Re: [PATCH] gcov: Add support for checksum field

["Martin Liška" <mliska@suse.cz>]

On 12/19/22 17:48, Peter Oberparleiter wrote:
> On 19.12.2022 16:06, Rickard Andersson wrote:
>> From: Rickard x Andersson <rickaran@axis.com>
>>
>> In GCC version 12.1 a checksum field was added.
> 
> Thanks for the patch!
> 
> In another e-mail you mentioned that this patch fixes a kernel crash
> during boot when using gcov-kernel with GCC 12. Please add this
> information to the commit message and if possible the platform on which
> this occurs.
> 
> Also this patch fixes a missing piece from a previous patch, so please add:
> 
> Fixes: 977ef30a7d88 ("gcov: support GCC 12.1 and newer compilers")
> Cc: <stable@vger.kernel.org>
> 
> Finally I reviewed and tested the patch and it looks good to me, so
> please add:
> 
> Reviewed-by: Peter Oberparleiter <oberpar@linux.ibm.com>
> Tested-by: Peter Oberparleiter <oberpar@linux.ibm.com>
> 
> Please resend with these commit message changes. Thanks!
> 
> For the record: I wondered why my testing of the previous patch with GCC
> 12 didn't catch this. It turns out that this crash does not occur on
> architectures with 8-byte pointer alignment such as s390x where I
> performed my tests. Consider this pahole output on s390x without the patch:
> 
> struct gcov_info {
> [...]
>         unsigned int       stamp;                /*    16     4 */
>         /* XXX 4 bytes hole, try to pack */
>         const char  *      filename;             /*    24     8 */
> [...]
> }
> 
> And with the patch:
> 
> struct gcov_info {
> [...]
>        unsigned int       stamp;                /*    16     4 */
>        unsigned int       checksum;             /*    20     4 */
>        const char  *      filename;             /*    24     8 */
> [...]
> }
> 
> As can be seen, the offset of the filename and subsequent fields does
> not change because the new field fills an alignment hole. It would
> change (resulting in a crash during boot) if the alignment-requirement
> of the const char *filename field would be different.

Hello.

Thank you Peter for the deep analysis and it finally explains why I didn't notice
while I was working on 977ef30a7d88.

Please add my review:

Reviewed-By: Martin Liska <mliska@suse.cz>

Thanks,
Martin

> 
>>
>> Signed-off-by: Rickard x Andersson <rickaran@axis.com>
>> ---
>>  kernel/gcov/gcc_4_7.c | 5 +++++
>>  1 file changed, 5 insertions(+)
>>
>> diff --git a/kernel/gcov/gcc_4_7.c b/kernel/gcov/gcc_4_7.c
>> index c699feda21ac..04880d8fba25 100644
>> --- a/kernel/gcov/gcc_4_7.c
>> +++ b/kernel/gcov/gcc_4_7.c
>> @@ -85,6 +85,7 @@ struct gcov_fn_info {
>>   * @version: gcov version magic indicating the gcc version used for compilation
>>   * @next: list head for a singly-linked list
>>   * @stamp: uniquifying time stamp
>> + * @checksum: unique object checksum
>>   * @filename: name of the associated gcov data file
>>   * @merge: merge functions (null for unused counter type)
>>   * @n_functions: number of instrumented functions
>> @@ -97,6 +98,10 @@ struct gcov_info {
>>  	unsigned int version;
>>  	struct gcov_info *next;
>>  	unsigned int stamp;
>> + /* Since GCC 12.1 a checksum field is added. */
>> +#if (__GNUC__ >= 12)
>> +	unsigned int checksum;
>> +#endif
>>  	const char *filename;
>>  	void (*merge[GCOV_COUNTERS])(gcov_type *, unsigned int);
>>  	unsigned int n_functions;
>