Re: 2.6.16-rc1-mm3

From: Eric Dumazet <dada1@cosmosbay.com>
To: Andy Whitcroft <apw@shadowen.org>
Cc: Andrew Morton <akpm@osdl.org>,
	penberg@cs.helsinki.fi, linux-kernel@vger.kernel.org
Subject: Re: 2.6.16-rc1-mm3
Date: Fri, 27 Jan 2006 11:37:49 +0100	[thread overview]
Message-ID: <43D9F7FD.8010102@cosmosbay.com> (raw)
In-Reply-To: <43D9F20F.1000906@shadowen.org>

Andy Whitcroft a écrit :
> Eric Dumazet wrote:
>> The NULL choice was maybe wrong. We might need more than one page to
>> fully catch all accesses. Something like 32KB.
> 
> The crash behavoir is handy to catch that the problem exists, and is
> very cheap (0 cost) at run time.  However, once its known I think we
> need something more targetted to allow tracking of the cause.  Perhaps
> we could set the offset thingy to -1 or something and simply do
> something like the following in per_cpu():
> 
> 	if (__per_cpu_offset[i] == -1)
> 		BUG();
> 	else
> 		*RELOC_HIDE(&per_cpu__##var, __per_cpu_offset[cpu])
> 

Yes we can set __per_cpu_offset[not_possible_cpu] to 0, because 
[__per_cpu_start,__per_cpu_end) is in init section and should be discarded in 
free_initmem(). I'm not sure if the freed virtual space can later be reused.

Eric