On Wed, 29 Nov 2017 09:58:44 +0800, Phil Reid wrote: > G'day Jean, > > Thanks for looking at this. > > On 29/11/2017 06:08, Jean Delvare wrote: > > Hi Phil, > > > > On Tue, 28 Nov 2017 11:38:28 +0800, Phil Reid wrote: > >> i2c-smbus now only contains code related to the smbus_alert driver. > >> Other smbus protocols have been moved from this into the i2c-core-smbus. > >> Change the Kconfig variable name to reflect this. > > > > Not really. i2c-core-smbus.c contains essentially SMBus transaction > > helpers, which have never been in i2c-smbus.c. They aren't really SMBus > > protocols, more like a subset of I2C transactions suitable for general > > purpose. The only really SMBus protocol specific portion is relative to > > SMBus Alert, and that's only a small part of it. The other supported > > SMBus-specific protocol, Host Notify, is in i2c-core-base.c. > > > > My original intent was to have all the SMBus protocols specific code in > > one file, controlled by one Kconfig option, which could be built as a > > separate module. I wasn't certain it would be viable, it was a bet > > which I lost, as it turns out there are too many dependencies. > > > > If the code can no longer build as a separate module, there is no > > benefit in having it in a separate file. Let's just merge it into > > i2c-code-smbus.c, where the rest of the SMBus alert code already is. > > That would make things more simple. > I had to put of_i2c_setup_smbus_alert in there to deal with dependencies. > This was seen as a little messy at the time. It was accepted provided > that I clean things up somehow. > > The question is how. This series is just one proposal for that to stir some > discussion. Your approach is certainly the simpler one. OK, can you please give it a try and wee how it compares with your own approach? > > I also don't think renaming this configuration option and moving code > > to a separate file (as your patch 3/3 does) makes sense. Having a > > separate option for each SMBus-specific protocol seems overkill to me, > > having one for all of them was and still is more reasonable. I wonder > > why the SMBus Notify code does NOT depend on CONFIG_I2C_SMBUS, and why > > it is in i2c-core-base.c instead of i2c-core-smbus.c. Wolfram? > > > > And more generally, renaming a Kconfig option has a non-trivial cost > > for distribution kernel maintainers, so it shall not be done lightly. > > So you need a clear and strong rationale. > > My only rational is that it didn't encompass all the SMBUS specific protocols. > It currently seems a little ambiguous to me. Especially with regard the current > Kconfig help description is not accurate with regard the Host Notify protocol. I agree, the I2C_SMBUS Kconfig help text no longer reflects the reality since over a year now. But I think the text is how things should be and the code is what should be fixed (that is, all the SMBus Host Notify code should be left out when I2C_SMBUS isn't set, and it should live in i2c-core-smbus.c.) But I did not look at the details, maybe this isn't possible for some reason. Maybe you can look into it after you are done with moving the SMBus Alert code into the i2c core? If you don't, I will. Thanks, -- Jean Delvare SUSE L3 Support
