Doubts on async_tx xor offload for raid5/6

Doubts on async_tx xor offload for raid5/6

[Vikas Aggarwal]

Hi  list,
Will appreciate if someone can clear my doubts with "some theoretical
examples". I am trying to improve XOR/GF offload for a multi-core SoC
(currently it has single XOR/GF  channel).

1) What is the purpose of device_prep_dma_interrupt  callback ?

2) My driver currently polls for posting  xor  completions
(dma_cookie_complete) and does'nt use  device_prep_dma_interrupt
callback at all.  What  gains/loss in terms of  latency/cpu
idleness/WRITE throughput  I can expect  by implementing this callback
in my async_tx driver.

3) Purpose of DMA_ACK as I read - it is  for higher layers to inform
dma driver that descriptors can now be freed. Can someone explain this
with an example as applicable with raid5/6 clients.

4) With example - why dma_run_dependencies(tx) needed  after the
hardware engine post completion for a descriptor.

5) Purpose of  tx->callback(cb_arg) - again with an example from a
raid5/6 offload perspective.

I want to use offload engine efficiently with multithreaded raid5/6.

I tried to dig through dma/drivers, crypto/async_xor.c, online
archives, linux/Documentation etc but could not get satisfactory
answers. Added two mail-lists as saw issues being discussed on both
ones.

Thanks and
Best Regards
Vikas Aggarwal

Re: Doubts on async_tx xor offload for raid5/6

[Vinod Koul]

On Thu, Jun 02, 2016 at 11:54:45AM +0530, Vikas Aggarwal wrote:
> Hi  list,
> Will appreciate if someone can clear my doubts with "some theoretical
> examples". I am trying to improve XOR/GF offload for a multi-core SoC
> (currently it has single XOR/GF  channel).
> 
> 1) What is the purpose of device_prep_dma_interrupt  callback ?
> 
> 2) My driver currently polls for posting  xor  completions
> (dma_cookie_complete) and does'nt use  device_prep_dma_interrupt
> callback at all.  What  gains/loss in terms of  latency/cpu
> idleness/WRITE throughput  I can expect  by implementing this callback
> in my async_tx driver.
> 
> 3) Purpose of DMA_ACK as I read - it is  for higher layers to inform
> dma driver that descriptors can now be freed. Can someone explain this
> with an example as applicable with raid5/6 clients.
> 
> 4) With example - why dma_run_dependencies(tx) needed  after the
> hardware engine post completion for a descriptor.
> 
> 5) Purpose of  tx->callback(cb_arg) - again with an example from a
> raid5/6 offload perspective.
> 
> I want to use offload engine efficiently with multithreaded raid5/6.
> 
> I tried to dig through dma/drivers, crypto/async_xor.c, online
> archives, linux/Documentation etc but could not get satisfactory
> answers. Added two mail-lists as saw issues being discussed on both
> ones.

Did you read Documentation/crypto/async-tx-api.txt

Also you should refer to code in crpto for dmaengine usage and other
RAID examples.

It would help if you post the code and ask people what you are trying to
improve..

-- 
~Vinod