blob: d28fc0cd3d9fd2e0a77aaa27d0343202d46bef88 [file] [log] [blame]
bellard54936002003-05-13 00:25:15 +00001/*
Blue Swirl5b6dd862012-12-02 16:04:43 +00002 * Virtual page mapping
ths5fafdf22007-09-16 21:08:06 +00003 *
bellard54936002003-05-13 00:25:15 +00004 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
Blue Swirl8167ee82009-07-16 20:47:01 +000017 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
bellard54936002003-05-13 00:25:15 +000018 */
Peter Maydell7b31bbc2016-01-26 18:16:56 +000019#include "qemu/osdep.h"
Markus Armbrusterda34e652016-03-14 09:01:28 +010020#include "qapi/error.h"
bellard54936002003-05-13 00:25:15 +000021
Veronia Bahaaf348b6d2016-03-20 19:16:19 +020022#include "qemu/cutils.h"
bellard6180a182003-09-30 21:04:53 +000023#include "cpu.h"
Paolo Bonzini63c91552016-03-15 13:18:37 +010024#include "exec/exec-all.h"
Juan Quintela51180422017-04-24 20:50:19 +020025#include "exec/target_page.h"
bellardb67d9a52008-05-23 09:57:34 +000026#include "tcg.h"
Paolo Bonzini741da0d2014-06-27 08:40:04 +020027#include "hw/qdev-core.h"
Fam Zhengc7e002c2017-07-14 10:15:08 +080028#include "hw/qdev-properties.h"
Michael S. Tsirkin4485bd22015-03-11 07:56:34 +010029#if !defined(CONFIG_USER_ONLY)
Marcel Apfelbaum47c8ca52015-02-04 17:43:54 +020030#include "hw/boards.h"
Paolo Bonzini33c11872016-03-15 16:58:45 +010031#include "hw/xen/xen.h"
Michael S. Tsirkin4485bd22015-03-11 07:56:34 +010032#endif
Paolo Bonzini9c17d612012-12-17 18:20:04 +010033#include "sysemu/kvm.h"
Markus Armbruster2ff3de62013-07-04 15:09:22 +020034#include "sysemu/sysemu.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010035#include "qemu/timer.h"
36#include "qemu/config-file.h"
Andreas Färber75a34032013-09-02 16:57:02 +020037#include "qemu/error-report.h"
pbrook53a59602006-03-25 19:31:22 +000038#if defined(CONFIG_USER_ONLY)
Markus Armbrustera9c94272016-06-22 19:11:19 +020039#include "qemu.h"
Jun Nakajima432d2682010-08-31 16:41:25 +010040#else /* !CONFIG_USER_ONLY */
Paolo Bonzini741da0d2014-06-27 08:40:04 +020041#include "hw/hw.h"
42#include "exec/memory.h"
Paolo Bonzinidf43d492016-03-16 10:24:54 +010043#include "exec/ioport.h"
Paolo Bonzini741da0d2014-06-27 08:40:04 +020044#include "sysemu/dma.h"
Alexey Kardashevskiy9c607662017-03-02 13:36:11 +110045#include "sysemu/numa.h"
Christian Borntraeger79ca7a12017-03-07 15:19:08 +010046#include "sysemu/hw_accel.h"
Paolo Bonzini741da0d2014-06-27 08:40:04 +020047#include "exec/address-spaces.h"
Paolo Bonzini9c17d612012-12-17 18:20:04 +010048#include "sysemu/xen-mapcache.h"
Daniel P. Berrange0ab8ed12017-01-25 16:14:15 +000049#include "trace-root.h"
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +000050
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +000051#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +000052#include <linux/falloc.h>
53#endif
54
pbrook53a59602006-03-25 19:31:22 +000055#endif
Mike Day0dc3f442013-09-05 14:41:35 -040056#include "qemu/rcu_queue.h"
Jan Kiszka4840f102015-06-18 18:47:22 +020057#include "qemu/main-loop.h"
Blue Swirl5b6dd862012-12-02 16:04:43 +000058#include "translate-all.h"
Pavel Dovgalyuk76159362015-09-17 19:25:07 +030059#include "sysemu/replay.h"
Blue Swirl0cac1b62012-04-09 16:50:52 +000060
Paolo Bonzini022c62c2012-12-17 18:19:49 +010061#include "exec/memory-internal.h"
Juan Quintela220c3eb2013-10-14 17:13:59 +020062#include "exec/ram_addr.h"
Paolo Bonzini508127e2016-01-07 16:55:28 +030063#include "exec/log.h"
Avi Kivity67d95c12011-12-15 15:25:22 +020064
Bharata B Rao9dfeca72016-05-12 09:18:12 +053065#include "migration/vmstate.h"
66
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +020067#include "qemu/range.h"
Michael S. Tsirkin794e8f32015-09-24 14:41:17 +030068#ifndef _WIN32
69#include "qemu/mmap-alloc.h"
70#endif
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +020071
Peter Xube9b23c2017-05-12 12:17:41 +080072#include "monitor/monitor.h"
73
blueswir1db7b5422007-05-26 17:36:03 +000074//#define DEBUG_SUBPAGE
ths1196be32007-03-17 15:17:58 +000075
pbrook99773bd2006-04-16 15:14:59 +000076#if !defined(CONFIG_USER_ONLY)
Mike Day0dc3f442013-09-05 14:41:35 -040077/* ram_list is read under rcu_read_lock()/rcu_read_unlock(). Writes
78 * are protected by the ramlist lock.
79 */
Mike Day0d53d9f2015-01-21 13:45:24 +010080RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) };
Avi Kivity62152b82011-07-26 14:26:14 +030081
82static MemoryRegion *system_memory;
Avi Kivity309cb472011-08-08 16:09:03 +030083static MemoryRegion *system_io;
Avi Kivity62152b82011-07-26 14:26:14 +030084
Avi Kivityf6790af2012-10-02 20:13:51 +020085AddressSpace address_space_io;
86AddressSpace address_space_memory;
Avi Kivity2673a5d2012-10-02 18:49:28 +020087
Paolo Bonzini0844e002013-05-24 14:37:28 +020088MemoryRegion io_mem_rom, io_mem_notdirty;
Jan Kiszkaacc9d802013-05-26 21:55:37 +020089static MemoryRegion io_mem_unassigned;
Avi Kivity0e0df1e2012-01-02 00:32:15 +020090
Paolo Bonzini7bd4f432014-05-14 17:43:22 +080091/* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
92#define RAM_PREALLOC (1 << 0)
93
Paolo Bonzinidbcb8982014-06-10 19:15:24 +080094/* RAM is mmap-ed with MAP_SHARED */
95#define RAM_SHARED (1 << 1)
96
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +020097/* Only a portion of RAM (used_length) is actually used, and migrated.
98 * This used_length size can change across reboots.
99 */
100#define RAM_RESIZEABLE (1 << 2)
101
pbrooke2eef172008-06-08 01:09:01 +0000102#endif
bellard9fa3e852004-01-04 18:06:42 +0000103
Peter Maydell20bccb82016-10-24 16:26:49 +0100104#ifdef TARGET_PAGE_BITS_VARY
105int target_page_bits;
106bool target_page_bits_decided;
107#endif
108
Andreas Färberbdc44642013-06-24 23:50:24 +0200109struct CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus);
bellard6a00d602005-11-21 23:25:50 +0000110/* current CPU in the current thread. It is only valid inside
111 cpu_exec() */
Paolo Bonzinif240eb62015-08-26 00:17:58 +0200112__thread CPUState *current_cpu;
pbrook2e70f6e2008-06-29 01:03:05 +0000113/* 0 = Do not count executed instructions.
thsbf20dc02008-06-30 17:22:19 +0000114 1 = Precise instruction counting.
pbrook2e70f6e2008-06-29 01:03:05 +0000115 2 = Adaptive rate instruction counting. */
Paolo Bonzini5708fc62012-11-26 15:36:40 +0100116int use_icount;
bellard6a00d602005-11-21 23:25:50 +0000117
Yang Zhonga0be0c52017-07-03 18:12:13 +0800118uintptr_t qemu_host_page_size;
119intptr_t qemu_host_page_mask;
Yang Zhonga0be0c52017-07-03 18:12:13 +0800120
Peter Maydell20bccb82016-10-24 16:26:49 +0100121bool set_preferred_target_page_bits(int bits)
122{
123 /* The target page size is the lowest common denominator for all
124 * the CPUs in the system, so we can only make it smaller, never
125 * larger. And we can't make it smaller once we've committed to
126 * a particular size.
127 */
128#ifdef TARGET_PAGE_BITS_VARY
129 assert(bits >= TARGET_PAGE_BITS_MIN);
130 if (target_page_bits == 0 || target_page_bits > bits) {
131 if (target_page_bits_decided) {
132 return false;
133 }
134 target_page_bits = bits;
135 }
136#endif
137 return true;
138}
139
pbrooke2eef172008-06-08 01:09:01 +0000140#if !defined(CONFIG_USER_ONLY)
Avi Kivity4346ae32012-02-10 17:00:01 +0200141
Peter Maydell20bccb82016-10-24 16:26:49 +0100142static void finalize_target_page_bits(void)
143{
144#ifdef TARGET_PAGE_BITS_VARY
145 if (target_page_bits == 0) {
146 target_page_bits = TARGET_PAGE_BITS_MIN;
147 }
148 target_page_bits_decided = true;
149#endif
150}
151
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200152typedef struct PhysPageEntry PhysPageEntry;
153
154struct PhysPageEntry {
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200155 /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200156 uint32_t skip : 6;
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200157 /* index into phys_sections (!skip) or phys_map_nodes (skip) */
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200158 uint32_t ptr : 26;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200159};
160
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200161#define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6)
162
Paolo Bonzini03f49952013-11-07 17:14:36 +0100163/* Size of the L2 (and L3, etc) page tables. */
Paolo Bonzini57271d62013-11-07 17:14:37 +0100164#define ADDR_SPACE_BITS 64
Paolo Bonzini03f49952013-11-07 17:14:36 +0100165
Michael S. Tsirkin026736c2013-11-13 20:13:03 +0200166#define P_L2_BITS 9
Paolo Bonzini03f49952013-11-07 17:14:36 +0100167#define P_L2_SIZE (1 << P_L2_BITS)
168
169#define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1)
170
171typedef PhysPageEntry Node[P_L2_SIZE];
Paolo Bonzini0475d942013-05-29 12:28:21 +0200172
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200173typedef struct PhysPageMap {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100174 struct rcu_head rcu;
175
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200176 unsigned sections_nb;
177 unsigned sections_nb_alloc;
178 unsigned nodes_nb;
179 unsigned nodes_nb_alloc;
180 Node *nodes;
181 MemoryRegionSection *sections;
182} PhysPageMap;
183
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200184struct AddressSpaceDispatch {
Fam Zheng729633c2016-03-01 14:18:24 +0800185 MemoryRegionSection *mru_section;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200186 /* This is a multi-level map on the physical address space.
187 * The bottom level has pointers to MemoryRegionSections.
188 */
189 PhysPageEntry phys_map;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200190 PhysPageMap map;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200191};
192
Jan Kiszka90260c62013-05-26 21:46:51 +0200193#define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
194typedef struct subpage_t {
195 MemoryRegion iomem;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +1000196 FlatView *fv;
Jan Kiszka90260c62013-05-26 21:46:51 +0200197 hwaddr base;
Vijaya Kumar K2615fab2016-10-24 16:26:49 +0100198 uint16_t sub_section[];
Jan Kiszka90260c62013-05-26 21:46:51 +0200199} subpage_t;
200
Liu Ping Fanb41aac42013-05-29 11:09:17 +0200201#define PHYS_SECTION_UNASSIGNED 0
202#define PHYS_SECTION_NOTDIRTY 1
203#define PHYS_SECTION_ROM 2
204#define PHYS_SECTION_WATCH 3
Avi Kivity5312bd82012-02-12 18:32:55 +0200205
pbrooke2eef172008-06-08 01:09:01 +0000206static void io_mem_init(void);
Avi Kivity62152b82011-07-26 14:26:14 +0300207static void memory_map_init(void);
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000208static void tcg_commit(MemoryListener *listener);
pbrooke2eef172008-06-08 01:09:01 +0000209
Avi Kivity1ec9b902012-01-02 12:47:48 +0200210static MemoryRegion io_mem_watch;
Peter Maydell32857f42015-10-01 15:29:50 +0100211
212/**
213 * CPUAddressSpace: all the information a CPU needs about an AddressSpace
214 * @cpu: the CPU whose AddressSpace this is
215 * @as: the AddressSpace itself
216 * @memory_dispatch: its dispatch pointer (cached, RCU protected)
217 * @tcg_as_listener: listener for tracking changes to the AddressSpace
218 */
219struct CPUAddressSpace {
220 CPUState *cpu;
221 AddressSpace *as;
222 struct AddressSpaceDispatch *memory_dispatch;
223 MemoryListener tcg_as_listener;
224};
225
Gerd Hoffmann8deaf122017-04-21 11:16:25 +0200226struct DirtyBitmapSnapshot {
227 ram_addr_t start;
228 ram_addr_t end;
229 unsigned long dirty[];
230};
231
pbrook6658ffb2007-03-16 23:58:11 +0000232#endif
bellard54936002003-05-13 00:25:15 +0000233
Paul Brook6d9a1302010-02-28 23:55:53 +0000234#if !defined(CONFIG_USER_ONLY)
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200235
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200236static void phys_map_node_reserve(PhysPageMap *map, unsigned nodes)
Avi Kivityf7bf5462012-02-13 20:12:05 +0200237{
Peter Lieven101420b2016-07-15 12:03:50 +0200238 static unsigned alloc_hint = 16;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200239 if (map->nodes_nb + nodes > map->nodes_nb_alloc) {
Peter Lieven101420b2016-07-15 12:03:50 +0200240 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, alloc_hint);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200241 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, map->nodes_nb + nodes);
242 map->nodes = g_renew(Node, map->nodes, map->nodes_nb_alloc);
Peter Lieven101420b2016-07-15 12:03:50 +0200243 alloc_hint = map->nodes_nb_alloc;
Avi Kivityf7bf5462012-02-13 20:12:05 +0200244 }
245}
246
Paolo Bonzinidb946042015-05-21 15:12:29 +0200247static uint32_t phys_map_node_alloc(PhysPageMap *map, bool leaf)
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200248{
249 unsigned i;
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200250 uint32_t ret;
Paolo Bonzinidb946042015-05-21 15:12:29 +0200251 PhysPageEntry e;
252 PhysPageEntry *p;
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200253
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200254 ret = map->nodes_nb++;
Paolo Bonzinidb946042015-05-21 15:12:29 +0200255 p = map->nodes[ret];
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200256 assert(ret != PHYS_MAP_NODE_NIL);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200257 assert(ret != map->nodes_nb_alloc);
Paolo Bonzinidb946042015-05-21 15:12:29 +0200258
259 e.skip = leaf ? 0 : 1;
260 e.ptr = leaf ? PHYS_SECTION_UNASSIGNED : PHYS_MAP_NODE_NIL;
Paolo Bonzini03f49952013-11-07 17:14:36 +0100261 for (i = 0; i < P_L2_SIZE; ++i) {
Paolo Bonzinidb946042015-05-21 15:12:29 +0200262 memcpy(&p[i], &e, sizeof(e));
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200263 }
Avi Kivityf7bf5462012-02-13 20:12:05 +0200264 return ret;
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200265}
266
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200267static void phys_page_set_level(PhysPageMap *map, PhysPageEntry *lp,
268 hwaddr *index, hwaddr *nb, uint16_t leaf,
Avi Kivity29990972012-02-13 20:21:20 +0200269 int level)
Avi Kivityf7bf5462012-02-13 20:12:05 +0200270{
271 PhysPageEntry *p;
Paolo Bonzini03f49952013-11-07 17:14:36 +0100272 hwaddr step = (hwaddr)1 << (level * P_L2_BITS);
Avi Kivityf7bf5462012-02-13 20:12:05 +0200273
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200274 if (lp->skip && lp->ptr == PHYS_MAP_NODE_NIL) {
Paolo Bonzinidb946042015-05-21 15:12:29 +0200275 lp->ptr = phys_map_node_alloc(map, level == 0);
Avi Kivityf7bf5462012-02-13 20:12:05 +0200276 }
Paolo Bonzinidb946042015-05-21 15:12:29 +0200277 p = map->nodes[lp->ptr];
Paolo Bonzini03f49952013-11-07 17:14:36 +0100278 lp = &p[(*index >> (level * P_L2_BITS)) & (P_L2_SIZE - 1)];
Avi Kivityf7bf5462012-02-13 20:12:05 +0200279
Paolo Bonzini03f49952013-11-07 17:14:36 +0100280 while (*nb && lp < &p[P_L2_SIZE]) {
Avi Kivity07f07b32012-02-13 20:45:32 +0200281 if ((*index & (step - 1)) == 0 && *nb >= step) {
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200282 lp->skip = 0;
Avi Kivityc19e8802012-02-13 20:25:31 +0200283 lp->ptr = leaf;
Avi Kivity07f07b32012-02-13 20:45:32 +0200284 *index += step;
285 *nb -= step;
Avi Kivity29990972012-02-13 20:21:20 +0200286 } else {
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200287 phys_page_set_level(map, lp, index, nb, leaf, level - 1);
Avi Kivity29990972012-02-13 20:21:20 +0200288 }
289 ++lp;
Avi Kivityf7bf5462012-02-13 20:12:05 +0200290 }
291}
292
Avi Kivityac1970f2012-10-03 16:22:53 +0200293static void phys_page_set(AddressSpaceDispatch *d,
Avi Kivitya8170e52012-10-23 12:30:10 +0200294 hwaddr index, hwaddr nb,
Avi Kivity29990972012-02-13 20:21:20 +0200295 uint16_t leaf)
bellard92e873b2004-05-21 14:52:29 +0000296{
Avi Kivity29990972012-02-13 20:21:20 +0200297 /* Wildly overreserve - it doesn't matter much. */
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200298 phys_map_node_reserve(&d->map, 3 * P_L2_LEVELS);
bellard92e873b2004-05-21 14:52:29 +0000299
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200300 phys_page_set_level(&d->map, &d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1);
bellard92e873b2004-05-21 14:52:29 +0000301}
302
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200303/* Compact a non leaf page entry. Simply detect that the entry has a single child,
304 * and update our entry so we can skip it and go directly to the destination.
305 */
Marc-André Lureauefee6782016-09-28 16:37:20 +0400306static void phys_page_compact(PhysPageEntry *lp, Node *nodes)
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200307{
308 unsigned valid_ptr = P_L2_SIZE;
309 int valid = 0;
310 PhysPageEntry *p;
311 int i;
312
313 if (lp->ptr == PHYS_MAP_NODE_NIL) {
314 return;
315 }
316
317 p = nodes[lp->ptr];
318 for (i = 0; i < P_L2_SIZE; i++) {
319 if (p[i].ptr == PHYS_MAP_NODE_NIL) {
320 continue;
321 }
322
323 valid_ptr = i;
324 valid++;
325 if (p[i].skip) {
Marc-André Lureauefee6782016-09-28 16:37:20 +0400326 phys_page_compact(&p[i], nodes);
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200327 }
328 }
329
330 /* We can only compress if there's only one child. */
331 if (valid != 1) {
332 return;
333 }
334
335 assert(valid_ptr < P_L2_SIZE);
336
337 /* Don't compress if it won't fit in the # of bits we have. */
338 if (lp->skip + p[valid_ptr].skip >= (1 << 3)) {
339 return;
340 }
341
342 lp->ptr = p[valid_ptr].ptr;
343 if (!p[valid_ptr].skip) {
344 /* If our only child is a leaf, make this a leaf. */
345 /* By design, we should have made this node a leaf to begin with so we
346 * should never reach here.
347 * But since it's so simple to handle this, let's do it just in case we
348 * change this rule.
349 */
350 lp->skip = 0;
351 } else {
352 lp->skip += p[valid_ptr].skip;
353 }
354}
355
Alexey Kardashevskiy8629d3f2017-09-21 18:51:00 +1000356void address_space_dispatch_compact(AddressSpaceDispatch *d)
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200357{
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200358 if (d->phys_map.skip) {
Marc-André Lureauefee6782016-09-28 16:37:20 +0400359 phys_page_compact(&d->phys_map, d->map.nodes);
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200360 }
361}
362
Fam Zheng29cb5332016-03-01 14:18:23 +0800363static inline bool section_covers_addr(const MemoryRegionSection *section,
364 hwaddr addr)
365{
366 /* Memory topology clips a memory region to [0, 2^64); size.hi > 0 means
367 * the section must cover the entire address space.
368 */
Richard Henderson258dfaa2016-06-29 15:48:03 -0700369 return int128_gethi(section->size) ||
Fam Zheng29cb5332016-03-01 14:18:23 +0800370 range_covers_byte(section->offset_within_address_space,
Richard Henderson258dfaa2016-06-29 15:48:03 -0700371 int128_getlo(section->size), addr);
Fam Zheng29cb5332016-03-01 14:18:23 +0800372}
373
Peter Xu003a0cf2017-05-15 16:50:57 +0800374static MemoryRegionSection *phys_page_find(AddressSpaceDispatch *d, hwaddr addr)
bellard92e873b2004-05-21 14:52:29 +0000375{
Peter Xu003a0cf2017-05-15 16:50:57 +0800376 PhysPageEntry lp = d->phys_map, *p;
377 Node *nodes = d->map.nodes;
378 MemoryRegionSection *sections = d->map.sections;
Michael S. Tsirkin97115a82013-11-13 20:08:19 +0200379 hwaddr index = addr >> TARGET_PAGE_BITS;
Avi Kivity31ab2b42012-02-13 16:44:19 +0200380 int i;
Avi Kivityf1f6e3b2011-11-20 17:52:22 +0200381
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200382 for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) {
Avi Kivityc19e8802012-02-13 20:25:31 +0200383 if (lp.ptr == PHYS_MAP_NODE_NIL) {
Paolo Bonzini9affd6f2013-05-29 12:09:47 +0200384 return &sections[PHYS_SECTION_UNASSIGNED];
Avi Kivity31ab2b42012-02-13 16:44:19 +0200385 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +0200386 p = nodes[lp.ptr];
Paolo Bonzini03f49952013-11-07 17:14:36 +0100387 lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)];
Avi Kivityf1f6e3b2011-11-20 17:52:22 +0200388 }
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200389
Fam Zheng29cb5332016-03-01 14:18:23 +0800390 if (section_covers_addr(&sections[lp.ptr], addr)) {
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200391 return &sections[lp.ptr];
392 } else {
393 return &sections[PHYS_SECTION_UNASSIGNED];
394 }
Avi Kivityf3705d52012-03-08 16:16:34 +0200395}
396
Blue Swirle5548612012-04-21 13:08:33 +0000397bool memory_region_is_unassigned(MemoryRegion *mr)
398{
Paolo Bonzini2a8e7492013-05-24 14:34:08 +0200399 return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device
Blue Swirle5548612012-04-21 13:08:33 +0000400 && mr != &io_mem_watch;
401}
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200402
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100403/* Called from RCU critical section */
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200404static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
Jan Kiszka90260c62013-05-26 21:46:51 +0200405 hwaddr addr,
406 bool resolve_subpage)
Jan Kiszka9f029602013-05-06 16:48:02 +0200407{
Fam Zheng729633c2016-03-01 14:18:24 +0800408 MemoryRegionSection *section = atomic_read(&d->mru_section);
Jan Kiszka90260c62013-05-26 21:46:51 +0200409 subpage_t *subpage;
410
Paolo Bonzini07c114b2017-11-15 15:11:03 +0100411 if (!section || section == &d->map.sections[PHYS_SECTION_UNASSIGNED] ||
412 !section_covers_addr(section, addr)) {
Peter Xu003a0cf2017-05-15 16:50:57 +0800413 section = phys_page_find(d, addr);
Paolo Bonzini07c114b2017-11-15 15:11:03 +0100414 atomic_set(&d->mru_section, section);
Fam Zheng729633c2016-03-01 14:18:24 +0800415 }
Jan Kiszka90260c62013-05-26 21:46:51 +0200416 if (resolve_subpage && section->mr->subpage) {
417 subpage = container_of(section->mr, subpage_t, iomem);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200418 section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]];
Jan Kiszka90260c62013-05-26 21:46:51 +0200419 }
420 return section;
Jan Kiszka9f029602013-05-06 16:48:02 +0200421}
422
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100423/* Called from RCU critical section */
Jan Kiszka90260c62013-05-26 21:46:51 +0200424static MemoryRegionSection *
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200425address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
Jan Kiszka90260c62013-05-26 21:46:51 +0200426 hwaddr *plen, bool resolve_subpage)
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200427{
428 MemoryRegionSection *section;
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200429 MemoryRegion *mr;
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100430 Int128 diff;
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200431
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200432 section = address_space_lookup_region(d, addr, resolve_subpage);
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200433 /* Compute offset within MemoryRegionSection */
434 addr -= section->offset_within_address_space;
435
436 /* Compute offset within MemoryRegion */
437 *xlat = addr + section->offset_within_region;
438
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200439 mr = section->mr;
Paolo Bonzinib242e0e2015-07-04 00:24:51 +0200440
441 /* MMIO registers can be expected to perform full-width accesses based only
442 * on their address, without considering adjacent registers that could
443 * decode to completely different MemoryRegions. When such registers
444 * exist (e.g. I/O ports 0xcf8 and 0xcf9 on most PC chipsets), MMIO
445 * regions overlap wildly. For this reason we cannot clamp the accesses
446 * here.
447 *
448 * If the length is small (as is the case for address_space_ldl/stl),
449 * everything works fine. If the incoming length is large, however,
450 * the caller really has to do the clamping through memory_access_size.
451 */
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200452 if (memory_region_is_ram(mr)) {
Paolo Bonzinie4a511f2015-06-17 10:36:54 +0200453 diff = int128_sub(section->size, int128_make64(addr));
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200454 *plen = int128_get64(int128_min(diff, int128_make64(*plen)));
455 }
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200456 return section;
457}
Jan Kiszka90260c62013-05-26 21:46:51 +0200458
Peter Xud5e5faf2017-10-10 11:42:45 +0200459/**
460 * flatview_do_translate - translate an address in FlatView
461 *
462 * @fv: the flat view that we want to translate on
463 * @addr: the address to be translated in above address space
464 * @xlat: the translated address offset within memory region. It
465 * cannot be @NULL.
466 * @plen_out: valid read/write length of the translated address. It
467 * can be @NULL when we don't care about it.
468 * @page_mask_out: page mask for the translated address. This
469 * should only be meaningful for IOMMU translated
470 * addresses, since there may be huge pages that this bit
471 * would tell. It can be @NULL if we don't care about it.
472 * @is_write: whether the translation operation is for write
473 * @is_mmio: whether this can be MMIO, set true if it can
474 *
475 * This function is called from RCU critical section
476 */
Alexey Kardashevskiy16620682017-09-21 18:50:58 +1000477static MemoryRegionSection flatview_do_translate(FlatView *fv,
478 hwaddr addr,
479 hwaddr *xlat,
Peter Xud5e5faf2017-10-10 11:42:45 +0200480 hwaddr *plen_out,
481 hwaddr *page_mask_out,
Alexey Kardashevskiy16620682017-09-21 18:50:58 +1000482 bool is_write,
483 bool is_mmio,
484 AddressSpace **target_as)
Jan Kiszka90260c62013-05-26 21:46:51 +0200485{
Avi Kivity30951152012-10-30 13:47:46 +0200486 IOMMUTLBEntry iotlb;
487 MemoryRegionSection *section;
Alexey Kardashevskiy3df9d742017-07-11 13:56:19 +1000488 IOMMUMemoryRegion *iommu_mr;
Alexey Kardashevskiy1221a472017-07-11 13:56:20 +1000489 IOMMUMemoryRegionClass *imrc;
Peter Xud5e5faf2017-10-10 11:42:45 +0200490 hwaddr page_mask = (hwaddr)(-1);
491 hwaddr plen = (hwaddr)(-1);
492
493 if (plen_out) {
494 plen = *plen_out;
495 }
Avi Kivity30951152012-10-30 13:47:46 +0200496
497 for (;;) {
Alexey Kardashevskiy16620682017-09-21 18:50:58 +1000498 section = address_space_translate_internal(
499 flatview_to_dispatch(fv), addr, &addr,
Peter Xud5e5faf2017-10-10 11:42:45 +0200500 &plen, is_mmio);
Avi Kivity30951152012-10-30 13:47:46 +0200501
Alexey Kardashevskiy3df9d742017-07-11 13:56:19 +1000502 iommu_mr = memory_region_get_iommu(section->mr);
503 if (!iommu_mr) {
Avi Kivity30951152012-10-30 13:47:46 +0200504 break;
505 }
Alexey Kardashevskiy1221a472017-07-11 13:56:20 +1000506 imrc = memory_region_get_iommu_class_nocheck(iommu_mr);
Avi Kivity30951152012-10-30 13:47:46 +0200507
Alexey Kardashevskiy1221a472017-07-11 13:56:20 +1000508 iotlb = imrc->translate(iommu_mr, addr, is_write ?
509 IOMMU_WO : IOMMU_RO);
Avi Kivity30951152012-10-30 13:47:46 +0200510 addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
511 | (addr & iotlb.addr_mask));
Peter Xud5e5faf2017-10-10 11:42:45 +0200512 page_mask &= iotlb.addr_mask;
513 plen = MIN(plen, (addr | iotlb.addr_mask) - addr + 1);
Avi Kivity30951152012-10-30 13:47:46 +0200514 if (!(iotlb.perm & (1 << is_write))) {
Peter Xua7640402017-05-17 16:57:42 +0800515 goto translate_fail;
Avi Kivity30951152012-10-30 13:47:46 +0200516 }
517
Alexey Kardashevskiy16620682017-09-21 18:50:58 +1000518 fv = address_space_to_flatview(iotlb.target_as);
Alexey Kardashevskiye76bb182017-09-21 18:50:53 +1000519 *target_as = iotlb.target_as;
Avi Kivity30951152012-10-30 13:47:46 +0200520 }
521
Peter Xua7640402017-05-17 16:57:42 +0800522 *xlat = addr;
523
Peter Xud5e5faf2017-10-10 11:42:45 +0200524 if (page_mask == (hwaddr)(-1)) {
525 /* Not behind an IOMMU, use default page size. */
526 page_mask = ~TARGET_PAGE_MASK;
527 }
528
529 if (page_mask_out) {
530 *page_mask_out = page_mask;
531 }
532
533 if (plen_out) {
534 *plen_out = plen;
535 }
536
Peter Xua7640402017-05-17 16:57:42 +0800537 return *section;
538
539translate_fail:
540 return (MemoryRegionSection) { .mr = &io_mem_unassigned };
541}
542
543/* Called from RCU critical section */
544IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
545 bool is_write)
546{
547 MemoryRegionSection section;
Peter Xu076a93d2017-10-10 11:42:46 +0200548 hwaddr xlat, page_mask;
Peter Xua7640402017-05-17 16:57:42 +0800549
Peter Xu076a93d2017-10-10 11:42:46 +0200550 /*
551 * This can never be MMIO, and we don't really care about plen,
552 * but page mask.
553 */
554 section = flatview_do_translate(address_space_to_flatview(as), addr, &xlat,
555 NULL, &page_mask, is_write, false, &as);
Peter Xua7640402017-05-17 16:57:42 +0800556
557 /* Illegal translation */
558 if (section.mr == &io_mem_unassigned) {
559 goto iotlb_fail;
560 }
561
562 /* Convert memory region offset into address space offset */
563 xlat += section.offset_within_address_space -
564 section.offset_within_region;
565
Peter Xua7640402017-05-17 16:57:42 +0800566 return (IOMMUTLBEntry) {
Alexey Kardashevskiye76bb182017-09-21 18:50:53 +1000567 .target_as = as,
Peter Xu076a93d2017-10-10 11:42:46 +0200568 .iova = addr & ~page_mask,
569 .translated_addr = xlat & ~page_mask,
570 .addr_mask = page_mask,
Peter Xua7640402017-05-17 16:57:42 +0800571 /* IOTLBs are for DMAs, and DMA only allows on RAMs. */
572 .perm = IOMMU_RW,
573 };
574
575iotlb_fail:
576 return (IOMMUTLBEntry) {0};
577}
578
579/* Called from RCU critical section */
Alexey Kardashevskiy16620682017-09-21 18:50:58 +1000580MemoryRegion *flatview_translate(FlatView *fv, hwaddr addr, hwaddr *xlat,
581 hwaddr *plen, bool is_write)
Peter Xua7640402017-05-17 16:57:42 +0800582{
583 MemoryRegion *mr;
584 MemoryRegionSection section;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +1000585 AddressSpace *as = NULL;
Peter Xua7640402017-05-17 16:57:42 +0800586
587 /* This can be MMIO, so setup MMIO bit. */
Peter Xud5e5faf2017-10-10 11:42:45 +0200588 section = flatview_do_translate(fv, addr, xlat, plen, NULL,
589 is_write, true, &as);
Peter Xua7640402017-05-17 16:57:42 +0800590 mr = section.mr;
591
Alexey Kardashevskiyfe680d02014-05-07 13:40:39 +0000592 if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100593 hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr;
Peter Crosthwaite23820db2015-03-16 22:35:54 -0700594 *plen = MIN(page, *plen);
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100595 }
596
Avi Kivity30951152012-10-30 13:47:46 +0200597 return mr;
Jan Kiszka90260c62013-05-26 21:46:51 +0200598}
599
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100600/* Called from RCU critical section */
Jan Kiszka90260c62013-05-26 21:46:51 +0200601MemoryRegionSection *
Peter Maydelld7898cd2016-01-21 14:15:05 +0000602address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr,
Paolo Bonzini9d82b5a2013-08-16 08:26:30 +0200603 hwaddr *xlat, hwaddr *plen)
Jan Kiszka90260c62013-05-26 21:46:51 +0200604{
Avi Kivity30951152012-10-30 13:47:46 +0200605 MemoryRegionSection *section;
Alex Bennéef35e44e2016-10-21 16:34:18 +0100606 AddressSpaceDispatch *d = atomic_rcu_read(&cpu->cpu_ases[asidx].memory_dispatch);
Peter Maydelld7898cd2016-01-21 14:15:05 +0000607
608 section = address_space_translate_internal(d, addr, xlat, plen, false);
Avi Kivity30951152012-10-30 13:47:46 +0200609
Alexey Kardashevskiy3df9d742017-07-11 13:56:19 +1000610 assert(!memory_region_is_iommu(section->mr));
Avi Kivity30951152012-10-30 13:47:46 +0200611 return section;
Jan Kiszka90260c62013-05-26 21:46:51 +0200612}
bellard9fa3e852004-01-04 18:06:42 +0000613#endif
bellardfd6ce8f2003-05-14 19:00:11 +0000614
Andreas Färberb170fce2013-01-20 20:23:22 +0100615#if !defined(CONFIG_USER_ONLY)
pbrook9656f322008-07-01 20:01:19 +0000616
Juan Quintelae59fb372009-09-29 22:48:21 +0200617static int cpu_common_post_load(void *opaque, int version_id)
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200618{
Andreas Färber259186a2013-01-17 18:51:17 +0100619 CPUState *cpu = opaque;
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200620
aurel323098dba2009-03-07 21:28:24 +0000621 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
622 version_id is increased. */
Andreas Färber259186a2013-01-17 18:51:17 +0100623 cpu->interrupt_request &= ~0x01;
Alex Bennéed10eb082016-11-14 14:17:28 +0000624 tlb_flush(cpu);
pbrook9656f322008-07-01 20:01:19 +0000625
Pavel Dovgalyuk15a356c2018-01-10 16:48:46 +0300626 /* loadvm has just updated the content of RAM, bypassing the
627 * usual mechanisms that ensure we flush TBs for writes to
628 * memory we've translated code from. So we must flush all TBs,
629 * which will now be stale.
630 */
631 tb_flush(cpu);
632
pbrook9656f322008-07-01 20:01:19 +0000633 return 0;
634}
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200635
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400636static int cpu_common_pre_load(void *opaque)
637{
638 CPUState *cpu = opaque;
639
Paolo Bonziniadee6422014-12-19 12:53:14 +0100640 cpu->exception_index = -1;
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400641
642 return 0;
643}
644
645static bool cpu_common_exception_index_needed(void *opaque)
646{
647 CPUState *cpu = opaque;
648
Paolo Bonziniadee6422014-12-19 12:53:14 +0100649 return tcg_enabled() && cpu->exception_index != -1;
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400650}
651
652static const VMStateDescription vmstate_cpu_common_exception_index = {
653 .name = "cpu_common/exception_index",
654 .version_id = 1,
655 .minimum_version_id = 1,
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200656 .needed = cpu_common_exception_index_needed,
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400657 .fields = (VMStateField[]) {
658 VMSTATE_INT32(exception_index, CPUState),
659 VMSTATE_END_OF_LIST()
660 }
661};
662
Andrey Smetaninbac05aa2015-07-03 15:01:44 +0300663static bool cpu_common_crash_occurred_needed(void *opaque)
664{
665 CPUState *cpu = opaque;
666
667 return cpu->crash_occurred;
668}
669
670static const VMStateDescription vmstate_cpu_common_crash_occurred = {
671 .name = "cpu_common/crash_occurred",
672 .version_id = 1,
673 .minimum_version_id = 1,
674 .needed = cpu_common_crash_occurred_needed,
675 .fields = (VMStateField[]) {
676 VMSTATE_BOOL(crash_occurred, CPUState),
677 VMSTATE_END_OF_LIST()
678 }
679};
680
Andreas Färber1a1562f2013-06-17 04:09:11 +0200681const VMStateDescription vmstate_cpu_common = {
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200682 .name = "cpu_common",
683 .version_id = 1,
684 .minimum_version_id = 1,
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400685 .pre_load = cpu_common_pre_load,
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200686 .post_load = cpu_common_post_load,
Juan Quintela35d08452014-04-16 16:01:33 +0200687 .fields = (VMStateField[]) {
Andreas Färber259186a2013-01-17 18:51:17 +0100688 VMSTATE_UINT32(halted, CPUState),
689 VMSTATE_UINT32(interrupt_request, CPUState),
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200690 VMSTATE_END_OF_LIST()
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400691 },
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200692 .subsections = (const VMStateDescription*[]) {
693 &vmstate_cpu_common_exception_index,
Andrey Smetaninbac05aa2015-07-03 15:01:44 +0300694 &vmstate_cpu_common_crash_occurred,
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200695 NULL
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200696 }
697};
Andreas Färber1a1562f2013-06-17 04:09:11 +0200698
pbrook9656f322008-07-01 20:01:19 +0000699#endif
700
Andreas Färber38d8f5c2012-12-17 19:47:15 +0100701CPUState *qemu_get_cpu(int index)
Glauber Costa950f1472009-06-09 12:15:18 -0400702{
Andreas Färberbdc44642013-06-24 23:50:24 +0200703 CPUState *cpu;
Glauber Costa950f1472009-06-09 12:15:18 -0400704
Andreas Färberbdc44642013-06-24 23:50:24 +0200705 CPU_FOREACH(cpu) {
Andreas Färber55e5c282012-12-17 06:18:02 +0100706 if (cpu->cpu_index == index) {
Andreas Färberbdc44642013-06-24 23:50:24 +0200707 return cpu;
Andreas Färber55e5c282012-12-17 06:18:02 +0100708 }
Glauber Costa950f1472009-06-09 12:15:18 -0400709 }
710
Andreas Färberbdc44642013-06-24 23:50:24 +0200711 return NULL;
Glauber Costa950f1472009-06-09 12:15:18 -0400712}
713
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000714#if !defined(CONFIG_USER_ONLY)
Peter Xu80ceb072017-11-23 17:23:32 +0800715void cpu_address_space_init(CPUState *cpu, int asidx,
716 const char *prefix, MemoryRegion *mr)
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000717{
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000718 CPUAddressSpace *newas;
Peter Xu80ceb072017-11-23 17:23:32 +0800719 AddressSpace *as = g_new0(AddressSpace, 1);
Peter Xu87a621d2017-11-23 17:23:33 +0800720 char *as_name;
Peter Xu80ceb072017-11-23 17:23:32 +0800721
722 assert(mr);
Peter Xu87a621d2017-11-23 17:23:33 +0800723 as_name = g_strdup_printf("%s-%d", prefix, cpu->cpu_index);
724 address_space_init(as, mr, as_name);
725 g_free(as_name);
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000726
727 /* Target code should have set num_ases before calling us */
728 assert(asidx < cpu->num_ases);
729
Peter Maydell56943e82016-01-21 14:15:04 +0000730 if (asidx == 0) {
731 /* address space 0 gets the convenience alias */
732 cpu->as = as;
733 }
734
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000735 /* KVM cannot currently support multiple address spaces. */
736 assert(asidx == 0 || !kvm_enabled());
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000737
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000738 if (!cpu->cpu_ases) {
739 cpu->cpu_ases = g_new0(CPUAddressSpace, cpu->num_ases);
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000740 }
Peter Maydell32857f42015-10-01 15:29:50 +0100741
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000742 newas = &cpu->cpu_ases[asidx];
743 newas->cpu = cpu;
744 newas->as = as;
Peter Maydell56943e82016-01-21 14:15:04 +0000745 if (tcg_enabled()) {
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000746 newas->tcg_as_listener.commit = tcg_commit;
747 memory_listener_register(&newas->tcg_as_listener, as);
Peter Maydell56943e82016-01-21 14:15:04 +0000748 }
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000749}
Peter Maydell651a5bc2016-01-21 14:15:05 +0000750
751AddressSpace *cpu_get_address_space(CPUState *cpu, int asidx)
752{
753 /* Return the AddressSpace corresponding to the specified index */
754 return cpu->cpu_ases[asidx].as;
755}
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000756#endif
757
Laurent Vivier7bbc1242016-10-20 13:26:04 +0200758void cpu_exec_unrealizefn(CPUState *cpu)
Bharata B Rao1c59eb32016-05-12 09:18:11 +0530759{
Bharata B Rao9dfeca72016-05-12 09:18:12 +0530760 CPUClass *cc = CPU_GET_CLASS(cpu);
761
Paolo Bonzini267f6852016-08-28 03:45:14 +0200762 cpu_list_remove(cpu);
Bharata B Rao9dfeca72016-05-12 09:18:12 +0530763
764 if (cc->vmsd != NULL) {
765 vmstate_unregister(NULL, cc->vmsd, cpu);
766 }
767 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
768 vmstate_unregister(NULL, &vmstate_cpu_common, cpu);
769 }
Bharata B Rao1c59eb32016-05-12 09:18:11 +0530770}
771
Fam Zhengc7e002c2017-07-14 10:15:08 +0800772Property cpu_common_props[] = {
773#ifndef CONFIG_USER_ONLY
774 /* Create a memory property for softmmu CPU object,
775 * so users can wire up its memory. (This can't go in qom/cpu.c
776 * because that file is compiled only once for both user-mode
777 * and system builds.) The default if no link is set up is to use
778 * the system address space.
779 */
780 DEFINE_PROP_LINK("memory", CPUState, memory, TYPE_MEMORY_REGION,
781 MemoryRegion *),
782#endif
783 DEFINE_PROP_END_OF_LIST(),
784};
785
Laurent Vivier39e329e2016-10-20 13:26:02 +0200786void cpu_exec_initfn(CPUState *cpu)
bellardfd6ce8f2003-05-14 19:00:11 +0000787{
Peter Maydell56943e82016-01-21 14:15:04 +0000788 cpu->as = NULL;
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000789 cpu->num_ases = 0;
Peter Maydell56943e82016-01-21 14:15:04 +0000790
Eduardo Habkost291135b2015-04-27 17:00:33 -0300791#ifndef CONFIG_USER_ONLY
Eduardo Habkost291135b2015-04-27 17:00:33 -0300792 cpu->thread_id = qemu_get_thread_id();
Peter Crosthwaite6731d862016-01-21 14:15:06 +0000793 cpu->memory = system_memory;
794 object_ref(OBJECT(cpu->memory));
Eduardo Habkost291135b2015-04-27 17:00:33 -0300795#endif
Laurent Vivier39e329e2016-10-20 13:26:02 +0200796}
797
Laurent Vivierce5b1bb2016-10-20 13:26:03 +0200798void cpu_exec_realizefn(CPUState *cpu, Error **errp)
Laurent Vivier39e329e2016-10-20 13:26:02 +0200799{
Richard Henderson55c3cee2017-10-15 19:02:42 -0700800 CPUClass *cc = CPU_GET_CLASS(cpu);
Emilio G. Cota2dda6352017-11-13 13:55:25 +0000801 static bool tcg_target_initialized;
Eduardo Habkost291135b2015-04-27 17:00:33 -0300802
Paolo Bonzini267f6852016-08-28 03:45:14 +0200803 cpu_list_add(cpu);
Igor Mammedov1bc7e522016-07-25 11:59:19 +0200804
Emilio G. Cota2dda6352017-11-13 13:55:25 +0000805 if (tcg_enabled() && !tcg_target_initialized) {
806 tcg_target_initialized = true;
Richard Henderson55c3cee2017-10-15 19:02:42 -0700807 cc->tcg_initialize();
808 }
809
Igor Mammedov1bc7e522016-07-25 11:59:19 +0200810#ifndef CONFIG_USER_ONLY
Andreas Färbere0d47942013-07-29 04:07:50 +0200811 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200812 vmstate_register(NULL, cpu->cpu_index, &vmstate_cpu_common, cpu);
Andreas Färbere0d47942013-07-29 04:07:50 +0200813 }
Andreas Färberb170fce2013-01-20 20:23:22 +0100814 if (cc->vmsd != NULL) {
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200815 vmstate_register(NULL, cpu->cpu_index, cc->vmsd, cpu);
Andreas Färberb170fce2013-01-20 20:23:22 +0100816 }
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200817#endif
bellardfd6ce8f2003-05-14 19:00:11 +0000818}
819
Pranith Kumar406bc332017-07-12 17:51:42 -0400820#if defined(CONFIG_USER_ONLY)
Andreas Färber00b941e2013-06-29 18:55:54 +0200821static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
Paul Brook94df27f2010-02-28 23:47:45 +0000822{
Pranith Kumar406bc332017-07-12 17:51:42 -0400823 mmap_lock();
824 tb_lock();
825 tb_invalidate_phys_page_range(pc, pc + 1, 0);
826 tb_unlock();
827 mmap_unlock();
Paul Brook94df27f2010-02-28 23:47:45 +0000828}
Pranith Kumar406bc332017-07-12 17:51:42 -0400829#else
830static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
831{
832 MemTxAttrs attrs;
833 hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs);
834 int asidx = cpu_asidx_from_attrs(cpu, attrs);
835 if (phys != -1) {
836 /* Locks grabbed by tb_invalidate_phys_addr */
837 tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as,
838 phys | (pc & ~TARGET_PAGE_MASK));
839 }
840}
841#endif
bellardd720b932004-04-25 17:57:43 +0000842
Paul Brookc527ee82010-03-01 03:31:14 +0000843#if defined(CONFIG_USER_ONLY)
Andreas Färber75a34032013-09-02 16:57:02 +0200844void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
Paul Brookc527ee82010-03-01 03:31:14 +0000845
846{
847}
848
Peter Maydell3ee887e2014-09-12 14:06:48 +0100849int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
850 int flags)
851{
852 return -ENOSYS;
853}
854
855void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
856{
857}
858
Andreas Färber75a34032013-09-02 16:57:02 +0200859int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
Paul Brookc527ee82010-03-01 03:31:14 +0000860 int flags, CPUWatchpoint **watchpoint)
861{
862 return -ENOSYS;
863}
864#else
pbrook6658ffb2007-03-16 23:58:11 +0000865/* Add a watchpoint. */
Andreas Färber75a34032013-09-02 16:57:02 +0200866int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
aliguoria1d1bb32008-11-18 20:07:32 +0000867 int flags, CPUWatchpoint **watchpoint)
pbrook6658ffb2007-03-16 23:58:11 +0000868{
aliguoric0ce9982008-11-25 22:13:57 +0000869 CPUWatchpoint *wp;
pbrook6658ffb2007-03-16 23:58:11 +0000870
Peter Maydell05068c02014-09-12 14:06:48 +0100871 /* forbid ranges which are empty or run off the end of the address space */
Max Filippov07e28632014-09-17 22:03:36 -0700872 if (len == 0 || (addr + len - 1) < addr) {
Andreas Färber75a34032013-09-02 16:57:02 +0200873 error_report("tried to set invalid watchpoint at %"
874 VADDR_PRIx ", len=%" VADDR_PRIu, addr, len);
aliguorib4051332008-11-18 20:14:20 +0000875 return -EINVAL;
876 }
Anthony Liguori7267c092011-08-20 22:09:37 -0500877 wp = g_malloc(sizeof(*wp));
pbrook6658ffb2007-03-16 23:58:11 +0000878
aliguoria1d1bb32008-11-18 20:07:32 +0000879 wp->vaddr = addr;
Peter Maydell05068c02014-09-12 14:06:48 +0100880 wp->len = len;
aliguoria1d1bb32008-11-18 20:07:32 +0000881 wp->flags = flags;
882
aliguori2dc9f412008-11-18 20:56:59 +0000883 /* keep all GDB-injected watchpoints in front */
Andreas Färberff4700b2013-08-26 18:23:18 +0200884 if (flags & BP_GDB) {
885 QTAILQ_INSERT_HEAD(&cpu->watchpoints, wp, entry);
886 } else {
887 QTAILQ_INSERT_TAIL(&cpu->watchpoints, wp, entry);
888 }
aliguoria1d1bb32008-11-18 20:07:32 +0000889
Andreas Färber31b030d2013-09-04 01:29:02 +0200890 tlb_flush_page(cpu, addr);
aliguoria1d1bb32008-11-18 20:07:32 +0000891
892 if (watchpoint)
893 *watchpoint = wp;
894 return 0;
pbrook6658ffb2007-03-16 23:58:11 +0000895}
896
aliguoria1d1bb32008-11-18 20:07:32 +0000897/* Remove a specific watchpoint. */
Andreas Färber75a34032013-09-02 16:57:02 +0200898int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
aliguoria1d1bb32008-11-18 20:07:32 +0000899 int flags)
pbrook6658ffb2007-03-16 23:58:11 +0000900{
aliguoria1d1bb32008-11-18 20:07:32 +0000901 CPUWatchpoint *wp;
pbrook6658ffb2007-03-16 23:58:11 +0000902
Andreas Färberff4700b2013-08-26 18:23:18 +0200903 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +0100904 if (addr == wp->vaddr && len == wp->len
aliguori6e140f22008-11-18 20:37:55 +0000905 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
Andreas Färber75a34032013-09-02 16:57:02 +0200906 cpu_watchpoint_remove_by_ref(cpu, wp);
pbrook6658ffb2007-03-16 23:58:11 +0000907 return 0;
908 }
909 }
aliguoria1d1bb32008-11-18 20:07:32 +0000910 return -ENOENT;
pbrook6658ffb2007-03-16 23:58:11 +0000911}
912
aliguoria1d1bb32008-11-18 20:07:32 +0000913/* Remove a specific watchpoint by reference. */
Andreas Färber75a34032013-09-02 16:57:02 +0200914void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
aliguoria1d1bb32008-11-18 20:07:32 +0000915{
Andreas Färberff4700b2013-08-26 18:23:18 +0200916 QTAILQ_REMOVE(&cpu->watchpoints, watchpoint, entry);
edgar_igl7d03f822008-05-17 18:58:29 +0000917
Andreas Färber31b030d2013-09-04 01:29:02 +0200918 tlb_flush_page(cpu, watchpoint->vaddr);
aliguoria1d1bb32008-11-18 20:07:32 +0000919
Anthony Liguori7267c092011-08-20 22:09:37 -0500920 g_free(watchpoint);
edgar_igl7d03f822008-05-17 18:58:29 +0000921}
922
aliguoria1d1bb32008-11-18 20:07:32 +0000923/* Remove all matching watchpoints. */
Andreas Färber75a34032013-09-02 16:57:02 +0200924void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
aliguoria1d1bb32008-11-18 20:07:32 +0000925{
aliguoric0ce9982008-11-25 22:13:57 +0000926 CPUWatchpoint *wp, *next;
aliguoria1d1bb32008-11-18 20:07:32 +0000927
Andreas Färberff4700b2013-08-26 18:23:18 +0200928 QTAILQ_FOREACH_SAFE(wp, &cpu->watchpoints, entry, next) {
Andreas Färber75a34032013-09-02 16:57:02 +0200929 if (wp->flags & mask) {
930 cpu_watchpoint_remove_by_ref(cpu, wp);
931 }
aliguoric0ce9982008-11-25 22:13:57 +0000932 }
aliguoria1d1bb32008-11-18 20:07:32 +0000933}
Peter Maydell05068c02014-09-12 14:06:48 +0100934
935/* Return true if this watchpoint address matches the specified
936 * access (ie the address range covered by the watchpoint overlaps
937 * partially or completely with the address range covered by the
938 * access).
939 */
940static inline bool cpu_watchpoint_address_matches(CPUWatchpoint *wp,
941 vaddr addr,
942 vaddr len)
943{
944 /* We know the lengths are non-zero, but a little caution is
945 * required to avoid errors in the case where the range ends
946 * exactly at the top of the address space and so addr + len
947 * wraps round to zero.
948 */
949 vaddr wpend = wp->vaddr + wp->len - 1;
950 vaddr addrend = addr + len - 1;
951
952 return !(addr > wpend || wp->vaddr > addrend);
953}
954
Paul Brookc527ee82010-03-01 03:31:14 +0000955#endif
aliguoria1d1bb32008-11-18 20:07:32 +0000956
957/* Add a breakpoint. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200958int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags,
aliguoria1d1bb32008-11-18 20:07:32 +0000959 CPUBreakpoint **breakpoint)
bellard4c3a88a2003-07-26 12:06:08 +0000960{
aliguoric0ce9982008-11-25 22:13:57 +0000961 CPUBreakpoint *bp;
ths3b46e622007-09-17 08:09:54 +0000962
Anthony Liguori7267c092011-08-20 22:09:37 -0500963 bp = g_malloc(sizeof(*bp));
aliguoria1d1bb32008-11-18 20:07:32 +0000964
965 bp->pc = pc;
966 bp->flags = flags;
967
aliguori2dc9f412008-11-18 20:56:59 +0000968 /* keep all GDB-injected breakpoints in front */
Andreas Färber00b941e2013-06-29 18:55:54 +0200969 if (flags & BP_GDB) {
Andreas Färberf0c3c502013-08-26 21:22:53 +0200970 QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry);
Andreas Färber00b941e2013-06-29 18:55:54 +0200971 } else {
Andreas Färberf0c3c502013-08-26 21:22:53 +0200972 QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry);
Andreas Färber00b941e2013-06-29 18:55:54 +0200973 }
aliguoria1d1bb32008-11-18 20:07:32 +0000974
Andreas Färberf0c3c502013-08-26 21:22:53 +0200975 breakpoint_invalidate(cpu, pc);
aliguoria1d1bb32008-11-18 20:07:32 +0000976
Andreas Färber00b941e2013-06-29 18:55:54 +0200977 if (breakpoint) {
aliguoria1d1bb32008-11-18 20:07:32 +0000978 *breakpoint = bp;
Andreas Färber00b941e2013-06-29 18:55:54 +0200979 }
aliguoria1d1bb32008-11-18 20:07:32 +0000980 return 0;
aliguoria1d1bb32008-11-18 20:07:32 +0000981}
982
983/* Remove a specific breakpoint. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200984int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags)
aliguoria1d1bb32008-11-18 20:07:32 +0000985{
aliguoria1d1bb32008-11-18 20:07:32 +0000986 CPUBreakpoint *bp;
987
Andreas Färberf0c3c502013-08-26 21:22:53 +0200988 QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) {
aliguoria1d1bb32008-11-18 20:07:32 +0000989 if (bp->pc == pc && bp->flags == flags) {
Andreas Färberb3310ab2013-09-02 17:26:20 +0200990 cpu_breakpoint_remove_by_ref(cpu, bp);
bellard4c3a88a2003-07-26 12:06:08 +0000991 return 0;
aliguoria1d1bb32008-11-18 20:07:32 +0000992 }
bellard4c3a88a2003-07-26 12:06:08 +0000993 }
aliguoria1d1bb32008-11-18 20:07:32 +0000994 return -ENOENT;
bellard4c3a88a2003-07-26 12:06:08 +0000995}
996
aliguoria1d1bb32008-11-18 20:07:32 +0000997/* Remove a specific breakpoint by reference. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200998void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *breakpoint)
bellard4c3a88a2003-07-26 12:06:08 +0000999{
Andreas Färberf0c3c502013-08-26 21:22:53 +02001000 QTAILQ_REMOVE(&cpu->breakpoints, breakpoint, entry);
1001
1002 breakpoint_invalidate(cpu, breakpoint->pc);
aliguoria1d1bb32008-11-18 20:07:32 +00001003
Anthony Liguori7267c092011-08-20 22:09:37 -05001004 g_free(breakpoint);
aliguoria1d1bb32008-11-18 20:07:32 +00001005}
1006
1007/* Remove all matching breakpoints. */
Andreas Färberb3310ab2013-09-02 17:26:20 +02001008void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
aliguoria1d1bb32008-11-18 20:07:32 +00001009{
aliguoric0ce9982008-11-25 22:13:57 +00001010 CPUBreakpoint *bp, *next;
aliguoria1d1bb32008-11-18 20:07:32 +00001011
Andreas Färberf0c3c502013-08-26 21:22:53 +02001012 QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) {
Andreas Färberb3310ab2013-09-02 17:26:20 +02001013 if (bp->flags & mask) {
1014 cpu_breakpoint_remove_by_ref(cpu, bp);
1015 }
aliguoric0ce9982008-11-25 22:13:57 +00001016 }
bellard4c3a88a2003-07-26 12:06:08 +00001017}
1018
bellardc33a3462003-07-29 20:50:33 +00001019/* enable or disable single step mode. EXCP_DEBUG is returned by the
1020 CPU loop after each instruction */
Andreas Färber3825b282013-06-24 18:41:06 +02001021void cpu_single_step(CPUState *cpu, int enabled)
bellardc33a3462003-07-29 20:50:33 +00001022{
Andreas Färbered2803d2013-06-21 20:20:45 +02001023 if (cpu->singlestep_enabled != enabled) {
1024 cpu->singlestep_enabled = enabled;
1025 if (kvm_enabled()) {
Stefan Weil38e478e2013-07-25 20:50:21 +02001026 kvm_update_guest_debug(cpu, 0);
Andreas Färbered2803d2013-06-21 20:20:45 +02001027 } else {
Stuart Bradyccbb4d42009-05-03 12:15:06 +01001028 /* must flush all the translated code to avoid inconsistencies */
aliguorie22a25c2009-03-12 20:12:48 +00001029 /* XXX: only flush what is necessary */
Peter Crosthwaitebbd77c12015-06-23 19:31:15 -07001030 tb_flush(cpu);
aliguorie22a25c2009-03-12 20:12:48 +00001031 }
bellardc33a3462003-07-29 20:50:33 +00001032 }
bellardc33a3462003-07-29 20:50:33 +00001033}
1034
Andreas Färbera47dddd2013-09-03 17:38:47 +02001035void cpu_abort(CPUState *cpu, const char *fmt, ...)
bellard75012672003-06-21 13:11:07 +00001036{
1037 va_list ap;
pbrook493ae1f2007-11-23 16:53:59 +00001038 va_list ap2;
bellard75012672003-06-21 13:11:07 +00001039
1040 va_start(ap, fmt);
pbrook493ae1f2007-11-23 16:53:59 +00001041 va_copy(ap2, ap);
bellard75012672003-06-21 13:11:07 +00001042 fprintf(stderr, "qemu: fatal: ");
1043 vfprintf(stderr, fmt, ap);
1044 fprintf(stderr, "\n");
Andreas Färber878096e2013-05-27 01:33:50 +02001045 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP);
Paolo Bonzini013a2942015-11-13 13:16:27 +01001046 if (qemu_log_separate()) {
Richard Henderson1ee73212016-09-22 15:17:10 -07001047 qemu_log_lock();
aliguori93fcfe32009-01-15 22:34:14 +00001048 qemu_log("qemu: fatal: ");
1049 qemu_log_vprintf(fmt, ap2);
1050 qemu_log("\n");
Andreas Färbera0762852013-06-16 07:28:50 +02001051 log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP);
aliguori31b1a7b2009-01-15 22:35:09 +00001052 qemu_log_flush();
Richard Henderson1ee73212016-09-22 15:17:10 -07001053 qemu_log_unlock();
aliguori93fcfe32009-01-15 22:34:14 +00001054 qemu_log_close();
balrog924edca2007-06-10 14:07:13 +00001055 }
pbrook493ae1f2007-11-23 16:53:59 +00001056 va_end(ap2);
j_mayerf9373292007-09-29 12:18:20 +00001057 va_end(ap);
Pavel Dovgalyuk76159362015-09-17 19:25:07 +03001058 replay_finish();
Riku Voipiofd052bf2010-01-25 14:30:49 +02001059#if defined(CONFIG_USER_ONLY)
1060 {
1061 struct sigaction act;
1062 sigfillset(&act.sa_mask);
1063 act.sa_handler = SIG_DFL;
1064 sigaction(SIGABRT, &act, NULL);
1065 }
1066#endif
bellard75012672003-06-21 13:11:07 +00001067 abort();
1068}
1069
bellard01243112004-01-04 15:48:17 +00001070#if !defined(CONFIG_USER_ONLY)
Mike Day0dc3f442013-09-05 14:41:35 -04001071/* Called from RCU critical section */
Paolo Bonzini041603f2013-09-09 17:49:45 +02001072static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
1073{
1074 RAMBlock *block;
1075
Paolo Bonzini43771532013-09-09 17:58:40 +02001076 block = atomic_rcu_read(&ram_list.mru_block);
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001077 if (block && addr - block->offset < block->max_length) {
Paolo Bonzini68851b92015-10-22 13:51:30 +02001078 return block;
Paolo Bonzini041603f2013-09-09 17:49:45 +02001079 }
Peter Xu99e15582017-05-12 12:17:39 +08001080 RAMBLOCK_FOREACH(block) {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001081 if (addr - block->offset < block->max_length) {
Paolo Bonzini041603f2013-09-09 17:49:45 +02001082 goto found;
1083 }
1084 }
1085
1086 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
1087 abort();
1088
1089found:
Paolo Bonzini43771532013-09-09 17:58:40 +02001090 /* It is safe to write mru_block outside the iothread lock. This
1091 * is what happens:
1092 *
1093 * mru_block = xxx
1094 * rcu_read_unlock()
1095 * xxx removed from list
1096 * rcu_read_lock()
1097 * read mru_block
1098 * mru_block = NULL;
1099 * call_rcu(reclaim_ramblock, xxx);
1100 * rcu_read_unlock()
1101 *
1102 * atomic_rcu_set is not needed here. The block was already published
1103 * when it was placed into the list. Here we're just making an extra
1104 * copy of the pointer.
1105 */
Paolo Bonzini041603f2013-09-09 17:49:45 +02001106 ram_list.mru_block = block;
1107 return block;
1108}
1109
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001110static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length)
bellard1ccde1c2004-02-06 19:46:14 +00001111{
Peter Crosthwaite9a135652015-09-10 22:39:41 -07001112 CPUState *cpu;
Paolo Bonzini041603f2013-09-09 17:49:45 +02001113 ram_addr_t start1;
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001114 RAMBlock *block;
1115 ram_addr_t end;
1116
1117 end = TARGET_PAGE_ALIGN(start + length);
1118 start &= TARGET_PAGE_MASK;
bellardf23db162005-08-21 19:12:28 +00001119
Mike Day0dc3f442013-09-05 14:41:35 -04001120 rcu_read_lock();
Paolo Bonzini041603f2013-09-09 17:49:45 +02001121 block = qemu_get_ram_block(start);
1122 assert(block == qemu_get_ram_block(end - 1));
Michael S. Tsirkin1240be22014-11-12 11:44:41 +02001123 start1 = (uintptr_t)ramblock_ptr(block, start - block->offset);
Peter Crosthwaite9a135652015-09-10 22:39:41 -07001124 CPU_FOREACH(cpu) {
1125 tlb_reset_dirty(cpu, start1, length);
1126 }
Mike Day0dc3f442013-09-05 14:41:35 -04001127 rcu_read_unlock();
Juan Quintelad24981d2012-05-22 00:42:40 +02001128}
1129
1130/* Note: start and end must be within the same ram block. */
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001131bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start,
1132 ram_addr_t length,
1133 unsigned client)
Juan Quintelad24981d2012-05-22 00:42:40 +02001134{
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001135 DirtyMemoryBlocks *blocks;
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001136 unsigned long end, page;
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001137 bool dirty = false;
Juan Quintelad24981d2012-05-22 00:42:40 +02001138
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001139 if (length == 0) {
1140 return false;
1141 }
1142
1143 end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
1144 page = start >> TARGET_PAGE_BITS;
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001145
1146 rcu_read_lock();
1147
1148 blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
1149
1150 while (page < end) {
1151 unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE;
1152 unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE;
1153 unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset);
1154
1155 dirty |= bitmap_test_and_clear_atomic(blocks->blocks[idx],
1156 offset, num);
1157 page += num;
1158 }
1159
1160 rcu_read_unlock();
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001161
1162 if (dirty && tcg_enabled()) {
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001163 tlb_reset_dirty_range_all(start, length);
Juan Quintelad24981d2012-05-22 00:42:40 +02001164 }
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001165
1166 return dirty;
bellard1ccde1c2004-02-06 19:46:14 +00001167}
1168
Gerd Hoffmann8deaf122017-04-21 11:16:25 +02001169DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty
1170 (ram_addr_t start, ram_addr_t length, unsigned client)
1171{
1172 DirtyMemoryBlocks *blocks;
1173 unsigned long align = 1UL << (TARGET_PAGE_BITS + BITS_PER_LEVEL);
1174 ram_addr_t first = QEMU_ALIGN_DOWN(start, align);
1175 ram_addr_t last = QEMU_ALIGN_UP(start + length, align);
1176 DirtyBitmapSnapshot *snap;
1177 unsigned long page, end, dest;
1178
1179 snap = g_malloc0(sizeof(*snap) +
1180 ((last - first) >> (TARGET_PAGE_BITS + 3)));
1181 snap->start = first;
1182 snap->end = last;
1183
1184 page = first >> TARGET_PAGE_BITS;
1185 end = last >> TARGET_PAGE_BITS;
1186 dest = 0;
1187
1188 rcu_read_lock();
1189
1190 blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
1191
1192 while (page < end) {
1193 unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE;
1194 unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE;
1195 unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset);
1196
1197 assert(QEMU_IS_ALIGNED(offset, (1 << BITS_PER_LEVEL)));
1198 assert(QEMU_IS_ALIGNED(num, (1 << BITS_PER_LEVEL)));
1199 offset >>= BITS_PER_LEVEL;
1200
1201 bitmap_copy_and_clear_atomic(snap->dirty + dest,
1202 blocks->blocks[idx] + offset,
1203 num);
1204 page += num;
1205 dest += num >> BITS_PER_LEVEL;
1206 }
1207
1208 rcu_read_unlock();
1209
1210 if (tcg_enabled()) {
1211 tlb_reset_dirty_range_all(start, length);
1212 }
1213
1214 return snap;
1215}
1216
1217bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap,
1218 ram_addr_t start,
1219 ram_addr_t length)
1220{
1221 unsigned long page, end;
1222
1223 assert(start >= snap->start);
1224 assert(start + length <= snap->end);
1225
1226 end = TARGET_PAGE_ALIGN(start + length - snap->start) >> TARGET_PAGE_BITS;
1227 page = (start - snap->start) >> TARGET_PAGE_BITS;
1228
1229 while (page < end) {
1230 if (test_bit(page, snap->dirty)) {
1231 return true;
1232 }
1233 page++;
1234 }
1235 return false;
1236}
1237
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01001238/* Called from RCU critical section */
Andreas Färberbb0e6272013-09-03 13:32:01 +02001239hwaddr memory_region_section_get_iotlb(CPUState *cpu,
Paolo Bonzini149f54b2013-05-24 12:59:37 +02001240 MemoryRegionSection *section,
1241 target_ulong vaddr,
1242 hwaddr paddr, hwaddr xlat,
1243 int prot,
1244 target_ulong *address)
Blue Swirle5548612012-04-21 13:08:33 +00001245{
Avi Kivitya8170e52012-10-23 12:30:10 +02001246 hwaddr iotlb;
Blue Swirle5548612012-04-21 13:08:33 +00001247 CPUWatchpoint *wp;
1248
Blue Swirlcc5bea62012-04-14 14:56:48 +00001249 if (memory_region_is_ram(section->mr)) {
Blue Swirle5548612012-04-21 13:08:33 +00001250 /* Normal RAM. */
Paolo Bonzinie4e69792016-03-01 10:44:50 +01001251 iotlb = memory_region_get_ram_addr(section->mr) + xlat;
Blue Swirle5548612012-04-21 13:08:33 +00001252 if (!section->readonly) {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001253 iotlb |= PHYS_SECTION_NOTDIRTY;
Blue Swirle5548612012-04-21 13:08:33 +00001254 } else {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001255 iotlb |= PHYS_SECTION_ROM;
Blue Swirle5548612012-04-21 13:08:33 +00001256 }
1257 } else {
Peter Maydell0b8e2c12015-07-20 12:27:16 +01001258 AddressSpaceDispatch *d;
1259
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10001260 d = flatview_to_dispatch(section->fv);
Peter Maydell0b8e2c12015-07-20 12:27:16 +01001261 iotlb = section - d->map.sections;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02001262 iotlb += xlat;
Blue Swirle5548612012-04-21 13:08:33 +00001263 }
1264
1265 /* Make accesses to pages with watchpoints go via the
1266 watchpoint trap routines. */
Andreas Färberff4700b2013-08-26 18:23:18 +02001267 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +01001268 if (cpu_watchpoint_address_matches(wp, vaddr, TARGET_PAGE_SIZE)) {
Blue Swirle5548612012-04-21 13:08:33 +00001269 /* Avoid trapping reads of pages with a write breakpoint. */
1270 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001271 iotlb = PHYS_SECTION_WATCH + paddr;
Blue Swirle5548612012-04-21 13:08:33 +00001272 *address |= TLB_MMIO;
1273 break;
1274 }
1275 }
1276 }
1277
1278 return iotlb;
1279}
bellard9fa3e852004-01-04 18:06:42 +00001280#endif /* defined(CONFIG_USER_ONLY) */
1281
pbrooke2eef172008-06-08 01:09:01 +00001282#if !defined(CONFIG_USER_ONLY)
pbrook8da3ff12008-12-01 18:59:50 +00001283
Anthony Liguoric227f092009-10-01 16:12:16 -05001284static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
Avi Kivity5312bd82012-02-12 18:32:55 +02001285 uint16_t section);
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10001286static subpage_t *subpage_init(FlatView *fv, hwaddr base);
Avi Kivity54688b12012-02-09 17:34:32 +02001287
Igor Mammedova2b257d2014-10-31 16:38:37 +00001288static void *(*phys_mem_alloc)(size_t size, uint64_t *align) =
1289 qemu_anon_ram_alloc;
Markus Armbruster91138032013-07-31 15:11:08 +02001290
1291/*
1292 * Set a custom physical guest memory alloator.
1293 * Accelerators with unusual needs may need this. Hopefully, we can
1294 * get rid of it eventually.
1295 */
Igor Mammedova2b257d2014-10-31 16:38:37 +00001296void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align))
Markus Armbruster91138032013-07-31 15:11:08 +02001297{
1298 phys_mem_alloc = alloc;
1299}
1300
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001301static uint16_t phys_section_add(PhysPageMap *map,
1302 MemoryRegionSection *section)
Avi Kivity5312bd82012-02-12 18:32:55 +02001303{
Paolo Bonzini68f3f652013-05-07 11:30:23 +02001304 /* The physical section number is ORed with a page-aligned
1305 * pointer to produce the iotlb entries. Thus it should
1306 * never overflow into the page-aligned value.
1307 */
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001308 assert(map->sections_nb < TARGET_PAGE_SIZE);
Paolo Bonzini68f3f652013-05-07 11:30:23 +02001309
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001310 if (map->sections_nb == map->sections_nb_alloc) {
1311 map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16);
1312 map->sections = g_renew(MemoryRegionSection, map->sections,
1313 map->sections_nb_alloc);
Avi Kivity5312bd82012-02-12 18:32:55 +02001314 }
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001315 map->sections[map->sections_nb] = *section;
Paolo Bonzinidfde4e62013-05-06 10:46:11 +02001316 memory_region_ref(section->mr);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001317 return map->sections_nb++;
Avi Kivity5312bd82012-02-12 18:32:55 +02001318}
1319
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001320static void phys_section_destroy(MemoryRegion *mr)
1321{
Don Slutz55b4e802015-11-30 17:11:04 -05001322 bool have_sub_page = mr->subpage;
1323
Paolo Bonzinidfde4e62013-05-06 10:46:11 +02001324 memory_region_unref(mr);
1325
Don Slutz55b4e802015-11-30 17:11:04 -05001326 if (have_sub_page) {
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001327 subpage_t *subpage = container_of(mr, subpage_t, iomem);
Peter Crosthwaiteb4fefef2014-06-05 23:15:52 -07001328 object_unref(OBJECT(&subpage->iomem));
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001329 g_free(subpage);
1330 }
1331}
1332
Paolo Bonzini60926662013-05-29 12:30:26 +02001333static void phys_sections_free(PhysPageMap *map)
Avi Kivity5312bd82012-02-12 18:32:55 +02001334{
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02001335 while (map->sections_nb > 0) {
1336 MemoryRegionSection *section = &map->sections[--map->sections_nb];
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001337 phys_section_destroy(section->mr);
1338 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02001339 g_free(map->sections);
1340 g_free(map->nodes);
Avi Kivity5312bd82012-02-12 18:32:55 +02001341}
1342
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001343static void register_subpage(FlatView *fv, MemoryRegionSection *section)
Avi Kivity0f0cb162012-02-13 17:14:32 +02001344{
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001345 AddressSpaceDispatch *d = flatview_to_dispatch(fv);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001346 subpage_t *subpage;
Avi Kivitya8170e52012-10-23 12:30:10 +02001347 hwaddr base = section->offset_within_address_space
Avi Kivity0f0cb162012-02-13 17:14:32 +02001348 & TARGET_PAGE_MASK;
Peter Xu003a0cf2017-05-15 16:50:57 +08001349 MemoryRegionSection *existing = phys_page_find(d, base);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001350 MemoryRegionSection subsection = {
1351 .offset_within_address_space = base,
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001352 .size = int128_make64(TARGET_PAGE_SIZE),
Avi Kivity0f0cb162012-02-13 17:14:32 +02001353 };
Avi Kivitya8170e52012-10-23 12:30:10 +02001354 hwaddr start, end;
Avi Kivity0f0cb162012-02-13 17:14:32 +02001355
Avi Kivityf3705d52012-03-08 16:16:34 +02001356 assert(existing->mr->subpage || existing->mr == &io_mem_unassigned);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001357
Avi Kivityf3705d52012-03-08 16:16:34 +02001358 if (!(existing->mr->subpage)) {
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10001359 subpage = subpage_init(fv, base);
1360 subsection.fv = fv;
Avi Kivity0f0cb162012-02-13 17:14:32 +02001361 subsection.mr = &subpage->iomem;
Avi Kivityac1970f2012-10-03 16:22:53 +02001362 phys_page_set(d, base >> TARGET_PAGE_BITS, 1,
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001363 phys_section_add(&d->map, &subsection));
Avi Kivity0f0cb162012-02-13 17:14:32 +02001364 } else {
Avi Kivityf3705d52012-03-08 16:16:34 +02001365 subpage = container_of(existing->mr, subpage_t, iomem);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001366 }
1367 start = section->offset_within_address_space & ~TARGET_PAGE_MASK;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001368 end = start + int128_get64(section->size) - 1;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001369 subpage_register(subpage, start, end,
1370 phys_section_add(&d->map, section));
Avi Kivity0f0cb162012-02-13 17:14:32 +02001371}
1372
1373
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001374static void register_multipage(FlatView *fv,
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001375 MemoryRegionSection *section)
bellard33417e72003-08-10 21:47:01 +00001376{
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001377 AddressSpaceDispatch *d = flatview_to_dispatch(fv);
Avi Kivitya8170e52012-10-23 12:30:10 +02001378 hwaddr start_addr = section->offset_within_address_space;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001379 uint16_t section_index = phys_section_add(&d->map, section);
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001380 uint64_t num_pages = int128_get64(int128_rshift(section->size,
1381 TARGET_PAGE_BITS));
Avi Kivitydd811242012-01-02 12:17:03 +02001382
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001383 assert(num_pages);
1384 phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index);
bellard33417e72003-08-10 21:47:01 +00001385}
1386
Alexey Kardashevskiy8629d3f2017-09-21 18:51:00 +10001387void flatview_add_to_dispatch(FlatView *fv, MemoryRegionSection *section)
Avi Kivity0f0cb162012-02-13 17:14:32 +02001388{
Paolo Bonzini99b9cc02013-05-27 13:18:01 +02001389 MemoryRegionSection now = *section, remain = *section;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001390 Int128 page_size = int128_make64(TARGET_PAGE_SIZE);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001391
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001392 if (now.offset_within_address_space & ~TARGET_PAGE_MASK) {
1393 uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space)
1394 - now.offset_within_address_space;
1395
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001396 now.size = int128_min(int128_make64(left), now.size);
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001397 register_subpage(fv, &now);
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001398 } else {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001399 now.size = int128_zero();
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001400 }
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001401 while (int128_ne(remain.size, now.size)) {
1402 remain.size = int128_sub(remain.size, now.size);
1403 remain.offset_within_address_space += int128_get64(now.size);
1404 remain.offset_within_region += int128_get64(now.size);
Tyler Hall69b67642012-07-25 18:45:04 -04001405 now = remain;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001406 if (int128_lt(remain.size, page_size)) {
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001407 register_subpage(fv, &now);
Hu Tao88266242013-08-29 18:21:16 +08001408 } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001409 now.size = page_size;
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001410 register_subpage(fv, &now);
Tyler Hall69b67642012-07-25 18:45:04 -04001411 } else {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001412 now.size = int128_and(now.size, int128_neg(page_size));
Alexey Kardashevskiy99503222017-09-21 18:50:59 +10001413 register_multipage(fv, &now);
Tyler Hall69b67642012-07-25 18:45:04 -04001414 }
Avi Kivity0f0cb162012-02-13 17:14:32 +02001415 }
1416}
1417
Sheng Yang62a27442010-01-26 19:21:16 +08001418void qemu_flush_coalesced_mmio_buffer(void)
1419{
1420 if (kvm_enabled())
1421 kvm_flush_coalesced_mmio_buffer();
1422}
1423
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001424void qemu_mutex_lock_ramlist(void)
1425{
1426 qemu_mutex_lock(&ram_list.mutex);
1427}
1428
1429void qemu_mutex_unlock_ramlist(void)
1430{
1431 qemu_mutex_unlock(&ram_list.mutex);
1432}
1433
Peter Xube9b23c2017-05-12 12:17:41 +08001434void ram_block_dump(Monitor *mon)
1435{
1436 RAMBlock *block;
1437 char *psize;
1438
1439 rcu_read_lock();
1440 monitor_printf(mon, "%24s %8s %18s %18s %18s\n",
1441 "Block Name", "PSize", "Offset", "Used", "Total");
1442 RAMBLOCK_FOREACH(block) {
1443 psize = size_to_str(block->page_size);
1444 monitor_printf(mon, "%24s %8s 0x%016" PRIx64 " 0x%016" PRIx64
1445 " 0x%016" PRIx64 "\n", block->idstr, psize,
1446 (uint64_t)block->offset,
1447 (uint64_t)block->used_length,
1448 (uint64_t)block->max_length);
1449 g_free(psize);
1450 }
1451 rcu_read_unlock();
1452}
1453
Markus Armbrustere1e84ba2013-07-31 15:11:10 +02001454#ifdef __linux__
Alexey Kardashevskiy9c607662017-03-02 13:36:11 +11001455/*
1456 * FIXME TOCTTOU: this iterates over memory backends' mem-path, which
1457 * may or may not name the same files / on the same filesystem now as
1458 * when we actually open and map them. Iterate over the file
1459 * descriptors instead, and use qemu_fd_getpagesize().
1460 */
1461static int find_max_supported_pagesize(Object *obj, void *opaque)
1462{
1463 char *mem_path;
1464 long *hpsize_min = opaque;
1465
1466 if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
1467 mem_path = object_property_get_str(obj, "mem-path", NULL);
1468 if (mem_path) {
1469 long hpsize = qemu_mempath_getpagesize(mem_path);
1470 if (hpsize < *hpsize_min) {
1471 *hpsize_min = hpsize;
1472 }
1473 } else {
1474 *hpsize_min = getpagesize();
1475 }
1476 }
1477
1478 return 0;
1479}
1480
1481long qemu_getrampagesize(void)
1482{
1483 long hpsize = LONG_MAX;
1484 long mainrampagesize;
1485 Object *memdev_root;
1486
1487 if (mem_path) {
1488 mainrampagesize = qemu_mempath_getpagesize(mem_path);
1489 } else {
1490 mainrampagesize = getpagesize();
1491 }
1492
1493 /* it's possible we have memory-backend objects with
1494 * hugepage-backed RAM. these may get mapped into system
1495 * address space via -numa parameters or memory hotplug
1496 * hooks. we want to take these into account, but we
1497 * also want to make sure these supported hugepage
1498 * sizes are applicable across the entire range of memory
1499 * we may boot from, so we take the min across all
1500 * backends, and assume normal pages in cases where a
1501 * backend isn't backed by hugepages.
1502 */
1503 memdev_root = object_resolve_path("/objects", NULL);
1504 if (memdev_root) {
1505 object_child_foreach(memdev_root, find_max_supported_pagesize, &hpsize);
1506 }
1507 if (hpsize == LONG_MAX) {
1508 /* No additional memory regions found ==> Report main RAM page size */
1509 return mainrampagesize;
1510 }
1511
1512 /* If NUMA is disabled or the NUMA nodes are not backed with a
1513 * memory-backend, then there is at least one node using "normal" RAM,
1514 * so if its page size is smaller we have got to report that size instead.
1515 */
1516 if (hpsize > mainrampagesize &&
1517 (nb_numa_nodes == 0 || numa_info[0].node_memdev == NULL)) {
1518 static bool warned;
1519 if (!warned) {
1520 error_report("Huge page support disabled (n/a for main memory).");
1521 warned = true;
1522 }
1523 return mainrampagesize;
1524 }
1525
1526 return hpsize;
1527}
1528#else
1529long qemu_getrampagesize(void)
1530{
1531 return getpagesize();
1532}
1533#endif
1534
1535#ifdef __linux__
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001536static int64_t get_file_size(int fd)
1537{
1538 int64_t size = lseek(fd, 0, SEEK_END);
1539 if (size < 0) {
1540 return -errno;
1541 }
1542 return size;
1543}
1544
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001545static int file_ram_open(const char *path,
1546 const char *region_name,
1547 bool *created,
1548 Error **errp)
Marcelo Tosattic9027602010-03-01 20:25:08 -03001549{
1550 char *filename;
Peter Feiner8ca761f2013-03-04 13:54:25 -05001551 char *sanitized_name;
1552 char *c;
Paolo Bonzini5c3ece72016-03-17 15:53:13 +01001553 int fd = -1;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001554
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001555 *created = false;
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001556 for (;;) {
1557 fd = open(path, O_RDWR);
1558 if (fd >= 0) {
1559 /* @path names an existing file, use it */
1560 break;
1561 }
1562 if (errno == ENOENT) {
1563 /* @path names a file that doesn't exist, create it */
1564 fd = open(path, O_RDWR | O_CREAT | O_EXCL, 0644);
1565 if (fd >= 0) {
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001566 *created = true;
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001567 break;
1568 }
1569 } else if (errno == EISDIR) {
1570 /* @path names a directory, create a file there */
1571 /* Make name safe to use with mkstemp by replacing '/' with '_'. */
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001572 sanitized_name = g_strdup(region_name);
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001573 for (c = sanitized_name; *c != '\0'; c++) {
1574 if (*c == '/') {
1575 *c = '_';
1576 }
1577 }
1578
1579 filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path,
1580 sanitized_name);
1581 g_free(sanitized_name);
1582
1583 fd = mkstemp(filename);
1584 if (fd >= 0) {
1585 unlink(filename);
1586 g_free(filename);
1587 break;
1588 }
1589 g_free(filename);
1590 }
1591 if (errno != EEXIST && errno != EINTR) {
1592 error_setg_errno(errp, errno,
1593 "can't open backing store %s for guest RAM",
1594 path);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001595 return -1;
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001596 }
1597 /*
1598 * Try again on EINTR and EEXIST. The latter happens when
1599 * something else creates the file between our two open().
1600 */
1601 }
1602
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001603 return fd;
1604}
1605
1606static void *file_ram_alloc(RAMBlock *block,
1607 ram_addr_t memory,
1608 int fd,
1609 bool truncate,
1610 Error **errp)
1611{
1612 void *area;
1613
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001614 block->page_size = qemu_fd_getpagesize(fd);
Haozhong Zhang83606682016-10-24 20:49:37 +08001615 block->mr->align = block->page_size;
1616#if defined(__s390x__)
1617 if (kvm_enabled()) {
1618 block->mr->align = MAX(block->mr->align, QEMU_VMALLOC_ALIGN);
1619 }
1620#endif
Marcelo Tosattic9027602010-03-01 20:25:08 -03001621
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001622 if (memory < block->page_size) {
Hu Tao557529d2014-09-09 13:28:00 +08001623 error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to "
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001624 "or larger than page size 0x%zx",
1625 memory, block->page_size);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001626 return NULL;
Haozhong Zhang1775f112016-11-02 09:05:51 +08001627 }
1628
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001629 memory = ROUND_UP(memory, block->page_size);
Marcelo Tosattic9027602010-03-01 20:25:08 -03001630
1631 /*
1632 * ftruncate is not supported by hugetlbfs in older
1633 * hosts, so don't bother bailing out on errors.
1634 * If anything goes wrong with it under other filesystems,
1635 * mmap will fail.
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001636 *
1637 * Do not truncate the non-empty backend file to avoid corrupting
1638 * the existing data in the file. Disabling shrinking is not
1639 * enough. For example, the current vNVDIMM implementation stores
1640 * the guest NVDIMM labels at the end of the backend file. If the
1641 * backend file is later extended, QEMU will not be able to find
1642 * those labels. Therefore, extending the non-empty backend file
1643 * is disabled as well.
Marcelo Tosattic9027602010-03-01 20:25:08 -03001644 */
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001645 if (truncate && ftruncate(fd, memory)) {
Yoshiaki Tamura9742bf22010-08-18 13:30:13 +09001646 perror("ftruncate");
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001647 }
Marcelo Tosattic9027602010-03-01 20:25:08 -03001648
Dominik Dingeld2f39ad2016-04-25 13:55:38 +02001649 area = qemu_ram_mmap(fd, memory, block->mr->align,
1650 block->flags & RAM_SHARED);
Marcelo Tosattic9027602010-03-01 20:25:08 -03001651 if (area == MAP_FAILED) {
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001652 error_setg_errno(errp, errno,
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001653 "unable to map backing store for guest RAM");
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001654 return NULL;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001655 }
Marcelo Tosattief36fa12013-10-28 18:51:46 -02001656
1657 if (mem_prealloc) {
Jitendra Kolhe1e356fc2017-02-24 09:01:43 +05301658 os_mem_prealloc(fd, area, memory, smp_cpus, errp);
Igor Mammedov056b68a2016-07-20 11:54:03 +02001659 if (errp && *errp) {
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001660 qemu_ram_munmap(area, memory);
1661 return NULL;
Igor Mammedov056b68a2016-07-20 11:54:03 +02001662 }
Marcelo Tosattief36fa12013-10-28 18:51:46 -02001663 }
1664
Alex Williamson04b16652010-07-02 11:13:17 -06001665 block->fd = fd;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001666 return area;
1667}
1668#endif
1669
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001670/* Allocate space within the ram_addr_t space that governs the
1671 * dirty bitmaps.
1672 * Called with the ramlist lock held.
1673 */
Alex Williamsond17b5282010-06-25 11:08:38 -06001674static ram_addr_t find_ram_offset(ram_addr_t size)
1675{
Alex Williamson04b16652010-07-02 11:13:17 -06001676 RAMBlock *block, *next_block;
Alex Williamson3e837b22011-10-31 08:54:09 -06001677 ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX;
Alex Williamson04b16652010-07-02 11:13:17 -06001678
Stefan Hajnoczi49cd9ac2013-03-11 10:20:21 +01001679 assert(size != 0); /* it would hand out same offset multiple times */
1680
Mike Day0dc3f442013-09-05 14:41:35 -04001681 if (QLIST_EMPTY_RCU(&ram_list.blocks)) {
Alex Williamson04b16652010-07-02 11:13:17 -06001682 return 0;
Mike Day0d53d9f2015-01-21 13:45:24 +01001683 }
Alex Williamson04b16652010-07-02 11:13:17 -06001684
Peter Xu99e15582017-05-12 12:17:39 +08001685 RAMBLOCK_FOREACH(block) {
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001686 ram_addr_t candidate, next = RAM_ADDR_MAX;
Alex Williamson04b16652010-07-02 11:13:17 -06001687
Dr. David Alan Gilbert801110a2018-01-05 17:01:38 +00001688 /* Align blocks to start on a 'long' in the bitmap
1689 * which makes the bitmap sync'ing take the fast path.
1690 */
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001691 candidate = block->offset + block->max_length;
Dr. David Alan Gilbert801110a2018-01-05 17:01:38 +00001692 candidate = ROUND_UP(candidate, BITS_PER_LONG << TARGET_PAGE_BITS);
Alex Williamson04b16652010-07-02 11:13:17 -06001693
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001694 /* Search for the closest following block
1695 * and find the gap.
1696 */
Peter Xu99e15582017-05-12 12:17:39 +08001697 RAMBLOCK_FOREACH(next_block) {
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001698 if (next_block->offset >= candidate) {
Alex Williamson04b16652010-07-02 11:13:17 -06001699 next = MIN(next, next_block->offset);
1700 }
1701 }
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001702
1703 /* If it fits remember our place and remember the size
1704 * of gap, but keep going so that we might find a smaller
1705 * gap to fill so avoiding fragmentation.
1706 */
1707 if (next - candidate >= size && next - candidate < mingap) {
1708 offset = candidate;
1709 mingap = next - candidate;
Alex Williamson04b16652010-07-02 11:13:17 -06001710 }
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001711
1712 trace_find_ram_offset_loop(size, candidate, offset, next, mingap);
Alex Williamson04b16652010-07-02 11:13:17 -06001713 }
Alex Williamson3e837b22011-10-31 08:54:09 -06001714
1715 if (offset == RAM_ADDR_MAX) {
1716 fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n",
1717 (uint64_t)size);
1718 abort();
1719 }
1720
Dr. David Alan Gilbert154cc9e2018-01-05 17:01:37 +00001721 trace_find_ram_offset(size, offset);
1722
Alex Williamson04b16652010-07-02 11:13:17 -06001723 return offset;
1724}
1725
Juan Quintelab8c48992017-03-21 17:44:30 +01001726unsigned long last_ram_page(void)
Alex Williamson04b16652010-07-02 11:13:17 -06001727{
Alex Williamsond17b5282010-06-25 11:08:38 -06001728 RAMBlock *block;
1729 ram_addr_t last = 0;
1730
Mike Day0dc3f442013-09-05 14:41:35 -04001731 rcu_read_lock();
Peter Xu99e15582017-05-12 12:17:39 +08001732 RAMBLOCK_FOREACH(block) {
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001733 last = MAX(last, block->offset + block->max_length);
Mike Day0d53d9f2015-01-21 13:45:24 +01001734 }
Mike Day0dc3f442013-09-05 14:41:35 -04001735 rcu_read_unlock();
Juan Quintelab8c48992017-03-21 17:44:30 +01001736 return last >> TARGET_PAGE_BITS;
Alex Williamsond17b5282010-06-25 11:08:38 -06001737}
1738
Jason Baronddb97f12012-08-02 15:44:16 -04001739static void qemu_ram_setup_dump(void *addr, ram_addr_t size)
1740{
1741 int ret;
Jason Baronddb97f12012-08-02 15:44:16 -04001742
1743 /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */
Marcel Apfelbaum47c8ca52015-02-04 17:43:54 +02001744 if (!machine_dump_guest_core(current_machine)) {
Jason Baronddb97f12012-08-02 15:44:16 -04001745 ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP);
1746 if (ret) {
1747 perror("qemu_madvise");
1748 fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, "
1749 "but dump_guest_core=off specified\n");
1750 }
1751 }
1752}
1753
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00001754const char *qemu_ram_get_idstr(RAMBlock *rb)
1755{
1756 return rb->idstr;
1757}
1758
Dr. David Alan Gilbert463a4ac2017-03-07 18:36:36 +00001759bool qemu_ram_is_shared(RAMBlock *rb)
1760{
1761 return rb->flags & RAM_SHARED;
1762}
1763
Mike Dayae3a7042013-09-05 14:41:35 -04001764/* Called with iothread lock held. */
Gongleifa53a0e2016-05-10 10:04:59 +08001765void qemu_ram_set_idstr(RAMBlock *new_block, const char *name, DeviceState *dev)
Hu Tao20cfe882014-04-02 15:13:26 +08001766{
Gongleifa53a0e2016-05-10 10:04:59 +08001767 RAMBlock *block;
Hu Tao20cfe882014-04-02 15:13:26 +08001768
Avi Kivityc5705a72011-12-20 15:59:12 +02001769 assert(new_block);
1770 assert(!new_block->idstr[0]);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001771
Anthony Liguori09e5ab62012-02-03 12:28:43 -06001772 if (dev) {
1773 char *id = qdev_get_dev_path(dev);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001774 if (id) {
1775 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
Anthony Liguori7267c092011-08-20 22:09:37 -05001776 g_free(id);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001777 }
1778 }
1779 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
1780
Gongleiab0a9952016-05-10 10:05:00 +08001781 rcu_read_lock();
Peter Xu99e15582017-05-12 12:17:39 +08001782 RAMBLOCK_FOREACH(block) {
Gongleifa53a0e2016-05-10 10:04:59 +08001783 if (block != new_block &&
1784 !strcmp(block->idstr, new_block->idstr)) {
Cam Macdonell84b89d72010-07-26 18:10:57 -06001785 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
1786 new_block->idstr);
1787 abort();
1788 }
1789 }
Mike Day0dc3f442013-09-05 14:41:35 -04001790 rcu_read_unlock();
Avi Kivityc5705a72011-12-20 15:59:12 +02001791}
1792
Mike Dayae3a7042013-09-05 14:41:35 -04001793/* Called with iothread lock held. */
Gongleifa53a0e2016-05-10 10:04:59 +08001794void qemu_ram_unset_idstr(RAMBlock *block)
Hu Tao20cfe882014-04-02 15:13:26 +08001795{
Mike Dayae3a7042013-09-05 14:41:35 -04001796 /* FIXME: arch_init.c assumes that this is not called throughout
1797 * migration. Ignore the problem since hot-unplug during migration
1798 * does not work anyway.
1799 */
Hu Tao20cfe882014-04-02 15:13:26 +08001800 if (block) {
1801 memset(block->idstr, 0, sizeof(block->idstr));
1802 }
1803}
1804
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001805size_t qemu_ram_pagesize(RAMBlock *rb)
1806{
1807 return rb->page_size;
1808}
1809
Dr. David Alan Gilbert67f11b52017-02-24 18:28:34 +00001810/* Returns the largest size of page in use */
1811size_t qemu_ram_pagesize_largest(void)
1812{
1813 RAMBlock *block;
1814 size_t largest = 0;
1815
Peter Xu99e15582017-05-12 12:17:39 +08001816 RAMBLOCK_FOREACH(block) {
Dr. David Alan Gilbert67f11b52017-02-24 18:28:34 +00001817 largest = MAX(largest, qemu_ram_pagesize(block));
1818 }
1819
1820 return largest;
1821}
1822
Luiz Capitulino8490fc72012-09-05 16:50:16 -03001823static int memory_try_enable_merging(void *addr, size_t len)
1824{
Marcel Apfelbaum75cc7f02015-02-04 17:43:55 +02001825 if (!machine_mem_merge(current_machine)) {
Luiz Capitulino8490fc72012-09-05 16:50:16 -03001826 /* disabled by the user */
1827 return 0;
1828 }
1829
1830 return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE);
1831}
1832
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001833/* Only legal before guest might have detected the memory size: e.g. on
1834 * incoming migration, or right after reset.
1835 *
1836 * As memory core doesn't know how is memory accessed, it is up to
1837 * resize callback to update device state and/or add assertions to detect
1838 * misuse, if necessary.
1839 */
Gongleifa53a0e2016-05-10 10:04:59 +08001840int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp)
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001841{
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001842 assert(block);
1843
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00001844 newsize = HOST_PAGE_ALIGN(newsize);
Michael S. Tsirkin129ddaf2015-02-17 10:15:30 +01001845
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001846 if (block->used_length == newsize) {
1847 return 0;
1848 }
1849
1850 if (!(block->flags & RAM_RESIZEABLE)) {
1851 error_setg_errno(errp, EINVAL,
1852 "Length mismatch: %s: 0x" RAM_ADDR_FMT
1853 " in != 0x" RAM_ADDR_FMT, block->idstr,
1854 newsize, block->used_length);
1855 return -EINVAL;
1856 }
1857
1858 if (block->max_length < newsize) {
1859 error_setg_errno(errp, EINVAL,
1860 "Length too large: %s: 0x" RAM_ADDR_FMT
1861 " > 0x" RAM_ADDR_FMT, block->idstr,
1862 newsize, block->max_length);
1863 return -EINVAL;
1864 }
1865
1866 cpu_physical_memory_clear_dirty_range(block->offset, block->used_length);
1867 block->used_length = newsize;
Paolo Bonzini58d27072015-03-23 11:56:01 +01001868 cpu_physical_memory_set_dirty_range(block->offset, block->used_length,
1869 DIRTY_CLIENTS_ALL);
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001870 memory_region_set_size(block->mr, newsize);
1871 if (block->resized) {
1872 block->resized(block->idstr, newsize, block->host);
1873 }
1874 return 0;
1875}
1876
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001877/* Called with ram_list.mutex held */
1878static void dirty_memory_extend(ram_addr_t old_ram_size,
1879 ram_addr_t new_ram_size)
1880{
1881 ram_addr_t old_num_blocks = DIV_ROUND_UP(old_ram_size,
1882 DIRTY_MEMORY_BLOCK_SIZE);
1883 ram_addr_t new_num_blocks = DIV_ROUND_UP(new_ram_size,
1884 DIRTY_MEMORY_BLOCK_SIZE);
1885 int i;
1886
1887 /* Only need to extend if block count increased */
1888 if (new_num_blocks <= old_num_blocks) {
1889 return;
1890 }
1891
1892 for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
1893 DirtyMemoryBlocks *old_blocks;
1894 DirtyMemoryBlocks *new_blocks;
1895 int j;
1896
1897 old_blocks = atomic_rcu_read(&ram_list.dirty_memory[i]);
1898 new_blocks = g_malloc(sizeof(*new_blocks) +
1899 sizeof(new_blocks->blocks[0]) * new_num_blocks);
1900
1901 if (old_num_blocks) {
1902 memcpy(new_blocks->blocks, old_blocks->blocks,
1903 old_num_blocks * sizeof(old_blocks->blocks[0]));
1904 }
1905
1906 for (j = old_num_blocks; j < new_num_blocks; j++) {
1907 new_blocks->blocks[j] = bitmap_new(DIRTY_MEMORY_BLOCK_SIZE);
1908 }
1909
1910 atomic_rcu_set(&ram_list.dirty_memory[i], new_blocks);
1911
1912 if (old_blocks) {
1913 g_free_rcu(old_blocks, rcu);
1914 }
1915 }
1916}
1917
Fam Zheng528f46a2016-03-01 14:18:18 +08001918static void ram_block_add(RAMBlock *new_block, Error **errp)
Avi Kivityc5705a72011-12-20 15:59:12 +02001919{
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001920 RAMBlock *block;
Mike Day0d53d9f2015-01-21 13:45:24 +01001921 RAMBlock *last_block = NULL;
Juan Quintela2152f5c2013-10-08 13:52:02 +02001922 ram_addr_t old_ram_size, new_ram_size;
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001923 Error *err = NULL;
Juan Quintela2152f5c2013-10-08 13:52:02 +02001924
Juan Quintelab8c48992017-03-21 17:44:30 +01001925 old_ram_size = last_ram_page();
Avi Kivityc5705a72011-12-20 15:59:12 +02001926
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001927 qemu_mutex_lock_ramlist();
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001928 new_block->offset = find_ram_offset(new_block->max_length);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001929
1930 if (!new_block->host) {
1931 if (xen_enabled()) {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001932 xen_ram_alloc(new_block->offset, new_block->max_length,
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001933 new_block->mr, &err);
1934 if (err) {
1935 error_propagate(errp, err);
1936 qemu_mutex_unlock_ramlist();
Paolo Bonzini39c350e2016-03-09 18:14:01 +01001937 return;
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001938 }
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001939 } else {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001940 new_block->host = phys_mem_alloc(new_block->max_length,
Igor Mammedova2b257d2014-10-31 16:38:37 +00001941 &new_block->mr->align);
Markus Armbruster39228252013-07-31 15:11:11 +02001942 if (!new_block->host) {
Hu Taoef701d72014-09-09 13:27:54 +08001943 error_setg_errno(errp, errno,
1944 "cannot set up guest memory '%s'",
1945 memory_region_name(new_block->mr));
1946 qemu_mutex_unlock_ramlist();
Paolo Bonzini39c350e2016-03-09 18:14:01 +01001947 return;
Markus Armbruster39228252013-07-31 15:11:11 +02001948 }
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001949 memory_try_enable_merging(new_block->host, new_block->max_length);
Yoshiaki Tamura6977dfe2010-08-18 15:41:49 +09001950 }
1951 }
Cam Macdonell84b89d72010-07-26 18:10:57 -06001952
Li Zhijiandd631692015-07-02 20:18:06 +08001953 new_ram_size = MAX(old_ram_size,
1954 (new_block->offset + new_block->max_length) >> TARGET_PAGE_BITS);
1955 if (new_ram_size > old_ram_size) {
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001956 dirty_memory_extend(old_ram_size, new_ram_size);
Li Zhijiandd631692015-07-02 20:18:06 +08001957 }
Mike Day0d53d9f2015-01-21 13:45:24 +01001958 /* Keep the list sorted from biggest to smallest block. Unlike QTAILQ,
1959 * QLIST (which has an RCU-friendly variant) does not have insertion at
1960 * tail, so save the last element in last_block.
1961 */
Peter Xu99e15582017-05-12 12:17:39 +08001962 RAMBLOCK_FOREACH(block) {
Mike Day0d53d9f2015-01-21 13:45:24 +01001963 last_block = block;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001964 if (block->max_length < new_block->max_length) {
Paolo Bonziniabb26d62012-11-14 16:00:51 +01001965 break;
1966 }
1967 }
1968 if (block) {
Mike Day0dc3f442013-09-05 14:41:35 -04001969 QLIST_INSERT_BEFORE_RCU(block, new_block, next);
Mike Day0d53d9f2015-01-21 13:45:24 +01001970 } else if (last_block) {
Mike Day0dc3f442013-09-05 14:41:35 -04001971 QLIST_INSERT_AFTER_RCU(last_block, new_block, next);
Mike Day0d53d9f2015-01-21 13:45:24 +01001972 } else { /* list is empty */
Mike Day0dc3f442013-09-05 14:41:35 -04001973 QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next);
Paolo Bonziniabb26d62012-11-14 16:00:51 +01001974 }
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01001975 ram_list.mru_block = NULL;
Cam Macdonell84b89d72010-07-26 18:10:57 -06001976
Mike Day0dc3f442013-09-05 14:41:35 -04001977 /* Write list before version */
1978 smp_wmb();
Umesh Deshpandef798b072011-08-18 11:41:17 -07001979 ram_list.version++;
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001980 qemu_mutex_unlock_ramlist();
Umesh Deshpandef798b072011-08-18 11:41:17 -07001981
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001982 cpu_physical_memory_set_dirty_range(new_block->offset,
Paolo Bonzini58d27072015-03-23 11:56:01 +01001983 new_block->used_length,
1984 DIRTY_CLIENTS_ALL);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001985
Paolo Bonzinia904c912015-01-21 16:18:35 +01001986 if (new_block->host) {
1987 qemu_ram_setup_dump(new_block->host, new_block->max_length);
1988 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_HUGEPAGE);
Cao jinc2cd6272016-09-12 14:34:56 +08001989 /* MADV_DONTFORK is also needed by KVM in absence of synchronous MMU */
Paolo Bonzinia904c912015-01-21 16:18:35 +01001990 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_DONTFORK);
Paolo Bonzini0987d732016-12-21 00:31:36 +08001991 ram_block_notify_add(new_block->host, new_block->max_length);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001992 }
Cam Macdonell84b89d72010-07-26 18:10:57 -06001993}
1994
Paolo Bonzini0b183fc2014-05-14 17:43:19 +08001995#ifdef __linux__
Marc-André Lureau38b33622017-06-02 18:12:23 +04001996RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, MemoryRegion *mr,
1997 bool share, int fd,
1998 Error **errp)
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001999{
2000 RAMBlock *new_block;
Hu Taoef701d72014-09-09 13:27:54 +08002001 Error *local_err = NULL;
Marc-André Lureau8d37b032017-06-02 18:12:22 +04002002 int64_t file_size;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002003
2004 if (xen_enabled()) {
Paolo Bonzini7f56e742014-05-14 17:43:20 +08002005 error_setg(errp, "-mem-path not supported with Xen");
Fam Zheng528f46a2016-03-01 14:18:18 +08002006 return NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002007 }
2008
Marc-André Lureaue45e7ae2017-06-02 18:12:21 +04002009 if (kvm_enabled() && !kvm_has_sync_mmu()) {
2010 error_setg(errp,
2011 "host lacks kvm mmu notifiers, -mem-path unsupported");
2012 return NULL;
2013 }
2014
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002015 if (phys_mem_alloc != qemu_anon_ram_alloc) {
2016 /*
2017 * file_ram_alloc() needs to allocate just like
2018 * phys_mem_alloc, but we haven't bothered to provide
2019 * a hook there.
2020 */
Paolo Bonzini7f56e742014-05-14 17:43:20 +08002021 error_setg(errp,
2022 "-mem-path not supported with this accelerator");
Fam Zheng528f46a2016-03-01 14:18:18 +08002023 return NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002024 }
2025
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00002026 size = HOST_PAGE_ALIGN(size);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04002027 file_size = get_file_size(fd);
2028 if (file_size > 0 && file_size < size) {
2029 error_setg(errp, "backing store %s size 0x%" PRIx64
2030 " does not match 'size' option 0x" RAM_ADDR_FMT,
2031 mem_path, file_size, size);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04002032 return NULL;
2033 }
2034
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002035 new_block = g_malloc0(sizeof(*new_block));
2036 new_block->mr = mr;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002037 new_block->used_length = size;
2038 new_block->max_length = size;
Paolo Bonzinidbcb8982014-06-10 19:15:24 +08002039 new_block->flags = share ? RAM_SHARED : 0;
Marc-André Lureau8d37b032017-06-02 18:12:22 +04002040 new_block->host = file_ram_alloc(new_block, size, fd, !file_size, errp);
Paolo Bonzini7f56e742014-05-14 17:43:20 +08002041 if (!new_block->host) {
2042 g_free(new_block);
Fam Zheng528f46a2016-03-01 14:18:18 +08002043 return NULL;
Paolo Bonzini7f56e742014-05-14 17:43:20 +08002044 }
2045
Fam Zheng528f46a2016-03-01 14:18:18 +08002046 ram_block_add(new_block, &local_err);
Hu Taoef701d72014-09-09 13:27:54 +08002047 if (local_err) {
2048 g_free(new_block);
2049 error_propagate(errp, local_err);
Fam Zheng528f46a2016-03-01 14:18:18 +08002050 return NULL;
Hu Taoef701d72014-09-09 13:27:54 +08002051 }
Fam Zheng528f46a2016-03-01 14:18:18 +08002052 return new_block;
Marc-André Lureau38b33622017-06-02 18:12:23 +04002053
2054}
2055
2056
2057RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
2058 bool share, const char *mem_path,
2059 Error **errp)
2060{
2061 int fd;
2062 bool created;
2063 RAMBlock *block;
2064
2065 fd = file_ram_open(mem_path, memory_region_name(mr), &created, errp);
2066 if (fd < 0) {
2067 return NULL;
2068 }
2069
2070 block = qemu_ram_alloc_from_fd(size, mr, share, fd, errp);
2071 if (!block) {
2072 if (created) {
2073 unlink(mem_path);
2074 }
2075 close(fd);
2076 return NULL;
2077 }
2078
2079 return block;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002080}
Paolo Bonzini0b183fc2014-05-14 17:43:19 +08002081#endif
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002082
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002083static
Fam Zheng528f46a2016-03-01 14:18:18 +08002084RAMBlock *qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size,
2085 void (*resized)(const char*,
2086 uint64_t length,
2087 void *host),
2088 void *host, bool resizeable,
2089 MemoryRegion *mr, Error **errp)
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002090{
2091 RAMBlock *new_block;
Hu Taoef701d72014-09-09 13:27:54 +08002092 Error *local_err = NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002093
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00002094 size = HOST_PAGE_ALIGN(size);
2095 max_size = HOST_PAGE_ALIGN(max_size);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002096 new_block = g_malloc0(sizeof(*new_block));
2097 new_block->mr = mr;
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002098 new_block->resized = resized;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002099 new_block->used_length = size;
2100 new_block->max_length = max_size;
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002101 assert(max_size >= size);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002102 new_block->fd = -1;
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01002103 new_block->page_size = getpagesize();
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002104 new_block->host = host;
2105 if (host) {
Paolo Bonzini7bd4f432014-05-14 17:43:22 +08002106 new_block->flags |= RAM_PREALLOC;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002107 }
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002108 if (resizeable) {
2109 new_block->flags |= RAM_RESIZEABLE;
2110 }
Fam Zheng528f46a2016-03-01 14:18:18 +08002111 ram_block_add(new_block, &local_err);
Hu Taoef701d72014-09-09 13:27:54 +08002112 if (local_err) {
2113 g_free(new_block);
2114 error_propagate(errp, local_err);
Fam Zheng528f46a2016-03-01 14:18:18 +08002115 return NULL;
Hu Taoef701d72014-09-09 13:27:54 +08002116 }
Fam Zheng528f46a2016-03-01 14:18:18 +08002117 return new_block;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002118}
2119
Fam Zheng528f46a2016-03-01 14:18:18 +08002120RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002121 MemoryRegion *mr, Error **errp)
2122{
2123 return qemu_ram_alloc_internal(size, size, NULL, host, false, mr, errp);
2124}
2125
Fam Zheng528f46a2016-03-01 14:18:18 +08002126RAMBlock *qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr, Error **errp)
pbrook94a6b542009-04-11 17:15:54 +00002127{
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002128 return qemu_ram_alloc_internal(size, size, NULL, NULL, false, mr, errp);
2129}
2130
Fam Zheng528f46a2016-03-01 14:18:18 +08002131RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t maxsz,
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002132 void (*resized)(const char*,
2133 uint64_t length,
2134 void *host),
2135 MemoryRegion *mr, Error **errp)
2136{
2137 return qemu_ram_alloc_internal(size, maxsz, resized, NULL, true, mr, errp);
pbrook94a6b542009-04-11 17:15:54 +00002138}
bellarde9a1ab12007-02-08 23:08:38 +00002139
Paolo Bonzini43771532013-09-09 17:58:40 +02002140static void reclaim_ramblock(RAMBlock *block)
2141{
2142 if (block->flags & RAM_PREALLOC) {
2143 ;
2144 } else if (xen_enabled()) {
2145 xen_invalidate_map_cache_entry(block->host);
2146#ifndef _WIN32
2147 } else if (block->fd >= 0) {
Eduardo Habkost2f3a2bb2015-11-06 20:11:21 -02002148 qemu_ram_munmap(block->host, block->max_length);
Paolo Bonzini43771532013-09-09 17:58:40 +02002149 close(block->fd);
2150#endif
2151 } else {
2152 qemu_anon_ram_free(block->host, block->max_length);
2153 }
2154 g_free(block);
2155}
2156
Fam Zhengf1060c52016-03-01 14:18:22 +08002157void qemu_ram_free(RAMBlock *block)
bellarde9a1ab12007-02-08 23:08:38 +00002158{
Marc-André Lureau85bc2a12016-03-29 13:20:51 +02002159 if (!block) {
2160 return;
2161 }
2162
Paolo Bonzini0987d732016-12-21 00:31:36 +08002163 if (block->host) {
2164 ram_block_notify_remove(block->host, block->max_length);
2165 }
2166
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07002167 qemu_mutex_lock_ramlist();
Fam Zhengf1060c52016-03-01 14:18:22 +08002168 QLIST_REMOVE_RCU(block, next);
2169 ram_list.mru_block = NULL;
2170 /* Write list before version */
2171 smp_wmb();
2172 ram_list.version++;
2173 call_rcu(block, reclaim_ramblock, rcu);
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07002174 qemu_mutex_unlock_ramlist();
bellarde9a1ab12007-02-08 23:08:38 +00002175}
2176
Huang Yingcd19cfa2011-03-02 08:56:19 +01002177#ifndef _WIN32
2178void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
2179{
2180 RAMBlock *block;
2181 ram_addr_t offset;
2182 int flags;
2183 void *area, *vaddr;
2184
Peter Xu99e15582017-05-12 12:17:39 +08002185 RAMBLOCK_FOREACH(block) {
Huang Yingcd19cfa2011-03-02 08:56:19 +01002186 offset = addr - block->offset;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002187 if (offset < block->max_length) {
Michael S. Tsirkin1240be22014-11-12 11:44:41 +02002188 vaddr = ramblock_ptr(block, offset);
Paolo Bonzini7bd4f432014-05-14 17:43:22 +08002189 if (block->flags & RAM_PREALLOC) {
Huang Yingcd19cfa2011-03-02 08:56:19 +01002190 ;
Markus Armbrusterdfeaf2a2013-07-31 15:11:05 +02002191 } else if (xen_enabled()) {
2192 abort();
Huang Yingcd19cfa2011-03-02 08:56:19 +01002193 } else {
2194 flags = MAP_FIXED;
Markus Armbruster3435f392013-07-31 15:11:07 +02002195 if (block->fd >= 0) {
Paolo Bonzinidbcb8982014-06-10 19:15:24 +08002196 flags |= (block->flags & RAM_SHARED ?
2197 MAP_SHARED : MAP_PRIVATE);
Markus Armbruster3435f392013-07-31 15:11:07 +02002198 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2199 flags, block->fd, offset);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002200 } else {
Markus Armbruster2eb9fba2013-07-31 15:11:09 +02002201 /*
2202 * Remap needs to match alloc. Accelerators that
2203 * set phys_mem_alloc never remap. If they did,
2204 * we'd need a remap hook here.
2205 */
2206 assert(phys_mem_alloc == qemu_anon_ram_alloc);
2207
Huang Yingcd19cfa2011-03-02 08:56:19 +01002208 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
2209 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2210 flags, -1, 0);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002211 }
2212 if (area != vaddr) {
Anthony PERARDf15fbc42011-07-20 08:17:42 +00002213 fprintf(stderr, "Could not remap addr: "
2214 RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n",
Huang Yingcd19cfa2011-03-02 08:56:19 +01002215 length, addr);
2216 exit(1);
2217 }
Luiz Capitulino8490fc72012-09-05 16:50:16 -03002218 memory_try_enable_merging(vaddr, length);
Jason Baronddb97f12012-08-02 15:44:16 -04002219 qemu_ram_setup_dump(vaddr, length);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002220 }
Huang Yingcd19cfa2011-03-02 08:56:19 +01002221 }
2222 }
2223}
2224#endif /* !_WIN32 */
2225
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002226/* Return a host pointer to ram allocated with qemu_ram_alloc.
Mike Dayae3a7042013-09-05 14:41:35 -04002227 * This should not be used for general purpose DMA. Use address_space_map
2228 * or address_space_rw instead. For local memory (e.g. video ram) that the
2229 * device owns, use memory_region_get_ram_ptr.
Mike Day0dc3f442013-09-05 14:41:35 -04002230 *
Paolo Bonzini49b24af2015-12-16 10:30:47 +01002231 * Called within RCU critical section.
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002232 */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002233void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr)
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002234{
Gonglei3655cb92016-02-20 10:35:20 +08002235 RAMBlock *block = ram_block;
2236
2237 if (block == NULL) {
2238 block = qemu_get_ram_block(addr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002239 addr -= block->offset;
Gonglei3655cb92016-02-20 10:35:20 +08002240 }
Mike Dayae3a7042013-09-05 14:41:35 -04002241
2242 if (xen_enabled() && block->host == NULL) {
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002243 /* We need to check if the requested address is in the RAM
2244 * because we don't want to map the entire memory in QEMU.
2245 * In that case just map until the end of the page.
2246 */
2247 if (block->offset == 0) {
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002248 return xen_map_cache(addr, 0, 0, false);
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002249 }
Mike Dayae3a7042013-09-05 14:41:35 -04002250
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002251 block->host = xen_map_cache(block->offset, block->max_length, 1, false);
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002252 }
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002253 return ramblock_ptr(block, addr);
pbrookdc828ca2009-04-09 22:21:07 +00002254}
2255
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002256/* Return a host pointer to guest's ram. Similar to qemu_map_ram_ptr
Mike Dayae3a7042013-09-05 14:41:35 -04002257 * but takes a size argument.
Mike Day0dc3f442013-09-05 14:41:35 -04002258 *
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002259 * Called within RCU critical section.
Mike Dayae3a7042013-09-05 14:41:35 -04002260 */
Gonglei3655cb92016-02-20 10:35:20 +08002261static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr,
Anthony PERARDf5aa69b2017-07-26 17:53:26 +01002262 hwaddr *size, bool lock)
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002263{
Gonglei3655cb92016-02-20 10:35:20 +08002264 RAMBlock *block = ram_block;
Stefano Stabellini8ab934f2011-06-27 18:26:06 +01002265 if (*size == 0) {
2266 return NULL;
2267 }
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002268
Gonglei3655cb92016-02-20 10:35:20 +08002269 if (block == NULL) {
2270 block = qemu_get_ram_block(addr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002271 addr -= block->offset;
Gonglei3655cb92016-02-20 10:35:20 +08002272 }
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002273 *size = MIN(*size, block->max_length - addr);
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002274
2275 if (xen_enabled() && block->host == NULL) {
2276 /* We need to check if the requested address is in the RAM
2277 * because we don't want to map the entire memory in QEMU.
2278 * In that case just map the requested area.
2279 */
2280 if (block->offset == 0) {
Anthony PERARDf5aa69b2017-07-26 17:53:26 +01002281 return xen_map_cache(addr, *size, lock, lock);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002282 }
2283
Anthony PERARDf5aa69b2017-07-26 17:53:26 +01002284 block->host = xen_map_cache(block->offset, block->max_length, 1, lock);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002285 }
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002286
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002287 return ramblock_ptr(block, addr);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002288}
2289
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002290/*
2291 * Translates a host ptr back to a RAMBlock, a ram_addr and an offset
2292 * in that RAMBlock.
2293 *
2294 * ptr: Host pointer to look up
2295 * round_offset: If true round the result offset down to a page boundary
2296 * *ram_addr: set to result ram_addr
2297 * *offset: set to result offset within the RAMBlock
2298 *
2299 * Returns: RAMBlock (or NULL if not found)
Mike Dayae3a7042013-09-05 14:41:35 -04002300 *
2301 * By the time this function returns, the returned pointer is not protected
2302 * by RCU anymore. If the caller is not within an RCU critical section and
2303 * does not hold the iothread lock, it must have other means of protecting the
2304 * pointer, such as a reference to the region that includes the incoming
2305 * ram_addr_t.
2306 */
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002307RAMBlock *qemu_ram_block_from_host(void *ptr, bool round_offset,
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002308 ram_addr_t *offset)
pbrook5579c7f2009-04-11 14:47:08 +00002309{
pbrook94a6b542009-04-11 17:15:54 +00002310 RAMBlock *block;
2311 uint8_t *host = ptr;
2312
Jan Kiszka868bb332011-06-21 22:59:09 +02002313 if (xen_enabled()) {
Paolo Bonzinif615f392016-05-26 10:07:50 +02002314 ram_addr_t ram_addr;
Mike Day0dc3f442013-09-05 14:41:35 -04002315 rcu_read_lock();
Paolo Bonzinif615f392016-05-26 10:07:50 +02002316 ram_addr = xen_ram_addr_from_mapcache(ptr);
2317 block = qemu_get_ram_block(ram_addr);
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002318 if (block) {
Anthony PERARDd6b6aec2016-06-09 16:56:17 +01002319 *offset = ram_addr - block->offset;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002320 }
Mike Day0dc3f442013-09-05 14:41:35 -04002321 rcu_read_unlock();
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002322 return block;
Stefano Stabellini712c2b42011-05-19 18:35:46 +01002323 }
2324
Mike Day0dc3f442013-09-05 14:41:35 -04002325 rcu_read_lock();
2326 block = atomic_rcu_read(&ram_list.mru_block);
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002327 if (block && block->host && host - block->host < block->max_length) {
Paolo Bonzini23887b72013-05-06 14:28:39 +02002328 goto found;
2329 }
2330
Peter Xu99e15582017-05-12 12:17:39 +08002331 RAMBLOCK_FOREACH(block) {
Jun Nakajima432d2682010-08-31 16:41:25 +01002332 /* This case append when the block is not mapped. */
2333 if (block->host == NULL) {
2334 continue;
2335 }
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002336 if (host - block->host < block->max_length) {
Paolo Bonzini23887b72013-05-06 14:28:39 +02002337 goto found;
Alex Williamsonf471a172010-06-11 11:11:42 -06002338 }
pbrook94a6b542009-04-11 17:15:54 +00002339 }
Jun Nakajima432d2682010-08-31 16:41:25 +01002340
Mike Day0dc3f442013-09-05 14:41:35 -04002341 rcu_read_unlock();
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002342 return NULL;
Paolo Bonzini23887b72013-05-06 14:28:39 +02002343
2344found:
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002345 *offset = (host - block->host);
2346 if (round_offset) {
2347 *offset &= TARGET_PAGE_MASK;
2348 }
Mike Day0dc3f442013-09-05 14:41:35 -04002349 rcu_read_unlock();
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002350 return block;
2351}
2352
Dr. David Alan Gilberte3dd7492015-11-05 18:10:33 +00002353/*
2354 * Finds the named RAMBlock
2355 *
2356 * name: The name of RAMBlock to find
2357 *
2358 * Returns: RAMBlock (or NULL if not found)
2359 */
2360RAMBlock *qemu_ram_block_by_name(const char *name)
2361{
2362 RAMBlock *block;
2363
Peter Xu99e15582017-05-12 12:17:39 +08002364 RAMBLOCK_FOREACH(block) {
Dr. David Alan Gilberte3dd7492015-11-05 18:10:33 +00002365 if (!strcmp(name, block->idstr)) {
2366 return block;
2367 }
2368 }
2369
2370 return NULL;
2371}
2372
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002373/* Some of the softmmu routines need to translate from a host pointer
2374 (typically a TLB entry) back to a ram offset. */
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002375ram_addr_t qemu_ram_addr_from_host(void *ptr)
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002376{
2377 RAMBlock *block;
Paolo Bonzinif615f392016-05-26 10:07:50 +02002378 ram_addr_t offset;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002379
Paolo Bonzinif615f392016-05-26 10:07:50 +02002380 block = qemu_ram_block_from_host(ptr, false, &offset);
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002381 if (!block) {
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002382 return RAM_ADDR_INVALID;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002383 }
2384
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002385 return block->offset + offset;
Marcelo Tosattie8902612010-10-11 15:31:19 -03002386}
Alex Williamsonf471a172010-06-11 11:11:42 -06002387
Peter Maydell27266272017-11-20 18:08:27 +00002388/* Called within RCU critical section. */
2389void memory_notdirty_write_prepare(NotDirtyInfo *ndi,
2390 CPUState *cpu,
2391 vaddr mem_vaddr,
2392 ram_addr_t ram_addr,
2393 unsigned size)
2394{
2395 ndi->cpu = cpu;
2396 ndi->ram_addr = ram_addr;
2397 ndi->mem_vaddr = mem_vaddr;
2398 ndi->size = size;
2399 ndi->locked = false;
2400
2401 assert(tcg_enabled());
2402 if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) {
2403 ndi->locked = true;
2404 tb_lock();
2405 tb_invalidate_phys_page_fast(ram_addr, size);
2406 }
2407}
2408
2409/* Called within RCU critical section. */
2410void memory_notdirty_write_complete(NotDirtyInfo *ndi)
2411{
2412 if (ndi->locked) {
2413 tb_unlock();
2414 }
2415
2416 /* Set both VGA and migration bits for simplicity and to remove
2417 * the notdirty callback faster.
2418 */
2419 cpu_physical_memory_set_dirty_range(ndi->ram_addr, ndi->size,
2420 DIRTY_CLIENTS_NOCODE);
2421 /* we remove the notdirty callback only if the code has been
2422 flushed */
2423 if (!cpu_physical_memory_is_clean(ndi->ram_addr)) {
2424 tlb_set_dirty(ndi->cpu, ndi->mem_vaddr);
2425 }
2426}
2427
Paolo Bonzini49b24af2015-12-16 10:30:47 +01002428/* Called within RCU critical section. */
Avi Kivitya8170e52012-10-23 12:30:10 +02002429static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002430 uint64_t val, unsigned size)
bellard1ccde1c2004-02-06 19:46:14 +00002431{
Peter Maydell27266272017-11-20 18:08:27 +00002432 NotDirtyInfo ndi;
Alex Bennéeba051fb2016-10-27 16:10:16 +01002433
Peter Maydell27266272017-11-20 18:08:27 +00002434 memory_notdirty_write_prepare(&ndi, current_cpu, current_cpu->mem_io_vaddr,
2435 ram_addr, size);
2436
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002437 switch (size) {
2438 case 1:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002439 stb_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002440 break;
2441 case 2:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002442 stw_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002443 break;
2444 case 4:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002445 stl_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002446 break;
Andrew Baumannad528782017-10-13 11:19:13 -07002447 case 8:
2448 stq_p(qemu_map_ram_ptr(NULL, ram_addr), val);
2449 break;
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002450 default:
2451 abort();
2452 }
Peter Maydell27266272017-11-20 18:08:27 +00002453 memory_notdirty_write_complete(&ndi);
bellard1ccde1c2004-02-06 19:46:14 +00002454}
2455
Paolo Bonzinib018ddf2013-05-24 14:48:38 +02002456static bool notdirty_mem_accepts(void *opaque, hwaddr addr,
2457 unsigned size, bool is_write)
2458{
2459 return is_write;
2460}
2461
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002462static const MemoryRegionOps notdirty_mem_ops = {
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002463 .write = notdirty_mem_write,
Paolo Bonzinib018ddf2013-05-24 14:48:38 +02002464 .valid.accepts = notdirty_mem_accepts,
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002465 .endianness = DEVICE_NATIVE_ENDIAN,
Andrew Baumannad528782017-10-13 11:19:13 -07002466 .valid = {
2467 .min_access_size = 1,
2468 .max_access_size = 8,
2469 .unaligned = false,
2470 },
2471 .impl = {
2472 .min_access_size = 1,
2473 .max_access_size = 8,
2474 .unaligned = false,
2475 },
bellard1ccde1c2004-02-06 19:46:14 +00002476};
2477
pbrook0f459d12008-06-09 00:20:13 +00002478/* Generate a debug exception if a watchpoint has been hit. */
Peter Maydell66b9b432015-04-26 16:49:24 +01002479static void check_watchpoint(int offset, int len, MemTxAttrs attrs, int flags)
pbrook0f459d12008-06-09 00:20:13 +00002480{
Andreas Färber93afead2013-08-26 03:41:01 +02002481 CPUState *cpu = current_cpu;
Sergey Fedorov568496c2016-02-11 11:17:32 +00002482 CPUClass *cc = CPU_GET_CLASS(cpu);
pbrook0f459d12008-06-09 00:20:13 +00002483 target_ulong vaddr;
aliguoria1d1bb32008-11-18 20:07:32 +00002484 CPUWatchpoint *wp;
pbrook0f459d12008-06-09 00:20:13 +00002485
Paolo Bonzini5aa1ef72017-07-03 17:50:40 +02002486 assert(tcg_enabled());
Andreas Färberff4700b2013-08-26 18:23:18 +02002487 if (cpu->watchpoint_hit) {
aliguori06d55cc2008-11-18 20:24:06 +00002488 /* We re-entered the check after replacing the TB. Now raise
2489 * the debug interrupt so that is will trigger after the
2490 * current instruction. */
Andreas Färber93afead2013-08-26 03:41:01 +02002491 cpu_interrupt(cpu, CPU_INTERRUPT_DEBUG);
aliguori06d55cc2008-11-18 20:24:06 +00002492 return;
2493 }
Andreas Färber93afead2013-08-26 03:41:01 +02002494 vaddr = (cpu->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
Julian Brown40612002017-02-07 18:29:59 +00002495 vaddr = cc->adjust_watchpoint_address(cpu, vaddr, len);
Andreas Färberff4700b2013-08-26 18:23:18 +02002496 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +01002497 if (cpu_watchpoint_address_matches(wp, vaddr, len)
2498 && (wp->flags & flags)) {
Peter Maydell08225672014-09-12 14:06:48 +01002499 if (flags == BP_MEM_READ) {
2500 wp->flags |= BP_WATCHPOINT_HIT_READ;
2501 } else {
2502 wp->flags |= BP_WATCHPOINT_HIT_WRITE;
2503 }
2504 wp->hitaddr = vaddr;
Peter Maydell66b9b432015-04-26 16:49:24 +01002505 wp->hitattrs = attrs;
Andreas Färberff4700b2013-08-26 18:23:18 +02002506 if (!cpu->watchpoint_hit) {
Sergey Fedorov568496c2016-02-11 11:17:32 +00002507 if (wp->flags & BP_CPU &&
2508 !cc->debug_check_watchpoint(cpu, wp)) {
2509 wp->flags &= ~BP_WATCHPOINT_HIT;
2510 continue;
2511 }
Andreas Färberff4700b2013-08-26 18:23:18 +02002512 cpu->watchpoint_hit = wp;
KONRAD Frederica5e99822016-10-27 16:10:06 +01002513
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002514 /* Both tb_lock and iothread_mutex will be reset when
2515 * cpu_loop_exit or cpu_loop_exit_noexc longjmp
2516 * back into the cpu_exec main loop.
KONRAD Frederica5e99822016-10-27 16:10:06 +01002517 */
2518 tb_lock();
Andreas Färber239c51a2013-09-01 17:12:23 +02002519 tb_check_watchpoint(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002520 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
Andreas Färber27103422013-08-26 08:31:06 +02002521 cpu->exception_index = EXCP_DEBUG;
Andreas Färber5638d182013-08-27 17:52:12 +02002522 cpu_loop_exit(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002523 } else {
Richard Henderson9b990ee2017-10-13 10:50:02 -07002524 /* Force execution of one insn next time. */
2525 cpu->cflags_next_tb = 1 | curr_cflags();
Peter Maydell6886b982016-05-17 15:18:04 +01002526 cpu_loop_exit_noexc(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002527 }
aliguori06d55cc2008-11-18 20:24:06 +00002528 }
aliguori6e140f22008-11-18 20:37:55 +00002529 } else {
2530 wp->flags &= ~BP_WATCHPOINT_HIT;
pbrook0f459d12008-06-09 00:20:13 +00002531 }
2532 }
2533}
2534
pbrook6658ffb2007-03-16 23:58:11 +00002535/* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
2536 so these check for a hit then pass through to the normal out-of-line
2537 phys routines. */
Peter Maydell66b9b432015-04-26 16:49:24 +01002538static MemTxResult watch_mem_read(void *opaque, hwaddr addr, uint64_t *pdata,
2539 unsigned size, MemTxAttrs attrs)
pbrook6658ffb2007-03-16 23:58:11 +00002540{
Peter Maydell66b9b432015-04-26 16:49:24 +01002541 MemTxResult res;
2542 uint64_t data;
Peter Maydell79ed0412016-01-21 14:15:06 +00002543 int asidx = cpu_asidx_from_attrs(current_cpu, attrs);
2544 AddressSpace *as = current_cpu->cpu_ases[asidx].as;
pbrook6658ffb2007-03-16 23:58:11 +00002545
Peter Maydell66b9b432015-04-26 16:49:24 +01002546 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_READ);
Avi Kivity1ec9b902012-01-02 12:47:48 +02002547 switch (size) {
Max Filippov67364152012-01-29 00:01:40 +04002548 case 1:
Peter Maydell79ed0412016-01-21 14:15:06 +00002549 data = address_space_ldub(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002550 break;
2551 case 2:
Peter Maydell79ed0412016-01-21 14:15:06 +00002552 data = address_space_lduw(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002553 break;
2554 case 4:
Peter Maydell79ed0412016-01-21 14:15:06 +00002555 data = address_space_ldl(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002556 break;
Paolo Bonzini306526b2017-10-17 14:16:05 +02002557 case 8:
2558 data = address_space_ldq(as, addr, attrs, &res);
2559 break;
Avi Kivity1ec9b902012-01-02 12:47:48 +02002560 default: abort();
2561 }
Peter Maydell66b9b432015-04-26 16:49:24 +01002562 *pdata = data;
2563 return res;
2564}
2565
2566static MemTxResult watch_mem_write(void *opaque, hwaddr addr,
2567 uint64_t val, unsigned size,
2568 MemTxAttrs attrs)
2569{
2570 MemTxResult res;
Peter Maydell79ed0412016-01-21 14:15:06 +00002571 int asidx = cpu_asidx_from_attrs(current_cpu, attrs);
2572 AddressSpace *as = current_cpu->cpu_ases[asidx].as;
Peter Maydell66b9b432015-04-26 16:49:24 +01002573
2574 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_WRITE);
2575 switch (size) {
2576 case 1:
Peter Maydell79ed0412016-01-21 14:15:06 +00002577 address_space_stb(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002578 break;
2579 case 2:
Peter Maydell79ed0412016-01-21 14:15:06 +00002580 address_space_stw(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002581 break;
2582 case 4:
Peter Maydell79ed0412016-01-21 14:15:06 +00002583 address_space_stl(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002584 break;
Paolo Bonzini306526b2017-10-17 14:16:05 +02002585 case 8:
2586 address_space_stq(as, addr, val, attrs, &res);
2587 break;
Peter Maydell66b9b432015-04-26 16:49:24 +01002588 default: abort();
2589 }
2590 return res;
pbrook6658ffb2007-03-16 23:58:11 +00002591}
2592
Avi Kivity1ec9b902012-01-02 12:47:48 +02002593static const MemoryRegionOps watch_mem_ops = {
Peter Maydell66b9b432015-04-26 16:49:24 +01002594 .read_with_attrs = watch_mem_read,
2595 .write_with_attrs = watch_mem_write,
Avi Kivity1ec9b902012-01-02 12:47:48 +02002596 .endianness = DEVICE_NATIVE_ENDIAN,
Paolo Bonzini306526b2017-10-17 14:16:05 +02002597 .valid = {
2598 .min_access_size = 1,
2599 .max_access_size = 8,
2600 .unaligned = false,
2601 },
2602 .impl = {
2603 .min_access_size = 1,
2604 .max_access_size = 8,
2605 .unaligned = false,
2606 },
pbrook6658ffb2007-03-16 23:58:11 +00002607};
pbrook6658ffb2007-03-16 23:58:11 +00002608
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002609static MemTxResult flatview_write(FlatView *fv, hwaddr addr, MemTxAttrs attrs,
2610 const uint8_t *buf, int len);
2611static bool flatview_access_valid(FlatView *fv, hwaddr addr, int len,
2612 bool is_write);
2613
Peter Maydellf25a49e2015-04-26 16:49:24 +01002614static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data,
2615 unsigned len, MemTxAttrs attrs)
blueswir1db7b5422007-05-26 17:36:03 +00002616{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002617 subpage_t *subpage = opaque;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002618 uint8_t buf[8];
Peter Maydell5c9eb022015-04-26 16:49:24 +01002619 MemTxResult res;
Paolo Bonzini791af8c2013-05-24 16:10:39 +02002620
blueswir1db7b5422007-05-26 17:36:03 +00002621#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002622 printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__,
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002623 subpage, len, addr);
blueswir1db7b5422007-05-26 17:36:03 +00002624#endif
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002625 res = flatview_read(subpage->fv, addr + subpage->base, attrs, buf, len);
Peter Maydell5c9eb022015-04-26 16:49:24 +01002626 if (res) {
2627 return res;
Peter Maydellf25a49e2015-04-26 16:49:24 +01002628 }
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002629 switch (len) {
2630 case 1:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002631 *data = ldub_p(buf);
2632 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002633 case 2:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002634 *data = lduw_p(buf);
2635 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002636 case 4:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002637 *data = ldl_p(buf);
2638 return MEMTX_OK;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002639 case 8:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002640 *data = ldq_p(buf);
2641 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002642 default:
2643 abort();
2644 }
blueswir1db7b5422007-05-26 17:36:03 +00002645}
2646
Peter Maydellf25a49e2015-04-26 16:49:24 +01002647static MemTxResult subpage_write(void *opaque, hwaddr addr,
2648 uint64_t value, unsigned len, MemTxAttrs attrs)
blueswir1db7b5422007-05-26 17:36:03 +00002649{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002650 subpage_t *subpage = opaque;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002651 uint8_t buf[8];
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002652
blueswir1db7b5422007-05-26 17:36:03 +00002653#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002654 printf("%s: subpage %p len %u addr " TARGET_FMT_plx
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002655 " value %"PRIx64"\n",
2656 __func__, subpage, len, addr, value);
blueswir1db7b5422007-05-26 17:36:03 +00002657#endif
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002658 switch (len) {
2659 case 1:
2660 stb_p(buf, value);
2661 break;
2662 case 2:
2663 stw_p(buf, value);
2664 break;
2665 case 4:
2666 stl_p(buf, value);
2667 break;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002668 case 8:
2669 stq_p(buf, value);
2670 break;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002671 default:
2672 abort();
2673 }
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002674 return flatview_write(subpage->fv, addr + subpage->base, attrs, buf, len);
blueswir1db7b5422007-05-26 17:36:03 +00002675}
2676
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002677static bool subpage_accepts(void *opaque, hwaddr addr,
Amos Kong016e9d62013-09-27 09:25:38 +08002678 unsigned len, bool is_write)
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002679{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002680 subpage_t *subpage = opaque;
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002681#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002682 printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n",
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002683 __func__, subpage, is_write ? 'w' : 'r', len, addr);
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002684#endif
2685
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002686 return flatview_access_valid(subpage->fv, addr + subpage->base,
2687 len, is_write);
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002688}
2689
Avi Kivity70c68e42012-01-02 12:32:48 +02002690static const MemoryRegionOps subpage_ops = {
Peter Maydellf25a49e2015-04-26 16:49:24 +01002691 .read_with_attrs = subpage_read,
2692 .write_with_attrs = subpage_write,
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002693 .impl.min_access_size = 1,
2694 .impl.max_access_size = 8,
2695 .valid.min_access_size = 1,
2696 .valid.max_access_size = 8,
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002697 .valid.accepts = subpage_accepts,
Avi Kivity70c68e42012-01-02 12:32:48 +02002698 .endianness = DEVICE_NATIVE_ENDIAN,
blueswir1db7b5422007-05-26 17:36:03 +00002699};
2700
Anthony Liguoric227f092009-10-01 16:12:16 -05002701static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
Avi Kivity5312bd82012-02-12 18:32:55 +02002702 uint16_t section)
blueswir1db7b5422007-05-26 17:36:03 +00002703{
2704 int idx, eidx;
2705
2706 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
2707 return -1;
2708 idx = SUBPAGE_IDX(start);
2709 eidx = SUBPAGE_IDX(end);
2710#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002711 printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n",
2712 __func__, mmio, start, end, idx, eidx, section);
blueswir1db7b5422007-05-26 17:36:03 +00002713#endif
blueswir1db7b5422007-05-26 17:36:03 +00002714 for (; idx <= eidx; idx++) {
Avi Kivity5312bd82012-02-12 18:32:55 +02002715 mmio->sub_section[idx] = section;
blueswir1db7b5422007-05-26 17:36:03 +00002716 }
2717
2718 return 0;
2719}
2720
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002721static subpage_t *subpage_init(FlatView *fv, hwaddr base)
blueswir1db7b5422007-05-26 17:36:03 +00002722{
Anthony Liguoric227f092009-10-01 16:12:16 -05002723 subpage_t *mmio;
blueswir1db7b5422007-05-26 17:36:03 +00002724
Vijaya Kumar K2615fab2016-10-24 16:26:49 +01002725 mmio = g_malloc0(sizeof(subpage_t) + TARGET_PAGE_SIZE * sizeof(uint16_t));
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002726 mmio->fv = fv;
aliguori1eec6142009-02-05 22:06:18 +00002727 mmio->base = base;
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002728 memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio,
Peter Crosthwaiteb4fefef2014-06-05 23:15:52 -07002729 NULL, TARGET_PAGE_SIZE);
Avi Kivityb3b00c72012-01-02 13:20:11 +02002730 mmio->iomem.subpage = true;
blueswir1db7b5422007-05-26 17:36:03 +00002731#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002732 printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__,
2733 mmio, base, TARGET_PAGE_SIZE);
blueswir1db7b5422007-05-26 17:36:03 +00002734#endif
Liu Ping Fanb41aac42013-05-29 11:09:17 +02002735 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED);
blueswir1db7b5422007-05-26 17:36:03 +00002736
2737 return mmio;
2738}
2739
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002740static uint16_t dummy_section(PhysPageMap *map, FlatView *fv, MemoryRegion *mr)
Avi Kivity5312bd82012-02-12 18:32:55 +02002741{
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002742 assert(fv);
Avi Kivity5312bd82012-02-12 18:32:55 +02002743 MemoryRegionSection section = {
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002744 .fv = fv,
Avi Kivity5312bd82012-02-12 18:32:55 +02002745 .mr = mr,
2746 .offset_within_address_space = 0,
2747 .offset_within_region = 0,
Paolo Bonzini052e87b2013-05-27 10:08:27 +02002748 .size = int128_2_64(),
Avi Kivity5312bd82012-02-12 18:32:55 +02002749 };
2750
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002751 return phys_section_add(map, &section);
Avi Kivity5312bd82012-02-12 18:32:55 +02002752}
2753
Peter Maydell8af36742017-12-13 17:52:28 +00002754static void readonly_mem_write(void *opaque, hwaddr addr,
2755 uint64_t val, unsigned size)
2756{
2757 /* Ignore any write to ROM. */
2758}
2759
2760static bool readonly_mem_accepts(void *opaque, hwaddr addr,
2761 unsigned size, bool is_write)
2762{
2763 return is_write;
2764}
2765
2766/* This will only be used for writes, because reads are special cased
2767 * to directly access the underlying host ram.
2768 */
2769static const MemoryRegionOps readonly_mem_ops = {
2770 .write = readonly_mem_write,
2771 .valid.accepts = readonly_mem_accepts,
2772 .endianness = DEVICE_NATIVE_ENDIAN,
2773 .valid = {
2774 .min_access_size = 1,
2775 .max_access_size = 8,
2776 .unaligned = false,
2777 },
2778 .impl = {
2779 .min_access_size = 1,
2780 .max_access_size = 8,
2781 .unaligned = false,
2782 },
2783};
2784
Peter Maydella54c87b2016-01-21 14:15:05 +00002785MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index, MemTxAttrs attrs)
Avi Kivityaa102232012-03-08 17:06:55 +02002786{
Peter Maydella54c87b2016-01-21 14:15:05 +00002787 int asidx = cpu_asidx_from_attrs(cpu, attrs);
2788 CPUAddressSpace *cpuas = &cpu->cpu_ases[asidx];
Peter Maydell32857f42015-10-01 15:29:50 +01002789 AddressSpaceDispatch *d = atomic_rcu_read(&cpuas->memory_dispatch);
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002790 MemoryRegionSection *sections = d->map.sections;
Paolo Bonzini9d82b5a2013-08-16 08:26:30 +02002791
2792 return sections[index & ~TARGET_PAGE_MASK].mr;
Avi Kivityaa102232012-03-08 17:06:55 +02002793}
2794
Avi Kivitye9179ce2009-06-14 11:38:52 +03002795static void io_mem_init(void)
2796{
Peter Maydell8af36742017-12-13 17:52:28 +00002797 memory_region_init_io(&io_mem_rom, NULL, &readonly_mem_ops,
2798 NULL, NULL, UINT64_MAX);
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002799 memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002800 NULL, UINT64_MAX);
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002801
2802 /* io_mem_notdirty calls tb_invalidate_phys_page_fast,
2803 * which can be called without the iothread mutex.
2804 */
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002805 memory_region_init_io(&io_mem_notdirty, NULL, &notdirty_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002806 NULL, UINT64_MAX);
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002807 memory_region_clear_global_locking(&io_mem_notdirty);
2808
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002809 memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002810 NULL, UINT64_MAX);
Avi Kivitye9179ce2009-06-14 11:38:52 +03002811}
2812
Alexey Kardashevskiy8629d3f2017-09-21 18:51:00 +10002813AddressSpaceDispatch *address_space_dispatch_new(FlatView *fv)
Avi Kivityac1970f2012-10-03 16:22:53 +02002814{
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002815 AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1);
2816 uint16_t n;
2817
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002818 n = dummy_section(&d->map, fv, &io_mem_unassigned);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002819 assert(n == PHYS_SECTION_UNASSIGNED);
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002820 n = dummy_section(&d->map, fv, &io_mem_notdirty);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002821 assert(n == PHYS_SECTION_NOTDIRTY);
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002822 n = dummy_section(&d->map, fv, &io_mem_rom);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002823 assert(n == PHYS_SECTION_ROM);
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002824 n = dummy_section(&d->map, fv, &io_mem_watch);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002825 assert(n == PHYS_SECTION_WATCH);
Paolo Bonzini00752702013-05-29 12:13:54 +02002826
Michael S. Tsirkin9736e552013-11-11 14:42:43 +02002827 d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 };
Alexey Kardashevskiy66a6df12017-09-21 18:50:56 +10002828
2829 return d;
Paolo Bonzini00752702013-05-29 12:13:54 +02002830}
2831
Alexey Kardashevskiy66a6df12017-09-21 18:50:56 +10002832void address_space_dispatch_free(AddressSpaceDispatch *d)
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002833{
2834 phys_sections_free(&d->map);
2835 g_free(d);
2836}
2837
Avi Kivity1d711482012-10-02 18:54:45 +02002838static void tcg_commit(MemoryListener *listener)
Avi Kivity50c1e142012-02-08 21:36:02 +02002839{
Peter Maydell32857f42015-10-01 15:29:50 +01002840 CPUAddressSpace *cpuas;
2841 AddressSpaceDispatch *d;
Avi Kivity117712c2012-02-12 21:23:17 +02002842
2843 /* since each CPU stores ram addresses in its TLB cache, we must
2844 reset the modified entries */
Peter Maydell32857f42015-10-01 15:29:50 +01002845 cpuas = container_of(listener, CPUAddressSpace, tcg_as_listener);
2846 cpu_reloading_memory_map();
2847 /* The CPU and TLB are protected by the iothread lock.
2848 * We reload the dispatch pointer now because cpu_reloading_memory_map()
2849 * may have split the RCU critical section.
2850 */
Alexey Kardashevskiy66a6df12017-09-21 18:50:56 +10002851 d = address_space_to_dispatch(cpuas->as);
Alex Bennéef35e44e2016-10-21 16:34:18 +01002852 atomic_rcu_set(&cpuas->memory_dispatch, d);
Alex Bennéed10eb082016-11-14 14:17:28 +00002853 tlb_flush(cpuas->cpu);
Avi Kivity50c1e142012-02-08 21:36:02 +02002854}
2855
Avi Kivity62152b82011-07-26 14:26:14 +03002856static void memory_map_init(void)
2857{
Anthony Liguori7267c092011-08-20 22:09:37 -05002858 system_memory = g_malloc(sizeof(*system_memory));
Paolo Bonzini03f49952013-11-07 17:14:36 +01002859
Paolo Bonzini57271d62013-11-07 17:14:37 +01002860 memory_region_init(system_memory, NULL, "system", UINT64_MAX);
Alexey Kardashevskiy7dca8042013-04-29 16:25:51 +00002861 address_space_init(&address_space_memory, system_memory, "memory");
Avi Kivity309cb472011-08-08 16:09:03 +03002862
Anthony Liguori7267c092011-08-20 22:09:37 -05002863 system_io = g_malloc(sizeof(*system_io));
Jan Kiszka3bb28b72013-09-02 18:43:30 +02002864 memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io",
2865 65536);
Alexey Kardashevskiy7dca8042013-04-29 16:25:51 +00002866 address_space_init(&address_space_io, system_io, "I/O");
Avi Kivity62152b82011-07-26 14:26:14 +03002867}
2868
2869MemoryRegion *get_system_memory(void)
2870{
2871 return system_memory;
2872}
2873
Avi Kivity309cb472011-08-08 16:09:03 +03002874MemoryRegion *get_system_io(void)
2875{
2876 return system_io;
2877}
2878
pbrooke2eef172008-06-08 01:09:01 +00002879#endif /* !defined(CONFIG_USER_ONLY) */
2880
bellard13eb76e2004-01-24 15:23:36 +00002881/* physical memory access (slow version, mainly for debug) */
2882#if defined(CONFIG_USER_ONLY)
Andreas Färberf17ec442013-06-29 19:40:58 +02002883int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
Paul Brooka68fe892010-03-01 00:08:59 +00002884 uint8_t *buf, int len, int is_write)
bellard13eb76e2004-01-24 15:23:36 +00002885{
2886 int l, flags;
2887 target_ulong page;
pbrook53a59602006-03-25 19:31:22 +00002888 void * p;
bellard13eb76e2004-01-24 15:23:36 +00002889
2890 while (len > 0) {
2891 page = addr & TARGET_PAGE_MASK;
2892 l = (page + TARGET_PAGE_SIZE) - addr;
2893 if (l > len)
2894 l = len;
2895 flags = page_get_flags(page);
2896 if (!(flags & PAGE_VALID))
Paul Brooka68fe892010-03-01 00:08:59 +00002897 return -1;
bellard13eb76e2004-01-24 15:23:36 +00002898 if (is_write) {
2899 if (!(flags & PAGE_WRITE))
Paul Brooka68fe892010-03-01 00:08:59 +00002900 return -1;
bellard579a97f2007-11-11 14:26:47 +00002901 /* XXX: this code should not depend on lock_user */
aurel3272fb7da2008-04-27 23:53:45 +00002902 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
Paul Brooka68fe892010-03-01 00:08:59 +00002903 return -1;
aurel3272fb7da2008-04-27 23:53:45 +00002904 memcpy(p, buf, l);
2905 unlock_user(p, addr, l);
bellard13eb76e2004-01-24 15:23:36 +00002906 } else {
2907 if (!(flags & PAGE_READ))
Paul Brooka68fe892010-03-01 00:08:59 +00002908 return -1;
bellard579a97f2007-11-11 14:26:47 +00002909 /* XXX: this code should not depend on lock_user */
aurel3272fb7da2008-04-27 23:53:45 +00002910 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
Paul Brooka68fe892010-03-01 00:08:59 +00002911 return -1;
aurel3272fb7da2008-04-27 23:53:45 +00002912 memcpy(buf, p, l);
aurel325b257572008-04-28 08:54:59 +00002913 unlock_user(p, addr, 0);
bellard13eb76e2004-01-24 15:23:36 +00002914 }
2915 len -= l;
2916 buf += l;
2917 addr += l;
2918 }
Paul Brooka68fe892010-03-01 00:08:59 +00002919 return 0;
bellard13eb76e2004-01-24 15:23:36 +00002920}
bellard8df1cd02005-01-28 22:37:22 +00002921
bellard13eb76e2004-01-24 15:23:36 +00002922#else
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002923
Paolo Bonzini845b6212015-03-23 11:45:53 +01002924static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr,
Avi Kivitya8170e52012-10-23 12:30:10 +02002925 hwaddr length)
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002926{
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002927 uint8_t dirty_log_mask = memory_region_get_dirty_log_mask(mr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002928 addr += memory_region_get_ram_addr(mr);
2929
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002930 /* No early return if dirty_log_mask is or becomes 0, because
2931 * cpu_physical_memory_set_dirty_range will still call
2932 * xen_modified_memory.
2933 */
2934 if (dirty_log_mask) {
2935 dirty_log_mask =
2936 cpu_physical_memory_range_includes_clean(addr, length, dirty_log_mask);
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002937 }
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002938 if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) {
Paolo Bonzini5aa1ef72017-07-03 17:50:40 +02002939 assert(tcg_enabled());
Alex Bennéeba051fb2016-10-27 16:10:16 +01002940 tb_lock();
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002941 tb_invalidate_phys_range(addr, addr + length);
Alex Bennéeba051fb2016-10-27 16:10:16 +01002942 tb_unlock();
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002943 dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE);
2944 }
2945 cpu_physical_memory_set_dirty_range(addr, length, dirty_log_mask);
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002946}
2947
Richard Henderson23326162013-07-08 14:55:59 -07002948static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr)
Paolo Bonzini82f25632013-05-24 11:59:43 +02002949{
Paolo Bonzinie1622f42013-07-17 13:17:41 +02002950 unsigned access_size_max = mr->ops->valid.max_access_size;
Richard Henderson23326162013-07-08 14:55:59 -07002951
2952 /* Regions are assumed to support 1-4 byte accesses unless
2953 otherwise specified. */
Richard Henderson23326162013-07-08 14:55:59 -07002954 if (access_size_max == 0) {
2955 access_size_max = 4;
Paolo Bonzini82f25632013-05-24 11:59:43 +02002956 }
Richard Henderson23326162013-07-08 14:55:59 -07002957
2958 /* Bound the maximum access by the alignment of the address. */
2959 if (!mr->ops->impl.unaligned) {
2960 unsigned align_size_max = addr & -addr;
2961 if (align_size_max != 0 && align_size_max < access_size_max) {
2962 access_size_max = align_size_max;
2963 }
2964 }
2965
2966 /* Don't attempt accesses larger than the maximum. */
2967 if (l > access_size_max) {
2968 l = access_size_max;
2969 }
Peter Maydell6554f5c2015-07-24 13:33:10 +01002970 l = pow2floor(l);
Richard Henderson23326162013-07-08 14:55:59 -07002971
2972 return l;
Paolo Bonzini82f25632013-05-24 11:59:43 +02002973}
2974
Jan Kiszka4840f102015-06-18 18:47:22 +02002975static bool prepare_mmio_access(MemoryRegion *mr)
Paolo Bonzini125b3802015-06-18 18:47:21 +02002976{
Jan Kiszka4840f102015-06-18 18:47:22 +02002977 bool unlocked = !qemu_mutex_iothread_locked();
2978 bool release_lock = false;
2979
2980 if (unlocked && mr->global_locking) {
2981 qemu_mutex_lock_iothread();
2982 unlocked = false;
2983 release_lock = true;
Paolo Bonzini125b3802015-06-18 18:47:21 +02002984 }
Jan Kiszka4840f102015-06-18 18:47:22 +02002985 if (mr->flush_coalesced_mmio) {
2986 if (unlocked) {
2987 qemu_mutex_lock_iothread();
2988 }
2989 qemu_flush_coalesced_mmio_buffer();
2990 if (unlocked) {
2991 qemu_mutex_unlock_iothread();
2992 }
2993 }
2994
2995 return release_lock;
Paolo Bonzini125b3802015-06-18 18:47:21 +02002996}
2997
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002998/* Called within RCU critical section. */
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10002999static MemTxResult flatview_write_continue(FlatView *fv, hwaddr addr,
3000 MemTxAttrs attrs,
3001 const uint8_t *buf,
3002 int len, hwaddr addr1,
3003 hwaddr l, MemoryRegion *mr)
bellard13eb76e2004-01-24 15:23:36 +00003004{
bellard13eb76e2004-01-24 15:23:36 +00003005 uint8_t *ptr;
Paolo Bonzini791af8c2013-05-24 16:10:39 +02003006 uint64_t val;
Peter Maydell3b643492015-04-26 16:49:23 +01003007 MemTxResult result = MEMTX_OK;
Jan Kiszka4840f102015-06-18 18:47:22 +02003008 bool release_lock = false;
ths3b46e622007-09-17 08:09:54 +00003009
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003010 for (;;) {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003011 if (!memory_access_is_direct(mr, true)) {
3012 release_lock |= prepare_mmio_access(mr);
3013 l = memory_access_size(mr, l, addr1);
3014 /* XXX: could force current_cpu to NULL to avoid
3015 potential bugs */
3016 switch (l) {
3017 case 8:
3018 /* 64 bit write access */
3019 val = ldq_p(buf);
3020 result |= memory_region_dispatch_write(mr, addr1, val, 8,
3021 attrs);
3022 break;
3023 case 4:
3024 /* 32 bit write access */
Ladi Prosek6da67de2017-01-26 15:22:37 +01003025 val = (uint32_t)ldl_p(buf);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003026 result |= memory_region_dispatch_write(mr, addr1, val, 4,
3027 attrs);
3028 break;
3029 case 2:
3030 /* 16 bit write access */
3031 val = lduw_p(buf);
3032 result |= memory_region_dispatch_write(mr, addr1, val, 2,
3033 attrs);
3034 break;
3035 case 1:
3036 /* 8 bit write access */
3037 val = ldub_p(buf);
3038 result |= memory_region_dispatch_write(mr, addr1, val, 1,
3039 attrs);
3040 break;
3041 default:
3042 abort();
bellard13eb76e2004-01-24 15:23:36 +00003043 }
3044 } else {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003045 /* RAM case */
Anthony PERARDf5aa69b2017-07-26 17:53:26 +01003046 ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l, false);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003047 memcpy(ptr, buf, l);
3048 invalidate_and_set_dirty(mr, addr1, l);
bellard13eb76e2004-01-24 15:23:36 +00003049 }
Jan Kiszka4840f102015-06-18 18:47:22 +02003050
3051 if (release_lock) {
3052 qemu_mutex_unlock_iothread();
3053 release_lock = false;
3054 }
3055
bellard13eb76e2004-01-24 15:23:36 +00003056 len -= l;
3057 buf += l;
3058 addr += l;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003059
3060 if (!len) {
3061 break;
3062 }
3063
3064 l = len;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003065 mr = flatview_translate(fv, addr, &addr1, &l, true);
bellard13eb76e2004-01-24 15:23:36 +00003066 }
Paolo Bonzinifd8aaa72013-05-21 09:56:55 +02003067
Peter Maydell3b643492015-04-26 16:49:23 +01003068 return result;
bellard13eb76e2004-01-24 15:23:36 +00003069}
bellard8df1cd02005-01-28 22:37:22 +00003070
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003071static MemTxResult flatview_write(FlatView *fv, hwaddr addr, MemTxAttrs attrs,
3072 const uint8_t *buf, int len)
Avi Kivityac1970f2012-10-03 16:22:53 +02003073{
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003074 hwaddr l;
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003075 hwaddr addr1;
3076 MemoryRegion *mr;
3077 MemTxResult result = MEMTX_OK;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003078
3079 if (len > 0) {
3080 rcu_read_lock();
3081 l = len;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003082 mr = flatview_translate(fv, addr, &addr1, &l, true);
3083 result = flatview_write_continue(fv, addr, attrs, buf, len,
3084 addr1, l, mr);
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003085 rcu_read_unlock();
3086 }
3087
3088 return result;
3089}
3090
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003091MemTxResult address_space_write(AddressSpace *as, hwaddr addr,
3092 MemTxAttrs attrs,
3093 const uint8_t *buf, int len)
3094{
3095 return flatview_write(address_space_to_flatview(as), addr, attrs, buf, len);
3096}
3097
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003098/* Called within RCU critical section. */
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003099MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr,
3100 MemTxAttrs attrs, uint8_t *buf,
3101 int len, hwaddr addr1, hwaddr l,
3102 MemoryRegion *mr)
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003103{
3104 uint8_t *ptr;
3105 uint64_t val;
3106 MemTxResult result = MEMTX_OK;
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003107 bool release_lock = false;
3108
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003109 for (;;) {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003110 if (!memory_access_is_direct(mr, false)) {
3111 /* I/O case */
3112 release_lock |= prepare_mmio_access(mr);
3113 l = memory_access_size(mr, l, addr1);
3114 switch (l) {
3115 case 8:
3116 /* 64 bit read access */
3117 result |= memory_region_dispatch_read(mr, addr1, &val, 8,
3118 attrs);
3119 stq_p(buf, val);
3120 break;
3121 case 4:
3122 /* 32 bit read access */
3123 result |= memory_region_dispatch_read(mr, addr1, &val, 4,
3124 attrs);
3125 stl_p(buf, val);
3126 break;
3127 case 2:
3128 /* 16 bit read access */
3129 result |= memory_region_dispatch_read(mr, addr1, &val, 2,
3130 attrs);
3131 stw_p(buf, val);
3132 break;
3133 case 1:
3134 /* 8 bit read access */
3135 result |= memory_region_dispatch_read(mr, addr1, &val, 1,
3136 attrs);
3137 stb_p(buf, val);
3138 break;
3139 default:
3140 abort();
3141 }
3142 } else {
3143 /* RAM case */
Anthony PERARDf5aa69b2017-07-26 17:53:26 +01003144 ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l, false);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003145 memcpy(buf, ptr, l);
3146 }
3147
3148 if (release_lock) {
3149 qemu_mutex_unlock_iothread();
3150 release_lock = false;
3151 }
3152
3153 len -= l;
3154 buf += l;
3155 addr += l;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003156
3157 if (!len) {
3158 break;
3159 }
3160
3161 l = len;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003162 mr = flatview_translate(fv, addr, &addr1, &l, false);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003163 }
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003164
3165 return result;
3166}
3167
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003168MemTxResult flatview_read_full(FlatView *fv, hwaddr addr,
3169 MemTxAttrs attrs, uint8_t *buf, int len)
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003170{
3171 hwaddr l;
3172 hwaddr addr1;
3173 MemoryRegion *mr;
3174 MemTxResult result = MEMTX_OK;
3175
3176 if (len > 0) {
3177 rcu_read_lock();
3178 l = len;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003179 mr = flatview_translate(fv, addr, &addr1, &l, false);
3180 result = flatview_read_continue(fv, addr, attrs, buf, len,
3181 addr1, l, mr);
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003182 rcu_read_unlock();
3183 }
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003184
3185 return result;
Avi Kivityac1970f2012-10-03 16:22:53 +02003186}
3187
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003188static MemTxResult flatview_rw(FlatView *fv, hwaddr addr, MemTxAttrs attrs,
3189 uint8_t *buf, int len, bool is_write)
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003190{
3191 if (is_write) {
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003192 return flatview_write(fv, addr, attrs, (uint8_t *)buf, len);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003193 } else {
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003194 return flatview_read(fv, addr, attrs, (uint8_t *)buf, len);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003195 }
3196}
Avi Kivityac1970f2012-10-03 16:22:53 +02003197
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003198MemTxResult address_space_rw(AddressSpace *as, hwaddr addr,
3199 MemTxAttrs attrs, uint8_t *buf,
3200 int len, bool is_write)
3201{
3202 return flatview_rw(address_space_to_flatview(as),
3203 addr, attrs, buf, len, is_write);
3204}
3205
Avi Kivitya8170e52012-10-23 12:30:10 +02003206void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf,
Avi Kivityac1970f2012-10-03 16:22:53 +02003207 int len, int is_write)
3208{
Peter Maydell5c9eb022015-04-26 16:49:24 +01003209 address_space_rw(&address_space_memory, addr, MEMTXATTRS_UNSPECIFIED,
3210 buf, len, is_write);
Avi Kivityac1970f2012-10-03 16:22:53 +02003211}
3212
Alexander Graf582b55a2013-12-11 14:17:44 +01003213enum write_rom_type {
3214 WRITE_DATA,
3215 FLUSH_CACHE,
3216};
3217
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003218static inline void cpu_physical_memory_write_rom_internal(AddressSpace *as,
Alexander Graf582b55a2013-12-11 14:17:44 +01003219 hwaddr addr, const uint8_t *buf, int len, enum write_rom_type type)
bellardd0ecd2a2006-04-23 17:14:48 +00003220{
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003221 hwaddr l;
bellardd0ecd2a2006-04-23 17:14:48 +00003222 uint8_t *ptr;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003223 hwaddr addr1;
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003224 MemoryRegion *mr;
ths3b46e622007-09-17 08:09:54 +00003225
Paolo Bonzini41063e12015-03-18 14:21:43 +01003226 rcu_read_lock();
bellardd0ecd2a2006-04-23 17:14:48 +00003227 while (len > 0) {
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003228 l = len;
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003229 mr = address_space_translate(as, addr, &addr1, &l, true);
ths3b46e622007-09-17 08:09:54 +00003230
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003231 if (!(memory_region_is_ram(mr) ||
3232 memory_region_is_romd(mr))) {
Paolo Bonzinib242e0e2015-07-04 00:24:51 +02003233 l = memory_access_size(mr, l, addr1);
bellardd0ecd2a2006-04-23 17:14:48 +00003234 } else {
bellardd0ecd2a2006-04-23 17:14:48 +00003235 /* ROM/RAM case */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01003236 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
Alexander Graf582b55a2013-12-11 14:17:44 +01003237 switch (type) {
3238 case WRITE_DATA:
3239 memcpy(ptr, buf, l);
Paolo Bonzini845b6212015-03-23 11:45:53 +01003240 invalidate_and_set_dirty(mr, addr1, l);
Alexander Graf582b55a2013-12-11 14:17:44 +01003241 break;
3242 case FLUSH_CACHE:
3243 flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l);
3244 break;
3245 }
bellardd0ecd2a2006-04-23 17:14:48 +00003246 }
3247 len -= l;
3248 buf += l;
3249 addr += l;
3250 }
Paolo Bonzini41063e12015-03-18 14:21:43 +01003251 rcu_read_unlock();
bellardd0ecd2a2006-04-23 17:14:48 +00003252}
3253
Alexander Graf582b55a2013-12-11 14:17:44 +01003254/* used for ROM loading : can write in RAM and ROM */
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003255void cpu_physical_memory_write_rom(AddressSpace *as, hwaddr addr,
Alexander Graf582b55a2013-12-11 14:17:44 +01003256 const uint8_t *buf, int len)
3257{
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003258 cpu_physical_memory_write_rom_internal(as, addr, buf, len, WRITE_DATA);
Alexander Graf582b55a2013-12-11 14:17:44 +01003259}
3260
3261void cpu_flush_icache_range(hwaddr start, int len)
3262{
3263 /*
3264 * This function should do the same thing as an icache flush that was
3265 * triggered from within the guest. For TCG we are always cache coherent,
3266 * so there is no need to flush anything. For KVM / Xen we need to flush
3267 * the host's instruction cache at least.
3268 */
3269 if (tcg_enabled()) {
3270 return;
3271 }
3272
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003273 cpu_physical_memory_write_rom_internal(&address_space_memory,
3274 start, NULL, len, FLUSH_CACHE);
Alexander Graf582b55a2013-12-11 14:17:44 +01003275}
3276
aliguori6d16c2f2009-01-22 16:59:11 +00003277typedef struct {
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003278 MemoryRegion *mr;
aliguori6d16c2f2009-01-22 16:59:11 +00003279 void *buffer;
Avi Kivitya8170e52012-10-23 12:30:10 +02003280 hwaddr addr;
3281 hwaddr len;
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003282 bool in_use;
aliguori6d16c2f2009-01-22 16:59:11 +00003283} BounceBuffer;
3284
3285static BounceBuffer bounce;
3286
aliguoriba223c22009-01-22 16:59:16 +00003287typedef struct MapClient {
Fam Zhenge95205e2015-03-16 17:03:37 +08003288 QEMUBH *bh;
Blue Swirl72cf2d42009-09-12 07:36:22 +00003289 QLIST_ENTRY(MapClient) link;
aliguoriba223c22009-01-22 16:59:16 +00003290} MapClient;
3291
Fam Zheng38e047b2015-03-16 17:03:35 +08003292QemuMutex map_client_list_lock;
Blue Swirl72cf2d42009-09-12 07:36:22 +00003293static QLIST_HEAD(map_client_list, MapClient) map_client_list
3294 = QLIST_HEAD_INITIALIZER(map_client_list);
aliguoriba223c22009-01-22 16:59:16 +00003295
Fam Zhenge95205e2015-03-16 17:03:37 +08003296static void cpu_unregister_map_client_do(MapClient *client)
aliguoriba223c22009-01-22 16:59:16 +00003297{
Blue Swirl72cf2d42009-09-12 07:36:22 +00003298 QLIST_REMOVE(client, link);
Anthony Liguori7267c092011-08-20 22:09:37 -05003299 g_free(client);
aliguoriba223c22009-01-22 16:59:16 +00003300}
3301
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003302static void cpu_notify_map_clients_locked(void)
aliguoriba223c22009-01-22 16:59:16 +00003303{
3304 MapClient *client;
3305
Blue Swirl72cf2d42009-09-12 07:36:22 +00003306 while (!QLIST_EMPTY(&map_client_list)) {
3307 client = QLIST_FIRST(&map_client_list);
Fam Zhenge95205e2015-03-16 17:03:37 +08003308 qemu_bh_schedule(client->bh);
3309 cpu_unregister_map_client_do(client);
aliguoriba223c22009-01-22 16:59:16 +00003310 }
3311}
3312
Fam Zhenge95205e2015-03-16 17:03:37 +08003313void cpu_register_map_client(QEMUBH *bh)
bellardd0ecd2a2006-04-23 17:14:48 +00003314{
3315 MapClient *client = g_malloc(sizeof(*client));
3316
Fam Zheng38e047b2015-03-16 17:03:35 +08003317 qemu_mutex_lock(&map_client_list_lock);
Fam Zhenge95205e2015-03-16 17:03:37 +08003318 client->bh = bh;
bellardd0ecd2a2006-04-23 17:14:48 +00003319 QLIST_INSERT_HEAD(&map_client_list, client, link);
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003320 if (!atomic_read(&bounce.in_use)) {
3321 cpu_notify_map_clients_locked();
3322 }
Fam Zheng38e047b2015-03-16 17:03:35 +08003323 qemu_mutex_unlock(&map_client_list_lock);
bellardd0ecd2a2006-04-23 17:14:48 +00003324}
3325
Fam Zheng38e047b2015-03-16 17:03:35 +08003326void cpu_exec_init_all(void)
3327{
3328 qemu_mutex_init(&ram_list.mutex);
Peter Maydell20bccb82016-10-24 16:26:49 +01003329 /* The data structures we set up here depend on knowing the page size,
3330 * so no more changes can be made after this point.
3331 * In an ideal world, nothing we did before we had finished the
3332 * machine setup would care about the target page size, and we could
3333 * do this much later, rather than requiring board models to state
3334 * up front what their requirements are.
3335 */
3336 finalize_target_page_bits();
Fam Zheng38e047b2015-03-16 17:03:35 +08003337 io_mem_init();
Paolo Bonzini680a4782015-11-02 09:23:52 +01003338 memory_map_init();
Fam Zheng38e047b2015-03-16 17:03:35 +08003339 qemu_mutex_init(&map_client_list_lock);
3340}
3341
Fam Zhenge95205e2015-03-16 17:03:37 +08003342void cpu_unregister_map_client(QEMUBH *bh)
bellardd0ecd2a2006-04-23 17:14:48 +00003343{
Fam Zhenge95205e2015-03-16 17:03:37 +08003344 MapClient *client;
bellardd0ecd2a2006-04-23 17:14:48 +00003345
Fam Zhenge95205e2015-03-16 17:03:37 +08003346 qemu_mutex_lock(&map_client_list_lock);
3347 QLIST_FOREACH(client, &map_client_list, link) {
3348 if (client->bh == bh) {
3349 cpu_unregister_map_client_do(client);
3350 break;
3351 }
3352 }
3353 qemu_mutex_unlock(&map_client_list_lock);
bellardd0ecd2a2006-04-23 17:14:48 +00003354}
3355
3356static void cpu_notify_map_clients(void)
3357{
Fam Zheng38e047b2015-03-16 17:03:35 +08003358 qemu_mutex_lock(&map_client_list_lock);
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003359 cpu_notify_map_clients_locked();
Fam Zheng38e047b2015-03-16 17:03:35 +08003360 qemu_mutex_unlock(&map_client_list_lock);
aliguori6d16c2f2009-01-22 16:59:11 +00003361}
3362
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003363static bool flatview_access_valid(FlatView *fv, hwaddr addr, int len,
3364 bool is_write)
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003365{
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003366 MemoryRegion *mr;
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003367 hwaddr l, xlat;
3368
Paolo Bonzini41063e12015-03-18 14:21:43 +01003369 rcu_read_lock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003370 while (len > 0) {
3371 l = len;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003372 mr = flatview_translate(fv, addr, &xlat, &l, is_write);
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003373 if (!memory_access_is_direct(mr, is_write)) {
3374 l = memory_access_size(mr, l, addr);
3375 if (!memory_region_access_valid(mr, xlat, l, is_write)) {
Roman Kapl5ad4a2b2017-01-09 12:09:21 +01003376 rcu_read_unlock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003377 return false;
3378 }
3379 }
3380
3381 len -= l;
3382 addr += l;
3383 }
Paolo Bonzini41063e12015-03-18 14:21:43 +01003384 rcu_read_unlock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003385 return true;
3386}
3387
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003388bool address_space_access_valid(AddressSpace *as, hwaddr addr,
3389 int len, bool is_write)
3390{
3391 return flatview_access_valid(address_space_to_flatview(as),
3392 addr, len, is_write);
3393}
3394
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003395static hwaddr
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003396flatview_extend_translation(FlatView *fv, hwaddr addr,
3397 hwaddr target_len,
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003398 MemoryRegion *mr, hwaddr base, hwaddr len,
3399 bool is_write)
3400{
3401 hwaddr done = 0;
3402 hwaddr xlat;
3403 MemoryRegion *this_mr;
3404
3405 for (;;) {
3406 target_len -= len;
3407 addr += len;
3408 done += len;
3409 if (target_len == 0) {
3410 return done;
3411 }
3412
3413 len = target_len;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003414 this_mr = flatview_translate(fv, addr, &xlat,
3415 &len, is_write);
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003416 if (this_mr != mr || xlat != base + done) {
3417 return done;
3418 }
3419 }
3420}
3421
aliguori6d16c2f2009-01-22 16:59:11 +00003422/* Map a physical memory region into a host virtual address.
3423 * May map a subset of the requested range, given by and returned in *plen.
3424 * May return NULL if resources needed to perform the mapping are exhausted.
3425 * Use only for reads OR writes - not for read-modify-write operations.
aliguoriba223c22009-01-22 16:59:16 +00003426 * Use cpu_register_map_client() to know when retrying the map operation is
3427 * likely to succeed.
aliguori6d16c2f2009-01-22 16:59:11 +00003428 */
Avi Kivityac1970f2012-10-03 16:22:53 +02003429void *address_space_map(AddressSpace *as,
Avi Kivitya8170e52012-10-23 12:30:10 +02003430 hwaddr addr,
3431 hwaddr *plen,
Avi Kivityac1970f2012-10-03 16:22:53 +02003432 bool is_write)
aliguori6d16c2f2009-01-22 16:59:11 +00003433{
Avi Kivitya8170e52012-10-23 12:30:10 +02003434 hwaddr len = *plen;
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003435 hwaddr l, xlat;
3436 MemoryRegion *mr;
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01003437 void *ptr;
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003438 FlatView *fv = address_space_to_flatview(as);
aliguori6d16c2f2009-01-22 16:59:11 +00003439
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003440 if (len == 0) {
3441 return NULL;
3442 }
aliguori6d16c2f2009-01-22 16:59:11 +00003443
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003444 l = len;
Paolo Bonzini41063e12015-03-18 14:21:43 +01003445 rcu_read_lock();
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003446 mr = flatview_translate(fv, addr, &xlat, &l, is_write);
Paolo Bonzini41063e12015-03-18 14:21:43 +01003447
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003448 if (!memory_access_is_direct(mr, is_write)) {
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003449 if (atomic_xchg(&bounce.in_use, true)) {
Paolo Bonzini41063e12015-03-18 14:21:43 +01003450 rcu_read_unlock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003451 return NULL;
aliguori6d16c2f2009-01-22 16:59:11 +00003452 }
Kevin Wolfe85d9db2013-07-22 14:30:23 +02003453 /* Avoid unbounded allocations */
3454 l = MIN(l, TARGET_PAGE_SIZE);
3455 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l);
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003456 bounce.addr = addr;
3457 bounce.len = l;
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003458
3459 memory_region_ref(mr);
3460 bounce.mr = mr;
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003461 if (!is_write) {
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003462 flatview_read(fv, addr, MEMTXATTRS_UNSPECIFIED,
Peter Maydell5c9eb022015-04-26 16:49:24 +01003463 bounce.buffer, l);
Stefano Stabellini8ab934f2011-06-27 18:26:06 +01003464 }
aliguori6d16c2f2009-01-22 16:59:11 +00003465
Paolo Bonzini41063e12015-03-18 14:21:43 +01003466 rcu_read_unlock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003467 *plen = l;
3468 return bounce.buffer;
3469 }
3470
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003471
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003472 memory_region_ref(mr);
Alexey Kardashevskiy16620682017-09-21 18:50:58 +10003473 *plen = flatview_extend_translation(fv, addr, len, mr, xlat,
3474 l, is_write);
Anthony PERARDf5aa69b2017-07-26 17:53:26 +01003475 ptr = qemu_ram_ptr_length(mr->ram_block, xlat, plen, true);
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01003476 rcu_read_unlock();
3477
3478 return ptr;
aliguori6d16c2f2009-01-22 16:59:11 +00003479}
3480
Avi Kivityac1970f2012-10-03 16:22:53 +02003481/* Unmaps a memory region previously mapped by address_space_map().
aliguori6d16c2f2009-01-22 16:59:11 +00003482 * Will also mark the memory as dirty if is_write == 1. access_len gives
3483 * the amount of memory that was actually read or written by the caller.
3484 */
Avi Kivitya8170e52012-10-23 12:30:10 +02003485void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
3486 int is_write, hwaddr access_len)
aliguori6d16c2f2009-01-22 16:59:11 +00003487{
3488 if (buffer != bounce.buffer) {
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003489 MemoryRegion *mr;
3490 ram_addr_t addr1;
3491
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01003492 mr = memory_region_from_host(buffer, &addr1);
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003493 assert(mr != NULL);
aliguori6d16c2f2009-01-22 16:59:11 +00003494 if (is_write) {
Paolo Bonzini845b6212015-03-23 11:45:53 +01003495 invalidate_and_set_dirty(mr, addr1, access_len);
aliguori6d16c2f2009-01-22 16:59:11 +00003496 }
Jan Kiszka868bb332011-06-21 22:59:09 +02003497 if (xen_enabled()) {
Jan Kiszkae41d7c62011-06-21 22:59:08 +02003498 xen_invalidate_map_cache_entry(buffer);
Anthony PERARD050a0dd2010-09-16 13:57:49 +01003499 }
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003500 memory_region_unref(mr);
aliguori6d16c2f2009-01-22 16:59:11 +00003501 return;
3502 }
3503 if (is_write) {
Peter Maydell5c9eb022015-04-26 16:49:24 +01003504 address_space_write(as, bounce.addr, MEMTXATTRS_UNSPECIFIED,
3505 bounce.buffer, access_len);
aliguori6d16c2f2009-01-22 16:59:11 +00003506 }
Herve Poussineauf8a83242010-01-24 21:23:56 +00003507 qemu_vfree(bounce.buffer);
aliguori6d16c2f2009-01-22 16:59:11 +00003508 bounce.buffer = NULL;
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003509 memory_region_unref(bounce.mr);
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003510 atomic_mb_set(&bounce.in_use, false);
aliguoriba223c22009-01-22 16:59:16 +00003511 cpu_notify_map_clients();
aliguori6d16c2f2009-01-22 16:59:11 +00003512}
bellardd0ecd2a2006-04-23 17:14:48 +00003513
Avi Kivitya8170e52012-10-23 12:30:10 +02003514void *cpu_physical_memory_map(hwaddr addr,
3515 hwaddr *plen,
Avi Kivityac1970f2012-10-03 16:22:53 +02003516 int is_write)
3517{
3518 return address_space_map(&address_space_memory, addr, plen, is_write);
3519}
3520
Avi Kivitya8170e52012-10-23 12:30:10 +02003521void cpu_physical_memory_unmap(void *buffer, hwaddr len,
3522 int is_write, hwaddr access_len)
Avi Kivityac1970f2012-10-03 16:22:53 +02003523{
3524 return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len);
3525}
3526
Paolo Bonzini0ce265f2016-11-22 11:34:02 +01003527#define ARG1_DECL AddressSpace *as
3528#define ARG1 as
3529#define SUFFIX
3530#define TRANSLATE(...) address_space_translate(as, __VA_ARGS__)
3531#define IS_DIRECT(mr, is_write) memory_access_is_direct(mr, is_write)
3532#define MAP_RAM(mr, ofs) qemu_map_ram_ptr((mr)->ram_block, ofs)
3533#define INVALIDATE(mr, ofs, len) invalidate_and_set_dirty(mr, ofs, len)
3534#define RCU_READ_LOCK(...) rcu_read_lock()
3535#define RCU_READ_UNLOCK(...) rcu_read_unlock()
3536#include "memory_ldst.inc.c"
Alexander Graf1e78bcc2011-07-06 09:09:23 +02003537
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003538int64_t address_space_cache_init(MemoryRegionCache *cache,
3539 AddressSpace *as,
3540 hwaddr addr,
3541 hwaddr len,
3542 bool is_write)
3543{
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003544 cache->len = len;
3545 cache->as = as;
3546 cache->xlat = addr;
3547 return len;
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003548}
3549
3550void address_space_cache_invalidate(MemoryRegionCache *cache,
3551 hwaddr addr,
3552 hwaddr access_len)
3553{
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003554}
3555
3556void address_space_cache_destroy(MemoryRegionCache *cache)
3557{
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003558 cache->as = NULL;
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003559}
3560
3561#define ARG1_DECL MemoryRegionCache *cache
3562#define ARG1 cache
3563#define SUFFIX _cached
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003564#define TRANSLATE(addr, ...) \
3565 address_space_translate(cache->as, cache->xlat + (addr), __VA_ARGS__)
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003566#define IS_DIRECT(mr, is_write) true
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003567#define MAP_RAM(mr, ofs) qemu_map_ram_ptr((mr)->ram_block, ofs)
3568#define INVALIDATE(mr, ofs, len) invalidate_and_set_dirty(mr, ofs, len)
3569#define RCU_READ_LOCK() rcu_read_lock()
3570#define RCU_READ_UNLOCK() rcu_read_unlock()
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003571#include "memory_ldst.inc.c"
3572
aliguori5e2972f2009-03-28 17:51:36 +00003573/* virtual memory access for debug (includes writing to ROM) */
Andreas Färberf17ec442013-06-29 19:40:58 +02003574int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
bellardb448f2f2004-02-25 23:24:04 +00003575 uint8_t *buf, int len, int is_write)
bellard13eb76e2004-01-24 15:23:36 +00003576{
3577 int l;
Avi Kivitya8170e52012-10-23 12:30:10 +02003578 hwaddr phys_addr;
j_mayer9b3c35e2007-04-07 11:21:28 +00003579 target_ulong page;
bellard13eb76e2004-01-24 15:23:36 +00003580
Christian Borntraeger79ca7a12017-03-07 15:19:08 +01003581 cpu_synchronize_state(cpu);
bellard13eb76e2004-01-24 15:23:36 +00003582 while (len > 0) {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003583 int asidx;
3584 MemTxAttrs attrs;
3585
bellard13eb76e2004-01-24 15:23:36 +00003586 page = addr & TARGET_PAGE_MASK;
Peter Maydell5232e4c2016-01-21 14:15:06 +00003587 phys_addr = cpu_get_phys_page_attrs_debug(cpu, page, &attrs);
3588 asidx = cpu_asidx_from_attrs(cpu, attrs);
bellard13eb76e2004-01-24 15:23:36 +00003589 /* if no physical page mapped, return an error */
3590 if (phys_addr == -1)
3591 return -1;
3592 l = (page + TARGET_PAGE_SIZE) - addr;
3593 if (l > len)
3594 l = len;
aliguori5e2972f2009-03-28 17:51:36 +00003595 phys_addr += (addr & ~TARGET_PAGE_MASK);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003596 if (is_write) {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003597 cpu_physical_memory_write_rom(cpu->cpu_ases[asidx].as,
3598 phys_addr, buf, l);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003599 } else {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003600 address_space_rw(cpu->cpu_ases[asidx].as, phys_addr,
3601 MEMTXATTRS_UNSPECIFIED,
Peter Maydell5c9eb022015-04-26 16:49:24 +01003602 buf, l, 0);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003603 }
bellard13eb76e2004-01-24 15:23:36 +00003604 len -= l;
3605 buf += l;
3606 addr += l;
3607 }
3608 return 0;
3609}
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003610
3611/*
3612 * Allows code that needs to deal with migration bitmaps etc to still be built
3613 * target independent.
3614 */
Juan Quintela20afaed2017-03-21 09:09:14 +01003615size_t qemu_target_page_size(void)
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003616{
Juan Quintela20afaed2017-03-21 09:09:14 +01003617 return TARGET_PAGE_SIZE;
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003618}
3619
Juan Quintela46d702b2017-04-24 21:03:48 +02003620int qemu_target_page_bits(void)
3621{
3622 return TARGET_PAGE_BITS;
3623}
3624
3625int qemu_target_page_bits_min(void)
3626{
3627 return TARGET_PAGE_BITS_MIN;
3628}
Paul Brooka68fe892010-03-01 00:08:59 +00003629#endif
bellard13eb76e2004-01-24 15:23:36 +00003630
Blue Swirl8e4a4242013-01-06 18:30:17 +00003631/*
3632 * A helper function for the _utterly broken_ virtio device model to find out if
3633 * it's running on a big endian machine. Don't do this at home kids!
3634 */
Greg Kurz98ed8ec2014-06-24 19:26:29 +02003635bool target_words_bigendian(void);
3636bool target_words_bigendian(void)
Blue Swirl8e4a4242013-01-06 18:30:17 +00003637{
3638#if defined(TARGET_WORDS_BIGENDIAN)
3639 return true;
3640#else
3641 return false;
3642#endif
3643}
3644
Wen Congyang76f35532012-05-07 12:04:18 +08003645#ifndef CONFIG_USER_ONLY
Avi Kivitya8170e52012-10-23 12:30:10 +02003646bool cpu_physical_memory_is_io(hwaddr phys_addr)
Wen Congyang76f35532012-05-07 12:04:18 +08003647{
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003648 MemoryRegion*mr;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003649 hwaddr l = 1;
Paolo Bonzini41063e12015-03-18 14:21:43 +01003650 bool res;
Wen Congyang76f35532012-05-07 12:04:18 +08003651
Paolo Bonzini41063e12015-03-18 14:21:43 +01003652 rcu_read_lock();
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003653 mr = address_space_translate(&address_space_memory,
3654 phys_addr, &phys_addr, &l, false);
Wen Congyang76f35532012-05-07 12:04:18 +08003655
Paolo Bonzini41063e12015-03-18 14:21:43 +01003656 res = !(memory_region_is_ram(mr) || memory_region_is_romd(mr));
3657 rcu_read_unlock();
3658 return res;
Wen Congyang76f35532012-05-07 12:04:18 +08003659}
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003660
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003661int qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003662{
3663 RAMBlock *block;
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003664 int ret = 0;
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003665
Mike Day0dc3f442013-09-05 14:41:35 -04003666 rcu_read_lock();
Peter Xu99e15582017-05-12 12:17:39 +08003667 RAMBLOCK_FOREACH(block) {
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003668 ret = func(block->idstr, block->host, block->offset,
3669 block->used_length, opaque);
3670 if (ret) {
3671 break;
3672 }
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003673 }
Mike Day0dc3f442013-09-05 14:41:35 -04003674 rcu_read_unlock();
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003675 return ret;
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003676}
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003677
3678/*
3679 * Unmap pages of memory from start to start+length such that
3680 * they a) read as 0, b) Trigger whatever fault mechanism
3681 * the OS provides for postcopy.
3682 * The pages must be unmapped by the end of the function.
3683 * Returns: 0 on success, none-0 on failure
3684 *
3685 */
3686int ram_block_discard_range(RAMBlock *rb, uint64_t start, size_t length)
3687{
3688 int ret = -1;
3689
3690 uint8_t *host_startaddr = rb->host + start;
3691
3692 if ((uintptr_t)host_startaddr & (rb->page_size - 1)) {
3693 error_report("ram_block_discard_range: Unaligned start address: %p",
3694 host_startaddr);
3695 goto err;
3696 }
3697
3698 if ((start + length) <= rb->used_length) {
3699 uint8_t *host_endaddr = host_startaddr + length;
3700 if ((uintptr_t)host_endaddr & (rb->page_size - 1)) {
3701 error_report("ram_block_discard_range: Unaligned end address: %p",
3702 host_endaddr);
3703 goto err;
3704 }
3705
3706 errno = ENOTSUP; /* If we are missing MADVISE etc */
3707
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003708 if (rb->page_size == qemu_host_page_size) {
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003709#if defined(CONFIG_MADVISE)
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003710 /* Note: We need the madvise MADV_DONTNEED behaviour of definitely
3711 * freeing the page.
3712 */
3713 ret = madvise(host_startaddr, length, MADV_DONTNEED);
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003714#endif
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003715 } else {
3716 /* Huge page case - unfortunately it can't do DONTNEED, but
3717 * it can do the equivalent by FALLOC_FL_PUNCH_HOLE in the
3718 * huge page file.
3719 */
3720#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
3721 ret = fallocate(rb->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
3722 start, length);
3723#endif
3724 }
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003725 if (ret) {
3726 ret = -errno;
3727 error_report("ram_block_discard_range: Failed to discard range "
3728 "%s:%" PRIx64 " +%zx (%d)",
3729 rb->idstr, start, length, ret);
3730 }
3731 } else {
3732 error_report("ram_block_discard_range: Overrun block '%s' (%" PRIu64
3733 "/%zx/" RAM_ADDR_FMT")",
3734 rb->idstr, start, length, rb->used_length);
3735 }
3736
3737err:
3738 return ret;
3739}
3740
Peter Maydellec3f8c92013-06-27 20:53:38 +01003741#endif
Yang Zhonga0be0c52017-07-03 18:12:13 +08003742
3743void page_size_init(void)
3744{
3745 /* NOTE: we can always suppose that qemu_host_page_size >=
3746 TARGET_PAGE_SIZE */
Yang Zhonga0be0c52017-07-03 18:12:13 +08003747 if (qemu_host_page_size == 0) {
3748 qemu_host_page_size = qemu_real_host_page_size;
3749 }
3750 if (qemu_host_page_size < TARGET_PAGE_SIZE) {
3751 qemu_host_page_size = TARGET_PAGE_SIZE;
3752 }
3753 qemu_host_page_mask = -(intptr_t)qemu_host_page_size;
3754}
Alexey Kardashevskiy5e8fd942017-09-21 18:51:06 +10003755
3756#if !defined(CONFIG_USER_ONLY)
3757
3758static void mtree_print_phys_entries(fprintf_function mon, void *f,
3759 int start, int end, int skip, int ptr)
3760{
3761 if (start == end - 1) {
3762 mon(f, "\t%3d ", start);
3763 } else {
3764 mon(f, "\t%3d..%-3d ", start, end - 1);
3765 }
3766 mon(f, " skip=%d ", skip);
3767 if (ptr == PHYS_MAP_NODE_NIL) {
3768 mon(f, " ptr=NIL");
3769 } else if (!skip) {
3770 mon(f, " ptr=#%d", ptr);
3771 } else {
3772 mon(f, " ptr=[%d]", ptr);
3773 }
3774 mon(f, "\n");
3775}
3776
3777#define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \
3778 int128_sub((size), int128_one())) : 0)
3779
3780void mtree_print_dispatch(fprintf_function mon, void *f,
3781 AddressSpaceDispatch *d, MemoryRegion *root)
3782{
3783 int i;
3784
3785 mon(f, " Dispatch\n");
3786 mon(f, " Physical sections\n");
3787
3788 for (i = 0; i < d->map.sections_nb; ++i) {
3789 MemoryRegionSection *s = d->map.sections + i;
3790 const char *names[] = { " [unassigned]", " [not dirty]",
3791 " [ROM]", " [watch]" };
3792
3793 mon(f, " #%d @" TARGET_FMT_plx ".." TARGET_FMT_plx " %s%s%s%s%s",
3794 i,
3795 s->offset_within_address_space,
3796 s->offset_within_address_space + MR_SIZE(s->mr->size),
3797 s->mr->name ? s->mr->name : "(noname)",
3798 i < ARRAY_SIZE(names) ? names[i] : "",
3799 s->mr == root ? " [ROOT]" : "",
3800 s == d->mru_section ? " [MRU]" : "",
3801 s->mr->is_iommu ? " [iommu]" : "");
3802
3803 if (s->mr->alias) {
3804 mon(f, " alias=%s", s->mr->alias->name ?
3805 s->mr->alias->name : "noname");
3806 }
3807 mon(f, "\n");
3808 }
3809
3810 mon(f, " Nodes (%d bits per level, %d levels) ptr=[%d] skip=%d\n",
3811 P_L2_BITS, P_L2_LEVELS, d->phys_map.ptr, d->phys_map.skip);
3812 for (i = 0; i < d->map.nodes_nb; ++i) {
3813 int j, jprev;
3814 PhysPageEntry prev;
3815 Node *n = d->map.nodes + i;
3816
3817 mon(f, " [%d]\n", i);
3818
3819 for (j = 0, jprev = 0, prev = *n[0]; j < ARRAY_SIZE(*n); ++j) {
3820 PhysPageEntry *pe = *n + j;
3821
3822 if (pe->ptr == prev.ptr && pe->skip == prev.skip) {
3823 continue;
3824 }
3825
3826 mtree_print_phys_entries(mon, f, jprev, j, prev.skip, prev.ptr);
3827
3828 jprev = j;
3829 prev = *pe;
3830 }
3831
3832 if (jprev != ARRAY_SIZE(*n)) {
3833 mtree_print_phys_entries(mon, f, jprev, j, prev.skip, prev.ptr);
3834 }
3835 }
3836}
3837
3838#endif