Simple, zero-copy DMA to/from userspace.
-
Specify which dmaengine-compatible DMA channels you'd like to create userspace-accessible device files for in your device tree:
ezdma0 { compatible = "ezdma"; dmas = <&loopback_dma 0 &loopback_dma 1>; dma-names = "loop_tx", "loop_rx"; ezdma,dirs = <2 1>; // direction of DMA channel: // 1 = RX (dev->cpu), 2 = TX (cpu->dev) };
Note that it's possible to specify multiple nodes which reference the ezdma module in their "compatible" strings -- you'll get an entry in /dev for every name you've specified in "dma-names" across all of your nodes.
-
After inserting the ezdma module, two devices, as named in your
dma-names
above, will become available:/dev/loop_tx /dev/loop_rx
-
Sending data is as simple as:
int tx_fd = open("/dev/loop_tx", O_WRONLY); int rx_fd = open("/dev/loop_rx", O_RDONLY); write(tx_fd, tx_buf, xfer_size); // send a DMA transaction read (rx_fd, rx_buf, xfer_size);
See Documentation/devicetree/bindings/dma/ezdma.txt for additional example info.
A Makefile for out-of-tree building is supplied. You just need to point it to the top-level directory of a kernel tree that you've already compiled.
make KERNELDIR=/path/to/your/kernel
Currently the Makefile assumes you want to cross-compile for ARM by default, but you're free to override the ARCH
and/or CROSS_COMPILE
variables on the command line or on in your environment. (I'd be interested to hear how it works on non-ARM platforms, as well!)
Once you've compiled, transfer the ezdma.ko module to your target system, and run:
insmod ezdma.ko
For each DMA channel you've added (in the "dma-names" field in the device tree), you should see a message like the following printed in dmesg:
ezdma: loop_tx (TX) available
ezdma: loop_rx (RX) available
The /dev/loop_tx
and /dev/loop_rx
devices should now be available, as shown in point 2 of the example above.
See examples/loopback for example C code, Zynq PL block diagram (TCL file for Vivado 2014.2), and example device tree entries used to perform a simple "loopback" test where AXI-Stream packets are written to and read from the Zynq PL. The example code here achieves ~20 MBytes/sec throughput, but that's sending and receiving 1 packet at a time (2048-byte packets, 100 MHz clock in the PL). Streaming (independently sending and receiving) could probably achieve at least double the throughput. Also, the packet size can be increased to increase overall througput (if your application allows it).
So far, I've used the ezdma module with:
- Xilinx AXI DMA on Zynq-7000 SoC. (ZedBoard)
I hope to add more CPUs/DMA devices above as this module gets used in other places. If you've had success on other platforms, let me know and I can post it here (or send a pull request with example code).
- Allow a forced transaction size to be specified in sysfs (such that DMA transfers are always performed in increments of the given size).
- This would be useful when sending AXI stream packets of a particular size and would allow simple usages like
cat packet_file > /dev/my_tx
orcat /dev/my_rx > packet_file
.
- This would be useful when sending AXI stream packets of a particular size and would allow simple usages like
- Add user-space scatter-gather
readv()
/writev()
support.
In the future, I hope to refine and contribute ezdma to the mainline kernel.
Copyright (C) 2015 Jeremy Trimble
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, see http://www.gnu.org/licenses/.