From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933841AbbBIXqY (ORCPT ); Mon, 9 Feb 2015 18:46:24 -0500 Received: from mail-ig0-f175.google.com ([209.85.213.175]:41461 "EHLO mail-ig0-f175.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933171AbbBIXqV (ORCPT ); Mon, 9 Feb 2015 18:46:21 -0500 MIME-Version: 1.0 In-Reply-To: References: <100e01d04493$11a2fd40$34e8f7c0$%opasiak@samsung.com> Date: Tue, 10 Feb 2015 01:46:20 +0200 Message-ID: Subject: Re: [PATCH 1/2] usb: gadget: udc-core: independent registration of gadgets and gadget drivers From: Ruslan Bilovol To: Alan Stern Cc: Krzysztof Opasiak , Peter Chen , "linux-usb@vger.kernel.org" , "linux-kernel@vger.kernel.org" , "Balbi, Felipe" , "gregkh@linuxfoundation.org" , Andrzej Pietrasiewicz Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi guys, On Mon, Feb 9, 2015 at 10:00 PM, Alan Stern wrote: > On Mon, 9 Feb 2015, Krzysztof Opasiak wrote: > >> > Why bother matching by name? Why not simply take the first >> > available >> > UDC? >> >> Because you may have more than one udc. This would allow to pick one by >> name just like using configfs interface. > > Clearly it would be more flexible to allow the user to do the matching, > the way configfs does (sysfs too). > >> > > Main feature of my path is not only deferred binding of gadget >> > driver, >> > > but also possibility to register/unregister udc at any time. >> > > This is useful for user who can load, for example, udc module >> > > if needed and unload it safely, not touching gadget driver. >> > >> > We can already do that with the existing code. There's no need for >> > a >> > patch. >> > >> > Also, it's not clear that the existing gadget drivers will work >> > properly if they are unbound from one UDC and then bound again to >> > another one. They were not written with that sort of thing in >> > mind. >> > >> >> What you have described is one of basics configfs features. >> You should be able to bind and unbind your gadget whenever you want >> and it should work properly after doing: >> >> ## create gadget >> $ echo "udc.0" > UDC >> $ echo "" > UDC >> $ echo "udc.1" > UDC >> >> Function shouldn't care which udc it has been bound previously. >> Only current one is important and on each unbind each function >> should cleanup its state and prepare to be bound to another udc. >> Configfs interface doesn't prohibit this and I haven't seen any >> info about such restriction. Thank you Krzysztof for this explanation that makes things more clear for reviewers. > > It's not prohibited, but it also was never required. Therefore it may > not be implemented in all gadget drivers. > >> If some function is not working in >> such situation there is a bug in that function and it should be fixed. > > That's fine. I wasn't pointing out a fundamental limitation, just a > fact that will have to be taken into account. I also don't see any restrictions to bind/unbind gadget driver few times and in case of such bug we just can fix it. I didn't see any issue with gadget drivers that I used for verification, however, to be honest, I didn't test it with all possible ones. Anyway, it should work in legacy way (one gadget driver is bound to one uds) so current behavior at least is not broken. > > Anyway, instead of going through all this, why not do what I suggested > earlier (see http://marc.info/?l=linux-usb&m=139888691230119&w=2) and > create a "gadget" bus type? That would give userspace explicit control > over which gadget driver was bound to which UDC. Exactly, my patch transforms udc-core to something like tiny bus with very basic functionality. But in mentioned mailthread I see good ideas that is possible to implement with small overhead. > > Or maybe that's not a very good fit with the existing code, since most > gadget drivers assume they will be bound to only one UDC at a time. So > instead, why not create a sysfs interface that allows userspace to > control which gadget drivers are bound to which UDCs? > > As I recall, the original problem people were complaining about was > deferred probing. They didn't need fancy matching; all they wanted was > the ability to bind a gadget driver to a UDC some time after the gadget > driver was loaded and initialized. Something simple like Robert > Baldyga's patch is enough to do that. This simplicity is also a limitation. As I mentioned before, sometimes it is needed to be able to bind/unbind gadget driver multiple times to different UDCs. A real case I faced recently is SoC with HighSpeed-only and SuperSpeed-only UDCs. SuperSpeed-only UDC can't work on USB 2.0 speeds and vice versa. The system has USB3.0 USB connector with soldered HS lines from mentioned HS-only and SS lines from SS-only UDCs. Each UDC has it's own device driver, so if we want to use both of them with one gadget driver, we must be able to bind/unbind it multiple times to different UDCs. Another one usecase is users who can unload udc drivers without carrying about gadget drivers. Third usecase is, for example, developers who can switch between dummy_hcd and real UDC hardware without unloading gadget driver. I'll work on improved version and will send new patch soon... Best regards, Ruslan > > Alan Stern >