Bluetooth: When HCI work queue is drained, only queue chained work
Subsystems (4)
PATCH v3 1/1 Bluetooth: When HCI work queue is drained, only queue chained work
The HCI command, event, and data packet processing workqueue is drained to avoid deadlock in commit76727c02c1e1 ("Bluetooth: Call drain_workqueue() before resetting state").
There is another delayed work, which will queue command to this drained workqueue. Which results in the following error report:
Bluetooth: hci2: command 0x040f tx timeout
WARNING: CPU: 1 PID: 18374 at kernel/workqueue.c:1438 __queue_work+0xdad/0x1140 Workqueue: events hci_cmd_timeout RIP: 0010:__queue_work+0xdad/0x1140 RSP: 0000:ffffc90002cffc60 EFLAGS: 00010093 RAX: 0000000000000000 RBX: ffff8880b9d3ec00 RCX: 0000000000000000 RDX: ffff888024ba0000 RSI: ffffffff814e048d RDI: ffff8880b9d3ec08 RBP: 0000000000000008 R08: 0000000000000000 R09: 00000000b9d39700 R10: ffffffff814f73c6 R11: 0000000000000000 R12: ffff88807cce4c60 R13: 0000000000000000 R14: ffff8880796d8800 R15: ffff8880796d8800 FS: 0000000000000000(0000) GS:ffff8880b9d00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 000000c0174b4000 CR3: 000000007cae9000 CR4: 00000000003506e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: <TASK> ? queue_work_on+0xcb/0x110 ? lockdep_hardirqs_off+0x90/0xd0 queue_work_on+0xee/0x110 process_one_work+0x996/0x1610 ? pwq_dec_nr_in_flight+0x2a0/0x2a0 ? rwlock_bug.part.0+0x90/0x90 ? _raw_spin_lock_irq+0x41/0x50 worker_thread+0x665/0x1080 ? process_one_work+0x1610/0x1610 kthread+0x2e9/0x3a0 ? kthread_complete_and_exit+0x40/0x40 ret_from_fork+0x1f/0x30 </TASK>
To fix this, we can add a new HCI_DRAIN_WQ flag, and don't queue the timeout workqueue while command workqueue is draining.
Changelog:
v1 -> v2:
- Move the workqueue drain flag to controller flags, and use hci_dev_*_flag.
- Add missing ncmd_timer cancel.
- Clear DRAIN_WORKQUEUE flag after device command flushed.
v2 -> v3:
- Fix typos in comments and adjust the description.
--- include/net/bluetooth/hci.h | 1 + net/bluetooth/hci_core.c | 10 +++++++++- net/bluetooth/hci_event.c | 5 +++-- 3 files changed, 13 insertions(+), 3 deletions(-)
include/net/bluetooth/hci.h +1 -0
@@ -361,6 +361,7 @@ enum {HCI_QUALITY_REPORT ,HCI_OFFLOAD_CODECS_ENABLED ,HCI_LE_SIMULTANEOUS_ROLES , +HCI_CMD_DRAIN_WORKQUEUE , __HCI_NUM_FLAGS, };
net/bluetooth/hci_core.c +9 -1
@@ -593,6 +593,11 @@ static int hci_dev_do_reset(struct hci_dev *hdev)skb_queue_purge (&hdev->rx_q);skb_queue_purge (&hdev->cmd_q); +/* Cancel these to avoid queueing non-chained pending work */ +hci_dev_set_flag (hdev,HCI_CMD_DRAIN_WORKQUEUE ); +cancel_delayed_work (&hdev->cmd_timer); +cancel_delayed_work (&hdev->ncmd_timer); +/* Avoid potential lockdep warnings from the *_flush() calls by * ensuring the workqueue is empty up front. */
@@ -606,6 +611,8 @@ static int hci_dev_do_reset(struct hci_dev *hdev)if (hdev->flush) hdev->flush (hdev); +hci_dev_clear_flag (hdev,HCI_CMD_DRAIN_WORKQUEUE ); +atomic_set (&hdev->cmd_cnt,1 ); hdev->acl_cnt =0 ; hdev->sco_cnt =0 ; hdev->le_cnt =0 ;
@@ -3861,7 +3868,8 @@ static void hci_cmd_work(struct work_struct *work)if (res <0 )__hci_cmd_sync_cancel (hdev, -res); -if (test_bit (HCI_RESET , &hdev->flags)) +if (test_bit (HCI_RESET , &hdev->flags) || +hci_dev_test_flag (hdev,HCI_CMD_DRAIN_WORKQUEUE ))cancel_delayed_work (&hdev->cmd_timer);else schedule_delayed_work(&hdev->cmd_timer,
net/bluetooth/hci_event.c +3 -2
@@ -3768,8 +3768,9 @@ static inline void handle_cmd_cnt_and_timer(struct hci_dev *hdev, u8 ncmd)cancel_delayed_work (&hdev->ncmd_timer);atomic_set (&hdev->cmd_cnt,1 ); }else { -schedule_delayed_work (&hdev->ncmd_timer, -HCI_NCMD_TIMEOUT ); +if (!hci_dev_test_flag (hdev,HCI_CMD_DRAIN_WORKQUEUE )) +schedule_delayed_work (&hdev->ncmd_timer, +HCI_NCMD_TIMEOUT ); } } }
Signature · 1 line
2.24.3 (Apple Git-128)
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=647101
---Test result---
Test Summary:
CheckPatch PASS 1.88 seconds GitLint PASS 1.05 seconds SubjectPrefix PASS 0.86 seconds BuildKernel PASS 30.46 seconds BuildKernel32 PASS 27.47 seconds Incremental Build with patchesPASS 38.46 seconds TestRunner: Setup PASS 449.95 seconds TestRunner: l2cap-tester PASS 16.39 seconds TestRunner: bnep-tester PASS 5.83 seconds TestRunner: mgmt-tester PASS 98.99 seconds TestRunner: rfcomm-tester PASS 9.42 seconds TestRunner: sco-tester PASS 9.15 seconds TestRunner: smp-tester PASS 8.97 seconds TestRunner: userchan-tester PASS 6.00 seconds
---
Regards,
Linux Bluetooth
This patch was applied to bluetooth/bluetooth-next.git (master) by Marcel Holtmann <[email protected]>:
Hello:
This patch was applied to bluetooth/bluetooth-next.git (master) by Marcel Holtmann <[email protected]>:
Quoted 8 lines · On Fri, 3 Jun 2022 16:19:14 +0800 you wrote:
The HCI command, event, and data packet processing workqueue is drained to avoid deadlock in commit
76727c02c1e1("Bluetooth: Calldrain_workqueue()before resetting state").There is another delayed work, which will queue command to this drained workqueue. Which results in the following error report:
[...]
Here is the summary with links:
- [v3] Bluetooth: When HCI work queue is drained, only queue chained work
https://git.kernel.org/bluetooth/bluetooth-next/c/dba7abaead13
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html