Re: Adding pool_cache_invalidate_local() to pool_cache(9) API

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Fri, Oct 09, 2009 at 12:28:56AM -0700, Jason Thorpe wrote:
> [...]  If per-CPU cached copies are not included in that, then a subsequent 
> allocation could return an incorrectly-constructed object, leading to errant 
> behavior.  There are subsystems in the kernel that depend on the 
> pool_cache_invalidate() semantics I describe; see 
> arch/alpha/alpha/pmap.c:pmap_growkernel() for an example.  The L1 PTPs are 
> cached in constructed form (with the L1 PTEs for the kernel portion of the 
> address space already initialized).  If PTPs are added to the kernel pmap in 
> such a way as to require an additional L1 PTE to link them up, then 
> already-constructed-but-free L1 PTPs need to be discarded since they will be 
> missing part of the kernel's address space.


As pointed out by Jean-Yves, this is the documented behavior of
pool_cache_invalidate(). x86 pmap uses it in a similar way, and checks
that items allocated from the cache are not stale before use.

-- 
Manuel Bouyer, LIP6, Universite Paris VI.           
Manuel.Bouyer%lip6.fr@localhost
     NetBSD: 26 ans d'experience feront toujours la difference
--