On Tuesday, November 11, 2014 4:35 AM, Ian Abbott wrote: > On 11/11/14 00:57, H Hartley Sweeten wrote: >> Currently the DMA buffers are switched before writing the current samples to >> the async buffer. This works but when the EOA event happens we end up with an >> outstanding DMA operation in progress that gets terminated by the (*cancel). >> >> Avoid the outstanding DMA operation by switching the DMA buffers after writing >> the samples. The driver will detect the EOA event and not retart the DMA. > > I think the DMA buffer is switched early to try and avoid overruns in > the hardware while the other buffer is being processed (by > defragment_dma_buffer() and comedi_buf_write_samples()), so this patch > is likely to reduce the maximum error-free throughput. > > The combination of defragment_dma_buffer() and > comedi_buf_write_samples() is a bit inefficient anyway, so perhaps we > can drop this patch for now, and do a "no switch needed" test as part of > another set of changes. (I can work on that if you want.) I'm fine with that. I assume you are ok with the other 3 patches? Greg, please drop this patch (4/4). Thanks, Hartley _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
