Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Content-Type: text/plain;
        charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 13.0 \(3601.0.10\))
Subject: Re: [PATCH v6 4/4] Bluetooth: hci_bcm: Support pcm params in dts
From:   Marcel Holtmann <marcel@holtmann.org>
In-Reply-To: <CANFp7mX8zsFBjQ4pYyBnG4KbTv-0m004LPmYAtLmGRm3rkfoRA@mail.gmail.com>
Date:   Wed, 20 Nov 2019 00:42:55 +0100
Cc:     Johan Hedberg <johan.hedberg@gmail.com>,
        Rob Herring <robh+dt@kernel.org>,
        linux-bluetooth@vger.kernel.org,
        Douglas Anderson <dianders@chromium.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Transfer-Encoding: 8BIT
Message-Id: <A98BB7FA-064A-4824-81F1-E5C8F204BBD0@holtmann.org>
References: <20191118192123.82430-1-abhishekpandit@chromium.org>
 <20191118110335.v6.4.I3e900de9478b68e5e4475e747d1c46fdd28313fa@changeid>
 <963EC7CC-B81D-4667-A681-2CE49D17CB1E@holtmann.org>
 <CANFp7mX8zsFBjQ4pYyBnG4KbTv-0m004LPmYAtLmGRm3rkfoRA@mail.gmail.com>
To:     Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Abhishek,

>>> BCM chips may require configuration of PCM to operate correctly and
>>> there is a vendor specific HCI command to do this. Add support in the
>>> hci_bcm driver to parse this from devicetree and configure the chip.
>>> 
>>> Signed-off-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
>>> ---
>>> 
>>> Changes in v6:
>>> - Added btbcm_read_pcm_int_params and change pcm params to first read
>>> the pcm params before setting it
>>> 
>>> Changes in v5:
>>> - Rename parameters to bt-* and read as integer instead of bytestring
>>> - Update documentation with defaults and put values in header
>>> - Changed patch order
>>> 
>>> Changes in v4:
>>> - Fix incorrect function name in hci_bcm
>>> 
>>> Changes in v3:
>>> - Change disallow baudrate setting to return -EBUSY if called before
>>> ready. bcm_proto is no longer modified and is back to being const.
>>> - Changed btbcm_set_pcm_params to btbcm_set_pcm_int_params
>>> - Changed brcm,sco-routing to brcm,bt-sco-routing
>>> 
>>> Changes in v2:
>>> - Use match data to disallow baudrate setting
>>> - Parse pcm parameters by name instead of as a byte string
>>> - Fix prefix for dt-bindings commit
>>> 
>>> drivers/bluetooth/hci_bcm.c | 57 +++++++++++++++++++++++++++++++++++++
>>> 1 file changed, 57 insertions(+)
>>> 
>>> diff --git a/drivers/bluetooth/hci_bcm.c b/drivers/bluetooth/hci_bcm.c
>>> index ee40003008d8..2ce3fac2c5dd 100644
>>> --- a/drivers/bluetooth/hci_bcm.c
>>> +++ b/drivers/bluetooth/hci_bcm.c
>>> @@ -25,6 +25,7 @@
>>> #include <linux/pm_runtime.h>
>>> #include <linux/serdev.h>
>>> 
>>> +#include <dt-bindings/bluetooth/brcm.h>
>>> #include <net/bluetooth/bluetooth.h>
>>> #include <net/bluetooth/hci_core.h>
>>> 
>>> @@ -88,6 +89,7 @@ struct bcm_device_data {
>>> *    used to disable flow control during runtime suspend and system sleep
>>> * @is_suspended: whether flow control is currently disabled
>>> * @no_early_set_baudrate: don't set_baudrate before setup()
>>> + * @pcm_params: PCM and routing parameters
>>> */
>>> struct bcm_device {
>>>      /* Must be the first member, hci_serdev.c expects this. */
>>> @@ -122,6 +124,8 @@ struct bcm_device {
>>>      bool                    is_suspended;
>>> #endif
>>>      bool                    no_early_set_baudrate;
>>> +
>>> +     struct bcm_set_pcm_int_params   pcm_params;
>>> };
>>> 
>>> /* generic bcm uart resources */
>>> @@ -541,6 +545,7 @@ static int bcm_flush(struct hci_uart *hu)
>>> static int bcm_setup(struct hci_uart *hu)
>>> {
>>>      struct bcm_data *bcm = hu->priv;
>>> +     struct bcm_set_pcm_int_params p;
>>>      char fw_name[64];
>>>      const struct firmware *fw;
>>>      unsigned int speed;
>>> @@ -594,6 +599,31 @@ static int bcm_setup(struct hci_uart *hu)
>>>                      host_set_baudrate(hu, speed);
>>>      }
>>> 
>>> +     /* PCM parameters if any*/
>>> +     err = btbcm_read_pcm_int_params(hu->hdev, &p);
>>> +     if (!err) {
>>> +             if (bcm->dev->pcm_params.routing == 0xff)
>>> +                     bcm->dev->pcm_params.routing = p.routing;
>>> +             if (bcm->dev->pcm_params.rate == 0xff)
>>> +                     bcm->dev->pcm_params.rate = p.rate;
>>> +             if (bcm->dev->pcm_params.frame_sync == 0xff)
>>> +                     bcm->dev->pcm_params.frame_sync = p.frame_sync;
>>> +             if (bcm->dev->pcm_params.sync_mode == 0xff)
>>> +                     bcm->dev->pcm_params.sync_mode = p.sync_mode;
>>> +             if (bcm->dev->pcm_params.clock_mode == 0xff)
>>> +                     bcm->dev->pcm_params.clock_mode = p.clock_mode;
>> 
>> Frankly, I wouldn’t bother here. If the read HCI command failed, then we abort bcm_setup and fail the whole procedure. These commands have been around the first Broadcom chips and you can assume they are present. And if at some point they do fail, I want to know about it.
> Ok -- will change to return error if it fails.
> 
>> 
>>> +
>>> +             /* Write only when there are changes */
>>> +             if (memcmp(&p, &bcm->dev->pcm_params, sizeof(p)))
>>> +                     err = btbcm_write_pcm_int_params(hu->hdev,
>>> +                                                      &bcm->dev->pcm_params);
>>> +
>>> +             if (err)
>>> +                     bt_dev_warn(hu->hdev, "BCM: Write pcm params failed (%d)",
>>> +                                 err);
>>> +     } else
>>> +             bt_dev_warn(hu->hdev, "BCM: Read pcm params failed (%d)", err);
>>> +
>>> finalize:
>>>      release_firmware(fw);
>>> 
>>> @@ -1128,9 +1158,36 @@ static int bcm_acpi_probe(struct bcm_device *dev)
>>> }
>>> #endif /* CONFIG_ACPI */
>>> 
>>> +static int property_read_u8(struct device *dev, const char *prop, u8 *value)
>>> +{
>>> +     int err;
>>> +     u32 tmp;
>>> +
>>> +     err = device_property_read_u32(dev, prop, &tmp);
>>> +
>>> +     if (!err)
>>> +             *value = (u8)tmp;
>>> +
>>> +     return err;
>>> +}
>> 
>> I think this really needs to be done in the generic property code if this is wanted.
> Yes, this should be device_property_read_u8. For some reason, I
> thought that wasn't working before (I'll have to retest it with
> straight integer values).
> 
>> 
>>> +
>>> static int bcm_of_probe(struct bcm_device *bdev)
>>> {
>>>      device_property_read_u32(bdev->dev, "max-speed", &bdev->oper_speed);
>>> +
>>> +     memset(&bdev->pcm_params, 0xff, sizeof(bdev->pcm_params));
>> 
>> Scrap this memset. We will read the values first.
> 
> I added this memset is bcm_of_probe occurs before patchram and without
> setting some magic value in the pcm_params, we don't know which values
> are valid (since 0 has some meaning in the params).
> It doesn't make sense to me to read pcm params outside setup (I want
> patchram to complete first) and it doesn't make sense to do property
> reads inside setup.

I wonder if we should just fail if bt-sco-routing is PCM and not all values are provided in the DT. Looks like there is no clean way of doing this gracefully otherwise.

Regards

Marcel