Hello all,
First post, in at the deep end.

I want to be able to allocate a number of coherent DMA buffers
to be used as a cyclic ring buffer and am having trouble finding
an answer.

Some things I do not want to do or cannot do for various reasons.
I don't want to allocate the buffers at boot time.
I didn't particually want a large single cyclic buffer.
I can't afford to re-submit a sg list.

I can allocate a single cyclic buffer and get the callback notifications to work. I can make a scatted gather list, submit it and that all works.

What I really want to do is to make a cyclic scatter gather list and I just cant see how to do it, that is to say without delving into the dmaengine code.

Maybe I misunderstand dmaengine_prep_dma_cyclic. As far as I can see you can only submit a single buffer and set the callback granularty.

This is for ARM at91sam9x5 with Linux-3.18

This would be a kernel driver concern, and I feel quite certain that there will be existing kernel drivers that do this. Before striving to write code, find code that you can steal.