1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/mm.h>
3 #include <linux/mmzone.h>
4 #include <linux/memblock.h>
5 #include <linux/page_ext.h>
6 #include <linux/memory.h>
7 #include <linux/vmalloc.h>
8 #include <linux/kmemleak.h>
9 #include <linux/page_owner.h>
10 #include <linux/page_idle.h>
11 #include <linux/page_table_check.h>
12 #include <linux/rcupdate.h>
13 
14 /*
15  * struct page extension
16  *
17  * This is the feature to manage memory for extended data per page.
18  *
19  * Until now, we must modify struct page itself to store extra data per page.
20  * This requires rebuilding the kernel and it is really time consuming process.
21  * And, sometimes, rebuild is impossible due to third party module dependency.
22  * At last, enlarging struct page could cause un-wanted system behaviour change.
23  *
24  * This feature is intended to overcome above mentioned problems. This feature
25  * allocates memory for extended data per page in certain place rather than
26  * the struct page itself. This memory can be accessed by the accessor
27  * functions provided by this code. During the boot process, it checks whether
28  * allocation of huge chunk of memory is needed or not. If not, it avoids
29  * allocating memory at all. With this advantage, we can include this feature
30  * into the kernel in default and can avoid rebuild and solve related problems.
31  *
32  * To help these things to work well, there are two callbacks for clients. One
33  * is the need callback which is mandatory if user wants to avoid useless
34  * memory allocation at boot-time. The other is optional, init callback, which
35  * is used to do proper initialization after memory is allocated.
36  *
37  * The need callback is used to decide whether extended memory allocation is
38  * needed or not. Sometimes users want to deactivate some features in this
39  * boot and extra memory would be unnecessary. In this case, to avoid
40  * allocating huge chunk of memory, each clients represent their need of
41  * extra memory through the need callback. If one of the need callbacks
42  * returns true, it means that someone needs extra memory so that
43  * page extension core should allocates memory for page extension. If
44  * none of need callbacks return true, memory isn't needed at all in this boot
45  * and page extension core can skip to allocate memory. As result,
46  * none of memory is wasted.
47  *
48  * When need callback returns true, page_ext checks if there is a request for
49  * extra memory through size in struct page_ext_operations. If it is non-zero,
50  * extra space is allocated for each page_ext entry and offset is returned to
51  * user through offset in struct page_ext_operations.
52  *
53  * The init callback is used to do proper initialization after page extension
54  * is completely initialized. In sparse memory system, extra memory is
55  * allocated some time later than memmap is allocated. In other words, lifetime
56  * of memory for page extension isn't same with memmap for struct page.
57  * Therefore, clients can't store extra data until page extension is
58  * initialized, even if pages are allocated and used freely. This could
59  * cause inadequate state of extra data per page, so, to prevent it, client
60  * can utilize this callback to initialize the state of it correctly.
61  */
62 
63 #ifdef CONFIG_SPARSEMEM
64 #define PAGE_EXT_INVALID       (0x1)
65 #endif
66 
67 #if defined(CONFIG_PAGE_IDLE_FLAG) && !defined(CONFIG_64BIT)
need_page_idle(void)68 static bool need_page_idle(void)
69 {
70 	return true;
71 }
72 static struct page_ext_operations page_idle_ops __initdata = {
73 	.need = need_page_idle,
74 	.need_shared_flags = true,
75 };
76 #endif
77 
78 static struct page_ext_operations *page_ext_ops[] __initdata = {
79 #ifdef CONFIG_PAGE_OWNER
80 	&page_owner_ops,
81 #endif
82 #if defined(CONFIG_PAGE_IDLE_FLAG) && !defined(CONFIG_64BIT)
83 	&page_idle_ops,
84 #endif
85 #ifdef CONFIG_PAGE_TABLE_CHECK
86 	&page_table_check_ops,
87 #endif
88 };
89 
90 unsigned long page_ext_size;
91 
92 static unsigned long total_usage;
93 static struct page_ext *lookup_page_ext(const struct page *page);
94 
95 bool early_page_ext __meminitdata;
setup_early_page_ext(char * str)96 static int __init setup_early_page_ext(char *str)
97 {
98 	early_page_ext = true;
99 	return 0;
100 }
101 early_param("early_page_ext", setup_early_page_ext);
102 
invoke_need_callbacks(void)103 static bool __init invoke_need_callbacks(void)
104 {
105 	int i;
106 	int entries = ARRAY_SIZE(page_ext_ops);
107 	bool need = false;
108 
109 	for (i = 0; i < entries; i++) {
110 		if (page_ext_ops[i]->need()) {
111 			if (page_ext_ops[i]->need_shared_flags) {
112 				page_ext_size = sizeof(struct page_ext);
113 				break;
114 			}
115 		}
116 	}
117 
118 	for (i = 0; i < entries; i++) {
119 		if (page_ext_ops[i]->need()) {
120 			page_ext_ops[i]->offset = page_ext_size;
121 			page_ext_size += page_ext_ops[i]->size;
122 			need = true;
123 		}
124 	}
125 
126 	return need;
127 }
128 
invoke_init_callbacks(void)129 static void __init invoke_init_callbacks(void)
130 {
131 	int i;
132 	int entries = ARRAY_SIZE(page_ext_ops);
133 
134 	for (i = 0; i < entries; i++) {
135 		if (page_ext_ops[i]->init)
136 			page_ext_ops[i]->init();
137 	}
138 }
139 
140 #ifndef CONFIG_SPARSEMEM
page_ext_init_flatmem_late(void)141 void __init page_ext_init_flatmem_late(void)
142 {
143 	invoke_init_callbacks();
144 }
145 #endif
146 
get_entry(void * base,unsigned long index)147 static inline struct page_ext *get_entry(void *base, unsigned long index)
148 {
149 	return base + page_ext_size * index;
150 }
151 
152 /**
153  * page_ext_get() - Get the extended information for a page.
154  * @page: The page we're interested in.
155  *
156  * Ensures that the page_ext will remain valid until page_ext_put()
157  * is called.
158  *
159  * Return: NULL if no page_ext exists for this page.
160  * Context: Any context.  Caller may not sleep until they have called
161  * page_ext_put().
162  */
page_ext_get(struct page * page)163 struct page_ext *page_ext_get(struct page *page)
164 {
165 	struct page_ext *page_ext;
166 
167 	rcu_read_lock();
168 	page_ext = lookup_page_ext(page);
169 	if (!page_ext) {
170 		rcu_read_unlock();
171 		return NULL;
172 	}
173 
174 	return page_ext;
175 }
176 
177 /**
178  * page_ext_put() - Working with page extended information is done.
179  * @page_ext: Page extended information received from page_ext_get().
180  *
181  * The page extended information of the page may not be valid after this
182  * function is called.
183  *
184  * Return: None.
185  * Context: Any context with corresponding page_ext_get() is called.
186  */
page_ext_put(struct page_ext * page_ext)187 void page_ext_put(struct page_ext *page_ext)
188 {
189 	if (unlikely(!page_ext))
190 		return;
191 
192 	rcu_read_unlock();
193 }
194 #ifndef CONFIG_SPARSEMEM
195 
196 
pgdat_page_ext_init(struct pglist_data * pgdat)197 void __meminit pgdat_page_ext_init(struct pglist_data *pgdat)
198 {
199 	pgdat->node_page_ext = NULL;
200 }
201 
lookup_page_ext(const struct page * page)202 static struct page_ext *lookup_page_ext(const struct page *page)
203 {
204 	unsigned long pfn = page_to_pfn(page);
205 	unsigned long index;
206 	struct page_ext *base;
207 
208 	WARN_ON_ONCE(!rcu_read_lock_held());
209 	base = NODE_DATA(page_to_nid(page))->node_page_ext;
210 	/*
211 	 * The sanity checks the page allocator does upon freeing a
212 	 * page can reach here before the page_ext arrays are
213 	 * allocated when feeding a range of pages to the allocator
214 	 * for the first time during bootup or memory hotplug.
215 	 */
216 	if (unlikely(!base))
217 		return NULL;
218 	index = pfn - round_down(node_start_pfn(page_to_nid(page)),
219 					MAX_ORDER_NR_PAGES);
220 	return get_entry(base, index);
221 }
222 
alloc_node_page_ext(int nid)223 static int __init alloc_node_page_ext(int nid)
224 {
225 	struct page_ext *base;
226 	unsigned long table_size;
227 	unsigned long nr_pages;
228 
229 	nr_pages = NODE_DATA(nid)->node_spanned_pages;
230 	if (!nr_pages)
231 		return 0;
232 
233 	/*
234 	 * Need extra space if node range is not aligned with
235 	 * MAX_ORDER_NR_PAGES. When page allocator's buddy algorithm
236 	 * checks buddy's status, range could be out of exact node range.
237 	 */
238 	if (!IS_ALIGNED(node_start_pfn(nid), MAX_ORDER_NR_PAGES) ||
239 		!IS_ALIGNED(node_end_pfn(nid), MAX_ORDER_NR_PAGES))
240 		nr_pages += MAX_ORDER_NR_PAGES;
241 
242 	table_size = page_ext_size * nr_pages;
243 
244 	base = memblock_alloc_try_nid(
245 			table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
246 			MEMBLOCK_ALLOC_ACCESSIBLE, nid);
247 	if (!base)
248 		return -ENOMEM;
249 	NODE_DATA(nid)->node_page_ext = base;
250 	total_usage += table_size;
251 	return 0;
252 }
253 
page_ext_init_flatmem(void)254 void __init page_ext_init_flatmem(void)
255 {
256 
257 	int nid, fail;
258 
259 	if (!invoke_need_callbacks())
260 		return;
261 
262 	for_each_online_node(nid)  {
263 		fail = alloc_node_page_ext(nid);
264 		if (fail)
265 			goto fail;
266 	}
267 	pr_info("allocated %ld bytes of page_ext\n", total_usage);
268 	return;
269 
270 fail:
271 	pr_crit("allocation of page_ext failed.\n");
272 	panic("Out of memory");
273 }
274 
275 #else /* CONFIG_SPARSEMEM */
page_ext_invalid(struct page_ext * page_ext)276 static bool page_ext_invalid(struct page_ext *page_ext)
277 {
278 	return !page_ext || (((unsigned long)page_ext & PAGE_EXT_INVALID) == PAGE_EXT_INVALID);
279 }
280 
lookup_page_ext(const struct page * page)281 static struct page_ext *lookup_page_ext(const struct page *page)
282 {
283 	unsigned long pfn = page_to_pfn(page);
284 	struct mem_section *section = __pfn_to_section(pfn);
285 	struct page_ext *page_ext = READ_ONCE(section->page_ext);
286 
287 	WARN_ON_ONCE(!rcu_read_lock_held());
288 	/*
289 	 * The sanity checks the page allocator does upon freeing a
290 	 * page can reach here before the page_ext arrays are
291 	 * allocated when feeding a range of pages to the allocator
292 	 * for the first time during bootup or memory hotplug.
293 	 */
294 	if (page_ext_invalid(page_ext))
295 		return NULL;
296 	return get_entry(page_ext, pfn);
297 }
298 
alloc_page_ext(size_t size,int nid)299 static void *__meminit alloc_page_ext(size_t size, int nid)
300 {
301 	gfp_t flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN;
302 	void *addr = NULL;
303 
304 	addr = alloc_pages_exact_nid(nid, size, flags);
305 	if (addr) {
306 		kmemleak_alloc(addr, size, 1, flags);
307 		return addr;
308 	}
309 
310 	addr = vzalloc_node(size, nid);
311 
312 	return addr;
313 }
314 
init_section_page_ext(unsigned long pfn,int nid)315 static int __meminit init_section_page_ext(unsigned long pfn, int nid)
316 {
317 	struct mem_section *section;
318 	struct page_ext *base;
319 	unsigned long table_size;
320 
321 	section = __pfn_to_section(pfn);
322 
323 	if (section->page_ext)
324 		return 0;
325 
326 	table_size = page_ext_size * PAGES_PER_SECTION;
327 	base = alloc_page_ext(table_size, nid);
328 
329 	/*
330 	 * The value stored in section->page_ext is (base - pfn)
331 	 * and it does not point to the memory block allocated above,
332 	 * causing kmemleak false positives.
333 	 */
334 	kmemleak_not_leak(base);
335 
336 	if (!base) {
337 		pr_err("page ext allocation failure\n");
338 		return -ENOMEM;
339 	}
340 
341 	/*
342 	 * The passed "pfn" may not be aligned to SECTION.  For the calculation
343 	 * we need to apply a mask.
344 	 */
345 	pfn &= PAGE_SECTION_MASK;
346 	section->page_ext = (void *)base - page_ext_size * pfn;
347 	total_usage += table_size;
348 	return 0;
349 }
350 
free_page_ext(void * addr)351 static void free_page_ext(void *addr)
352 {
353 	if (is_vmalloc_addr(addr)) {
354 		vfree(addr);
355 	} else {
356 		struct page *page = virt_to_page(addr);
357 		size_t table_size;
358 
359 		table_size = page_ext_size * PAGES_PER_SECTION;
360 
361 		BUG_ON(PageReserved(page));
362 		kmemleak_free(addr);
363 		free_pages_exact(addr, table_size);
364 	}
365 }
366 
__free_page_ext(unsigned long pfn)367 static void __free_page_ext(unsigned long pfn)
368 {
369 	struct mem_section *ms;
370 	struct page_ext *base;
371 
372 	ms = __pfn_to_section(pfn);
373 	if (!ms || !ms->page_ext)
374 		return;
375 
376 	base = READ_ONCE(ms->page_ext);
377 	/*
378 	 * page_ext here can be valid while doing the roll back
379 	 * operation in online_page_ext().
380 	 */
381 	if (page_ext_invalid(base))
382 		base = (void *)base - PAGE_EXT_INVALID;
383 	WRITE_ONCE(ms->page_ext, NULL);
384 
385 	base = get_entry(base, pfn);
386 	free_page_ext(base);
387 }
388 
__invalidate_page_ext(unsigned long pfn)389 static void __invalidate_page_ext(unsigned long pfn)
390 {
391 	struct mem_section *ms;
392 	void *val;
393 
394 	ms = __pfn_to_section(pfn);
395 	if (!ms || !ms->page_ext)
396 		return;
397 	val = (void *)ms->page_ext + PAGE_EXT_INVALID;
398 	WRITE_ONCE(ms->page_ext, val);
399 }
400 
online_page_ext(unsigned long start_pfn,unsigned long nr_pages,int nid)401 static int __meminit online_page_ext(unsigned long start_pfn,
402 				unsigned long nr_pages,
403 				int nid)
404 {
405 	unsigned long start, end, pfn;
406 	int fail = 0;
407 
408 	start = SECTION_ALIGN_DOWN(start_pfn);
409 	end = SECTION_ALIGN_UP(start_pfn + nr_pages);
410 
411 	if (nid == NUMA_NO_NODE) {
412 		/*
413 		 * In this case, "nid" already exists and contains valid memory.
414 		 * "start_pfn" passed to us is a pfn which is an arg for
415 		 * online__pages(), and start_pfn should exist.
416 		 */
417 		nid = pfn_to_nid(start_pfn);
418 		VM_BUG_ON(!node_online(nid));
419 	}
420 
421 	for (pfn = start; !fail && pfn < end; pfn += PAGES_PER_SECTION)
422 		fail = init_section_page_ext(pfn, nid);
423 	if (!fail)
424 		return 0;
425 
426 	/* rollback */
427 	for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
428 		__free_page_ext(pfn);
429 
430 	return -ENOMEM;
431 }
432 
offline_page_ext(unsigned long start_pfn,unsigned long nr_pages)433 static int __meminit offline_page_ext(unsigned long start_pfn,
434 				unsigned long nr_pages)
435 {
436 	unsigned long start, end, pfn;
437 
438 	start = SECTION_ALIGN_DOWN(start_pfn);
439 	end = SECTION_ALIGN_UP(start_pfn + nr_pages);
440 
441 	/*
442 	 * Freeing of page_ext is done in 3 steps to avoid
443 	 * use-after-free of it:
444 	 * 1) Traverse all the sections and mark their page_ext
445 	 *    as invalid.
446 	 * 2) Wait for all the existing users of page_ext who
447 	 *    started before invalidation to finish.
448 	 * 3) Free the page_ext.
449 	 */
450 	for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
451 		__invalidate_page_ext(pfn);
452 
453 	synchronize_rcu();
454 
455 	for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION)
456 		__free_page_ext(pfn);
457 	return 0;
458 
459 }
460 
page_ext_callback(struct notifier_block * self,unsigned long action,void * arg)461 static int __meminit page_ext_callback(struct notifier_block *self,
462 			       unsigned long action, void *arg)
463 {
464 	struct memory_notify *mn = arg;
465 	int ret = 0;
466 
467 	switch (action) {
468 	case MEM_GOING_ONLINE:
469 		ret = online_page_ext(mn->start_pfn,
470 				   mn->nr_pages, mn->status_change_nid);
471 		break;
472 	case MEM_OFFLINE:
473 		offline_page_ext(mn->start_pfn,
474 				mn->nr_pages);
475 		break;
476 	case MEM_CANCEL_ONLINE:
477 		offline_page_ext(mn->start_pfn,
478 				mn->nr_pages);
479 		break;
480 	case MEM_GOING_OFFLINE:
481 		break;
482 	case MEM_ONLINE:
483 	case MEM_CANCEL_OFFLINE:
484 		break;
485 	}
486 
487 	return notifier_from_errno(ret);
488 }
489 
page_ext_init(void)490 void __init page_ext_init(void)
491 {
492 	unsigned long pfn;
493 	int nid;
494 
495 	if (!invoke_need_callbacks())
496 		return;
497 
498 	for_each_node_state(nid, N_MEMORY) {
499 		unsigned long start_pfn, end_pfn;
500 
501 		start_pfn = node_start_pfn(nid);
502 		end_pfn = node_end_pfn(nid);
503 		/*
504 		 * start_pfn and end_pfn may not be aligned to SECTION and the
505 		 * page->flags of out of node pages are not initialized.  So we
506 		 * scan [start_pfn, the biggest section's pfn < end_pfn) here.
507 		 */
508 		for (pfn = start_pfn; pfn < end_pfn;
509 			pfn = ALIGN(pfn + 1, PAGES_PER_SECTION)) {
510 
511 			if (!pfn_valid(pfn))
512 				continue;
513 			/*
514 			 * Nodes's pfns can be overlapping.
515 			 * We know some arch can have a nodes layout such as
516 			 * -------------pfn-------------->
517 			 * N0 | N1 | N2 | N0 | N1 | N2|....
518 			 */
519 			if (pfn_to_nid(pfn) != nid)
520 				continue;
521 			if (init_section_page_ext(pfn, nid))
522 				goto oom;
523 			cond_resched();
524 		}
525 	}
526 	hotplug_memory_notifier(page_ext_callback, DEFAULT_CALLBACK_PRI);
527 	pr_info("allocated %ld bytes of page_ext\n", total_usage);
528 	invoke_init_callbacks();
529 	return;
530 
531 oom:
532 	panic("Out of memory");
533 }
534 
pgdat_page_ext_init(struct pglist_data * pgdat)535 void __meminit pgdat_page_ext_init(struct pglist_data *pgdat)
536 {
537 }
538 
539 #endif
540