svn commit: r365445 - head/sys/cam/mmc

[Bjoern A. Zeeb]

On 8 Sep 2020, at 7:10, Andriy Gapon wrote:

> On 08/09/2020 08:46, Andriy Gapon wrote:
>> Author: avg
>> Date: Tue Sep  8 05:46:10 2020
>> New Revision: 365445
>> URL: https://svnweb.freebsd.org/changeset/base/365445
>>
>> Log:
>>   mmc_da: make sure that part_index is not used uninitialized in 
>> sddastart
> [snip]
>> Modified: head/sys/cam/mmc/mmc_da.c
>> ==============================================================================
>> --- head/sys/cam/mmc/mmc_da.c	Tue Sep  8 04:44:37 2020	(r365444)
>> +++ head/sys/cam/mmc/mmc_da.c	Tue Sep  8 05:46:10 2020	(r365445)
>> @@ -1808,6 +1808,7 @@ sddastart(struct cam_periph *periph, union ccb 
>> *start_
>>  	}
>>
>>  	/* Find partition that has outstanding commands.  Prefer current 
>> partition. */
>> +	part_index = softc->part_curr;
>>  	part = softc->part[softc->part_curr];
>>  	bp = bioq_first(&part->bio_queue);
>>  	if (bp == NULL) {
>>
>
> One thing that concerns me is that it was obvious (to a human) that 
> part_index
> could be used uninitialized if bp was not NULL.
> Yet, there was no warning or error from the compiler when I built that 
> code for
> armv7.
>
> I wonder if we disable some relevant warnings for that architecture.
> Or if the compiler (clang 11) could not figure that out.


I had just finished dissecting my hang in g_waitidle on boot on the 
nanopi-m4v2/arm64 this morning and when I saw your commits as I came 
back home .. guess what:

364132          boots
364219          boots
364263          boots
364274          boots
364283          boots
364284          no			<< 
https://svnweb.freebsd.org/base?view=revision&revision=364284  (clang 
merge)
364285          no
364307          no
364482          no
364657          no
365335          no
365364          no
365448          boots

/bz