diff mbox series

spmi: Add check for remove callback in spmi_drv_remove API

Message ID 1670145780-13111-2-git-send-email-quic_jprakash@quicinc.com
State Accepted
Commit b56eef3e16d888883fefab47425036de80dd38fc
Headers show
Series spmi: Add check for remove callback in spmi_drv_remove API | expand

Commit Message

Jishnu Prakash Dec. 4, 2022, 9:23 a.m. UTC
Add a check for remove callback presence before calling it for a
spmi driver, to avoid NULL pointer dereference error if remove callback
has not been specified for that SPMI driver.

Signed-off-by: Jishnu Prakash <quic_jprakash@quicinc.com>
---
 drivers/spmi/spmi.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Jishnu Prakash Dec. 13, 2022, 1:42 p.m. UTC | #1
Hi Greg

These are two SPMI drivers without remove callbacks defined:

drivers/mfd/qcom-spmi-pmic.c
drivers/mfd/hi6421-spmi-pmic.c

We made this change after noticing an issue internally with the first 
one above, there was a crash when trying to remove it with rmmod, which 
is fixed by this change. In addition, since the probe of the QCOM SPMI 
PMIC driver uses devm_ functions throughout, we could see that with this 
change, when we remove the device with rmmod, the cleanup does happen 
correctly even though there is no remove function defined in the driver. 
The last function called in the probe of our SPMI PMIC driver is 
devm_of_platform_populate(), to probe all the PMIC peripheral drivers 
under this one, and when this driver module was  removed with rmmod, we 
could see that the individual PMIC drivers under it also got depopulated 
with their remove APIs getting called.

If it is possible for a SPMI driver to be removed correctly by rmmod 
without having a remove API defined, this change should be right, what 
do you think?

Thanks,

Jishnu

On 12/13/2022 5:34 PM, Greg KH wrote:
> On Sun, Dec 04, 2022 at 02:53:00PM +0530, Jishnu Prakash wrote:
>> Add a check for remove callback presence before calling it for a
>> spmi driver, to avoid NULL pointer dereference error if remove callback
>> has not been specified for that SPMI driver.
>>
>> Signed-off-by: Jishnu Prakash <quic_jprakash@quicinc.com>
>> ---
>>   drivers/spmi/spmi.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/spmi/spmi.c b/drivers/spmi/spmi.c
>> index a456ce5..6b34356 100644
>> --- a/drivers/spmi/spmi.c
>> +++ b/drivers/spmi/spmi.c
>> @@ -350,7 +350,8 @@ static void spmi_drv_remove(struct device *dev)
>>   	const struct spmi_driver *sdrv = to_spmi_driver(dev->driver);
>>   
>>   	pm_runtime_get_sync(dev);
>> -	sdrv->remove(to_spmi_device(dev));
>> +	if (sdrv->remove)
>> +		sdrv->remove(to_spmi_device(dev));
>>   	pm_runtime_put_noidle(dev);
>>   
>>   	pm_runtime_disable(dev);
> 
> What in-kernel spmi driver does not have a remove function set that
> requires this change?
> 
> thanks,
> 
> greg k-h
Jishnu Prakash Dec. 21, 2022, 5:39 a.m. UTC | #2
Hi Greg

On 12/13/2022 8:39 PM, Greg KH wrote:
> On Tue, Dec 13, 2022 at 07:12:10PM +0530, Jishnu Prakash wrote:
>> Hi Greg
> 
> Hi, please do not top-post :(
> 
>> These are two SPMI drivers without remove callbacks defined:
>>
>> drivers/mfd/qcom-spmi-pmic.c
>> drivers/mfd/hi6421-spmi-pmic.c
> 
> Great, they should be fixed up now, right?
> 

Our QCOM SPMI PMIC driver allocates resources in its probe using only 
devm_() APIs and does not require any other cleanup. It doesn't seem 
right to add an empty remove callback to it just to avoid this crash, it
seems the better solution architecturally is to call the remove function 
only if it's defined.

In addition, I could see that other buses like PCI and AMBA do check for 
the remove API being defined for their drivers before calling it:

AMBA example: 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/amba/bus.c#n328

PCI example: 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/pci-driver.c#n474


>> We made this change after noticing an issue internally with the first one
>> above, there was a crash when trying to remove it with rmmod, which is fixed
>> by this change.
> 
> Then please say that in the changelog text, otherwise we have no idea
> _why_ this is needed.  All you said was "add this new check _IF_" and we
> have no idea what the answer to "if" is :(
> 

I have uploaded the change with an updated title and commit text, can 
you please have a look?

> thanks,
> 
> greg k-h
Thanks,
Jishnu
diff mbox series

Patch

diff --git a/drivers/spmi/spmi.c b/drivers/spmi/spmi.c
index a456ce5..6b34356 100644
--- a/drivers/spmi/spmi.c
+++ b/drivers/spmi/spmi.c
@@ -350,7 +350,8 @@  static void spmi_drv_remove(struct device *dev)
 	const struct spmi_driver *sdrv = to_spmi_driver(dev->driver);
 
 	pm_runtime_get_sync(dev);
-	sdrv->remove(to_spmi_device(dev));
+	if (sdrv->remove)
+		sdrv->remove(to_spmi_device(dev));
 	pm_runtime_put_noidle(dev);
 
 	pm_runtime_disable(dev);