On Fri, Jan 11, 2013 at 03:13:31PM +0900, Chanho Min wrote: > In DMA support, Received data is not triggered until the DMA buffer is filled. > In order to actually use Rx DMA, We would like to suggest the use of the timer > for polling DMA buffer. It makes possible character-level trigger. > In our test, no data loss occurred at high-baudrate as compared with interrupt > driven (We tested with 3Mbps). We know it will be because of the increased Rx > buffer rather than DMA effect. but,It is very useful for high speed uart device. > We changes: > > - CONFIG_SERIAL_AMBA_PL011_DMAPOLL is added to select this feature. > > - add timer and last_residue for polling. Every polling, timer handler checks > the residue in the dma buffer and transfer data to the tty. Also, last_residue > is updated for the next polling. > > - We use consistent DMA mappings instead of streaming DMA mappings to avoid from > the frequent cache operation. > > - pl011_dma_rx_chars is modified. the pending size is recalculated because data > can be taken by polling. > > - add module parameter for adjusting polling interval. Here's the question: what is the effect of having to poll every tick on the overall system? What I'm getting at is: what is the cost of that polling with DMA enabled over not having DMA on the receive side at all? What is the trade-off vs baud rate? Should we scale the polling interval according to baud rate? -- To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
