[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20191218121837.GD21485@dhcp22.suse.cz>
Date: Wed, 18 Dec 2019 13:18:37 +0100
From: Michal Hocko <mhocko@...nel.org>
To: Mike Kravetz <mike.kravetz@...cle.com>
Cc: Waiman Long <longman@...hat.com>,
Eric B Munson <emunson@...mai.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Matthew Wilcox <willy@...radead.org>,
linux-kernel@...r.kernel.org, linux-mm@...ck.org,
aneesh.kumar@...ux.ibm.com, Jarod Wilson <jarod@...hat.com>
Subject: hugetlbfs testing coverage (was: Re: [PATCH v2] mm/hugetlb: defer
free_huge_page() to a workqueue)
On Tue 17-12-19 10:05:06, Mike Kravetz wrote:
> Cc: Eric
>
> On 12/17/19 1:00 AM, Michal Hocko wrote:
> > On Mon 16-12-19 13:44:53, Waiman Long wrote:
> >> On 12/16/19 10:38 AM, Waiman Long wrote:
> > [...]
> >>>> Can you extract guts of the testcase and integrate them into hugetlb
> >>>> test suite?
> >>
> >> BTW, what hugetlb test suite are you talking about?
> >
> > I was using tests from libhugetlbfs package in the past. There are few
> > tests in LTP project but the libhugetlbfs coverage used to cover the
> > largest part of the functionality.
> >
> > Is there any newer home for the package than [1], Mike? Btw. would it
> > mak sense to migrate those tests to a more common place, LTP or kernel
> > selftests?
>
> That is the latest home/release for libhugetlbfs.
>
> The libhugetlbfs test suite is somewhat strange in that I suspect it started
> as testing for libhugetlbfs itself. When it was written, the thought may have
> been that people would use libhugetlfs as the primary interface to hugetlb
> pages. That is not the case today. Over time, hugetlbfs tests not associated
> with libhugetlbfs were added.
>
> If we want to migrate libhugetlbfs tests, then I think we would only want to
> migrate the non-libhugetlbfs test cases. Although, the libhugetlbfs specific
> tests are useful as they 'could' point out regressions.
Yeah, I can second that. I remember using the suite and it pointed to
real issues when I was touching the area in the past. So if we can get
as many tests to be independent on the library and integrate it to some
existing testing project - be it kernel selftest or LTP - then it would
be really great and I assume the testing coverage of the hugetlb
functionality would increase dramatically.
--
Michal Hocko
SUSE Labs
Powered by blists - more mailing lists