Re: [PATCH] fixp-arith: prevent division by zero in fixp_sin32_rad()

[Sergey Shtylyov <s.shtylyov@omp.ru>]

On 3/13/24 9:23 PM, Sergey Shtylyov wrote:
[...]

>> The parameter twopi can have a zero value. It is necessary
>> to prevent division by zero.
>>
>> Found by Linux Verification Center (linuxtesting.org) with Svace.
>>
>> Signed-off-by: Roman Smirnov <r.smirnov@omp.ru>
>> ---
>>  include/linux/fixp-arith.h | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/include/linux/fixp-arith.h b/include/linux/fixp-arith.h
>> index e485fb0c1201..434eee8bd9ef 100644
>> --- a/include/linux/fixp-arith.h
>> +++ b/include/linux/fixp-arith.h
>> @@ -118,8 +118,9 @@ static inline s32 fixp_sin32_rad(u32 radians, u32 twopi)
>>  
>>  	/*
>>  	 * Avoid too large values for twopi, as we don't want overflows.
>> +	 * Also prevent division by zero.
>>  	 */
>> -	BUG_ON(twopi > 1 << 18);
>> +	BUG_ON(!twopi || twopi > 1 << 18);
> 
>     Not really sure ATM that BUG_ON() is better than division by 0
> (which should cause exception 0 on x86 but I'm not sure about e.g.
> ARM32/64)...

   Now I know what happens there: on arch/arm/ an pr_err() will be called
followed by dump_stack(); on arch/arm64/ division instruction will silently
yield 0 result...

>>  
>>  	degrees = (radians * 360) / twopi;
>>  	tmp = radians - (degrees * twopi) / 360;
>>
> 
> Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>

   The patch does seem legitimate...

]...]

MBR, Sergey