Skip to contentSkip to conversation

Bluetooth: Add timeout in disconnect when power off

PATCH In review Howard Chung 5 participants 9 messages Jul 24, 2023 – Jul 26, 2023 1 message not in the archive

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Bluetooth: Add timeout in disconnect when power off

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when it is asked to reset itself. This could happen on powering off adapter.

One possible issue is that if a connection exists, and then powering off and on adapter within a short time, then our host stack assumes the conntection was disconnected but this might not be true, so if we issue a connection to the peer, it will fail with ACL Already Connected error.

This CL makes the host stack to wait for |HCI_EV_DISCONN_COMPLETE| when powering off with a configurable timeout unless the timeout is set to 0.

---
Hi upstream maintainers, this is tested with an AX211 device and Logi K580 keyboard by the following procedures:
1. pair the peer and stay connected.
2. power off and on immediately
3. observe that the btsnoop log is consistent with the configured timeout.

 include/net/bluetooth/hci_core.h |  1 +
 net/bluetooth/hci_core.c         |  2 +-
 net/bluetooth/hci_sync.c         | 38 +++++++++++++++++++++++---------
 net/bluetooth/mgmt_config.c      |  6 +++++
 4 files changed, 35 insertions(+), 12 deletions(-)
include/net/bluetooth/hci_core.h +1 -0
index 8200a6689b39..ce44f9c60059 100644
@@ -432,6 +432,7 @@ struct hci_dev {
 	__u16		advmon_allowlist_duration;
 	__u16		advmon_no_filter_duration;
 	__u8		enable_advmon_interleave_scan;
+	__u16		discon_on_poweroff_timeout;
 
 	__u16		devid_source;
 	__u16		devid_vendor;
net/bluetooth/hci_core.c +1 -1
index 0fefa6788911..769865494f45 100644
@@ -2479,7 +2479,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
 	hdev->adv_instance_cnt = 0;
 	hdev->cur_adv_instance = 0x00;
 	hdev->adv_instance_timeout = 0;
-
+	hdev->discon_on_poweroff_timeout = 0;	/* Default to no timeout */
 	hdev->advmon_allowlist_duration = 300;
 	hdev->advmon_no_filter_duration = 500;
 	hdev->enable_advmon_interleave_scan = 0x00;	/* Default to disable */
net/bluetooth/hci_sync.c +27 -11
index 3348a1b0e3f7..260e9f05359c 100644
@@ -5250,6 +5250,8 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
 			       u8 reason)
 {
 	struct hci_cp_disconnect cp;
+	unsigned long timeout;
+	int err;
 
 	if (conn->type == AMP_LINK)
 		return hci_disconnect_phy_link_sync(hdev, conn->handle, reason);
@@ -5258,19 +5260,33 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
 	cp.handle = cpu_to_le16(conn->handle);
 	cp.reason = reason;
 
-	/* Wait for HCI_EV_DISCONN_COMPLETE, not HCI_EV_CMD_STATUS, when the
-	 * reason is anything but HCI_ERROR_REMOTE_POWER_OFF. This reason is
-	 * used when suspending or powering off, where we don't want to wait
-	 * for the peer's response.
+	/* The HCI_ERROR_REMOTE_POWER_OFF is used when suspending or powering off,
+	 * so we don't want to waste time waiting for the reply of the peer.
+	 * However, if the configuration specified, we'll wait some time to give the
+	 * controller chance to actually send the disconnect command.
 	 */
-	if (reason != HCI_ERROR_REMOTE_POWER_OFF)
-		return __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
-						sizeof(cp), &cp,
-						HCI_EV_DISCONN_COMPLETE,
-						HCI_CMD_TIMEOUT, NULL);
+	if (reason == HCI_ERROR_REMOTE_POWER_OFF && !hdev->discon_on_poweroff_timeout) {
+		return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT,
+					     sizeof(cp), &cp, HCI_CMD_TIMEOUT);
+	}
 
-	return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT, sizeof(cp), &cp,
-				     HCI_CMD_TIMEOUT);
+	if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+		timeout = msecs_to_jiffies(hdev->discon_on_poweroff_timeout);
+	else
+		timeout = HCI_CMD_TIMEOUT;
+
+	err = __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
+				       sizeof(cp), &cp,
+				       HCI_EV_DISCONN_COMPLETE,
+				       timeout, NULL);
+
+	/* Ignore the error in suspending or powering off case to avoid the procedure being
+	 * aborted.
+	 */
+	if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+		return 0;
+
+	return err;
 }
 
 static int hci_le_connect_cancel_sync(struct hci_dev *hdev,
net/bluetooth/mgmt_config.c +6 -0
index 6ef701c27da4..f3194e3642d9 100644
@@ -78,6 +78,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
 		HDEV_PARAM_U16(advmon_allowlist_duration);
 		HDEV_PARAM_U16(advmon_no_filter_duration);
 		HDEV_PARAM_U8(enable_advmon_interleave_scan);
+		HDEV_PARAM_U16(discon_on_poweroff_timeout);
 	} __packed rp = {
 		TLV_SET_U16(0x0000, def_page_scan_type),
 		TLV_SET_U16(0x0001, def_page_scan_int),
@@ -111,6 +112,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
 		TLV_SET_U16(0x001d, advmon_allowlist_duration),
 		TLV_SET_U16(0x001e, advmon_no_filter_duration),
 		TLV_SET_U8(0x001f, enable_advmon_interleave_scan),
+		TLV_SET_U16(0x0020, discon_on_poweroff_timeout),
 	};
 
 	bt_dev_dbg(hdev, "sock %p", sk);
@@ -186,6 +188,7 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
 		case 0x001b:
 		case 0x001d:
 		case 0x001e:
+		case 0x0020:
 			exp_type_len = sizeof(u16);
 			break;
 		case 0x001f:
@@ -314,6 +317,9 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
 		case 0x0001f:
 			hdev->enable_advmon_interleave_scan = TLV_GET_U8(buffer);
 			break;
+		case 0x00020:
+			hdev->discon_on_poweroff_timeout = TLV_GET_LE16(buffer);
+			break;
 		default:
 			bt_dev_warn(hdev, "unsupported parameter %u", type);
 			break;
Signature · 1 line
2.41.0.487.g6d72f3e995-goog
#1 lore raw

This is automated email and please do not reply to this email!

Dear submitter,

Thank you for submitting the patches to the linux bluetooth mailing list. This is a CI test results with your patch series:
PW Link:https://patchwork.kernel.org/project/bluetooth/list/?series=768858

---Test result---

Test Summary:

CheckPatch                    PASS      1.32 seconds
GitLint                       PASS      0.30 seconds
SubjectPrefix                 PASS      0.09 seconds
BuildKernel                   PASS      40.15 seconds
CheckAllWarning               PASS      44.04 seconds
CheckSparse                   PASS      49.87 seconds
CheckSmatch                   PASS      134.79 seconds
BuildKernel32                 PASS      38.62 seconds
TestRunnerSetup               PASS      584.62 seconds
TestRunner_l2cap-tester       PASS      28.42 seconds
TestRunner_iso-tester         PASS      56.98 seconds
TestRunner_bnep-tester        PASS      12.70 seconds
TestRunner_mgmt-tester        PASS      247.32 seconds
TestRunner_rfcomm-tester      PASS      19.68 seconds
TestRunner_sco-tester         PASS      19.99 seconds
TestRunner_ioctl-tester       PASS      21.88 seconds
TestRunner_mesh-tester        PASS      16.37 seconds
TestRunner_smp-tester         PASS      17.27 seconds
TestRunner_userchan-tester    PASS      13.62 seconds
IncrementalBuild              PASS      36.82 seconds

---
Regards,
Linux Bluetooth

#2 lore raw
In reply to Howard Chung

Dear Howard,

Thank you for your patch. Some minor nits.

Am 24.07.23 um 13:12 schrieb Howard Chung:

Quoted 3 lines

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when

1. s/been/be/ or *have been*?
2. controller

Quoted 4 lines

it is asked to reset itself. This could happen on powering off adapter.

One possible issue is that if a connection exists, and then powering off and on adapter within a short time, then our host stack assumes the

I do not understand the part of the first comma.

Quoted 1 line

conntection was disconnected but this might not be true, so if we issue

connection

Quoted 7 lines

a connection to the peer, it will fail with ACL Already Connected error.

This CL makes the host stack to wait for |HCI_EV_DISCONN_COMPLETE| when powering off with a configurable timeout unless the timeout is set to 0.

Quoted 7 lines

---
Hi upstream maintainers, this is tested with an AX211 device and Logi K580 keyboard by the following procedures:
1. pair the peer and stay connected.
2. power off and on immediately
3. observe that the btsnoop log is consistent with the configured timeout.

It’d be great to have this in the commit message.

Quoted 11 lines
  include/net/bluetooth/hci_core.h |  1 +
  net/bluetooth/hci_core.c         |  2 +-
  net/bluetooth/hci_sync.c         | 38 +++++++++++++++++++++++---------
  net/bluetooth/mgmt_config.c      |  6 +++++
  4 files changed, 35 insertions(+), 12 deletions(-)
include/net/bluetooth/hci_core.h +0 -0
index 8200a6689b39..ce44f9c60059 100644

include/net/bluetooth/hci_core.h struct hci_dev { · line 435

Quoted 4 lines
include/net/bluetooth/hci_core.h +1 -0
@@ -432,6 +432,7 @@ struct hci_dev {
  	__u16		advmon_allowlist_duration;
  	__u16		advmon_no_filter_duration;
  	__u8		enable_advmon_interleave_scan;
+	__u16		discon_on_poweroff_timeout;

I’d append the unit to the variable name: discon_on_poweroff_timeout_ms.

Quoted 21 lines
include/net/bluetooth/hci_core.h +0 -0
  	__u16		devid_source;
  	__u16		devid_vendor;
net/bluetooth/hci_core.c +1 -1
index 0fefa6788911..769865494f45 100644
@@ -2479,7 +2479,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
  	hdev->adv_instance_cnt = 0;
  	hdev->cur_adv_instance = 0x00;
  	hdev->adv_instance_timeout = 0;
-
+	hdev->discon_on_poweroff_timeout = 0;	/* Default to no timeout */
  	hdev->advmon_allowlist_duration = 300;
  	hdev->advmon_no_filter_duration = 500;
  	hdev->enable_advmon_interleave_scan = 0x00;	/* Default to disable */
net/bluetooth/hci_sync.c +0 -0
index 3348a1b0e3f7..260e9f05359c 100644

net/bluetooth/hci_sync.c static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn, · line 5253

Quoted 4 lines
net/bluetooth/hci_sync.c +1 -0
@@ -5250,6 +5250,8 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
  			       u8 reason)
  {
  	struct hci_cp_disconnect cp;
+	unsigned long timeout;

Ditto.

Quoted 5 lines
net/bluetooth/hci_sync.c +1 -0
+	int err;
 
  	if (conn->type == AMP_LINK)
  		return hci_disconnect_phy_link_sync(hdev, conn->handle, reason);

net/bluetooth/hci_sync.c static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn, · line 5265

Quoted 10 lines
net/bluetooth/hci_sync.c +3 -4
@@ -5258,19 +5260,33 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
  	cp.handle = cpu_to_le16(conn->handle);
  	cp.reason = reason;
 
-	/* Wait for HCI_EV_DISCONN_COMPLETE, not HCI_EV_CMD_STATUS, when the
-	 * reason is anything but HCI_ERROR_REMOTE_POWER_OFF. This reason is
-	 * used when suspending or powering off, where we don't want to wait
-	 * for the peer's response.
+	/* The HCI_ERROR_REMOTE_POWER_OFF is used when suspending or powering off,
+	 * so we don't want to waste time waiting for the reply of the peer.
+	 * However, if the configuration specified, we'll wait some time to give the

“if the configuration specified” sounds strange to me.

Kind regards,

Paul

Quoted 72 lines
net/bluetooth/hci_sync.c +22 -7
+	 * controller chance to actually send the disconnect command.
  	 */
-	if (reason != HCI_ERROR_REMOTE_POWER_OFF)
-		return __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
-						sizeof(cp), &cp,
-						HCI_EV_DISCONN_COMPLETE,
-						HCI_CMD_TIMEOUT, NULL);
+	if (reason == HCI_ERROR_REMOTE_POWER_OFF && !hdev->discon_on_poweroff_timeout) {
+		return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT,
+					     sizeof(cp), &cp, HCI_CMD_TIMEOUT);
+	}
 
-	return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT, sizeof(cp), &cp,
-				     HCI_CMD_TIMEOUT);
+	if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+		timeout = msecs_to_jiffies(hdev->discon_on_poweroff_timeout);
+	else
+		timeout = HCI_CMD_TIMEOUT;
+
+	err = __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
+				       sizeof(cp), &cp,
+				       HCI_EV_DISCONN_COMPLETE,
+				       timeout, NULL);
+
+	/* Ignore the error in suspending or powering off case to avoid the procedure being
+	 * aborted.
+	 */
+	if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+		return 0;
+
+	return err;
  }
 
  static int hci_le_connect_cancel_sync(struct hci_dev *hdev,
net/bluetooth/mgmt_config.c +6 -0
index 6ef701c27da4..f3194e3642d9 100644
@@ -78,6 +78,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
  		HDEV_PARAM_U16(advmon_allowlist_duration);
  		HDEV_PARAM_U16(advmon_no_filter_duration);
  		HDEV_PARAM_U8(enable_advmon_interleave_scan);
+		HDEV_PARAM_U16(discon_on_poweroff_timeout);
  	} __packed rp = {
  		TLV_SET_U16(0x0000, def_page_scan_type),
  		TLV_SET_U16(0x0001, def_page_scan_int),
@@ -111,6 +112,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
  		TLV_SET_U16(0x001d, advmon_allowlist_duration),
  		TLV_SET_U16(0x001e, advmon_no_filter_duration),
  		TLV_SET_U8(0x001f, enable_advmon_interleave_scan),
+		TLV_SET_U16(0x0020, discon_on_poweroff_timeout),
  	};
 
  	bt_dev_dbg(hdev, "sock %p", sk);
@@ -186,6 +188,7 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
  		case 0x001b:
  		case 0x001d:
  		case 0x001e:
+		case 0x0020:
  			exp_type_len = sizeof(u16);
  			break;
  		case 0x001f:
@@ -314,6 +317,9 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
  		case 0x0001f:
  			hdev->enable_advmon_interleave_scan = TLV_GET_U8(buffer);
  			break;
+		case 0x00020:
+			hdev->discon_on_poweroff_timeout = TLV_GET_LE16(buffer);
+			break;
  		default:
  			bt_dev_warn(hdev, "unsupported parameter %u", type);
  			break;
#3 lore raw
In reply to Howard Chung

Hi Howard,

Quoted 10 lines · Howard Chung

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when it is asked to reset itself. This could happen on powering off adapter.

One possible issue is that if a connection exists, and then powering off and on adapter within a short time, then our host stack assumes the conntection was disconnected but this might not be true, so if we issue a connection to the peer, it will fail with ACL Already Connected error.

That sounds more like a bug in the controller though, the spec is quite clear that it must reset the link-layer state:

BLUETOOTH CORE SPECIFICATION Version 5.3 | Vol 4, Part E page 1972

...HCI_Reset command shall reset the Link Manager, Baseband and Link Layer.

So it sounds like the controller shall perform and the necessary procedures before it respond with a Command Complete.

Quoted 143 lines

This CL makes the host stack to wait for |HCI_EV_DISCONN_COMPLETE| when powering off with a configurable timeout unless the timeout is set to 0.

---
Hi upstream maintainers, this is tested with an AX211 device and Logi K580 keyboard by the following procedures:
1. pair the peer and stay connected.
2. power off and on immediately
3. observe that the btsnoop log is consistent with the configured timeout.

 include/net/bluetooth/hci_core.h |  1 +
 net/bluetooth/hci_core.c         |  2 +-
 net/bluetooth/hci_sync.c         | 38 +++++++++++++++++++++++---------
 net/bluetooth/mgmt_config.c      |  6 +++++
 4 files changed, 35 insertions(+), 12 deletions(-)
include/net/bluetooth/hci_core.h +1 -0
index 8200a6689b39..ce44f9c60059 100644
@@ -432,6 +432,7 @@ struct hci_dev {
        __u16           advmon_allowlist_duration;
        __u16           advmon_no_filter_duration;
        __u8            enable_advmon_interleave_scan;
+       __u16           discon_on_poweroff_timeout;
 
        __u16           devid_source;
        __u16           devid_vendor;
net/bluetooth/hci_core.c +1 -1
index 0fefa6788911..769865494f45 100644
@@ -2479,7 +2479,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
        hdev->adv_instance_cnt = 0;
        hdev->cur_adv_instance = 0x00;
        hdev->adv_instance_timeout = 0;
-
+       hdev->discon_on_poweroff_timeout = 0;   /* Default to no timeout */
        hdev->advmon_allowlist_duration = 300;
        hdev->advmon_no_filter_duration = 500;
        hdev->enable_advmon_interleave_scan = 0x00;     /* Default to disable */
net/bluetooth/hci_sync.c +27 -11
index 3348a1b0e3f7..260e9f05359c 100644
@@ -5250,6 +5250,8 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
                               u8 reason)
 {
        struct hci_cp_disconnect cp;
+       unsigned long timeout;
+       int err;
 
        if (conn->type == AMP_LINK)
                return hci_disconnect_phy_link_sync(hdev, conn->handle, reason);
@@ -5258,19 +5260,33 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
        cp.handle = cpu_to_le16(conn->handle);
        cp.reason = reason;
 
-       /* Wait for HCI_EV_DISCONN_COMPLETE, not HCI_EV_CMD_STATUS, when the
-        * reason is anything but HCI_ERROR_REMOTE_POWER_OFF. This reason is
-        * used when suspending or powering off, where we don't want to wait
-        * for the peer's response.
+       /* The HCI_ERROR_REMOTE_POWER_OFF is used when suspending or powering off,
+        * so we don't want to waste time waiting for the reply of the peer.
+        * However, if the configuration specified, we'll wait some time to give the
+        * controller chance to actually send the disconnect command.
         */
-       if (reason != HCI_ERROR_REMOTE_POWER_OFF)
-               return __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
-                                               sizeof(cp), &cp,
-                                               HCI_EV_DISCONN_COMPLETE,
-                                               HCI_CMD_TIMEOUT, NULL);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF && !hdev->discon_on_poweroff_timeout) {
+               return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT,
+                                            sizeof(cp), &cp, HCI_CMD_TIMEOUT);
+       }
 
-       return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT, sizeof(cp), &cp,
-                                    HCI_CMD_TIMEOUT);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               timeout = msecs_to_jiffies(hdev->discon_on_poweroff_timeout);
+       else
+               timeout = HCI_CMD_TIMEOUT;
+
+       err = __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
+                                      sizeof(cp), &cp,
+                                      HCI_EV_DISCONN_COMPLETE,
+                                      timeout, NULL);
+
+       /* Ignore the error in suspending or powering off case to avoid the procedure being
+        * aborted.
+        */
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               return 0;
+
+       return err;
 }
 
 static int hci_le_connect_cancel_sync(struct hci_dev *hdev,
net/bluetooth/mgmt_config.c +6 -1
index 6ef701c27da4..f3194e3642d9 100644
@@ -78,6 +78,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                HDEV_PARAM_U16(advmon_allowlist_duration);
                HDEV_PARAM_U16(advmon_no_filter_duration);
                HDEV_PARAM_U8(enable_advmon_interleave_scan);
+               HDEV_PARAM_U16(discon_on_poweroff_timeout);
        } __packed rp = {
                TLV_SET_U16(0x0000, def_page_scan_type),
                TLV_SET_U16(0x0001, def_page_scan_int),
@@ -111,6 +112,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                TLV_SET_U16(0x001d, advmon_allowlist_duration),
                TLV_SET_U16(0x001e, advmon_no_filter_duration),
                TLV_SET_U8(0x001f, enable_advmon_interleave_scan),
+               TLV_SET_U16(0x0020, discon_on_poweroff_timeout),
        };
 
        bt_dev_dbg(hdev, "sock %p", sk);
@@ -186,6 +188,7 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x001b:
                case 0x001d:
                case 0x001e:
+               case 0x0020:
                        exp_type_len = sizeof(u16);
                        break;
                case 0x001f:
@@ -314,6 +317,9 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x0001f:
                        hdev->enable_advmon_interleave_scan = TLV_GET_U8(buffer);
                        break;
+               case 0x00020:
+                       hdev->discon_on_poweroff_timeout = TLV_GET_LE16(buffer);
+                       break;
                default:
                        bt_dev_warn(hdev, "unsupported parameter %u", type);
                        break;
--

2.41.0.487.g6d72f3e995-goog

Signature · 1 line
Luiz Augusto von Dentz
#4 lore raw

Thanks Paul for the review. I'll fix them in the next patch.

Hi Luiz,

I think the controller did reset its internal state in the link layer, so the next time we asked the controller to create a connection to the same remote, the controller didn't reply the error in the command complete event but the connection was rejected by the remote. If we really want the controller to complete the disconnect procedure on reset, the power-off procedure could be stuck for tens of seconds if the remote was not responding. A similar problem was found and solved in this patch [Bluetooth: hci_sync: Don't wait peer's reply when powering off].

The other reason why we think this is better to be solved in the software layer is this is not only observed in Intel's AX211 but also MTK's MT7921 and maybe more.

Since this is a tradeoff problem, I make it a configurable parameter so that each user can decide a good number from their perspective. If it is 0, which is the default value, the behavior should remain the same.

Do you think this sounds good to you?

Thanks,
Howard

Quoted 173 lines · Luiz Augusto von Dentz

Hi Howard,

Quoted 10 lines · Howard Chung

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when it is asked to reset itself. This could happen on powering off adapter.

One possible issue is that if a connection exists, and then powering off and on adapter within a short time, then our host stack assumes the conntection was disconnected but this might not be true, so if we issue a connection to the peer, it will fail with ACL Already Connected error.

That sounds more like a bug in the controller though, the spec is quite clear that it must reset the link-layer state:

BLUETOOTH CORE SPECIFICATION Version 5.3 | Vol 4, Part E page 1972

...HCI_Reset command shall reset the Link Manager, Baseband and Link Layer.

So it sounds like the controller shall perform and the necessary procedures before it respond with a Command Complete.

Quoted 141 lines

This CL makes the host stack to wait for |HCI_EV_DISCONN_COMPLETE| when powering off with a configurable timeout unless the timeout is set to 0.

---
Hi upstream maintainers, this is tested with an AX211 device and Logi K580 keyboard by the following procedures:
1. pair the peer and stay connected.
2. power off and on immediately
3. observe that the btsnoop log is consistent with the configured timeout.

 include/net/bluetooth/hci_core.h |  1 +
 net/bluetooth/hci_core.c         |  2 +-
 net/bluetooth/hci_sync.c         | 38 +++++++++++++++++++++++---------
 net/bluetooth/mgmt_config.c      |  6 +++++
 4 files changed, 35 insertions(+), 12 deletions(-)
include/net/bluetooth/hci_core.h +1 -0
index 8200a6689b39..ce44f9c60059 100644
@@ -432,6 +432,7 @@ struct hci_dev {
        __u16           advmon_allowlist_duration;
        __u16           advmon_no_filter_duration;
        __u8            enable_advmon_interleave_scan;
+       __u16           discon_on_poweroff_timeout;
 
        __u16           devid_source;
        __u16           devid_vendor;
net/bluetooth/hci_core.c +1 -1
index 0fefa6788911..769865494f45 100644
@@ -2479,7 +2479,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
        hdev->adv_instance_cnt = 0;
        hdev->cur_adv_instance = 0x00;
        hdev->adv_instance_timeout = 0;
-
+       hdev->discon_on_poweroff_timeout = 0;   /* Default to no timeout */
        hdev->advmon_allowlist_duration = 300;
        hdev->advmon_no_filter_duration = 500;
        hdev->enable_advmon_interleave_scan = 0x00;     /* Default to disable */
net/bluetooth/hci_sync.c +27 -11
index 3348a1b0e3f7..260e9f05359c 100644
@@ -5250,6 +5250,8 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
                               u8 reason)
 {
        struct hci_cp_disconnect cp;
+       unsigned long timeout;
+       int err;
 
        if (conn->type == AMP_LINK)
                return hci_disconnect_phy_link_sync(hdev, conn->handle, reason);
@@ -5258,19 +5260,33 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
        cp.handle = cpu_to_le16(conn->handle);
        cp.reason = reason;
 
-       /* Wait for HCI_EV_DISCONN_COMPLETE, not HCI_EV_CMD_STATUS, when the
-        * reason is anything but HCI_ERROR_REMOTE_POWER_OFF. This reason is
-        * used when suspending or powering off, where we don't want to wait
-        * for the peer's response.
+       /* The HCI_ERROR_REMOTE_POWER_OFF is used when suspending or powering off,
+        * so we don't want to waste time waiting for the reply of the peer.
+        * However, if the configuration specified, we'll wait some time to give the
+        * controller chance to actually send the disconnect command.
         */
-       if (reason != HCI_ERROR_REMOTE_POWER_OFF)
-               return __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
-                                               sizeof(cp), &cp,
-                                               HCI_EV_DISCONN_COMPLETE,
-                                               HCI_CMD_TIMEOUT, NULL);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF && !hdev->discon_on_poweroff_timeout) {
+               return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT,
+                                            sizeof(cp), &cp, HCI_CMD_TIMEOUT);
+       }
 
-       return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT, sizeof(cp), &cp,
-                                    HCI_CMD_TIMEOUT);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               timeout = msecs_to_jiffies(hdev->discon_on_poweroff_timeout);
+       else
+               timeout = HCI_CMD_TIMEOUT;
+
+       err = __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
+                                      sizeof(cp), &cp,
+                                      HCI_EV_DISCONN_COMPLETE,
+                                      timeout, NULL);
+
+       /* Ignore the error in suspending or powering off case to avoid the procedure being
+        * aborted.
+        */
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               return 0;
+
+       return err;
 }
 
 static int hci_le_connect_cancel_sync(struct hci_dev *hdev,
net/bluetooth/mgmt_config.c +6 -1
index 6ef701c27da4..f3194e3642d9 100644
@@ -78,6 +78,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                HDEV_PARAM_U16(advmon_allowlist_duration);
                HDEV_PARAM_U16(advmon_no_filter_duration);
                HDEV_PARAM_U8(enable_advmon_interleave_scan);
+               HDEV_PARAM_U16(discon_on_poweroff_timeout);
        } __packed rp = {
                TLV_SET_U16(0x0000, def_page_scan_type),
                TLV_SET_U16(0x0001, def_page_scan_int),
@@ -111,6 +112,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                TLV_SET_U16(0x001d, advmon_allowlist_duration),
                TLV_SET_U16(0x001e, advmon_no_filter_duration),
                TLV_SET_U8(0x001f, enable_advmon_interleave_scan),
+               TLV_SET_U16(0x0020, discon_on_poweroff_timeout),
        };
 
        bt_dev_dbg(hdev, "sock %p", sk);
@@ -186,6 +188,7 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x001b:
                case 0x001d:
                case 0x001e:
+               case 0x0020:
                        exp_type_len = sizeof(u16);
                        break;
                case 0x001f:
@@ -314,6 +317,9 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x0001f:
                        hdev->enable_advmon_interleave_scan = TLV_GET_U8(buffer);
                        break;
+               case 0x00020:
+                       hdev->discon_on_poweroff_timeout = TLV_GET_LE16(buffer);
+                       break;
                default:
                        bt_dev_warn(hdev, "unsupported parameter %u", type);
                        break;
--
Quoted 2 lines

2.41.0.487.g6d72f3e995-goog

--
Luiz Augusto von Dentz

A message that is not in the archive (sent off-list or not yet received): CAPHZWUejzqz5=tHPwib6gfSjcd6K90dyfw8m_3CgaLqjdYRa3A@mail.gmail.com

#6 lore raw

Hi Yun-hao,

Quoted 8 lines · Yun-hao Chung

Thanks Paul for the review. I'll fix them in the next patch.

Hi Luiz,

I think the controller did reset its internal state in the link layer, so the next time we asked the controller to create a connection to the same remote, the controller didn't reply the error in the command complete event but the connection was rejected by the remote. If we really want the controller to complete the disconnect procedure on reset, the power-off procedure could be stuck for tens of seconds if the remote was not responding. A similar problem was found and solved in this patch [Bluetooth: hci_sync: Don't wait peer's reply when powering off].

The other reason why we think this is better to be solved in the software layer is this is not only observed in Intel's AX211 but also MTK's MT7921 and maybe more.

Interesting, but in this case shouldn't the link-layer attempt to cleanup _before_ it reset the state? Because as far I recall doing a HCI_Reset shall cause a Disconnection even if not initiated by the host stack with HCI_Disconnect, also there is something really odd with the fact that the link-layer is generating such an error, because in case that happens isn't it clear the states are out of sync and it shall disconnect or something, I'm also curious what would happen if one quickly toggle power then the same problem should manifest, at least I don't see a difference between HCI_Reset and power off/on within the connection supervision timeout.

Quoted 186 lines

Since this is a tradeoff problem, I make it a configurable parameter so that each user can decide a good number from their perspective. If it is 0, which is the default value, the behavior should remain the same.

Do you think this sounds good to you?

Thanks,
Howard

Quoted 173 lines · Luiz Augusto von Dentz

Hi Howard,

Quoted 10 lines · Howard Chung

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when it is asked to reset itself. This could happen on powering off adapter.

One possible issue is that if a connection exists, and then powering off and on adapter within a short time, then our host stack assumes the conntection was disconnected but this might not be true, so if we issue a connection to the peer, it will fail with ACL Already Connected error.

That sounds more like a bug in the controller though, the spec is quite clear that it must reset the link-layer state:

BLUETOOTH CORE SPECIFICATION Version 5.3 | Vol 4, Part E page 1972

...HCI_Reset command shall reset the Link Manager, Baseband and Link Layer.

So it sounds like the controller shall perform and the necessary procedures before it respond with a Command Complete.

Quoted 141 lines

This CL makes the host stack to wait for |HCI_EV_DISCONN_COMPLETE| when powering off with a configurable timeout unless the timeout is set to 0.

---
Hi upstream maintainers, this is tested with an AX211 device and Logi K580 keyboard by the following procedures:
1. pair the peer and stay connected.
2. power off and on immediately
3. observe that the btsnoop log is consistent with the configured timeout.

 include/net/bluetooth/hci_core.h |  1 +
 net/bluetooth/hci_core.c         |  2 +-
 net/bluetooth/hci_sync.c         | 38 +++++++++++++++++++++++---------
 net/bluetooth/mgmt_config.c      |  6 +++++
 4 files changed, 35 insertions(+), 12 deletions(-)
include/net/bluetooth/hci_core.h +1 -0
index 8200a6689b39..ce44f9c60059 100644
@@ -432,6 +432,7 @@ struct hci_dev {
        __u16           advmon_allowlist_duration;
        __u16           advmon_no_filter_duration;
        __u8            enable_advmon_interleave_scan;
+       __u16           discon_on_poweroff_timeout;
 
        __u16           devid_source;
        __u16           devid_vendor;
net/bluetooth/hci_core.c +1 -1
index 0fefa6788911..769865494f45 100644
@@ -2479,7 +2479,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
        hdev->adv_instance_cnt = 0;
        hdev->cur_adv_instance = 0x00;
        hdev->adv_instance_timeout = 0;
-
+       hdev->discon_on_poweroff_timeout = 0;   /* Default to no timeout */
        hdev->advmon_allowlist_duration = 300;
        hdev->advmon_no_filter_duration = 500;
        hdev->enable_advmon_interleave_scan = 0x00;     /* Default to disable */
net/bluetooth/hci_sync.c +27 -11
index 3348a1b0e3f7..260e9f05359c 100644
@@ -5250,6 +5250,8 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
                               u8 reason)
 {
        struct hci_cp_disconnect cp;
+       unsigned long timeout;
+       int err;
 
        if (conn->type == AMP_LINK)
                return hci_disconnect_phy_link_sync(hdev, conn->handle, reason);
@@ -5258,19 +5260,33 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
        cp.handle = cpu_to_le16(conn->handle);
        cp.reason = reason;
 
-       /* Wait for HCI_EV_DISCONN_COMPLETE, not HCI_EV_CMD_STATUS, when the
-        * reason is anything but HCI_ERROR_REMOTE_POWER_OFF. This reason is
-        * used when suspending or powering off, where we don't want to wait
-        * for the peer's response.
+       /* The HCI_ERROR_REMOTE_POWER_OFF is used when suspending or powering off,
+        * so we don't want to waste time waiting for the reply of the peer.
+        * However, if the configuration specified, we'll wait some time to give the
+        * controller chance to actually send the disconnect command.
         */
-       if (reason != HCI_ERROR_REMOTE_POWER_OFF)
-               return __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
-                                               sizeof(cp), &cp,
-                                               HCI_EV_DISCONN_COMPLETE,
-                                               HCI_CMD_TIMEOUT, NULL);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF && !hdev->discon_on_poweroff_timeout) {
+               return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT,
+                                            sizeof(cp), &cp, HCI_CMD_TIMEOUT);
+       }
 
-       return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT, sizeof(cp), &cp,
-                                    HCI_CMD_TIMEOUT);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               timeout = msecs_to_jiffies(hdev->discon_on_poweroff_timeout);
+       else
+               timeout = HCI_CMD_TIMEOUT;
+
+       err = __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
+                                      sizeof(cp), &cp,
+                                      HCI_EV_DISCONN_COMPLETE,
+                                      timeout, NULL);
+
+       /* Ignore the error in suspending or powering off case to avoid the procedure being
+        * aborted.
+        */
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               return 0;
+
+       return err;
 }
 
 static int hci_le_connect_cancel_sync(struct hci_dev *hdev,
net/bluetooth/mgmt_config.c +6 -1
index 6ef701c27da4..f3194e3642d9 100644
@@ -78,6 +78,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                HDEV_PARAM_U16(advmon_allowlist_duration);
                HDEV_PARAM_U16(advmon_no_filter_duration);
                HDEV_PARAM_U8(enable_advmon_interleave_scan);
+               HDEV_PARAM_U16(discon_on_poweroff_timeout);
        } __packed rp = {
                TLV_SET_U16(0x0000, def_page_scan_type),
                TLV_SET_U16(0x0001, def_page_scan_int),
@@ -111,6 +112,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                TLV_SET_U16(0x001d, advmon_allowlist_duration),
                TLV_SET_U16(0x001e, advmon_no_filter_duration),
                TLV_SET_U8(0x001f, enable_advmon_interleave_scan),
+               TLV_SET_U16(0x0020, discon_on_poweroff_timeout),
        };
 
        bt_dev_dbg(hdev, "sock %p", sk);
@@ -186,6 +188,7 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x001b:
                case 0x001d:
                case 0x001e:
+               case 0x0020:
                        exp_type_len = sizeof(u16);
                        break;
                case 0x001f:
@@ -314,6 +317,9 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x0001f:
                        hdev->enable_advmon_interleave_scan = TLV_GET_U8(buffer);
                        break;
+               case 0x00020:
+                       hdev->discon_on_poweroff_timeout = TLV_GET_LE16(buffer);
+                       break;
                default:
                        bt_dev_warn(hdev, "unsupported parameter %u", type);
                        break;
--
Quoted 2 lines

2.41.0.487.g6d72f3e995-goog

--
Luiz Augusto von Dentz

Signature · 1 line
Luiz Augusto von Dentz
#7 lore raw
Quoted 13 lines · Luiz Augusto von Dentz

Hi Yun-hao,

Quoted 8 lines · Yun-hao Chung

Thanks Paul for the review. I'll fix them in the next patch.

Hi Luiz,

I think the controller did reset its internal state in the link layer, so the next time we asked the controller to create a connection to the same remote, the controller didn't reply the error in the command complete event but the connection was rejected by the remote. If we really want the controller to complete the disconnect procedure on reset, the power-off procedure could be stuck for tens of seconds if the remote was not responding. A similar problem was found and solved in this patch [Bluetooth: hci_sync: Don't wait peer's reply when powering off].

The other reason why we think this is better to be solved in the software layer is this is not only observed in Intel's AX211 but also MTK's MT7921 and maybe more.

Quoted 10 lines

Interesting, but in this case shouldn't the link-layer attempt to cleanup _before_ it reset the state? Because as far I recall doing a HCI_Reset shall cause a Disconnection even if not initiated by the host stack with HCI_Disconnect, also there is something really odd with the fact that the link-layer is generating such an error, because in case that happens isn't it clear the states are out of sync and it shall disconnect or something, I'm also curious what would happen if one quickly toggle power then the same problem should manifest, at least I don't see a difference between HCI_Reset and power off/on within the connection supervision timeout.

Do you think it's better to do this instead: Set HCI Supervision timeout with a configurable or just a small value and then it's more safe to wait for disconnect complete because the result should be returned within the supervision timeout.
In this way, we don't need the controller to do a graceful disconnection on HCI_Reset and the controller state won't be out-of-sync.

Quoted 192 lines
Quoted 13 lines

Since this is a tradeoff problem, I make it a configurable parameter so that each user can decide a good number from their perspective. If it is 0, which is the default value, the behavior should remain the same.

Do you think this sounds good to you?

Thanks,
Howard

Quoted 174 lines
Quoted 173 lines · Luiz Augusto von Dentz

Hi Howard,

Quoted 10 lines · Howard Chung

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when it is asked to reset itself. This could happen on powering off adapter.

One possible issue is that if a connection exists, and then powering off and on adapter within a short time, then our host stack assumes the conntection was disconnected but this might not be true, so if we issue a connection to the peer, it will fail with ACL Already Connected error.

That sounds more like a bug in the controller though, the spec is quite clear that it must reset the link-layer state:

BLUETOOTH CORE SPECIFICATION Version 5.3 | Vol 4, Part E page 1972

...HCI_Reset command shall reset the Link Manager, Baseband and Link Layer.

So it sounds like the controller shall perform and the necessary procedures before it respond with a Command Complete.

Quoted 141 lines

This CL makes the host stack to wait for |HCI_EV_DISCONN_COMPLETE| when powering off with a configurable timeout unless the timeout is set to 0.

---
Hi upstream maintainers, this is tested with an AX211 device and Logi K580 keyboard by the following procedures:
1. pair the peer and stay connected.
2. power off and on immediately
3. observe that the btsnoop log is consistent with the configured timeout.

 include/net/bluetooth/hci_core.h |  1 +
 net/bluetooth/hci_core.c         |  2 +-
 net/bluetooth/hci_sync.c         | 38 +++++++++++++++++++++++---------
 net/bluetooth/mgmt_config.c      |  6 +++++
 4 files changed, 35 insertions(+), 12 deletions(-)
include/net/bluetooth/hci_core.h +1 -0
index 8200a6689b39..ce44f9c60059 100644
@@ -432,6 +432,7 @@ struct hci_dev {
        __u16           advmon_allowlist_duration;
        __u16           advmon_no_filter_duration;
        __u8            enable_advmon_interleave_scan;
+       __u16           discon_on_poweroff_timeout;
 
        __u16           devid_source;
        __u16           devid_vendor;
net/bluetooth/hci_core.c +1 -1
index 0fefa6788911..769865494f45 100644
@@ -2479,7 +2479,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
        hdev->adv_instance_cnt = 0;
        hdev->cur_adv_instance = 0x00;
        hdev->adv_instance_timeout = 0;
-
+       hdev->discon_on_poweroff_timeout = 0;   /* Default to no timeout */
        hdev->advmon_allowlist_duration = 300;
        hdev->advmon_no_filter_duration = 500;
        hdev->enable_advmon_interleave_scan = 0x00;     /* Default to disable */
net/bluetooth/hci_sync.c +27 -11
index 3348a1b0e3f7..260e9f05359c 100644
@@ -5250,6 +5250,8 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
                               u8 reason)
 {
        struct hci_cp_disconnect cp;
+       unsigned long timeout;
+       int err;
 
        if (conn->type == AMP_LINK)
                return hci_disconnect_phy_link_sync(hdev, conn->handle, reason);
@@ -5258,19 +5260,33 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
        cp.handle = cpu_to_le16(conn->handle);
        cp.reason = reason;
 
-       /* Wait for HCI_EV_DISCONN_COMPLETE, not HCI_EV_CMD_STATUS, when the
-        * reason is anything but HCI_ERROR_REMOTE_POWER_OFF. This reason is
-        * used when suspending or powering off, where we don't want to wait
-        * for the peer's response.
+       /* The HCI_ERROR_REMOTE_POWER_OFF is used when suspending or powering off,
+        * so we don't want to waste time waiting for the reply of the peer.
+        * However, if the configuration specified, we'll wait some time to give the
+        * controller chance to actually send the disconnect command.
         */
-       if (reason != HCI_ERROR_REMOTE_POWER_OFF)
-               return __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
-                                               sizeof(cp), &cp,
-                                               HCI_EV_DISCONN_COMPLETE,
-                                               HCI_CMD_TIMEOUT, NULL);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF && !hdev->discon_on_poweroff_timeout) {
+               return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT,
+                                            sizeof(cp), &cp, HCI_CMD_TIMEOUT);
+       }
 
-       return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT, sizeof(cp), &cp,
-                                    HCI_CMD_TIMEOUT);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               timeout = msecs_to_jiffies(hdev->discon_on_poweroff_timeout);
+       else
+               timeout = HCI_CMD_TIMEOUT;
+
+       err = __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
+                                      sizeof(cp), &cp,
+                                      HCI_EV_DISCONN_COMPLETE,
+                                      timeout, NULL);
+
+       /* Ignore the error in suspending or powering off case to avoid the procedure being
+        * aborted.
+        */
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               return 0;
+
+       return err;
 }
 
 static int hci_le_connect_cancel_sync(struct hci_dev *hdev,
net/bluetooth/mgmt_config.c +6 -1
index 6ef701c27da4..f3194e3642d9 100644
@@ -78,6 +78,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                HDEV_PARAM_U16(advmon_allowlist_duration);
                HDEV_PARAM_U16(advmon_no_filter_duration);
                HDEV_PARAM_U8(enable_advmon_interleave_scan);
+               HDEV_PARAM_U16(discon_on_poweroff_timeout);
        } __packed rp = {
                TLV_SET_U16(0x0000, def_page_scan_type),
                TLV_SET_U16(0x0001, def_page_scan_int),
@@ -111,6 +112,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                TLV_SET_U16(0x001d, advmon_allowlist_duration),
                TLV_SET_U16(0x001e, advmon_no_filter_duration),
                TLV_SET_U8(0x001f, enable_advmon_interleave_scan),
+               TLV_SET_U16(0x0020, discon_on_poweroff_timeout),
        };
 
        bt_dev_dbg(hdev, "sock %p", sk);
@@ -186,6 +188,7 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x001b:
                case 0x001d:
                case 0x001e:
+               case 0x0020:
                        exp_type_len = sizeof(u16);
                        break;
                case 0x001f:
@@ -314,6 +317,9 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x0001f:
                        hdev->enable_advmon_interleave_scan = TLV_GET_U8(buffer);
                        break;
+               case 0x00020:
+                       hdev->discon_on_poweroff_timeout = TLV_GET_LE16(buffer);
+                       break;
                default:
                        bt_dev_warn(hdev, "unsupported parameter %u", type);
                        break;
--
Quoted 2 lines

2.41.0.487.g6d72f3e995-goog

--
Luiz Augusto von Dentz

--
Luiz Augusto von Dentz

#8 lore raw

Hi Yun-hao,

Quoted 27 lines · Yun-hao Chung
Quoted 13 lines · Luiz Augusto von Dentz

Hi Yun-hao,

Quoted 8 lines · Yun-hao Chung

Thanks Paul for the review. I'll fix them in the next patch.

Hi Luiz,

I think the controller did reset its internal state in the link layer, so the next time we asked the controller to create a connection to the same remote, the controller didn't reply the error in the command complete event but the connection was rejected by the remote. If we really want the controller to complete the disconnect procedure on reset, the power-off procedure could be stuck for tens of seconds if the remote was not responding. A similar problem was found and solved in this patch [Bluetooth: hci_sync: Don't wait peer's reply when powering off].

The other reason why we think this is better to be solved in the software layer is this is not only observed in Intel's AX211 but also MTK's MT7921 and maybe more.

Quoted 10 lines

Interesting, but in this case shouldn't the link-layer attempt to cleanup _before_ it reset the state? Because as far I recall doing a HCI_Reset shall cause a Disconnection even if not initiated by the host stack with HCI_Disconnect, also there is something really odd with the fact that the link-layer is generating such an error, because in case that happens isn't it clear the states are out of sync and it shall disconnect or something, I'm also curious what would happen if one quickly toggle power then the same problem should manifest, at least I don't see a difference between HCI_Reset and power off/on within the connection supervision timeout.

Quoted 7 lines

Do you think it's better to do this instead: Set HCI Supervision timeout with a configurable or just a small value and then it's more safe to wait for disconnect complete because the result should be returned within the supervision timeout.
In this way, we don't need the controller to do a graceful disconnection on HCI_Reset and the controller state won't be out-of-sync.

I think it is worth trying at least, not sure if we won't get into the same situation as to wait for the command to complete though, anyway the default of supervision timeout is something like 20 seconds which is way too long compared to our command timeout.

Quoted 192 lines
Quoted 192 lines
Quoted 13 lines

Since this is a tradeoff problem, I make it a configurable parameter so that each user can decide a good number from their perspective. If it is 0, which is the default value, the behavior should remain the same.

Do you think this sounds good to you?

Thanks,
Howard

Quoted 174 lines
Quoted 173 lines · Luiz Augusto von Dentz

Hi Howard,

Quoted 10 lines · Howard Chung

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when it is asked to reset itself. This could happen on powering off adapter.

One possible issue is that if a connection exists, and then powering off and on adapter within a short time, then our host stack assumes the conntection was disconnected but this might not be true, so if we issue a connection to the peer, it will fail with ACL Already Connected error.

That sounds more like a bug in the controller though, the spec is quite clear that it must reset the link-layer state:

BLUETOOTH CORE SPECIFICATION Version 5.3 | Vol 4, Part E page 1972

...HCI_Reset command shall reset the Link Manager, Baseband and Link Layer.

So it sounds like the controller shall perform and the necessary procedures before it respond with a Command Complete.

Quoted 141 lines

This CL makes the host stack to wait for |HCI_EV_DISCONN_COMPLETE| when powering off with a configurable timeout unless the timeout is set to 0.

---
Hi upstream maintainers, this is tested with an AX211 device and Logi K580 keyboard by the following procedures:
1. pair the peer and stay connected.
2. power off and on immediately
3. observe that the btsnoop log is consistent with the configured timeout.

 include/net/bluetooth/hci_core.h |  1 +
 net/bluetooth/hci_core.c         |  2 +-
 net/bluetooth/hci_sync.c         | 38 +++++++++++++++++++++++---------
 net/bluetooth/mgmt_config.c      |  6 +++++
 4 files changed, 35 insertions(+), 12 deletions(-)
include/net/bluetooth/hci_core.h +1 -0
index 8200a6689b39..ce44f9c60059 100644
@@ -432,6 +432,7 @@ struct hci_dev {
        __u16           advmon_allowlist_duration;
        __u16           advmon_no_filter_duration;
        __u8            enable_advmon_interleave_scan;
+       __u16           discon_on_poweroff_timeout;
 
        __u16           devid_source;
        __u16           devid_vendor;
net/bluetooth/hci_core.c +1 -1
index 0fefa6788911..769865494f45 100644
@@ -2479,7 +2479,7 @@ struct hci_dev *hci_alloc_dev_priv(int sizeof_priv)
        hdev->adv_instance_cnt = 0;
        hdev->cur_adv_instance = 0x00;
        hdev->adv_instance_timeout = 0;
-
+       hdev->discon_on_poweroff_timeout = 0;   /* Default to no timeout */
        hdev->advmon_allowlist_duration = 300;
        hdev->advmon_no_filter_duration = 500;
        hdev->enable_advmon_interleave_scan = 0x00;     /* Default to disable */
net/bluetooth/hci_sync.c +27 -11
index 3348a1b0e3f7..260e9f05359c 100644
@@ -5250,6 +5250,8 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
                               u8 reason)
 {
        struct hci_cp_disconnect cp;
+       unsigned long timeout;
+       int err;
 
        if (conn->type == AMP_LINK)
                return hci_disconnect_phy_link_sync(hdev, conn->handle, reason);
@@ -5258,19 +5260,33 @@ static int hci_disconnect_sync(struct hci_dev *hdev, struct hci_conn *conn,
        cp.handle = cpu_to_le16(conn->handle);
        cp.reason = reason;
 
-       /* Wait for HCI_EV_DISCONN_COMPLETE, not HCI_EV_CMD_STATUS, when the
-        * reason is anything but HCI_ERROR_REMOTE_POWER_OFF. This reason is
-        * used when suspending or powering off, where we don't want to wait
-        * for the peer's response.
+       /* The HCI_ERROR_REMOTE_POWER_OFF is used when suspending or powering off,
+        * so we don't want to waste time waiting for the reply of the peer.
+        * However, if the configuration specified, we'll wait some time to give the
+        * controller chance to actually send the disconnect command.
         */
-       if (reason != HCI_ERROR_REMOTE_POWER_OFF)
-               return __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
-                                               sizeof(cp), &cp,
-                                               HCI_EV_DISCONN_COMPLETE,
-                                               HCI_CMD_TIMEOUT, NULL);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF && !hdev->discon_on_poweroff_timeout) {
+               return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT,
+                                            sizeof(cp), &cp, HCI_CMD_TIMEOUT);
+       }
 
-       return __hci_cmd_sync_status(hdev, HCI_OP_DISCONNECT, sizeof(cp), &cp,
-                                    HCI_CMD_TIMEOUT);
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               timeout = msecs_to_jiffies(hdev->discon_on_poweroff_timeout);
+       else
+               timeout = HCI_CMD_TIMEOUT;
+
+       err = __hci_cmd_sync_status_sk(hdev, HCI_OP_DISCONNECT,
+                                      sizeof(cp), &cp,
+                                      HCI_EV_DISCONN_COMPLETE,
+                                      timeout, NULL);
+
+       /* Ignore the error in suspending or powering off case to avoid the procedure being
+        * aborted.
+        */
+       if (reason == HCI_ERROR_REMOTE_POWER_OFF)
+               return 0;
+
+       return err;
 }
 
 static int hci_le_connect_cancel_sync(struct hci_dev *hdev,
net/bluetooth/mgmt_config.c +6 -1
index 6ef701c27da4..f3194e3642d9 100644
@@ -78,6 +78,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                HDEV_PARAM_U16(advmon_allowlist_duration);
                HDEV_PARAM_U16(advmon_no_filter_duration);
                HDEV_PARAM_U8(enable_advmon_interleave_scan);
+               HDEV_PARAM_U16(discon_on_poweroff_timeout);
        } __packed rp = {
                TLV_SET_U16(0x0000, def_page_scan_type),
                TLV_SET_U16(0x0001, def_page_scan_int),
@@ -111,6 +112,7 @@ int read_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                TLV_SET_U16(0x001d, advmon_allowlist_duration),
                TLV_SET_U16(0x001e, advmon_no_filter_duration),
                TLV_SET_U8(0x001f, enable_advmon_interleave_scan),
+               TLV_SET_U16(0x0020, discon_on_poweroff_timeout),
        };
 
        bt_dev_dbg(hdev, "sock %p", sk);
@@ -186,6 +188,7 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x001b:
                case 0x001d:
                case 0x001e:
+               case 0x0020:
                        exp_type_len = sizeof(u16);
                        break;
                case 0x001f:
@@ -314,6 +317,9 @@ int set_def_system_config(struct sock *sk, struct hci_dev *hdev, void *data,
                case 0x0001f:
                        hdev->enable_advmon_interleave_scan = TLV_GET_U8(buffer);
                        break;
+               case 0x00020:
+                       hdev->discon_on_poweroff_timeout = TLV_GET_LE16(buffer);
+                       break;
                default:
                        bt_dev_warn(hdev, "unsupported parameter %u", type);
                        break;
--
Quoted 2 lines

2.41.0.487.g6d72f3e995-goog

--
Luiz Augusto von Dentz

--
Luiz Augusto von Dentz

Signature · 1 line
Luiz Augusto von Dentz
#9 lore raw
In reply to Howard Chung
Quoted 3 lines · Howard Chung

For some controllers, it is known that when the HCI disconnect and HCI Reset are too close to each other, the LMP disconnect command might not been sent out yet and the command will be dropped by the controoler when

nit: controoler -> controller

...