Re: [Patch] lkdtm: avoid calling lkdtm_do_action() with spin lock held

[Cong Wang <xiyou.wangcong@gmail.com>]

On 01/31/2012 11:35 PM, Arnd Bergmann wrote:
> On Tuesday 31 January 2012, Cong Wang wrote:
>> @@ -323,14 +323,16 @@ static void lkdtm_do_action(enum ctype which)
>>          }
>>          case CT_WRITE_AFTER_FREE: {
>>                  size_t len = 1024;
>> -               u32 *data = kmalloc(len, GFP_KERNEL);
>> +               u32 *data = kmalloc(len, GFP_ATOMIC);
>>
>>                  kfree(data);
>> -               schedule();
>> +               udelay(100);
>>                  memset(data, 0x78, len);
>>                  break;
>>          }
>
> I can't think of why the udelay would have any positive effect here,
> if the idea of the schedule was to let some other process allocate and
> use the memory.


Hmm, on SMP udelay on this CPU will give a chance to other CPU's to use 
that memory, right?

>
> Can't you just get rid of the count_lock if you use an atomic_t for the
> count and use appropriate accesses on it?
>

Good idea, will do.