>Number:         36204
>Category:       kern
>Synopsis:       makphy failed to work on Intel S3000AHLX server board
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Apr 24 10:50:00 +0000 2007
>Originator:     Wolfgang Stukenbrock
>Release:        NetBSD current
>Organization:
Dr. Nagler & Company GmbH
	
>Environment:
	
	
System: NetBSD test-s0 3.1 NetBSD 3.1 (test-s0) #0: Tue Apr 3 11:33:43 CEST 2007 root@test-s0:/usr/src/sys/arch/i386/compile/test-s0 i386
Architecture: i386 / amd64
Machine: i386 / amd64
>Description:
	There are two onbaord Gbit-Interfaces on this server board:
	wm0 at pci4 dev 0 function 0: Intel i82573E  IAMT, rev. 3
	wm1 at pci5 dev 5 function 0: Intel i82541GI 1000BASE-T Ethernet, rev. 5
	The i82541GI has an igphy and works fine.
	The i82573E has a makphy and fails to establisch connections to the network.
	I've added some debug output to see what is gooing on, and here is a short
	discription of what I've seen:
	After the system ahs booted the iterface will be reset (reset command in BMCR).
	Before reset it has the value 0x1140 and will be set to 0x0 after reset.
	Then the startup will setup auto-negotiation and will end up in the following
	setting of the control-registers.
	pscr 0xb60 - pssr 0x6d0c - bmcr 0x1000 epsc 0xd60 anar 0xde1 anap 0x45e1
	Interpretion:
	pscr - the PSCR_MDI_XOVER_MODE is to 0x3 ?? - not reflected int the headerfile.
	       I've also tried 0x2 - documented for AUTO without success
	       The PSCR_MBO is 0 - coment "must be one" ???
	       PSCR_CRS_ON_TX set - as done in the reset code (change from 3.1 to current)
	       The bits 8 and 9 are both set - not ducumented in headerfile are set. This
	       are energie mode bits in the FreeeBSD implementation ...
	pssr - The PHY report sucessfull link up with MDI 100TX fdx
	       speed and duplex are indicated as resolved.
	       cable length is 0x2 - whatever this means ...
	       some other (unknown to the headerfile) bits are set.
	bmcr - set to BMCR_AUTOEN
	anar - looks good
	anap - some information about the partner is retrieve - looks good
	epsc - ??? EPSC_TX_CLK = 0xd6 if it has so much bits ...

	The link LED in the switch is on and if packets are send to the network
	the activity LED indicates some data, but noone seems to get this data.
	I've also dump the mii_anegticks and mii_ticks and found a strange behavior,
	but I think this is another bug or feature of makphy.c.
	mii_anegticks is set to 10 (because the interface can do 1000).
	I've places a kernel-printf in the status routine and mii_ticks starts 0 on boot,
	will reach 3 after auto-neg.
	If I do a "ifconfig wm0 media auto" mii_ticks gets incremented by 2 or 3 each time
	the comand is issued. A reset of the interface is done after it has reached 10.
	I think there is a reset of mii_ticks ticks missing when starting auto-neg.

	I've replaced the switch with a 10MBit hub, and the interface does auto-neg
	to 10MBit, hdx and works fine. Packets are send and recieved as expected.
	The same will happen if I connect the 100MBit switch again and use
	"ifconfig wm0 media utp", "ifconfig wm0 media utp mediaopt fdx" or
	"ifconfig wm0 media utp mediaopt fdx,flow". The interface works.
	Here the setting of the registers of theese three settings in that order:
	pscr 0xb60 - pssr 0xd00 - bmcr 0x0 epsc 0xd60 anar 0x21 anap 0x0
	pscr 0xb60 - pssr 0x2d00 - bmcr 0x100 epsc 0xd60 anar 0x41 anap 0x0
	pscr 0xb60 - pssr 0x2d00 - bmcr 0x100 epsc 0xd60 anar 0xc41 anap 0x0
	The speed and mediaopt settings are reflectd in anar but no information
	is recieved from the peer! is this OK ???????

	No communication will be established useing 100tx instead of utp.
	Here is the debug output again for that:
	pscr 0xb60 - pssr 0x4d00 - bmcr 0x2000 epcr 0xd60 anar 0x81 anap 0x0
	pscr 0xb60 - pssr 0x6d00 - bmcr 0x2100 epcr 0xd60 anar 0x101 anap 0x0
	pscr 0xb60 - pssr 0x6d40 - bmcr 0x2100 epcr 0xd60 anar 0xd01 anap 0x0
	Again no anap information is retrieved - bmcr, anar looks good to me.
	pssr: "link up" is reported, but no communication can be done ...
	      speed and duplex mode look good

	I've tried the following version of makphy.c:
	$NetBSD: makphy.c,v 1.23 2007/02/23 03:03:10 msaitoh Exp $

	I need to get it up and are willing to spend additional time in debugging,
	but I've no datasheets etc..
	I've alreay looked into the FreeBSD implementation and tried some varaints
	from there without success. (e.g. other clock-bit setting)
>How-To-Repeat:
	Try to run the wm0 on S3000AHLX server board ...
>Fix:
	Not known to me up to now.
	If some have a datasheet for this PHY, I will help to locate
	the problem.

>Unformatted: