zdtm/static/maps06: removed hardcoded page-size

Submitted by Adrian Reber on June 5, 2018, 7:58 p.m.

Details

Message ID 1528228739-27731-1-git-send-email-adrian@lisas.de
State Accepted
Series "zdtm/static/maps06: removed hardcoded page-size"
Commit 0bb02253e67c75bb3829d993b6622e9742d1fe22
Headers show

Commit Message

Adrian Reber June 5, 2018, 7:58 p.m.
From: Adrian Reber <areber@redhat.com>

zdtm/static/maps06 failed on systems with different page-size than 4096.
This changes maps06 to use sysconf(_SC_PAGESIZE) instead.

Signed-off-by: Adrian Reber <areber@redhat.com>
---
 test/zdtm/static/maps06.c | 25 +++++++++++++++----------
 1 file changed, 15 insertions(+), 10 deletions(-)

Patch hide | download patch | download mbox

diff --git a/test/zdtm/static/maps06.c b/test/zdtm/static/maps06.c
index 14dd90f..7480d6b 100644
--- a/test/zdtm/static/maps06.c
+++ b/test/zdtm/static/maps06.c
@@ -10,12 +10,12 @@  const char *test_author	= "Andrei Vagin <avagin@openvz.org>";
 char *filename;
 TEST_OPTION(filename, string, "file name", 1);
 
-#define TEST_SIZE 10240
-
 int main(int argc, char ** argv)
 {
 	void *start;
 	int fd, i;
+	int ps = sysconf(_SC_PAGESIZE);
+	int test_size;
 
 	test_init(argc, argv);
 
@@ -23,21 +23,26 @@  int main(int argc, char ** argv)
 	if (fd < 0)
 		return 1;
 
-	ftruncate(fd, 4096);
+	ftruncate(fd, ps);
+
+	if (ps == 0x1000)
+		test_size = 10240;
+	else
+		test_size = 512;
 
-	start = mmap(0, 4096 * TEST_SIZE * 4, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
+	start = mmap(0, ps * test_size * 4, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
 	if (start == MAP_FAILED)
 		return 1;
 
-	for (i = 0; i < TEST_SIZE; i++) {
+	for (i = 0; i < test_size; i++) {
 		int *addr;
-		addr = mmap(start + i * 3 * 4096, 4096,
+		addr = mmap(start + i * 3 * ps, ps,
 				PROT_READ | PROT_WRITE,
 				MAP_PRIVATE | MAP_FILE | MAP_FIXED, fd, 0);
 		if (addr == MAP_FAILED)
 			return 1;
 		addr[0] = i * 2;
-		addr = mmap(start + (i * 3 + 1) * 4096, 4096,
+		addr = mmap(start + (i * 3 + 1) * ps, ps,
 				PROT_READ | PROT_WRITE,
 				MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
 		if (addr == MAP_FAILED)
@@ -49,12 +54,12 @@  int main(int argc, char ** argv)
 
 	test_waitsig();
 
-	for (i = 0; i < TEST_SIZE; i++) {
+	for (i = 0; i < test_size; i++) {
 		int *addr;
-		addr = start + i * 3 * 4096;
+		addr = start + i * 3 * ps;
 		if (addr[0] != i * 2)
 			fail();
-		addr = start + (i * 3  + 1) * 4096;
+		addr = start + (i * 3  + 1) * ps;
 		if (addr[0] != i)
 			fail();
 	}

Comments

Dmitry Safonov June 6, 2018, 2:44 p.m.
Hi Adrian,

Thanks for doing this!

2018-06-05 20:58 GMT+01:00 Adrian Reber <adrian@lisas.de>:
[..]
> @@ -23,21 +23,26 @@ int main(int argc, char ** argv)
>         if (fd < 0)
>                 return 1;
>
> -       ftruncate(fd, 4096);
> +       ftruncate(fd, ps);
> +
> +       if (ps == 0x1000)
> +               test_size = 10240;
> +       else
> +               test_size = 512;

Is it worth to calculate test_size dynamically based on ps?
Like will the test work on both 16k and 64k pages?

Other than that LGTM,

Reviewed-by: Dmitry Safonov <0x7f454c46@gmail.com>
Adrian Reber June 6, 2018, 2:52 p.m.
On Wed, Jun 06, 2018 at 03:44:32PM +0100, Dmitry Safonov wrote:
> 2018-06-05 20:58 GMT+01:00 Adrian Reber <adrian@lisas.de>:
> [..]
> > @@ -23,21 +23,26 @@ int main(int argc, char ** argv)
> >         if (fd < 0)
> >                 return 1;
> >
> > -       ftruncate(fd, 4096);
> > +       ftruncate(fd, ps);
> > +
> > +       if (ps == 0x1000)
> > +               test_size = 10240;
> > +       else
> > +               test_size = 512;
> 
> Is it worth to calculate test_size dynamically based on ps?
> Like will the test work on both 16k and 64k pages?

Without that the test failed as the result (ps * test_size *4) for mmap()'s
length overflowed. I have not checked if it works when changing ps or
test_size to a 64bit value.

		Adrian
Dmitry Safonov June 6, 2018, 3:12 p.m.
2018-06-06 15:52 GMT+01:00 Adrian Reber <areber@redhat.com>:
> On Wed, Jun 06, 2018 at 03:44:32PM +0100, Dmitry Safonov wrote:
>> 2018-06-05 20:58 GMT+01:00 Adrian Reber <adrian@lisas.de>:
>> [..]
>> > @@ -23,21 +23,26 @@ int main(int argc, char ** argv)
>> >         if (fd < 0)
>> >                 return 1;
>> >
>> > -       ftruncate(fd, 4096);
>> > +       ftruncate(fd, ps);
>> > +
>> > +       if (ps == 0x1000)
>> > +               test_size = 10240;
>> > +       else
>> > +               test_size = 512;
>>
>> Is it worth to calculate test_size dynamically based on ps?
>> Like will the test work on both 16k and 64k pages?
>
> Without that the test failed as the result (ps * test_size *4) for mmap()'s
> length overflowed. I have not checked if it works when changing ps or
> test_size to a 64bit value.

No, what I meant is that you've chosen 512 static value.
Will the test work both for 16k and 64k pages?
Or test_size will differ between 16k/64k - then probably
it would be worth to calculate it with a division.

Thanks,
             Dmitry
Adrian Reber June 6, 2018, 3:47 p.m.
On Wed, Jun 06, 2018 at 04:12:51PM +0100, Dmitry Safonov wrote:
> 2018-06-06 15:52 GMT+01:00 Adrian Reber <areber@redhat.com>:
> > On Wed, Jun 06, 2018 at 03:44:32PM +0100, Dmitry Safonov wrote:
> >> 2018-06-05 20:58 GMT+01:00 Adrian Reber <adrian@lisas.de>:
> >> [..]
> >> > @@ -23,21 +23,26 @@ int main(int argc, char ** argv)
> >> >         if (fd < 0)
> >> >                 return 1;
> >> >
> >> > -       ftruncate(fd, 4096);
> >> > +       ftruncate(fd, ps);
> >> > +
> >> > +       if (ps == 0x1000)
> >> > +               test_size = 10240;
> >> > +       else
> >> > +               test_size = 512;
> >>
> >> Is it worth to calculate test_size dynamically based on ps?
> >> Like will the test work on both 16k and 64k pages?
> >
> > Without that the test failed as the result (ps * test_size *4) for mmap()'s
> > length overflowed. I have not checked if it works when changing ps or
> > test_size to a 64bit value.
> 
> No, what I meant is that you've chosen 512 static value.
> Will the test work both for 16k and 64k pages?
> Or test_size will differ between 16k/64k - then probably
> it would be worth to calculate it with a division.

I first thought that was what you were asking, but then I was not so
sure any more. Yes, for 16K pages it is not perfect. That is correct.

I only tested with 64K a 4K pages.

I thought about a division, but wanted to keep it simple.

I can do a new version with a division.

		Adrian
Dmitry Safonov June 6, 2018, 4:06 p.m.
2018-06-06 16:47 GMT+01:00 Adrian Reber <areber@redhat.com>:
> On Wed, Jun 06, 2018 at 04:12:51PM +0100, Dmitry Safonov wrote:
>> 2018-06-06 15:52 GMT+01:00 Adrian Reber <areber@redhat.com>:
>> > On Wed, Jun 06, 2018 at 03:44:32PM +0100, Dmitry Safonov wrote:
>> >> 2018-06-05 20:58 GMT+01:00 Adrian Reber <adrian@lisas.de>:
>> >> [..]
>> >> > @@ -23,21 +23,26 @@ int main(int argc, char ** argv)
>> >> >         if (fd < 0)
>> >> >                 return 1;
>> >> >
>> >> > -       ftruncate(fd, 4096);
>> >> > +       ftruncate(fd, ps);
>> >> > +
>> >> > +       if (ps == 0x1000)
>> >> > +               test_size = 10240;
>> >> > +       else
>> >> > +               test_size = 512;
>> >>
>> >> Is it worth to calculate test_size dynamically based on ps?
>> >> Like will the test work on both 16k and 64k pages?
>> >
>> > Without that the test failed as the result (ps * test_size *4) for mmap()'s
>> > length overflowed. I have not checked if it works when changing ps or
>> > test_size to a 64bit value.
>>
>> No, what I meant is that you've chosen 512 static value.
>> Will the test work both for 16k and 64k pages?
>> Or test_size will differ between 16k/64k - then probably
>> it would be worth to calculate it with a division.
>
> I first thought that was what you were asking, but then I was not so
> sure any more. Yes, for 16K pages it is not perfect. That is correct.
>
> I only tested with 64K a 4K pages.
>
> I thought about a division, but wanted to keep it simple.
>
> I can do a new version with a division.

I'm not really sure if it's worth to create a new version with a division.
If it you expect it to work with the current code on 16k, it's fine for me.
I haven't carefully looked at code to catch if it will or will not work
with 512 value.

Thanks,
             Dmitry
Andrey Vagin June 18, 2018, 6:02 p.m.
Applied, thanks!

On Tue, Jun 05, 2018 at 07:58:59PM +0000, Adrian Reber wrote:
> From: Adrian Reber <areber@redhat.com>
> 
> zdtm/static/maps06 failed on systems with different page-size than 4096.
> This changes maps06 to use sysconf(_SC_PAGESIZE) instead.
> 
> Signed-off-by: Adrian Reber <areber@redhat.com>
> ---
>  test/zdtm/static/maps06.c | 25 +++++++++++++++----------
>  1 file changed, 15 insertions(+), 10 deletions(-)
> 
> diff --git a/test/zdtm/static/maps06.c b/test/zdtm/static/maps06.c
> index 14dd90f..7480d6b 100644
> --- a/test/zdtm/static/maps06.c
> +++ b/test/zdtm/static/maps06.c
> @@ -10,12 +10,12 @@ const char *test_author	= "Andrei Vagin <avagin@openvz.org>";
>  char *filename;
>  TEST_OPTION(filename, string, "file name", 1);
>  
> -#define TEST_SIZE 10240
> -
>  int main(int argc, char ** argv)
>  {
>  	void *start;
>  	int fd, i;
> +	int ps = sysconf(_SC_PAGESIZE);
> +	int test_size;
>  
>  	test_init(argc, argv);
>  
> @@ -23,21 +23,26 @@ int main(int argc, char ** argv)
>  	if (fd < 0)
>  		return 1;
>  
> -	ftruncate(fd, 4096);
> +	ftruncate(fd, ps);
> +
> +	if (ps == 0x1000)
> +		test_size = 10240;
> +	else
> +		test_size = 512;
>  
> -	start = mmap(0, 4096 * TEST_SIZE * 4, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
> +	start = mmap(0, ps * test_size * 4, PROT_NONE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
>  	if (start == MAP_FAILED)
>  		return 1;
>  
> -	for (i = 0; i < TEST_SIZE; i++) {
> +	for (i = 0; i < test_size; i++) {
>  		int *addr;
> -		addr = mmap(start + i * 3 * 4096, 4096,
> +		addr = mmap(start + i * 3 * ps, ps,
>  				PROT_READ | PROT_WRITE,
>  				MAP_PRIVATE | MAP_FILE | MAP_FIXED, fd, 0);
>  		if (addr == MAP_FAILED)
>  			return 1;
>  		addr[0] = i * 2;
> -		addr = mmap(start + (i * 3 + 1) * 4096, 4096,
> +		addr = mmap(start + (i * 3 + 1) * ps, ps,
>  				PROT_READ | PROT_WRITE,
>  				MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
>  		if (addr == MAP_FAILED)
> @@ -49,12 +54,12 @@ int main(int argc, char ** argv)
>  
>  	test_waitsig();
>  
> -	for (i = 0; i < TEST_SIZE; i++) {
> +	for (i = 0; i < test_size; i++) {
>  		int *addr;
> -		addr = start + i * 3 * 4096;
> +		addr = start + i * 3 * ps;
>  		if (addr[0] != i * 2)
>  			fail();
> -		addr = start + (i * 3  + 1) * 4096;
> +		addr = start + (i * 3  + 1) * ps;
>  		if (addr[0] != i)
>  			fail();
>  	}
> -- 
> 1.8.3.1
>