Linux-Bluetooth Archive on lore.kernel.org
 help / color / Atom feed
* [PATCH v2] Bluetooth: Fix suspend notifier race
@ 2020-07-28 16:52 Abhishek Pandit-Subedi
  2020-07-28 16:53 ` Abhishek Pandit-Subedi
  0 siblings, 1 reply; 2+ messages in thread
From: Abhishek Pandit-Subedi @ 2020-07-28 16:52 UTC (permalink / raw)
  To: marcel
  Cc: chromeos-bluetooth-upstreaming, linux-bluetooth,
	Abhishek Pandit-Subedi, Miao-chen Chou, David S. Miller,
	Johan Hedberg, netdev, linux-kernel, Jakub Kicinski

Unregister from suspend notifications and cancel suspend preparations
before running hci_dev_do_close. Otherwise, the suspend notifier may
race with unregister and cause cmd_timeout even after hdev has been
freed.

Below is the trace from when this panic was seen:

[  832.578518] Bluetooth: hci_core.c:hci_cmd_timeout() hci0: command 0x0c05 tx timeout
[  832.586200] BUG: kernel NULL pointer dereference, address: 0000000000000000
[  832.586203] #PF: supervisor read access in kernel mode
[  832.586205] #PF: error_code(0x0000) - not-present page
[  832.586206] PGD 0 P4D 0
[  832.586210] PM: suspend exit
[  832.608870] Oops: 0000 [#1] PREEMPT SMP NOPTI
[  832.613232] CPU: 3 PID: 10755 Comm: kworker/3:7 Not tainted 5.4.44-04894-g1e9dbb96a161 #1
[  832.630036] Workqueue: events hci_cmd_timeout [bluetooth]
[  832.630046] RIP: 0010:__queue_work+0xf0/0x374
[  832.630051] RSP: 0018:ffff9b5285f1fdf8 EFLAGS: 00010046
[  832.674033] RAX: ffff8a97681bac00 RBX: 0000000000000000 RCX: ffff8a976a000600
[  832.681162] RDX: 0000000000000000 RSI: 0000000000000009 RDI: ffff8a976a000748
[  832.688289] RBP: ffff9b5285f1fe38 R08: 0000000000000000 R09: ffff8a97681bac00
[  832.695418] R10: 0000000000000002 R11: ffff8a976a0006d8 R12: ffff8a9745107600
[  832.698045] usb 1-6: new full-speed USB device number 119 using xhci_hcd
[  832.702547] R13: ffff8a9673658850 R14: 0000000000000040 R15: 000000000000001e
[  832.702549] FS:  0000000000000000(0000) GS:ffff8a976af80000(0000) knlGS:0000000000000000
[  832.702550] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  832.702550] CR2: 0000000000000000 CR3: 000000010415a000 CR4: 00000000003406e0
[  832.702551] Call Trace:
[  832.702558]  queue_work_on+0x3f/0x68
[  832.702562]  process_one_work+0x1db/0x396
[  832.747397]  worker_thread+0x216/0x375
[  832.751147]  kthread+0x138/0x140
[  832.754377]  ? pr_cont_work+0x58/0x58
[  832.758037]  ? kthread_blkcg+0x2e/0x2e
[  832.761787]  ret_from_fork+0x22/0x40
[  832.846191] ---[ end trace fa93f466da517212 ]---

Signed-off-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
Reviewed-by: Miao-chen Chou <mcchou@chromium.org>
---
Hi Marcel,

This fixes a race between hci_unregister_dev and the suspend notifier.

The suspend notifier handler seemed to be scheduling commands even after
it was cleaned up and this was resulting in a panic in cmd_timeout (when
it tries to requeue the cmd_timer).

This was tested on 5.4 kernel with a suspend+resume stress test for 500+
iterations. I also confirmed that after a usb disconnect, the suspend
notifier times out before the USB device is probed again (fixing the
original race between the usb_disconnect + probe and the notifier).

Thanks
Abhishek


Changes in v2:
* Moved oops into commit message

 net/bluetooth/hci_core.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 5394ab56c915a9..4ba23b821cbf4a 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -3767,9 +3767,10 @@ void hci_unregister_dev(struct hci_dev *hdev)
 
 	cancel_work_sync(&hdev->power_on);
 
-	hci_dev_do_close(hdev);
-
 	unregister_pm_notifier(&hdev->suspend_notifier);
+	cancel_work_sync(&hdev->suspend_prepare);
+
+	hci_dev_do_close(hdev);
 
 	if (!test_bit(HCI_INIT, &hdev->flags) &&
 	    !hci_dev_test_flag(hdev, HCI_SETUP) &&
-- 
2.28.0.rc0.142.g3c755180ce-goog


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH v2] Bluetooth: Fix suspend notifier race
  2020-07-28 16:52 [PATCH v2] Bluetooth: Fix suspend notifier race Abhishek Pandit-Subedi
@ 2020-07-28 16:53 ` Abhishek Pandit-Subedi
  0 siblings, 0 replies; 2+ messages in thread
From: Abhishek Pandit-Subedi @ 2020-07-28 16:53 UTC (permalink / raw)
  To: Marcel Holtmann
  Cc: ChromeOS Bluetooth Upstreaming, Bluez mailing list,
	Miao-chen Chou, David S. Miller, Johan Hedberg, netdev, LKML,
	Jakub Kicinski

I sent this a bit too quick without a Fixes tag. Please disregard. v3 coming up.

On Tue, Jul 28, 2020 at 9:53 AM Abhishek Pandit-Subedi
<abhishekpandit@chromium.org> wrote:
>
> Unregister from suspend notifications and cancel suspend preparations
> before running hci_dev_do_close. Otherwise, the suspend notifier may
> race with unregister and cause cmd_timeout even after hdev has been
> freed.
>
> Below is the trace from when this panic was seen:
>
> [  832.578518] Bluetooth: hci_core.c:hci_cmd_timeout() hci0: command 0x0c05 tx timeout
> [  832.586200] BUG: kernel NULL pointer dereference, address: 0000000000000000
> [  832.586203] #PF: supervisor read access in kernel mode
> [  832.586205] #PF: error_code(0x0000) - not-present page
> [  832.586206] PGD 0 P4D 0
> [  832.586210] PM: suspend exit
> [  832.608870] Oops: 0000 [#1] PREEMPT SMP NOPTI
> [  832.613232] CPU: 3 PID: 10755 Comm: kworker/3:7 Not tainted 5.4.44-04894-g1e9dbb96a161 #1
> [  832.630036] Workqueue: events hci_cmd_timeout [bluetooth]
> [  832.630046] RIP: 0010:__queue_work+0xf0/0x374
> [  832.630051] RSP: 0018:ffff9b5285f1fdf8 EFLAGS: 00010046
> [  832.674033] RAX: ffff8a97681bac00 RBX: 0000000000000000 RCX: ffff8a976a000600
> [  832.681162] RDX: 0000000000000000 RSI: 0000000000000009 RDI: ffff8a976a000748
> [  832.688289] RBP: ffff9b5285f1fe38 R08: 0000000000000000 R09: ffff8a97681bac00
> [  832.695418] R10: 0000000000000002 R11: ffff8a976a0006d8 R12: ffff8a9745107600
> [  832.698045] usb 1-6: new full-speed USB device number 119 using xhci_hcd
> [  832.702547] R13: ffff8a9673658850 R14: 0000000000000040 R15: 000000000000001e
> [  832.702549] FS:  0000000000000000(0000) GS:ffff8a976af80000(0000) knlGS:0000000000000000
> [  832.702550] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [  832.702550] CR2: 0000000000000000 CR3: 000000010415a000 CR4: 00000000003406e0
> [  832.702551] Call Trace:
> [  832.702558]  queue_work_on+0x3f/0x68
> [  832.702562]  process_one_work+0x1db/0x396
> [  832.747397]  worker_thread+0x216/0x375
> [  832.751147]  kthread+0x138/0x140
> [  832.754377]  ? pr_cont_work+0x58/0x58
> [  832.758037]  ? kthread_blkcg+0x2e/0x2e
> [  832.761787]  ret_from_fork+0x22/0x40
> [  832.846191] ---[ end trace fa93f466da517212 ]---
>
> Signed-off-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
> Reviewed-by: Miao-chen Chou <mcchou@chromium.org>
> ---
> Hi Marcel,
>
> This fixes a race between hci_unregister_dev and the suspend notifier.
>
> The suspend notifier handler seemed to be scheduling commands even after
> it was cleaned up and this was resulting in a panic in cmd_timeout (when
> it tries to requeue the cmd_timer).
>
> This was tested on 5.4 kernel with a suspend+resume stress test for 500+
> iterations. I also confirmed that after a usb disconnect, the suspend
> notifier times out before the USB device is probed again (fixing the
> original race between the usb_disconnect + probe and the notifier).
>
> Thanks
> Abhishek
>
>
> Changes in v2:
> * Moved oops into commit message
>
>  net/bluetooth/hci_core.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
> index 5394ab56c915a9..4ba23b821cbf4a 100644
> --- a/net/bluetooth/hci_core.c
> +++ b/net/bluetooth/hci_core.c
> @@ -3767,9 +3767,10 @@ void hci_unregister_dev(struct hci_dev *hdev)
>
>         cancel_work_sync(&hdev->power_on);
>
> -       hci_dev_do_close(hdev);
> -
>         unregister_pm_notifier(&hdev->suspend_notifier);
> +       cancel_work_sync(&hdev->suspend_prepare);
> +
> +       hci_dev_do_close(hdev);
>
>         if (!test_bit(HCI_INIT, &hdev->flags) &&
>             !hci_dev_test_flag(hdev, HCI_SETUP) &&
> --
> 2.28.0.rc0.142.g3c755180ce-goog
>

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, back to index

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-07-28 16:52 [PATCH v2] Bluetooth: Fix suspend notifier race Abhishek Pandit-Subedi
2020-07-28 16:53 ` Abhishek Pandit-Subedi

Linux-Bluetooth Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-bluetooth/0 linux-bluetooth/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-bluetooth linux-bluetooth/ https://lore.kernel.org/linux-bluetooth \
		linux-bluetooth@vger.kernel.org
	public-inbox-index linux-bluetooth

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-bluetooth


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git