Re: 2.6.22-rc4-mm1

["Dan Williams" <dan.j.williams@intel.com>]

On 6/7/07, Andrew Morton <akpm@linux-foundation.org> wrote:
> On Thu, 7 Jun 2007 17:01:08 +1000 Herbert Xu <herbert@gondor.apana.org.au> wrote:
>
> > On Thu, Jun 07, 2007 at 08:54:50AM +0200, Jan Engelhardt wrote:
> > >
> > > /me points at Herbert
> > > Andrew would not add options between the "menuconfig CRYPTO" and
> > > the "if CRYPTO" line... :)
> >
> > Actually this patch is not even in my tree :)
>
> uh, OK, sorry.
>
> > > Index: linux-2.6.22-rc4/crypto/Kconfig
> > > ===================================================================
> > > --- linux-2.6.22-rc4.orig/crypto/Kconfig
> > > +++ linux-2.6.22-rc4/crypto/Kconfig
> > > @@ -7,6 +7,8 @@ menuconfig CRYPTO
> > >     help
> > >       This option provides the core Cryptographic API.
> > >
> > > +if CRYPTO
> > > +
> > >  #
> > >  # Generic algorithms support
> > >  #
> > > @@ -18,8 +20,6 @@ config XOR_BLOCKS
> > >  #
> > >  source "crypto/async_tx/Kconfig"
> >
> > Andrew, do you want me to pick the async_tx stuff up instead?
> >
>
It would be very helpful to have a clear merge path for dmaengine
changes and the async offload api.  Neil has been extremely helpful
reviewing the raid specific changes, and I received his "Acked-by" for
the changes to MD[1].  However I have thus far been unable to attract
someone to 'ack/nak' the async_tx api and the changes to drivers/dma/
[2].  Jeff commented on an early revision...

I have recently gravitated to Herbert and the crypto directory since
async_tx and crypto have some structural similarities [3].

> I wouldn't recommend it.  It's an ongoing source of bustage frankly, has a
> habit of getting unpleasantly tangled with git-ioat.patch and afaik is
> still awaiting a go-ahead from Neil.
>

Sorry, the crypto/Kconfig bustage was a goof on my part as I moved the
async_tx files from drivers/dma/, to the top-level directory, and
finally to crypto/.  Hopefully these recent build breakages I have
caused in -mm have not put the series in too negative a light...

I was hoping the git-ioat.patch situation would be solved by me
rebasing my series on a version of mainline with Chris' changes
merged, but his attempts over the past two merge windows were ignored.
 Should my series wait outside of -mm until git-ioat.patch makes
forward progress?

Overall, I feel that async_tx is perhaps justifiably receiving the
silent treatment because offload engines are not a mainstream
occurrence.  Currently only people with an Xscale IOP or a PPC 440spe
[4] will notice that mainline lacks support for all the features of
their platform.  I see async_tx as a nod to the embedded space where
offload engines act to make up for the absence of multi-Ghz CPUs with
streaming SIMD instructions.

Herbert's offer is greatly appreciated as it will give guidance to the
parts of the series outside of Neil's purview.

Regards,
Dan

[1]: The ack from Neil was in an offlist message for the MD specific
portion of the series
[2]: I asked DaveM and netdev to take a look at the two patches in the
series that change drivers/dma/ and net/core/dev.c since that was the
original merge path for I/OAT and dmaengine
[3]: async_tx is similar to crypto in that they both provide a library
of memory transforms that can in some cases be carried out by
hardware.
[4]: async_tx has attracted at least one other developer that I know
about to write a driver for their engines:
http://marc.info/?l=linux-raid&m=117400143317440&w=2