Re: [PATCH] selftests/vm: fix link error for transhuge-stress test

[Andrey Skvortsov <andrej.skvortzov@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1725 bytes --]

On Thu, Jan 08, 2015 at 08:55:25AM -0700, Shuah Khan wrote:

> On 01/07/2015 01:10 PM, Andrey Skvortsov wrote:
> > man page for clock_gettime says 'Link with -lrt'. So I think the
> > error message is correct.
> > 
> 
> Thanks for fixing it. Applied to linux-kselftest fixes branch
> 


Hi Shuah,

thanks for taking the patch.

sorry for the late reply. I've just checked e-mail. Different
timezone. =)

I have couple questions about selftests/vm/hugetlbfstest. It is called by
make run_tests, but fails always on my machine.
What does it actually test? if I understand the code right,
it tests whether RSS (read from /proc/self/statm) grows/shrinks not much during
allocation/deallocation of 16Mb. 

Here are couple of questions about some tests in the code.
1)
hugetlbfstest.c:75
	do_mmap(-1, MAP_ANONYMOUS, 1);

it does not use hugetlbfs. So it does not correspond to the
actual test name. That confused me a little. Does it test usage of
transparent huge pages? 

2) 
hugetlbfstest.c:78
	do_mmap(hugefd, 0, 1);

Despite the fact, that hugefd is open at wrong location
(run_tests script mounts hugetlbfs at ./huge) the test actually does
use hugetlbfs. Unfortunately on my machine it fails always too. 
RSS does not grow after allocation, but is the same as before.
So assertion 'llabs(after - before - length) < 0x40000' fails.
But hugepages are definitely allocated, I can see that HugePages_Free decreases in
/proc/meminfo. To find this out I placed delays in the test code.
Does rss field in /proc/statm counts only normal non-huge pages? 

        
-- 
Best regards,
Andrey Skvortsov

Secure e-mail with gnupg: See http://www.gnupg.org/
PGP Key ID: 0x57A3AEAD



[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]