[v2,0/2] wifi: ath11k: fix event locking

  RCU lockdep reported suspicious RCU usage when accessing the temperature
sensor. Inspection revealed that the DFS radar event code was also
missing the required RCU read-side critical section marking.

Johan


Changes in v2
 - add the missing rcu_read_unlock() to an
   ath11k_wmi_pdev_temperature_event() error path as noticed by Jeff


Johan Hovold (2):
  wifi: ath11k: fix temperature event locking
  wifi: ath11k: fix dfs radar event locking

 drivers/net/wireless/ath/ath11k/wmi.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)
  

Johan Hovold <johan+linaro@kernel.org> writes:

> RCU lockdep reported suspicious RCU usage when accessing the temperature
> sensor. Inspection revealed that the DFS radar event code was also
> missing the required RCU read-side critical section marking.
>
> Johan
>
>
> Changes in v2
>  - add the missing rcu_read_unlock() to an
>    ath11k_wmi_pdev_temperature_event() error path as noticed by Jeff
>
>
> Johan Hovold (2):
>   wifi: ath11k: fix temperature event locking
>   wifi: ath11k: fix dfs radar event locking

Thanks for the fixes. I really like using lockdep_assert_held() to
document if a function requires some lock held, is there anything
similar for RCU?
  

On Tue, Oct 24, 2023 at 05:07:38PM +0300, Kalle Valo wrote:
> Johan Hovold <johan+linaro@kernel.org> writes:
> 
> > RCU lockdep reported suspicious RCU usage when accessing the temperature
> > sensor. Inspection revealed that the DFS radar event code was also
> > missing the required RCU read-side critical section marking.
> >
> > Johan
> >
> >
> > Changes in v2
> >  - add the missing rcu_read_unlock() to an
> >    ath11k_wmi_pdev_temperature_event() error path as noticed by Jeff
> >
> >
> > Johan Hovold (2):
> >   wifi: ath11k: fix temperature event locking
> >   wifi: ath11k: fix dfs radar event locking
> 
> Thanks for the fixes. I really like using lockdep_assert_held() to
> document if a function requires some lock held, is there anything
> similar for RCU?

Not really, but the checking is instead built into the primitives like
rcu_dereference() and enabled whenever CONFIG_PROVE_RCU is set.

For some special cases, we have open-coded checks like:

	RCU_LOCKDEP_WARN(!rcu_read_lock_held());

which similarly depend on CONFIG_PROVE_RCU or simply

	WARN_ON_ONCE(!rcu_read_lock_held());

Johan
  

(old discussion, changing title)

Johan Hovold <johan@kernel.org> writes:

> On Tue, Oct 24, 2023 at 05:07:38PM +0300, Kalle Valo wrote:
>
>> Johan Hovold <johan+linaro@kernel.org> writes:
>> 
>> > RCU lockdep reported suspicious RCU usage when accessing the temperature
>> > sensor. Inspection revealed that the DFS radar event code was also
>> > missing the required RCU read-side critical section marking.
>> >
>> > Johan
>> >
>> >
>> > Changes in v2
>> >  - add the missing rcu_read_unlock() to an
>> >    ath11k_wmi_pdev_temperature_event() error path as noticed by Jeff
>> >
>> >
>> > Johan Hovold (2):
>> >   wifi: ath11k: fix temperature event locking
>> >   wifi: ath11k: fix dfs radar event locking
>> 
>> Thanks for the fixes. I really like using lockdep_assert_held() to
>> document if a function requires some lock held, is there anything
>> similar for RCU?
>
> Not really, but the checking is instead built into the primitives like
> rcu_dereference() and enabled whenever CONFIG_PROVE_RCU is set.
>
> For some special cases, we have open-coded checks like:
>
> 	RCU_LOCKDEP_WARN(!rcu_read_lock_held());
>
> which similarly depend on CONFIG_PROVE_RCU or simply
>
> 	WARN_ON_ONCE(!rcu_read_lock_held());

I just found out that sparse has __must_hold():

https://lore.kernel.org/linux-wireless/87sf31hhfp.fsf@kernel.org/

That looks promising, should we start using that in ath11k and ath12k to
check our RCU usage?
  

Hi Kalle,

On Sat, Jan 13, 2024 at 12:26:53PM +0200, Kalle Valo wrote:
> (old discussion, changing title)
> 
> Johan Hovold <johan@kernel.org> writes:
> 
> > On Tue, Oct 24, 2023 at 05:07:38PM +0300, Kalle Valo wrote:

> >> Thanks for the fixes. I really like using lockdep_assert_held() to
> >> document if a function requires some lock held, is there anything
> >> similar for RCU?
> >
> > Not really, but the checking is instead built into the primitives like
> > rcu_dereference() and enabled whenever CONFIG_PROVE_RCU is set.
> >
> > For some special cases, we have open-coded checks like:
> >
> > 	RCU_LOCKDEP_WARN(!rcu_read_lock_held());
> >
> > which similarly depend on CONFIG_PROVE_RCU or simply
> >
> > 	WARN_ON_ONCE(!rcu_read_lock_held());
> 
> I just found out that sparse has __must_hold():
> 
> https://lore.kernel.org/linux-wireless/87sf31hhfp.fsf@kernel.org/
> 
> That looks promising, should we start using that in ath11k and ath12k to
> check our RCU usage?

I see that Johannes already commented on this in the thread above.

I'm pretty sure smatch can't be used for this.

Johan
  

On Mon, Jan 22, 2024 at 02:10:17PM +0100, Johan Hovold wrote:
> On Sat, Jan 13, 2024 at 12:26:53PM +0200, Kalle Valo wrote:

> > I just found out that sparse has __must_hold():
> > 
> > https://lore.kernel.org/linux-wireless/87sf31hhfp.fsf@kernel.org/
> > 
> > That looks promising, should we start using that in ath11k and ath12k to
> > check our RCU usage?
> 
> I see that Johannes already commented on this in the thread above.
> 
> I'm pretty sure smatch can't be used for this.

I meant "sparse"...

Johan
  

Johan Hovold <johan@kernel.org> writes:

> On Mon, Jan 22, 2024 at 02:10:17PM +0100, Johan Hovold wrote:
>> On Sat, Jan 13, 2024 at 12:26:53PM +0200, Kalle Valo wrote:
>
>> > I just found out that sparse has __must_hold():
>> > 
>> > https://lore.kernel.org/linux-wireless/87sf31hhfp.fsf@kernel.org/
>> > 
>> > That looks promising, should we start using that in ath11k and ath12k to
>> > check our RCU usage?
>> 
>> I see that Johannes already commented on this in the thread above.
>> 
>> I'm pretty sure smatch can't be used for this.
>
> I meant "sparse"...

Yeah, that was a disappointment. I should have tested it first :)