Re: [LTP] [PATCH] fix false alarm in pselect01

Re: [LTP] [PATCH] fix false alarm in pselect01

[Caspar Zhang]

[-- Attachment #1: Type: text/plain, Size: 1330 bytes --]


----- "Caspar Zhang" <czhang@redhat.com> wrote:

> Hi all,
> 
>    In my recent tests, not all desired time is equal to exact
> execution time,
>    sometimes it is +/-1s, and in a few times it becomes +/-2s. In most
> cases,
>    +/-2s is still a valid value between exact execute time and desired
> time.
> 
>    Signed-off-by: Caspar Zhang <czhang@redhat.com>
> 
> --- a/testcases/kernel/syscalls/pselect/pselect01.c     2010-04-01
> 14:23:10.000000000 +0800
> +++ b/testcases/kernel/syscalls/pselect/pselect01.c     2010-07-15
> 16:03:40.969426078 +0800
> @@ -109,7 +109,8 @@
>                             NULL);
>                 end = time(&t);
> 
> -               if (((end - start) == total_sec) || ((end - start) ==
> total_sec + 1))
> +               /* Is it still valid in range -2~2s? */
> +               if ((end - start - total_sec <= 2) && (end - start -
> total_sec >= -2))

Sorry for mistake, since end, start are unsigned int, use 
end-start-total_sec >= -2 will cause error, add (int)before 
(end-start) instead. New patch attached.

Thanks,
Caspar


-- 
Kernel Associate Quality Engineer
Red Hat Inc. (Beijing R&D Branch)

Red Hat China R&D Branch Unit 907, North Tower C, 
  Raycom Infotech Park, No.2 Kexueyuan Nanlu, 
  Haidian District, Beijing 100190

TEL: +86-10-62608150
Web: http://www.redhat.com/

[-- Attachment #2: syscalls-pselect01-fix-false-alarm.patch --]
[-- Type: text/x-patch, Size: 850 bytes --]

Hi all,
   
   In my recent tests, not all desired time is equal to exact execution time, 
   sometimes it is +/-1s, and in a few times it becomes +/-2s. In most cases,
   +/-2s is still a valid value between exact execute time and desired time.
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

--- a/testcases/kernel/syscalls/pselect/pselect01.c	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-15 16:03:40.969426078 +0800
@@ -109,7 +109,8 @@
 			    NULL);
 		end = time(&t);
 
-		if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+		/* Is it still valid in range -2~2s? */
+		if (((int)(end - start) - total_sec <= 2) && ((int)(end - start) - total_sec >= -2))
 			tst_resm(TPASS, "Sleep time was correct");
 		else
 			tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",

[-- Attachment #3: Type: text/plain, Size: 235 bytes --]

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

[-- Attachment #4: Type: text/plain, Size: 155 bytes --]

_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Garrett Cooper]

On Sun, Jul 18, 2010 at 12:25 AM, Caspar Zhang <czhang@redhat.com> wrote:
>
> ----- "Caspar Zhang" <czhang@redhat.com> wrote:
>
>> Hi all,
>>
>>    In my recent tests, not all desired time is equal to exact
>> execution time,
>>    sometimes it is +/-1s, and in a few times it becomes +/-2s. In most
>> cases,
>>    +/-2s is still a valid value between exact execute time and desired
>> time.
>>
>>    Signed-off-by: Caspar Zhang <czhang@redhat.com>
>>
>> --- a/testcases/kernel/syscalls/pselect/pselect01.c     2010-04-01
>> 14:23:10.000000000 +0800
>> +++ b/testcases/kernel/syscalls/pselect/pselect01.c     2010-07-15
>> 16:03:40.969426078 +0800
>> @@ -109,7 +109,8 @@
>>                             NULL);
>>                 end = time(&t);
>>
>> -               if (((end - start) == total_sec) || ((end - start) ==
>> total_sec + 1))
>> +               /* Is it still valid in range -2~2s? */
>> +               if ((end - start - total_sec <= 2) && (end - start -
>> total_sec >= -2))
>
> Sorry for mistake, since end, start are unsigned int, use
> end-start-total_sec >= -2 will cause error, add (int)before
> (end-start) instead. New patch attached.

    The fact that the test is using time(3) to calculate the elapsed
period instead of clock_gettime is dubious at best, because the
precision of time is in seconds. Maybe it should be converted to
clock_gettime calls for improved precision?
Thanks,
-Garrett

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Caspar Zhang]

[-- Attachment #1: Type: text/plain, Size: 2699 bytes --]


----- "Garrett Cooper" <yanegomi@gmail.com> wrote:

> On Wed, Jul 21, 2010 at 2:40 AM, Caspar Zhang <czhang@redhat.com>
> wrote:
> > Hi Garrett, I tested my patch yesterday but the result
> > is not so good. After 1000 times running, the fix:
> >
> > +               /* do a rounding */
> > +               real_sec = (int)(0.5 + (tv_end.tv_sec -
> tv_start.tv_sec +
> > +                               1e-9 * (tv_end.tv_nsec -
> tv_start.tv_nsec)));
> > +               if (real_sec == total_sec)
> >
> > got failed for once. so I set the allowed error to 20%(see the
> > attached patch). Then re-ran it for over 5000 times, without
> > a failure.

Hi, seems that this changing not ACKed or committed, re-attach
the patch .

Thanks,
Caspar

> >
> > Then the nsec part, 10% allowed error seems not enough,
> > the test failed for twice among 1000 times. So I set the
> > range to 20%, too.
> 
> Fair enough... change committed.



Hi,
        Tests can be easily failing when using real_sec == total_sec.
        Changed to the similar way as in nsec part does, allow 20%
        error.

        Signed-off-by: Caspar Zhang <czhang@redhat.com>

--- a/testcases/kernel/syscalls/pselect/pselect01.c     2010-07-21 18:08:25.139600055 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c     2010-07-22 16:17:06.552474774 +0800
@@ -66,7 +66,8 @@
        int retval;
        struct timespec tv, tv_start, tv_end;
        long real_nsec, total_nsec;
-       int real_sec, total_sec;
+       double real_sec;
+       int total_sec;
 
        setup();
 
@@ -107,13 +108,13 @@
                            NULL);
                clock_gettime(CLOCK_REALTIME, &tv_end);
 
-               /* do a rounding */
-               real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec + 
-                               1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec)));
-               if (real_sec == total_sec)
+               real_sec = (tv_end.tv_sec - tv_start.tv_sec + 
+                               1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec));
+               /* allow about 20% error */
+               if (abs(real_sec - total_sec) < 0.2 * total_sec)
                        tst_resm(TPASS, "Sleep time was correct");
                else
-                       tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
+                       tst_resm(TFAIL, "Sleep time was incorrect:%d != %lf",
                                 total_sec, real_sec);
        }

-- 
Quality Assurance Associate (Kernel) in
Red Hat Inc. (Beijing R&D Branch)

TEL: +86-10-62608150
Web: http://www.redhat.com/

[-- Attachment #2: pselect-error-range.patch --]
[-- Type: text/x-patch, Size: 1177 bytes --]

Hi,
	Tests can be easily failing when using real_sec == total_sec.
	Changed to the similar way as in nsec part does, allow 20%
	error.

	Signed-off-by: Caspar Zhang <czhang@redhat.com>

--- a/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-21 18:08:25.139600055 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-22 16:17:06.552474774 +0800
@@ -66,7 +66,8 @@
 	int retval;
 	struct timespec tv, tv_start, tv_end;
 	long real_nsec, total_nsec;
-	int real_sec, total_sec;
+	double real_sec;
+	int total_sec;
 
 	setup();
 
@@ -107,13 +108,13 @@
 			    NULL);
 		clock_gettime(CLOCK_REALTIME, &tv_end);
 
-		/* do a rounding */
-		real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec + 
-				1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec)));
-		if (real_sec == total_sec)
+		real_sec = (tv_end.tv_sec - tv_start.tv_sec + 
+				1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec));
+		/* allow about 20% error */
+		if (abs(real_sec - total_sec) < 0.2 * total_sec)
 			tst_resm(TPASS, "Sleep time was correct");
 		else
-			tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
+			tst_resm(TFAIL, "Sleep time was incorrect:%d != %lf",
 				 total_sec, real_sec);
 	}
 

[-- Attachment #3: Type: text/plain, Size: 235 bytes --]

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

[-- Attachment #4: Type: text/plain, Size: 155 bytes --]

_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Garrett Cooper]

On Wed, Jul 21, 2010 at 2:40 AM, Caspar Zhang <czhang@redhat.com> wrote:
>
> ----- "Garrett Cooper" <yanegomi@gmail.com> wrote:
>
>> ACK (and thanks for the hard work)!
>> -Garrett
>
> Hi Garrett, I tested my patch yesterday but the result
> is not so good. After 1000 times running, the fix:
>
> +               /* do a rounding */
> +               real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec +
> +                               1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec)));
> +               if (real_sec == total_sec)
>
> got failed for once. so I set the allowed error to 20%(see the
> attached patch). Then re-ran it for over 5000 times, without
> a failure.
>
> Then the nsec part, 10% allowed error seems not enough,
> the test failed for twice among 1000 times. So I set the
> range to 20%, too.

Fair enough... change committed.
Thanks!
-Garrett

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Caspar Zhang]

[-- Attachment #1: Type: text/plain, Size: 791 bytes --]


----- "Garrett Cooper" <yanegomi@gmail.com> wrote:

> ACK (and thanks for the hard work)!
> -Garrett

Hi Garrett, I tested my patch yesterday but the result
is not so good. After 1000 times running, the fix:

+		/* do a rounding */
+		real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec + 
+				1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec)));
+		if (real_sec == total_sec)

got failed for once. so I set the allowed error to 20%(see the
attached patch). Then re-ran it for over 5000 times, without
a failure.

Then the nsec part, 10% allowed error seems not enough, 
the test failed for twice among 1000 times. So I set the
range to 20%, too.

Thanks,
Caspar

-- 
Quality Assurance Associate (Kernel) in
Red Hat Inc. (Beijing R&D Branch)

TEL: +86-10-62608150
Web: http://www.redhat.com/

[-- Attachment #2: syscalls-pselect01-fix-false-alarm.patch --]
[-- Type: text/x-patch, Size: 3854 bytes --]

Hi all,
   
   1. In my recent tests, not all desired time be equal to exact execution 
   time in sec test part, use clock_gettime to improve precision, and allow
   the error range to be 20%.

   2. gettimeofday is marked as obsolete in POSIX.1-2008 according to
   man gettimeofday, so change gettimeofday to clock_gettime in nsec test
   part. 
   
   3. 10% range in nsec test part seems not enough.I ran the test for about 
   1000 times, the nsec test part failed twice. Changed the value to 20%.
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

diff -Naur a/testcases/kernel/syscalls/pselect/Makefile b/testcases/kernel/syscalls/pselect/Makefile
--- a/testcases/kernel/syscalls/pselect/Makefile	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/Makefile	2010-07-20 16:59:45.983475123 +0800
@@ -21,6 +21,7 @@
 include $(top_srcdir)/include/mk/testcases.mk
 include $(abs_srcdir)/../utils/newer_64.mk
 
+LDLIBS			+= -lpthread -lrt
 %_64: CPPFLAGS += -D_FILE_OFFSET_BITS=64
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
diff -Naur a/testcases/kernel/syscalls/pselect/pselect01.c b/testcases/kernel/syscalls/pselect/pselect01.c
--- a/testcases/kernel/syscalls/pselect/pselect01.c	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-20 17:00:01.498601028 +0800
@@ -61,14 +61,13 @@
 
 int main()
 {
-	int ret_pselect, total_sec, fd, total_nsec;
+	int ret_pselect, fd;
 	fd_set readfds;
-	struct timespec tv;
 	int retval;
-	time_t t;
-	unsigned start, end;
-	struct timeval tv_start, tv_end;
-	int real_usec;
+	struct timespec tv, tv_start, tv_end;
+	long real_nsec, total_nsec;
+	double real_sec;
+	int total_sec;
 
 	setup();
 
@@ -103,17 +102,20 @@
 		tst_resm(TINFO,
 			 "Testing basic pselect sanity,Sleeping for %jd secs",
 			 (intmax_t)tv.tv_sec);
-		start = time(&t);
+		clock_gettime(CLOCK_REALTIME, &tv_start);
 		retval =
-		    pselect(0, &readfds, NULL, NULL, (struct timespec *)&tv,
+		    pselect(0, &readfds, NULL, NULL, &tv,
 			    NULL);
-		end = time(&t);
+		clock_gettime(CLOCK_REALTIME, &tv_end);
 
-		if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+		real_sec = (tv_end.tv_sec - tv_start.tv_sec + 
+				1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec));
+		/* allow about 20% error */
+		if (abs(real_sec - total_sec) < 0.2 * total_sec)
 			tst_resm(TPASS, "Sleep time was correct");
 		else
-			tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
-				 total_sec, (end - start));
+			tst_resm(TFAIL, "Sleep time was incorrect:%d != %lf",
+				 total_sec, real_sec);
 	}
 
 #ifdef DEBUG
@@ -129,19 +131,19 @@
 		tst_resm(TINFO,
 			 "Testing basic pselect sanity,Sleeping for %ld nano secs",
 			 tv.tv_nsec);
-		gettimeofday(&tv_start, NULL);
+		clock_gettime(CLOCK_REALTIME, &tv_start);
 		retval =
 		    pselect(0, &readfds, NULL, NULL, &tv,
 			    NULL);
-		gettimeofday(&tv_end, NULL);
+		clock_gettime(CLOCK_REALTIME, &tv_end);
 
 		/* Changed total_sec compare to an at least vs an exact compare */
 
-		real_usec = (tv_end.tv_sec - tv_start.tv_sec) * 1e6 +
-			tv_end.tv_usec - tv_start.tv_usec;
+		real_nsec = (tv_end.tv_sec - tv_start.tv_sec) * 1e9 +
+			tv_end.tv_nsec - tv_start.tv_nsec;
 
-		/* allow 10% error*/
-		if (abs(real_usec - tv.tv_nsec / 1000) < 0.1 * total_nsec / 1000)
+		/* allow 20% error*/
+		if (abs(real_nsec - tv.tv_nsec) < 0.2 * total_nsec)
 			tst_resm(TPASS, "Sleep time was correct");
 		else {
 			tst_resm(TWARN,
@@ -150,8 +152,8 @@
 				 "due to the limitation of the way it calculates the");
 			tst_resm(TWARN, "system call execution time.");
 			tst_resm(TFAIL,
-				"Sleep time was incorrect:%d usec vs expected %d usec",
-				 real_usec, total_nsec / 1000);
+				"Sleep time was incorrect:%ld nsec vs expected %ld nsec",
+				 real_nsec, total_nsec);
 		}
 	}
 	cleanup();

[-- Attachment #3: Type: text/plain, Size: 235 bytes --]

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

[-- Attachment #4: Type: text/plain, Size: 155 bytes --]

_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Garrett Cooper]

On Mon, Jul 19, 2010 at 1:32 AM, Caspar Zhang <czhang@redhat.com> wrote:
>
> ----- "Garrett Cooper" <yanegomi@gmail.com> wrote:
>
>> On Sun, Jul 18, 2010 at 11:55 PM, Caspar Zhang <czhang@redhat.com>
>> wrote:
>> >
>> > ----- "Garrett Cooper" <yanegomi@gmail.com> wrote:
>> >
>> >>
>> >>     The fact that the test is using time(3) to calculate the
>> elapsed
>> >> period instead of clock_gettime is dubious at best, because the
>> >> precision of time is in seconds. Maybe it should be converted to
>> >> clock_gettime calls for improved precision?
>> >
>> > Use clock_gettime instead of time(3), patch attached.
>>
>> Looks better, but being strict it's still buggy. POSIX defines struct
>> timespec as:
>>
>> The <time.h> header shall declare the structure timespec, which has
>> at
>> least the following members:
>>
>> time_t  tv_sec    Seconds.
>> long    tv_nsec   Nanoseconds.
>>
>> So you're losing precision in one case, and time_t can be variable
>> (the spec for types.h is rather fluid):
>>
>> time_t and clock_t shall be integer or real-floating types.
>>
>> The (struct timeval *) casts are also unnecessary because you're
>> getting the address of a struct timeval object (which is in both
>> cases
>> struct timeval*).
>>
>> > Another question, since POSIX.1-2008 marks gettimeofday() as
>> obsolete
>> > (man gettimeofday), should the usec test part be changed to
>> > clock_getime as well?
>>
>> Sure, if you want.
>
> Updated patch, FYI.
>
> Thanks,
> Caspar
>
>
>
> Hi all,
>
>   1. In my recent tests, not all desired time be equal to exact execution
>   time in sec test part, use clock_gettime to improve precision.
>   2. gettimeofday is marked as obsolete in POSIX.1-2008 according to
>   man gettimeofday, so change gettimeofday to clock_gettime in nsec test
>   part.
>
>   Signed-off-by: Caspar Zhang <czhang@redhat.com>
>
> diff -Naur a/testcases/kernel/syscalls/pselect/Makefile b/testcases/kernel/syscalls/pselect/Makefile
> --- a/testcases/kernel/syscalls/pselect/Makefile        2010-04-01 14:23:10.000000000 +0800
> +++ b/testcases/kernel/syscalls/pselect/Makefile        2010-07-19 14:54:38.984482305 +0800
> @@ -21,6 +21,7 @@
>  include $(top_srcdir)/include/mk/testcases.mk
>  include $(abs_srcdir)/../utils/newer_64.mk
>
> +LDLIBS                 += -lpthread -lrt
>  %_64: CPPFLAGS += -D_FILE_OFFSET_BITS=64
>
>  include $(top_srcdir)/include/mk/generic_leaf_target.mk
> diff -Naur a/testcases/kernel/syscalls/pselect/pselect01.c b/testcases/kernel/syscalls/pselect/pselect01.c
> --- a/testcases/kernel/syscalls/pselect/pselect01.c     2010-04-01 14:23:10.000000000 +0800
> +++ b/testcases/kernel/syscalls/pselect/pselect01.c     2010-07-19 16:08:32.402482397 +0800
> @@ -61,14 +61,12 @@
>
>  int main()
>  {
> -       int ret_pselect, total_sec, fd, total_nsec;
> +       int ret_pselect, fd;
>        fd_set readfds;
> -       struct timespec tv;
>        int retval;
> -       time_t t;
> -       unsigned start, end;
> -       struct timeval tv_start, tv_end;
> -       int real_usec;
> +       struct timespec tv, tv_start, tv_end;
> +       long real_nsec, total_nsec;
> +       int real_sec, total_sec;
>
>        setup();
>
> @@ -103,17 +101,20 @@
>                tst_resm(TINFO,
>                         "Testing basic pselect sanity,Sleeping for %jd secs",
>                         (intmax_t)tv.tv_sec);
> -               start = time(&t);
> +               clock_gettime(CLOCK_REALTIME, &tv_start);
>                retval =
> -                   pselect(0, &readfds, NULL, NULL, (struct timespec *)&tv,
> +                   pselect(0, &readfds, NULL, NULL, &tv,
>                            NULL);
> -               end = time(&t);
> +               clock_gettime(CLOCK_REALTIME, &tv_end);
>
> -               if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
> +               /* do a rounding */
> +               real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec +
> +                               1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec)));
> +               if (real_sec == total_sec)
>                        tst_resm(TPASS, "Sleep time was correct");
>                else
>                        tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
> -                                total_sec, (end - start));
> +                                total_sec, real_sec);
>        }
>
>  #ifdef DEBUG
> @@ -129,19 +130,19 @@
>                tst_resm(TINFO,
>                         "Testing basic pselect sanity,Sleeping for %ld nano secs",
>                         tv.tv_nsec);
> -               gettimeofday(&tv_start, NULL);
> +               clock_gettime(CLOCK_REALTIME, &tv_start);
>                retval =
>                    pselect(0, &readfds, NULL, NULL, &tv,
>                            NULL);
> -               gettimeofday(&tv_end, NULL);
> +               clock_gettime(CLOCK_REALTIME, &tv_end);
>
>                /* Changed total_sec compare to an at least vs an exact compare */
>
> -               real_usec = (tv_end.tv_sec - tv_start.tv_sec) * 1e6 +
> -                       tv_end.tv_usec - tv_start.tv_usec;
> +               real_nsec = (tv_end.tv_sec - tv_start.tv_sec) * 1e9 +
> +                       tv_end.tv_nsec - tv_start.tv_nsec;
>
>                /* allow 10% error*/
> -               if (abs(real_usec - tv.tv_nsec / 1000) < 0.1 * total_nsec / 1000)
> +               if (abs(real_nsec - tv.tv_nsec) < 0.1 * total_nsec)
>                        tst_resm(TPASS, "Sleep time was correct");
>                else {
>                        tst_resm(TWARN,
> @@ -150,8 +151,8 @@
>                                 "due to the limitation of the way it calculates the");
>                        tst_resm(TWARN, "system call execution time.");
>                        tst_resm(TFAIL,
> -                               "Sleep time was incorrect:%d usec vs expected %d usec",
> -                                real_usec, total_nsec / 1000);
> +                               "Sleep time was incorrect:%ld nsec vs expected %ld nsec",
> +                                real_nsec, total_nsec);
>                }
>        }
>        cleanup();

ACK (and thanks for the hard work)!
-Garrett

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Caspar Zhang]

[-- Attachment #1: Type: text/plain, Size: 6336 bytes --]


----- "Garrett Cooper" <yanegomi@gmail.com> wrote:

> On Sun, Jul 18, 2010 at 11:55 PM, Caspar Zhang <czhang@redhat.com>
> wrote:
> >
> > ----- "Garrett Cooper" <yanegomi@gmail.com> wrote:
> >
> >>
> >>     The fact that the test is using time(3) to calculate the
> elapsed
> >> period instead of clock_gettime is dubious at best, because the
> >> precision of time is in seconds. Maybe it should be converted to
> >> clock_gettime calls for improved precision?
> >
> > Use clock_gettime instead of time(3), patch attached.
> 
> Looks better, but being strict it's still buggy. POSIX defines struct
> timespec as:
> 
> The <time.h> header shall declare the structure timespec, which has
> at
> least the following members:
> 
> time_t  tv_sec    Seconds.
> long    tv_nsec   Nanoseconds.
> 
> So you're losing precision in one case, and time_t can be variable
> (the spec for types.h is rather fluid):
> 
> time_t and clock_t shall be integer or real-floating types.
> 
> The (struct timeval *) casts are also unnecessary because you're
> getting the address of a struct timeval object (which is in both
> cases
> struct timeval*).
> 
> > Another question, since POSIX.1-2008 marks gettimeofday() as
> obsolete
> > (man gettimeofday), should the usec test part be changed to
> > clock_getime as well?
> 
> Sure, if you want.

Updated patch, FYI.

Thanks,
Caspar



Hi all,
   
   1. In my recent tests, not all desired time be equal to exact execution 
   time in sec test part, use clock_gettime to improve precision. 
   2. gettimeofday is marked as obsolete in POSIX.1-2008 according to
   man gettimeofday, so change gettimeofday to clock_gettime in nsec test
   part.
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

diff -Naur a/testcases/kernel/syscalls/pselect/Makefile b/testcases/kernel/syscalls/pselect/Makefile
--- a/testcases/kernel/syscalls/pselect/Makefile        2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/Makefile        2010-07-19 14:54:38.984482305 +0800
@@ -21,6 +21,7 @@
 include $(top_srcdir)/include/mk/testcases.mk
 include $(abs_srcdir)/../utils/newer_64.mk
 
+LDLIBS                 += -lpthread -lrt
 %_64: CPPFLAGS += -D_FILE_OFFSET_BITS=64
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
diff -Naur a/testcases/kernel/syscalls/pselect/pselect01.c b/testcases/kernel/syscalls/pselect/pselect01.c
--- a/testcases/kernel/syscalls/pselect/pselect01.c     2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c     2010-07-19 16:08:32.402482397 +0800
@@ -61,14 +61,12 @@
 
 int main()
 {
-       int ret_pselect, total_sec, fd, total_nsec;
+       int ret_pselect, fd;
        fd_set readfds;
-       struct timespec tv;
        int retval;
-       time_t t;
-       unsigned start, end;
-       struct timeval tv_start, tv_end;
-       int real_usec;
+       struct timespec tv, tv_start, tv_end;
+       long real_nsec, total_nsec;
+       int real_sec, total_sec;
 
        setup();
 
@@ -103,17 +101,20 @@
                tst_resm(TINFO,
                         "Testing basic pselect sanity,Sleeping for %jd secs",
                         (intmax_t)tv.tv_sec);
-               start = time(&t);
+               clock_gettime(CLOCK_REALTIME, &tv_start);
                retval =
-                   pselect(0, &readfds, NULL, NULL, (struct timespec *)&tv,
+                   pselect(0, &readfds, NULL, NULL, &tv,
                            NULL);
-               end = time(&t);
+               clock_gettime(CLOCK_REALTIME, &tv_end);
 
-               if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+               /* do a rounding */
+               real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec + 
+                               1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec)));
+               if (real_sec == total_sec)
                        tst_resm(TPASS, "Sleep time was correct");
                else
                        tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
-                                total_sec, (end - start));
+                                total_sec, real_sec);
        }
 
 #ifdef DEBUG
@@ -129,19 +130,19 @@
                tst_resm(TINFO,
                         "Testing basic pselect sanity,Sleeping for %ld nano secs",
                         tv.tv_nsec);
-               gettimeofday(&tv_start, NULL);
+               clock_gettime(CLOCK_REALTIME, &tv_start);
                retval =
                    pselect(0, &readfds, NULL, NULL, &tv,
                            NULL);
-               gettimeofday(&tv_end, NULL);
+               clock_gettime(CLOCK_REALTIME, &tv_end);
 
                /* Changed total_sec compare to an at least vs an exact compare */
 
-               real_usec = (tv_end.tv_sec - tv_start.tv_sec) * 1e6 +
-                       tv_end.tv_usec - tv_start.tv_usec;
+               real_nsec = (tv_end.tv_sec - tv_start.tv_sec) * 1e9 +
+                       tv_end.tv_nsec - tv_start.tv_nsec;
 
                /* allow 10% error*/
-               if (abs(real_usec - tv.tv_nsec / 1000) < 0.1 * total_nsec / 1000)
+               if (abs(real_nsec - tv.tv_nsec) < 0.1 * total_nsec)
                        tst_resm(TPASS, "Sleep time was correct");
                else {
                        tst_resm(TWARN,
@@ -150,8 +151,8 @@
                                 "due to the limitation of the way it calculates the");
                        tst_resm(TWARN, "system call execution time.");
                        tst_resm(TFAIL,
-                               "Sleep time was incorrect:%d usec vs expected %d usec",
-                                real_usec, total_nsec / 1000);
+                               "Sleep time was incorrect:%ld nsec vs expected %ld nsec",
+                                real_nsec, total_nsec);
                }
        }
        cleanup();


-- 
Kernel Associate Quality Engineer
Red Hat Inc. (Beijing R&D Branch)

Red Hat China R&D Branch Unit 907, North Tower C, 
  Raycom Infotech Park, No.2 Kexueyuan Nanlu, 
  Haidian District, Beijing 100190

TEL: +86-10-62608150
Web: http://www.redhat.com/

[-- Attachment #2: syscalls-pselect01-fix-false-alarm.patch --]
[-- Type: text/x-patch, Size: 3550 bytes --]

Hi all,
   
   1. In my recent tests, not all desired time be equal to exact execution 
   time in sec test part, use clock_gettime to improve precision. 
   2. gettimeofday is marked as obsolete in POSIX.1-2008 according to
   man gettimeofday, so change gettimeofday to clock_gettime in nsec test
   part.
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

diff -Naur a/testcases/kernel/syscalls/pselect/Makefile b/testcases/kernel/syscalls/pselect/Makefile
--- a/testcases/kernel/syscalls/pselect/Makefile	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/Makefile	2010-07-19 14:54:38.984482305 +0800
@@ -21,6 +21,7 @@
 include $(top_srcdir)/include/mk/testcases.mk
 include $(abs_srcdir)/../utils/newer_64.mk
 
+LDLIBS			+= -lpthread -lrt
 %_64: CPPFLAGS += -D_FILE_OFFSET_BITS=64
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
diff -Naur a/testcases/kernel/syscalls/pselect/pselect01.c b/testcases/kernel/syscalls/pselect/pselect01.c
--- a/testcases/kernel/syscalls/pselect/pselect01.c	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-19 16:08:32.402482397 +0800
@@ -61,14 +61,12 @@
 
 int main()
 {
-	int ret_pselect, total_sec, fd, total_nsec;
+	int ret_pselect, fd;
 	fd_set readfds;
-	struct timespec tv;
 	int retval;
-	time_t t;
-	unsigned start, end;
-	struct timeval tv_start, tv_end;
-	int real_usec;
+	struct timespec tv, tv_start, tv_end;
+	long real_nsec, total_nsec;
+	int real_sec, total_sec;
 
 	setup();
 
@@ -103,17 +101,20 @@
 		tst_resm(TINFO,
 			 "Testing basic pselect sanity,Sleeping for %jd secs",
 			 (intmax_t)tv.tv_sec);
-		start = time(&t);
+		clock_gettime(CLOCK_REALTIME, &tv_start);
 		retval =
-		    pselect(0, &readfds, NULL, NULL, (struct timespec *)&tv,
+		    pselect(0, &readfds, NULL, NULL, &tv,
 			    NULL);
-		end = time(&t);
+		clock_gettime(CLOCK_REALTIME, &tv_end);
 
-		if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+		/* do a rounding */
+		real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec + 
+				1e-9 * (tv_end.tv_nsec - tv_start.tv_nsec)));
+		if (real_sec == total_sec)
 			tst_resm(TPASS, "Sleep time was correct");
 		else
 			tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
-				 total_sec, (end - start));
+				 total_sec, real_sec);
 	}
 
 #ifdef DEBUG
@@ -129,19 +130,19 @@
 		tst_resm(TINFO,
 			 "Testing basic pselect sanity,Sleeping for %ld nano secs",
 			 tv.tv_nsec);
-		gettimeofday(&tv_start, NULL);
+		clock_gettime(CLOCK_REALTIME, &tv_start);
 		retval =
 		    pselect(0, &readfds, NULL, NULL, &tv,
 			    NULL);
-		gettimeofday(&tv_end, NULL);
+		clock_gettime(CLOCK_REALTIME, &tv_end);
 
 		/* Changed total_sec compare to an at least vs an exact compare */
 
-		real_usec = (tv_end.tv_sec - tv_start.tv_sec) * 1e6 +
-			tv_end.tv_usec - tv_start.tv_usec;
+		real_nsec = (tv_end.tv_sec - tv_start.tv_sec) * 1e9 +
+			tv_end.tv_nsec - tv_start.tv_nsec;
 
 		/* allow 10% error*/
-		if (abs(real_usec - tv.tv_nsec / 1000) < 0.1 * total_nsec / 1000)
+		if (abs(real_nsec - tv.tv_nsec) < 0.1 * total_nsec)
 			tst_resm(TPASS, "Sleep time was correct");
 		else {
 			tst_resm(TWARN,
@@ -150,8 +151,8 @@
 				 "due to the limitation of the way it calculates the");
 			tst_resm(TWARN, "system call execution time.");
 			tst_resm(TFAIL,
-				"Sleep time was incorrect:%d usec vs expected %d usec",
-				 real_usec, total_nsec / 1000);
+				"Sleep time was incorrect:%ld nsec vs expected %ld nsec",
+				 real_nsec, total_nsec);
 		}
 	}
 	cleanup();

[-- Attachment #3: Type: text/plain, Size: 235 bytes --]

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

[-- Attachment #4: Type: text/plain, Size: 155 bytes --]

_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Garrett Cooper]

On Sun, Jul 18, 2010 at 11:55 PM, Caspar Zhang <czhang@redhat.com> wrote:
>
> ----- "Garrett Cooper" <yanegomi@gmail.com> wrote:
>
>>
>>     The fact that the test is using time(3) to calculate the elapsed
>> period instead of clock_gettime is dubious at best, because the
>> precision of time is in seconds. Maybe it should be converted to
>> clock_gettime calls for improved precision?
>
> Use clock_gettime instead of time(3), patch attached.

Looks better, but being strict it's still buggy. POSIX defines struct
timespec as:

The <time.h> header shall declare the structure timespec, which has at
least the following members:

time_t  tv_sec    Seconds.
long    tv_nsec   Nanoseconds.

So you're losing precision in one case, and time_t can be variable
(the spec for types.h is rather fluid):

time_t and clock_t shall be integer or real-floating types.

The (struct timeval *) casts are also unnecessary because you're
getting the address of a struct timeval object (which is in both cases
struct timeval*).

> Another question, since POSIX.1-2008 marks gettimeofday() as obsolete
> (man gettimeofday), should the usec test part be changed to
> clock_getime as well?

Sure, if you want.

Thanks!
-Garrett

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Caspar Zhang]

[-- Attachment #1: Type: text/plain, Size: 905 bytes --]


----- "Caspar Zhang" <czhang@redhat.com> wrote:

> ----- "Garrett Cooper" <yanegomi@gmail.com> wrote:
> 
> >
> >     The fact that the test is using time(3) to calculate the elapsed
> > period instead of clock_gettime is dubious at best, because the
> > precision of time is in seconds. Maybe it should be converted to
> > clock_gettime calls for improved precision?

s/1e-6/1e-9 in the patch, my typo :-|

Thanks,
Caspar

> 
> Use clock_gettime instead of time(3), patch attached.
> 
> Another question, since POSIX.1-2008 marks gettimeofday() as obsolete
> (man gettimeofday), should the usec test part be changed to
> clock_getime as well?
> 


-- 
Kernel Associate Quality Engineer
Red Hat Inc. (Beijing R&D Branch)

Red Hat China R&D Branch Unit 907, North Tower C, 
  Raycom Infotech Park, No.2 Kexueyuan Nanlu, 
  Haidian District, Beijing 100190

TEL: +86-10-62608150
Web: http://www.redhat.com/

[-- Attachment #2: syscalls-pselect01-fix-false-alarm.patch --]
[-- Type: text/x-patch, Size: 1795 bytes --]

Hi all,
   
   In my recent tests, not all desired time be equal to exact execution time, 
   use clock_gettime to improve precision. 
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

--- a/testcases/kernel/syscalls/pselect/Makefile	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/Makefile	2010-07-19 14:54:38.984482305 +0800
@@ -21,6 +21,7 @@
 include $(top_srcdir)/include/mk/testcases.mk
 include $(abs_srcdir)/../utils/newer_64.mk
 
+LDLIBS			+= -lpthread -lrt
 %_64: CPPFLAGS += -D_FILE_OFFSET_BITS=64
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
--- a/testcases/kernel/syscalls/pselect/pselect01.c	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-19 15:04:24.175607371 +0800
@@ -65,10 +65,9 @@
 	fd_set readfds;
 	struct timespec tv;
 	int retval;
-	time_t t;
-	unsigned start, end;
+	struct timespec tp_start, tp_end;
 	struct timeval tv_start, tv_end;
-	int real_usec;
+	int real_usec, real_sec;
 
 	setup();
 
@@ -103,17 +102,20 @@
 		tst_resm(TINFO,
 			 "Testing basic pselect sanity,Sleeping for %jd secs",
 			 (intmax_t)tv.tv_sec);
-		start = time(&t);
+		clock_gettime(CLOCK_REALTIME, &tp_start);
 		retval =
 		    pselect(0, &readfds, NULL, NULL, (struct timespec *)&tv,
 			    NULL);
-		end = time(&t);
+		clock_gettime(CLOCK_REALTIME, &tp_end);
 
-		if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+		/* do a rounding */
+		real_sec = (int)(0.5 + (tp_end.tv_sec - tp_start.tv_sec + 
+				1e-9 * (tp_end.tv_nsec - tp_start.tv_nsec)));
+		if (real_sec == total_sec)
 			tst_resm(TPASS, "Sleep time was correct");
 		else
 			tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
-				 total_sec, (end - start));
+				 total_sec, real_sec);
 	}
 
 #ifdef DEBUG

[-- Attachment #3: Type: text/plain, Size: 235 bytes --]

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

[-- Attachment #4: Type: text/plain, Size: 155 bytes --]

_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

Re: [LTP] [PATCH] fix false alarm in pselect01

[Caspar Zhang]

[-- Attachment #1: Type: text/plain, Size: 831 bytes --]


----- "Garrett Cooper" <yanegomi@gmail.com> wrote:

> 
>     The fact that the test is using time(3) to calculate the elapsed
> period instead of clock_gettime is dubious at best, because the
> precision of time is in seconds. Maybe it should be converted to
> clock_gettime calls for improved precision?

Use clock_gettime instead of time(3), patch attached. 

Another question, since POSIX.1-2008 marks gettimeofday() as obsolete
(man gettimeofday), should the usec test part be changed to
clock_getime as well?

Thanks,
Caspar

-- 
/---------------------------------------\
| Name: Caspar Zhang                    |
| Team: Kernel-QE                       |
| IRC: caspar @ #kernel-qe, #eng-china  |
| Phone: +86-10-6260-8150               |
| Cellphone: +86-1381-073-0090          |
\---------------------------------------/


[-- Attachment #2: syscalls-pselect01-fix-false-alarm.patch --]
[-- Type: text/x-patch, Size: 1797 bytes --]

Hi all,
   
   In my recent tests, not all desired time be equal to exact execution time, 
   use clock_gettime to improve precision. 
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

--- a/testcases/kernel/syscalls/pselect/Makefile	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/Makefile	2010-07-19 14:54:38.984482305 +0800
@@ -21,6 +21,7 @@
 include $(top_srcdir)/include/mk/testcases.mk
 include $(abs_srcdir)/../utils/newer_64.mk
 
+LDLIBS			+= -lpthread -lrt
 %_64: CPPFLAGS += -D_FILE_OFFSET_BITS=64
 
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
--- a/testcases/kernel/syscalls/pselect/pselect01.c	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-19 14:25:31.900419665 +0800
@@ -65,10 +65,8 @@
 	fd_set readfds;
 	struct timespec tv;
 	int retval;
-	time_t t;
-	unsigned start, end;
 	struct timeval tv_start, tv_end;
-	int real_usec;
+	int real_usec, real_sec;
 
 	setup();
 
@@ -103,17 +101,20 @@
 		tst_resm(TINFO,
 			 "Testing basic pselect sanity,Sleeping for %jd secs",
 			 (intmax_t)tv.tv_sec);
-		start = time(&t);
+		clock_gettime(CLOCK_REALTIME, (struct timespec *)&tv_start);
 		retval =
 		    pselect(0, &readfds, NULL, NULL, (struct timespec *)&tv,
 			    NULL);
-		end = time(&t);
+		clock_gettime(CLOCK_REALTIME, (struct timespec *)&tv_end);
 
-		if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+		/* do a rounding */
+		real_sec = (int)(0.5 + (tv_end.tv_sec - tv_start.tv_sec + 
+				1e-6 * (tv_end.tv_usec - tv_start.tv_usec)));
+		if (real_sec == total_sec)
 			tst_resm(TPASS, "Sleep time was correct");
 		else
 			tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",
-				 total_sec, (end - start));
+				 total_sec, real_sec);
 	}
 
 #ifdef DEBUG

[-- Attachment #3: Type: text/plain, Size: 235 bytes --]

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

[-- Attachment #4: Type: text/plain, Size: 155 bytes --]

_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list

[LTP] [PATCH] fix false alarm in pselect01

[Caspar Zhang]

[-- Attachment #1: Type: text/plain, Size: 1226 bytes --]

Hi all,
   
   In my recent tests, not all desired time is equal to exact execution time, 
   sometimes it is +/-1s, and in a few times it becomes +/-2s. In most cases,
   +/-2s is still a valid value between exact execute time and desired time.
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

--- a/testcases/kernel/syscalls/pselect/pselect01.c     2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c     2010-07-15 16:03:40.969426078 +0800
@@ -109,7 +109,8 @@
                            NULL);
                end = time(&t);
 
-               if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+               /* Is it still valid in range -2~2s? */
+               if ((end - start - total_sec <= 2) && (end - start - total_sec >= -2))
                        tst_resm(TPASS, "Sleep time was correct");
                else
                        tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",

-- 
Kernel Associate Quality Engineer
Red Hat Inc. (Beijing R&D Branch)

Red Hat China R&D Branch Unit 907, North Tower C, 
  Raycom Infotech Park, No.2 Kexueyuan Nanlu, 
  Haidian District, Beijing 100190

TEL: +86-10-62608150
Web: http://www.redhat.com/

[-- Attachment #2: syscalls-pselect01-fix-false-alarm.patch --]
[-- Type: text/x-patch, Size: 836 bytes --]

Hi all,
   
   In my recent tests, not all desired time is equal to exact execution time, 
   sometimes it is +/-1s, and in a few times it becomes +/-2s. In most cases,
   +/-2s is still a valid value between exact execute time and desired time.
   
   Signed-off-by: Caspar Zhang <czhang@redhat.com>

--- a/testcases/kernel/syscalls/pselect/pselect01.c	2010-04-01 14:23:10.000000000 +0800
+++ b/testcases/kernel/syscalls/pselect/pselect01.c	2010-07-15 16:03:40.969426078 +0800
@@ -109,7 +109,8 @@
 			    NULL);
 		end = time(&t);
 
-		if (((end - start) == total_sec) || ((end - start) == total_sec + 1))
+		/* Is it still valid in range -2~2s? */
+		if ((end - start - total_sec <= 2) && (end - start - total_sec >= -2))
 			tst_resm(TPASS, "Sleep time was correct");
 		else
 			tst_resm(TFAIL, "Sleep time was incorrect:%d != %d",

[-- Attachment #3: Type: text/plain, Size: 235 bytes --]

------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first

[-- Attachment #4: Type: text/plain, Size: 155 bytes --]

_______________________________________________
Ltp-list mailing list
Ltp-list@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ltp-list