On 12/06/2022 21:23:47+0200, Stephen Kitt wrote: > Hi Alexandre, > > Le 11/06/2022 16:48, Alexandre Belloni a écrit : > > On 10/06/2022 18:23:43+0200, Stephen Kitt wrote: > > > All these drivers have an i2c probe function which doesn't use the > > > "struct i2c_device_id *id" parameter, so they can trivially be > > > converted to the "probe_new" style of probe with a single argument. > > > > > > > I think you should explain why you want to do that as the trend is to do > > the exact opposite to allow support for those RTCS on x86 systems. > > Indeed, I should have given more context. The idea is to continue the > transition started with https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b8a1a4cd5a98a2adf8dfd6902cd98e57d910ee12 > (in 2016...). I²C drivers using probe() involve a call to i2c_match_id: in > drivers/i2c/i2c-core-base.c, > > /* > * When there are no more users of probe(), > * rename probe_new to probe. > */ > if (driver->probe_new) > status = driver->probe_new(client); > else if (driver->probe) > status = driver->probe(client, > i2c_match_id(driver->id_table, > client)); > else > status = -EINVAL; > > Many drivers don't actually need the second parameter, so instead of having > probe() with both parameters, the goal is to switch to a probe function with > only "struct i2c_client *". Probe functions that *do* need the "struct > i2c_device_id" can call i2c_match_id themselves (as is done currently with > of_match_id). > > I discussed this briefly with Wolfram beginning of June at Kernel Recipes, > and as I understood it the plan was still to continue with this transition > (in fact, the plan was for *me* to continue with this transition). > > Do you have an example of a change requiring id to support RTCs on x86 > systems? I checked and your patch is actually fine, the issue I referred to is: 1c1b3098ae1e ("rtc: pcf85063: add i2c_device_id name matching support") -- Alexandre Belloni, co-owner and COO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
