Re: [PATCH v7 4/8] Bluetooth: btqca: Redefine qca_uart_setup() to generic function.

[Matthias Kaehlcke <mka@chromium.org>]

On Fri, Jun 22, 2018 at 08:41:01PM +0530, Balakrishna Godavarthi wrote:
> Hi Matthias,
> 
> On 2018-06-22 03:39, Matthias Kaehlcke wrote:
> > On Thu, Jun 21, 2018 at 04:50:28PM +0530, Balakrishna Godavarthi wrote:
> > > Hi Matthias,
> > > 
> > > On 2018-06-21 05:03, Matthias Kaehlcke wrote:
> > > > Hi Balakrishna,
> > > >
> > > > On Wed, Jun 20, 2018 at 04:23:25PM +0530, Balakrishna Godavarthi wrote:
> > > > > Hi Matthias,
> > > > >
> > > > > On 2018-06-20 01:33, Matthias Kaehlcke wrote:
> > > > > > On Tue, Jun 19, 2018 at 12:39:07PM +0530, Balakrishna Godavarthi wrote:
> > > > > > > Hi Matthias,
> > > > > > >
> > > > > > > Please find my comments inline.
> > > > > > >
> > > > > > > On 2018-06-19 02:49, Matthias Kaehlcke wrote:
> > > > > > > > On Sat, Jun 16, 2018 at 11:57:14AM +0530, Balakrishna Godavarthi wrote:
> > > > > > > >
> > > > > > > > > @@ -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);
> > > > > > > > >  	}
> > > > > > > >
> > > > > > > > Why is this needed? A caller that passes in soc_ver = 0 could just
> > > > > > > > call qca_read_soc_version() themselves.
> > > > > > > [Bala]: hci_qca support two chip one ROME (already up-streamed) and
> > > > > > > other
> > > > > > > wcn3990.
> > > > > > >         there setup sequence differs.
> > > > > > >
> > > > > > >         wcn3900:
> > > > > > >         1. set baudrate to 115200
> > > > > > >         2. read soc version
> > > > > > >         3. set baudarte to 3.2Mbps
> > > > > > >         4. download firmware. (calls qca_uart_setup)
> > > > > > >
> > > > > > >        ROME:
> > > > > > >        1. set baudrate to 3.0 mbps
> > > > > > >        2. calls qca_uart_setup to read soc version and download
> > > > > > > firmware.
> > > > > > >
> > > > > > >       so to make use of existing function qca_setup_uart(). passing
> > > > > > > soc_ver
> > > > > > > as argument.
> > > > > > >       if soc_ver is zero, then read soc version (i.e. for ROME chip).
> > > > > > >
> > > > > > >       Pls let me know, if you have any queries.
> > > > > >
> > > > > > I don't really understand this argumentation. Let's look at the code
> > > > > > at the end of this series, where both Rome and wcn3900 are supported:
> > > > > >
> > > > > > https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/1045608/6/drivers/bluetooth/hci_qca.c#1059
> > > > > >
> > > > > > The version could be read (for Rome only) after setting the operating
> > > > > > speed:
> > > > >
> > > > > [Bala]: yes we can read SoC version after setting operating baud
> > > > > rate. an
> > > > > advice from SoC vendor that, read SoC version before setting operating
> > > > > speed.
> > > > >         one advantage from reading SoC version before setting
> > > > > operating
> > > > > baudrate,it will helps us to understand whether the soc is
> > > > > responding to any
> > > > > commands in default baud rate
> > > > >         before setting operating speed.
> > > >
> > > > Is the recommendation for Rome or wcn3990? I was referring to Rome and
> > > > the current code sets the operating speed and then reads the SoC
> > > > version (inside qca_uart_setup())
> > > >
> > > 
> > > [Bala]: my statement is wrt to wcn3990. yes, ROME will set operating
> > > speed
> > > and read version.
> > > 
> > > > > > https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/1045608/6/drivers/bluetooth/hci_qca.c#1111
> > > > > >
> > > > > > I concede that having two "if (qcadev->btsoc_type == XYZ)" branches in
> > > > > > qca_setup() is kinda ugly, but I think it's still better than the
> > > > > > conditional call of qca_read_soc_version() in qca_uart_setup().
> > > > >
> > > > > [Bala]: we require an if-else block  for soc type.As wcn3900 support
> > > > > hardware flow control where as ROME is not supporting hardware flow
> > > > > control.
> > > >
> > > > Since you mention if-else and flow control I'm not sure if you
> > > > understood correctly what I suggested. My suggestion is to add
> > > > something like this:
> > > >
> > > > if (qcadev->btsoc_type == QCA_ROME) {
> > > > 	ret = qca_read_soc_version(hdev, &soc_ver);
> > > > 	if (ret < 0 || soc_ver == 0) {
> > > > 	   	// Note mka@: we might want to skip this log,
> > > > 		// qca_read_soc_version() already logs in all error paths
> > > > 		bt_dev_err(hdev, "Failed to get version %d", ret);
> > > > 		return ret;
> > > > 	}
> > > > }
> > > >
> > > > right here:
> > > >
> > > > https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/1045608/6/drivers/bluetooth/hci_qca.c#1111
> > > >
> > > > which is functionally equivalent to the current code.
> > > >
> > > > You could even do the error check in the common code (just checking
> > > > for 'soc_ver == 0', assuming that qca_read_soc_version() doesn't set
> > > > it to something different in case of error).
> > > >
> > > > Anyway, I won't insist if you prefer it as is and Marcel is ok with it.
> > > 
> > > [Bala]: thanks for clarifying my doubt, but if remove the
> > > qca_read_soc_version() from qca_uart_setup() and handle it
> > > qca_setup().
> > >         in that case, we will have common blocks of code, when we
> > > integrate
> > > wcn3990 into existing qca_setup()
> > 
> > Excellent!
> > 
> > >        static int qca_setup(struct hci_uart *hu)
> > >       {
> > >         ...
> > > 
> > >         qcadev = serdev_device_get_drvdata(hu->serdev);
> > > 
> > >         /* Patch downloading has to be done without IBS mode */
> > >         clear_bit(STATE_IN_BAND_SLEEP_ENABLED, &qca->flags);
> > > 
> > >         /* Setup initial baudrate */
> > >         speed = qca_get_speed(hu, QCA_INIT_SPEED);
> > >         if (speed)
> > >                 qca_set_speed(hu, speed, QCA_INIT_SPEED);
> > 
> > I said earlier that a _set_speed() followed by a _get_speed() is a bit
> > odd, and that it would be better to get the speed first and pass it as
> > a parameter. After looking again over the larger code I'm not
> > convinced anymore this is such a good idea, since the above pattern is
> > repeated multiple times, when it could be just:
> > 
> > qca_set_speed(hu, QCA_INIT_SPEED)
> > 
> > (+ error handling, which is also missing in the code above)
> > 
> > Just a thought, not asking you to change it back, do whatever seems
> > best to you.
> > 
> > >         if (qcadev->btsoc_type == QCA_WCN3990) {
> > >                 bt_dev_dbg(hdev, "setting up wcn3990");
> > >                 hci_uart_set_flow_control(hu, true);
> > >                 ret = qca_send_vendor_cmd(hdev,
> > > QCA_WCN3990_POWERON_PULSE);
> > >                 if (ret) {
> > >                         bt_dev_err(hdev, "failed to send power on
> > > command");
> > >                         return ret;
> > >                 }
> > > 
> > >                 serdev_device_close(hu->serdev);
> > >                 ret = serdev_device_open(hu->serdev);
> > >                 if (ret) {
> > >                         bt_dev_err(hdev, "failed to open port");
> > >                         return ret;
> > >                 }
> > > 
> > >                 msleep(100);
> > >                 speed = qca_get_speed(hu, QCA_INIT_SPEED);
> > >                 if (speed)
> > >                         qca_set_speed(hu, speed, QCA_INIT_SPEED);
> > > 
> > >                 hci_uart_set_flow_control(hu, false);
> > >         } else {
> > >                 bt_dev_info(hdev, "ROME setup");
> > >                 /* Setup user speed if needed */
> > >                 speed = qca_get_speed(hu, QCA_OPER_SPEED);
> > >                 if (speed) {
> > >                         ret = qca_set_speed(hu, speed,
> > > QCA_OPER_SPEED);
> > >                         if (ret)
> > >                                 return ret;
> > > 
> > >                         qca_baudrate = qca_get_baudrate_value(speed);
> > 
> > It seems setting 'qca_baudrate' could also be done in the common
> > path. Would require an extra qca_get_speed(hu, QCA_OPER_SPEED) call,
> > in exchange this would be done in a single place closer to where
> > 'qca_baudrate' is actually used.
> > 
> > Actually as it is now 'qca_baudrate' could remain uninitialized for
> > Rome if no operating speed is defined, which judging from
> > qca_get_baudrate_value() should result in QCA_BAUDRATE_115200.
> > 
> > >                 }
> > >         }
> > > 
> > >         ret = qca_read_soc_version(hdev, &soc_ver);
> > >         if (ret < 0 || soc_ver == 0) {
> > >                 bt_dev_err(hdev, "Failed to get version %d", ret);
> > >                 return ret;
> > >         }
> > >         bt_dev_info(hdev, "wcn3990 controller version 0x%08x",
> > > soc_ver);
> > > 
> > > 
> > >         if (qcadev->btsoc_type == QCA_WCN3990) {
> > >                 hci_uart_set_flow_control(hu, true);
> > >                 /* Setup user speed if needed */
> > >                 speed = qca_get_speed(hu, QCA_OPER_SPEED);
> > >                 if (speed) {
> > >                         ret = qca_set_speed(hu, speed,
> > > QCA_OPER_SPEED);
> > >                         if (ret)
> > >                                 return ret;
> > > 
> > >                 qca_baudrate = qca_get_baudrate_value(speed);
> > > 
> > >         }
> > > 
> > >         /* Setup patch / NVM configurations */
> > >         ret = qca_uart_setup(hdev, qca_baudrate, qcadev->btsoc_type,
> > > soc_ver);
> > >         if (!ret) {
> > >                 set_bit(STATE_IN_BAND_SLEEP_ENABLED, &qca->flags);
> > >                 qca_debugfs_init(hdev);
> > >         } else if (ret == -ENOENT) {
> > >                 /* No patch/nvm-config found, run with original
> > > fw/config */
> > >                 ret = 0;
> > >         } else if (ret == -EAGAIN) {
> > >                 /*
> > >       ...
> > >       }
> > 
> > Overall looks good and should still become a bit more compact if the
> > handling of flow control is done in _set_speed() as you suggested in
> > "[PATCH v7 8/8] Bluetooth: hci_qca: Add support for Qualcomm Bluetooth
> > chip wcn3990"
> 
> [Bala}: to over come calling qca_get_speed() for qca_baudrate.
>         i prefer passing qca_baudrate as pointer argument to
> qca_set_speed().
>         that  will actually helps to lower function calls. intern increase
> the speed of execution too.

I argued earlier against something similar, where speed was a pointer
that was set inside _set_speed(). IMO this interface is unintuitive
and makes the code harder to read, even if it saves a function
call. The optimization isn't really an important factor here,
_get_speed() is cheap and it isn't called in a hot code path.

It's also an option to add an enum qca_speed_type parameter to
qca_get_baudrate_value() and hide the _get_speed() call there.