On Wed, Sep 23, 2020 at 12:01 PM Carlos O'Donell <carlos@redhat.com> wrote:
>
> On 6/12/20 4:10 PM, H.J. Lu via Libc-alpha wrote:
> > Add a strncmp testcase to cover cases where one of strings ends on the
> > page boundary.
>
> I would like to see this sequence of 4 patches committed because they *do*
> correctly regression test swbz#25933, and I'd like to see this not regress
> again.
>
> However, I share Paul's concerns over the magic numbers, so I have made
> some concrete suggestions for comments.
>
> OK with the following changes:
> - Add all suggested comments.
> - s1a iterates over [30,32).
>
> Reviewed-by: Carlos O'Donell <carlos@redhat.com>
>
> > ---
> >  string/test-strncmp.c | 25 +++++++++++++++++++++++++
> >  1 file changed, 25 insertions(+)
> >
> > diff --git a/string/test-strncmp.c b/string/test-strncmp.c
> > index d961ac4493..d0928a2864 100644
> > --- a/string/test-strncmp.c
> > +++ b/string/test-strncmp.c
> > @@ -403,6 +403,30 @@ check2 (void)
> >    free (s2);
> >  }
> >
> > +static void
> > +check3 (void)
> > +{
> > +  size_t size = 32 * 4;
>
> Add a comment:
>
> /* To trigger bug 25933 we need a size that is equal to the
>    vector length times 4. In the case of AVX2 for Intel we
>    need 32 * 4. We make this test generic and run it for all
>    architectures as additional boundary testing for such
>    related algorithms.  */
>
> This is my understanding, that we need 32*4 to trigger the bug.
>
> > +  CHAR *s1 = (CHAR *) (buf1 + (BUF1PAGES - 1) * page_size);
> > +  CHAR *s2 = (CHAR *) (buf2 + (BUF1PAGES - 1) * page_size);
>
> We set s1 and s2 to point into the buffers at a point 1 page
> before the end. We expect the test to add 1 page PROT_NONE at
> the end. Thus s1 and s2 by default point to two pages, the
> first page PROT_READ|PROT_WRITE and the second page PROT_NONE.
> So far so good. No comment required. You have to understand
> how these test cases work to read this.
>
> > +  int exp_result;
> > +
> > +  memset (s1, 'a', page_size);
> > +  memset (s2, 'a', page_size);
>
> OK. Fill both full of 'a'.
>
> > +  s1[(page_size / CHARBYTES) - 1] = (CHAR) 0;
>
> OK. Null terminate s1.
>
> Note that s2 is not null terminated.
>
> > +
>
> Add comment:
>
> /* Iterate over a size that is just below where we expect
>    the bug to trigger up to the size we expect will trigger
>    the bug e.g. [99-128].  Likewise iterate the start of
>    two strings between 30 and 31 bytes away from the
>    boundary to simulate alignment changes.  */
>
> > +  for (size_t s = 99; s <= size; s++)
>
> OK. A bit of belt-and-suspenders here we make s range
> from 99-128, so we iterate just before the size we care
> about up to the size we expect to trigger the bug.
>
> > +    for (size_t s1a = 31; s1a < 32; s1a++)
>
> s1a iterates over [31,32) e.g. 31
>
> > +      for (size_t s2a = 30; s2a < 32; s2a++)
>
> s2a iterates over [30,32) e.g. 30, 31
>
> > +     {
> > +       CHAR *s1p = s1 + (page_size / CHARBYTES - s) - s1a;
>
> Set the pointer back from the PROT_NONE page by "s+s1a" bytes.
>
> > +       CHAR *s2p = s2 + (page_size / CHARBYTES - s) - s2a;
>
> Set the pointer back from the PROT_NONE page by "s+s2a" bytes.
>
> > +       exp_result = SIMPLE_STRNCMP (s1p, s2p, s);
>
> Then compare.
>
> This code comes from Adhemerval's testing in comment #2,
> where the test catches a second loop that has similar problems.
>
> At most we test [30 sizes]x[1 offset for s1a]x[2 offets for s2a] = 60 tests.
>
> Suggest:
>
> s1a iterate over [30,32) like s2a for the sake of simplicity.
>
> > +       FOR_EACH_IMPL (impl, 0)
> > +         check_result (impl, s1p, s2p, s, exp_result);
> > +     }
> > +}
> > +
> >  int
> >  test_main (void)
> >  {
> > @@ -412,6 +436,7 @@ test_main (void)
> >
> >    check1 ();
> >    check2 ();
> > +  check3 ();
>
> OK.
>
> >
> >    printf ("%23s", "");
> >    FOR_EACH_IMPL (impl, 0)
> >
>

Here is the updated patch I am checking in.

Thanks.

-- 
H.J.