[PATCH] uqmi: fix compilation with GCC12

[Hauke Mehrtens]

On 6/11/22 16:47, Bjørn Mork wrote:
> e9hack <e9hack at gmail.com> writes:
> 
>> The 'dangling pointer' issue can be fix without using malloc().
>>
>> --- a/dev.c	2022-05-04 02:18:17.000000000 +0200
>> +++ b/dev.c	2022-06-11 08:48:21.185567953 +0200
>> @@ -206,6 +206,7 @@ void qmi_request_cancel(struct qmi_dev *
>>   int qmi_request_wait(struct qmi_dev *qmi, struct qmi_request *req)
>>   {
>>   	bool complete = false;
>> +	bool *c;
>>   	bool cancelled;
>>     	if (!req->pending)
>> @@ -226,8 +227,10 @@ int qmi_request_wait(struct qmi_dev *qmi
>>   		uloop_cancelled = cancelled;
>>   	}
>>   -	if (req->complete == &complete)
>> -		req->complete = NULL;
>> +	c = req->complete;
>> +	req->complete = NULL;
>> +	if (c != &complete)
>> +		req->complete = c;
>>     	return req->ret;
>>   }
> 
> How about just fixing GCC instead?  Having all sorts of funny and
> pointless code just to avoid bogus compiler warnings is kind of stupid,
> isn't it?
> 
> If GCC can't do better that this, then it's much better to disable that
> warning.

GCC complains about this code because the pointer is only removed 
conditionally
-------------------------------------------
	if (req->complete == &complete)
		req->complete = NULL;
-------------------------------------------
https://git.openwrt.org/?p=project/uqmi.git;a=blob;f=dev.c;h=bd1020790f844fd364fd753135acd8f53f34d996;hb=HEAD#l229

When you make this part unconditionally it does not complain any more.

Is it possible that something changes the req->complete pointer address 
in between?

Hauke