Re: [PATCH 01/04] Adding cipher mode context information to crypto_tfm

[Fruhwirth Clemens <clemens@endorphin.org>]

On Tue, 2005-02-08 at 19:09 -0500, James Morris wrote:
> On Wed, 9 Feb 2005, Fruhwirth Clemens wrote:
> 
> > > You can't call kmap() in softirq context (why was it even trying?):
> > 
> > Why not? What's the alternative, then?
> 
> It can sleep in map_new_virtual().
> 
> The alternative is to use atomic kmaps.  For this code, unless you can 
> point to something concrete in the existing kernel which would benefit 
> from passing an arbitrary number of scatterlists in, just code for the 
> case of processing two at once (input & output).

kmap_atomic doesn't qualify as alternative. It's limited to two free
mappings. There must be something else to kmap in softirq. I really
can't imagine such a limitation.

Where is the documentation for the highmem stuff anyway? It's stunning
that there is not a single useful hit in ./Documentation for kmap_atomic
or kmap. The web is also close-lipped about that issue.

I can't code for the case of two. Because, first, that's the idea of
generic in the name "generic scatterwalk", second, I need at least 3
scatterlists in parallel for LRW.

-- 
Fruhwirth Clemens <clemens@endorphin.org>  http://clemens.endorphin.org