Re: [PATCH] x86/spinlocks: Fix regression in spinlock contention detection

[Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>]

On 05/05/2015 09:02 PM, Waiman Long wrote:
> On 05/05/2015 11:25 AM, Raghavendra K T wrote:
>> On 05/05/2015 07:33 PM, Tahsin Erdogan wrote:
>>> The conversion to signed happens with types shorter than int (__ticket_t
>>> is either u8 or u16).
>>>
>>> By changing Raghavendra's program to use unsigned short int, you can see
>>> the problem:
>>>
>>> ================
>>> #include <stdio.h>
>>>
>>> #define LOCK_INC 2
>>>
>>> int main()
>>> {
>>>          unsigned short int head = 32700, tail=2;
>>>
>>>          if ((tail - head) > LOCK_INC)
>>>                  printf(" tail - head > LOCK_INC \n");
>>>          else
>>>                  printf(" tail - head < LOCK_INC \n");
>>>
>>>          return 0;
>>> }
>>>
>>> ================
>>> gcc -g -o t main.c
>>> ./t
>>>   tail - head < LOCK_INC
>>>
>>> However, having just unsigned int returns the opposite result (unsigned
>>> int head = 32700, tail=2;)
>>>
>>
>> Interestingly,
>>
>> #include <stdio.h>
>>
>> //#define LOCK_INC ((unsigned int)2) // case 1
>> #define LOCK_INC 2 //case 2
>>
>> int main()
>> {
>>     unsigned short int head = 32700, tail=2;
>>
>>     if ((tail - head) > LOCK_INC)
>>         printf(" tail - head > LOCK_INC \n");
>>     else
>>         printf(" tail - head < LOCK_INC \n");
>>
>>     return 0;
>> }
>>
>> case 1 works here (PeterZ's stricter version)
>>
>> case 2 gives tail - head < LOCK_INC
>>
>> But is it not that we have case 1 we are looking here ?
>>
>>
>
> __TICKET_LOCK_INC is currently ((unsigned short)2), not ((unsigned
> int)2). That makes a difference.
>

aah missed that part :). That makes sense.

Good catch Tahsin.