Hi Greg, Jiri, When submitting a DMA request fails, the sh-sci driver is supposed to fall back to PIO. However, this never really worked due to various reasons (sh-sci driver issues and dmaengine framework limitations). There are three places where DMA submission can fail, and the driver should fall back to PIO: 1. sci_dma_rx_complete(), 2. sci_submit_rx(), 3. work_fn_tx(). This patch series fixes fallback to PIO in the receive path (cases 1 and 2). Fallback to PIO in the transmit path (case 3) already seems to work fine. Changes compared to v3: - Let sci_submit_rx() return -EAGAIN instead of -1 on failure, - Check for negative error in sci_submit_rx() caller. Changes compared to v2: - Add missing definition of "u16 scr" to sci_dma_rx_complete(), - Move label handle_pio inside #ifdef to kill defined but not used compiler warning when CONFIG_SERIAL_SH_SCI_DMA=n, - Move call to dmaengine_terminate_async() in sci_dma_rx_complete() inside the spinlock, for symmetry with sci_submit_rx(), - Move the call sci_submit_rx() in sci_rx_interrupt() up, as it may fail, rendering the modification of scr unused, - Split in multiple patches, - Drop RFC status. Changes compared to v1: - Fix fallback in sci_dma_rx_complete(), - Fallback in the transmit path already works fine, - Widen audience, but keep RFC. This has been tested on r8a7791/koelsch, using SCIF1 on debug serial 1, and SCIFA3 on EXIO-B, by introducing random failures in DMA submission code. Thanks! Geert Uytterhoeven (4): serial: sh-sci: Fix locking in sci_submit_rx() serial: sh-sci: Fix crash in rx_timer_fn() on PIO fallback serial: sh-sci: Resume PIO in sci_rx_interrupt() on DMA failure serial: sh-sci: Fix fallback to PIO in sci_dma_rx_complete() drivers/tty/serial/sh-sci.c | 39 +++++++++++++++++++++++++++---------- 1 file changed, 29 insertions(+), 10 deletions(-) -- 2.17.1 Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds