On Fri, Dec 9, 2022 at 3:26 AM Linus Walleij <linus.walleij@xxxxxxxxxx> wrote: > > On Fri, Dec 9, 2022 at 5:39 AM Chester Lin <clin@xxxxxxxx> wrote: > > > > Hi Linus and Fabio, > > > > Thanks for your time to review this patch! > > > > On Thu, Dec 08, 2022 at 10:37:36PM +0100, Linus Walleij wrote: > > > On Thu, Dec 8, 2022 at 12:04 AM Fabio Estevam <festevam@xxxxxxxxx> wrote: > > > > > > > In other imx8m pinctrl drivers we pass: > > > (...) > > > > > +module_platform_driver(s32g_pinctrl_driver); > > > > > > > > And we also register it in arch_initcall() level. > > > > > > Do you really need that though? This driver certainly does not. > > > > > > I was under the impression that recent changes to the probe-order > > > logic has made most explicit arch_ etc initcall orderings surplus. > > > > > > > Could bool/tristate options in the Kconfig be the key point? > > > > Based on current design I prefer to build the s32g2 pinctrl driver as built-in > > rather than a loadable module. IIUC, when the driver is not built as module > > then the initcall ordering should still matter. > > It is true that if you compile something into a module then all initicalls > are the same: they are called when the module is loaded. > > But the remaining initcalls used to be assigned to core, arch, subsystem > etc in order for resources (such as clocks, regulators or pins) to be > available before the drivers that need them get probed. > > However there was first deferred probe to partially solve the problem > and recently a large and refined series that use the dependencies in > the device tree to resolve probe order. > > Saravana Kannan has been working tirelessly at this, issueing > git log --oneline --author="Saravana Kannan" > you will see the scope of this work. > Thanks Linus. For a system using DT, fw_devlink's goal is to make module load ordering or driver registration ordering irrelevant to proper functioning of the kernel/drivers. It should automatically figure out the dependencies and have the devices probe in the right order. It's already true for at least 80% of the cases for a system using DT. There are some known issues I'm either working on or have on my To do list. So, if you see a case where fw_devlink is not handling it correctly, please let me know. -Saravana
