sis5595_pci_probe() registers platform driver callbacks and just then
initializes global pointer variable s_bridge. sis5595_probe() may
dereference it before this happens that can result in null pointer
dereference.
We can not swap registration of platform driver callbacks with
initialization of s_bridge since sm_sis5595_exit() assumes the
current order. Perhaps it has sense to implement a pci_driver.remove
callback that will take care about deregistration of platform driver
callbacks.
Found by Linux Driver Verification project (linuxtesting.org).
On 8/4/20 4:50 AM, Evgeny Novikov wrote:
> sis5595_pci_probe() registers platform driver callbacks and just then
> initializes global pointer variable s_bridge. sis5595_probe() may
> dereference it before this happens that can result in null pointer
> dereference.
>
sis5595_probe() is only called after the device is registered,
which happens in sis5595_device_add() after s_bridge is set. This is
a southbridge, so there won't be any hot insertion/removal events.
> We can not swap registration of platform driver callbacks with
> initialization of s_bridge since sm_sis5595_exit() assumes the
> current order. Perhaps it has sense to implement a pci_driver.remove
> callback that will take care about deregistration of platform driver
> callbacks.
>
Agreed regarding the remove function. However, given the age of the chip,
I'd rather remove the driver than spending time on cleanup efforts.
This looks like a perfect candidate for depreciation.
Guenter
04.08.2020, 17:33, "Guenter Roeck" <[email protected]>:
> On 8/4/20 4:50 AM, Evgeny Novikov wrote:
>> sis5595_pci_probe() registers platform driver callbacks and just then
>> initializes global pointer variable s_bridge. sis5595_probe() may
>> dereference it before this happens that can result in null pointer
>> dereference.
>
> sis5595_probe() is only called after the device is registered,
> which happens in sis5595_device_add() after s_bridge is set. This is
> a southbridge, so there won't be any hot insertion/removal events.
>
Thank you for this hint. We need to tune our models appropriately.
>> We can not swap registration of platform driver callbacks with
>> initialization of s_bridge since sm_sis5595_exit() assumes the
>> current order. Perhaps it has sense to implement a pci_driver.remove
>> callback that will take care about deregistration of platform driver
>> callbacks.
>
> Agreed regarding the remove function. However, given the age of the chip,
> I'd rather remove the driver than spending time on cleanup efforts.
> This looks like a perfect candidate for depreciation.
>
This is completely up to you. Anyway the driver does not have the bug.
Best regards,
Evgeny
> Guenter