spi: stm32-qspi: Trigger DMA only if more than 4 bytes to transfer
In order to optimize accesses to spi flashes, trigger a DMA only
if more than 4 bytes has to be transferred.
DMA transfer preparation's cost becomes negligible above 4 bytes to
transfer. Below this threshold, indirect transfer give more throughput.
mtd_speedtest shows that page write throughtput increases :
- from 779 to 853 KiB/s (~9.5%) with s25fl512s SPI-NOR.
- from 5283 to 5666 KiB/s (~7.25%) with Micron SPI-NAND.
Signed-off-by: Christophe Kerello <christophe.kerello@foss.st.com>
Signed-off-by: Patrice Chotard <patrice.chotard@foss.st.com>
Link: https://lore.kernel.org/r/20210419121541.11617-3-patrice.chotard@foss.st.com
Signed-off-by: Mark Brown <broonie@kernel.org>
diff --git a/drivers/spi/spi-stm32-qspi.c b/drivers/spi/spi-stm32-qspi.c
index 2786470..6e74d6b 100644
--- a/drivers/spi/spi-stm32-qspi.c
+++ b/drivers/spi/spi-stm32-qspi.c
@@ -269,8 +269,9 @@ static int stm32_qspi_tx(struct stm32_qspi *qspi, const struct spi_mem_op *op)
if (qspi->fmode == CCR_FMODE_MM)
return stm32_qspi_tx_mm(qspi, op);
- else if ((op->data.dir == SPI_MEM_DATA_IN && qspi->dma_chrx) ||
- (op->data.dir == SPI_MEM_DATA_OUT && qspi->dma_chtx))
+ else if (((op->data.dir == SPI_MEM_DATA_IN && qspi->dma_chrx) ||
+ (op->data.dir == SPI_MEM_DATA_OUT && qspi->dma_chtx)) &&
+ op->data.nbytes > 4)
if (!stm32_qspi_tx_dma(qspi, op))
return 0;
