Is number of process_madvise()-able ranges limited to 8? (was Re: [PATCH 1/6] prctl: introduce PR_THP_POLICY_DEFAULT_HUGE for the process)

[SeongJae Park <sj@kernel.org>]

Hi Lorenzo,

On Fri, 16 May 2025 13:57:18 +0100 Lorenzo Stoakes <lorenzo.stoakes@oracle.com> wrote:
[...]
> Right now madvise() has limited utility because:
[...]
> - While you can perform multiple operations at once via process_madvise(),
>   even to the current process (after my changes to extend it), it's limited
>   to a single advice over 8 ranges.

I'm bit confused by the last part, since I'm understanding your point as 'vlen'
parameter of process_madvise() is limited to 8, but my test code below succeeds
with 'vlen' parameter value 512.  Could you please enlighten me?

Attaching my test code below.  You could simply run it as below.

    gcc test.c && ./a.out

==== Attachment 0 (test.c) ====
#define _GNU_SOURCE

#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <sys/syscall.h>
#include <sys/uio.h>
#include <unistd.h>

#define SZ_PAGE (4096)
#define NR_PAGES (512)
#define MMAP_SZ	(SZ_PAGE * NR_PAGES)

int main(void)
{
	char *buf;
	unsigned int i;
	int ret;
	pid_t pid = getpid();
	int pidfd = syscall(SYS_pidfd_open, pid, 0);
	struct iovec *vec;

	buf = mmap(NULL, MMAP_SZ, PROT_READ | PROT_WRITE, MAP_PRIVATE |
			MAP_ANON, -1, 0);
	if (buf == MAP_FAILED) {
		printf("mmap fail\n");
		return -1;
	}

	for (i = 0; i < MMAP_SZ; i++)
		buf[i] = 123;

	vec = malloc(sizeof(*vec) * NR_PAGES);
	for (i = 0; i < NR_PAGES; i++) {
		vec[i].iov_base = &buf[i * SZ_PAGE];
		vec[i].iov_len = SZ_PAGE;
	}

	ret = syscall(SYS_process_madvise, pidfd, vec, NR_PAGES,
			MADV_DONTNEED, 0);
	if (ret != MMAP_SZ) {
		printf("process_madvise fail\n");
		return -1;
	}

	ret = munmap(buf, MMAP_SZ);
	if (ret) {
		printf("munmap failed\n");
		return -1;
	}

	close(pidfd);
	printf("good\n");
	return 0;
}