ssb: 2.6.34.1 -> 2.6.34.2 regression

ssb: 2.6.34.1 -> 2.6.34.2 regression

[Michael Büsch]

Hi,

The SPROM changes applied to SSB in 2.6.34.2 completely broke SPROM
extraction. This leads to broken wireless for my BCM4306 card.
All other BCM43xx and BCM44xx most likely are affected, too.
Symptoms should be wrong MAC addresses and broken transmission (for 43xx).

The patch changed the SPEX() macro and removed the offset subtraction,
but did not remove the actual 0x1000 offset from the SPROM defines.
This results in random memory poking rather than proper SPROM data
extraction.

I applied the following to my kernel to get it working again. However
this is not a proper fix. I suggest that the -stable SSB patch
is dropped and mainline patches being applied. They seem to
be correct (However I did not test mainline, yet).


Index: linux-2.6.34/drivers/ssb/pci.c
===================================================================
--- linux-2.6.34.orig/drivers/ssb/pci.c	2010-08-05 23:26:07.000000000 +0200
+++ linux-2.6.34/drivers/ssb/pci.c	2010-08-07 00:17:26.000000000 +0200
@@ -169,7 +169,7 @@ err_pci:
  }

  /* Get the word-offset for a SSB_SPROM_XXX define. */
-#define SPOFF(offset)	((offset) / sizeof(u16))
+#define SPOFF(offset)	(((offset) - SSB_SPROM_BASE1) / sizeof(u16))
  /* Helper to extract some _offset, which is one of the SSB_SPROM_XXX 
defines. */
  #define SPEX16(_outvar, _offset, _mask, _shift)	\
  	out->_outvar = ((in[SPOFF(_offset)] & (_mask)) >> (_shift))
-- 
Greetings Michael.

Re: [stable] ssb: 2.6.34.1 -> 2.6.34.2 regression

[Greg KH]

On Sat, Aug 07, 2010 at 12:53:06AM +0200, Michael Büsch wrote:
> Hi,
> 
> The SPROM changes applied to SSB in 2.6.34.2 completely broke SPROM
> extraction. This leads to broken wireless for my BCM4306 card.
> All other BCM43xx and BCM44xx most likely are affected, too.
> Symptoms should be wrong MAC addresses and broken transmission (for 43xx).
> 
> The patch changed the SPEX() macro and removed the offset subtraction,
> but did not remove the actual 0x1000 offset from the SPROM defines.
> This results in random memory poking rather than proper SPROM data
> extraction.
> 
> I applied the following to my kernel to get it working again. However
> this is not a proper fix. I suggest that the -stable SSB patch
> is dropped and mainline patches being applied. They seem to
> be correct (However I did not test mainline, yet).

That's already done, can you please test the 2.6.34.3-rc1 kernel that
was announced a few hours ago?  I should have fixed all of the issues
there, if not, please let me know.

thanks,
greg k-h

Re: [stable] ssb: 2.6.34.1 -> 2.6.34.2 regression

[Michael Büsch]

On 08/07/2010 01:56 AM, Greg KH wrote:
> On Sat, Aug 07, 2010 at 12:53:06AM +0200, Michael Büsch wrote:
>> Hi,
>>
>> The SPROM changes applied to SSB in 2.6.34.2 completely broke SPROM
>> extraction. This leads to broken wireless for my BCM4306 card.
>> All other BCM43xx and BCM44xx most likely are affected, too.
>> Symptoms should be wrong MAC addresses and broken transmission (for 43xx).
>>
>> The patch changed the SPEX() macro and removed the offset subtraction,
>> but did not remove the actual 0x1000 offset from the SPROM defines.
>> This results in random memory poking rather than proper SPROM data
>> extraction.
>>
>> I applied the following to my kernel to get it working again. However
>> this is not a proper fix. I suggest that the -stable SSB patch
>> is dropped and mainline patches being applied. They seem to
>> be correct (However I did not test mainline, yet).
>
> That's already done, can you please test the 2.6.34.3-rc1 kernel that
> was announced a few hours ago?  I should have fixed all of the issues
> there, if not, please let me know.

That one works. Thanks.

-- 
Greetings Michael.

Re: ssb: 2.6.34.1 -> 2.6.34.2 regression

[Rafał Miłecki]

2010/8/7 Michael Büsch <mb@bu3sch.de>:
> The SPROM changes applied to SSB in 2.6.34.2 completely broke SPROM
> extraction. This leads to broken wireless for my BCM4306 card.
> All other BCM43xx and BCM44xx most likely are affected, too.
> Symptoms should be wrong MAC addresses and broken transmission (for 43xx).
>
> The patch changed the SPEX() macro and removed the offset subtraction,
> but did not remove the actual 0x1000 offset from the SPROM defines.
> This results in random memory poking rather than proper SPROM data
> extraction.
>
> I applied the following to my kernel to get it working again. However
> this is not a proper fix. I suggest that the -stable SSB patch
> is dropped and mainline patches being applied. They seem to
> be correct (However I did not test mainline, yet).

The original patch didn't simply drop subtraction of offset, it also
modified defines in .h:
http://marc.info/?l=linux-wireless&m=127006884129062&w=4

It's Larry combined patch that you see in .34.2 and it indeed lacks
changes in defines. We already explained that situation, Greg will
revert it and apply correct patches instead.

-- 
Rafał