Re: [LTP] Regarding shmat01 syscall test

[Wanlong G. <gao...@cn...>]

On 10/24/2012 03:49 PM, Jan Stancek wrote:
> 
> 
> ----- Original Message -----
>> From: "Wanlong Gao" <gao...@cn...>
>> To: "Jan Stancek" <jst...@re...>
>> Cc: ltp...@li..., "Om Prakash PAL" <omp...@st...>
>> Sent: Wednesday, 24 October, 2012 9:03:16 AM
>> Subject: Re: [LTP] Regarding shmat01 syscall test
>>
>> On 10/24/2012 02:43 PM, Jan Stancek wrote:
>>>
>>>
>>> ----- Original Message -----
>>>> From: "Wanlong Gao" <gao...@cn...>
>>>> To: "Om Prakash PAL" <omp...@st...>
>>>> Cc: ltp...@li...
>>>> Sent: Wednesday, 24 October, 2012 2:45:47 AM
>>>> Subject: Re: [LTP] Regarding shmat01 syscall test
>>>>
>>>> On 10/23/2012 06:05 PM, Om Prakash PAL wrote:
>>>>>
>>>>>
>>>>> -----Original Message-----
>>>>> From: Wanlong Gao [mailto:gao...@cn...]
>>>>> Sent: Tuesday, October 23, 2012 3:07 PM
>>>>> To: Om Prakash PAL
>>>>> Cc: ltp...@li...
>>>>> Subject: Re: [LTP] Regarding shmat01 syscall test
>>>>>
>>>>> On 10/23/2012 05:24 PM, Om Prakash PAL wrote:
>>>>>> Hi,
>>>>>>
>>>>>> I am working on syscall test:  shmat01.c
>>>>>>
>>>>>> I have some confusion:
>>>>>>
>>>>>>  
>>>>>>
>>>>>> In setup() : it is  allocating shared memory by shmget() and
>>>>>> then
>>>>>> attaching by shmat() and after  that detaching the attached
>>>>>> address (i.e. shmdt())
>>>>>>
>>>>>>  
>>>>>>
>>>>>>         if (shmdt((const void *)base_addr) == -1) {
>>>>>>
>>>>>>                 tst_brkm(TBROK, cleanup, "Couldn't detach shared
>>>>>>                 memory");
>>>>>>
>>>>>>         }
>>>>>>
>>>>>>  
>>>>>>
>>>>>> And again in main function it is using same "base_addr" as
>>>>>> attaching address,
>>>>>>
>>>>>>  
>>>>>>
>>>>>> addr = shmat(*(TC[i].shmid), base_addr+TC[i].offset,
>>>>>>
>>>>>>                                      TC[i].flags);
>>>>>>
>>>>>> how can we ensure(100%) that base_addr (virtual) will be free
>>>>>> till
>>>>>> this point for attaching?.
>>>>>
>>>>> Maybe we can't, but I didn't see any fail on this. Did you see
>>>>> any
>>>>> testing failure here?
>>>>>
>>>>> Yes, I got some failure and the reason of failure is : the
>>>>> address
>>>>> at which we want to attach is busy.
>>>>
>>>> OK, please feel free to send a patch, or can you tell us how to
>>>> reproduce it?
>>>
>>> I recall I could reproduce it, if I added single printf:
>>> http://article.gmane.org/gmane.linux.ltp/16480
>>
>> Do you get a solution? Send out a patch?
> 
> No, I haven't send any patch.
> 
> About solution:
> I'm thinking, that instead of probing with shmat, we can mmap large chunk of memory,
> and then set base_addr somewhere in the middle and unmap the chunk.
> That is, using address that get_unmapped_area() is unlikely to pick.

This idea seems good, bug how can you decide the size of this "chunk of memory"?

Thanks,
Wanlong Gao

> 
> Regards,
> Jan
> 
>>
>>
>> Thanks,
>> Wanlong Gao
>>
>>>
>>> Regards,
>>> Jan
>>>
>>
>>
>