Hi, On Thu, Dec 8, 2022 at 6:25 AM Vijaya Krishna Nivarthi <quic_vnivarth@xxxxxxxxxxx> wrote: > > SE DMA mode can be used for larger transfers and FIFO mode > for smaller transfers. > > Signed-off-by: Vijaya Krishna Nivarthi <quic_vnivarth@xxxxxxxxxxx> > --- > v4 -> v3: > - remove cur_m_cmd field > - add dma_addr fields to spi_geni_master > - dont use dma_addr from xfer, use above > - dev_warn instead of dev_err in handle_timeout > - set xfer_mode to FIFO in set_cs > > v2 -> v3: > - indentation change > - readability change > > v1 -> v2: > - replace writel_relaxed with writel > - corrected order of performing FSM reset and dma_unprep > - added more comments in geni_can_dma > - removed redundant change in spi_geni_init(DMA as default xfer_mode) > - removed redundant initialisations in setup_se_xfer > - removed redundant null checks in setup_se_xfer > - added dma_tx_prep failure handling(rx un_map) in setup_se_xfer > - move handing return of setip_se_xfer to transfer_one > - apply irq error handling for DMA mode too in geni_spi_isr > --- > drivers/spi/spi-geni-qcom.c | 211 ++++++++++++++++++++++++++++++++++---------- > 1 file changed, 165 insertions(+), 46 deletions(-) This looks reasonable to me now, thanks. Reviewed-by: Douglas Anderson <dianders@xxxxxxxxxxxx> A note to Mark since he was also concerned about the fact that we aren't using the SPI framework to manage the DMA [1] and Vijay didn't respond directly to him on this issue. The issue here has to do with the fact that the driver wanted to use geni_se_tx_dma_prep() / geni_se_rx_dma_prep() and that function overlaps with what the SPI framework does. Eventually it seems like geni_se_tx_dma_prep() / geni_se_rx_dma_prep() should be reworked so as not to overlap, but I agree with Vijay that it can be future work. I'm at least happy now that the driver is using its own storage for the dma_addr_t variables now.Obviously, though, the final say is yours. :-) [1] https://lore.kernel.org/r/Y4kxVP97C66oi0Bi@xxxxxxxxxxxxx
