Actions
Forum Thread Details
  • State Not Answered
  • Replies 2 replies
  • Subscribers 314 subscribers
  • Views 247 views
  • Users 0 members are here
Related

MicroZed continuous dma

raymadigan

I am working on a project with the MicroZed that uses dma to transfer data from the PL to the PS.  The dma is configured as interrupt driven and uses scatter gather.  The dma is one direction, from the PL to the PS.

My solution generally works, but when the original set of Bd's are used, I try to recycle them, the dma continues as if nothing is wrong.  The problem is, it doesn't write the new data into the Bd.  When the interrupt occurs and it tells me which Bd to look at, the data is the same as it was for the first time it was used.  If I clear the data after I read it the first time, the data is always zero for subsequent transfers.

If anyone knows about this, please respond and I will submit code and more information.

  • 0

    Hello Raymadigan,

    Firstly, I would suggest that you start with the DMA engine located in the Supporting Documents of our Zynq HW Speedway

    http://zedboard.org/course/developing-zynq%C2%AE-7000-all-programmable-soc-hardware-vivado-20133-201441-20152-and-20162

    From there, make sure you check your parameters to ensure your configuration is correct. So in a way you are starting with a known working example and going from there. 

    Anotother suggestion I can offer is to CAN data in the PL and see what you get on the PS side. Put in a pattern that changes, that you know is triggering from say a button on your demo board. Then send one DMA dump, check, send a second DMA dump of a different pattern, check, do a third, etc.

    Thats how I would go about debugging your project.

    --Josh

  • 0 in reply to jafoste4

    I looked at the speedway designs and all of them send small blocks of data from the PL to the PS.  I need to send a continuous stream of data.  The example I am using came with the SDK and I figured out the reason all of the blocks were zero after the first time they were used.  After the block is used it has to have it's cache flushed before it can be reused.  

    At this point I can transfer around 10K blocks of 256 32 bit elements and am down to about 35 faults.  I flush the cache for the specific range for each interrupt and maybe I am missing some for some reason.  The faults are very sporadic and I can't find a pattern at this point.