On 10/5/26 2:24 AM, Jaeyeon Lee wrote:
> Every test is expected to reserve the HugeTLB pages it needs.
> hugetlb-read-hwpoison was not converted, so on a system without
> pre-reserved huge pages the MAP_POPULATE mmap() fails with ENOMEM and
> every test case is skipped.
>
> Reserve the pages with hugetlb_setup_default(). Each chunk size runs two
> HWPOISON tests and a poisoned huge page can't be reused, so reserve two
> pages per chunk size. The pool size is restored on exit. The poisoned
> pages remain allocated until reboot.
>
> With the patch, all 12 cases pass on a system with no huge pages
> reserved (tested on Fedora, x86_64, CONFIG_MEMORY_FAILURE=y).
>
> Assisted-by: LLM
> Signed-off-by: Jaeyeon Lee <[email protected]>
> ---
Looks good to me. Before the patch, all 12 test cases were skipping with
ENOMEM. After the patch, all test cases passed.
Tested-by: Sarthak Sharma <[email protected]>
Reviewed-by: Sarthak Sharma <[email protected]>
> tools/testing/selftests/mm/hugetlb-read-hwpoison.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/tools/testing/selftests/mm/hugetlb-read-hwpoison.c
> b/tools/testing/selftests/mm/hugetlb-read-hwpoison.c
> index 70b24e3660c4..31748a24448f 100644
> --- a/tools/testing/selftests/mm/hugetlb-read-hwpoison.c
> +++ b/tools/testing/selftests/mm/hugetlb-read-hwpoison.c
> @@ -13,6 +13,7 @@
> #include <signal.h>
>
> #include "kselftest.h"
> +#include "vm_util.h"
>
> #define MAX_WRITE_READ_CHUNK_SIZE (getpagesize() * 16)
> #define MAX(a, b) (((a) > (b)) ? (a) : (b))
> @@ -281,6 +282,13 @@ int main(void)
> size_t i;
>
> ksft_print_header();
> +
> + if (!hugetlb_available())
> + ksft_exit_skip("HugeTLB is not available\n");
> +
> + if (!hugetlb_setup_default(ARRAY_SIZE(wr_chunk_sizes) * 2))
> + ksft_exit_skip("Not enough free huge pages\n");
> +
> ksft_set_plan(ARRAY_SIZE(wr_chunk_sizes) * 3);
>
> signal(SIGBUS, sigbus_handler);
