Re: [PATCH 08/19] dmaengine: enable multiple clients and operations

[Jeff Garzik <jeff@garzik.org>]

Dan Williams wrote:
> On 9/11/06, Roland Dreier <rdreier@cisco.com> wrote:
>>     Jeff> Are we really going to add a set of hooks for each DMA
>>     Jeff> engine whizbang feature?
> ...ok, but at some level we are going to need a file that has:
> EXPORT_SYMBOL_GPL(dma_whizbang_op1)
> . . .
> EXPORT_SYMBOL_GPL(dma_whizbang_opX)
> correct?

If properly modularized, you'll have multiple files with such exports.

Or perhaps you won't have such exports at all, if it is hidden inside a 
module-specific struct-of-hooks.


> I understand what you are saying Jeff, the implementation can be made
> better, but something I think is valuable is the ability to write
> clients once like NET_DMA and RAID5_DMA and have them run without
> modification on any platform that can provide the engine interface
> rather than needing a client per architecture
> IOP_RAID5_DMA...FOO_X_RAID5_DMA.

It depends on the situation.

The hardware capabilities exported by each platform[or device] vary 
greatly, not only in the raw capabilities provided, but also in the 
level of offload.

In general, we don't want to see hardware-specific stuff in generic 
code, though...


> Or is this an example of the where "Do What You Must, And No More"
> comes in, i.e. don't worry about making a generic RAID5_DMA while
> there is only one implementation existence?

> I also want to pose the question of whether the dmaengine interface
> should handle cryptographic transforms?  We already have Acrypto:
> http://tservice.net.ru/~s0mbre/blog/devel/acrypto/index.html.  At the
> same time since IOPs can do Galois Field multiplication and XOR it
> would be nice to take advantage of that for crypto acceleration, but
> this does not fit the model of a device that Acrypto supports.

It would be quite interesting to see where the synergies are between the 
two, at the very least.  "async [transform|sum]" is a superset of "async 
crypto" after all.

	Jeff