Return-Path: Subject: RE: [PATCH 1/2] drivers:bluetooth: TI_ST bluetooth driver From: Marcel Holtmann To: "Savoy, Pavan" Cc: "linux-bluetooth@vger.kernel.org" , "johan.hedberg@gmail.com" , "greg@kroah.com" , "linux-kernel@vger.kernel.org" In-Reply-To: <19F8576C6E063C45BE387C64729E739404AA21D1BD@dbde02.ent.ti.com> References: <1286404493-23816-1-git-send-email-pavan-savoy@ti.com> <1286404493-23816-2-git-send-email-pavan-savoy@ti.com> <1286445948.6145.70.camel@aeonflux> <19F8576C6E063C45BE387C64729E739404AA21D19E@dbde02.ent.ti.com> <1286464659.6145.144.camel@aeonflux> <19F8576C6E063C45BE387C64729E739404AA21D1B4@dbde02.ent.ti.com> <1286465843.6145.157.camel@aeonflux> <19F8576C6E063C45BE387C64729E739404AA21D1BD@dbde02.ent.ti.com> Content-Type: text/plain; charset="UTF-8" Date: Thu, 07 Oct 2010 17:49:36 +0200 Message-ID: <1286466576.6145.161.camel@aeonflux> Mime-Version: 1.0 Sender: linux-bluetooth-owner@vger.kernel.org List-ID: Hi Pavan, > > > > > > Registering the Bluetooth HCI driver in module_init/module_exit is not > > > > > > acceptable. Turn your shared transport into a proper bus. > > > > > > > > > > Yes, you did comment on it before, I remember, I did prototype the > > driver as > > > > > a bus driver, However I didn't find any advantages by converting it to a > > bus > > > > > driver. > > > > > As in, currently the shared transport driver is a line discipline driver > > > > because > > > > > it is the only way it can communicate over TTY without being tightly > > coupled > > > > with the UART driver. > > > > > > > > > > > We want to be able to have generic kernels where this module is > > enabled, > > > > > > but no Shared Transport is available. > > > > > > > > > > Oh if this is the reason I cannot have hci_register/_unregister in > > > > module_init/_exit, Can I do this module "depends" on TI_ST, Then it would > > not > > > > > even be visible to build if TI_ST is not selected. > > > > > > > > this is not helping either. Then TI_ST can not be selected and so you > > > > still end up with some weird platform specific kernels. We don't want > > > > that. We want generic kernels that can detect the hardware they are > > > > running on. > > > > > > > > As I said, I will not accept this driver if it registers HCI device in > > > > module_init. No other driver is doing this and it is in general a really > > > > really really bad idea. > > > > > > > > > > Ok, now I am beginning to get what you say, Let me check, may be what > > > I can do is, have something like a st_prepare() function called in the > > > module_init, and a _probe function of the bluetooth driver will be called, > > > _ONLY_ if the _probe of my platform driver has been called.. > > > Do you think this would be a good idea? > > > > > > Note: the TI_ST driver is also a platform device driver, so that TI_ST's > > > Probe is not called, if a arch/xx/board-xx doesn't add it. > > > > that that should be your bus right there. > > I understand the perspective, but "bus" is not device-driver type of model right? I mean I need a device which will be added in some platform specific > board file, and the driver in my driver core file. > > > Let me repeat this. If you register the HCI device in module_init then > > it will be registered on all platform this module is selected. Even if > > the kernel runs on x86. And that is not acceptable. Registering devices > > in module_init is a bad idea no matter what. That is why all other > > drivers just register a driver here and not a device. > > I did initially think about making each of the protocol drivers a > platform devices as well. > As in Bluetooth/FM/GPS TI_ST driver would also be a platform device and its _probe doing the HCI/v4L2/character device registration. > > So which one do you think makes more sense here? > 1. Do I EXPORT a new symbol called st_prepare? And allow hci registration there? > > 2. Or make Bluetooth device a platform device and this driver a platform driver > and add this Bluetooth device only when I add TI_ST platform device? then make them a platform device. Since you do need a proper parent for these devices anyway. Otherwise a lot of logic within sysfs will fail. Regards Marcel