On Thu, Jul 23, 2015 at 05:10:42PM +0800, Leilk Liu wrote: This basically seems fine but there's a couple of issues that should be relatively easy to fix: > + mdata->cur_transfer = xfer; > + mtk_spi_prepare_transfer(master, xfer); > + mtk_spi_setup_packet(master, xfer); > + > + cnt = (xfer->len % 4) ? (xfer->len / 4 + 1) : (xfer->len / 4); Please write this as an if statement for legibility. > +static bool mtk_spi_can_dma(struct spi_master *master, > + struct spi_device *spi, > + struct spi_transfer *xfer) > +{ > + struct mtk_spi *mdata = spi_master_get_devdata(master); > + > + if (xfer->len > MTK_SPI_MAX_FIFO_SIZE) > + mdata->use_dma = true; > + else > + mdata->use_dma = false; > + > + return mdata->use_dma; > +} This is broken since can_dma() can be called multiple transfers before actually doing a transfer (the current implementation loops over all transfers in a message before starting the message) - you can't store any local data. The transfer_one() function should do another can_dma() check to decide if it can DMA, it shouldn't rely on driver global data. > + if (!mdata->use_dma) { > + if (trans->rx_buf) { This should be a variable set when doing the transfer (or perhaps based on checking spi->cur_xfer with can_dma()). > + for (i = 0; i < trans->len; i++) { > + if (i % 4 == 0) > + reg_val = > + readl(mdata->base + SPI_RX_DATA_REG); > + *((u8 *)(trans->rx_buf + i)) = > + (reg_val >> ((i % 4) * 8)) & 0xff; This isn't the clearest code ever... a comment would help. > + if (mdata->tx_sgl && (mdata->tx_sgl_len == 0)) { > + mdata->tx_sgl = sg_next(mdata->tx_sgl); > + if (mdata->tx_sgl) { > + trans->tx_dma = sg_dma_address(mdata->tx_sgl); > + mdata->tx_sgl_len = sg_dma_len(mdata->tx_sgl); > + } > + } There's a *lot* of code in this interrupt handler, and a lot of it looks an awful lot like the contents of mtk_spi_dma_transfer() has been cut'n'pasted in. The shared code should all be factored out into a function called from both places.
Attachment:
signature.asc
Description: Digital signature