Data TLB error exception

Data TLB error exception

[Damjan Lampret]

Hello,

I am playing with 2.2.13 kernel on custom MPC860 board and I have a problem
with DTLB error (1400h) exception. I downloaded my kernel via BDM and it
boots ok until it starts /sbin/init (or any other user application). Most of
the time I get DTLB error usually in do_load_elf_binary() followed by data
access exception (data access exception normaly follows DTLB error exception
as far as I can tell from kernel sources) and a fatal machine check
exception in data access exception.

Another strange thing was happening with caches enabled when kernel hanged
almost instantly after booting to the point where it starts init. When I
disabled insn/data caches it is much more stable but it still hangs after a
couple of minutes. Althougth in this case I don't get DTLB error exception
AFAIK (I checked this only once). Perhaps this has something to do with BDM
(MPC860 that I have is revision 1 and it does not have a commonly known data
cache bug AFAIK)?

Any idea what is happening with my system? I would be happy to provide more
information if someone would care to help me.

regards, Damjan

PS Also I have MBX860 that I could use for a reference. But I get "Network
Boot Logic Error \ Packet Status: 0500" when trying to download kernel via
TFTP. Also nping in EPPC-Bug is not working. In fact I can't make any
network activity. Any idea? (I didn't yet tried to download kernel via
serial)


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Data TLB error exception

[Dan Malek]

Damjan Lampret wrote:

> I am playing with 2.2.13 kernel on custom MPC860 board and I have a problem
> with DTLB error (1400h) exception. I downloaded my kernel via BDM and it

BDMs with MMU and caches are a bad combination.  With the exception of
KGDB, there are no debuggers that understand the Linux kernel once
the MMU is enabled.  The debugger tools normally used with BDMs trap
MMU exceptions as a "bad thing" or at least try to invoke some monitor
support code that is supposed to reside in the kernel.

The MMUs and caches work fine on the 8xx processors when running Linux/PPC.
I have been running them all day, every day, for years and have several
products deployed in the field without trouble.

If you are using BDMs for downloading Linux, that is all you can do with
them.  Once the kernel executes the first few lines of code to enable
the MMU, you are done.  If you can't disable _all_ debugging from the
BDM tool, including show cycles, you can't use it with Linux.


	-- Dan

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Data TLB error exception

[Damjan Lampret]

> Damjan Lampret wrote:
>
> > I am playing with 2.2.13 kernel on custom MPC860 board and I have a
problem
> > with DTLB error (1400h) exception. I downloaded my kernel via BDM and it
>
> BDMs with MMU and caches are a bad combination.  With the exception of
> KGDB, there are no debuggers that understand the Linux kernel once
> the MMU is enabled.  The debugger tools normally used with BDMs trap
> MMU exceptions as a "bad thing" or at least try to invoke some monitor
> support code that is supposed to reside in the kernel.
>

I use EST's VisionICE. I use it to download code and to check which
exceptions occur. Of course I also assumed that debugger is having problems
with exceptions so I disabled detection in DER but my target still hangs. I
am quite sure that visionICE is not executing any "monitor support" code.
Like most debuggers it does not support MMU so I need to add 0xc0000000 to
all addresses when analyzing kernel code.

> The MMUs and caches work fine on the 8xx processors when running
Linux/PPC.
> I have been running them all day, every day, for years and have several
> products deployed in the field without trouble.
>

Yes, apparently there is a problem running target with BDM and caches
enabled. As I said in my original email when I disabled caches target is
much more stable (but still hangs after a couple of minutes). Perhaps this
is still because of a debugger. Although I don't understand how a debugger
can harm target when target is not in debug mode (when running "real-time").

> If you are using BDMs for downloading Linux, that is all you can do with
> them.  Once the kernel executes the first few lines of code to enable

I do use it for downloading code.

> the MMU, you are done.  If you can't disable _all_ debugging from the
> BDM tool, including show cycles, you can't use it with Linux.
>

You can disable and I did. Target still hangs. What I don't understand is
why after DTLB error exception. Everytime I get machine check in data access
exception handler. I'll backtrace it and post my results to the list.

regards, Damjan


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Data TLB error exception

[Damjan Lampret]

----- Original Message -----
From: Wolfgang Denk <wd@denx.de>
To: Damjan Lampret <damjan.lampret@arnes.si>
Sent: Tuesday, February 29, 2000 7:27 PM
Subject: Re: Data TLB error exception


[cut]
> > I suspect there is something wrong with my hub (MBX is connected to a
hub).
>
> Try a cross-over cable first - then you can be  sure  which  part  is
> failing.
>
> Wolfgang Denk
>

Perhaps this will save someone a couple of days of troubles...

Once in a while strange things happen. I have PINE Technology's 8-port
ethernet hub that works ok. Hub like any other. I have various custom MPC8xx
boards and PCs connected to it. Everything works OK. So I connected MBX860
to it and tried to download linux via tftp. My MBX was complaining with
'Network Error 0019' or something like that. I disconnected everything
except MBX and my working PC. It didn't work. Lost a couple of weeks trying
to figure out what is wrong. Lately I got another hub. This one is 8-port
3com's Office Connect. Since then MBX860 works great. And don't tell me to
check my cables or something like that. I tried everything except another
hub.

regards, Damjan

PS GCC from Dan's mbxroot.tar.gz is not working properly on my MBX. Other
simple programs from /bin and /usr/bin etc work except GCC which hangs
console (it outputs proper MPC8xx assembly in /tmp and then stops; kernel is
still running properly). I tried several GCCs from Dan's mbxroot package. Is
any of them properly compiled for native work on Linux/MPC860 (I found gcc
in several places) ? I have also compiled egcs-2.90.25 for cross work on
Linux/i386 but binaries segfault on Linux/MBX. Assembly is properly
generated by I suspect that not proper crt* are linked to final binary.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/