There is really no way to abort a started cycle in Wishbone. Once a masters starts a cycle it has to complete it. Slaves can, on the other hand acknowledge, retry, or error out.
But a master has to figure this out before hand. As there is no protocol for aborting a started cycle, slaves can not expect this behaviour. So your suggested 'abortion' may cause havoc on some slaves. IPersonally I think it should be fixed in the master (do not start a cycle until you need one). However if really required, then we may need to think about how this would work. We'd still need to support slaves that don't support cycle abortion. Richard On Oct 9, 2012, at 21:04 , Franck Jullien wrote: > 2012/10/9 Stefan Wallentowitz <[email protected]>: >> On Mon, 2012-10-08 at 22:46 +0200, Franck Jullien wrote: >>> Hi, >>> >>> As I was investigating on an IC problem on my Altera board, I found a >>> wrong behavior (that's what I think) on the wb_cyc_o output from the >>> or1200_wb_biu. >>> At some point, when the biu_cyc_i to the biu is desaserted, the biu >>> fsm is still waiting for an ack from the wishbone and the biu_cyc_i >>> desasertion is not followed to the wb side. >>> >>> Attached patch fix this. >> >> Hi, >> >> just for curiosity as we recently faced the same "issue". Doesn't >> de-asserting the cycle and strobe without waiting for the ack break the >> protocol? >> We had the case that a second level cache was connected at the BIU and >> because of internal changes to the L1 cache the BIU incorrectly aborted >> a transaction. As the fetching at L2 took kind of long the corresponding >> ack was then incorrectly attributed to the next transaction of the BIU. >> From what I see, the Bus Master can not easily abort a bus transaction >> or am I wrong? >> >> Bye, >> Stefan >> >> -- >> Stefan Wallentowitz _/ _/_/_/ _/_/_/ >> Institute for Integrated Systems (LIS) _/ _/ _/ >> Technische Universität München _/ _/ _/_/ >> D-80290 München _/ _/ _/ >> fon +49 89 289 22963 _/_/_/_/ _/_/_/ _/_/_/ >> >> [email protected] http://www.lis.ei.tum.de > > I would say that too. From what I known, de-asserting the cycle and > strobe is the only way for a master to abort a bus transaction (but > I'm not a wb expert). > So I guess it's ok.... > > Franck. > _______________________________________________ > Openrisc mailing list > [email protected] > http://lists.opencores.org/listinfo/openrisc _______________________________________________ OpenRISC mailing list [email protected] http://lists.openrisc.net/listinfo/openrisc
