Message-ID: <AANLkTin72z7AmfkMMH01siwkNNNvHM+1+qXLodmaUx5q@mail.gmail.com>
Date:	Thu, 28 Oct 2010 12:32:09 +0200
From:	Par-Gunnar Hjalmdahl <pghatwork@...il.com>
To:	Arnd Bergmann <arnd@...db.de>
Cc:	linus.walleij@...ricsson.com, linux-bluetooth@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/9] mfd: Add support for the ST-Ericsson CG2900.

Thanks again for your comments Arnd.
Sorry for not answering your mail earlier but I've spent the last few
days changing a few hundred debug comments. :-)
See answers below.

/P-G

2010/10/26 Arnd Bergmann <arnd@...db.de>:
> On Tuesday 26 October 2010, Par-Gunnar Hjalmdahl wrote:
>
>> 2010/10/22 Arnd Bergmann <arnd@...db.de>:
>> > On Friday 22 October 2010 12:35:16 Par-Gunnar Hjalmdahl wrote:
>> >> This patch adds support for the ST-Ericsson CG2900 Connectivity
>> >> Combo controller.
>> >> This patch adds the central framework to be able to register CG2900 users,
>> >> transports, and chip handlers.
>> >>
>> >> Signed-off-by: Par-Gunnar Hjalmdahl <par-gunnar.p.hjalmdahl@...ricsson.com>
>> >
>> > Looks ok from a coding style perspective, but some important information is
>> > missing from the description:
>> >
>> > * What is a CG2900?
>> > * Why is it a MFD device rather than e.g. a bus or a subsystem?
>> >
>>
>> I will add some more info:
>>  - CG2900 is a GPS, Bluetooth, FM controller
>>  - I do not really know what makes a device qualify as a certain type,
>> but at least for us this is a multifunction device. But I can't really
>> say that it isn't really a bus (could be stated as multifunction HCI
>> bus). I guess it could be qualified as a subsystem as well, but I
>> cannot give you a reason to have it as either.
>
> It's not often completely clear, but I would draw the distinction like
> this:
>
> * A multi-function device is a single device that sits on a given bus
> with one host I/O interface and provides functionality to different
> logical subsystems like serial or alsa. A typical case for an MFD would
> be a to solve the problem of sharing resources between the child drivers
> because you cannot bind one device to two drivers. If CG2900 is a single
> piece of silicon that always looks the same way, it's probably an MFD.
>
> * A bus is a much more general abstraction which has multiple devices
> and multiple device types behind it. The bus has no idea what devices
> are attached to it at compile time, so you either probe the devices at
> boot time or define a set of devices in a board definition (platform
> devices). Driver modules then register a struct device_driver for the
> bus, which gives all matching devices to the bus. If you can have
> future CG2900 compatible devices that need to have other drivers, e.g.
> for HID or video, it should probably be a bus.
>
> * A subsystem is less clearly defined, I would call bluetooth a subsystem
> instead of just a bus because it not only matches devices with drivers
> but also has its own user-level interfaces for raw communication.
> If you want to be able to have drivers in the kernel and in user space,
> you might want to consider starting a new drivers/cg2900 subsystem, or
> integrating into the bluetooth subsystem if that makes sense.
>

Thanks for this explanation. I would like to see this as a MFD at the
moment since it is most definitely a silicon, as stated by Linus
Walleij in a separate answer.
In the future this could however be extended to be a bus, when we
could use more chips with this driver.

>> >> +config MFD_CG2900
>> >> +     tristate "Support ST-Ericsson CG2900 main structure"
>> >> +     depends on NET
>> >> +     help
>> >> +       Support for ST-Ericsson CG2900 Connectivity Combo controller main structure.
>> >> +       Supports multiple functionalities muxed over a Bluetooth HCI H:4 interface.
>> >> +       CG2900 support Bluetooth, FM radio, and GPS.
>> >> +
>> >
>> > Can you explain what it means to mux over a H:4 interface? Does this mean
>> > you use bluetooth infrastructure that is designed for wireless communication
>> > in order to connect on-board or on-chip devices?
>> >
>>
>> The H:4 protocol is defined in the Bluetooth Core specification. It
>> uses a single byte first in the data packet to determine an HCI
>> channel. In the Bluetooth Core specification there are 4 channels
>> specified, 0x01-0x04 (0x01 is BT command, etc). The rest of the
>> channels are reserved, but these have been used on a proprietary basis
>> to transport data to different IPs within the controller. In our API
>> we have chosen to have a channel separation on an API basis and the
>> H:4 byte is then added within the driver. So we have multiple channels
>> coming in from the users that we mux onto a single data transport.
>> That's what I mean in the text. I guess I could rewrite it a bit to
>> make it clearer.
>
> I still don't understand it, though that may be a problem on my side.
>
> Does that mean that cg2900 is to the host side a standard bluetooth HCI,
> but it uses extensions to the HCI in order to make the other functions
> available to the host?
>
> What chapter in the bluetooth core specification describes this?
>

Yes, exactly. Just for Bluetooth functionality you could more or less
use existing Bluetooth UART drivers (you still have to enable chip and
so on separately though), but in order to reach FM and GPS this
protocol has been extended. In order to have an energy efficient
solution there must also be a common driver that can shut off the chip
when not in use, hence this CG2900 driver.
You can read about the original protocol in the Bluetooth Core
specification Volume 4 Part A - UART Transport Layer.
This way to extend functionality upon the existing protocol is common
amongst chip manufacturers for combo chips including Bluetooth
functionality. Texas Instruments has already posted a driver for their
chip (in drivers/staging/ti-st) that uses a similar protocol.

>> >> +
>> >> +/**
>> >> + * add_h4_user() - Add H4 user to user storage based on supplied H4 channel.
>> >> + * @dev:     Stored CG2900 device.
>> >> + * @name:    Device name to identify different devices that are using
>> >> + *           the same H4 channel.
>> >> + *
>> >> + * Returns:
>> >> + *   0 if there is no error.
>> >> + *   -EINVAL if NULL pointer or bad channel is supplied.
>> >> + *   -EBUSY if there already is a user for supplied channel.
>> >> + */
>> >> +static int add_h4_user(struct cg2900_device *dev, const char * const name)
>> >> +{
>> >
>> > Where does that name come from? Why not just use an enum if the name is
>> > only use for strncmp?
>> >
>>
>> At a point in time we used to have an enum, but from what I can see it
>> is easier to keep an API stable if you use name lookup instead. If we
>> want to have minimal changes to the API in the future this is quite a
>> flexible solution, but this code should be moved to each respective
>> chip handler.
>
> You haven't really answered the first question, where the name comes from
> (I guess the question was not too clear). Does this e.g. get passed by a
> user application, or does it come from a hardware description of some
> sort?
>
> It looks a bit like a handcoded version of the code we use for device/driver
> matching in drivers/core. If that's the case, it would be better to use
> the existing infrastructure and create a bus with devices that can be
> matched with drivers.
>

Oops, I misunderstood your question. The name is supplied by the user
calling cg2900_register_user. The names are defined for reference in
include/linux/mfd/cg2900.h.
We used to have an enum to define this, but since we thought that
using name lookup to be more future safe (and as well more "Kernel
like") we chose to use that instead. It can have been a bad choice.
Problem here is to some extent that we are beginners at Linux coding.
A straight minimal API seemed to be easiest way to solve it. As you
say we could probably use device registration as you state (we are
already using MFD cells so we could possibly extend that) but I don't
know if that would improve our code.
We also have a lot of internal dependencies (like our FM driver that
will be submitted to the Kernel community) so would like to avoid
changing APIs if possible. But if you make this an absolute demand I
guess we will have to follow.

>> >> +static ssize_t test_char_dev_read(struct file *filp, char __user *buf,
>> >> +                               size_t count, loff_t *f_pos)
>> >> +{
>> >> +     struct sk_buff *skb;
>> >> +     int bytes_to_copy;
>> >> +     int err;
>> >> +     struct sk_buff_head *rx_queue = &core_info->test_char_dev->rx_queue;
>> >
>> > What is this read/write interface used for?
>> >
>> > The name suggests that this is only for testing and not an essential
>> > part of your driver. Could this be made a separate driver?
>> >
>> > It also looks like you do socket communication here, can't you use
>> > a proper AF_BLUETOOTH socket to do the same?
>> >
>>
>> This interface can be used for module testing where you can have a
>> test engine in user space that acts as a chip. Yes, it could be made
>> into a separate driver but it would be a very small driver. Don't know
>> if it will be worth having it in a separate file...
>
> I don't think file size is a major argument here. If it's used for
> testing, I'd make it a separate module that defaults to not being
> built, in order to prevent people from relying on what you intended
> as a testing interface.
>

OK. We can do this.

>> We use the sk_buffer, which I guess is the reason that you mention
>> sockets. We could of course use a socket instead of a char dev, both
>> here and in the cg2900_char_dev file (which of course then would be
>> renamed). It was just a choice we made during development. We think
>> that the transport as such is more like a streaming interface, but I
>> see no real problem to have sockets. We already have several users
>> using char dev though so I would prefer to keep char dev rather than
>> switching to sockets unless you have a strong reason for this.
>> I see no reason to use AF_BLUETOOTH though, the transport is not only
>> for Bluetooth.
>
> The question here is what is appropriate. My guess was that the frame
> format was the one used in bluetooth sockets. If that's not the
> case, AF_BLUETOOTH would obviously be wrong.
>
> For the test interface, it doesn't matter that much what you use, but
> for the cg2900_char_dev we should make sure that you considered all
> the options and made a reasonable decision. What does the data format
> look like there? My impression is that the character device might be
> too low-level and it could be better to use some structured interface
> like a socket, but it really depends on what it actually does.
>
> In particular, if you can address multiple hardware units over one
> character device, it may be better to have separate devices for each
> of them.
>

What is transmitted over each char dev is individual for that channel.
For BT channels this will be BT data (normally char dev is not used
for Bluetooth since BT stack connection is done directly in Kernel
through btcg2900.c), for FM channel this will be FM data, etc. But
most of all it will not be the same format as the sockets provided by
the Bluetooth stack (at least I think so) since BT stack have
protocols below the sockets (such as L2CAP). What is sent over the
char dev is the same data as sent to the chip minus the H4 header (the
first byte).
The CG2900 driver is currently written only to support one CG2900
chip, but of course that could be extended in the future.

>> >> +     CG2900_INFO("test_char_dev_read");
>> >> +
>> >> +     if (skb_queue_empty(rx_queue))
>> >> +             wait_event_interruptible(char_wait_queue,
>> >> +                                      !(skb_queue_empty(rx_queue)));
>> >> +
>> >> +     skb = skb_dequeue(rx_queue);
>> >> +     if (!skb) {
>> >> +             CG2900_INFO("skb queue is empty - return with zero bytes");
>> >> +             bytes_to_copy = 0;
>> >> +             goto finished;
>> >> +     }
>> >> +
>> >> +     bytes_to_copy = min(count, skb->len);
>> >> +     err = copy_to_user(buf, skb->data, bytes_to_copy);
>> >> +     if (err) {
>> >> +             skb_queue_head(rx_queue, skb);
>> >> +             return -EFAULT;
>> >> +     }
>> >> +
>> >> +     skb_pull(skb, bytes_to_copy);
>> >> +
>> >> +     if (skb->len > 0)
>> >> +             skb_queue_head(rx_queue, skb);
>> >> +     else
>> >> +             kfree_skb(skb);
>> >> +
>> >> +finished:
>> >> +     return bytes_to_copy;
>> >> +}
>> >
>> > This does not handle short/continued reads.
>> >
>>
>> As I mentioned above this interface is used for testing and we
>> therefore have some restriction of usage. I don't think we need to
>> impose all things necessary for a full interface upon it.
>
> This is where it gets complicated. We don't normally add interfaces
> to the kernel unless we expect to fully support them for the forseeable
> future. It has happened more than enough in the past that somebody
> introduced a debugging interface which subsequently became a supported
> ABI because some application started relying on it.
>
> Is the code that uses the test interface even publically available?
> If not, I would recommend saving the trouble of arguing about it
> and leaving out the interface.
>

OK. Good idea. I will split out the test code to a separate file and
leave it out of next patch set.

>> >> +     #define CG2900_ERR(fmt, arg...)                                 \
>> >> +     do {                                                            \
>> >> +             if (cg2900_debug_level >= 1)                            \
>> >> +                     pr_err("CG2900 %s: " fmt "\n" , __func__ , ## arg); \
>> >> +     } while (0)
>> >> +
>> >> +#endif /* NDEBUG */
>> >
>> > You'll feel relieved when all this is gone...
>> >
>>
>> The only thing is it's been pretty nice to have it, but I will remove it.
>> Is it OK to keep defines so we can have "CG2900" in front and "\n"
>> after the text?
>
> Ideally, you would use the dev_* functions instead of the pr_* functions,
> which print the name of the device before the message. I also wouldn't
> add the "\n" in the macro because all the regular printing functions don't
> do it. It will likely only confuse people and the binary remains the same.
>
>        Arnd
>

OK. I've updated the code (will come in next patch set). I use dev_dbg
where possible.

/P-G
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists