Re: Trouble with make-install from a NFS mount

[Gregory Haskins]

Randy Dunlap wrote:
> Gregory Haskins wrote:
>   
>> Randy Dunlap wrote:
>>     
>>> Gregory Haskins wrote:
>>>   
>>>       
>>>>
>>>> Looking into this, I couldnt track down where the "filechk" magic was
>>>> coming from (any hints from the kbuild gurus?).  So I thought that I
>>>>         
>
> filechk is defined in scripts/Kbuild.include .
>
>   

ah, thx.

>>>
>>> See this patch, already applied/merged:
>>>
>>> Gitweb:     http://git.kernel.org/linus/556b0f58bbcdc96ba8ed67001b4e57c50198da89
>>> Commit:     556b0f58bbcdc96ba8ed67001b4e57c50198da89
>>> Parent:     8e320d02718d2872d52ef88a69a493e420494269
>>> Author:     David Woodhouse <David.Woodhouse@xxxxxxxxx>
>>> AuthorDate: Sat Jan 10 14:53:15 2009 +0000
>>> Committer:  David Woodhouse <David.Woodhouse@xxxxxxxxx>
>>> CommitDate: Mon Apr 6 14:27:17 2009 -0700
>>>
>>>     Revert "fix modules_install via NFS"
>>>     
>>>     This reverts commit 8b249b6856f16f09b0e5b79ce5f4d435e439b9d6.
>>>     
>>>     This 'fix' is not necessary; we just need to undo the damage caused
>>>     accidentally by Igor/Mauro in 4b29631db33292d416dc395c56122ea865e7635c
>>>     ("V4L/DVB (9533): cx88: Add support for TurboSight TBS8910 DVB-S PCI card")
>>>     
>>>     Signed-off-by: David Woodhouse <David.Woodhouse@xxxxxxxxx>
>>>
>>>
>>>   
>>>       
>> Thanks Randy.  I was running on 0882e8dd3aad33eca41696d463bb896e6c8817eb
>> so I had these patches all applied.  I think you were trying to tell me
>> that perhaps a similar fix could be found among these patches, but I was
>> unfortunately not able to find any and this is still broken for me.
>>     
>
> No, I wasn't trying to say that. 
Ah, sorry for the confusion.

>  Your very first paragraph said
> "regression in 30-rc1" and David's patch fixes that problem.
>
> So is there still some other problem in linus-current?
>
>   
Yes, this is still broken in current, though I think its a slightly
different problem than the one David addressed.  I think he addressed
"make modules_install", whereas I see the issue with "make install".

The problem seems to be that 30-rcX thinks that there are dependencies
to update (such as include/config/kernel.release and version.h.tmp)
during a make-install operation.  Since I am on a root-squashed NFS
mount, these files cannot be updated and thus the install fails. 
Previous trees prior to linux-current (at least up to 2.6.29) did not
try to update these files so the install would succeed.

FWIW, I ran the make with -d, and at least part of the reason for the
dep-failure was the FORCE psuedo dependency against kernel.release and
version.h.  However, I didn't find that FORCE was a recent addition to
the Makefile so I am not sure why the behavior suddenly changed.

I can try things to debug this for anyone who is interested in sending
me suggestions.  However, I am currently unblocked by the issue since I
added the posted "push" script to my workflow, so this is not critical
to me anymore.

-Greg

Attachment: signature.asc
Description: OpenPGP digital signature