Re: [PATCH 16/16] selftests/mm: Create MAP_BELOW_HINT test

On Wed, Aug 28, 2024 at 06:48:33PM +0100, Lorenzo Stoakes wrote:
> On Tue, Aug 27, 2024 at 10:49:22PM GMT, Charlie Jenkins wrote:
> > Add a selftest for MAP_BELOW_HINT that maps until it runs out of space
> > below the hint address.
> >
> > Signed-off-by: Charlie Jenkins <[email protected]>
> > ---
> >  tools/testing/selftests/mm/Makefile         |  1 +
> >  tools/testing/selftests/mm/map_below_hint.c | 29 +++++++++++++++++++++++++++++
> >  2 files changed, 30 insertions(+)
> >
> > diff --git a/tools/testing/selftests/mm/Makefile b/tools/testing/selftests/mm/Makefile
> > index cfad627e8d94..4e2de85267b5 100644
> > --- a/tools/testing/selftests/mm/Makefile
> > +++ b/tools/testing/selftests/mm/Makefile
> > @@ -50,6 +50,7 @@ TEST_GEN_FILES += hugepage-shm
> >  TEST_GEN_FILES += hugepage-vmemmap
> >  TEST_GEN_FILES += khugepaged
> >  TEST_GEN_FILES += madv_populate
> > +TEST_GEN_FILES += map_below_hint
> >  TEST_GEN_FILES += map_fixed_noreplace
> >  TEST_GEN_FILES += map_hugetlb
> >  TEST_GEN_FILES += map_populate
> > diff --git a/tools/testing/selftests/mm/map_below_hint.c b/tools/testing/selftests/mm/map_below_hint.c
> > new file mode 100644
> > index 000000000000..305274c5af49
> > --- /dev/null
> > +++ b/tools/testing/selftests/mm/map_below_hint.c
> > @@ -0,0 +1,29 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * Test the MAP_BELOW_HINT mmap flag.
> > + */
> > +#include <sys/mman.h>
> > +#include "../kselftest.h"
> > +
> > +#define ADDR 0x1000000UL
> > +#define LENGTH (ADDR / 100)
> > +
> > +#define MAP_BELOW_HINT	  0x8000000	/* Not defined in all libc */
> > +
> > +/*
> > + * Map memory with MAP_BELOW_HINT until no memory left. Ensure that all returned
> > + * addresses are below the hint.
> > + */
> > +int main(int argc, char **argv)
> > +{
> > +	void *addr;
> > +
> > +	do {
> > +		addr = mmap((void *)ADDR, LENGTH, MAP_ANONYMOUS, MAP_BELOW_HINT, -1, 0);
> 
> How can this be correct? mmap() has parameters:
> 
>        void *mmap(void addr[.length], size_t length, int prot, int flags,
>                   int fd, off_t offset);
> 
> You'r setting prot = MAP_ANONYMOUS, flags = MAP_BELOW_HINT...
> 
> This surely should be:
> 
> mmap(..., PROT_READ | PROT_WRITE, MAP_ANON | MAP_PRIVATE | MAP_BELOW_HINT, -1, 0);
> 
> > +	} while (addr == MAP_FAILED && (unsigned long)addr <= ADDR);
> 
> How can addr == MAP_FAILED (i.e. ~0) and addr <= ADDR? This will just loop
> through once...
> 
> If you want to make sure you're getting mappings only below the hint until
> you start getting MAP_FAILED's you'll need to handle this more robustly.
> 
> > +
> > +	if (addr != MAP_FAILED && (unsigned long)addr > ADDR)
> > +		ksft_exit_fail_msg("mmap returned address above hint with MAP_BELOW_HINT\n");
> 
> This is just going to fail because your flags are wrong then wrongly claim
> to have passed...

I obviously didn't spend enough time thinking about this test case...
You are correct that I wrote this incorrectly. I will make a proper test case
and send out a new version.

- Charlie

> 
> > +
> > +	ksft_test_result_pass("MAP_BELOW_HINT works\n");
> > +}
> >
> > --
> > 2.45.0
> >