xref: /openbmc/qemu/hw/virtio/virtio-mem.c (revision 4f7689f0)
1 /*
2  * Virtio MEM device
3  *
4  * Copyright (C) 2020 Red Hat, Inc.
5  *
6  * Authors:
7  *  David Hildenbrand <david@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.
10  * See the COPYING file in the top-level directory.
11  */
12 
13 #include "qemu/osdep.h"
14 #include "qemu/iov.h"
15 #include "qemu/cutils.h"
16 #include "qemu/error-report.h"
17 #include "qemu/units.h"
18 #include "sysemu/numa.h"
19 #include "sysemu/sysemu.h"
20 #include "sysemu/reset.h"
21 #include "sysemu/runstate.h"
22 #include "hw/virtio/virtio.h"
23 #include "hw/virtio/virtio-bus.h"
24 #include "hw/virtio/virtio-mem.h"
25 #include "qapi/error.h"
26 #include "qapi/visitor.h"
27 #include "exec/ram_addr.h"
28 #include "migration/misc.h"
29 #include "hw/boards.h"
30 #include "hw/qdev-properties.h"
31 #include CONFIG_DEVICES
32 #include "trace.h"
33 
34 static const VMStateDescription vmstate_virtio_mem_device_early;
35 
36 /*
37  * We only had legacy x86 guests that did not support
38  * VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE. Other targets don't have legacy guests.
39  */
40 #if defined(TARGET_X86_64) || defined(TARGET_I386)
41 #define VIRTIO_MEM_HAS_LEGACY_GUESTS
42 #endif
43 
44 /*
45  * Let's not allow blocks smaller than 1 MiB, for example, to keep the tracking
46  * bitmap small.
47  */
48 #define VIRTIO_MEM_MIN_BLOCK_SIZE ((uint32_t)(1 * MiB))
49 
50 static uint32_t virtio_mem_default_thp_size(void)
51 {
52     uint32_t default_thp_size = VIRTIO_MEM_MIN_BLOCK_SIZE;
53 
54 #if defined(__x86_64__) || defined(__arm__) || defined(__powerpc64__)
55     default_thp_size = 2 * MiB;
56 #elif defined(__aarch64__)
57     if (qemu_real_host_page_size() == 4 * KiB) {
58         default_thp_size = 2 * MiB;
59     } else if (qemu_real_host_page_size() == 16 * KiB) {
60         default_thp_size = 32 * MiB;
61     } else if (qemu_real_host_page_size() == 64 * KiB) {
62         default_thp_size = 512 * MiB;
63     }
64 #endif
65 
66     return default_thp_size;
67 }
68 
69 /*
70  * We want to have a reasonable default block size such that
71  * 1. We avoid splitting THPs when unplugging memory, which degrades
72  *    performance.
73  * 2. We avoid placing THPs for plugged blocks that also cover unplugged
74  *    blocks.
75  *
76  * The actual THP size might differ between Linux kernels, so we try to probe
77  * it. In the future (if we ever run into issues regarding 2.), we might want
78  * to disable THP in case we fail to properly probe the THP size, or if the
79  * block size is configured smaller than the THP size.
80  */
81 static uint32_t thp_size;
82 
83 #define HPAGE_PMD_SIZE_PATH "/sys/kernel/mm/transparent_hugepage/hpage_pmd_size"
84 static uint32_t virtio_mem_thp_size(void)
85 {
86     gchar *content = NULL;
87     const char *endptr;
88     uint64_t tmp;
89 
90     if (thp_size) {
91         return thp_size;
92     }
93 
94     /*
95      * Try to probe the actual THP size, fallback to (sane but eventually
96      * incorrect) default sizes.
97      */
98     if (g_file_get_contents(HPAGE_PMD_SIZE_PATH, &content, NULL, NULL) &&
99         !qemu_strtou64(content, &endptr, 0, &tmp) &&
100         (!endptr || *endptr == '\n')) {
101         /* Sanity-check the value and fallback to something reasonable. */
102         if (!tmp || !is_power_of_2(tmp)) {
103             warn_report("Read unsupported THP size: %" PRIx64, tmp);
104         } else {
105             thp_size = tmp;
106         }
107     }
108 
109     if (!thp_size) {
110         thp_size = virtio_mem_default_thp_size();
111         warn_report("Could not detect THP size, falling back to %" PRIx64
112                     "  MiB.", thp_size / MiB);
113     }
114 
115     g_free(content);
116     return thp_size;
117 }
118 
119 static uint64_t virtio_mem_default_block_size(RAMBlock *rb)
120 {
121     const uint64_t page_size = qemu_ram_pagesize(rb);
122 
123     /* We can have hugetlbfs with a page size smaller than the THP size. */
124     if (page_size == qemu_real_host_page_size()) {
125         return MAX(page_size, virtio_mem_thp_size());
126     }
127     return MAX(page_size, VIRTIO_MEM_MIN_BLOCK_SIZE);
128 }
129 
130 #if defined(VIRTIO_MEM_HAS_LEGACY_GUESTS)
131 static bool virtio_mem_has_shared_zeropage(RAMBlock *rb)
132 {
133     /*
134      * We only have a guaranteed shared zeropage on ordinary MAP_PRIVATE
135      * anonymous RAM. In any other case, reading unplugged *can* populate a
136      * fresh page, consuming actual memory.
137      */
138     return !qemu_ram_is_shared(rb) && qemu_ram_get_fd(rb) < 0 &&
139            qemu_ram_pagesize(rb) == qemu_real_host_page_size();
140 }
141 #endif /* VIRTIO_MEM_HAS_LEGACY_GUESTS */
142 
143 /*
144  * Size the usable region bigger than the requested size if possible. Esp.
145  * Linux guests will only add (aligned) memory blocks in case they fully
146  * fit into the usable region, but plug+online only a subset of the pages.
147  * The memory block size corresponds mostly to the section size.
148  *
149  * This allows e.g., to add 20MB with a section size of 128MB on x86_64, and
150  * a section size of 512MB on arm64 (as long as the start address is properly
151  * aligned, similar to ordinary DIMMs).
152  *
153  * We can change this at any time and maybe even make it configurable if
154  * necessary (as the section size can change). But it's more likely that the
155  * section size will rather get smaller and not bigger over time.
156  */
157 #if defined(TARGET_X86_64) || defined(TARGET_I386)
158 #define VIRTIO_MEM_USABLE_EXTENT (2 * (128 * MiB))
159 #elif defined(TARGET_ARM)
160 #define VIRTIO_MEM_USABLE_EXTENT (2 * (512 * MiB))
161 #else
162 #error VIRTIO_MEM_USABLE_EXTENT not defined
163 #endif
164 
165 static bool virtio_mem_is_busy(void)
166 {
167     /*
168      * Postcopy cannot handle concurrent discards and we don't want to migrate
169      * pages on-demand with stale content when plugging new blocks.
170      *
171      * For precopy, we don't want unplugged blocks in our migration stream, and
172      * when plugging new blocks, the page content might differ between source
173      * and destination (observable by the guest when not initializing pages
174      * after plugging them) until we're running on the destination (as we didn't
175      * migrate these blocks when they were unplugged).
176      */
177     return migration_in_incoming_postcopy() || !migration_is_idle();
178 }
179 
180 typedef int (*virtio_mem_range_cb)(const VirtIOMEM *vmem, void *arg,
181                                    uint64_t offset, uint64_t size);
182 
183 static int virtio_mem_for_each_unplugged_range(const VirtIOMEM *vmem, void *arg,
184                                                virtio_mem_range_cb cb)
185 {
186     unsigned long first_zero_bit, last_zero_bit;
187     uint64_t offset, size;
188     int ret = 0;
189 
190     first_zero_bit = find_first_zero_bit(vmem->bitmap, vmem->bitmap_size);
191     while (first_zero_bit < vmem->bitmap_size) {
192         offset = first_zero_bit * vmem->block_size;
193         last_zero_bit = find_next_bit(vmem->bitmap, vmem->bitmap_size,
194                                       first_zero_bit + 1) - 1;
195         size = (last_zero_bit - first_zero_bit + 1) * vmem->block_size;
196 
197         ret = cb(vmem, arg, offset, size);
198         if (ret) {
199             break;
200         }
201         first_zero_bit = find_next_zero_bit(vmem->bitmap, vmem->bitmap_size,
202                                             last_zero_bit + 2);
203     }
204     return ret;
205 }
206 
207 static int virtio_mem_for_each_plugged_range(const VirtIOMEM *vmem, void *arg,
208                                              virtio_mem_range_cb cb)
209 {
210     unsigned long first_bit, last_bit;
211     uint64_t offset, size;
212     int ret = 0;
213 
214     first_bit = find_first_bit(vmem->bitmap, vmem->bitmap_size);
215     while (first_bit < vmem->bitmap_size) {
216         offset = first_bit * vmem->block_size;
217         last_bit = find_next_zero_bit(vmem->bitmap, vmem->bitmap_size,
218                                       first_bit + 1) - 1;
219         size = (last_bit - first_bit + 1) * vmem->block_size;
220 
221         ret = cb(vmem, arg, offset, size);
222         if (ret) {
223             break;
224         }
225         first_bit = find_next_bit(vmem->bitmap, vmem->bitmap_size,
226                                   last_bit + 2);
227     }
228     return ret;
229 }
230 
231 /*
232  * Adjust the memory section to cover the intersection with the given range.
233  *
234  * Returns false if the intersection is empty, otherwise returns true.
235  */
236 static bool virtio_mem_intersect_memory_section(MemoryRegionSection *s,
237                                                 uint64_t offset, uint64_t size)
238 {
239     uint64_t start = MAX(s->offset_within_region, offset);
240     uint64_t end = MIN(s->offset_within_region + int128_get64(s->size),
241                        offset + size);
242 
243     if (end <= start) {
244         return false;
245     }
246 
247     s->offset_within_address_space += start - s->offset_within_region;
248     s->offset_within_region = start;
249     s->size = int128_make64(end - start);
250     return true;
251 }
252 
253 typedef int (*virtio_mem_section_cb)(MemoryRegionSection *s, void *arg);
254 
255 static int virtio_mem_for_each_plugged_section(const VirtIOMEM *vmem,
256                                                MemoryRegionSection *s,
257                                                void *arg,
258                                                virtio_mem_section_cb cb)
259 {
260     unsigned long first_bit, last_bit;
261     uint64_t offset, size;
262     int ret = 0;
263 
264     first_bit = s->offset_within_region / vmem->block_size;
265     first_bit = find_next_bit(vmem->bitmap, vmem->bitmap_size, first_bit);
266     while (first_bit < vmem->bitmap_size) {
267         MemoryRegionSection tmp = *s;
268 
269         offset = first_bit * vmem->block_size;
270         last_bit = find_next_zero_bit(vmem->bitmap, vmem->bitmap_size,
271                                       first_bit + 1) - 1;
272         size = (last_bit - first_bit + 1) * vmem->block_size;
273 
274         if (!virtio_mem_intersect_memory_section(&tmp, offset, size)) {
275             break;
276         }
277         ret = cb(&tmp, arg);
278         if (ret) {
279             break;
280         }
281         first_bit = find_next_bit(vmem->bitmap, vmem->bitmap_size,
282                                   last_bit + 2);
283     }
284     return ret;
285 }
286 
287 static int virtio_mem_for_each_unplugged_section(const VirtIOMEM *vmem,
288                                                  MemoryRegionSection *s,
289                                                  void *arg,
290                                                  virtio_mem_section_cb cb)
291 {
292     unsigned long first_bit, last_bit;
293     uint64_t offset, size;
294     int ret = 0;
295 
296     first_bit = s->offset_within_region / vmem->block_size;
297     first_bit = find_next_zero_bit(vmem->bitmap, vmem->bitmap_size, first_bit);
298     while (first_bit < vmem->bitmap_size) {
299         MemoryRegionSection tmp = *s;
300 
301         offset = first_bit * vmem->block_size;
302         last_bit = find_next_bit(vmem->bitmap, vmem->bitmap_size,
303                                  first_bit + 1) - 1;
304         size = (last_bit - first_bit + 1) * vmem->block_size;
305 
306         if (!virtio_mem_intersect_memory_section(&tmp, offset, size)) {
307             break;
308         }
309         ret = cb(&tmp, arg);
310         if (ret) {
311             break;
312         }
313         first_bit = find_next_zero_bit(vmem->bitmap, vmem->bitmap_size,
314                                        last_bit + 2);
315     }
316     return ret;
317 }
318 
319 static int virtio_mem_notify_populate_cb(MemoryRegionSection *s, void *arg)
320 {
321     RamDiscardListener *rdl = arg;
322 
323     return rdl->notify_populate(rdl, s);
324 }
325 
326 static int virtio_mem_notify_discard_cb(MemoryRegionSection *s, void *arg)
327 {
328     RamDiscardListener *rdl = arg;
329 
330     rdl->notify_discard(rdl, s);
331     return 0;
332 }
333 
334 static void virtio_mem_notify_unplug(VirtIOMEM *vmem, uint64_t offset,
335                                      uint64_t size)
336 {
337     RamDiscardListener *rdl;
338 
339     QLIST_FOREACH(rdl, &vmem->rdl_list, next) {
340         MemoryRegionSection tmp = *rdl->section;
341 
342         if (!virtio_mem_intersect_memory_section(&tmp, offset, size)) {
343             continue;
344         }
345         rdl->notify_discard(rdl, &tmp);
346     }
347 }
348 
349 static int virtio_mem_notify_plug(VirtIOMEM *vmem, uint64_t offset,
350                                   uint64_t size)
351 {
352     RamDiscardListener *rdl, *rdl2;
353     int ret = 0;
354 
355     QLIST_FOREACH(rdl, &vmem->rdl_list, next) {
356         MemoryRegionSection tmp = *rdl->section;
357 
358         if (!virtio_mem_intersect_memory_section(&tmp, offset, size)) {
359             continue;
360         }
361         ret = rdl->notify_populate(rdl, &tmp);
362         if (ret) {
363             break;
364         }
365     }
366 
367     if (ret) {
368         /* Notify all already-notified listeners. */
369         QLIST_FOREACH(rdl2, &vmem->rdl_list, next) {
370             MemoryRegionSection tmp = *rdl2->section;
371 
372             if (rdl2 == rdl) {
373                 break;
374             }
375             if (!virtio_mem_intersect_memory_section(&tmp, offset, size)) {
376                 continue;
377             }
378             rdl2->notify_discard(rdl2, &tmp);
379         }
380     }
381     return ret;
382 }
383 
384 static void virtio_mem_notify_unplug_all(VirtIOMEM *vmem)
385 {
386     RamDiscardListener *rdl;
387 
388     if (!vmem->size) {
389         return;
390     }
391 
392     QLIST_FOREACH(rdl, &vmem->rdl_list, next) {
393         if (rdl->double_discard_supported) {
394             rdl->notify_discard(rdl, rdl->section);
395         } else {
396             virtio_mem_for_each_plugged_section(vmem, rdl->section, rdl,
397                                                 virtio_mem_notify_discard_cb);
398         }
399     }
400 }
401 
402 static bool virtio_mem_is_range_plugged(const VirtIOMEM *vmem,
403                                         uint64_t start_gpa, uint64_t size)
404 {
405     const unsigned long first_bit = (start_gpa - vmem->addr) / vmem->block_size;
406     const unsigned long last_bit = first_bit + (size / vmem->block_size) - 1;
407     unsigned long found_bit;
408 
409     /* We fake a shorter bitmap to avoid searching too far. */
410     found_bit = find_next_zero_bit(vmem->bitmap, last_bit + 1, first_bit);
411     return found_bit > last_bit;
412 }
413 
414 static bool virtio_mem_is_range_unplugged(const VirtIOMEM *vmem,
415                                           uint64_t start_gpa, uint64_t size)
416 {
417     const unsigned long first_bit = (start_gpa - vmem->addr) / vmem->block_size;
418     const unsigned long last_bit = first_bit + (size / vmem->block_size) - 1;
419     unsigned long found_bit;
420 
421     /* We fake a shorter bitmap to avoid searching too far. */
422     found_bit = find_next_bit(vmem->bitmap, last_bit + 1, first_bit);
423     return found_bit > last_bit;
424 }
425 
426 static void virtio_mem_set_range_plugged(VirtIOMEM *vmem, uint64_t start_gpa,
427                                          uint64_t size)
428 {
429     const unsigned long bit = (start_gpa - vmem->addr) / vmem->block_size;
430     const unsigned long nbits = size / vmem->block_size;
431 
432     bitmap_set(vmem->bitmap, bit, nbits);
433 }
434 
435 static void virtio_mem_set_range_unplugged(VirtIOMEM *vmem, uint64_t start_gpa,
436                                            uint64_t size)
437 {
438     const unsigned long bit = (start_gpa - vmem->addr) / vmem->block_size;
439     const unsigned long nbits = size / vmem->block_size;
440 
441     bitmap_clear(vmem->bitmap, bit, nbits);
442 }
443 
444 static void virtio_mem_send_response(VirtIOMEM *vmem, VirtQueueElement *elem,
445                                      struct virtio_mem_resp *resp)
446 {
447     VirtIODevice *vdev = VIRTIO_DEVICE(vmem);
448     VirtQueue *vq = vmem->vq;
449 
450     trace_virtio_mem_send_response(le16_to_cpu(resp->type));
451     iov_from_buf(elem->in_sg, elem->in_num, 0, resp, sizeof(*resp));
452 
453     virtqueue_push(vq, elem, sizeof(*resp));
454     virtio_notify(vdev, vq);
455 }
456 
457 static void virtio_mem_send_response_simple(VirtIOMEM *vmem,
458                                             VirtQueueElement *elem,
459                                             uint16_t type)
460 {
461     struct virtio_mem_resp resp = {
462         .type = cpu_to_le16(type),
463     };
464 
465     virtio_mem_send_response(vmem, elem, &resp);
466 }
467 
468 static bool virtio_mem_valid_range(const VirtIOMEM *vmem, uint64_t gpa,
469                                    uint64_t size)
470 {
471     if (!QEMU_IS_ALIGNED(gpa, vmem->block_size)) {
472         return false;
473     }
474     if (gpa + size < gpa || !size) {
475         return false;
476     }
477     if (gpa < vmem->addr || gpa >= vmem->addr + vmem->usable_region_size) {
478         return false;
479     }
480     if (gpa + size > vmem->addr + vmem->usable_region_size) {
481         return false;
482     }
483     return true;
484 }
485 
486 static int virtio_mem_set_block_state(VirtIOMEM *vmem, uint64_t start_gpa,
487                                       uint64_t size, bool plug)
488 {
489     const uint64_t offset = start_gpa - vmem->addr;
490     RAMBlock *rb = vmem->memdev->mr.ram_block;
491     int ret = 0;
492 
493     if (virtio_mem_is_busy()) {
494         return -EBUSY;
495     }
496 
497     if (!plug) {
498         if (ram_block_discard_range(rb, offset, size)) {
499             return -EBUSY;
500         }
501         virtio_mem_notify_unplug(vmem, offset, size);
502         virtio_mem_set_range_unplugged(vmem, start_gpa, size);
503         return 0;
504     }
505 
506     if (vmem->prealloc) {
507         void *area = memory_region_get_ram_ptr(&vmem->memdev->mr) + offset;
508         int fd = memory_region_get_fd(&vmem->memdev->mr);
509         Error *local_err = NULL;
510 
511         qemu_prealloc_mem(fd, area, size, 1, NULL, &local_err);
512         if (local_err) {
513             static bool warned;
514 
515             /*
516              * Warn only once, we don't want to fill the log with these
517              * warnings.
518              */
519             if (!warned) {
520                 warn_report_err(local_err);
521                 warned = true;
522             } else {
523                 error_free(local_err);
524             }
525             ret = -EBUSY;
526         }
527     }
528 
529     if (!ret) {
530         ret = virtio_mem_notify_plug(vmem, offset, size);
531     }
532     if (ret) {
533         /* Could be preallocation or a notifier populated memory. */
534         ram_block_discard_range(vmem->memdev->mr.ram_block, offset, size);
535         return -EBUSY;
536     }
537 
538     virtio_mem_set_range_plugged(vmem, start_gpa, size);
539     return 0;
540 }
541 
542 static int virtio_mem_state_change_request(VirtIOMEM *vmem, uint64_t gpa,
543                                            uint16_t nb_blocks, bool plug)
544 {
545     const uint64_t size = nb_blocks * vmem->block_size;
546     int ret;
547 
548     if (!virtio_mem_valid_range(vmem, gpa, size)) {
549         return VIRTIO_MEM_RESP_ERROR;
550     }
551 
552     if (plug && (vmem->size + size > vmem->requested_size)) {
553         return VIRTIO_MEM_RESP_NACK;
554     }
555 
556     /* test if really all blocks are in the opposite state */
557     if ((plug && !virtio_mem_is_range_unplugged(vmem, gpa, size)) ||
558         (!plug && !virtio_mem_is_range_plugged(vmem, gpa, size))) {
559         return VIRTIO_MEM_RESP_ERROR;
560     }
561 
562     ret = virtio_mem_set_block_state(vmem, gpa, size, plug);
563     if (ret) {
564         return VIRTIO_MEM_RESP_BUSY;
565     }
566     if (plug) {
567         vmem->size += size;
568     } else {
569         vmem->size -= size;
570     }
571     notifier_list_notify(&vmem->size_change_notifiers, &vmem->size);
572     return VIRTIO_MEM_RESP_ACK;
573 }
574 
575 static void virtio_mem_plug_request(VirtIOMEM *vmem, VirtQueueElement *elem,
576                                     struct virtio_mem_req *req)
577 {
578     const uint64_t gpa = le64_to_cpu(req->u.plug.addr);
579     const uint16_t nb_blocks = le16_to_cpu(req->u.plug.nb_blocks);
580     uint16_t type;
581 
582     trace_virtio_mem_plug_request(gpa, nb_blocks);
583     type = virtio_mem_state_change_request(vmem, gpa, nb_blocks, true);
584     virtio_mem_send_response_simple(vmem, elem, type);
585 }
586 
587 static void virtio_mem_unplug_request(VirtIOMEM *vmem, VirtQueueElement *elem,
588                                       struct virtio_mem_req *req)
589 {
590     const uint64_t gpa = le64_to_cpu(req->u.unplug.addr);
591     const uint16_t nb_blocks = le16_to_cpu(req->u.unplug.nb_blocks);
592     uint16_t type;
593 
594     trace_virtio_mem_unplug_request(gpa, nb_blocks);
595     type = virtio_mem_state_change_request(vmem, gpa, nb_blocks, false);
596     virtio_mem_send_response_simple(vmem, elem, type);
597 }
598 
599 static void virtio_mem_resize_usable_region(VirtIOMEM *vmem,
600                                             uint64_t requested_size,
601                                             bool can_shrink)
602 {
603     uint64_t newsize = MIN(memory_region_size(&vmem->memdev->mr),
604                            requested_size + VIRTIO_MEM_USABLE_EXTENT);
605 
606     /* The usable region size always has to be multiples of the block size. */
607     newsize = QEMU_ALIGN_UP(newsize, vmem->block_size);
608 
609     if (!requested_size) {
610         newsize = 0;
611     }
612 
613     if (newsize < vmem->usable_region_size && !can_shrink) {
614         return;
615     }
616 
617     trace_virtio_mem_resized_usable_region(vmem->usable_region_size, newsize);
618     vmem->usable_region_size = newsize;
619 }
620 
621 static int virtio_mem_unplug_all(VirtIOMEM *vmem)
622 {
623     RAMBlock *rb = vmem->memdev->mr.ram_block;
624 
625     if (vmem->size) {
626         if (virtio_mem_is_busy()) {
627             return -EBUSY;
628         }
629         if (ram_block_discard_range(rb, 0, qemu_ram_get_used_length(rb))) {
630             return -EBUSY;
631         }
632         virtio_mem_notify_unplug_all(vmem);
633 
634         bitmap_clear(vmem->bitmap, 0, vmem->bitmap_size);
635         vmem->size = 0;
636         notifier_list_notify(&vmem->size_change_notifiers, &vmem->size);
637     }
638 
639     trace_virtio_mem_unplugged_all();
640     virtio_mem_resize_usable_region(vmem, vmem->requested_size, true);
641     return 0;
642 }
643 
644 static void virtio_mem_unplug_all_request(VirtIOMEM *vmem,
645                                           VirtQueueElement *elem)
646 {
647     trace_virtio_mem_unplug_all_request();
648     if (virtio_mem_unplug_all(vmem)) {
649         virtio_mem_send_response_simple(vmem, elem, VIRTIO_MEM_RESP_BUSY);
650     } else {
651         virtio_mem_send_response_simple(vmem, elem, VIRTIO_MEM_RESP_ACK);
652     }
653 }
654 
655 static void virtio_mem_state_request(VirtIOMEM *vmem, VirtQueueElement *elem,
656                                      struct virtio_mem_req *req)
657 {
658     const uint16_t nb_blocks = le16_to_cpu(req->u.state.nb_blocks);
659     const uint64_t gpa = le64_to_cpu(req->u.state.addr);
660     const uint64_t size = nb_blocks * vmem->block_size;
661     struct virtio_mem_resp resp = {
662         .type = cpu_to_le16(VIRTIO_MEM_RESP_ACK),
663     };
664 
665     trace_virtio_mem_state_request(gpa, nb_blocks);
666     if (!virtio_mem_valid_range(vmem, gpa, size)) {
667         virtio_mem_send_response_simple(vmem, elem, VIRTIO_MEM_RESP_ERROR);
668         return;
669     }
670 
671     if (virtio_mem_is_range_plugged(vmem, gpa, size)) {
672         resp.u.state.state = cpu_to_le16(VIRTIO_MEM_STATE_PLUGGED);
673     } else if (virtio_mem_is_range_unplugged(vmem, gpa, size)) {
674         resp.u.state.state = cpu_to_le16(VIRTIO_MEM_STATE_UNPLUGGED);
675     } else {
676         resp.u.state.state = cpu_to_le16(VIRTIO_MEM_STATE_MIXED);
677     }
678     trace_virtio_mem_state_response(le16_to_cpu(resp.u.state.state));
679     virtio_mem_send_response(vmem, elem, &resp);
680 }
681 
682 static void virtio_mem_handle_request(VirtIODevice *vdev, VirtQueue *vq)
683 {
684     const int len = sizeof(struct virtio_mem_req);
685     VirtIOMEM *vmem = VIRTIO_MEM(vdev);
686     VirtQueueElement *elem;
687     struct virtio_mem_req req;
688     uint16_t type;
689 
690     while (true) {
691         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
692         if (!elem) {
693             return;
694         }
695 
696         if (iov_to_buf(elem->out_sg, elem->out_num, 0, &req, len) < len) {
697             virtio_error(vdev, "virtio-mem protocol violation: invalid request"
698                          " size: %d", len);
699             virtqueue_detach_element(vq, elem, 0);
700             g_free(elem);
701             return;
702         }
703 
704         if (iov_size(elem->in_sg, elem->in_num) <
705             sizeof(struct virtio_mem_resp)) {
706             virtio_error(vdev, "virtio-mem protocol violation: not enough space"
707                          " for response: %zu",
708                          iov_size(elem->in_sg, elem->in_num));
709             virtqueue_detach_element(vq, elem, 0);
710             g_free(elem);
711             return;
712         }
713 
714         type = le16_to_cpu(req.type);
715         switch (type) {
716         case VIRTIO_MEM_REQ_PLUG:
717             virtio_mem_plug_request(vmem, elem, &req);
718             break;
719         case VIRTIO_MEM_REQ_UNPLUG:
720             virtio_mem_unplug_request(vmem, elem, &req);
721             break;
722         case VIRTIO_MEM_REQ_UNPLUG_ALL:
723             virtio_mem_unplug_all_request(vmem, elem);
724             break;
725         case VIRTIO_MEM_REQ_STATE:
726             virtio_mem_state_request(vmem, elem, &req);
727             break;
728         default:
729             virtio_error(vdev, "virtio-mem protocol violation: unknown request"
730                          " type: %d", type);
731             virtqueue_detach_element(vq, elem, 0);
732             g_free(elem);
733             return;
734         }
735 
736         g_free(elem);
737     }
738 }
739 
740 static void virtio_mem_get_config(VirtIODevice *vdev, uint8_t *config_data)
741 {
742     VirtIOMEM *vmem = VIRTIO_MEM(vdev);
743     struct virtio_mem_config *config = (void *) config_data;
744 
745     config->block_size = cpu_to_le64(vmem->block_size);
746     config->node_id = cpu_to_le16(vmem->node);
747     config->requested_size = cpu_to_le64(vmem->requested_size);
748     config->plugged_size = cpu_to_le64(vmem->size);
749     config->addr = cpu_to_le64(vmem->addr);
750     config->region_size = cpu_to_le64(memory_region_size(&vmem->memdev->mr));
751     config->usable_region_size = cpu_to_le64(vmem->usable_region_size);
752 }
753 
754 static uint64_t virtio_mem_get_features(VirtIODevice *vdev, uint64_t features,
755                                         Error **errp)
756 {
757     MachineState *ms = MACHINE(qdev_get_machine());
758     VirtIOMEM *vmem = VIRTIO_MEM(vdev);
759 
760     if (ms->numa_state) {
761 #if defined(CONFIG_ACPI)
762         virtio_add_feature(&features, VIRTIO_MEM_F_ACPI_PXM);
763 #endif
764     }
765     assert(vmem->unplugged_inaccessible != ON_OFF_AUTO_AUTO);
766     if (vmem->unplugged_inaccessible == ON_OFF_AUTO_ON) {
767         virtio_add_feature(&features, VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE);
768     }
769     return features;
770 }
771 
772 static int virtio_mem_validate_features(VirtIODevice *vdev)
773 {
774     if (virtio_host_has_feature(vdev, VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE) &&
775         !virtio_vdev_has_feature(vdev, VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE)) {
776         return -EFAULT;
777     }
778     return 0;
779 }
780 
781 static void virtio_mem_system_reset(void *opaque)
782 {
783     VirtIOMEM *vmem = VIRTIO_MEM(opaque);
784 
785     /*
786      * During usual resets, we will unplug all memory and shrink the usable
787      * region size. This is, however, not possible in all scenarios. Then,
788      * the guest has to deal with this manually (VIRTIO_MEM_REQ_UNPLUG_ALL).
789      */
790     virtio_mem_unplug_all(vmem);
791 }
792 
793 static void virtio_mem_device_realize(DeviceState *dev, Error **errp)
794 {
795     MachineState *ms = MACHINE(qdev_get_machine());
796     int nb_numa_nodes = ms->numa_state ? ms->numa_state->num_nodes : 0;
797     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
798     VirtIOMEM *vmem = VIRTIO_MEM(dev);
799     uint64_t page_size;
800     RAMBlock *rb;
801     int ret;
802 
803     if (!vmem->memdev) {
804         error_setg(errp, "'%s' property is not set", VIRTIO_MEM_MEMDEV_PROP);
805         return;
806     } else if (host_memory_backend_is_mapped(vmem->memdev)) {
807         error_setg(errp, "'%s' property specifies a busy memdev: %s",
808                    VIRTIO_MEM_MEMDEV_PROP,
809                    object_get_canonical_path_component(OBJECT(vmem->memdev)));
810         return;
811     } else if (!memory_region_is_ram(&vmem->memdev->mr) ||
812         memory_region_is_rom(&vmem->memdev->mr) ||
813         !vmem->memdev->mr.ram_block) {
814         error_setg(errp, "'%s' property specifies an unsupported memdev",
815                    VIRTIO_MEM_MEMDEV_PROP);
816         return;
817     } else if (vmem->memdev->prealloc) {
818         error_setg(errp, "'%s' property specifies a memdev with preallocation"
819                    " enabled: %s. Instead, specify 'prealloc=on' for the"
820                    " virtio-mem device. ", VIRTIO_MEM_MEMDEV_PROP,
821                    object_get_canonical_path_component(OBJECT(vmem->memdev)));
822         return;
823     }
824 
825     if ((nb_numa_nodes && vmem->node >= nb_numa_nodes) ||
826         (!nb_numa_nodes && vmem->node)) {
827         error_setg(errp, "'%s' property has value '%" PRIu32 "', which exceeds"
828                    "the number of numa nodes: %d", VIRTIO_MEM_NODE_PROP,
829                    vmem->node, nb_numa_nodes ? nb_numa_nodes : 1);
830         return;
831     }
832 
833     if (enable_mlock) {
834         error_setg(errp, "Incompatible with mlock");
835         return;
836     }
837 
838     rb = vmem->memdev->mr.ram_block;
839     page_size = qemu_ram_pagesize(rb);
840 
841 #if defined(VIRTIO_MEM_HAS_LEGACY_GUESTS)
842     switch (vmem->unplugged_inaccessible) {
843     case ON_OFF_AUTO_AUTO:
844         if (virtio_mem_has_shared_zeropage(rb)) {
845             vmem->unplugged_inaccessible = ON_OFF_AUTO_OFF;
846         } else {
847             vmem->unplugged_inaccessible = ON_OFF_AUTO_ON;
848         }
849         break;
850     case ON_OFF_AUTO_OFF:
851         if (!virtio_mem_has_shared_zeropage(rb)) {
852             warn_report("'%s' property set to 'off' with a memdev that does"
853                         " not support the shared zeropage.",
854                         VIRTIO_MEM_UNPLUGGED_INACCESSIBLE_PROP);
855         }
856         break;
857     default:
858         break;
859     }
860 #else /* VIRTIO_MEM_HAS_LEGACY_GUESTS */
861     vmem->unplugged_inaccessible = ON_OFF_AUTO_ON;
862 #endif /* VIRTIO_MEM_HAS_LEGACY_GUESTS */
863 
864     /*
865      * If the block size wasn't configured by the user, use a sane default. This
866      * allows using hugetlbfs backends of any page size without manual
867      * intervention.
868      */
869     if (!vmem->block_size) {
870         vmem->block_size = virtio_mem_default_block_size(rb);
871     }
872 
873     if (vmem->block_size < page_size) {
874         error_setg(errp, "'%s' property has to be at least the page size (0x%"
875                    PRIx64 ")", VIRTIO_MEM_BLOCK_SIZE_PROP, page_size);
876         return;
877     } else if (vmem->block_size < virtio_mem_default_block_size(rb)) {
878         warn_report("'%s' property is smaller than the default block size (%"
879                     PRIx64 " MiB)", VIRTIO_MEM_BLOCK_SIZE_PROP,
880                     virtio_mem_default_block_size(rb) / MiB);
881     }
882     if (!QEMU_IS_ALIGNED(vmem->requested_size, vmem->block_size)) {
883         error_setg(errp, "'%s' property has to be multiples of '%s' (0x%" PRIx64
884                    ")", VIRTIO_MEM_REQUESTED_SIZE_PROP,
885                    VIRTIO_MEM_BLOCK_SIZE_PROP, vmem->block_size);
886         return;
887     } else if (!QEMU_IS_ALIGNED(vmem->addr, vmem->block_size)) {
888         error_setg(errp, "'%s' property has to be multiples of '%s' (0x%" PRIx64
889                    ")", VIRTIO_MEM_ADDR_PROP, VIRTIO_MEM_BLOCK_SIZE_PROP,
890                    vmem->block_size);
891         return;
892     } else if (!QEMU_IS_ALIGNED(memory_region_size(&vmem->memdev->mr),
893                                 vmem->block_size)) {
894         error_setg(errp, "'%s' property memdev size has to be multiples of"
895                    "'%s' (0x%" PRIx64 ")", VIRTIO_MEM_MEMDEV_PROP,
896                    VIRTIO_MEM_BLOCK_SIZE_PROP, vmem->block_size);
897         return;
898     }
899 
900     if (ram_block_coordinated_discard_require(true)) {
901         error_setg(errp, "Discarding RAM is disabled");
902         return;
903     }
904 
905     /*
906      * We don't know at this point whether shared RAM is migrated using
907      * QEMU or migrated using the file content. "x-ignore-shared" will be
908      * configured after realizing the device. So in case we have an
909      * incoming migration, simply always skip the discard step.
910      *
911      * Otherwise, make sure that we start with a clean slate: either the
912      * memory backend might get reused or the shared file might still have
913      * memory allocated.
914      */
915     if (!runstate_check(RUN_STATE_INMIGRATE)) {
916         ret = ram_block_discard_range(rb, 0, qemu_ram_get_used_length(rb));
917         if (ret) {
918             error_setg_errno(errp, -ret, "Unexpected error discarding RAM");
919             ram_block_coordinated_discard_require(false);
920             return;
921         }
922     }
923 
924     virtio_mem_resize_usable_region(vmem, vmem->requested_size, true);
925 
926     vmem->bitmap_size = memory_region_size(&vmem->memdev->mr) /
927                         vmem->block_size;
928     vmem->bitmap = bitmap_new(vmem->bitmap_size);
929 
930     virtio_init(vdev, VIRTIO_ID_MEM, sizeof(struct virtio_mem_config));
931     vmem->vq = virtio_add_queue(vdev, 128, virtio_mem_handle_request);
932 
933     host_memory_backend_set_mapped(vmem->memdev, true);
934     vmstate_register_ram(&vmem->memdev->mr, DEVICE(vmem));
935     if (vmem->early_migration) {
936         vmstate_register(VMSTATE_IF(vmem), VMSTATE_INSTANCE_ID_ANY,
937                          &vmstate_virtio_mem_device_early, vmem);
938     }
939     qemu_register_reset(virtio_mem_system_reset, vmem);
940 
941     /*
942      * Set ourselves as RamDiscardManager before the plug handler maps the
943      * memory region and exposes it via an address space.
944      */
945     memory_region_set_ram_discard_manager(&vmem->memdev->mr,
946                                           RAM_DISCARD_MANAGER(vmem));
947 }
948 
949 static void virtio_mem_device_unrealize(DeviceState *dev)
950 {
951     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
952     VirtIOMEM *vmem = VIRTIO_MEM(dev);
953 
954     /*
955      * The unplug handler unmapped the memory region, it cannot be
956      * found via an address space anymore. Unset ourselves.
957      */
958     memory_region_set_ram_discard_manager(&vmem->memdev->mr, NULL);
959     qemu_unregister_reset(virtio_mem_system_reset, vmem);
960     if (vmem->early_migration) {
961         vmstate_unregister(VMSTATE_IF(vmem), &vmstate_virtio_mem_device_early,
962                            vmem);
963     }
964     vmstate_unregister_ram(&vmem->memdev->mr, DEVICE(vmem));
965     host_memory_backend_set_mapped(vmem->memdev, false);
966     virtio_del_queue(vdev, 0);
967     virtio_cleanup(vdev);
968     g_free(vmem->bitmap);
969     ram_block_coordinated_discard_require(false);
970 }
971 
972 static int virtio_mem_discard_range_cb(const VirtIOMEM *vmem, void *arg,
973                                        uint64_t offset, uint64_t size)
974 {
975     RAMBlock *rb = vmem->memdev->mr.ram_block;
976 
977     return ram_block_discard_range(rb, offset, size) ? -EINVAL : 0;
978 }
979 
980 static int virtio_mem_restore_unplugged(VirtIOMEM *vmem)
981 {
982     /* Make sure all memory is really discarded after migration. */
983     return virtio_mem_for_each_unplugged_range(vmem, NULL,
984                                                virtio_mem_discard_range_cb);
985 }
986 
987 static int virtio_mem_post_load(void *opaque, int version_id)
988 {
989     VirtIOMEM *vmem = VIRTIO_MEM(opaque);
990     RamDiscardListener *rdl;
991     int ret;
992 
993     /*
994      * We started out with all memory discarded and our memory region is mapped
995      * into an address space. Replay, now that we updated the bitmap.
996      */
997     QLIST_FOREACH(rdl, &vmem->rdl_list, next) {
998         ret = virtio_mem_for_each_plugged_section(vmem, rdl->section, rdl,
999                                                  virtio_mem_notify_populate_cb);
1000         if (ret) {
1001             return ret;
1002         }
1003     }
1004 
1005     /*
1006      * If shared RAM is migrated using the file content and not using QEMU,
1007      * don't mess with preallocation and postcopy.
1008      */
1009     if (migrate_ram_is_ignored(vmem->memdev->mr.ram_block)) {
1010         return 0;
1011     }
1012 
1013     if (vmem->prealloc && !vmem->early_migration) {
1014         warn_report("Proper preallocation with migration requires a newer QEMU machine");
1015     }
1016 
1017     if (migration_in_incoming_postcopy()) {
1018         return 0;
1019     }
1020 
1021     return virtio_mem_restore_unplugged(vmem);
1022 }
1023 
1024 static int virtio_mem_prealloc_range_cb(const VirtIOMEM *vmem, void *arg,
1025                                         uint64_t offset, uint64_t size)
1026 {
1027     void *area = memory_region_get_ram_ptr(&vmem->memdev->mr) + offset;
1028     int fd = memory_region_get_fd(&vmem->memdev->mr);
1029     Error *local_err = NULL;
1030 
1031     qemu_prealloc_mem(fd, area, size, 1, NULL, &local_err);
1032     if (local_err) {
1033         error_report_err(local_err);
1034         return -ENOMEM;
1035     }
1036     return 0;
1037 }
1038 
1039 static int virtio_mem_post_load_early(void *opaque, int version_id)
1040 {
1041     VirtIOMEM *vmem = VIRTIO_MEM(opaque);
1042     RAMBlock *rb = vmem->memdev->mr.ram_block;
1043     int ret;
1044 
1045     if (!vmem->prealloc) {
1046         return 0;
1047     }
1048 
1049     /*
1050      * If shared RAM is migrated using the file content and not using QEMU,
1051      * don't mess with preallocation and postcopy.
1052      */
1053     if (migrate_ram_is_ignored(rb)) {
1054         return 0;
1055     }
1056 
1057     /*
1058      * We restored the bitmap and verified that the basic properties
1059      * match on source and destination, so we can go ahead and preallocate
1060      * memory for all plugged memory blocks, before actual RAM migration starts
1061      * touching this memory.
1062      */
1063     ret = virtio_mem_for_each_plugged_range(vmem, NULL,
1064                                             virtio_mem_prealloc_range_cb);
1065     if (ret) {
1066         return ret;
1067     }
1068 
1069     /*
1070      * This is tricky: postcopy wants to start with a clean slate. On
1071      * POSTCOPY_INCOMING_ADVISE, postcopy code discards all (ordinarily
1072      * preallocated) RAM such that postcopy will work as expected later.
1073      *
1074      * However, we run after POSTCOPY_INCOMING_ADVISE -- but before actual
1075      * RAM migration. So let's discard all memory again. This looks like an
1076      * expensive NOP, but actually serves a purpose: we made sure that we
1077      * were able to allocate all required backend memory once. We cannot
1078      * guarantee that the backend memory we will free will remain free
1079      * until we need it during postcopy, but at least we can catch the
1080      * obvious setup issues this way.
1081      */
1082     if (migration_incoming_postcopy_advised()) {
1083         if (ram_block_discard_range(rb, 0, qemu_ram_get_used_length(rb))) {
1084             return -EBUSY;
1085         }
1086     }
1087     return 0;
1088 }
1089 
1090 typedef struct VirtIOMEMMigSanityChecks {
1091     VirtIOMEM *parent;
1092     uint64_t addr;
1093     uint64_t region_size;
1094     uint64_t block_size;
1095     uint32_t node;
1096 } VirtIOMEMMigSanityChecks;
1097 
1098 static int virtio_mem_mig_sanity_checks_pre_save(void *opaque)
1099 {
1100     VirtIOMEMMigSanityChecks *tmp = opaque;
1101     VirtIOMEM *vmem = tmp->parent;
1102 
1103     tmp->addr = vmem->addr;
1104     tmp->region_size = memory_region_size(&vmem->memdev->mr);
1105     tmp->block_size = vmem->block_size;
1106     tmp->node = vmem->node;
1107     return 0;
1108 }
1109 
1110 static int virtio_mem_mig_sanity_checks_post_load(void *opaque, int version_id)
1111 {
1112     VirtIOMEMMigSanityChecks *tmp = opaque;
1113     VirtIOMEM *vmem = tmp->parent;
1114     const uint64_t new_region_size = memory_region_size(&vmem->memdev->mr);
1115 
1116     if (tmp->addr != vmem->addr) {
1117         error_report("Property '%s' changed from 0x%" PRIx64 " to 0x%" PRIx64,
1118                      VIRTIO_MEM_ADDR_PROP, tmp->addr, vmem->addr);
1119         return -EINVAL;
1120     }
1121     /*
1122      * Note: Preparation for resizable memory regions. The maximum size
1123      * of the memory region must not change during migration.
1124      */
1125     if (tmp->region_size != new_region_size) {
1126         error_report("Property '%s' size changed from 0x%" PRIx64 " to 0x%"
1127                      PRIx64, VIRTIO_MEM_MEMDEV_PROP, tmp->region_size,
1128                      new_region_size);
1129         return -EINVAL;
1130     }
1131     if (tmp->block_size != vmem->block_size) {
1132         error_report("Property '%s' changed from 0x%" PRIx64 " to 0x%" PRIx64,
1133                      VIRTIO_MEM_BLOCK_SIZE_PROP, tmp->block_size,
1134                      vmem->block_size);
1135         return -EINVAL;
1136     }
1137     if (tmp->node != vmem->node) {
1138         error_report("Property '%s' changed from %" PRIu32 " to %" PRIu32,
1139                      VIRTIO_MEM_NODE_PROP, tmp->node, vmem->node);
1140         return -EINVAL;
1141     }
1142     return 0;
1143 }
1144 
1145 static const VMStateDescription vmstate_virtio_mem_sanity_checks = {
1146     .name = "virtio-mem-device/sanity-checks",
1147     .pre_save = virtio_mem_mig_sanity_checks_pre_save,
1148     .post_load = virtio_mem_mig_sanity_checks_post_load,
1149     .fields = (VMStateField[]) {
1150         VMSTATE_UINT64(addr, VirtIOMEMMigSanityChecks),
1151         VMSTATE_UINT64(region_size, VirtIOMEMMigSanityChecks),
1152         VMSTATE_UINT64(block_size, VirtIOMEMMigSanityChecks),
1153         VMSTATE_UINT32(node, VirtIOMEMMigSanityChecks),
1154         VMSTATE_END_OF_LIST(),
1155     },
1156 };
1157 
1158 static bool virtio_mem_vmstate_field_exists(void *opaque, int version_id)
1159 {
1160     const VirtIOMEM *vmem = VIRTIO_MEM(opaque);
1161 
1162     /* With early migration, these fields were already migrated. */
1163     return !vmem->early_migration;
1164 }
1165 
1166 static const VMStateDescription vmstate_virtio_mem_device = {
1167     .name = "virtio-mem-device",
1168     .minimum_version_id = 1,
1169     .version_id = 1,
1170     .priority = MIG_PRI_VIRTIO_MEM,
1171     .post_load = virtio_mem_post_load,
1172     .fields = (VMStateField[]) {
1173         VMSTATE_WITH_TMP_TEST(VirtIOMEM, virtio_mem_vmstate_field_exists,
1174                               VirtIOMEMMigSanityChecks,
1175                               vmstate_virtio_mem_sanity_checks),
1176         VMSTATE_UINT64(usable_region_size, VirtIOMEM),
1177         VMSTATE_UINT64_TEST(size, VirtIOMEM, virtio_mem_vmstate_field_exists),
1178         VMSTATE_UINT64(requested_size, VirtIOMEM),
1179         VMSTATE_BITMAP_TEST(bitmap, VirtIOMEM, virtio_mem_vmstate_field_exists,
1180                             0, bitmap_size),
1181         VMSTATE_END_OF_LIST()
1182     },
1183 };
1184 
1185 /*
1186  * Transfer properties that are immutable while migration is active early,
1187  * such that we have have this information around before migrating any RAM
1188  * content.
1189  *
1190  * Note that virtio_mem_is_busy() makes sure these properties can no longer
1191  * change on the migration source until migration completed.
1192  *
1193  * With QEMU compat machines, we transmit these properties later, via
1194  * vmstate_virtio_mem_device instead -- see virtio_mem_vmstate_field_exists().
1195  */
1196 static const VMStateDescription vmstate_virtio_mem_device_early = {
1197     .name = "virtio-mem-device-early",
1198     .minimum_version_id = 1,
1199     .version_id = 1,
1200     .early_setup = true,
1201     .post_load = virtio_mem_post_load_early,
1202     .fields = (VMStateField[]) {
1203         VMSTATE_WITH_TMP(VirtIOMEM, VirtIOMEMMigSanityChecks,
1204                          vmstate_virtio_mem_sanity_checks),
1205         VMSTATE_UINT64(size, VirtIOMEM),
1206         VMSTATE_BITMAP(bitmap, VirtIOMEM, 0, bitmap_size),
1207         VMSTATE_END_OF_LIST()
1208     },
1209 };
1210 
1211 static const VMStateDescription vmstate_virtio_mem = {
1212     .name = "virtio-mem",
1213     .minimum_version_id = 1,
1214     .version_id = 1,
1215     .fields = (VMStateField[]) {
1216         VMSTATE_VIRTIO_DEVICE,
1217         VMSTATE_END_OF_LIST()
1218     },
1219 };
1220 
1221 static void virtio_mem_fill_device_info(const VirtIOMEM *vmem,
1222                                         VirtioMEMDeviceInfo *vi)
1223 {
1224     vi->memaddr = vmem->addr;
1225     vi->node = vmem->node;
1226     vi->requested_size = vmem->requested_size;
1227     vi->size = vmem->size;
1228     vi->max_size = memory_region_size(&vmem->memdev->mr);
1229     vi->block_size = vmem->block_size;
1230     vi->memdev = object_get_canonical_path(OBJECT(vmem->memdev));
1231 }
1232 
1233 static MemoryRegion *virtio_mem_get_memory_region(VirtIOMEM *vmem, Error **errp)
1234 {
1235     if (!vmem->memdev) {
1236         error_setg(errp, "'%s' property must be set", VIRTIO_MEM_MEMDEV_PROP);
1237         return NULL;
1238     }
1239 
1240     return &vmem->memdev->mr;
1241 }
1242 
1243 static void virtio_mem_add_size_change_notifier(VirtIOMEM *vmem,
1244                                                 Notifier *notifier)
1245 {
1246     notifier_list_add(&vmem->size_change_notifiers, notifier);
1247 }
1248 
1249 static void virtio_mem_remove_size_change_notifier(VirtIOMEM *vmem,
1250                                                    Notifier *notifier)
1251 {
1252     notifier_remove(notifier);
1253 }
1254 
1255 static void virtio_mem_get_size(Object *obj, Visitor *v, const char *name,
1256                                 void *opaque, Error **errp)
1257 {
1258     const VirtIOMEM *vmem = VIRTIO_MEM(obj);
1259     uint64_t value = vmem->size;
1260 
1261     visit_type_size(v, name, &value, errp);
1262 }
1263 
1264 static void virtio_mem_get_requested_size(Object *obj, Visitor *v,
1265                                           const char *name, void *opaque,
1266                                           Error **errp)
1267 {
1268     const VirtIOMEM *vmem = VIRTIO_MEM(obj);
1269     uint64_t value = vmem->requested_size;
1270 
1271     visit_type_size(v, name, &value, errp);
1272 }
1273 
1274 static void virtio_mem_set_requested_size(Object *obj, Visitor *v,
1275                                           const char *name, void *opaque,
1276                                           Error **errp)
1277 {
1278     VirtIOMEM *vmem = VIRTIO_MEM(obj);
1279     uint64_t value;
1280 
1281     if (!visit_type_size(v, name, &value, errp)) {
1282         return;
1283     }
1284 
1285     /*
1286      * The block size and memory backend are not fixed until the device was
1287      * realized. realize() will verify these properties then.
1288      */
1289     if (DEVICE(obj)->realized) {
1290         if (!QEMU_IS_ALIGNED(value, vmem->block_size)) {
1291             error_setg(errp, "'%s' has to be multiples of '%s' (0x%" PRIx64
1292                        ")", name, VIRTIO_MEM_BLOCK_SIZE_PROP,
1293                        vmem->block_size);
1294             return;
1295         } else if (value > memory_region_size(&vmem->memdev->mr)) {
1296             error_setg(errp, "'%s' cannot exceed the memory backend size"
1297                        "(0x%" PRIx64 ")", name,
1298                        memory_region_size(&vmem->memdev->mr));
1299             return;
1300         }
1301 
1302         if (value != vmem->requested_size) {
1303             virtio_mem_resize_usable_region(vmem, value, false);
1304             vmem->requested_size = value;
1305         }
1306         /*
1307          * Trigger a config update so the guest gets notified. We trigger
1308          * even if the size didn't change (especially helpful for debugging).
1309          */
1310         virtio_notify_config(VIRTIO_DEVICE(vmem));
1311     } else {
1312         vmem->requested_size = value;
1313     }
1314 }
1315 
1316 static void virtio_mem_get_block_size(Object *obj, Visitor *v, const char *name,
1317                                       void *opaque, Error **errp)
1318 {
1319     const VirtIOMEM *vmem = VIRTIO_MEM(obj);
1320     uint64_t value = vmem->block_size;
1321 
1322     /*
1323      * If not configured by the user (and we're not realized yet), use the
1324      * default block size we would use with the current memory backend.
1325      */
1326     if (!value) {
1327         if (vmem->memdev && memory_region_is_ram(&vmem->memdev->mr)) {
1328             value = virtio_mem_default_block_size(vmem->memdev->mr.ram_block);
1329         } else {
1330             value = virtio_mem_thp_size();
1331         }
1332     }
1333 
1334     visit_type_size(v, name, &value, errp);
1335 }
1336 
1337 static void virtio_mem_set_block_size(Object *obj, Visitor *v, const char *name,
1338                                       void *opaque, Error **errp)
1339 {
1340     VirtIOMEM *vmem = VIRTIO_MEM(obj);
1341     uint64_t value;
1342 
1343     if (DEVICE(obj)->realized) {
1344         error_setg(errp, "'%s' cannot be changed", name);
1345         return;
1346     }
1347 
1348     if (!visit_type_size(v, name, &value, errp)) {
1349         return;
1350     }
1351 
1352     if (value < VIRTIO_MEM_MIN_BLOCK_SIZE) {
1353         error_setg(errp, "'%s' property has to be at least 0x%" PRIx32, name,
1354                    VIRTIO_MEM_MIN_BLOCK_SIZE);
1355         return;
1356     } else if (!is_power_of_2(value)) {
1357         error_setg(errp, "'%s' property has to be a power of two", name);
1358         return;
1359     }
1360     vmem->block_size = value;
1361 }
1362 
1363 static void virtio_mem_instance_init(Object *obj)
1364 {
1365     VirtIOMEM *vmem = VIRTIO_MEM(obj);
1366 
1367     notifier_list_init(&vmem->size_change_notifiers);
1368     QLIST_INIT(&vmem->rdl_list);
1369 
1370     object_property_add(obj, VIRTIO_MEM_SIZE_PROP, "size", virtio_mem_get_size,
1371                         NULL, NULL, NULL);
1372     object_property_add(obj, VIRTIO_MEM_REQUESTED_SIZE_PROP, "size",
1373                         virtio_mem_get_requested_size,
1374                         virtio_mem_set_requested_size, NULL, NULL);
1375     object_property_add(obj, VIRTIO_MEM_BLOCK_SIZE_PROP, "size",
1376                         virtio_mem_get_block_size, virtio_mem_set_block_size,
1377                         NULL, NULL);
1378 }
1379 
1380 static Property virtio_mem_properties[] = {
1381     DEFINE_PROP_UINT64(VIRTIO_MEM_ADDR_PROP, VirtIOMEM, addr, 0),
1382     DEFINE_PROP_UINT32(VIRTIO_MEM_NODE_PROP, VirtIOMEM, node, 0),
1383     DEFINE_PROP_BOOL(VIRTIO_MEM_PREALLOC_PROP, VirtIOMEM, prealloc, false),
1384     DEFINE_PROP_LINK(VIRTIO_MEM_MEMDEV_PROP, VirtIOMEM, memdev,
1385                      TYPE_MEMORY_BACKEND, HostMemoryBackend *),
1386 #if defined(VIRTIO_MEM_HAS_LEGACY_GUESTS)
1387     DEFINE_PROP_ON_OFF_AUTO(VIRTIO_MEM_UNPLUGGED_INACCESSIBLE_PROP, VirtIOMEM,
1388                             unplugged_inaccessible, ON_OFF_AUTO_ON),
1389 #endif
1390     DEFINE_PROP_BOOL(VIRTIO_MEM_EARLY_MIGRATION_PROP, VirtIOMEM,
1391                      early_migration, true),
1392     DEFINE_PROP_END_OF_LIST(),
1393 };
1394 
1395 static uint64_t virtio_mem_rdm_get_min_granularity(const RamDiscardManager *rdm,
1396                                                    const MemoryRegion *mr)
1397 {
1398     const VirtIOMEM *vmem = VIRTIO_MEM(rdm);
1399 
1400     g_assert(mr == &vmem->memdev->mr);
1401     return vmem->block_size;
1402 }
1403 
1404 static bool virtio_mem_rdm_is_populated(const RamDiscardManager *rdm,
1405                                         const MemoryRegionSection *s)
1406 {
1407     const VirtIOMEM *vmem = VIRTIO_MEM(rdm);
1408     uint64_t start_gpa = vmem->addr + s->offset_within_region;
1409     uint64_t end_gpa = start_gpa + int128_get64(s->size);
1410 
1411     g_assert(s->mr == &vmem->memdev->mr);
1412 
1413     start_gpa = QEMU_ALIGN_DOWN(start_gpa, vmem->block_size);
1414     end_gpa = QEMU_ALIGN_UP(end_gpa, vmem->block_size);
1415 
1416     if (!virtio_mem_valid_range(vmem, start_gpa, end_gpa - start_gpa)) {
1417         return false;
1418     }
1419 
1420     return virtio_mem_is_range_plugged(vmem, start_gpa, end_gpa - start_gpa);
1421 }
1422 
1423 struct VirtIOMEMReplayData {
1424     void *fn;
1425     void *opaque;
1426 };
1427 
1428 static int virtio_mem_rdm_replay_populated_cb(MemoryRegionSection *s, void *arg)
1429 {
1430     struct VirtIOMEMReplayData *data = arg;
1431 
1432     return ((ReplayRamPopulate)data->fn)(s, data->opaque);
1433 }
1434 
1435 static int virtio_mem_rdm_replay_populated(const RamDiscardManager *rdm,
1436                                            MemoryRegionSection *s,
1437                                            ReplayRamPopulate replay_fn,
1438                                            void *opaque)
1439 {
1440     const VirtIOMEM *vmem = VIRTIO_MEM(rdm);
1441     struct VirtIOMEMReplayData data = {
1442         .fn = replay_fn,
1443         .opaque = opaque,
1444     };
1445 
1446     g_assert(s->mr == &vmem->memdev->mr);
1447     return virtio_mem_for_each_plugged_section(vmem, s, &data,
1448                                             virtio_mem_rdm_replay_populated_cb);
1449 }
1450 
1451 static int virtio_mem_rdm_replay_discarded_cb(MemoryRegionSection *s,
1452                                               void *arg)
1453 {
1454     struct VirtIOMEMReplayData *data = arg;
1455 
1456     ((ReplayRamDiscard)data->fn)(s, data->opaque);
1457     return 0;
1458 }
1459 
1460 static void virtio_mem_rdm_replay_discarded(const RamDiscardManager *rdm,
1461                                             MemoryRegionSection *s,
1462                                             ReplayRamDiscard replay_fn,
1463                                             void *opaque)
1464 {
1465     const VirtIOMEM *vmem = VIRTIO_MEM(rdm);
1466     struct VirtIOMEMReplayData data = {
1467         .fn = replay_fn,
1468         .opaque = opaque,
1469     };
1470 
1471     g_assert(s->mr == &vmem->memdev->mr);
1472     virtio_mem_for_each_unplugged_section(vmem, s, &data,
1473                                           virtio_mem_rdm_replay_discarded_cb);
1474 }
1475 
1476 static void virtio_mem_rdm_register_listener(RamDiscardManager *rdm,
1477                                              RamDiscardListener *rdl,
1478                                              MemoryRegionSection *s)
1479 {
1480     VirtIOMEM *vmem = VIRTIO_MEM(rdm);
1481     int ret;
1482 
1483     g_assert(s->mr == &vmem->memdev->mr);
1484     rdl->section = memory_region_section_new_copy(s);
1485 
1486     QLIST_INSERT_HEAD(&vmem->rdl_list, rdl, next);
1487     ret = virtio_mem_for_each_plugged_section(vmem, rdl->section, rdl,
1488                                               virtio_mem_notify_populate_cb);
1489     if (ret) {
1490         error_report("%s: Replaying plugged ranges failed: %s", __func__,
1491                      strerror(-ret));
1492     }
1493 }
1494 
1495 static void virtio_mem_rdm_unregister_listener(RamDiscardManager *rdm,
1496                                                RamDiscardListener *rdl)
1497 {
1498     VirtIOMEM *vmem = VIRTIO_MEM(rdm);
1499 
1500     g_assert(rdl->section->mr == &vmem->memdev->mr);
1501     if (vmem->size) {
1502         if (rdl->double_discard_supported) {
1503             rdl->notify_discard(rdl, rdl->section);
1504         } else {
1505             virtio_mem_for_each_plugged_section(vmem, rdl->section, rdl,
1506                                                 virtio_mem_notify_discard_cb);
1507         }
1508     }
1509 
1510     memory_region_section_free_copy(rdl->section);
1511     rdl->section = NULL;
1512     QLIST_REMOVE(rdl, next);
1513 }
1514 
1515 static void virtio_mem_unplug_request_check(VirtIOMEM *vmem, Error **errp)
1516 {
1517     if (vmem->unplugged_inaccessible == ON_OFF_AUTO_OFF) {
1518         /*
1519          * We could allow it with a usable region size of 0, but let's just
1520          * not care about that legacy setting.
1521          */
1522         error_setg(errp, "virtio-mem device cannot get unplugged while"
1523                    " '" VIRTIO_MEM_UNPLUGGED_INACCESSIBLE_PROP "' != 'on'");
1524         return;
1525     }
1526 
1527     if (vmem->size) {
1528         error_setg(errp, "virtio-mem device cannot get unplugged while"
1529                    " '" VIRTIO_MEM_SIZE_PROP "' != '0'");
1530         return;
1531     }
1532     if (vmem->requested_size) {
1533         error_setg(errp, "virtio-mem device cannot get unplugged while"
1534                    " '" VIRTIO_MEM_REQUESTED_SIZE_PROP "' != '0'");
1535         return;
1536     }
1537 }
1538 
1539 static void virtio_mem_class_init(ObjectClass *klass, void *data)
1540 {
1541     DeviceClass *dc = DEVICE_CLASS(klass);
1542     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1543     VirtIOMEMClass *vmc = VIRTIO_MEM_CLASS(klass);
1544     RamDiscardManagerClass *rdmc = RAM_DISCARD_MANAGER_CLASS(klass);
1545 
1546     device_class_set_props(dc, virtio_mem_properties);
1547     dc->vmsd = &vmstate_virtio_mem;
1548 
1549     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1550     vdc->realize = virtio_mem_device_realize;
1551     vdc->unrealize = virtio_mem_device_unrealize;
1552     vdc->get_config = virtio_mem_get_config;
1553     vdc->get_features = virtio_mem_get_features;
1554     vdc->validate_features = virtio_mem_validate_features;
1555     vdc->vmsd = &vmstate_virtio_mem_device;
1556 
1557     vmc->fill_device_info = virtio_mem_fill_device_info;
1558     vmc->get_memory_region = virtio_mem_get_memory_region;
1559     vmc->add_size_change_notifier = virtio_mem_add_size_change_notifier;
1560     vmc->remove_size_change_notifier = virtio_mem_remove_size_change_notifier;
1561     vmc->unplug_request_check = virtio_mem_unplug_request_check;
1562 
1563     rdmc->get_min_granularity = virtio_mem_rdm_get_min_granularity;
1564     rdmc->is_populated = virtio_mem_rdm_is_populated;
1565     rdmc->replay_populated = virtio_mem_rdm_replay_populated;
1566     rdmc->replay_discarded = virtio_mem_rdm_replay_discarded;
1567     rdmc->register_listener = virtio_mem_rdm_register_listener;
1568     rdmc->unregister_listener = virtio_mem_rdm_unregister_listener;
1569 }
1570 
1571 static const TypeInfo virtio_mem_info = {
1572     .name = TYPE_VIRTIO_MEM,
1573     .parent = TYPE_VIRTIO_DEVICE,
1574     .instance_size = sizeof(VirtIOMEM),
1575     .instance_init = virtio_mem_instance_init,
1576     .class_init = virtio_mem_class_init,
1577     .class_size = sizeof(VirtIOMEMClass),
1578     .interfaces = (InterfaceInfo[]) {
1579         { TYPE_RAM_DISCARD_MANAGER },
1580         { }
1581     },
1582 };
1583 
1584 static void virtio_register_types(void)
1585 {
1586     type_register_static(&virtio_mem_info);
1587 }
1588 
1589 type_init(virtio_register_types)
1590