BCM63xx merge progress

BCM63xx merge progress

[Hector Martin]

Hello, Maxime and Florian,

I saw that partial BCM63xx support has been merged into mainline. As I
am interested in developing a driver for this platform, I would like to
inquire as to the state of the remaining bits to merge. As it stands
now, the code merged into mainline is rather broken and incomplete.

I'll be happy to grab the code myself and get the patches in proper
merging order for mainline if nobody else is working / has time for it
at the moment, so I'm asking first.

I have a BCM63xx router and will be testing whatever I work on, both
compiling and running. My current reference is the OpenWRT patchset,
which is against an older kernel but seems to be working pretty well on
the actual hardware. I've also seen the linux-bcm63xx.git tree. Between
these I should be able to put together a decent set of commits that
cleanly apply onto current mainline.

Please let me know what your thoughts are, and whether I should proceed
and work on this myself or whether you have other plans.

-- 
Hector Martin (hector@marcansoft.com)
Public Key: http://www.marcansoft.com/marcan.asc

Re: BCM63xx merge progress

[Florian Fainelli]

Hello Hector,

Le mardi 1 décembre 2009 23:23:10, Hector Martin a écrit :
> Hello, Maxime and Florian,
> 
> I saw that partial BCM63xx support has been merged into mainline. As I
> am interested in developing a driver for this platform, I would like to
> inquire as to the state of the remaining bits to merge. As it stands
> now, the code merged into mainline is rather broken and incomplete.
> 
> I'll be happy to grab the code myself and get the patches in proper
> merging order for mainline if nobody else is working / has time for it
> at the moment, so I'm asking first.

Everything that Maxime sent is, or will be merged either in 2.6.32 or in 
2.6.33, that includes:
- arch/mips/bcm63xx (will be in 2.6.32)
- USB driver bits (2.6.33)
- Ethernet MAC driver bits (2.6.32)
- Ethernet PHY driver bits (2.6.32)
- UART driver (2.6.32)
- PCMCIA support (2.6.32)

> 
> I have a BCM63xx router and will be testing whatever I work on, both
> compiling and running. My current reference is the OpenWRT patchset,
> which is against an older kernel but seems to be working pretty well on
> the actual hardware. I've also seen the linux-bcm63xx.git tree. Between
> these I should be able to put together a decent set of commits that
> cleanly apply onto current mainline.

We maintain a couple of different patches for OpenWrt, specifically the mtd 
partition parser since we provide images, so the box should boot from Flash. 
That driver is not in a mergable state at the moment. There is also a embryon 
of a SPI driver and a watchdog driver, which I will probably submit once 
cleaned up.

> 
> Please let me know what your thoughts are, and whether I should proceed
> and work on this myself or whether you have other plans.
> 

My plans are currently to maintain the bcm63xx board code and drivers in a 
working state for the community, probably not much.
-- 
Best regards, Florian Fainelli
Email: florian@openwrt.org
Web: http://openwrt.org
IRC: [florian] on irc.freenode.net
-------------------------------

Re: BCM63xx merge progress

[Maxime Bizon]

On Tue, 2009-12-01 at 23:23 +0100, Hector Martin wrote:

Hi,

> I saw that partial BCM63xx support has been merged into mainline. As I
> am interested in developing a driver for this platform, I would like
> to inquire as to the state of the remaining bits to merge. As it
> stands now, the code merged into mainline is rather broken and
> incomplete. 

Nice & useful review, thanks ;)

Which part is broken/missing on your side ?

-- 
Maxime

Re: BCM63xx merge progress

[Hector Martin]

Maxime Bizon wrote:
> Which part is broken/missing on your side ?
> 

Well, there's a commit (4059ddb4) that claims to "Add USB OHCI support",
yet it only adds a #include to ohci-hcd.c and a few .h bits. The actual
ohci-bcm63xx.c is missing, which means the OHCI driver probably won't
even compile. It's also missing the relevant platform device stuff in
arch/. This concerned me somewhat, it looked like something weird had
happened and the merge was incomplete.

Florian Fainelli wrote:
> Everything that Maxime sent is, or will be merged either in 2.6.32 or
> in 2.6.33, that includes:
> - arch/mips/bcm63xx (will be in 2.6.32)
> - USB driver bits (2.6.33)
> - Ethernet MAC driver bits (2.6.32)
> - Ethernet PHY driver bits (2.6.32)
> - UART driver (2.6.32)
> - PCMCIA support (2.6.32)

I can see the 2.6.32 stuff already. Regarding USB, is there a repo with
the commit that will be in .33? Otherwise I'll just have to merge it
locally and work from there for the moment. What's up with the weirdo
commit mentioned above?

> We maintain a couple of different patches for OpenWrt, specifically
> the mtd partition parser since we provide images, so the box should
> boot from Flash. That driver is not in a mergable state at the moment.
> There is also a embryon of a SPI driver and a watchdog driver, which I
> will probably submit once cleaned up.

Another tidbit I saw in the OpenWRT set is the reset button GPIO
support. I want to have that (eventually) too.

Basically, my dilemma is simply that I want to work off of mainline
(especially since stuff has been partially merged already and changes
have been made), yet I need the missing patches in order to make this
work completely. So eventually I need a tree that contains everything I
need, whether a few individual commits are not really ready for mainline
or not. In doing this, I obviously want to keep in sync with any work
that has been done already as much as possible. So please do tell if
there's a specific place I should be looking at for mainline-compatible
versions of these patches, or anything closer to what will eventually be
merged. As I said, I don't mind doing the merges myself, but then it
might make rebasing onto mainline harder if your official patches are
merged in a substantially different way once they do make it to
mainline. And if I do end up merging stuff, I might as well send it to
you to save you some time.

-- 
Hector Martin (hector@marcansoft.com)
Public Key: http://www.marcansoft.com/marcan.asc

Re: BCM63xx merge progress

[Maxime Bizon]

On Wed, 2009-12-02 at 01:25 +0100, Hector Martin wrote:

> Well, there's a commit (4059ddb4) that claims to "Add USB OHCI support",
> yet it only adds a #include to ohci-hcd.c and a few .h bits. The actual
> ohci-bcm63xx.c is missing, which means the OHCI driver probably won't
> even compile. It's also missing the relevant platform device stuff in

I cannot find this commit in linus' tree nor linux-mips master, where
does it come from ?

The remaining two patches that were not merged in 2.6.32 are on top of
this tree: http://www.linux-mips.org/git?p=linux-bcm63xx.git

They should apply with little difficulty on current upstream kernel and
besides the *_be stuffs that need to be cleaned up, they were ACKed by
the usb maintainer.

-- 
Maxime

Re: BCM63xx merge progress

[Hector Martin]

Maxime Bizon wrote:
> On Wed, 2009-12-02 at 01:25 +0100, Hector Martin wrote:
> 
>> Well, there's a commit (4059ddb4) that claims to "Add USB OHCI support",
>> yet it only adds a #include to ohci-hcd.c and a few .h bits. The actual
>> ohci-bcm63xx.c is missing, which means the OHCI driver probably won't
>> even compile. It's also missing the relevant platform device stuff in
> 
> I cannot find this commit in linus' tree nor linux-mips master, where
> does it come from ?

Okay, nevermind, that was a screwup by me. At some point one of my
OpenWRT patch test commits made its way onto the wrong local branch, and
I didn't notice where it lived. Oops. (This commit was meant to be
incomplete, it comes straight from the OpenWRT patch). I think I found
the mips-bcm63xx (proper) patch, tried to search for it on my local
repo, and this thing popped up and I didn't look at it closely enough.
This is what I get for trying to work at 4am.

I take back the 'broken' then, we're just left with 'incomplete'. Makes
quite a bit more sense now, sorry for the confusion.

> The remaining two patches that were not merged in 2.6.32 are on top of
> this tree: http://www.linux-mips.org/git?p=linux-bcm63xx.git
> 
> They should apply with little difficulty on current upstream kernel and
> besides the *_be stuffs that need to be cleaned up, they were ACKed by
> the usb maintainer.

Cool, I'll see about applying them and working off of that. Thanks for
the confirmation.

FWIW, I'm trying to reverse engineer Broadcom's DSL driver, which I hope
will be useful so we can finally have a properly working OpenWRT
firmware for DSL routers based on this firmware.

-- 
Hector Martin (hector@marcansoft.com)
Public Key: http://www.marcansoft.com/marcan.asc

Re: BCM63xx merge progress

[Florian Fainelli]

On Wednesday 02 December 2009 03:45:59 Hector Martin wrote:
> Maxime Bizon wrote:
> > On Wed, 2009-12-02 at 01:25 +0100, Hector Martin wrote:
> >> Well, there's a commit (4059ddb4) that claims to "Add USB OHCI support",
> >> yet it only adds a #include to ohci-hcd.c and a few .h bits. The actual
> >> ohci-bcm63xx.c is missing, which means the OHCI driver probably won't
> >> even compile. It's also missing the relevant platform device stuff in
> >
> > I cannot find this commit in linus' tree nor linux-mips master, where
> > does it come from ?
> 
> Okay, nevermind, that was a screwup by me. At some point one of my
> OpenWRT patch test commits made its way onto the wrong local branch, and
> I didn't notice where it lived. Oops. (This commit was meant to be
> incomplete, it comes straight from the OpenWRT patch). I think I found
> the mips-bcm63xx (proper) patch, tried to search for it on my local
> repo, and this thing popped up and I didn't look at it closely enough.
> This is what I get for trying to work at 4am.
> 
> I take back the 'broken' then, we're just left with 'incomplete'. Makes
> quite a bit more sense now, sorry for the confusion.
> 
> > The remaining two patches that were not merged in 2.6.32 are on top of
> > this tree: http://www.linux-mips.org/git?p=linux-bcm63xx.git
> >
> > They should apply with little difficulty on current upstream kernel and
> > besides the *_be stuffs that need to be cleaned up, they were ACKed by
> > the usb maintainer.
> 
> Cool, I'll see about applying them and working off of that. Thanks for
> the confirmation.
> 
> FWIW, I'm trying to reverse engineer Broadcom's DSL driver, which I hope
> will be useful so we can finally have a properly working OpenWRT
> firmware for DSL routers based on this firmware.

Good luck :)
-- 
WBR, Florian

Re: BCM63xx merge progress

[Florian Fainelli]

On Wednesday 02 December 2009 01:25:49 Hector Martin wrote:
> Maxime Bizon wrote:
> > Which part is broken/missing on your side ?
> 
> Well, there's a commit (4059ddb4) that claims to "Add USB OHCI support",
> yet it only adds a #include to ohci-hcd.c and a few .h bits. The actual
> ohci-bcm63xx.c is missing, which means the OHCI driver probably won't
> even compile. It's also missing the relevant platform device stuff in
> arch/. This concerned me somewhat, it looked like something weird had
> happened and the merge was incomplete.
> 
> Florian Fainelli wrote:
> > Everything that Maxime sent is, or will be merged either in 2.6.32 or
> > in 2.6.33, that includes:
> > - arch/mips/bcm63xx (will be in 2.6.32)
> > - USB driver bits (2.6.33)
> > - Ethernet MAC driver bits (2.6.32)
> > - Ethernet PHY driver bits (2.6.32)
> > - UART driver (2.6.32)
> > - PCMCIA support (2.6.32)
> 
> I can see the 2.6.32 stuff already. Regarding USB, is there a repo with
> the commit that will be in .33? Otherwise I'll just have to merge it
> locally and work from there for the moment. What's up with the weirdo
> commit mentioned above?
> 
> > We maintain a couple of different patches for OpenWrt, specifically
> > the mtd partition parser since we provide images, so the box should
> > boot from Flash. That driver is not in a mergable state at the moment.
> > There is also a embryon of a SPI driver and a watchdog driver, which I
> > will probably submit once cleaned up.
> 
> Another tidbit I saw in the OpenWRT set is the reset button GPIO
> support. I want to have that (eventually) too.

The gpio-buttons driver that we use in OpenWrt to do that is not that perfect 
and handling reset and other buttons is a distribution side problem since not 
evryone wants to map a button to a functionnality.

> 
> Basically, my dilemma is simply that I want to work off of mainline
> (especially since stuff has been partially merged already and changes
> have been made), yet I need the missing patches in order to make this
> work completely. So eventually I need a tree that contains everything I
> need, whether a few individual commits are not really ready for mainline
> or not. In doing this, I obviously want to keep in sync with any work
> that has been done already as much as possible. So please do tell if
> there's a specific place I should be looking at for mainline-compatible
> versions of these patches, or anything closer to what will eventually be
> merged. As I said, I don't mind doing the merges myself, but then it
> might make rebasing onto mainline harder if your official patches are
> merged in a substantially different way once they do make it to
> mainline. And if I do end up merging stuff, I might as well send it to
> you to save you some time.

2.6.33-rc will have the remaining bits, so if you should start working with 
something, better wait a for couple of days. Depending on what you are 
planning to work on, you really do not need to have everything merged.
-- 
WBR, Florian