RE: MPC5200B SPI codec error when there is a heavy ethernet

RE: MPC5200B SPI codec error when there is a heavy ethernet

[Albrecht Moll]

> Txema wrote on 15 Jan 2008
> We have a MPC5200B based custom board running a 2.4.25 Denx linux
> kernel. We use the PSC6 in SPI-codec mode to exchange some bytes
> periodically in full-duplex with a peripheral.
>
> In 'normal' conditions this communication runs fine. But when we start
> sending packets over the Ethernet as a background activity, we see on
> the oscilloscope that one of the SPI transfers is interrupted.

Also just a shot in the dark:
I remember someone having similar problems with UART in AC97 mode.
(http://marc.info/?l=linuxppc-embedded&m=117932912618680&w=2)
I associated this with other FEC problems which boiled down to an issue
with
MPC52xx_XLB_CFG_BSDIS.

RE: MPC5200B SPI codec error when there is a heavy ethernet

[TXEMA LOPEZ]

Hi all,=20

The engineers of Freescale have recognized a problem in the PSC SPI =
Slave Select signal when there is a heavy ethernet loading. Textually =
they say:

" I have tested the MPC5200B FEC and PSC6 SPI.=20
You are right. If there is heavy Ethernet loading, the PSC SPI can stop =
during transmission with SS goes high.
Use a general-purpose output as SPI slave select signal instead PSC SPI =
SS signal.
Factory is informed about similar incorrect behaviour of the PSC SPI =
slave select."

We have repeated the error in three scenarios:
	In our MPC5200B custom board with a Denx 2.4.25 kernel.
	In a Lite5200B with a Denx 2.4.25 kernel.
	In a Lite5200B with the freescale bsp: =
mpc5200_lite5200b_20070203_ltib-rc4. It's a 2.6.16 kernel version.

We have checked the PSC3 and PSC6 and the behaviour is the same.

So, it seems is a chip's bug and we must avoid to use the SS signal with =
the PSC SPI if we want to communicate by ethernet. I think it's a =
probabilistic error and in case there is some traffic in ethernet and a =
transmission by SPI at the same time it could happen.


Best regards,
Txema L=F3pez.

=09
=09
=09

Re: MPC5200B SPI codec error when there is a heavy ethernet

[Juergen Beisert]

On Monday 28 January 2008 09:37, TXEMA LOPEZ wrote:
> The engineers of Freescale have recognized a problem in the PSC SPI Slave
> Select signal when there is a heavy ethernet loading. Textually they say:
>
> " I have tested the MPC5200B FEC and PSC6 SPI.
> You are right. If there is heavy Ethernet loading, the PSC SPI can stop
> during transmission with SS goes high. Use a general-purpose output as SPI
> slave select signal instead PSC SPI SS signal. Factory is informed about
> similar incorrect behaviour of the PSC SPI slave select."
>
> We have repeated the error in three scenarios:
> 	In our MPC5200B custom board with a Denx 2.4.25 kernel.
> 	In a Lite5200B with a Denx 2.4.25 kernel.
> 	In a Lite5200B with the freescale bsp:
> mpc5200_lite5200b_20070203_ltib-rc4. It's a 2.6.16 kernel version.
>
> We have checked the PSC3 and PSC6 and the behaviour is the same.
>
> So, it seems is a chip's bug and we must avoid to use the SS signal with
> the PSC SPI if we want to communicate by ethernet. I think it's a
> probabilistic error and in case there is some traffic in ethernet and a
> transmission by SPI at the same time it could happen.

There seems also another bug: If the PSC based SPI unit runs as SPI slave, =
and=20
some data in the send FIFO waits for transmission, the MISO line is active=
=20
even if SS input is high and blocks the bus. It does not happen, of the sen=
d=20
=46IFO is empty.

JB
=2D-=20
Dipl.-Ing. Juergen Beisert | http://www.pengutronix.de
=A0Pengutronix - Linux Solutions for Science and Industry
=A0   Handelsregister: Amtsgericht Hildesheim, HRA 2686
=A0 =A0 =A0    Vertretung Sued/Muenchen, Germany
   Phone: +49-8766-939 228 |  Fax: +49-5121-206917-9