xref: /linux/drivers/virtio/virtio_mem.c (revision cbac924200b838cfb8d8b1415113d788089dc50b)
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 
25 #include <acpi/acpi_numa.h>
26 
27 static bool unplug_online = true;
28 module_param(unplug_online, bool, 0644);
29 MODULE_PARM_DESC(unplug_online, "Try to unplug online memory");
30 
31 static bool force_bbm;
32 module_param(force_bbm, bool, 0444);
33 MODULE_PARM_DESC(force_bbm,
34 		"Force Big Block Mode. Default is 0 (auto-selection)");
35 
36 static unsigned long bbm_block_size;
37 module_param(bbm_block_size, ulong, 0444);
38 MODULE_PARM_DESC(bbm_block_size,
39 		 "Big Block size in bytes. Default is 0 (auto-detection).");
40 
41 static bool bbm_safe_unplug = true;
42 module_param(bbm_safe_unplug, bool, 0444);
43 MODULE_PARM_DESC(bbm_safe_unplug,
44 	     "Use a safe unplug mechanism in BBM, avoiding long/endless loops");
45 
46 /*
47  * virtio-mem currently supports the following modes of operation:
48  *
49  * * Sub Block Mode (SBM): A Linux memory block spans 2..X subblocks (SB). The
50  *   size of a Sub Block (SB) is determined based on the device block size, the
51  *   pageblock size, and the maximum allocation granularity of the buddy.
52  *   Subblocks within a Linux memory block might either be plugged or unplugged.
53  *   Memory is added/removed to Linux MM in Linux memory block granularity.
54  *
55  * * Big Block Mode (BBM): A Big Block (BB) spans 1..X Linux memory blocks.
56  *   Memory is added/removed to Linux MM in Big Block granularity.
57  *
58  * The mode is determined automatically based on the Linux memory block size
59  * and the device block size.
60  *
61  * User space / core MM (auto onlining) is responsible for onlining added
62  * Linux memory blocks - and for selecting a zone. Linux Memory Blocks are
63  * always onlined separately, and all memory within a Linux memory block is
64  * onlined to the same zone - virtio-mem relies on this behavior.
65  */
66 
67 /*
68  * State of a Linux memory block in SBM.
69  */
70 enum virtio_mem_sbm_mb_state {
71 	/* Unplugged, not added to Linux. Can be reused later. */
72 	VIRTIO_MEM_SBM_MB_UNUSED = 0,
73 	/* (Partially) plugged, not added to Linux. Error on add_memory(). */
74 	VIRTIO_MEM_SBM_MB_PLUGGED,
75 	/* Fully plugged, fully added to Linux, offline. */
76 	VIRTIO_MEM_SBM_MB_OFFLINE,
77 	/* Partially plugged, fully added to Linux, offline. */
78 	VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
79 	/* Fully plugged, fully added to Linux, onlined to a kernel zone. */
80 	VIRTIO_MEM_SBM_MB_KERNEL,
81 	/* Partially plugged, fully added to Linux, online to a kernel zone */
82 	VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
83 	/* Fully plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
84 	VIRTIO_MEM_SBM_MB_MOVABLE,
85 	/* Partially plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
86 	VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
87 	VIRTIO_MEM_SBM_MB_COUNT
88 };
89 
90 /*
91  * State of a Big Block (BB) in BBM, covering 1..X Linux memory blocks.
92  */
93 enum virtio_mem_bbm_bb_state {
94 	/* Unplugged, not added to Linux. Can be reused later. */
95 	VIRTIO_MEM_BBM_BB_UNUSED = 0,
96 	/* Plugged, not added to Linux. Error on add_memory(). */
97 	VIRTIO_MEM_BBM_BB_PLUGGED,
98 	/* Plugged and added to Linux. */
99 	VIRTIO_MEM_BBM_BB_ADDED,
100 	/* All online parts are fake-offline, ready to remove. */
101 	VIRTIO_MEM_BBM_BB_FAKE_OFFLINE,
102 	VIRTIO_MEM_BBM_BB_COUNT
103 };
104 
105 struct virtio_mem {
106 	struct virtio_device *vdev;
107 
108 	/* We might first have to unplug all memory when starting up. */
109 	bool unplug_all_required;
110 
111 	/* Workqueue that processes the plug/unplug requests. */
112 	struct work_struct wq;
113 	atomic_t wq_active;
114 	atomic_t config_changed;
115 
116 	/* Virtqueue for guest->host requests. */
117 	struct virtqueue *vq;
118 
119 	/* Wait for a host response to a guest request. */
120 	wait_queue_head_t host_resp;
121 
122 	/* Space for one guest request and the host response. */
123 	struct virtio_mem_req req;
124 	struct virtio_mem_resp resp;
125 
126 	/* The current size of the device. */
127 	uint64_t plugged_size;
128 	/* The requested size of the device. */
129 	uint64_t requested_size;
130 
131 	/* The device block size (for communicating with the device). */
132 	uint64_t device_block_size;
133 	/* The determined node id for all memory of the device. */
134 	int nid;
135 	/* Physical start address of the memory region. */
136 	uint64_t addr;
137 	/* Maximum region size in bytes. */
138 	uint64_t region_size;
139 
140 	/* The parent resource for all memory added via this device. */
141 	struct resource *parent_resource;
142 	/*
143 	 * Copy of "System RAM (virtio_mem)" to be used for
144 	 * add_memory_driver_managed().
145 	 */
146 	const char *resource_name;
147 	/* Memory group identification. */
148 	int mgid;
149 
150 	/*
151 	 * We don't want to add too much memory if it's not getting onlined,
152 	 * to avoid running OOM. Besides this threshold, we allow to have at
153 	 * least two offline blocks at a time (whatever is bigger).
154 	 */
155 #define VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD		(1024 * 1024 * 1024)
156 	atomic64_t offline_size;
157 	uint64_t offline_threshold;
158 
159 	/* If set, the driver is in SBM, otherwise in BBM. */
160 	bool in_sbm;
161 
162 	union {
163 		struct {
164 			/* Id of the first memory block of this device. */
165 			unsigned long first_mb_id;
166 			/* Id of the last usable memory block of this device. */
167 			unsigned long last_usable_mb_id;
168 			/* Id of the next memory bock to prepare when needed. */
169 			unsigned long next_mb_id;
170 
171 			/* The subblock size. */
172 			uint64_t sb_size;
173 			/* The number of subblocks per Linux memory block. */
174 			uint32_t sbs_per_mb;
175 
176 			/* Summary of all memory block states. */
177 			unsigned long mb_count[VIRTIO_MEM_SBM_MB_COUNT];
178 
179 			/*
180 			 * One byte state per memory block. Allocated via
181 			 * vmalloc(). Resized (alloc+copy+free) on demand.
182 			 *
183 			 * With 128 MiB memory blocks, we have states for 512
184 			 * GiB of memory in one 4 KiB page.
185 			 */
186 			uint8_t *mb_states;
187 
188 			/*
189 			 * Bitmap: one bit per subblock. Allocated similar to
190 			 * sbm.mb_states.
191 			 *
192 			 * A set bit means the corresponding subblock is
193 			 * plugged, otherwise it's unblocked.
194 			 *
195 			 * With 4 MiB subblocks, we manage 128 GiB of memory
196 			 * in one 4 KiB page.
197 			 */
198 			unsigned long *sb_states;
199 		} sbm;
200 
201 		struct {
202 			/* Id of the first big block of this device. */
203 			unsigned long first_bb_id;
204 			/* Id of the last usable big block of this device. */
205 			unsigned long last_usable_bb_id;
206 			/* Id of the next device bock to prepare when needed. */
207 			unsigned long next_bb_id;
208 
209 			/* Summary of all big block states. */
210 			unsigned long bb_count[VIRTIO_MEM_BBM_BB_COUNT];
211 
212 			/* One byte state per big block. See sbm.mb_states. */
213 			uint8_t *bb_states;
214 
215 			/* The block size used for plugging/adding/removing. */
216 			uint64_t bb_size;
217 		} bbm;
218 	};
219 
220 	/*
221 	 * Mutex that protects the sbm.mb_count, sbm.mb_states,
222 	 * sbm.sb_states, bbm.bb_count, and bbm.bb_states
223 	 *
224 	 * When this lock is held the pointers can't change, ONLINE and
225 	 * OFFLINE blocks can't change the state and no subblocks will get
226 	 * plugged/unplugged.
227 	 *
228 	 * In kdump mode, used to serialize requests, last_block_addr and
229 	 * last_block_plugged.
230 	 */
231 	struct mutex hotplug_mutex;
232 	bool hotplug_active;
233 
234 	/* An error occurred we cannot handle - stop processing requests. */
235 	bool broken;
236 
237 	/* Cached valued of is_kdump_kernel() when the device was probed. */
238 	bool in_kdump;
239 
240 	/* The driver is being removed. */
241 	spinlock_t removal_lock;
242 	bool removing;
243 
244 	/* Timer for retrying to plug/unplug memory. */
245 	struct hrtimer retry_timer;
246 	unsigned int retry_timer_ms;
247 #define VIRTIO_MEM_RETRY_TIMER_MIN_MS		50000
248 #define VIRTIO_MEM_RETRY_TIMER_MAX_MS		300000
249 
250 	/* Memory notifier (online/offline events). */
251 	struct notifier_block memory_notifier;
252 
253 #ifdef CONFIG_PROC_VMCORE
254 	/* vmcore callback for /proc/vmcore handling in kdump mode */
255 	struct vmcore_cb vmcore_cb;
256 	uint64_t last_block_addr;
257 	bool last_block_plugged;
258 #endif /* CONFIG_PROC_VMCORE */
259 
260 	/* Next device in the list of virtio-mem devices. */
261 	struct list_head next;
262 };
263 
264 /*
265  * We have to share a single online_page callback among all virtio-mem
266  * devices. We use RCU to iterate the list in the callback.
267  */
268 static DEFINE_MUTEX(virtio_mem_mutex);
269 static LIST_HEAD(virtio_mem_devices);
270 
271 static void virtio_mem_online_page_cb(struct page *page, unsigned int order);
272 static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
273 						  unsigned long nr_pages);
274 static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
275 						   unsigned long nr_pages);
276 static void virtio_mem_retry(struct virtio_mem *vm);
277 static int virtio_mem_create_resource(struct virtio_mem *vm);
278 static void virtio_mem_delete_resource(struct virtio_mem *vm);
279 
280 /*
281  * Register a virtio-mem device so it will be considered for the online_page
282  * callback.
283  */
284 static int register_virtio_mem_device(struct virtio_mem *vm)
285 {
286 	int rc = 0;
287 
288 	/* First device registers the callback. */
289 	mutex_lock(&virtio_mem_mutex);
290 	if (list_empty(&virtio_mem_devices))
291 		rc = set_online_page_callback(&virtio_mem_online_page_cb);
292 	if (!rc)
293 		list_add_rcu(&vm->next, &virtio_mem_devices);
294 	mutex_unlock(&virtio_mem_mutex);
295 
296 	return rc;
297 }
298 
299 /*
300  * Unregister a virtio-mem device so it will no longer be considered for the
301  * online_page callback.
302  */
303 static void unregister_virtio_mem_device(struct virtio_mem *vm)
304 {
305 	/* Last device unregisters the callback. */
306 	mutex_lock(&virtio_mem_mutex);
307 	list_del_rcu(&vm->next);
308 	if (list_empty(&virtio_mem_devices))
309 		restore_online_page_callback(&virtio_mem_online_page_cb);
310 	mutex_unlock(&virtio_mem_mutex);
311 
312 	synchronize_rcu();
313 }
314 
315 /*
316  * Calculate the memory block id of a given address.
317  */
318 static unsigned long virtio_mem_phys_to_mb_id(unsigned long addr)
319 {
320 	return addr / memory_block_size_bytes();
321 }
322 
323 /*
324  * Calculate the physical start address of a given memory block id.
325  */
326 static unsigned long virtio_mem_mb_id_to_phys(unsigned long mb_id)
327 {
328 	return mb_id * memory_block_size_bytes();
329 }
330 
331 /*
332  * Calculate the big block id of a given address.
333  */
334 static unsigned long virtio_mem_phys_to_bb_id(struct virtio_mem *vm,
335 					      uint64_t addr)
336 {
337 	return addr / vm->bbm.bb_size;
338 }
339 
340 /*
341  * Calculate the physical start address of a given big block id.
342  */
343 static uint64_t virtio_mem_bb_id_to_phys(struct virtio_mem *vm,
344 					 unsigned long bb_id)
345 {
346 	return bb_id * vm->bbm.bb_size;
347 }
348 
349 /*
350  * Calculate the subblock id of a given address.
351  */
352 static unsigned long virtio_mem_phys_to_sb_id(struct virtio_mem *vm,
353 					      unsigned long addr)
354 {
355 	const unsigned long mb_id = virtio_mem_phys_to_mb_id(addr);
356 	const unsigned long mb_addr = virtio_mem_mb_id_to_phys(mb_id);
357 
358 	return (addr - mb_addr) / vm->sbm.sb_size;
359 }
360 
361 /*
362  * Set the state of a big block, taking care of the state counter.
363  */
364 static void virtio_mem_bbm_set_bb_state(struct virtio_mem *vm,
365 					unsigned long bb_id,
366 					enum virtio_mem_bbm_bb_state state)
367 {
368 	const unsigned long idx = bb_id - vm->bbm.first_bb_id;
369 	enum virtio_mem_bbm_bb_state old_state;
370 
371 	old_state = vm->bbm.bb_states[idx];
372 	vm->bbm.bb_states[idx] = state;
373 
374 	BUG_ON(vm->bbm.bb_count[old_state] == 0);
375 	vm->bbm.bb_count[old_state]--;
376 	vm->bbm.bb_count[state]++;
377 }
378 
379 /*
380  * Get the state of a big block.
381  */
382 static enum virtio_mem_bbm_bb_state virtio_mem_bbm_get_bb_state(struct virtio_mem *vm,
383 								unsigned long bb_id)
384 {
385 	return vm->bbm.bb_states[bb_id - vm->bbm.first_bb_id];
386 }
387 
388 /*
389  * Prepare the big block state array for the next big block.
390  */
391 static int virtio_mem_bbm_bb_states_prepare_next_bb(struct virtio_mem *vm)
392 {
393 	unsigned long old_bytes = vm->bbm.next_bb_id - vm->bbm.first_bb_id;
394 	unsigned long new_bytes = old_bytes + 1;
395 	int old_pages = PFN_UP(old_bytes);
396 	int new_pages = PFN_UP(new_bytes);
397 	uint8_t *new_array;
398 
399 	if (vm->bbm.bb_states && old_pages == new_pages)
400 		return 0;
401 
402 	new_array = vzalloc(new_pages * PAGE_SIZE);
403 	if (!new_array)
404 		return -ENOMEM;
405 
406 	mutex_lock(&vm->hotplug_mutex);
407 	if (vm->bbm.bb_states)
408 		memcpy(new_array, vm->bbm.bb_states, old_pages * PAGE_SIZE);
409 	vfree(vm->bbm.bb_states);
410 	vm->bbm.bb_states = new_array;
411 	mutex_unlock(&vm->hotplug_mutex);
412 
413 	return 0;
414 }
415 
416 #define virtio_mem_bbm_for_each_bb(_vm, _bb_id, _state) \
417 	for (_bb_id = vm->bbm.first_bb_id; \
418 	     _bb_id < vm->bbm.next_bb_id && _vm->bbm.bb_count[_state]; \
419 	     _bb_id++) \
420 		if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
421 
422 #define virtio_mem_bbm_for_each_bb_rev(_vm, _bb_id, _state) \
423 	for (_bb_id = vm->bbm.next_bb_id - 1; \
424 	     _bb_id >= vm->bbm.first_bb_id && _vm->bbm.bb_count[_state]; \
425 	     _bb_id--) \
426 		if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
427 
428 /*
429  * Set the state of a memory block, taking care of the state counter.
430  */
431 static void virtio_mem_sbm_set_mb_state(struct virtio_mem *vm,
432 					unsigned long mb_id, uint8_t state)
433 {
434 	const unsigned long idx = mb_id - vm->sbm.first_mb_id;
435 	uint8_t old_state;
436 
437 	old_state = vm->sbm.mb_states[idx];
438 	vm->sbm.mb_states[idx] = state;
439 
440 	BUG_ON(vm->sbm.mb_count[old_state] == 0);
441 	vm->sbm.mb_count[old_state]--;
442 	vm->sbm.mb_count[state]++;
443 }
444 
445 /*
446  * Get the state of a memory block.
447  */
448 static uint8_t virtio_mem_sbm_get_mb_state(struct virtio_mem *vm,
449 					   unsigned long mb_id)
450 {
451 	const unsigned long idx = mb_id - vm->sbm.first_mb_id;
452 
453 	return vm->sbm.mb_states[idx];
454 }
455 
456 /*
457  * Prepare the state array for the next memory block.
458  */
459 static int virtio_mem_sbm_mb_states_prepare_next_mb(struct virtio_mem *vm)
460 {
461 	int old_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id);
462 	int new_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id + 1);
463 	uint8_t *new_array;
464 
465 	if (vm->sbm.mb_states && old_pages == new_pages)
466 		return 0;
467 
468 	new_array = vzalloc(new_pages * PAGE_SIZE);
469 	if (!new_array)
470 		return -ENOMEM;
471 
472 	mutex_lock(&vm->hotplug_mutex);
473 	if (vm->sbm.mb_states)
474 		memcpy(new_array, vm->sbm.mb_states, old_pages * PAGE_SIZE);
475 	vfree(vm->sbm.mb_states);
476 	vm->sbm.mb_states = new_array;
477 	mutex_unlock(&vm->hotplug_mutex);
478 
479 	return 0;
480 }
481 
482 #define virtio_mem_sbm_for_each_mb(_vm, _mb_id, _state) \
483 	for (_mb_id = _vm->sbm.first_mb_id; \
484 	     _mb_id < _vm->sbm.next_mb_id && _vm->sbm.mb_count[_state]; \
485 	     _mb_id++) \
486 		if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
487 
488 #define virtio_mem_sbm_for_each_mb_rev(_vm, _mb_id, _state) \
489 	for (_mb_id = _vm->sbm.next_mb_id - 1; \
490 	     _mb_id >= _vm->sbm.first_mb_id && _vm->sbm.mb_count[_state]; \
491 	     _mb_id--) \
492 		if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
493 
494 /*
495  * Calculate the bit number in the subblock bitmap for the given subblock
496  * inside the given memory block.
497  */
498 static int virtio_mem_sbm_sb_state_bit_nr(struct virtio_mem *vm,
499 					  unsigned long mb_id, int sb_id)
500 {
501 	return (mb_id - vm->sbm.first_mb_id) * vm->sbm.sbs_per_mb + sb_id;
502 }
503 
504 /*
505  * Mark all selected subblocks plugged.
506  *
507  * Will not modify the state of the memory block.
508  */
509 static void virtio_mem_sbm_set_sb_plugged(struct virtio_mem *vm,
510 					  unsigned long mb_id, int sb_id,
511 					  int count)
512 {
513 	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
514 
515 	__bitmap_set(vm->sbm.sb_states, bit, count);
516 }
517 
518 /*
519  * Mark all selected subblocks unplugged.
520  *
521  * Will not modify the state of the memory block.
522  */
523 static void virtio_mem_sbm_set_sb_unplugged(struct virtio_mem *vm,
524 					    unsigned long mb_id, int sb_id,
525 					    int count)
526 {
527 	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
528 
529 	__bitmap_clear(vm->sbm.sb_states, bit, count);
530 }
531 
532 /*
533  * Test if all selected subblocks are plugged.
534  */
535 static bool virtio_mem_sbm_test_sb_plugged(struct virtio_mem *vm,
536 					   unsigned long mb_id, int sb_id,
537 					   int count)
538 {
539 	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
540 
541 	if (count == 1)
542 		return test_bit(bit, vm->sbm.sb_states);
543 
544 	/* TODO: Helper similar to bitmap_set() */
545 	return find_next_zero_bit(vm->sbm.sb_states, bit + count, bit) >=
546 	       bit + count;
547 }
548 
549 /*
550  * Test if all selected subblocks are unplugged.
551  */
552 static bool virtio_mem_sbm_test_sb_unplugged(struct virtio_mem *vm,
553 					     unsigned long mb_id, int sb_id,
554 					     int count)
555 {
556 	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
557 
558 	/* TODO: Helper similar to bitmap_set() */
559 	return find_next_bit(vm->sbm.sb_states, bit + count, bit) >=
560 	       bit + count;
561 }
562 
563 /*
564  * Find the first unplugged subblock. Returns vm->sbm.sbs_per_mb in case there is
565  * none.
566  */
567 static int virtio_mem_sbm_first_unplugged_sb(struct virtio_mem *vm,
568 					    unsigned long mb_id)
569 {
570 	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, 0);
571 
572 	return find_next_zero_bit(vm->sbm.sb_states,
573 				  bit + vm->sbm.sbs_per_mb, bit) - bit;
574 }
575 
576 /*
577  * Prepare the subblock bitmap for the next memory block.
578  */
579 static int virtio_mem_sbm_sb_states_prepare_next_mb(struct virtio_mem *vm)
580 {
581 	const unsigned long old_nb_mb = vm->sbm.next_mb_id - vm->sbm.first_mb_id;
582 	const unsigned long old_nb_bits = old_nb_mb * vm->sbm.sbs_per_mb;
583 	const unsigned long new_nb_bits = (old_nb_mb + 1) * vm->sbm.sbs_per_mb;
584 	int old_pages = PFN_UP(BITS_TO_LONGS(old_nb_bits) * sizeof(long));
585 	int new_pages = PFN_UP(BITS_TO_LONGS(new_nb_bits) * sizeof(long));
586 	unsigned long *new_bitmap, *old_bitmap;
587 
588 	if (vm->sbm.sb_states && old_pages == new_pages)
589 		return 0;
590 
591 	new_bitmap = vzalloc(new_pages * PAGE_SIZE);
592 	if (!new_bitmap)
593 		return -ENOMEM;
594 
595 	mutex_lock(&vm->hotplug_mutex);
596 	if (vm->sbm.sb_states)
597 		memcpy(new_bitmap, vm->sbm.sb_states, old_pages * PAGE_SIZE);
598 
599 	old_bitmap = vm->sbm.sb_states;
600 	vm->sbm.sb_states = new_bitmap;
601 	mutex_unlock(&vm->hotplug_mutex);
602 
603 	vfree(old_bitmap);
604 	return 0;
605 }
606 
607 /*
608  * Test if we could add memory without creating too much offline memory -
609  * to avoid running OOM if memory is getting onlined deferred.
610  */
611 static bool virtio_mem_could_add_memory(struct virtio_mem *vm, uint64_t size)
612 {
613 	if (WARN_ON_ONCE(size > vm->offline_threshold))
614 		return false;
615 
616 	return atomic64_read(&vm->offline_size) + size <= vm->offline_threshold;
617 }
618 
619 /*
620  * Try adding memory to Linux. Will usually only fail if out of memory.
621  *
622  * Must not be called with the vm->hotplug_mutex held (possible deadlock with
623  * onlining code).
624  *
625  * Will not modify the state of memory blocks in virtio-mem.
626  */
627 static int virtio_mem_add_memory(struct virtio_mem *vm, uint64_t addr,
628 				 uint64_t size)
629 {
630 	int rc;
631 
632 	/*
633 	 * When force-unloading the driver and we still have memory added to
634 	 * Linux, the resource name has to stay.
635 	 */
636 	if (!vm->resource_name) {
637 		vm->resource_name = kstrdup_const("System RAM (virtio_mem)",
638 						  GFP_KERNEL);
639 		if (!vm->resource_name)
640 			return -ENOMEM;
641 	}
642 
643 	dev_dbg(&vm->vdev->dev, "adding memory: 0x%llx - 0x%llx\n", addr,
644 		addr + size - 1);
645 	/* Memory might get onlined immediately. */
646 	atomic64_add(size, &vm->offline_size);
647 	rc = add_memory_driver_managed(vm->mgid, addr, size, vm->resource_name,
648 				       MHP_MERGE_RESOURCE | MHP_NID_IS_MGID);
649 	if (rc) {
650 		atomic64_sub(size, &vm->offline_size);
651 		dev_warn(&vm->vdev->dev, "adding memory failed: %d\n", rc);
652 		/*
653 		 * TODO: Linux MM does not properly clean up yet in all cases
654 		 * where adding of memory failed - especially on -ENOMEM.
655 		 */
656 	}
657 	return rc;
658 }
659 
660 /*
661  * See virtio_mem_add_memory(): Try adding a single Linux memory block.
662  */
663 static int virtio_mem_sbm_add_mb(struct virtio_mem *vm, unsigned long mb_id)
664 {
665 	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
666 	const uint64_t size = memory_block_size_bytes();
667 
668 	return virtio_mem_add_memory(vm, addr, size);
669 }
670 
671 /*
672  * See virtio_mem_add_memory(): Try adding a big block.
673  */
674 static int virtio_mem_bbm_add_bb(struct virtio_mem *vm, unsigned long bb_id)
675 {
676 	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
677 	const uint64_t size = vm->bbm.bb_size;
678 
679 	return virtio_mem_add_memory(vm, addr, size);
680 }
681 
682 /*
683  * Try removing memory from Linux. Will only fail if memory blocks aren't
684  * offline.
685  *
686  * Must not be called with the vm->hotplug_mutex held (possible deadlock with
687  * onlining code).
688  *
689  * Will not modify the state of memory blocks in virtio-mem.
690  */
691 static int virtio_mem_remove_memory(struct virtio_mem *vm, uint64_t addr,
692 				    uint64_t size)
693 {
694 	int rc;
695 
696 	dev_dbg(&vm->vdev->dev, "removing memory: 0x%llx - 0x%llx\n", addr,
697 		addr + size - 1);
698 	rc = remove_memory(addr, size);
699 	if (!rc) {
700 		atomic64_sub(size, &vm->offline_size);
701 		/*
702 		 * We might have freed up memory we can now unplug, retry
703 		 * immediately instead of waiting.
704 		 */
705 		virtio_mem_retry(vm);
706 	} else {
707 		dev_dbg(&vm->vdev->dev, "removing memory failed: %d\n", rc);
708 	}
709 	return rc;
710 }
711 
712 /*
713  * See virtio_mem_remove_memory(): Try removing a single Linux memory block.
714  */
715 static int virtio_mem_sbm_remove_mb(struct virtio_mem *vm, unsigned long mb_id)
716 {
717 	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
718 	const uint64_t size = memory_block_size_bytes();
719 
720 	return virtio_mem_remove_memory(vm, addr, size);
721 }
722 
723 /*
724  * Try offlining and removing memory from Linux.
725  *
726  * Must not be called with the vm->hotplug_mutex held (possible deadlock with
727  * onlining code).
728  *
729  * Will not modify the state of memory blocks in virtio-mem.
730  */
731 static int virtio_mem_offline_and_remove_memory(struct virtio_mem *vm,
732 						uint64_t addr,
733 						uint64_t size)
734 {
735 	int rc;
736 
737 	dev_dbg(&vm->vdev->dev,
738 		"offlining and removing memory: 0x%llx - 0x%llx\n", addr,
739 		addr + size - 1);
740 
741 	rc = offline_and_remove_memory(addr, size);
742 	if (!rc) {
743 		atomic64_sub(size, &vm->offline_size);
744 		/*
745 		 * We might have freed up memory we can now unplug, retry
746 		 * immediately instead of waiting.
747 		 */
748 		virtio_mem_retry(vm);
749 	} else {
750 		dev_dbg(&vm->vdev->dev,
751 			"offlining and removing memory failed: %d\n", rc);
752 	}
753 	return rc;
754 }
755 
756 /*
757  * See virtio_mem_offline_and_remove_memory(): Try offlining and removing
758  * a single Linux memory block.
759  */
760 static int virtio_mem_sbm_offline_and_remove_mb(struct virtio_mem *vm,
761 						unsigned long mb_id)
762 {
763 	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
764 	const uint64_t size = memory_block_size_bytes();
765 
766 	return virtio_mem_offline_and_remove_memory(vm, addr, size);
767 }
768 
769 /*
770  * See virtio_mem_offline_and_remove_memory(): Try to offline and remove a
771  * all Linux memory blocks covered by the big block.
772  */
773 static int virtio_mem_bbm_offline_and_remove_bb(struct virtio_mem *vm,
774 						unsigned long bb_id)
775 {
776 	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
777 	const uint64_t size = vm->bbm.bb_size;
778 
779 	return virtio_mem_offline_and_remove_memory(vm, addr, size);
780 }
781 
782 /*
783  * Trigger the workqueue so the device can perform its magic.
784  */
785 static void virtio_mem_retry(struct virtio_mem *vm)
786 {
787 	unsigned long flags;
788 
789 	spin_lock_irqsave(&vm->removal_lock, flags);
790 	if (!vm->removing)
791 		queue_work(system_freezable_wq, &vm->wq);
792 	spin_unlock_irqrestore(&vm->removal_lock, flags);
793 }
794 
795 static int virtio_mem_translate_node_id(struct virtio_mem *vm, uint16_t node_id)
796 {
797 	int node = NUMA_NO_NODE;
798 
799 #if defined(CONFIG_ACPI_NUMA)
800 	if (virtio_has_feature(vm->vdev, VIRTIO_MEM_F_ACPI_PXM))
801 		node = pxm_to_node(node_id);
802 #endif
803 	return node;
804 }
805 
806 /*
807  * Test if a virtio-mem device overlaps with the given range. Can be called
808  * from (notifier) callbacks lockless.
809  */
810 static bool virtio_mem_overlaps_range(struct virtio_mem *vm, uint64_t start,
811 				      uint64_t size)
812 {
813 	return start < vm->addr + vm->region_size && vm->addr < start + size;
814 }
815 
816 /*
817  * Test if a virtio-mem device contains a given range. Can be called from
818  * (notifier) callbacks lockless.
819  */
820 static bool virtio_mem_contains_range(struct virtio_mem *vm, uint64_t start,
821 				      uint64_t size)
822 {
823 	return start >= vm->addr && start + size <= vm->addr + vm->region_size;
824 }
825 
826 static int virtio_mem_sbm_notify_going_online(struct virtio_mem *vm,
827 					      unsigned long mb_id)
828 {
829 	switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
830 	case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
831 	case VIRTIO_MEM_SBM_MB_OFFLINE:
832 		return NOTIFY_OK;
833 	default:
834 		break;
835 	}
836 	dev_warn_ratelimited(&vm->vdev->dev,
837 			     "memory block onlining denied\n");
838 	return NOTIFY_BAD;
839 }
840 
841 static void virtio_mem_sbm_notify_offline(struct virtio_mem *vm,
842 					  unsigned long mb_id)
843 {
844 	switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
845 	case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
846 	case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
847 		virtio_mem_sbm_set_mb_state(vm, mb_id,
848 					    VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
849 		break;
850 	case VIRTIO_MEM_SBM_MB_KERNEL:
851 	case VIRTIO_MEM_SBM_MB_MOVABLE:
852 		virtio_mem_sbm_set_mb_state(vm, mb_id,
853 					    VIRTIO_MEM_SBM_MB_OFFLINE);
854 		break;
855 	default:
856 		BUG();
857 		break;
858 	}
859 }
860 
861 static void virtio_mem_sbm_notify_online(struct virtio_mem *vm,
862 					 unsigned long mb_id,
863 					 unsigned long start_pfn)
864 {
865 	const bool is_movable = page_zonenum(pfn_to_page(start_pfn)) ==
866 				ZONE_MOVABLE;
867 	int new_state;
868 
869 	switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
870 	case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
871 		new_state = VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL;
872 		if (is_movable)
873 			new_state = VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL;
874 		break;
875 	case VIRTIO_MEM_SBM_MB_OFFLINE:
876 		new_state = VIRTIO_MEM_SBM_MB_KERNEL;
877 		if (is_movable)
878 			new_state = VIRTIO_MEM_SBM_MB_MOVABLE;
879 		break;
880 	default:
881 		BUG();
882 		break;
883 	}
884 	virtio_mem_sbm_set_mb_state(vm, mb_id, new_state);
885 }
886 
887 static void virtio_mem_sbm_notify_going_offline(struct virtio_mem *vm,
888 						unsigned long mb_id)
889 {
890 	const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
891 	unsigned long pfn;
892 	int sb_id;
893 
894 	for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
895 		if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
896 			continue;
897 		pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
898 			       sb_id * vm->sbm.sb_size);
899 		virtio_mem_fake_offline_going_offline(pfn, nr_pages);
900 	}
901 }
902 
903 static void virtio_mem_sbm_notify_cancel_offline(struct virtio_mem *vm,
904 						 unsigned long mb_id)
905 {
906 	const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
907 	unsigned long pfn;
908 	int sb_id;
909 
910 	for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
911 		if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
912 			continue;
913 		pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
914 			       sb_id * vm->sbm.sb_size);
915 		virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
916 	}
917 }
918 
919 static void virtio_mem_bbm_notify_going_offline(struct virtio_mem *vm,
920 						unsigned long bb_id,
921 						unsigned long pfn,
922 						unsigned long nr_pages)
923 {
924 	/*
925 	 * When marked as "fake-offline", all online memory of this device block
926 	 * is allocated by us. Otherwise, we don't have any memory allocated.
927 	 */
928 	if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
929 	    VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
930 		return;
931 	virtio_mem_fake_offline_going_offline(pfn, nr_pages);
932 }
933 
934 static void virtio_mem_bbm_notify_cancel_offline(struct virtio_mem *vm,
935 						 unsigned long bb_id,
936 						 unsigned long pfn,
937 						 unsigned long nr_pages)
938 {
939 	if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
940 	    VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
941 		return;
942 	virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
943 }
944 
945 /*
946  * This callback will either be called synchronously from add_memory() or
947  * asynchronously (e.g., triggered via user space). We have to be careful
948  * with locking when calling add_memory().
949  */
950 static int virtio_mem_memory_notifier_cb(struct notifier_block *nb,
951 					 unsigned long action, void *arg)
952 {
953 	struct virtio_mem *vm = container_of(nb, struct virtio_mem,
954 					     memory_notifier);
955 	struct memory_notify *mhp = arg;
956 	const unsigned long start = PFN_PHYS(mhp->start_pfn);
957 	const unsigned long size = PFN_PHYS(mhp->nr_pages);
958 	int rc = NOTIFY_OK;
959 	unsigned long id;
960 
961 	if (!virtio_mem_overlaps_range(vm, start, size))
962 		return NOTIFY_DONE;
963 
964 	if (vm->in_sbm) {
965 		id = virtio_mem_phys_to_mb_id(start);
966 		/*
967 		 * In SBM, we add memory in separate memory blocks - we expect
968 		 * it to be onlined/offlined in the same granularity. Bail out
969 		 * if this ever changes.
970 		 */
971 		if (WARN_ON_ONCE(size != memory_block_size_bytes() ||
972 				 !IS_ALIGNED(start, memory_block_size_bytes())))
973 			return NOTIFY_BAD;
974 	} else {
975 		id = virtio_mem_phys_to_bb_id(vm, start);
976 		/*
977 		 * In BBM, we only care about onlining/offlining happening
978 		 * within a single big block, we don't care about the
979 		 * actual granularity as we don't track individual Linux
980 		 * memory blocks.
981 		 */
982 		if (WARN_ON_ONCE(id != virtio_mem_phys_to_bb_id(vm, start + size - 1)))
983 			return NOTIFY_BAD;
984 	}
985 
986 	/*
987 	 * Avoid circular locking lockdep warnings. We lock the mutex
988 	 * e.g., in MEM_GOING_ONLINE and unlock it in MEM_ONLINE. The
989 	 * blocking_notifier_call_chain() has it's own lock, which gets unlocked
990 	 * between both notifier calls and will bail out. False positive.
991 	 */
992 	lockdep_off();
993 
994 	switch (action) {
995 	case MEM_GOING_OFFLINE:
996 		mutex_lock(&vm->hotplug_mutex);
997 		if (vm->removing) {
998 			rc = notifier_from_errno(-EBUSY);
999 			mutex_unlock(&vm->hotplug_mutex);
1000 			break;
1001 		}
1002 		vm->hotplug_active = true;
1003 		if (vm->in_sbm)
1004 			virtio_mem_sbm_notify_going_offline(vm, id);
1005 		else
1006 			virtio_mem_bbm_notify_going_offline(vm, id,
1007 							    mhp->start_pfn,
1008 							    mhp->nr_pages);
1009 		break;
1010 	case MEM_GOING_ONLINE:
1011 		mutex_lock(&vm->hotplug_mutex);
1012 		if (vm->removing) {
1013 			rc = notifier_from_errno(-EBUSY);
1014 			mutex_unlock(&vm->hotplug_mutex);
1015 			break;
1016 		}
1017 		vm->hotplug_active = true;
1018 		if (vm->in_sbm)
1019 			rc = virtio_mem_sbm_notify_going_online(vm, id);
1020 		break;
1021 	case MEM_OFFLINE:
1022 		if (vm->in_sbm)
1023 			virtio_mem_sbm_notify_offline(vm, id);
1024 
1025 		atomic64_add(size, &vm->offline_size);
1026 		/*
1027 		 * Trigger the workqueue. Now that we have some offline memory,
1028 		 * maybe we can handle pending unplug requests.
1029 		 */
1030 		if (!unplug_online)
1031 			virtio_mem_retry(vm);
1032 
1033 		vm->hotplug_active = false;
1034 		mutex_unlock(&vm->hotplug_mutex);
1035 		break;
1036 	case MEM_ONLINE:
1037 		if (vm->in_sbm)
1038 			virtio_mem_sbm_notify_online(vm, id, mhp->start_pfn);
1039 
1040 		atomic64_sub(size, &vm->offline_size);
1041 		/*
1042 		 * Start adding more memory once we onlined half of our
1043 		 * threshold. Don't trigger if it's possibly due to our actipn
1044 		 * (e.g., us adding memory which gets onlined immediately from
1045 		 * the core).
1046 		 */
1047 		if (!atomic_read(&vm->wq_active) &&
1048 		    virtio_mem_could_add_memory(vm, vm->offline_threshold / 2))
1049 			virtio_mem_retry(vm);
1050 
1051 		vm->hotplug_active = false;
1052 		mutex_unlock(&vm->hotplug_mutex);
1053 		break;
1054 	case MEM_CANCEL_OFFLINE:
1055 		if (!vm->hotplug_active)
1056 			break;
1057 		if (vm->in_sbm)
1058 			virtio_mem_sbm_notify_cancel_offline(vm, id);
1059 		else
1060 			virtio_mem_bbm_notify_cancel_offline(vm, id,
1061 							     mhp->start_pfn,
1062 							     mhp->nr_pages);
1063 		vm->hotplug_active = false;
1064 		mutex_unlock(&vm->hotplug_mutex);
1065 		break;
1066 	case MEM_CANCEL_ONLINE:
1067 		if (!vm->hotplug_active)
1068 			break;
1069 		vm->hotplug_active = false;
1070 		mutex_unlock(&vm->hotplug_mutex);
1071 		break;
1072 	default:
1073 		break;
1074 	}
1075 
1076 	lockdep_on();
1077 
1078 	return rc;
1079 }
1080 
1081 /*
1082  * Set a range of pages PG_offline. Remember pages that were never onlined
1083  * (via generic_online_page()) using PageDirty().
1084  */
1085 static void virtio_mem_set_fake_offline(unsigned long pfn,
1086 					unsigned long nr_pages, bool onlined)
1087 {
1088 	page_offline_begin();
1089 	for (; nr_pages--; pfn++) {
1090 		struct page *page = pfn_to_page(pfn);
1091 
1092 		__SetPageOffline(page);
1093 		if (!onlined) {
1094 			SetPageDirty(page);
1095 			/* FIXME: remove after cleanups */
1096 			ClearPageReserved(page);
1097 		}
1098 	}
1099 	page_offline_end();
1100 }
1101 
1102 /*
1103  * Clear PG_offline from a range of pages. If the pages were never onlined,
1104  * (via generic_online_page()), clear PageDirty().
1105  */
1106 static void virtio_mem_clear_fake_offline(unsigned long pfn,
1107 					  unsigned long nr_pages, bool onlined)
1108 {
1109 	for (; nr_pages--; pfn++) {
1110 		struct page *page = pfn_to_page(pfn);
1111 
1112 		__ClearPageOffline(page);
1113 		if (!onlined)
1114 			ClearPageDirty(page);
1115 	}
1116 }
1117 
1118 /*
1119  * Release a range of fake-offline pages to the buddy, effectively
1120  * fake-onlining them.
1121  */
1122 static void virtio_mem_fake_online(unsigned long pfn, unsigned long nr_pages)
1123 {
1124 	unsigned long order = MAX_ORDER - 1;
1125 	unsigned long i;
1126 
1127 	/*
1128 	 * We might get called for ranges that don't cover properly aligned
1129 	 * MAX_ORDER - 1 pages; however, we can only online properly aligned
1130 	 * pages with an order of MAX_ORDER - 1 at maximum.
1131 	 */
1132 	while (!IS_ALIGNED(pfn | nr_pages, 1 << order))
1133 		order--;
1134 
1135 	for (i = 0; i < nr_pages; i += 1 << order) {
1136 		struct page *page = pfn_to_page(pfn + i);
1137 
1138 		/*
1139 		 * If the page is PageDirty(), it was kept fake-offline when
1140 		 * onlining the memory block. Otherwise, it was allocated
1141 		 * using alloc_contig_range(). All pages in a subblock are
1142 		 * alike.
1143 		 */
1144 		if (PageDirty(page)) {
1145 			virtio_mem_clear_fake_offline(pfn + i, 1 << order, false);
1146 			generic_online_page(page, order);
1147 		} else {
1148 			virtio_mem_clear_fake_offline(pfn + i, 1 << order, true);
1149 			free_contig_range(pfn + i, 1 << order);
1150 			adjust_managed_page_count(page, 1 << order);
1151 		}
1152 	}
1153 }
1154 
1155 /*
1156  * Try to allocate a range, marking pages fake-offline, effectively
1157  * fake-offlining them.
1158  */
1159 static int virtio_mem_fake_offline(unsigned long pfn, unsigned long nr_pages)
1160 {
1161 	const bool is_movable = page_zonenum(pfn_to_page(pfn)) ==
1162 				ZONE_MOVABLE;
1163 	int rc, retry_count;
1164 
1165 	/*
1166 	 * TODO: We want an alloc_contig_range() mode that tries to allocate
1167 	 * harder (e.g., dealing with temporarily pinned pages, PCP), especially
1168 	 * with ZONE_MOVABLE. So for now, retry a couple of times with
1169 	 * ZONE_MOVABLE before giving up - because that zone is supposed to give
1170 	 * some guarantees.
1171 	 */
1172 	for (retry_count = 0; retry_count < 5; retry_count++) {
1173 		rc = alloc_contig_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE,
1174 					GFP_KERNEL);
1175 		if (rc == -ENOMEM)
1176 			/* whoops, out of memory */
1177 			return rc;
1178 		else if (rc && !is_movable)
1179 			break;
1180 		else if (rc)
1181 			continue;
1182 
1183 		virtio_mem_set_fake_offline(pfn, nr_pages, true);
1184 		adjust_managed_page_count(pfn_to_page(pfn), -nr_pages);
1185 		return 0;
1186 	}
1187 
1188 	return -EBUSY;
1189 }
1190 
1191 /*
1192  * Handle fake-offline pages when memory is going offline - such that the
1193  * pages can be skipped by mm-core when offlining.
1194  */
1195 static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
1196 						  unsigned long nr_pages)
1197 {
1198 	struct page *page;
1199 	unsigned long i;
1200 
1201 	/*
1202 	 * Drop our reference to the pages so the memory can get offlined
1203 	 * and add the unplugged pages to the managed page counters (so
1204 	 * offlining code can correctly subtract them again).
1205 	 */
1206 	adjust_managed_page_count(pfn_to_page(pfn), nr_pages);
1207 	/* Drop our reference to the pages so the memory can get offlined. */
1208 	for (i = 0; i < nr_pages; i++) {
1209 		page = pfn_to_page(pfn + i);
1210 		if (WARN_ON(!page_ref_dec_and_test(page)))
1211 			dump_page(page, "fake-offline page referenced");
1212 	}
1213 }
1214 
1215 /*
1216  * Handle fake-offline pages when memory offlining is canceled - to undo
1217  * what we did in virtio_mem_fake_offline_going_offline().
1218  */
1219 static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
1220 						   unsigned long nr_pages)
1221 {
1222 	unsigned long i;
1223 
1224 	/*
1225 	 * Get the reference we dropped when going offline and subtract the
1226 	 * unplugged pages from the managed page counters.
1227 	 */
1228 	adjust_managed_page_count(pfn_to_page(pfn), -nr_pages);
1229 	for (i = 0; i < nr_pages; i++)
1230 		page_ref_inc(pfn_to_page(pfn + i));
1231 }
1232 
1233 static void virtio_mem_online_page(struct virtio_mem *vm,
1234 				   struct page *page, unsigned int order)
1235 {
1236 	const unsigned long start = page_to_phys(page);
1237 	const unsigned long end = start + PFN_PHYS(1 << order);
1238 	unsigned long addr, next, id, sb_id, count;
1239 	bool do_online;
1240 
1241 	/*
1242 	 * We can get called with any order up to MAX_ORDER - 1. If our
1243 	 * subblock size is smaller than that and we have a mixture of plugged
1244 	 * and unplugged subblocks within such a page, we have to process in
1245 	 * smaller granularity. In that case we'll adjust the order exactly once
1246 	 * within the loop.
1247 	 */
1248 	for (addr = start; addr < end; ) {
1249 		next = addr + PFN_PHYS(1 << order);
1250 
1251 		if (vm->in_sbm) {
1252 			id = virtio_mem_phys_to_mb_id(addr);
1253 			sb_id = virtio_mem_phys_to_sb_id(vm, addr);
1254 			count = virtio_mem_phys_to_sb_id(vm, next - 1) - sb_id + 1;
1255 
1256 			if (virtio_mem_sbm_test_sb_plugged(vm, id, sb_id, count)) {
1257 				/* Fully plugged. */
1258 				do_online = true;
1259 			} else if (count == 1 ||
1260 				   virtio_mem_sbm_test_sb_unplugged(vm, id, sb_id, count)) {
1261 				/* Fully unplugged. */
1262 				do_online = false;
1263 			} else {
1264 				/*
1265 				 * Mixture, process sub-blocks instead. This
1266 				 * will be at least the size of a pageblock.
1267 				 * We'll run into this case exactly once.
1268 				 */
1269 				order = ilog2(vm->sbm.sb_size) - PAGE_SHIFT;
1270 				do_online = virtio_mem_sbm_test_sb_plugged(vm, id, sb_id, 1);
1271 				continue;
1272 			}
1273 		} else {
1274 			/*
1275 			 * If the whole block is marked fake offline, keep
1276 			 * everything that way.
1277 			 */
1278 			id = virtio_mem_phys_to_bb_id(vm, addr);
1279 			do_online = virtio_mem_bbm_get_bb_state(vm, id) !=
1280 				    VIRTIO_MEM_BBM_BB_FAKE_OFFLINE;
1281 		}
1282 
1283 		if (do_online)
1284 			generic_online_page(pfn_to_page(PFN_DOWN(addr)), order);
1285 		else
1286 			virtio_mem_set_fake_offline(PFN_DOWN(addr), 1 << order,
1287 						    false);
1288 		addr = next;
1289 	}
1290 }
1291 
1292 static void virtio_mem_online_page_cb(struct page *page, unsigned int order)
1293 {
1294 	const unsigned long addr = page_to_phys(page);
1295 	struct virtio_mem *vm;
1296 
1297 	rcu_read_lock();
1298 	list_for_each_entry_rcu(vm, &virtio_mem_devices, next) {
1299 		/*
1300 		 * Pages we're onlining will never cross memory blocks and,
1301 		 * therefore, not virtio-mem devices.
1302 		 */
1303 		if (!virtio_mem_contains_range(vm, addr, PFN_PHYS(1 << order)))
1304 			continue;
1305 
1306 		/*
1307 		 * virtio_mem_set_fake_offline() might sleep. We can safely
1308 		 * drop the RCU lock at this point because the device
1309 		 * cannot go away. See virtio_mem_remove() how races
1310 		 * between memory onlining and device removal are handled.
1311 		 */
1312 		rcu_read_unlock();
1313 
1314 		virtio_mem_online_page(vm, page, order);
1315 		return;
1316 	}
1317 	rcu_read_unlock();
1318 
1319 	/* not virtio-mem memory, but e.g., a DIMM. online it */
1320 	generic_online_page(page, order);
1321 }
1322 
1323 static uint64_t virtio_mem_send_request(struct virtio_mem *vm,
1324 					const struct virtio_mem_req *req)
1325 {
1326 	struct scatterlist *sgs[2], sg_req, sg_resp;
1327 	unsigned int len;
1328 	int rc;
1329 
1330 	/* don't use the request residing on the stack (vaddr) */
1331 	vm->req = *req;
1332 
1333 	/* out: buffer for request */
1334 	sg_init_one(&sg_req, &vm->req, sizeof(vm->req));
1335 	sgs[0] = &sg_req;
1336 
1337 	/* in: buffer for response */
1338 	sg_init_one(&sg_resp, &vm->resp, sizeof(vm->resp));
1339 	sgs[1] = &sg_resp;
1340 
1341 	rc = virtqueue_add_sgs(vm->vq, sgs, 1, 1, vm, GFP_KERNEL);
1342 	if (rc < 0)
1343 		return rc;
1344 
1345 	virtqueue_kick(vm->vq);
1346 
1347 	/* wait for a response */
1348 	wait_event(vm->host_resp, virtqueue_get_buf(vm->vq, &len));
1349 
1350 	return virtio16_to_cpu(vm->vdev, vm->resp.type);
1351 }
1352 
1353 static int virtio_mem_send_plug_request(struct virtio_mem *vm, uint64_t addr,
1354 					uint64_t size)
1355 {
1356 	const uint64_t nb_vm_blocks = size / vm->device_block_size;
1357 	const struct virtio_mem_req req = {
1358 		.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_PLUG),
1359 		.u.plug.addr = cpu_to_virtio64(vm->vdev, addr),
1360 		.u.plug.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
1361 	};
1362 	int rc = -ENOMEM;
1363 
1364 	if (atomic_read(&vm->config_changed))
1365 		return -EAGAIN;
1366 
1367 	dev_dbg(&vm->vdev->dev, "plugging memory: 0x%llx - 0x%llx\n", addr,
1368 		addr + size - 1);
1369 
1370 	switch (virtio_mem_send_request(vm, &req)) {
1371 	case VIRTIO_MEM_RESP_ACK:
1372 		vm->plugged_size += size;
1373 		return 0;
1374 	case VIRTIO_MEM_RESP_NACK:
1375 		rc = -EAGAIN;
1376 		break;
1377 	case VIRTIO_MEM_RESP_BUSY:
1378 		rc = -ETXTBSY;
1379 		break;
1380 	case VIRTIO_MEM_RESP_ERROR:
1381 		rc = -EINVAL;
1382 		break;
1383 	default:
1384 		break;
1385 	}
1386 
1387 	dev_dbg(&vm->vdev->dev, "plugging memory failed: %d\n", rc);
1388 	return rc;
1389 }
1390 
1391 static int virtio_mem_send_unplug_request(struct virtio_mem *vm, uint64_t addr,
1392 					  uint64_t size)
1393 {
1394 	const uint64_t nb_vm_blocks = size / vm->device_block_size;
1395 	const struct virtio_mem_req req = {
1396 		.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_UNPLUG),
1397 		.u.unplug.addr = cpu_to_virtio64(vm->vdev, addr),
1398 		.u.unplug.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
1399 	};
1400 	int rc = -ENOMEM;
1401 
1402 	if (atomic_read(&vm->config_changed))
1403 		return -EAGAIN;
1404 
1405 	dev_dbg(&vm->vdev->dev, "unplugging memory: 0x%llx - 0x%llx\n", addr,
1406 		addr + size - 1);
1407 
1408 	switch (virtio_mem_send_request(vm, &req)) {
1409 	case VIRTIO_MEM_RESP_ACK:
1410 		vm->plugged_size -= size;
1411 		return 0;
1412 	case VIRTIO_MEM_RESP_BUSY:
1413 		rc = -ETXTBSY;
1414 		break;
1415 	case VIRTIO_MEM_RESP_ERROR:
1416 		rc = -EINVAL;
1417 		break;
1418 	default:
1419 		break;
1420 	}
1421 
1422 	dev_dbg(&vm->vdev->dev, "unplugging memory failed: %d\n", rc);
1423 	return rc;
1424 }
1425 
1426 static int virtio_mem_send_unplug_all_request(struct virtio_mem *vm)
1427 {
1428 	const struct virtio_mem_req req = {
1429 		.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_UNPLUG_ALL),
1430 	};
1431 	int rc = -ENOMEM;
1432 
1433 	dev_dbg(&vm->vdev->dev, "unplugging all memory");
1434 
1435 	switch (virtio_mem_send_request(vm, &req)) {
1436 	case VIRTIO_MEM_RESP_ACK:
1437 		vm->unplug_all_required = false;
1438 		vm->plugged_size = 0;
1439 		/* usable region might have shrunk */
1440 		atomic_set(&vm->config_changed, 1);
1441 		return 0;
1442 	case VIRTIO_MEM_RESP_BUSY:
1443 		rc = -ETXTBSY;
1444 		break;
1445 	default:
1446 		break;
1447 	}
1448 
1449 	dev_dbg(&vm->vdev->dev, "unplugging all memory failed: %d\n", rc);
1450 	return rc;
1451 }
1452 
1453 /*
1454  * Plug selected subblocks. Updates the plugged state, but not the state
1455  * of the memory block.
1456  */
1457 static int virtio_mem_sbm_plug_sb(struct virtio_mem *vm, unsigned long mb_id,
1458 				  int sb_id, int count)
1459 {
1460 	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
1461 			      sb_id * vm->sbm.sb_size;
1462 	const uint64_t size = count * vm->sbm.sb_size;
1463 	int rc;
1464 
1465 	rc = virtio_mem_send_plug_request(vm, addr, size);
1466 	if (!rc)
1467 		virtio_mem_sbm_set_sb_plugged(vm, mb_id, sb_id, count);
1468 	return rc;
1469 }
1470 
1471 /*
1472  * Unplug selected subblocks. Updates the plugged state, but not the state
1473  * of the memory block.
1474  */
1475 static int virtio_mem_sbm_unplug_sb(struct virtio_mem *vm, unsigned long mb_id,
1476 				    int sb_id, int count)
1477 {
1478 	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
1479 			      sb_id * vm->sbm.sb_size;
1480 	const uint64_t size = count * vm->sbm.sb_size;
1481 	int rc;
1482 
1483 	rc = virtio_mem_send_unplug_request(vm, addr, size);
1484 	if (!rc)
1485 		virtio_mem_sbm_set_sb_unplugged(vm, mb_id, sb_id, count);
1486 	return rc;
1487 }
1488 
1489 /*
1490  * Request to unplug a big block.
1491  *
1492  * Will not modify the state of the big block.
1493  */
1494 static int virtio_mem_bbm_unplug_bb(struct virtio_mem *vm, unsigned long bb_id)
1495 {
1496 	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
1497 	const uint64_t size = vm->bbm.bb_size;
1498 
1499 	return virtio_mem_send_unplug_request(vm, addr, size);
1500 }
1501 
1502 /*
1503  * Request to plug a big block.
1504  *
1505  * Will not modify the state of the big block.
1506  */
1507 static int virtio_mem_bbm_plug_bb(struct virtio_mem *vm, unsigned long bb_id)
1508 {
1509 	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
1510 	const uint64_t size = vm->bbm.bb_size;
1511 
1512 	return virtio_mem_send_plug_request(vm, addr, size);
1513 }
1514 
1515 /*
1516  * Unplug the desired number of plugged subblocks of a offline or not-added
1517  * memory block. Will fail if any subblock cannot get unplugged (instead of
1518  * skipping it).
1519  *
1520  * Will not modify the state of the memory block.
1521  *
1522  * Note: can fail after some subblocks were unplugged.
1523  */
1524 static int virtio_mem_sbm_unplug_any_sb_raw(struct virtio_mem *vm,
1525 					    unsigned long mb_id, uint64_t *nb_sb)
1526 {
1527 	int sb_id, count;
1528 	int rc;
1529 
1530 	sb_id = vm->sbm.sbs_per_mb - 1;
1531 	while (*nb_sb) {
1532 		/* Find the next candidate subblock */
1533 		while (sb_id >= 0 &&
1534 		       virtio_mem_sbm_test_sb_unplugged(vm, mb_id, sb_id, 1))
1535 			sb_id--;
1536 		if (sb_id < 0)
1537 			break;
1538 		/* Try to unplug multiple subblocks at a time */
1539 		count = 1;
1540 		while (count < *nb_sb && sb_id > 0 &&
1541 		       virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id - 1, 1)) {
1542 			count++;
1543 			sb_id--;
1544 		}
1545 
1546 		rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
1547 		if (rc)
1548 			return rc;
1549 		*nb_sb -= count;
1550 		sb_id--;
1551 	}
1552 
1553 	return 0;
1554 }
1555 
1556 /*
1557  * Unplug all plugged subblocks of an offline or not-added memory block.
1558  *
1559  * Will not modify the state of the memory block.
1560  *
1561  * Note: can fail after some subblocks were unplugged.
1562  */
1563 static int virtio_mem_sbm_unplug_mb(struct virtio_mem *vm, unsigned long mb_id)
1564 {
1565 	uint64_t nb_sb = vm->sbm.sbs_per_mb;
1566 
1567 	return virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, &nb_sb);
1568 }
1569 
1570 /*
1571  * Prepare tracking data for the next memory block.
1572  */
1573 static int virtio_mem_sbm_prepare_next_mb(struct virtio_mem *vm,
1574 					  unsigned long *mb_id)
1575 {
1576 	int rc;
1577 
1578 	if (vm->sbm.next_mb_id > vm->sbm.last_usable_mb_id)
1579 		return -ENOSPC;
1580 
1581 	/* Resize the state array if required. */
1582 	rc = virtio_mem_sbm_mb_states_prepare_next_mb(vm);
1583 	if (rc)
1584 		return rc;
1585 
1586 	/* Resize the subblock bitmap if required. */
1587 	rc = virtio_mem_sbm_sb_states_prepare_next_mb(vm);
1588 	if (rc)
1589 		return rc;
1590 
1591 	vm->sbm.mb_count[VIRTIO_MEM_SBM_MB_UNUSED]++;
1592 	*mb_id = vm->sbm.next_mb_id++;
1593 	return 0;
1594 }
1595 
1596 /*
1597  * Try to plug the desired number of subblocks and add the memory block
1598  * to Linux.
1599  *
1600  * Will modify the state of the memory block.
1601  */
1602 static int virtio_mem_sbm_plug_and_add_mb(struct virtio_mem *vm,
1603 					  unsigned long mb_id, uint64_t *nb_sb)
1604 {
1605 	const int count = min_t(int, *nb_sb, vm->sbm.sbs_per_mb);
1606 	int rc;
1607 
1608 	if (WARN_ON_ONCE(!count))
1609 		return -EINVAL;
1610 
1611 	/*
1612 	 * Plug the requested number of subblocks before adding it to linux,
1613 	 * so that onlining will directly online all plugged subblocks.
1614 	 */
1615 	rc = virtio_mem_sbm_plug_sb(vm, mb_id, 0, count);
1616 	if (rc)
1617 		return rc;
1618 
1619 	/*
1620 	 * Mark the block properly offline before adding it to Linux,
1621 	 * so the memory notifiers will find the block in the right state.
1622 	 */
1623 	if (count == vm->sbm.sbs_per_mb)
1624 		virtio_mem_sbm_set_mb_state(vm, mb_id,
1625 					    VIRTIO_MEM_SBM_MB_OFFLINE);
1626 	else
1627 		virtio_mem_sbm_set_mb_state(vm, mb_id,
1628 					    VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
1629 
1630 	/* Add the memory block to linux - if that fails, try to unplug. */
1631 	rc = virtio_mem_sbm_add_mb(vm, mb_id);
1632 	if (rc) {
1633 		int new_state = VIRTIO_MEM_SBM_MB_UNUSED;
1634 
1635 		if (virtio_mem_sbm_unplug_sb(vm, mb_id, 0, count))
1636 			new_state = VIRTIO_MEM_SBM_MB_PLUGGED;
1637 		virtio_mem_sbm_set_mb_state(vm, mb_id, new_state);
1638 		return rc;
1639 	}
1640 
1641 	*nb_sb -= count;
1642 	return 0;
1643 }
1644 
1645 /*
1646  * Try to plug the desired number of subblocks of a memory block that
1647  * is already added to Linux.
1648  *
1649  * Will modify the state of the memory block.
1650  *
1651  * Note: Can fail after some subblocks were successfully plugged.
1652  */
1653 static int virtio_mem_sbm_plug_any_sb(struct virtio_mem *vm,
1654 				      unsigned long mb_id, uint64_t *nb_sb)
1655 {
1656 	const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
1657 	unsigned long pfn, nr_pages;
1658 	int sb_id, count;
1659 	int rc;
1660 
1661 	if (WARN_ON_ONCE(!*nb_sb))
1662 		return -EINVAL;
1663 
1664 	while (*nb_sb) {
1665 		sb_id = virtio_mem_sbm_first_unplugged_sb(vm, mb_id);
1666 		if (sb_id >= vm->sbm.sbs_per_mb)
1667 			break;
1668 		count = 1;
1669 		while (count < *nb_sb &&
1670 		       sb_id + count < vm->sbm.sbs_per_mb &&
1671 		       !virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id + count, 1))
1672 			count++;
1673 
1674 		rc = virtio_mem_sbm_plug_sb(vm, mb_id, sb_id, count);
1675 		if (rc)
1676 			return rc;
1677 		*nb_sb -= count;
1678 		if (old_state == VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL)
1679 			continue;
1680 
1681 		/* fake-online the pages if the memory block is online */
1682 		pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
1683 			       sb_id * vm->sbm.sb_size);
1684 		nr_pages = PFN_DOWN(count * vm->sbm.sb_size);
1685 		virtio_mem_fake_online(pfn, nr_pages);
1686 	}
1687 
1688 	if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
1689 		virtio_mem_sbm_set_mb_state(vm, mb_id, old_state - 1);
1690 
1691 	return 0;
1692 }
1693 
1694 static int virtio_mem_sbm_plug_request(struct virtio_mem *vm, uint64_t diff)
1695 {
1696 	const int mb_states[] = {
1697 		VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
1698 		VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
1699 		VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
1700 	};
1701 	uint64_t nb_sb = diff / vm->sbm.sb_size;
1702 	unsigned long mb_id;
1703 	int rc, i;
1704 
1705 	if (!nb_sb)
1706 		return 0;
1707 
1708 	/* Don't race with onlining/offlining */
1709 	mutex_lock(&vm->hotplug_mutex);
1710 
1711 	for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
1712 		virtio_mem_sbm_for_each_mb(vm, mb_id, mb_states[i]) {
1713 			rc = virtio_mem_sbm_plug_any_sb(vm, mb_id, &nb_sb);
1714 			if (rc || !nb_sb)
1715 				goto out_unlock;
1716 			cond_resched();
1717 		}
1718 	}
1719 
1720 	/*
1721 	 * We won't be working on online/offline memory blocks from this point,
1722 	 * so we can't race with memory onlining/offlining. Drop the mutex.
1723 	 */
1724 	mutex_unlock(&vm->hotplug_mutex);
1725 
1726 	/* Try to plug and add unused blocks */
1727 	virtio_mem_sbm_for_each_mb(vm, mb_id, VIRTIO_MEM_SBM_MB_UNUSED) {
1728 		if (!virtio_mem_could_add_memory(vm, memory_block_size_bytes()))
1729 			return -ENOSPC;
1730 
1731 		rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, &nb_sb);
1732 		if (rc || !nb_sb)
1733 			return rc;
1734 		cond_resched();
1735 	}
1736 
1737 	/* Try to prepare, plug and add new blocks */
1738 	while (nb_sb) {
1739 		if (!virtio_mem_could_add_memory(vm, memory_block_size_bytes()))
1740 			return -ENOSPC;
1741 
1742 		rc = virtio_mem_sbm_prepare_next_mb(vm, &mb_id);
1743 		if (rc)
1744 			return rc;
1745 		rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, &nb_sb);
1746 		if (rc)
1747 			return rc;
1748 		cond_resched();
1749 	}
1750 
1751 	return 0;
1752 out_unlock:
1753 	mutex_unlock(&vm->hotplug_mutex);
1754 	return rc;
1755 }
1756 
1757 /*
1758  * Plug a big block and add it to Linux.
1759  *
1760  * Will modify the state of the big block.
1761  */
1762 static int virtio_mem_bbm_plug_and_add_bb(struct virtio_mem *vm,
1763 					  unsigned long bb_id)
1764 {
1765 	int rc;
1766 
1767 	if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
1768 			 VIRTIO_MEM_BBM_BB_UNUSED))
1769 		return -EINVAL;
1770 
1771 	rc = virtio_mem_bbm_plug_bb(vm, bb_id);
1772 	if (rc)
1773 		return rc;
1774 	virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED);
1775 
1776 	rc = virtio_mem_bbm_add_bb(vm, bb_id);
1777 	if (rc) {
1778 		if (!virtio_mem_bbm_unplug_bb(vm, bb_id))
1779 			virtio_mem_bbm_set_bb_state(vm, bb_id,
1780 						    VIRTIO_MEM_BBM_BB_UNUSED);
1781 		else
1782 			/* Retry from the main loop. */
1783 			virtio_mem_bbm_set_bb_state(vm, bb_id,
1784 						    VIRTIO_MEM_BBM_BB_PLUGGED);
1785 		return rc;
1786 	}
1787 	return 0;
1788 }
1789 
1790 /*
1791  * Prepare tracking data for the next big block.
1792  */
1793 static int virtio_mem_bbm_prepare_next_bb(struct virtio_mem *vm,
1794 					  unsigned long *bb_id)
1795 {
1796 	int rc;
1797 
1798 	if (vm->bbm.next_bb_id > vm->bbm.last_usable_bb_id)
1799 		return -ENOSPC;
1800 
1801 	/* Resize the big block state array if required. */
1802 	rc = virtio_mem_bbm_bb_states_prepare_next_bb(vm);
1803 	if (rc)
1804 		return rc;
1805 
1806 	vm->bbm.bb_count[VIRTIO_MEM_BBM_BB_UNUSED]++;
1807 	*bb_id = vm->bbm.next_bb_id;
1808 	vm->bbm.next_bb_id++;
1809 	return 0;
1810 }
1811 
1812 static int virtio_mem_bbm_plug_request(struct virtio_mem *vm, uint64_t diff)
1813 {
1814 	uint64_t nb_bb = diff / vm->bbm.bb_size;
1815 	unsigned long bb_id;
1816 	int rc;
1817 
1818 	if (!nb_bb)
1819 		return 0;
1820 
1821 	/* Try to plug and add unused big blocks */
1822 	virtio_mem_bbm_for_each_bb(vm, bb_id, VIRTIO_MEM_BBM_BB_UNUSED) {
1823 		if (!virtio_mem_could_add_memory(vm, vm->bbm.bb_size))
1824 			return -ENOSPC;
1825 
1826 		rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
1827 		if (!rc)
1828 			nb_bb--;
1829 		if (rc || !nb_bb)
1830 			return rc;
1831 		cond_resched();
1832 	}
1833 
1834 	/* Try to prepare, plug and add new big blocks */
1835 	while (nb_bb) {
1836 		if (!virtio_mem_could_add_memory(vm, vm->bbm.bb_size))
1837 			return -ENOSPC;
1838 
1839 		rc = virtio_mem_bbm_prepare_next_bb(vm, &bb_id);
1840 		if (rc)
1841 			return rc;
1842 		rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
1843 		if (!rc)
1844 			nb_bb--;
1845 		if (rc)
1846 			return rc;
1847 		cond_resched();
1848 	}
1849 
1850 	return 0;
1851 }
1852 
1853 /*
1854  * Try to plug the requested amount of memory.
1855  */
1856 static int virtio_mem_plug_request(struct virtio_mem *vm, uint64_t diff)
1857 {
1858 	if (vm->in_sbm)
1859 		return virtio_mem_sbm_plug_request(vm, diff);
1860 	return virtio_mem_bbm_plug_request(vm, diff);
1861 }
1862 
1863 /*
1864  * Unplug the desired number of plugged subblocks of an offline memory block.
1865  * Will fail if any subblock cannot get unplugged (instead of skipping it).
1866  *
1867  * Will modify the state of the memory block. Might temporarily drop the
1868  * hotplug_mutex.
1869  *
1870  * Note: Can fail after some subblocks were successfully unplugged.
1871  */
1872 static int virtio_mem_sbm_unplug_any_sb_offline(struct virtio_mem *vm,
1873 						unsigned long mb_id,
1874 						uint64_t *nb_sb)
1875 {
1876 	int rc;
1877 
1878 	rc = virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, nb_sb);
1879 
1880 	/* some subblocks might have been unplugged even on failure */
1881 	if (!virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb))
1882 		virtio_mem_sbm_set_mb_state(vm, mb_id,
1883 					    VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
1884 	if (rc)
1885 		return rc;
1886 
1887 	if (virtio_mem_sbm_test_sb_unplugged(vm, mb_id, 0, vm->sbm.sbs_per_mb)) {
1888 		/*
1889 		 * Remove the block from Linux - this should never fail.
1890 		 * Hinder the block from getting onlined by marking it
1891 		 * unplugged. Temporarily drop the mutex, so
1892 		 * any pending GOING_ONLINE requests can be serviced/rejected.
1893 		 */
1894 		virtio_mem_sbm_set_mb_state(vm, mb_id,
1895 					    VIRTIO_MEM_SBM_MB_UNUSED);
1896 
1897 		mutex_unlock(&vm->hotplug_mutex);
1898 		rc = virtio_mem_sbm_remove_mb(vm, mb_id);
1899 		BUG_ON(rc);
1900 		mutex_lock(&vm->hotplug_mutex);
1901 	}
1902 	return 0;
1903 }
1904 
1905 /*
1906  * Unplug the given plugged subblocks of an online memory block.
1907  *
1908  * Will modify the state of the memory block.
1909  */
1910 static int virtio_mem_sbm_unplug_sb_online(struct virtio_mem *vm,
1911 					   unsigned long mb_id, int sb_id,
1912 					   int count)
1913 {
1914 	const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size) * count;
1915 	const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
1916 	unsigned long start_pfn;
1917 	int rc;
1918 
1919 	start_pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
1920 			     sb_id * vm->sbm.sb_size);
1921 
1922 	rc = virtio_mem_fake_offline(start_pfn, nr_pages);
1923 	if (rc)
1924 		return rc;
1925 
1926 	/* Try to unplug the allocated memory */
1927 	rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
1928 	if (rc) {
1929 		/* Return the memory to the buddy. */
1930 		virtio_mem_fake_online(start_pfn, nr_pages);
1931 		return rc;
1932 	}
1933 
1934 	switch (old_state) {
1935 	case VIRTIO_MEM_SBM_MB_KERNEL:
1936 		virtio_mem_sbm_set_mb_state(vm, mb_id,
1937 					    VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL);
1938 		break;
1939 	case VIRTIO_MEM_SBM_MB_MOVABLE:
1940 		virtio_mem_sbm_set_mb_state(vm, mb_id,
1941 					    VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL);
1942 		break;
1943 	}
1944 
1945 	return 0;
1946 }
1947 
1948 /*
1949  * Unplug the desired number of plugged subblocks of an online memory block.
1950  * Will skip subblock that are busy.
1951  *
1952  * Will modify the state of the memory block. Might temporarily drop the
1953  * hotplug_mutex.
1954  *
1955  * Note: Can fail after some subblocks were successfully unplugged. Can
1956  *       return 0 even if subblocks were busy and could not get unplugged.
1957  */
1958 static int virtio_mem_sbm_unplug_any_sb_online(struct virtio_mem *vm,
1959 					       unsigned long mb_id,
1960 					       uint64_t *nb_sb)
1961 {
1962 	int rc, sb_id;
1963 
1964 	/* If possible, try to unplug the complete block in one shot. */
1965 	if (*nb_sb >= vm->sbm.sbs_per_mb &&
1966 	    virtio_mem_sbm_test_sb_plugged(vm, mb_id, 0, vm->sbm.sbs_per_mb)) {
1967 		rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, 0,
1968 						     vm->sbm.sbs_per_mb);
1969 		if (!rc) {
1970 			*nb_sb -= vm->sbm.sbs_per_mb;
1971 			goto unplugged;
1972 		} else if (rc != -EBUSY)
1973 			return rc;
1974 	}
1975 
1976 	/* Fallback to single subblocks. */
1977 	for (sb_id = vm->sbm.sbs_per_mb - 1; sb_id >= 0 && *nb_sb; sb_id--) {
1978 		/* Find the next candidate subblock */
1979 		while (sb_id >= 0 &&
1980 		       !virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, 1))
1981 			sb_id--;
1982 		if (sb_id < 0)
1983 			break;
1984 
1985 		rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, sb_id, 1);
1986 		if (rc == -EBUSY)
1987 			continue;
1988 		else if (rc)
1989 			return rc;
1990 		*nb_sb -= 1;
1991 	}
1992 
1993 unplugged:
1994 	/*
1995 	 * Once all subblocks of a memory block were unplugged, offline and
1996 	 * remove it. This will usually not fail, as no memory is in use
1997 	 * anymore - however some other notifiers might NACK the request.
1998 	 */
1999 	if (virtio_mem_sbm_test_sb_unplugged(vm, mb_id, 0, vm->sbm.sbs_per_mb)) {
2000 		mutex_unlock(&vm->hotplug_mutex);
2001 		rc = virtio_mem_sbm_offline_and_remove_mb(vm, mb_id);
2002 		mutex_lock(&vm->hotplug_mutex);
2003 		if (!rc)
2004 			virtio_mem_sbm_set_mb_state(vm, mb_id,
2005 						    VIRTIO_MEM_SBM_MB_UNUSED);
2006 	}
2007 
2008 	return 0;
2009 }
2010 
2011 /*
2012  * Unplug the desired number of plugged subblocks of a memory block that is
2013  * already added to Linux. Will skip subblock of online memory blocks that are
2014  * busy (by the OS). Will fail if any subblock that's not busy cannot get
2015  * unplugged.
2016  *
2017  * Will modify the state of the memory block. Might temporarily drop the
2018  * hotplug_mutex.
2019  *
2020  * Note: Can fail after some subblocks were successfully unplugged. Can
2021  *       return 0 even if subblocks were busy and could not get unplugged.
2022  */
2023 static int virtio_mem_sbm_unplug_any_sb(struct virtio_mem *vm,
2024 					unsigned long mb_id,
2025 					uint64_t *nb_sb)
2026 {
2027 	const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
2028 
2029 	switch (old_state) {
2030 	case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
2031 	case VIRTIO_MEM_SBM_MB_KERNEL:
2032 	case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
2033 	case VIRTIO_MEM_SBM_MB_MOVABLE:
2034 		return virtio_mem_sbm_unplug_any_sb_online(vm, mb_id, nb_sb);
2035 	case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
2036 	case VIRTIO_MEM_SBM_MB_OFFLINE:
2037 		return virtio_mem_sbm_unplug_any_sb_offline(vm, mb_id, nb_sb);
2038 	}
2039 	return -EINVAL;
2040 }
2041 
2042 static int virtio_mem_sbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
2043 {
2044 	const int mb_states[] = {
2045 		VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
2046 		VIRTIO_MEM_SBM_MB_OFFLINE,
2047 		VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
2048 		VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
2049 		VIRTIO_MEM_SBM_MB_MOVABLE,
2050 		VIRTIO_MEM_SBM_MB_KERNEL,
2051 	};
2052 	uint64_t nb_sb = diff / vm->sbm.sb_size;
2053 	unsigned long mb_id;
2054 	int rc, i;
2055 
2056 	if (!nb_sb)
2057 		return 0;
2058 
2059 	/*
2060 	 * We'll drop the mutex a couple of times when it is safe to do so.
2061 	 * This might result in some blocks switching the state (online/offline)
2062 	 * and we could miss them in this run - we will retry again later.
2063 	 */
2064 	mutex_lock(&vm->hotplug_mutex);
2065 
2066 	/*
2067 	 * We try unplug from partially plugged blocks first, to try removing
2068 	 * whole memory blocks along with metadata. We prioritize ZONE_MOVABLE
2069 	 * as it's more reliable to unplug memory and remove whole memory
2070 	 * blocks, and we don't want to trigger a zone imbalances by
2071 	 * accidentially removing too much kernel memory.
2072 	 */
2073 	for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
2074 		virtio_mem_sbm_for_each_mb_rev(vm, mb_id, mb_states[i]) {
2075 			rc = virtio_mem_sbm_unplug_any_sb(vm, mb_id, &nb_sb);
2076 			if (rc || !nb_sb)
2077 				goto out_unlock;
2078 			mutex_unlock(&vm->hotplug_mutex);
2079 			cond_resched();
2080 			mutex_lock(&vm->hotplug_mutex);
2081 		}
2082 		if (!unplug_online && i == 1) {
2083 			mutex_unlock(&vm->hotplug_mutex);
2084 			return 0;
2085 		}
2086 	}
2087 
2088 	mutex_unlock(&vm->hotplug_mutex);
2089 	return nb_sb ? -EBUSY : 0;
2090 out_unlock:
2091 	mutex_unlock(&vm->hotplug_mutex);
2092 	return rc;
2093 }
2094 
2095 /*
2096  * Try to offline and remove a big block from Linux and unplug it. Will fail
2097  * with -EBUSY if some memory is busy and cannot get unplugged.
2098  *
2099  * Will modify the state of the memory block. Might temporarily drop the
2100  * hotplug_mutex.
2101  */
2102 static int virtio_mem_bbm_offline_remove_and_unplug_bb(struct virtio_mem *vm,
2103 						       unsigned long bb_id)
2104 {
2105 	const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2106 	const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2107 	unsigned long end_pfn = start_pfn + nr_pages;
2108 	unsigned long pfn;
2109 	struct page *page;
2110 	int rc;
2111 
2112 	if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
2113 			 VIRTIO_MEM_BBM_BB_ADDED))
2114 		return -EINVAL;
2115 
2116 	if (bbm_safe_unplug) {
2117 		/*
2118 		 * Start by fake-offlining all memory. Once we marked the device
2119 		 * block as fake-offline, all newly onlined memory will
2120 		 * automatically be kept fake-offline. Protect from concurrent
2121 		 * onlining/offlining until we have a consistent state.
2122 		 */
2123 		mutex_lock(&vm->hotplug_mutex);
2124 		virtio_mem_bbm_set_bb_state(vm, bb_id,
2125 					    VIRTIO_MEM_BBM_BB_FAKE_OFFLINE);
2126 
2127 		for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2128 			page = pfn_to_online_page(pfn);
2129 			if (!page)
2130 				continue;
2131 
2132 			rc = virtio_mem_fake_offline(pfn, PAGES_PER_SECTION);
2133 			if (rc) {
2134 				end_pfn = pfn;
2135 				goto rollback_safe_unplug;
2136 			}
2137 		}
2138 		mutex_unlock(&vm->hotplug_mutex);
2139 	}
2140 
2141 	rc = virtio_mem_bbm_offline_and_remove_bb(vm, bb_id);
2142 	if (rc) {
2143 		if (bbm_safe_unplug) {
2144 			mutex_lock(&vm->hotplug_mutex);
2145 			goto rollback_safe_unplug;
2146 		}
2147 		return rc;
2148 	}
2149 
2150 	rc = virtio_mem_bbm_unplug_bb(vm, bb_id);
2151 	if (rc)
2152 		virtio_mem_bbm_set_bb_state(vm, bb_id,
2153 					    VIRTIO_MEM_BBM_BB_PLUGGED);
2154 	else
2155 		virtio_mem_bbm_set_bb_state(vm, bb_id,
2156 					    VIRTIO_MEM_BBM_BB_UNUSED);
2157 	return rc;
2158 
2159 rollback_safe_unplug:
2160 	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2161 		page = pfn_to_online_page(pfn);
2162 		if (!page)
2163 			continue;
2164 		virtio_mem_fake_online(pfn, PAGES_PER_SECTION);
2165 	}
2166 	virtio_mem_bbm_set_bb_state(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED);
2167 	mutex_unlock(&vm->hotplug_mutex);
2168 	return rc;
2169 }
2170 
2171 /*
2172  * Test if a big block is completely offline.
2173  */
2174 static bool virtio_mem_bbm_bb_is_offline(struct virtio_mem *vm,
2175 					 unsigned long bb_id)
2176 {
2177 	const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2178 	const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2179 	unsigned long pfn;
2180 
2181 	for (pfn = start_pfn; pfn < start_pfn + nr_pages;
2182 	     pfn += PAGES_PER_SECTION) {
2183 		if (pfn_to_online_page(pfn))
2184 			return false;
2185 	}
2186 
2187 	return true;
2188 }
2189 
2190 /*
2191  * Test if a big block is completely onlined to ZONE_MOVABLE (or offline).
2192  */
2193 static bool virtio_mem_bbm_bb_is_movable(struct virtio_mem *vm,
2194 					 unsigned long bb_id)
2195 {
2196 	const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2197 	const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2198 	struct page *page;
2199 	unsigned long pfn;
2200 
2201 	for (pfn = start_pfn; pfn < start_pfn + nr_pages;
2202 	     pfn += PAGES_PER_SECTION) {
2203 		page = pfn_to_online_page(pfn);
2204 		if (!page)
2205 			continue;
2206 		if (page_zonenum(page) != ZONE_MOVABLE)
2207 			return false;
2208 	}
2209 
2210 	return true;
2211 }
2212 
2213 static int virtio_mem_bbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
2214 {
2215 	uint64_t nb_bb = diff / vm->bbm.bb_size;
2216 	uint64_t bb_id;
2217 	int rc, i;
2218 
2219 	if (!nb_bb)
2220 		return 0;
2221 
2222 	/*
2223 	 * Try to unplug big blocks. Similar to SBM, start with offline
2224 	 * big blocks.
2225 	 */
2226 	for (i = 0; i < 3; i++) {
2227 		virtio_mem_bbm_for_each_bb_rev(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED) {
2228 			cond_resched();
2229 
2230 			/*
2231 			 * As we're holding no locks, these checks are racy,
2232 			 * but we don't care.
2233 			 */
2234 			if (i == 0 && !virtio_mem_bbm_bb_is_offline(vm, bb_id))
2235 				continue;
2236 			if (i == 1 && !virtio_mem_bbm_bb_is_movable(vm, bb_id))
2237 				continue;
2238 			rc = virtio_mem_bbm_offline_remove_and_unplug_bb(vm, bb_id);
2239 			if (rc == -EBUSY)
2240 				continue;
2241 			if (!rc)
2242 				nb_bb--;
2243 			if (rc || !nb_bb)
2244 				return rc;
2245 		}
2246 		if (i == 0 && !unplug_online)
2247 			return 0;
2248 	}
2249 
2250 	return nb_bb ? -EBUSY : 0;
2251 }
2252 
2253 /*
2254  * Try to unplug the requested amount of memory.
2255  */
2256 static int virtio_mem_unplug_request(struct virtio_mem *vm, uint64_t diff)
2257 {
2258 	if (vm->in_sbm)
2259 		return virtio_mem_sbm_unplug_request(vm, diff);
2260 	return virtio_mem_bbm_unplug_request(vm, diff);
2261 }
2262 
2263 /*
2264  * Try to unplug all blocks that couldn't be unplugged before, for example,
2265  * because the hypervisor was busy.
2266  */
2267 static int virtio_mem_unplug_pending_mb(struct virtio_mem *vm)
2268 {
2269 	unsigned long id;
2270 	int rc;
2271 
2272 	if (!vm->in_sbm) {
2273 		virtio_mem_bbm_for_each_bb(vm, id,
2274 					   VIRTIO_MEM_BBM_BB_PLUGGED) {
2275 			rc = virtio_mem_bbm_unplug_bb(vm, id);
2276 			if (rc)
2277 				return rc;
2278 			virtio_mem_bbm_set_bb_state(vm, id,
2279 						    VIRTIO_MEM_BBM_BB_UNUSED);
2280 		}
2281 		return 0;
2282 	}
2283 
2284 	virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_PLUGGED) {
2285 		rc = virtio_mem_sbm_unplug_mb(vm, id);
2286 		if (rc)
2287 			return rc;
2288 		virtio_mem_sbm_set_mb_state(vm, id,
2289 					    VIRTIO_MEM_SBM_MB_UNUSED);
2290 	}
2291 
2292 	return 0;
2293 }
2294 
2295 /*
2296  * Update all parts of the config that could have changed.
2297  */
2298 static void virtio_mem_refresh_config(struct virtio_mem *vm)
2299 {
2300 	const struct range pluggable_range = mhp_get_pluggable_range(true);
2301 	uint64_t new_plugged_size, usable_region_size, end_addr;
2302 
2303 	/* the plugged_size is just a reflection of what _we_ did previously */
2304 	virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
2305 			&new_plugged_size);
2306 	if (WARN_ON_ONCE(new_plugged_size != vm->plugged_size))
2307 		vm->plugged_size = new_plugged_size;
2308 
2309 	/* calculate the last usable memory block id */
2310 	virtio_cread_le(vm->vdev, struct virtio_mem_config,
2311 			usable_region_size, &usable_region_size);
2312 	end_addr = min(vm->addr + usable_region_size - 1,
2313 		       pluggable_range.end);
2314 
2315 	if (vm->in_sbm) {
2316 		vm->sbm.last_usable_mb_id = virtio_mem_phys_to_mb_id(end_addr);
2317 		if (!IS_ALIGNED(end_addr + 1, memory_block_size_bytes()))
2318 			vm->sbm.last_usable_mb_id--;
2319 	} else {
2320 		vm->bbm.last_usable_bb_id = virtio_mem_phys_to_bb_id(vm,
2321 								     end_addr);
2322 		if (!IS_ALIGNED(end_addr + 1, vm->bbm.bb_size))
2323 			vm->bbm.last_usable_bb_id--;
2324 	}
2325 	/*
2326 	 * If we cannot plug any of our device memory (e.g., nothing in the
2327 	 * usable region is addressable), the last usable memory block id will
2328 	 * be smaller than the first usable memory block id. We'll stop
2329 	 * attempting to add memory with -ENOSPC from our main loop.
2330 	 */
2331 
2332 	/* see if there is a request to change the size */
2333 	virtio_cread_le(vm->vdev, struct virtio_mem_config, requested_size,
2334 			&vm->requested_size);
2335 
2336 	dev_info(&vm->vdev->dev, "plugged size: 0x%llx", vm->plugged_size);
2337 	dev_info(&vm->vdev->dev, "requested size: 0x%llx", vm->requested_size);
2338 }
2339 
2340 /*
2341  * Workqueue function for handling plug/unplug requests and config updates.
2342  */
2343 static void virtio_mem_run_wq(struct work_struct *work)
2344 {
2345 	struct virtio_mem *vm = container_of(work, struct virtio_mem, wq);
2346 	uint64_t diff;
2347 	int rc;
2348 
2349 	if (unlikely(vm->in_kdump)) {
2350 		dev_warn_once(&vm->vdev->dev,
2351 			     "unexpected workqueue run in kdump kernel\n");
2352 		return;
2353 	}
2354 
2355 	hrtimer_cancel(&vm->retry_timer);
2356 
2357 	if (vm->broken)
2358 		return;
2359 
2360 	atomic_set(&vm->wq_active, 1);
2361 retry:
2362 	rc = 0;
2363 
2364 	/* Make sure we start with a clean state if there are leftovers. */
2365 	if (unlikely(vm->unplug_all_required))
2366 		rc = virtio_mem_send_unplug_all_request(vm);
2367 
2368 	if (atomic_read(&vm->config_changed)) {
2369 		atomic_set(&vm->config_changed, 0);
2370 		virtio_mem_refresh_config(vm);
2371 	}
2372 
2373 	/* Unplug any leftovers from previous runs */
2374 	if (!rc)
2375 		rc = virtio_mem_unplug_pending_mb(vm);
2376 
2377 	if (!rc && vm->requested_size != vm->plugged_size) {
2378 		if (vm->requested_size > vm->plugged_size) {
2379 			diff = vm->requested_size - vm->plugged_size;
2380 			rc = virtio_mem_plug_request(vm, diff);
2381 		} else {
2382 			diff = vm->plugged_size - vm->requested_size;
2383 			rc = virtio_mem_unplug_request(vm, diff);
2384 		}
2385 	}
2386 
2387 	switch (rc) {
2388 	case 0:
2389 		vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
2390 		break;
2391 	case -ENOSPC:
2392 		/*
2393 		 * We cannot add any more memory (alignment, physical limit)
2394 		 * or we have too many offline memory blocks.
2395 		 */
2396 		break;
2397 	case -ETXTBSY:
2398 		/*
2399 		 * The hypervisor cannot process our request right now
2400 		 * (e.g., out of memory, migrating);
2401 		 */
2402 	case -EBUSY:
2403 		/*
2404 		 * We cannot free up any memory to unplug it (all plugged memory
2405 		 * is busy).
2406 		 */
2407 	case -ENOMEM:
2408 		/* Out of memory, try again later. */
2409 		hrtimer_start(&vm->retry_timer, ms_to_ktime(vm->retry_timer_ms),
2410 			      HRTIMER_MODE_REL);
2411 		break;
2412 	case -EAGAIN:
2413 		/* Retry immediately (e.g., the config changed). */
2414 		goto retry;
2415 	default:
2416 		/* Unknown error, mark as broken */
2417 		dev_err(&vm->vdev->dev,
2418 			"unknown error, marking device broken: %d\n", rc);
2419 		vm->broken = true;
2420 	}
2421 
2422 	atomic_set(&vm->wq_active, 0);
2423 }
2424 
2425 static enum hrtimer_restart virtio_mem_timer_expired(struct hrtimer *timer)
2426 {
2427 	struct virtio_mem *vm = container_of(timer, struct virtio_mem,
2428 					     retry_timer);
2429 
2430 	virtio_mem_retry(vm);
2431 	vm->retry_timer_ms = min_t(unsigned int, vm->retry_timer_ms * 2,
2432 				   VIRTIO_MEM_RETRY_TIMER_MAX_MS);
2433 	return HRTIMER_NORESTART;
2434 }
2435 
2436 static void virtio_mem_handle_response(struct virtqueue *vq)
2437 {
2438 	struct virtio_mem *vm = vq->vdev->priv;
2439 
2440 	wake_up(&vm->host_resp);
2441 }
2442 
2443 static int virtio_mem_init_vq(struct virtio_mem *vm)
2444 {
2445 	struct virtqueue *vq;
2446 
2447 	vq = virtio_find_single_vq(vm->vdev, virtio_mem_handle_response,
2448 				   "guest-request");
2449 	if (IS_ERR(vq))
2450 		return PTR_ERR(vq);
2451 	vm->vq = vq;
2452 
2453 	return 0;
2454 }
2455 
2456 static int virtio_mem_init_hotplug(struct virtio_mem *vm)
2457 {
2458 	const struct range pluggable_range = mhp_get_pluggable_range(true);
2459 	uint64_t unit_pages, sb_size, addr;
2460 	int rc;
2461 
2462 	/* bad device setup - warn only */
2463 	if (!IS_ALIGNED(vm->addr, memory_block_size_bytes()))
2464 		dev_warn(&vm->vdev->dev,
2465 			 "The alignment of the physical start address can make some memory unusable.\n");
2466 	if (!IS_ALIGNED(vm->addr + vm->region_size, memory_block_size_bytes()))
2467 		dev_warn(&vm->vdev->dev,
2468 			 "The alignment of the physical end address can make some memory unusable.\n");
2469 	if (vm->addr < pluggable_range.start ||
2470 	    vm->addr + vm->region_size - 1 > pluggable_range.end)
2471 		dev_warn(&vm->vdev->dev,
2472 			 "Some device memory is not addressable/pluggable. This can make some memory unusable.\n");
2473 
2474 	/* Prepare the offline threshold - make sure we can add two blocks. */
2475 	vm->offline_threshold = max_t(uint64_t, 2 * memory_block_size_bytes(),
2476 				      VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD);
2477 
2478 	/*
2479 	 * TODO: once alloc_contig_range() works reliably with pageblock
2480 	 * granularity on ZONE_NORMAL, use pageblock_nr_pages instead.
2481 	 */
2482 	sb_size = PAGE_SIZE * MAX_ORDER_NR_PAGES;
2483 	sb_size = max_t(uint64_t, vm->device_block_size, sb_size);
2484 
2485 	if (sb_size < memory_block_size_bytes() && !force_bbm) {
2486 		/* SBM: At least two subblocks per Linux memory block. */
2487 		vm->in_sbm = true;
2488 		vm->sbm.sb_size = sb_size;
2489 		vm->sbm.sbs_per_mb = memory_block_size_bytes() /
2490 				     vm->sbm.sb_size;
2491 
2492 		/* Round up to the next full memory block */
2493 		addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
2494 		       memory_block_size_bytes() - 1;
2495 		vm->sbm.first_mb_id = virtio_mem_phys_to_mb_id(addr);
2496 		vm->sbm.next_mb_id = vm->sbm.first_mb_id;
2497 	} else {
2498 		/* BBM: At least one Linux memory block. */
2499 		vm->bbm.bb_size = max_t(uint64_t, vm->device_block_size,
2500 					memory_block_size_bytes());
2501 
2502 		if (bbm_block_size) {
2503 			if (!is_power_of_2(bbm_block_size)) {
2504 				dev_warn(&vm->vdev->dev,
2505 					 "bbm_block_size is not a power of 2");
2506 			} else if (bbm_block_size < vm->bbm.bb_size) {
2507 				dev_warn(&vm->vdev->dev,
2508 					 "bbm_block_size is too small");
2509 			} else {
2510 				vm->bbm.bb_size = bbm_block_size;
2511 			}
2512 		}
2513 
2514 		/* Round up to the next aligned big block */
2515 		addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
2516 		       vm->bbm.bb_size - 1;
2517 		vm->bbm.first_bb_id = virtio_mem_phys_to_bb_id(vm, addr);
2518 		vm->bbm.next_bb_id = vm->bbm.first_bb_id;
2519 
2520 		/* Make sure we can add two big blocks. */
2521 		vm->offline_threshold = max_t(uint64_t, 2 * vm->bbm.bb_size,
2522 					      vm->offline_threshold);
2523 	}
2524 
2525 	dev_info(&vm->vdev->dev, "memory block size: 0x%lx",
2526 		 memory_block_size_bytes());
2527 	if (vm->in_sbm)
2528 		dev_info(&vm->vdev->dev, "subblock size: 0x%llx",
2529 			 (unsigned long long)vm->sbm.sb_size);
2530 	else
2531 		dev_info(&vm->vdev->dev, "big block size: 0x%llx",
2532 			 (unsigned long long)vm->bbm.bb_size);
2533 
2534 	/* create the parent resource for all memory */
2535 	rc = virtio_mem_create_resource(vm);
2536 	if (rc)
2537 		return rc;
2538 
2539 	/* use a single dynamic memory group to cover the whole memory device */
2540 	if (vm->in_sbm)
2541 		unit_pages = PHYS_PFN(memory_block_size_bytes());
2542 	else
2543 		unit_pages = PHYS_PFN(vm->bbm.bb_size);
2544 	rc = memory_group_register_dynamic(vm->nid, unit_pages);
2545 	if (rc < 0)
2546 		goto out_del_resource;
2547 	vm->mgid = rc;
2548 
2549 	/*
2550 	 * If we still have memory plugged, we have to unplug all memory first.
2551 	 * Registering our parent resource makes sure that this memory isn't
2552 	 * actually in use (e.g., trying to reload the driver).
2553 	 */
2554 	if (vm->plugged_size) {
2555 		vm->unplug_all_required = true;
2556 		dev_info(&vm->vdev->dev, "unplugging all memory is required\n");
2557 	}
2558 
2559 	/* register callbacks */
2560 	vm->memory_notifier.notifier_call = virtio_mem_memory_notifier_cb;
2561 	rc = register_memory_notifier(&vm->memory_notifier);
2562 	if (rc)
2563 		goto out_unreg_group;
2564 	rc = register_virtio_mem_device(vm);
2565 	if (rc)
2566 		goto out_unreg_mem;
2567 
2568 	return 0;
2569 out_unreg_mem:
2570 	unregister_memory_notifier(&vm->memory_notifier);
2571 out_unreg_group:
2572 	memory_group_unregister(vm->mgid);
2573 out_del_resource:
2574 	virtio_mem_delete_resource(vm);
2575 	return rc;
2576 }
2577 
2578 #ifdef CONFIG_PROC_VMCORE
2579 static int virtio_mem_send_state_request(struct virtio_mem *vm, uint64_t addr,
2580 					 uint64_t size)
2581 {
2582 	const uint64_t nb_vm_blocks = size / vm->device_block_size;
2583 	const struct virtio_mem_req req = {
2584 		.type = cpu_to_virtio16(vm->vdev, VIRTIO_MEM_REQ_STATE),
2585 		.u.state.addr = cpu_to_virtio64(vm->vdev, addr),
2586 		.u.state.nb_blocks = cpu_to_virtio16(vm->vdev, nb_vm_blocks),
2587 	};
2588 	int rc = -ENOMEM;
2589 
2590 	dev_dbg(&vm->vdev->dev, "requesting state: 0x%llx - 0x%llx\n", addr,
2591 		addr + size - 1);
2592 
2593 	switch (virtio_mem_send_request(vm, &req)) {
2594 	case VIRTIO_MEM_RESP_ACK:
2595 		return virtio16_to_cpu(vm->vdev, vm->resp.u.state.state);
2596 	case VIRTIO_MEM_RESP_ERROR:
2597 		rc = -EINVAL;
2598 		break;
2599 	default:
2600 		break;
2601 	}
2602 
2603 	dev_dbg(&vm->vdev->dev, "requesting state failed: %d\n", rc);
2604 	return rc;
2605 }
2606 
2607 static bool virtio_mem_vmcore_pfn_is_ram(struct vmcore_cb *cb,
2608 					 unsigned long pfn)
2609 {
2610 	struct virtio_mem *vm = container_of(cb, struct virtio_mem,
2611 					     vmcore_cb);
2612 	uint64_t addr = PFN_PHYS(pfn);
2613 	bool is_ram;
2614 	int rc;
2615 
2616 	if (!virtio_mem_contains_range(vm, addr, PAGE_SIZE))
2617 		return true;
2618 	if (!vm->plugged_size)
2619 		return false;
2620 
2621 	/*
2622 	 * We have to serialize device requests and access to the information
2623 	 * about the block queried last.
2624 	 */
2625 	mutex_lock(&vm->hotplug_mutex);
2626 
2627 	addr = ALIGN_DOWN(addr, vm->device_block_size);
2628 	if (addr != vm->last_block_addr) {
2629 		rc = virtio_mem_send_state_request(vm, addr,
2630 						   vm->device_block_size);
2631 		/* On any kind of error, we're going to signal !ram. */
2632 		if (rc == VIRTIO_MEM_STATE_PLUGGED)
2633 			vm->last_block_plugged = true;
2634 		else
2635 			vm->last_block_plugged = false;
2636 		vm->last_block_addr = addr;
2637 	}
2638 
2639 	is_ram = vm->last_block_plugged;
2640 	mutex_unlock(&vm->hotplug_mutex);
2641 	return is_ram;
2642 }
2643 #endif /* CONFIG_PROC_VMCORE */
2644 
2645 static int virtio_mem_init_kdump(struct virtio_mem *vm)
2646 {
2647 #ifdef CONFIG_PROC_VMCORE
2648 	dev_info(&vm->vdev->dev, "memory hot(un)plug disabled in kdump kernel\n");
2649 	vm->vmcore_cb.pfn_is_ram = virtio_mem_vmcore_pfn_is_ram;
2650 	register_vmcore_cb(&vm->vmcore_cb);
2651 	return 0;
2652 #else /* CONFIG_PROC_VMCORE */
2653 	dev_warn(&vm->vdev->dev, "disabled in kdump kernel without vmcore\n");
2654 	return -EBUSY;
2655 #endif /* CONFIG_PROC_VMCORE */
2656 }
2657 
2658 static int virtio_mem_init(struct virtio_mem *vm)
2659 {
2660 	uint16_t node_id;
2661 
2662 	if (!vm->vdev->config->get) {
2663 		dev_err(&vm->vdev->dev, "config access disabled\n");
2664 		return -EINVAL;
2665 	}
2666 
2667 	/* Fetch all properties that can't change. */
2668 	virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
2669 			&vm->plugged_size);
2670 	virtio_cread_le(vm->vdev, struct virtio_mem_config, block_size,
2671 			&vm->device_block_size);
2672 	virtio_cread_le(vm->vdev, struct virtio_mem_config, node_id,
2673 			&node_id);
2674 	vm->nid = virtio_mem_translate_node_id(vm, node_id);
2675 	virtio_cread_le(vm->vdev, struct virtio_mem_config, addr, &vm->addr);
2676 	virtio_cread_le(vm->vdev, struct virtio_mem_config, region_size,
2677 			&vm->region_size);
2678 
2679 	/* Determine the nid for the device based on the lowest address. */
2680 	if (vm->nid == NUMA_NO_NODE)
2681 		vm->nid = memory_add_physaddr_to_nid(vm->addr);
2682 
2683 	dev_info(&vm->vdev->dev, "start address: 0x%llx", vm->addr);
2684 	dev_info(&vm->vdev->dev, "region size: 0x%llx", vm->region_size);
2685 	dev_info(&vm->vdev->dev, "device block size: 0x%llx",
2686 		 (unsigned long long)vm->device_block_size);
2687 	if (vm->nid != NUMA_NO_NODE && IS_ENABLED(CONFIG_NUMA))
2688 		dev_info(&vm->vdev->dev, "nid: %d", vm->nid);
2689 
2690 	/*
2691 	 * We don't want to (un)plug or reuse any memory when in kdump. The
2692 	 * memory is still accessible (but not exposed to Linux).
2693 	 */
2694 	if (vm->in_kdump)
2695 		return virtio_mem_init_kdump(vm);
2696 	return virtio_mem_init_hotplug(vm);
2697 }
2698 
2699 static int virtio_mem_create_resource(struct virtio_mem *vm)
2700 {
2701 	/*
2702 	 * When force-unloading the driver and removing the device, we
2703 	 * could have a garbage pointer. Duplicate the string.
2704 	 */
2705 	const char *name = kstrdup(dev_name(&vm->vdev->dev), GFP_KERNEL);
2706 
2707 	if (!name)
2708 		return -ENOMEM;
2709 
2710 	/* Disallow mapping device memory via /dev/mem completely. */
2711 	vm->parent_resource = __request_mem_region(vm->addr, vm->region_size,
2712 						   name, IORESOURCE_SYSTEM_RAM |
2713 						   IORESOURCE_EXCLUSIVE);
2714 	if (!vm->parent_resource) {
2715 		kfree(name);
2716 		dev_warn(&vm->vdev->dev, "could not reserve device region\n");
2717 		dev_info(&vm->vdev->dev,
2718 			 "reloading the driver is not supported\n");
2719 		return -EBUSY;
2720 	}
2721 
2722 	/* The memory is not actually busy - make add_memory() work. */
2723 	vm->parent_resource->flags &= ~IORESOURCE_BUSY;
2724 	return 0;
2725 }
2726 
2727 static void virtio_mem_delete_resource(struct virtio_mem *vm)
2728 {
2729 	const char *name;
2730 
2731 	if (!vm->parent_resource)
2732 		return;
2733 
2734 	name = vm->parent_resource->name;
2735 	release_resource(vm->parent_resource);
2736 	kfree(vm->parent_resource);
2737 	kfree(name);
2738 	vm->parent_resource = NULL;
2739 }
2740 
2741 static int virtio_mem_range_has_system_ram(struct resource *res, void *arg)
2742 {
2743 	return 1;
2744 }
2745 
2746 static bool virtio_mem_has_memory_added(struct virtio_mem *vm)
2747 {
2748 	const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
2749 
2750 	return walk_iomem_res_desc(IORES_DESC_NONE, flags, vm->addr,
2751 				   vm->addr + vm->region_size, NULL,
2752 				   virtio_mem_range_has_system_ram) == 1;
2753 }
2754 
2755 static int virtio_mem_probe(struct virtio_device *vdev)
2756 {
2757 	struct virtio_mem *vm;
2758 	int rc;
2759 
2760 	BUILD_BUG_ON(sizeof(struct virtio_mem_req) != 24);
2761 	BUILD_BUG_ON(sizeof(struct virtio_mem_resp) != 10);
2762 
2763 	vdev->priv = vm = kzalloc(sizeof(*vm), GFP_KERNEL);
2764 	if (!vm)
2765 		return -ENOMEM;
2766 
2767 	init_waitqueue_head(&vm->host_resp);
2768 	vm->vdev = vdev;
2769 	INIT_WORK(&vm->wq, virtio_mem_run_wq);
2770 	mutex_init(&vm->hotplug_mutex);
2771 	INIT_LIST_HEAD(&vm->next);
2772 	spin_lock_init(&vm->removal_lock);
2773 	hrtimer_init(&vm->retry_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2774 	vm->retry_timer.function = virtio_mem_timer_expired;
2775 	vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
2776 	vm->in_kdump = is_kdump_kernel();
2777 
2778 	/* register the virtqueue */
2779 	rc = virtio_mem_init_vq(vm);
2780 	if (rc)
2781 		goto out_free_vm;
2782 
2783 	/* initialize the device by querying the config */
2784 	rc = virtio_mem_init(vm);
2785 	if (rc)
2786 		goto out_del_vq;
2787 
2788 	virtio_device_ready(vdev);
2789 
2790 	/* trigger a config update to start processing the requested_size */
2791 	if (!vm->in_kdump) {
2792 		atomic_set(&vm->config_changed, 1);
2793 		queue_work(system_freezable_wq, &vm->wq);
2794 	}
2795 
2796 	return 0;
2797 out_del_vq:
2798 	vdev->config->del_vqs(vdev);
2799 out_free_vm:
2800 	kfree(vm);
2801 	vdev->priv = NULL;
2802 
2803 	return rc;
2804 }
2805 
2806 static void virtio_mem_deinit_hotplug(struct virtio_mem *vm)
2807 {
2808 	unsigned long mb_id;
2809 	int rc;
2810 
2811 	/*
2812 	 * Make sure the workqueue won't be triggered anymore and no memory
2813 	 * blocks can be onlined/offlined until we're finished here.
2814 	 */
2815 	mutex_lock(&vm->hotplug_mutex);
2816 	spin_lock_irq(&vm->removal_lock);
2817 	vm->removing = true;
2818 	spin_unlock_irq(&vm->removal_lock);
2819 	mutex_unlock(&vm->hotplug_mutex);
2820 
2821 	/* wait until the workqueue stopped */
2822 	cancel_work_sync(&vm->wq);
2823 	hrtimer_cancel(&vm->retry_timer);
2824 
2825 	if (vm->in_sbm) {
2826 		/*
2827 		 * After we unregistered our callbacks, user space can online
2828 		 * partially plugged offline blocks. Make sure to remove them.
2829 		 */
2830 		virtio_mem_sbm_for_each_mb(vm, mb_id,
2831 					   VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL) {
2832 			rc = virtio_mem_sbm_remove_mb(vm, mb_id);
2833 			BUG_ON(rc);
2834 			virtio_mem_sbm_set_mb_state(vm, mb_id,
2835 						    VIRTIO_MEM_SBM_MB_UNUSED);
2836 		}
2837 		/*
2838 		 * After we unregistered our callbacks, user space can no longer
2839 		 * offline partially plugged online memory blocks. No need to
2840 		 * worry about them.
2841 		 */
2842 	}
2843 
2844 	/* unregister callbacks */
2845 	unregister_virtio_mem_device(vm);
2846 	unregister_memory_notifier(&vm->memory_notifier);
2847 
2848 	/*
2849 	 * There is no way we could reliably remove all memory we have added to
2850 	 * the system. And there is no way to stop the driver/device from going
2851 	 * away. Warn at least.
2852 	 */
2853 	if (virtio_mem_has_memory_added(vm)) {
2854 		dev_warn(&vm->vdev->dev,
2855 			 "device still has system memory added\n");
2856 	} else {
2857 		virtio_mem_delete_resource(vm);
2858 		kfree_const(vm->resource_name);
2859 		memory_group_unregister(vm->mgid);
2860 	}
2861 
2862 	/* remove all tracking data - no locking needed */
2863 	if (vm->in_sbm) {
2864 		vfree(vm->sbm.mb_states);
2865 		vfree(vm->sbm.sb_states);
2866 	} else {
2867 		vfree(vm->bbm.bb_states);
2868 	}
2869 }
2870 
2871 static void virtio_mem_deinit_kdump(struct virtio_mem *vm)
2872 {
2873 #ifdef CONFIG_PROC_VMCORE
2874 	unregister_vmcore_cb(&vm->vmcore_cb);
2875 #endif /* CONFIG_PROC_VMCORE */
2876 }
2877 
2878 static void virtio_mem_remove(struct virtio_device *vdev)
2879 {
2880 	struct virtio_mem *vm = vdev->priv;
2881 
2882 	if (vm->in_kdump)
2883 		virtio_mem_deinit_kdump(vm);
2884 	else
2885 		virtio_mem_deinit_hotplug(vm);
2886 
2887 	/* reset the device and cleanup the queues */
2888 	virtio_reset_device(vdev);
2889 	vdev->config->del_vqs(vdev);
2890 
2891 	kfree(vm);
2892 	vdev->priv = NULL;
2893 }
2894 
2895 static void virtio_mem_config_changed(struct virtio_device *vdev)
2896 {
2897 	struct virtio_mem *vm = vdev->priv;
2898 
2899 	if (unlikely(vm->in_kdump))
2900 		return;
2901 
2902 	atomic_set(&vm->config_changed, 1);
2903 	virtio_mem_retry(vm);
2904 }
2905 
2906 #ifdef CONFIG_PM_SLEEP
2907 static int virtio_mem_freeze(struct virtio_device *vdev)
2908 {
2909 	/*
2910 	 * When restarting the VM, all memory is usually unplugged. Don't
2911 	 * allow to suspend/hibernate.
2912 	 */
2913 	dev_err(&vdev->dev, "save/restore not supported.\n");
2914 	return -EPERM;
2915 }
2916 
2917 static int virtio_mem_restore(struct virtio_device *vdev)
2918 {
2919 	return -EPERM;
2920 }
2921 #endif
2922 
2923 static unsigned int virtio_mem_features[] = {
2924 #if defined(CONFIG_NUMA) && defined(CONFIG_ACPI_NUMA)
2925 	VIRTIO_MEM_F_ACPI_PXM,
2926 #endif
2927 	VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE,
2928 };
2929 
2930 static const struct virtio_device_id virtio_mem_id_table[] = {
2931 	{ VIRTIO_ID_MEM, VIRTIO_DEV_ANY_ID },
2932 	{ 0 },
2933 };
2934 
2935 static struct virtio_driver virtio_mem_driver = {
2936 	.feature_table = virtio_mem_features,
2937 	.feature_table_size = ARRAY_SIZE(virtio_mem_features),
2938 	.driver.name = KBUILD_MODNAME,
2939 	.driver.owner = THIS_MODULE,
2940 	.id_table = virtio_mem_id_table,
2941 	.probe = virtio_mem_probe,
2942 	.remove = virtio_mem_remove,
2943 	.config_changed = virtio_mem_config_changed,
2944 #ifdef CONFIG_PM_SLEEP
2945 	.freeze	=	virtio_mem_freeze,
2946 	.restore =	virtio_mem_restore,
2947 #endif
2948 };
2949 
2950 module_virtio_driver(virtio_mem_driver);
2951 MODULE_DEVICE_TABLE(virtio, virtio_mem_id_table);
2952 MODULE_AUTHOR("David Hildenbrand <david@redhat.com>");
2953 MODULE_DESCRIPTION("Virtio-mem driver");
2954 MODULE_LICENSE("GPL");
2955