mips part, as long as the
b44 part has been thoroughly tested.  I wonder if Ralf has an opinion?

John
-- 
John W. Linville
linville-2XuSBdqkA4R54TAoqtyWWQ@public.gmane.org

Re: Merging SSB upstream

[Michael Buesch]

On Sunday 06 May 2007 04:00:51 John W. Linville wrote:
> On Sun, May 06, 2007 at 03:03:17AM +0200, Michael Buesch wrote:
> > So, now that mac80211 is merged upstream, I think it's
> > time to merge SSB and the b44-ssb port upstream.
> > Note that bcm43xx-mac80211 is _not_ ready for upstream, yet.
>  
> ACK, unfortunately.
> 
> > What do you think? I'd like to merge ssb as-is, although
> > the embedded-device parts are not quite finished, yet.
> > But they don't interfere with the non-embedded parts used
> > by b44 and the bcm43xx PCI cards.
> 
> How much testing have you (and others) done w/ b44?

Well, it works for me with my b44 card I got from broadcom.
There are really only 2 or 3 different b44 cards, so chances are damn
high it will work for all.

> I had to remove 
> the b44 ssb changes from fedora because a) users reported problems;

Which problems were that? The 2 compile issues?
Trivial to fix if that's the only issue. ;)

-- 
Greetings Michael.

Re: Merging SSB upstream

[John W. Linville]

On Sun, May 06, 2007 at 11:44:27AM +0200, Michael Buesch wrote:
> On Sunday 06 May 2007 04:00:51 John W. Linville wrote:

> > I had to remove 
> > the b44 ssb changes from fedora because a) users reported problems;
> 
> Which problems were that? The 2 compile issues?
> Trivial to fix if that's the only issue. ;)

I knew you would ask that... :-)

I don't think there was a bugzilla, but Dave Jones forwarded an email
to me from "MASAO TAKAHASHI" in late February.  Takahashi-san (forgive
me if I did that wrong) was complaining about tx timeouts after I
had added the full wireless-dev patchset to rawhide.  Removing the
b44 bits of the patch seemed to remove the problem.

That's all the info I have.  Perhaps Dave or Takahashi-san can add
to the description?

John
-- 
John W. Linville
linville@tuxdriver.com

Re: Merging SSB upstream

[Michael Buesch]

On Sunday 06 May 2007 19:38:33 John W. Linville wrote:
> On Sun, May 06, 2007 at 11:44:27AM +0200, Michael Buesch wrote:
> > On Sunday 06 May 2007 04:00:51 John W. Linville wrote:
> 
> > > I had to remove 
> > > the b44 ssb changes from fedora because a) users reported problems;
> > 
> > Which problems were that? The 2 compile issues?
> > Trivial to fix if that's the only issue. ;)
> 
> I knew you would ask that... :-)

:P

> I don't think there was a bugzilla, but Dave Jones forwarded an email
> to me from "MASAO TAKAHASHI" in late February.  Takahashi-san (forgive
> me if I did that wrong) was complaining about tx timeouts after I
> had added the full wireless-dev patchset to rawhide.  Removing the
> b44 bits of the patch seemed to remove the problem.
> 
> That's all the info I have.  Perhaps Dave or Takahashi-san can add
> to the description?

Hm, interesting issue.
But I'm not convinced that it's caused by the SSB port, though.
What the SSB port essentially does is modifying small areas in the
init and exit paths. The modified things in the TX and RX hotpaths
are really tiny and trivial. TX timeout sounds like something in the
TX/RX paths is going wrong.
But anyway, maybe I got something wrong.
I'll run some burn-in tests on it now.

-- 
Greetings Michael.

Re: Merging SSB upstream

[Jeff Garzik]

Michael Buesch wrote:
> So, now that mac80211 is merged upstream, I think it's
> time to merge SSB and the b44-ssb port upstream.
> Note that bcm43xx-mac80211 is _not_ ready for upstream, yet.
> 
> What do you think? I'd like to merge ssb as-is, although
> the embedded-device parts are not quite finished, yet.
> But they don't interfere with the non-embedded parts used
> by b44 and the bcm43xx PCI cards.
> So we _could_ remove the ssb-mips code, but I don't like to
> do that for better maintainability. It doesn't hurt anyone IMO.

What does Ralf (MIPS maintainer) and Gary (Broadcom maintainer) think?

For my part, I'm not going to render even a tentative opinion without a 
link to actual code.

Last I saw of the code, and descriptions in IRC, it sounded sane.

	Jeff

Re: Merging SSB upstream

[Michael Buesch]

On Sunday 06 May 2007 05:11:44 Jeff Garzik wrote:
> Michael Buesch wrote:
> > So, now that mac80211 is merged upstream, I think it's
> > time to merge SSB and the b44-ssb port upstream.
> > Note that bcm43xx-mac80211 is _not_ ready for upstream, yet.
> > 
> > What do you think? I'd like to merge ssb as-is, although
> > the embedded-device parts are not quite finished, yet.
> > But they don't interfere with the non-embedded parts used
> > by b44 and the bcm43xx PCI cards.
> > So we _could_ remove the ssb-mips code, but I don't like to
> > do that for better maintainability. It doesn't hurt anyone IMO.
> 
> What does Ralf (MIPS maintainer) and Gary (Broadcom maintainer) think?

I don't know. I wanted to get all these opinions with this mail. :)

> For my part, I'm not going to render even a tentative opinion without a 
> link to actual code.

All code is in wireless tree drivers/ssb
http://bu3sch.de/gitweb?p=wireless-dev.git;a=tree;f=drivers/ssb;h=681cd93bf166670efbdec471b78137d3d0f26537;hb=HEAD

-- 
Greetings Michael.

Re: Merging SSB upstream

[Gary Zambrano]

On Sat, 2007-05-05 at 23:11 -0400, Jeff Garzik wrote:
> Michael Buesch wrote:
> > So, now that mac80211 is merged upstream, I think it's
> > time to merge SSB and the b44-ssb port upstream.
> > Note that bcm43xx-mac80211 is _not_ ready for upstream, yet.
> > 
> > What do you think? I'd like to merge ssb as-is, although
> > the embedded-device parts are not quite finished, yet.
> > But they don't interfere with the non-embedded parts used
> > by b44 and the bcm43xx PCI cards.
> > So we _could_ remove the ssb-mips code, but I don't like to
> > do that for better maintainability. It doesn't hurt anyone IMO.
> 
> What does Ralf (MIPS maintainer) and Gary (Broadcom maintainer) think?
> 
> For my part, I'm not going to render even a tentative opinion without a 
> link to actual code.
> 
> Last I saw of the code, and descriptions in IRC, it sounded sane.
> 
> 	Jeff
> 

I would like to put some test mileage behind the ssb.
We had a hard time testing it a while back, so we will try the latest.
Thanks,
Gary

Re: Merging SSB upstream

[Michael Buesch]

On Monday 07 May 2007 18:43:18 Gary Zambrano wrote:
> On Sat, 2007-05-05 at 23:11 -0400, Jeff Garzik wrote:
> > Michael Buesch wrote:
> > > So, now that mac80211 is merged upstream, I think it's
> > > time to merge SSB and the b44-ssb port upstream.
> > > Note that bcm43xx-mac80211 is _not_ ready for upstream, yet.
> > > 
> > > What do you think? I'd like to merge ssb as-is, although
> > > the embedded-device parts are not quite finished, yet.
> > > But they don't interfere with the non-embedded parts used
> > > by b44 and the bcm43xx PCI cards.
> > > So we _could_ remove the ssb-mips code, but I don't like to
> > > do that for better maintainability. It doesn't hurt anyone IMO.
> > 
> > What does Ralf (MIPS maintainer) and Gary (Broadcom maintainer) think?
> > 
> > For my part, I'm not going to render even a tentative opinion without a 
> > link to actual code.
> > 
> > Last I saw of the code, and descriptions in IRC, it sounded sane.
> > 
> > 	Jeff
> > 
> 
> I would like to put some test mileage behind the ssb.
> We had a hard time testing it a while back, so we will try the latest.

Ok, nice to hear. :)
I stresstested latest ssb on my b44 card and it works fine here.
No TX timeouts or something.

-- 
Greetings Michael.