1	// SPDX-License-Identifier: GPL-2.0-or-later
2	/*
3	* Virtio-mem device driver.
4	*
5	* Copyright Red Hat, Inc. 2020
6	*
7	* Author(s): David Hildenbrand <david@redhat.com>
8	*/
9
10	#include <linux/virtio.h>
11	#include <linux/virtio_mem.h>
12	#include <linux/workqueue.h>
13	#include <linux/slab.h>
14	#include <linux/module.h>
15	#include <linux/mm.h>
16	#include <linux/memory_hotplug.h>
17	#include <linux/memory.h>
18	#include <linux/hrtimer.h>
19	#include <linux/crash_dump.h>
20	#include <linux/mutex.h>
21	#include <linux/bitmap.h>
22	#include <linux/lockdep.h>
23	#include <linux/log2.h>
24	#include <linux/vmalloc.h>
25	#include <linux/suspend.h>
26
27	#include <acpi/acpi_numa.h>
28
29	static bool unplug_online = true;
30	module_param(unplug_online, bool, 0644);
31	MODULE_PARM_DESC(unplug_online, "Try to unplug online memory");
32
33	static bool force_bbm;
34	module_param(force_bbm, bool, 0444);
35	MODULE_PARM_DESC(force_bbm,
36	"Force Big Block Mode. Default is 0 (auto-selection)");
37
38	static unsigned long bbm_block_size;
39	module_param(bbm_block_size, ulong, 0444);
40	MODULE_PARM_DESC(bbm_block_size,
41	"Big Block size in bytes. Default is 0 (auto-detection).");
42
43	/*
44	* virtio-mem currently supports the following modes of operation:
45	*
46	* * Sub Block Mode (SBM): A Linux memory block spans 2..X subblocks (SB). The
47	* size of a Sub Block (SB) is determined based on the device block size, the
48	* pageblock size, and the maximum allocation granularity of the buddy.
49	* Subblocks within a Linux memory block might either be plugged or unplugged.
50	* Memory is added/removed to Linux MM in Linux memory block granularity.
51	*
52	* * Big Block Mode (BBM): A Big Block (BB) spans 1..X Linux memory blocks.
53	* Memory is added/removed to Linux MM in Big Block granularity.
54	*
55	* The mode is determined automatically based on the Linux memory block size
56	* and the device block size.
57	*
58	* User space / core MM (auto onlining) is responsible for onlining added
59	* Linux memory blocks - and for selecting a zone. Linux Memory Blocks are
60	* always onlined separately, and all memory within a Linux memory block is
61	* onlined to the same zone - virtio-mem relies on this behavior.
62	*/
63
64	/*
65	* State of a Linux memory block in SBM.
66	*/
67	enum virtio_mem_sbm_mb_state {
68	/* Unplugged, not added to Linux. Can be reused later. */
69	VIRTIO_MEM_SBM_MB_UNUSED = 0,
70	/* (Partially) plugged, not added to Linux. Error on add_memory(). */
71	VIRTIO_MEM_SBM_MB_PLUGGED,
72	/* Fully plugged, fully added to Linux, offline. */
73	VIRTIO_MEM_SBM_MB_OFFLINE,
74	/* Partially plugged, fully added to Linux, offline. */
75	VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
76	/* Fully plugged, fully added to Linux, onlined to a kernel zone. */
77	VIRTIO_MEM_SBM_MB_KERNEL,
78	/* Partially plugged, fully added to Linux, online to a kernel zone */
79	VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
80	/* Fully plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
81	VIRTIO_MEM_SBM_MB_MOVABLE,
82	/* Partially plugged, fully added to Linux, onlined to ZONE_MOVABLE. */
83	VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
84	VIRTIO_MEM_SBM_MB_COUNT
85	};
86
87	/*
88	* State of a Big Block (BB) in BBM, covering 1..X Linux memory blocks.
89	*/
90	enum virtio_mem_bbm_bb_state {
91	/* Unplugged, not added to Linux. Can be reused later. */
92	VIRTIO_MEM_BBM_BB_UNUSED = 0,
93	/* Plugged, not added to Linux. Error on add_memory(). */
94	VIRTIO_MEM_BBM_BB_PLUGGED,
95	/* Plugged and added to Linux. */
96	VIRTIO_MEM_BBM_BB_ADDED,
97	/* All online parts are fake-offline, ready to remove. */
98	VIRTIO_MEM_BBM_BB_FAKE_OFFLINE,
99	VIRTIO_MEM_BBM_BB_COUNT
100	};
101
102	struct virtio_mem {
103	struct virtio_device *vdev;
104
105	/* We might first have to unplug all memory when starting up. */
106	bool unplug_all_required;
107
108	/* Workqueue that processes the plug/unplug requests. */
109	struct work_struct wq;
110	atomic_t wq_active;
111	atomic_t config_changed;
112
113	/* Virtqueue for guest->host requests. */
114	struct virtqueue *vq;
115
116	/* Wait for a host response to a guest request. */
117	wait_queue_head_t host_resp;
118
119	/* Space for one guest request and the host response. */
120	struct virtio_mem_req req;
121	struct virtio_mem_resp resp;
122
123	/* The current size of the device. */
124	uint64_t plugged_size;
125	/* The requested size of the device. */
126	uint64_t requested_size;
127
128	/* The device block size (for communicating with the device). */
129	uint64_t device_block_size;
130	/* The determined node id for all memory of the device. */
131	int nid;
132	/* Physical start address of the memory region. */
133	uint64_t addr;
134	/* Maximum region size in bytes. */
135	uint64_t region_size;
136	/* Usable region size in bytes. */
137	uint64_t usable_region_size;
138
139	/* The parent resource for all memory added via this device. */
140	struct resource *parent_resource;
141	/*
142	* Copy of "System RAM (virtio_mem)" to be used for
143	* add_memory_driver_managed().
144	*/
145	const char *resource_name;
146	/* Memory group identification. */
147	int mgid;
148
149	/*
150	* We don't want to add too much memory if it's not getting onlined,
151	* to avoid running OOM. Besides this threshold, we allow to have at
152	* least two offline blocks at a time (whatever is bigger).
153	*/
154	#define VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD (1024 * 1024 * 1024)
155	atomic64_t offline_size;
156	uint64_t offline_threshold;
157
158	/* If set, the driver is in SBM, otherwise in BBM. */
159	bool in_sbm;
160
161	union {
162	struct {
163	/* Id of the first memory block of this device. */
164	unsigned long first_mb_id;
165	/* Id of the last usable memory block of this device. */
166	unsigned long last_usable_mb_id;
167	/* Id of the next memory bock to prepare when needed. */
168	unsigned long next_mb_id;
169
170	/* The subblock size. */
171	uint64_t sb_size;
172	/* The number of subblocks per Linux memory block. */
173	uint32_t sbs_per_mb;
174
175	/*
176	* Some of the Linux memory blocks tracked as "partially
177	* plugged" are completely unplugged and can be offlined
178	* and removed -- which previously failed.
179	*/
180	bool have_unplugged_mb;
181
182	/* Summary of all memory block states. */
183	unsigned long mb_count[VIRTIO_MEM_SBM_MB_COUNT];
184
185	/*
186	* One byte state per memory block. Allocated via
187	* vmalloc(). Resized (alloc+copy+free) on demand.
188	*
189	* With 128 MiB memory blocks, we have states for 512
190	* GiB of memory in one 4 KiB page.
191	*/
192	uint8_t *mb_states;
193
194	/*
195	* Bitmap: one bit per subblock. Allocated similar to
196	* sbm.mb_states.
197	*
198	* A set bit means the corresponding subblock is
199	* plugged, otherwise it's unblocked.
200	*
201	* With 4 MiB subblocks, we manage 128 GiB of memory
202	* in one 4 KiB page.
203	*/
204	unsigned long *sb_states;
205	} sbm;
206
207	struct {
208	/* Id of the first big block of this device. */
209	unsigned long first_bb_id;
210	/* Id of the last usable big block of this device. */
211	unsigned long last_usable_bb_id;
212	/* Id of the next device bock to prepare when needed. */
213	unsigned long next_bb_id;
214
215	/* Summary of all big block states. */
216	unsigned long bb_count[VIRTIO_MEM_BBM_BB_COUNT];
217
218	/* One byte state per big block. See sbm.mb_states. */
219	uint8_t *bb_states;
220
221	/* The block size used for plugging/adding/removing. */
222	uint64_t bb_size;
223	} bbm;
224	};
225
226	/*
227	* Mutex that protects the sbm.mb_count, sbm.mb_states,
228	* sbm.sb_states, bbm.bb_count, and bbm.bb_states
229	*
230	* When this lock is held the pointers can't change, ONLINE and
231	* OFFLINE blocks can't change the state and no subblocks will get
232	* plugged/unplugged.
233	*
234	* In kdump mode, used to serialize requests, last_block_addr and
235	* last_block_plugged.
236	*/
237	struct mutex hotplug_mutex;
238	bool hotplug_active;
239
240	/* An error occurred we cannot handle - stop processing requests. */
241	bool broken;
242
243	/* Cached valued of is_kdump_kernel() when the device was probed. */
244	bool in_kdump;
245
246	/* The driver is being removed. */
247	spinlock_t removal_lock;
248	bool removing;
249
250	/* Timer for retrying to plug/unplug memory. */
251	struct hrtimer retry_timer;
252	unsigned int retry_timer_ms;
253	#define VIRTIO_MEM_RETRY_TIMER_MIN_MS 50000
254	#define VIRTIO_MEM_RETRY_TIMER_MAX_MS 300000
255
256	/* Memory notifier (online/offline events). */
257	struct notifier_block memory_notifier;
258
259	/* Notifier to block hibernation image storing/reloading. */
260	struct notifier_block pm_notifier;
261
262	#ifdef CONFIG_PROC_VMCORE
263	/* vmcore callback for /proc/vmcore handling in kdump mode */
264	struct vmcore_cb vmcore_cb;
265	uint64_t last_block_addr;
266	bool last_block_plugged;
267	#endif /* CONFIG_PROC_VMCORE */
268
269	/* Next device in the list of virtio-mem devices. */
270	struct list_head next;
271	};
272
273	/*
274	* We have to share a single online_page callback among all virtio-mem
275	* devices. We use RCU to iterate the list in the callback.
276	*/
277	static DEFINE_MUTEX(virtio_mem_mutex);
278	static LIST_HEAD(virtio_mem_devices);
279
280	static void virtio_mem_online_page_cb(struct page *page, unsigned int order);
281	static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
282	unsigned long nr_pages);
283	static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
284	unsigned long nr_pages);
285	static void virtio_mem_retry(struct virtio_mem *vm);
286	static int virtio_mem_create_resource(struct virtio_mem *vm);
287	static void virtio_mem_delete_resource(struct virtio_mem *vm);
288
289	/*
290	* Register a virtio-mem device so it will be considered for the online_page
291	* callback.
292	*/
293	static int register_virtio_mem_device(struct virtio_mem *vm)
294	{
295	int rc = 0;
296
297	/* First device registers the callback. */
298	mutex_lock(&virtio_mem_mutex);
299	if (list_empty(head: &virtio_mem_devices))
300	rc = set_online_page_callback(&virtio_mem_online_page_cb);
301	if (!rc)
302	list_add_rcu(new: &vm->next, head: &virtio_mem_devices);
303	mutex_unlock(lock: &virtio_mem_mutex);
304
305	return rc;
306	}
307
308	/*
309	* Unregister a virtio-mem device so it will no longer be considered for the
310	* online_page callback.
311	*/
312	static void unregister_virtio_mem_device(struct virtio_mem *vm)
313	{
314	/* Last device unregisters the callback. */
315	mutex_lock(&virtio_mem_mutex);
316	list_del_rcu(entry: &vm->next);
317	if (list_empty(head: &virtio_mem_devices))
318	restore_online_page_callback(callback: &virtio_mem_online_page_cb);
319	mutex_unlock(lock: &virtio_mem_mutex);
320
321	synchronize_rcu();
322	}
323
324	/*
325	* Calculate the memory block id of a given address.
326	*/
327	static unsigned long virtio_mem_phys_to_mb_id(unsigned long addr)
328	{
329	return addr / memory_block_size_bytes();
330	}
331
332	/*
333	* Calculate the physical start address of a given memory block id.
334	*/
335	static unsigned long virtio_mem_mb_id_to_phys(unsigned long mb_id)
336	{
337	return mb_id * memory_block_size_bytes();
338	}
339
340	/*
341	* Calculate the big block id of a given address.
342	*/
343	static unsigned long virtio_mem_phys_to_bb_id(struct virtio_mem *vm,
344	uint64_t addr)
345	{
346	return addr / vm->bbm.bb_size;
347	}
348
349	/*
350	* Calculate the physical start address of a given big block id.
351	*/
352	static uint64_t virtio_mem_bb_id_to_phys(struct virtio_mem *vm,
353	unsigned long bb_id)
354	{
355	return bb_id * vm->bbm.bb_size;
356	}
357
358	/*
359	* Calculate the subblock id of a given address.
360	*/
361	static unsigned long virtio_mem_phys_to_sb_id(struct virtio_mem *vm,
362	unsigned long addr)
363	{
364	const unsigned long mb_id = virtio_mem_phys_to_mb_id(addr);
365	const unsigned long mb_addr = virtio_mem_mb_id_to_phys(mb_id);
366
367	return (addr - mb_addr) / vm->sbm.sb_size;
368	}
369
370	/*
371	* Set the state of a big block, taking care of the state counter.
372	*/
373	static void virtio_mem_bbm_set_bb_state(struct virtio_mem *vm,
374	unsigned long bb_id,
375	enum virtio_mem_bbm_bb_state state)
376	{
377	const unsigned long idx = bb_id - vm->bbm.first_bb_id;
378	enum virtio_mem_bbm_bb_state old_state;
379
380	old_state = vm->bbm.bb_states[idx];
381	vm->bbm.bb_states[idx] = state;
382
383	BUG_ON(vm->bbm.bb_count[old_state] == 0);
384	vm->bbm.bb_count[old_state]--;
385	vm->bbm.bb_count[state]++;
386	}
387
388	/*
389	* Get the state of a big block.
390	*/
391	static enum virtio_mem_bbm_bb_state virtio_mem_bbm_get_bb_state(struct virtio_mem *vm,
392	unsigned long bb_id)
393	{
394	return vm->bbm.bb_states[bb_id - vm->bbm.first_bb_id];
395	}
396
397	/*
398	* Prepare the big block state array for the next big block.
399	*/
400	static int virtio_mem_bbm_bb_states_prepare_next_bb(struct virtio_mem *vm)
401	{
402	unsigned long old_bytes = vm->bbm.next_bb_id - vm->bbm.first_bb_id;
403	unsigned long new_bytes = old_bytes + 1;
404	int old_pages = PFN_UP(old_bytes);
405	int new_pages = PFN_UP(new_bytes);
406	uint8_t *new_array;
407
408	if (vm->bbm.bb_states && old_pages == new_pages)
409	return 0;
410
411	new_array = vzalloc(new_pages * PAGE_SIZE);
412	if (!new_array)
413	return -ENOMEM;
414
415	mutex_lock(&vm->hotplug_mutex);
416	if (vm->bbm.bb_states)
417	memcpy(new_array, vm->bbm.bb_states, old_pages * PAGE_SIZE);
418	vfree(addr: vm->bbm.bb_states);
419	vm->bbm.bb_states = new_array;
420	mutex_unlock(lock: &vm->hotplug_mutex);
421
422	return 0;
423	}
424
425	#define virtio_mem_bbm_for_each_bb(_vm, _bb_id, _state) \
426	for (_bb_id = vm->bbm.first_bb_id; \
427	_bb_id < vm->bbm.next_bb_id && _vm->bbm.bb_count[_state]; \
428	_bb_id++) \
429	if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
430
431	#define virtio_mem_bbm_for_each_bb_rev(_vm, _bb_id, _state) \
432	for (_bb_id = vm->bbm.next_bb_id - 1; \
433	_bb_id >= vm->bbm.first_bb_id && _vm->bbm.bb_count[_state]; \
434	_bb_id--) \
435	if (virtio_mem_bbm_get_bb_state(_vm, _bb_id) == _state)
436
437	/*
438	* Set the state of a memory block, taking care of the state counter.
439	*/
440	static void virtio_mem_sbm_set_mb_state(struct virtio_mem *vm,
441	unsigned long mb_id, uint8_t state)
442	{
443	const unsigned long idx = mb_id - vm->sbm.first_mb_id;
444	uint8_t old_state;
445
446	old_state = vm->sbm.mb_states[idx];
447	vm->sbm.mb_states[idx] = state;
448
449	BUG_ON(vm->sbm.mb_count[old_state] == 0);
450	vm->sbm.mb_count[old_state]--;
451	vm->sbm.mb_count[state]++;
452	}
453
454	/*
455	* Get the state of a memory block.
456	*/
457	static uint8_t virtio_mem_sbm_get_mb_state(struct virtio_mem *vm,
458	unsigned long mb_id)
459	{
460	const unsigned long idx = mb_id - vm->sbm.first_mb_id;
461
462	return vm->sbm.mb_states[idx];
463	}
464
465	/*
466	* Prepare the state array for the next memory block.
467	*/
468	static int virtio_mem_sbm_mb_states_prepare_next_mb(struct virtio_mem *vm)
469	{
470	int old_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id);
471	int new_pages = PFN_UP(vm->sbm.next_mb_id - vm->sbm.first_mb_id + 1);
472	uint8_t *new_array;
473
474	if (vm->sbm.mb_states && old_pages == new_pages)
475	return 0;
476
477	new_array = vzalloc(new_pages * PAGE_SIZE);
478	if (!new_array)
479	return -ENOMEM;
480
481	mutex_lock(&vm->hotplug_mutex);
482	if (vm->sbm.mb_states)
483	memcpy(new_array, vm->sbm.mb_states, old_pages * PAGE_SIZE);
484	vfree(addr: vm->sbm.mb_states);
485	vm->sbm.mb_states = new_array;
486	mutex_unlock(lock: &vm->hotplug_mutex);
487
488	return 0;
489	}
490
491	#define virtio_mem_sbm_for_each_mb(_vm, _mb_id, _state) \
492	for (_mb_id = _vm->sbm.first_mb_id; \
493	_mb_id < _vm->sbm.next_mb_id && _vm->sbm.mb_count[_state]; \
494	_mb_id++) \
495	if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
496
497	#define virtio_mem_sbm_for_each_mb_rev(_vm, _mb_id, _state) \
498	for (_mb_id = _vm->sbm.next_mb_id - 1; \
499	_mb_id >= _vm->sbm.first_mb_id && _vm->sbm.mb_count[_state]; \
500	_mb_id--) \
501	if (virtio_mem_sbm_get_mb_state(_vm, _mb_id) == _state)
502
503	/*
504	* Calculate the bit number in the subblock bitmap for the given subblock
505	* inside the given memory block.
506	*/
507	static int virtio_mem_sbm_sb_state_bit_nr(struct virtio_mem *vm,
508	unsigned long mb_id, int sb_id)
509	{
510	return (mb_id - vm->sbm.first_mb_id) * vm->sbm.sbs_per_mb + sb_id;
511	}
512
513	/*
514	* Mark all selected subblocks plugged.
515	*
516	* Will not modify the state of the memory block.
517	*/
518	static void virtio_mem_sbm_set_sb_plugged(struct virtio_mem *vm,
519	unsigned long mb_id, int sb_id,
520	int count)
521	{
522	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
523
524	__bitmap_set(map: vm->sbm.sb_states, start: bit, len: count);
525	}
526
527	/*
528	* Mark all selected subblocks unplugged.
529	*
530	* Will not modify the state of the memory block.
531	*/
532	static void virtio_mem_sbm_set_sb_unplugged(struct virtio_mem *vm,
533	unsigned long mb_id, int sb_id,
534	int count)
535	{
536	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
537
538	__bitmap_clear(map: vm->sbm.sb_states, start: bit, len: count);
539	}
540
541	/*
542	* Test if all selected subblocks are plugged.
543	*/
544	static bool virtio_mem_sbm_test_sb_plugged(struct virtio_mem *vm,
545	unsigned long mb_id, int sb_id,
546	int count)
547	{
548	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
549
550	if (count == 1)
551	return test_bit(bit, vm->sbm.sb_states);
552
553	/* TODO: Helper similar to bitmap_set() */
554	return find_next_zero_bit(addr: vm->sbm.sb_states, size: bit + count, offset: bit) >=
555	bit + count;
556	}
557
558	/*
559	* Test if all selected subblocks are unplugged.
560	*/
561	static bool virtio_mem_sbm_test_sb_unplugged(struct virtio_mem *vm,
562	unsigned long mb_id, int sb_id,
563	int count)
564	{
565	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id);
566
567	/* TODO: Helper similar to bitmap_set() */
568	return find_next_bit(addr: vm->sbm.sb_states, size: bit + count, offset: bit) >=
569	bit + count;
570	}
571
572	/*
573	* Find the first unplugged subblock. Returns vm->sbm.sbs_per_mb in case there is
574	* none.
575	*/
576	static int virtio_mem_sbm_first_unplugged_sb(struct virtio_mem *vm,
577	unsigned long mb_id)
578	{
579	const int bit = virtio_mem_sbm_sb_state_bit_nr(vm, mb_id, sb_id: 0);
580
581	return find_next_zero_bit(addr: vm->sbm.sb_states,
582	size: bit + vm->sbm.sbs_per_mb, offset: bit) - bit;
583	}
584
585	/*
586	* Prepare the subblock bitmap for the next memory block.
587	*/
588	static int virtio_mem_sbm_sb_states_prepare_next_mb(struct virtio_mem *vm)
589	{
590	const unsigned long old_nb_mb = vm->sbm.next_mb_id - vm->sbm.first_mb_id;
591	const unsigned long old_nb_bits = old_nb_mb * vm->sbm.sbs_per_mb;
592	const unsigned long new_nb_bits = (old_nb_mb + 1) * vm->sbm.sbs_per_mb;
593	int old_pages = PFN_UP(BITS_TO_LONGS(old_nb_bits) * sizeof(long));
594	int new_pages = PFN_UP(BITS_TO_LONGS(new_nb_bits) * sizeof(long));
595	unsigned long *new_bitmap, *old_bitmap;
596
597	if (vm->sbm.sb_states && old_pages == new_pages)
598	return 0;
599
600	new_bitmap = vzalloc(new_pages * PAGE_SIZE);
601	if (!new_bitmap)
602	return -ENOMEM;
603
604	mutex_lock(&vm->hotplug_mutex);
605	if (vm->sbm.sb_states)
606	memcpy(new_bitmap, vm->sbm.sb_states, old_pages * PAGE_SIZE);
607
608	old_bitmap = vm->sbm.sb_states;
609	vm->sbm.sb_states = new_bitmap;
610	mutex_unlock(lock: &vm->hotplug_mutex);
611
612	vfree(addr: old_bitmap);
613	return 0;
614	}
615
616	/*
617	* Test if we could add memory without creating too much offline memory -
618	* to avoid running OOM if memory is getting onlined deferred.
619	*/
620	static bool virtio_mem_could_add_memory(struct virtio_mem *vm, uint64_t size)
621	{
622	if (WARN_ON_ONCE(size > vm->offline_threshold))
623	return false;
624
625	return atomic64_read(v: &vm->offline_size) + size <= vm->offline_threshold;
626	}
627
628	/*
629	* Try adding memory to Linux. Will usually only fail if out of memory.
630	*
631	* Must not be called with the vm->hotplug_mutex held (possible deadlock with
632	* onlining code).
633	*
634	* Will not modify the state of memory blocks in virtio-mem.
635	*/
636	static int virtio_mem_add_memory(struct virtio_mem *vm, uint64_t addr,
637	uint64_t size)
638	{
639	int rc;
640
641	/*
642	* When force-unloading the driver and we still have memory added to
643	* Linux, the resource name has to stay.
644	*/
645	if (!vm->resource_name) {
646	vm->resource_name = kstrdup_const(s: "System RAM (virtio_mem)",
647	GFP_KERNEL);
648	if (!vm->resource_name)
649	return -ENOMEM;
650	}
651
652	dev_dbg(&vm->vdev->dev, "adding memory: 0x%llx - 0x%llx\n", addr,
653	addr + size - 1);
654	/* Memory might get onlined immediately. */
655	atomic64_add(i: size, v: &vm->offline_size);
656	rc = add_memory_driver_managed(nid: vm->mgid, start: addr, size, resource_name: vm->resource_name,
657	MHP_MERGE_RESOURCE | MHP_NID_IS_MGID);
658	if (rc) {
659	atomic64_sub(i: size, v: &vm->offline_size);
660	dev_warn(&vm->vdev->dev, "adding memory failed: %d\n", rc);
661	/*
662	* TODO: Linux MM does not properly clean up yet in all cases
663	* where adding of memory failed - especially on -ENOMEM.
664	*/
665	}
666	return rc;
667	}
668
669	/*
670	* See virtio_mem_add_memory(): Try adding a single Linux memory block.
671	*/
672	static int virtio_mem_sbm_add_mb(struct virtio_mem *vm, unsigned long mb_id)
673	{
674	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
675	const uint64_t size = memory_block_size_bytes();
676
677	return virtio_mem_add_memory(vm, addr, size);
678	}
679
680	/*
681	* See virtio_mem_add_memory(): Try adding a big block.
682	*/
683	static int virtio_mem_bbm_add_bb(struct virtio_mem *vm, unsigned long bb_id)
684	{
685	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
686	const uint64_t size = vm->bbm.bb_size;
687
688	return virtio_mem_add_memory(vm, addr, size);
689	}
690
691	/*
692	* Try removing memory from Linux. Will only fail if memory blocks aren't
693	* offline.
694	*
695	* Must not be called with the vm->hotplug_mutex held (possible deadlock with
696	* onlining code).
697	*
698	* Will not modify the state of memory blocks in virtio-mem.
699	*/
700	static int virtio_mem_remove_memory(struct virtio_mem *vm, uint64_t addr,
701	uint64_t size)
702	{
703	int rc;
704
705	dev_dbg(&vm->vdev->dev, "removing memory: 0x%llx - 0x%llx\n", addr,
706	addr + size - 1);
707	rc = remove_memory(start: addr, size);
708	if (!rc) {
709	atomic64_sub(i: size, v: &vm->offline_size);
710	/*
711	* We might have freed up memory we can now unplug, retry
712	* immediately instead of waiting.
713	*/
714	virtio_mem_retry(vm);
715	} else {
716	dev_dbg(&vm->vdev->dev, "removing memory failed: %d\n", rc);
717	}
718	return rc;
719	}
720
721	/*
722	* See virtio_mem_remove_memory(): Try removing a single Linux memory block.
723	*/
724	static int virtio_mem_sbm_remove_mb(struct virtio_mem *vm, unsigned long mb_id)
725	{
726	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
727	const uint64_t size = memory_block_size_bytes();
728
729	return virtio_mem_remove_memory(vm, addr, size);
730	}
731
732	/*
733	* Try offlining and removing memory from Linux.
734	*
735	* Must not be called with the vm->hotplug_mutex held (possible deadlock with
736	* onlining code).
737	*
738	* Will not modify the state of memory blocks in virtio-mem.
739	*/
740	static int virtio_mem_offline_and_remove_memory(struct virtio_mem *vm,
741	uint64_t addr,
742	uint64_t size)
743	{
744	int rc;
745
746	dev_dbg(&vm->vdev->dev,
747	"offlining and removing memory: 0x%llx - 0x%llx\n", addr,
748	addr + size - 1);
749
750	rc = offline_and_remove_memory(start: addr, size);
751	if (!rc) {
752	atomic64_sub(i: size, v: &vm->offline_size);
753	/*
754	* We might have freed up memory we can now unplug, retry
755	* immediately instead of waiting.
756	*/
757	virtio_mem_retry(vm);
758	return 0;
759	}
760	dev_dbg(&vm->vdev->dev, "offlining and removing memory failed: %d\n", rc);
761	/*
762	* We don't really expect this to fail, because we fake-offlined all
763	* memory already. But it could fail in corner cases.
764	*/
765	WARN_ON_ONCE(rc != -ENOMEM && rc != -EBUSY);
766	return rc == -ENOMEM ? -ENOMEM : -EBUSY;
767	}
768
769	/*
770	* See virtio_mem_offline_and_remove_memory(): Try offlining and removing
771	* a single Linux memory block.
772	*/
773	static int virtio_mem_sbm_offline_and_remove_mb(struct virtio_mem *vm,
774	unsigned long mb_id)
775	{
776	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id);
777	const uint64_t size = memory_block_size_bytes();
778
779	return virtio_mem_offline_and_remove_memory(vm, addr, size);
780	}
781
782	/*
783	* Try (offlining and) removing memory from Linux in case all subblocks are
784	* unplugged. Can be called on online and offline memory blocks.
785	*
786	* May modify the state of memory blocks in virtio-mem.
787	*/
788	static int virtio_mem_sbm_try_remove_unplugged_mb(struct virtio_mem *vm,
789	unsigned long mb_id)
790	{
791	int rc;
792
793	/*
794	* Once all subblocks of a memory block were unplugged, offline and
795	* remove it.
796	*/
797	if (!virtio_mem_sbm_test_sb_unplugged(vm, mb_id, sb_id: 0, count: vm->sbm.sbs_per_mb))
798	return 0;
799
800	/* offline_and_remove_memory() works for online and offline memory. */
801	mutex_unlock(lock: &vm->hotplug_mutex);
802	rc = virtio_mem_sbm_offline_and_remove_mb(vm, mb_id);
803	mutex_lock(&vm->hotplug_mutex);
804	if (!rc)
805	virtio_mem_sbm_set_mb_state(vm, mb_id,
806	state: VIRTIO_MEM_SBM_MB_UNUSED);
807	return rc;
808	}
809
810	/*
811	* See virtio_mem_offline_and_remove_memory(): Try to offline and remove a
812	* all Linux memory blocks covered by the big block.
813	*/
814	static int virtio_mem_bbm_offline_and_remove_bb(struct virtio_mem *vm,
815	unsigned long bb_id)
816	{
817	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
818	const uint64_t size = vm->bbm.bb_size;
819
820	return virtio_mem_offline_and_remove_memory(vm, addr, size);
821	}
822
823	/*
824	* Trigger the workqueue so the device can perform its magic.
825	*/
826	static void virtio_mem_retry(struct virtio_mem *vm)
827	{
828	unsigned long flags;
829
830	spin_lock_irqsave(&vm->removal_lock, flags);
831	if (!vm->removing)
832	queue_work(wq: system_freezable_wq, work: &vm->wq);
833	spin_unlock_irqrestore(lock: &vm->removal_lock, flags);
834	}
835
836	static int virtio_mem_translate_node_id(struct virtio_mem *vm, uint16_t node_id)
837	{
838	int node = NUMA_NO_NODE;
839
840	#if defined(CONFIG_ACPI_NUMA)
841	if (virtio_has_feature(vdev: vm->vdev, VIRTIO_MEM_F_ACPI_PXM))
842	node = pxm_to_node(node_id);
843	#endif
844	return node;
845	}
846
847	/*
848	* Test if a virtio-mem device overlaps with the given range. Can be called
849	* from (notifier) callbacks lockless.
850	*/
851	static bool virtio_mem_overlaps_range(struct virtio_mem *vm, uint64_t start,
852	uint64_t size)
853	{
854	return start < vm->addr + vm->region_size && vm->addr < start + size;
855	}
856
857	/*
858	* Test if a virtio-mem device contains a given range. Can be called from
859	* (notifier) callbacks lockless.
860	*/
861	static bool virtio_mem_contains_range(struct virtio_mem *vm, uint64_t start,
862	uint64_t size)
863	{
864	return start >= vm->addr && start + size <= vm->addr + vm->region_size;
865	}
866
867	static int virtio_mem_sbm_notify_going_online(struct virtio_mem *vm,
868	unsigned long mb_id)
869	{
870	switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
871	case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
872	case VIRTIO_MEM_SBM_MB_OFFLINE:
873	return NOTIFY_OK;
874	default:
875	break;
876	}
877	dev_warn_ratelimited(&vm->vdev->dev,
878	"memory block onlining denied\n");
879	return NOTIFY_BAD;
880	}
881
882	static void virtio_mem_sbm_notify_offline(struct virtio_mem *vm,
883	unsigned long mb_id)
884	{
885	switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
886	case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
887	case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
888	virtio_mem_sbm_set_mb_state(vm, mb_id,
889	state: VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
890	break;
891	case VIRTIO_MEM_SBM_MB_KERNEL:
892	case VIRTIO_MEM_SBM_MB_MOVABLE:
893	virtio_mem_sbm_set_mb_state(vm, mb_id,
894	state: VIRTIO_MEM_SBM_MB_OFFLINE);
895	break;
896	default:
897	BUG();
898	break;
899	}
900	}
901
902	static void virtio_mem_sbm_notify_online(struct virtio_mem *vm,
903	unsigned long mb_id,
904	unsigned long start_pfn)
905	{
906	const bool is_movable = is_zone_movable_page(pfn_to_page(start_pfn));
907	int new_state;
908
909	switch (virtio_mem_sbm_get_mb_state(vm, mb_id)) {
910	case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
911	new_state = VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL;
912	if (is_movable)
913	new_state = VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL;
914	break;
915	case VIRTIO_MEM_SBM_MB_OFFLINE:
916	new_state = VIRTIO_MEM_SBM_MB_KERNEL;
917	if (is_movable)
918	new_state = VIRTIO_MEM_SBM_MB_MOVABLE;
919	break;
920	default:
921	BUG();
922	break;
923	}
924	virtio_mem_sbm_set_mb_state(vm, mb_id, state: new_state);
925	}
926
927	static void virtio_mem_sbm_notify_going_offline(struct virtio_mem *vm,
928	unsigned long mb_id)
929	{
930	const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
931	unsigned long pfn;
932	int sb_id;
933
934	for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
935	if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, count: 1))
936	continue;
937	pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
938	sb_id * vm->sbm.sb_size);
939	virtio_mem_fake_offline_going_offline(pfn, nr_pages);
940	}
941	}
942
943	static void virtio_mem_sbm_notify_cancel_offline(struct virtio_mem *vm,
944	unsigned long mb_id)
945	{
946	const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size);
947	unsigned long pfn;
948	int sb_id;
949
950	for (sb_id = 0; sb_id < vm->sbm.sbs_per_mb; sb_id++) {
951	if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, count: 1))
952	continue;
953	pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
954	sb_id * vm->sbm.sb_size);
955	virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
956	}
957	}
958
959	static void virtio_mem_bbm_notify_going_offline(struct virtio_mem *vm,
960	unsigned long bb_id,
961	unsigned long pfn,
962	unsigned long nr_pages)
963	{
964	/*
965	* When marked as "fake-offline", all online memory of this device block
966	* is allocated by us. Otherwise, we don't have any memory allocated.
967	*/
968	if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
969	VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
970	return;
971	virtio_mem_fake_offline_going_offline(pfn, nr_pages);
972	}
973
974	static void virtio_mem_bbm_notify_cancel_offline(struct virtio_mem *vm,
975	unsigned long bb_id,
976	unsigned long pfn,
977	unsigned long nr_pages)
978	{
979	if (virtio_mem_bbm_get_bb_state(vm, bb_id) !=
980	VIRTIO_MEM_BBM_BB_FAKE_OFFLINE)
981	return;
982	virtio_mem_fake_offline_cancel_offline(pfn, nr_pages);
983	}
984
985	/*
986	* This callback will either be called synchronously from add_memory() or
987	* asynchronously (e.g., triggered via user space). We have to be careful
988	* with locking when calling add_memory().
989	*/
990	static int virtio_mem_memory_notifier_cb(struct notifier_block *nb,
991	unsigned long action, void *arg)
992	{
993	struct virtio_mem *vm = container_of(nb, struct virtio_mem,
994	memory_notifier);
995	struct memory_notify *mhp = arg;
996	const unsigned long start = PFN_PHYS(mhp->start_pfn);
997	const unsigned long size = PFN_PHYS(mhp->nr_pages);
998	int rc = NOTIFY_OK;
999	unsigned long id;
1000
1001	if (!virtio_mem_overlaps_range(vm, start, size))
1002	return NOTIFY_DONE;
1003
1004	if (vm->in_sbm) {
1005	id = virtio_mem_phys_to_mb_id(addr: start);
1006	/*
1007	* In SBM, we add memory in separate memory blocks - we expect
1008	* it to be onlined/offlined in the same granularity. Bail out
1009	* if this ever changes.
1010	*/
1011	if (WARN_ON_ONCE(size != memory_block_size_bytes() ||
1012	!IS_ALIGNED(start, memory_block_size_bytes())))
1013	return NOTIFY_BAD;
1014	} else {
1015	id = virtio_mem_phys_to_bb_id(vm, addr: start);
1016	/*
1017	* In BBM, we only care about onlining/offlining happening
1018	* within a single big block, we don't care about the
1019	* actual granularity as we don't track individual Linux
1020	* memory blocks.
1021	*/
1022	if (WARN_ON_ONCE(id != virtio_mem_phys_to_bb_id(vm, start + size - 1)))
1023	return NOTIFY_BAD;
1024	}
1025
1026	/*
1027	* Avoid circular locking lockdep warnings. We lock the mutex
1028	* e.g., in MEM_GOING_ONLINE and unlock it in MEM_ONLINE. The
1029	* blocking_notifier_call_chain() has it's own lock, which gets unlocked
1030	* between both notifier calls and will bail out. False positive.
1031	*/
1032	lockdep_off();
1033
1034	switch (action) {
1035	case MEM_GOING_OFFLINE:
1036	mutex_lock(&vm->hotplug_mutex);
1037	if (vm->removing) {
1038	rc = notifier_from_errno(err: -EBUSY);
1039	mutex_unlock(lock: &vm->hotplug_mutex);
1040	break;
1041	}
1042	vm->hotplug_active = true;
1043	if (vm->in_sbm)
1044	virtio_mem_sbm_notify_going_offline(vm, mb_id: id);
1045	else
1046	virtio_mem_bbm_notify_going_offline(vm, bb_id: id,
1047	pfn: mhp->start_pfn,
1048	nr_pages: mhp->nr_pages);
1049	break;
1050	case MEM_GOING_ONLINE:
1051	mutex_lock(&vm->hotplug_mutex);
1052	if (vm->removing) {
1053	rc = notifier_from_errno(err: -EBUSY);
1054	mutex_unlock(lock: &vm->hotplug_mutex);
1055	break;
1056	}
1057	vm->hotplug_active = true;
1058	if (vm->in_sbm)
1059	rc = virtio_mem_sbm_notify_going_online(vm, mb_id: id);
1060	break;
1061	case MEM_OFFLINE:
1062	if (vm->in_sbm)
1063	virtio_mem_sbm_notify_offline(vm, mb_id: id);
1064
1065	atomic64_add(i: size, v: &vm->offline_size);
1066	/*
1067	* Trigger the workqueue. Now that we have some offline memory,
1068	* maybe we can handle pending unplug requests.
1069	*/
1070	if (!unplug_online)
1071	virtio_mem_retry(vm);
1072
1073	vm->hotplug_active = false;
1074	mutex_unlock(lock: &vm->hotplug_mutex);
1075	break;
1076	case MEM_ONLINE:
1077	if (vm->in_sbm)
1078	virtio_mem_sbm_notify_online(vm, mb_id: id, start_pfn: mhp->start_pfn);
1079
1080	atomic64_sub(i: size, v: &vm->offline_size);
1081	/*
1082	* Start adding more memory once we onlined half of our
1083	* threshold. Don't trigger if it's possibly due to our actipn
1084	* (e.g., us adding memory which gets onlined immediately from
1085	* the core).
1086	*/
1087	if (!atomic_read(v: &vm->wq_active) &&
1088	virtio_mem_could_add_memory(vm, size: vm->offline_threshold / 2))
1089	virtio_mem_retry(vm);
1090
1091	vm->hotplug_active = false;
1092	mutex_unlock(lock: &vm->hotplug_mutex);
1093	break;
1094	case MEM_CANCEL_OFFLINE:
1095	if (!vm->hotplug_active)
1096	break;
1097	if (vm->in_sbm)
1098	virtio_mem_sbm_notify_cancel_offline(vm, mb_id: id);
1099	else
1100	virtio_mem_bbm_notify_cancel_offline(vm, bb_id: id,
1101	pfn: mhp->start_pfn,
1102	nr_pages: mhp->nr_pages);
1103	vm->hotplug_active = false;
1104	mutex_unlock(lock: &vm->hotplug_mutex);
1105	break;
1106	case MEM_CANCEL_ONLINE:
1107	if (!vm->hotplug_active)
1108	break;
1109	vm->hotplug_active = false;
1110	mutex_unlock(lock: &vm->hotplug_mutex);
1111	break;
1112	default:
1113	break;
1114	}
1115
1116	lockdep_on();
1117
1118	return rc;
1119	}
1120
1121	static int virtio_mem_pm_notifier_cb(struct notifier_block *nb,
1122	unsigned long action, void *arg)
1123	{
1124	struct virtio_mem *vm = container_of(nb, struct virtio_mem,
1125	pm_notifier);
1126	switch (action) {
1127	case PM_HIBERNATION_PREPARE:
1128	case PM_RESTORE_PREPARE:
1129	/*
1130	* When restarting the VM, all memory is unplugged. Don't
1131	* allow to hibernate and restore from an image.
1132	*/
1133	dev_err(&vm->vdev->dev, "hibernation is not supported.\n");
1134	return NOTIFY_BAD;
1135	default:
1136	return NOTIFY_OK;
1137	}
1138	}
1139
1140	/*
1141	* Set a range of pages PG_offline. Remember pages that were never onlined
1142	* (via generic_online_page()) using PageDirty().
1143	*/
1144	static void virtio_mem_set_fake_offline(unsigned long pfn,
1145	unsigned long nr_pages, bool onlined)
1146	{
1147	page_offline_begin();
1148	for (; nr_pages--; pfn++) {
1149	struct page *page = pfn_to_page(pfn);
1150
1151	if (!onlined)
1152	/*
1153	* Pages that have not been onlined yet were initialized
1154	* to PageOffline(). Remember that we have to route them
1155	* through generic_online_page().
1156	*/
1157	SetPageDirty(page);
1158	else
1159	__SetPageOffline(page);
1160	VM_WARN_ON_ONCE(!PageOffline(page));
1161	}
1162	page_offline_end();
1163	}
1164
1165	/*
1166	* Clear PG_offline from a range of pages. If the pages were never onlined,
1167	* (via generic_online_page()), clear PageDirty().
1168	*/
1169	static void virtio_mem_clear_fake_offline(unsigned long pfn,
1170	unsigned long nr_pages, bool onlined)
1171	{
1172	for (; nr_pages--; pfn++) {
1173	struct page *page = pfn_to_page(pfn);
1174
1175	if (!onlined)
1176	/* generic_online_page() will clear PageOffline(). */
1177	ClearPageDirty(page);
1178	else
1179	__ClearPageOffline(page);
1180	}
1181	}
1182
1183	/*
1184	* Release a range of fake-offline pages to the buddy, effectively
1185	* fake-onlining them.
1186	*/
1187	static void virtio_mem_fake_online(unsigned long pfn, unsigned long nr_pages)
1188	{
1189	unsigned long order = MAX_PAGE_ORDER;
1190	unsigned long i;
1191
1192	/*
1193	* We might get called for ranges that don't cover properly aligned
1194	* MAX_PAGE_ORDER pages; however, we can only online properly aligned
1195	* pages with an order of MAX_PAGE_ORDER at maximum.
1196	*/
1197	while (!IS_ALIGNED(pfn | nr_pages, 1 << order))
1198	order--;
1199
1200	for (i = 0; i < nr_pages; i += 1 << order) {
1201	struct page *page = pfn_to_page(pfn + i);
1202
1203	/*
1204	* If the page is PageDirty(), it was kept fake-offline when
1205	* onlining the memory block. Otherwise, it was allocated
1206	* using alloc_contig_range(). All pages in a subblock are
1207	* alike.
1208	*/
1209	if (PageDirty(page)) {
1210	virtio_mem_clear_fake_offline(pfn: pfn + i, nr_pages: 1 << order, onlined: false);
1211	generic_online_page(page, order);
1212	} else {
1213	virtio_mem_clear_fake_offline(pfn: pfn + i, nr_pages: 1 << order, onlined: true);
1214	free_contig_range(pfn: pfn + i, nr_pages: 1 << order);
1215	adjust_managed_page_count(page, count: 1 << order);
1216	}
1217	}
1218	}
1219
1220	/*
1221	* Try to allocate a range, marking pages fake-offline, effectively
1222	* fake-offlining them.
1223	*/
1224	static int virtio_mem_fake_offline(struct virtio_mem *vm, unsigned long pfn,
1225	unsigned long nr_pages)
1226	{
1227	const bool is_movable = is_zone_movable_page(pfn_to_page(pfn));
1228	int rc, retry_count;
1229
1230	/*
1231	* TODO: We want an alloc_contig_range() mode that tries to allocate
1232	* harder (e.g., dealing with temporarily pinned pages, PCP), especially
1233	* with ZONE_MOVABLE. So for now, retry a couple of times with
1234	* ZONE_MOVABLE before giving up - because that zone is supposed to give
1235	* some guarantees.
1236	*/
1237	for (retry_count = 0; retry_count < 5; retry_count++) {
1238	/*
1239	* If the config changed, stop immediately and go back to the
1240	* main loop: avoid trying to keep unplugging if the device
1241	* might have decided to not remove any more memory.
1242	*/
1243	if (atomic_read(v: &vm->config_changed))
1244	return -EAGAIN;
1245
1246	rc = alloc_contig_range(pfn, pfn + nr_pages, ACR_FLAGS_NONE,
1247	GFP_KERNEL);
1248	if (rc == -ENOMEM)
1249	/* whoops, out of memory */
1250	return rc;
1251	else if (rc && !is_movable)
1252	break;
1253	else if (rc)
1254	continue;
1255
1256	virtio_mem_set_fake_offline(pfn, nr_pages, onlined: true);
1257	adjust_managed_page_count(pfn_to_page(pfn), count: -nr_pages);
1258	return 0;
1259	}
1260
1261	return -EBUSY;
1262	}
1263
1264	/*
1265	* Handle fake-offline pages when memory is going offline - such that the
1266	* pages can be skipped by mm-core when offlining.
1267	*/
1268	static void virtio_mem_fake_offline_going_offline(unsigned long pfn,
1269	unsigned long nr_pages)
1270	{
1271	struct page *page;
1272	unsigned long i;
1273
1274	/* Drop our reference to the pages so the memory can get offlined. */
1275	for (i = 0; i < nr_pages; i++) {
1276	page = pfn_to_page(pfn + i);
1277	if (WARN_ON(!page_ref_dec_and_test(page)))
1278	dump_page(page, reason: "fake-offline page referenced");
1279	}
1280	}
1281
1282	/*
1283	* Handle fake-offline pages when memory offlining is canceled - to undo
1284	* what we did in virtio_mem_fake_offline_going_offline().
1285	*/
1286	static void virtio_mem_fake_offline_cancel_offline(unsigned long pfn,
1287	unsigned long nr_pages)
1288	{
1289	unsigned long i;
1290
1291	/*
1292	* Get the reference again that we dropped via page_ref_dec_and_test()
1293	* when going offline.
1294	*/
1295	for (i = 0; i < nr_pages; i++)
1296	page_ref_inc(pfn_to_page(pfn + i));
1297	}
1298
1299	static void virtio_mem_online_page(struct virtio_mem *vm,
1300	struct page *page, unsigned int order)
1301	{
1302	const unsigned long start = page_to_phys(page);
1303	const unsigned long end = start + PFN_PHYS(1 << order);
1304	unsigned long addr, next, id, sb_id, count;
1305	bool do_online;
1306
1307	/*
1308	* We can get called with any order up to MAX_PAGE_ORDER. If our subblock
1309	* size is smaller than that and we have a mixture of plugged and
1310	* unplugged subblocks within such a page, we have to process in
1311	* smaller granularity. In that case we'll adjust the order exactly once
1312	* within the loop.
1313	*/
1314	for (addr = start; addr < end; ) {
1315	next = addr + PFN_PHYS(1 << order);
1316
1317	if (vm->in_sbm) {
1318	id = virtio_mem_phys_to_mb_id(addr);
1319	sb_id = virtio_mem_phys_to_sb_id(vm, addr);
1320	count = virtio_mem_phys_to_sb_id(vm, addr: next - 1) - sb_id + 1;
1321
1322	if (virtio_mem_sbm_test_sb_plugged(vm, mb_id: id, sb_id, count)) {
1323	/* Fully plugged. */
1324	do_online = true;
1325	} else if (count == 1 ||
1326	virtio_mem_sbm_test_sb_unplugged(vm, mb_id: id, sb_id, count)) {
1327	/* Fully unplugged. */
1328	do_online = false;
1329	} else {
1330	/*
1331	* Mixture, process sub-blocks instead. This
1332	* will be at least the size of a pageblock.
1333	* We'll run into this case exactly once.
1334	*/
1335	order = ilog2(vm->sbm.sb_size) - PAGE_SHIFT;
1336	do_online = virtio_mem_sbm_test_sb_plugged(vm, mb_id: id, sb_id, count: 1);
1337	continue;
1338	}
1339	} else {
1340	/*
1341	* If the whole block is marked fake offline, keep
1342	* everything that way.
1343	*/
1344	id = virtio_mem_phys_to_bb_id(vm, addr);
1345	do_online = virtio_mem_bbm_get_bb_state(vm, bb_id: id) !=
1346	VIRTIO_MEM_BBM_BB_FAKE_OFFLINE;
1347	}
1348
1349	if (do_online)
1350	generic_online_page(pfn_to_page(PFN_DOWN(addr)), order);
1351	else
1352	virtio_mem_set_fake_offline(PFN_DOWN(addr), nr_pages: 1 << order,
1353	onlined: false);
1354	addr = next;
1355	}
1356	}
1357
1358	static void virtio_mem_online_page_cb(struct page *page, unsigned int order)
1359	{
1360	const unsigned long addr = page_to_phys(page);
1361	struct virtio_mem *vm;
1362
1363	rcu_read_lock();
1364	list_for_each_entry_rcu(vm, &virtio_mem_devices, next) {
1365	/*
1366	* Pages we're onlining will never cross memory blocks and,
1367	* therefore, not virtio-mem devices.
1368	*/
1369	if (!virtio_mem_contains_range(vm, start: addr, PFN_PHYS(1 << order)))
1370	continue;
1371
1372	/*
1373	* virtio_mem_set_fake_offline() might sleep. We can safely
1374	* drop the RCU lock at this point because the device
1375	* cannot go away. See virtio_mem_remove() how races
1376	* between memory onlining and device removal are handled.
1377	*/
1378	rcu_read_unlock();
1379
1380	virtio_mem_online_page(vm, page, order);
1381	return;
1382	}
1383	rcu_read_unlock();
1384
1385	/* not virtio-mem memory, but e.g., a DIMM. online it */
1386	generic_online_page(page, order);
1387	}
1388
1389	static uint64_t virtio_mem_send_request(struct virtio_mem *vm,
1390	const struct virtio_mem_req *req)
1391	{
1392	struct scatterlist *sgs[2], sg_req, sg_resp;
1393	unsigned int len;
1394	int rc;
1395
1396	/* don't use the request residing on the stack (vaddr) */
1397	vm->req = *req;
1398
1399	/* out: buffer for request */
1400	sg_init_one(&sg_req, &vm->req, sizeof(vm->req));
1401	sgs[0] = &sg_req;
1402
1403	/* in: buffer for response */
1404	sg_init_one(&sg_resp, &vm->resp, sizeof(vm->resp));
1405	sgs[1] = &sg_resp;
1406
1407	rc = virtqueue_add_sgs(vq: vm->vq, sgs, out_sgs: 1, in_sgs: 1, data: vm, GFP_KERNEL);
1408	if (rc < 0)
1409	return rc;
1410
1411	virtqueue_kick(vq: vm->vq);
1412
1413	/* wait for a response */
1414	wait_event(vm->host_resp, virtqueue_get_buf(vm->vq, &len));
1415
1416	return virtio16_to_cpu(vdev: vm->vdev, val: vm->resp.type);
1417	}
1418
1419	static int virtio_mem_send_plug_request(struct virtio_mem *vm, uint64_t addr,
1420	uint64_t size)
1421	{
1422	const uint64_t nb_vm_blocks = size / vm->device_block_size;
1423	const struct virtio_mem_req req = {
1424	.type = cpu_to_virtio16(vdev: vm->vdev, VIRTIO_MEM_REQ_PLUG),
1425	.u.plug.addr = cpu_to_virtio64(vdev: vm->vdev, val: addr),
1426	.u.plug.nb_blocks = cpu_to_virtio16(vdev: vm->vdev, val: nb_vm_blocks),
1427	};
1428	int rc = -ENOMEM;
1429
1430	if (atomic_read(v: &vm->config_changed))
1431	return -EAGAIN;
1432
1433	dev_dbg(&vm->vdev->dev, "plugging memory: 0x%llx - 0x%llx\n", addr,
1434	addr + size - 1);
1435
1436	switch (virtio_mem_send_request(vm, req: &req)) {
1437	case VIRTIO_MEM_RESP_ACK:
1438	vm->plugged_size += size;
1439	return 0;
1440	case VIRTIO_MEM_RESP_NACK:
1441	rc = -EAGAIN;
1442	break;
1443	case VIRTIO_MEM_RESP_BUSY:
1444	rc = -ETXTBSY;
1445	break;
1446	case VIRTIO_MEM_RESP_ERROR:
1447	rc = -EINVAL;
1448	break;
1449	default:
1450	break;
1451	}
1452
1453	dev_dbg(&vm->vdev->dev, "plugging memory failed: %d\n", rc);
1454	return rc;
1455	}
1456
1457	static int virtio_mem_send_unplug_request(struct virtio_mem *vm, uint64_t addr,
1458	uint64_t size)
1459	{
1460	const uint64_t nb_vm_blocks = size / vm->device_block_size;
1461	const struct virtio_mem_req req = {
1462	.type = cpu_to_virtio16(vdev: vm->vdev, VIRTIO_MEM_REQ_UNPLUG),
1463	.u.unplug.addr = cpu_to_virtio64(vdev: vm->vdev, val: addr),
1464	.u.unplug.nb_blocks = cpu_to_virtio16(vdev: vm->vdev, val: nb_vm_blocks),
1465	};
1466	int rc = -ENOMEM;
1467
1468	if (atomic_read(v: &vm->config_changed))
1469	return -EAGAIN;
1470
1471	dev_dbg(&vm->vdev->dev, "unplugging memory: 0x%llx - 0x%llx\n", addr,
1472	addr + size - 1);
1473
1474	switch (virtio_mem_send_request(vm, req: &req)) {
1475	case VIRTIO_MEM_RESP_ACK:
1476	vm->plugged_size -= size;
1477	return 0;
1478	case VIRTIO_MEM_RESP_BUSY:
1479	rc = -ETXTBSY;
1480	break;
1481	case VIRTIO_MEM_RESP_ERROR:
1482	rc = -EINVAL;
1483	break;
1484	default:
1485	break;
1486	}
1487
1488	dev_dbg(&vm->vdev->dev, "unplugging memory failed: %d\n", rc);
1489	return rc;
1490	}
1491
1492	static int virtio_mem_send_unplug_all_request(struct virtio_mem *vm)
1493	{
1494	const struct virtio_mem_req req = {
1495	.type = cpu_to_virtio16(vdev: vm->vdev, VIRTIO_MEM_REQ_UNPLUG_ALL),
1496	};
1497	int rc = -ENOMEM;
1498
1499	dev_dbg(&vm->vdev->dev, "unplugging all memory");
1500
1501	switch (virtio_mem_send_request(vm, req: &req)) {
1502	case VIRTIO_MEM_RESP_ACK:
1503	vm->unplug_all_required = false;
1504	vm->plugged_size = 0;
1505	/* usable region might have shrunk */
1506	atomic_set(v: &vm->config_changed, i: 1);
1507	return 0;
1508	case VIRTIO_MEM_RESP_BUSY:
1509	rc = -ETXTBSY;
1510	break;
1511	default:
1512	break;
1513	}
1514
1515	dev_dbg(&vm->vdev->dev, "unplugging all memory failed: %d\n", rc);
1516	return rc;
1517	}
1518
1519	/*
1520	* Plug selected subblocks. Updates the plugged state, but not the state
1521	* of the memory block.
1522	*/
1523	static int virtio_mem_sbm_plug_sb(struct virtio_mem *vm, unsigned long mb_id,
1524	int sb_id, int count)
1525	{
1526	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
1527	sb_id * vm->sbm.sb_size;
1528	const uint64_t size = count * vm->sbm.sb_size;
1529	int rc;
1530
1531	rc = virtio_mem_send_plug_request(vm, addr, size);
1532	if (!rc)
1533	virtio_mem_sbm_set_sb_plugged(vm, mb_id, sb_id, count);
1534	return rc;
1535	}
1536
1537	/*
1538	* Unplug selected subblocks. Updates the plugged state, but not the state
1539	* of the memory block.
1540	*/
1541	static int virtio_mem_sbm_unplug_sb(struct virtio_mem *vm, unsigned long mb_id,
1542	int sb_id, int count)
1543	{
1544	const uint64_t addr = virtio_mem_mb_id_to_phys(mb_id) +
1545	sb_id * vm->sbm.sb_size;
1546	const uint64_t size = count * vm->sbm.sb_size;
1547	int rc;
1548
1549	rc = virtio_mem_send_unplug_request(vm, addr, size);
1550	if (!rc)
1551	virtio_mem_sbm_set_sb_unplugged(vm, mb_id, sb_id, count);
1552	return rc;
1553	}
1554
1555	/*
1556	* Request to unplug a big block.
1557	*
1558	* Will not modify the state of the big block.
1559	*/
1560	static int virtio_mem_bbm_unplug_bb(struct virtio_mem *vm, unsigned long bb_id)
1561	{
1562	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
1563	const uint64_t size = vm->bbm.bb_size;
1564
1565	return virtio_mem_send_unplug_request(vm, addr, size);
1566	}
1567
1568	/*
1569	* Request to plug a big block.
1570	*
1571	* Will not modify the state of the big block.
1572	*/
1573	static int virtio_mem_bbm_plug_bb(struct virtio_mem *vm, unsigned long bb_id)
1574	{
1575	const uint64_t addr = virtio_mem_bb_id_to_phys(vm, bb_id);
1576	const uint64_t size = vm->bbm.bb_size;
1577
1578	return virtio_mem_send_plug_request(vm, addr, size);
1579	}
1580
1581	/*
1582	* Unplug the desired number of plugged subblocks of a offline or not-added
1583	* memory block. Will fail if any subblock cannot get unplugged (instead of
1584	* skipping it).
1585	*
1586	* Will not modify the state of the memory block.
1587	*
1588	* Note: can fail after some subblocks were unplugged.
1589	*/
1590	static int virtio_mem_sbm_unplug_any_sb_raw(struct virtio_mem *vm,
1591	unsigned long mb_id, uint64_t *nb_sb)
1592	{
1593	int sb_id, count;
1594	int rc;
1595
1596	sb_id = vm->sbm.sbs_per_mb - 1;
1597	while (*nb_sb) {
1598	/* Find the next candidate subblock */
1599	while (sb_id >= 0 &&
1600	virtio_mem_sbm_test_sb_unplugged(vm, mb_id, sb_id, count: 1))
1601	sb_id--;
1602	if (sb_id < 0)
1603	break;
1604	/* Try to unplug multiple subblocks at a time */
1605	count = 1;
1606	while (count < *nb_sb && sb_id > 0 &&
1607	virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id: sb_id - 1, count: 1)) {
1608	count++;
1609	sb_id--;
1610	}
1611
1612	rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
1613	if (rc)
1614	return rc;
1615	*nb_sb -= count;
1616	sb_id--;
1617	}
1618
1619	return 0;
1620	}
1621
1622	/*
1623	* Unplug all plugged subblocks of an offline or not-added memory block.
1624	*
1625	* Will not modify the state of the memory block.
1626	*
1627	* Note: can fail after some subblocks were unplugged.
1628	*/
1629	static int virtio_mem_sbm_unplug_mb(struct virtio_mem *vm, unsigned long mb_id)
1630	{
1631	uint64_t nb_sb = vm->sbm.sbs_per_mb;
1632
1633	return virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, nb_sb: &nb_sb);
1634	}
1635
1636	/*
1637	* Prepare tracking data for the next memory block.
1638	*/
1639	static int virtio_mem_sbm_prepare_next_mb(struct virtio_mem *vm,
1640	unsigned long *mb_id)
1641	{
1642	int rc;
1643
1644	if (vm->sbm.next_mb_id > vm->sbm.last_usable_mb_id)
1645	return -ENOSPC;
1646
1647	/* Resize the state array if required. */
1648	rc = virtio_mem_sbm_mb_states_prepare_next_mb(vm);
1649	if (rc)
1650	return rc;
1651
1652	/* Resize the subblock bitmap if required. */
1653	rc = virtio_mem_sbm_sb_states_prepare_next_mb(vm);
1654	if (rc)
1655	return rc;
1656
1657	vm->sbm.mb_count[VIRTIO_MEM_SBM_MB_UNUSED]++;
1658	*mb_id = vm->sbm.next_mb_id++;
1659	return 0;
1660	}
1661
1662	/*
1663	* Try to plug the desired number of subblocks and add the memory block
1664	* to Linux.
1665	*
1666	* Will modify the state of the memory block.
1667	*/
1668	static int virtio_mem_sbm_plug_and_add_mb(struct virtio_mem *vm,
1669	unsigned long mb_id, uint64_t *nb_sb)
1670	{
1671	const int count = min_t(int, *nb_sb, vm->sbm.sbs_per_mb);
1672	int rc;
1673
1674	if (WARN_ON_ONCE(!count))
1675	return -EINVAL;
1676
1677	/*
1678	* Plug the requested number of subblocks before adding it to linux,
1679	* so that onlining will directly online all plugged subblocks.
1680	*/
1681	rc = virtio_mem_sbm_plug_sb(vm, mb_id, sb_id: 0, count);
1682	if (rc)
1683	return rc;
1684
1685	/*
1686	* Mark the block properly offline before adding it to Linux,
1687	* so the memory notifiers will find the block in the right state.
1688	*/
1689	if (count == vm->sbm.sbs_per_mb)
1690	virtio_mem_sbm_set_mb_state(vm, mb_id,
1691	state: VIRTIO_MEM_SBM_MB_OFFLINE);
1692	else
1693	virtio_mem_sbm_set_mb_state(vm, mb_id,
1694	state: VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
1695
1696	/* Add the memory block to linux - if that fails, try to unplug. */
1697	rc = virtio_mem_sbm_add_mb(vm, mb_id);
1698	if (rc) {
1699	int new_state = VIRTIO_MEM_SBM_MB_UNUSED;
1700
1701	if (virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id: 0, count))
1702	new_state = VIRTIO_MEM_SBM_MB_PLUGGED;
1703	virtio_mem_sbm_set_mb_state(vm, mb_id, state: new_state);
1704	return rc;
1705	}
1706
1707	*nb_sb -= count;
1708	return 0;
1709	}
1710
1711	/*
1712	* Try to plug the desired number of subblocks of a memory block that
1713	* is already added to Linux.
1714	*
1715	* Will modify the state of the memory block.
1716	*
1717	* Note: Can fail after some subblocks were successfully plugged.
1718	*/
1719	static int virtio_mem_sbm_plug_any_sb(struct virtio_mem *vm,
1720	unsigned long mb_id, uint64_t *nb_sb)
1721	{
1722	const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
1723	unsigned long pfn, nr_pages;
1724	int sb_id, count;
1725	int rc;
1726
1727	if (WARN_ON_ONCE(!*nb_sb))
1728	return -EINVAL;
1729
1730	while (*nb_sb) {
1731	sb_id = virtio_mem_sbm_first_unplugged_sb(vm, mb_id);
1732	if (sb_id >= vm->sbm.sbs_per_mb)
1733	break;
1734	count = 1;
1735	while (count < *nb_sb &&
1736	sb_id + count < vm->sbm.sbs_per_mb &&
1737	!virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id: sb_id + count, count: 1))
1738	count++;
1739
1740	rc = virtio_mem_sbm_plug_sb(vm, mb_id, sb_id, count);
1741	if (rc)
1742	return rc;
1743	*nb_sb -= count;
1744	if (old_state == VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL)
1745	continue;
1746
1747	/* fake-online the pages if the memory block is online */
1748	pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
1749	sb_id * vm->sbm.sb_size);
1750	nr_pages = PFN_DOWN(count * vm->sbm.sb_size);
1751	virtio_mem_fake_online(pfn, nr_pages);
1752	}
1753
1754	if (virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id: 0, count: vm->sbm.sbs_per_mb))
1755	virtio_mem_sbm_set_mb_state(vm, mb_id, state: old_state - 1);
1756
1757	return 0;
1758	}
1759
1760	static int virtio_mem_sbm_plug_request(struct virtio_mem *vm, uint64_t diff)
1761	{
1762	const int mb_states[] = {
1763	VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
1764	VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
1765	VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
1766	};
1767	uint64_t nb_sb = diff / vm->sbm.sb_size;
1768	unsigned long mb_id;
1769	int rc, i;
1770
1771	if (!nb_sb)
1772	return 0;
1773
1774	/* Don't race with onlining/offlining */
1775	mutex_lock(&vm->hotplug_mutex);
1776
1777	for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
1778	virtio_mem_sbm_for_each_mb(vm, mb_id, mb_states[i]) {
1779	rc = virtio_mem_sbm_plug_any_sb(vm, mb_id, nb_sb: &nb_sb);
1780	if (rc || !nb_sb)
1781	goto out_unlock;
1782	cond_resched();
1783	}
1784	}
1785
1786	/*
1787	* We won't be working on online/offline memory blocks from this point,
1788	* so we can't race with memory onlining/offlining. Drop the mutex.
1789	*/
1790	mutex_unlock(lock: &vm->hotplug_mutex);
1791
1792	/* Try to plug and add unused blocks */
1793	virtio_mem_sbm_for_each_mb(vm, mb_id, VIRTIO_MEM_SBM_MB_UNUSED) {
1794	if (!virtio_mem_could_add_memory(vm, size: memory_block_size_bytes()))
1795	return -ENOSPC;
1796
1797	rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, nb_sb: &nb_sb);
1798	if (rc || !nb_sb)
1799	return rc;
1800	cond_resched();
1801	}
1802
1803	/* Try to prepare, plug and add new blocks */
1804	while (nb_sb) {
1805	if (!virtio_mem_could_add_memory(vm, size: memory_block_size_bytes()))
1806	return -ENOSPC;
1807
1808	rc = virtio_mem_sbm_prepare_next_mb(vm, mb_id: &mb_id);
1809	if (rc)
1810	return rc;
1811	rc = virtio_mem_sbm_plug_and_add_mb(vm, mb_id, nb_sb: &nb_sb);
1812	if (rc)
1813	return rc;
1814	cond_resched();
1815	}
1816
1817	return 0;
1818	out_unlock:
1819	mutex_unlock(lock: &vm->hotplug_mutex);
1820	return rc;
1821	}
1822
1823	/*
1824	* Plug a big block and add it to Linux.
1825	*
1826	* Will modify the state of the big block.
1827	*/
1828	static int virtio_mem_bbm_plug_and_add_bb(struct virtio_mem *vm,
1829	unsigned long bb_id)
1830	{
1831	int rc;
1832
1833	if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
1834	VIRTIO_MEM_BBM_BB_UNUSED))
1835	return -EINVAL;
1836
1837	rc = virtio_mem_bbm_plug_bb(vm, bb_id);
1838	if (rc)
1839	return rc;
1840	virtio_mem_bbm_set_bb_state(vm, bb_id, state: VIRTIO_MEM_BBM_BB_ADDED);
1841
1842	rc = virtio_mem_bbm_add_bb(vm, bb_id);
1843	if (rc) {
1844	if (!virtio_mem_bbm_unplug_bb(vm, bb_id))
1845	virtio_mem_bbm_set_bb_state(vm, bb_id,
1846	state: VIRTIO_MEM_BBM_BB_UNUSED);
1847	else
1848	/* Retry from the main loop. */
1849	virtio_mem_bbm_set_bb_state(vm, bb_id,
1850	state: VIRTIO_MEM_BBM_BB_PLUGGED);
1851	return rc;
1852	}
1853	return 0;
1854	}
1855
1856	/*
1857	* Prepare tracking data for the next big block.
1858	*/
1859	static int virtio_mem_bbm_prepare_next_bb(struct virtio_mem *vm,
1860	unsigned long *bb_id)
1861	{
1862	int rc;
1863
1864	if (vm->bbm.next_bb_id > vm->bbm.last_usable_bb_id)
1865	return -ENOSPC;
1866
1867	/* Resize the big block state array if required. */
1868	rc = virtio_mem_bbm_bb_states_prepare_next_bb(vm);
1869	if (rc)
1870	return rc;
1871
1872	vm->bbm.bb_count[VIRTIO_MEM_BBM_BB_UNUSED]++;
1873	*bb_id = vm->bbm.next_bb_id;
1874	vm->bbm.next_bb_id++;
1875	return 0;
1876	}
1877
1878	static int virtio_mem_bbm_plug_request(struct virtio_mem *vm, uint64_t diff)
1879	{
1880	uint64_t nb_bb = diff / vm->bbm.bb_size;
1881	unsigned long bb_id;
1882	int rc;
1883
1884	if (!nb_bb)
1885	return 0;
1886
1887	/* Try to plug and add unused big blocks */
1888	virtio_mem_bbm_for_each_bb(vm, bb_id, VIRTIO_MEM_BBM_BB_UNUSED) {
1889	if (!virtio_mem_could_add_memory(vm, size: vm->bbm.bb_size))
1890	return -ENOSPC;
1891
1892	rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
1893	if (!rc)
1894	nb_bb--;
1895	if (rc || !nb_bb)
1896	return rc;
1897	cond_resched();
1898	}
1899
1900	/* Try to prepare, plug and add new big blocks */
1901	while (nb_bb) {
1902	if (!virtio_mem_could_add_memory(vm, size: vm->bbm.bb_size))
1903	return -ENOSPC;
1904
1905	rc = virtio_mem_bbm_prepare_next_bb(vm, bb_id: &bb_id);
1906	if (rc)
1907	return rc;
1908	rc = virtio_mem_bbm_plug_and_add_bb(vm, bb_id);
1909	if (!rc)
1910	nb_bb--;
1911	if (rc)
1912	return rc;
1913	cond_resched();
1914	}
1915
1916	return 0;
1917	}
1918
1919	/*
1920	* Try to plug the requested amount of memory.
1921	*/
1922	static int virtio_mem_plug_request(struct virtio_mem *vm, uint64_t diff)
1923	{
1924	if (vm->in_sbm)
1925	return virtio_mem_sbm_plug_request(vm, diff);
1926	return virtio_mem_bbm_plug_request(vm, diff);
1927	}
1928
1929	/*
1930	* Unplug the desired number of plugged subblocks of an offline memory block.
1931	* Will fail if any subblock cannot get unplugged (instead of skipping it).
1932	*
1933	* Will modify the state of the memory block. Might temporarily drop the
1934	* hotplug_mutex.
1935	*
1936	* Note: Can fail after some subblocks were successfully unplugged.
1937	*/
1938	static int virtio_mem_sbm_unplug_any_sb_offline(struct virtio_mem *vm,
1939	unsigned long mb_id,
1940	uint64_t *nb_sb)
1941	{
1942	int rc;
1943
1944	rc = virtio_mem_sbm_unplug_any_sb_raw(vm, mb_id, nb_sb);
1945
1946	/* some subblocks might have been unplugged even on failure */
1947	if (!virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id: 0, count: vm->sbm.sbs_per_mb))
1948	virtio_mem_sbm_set_mb_state(vm, mb_id,
1949	state: VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL);
1950	if (rc)
1951	return rc;
1952
1953	if (virtio_mem_sbm_test_sb_unplugged(vm, mb_id, sb_id: 0, count: vm->sbm.sbs_per_mb)) {
1954	/*
1955	* Remove the block from Linux - this should never fail.
1956	* Hinder the block from getting onlined by marking it
1957	* unplugged. Temporarily drop the mutex, so
1958	* any pending GOING_ONLINE requests can be serviced/rejected.
1959	*/
1960	virtio_mem_sbm_set_mb_state(vm, mb_id,
1961	state: VIRTIO_MEM_SBM_MB_UNUSED);
1962
1963	mutex_unlock(lock: &vm->hotplug_mutex);
1964	rc = virtio_mem_sbm_remove_mb(vm, mb_id);
1965	BUG_ON(rc);
1966	mutex_lock(&vm->hotplug_mutex);
1967	}
1968	return 0;
1969	}
1970
1971	/*
1972	* Unplug the given plugged subblocks of an online memory block.
1973	*
1974	* Will modify the state of the memory block.
1975	*/
1976	static int virtio_mem_sbm_unplug_sb_online(struct virtio_mem *vm,
1977	unsigned long mb_id, int sb_id,
1978	int count)
1979	{
1980	const unsigned long nr_pages = PFN_DOWN(vm->sbm.sb_size) * count;
1981	const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
1982	unsigned long start_pfn;
1983	int rc;
1984
1985	start_pfn = PFN_DOWN(virtio_mem_mb_id_to_phys(mb_id) +
1986	sb_id * vm->sbm.sb_size);
1987
1988	rc = virtio_mem_fake_offline(vm, pfn: start_pfn, nr_pages);
1989	if (rc)
1990	return rc;
1991
1992	/* Try to unplug the allocated memory */
1993	rc = virtio_mem_sbm_unplug_sb(vm, mb_id, sb_id, count);
1994	if (rc) {
1995	/* Return the memory to the buddy. */
1996	virtio_mem_fake_online(pfn: start_pfn, nr_pages);
1997	return rc;
1998	}
1999
2000	switch (old_state) {
2001	case VIRTIO_MEM_SBM_MB_KERNEL:
2002	virtio_mem_sbm_set_mb_state(vm, mb_id,
2003	state: VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL);
2004	break;
2005	case VIRTIO_MEM_SBM_MB_MOVABLE:
2006	virtio_mem_sbm_set_mb_state(vm, mb_id,
2007	state: VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL);
2008	break;
2009	}
2010
2011	return 0;
2012	}
2013
2014	/*
2015	* Unplug the desired number of plugged subblocks of an online memory block.
2016	* Will skip subblock that are busy.
2017	*
2018	* Will modify the state of the memory block. Might temporarily drop the
2019	* hotplug_mutex.
2020	*
2021	* Note: Can fail after some subblocks were successfully unplugged. Can
2022	* return 0 even if subblocks were busy and could not get unplugged.
2023	*/
2024	static int virtio_mem_sbm_unplug_any_sb_online(struct virtio_mem *vm,
2025	unsigned long mb_id,
2026	uint64_t *nb_sb)
2027	{
2028	int rc, sb_id;
2029
2030	/* If possible, try to unplug the complete block in one shot. */
2031	if (*nb_sb >= vm->sbm.sbs_per_mb &&
2032	virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id: 0, count: vm->sbm.sbs_per_mb)) {
2033	rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, sb_id: 0,
2034	count: vm->sbm.sbs_per_mb);
2035	if (!rc) {
2036	*nb_sb -= vm->sbm.sbs_per_mb;
2037	goto unplugged;
2038	} else if (rc != -EBUSY)
2039	return rc;
2040	}
2041
2042	/* Fallback to single subblocks. */
2043	for (sb_id = vm->sbm.sbs_per_mb - 1; sb_id >= 0 && *nb_sb; sb_id--) {
2044	/* Find the next candidate subblock */
2045	while (sb_id >= 0 &&
2046	!virtio_mem_sbm_test_sb_plugged(vm, mb_id, sb_id, count: 1))
2047	sb_id--;
2048	if (sb_id < 0)
2049	break;
2050
2051	rc = virtio_mem_sbm_unplug_sb_online(vm, mb_id, sb_id, count: 1);
2052	if (rc == -EBUSY)
2053	continue;
2054	else if (rc)
2055	return rc;
2056	*nb_sb -= 1;
2057	}
2058
2059	unplugged:
2060	rc = virtio_mem_sbm_try_remove_unplugged_mb(vm, mb_id);
2061	if (rc)
2062	vm->sbm.have_unplugged_mb = 1;
2063	/* Ignore errors, this is not critical. We'll retry later. */
2064	return 0;
2065	}
2066
2067	/*
2068	* Unplug the desired number of plugged subblocks of a memory block that is
2069	* already added to Linux. Will skip subblock of online memory blocks that are
2070	* busy (by the OS). Will fail if any subblock that's not busy cannot get
2071	* unplugged.
2072	*
2073	* Will modify the state of the memory block. Might temporarily drop the
2074	* hotplug_mutex.
2075	*
2076	* Note: Can fail after some subblocks were successfully unplugged. Can
2077	* return 0 even if subblocks were busy and could not get unplugged.
2078	*/
2079	static int virtio_mem_sbm_unplug_any_sb(struct virtio_mem *vm,
2080	unsigned long mb_id,
2081	uint64_t *nb_sb)
2082	{
2083	const int old_state = virtio_mem_sbm_get_mb_state(vm, mb_id);
2084
2085	switch (old_state) {
2086	case VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL:
2087	case VIRTIO_MEM_SBM_MB_KERNEL:
2088	case VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL:
2089	case VIRTIO_MEM_SBM_MB_MOVABLE:
2090	return virtio_mem_sbm_unplug_any_sb_online(vm, mb_id, nb_sb);
2091	case VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL:
2092	case VIRTIO_MEM_SBM_MB_OFFLINE:
2093	return virtio_mem_sbm_unplug_any_sb_offline(vm, mb_id, nb_sb);
2094	}
2095	return -EINVAL;
2096	}
2097
2098	static int virtio_mem_sbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
2099	{
2100	const int mb_states[] = {
2101	VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL,
2102	VIRTIO_MEM_SBM_MB_OFFLINE,
2103	VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL,
2104	VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL,
2105	VIRTIO_MEM_SBM_MB_MOVABLE,
2106	VIRTIO_MEM_SBM_MB_KERNEL,
2107	};
2108	uint64_t nb_sb = diff / vm->sbm.sb_size;
2109	unsigned long mb_id;
2110	int rc, i;
2111
2112	if (!nb_sb)
2113	return 0;
2114
2115	/*
2116	* We'll drop the mutex a couple of times when it is safe to do so.
2117	* This might result in some blocks switching the state (online/offline)
2118	* and we could miss them in this run - we will retry again later.
2119	*/
2120	mutex_lock(&vm->hotplug_mutex);
2121
2122	/*
2123	* We try unplug from partially plugged blocks first, to try removing
2124	* whole memory blocks along with metadata. We prioritize ZONE_MOVABLE
2125	* as it's more reliable to unplug memory and remove whole memory
2126	* blocks, and we don't want to trigger a zone imbalances by
2127	* accidentially removing too much kernel memory.
2128	*/
2129	for (i = 0; i < ARRAY_SIZE(mb_states); i++) {
2130	virtio_mem_sbm_for_each_mb_rev(vm, mb_id, mb_states[i]) {
2131	rc = virtio_mem_sbm_unplug_any_sb(vm, mb_id, nb_sb: &nb_sb);
2132	if (rc || !nb_sb)
2133	goto out_unlock;
2134	mutex_unlock(lock: &vm->hotplug_mutex);
2135	cond_resched();
2136	mutex_lock(&vm->hotplug_mutex);
2137	}
2138	if (!unplug_online && i == 1) {
2139	mutex_unlock(lock: &vm->hotplug_mutex);
2140	return 0;
2141	}
2142	}
2143
2144	mutex_unlock(lock: &vm->hotplug_mutex);
2145	return nb_sb ? -EBUSY : 0;
2146	out_unlock:
2147	mutex_unlock(lock: &vm->hotplug_mutex);
2148	return rc;
2149	}
2150
2151	/*
2152	* Try to offline and remove a big block from Linux and unplug it. Will fail
2153	* with -EBUSY if some memory is busy and cannot get unplugged.
2154	*
2155	* Will modify the state of the memory block. Might temporarily drop the
2156	* hotplug_mutex.
2157	*/
2158	static int virtio_mem_bbm_offline_remove_and_unplug_bb(struct virtio_mem *vm,
2159	unsigned long bb_id)
2160	{
2161	const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2162	const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2163	unsigned long end_pfn = start_pfn + nr_pages;
2164	unsigned long pfn;
2165	struct page *page;
2166	int rc;
2167
2168	if (WARN_ON_ONCE(virtio_mem_bbm_get_bb_state(vm, bb_id) !=
2169	VIRTIO_MEM_BBM_BB_ADDED))
2170	return -EINVAL;
2171
2172	/*
2173	* Start by fake-offlining all memory. Once we marked the device
2174	* block as fake-offline, all newly onlined memory will
2175	* automatically be kept fake-offline. Protect from concurrent
2176	* onlining/offlining until we have a consistent state.
2177	*/
2178	mutex_lock(&vm->hotplug_mutex);
2179	virtio_mem_bbm_set_bb_state(vm, bb_id, state: VIRTIO_MEM_BBM_BB_FAKE_OFFLINE);
2180
2181	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2182	page = pfn_to_online_page(pfn);
2183	if (!page)
2184	continue;
2185
2186	rc = virtio_mem_fake_offline(vm, pfn, PAGES_PER_SECTION);
2187	if (rc) {
2188	end_pfn = pfn;
2189	goto rollback;
2190	}
2191	}
2192	mutex_unlock(lock: &vm->hotplug_mutex);
2193
2194	rc = virtio_mem_bbm_offline_and_remove_bb(vm, bb_id);
2195	if (rc) {
2196	mutex_lock(&vm->hotplug_mutex);
2197	goto rollback;
2198	}
2199
2200	rc = virtio_mem_bbm_unplug_bb(vm, bb_id);
2201	if (rc)
2202	virtio_mem_bbm_set_bb_state(vm, bb_id,
2203	state: VIRTIO_MEM_BBM_BB_PLUGGED);
2204	else
2205	virtio_mem_bbm_set_bb_state(vm, bb_id,
2206	state: VIRTIO_MEM_BBM_BB_UNUSED);
2207	return rc;
2208
2209	rollback:
2210	for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2211	page = pfn_to_online_page(pfn);
2212	if (!page)
2213	continue;
2214	virtio_mem_fake_online(pfn, PAGES_PER_SECTION);
2215	}
2216	virtio_mem_bbm_set_bb_state(vm, bb_id, state: VIRTIO_MEM_BBM_BB_ADDED);
2217	mutex_unlock(lock: &vm->hotplug_mutex);
2218	return rc;
2219	}
2220
2221	/*
2222	* Test if a big block is completely offline.
2223	*/
2224	static bool virtio_mem_bbm_bb_is_offline(struct virtio_mem *vm,
2225	unsigned long bb_id)
2226	{
2227	const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2228	const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2229	unsigned long pfn;
2230
2231	for (pfn = start_pfn; pfn < start_pfn + nr_pages;
2232	pfn += PAGES_PER_SECTION) {
2233	if (pfn_to_online_page(pfn))
2234	return false;
2235	}
2236
2237	return true;
2238	}
2239
2240	/*
2241	* Test if a big block is completely onlined to ZONE_MOVABLE (or offline).
2242	*/
2243	static bool virtio_mem_bbm_bb_is_movable(struct virtio_mem *vm,
2244	unsigned long bb_id)
2245	{
2246	const unsigned long start_pfn = PFN_DOWN(virtio_mem_bb_id_to_phys(vm, bb_id));
2247	const unsigned long nr_pages = PFN_DOWN(vm->bbm.bb_size);
2248	struct page *page;
2249	unsigned long pfn;
2250
2251	for (pfn = start_pfn; pfn < start_pfn + nr_pages;
2252	pfn += PAGES_PER_SECTION) {
2253	page = pfn_to_online_page(pfn);
2254	if (!page)
2255	continue;
2256	if (page_zonenum(page) != ZONE_MOVABLE)
2257	return false;
2258	}
2259
2260	return true;
2261	}
2262
2263	static int virtio_mem_bbm_unplug_request(struct virtio_mem *vm, uint64_t diff)
2264	{
2265	uint64_t nb_bb = diff / vm->bbm.bb_size;
2266	uint64_t bb_id;
2267	int rc, i;
2268
2269	if (!nb_bb)
2270	return 0;
2271
2272	/*
2273	* Try to unplug big blocks. Similar to SBM, start with offline
2274	* big blocks.
2275	*/
2276	for (i = 0; i < 3; i++) {
2277	virtio_mem_bbm_for_each_bb_rev(vm, bb_id, VIRTIO_MEM_BBM_BB_ADDED) {
2278	cond_resched();
2279
2280	/*
2281	* As we're holding no locks, these checks are racy,
2282	* but we don't care.
2283	*/
2284	if (i == 0 && !virtio_mem_bbm_bb_is_offline(vm, bb_id))
2285	continue;
2286	if (i == 1 && !virtio_mem_bbm_bb_is_movable(vm, bb_id))
2287	continue;
2288	rc = virtio_mem_bbm_offline_remove_and_unplug_bb(vm, bb_id);
2289	if (rc == -EBUSY)
2290	continue;
2291	if (!rc)
2292	nb_bb--;
2293	if (rc || !nb_bb)
2294	return rc;
2295	}
2296	if (i == 0 && !unplug_online)
2297	return 0;
2298	}
2299
2300	return nb_bb ? -EBUSY : 0;
2301	}
2302
2303	/*
2304	* Try to unplug the requested amount of memory.
2305	*/
2306	static int virtio_mem_unplug_request(struct virtio_mem *vm, uint64_t diff)
2307	{
2308	if (vm->in_sbm)
2309	return virtio_mem_sbm_unplug_request(vm, diff);
2310	return virtio_mem_bbm_unplug_request(vm, diff);
2311	}
2312
2313	/*
2314	* Try to unplug all blocks that couldn't be unplugged before, for example,
2315	* because the hypervisor was busy. Further, offline and remove any memory
2316	* blocks where we previously failed.
2317	*/
2318	static int virtio_mem_cleanup_pending_mb(struct virtio_mem *vm)
2319	{
2320	unsigned long id;
2321	int rc = 0;
2322
2323	if (!vm->in_sbm) {
2324	virtio_mem_bbm_for_each_bb(vm, id,
2325	VIRTIO_MEM_BBM_BB_PLUGGED) {
2326	rc = virtio_mem_bbm_unplug_bb(vm, bb_id: id);
2327	if (rc)
2328	return rc;
2329	virtio_mem_bbm_set_bb_state(vm, bb_id: id,
2330	state: VIRTIO_MEM_BBM_BB_UNUSED);
2331	}
2332	return 0;
2333	}
2334
2335	virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_PLUGGED) {
2336	rc = virtio_mem_sbm_unplug_mb(vm, mb_id: id);
2337	if (rc)
2338	return rc;
2339	virtio_mem_sbm_set_mb_state(vm, mb_id: id,
2340	state: VIRTIO_MEM_SBM_MB_UNUSED);
2341	}
2342
2343	if (!vm->sbm.have_unplugged_mb)
2344	return 0;
2345
2346	/*
2347	* Let's retry (offlining and) removing completely unplugged Linux
2348	* memory blocks.
2349	*/
2350	vm->sbm.have_unplugged_mb = false;
2351
2352	mutex_lock(&vm->hotplug_mutex);
2353	virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_MOVABLE_PARTIAL)
2354	rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, mb_id: id);
2355	virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_KERNEL_PARTIAL)
2356	rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, mb_id: id);
2357	virtio_mem_sbm_for_each_mb(vm, id, VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL)
2358	rc |= virtio_mem_sbm_try_remove_unplugged_mb(vm, mb_id: id);
2359	mutex_unlock(lock: &vm->hotplug_mutex);
2360
2361	if (rc)
2362	vm->sbm.have_unplugged_mb = true;
2363	/* Ignore errors, this is not critical. We'll retry later. */
2364	return 0;
2365	}
2366
2367	/*
2368	* Update all parts of the config that could have changed.
2369	*/
2370	static void virtio_mem_refresh_config(struct virtio_mem *vm)
2371	{
2372	const struct range pluggable_range = mhp_get_pluggable_range(need_mapping: true);
2373	uint64_t new_plugged_size, end_addr;
2374
2375	/* the plugged_size is just a reflection of what _we_ did previously */
2376	virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
2377	&new_plugged_size);
2378	if (WARN_ON_ONCE(new_plugged_size != vm->plugged_size))
2379	vm->plugged_size = new_plugged_size;
2380
2381	/* calculate the last usable memory block id */
2382	virtio_cread_le(vm->vdev, struct virtio_mem_config,
2383	usable_region_size, &vm->usable_region_size);
2384	end_addr = min(vm->addr + vm->usable_region_size - 1,
2385	pluggable_range.end);
2386
2387	if (vm->in_sbm) {
2388	vm->sbm.last_usable_mb_id = virtio_mem_phys_to_mb_id(addr: end_addr);
2389	if (!IS_ALIGNED(end_addr + 1, memory_block_size_bytes()))
2390	vm->sbm.last_usable_mb_id--;
2391	} else {
2392	vm->bbm.last_usable_bb_id = virtio_mem_phys_to_bb_id(vm,
2393	addr: end_addr);
2394	if (!IS_ALIGNED(end_addr + 1, vm->bbm.bb_size))
2395	vm->bbm.last_usable_bb_id--;
2396	}
2397	/*
2398	* If we cannot plug any of our device memory (e.g., nothing in the
2399	* usable region is addressable), the last usable memory block id will
2400	* be smaller than the first usable memory block id. We'll stop
2401	* attempting to add memory with -ENOSPC from our main loop.
2402	*/
2403
2404	/* see if there is a request to change the size */
2405	virtio_cread_le(vm->vdev, struct virtio_mem_config, requested_size,
2406	&vm->requested_size);
2407
2408	dev_info(&vm->vdev->dev, "plugged size: 0x%llx", vm->plugged_size);
2409	dev_info(&vm->vdev->dev, "requested size: 0x%llx", vm->requested_size);
2410	}
2411
2412	/*
2413	* Workqueue function for handling plug/unplug requests and config updates.
2414	*/
2415	static void virtio_mem_run_wq(struct work_struct *work)
2416	{
2417	struct virtio_mem *vm = container_of(work, struct virtio_mem, wq);
2418	uint64_t diff;
2419	int rc;
2420
2421	if (unlikely(vm->in_kdump)) {
2422	dev_warn_once(&vm->vdev->dev,
2423	"unexpected workqueue run in kdump kernel\n");
2424	return;
2425	}
2426
2427	hrtimer_cancel(timer: &vm->retry_timer);
2428
2429	if (vm->broken)
2430	return;
2431
2432	atomic_set(v: &vm->wq_active, i: 1);
2433	retry:
2434	rc = 0;
2435
2436	/* Make sure we start with a clean state if there are leftovers. */
2437	if (unlikely(vm->unplug_all_required))
2438	rc = virtio_mem_send_unplug_all_request(vm);
2439
2440	if (atomic_read(v: &vm->config_changed)) {
2441	atomic_set(v: &vm->config_changed, i: 0);
2442	virtio_mem_refresh_config(vm);
2443	}
2444
2445	/* Cleanup any leftovers from previous runs */
2446	if (!rc)
2447	rc = virtio_mem_cleanup_pending_mb(vm);
2448
2449	if (!rc && vm->requested_size != vm->plugged_size) {
2450	if (vm->requested_size > vm->plugged_size) {
2451	diff = vm->requested_size - vm->plugged_size;
2452	rc = virtio_mem_plug_request(vm, diff);
2453	} else {
2454	diff = vm->plugged_size - vm->requested_size;
2455	rc = virtio_mem_unplug_request(vm, diff);
2456	}
2457	}
2458
2459	/*
2460	* Keep retrying to offline and remove completely unplugged Linux
2461	* memory blocks.
2462	*/
2463	if (!rc && vm->in_sbm && vm->sbm.have_unplugged_mb)
2464	rc = -EBUSY;
2465
2466	switch (rc) {
2467	case 0:
2468	vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
2469	break;
2470	case -ENOSPC:
2471	/*
2472	* We cannot add any more memory (alignment, physical limit)
2473	* or we have too many offline memory blocks.
2474	*/
2475	break;
2476	case -ETXTBSY:
2477	/*
2478	* The hypervisor cannot process our request right now
2479	* (e.g., out of memory, migrating);
2480	*/
2481	case -EBUSY:
2482	/*
2483	* We cannot free up any memory to unplug it (all plugged memory
2484	* is busy).
2485	*/
2486	case -ENOMEM:
2487	/* Out of memory, try again later. */
2488	hrtimer_start(timer: &vm->retry_timer, tim: ms_to_ktime(ms: vm->retry_timer_ms),
2489	mode: HRTIMER_MODE_REL);
2490	break;
2491	case -EAGAIN:
2492	/* Retry immediately (e.g., the config changed). */
2493	goto retry;
2494	default:
2495	/* Unknown error, mark as broken */
2496	dev_err(&vm->vdev->dev,
2497	"unknown error, marking device broken: %d\n", rc);
2498	vm->broken = true;
2499	}
2500
2501	atomic_set(v: &vm->wq_active, i: 0);
2502	}
2503
2504	static enum hrtimer_restart virtio_mem_timer_expired(struct hrtimer *timer)
2505	{
2506	struct virtio_mem *vm = container_of(timer, struct virtio_mem,
2507	retry_timer);
2508
2509	virtio_mem_retry(vm);
2510	vm->retry_timer_ms = min_t(unsigned int, vm->retry_timer_ms * 2,
2511	VIRTIO_MEM_RETRY_TIMER_MAX_MS);
2512	return HRTIMER_NORESTART;
2513	}
2514
2515	static void virtio_mem_handle_response(struct virtqueue *vq)
2516	{
2517	struct virtio_mem *vm = vq->vdev->priv;
2518
2519	wake_up(&vm->host_resp);
2520	}
2521
2522	static int virtio_mem_init_vq(struct virtio_mem *vm)
2523	{
2524	struct virtqueue *vq;
2525
2526	vq = virtio_find_single_vq(vdev: vm->vdev, c: virtio_mem_handle_response,
2527	n: "guest-request");
2528	if (IS_ERR(ptr: vq))
2529	return PTR_ERR(ptr: vq);
2530	vm->vq = vq;
2531
2532	return 0;
2533	}
2534
2535	static int virtio_mem_init_hotplug(struct virtio_mem *vm)
2536	{
2537	const struct range pluggable_range = mhp_get_pluggable_range(need_mapping: true);
2538	uint64_t unit_pages, sb_size, addr;
2539	int rc;
2540
2541	/* bad device setup - warn only */
2542	if (!IS_ALIGNED(vm->addr, memory_block_size_bytes()))
2543	dev_warn(&vm->vdev->dev,
2544	"The alignment of the physical start address can make some memory unusable.\n");
2545	if (!IS_ALIGNED(vm->addr + vm->region_size, memory_block_size_bytes()))
2546	dev_warn(&vm->vdev->dev,
2547	"The alignment of the physical end address can make some memory unusable.\n");
2548	if (vm->addr < pluggable_range.start ||
2549	vm->addr + vm->region_size - 1 > pluggable_range.end)
2550	dev_warn(&vm->vdev->dev,
2551	"Some device memory is not addressable/pluggable. This can make some memory unusable.\n");
2552
2553	/* Prepare the offline threshold - make sure we can add two blocks. */
2554	vm->offline_threshold = max_t(uint64_t, 2 * memory_block_size_bytes(),
2555	VIRTIO_MEM_DEFAULT_OFFLINE_THRESHOLD);
2556
2557	/*
2558	* alloc_contig_range() works reliably with pageblock
2559	* granularity on ZONE_NORMAL, use pageblock_nr_pages.
2560	*/
2561	sb_size = PAGE_SIZE * pageblock_nr_pages;
2562	sb_size = max_t(uint64_t, vm->device_block_size, sb_size);
2563
2564	if (sb_size < memory_block_size_bytes() && !force_bbm) {
2565	/* SBM: At least two subblocks per Linux memory block. */
2566	vm->in_sbm = true;
2567	vm->sbm.sb_size = sb_size;
2568	vm->sbm.sbs_per_mb = memory_block_size_bytes() /
2569	vm->sbm.sb_size;
2570
2571	/* Round up to the next full memory block */
2572	addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
2573	memory_block_size_bytes() - 1;
2574	vm->sbm.first_mb_id = virtio_mem_phys_to_mb_id(addr);
2575	vm->sbm.next_mb_id = vm->sbm.first_mb_id;
2576	} else {
2577	/* BBM: At least one Linux memory block. */
2578	vm->bbm.bb_size = max_t(uint64_t, vm->device_block_size,
2579	memory_block_size_bytes());
2580
2581	if (bbm_block_size) {
2582	if (!is_power_of_2(n: bbm_block_size)) {
2583	dev_warn(&vm->vdev->dev,
2584	"bbm_block_size is not a power of 2");
2585	} else if (bbm_block_size < vm->bbm.bb_size) {
2586	dev_warn(&vm->vdev->dev,
2587	"bbm_block_size is too small");
2588	} else {
2589	vm->bbm.bb_size = bbm_block_size;
2590	}
2591	}
2592
2593	/* Round up to the next aligned big block */
2594	addr = max_t(uint64_t, vm->addr, pluggable_range.start) +
2595	vm->bbm.bb_size - 1;
2596	vm->bbm.first_bb_id = virtio_mem_phys_to_bb_id(vm, addr);
2597	vm->bbm.next_bb_id = vm->bbm.first_bb_id;
2598
2599	/* Make sure we can add two big blocks. */
2600	vm->offline_threshold = max_t(uint64_t, 2 * vm->bbm.bb_size,
2601	vm->offline_threshold);
2602	}
2603
2604	dev_info(&vm->vdev->dev, "memory block size: 0x%lx",
2605	memory_block_size_bytes());
2606	if (vm->in_sbm)
2607	dev_info(&vm->vdev->dev, "subblock size: 0x%llx",
2608	(unsigned long long)vm->sbm.sb_size);
2609	else
2610	dev_info(&vm->vdev->dev, "big block size: 0x%llx",
2611	(unsigned long long)vm->bbm.bb_size);
2612
2613	/* create the parent resource for all memory */
2614	rc = virtio_mem_create_resource(vm);
2615	if (rc)
2616	return rc;
2617
2618	/* use a single dynamic memory group to cover the whole memory device */
2619	if (vm->in_sbm)
2620	unit_pages = PHYS_PFN(memory_block_size_bytes());
2621	else
2622	unit_pages = PHYS_PFN(vm->bbm.bb_size);
2623	rc = memory_group_register_dynamic(nid: vm->nid, unit_pages);
2624	if (rc < 0)
2625	goto out_del_resource;
2626	vm->mgid = rc;
2627
2628	/*
2629	* If we still have memory plugged, we have to unplug all memory first.
2630	* Registering our parent resource makes sure that this memory isn't
2631	* actually in use (e.g., trying to reload the driver).
2632	*/
2633	if (vm->plugged_size) {
2634	vm->unplug_all_required = true;
2635	dev_info(&vm->vdev->dev, "unplugging all memory is required\n");
2636	}
2637
2638	/* register callbacks */
2639	vm->memory_notifier.notifier_call = virtio_mem_memory_notifier_cb;
2640	rc = register_memory_notifier(nb: &vm->memory_notifier);
2641	if (rc)
2642	goto out_unreg_group;
2643	/* Block hibernation as early as possible. */
2644	vm->pm_notifier.priority = INT_MAX;
2645	vm->pm_notifier.notifier_call = virtio_mem_pm_notifier_cb;
2646	rc = register_pm_notifier(nb: &vm->pm_notifier);
2647	if (rc)
2648	goto out_unreg_mem;
2649	rc = register_virtio_mem_device(vm);
2650	if (rc)
2651	goto out_unreg_pm;
2652
2653	virtio_device_ready(dev: vm->vdev);
2654	return 0;
2655	out_unreg_pm:
2656	unregister_pm_notifier(nb: &vm->pm_notifier);
2657	out_unreg_mem:
2658	unregister_memory_notifier(nb: &vm->memory_notifier);
2659	out_unreg_group:
2660	memory_group_unregister(mgid: vm->mgid);
2661	out_del_resource:
2662	virtio_mem_delete_resource(vm);
2663	return rc;
2664	}
2665
2666	#ifdef CONFIG_PROC_VMCORE
2667	static int virtio_mem_send_state_request(struct virtio_mem *vm, uint64_t addr,
2668	uint64_t size)
2669	{
2670	const uint64_t nb_vm_blocks = size / vm->device_block_size;
2671	const struct virtio_mem_req req = {
2672	.type = cpu_to_virtio16(vdev: vm->vdev, VIRTIO_MEM_REQ_STATE),
2673	.u.state.addr = cpu_to_virtio64(vdev: vm->vdev, val: addr),
2674	.u.state.nb_blocks = cpu_to_virtio16(vdev: vm->vdev, val: nb_vm_blocks),
2675	};
2676	int rc = -ENOMEM;
2677
2678	dev_dbg(&vm->vdev->dev, "requesting state: 0x%llx - 0x%llx\n", addr,
2679	addr + size - 1);
2680
2681	switch (virtio_mem_send_request(vm, req: &req)) {
2682	case VIRTIO_MEM_RESP_ACK:
2683	return virtio16_to_cpu(vdev: vm->vdev, val: vm->resp.u.state.state);
2684	case VIRTIO_MEM_RESP_ERROR:
2685	rc = -EINVAL;
2686	break;
2687	default:
2688	break;
2689	}
2690
2691	dev_dbg(&vm->vdev->dev, "requesting state failed: %d\n", rc);
2692	return rc;
2693	}
2694
2695	static bool virtio_mem_vmcore_pfn_is_ram(struct vmcore_cb *cb,
2696	unsigned long pfn)
2697	{
2698	struct virtio_mem *vm = container_of(cb, struct virtio_mem,
2699	vmcore_cb);
2700	uint64_t addr = PFN_PHYS(pfn);
2701	bool is_ram;
2702	int rc;
2703
2704	if (!virtio_mem_contains_range(vm, start: addr, PAGE_SIZE))
2705	return true;
2706	if (!vm->plugged_size)
2707	return false;
2708
2709	/*
2710	* We have to serialize device requests and access to the information
2711	* about the block queried last.
2712	*/
2713	mutex_lock(&vm->hotplug_mutex);
2714
2715	addr = ALIGN_DOWN(addr, vm->device_block_size);
2716	if (addr != vm->last_block_addr) {
2717	rc = virtio_mem_send_state_request(vm, addr,
2718	size: vm->device_block_size);
2719	/* On any kind of error, we're going to signal !ram. */
2720	if (rc == VIRTIO_MEM_STATE_PLUGGED)
2721	vm->last_block_plugged = true;
2722	else
2723	vm->last_block_plugged = false;
2724	vm->last_block_addr = addr;
2725	}
2726
2727	is_ram = vm->last_block_plugged;
2728	mutex_unlock(lock: &vm->hotplug_mutex);
2729	return is_ram;
2730	}
2731
2732	#ifdef CONFIG_PROC_VMCORE_DEVICE_RAM
2733	static int virtio_mem_vmcore_add_device_ram(struct virtio_mem *vm,
2734	struct list_head *list, uint64_t start, uint64_t end)
2735	{
2736	int rc;
2737
2738	rc = vmcore_alloc_add_range(list, start, end - start);
2739	if (rc)
2740	dev_err(&vm->vdev->dev,
2741	"Error adding device RAM range: %d\n", rc);
2742	return rc;
2743	}
2744
2745	static int virtio_mem_vmcore_get_device_ram(struct vmcore_cb *cb,
2746	struct list_head *list)
2747	{
2748	struct virtio_mem *vm = container_of(cb, struct virtio_mem,
2749	vmcore_cb);
2750	const uint64_t device_start = vm->addr;
2751	const uint64_t device_end = vm->addr + vm->usable_region_size;
2752	uint64_t chunk_size, cur_start, cur_end, plugged_range_start = 0;
2753	LIST_HEAD(tmp_list);
2754	int rc;
2755
2756	if (!vm->plugged_size)
2757	return 0;
2758
2759	/* Process memory sections, unless the device block size is bigger. */
2760	chunk_size = max_t(uint64_t, PFN_PHYS(PAGES_PER_SECTION),
2761	vm->device_block_size);
2762
2763	mutex_lock(&vm->hotplug_mutex);
2764
2765	/*
2766	* We process larger chunks and indicate the complete chunk if any
2767	* block in there is plugged. This reduces the number of pfn_is_ram()
2768	* callbacks and mimic what is effectively being done when the old
2769	* kernel would add complete memory sections/blocks to the elfcore hdr.
2770	*/
2771	cur_start = device_start;
2772	for (cur_start = device_start; cur_start < device_end; cur_start = cur_end) {
2773	cur_end = ALIGN_DOWN(cur_start + chunk_size, chunk_size);
2774	cur_end = min_t(uint64_t, cur_end, device_end);
2775
2776	rc = virtio_mem_send_state_request(vm, cur_start,
2777	cur_end - cur_start);
2778
2779	if (rc < 0) {
2780	dev_err(&vm->vdev->dev,
2781	"Error querying block states: %d\n", rc);
2782	goto out;
2783	} else if (rc != VIRTIO_MEM_STATE_UNPLUGGED) {
2784	/* Merge ranges with plugged memory. */
2785	if (!plugged_range_start)
2786	plugged_range_start = cur_start;
2787	continue;
2788	}
2789
2790	/* Flush any plugged range. */
2791	if (plugged_range_start) {
2792	rc = virtio_mem_vmcore_add_device_ram(vm, &tmp_list,
2793	plugged_range_start,
2794	cur_start);
2795	if (rc)
2796	goto out;
2797	plugged_range_start = 0;
2798	}
2799	}
2800
2801	/* Flush any plugged range. */
2802	if (plugged_range_start)
2803	rc = virtio_mem_vmcore_add_device_ram(vm, &tmp_list,
2804	plugged_range_start,
2805	cur_start);
2806	out:
2807	mutex_unlock(&vm->hotplug_mutex);
2808	if (rc < 0) {
2809	vmcore_free_ranges(&tmp_list);
2810	return rc;
2811	}
2812	list_splice_tail(&tmp_list, list);
2813	return 0;
2814	}
2815	#endif /* CONFIG_PROC_VMCORE_DEVICE_RAM */
2816	#endif /* CONFIG_PROC_VMCORE */
2817
2818	static int virtio_mem_init_kdump(struct virtio_mem *vm)
2819	{
2820	/* We must be prepared to receive a callback immediately. */
2821	virtio_device_ready(dev: vm->vdev);
2822	#ifdef CONFIG_PROC_VMCORE
2823	dev_info(&vm->vdev->dev, "memory hot(un)plug disabled in kdump kernel\n");
2824	vm->vmcore_cb.pfn_is_ram = virtio_mem_vmcore_pfn_is_ram;
2825	#ifdef CONFIG_PROC_VMCORE_DEVICE_RAM
2826	vm->vmcore_cb.get_device_ram = virtio_mem_vmcore_get_device_ram;
2827	#endif /* CONFIG_PROC_VMCORE_DEVICE_RAM */
2828	register_vmcore_cb(cb: &vm->vmcore_cb);
2829	return 0;
2830	#else /* CONFIG_PROC_VMCORE */
2831	dev_warn(&vm->vdev->dev, "disabled in kdump kernel without vmcore\n");
2832	return -EBUSY;
2833	#endif /* CONFIG_PROC_VMCORE */
2834	}
2835
2836	static int virtio_mem_init(struct virtio_mem *vm)
2837	{
2838	uint16_t node_id;
2839
2840	if (!vm->vdev->config->get) {
2841	dev_err(&vm->vdev->dev, "config access disabled\n");
2842	return -EINVAL;
2843	}
2844
2845	/* Fetch all properties that can't change. */
2846	virtio_cread_le(vm->vdev, struct virtio_mem_config, plugged_size,
2847	&vm->plugged_size);
2848	virtio_cread_le(vm->vdev, struct virtio_mem_config, block_size,
2849	&vm->device_block_size);
2850	virtio_cread_le(vm->vdev, struct virtio_mem_config, node_id,
2851	&node_id);
2852	vm->nid = virtio_mem_translate_node_id(vm, node_id);
2853	virtio_cread_le(vm->vdev, struct virtio_mem_config, addr, &vm->addr);
2854	virtio_cread_le(vm->vdev, struct virtio_mem_config, region_size,
2855	&vm->region_size);
2856	virtio_cread_le(vm->vdev, struct virtio_mem_config, usable_region_size,
2857	&vm->usable_region_size);
2858
2859	/* Determine the nid for the device based on the lowest address. */
2860	if (vm->nid == NUMA_NO_NODE)
2861	vm->nid = memory_add_physaddr_to_nid(start: vm->addr);
2862
2863	dev_info(&vm->vdev->dev, "start address: 0x%llx", vm->addr);
2864	dev_info(&vm->vdev->dev, "region size: 0x%llx", vm->region_size);
2865	dev_info(&vm->vdev->dev, "device block size: 0x%llx",
2866	(unsigned long long)vm->device_block_size);
2867	if (vm->nid != NUMA_NO_NODE && IS_ENABLED(CONFIG_NUMA))
2868	dev_info(&vm->vdev->dev, "nid: %d", vm->nid);
2869
2870	/*
2871	* We don't want to (un)plug or reuse any memory when in kdump. The
2872	* memory is still accessible (but not exposed to Linux).
2873	*/
2874	if (vm->in_kdump)
2875	return virtio_mem_init_kdump(vm);
2876	return virtio_mem_init_hotplug(vm);
2877	}
2878
2879	static int virtio_mem_create_resource(struct virtio_mem *vm)
2880	{
2881	/*
2882	* When force-unloading the driver and removing the device, we
2883	* could have a garbage pointer. Duplicate the string.
2884	*/
2885	const char *name = kstrdup(s: dev_name(dev: &vm->vdev->dev), GFP_KERNEL);
2886
2887	if (!name)
2888	return -ENOMEM;
2889
2890	/* Disallow mapping device memory via /dev/mem completely. */
2891	vm->parent_resource = __request_mem_region(vm->addr, vm->region_size,
2892	name, IORESOURCE_SYSTEM_RAM |
2893	IORESOURCE_EXCLUSIVE);
2894	if (!vm->parent_resource) {
2895	kfree(objp: name);
2896	dev_warn(&vm->vdev->dev, "could not reserve device region\n");
2897	dev_info(&vm->vdev->dev,
2898	"reloading the driver is not supported\n");
2899	return -EBUSY;
2900	}
2901
2902	/* The memory is not actually busy - make add_memory() work. */
2903	vm->parent_resource->flags &= ~IORESOURCE_BUSY;
2904	return 0;
2905	}
2906
2907	static void virtio_mem_delete_resource(struct virtio_mem *vm)
2908	{
2909	const char *name;
2910
2911	if (!vm->parent_resource)
2912	return;
2913
2914	name = vm->parent_resource->name;
2915	release_resource(new: vm->parent_resource);
2916	kfree(objp: vm->parent_resource);
2917	kfree(objp: name);
2918	vm->parent_resource = NULL;
2919	}
2920
2921	static int virtio_mem_range_has_system_ram(struct resource *res, void *arg)
2922	{
2923	return 1;
2924	}
2925
2926	static bool virtio_mem_has_memory_added(struct virtio_mem *vm)
2927	{
2928	const unsigned long flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
2929
2930	return walk_iomem_res_desc(desc: IORES_DESC_NONE, flags, start: vm->addr,
2931	end: vm->addr + vm->region_size, NULL,
2932	func: virtio_mem_range_has_system_ram) == 1;
2933	}
2934
2935	static int virtio_mem_probe(struct virtio_device *vdev)
2936	{
2937	struct virtio_mem *vm;
2938	int rc;
2939
2940	BUILD_BUG_ON(sizeof(struct virtio_mem_req) != 24);
2941	BUILD_BUG_ON(sizeof(struct virtio_mem_resp) != 10);
2942
2943	vdev->priv = vm = kzalloc(sizeof(*vm), GFP_KERNEL);
2944	if (!vm)
2945	return -ENOMEM;
2946
2947	init_waitqueue_head(&vm->host_resp);
2948	vm->vdev = vdev;
2949	INIT_WORK(&vm->wq, virtio_mem_run_wq);
2950	mutex_init(&vm->hotplug_mutex);
2951	INIT_LIST_HEAD(list: &vm->next);
2952	spin_lock_init(&vm->removal_lock);
2953	hrtimer_setup(timer: &vm->retry_timer, function: virtio_mem_timer_expired, CLOCK_MONOTONIC,
2954	mode: HRTIMER_MODE_REL);
2955	vm->retry_timer_ms = VIRTIO_MEM_RETRY_TIMER_MIN_MS;
2956	vm->in_kdump = is_kdump_kernel();
2957
2958	/* register the virtqueue */
2959	rc = virtio_mem_init_vq(vm);
2960	if (rc)
2961	goto out_free_vm;
2962
2963	/* initialize the device by querying the config */
2964	rc = virtio_mem_init(vm);
2965	if (rc)
2966	goto out_del_vq;
2967
2968	/* trigger a config update to start processing the requested_size */
2969	if (!vm->in_kdump) {
2970	atomic_set(v: &vm->config_changed, i: 1);
2971	queue_work(wq: system_freezable_wq, work: &vm->wq);
2972	}
2973
2974	return 0;
2975	out_del_vq:
2976	vdev->config->del_vqs(vdev);
2977	out_free_vm:
2978	kfree(objp: vm);
2979	vdev->priv = NULL;
2980
2981	return rc;
2982	}
2983
2984	static void virtio_mem_deinit_hotplug(struct virtio_mem *vm)
2985	{
2986	unsigned long mb_id;
2987	int rc;
2988
2989	/*
2990	* Make sure the workqueue won't be triggered anymore and no memory
2991	* blocks can be onlined/offlined until we're finished here.
2992	*/
2993	mutex_lock(&vm->hotplug_mutex);
2994	spin_lock_irq(lock: &vm->removal_lock);
2995	vm->removing = true;
2996	spin_unlock_irq(lock: &vm->removal_lock);
2997	mutex_unlock(lock: &vm->hotplug_mutex);
2998
2999	/* wait until the workqueue stopped */
3000	cancel_work_sync(work: &vm->wq);
3001	hrtimer_cancel(timer: &vm->retry_timer);
3002
3003	if (vm->in_sbm) {
3004	/*
3005	* After we unregistered our callbacks, user space can online
3006	* partially plugged offline blocks. Make sure to remove them.
3007	*/
3008	virtio_mem_sbm_for_each_mb(vm, mb_id,
3009	VIRTIO_MEM_SBM_MB_OFFLINE_PARTIAL) {
3010	rc = virtio_mem_sbm_remove_mb(vm, mb_id);
3011	BUG_ON(rc);
3012	virtio_mem_sbm_set_mb_state(vm, mb_id,
3013	state: VIRTIO_MEM_SBM_MB_UNUSED);
3014	}
3015	/*
3016	* After we unregistered our callbacks, user space can no longer
3017	* offline partially plugged online memory blocks. No need to
3018	* worry about them.
3019	*/
3020	}
3021
3022	/* unregister callbacks */
3023	unregister_virtio_mem_device(vm);
3024	unregister_pm_notifier(nb: &vm->pm_notifier);
3025	unregister_memory_notifier(nb: &vm->memory_notifier);
3026
3027	/*
3028	* There is no way we could reliably remove all memory we have added to
3029	* the system. And there is no way to stop the driver/device from going
3030	* away. Warn at least.
3031	*/
3032	if (virtio_mem_has_memory_added(vm)) {
3033	dev_warn(&vm->vdev->dev,
3034	"device still has system memory added\n");
3035	} else {
3036	virtio_mem_delete_resource(vm);
3037	kfree_const(x: vm->resource_name);
3038	memory_group_unregister(mgid: vm->mgid);
3039	}
3040
3041	/* remove all tracking data - no locking needed */
3042	if (vm->in_sbm) {
3043	vfree(addr: vm->sbm.mb_states);
3044	vfree(addr: vm->sbm.sb_states);
3045	} else {
3046	vfree(addr: vm->bbm.bb_states);
3047	}
3048	}
3049
3050	static void virtio_mem_deinit_kdump(struct virtio_mem *vm)
3051	{
3052	#ifdef CONFIG_PROC_VMCORE
3053	unregister_vmcore_cb(cb: &vm->vmcore_cb);
3054	#endif /* CONFIG_PROC_VMCORE */
3055	}
3056
3057	static void virtio_mem_remove(struct virtio_device *vdev)
3058	{
3059	struct virtio_mem *vm = vdev->priv;
3060
3061	if (vm->in_kdump)
3062	virtio_mem_deinit_kdump(vm);
3063	else
3064	virtio_mem_deinit_hotplug(vm);
3065
3066	/* reset the device and cleanup the queues */
3067	virtio_reset_device(dev: vdev);
3068	vdev->config->del_vqs(vdev);
3069
3070	kfree(objp: vm);
3071	vdev->priv = NULL;
3072	}
3073
3074	static void virtio_mem_config_changed(struct virtio_device *vdev)
3075	{
3076	struct virtio_mem *vm = vdev->priv;
3077
3078	if (unlikely(vm->in_kdump))
3079	return;
3080
3081	atomic_set(v: &vm->config_changed, i: 1);
3082	virtio_mem_retry(vm);
3083	}
3084
3085	#ifdef CONFIG_PM_SLEEP
3086	static int virtio_mem_freeze(struct virtio_device *vdev)
3087	{
3088	struct virtio_mem *vm = vdev->priv;
3089
3090	/*
3091	* We block hibernation using the PM notifier completely. The workqueue
3092	* is already frozen by the PM core at this point, so we simply
3093	* reset the device and cleanup the queues.
3094	*/
3095	if (pm_suspend_target_state != PM_SUSPEND_TO_IDLE &&
3096	vm->plugged_size &&
3097	!virtio_has_feature(vdev: vm->vdev, VIRTIO_MEM_F_PERSISTENT_SUSPEND)) {
3098	dev_err(&vm->vdev->dev,
3099	"suspending with plugged memory is not supported\n");
3100	return -EPERM;
3101	}
3102
3103	virtio_reset_device(dev: vdev);
3104	vdev->config->del_vqs(vdev);
3105	vm->vq = NULL;
3106	return 0;
3107	}
3108
3109	static int virtio_mem_restore(struct virtio_device *vdev)
3110	{
3111	struct virtio_mem *vm = vdev->priv;
3112	int ret;
3113
3114	ret = virtio_mem_init_vq(vm);
3115	if (ret)
3116	return ret;
3117	virtio_device_ready(dev: vdev);
3118
3119	/* Let's check if anything changed. */
3120	virtio_mem_config_changed(vdev);
3121	return 0;
3122	}
3123	#endif
3124
3125	static unsigned int virtio_mem_features[] = {
3126	#if defined(CONFIG_NUMA) && defined(CONFIG_ACPI_NUMA)
3127	VIRTIO_MEM_F_ACPI_PXM,
3128	#endif
3129	VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE,
3130	VIRTIO_MEM_F_PERSISTENT_SUSPEND,
3131	};
3132
3133	static const struct virtio_device_id virtio_mem_id_table[] = {
3134	{ VIRTIO_ID_MEM, VIRTIO_DEV_ANY_ID },
3135	{ 0 },
3136	};
3137
3138	static struct virtio_driver virtio_mem_driver = {
3139	.feature_table = virtio_mem_features,
3140	.feature_table_size = ARRAY_SIZE(virtio_mem_features),
3141	.driver.name = KBUILD_MODNAME,
3142	.id_table = virtio_mem_id_table,
3143	.probe = virtio_mem_probe,
3144	.remove = virtio_mem_remove,
3145	.config_changed = virtio_mem_config_changed,
3146	#ifdef CONFIG_PM_SLEEP
3147	.freeze = virtio_mem_freeze,
3148	.restore = virtio_mem_restore,
3149	#endif
3150	};
3151
3152	module_virtio_driver(virtio_mem_driver);
3153	MODULE_DEVICE_TABLE(virtio, virtio_mem_id_table);
3154	MODULE_AUTHOR("David Hildenbrand <david@redhat.com>");
3155	MODULE_DESCRIPTION("Virtio-mem driver");
3156	MODULE_LICENSE("GPL");
3157