problem with standard libc on 8xx

problem with standard libc on 8xx

[Demke, Torsten]

Hello all,

I created my own root-fs for a MPC8xx platform
using standard Debian packages (including libc).
When I start the system, the init scripts hangs
whenever it executes a function (exceptt some simple
thinks like 'echo').
- the same root-fs starts without problem on a MPC8245 platform.
- the root-fs starts when I exchange the libc with the libraries
from the ELDK.

I guess that the Debian libc uses floating point - which
is not available on MPC8xx - but I thougt the kernel would
emulate this calls (CONFIG_MATH_EMULATION=y).

Any hints?

Regards,
Torsten

  Torsten Demke
  Software Engineer
  Force Computers GmbH
  - A Solectron Company -
  Lilienthalstr. 15
  D-85579 Neubiberg/München
  Tel. ++49/89/60814-404
  Fax ++49/89/60814-112
  mailto:torsten.demke@fci.com
  www.forcecomputers.com


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

Re: problem with standard libc on 8xx

[Wolfgang Denk]

In message <9CFB9DA5261CD611A29B00508B78904807F880ED@ex-deu-munich02.force.de> you wrote:
>
> I created my own root-fs for a MPC8xx platform
> using standard Debian packages (including libc).
> When I start the system, the init scripts hangs
...
> I guess that the Debian libc uses floating point - which

Thi sis definitely one cause of the problem.  Another  one  might  be
optimization  for  a  incorrect  cache  line  size  (32 instead of 16
bytes).

> is not available on MPC8xx - but I thougt the kernel would
> emulate this calls (CONFIG_MATH_EMULATION=y).

This has never been working reliably for any real-life code. And it's
awfully slow.

Also note that you CANNOT mix standard binaries (from Debian) and 8xx
specific libraries (from ELDK) -  this  will  crash  as  soon  as  an
application attempts to use floating point.


Best regards,

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
Above all else -- sky.

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

RE: problem with standard libc on 8xx

[Demke, Torsten]

Hello Wolfgang,

so what is your proposal, to solve the problem?
I would like to use Debian binaries, but I cannot
use the floating point emulation of the kernel!?

Regards,
Torsten


> -----Original Message-----
> From: Wolfgang Denk [mailto:wd@denx.de]
> Sent: Freitag, 27. Februar 2004 10:43
> To: Demke, Torsten
> Cc: Linuxppc-Embedded (E-mail)
> Subject: Re: problem with standard libc on 8xx
>
>
> In message
> <9CFB9DA5261CD611A29B00508B78904807F880ED@ex-deu-munich02.forc
> e.de> you wrote:
> >
> > I created my own root-fs for a MPC8xx platform
> > using standard Debian packages (including libc).
> > When I start the system, the init scripts hangs
> ...
> > I guess that the Debian libc uses floating point - which
>
> Thi sis definitely one cause of the problem.  Another  one  might  be
> optimization  for  a  incorrect  cache  line  size  (32 instead of 16
> bytes).
>
> > is not available on MPC8xx - but I thougt the kernel would
> > emulate this calls (CONFIG_MATH_EMULATION=y).
>
> This has never been working reliably for any real-life code. And it's
> awfully slow.
>
> Also note that you CANNOT mix standard binaries (from Debian) and 8xx
> specific libraries (from ELDK) -  this  will  crash  as  soon  as  an
> application attempts to use floating point.
>
>
> Best regards,
>
> Wolfgang Denk
>
> --
> Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
> Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
> Above all else -- sky.
>

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

RE: problem with standard libc on 8xx

[Joakim Tjernlund]

> In message
> <9CFB9DA5261CD611A29B00508B78904807F880ED@ex-deu-munich02.forc
> e.de> you wrote:
> >
> > I created my own root-fs for a MPC8xx platform
> > using standard Debian packages (including libc).
> > When I start the system, the init scripts hangs
> ...
> > I guess that the Debian libc uses floating point - which
>
> Thi sis definitely one cause of the problem.  Another  one  might  be
> optimization  for  a  incorrect  cache  line  size  (32 instead of 16
> bytes).

Also, you must make sure you don't use any cache instructions such
as dcbz. I believe the optimized memset function i glibc uses dcbz.

The mpc8xx CPU has a HW bug which makes the TLB exceptions fail
for most cache instructions. There is a workaround for this, but it
is somewhat ugly. Search the embedded archives if you want to known more.

   Jocke

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

Re: problem with standard libc on 8xx

[Wolfgang Denk]

In message <9CFB9DA5261CD611A29B00508B78904807F88115@ex-deu-munich02.force.de> you wrote:
>
> so what is your proposal, to solve the problem?

Use libraries and binaries that were built for a 8xx system.

> I would like to use Debian binaries, but I cannot
> use the floating point emulation of the kernel!?

I'm afraid you cannot.

Best regards,

Wolfgang Denk

--
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd@denx.de
Old programmers never die, they just branch to a new address.

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