Re: More RT Test Programs

Re: More RT Test Programs

[John Kacur]

On Mon, 14 Dec 2009, Carsten Emde wrote:

> John,
> 
> > So, for now I am going to at least temporarily submit a patch that
> > removes building the modules from the main makefile.
> That's okay.
> 
> > Of course you can still cd to src/backfire and build the module there
> > without any problems.
> Yes. Shouldn't we then explain this a little bit better in sendme?
> 
> Carsten.
> 
> 
> Signed-off-by: Carsten Emde <C.Emde@osadl.org>
> 
> --- sendme-old.c        2009-12-14 22:59:45.114507575 +0100
> +++ sendme.c    2009-12-14 23:07:48.714144429 +0100
> @@ -232,7 +232,12 @@
>        path = open("/dev/backfire", O_RDWR);
>        if (path < 0) {
>                fprintf(stderr, "ERROR: Could not access backfire device, "
> -                               "try 'modprobe backfire'\n");
> +                               "try 'modprobe backfire'.\n"
> +                               "If the module backfire can't be loaded, "
> +                               "it may need to be built first.\n"
> +                               "Execute 'cd src/backfire; make' in the "
> +                               "rt-tests directory (requires rt-tests\n"
> +                               "sources and kernel-devel package).\n");
>                return 1;
>        }
>        fl.l_type = F_WRLCK;
> 

Thanks Carsten - I applied it to my git tree and pushed it.
Clark - pls pull.

Re: More RT Test Programs

[Carsten Emde]

Clark,

by some reason, this patch didn't make it into 0.57. Would you mind to
pull it from John's tree?

Thanks,

	Carsten.

Re: More RT Test Programs

[John Kacur]

On Sun, Dec 20, 2009 at 9:42 PM, Carsten Emde <Carsten.Emde@osadl.org> wrote:
> Clark,
>
> by some reason, this patch didn't make it into 0.57. Would you mind to
> pull it from John's tree?
>
> Thanks,
>
>        Carsten.

Hi Carsten - I think that Clark pulled my tree after I grabbed this
last patch. We are aware of it, and it's lined-up for the
next round of updates. I'll be posting some more tomorrow.

Thanks.
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: More RT Test Programs

[Carsten Emde]

[-- Attachment #1: Type: text/plain, Size: 1121 bytes --]

John,

> [..] I'll be posting some more tomorrow.
Okay, great.

I have found an issue with all source files of the new tests. At some
stage between the original files and Clark's tree, the line delimiters
changed from NL to CR/NL. Would be great, if you could move this back to NL.

A second issue is related to the removal of the getcpu() definition.
There is a leftover in svsematest.c which should also be removed for
consistency. However, this and the other tests no longer compile in EL5
without this definition - should we better revert this and provide a
working recognition through "grep sched_getcpu
/usr/include/bits/sched.h"? Or is EL5 irrelevant in this context? I
still have an EL5 test machine where all RT kernels are tested.

Another issue is in the file backfire.c After you removed the line
>> -	sed s/__VERSION_STRING__/$(VERSION_STRING)/ [..]
from the Makefile, the driver is producing the rather ugly syslog message:
  backfire driver v__VERSION_STRING__
instead of
  backfire driver v0.57
or similar.

We better completely remove the version display.

Signed-off-by: Carsten Emde <C.Emde@osadl.org>

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: fix-backfire-version-display.patch --]
[-- Type: text/x-patch; name="fix-backfire-version-display.patch", Size: 396 bytes --]

--- backfire.c-orig	2009-12-20 22:40:48.000000000 +0100
+++ backfire.c	2009-12-20 22:41:40.000000000 +0100
@@ -132,7 +132,7 @@
 	if (ret)
 		printk(KERN_ERR "backfire: can't register dynamic misc device\n");
 	else
-		printk(KERN_INFO "backfire driver v__VERSION_STRING__ misc device %d\n",
+		printk(KERN_INFO "backfire driver misc device %d\n",
 			backfire_dev.minor);
 	return ret;
 }

Re: More RT Test Programs

[John Kacur]

(Carsten - resend to cc all)

On Mon, Dec 21, 2009 at 12:49 AM, Carsten Emde <Carsten.Emde@osadl.org> wrote:
> John,
>
>> [..] I'll be posting some more tomorrow.
> Okay, great.
>
> I have found an issue with all source files of the new tests. At some
> stage between the original files and Clark's tree, the line delimiters
> changed from NL to CR/NL. Would be great, if you could move this back to NL.

Weird - I'll remove them.
It's not coming from you is it? The patch that you attached here has
^M at the end
of each line. I can run that through a script to remove it - but you
should look into
where that is coming from on your end.

>
> A second issue is related to the removal of the getcpu() definition.
> There is a leftover in svsematest.c which should also be removed for
> consistency. However, this and the other tests no longer compile in EL5
> without this definition - should we better revert this and provide a
> working recognition through "grep sched_getcpu
> /usr/include/bits/sched.h"? Or is EL5 irrelevant in this context? I
> still have an EL5 test machine where all RT kernels are tested.


Yup - I've been struggling with a resolution for this issue, I have something
to post tomorrow after my testing is complete. EL5 is not irrelevant to us.
Do you test 32-bit as well as 64-bit?


>
> Another issue is in the file backfire.c After you removed the line
>>> -    sed s/__VERSION_STRING__/$(VERSION_STRING)/ [..]
> from the Makefile, the driver is producing the rather ugly syslog message:
>  backfire driver v__VERSION_STRING__
> instead of
>  backfire driver v0.57
> or similar.
>
> We better completely remove the version display.

Thanks for the patch, I will apply it.

>
> Signed-off-by: Carsten Emde <C.Emde@osadl.org>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: More RT Test Programs

[Carsten Emde]

[-- Attachment #1: Type: text/plain, Size: 1838 bytes --]

John,

>> I have found an issue with all source files of the new tests. At some
>> stage between the original files and Clark's tree, the line delimiters
>> changed from NL to CR/NL. Would be great, if you could move this back to NL.
> Weird - I'll remove them.
> It's not coming from you is it? The patch that you attached here has
> ^M at the end of each line.
Yes, this is because of the ^Ms in the original file. I had to use
diff's -w option.

> I can run that through a script to remove it - but you
> should look into where that is coming from on your end.
Okay. Here comes another patch. Let's see in what shape it arrives at
your side. I converted the files and did not use diff's -w option this
time. You need to remove the ^Ms from the patched files; otherwise, the
patch will not apply.

>> A second issue is related to the removal of the getcpu() definition.
>> There is a leftover in svsematest.c which should also be removed for
>> consistency. However, this and the other tests no longer compile in EL5
>> without this definition - should we better revert this and provide a
>> working recognition through "grep sched_getcpu
>> /usr/include/bits/sched.h"? Or is EL5 irrelevant in this context? I
>> still have an EL5 test machine where all RT kernels are tested.
> Yup - I've been struggling with a resolution for this issue, I have something
> to post tomorrow after my testing is complete. EL5 is not irrelevant to us.
Below comes a proposal of a patch that works here on EL5.

> Do you test 32-bit as well as 64-bit?
No - currently, I have only an F11 machine that runs 64-bit. Is there an
issue with EL5 64-bit?

	Carsten.


-=-------------------------------------------------------------------=-
Generate getcpu.h to use getcpu() instead of sched_getcpu(), if needed.

Signed-off-by: Carsten Emde <C.Emde@osadl.org>


[-- Attachment #2: use-getcpu-if-sched_getcpu-not-available.patch --]
[-- Type: text/x-patch, Size: 4845 bytes --]

Index: rt-tests/Makefile
===================================================================
--- rt-tests.orig/Makefile
+++ rt-tests/Makefile
@@ -20,7 +20,11 @@ endif
 UTILS	= src/lib/rt-utils.o
 
 .PHONY: all
-all: $(TARGETS)
+all:  src/lib/getcpu.h $(TARGETS)
+
+src/lib/getcpu.h:
+	@touch $@
+	@grep -q sched_getcpu /usr/include/bits/sched.h && echo "#define HAS_SCHED_GETCPU" >$@; true
 
 cyclictest: src/cyclictest/cyclictest.c $(UTILS)
 	$(CC) $(CFLAGS) -D VERSION_STRING=$(VERSION_STRING) $^ -o $@ $(LIBS)
@@ -59,7 +63,7 @@ CLEANUP += $(if $(wildcard .git), Change
 .PHONY: clean
 clean:
 	for F in $(CLEANUP); do find -type f -name $$F | xargs rm -f; done
-	rm -f hwlatdetect
+	rm -f hwlatdetect src/lib/getcpu.h
 
 .PHONY: distclean
 distclean: clean
Index: rt-tests/src/backfire/sendme.c
===================================================================
--- rt-tests.orig/src/backfire/sendme.c
+++ rt-tests/src/backfire/sendme.c
@@ -30,6 +30,7 @@
 #include <string.h>
 #include <time.h>
 #include "rt-utils.h"
+#include "getcpu.h"
 
 #define _GNU_SOURCE
 #include <utmpx.h>
@@ -38,11 +39,17 @@
 #include <sys/time.h>
 #include <sys/mman.h>
 
+#include <linux/unistd.h>
+
 #define USEC_PER_SEC		1000000
 #define NSEC_PER_SEC		1000000000
 
 #define SIGTEST SIGHUP
 
+#ifndef HAS_SCHED_GETCPU
+#define getcpu(cpu, node, cache) syscall(__NR_getcpu, cpu, node, cache)
+#endif
+
 enum {
 	AFFINITY_UNSPECIFIED,
 	AFFINITY_SPECIFIED,
@@ -197,8 +204,15 @@ int main(int argc, char *argv[])
 
 	if (setaffinity != AFFINITY_UNSPECIFIED) {
 		CPU_ZERO(&mask);
-		if (setaffinity == AFFINITY_USECURRENT)
+		if (setaffinity == AFFINITY_USECURRENT) {
+#ifdef HAS_SCHED_GETCPU
 			affinity = sched_getcpu();
+#else
+			int c, s;
+                        s = getcpu(&c, NULL, NULL);
+		        affinity = (s == -1) ? s : c;
+#endif
+                }
 		CPU_SET(affinity, &mask);
 		if (sched_setaffinity(0, sizeof(mask), &mask) == -1)
 			fprintf(stderr,	"WARNING: Could not set CPU affinity "
Index: rt-tests/src/ptsematest/ptsematest.c
===================================================================
--- rt-tests.orig/src/ptsematest/ptsematest.c
+++ rt-tests/src/ptsematest/ptsematest.c
@@ -35,11 +35,15 @@
 #include <linux/unistd.h>
 #include <utmpx.h>
 #include "rt-utils.h"
+#include "getcpu.h"
 
 #define __USE_GNU
 #include <pthread.h>
 
 #define gettid() syscall(__NR_gettid)
+#ifndef HAS_SCHED_GETCPU
+#define getcpu(cpu, node, cache) syscall(__NR_getcpu, cpu, node, cache)
+#endif
 
 #define USEC_PER_SEC 1000000
 
@@ -108,7 +112,13 @@ void *semathread(void *param)
 			if(par->max_cycles && par->samples >= par->max_cycles)
 				par->shutdown = 1;
 			if (mustgetcpu) {
+#ifdef HAS_SCHED_GETCPU
 				par->cpu = sched_getcpu();
+#else
+				int c, s;
+	                        s = getcpu(&c, NULL, NULL);
+			        par->cpu = (s == -1) ? s : c;
+#endif
 			}
 		} else {
 			/* Receiver */
@@ -150,7 +160,13 @@ void *semathread(void *param)
 			if (par->max_cycles && par->samples >= par->max_cycles)
 				par->shutdown = 1;
 			if (mustgetcpu) {
+#ifdef HAS_SCHED_GETCPU
 				par->cpu = sched_getcpu();
+#else
+				int c, s;
+	                        s = getcpu(&c, NULL, NULL);
+			        par->cpu = (s == -1) ? s : c;
+#endif
 		        }
 			nanosleep(&par->delay, NULL);
 			pthread_mutex_unlock(&syncmutex[par->num]);
Index: rt-tests/src/sigwaittest/sigwaittest.c
===================================================================
--- rt-tests.orig/src/sigwaittest/sigwaittest.c
+++ rt-tests/src/sigwaittest/sigwaittest.c
@@ -37,11 +37,15 @@
 #include <linux/unistd.h>
 #include <utmpx.h>
 #include "rt-utils.h"
+#include "getcpu.h"
 
 #define __USE_GNU
 #include <pthread.h>
 
 #define gettid() syscall(__NR_gettid)
+#ifndef HAS_SCHED_GETCPU
+#define getcpu(cpu, node, cache) syscall(__NR_getcpu, cpu, node, cache)
+#endif
 
 #define USEC_PER_SEC 1000000
 
@@ -139,7 +143,13 @@ void *semathread(void *param)
 				par->shutdown = 1;
 
 			if (mustgetcpu) {
+#ifdef HAS_SCHED_GETCPU
 				par->cpu = sched_getcpu();
+#else
+				int c, s;
+	                        s = getcpu(&c, NULL, NULL);
+			        par->cpu = (s == -1) ? s : c;
+#endif
 			}
 			sigwait(&sigset, &sig);
 		} else {
@@ -163,7 +173,13 @@ void *semathread(void *param)
 				par->shutdown = 1;
 
 			if (mustgetcpu) {
+#ifdef HAS_SCHED_GETCPU
 				par->cpu = sched_getcpu();
+#else
+				int c, s;
+	                        s = getcpu(&c, NULL, NULL);
+			        par->cpu = (s == -1) ? s : c;
+#endif
 		        }
 			/*
 			 * Latency is the time spent between sending and
Index: rt-tests/src/svsematest/svsematest.c
===================================================================
--- rt-tests.orig/src/svsematest/svsematest.c
+++ rt-tests/src/svsematest/svsematest.c
@@ -42,8 +42,7 @@
 #include <sys/time.h>
 #include <sys/mman.h>
 #include "rt-utils.h"