diff mbox

clk: fix a panic error caused by accessing NULL pointer

Message ID 20171120191219.GD18379@codeaurora.org (mailing list archive)
State Not Applicable
Headers show

Commit Message

Stephen Boyd Nov. 20, 2017, 7:12 p.m. UTC
On 11/20, Chunyan Zhang wrote:
> From: Cai Li <cai.li@spreadtrum.com>
> 
> In some cases the clock parent would be set NULL when doing re-parent,
> it will cause a NULL pointer accessing if clk_set trace event is enabled,
> since the trace event function would not check the input parameter.
> 
> Signed-off-by: Cai Li <cai.li@spreadtrum.com>
> Signed-off-by: Chunyan Zhang <chunyan.zhang@spreadtrum.com>

Fixes: tag?

> ---
>  drivers/clk/clk.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> index c8d83ac..64efaf0 100644
> --- a/drivers/clk/clk.c
> +++ b/drivers/clk/clk.c
> @@ -1242,13 +1242,12 @@ static int __clk_set_parent(struct clk_core *core, struct clk_core *parent,
>  
>  	old_parent = __clk_set_parent_before(core, parent);
>  
> -	trace_clk_set_parent(core, parent);
> -
>  	/* change clock input source */
> -	if (parent && core->ops->set_parent)
> +	if (parent && core->ops->set_parent) {
> +		trace_clk_set_parent(core, parent);
>  		ret = core->ops->set_parent(core->hw, p_index);
> -
> -	trace_clk_set_parent_complete(core, parent);
> +		trace_clk_set_parent_complete(core, parent);
> +	}

Is the problem that parent may be NULL and the tracepoint
dereferences it? Perhaps we need to update the tracepoint code
instead so that we always see that the tracepoint is called even
if we don't actually touch the hardware. Something like the patch
below instead.

---8<----

Comments

Chunyan Zhang Nov. 21, 2017, 8:57 a.m. UTC | #1
On 21 November 2017 at 03:12, Stephen Boyd <sboyd@codeaurora.org> wrote:
> On 11/20, Chunyan Zhang wrote:
>> From: Cai Li <cai.li@spreadtrum.com>
>>
>> In some cases the clock parent would be set NULL when doing re-parent,
>> it will cause a NULL pointer accessing if clk_set trace event is enabled,
>> since the trace event function would not check the input parameter.
>>
>> Signed-off-by: Cai Li <cai.li@spreadtrum.com>
>> Signed-off-by: Chunyan Zhang <chunyan.zhang@spreadtrum.com>
>
> Fixes: tag?
>
>> ---
>>  drivers/clk/clk.c | 9 ++++-----
>>  1 file changed, 4 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
>> index c8d83ac..64efaf0 100644
>> --- a/drivers/clk/clk.c
>> +++ b/drivers/clk/clk.c
>> @@ -1242,13 +1242,12 @@ static int __clk_set_parent(struct clk_core *core, struct clk_core *parent,
>>
>>       old_parent = __clk_set_parent_before(core, parent);
>>
>> -     trace_clk_set_parent(core, parent);
>> -
>>       /* change clock input source */
>> -     if (parent && core->ops->set_parent)
>> +     if (parent && core->ops->set_parent) {
>> +             trace_clk_set_parent(core, parent);
>>               ret = core->ops->set_parent(core->hw, p_index);
>> -
>> -     trace_clk_set_parent_complete(core, parent);
>> +             trace_clk_set_parent_complete(core, parent);
>> +     }
>
> Is the problem that parent may be NULL and the tracepoint
> dereferences it?

Yes, I think that probably is uncommon usage though, it revealed that
the tracepoint could be stronger :)

> Perhaps we need to update the tracepoint code
> instead so that we always see that the tracepoint is called even
> if we don't actually touch the hardware. Something like the patch
> below instead.

Ok, we will cook a new patch according to your comments.

Thanks,
Chunyan

>
> ---8<----
> diff --git a/include/trace/events/clk.h b/include/trace/events/clk.h
> index 758607226bfd..5a85ea2090c4 100644
> --- a/include/trace/events/clk.h
> +++ b/include/trace/events/clk.h
> @@ -139,7 +139,7 @@ DECLARE_EVENT_CLASS(clk_parent,
>
>         TP_fast_assign(
>                 __assign_str(name, core->name);
> -               __assign_str(pname, parent->name);
> +               __assign_str(pname, parent ? parent->name : NULL);
>         ),
>
>         TP_printk("%s %s", __get_str(name), __get_str(pname))
> --
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
> a Linux Foundation Collaborative Project
--
To unsubscribe from this list: send the line "unsubscribe linux-clk" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Chunyan Zhang Nov. 21, 2017, 9:21 a.m. UTC | #2
On 21 November 2017 at 16:57, Chunyan Zhang <zhang.lyra@gmail.com> wrote:
> On 21 November 2017 at 03:12, Stephen Boyd <sboyd@codeaurora.org> wrote:
>> On 11/20, Chunyan Zhang wrote:
>>> From: Cai Li <cai.li@spreadtrum.com>
>>>
>>> In some cases the clock parent would be set NULL when doing re-parent,
>>> it will cause a NULL pointer accessing if clk_set trace event is enabled,
>>> since the trace event function would not check the input parameter.
>>>
>>> Signed-off-by: Cai Li <cai.li@spreadtrum.com>
>>> Signed-off-by: Chunyan Zhang <chunyan.zhang@spreadtrum.com>
>>
>> Fixes: tag?
>>
>>> ---
>>>  drivers/clk/clk.c | 9 ++++-----
>>>  1 file changed, 4 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
>>> index c8d83ac..64efaf0 100644
>>> --- a/drivers/clk/clk.c
>>> +++ b/drivers/clk/clk.c
>>> @@ -1242,13 +1242,12 @@ static int __clk_set_parent(struct clk_core *core, struct clk_core *parent,
>>>
>>>       old_parent = __clk_set_parent_before(core, parent);
>>>
>>> -     trace_clk_set_parent(core, parent);
>>> -
>>>       /* change clock input source */
>>> -     if (parent && core->ops->set_parent)
>>> +     if (parent && core->ops->set_parent) {
>>> +             trace_clk_set_parent(core, parent);
>>>               ret = core->ops->set_parent(core->hw, p_index);
>>> -
>>> -     trace_clk_set_parent_complete(core, parent);
>>> +             trace_clk_set_parent_complete(core, parent);
>>> +     }
>>
>> Is the problem that parent may be NULL and the tracepoint
>> dereferences it?
>
> Yes, I think that probably is uncommon usage though, it revealed that
> the tracepoint could be stronger :)

The reason we need to set the parent as NULL is to disable the clk for
the purpose of saving power.

>
>> Perhaps we need to update the tracepoint code
>> instead so that we always see that the tracepoint is called even
>> if we don't actually touch the hardware. Something like the patch
>> below instead.
>
> Ok, we will cook a new patch according to your comments.
>
> Thanks,
> Chunyan
>
>>
>> ---8<----
>> diff --git a/include/trace/events/clk.h b/include/trace/events/clk.h
>> index 758607226bfd..5a85ea2090c4 100644
>> --- a/include/trace/events/clk.h
>> +++ b/include/trace/events/clk.h
>> @@ -139,7 +139,7 @@ DECLARE_EVENT_CLASS(clk_parent,
>>
>>         TP_fast_assign(
>>                 __assign_str(name, core->name);
>> -               __assign_str(pname, parent->name);
>> +               __assign_str(pname, parent ? parent->name : NULL);
>>         ),
>>
>>         TP_printk("%s %s", __get_str(name), __get_str(pname))
>> --
>> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
>> a Linux Foundation Collaborative Project
--
To unsubscribe from this list: send the line "unsubscribe linux-clk" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Stephen Boyd Dec. 5, 2017, 11:28 p.m. UTC | #3
On 11/21, Chunyan Zhang wrote:
> On 21 November 2017 at 16:57, Chunyan Zhang <zhang.lyra@gmail.com> wrote:
> > On 21 November 2017 at 03:12, Stephen Boyd <sboyd@codeaurora.org> wrote:
> >> On 11/20, Chunyan Zhang wrote:
> >>> From: Cai Li <cai.li@spreadtrum.com>
> >>>
> >>> In some cases the clock parent would be set NULL when doing re-parent,
> >>> it will cause a NULL pointer accessing if clk_set trace event is enabled,
> >>> since the trace event function would not check the input parameter.
> >>>
> >>> Signed-off-by: Cai Li <cai.li@spreadtrum.com>
> >>> Signed-off-by: Chunyan Zhang <chunyan.zhang@spreadtrum.com>
> >>
> >> Fixes: tag?
> >>
> >>> ---
> >>>  drivers/clk/clk.c | 9 ++++-----
> >>>  1 file changed, 4 insertions(+), 5 deletions(-)
> >>>
> >>> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> >>> index c8d83ac..64efaf0 100644
> >>> --- a/drivers/clk/clk.c
> >>> +++ b/drivers/clk/clk.c
> >>> @@ -1242,13 +1242,12 @@ static int __clk_set_parent(struct clk_core *core, struct clk_core *parent,
> >>>
> >>>       old_parent = __clk_set_parent_before(core, parent);
> >>>
> >>> -     trace_clk_set_parent(core, parent);
> >>> -
> >>>       /* change clock input source */
> >>> -     if (parent && core->ops->set_parent)
> >>> +     if (parent && core->ops->set_parent) {
> >>> +             trace_clk_set_parent(core, parent);
> >>>               ret = core->ops->set_parent(core->hw, p_index);
> >>> -
> >>> -     trace_clk_set_parent_complete(core, parent);
> >>> +             trace_clk_set_parent_complete(core, parent);
> >>> +     }
> >>
> >> Is the problem that parent may be NULL and the tracepoint
> >> dereferences it?
> >
> > Yes, I think that probably is uncommon usage though, it revealed that
> > the tracepoint could be stronger :)
> 
> The reason we need to set the parent as NULL is to disable the clk for
> the purpose of saving power.

Do you have drivers calling set_parent with NULL to save power?
Seems sort of odd. Why not do something when disabling the clk in
clk_disable() path instead? Either way, I'll apply the patch to
clk-fixes.
Chunyan Zhang Dec. 6, 2017, 7:38 a.m. UTC | #4
On 6 December 2017 at 07:28, Stephen Boyd <sboyd@codeaurora.org> wrote:
> On 11/21, Chunyan Zhang wrote:
>> On 21 November 2017 at 16:57, Chunyan Zhang <zhang.lyra@gmail.com> wrote:
>> > On 21 November 2017 at 03:12, Stephen Boyd <sboyd@codeaurora.org> wrote:
>> >> On 11/20, Chunyan Zhang wrote:
>> >>> From: Cai Li <cai.li@spreadtrum.com>
>> >>>
>> >>> In some cases the clock parent would be set NULL when doing re-parent,
>> >>> it will cause a NULL pointer accessing if clk_set trace event is enabled,
>> >>> since the trace event function would not check the input parameter.
>> >>>
>> >>> Signed-off-by: Cai Li <cai.li@spreadtrum.com>
>> >>> Signed-off-by: Chunyan Zhang <chunyan.zhang@spreadtrum.com>
>> >>
>> >> Fixes: tag?
>> >>
>> >>> ---
>> >>>  drivers/clk/clk.c | 9 ++++-----
>> >>>  1 file changed, 4 insertions(+), 5 deletions(-)
>> >>>
>> >>> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
>> >>> index c8d83ac..64efaf0 100644
>> >>> --- a/drivers/clk/clk.c
>> >>> +++ b/drivers/clk/clk.c
>> >>> @@ -1242,13 +1242,12 @@ static int __clk_set_parent(struct clk_core *core, struct clk_core *parent,
>> >>>
>> >>>       old_parent = __clk_set_parent_before(core, parent);
>> >>>
>> >>> -     trace_clk_set_parent(core, parent);
>> >>> -
>> >>>       /* change clock input source */
>> >>> -     if (parent && core->ops->set_parent)
>> >>> +     if (parent && core->ops->set_parent) {
>> >>> +             trace_clk_set_parent(core, parent);
>> >>>               ret = core->ops->set_parent(core->hw, p_index);
>> >>> -
>> >>> -     trace_clk_set_parent_complete(core, parent);
>> >>> +             trace_clk_set_parent_complete(core, parent);
>> >>> +     }
>> >>
>> >> Is the problem that parent may be NULL and the tracepoint
>> >> dereferences it?
>> >
>> > Yes, I think that probably is uncommon usage though, it revealed that
>> > the tracepoint could be stronger :)
>>
>> The reason we need to set the parent as NULL is to disable the clk for
>> the purpose of saving power.
>
> Do you have drivers calling set_parent with NULL to save power?
> Seems sort of odd. Why not do something when disabling the clk in

It's a mux clk in the same address area with a device, when powerring
off the device module, the registers of this mux are reset with
default value by hardware.  When the device resumed afterward, the mux
clock needs to recover its previous clk parent, but the clk core would
find it is setting a same clk parent, the function
clk_core_set_parent() [1] will return directly without writing the
exact register to set the parent again.

So setting the clk parent with NULL is just a workaround, we will
appreciate if you have other better solution for us.

Thanks,
Chunyan

[1] http://elixir.free-electrons.com/linux/v4.15-rc1/source/drivers/clk/clk.c#L1886

> clk_disable() path instead? Either way, I'll apply the patch to
> clk-fixes.
>
> --
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
> a Linux Foundation Collaborative Project
--
To unsubscribe from this list: send the line "unsubscribe linux-clk" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/include/trace/events/clk.h b/include/trace/events/clk.h
index 758607226bfd..5a85ea2090c4 100644
--- a/include/trace/events/clk.h
+++ b/include/trace/events/clk.h
@@ -139,7 +139,7 @@  DECLARE_EVENT_CLASS(clk_parent,
 
 	TP_fast_assign(
 		__assign_str(name, core->name);
-		__assign_str(pname, parent->name);
+		__assign_str(pname, parent ? parent->name : NULL);
 	),
 
 	TP_printk("%s %s", __get_str(name), __get_str(pname))