Hi!
There is a potential data race in
drivers/net/wireless/realtek/rtlwifi/rtl8188ee/rtl8188ee.ko.
In the function rtl88ee_gpio_radio_on_off_checking the flag
ppsc->rfchange_inprogress is set with a spinlock protection. In the
function rtl_ps_set_rf_state the flag is read also under a spinlock. But
the function rtl88e_dm_watchdog read it without any locks. As a result
rtl88e_dm_watchdog may execute the succeeding code while changing (with
the flag rfchange_inprogress == true). I do not exactly determine the
consequences, but likely they are not good if there exists such check.
Could anybody more confident confirm this?
The function rtl_ps_set_rf_state is always called with its parameter
[protect_or_not == false]. Is this flag really necessary, if the value
'true' is never used? The function is also set the flag
ppsc->rfchange_inprogress and may affect the rtl88e_dm_watchdog as in
the previous case.
--
Pavel Andrianov
Linux Verification Center, ISPRAS
web: http://linuxtesting.org
e-mail: [email protected]
On Friday 10 June 2016 01:51 PM, Pavel Andrianov wrote:
> Hi!
>
> There is a potential data race in drivers/net/wireless/realtek/rtlwifi/rtl8188ee/rtl8188ee.ko.
>
> In the function rtl88ee_gpio_radio_on_off_checking the flag ppsc->rfchange_inprogress is set with a spinlock protection. In the function rtl_ps_set_rf_state the flag is read also under a spinlock. But the function rtl88e_dm_watchdog read it without any locks. As a result rtl88e_dm_watchdog may execute the succeeding code while changing (with the flag rfchange_inprogress == true). I do not exactly determine the consequences, but likely they are not good if there exists such check. Could anybody more confident confirm this?
>
> The function rtl_ps_set_rf_state is always called with its parameter [protect_or_not == false]. Is this flag really necessary, if the value 'true' is never used? The function is also set the flag ppsc->rfchange_inprogress and may affect the rtl88e_dm_watchdog as in the previous case.
I think the patch was sent sometime ago for removing the parameter. But I am not sure why it's not applied.
May be Larry can have better idea about this.
Here, is link to the patch: http://linux-wireless.vger.kernel.narkive.com/mu4t9xxr/patch-3-4-rtlwifi-rtl8192cu-remove-unused-parameter
--
Vaishali
On 06/24/2016 09:17 AM, Vaishali Thakkar wrote:
>
>
> On Friday 10 June 2016 01:51 PM, Pavel Andrianov wrote:
>> Hi!
>>
>> There is a potential data race in drivers/net/wireless/realtek/rtlwifi/rtl8188ee/rtl8188ee.ko.
>>
>> In the function rtl88ee_gpio_radio_on_off_checking the flag ppsc->rfchange_inprogress is set with a spinlock protection. In the function rtl_ps_set_rf_state the flag is read also under a spinlock. But the function rtl88e_dm_watchdog read it without any locks. As a result rtl88e_dm_watchdog may execute the succeeding code while changing (with the flag rfchange_inprogress == true). I do not exactly determine the consequences, but likely they are not good if there exists such check. Could anybody more confident confirm this?
>>
>> The function rtl_ps_set_rf_state is always called with its parameter [protect_or_not == false]. Is this flag really necessary, if the value 'true' is never used? The function is also set the flag ppsc->rfchange_inprogress and may affect the rtl88e_dm_watchdog as in the previous case.
>
> I think the patch was sent sometime ago for removing the parameter. But I am not sure why it's not applied.
> May be Larry can have better idea about this.
>
> Here, is link to the patch: http://linux-wireless.vger.kernel.narkive.com/mu4t9xxr/patch-3-4-rtlwifi-rtl8192cu-remove-unused-parameter
The patch for rtl8192cu was applied as commit 4b9d8d67b44a on Jun 20 2011, but
the unused parameter was reintroduced as part of an update of the power-save
code with commit d3feae41a347 on Sep 22 2014. My recollection is that Realtek
envisioned a driver that needed this parameter to be true. As none has yet been
introduced, I will prepare a patch to remove it again.
I am also testing a patch to remove the race condition in rtl8188ee.
Larry