Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Virtio-mem device driver.
4 *
5 * Copyright Red Hat, Inc. 2020
6 *
7 * Author(s): David Hildenbrand <david@redhat.com>
8 */
9
10#include <linux/virtio.h>
11#include <linux/virtio_mem.h>
12#include <linux/workqueue.h>
13#include <linux/slab.h>
14#include <linux/module.h>
15#include <linux/mm.h>
16#include <linux/memory_hotplug.h>
17#include <linux/memory.h>
18#include <linux/hrtimer.h>
19#include <linux/crash_dump.h>
20#include <linux/mutex.h>
21#include <linux/bitmap.h>
22#include <linux/lockdep.h>
23#include <linux/log2.h>
24#include <linux/vmalloc.h>
25#include <linux/suspend.h>
26
27#include <acpi/acpi_numa.h>
28
29static bool unplug_online = true;
30module_param(unplug_online, bool, 0644);
31MODULE_PARM_DESC(unplug_online, "Try to unplug online memory");
32
33static bool force_bbm;
34module_param(force_bbm, bool, 0444);
35MODULE_PARM_DESC(force_bbm,
36 "Force Big Block Mode. Default is 0 (auto-selection)");
37
38static unsigned long bbm_block_size;
39module_param(bbm_block_size, ulong, 0444);
40MODULE_PARM_DESC(bbm_block_size,
41 "Big Block size in bytes. Default is 0 (auto-detection).");
42
43/*
44 * virtio-mem currently supports the following modes of operation:
45 *
46 * * Sub Block Mode (SBM): A Linux memory block spans 2..X subblocks (SB). The
47 * size of a Sub Block (SB) is determined based on the device block size, the
48 * pageblock size, and the maximum allocation granularity of the buddy.
49 * Subblocks within a Linux memory block might either be plugged or unplugged.
50 * Memory is added/removed to Linux MM in Linux memory block granularity.
51 *
52 * * Big Block Mode (BBM): A Big Block (BB) spans 1..X Linux memory blocks.
53 * Memory is added/removed to Linux MM in Big Block granularity.
54 *
55 * The mode is determined automatically based on the Linux memory block size
56 * and the device block size.
57 *
58 * User space / core MM (auto onlining) is responsible for onlining added
59 * Linux memory blocks - and for selecting a zone. Linux Memory Blocks are
60 * always onlined separately, and all memory within a Linux memory block is
61 * onlined to the same zone - virtio-mem relies on this behavior.
62 */
63
64/*
65 * State of a Linux memory block in SBM.
66 */
67enum virtio_mem_sbm_mb_state {
68 /* Unplugged, not added to Linux. Can be reused later. */
69 VIRTIO_MEM_SBM_MB_UNUSED = 0,
70 /* (Partially) plugged, not added to Linux. Error on add_memory(). */
71 VIRTIO_MEM_SBM_MB_PLUGGED,
72 /* Fully plugged, fully added to Linux, offline. */
73 VIRTIO_MEM_SBM_MB_OFFLINE,
74 /* Partially plugged, fully added to Linux, offline. */
75 VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
76 /* Fully plugged, fully added to Linux, onlined to a kernel zone. */
77 VIRTIO_MEM_SBM_MB_KERNEL,
78 /* Partially plugged, fully added to Linux, online to a kernel zone */
79 VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
80 /* Fully plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
81 VIRTIO_MEM_SBM_MB_MOVABLE,
82 /* Partially plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
83 VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
84 VIRTIO_MEM_SBM_MB_COUNT
85};
86
87/*
88 * State of a Big Block (BB) in BBM, covering 1..X Linux memory blocks.
89 */
90enum virtio_mem_bbm_bb_state {
91 /* Unplugged, not added to Linux. Can be reused later. */
92 VIRTIO_MEM_BBM_BB_UNUSED = 0,
93 /* Plugged, not added to Linux. Error on add_memory(). */
94 VIRTIO_MEM_BBM_BB_PLUGGED,
95 /* Plugged and added to Linux. */
96 VIRTIO_MEM_BBM_BB_ADDED,
97 /* All online parts are fake-offline, ready to remove. */
98 VIRTIO_MEM_BBM_BB_FAKE_OFFLINE,
99 VIRTIO_MEM_BBM_BB_COUNT
100};
101
102struct virtio_mem {
103 struct virtio_device *vdev;
104
105 /* We might first have to unplug all memory when starting up. */
106 bool unplug_all_required;
107
108 /* Workqueue that processes the plug/unplug requests. */
109 struct work_struct wq;
110 atomic_t wq_active;
111 atomic_t config_changed;
112
113 /* Virtqueue for guest->host requests. */
114 struct virtqueue *vq;
115
116 /* Wait for a host response to a guest request. */
117 wait_queue_head_t host_resp;
118
119 /* Space for one guest request and the host response. */
120 struct virtio_mem_req req;
121 struct virtio_mem_resp resp;
122
123 /* The current size of the device. */
124 uint64_t plugged_size;
125 /* The requested size of the device. */
126 uint64_t requested_size;
127
128 /* The device block size (for communicating with the device). */
129 uint64_t device_block_size;
130 /* The determined node id for all memory of the device. */
131 int nid;
132 /* Physical start address of the memory region. */
133 uint64_t addr;
134 /* Maximum region size in bytes. */
135 uint64_t region_size;
136
137 /* The parent resource for all memory added via this device. */
138 struct resource *parent_resource;
139 /*
140 * Copy of "System RAM (virtio_mem)" to be used for
141 * add_memory_driver_managed().
142 */
143 const char *resource_name;
144 /* Memory group identification. */
145 int mgid;
146
147 /*
148 * We don't want to add too much memory if it's not getting onlined,
149 * to avoid running OOM. Besides this threshold, we allow to have at
150 * least two offline blocks at a time (whatever is bigger).
151 */
152#define VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD (1024 * 1024 * 1024)
153 atomic64_t offline_size;
154 uint64_t offline_threshold;
155
156 /* If set, the driver is in SBM, otherwise in BBM. */
157 bool in_sbm;
158
159 union {
160 struct {
161 /* Id of the first memory block of this device. */
162 unsigned long first_mb_id;
163 /* Id of the last usable memory block of this device. */
164 unsigned long last_usable_mb_id;
165 /* Id of the next memory bock to prepare when needed. */
166 unsigned long next_mb_id;
167
168 /* The subblock size. */
169 uint64_t sb_size;
170 /* The number of subblocks per Linux memory block. */
171 uint32_t sbs_per_mb;
172
173 /*
174 * Some of the Linux memory blocks tracked as "partially
175 * plugged" are completely unplugged and can be offlined
176 * and removed -- which previously failed.
177 */
178 bool have_unplugged_mb;
179
180 /* Summary of all memory block states. */
181 unsigned long mb_count[VIRTIO_MEM_SBM_MB_COUNT];
182
183 /*
184 * One byte state per memory block. Allocated via
185 * vmalloc(). Resized (alloc+copy+free) on demand.
186 *
187 * With 128 MiB memory blocks, we have states for 512
188 * GiB of memory in one 4 KiB page.
189 */
190 uint8_t *mb_states;
191
192 /*
193 * Bitmap: one bit per subblock. Allocated similar to
194 * sbm.mb_states.
195 *
196 * A set bit means the corresponding subblock is
197 * plugged, otherwise it's unblocked.
198 *
199 * With 4 MiB subblocks, we manage 128 GiB of memory
200 * in one 4 KiB page.
201 */
202 unsigned long *sb_states;
203 } sbm;
204
205 struct {
206 /* Id of the first big block of this device. */
207 unsigned long first_bb_id;
208 /* Id of the last usable big block of this device. */
209 unsigned long last_usable_bb_id;
210 /* Id of the next device bock to prepare when needed. */
211 unsigned long next_bb_id;
212
213 /* Summary of all big block states. */
214 unsigned long bb_count[VIRTIO_MEM_BBM_BB_COUNT];
215
216 /* One byte state per big block. See sbm.mb_states. */
217 uint8_t *bb_states;
218
219 /* The block size used for plugging/adding/removing. */
220 uint64_t bb_size;
221 } bbm;
222 };
223
224 /*
225 * Mutex that protects the sbm.mb_count, sbm.mb_states,
226 * sbm.sb_states, bbm.bb_count, and bbm.bb_states
227 *
228 * When this lock is held the pointers can't change, ONLINE and
229 * OFFLINE blocks can't change the state and no subblocks will get
230 * plugged/unplugged.
231 *
232 * In kdump mode, used to serialize requests, last_block_addr and
233 * last_block_plugged.
234 */
235 struct mutex hotplug_mutex;
236 bool hotplug_active;
237
238 /* An error occurred we cannot handle - stop processing requests. */
239 bool broken;
240
241 /* Cached valued of is_kdump_kernel() when the device was probed. */
242 bool in_kdump;
243
244 /* The driver is being removed. */
245 spinlock_t removal_lock;
246 bool removing;
247
248 /* Timer for retrying to plug/unplug memory. */
249 struct hrtimer retry_timer;
250 unsigned int retry_timer_ms;
251#define VIRTIO_MEM_RETRY_TIMER_MIN_MS 50000
252#define VIRTIO_MEM_RETRY_TIMER_MAX_MS 300000
253
254 /* Memory notifier (online/offline events). */
255 struct notifier_block memory_notifier;
256
257 /* Notifier to block hibernation image storing/reloading. */
258 struct notifier_block pm_notifier;
259
260#ifdef CONFIG_PROC_VMCORE
261 /* vmcore callback for /proc/vmcore handling in kdump mode */
262 struct vmcore_cb vmcore_cb;
263 uint64_t last_block_addr;
264 bool last_block_plugged;
265#endif /* CONFIG_PROC_VMCORE */
266
267 /* Next device in the list of virtio-mem devices. */
268 struct list_head next;
269};
270
271/*
272 * We have to share a single online_page callback among all virtio-mem
273 * devices. We use RCU to iterate the list in the callback.
274 */
275static DEFINE_MUTEX(virtio_mem_mutex);
276static LIST_HEAD(virtio_mem_devices);
277
278static void virtio_mem_online_page_cb(struct page *page, unsigned int order);
279static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
280 unsigned long nr_pages);
281static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
282 unsigned long nr_pages);
283static void virtio_mem_retry(struct virtio_mem *vm);
284static int virtio_mem_create_resource(struct virtio_mem *vm);
285static void virtio_mem_delete_resource(struct virtio_mem *vm);
286
287/*
288 * Register a virtio-mem device so it will be considered for the online_page
289 * callback.
290 */
291static int register_virtio_mem_device(struct virtio_mem *vm)
292{
293 int rc = 0;
294
295 /* First device registers the callback. */
296 mutex_lock(&virtio_mem_mutex);
297 if (list_empty(&virtio_mem_devices))
298 rc = set_online_page_callback(&virtio_mem_online_page_cb);
299 if (!rc)
300 list_add_rcu(&vm->next, &virtio_mem_devices);
301 mutex_unlock(&virtio_mem_mutex);
302
303 return rc;
304}
305
306/*
307 * Unregister a virtio-mem device so it will no longer be considered for the
308 * online_page callback.
309 */
310static void unregister_virtio_mem_device(struct virtio_mem *vm)
311{
312 /* Last device unregisters the callback. */
313 mutex_lock(&virtio_mem_mutex);
314 list_del_rcu(&vm->next);
315 if (list_empty(&virtio_mem_devices))
316 restore_online_page_callback(&virtio_mem_online_page_cb);
317 mutex_unlock(&virtio_mem_mutex);
318
319 synchronize_rcu();
320}
321
322/*
323 * Calculate the memory block id of a given address.
324 */
325static unsigned long virtio_mem_phys_to_mb_id(unsigned long addr)
326{
327 return addr / memory_block_size_bytes();
328}
329
330/*
331 * Calculate the physical start address of a given memory block id.
332 */
333static unsigned long virtio_mem_mb_id_to_phys(unsigned long mb_id)
334{
335 return mb_id * memory_block_size_bytes();
336}
337
338/*
339 * Calculate the big block id of a given address.
340 */
341static unsigned long virtio_mem_phys_to_bb_id(struct virtio_mem *vm,
342 uint64_t addr)
343{
344 return addr / vm->bbm.bb_size;
345}
346
347/*
348 * Calculate the physical start address of a given big block id.
349 */
350static uint64_t virtio_mem_bb_id_to_phys(struct virtio_mem *vm,
351 unsigned long bb_id)
352{
353 return bb_id * vm->bbm.bb_size;
354}
355
356/*
357 * Calculate the subblock id of a given address.
358 */
359static unsigned long virtio_mem_phys_to_sb_id(struct virtio_mem *vm,
360 unsigned long addr)
361{
362 const unsigned long mb_id = virtio_mem_phys_to_mb_id(addr);
363 const unsigned long mb_addr = virtio_mem_mb_id_to_phys(mb_id);
364
365 return (addr - mb_addr) / vm->sbm.sb_size;
366}
367
368/*
369 * Set the state of a big block, taking care of the state counter.
370 */
371static void virtio_mem_bbm_set_bb_state(struct virtio_mem *vm,
372 unsigned long bb_id,
373 enum virtio_mem_bbm_bb_state state)
374{
375 const unsigned long idx = bb_id - vm->bbm.first_bb_id;
376 enum virtio_mem_bbm_bb_state old_state;
377
378 old_state = vm->bbm.bb_states[idx];
379 vm->bbm.bb_states[idx] = state;
380
381 BUG_ON(vm->bbm.bb_count[old_state] == 0);
382 vm->bbm.bb_count[old_state]--;
383 vm->bbm.bb_count[state]++;
384}
385
386/*
387 * Get the state of a big block.
388 */
389static enum virtio_mem_bbm_bb_state virtio_mem_bbm_get_bb_state(struct virtio_mem *vm,
390 unsigned long bb_id)
391{
392 return vm->bbm.bb_states[bb_id - vm->bbm.first_bb_id];
393}
394
395/*
396 * Prepare the big block state array for the next big block.
397 */
398static int virtio_mem_bbm_bb_states_prepare_next_bb(struct virtio_mem *vm)
399{
400 unsigned long old_bytes = vm->bbm.next_bb_id - vm->bbm.first_bb_id;
401 unsigned long new_bytes = old_bytes + 1;
402 int old_pages = PFN_UP(old_bytes);
403 int new_pages = PFN_UP(new_bytes);
404 uint8_t *new_array;
405
406 if (vm->bbm.bb_states && old_pages == new_pages)
407 return 0;
408
409 new_array = vzalloc(new_pages * PAGE_SIZE);
410 if (!new_array)
411 return -ENOMEM;
412
413 mutex_lock(&vm->hotplug_mutex);
414 if (vm->bbm.bb_states)
415 memcpy(new_array, vm->bbm.bb_states, old_pages * PAGE_SIZE);
416 vfree(vm->bbm.bb_states);
417 vm->bbm.bb_states = new_array;
418 mutex_unlock(&vm->hotplug_mutex);
419
420 return 0;
421}
422
423#define virtio_mem_bbm_for_each_bb(_vm, _bb_id, _state) \
424 for (_bb_id = vm->bbm.first_bb_id; \
425 _bb_id < vm->bbm.next_bb_id && _vm->bbm.bb_count[_state]; \
426 _bb_id++) \
427 if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
428
429#define virtio_mem_bbm_for_each_bb_rev(_vm, _bb_id, _state) \
430 for (_bb_id = vm->bbm.next_bb_id - 1; \
431 _bb_id >= vm->bbm.first_bb_id && _vm->bbm.bb_count[_state]; \
432 _bb_id--) \
433 if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
434
435/*
436 * Set the state of a memory block, taking care of the state counter.
437 */
438static void virtio_mem_sbm_set_mb_state(struct virtio_mem *vm,
439 unsigned long mb_id, uint8_t state)
440{
441 const unsigned long idx = mb_id - vm->sbm.first_mb_id;
442 uint8_t old_state;
443
444 old_state = vm->sbm.mb_states[idx];
445 vm->sbm.mb_states[idx] = state;
446
447 BUG_ON(vm->sbm.mb_count[old_state] == 0);
448 vm->sbm.mb_count[old_state]--;
449 vm->sbm.mb_count[state]++;
450}
451
452/*
453 * Get the state of a memory block.
454 */
455static uint8_t virtio_mem_sbm_get_mb_state(struct virtio_mem *vm,
456 unsigned long mb_id)
457{
458 const unsigned long idx = mb_id - vm->sbm.first_mb_id;
459
460 return vm->sbm.mb_states[idx];
461}
462
463/*
464 * Prepare the state array for the next memory block.
465 */
466static int virtio_mem_sbm_mb_states_prepare_next_mb(struct virtio_mem *vm)
467{
468 int old_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id);
469 int new_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id + 1);
470 uint8_t *new_array;
471
472 if (vm->sbm.mb_states && old_pages == new_pages)
473 return 0;
474
475 new_array = vzalloc(new_pages * PAGE_SIZE);
476 if (!new_array)
477 return -ENOMEM;
478
479 mutex_lock(&vm->hotplug_mutex);
480 if (vm->sbm.mb_states)
481 memcpy(new_array, vm->sbm.mb_states, old_pages * PAGE_SIZE);
482 vfree(vm->sbm.mb_states);
483 vm->sbm.mb_states = new_array;
484 mutex_unlock(&vm->hotplug_mutex);
485
486 return 0;
487}
488
489#define virtio_mem_sbm_for_each_mb(_vm, _mb_id, _state) \
490 for (_mb_id = _vm->sbm.first_mb_id; \
491 _mb_id < _vm->sbm.next_mb_id && _vm->sbm.mb_count[_state]; \
492 _mb_id++) \
493 if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
494
495#define virtio_mem_sbm_for_each_mb_rev(_vm, _mb_id, _state) \
496 for (_mb_id = _vm->sbm.next_mb_id - 1; \
497 _mb_id >= _vm->sbm.first_mb_id && _vm->sbm.mb_count[_state]; \
498 _mb_id--) \
499 if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
500
501/*
502 * Calculate the bit number in the subblock bitmap for the given subblock
503 * inside the given memory block.
504 */
505static int virtio_mem_sbm_sb_state_bit_nr(struct virtio_mem *vm,
506 unsigned long mb_id, int sb_id)
507{
508 return (mb_id - vm->sbm.first_mb_id) * vm->sbm.sbs_per_mb + sb_id;
509}
510
511/*
512 * Mark all selected subblocks plugged.
513 *
514 * Will not modify the state of the memory block.
515 */
516static void virtio_mem_sbm_set_sb_plugged(struct virtio_mem *vm,
517 unsigned long mb_id, int sb_id,
518 int count)
519{
520 const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
521
522 __bitmap_set(vm->sbm.sb_states, bit, count);
523}
524
525/*
526 * Mark all selected subblocks unplugged.
527 *
528 * Will not modify the state of the memory block.
529 */
530static void virtio_mem_sbm_set_sb_unplugged(struct virtio_mem *vm,
531 unsigned long mb_id, int sb_id,
532 int count)
533{
534 const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
535
536 __bitmap_clear(vm->sbm.sb_states, bit, count);
537}
538
539/*
540 * Test if all selected subblocks are plugged.
541 */
542static bool virtio_mem_sbm_test_sb_plugged(struct virtio_mem *vm,
543 unsigned long mb_id, int sb_id,
544 int count)
545{
546 const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
547
548 if (count == 1)
549 return test_bit(bit, vm->sbm.sb_states);
550
551 /* TODO: Helper similar to bitmap_set() */
552 return find_next_zero_bit(vm->sbm.sb_states, bit + count, bit) >=
553 bit + count;
554}
555
556/*
557 * Test if all selected subblocks are unplugged.
558 */
559static bool virtio_mem_sbm_test_sb_unplugged(struct virtio_mem *vm,
560 unsigned long mb_id, int sb_id,
561 int count)
562{
563 const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
564
565 /* TODO: Helper similar to bitmap_set() */
566 return find_next_bit(vm->sbm.sb_states, bit + count, bit) >=
567 bit + count;
568}
569
570/*
571 * Find the first unplugged subblock. Returns vm->sbm.sbs_per_mb in case there is
572 * none.
573 */
574static int virtio_mem_sbm_first_unplugged_sb(struct virtio_mem *vm,
575 unsigned long mb_id)
576{
577 const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, 0);
578
579 return find_next_zero_bit(vm->sbm.sb_states,
580 bit + vm->sbm.sbs_per_mb, bit) - bit;
581}
582
583/*
584 * Prepare the subblock bitmap for the next memory block.
585 */
586static int virtio_mem_sbm_sb_states_prepare_next_mb(struct virtio_mem *vm)
587{
588 const unsigned long old_nb_mb = vm->sbm.next_mb_id - vm->sbm.first_mb_id;
589 const unsigned long old_nb_bits = old_nb_mb * vm->sbm.sbs_per_mb;
590 const unsigned long new_nb_bits = (old_nb_mb + 1) * vm->sbm.sbs_per_mb;
591 int old_pages = PFN_UP(BITS_TO_LONGS(old_nb_bits) * sizeof(long));
592 int new_pages = PFN_UP(BITS_TO_LONGS(new_nb_bits) * sizeof(long));
593 unsigned long *new_bitmap, *old_bitmap;
594
595 if (vm->sbm.sb_states && old_pages == new_pages)
596 return 0;
597
598 new_bitmap = vzalloc(new_pages * PAGE_SIZE);
599 if (!new_bitmap)
600 return -ENOMEM;
601
602 mutex_lock(&vm->hotplug_mutex);
603 if (vm->sbm.sb_states)
604 memcpy(new_bitmap, vm->sbm.sb_states, old_pages * PAGE_SIZE);
605
606 old_bitmap = vm->sbm.sb_states;
607 vm->sbm.sb_states = new_bitmap;
608 mutex_unlock(&vm->hotplug_mutex);
609
610 vfree(old_bitmap);
611 return 0;
612}
613
614/*
615 * Test if we could add memory without creating too much offline memory -
616 * to avoid running OOM if memory is getting onlined deferred.
617 */
618static bool virtio_mem_could_add_memory(struct virtio_mem *vm, uint64_t size)
619{
620 if (WARN_ON_ONCE(size > vm->offline_threshold))
621 return false;
622
623 return atomic64_read(&vm->offline_size) + size <= vm->offline_threshold;
624}
625
626/*
627 * Try adding memory to Linux. Will usually only fail if out of memory.
628 *
629 * Must not be called with the vm->hotplug_mutex held (possible deadlock with
630 * onlining code).
631 *
632 * Will not modify the state of memory blocks in virtio-mem.
633 */
634static int virtio_mem_add_memory(struct virtio_mem *vm, uint64_t addr,
635 uint64_t size)
636{
637 int rc;
638
639 /*
640 * When force-unloading the driver and we still have memory added to
641 * Linux, the resource name has to stay.
642 */
643 if (!vm->resource_name) {
644 vm->resource_name = kstrdup_const("System RAM (virtio_mem)",
645 GFP_KERNEL);
646 if (!vm->resource_name)
647 return -ENOMEM;
648 }
649
650 dev_dbg(&vm->vdev->dev, "adding memory: 0x%llx - 0x%llx\n", addr,
651 addr + size - 1);
652 /* Memory might get onlined immediately. */
653 atomic64_add(size, &vm->offline_size);
654 rc = add_memory_driver_managed(vm->mgid, addr, size, vm->resource_name,
655 MHP_MERGE_RESOURCE | MHP_NID_IS_MGID);
656 if (rc) {
657 atomic64_sub(size, &vm->offline_size);
658 dev_warn(&vm->vdev->dev, "adding memory failed: %d\n", rc);
659 /*
660 * TODO: Linux MM does not properly clean up yet in all cases
661 * where adding of memory failed - especially on -ENOMEM.
662 */
663 }
664 return rc;
665}
666
667/*
668 * See virtio_mem_add_memory(): Try adding a single Linux memory block.
669 */
670static int virtio_mem_sbm_add_mb(struct virtio_mem *vm, unsigned long mb_id)
671{
672 const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
673 const uint64_t size = memory_block_size_bytes();
674
675 return virtio_mem_add_memory(vm, addr, size);
676}
677
678/*
679 * See virtio_mem_add_memory(): Try adding a big block.
680 */
681static int virtio_mem_bbm_add_bb(struct virtio_mem *vm, unsigned long bb_id)
682{
683 const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
684 const uint64_t size = vm->bbm.bb_size;
685
686 return virtio_mem_add_memory(vm, addr, size);
687}
688
689/*
690 * Try removing memory from Linux. Will only fail if memory blocks aren't
691 * offline.
692 *
693 * Must not be called with the vm->hotplug_mutex held (possible deadlock with
694 * onlining code).
695 *
696 * Will not modify the state of memory blocks in virtio-mem.
697 */
698static int virtio_mem_remove_memory(struct virtio_mem *vm, uint64_t addr,
699 uint64_t size)
700{
701 int rc;
702
703 dev_dbg(&vm->vdev->dev, "removing memory: 0x%llx - 0x%llx\n", addr,
704 addr + size - 1);
705 rc = remove_memory(addr, size);
706 if (!rc) {
707 atomic64_sub(size, &vm->offline_size);
708 /*
709 * We might have freed up memory we can now unplug, retry
710 * immediately instead of waiting.
711 */
712 virtio_mem_retry(vm);
713 } else {
714 dev_dbg(&vm->vdev->dev, "removing memory failed: %d\n", rc);
715 }
716 return rc;
717}
718
719/*
720 * See virtio_mem_remove_memory(): Try removing a single Linux memory block.
721 */
722static int virtio_mem_sbm_remove_mb(struct virtio_mem *vm, unsigned long mb_id)
723{
724 const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
725 const uint64_t size = memory_block_size_bytes();
726
727 return virtio_mem_remove_memory(vm, addr, size);
728}
729
730/*
731 * Try offlining and removing memory from Linux.
732 *
733 * Must not be called with the vm->hotplug_mutex held (possible deadlock with
734 * onlining code).
735 *
736 * Will not modify the state of memory blocks in virtio-mem.
737 */
738static int virtio_mem_offline_and_remove_memory(struct virtio_mem *vm,
739 uint64_t addr,
740 uint64_t size)
741{
742 int rc;
743
744 dev_dbg(&vm->vdev->dev,
745 "offlining and removing memory: 0x%llx - 0x%llx\n", addr,
746 addr + size - 1);
747
748 rc = offline_and_remove_memory(addr, size);
749 if (!rc) {
750 atomic64_sub(size, &vm->offline_size);
751 /*
752 * We might have freed up memory we can now unplug, retry
753 * immediately instead of waiting.
754 */
755 virtio_mem_retry(vm);
756 return 0;
757 }
758 dev_dbg(&vm->vdev->dev, "offlining and removing memory failed: %d\n", rc);
759 /*
760 * We don't really expect this to fail, because we fake-offlined all
761 * memory already. But it could fail in corner cases.
762 */
763 WARN_ON_ONCE(rc != -ENOMEM && rc != -EBUSY);
764 return rc == -ENOMEM ? -ENOMEM : -EBUSY;
765}
766
767/*
768 * See virtio_mem_offline_and_remove_memory(): Try offlining and removing
769 * a single Linux memory block.
770 */
771static int virtio_mem_sbm_offline_and_remove_mb(struct virtio_mem *vm,
772 unsigned long mb_id)
773{
774 const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
775 const uint64_t size = memory_block_size_bytes();
776
777 return virtio_mem_offline_and_remove_memory(vm, addr, size);
778}
779
780/*
781 * Try (offlining and) removing memory from Linux in case all subblocks are
782 * unplugged. Can be called on online and offline memory blocks.
783 *
784 * May modify the state of memory blocks in virtio-mem.
785 */
786static int virtio_mem_sbm_try_remove_unplugged_mb(struct virtio_mem *vm,
787 unsigned long mb_id)
788{
789 int rc;
790
791 /*
792 * Once all subblocks of a memory block were unplugged, offline and
793 * remove it.
794 */
795 if (!virtio_mem_sbm_test_sb_unplugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
796 return 0;
797
798 /* offline_and_remove_memory() works for online and offline memory. */
799 mutex_unlock(&vm->hotplug_mutex);
800 rc = virtio_mem_sbm_offline_and_remove_mb(vm, mb_id);
801 mutex_lock(&vm->hotplug_mutex);
802 if (!rc)
803 virtio_mem_sbm_set_mb_state(vm, mb_id,
804 VIRTIO_MEM_SBM_MB_UNUSED);
805 return rc;
806}
807
808/*
809 * See virtio_mem_offline_and_remove_memory(): Try to offline and remove a
810 * all Linux memory blocks covered by the big block.
811 */
812static int virtio_mem_bbm_offline_and_remove_bb(struct virtio_mem *vm,
813 unsigned long bb_id)
814{
815 const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
816 const uint64_t size = vm->bbm.bb_size;
817
818 return virtio_mem_offline_and_remove_memory(vm, addr, size);
819}
820
821/*
822 * Trigger the workqueue so the device can perform its magic.
823 */
824static void virtio_mem_retry(struct virtio_mem *vm)
825{
826 unsigned long flags;
827
828 spin_lock_irqsave(&vm->removal_lock, flags);
829 if (!vm->removing)
830 queue_work(system_freezable_wq, &vm->wq);
831 spin_unlock_irqrestore(&vm->removal_lock, flags);
832}
833
834static int virtio_mem_translate_node_id(struct virtio_mem *vm, uint16_t node_id)
835{
836 int node = NUMA_NO_NODE;
837
838#if defined(CONFIG_ACPI_NUMA)
839 if (virtio_has_feature(vm->vdev, VIRTIO_MEM_F_ACPI_PXM))
840 node = pxm_to_node(node_id);
841#endif
842 return node;
843}
844
845/*
846 * Test if a virtio-mem device overlaps with the given range. Can be called
847 * from (notifier) callbacks lockless.
848 */
849static bool virtio_mem_overlaps_range(struct virtio_mem *vm, uint64_t start,
850 uint64_t size)
851{
852 return start < vm->addr + vm->region_size && vm->addr < start + size;
853}
854
855/*
856 * Test if a virtio-mem device contains a given range. Can be called from
857 * (notifier) callbacks lockless.
858 */
859static bool virtio_mem_contains_range(struct virtio_mem *vm, uint64_t start,
860 uint64_t size)
861{
862 return start >= vm->addr && start + size <= vm->addr + vm->region_size;
863}
864
865static int virtio_mem_sbm_notify_going_online(struct virtio_mem *vm,
866 unsigned long mb_id)
867{
868 switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
869 case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
870 case VIRTIO_MEM_SBM_MB_OFFLINE:
871 return NOTIFY_OK;
872 default:
873 break;
874 }
875 dev_warn_ratelimited(&vm->vdev->dev,
876 "memory block onlining denied\n");
877 return NOTIFY_BAD;
878}
879
880static void virtio_mem_sbm_notify_offline(struct virtio_mem *vm,
881 unsigned long mb_id)
882{
883 switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
884 case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
885 case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
886 virtio_mem_sbm_set_mb_state(vm, mb_id,
887 VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
888 break;
889 case VIRTIO_MEM_SBM_MB_KERNEL:
890 case VIRTIO_MEM_SBM_MB_MOVABLE:
891 virtio_mem_sbm_set_mb_state(vm, mb_id,
892 VIRTIO_MEM_SBM_MB_OFFLINE);
893 break;
894 default:
895 BUG();
896 break;
897 }
898}
899
900static void virtio_mem_sbm_notify_online(struct virtio_mem *vm,
901 unsigned long mb_id,
902 unsigned long start_pfn)
903{
904 const bool is_movable = is_zone_movable_page(pfn_to_page(start_pfn));
905 int new_state;
906
907 switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
908 case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
909 new_state = VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL;
910 if (is_movable)
911 new_state = VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL;
912 break;
913 case VIRTIO_MEM_SBM_MB_OFFLINE:
914 new_state = VIRTIO_MEM_SBM_MB_KERNEL;
915 if (is_movable)
916 new_state = VIRTIO_MEM_SBM_MB_MOVABLE;
917 break;
918 default:
919 BUG();
920 break;
921 }
922 virtio_mem_sbm_set_mb_state(vm, mb_id, new_state);
923}
924
925static void virtio_mem_sbm_notify_going_offline(struct virtio_mem *vm,
926 unsigned long mb_id)
927{
928 const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
929 unsigned long pfn;
930 int sb_id;
931
932 for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
933 if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
934 continue;
935 pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
936 sb_id * vm->sbm.sb_size);
937 virtio_mem_fake_offline_going_offline(pfn, nr_pages);
938 }
939}
940
941static void virtio_mem_sbm_notify_cancel_offline(struct virtio_mem *vm,
942 unsigned long mb_id)
943{
944 const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
945 unsigned long pfn;
946 int sb_id;
947
948 for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
949 if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
950 continue;
951 pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
952 sb_id * vm->sbm.sb_size);
953 virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
954 }
955}
956
957static void virtio_mem_bbm_notify_going_offline(struct virtio_mem *vm,
958 unsigned long bb_id,
959 unsigned long pfn,
960 unsigned long nr_pages)
961{
962 /*
963 * When marked as "fake-offline", all online memory of this device block
964 * is allocated by us. Otherwise, we don't have any memory allocated.
965 */
966 if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
967 VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
968 return;
969 virtio_mem_fake_offline_going_offline(pfn, nr_pages);
970}
971
972static void virtio_mem_bbm_notify_cancel_offline(struct virtio_mem *vm,
973 unsigned long bb_id,
974 unsigned long pfn,
975 unsigned long nr_pages)
976{
977 if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
978 VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
979 return;
980 virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
981}
982
983/*
984 * This callback will either be called synchronously from add_memory() or
985 * asynchronously (e.g., triggered via user space). We have to be careful
986 * with locking when calling add_memory().
987 */
988static int virtio_mem_memory_notifier_cb(struct notifier_block *nb,
989 unsigned long action, void *arg)
990{
991 struct virtio_mem *vm = container_of(nb, struct virtio_mem,
992 memory_notifier);
993 struct memory_notify *mhp = arg;
994 const unsigned long start = PFN_PHYS(mhp->start_pfn);
995 const unsigned long size = PFN_PHYS(mhp->nr_pages);
996 int rc = NOTIFY_OK;
997 unsigned long id;
998
999 if (!virtio_mem_overlaps_range(vm, start, size))
1000 return NOTIFY_DONE;
1001
1002 if (vm->in_sbm) {
1003 id = virtio_mem_phys_to_mb_id(start);
1004 /*
1005 * In SBM, we add memory in separate memory blocks - we expect
1006 * it to be onlined/offlined in the same granularity. Bail out
1007 * if this ever changes.
1008 */
1009 if (WARN_ON_ONCE(size != memory_block_size_bytes() ||
1010 !IS_ALIGNED(start, memory_block_size_bytes())))
1011 return NOTIFY_BAD;
1012 } else {
1013 id = virtio_mem_phys_to_bb_id(vm, start);
1014 /*
1015 * In BBM, we only care about onlining/offlining happening
1016 * within a single big block, we don't care about the
1017 * actual granularity as we don't track individual Linux
1018 * memory blocks.
1019 */
1020 if (WARN_ON_ONCE(id != virtio_mem_phys_to_bb_id(vm, start + size - 1)))
1021 return NOTIFY_BAD;
1022 }
1023
1024 /*
1025 * Avoid circular locking lockdep warnings. We lock the mutex
1026 * e.g., in MEM_GOING_ONLINE and unlock it in MEM_ONLINE. The
1027 * blocking_notifier_call_chain() has it's own lock, which gets unlocked
1028 * between both notifier calls and will bail out. False positive.
1029 */
1030 lockdep_off();
1031
1032 switch (action) {
1033 case MEM_GOING_OFFLINE:
1034 mutex_lock(&vm->hotplug_mutex);
1035 if (vm->removing) {
1036 rc = notifier_from_errno(-EBUSY);
1037 mutex_unlock(&vm->hotplug_mutex);
1038 break;
1039 }
1040 vm->hotplug_active = true;
1041 if (vm->in_sbm)
1042 virtio_mem_sbm_notify_going_offline(vm, id);
1043 else
1044 virtio_mem_bbm_notify_going_offline(vm, id,
1045 mhp->start_pfn,
1046 mhp->nr_pages);
1047 break;
1048 case MEM_GOING_ONLINE:
1049 mutex_lock(&vm->hotplug_mutex);
1050 if (vm->removing) {
1051 rc = notifier_from_errno(-EBUSY);
1052 mutex_unlock(&vm->hotplug_mutex);
1053 break;
1054 }
1055 vm->hotplug_active = true;
1056 if (vm->in_sbm)
1057 rc = virtio_mem_sbm_notify_going_online(vm, id);
1058 break;
1059 case MEM_OFFLINE:
1060 if (vm->in_sbm)
1061 virtio_mem_sbm_notify_offline(vm, id);
1062
1063 atomic64_add(size, &vm->offline_size);
1064 /*
1065 * Trigger the workqueue. Now that we have some offline memory,
1066 * maybe we can handle pending unplug requests.
1067 */
1068 if (!unplug_online)
1069 virtio_mem_retry(vm);
1070
1071 vm->hotplug_active = false;
1072 mutex_unlock(&vm->hotplug_mutex);
1073 break;
1074 case MEM_ONLINE:
1075 if (vm->in_sbm)
1076 virtio_mem_sbm_notify_online(vm, id, mhp->start_pfn);
1077
1078 atomic64_sub(size, &vm->offline_size);
1079 /*
1080 * Start adding more memory once we onlined half of our
1081 * threshold. Don't trigger if it's possibly due to our actipn
1082 * (e.g., us adding memory which gets onlined immediately from
1083 * the core).
1084 */
1085 if (!atomic_read(&vm->wq_active) &&
1086 virtio_mem_could_add_memory(vm, vm->offline_threshold / 2))
1087 virtio_mem_retry(vm);
1088
1089 vm->hotplug_active = false;
1090 mutex_unlock(&vm->hotplug_mutex);
1091 break;
1092 case MEM_CANCEL_OFFLINE:
1093 if (!vm->hotplug_active)
1094 break;
1095 if (vm->in_sbm)
1096 virtio_mem_sbm_notify_cancel_offline(vm, id);
1097 else
1098 virtio_mem_bbm_notify_cancel_offline(vm, id,
1099 mhp->start_pfn,
1100 mhp->nr_pages);
1101 vm->hotplug_active = false;
1102 mutex_unlock(&vm->hotplug_mutex);
1103 break;
1104 case MEM_CANCEL_ONLINE:
1105 if (!vm->hotplug_active)
1106 break;
1107 vm->hotplug_active = false;
1108 mutex_unlock(&vm->hotplug_mutex);
1109 break;
1110 default:
1111 break;
1112 }
1113
1114 lockdep_on();
1115
1116 return rc;
1117}
1118
1119static int virtio_mem_pm_notifier_cb(struct notifier_block *nb,
1120 unsigned long action, void *arg)
1121{
1122 struct virtio_mem *vm = container_of(nb, struct virtio_mem,
1123 pm_notifier);
1124 switch (action) {
1125 case PM_HIBERNATION_PREPARE:
1126 case PM_RESTORE_PREPARE:
1127 /*
1128 * When restarting the VM, all memory is unplugged. Don't
1129 * allow to hibernate and restore from an image.
1130 */
1131 dev_err(&vm->vdev->dev, "hibernation is not supported.\n");
1132 return NOTIFY_BAD;
1133 default:
1134 return NOTIFY_OK;
1135 }
1136}
1137
1138/*
1139 * Set a range of pages PG_offline. Remember pages that were never onlined
1140 * (via generic_online_page()) using PageDirty().
1141 */
1142static void virtio_mem_set_fake_offline(unsigned long pfn,
1143 unsigned long nr_pages, bool onlined)
1144{
1145 page_offline_begin();
1146 for (; nr_pages--; pfn++) {
1147 struct page *page = pfn_to_page(pfn);
1148
1149 if (!onlined)
1150 /*
1151 * Pages that have not been onlined yet were initialized
1152 * to PageOffline(). Remember that we have to route them
1153 * through generic_online_page().
1154 */
1155 SetPageDirty(page);
1156 else
1157 __SetPageOffline(page);
1158 VM_WARN_ON_ONCE(!PageOffline(page));
1159 }
1160 page_offline_end();
1161}
1162
1163/*
1164 * Clear PG_offline from a range of pages. If the pages were never onlined,
1165 * (via generic_online_page()), clear PageDirty().
1166 */
1167static void virtio_mem_clear_fake_offline(unsigned long pfn,
1168 unsigned long nr_pages, bool onlined)
1169{
1170 for (; nr_pages--; pfn++) {
1171 struct page *page = pfn_to_page(pfn);
1172
1173 if (!onlined)
1174 /* generic_online_page() will clear PageOffline(). */
1175 ClearPageDirty(page);
1176 else
1177 __ClearPageOffline(page);
1178 }
1179}
1180
1181/*
1182 * Release a range of fake-offline pages to the buddy, effectively
1183 * fake-onlining them.
1184 */
1185static void virtio_mem_fake_online(unsigned long pfn, unsigned long nr_pages)
1186{
1187 unsigned long order = MAX_PAGE_ORDER;
1188 unsigned long i;
1189
1190 /*
1191 * We might get called for ranges that don't cover properly aligned
1192 * MAX_PAGE_ORDER pages; however, we can only online properly aligned
1193 * pages with an order of MAX_PAGE_ORDER at maximum.
1194 */
1195 while (!IS_ALIGNED(pfn | nr_pages, 1 << order))
1196 order--;
1197
1198 for (i = 0; i < nr_pages; i += 1 << order) {
1199 struct page *page = pfn_to_page(pfn + i);
1200
1201 /*
1202 * If the page is PageDirty(), it was kept fake-offline when
1203 * onlining the memory block. Otherwise, it was allocated
1204 * using alloc_contig_range(). All pages in a subblock are
1205 * alike.
1206 */
1207 if (PageDirty(page)) {
1208 virtio_mem_clear_fake_offline(pfn + i, 1 << order, false);
1209 generic_online_page(page, order);
1210 } else {
1211 virtio_mem_clear_fake_offline(pfn + i, 1 << order, true);
1212 free_contig_range(pfn + i, 1 << order);
1213 adjust_managed_page_count(page, 1 << order);
1214 }
1215 }
1216}
1217
1218/*
1219 * Try to allocate a range, marking pages fake-offline, effectively
1220 * fake-offlining them.
1221 */
1222static int virtio_mem_fake_offline(struct virtio_mem *vm, unsigned long pfn,
1223 unsigned long nr_pages)
1224{
1225 const bool is_movable = is_zone_movable_page(pfn_to_page(pfn));
1226 int rc, retry_count;
1227
1228 /*
1229 * TODO: We want an alloc_contig_range() mode that tries to allocate
1230 * harder (e.g., dealing with temporarily pinned pages, PCP), especially
1231 * with ZONE_MOVABLE. So for now, retry a couple of times with
1232 * ZONE_MOVABLE before giving up - because that zone is supposed to give
1233 * some guarantees.
1234 */
1235 for (retry_count = 0; retry_count < 5; retry_count++) {
1236 /*
1237 * If the config changed, stop immediately and go back to the
1238 * main loop: avoid trying to keep unplugging if the device
1239 * might have decided to not remove any more memory.
1240 */
1241 if (atomic_read(&vm->config_changed))
1242 return -EAGAIN;
1243
1244 rc = alloc_contig_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE,
1245 GFP_KERNEL);
1246 if (rc == -ENOMEM)
1247 /* whoops, out of memory */
1248 return rc;
1249 else if (rc && !is_movable)
1250 break;
1251 else if (rc)
1252 continue;
1253
1254 virtio_mem_set_fake_offline(pfn, nr_pages, true);
1255 adjust_managed_page_count(pfn_to_page(pfn), -nr_pages);
1256 return 0;
1257 }
1258
1259 return -EBUSY;
1260}
1261
1262/*
1263 * Handle fake-offline pages when memory is going offline - such that the
1264 * pages can be skipped by mm-core when offlining.
1265 */
1266static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
1267 unsigned long nr_pages)
1268{
1269 struct page *page;
1270 unsigned long i;
1271
1272 /* Drop our reference to the pages so the memory can get offlined. */
1273 for (i = 0; i < nr_pages; i++) {
1274 page = pfn_to_page(pfn + i);
1275 if (WARN_ON(!page_ref_dec_and_test(page)))
1276 dump_page(page, "fake-offline page referenced");
1277 }
1278}
1279
1280/*
1281 * Handle fake-offline pages when memory offlining is canceled - to undo
1282 * what we did in virtio_mem_fake_offline_going_offline().
1283 */
1284static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
1285 unsigned long nr_pages)
1286{
1287 unsigned long i;
1288
1289 /*
1290 * Get the reference again that we dropped via page_ref_dec_and_test()
1291 * when going offline.
1292 */
1293 for (i = 0; i < nr_pages; i++)
1294 page_ref_inc(pfn_to_page(pfn + i));
1295}
1296
1297static void virtio_mem_online_page(struct virtio_mem *vm,
1298 struct page *page, unsigned int order)
1299{
1300 const unsigned long start = page_to_phys(page);
1301 const unsigned long end = start + PFN_PHYS(1 << order);
1302 unsigned long addr, next, id, sb_id, count;
1303 bool do_online;
1304
1305 /*
1306 * We can get called with any order up to MAX_PAGE_ORDER. If our subblock
1307 * size is smaller than that and we have a mixture of plugged and
1308 * unplugged subblocks within such a page, we have to process in
1309 * smaller granularity. In that case we'll adjust the order exactly once
1310 * within the loop.
1311 */
1312 for (addr = start; addr < end; ) {
1313 next = addr + PFN_PHYS(1 << order);
1314
1315 if (vm->in_sbm) {
1316 id = virtio_mem_phys_to_mb_id(addr);
1317 sb_id = virtio_mem_phys_to_sb_id(vm, addr);
1318 count = virtio_mem_phys_to_sb_id(vm, next - 1) - sb_id + 1;
1319
1320 if (virtio_mem_sbm_test_sb_plugged(vm, id, sb_id, count)) {
1321 /* Fully plugged. */
1322 do_online = true;
1323 } else if (count == 1 ||
1324 virtio_mem_sbm_test_sb_unplugged(vm, id, sb_id, count)) {
1325 /* Fully unplugged. */
1326 do_online = false;
1327 } else {
1328 /*
1329 * Mixture, process sub-blocks instead. This
1330 * will be at least the size of a pageblock.
1331 * We'll run into this case exactly once.
1332 */
1333 order = ilog2(vm->sbm.sb_size) - PAGE_SHIFT;
1334 do_online = virtio_mem_sbm_test_sb_plugged(vm, id, sb_id, 1);
1335 continue;
1336 }
1337 } else {
1338 /*
1339 * If the whole block is marked fake offline, keep
1340 * everything that way.
1341 */
1342 id = virtio_mem_phys_to_bb_id(vm, addr);
1343 do_online = virtio_mem_bbm_get_bb_state(vm, id) !=
1344 VIRTIO_MEM_BBM_BB_FAKE_OFFLINE;
1345 }
1346
1347 if (do_online)
1348 generic_online_page(pfn_to_page(PFN_DOWN(addr)), order);
1349 else
1350 virtio_mem_set_fake_offline(PFN_DOWN(addr), 1 << order,
1351 false);
1352 addr = next;
1353 }
1354}
1355
1356static void virtio_mem_online_page_cb(struct page *page, unsigned int order)
1357{
1358 const unsigned long addr = page_to_phys(page);
1359 struct virtio_mem *vm;
1360
1361 rcu_read_lock();
1362 list_for_each_entry_rcu(vm, &virtio_mem_devices, next) {
1363 /*
1364 * Pages we're onlining will never cross memory blocks and,
1365 * therefore, not virtio-mem devices.
1366 */
1367 if (!virtio_mem_contains_range(vm, addr, PFN_PHYS(1 << order)))
1368 continue;
1369
1370 /*
1371 * virtio_mem_set_fake_offline() might sleep. We can safely
1372 * drop the RCU lock at this point because the device
1373 * cannot go away. See virtio_mem_remove() how races
1374 * between memory onlining and device removal are handled.
1375 */
1376 rcu_read_unlock();
1377
1378 virtio_mem_online_page(vm, page, order);
1379 return;
1380 }
1381 rcu_read_unlock();
1382
1383 /* not virtio-mem memory, but e.g., a DIMM. online it */
1384 generic_online_page(page, order);
1385}
1386
1387static uint64_t virtio_mem_send_request(struct virtio_mem *vm,
1388 const struct virtio_mem_req *req)
1389{
1390 struct scatterlist *sgs[2], sg_req, sg_resp;
1391 unsigned int len;
1392 int rc;
1393
1394 /* don't use the request residing on the stack (vaddr) */
1395 vm->req = *req;
1396
1397 /* out: buffer for request */
1398 sg_init_one(&sg_req, &vm->req, sizeof(vm->req));
1399 sgs[0] = &sg_req;
1400
1401 /* in: buffer for response */
1402 sg_init_one(&sg_resp, &vm->resp, sizeof(vm->resp));
1403 sgs[1] = &sg_resp;
1404
1405 rc = virtqueue_add_sgs(vm->vq, sgs, 1, 1, vm, GFP_KERNEL);
1406 if (rc < 0)
1407 return rc;
1408
1409 virtqueue_kick(vm->vq);
1410
1411 /* wait for a response */
1412 wait_event(vm->host_resp, virtqueue_get_buf(vm->vq, &len));
1413
1414 return virtio16_to_cpu(vm->vdev, vm->resp.type);
1415}
1416
1417static int virtio_mem_send_plug_request(struct virtio_mem *vm, uint64_t addr,
1418 uint64_t size)
1419{
1420 const uint64_t nb_vm_blocks = size / vm->device_block_size;
1421 const struct virtio_mem_req req = {
1422 .type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_PLUG),
1423 .u.plug.addr = cpu_to_virtio64(vm->vdev, addr),
1424 .u.plug.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
1425 };
1426 int rc = -ENOMEM;
1427
1428 if (atomic_read(&vm->config_changed))
1429 return -EAGAIN;
1430
1431 dev_dbg(&vm->vdev->dev, "plugging memory: 0x%llx - 0x%llx\n", addr,
1432 addr + size - 1);
1433
1434 switch (virtio_mem_send_request(vm, &req)) {
1435 case VIRTIO_MEM_RESP_ACK:
1436 vm->plugged_size += size;
1437 return 0;
1438 case VIRTIO_MEM_RESP_NACK:
1439 rc = -EAGAIN;
1440 break;
1441 case VIRTIO_MEM_RESP_BUSY:
1442 rc = -ETXTBSY;
1443 break;
1444 case VIRTIO_MEM_RESP_ERROR:
1445 rc = -EINVAL;
1446 break;
1447 default:
1448 break;
1449 }
1450
1451 dev_dbg(&vm->vdev->dev, "plugging memory failed: %d\n", rc);
1452 return rc;
1453}
1454
1455static int virtio_mem_send_unplug_request(struct virtio_mem *vm, uint64_t addr,
1456 uint64_t size)
1457{
1458 const uint64_t nb_vm_blocks = size / vm->device_block_size;
1459 const struct virtio_mem_req req = {
1460 .type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_UNPLUG),
1461 .u.unplug.addr = cpu_to_virtio64(vm->vdev, addr),
1462 .u.unplug.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
1463 };
1464 int rc = -ENOMEM;
1465
1466 if (atomic_read(&vm->config_changed))
1467 return -EAGAIN;
1468
1469 dev_dbg(&vm->vdev->dev, "unplugging memory: 0x%llx - 0x%llx\n", addr,
1470 addr + size - 1);
1471
1472 switch (virtio_mem_send_request(vm, &req)) {
1473 case VIRTIO_MEM_RESP_ACK:
1474 vm->plugged_size -= size;
1475 return 0;
1476 case VIRTIO_MEM_RESP_BUSY:
1477 rc = -ETXTBSY;
1478 break;
1479 case VIRTIO_MEM_RESP_ERROR:
1480 rc = -EINVAL;
1481 break;
1482 default:
1483 break;
1484 }
1485
1486 dev_dbg(&vm->vdev->dev, "unplugging memory failed: %d\n", rc);
1487 return rc;
1488}
1489
1490static int virtio_mem_send_unplug_all_request(struct virtio_mem *vm)
1491{
1492 const struct virtio_mem_req req = {
1493 .type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_UNPLUG_ALL),
1494 };
1495 int rc = -ENOMEM;
1496
1497 dev_dbg(&vm->vdev->dev, "unplugging all memory");
1498
1499 switch (virtio_mem_send_request(vm, &req)) {
1500 case VIRTIO_MEM_RESP_ACK:
1501 vm->unplug_all_required = false;
1502 vm->plugged_size = 0;
1503 /* usable region might have shrunk */
1504 atomic_set(&vm->config_changed, 1);
1505 return 0;
1506 case VIRTIO_MEM_RESP_BUSY:
1507 rc = -ETXTBSY;
1508 break;
1509 default:
1510 break;
1511 }
1512
1513 dev_dbg(&vm->vdev->dev, "unplugging all memory failed: %d\n", rc);
1514 return rc;
1515}
1516
1517/*
1518 * Plug selected subblocks. Updates the plugged state, but not the state
1519 * of the memory block.
1520 */
1521static int virtio_mem_sbm_plug_sb(struct virtio_mem *vm, unsigned long mb_id,
1522 int sb_id, int count)
1523{
1524 const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
1525 sb_id * vm->sbm.sb_size;
1526 const uint64_t size = count * vm->sbm.sb_size;
1527 int rc;
1528
1529 rc = virtio_mem_send_plug_request(vm, addr, size);
1530 if (!rc)
1531 virtio_mem_sbm_set_sb_plugged(vm, mb_id, sb_id, count);
1532 return rc;
1533}
1534
1535/*
1536 * Unplug selected subblocks. Updates the plugged state, but not the state
1537 * of the memory block.
1538 */
1539static int virtio_mem_sbm_unplug_sb(struct virtio_mem *vm, unsigned long mb_id,
1540 int sb_id, int count)
1541{
1542 const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
1543 sb_id * vm->sbm.sb_size;
1544 const uint64_t size = count * vm->sbm.sb_size;
1545 int rc;
1546
1547 rc = virtio_mem_send_unplug_request(vm, addr, size);
1548 if (!rc)
1549 virtio_mem_sbm_set_sb_unplugged(vm, mb_id, sb_id, count);
1550 return rc;
1551}
1552
1553/*
1554 * Request to unplug a big block.
1555 *
1556 * Will not modify the state of the big block.
1557 */
1558static int virtio_mem_bbm_unplug_bb(struct virtio_mem *vm, unsigned long bb_id)
1559{
1560 const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
1561 const uint64_t size = vm->bbm.bb_size;
1562
1563 return virtio_mem_send_unplug_request(vm, addr, size);
1564}
1565
1566/*
1567 * Request to plug a big block.
1568 *
1569 * Will not modify the state of the big block.
1570 */
1571static int virtio_mem_bbm_plug_bb(struct virtio_mem *vm, unsigned long bb_id)
1572{
1573 const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
1574 const uint64_t size = vm->bbm.bb_size;
1575
1576 return virtio_mem_send_plug_request(vm, addr, size);
1577}
1578
1579/*
1580 * Unplug the desired number of plugged subblocks of a offline or not-added
1581 * memory block. Will fail if any subblock cannot get unplugged (instead of
1582 * skipping it).
1583 *
1584 * Will not modify the state of the memory block.
1585 *
1586 * Note: can fail after some subblocks were unplugged.
1587 */
1588static int virtio_mem_sbm_unplug_any_sb_raw(struct virtio_mem *vm,
1589 unsigned long mb_id, uint64_t *nb_sb)
1590{
1591 int sb_id, count;
1592 int rc;
1593
1594 sb_id = vm->sbm.sbs_per_mb - 1;
1595 while (*nb_sb) {
1596 /* Find the next candidate subblock */
1597 while (sb_id >= 0 &&
1598 virtio_mem_sbm_test_sb_unplugged(vm, mb_id, sb_id, 1))
1599 sb_id--;
1600 if (sb_id < 0)
1601 break;
1602 /* Try to unplug multiple subblocks at a time */
1603 count = 1;
1604 while (count < *nb_sb && sb_id > 0 &&
1605 virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id - 1, 1)) {
1606 count++;
1607 sb_id--;
1608 }
1609
1610 rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
1611 if (rc)
1612 return rc;
1613 *nb_sb -= count;
1614 sb_id--;
1615 }
1616
1617 return 0;
1618}
1619
1620/*
1621 * Unplug all plugged subblocks of an offline or not-added memory block.
1622 *
1623 * Will not modify the state of the memory block.
1624 *
1625 * Note: can fail after some subblocks were unplugged.
1626 */
1627static int virtio_mem_sbm_unplug_mb(struct virtio_mem *vm, unsigned long mb_id)
1628{
1629 uint64_t nb_sb = vm->sbm.sbs_per_mb;
1630
1631 return virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, &nb_sb);
1632}
1633
1634/*
1635 * Prepare tracking data for the next memory block.
1636 */
1637static int virtio_mem_sbm_prepare_next_mb(struct virtio_mem *vm,
1638 unsigned long *mb_id)
1639{
1640 int rc;
1641
1642 if (vm->sbm.next_mb_id > vm->sbm.last_usable_mb_id)
1643 return -ENOSPC;
1644
1645 /* Resize the state array if required. */
1646 rc = virtio_mem_sbm_mb_states_prepare_next_mb(vm);
1647 if (rc)
1648 return rc;
1649
1650 /* Resize the subblock bitmap if required. */
1651 rc = virtio_mem_sbm_sb_states_prepare_next_mb(vm);
1652 if (rc)
1653 return rc;
1654
1655 vm->sbm.mb_count[VIRTIO_MEM_SBM_MB_UNUSED]++;
1656 *mb_id = vm->sbm.next_mb_id++;
1657 return 0;
1658}
1659
1660/*
1661 * Try to plug the desired number of subblocks and add the memory block
1662 * to Linux.
1663 *
1664 * Will modify the state of the memory block.
1665 */
1666static int virtio_mem_sbm_plug_and_add_mb(struct virtio_mem *vm,
1667 unsigned long mb_id, uint64_t *nb_sb)
1668{
1669 const int count = min_t(int, *nb_sb, vm->sbm.sbs_per_mb);
1670 int rc;
1671
1672 if (WARN_ON_ONCE(!count))
1673 return -EINVAL;
1674
1675 /*
1676 * Plug the requested number of subblocks before adding it to linux,
1677 * so that onlining will directly online all plugged subblocks.
1678 */
1679 rc = virtio_mem_sbm_plug_sb(vm, mb_id, 0, count);
1680 if (rc)
1681 return rc;
1682
1683 /*
1684 * Mark the block properly offline before adding it to Linux,
1685 * so the memory notifiers will find the block in the right state.
1686 */
1687 if (count == vm->sbm.sbs_per_mb)
1688 virtio_mem_sbm_set_mb_state(vm, mb_id,
1689 VIRTIO_MEM_SBM_MB_OFFLINE);
1690 else
1691 virtio_mem_sbm_set_mb_state(vm, mb_id,
1692 VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
1693
1694 /* Add the memory block to linux - if that fails, try to unplug. */
1695 rc = virtio_mem_sbm_add_mb(vm, mb_id);
1696 if (rc) {
1697 int new_state = VIRTIO_MEM_SBM_MB_UNUSED;
1698
1699 if (virtio_mem_sbm_unplug_sb(vm, mb_id, 0, count))
1700 new_state = VIRTIO_MEM_SBM_MB_PLUGGED;
1701 virtio_mem_sbm_set_mb_state(vm, mb_id, new_state);
1702 return rc;
1703 }
1704
1705 *nb_sb -= count;
1706 return 0;
1707}
1708
1709/*
1710 * Try to plug the desired number of subblocks of a memory block that
1711 * is already added to Linux.
1712 *
1713 * Will modify the state of the memory block.
1714 *
1715 * Note: Can fail after some subblocks were successfully plugged.
1716 */
1717static int virtio_mem_sbm_plug_any_sb(struct virtio_mem *vm,
1718 unsigned long mb_id, uint64_t *nb_sb)
1719{
1720 const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
1721 unsigned long pfn, nr_pages;
1722 int sb_id, count;
1723 int rc;
1724
1725 if (WARN_ON_ONCE(!*nb_sb))
1726 return -EINVAL;
1727
1728 while (*nb_sb) {
1729 sb_id = virtio_mem_sbm_first_unplugged_sb(vm, mb_id);
1730 if (sb_id >= vm->sbm.sbs_per_mb)
1731 break;
1732 count = 1;
1733 while (count < *nb_sb &&
1734 sb_id + count < vm->sbm.sbs_per_mb &&
1735 !virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id + count, 1))
1736 count++;
1737
1738 rc = virtio_mem_sbm_plug_sb(vm, mb_id, sb_id, count);
1739 if (rc)
1740 return rc;
1741 *nb_sb -= count;
1742 if (old_state == VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL)
1743 continue;
1744
1745 /* fake-online the pages if the memory block is online */
1746 pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
1747 sb_id * vm->sbm.sb_size);
1748 nr_pages = PFN_DOWN(count * vm->sbm.sb_size);
1749 virtio_mem_fake_online(pfn, nr_pages);
1750 }
1751
1752 if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
1753 virtio_mem_sbm_set_mb_state(vm, mb_id, old_state - 1);
1754
1755 return 0;
1756}
1757
1758static int virtio_mem_sbm_plug_request(struct virtio_mem *vm, uint64_t diff)
1759{
1760 const int mb_states[] = {
1761 VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
1762 VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
1763 VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
1764 };
1765 uint64_t nb_sb = diff / vm->sbm.sb_size;
1766 unsigned long mb_id;
1767 int rc, i;
1768
1769 if (!nb_sb)
1770 return 0;
1771
1772 /* Don't race with onlining/offlining */
1773 mutex_lock(&vm->hotplug_mutex);
1774
1775 for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
1776 virtio_mem_sbm_for_each_mb(vm, mb_id, mb_states[i]) {
1777 rc = virtio_mem_sbm_plug_any_sb(vm, mb_id, &nb_sb);
1778 if (rc || !nb_sb)
1779 goto out_unlock;
1780 cond_resched();
1781 }
1782 }
1783
1784 /*
1785 * We won't be working on online/offline memory blocks from this point,
1786 * so we can't race with memory onlining/offlining. Drop the mutex.
1787 */
1788 mutex_unlock(&vm->hotplug_mutex);
1789
1790 /* Try to plug and add unused blocks */
1791 virtio_mem_sbm_for_each_mb(vm, mb_id, VIRTIO_MEM_SBM_MB_UNUSED) {
1792 if (!virtio_mem_could_add_memory(vm, memory_block_size_bytes()))
1793 return -ENOSPC;
1794
1795 rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, &nb_sb);
1796 if (rc || !nb_sb)
1797 return rc;
1798 cond_resched();
1799 }
1800
1801 /* Try to prepare, plug and add new blocks */
1802 while (nb_sb) {
1803 if (!virtio_mem_could_add_memory(vm, memory_block_size_bytes()))
1804 return -ENOSPC;
1805
1806 rc = virtio_mem_sbm_prepare_next_mb(vm, &mb_id);
1807 if (rc)
1808 return rc;
1809 rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, &nb_sb);
1810 if (rc)
1811 return rc;
1812 cond_resched();
1813 }
1814
1815 return 0;
1816out_unlock:
1817 mutex_unlock(&vm->hotplug_mutex);
1818 return rc;
1819}
1820
1821/*
1822 * Plug a big block and add it to Linux.
1823 *
1824 * Will modify the state of the big block.
1825 */
1826static int virtio_mem_bbm_plug_and_add_bb(struct virtio_mem *vm,
1827 unsigned long bb_id)
1828{
1829 int rc;
1830
1831 if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
1832 VIRTIO_MEM_BBM_BB_UNUSED))
1833 return -EINVAL;
1834
1835 rc = virtio_mem_bbm_plug_bb(vm, bb_id);
1836 if (rc)
1837 return rc;
1838 virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED);
1839
1840 rc = virtio_mem_bbm_add_bb(vm, bb_id);
1841 if (rc) {
1842 if (!virtio_mem_bbm_unplug_bb(vm, bb_id))
1843 virtio_mem_bbm_set_bb_state(vm, bb_id,
1844 VIRTIO_MEM_BBM_BB_UNUSED);
1845 else
1846 /* Retry from the main loop. */
1847 virtio_mem_bbm_set_bb_state(vm, bb_id,
1848 VIRTIO_MEM_BBM_BB_PLUGGED);
1849 return rc;
1850 }
1851 return 0;
1852}
1853
1854/*
1855 * Prepare tracking data for the next big block.
1856 */
1857static int virtio_mem_bbm_prepare_next_bb(struct virtio_mem *vm,
1858 unsigned long *bb_id)
1859{
1860 int rc;
1861
1862 if (vm->bbm.next_bb_id > vm->bbm.last_usable_bb_id)
1863 return -ENOSPC;
1864
1865 /* Resize the big block state array if required. */
1866 rc = virtio_mem_bbm_bb_states_prepare_next_bb(vm);
1867 if (rc)
1868 return rc;
1869
1870 vm->bbm.bb_count[VIRTIO_MEM_BBM_BB_UNUSED]++;
1871 *bb_id = vm->bbm.next_bb_id;
1872 vm->bbm.next_bb_id++;
1873 return 0;
1874}
1875
1876static int virtio_mem_bbm_plug_request(struct virtio_mem *vm, uint64_t diff)
1877{
1878 uint64_t nb_bb = diff / vm->bbm.bb_size;
1879 unsigned long bb_id;
1880 int rc;
1881
1882 if (!nb_bb)
1883 return 0;
1884
1885 /* Try to plug and add unused big blocks */
1886 virtio_mem_bbm_for_each_bb(vm, bb_id, VIRTIO_MEM_BBM_BB_UNUSED) {
1887 if (!virtio_mem_could_add_memory(vm, vm->bbm.bb_size))
1888 return -ENOSPC;
1889
1890 rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
1891 if (!rc)
1892 nb_bb--;
1893 if (rc || !nb_bb)
1894 return rc;
1895 cond_resched();
1896 }
1897
1898 /* Try to prepare, plug and add new big blocks */
1899 while (nb_bb) {
1900 if (!virtio_mem_could_add_memory(vm, vm->bbm.bb_size))
1901 return -ENOSPC;
1902
1903 rc = virtio_mem_bbm_prepare_next_bb(vm, &bb_id);
1904 if (rc)
1905 return rc;
1906 rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
1907 if (!rc)
1908 nb_bb--;
1909 if (rc)
1910 return rc;
1911 cond_resched();
1912 }
1913
1914 return 0;
1915}
1916
1917/*
1918 * Try to plug the requested amount of memory.
1919 */
1920static int virtio_mem_plug_request(struct virtio_mem *vm, uint64_t diff)
1921{
1922 if (vm->in_sbm)
1923 return virtio_mem_sbm_plug_request(vm, diff);
1924 return virtio_mem_bbm_plug_request(vm, diff);
1925}
1926
1927/*
1928 * Unplug the desired number of plugged subblocks of an offline memory block.
1929 * Will fail if any subblock cannot get unplugged (instead of skipping it).
1930 *
1931 * Will modify the state of the memory block. Might temporarily drop the
1932 * hotplug_mutex.
1933 *
1934 * Note: Can fail after some subblocks were successfully unplugged.
1935 */
1936static int virtio_mem_sbm_unplug_any_sb_offline(struct virtio_mem *vm,
1937 unsigned long mb_id,
1938 uint64_t *nb_sb)
1939{
1940 int rc;
1941
1942 rc = virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, nb_sb);
1943
1944 /* some subblocks might have been unplugged even on failure */
1945 if (!virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
1946 virtio_mem_sbm_set_mb_state(vm, mb_id,
1947 VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
1948 if (rc)
1949 return rc;
1950
1951 if (virtio_mem_sbm_test_sb_unplugged(vm, mb_id, 0, vm->sbm.sbs_per_mb)) {
1952 /*
1953 * Remove the block from Linux - this should never fail.
1954 * Hinder the block from getting onlined by marking it
1955 * unplugged. Temporarily drop the mutex, so
1956 * any pending GOING_ONLINE requests can be serviced/rejected.
1957 */
1958 virtio_mem_sbm_set_mb_state(vm, mb_id,
1959 VIRTIO_MEM_SBM_MB_UNUSED);
1960
1961 mutex_unlock(&vm->hotplug_mutex);
1962 rc = virtio_mem_sbm_remove_mb(vm, mb_id);
1963 BUG_ON(rc);
1964 mutex_lock(&vm->hotplug_mutex);
1965 }
1966 return 0;
1967}
1968
1969/*
1970 * Unplug the given plugged subblocks of an online memory block.
1971 *
1972 * Will modify the state of the memory block.
1973 */
1974static int virtio_mem_sbm_unplug_sb_online(struct virtio_mem *vm,
1975 unsigned long mb_id, int sb_id,
1976 int count)
1977{
1978 const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size) * count;
1979 const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
1980 unsigned long start_pfn;
1981 int rc;
1982
1983 start_pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
1984 sb_id * vm->sbm.sb_size);
1985
1986 rc = virtio_mem_fake_offline(vm, start_pfn, nr_pages);
1987 if (rc)
1988 return rc;
1989
1990 /* Try to unplug the allocated memory */
1991 rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
1992 if (rc) {
1993 /* Return the memory to the buddy. */
1994 virtio_mem_fake_online(start_pfn, nr_pages);
1995 return rc;
1996 }
1997
1998 switch (old_state) {
1999 case VIRTIO_MEM_SBM_MB_KERNEL:
2000 virtio_mem_sbm_set_mb_state(vm, mb_id,
2001 VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL);
2002 break;
2003 case VIRTIO_MEM_SBM_MB_MOVABLE:
2004 virtio_mem_sbm_set_mb_state(vm, mb_id,
2005 VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL);
2006 break;
2007 }
2008
2009 return 0;
2010}
2011
2012/*
2013 * Unplug the desired number of plugged subblocks of an online memory block.
2014 * Will skip subblock that are busy.
2015 *
2016 * Will modify the state of the memory block. Might temporarily drop the
2017 * hotplug_mutex.
2018 *
2019 * Note: Can fail after some subblocks were successfully unplugged. Can
2020 * return 0 even if subblocks were busy and could not get unplugged.
2021 */
2022static int virtio_mem_sbm_unplug_any_sb_online(struct virtio_mem *vm,
2023 unsigned long mb_id,
2024 uint64_t *nb_sb)
2025{
2026 int rc, sb_id;
2027
2028 /* If possible, try to unplug the complete block in one shot. */
2029 if (*nb_sb >= vm->sbm.sbs_per_mb &&
2030 virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb)) {
2031 rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, 0,
2032 vm->sbm.sbs_per_mb);
2033 if (!rc) {
2034 *nb_sb -= vm->sbm.sbs_per_mb;
2035 goto unplugged;
2036 } else if (rc != -EBUSY)
2037 return rc;
2038 }
2039
2040 /* Fallback to single subblocks. */
2041 for (sb_id = vm->sbm.sbs_per_mb - 1; sb_id >= 0 && *nb_sb; sb_id--) {
2042 /* Find the next candidate subblock */
2043 while (sb_id >= 0 &&
2044 !virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
2045 sb_id--;
2046 if (sb_id < 0)
2047 break;
2048
2049 rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, sb_id, 1);
2050 if (rc == -EBUSY)
2051 continue;
2052 else if (rc)
2053 return rc;
2054 *nb_sb -= 1;
2055 }
2056
2057unplugged:
2058 rc = virtio_mem_sbm_try_remove_unplugged_mb(vm, mb_id);
2059 if (rc)
2060 vm->sbm.have_unplugged_mb = 1;
2061 /* Ignore errors, this is not critical. We'll retry later. */
2062 return 0;
2063}
2064
2065/*
2066 * Unplug the desired number of plugged subblocks of a memory block that is
2067 * already added to Linux. Will skip subblock of online memory blocks that are
2068 * busy (by the OS). Will fail if any subblock that's not busy cannot get
2069 * unplugged.
2070 *
2071 * Will modify the state of the memory block. Might temporarily drop the
2072 * hotplug_mutex.
2073 *
2074 * Note: Can fail after some subblocks were successfully unplugged. Can
2075 * return 0 even if subblocks were busy and could not get unplugged.
2076 */
2077static int virtio_mem_sbm_unplug_any_sb(struct virtio_mem *vm,
2078 unsigned long mb_id,
2079 uint64_t *nb_sb)
2080{
2081 const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
2082
2083 switch (old_state) {
2084 case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
2085 case VIRTIO_MEM_SBM_MB_KERNEL:
2086 case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
2087 case VIRTIO_MEM_SBM_MB_MOVABLE:
2088 return virtio_mem_sbm_unplug_any_sb_online(vm, mb_id, nb_sb);
2089 case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
2090 case VIRTIO_MEM_SBM_MB_OFFLINE:
2091 return virtio_mem_sbm_unplug_any_sb_offline(vm, mb_id, nb_sb);
2092 }
2093 return -EINVAL;
2094}
2095
2096static int virtio_mem_sbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
2097{
2098 const int mb_states[] = {
2099 VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
2100 VIRTIO_MEM_SBM_MB_OFFLINE,
2101 VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
2102 VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
2103 VIRTIO_MEM_SBM_MB_MOVABLE,
2104 VIRTIO_MEM_SBM_MB_KERNEL,
2105 };
2106 uint64_t nb_sb = diff / vm->sbm.sb_size;
2107 unsigned long mb_id;
2108 int rc, i;
2109
2110 if (!nb_sb)
2111 return 0;
2112
2113 /*
2114 * We'll drop the mutex a couple of times when it is safe to do so.
2115 * This might result in some blocks switching the state (online/offline)
2116 * and we could miss them in this run - we will retry again later.
2117 */
2118 mutex_lock(&vm->hotplug_mutex);
2119
2120 /*
2121 * We try unplug from partially plugged blocks first, to try removing
2122 * whole memory blocks along with metadata. We prioritize ZONE_MOVABLE
2123 * as it's more reliable to unplug memory and remove whole memory
2124 * blocks, and we don't want to trigger a zone imbalances by
2125 * accidentially removing too much kernel memory.
2126 */
2127 for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
2128 virtio_mem_sbm_for_each_mb_rev(vm, mb_id, mb_states[i]) {
2129 rc = virtio_mem_sbm_unplug_any_sb(vm, mb_id, &nb_sb);
2130 if (rc || !nb_sb)
2131 goto out_unlock;
2132 mutex_unlock(&vm->hotplug_mutex);
2133 cond_resched();
2134 mutex_lock(&vm->hotplug_mutex);
2135 }
2136 if (!unplug_online && i == 1) {
2137 mutex_unlock(&vm->hotplug_mutex);
2138 return 0;
2139 }
2140 }
2141
2142 mutex_unlock(&vm->hotplug_mutex);
2143 return nb_sb ? -EBUSY : 0;
2144out_unlock:
2145 mutex_unlock(&vm->hotplug_mutex);
2146 return rc;
2147}
2148
2149/*
2150 * Try to offline and remove a big block from Linux and unplug it. Will fail
2151 * with -EBUSY if some memory is busy and cannot get unplugged.
2152 *
2153 * Will modify the state of the memory block. Might temporarily drop the
2154 * hotplug_mutex.
2155 */
2156static int virtio_mem_bbm_offline_remove_and_unplug_bb(struct virtio_mem *vm,
2157 unsigned long bb_id)
2158{
2159 const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2160 const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2161 unsigned long end_pfn = start_pfn + nr_pages;
2162 unsigned long pfn;
2163 struct page *page;
2164 int rc;
2165
2166 if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
2167 VIRTIO_MEM_BBM_BB_ADDED))
2168 return -EINVAL;
2169
2170 /*
2171 * Start by fake-offlining all memory. Once we marked the device
2172 * block as fake-offline, all newly onlined memory will
2173 * automatically be kept fake-offline. Protect from concurrent
2174 * onlining/offlining until we have a consistent state.
2175 */
2176 mutex_lock(&vm->hotplug_mutex);
2177 virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_FAKE_OFFLINE);
2178
2179 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2180 page = pfn_to_online_page(pfn);
2181 if (!page)
2182 continue;
2183
2184 rc = virtio_mem_fake_offline(vm, pfn, PAGES_PER_SECTION);
2185 if (rc) {
2186 end_pfn = pfn;
2187 goto rollback;
2188 }
2189 }
2190 mutex_unlock(&vm->hotplug_mutex);
2191
2192 rc = virtio_mem_bbm_offline_and_remove_bb(vm, bb_id);
2193 if (rc) {
2194 mutex_lock(&vm->hotplug_mutex);
2195 goto rollback;
2196 }
2197
2198 rc = virtio_mem_bbm_unplug_bb(vm, bb_id);
2199 if (rc)
2200 virtio_mem_bbm_set_bb_state(vm, bb_id,
2201 VIRTIO_MEM_BBM_BB_PLUGGED);
2202 else
2203 virtio_mem_bbm_set_bb_state(vm, bb_id,
2204 VIRTIO_MEM_BBM_BB_UNUSED);
2205 return rc;
2206
2207rollback:
2208 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2209 page = pfn_to_online_page(pfn);
2210 if (!page)
2211 continue;
2212 virtio_mem_fake_online(pfn, PAGES_PER_SECTION);
2213 }
2214 virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED);
2215 mutex_unlock(&vm->hotplug_mutex);
2216 return rc;
2217}
2218
2219/*
2220 * Test if a big block is completely offline.
2221 */
2222static bool virtio_mem_bbm_bb_is_offline(struct virtio_mem *vm,
2223 unsigned long bb_id)
2224{
2225 const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2226 const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2227 unsigned long pfn;
2228
2229 for (pfn = start_pfn; pfn < start_pfn + nr_pages;
2230 pfn += PAGES_PER_SECTION) {
2231 if (pfn_to_online_page(pfn))
2232 return false;
2233 }
2234
2235 return true;
2236}
2237
2238/*
2239 * Test if a big block is completely onlined to ZONE_MOVABLE (or offline).
2240 */
2241static bool virtio_mem_bbm_bb_is_movable(struct virtio_mem *vm,
2242 unsigned long bb_id)
2243{
2244 const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2245 const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2246 struct page *page;
2247 unsigned long pfn;
2248
2249 for (pfn = start_pfn; pfn < start_pfn + nr_pages;
2250 pfn += PAGES_PER_SECTION) {
2251 page = pfn_to_online_page(pfn);
2252 if (!page)
2253 continue;
2254 if (page_zonenum(page) != ZONE_MOVABLE)
2255 return false;
2256 }
2257
2258 return true;
2259}
2260
2261static int virtio_mem_bbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
2262{
2263 uint64_t nb_bb = diff / vm->bbm.bb_size;
2264 uint64_t bb_id;
2265 int rc, i;
2266
2267 if (!nb_bb)
2268 return 0;
2269
2270 /*
2271 * Try to unplug big blocks. Similar to SBM, start with offline
2272 * big blocks.
2273 */
2274 for (i = 0; i < 3; i++) {
2275 virtio_mem_bbm_for_each_bb_rev(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED) {
2276 cond_resched();
2277
2278 /*
2279 * As we're holding no locks, these checks are racy,
2280 * but we don't care.
2281 */
2282 if (i == 0 && !virtio_mem_bbm_bb_is_offline(vm, bb_id))
2283 continue;
2284 if (i == 1 && !virtio_mem_bbm_bb_is_movable(vm, bb_id))
2285 continue;
2286 rc = virtio_mem_bbm_offline_remove_and_unplug_bb(vm, bb_id);
2287 if (rc == -EBUSY)
2288 continue;
2289 if (!rc)
2290 nb_bb--;
2291 if (rc || !nb_bb)
2292 return rc;
2293 }
2294 if (i == 0 && !unplug_online)
2295 return 0;
2296 }
2297
2298 return nb_bb ? -EBUSY : 0;
2299}
2300
2301/*
2302 * Try to unplug the requested amount of memory.
2303 */
2304static int virtio_mem_unplug_request(struct virtio_mem *vm, uint64_t diff)
2305{
2306 if (vm->in_sbm)
2307 return virtio_mem_sbm_unplug_request(vm, diff);
2308 return virtio_mem_bbm_unplug_request(vm, diff);
2309}
2310
2311/*
2312 * Try to unplug all blocks that couldn't be unplugged before, for example,
2313 * because the hypervisor was busy. Further, offline and remove any memory
2314 * blocks where we previously failed.
2315 */
2316static int virtio_mem_cleanup_pending_mb(struct virtio_mem *vm)
2317{
2318 unsigned long id;
2319 int rc = 0;
2320
2321 if (!vm->in_sbm) {
2322 virtio_mem_bbm_for_each_bb(vm, id,
2323 VIRTIO_MEM_BBM_BB_PLUGGED) {
2324 rc = virtio_mem_bbm_unplug_bb(vm, id);
2325 if (rc)
2326 return rc;
2327 virtio_mem_bbm_set_bb_state(vm, id,
2328 VIRTIO_MEM_BBM_BB_UNUSED);
2329 }
2330 return 0;
2331 }
2332
2333 virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_PLUGGED) {
2334 rc = virtio_mem_sbm_unplug_mb(vm, id);
2335 if (rc)
2336 return rc;
2337 virtio_mem_sbm_set_mb_state(vm, id,
2338 VIRTIO_MEM_SBM_MB_UNUSED);
2339 }
2340
2341 if (!vm->sbm.have_unplugged_mb)
2342 return 0;
2343
2344 /*
2345 * Let's retry (offlining and) removing completely unplugged Linux
2346 * memory blocks.
2347 */
2348 vm->sbm.have_unplugged_mb = false;
2349
2350 mutex_lock(&vm->hotplug_mutex);
2351 virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL)
2352 rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, id);
2353 virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL)
2354 rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, id);
2355 virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL)
2356 rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, id);
2357 mutex_unlock(&vm->hotplug_mutex);
2358
2359 if (rc)
2360 vm->sbm.have_unplugged_mb = true;
2361 /* Ignore errors, this is not critical. We'll retry later. */
2362 return 0;
2363}
2364
2365/*
2366 * Update all parts of the config that could have changed.
2367 */
2368static void virtio_mem_refresh_config(struct virtio_mem *vm)
2369{
2370 const struct range pluggable_range = mhp_get_pluggable_range(true);
2371 uint64_t new_plugged_size, usable_region_size, end_addr;
2372
2373 /* the plugged_size is just a reflection of what _we_ did previously */
2374 virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
2375 &new_plugged_size);
2376 if (WARN_ON_ONCE(new_plugged_size != vm->plugged_size))
2377 vm->plugged_size = new_plugged_size;
2378
2379 /* calculate the last usable memory block id */
2380 virtio_cread_le(vm->vdev, struct virtio_mem_config,
2381 usable_region_size, &usable_region_size);
2382 end_addr = min(vm->addr + usable_region_size - 1,
2383 pluggable_range.end);
2384
2385 if (vm->in_sbm) {
2386 vm->sbm.last_usable_mb_id = virtio_mem_phys_to_mb_id(end_addr);
2387 if (!IS_ALIGNED(end_addr + 1, memory_block_size_bytes()))
2388 vm->sbm.last_usable_mb_id--;
2389 } else {
2390 vm->bbm.last_usable_bb_id = virtio_mem_phys_to_bb_id(vm,
2391 end_addr);
2392 if (!IS_ALIGNED(end_addr + 1, vm->bbm.bb_size))
2393 vm->bbm.last_usable_bb_id--;
2394 }
2395 /*
2396 * If we cannot plug any of our device memory (e.g., nothing in the
2397 * usable region is addressable), the last usable memory block id will
2398 * be smaller than the first usable memory block id. We'll stop
2399 * attempting to add memory with -ENOSPC from our main loop.
2400 */
2401
2402 /* see if there is a request to change the size */
2403 virtio_cread_le(vm->vdev, struct virtio_mem_config, requested_size,
2404 &vm->requested_size);
2405
2406 dev_info(&vm->vdev->dev, "plugged size: 0x%llx", vm->plugged_size);
2407 dev_info(&vm->vdev->dev, "requested size: 0x%llx", vm->requested_size);
2408}
2409
2410/*
2411 * Workqueue function for handling plug/unplug requests and config updates.
2412 */
2413static void virtio_mem_run_wq(struct work_struct *work)
2414{
2415 struct virtio_mem *vm = container_of(work, struct virtio_mem, wq);
2416 uint64_t diff;
2417 int rc;
2418
2419 if (unlikely(vm->in_kdump)) {
2420 dev_warn_once(&vm->vdev->dev,
2421 "unexpected workqueue run in kdump kernel\n");
2422 return;
2423 }
2424
2425 hrtimer_cancel(&vm->retry_timer);
2426
2427 if (vm->broken)
2428 return;
2429
2430 atomic_set(&vm->wq_active, 1);
2431retry:
2432 rc = 0;
2433
2434 /* Make sure we start with a clean state if there are leftovers. */
2435 if (unlikely(vm->unplug_all_required))
2436 rc = virtio_mem_send_unplug_all_request(vm);
2437
2438 if (atomic_read(&vm->config_changed)) {
2439 atomic_set(&vm->config_changed, 0);
2440 virtio_mem_refresh_config(vm);
2441 }
2442
2443 /* Cleanup any leftovers from previous runs */
2444 if (!rc)
2445 rc = virtio_mem_cleanup_pending_mb(vm);
2446
2447 if (!rc && vm->requested_size != vm->plugged_size) {
2448 if (vm->requested_size > vm->plugged_size) {
2449 diff = vm->requested_size - vm->plugged_size;
2450 rc = virtio_mem_plug_request(vm, diff);
2451 } else {
2452 diff = vm->plugged_size - vm->requested_size;
2453 rc = virtio_mem_unplug_request(vm, diff);
2454 }
2455 }
2456
2457 /*
2458 * Keep retrying to offline and remove completely unplugged Linux
2459 * memory blocks.
2460 */
2461 if (!rc && vm->in_sbm && vm->sbm.have_unplugged_mb)
2462 rc = -EBUSY;
2463
2464 switch (rc) {
2465 case 0:
2466 vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
2467 break;
2468 case -ENOSPC:
2469 /*
2470 * We cannot add any more memory (alignment, physical limit)
2471 * or we have too many offline memory blocks.
2472 */
2473 break;
2474 case -ETXTBSY:
2475 /*
2476 * The hypervisor cannot process our request right now
2477 * (e.g., out of memory, migrating);
2478 */
2479 case -EBUSY:
2480 /*
2481 * We cannot free up any memory to unplug it (all plugged memory
2482 * is busy).
2483 */
2484 case -ENOMEM:
2485 /* Out of memory, try again later. */
2486 hrtimer_start(&vm->retry_timer, ms_to_ktime(vm->retry_timer_ms),
2487 HRTIMER_MODE_REL);
2488 break;
2489 case -EAGAIN:
2490 /* Retry immediately (e.g., the config changed). */
2491 goto retry;
2492 default:
2493 /* Unknown error, mark as broken */
2494 dev_err(&vm->vdev->dev,
2495 "unknown error, marking device broken: %d\n", rc);
2496 vm->broken = true;
2497 }
2498
2499 atomic_set(&vm->wq_active, 0);
2500}
2501
2502static enum hrtimer_restart virtio_mem_timer_expired(struct hrtimer *timer)
2503{
2504 struct virtio_mem *vm = container_of(timer, struct virtio_mem,
2505 retry_timer);
2506
2507 virtio_mem_retry(vm);
2508 vm->retry_timer_ms = min_t(unsigned int, vm->retry_timer_ms * 2,
2509 VIRTIO_MEM_RETRY_TIMER_MAX_MS);
2510 return HRTIMER_NORESTART;
2511}
2512
2513static void virtio_mem_handle_response(struct virtqueue *vq)
2514{
2515 struct virtio_mem *vm = vq->vdev->priv;
2516
2517 wake_up(&vm->host_resp);
2518}
2519
2520static int virtio_mem_init_vq(struct virtio_mem *vm)
2521{
2522 struct virtqueue *vq;
2523
2524 vq = virtio_find_single_vq(vm->vdev, virtio_mem_handle_response,
2525 "guest-request");
2526 if (IS_ERR(vq))
2527 return PTR_ERR(vq);
2528 vm->vq = vq;
2529
2530 return 0;
2531}
2532
2533static int virtio_mem_init_hotplug(struct virtio_mem *vm)
2534{
2535 const struct range pluggable_range = mhp_get_pluggable_range(true);
2536 uint64_t unit_pages, sb_size, addr;
2537 int rc;
2538
2539 /* bad device setup - warn only */
2540 if (!IS_ALIGNED(vm->addr, memory_block_size_bytes()))
2541 dev_warn(&vm->vdev->dev,
2542 "The alignment of the physical start address can make some memory unusable.\n");
2543 if (!IS_ALIGNED(vm->addr + vm->region_size, memory_block_size_bytes()))
2544 dev_warn(&vm->vdev->dev,
2545 "The alignment of the physical end address can make some memory unusable.\n");
2546 if (vm->addr < pluggable_range.start ||
2547 vm->addr + vm->region_size - 1 > pluggable_range.end)
2548 dev_warn(&vm->vdev->dev,
2549 "Some device memory is not addressable/pluggable. This can make some memory unusable.\n");
2550
2551 /* Prepare the offline threshold - make sure we can add two blocks. */
2552 vm->offline_threshold = max_t(uint64_t, 2 * memory_block_size_bytes(),
2553 VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD);
2554
2555 /*
2556 * alloc_contig_range() works reliably with pageblock
2557 * granularity on ZONE_NORMAL, use pageblock_nr_pages.
2558 */
2559 sb_size = PAGE_SIZE * pageblock_nr_pages;
2560 sb_size = max_t(uint64_t, vm->device_block_size, sb_size);
2561
2562 if (sb_size < memory_block_size_bytes() && !force_bbm) {
2563 /* SBM: At least two subblocks per Linux memory block. */
2564 vm->in_sbm = true;
2565 vm->sbm.sb_size = sb_size;
2566 vm->sbm.sbs_per_mb = memory_block_size_bytes() /
2567 vm->sbm.sb_size;
2568
2569 /* Round up to the next full memory block */
2570 addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
2571 memory_block_size_bytes() - 1;
2572 vm->sbm.first_mb_id = virtio_mem_phys_to_mb_id(addr);
2573 vm->sbm.next_mb_id = vm->sbm.first_mb_id;
2574 } else {
2575 /* BBM: At least one Linux memory block. */
2576 vm->bbm.bb_size = max_t(uint64_t, vm->device_block_size,
2577 memory_block_size_bytes());
2578
2579 if (bbm_block_size) {
2580 if (!is_power_of_2(bbm_block_size)) {
2581 dev_warn(&vm->vdev->dev,
2582 "bbm_block_size is not a power of 2");
2583 } else if (bbm_block_size < vm->bbm.bb_size) {
2584 dev_warn(&vm->vdev->dev,
2585 "bbm_block_size is too small");
2586 } else {
2587 vm->bbm.bb_size = bbm_block_size;
2588 }
2589 }
2590
2591 /* Round up to the next aligned big block */
2592 addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
2593 vm->bbm.bb_size - 1;
2594 vm->bbm.first_bb_id = virtio_mem_phys_to_bb_id(vm, addr);
2595 vm->bbm.next_bb_id = vm->bbm.first_bb_id;
2596
2597 /* Make sure we can add two big blocks. */
2598 vm->offline_threshold = max_t(uint64_t, 2 * vm->bbm.bb_size,
2599 vm->offline_threshold);
2600 }
2601
2602 dev_info(&vm->vdev->dev, "memory block size: 0x%lx",
2603 memory_block_size_bytes());
2604 if (vm->in_sbm)
2605 dev_info(&vm->vdev->dev, "subblock size: 0x%llx",
2606 (unsigned long long)vm->sbm.sb_size);
2607 else
2608 dev_info(&vm->vdev->dev, "big block size: 0x%llx",
2609 (unsigned long long)vm->bbm.bb_size);
2610
2611 /* create the parent resource for all memory */
2612 rc = virtio_mem_create_resource(vm);
2613 if (rc)
2614 return rc;
2615
2616 /* use a single dynamic memory group to cover the whole memory device */
2617 if (vm->in_sbm)
2618 unit_pages = PHYS_PFN(memory_block_size_bytes());
2619 else
2620 unit_pages = PHYS_PFN(vm->bbm.bb_size);
2621 rc = memory_group_register_dynamic(vm->nid, unit_pages);
2622 if (rc < 0)
2623 goto out_del_resource;
2624 vm->mgid = rc;
2625
2626 /*
2627 * If we still have memory plugged, we have to unplug all memory first.
2628 * Registering our parent resource makes sure that this memory isn't
2629 * actually in use (e.g., trying to reload the driver).
2630 */
2631 if (vm->plugged_size) {
2632 vm->unplug_all_required = true;
2633 dev_info(&vm->vdev->dev, "unplugging all memory is required\n");
2634 }
2635
2636 /* register callbacks */
2637 vm->memory_notifier.notifier_call = virtio_mem_memory_notifier_cb;
2638 rc = register_memory_notifier(&vm->memory_notifier);
2639 if (rc)
2640 goto out_unreg_group;
2641 /* Block hibernation as early as possible. */
2642 vm->pm_notifier.priority = INT_MAX;
2643 vm->pm_notifier.notifier_call = virtio_mem_pm_notifier_cb;
2644 rc = register_pm_notifier(&vm->pm_notifier);
2645 if (rc)
2646 goto out_unreg_mem;
2647 rc = register_virtio_mem_device(vm);
2648 if (rc)
2649 goto out_unreg_pm;
2650
2651 return 0;
2652out_unreg_pm:
2653 unregister_pm_notifier(&vm->pm_notifier);
2654out_unreg_mem:
2655 unregister_memory_notifier(&vm->memory_notifier);
2656out_unreg_group:
2657 memory_group_unregister(vm->mgid);
2658out_del_resource:
2659 virtio_mem_delete_resource(vm);
2660 return rc;
2661}
2662
2663#ifdef CONFIG_PROC_VMCORE
2664static int virtio_mem_send_state_request(struct virtio_mem *vm, uint64_t addr,
2665 uint64_t size)
2666{
2667 const uint64_t nb_vm_blocks = size / vm->device_block_size;
2668 const struct virtio_mem_req req = {
2669 .type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_STATE),
2670 .u.state.addr = cpu_to_virtio64(vm->vdev, addr),
2671 .u.state.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
2672 };
2673 int rc = -ENOMEM;
2674
2675 dev_dbg(&vm->vdev->dev, "requesting state: 0x%llx - 0x%llx\n", addr,
2676 addr + size - 1);
2677
2678 switch (virtio_mem_send_request(vm, &req)) {
2679 case VIRTIO_MEM_RESP_ACK:
2680 return virtio16_to_cpu(vm->vdev, vm->resp.u.state.state);
2681 case VIRTIO_MEM_RESP_ERROR:
2682 rc = -EINVAL;
2683 break;
2684 default:
2685 break;
2686 }
2687
2688 dev_dbg(&vm->vdev->dev, "requesting state failed: %d\n", rc);
2689 return rc;
2690}
2691
2692static bool virtio_mem_vmcore_pfn_is_ram(struct vmcore_cb *cb,
2693 unsigned long pfn)
2694{
2695 struct virtio_mem *vm = container_of(cb, struct virtio_mem,
2696 vmcore_cb);
2697 uint64_t addr = PFN_PHYS(pfn);
2698 bool is_ram;
2699 int rc;
2700
2701 if (!virtio_mem_contains_range(vm, addr, PAGE_SIZE))
2702 return true;
2703 if (!vm->plugged_size)
2704 return false;
2705
2706 /*
2707 * We have to serialize device requests and access to the information
2708 * about the block queried last.
2709 */
2710 mutex_lock(&vm->hotplug_mutex);
2711
2712 addr = ALIGN_DOWN(addr, vm->device_block_size);
2713 if (addr != vm->last_block_addr) {
2714 rc = virtio_mem_send_state_request(vm, addr,
2715 vm->device_block_size);
2716 /* On any kind of error, we're going to signal !ram. */
2717 if (rc == VIRTIO_MEM_STATE_PLUGGED)
2718 vm->last_block_plugged = true;
2719 else
2720 vm->last_block_plugged = false;
2721 vm->last_block_addr = addr;
2722 }
2723
2724 is_ram = vm->last_block_plugged;
2725 mutex_unlock(&vm->hotplug_mutex);
2726 return is_ram;
2727}
2728#endif /* CONFIG_PROC_VMCORE */
2729
2730static int virtio_mem_init_kdump(struct virtio_mem *vm)
2731{
2732#ifdef CONFIG_PROC_VMCORE
2733 dev_info(&vm->vdev->dev, "memory hot(un)plug disabled in kdump kernel\n");
2734 vm->vmcore_cb.pfn_is_ram = virtio_mem_vmcore_pfn_is_ram;
2735 register_vmcore_cb(&vm->vmcore_cb);
2736 return 0;
2737#else /* CONFIG_PROC_VMCORE */
2738 dev_warn(&vm->vdev->dev, "disabled in kdump kernel without vmcore\n");
2739 return -EBUSY;
2740#endif /* CONFIG_PROC_VMCORE */
2741}
2742
2743static int virtio_mem_init(struct virtio_mem *vm)
2744{
2745 uint16_t node_id;
2746
2747 if (!vm->vdev->config->get) {
2748 dev_err(&vm->vdev->dev, "config access disabled\n");
2749 return -EINVAL;
2750 }
2751
2752 /* Fetch all properties that can't change. */
2753 virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
2754 &vm->plugged_size);
2755 virtio_cread_le(vm->vdev, struct virtio_mem_config, block_size,
2756 &vm->device_block_size);
2757 virtio_cread_le(vm->vdev, struct virtio_mem_config, node_id,
2758 &node_id);
2759 vm->nid = virtio_mem_translate_node_id(vm, node_id);
2760 virtio_cread_le(vm->vdev, struct virtio_mem_config, addr, &vm->addr);
2761 virtio_cread_le(vm->vdev, struct virtio_mem_config, region_size,
2762 &vm->region_size);
2763
2764 /* Determine the nid for the device based on the lowest address. */
2765 if (vm->nid == NUMA_NO_NODE)
2766 vm->nid = memory_add_physaddr_to_nid(vm->addr);
2767
2768 dev_info(&vm->vdev->dev, "start address: 0x%llx", vm->addr);
2769 dev_info(&vm->vdev->dev, "region size: 0x%llx", vm->region_size);
2770 dev_info(&vm->vdev->dev, "device block size: 0x%llx",
2771 (unsigned long long)vm->device_block_size);
2772 if (vm->nid != NUMA_NO_NODE && IS_ENABLED(CONFIG_NUMA))
2773 dev_info(&vm->vdev->dev, "nid: %d", vm->nid);
2774
2775 /*
2776 * We don't want to (un)plug or reuse any memory when in kdump. The
2777 * memory is still accessible (but not exposed to Linux).
2778 */
2779 if (vm->in_kdump)
2780 return virtio_mem_init_kdump(vm);
2781 return virtio_mem_init_hotplug(vm);
2782}
2783
2784static int virtio_mem_create_resource(struct virtio_mem *vm)
2785{
2786 /*
2787 * When force-unloading the driver and removing the device, we
2788 * could have a garbage pointer. Duplicate the string.
2789 */
2790 const char *name = kstrdup(dev_name(&vm->vdev->dev), GFP_KERNEL);
2791
2792 if (!name)
2793 return -ENOMEM;
2794
2795 /* Disallow mapping device memory via /dev/mem completely. */
2796 vm->parent_resource = __request_mem_region(vm->addr, vm->region_size,
2797 name, IORESOURCE_SYSTEM_RAM |
2798 IORESOURCE_EXCLUSIVE);
2799 if (!vm->parent_resource) {
2800 kfree(name);
2801 dev_warn(&vm->vdev->dev, "could not reserve device region\n");
2802 dev_info(&vm->vdev->dev,
2803 "reloading the driver is not supported\n");
2804 return -EBUSY;
2805 }
2806
2807 /* The memory is not actually busy - make add_memory() work. */
2808 vm->parent_resource->flags &= ~IORESOURCE_BUSY;
2809 return 0;
2810}
2811
2812static void virtio_mem_delete_resource(struct virtio_mem *vm)
2813{
2814 const char *name;
2815
2816 if (!vm->parent_resource)
2817 return;
2818
2819 name = vm->parent_resource->name;
2820 release_resource(vm->parent_resource);
2821 kfree(vm->parent_resource);
2822 kfree(name);
2823 vm->parent_resource = NULL;
2824}
2825
2826static int virtio_mem_range_has_system_ram(struct resource *res, void *arg)
2827{
2828 return 1;
2829}
2830
2831static bool virtio_mem_has_memory_added(struct virtio_mem *vm)
2832{
2833 const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
2834
2835 return walk_iomem_res_desc(IORES_DESC_NONE, flags, vm->addr,
2836 vm->addr + vm->region_size, NULL,
2837 virtio_mem_range_has_system_ram) == 1;
2838}
2839
2840static int virtio_mem_probe(struct virtio_device *vdev)
2841{
2842 struct virtio_mem *vm;
2843 int rc;
2844
2845 BUILD_BUG_ON(sizeof(struct virtio_mem_req) != 24);
2846 BUILD_BUG_ON(sizeof(struct virtio_mem_resp) != 10);
2847
2848 vdev->priv = vm = kzalloc(sizeof(*vm), GFP_KERNEL);
2849 if (!vm)
2850 return -ENOMEM;
2851
2852 init_waitqueue_head(&vm->host_resp);
2853 vm->vdev = vdev;
2854 INIT_WORK(&vm->wq, virtio_mem_run_wq);
2855 mutex_init(&vm->hotplug_mutex);
2856 INIT_LIST_HEAD(&vm->next);
2857 spin_lock_init(&vm->removal_lock);
2858 hrtimer_init(&vm->retry_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2859 vm->retry_timer.function = virtio_mem_timer_expired;
2860 vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
2861 vm->in_kdump = is_kdump_kernel();
2862
2863 /* register the virtqueue */
2864 rc = virtio_mem_init_vq(vm);
2865 if (rc)
2866 goto out_free_vm;
2867
2868 /* initialize the device by querying the config */
2869 rc = virtio_mem_init(vm);
2870 if (rc)
2871 goto out_del_vq;
2872
2873 virtio_device_ready(vdev);
2874
2875 /* trigger a config update to start processing the requested_size */
2876 if (!vm->in_kdump) {
2877 atomic_set(&vm->config_changed, 1);
2878 queue_work(system_freezable_wq, &vm->wq);
2879 }
2880
2881 return 0;
2882out_del_vq:
2883 vdev->config->del_vqs(vdev);
2884out_free_vm:
2885 kfree(vm);
2886 vdev->priv = NULL;
2887
2888 return rc;
2889}
2890
2891static void virtio_mem_deinit_hotplug(struct virtio_mem *vm)
2892{
2893 unsigned long mb_id;
2894 int rc;
2895
2896 /*
2897 * Make sure the workqueue won't be triggered anymore and no memory
2898 * blocks can be onlined/offlined until we're finished here.
2899 */
2900 mutex_lock(&vm->hotplug_mutex);
2901 spin_lock_irq(&vm->removal_lock);
2902 vm->removing = true;
2903 spin_unlock_irq(&vm->removal_lock);
2904 mutex_unlock(&vm->hotplug_mutex);
2905
2906 /* wait until the workqueue stopped */
2907 cancel_work_sync(&vm->wq);
2908 hrtimer_cancel(&vm->retry_timer);
2909
2910 if (vm->in_sbm) {
2911 /*
2912 * After we unregistered our callbacks, user space can online
2913 * partially plugged offline blocks. Make sure to remove them.
2914 */
2915 virtio_mem_sbm_for_each_mb(vm, mb_id,
2916 VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL) {
2917 rc = virtio_mem_sbm_remove_mb(vm, mb_id);
2918 BUG_ON(rc);
2919 virtio_mem_sbm_set_mb_state(vm, mb_id,
2920 VIRTIO_MEM_SBM_MB_UNUSED);
2921 }
2922 /*
2923 * After we unregistered our callbacks, user space can no longer
2924 * offline partially plugged online memory blocks. No need to
2925 * worry about them.
2926 */
2927 }
2928
2929 /* unregister callbacks */
2930 unregister_virtio_mem_device(vm);
2931 unregister_pm_notifier(&vm->pm_notifier);
2932 unregister_memory_notifier(&vm->memory_notifier);
2933
2934 /*
2935 * There is no way we could reliably remove all memory we have added to
2936 * the system. And there is no way to stop the driver/device from going
2937 * away. Warn at least.
2938 */
2939 if (virtio_mem_has_memory_added(vm)) {
2940 dev_warn(&vm->vdev->dev,
2941 "device still has system memory added\n");
2942 } else {
2943 virtio_mem_delete_resource(vm);
2944 kfree_const(vm->resource_name);
2945 memory_group_unregister(vm->mgid);
2946 }
2947
2948 /* remove all tracking data - no locking needed */
2949 if (vm->in_sbm) {
2950 vfree(vm->sbm.mb_states);
2951 vfree(vm->sbm.sb_states);
2952 } else {
2953 vfree(vm->bbm.bb_states);
2954 }
2955}
2956
2957static void virtio_mem_deinit_kdump(struct virtio_mem *vm)
2958{
2959#ifdef CONFIG_PROC_VMCORE
2960 unregister_vmcore_cb(&vm->vmcore_cb);
2961#endif /* CONFIG_PROC_VMCORE */
2962}
2963
2964static void virtio_mem_remove(struct virtio_device *vdev)
2965{
2966 struct virtio_mem *vm = vdev->priv;
2967
2968 if (vm->in_kdump)
2969 virtio_mem_deinit_kdump(vm);
2970 else
2971 virtio_mem_deinit_hotplug(vm);
2972
2973 /* reset the device and cleanup the queues */
2974 virtio_reset_device(vdev);
2975 vdev->config->del_vqs(vdev);
2976
2977 kfree(vm);
2978 vdev->priv = NULL;
2979}
2980
2981static void virtio_mem_config_changed(struct virtio_device *vdev)
2982{
2983 struct virtio_mem *vm = vdev->priv;
2984
2985 if (unlikely(vm->in_kdump))
2986 return;
2987
2988 atomic_set(&vm->config_changed, 1);
2989 virtio_mem_retry(vm);
2990}
2991
2992#ifdef CONFIG_PM_SLEEP
2993static int virtio_mem_freeze(struct virtio_device *vdev)
2994{
2995 struct virtio_mem *vm = vdev->priv;
2996
2997 /*
2998 * We block hibernation using the PM notifier completely. The workqueue
2999 * is already frozen by the PM core at this point, so we simply
3000 * reset the device and cleanup the queues.
3001 */
3002 if (pm_suspend_target_state != PM_SUSPEND_TO_IDLE &&
3003 vm->plugged_size &&
3004 !virtio_has_feature(vm->vdev, VIRTIO_MEM_F_PERSISTENT_SUSPEND)) {
3005 dev_err(&vm->vdev->dev,
3006 "suspending with plugged memory is not supported\n");
3007 return -EPERM;
3008 }
3009
3010 virtio_reset_device(vdev);
3011 vdev->config->del_vqs(vdev);
3012 vm->vq = NULL;
3013 return 0;
3014}
3015
3016static int virtio_mem_restore(struct virtio_device *vdev)
3017{
3018 struct virtio_mem *vm = vdev->priv;
3019 int ret;
3020
3021 ret = virtio_mem_init_vq(vm);
3022 if (ret)
3023 return ret;
3024 virtio_device_ready(vdev);
3025
3026 /* Let's check if anything changed. */
3027 virtio_mem_config_changed(vdev);
3028 return 0;
3029}
3030#endif
3031
3032static unsigned int virtio_mem_features[] = {
3033#if defined(CONFIG_NUMA) && defined(CONFIG_ACPI_NUMA)
3034 VIRTIO_MEM_F_ACPI_PXM,
3035#endif
3036 VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE,
3037 VIRTIO_MEM_F_PERSISTENT_SUSPEND,
3038};
3039
3040static const struct virtio_device_id virtio_mem_id_table[] = {
3041 { VIRTIO_ID_MEM, VIRTIO_DEV_ANY_ID },
3042 { 0 },
3043};
3044
3045static struct virtio_driver virtio_mem_driver = {
3046 .feature_table = virtio_mem_features,
3047 .feature_table_size = ARRAY_SIZE(virtio_mem_features),
3048 .driver.name = KBUILD_MODNAME,
3049 .id_table = virtio_mem_id_table,
3050 .probe = virtio_mem_probe,
3051 .remove = virtio_mem_remove,
3052 .config_changed = virtio_mem_config_changed,
3053#ifdef CONFIG_PM_SLEEP
3054 .freeze = virtio_mem_freeze,
3055 .restore = virtio_mem_restore,
3056#endif
3057};
3058
3059module_virtio_driver(virtio_mem_driver);
3060MODULE_DEVICE_TABLE(virtio, virtio_mem_id_table);
3061MODULE_AUTHOR("David Hildenbrand <david@redhat.com>");
3062MODULE_DESCRIPTION("Virtio-mem driver");
3063MODULE_LICENSE("GPL");