All of lore.kernel.org
 help / color / mirror / Atom feed
From: Marcel Holtmann <marcel@holtmann.org>
To: Dean Jenkins <Dean_Jenkins@mentor.com>
Cc: "Gustavo F. Padovan" <gustavo@padovan.org>,
	Johan Hedberg <johan.hedberg@gmail.com>,
	linux-bluetooth@vger.kernel.org
Subject: Re: [RFC V1 04/16] Bluetooth: hci_ldisc: Add HCI RESET comment to hci_unregister_dev() call
Date: Thu, 30 Mar 2017 12:11:36 +0200	[thread overview]
Message-ID: <5EAAF4AF-1814-4FC2-83D7-701C93F6B779@holtmann.org> (raw)
In-Reply-To: <1490723429-28870-5-git-send-email-Dean_Jenkins@mentor.com>

Hi Dean,

> This commit relates to the HCI UART quirk HCI_QUIRK_RESET_ON_CLOSE
> which is enabled by default and can be disabled by setting hdev_flags
> flag HCI_UART_RESET_ON_INIT via ioctl HCIUARTSETFLAGS from userland.
> 
> The implementation of HCI_QUIRK_RESET_ON_CLOSE is broken for the BCSP
> Data Link protocol layer because it can be seen that
> hci_unregister_dev() takes 2 seconds to run due to BCSP communications
> failure. Suspect that sending of the HCI RESET command is broken
> for the other Data Link protocols as well.
> 
> Therefore, add a comment to inform that the call to
> hci_unregister_dev() in hci_uart_tty_close() may send a HCI RESET
> command. This transmission needs the HCI UART driver to remain
> operational including having the Data Link protocol being bound to
> the HCI UART driver. If the transmission fails, then
> hci_unregister_dev() waits HCI_CMD_TIMEOUT (2) seconds for the
> timeout to occur which is undesirable.
> 
> The current software has a call to hci_unregister_dev(hdev) in
> hci_uart_tty_close() which can cause an attempt at sending the
> command HCI RESET to occur through the following callstack:
> 
> hci_uart_tty_close()
> hci_unregister_dev()
> hci_dev_do_close()
> __hci_req_sync() to queue up command HCI RESET
> which adds a work-item to the hdev->workqueue and waits 2 seconds
> for a response
> 
> Subsequently
> hdev->workqueue runs and processes the work-item by calling
> hci_cmd_work()
> hci_send_frame()
> hci_uart_send_frame() to enqueue into the Data Link protocol layer
> 
> No protocol response comes so hci_unregister_dev() takes 2 seconds to
> run!
> 
> In fact, this hidden transmission of the HCI RESET command is most
> likely the root cause of why hci_uart_tty_close() crashed in the past
> and was "fixed" with multiple workarounds in the past.
> 
> The underlying design flaw is that hci_uart_tty_close() is interfering
> with various aspects of transmitting and receiving HCI messages
> whilst hci_unregister_dev() is trying to use the Data Link protocol
> to send the HCI RESET command. Consequently, the design flaw
> causes the Data Link protocol to retransmit as no reply comes from
> the Bluetooth Radio module over the UART link.
> 
> Over the many years of "fixes", this caused the logic in
> hci_uart_tty_close() to become over-complex.
> 
> Signed-off-by: Dean Jenkins <Dean_Jenkins@mentor.com>
> ---
> drivers/bluetooth/hci_ldisc.c | 6 ++++++
> 1 file changed, 6 insertions(+)
> 
> diff --git a/drivers/bluetooth/hci_ldisc.c b/drivers/bluetooth/hci_ldisc.c
> index 1887ad4..c78c9dc 100644
> --- a/drivers/bluetooth/hci_ldisc.c
> +++ b/drivers/bluetooth/hci_ldisc.c
> @@ -499,6 +499,12 @@ static void hci_uart_tty_close(struct tty_struct *tty)
> 	if (test_and_clear_bit(HCI_UART_PROTO_READY, &hu->flags)) {
> 		if (hdev) {
> 			if (test_bit(HCI_UART_REGISTERED, &hu->flags))
> +				/* Note hci_unregister_dev() may try to send
> +				 * a HCI RESET command. If the transmission
> +				 * fails then hci_unregister_dev() waits
> +				 * HCI_CMD_TIMEOUT (2) seconds for the timeout
> +				 * to occur.
> +				 */
> 				hci_unregister_dev(hdev);

lets put { } around this if statement. I know it is no needed, but we generally do that for clarity.

Regards

Marcel


  reply	other threads:[~2017-03-30 10:11 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-03-28 17:50 [RFC V1 00/16] hci_ldisc hci_uart_tty_close() fixes Dean Jenkins
2017-03-28 17:50 ` [RFC V1 01/16] Bluetooth: hci_ldisc: Add missing return in hci_uart_init_work() Dean Jenkins
2017-03-28 17:50 ` [RFC V1 02/16] Bluetooth: hci_ldisc: Ensure hu->hdev set to NULL before freeing hdev Dean Jenkins
2017-03-28 17:50 ` [RFC V1 03/16] Bluetooth: hci_ldisc: Add missing clear HCI_UART_PROTO_READY Dean Jenkins
2017-03-28 17:50 ` [RFC V1 04/16] Bluetooth: hci_ldisc: Add HCI RESET comment to hci_unregister_dev() call Dean Jenkins
2017-03-30 10:11   ` Marcel Holtmann [this message]
2017-03-28 17:50 ` [RFC V1 05/16] Bluetooth: hci_ldisc: Add protocol check to hci_uart_send_frame() Dean Jenkins
2017-03-28 17:50 ` [RFC V1 06/16] Bluetooth: hci_ldisc: Add protocol check to hci_uart_dequeue() Dean Jenkins
2017-03-28 17:50 ` [RFC V1 07/16] Bluetooth: hci_ldisc: Add protocol check to hci_uart_tx_wakeup() Dean Jenkins
2017-03-30 10:11   ` Marcel Holtmann
2017-03-28 17:50 ` [RFC V1 08/16] Bluetooth: hci_ldisc: Separate flag handling in hci_uart_tty_close() Dean Jenkins
2017-03-28 17:50 ` [RFC V1 09/16] Bluetooth: hci_ldisc: Tidy-up HCI_UART_REGISTERED " Dean Jenkins
2017-03-28 17:50 ` [RFC V1 10/16] Bluetooth: hci_ldisc: hci_uart_tty_close() detach tty after last msg Dean Jenkins
2017-03-28 17:50 ` [RFC V1 11/16] Bluetooth: hci_ldisc: hci_uart_tty_close() move hci_uart_close() Dean Jenkins
2017-03-28 17:50 ` [RFC V1 12/16] Bluetooth: hci_ldisc: hci_uart_tty_close() move cancel_work_sync() Dean Jenkins
2017-03-28 17:50 ` [RFC V1 13/16] Bluetooth: hci_ldisc: hci_uart_tty_close() free hu->tx_skb Dean Jenkins
2017-03-28 17:50 ` [RFC V1 14/16] Bluetooth: hci_ldisc: Simplify flushing Dean Jenkins
2017-03-28 17:50 ` [RFC V1 15/16] Bluetooth: hci_ldisc: Use rwlocking to avoid closing proto races Dean Jenkins
2017-03-28 17:50 ` [RFC V1 16/16] Bluetooth: hci_ldisc: Add ioctl_mutex avoiding concurrency Dean Jenkins
2017-03-30 10:11 ` [RFC V1 00/16] hci_ldisc hci_uart_tty_close() fixes Marcel Holtmann
2017-04-03 15:09   ` Dean Jenkins
2017-04-03 15:51     ` Marcel Holtmann
2017-04-04 20:36       ` Dean Jenkins
2017-04-05 15:28         ` Dean Jenkins
2017-04-06  7:23           ` Marcel Holtmann

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=5EAAF4AF-1814-4FC2-83D7-701C93F6B779@holtmann.org \
    --to=marcel@holtmann.org \
    --cc=Dean_Jenkins@mentor.com \
    --cc=gustavo@padovan.org \
    --cc=johan.hedberg@gmail.com \
    --cc=linux-bluetooth@vger.kernel.org \
    --subject='Re: [RFC V1 04/16] Bluetooth: hci_ldisc: Add HCI RESET comment to hci_unregister_dev() call' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.