xref: /openbmc/qemu/hw/mem/memory-device.c (revision b91a0fa7)
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     Error *err = NULL;
103     GSList *list = NULL, *item;
104     Range as, new = range_empty;
105 
106     if (!ms->device_memory) {
107         error_setg(errp, "memory devices (e.g. for memory hotplug) are not "
108                          "supported by the machine");
109         return 0;
110     }
111 
112     if (!memory_region_size(&ms->device_memory->mr)) {
113         error_setg(errp, "memory devices (e.g. for memory hotplug) are not "
114                          "enabled, please specify the maxmem option");
115         return 0;
116     }
117     range_init_nofail(&as, ms->device_memory->base,
118                       memory_region_size(&ms->device_memory->mr));
119 
120     /* start of address space indicates the maximum alignment we expect */
121     if (!QEMU_IS_ALIGNED(range_lob(&as), align)) {
122         warn_report("the alignment (0x%" PRIx64 ") exceeds the expected"
123                     " maximum alignment, memory will get fragmented and not"
124                     " all 'maxmem' might be usable for memory devices.",
125                     align);
126     }
127 
128     memory_device_check_addable(ms, size, &err);
129     if (err) {
130         error_propagate(errp, err);
131         return 0;
132     }
133 
134     if (hint && !QEMU_IS_ALIGNED(*hint, align)) {
135         error_setg(errp, "address must be aligned to 0x%" PRIx64 " bytes",
136                    align);
137         return 0;
138     }
139 
140     if (!QEMU_IS_ALIGNED(size, align)) {
141         error_setg(errp, "backend memory size must be multiple of 0x%"
142                    PRIx64, align);
143         return 0;
144     }
145 
146     if (hint) {
147         if (range_init(&new, *hint, size) || !range_contains_range(&as, &new)) {
148             error_setg(errp, "can't add memory device [0x%" PRIx64 ":0x%" PRIx64
149                        "], usable range for memory devices [0x%" PRIx64 ":0x%"
150                        PRIx64 "]", *hint, size, range_lob(&as),
151                        range_size(&as));
152             return 0;
153         }
154     } else {
155         if (range_init(&new, QEMU_ALIGN_UP(range_lob(&as), align), size)) {
156             error_setg(errp, "can't add memory device, device too big");
157             return 0;
158         }
159     }
160 
161     /* find address range that will fit new memory device */
162     object_child_foreach(OBJECT(ms), memory_device_build_list, &list);
163     for (item = list; item; item = g_slist_next(item)) {
164         const MemoryDeviceState *md = item->data;
165         const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(OBJECT(md));
166         uint64_t next_addr;
167         Range tmp;
168 
169         range_init_nofail(&tmp, mdc->get_addr(md),
170                           memory_device_get_region_size(md, &error_abort));
171 
172         if (range_overlaps_range(&tmp, &new)) {
173             if (hint) {
174                 const DeviceState *d = DEVICE(md);
175                 error_setg(errp, "address range conflicts with memory device"
176                            " id='%s'", d->id ? d->id : "(unnamed)");
177                 goto out;
178             }
179 
180             next_addr = QEMU_ALIGN_UP(range_upb(&tmp) + 1, align);
181             if (!next_addr || range_init(&new, next_addr, range_size(&new))) {
182                 range_make_empty(&new);
183                 break;
184             }
185         } else if (range_lob(&tmp) > range_upb(&new)) {
186             break;
187         }
188     }
189 
190     if (!range_contains_range(&as, &new)) {
191         error_setg(errp, "could not find position in guest address space for "
192                    "memory device - memory fragmented due to alignments");
193     }
194 out:
195     g_slist_free(list);
196     return range_lob(&new);
197 }
198 
199 MemoryDeviceInfoList *qmp_memory_device_list(void)
200 {
201     GSList *devices = NULL, *item;
202     MemoryDeviceInfoList *list = NULL, **tail = &list;
203 
204     object_child_foreach(qdev_get_machine(), memory_device_build_list,
205                          &devices);
206 
207     for (item = devices; item; item = g_slist_next(item)) {
208         const MemoryDeviceState *md = MEMORY_DEVICE(item->data);
209         const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(item->data);
210         MemoryDeviceInfo *info = g_new0(MemoryDeviceInfo, 1);
211 
212         mdc->fill_device_info(md, info);
213 
214         QAPI_LIST_APPEND(tail, info);
215     }
216 
217     g_slist_free(devices);
218 
219     return list;
220 }
221 
222 static int memory_device_plugged_size(Object *obj, void *opaque)
223 {
224     uint64_t *size = opaque;
225 
226     if (object_dynamic_cast(obj, TYPE_MEMORY_DEVICE)) {
227         const DeviceState *dev = DEVICE(obj);
228         const MemoryDeviceState *md = MEMORY_DEVICE(obj);
229         const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(obj);
230 
231         if (dev->realized) {
232             *size += mdc->get_plugged_size(md, &error_abort);
233         }
234     }
235 
236     object_child_foreach(obj, memory_device_plugged_size, opaque);
237     return 0;
238 }
239 
240 uint64_t get_plugged_memory_size(void)
241 {
242     uint64_t size = 0;
243 
244     memory_device_plugged_size(qdev_get_machine(), &size);
245 
246     return size;
247 }
248 
249 void memory_device_pre_plug(MemoryDeviceState *md, MachineState *ms,
250                             const uint64_t *legacy_align, Error **errp)
251 {
252     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
253     Error *local_err = NULL;
254     uint64_t addr, align = 0;
255     MemoryRegion *mr;
256 
257     mr = mdc->get_memory_region(md, &local_err);
258     if (local_err) {
259         goto out;
260     }
261 
262     if (legacy_align) {
263         align = *legacy_align;
264     } else {
265         if (mdc->get_min_alignment) {
266             align = mdc->get_min_alignment(md);
267         }
268         align = MAX(align, memory_region_get_alignment(mr));
269     }
270     addr = mdc->get_addr(md);
271     addr = memory_device_get_free_addr(ms, !addr ? NULL : &addr, align,
272                                        memory_region_size(mr), &local_err);
273     if (local_err) {
274         goto out;
275     }
276     mdc->set_addr(md, addr, &local_err);
277     if (!local_err) {
278         trace_memory_device_pre_plug(DEVICE(md)->id ? DEVICE(md)->id : "",
279                                      addr);
280     }
281 out:
282     error_propagate(errp, local_err);
283 }
284 
285 void memory_device_plug(MemoryDeviceState *md, MachineState *ms)
286 {
287     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
288     const uint64_t addr = mdc->get_addr(md);
289     MemoryRegion *mr;
290 
291     /*
292      * We expect that a previous call to memory_device_pre_plug() succeeded, so
293      * it can't fail at this point.
294      */
295     mr = mdc->get_memory_region(md, &error_abort);
296     g_assert(ms->device_memory);
297 
298     memory_region_add_subregion(&ms->device_memory->mr,
299                                 addr - ms->device_memory->base, mr);
300     trace_memory_device_plug(DEVICE(md)->id ? DEVICE(md)->id : "", addr);
301 }
302 
303 void memory_device_unplug(MemoryDeviceState *md, MachineState *ms)
304 {
305     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
306     MemoryRegion *mr;
307 
308     /*
309      * We expect that a previous call to memory_device_pre_plug() succeeded, so
310      * it can't fail at this point.
311      */
312     mr = mdc->get_memory_region(md, &error_abort);
313     g_assert(ms->device_memory);
314 
315     memory_region_del_subregion(&ms->device_memory->mr, mr);
316     trace_memory_device_unplug(DEVICE(md)->id ? DEVICE(md)->id : "",
317                                mdc->get_addr(md));
318 }
319 
320 uint64_t memory_device_get_region_size(const MemoryDeviceState *md,
321                                        Error **errp)
322 {
323     const MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(md);
324     MemoryRegion *mr;
325 
326     /* dropping const here is fine as we don't touch the memory region */
327     mr = mdc->get_memory_region((MemoryDeviceState *)md, errp);
328     if (!mr) {
329         return 0;
330     }
331 
332     return memory_region_size(mr);
333 }
334 
335 static const TypeInfo memory_device_info = {
336     .name          = TYPE_MEMORY_DEVICE,
337     .parent        = TYPE_INTERFACE,
338     .class_size = sizeof(MemoryDeviceClass),
339 };
340 
341 static void memory_device_register_types(void)
342 {
343     type_register_static(&memory_device_info);
344 }
345 
346 type_init(memory_device_register_types)
347