Re: misc oprofile-tests updates

Will Schmidt <[email protected]>
Newsgroups gmane.linux.oprofile
Message-ID <[email protected]>
On Fri, 2016-04-01 at 17:21 +0200, Michael Petlan wrote:
> On Wed, 30 Mar 2016, Will Schmidt wrote:
> > Hi Oprofile devs,
> >
> > A handful of changes to the oprofile-tests suite, for consideration.
> >
> > I stumbled across an issue while running the tests, dug in a bit, and came
> > out with this short series of patches to improve things.  Most of the
> > changes are cosmetic, but a couple will affect the suite behavior, most
> > notably that more of the tests will be runnable as a normal user, and we
> > will successfully run more tests as root when the kptr_restrict mode is
> > in restricted mode, where we previously would have punted.  (Additional
> > detail in the patch descriptions.)
> >
> > I've tested these changes against an older (0.9.8) install, as well as
> > against latest (1.2.0git), no new issues in my testing.
> >
> > As a user:
> > before:  49/49 tests passed.
> > after:   74/78 to 78/78 tests passed.  (depending on the value of kptr_restrict)
> >
> > As root:
> > No notable changes.  My latest run showed
> > after:  113/117 to 117/117 tests passed.  (depending on kptr_restrict)
> >
> > Thanks,
> > -Will
> >
> 
> Hi,
> 
> I have run the testsuite with your patches on a few machines and I like
> the changes. Thank you for them.
> 
> I am interested in the failures related to the value of kptr_restrict.
> Could you please provide more information about your testing?
>
> Which kptr_restrict values it passes with? Which values it fails with?
> Which kernel versions, archs, systems have you used?

I'm mostly power focused, so ppc64 and ppc64le arches on power7 and
power8 hardware.    I do some sniff tests on x86.  The distro flavor
varies, mostly fedora based, but some RHEL/SuSE/Ubuntu installs
sprinkled in from time to time. 

For the kernel versions, at a glance the oldest I've run against is
2.6.32.*, and newest is 4.2.0.*.  All distro provided kernels.

> I have met an issue with incorrect kallsyms detection (see the patch
> I have sent recently). I noticed that issue by running the patched suite
> where it was reporting "set kptr_restrict to 0" despite having it 0.

Is that the blurb that now looks like so: ? 
        fail  "    Failed to obtain the kernel symbols from kallsyms."
        print "Verify the test is being run as root or that "
        print "/proc/sys/kernel/kptr_restrict is set to 0."
        send "The current value of kptr_restrict is: "

The alternative path into that error message would be if something else
went wrong such that the expected symbol was not found in the
opreport.  


> Also please try to run
> 
> grep -e ' _stext' -e ' _text' -e ' _etext' /proc/kallsyms

Each of my boxes returns roughly the following.  
power:
c000000000000000 T _stext
c000000000000000 T _text
c000000000950000 R _etext
x86:
ffffffff81000000 T _text
ffffffff810001c8 T _stext
ffffffff8177a6f2 T _etext

when kptr_restrict is 2, the values are always zero'd out, regardless of
if I run that grep as root or user.
when kptr_restrict is 1, the values are zero for user, valid for root.
and when kptr_restrict is 0, values are always valid.

> on the box(es) where you tested the suite and send the output. In case
> your kernel's kallsyms file has the _stext symbol, could you also try
> to run the tests against oprofile with my patch if anything changes.

I don't see any obvious changes in output with your patch applied.  The
_stext and _text values on the power systems appear to always be the
same, at the bottom of the kernel address space, so would not make a
difference there.  On the x86* boxes I've got, that value varies just a
little bit,.. I'm not sure how significant that delta is in the grand
scheme of things, I don't expect a whole lot of symbols are in the
*[000]-*[1c8] range that I'd be missing. :-)


What arch/kernel versions are you seeing the problem with?  I'll look
closer/deeper if I can.  

Thanks,
-Will


> 
> Thank you!
> 
> Michael
> 
> >
> > ------------------------------------------------------------------------------
> > Transform Data into Opportunity.
> > Accelerate data analysis in your applications with
> > Intel Data Analytics Acceleration Library.
> > Click to learn more.
> > http://pubads.g.doubleclick.net/gampad/clk?id=278785471&iu=/4140
> > _______________________________________________
> > oprofile-list mailing list
> > [email protected]
> > https://lists.sourceforge.net/lists/listinfo/oprofile-list
> >
> 



------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785471&iu=/4140
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.