[LTP] [PATCH] Fixed hugeshmat05 Test Failure with 1GB Hugepages.

[Cyril Hrubis]

Hi!
> Modified the test to detect large hugepage sizes (≥1GB) and adjust test
> sizes accordingly.
> 
> Signed-off-by: Pavithra <pavrampu@linux.ibm.com>
> ---
>  .../mem/hugetlb/hugeshmat/hugeshmat05.c       | 45 +++++++++++++++----
>  1 file changed, 36 insertions(+), 9 deletions(-)
> 
> diff --git a/testcases/kernel/mem/hugetlb/hugeshmat/hugeshmat05.c b/testcases/kernel/mem/hugetlb/hugeshmat/hugeshmat05.c
> index 3b2ae351c..870a61ec1 100644
> --- a/testcases/kernel/mem/hugetlb/hugeshmat/hugeshmat05.c
> +++ b/testcases/kernel/mem/hugetlb/hugeshmat/hugeshmat05.c
> @@ -37,9 +37,13 @@ void setup(void)
>  {
>  	page_size = getpagesize();
>  	hpage_size = SAFE_READ_MEMINFO("Hugepagesize:") * 1024;
> +
> +	tst_res(TINFO, "Page size: %ld bytes", page_size);
> +	tst_res(TINFO, "Hugepage size: %ld bytes (%ld MB)",
> +		hpage_size, hpage_size / (1024 * 1024));
>  }
>  
> -void shm_test(int size)
> +void shm_test(long size)
>  {
>  	int shmid;
>  	char *shmaddr;
> @@ -56,7 +60,7 @@ void shm_test(int size)
>  	}
>  
>  	shmaddr[0] = 1;
> -	tst_res(TINFO, "allocated %d huge bytes", size);
> +	tst_res(TINFO, "allocated %ld huge bytes", size);
>  
>  	if (shmdt((const void *)shmaddr) != 0) {
>  		shmctl(shmid, IPC_RMID, NULL);
> @@ -69,16 +73,39 @@ void shm_test(int size)
>  static void test_hugeshmat(void)
>  {
>  	unsigned int i;
> +	long tst_sizes[4];
> +
> +	/*
> +	 * For large hugepage sizes (e.g., 1GB), we need to ensure
> +	 * test sizes are within reasonable bounds and properly aligned.
> +	 * The original test used N*hpage_size which could be 4GB for 1GB pages.
> +	 *
> +	 * We adjust the test to use N/2 multiplier for large hugepages
> +	 * to avoid excessive memory requirements while still testing the
> +	 * alignment boundary conditions with multiple pages.
> +	 */
> +	if (hpage_size >= 1024 * 1024 * 1024) {
> +		/* For 1GB or larger hugepages, use N/2 pages (2 pages for N=4) */
> +		long multiplier = N / 2;

Two obvious comments for this if? Really?

> -	const int tst_sizes[] = {
> -		N * hpage_size - page_size,
> -		N * hpage_size - page_size - 1,
> -		hpage_size,
> -		hpage_size + 1
> -	};
> +		tst_sizes[0] = multiplier * hpage_size - page_size;
> +		tst_sizes[1] = multiplier * hpage_size - page_size - 1;
> +		tst_sizes[2] = hpage_size;
> +		tst_sizes[3] = hpage_size + 1;
> +		tst_res(TINFO, "Using N/2=%ld hugepage test sizes for large hugepages", multiplier);
> +	} else {
> +		/* For smaller hugepages (2MB, 16MB, etc.), use original test */
> +		tst_sizes[0] = N * hpage_size - page_size;
> +		tst_sizes[1] = N * hpage_size - page_size - 1;
> +		tst_sizes[2] = hpage_size;
> +		tst_sizes[3] = hpage_size + 1;
> +		tst_res(TINFO, "Using N=%d hugepage test sizes", N);
> +	}

This is ugly. Why can't we set just the multiplier and initialize the
array only once?

> -	for (i = 0; i < ARRAY_SIZE(tst_sizes); ++i)
> +	for (i = 0; i < ARRAY_SIZE(tst_sizes); ++i) {
> +		tst_res(TINFO, "Testing size: %ld bytes", tst_sizes[i]);

This patch adds way too much debugging output. The shm_test already
prints the size.

>  		shm_test(tst_sizes[i]);
> +	}
>  
>  	tst_res(TPASS, "No regression found.");
>  }
> -- 
> 2.53.0
> 
> 
> -- 
> Mailing list info: https://lists.linux.it/listinfo/ltp

-- 
Cyril Hrubis
chrubis@suse.cz