Re: [PATCH V2 5/6] rtlwifi: btcoexist: Add routines for RTL8812AE kernel socket communications

[Marcel Holtmann <marcel@holtmann.org>]

Hi Kalle,

> I'm adding bluetooth list to the discussion. Full patch is available
> here:
> 
> https://patchwork.kernel.org/patch/5712591/
> 
> Larry Finger <Larry.Finger@lwfinger.net> writes:
> 
>> From: Troy Tan <troy_tan@realsil.com.cn>
>> 
>> This patch adds the routines used to communicate between the RTL8812AE (wifi)
>> device and the RTL8761AU (bluetooth) device that are part of the same chip.
>> Unlike other similar dual-function devices, this chip does not contain special
>> hardware that lets the firmware pass coexistence info from one part to the
>> other. As a result, this driver implements such communication as a kernel
>> socket.
>> 
>> Signed-off-by: Troy Tan <troy_tan@realsil.com.cn>
>> Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
>> ---
>> V2 - Add comments explaining the routine that sends a message via the
>> socket.
> 
> The commit log is not still explaining that much about the actual
> functionality, so I investigated on my own:
> 
>> +static u8 rtl_btcoex_create_kernel_socket(struct rtl_priv *rtlpriv,
>> +					  u8 is_invite)
>> +{
>> +	struct bt_coex_info *pcoex_info = &rtlpriv->coex_info;
>> +	s8 kernel_socket_err;
>> +
>> +	BTC_PRINT(BTC_MSG_SOCKET, SOCKET_CRITICAL,
>> +		  "%s CONNECT_PORT %d\n", __func__, CONNECT_PORT);
>> +
>> +	if (!pcoex_info) {
>> +		BTC_PRINT(BTC_MSG_SOCKET, SOCKET_CRITICAL, "coex_info: NULL\n");
>> +		return _FAIL;
>> +	}
>> +
>> +	kernel_socket_err = sock_create(PF_INET, SOCK_DGRAM, 0,
>> +					&pcoex_info->udpsock);
>> +	BTC_PRINT(BTC_MSG_SOCKET, SOCKET_CRITICAL,
>> +		  "binding socket, err = %d\n", kernel_socket_err);
>> +
>> +	if (kernel_socket_err < 0) {
>> +		BTC_PRINT(BTC_MSG_SOCKET, SOCKET_CRITICAL,
>> +			  "Error during creation of socket error:%d\n",
>> +			  kernel_socket_err);
>> +		return _FAIL;
>> +	}
>> +	memset(&pcoex_info->sin, 0, sizeof(pcoex_info->sin));
>> +	pcoex_info->sin.sin_family = AF_INET;
>> +	pcoex_info->sin.sin_port = htons(CONNECT_PORT);
>> +	pcoex_info->sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
>> +
>> +	memset(&pcoex_info->bt_addr, 0, sizeof(pcoex_info->bt_addr));
>> +	pcoex_info->bt_addr.sin_family = AF_INET;
>> +	pcoex_info->bt_addr.sin_port = htons(CONNECT_PORT_BT);
>> +	pcoex_info->bt_addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
>> +
>> +	pcoex_info->sk_store = NULL;
>> +
>> +	kernel_socket_err =
>> +	   pcoex_info->udpsock->ops->bind(pcoex_info->udpsock,
>> +					  (struct sockaddr *)&pcoex_info->sin,
>> +					  sizeof(pcoex_info->sin));
>> +	if (kernel_socket_err == 0) {
>> +		BTC_PRINT(BTC_MSG_SOCKET, SOCKET_CRITICAL,
>> +			  "binding socket success\n");
>> +		pcoex_info->udpsock->sk->sk_data_ready =
>> +			rtl_btcoex_recvmsg_int;
>> +		pcoex_info->sock_open |=  KERNEL_SOCKET_OK;
>> +		pcoex_info->bt_attend = false;
>> +		BTC_PRINT(BTC_MSG_SOCKET, SOCKET_CRITICAL,
>> +			  "WIFI sending attend_req\n");
>> +		rtl_btcoex_sendmsgbysocket(rtlpriv, attend_req,
>> +					   sizeof(attend_req), true);
>> +		return _SUCCESS;
> 
> So the wireless driver communicates with the bluetooth driver (which is
> not in upstream) via a localhost UDP connection?

I think the first order of business should be to get the Bluetooth driver upstream. Until that has happened this is all kinda pointless discussion.

>> +#define CONNECT_PORT 30000
>> +#define CONNECT_PORT_BT 30001
> 
> And these are the UDP ports used.
> 
>> +struct hci_link_info {
>> +	u16		connect_handle;
>> +	u8		incoming_traffic_mode;
>> +	u8		outgoing_traffic_mode;
>> +	u8		bt_profile;
>> +	u8		bt_corespec;
>> +	s8		bt_RSSI;
>> +	u8		traffic_profile;
>> +	u8		link_role;
>> +};
>> +
>> +#define	MAX_BT_ACL_LINK_NUM		8
>> +
>> +struct hci_ext_config {
>> +	struct hci_link_info	acl_link[MAX_BT_ACL_LINK_NUM];
>> +	u8	bt_operation_code;
>> +	u16	current_connect_handle;
>> +	u8	current_incoming_traffic_mode;
>> +	u8	current_outgoing_traffic_mode;
>> +
>> +	u8	number_of_acl;
>> +	u8	number_of_sco;
>> +	u8	current_bt_status;
>> +	u16	hci_ext_ver;
>> +	bool	enable_wifi_scan_notify;
>> +};
> 
> [...]
> 
>> +enum HCI_STATUS {
>> +	/* Success */
>> +	HCI_STATUS_SUCCESS				= 0x00,
>> +	/* Unknown HCI Command */
>> +	HCI_STATUS_UNKNOW_HCI_CMD			= 0x01,
>> +	/* Unknown Connection Identifier */
>> +	HCI_STATUS_UNKNOW_CONNECT_ID			= 0X02,
>> +	/* Hardware Failure */
>> +	HCI_STATUS_HW_FAIL				= 0X03,
>> +	/* Page Timeout */
>> +	HCI_STATUS_PAGE_TIMEOUT				= 0X04,
>> +	/* Authentication Failure */
>> +	HCI_STATUS_AUTH_FAIL				= 0X05,
>> +	/* PIN or Key Missing */
>> +	HCI_STATUS_PIN_OR_KEY_MISSING			= 0X06,
>> +	/* Memory Capacity Exceeded */
>> +	HCI_STATUS_MEM_CAP_EXCEED			= 0X07,
>> +	/* Connection Timeout */
>> +	HCI_STATUS_CONNECT_TIMEOUT			= 0X08,
>> +	/* Connection Limit Exceeded */
> 
> And here's part of the information exchanged between the drivers.
> 
> This is something which needs to be properly reviewed both in wireless
> and bluetooth lists, a wireless driver cannot just invent these on their
> own. And using UDP sockets for this is in my opinion just horrible.
> 
> I know there's a general need for something similar like this, but it
> needs to properly discussed and designed.

This is just insane. Clear NAK.

Two kernel modules will not use UDP ports over the loopback interface to communicate with each other.

Regards

Marcel