Hi, On Mon, Jun 07, 2010 at 20:39:05, Brian Niebuhr wrote: > NOTE > > This patch requires the EDMA patch at: > > http://linux.davincidsp.com/pipermail/davinci-linux-open-source/2010-March/018022.html > > which has since been reverted out of the davinci kernel. > > > INTRODUCTION > > I have been working on a custom OMAP-L138 board that has multiple spi > devices (seven) on one controller. These devices have a wide range of > transfer parameters (speed, phase, polarity, internal and gpio chip > selects). During my testing I found multiple errors in the davinci spi > driver as a result of this complex setup. The primary issues were: > > 1. There is a race condition due to the SPIBUF read busy-waits for slow > devices > 2. I found some DMA transfer length errors under some conditions > 3. The chip select code caused extra byte transfers (with no chip > select active) due to writes to SPIDAT1 > 4. Several issues prevented using multiple SPI devices, especially > the DMA code, as disucussed previously on the davinci list. > > The fixes to these problems were not simple. I ended up making fairly > large changes to the driver, and those changes are contained in these > patches. The full list of changes follows. > > CHANGE LIST > > 1. davinci_spi_chipelect() now performs both activation and deactivation > of chip selects. This lets spi_bitbang fully control chip > select activation, as intended by the SPI API. > 2. Chip select activation does not cause extra writes to the SPI bus > 3. Chip select activation does not use SPIDEF for control. This change > will also allow for implementation of inverted (active high) > chip selects in the future. > 4. Added back gpio chip select capability from the old driver > 5. Fixed prescale calculation for non-integer fractions of spi clock > 6. Allow specification of SPI transfer parameters on a per-device > (instead of per-controller) basis > 7. Allow specification of polled, interrupt-based, or DMA operation on > a per-device basis > 8. Allow DMA with when more than one device is connected > 9. Combined pio and dma txrx_bufs functions into one since they share > large parts of their functionality, and to simplify item (8). > 10. Use only SPIFMT0 to allow more than 4 devices > > TESTING > > I have tested the driver using a custom SPI stress test on my > OMAP-L138-based board with three devices connected. I have tested > configurations with all three devices polled, all three interrupt-based, > all three DMA, and a mixture. > > I have compiled with the davinci_all_defconfig, but I don't have EVMs > for the other davinci platforms to test with. >
I was testing this series on OMAP-L137, OMAP-L138 and DM365 EVMs and I found that on all these EVMS, Interrupt mode has problems. Driver works fine in DMA and POLL modes. Some problems which I had observed with the old driver are gone now. Regards, Sudhakar _______________________________________________ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
