blob: 248e2ba4ac5900b6d347854ac989a1f9f71343c2 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Dave Hansen3947be12005-10-29 18:16:54 -07002/*
3 * linux/mm/memory_hotplug.c
4 *
5 * Copyright (C)
6 */
7
Dave Hansen3947be12005-10-29 18:16:54 -07008#include <linux/stddef.h>
9#include <linux/mm.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010010#include <linux/sched/signal.h>
Dave Hansen3947be12005-10-29 18:16:54 -070011#include <linux/swap.h>
12#include <linux/interrupt.h>
13#include <linux/pagemap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070014#include <linux/compiler.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040015#include <linux/export.h>
Dave Hansen3947be12005-10-29 18:16:54 -070016#include <linux/pagevec.h>
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -070017#include <linux/writeback.h>
Dave Hansen3947be12005-10-29 18:16:54 -070018#include <linux/slab.h>
19#include <linux/sysctl.h>
20#include <linux/cpu.h>
21#include <linux/memory.h>
Dan Williams4b94ffd2016-01-15 16:56:22 -080022#include <linux/memremap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070023#include <linux/memory_hotplug.h>
24#include <linux/highmem.h>
25#include <linux/vmalloc.h>
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -070026#include <linux/ioport.h>
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -070027#include <linux/delay.h>
28#include <linux/migrate.h>
29#include <linux/page-isolation.h>
Badari Pulavarty71088782008-10-18 20:25:58 -070030#include <linux/pfn.h>
Andi Kleen6ad696d2009-11-17 14:06:22 -080031#include <linux/suspend.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080032#include <linux/mm_inline.h>
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -080033#include <linux/firmware-map.h>
Tang Chen60a5a192013-02-22 16:33:14 -080034#include <linux/stop_machine.h>
Naoya Horiguchic8721bb2013-09-11 14:22:09 -070035#include <linux/hugetlb.h>
Tang Chenc5320922013-11-12 15:08:10 -080036#include <linux/memblock.h>
Vlastimil Babka698b1b32016-03-17 14:18:08 -070037#include <linux/compaction.h>
Michal Hockob15c8722018-12-28 00:38:01 -080038#include <linux/rmap.h>
Dave Hansen3947be12005-10-29 18:16:54 -070039
40#include <asm/tlbflush.h>
41
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030042#include "internal.h"
Dan Williamse900a912019-05-14 15:41:28 -070043#include "shuffle.h"
Adrian Bunk1e5ad9a2008-04-28 20:40:08 +030044
Oscar Salvadore3a9d9f2021-05-04 18:39:48 -070045
46/*
47 * memory_hotplug.memmap_on_memory parameter
48 */
49static bool memmap_on_memory __ro_after_init;
50#ifdef CONFIG_MHP_MEMMAP_ON_MEMORY
51module_param(memmap_on_memory, bool, 0444);
52MODULE_PARM_DESC(memmap_on_memory, "Enable memmap on memory for memory hotplug");
53#endif
Oscar Salvadora08a2ae2021-05-04 18:39:42 -070054
David Hildenbrande83a4372021-09-07 19:55:23 -070055enum {
56 ONLINE_POLICY_CONTIG_ZONES = 0,
57 ONLINE_POLICY_AUTO_MOVABLE,
58};
59
60const char *online_policy_to_str[] = {
61 [ONLINE_POLICY_CONTIG_ZONES] = "contig-zones",
62 [ONLINE_POLICY_AUTO_MOVABLE] = "auto-movable",
63};
64
65static int set_online_policy(const char *val, const struct kernel_param *kp)
66{
67 int ret = sysfs_match_string(online_policy_to_str, val);
68
69 if (ret < 0)
70 return ret;
71 *((int *)kp->arg) = ret;
72 return 0;
73}
74
75static int get_online_policy(char *buffer, const struct kernel_param *kp)
76{
77 return sprintf(buffer, "%s\n", online_policy_to_str[*((int *)kp->arg)]);
78}
79
80/*
81 * memory_hotplug.online_policy: configure online behavior when onlining without
82 * specifying a zone (MMOP_ONLINE)
83 *
84 * "contig-zones": keep zone contiguous
85 * "auto-movable": online memory to ZONE_MOVABLE if the configuration
86 * (auto_movable_ratio, auto_movable_numa_aware) allows for it
87 */
88static int online_policy __read_mostly = ONLINE_POLICY_CONTIG_ZONES;
89static const struct kernel_param_ops online_policy_ops = {
90 .set = set_online_policy,
91 .get = get_online_policy,
92};
93module_param_cb(online_policy, &online_policy_ops, &online_policy, 0644);
94MODULE_PARM_DESC(online_policy,
95 "Set the online policy (\"contig-zones\", \"auto-movable\") "
96 "Default: \"contig-zones\"");
97
98/*
99 * memory_hotplug.auto_movable_ratio: specify maximum MOVABLE:KERNEL ratio
100 *
101 * The ratio represent an upper limit and the kernel might decide to not
102 * online some memory to ZONE_MOVABLE -- e.g., because hotplugged KERNEL memory
103 * doesn't allow for more MOVABLE memory.
104 */
105static unsigned int auto_movable_ratio __read_mostly = 301;
106module_param(auto_movable_ratio, uint, 0644);
107MODULE_PARM_DESC(auto_movable_ratio,
108 "Set the maximum ratio of MOVABLE:KERNEL memory in the system "
109 "in percent for \"auto-movable\" online policy. Default: 301");
110
111/*
112 * memory_hotplug.auto_movable_numa_aware: consider numa node stats
113 */
114#ifdef CONFIG_NUMA
115static bool auto_movable_numa_aware __read_mostly = true;
116module_param(auto_movable_numa_aware, bool, 0644);
117MODULE_PARM_DESC(auto_movable_numa_aware,
118 "Consider numa node stats in addition to global stats in "
119 "\"auto-movable\" online policy. Default: true");
120#endif /* CONFIG_NUMA */
121
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700122/*
123 * online_page_callback contains pointer to current page onlining function.
124 * Initially it is generic_online_page(). If it is required it could be
125 * changed by calling set_online_page_callback() for callback registration
126 * and restore_online_page_callback() for generic callback restore.
127 */
128
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700129static online_page_callback_t online_page_callback = generic_online_page;
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700130static DEFINE_MUTEX(online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700131
Thomas Gleixner3f906ba2017-07-10 15:50:09 -0700132DEFINE_STATIC_PERCPU_RWSEM(mem_hotplug_lock);
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800133
Thomas Gleixner3f906ba2017-07-10 15:50:09 -0700134void get_online_mems(void)
135{
136 percpu_down_read(&mem_hotplug_lock);
137}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700138
Thomas Gleixner3f906ba2017-07-10 15:50:09 -0700139void put_online_mems(void)
140{
141 percpu_up_read(&mem_hotplug_lock);
142}
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700143
Michal Hocko49323812017-07-06 15:41:05 -0700144bool movable_node_enabled = false;
145
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -0700146#ifndef CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
Anshuman Khandual1adf8b42021-02-25 17:17:13 -0800147int mhp_default_online_type = MMOP_OFFLINE;
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -0700148#else
Anshuman Khandual1adf8b42021-02-25 17:17:13 -0800149int mhp_default_online_type = MMOP_ONLINE;
Vitaly Kuznetsov8604d9e2016-05-19 17:13:03 -0700150#endif
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -0700151
Vitaly Kuznetsov86dd9952016-05-19 17:13:06 -0700152static int __init setup_memhp_default_state(char *str)
153{
Anshuman Khandual1adf8b42021-02-25 17:17:13 -0800154 const int online_type = mhp_online_type_from_str(str);
David Hildenbrand5f47adf2020-04-06 20:07:44 -0700155
156 if (online_type >= 0)
Anshuman Khandual1adf8b42021-02-25 17:17:13 -0800157 mhp_default_online_type = online_type;
Vitaly Kuznetsov86dd9952016-05-19 17:13:06 -0700158
159 return 1;
160}
161__setup("memhp_default_state=", setup_memhp_default_state);
162
David Rientjes30467e02015-04-14 15:45:11 -0700163void mem_hotplug_begin(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700164{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -0700165 cpus_read_lock();
166 percpu_down_write(&mem_hotplug_lock);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700167}
168
David Rientjes30467e02015-04-14 15:45:11 -0700169void mem_hotplug_done(void)
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700170{
Thomas Gleixner3f906ba2017-07-10 15:50:09 -0700171 percpu_up_write(&mem_hotplug_lock);
172 cpus_read_unlock();
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700173}
KOSAKI Motohiro20d6c962010-12-02 14:31:19 -0800174
Juergen Gross357b4da2019-02-14 11:42:39 +0100175u64 max_mem_size = U64_MAX;
176
Keith Mannthey45e0b782006-09-30 23:27:09 -0700177/* add this memory to iomem resource */
David Hildenbrand7b7b2722020-06-04 16:48:41 -0700178static struct resource *register_memory_resource(u64 start, u64 size,
179 const char *resource_name)
Keith Mannthey45e0b782006-09-30 23:27:09 -0700180{
Dave Hansen27941292019-02-25 10:57:36 -0800181 struct resource *res;
182 unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
David Hildenbrand7b7b2722020-06-04 16:48:41 -0700183
184 if (strcmp(resource_name, "System RAM"))
David Hildenbrand7cf603d2020-10-15 20:08:33 -0700185 flags |= IORESOURCE_SYSRAM_DRIVER_MANAGED;
Juergen Gross357b4da2019-02-14 11:42:39 +0100186
Anshuman Khandualbca3fea2021-02-25 17:17:33 -0800187 if (!mhp_range_allowed(start, size, true))
188 return ERR_PTR(-E2BIG);
189
Baoquan Hef3cd4c82020-04-06 20:06:50 -0700190 /*
191 * Make sure value parsed from 'mem=' only restricts memory adding
192 * while booting, so that memory hotplug won't be impacted. Please
193 * refer to document of 'mem=' in kernel-parameters.txt for more
194 * details.
195 */
196 if (start + size > max_mem_size && system_state < SYSTEM_RUNNING)
Juergen Gross357b4da2019-02-14 11:42:39 +0100197 return ERR_PTR(-E2BIG);
198
Dave Hansen27941292019-02-25 10:57:36 -0800199 /*
200 * Request ownership of the new memory range. This might be
201 * a child of an existing resource that was present but
202 * not marked as busy.
203 */
204 res = __request_region(&iomem_resource, start, size,
205 resource_name, flags);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700206
Dave Hansen27941292019-02-25 10:57:36 -0800207 if (!res) {
208 pr_debug("Unable to reserve System RAM region: %016llx->%016llx\n",
209 start, start + size);
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -0800210 return ERR_PTR(-EEXIST);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700211 }
212 return res;
213}
214
215static void release_memory_resource(struct resource *res)
216{
217 if (!res)
218 return;
219 release_resource(res);
220 kfree(res);
Keith Mannthey45e0b782006-09-30 23:27:09 -0700221}
222
Keith Mannthey53947022006-09-30 23:27:08 -0700223#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
Dan Williams7ea62162019-07-18 15:58:22 -0700224static int check_pfn_span(unsigned long pfn, unsigned long nr_pages,
225 const char *reason)
226{
227 /*
228 * Disallow all operations smaller than a sub-section and only
229 * allow operations smaller than a section for
230 * SPARSEMEM_VMEMMAP. Note that check_hotplug_memory_range()
231 * enforces a larger memory_block_size_bytes() granularity for
232 * memory that will be marked online, so this check should only
233 * fire for direct arch_{add,remove}_memory() users outside of
234 * add_memory_resource().
235 */
236 unsigned long min_align;
237
238 if (IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
239 min_align = PAGES_PER_SUBSECTION;
240 else
241 min_align = PAGES_PER_SECTION;
242 if (!IS_ALIGNED(pfn, min_align)
243 || !IS_ALIGNED(nr_pages, min_align)) {
244 WARN(1, "Misaligned __%s_pages start: %#lx end: #%lx\n",
245 reason, pfn, pfn + nr_pages - 1);
246 return -EINVAL;
247 }
248 return 0;
249}
250
David Rientjes4edd7ce2013-04-29 15:08:22 -0700251/*
Dan Williams9f605f22021-02-25 17:16:57 -0800252 * Return page for the valid pfn only if the page is online. All pfn
253 * walkers which rely on the fully initialized page->flags and others
254 * should use this rather than pfn_valid && pfn_to_page
255 */
256struct page *pfn_to_online_page(unsigned long pfn)
257{
258 unsigned long nr = pfn_to_section_nr(pfn);
Dan Williams1f90a342021-02-25 17:17:05 -0800259 struct dev_pagemap *pgmap;
Dan Williams9f9b02e2021-02-25 17:17:01 -0800260 struct mem_section *ms;
Dan Williams9f605f22021-02-25 17:16:57 -0800261
Dan Williams9f9b02e2021-02-25 17:17:01 -0800262 if (nr >= NR_MEM_SECTIONS)
263 return NULL;
264
265 ms = __nr_to_section(nr);
266 if (!online_section(ms))
267 return NULL;
268
269 /*
270 * Save some code text when online_section() +
271 * pfn_section_valid() are sufficient.
272 */
273 if (IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) && !pfn_valid(pfn))
274 return NULL;
275
276 if (!pfn_section_valid(ms, pfn))
277 return NULL;
278
Dan Williams1f90a342021-02-25 17:17:05 -0800279 if (!online_device_section(ms))
280 return pfn_to_page(pfn);
281
282 /*
283 * Slowpath: when ZONE_DEVICE collides with
284 * ZONE_{NORMAL,MOVABLE} within the same section some pfns in
285 * the section may be 'offline' but 'valid'. Only
286 * get_dev_pagemap() can determine sub-section online status.
287 */
288 pgmap = get_dev_pagemap(pfn, NULL);
289 put_dev_pagemap(pgmap);
290
291 /* The presence of a pgmap indicates ZONE_DEVICE offline pfn */
292 if (pgmap)
293 return NULL;
294
Dan Williams9f9b02e2021-02-25 17:17:01 -0800295 return pfn_to_page(pfn);
Dan Williams9f605f22021-02-25 17:16:57 -0800296}
297EXPORT_SYMBOL_GPL(pfn_to_online_page);
298
299/*
David Rientjes4edd7ce2013-04-29 15:08:22 -0700300 * Reasonably generic function for adding memory. It is
301 * expected that archs that support memory hotplug will
302 * call this function after deciding the zone to which to
303 * add the new pages.
304 */
Dan Williams7ea62162019-07-18 15:58:22 -0700305int __ref __add_pages(int nid, unsigned long pfn, unsigned long nr_pages,
Logan Gunthorpef5637d32020-04-10 14:33:21 -0700306 struct mhp_params *params)
David Rientjes4edd7ce2013-04-29 15:08:22 -0700307{
David Hildenbrand6cdd0b32020-04-06 20:06:56 -0700308 const unsigned long end_pfn = pfn + nr_pages;
309 unsigned long cur_nr_pages;
Dan Williams9a845032019-07-18 15:58:43 -0700310 int err;
Logan Gunthorpef5637d32020-04-10 14:33:21 -0700311 struct vmem_altmap *altmap = params->altmap;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800312
Logan Gunthorpebfeb0222020-04-10 14:33:36 -0700313 if (WARN_ON_ONCE(!params->pgprot.pgprot))
314 return -EINVAL;
315
Anshuman Khandualbca3fea2021-02-25 17:17:33 -0800316 VM_BUG_ON(!mhp_range_allowed(PFN_PHYS(pfn), nr_pages * PAGE_SIZE, false));
Alastair D'Silvadca44362019-11-30 17:53:48 -0800317
Dan Williams4b94ffd2016-01-15 16:56:22 -0800318 if (altmap) {
319 /*
320 * Validate altmap is within bounds of the total request
321 */
Dan Williams7ea62162019-07-18 15:58:22 -0700322 if (altmap->base_pfn != pfn
Dan Williams4b94ffd2016-01-15 16:56:22 -0800323 || vmem_altmap_offset(altmap) > nr_pages) {
324 pr_warn_once("memory add fail, invalid altmap\n");
Dan Williams7ea62162019-07-18 15:58:22 -0700325 return -EINVAL;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800326 }
327 altmap->alloc = 0;
328 }
329
Dan Williams7ea62162019-07-18 15:58:22 -0700330 err = check_pfn_span(pfn, nr_pages, "add");
331 if (err)
332 return err;
333
David Hildenbrand6cdd0b32020-04-06 20:06:56 -0700334 for (; pfn < end_pfn; pfn += cur_nr_pages) {
335 /* Select all remaining pages up to the next section boundary */
336 cur_nr_pages = min(end_pfn - pfn,
337 SECTION_ALIGN_UP(pfn + 1) - pfn);
338 err = sparse_add_section(nid, pfn, cur_nr_pages, altmap);
Dan Williamsba72b4c2019-07-18 15:58:26 -0700339 if (err)
340 break;
Michal Hockof64ac5e2017-10-03 16:16:16 -0700341 cond_resched();
David Rientjes4edd7ce2013-04-29 15:08:22 -0700342 }
Zhu Guihuac435a392015-06-24 16:58:42 -0700343 vmemmap_populate_print_last();
David Rientjes4edd7ce2013-04-29 15:08:22 -0700344 return err;
345}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700346
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800347/* find the smallest valid pfn in the range [start_pfn, end_pfn) */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700348static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800349 unsigned long start_pfn,
350 unsigned long end_pfn)
351{
Dan Williams49ba3c62019-07-18 15:58:07 -0700352 for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SUBSECTION) {
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800353 if (unlikely(!pfn_to_online_page(start_pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800354 continue;
355
356 if (unlikely(pfn_to_nid(start_pfn) != nid))
357 continue;
358
David Hildenbrand9b051582020-02-03 17:34:12 -0800359 if (zone != page_zone(pfn_to_page(start_pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800360 continue;
361
362 return start_pfn;
363 }
364
365 return 0;
366}
367
368/* find the biggest valid pfn in the range [start_pfn, end_pfn). */
YASUAKI ISHIMATSUd09b0132017-10-03 16:16:32 -0700369static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800370 unsigned long start_pfn,
371 unsigned long end_pfn)
372{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800373 unsigned long pfn;
374
375 /* pfn is the end pfn of a memory section. */
376 pfn = end_pfn - 1;
Dan Williams49ba3c62019-07-18 15:58:07 -0700377 for (; pfn >= start_pfn; pfn -= PAGES_PER_SUBSECTION) {
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800378 if (unlikely(!pfn_to_online_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800379 continue;
380
381 if (unlikely(pfn_to_nid(pfn) != nid))
382 continue;
383
David Hildenbrand9b051582020-02-03 17:34:12 -0800384 if (zone != page_zone(pfn_to_page(pfn)))
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800385 continue;
386
387 return pfn;
388 }
389
390 return 0;
391}
392
393static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
394 unsigned long end_pfn)
395{
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800396 unsigned long pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800397 int nid = zone_to_nid(zone);
398
David Hildenbrand5d120712020-02-03 17:34:19 -0800399 if (zone->zone_start_pfn == start_pfn) {
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800400 /*
401 * If the section is smallest section in the zone, it need
402 * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
403 * In this case, we find second smallest valid mem_section
404 * for shrinking zone.
405 */
406 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
David Hildenbrand5d120712020-02-03 17:34:19 -0800407 zone_end_pfn(zone));
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800408 if (pfn) {
David Hildenbrand5d120712020-02-03 17:34:19 -0800409 zone->spanned_pages = zone_end_pfn(zone) - pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800410 zone->zone_start_pfn = pfn;
David Hildenbrand950b68d2020-02-03 17:34:16 -0800411 } else {
412 zone->zone_start_pfn = 0;
413 zone->spanned_pages = 0;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800414 }
David Hildenbrand5d120712020-02-03 17:34:19 -0800415 } else if (zone_end_pfn(zone) == end_pfn) {
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800416 /*
417 * If the section is biggest section in the zone, it need
418 * shrink zone->spanned_pages.
419 * In this case, we find second biggest valid mem_section for
420 * shrinking zone.
421 */
David Hildenbrand5d120712020-02-03 17:34:19 -0800422 pfn = find_biggest_section_pfn(nid, zone, zone->zone_start_pfn,
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800423 start_pfn);
424 if (pfn)
David Hildenbrand5d120712020-02-03 17:34:19 -0800425 zone->spanned_pages = pfn - zone->zone_start_pfn + 1;
David Hildenbrand950b68d2020-02-03 17:34:16 -0800426 else {
427 zone->zone_start_pfn = 0;
428 zone->spanned_pages = 0;
429 }
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800430 }
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800431}
432
David Hildenbrand00d6c012019-10-18 20:19:33 -0700433static void update_pgdat_span(struct pglist_data *pgdat)
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800434{
David Hildenbrand00d6c012019-10-18 20:19:33 -0700435 unsigned long node_start_pfn = 0, node_end_pfn = 0;
436 struct zone *zone;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800437
David Hildenbrand00d6c012019-10-18 20:19:33 -0700438 for (zone = pgdat->node_zones;
439 zone < pgdat->node_zones + MAX_NR_ZONES; zone++) {
Miaohe Lin6c922cf2021-02-25 17:17:21 -0800440 unsigned long end_pfn = zone_end_pfn(zone);
David Hildenbrand00d6c012019-10-18 20:19:33 -0700441
442 /* No need to lock the zones, they can't change. */
David Hildenbrand656d5712019-11-05 21:17:10 -0800443 if (!zone->spanned_pages)
444 continue;
445 if (!node_end_pfn) {
446 node_start_pfn = zone->zone_start_pfn;
Miaohe Lin6c922cf2021-02-25 17:17:21 -0800447 node_end_pfn = end_pfn;
David Hildenbrand656d5712019-11-05 21:17:10 -0800448 continue;
449 }
450
Miaohe Lin6c922cf2021-02-25 17:17:21 -0800451 if (end_pfn > node_end_pfn)
452 node_end_pfn = end_pfn;
David Hildenbrand00d6c012019-10-18 20:19:33 -0700453 if (zone->zone_start_pfn < node_start_pfn)
454 node_start_pfn = zone->zone_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800455 }
456
David Hildenbrand00d6c012019-10-18 20:19:33 -0700457 pgdat->node_start_pfn = node_start_pfn;
458 pgdat->node_spanned_pages = node_end_pfn - node_start_pfn;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800459}
460
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800461void __ref remove_pfn_range_from_zone(struct zone *zone,
462 unsigned long start_pfn,
463 unsigned long nr_pages)
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800464{
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700465 const unsigned long end_pfn = start_pfn + nr_pages;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800466 struct pglist_data *pgdat = zone->zone_pgdat;
Oscar Salvador27cacaa2021-06-30 18:52:46 -0700467 unsigned long pfn, cur_nr_pages;
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800468
David Hildenbrandd33695b2020-02-03 17:34:09 -0800469 /* Poison struct pages because they are now uninitialized again. */
Ben Widawskyb7e3deb2020-06-25 20:30:51 -0700470 for (pfn = start_pfn; pfn < end_pfn; pfn += cur_nr_pages) {
471 cond_resched();
472
473 /* Select all remaining pages up to the next section boundary */
474 cur_nr_pages =
475 min(end_pfn - pfn, SECTION_ALIGN_UP(pfn + 1) - pfn);
476 page_init_poison(pfn_to_page(pfn),
477 sizeof(struct page) * cur_nr_pages);
478 }
David Hildenbrandd33695b2020-02-03 17:34:09 -0800479
David Hildenbrand7ce700b2019-11-21 17:53:56 -0800480#ifdef CONFIG_ZONE_DEVICE
481 /*
482 * Zone shrinking code cannot properly deal with ZONE_DEVICE. So
483 * we will not try to shrink the zones - which is okay as
484 * set_zone_contiguous() cannot deal with ZONE_DEVICE either way.
485 */
486 if (zone_idx(zone) == ZONE_DEVICE)
487 return;
488#endif
489
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800490 clear_zone_contiguous(zone);
491
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800492 shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
David Hildenbrand00d6c012019-10-18 20:19:33 -0700493 update_pgdat_span(pgdat);
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800494
495 set_zone_contiguous(zone);
Yasuaki Ishimatsu815121d2013-02-22 16:33:12 -0800496}
497
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800498static void __remove_section(unsigned long pfn, unsigned long nr_pages,
499 unsigned long map_offset,
500 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700501{
chenqiwu10404902020-04-06 20:07:48 -0700502 struct mem_section *ms = __pfn_to_section(pfn);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700503
David Hildenbrand9d1d8872019-05-13 17:21:41 -0700504 if (WARN_ON_ONCE(!valid_section(ms)))
505 return;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700506
Dan Williamsba72b4c2019-07-18 15:58:26 -0700507 sparse_remove_section(ms, pfn, nr_pages, map_offset, altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700508}
509
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700510/**
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800511 * __remove_pages() - remove sections of pages
Dan Williams7ea62162019-07-18 15:58:22 -0700512 * @pfn: starting pageframe (must be aligned to start of a section)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700513 * @nr_pages: number of pages to remove (must be multiple of section size)
Mike Rapoporte8b098f2018-04-05 16:24:57 -0700514 * @altmap: alternative device page map or %NULL if default memmap is used
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700515 *
516 * Generic helper function to remove section mappings and sysfs entries
517 * for the section of the memory we are removing. Caller needs to make
518 * sure that pages are marked reserved and zones are adjust properly by
519 * calling offline_pages().
520 */
David Hildenbrandfeee6b22020-01-04 12:59:33 -0800521void __remove_pages(unsigned long pfn, unsigned long nr_pages,
522 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700523{
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800524 const unsigned long end_pfn = pfn + nr_pages;
525 unsigned long cur_nr_pages;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800526 unsigned long map_offset = 0;
Dan Williams4b94ffd2016-01-15 16:56:22 -0800527
Dan Williams96da4352019-07-18 15:58:15 -0700528 map_offset = vmem_altmap_offset(altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700529
Dan Williams7ea62162019-07-18 15:58:22 -0700530 if (check_pfn_span(pfn, nr_pages, "remove"))
531 return;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700532
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800533 for (; pfn < end_pfn; pfn += cur_nr_pages) {
Michal Hockodd33ad72018-11-02 15:48:46 -0700534 cond_resched();
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800535 /* Select all remaining pages up to the next section boundary */
David Hildenbranda11b9412020-04-06 20:06:53 -0700536 cur_nr_pages = min(end_pfn - pfn,
537 SECTION_ALIGN_UP(pfn + 1) - pfn);
David Hildenbrand52fb87c2020-02-03 17:34:23 -0800538 __remove_section(pfn, cur_nr_pages, map_offset, altmap);
Dan Williams4b94ffd2016-01-15 16:56:22 -0800539 map_offset = 0;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700540 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700541}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700542
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700543int set_online_page_callback(online_page_callback_t callback)
544{
545 int rc = -EINVAL;
546
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700547 get_online_mems();
548 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700549
550 if (online_page_callback == generic_online_page) {
551 online_page_callback = callback;
552 rc = 0;
553 }
554
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700555 mutex_unlock(&online_page_callback_lock);
556 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700557
558 return rc;
559}
560EXPORT_SYMBOL_GPL(set_online_page_callback);
561
562int restore_online_page_callback(online_page_callback_t callback)
563{
564 int rc = -EINVAL;
565
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700566 get_online_mems();
567 mutex_lock(&online_page_callback_lock);
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700568
569 if (online_page_callback == callback) {
570 online_page_callback = generic_online_page;
571 rc = 0;
572 }
573
Vladimir Davydovbfc8c902014-06-04 16:07:18 -0700574 mutex_unlock(&online_page_callback_lock);
575 put_online_mems();
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700576
577 return rc;
578}
579EXPORT_SYMBOL_GPL(restore_online_page_callback);
580
David Hildenbrand18db1492019-11-30 17:53:51 -0800581void generic_online_page(struct page *page, unsigned int order)
Daniel Kiper9d0ad8c2011-07-25 17:12:05 -0700582{
Vlastimil Babkac87cbc12020-03-05 22:28:42 -0800583 /*
584 * Freeing the page with debug_pagealloc enabled will try to unmap it,
585 * so we should map it first. This is better than introducing a special
586 * case in page freeing fast path.
587 */
Mike Rapoport77bc7fd2020-12-14 19:10:20 -0800588 debug_pagealloc_map_pages(page, 1 << order);
Arun KSa9cd4102019-03-05 15:42:14 -0800589 __free_pages_core(page, order);
590 totalram_pages_add(1UL << order);
591#ifdef CONFIG_HIGHMEM
592 if (PageHighMem(page))
593 totalhigh_pages_add(1UL << order);
594#endif
595}
David Hildenbrand18db1492019-11-30 17:53:51 -0800596EXPORT_SYMBOL_GPL(generic_online_page);
Arun KSa9cd4102019-03-05 15:42:14 -0800597
David Hildenbrandaac65322020-10-15 20:08:11 -0700598static void online_pages_range(unsigned long start_pfn, unsigned long nr_pages)
Dave Hansen3947be12005-10-29 18:16:54 -0700599{
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700600 const unsigned long end_pfn = start_pfn + nr_pages;
601 unsigned long pfn;
Michal Hocko2d070ea2017-07-06 15:37:56 -0700602
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700603 /*
David Hildenbrandaac65322020-10-15 20:08:11 -0700604 * Online the pages in MAX_ORDER - 1 aligned chunks. The callback might
605 * decide to not expose all pages to the buddy (e.g., expose them
606 * later). We account all pages as being online and belonging to this
607 * zone ("present").
Oscar Salvadora08a2ae2021-05-04 18:39:42 -0700608 * When using memmap_on_memory, the range might not be aligned to
609 * MAX_ORDER_NR_PAGES - 1, but pageblock aligned. __ffs() will detect
610 * this and the first chunk to online will be pageblock_nr_pages.
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700611 */
Oscar Salvadora08a2ae2021-05-04 18:39:42 -0700612 for (pfn = start_pfn; pfn < end_pfn;) {
613 int order = min(MAX_ORDER - 1UL, __ffs(pfn));
614
615 (*online_page_callback)(pfn_to_page(pfn), order);
616 pfn += (1UL << order);
617 }
Michal Hocko2d070ea2017-07-06 15:37:56 -0700618
David Hildenbrandb2c2ab22019-09-23 15:36:02 -0700619 /* mark all involved sections as online */
620 online_mem_sections(start_pfn, end_pfn);
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -0700621}
622
Lai Jiangshand9713672012-12-11 16:01:03 -0800623/* check which state of node_states will be changed when online memory */
624static void node_states_check_changes_online(unsigned long nr_pages,
625 struct zone *zone, struct memory_notify *arg)
626{
627 int nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -0800628
Anshuman Khandual98fa15f2019-03-05 15:42:58 -0800629 arg->status_change_nid = NUMA_NO_NODE;
630 arg->status_change_nid_normal = NUMA_NO_NODE;
631 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -0800632
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800633 if (!node_state(nid, N_MEMORY))
Lai Jiangshand9713672012-12-11 16:01:03 -0800634 arg->status_change_nid = nid;
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700635 if (zone_idx(zone) <= ZONE_NORMAL && !node_state(nid, N_NORMAL_MEMORY))
636 arg->status_change_nid_normal = nid;
637#ifdef CONFIG_HIGHMEM
Baoquan Hed3ba3ae2019-05-13 17:17:35 -0700638 if (zone_idx(zone) <= ZONE_HIGHMEM && !node_state(nid, N_HIGH_MEMORY))
Oscar Salvador8efe33f2018-10-26 15:07:34 -0700639 arg->status_change_nid_high = nid;
640#endif
Lai Jiangshand9713672012-12-11 16:01:03 -0800641}
642
643static void node_states_set_node(int node, struct memory_notify *arg)
644{
645 if (arg->status_change_nid_normal >= 0)
646 node_set_state(node, N_NORMAL_MEMORY);
647
Lai Jiangshan6715ddf2012-12-12 13:51:49 -0800648 if (arg->status_change_nid_high >= 0)
649 node_set_state(node, N_HIGH_MEMORY);
650
Oscar Salvador83d83612018-10-26 15:07:25 -0700651 if (arg->status_change_nid >= 0)
652 node_set_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -0800653}
654
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700655static void __meminit resize_zone_range(struct zone *zone, unsigned long start_pfn,
656 unsigned long nr_pages)
657{
658 unsigned long old_end_pfn = zone_end_pfn(zone);
659
660 if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
661 zone->zone_start_pfn = start_pfn;
662
663 zone->spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - zone->zone_start_pfn;
664}
665
666static void __meminit resize_pgdat_range(struct pglist_data *pgdat, unsigned long start_pfn,
667 unsigned long nr_pages)
668{
669 unsigned long old_end_pfn = pgdat_end_pfn(pgdat);
670
671 if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
672 pgdat->node_start_pfn = start_pfn;
673
674 pgdat->node_spanned_pages = max(start_pfn + nr_pages, old_end_pfn) - pgdat->node_start_pfn;
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700675
David Hildenbrand3fccb742019-09-23 15:35:37 -0700676}
Dan Williams1f90a342021-02-25 17:17:05 -0800677
678static void section_taint_zone_device(unsigned long pfn)
679{
680 struct mem_section *ms = __pfn_to_section(pfn);
681
682 ms->section_mem_map |= SECTION_TAINT_ZONE_DEVICE;
683}
684
David Hildenbrand3fccb742019-09-23 15:35:37 -0700685/*
686 * Associate the pfn range with the given zone, initializing the memmaps
687 * and resizing the pgdat/zone data to span the added pages. After this
688 * call, all affected pages are PG_reserved.
David Hildenbrandd882c002020-10-15 20:08:19 -0700689 *
690 * All aligned pageblocks are initialized to the specified migratetype
691 * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
692 * zone stats (e.g., nr_isolate_pageblock) are touched.
David Hildenbrand3fccb742019-09-23 15:35:37 -0700693 */
Christoph Hellwiga99583e2017-12-29 08:53:57 +0100694void __ref move_pfn_range_to_zone(struct zone *zone, unsigned long start_pfn,
David Hildenbrandd882c002020-10-15 20:08:19 -0700695 unsigned long nr_pages,
696 struct vmem_altmap *altmap, int migratetype)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700697{
698 struct pglist_data *pgdat = zone->zone_pgdat;
699 int nid = pgdat->node_id;
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700700
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700701 clear_zone_contiguous(zone);
702
Wei Yangfa004ab2018-12-28 00:37:10 -0800703 if (zone_is_empty(zone))
704 init_currently_empty_zone(zone, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700705 resize_zone_range(zone, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700706 resize_pgdat_range(pgdat, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700707
708 /*
Dan Williams1f90a342021-02-25 17:17:05 -0800709 * Subsection population requires care in pfn_to_online_page().
710 * Set the taint to enable the slow path detection of
711 * ZONE_DEVICE pages in an otherwise ZONE_{NORMAL,MOVABLE}
712 * section.
713 */
714 if (zone_is_zone_device(zone)) {
715 if (!IS_ALIGNED(start_pfn, PAGES_PER_SECTION))
716 section_taint_zone_device(start_pfn);
717 if (!IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION))
718 section_taint_zone_device(start_pfn + nr_pages);
719 }
720
721 /*
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700722 * TODO now we have a visible range of pages which are not associated
723 * with their zone properly. Not nice but set_pfnblock_flags_mask
724 * expects the zone spans the pfn range. All the pages in the range
725 * are reserved so nobody should be touching them so we should be safe
726 */
Baoquan Heab28cb62021-02-24 12:06:14 -0800727 memmap_init_range(nr_pages, nid, zone_idx(zone), start_pfn, 0,
David Hildenbrandd882c002020-10-15 20:08:19 -0700728 MEMINIT_HOTPLUG, altmap, migratetype);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700729
730 set_zone_contiguous(zone);
731}
732
David Hildenbrande83a4372021-09-07 19:55:23 -0700733struct auto_movable_stats {
734 unsigned long kernel_early_pages;
735 unsigned long movable_pages;
736};
737
738static void auto_movable_stats_account_zone(struct auto_movable_stats *stats,
739 struct zone *zone)
740{
741 if (zone_idx(zone) == ZONE_MOVABLE) {
742 stats->movable_pages += zone->present_pages;
743 } else {
744 stats->kernel_early_pages += zone->present_early_pages;
745#ifdef CONFIG_CMA
746 /*
747 * CMA pages (never on hotplugged memory) behave like
748 * ZONE_MOVABLE.
749 */
750 stats->movable_pages += zone->cma_pages;
751 stats->kernel_early_pages -= zone->cma_pages;
752#endif /* CONFIG_CMA */
753 }
754}
755
756static bool auto_movable_can_online_movable(int nid, unsigned long nr_pages)
757{
758 struct auto_movable_stats stats = {};
759 unsigned long kernel_early_pages, movable_pages;
760 pg_data_t *pgdat = NODE_DATA(nid);
761 struct zone *zone;
762 int i;
763
764 /* Walk all relevant zones and collect MOVABLE vs. KERNEL stats. */
765 if (nid == NUMA_NO_NODE) {
766 /* TODO: cache values */
767 for_each_populated_zone(zone)
768 auto_movable_stats_account_zone(&stats, zone);
769 } else {
770 for (i = 0; i < MAX_NR_ZONES; i++) {
771 zone = pgdat->node_zones + i;
772 if (populated_zone(zone))
773 auto_movable_stats_account_zone(&stats, zone);
774 }
775 }
776
777 kernel_early_pages = stats.kernel_early_pages;
778 movable_pages = stats.movable_pages;
779
780 /*
781 * Test if we could online the given number of pages to ZONE_MOVABLE
782 * and still stay in the configured ratio.
783 */
784 movable_pages += nr_pages;
785 return movable_pages <= (auto_movable_ratio * kernel_early_pages) / 100;
786}
787
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700788/*
Michal Hockoc246a212017-07-06 15:38:18 -0700789 * Returns a default kernel memory zone for the given pfn range.
790 * If no kernel zone covers this pfn range it will automatically go
791 * to the ZONE_NORMAL.
792 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700793static struct zone *default_kernel_zone_for_pfn(int nid, unsigned long start_pfn,
Michal Hockoc246a212017-07-06 15:38:18 -0700794 unsigned long nr_pages)
795{
796 struct pglist_data *pgdat = NODE_DATA(nid);
797 int zid;
798
799 for (zid = 0; zid <= ZONE_NORMAL; zid++) {
800 struct zone *zone = &pgdat->node_zones[zid];
801
802 if (zone_intersects(zone, start_pfn, nr_pages))
803 return zone;
804 }
805
806 return &pgdat->node_zones[ZONE_NORMAL];
807}
808
David Hildenbrande83a4372021-09-07 19:55:23 -0700809/*
810 * Determine to which zone to online memory dynamically based on user
811 * configuration and system stats. We care about the following ratio:
812 *
813 * MOVABLE : KERNEL
814 *
815 * Whereby MOVABLE is memory in ZONE_MOVABLE and KERNEL is memory in
816 * one of the kernel zones. CMA pages inside one of the kernel zones really
817 * behaves like ZONE_MOVABLE, so we treat them accordingly.
818 *
819 * We don't allow for hotplugged memory in a KERNEL zone to increase the
820 * amount of MOVABLE memory we can have, so we end up with:
821 *
822 * MOVABLE : KERNEL_EARLY
823 *
824 * Whereby KERNEL_EARLY is memory in one of the kernel zones, available sinze
825 * boot. We base our calculation on KERNEL_EARLY internally, because:
826 *
827 * a) Hotplugged memory in one of the kernel zones can sometimes still get
828 * hotunplugged, especially when hot(un)plugging individual memory blocks.
829 * There is no coordination across memory devices, therefore "automatic"
830 * hotunplugging, as implemented in hypervisors, could result in zone
831 * imbalances.
832 * b) Early/boot memory in one of the kernel zones can usually not get
833 * hotunplugged again (e.g., no firmware interface to unplug, fragmented
834 * with unmovable allocations). While there are corner cases where it might
835 * still work, it is barely relevant in practice.
836 *
837 * We rely on "present pages" instead of "managed pages", as the latter is
838 * highly unreliable and dynamic in virtualized environments, and does not
839 * consider boot time allocations. For example, memory ballooning adjusts the
840 * managed pages when inflating/deflating the balloon, and balloon compaction
841 * can even migrate inflated pages between zones.
842 *
843 * Using "present pages" is better but some things to keep in mind are:
844 *
845 * a) Some memblock allocations, such as for the crashkernel area, are
846 * effectively unused by the kernel, yet they account to "present pages".
847 * Fortunately, these allocations are comparatively small in relevant setups
848 * (e.g., fraction of system memory).
849 * b) Some hotplugged memory blocks in virtualized environments, esecially
850 * hotplugged by virtio-mem, look like they are completely present, however,
851 * only parts of the memory block are actually currently usable.
852 * "present pages" is an upper limit that can get reached at runtime. As
853 * we base our calculations on KERNEL_EARLY, this is not an issue.
854 */
David Hildenbrand445fcf72021-09-07 19:55:45 -0700855static struct zone *auto_movable_zone_for_pfn(int nid,
856 struct memory_group *group,
857 unsigned long pfn,
David Hildenbrande83a4372021-09-07 19:55:23 -0700858 unsigned long nr_pages)
859{
David Hildenbrand445fcf72021-09-07 19:55:45 -0700860 unsigned long online_pages = 0, max_pages, end_pfn;
861 struct page *page;
862
David Hildenbrande83a4372021-09-07 19:55:23 -0700863 if (!auto_movable_ratio)
864 goto kernel_zone;
865
David Hildenbrand445fcf72021-09-07 19:55:45 -0700866 if (group && !group->is_dynamic) {
867 max_pages = group->s.max_pages;
868 online_pages = group->present_movable_pages;
869
870 /* If anything is !MOVABLE online the rest !MOVABLE. */
871 if (group->present_kernel_pages)
872 goto kernel_zone;
873 } else if (!group || group->d.unit_pages == nr_pages) {
874 max_pages = nr_pages;
875 } else {
876 max_pages = group->d.unit_pages;
877 /*
878 * Take a look at all online sections in the current unit.
879 * We can safely assume that all pages within a section belong
880 * to the same zone, because dynamic memory groups only deal
881 * with hotplugged memory.
882 */
883 pfn = ALIGN_DOWN(pfn, group->d.unit_pages);
884 end_pfn = pfn + group->d.unit_pages;
885 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
886 page = pfn_to_online_page(pfn);
887 if (!page)
888 continue;
889 /* If anything is !MOVABLE online the rest !MOVABLE. */
890 if (page_zonenum(page) != ZONE_MOVABLE)
891 goto kernel_zone;
892 online_pages += PAGES_PER_SECTION;
893 }
894 }
895
896 /*
897 * Online MOVABLE if we could *currently* online all remaining parts
898 * MOVABLE. We expect to (add+) online them immediately next, so if
899 * nobody interferes, all will be MOVABLE if possible.
900 */
901 nr_pages = max_pages - online_pages;
David Hildenbrande83a4372021-09-07 19:55:23 -0700902 if (!auto_movable_can_online_movable(NUMA_NO_NODE, nr_pages))
903 goto kernel_zone;
904
905#ifdef CONFIG_NUMA
906 if (auto_movable_numa_aware &&
907 !auto_movable_can_online_movable(nid, nr_pages))
908 goto kernel_zone;
909#endif /* CONFIG_NUMA */
910
911 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
912kernel_zone:
913 return default_kernel_zone_for_pfn(nid, pfn, nr_pages);
914}
915
Michal Hockoc6f03e22017-09-06 16:19:40 -0700916static inline struct zone *default_zone_for_pfn(int nid, unsigned long start_pfn,
917 unsigned long nr_pages)
Michal Hockoe5e68932017-09-06 16:19:37 -0700918{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700919 struct zone *kernel_zone = default_kernel_zone_for_pfn(nid, start_pfn,
920 nr_pages);
921 struct zone *movable_zone = &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
922 bool in_kernel = zone_intersects(kernel_zone, start_pfn, nr_pages);
923 bool in_movable = zone_intersects(movable_zone, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700924
925 /*
Michal Hockoc6f03e22017-09-06 16:19:40 -0700926 * We inherit the existing zone in a simple case where zones do not
927 * overlap in the given range
Michal Hockoe5e68932017-09-06 16:19:37 -0700928 */
Michal Hockoc6f03e22017-09-06 16:19:40 -0700929 if (in_kernel ^ in_movable)
930 return (in_kernel) ? kernel_zone : movable_zone;
Michal Hockoe5e68932017-09-06 16:19:37 -0700931
Michal Hockoc6f03e22017-09-06 16:19:40 -0700932 /*
933 * If the range doesn't belong to any zone or two zones overlap in the
934 * given range then we use movable zone only if movable_node is
935 * enabled because we always online to a kernel zone by default.
936 */
937 return movable_node_enabled ? movable_zone : kernel_zone;
Michal Hocko9f123ab2017-07-10 15:48:37 -0700938}
939
David Hildenbrand7cf209b2021-09-07 19:54:59 -0700940struct zone *zone_for_pfn_range(int online_type, int nid,
David Hildenbrand445fcf72021-09-07 19:55:45 -0700941 struct memory_group *group, unsigned long start_pfn,
942 unsigned long nr_pages)
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700943{
Michal Hockoc6f03e22017-09-06 16:19:40 -0700944 if (online_type == MMOP_ONLINE_KERNEL)
945 return default_kernel_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockof1dd2cd2017-07-06 15:38:11 -0700946
Michal Hockoc6f03e22017-09-06 16:19:40 -0700947 if (online_type == MMOP_ONLINE_MOVABLE)
948 return &NODE_DATA(nid)->node_zones[ZONE_MOVABLE];
Reza Arbabdf429ac2016-07-26 15:22:23 -0700949
David Hildenbrande83a4372021-09-07 19:55:23 -0700950 if (online_policy == ONLINE_POLICY_AUTO_MOVABLE)
David Hildenbrand445fcf72021-09-07 19:55:45 -0700951 return auto_movable_zone_for_pfn(nid, group, start_pfn, nr_pages);
David Hildenbrande83a4372021-09-07 19:55:23 -0700952
Michal Hockoc6f03e22017-09-06 16:19:40 -0700953 return default_zone_for_pfn(nid, start_pfn, nr_pages);
Michal Hockoe5e68932017-09-06 16:19:37 -0700954}
955
Oscar Salvadora08a2ae2021-05-04 18:39:42 -0700956/*
957 * This function should only be called by memory_block_{online,offline},
958 * and {online,offline}_pages.
959 */
David Hildenbrand836809e2021-09-07 19:55:30 -0700960void adjust_present_page_count(struct page *page, struct memory_group *group,
961 long nr_pages)
David Hildenbrandf9901142021-05-04 18:39:39 -0700962{
David Hildenbrand4b097002021-09-07 19:55:19 -0700963 struct zone *zone = page_zone(page);
David Hildenbrand836809e2021-09-07 19:55:30 -0700964 const bool movable = zone_idx(zone) == ZONE_MOVABLE;
David Hildenbrand4b097002021-09-07 19:55:19 -0700965
966 /*
967 * We only support onlining/offlining/adding/removing of complete
968 * memory blocks; therefore, either all is either early or hotplugged.
969 */
970 if (early_section(__pfn_to_section(page_to_pfn(page))))
971 zone->present_early_pages += nr_pages;
David Hildenbrandf9901142021-05-04 18:39:39 -0700972 zone->present_pages += nr_pages;
David Hildenbrandf9901142021-05-04 18:39:39 -0700973 zone->zone_pgdat->node_present_pages += nr_pages;
David Hildenbrand836809e2021-09-07 19:55:30 -0700974
975 if (group && movable)
976 group->present_movable_pages += nr_pages;
977 else if (group && !movable)
978 group->present_kernel_pages += nr_pages;
David Hildenbrandf9901142021-05-04 18:39:39 -0700979}
980
Oscar Salvadora08a2ae2021-05-04 18:39:42 -0700981int mhp_init_memmap_on_memory(unsigned long pfn, unsigned long nr_pages,
982 struct zone *zone)
983{
984 unsigned long end_pfn = pfn + nr_pages;
985 int ret;
986
987 ret = kasan_add_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
988 if (ret)
989 return ret;
990
991 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_UNMOVABLE);
992
993 /*
994 * It might be that the vmemmap_pages fully span sections. If that is
995 * the case, mark those sections online here as otherwise they will be
996 * left offline.
997 */
998 if (nr_pages >= PAGES_PER_SECTION)
999 online_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
1000
1001 return ret;
1002}
1003
1004void mhp_deinit_memmap_on_memory(unsigned long pfn, unsigned long nr_pages)
1005{
1006 unsigned long end_pfn = pfn + nr_pages;
1007
1008 /*
1009 * It might be that the vmemmap_pages fully span sections. If that is
1010 * the case, mark those sections offline here as otherwise they will be
1011 * left online.
1012 */
1013 if (nr_pages >= PAGES_PER_SECTION)
1014 offline_mem_sections(pfn, ALIGN_DOWN(end_pfn, PAGES_PER_SECTION));
1015
1016 /*
1017 * The pages associated with this vmemmap have been offlined, so
1018 * we can reset its state here.
1019 */
1020 remove_pfn_range_from_zone(page_zone(pfn_to_page(pfn)), pfn, nr_pages);
1021 kasan_remove_zero_shadow(__va(PFN_PHYS(pfn)), PFN_PHYS(nr_pages));
1022}
1023
David Hildenbrand836809e2021-09-07 19:55:30 -07001024int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
1025 struct zone *zone, struct memory_group *group)
KAMEZAWA Hiroyuki75884fb2007-10-16 01:26:10 -07001026{
Cody P Schaferaa472282013-07-03 15:02:10 -07001027 unsigned long flags;
Yasunori Goto68113782006-06-23 02:03:11 -07001028 int need_zonelists_rebuild = 0;
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07001029 const int nid = zone_to_nid(zone);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001030 int ret;
1031 struct memory_notify arg;
Dave Hansen3947be12005-10-29 18:16:54 -07001032
Oscar Salvadordd8e2f22021-05-04 18:39:36 -07001033 /*
1034 * {on,off}lining is constrained to full memory sections (or more
Zhen Lei041711c2021-06-30 18:53:17 -07001035 * precisely to memory blocks from the user space POV).
Oscar Salvadordd8e2f22021-05-04 18:39:36 -07001036 * memmap_on_memory is an exception because it reserves initial part
1037 * of the physical memory space for vmemmaps. That space is pageblock
1038 * aligned.
1039 */
David Hildenbrand4986fac2020-10-15 20:07:50 -07001040 if (WARN_ON_ONCE(!nr_pages ||
Oscar Salvadordd8e2f22021-05-04 18:39:36 -07001041 !IS_ALIGNED(pfn, pageblock_nr_pages) ||
1042 !IS_ALIGNED(pfn + nr_pages, PAGES_PER_SECTION)))
David Hildenbrand4986fac2020-10-15 20:07:50 -07001043 return -EINVAL;
1044
David Hildenbrand381eab42018-10-30 15:10:29 -07001045 mem_hotplug_begin();
1046
Michal Hockof1dd2cd2017-07-06 15:38:11 -07001047 /* associate pfn range with the zone */
David Hildenbrandb30c5922020-10-15 20:08:23 -07001048 move_pfn_range_to_zone(zone, pfn, nr_pages, NULL, MIGRATE_ISOLATE);
Michal Hockof1dd2cd2017-07-06 15:38:11 -07001049
Yasunori Goto7b78d332007-10-21 16:41:36 -07001050 arg.start_pfn = pfn;
1051 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001052 node_states_check_changes_online(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001053
Yasunori Goto7b78d332007-10-21 16:41:36 -07001054 ret = memory_notify(MEM_GOING_ONLINE, &arg);
1055 ret = notifier_to_errno(ret);
Chen Yuconge33e33b2016-03-17 14:19:35 -07001056 if (ret)
1057 goto failed_addition;
1058
Dave Hansen3947be12005-10-29 18:16:54 -07001059 /*
David Hildenbrandb30c5922020-10-15 20:08:23 -07001060 * Fixup the number of isolated pageblocks before marking the sections
1061 * onlining, such that undo_isolate_page_range() works correctly.
1062 */
1063 spin_lock_irqsave(&zone->lock, flags);
1064 zone->nr_isolate_pageblock += nr_pages / pageblock_nr_pages;
1065 spin_unlock_irqrestore(&zone->lock, flags);
1066
1067 /*
Yasunori Goto68113782006-06-23 02:03:11 -07001068 * If this zone is not populated, then it is not in zonelist.
1069 * This means the page allocator ignores this zone.
1070 * So, zonelist must be updated after online.
1071 */
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001072 if (!populated_zone(zone)) {
Yasunori Goto68113782006-06-23 02:03:11 -07001073 need_zonelists_rebuild = 1;
Michal Hocko72675e12017-09-06 16:20:24 -07001074 setup_zone_pageset(zone);
Wen Congyang6dcd73d2012-12-11 16:01:01 -08001075 }
Yasunori Goto68113782006-06-23 02:03:11 -07001076
David Hildenbrandaac65322020-10-15 20:08:11 -07001077 online_pages_range(pfn, nr_pages);
David Hildenbrand836809e2021-09-07 19:55:30 -07001078 adjust_present_page_count(pfn_to_page(pfn), group, nr_pages);
Cody P Schaferaa472282013-07-03 15:02:10 -07001079
David Hildenbrandb30c5922020-10-15 20:08:23 -07001080 node_states_set_node(nid, &arg);
1081 if (need_zonelists_rebuild)
1082 build_all_zonelists(NULL);
David Hildenbrandb30c5922020-10-15 20:08:23 -07001083
1084 /* Basic onlining is complete, allow allocation of onlined pages. */
1085 undo_isolate_page_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE);
1086
David Hildenbrand93146d92020-08-06 23:25:35 -07001087 /*
David Hildenbrandb86c5fc2020-10-15 20:09:39 -07001088 * Freshly onlined pages aren't shuffled (e.g., all pages are placed to
1089 * the tail of the freelist when undoing isolation). Shuffle the whole
1090 * zone to make sure the just onlined pages are properly distributed
1091 * across the whole freelist - to create an initial shuffle.
David Hildenbrand93146d92020-08-06 23:25:35 -07001092 */
Dan Williamse900a912019-05-14 15:41:28 -07001093 shuffle_zone(zone);
1094
Mel Gormanb92ca182021-06-28 19:42:12 -07001095 /* reinitialise watermarks and update pcp limits */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001096 init_per_zone_wmark_min();
1097
David Hildenbrandca9a46f2019-09-23 15:36:08 -07001098 kswapd_run(nid);
1099 kcompactd_run(nid);
Dave Hansen61b13992005-10-29 18:16:56 -07001100
Chandra Seetharaman2d1d43f2006-09-29 02:01:25 -07001101 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001102
David Hildenbrandca9a46f2019-09-23 15:36:08 -07001103 memory_notify(MEM_ONLINE, &arg);
David Hildenbrand381eab42018-10-30 15:10:29 -07001104 mem_hotplug_done();
David Rientjes30467e02015-04-14 15:45:11 -07001105 return 0;
Chen Yuconge33e33b2016-03-17 14:19:35 -07001106
1107failed_addition:
1108 pr_debug("online_pages [mem %#010llx-%#010llx] failed\n",
1109 (unsigned long long) pfn << PAGE_SHIFT,
1110 (((unsigned long long) pfn + nr_pages) << PAGE_SHIFT) - 1);
1111 memory_notify(MEM_CANCEL_ONLINE, &arg);
David Hildenbrandfeee6b22020-01-04 12:59:33 -08001112 remove_pfn_range_from_zone(zone, pfn, nr_pages);
David Hildenbrand381eab42018-10-30 15:10:29 -07001113 mem_hotplug_done();
Chen Yuconge33e33b2016-03-17 14:19:35 -07001114 return ret;
Dave Hansen3947be12005-10-29 18:16:54 -07001115}
Keith Mannthey53947022006-09-30 23:27:08 -07001116#endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
Yasunori Gotobc02af92006-06-27 02:53:30 -07001117
Tang Chen0bd85422014-11-13 15:19:41 -08001118static void reset_node_present_pages(pg_data_t *pgdat)
1119{
1120 struct zone *z;
1121
1122 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1123 z->present_pages = 0;
1124
1125 pgdat->node_present_pages = 0;
1126}
1127
Hidetoshi Setoe1319332009-11-17 14:06:18 -08001128/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
David Hildenbrandc68ab182020-06-04 16:48:35 -07001129static pg_data_t __ref *hotadd_new_pgdat(int nid)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001130{
1131 struct pglist_data *pgdat;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001132
Tang Chena1e565a2013-02-22 16:33:18 -08001133 pgdat = NODE_DATA(nid);
1134 if (!pgdat) {
1135 pgdat = arch_alloc_nodedata(nid);
1136 if (!pgdat)
1137 return NULL;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001138
Wei Yang33fce012019-09-23 15:35:52 -07001139 pgdat->per_cpu_nodestats =
1140 alloc_percpu(struct per_cpu_nodestat);
Tang Chena1e565a2013-02-22 16:33:18 -08001141 arch_refresh_nodedata(nid, pgdat);
Gu Zhengb0dc3a32015-03-25 15:55:20 -07001142 } else {
Wei Yang33fce012019-09-23 15:35:52 -07001143 int cpu;
Mel Gormane716f2e2017-05-03 14:53:45 -07001144 /*
Joonsoo Kim97a225e2020-06-03 15:59:01 -07001145 * Reset the nr_zones, order and highest_zoneidx before reuse.
1146 * Note that kswapd will init kswapd_highest_zoneidx properly
Mel Gormane716f2e2017-05-03 14:53:45 -07001147 * when it starts in the near future.
1148 */
Gu Zhengb0dc3a32015-03-25 15:55:20 -07001149 pgdat->nr_zones = 0;
Mel Gorman38087d92016-07-28 15:45:49 -07001150 pgdat->kswapd_order = 0;
Joonsoo Kim97a225e2020-06-03 15:59:01 -07001151 pgdat->kswapd_highest_zoneidx = 0;
Wei Yang33fce012019-09-23 15:35:52 -07001152 for_each_online_cpu(cpu) {
1153 struct per_cpu_nodestat *p;
1154
1155 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
1156 memset(p, 0, sizeof(*p));
1157 }
Tang Chena1e565a2013-02-22 16:33:18 -08001158 }
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001159
1160 /* we can use NODE_DATA(nid) from here */
Oscar Salvador03e85f92018-08-21 21:53:43 -07001161 pgdat->node_id = nid;
David Hildenbrandc68ab182020-06-04 16:48:35 -07001162 pgdat->node_start_pfn = 0;
Oscar Salvador03e85f92018-08-21 21:53:43 -07001163
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001164 /* init node's zones as empty zones, we don't have any present pages.*/
Oscar Salvador03e85f92018-08-21 21:53:43 -07001165 free_area_init_core_hotplug(nid);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001166
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001167 /*
1168 * The node we allocated has no zone fallback lists. For avoiding
1169 * to access not-initialized zonelist, build here.
1170 */
Michal Hocko72675e12017-09-06 16:20:24 -07001171 build_all_zonelists(pgdat);
KAMEZAWA Hiroyuki959ecc42011-06-15 15:08:38 -07001172
Tang Chenf784a3f2014-11-13 15:19:39 -08001173 /*
Tang Chen0bd85422014-11-13 15:19:41 -08001174 * When memory is hot-added, all the memory is in offline state. So
1175 * clear all zones' present_pages because they will be updated in
1176 * online_pages() and offline_pages().
1177 */
Oscar Salvador03e85f92018-08-21 21:53:43 -07001178 reset_node_managed_pages(pgdat);
Tang Chen0bd85422014-11-13 15:19:41 -08001179 reset_node_present_pages(pgdat);
1180
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001181 return pgdat;
1182}
1183
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001184static void rollback_node_hotadd(int nid)
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001185{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001186 pg_data_t *pgdat = NODE_DATA(nid);
1187
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001188 arch_refresh_nodedata(nid, NULL);
Reza Arbab58301692016-08-11 15:33:12 -07001189 free_percpu(pgdat->per_cpu_nodestats);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001190 arch_free_nodedata(pgdat);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001191}
1192
KAMEZAWA Hiroyuki0a547032006-06-27 02:53:35 -07001193
Mel Gormanba2d2662021-06-30 18:53:35 -07001194/*
1195 * __try_online_node - online a node if offlined
Mike Rapoporte8b098f2018-04-05 16:24:57 -07001196 * @nid: the node ID
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001197 * @set_node_online: Whether we want to online the node
minskey guocf234222010-05-24 14:32:41 -07001198 * called by cpu_up() to online a node without onlined memory.
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001199 *
1200 * Returns:
1201 * 1 -> a new node has been allocated
1202 * 0 -> the node is already online
1203 * -ENOMEM -> the node could not be allocated
minskey guocf234222010-05-24 14:32:41 -07001204 */
David Hildenbrandc68ab182020-06-04 16:48:35 -07001205static int __try_online_node(int nid, bool set_node_online)
minskey guocf234222010-05-24 14:32:41 -07001206{
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001207 pg_data_t *pgdat;
1208 int ret = 1;
minskey guocf234222010-05-24 14:32:41 -07001209
Toshi Kani01b0f192013-11-12 15:07:25 -08001210 if (node_online(nid))
1211 return 0;
1212
David Hildenbrandc68ab182020-06-04 16:48:35 -07001213 pgdat = hotadd_new_pgdat(nid);
David Rientjes7553e8f2011-06-22 18:13:01 -07001214 if (!pgdat) {
Toshi Kani01b0f192013-11-12 15:07:25 -08001215 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
minskey guocf234222010-05-24 14:32:41 -07001216 ret = -ENOMEM;
1217 goto out;
1218 }
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001219
1220 if (set_node_online) {
1221 node_set_online(nid);
1222 ret = register_one_node(nid);
1223 BUG_ON(ret);
1224 }
minskey guocf234222010-05-24 14:32:41 -07001225out:
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001226 return ret;
1227}
1228
1229/*
1230 * Users of this function always want to online/register the node
1231 */
1232int try_online_node(int nid)
1233{
1234 int ret;
1235
1236 mem_hotplug_begin();
David Hildenbrandc68ab182020-06-04 16:48:35 -07001237 ret = __try_online_node(nid, true);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001238 mem_hotplug_done();
minskey guocf234222010-05-24 14:32:41 -07001239 return ret;
1240}
1241
Toshi Kani27356f52013-09-11 14:21:49 -07001242static int check_hotplug_memory_range(u64 start, u64 size)
1243{
Pavel Tatashinba325582018-04-05 16:22:39 -07001244 /* memory range must be block size aligned */
David Hildenbrandcec3ebd2019-07-18 15:56:25 -07001245 if (!size || !IS_ALIGNED(start, memory_block_size_bytes()) ||
1246 !IS_ALIGNED(size, memory_block_size_bytes())) {
Pavel Tatashinba325582018-04-05 16:22:39 -07001247 pr_err("Block size [%#lx] unaligned hotplug range: start %#llx, size %#llx",
David Hildenbrandcec3ebd2019-07-18 15:56:25 -07001248 memory_block_size_bytes(), start, size);
Toshi Kani27356f52013-09-11 14:21:49 -07001249 return -EINVAL;
1250 }
1251
1252 return 0;
1253}
1254
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001255static int online_memory_block(struct memory_block *mem, void *arg)
1256{
Anshuman Khandual1adf8b42021-02-25 17:17:13 -08001257 mem->online_type = mhp_default_online_type;
Nathan Fontenotdc18d702017-02-24 15:00:02 -08001258 return device_online(&mem->dev);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001259}
1260
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07001261bool mhp_supports_memmap_on_memory(unsigned long size)
1262{
1263 unsigned long nr_vmemmap_pages = size / PAGE_SIZE;
1264 unsigned long vmemmap_size = nr_vmemmap_pages * sizeof(struct page);
1265 unsigned long remaining_size = size - vmemmap_size;
1266
1267 /*
1268 * Besides having arch support and the feature enabled at runtime, we
1269 * need a few more assumptions to hold true:
1270 *
1271 * a) We span a single memory block: memory onlining/offlinin;g happens
1272 * in memory block granularity. We don't want the vmemmap of online
1273 * memory blocks to reside on offline memory blocks. In the future,
1274 * we might want to support variable-sized memory blocks to make the
1275 * feature more versatile.
1276 *
1277 * b) The vmemmap pages span complete PMDs: We don't want vmemmap code
1278 * to populate memory from the altmap for unrelated parts (i.e.,
1279 * other memory blocks)
1280 *
1281 * c) The vmemmap pages (and thereby the pages that will be exposed to
1282 * the buddy) have to cover full pageblocks: memory onlining/offlining
1283 * code requires applicable ranges to be page-aligned, for example, to
1284 * set the migratetypes properly.
1285 *
1286 * TODO: Although we have a check here to make sure that vmemmap pages
1287 * fully populate a PMD, it is not the right place to check for
1288 * this. A much better solution involves improving vmemmap code
1289 * to fallback to base pages when trying to populate vmemmap using
1290 * altmap as an alternative source of memory, and we do not exactly
1291 * populate a single PMD.
1292 */
1293 return memmap_on_memory &&
Muchun Song2d7a2172021-06-30 18:48:25 -07001294 !hugetlb_free_vmemmap_enabled &&
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07001295 IS_ENABLED(CONFIG_MHP_MEMMAP_ON_MEMORY) &&
1296 size == memory_block_size_bytes() &&
1297 IS_ALIGNED(vmemmap_size, PMD_SIZE) &&
1298 IS_ALIGNED(remaining_size, (pageblock_nr_pages << PAGE_SHIFT));
1299}
1300
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001301/*
1302 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1303 * and online/offline operations (triggered e.g. by sysfs).
1304 *
1305 * we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG
1306 */
David Hildenbrandb6117192020-10-15 20:08:44 -07001307int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
Yasunori Gotobc02af92006-06-27 02:53:30 -07001308{
Catalin Marinasd15dfd32021-03-09 12:26:01 +00001309 struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07001310 struct vmem_altmap mhp_altmap = {};
David Hildenbrand028fc572021-09-07 19:55:26 -07001311 struct memory_group *group = NULL;
David Vrabel62cedb92015-06-25 16:35:49 +01001312 u64 start, size;
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001313 bool new_node = false;
Yasunori Gotobc02af92006-06-27 02:53:30 -07001314 int ret;
1315
David Vrabel62cedb92015-06-25 16:35:49 +01001316 start = res->start;
1317 size = resource_size(res);
1318
Toshi Kani27356f52013-09-11 14:21:49 -07001319 ret = check_hotplug_memory_range(start, size);
1320 if (ret)
1321 return ret;
1322
David Hildenbrand028fc572021-09-07 19:55:26 -07001323 if (mhp_flags & MHP_NID_IS_MGID) {
1324 group = memory_group_find_by_id(nid);
1325 if (!group)
1326 return -EINVAL;
1327 nid = group->nid;
1328 }
1329
Vishal Vermafa6d9ec2020-06-04 16:48:25 -07001330 if (!node_possible(nid)) {
1331 WARN(1, "node %d was absent from the node_possible_map\n", nid);
1332 return -EINVAL;
1333 }
1334
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001335 mem_hotplug_begin();
Nathan Zimmerac13c462014-01-23 15:53:26 -08001336
David Hildenbrand52219ae2020-06-04 16:48:38 -07001337 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1338 memblock_add_node(start, size, nid);
Tang Chen7f36e3e2015-09-04 15:42:32 -07001339
David Hildenbrandc68ab182020-06-04 16:48:35 -07001340 ret = __try_online_node(nid, false);
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001341 if (ret < 0)
1342 goto error;
1343 new_node = ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001344
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07001345 /*
1346 * Self hosted memmap array
1347 */
1348 if (mhp_flags & MHP_MEMMAP_ON_MEMORY) {
1349 if (!mhp_supports_memmap_on_memory(size)) {
1350 ret = -EINVAL;
1351 goto error;
1352 }
1353 mhp_altmap.free = PHYS_PFN(size);
1354 mhp_altmap.base_pfn = PHYS_PFN(start);
1355 params.altmap = &mhp_altmap;
1356 }
1357
Yasunori Gotobc02af92006-06-27 02:53:30 -07001358 /* call arch's memory hotadd */
Logan Gunthorpef5637d32020-04-10 14:33:21 -07001359 ret = arch_add_memory(nid, start, size, &params);
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001360 if (ret < 0)
1361 goto error;
1362
David Hildenbranddb051a02019-07-18 15:56:56 -07001363 /* create memory block devices after memory was added */
David Hildenbrand028fc572021-09-07 19:55:26 -07001364 ret = create_memory_block_devices(start, size, mhp_altmap.alloc,
1365 group);
David Hildenbranddb051a02019-07-18 15:56:56 -07001366 if (ret) {
David Hildenbrand65a2aa52021-09-07 19:55:04 -07001367 arch_remove_memory(start, size, NULL);
David Hildenbranddb051a02019-07-18 15:56:56 -07001368 goto error;
1369 }
1370
Tang Chena1e565a2013-02-22 16:33:18 -08001371 if (new_node) {
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001372 /* If sysfs file of new node can't be created, cpu on the node
Yasunori Goto0fc44152006-06-27 02:53:38 -07001373 * can't be hot-added. There is no rollback way now.
1374 * So, check by BUG_ON() to catch it reluctantly..
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001375 * We online node here. We can't roll back from here.
Yasunori Goto0fc44152006-06-27 02:53:38 -07001376 */
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001377 node_set_online(nid);
1378 ret = __register_one_node(nid);
Yasunori Goto0fc44152006-06-27 02:53:38 -07001379 BUG_ON(ret);
1380 }
1381
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001382 /* link memory sections under this node.*/
Laurent Dufour90c7eae2020-10-15 20:09:15 -07001383 link_mem_sections(nid, PFN_DOWN(start), PFN_UP(start + size - 1),
1384 MEMINIT_HOTPLUG);
Oscar Salvadord5b6f6a32018-08-17 15:46:18 -07001385
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001386 /* create new memmap entry */
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001387 if (!strcmp(res->name, "System RAM"))
1388 firmware_map_add_hotplug(start, start + size, "System RAM");
akpm@linux-foundation.orgd96ae532010-03-05 13:41:58 -08001389
David Hildenbrand381eab42018-10-30 15:10:29 -07001390 /* device_online() will take the lock when calling online_pages() */
1391 mem_hotplug_done();
1392
David Hildenbrand9ca65512020-10-15 20:08:49 -07001393 /*
1394 * In case we're allowed to merge the resource, flag it and trigger
1395 * merging now that adding succeeded.
1396 */
David Hildenbrand26011262021-02-25 17:17:17 -08001397 if (mhp_flags & MHP_MERGE_RESOURCE)
David Hildenbrand9ca65512020-10-15 20:08:49 -07001398 merge_system_ram_resource(res);
1399
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001400 /* online pages if requested */
Anshuman Khandual1adf8b42021-02-25 17:17:13 -08001401 if (mhp_default_online_type != MMOP_OFFLINE)
David Hildenbrandfbcf73c2019-07-18 15:57:46 -07001402 walk_memory_blocks(start, size, NULL, online_memory_block);
Vitaly Kuznetsov31bc3852016-03-15 14:56:48 -07001403
David Hildenbrand381eab42018-10-30 15:10:29 -07001404 return ret;
Yasunori Goto9af3c2d2006-06-27 02:53:34 -07001405error:
1406 /* rollback pgdat allocation and others */
Oscar Salvadorb9ff0362018-08-17 15:46:15 -07001407 if (new_node)
1408 rollback_node_hotadd(nid);
David Hildenbrand52219ae2020-06-04 16:48:38 -07001409 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
1410 memblock_remove(start, size);
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07001411 mem_hotplug_done();
Yasunori Gotobc02af92006-06-27 02:53:30 -07001412 return ret;
1413}
David Vrabel62cedb92015-06-25 16:35:49 +01001414
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001415/* requires device_hotplug_lock, see add_memory_resource() */
David Hildenbrandb6117192020-10-15 20:08:44 -07001416int __ref __add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
David Vrabel62cedb92015-06-25 16:35:49 +01001417{
1418 struct resource *res;
1419 int ret;
1420
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001421 res = register_memory_resource(start, size, "System RAM");
Vitaly Kuznetsov6f754ba2016-01-14 15:21:55 -08001422 if (IS_ERR(res))
1423 return PTR_ERR(res);
David Vrabel62cedb92015-06-25 16:35:49 +01001424
David Hildenbrandb6117192020-10-15 20:08:44 -07001425 ret = add_memory_resource(nid, res, mhp_flags);
David Vrabel62cedb92015-06-25 16:35:49 +01001426 if (ret < 0)
1427 release_memory_resource(res);
1428 return ret;
1429}
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001430
David Hildenbrandb6117192020-10-15 20:08:44 -07001431int add_memory(int nid, u64 start, u64 size, mhp_t mhp_flags)
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001432{
1433 int rc;
1434
1435 lock_device_hotplug();
David Hildenbrandb6117192020-10-15 20:08:44 -07001436 rc = __add_memory(nid, start, size, mhp_flags);
David Hildenbrand8df1d0e2018-10-30 15:10:24 -07001437 unlock_device_hotplug();
1438
1439 return rc;
1440}
Yasunori Gotobc02af92006-06-27 02:53:30 -07001441EXPORT_SYMBOL_GPL(add_memory);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001442
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001443/*
1444 * Add special, driver-managed memory to the system as system RAM. Such
1445 * memory is not exposed via the raw firmware-provided memmap as system
1446 * RAM, instead, it is detected and added by a driver - during cold boot,
1447 * after a reboot, and after kexec.
1448 *
1449 * Reasons why this memory should not be used for the initial memmap of a
1450 * kexec kernel or for placing kexec images:
1451 * - The booting kernel is in charge of determining how this memory will be
1452 * used (e.g., use persistent memory as system RAM)
1453 * - Coordination with a hypervisor is required before this memory
1454 * can be used (e.g., inaccessible parts).
1455 *
1456 * For this memory, no entries in /sys/firmware/memmap ("raw firmware-provided
1457 * memory map") are created. Also, the created memory resource is flagged
David Hildenbrand7cf603d2020-10-15 20:08:33 -07001458 * with IORESOURCE_SYSRAM_DRIVER_MANAGED, so in-kernel users can special-case
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001459 * this memory as well (esp., not place kexec images onto it).
1460 *
1461 * The resource_name (visible via /proc/iomem) has to have the format
1462 * "System RAM ($DRIVER)".
1463 */
1464int add_memory_driver_managed(int nid, u64 start, u64 size,
David Hildenbrandb6117192020-10-15 20:08:44 -07001465 const char *resource_name, mhp_t mhp_flags)
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001466{
1467 struct resource *res;
1468 int rc;
1469
1470 if (!resource_name ||
1471 strstr(resource_name, "System RAM (") != resource_name ||
1472 resource_name[strlen(resource_name) - 1] != ')')
1473 return -EINVAL;
1474
1475 lock_device_hotplug();
1476
1477 res = register_memory_resource(start, size, resource_name);
1478 if (IS_ERR(res)) {
1479 rc = PTR_ERR(res);
1480 goto out_unlock;
1481 }
1482
David Hildenbrandb6117192020-10-15 20:08:44 -07001483 rc = add_memory_resource(nid, res, mhp_flags);
David Hildenbrand7b7b2722020-06-04 16:48:41 -07001484 if (rc < 0)
1485 release_memory_resource(res);
1486
1487out_unlock:
1488 unlock_device_hotplug();
1489 return rc;
1490}
1491EXPORT_SYMBOL_GPL(add_memory_driver_managed);
1492
Anshuman Khandualbca3fea2021-02-25 17:17:33 -08001493/*
1494 * Platforms should define arch_get_mappable_range() that provides
1495 * maximum possible addressable physical memory range for which the
1496 * linear mapping could be created. The platform returned address
1497 * range must adhere to these following semantics.
1498 *
1499 * - range.start <= range.end
1500 * - Range includes both end points [range.start..range.end]
1501 *
1502 * There is also a fallback definition provided here, allowing the
1503 * entire possible physical address range in case any platform does
1504 * not define arch_get_mappable_range().
1505 */
1506struct range __weak arch_get_mappable_range(void)
1507{
1508 struct range mhp_range = {
1509 .start = 0UL,
1510 .end = -1ULL,
1511 };
1512 return mhp_range;
1513}
1514
1515struct range mhp_get_pluggable_range(bool need_mapping)
1516{
1517 const u64 max_phys = (1ULL << MAX_PHYSMEM_BITS) - 1;
1518 struct range mhp_range;
1519
1520 if (need_mapping) {
1521 mhp_range = arch_get_mappable_range();
1522 if (mhp_range.start > max_phys) {
1523 mhp_range.start = 0;
1524 mhp_range.end = 0;
1525 }
1526 mhp_range.end = min_t(u64, mhp_range.end, max_phys);
1527 } else {
1528 mhp_range.start = 0;
1529 mhp_range.end = max_phys;
1530 }
1531 return mhp_range;
1532}
1533EXPORT_SYMBOL_GPL(mhp_get_pluggable_range);
1534
1535bool mhp_range_allowed(u64 start, u64 size, bool need_mapping)
1536{
1537 struct range mhp_range = mhp_get_pluggable_range(need_mapping);
1538 u64 end = start + size;
1539
1540 if (start < end && start >= mhp_range.start && (end - 1) <= mhp_range.end)
1541 return true;
1542
1543 pr_warn("Hotplug memory [%#llx-%#llx] exceeds maximum addressable range [%#llx-%#llx]\n",
1544 start, end, mhp_range.start, mhp_range.end);
1545 return false;
1546}
1547
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001548#ifdef CONFIG_MEMORY_HOTREMOVE
1549/*
David Hildenbrand92917992020-02-03 17:34:26 -08001550 * Confirm all pages in a range [start, end) belong to the same zone (skipping
1551 * memory holes). When true, return the zone.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001552 */
David Hildenbrand92917992020-02-03 17:34:26 -08001553struct zone *test_pages_in_a_zone(unsigned long start_pfn,
1554 unsigned long end_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001555{
Andrew Banman5f0f2882015-12-29 14:54:25 -08001556 unsigned long pfn, sec_end_pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001557 struct zone *zone = NULL;
1558 struct page *page;
Mike Rapoport673d40c2021-09-07 19:54:55 -07001559
Toshi Kanideb88a22017-02-03 13:13:20 -08001560 for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn + 1);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001561 pfn < end_pfn;
Toshi Kanideb88a22017-02-03 13:13:20 -08001562 pfn = sec_end_pfn, sec_end_pfn += PAGES_PER_SECTION) {
Andrew Banman5f0f2882015-12-29 14:54:25 -08001563 /* Make sure the memory section is present first */
1564 if (!present_section_nr(pfn_to_section_nr(pfn)))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001565 continue;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001566 for (; pfn < sec_end_pfn && pfn < end_pfn;
1567 pfn += MAX_ORDER_NR_PAGES) {
Mikhail Zaslonko24feb472019-02-01 14:20:38 -08001568 /* Check if we got outside of the zone */
Mike Rapoport673d40c2021-09-07 19:54:55 -07001569 if (zone && !zone_spans_pfn(zone, pfn))
David Hildenbrand92917992020-02-03 17:34:26 -08001570 return NULL;
Mike Rapoport673d40c2021-09-07 19:54:55 -07001571 page = pfn_to_page(pfn);
Andrew Banman5f0f2882015-12-29 14:54:25 -08001572 if (zone && page_zone(page) != zone)
David Hildenbrand92917992020-02-03 17:34:26 -08001573 return NULL;
Andrew Banman5f0f2882015-12-29 14:54:25 -08001574 zone = page_zone(page);
1575 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001576 }
Toshi Kanideb88a22017-02-03 13:13:20 -08001577
David Hildenbrand92917992020-02-03 17:34:26 -08001578 return zone;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001579}
1580
1581/*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001582 * Scan pfn range [start,end) to find movable/migratable pages (LRU pages,
David Hildenbrandaa218792020-05-07 16:01:30 +02001583 * non-lru movable pages and hugepages). Will skip over most unmovable
1584 * pages (esp., pages that can be skipped when offlining), but bail out on
1585 * definitely unmovable pages.
1586 *
1587 * Returns:
1588 * 0 in case a movable page is found and movable_pfn was updated.
1589 * -ENOENT in case no movable page was found.
1590 * -EBUSY in case a definitely unmovable page was found.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001591 */
David Hildenbrandaa218792020-05-07 16:01:30 +02001592static int scan_movable_pages(unsigned long start, unsigned long end,
1593 unsigned long *movable_pfn)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001594{
1595 unsigned long pfn;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001596
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001597 for (pfn = start; pfn < end; pfn++) {
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001598 struct page *page, *head;
1599 unsigned long skip;
1600
1601 if (!pfn_valid(pfn))
1602 continue;
1603 page = pfn_to_page(pfn);
1604 if (PageLRU(page))
David Hildenbrandaa218792020-05-07 16:01:30 +02001605 goto found;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001606 if (__PageMovable(page))
David Hildenbrandaa218792020-05-07 16:01:30 +02001607 goto found;
1608
1609 /*
1610 * PageOffline() pages that are not marked __PageMovable() and
1611 * have a reference count > 0 (after MEM_GOING_OFFLINE) are
1612 * definitely unmovable. If their reference count would be 0,
1613 * they could at least be skipped when offlining memory.
1614 */
1615 if (PageOffline(page) && page_count(page))
1616 return -EBUSY;
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001617
1618 if (!PageHuge(page))
1619 continue;
1620 head = compound_head(page);
Mike Kravetz8f251a32021-02-24 12:08:56 -08001621 /*
1622 * This test is racy as we hold no reference or lock. The
1623 * hugetlb page could have been free'ed and head is no longer
1624 * a hugetlb page before the following check. In such unlikely
1625 * cases false positives and negatives are possible. Calling
1626 * code must deal with these scenarios.
1627 */
1628 if (HPageMigratable(head))
David Hildenbrandaa218792020-05-07 16:01:30 +02001629 goto found;
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001630 skip = compound_nr(head) - (page - head);
Oscar Salvadoreeb0efd2019-02-01 14:20:47 -08001631 pfn += skip - 1;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001632 }
David Hildenbrandaa218792020-05-07 16:01:30 +02001633 return -ENOENT;
1634found:
1635 *movable_pfn = pfn;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001636 return 0;
1637}
1638
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001639static int
1640do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1641{
1642 unsigned long pfn;
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07001643 struct page *page, *head;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001644 int ret = 0;
1645 LIST_HEAD(source);
Liam Mark786dee82021-06-30 18:52:43 -07001646 static DEFINE_RATELIMIT_STATE(migrate_rs, DEFAULT_RATELIMIT_INTERVAL,
1647 DEFAULT_RATELIMIT_BURST);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001648
Michal Hockoa85009c2018-12-28 00:38:29 -08001649 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001650 if (!pfn_valid(pfn))
1651 continue;
1652 page = pfn_to_page(pfn);
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07001653 head = compound_head(page);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001654
1655 if (PageHuge(page)) {
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -07001656 pfn = page_to_pfn(head) + compound_nr(head) - 1;
Oscar Salvadordaf35382019-03-05 15:48:53 -08001657 isolate_huge_page(head, &source);
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001658 continue;
Michal Hocko94723aa2018-04-10 16:30:07 -07001659 } else if (PageTransHuge(page))
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07001660 pfn = page_to_pfn(head) + thp_nr_pages(page) - 1;
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001661
Michal Hockob15c8722018-12-28 00:38:01 -08001662 /*
1663 * HWPoison pages have elevated reference counts so the migration would
1664 * fail on them. It also doesn't make any sense to migrate them in the
1665 * first place. Still try to unmap such a page in case it is still mapped
1666 * (e.g. current hwpoison implementation doesn't unmap KSM pages but keep
1667 * the unmap as the catch all safety net).
1668 */
1669 if (PageHWPoison(page)) {
1670 if (WARN_ON(PageLRU(page)))
1671 isolate_lru_page(page);
1672 if (page_mapped(page))
Shakeel Butt013339d2020-12-14 19:06:39 -08001673 try_to_unmap(page, TTU_IGNORE_MLOCK);
Michal Hockob15c8722018-12-28 00:38:01 -08001674 continue;
1675 }
1676
Konstantin Khlebnikov700c2a42011-05-24 17:12:19 -07001677 if (!get_page_unless_zero(page))
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001678 continue;
1679 /*
Yisheng Xie0efadf42017-02-24 14:57:39 -08001680 * We can skip free pages. And we can deal with pages on
1681 * LRU and non-lru movable pages.
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001682 */
Yisheng Xie0efadf42017-02-24 14:57:39 -08001683 if (PageLRU(page))
1684 ret = isolate_lru_page(page);
1685 else
1686 ret = isolate_movable_page(page, ISOLATE_UNEVICTABLE);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001687 if (!ret) { /* Success */
Nick Piggin62695a82008-10-18 20:26:09 -07001688 list_add_tail(&page->lru, &source);
Yisheng Xie0efadf42017-02-24 14:57:39 -08001689 if (!__PageMovable(page))
1690 inc_node_page_state(page, NR_ISOLATED_ANON +
Huang Ying9de4f222020-04-06 20:04:41 -07001691 page_is_file_lru(page));
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001692
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001693 } else {
Liam Mark786dee82021-06-30 18:52:43 -07001694 if (__ratelimit(&migrate_rs)) {
1695 pr_warn("failed to isolate pfn %lx\n", pfn);
1696 dump_page(page, "isolation failed");
1697 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001698 }
Oscar Salvador17230582019-02-01 14:19:57 -08001699 put_page(page);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001700 }
Bob Liuf3ab2632010-10-26 14:22:10 -07001701 if (!list_empty(&source)) {
Joonsoo Kim203e6e52020-10-17 16:14:00 -07001702 nodemask_t nmask = node_states[N_MEMORY];
1703 struct migration_target_control mtc = {
1704 .nmask = &nmask,
1705 .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
1706 };
1707
1708 /*
1709 * We have checked that migration range is on a single zone so
1710 * we can use the nid of the first page to all the others.
1711 */
1712 mtc.nid = page_to_nid(list_first_entry(&source, struct page, lru));
1713
1714 /*
1715 * try to allocate from a different node but reuse this node
1716 * if there are no other online nodes to be used (e.g. we are
1717 * offlining a part of the only existing node)
1718 */
1719 node_clear(mtc.nid, nmask);
1720 if (nodes_empty(nmask))
1721 node_set(mtc.nid, nmask);
1722 ret = migrate_pages(&source, alloc_migration_target, NULL,
1723 (unsigned long)&mtc, MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001724 if (ret) {
1725 list_for_each_entry(page, &source, lru) {
Liam Mark786dee82021-06-30 18:52:43 -07001726 if (__ratelimit(&migrate_rs)) {
1727 pr_warn("migrating pfn %lx failed ret:%d\n",
1728 page_to_pfn(page), ret);
1729 dump_page(page, "migration failure");
1730 }
Michal Hocko2932c8b2018-12-28 00:33:53 -08001731 }
Naoya Horiguchic8721bb2013-09-11 14:22:09 -07001732 putback_movable_pages(&source);
Michal Hocko2932c8b2018-12-28 00:33:53 -08001733 }
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001734 }
Oscar Salvador17230582019-02-01 14:19:57 -08001735
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001736 return ret;
1737}
1738
Tang Chenc5320922013-11-12 15:08:10 -08001739static int __init cmdline_parse_movable_node(char *p)
1740{
Tang Chen55ac5902014-01-21 15:49:35 -08001741 movable_node_enabled = true;
Tang Chenc5320922013-11-12 15:08:10 -08001742 return 0;
1743}
1744early_param("movable_node", cmdline_parse_movable_node);
1745
Lai Jiangshand9713672012-12-11 16:01:03 -08001746/* check which state of node_states will be changed when offline memory */
1747static void node_states_check_changes_offline(unsigned long nr_pages,
1748 struct zone *zone, struct memory_notify *arg)
1749{
1750 struct pglist_data *pgdat = zone->zone_pgdat;
1751 unsigned long present_pages = 0;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001752 enum zone_type zt;
1753
Anshuman Khandual98fa15f2019-03-05 15:42:58 -08001754 arg->status_change_nid = NUMA_NO_NODE;
1755 arg->status_change_nid_normal = NUMA_NO_NODE;
1756 arg->status_change_nid_high = NUMA_NO_NODE;
Lai Jiangshand9713672012-12-11 16:01:03 -08001757
1758 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001759 * Check whether node_states[N_NORMAL_MEMORY] will be changed.
1760 * If the memory to be offline is within the range
1761 * [0..ZONE_NORMAL], and it is the last present memory there,
1762 * the zones in that range will become empty after the offlining,
1763 * thus we can determine that we need to clear the node from
1764 * node_states[N_NORMAL_MEMORY].
Lai Jiangshand9713672012-12-11 16:01:03 -08001765 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001766 for (zt = 0; zt <= ZONE_NORMAL; zt++)
Lai Jiangshand9713672012-12-11 16:01:03 -08001767 present_pages += pgdat->node_zones[zt].present_pages;
Oscar Salvador86b27be2018-10-26 15:07:38 -07001768 if (zone_idx(zone) <= ZONE_NORMAL && nr_pages >= present_pages)
Lai Jiangshand9713672012-12-11 16:01:03 -08001769 arg->status_change_nid_normal = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001770
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001771#ifdef CONFIG_HIGHMEM
1772 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001773 * node_states[N_HIGH_MEMORY] contains nodes which
1774 * have normal memory or high memory.
1775 * Here we add the present_pages belonging to ZONE_HIGHMEM.
1776 * If the zone is within the range of [0..ZONE_HIGHMEM), and
1777 * we determine that the zones in that range become empty,
1778 * we need to clear the node for N_HIGH_MEMORY.
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001779 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001780 present_pages += pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1781 if (zone_idx(zone) <= ZONE_HIGHMEM && nr_pages >= present_pages)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001782 arg->status_change_nid_high = zone_to_nid(zone);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001783#endif
1784
Lai Jiangshand9713672012-12-11 16:01:03 -08001785 /*
Oscar Salvador86b27be2018-10-26 15:07:38 -07001786 * We have accounted the pages from [0..ZONE_NORMAL), and
1787 * in case of CONFIG_HIGHMEM the pages from ZONE_HIGHMEM
1788 * as well.
1789 * Here we count the possible pages from ZONE_MOVABLE.
1790 * If after having accounted all the pages, we see that the nr_pages
1791 * to be offlined is over or equal to the accounted pages,
1792 * we know that the node will become empty, and so, we can clear
1793 * it for N_MEMORY as well.
Lai Jiangshand9713672012-12-11 16:01:03 -08001794 */
Oscar Salvador86b27be2018-10-26 15:07:38 -07001795 present_pages += pgdat->node_zones[ZONE_MOVABLE].present_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001796
Lai Jiangshand9713672012-12-11 16:01:03 -08001797 if (nr_pages >= present_pages)
1798 arg->status_change_nid = zone_to_nid(zone);
Lai Jiangshand9713672012-12-11 16:01:03 -08001799}
1800
1801static void node_states_clear_node(int node, struct memory_notify *arg)
1802{
1803 if (arg->status_change_nid_normal >= 0)
1804 node_clear_state(node, N_NORMAL_MEMORY);
1805
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001806 if (arg->status_change_nid_high >= 0)
Lai Jiangshand9713672012-12-11 16:01:03 -08001807 node_clear_state(node, N_HIGH_MEMORY);
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001808
Oscar Salvadorcf01f6f52018-10-26 15:07:28 -07001809 if (arg->status_change_nid >= 0)
Lai Jiangshan6715ddf2012-12-12 13:51:49 -08001810 node_clear_state(node, N_MEMORY);
Lai Jiangshand9713672012-12-11 16:01:03 -08001811}
1812
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001813static int count_system_ram_pages_cb(unsigned long start_pfn,
1814 unsigned long nr_pages, void *data)
1815{
1816 unsigned long *nr_system_ram_pages = data;
1817
1818 *nr_system_ram_pages += nr_pages;
1819 return 0;
1820}
1821
David Hildenbrand836809e2021-09-07 19:55:30 -07001822int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages,
1823 struct memory_group *group)
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001824{
David Hildenbrand73a11c92020-10-15 20:07:46 -07001825 const unsigned long end_pfn = start_pfn + nr_pages;
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001826 unsigned long pfn, system_ram_pages = 0;
Cody P Schaferd7029092013-07-03 15:02:11 -07001827 unsigned long flags;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001828 struct zone *zone;
Yasunori Goto7b78d332007-10-21 16:41:36 -07001829 struct memory_notify arg;
David Hildenbrandea151532020-10-15 20:08:03 -07001830 int ret, node;
Michal Hocko79605092018-12-28 00:33:49 -08001831 char *reason;
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001832
Oscar Salvadordd8e2f22021-05-04 18:39:36 -07001833 /*
1834 * {on,off}lining is constrained to full memory sections (or more
Zhen Lei041711c2021-06-30 18:53:17 -07001835 * precisely to memory blocks from the user space POV).
Oscar Salvadordd8e2f22021-05-04 18:39:36 -07001836 * memmap_on_memory is an exception because it reserves initial part
1837 * of the physical memory space for vmemmaps. That space is pageblock
1838 * aligned.
1839 */
David Hildenbrand4986fac2020-10-15 20:07:50 -07001840 if (WARN_ON_ONCE(!nr_pages ||
Oscar Salvadordd8e2f22021-05-04 18:39:36 -07001841 !IS_ALIGNED(start_pfn, pageblock_nr_pages) ||
1842 !IS_ALIGNED(start_pfn + nr_pages, PAGES_PER_SECTION)))
David Hildenbrand4986fac2020-10-15 20:07:50 -07001843 return -EINVAL;
1844
David Hildenbrand381eab42018-10-30 15:10:29 -07001845 mem_hotplug_begin();
1846
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001847 /*
1848 * Don't allow to offline memory blocks that contain holes.
1849 * Consequently, memory blocks with holes can never get onlined
1850 * via the hotplug path - online_pages() - as hotplugged memory has
1851 * no holes. This way, we e.g., don't have to worry about marking
1852 * memory holes PG_reserved, don't need pfn_valid() checks, and can
1853 * avoid using walk_system_ram_range() later.
1854 */
David Hildenbrand73a11c92020-10-15 20:07:46 -07001855 walk_system_ram_range(start_pfn, nr_pages, &system_ram_pages,
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001856 count_system_ram_pages_cb);
David Hildenbrand73a11c92020-10-15 20:07:46 -07001857 if (system_ram_pages != nr_pages) {
David Hildenbrandc5e79ef2019-11-30 17:54:17 -08001858 ret = -EINVAL;
1859 reason = "memory holes";
1860 goto failed_removal;
1861 }
1862
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001863 /* This makes hotplug much easier...and readable.
1864 we assume this for now. .*/
David Hildenbrand92917992020-02-03 17:34:26 -08001865 zone = test_pages_in_a_zone(start_pfn, end_pfn);
1866 if (!zone) {
Michal Hocko79605092018-12-28 00:33:49 -08001867 ret = -EINVAL;
1868 reason = "multizone range";
1869 goto failed_removal;
David Hildenbrand381eab42018-10-30 15:10:29 -07001870 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001871 node = zone_to_nid(zone);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001872
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -08001873 /*
1874 * Disable pcplists so that page isolation cannot race with freeing
1875 * in a way that pages from isolated pageblock are left on pcplists.
1876 */
1877 zone_pcp_disable(zone);
Minchan Kimd479960e2021-05-04 18:36:54 -07001878 lru_cache_disable();
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -08001879
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001880 /* set above range as isolated */
Wen Congyangb023f462012-12-11 16:00:45 -08001881 ret = start_isolate_page_range(start_pfn, end_pfn,
Michal Hockod381c542018-12-28 00:33:56 -08001882 MIGRATE_MOVABLE,
David Hildenbrand756d25b2019-11-30 17:54:07 -08001883 MEMORY_OFFLINE | REPORT_FAILURE);
David Hildenbrand3fa0c7c2020-10-15 20:08:07 -07001884 if (ret) {
Michal Hocko79605092018-12-28 00:33:49 -08001885 reason = "failure to isolate range";
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -08001886 goto failed_removal_pcplists_disabled;
David Hildenbrand381eab42018-10-30 15:10:29 -07001887 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001888
1889 arg.start_pfn = start_pfn;
1890 arg.nr_pages = nr_pages;
Lai Jiangshand9713672012-12-11 16:01:03 -08001891 node_states_check_changes_offline(nr_pages, zone, &arg);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001892
1893 ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1894 ret = notifier_to_errno(ret);
Michal Hocko79605092018-12-28 00:33:49 -08001895 if (ret) {
1896 reason = "notifier failure";
1897 goto failed_removal_isolated;
1898 }
Yasunori Goto7b78d332007-10-21 16:41:36 -07001899
Michal Hockobb8965b2018-12-28 00:38:32 -08001900 do {
David Hildenbrandaa218792020-05-07 16:01:30 +02001901 pfn = start_pfn;
1902 do {
Michal Hockobb8965b2018-12-28 00:38:32 -08001903 if (signal_pending(current)) {
1904 ret = -EINTR;
1905 reason = "signal backoff";
1906 goto failed_removal_isolated;
1907 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001908
Michal Hockobb8965b2018-12-28 00:38:32 -08001909 cond_resched();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001910
David Hildenbrandaa218792020-05-07 16:01:30 +02001911 ret = scan_movable_pages(pfn, end_pfn, &pfn);
1912 if (!ret) {
Michal Hockobb8965b2018-12-28 00:38:32 -08001913 /*
1914 * TODO: fatal migration failures should bail
1915 * out
1916 */
1917 do_migrate_range(pfn, end_pfn);
1918 }
David Hildenbrandaa218792020-05-07 16:01:30 +02001919 } while (!ret);
1920
1921 if (ret != -ENOENT) {
1922 reason = "unmovable page";
1923 goto failed_removal_isolated;
Michal Hockobb8965b2018-12-28 00:38:32 -08001924 }
Michal Hocko72b39cf2017-11-15 17:33:34 -08001925
Michal Hockobb8965b2018-12-28 00:38:32 -08001926 /*
1927 * Dissolve free hugepages in the memory block before doing
1928 * offlining actually in order to make hugetlbfs's object
1929 * counting consistent.
1930 */
1931 ret = dissolve_free_huge_pages(start_pfn, end_pfn);
1932 if (ret) {
1933 reason = "failure to dissolve huge pages";
1934 goto failed_removal_isolated;
1935 }
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001936
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001937 ret = test_pages_isolated(start_pfn, end_pfn, MEMORY_OFFLINE);
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -08001938
Michal Hocko5557c762019-05-13 17:21:24 -07001939 } while (ret);
Michal Hockobb8965b2018-12-28 00:38:32 -08001940
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001941 /* Mark all sections offline and remove free pages from the buddy. */
1942 __offline_isolated_pages(start_pfn, end_pfn);
Laurent Dufour7c330232020-12-15 20:42:26 -08001943 pr_debug("Offlined Pages %ld\n", nr_pages);
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001944
Qian Cai9b7ea462019-03-28 20:43:34 -07001945 /*
David Hildenbrandb30c5922020-10-15 20:08:23 -07001946 * The memory sections are marked offline, and the pageblock flags
1947 * effectively stale; nobody should be touching them. Fixup the number
1948 * of isolated pageblocks, memory onlining will properly revert this.
Qian Cai9b7ea462019-03-28 20:43:34 -07001949 */
1950 spin_lock_irqsave(&zone->lock, flags);
David Hildenbrandea151532020-10-15 20:08:03 -07001951 zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
Qian Cai9b7ea462019-03-28 20:43:34 -07001952 spin_unlock_irqrestore(&zone->lock, flags);
1953
Minchan Kimd479960e2021-05-04 18:36:54 -07001954 lru_cache_enable();
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -08001955 zone_pcp_enable(zone);
1956
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001957 /* removal success */
David Hildenbrand0a1a9a02020-10-15 20:07:54 -07001958 adjust_managed_page_count(pfn_to_page(start_pfn), -nr_pages);
David Hildenbrand836809e2021-09-07 19:55:30 -07001959 adjust_present_page_count(pfn_to_page(start_pfn), group, -nr_pages);
Yasunori Goto7b78d332007-10-21 16:41:36 -07001960
Mel Gormanb92ca182021-06-28 19:42:12 -07001961 /* reinitialise watermarks and update pcp limits */
KOSAKI Motohiro1b79acc2011-05-24 17:11:32 -07001962 init_per_zone_wmark_min();
1963
Xishi Qiu1e8537b2012-10-08 16:31:51 -07001964 if (!populated_zone(zone)) {
Jiang Liu340175b2012-07-31 16:43:32 -07001965 zone_pcp_reset(zone);
Michal Hocko72675e12017-09-06 16:20:24 -07001966 build_all_zonelists(NULL);
Mel Gormanb92ca182021-06-28 19:42:12 -07001967 }
Jiang Liu340175b2012-07-31 16:43:32 -07001968
Lai Jiangshand9713672012-12-11 16:01:03 -08001969 node_states_clear_node(node, &arg);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001970 if (arg.status_change_nid >= 0) {
David Rientjes8fe23e02009-12-14 17:58:33 -08001971 kswapd_stop(node);
Vlastimil Babka698b1b32016-03-17 14:18:08 -07001972 kcompactd_stop(node);
1973 }
Minchan Kimbce73942009-06-16 15:32:50 -07001974
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001975 writeback_set_ratelimit();
Yasunori Goto7b78d332007-10-21 16:41:36 -07001976
1977 memory_notify(MEM_OFFLINE, &arg);
David Hildenbrandfeee6b22020-01-04 12:59:33 -08001978 remove_pfn_range_from_zone(zone, start_pfn, nr_pages);
David Hildenbrand381eab42018-10-30 15:10:29 -07001979 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001980 return 0;
1981
Michal Hocko79605092018-12-28 00:33:49 -08001982failed_removal_isolated:
1983 undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
Qian Caic4efe482019-03-28 20:44:16 -07001984 memory_notify(MEM_CANCEL_OFFLINE, &arg);
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -08001985failed_removal_pcplists_disabled:
Miaohe Lin946746d12021-08-25 12:17:55 -07001986 lru_cache_enable();
Vlastimil Babkaec6e8c7e2020-12-14 19:10:59 -08001987 zone_pcp_enable(zone);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001988failed_removal:
Michal Hocko79605092018-12-28 00:33:49 -08001989 pr_debug("memory offlining [mem %#010llx-%#010llx] failed due to %s\n",
Chen Yuconge33e33b2016-03-17 14:19:35 -07001990 (unsigned long long) start_pfn << PAGE_SHIFT,
Michal Hocko79605092018-12-28 00:33:49 -08001991 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1,
1992 reason);
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001993 /* pushback to free area */
David Hildenbrand381eab42018-10-30 15:10:29 -07001994 mem_hotplug_done();
KAMEZAWA Hiroyuki0c0e6192007-10-16 01:26:12 -07001995 return ret;
1996}
Badari Pulavarty71088782008-10-18 20:25:58 -07001997
Xishi Qiud6de9d52013-11-12 15:07:20 -08001998static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
Wen Congyangbbc76be2013-02-22 16:32:54 -08001999{
2000 int ret = !is_memblock_offlined(mem);
David Hildenbrande1c158e2021-09-07 19:55:09 -07002001 int *nid = arg;
Wen Congyangbbc76be2013-02-22 16:32:54 -08002002
David Hildenbrande1c158e2021-09-07 19:55:09 -07002003 *nid = mem->nid;
Randy Dunlap349daa02013-04-29 15:08:49 -07002004 if (unlikely(ret)) {
2005 phys_addr_t beginpa, endpa;
2006
2007 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
David Hildenbrandb6c88d32019-09-23 15:35:49 -07002008 endpa = beginpa + memory_block_size_bytes() - 1;
Joe Perches756a0252016-03-17 14:19:47 -07002009 pr_warn("removing memory fails, because memory [%pa-%pa] is onlined\n",
Randy Dunlap349daa02013-04-29 15:08:49 -07002010 &beginpa, &endpa);
Wen Congyangbbc76be2013-02-22 16:32:54 -08002011
Pavel Tatashineca499a2019-07-16 16:30:31 -07002012 return -EBUSY;
2013 }
2014 return 0;
Wen Congyangbbc76be2013-02-22 16:32:54 -08002015}
2016
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07002017static int get_nr_vmemmap_pages_cb(struct memory_block *mem, void *arg)
2018{
2019 /*
2020 * If not set, continue with the next block.
2021 */
2022 return mem->nr_vmemmap_pages;
2023}
2024
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002025static int check_cpu_on_node(pg_data_t *pgdat)
Tang Chen60a5a192013-02-22 16:33:14 -08002026{
Tang Chen60a5a192013-02-22 16:33:14 -08002027 int cpu;
2028
2029 for_each_present_cpu(cpu) {
2030 if (cpu_to_node(cpu) == pgdat->node_id)
2031 /*
2032 * the cpu on this node isn't removed, and we can't
2033 * offline this node.
2034 */
2035 return -EBUSY;
2036 }
2037
2038 return 0;
2039}
2040
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08002041static int check_no_memblock_for_node_cb(struct memory_block *mem, void *arg)
2042{
2043 int nid = *(int *)arg;
2044
2045 /*
2046 * If a memory block belongs to multiple nodes, the stored nid is not
2047 * reliable. However, such blocks are always online (e.g., cannot get
2048 * offlined) and, therefore, are still spanned by the node.
2049 */
2050 return mem->nid == nid ? -EEXIST : 0;
2051}
2052
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002053/**
2054 * try_offline_node
Mike Rapoporte8b098f2018-04-05 16:24:57 -07002055 * @nid: the node ID
Toshi Kani0f1cfe92013-09-11 14:21:50 -07002056 *
2057 * Offline a node if all memory sections and cpus of the node are removed.
2058 *
2059 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2060 * and online/offline operations before this call.
2061 */
Wen Congyang90b30cd2013-02-22 16:33:27 -08002062void try_offline_node(int nid)
Tang Chen60a5a192013-02-22 16:33:14 -08002063{
Wen Congyangd822b862013-02-22 16:33:16 -08002064 pg_data_t *pgdat = NODE_DATA(nid);
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08002065 int rc;
Tang Chen60a5a192013-02-22 16:33:14 -08002066
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08002067 /*
2068 * If the node still spans pages (especially ZONE_DEVICE), don't
2069 * offline it. A node spans memory after move_pfn_range_to_zone(),
2070 * e.g., after the memory block was onlined.
2071 */
2072 if (pgdat->node_spanned_pages)
Tang Chen60a5a192013-02-22 16:33:14 -08002073 return;
David Hildenbrand2c91f8f2019-11-15 17:34:57 -08002074
2075 /*
2076 * Especially offline memory blocks might not be spanned by the
2077 * node. They will get spanned by the node once they get onlined.
2078 * However, they link to the node in sysfs and can get onlined later.
2079 */
2080 rc = for_each_memory_block(&nid, check_no_memblock_for_node_cb);
2081 if (rc)
2082 return;
Tang Chen60a5a192013-02-22 16:33:14 -08002083
Michal Hocko46a36792018-12-28 00:34:13 -08002084 if (check_cpu_on_node(pgdat))
Tang Chen60a5a192013-02-22 16:33:14 -08002085 return;
2086
2087 /*
2088 * all memory/cpu of this node are removed, we can offline this
2089 * node now.
2090 */
2091 node_set_offline(nid);
2092 unregister_one_node(nid);
2093}
Wen Congyang90b30cd2013-02-22 16:33:27 -08002094EXPORT_SYMBOL(try_offline_node);
Tang Chen60a5a192013-02-22 16:33:14 -08002095
David Hildenbrande1c158e2021-09-07 19:55:09 -07002096static int __ref try_remove_memory(u64 start, u64 size)
Wen Congyangbbc76be2013-02-22 16:32:54 -08002097{
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07002098 struct vmem_altmap mhp_altmap = {};
2099 struct vmem_altmap *altmap = NULL;
2100 unsigned long nr_vmemmap_pages;
David Hildenbrande1c158e2021-09-07 19:55:09 -07002101 int rc = 0, nid = NUMA_NO_NODE;
Wen Congyang993c1aa2013-02-22 16:32:50 -08002102
Toshi Kani27356f52013-09-11 14:21:49 -07002103 BUG_ON(check_hotplug_memory_range(start, size));
2104
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002105 /*
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002106 * All memory blocks must be offlined before removing memory. Check
Pavel Tatashineca499a2019-07-16 16:30:31 -07002107 * whether all memory blocks in question are offline and return error
Rafael J. Wysocki242831e2013-05-27 12:58:46 +02002108 * if this is not the case.
David Hildenbrande1c158e2021-09-07 19:55:09 -07002109 *
2110 * While at it, determine the nid. Note that if we'd have mixed nodes,
2111 * we'd only try to offline the last determined one -- which is good
2112 * enough for the cases we care about.
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002113 */
David Hildenbrande1c158e2021-09-07 19:55:09 -07002114 rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
Pavel Tatashineca499a2019-07-16 16:30:31 -07002115 if (rc)
Jia Heb4223a52020-08-11 18:32:20 -07002116 return rc;
Yasuaki Ishimatsu6677e3e2013-02-22 16:32:52 -08002117
Oscar Salvadora08a2ae2021-05-04 18:39:42 -07002118 /*
2119 * We only support removing memory added with MHP_MEMMAP_ON_MEMORY in
2120 * the same granularity it was added - a single memory block.
2121 */
2122 if (memmap_on_memory) {
2123 nr_vmemmap_pages = walk_memory_blocks(start, size, NULL,
2124 get_nr_vmemmap_pages_cb);
2125 if (nr_vmemmap_pages) {
2126 if (size != memory_block_size_bytes()) {
2127 pr_warn("Refuse to remove %#llx - %#llx,"
2128 "wrong granularity\n",
2129 start, start + size);
2130 return -EINVAL;
2131 }
2132
2133 /*
2134 * Let remove_pmd_table->free_hugepage_table do the
2135 * right thing if we used vmem_altmap when hot-adding
2136 * the range.
2137 */
2138 mhp_altmap.alloc = nr_vmemmap_pages;
2139 altmap = &mhp_altmap;
2140 }
2141 }
2142
Yasuaki Ishimatsu46c66c42013-02-22 16:32:56 -08002143 /* remove memmap entry */
2144 firmware_map_remove(start, start + size, "System RAM");
2145
Dan Williamsf1037ec2020-01-30 22:11:17 -08002146 /*
2147 * Memory block device removal under the device_hotplug_lock is
2148 * a barrier against racing online attempts.
2149 */
David Hildenbrand4c4b7f92019-07-18 15:57:06 -07002150 remove_memory_block_devices(start, size);
2151
Dan Williamsf1037ec2020-01-30 22:11:17 -08002152 mem_hotplug_begin();
2153
David Hildenbrand65a2aa52021-09-07 19:55:04 -07002154 arch_remove_memory(start, size, altmap);
David Hildenbrand52219ae2020-06-04 16:48:38 -07002155
2156 if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
2157 memblock_free(start, size);
2158 memblock_remove(start, size);
2159 }
2160
David Hildenbrandcb8e3c82020-10-15 20:09:12 -07002161 release_mem_region_adjustable(start, size);
Wen Congyang24d335c2013-02-22 16:32:58 -08002162
David Hildenbrande1c158e2021-09-07 19:55:09 -07002163 if (nid != NUMA_NO_NODE)
2164 try_offline_node(nid);
Tang Chen60a5a192013-02-22 16:33:14 -08002165
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07002166 mem_hotplug_done();
Jia Heb4223a52020-08-11 18:32:20 -07002167 return 0;
Badari Pulavarty71088782008-10-18 20:25:58 -07002168}
David Hildenbrandd15e5922018-10-30 15:10:18 -07002169
Pavel Tatashineca499a2019-07-16 16:30:31 -07002170/**
Mel Gorman5640c9c2021-06-30 18:53:38 -07002171 * __remove_memory - Remove memory if every memory block is offline
Pavel Tatashineca499a2019-07-16 16:30:31 -07002172 * @start: physical address of the region to remove
2173 * @size: size of the region to remove
2174 *
2175 * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2176 * and online/offline operations before this call, as required by
2177 * try_offline_node().
2178 */
David Hildenbrande1c158e2021-09-07 19:55:09 -07002179void __remove_memory(u64 start, u64 size)
David Hildenbrandd15e5922018-10-30 15:10:18 -07002180{
Pavel Tatashineca499a2019-07-16 16:30:31 -07002181
2182 /*
Souptick Joarder29a90db2019-09-23 15:36:18 -07002183 * trigger BUG() if some memory is not offlined prior to calling this
Pavel Tatashineca499a2019-07-16 16:30:31 -07002184 * function
2185 */
David Hildenbrande1c158e2021-09-07 19:55:09 -07002186 if (try_remove_memory(start, size))
Pavel Tatashineca499a2019-07-16 16:30:31 -07002187 BUG();
2188}
2189
2190/*
2191 * Remove memory if every memory block is offline, otherwise return -EBUSY is
2192 * some memory is not offline
2193 */
David Hildenbrande1c158e2021-09-07 19:55:09 -07002194int remove_memory(u64 start, u64 size)
Pavel Tatashineca499a2019-07-16 16:30:31 -07002195{
2196 int rc;
2197
David Hildenbrandd15e5922018-10-30 15:10:18 -07002198 lock_device_hotplug();
David Hildenbrande1c158e2021-09-07 19:55:09 -07002199 rc = try_remove_memory(start, size);
David Hildenbrandd15e5922018-10-30 15:10:18 -07002200 unlock_device_hotplug();
Pavel Tatashineca499a2019-07-16 16:30:31 -07002201
2202 return rc;
David Hildenbrandd15e5922018-10-30 15:10:18 -07002203}
Badari Pulavarty71088782008-10-18 20:25:58 -07002204EXPORT_SYMBOL_GPL(remove_memory);
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002205
David Hildenbrand8dc4bb52020-11-12 14:38:13 +01002206static int try_offline_memory_block(struct memory_block *mem, void *arg)
2207{
2208 uint8_t online_type = MMOP_ONLINE_KERNEL;
2209 uint8_t **online_types = arg;
2210 struct page *page;
2211 int rc;
2212
2213 /*
2214 * Sense the online_type via the zone of the memory block. Offlining
2215 * with multiple zones within one memory block will be rejected
2216 * by offlining code ... so we don't care about that.
2217 */
2218 page = pfn_to_online_page(section_nr_to_pfn(mem->start_section_nr));
2219 if (page && zone_idx(page_zone(page)) == ZONE_MOVABLE)
2220 online_type = MMOP_ONLINE_MOVABLE;
2221
2222 rc = device_offline(&mem->dev);
2223 /*
2224 * Default is MMOP_OFFLINE - change it only if offlining succeeded,
2225 * so try_reonline_memory_block() can do the right thing.
2226 */
2227 if (!rc)
2228 **online_types = online_type;
2229
2230 (*online_types)++;
2231 /* Ignore if already offline. */
2232 return rc < 0 ? rc : 0;
2233}
2234
2235static int try_reonline_memory_block(struct memory_block *mem, void *arg)
2236{
2237 uint8_t **online_types = arg;
2238 int rc;
2239
2240 if (**online_types != MMOP_OFFLINE) {
2241 mem->online_type = **online_types;
2242 rc = device_online(&mem->dev);
2243 if (rc < 0)
2244 pr_warn("%s: Failed to re-online memory: %d",
2245 __func__, rc);
2246 }
2247
2248 /* Continue processing all remaining memory blocks. */
2249 (*online_types)++;
2250 return 0;
2251}
2252
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002253/*
David Hildenbrand8dc4bb52020-11-12 14:38:13 +01002254 * Try to offline and remove memory. Might take a long time to finish in case
2255 * memory is still in use. Primarily useful for memory devices that logically
2256 * unplugged all memory (so it's no longer in use) and want to offline + remove
2257 * that memory.
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002258 */
David Hildenbrande1c158e2021-09-07 19:55:09 -07002259int offline_and_remove_memory(u64 start, u64 size)
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002260{
David Hildenbrand8dc4bb52020-11-12 14:38:13 +01002261 const unsigned long mb_count = size / memory_block_size_bytes();
2262 uint8_t *online_types, *tmp;
2263 int rc;
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002264
2265 if (!IS_ALIGNED(start, memory_block_size_bytes()) ||
David Hildenbrand8dc4bb52020-11-12 14:38:13 +01002266 !IS_ALIGNED(size, memory_block_size_bytes()) || !size)
2267 return -EINVAL;
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002268
2269 /*
David Hildenbrand8dc4bb52020-11-12 14:38:13 +01002270 * We'll remember the old online type of each memory block, so we can
2271 * try to revert whatever we did when offlining one memory block fails
2272 * after offlining some others succeeded.
2273 */
2274 online_types = kmalloc_array(mb_count, sizeof(*online_types),
2275 GFP_KERNEL);
2276 if (!online_types)
2277 return -ENOMEM;
2278 /*
2279 * Initialize all states to MMOP_OFFLINE, so when we abort processing in
2280 * try_offline_memory_block(), we'll skip all unprocessed blocks in
2281 * try_reonline_memory_block().
2282 */
2283 memset(online_types, MMOP_OFFLINE, mb_count);
2284
2285 lock_device_hotplug();
2286
2287 tmp = online_types;
2288 rc = walk_memory_blocks(start, size, &tmp, try_offline_memory_block);
2289
2290 /*
2291 * In case we succeeded to offline all memory, remove it.
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002292 * This cannot fail as it cannot get onlined in the meantime.
2293 */
2294 if (!rc) {
David Hildenbrande1c158e2021-09-07 19:55:09 -07002295 rc = try_remove_memory(start, size);
David Hildenbrand8dc4bb52020-11-12 14:38:13 +01002296 if (rc)
2297 pr_err("%s: Failed to remove memory: %d", __func__, rc);
2298 }
2299
2300 /*
2301 * Rollback what we did. While memory onlining might theoretically fail
2302 * (nacked by a notifier), it barely ever happens.
2303 */
2304 if (rc) {
2305 tmp = online_types;
2306 walk_memory_blocks(start, size, &tmp,
2307 try_reonline_memory_block);
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002308 }
2309 unlock_device_hotplug();
2310
David Hildenbrand8dc4bb52020-11-12 14:38:13 +01002311 kfree(online_types);
David Hildenbrand08b3acd2020-05-07 16:01:32 +02002312 return rc;
2313}
2314EXPORT_SYMBOL_GPL(offline_and_remove_memory);
Rafael J. Wysockiaba6efc2013-06-01 22:24:07 +02002315#endif /* CONFIG_MEMORY_HOTREMOVE */