Ethernet POST freeze on 440EPx boards (PMC440, sequoia)

Hi,

I noticed some strange issues with our PMC440 (PPC440PEx based) board and also with the sequoia eval platform.

In a certain configuration these boards stuck during the Ethernet POST tests. When they got stuck, it is even not possible to attach with a BDI2000.

On the console you only see this:

U-Boot 1.3.1 (May 30 2008 - 17:01:42)

CPU: ? AMCC PowerPC 440EPx Rev. A at 533.333 MHz (PLB=133, OPB=66, EBC=66 MHz) ? ? ? ?Security/Kasumi support
? ? ? ?Bootstrap Option H - Boot ROM Location I2C (Addr 0x52)
? ? ? ?Internal PCI arbiter enabled, PCI async ext clock used
? ? ? ?32 kB I-Cache 32 kB D-Cache
Board: Sequoia - AMCC PPC440EPx Evaluation Board, Rev. F, PCI=33 MHz
I2C: ? ready
DTT: ? 1 is 38 C
DRAM: ?256 MB
FLASH: 64 MB
NAND: ?32 MiB
PCI: ? Bus Dev VenId DevId Class Int
? ? ? ? 00 ?0c ?168c ?0013 ?0200 ?43
In: ? ?serial
Out: ? serial
Err: ? serial
USB: ? Host(int phy) Device(ext phy)
Net: ? ppc_4xx_eth0, ppc_4xx_eth1

I first noticed this behavior on our PMC440 boards. These boards are similiar to the Sequoia platform (256MB DDR2 RAM, 2x Gigabit Ethernet, ...). Playing with some printfs I found out that the board got stuck in the Ethernet POST. When I disable U-Boot's LOG_BUFFER feature the problem dissapears. First I never noticed this bahavior on the sequoia platform and I thought about a PMC440 specific issue. Then I compiled U-Boot 1.3.1 for the sequoia platform (1.3.1 is used on PMC440 until now). Last week I got the same issue on the sequoia platform. Not that often as on the PMC440, but the same issue. Because it is easy to reproduce on PMC440 boards, I played a little bit with different configurations:

1) With LOG_BUFFER enabled: often, about every 2nd boot
2) Without LOG_BUFFER (POST messages come out in the console) -> issue never seens
3) Modified 4xx Ether POST with RX Buffers in OCM -> issue never seens
4) DCACHE turned on -> issue never seens

On the sequoia board I've only seens it after a reboot from Linux. But on PMC440 also after poweron. This must not have any meaning because I do more testing with our board than with the eval board :-)

printf debugging showed up that the boards get stuck in post/cpu/ppc4xx/ether.c in test_ctlr()/ether_post_send(). Some packets are send, recevied and checked correctly. Then suddenly at a random packet size ether_post_send() freezes the board.

Did anybody else see this behavior?
Did we miss any EMAC errata?

Matthias

Hi Matthias,

Oh, this hints to some deeper problems, so maybe one should check all erratas (again).

[...]

The last time I had problems looking similar to what you see, these two commits from Anatolij solved my problems:

commit 5e3dca577b7c1bf58bd2b48449b18b7e7dcd8e04

Author: Anatolij Gustschin <[email protected]>
Date:   Thu Apr 17 18:18:00 2008 +0200

Fix crash on sequoia in ppc_4xx_eth_init

Currently U-Boot crashes in ppc_4xx_eth_init on sequoia with cache enabled (TLB Parity exeption). This patch fixes the problem.

commit accf7355767dc7f6b85d88bb1c75c9d95e84ba5b

Author: Anatolij Gustschin <[email protected]>
Date:   Thu Apr 17 18:15:27 2008 +0200

ppc4xx: Fix crash on sequoia with cache enabled

Currently U-Boot crashes on sequoia board in CPU POST if cache is enabled (CONFIG_4xx_DCACHE defined). The cache won't be disabled by change_tlb before CPU POST because there is an insufficient adress range check since CFG_MEM_TOP_HIDE was introduced. This patch tries to fix this problem.

Cheers
Detlev

But in terms of creative  information, information that people can use
or enjoy, and that will be  used and enjoyed more  the more people who
have it, always we should encourage the copying.
	                                    -- Richard M. Stallman
--
DENX Software Engineering GmbH,      MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich,  Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-40 Fax: (+49)-8142-66989-80 Email: dzu at denx.de