Williams, Dan J wrote:

The dmaengine subsystem collects and advertises dma channels for two
classes of users in the kernel, memory-to-memory offload and
traditional device-to-memory DMA.  The original design was driven by
the memory-to-memory case and is starting to show its limitations now
that more device-to-memory DMA users are being planned.  The primary
difference between the two classes is that memory-to-memory offload
is very amenable to channel sharing and is tolerant of dynamic
channel changes.  Compare this to the device-to-memory case where a
channel must be dedicated to a device and may have platform-specific
reasons why it cannot talk to a different device. 

This rework allows channels to be targeted to a public (mem-to-mem)
pool or be reserved for an exclusive private (dev-to-mem) allocation.
See [PATCH 1/13] for documentation of the changes.  A nice side
effect of the rework is: 

 24 files changed, 679 insertions(+), 1108 deletions(-)

All review welcomed, especially around the dma_slave changes, or
performance impacts of dma_find_channel.

These patches are currently on async_tx.git/upstream, and barring any
brown-paper-bag issues will move to linux-next via async_tx.git/next.

  git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx.git
upstream 

---
Dan Williams (13):
      dmaengine: kill enum dma_state_client
      dmaengine: remove 'bigref' infrastructure
      dmaengine: kill struct dma_client and supporting infrastructure
      dmaengine: replace dma_async_client_register with dmaengine_get
      atmel-mci: convert to dma_request_channel and down-level
      dma_slave dmatest: convert to dma_request_channel
      dmaengine: introduce dma_request_channel and private channels
      net_dma: convert to dma_find_channel
      dmaengine: provide a common 'issue_pending_all' implementation
      dmaengine: centralize channel allocation, introduce
      dma_find_channel dmaengine: up-level reference counting to the
      module level dmaengine: remove dependency on async_tx
      async_tx, dmaengine: document channel allocation and api rework

 Documentation/crypto/async-tx-api.txt |  135 +++----
 Documentation/dmaengine.txt           |    1
 arch/avr32/include/asm/atmel-mci.h    |    6
 arch/avr32/mach-at32ap/at32ap700x.c   |   15 -
 crypto/async_tx/async_tx.c            |  350 ------------------
 drivers/dma/Kconfig                   |    2
 drivers/dma/dmaengine.c               |  637
 +++++++++++++++++++++++---------- drivers/dma/dmatest.c             
 |  111 ++---- drivers/dma/dw_dmac.c                 |   28 -
 drivers/dma/fsldma.c                  |    3
 drivers/dma/ioat_dma.c                |    5
 drivers/dma/iop-adma.c                |   11 -
 drivers/dma/mv_xor.c                  |   11 -
 drivers/mmc/host/atmel-mci.c          |  103 +----
 include/linux/async_tx.h              |   17 -
 include/linux/dmaengine.h             |  148 ++------
 include/linux/dw_dmac.h               |   31 +-
 include/linux/netdevice.h             |    3
 include/net/netdma.h                  |   11 -
 net/core/dev.c                        |  148 --------
 net/ipv4/tcp.c                        |    5
 net/ipv4/tcp_input.c                  |    2
 net/ipv4/tcp_ipv4.c                   |    2
 net/ipv6/tcp_ipv6.c                   |    2
 24 files changed, 679 insertions(+), 1108 deletions(-)
 create mode 100644 Documentation/dmaengine.txt

Dan,

First of all sorry for delay in my feedback on this.
I have been absent for couple of weeks and last days
I was struggling with some ioatdma hot issues.

I have walked through these patches and generally
I like the whole redesign/reduction idea.
There are couple of concerns/questions (+ minor comments) from my side
to three of these patches: 3, 7 and 11.
I will send them in a minute.
The rest of them looks good to me at this point.

Next week I am planning to run some tests on my I/OAT setup,
to see how this new dmaengine would behave in this environment.
I will let you know.

Regards,
Maciej