Re: [PATCH v8 3/7] Bluetooth: btqca: Redefine qca_uart_setup() to generic function.

[Matthias Kaehlcke]

On Tue, Jun 26, 2018 at 06:53:47AM +0530, Balakrishna Godavarthi wrote:
> Hi Matthias,
> 
> On 2018-06-26 04:50, Matthias Kaehlcke wrote:
> > On Mon, Jun 25, 2018 at 07:10:09PM +0530, Balakrishna Godavarthi wrote:
> > > Redefinition of qca_uart_setup will help future Qualcomm Bluetooth
> > > SoC, to use the same function instead of duplicating the function.
> > > Added new arguments soc_type and soc_ver to the functions.
> > > 
> > > These arguments will help to decide type of firmware files
> > > to be loaded into Bluetooth chip.
> > > soc_type holds the Bluetooth chip connected to APPS processor.
> > > soc_ver holds the Bluetooth chip version.
> > > 
> > > Signed-off-by: Balakrishna Godavarthi <bgodavar@xxxxxxxxxxxxxx>
> > > ---
> > > Changes in v8:
> > >     * updated soc_type with enum.
> > > 
> > > Changes in v7:
> > >     * initial patch
> > >     * redefined qca_uart_setup function to generic.
> > > ---
> > >  drivers/bluetooth/btqca.c   | 23 ++++++++++++-----------
> > >  drivers/bluetooth/btqca.h   | 13 +++++++++++--
> > >  drivers/bluetooth/hci_qca.c |  3 ++-
> > >  3 files changed, 25 insertions(+), 14 deletions(-)
> > > 
> > > diff --git a/drivers/bluetooth/btqca.c b/drivers/bluetooth/btqca.c
> > > index c5cf9cab438a..3b25be1be19c 100644
> > > --- a/drivers/bluetooth/btqca.c
> > > +++ b/drivers/bluetooth/btqca.c
> > > @@ -327,9 +327,9 @@ int qca_set_bdaddr_rome(struct hci_dev *hdev,
> > > const bdaddr_t *bdaddr)
> > >  }
> > >  EXPORT_SYMBOL_GPL(qca_set_bdaddr_rome);
> > > 
> > > -int qca_uart_setup(struct hci_dev *hdev, uint8_t baudrate)
> > > +int qca_uart_setup(struct hci_dev *hdev, uint8_t baudrate,
> > > +		   enum qca_btsoc_type soc_type, u32 soc_ver)
> > >  {
> > > -	u32 rome_ver = 0;
> > >  	struct rome_config config;
> > >  	int err;
> > > 
> > > @@ -337,19 +337,20 @@ int qca_uart_setup(struct hci_dev *hdev,
> > > uint8_t baudrate)
> > > 
> > >  	config.user_baud_rate = baudrate;
> > > 
> > > -	/* Get QCA version information */
> > > -	err = qca_read_soc_version(hdev, &rome_ver);
> > > -	if (err < 0 || rome_ver == 0) {
> > > -		bt_dev_err(hdev, "QCA Failed to get version %d", err);
> > > -		return err;
> > > +	if (!soc_ver) {
> > > +		/* Get QCA version information */
> > > +		err = qca_read_soc_version(hdev, &soc_ver);
> > > +		if (err < 0 || soc_ver == 0) {
> > > +			bt_dev_err(hdev, "QCA Failed to get version (%d)", err);
> > > +			return err;
> > > +		}
> > > +		bt_dev_info(hdev, "QCA controller version 0x%08x", soc_ver);
> > >  	}
> > 
> > I thought we agreed in the discussion on "[v7,4/8] Bluetooth: btqca:
> > Redefine qca_uart_setup() to generic function" to call
> > qca_read_soc_version() in common code. Did I misinterpret that?
> > 
> [Bala]: After integrating wcn3990, calling qca_read_soc_version() in
> qca_setup()
>         is not preferable. as we will have multiple common blocks of code in
> qca_setup.
>         calling function to set an operator speed is required in the both
> the if -else blcoks

We can probably agree that there is no ideal solution, there is some
ugliness in on way or the other. IMO the conditional
qca_read_soc_version() in qca_uart_setup() based on the vale of
'soc_ver' is far worse than a small piece of redundant code.

If qca_read_soc_version() was done in qca_setup() the code could look
something like this:

static int qca_setup(struct hci_uart *hu)
{
	...
	if (qcadev->btsoc_type == QCA_WCN3990) {
		...
		qca_read_soc_version();
		ret = qca_set_speed(hu, QCA_OPER_SPEED);
                if (ret)
                        return ret;
	} else {
		ret = qca_set_speed(hu, QCA_OPER_SPEED);
                if (ret)
                        return ret;
                qca_read_soc_version();
	}

	speed = qca_get_speed(hu, QCA_OPER_SPEED);
	qca_baudrate = qca_get_baudrate_value(speed);

	 /* Setup patch / NVM configurations */
        ret = qca_uart_setup(hdev, qca_baudrate, qcadev->btsoc_type, soc_ver);
	...
}

Yes, 'qca_set_speed(hu, QCA_OPER_SPEED)' and the error handling is
redundant, but it's only 3 lines of trivial code in exchange for
making qca_uart_setup() more consistent and not spreading
the qca_read_soc_version() calls over multiple files, depending on the
SoC version.

If you are super-convinced that the split is superior leave it as is,
I might already be doing too much bike-shedding, and after all it
isn't my code.

> > > diff --git a/drivers/bluetooth/btqca.h b/drivers/bluetooth/btqca.h
> > > index 5c9851b11838..24d6667eecf1 100644
> > > --- a/drivers/bluetooth/btqca.h
> > > +++ b/drivers/bluetooth/btqca.h
> > > ...
> > > -static inline int qca_uart_setup(struct hci_dev *hdev, uint8_t
> > > baudrate)
> > > +static inline int qca_uart_setup(struct hci_dev *hdev, uint8_t
> > > baudrate,
> > > +				 enum qca_btsoc_type soc_type, u32 soc_ver);
> > 
> > Remove trailing semicolon.
> 
> [Bala]: i didn't get you.

Sorry, I should have left more context:

> static inline int qca_uart_setup(struct hci_dev *hdev, uint8_t baudrate,
> 				 enum qca_btsoc_type soc_type, u32 soc_ver);
> {
> 	return -EOPNOTSUPP;
> }

This is a function definition, not just a declaration. The semicolon
would make it a declaration and make the compiler unhappy about a
function body where it doesn't expect it.