[LTP] [PATCH v4 1/4] syscalls/madvise04: new test for madvise(MADV_DONTNEED)

zhaogongyi zhaogongyi@huawei.com
Mon Oct 24 11:29:41 CEST 2022


Hi,

> 
> Hello,
> 
> Zhao Gongyi via ltp <ltp@lists.linux.it> writes:
> 
> > Check that madvise(2) MADV_DONTNEED operation applied to Huge TLB
> > pages successfully after kernel version 5.18, and will result in
> > zero-fill-on-demand pages for anonymous private mappings.
> >
> > Signed-off-by: Zhao Gongyi <zhaogongyi@huawei.com>
> > ---
> >  runtest/syscalls                              |  1 +
> >  testcases/kernel/syscalls/madvise/.gitignore  |  1 +
> > testcases/kernel/syscalls/madvise/madvise04.c | 62
> +++++++++++++++++++
> >  3 files changed, 64 insertions(+)
> >  create mode 100644 testcases/kernel/syscalls/madvise/madvise04.c
> >
> > diff --git a/runtest/syscalls b/runtest/syscalls index
> > c81764df4..eb1910cec 100644
> > --- a/runtest/syscalls
> > +++ b/runtest/syscalls
> > @@ -941,6 +941,7 @@ mincore04 mincore04
> >  madvise01 madvise01
> >  madvise02 madvise02
> >  madvise03 madvise03
> > +madvise04 madvise04
> >  madvise05 madvise05
> >  madvise06 madvise06
> >  madvise07 madvise07
> > diff --git a/testcases/kernel/syscalls/madvise/.gitignore
> > b/testcases/kernel/syscalls/madvise/.gitignore
> > index f4bfdfefe..db8ce47c1 100644
> > --- a/testcases/kernel/syscalls/madvise/.gitignore
> > +++ b/testcases/kernel/syscalls/madvise/.gitignore
> > @@ -1,6 +1,7 @@
> >  /madvise01
> >  /madvise02
> >  /madvise03
> > +/madvise04
> >  /madvise05
> >  /madvise06
> >  /madvise07
> > diff --git a/testcases/kernel/syscalls/madvise/madvise04.c
> > b/testcases/kernel/syscalls/madvise/madvise04.c
> > new file mode 100644
> > index 000000000..a970fb33e
> > --- /dev/null
> > +++ b/testcases/kernel/syscalls/madvise/madvise04.c
> > @@ -0,0 +1,62 @@
> > +// SPDX-License-Identifier: GPL-2.0-or-later
> > +/*
> > + * Copyright (c) Huawei Technologies Co., Ltd. 2022. All rights reserved.
> > + * Author: Zhao Gongyi <zhaogongyi@huawei.com>  */
> > +
> > +/*\
> > + * [Description]
> > + *
> > + * Check that madvise(2) MADV_DONTNEED operation applied to Huge
> TLB
> > +pages
> > + * successfully after kernel version 5.18, and will result in
> > + * zero-fill-on-demand pages for anonymous private mappings.
> > + */
> > +
> > +#include "tst_test.h"
> > +
> > +#define MAP_SIZE (8 * 1024)
> > +
> > +static char *addr;
> > +static int mapsz;
> > +
> > +static void run(void)
> > +{
> > +	TST_EXP_PASS(madvise(addr, mapsz, MADV_DONTNEED));
> > +	for (int i = 0; i < mapsz; i++) {
> > +		if (addr[i]) {
> > +			tst_res(TFAIL,
> > +				"There are no zero-fill-on-demand pages "
> > +				"for anonymous private mappings");
> > +			return;
> > +		}
> > +	}
> > +
> > +	tst_res(TPASS, "There are zero-fill-on-demand pages "
> > +		       "for anonymous private mappings"); }
> > +
> > +static void setup(void)
> > +{
> > +	mapsz = tst_get_hugepage_size();
> > +	addr = SAFE_MMAP(NULL, mapsz,
> > +			PROT_READ | PROT_WRITE,
> > +			MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB,
> > +			-1, 0);
> > +	memset(addr, 1, mapsz);
> 
> If we only do memset here then we are only testing the zero-fill feature on
> the first iteration.

Yes, thanks. It need to move to run().

> 
> > +}
> > +
> > +static void cleanup(void)
> > +{
> > +	if (addr)
> > +		SAFE_MUNMAP(addr, mapsz);
> > +}
> > +
> > +static struct tst_test test = {
> > +	.test_all = run,
> > +	.setup = setup,
> > +	.cleanup = cleanup,
> > +	.min_kver = "5.18",
> 
> What happens before 5.18? Could we try applying MADV_DONTNEED and
> return TCONF instead?

MADV_DONTNEED is not supported before 5.18. 
Try applying MADV_DONTNEED and return TCONF seems more better, thanks.

> 
> > +	.needs_root = 1,
> 
> Why does this need root?

When we get or set hugepages, we need to read or write '/proc/sys/vm/nr_hugepages', so we need to set needs_root here.

Regards,
Gongyi


More information about the ltp mailing list