[LTP] [PATCH 3/3] [PATCH] [PATCH] Migrating the libhugetlbfs/testcases/straddle_4GB.c v3

[Petr Vorel]

Hi Pavithra, Geetika,

I very briefly looked into the original source file (I'd personally put that
link into the commit message).

> This testcase attempts to map a memory range that straddles the 4GB boundary using mmap() with and without the MAP_FIXED flag.

> Changes in v3:

FYI I tried to search in the mailing history [2] [3], but I found only v1, sent
by Geetika. Is there any other discussion than I found?

[1] https://github.com/libhugetlbfs/libhugetlbfs/blob/master/tests/straddle_4GB.c
[2] https://lore.kernel.org/ltp/?q=straddle_4GB.c
[3] https://lore.kernel.org/ltp/?q=hugemmap40

> - Added magic definition to include/tst_fs.h as separate patch.
> - Moved CFLAGS to make file.
> - Added read_maps definition to separate patch.
> - Removed \n from tst_res

> diff --git a/testcases/kernel/mem/hugetlb/hugemmap/hugemmap40.c b/testcases/kernel/mem/hugetlb/hugemmap/hugemmap40.c
> new file mode 100644
> index 000000000..7b4ad7b05
> --- /dev/null
> +++ b/testcases/kernel/mem/hugetlb/hugemmap/hugemmap40.c
> @@ -0,0 +1,147 @@
> +// SPDX-License-Identifier: LGPL-2.1-or-later
> +/*
> + * Copyright (C) 2005-2006 David Gibson & Adam Litke, IBM Corporation.
> + * Copyright (C) 2006 Hugh Dickins <hugh@veritas.com>
> + */
> +
> +/*\
> + * [Description]
nit: we don't use "[Description]" any more, because /*\ is enough for parsing
comments.
Please remove the line.

> + *
> + * Test tries to allocate hugepages to cover a memory range that straddles the
> + * 4GB boundary, using mmap(2) with and without MAP_FIXED.
nit: Could you please use :man2:`mmap` (this converts to a man page link [4] in
our docs, see e.g. stack_clash [5]).

[4] https://man7.org/linux/man-pages/man2/mmap.2.html
[5] https://linux-test-project.readthedocs.io/en/latest/users/test_catalog.html#stack-clash

> + */
> +
> +#define MAPS_BUF_SZ 4096
> +#define FOURGB (1UL << 32)
FYI we also have TST_GB in include/tst_fs.h.

> +#define MNTPOINT "hugetlbfs/"
> +
> +#include "hugetlb.h"
> +
> +static unsigned long hpage_size;
> +static int fd = -1;
> +static unsigned long straddle_addr;
> +
> +static int test_addr_huge(void *p)
> +{
> +	char name[256];
> +	char *dirend;
> +	int ret;
> +	struct statfs64 sb;
> +
> +	ret = read_maps((unsigned long)p, name);
> +	if (ret < 0)
> +		return ret;
> +
> +	if (ret == 0) {
> +		tst_res(TINFO, "Couldn't find address %p in /proc/self/maps", p);
Having this as info message which is later followed by tst_brk(TFAIL) is is not
optimal. I suspect test is unnecessary complicated, but I'm not sure now how to
improve it.

> +		return -1;
> +	}
> +
> +	/* looks like a filename? */
> +	if (name[0] != '/')
> +		return 0;
> +
> +	/* Truncate the filename portion */
> +
nit: empty line above, please remove it.
> +	dirend = strrchr(name, '/');
> +	if (dirend && dirend > name)
> +		*dirend = '\0';
> +
> +	ret = statfs64(name, &sb);
> +	if (ret)
> +		return -1;
I guess we really need statfs64() not just statfs(), right?
In case of the failure test should quit with tst_brk(TBROK | TERRNO, "...").

> +
> +	return (sb.f_type == HUGETLBFS_MAGIC);
> +}
> +
> +static void run_test(void)
> +{
> +	void *p;
> +
> +	/* We first try to get the mapping without MAP_FIXED */
> +	tst_res(TINFO, "Mapping without MAP_FIXED at %lx...", straddle_addr);
> +	p = mmap((void *)straddle_addr, 2*hpage_size, PROT_READ|PROT_WRITE,
> +			MAP_SHARED, fd, 0);
> +	if (p == (void *)straddle_addr) {
> +		/* These tests irrelevant if we didn't get the straddle address*/
> +		if (test_addr_huge(p) != 1) {
> +			tst_brk(TFAIL, "1st Mapped address is not hugepage");
> +			goto windup;
FYI tst_brk() quits testing (test executes only cleanup function. You need to
use tst_res(TFAIL, ...).

> +		}
> +		if (test_addr_huge(p + hpage_size) != 1) {
> +			tst_brk(TFAIL, "2nd Mapped address is not hugepage");
> +			goto windup;
> +		}
> +		memset(p, 0, hpage_size);
> +		memset(p + hpage_size, 0, hpage_size);
> +		tst_res(TPASS, "Mapping without MAP_FIXED at %lx... completed", straddle_addr);
> +	} else {
> +		tst_res(TINFO, "Got %p instead, never mind, let's move to mapping with MAP_FIXED", p);
> +		munmap(p, 2*hpage_size);
We have SAFE_MUNMAP().
> +	}
> +
> +	tst_res(TINFO, "Mapping with MAP_FIXED at %lx...", straddle_addr);
> +	p = mmap((void *)straddle_addr, 2*hpage_size, PROT_READ|PROT_WRITE,
> +				MAP_SHARED|MAP_FIXED, fd, 0);
> +
> +	if (p == MAP_FAILED) {
> +		/* this area crosses last low slice and first high slice */
> +		unsigned long below_start = FOURGB - 256L*1024*1024;
> +		unsigned long above_end = 1024L*1024*1024*1024;
> +
> +		if (tst_mapping_in_range(below_start, above_end) == 1) {
> +			tst_res(TINFO, "region (4G-256M)-1T is not free");
> +			tst_res(TINFO, "mmap() failed: %s", strerror(errno));
> +			tst_res(TWARN, "Pass Inconclusive!");
I don't understand "Pass Inconclusive!". Could it be just described what happen?

We have TERRNO which we use instead strerror(). Instead of all 3 tst_res()
mesagess above I would use something like:
			tst_res(TFAIL | TERRNO, "region (4G-256M)-1T is not free");

> +			goto windup;
> +		} else
> +			tst_res(TFAIL, "mmap() FIXED failed: %s", strerror(errno));
I also don't understand "FIXED". I guess it was supposed to be "mmap() with MAP_FIXED failed".
Does it even make sense to continue on mmap() failure? The original test does,
but I'm not sure if that's ok.

very nit: if you use { } on if, they should be also on else.
} else {
> 		tst_res(TFAIL | TERRNO, "mmap() failed: %s", strerror(errno));
}


> +	}
> +
The code below uses wrong indent
> +		if (p != (void *)straddle_addr) {
> +			tst_res(TINFO, "got %p instead", p);
> +			tst_res(TFAIL, "Wrong address with MAP_FIXED");
> +			goto windup;
> +		}
> +
> +		if (test_addr_huge(p) != 1) {
> +			tst_brk(TFAIL, "1st Mapped address is not hugepage");
> +			goto windup;
> +		}
> +
> +		if (test_addr_huge(p + hpage_size) != 1) {
> +			tst_brk(TFAIL, "2nd Mapped address is not hugepage");
> +			goto windup;
> +		}
> +		memset(p, 0, hpage_size);
> +		memset(p + hpage_size, 0, hpage_size);
Why is this memset() for? If needed for testing, shouldn't it be at the
beginning?

> +		tst_res(TPASS, "Mapping with MAP_FIXED at %lx... completed", straddle_addr);
> +
> +windup:
> +	SAFE_MUNMAP(p, 2*hpage_size);
> +}
> +
> +static void setup(void)
> +{
> +	straddle_addr = FOURGB - hpage_size;
> +	hpage_size = tst_get_hugepage_size();
> +	fd = tst_creat_unlinked(MNTPOINT, 0, 0600);

> +	if (hpage_size > FOURGB)
> +		tst_brk(TCONF, "Huge page size is too large!");
I suppose the original check is really needed.

The original test does this check for this:

static inline long check_hugepagesize()
{
	long __hpage_size = gethugepagesize();
	if (__hpage_size < 0) {
		if (errno == ENOSYS)
			CONFIG("No hugepage kernel support\n");
		else if (errno == EOVERFLOW)
			CONFIG("Hugepage size too large");
		else
			CONFIG("Hugepage size (%s)", strerror(errno));
	}
	return __hpage_size;
}

Kind regards,
Petr

[6] https://github.com/libhugetlbfs/libhugetlbfs/blob/master/tests/hugetests.h#L126-L138

> +}
> +
> +static void cleanup(void)
> +{
> +	if (fd >= 0)
> +		SAFE_CLOSE(fd);
> +}
> +
> +static struct tst_test test = {
> +	.needs_root = 1,
> +	.mntpoint = MNTPOINT,
> +	.needs_hugetlbfs = 1,
> +	.hugepages = {2, TST_NEEDS},
> +	.setup = setup,
> +	.cleanup = cleanup,
> +	.test_all = run_test,
> +};