Re: DMA Engine: Transfer From Userspace

[Thomas Ruf]

> On 26 June 2020 at 12:29 Peter Ujfalusi <peter.ujfalusi@xxxxxx> wrote:
> 
> On 24/06/2020 16.58, Thomas Ruf wrote:
> > 
> >> On 24 June 2020 at 14:07 Peter Ujfalusi <peter.ujfalusi@xxxxxx> wrote:
> >> On 24/06/2020 12.38, Vinod Koul wrote:
> >>> On 24-06-20, 11:30, Thomas Ruf wrote:
> >>>
> >>>> To make it short - i have two questions:
> >>>> - what are the chances to revive DMA_SG?
> >>>
> >>> 100%, if we have a in-kernel user
> >>
> >> Most DMAs can not handle differently provisioned sg_list for src and dst.
> >> Even if they could handle non symmetric SG setup it requires entirely
> >> different setup (two independent channels sending the data to each
> >> other, one reads, the other writes?).
> > 
> > Ok, i implemented that using zynqmp_dma on a Xilinx Zynq platform (obviously ;-) and it works nicely for us.
> 
> I see, if the HW does not support it then something along the lines of
> what the atc_prep_dma_sg did can be implemented for most engines.
> 
> In essence: create a new set of sg_list which is symmetric.

Sorry, not sure if i understand you right?
You suggest that in case DMA_SG gets revived we should restrict the support to symmetric sg_lists?
Just had a glance at the deleted code and the *_prep_dma_sg of these drivers had code to support asymmetric lists and by that "unaligend" memory (relative to page start):
at_hdmac.c         
dmaengine.c        
dmatest.c          
fsldma.c           
mv_xor.c           
nbpfaxi.c          
ste_dma40.c        
xgene-dma.c        
xilinx/zynqmp_dma.c

Why not just revive that and keep this nice functionality? ;-)

> > Don't think that it uses two channels from what a saw in their implementation.
> 
> I believe it was breaking it up like atc_prep_dma_sg did.
> 
> > Of course that was on kernel 4.19.x where DMA_SG was still available.
> > 
> >>>> - what are the chances to get my driver for memcpy like transfers from
> >>>> user space using DMA_SG upstream? ("dma-sg-proxy")
> >>>
> >>> pretty bleak IMHO.
> >>
> >> fwiw, I also get requests time-to-time to DMA memcpy support from user
> >> space from companies trying to move from bare-metal code to Linux.
> >>
> >> What could be plausible is a generic dmabuf-to-dmabuf copy driver (V4L2
> >> can provide dma-buf, DRM can also).
> >> If there is a DMA memcpy channel available, use that, otherwise use some
> >> method to do the copy, user space should not care how it is done.
> > 
> > Yes, i'm using it together with a v4l2 capture driver and also saw the dma-buf thing but did not find a way how to bring this together with "ordinary user memory".
> 
> One of the aim of dma-buf is to share buffers between drivers and user
> space (among drivers and/or drivers and userspace), but I might be
> missing something.
> 
> > For me the root of my problem seems to be that dma_alloc_coherent leads to uncached memory on ARM platforms.
> 
> It depends, but in most cases that is true.
> 
> > But maybe i am doing it all wrong ;-)
> > 
> >> Where things are going to get a bit more trickier is when the copy needs
> >> to be triggered by other DMA channel (completion of a frame reception
> >> triggering an interleaved sub-frame extraction copy).
> >> You don't want to extract from a buffer which can be modified while the
> >> other channel is writing to it.
> > 
> > I think that would be no problem in case of our v4l2 capture driver doing both DMAs:
> > Framebuffer DMA for streaming and Zynqmp DMA (using DMA_SG) to get it to "ordinary user memory".
> > But as i wrote before i prefer to do the "logic and management" in userspace so the capture driver is just using the first DMA and the "dma-sg-proxy" driver is only used as a memcpy replacement.
> > As said this is all working fine with kernel 4.19.x but now we are stuck :-(
> > 
> >> In Linux the DMA is used for kernel and user space can only use it
> >> implicitly via standard subsystems.
> >> Misused DMA can be very dangerous and giving full access to program a
> >> transfer can open a can of worms.
> > 
> > Fully understand that!
> > But i also hope you understand that we are developing a "closed system" and do not have a problem with that at all.
> > We are also willing to bring that driver upstream for anyone doing the same but of course this should not affect security of any desktop or server systems.
> > Maybe we just need the right place for that driver?!
> 
> What might be plausible is to introduce hw offloading support for memcpy
> type of operations in a similar fashion how for example crypto does it?

Sounds good to me, my proxy driver implementation could be a good start for that, too!
 
> The issue with a user space implemented logic is that it is not portable
> between systems with different DMAs. It might be that on one DMA the
> setup takes longer than do a CPU copy of X bytes, on the other DMA it
> might be significantly less or higher.

Fully agree with that!
I was also unsure how my approach will perform but in our case the latency was increased by ~20%, cpu load roughly stayed the same, of course this was the benchmark from user memory to user memory.
>From uncached to user memory the DMA was around 15 times faster.

> Using CPU vs DMA for a copy in certain lengths and setups should not be
> a concern of the user space.

Also fully agree with that!

> Yes, you have a closed system with controlled parameters, but a generic
> mem2mem_offload framework should be usable on other setups and the same
> binary should be working on different DMAs where one is not efficient
> for <512 bytes, the other shows benefits under 128bytes.

Usable: of course
"Faster": not necessarily as long as it is an option

Thanks for your valuable input and suggestions!

best regards,
Thomas