warning: division by zero

warning: division by zero

[Randy Dunlap]

In Linux kernel, we get that message 2 times (in x86_64 allmodconfig):

../fs/exofs/ore_raid.c:95:64: warning: division by zero
../fs/exofs/ore_raid.c:116:36: warning: division by zero

when dividing by variable sizeof_a1pa, where that is:
	const unsigned sizeof__a1pa = sizeof(_aab->__a1pa[0]);

which is in this struct:
	struct _alloc_all_bytes {
		struct __alloc_stripe_pages_2d {
			struct __stripe_pages_2d sp2d;
			struct __1_page_stripe _1p_stripes[pages_in_unit];
		} __asp2d;
		struct __alloc_1p_arrays {
			struct page *pages[group_width];
			struct page *scribble[group_width];
			char page_is_read[data_devs];
		} __a1pa[pages_in_unit];
	} *_aab;

where 'pages_in_unit' is a function parameter. So, yes, it could have a value
of zero.

Is the warning message considered correct or useful as is?  If so, OK, no problem.
Although I would rather see something more like:
	warning: variable length array on stack

thanks,
-- 
~Randy

Re: warning: division by zero

[Luc Van Oostenryck]

On Tue, Feb 13, 2018 at 04:26:04PM -0800, Randy Dunlap wrote:
> In Linux kernel, we get that message 2 times (in x86_64 allmodconfig):
> 
> ../fs/exofs/ore_raid.c:95:64: warning: division by zero
> ../fs/exofs/ore_raid.c:116:36: warning: division by zero
> 
> when dividing by variable sizeof_a1pa, where that is:
> 	const unsigned sizeof__a1pa = sizeof(_aab->__a1pa[0]);
> 
> which is in this struct:
> 	struct _alloc_all_bytes {
> 		struct __alloc_stripe_pages_2d {
> 			struct __stripe_pages_2d sp2d;
> 			struct __1_page_stripe _1p_stripes[pages_in_unit];
> 		} __asp2d;
> 		struct __alloc_1p_arrays {
> 			struct page *pages[group_width];
> 			struct page *scribble[group_width];
> 			char page_is_read[data_devs];
> 		} __a1pa[pages_in_unit];
> 	} *_aab;
> 
> where 'pages_in_unit' is a function parameter. So, yes, it could have a value
> of zero.
> 
> Is the warning message considered correct or useful as is?  If so, OK, no problem.

It's a division by zero, not because the parameter can be zero
but because currently sparse doesn't handle the size of VLAs which
are always considered as zero-sized.
So, no I don't consider this warning as correct and even less useful.

I have a WIP series to handle VLA's sizeof, if there is some interests
I can boost a bit its priority.

> Although I would rather see something more like:
> 	warning: variable length array on stack

That's of course something different that could indeed be quite useful.

-- Luc

Re: warning: division by zero

[Randy Dunlap]

On 02/13/2018 04:39 PM, Luc Van Oostenryck wrote:
> On Tue, Feb 13, 2018 at 04:26:04PM -0800, Randy Dunlap wrote:
>> In Linux kernel, we get that message 2 times (in x86_64 allmodconfig):
>>
>> ../fs/exofs/ore_raid.c:95:64: warning: division by zero
>> ../fs/exofs/ore_raid.c:116:36: warning: division by zero
>>
>> when dividing by variable sizeof_a1pa, where that is:
>> 	const unsigned sizeof__a1pa = sizeof(_aab->__a1pa[0]);
>>
>> which is in this struct:
>> 	struct _alloc_all_bytes {
>> 		struct __alloc_stripe_pages_2d {
>> 			struct __stripe_pages_2d sp2d;
>> 			struct __1_page_stripe _1p_stripes[pages_in_unit];
>> 		} __asp2d;
>> 		struct __alloc_1p_arrays {
>> 			struct page *pages[group_width];
>> 			struct page *scribble[group_width];
>> 			char page_is_read[data_devs];
>> 		} __a1pa[pages_in_unit];
>> 	} *_aab;
>>
>> where 'pages_in_unit' is a function parameter. So, yes, it could have a value
>> of zero.
>>
>> Is the warning message considered correct or useful as is?  If so, OK, no problem.
> 
> It's a division by zero, not because the parameter can be zero
> but because currently sparse doesn't handle the size of VLAs which
> are always considered as zero-sized.
> So, no I don't consider this warning as correct and even less useful.
> 
> I have a WIP series to handle VLA's sizeof, if there is some interests
> I can boost a bit its priority.

No hurry. Thanks for your comments.

>> Although I would rather see something more like:
>> 	warning: variable length array on stack
> 
> That's of course something different that could indeed be quite useful.


-- 
~Randy