xref: /openbmc/qemu/include/exec/ram_addr.h (revision 4fe6d78b)
1 /*
2  * Declarations for cpu physical memory functions
3  *
4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
5  *
6  * Authors:
7  *  Avi Kivity <avi@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2 or
10  * later.  See the COPYING file in the top-level directory.
11  *
12  */
13 
14 /*
15  * This header is for use by exec.c and memory.c ONLY.  Do not include it.
16  * The functions declared here will be removed soon.
17  */
18 
19 #ifndef RAM_ADDR_H
20 #define RAM_ADDR_H
21 
22 #ifndef CONFIG_USER_ONLY
23 #include "hw/xen/xen.h"
24 #include "exec/ramlist.h"
25 
26 struct RAMBlock {
27     struct rcu_head rcu;
28     struct MemoryRegion *mr;
29     uint8_t *host;
30     ram_addr_t offset;
31     ram_addr_t used_length;
32     ram_addr_t max_length;
33     void (*resized)(const char*, uint64_t length, void *host);
34     uint32_t flags;
35     /* Protected by iothread lock.  */
36     char idstr[256];
37     /* RCU-enabled, writes protected by the ramlist lock */
38     QLIST_ENTRY(RAMBlock) next;
39     QLIST_HEAD(, RAMBlockNotifier) ramblock_notifiers;
40     int fd;
41     size_t page_size;
42     /* dirty bitmap used during migration */
43     unsigned long *bmap;
44     /* bitmap of pages that haven't been sent even once
45      * only maintained and used in postcopy at the moment
46      * where it's used to send the dirtymap at the start
47      * of the postcopy phase
48      */
49     unsigned long *unsentmap;
50     /* bitmap of already received pages in postcopy */
51     unsigned long *receivedmap;
52 };
53 
54 static inline bool offset_in_ramblock(RAMBlock *b, ram_addr_t offset)
55 {
56     return (b && b->host && offset < b->used_length) ? true : false;
57 }
58 
59 static inline void *ramblock_ptr(RAMBlock *block, ram_addr_t offset)
60 {
61     assert(offset_in_ramblock(block, offset));
62     return (char *)block->host + offset;
63 }
64 
65 static inline unsigned long int ramblock_recv_bitmap_offset(void *host_addr,
66                                                             RAMBlock *rb)
67 {
68     uint64_t host_addr_offset =
69             (uint64_t)(uintptr_t)(host_addr - (void *)rb->host);
70     return host_addr_offset >> TARGET_PAGE_BITS;
71 }
72 
73 long qemu_getrampagesize(void);
74 unsigned long last_ram_page(void);
75 RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
76                                    bool share, const char *mem_path,
77                                    Error **errp);
78 RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, MemoryRegion *mr,
79                                  bool share, int fd,
80                                  Error **errp);
81 RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
82                                   MemoryRegion *mr, Error **errp);
83 RAMBlock *qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr, Error **errp);
84 RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t max_size,
85                                     void (*resized)(const char*,
86                                                     uint64_t length,
87                                                     void *host),
88                                     MemoryRegion *mr, Error **errp);
89 void qemu_ram_free(RAMBlock *block);
90 
91 int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp);
92 
93 #define DIRTY_CLIENTS_ALL     ((1 << DIRTY_MEMORY_NUM) - 1)
94 #define DIRTY_CLIENTS_NOCODE  (DIRTY_CLIENTS_ALL & ~(1 << DIRTY_MEMORY_CODE))
95 
96 static inline bool cpu_physical_memory_get_dirty(ram_addr_t start,
97                                                  ram_addr_t length,
98                                                  unsigned client)
99 {
100     DirtyMemoryBlocks *blocks;
101     unsigned long end, page;
102     unsigned long idx, offset, base;
103     bool dirty = false;
104 
105     assert(client < DIRTY_MEMORY_NUM);
106 
107     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
108     page = start >> TARGET_PAGE_BITS;
109 
110     rcu_read_lock();
111 
112     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
113 
114     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
115     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
116     base = page - offset;
117     while (page < end) {
118         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
119         unsigned long num = next - base;
120         unsigned long found = find_next_bit(blocks->blocks[idx], num, offset);
121         if (found < num) {
122             dirty = true;
123             break;
124         }
125 
126         page = next;
127         idx++;
128         offset = 0;
129         base += DIRTY_MEMORY_BLOCK_SIZE;
130     }
131 
132     rcu_read_unlock();
133 
134     return dirty;
135 }
136 
137 static inline bool cpu_physical_memory_all_dirty(ram_addr_t start,
138                                                  ram_addr_t length,
139                                                  unsigned client)
140 {
141     DirtyMemoryBlocks *blocks;
142     unsigned long end, page;
143     unsigned long idx, offset, base;
144     bool dirty = true;
145 
146     assert(client < DIRTY_MEMORY_NUM);
147 
148     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
149     page = start >> TARGET_PAGE_BITS;
150 
151     rcu_read_lock();
152 
153     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
154 
155     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
156     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
157     base = page - offset;
158     while (page < end) {
159         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
160         unsigned long num = next - base;
161         unsigned long found = find_next_zero_bit(blocks->blocks[idx], num, offset);
162         if (found < num) {
163             dirty = false;
164             break;
165         }
166 
167         page = next;
168         idx++;
169         offset = 0;
170         base += DIRTY_MEMORY_BLOCK_SIZE;
171     }
172 
173     rcu_read_unlock();
174 
175     return dirty;
176 }
177 
178 static inline bool cpu_physical_memory_get_dirty_flag(ram_addr_t addr,
179                                                       unsigned client)
180 {
181     return cpu_physical_memory_get_dirty(addr, 1, client);
182 }
183 
184 static inline bool cpu_physical_memory_is_clean(ram_addr_t addr)
185 {
186     bool vga = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_VGA);
187     bool code = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_CODE);
188     bool migration =
189         cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_MIGRATION);
190     return !(vga && code && migration);
191 }
192 
193 static inline uint8_t cpu_physical_memory_range_includes_clean(ram_addr_t start,
194                                                                ram_addr_t length,
195                                                                uint8_t mask)
196 {
197     uint8_t ret = 0;
198 
199     if (mask & (1 << DIRTY_MEMORY_VGA) &&
200         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_VGA)) {
201         ret |= (1 << DIRTY_MEMORY_VGA);
202     }
203     if (mask & (1 << DIRTY_MEMORY_CODE) &&
204         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_CODE)) {
205         ret |= (1 << DIRTY_MEMORY_CODE);
206     }
207     if (mask & (1 << DIRTY_MEMORY_MIGRATION) &&
208         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_MIGRATION)) {
209         ret |= (1 << DIRTY_MEMORY_MIGRATION);
210     }
211     return ret;
212 }
213 
214 static inline void cpu_physical_memory_set_dirty_flag(ram_addr_t addr,
215                                                       unsigned client)
216 {
217     unsigned long page, idx, offset;
218     DirtyMemoryBlocks *blocks;
219 
220     assert(client < DIRTY_MEMORY_NUM);
221 
222     page = addr >> TARGET_PAGE_BITS;
223     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
224     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
225 
226     rcu_read_lock();
227 
228     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
229 
230     set_bit_atomic(offset, blocks->blocks[idx]);
231 
232     rcu_read_unlock();
233 }
234 
235 static inline void cpu_physical_memory_set_dirty_range(ram_addr_t start,
236                                                        ram_addr_t length,
237                                                        uint8_t mask)
238 {
239     DirtyMemoryBlocks *blocks[DIRTY_MEMORY_NUM];
240     unsigned long end, page;
241     unsigned long idx, offset, base;
242     int i;
243 
244     if (!mask && !xen_enabled()) {
245         return;
246     }
247 
248     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
249     page = start >> TARGET_PAGE_BITS;
250 
251     rcu_read_lock();
252 
253     for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
254         blocks[i] = atomic_rcu_read(&ram_list.dirty_memory[i]);
255     }
256 
257     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
258     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
259     base = page - offset;
260     while (page < end) {
261         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
262 
263         if (likely(mask & (1 << DIRTY_MEMORY_MIGRATION))) {
264             bitmap_set_atomic(blocks[DIRTY_MEMORY_MIGRATION]->blocks[idx],
265                               offset, next - page);
266         }
267         if (unlikely(mask & (1 << DIRTY_MEMORY_VGA))) {
268             bitmap_set_atomic(blocks[DIRTY_MEMORY_VGA]->blocks[idx],
269                               offset, next - page);
270         }
271         if (unlikely(mask & (1 << DIRTY_MEMORY_CODE))) {
272             bitmap_set_atomic(blocks[DIRTY_MEMORY_CODE]->blocks[idx],
273                               offset, next - page);
274         }
275 
276         page = next;
277         idx++;
278         offset = 0;
279         base += DIRTY_MEMORY_BLOCK_SIZE;
280     }
281 
282     rcu_read_unlock();
283 
284     xen_hvm_modified_memory(start, length);
285 }
286 
287 #if !defined(_WIN32)
288 static inline void cpu_physical_memory_set_dirty_lebitmap(unsigned long *bitmap,
289                                                           ram_addr_t start,
290                                                           ram_addr_t pages)
291 {
292     unsigned long i, j;
293     unsigned long page_number, c;
294     hwaddr addr;
295     ram_addr_t ram_addr;
296     unsigned long len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
297     unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE;
298     unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
299 
300     /* start address is aligned at the start of a word? */
301     if ((((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) &&
302         (hpratio == 1)) {
303         unsigned long **blocks[DIRTY_MEMORY_NUM];
304         unsigned long idx;
305         unsigned long offset;
306         long k;
307         long nr = BITS_TO_LONGS(pages);
308 
309         idx = (start >> TARGET_PAGE_BITS) / DIRTY_MEMORY_BLOCK_SIZE;
310         offset = BIT_WORD((start >> TARGET_PAGE_BITS) %
311                           DIRTY_MEMORY_BLOCK_SIZE);
312 
313         rcu_read_lock();
314 
315         for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
316             blocks[i] = atomic_rcu_read(&ram_list.dirty_memory[i])->blocks;
317         }
318 
319         for (k = 0; k < nr; k++) {
320             if (bitmap[k]) {
321                 unsigned long temp = leul_to_cpu(bitmap[k]);
322 
323                 atomic_or(&blocks[DIRTY_MEMORY_MIGRATION][idx][offset], temp);
324                 atomic_or(&blocks[DIRTY_MEMORY_VGA][idx][offset], temp);
325                 if (tcg_enabled()) {
326                     atomic_or(&blocks[DIRTY_MEMORY_CODE][idx][offset], temp);
327                 }
328             }
329 
330             if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
331                 offset = 0;
332                 idx++;
333             }
334         }
335 
336         rcu_read_unlock();
337 
338         xen_hvm_modified_memory(start, pages << TARGET_PAGE_BITS);
339     } else {
340         uint8_t clients = tcg_enabled() ? DIRTY_CLIENTS_ALL : DIRTY_CLIENTS_NOCODE;
341         /*
342          * bitmap-traveling is faster than memory-traveling (for addr...)
343          * especially when most of the memory is not dirty.
344          */
345         for (i = 0; i < len; i++) {
346             if (bitmap[i] != 0) {
347                 c = leul_to_cpu(bitmap[i]);
348                 do {
349                     j = ctzl(c);
350                     c &= ~(1ul << j);
351                     page_number = (i * HOST_LONG_BITS + j) * hpratio;
352                     addr = page_number * TARGET_PAGE_SIZE;
353                     ram_addr = start + addr;
354                     cpu_physical_memory_set_dirty_range(ram_addr,
355                                        TARGET_PAGE_SIZE * hpratio, clients);
356                 } while (c != 0);
357             }
358         }
359     }
360 }
361 #endif /* not _WIN32 */
362 
363 bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start,
364                                               ram_addr_t length,
365                                               unsigned client);
366 
367 DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty
368     (ram_addr_t start, ram_addr_t length, unsigned client);
369 
370 bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap,
371                                             ram_addr_t start,
372                                             ram_addr_t length);
373 
374 static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start,
375                                                          ram_addr_t length)
376 {
377     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_MIGRATION);
378     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_VGA);
379     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_CODE);
380 }
381 
382 
383 static inline
384 uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb,
385                                                ram_addr_t start,
386                                                ram_addr_t length,
387                                                uint64_t *real_dirty_pages)
388 {
389     ram_addr_t addr;
390     unsigned long word = BIT_WORD((start + rb->offset) >> TARGET_PAGE_BITS);
391     uint64_t num_dirty = 0;
392     unsigned long *dest = rb->bmap;
393 
394     /* start address and length is aligned at the start of a word? */
395     if (((word * BITS_PER_LONG) << TARGET_PAGE_BITS) ==
396          (start + rb->offset) &&
397         !(length & ((BITS_PER_LONG << TARGET_PAGE_BITS) - 1))) {
398         int k;
399         int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
400         unsigned long * const *src;
401         unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE;
402         unsigned long offset = BIT_WORD((word * BITS_PER_LONG) %
403                                         DIRTY_MEMORY_BLOCK_SIZE);
404         unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
405 
406         rcu_read_lock();
407 
408         src = atomic_rcu_read(
409                 &ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION])->blocks;
410 
411         for (k = page; k < page + nr; k++) {
412             if (src[idx][offset]) {
413                 unsigned long bits = atomic_xchg(&src[idx][offset], 0);
414                 unsigned long new_dirty;
415                 *real_dirty_pages += ctpopl(bits);
416                 new_dirty = ~dest[k];
417                 dest[k] |= bits;
418                 new_dirty &= bits;
419                 num_dirty += ctpopl(new_dirty);
420             }
421 
422             if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
423                 offset = 0;
424                 idx++;
425             }
426         }
427 
428         rcu_read_unlock();
429     } else {
430         ram_addr_t offset = rb->offset;
431 
432         for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
433             if (cpu_physical_memory_test_and_clear_dirty(
434                         start + addr + offset,
435                         TARGET_PAGE_SIZE,
436                         DIRTY_MEMORY_MIGRATION)) {
437                 *real_dirty_pages += 1;
438                 long k = (start + addr) >> TARGET_PAGE_BITS;
439                 if (!test_and_set_bit(k, dest)) {
440                     num_dirty++;
441                 }
442             }
443         }
444     }
445 
446     return num_dirty;
447 }
448 #endif
449 #endif
450