On Tue, Aug 30, 2011 at 9:31 AM, Jonathan Cameron <jic23@xxxxxxxxx> wrote: > On 08/30/11 17:14, Grant Grundler wrote: >> On Tue, Aug 30, 2011 at 3:05 AM, Jonathan Cameron <jic23@xxxxxxxxx> wrote: >> .... >>>> I did change one basic behavior that I think was also broken: cache >>>> the value regardless of if the transaction completed successfully or >>>> not. >>> Don't do that. That means userspace will get an invalid value if it reads >>> in the meantime. If you have an error on a hardware bus - tell userspace about >>> it and don't 'guess' what is in the register. >> >> Ah ok - I didn't know this was part of the path to/from user space. > It plausibly isn't, I didn't go over it closely enough to be sure. Ok. After thinking about it more, it might not matter anyway. > Even if not, it's a nasty complexity that will bite someone at some stage. > Mostly comes down to code doing what other code does in the same situation > rather than trying to be clever. The whole error handling path one ugly bit of "nasty complexity". If a register write fails, the register contents will be unknown. We will have no idea what value is in the register or if the device is even still alive. The cached value can't represent "known HW state" at that point and should probably be marked "invalid" until either the value is succesfully read or the next write succeeds. BTW, I didn't consider this particularly clever - just the code was easier to read. :) I don't think it really matters how the "cached value" is handled as long as the error gets returned. I'm just going to implement whatever you prefer. :) cheers, grant _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
