Message-ID: <20181016210245.GU22824@google.com>
Date:   Tue, 16 Oct 2018 14:02:45 -0700
From:   Matthias Kaehlcke <mka@...omium.org>
To:     Marcel Holtmann <marcel@...tmann.org>
Cc:     "Rafael J . Wysocki" <rafael@...nel.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Sinan Kaya <okaya@...eaurora.org>,
        Balakrishna Godavarthi <bgodavar@...eaurora.org>,
        Sakari Ailus <sakari.ailus@...ux.intel.com>,
        Marcin Wojtas <mw@...ihalf.com>,
        Andy Shevchenko Andy Shevchenko <andy.shevchenko@...il.com>,
        Johan Hedberg <johan.hedberg@...il.com>,
        LKML <linux-kernel@...r.kernel.org>,
        Bluez mailing list <linux-bluetooth@...r.kernel.org>,
        Loic Poulain <loic.poulain@...aro.org>,
        Brian Norris <briannorris@...omium.org>
Subject: Re: [PATCH v4 1/2] Bluetooth: Add device_get_bd_address()

Hi Marcel,

On Tue, Oct 16, 2018 at 08:52:07AM +0200, Marcel Holtmann wrote:
> Hi Matthias,
> 
> >>>>>> void bt_sock_reclassify_lock(struct sock *sk, int proto);
> >>>>>> 
> >>>>>> +int device_get_bd_address(struct device *dev, bdaddr_t *bd_addr);
> >>>>> 
> >>>>> Maybe change the API name to start with bt_ and get rid of device_?
> >>>> 
> >>>> device_ indicates that we get the BD_ADDR for a 'struct device' and
> >>>> not for e.g. a 'struct fwnode_handle'.
> >>>> 
> >>>> Anyway with this version of the patch fwnode_get_bd_address() has been
> >>>> scrapped and it might never be introduced again, so I'm open to change
> >>>> the name to bt_ if there is a general preference for it.
> >>> 
> >>> Marcel, can you live with this being added to the Bluetooth code base
> >>> instead of property? Also if you'd prefer the function to be named
> >>> bt_get_bd_address() let me know.
> >> 
> >> explain to me again why this is useful?
> > 
> > The official binding for providing the BD_ADDR through the device tree
> > is the property 'local-bd-address'. device_get_bd_address() provides a
> > common API to retrieve the BD_ADDR instead of requiring BT drivers to
> > use the lower level fwnode_property_read_u8_array(). It also avoids
> > repeating the check for an all zeroes BD_ADDR in multiple drivers.
> > 
> >> I am failing to see the benefit if this is not part of the property.h API.
> > 
> > My understanding is that the intention of property.h it to provide an
> > API for common property types used by drivers from different
> > subsystems, hence the implementation 'lives' in drivers/base.
> > Obtaining the BD_ADDR is clearly limited to the Bluetooth subsystem,
> > and drivers/base doesn't seem to be the right place for it. It's true,
> > device_get_mac_address() lives in the common property code, but that
> > doesn't necessarily mean it really should be there and we should do
> > the same. I agree with Sakari that the the approach taken by V4L2
> > (drivers/media/v4l2-core/v4l2-fwnode.c) seems more appropriate.
> > 
> > That said I wouldn't raise opposition if the maintainers of
> > drivers/base agreed to add device_get_mac_address() there, however so
> > far several recent authors of property.[ch] have raised objections.
> 
> so if this is not in drivers/base/ then what is the point in making
> each driver do this? If it is a common property, then it can be well
> handled in the Bluetooth core when setting up the hardware.

Agreed, it would be better if this can be handled in the Bluetooth core.

> This whole BD_ADDR via DT is stupid anyway. Just so that is clear
> up-front. It has been a total hack and fully relies on boot loaders
> doing too much magic and then using DT to hide this magic. The
> BD_ADDR is required to be unique and that means no user will ever
> create a DT with that set. The boot loader always has to read some
> magic value and then convert it and merge it into the actual DT
> provided to the kernel. The clean part would be just to have proper
> APIs to read the memory of the persistent / programmed BD_ADDR and
> then access that.

Yes, the DT approach relies on the bootloader which isn't ideal. Part
of the problem is that AFAIK there is currently no standard way for
storing/retrieving persistent, board specific values like BD ADDR, so
different custom mechanisms are used, which tend to be incompatible
with each other (e.g. Chrome OS uses VPD:
https://chromium.googlesource.com/chromiumos/platform/vpd/)

Using the bootloader & DT is a pragmatic approach, since the DT is
well established and the bootloader often already has DT awareness.

That said I agree that a common solution would make our lives easier.

> That all said, we have hdev->set_bdaddr address and the
> HCI_QUIRK_INVALID_BDADDR to mark the controller as not fully set
> up. And then actually user space can deal with getting the correct
> address and providing it. The code is already there that handles all
> of this if the BD_ADDR comes from user space. Actually hacking this
> into the driver and doing that in the hdev->setup callback is quirky
> to begin with. A user space provided address will just overwrite
> that.
> 
> If you really want to make this generic, then introduce
> HCI_QUIRK_USE_BDADDR_PROPERTY that a driver can set and then do that
> all in hci_dev_do_open() so that if no user space provided BD_ADDR
> is available, it is read from local-bt-address property and if that
> is not available or empty, then mark the the device as unconfigured.
> 
> I am intentionally saying unconfigured when you set
> HCI_QUIRK_USE_BDADDR_PROPERTY since I assume that the logic that we
> have behind HCI_QUIRK_INVALID_BDADDR is implied and whatever address
> comes back via Read_BD_Address is invalid. Otherwise this hardware
> should not set HCI_QUIRK_USE_BDADDR_PROPERTY at all.

Thanks for proposing this generic alternative solution and providing
details!

I'm not really experienced with hacking the Bluetooth core and don't
understand your proposal entirely:

I get the part of setting the quirk in the driver, checking for it in
hci_dev_do_open(), reading the address from 'local-bd-address',
setting it with hdev->bd_addr and marking the device as unconfigured
if the address is empty/unavailable. I interpret that you suggest that
'local-bd-address' should only be used if user space doesn't provide a
BD_ADDR. It is not evident to me where a user space provided address
is set, in any case it doesn't seem to be in hci_dev_do_open(), my
uneducated guess is that the address is set with the management
command SET_PUBLIC_ADDRESS. Could you clarify on this?

I also wonder how to identify the DT node corresponding to an HCI
device, for hci_qca it's the node of hdev->dev.parent, but I'm not
sure if that is universally true. If it isn't looking for the first
parent with a DT node could be an option.

Thanks

Matthias

Powered by blists - more mailing lists