On Fri, Jan 27, 2023 at 11:10:25AM +0100, Uwe Kleine-König wrote: > Hello, > > On Wed, Nov 23, 2022 at 09:09:12AM +0100, Uwe Kleine-König wrote: > > On Wed, Nov 23, 2022 at 07:36:52AM +0100, Jiri Slaby wrote: > > > On 21. 11. 22, 8:07, Uwe Kleine-König wrote: > > > > On Mon, Nov 21, 2022 at 07:03:41AM +0100, Jiri Slaby wrote: > > > > > On 18. 11. 22, 23:45, Uwe Kleine-König wrote: > > > > > > From: Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx> > > > > > > > > > > > > .probe_new() doesn't get the i2c_device_id * parameter, so determine > > > > > > that explicitly in the probe function. > > > > > > > > > > I wonder why -- is this a new approach to probe functions? Or is only i2c > > > > > affected? And why? Could you point to the commit introducing and describing > > > > > the change in the i2c core? > > > > > > > > I didn't sent the cover letter to all recipents of the individual > > > > patches, so flow of information is a bit rough. Sorry about that. > > > > > > > > You can find it at > > > > https://lore.kernel.org/lkml/20221118224540.619276-1-uwe@xxxxxxxxxxxxxxxxx/, > > > > it should answer your question. > > > > > > Yes, I looked up that beforehand, but was no more clever after reading it. > > > > > > > The short version is: The i2c framework does a more or less expensive > > > > lookup for each call to .probe() to provide the id parameter. A relevant > > > > part of the drivers however doesn't use this parameter, so the idea is > > > > to let the drivers who actually need it, determine it themselves. > > > > > > > > Statistics for the current state of this series in my tree: > > > > Among the 602 converted drivers, 404 don't make use of the parameter. > > > > > > So doesn't it make sense to provide both probe with no id and "probe_id" > > > then? 200 is quite a few (a third to be precise). > > > > Having the probe callback with the id parameter is only temporary. As > > soon as all drivers are converted, the variant with the id parameter > > will go away. > > > > > BTW is this a performance issue? I.e. does it slow down the boot? > > > > I don't know the start motivation for Lee (who triggered the conversion > > in b8a1a4cd5a98 ("i2c: Provide a temporary .probe_new() call-back > > type")). > > Looking at the git history, he created 1e98dcd77970 ("mfd: 88pm860x: > > Move over to new I2C device .probe() call") converting a driver that > > doesn't benefit immensely. The lookup is more expensive for drivers with > > big .id_table, the converted driver has only one entry. > > > > I think in the end is a mixture between: > > > > - A big part of the drivers doesn't benefit from the lookup. > > - For most other busses the probe function only gets a device parameter > > and no id (spi, platform, i3c). There are counter examples though: > > amba, usb. Didn't check further. > > The discussion somehow ended here without a real result. > > As of today's next master there are only 9 drivers left using .probe(). > So I'd like to stop this discussion and ask to apply the conversion for > the sc16is7xx driver to be able to complete the conversion. > > My plan is to drop the .probe callback as it is today after the next > merge window. So I ask the serial maintainers to either take the patch > under discussion for the next merge window or accept that the conversion > is done together with the patch that drops .probe() that probably will > go in via the i2c tree. I don't see the patch anymore, so I have no objection for it going through the i2c tree. thanks, greg k-h
