On Mon, 25 Oct 2004 22:39:30 -0700, Peter Ryser <peter.ryser at xilinx.com> wrote:
jcm>> btw Peter, I've been moaning about the weird interrupt behaviour of jcm>> the SysACE on e.g. the Memec board for some time (see archives for jcm>> various other people's experiences too) - and I've spoken to several jcm>> folks at Xilinx without getting very far. If there's a fundamental jcm>> problem then it probably makes sense for us to help get it fixed. > Hmm, I'm not aware of any interrupt problems with System ACE CF. It issues an extra interrupt after each sector write on our hardware, in what seems to be a reasonably reproducible but not entirely predictable fashion. This is after the interrupt we expect to get on sector completion - it causes the Monta driver to go belly up because it has no handling to check for interrupt on QUEUE_EMPTY or similar error conditions. I have every confidence that it works on the ML300, just not on other boards, including both the Memec board and our in house prototypes. Since it's not a major issue right now to have good performance from the SysACE (just a reliable driver), I've created the patch I posted previously to use polled IO instead. > There is now a MVL preview kit for the Memec board. You might want to > try it out and see whether the problems you observe persist. Is this the one which doesn't actually use the SystemACE for a filesystem? > Looking at the postings back in April I can only say that I never saw > any similar problems. Others have done so. I'll check with the hardware guys and see if I can't send you a simple hardware case that falls over at some point - I'm slightly constrained to release hardware designs and so on until after this thing is ready to ship. > The implementation of System ACE CF functionality in U-Boot is written > for polled operation. In that case Stephen Williams wrote the actual > driver and we decided not to publish our own driver but reuse his. So it probably doesn't use the Xilinx HAL, that can only be a good thing :-) Jon.
