On Friday 12 June 2009 14:59:03 Guennadi Liakhovetski wrote: > On Fri, 12 Jun 2009, Hans Verkuil wrote: > > > > 1. it is very unusual that the board designer has to mandate what signal > > > polarity has to be used - only when there's additional logic between the > > > capture device and the host. So, we shouldn't overload all boards with > > > this information. Board-code authors will be grateful to us! > > > > I talked to my colleague who actually designs boards like that about what > > he would prefer. His opinion is that he wants to set this himself, rather > > than leave it as the result of a software negotiation. It simplifies > > verification and debugging the hardware, and in addition there may be > > cases where subtle timing differences between e.g. sampling on a falling > > edge vs rising edge can actually become an important factor, particularly > > on high frequencies. > > I'd say this is different. You're talking about cases where you _want_ to > be able to configure it explicitly, I am saying you do not have to _force_ > all to do this. Now, this selection only makes sense if both are > configurable, right? In this case, e.g., pxa270 driver does support > platform-specified preference. So, if both the host and the client can > configure either polarity in the software you _can_ still specify the > preferred one in platform data and it will be used. > > I think, the ability to specify inverters and the preferred polarity > should cover all possible cases. In my opinion you should always want to set this explicitly. This is not something you want to leave to chance. Say you autoconfigure this. Now someone either changes the autoconf algorithm, or a previously undocumented register was discovered for the i2c device and it can suddenly configure the polarity of some signal that was previously thought to be fixed, or something else happens causing a different polarity to be negotiated. Suddenly the board doesn't work because it was never verified or tested with that different polarity. Or worse: it glitches only 0.001% of the time. That's going to be a nasty bug to find. You generally verify your board with specific bus settings, and that's what should also be configured explicitly. Sure, it is nice not to have to think about this. The problem is that I believe that you *have* to think about it. The longer I think about this, the more convinced I am that relying on autoconfiguration is a bad design. Regards, Hans -- Hans Verkuil - video4linux developer - sponsored by TANDBERG Telecom -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
