On 05/17/2010 09:44 AM, Marc Kleine-Budde wrote: > Wolfgang Grandegger wrote: >> Hi Marc, >> >> On 05/17/2010 12:46 AM, Marc Kleine-Budde wrote: >>> Hey Wolfgang, >>> >>> Marc Kleine-Budde wrote: >>>>>> The patch series can be viewed as as git tree under: >>>>>> http://git.pengutronix.de/?p=mkl/berlios-can-utils.git;a=shortlog;h=refs/heads/pu >>>>> I would prefer to make the struct can_bittiming_const compatible with >>>>> the one used by the kernel: >>>>> >>>>> http://lxr.linux.no/#linux+v2.6.33/drivers/net/can/sja1000/sja1000.c#L75 >>> The common members have now the same names, but there's still the printf_btr >>> and I add a ref_clock in a later patch..... >>> >>>>> I forgot to mention that, sorry. Should not be a big deal, though. >>>> There're more patches to come. And one of this does exactly this. I just >>>> have to write a decent commit message for the last patch, then I'll send >>>> it. >>> Here they are...they are based on the other series I posted earlier. >>> >>> It's available as a git tree at: >>> http://git.pengutronix.de/?p=mkl/berlios-can-utils.git;a=shortlog;h=refs/heads/rfc/better-bitrate-calculation >> >> Looks good at a first glance and they could go in immediately. We can >> add fixes later on, if necessary. One requirement of the algorithm is >> that it *must* reproduce the CIA recommanded values for the SJA1000 for >> the standard bit-rates. We should not expect that it works for any > > Have a look at patch 5. In it's patch description is a diff of output of > the original and the improved version. As you can see no changes for the > sja.
Ah, right. > The original algorithm find optimal values for the sja (no bitrate error > and no sample point error). My patch just optimizes the sample point in > cases where you cannot avoid bitrate errors. Good. At some point we may want to update the algorithm in the kernel as well. >> bitrate, especially with exotic clock frequencies. Those require manual >> fiddling with the bit-timing parameters, e.g. SJW, which is non-trivial. > > I tend to call them "real live" clock values :) Yes, that's the real fun. Wolfgang. _______________________________________________ Socketcan-core mailing list [email protected] https://lists.berlios.de/mailman/listinfo/socketcan-core
