Return-Path: Content-Type: text/plain; charset=windows-1252 Mime-Version: 1.0 (Mac OS X Mail 8.2 \(2098\)) Subject: Re: [PATCH] bluetoothd: add option to automatically power on the first adapter found From: Marcel Holtmann In-Reply-To: <552CD0F2.7080605@ahsoftware.de> Date: Tue, 14 Apr 2015 06:50:11 -0700 Cc: linux-bluetooth@vger.kernel.org Message-Id: <35E686A5-9C85-421D-BB8B-DF14F7D1A276@holtmann.org> References: <1428685062-2911-1-git-send-email-holler@ahsoftware.de> <97E0530A-49BA-4776-BF4E-DF500F28116E@holtmann.org> <55280525.7040902@ahsoftware.de> <55289DD3.1080509@ahsoftware.de> <5528AC01.80804@ahsoftware.de> <55295063.2010106@ahsoftware.de> <72FEAB70-4102-4048-94E0-4AA0B4D67217@holtmann.org> <552A397D.50709@ahsoftware.de> <273C2B2C-FE29-4E28-8B50-7A9BE82C098D@holtmann.org> <552B87EE.5040800@ahsoftware.de> <552C1413.1000504@ahsoftware.de> <552CD0F2.7080605@ahsoftware.de> To: Alexander Holler Sender: linux-bluetooth-owner@vger.kernel.org List-ID: Hi Alexander, >>>>>>> As said, this (currently) needs that dbus is ready and bluetoothctl can communicate with bluetoothd. And on a reasonable fast machine (4 core x86_64 2ghz+) this needs more than 4 seconds after startup here (invoked through /etc/rc.d/rc.local on f21). >>>>>> >>>>>> Szymon send an example on how to add wait-for-adapter in bluetoothctl. Together with that and a proper systemd unit, I bet you that this does not take 4 seconds. >>>>>> >>>>>> If you insist on doing old fashion rc.local stuff, then I can not help you. You can look into systemd target definitions, but I am pretty sure it ends up being started after something else happened. >>>>> >>>>> At least that old fashioned stuff worked. I even still have bt-alsa-devices on systems where I stick to bluez 4.0. >>>>> >>>>> And it would make me extremly wonder, if it would be any faster when using a systemd unit. Remember, I did in rc.local: >>>>> >>>>> sleep 4 >>>>> echo "power on" | /bin/bluetoothctl >>>>> >>>>> And if that fails, then how could a systemd unit which calls bluetoothctl be faster? There is absolutely nothing in those 2 lines which slows down the necessary prerequisits for bluetoothctl. >>>> >>>> except the sleep 4 that you put in. So you are going on for rant this takes 4 seconds and the reason it takes 4 seconds since you put a sleep 4 in there. Szymon gave you a simple solution for removing the sleep and we are still discussing this. >>> >>> I've run into the problem, because those 4 second weren't enough after an update of bluez. I then had to use 6 seconds, otherwise that construct failed. And because that ugly sleep was just an extremly ugly workaround for the missing easy-to-use knob to reliable turn on bt on Linux with bluez 5.x, I've now written that patch. >>> >>> But it's senseless trying to discuss further. If you think people have to write udev-rules or systemd units in order to use bluetooth on Linux, there is just nothing I could add to such a discussion. >> >> you did read my responses here? I mean there are so many possible options. Szymon provided an addition for bluetoothctl command where you could just have replaced sleep 4 with the new command. And you could have it working perfectly all the time. >> >>> And you might want to decide if you want a configuration less bluez (something I think I've read several times and which is likely the reason that make install doesn't install at least an example for configuration), or if you still want to support some configuration as mentioned in the comment in regard to a [policy] section. >> >> Upstream is not responsible for personal or distribution specific customization. We provide configuration via main.conf and I am repeating myself now for the 3rd time here, I am fine with a policy setting for auto-power on found controllers. The distribution has to do the work to put a proper main.conf in place if they choose to. It is really up to them. This is not something upstream should do in the first place. >> >> bluetoothd has to work properly with default values if the configuration is present or not. Failing because a configuration file is not present and you can not run a proper sane default is plain stupid. That is not how we write software. We are doing proper defaults. > > Sorry, I'm sick of having to teach others about the usability problems of bluez. > > E.g. since the configuration knob for hid2hci disappeared (I think that > happened with bluez 4.x), it's still done wrong by many people and > distributions. E.g. on Ubuntu the udev-rule for hid2hci is still wrong because nobody explained them that the udev-rule is not the problem but the non-existing knob to turn off hid2hci if it is installed. Not to mention that it should be turned off by default (until the user turned it on, hoping he is aware of the implications). > > Just the same problem which happened on Fedora and Gentoo too (until I've explained the problem to the maintainers and now hid2hci isnt installed by default there). It's just to hard to understand (and to explain) that installing a command-line utility (by default) might render a system unusable. And users aren't the problem here, if even maintainers get it wrong. > > And almost anyone runs into the problem that bt-adapters are now powered > off by default and people have not idea how to change that policy or even how to manually turn on the adapter. And that now happens since 5.x was thrown at users, which is already 30 releases old. > >> >> Please respect the upstream decision not to install main.conf in /etc/bluetooth/ and leave this up to the distribution and packagers. In case you wonder this is important for stateless installations and systems that support factory reset. >> >> I am happy to take a patch that implements the auto-power on properly and without race conditions and makes it configurable based on the main.conf policy section. And I said this many times now. The amount of emails exchanged on this topic where I agreed to take a patch if it would be done race free and in sync with how upstream operates is amazing. >> > > Sorry, but I've no idea about which races you are talking. Of course, I haven't fixed the already existing race(s) in bluetoothd with the missing lock for the adapter list (at least I haven't seen or found any lock for that list, otherwise I would have used it). But that still might be my problem, as I'm not very familiar with bluetoothd and especially it's locking strategy (if there is any, which I hope). it is a single threaded process. It does not need a lock for the adapter list. Your race power on vs keys and known devices programmed into the kernel. It needs to be done in the right order. > I also have noticed that bluetoothd and bluetoothctl seem to have a different idea about the default adapter (at least that is_default variable in bluetoothd doesn't seem to be in sync with what bluetoothctl displayed as default adapter when I've tried the stuff with two adapters). Or I just have missed how that is_default was changed during my few tests. Otherwise I would have named that option I've added different ("default adapter" instead of "first adapter"). That sounds like a bug to me that should be fixed. I wonder why bluetoothd actually has to carry the concept of the default adapter at all anymore. Regards Marcel