Re: ELDK 4.2/kilauea/3.5+ kernel broken

5 messages, 3 authors, 2012-10-19 · open the first message on its own page

Re: ELDK 4.2/kilauea/3.5+ kernel broken

From: Robert Berger <hidden>
Date: 2012-10-18 17:45:13 

FYI:

When I replace arch/powerpc/sysdev/ppc4xx_msi.c from a 3.6 kernel with a
ppc4xx_msi.c from 47da421981571c69ef29740cc55fa7248682e167 it boots from
nfs with a defconfig, so this seems to be the guilty one.

Please find attached the difference between the good and bad
ppc4xx_msi.c files.

I'll be happy to test your patches;)

Regards,

Robert

Re: ELDK 4.2/kilauea/3.5+ kernel broken

From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Date: 2012-10-18 19:03:58 

On Thu, 2012-10-18 at 20:45 +0300, Robert Berger wrote:
-       mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH addr */
-       mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low addr */
+       SDR0_WRITE(sdr_addr, (u64)res.start >> 32);      /*HIGH addr */
+       SDR0_WRITE(sdr_addr + 1, res.start & 0xFFFFFFFF); /* Low addr */
+
 
        msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
-       if (!msi->msi_dev)
+       if (msi->msi_dev)
                return -ENODEV; 
The above changes look bad. The first one is stupid, the second one is clearly broken.

The diff us from good to bad right ? Looks like somebody added a very busted patch.

If I look at the code in current upstream, I see:

	mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));	/*HIGH addr */
	mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));	/* Low addr */

	msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
	if (!msi->msi_dev)
		return -ENODEV;


Which looks correct. So this might be something specific to ELDK ?

Cheers,
Ben.

Re: ELDK 4.2/kilauea/3.5+ kernel broken

From: Robert Berger <hidden>
Date: 2012-10-18 20:05:31 

Hi,

On 10/18/2012 10:03 PM, Benjamin Herrenschmidt wrote:
On Thu, 2012-10-18 at 20:45 +0300, Robert Berger wrote:
quoted
-       mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH addr */
-       mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low addr */
+       SDR0_WRITE(sdr_addr, (u64)res.start >> 32);      /*HIGH addr */
+       SDR0_WRITE(sdr_addr + 1, res.start & 0xFFFFFFFF); /* Low addr */
+
 
        msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
-       if (!msi->msi_dev)
+       if (msi->msi_dev)
                return -ENODEV; 
The above changes look bad. The first one is stupid, the second one is clearly broken.

The diff us from good to bad right ? Looks like somebody added a very busted patch.
this (from 3.6) does not work:

-       mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH
addr */
-       mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low
addr */

The good old file (which works) is this:

+       SDR0_WRITE(sdr_addr, (u64)res.start >> 32);      /*HIGH addr */
+       SDR0_WRITE(sdr_addr + 1, res.start & 0xFFFFFFFF); /* Low addr */
+

If I look at the code in current upstream, I see:

	mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));	/*HIGH addr */
	mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));	/* Low addr */

	msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
	if (!msi->msi_dev)
		return -ENODEV;


Which looks correct. So this might be something specific to ELDK ?
I will be on the road from tomorrow for a week or so, but maybe I can
isolate the exact lines which break it. I can also try a newer compiler
to see if this changes anything.

Is there someone out there with a kilauea board who can boot a 3.6.
mainline kernel with a default config with a rootfs over nfs?

Cheers,
Ben.

Regards,

Robert

Re: ELDK 4.2/kilauea/3.5+ kernel broken

From: Mai La <hidden>
Date: 2012-10-19 03:24:09 

Hi,

My patch was:

@@ -150,12 +157,11 @@ static int ppc4xx_setup_pcieh_hw(struct
platform_device *dev,
        if (!sdr_addr)
                return -1;

-       SDR0_WRITE(sdr_addr, (u64)res.start >> 32);      /*HIGH addr */
-       SDR0_WRITE(sdr_addr + 1, res.start & 0xFFFFFFFF); /* Low addr */
-
+       mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH addr
*/
+       mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low addr
*/

        msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
-       if (msi->msi_dev)
+       if (!msi->msi_dev)
                return -ENODEV;

        msi->msi_regs = of_iomap(msi->msi_dev, 0);


1. The first few lines: change from SDR0_WRITE to mtdcri since the old one
cause crash. I use ELDK 4.2.

2. The second one should mean that: if not find any node then return error.
So it should be "!msi->msi_dev"

Regards,
Mai La.


On Fri, Oct 19, 2012 at 3:05 AM, Robert Berger <robert.karl.berger@gmail.com
wrote:
Hi,

On 10/18/2012 10:03 PM, Benjamin Herrenschmidt wrote:
quoted
On Thu, 2012-10-18 at 20:45 +0300, Robert Berger wrote:
quoted
-       mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH
addr */
quoted
quoted
-       mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low
addr */
quoted
quoted
+       SDR0_WRITE(sdr_addr, (u64)res.start >> 32);      /*HIGH addr */
+       SDR0_WRITE(sdr_addr + 1, res.start & 0xFFFFFFFF); /* Low addr */
+

        msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
-       if (!msi->msi_dev)
+       if (msi->msi_dev)
                return -ENODEV;
The above changes look bad. The first one is stupid, the second one is
clearly broken.
quoted
The diff us from good to bad right ? Looks like somebody added a very
busted patch.

this (from 3.6) does not work:

-       mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH
addr */
-       mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low
addr */

The good old file (which works) is this:

+       SDR0_WRITE(sdr_addr, (u64)res.start >> 32);      /*HIGH addr */
+       SDR0_WRITE(sdr_addr + 1, res.start & 0xFFFFFFFF); /* Low addr */
+

quoted
If I look at the code in current upstream, I see:

      mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH
addr */
quoted
      mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low
addr */
quoted
      msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
      if (!msi->msi_dev)
              return -ENODEV;


Which looks correct. So this might be something specific to ELDK ?
I will be on the road from tomorrow for a week or so, but maybe I can
isolate the exact lines which break it. I can also try a newer compiler
to see if this changes anything.

Is there someone out there with a kilauea board who can boot a 3.6.
mainline kernel with a default config with a rootfs over nfs?

quoted
Cheers,
Ben.

Regards,

Robert





  

    
Re: ELDK 4.2/kilauea/3.5+ kernel broken

    

      From: Robert Berger <hidden>

      Date: 2012-10-19 06:36:03 
    

  

  
Hi,

On 10/19/2012 06:16 AM, Mai La wrote:
quoted hunk
Hi,

My patch was:

@@ -150,12 +157,11 @@ static int ppc4xx_setup_pcieh_hw(struct
platform_device *dev,
        if (!sdr_addr)
                return -1;

-       SDR0_WRITE(sdr_addr, (u64)res.start >> 32);      /*HIGH addr */
-       SDR0_WRITE(sdr_addr + 1, res.start & 0xFFFFFFFF); /* Low addr */
-
+       mtdcri(SDR0, *sdr_addr, upper_32_bits(res.start));      /*HIGH
addr */
+       mtdcri(SDR0, *sdr_addr + 1, lower_32_bits(res.start));  /* Low
addr */

        msi->msi_dev = of_find_node_by_name(NULL, "ppc4xx-msi");
-       if (msi->msi_dev)
+       if (!msi->msi_dev)
                return -ENODEV;

        msi->msi_regs = of_iomap(msi->msi_dev, 0);


1. The first few lines: change from SDR0_WRITE to mtdcri since the old
one cause crash. I use ELDK 4.2.
The old one does not cause a crash for me.  As I said on a kilauea board
with ELDK 4.2 a 3.6 kernel, default config and everything reverted to
the good old file I can boot happily with a rootfs from nfs.

If I use the file as it is in 3.6 I don't see the kernel booting. but it
crashes.

2. The second one should mean that: if not find any node then return
error. So it should be "!msi->msi_dev"
In the 3.6 kernel it's with ! the old file (which works for me) is
without the !

... very strange ...

Regards,
Mai La.

Regards,

Robert




    
  

  
    

      

        Keyboard shortcuts
      

      

        
          
hback out one level

          
jnext message in thread

          
kprevious message in thread

          
ldrill in

          
Escclose help / fold thread tree

          
?toggle this help