Return-Path: Content-Type: text/plain; charset=utf-8 Mime-Version: 1.0 (Mac OS X Mail 10.3 \(3273\)) Subject: Re: btattach and 3wire doesn't work From: Marcel Holtmann In-Reply-To: Date: Wed, 20 Sep 2017 18:54:12 +0200 Cc: Bluez mailing list Message-Id: <292E97AC-D1B0-440E-BA2D-6C50395C0675@holtmann.org> References: <5467450F-B5AA-4BD5-874F-2ABC7F05D821@holtmann.org> <5D4329D8-61EB-420B-8985-E4204D272E65@holtmann.org> To: Emil Lenngren Sender: linux-bluetooth-owner@vger.kernel.org List-ID: Hi Emil, >>>>> I'm trying to use the new btattach utility to attach a controller >>>>> which uses the 3-wire UART protocol. The older hciattach works great >>>>> but btattach doesn't. >>>>> This is the command I use including the output: >>>>> >>>>> # btattach -B /dev/ttyS1 -P 3wire -S 500000 >>>>> Attaching Primary controller to /dev/ttyS1 >>>>> Switched line discipline from 0 to 15 >>>>> Failed to get device id: Protocol driver not attached >>>>> No controller attached >>>>> >>>>> By reading the source at >>>>> https://git.kernel.org/pub/scm/bluetooth/bluez.git/tree/tools/btattach.c >>>>> I see that it differs a bit from hciattach. Notably, it calls the >>>>> HCIUARTGETDEVICE ioctl function directly after the HCIUARTSETPROTO >>>>> ioctl call. The hciattach utility does not use HCIUARTGETDEVICE at >>>>> all. >>>>> >>>>> The cause of the problem is that when using 3wire, the hci does not >>>>> get registered immediately (which btattach thinks) but instead after >>>>> the handshake is complete. >>>>> >>>>> If I run btattach in gdb and break before the HCIUARTGETDEVICE ioctl >>>>> call, wait a few seconds and then resume, it works as expected. What >>>>> is the purpose of HCIUARTGETDEVICE? I see it's only being used when >>>>> the Raw option is used (which is by the way not mentioned in the Usage >>>>> docs). >>>>> >>>>> Another issue I have is that hciattach/btattach only support a few >>>>> different baud rates. I want to use 750000 bit/s. Is there a purpose >>>>> of only allowing some? Currently I use hciattach with a dummy baud >>>>> rate followed by >>>>> https://gist.github.com/lategoodbye/f2d76134aa6c404cd92c and that >>>>> works. >>>> >>>> which hardware is this? I always wanted to fix some of the 3-wire support to make sure it fully works with btattach and everything is done inside the kernel correctly. Including proper support for hdev->setup for vendor firmware loading etc. >>> >>> >>> The hardware is an nRF52 device running a generic 3wire >>> implementation. So no custom setup needed here. >>> The 3wire support otherwise in the kernel seems to work very nice as >>> long as I've been testing this. The only thing I miss myself is the >>> lack of the CRC check defined optionally by the 3wire specification. >> >> who is providing a 3-wire HCI controller firmware for the nRF52? I know with Zephyr and Mynewt you can turn them into a generic HCI controller, but we have not gotten around to add 3-wire support to Zephyr yet. > > I have written my own implementation according to the Core v5.0 > specification. I thought it was more nice to have a 3wire interface > rather than h4 since that adds better error checking, automatic resync > upon chip reset etc. any chance you want to open source that and share it. Or consider porting the 3-wire support into Zephyr? Anyway, have you tried —noflowctrl option with btattach. If I remember correctly then 3-wire is requiring that flow control is turned off. Regards Marcel