Re: [PATCH 5/5] platform/x86: think-lmi: mutex protection around multiple WMI calls

["Mark Pearson" <mpearson-lenovo@squebb.ca>]

On Fri, May 26, 2023, at 4:12 AM, Hans de Goede wrote:
> Hi,
>
> On 5/25/23 21:50, Mark Pearson wrote:
>> 
>> 
>> On Thu, May 25, 2023, at 3:41 PM, Hans de Goede wrote:
>>> Hi Mark,
>>>
>>> On 5/25/23 21:31, Mark Pearson wrote:
>>>> Add mutex protection around cases where an operation needs multiple
>>>> WMI calls - e.g. setting password.
>>>>
>>>> Signed-off-by: Mark Pearson <mpearson-lenovo@squebb.ca>
>>>> ---
>>>> Changes in V2: New commit added after review of other patches in series.
>>>>
>>>>  drivers/platform/x86/think-lmi.c | 46 ++++++++++++++++++++++++--------
>>>>  1 file changed, 35 insertions(+), 11 deletions(-)
>>>>
>>>> diff --git a/drivers/platform/x86/think-lmi.c b/drivers/platform/x86/think-lmi.c
>>>> index 64cd453d6e7d..f3e1e4dacba2 100644
>>>> --- a/drivers/platform/x86/think-lmi.c
>>>> +++ b/drivers/platform/x86/think-lmi.c
>>>> @@ -14,6 +14,7 @@
>>>>  #include <linux/acpi.h>
>>>>  #include <linux/errno.h>
>>>>  #include <linux/fs.h>
>>>> +#include <linux/mutex.h>
>>>>  #include <linux/string.h>
>>>>  #include <linux/types.h>
>>>>  #include <linux/dmi.h>
>>>> @@ -195,6 +196,7 @@ static const char * const level_options[] = {
>>>>  };
>>>>  static struct think_lmi tlmi_priv;
>>>>  static struct class *fw_attr_class;
>>>> +static DEFINE_MUTEX(tlmi_mutex);
>>>>  
>>>>  /* ------ Utility functions ------------*/
>>>>  /* Strip out CR if one is present */
>>>> @@ -463,23 +465,32 @@ static ssize_t new_password_store(struct kobject *kobj,
>>>>  			sprintf(pwd_type, "%s", setting->pwd_type);
>>>>  		}
>>>>  
>>>> +		mutex_lock(&tlmi_mutex);
>>>>  		ret = tlmi_opcode_setting("WmiOpcodePasswordType", pwd_type);
>>>> -		if (ret)
>>>> +		if (ret) {
>>>> +			mutex_unlock(&tlmi_mutex);
>>>>  			goto out;
>>>> -
>>>> +		}
>>>>  		if (tlmi_priv.pwd_admin->valid) {
>>>>  			ret = tlmi_opcode_setting("WmiOpcodePasswordAdmin",
>>>>  					tlmi_priv.pwd_admin->password);
>>>> -			if (ret)
>>>> +			if (ret) {
>>>> +				mutex_unlock(&tlmi_mutex);
>>>>  				goto out;
>>>> +			}
>>>>  		}
>>>>  		ret = tlmi_opcode_setting("WmiOpcodePasswordCurrent01", setting->password);
>>>> -		if (ret)
>>>> +		if (ret) {
>>>> +			mutex_unlock(&tlmi_mutex);
>>>>  			goto out;
>>>> +		}
>>>>  		ret = tlmi_opcode_setting("WmiOpcodePasswordNew01", new_pwd);
>>>> -		if (ret)
>>>> +		if (ret) {
>>>> +			mutex_unlock(&tlmi_mutex);
>>>>  			goto out;
>>>> +		}
>>>>  		ret = tlmi_simple_call(LENOVO_OPCODE_IF_GUID, "WmiOpcodePasswordSetUpdate;");
>>>> +		mutex_unlock(&tlmi_mutex);
>>>>  	} else {
>>>>  		/* Format: 'PasswordType,CurrentPw,NewPw,Encoding,KbdLang;' */
>>>>  		auth_str = kasprintf(GFP_KERNEL, "%s,%s,%s,%s,%s;",
>>>
>>>
>>> I haven't take a really close / good look yet. But at a first glance
>>> I think it would be cleaner to just take the mutex at the top
>>> and unlock it after the out label to which all the existing goto-s
>>> already go ?
>>>
>> I did consider that - and it was in my first implementation; but then I got concerned
>> about if the mutex_unlock could potentially get called without mutex_lock having been 
>> called beforehand. I couldn't find any good reference as to whether that was safe or not.
>> 
>> I ended up deciding that a few extra brackets and unlock calls wasn't that ugly and was 'safer'...and 
>> so went that route.
>> 
>> Happy to change it - but do you happen to know if it's safe to call unlock without a lock? If it is then
>> that implementation is cleaner.
>
> It is not allowed to unlock without a lock. But if you put the lock 
> directly after the malloc for which the out: does the free then there 
> should be no goto out paths which don't have the lock.
>
> E.g. for new_password_store() put it here:
>
>         new_pwd = kstrdup(buf, GFP_KERNEL);
>         if (!new_pwd)
>                 return -ENOMEM;
>
> 	mutex_lock(&tlmi_mutex);
>
> 	/* Strip out CR if one is present, setting password won't work if it 
> is present */
> 	...
>
> This does mean also taking the lock in the case where the new password 
> store is done with a single WMI call, but that is not an issue. It 
> makes things a tiny bit slower but WMI calls already are not fast and 
> it is not like we are going to change the password / settings 100-times 
> per second.
>
> And the same thing can be done in current_value_store():
>
>         new_setting = kstrdup(buf, GFP_KERNEL);
>         if (!new_setting)
>                 return -ENOMEM;
>
> 	mutex_lock(&tlmi_mutex);
>
>         /* Strip out CR if one is present */
>         ...
>

Yeah - you're right.
For some reason I was trying to do the lock only in the block of code that needed locking...but it makes more sense to do it earlier. I'll update.
Thanks!
Mark