Hi Sean,
>>>>> +
>>>>> +static int mtk_wmt_cmd_sync(struct hci_uart *hu, u8 opcode, u8 flag, u16
>>>>> plen,
>>>>> + const void *param)
>>>>> +{
>>>>> + struct mtk_bt_dev *btdev = hu->priv;
>>>>> + struct hci_command_hdr *hhdr;
>>>>> + struct hci_acl_hdr *ahdr;
>>>>> + struct mtk_wmt_hdr *whdr;
>>>>> + struct sk_buff *skb;
>>>>> + int ret = 0;
>>>>> +
>>>>> + init_completion(&btdev->wmt_cmd);
>>>>> +
>>>>> + skb = bt_skb_alloc(plen + MTK_WMT_CMD_SIZE, GFP_KERNEL);
>>>>> + if (!skb)
>>>>> + return -ENOMEM;
>>>>> +
>>>>> + /*
>>>>> + * WMT data is carried in either ACL or HCI format with op code as
>>>>> + * 0xfc6f and followed by a WMT header and its actual payload.
>>>>> + */
>>>>
>>>> Please use net subsystem comment style.
>>>>
>>>>> + switch (opcode) {
>>>>> + case MTK_WMT_PATCH_DWNLD:
>>>>> + ahdr = skb_put(skb, HCI_ACL_HDR_SIZE);
>>>>> + ahdr->handle = cpu_to_le16(0xfc6f);
>>>>> + ahdr->dlen = cpu_to_le16(plen + MTK_WMT_HDR_SIZE);
>>>>> + break;
>>>>> + default:
>>>>> + hhdr = skb_put(skb, HCI_COMMAND_HDR_SIZE);
>>>>> + hhdr->opcode = cpu_to_le16(0xfc6f);
>>>>> + hhdr->plen = plen + MTK_WMT_HDR_SIZE;
>>>>> + break;
>>>>> + }
>>>>> +
>>>>> + hci_skb_pkt_type(skb) = opcode == MTK_WMT_PATCH_DWNLD ?
>>>>> + HCI_ACLDATA_PKT : HCI_COMMAND_PKT;
>>>>
>>>> Why not move that into the switch statement above.
>>>>
>>>>> +
>>>>> + /* Start to build a WMT header and its actual payload. */
>>>>> + whdr = skb_put(skb, MTK_WMT_HDR_SIZE);
>>>>> + whdr->dir = 1;
>>>>> + whdr->op = opcode;
>>>>> + whdr->dlen = cpu_to_le16(plen + 1);
>>>>> + whdr->flag = flag;
>>>>> + skb_put_data(skb, param, plen);
>>>>> +
>>>>> + mtk_enqueue(hu, skb);
>>>>> + hci_uart_tx_wakeup(hu);
>>>>> +
>>>>> + /*
>>>>> + * Waiting a WMT event response, while we must take care in case of
>>>>> + * failures for the wait.
>>>>> + */
>>>>> + ret = wait_for_completion_interruptible_timeout(&btdev->wmt_cmd, HZ);
>>>>> +
>>>>> + return ret > 0 ? 0 : ret < 0 ? ret : -ETIMEDOUT;
>>>>> +}
>>>>
>>>> All in all I am not convinced that this is super clean. I get that we need
>>>> something special for having this in the ACL data packets, but for the
>>>> standard HCI command I prefer that __hci_cmd_sync is used. I addition, it
>>>> seems that patch download is the only special case and that happens before
>>>> at the setup stage. So we could make things special for that. I need to
>>>> understand this a bit better. Can I get a btmon -w trace.log file from the
>>>> whole init procedure.
>>>>
>>>
>>> While i was trying to rewrite the driver based on btuart.c. you posted
>>> on RFC, I used __hci_cmd_sync_ev to replace such kinds of SoC specific
>>> hci command sending which I've done previously with mtk_wmt_cmd_sync.
>>>
>>> However, eventually, I got a cmd_timer timeout whose message printed
>>> on console as "Bluetooth: hci0: command 0xfc6f tx timeout".
>>>
>>> The mtk soc specific cmd/event I posted below, I dumped directly in
>>> driver, always uses cmd as opcode 0xfc6f, and its event id as 0xe4.
>>>
>>> It appears to the event id is not standard and thus it cannot cancel the
>>> cmd timer when the special hci event is being handled. This way can we
>>> can still use __hci_cmd_sync api ?
>>>
>>> [ 4.896200] hci tx: 00000000: 01 6f fc 05 01 07 01 00 04
>>> [ 4.904671] hci rx: 00000000: e4 05 02 07 01 00
>>> 00
>>> [ 4.912859] Bluetooth: hci0 event 0xe4
>>>
>>>
>>> buildroot login: [ 6.914509] Bluetooth: hci0: command 0xfc6f tx
>>> timeout
>>> [ 6.919831] hci tx: 00000000: 01 6f fc 06 01 06 02 00 00
>>> 01 .o........
>>> [ 7.006631] hci rx: 00000000: e4 05 02 06 01 00
>>> 00 .......
>>> [ 7.014821] Bluetooth: hci0 event 0xe4
>>
>> can you just start btmon before loading the module / driver? It makes it a
>> lot easier since it will actually decode the basics for us. If there is a
>> bug within __hci_cmd_sync_ev, then we are going to fix it.
>>
>
> I'm happy to do with btmon. just the environment with buildroot the BT
> running on seems there's a missing support for btmon. I can start to use
> btmon once I change the environment to Debian.
>
>> So all the MTK vendor commands respond with a vendor event? Or are there
>> some that do the standard command status/complete handling?
>>
>
> yes, mtk controller after mt7622 (included), its MTK vendors command
> (opcode 0xfc6f) always respond with a vendor event id 0xe4. And they
> don't do any standard status/complete handling.
then we need to figure out where the __hci_cmd_sync_ev causes a problem. Since
normally that should just work for you.
> BTW, mtk controller before mt7622, such as mt7623, its MTK vendor
> command always go with completely specific format, not with hci format.
What does that mean? Do you have an example?
Regards
Marcel
