Dear USB CAN mailing list members, In a previous thread reviewing my changes to stop a DMA memory leak in the gs_usb driver, a discussion ensued about why it (and other USB CAN drivers were using DMA in the first place. At the time, it was suggested that we perform an exploration to determine if there is any performance benefit to DMA or if the DMA allocations could be replaced with normal kmalloc( )s instead. A method for gathering this information was proposed here: https://lore.kernel.org/linux-can/CAMZ6Rq+FSzy5ijQZhYyVJrbe86U9faD5aPFO4cezNkN9G-USzQ@xxxxxxxxxxxxxx/ A member of my team found some time and resources to work on it, and so I'm communicating his work to the list as an RFC on his behalf. We are seeking input from the mailing list regarding his testing methods as well as his patch. The short summary is that he couldn't identify any meaningful change in performance on his x86-64 hardware. We should be able to also repeat these tests on our ARM products under and older kernel and perhaps also on more modern ARM hardware such as a Raspberry Pi. This might give us the confidence we need to consider removing DMA allocations from the gs_usb driver and perhaps others. His detailed notes may be found here: https://www.dropbox.com/s/c3gm3oomihdze1n/USB%20CAN%20Driver%20DMA%20Removal%20Testing%20Doc.pdf?dl=0 Feedback is very much welcome. Best, Rhett Aultman
