MMU problems?

MMU problems?

[Nguyen Xuan Hoang]

Hi All,

I have seen a lot of discussions about MMU on MPC823, that make me worry. I just start and not up to see the real problem, look at the errata I don't see any problem about MMU. So is that really a bug? or just problem of misusing code? disable the cache will slow down the system... not using MMU seems to be difficult with Linux,
so what's the solution?  Hope somebody can give me a point.

Best regards
Hoang


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

Re: MMU problems?

[Dan Malek]

Nguyen Xuan Hoang wrote:

> I have seen a lot of discussions about MMU on MPC823,


There are no MMU or cache related problems on any of the MPC8xx
processors and stable versions of Linux.  These are just people
discussing the challenges of building custom boards and making
source code changes.

There are dozens of people shipping products with Linux running on
these processors.  You seldom hear from people that are happy and
successful.


	-- Dan

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

RE: MMU problems?

[Patrick Lerda]

Have you tried to do complex operation on a 8xx board like compiling the
kernel. This operation
works well on a PPC740 board, but on a 8xx after a long time the kernel goes
in an unstable
state. Theses bugs are related to the 8xx MMU code.

Patrick LERDA

> -----Message d'origine-----
> De:	Dan Malek [SMTP:dan@netx4.com]
> Date:	jeudi 22 juin 2000 05:56
> À:	Nguyen Xuan Hoang
> Cc:	linuxppc-embedded@lists.linuxppc.org
> Objet:	Re: MMU problems?
>
>
> Nguyen Xuan Hoang wrote:
>
> > I have seen a lot of discussions about MMU on MPC823,
>
>
> There are no MMU or cache related problems on any of the MPC8xx
> processors and stable versions of Linux.  These are just people
> discussing the challenges of building custom boards and making
> source code changes.
>
> There are dozens of people shipping products with Linux running on
> these processors.  You seldom hear from people that are happy and
> successful.
>
>
> 	-- Dan
>

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

RE: MMU problems?

[Lucinda Schafer]

I agree. I believe we ARE dealing with a kernel bug, and it is particular to
the MPC823.

Let's hear some success stories--embedded units in production, running in
the field for days, shutting down power and rebooting with no kernel panics.
That may help convince me we are dealing with a hardware issue.


Lucinda Schafer
Staff Software Engineer
Adaptive Micro-Ware, Inc.


-----Original Message-----
From: Patrick Lerda [mailto:Lerda@microprocess.com]
Sent: Thursday, June 22, 2000 2:01 AM
To: 'Dan Malek'; Nguyen Xuan Hoang
Cc: linuxppc-embedded@lists.linuxppc.org
Subject: RE: MMU problems?



Have you tried to do complex operation on a 8xx board like compiling the
kernel. This operation
works well on a PPC740 board, but on a 8xx after a long time the kernel goes
in an unstable
state. Theses bugs are related to the 8xx MMU code.

Patrick LERDA

> -----Message d'origine-----
> De:	Dan Malek [SMTP:dan@netx4.com]
> Date:	jeudi 22 juin 2000 05:56
> À:	Nguyen Xuan Hoang
> Cc:	linuxppc-embedded@lists.linuxppc.org
> Objet:	Re: MMU problems?
>
>
> Nguyen Xuan Hoang wrote:
>
> > I have seen a lot of discussions about MMU on MPC823,
>
>
> There are no MMU or cache related problems on any of the MPC8xx
> processors and stable versions of Linux.  These are just people
> discussing the challenges of building custom boards and making
> source code changes.
>
> There are dozens of people shipping products with Linux running on
> these processors.  You seldom hear from people that are happy and
> successful.
>
>
> 	-- Dan
>


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

Re: MMU problems?

[Daris Nevil]

Lucinda,

I am having good success with the MPC850.  Our board is able to
run without MMU problems, and shutting down is not a problem.
We are using a kernel leveraged from 2.2.5, so it is a bit older,
and I think some of the MMU code has changed since them.

Daris Nevil
SiSIC Inc/SNMC
www.snmc.com

Lucinda Schafer wrote:

> I agree. I believe we ARE dealing with a kernel bug, and it is particular to
> the MPC823.
>
> Let's hear some success stories--embedded units in production, running in
> the field for days, shutting down power and rebooting with no kernel panics.
> That may help convince me we are dealing with a hardware issue.
>
> Lucinda Schafer
> Staff Software Engineer
> Adaptive Micro-Ware, Inc.
>
> -----Original Message-----
> From: Patrick Lerda [mailto:Lerda@microprocess.com]
> Sent: Thursday, June 22, 2000 2:01 AM
> To: 'Dan Malek'; Nguyen Xuan Hoang
> Cc: linuxppc-embedded@lists.linuxppc.org
> Subject: RE: MMU problems?
>
> Have you tried to do complex operation on a 8xx board like compiling the
> kernel. This operation
> works well on a PPC740 board, but on a 8xx after a long time the kernel goes
> in an unstable
> state. Theses bugs are related to the 8xx MMU code.
>
> Patrick LERDA
>
> > -----Message d'origine-----
> > De:   Dan Malek [SMTP:dan@netx4.com]
> > Date: jeudi 22 juin 2000 05:56
> > À:    Nguyen Xuan Hoang
> > Cc:   linuxppc-embedded@lists.linuxppc.org
> > Objet:        Re: MMU problems?
> >
> >
> > Nguyen Xuan Hoang wrote:
> >
> > > I have seen a lot of discussions about MMU on MPC823,
> >
> >
> > There are no MMU or cache related problems on any of the MPC8xx
> > processors and stable versions of Linux.  These are just people
> > discussing the challenges of building custom boards and making
> > source code changes.
> >
> > There are dozens of people shipping products with Linux running on
> > these processors.  You seldom hear from people that are happy and
> > successful.
> >
> >
> >       -- Dan
> >
>


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

Re: MMU problems?

[Dan Malek]

Patrick Lerda wrote:
>
> Have you tried to do complex operation on a 8xx board like compiling the
> kernel.

Yes, and it runs just like any other system that doesn't have any
swap space and a minimal amount of memory.  Some things just don't
work well, and applications fail because few test for memory allocation
failure.

Are there bugs?  Perhaps.  Do I want them fixed?  Yes.

Do you want to help this FREE software effort?  If you are so sure
there is an MMU bug, find it, fix it, and send me the patch.  We would
all be grateful.  If you want dedicated help to assist your custom
product development so you can make money, you may have to pay one
of the very capable embedded Linux companies for this attention.

Thanks.


	-- Dan

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

Re: MMU problems?

[Wolfgang Denk]

In message <A109131318C4D1119AC20060088DECE330F4B1@amwmail.adaptivemicro.com> you wrote:
>
> I agree. I believe we ARE dealing with a kernel bug, and it is particular to
> the MPC823.

What makes you think so? Have you run the same  hardware  with  other
CPUs  without  problems?  Have  you  seen  other  "good" hardware (TQ
Components, Embedded Planet) having with similar problems,  and  only
when running with 823 CPUs?

> Let's hear some success stories--embedded units in production, running in
> the field for days, shutting down power and rebooting with no kernel panics.
> That may help convince me we are dealing with a hardware issue.

OK, then here is a success story.

I have several systems in test here, a mix of 823, 850, 860, -DE, -T,
-SR, and  some  development  systems  are  in  permanent  use  by  my
customers, and I can't say I see any differences between the CPU's.

Booting or shutting down was _never_ a  problem.  We  had  the  usual
crashes  whith  early  versions of 2.2.13 when running low on memory,
but that's fixed.

With recent kernels (for instance, the  2.2.13  version  from  MV)  I
haven't  problems  for  a _very_ long time (except for the 2.4.0 test
kernels - but I don't use these in production yet).


Did you check if environmental parameters make  any  difference?  How
stable  is  your power supply - are the voltages for your board clean
and within limits?

Did you measure CPU temperatures, or did you try cooling the CPU's? I
remember early versions of some TQ modules  (MPC821  +  Adaptec  SCSI
controller  on  a  credit  card  sized modules) which got too hot and
stopped working.


And yes, I just compiled (once more) a Linux  kernel  on  one  of  my
boxes  (that's  why  I reply so late - it takes more than 140 minutes
with a NFS based root filsystem and 16 MB of RAM). You are right,  it
makes  kind  of  a good regression test - it streeses all of CPU, I/O
and (with NFS root) networking.

You asked for success stories: It works for me.

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
"Virtual" means never knowing where your next byte is coming from.

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

RE: MMU problems?

[Patrick Lerda]

The things work better on our board now. The main problem, with 8xx kernel
patches are spread everywhere, and kernel like mpc8xx contains
a lot of useless changes that make the changes difficult to resync with the
main
serie. Kernel like 2.2.16 works properly and reliably on PPC7xx and PPC6xx
processors,
with a lot of PCI drivers functionnal now. I think the main effort will be
to resync 8xx kernels
with the standard ones. Now I have a patched 2.2.16 kernel that seems to
work on our board. Complex operations like compiling a kernel
take hours to complete on a 8xx board, and the bugs seems to have a low
probability.

(With the classic mpc8xx kernel, the system crash after only 20 minutes
while compiling
a kernel)

I have some success with theses kernel now, but I'm not sure if some bugs
remains, testing this system takes now hours... And this is quite
undebugable...

Patrick LERDA



> -----Message d'origine-----
> De:	Dan Malek [SMTP:dan@netx4.com]
> Date:	jeudi 22 juin 2000 17:58
> À:	Patrick Lerda
> Cc:	Nguyen Xuan Hoang; linuxppc-embedded@lists.linuxppc.org
> Objet:	Re: MMU problems?
>
> Patrick Lerda wrote:
> >
> > Have you tried to do complex operation on a 8xx board like compiling the
> > kernel.
>
> Yes, and it runs just like any other system that doesn't have any
> swap space and a minimal amount of memory.  Some things just don't
> work well, and applications fail because few test for memory allocation
> failure.
>
> Are there bugs?  Perhaps.  Do I want them fixed?  Yes.
>
> Do you want to help this FREE software effort?  If you are so sure
> there is an MMU bug, find it, fix it, and send me the patch.  We would
> all be grateful.  If you want dedicated help to assist your custom
> product development so you can make money, you may have to pay one
> of the very capable embedded Linux companies for this attention.
>
> Thanks.
>
>
> 	-- Dan

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

RE: MMU problems?

[Patrick Lerda]

I've now  a success with our board. It takes hours to compile with our
kernel configuration file,
on a NFS mounted partition, a PPC823 board with 64MB of memory.

Patrick LERDA


> -----Message d'origine-----
> De:	Wolfgang Denk [SMTP:wd@denx.de]
> Date:	jeudi 22 juin 2000 19:25
> À:	Lucinda Schafer
> Cc:	Patrick Lerda; 'Dan Malek'; Nguyen Xuan Hoang;
> linuxppc-embedded@lists.linuxppc.org
> Objet:	Re: MMU problems?
>
>
> In message
> <A109131318C4D1119AC20060088DECE330F4B1@amwmail.adaptivemicro.com> you
> wrote:
> >
> > I agree. I believe we ARE dealing with a kernel bug, and it is
> particular to
> > the MPC823.
>
> What makes you think so? Have you run the same  hardware  with  other
> CPUs  without  problems?  Have  you  seen  other  "good" hardware (TQ
> Components, Embedded Planet) having with similar problems,  and  only
> when running with 823 CPUs?
>
> > Let's hear some success stories--embedded units in production, running
> in
> > the field for days, shutting down power and rebooting with no kernel
> panics.
> > That may help convince me we are dealing with a hardware issue.
>
> OK, then here is a success story.
>
> I have several systems in test here, a mix of 823, 850, 860, -DE, -T,
> -SR, and  some  development  systems  are  in  permanent  use  by  my
> customers, and I can't say I see any differences between the CPU's.
>
> Booting or shutting down was _never_ a  problem.  We  had  the  usual
> crashes  whith  early  versions of 2.2.13 when running low on memory,
> but that's fixed.
>
> With recent kernels (for instance, the  2.2.13  version  from  MV)  I
> haven't  problems  for  a _very_ long time (except for the 2.4.0 test
> kernels - but I don't use these in production yet).
>
>
> Did you check if environmental parameters make  any  difference?  How
> stable  is  your power supply - are the voltages for your board clean
> and within limits?
>
> Did you measure CPU temperatures, or did you try cooling the CPU's? I
> remember early versions of some TQ modules  (MPC821  +  Adaptec  SCSI
> controller  on  a  credit  card  sized modules) which got too hot and
> stopped working.
>
>
> And yes, I just compiled (once more) a Linux  kernel  on  one  of  my
> boxes  (that's  why  I reply so late - it takes more than 140 minutes
> with a NFS based root filsystem and 16 MB of RAM). You are right,  it
> makes  kind  of  a good regression test - it streeses all of CPU, I/O
> and (with NFS root) networking.
>
> You asked for success stories: It works for me.
>
> Wolfgang Denk
>
> --
> Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
> Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
> "Virtual" means never knowing where your next byte is coming from.
>

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

Re: MMU problems?

[Wolfgang Denk]

In message <17B78BDF120BD411B70100500422FC63079FA0@IIS000> you wrote:
>
> The things work better on our board now. The main problem, with 8xx kernel
> patches are spread everywhere, and kernel like mpc8xx contains
> a lot of useless changes that make the changes difficult to resync with the

Nobody forces you to apply patches you don't need or consider useless.

And I really don't understand what you're complaining about:  if  you
just  need  a stable, production quality 2.2.x Linux kernel then grab
the latest MontaVista version, configure, build and run.

Dan (and others) have pointed this out several times before.

If you need something more recent than 2.2.13 for  some  reasons  you
can  try  using  the  BetKeeper  tree  (2.4.0-test1-ac21 or so at the
moment). This is where current work is going into.

> serie. Kernel like 2.2.16 works properly and reliably on PPC7xx and PPC6xx
> processors,
> with a lot of PCI drivers functionnal now. I think the main effort will be
> to resync 8xx kernels
> with the standard ones. Now I have a patched 2.2.16 kernel that seems to
> work on our board. Complex operations like compiling a kernel

Porting those changes back to 2.2.xx with x > 13 seems just  a  waste
of time to me. You will throw all this away as soon as 2.4 is stable.

Is there any special reason you can't use the 2.2.3 MV sources?

Have you had a look at the 2.4.0-test* sources?

> take hours to complete on a 8xx board, and the bugs seems to have a low
> probability.

Urghhhh... A bug is a bug is a bug, no matter how difficult it is  to
trigger  in test cases. It will become the show stopper in your first
presentation for the most important customer.


Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
Quote from the Boss after overriding the decision of a task force  he
created  to  find  a  solution:  "I'm  sorry  if  I ever gave you the
impression your input would have any effect on my  decision  for  the
outcome of this project!"

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

Re: MMU problems?

[Wolfgang Denk]

In message <17B78BDF120BD411B70100500422FC63079FA3@IIS000> you wrote:
>
> I think getting a reliable and working linux-2.2.x is not a waste of time.

You already _have_ a reliable and working linux-2.2.x, with x==13.

Download the MontaVista sources, configure, build, and run.

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
"Been through Hell? Whaddya bring back for me?"        - A. Brilliant

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