2014-10-31 05:23:06

by Chan-yeol Park

Subject: [RFC v2] Bluetooth : Fix hci_sync miss wakeup interrupt

From: Chan-yeol Park <[email protected]>

__hci_cmd_sync_ev(), __hci_req_sync() could miss wake_up_interrupt from
hci_req_sync_complete() because hci_cmd_work() workquee and its reponse
could be completed before they are ready to get the signal through
add_wait_queue(), set_current_state(TASK_INTERRUPTIBLE).

Signed-off-by: Chan-yeol Park <[email protected]>
Signed-off-by: Kyungmin Park <[email protected]>
---
net/bluetooth/hci_core.c | 18 +++++++++++-------
1 file changed, 11 insertions(+), 7 deletions(-)

diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index cb05d7f..c008f1f 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -1147,13 +1147,15 @@ struct sk_buff *__hci_cmd_sync_ev(struct hci_dev *hdev, u16 opcode, u32 plen,

hdev->req_status = HCI_REQ_PEND;

- err = hci_req_run(&req, hci_req_sync_complete);
- if (err < 0)
- return ERR_PTR(err);
-
add_wait_queue(&hdev->req_wait_q, &wait);
set_current_state(TASK_INTERRUPTIBLE);

+ err = hci_req_run(&req, hci_req_sync_complete);
+ if (err < 0) {
+ remove_wait_queue(&hdev->req_wait_q, &wait);
+ return ERR_PTR(err);
+ }
+
schedule_timeout(timeout);

remove_wait_queue(&hdev->req_wait_q, &wait);
@@ -1211,10 +1213,15 @@ static int __hci_req_sync(struct hci_dev *hdev,

func(&req, opt);

+ add_wait_queue(&hdev->req_wait_q, &wait);
+ set_current_state(TASK_INTERRUPTIBLE);
+
err = hci_req_run(&req, hci_req_sync_complete);
if (err < 0) {
hdev->req_status = 0;

+ remove_wait_queue(&hdev->req_wait_q, &wait);
+
/* ENODATA means the HCI request command queue is empty.
* This can happen when a request with conditionals doesn't
* trigger any commands to be sent. This is normal behavior
@@ -1226,9 +1233,6 @@ static int __hci_req_sync(struct hci_dev *hdev,
return err;
}

- add_wait_queue(&hdev->req_wait_q, &wait);
- set_current_state(TASK_INTERRUPTIBLE);
-
schedule_timeout(timeout);

remove_wait_queue(&hdev->req_wait_q, &wait);
--
1.9.1

2014-11-03 05:21:56

by Chan-yeol Park

Subject: Re: [RFC v2] Bluetooth : Fix hci_sync miss wakeup interrupt

Hi Johan,
On 11/02/2014 06:24 AM, Johan Hedberg wrote:
> This fixed a really frequent hang for me of init HCI commands when
> running mgmt-tester. Is it through the our user space testers that you
> were seeing this issue as well? I might have seen the problem in the
> past too but for some reason it started showing up really frequently
> now. Johan

In case of me, I've faced this bug while I execute "hciconfig hci0 down"
and hci_dev_do_close(). and it takes about 5 second(HCI_CMD_TIMEOUT).
As you mentioned this bug is shown recently. I guess some of patches
reveal our potential bug.

Thanks
Chanyeol

2014-11-01 21:24:09

by Johan Hedberg

Subject: Re: [RFC v2] Bluetooth : Fix hci_sync miss wakeup interrupt

Hi Chan-yeol,

On Fri, Oct 31, 2014, [email protected] wrote:
> __hci_cmd_sync_ev(), __hci_req_sync() could miss wake_up_interrupt from
> hci_req_sync_complete() because hci_cmd_work() workquee and its reponse
> could be completed before they are ready to get the signal through
> add_wait_queue(), set_current_state(TASK_INTERRUPTIBLE).
>
> Signed-off-by: Chan-yeol Park <[email protected]>
> Signed-off-by: Kyungmin Park <[email protected]>
> ---
> net/bluetooth/hci_core.c | 18 +++++++++++-------
> 1 file changed, 11 insertions(+), 7 deletions(-)

Applied to bluetooth-next. Thanks.

This fixed a really frequent hang for me of init HCI commands when
running mgmt-tester. Is it through the our user space testers that you
were seeing this issue as well? I might have seen the problem in the
past too but for some reason it started showing up really frequently
now.

Johan