xref: /openbmc/qemu/include/exec/ram_addr.h (revision fe7f9b8e)
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, bool share, MemoryRegion *mr,
84                          Error **errp);
85 RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t max_size,
86                                     void (*resized)(const char*,
87                                                     uint64_t length,
88                                                     void *host),
89                                     MemoryRegion *mr, Error **errp);
90 void qemu_ram_free(RAMBlock *block);
91 
92 int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp);
93 
94 #define DIRTY_CLIENTS_ALL     ((1 << DIRTY_MEMORY_NUM) - 1)
95 #define DIRTY_CLIENTS_NOCODE  (DIRTY_CLIENTS_ALL & ~(1 << DIRTY_MEMORY_CODE))
96 
97 static inline bool cpu_physical_memory_get_dirty(ram_addr_t start,
98                                                  ram_addr_t length,
99                                                  unsigned client)
100 {
101     DirtyMemoryBlocks *blocks;
102     unsigned long end, page;
103     unsigned long idx, offset, base;
104     bool dirty = false;
105 
106     assert(client < DIRTY_MEMORY_NUM);
107 
108     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
109     page = start >> TARGET_PAGE_BITS;
110 
111     rcu_read_lock();
112 
113     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
114 
115     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
116     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
117     base = page - offset;
118     while (page < end) {
119         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
120         unsigned long num = next - base;
121         unsigned long found = find_next_bit(blocks->blocks[idx], num, offset);
122         if (found < num) {
123             dirty = true;
124             break;
125         }
126 
127         page = next;
128         idx++;
129         offset = 0;
130         base += DIRTY_MEMORY_BLOCK_SIZE;
131     }
132 
133     rcu_read_unlock();
134 
135     return dirty;
136 }
137 
138 static inline bool cpu_physical_memory_all_dirty(ram_addr_t start,
139                                                  ram_addr_t length,
140                                                  unsigned client)
141 {
142     DirtyMemoryBlocks *blocks;
143     unsigned long end, page;
144     unsigned long idx, offset, base;
145     bool dirty = true;
146 
147     assert(client < DIRTY_MEMORY_NUM);
148 
149     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
150     page = start >> TARGET_PAGE_BITS;
151 
152     rcu_read_lock();
153 
154     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
155 
156     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
157     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
158     base = page - offset;
159     while (page < end) {
160         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
161         unsigned long num = next - base;
162         unsigned long found = find_next_zero_bit(blocks->blocks[idx], num, offset);
163         if (found < num) {
164             dirty = false;
165             break;
166         }
167 
168         page = next;
169         idx++;
170         offset = 0;
171         base += DIRTY_MEMORY_BLOCK_SIZE;
172     }
173 
174     rcu_read_unlock();
175 
176     return dirty;
177 }
178 
179 static inline bool cpu_physical_memory_get_dirty_flag(ram_addr_t addr,
180                                                       unsigned client)
181 {
182     return cpu_physical_memory_get_dirty(addr, 1, client);
183 }
184 
185 static inline bool cpu_physical_memory_is_clean(ram_addr_t addr)
186 {
187     bool vga = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_VGA);
188     bool code = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_CODE);
189     bool migration =
190         cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_MIGRATION);
191     return !(vga && code && migration);
192 }
193 
194 static inline uint8_t cpu_physical_memory_range_includes_clean(ram_addr_t start,
195                                                                ram_addr_t length,
196                                                                uint8_t mask)
197 {
198     uint8_t ret = 0;
199 
200     if (mask & (1 << DIRTY_MEMORY_VGA) &&
201         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_VGA)) {
202         ret |= (1 << DIRTY_MEMORY_VGA);
203     }
204     if (mask & (1 << DIRTY_MEMORY_CODE) &&
205         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_CODE)) {
206         ret |= (1 << DIRTY_MEMORY_CODE);
207     }
208     if (mask & (1 << DIRTY_MEMORY_MIGRATION) &&
209         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_MIGRATION)) {
210         ret |= (1 << DIRTY_MEMORY_MIGRATION);
211     }
212     return ret;
213 }
214 
215 static inline void cpu_physical_memory_set_dirty_flag(ram_addr_t addr,
216                                                       unsigned client)
217 {
218     unsigned long page, idx, offset;
219     DirtyMemoryBlocks *blocks;
220 
221     assert(client < DIRTY_MEMORY_NUM);
222 
223     page = addr >> TARGET_PAGE_BITS;
224     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
225     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
226 
227     rcu_read_lock();
228 
229     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
230 
231     set_bit_atomic(offset, blocks->blocks[idx]);
232 
233     rcu_read_unlock();
234 }
235 
236 static inline void cpu_physical_memory_set_dirty_range(ram_addr_t start,
237                                                        ram_addr_t length,
238                                                        uint8_t mask)
239 {
240     DirtyMemoryBlocks *blocks[DIRTY_MEMORY_NUM];
241     unsigned long end, page;
242     unsigned long idx, offset, base;
243     int i;
244 
245     if (!mask && !xen_enabled()) {
246         return;
247     }
248 
249     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
250     page = start >> TARGET_PAGE_BITS;
251 
252     rcu_read_lock();
253 
254     for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
255         blocks[i] = atomic_rcu_read(&ram_list.dirty_memory[i]);
256     }
257 
258     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
259     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
260     base = page - offset;
261     while (page < end) {
262         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
263 
264         if (likely(mask & (1 << DIRTY_MEMORY_MIGRATION))) {
265             bitmap_set_atomic(blocks[DIRTY_MEMORY_MIGRATION]->blocks[idx],
266                               offset, next - page);
267         }
268         if (unlikely(mask & (1 << DIRTY_MEMORY_VGA))) {
269             bitmap_set_atomic(blocks[DIRTY_MEMORY_VGA]->blocks[idx],
270                               offset, next - page);
271         }
272         if (unlikely(mask & (1 << DIRTY_MEMORY_CODE))) {
273             bitmap_set_atomic(blocks[DIRTY_MEMORY_CODE]->blocks[idx],
274                               offset, next - page);
275         }
276 
277         page = next;
278         idx++;
279         offset = 0;
280         base += DIRTY_MEMORY_BLOCK_SIZE;
281     }
282 
283     rcu_read_unlock();
284 
285     xen_hvm_modified_memory(start, length);
286 }
287 
288 #if !defined(_WIN32)
289 static inline void cpu_physical_memory_set_dirty_lebitmap(unsigned long *bitmap,
290                                                           ram_addr_t start,
291                                                           ram_addr_t pages)
292 {
293     unsigned long i, j;
294     unsigned long page_number, c;
295     hwaddr addr;
296     ram_addr_t ram_addr;
297     unsigned long len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
298     unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE;
299     unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
300 
301     /* start address is aligned at the start of a word? */
302     if ((((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) &&
303         (hpratio == 1)) {
304         unsigned long **blocks[DIRTY_MEMORY_NUM];
305         unsigned long idx;
306         unsigned long offset;
307         long k;
308         long nr = BITS_TO_LONGS(pages);
309 
310         idx = (start >> TARGET_PAGE_BITS) / DIRTY_MEMORY_BLOCK_SIZE;
311         offset = BIT_WORD((start >> TARGET_PAGE_BITS) %
312                           DIRTY_MEMORY_BLOCK_SIZE);
313 
314         rcu_read_lock();
315 
316         for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
317             blocks[i] = atomic_rcu_read(&ram_list.dirty_memory[i])->blocks;
318         }
319 
320         for (k = 0; k < nr; k++) {
321             if (bitmap[k]) {
322                 unsigned long temp = leul_to_cpu(bitmap[k]);
323 
324                 atomic_or(&blocks[DIRTY_MEMORY_MIGRATION][idx][offset], temp);
325                 atomic_or(&blocks[DIRTY_MEMORY_VGA][idx][offset], temp);
326                 if (tcg_enabled()) {
327                     atomic_or(&blocks[DIRTY_MEMORY_CODE][idx][offset], temp);
328                 }
329             }
330 
331             if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
332                 offset = 0;
333                 idx++;
334             }
335         }
336 
337         rcu_read_unlock();
338 
339         xen_hvm_modified_memory(start, pages << TARGET_PAGE_BITS);
340     } else {
341         uint8_t clients = tcg_enabled() ? DIRTY_CLIENTS_ALL : DIRTY_CLIENTS_NOCODE;
342         /*
343          * bitmap-traveling is faster than memory-traveling (for addr...)
344          * especially when most of the memory is not dirty.
345          */
346         for (i = 0; i < len; i++) {
347             if (bitmap[i] != 0) {
348                 c = leul_to_cpu(bitmap[i]);
349                 do {
350                     j = ctzl(c);
351                     c &= ~(1ul << j);
352                     page_number = (i * HOST_LONG_BITS + j) * hpratio;
353                     addr = page_number * TARGET_PAGE_SIZE;
354                     ram_addr = start + addr;
355                     cpu_physical_memory_set_dirty_range(ram_addr,
356                                        TARGET_PAGE_SIZE * hpratio, clients);
357                 } while (c != 0);
358             }
359         }
360     }
361 }
362 #endif /* not _WIN32 */
363 
364 bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start,
365                                               ram_addr_t length,
366                                               unsigned client);
367 
368 DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty
369     (ram_addr_t start, ram_addr_t length, unsigned client);
370 
371 bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap,
372                                             ram_addr_t start,
373                                             ram_addr_t length);
374 
375 static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start,
376                                                          ram_addr_t length)
377 {
378     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_MIGRATION);
379     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_VGA);
380     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_CODE);
381 }
382 
383 
384 static inline
385 uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb,
386                                                ram_addr_t start,
387                                                ram_addr_t length,
388                                                uint64_t *real_dirty_pages)
389 {
390     ram_addr_t addr;
391     unsigned long word = BIT_WORD((start + rb->offset) >> TARGET_PAGE_BITS);
392     uint64_t num_dirty = 0;
393     unsigned long *dest = rb->bmap;
394 
395     /* start address and length is aligned at the start of a word? */
396     if (((word * BITS_PER_LONG) << TARGET_PAGE_BITS) ==
397          (start + rb->offset) &&
398         !(length & ((BITS_PER_LONG << TARGET_PAGE_BITS) - 1))) {
399         int k;
400         int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
401         unsigned long * const *src;
402         unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE;
403         unsigned long offset = BIT_WORD((word * BITS_PER_LONG) %
404                                         DIRTY_MEMORY_BLOCK_SIZE);
405         unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
406 
407         rcu_read_lock();
408 
409         src = atomic_rcu_read(
410                 &ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION])->blocks;
411 
412         for (k = page; k < page + nr; k++) {
413             if (src[idx][offset]) {
414                 unsigned long bits = atomic_xchg(&src[idx][offset], 0);
415                 unsigned long new_dirty;
416                 *real_dirty_pages += ctpopl(bits);
417                 new_dirty = ~dest[k];
418                 dest[k] |= bits;
419                 new_dirty &= bits;
420                 num_dirty += ctpopl(new_dirty);
421             }
422 
423             if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
424                 offset = 0;
425                 idx++;
426             }
427         }
428 
429         rcu_read_unlock();
430     } else {
431         ram_addr_t offset = rb->offset;
432 
433         for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
434             if (cpu_physical_memory_test_and_clear_dirty(
435                         start + addr + offset,
436                         TARGET_PAGE_SIZE,
437                         DIRTY_MEMORY_MIGRATION)) {
438                 *real_dirty_pages += 1;
439                 long k = (start + addr) >> TARGET_PAGE_BITS;
440                 if (!test_and_set_bit(k, dest)) {
441                     num_dirty++;
442                 }
443             }
444         }
445     }
446 
447     return num_dirty;
448 }
449 #endif
450 #endif
451