xref: /linux/tools/testing/selftests/mm/split_huge_page_test.c (revision dad4964a34c20cb86dcbedfe64ef7fe0728346df)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * A test of splitting PMD THPs and PTE-mapped THPs from a specified virtual
4  * address range in a process via <debugfs>/split_huge_pages interface.
5  */
6 
7 #define _GNU_SOURCE
8 #include <assert.h>
9 #include <stdio.h>
10 #include <stdlib.h>
11 #include <stdarg.h>
12 #include <unistd.h>
13 #include <inttypes.h>
14 #include <string.h>
15 #include <fcntl.h>
16 #include <sys/mman.h>
17 #include <sys/mount.h>
18 #include <sys/param.h>
19 #include <malloc.h>
20 #include <stdbool.h>
21 #include <time.h>
22 #include "vm_util.h"
23 #include "kselftest.h"
24 #include "thp_settings.h"
25 
26 uint64_t pagesize;
27 unsigned int pageshift;
28 uint64_t pmd_pagesize;
29 unsigned int pmd_order;
30 int *expected_orders;
31 
32 #define SPLIT_DEBUGFS "/sys/kernel/debug/split_huge_pages"
33 #define SMAP_PATH "/proc/self/smaps"
34 #define INPUT_MAX 80
35 
36 #define PID_FMT "%d,0x%lx,0x%lx,%d"
37 #define PID_FMT_OFFSET "%d,0x%lx,0x%lx,%d,%d"
38 #define PATH_FMT "%s,0x%lx,0x%lx,%d"
39 
40 const char *pagemap_proc = "/proc/self/pagemap";
41 const char *kpageflags_proc = "/proc/kpageflags";
42 int pagemap_fd;
43 int kpageflags_fd;
44 
45 static bool is_backed_by_folio(char *vaddr, int order, int pagemap_fd,
46 		int kpageflags_fd)
47 {
48 	const uint64_t folio_head_flags = KPF_THP | KPF_COMPOUND_HEAD;
49 	const uint64_t folio_tail_flags = KPF_THP | KPF_COMPOUND_TAIL;
50 	const unsigned long nr_pages = 1UL << order;
51 	unsigned long pfn_head;
52 	uint64_t pfn_flags;
53 	unsigned long pfn;
54 	unsigned long i;
55 
56 	pfn = pagemap_get_pfn(pagemap_fd, vaddr);
57 
58 	/* non present page */
59 	if (pfn == -1UL)
60 		return false;
61 
62 	if (pageflags_get(pfn, kpageflags_fd, &pfn_flags))
63 		goto fail;
64 
65 	/* check for order-0 pages */
66 	if (!order) {
67 		if (pfn_flags & (folio_head_flags | folio_tail_flags))
68 			return false;
69 		return true;
70 	}
71 
72 	/* non THP folio */
73 	if (!(pfn_flags & KPF_THP))
74 		return false;
75 
76 	pfn_head = pfn & ~(nr_pages - 1);
77 
78 	if (pageflags_get(pfn_head, kpageflags_fd, &pfn_flags))
79 		goto fail;
80 
81 	/* head PFN has no compound_head flag set */
82 	if ((pfn_flags & folio_head_flags) != folio_head_flags)
83 		return false;
84 
85 	/* check all tail PFN flags */
86 	for (i = 1; i < nr_pages; i++) {
87 		if (pageflags_get(pfn_head + i, kpageflags_fd, &pfn_flags))
88 			goto fail;
89 		if ((pfn_flags & folio_tail_flags) != folio_tail_flags)
90 			return false;
91 	}
92 
93 	/*
94 	 * check the PFN after this folio, but if its flags cannot be obtained,
95 	 * assume this folio has the expected order
96 	 */
97 	if (pageflags_get(pfn_head + nr_pages, kpageflags_fd, &pfn_flags))
98 		return true;
99 
100 	/* If we find another tail page, then the folio is larger. */
101 	return (pfn_flags & folio_tail_flags) != folio_tail_flags;
102 fail:
103 	ksft_exit_fail_msg("Failed to get folio info\n");
104 	return false;
105 }
106 
107 static int vaddr_pageflags_get(char *vaddr, int pagemap_fd, int kpageflags_fd,
108 		uint64_t *flags)
109 {
110 	unsigned long pfn;
111 
112 	pfn = pagemap_get_pfn(pagemap_fd, vaddr);
113 
114 	/* non-present PFN */
115 	if (pfn == -1UL)
116 		return 1;
117 
118 	if (pageflags_get(pfn, kpageflags_fd, flags))
119 		return -1;
120 
121 	return 0;
122 }
123 
124 /*
125  * gather_after_split_folio_orders - scan through [vaddr_start, len) and record
126  * folio orders
127  *
128  * @vaddr_start: start vaddr
129  * @len: range length
130  * @pagemap_fd: file descriptor to /proc/<pid>/pagemap
131  * @kpageflags_fd: file descriptor to /proc/kpageflags
132  * @orders: output folio order array
133  * @nr_orders: folio order array size
134  *
135  * gather_after_split_folio_orders() scan through [vaddr_start, len) and check
136  * all folios within the range and record their orders. All order-0 pages will
137  * be recorded. Non-present vaddr is skipped.
138  *
139  * NOTE: the function is used to check folio orders after a split is performed,
140  * so it assumes [vaddr_start, len) fully maps to after-split folios within that
141  * range.
142  *
143  * Return: 0 - no error, -1 - unhandled cases
144  */
145 static int gather_after_split_folio_orders(char *vaddr_start, size_t len,
146 		int pagemap_fd, int kpageflags_fd, int orders[], int nr_orders)
147 {
148 	uint64_t page_flags = 0;
149 	int cur_order = -1;
150 	char *vaddr;
151 
152 	if (pagemap_fd == -1 || kpageflags_fd == -1)
153 		return -1;
154 	if (!orders)
155 		return -1;
156 	if (nr_orders <= 0)
157 		return -1;
158 
159 	for (vaddr = vaddr_start; vaddr < vaddr_start + len;) {
160 		char *next_folio_vaddr;
161 		int status;
162 
163 		status = vaddr_pageflags_get(vaddr, pagemap_fd, kpageflags_fd,
164 					&page_flags);
165 		if (status < 0)
166 			return -1;
167 
168 		/* skip non present vaddr */
169 		if (status == 1) {
170 			vaddr += psize();
171 			continue;
172 		}
173 
174 		/* all order-0 pages with possible false postive (non folio) */
175 		if (!(page_flags & (KPF_COMPOUND_HEAD | KPF_COMPOUND_TAIL))) {
176 			orders[0]++;
177 			vaddr += psize();
178 			continue;
179 		}
180 
181 		/* skip non thp compound pages */
182 		if (!(page_flags & KPF_THP)) {
183 			vaddr += psize();
184 			continue;
185 		}
186 
187 		/* vpn points to part of a THP at this point */
188 		if (page_flags & KPF_COMPOUND_HEAD)
189 			cur_order = 1;
190 		else {
191 			vaddr += psize();
192 			continue;
193 		}
194 
195 		next_folio_vaddr = vaddr + (1UL << (cur_order + pshift()));
196 
197 		if (next_folio_vaddr >= vaddr_start + len)
198 			break;
199 
200 		while ((status = vaddr_pageflags_get(next_folio_vaddr,
201 						     pagemap_fd, kpageflags_fd,
202 						     &page_flags)) >= 0) {
203 			/*
204 			 * non present vaddr, next compound head page, or
205 			 * order-0 page
206 			 */
207 			if (status == 1 ||
208 			    (page_flags & KPF_COMPOUND_HEAD) ||
209 			    !(page_flags & (KPF_COMPOUND_HEAD | KPF_COMPOUND_TAIL))) {
210 				if (cur_order < nr_orders) {
211 					orders[cur_order]++;
212 					cur_order = -1;
213 					vaddr = next_folio_vaddr;
214 				}
215 				break;
216 			}
217 
218 			cur_order++;
219 			next_folio_vaddr = vaddr + (1UL << (cur_order + pshift()));
220 		}
221 
222 		if (status < 0)
223 			return status;
224 	}
225 	if (cur_order > 0 && cur_order < nr_orders)
226 		orders[cur_order]++;
227 	return 0;
228 }
229 
230 static int check_after_split_folio_orders(char *vaddr_start, size_t len,
231 		int pagemap_fd, int kpageflags_fd, int orders[], int nr_orders)
232 {
233 	int *vaddr_orders;
234 	int status;
235 	int i;
236 
237 	vaddr_orders = (int *)malloc(sizeof(int) * nr_orders);
238 
239 	if (!vaddr_orders)
240 		ksft_exit_fail_msg("Cannot allocate memory for vaddr_orders");
241 
242 	memset(vaddr_orders, 0, sizeof(int) * nr_orders);
243 	status = gather_after_split_folio_orders(vaddr_start, len, pagemap_fd,
244 				     kpageflags_fd, vaddr_orders, nr_orders);
245 	if (status)
246 		ksft_exit_fail_msg("gather folio info failed\n");
247 
248 	for (i = 0; i < nr_orders; i++)
249 		if (vaddr_orders[i] != orders[i]) {
250 			ksft_print_msg("order %d: expected: %d got %d\n", i,
251 				       orders[i], vaddr_orders[i]);
252 			status = -1;
253 		}
254 
255 	free(vaddr_orders);
256 	return status;
257 }
258 
259 static void write_debugfs(const char *fmt, ...)
260 {
261 	char input[INPUT_MAX];
262 	int ret;
263 	va_list argp;
264 
265 	va_start(argp, fmt);
266 	ret = vsnprintf(input, INPUT_MAX, fmt, argp);
267 	va_end(argp);
268 
269 	if (ret >= INPUT_MAX)
270 		ksft_exit_fail_msg("%s: Debugfs input is too long\n", __func__);
271 
272 	write_file(SPLIT_DEBUGFS, input, ret + 1);
273 }
274 
275 static char *allocate_zero_filled_hugepage(size_t len)
276 {
277 	char *result;
278 	size_t i;
279 
280 	result = memalign(pmd_pagesize, len);
281 	if (!result) {
282 		printf("Fail to allocate memory\n");
283 		exit(EXIT_FAILURE);
284 	}
285 
286 	madvise(result, len, MADV_HUGEPAGE);
287 
288 	for (i = 0; i < len; i++)
289 		result[i] = (char)0;
290 
291 	return result;
292 }
293 
294 static void verify_rss_anon_split_huge_page_all_zeroes(char *one_page, int nr_hpages, size_t len)
295 {
296 	unsigned long rss_anon_before, rss_anon_after;
297 	size_t i;
298 
299 	if (!check_huge_anon(one_page, nr_hpages, pmd_pagesize))
300 		ksft_exit_fail_msg("No THP is allocated\n");
301 
302 	rss_anon_before = rss_anon();
303 	if (!rss_anon_before)
304 		ksft_exit_fail_msg("No RssAnon is allocated before split\n");
305 
306 	/* split all THPs */
307 	write_debugfs(PID_FMT, getpid(), (uint64_t)one_page,
308 		      (uint64_t)one_page + len, 0);
309 
310 	for (i = 0; i < len; i++)
311 		if (one_page[i] != (char)0)
312 			ksft_exit_fail_msg("%ld byte corrupted\n", i);
313 
314 	if (!check_huge_anon(one_page, 0, pmd_pagesize))
315 		ksft_exit_fail_msg("Still AnonHugePages not split\n");
316 
317 	rss_anon_after = rss_anon();
318 	if (rss_anon_after >= rss_anon_before)
319 		ksft_exit_fail_msg("Incorrect RssAnon value. Before: %ld After: %ld\n",
320 		       rss_anon_before, rss_anon_after);
321 }
322 
323 static void split_pmd_zero_pages(void)
324 {
325 	char *one_page;
326 	int nr_hpages = 4;
327 	size_t len = nr_hpages * pmd_pagesize;
328 
329 	one_page = allocate_zero_filled_hugepage(len);
330 	verify_rss_anon_split_huge_page_all_zeroes(one_page, nr_hpages, len);
331 	ksft_test_result_pass("Split zero filled huge pages successful\n");
332 	free(one_page);
333 }
334 
335 static void split_pmd_thp_to_order(int order)
336 {
337 	char *one_page;
338 	size_t len = 4 * pmd_pagesize;
339 	size_t i;
340 
341 	one_page = memalign(pmd_pagesize, len);
342 	if (!one_page)
343 		ksft_exit_fail_msg("Fail to allocate memory: %s\n", strerror(errno));
344 
345 	madvise(one_page, len, MADV_HUGEPAGE);
346 
347 	for (i = 0; i < len; i++)
348 		one_page[i] = (char)i;
349 
350 	if (!check_huge_anon(one_page, 4, pmd_pagesize))
351 		ksft_exit_fail_msg("No THP is allocated\n");
352 
353 	/* split all THPs */
354 	write_debugfs(PID_FMT, getpid(), (uint64_t)one_page,
355 		(uint64_t)one_page + len, order);
356 
357 	for (i = 0; i < len; i++)
358 		if (one_page[i] != (char)i)
359 			ksft_exit_fail_msg("%ld byte corrupted\n", i);
360 
361 	memset(expected_orders, 0, sizeof(int) * (pmd_order + 1));
362 	expected_orders[order] = 4 << (pmd_order - order);
363 
364 	if (check_after_split_folio_orders(one_page, len, pagemap_fd,
365 					   kpageflags_fd, expected_orders,
366 					   (pmd_order + 1)))
367 		ksft_exit_fail_msg("Unexpected THP split\n");
368 
369 	if (!check_huge_anon(one_page, 0, pmd_pagesize))
370 		ksft_exit_fail_msg("Still AnonHugePages not split\n");
371 
372 	ksft_test_result_pass("Split huge pages to order %d successful\n", order);
373 	free(one_page);
374 }
375 
376 static void split_pte_mapped_thp(void)
377 {
378 	const size_t nr_thps = 4;
379 	const size_t thp_area_size = nr_thps * pmd_pagesize;
380 	const size_t page_area_size = nr_thps * pagesize;
381 	char *thp_area, *tmp, *page_area = MAP_FAILED;
382 	size_t i;
383 
384 	thp_area = mmap((void *)(1UL << 30), thp_area_size, PROT_READ | PROT_WRITE,
385 			MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
386 	if (thp_area == MAP_FAILED) {
387 		ksft_test_result_fail("Fail to allocate memory: %s\n", strerror(errno));
388 		return;
389 	}
390 
391 	madvise(thp_area, thp_area_size, MADV_HUGEPAGE);
392 
393 	for (i = 0; i < thp_area_size; i++)
394 		thp_area[i] = (char)i;
395 
396 	if (!check_huge_anon(thp_area, nr_thps, pmd_pagesize)) {
397 		ksft_test_result_skip("Not all THPs allocated\n");
398 		goto out;
399 	}
400 
401 	/*
402 	 * To challenge spitting code, we will mremap a single page of each
403 	 * THP (page[i] of thp[i]) in the thp_area into page_area. This will
404 	 * replace the PMD mappings in the thp_area by PTE mappings first,
405 	 * but leaving the THP unsplit, to then create a page-sized hole in
406 	 * the thp_area.
407 	 * We will then manually trigger splitting of all THPs through the
408 	 * single mremap'ed pages of each THP in the page_area.
409 	 */
410 	page_area = mmap(NULL, page_area_size, PROT_READ | PROT_WRITE,
411 			MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
412 	if (page_area == MAP_FAILED) {
413 		ksft_test_result_fail("Fail to allocate memory: %s\n", strerror(errno));
414 		goto out;
415 	}
416 
417 	for (i = 0; i < nr_thps; i++) {
418 		tmp = mremap(thp_area + pmd_pagesize * i + pagesize * i,
419 			     pagesize, pagesize, MREMAP_MAYMOVE|MREMAP_FIXED,
420 			     page_area + pagesize * i);
421 		if (tmp != MAP_FAILED)
422 			continue;
423 		ksft_test_result_fail("mremap failed: %s\n", strerror(errno));
424 		goto out;
425 	}
426 
427 	/*
428 	 * Verify that our THPs were not split yet. Note that
429 	 * check_huge_anon() cannot be used as it checks for PMD mappings.
430 	 */
431 	for (i = 0; i < nr_thps; i++) {
432 		if (is_backed_by_folio(page_area + i * pagesize, pmd_order,
433 				       pagemap_fd, kpageflags_fd))
434 			continue;
435 		ksft_test_result_fail("THP %zu missing after mremap\n", i);
436 		goto out;
437 	}
438 
439 	/* Split all THPs through the remapped pages. */
440 	write_debugfs(PID_FMT, getpid(), (uint64_t)page_area,
441 		      (uint64_t)page_area + page_area_size, 0);
442 
443 	/* Corruption during mremap or split? */
444 	for (i = 0; i < page_area_size; i++) {
445 		if (page_area[i] == (char)i)
446 			continue;
447 		ksft_test_result_fail("%zu byte corrupted\n", i);
448 		goto out;
449 	}
450 
451 	/* Split failed? */
452 	for (i = 0; i < nr_thps; i++) {
453 		if (is_backed_by_folio(page_area + i * pagesize, 0,
454 				       pagemap_fd, kpageflags_fd))
455 			continue;
456 		ksft_test_result_fail("THP %zu not split\n", i);
457 	}
458 
459 	ksft_test_result_pass("Split PTE-mapped huge pages successful\n");
460 out:
461 	munmap(thp_area, thp_area_size);
462 	if (page_area != MAP_FAILED)
463 		munmap(page_area, page_area_size);
464 }
465 
466 static void split_file_backed_thp(int order)
467 {
468 	int status;
469 	int fd;
470 	char tmpfs_template[] = "/tmp/thp_split_XXXXXX";
471 	const char *tmpfs_loc = mkdtemp(tmpfs_template);
472 	char testfile[INPUT_MAX];
473 	ssize_t num_written, num_read;
474 	char *file_buf1, *file_buf2;
475 	uint64_t pgoff_start = 0, pgoff_end = 1024;
476 	int i;
477 
478 	ksft_print_msg("Please enable pr_debug in split_huge_pages_in_file() for more info.\n");
479 
480 	file_buf1 = (char *)malloc(pmd_pagesize);
481 	file_buf2 = (char *)malloc(pmd_pagesize);
482 
483 	if (!file_buf1 || !file_buf2) {
484 		ksft_print_msg("cannot allocate file buffers\n");
485 		goto out;
486 	}
487 
488 	for (i = 0; i < pmd_pagesize; i++)
489 		file_buf1[i] = (char)i;
490 	memset(file_buf2, 0, pmd_pagesize);
491 
492 	status = mount("tmpfs", tmpfs_loc, "tmpfs", 0, "huge=always,size=4m");
493 
494 	if (status)
495 		ksft_exit_fail_msg("Unable to create a tmpfs for testing\n");
496 
497 	status = snprintf(testfile, INPUT_MAX, "%s/thp_file", tmpfs_loc);
498 	if (status >= INPUT_MAX) {
499 		ksft_print_msg("Fail to create file-backed THP split testing file\n");
500 		goto cleanup;
501 	}
502 
503 	fd = open(testfile, O_CREAT|O_RDWR, 0664);
504 	if (fd == -1) {
505 		ksft_perror("Cannot open testing file");
506 		goto cleanup;
507 	}
508 
509 	/* write pmd size data to the file, so a file-backed THP can be allocated */
510 	num_written = write(fd, file_buf1, pmd_pagesize);
511 
512 	if (num_written == -1 || num_written != pmd_pagesize) {
513 		ksft_perror("Failed to write data to testing file");
514 		goto close_file;
515 	}
516 
517 	/* split the file-backed THP */
518 	write_debugfs(PATH_FMT, testfile, pgoff_start, pgoff_end, order);
519 
520 	/* check file content after split */
521 	status = lseek(fd, 0, SEEK_SET);
522 	if (status == -1) {
523 		ksft_perror("Cannot lseek file");
524 		goto close_file;
525 	}
526 
527 	num_read = read(fd, file_buf2, num_written);
528 	if (num_read == -1 || num_read != num_written) {
529 		ksft_perror("Cannot read file content back");
530 		goto close_file;
531 	}
532 
533 	if (strncmp(file_buf1, file_buf2, pmd_pagesize) != 0) {
534 		ksft_print_msg("File content changed\n");
535 		goto close_file;
536 	}
537 
538 	close(fd);
539 	status = unlink(testfile);
540 	if (status) {
541 		ksft_perror("Cannot remove testing file");
542 		goto cleanup;
543 	}
544 
545 	status = umount(tmpfs_loc);
546 	if (status) {
547 		rmdir(tmpfs_loc);
548 		ksft_exit_fail_msg("Unable to umount %s\n", tmpfs_loc);
549 	}
550 
551 	status = rmdir(tmpfs_loc);
552 	if (status)
553 		ksft_exit_fail_msg("cannot remove tmp dir: %s\n", strerror(errno));
554 
555 	ksft_print_msg("Please check dmesg for more information\n");
556 	ksft_test_result_pass("File-backed THP split to order %d test done\n", order);
557 	return;
558 
559 close_file:
560 	close(fd);
561 cleanup:
562 	umount(tmpfs_loc);
563 	rmdir(tmpfs_loc);
564 out:
565 	ksft_exit_fail_msg("Error occurred\n");
566 }
567 
568 static bool prepare_thp_fs(const char *xfs_path, char *thp_fs_template,
569 		const char **thp_fs_loc)
570 {
571 	if (xfs_path) {
572 		*thp_fs_loc = xfs_path;
573 		return false;
574 	}
575 
576 	*thp_fs_loc = mkdtemp(thp_fs_template);
577 
578 	if (!*thp_fs_loc)
579 		ksft_exit_fail_msg("cannot create temp folder\n");
580 
581 	return true;
582 }
583 
584 static void cleanup_thp_fs(const char *thp_fs_loc, bool created_tmp)
585 {
586 	int status;
587 
588 	if (!created_tmp)
589 		return;
590 
591 	status = rmdir(thp_fs_loc);
592 	if (status)
593 		ksft_exit_fail_msg("cannot remove tmp dir: %s\n",
594 				   strerror(errno));
595 }
596 
597 static int create_pagecache_thp_and_fd(const char *testfile, size_t fd_size,
598 		int *fd, char **addr)
599 {
600 	size_t i;
601 	unsigned char buf[1024];
602 
603 	srand(time(NULL));
604 
605 	*fd = open(testfile, O_CREAT | O_RDWR, 0664);
606 	if (*fd == -1)
607 		ksft_exit_fail_msg("Failed to create a file at %s\n", testfile);
608 
609 	assert(fd_size % sizeof(buf) == 0);
610 	for (i = 0; i < sizeof(buf); i++)
611 		buf[i] = (unsigned char)i;
612 	for (i = 0; i < fd_size; i += sizeof(buf))
613 		write(*fd, buf, sizeof(buf));
614 
615 	close(*fd);
616 	sync();
617 	*fd = open("/proc/sys/vm/drop_caches", O_WRONLY);
618 	if (*fd == -1) {
619 		ksft_perror("open drop_caches");
620 		goto err_out_unlink;
621 	}
622 	if (write(*fd, "3", 1) != 1) {
623 		ksft_perror("write to drop_caches");
624 		goto err_out_unlink;
625 	}
626 	close(*fd);
627 
628 	*fd = open(testfile, O_RDWR);
629 	if (*fd == -1) {
630 		ksft_perror("Failed to open testfile\n");
631 		goto err_out_unlink;
632 	}
633 
634 	*addr = mmap(NULL, fd_size, PROT_READ|PROT_WRITE, MAP_SHARED, *fd, 0);
635 	if (*addr == (char *)-1) {
636 		ksft_perror("cannot mmap");
637 		goto err_out_close;
638 	}
639 	madvise(*addr, fd_size, MADV_HUGEPAGE);
640 
641 	force_read_pages(*addr, fd_size / pmd_pagesize, pmd_pagesize);
642 
643 	if (!check_huge_file(*addr, fd_size / pmd_pagesize, pmd_pagesize)) {
644 		ksft_print_msg("No large pagecache folio generated, please provide a filesystem supporting large folio\n");
645 		munmap(*addr, fd_size);
646 		close(*fd);
647 		unlink(testfile);
648 		ksft_test_result_skip("Pagecache folio split skipped\n");
649 		return -2;
650 	}
651 	return 0;
652 err_out_close:
653 	close(*fd);
654 err_out_unlink:
655 	unlink(testfile);
656 	ksft_exit_fail_msg("Failed to create large pagecache folios\n");
657 	return -1;
658 }
659 
660 static void split_thp_in_pagecache_to_order_at(size_t fd_size,
661 		const char *fs_loc, int order, int offset)
662 {
663 	int fd;
664 	char *split_addr;
665 	char *addr;
666 	size_t i;
667 	char testfile[INPUT_MAX];
668 	int err = 0;
669 
670 	err = snprintf(testfile, INPUT_MAX, "%s/test", fs_loc);
671 
672 	if (err < 0)
673 		ksft_exit_fail_msg("cannot generate right test file name\n");
674 
675 	err = create_pagecache_thp_and_fd(testfile, fd_size, &fd, &addr);
676 	if (err)
677 		return;
678 
679 	err = 0;
680 
681 	memset(expected_orders, 0, sizeof(int) * (pmd_order + 1));
682 	/*
683 	 * use [split_addr, split_addr + pagesize) range to split THPs, since
684 	 * the debugfs function always split a range with pagesize step and
685 	 * providing a full [addr, addr + fd_size) range can trigger multiple
686 	 * splits, complicating after-split result checking.
687 	 */
688 	if (offset == -1) {
689 		for (split_addr = addr; split_addr < addr + fd_size; split_addr += pmd_pagesize)
690 			write_debugfs(PID_FMT, getpid(), (uint64_t)split_addr,
691 				      (uint64_t)split_addr + pagesize, order);
692 
693 		expected_orders[order] = fd_size / (pagesize << order);
694 	} else {
695 		int times = fd_size / pmd_pagesize;
696 
697 		for (split_addr = addr; split_addr < addr + fd_size; split_addr += pmd_pagesize)
698 			write_debugfs(PID_FMT_OFFSET, getpid(), (uint64_t)split_addr,
699 				      (uint64_t)split_addr + pagesize, order, offset);
700 
701 		for (i = order + 1; i < pmd_order; i++)
702 			expected_orders[i] = times;
703 		expected_orders[order] = 2 * times;
704 	}
705 
706 	for (i = 0; i < fd_size; i++)
707 		if (*(addr + i) != (char)i) {
708 			ksft_print_msg("%lu byte corrupted in the file\n", i);
709 			err = EXIT_FAILURE;
710 			goto out;
711 		}
712 
713 	if (check_after_split_folio_orders(addr, fd_size, pagemap_fd,
714 					   kpageflags_fd, expected_orders,
715 					   (pmd_order + 1))) {
716 		ksft_print_msg("Unexpected THP split\n");
717 		err = 1;
718 		goto out;
719 	}
720 
721 	if (!check_huge_file(addr, 0, pmd_pagesize)) {
722 		ksft_print_msg("Still FilePmdMapped not split\n");
723 		err = EXIT_FAILURE;
724 		goto out;
725 	}
726 
727 out:
728 	munmap(addr, fd_size);
729 	close(fd);
730 	unlink(testfile);
731 	if (offset == -1) {
732 		if (err)
733 			ksft_exit_fail_msg("Split PMD-mapped pagecache folio to order %d failed\n", order);
734 		ksft_test_result_pass("Split PMD-mapped pagecache folio to order %d passed\n", order);
735 	} else {
736 		if (err)
737 			ksft_exit_fail_msg("Split PMD-mapped pagecache folio to order %d at in-folio offset %d failed\n", order, offset);
738 		ksft_test_result_pass("Split PMD-mapped pagecache folio to order %d at in-folio offset %d passed\n", order, offset);
739 	}
740 }
741 
742 int main(int argc, char **argv)
743 {
744 	int i;
745 	size_t fd_size;
746 	char *optional_xfs_path = NULL;
747 	char fs_loc_template[] = "/tmp/thp_fs_XXXXXX";
748 	const char *fs_loc;
749 	bool created_tmp;
750 	int offset;
751 	unsigned int nr_pages;
752 	unsigned int tests;
753 
754 	ksft_print_header();
755 
756 	if (geteuid() != 0) {
757 		ksft_print_msg("Please run the benchmark as root\n");
758 		ksft_finished();
759 	}
760 
761 	if (!thp_is_enabled())
762 		ksft_exit_skip("Transparent Hugepages not available\n");
763 
764 	if (argc > 1)
765 		optional_xfs_path = argv[1];
766 
767 	pagesize = getpagesize();
768 	pageshift = ffs(pagesize) - 1;
769 	pmd_pagesize = read_pmd_pagesize();
770 	if (!pmd_pagesize)
771 		ksft_exit_fail_msg("Reading PMD pagesize failed\n");
772 
773 	nr_pages = pmd_pagesize / pagesize;
774 	pmd_order = sz2ord(pmd_pagesize, pagesize);
775 
776 	expected_orders = (int *)malloc(sizeof(int) * (pmd_order + 1));
777 	if (!expected_orders)
778 		ksft_exit_fail_msg("Fail to allocate memory: %s\n", strerror(errno));
779 
780 	tests = 2 + (pmd_order - 1) + (2 * pmd_order) + (pmd_order - 1) * 4 + 2;
781 	ksft_set_plan(tests);
782 
783 	pagemap_fd = open(pagemap_proc, O_RDONLY);
784 	if (pagemap_fd == -1)
785 		ksft_exit_fail_msg("read pagemap: %s\n", strerror(errno));
786 
787 	kpageflags_fd = open(kpageflags_proc, O_RDONLY);
788 	if (kpageflags_fd == -1)
789 		ksft_exit_fail_msg("read kpageflags: %s\n", strerror(errno));
790 
791 	fd_size = 2 * pmd_pagesize;
792 
793 	split_pmd_zero_pages();
794 
795 	for (i = 0; i < pmd_order; i++)
796 		if (i != 1)
797 			split_pmd_thp_to_order(i);
798 
799 	split_pte_mapped_thp();
800 	for (i = 0; i < pmd_order; i++)
801 		split_file_backed_thp(i);
802 
803 	created_tmp = prepare_thp_fs(optional_xfs_path, fs_loc_template,
804 			&fs_loc);
805 	for (i = pmd_order - 1; i >= 0; i--)
806 		split_thp_in_pagecache_to_order_at(fd_size, fs_loc, i, -1);
807 
808 	for (i = 0; i < pmd_order; i++)
809 		for (offset = 0;
810 		     offset < nr_pages;
811 		     offset += MAX(nr_pages / 4, 1 << i))
812 			split_thp_in_pagecache_to_order_at(fd_size, fs_loc, i, offset);
813 	cleanup_thp_fs(fs_loc, created_tmp);
814 
815 	close(pagemap_fd);
816 	close(kpageflags_fd);
817 	free(expected_orders);
818 
819 	ksft_finished();
820 
821 	return 0;
822 }
823