xref: /openbmc/qemu/hw/mem/memory-device.c (revision 3a6606c7)
1 /*
2  * Memory Device Interface
3  *
4  * Copyright ProfitBricks GmbH 2012
5  * Copyright (C) 2014 Red Hat Inc
6  * Copyright (c) 2018 Red Hat Inc
7  *
8  * This work is licensed under the terms of the GNU GPL, version 2 or later.
9  * See the COPYING file in the top-level directory.
10  */
11 
12 #include "qemu/osdep.h"
13 #include "hw/mem/memory-device.h"
14 #include "qapi/error.h"
15 #include "hw/boards.h"
16 #include "qemu/range.h"
17 #include "hw/virtio/vhost.h"
18 #include "sysemu/kvm.h"
19 #include "trace.h"
20 
21 static gint memory_device_addr_sort(gconstpointer a, gconstpointer b)
22 {
23     const MemoryDeviceState *md_a = MEMORY_DEVICE(a);
24     const MemoryDeviceState *md_b = MEMORY_DEVICE(b);
25     const MemoryDeviceClass *mdc_a = MEMORY_DEVICE_GET_CLASS(a);
26     const MemoryDeviceClass *mdc_b = MEMORY_DEVICE_GET_CLASS(b);
27     const uint64_t addr_a = mdc_a->get_addr(md_a);
28     const uint64_t addr_b = mdc_b->get_addr(md_b);
29 
30     if (addr_a > addr_b) {
31         return 1;
32     } else if (addr_a < addr_b) {
33         return -1;
34     }
35     return 0;
36 }
37 
38 static int memory_device_build_list(Object *obj, void *opaque)
39 {
40     GSList **list = opaque;
41 
42     if (object_dynamic_cast(obj, TYPE_MEMORY_DEVICE)) {
43         DeviceState *dev = DEVICE(obj);
44         if (dev->realized) { /* only realized memory devices matter */
45             *list = g_slist_insert_sorted(*list, dev, memory_device_addr_sort);
46         }
47     }
48 
49     object_child_foreach(obj, memory_device_build_list, opaque);
50     return 0;
51 }
52 
53 static int memory_device_used_region_size(Object *obj, void *opaque)
54 {
55     uint64_t *size = opaque;
56 
57     if (object_dynamic_cast(obj, TYPE_MEMORY_DEVICE)) {
58         const DeviceState *dev = DEVICE(obj);
59         const MemoryDeviceState *md = MEMORY_DEVICE(obj);
60 
61         if (dev->realized) {
62             *size += memory_device_get_region_size(md, &error_abort);
63         }
64     }
65 
66     object_child_foreach(obj, memory_device_used_region_size, opaque);
67     return 0;
68 }
69 
70 static void memory_device_check_addable(MachineState *ms, uint64_t size,
71                                         Error **errp)
72 {
73     uint64_t used_region_size = 0;
74 
75     /* we will need a new memory slot for kvm and vhost */
76     if (kvm_enabled() && !kvm_has_free_slot(ms)) {
77         error_setg(errp, "hypervisor has no free memory slots left");
78         return;
79     }
80     if (!vhost_has_free_slot()) {
81         error_setg(errp, "a used vhost backend has no free memory slots left");
82         return;
83     }
84 
85     /* will we exceed the total amount of memory specified */
86     memory_device_used_region_size(OBJECT(ms), &used_region_size);
87     if (used_region_size + size < used_region_size ||
88         used_region_size + size > ms->maxram_size - ms->ram_size) {
89         error_setg(errp, "not enough space, currently 0x%" PRIx64
90                    " in use of total space for memory devices 0x" RAM_ADDR_FMT,
91                    used_region_size, ms->maxram_size - ms->ram_size);
92         return;
93     }
94 
95 }
96 
97 static uint64_t memory_device_get_free_addr(MachineState *ms,
98                                             const uint64_t *hint,
99                                             uint64_t align, uint64_t size,
100                                             Error **errp)
101 {
102     GSList *list = NULL, *item;
103     Range as, new = range_empty;
104 
105     if (!ms->device_memory) {
106         error_setg(errp, "memory devices (e.g. for memory hotplug) are not "
107                          "supported by the machine");
108         return 0;
109     }
110 
111     if (!memory_region_size(&ms->device_memory->mr)) {
112         error_setg(errp, "memory devices (e.g. for memory hotplug) are not "
113                          "enabled, please specify the maxmem option");
114         return 0;
115     }
116     range_init_nofail(&as, ms->device_memory->base,
117                       memory_region_size(&ms->device_memory->mr));
118 
119     /* start of address space indicates the maximum alignment we expect */
120     if (!QEMU_IS_ALIGNED(range_lob(&as), align)) {
121         error_setg(errp, "the alignment (0x%" PRIx64 ") is not supported",
122                    align);
123         return 0;
124     }
125 
126     memory_device_check_addable(ms, size, errp);
127     if (*errp) {
128         return 0;
129     }
130 
131     if (hint && !QEMU_IS_ALIGNED(*hint, align)) {
132         error_setg(errp, "address must be aligned to 0x%" PRIx64 " bytes",
133                    align);
134         return 0;
135     }
136 
137     if (!QEMU_IS_ALIGNED(size, align)) {
138         error_setg(errp, "backend memory size must be multiple of 0x%"
139                    PRIx64, align);
140         return 0;
141     }
142 
143     if (hint) {
144         if (range_init(&new, *hint, size) || !range_contains_range(&as, &new)) {
145             error_setg(errp, "can't add memory device [0x%" PRIx64 ":0x%" PRIx64
146                        "], usable range for memory devices [0x%" PRIx64 ":0x%"
147                        PRIx64 "]", *hint, size, range_lob(&as),
148                        range_size(&as));
149             return 0;
150         }
151     } else {
152         if (range_init(&new, range_lob(&as), size)) {
153             error_setg(errp, "can't add memory device, device too big");
154             return 0;
155         }
156     }
157 
158     /* find address range that will fit new memory device */
159     object_child_foreach(OBJECT(ms), memory_device_build_list, &list);
160     for (item = list; item; item = g_slist_next(item)) {
161         const MemoryDeviceState *md = item->data;
162         const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(OBJECT(md));
163         uint64_t next_addr;
164         Range tmp;
165 
166         range_init_nofail(&tmp, mdc->get_addr(md),
167                           memory_device_get_region_size(md, &error_abort));
168 
169         if (range_overlaps_range(&tmp, &new)) {
170             if (hint) {
171                 const DeviceState *d = DEVICE(md);
172                 error_setg(errp, "address range conflicts with memory device"
173                            " id='%s'", d->id ? d->id : "(unnamed)");
174                 goto out;
175             }
176 
177             next_addr = QEMU_ALIGN_UP(range_upb(&tmp) + 1, align);
178             if (!next_addr || range_init(&new, next_addr, range_size(&new))) {
179                 range_make_empty(&new);
180                 break;
181             }
182         } else if (range_lob(&tmp) > range_upb(&new)) {
183             break;
184         }
185     }
186 
187     if (!range_contains_range(&as, &new)) {
188         error_setg(errp, "could not find position in guest address space for "
189                    "memory device - memory fragmented due to alignments");
190     }
191 out:
192     g_slist_free(list);
193     return range_lob(&new);
194 }
195 
196 MemoryDeviceInfoList *qmp_memory_device_list(void)
197 {
198     GSList *devices = NULL, *item;
199     MemoryDeviceInfoList *list = NULL, *prev = NULL;
200 
201     object_child_foreach(qdev_get_machine(), memory_device_build_list,
202                          &devices);
203 
204     for (item = devices; item; item = g_slist_next(item)) {
205         const MemoryDeviceState *md = MEMORY_DEVICE(item->data);
206         const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(item->data);
207         MemoryDeviceInfoList *elem = g_new0(MemoryDeviceInfoList, 1);
208         MemoryDeviceInfo *info = g_new0(MemoryDeviceInfo, 1);
209 
210         mdc->fill_device_info(md, info);
211 
212         elem->value = info;
213         elem->next = NULL;
214         if (prev) {
215             prev->next = elem;
216         } else {
217             list = elem;
218         }
219         prev = elem;
220     }
221 
222     g_slist_free(devices);
223 
224     return list;
225 }
226 
227 static int memory_device_plugged_size(Object *obj, void *opaque)
228 {
229     uint64_t *size = opaque;
230 
231     if (object_dynamic_cast(obj, TYPE_MEMORY_DEVICE)) {
232         const DeviceState *dev = DEVICE(obj);
233         const MemoryDeviceState *md = MEMORY_DEVICE(obj);
234         const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(obj);
235 
236         if (dev->realized) {
237             *size += mdc->get_plugged_size(md, &error_abort);
238         }
239     }
240 
241     object_child_foreach(obj, memory_device_plugged_size, opaque);
242     return 0;
243 }
244 
245 uint64_t get_plugged_memory_size(void)
246 {
247     uint64_t size = 0;
248 
249     memory_device_plugged_size(qdev_get_machine(), &size);
250 
251     return size;
252 }
253 
254 void memory_device_pre_plug(MemoryDeviceState *md, MachineState *ms,
255                             const uint64_t *legacy_align, Error **errp)
256 {
257     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
258     Error *local_err = NULL;
259     uint64_t addr, align;
260     MemoryRegion *mr;
261 
262     mr = mdc->get_memory_region(md, &local_err);
263     if (local_err) {
264         goto out;
265     }
266 
267     align = legacy_align ? *legacy_align : memory_region_get_alignment(mr);
268     addr = mdc->get_addr(md);
269     addr = memory_device_get_free_addr(ms, !addr ? NULL : &addr, align,
270                                        memory_region_size(mr), &local_err);
271     if (local_err) {
272         goto out;
273     }
274     mdc->set_addr(md, addr, &local_err);
275     if (!local_err) {
276         trace_memory_device_pre_plug(DEVICE(md)->id ? DEVICE(md)->id : "",
277                                      addr);
278     }
279 out:
280     error_propagate(errp, local_err);
281 }
282 
283 void memory_device_plug(MemoryDeviceState *md, MachineState *ms)
284 {
285     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
286     const uint64_t addr = mdc->get_addr(md);
287     MemoryRegion *mr;
288 
289     /*
290      * We expect that a previous call to memory_device_pre_plug() succeeded, so
291      * it can't fail at this point.
292      */
293     mr = mdc->get_memory_region(md, &error_abort);
294     g_assert(ms->device_memory);
295 
296     memory_region_add_subregion(&ms->device_memory->mr,
297                                 addr - ms->device_memory->base, mr);
298     trace_memory_device_plug(DEVICE(md)->id ? DEVICE(md)->id : "", addr);
299 }
300 
301 void memory_device_unplug(MemoryDeviceState *md, MachineState *ms)
302 {
303     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
304     MemoryRegion *mr;
305 
306     /*
307      * We expect that a previous call to memory_device_pre_plug() succeeded, so
308      * it can't fail at this point.
309      */
310     mr = mdc->get_memory_region(md, &error_abort);
311     g_assert(ms->device_memory);
312 
313     memory_region_del_subregion(&ms->device_memory->mr, mr);
314     trace_memory_device_unplug(DEVICE(md)->id ? DEVICE(md)->id : "",
315                                mdc->get_addr(md));
316 }
317 
318 uint64_t memory_device_get_region_size(const MemoryDeviceState *md,
319                                        Error **errp)
320 {
321     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
322     MemoryRegion *mr;
323 
324     /* dropping const here is fine as we don't touch the memory region */
325     mr = mdc->get_memory_region((MemoryDeviceState *)md, errp);
326     if (!mr) {
327         return 0;
328     }
329 
330     return memory_region_size(mr);
331 }
332 
333 static const TypeInfo memory_device_info = {
334     .name          = TYPE_MEMORY_DEVICE,
335     .parent        = TYPE_INTERFACE,
336     .class_size = sizeof(MemoryDeviceClass),
337 };
338 
339 static void memory_device_register_types(void)
340 {
341     type_register_static(&memory_device_info);
342 }
343 
344 type_init(memory_device_register_types)
345