Re: [LTP] LTP test reults for kernel 3.0.15 and 3.4

[Ritesh H. <rit...@gm...>]

Hi Chris

Sorry for such a long gap. I tried finding out that why that test case was
failing by running a normal C program with malloc implementation.

I found that, when I am trying to allocate a smaller memory chunk, memory
is getting allocated from the same address, but for a larger memory chunk,
it is allocating from some different address i.e if condition is getting
true (that means that test is failing).
Couldn't figure out, why this difference. Any idea on to where should I
look into?

And yes, I will surely try to fix if this is a broken test. But moreover,
in some other kernel version it was passing. So, need to figure out as to
what changes has been made for this.

Thank you
Ritesh


On Thu, Sep 20, 2012 at 7:59 PM, <ch...@su...> wrote:

> Hi!
> > Following are the lines of code where the test case is exiting, for
> kernel
> > 3.0.15 (ICS).
> >
> > 128         pm2=pm1;
> > 129         free(pm1);
> > 130
> > 131         if((pm1=(char *)calloc(memsize,1)) == NULL) {
> > 132                 tst_resm(TFAIL, "calloc did not alloc memory ");
> > 133                              tst_exit();
> > 134         }
> > 135
> > 136         if(pm1 != pm2) {
> > 137                 tst_resm(TINFO, "pm1=%p pm2=%p ", pm1,pm2);
> > 138                 tst_resm(TFAIL, "free did not dealloc memory ");
> > 139
> > tst_exit();                                                     // *Here
> > code is exited in kernel 3.0.15 while it is passing in case of kernel 3.4
> > (ICS)*.
> > 140         }
> >
> >
> >
> > *LTP version = "LTP-20100630"*
> >
> > at line 136 my test is failing. Now isnt this condition of the dangling
> > pointer. At line 129 we are freeing pm1 this means that pm2 becomes a
> > dangling pointer.
> > And how come after doing
> > pm1=(char *)calloc(memsize,1)
> > pm1 is equal to pm2 for kernel 3.4 (ICS) (Since this test case is passing
> > for that)??
> >
> > According to me, it should not be equal.
>
> Indeed this test seems to be wrong. My guess is that it was written with
> 'it works so and so it must be true' attitude as it's not unlikely to
> get same pointer if you request chunk of a same size directly after you
> free one.
>
> So my conclusion is that the test is broken and should be fixed. Do you
> care to send a patch (against latest git tree)?
>
> --
> Cyril Hrubis
> ch...@su...
>