Re: [PATCH v3] Bluetooth: Apply HCI_QUIRK_BROKEN_READ_TRANSMIT_POWER to CYW4373

[Aditya Garg]

Hi

> On 22 May 2024, at 12:34 PM, Nobuaki.Tsunashima@xxxxxxxxxxxx wrote:
> 
> Hi Aditya,
> 
>>> CYW4373 ROM FW has an issue that it claims LE_Read_Transmit_Power
>>> command as supported in a response of Read_Local_Supported_Command
>>> command but rejects the LE_Read_Transmit_Power command with "Unknown HCI Command"
>>> status. Due to the issue, Bluetooth driver of 5.15 and later kernel
>>> fails to hci up.
>> 
>> I remember the LE Transmit power issue came up in 5.11 kernel, so if you are getting the issue starting
>> from 5.15, you probably want to bisect.
> Yes, I've just found below commit added sending LE_Read_Transmit_Power command, introduced in v5.11.
> https://github.com/torvalds/linux/commit/7c395ea521e6c8d77f643be61bf2f0f3a1f5b3e8

Yes that's the same commit which had caused issues for me. But was the device working from 5.11 to 5.14? If yes, what broke it in 5.15?

I'm sorry if I am missing any information, I was added by Paul in this thread, probably because I introduced this quirk in the kernel to fix Bluetooth on my Mac.
> 
> Best Regards,
> Nobuaki Tsunashima
> 
> -----Original Message-----
> From: Aditya Garg <gargaditya08@xxxxxxxx>
> Sent: Wednesday, May 22, 2024 3:59 PM
> To: Paul Menzel <pmenzel@xxxxxxxxxxxxx>
> Cc: Tsunashima Nobuaki (SMD C3 JP RM WLS AE) <Nobuaki.Tsunashima@xxxxxxxxxxxx>; Marcel Holtmann <marcel@xxxxxxxxxxxx>; Luiz Augusto von Dentz <luiz.dentz@xxxxxxxxx>; linux-bluetooth@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH v3] Bluetooth: Apply HCI_QUIRK_BROKEN_READ_TRANSMIT_POWER to CYW4373
> 
> Caution: This e-mail originated outside Infineon Technologies. Do not click on links or open attachments unless you validate it is safe<https://intranet-content.infineon.com/explore/aboutinfineon/rules/informationsecurity/ug/SocialEngineering/Pages/SocialEngineeringElements_en.aspx>.
> 
> 
> 
> Hi
> 
>> On 22 May 2024, at 10:13 AM, Paul Menzel <pmenzel@xxxxxxxxxxxxx> wrote:
>> 
>> Dear Nobuaki,
>> 
>> 
>> Thank you for your patch and addressing the comments. Please note, that the time on the system you sent the patch from is in the future:
>> 
>>   Date: Wed, 22 May 2024 17:17:35 +0900
>> 
>> But:
>> 
>>   Received: from smtp9.infineon.com (smtp9.infineon.com [217.10.52.204])
>>       (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits))
>>       (No client certificate requested)
>>       by smtp.subspace.kernel.org (Postfix) with ESMTPS id 83EC328EA;
>>       Wed, 22 May 2024 01:28:45 +0000 (UTC)
>> 
>>>> Am 22.05.24 um 10:17 schrieb Nobuaki Tsunashima:
>>> From: Nobuaki Tsunashima <Nobuaki.Tsunashima@xxxxxxxxxxxx>
>> 
>> I forgot to add btbcm in the summary:
>> 
>> Bluetooth: btbcm: …
>> 
>>> CYW4373 ROM FW has an issue that it claims LE_Read_Transmit_Power
>>> command as supported in a response of Read_Local_Supported_Command
>>> command but rejects the LE_Read_Transmit_Power command with "Unknown HCI Command"
>>> status. Due to the issue, Bluetooth driver of 5.15 and later kernel
>>> fails to hci up.
> 
> I remember the LE Transmit power issue came up in 5.11 kernel, so if you are getting the issue starting from 5.15, you probably want to bisect.
>> 
>> As written in the other thread, it’d be great if you bisected the commit.
>> 
>>> Especially in USB i/f case, it would be difficult to download patch
>>> FW that includes Its fix unless hci is up.
>> 
>> lowercase: its
>> 
>> Which firmware versions are fixed?
>> 
>>> The patch forces the driver to skip LE_Read_Transmit_Power Command
>>> when it detects CYW4373 with ROM FW build.
>> 
>> Maybe add something like:
>> 
>> The driver already contains infrastructure to apply the quirk, but currently it only supports DMI based matching. Add support to match by chip id and baseline, which ….
>> 
>>> Signed-off-by: Nobuaki Tsunashima <Nobuaki.Tsunashima@xxxxxxxxxxxx>
>>> ---
>>> V2 -> V3: Fix a few coding style warnings and change the subject as more specific.
>>> V1 -> V2: Fix several coding style warnings.
>>> drivers/bluetooth/btbcm.c | 32 +++++++++++++++++++++++++++++++-  
>>> drivers/bluetooth/btusb.c |  4 ++++
>>> 2 files changed, 35 insertions(+), 1 deletion(-) diff --git
>>> a/drivers/bluetooth/btbcm.c b/drivers/bluetooth/btbcm.c index
>>> 0a5445ac5e1b..c763e368d6ad 100644
>>> --- a/drivers/bluetooth/btbcm.c
>>> +++ b/drivers/bluetooth/btbcm.c
>>> @@ -437,18 +437,48 @@ static const struct dmi_system_id disable_broken_read_transmit_power[] = {
>>>     { }
>>> };
>>> +struct bcm_chip_version_table {
>>> +    u8    chip_id;
>> 
>> Please use one space. (Please also check the line below.)
>> 
>>> +    u16 baseline;
>> 
>> Add a comment above the struct, what baseline means?
>> 
>>> +};
>>> +#define BCM_ROMFW_BASELINE_NUM    0xFFFF
>>> +static const struct bcm_chip_version_table disable_broken_read_transmit_power_by_chip_ver[] = {
>>> +    {0x87, BCM_ROMFW_BASELINE_NUM}        /* CYW4373/4373E */
>> 
>> Add one space after { and before }?
>> 
> You may want to rename the existing variable btbcm_is_disable_broken_read_tx_power to btbcm_is_disable_broken_read_tx_power_by_dmi to avoid confusion. Although, I'm not a maintainer so consider it as just a suggestion.
>> 
>>> +};
>>> +static bool btbcm_is_disable_broken_read_tx_power_by_chip_ver(u8
>>> +chip_id, u16 baseline) {
>>> +    int i;
>>> +    int table_size =
>>> +ARRAY_SIZE(disable_broken_read_transmit_power_by_chip_ver);
>> 
>> Use size_t?
>> 
>>> +    const struct bcm_chip_version_table *entry =
>>> +                        
>>> + &disable_broken_read_transmit_power_by_chip_ver[0];
>>> +
>>> +    for (i = 0 ; i < table_size ; i++, entry++)    {
>>> +        if ((chip_id == entry->chip_id) && (baseline == entry->baseline))
>>> +            return true;
>>> +    }
>>> +
>>> +    return false;
>>> +}
>>> +
>>> static int btbcm_read_info(struct hci_dev *hdev)  {
>>>     struct sk_buff *skb;
>>> +    u8 chip_id;
>>> +    u16 baseline;
>>>       /* Read Verbose Config Version Info */
>>>     skb = btbcm_read_verbose_config(hdev);
>>>     if (IS_ERR(skb))
>>>         return PTR_ERR(skb);
>>> -
>>> +    chip_id = skb->data[1];
>>> +    baseline = skb->data[3] | (skb->data[4] << 8);
>>>     bt_dev_info(hdev, "BCM: chip id %u", skb->data[1]);
>>>     kfree_skb(skb);
>>> +    /* Check Chip ID and disable broken Read LE Min/Max Tx Power */
>>> +    if (btbcm_is_disable_broken_read_tx_power_by_chip_ver(chip_id, baseline))
>>> +        set_bit(HCI_QUIRK_BROKEN_READ_TRANSMIT_POWER,
>>> + &hdev->quirks);
>>> +
>> 
>> Commit 801b4c027b44 (Bluetooth: btbcm: disable read tx power for some Macs with the T2 Security chip) added the check in `btbcm_print_controller_features()`? No idea, where the best place is.
> 
> I added the check in `btbcm_print_controller_features()` because the the issue was not being fixed at other places. I remember compiling and testing it at various other places. I'm not really sure why it specifically works in `btbcm_print_controller_features()`
>> 
>> 
>>>     return 0;
>>> }
>>> diff --git a/drivers/bluetooth/btusb.c b/drivers/bluetooth/btusb.c
>>> index d31edad7a056..52561c8d8828 100644
>>> --- a/drivers/bluetooth/btusb.c
>>> +++ b/drivers/bluetooth/btusb.c
>>> @@ -142,6 +142,10 @@ static const struct usb_device_id btusb_table[] = {
>>>     { USB_VENDOR_AND_INTERFACE_INFO(0x04ca, 0xff, 0x01, 0x01),
>>>       .driver_info = BTUSB_BCM_PATCHRAM },
>>> +    /* Cypress devices with vendor specific id */
>>> +    { USB_VENDOR_AND_INTERFACE_INFO(0x04b4, 0xff, 0x01, 0x01),
>>> +      .driver_info = BTUSB_BCM_PATCHRAM },
>>> +
>> 
>> Order 0x04b4 before 0x04ca?
>> 
>>>     /* Broadcom devices with vendor specific id */
>>>     { USB_VENDOR_AND_INTERFACE_INFO(0x0a5c, 0xff, 0x01, 0x01),
>>>       .driver_info = BTUSB_BCM_PATCHRAM },
>> 
>> 
>> Kind regards,
>> 
>> Paul
> 
> Regards
> 
> Aditya