Dear all, I'm trying to get optimal results with a XVME6300 Single Board Computer, using the VME_TSI148 driver from Martyn Welch, combined with the VME_USER device driver. So far, I can read/write a block of data (eg. 64 bytes, 16 LWORDS) from/to a slave on the VME-bus, but I encounter the following problems (as seen with a VME analyzer in the crate): - in SCT mode, each LWORD being read is separated approx 3 microseconds from eachother, while the duration for the read itself only takes 0.5 microseconds. In other words: there's lots of unused bandwith, causing slow performance. What could be causing the delay of 2.5 microseconds ? - The TSI148 bridge is using PCI-X on my-board, and is behind a PCIe-PCIX bridge. According to the TSI148 manual, every read will result in a 'split read', and the TSI148 should take further care of the read command. Does this mean, that the TSI148 will ensure that all my LWORDs are being read if I do a read of multiple bytes (eg. 64) ? If so, can the TSI148 itself be responsible for the 3 microseconds ? - BLT mode is no different than SCT mode according to the signal charts I get from the analyser? Each read/write modifies the address-lines, which shouldn't be the case, right? Could it be my slave card does not support BLT transfers? Apart from this: - Could somebody show me how to do DMA access with the VME_TSI148 driver ? - Is it realistic to modify the VME_USER driver to add DMA capabilities ? PS: using simple access to /dev/mem at the correct address, I was able to change the Requester mode from Release-When-Done to Release-On-Request, but this has no influence on all of the above... (maybe this can be added to vme_tsi148 ?) I know these are a lot of (rather difficult) questions, but I'm very grateful to everybody that can help me out here... Kind regards, Steven De Roo _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
