powerpc: Fix unitialized variable

On 11-12-2014 10:34, Adhemerval Zanella wrote:
> On 11-12-2014 09:58, Stefan Liebler wrote:
>> On 12/11/2014 12:54 PM, Andreas Schwab wrote:
>>> Stefan Liebler <stli@linux.vnet.ibm.com> writes:
>>>
>>>> +    /* There´s the following warning on S390:
>>> Please avoid non-ascii.
>>>
>>> Andreas.
>>>
>> Oh. Ok. Here is it without ´.
> I was about to send a similar patch.
>
>> ---
>> 2014-12-11  Stefan Liebler  <stli@linux.vnet.ibm.com>
>>
>>     resolv/res_send.c (send_vc): Disable warning resplen may
>>     be used uninitialized.
>>
>> res_send_11122014
>>
>>
>> diff --git a/resolv/res_send.c b/resolv/res_send.c
>> index af42b8a..22b322c 100644
>> --- a/resolv/res_send.c
>> +++ b/resolv/res_send.c
>> @@ -96,6 +96,7 @@ static const char rcsid[] = "$BINDId: res_send.c,v 8.38 2000/03/30 20:16:51 vixi
>>  #include <string.h>
>>  #include <unistd.h>
>>  #include <kernel-features.h>
>> +#include <libc-internal.h>
>>  
>>  #if PACKETSZ > 65536
>>  #define MAXPACKET       PACKETSZ
>> @@ -668,7 +669,17 @@ send_vc(res_state statp,
>>  	// int anssiz = *anssizp;
>>  	HEADER *anhp = (HEADER *) ans;
>>  	struct sockaddr_in6 *nsap = EXT(statp).nsaddrs[ns];
>> -	int truncating, connreset, resplen, n;
>> +	int truncating, connreset, n;
>> +	/* There is the following warning on S390:
>> +	   'resplen' may be used uninitialized in this function
>> +	   [-Wmaybe-uninitialized]
>> +	   This is a false positive according to:
>> +	   https://www.sourceware.org/ml/libc-alpha/2014-12/msg00323.html
>> +	 */
>> +	DIAG_PUSH_NEEDS_COMMENT;
>> +	DIAG_IGNORE_NEEDS_COMMENT (5, "-Wmaybe-uninitialized");
>> +	int resplen;
>> +	DIAG_POP_NEEDS_COMMENT;
> I would a comment saying it the compiler emits if architecture does not 
> define '_STRING_ARCH_unaligned', instead of just for S390. 
>
>>  	struct iovec iov[4];
>>  	u_short len;
>>  	u_short len2;
>> @@ -788,8 +799,11 @@ send_vc(res_state statp,
>>  			   reply.  We can try to use the rest
>>  			   of the user-provided buffer.  */
>>  #if _STRING_ARCH_unaligned
>> +			DIAG_PUSH_NEEDS_COMMENT;
>> +			DIAG_IGNORE_NEEDS_COMMENT (5, "-Wmaybe-uninitialized");
>>  			*anssizp2 = orig_anssizp - resplen;
>>  			*ansp2 = *ansp + resplen;
>> +			DIAG_POP_NEEDS_COMMENT;
> I am seeing the may be used uninitialized with GCC 4.7 and forward.
>
>>  #else
>>  			int aligned_resplen
>>  			  = ((resplen + __alignof__ (HEADER) - 1)

Here it is a more comprehensible patch that also suppress the warning for
powerpc (and possible other arches that do not define _STRING_ARCH_unaligned),
what do you think?

--

On Thu, 11 Dec 2014, Adhemerval Zanella wrote:

> +       int truncating, connreset, n;
> +       /* There is the following warning on some architectures:
> +          'resplen' may be used uninitialized in this function
> +          [-Wmaybe-uninitialized]
> +          This is a false positive according to:
> +          https://www.sourceware.org/ml/libc-alpha/2014-12/msg00323.html
> +        */
> +       DIAG_PUSH_NEEDS_COMMENT;
> +       DIAG_IGNORE_NEEDS_COMMENT (4.7, "-Wmaybe-uninitialized");
> +       int resplen;
> +       DIAG_POP_NEEDS_COMMENT;

* Do you actually need this here, or only around the use of the variable?

* An actual analysis of why the variable can't be used uninitialized would 
be better than a URL.  I.e., if buf2 == NULL then this code won't be 
executed; if buf2 != NULL, then first time round the loop recvresp1 and 
recvresp2 will be 0 so this code won't be executed but "thisresplenp = 
&resplen;" followed by "*thisresplenp = rlen;" will be executed so that 
subsequent times round the loop resplen has been initialized.

* The version number in DIAG_IGNORE_NEEDS_COMMENT is the most recent GCC 
version with which the issue has been observed, not the oldest.

* A conditional __GNUC_PREREQ (4, 7) is needed around the 
DIAG_IGNORE_NEEDS_COMMENT call because 4.6 doesn't have 
-Wmaybe-uninitialized (if the warnings appear with 4.6, a #else case to 
use -Wuninitialized instead with 4.6 will be needed).