[Xenomai] MDIO on Freescale MX35.

[Xenomai] MDIO on Freescale MX35.

[Chris Stone]

I am using Xenomai 2.6.0. We wish to run Xenomai on two different target boards, one uses the Freescale MPC8378 CPU,
which is powerpc, the other uses the Freescale MX35 which is arm11. This is for real products that are under development, 
so there are some factors that affect our choices. For instance, to simplify maintenance of these cards for many 
years to come, we want to use the same kernel version on the MPC8378 as we do on the MX35. For this reason, we 
chose to use kernel 3.0.13, because there are Adeos patches against kernel 3.0.13 for both arm and powerpc. However, 
I know that the Adeos patches and kernel 3.0.13 are not really supported by Xenomai 2.6.0 because they are only in the 
Git tree and not in the official tarball download. Nonetheless, the powerpc target has been running with no problem 
for some time now, and the card is actually in product verification now, which means that testing has not discovered 
any problems with Xenomai 2.6.0 on that target. I am now attempting bring Xenomai up on the mx35 card.

I am using the adeos patch adeos-ipipe-3.0.13-arm-1.18.08.patch. I tried following the README for the MXC, but the 
Freescale IMX git has no tag rel_imx_3.0.15_12.03.00. The denx tree has the tag, but the mxc patches as described in 
the README do not apply without errors. Even after fixing the patch rejects, I ended up with compilation errors, so I 
lost confidence in that approach. So, I applied the adeos-ipipe-3.0.13-arm-1.18.08.patch to the kernel.org 3.0.13 
kernel, without the pre and post patches, which went without error. The kernel boots, and many things work, but the 
FEC Ethernet is broken with an MDIO read timeout. 

I would like to debug this issue and provide whatever fixes back to the community. I don't really know where to 
start. I am looking for some guidance on what I should do to debug the issue.

Cheers,

Chris Stone
Embedded Linux Architect
Optelian
1 Brewer Hunt Way
Ottawa, Ontario K2K 2B5

Phone: 613-287-2000 x2106

www.optelian.com

Re: [Xenomai] MDIO on Freescale MX35.

[Gilles Chanteperdrix]

On 05/23/2012 05:43 PM, Chris Stone wrote:
> I am using Xenomai 2.6.0. We wish to run Xenomai on two different
> target boards, one uses the Freescale MPC8378 CPU, which is powerpc,
> the other uses the Freescale MX35 which is arm11. This is for real
> products that are under development, so there are some factors that
> affect our choices. For instance, to simplify maintenance of these
> cards for many years to come, we want to use the same kernel version
> on the MPC8378 as we do on the MX35. For this reason, we chose to use
> kernel 3.0.13, because there are Adeos patches against kernel 3.0.13
> for both arm and powerpc. However, I know that the Adeos patches and
> kernel 3.0.13 are not really supported by Xenomai 2.6.0 because they
> are only in the Git tree and not in the official tarball download.
> Nonetheless, the powerpc target has been running with no problem for
> some time now, and the card is actually in product verification now,
> which means that testing has not discovered any problems with Xenomai
> 2.6.0 on that target. I am now attempting bring Xenomai up on the
> mx35 card.
> 
> I am using the adeos patch adeos-ipipe-3.0.13-arm-1.18.08.patch. I
> tried following the README for the MXC, but the Freescale IMX git has
> no tag rel_imx_3.0.15_12.03.00. The denx tree has the tag, but the
> mxc patches as described in the README do not apply without errors.
> Even after fixing the patch rejects, I ended up with compilation
> errors, so I lost confidence in that approach.

This should not happen. This combination has been extensively tested on
imx51, imx53 and imx6. It should compile and you should get no rejects.
I will look into it.

> So, I applied the
> adeos-ipipe-3.0.13-arm-1.18.08.patch to the kernel.org 3.0.13 kernel,
> without the pre and post patches, which went without error. The
> kernel boots, and many things work, but the FEC Ethernet is broken
> with an MDIO read timeout.

Do you also get the MDIO timeout without the I-pipe and Xenomai patches?
If no, try applying the patch but not enabling CONFIG_IPIPE, then
enabling CONFIG_IPIPE without CONFIG_XENOMAI.

-- 
					    Gilles.

Re: [Xenomai] MDIO on Freescale MX35.

[Chris Stone]

With CONFIG_XENOMAI disabled, and CONFIG_IPIPE enabled the timeout still happens. With both CONFIG_XENOMAI and CONFIG_IPIPE disabled the MDIO timeout goes away. Thus the issue appears to be introduced by enabling ipipe.

Cheers,
   Chris.

-----Original Message-----
From: Gilles Chanteperdrix [mailto:gilles.chanteperdrix@xenomai.org] 
Sent: Wednesday, May 23, 2012 12:29 PM
To: Chris Stone
Cc: xenomai@xenomai.org
Subject: Re: [Xenomai] MDIO on Freescale MX35.

On 05/23/2012 05:43 PM, Chris Stone wrote:
> I am using Xenomai 2.6.0. We wish to run Xenomai on two different 
> target boards, one uses the Freescale MPC8378 CPU, which is powerpc, 
> the other uses the Freescale MX35 which is arm11. This is for real 
> products that are under development, so there are some factors that 
> affect our choices. For instance, to simplify maintenance of these 
> cards for many years to come, we want to use the same kernel version 
> on the MPC8378 as we do on the MX35. For this reason, we chose to use 
> kernel 3.0.13, because there are Adeos patches against kernel 3.0.13 
> for both arm and powerpc. However, I know that the Adeos patches and 
> kernel 3.0.13 are not really supported by Xenomai 2.6.0 because they 
> are only in the Git tree and not in the official tarball download.
> Nonetheless, the powerpc target has been running with no problem for 
> some time now, and the card is actually in product verification now, 
> which means that testing has not discovered any problems with Xenomai
> 2.6.0 on that target. I am now attempting bring Xenomai up on the
> mx35 card.
> 
> I am using the adeos patch adeos-ipipe-3.0.13-arm-1.18.08.patch. I 
> tried following the README for the MXC, but the Freescale IMX git has 
> no tag rel_imx_3.0.15_12.03.00. The denx tree has the tag, but the mxc 
> patches as described in the README do not apply without errors.
> Even after fixing the patch rejects, I ended up with compilation 
> errors, so I lost confidence in that approach.

This should not happen. This combination has been extensively tested on imx51, imx53 and imx6. It should compile and you should get no rejects.
I will look into it.

> So, I applied the
> adeos-ipipe-3.0.13-arm-1.18.08.patch to the kernel.org 3.0.13 kernel, 
> without the pre and post patches, which went without error. The kernel 
> boots, and many things work, but the FEC Ethernet is broken with an 
> MDIO read timeout.

Do you also get the MDIO timeout without the I-pipe and Xenomai patches?
If no, try applying the patch but not enabling CONFIG_IPIPE, then enabling CONFIG_IPIPE without CONFIG_XENOMAI.

-- 
					    Gilles.

Re: [Xenomai] MDIO on Freescale MX35.

[Gilles Chanteperdrix]

On 05/23/2012 06:34 PM, Chris Stone wrote:
> With CONFIG_XENOMAI disabled, and CONFIG_IPIPE enabled the timeout
> still happens. With both CONFIG_XENOMAI and CONFIG_IPIPE disabled the
> MDIO timeout goes away. Thus the issue appears to be introduced by
> enabling ipipe.

You can try enabling the I-pipe tracer, and triggering a trace freeze
when the MDIO timeout happen. The trace has to have sufficiently many
points to see the last interrupt corresponding to MDIO (if MDIO is
interrupt based).

-- 
					    Gilles.

Re: [Xenomai] MDIO on Freescale MX35.

[Chris Stone]

When I enable the I-pipe tracer, the problem goes away, i.e., there is no MDIO timeout, and
so the freeze is never executed, and so there is no frozen trace to examine. Does this
behavior give you any ideas about what the cause might be?

Cheers,
   Chris.


-----Original Message-----
From: Gilles Chanteperdrix [mailto:gilles.chanteperdrix@xenomai.org] 
Sent: Wednesday, May 23, 2012 12:37 PM
To: Chris Stone
Cc: xenomai@xenomai.org
Subject: Re: [Xenomai] MDIO on Freescale MX35.

On 05/23/2012 06:34 PM, Chris Stone wrote:
> With CONFIG_XENOMAI disabled, and CONFIG_IPIPE enabled the timeout 
> still happens. With both CONFIG_XENOMAI and CONFIG_IPIPE disabled the 
> MDIO timeout goes away. Thus the issue appears to be introduced by 
> enabling ipipe.

You can try enabling the I-pipe tracer, and triggering a trace freeze when the MDIO timeout happen. The trace has to have sufficiently many points to see the last interrupt corresponding to MDIO (if MDIO is interrupt based).

-- 
					    Gilles.

Re: [Xenomai] MDIO on Freescale MX35.

[Wolfgang Grandegger]

On 05/23/2012 05:43 PM, Chris Stone wrote:
> I am using Xenomai 2.6.0. We wish to run Xenomai on two different target boards, one uses the Freescale MPC8378 CPU,
> which is powerpc, the other uses the Freescale MX35 which is arm11. This is for real products that are under development, 
> so there are some factors that affect our choices. For instance, to simplify maintenance of these cards for many 
> years to come, we want to use the same kernel version on the MPC8378 as we do on the MX35. For this reason, we 
> chose to use kernel 3.0.13, because there are Adeos patches against kernel 3.0.13 for both arm and powerpc. However, 
> I know that the Adeos patches and kernel 3.0.13 are not really supported by Xenomai 2.6.0 because they are only in the 
> Git tree and not in the official tarball download. Nonetheless, the powerpc target has been running with no problem 
> for some time now, and the card is actually in product verification now, which means that testing has not discovered 
> any problems with Xenomai 2.6.0 on that target. I am now attempting bring Xenomai up on the mx35 card.
> 
> I am using the adeos patch adeos-ipipe-3.0.13-arm-1.18.08.patch. I tried following the README for the MXC, but the 
> Freescale IMX git has no tag rel_imx_3.0.15_12.03.00. The denx tree has the tag, but the mxc patches as described in 

The Freescale git repository has moved some time ago, you can find it
now here::

http://git.freescale.com/git/cgit.cgi/imx/linux-2.6-imx.git/refs/tags

DENX just fetches from that tree from time to time.

> the README do not apply without errors. Even after fixing the patch rejects, I ended up with compilation errors, so I 
> lost confidence in that approach. So, I applied the adeos-ipipe-3.0.13-arm-1.18.08.patch to the kernel.org 3.0.13 

Yes, that is true, unfortunately. A few other patches are required (for
the mx53) before applying the ADOES patches mentioned above.

> kernel, without the pre and post patches, which went without error. The kernel boots, and many things work, but the 
> FEC Ethernet is broken with an MDIO read timeout. 

That's a known issue in the FEC driver (on the mx53).

> I would like to debug this issue and provide whatever fixes back to the community. I don't really know where to 
> start. I am looking for some guidance on what I should do to debug the issue.

We have that working for the mx6q and mx53 and we can send the patches
if you like. But the Freescale kernel is in general a bad choice. It is
branched, adapted and tests usually just for *one* target platform. For
the tree above it's the mx6q. It does not compile for the mx53 and
that's why we need the patches and it might be worse for the mx35. If
you can, you should therefore use a mainline kernel and the mx35 is well
supported in 3.0.x, IIRC. What board do you need to support?

Wolfgang.