2018-06-01 10:41:27

by Ganapathi Bhat

[permalink] [raw]
Subject: [PATCH 1/2] Revert "mwifiex: handle race during mwifiex_usb_disconnect"

This reverts commit b817047ae70c0bd67b677b65d0d69d72cd6e9728.

We have a better fix for this issue, which will be sent on top
of this revert.

Signed-off-by: Ganapathi Bhat <[email protected]>
---
drivers/net/wireless/marvell/mwifiex/usb.c | 3 ---
1 file changed, 3 deletions(-)

diff --git a/drivers/net/wireless/marvell/mwifiex/usb.c b/drivers/net/wireless/marvell/mwifiex/usb.c
index 6e3cf98..bc475b8 100644
--- a/drivers/net/wireless/marvell/mwifiex/usb.c
+++ b/drivers/net/wireless/marvell/mwifiex/usb.c
@@ -644,9 +644,6 @@ static void mwifiex_usb_disconnect(struct usb_interface *intf)
MWIFIEX_FUNC_SHUTDOWN);
}

- if (adapter->workqueue)
- flush_workqueue(adapter->workqueue);
-
mwifiex_usb_free(card);

mwifiex_dbg(adapter, FATAL,
--
1.9.1


2018-06-01 10:41:37

by Ganapathi Bhat

[permalink] [raw]
Subject: [PATCH 2/2] mwifiex: handle race during mwifiex_usb_disconnect

Race condition is observed during rmmod of mwifiex_usb:

1. The rmmod thread will call mwifiex_usb_disconnect(), download
SHUTDOWN command and do wait_event_interruptible_timeout(),
waiting for response.

2. The main thread will handle the response and will do a
wake_up_interruptible(), unblocking rmmod thread.

3. On getting unblocked, rmmod thread will make rx_cmd.urb = NULL in
mwifiex_usb_free().

4. The main thread will try to resubmit rx_cmd.urb in
mwifiex_usb_submit_rx_urb(), which is NULL.

To fix this, move mwifiex_usb_free() from mwifiex_usb_disconnect
to mwifiex_unregister_dev(). Function mwifiex_unregister_dev() is
called after flushing the command and RX work queues.

Signed-off-by: Brian Norris <[email protected]>
Signed-off-by: Ganapathi Bhat <[email protected]>
---
drivers/net/wireless/marvell/mwifiex/usb.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/net/wireless/marvell/mwifiex/usb.c b/drivers/net/wireless/marvell/mwifiex/usb.c
index bc475b8..88f4c89 100644
--- a/drivers/net/wireless/marvell/mwifiex/usb.c
+++ b/drivers/net/wireless/marvell/mwifiex/usb.c
@@ -644,8 +644,6 @@ static void mwifiex_usb_disconnect(struct usb_interface *intf)
MWIFIEX_FUNC_SHUTDOWN);
}

- mwifiex_usb_free(card);
-
mwifiex_dbg(adapter, FATAL,
"%s: removing card\n", __func__);
mwifiex_remove_card(adapter);
@@ -1353,6 +1351,8 @@ static void mwifiex_unregister_dev(struct mwifiex_adapter *adapter)
{
struct usb_card_rec *card = (struct usb_card_rec *)adapter->card;

+ mwifiex_usb_free(card);
+
mwifiex_usb_cleanup_tx_aggr(adapter);

card->adapter = NULL;
--
1.9.1

2018-06-01 15:56:44

by Kalle Valo

[permalink] [raw]
Subject: Re: [PATCH 1/2] Revert "mwifiex: handle race during mwifiex_usb_disconnect"

Ganapathi Bhat <[email protected]> writes:

> This reverts commit b817047ae70c0bd67b677b65d0d69d72cd6e9728.
>
> We have a better fix for this issue, which will be sent on top
> of this revert.
>
> Signed-off-by: Ganapathi Bhat <[email protected]>

I'll queue both of these patches to 4.18.

--
Kalle Valo

2018-06-01 16:34:00

by Brian Norris

[permalink] [raw]
Subject: Re: [PATCH 2/2] mwifiex: handle race during mwifiex_usb_disconnect

Hi Ganapathi,

On Fri, Jun 01, 2018 at 04:11:20PM +0530, Ganapathi Bhat wrote:
> Race condition is observed during rmmod of mwifiex_usb:
>
> 1. The rmmod thread will call mwifiex_usb_disconnect(), download
> SHUTDOWN command and do wait_event_interruptible_timeout(),
> waiting for response.
>
> 2. The main thread will handle the response and will do a
> wake_up_interruptible(), unblocking rmmod thread.
>
> 3. On getting unblocked, rmmod thread will make rx_cmd.urb = NULL in
> mwifiex_usb_free().
>
> 4. The main thread will try to resubmit rx_cmd.urb in
> mwifiex_usb_submit_rx_urb(), which is NULL.
>
> To fix this, move mwifiex_usb_free() from mwifiex_usb_disconnect
> to mwifiex_unregister_dev(). Function mwifiex_unregister_dev() is
> called after flushing the command and RX work queues.
>
> Signed-off-by: Brian Norris <[email protected]>

^^ I'm not sure if that line is quite accurate. While I nearly spelled
out what the patch would look like, you wrote it.

Anyway, patch seems good to me, assuming it tests out OK for you:

Reviewed-by: Brian Norris <[email protected]>

and if Kalle hasn't applied this yet, an alternative to Signed-off-by:

Suggested-by: Brian Norris <[email protected]>

> Signed-off-by: Ganapathi Bhat <[email protected]>
> ---
> drivers/net/wireless/marvell/mwifiex/usb.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/wireless/marvell/mwifiex/usb.c b/drivers/net/wireless/marvell/mwifiex/usb.c
> index bc475b8..88f4c89 100644
> --- a/drivers/net/wireless/marvell/mwifiex/usb.c
> +++ b/drivers/net/wireless/marvell/mwifiex/usb.c
> @@ -644,8 +644,6 @@ static void mwifiex_usb_disconnect(struct usb_interface *intf)
> MWIFIEX_FUNC_SHUTDOWN);
> }
>
> - mwifiex_usb_free(card);
> -
> mwifiex_dbg(adapter, FATAL,
> "%s: removing card\n", __func__);
> mwifiex_remove_card(adapter);
> @@ -1353,6 +1351,8 @@ static void mwifiex_unregister_dev(struct mwifiex_adapter *adapter)
> {
> struct usb_card_rec *card = (struct usb_card_rec *)adapter->card;
>
> + mwifiex_usb_free(card);
> +
> mwifiex_usb_cleanup_tx_aggr(adapter);
>
> card->adapter = NULL;
> --
> 1.9.1
>

2018-06-18 14:41:28

by Kalle Valo

[permalink] [raw]
Subject: Re: [PATCH 2/2] mwifiex: handle race during mwifiex_usb_disconnect

Brian Norris <[email protected]> writes:

> Hi Ganapathi,
>
> On Fri, Jun 01, 2018 at 04:11:20PM +0530, Ganapathi Bhat wrote:
>> Race condition is observed during rmmod of mwifiex_usb:
>>
>> 1. The rmmod thread will call mwifiex_usb_disconnect(), download
>> SHUTDOWN command and do wait_event_interruptible_timeout(),
>> waiting for response.
>>
>> 2. The main thread will handle the response and will do a
>> wake_up_interruptible(), unblocking rmmod thread.
>>
>> 3. On getting unblocked, rmmod thread will make rx_cmd.urb = NULL in
>> mwifiex_usb_free().
>>
>> 4. The main thread will try to resubmit rx_cmd.urb in
>> mwifiex_usb_submit_rx_urb(), which is NULL.
>>
>> To fix this, move mwifiex_usb_free() from mwifiex_usb_disconnect
>> to mwifiex_unregister_dev(). Function mwifiex_unregister_dev() is
>> called after flushing the command and RX work queues.
>>
>> Signed-off-by: Brian Norris <[email protected]>
>
> ^^ I'm not sure if that line is quite accurate. While I nearly spelled
> out what the patch would look like, you wrote it.
>
> Anyway, patch seems good to me, assuming it tests out OK for you:
>
> Reviewed-by: Brian Norris <[email protected]>
>
> and if Kalle hasn't applied this yet, an alternative to Signed-off-by:
>
> Suggested-by: Brian Norris <[email protected]>

Ok, I'll change that.

--
Kalle Valo

2018-06-24 17:01:08

by Kalle Valo

[permalink] [raw]
Subject: Re: [1/2] Revert "mwifiex: handle race during mwifiex_usb_disconnect"

Ganapathi Bhat <[email protected]> wrote:

> This reverts commit b817047ae70c0bd67b677b65d0d69d72cd6e9728.
>
> We have a better fix for this issue, which will be sent on top
> of this revert.
>
> Signed-off-by: Ganapathi Bhat <[email protected]>

2 patches applied to wireless-drivers.git, thanks.

7e58e741c001 Revert "mwifiex: handle race during mwifiex_usb_disconnect"
f8c095f67962 mwifiex: handle race during mwifiex_usb_disconnect

--
https://patchwork.kernel.org/patch/10442949/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches