Any thoughts?
On Mon, May 11, 2009 at 12:13 PM, Pei Lin <telent997@xxxxxxxxx> wrote:
2009/5/8 Chetan Nanda <chetannanda@xxxxxxxxx>:
>because "volatile int j asm("ebx");" u objdump the elf
>
> On Fri, May 8, 2009 at 1:28 AM, seshikanth varma <seshikanthvarma@xxxxxxxxx>
> wrote:
>>
>> This is the way i have defined the variables:
>>
>> volatile int i asm("eax"); /* For using eax ebx ecx & edx
>> for i, j , k and l respectively.
>>
>> * All the threads shud use eax
>> for i, ebx for j ...
>>
>> */
>> volatile int j asm("ebx");
>> volatile int k asm("ecx");
>> volatile int l asm("edx");
>>
>> Though every thread has its own copy of registers,the registers eax, ebx
>> ecx and edx are unique by name. So, the threads should be writing to these
>> registers which means that these variables are shared.
>
>
> I am still confused, as every thread will be having its own register set and
> this set will get stored in its 'tast_struct' at each context switch and at
> next run registers will get populated from corresponding 'task_struct'.
>
> So how these variable will get shared between different threads? I am
> missing any basic thing?
will find the variables are in data section, without initial it in
.bss section. if u change code to this "register int j asm("ebx")
",run the app u will get the segmentation fault.
sandeep lahane said
"The race condition in OP's code can be easily reproduced by making the>>>>>>>>>>>yes, i also reproduced in my computer.
LOOPCOUNT to 10000 and running the program multiple times on a SMP
machine."
>
> Thanks,
> Chetan Nanda
>
>>
>> Regards,
>> Seshikanth
>>
>> On Fri, May 8, 2009 at 1:21 AM, Daniel Baluta <daniel.baluta@xxxxxxxxx>
>> wrote:
>>>
>>> On Thu, May 7, 2009 at 10:38 PM, seshikanth varma
>>> <seshikanthvarma@xxxxxxxxx> wrote:
>>> > Hi,
>>> > I am learning linux kernel. I have written a simple program to
>>> > understand
>>> > the usage of mutex variable.
>>> > Ideally the following program should generate race condition and should
>>> > produce the different values on i,j,k and l rather than i = j = k = l =
>>> > 100(LOOPCONSTANT * Number of threads) on usage of mutex. Here shared
>>> > variables between threads are i,j,k and l.
>>> > My kernel is not SMP. Can u please tell me where am i going wrong?
>>> >
>>> > ======================================================================================================
>>> >
>>> > #include <pthread.h>
>>> > #include <stdio.h>
>>> > #define checkResults(string, val) { \
>>> > if (val) { \
>>> > printf("Failed with %d at %s", val, string); \
>>> > exit(1); \
>>> > } \
>>> > }
>>> > //#define LOOPCONSTANT 100000
>>> > #define LOOPCONSTANT 10
>>> > #define THREADS 10
>>> >
>>> > pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
>>> > volatile int i asm("eax"); /* For using eax ebx ecx &
>>> > edx for
>>> > i, j , k and l respectively.
>>> >
>>> > * All the threads shud use
>>> > eax
>>> > for i, ebx for j ...
>>> >
>>> > */
>>> > volatile int j asm("ebx");
>>> > volatile int k asm("ecx");
>>> > volatile int l asm("edx");
>>>
>>> Are you sure that this variables are shared among threads?
>>> Please keep in mind that every thread has its own copy of registers.
>>>
>>> > int uselock=0;
>>> >
>>> > void *threadfunc(void *parm)
>>> > {
>>> > int loop = 0;
>>> > int rc;
>>> >
>>> > for (loop=0; loop<LOOPCONSTANT; ++loop) {
>>> > if (uselock) {
>>> > rc = pthread_mutex_lock(&mutex);
>>> > checkResults("pthread_mutex_lock()\n", rc);
>>> > }
>>> > ++i; ++j; ++k; ++l;
>>> > if (uselock) {
>>> > rc = pthread_mutex_unlock(&mutex);
>>> > checkResults("pthread_mutex_unlock()\n", rc);
>>> > }
>>> > }
>>> > return NULL;
>>> > }
>>> >
>>> > int main(int argc, char **argv)
>>> > {
>>> > pthread_t threadid[THREADS];
>>> > int rc=0;
>>> >
>>> > int loop=0;
>>> > pthread_attr_t pta;
>>> >
>>> > printf("Entering testcase\n");
>>> > printf("Give any number of parameters to show data corruption\n");
>>> > if (argc != 1) {
>>> > printf("A parameter was specified, no serialization is being
>>> > done!\n");
>>> > uselock = 0;
>>> > }
>>> >
>>> > pthread_attr_init(&pta);
>>> > pthread_attr_setdetachstate(&pta, PTHREAD_CREATE_JOINABLE);
>>> >
>>> > printf("Creating %d threads\n", THREADS);
>>> > for (loop=0; loop<THREADS; ++loop) {
>>> > rc = pthread_create(&threadid[loop], &pta, threadfunc, NULL);
>>> > checkResults("pthread_create()\n", rc);
>>> > }
>>> >
>>> > printf("Wait for results\n");
>>> > for (loop=0; loop<THREADS; ++loop) {
>>> > rc = pthread_join(threadid[loop], NULL);
>>> > checkResults("pthread_join()\n", rc);
>>> > }
>>> >
>>> > printf("Cleanup and show results\n");
>>> > pthread_attr_destroy(&pta);
>>> > pthread_mutex_destroy(&mutex);
>>> >
>>> > printf("\nUsing %d threads and LOOPCONSTANT = %d\n",
>>> > THREADS, LOOPCONSTANT);
>>> > printf("Values are: (should be %d)\n", THREADS * LOOPCONSTANT);
>>> > printf(" ==> seshikanth | %d, %d, %d, %d\n", i, j, k, l);
>>> >
>>> > printf("Main completed\n");
>>> > return 0;
>>> >
>>> > }
>>> >
>>> > ======================================================================================================
>>> >
>>> > O/p of the above program:
>>> > ---------------------------------------
>>> > with userlock = 0 (i.e., shud hit race condition)
>>> > --------------------------------------------------------------------
>>> > gcc check.c -pthread
>>> > bash-2.05b$ ./a.out
>>> > Creating 10 threads
>>> > Wait for results
>>> > Cleanup and show results
>>> > Using 10 threads and LOOPCONSTANT = 10
>>> > Values are: (should be 100)
>>> > ==> | 100, 100, 100, 100 <<<<<<<<<<<==========*******
>>> > Main completed
>>> > bash-2.05b$
>>> > =====================================
>>> > with userlock = 1;
>>> > -------------------------
>>> > I am getting the same o/p which is correct....
>>> >
>>> > Any help regarding this is greatly appreciated.
>>> > Thanks,
>>> > Seshikanth
>>> >
>>> >
>>
>
>
--
Regards,
Seshikanth
