Problems with PHY on STK5200

Problems with PHY on STK5200

[Jan-Hinnerk Dumjahn]

Hi,

I'm trying to compile a recent kernel from DENX "linuxppc_2_4_devel" for my 
STK5200 using ELDK 3.1.1.

It seems that the current kernel there is not configuring the PHY. It looks 
like the link is configured for full-duplex (although it should be configured 
half-duplex).
There is no informative output about the PHY after the line

eth0: Phy @ 0x0, type LXT971 (0x001378e2)

-------

I have traced this change of behaviour back to a commit on 2005-09-04 
22:41:06. This is a major restructuring/rework of the file 
"arch/ppc/5xxx_io/fec.c".

BTW: What is the normal way to reference a commit in a git-repository? I 
haven't seen any revision numbers.

The versions also show different program flow (with MPC5xxx_FEC_DEBUG set to 
2).
The old version goes:
- mpc5xxx_fec_init()
- mpc5xxx_fec_restart()
  - PHY gets identified here
- mpc5xxx_fec_open()
  - autonegotiation here
- mpc5xxx_fec_restart()

The new one goes
- mpc5xxx_fec_init()
- mpc5xxx_fec_open()
- mpc5xxx_fec_restart()
  - PHY gets identified here


The new version has some bad (or at least dead) code in the funtion 
mpc5xxx_fec_setup(). However, I don't understand enough of the drivers design 
to see how to fix this.

Can anybody explain what the driver should to? Or even better has some patch 
to make it work?

Thanks in advance
/Jan-Hinnerk Dumjahn

Re: Problems with PHY on STK5200

[Wolfgang Denk]

In message <200703011216.57611.jan-hinnerk.dumjahn@arcor.de> you wrote:
> 
> I'm trying to compile a recent kernel from DENX "linuxppc_2_4_devel" for my 
> STK5200 using ELDK 3.1.1.

It should compile and run fine.

> It seems that the current kernel there is not configuring the PHY. It looks 
> like the link is configured for full-duplex (although it should be configured 
> half-duplex).

What makes you think it should be half duplex only?

> BTW: What is the normal way to reference a commit in a git-repository? I 
> haven't seen any revision numbers.

You specify the git commit ID.

> Can anybody explain what the driver should to? Or even better has some patch 
> to make it work?

The driver should work :-) And it certainly does for me.

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH, HRB 165235 Munich, CEO: Wolfgang Denk
Office:  Kirchenstr. 5,       D-82194 Groebenzell,            Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@denx.de
Is not that the nature of men and women -- that the  pleasure  is  in
the learning of each other?
	-- Natira, the High Priestess of Yonada, "For the World is
	   Hollow and I Have Touched the Sky", stardate 5476.3.

Re: Problems with PHY on STK5200

[Jan-Hinnerk Dumjahn]

On Thursday 01 March 2007 19:57, Wolfgang Denk wrote:

> In message <200703011216.57611.jan-hinnerk.dumjahn@arcor.de> you wrote:
> > I'm trying to compile a recent kernel from DENX "linuxppc_2_4_devel" for
> > my STK5200 using ELDK 3.1.1.
>
> It should compile and run fine.

It is compiling fine and it is basically working, but I get lot of receive 
errors on the link, making it (almost) unusable for NFS root file system.

If I connect the STK5200 directly to my workstation using a crossover cable, 
everything works fine.

> > It seems that the current kernel there is not configuring the PHY. It
> > looks like the link is configured for full-duplex (although it should be
> > configured half-duplex).
>
> What makes you think it should be half duplex only?

The old version (up to and including cc891b92c2540487be1eeba4b5d38c913570d339) 
is negotiating half duplex. My linux workstation is also using half duplex if 
plugged into the same hub (a D-Link DFE-905DX).

However, I'm not 100% sure that the PHY on STK5200 is operating with 
full-duplex, because the link status is not reported by the kernel.

> > BTW: What is the normal way to reference a commit in a git-repository? I
> > haven't seen any revision numbers.
>
> You specify the git commit ID.

I hope the number above is a commit-ID...

> > Can anybody explain what the driver should to? Or even better has some
> > patch to make it work?
>
> The driver should work :-) And it certainly does for me.

It certainly doesn't work for me ;-(

All least newer versions of the kernel don't register "mdio_timer_callback()". 
So, link up report isn't working.
This is caused by two occurences of the following code in mpc5xxx_fec_setup().

if(!reinit) {
...
  if(reinit) {
...
  }
...
}

Nevertheless, the MDIO-code is doing something useful. Without CONFIG_USE_MDIO 
I don't get a single packet through the network interface (using either hub 
or crossover-cable).

I'm willing to do further tests and provide more information. However, I need 
some advice on the direction, because I don't understand enough of the code, 
yet.

Best regards,
  Jan-Hinnerk Dumjahn

Re: Problems with PHY on STK5200

[Jan-Hinnerk Dumjahn]

On Monday 05 March 2007 17:12, Jan-Hinnerk Dumjahn wrote:
> On Thursday 01 March 2007 19:57, Wolfgang Denk wrote:
> > In message <200703011216.57611.jan-hinnerk.dumjahn@arcor.de> you wrote:

[...]
> > > Can anybody explain what the driver should to? Or even better has some
> > > patch to make it work?
> >
> > The driver should work :-) And it certainly does for me.
>
> It certainly doesn't work for me ;-(
>
> All least newer versions of the kernel don't register
> "mdio_timer_callback()". So, link up report isn't working.
> This is caused by two occurences of the following code in
> mpc5xxx_fec_setup().
>
> if(!reinit) {
> ...
>   if(reinit) {
> ...
>   }
> ...
> }

I've done some more analysis: the following is a list of code/data in "fec.c" 
that is never used:

- lines 936-950 (if-clause)
- lines 1035-1069 (if-clause)
- mdio_timer_callback()
- mpc5xxx_fec_link_up()
- mii_display_status()
- phy_cmd_config
- mii_queue_config()
- mii_display_config()
- mii_do_cmd()

BTW: What is the status of the code guarded by CONFIG_USE_MDIO_NOT_YET?

Best regards
  Jan-Hinnerk Dumjahn