linux-kernel - Re: [Cbe-oss-dev] [RFC, PATCH] CELL Oprofile SPU profiling updated patch

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <45CB3BDF.4080408@us.ibm.com>
Date:	Thu, 08 Feb 2007 09:03:59 -0600
From:	Maynard Johnson <maynardj@...ibm.com>
To:	michael@...erman.id.au
CC:	Carl Love <cel@...ibm.com>, linuxppc-dev@...abs.org,
	linux-kernel@...r.kernel.org, cbe-oss-dev@...abs.org
Subject: Re: [Cbe-oss-dev] [RFC, PATCH] CELL Oprofile SPU profiling updated
 patch

Michael,
Thanks very much for the advice.  Both issues have been solved now, with 
your help.

-Maynard

Michael Ellerman wrote:

>On Wed, 2007-02-07 at 09:41 -0600, Maynard Johnson wrote:
>  
>
>>Carl Love wrote:
>>
>>    
>>
>>>Subject: Add support to OProfile for profiling Cell BE SPUs
>>>
>>>From: Maynard Johnson <maynardj@...ibm.com>
>>>
>>>This patch updates the existing arch/powerpc/oprofile/op_model_cell.c
>>>to add in the SPU profiling capabilities.  In addition, a 'cell' subdirectory
>>>was added to arch/powerpc/oprofile to hold Cell-specific SPU profiling
>>>code.
>>>
>>>Signed-off-by: Carl Love <carll@...ibm.com>
>>>Signed-off-by: Maynard Johnson <mpjohn@...ibm.com>
>>>
>>>Index: linux-2.6.20-rc1/arch/powerpc/configs/cell_defconfig
>>> 
>>>
>>>      
>>>
>>I've discovered more problems with the kref handling for the cached_info 
>>object that we store in the spu_context.  :-(
>>
>>When the OProfile module initially determines that no cached_info yet 
>>exists for a given spu_context, it creates the cached_info, inits the 
>>cached_info's kref (which increments the refcount) and does a kref_get 
>>(for SPUFS' ref) before passing the cached_info reference off to SUPFS 
>>to store into the spu_context.  When OProfile shuts down or the SPU job 
>>ends, OProfile gives up its ref to the cached_info with kref_put.  Then 
>>when SPUFS destroys the spu_context, it also gives up its ref.  HOWEVER 
>>. . . . If OProfile shuts down while the SPU job is still active _and_ 
>>_then_ is restarted while the job is still active, OProfile will find 
>>that the cached_info exists for the given spu_context, so it won't go 
>>through the process of creating it and doing kref_init on the kref.  
>>Under this scenario, OProfile does not own a ref of its own to the 
>>cached_info, and should not be doing a kref_put when done using the 
>>cached_info -- but it does, and so does SPUFS when the spu_context is 
>>destroyed.  The end result (with the code as currently written) is that 
>>an extra kref_put is done when the refcount is already down to zero.  To 
>>fix this, OProfile needs to detect when it finds an existing cached_info 
>>already stored in the spu_context.  Then, instead of creating a new one, 
>>it sets a reminder flag to be used later when it's done using the cached 
>>info to indicate whether or not it needs to call kref_put.
>>    
>>
>
>I think all you want to do is when oprofile finds the cached_info
>already existing, it does a kref_get(). After all it doesn't have a
>reference to it, so before it starts using it it must inc the ref count.
>
>  
>
>>Unfortunately, there's another problem (one that should have been 
>>obvious to me).  The cached_info's kref "release" function is 
>>destroy_cached_info(), defined in the OProfile module.  If the OProfile 
>>module is unloaded when SPUFS destroys the spu_context and calls 
>>kref_put on the cached_info's kref -- KABOOM!  The destroy_cached_info 
>>function (the second arg to kref_put) is not in memory, so we get a 
>>paging fault.  I see a couple options to solve this:
>>    1) Don't store the cached_info in the spu_context.  Basically, go 
>>back to the simplistic model of creating/deleting the cached_info on 
>>every SPU task activation/deactivation.
>>    2)  If there's some way to do this, force the OProfile module to 
>>stay loaded until SPUFS has destroyed its last spu_context that holds a 
>>cached_info object.
>>    
>>
>
>There is a mechanism for that, you just have each cached_info inc the
>module's refcount.
>
>Another option would be to have a mapping, in the oprofile code, from
>spu_contexts to cached_infos, ie. a hash table or something.
>
>cheers
>
>  
>


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/