On 2019-04-26 at 11:37 +0200, Lucas Stach wrote: > Am Freitag, den 26.04.2019, 09:22 +0000 schrieb Robin Gong: > > > > On 2019-04-26 at 11:07 +0200, Lucas Stach wrote: > > > > > > > > > Am Freitag, den 26.04.2019, 08:05 +0000 schrieb Robin Gong: > > > > > > > > static void mx51_setup_wml(struct spi_imx_data *spi_imx) > > > > { > > > > + u32 tx_wml = 0; > > > > + > > > With a wml of 0 you might set the maxburst of the TX dma channel > > > to > > > fifosize to minimize the performance impact of this workaround. > > > > > > Regards, > > > Lucas > > Unfortunately, this is a MUST part of errata which cause > > performance > > drop. > I'm not talking about changing the WML, but when the WML must be 0 > you > can safely increase the DMA burst size without overflowing the FIFO, > which might recover some of of the performance loss. > > Regards, > Lucas Good point, will add it in V3, thanks.