Hi, Alec, This is a serious bug. Even though the workaround seems to work (still pending Glenn's testing), having to worry about when to use ?wordwrite vs ?write is very unfriendly for users/programmers. It would be good to fix it sooner rather than later, IMHO.
Dave On Dec 10, 2012, at 11:12 AM, Alec Rust wrote: > Dave if the wordwrite workaround works lets stick to that for now. The > workaround Marc compiled is not really good for release. We'll work on a > proper release but for now use wordwrite if thats ok? > > Regards > Alec > > On Mon, Dec 10, 2012 at 8:42 PM, David MacMahon <dav...@astro.berkeley.edu> > wrote: > Hi, Marc, > > I can confirm that the wordwrite workaround works. Hopefully the fix for > byte enables (either to make tcpborphserver3 not use them or to make the > gateware support them) will not be too hard. > > Thanks, > Dave > > On Dec 10, 2012, at 7:06 AM, Marc Welz wrote: > > > Hello > > > >> We have picked up on transient on the register write with the latest memory > >> mapped TCPBORPH. It seems that sometimes, some of the bits goes high for a > >> short while and then settles at the required register value. > >> > >> We need to figure out where the issue is (gateware or software).> > > > > So we think we have found the problem - ?write operations in > > tcpborphserver3 rely on byte enables, which are not supported by the > > gateware. I'll rewrite ?write to use multiples of 4... but in the mean > > time you > > could try using ?wordwrite, which operates on words, and so doesn't use byte > > enables. That workardound should be simpler than installing older > > kernel/tcpborphserver combinations. > > > > regards > > > > marc > > > > >