blob: 85769e1464afdfba11fda2354bdfe8aa60d54d9f [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"
Stefan Weil777872e2014-02-23 18:02:08 +010021#ifndef _WIN32
bellardd5a8f072004-09-29 21:15:28 +000022#endif
bellard54936002003-05-13 00:25:15 +000023
Veronia Bahaaf348b6d2016-03-20 19:16:19 +020024#include "qemu/cutils.h"
bellard6180a182003-09-30 21:04:53 +000025#include "cpu.h"
Paolo Bonzini63c91552016-03-15 13:18:37 +010026#include "exec/exec-all.h"
bellardb67d9a52008-05-23 09:57:34 +000027#include "tcg.h"
Paolo Bonzini741da0d2014-06-27 08:40:04 +020028#include "hw/qdev-core.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
52#include <fcntl.h>
53#include <linux/falloc.h>
54#endif
55
pbrook53a59602006-03-25 19:31:22 +000056#endif
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +010057#include "exec/cpu-all.h"
Mike Day0dc3f442013-09-05 14:41:35 -040058#include "qemu/rcu_queue.h"
Jan Kiszka4840f102015-06-18 18:47:22 +020059#include "qemu/main-loop.h"
Blue Swirl5b6dd862012-12-02 16:04:43 +000060#include "translate-all.h"
Pavel Dovgalyuk76159362015-09-17 19:25:07 +030061#include "sysemu/replay.h"
Blue Swirl0cac1b62012-04-09 16:50:52 +000062
Paolo Bonzini022c62c2012-12-17 18:19:49 +010063#include "exec/memory-internal.h"
Juan Quintela220c3eb2013-10-14 17:13:59 +020064#include "exec/ram_addr.h"
Paolo Bonzini508127e2016-01-07 16:55:28 +030065#include "exec/log.h"
Avi Kivity67d95c12011-12-15 15:25:22 +020066
Bharata B Rao9dfeca72016-05-12 09:18:12 +053067#include "migration/vmstate.h"
68
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +020069#include "qemu/range.h"
Michael S. Tsirkin794e8f32015-09-24 14:41:17 +030070#ifndef _WIN32
71#include "qemu/mmap-alloc.h"
72#endif
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +020073
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
Peter Maydell20bccb82016-10-24 16:26:49 +0100118bool set_preferred_target_page_bits(int bits)
119{
120 /* The target page size is the lowest common denominator for all
121 * the CPUs in the system, so we can only make it smaller, never
122 * larger. And we can't make it smaller once we've committed to
123 * a particular size.
124 */
125#ifdef TARGET_PAGE_BITS_VARY
126 assert(bits >= TARGET_PAGE_BITS_MIN);
127 if (target_page_bits == 0 || target_page_bits > bits) {
128 if (target_page_bits_decided) {
129 return false;
130 }
131 target_page_bits = bits;
132 }
133#endif
134 return true;
135}
136
pbrooke2eef172008-06-08 01:09:01 +0000137#if !defined(CONFIG_USER_ONLY)
Avi Kivity4346ae32012-02-10 17:00:01 +0200138
Peter Maydell20bccb82016-10-24 16:26:49 +0100139static void finalize_target_page_bits(void)
140{
141#ifdef TARGET_PAGE_BITS_VARY
142 if (target_page_bits == 0) {
143 target_page_bits = TARGET_PAGE_BITS_MIN;
144 }
145 target_page_bits_decided = true;
146#endif
147}
148
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200149typedef struct PhysPageEntry PhysPageEntry;
150
151struct PhysPageEntry {
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200152 /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200153 uint32_t skip : 6;
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200154 /* index into phys_sections (!skip) or phys_map_nodes (skip) */
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200155 uint32_t ptr : 26;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200156};
157
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200158#define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6)
159
Paolo Bonzini03f49952013-11-07 17:14:36 +0100160/* Size of the L2 (and L3, etc) page tables. */
Paolo Bonzini57271d62013-11-07 17:14:37 +0100161#define ADDR_SPACE_BITS 64
Paolo Bonzini03f49952013-11-07 17:14:36 +0100162
Michael S. Tsirkin026736c2013-11-13 20:13:03 +0200163#define P_L2_BITS 9
Paolo Bonzini03f49952013-11-07 17:14:36 +0100164#define P_L2_SIZE (1 << P_L2_BITS)
165
166#define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1)
167
168typedef PhysPageEntry Node[P_L2_SIZE];
Paolo Bonzini0475d942013-05-29 12:28:21 +0200169
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200170typedef struct PhysPageMap {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100171 struct rcu_head rcu;
172
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200173 unsigned sections_nb;
174 unsigned sections_nb_alloc;
175 unsigned nodes_nb;
176 unsigned nodes_nb_alloc;
177 Node *nodes;
178 MemoryRegionSection *sections;
179} PhysPageMap;
180
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200181struct AddressSpaceDispatch {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100182 struct rcu_head rcu;
183
Fam Zheng729633c2016-03-01 14:18:24 +0800184 MemoryRegionSection *mru_section;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200185 /* This is a multi-level map on the physical address space.
186 * The bottom level has pointers to MemoryRegionSections.
187 */
188 PhysPageEntry phys_map;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200189 PhysPageMap map;
Jan Kiszkaacc9d802013-05-26 21:55:37 +0200190 AddressSpace *as;
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;
Jan Kiszkaacc9d802013-05-26 21:55:37 +0200196 AddressSpace *as;
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
356static void phys_page_compact_all(AddressSpaceDispatch *d, int nodes_nb)
357{
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
Michael S. Tsirkin97115a82013-11-13 20:08:19 +0200374static MemoryRegionSection *phys_page_find(PhysPageEntry lp, hwaddr addr,
Paolo Bonzini9affd6f2013-05-29 12:09:47 +0200375 Node *nodes, MemoryRegionSection *sections)
bellard92e873b2004-05-21 14:52:29 +0000376{
Avi Kivity31ab2b42012-02-13 16:44:19 +0200377 PhysPageEntry *p;
Michael S. Tsirkin97115a82013-11-13 20:08:19 +0200378 hwaddr index = addr >> TARGET_PAGE_BITS;
Avi Kivity31ab2b42012-02-13 16:44:19 +0200379 int i;
Avi Kivityf1f6e3b2011-11-20 17:52:22 +0200380
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200381 for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) {
Avi Kivityc19e8802012-02-13 20:25:31 +0200382 if (lp.ptr == PHYS_MAP_NODE_NIL) {
Paolo Bonzini9affd6f2013-05-29 12:09:47 +0200383 return &sections[PHYS_SECTION_UNASSIGNED];
Avi Kivity31ab2b42012-02-13 16:44:19 +0200384 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +0200385 p = nodes[lp.ptr];
Paolo Bonzini03f49952013-11-07 17:14:36 +0100386 lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)];
Avi Kivityf1f6e3b2011-11-20 17:52:22 +0200387 }
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200388
Fam Zheng29cb5332016-03-01 14:18:23 +0800389 if (section_covers_addr(&sections[lp.ptr], addr)) {
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200390 return &sections[lp.ptr];
391 } else {
392 return &sections[PHYS_SECTION_UNASSIGNED];
393 }
Avi Kivityf3705d52012-03-08 16:16:34 +0200394}
395
Blue Swirle5548612012-04-21 13:08:33 +0000396bool memory_region_is_unassigned(MemoryRegion *mr)
397{
Paolo Bonzini2a8e7492013-05-24 14:34:08 +0200398 return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device
Blue Swirle5548612012-04-21 13:08:33 +0000399 && mr != &io_mem_watch;
400}
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200401
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100402/* Called from RCU critical section */
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200403static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
Jan Kiszka90260c62013-05-26 21:46:51 +0200404 hwaddr addr,
405 bool resolve_subpage)
Jan Kiszka9f029602013-05-06 16:48:02 +0200406{
Fam Zheng729633c2016-03-01 14:18:24 +0800407 MemoryRegionSection *section = atomic_read(&d->mru_section);
Jan Kiszka90260c62013-05-26 21:46:51 +0200408 subpage_t *subpage;
Fam Zheng729633c2016-03-01 14:18:24 +0800409 bool update;
Jan Kiszka90260c62013-05-26 21:46:51 +0200410
Fam Zheng729633c2016-03-01 14:18:24 +0800411 if (section && section != &d->map.sections[PHYS_SECTION_UNASSIGNED] &&
412 section_covers_addr(section, addr)) {
413 update = false;
414 } else {
415 section = phys_page_find(d->phys_map, addr, d->map.nodes,
416 d->map.sections);
417 update = true;
418 }
Jan Kiszka90260c62013-05-26 21:46:51 +0200419 if (resolve_subpage && section->mr->subpage) {
420 subpage = container_of(section->mr, subpage_t, iomem);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200421 section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]];
Jan Kiszka90260c62013-05-26 21:46:51 +0200422 }
Fam Zheng729633c2016-03-01 14:18:24 +0800423 if (update) {
424 atomic_set(&d->mru_section, section);
425 }
Jan Kiszka90260c62013-05-26 21:46:51 +0200426 return section;
Jan Kiszka9f029602013-05-06 16:48:02 +0200427}
428
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100429/* Called from RCU critical section */
Jan Kiszka90260c62013-05-26 21:46:51 +0200430static MemoryRegionSection *
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200431address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
Jan Kiszka90260c62013-05-26 21:46:51 +0200432 hwaddr *plen, bool resolve_subpage)
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200433{
434 MemoryRegionSection *section;
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200435 MemoryRegion *mr;
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100436 Int128 diff;
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200437
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200438 section = address_space_lookup_region(d, addr, resolve_subpage);
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200439 /* Compute offset within MemoryRegionSection */
440 addr -= section->offset_within_address_space;
441
442 /* Compute offset within MemoryRegion */
443 *xlat = addr + section->offset_within_region;
444
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200445 mr = section->mr;
Paolo Bonzinib242e0e2015-07-04 00:24:51 +0200446
447 /* MMIO registers can be expected to perform full-width accesses based only
448 * on their address, without considering adjacent registers that could
449 * decode to completely different MemoryRegions. When such registers
450 * exist (e.g. I/O ports 0xcf8 and 0xcf9 on most PC chipsets), MMIO
451 * regions overlap wildly. For this reason we cannot clamp the accesses
452 * here.
453 *
454 * If the length is small (as is the case for address_space_ldl/stl),
455 * everything works fine. If the incoming length is large, however,
456 * the caller really has to do the clamping through memory_access_size.
457 */
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200458 if (memory_region_is_ram(mr)) {
Paolo Bonzinie4a511f2015-06-17 10:36:54 +0200459 diff = int128_sub(section->size, int128_make64(addr));
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200460 *plen = int128_get64(int128_min(diff, int128_make64(*plen)));
461 }
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200462 return section;
463}
Jan Kiszka90260c62013-05-26 21:46:51 +0200464
Paolo Bonzini41063e12015-03-18 14:21:43 +0100465/* Called from RCU critical section */
Jason Wang052c8fa2016-12-30 18:09:13 +0800466IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
467 bool is_write)
468{
469 IOMMUTLBEntry iotlb = {0};
470 MemoryRegionSection *section;
471 MemoryRegion *mr;
472
473 for (;;) {
474 AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
475 section = address_space_lookup_region(d, addr, false);
476 addr = addr - section->offset_within_address_space
477 + section->offset_within_region;
478 mr = section->mr;
479
480 if (!mr->iommu_ops) {
481 break;
482 }
483
484 iotlb = mr->iommu_ops->translate(mr, addr, is_write);
485 if (!(iotlb.perm & (1 << is_write))) {
486 iotlb.target_as = NULL;
487 break;
488 }
489
490 addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
491 | (addr & iotlb.addr_mask));
492 as = iotlb.target_as;
493 }
494
495 return iotlb;
496}
497
498/* Called from RCU critical section */
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +0200499MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
500 hwaddr *xlat, hwaddr *plen,
501 bool is_write)
Jan Kiszka90260c62013-05-26 21:46:51 +0200502{
Avi Kivity30951152012-10-30 13:47:46 +0200503 IOMMUTLBEntry iotlb;
504 MemoryRegionSection *section;
505 MemoryRegion *mr;
Avi Kivity30951152012-10-30 13:47:46 +0200506
507 for (;;) {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100508 AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
509 section = address_space_translate_internal(d, addr, &addr, plen, true);
Avi Kivity30951152012-10-30 13:47:46 +0200510 mr = section->mr;
511
512 if (!mr->iommu_ops) {
513 break;
514 }
515
Le Tan8d7b8cb2014-08-16 13:55:37 +0800516 iotlb = mr->iommu_ops->translate(mr, addr, is_write);
Avi Kivity30951152012-10-30 13:47:46 +0200517 addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
518 | (addr & iotlb.addr_mask));
Peter Crosthwaite23820db2015-03-16 22:35:54 -0700519 *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1);
Avi Kivity30951152012-10-30 13:47:46 +0200520 if (!(iotlb.perm & (1 << is_write))) {
521 mr = &io_mem_unassigned;
522 break;
523 }
524
525 as = iotlb.target_as;
526 }
527
Alexey Kardashevskiyfe680d02014-05-07 13:40:39 +0000528 if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100529 hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr;
Peter Crosthwaite23820db2015-03-16 22:35:54 -0700530 *plen = MIN(page, *plen);
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100531 }
532
Avi Kivity30951152012-10-30 13:47:46 +0200533 *xlat = addr;
534 return mr;
Jan Kiszka90260c62013-05-26 21:46:51 +0200535}
536
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100537/* Called from RCU critical section */
Jan Kiszka90260c62013-05-26 21:46:51 +0200538MemoryRegionSection *
Peter Maydelld7898cd2016-01-21 14:15:05 +0000539address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr,
Paolo Bonzini9d82b5a2013-08-16 08:26:30 +0200540 hwaddr *xlat, hwaddr *plen)
Jan Kiszka90260c62013-05-26 21:46:51 +0200541{
Avi Kivity30951152012-10-30 13:47:46 +0200542 MemoryRegionSection *section;
Alex Bennéef35e44e2016-10-21 16:34:18 +0100543 AddressSpaceDispatch *d = atomic_rcu_read(&cpu->cpu_ases[asidx].memory_dispatch);
Peter Maydelld7898cd2016-01-21 14:15:05 +0000544
545 section = address_space_translate_internal(d, addr, xlat, plen, false);
Avi Kivity30951152012-10-30 13:47:46 +0200546
547 assert(!section->mr->iommu_ops);
548 return section;
Jan Kiszka90260c62013-05-26 21:46:51 +0200549}
bellard9fa3e852004-01-04 18:06:42 +0000550#endif
bellardfd6ce8f2003-05-14 19:00:11 +0000551
Andreas Färberb170fce2013-01-20 20:23:22 +0100552#if !defined(CONFIG_USER_ONLY)
pbrook9656f322008-07-01 20:01:19 +0000553
Juan Quintelae59fb372009-09-29 22:48:21 +0200554static int cpu_common_post_load(void *opaque, int version_id)
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200555{
Andreas Färber259186a2013-01-17 18:51:17 +0100556 CPUState *cpu = opaque;
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200557
aurel323098dba2009-03-07 21:28:24 +0000558 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
559 version_id is increased. */
Andreas Färber259186a2013-01-17 18:51:17 +0100560 cpu->interrupt_request &= ~0x01;
Alex Bennéed10eb082016-11-14 14:17:28 +0000561 tlb_flush(cpu);
pbrook9656f322008-07-01 20:01:19 +0000562
563 return 0;
564}
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200565
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400566static int cpu_common_pre_load(void *opaque)
567{
568 CPUState *cpu = opaque;
569
Paolo Bonziniadee6422014-12-19 12:53:14 +0100570 cpu->exception_index = -1;
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400571
572 return 0;
573}
574
575static bool cpu_common_exception_index_needed(void *opaque)
576{
577 CPUState *cpu = opaque;
578
Paolo Bonziniadee6422014-12-19 12:53:14 +0100579 return tcg_enabled() && cpu->exception_index != -1;
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400580}
581
582static const VMStateDescription vmstate_cpu_common_exception_index = {
583 .name = "cpu_common/exception_index",
584 .version_id = 1,
585 .minimum_version_id = 1,
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200586 .needed = cpu_common_exception_index_needed,
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400587 .fields = (VMStateField[]) {
588 VMSTATE_INT32(exception_index, CPUState),
589 VMSTATE_END_OF_LIST()
590 }
591};
592
Andrey Smetaninbac05aa2015-07-03 15:01:44 +0300593static bool cpu_common_crash_occurred_needed(void *opaque)
594{
595 CPUState *cpu = opaque;
596
597 return cpu->crash_occurred;
598}
599
600static const VMStateDescription vmstate_cpu_common_crash_occurred = {
601 .name = "cpu_common/crash_occurred",
602 .version_id = 1,
603 .minimum_version_id = 1,
604 .needed = cpu_common_crash_occurred_needed,
605 .fields = (VMStateField[]) {
606 VMSTATE_BOOL(crash_occurred, CPUState),
607 VMSTATE_END_OF_LIST()
608 }
609};
610
Andreas Färber1a1562f2013-06-17 04:09:11 +0200611const VMStateDescription vmstate_cpu_common = {
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200612 .name = "cpu_common",
613 .version_id = 1,
614 .minimum_version_id = 1,
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400615 .pre_load = cpu_common_pre_load,
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200616 .post_load = cpu_common_post_load,
Juan Quintela35d08452014-04-16 16:01:33 +0200617 .fields = (VMStateField[]) {
Andreas Färber259186a2013-01-17 18:51:17 +0100618 VMSTATE_UINT32(halted, CPUState),
619 VMSTATE_UINT32(interrupt_request, CPUState),
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200620 VMSTATE_END_OF_LIST()
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400621 },
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200622 .subsections = (const VMStateDescription*[]) {
623 &vmstate_cpu_common_exception_index,
Andrey Smetaninbac05aa2015-07-03 15:01:44 +0300624 &vmstate_cpu_common_crash_occurred,
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200625 NULL
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200626 }
627};
Andreas Färber1a1562f2013-06-17 04:09:11 +0200628
pbrook9656f322008-07-01 20:01:19 +0000629#endif
630
Andreas Färber38d8f5c2012-12-17 19:47:15 +0100631CPUState *qemu_get_cpu(int index)
Glauber Costa950f1472009-06-09 12:15:18 -0400632{
Andreas Färberbdc44642013-06-24 23:50:24 +0200633 CPUState *cpu;
Glauber Costa950f1472009-06-09 12:15:18 -0400634
Andreas Färberbdc44642013-06-24 23:50:24 +0200635 CPU_FOREACH(cpu) {
Andreas Färber55e5c282012-12-17 06:18:02 +0100636 if (cpu->cpu_index == index) {
Andreas Färberbdc44642013-06-24 23:50:24 +0200637 return cpu;
Andreas Färber55e5c282012-12-17 06:18:02 +0100638 }
Glauber Costa950f1472009-06-09 12:15:18 -0400639 }
640
Andreas Färberbdc44642013-06-24 23:50:24 +0200641 return NULL;
Glauber Costa950f1472009-06-09 12:15:18 -0400642}
643
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000644#if !defined(CONFIG_USER_ONLY)
Peter Maydell56943e82016-01-21 14:15:04 +0000645void cpu_address_space_init(CPUState *cpu, AddressSpace *as, int asidx)
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000646{
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000647 CPUAddressSpace *newas;
648
649 /* Target code should have set num_ases before calling us */
650 assert(asidx < cpu->num_ases);
651
Peter Maydell56943e82016-01-21 14:15:04 +0000652 if (asidx == 0) {
653 /* address space 0 gets the convenience alias */
654 cpu->as = as;
655 }
656
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000657 /* KVM cannot currently support multiple address spaces. */
658 assert(asidx == 0 || !kvm_enabled());
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000659
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000660 if (!cpu->cpu_ases) {
661 cpu->cpu_ases = g_new0(CPUAddressSpace, cpu->num_ases);
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000662 }
Peter Maydell32857f42015-10-01 15:29:50 +0100663
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000664 newas = &cpu->cpu_ases[asidx];
665 newas->cpu = cpu;
666 newas->as = as;
Peter Maydell56943e82016-01-21 14:15:04 +0000667 if (tcg_enabled()) {
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000668 newas->tcg_as_listener.commit = tcg_commit;
669 memory_listener_register(&newas->tcg_as_listener, as);
Peter Maydell56943e82016-01-21 14:15:04 +0000670 }
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000671}
Peter Maydell651a5bc2016-01-21 14:15:05 +0000672
673AddressSpace *cpu_get_address_space(CPUState *cpu, int asidx)
674{
675 /* Return the AddressSpace corresponding to the specified index */
676 return cpu->cpu_ases[asidx].as;
677}
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000678#endif
679
Laurent Vivier7bbc1242016-10-20 13:26:04 +0200680void cpu_exec_unrealizefn(CPUState *cpu)
Bharata B Rao1c59eb32016-05-12 09:18:11 +0530681{
Bharata B Rao9dfeca72016-05-12 09:18:12 +0530682 CPUClass *cc = CPU_GET_CLASS(cpu);
683
Paolo Bonzini267f6852016-08-28 03:45:14 +0200684 cpu_list_remove(cpu);
Bharata B Rao9dfeca72016-05-12 09:18:12 +0530685
686 if (cc->vmsd != NULL) {
687 vmstate_unregister(NULL, cc->vmsd, cpu);
688 }
689 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
690 vmstate_unregister(NULL, &vmstate_cpu_common, cpu);
691 }
Bharata B Rao1c59eb32016-05-12 09:18:11 +0530692}
693
Laurent Vivier39e329e2016-10-20 13:26:02 +0200694void cpu_exec_initfn(CPUState *cpu)
bellardfd6ce8f2003-05-14 19:00:11 +0000695{
Peter Maydell56943e82016-01-21 14:15:04 +0000696 cpu->as = NULL;
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000697 cpu->num_ases = 0;
Peter Maydell56943e82016-01-21 14:15:04 +0000698
Eduardo Habkost291135b2015-04-27 17:00:33 -0300699#ifndef CONFIG_USER_ONLY
Eduardo Habkost291135b2015-04-27 17:00:33 -0300700 cpu->thread_id = qemu_get_thread_id();
Peter Crosthwaite6731d862016-01-21 14:15:06 +0000701
702 /* This is a softmmu CPU object, so create a property for it
703 * so users can wire up its memory. (This can't go in qom/cpu.c
704 * because that file is compiled only once for both user-mode
705 * and system builds.) The default if no link is set up is to use
706 * the system address space.
707 */
708 object_property_add_link(OBJECT(cpu), "memory", TYPE_MEMORY_REGION,
709 (Object **)&cpu->memory,
710 qdev_prop_allow_set_link_before_realize,
711 OBJ_PROP_LINK_UNREF_ON_RELEASE,
712 &error_abort);
713 cpu->memory = system_memory;
714 object_ref(OBJECT(cpu->memory));
Eduardo Habkost291135b2015-04-27 17:00:33 -0300715#endif
Laurent Vivier39e329e2016-10-20 13:26:02 +0200716}
717
Laurent Vivierce5b1bb2016-10-20 13:26:03 +0200718void cpu_exec_realizefn(CPUState *cpu, Error **errp)
Laurent Vivier39e329e2016-10-20 13:26:02 +0200719{
720 CPUClass *cc ATTRIBUTE_UNUSED = CPU_GET_CLASS(cpu);
Eduardo Habkost291135b2015-04-27 17:00:33 -0300721
Paolo Bonzini267f6852016-08-28 03:45:14 +0200722 cpu_list_add(cpu);
Igor Mammedov1bc7e522016-07-25 11:59:19 +0200723
724#ifndef CONFIG_USER_ONLY
Andreas Färbere0d47942013-07-29 04:07:50 +0200725 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200726 vmstate_register(NULL, cpu->cpu_index, &vmstate_cpu_common, cpu);
Andreas Färbere0d47942013-07-29 04:07:50 +0200727 }
Andreas Färberb170fce2013-01-20 20:23:22 +0100728 if (cc->vmsd != NULL) {
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200729 vmstate_register(NULL, cpu->cpu_index, cc->vmsd, cpu);
Andreas Färberb170fce2013-01-20 20:23:22 +0100730 }
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200731#endif
bellardfd6ce8f2003-05-14 19:00:11 +0000732}
733
Andreas Färber00b941e2013-06-29 18:55:54 +0200734static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
Paul Brook94df27f2010-02-28 23:47:45 +0000735{
Peter Maydella9353fe2016-12-06 18:07:09 +0000736 /* Flush the whole TB as this will not have race conditions
737 * even if we don't have proper locking yet.
738 * Ideally we would just invalidate the TBs for the
739 * specified PC.
740 */
741 tb_flush(cpu);
Paul Brook94df27f2010-02-28 23:47:45 +0000742}
bellardd720b932004-04-25 17:57:43 +0000743
Paul Brookc527ee82010-03-01 03:31:14 +0000744#if defined(CONFIG_USER_ONLY)
Andreas Färber75a34032013-09-02 16:57:02 +0200745void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
Paul Brookc527ee82010-03-01 03:31:14 +0000746
747{
748}
749
Peter Maydell3ee887e2014-09-12 14:06:48 +0100750int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
751 int flags)
752{
753 return -ENOSYS;
754}
755
756void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
757{
758}
759
Andreas Färber75a34032013-09-02 16:57:02 +0200760int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
Paul Brookc527ee82010-03-01 03:31:14 +0000761 int flags, CPUWatchpoint **watchpoint)
762{
763 return -ENOSYS;
764}
765#else
pbrook6658ffb2007-03-16 23:58:11 +0000766/* Add a watchpoint. */
Andreas Färber75a34032013-09-02 16:57:02 +0200767int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
aliguoria1d1bb32008-11-18 20:07:32 +0000768 int flags, CPUWatchpoint **watchpoint)
pbrook6658ffb2007-03-16 23:58:11 +0000769{
aliguoric0ce9982008-11-25 22:13:57 +0000770 CPUWatchpoint *wp;
pbrook6658ffb2007-03-16 23:58:11 +0000771
Peter Maydell05068c02014-09-12 14:06:48 +0100772 /* forbid ranges which are empty or run off the end of the address space */
Max Filippov07e28632014-09-17 22:03:36 -0700773 if (len == 0 || (addr + len - 1) < addr) {
Andreas Färber75a34032013-09-02 16:57:02 +0200774 error_report("tried to set invalid watchpoint at %"
775 VADDR_PRIx ", len=%" VADDR_PRIu, addr, len);
aliguorib4051332008-11-18 20:14:20 +0000776 return -EINVAL;
777 }
Anthony Liguori7267c092011-08-20 22:09:37 -0500778 wp = g_malloc(sizeof(*wp));
pbrook6658ffb2007-03-16 23:58:11 +0000779
aliguoria1d1bb32008-11-18 20:07:32 +0000780 wp->vaddr = addr;
Peter Maydell05068c02014-09-12 14:06:48 +0100781 wp->len = len;
aliguoria1d1bb32008-11-18 20:07:32 +0000782 wp->flags = flags;
783
aliguori2dc9f412008-11-18 20:56:59 +0000784 /* keep all GDB-injected watchpoints in front */
Andreas Färberff4700b2013-08-26 18:23:18 +0200785 if (flags & BP_GDB) {
786 QTAILQ_INSERT_HEAD(&cpu->watchpoints, wp, entry);
787 } else {
788 QTAILQ_INSERT_TAIL(&cpu->watchpoints, wp, entry);
789 }
aliguoria1d1bb32008-11-18 20:07:32 +0000790
Andreas Färber31b030d2013-09-04 01:29:02 +0200791 tlb_flush_page(cpu, addr);
aliguoria1d1bb32008-11-18 20:07:32 +0000792
793 if (watchpoint)
794 *watchpoint = wp;
795 return 0;
pbrook6658ffb2007-03-16 23:58:11 +0000796}
797
aliguoria1d1bb32008-11-18 20:07:32 +0000798/* Remove a specific watchpoint. */
Andreas Färber75a34032013-09-02 16:57:02 +0200799int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
aliguoria1d1bb32008-11-18 20:07:32 +0000800 int flags)
pbrook6658ffb2007-03-16 23:58:11 +0000801{
aliguoria1d1bb32008-11-18 20:07:32 +0000802 CPUWatchpoint *wp;
pbrook6658ffb2007-03-16 23:58:11 +0000803
Andreas Färberff4700b2013-08-26 18:23:18 +0200804 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +0100805 if (addr == wp->vaddr && len == wp->len
aliguori6e140f22008-11-18 20:37:55 +0000806 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
Andreas Färber75a34032013-09-02 16:57:02 +0200807 cpu_watchpoint_remove_by_ref(cpu, wp);
pbrook6658ffb2007-03-16 23:58:11 +0000808 return 0;
809 }
810 }
aliguoria1d1bb32008-11-18 20:07:32 +0000811 return -ENOENT;
pbrook6658ffb2007-03-16 23:58:11 +0000812}
813
aliguoria1d1bb32008-11-18 20:07:32 +0000814/* Remove a specific watchpoint by reference. */
Andreas Färber75a34032013-09-02 16:57:02 +0200815void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
aliguoria1d1bb32008-11-18 20:07:32 +0000816{
Andreas Färberff4700b2013-08-26 18:23:18 +0200817 QTAILQ_REMOVE(&cpu->watchpoints, watchpoint, entry);
edgar_igl7d03f822008-05-17 18:58:29 +0000818
Andreas Färber31b030d2013-09-04 01:29:02 +0200819 tlb_flush_page(cpu, watchpoint->vaddr);
aliguoria1d1bb32008-11-18 20:07:32 +0000820
Anthony Liguori7267c092011-08-20 22:09:37 -0500821 g_free(watchpoint);
edgar_igl7d03f822008-05-17 18:58:29 +0000822}
823
aliguoria1d1bb32008-11-18 20:07:32 +0000824/* Remove all matching watchpoints. */
Andreas Färber75a34032013-09-02 16:57:02 +0200825void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
aliguoria1d1bb32008-11-18 20:07:32 +0000826{
aliguoric0ce9982008-11-25 22:13:57 +0000827 CPUWatchpoint *wp, *next;
aliguoria1d1bb32008-11-18 20:07:32 +0000828
Andreas Färberff4700b2013-08-26 18:23:18 +0200829 QTAILQ_FOREACH_SAFE(wp, &cpu->watchpoints, entry, next) {
Andreas Färber75a34032013-09-02 16:57:02 +0200830 if (wp->flags & mask) {
831 cpu_watchpoint_remove_by_ref(cpu, wp);
832 }
aliguoric0ce9982008-11-25 22:13:57 +0000833 }
aliguoria1d1bb32008-11-18 20:07:32 +0000834}
Peter Maydell05068c02014-09-12 14:06:48 +0100835
836/* Return true if this watchpoint address matches the specified
837 * access (ie the address range covered by the watchpoint overlaps
838 * partially or completely with the address range covered by the
839 * access).
840 */
841static inline bool cpu_watchpoint_address_matches(CPUWatchpoint *wp,
842 vaddr addr,
843 vaddr len)
844{
845 /* We know the lengths are non-zero, but a little caution is
846 * required to avoid errors in the case where the range ends
847 * exactly at the top of the address space and so addr + len
848 * wraps round to zero.
849 */
850 vaddr wpend = wp->vaddr + wp->len - 1;
851 vaddr addrend = addr + len - 1;
852
853 return !(addr > wpend || wp->vaddr > addrend);
854}
855
Paul Brookc527ee82010-03-01 03:31:14 +0000856#endif
aliguoria1d1bb32008-11-18 20:07:32 +0000857
858/* Add a breakpoint. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200859int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags,
aliguoria1d1bb32008-11-18 20:07:32 +0000860 CPUBreakpoint **breakpoint)
bellard4c3a88a2003-07-26 12:06:08 +0000861{
aliguoric0ce9982008-11-25 22:13:57 +0000862 CPUBreakpoint *bp;
ths3b46e622007-09-17 08:09:54 +0000863
Anthony Liguori7267c092011-08-20 22:09:37 -0500864 bp = g_malloc(sizeof(*bp));
aliguoria1d1bb32008-11-18 20:07:32 +0000865
866 bp->pc = pc;
867 bp->flags = flags;
868
aliguori2dc9f412008-11-18 20:56:59 +0000869 /* keep all GDB-injected breakpoints in front */
Andreas Färber00b941e2013-06-29 18:55:54 +0200870 if (flags & BP_GDB) {
Andreas Färberf0c3c502013-08-26 21:22:53 +0200871 QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry);
Andreas Färber00b941e2013-06-29 18:55:54 +0200872 } else {
Andreas Färberf0c3c502013-08-26 21:22:53 +0200873 QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry);
Andreas Färber00b941e2013-06-29 18:55:54 +0200874 }
aliguoria1d1bb32008-11-18 20:07:32 +0000875
Andreas Färberf0c3c502013-08-26 21:22:53 +0200876 breakpoint_invalidate(cpu, pc);
aliguoria1d1bb32008-11-18 20:07:32 +0000877
Andreas Färber00b941e2013-06-29 18:55:54 +0200878 if (breakpoint) {
aliguoria1d1bb32008-11-18 20:07:32 +0000879 *breakpoint = bp;
Andreas Färber00b941e2013-06-29 18:55:54 +0200880 }
aliguoria1d1bb32008-11-18 20:07:32 +0000881 return 0;
aliguoria1d1bb32008-11-18 20:07:32 +0000882}
883
884/* Remove a specific breakpoint. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200885int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags)
aliguoria1d1bb32008-11-18 20:07:32 +0000886{
aliguoria1d1bb32008-11-18 20:07:32 +0000887 CPUBreakpoint *bp;
888
Andreas Färberf0c3c502013-08-26 21:22:53 +0200889 QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) {
aliguoria1d1bb32008-11-18 20:07:32 +0000890 if (bp->pc == pc && bp->flags == flags) {
Andreas Färberb3310ab2013-09-02 17:26:20 +0200891 cpu_breakpoint_remove_by_ref(cpu, bp);
bellard4c3a88a2003-07-26 12:06:08 +0000892 return 0;
aliguoria1d1bb32008-11-18 20:07:32 +0000893 }
bellard4c3a88a2003-07-26 12:06:08 +0000894 }
aliguoria1d1bb32008-11-18 20:07:32 +0000895 return -ENOENT;
bellard4c3a88a2003-07-26 12:06:08 +0000896}
897
aliguoria1d1bb32008-11-18 20:07:32 +0000898/* Remove a specific breakpoint by reference. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200899void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *breakpoint)
bellard4c3a88a2003-07-26 12:06:08 +0000900{
Andreas Färberf0c3c502013-08-26 21:22:53 +0200901 QTAILQ_REMOVE(&cpu->breakpoints, breakpoint, entry);
902
903 breakpoint_invalidate(cpu, breakpoint->pc);
aliguoria1d1bb32008-11-18 20:07:32 +0000904
Anthony Liguori7267c092011-08-20 22:09:37 -0500905 g_free(breakpoint);
aliguoria1d1bb32008-11-18 20:07:32 +0000906}
907
908/* Remove all matching breakpoints. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200909void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
aliguoria1d1bb32008-11-18 20:07:32 +0000910{
aliguoric0ce9982008-11-25 22:13:57 +0000911 CPUBreakpoint *bp, *next;
aliguoria1d1bb32008-11-18 20:07:32 +0000912
Andreas Färberf0c3c502013-08-26 21:22:53 +0200913 QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) {
Andreas Färberb3310ab2013-09-02 17:26:20 +0200914 if (bp->flags & mask) {
915 cpu_breakpoint_remove_by_ref(cpu, bp);
916 }
aliguoric0ce9982008-11-25 22:13:57 +0000917 }
bellard4c3a88a2003-07-26 12:06:08 +0000918}
919
bellardc33a3462003-07-29 20:50:33 +0000920/* enable or disable single step mode. EXCP_DEBUG is returned by the
921 CPU loop after each instruction */
Andreas Färber3825b282013-06-24 18:41:06 +0200922void cpu_single_step(CPUState *cpu, int enabled)
bellardc33a3462003-07-29 20:50:33 +0000923{
Andreas Färbered2803d2013-06-21 20:20:45 +0200924 if (cpu->singlestep_enabled != enabled) {
925 cpu->singlestep_enabled = enabled;
926 if (kvm_enabled()) {
Stefan Weil38e478e2013-07-25 20:50:21 +0200927 kvm_update_guest_debug(cpu, 0);
Andreas Färbered2803d2013-06-21 20:20:45 +0200928 } else {
Stuart Bradyccbb4d42009-05-03 12:15:06 +0100929 /* must flush all the translated code to avoid inconsistencies */
aliguorie22a25c2009-03-12 20:12:48 +0000930 /* XXX: only flush what is necessary */
Peter Crosthwaitebbd77c12015-06-23 19:31:15 -0700931 tb_flush(cpu);
aliguorie22a25c2009-03-12 20:12:48 +0000932 }
bellardc33a3462003-07-29 20:50:33 +0000933 }
bellardc33a3462003-07-29 20:50:33 +0000934}
935
Andreas Färbera47dddd2013-09-03 17:38:47 +0200936void cpu_abort(CPUState *cpu, const char *fmt, ...)
bellard75012672003-06-21 13:11:07 +0000937{
938 va_list ap;
pbrook493ae1f2007-11-23 16:53:59 +0000939 va_list ap2;
bellard75012672003-06-21 13:11:07 +0000940
941 va_start(ap, fmt);
pbrook493ae1f2007-11-23 16:53:59 +0000942 va_copy(ap2, ap);
bellard75012672003-06-21 13:11:07 +0000943 fprintf(stderr, "qemu: fatal: ");
944 vfprintf(stderr, fmt, ap);
945 fprintf(stderr, "\n");
Andreas Färber878096e2013-05-27 01:33:50 +0200946 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP);
Paolo Bonzini013a2942015-11-13 13:16:27 +0100947 if (qemu_log_separate()) {
Richard Henderson1ee73212016-09-22 15:17:10 -0700948 qemu_log_lock();
aliguori93fcfe32009-01-15 22:34:14 +0000949 qemu_log("qemu: fatal: ");
950 qemu_log_vprintf(fmt, ap2);
951 qemu_log("\n");
Andreas Färbera0762852013-06-16 07:28:50 +0200952 log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP);
aliguori31b1a7b2009-01-15 22:35:09 +0000953 qemu_log_flush();
Richard Henderson1ee73212016-09-22 15:17:10 -0700954 qemu_log_unlock();
aliguori93fcfe32009-01-15 22:34:14 +0000955 qemu_log_close();
balrog924edca2007-06-10 14:07:13 +0000956 }
pbrook493ae1f2007-11-23 16:53:59 +0000957 va_end(ap2);
j_mayerf9373292007-09-29 12:18:20 +0000958 va_end(ap);
Pavel Dovgalyuk76159362015-09-17 19:25:07 +0300959 replay_finish();
Riku Voipiofd052bf2010-01-25 14:30:49 +0200960#if defined(CONFIG_USER_ONLY)
961 {
962 struct sigaction act;
963 sigfillset(&act.sa_mask);
964 act.sa_handler = SIG_DFL;
965 sigaction(SIGABRT, &act, NULL);
966 }
967#endif
bellard75012672003-06-21 13:11:07 +0000968 abort();
969}
970
bellard01243112004-01-04 15:48:17 +0000971#if !defined(CONFIG_USER_ONLY)
Mike Day0dc3f442013-09-05 14:41:35 -0400972/* Called from RCU critical section */
Paolo Bonzini041603f2013-09-09 17:49:45 +0200973static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
974{
975 RAMBlock *block;
976
Paolo Bonzini43771532013-09-09 17:58:40 +0200977 block = atomic_rcu_read(&ram_list.mru_block);
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +0200978 if (block && addr - block->offset < block->max_length) {
Paolo Bonzini68851b92015-10-22 13:51:30 +0200979 return block;
Paolo Bonzini041603f2013-09-09 17:49:45 +0200980 }
Mike Day0dc3f442013-09-05 14:41:35 -0400981 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +0200982 if (addr - block->offset < block->max_length) {
Paolo Bonzini041603f2013-09-09 17:49:45 +0200983 goto found;
984 }
985 }
986
987 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
988 abort();
989
990found:
Paolo Bonzini43771532013-09-09 17:58:40 +0200991 /* It is safe to write mru_block outside the iothread lock. This
992 * is what happens:
993 *
994 * mru_block = xxx
995 * rcu_read_unlock()
996 * xxx removed from list
997 * rcu_read_lock()
998 * read mru_block
999 * mru_block = NULL;
1000 * call_rcu(reclaim_ramblock, xxx);
1001 * rcu_read_unlock()
1002 *
1003 * atomic_rcu_set is not needed here. The block was already published
1004 * when it was placed into the list. Here we're just making an extra
1005 * copy of the pointer.
1006 */
Paolo Bonzini041603f2013-09-09 17:49:45 +02001007 ram_list.mru_block = block;
1008 return block;
1009}
1010
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001011static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length)
bellard1ccde1c2004-02-06 19:46:14 +00001012{
Peter Crosthwaite9a135652015-09-10 22:39:41 -07001013 CPUState *cpu;
Paolo Bonzini041603f2013-09-09 17:49:45 +02001014 ram_addr_t start1;
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001015 RAMBlock *block;
1016 ram_addr_t end;
1017
1018 end = TARGET_PAGE_ALIGN(start + length);
1019 start &= TARGET_PAGE_MASK;
bellardf23db162005-08-21 19:12:28 +00001020
Mike Day0dc3f442013-09-05 14:41:35 -04001021 rcu_read_lock();
Paolo Bonzini041603f2013-09-09 17:49:45 +02001022 block = qemu_get_ram_block(start);
1023 assert(block == qemu_get_ram_block(end - 1));
Michael S. Tsirkin1240be22014-11-12 11:44:41 +02001024 start1 = (uintptr_t)ramblock_ptr(block, start - block->offset);
Peter Crosthwaite9a135652015-09-10 22:39:41 -07001025 CPU_FOREACH(cpu) {
1026 tlb_reset_dirty(cpu, start1, length);
1027 }
Mike Day0dc3f442013-09-05 14:41:35 -04001028 rcu_read_unlock();
Juan Quintelad24981d2012-05-22 00:42:40 +02001029}
1030
1031/* Note: start and end must be within the same ram block. */
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001032bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start,
1033 ram_addr_t length,
1034 unsigned client)
Juan Quintelad24981d2012-05-22 00:42:40 +02001035{
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001036 DirtyMemoryBlocks *blocks;
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001037 unsigned long end, page;
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001038 bool dirty = false;
Juan Quintelad24981d2012-05-22 00:42:40 +02001039
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001040 if (length == 0) {
1041 return false;
1042 }
1043
1044 end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
1045 page = start >> TARGET_PAGE_BITS;
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001046
1047 rcu_read_lock();
1048
1049 blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
1050
1051 while (page < end) {
1052 unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE;
1053 unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE;
1054 unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset);
1055
1056 dirty |= bitmap_test_and_clear_atomic(blocks->blocks[idx],
1057 offset, num);
1058 page += num;
1059 }
1060
1061 rcu_read_unlock();
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001062
1063 if (dirty && tcg_enabled()) {
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001064 tlb_reset_dirty_range_all(start, length);
Juan Quintelad24981d2012-05-22 00:42:40 +02001065 }
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001066
1067 return dirty;
bellard1ccde1c2004-02-06 19:46:14 +00001068}
1069
Gerd Hoffmann8deaf122017-04-21 11:16:25 +02001070DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty
1071 (ram_addr_t start, ram_addr_t length, unsigned client)
1072{
1073 DirtyMemoryBlocks *blocks;
1074 unsigned long align = 1UL << (TARGET_PAGE_BITS + BITS_PER_LEVEL);
1075 ram_addr_t first = QEMU_ALIGN_DOWN(start, align);
1076 ram_addr_t last = QEMU_ALIGN_UP(start + length, align);
1077 DirtyBitmapSnapshot *snap;
1078 unsigned long page, end, dest;
1079
1080 snap = g_malloc0(sizeof(*snap) +
1081 ((last - first) >> (TARGET_PAGE_BITS + 3)));
1082 snap->start = first;
1083 snap->end = last;
1084
1085 page = first >> TARGET_PAGE_BITS;
1086 end = last >> TARGET_PAGE_BITS;
1087 dest = 0;
1088
1089 rcu_read_lock();
1090
1091 blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
1092
1093 while (page < end) {
1094 unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE;
1095 unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE;
1096 unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset);
1097
1098 assert(QEMU_IS_ALIGNED(offset, (1 << BITS_PER_LEVEL)));
1099 assert(QEMU_IS_ALIGNED(num, (1 << BITS_PER_LEVEL)));
1100 offset >>= BITS_PER_LEVEL;
1101
1102 bitmap_copy_and_clear_atomic(snap->dirty + dest,
1103 blocks->blocks[idx] + offset,
1104 num);
1105 page += num;
1106 dest += num >> BITS_PER_LEVEL;
1107 }
1108
1109 rcu_read_unlock();
1110
1111 if (tcg_enabled()) {
1112 tlb_reset_dirty_range_all(start, length);
1113 }
1114
1115 return snap;
1116}
1117
1118bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap,
1119 ram_addr_t start,
1120 ram_addr_t length)
1121{
1122 unsigned long page, end;
1123
1124 assert(start >= snap->start);
1125 assert(start + length <= snap->end);
1126
1127 end = TARGET_PAGE_ALIGN(start + length - snap->start) >> TARGET_PAGE_BITS;
1128 page = (start - snap->start) >> TARGET_PAGE_BITS;
1129
1130 while (page < end) {
1131 if (test_bit(page, snap->dirty)) {
1132 return true;
1133 }
1134 page++;
1135 }
1136 return false;
1137}
1138
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01001139/* Called from RCU critical section */
Andreas Färberbb0e6272013-09-03 13:32:01 +02001140hwaddr memory_region_section_get_iotlb(CPUState *cpu,
Paolo Bonzini149f54b2013-05-24 12:59:37 +02001141 MemoryRegionSection *section,
1142 target_ulong vaddr,
1143 hwaddr paddr, hwaddr xlat,
1144 int prot,
1145 target_ulong *address)
Blue Swirle5548612012-04-21 13:08:33 +00001146{
Avi Kivitya8170e52012-10-23 12:30:10 +02001147 hwaddr iotlb;
Blue Swirle5548612012-04-21 13:08:33 +00001148 CPUWatchpoint *wp;
1149
Blue Swirlcc5bea62012-04-14 14:56:48 +00001150 if (memory_region_is_ram(section->mr)) {
Blue Swirle5548612012-04-21 13:08:33 +00001151 /* Normal RAM. */
Paolo Bonzinie4e69792016-03-01 10:44:50 +01001152 iotlb = memory_region_get_ram_addr(section->mr) + xlat;
Blue Swirle5548612012-04-21 13:08:33 +00001153 if (!section->readonly) {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001154 iotlb |= PHYS_SECTION_NOTDIRTY;
Blue Swirle5548612012-04-21 13:08:33 +00001155 } else {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001156 iotlb |= PHYS_SECTION_ROM;
Blue Swirle5548612012-04-21 13:08:33 +00001157 }
1158 } else {
Peter Maydell0b8e2c12015-07-20 12:27:16 +01001159 AddressSpaceDispatch *d;
1160
1161 d = atomic_rcu_read(&section->address_space->dispatch);
1162 iotlb = section - d->map.sections;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02001163 iotlb += xlat;
Blue Swirle5548612012-04-21 13:08:33 +00001164 }
1165
1166 /* Make accesses to pages with watchpoints go via the
1167 watchpoint trap routines. */
Andreas Färberff4700b2013-08-26 18:23:18 +02001168 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +01001169 if (cpu_watchpoint_address_matches(wp, vaddr, TARGET_PAGE_SIZE)) {
Blue Swirle5548612012-04-21 13:08:33 +00001170 /* Avoid trapping reads of pages with a write breakpoint. */
1171 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001172 iotlb = PHYS_SECTION_WATCH + paddr;
Blue Swirle5548612012-04-21 13:08:33 +00001173 *address |= TLB_MMIO;
1174 break;
1175 }
1176 }
1177 }
1178
1179 return iotlb;
1180}
bellard9fa3e852004-01-04 18:06:42 +00001181#endif /* defined(CONFIG_USER_ONLY) */
1182
pbrooke2eef172008-06-08 01:09:01 +00001183#if !defined(CONFIG_USER_ONLY)
pbrook8da3ff12008-12-01 18:59:50 +00001184
Anthony Liguoric227f092009-10-01 16:12:16 -05001185static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
Avi Kivity5312bd82012-02-12 18:32:55 +02001186 uint16_t section);
Jan Kiszkaacc9d802013-05-26 21:55:37 +02001187static subpage_t *subpage_init(AddressSpace *as, hwaddr base);
Avi Kivity54688b12012-02-09 17:34:32 +02001188
Igor Mammedova2b257d2014-10-31 16:38:37 +00001189static void *(*phys_mem_alloc)(size_t size, uint64_t *align) =
1190 qemu_anon_ram_alloc;
Markus Armbruster91138032013-07-31 15:11:08 +02001191
1192/*
1193 * Set a custom physical guest memory alloator.
1194 * Accelerators with unusual needs may need this. Hopefully, we can
1195 * get rid of it eventually.
1196 */
Igor Mammedova2b257d2014-10-31 16:38:37 +00001197void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align))
Markus Armbruster91138032013-07-31 15:11:08 +02001198{
1199 phys_mem_alloc = alloc;
1200}
1201
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001202static uint16_t phys_section_add(PhysPageMap *map,
1203 MemoryRegionSection *section)
Avi Kivity5312bd82012-02-12 18:32:55 +02001204{
Paolo Bonzini68f3f652013-05-07 11:30:23 +02001205 /* The physical section number is ORed with a page-aligned
1206 * pointer to produce the iotlb entries. Thus it should
1207 * never overflow into the page-aligned value.
1208 */
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001209 assert(map->sections_nb < TARGET_PAGE_SIZE);
Paolo Bonzini68f3f652013-05-07 11:30:23 +02001210
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001211 if (map->sections_nb == map->sections_nb_alloc) {
1212 map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16);
1213 map->sections = g_renew(MemoryRegionSection, map->sections,
1214 map->sections_nb_alloc);
Avi Kivity5312bd82012-02-12 18:32:55 +02001215 }
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001216 map->sections[map->sections_nb] = *section;
Paolo Bonzinidfde4e62013-05-06 10:46:11 +02001217 memory_region_ref(section->mr);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001218 return map->sections_nb++;
Avi Kivity5312bd82012-02-12 18:32:55 +02001219}
1220
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001221static void phys_section_destroy(MemoryRegion *mr)
1222{
Don Slutz55b4e802015-11-30 17:11:04 -05001223 bool have_sub_page = mr->subpage;
1224
Paolo Bonzinidfde4e62013-05-06 10:46:11 +02001225 memory_region_unref(mr);
1226
Don Slutz55b4e802015-11-30 17:11:04 -05001227 if (have_sub_page) {
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001228 subpage_t *subpage = container_of(mr, subpage_t, iomem);
Peter Crosthwaiteb4fefef2014-06-05 23:15:52 -07001229 object_unref(OBJECT(&subpage->iomem));
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001230 g_free(subpage);
1231 }
1232}
1233
Paolo Bonzini60926662013-05-29 12:30:26 +02001234static void phys_sections_free(PhysPageMap *map)
Avi Kivity5312bd82012-02-12 18:32:55 +02001235{
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02001236 while (map->sections_nb > 0) {
1237 MemoryRegionSection *section = &map->sections[--map->sections_nb];
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001238 phys_section_destroy(section->mr);
1239 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02001240 g_free(map->sections);
1241 g_free(map->nodes);
Avi Kivity5312bd82012-02-12 18:32:55 +02001242}
1243
Avi Kivityac1970f2012-10-03 16:22:53 +02001244static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section)
Avi Kivity0f0cb162012-02-13 17:14:32 +02001245{
1246 subpage_t *subpage;
Avi Kivitya8170e52012-10-23 12:30:10 +02001247 hwaddr base = section->offset_within_address_space
Avi Kivity0f0cb162012-02-13 17:14:32 +02001248 & TARGET_PAGE_MASK;
Michael S. Tsirkin97115a82013-11-13 20:08:19 +02001249 MemoryRegionSection *existing = phys_page_find(d->phys_map, base,
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001250 d->map.nodes, d->map.sections);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001251 MemoryRegionSection subsection = {
1252 .offset_within_address_space = base,
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001253 .size = int128_make64(TARGET_PAGE_SIZE),
Avi Kivity0f0cb162012-02-13 17:14:32 +02001254 };
Avi Kivitya8170e52012-10-23 12:30:10 +02001255 hwaddr start, end;
Avi Kivity0f0cb162012-02-13 17:14:32 +02001256
Avi Kivityf3705d52012-03-08 16:16:34 +02001257 assert(existing->mr->subpage || existing->mr == &io_mem_unassigned);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001258
Avi Kivityf3705d52012-03-08 16:16:34 +02001259 if (!(existing->mr->subpage)) {
Jan Kiszkaacc9d802013-05-26 21:55:37 +02001260 subpage = subpage_init(d->as, base);
Edgar E. Iglesias3be91e82013-11-07 18:42:51 +01001261 subsection.address_space = d->as;
Avi Kivity0f0cb162012-02-13 17:14:32 +02001262 subsection.mr = &subpage->iomem;
Avi Kivityac1970f2012-10-03 16:22:53 +02001263 phys_page_set(d, base >> TARGET_PAGE_BITS, 1,
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001264 phys_section_add(&d->map, &subsection));
Avi Kivity0f0cb162012-02-13 17:14:32 +02001265 } else {
Avi Kivityf3705d52012-03-08 16:16:34 +02001266 subpage = container_of(existing->mr, subpage_t, iomem);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001267 }
1268 start = section->offset_within_address_space & ~TARGET_PAGE_MASK;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001269 end = start + int128_get64(section->size) - 1;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001270 subpage_register(subpage, start, end,
1271 phys_section_add(&d->map, section));
Avi Kivity0f0cb162012-02-13 17:14:32 +02001272}
1273
1274
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001275static void register_multipage(AddressSpaceDispatch *d,
1276 MemoryRegionSection *section)
bellard33417e72003-08-10 21:47:01 +00001277{
Avi Kivitya8170e52012-10-23 12:30:10 +02001278 hwaddr start_addr = section->offset_within_address_space;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001279 uint16_t section_index = phys_section_add(&d->map, section);
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001280 uint64_t num_pages = int128_get64(int128_rshift(section->size,
1281 TARGET_PAGE_BITS));
Avi Kivitydd811242012-01-02 12:17:03 +02001282
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001283 assert(num_pages);
1284 phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index);
bellard33417e72003-08-10 21:47:01 +00001285}
1286
Avi Kivityac1970f2012-10-03 16:22:53 +02001287static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
Avi Kivity0f0cb162012-02-13 17:14:32 +02001288{
Paolo Bonzini89ae3372013-06-02 10:39:07 +02001289 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
Paolo Bonzini00752702013-05-29 12:13:54 +02001290 AddressSpaceDispatch *d = as->next_dispatch;
Paolo Bonzini99b9cc02013-05-27 13:18:01 +02001291 MemoryRegionSection now = *section, remain = *section;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001292 Int128 page_size = int128_make64(TARGET_PAGE_SIZE);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001293
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001294 if (now.offset_within_address_space & ~TARGET_PAGE_MASK) {
1295 uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space)
1296 - now.offset_within_address_space;
1297
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001298 now.size = int128_min(int128_make64(left), now.size);
Avi Kivityac1970f2012-10-03 16:22:53 +02001299 register_subpage(d, &now);
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001300 } else {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001301 now.size = int128_zero();
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001302 }
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001303 while (int128_ne(remain.size, now.size)) {
1304 remain.size = int128_sub(remain.size, now.size);
1305 remain.offset_within_address_space += int128_get64(now.size);
1306 remain.offset_within_region += int128_get64(now.size);
Tyler Hall69b67642012-07-25 18:45:04 -04001307 now = remain;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001308 if (int128_lt(remain.size, page_size)) {
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001309 register_subpage(d, &now);
Hu Tao88266242013-08-29 18:21:16 +08001310 } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001311 now.size = page_size;
Avi Kivityac1970f2012-10-03 16:22:53 +02001312 register_subpage(d, &now);
Tyler Hall69b67642012-07-25 18:45:04 -04001313 } else {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001314 now.size = int128_and(now.size, int128_neg(page_size));
Avi Kivityac1970f2012-10-03 16:22:53 +02001315 register_multipage(d, &now);
Tyler Hall69b67642012-07-25 18:45:04 -04001316 }
Avi Kivity0f0cb162012-02-13 17:14:32 +02001317 }
1318}
1319
Sheng Yang62a27442010-01-26 19:21:16 +08001320void qemu_flush_coalesced_mmio_buffer(void)
1321{
1322 if (kvm_enabled())
1323 kvm_flush_coalesced_mmio_buffer();
1324}
1325
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001326void qemu_mutex_lock_ramlist(void)
1327{
1328 qemu_mutex_lock(&ram_list.mutex);
1329}
1330
1331void qemu_mutex_unlock_ramlist(void)
1332{
1333 qemu_mutex_unlock(&ram_list.mutex);
1334}
1335
Markus Armbrustere1e84ba2013-07-31 15:11:10 +02001336#ifdef __linux__
Alexey Kardashevskiy9c607662017-03-02 13:36:11 +11001337/*
1338 * FIXME TOCTTOU: this iterates over memory backends' mem-path, which
1339 * may or may not name the same files / on the same filesystem now as
1340 * when we actually open and map them. Iterate over the file
1341 * descriptors instead, and use qemu_fd_getpagesize().
1342 */
1343static int find_max_supported_pagesize(Object *obj, void *opaque)
1344{
1345 char *mem_path;
1346 long *hpsize_min = opaque;
1347
1348 if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
1349 mem_path = object_property_get_str(obj, "mem-path", NULL);
1350 if (mem_path) {
1351 long hpsize = qemu_mempath_getpagesize(mem_path);
1352 if (hpsize < *hpsize_min) {
1353 *hpsize_min = hpsize;
1354 }
1355 } else {
1356 *hpsize_min = getpagesize();
1357 }
1358 }
1359
1360 return 0;
1361}
1362
1363long qemu_getrampagesize(void)
1364{
1365 long hpsize = LONG_MAX;
1366 long mainrampagesize;
1367 Object *memdev_root;
1368
1369 if (mem_path) {
1370 mainrampagesize = qemu_mempath_getpagesize(mem_path);
1371 } else {
1372 mainrampagesize = getpagesize();
1373 }
1374
1375 /* it's possible we have memory-backend objects with
1376 * hugepage-backed RAM. these may get mapped into system
1377 * address space via -numa parameters or memory hotplug
1378 * hooks. we want to take these into account, but we
1379 * also want to make sure these supported hugepage
1380 * sizes are applicable across the entire range of memory
1381 * we may boot from, so we take the min across all
1382 * backends, and assume normal pages in cases where a
1383 * backend isn't backed by hugepages.
1384 */
1385 memdev_root = object_resolve_path("/objects", NULL);
1386 if (memdev_root) {
1387 object_child_foreach(memdev_root, find_max_supported_pagesize, &hpsize);
1388 }
1389 if (hpsize == LONG_MAX) {
1390 /* No additional memory regions found ==> Report main RAM page size */
1391 return mainrampagesize;
1392 }
1393
1394 /* If NUMA is disabled or the NUMA nodes are not backed with a
1395 * memory-backend, then there is at least one node using "normal" RAM,
1396 * so if its page size is smaller we have got to report that size instead.
1397 */
1398 if (hpsize > mainrampagesize &&
1399 (nb_numa_nodes == 0 || numa_info[0].node_memdev == NULL)) {
1400 static bool warned;
1401 if (!warned) {
1402 error_report("Huge page support disabled (n/a for main memory).");
1403 warned = true;
1404 }
1405 return mainrampagesize;
1406 }
1407
1408 return hpsize;
1409}
1410#else
1411long qemu_getrampagesize(void)
1412{
1413 return getpagesize();
1414}
1415#endif
1416
1417#ifdef __linux__
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001418static int64_t get_file_size(int fd)
1419{
1420 int64_t size = lseek(fd, 0, SEEK_END);
1421 if (size < 0) {
1422 return -errno;
1423 }
1424 return size;
1425}
1426
Alex Williamson04b16652010-07-02 11:13:17 -06001427static void *file_ram_alloc(RAMBlock *block,
1428 ram_addr_t memory,
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001429 const char *path,
1430 Error **errp)
Marcelo Tosattic9027602010-03-01 20:25:08 -03001431{
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001432 bool unlink_on_error = false;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001433 char *filename;
Peter Feiner8ca761f2013-03-04 13:54:25 -05001434 char *sanitized_name;
1435 char *c;
Igor Mammedov056b68a2016-07-20 11:54:03 +02001436 void *area = MAP_FAILED;
Paolo Bonzini5c3ece72016-03-17 15:53:13 +01001437 int fd = -1;
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001438 int64_t file_size;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001439
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001440 if (kvm_enabled() && !kvm_has_sync_mmu()) {
1441 error_setg(errp,
1442 "host lacks kvm mmu notifiers, -mem-path unsupported");
1443 return NULL;
1444 }
1445
1446 for (;;) {
1447 fd = open(path, O_RDWR);
1448 if (fd >= 0) {
1449 /* @path names an existing file, use it */
1450 break;
1451 }
1452 if (errno == ENOENT) {
1453 /* @path names a file that doesn't exist, create it */
1454 fd = open(path, O_RDWR | O_CREAT | O_EXCL, 0644);
1455 if (fd >= 0) {
1456 unlink_on_error = true;
1457 break;
1458 }
1459 } else if (errno == EISDIR) {
1460 /* @path names a directory, create a file there */
1461 /* Make name safe to use with mkstemp by replacing '/' with '_'. */
1462 sanitized_name = g_strdup(memory_region_name(block->mr));
1463 for (c = sanitized_name; *c != '\0'; c++) {
1464 if (*c == '/') {
1465 *c = '_';
1466 }
1467 }
1468
1469 filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path,
1470 sanitized_name);
1471 g_free(sanitized_name);
1472
1473 fd = mkstemp(filename);
1474 if (fd >= 0) {
1475 unlink(filename);
1476 g_free(filename);
1477 break;
1478 }
1479 g_free(filename);
1480 }
1481 if (errno != EEXIST && errno != EINTR) {
1482 error_setg_errno(errp, errno,
1483 "can't open backing store %s for guest RAM",
1484 path);
1485 goto error;
1486 }
1487 /*
1488 * Try again on EINTR and EEXIST. The latter happens when
1489 * something else creates the file between our two open().
1490 */
1491 }
1492
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001493 block->page_size = qemu_fd_getpagesize(fd);
Haozhong Zhang83606682016-10-24 20:49:37 +08001494 block->mr->align = block->page_size;
1495#if defined(__s390x__)
1496 if (kvm_enabled()) {
1497 block->mr->align = MAX(block->mr->align, QEMU_VMALLOC_ALIGN);
1498 }
1499#endif
Marcelo Tosattic9027602010-03-01 20:25:08 -03001500
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001501 file_size = get_file_size(fd);
1502
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001503 if (memory < block->page_size) {
Hu Tao557529d2014-09-09 13:28:00 +08001504 error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to "
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001505 "or larger than page size 0x%zx",
1506 memory, block->page_size);
Hu Tao557529d2014-09-09 13:28:00 +08001507 goto error;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001508 }
1509
Haozhong Zhang1775f112016-11-02 09:05:51 +08001510 if (file_size > 0 && file_size < memory) {
1511 error_setg(errp, "backing store %s size 0x%" PRIx64
1512 " does not match 'size' option 0x" RAM_ADDR_FMT,
1513 path, file_size, memory);
1514 goto error;
1515 }
1516
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001517 memory = ROUND_UP(memory, block->page_size);
Marcelo Tosattic9027602010-03-01 20:25:08 -03001518
1519 /*
1520 * ftruncate is not supported by hugetlbfs in older
1521 * hosts, so don't bother bailing out on errors.
1522 * If anything goes wrong with it under other filesystems,
1523 * mmap will fail.
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001524 *
1525 * Do not truncate the non-empty backend file to avoid corrupting
1526 * the existing data in the file. Disabling shrinking is not
1527 * enough. For example, the current vNVDIMM implementation stores
1528 * the guest NVDIMM labels at the end of the backend file. If the
1529 * backend file is later extended, QEMU will not be able to find
1530 * those labels. Therefore, extending the non-empty backend file
1531 * is disabled as well.
Marcelo Tosattic9027602010-03-01 20:25:08 -03001532 */
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001533 if (!file_size && ftruncate(fd, memory)) {
Yoshiaki Tamura9742bf22010-08-18 13:30:13 +09001534 perror("ftruncate");
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001535 }
Marcelo Tosattic9027602010-03-01 20:25:08 -03001536
Dominik Dingeld2f39ad2016-04-25 13:55:38 +02001537 area = qemu_ram_mmap(fd, memory, block->mr->align,
1538 block->flags & RAM_SHARED);
Marcelo Tosattic9027602010-03-01 20:25:08 -03001539 if (area == MAP_FAILED) {
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001540 error_setg_errno(errp, errno,
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001541 "unable to map backing store for guest RAM");
Marcelo Tosattif9a49df2014-02-04 13:41:53 -05001542 goto error;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001543 }
Marcelo Tosattief36fa12013-10-28 18:51:46 -02001544
1545 if (mem_prealloc) {
Jitendra Kolhe1e356fc2017-02-24 09:01:43 +05301546 os_mem_prealloc(fd, area, memory, smp_cpus, errp);
Igor Mammedov056b68a2016-07-20 11:54:03 +02001547 if (errp && *errp) {
1548 goto error;
1549 }
Marcelo Tosattief36fa12013-10-28 18:51:46 -02001550 }
1551
Alex Williamson04b16652010-07-02 11:13:17 -06001552 block->fd = fd;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001553 return area;
Marcelo Tosattif9a49df2014-02-04 13:41:53 -05001554
1555error:
Igor Mammedov056b68a2016-07-20 11:54:03 +02001556 if (area != MAP_FAILED) {
1557 qemu_ram_munmap(area, memory);
1558 }
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001559 if (unlink_on_error) {
1560 unlink(path);
1561 }
Paolo Bonzini5c3ece72016-03-17 15:53:13 +01001562 if (fd != -1) {
1563 close(fd);
1564 }
Marcelo Tosattif9a49df2014-02-04 13:41:53 -05001565 return NULL;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001566}
1567#endif
1568
Mike Day0dc3f442013-09-05 14:41:35 -04001569/* Called with the ramlist lock held. */
Alex Williamsond17b5282010-06-25 11:08:38 -06001570static ram_addr_t find_ram_offset(ram_addr_t size)
1571{
Alex Williamson04b16652010-07-02 11:13:17 -06001572 RAMBlock *block, *next_block;
Alex Williamson3e837b22011-10-31 08:54:09 -06001573 ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX;
Alex Williamson04b16652010-07-02 11:13:17 -06001574
Stefan Hajnoczi49cd9ac2013-03-11 10:20:21 +01001575 assert(size != 0); /* it would hand out same offset multiple times */
1576
Mike Day0dc3f442013-09-05 14:41:35 -04001577 if (QLIST_EMPTY_RCU(&ram_list.blocks)) {
Alex Williamson04b16652010-07-02 11:13:17 -06001578 return 0;
Mike Day0d53d9f2015-01-21 13:45:24 +01001579 }
Alex Williamson04b16652010-07-02 11:13:17 -06001580
Mike Day0dc3f442013-09-05 14:41:35 -04001581 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Anthony PERARDf15fbc42011-07-20 08:17:42 +00001582 ram_addr_t end, next = RAM_ADDR_MAX;
Alex Williamson04b16652010-07-02 11:13:17 -06001583
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001584 end = block->offset + block->max_length;
Alex Williamson04b16652010-07-02 11:13:17 -06001585
Mike Day0dc3f442013-09-05 14:41:35 -04001586 QLIST_FOREACH_RCU(next_block, &ram_list.blocks, next) {
Alex Williamson04b16652010-07-02 11:13:17 -06001587 if (next_block->offset >= end) {
1588 next = MIN(next, next_block->offset);
1589 }
1590 }
1591 if (next - end >= size && next - end < mingap) {
Alex Williamson3e837b22011-10-31 08:54:09 -06001592 offset = end;
Alex Williamson04b16652010-07-02 11:13:17 -06001593 mingap = next - end;
1594 }
1595 }
Alex Williamson3e837b22011-10-31 08:54:09 -06001596
1597 if (offset == RAM_ADDR_MAX) {
1598 fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n",
1599 (uint64_t)size);
1600 abort();
1601 }
1602
Alex Williamson04b16652010-07-02 11:13:17 -06001603 return offset;
1604}
1605
Juan Quintelab8c48992017-03-21 17:44:30 +01001606unsigned long last_ram_page(void)
Alex Williamson04b16652010-07-02 11:13:17 -06001607{
Alex Williamsond17b5282010-06-25 11:08:38 -06001608 RAMBlock *block;
1609 ram_addr_t last = 0;
1610
Mike Day0dc3f442013-09-05 14:41:35 -04001611 rcu_read_lock();
1612 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001613 last = MAX(last, block->offset + block->max_length);
Mike Day0d53d9f2015-01-21 13:45:24 +01001614 }
Mike Day0dc3f442013-09-05 14:41:35 -04001615 rcu_read_unlock();
Juan Quintelab8c48992017-03-21 17:44:30 +01001616 return last >> TARGET_PAGE_BITS;
Alex Williamsond17b5282010-06-25 11:08:38 -06001617}
1618
Jason Baronddb97f12012-08-02 15:44:16 -04001619static void qemu_ram_setup_dump(void *addr, ram_addr_t size)
1620{
1621 int ret;
Jason Baronddb97f12012-08-02 15:44:16 -04001622
1623 /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */
Marcel Apfelbaum47c8ca52015-02-04 17:43:54 +02001624 if (!machine_dump_guest_core(current_machine)) {
Jason Baronddb97f12012-08-02 15:44:16 -04001625 ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP);
1626 if (ret) {
1627 perror("qemu_madvise");
1628 fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, "
1629 "but dump_guest_core=off specified\n");
1630 }
1631 }
1632}
1633
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00001634const char *qemu_ram_get_idstr(RAMBlock *rb)
1635{
1636 return rb->idstr;
1637}
1638
Dr. David Alan Gilbert463a4ac2017-03-07 18:36:36 +00001639bool qemu_ram_is_shared(RAMBlock *rb)
1640{
1641 return rb->flags & RAM_SHARED;
1642}
1643
Mike Dayae3a7042013-09-05 14:41:35 -04001644/* Called with iothread lock held. */
Gongleifa53a0e2016-05-10 10:04:59 +08001645void qemu_ram_set_idstr(RAMBlock *new_block, const char *name, DeviceState *dev)
Hu Tao20cfe882014-04-02 15:13:26 +08001646{
Gongleifa53a0e2016-05-10 10:04:59 +08001647 RAMBlock *block;
Hu Tao20cfe882014-04-02 15:13:26 +08001648
Avi Kivityc5705a72011-12-20 15:59:12 +02001649 assert(new_block);
1650 assert(!new_block->idstr[0]);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001651
Anthony Liguori09e5ab62012-02-03 12:28:43 -06001652 if (dev) {
1653 char *id = qdev_get_dev_path(dev);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001654 if (id) {
1655 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
Anthony Liguori7267c092011-08-20 22:09:37 -05001656 g_free(id);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001657 }
1658 }
1659 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
1660
Gongleiab0a9952016-05-10 10:05:00 +08001661 rcu_read_lock();
Mike Day0dc3f442013-09-05 14:41:35 -04001662 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Gongleifa53a0e2016-05-10 10:04:59 +08001663 if (block != new_block &&
1664 !strcmp(block->idstr, new_block->idstr)) {
Cam Macdonell84b89d72010-07-26 18:10:57 -06001665 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
1666 new_block->idstr);
1667 abort();
1668 }
1669 }
Mike Day0dc3f442013-09-05 14:41:35 -04001670 rcu_read_unlock();
Avi Kivityc5705a72011-12-20 15:59:12 +02001671}
1672
Mike Dayae3a7042013-09-05 14:41:35 -04001673/* Called with iothread lock held. */
Gongleifa53a0e2016-05-10 10:04:59 +08001674void qemu_ram_unset_idstr(RAMBlock *block)
Hu Tao20cfe882014-04-02 15:13:26 +08001675{
Mike Dayae3a7042013-09-05 14:41:35 -04001676 /* FIXME: arch_init.c assumes that this is not called throughout
1677 * migration. Ignore the problem since hot-unplug during migration
1678 * does not work anyway.
1679 */
Hu Tao20cfe882014-04-02 15:13:26 +08001680 if (block) {
1681 memset(block->idstr, 0, sizeof(block->idstr));
1682 }
1683}
1684
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001685size_t qemu_ram_pagesize(RAMBlock *rb)
1686{
1687 return rb->page_size;
1688}
1689
Dr. David Alan Gilbert67f11b52017-02-24 18:28:34 +00001690/* Returns the largest size of page in use */
1691size_t qemu_ram_pagesize_largest(void)
1692{
1693 RAMBlock *block;
1694 size_t largest = 0;
1695
1696 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
1697 largest = MAX(largest, qemu_ram_pagesize(block));
1698 }
1699
1700 return largest;
1701}
1702
Luiz Capitulino8490fc72012-09-05 16:50:16 -03001703static int memory_try_enable_merging(void *addr, size_t len)
1704{
Marcel Apfelbaum75cc7f02015-02-04 17:43:55 +02001705 if (!machine_mem_merge(current_machine)) {
Luiz Capitulino8490fc72012-09-05 16:50:16 -03001706 /* disabled by the user */
1707 return 0;
1708 }
1709
1710 return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE);
1711}
1712
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001713/* Only legal before guest might have detected the memory size: e.g. on
1714 * incoming migration, or right after reset.
1715 *
1716 * As memory core doesn't know how is memory accessed, it is up to
1717 * resize callback to update device state and/or add assertions to detect
1718 * misuse, if necessary.
1719 */
Gongleifa53a0e2016-05-10 10:04:59 +08001720int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp)
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001721{
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001722 assert(block);
1723
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00001724 newsize = HOST_PAGE_ALIGN(newsize);
Michael S. Tsirkin129ddaf2015-02-17 10:15:30 +01001725
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001726 if (block->used_length == newsize) {
1727 return 0;
1728 }
1729
1730 if (!(block->flags & RAM_RESIZEABLE)) {
1731 error_setg_errno(errp, EINVAL,
1732 "Length mismatch: %s: 0x" RAM_ADDR_FMT
1733 " in != 0x" RAM_ADDR_FMT, block->idstr,
1734 newsize, block->used_length);
1735 return -EINVAL;
1736 }
1737
1738 if (block->max_length < newsize) {
1739 error_setg_errno(errp, EINVAL,
1740 "Length too large: %s: 0x" RAM_ADDR_FMT
1741 " > 0x" RAM_ADDR_FMT, block->idstr,
1742 newsize, block->max_length);
1743 return -EINVAL;
1744 }
1745
1746 cpu_physical_memory_clear_dirty_range(block->offset, block->used_length);
1747 block->used_length = newsize;
Paolo Bonzini58d27072015-03-23 11:56:01 +01001748 cpu_physical_memory_set_dirty_range(block->offset, block->used_length,
1749 DIRTY_CLIENTS_ALL);
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001750 memory_region_set_size(block->mr, newsize);
1751 if (block->resized) {
1752 block->resized(block->idstr, newsize, block->host);
1753 }
1754 return 0;
1755}
1756
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001757/* Called with ram_list.mutex held */
1758static void dirty_memory_extend(ram_addr_t old_ram_size,
1759 ram_addr_t new_ram_size)
1760{
1761 ram_addr_t old_num_blocks = DIV_ROUND_UP(old_ram_size,
1762 DIRTY_MEMORY_BLOCK_SIZE);
1763 ram_addr_t new_num_blocks = DIV_ROUND_UP(new_ram_size,
1764 DIRTY_MEMORY_BLOCK_SIZE);
1765 int i;
1766
1767 /* Only need to extend if block count increased */
1768 if (new_num_blocks <= old_num_blocks) {
1769 return;
1770 }
1771
1772 for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
1773 DirtyMemoryBlocks *old_blocks;
1774 DirtyMemoryBlocks *new_blocks;
1775 int j;
1776
1777 old_blocks = atomic_rcu_read(&ram_list.dirty_memory[i]);
1778 new_blocks = g_malloc(sizeof(*new_blocks) +
1779 sizeof(new_blocks->blocks[0]) * new_num_blocks);
1780
1781 if (old_num_blocks) {
1782 memcpy(new_blocks->blocks, old_blocks->blocks,
1783 old_num_blocks * sizeof(old_blocks->blocks[0]));
1784 }
1785
1786 for (j = old_num_blocks; j < new_num_blocks; j++) {
1787 new_blocks->blocks[j] = bitmap_new(DIRTY_MEMORY_BLOCK_SIZE);
1788 }
1789
1790 atomic_rcu_set(&ram_list.dirty_memory[i], new_blocks);
1791
1792 if (old_blocks) {
1793 g_free_rcu(old_blocks, rcu);
1794 }
1795 }
1796}
1797
Fam Zheng528f46a2016-03-01 14:18:18 +08001798static void ram_block_add(RAMBlock *new_block, Error **errp)
Avi Kivityc5705a72011-12-20 15:59:12 +02001799{
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001800 RAMBlock *block;
Mike Day0d53d9f2015-01-21 13:45:24 +01001801 RAMBlock *last_block = NULL;
Juan Quintela2152f5c2013-10-08 13:52:02 +02001802 ram_addr_t old_ram_size, new_ram_size;
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001803 Error *err = NULL;
Juan Quintela2152f5c2013-10-08 13:52:02 +02001804
Juan Quintelab8c48992017-03-21 17:44:30 +01001805 old_ram_size = last_ram_page();
Avi Kivityc5705a72011-12-20 15:59:12 +02001806
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001807 qemu_mutex_lock_ramlist();
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001808 new_block->offset = find_ram_offset(new_block->max_length);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001809
1810 if (!new_block->host) {
1811 if (xen_enabled()) {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001812 xen_ram_alloc(new_block->offset, new_block->max_length,
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001813 new_block->mr, &err);
1814 if (err) {
1815 error_propagate(errp, err);
1816 qemu_mutex_unlock_ramlist();
Paolo Bonzini39c350e2016-03-09 18:14:01 +01001817 return;
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001818 }
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001819 } else {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001820 new_block->host = phys_mem_alloc(new_block->max_length,
Igor Mammedova2b257d2014-10-31 16:38:37 +00001821 &new_block->mr->align);
Markus Armbruster39228252013-07-31 15:11:11 +02001822 if (!new_block->host) {
Hu Taoef701d72014-09-09 13:27:54 +08001823 error_setg_errno(errp, errno,
1824 "cannot set up guest memory '%s'",
1825 memory_region_name(new_block->mr));
1826 qemu_mutex_unlock_ramlist();
Paolo Bonzini39c350e2016-03-09 18:14:01 +01001827 return;
Markus Armbruster39228252013-07-31 15:11:11 +02001828 }
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001829 memory_try_enable_merging(new_block->host, new_block->max_length);
Yoshiaki Tamura6977dfe2010-08-18 15:41:49 +09001830 }
1831 }
Cam Macdonell84b89d72010-07-26 18:10:57 -06001832
Li Zhijiandd631692015-07-02 20:18:06 +08001833 new_ram_size = MAX(old_ram_size,
1834 (new_block->offset + new_block->max_length) >> TARGET_PAGE_BITS);
1835 if (new_ram_size > old_ram_size) {
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001836 dirty_memory_extend(old_ram_size, new_ram_size);
Li Zhijiandd631692015-07-02 20:18:06 +08001837 }
Mike Day0d53d9f2015-01-21 13:45:24 +01001838 /* Keep the list sorted from biggest to smallest block. Unlike QTAILQ,
1839 * QLIST (which has an RCU-friendly variant) does not have insertion at
1840 * tail, so save the last element in last_block.
1841 */
Mike Day0dc3f442013-09-05 14:41:35 -04001842 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Mike Day0d53d9f2015-01-21 13:45:24 +01001843 last_block = block;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001844 if (block->max_length < new_block->max_length) {
Paolo Bonziniabb26d62012-11-14 16:00:51 +01001845 break;
1846 }
1847 }
1848 if (block) {
Mike Day0dc3f442013-09-05 14:41:35 -04001849 QLIST_INSERT_BEFORE_RCU(block, new_block, next);
Mike Day0d53d9f2015-01-21 13:45:24 +01001850 } else if (last_block) {
Mike Day0dc3f442013-09-05 14:41:35 -04001851 QLIST_INSERT_AFTER_RCU(last_block, new_block, next);
Mike Day0d53d9f2015-01-21 13:45:24 +01001852 } else { /* list is empty */
Mike Day0dc3f442013-09-05 14:41:35 -04001853 QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next);
Paolo Bonziniabb26d62012-11-14 16:00:51 +01001854 }
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01001855 ram_list.mru_block = NULL;
Cam Macdonell84b89d72010-07-26 18:10:57 -06001856
Mike Day0dc3f442013-09-05 14:41:35 -04001857 /* Write list before version */
1858 smp_wmb();
Umesh Deshpandef798b072011-08-18 11:41:17 -07001859 ram_list.version++;
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001860 qemu_mutex_unlock_ramlist();
Umesh Deshpandef798b072011-08-18 11:41:17 -07001861
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001862 cpu_physical_memory_set_dirty_range(new_block->offset,
Paolo Bonzini58d27072015-03-23 11:56:01 +01001863 new_block->used_length,
1864 DIRTY_CLIENTS_ALL);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001865
Paolo Bonzinia904c912015-01-21 16:18:35 +01001866 if (new_block->host) {
1867 qemu_ram_setup_dump(new_block->host, new_block->max_length);
1868 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_HUGEPAGE);
Cao jinc2cd6272016-09-12 14:34:56 +08001869 /* MADV_DONTFORK is also needed by KVM in absence of synchronous MMU */
Paolo Bonzinia904c912015-01-21 16:18:35 +01001870 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_DONTFORK);
Paolo Bonzini0987d732016-12-21 00:31:36 +08001871 ram_block_notify_add(new_block->host, new_block->max_length);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001872 }
Cam Macdonell84b89d72010-07-26 18:10:57 -06001873}
1874
Paolo Bonzini0b183fc2014-05-14 17:43:19 +08001875#ifdef __linux__
Fam Zheng528f46a2016-03-01 14:18:18 +08001876RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
1877 bool share, const char *mem_path,
1878 Error **errp)
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001879{
1880 RAMBlock *new_block;
Hu Taoef701d72014-09-09 13:27:54 +08001881 Error *local_err = NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001882
1883 if (xen_enabled()) {
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001884 error_setg(errp, "-mem-path not supported with Xen");
Fam Zheng528f46a2016-03-01 14:18:18 +08001885 return NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001886 }
1887
1888 if (phys_mem_alloc != qemu_anon_ram_alloc) {
1889 /*
1890 * file_ram_alloc() needs to allocate just like
1891 * phys_mem_alloc, but we haven't bothered to provide
1892 * a hook there.
1893 */
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001894 error_setg(errp,
1895 "-mem-path not supported with this accelerator");
Fam Zheng528f46a2016-03-01 14:18:18 +08001896 return NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001897 }
1898
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00001899 size = HOST_PAGE_ALIGN(size);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001900 new_block = g_malloc0(sizeof(*new_block));
1901 new_block->mr = mr;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001902 new_block->used_length = size;
1903 new_block->max_length = size;
Paolo Bonzinidbcb8982014-06-10 19:15:24 +08001904 new_block->flags = share ? RAM_SHARED : 0;
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001905 new_block->host = file_ram_alloc(new_block, size,
1906 mem_path, errp);
1907 if (!new_block->host) {
1908 g_free(new_block);
Fam Zheng528f46a2016-03-01 14:18:18 +08001909 return NULL;
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001910 }
1911
Fam Zheng528f46a2016-03-01 14:18:18 +08001912 ram_block_add(new_block, &local_err);
Hu Taoef701d72014-09-09 13:27:54 +08001913 if (local_err) {
1914 g_free(new_block);
1915 error_propagate(errp, local_err);
Fam Zheng528f46a2016-03-01 14:18:18 +08001916 return NULL;
Hu Taoef701d72014-09-09 13:27:54 +08001917 }
Fam Zheng528f46a2016-03-01 14:18:18 +08001918 return new_block;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001919}
Paolo Bonzini0b183fc2014-05-14 17:43:19 +08001920#endif
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001921
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001922static
Fam Zheng528f46a2016-03-01 14:18:18 +08001923RAMBlock *qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size,
1924 void (*resized)(const char*,
1925 uint64_t length,
1926 void *host),
1927 void *host, bool resizeable,
1928 MemoryRegion *mr, Error **errp)
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001929{
1930 RAMBlock *new_block;
Hu Taoef701d72014-09-09 13:27:54 +08001931 Error *local_err = NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001932
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00001933 size = HOST_PAGE_ALIGN(size);
1934 max_size = HOST_PAGE_ALIGN(max_size);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001935 new_block = g_malloc0(sizeof(*new_block));
1936 new_block->mr = mr;
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001937 new_block->resized = resized;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001938 new_block->used_length = size;
1939 new_block->max_length = max_size;
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001940 assert(max_size >= size);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001941 new_block->fd = -1;
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001942 new_block->page_size = getpagesize();
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001943 new_block->host = host;
1944 if (host) {
Paolo Bonzini7bd4f432014-05-14 17:43:22 +08001945 new_block->flags |= RAM_PREALLOC;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001946 }
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001947 if (resizeable) {
1948 new_block->flags |= RAM_RESIZEABLE;
1949 }
Fam Zheng528f46a2016-03-01 14:18:18 +08001950 ram_block_add(new_block, &local_err);
Hu Taoef701d72014-09-09 13:27:54 +08001951 if (local_err) {
1952 g_free(new_block);
1953 error_propagate(errp, local_err);
Fam Zheng528f46a2016-03-01 14:18:18 +08001954 return NULL;
Hu Taoef701d72014-09-09 13:27:54 +08001955 }
Fam Zheng528f46a2016-03-01 14:18:18 +08001956 return new_block;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001957}
1958
Fam Zheng528f46a2016-03-01 14:18:18 +08001959RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001960 MemoryRegion *mr, Error **errp)
1961{
1962 return qemu_ram_alloc_internal(size, size, NULL, host, false, mr, errp);
1963}
1964
Fam Zheng528f46a2016-03-01 14:18:18 +08001965RAMBlock *qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr, Error **errp)
pbrook94a6b542009-04-11 17:15:54 +00001966{
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001967 return qemu_ram_alloc_internal(size, size, NULL, NULL, false, mr, errp);
1968}
1969
Fam Zheng528f46a2016-03-01 14:18:18 +08001970RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t maxsz,
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001971 void (*resized)(const char*,
1972 uint64_t length,
1973 void *host),
1974 MemoryRegion *mr, Error **errp)
1975{
1976 return qemu_ram_alloc_internal(size, maxsz, resized, NULL, true, mr, errp);
pbrook94a6b542009-04-11 17:15:54 +00001977}
bellarde9a1ab12007-02-08 23:08:38 +00001978
Paolo Bonzini43771532013-09-09 17:58:40 +02001979static void reclaim_ramblock(RAMBlock *block)
1980{
1981 if (block->flags & RAM_PREALLOC) {
1982 ;
1983 } else if (xen_enabled()) {
1984 xen_invalidate_map_cache_entry(block->host);
1985#ifndef _WIN32
1986 } else if (block->fd >= 0) {
Eduardo Habkost2f3a2bb2015-11-06 20:11:21 -02001987 qemu_ram_munmap(block->host, block->max_length);
Paolo Bonzini43771532013-09-09 17:58:40 +02001988 close(block->fd);
1989#endif
1990 } else {
1991 qemu_anon_ram_free(block->host, block->max_length);
1992 }
1993 g_free(block);
1994}
1995
Fam Zhengf1060c52016-03-01 14:18:22 +08001996void qemu_ram_free(RAMBlock *block)
bellarde9a1ab12007-02-08 23:08:38 +00001997{
Marc-André Lureau85bc2a12016-03-29 13:20:51 +02001998 if (!block) {
1999 return;
2000 }
2001
Paolo Bonzini0987d732016-12-21 00:31:36 +08002002 if (block->host) {
2003 ram_block_notify_remove(block->host, block->max_length);
2004 }
2005
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07002006 qemu_mutex_lock_ramlist();
Fam Zhengf1060c52016-03-01 14:18:22 +08002007 QLIST_REMOVE_RCU(block, next);
2008 ram_list.mru_block = NULL;
2009 /* Write list before version */
2010 smp_wmb();
2011 ram_list.version++;
2012 call_rcu(block, reclaim_ramblock, rcu);
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07002013 qemu_mutex_unlock_ramlist();
bellarde9a1ab12007-02-08 23:08:38 +00002014}
2015
Huang Yingcd19cfa2011-03-02 08:56:19 +01002016#ifndef _WIN32
2017void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
2018{
2019 RAMBlock *block;
2020 ram_addr_t offset;
2021 int flags;
2022 void *area, *vaddr;
2023
Mike Day0dc3f442013-09-05 14:41:35 -04002024 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Huang Yingcd19cfa2011-03-02 08:56:19 +01002025 offset = addr - block->offset;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002026 if (offset < block->max_length) {
Michael S. Tsirkin1240be22014-11-12 11:44:41 +02002027 vaddr = ramblock_ptr(block, offset);
Paolo Bonzini7bd4f432014-05-14 17:43:22 +08002028 if (block->flags & RAM_PREALLOC) {
Huang Yingcd19cfa2011-03-02 08:56:19 +01002029 ;
Markus Armbrusterdfeaf2a2013-07-31 15:11:05 +02002030 } else if (xen_enabled()) {
2031 abort();
Huang Yingcd19cfa2011-03-02 08:56:19 +01002032 } else {
2033 flags = MAP_FIXED;
Markus Armbruster3435f392013-07-31 15:11:07 +02002034 if (block->fd >= 0) {
Paolo Bonzinidbcb8982014-06-10 19:15:24 +08002035 flags |= (block->flags & RAM_SHARED ?
2036 MAP_SHARED : MAP_PRIVATE);
Markus Armbruster3435f392013-07-31 15:11:07 +02002037 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2038 flags, block->fd, offset);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002039 } else {
Markus Armbruster2eb9fba2013-07-31 15:11:09 +02002040 /*
2041 * Remap needs to match alloc. Accelerators that
2042 * set phys_mem_alloc never remap. If they did,
2043 * we'd need a remap hook here.
2044 */
2045 assert(phys_mem_alloc == qemu_anon_ram_alloc);
2046
Huang Yingcd19cfa2011-03-02 08:56:19 +01002047 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
2048 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2049 flags, -1, 0);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002050 }
2051 if (area != vaddr) {
Anthony PERARDf15fbc42011-07-20 08:17:42 +00002052 fprintf(stderr, "Could not remap addr: "
2053 RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n",
Huang Yingcd19cfa2011-03-02 08:56:19 +01002054 length, addr);
2055 exit(1);
2056 }
Luiz Capitulino8490fc72012-09-05 16:50:16 -03002057 memory_try_enable_merging(vaddr, length);
Jason Baronddb97f12012-08-02 15:44:16 -04002058 qemu_ram_setup_dump(vaddr, length);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002059 }
Huang Yingcd19cfa2011-03-02 08:56:19 +01002060 }
2061 }
2062}
2063#endif /* !_WIN32 */
2064
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002065/* Return a host pointer to ram allocated with qemu_ram_alloc.
Mike Dayae3a7042013-09-05 14:41:35 -04002066 * This should not be used for general purpose DMA. Use address_space_map
2067 * or address_space_rw instead. For local memory (e.g. video ram) that the
2068 * device owns, use memory_region_get_ram_ptr.
Mike Day0dc3f442013-09-05 14:41:35 -04002069 *
Paolo Bonzini49b24af2015-12-16 10:30:47 +01002070 * Called within RCU critical section.
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002071 */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002072void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr)
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002073{
Gonglei3655cb92016-02-20 10:35:20 +08002074 RAMBlock *block = ram_block;
2075
2076 if (block == NULL) {
2077 block = qemu_get_ram_block(addr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002078 addr -= block->offset;
Gonglei3655cb92016-02-20 10:35:20 +08002079 }
Mike Dayae3a7042013-09-05 14:41:35 -04002080
2081 if (xen_enabled() && block->host == NULL) {
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002082 /* We need to check if the requested address is in the RAM
2083 * because we don't want to map the entire memory in QEMU.
2084 * In that case just map until the end of the page.
2085 */
2086 if (block->offset == 0) {
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002087 return xen_map_cache(addr, 0, 0, false);
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002088 }
Mike Dayae3a7042013-09-05 14:41:35 -04002089
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002090 block->host = xen_map_cache(block->offset, block->max_length, 1, false);
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002091 }
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002092 return ramblock_ptr(block, addr);
pbrookdc828ca2009-04-09 22:21:07 +00002093}
2094
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002095/* Return a host pointer to guest's ram. Similar to qemu_map_ram_ptr
Mike Dayae3a7042013-09-05 14:41:35 -04002096 * but takes a size argument.
Mike Day0dc3f442013-09-05 14:41:35 -04002097 *
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002098 * Called within RCU critical section.
Mike Dayae3a7042013-09-05 14:41:35 -04002099 */
Gonglei3655cb92016-02-20 10:35:20 +08002100static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr,
2101 hwaddr *size)
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002102{
Gonglei3655cb92016-02-20 10:35:20 +08002103 RAMBlock *block = ram_block;
Stefano Stabellini8ab934f2011-06-27 18:26:06 +01002104 if (*size == 0) {
2105 return NULL;
2106 }
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002107
Gonglei3655cb92016-02-20 10:35:20 +08002108 if (block == NULL) {
2109 block = qemu_get_ram_block(addr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002110 addr -= block->offset;
Gonglei3655cb92016-02-20 10:35:20 +08002111 }
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002112 *size = MIN(*size, block->max_length - addr);
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002113
2114 if (xen_enabled() && block->host == NULL) {
2115 /* We need to check if the requested address is in the RAM
2116 * because we don't want to map the entire memory in QEMU.
2117 * In that case just map the requested area.
2118 */
2119 if (block->offset == 0) {
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002120 return xen_map_cache(addr, *size, 1, true);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002121 }
2122
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002123 block->host = xen_map_cache(block->offset, block->max_length, 1, true);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002124 }
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002125
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002126 return ramblock_ptr(block, addr);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002127}
2128
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002129/*
2130 * Translates a host ptr back to a RAMBlock, a ram_addr and an offset
2131 * in that RAMBlock.
2132 *
2133 * ptr: Host pointer to look up
2134 * round_offset: If true round the result offset down to a page boundary
2135 * *ram_addr: set to result ram_addr
2136 * *offset: set to result offset within the RAMBlock
2137 *
2138 * Returns: RAMBlock (or NULL if not found)
Mike Dayae3a7042013-09-05 14:41:35 -04002139 *
2140 * By the time this function returns, the returned pointer is not protected
2141 * by RCU anymore. If the caller is not within an RCU critical section and
2142 * does not hold the iothread lock, it must have other means of protecting the
2143 * pointer, such as a reference to the region that includes the incoming
2144 * ram_addr_t.
2145 */
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002146RAMBlock *qemu_ram_block_from_host(void *ptr, bool round_offset,
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002147 ram_addr_t *offset)
pbrook5579c7f2009-04-11 14:47:08 +00002148{
pbrook94a6b542009-04-11 17:15:54 +00002149 RAMBlock *block;
2150 uint8_t *host = ptr;
2151
Jan Kiszka868bb332011-06-21 22:59:09 +02002152 if (xen_enabled()) {
Paolo Bonzinif615f392016-05-26 10:07:50 +02002153 ram_addr_t ram_addr;
Mike Day0dc3f442013-09-05 14:41:35 -04002154 rcu_read_lock();
Paolo Bonzinif615f392016-05-26 10:07:50 +02002155 ram_addr = xen_ram_addr_from_mapcache(ptr);
2156 block = qemu_get_ram_block(ram_addr);
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002157 if (block) {
Anthony PERARDd6b6aec2016-06-09 16:56:17 +01002158 *offset = ram_addr - block->offset;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002159 }
Mike Day0dc3f442013-09-05 14:41:35 -04002160 rcu_read_unlock();
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002161 return block;
Stefano Stabellini712c2b42011-05-19 18:35:46 +01002162 }
2163
Mike Day0dc3f442013-09-05 14:41:35 -04002164 rcu_read_lock();
2165 block = atomic_rcu_read(&ram_list.mru_block);
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002166 if (block && block->host && host - block->host < block->max_length) {
Paolo Bonzini23887b72013-05-06 14:28:39 +02002167 goto found;
2168 }
2169
Mike Day0dc3f442013-09-05 14:41:35 -04002170 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Jun Nakajima432d2682010-08-31 16:41:25 +01002171 /* This case append when the block is not mapped. */
2172 if (block->host == NULL) {
2173 continue;
2174 }
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002175 if (host - block->host < block->max_length) {
Paolo Bonzini23887b72013-05-06 14:28:39 +02002176 goto found;
Alex Williamsonf471a172010-06-11 11:11:42 -06002177 }
pbrook94a6b542009-04-11 17:15:54 +00002178 }
Jun Nakajima432d2682010-08-31 16:41:25 +01002179
Mike Day0dc3f442013-09-05 14:41:35 -04002180 rcu_read_unlock();
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002181 return NULL;
Paolo Bonzini23887b72013-05-06 14:28:39 +02002182
2183found:
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002184 *offset = (host - block->host);
2185 if (round_offset) {
2186 *offset &= TARGET_PAGE_MASK;
2187 }
Mike Day0dc3f442013-09-05 14:41:35 -04002188 rcu_read_unlock();
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002189 return block;
2190}
2191
Dr. David Alan Gilberte3dd7492015-11-05 18:10:33 +00002192/*
2193 * Finds the named RAMBlock
2194 *
2195 * name: The name of RAMBlock to find
2196 *
2197 * Returns: RAMBlock (or NULL if not found)
2198 */
2199RAMBlock *qemu_ram_block_by_name(const char *name)
2200{
2201 RAMBlock *block;
2202
2203 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
2204 if (!strcmp(name, block->idstr)) {
2205 return block;
2206 }
2207 }
2208
2209 return NULL;
2210}
2211
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002212/* Some of the softmmu routines need to translate from a host pointer
2213 (typically a TLB entry) back to a ram offset. */
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002214ram_addr_t qemu_ram_addr_from_host(void *ptr)
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002215{
2216 RAMBlock *block;
Paolo Bonzinif615f392016-05-26 10:07:50 +02002217 ram_addr_t offset;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002218
Paolo Bonzinif615f392016-05-26 10:07:50 +02002219 block = qemu_ram_block_from_host(ptr, false, &offset);
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002220 if (!block) {
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002221 return RAM_ADDR_INVALID;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002222 }
2223
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002224 return block->offset + offset;
Marcelo Tosattie8902612010-10-11 15:31:19 -03002225}
Alex Williamsonf471a172010-06-11 11:11:42 -06002226
Paolo Bonzini49b24af2015-12-16 10:30:47 +01002227/* Called within RCU critical section. */
Avi Kivitya8170e52012-10-23 12:30:10 +02002228static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002229 uint64_t val, unsigned size)
bellard1ccde1c2004-02-06 19:46:14 +00002230{
Alex Bennéeba051fb2016-10-27 16:10:16 +01002231 bool locked = false;
2232
Juan Quintela52159192013-10-08 12:44:04 +02002233 if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) {
Alex Bennéeba051fb2016-10-27 16:10:16 +01002234 locked = true;
2235 tb_lock();
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002236 tb_invalidate_phys_page_fast(ram_addr, size);
bellard3a7d9292005-08-21 09:26:42 +00002237 }
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002238 switch (size) {
2239 case 1:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002240 stb_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002241 break;
2242 case 2:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002243 stw_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002244 break;
2245 case 4:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002246 stl_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002247 break;
2248 default:
2249 abort();
2250 }
Alex Bennéeba051fb2016-10-27 16:10:16 +01002251
2252 if (locked) {
2253 tb_unlock();
2254 }
2255
Paolo Bonzini58d27072015-03-23 11:56:01 +01002256 /* Set both VGA and migration bits for simplicity and to remove
2257 * the notdirty callback faster.
2258 */
2259 cpu_physical_memory_set_dirty_range(ram_addr, size,
2260 DIRTY_CLIENTS_NOCODE);
bellardf23db162005-08-21 19:12:28 +00002261 /* we remove the notdirty callback only if the code has been
2262 flushed */
Juan Quintelaa2cd8c82013-10-10 11:20:22 +02002263 if (!cpu_physical_memory_is_clean(ram_addr)) {
Peter Crosthwaitebcae01e2015-09-10 22:39:42 -07002264 tlb_set_dirty(current_cpu, current_cpu->mem_io_vaddr);
Andreas Färber4917cf42013-05-27 05:17:50 +02002265 }
bellard1ccde1c2004-02-06 19:46:14 +00002266}
2267
Paolo Bonzinib018ddf2013-05-24 14:48:38 +02002268static bool notdirty_mem_accepts(void *opaque, hwaddr addr,
2269 unsigned size, bool is_write)
2270{
2271 return is_write;
2272}
2273
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002274static const MemoryRegionOps notdirty_mem_ops = {
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002275 .write = notdirty_mem_write,
Paolo Bonzinib018ddf2013-05-24 14:48:38 +02002276 .valid.accepts = notdirty_mem_accepts,
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002277 .endianness = DEVICE_NATIVE_ENDIAN,
bellard1ccde1c2004-02-06 19:46:14 +00002278};
2279
pbrook0f459d12008-06-09 00:20:13 +00002280/* Generate a debug exception if a watchpoint has been hit. */
Peter Maydell66b9b432015-04-26 16:49:24 +01002281static void check_watchpoint(int offset, int len, MemTxAttrs attrs, int flags)
pbrook0f459d12008-06-09 00:20:13 +00002282{
Andreas Färber93afead2013-08-26 03:41:01 +02002283 CPUState *cpu = current_cpu;
Sergey Fedorov568496c2016-02-11 11:17:32 +00002284 CPUClass *cc = CPU_GET_CLASS(cpu);
Andreas Färber93afead2013-08-26 03:41:01 +02002285 CPUArchState *env = cpu->env_ptr;
aliguori06d55cc2008-11-18 20:24:06 +00002286 target_ulong pc, cs_base;
pbrook0f459d12008-06-09 00:20:13 +00002287 target_ulong vaddr;
aliguoria1d1bb32008-11-18 20:07:32 +00002288 CPUWatchpoint *wp;
Emilio G. Cota89fee742016-04-07 13:19:22 -04002289 uint32_t cpu_flags;
pbrook0f459d12008-06-09 00:20:13 +00002290
Andreas Färberff4700b2013-08-26 18:23:18 +02002291 if (cpu->watchpoint_hit) {
aliguori06d55cc2008-11-18 20:24:06 +00002292 /* We re-entered the check after replacing the TB. Now raise
2293 * the debug interrupt so that is will trigger after the
2294 * current instruction. */
Andreas Färber93afead2013-08-26 03:41:01 +02002295 cpu_interrupt(cpu, CPU_INTERRUPT_DEBUG);
aliguori06d55cc2008-11-18 20:24:06 +00002296 return;
2297 }
Andreas Färber93afead2013-08-26 03:41:01 +02002298 vaddr = (cpu->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
Julian Brown40612002017-02-07 18:29:59 +00002299 vaddr = cc->adjust_watchpoint_address(cpu, vaddr, len);
Andreas Färberff4700b2013-08-26 18:23:18 +02002300 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +01002301 if (cpu_watchpoint_address_matches(wp, vaddr, len)
2302 && (wp->flags & flags)) {
Peter Maydell08225672014-09-12 14:06:48 +01002303 if (flags == BP_MEM_READ) {
2304 wp->flags |= BP_WATCHPOINT_HIT_READ;
2305 } else {
2306 wp->flags |= BP_WATCHPOINT_HIT_WRITE;
2307 }
2308 wp->hitaddr = vaddr;
Peter Maydell66b9b432015-04-26 16:49:24 +01002309 wp->hitattrs = attrs;
Andreas Färberff4700b2013-08-26 18:23:18 +02002310 if (!cpu->watchpoint_hit) {
Sergey Fedorov568496c2016-02-11 11:17:32 +00002311 if (wp->flags & BP_CPU &&
2312 !cc->debug_check_watchpoint(cpu, wp)) {
2313 wp->flags &= ~BP_WATCHPOINT_HIT;
2314 continue;
2315 }
Andreas Färberff4700b2013-08-26 18:23:18 +02002316 cpu->watchpoint_hit = wp;
KONRAD Frederica5e99822016-10-27 16:10:06 +01002317
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002318 /* Both tb_lock and iothread_mutex will be reset when
2319 * cpu_loop_exit or cpu_loop_exit_noexc longjmp
2320 * back into the cpu_exec main loop.
KONRAD Frederica5e99822016-10-27 16:10:06 +01002321 */
2322 tb_lock();
Andreas Färber239c51a2013-09-01 17:12:23 +02002323 tb_check_watchpoint(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002324 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
Andreas Färber27103422013-08-26 08:31:06 +02002325 cpu->exception_index = EXCP_DEBUG;
Andreas Färber5638d182013-08-27 17:52:12 +02002326 cpu_loop_exit(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002327 } else {
2328 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
Andreas Färber648f0342013-09-01 17:43:17 +02002329 tb_gen_code(cpu, pc, cs_base, cpu_flags, 1);
Peter Maydell6886b982016-05-17 15:18:04 +01002330 cpu_loop_exit_noexc(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002331 }
aliguori06d55cc2008-11-18 20:24:06 +00002332 }
aliguori6e140f22008-11-18 20:37:55 +00002333 } else {
2334 wp->flags &= ~BP_WATCHPOINT_HIT;
pbrook0f459d12008-06-09 00:20:13 +00002335 }
2336 }
2337}
2338
pbrook6658ffb2007-03-16 23:58:11 +00002339/* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
2340 so these check for a hit then pass through to the normal out-of-line
2341 phys routines. */
Peter Maydell66b9b432015-04-26 16:49:24 +01002342static MemTxResult watch_mem_read(void *opaque, hwaddr addr, uint64_t *pdata,
2343 unsigned size, MemTxAttrs attrs)
pbrook6658ffb2007-03-16 23:58:11 +00002344{
Peter Maydell66b9b432015-04-26 16:49:24 +01002345 MemTxResult res;
2346 uint64_t data;
Peter Maydell79ed0412016-01-21 14:15:06 +00002347 int asidx = cpu_asidx_from_attrs(current_cpu, attrs);
2348 AddressSpace *as = current_cpu->cpu_ases[asidx].as;
pbrook6658ffb2007-03-16 23:58:11 +00002349
Peter Maydell66b9b432015-04-26 16:49:24 +01002350 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_READ);
Avi Kivity1ec9b902012-01-02 12:47:48 +02002351 switch (size) {
Max Filippov67364152012-01-29 00:01:40 +04002352 case 1:
Peter Maydell79ed0412016-01-21 14:15:06 +00002353 data = address_space_ldub(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002354 break;
2355 case 2:
Peter Maydell79ed0412016-01-21 14:15:06 +00002356 data = address_space_lduw(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002357 break;
2358 case 4:
Peter Maydell79ed0412016-01-21 14:15:06 +00002359 data = address_space_ldl(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002360 break;
Avi Kivity1ec9b902012-01-02 12:47:48 +02002361 default: abort();
2362 }
Peter Maydell66b9b432015-04-26 16:49:24 +01002363 *pdata = data;
2364 return res;
2365}
2366
2367static MemTxResult watch_mem_write(void *opaque, hwaddr addr,
2368 uint64_t val, unsigned size,
2369 MemTxAttrs attrs)
2370{
2371 MemTxResult res;
Peter Maydell79ed0412016-01-21 14:15:06 +00002372 int asidx = cpu_asidx_from_attrs(current_cpu, attrs);
2373 AddressSpace *as = current_cpu->cpu_ases[asidx].as;
Peter Maydell66b9b432015-04-26 16:49:24 +01002374
2375 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_WRITE);
2376 switch (size) {
2377 case 1:
Peter Maydell79ed0412016-01-21 14:15:06 +00002378 address_space_stb(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002379 break;
2380 case 2:
Peter Maydell79ed0412016-01-21 14:15:06 +00002381 address_space_stw(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002382 break;
2383 case 4:
Peter Maydell79ed0412016-01-21 14:15:06 +00002384 address_space_stl(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002385 break;
2386 default: abort();
2387 }
2388 return res;
pbrook6658ffb2007-03-16 23:58:11 +00002389}
2390
Avi Kivity1ec9b902012-01-02 12:47:48 +02002391static const MemoryRegionOps watch_mem_ops = {
Peter Maydell66b9b432015-04-26 16:49:24 +01002392 .read_with_attrs = watch_mem_read,
2393 .write_with_attrs = watch_mem_write,
Avi Kivity1ec9b902012-01-02 12:47:48 +02002394 .endianness = DEVICE_NATIVE_ENDIAN,
pbrook6658ffb2007-03-16 23:58:11 +00002395};
pbrook6658ffb2007-03-16 23:58:11 +00002396
Peter Maydellf25a49e2015-04-26 16:49:24 +01002397static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data,
2398 unsigned len, MemTxAttrs attrs)
blueswir1db7b5422007-05-26 17:36:03 +00002399{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002400 subpage_t *subpage = opaque;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002401 uint8_t buf[8];
Peter Maydell5c9eb022015-04-26 16:49:24 +01002402 MemTxResult res;
Paolo Bonzini791af8c2013-05-24 16:10:39 +02002403
blueswir1db7b5422007-05-26 17:36:03 +00002404#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002405 printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__,
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002406 subpage, len, addr);
blueswir1db7b5422007-05-26 17:36:03 +00002407#endif
Peter Maydell5c9eb022015-04-26 16:49:24 +01002408 res = address_space_read(subpage->as, addr + subpage->base,
2409 attrs, buf, len);
2410 if (res) {
2411 return res;
Peter Maydellf25a49e2015-04-26 16:49:24 +01002412 }
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002413 switch (len) {
2414 case 1:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002415 *data = ldub_p(buf);
2416 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002417 case 2:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002418 *data = lduw_p(buf);
2419 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002420 case 4:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002421 *data = ldl_p(buf);
2422 return MEMTX_OK;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002423 case 8:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002424 *data = ldq_p(buf);
2425 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002426 default:
2427 abort();
2428 }
blueswir1db7b5422007-05-26 17:36:03 +00002429}
2430
Peter Maydellf25a49e2015-04-26 16:49:24 +01002431static MemTxResult subpage_write(void *opaque, hwaddr addr,
2432 uint64_t value, unsigned len, MemTxAttrs attrs)
blueswir1db7b5422007-05-26 17:36:03 +00002433{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002434 subpage_t *subpage = opaque;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002435 uint8_t buf[8];
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002436
blueswir1db7b5422007-05-26 17:36:03 +00002437#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002438 printf("%s: subpage %p len %u addr " TARGET_FMT_plx
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002439 " value %"PRIx64"\n",
2440 __func__, subpage, len, addr, value);
blueswir1db7b5422007-05-26 17:36:03 +00002441#endif
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002442 switch (len) {
2443 case 1:
2444 stb_p(buf, value);
2445 break;
2446 case 2:
2447 stw_p(buf, value);
2448 break;
2449 case 4:
2450 stl_p(buf, value);
2451 break;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002452 case 8:
2453 stq_p(buf, value);
2454 break;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002455 default:
2456 abort();
2457 }
Peter Maydell5c9eb022015-04-26 16:49:24 +01002458 return address_space_write(subpage->as, addr + subpage->base,
2459 attrs, buf, len);
blueswir1db7b5422007-05-26 17:36:03 +00002460}
2461
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002462static bool subpage_accepts(void *opaque, hwaddr addr,
Amos Kong016e9d62013-09-27 09:25:38 +08002463 unsigned len, bool is_write)
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002464{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002465 subpage_t *subpage = opaque;
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002466#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002467 printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n",
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002468 __func__, subpage, is_write ? 'w' : 'r', len, addr);
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002469#endif
2470
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002471 return address_space_access_valid(subpage->as, addr + subpage->base,
Amos Kong016e9d62013-09-27 09:25:38 +08002472 len, is_write);
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002473}
2474
Avi Kivity70c68e42012-01-02 12:32:48 +02002475static const MemoryRegionOps subpage_ops = {
Peter Maydellf25a49e2015-04-26 16:49:24 +01002476 .read_with_attrs = subpage_read,
2477 .write_with_attrs = subpage_write,
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002478 .impl.min_access_size = 1,
2479 .impl.max_access_size = 8,
2480 .valid.min_access_size = 1,
2481 .valid.max_access_size = 8,
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002482 .valid.accepts = subpage_accepts,
Avi Kivity70c68e42012-01-02 12:32:48 +02002483 .endianness = DEVICE_NATIVE_ENDIAN,
blueswir1db7b5422007-05-26 17:36:03 +00002484};
2485
Anthony Liguoric227f092009-10-01 16:12:16 -05002486static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
Avi Kivity5312bd82012-02-12 18:32:55 +02002487 uint16_t section)
blueswir1db7b5422007-05-26 17:36:03 +00002488{
2489 int idx, eidx;
2490
2491 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
2492 return -1;
2493 idx = SUBPAGE_IDX(start);
2494 eidx = SUBPAGE_IDX(end);
2495#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002496 printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n",
2497 __func__, mmio, start, end, idx, eidx, section);
blueswir1db7b5422007-05-26 17:36:03 +00002498#endif
blueswir1db7b5422007-05-26 17:36:03 +00002499 for (; idx <= eidx; idx++) {
Avi Kivity5312bd82012-02-12 18:32:55 +02002500 mmio->sub_section[idx] = section;
blueswir1db7b5422007-05-26 17:36:03 +00002501 }
2502
2503 return 0;
2504}
2505
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002506static subpage_t *subpage_init(AddressSpace *as, hwaddr base)
blueswir1db7b5422007-05-26 17:36:03 +00002507{
Anthony Liguoric227f092009-10-01 16:12:16 -05002508 subpage_t *mmio;
blueswir1db7b5422007-05-26 17:36:03 +00002509
Vijaya Kumar K2615fab2016-10-24 16:26:49 +01002510 mmio = g_malloc0(sizeof(subpage_t) + TARGET_PAGE_SIZE * sizeof(uint16_t));
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002511 mmio->as = as;
aliguori1eec6142009-02-05 22:06:18 +00002512 mmio->base = base;
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002513 memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio,
Peter Crosthwaiteb4fefef2014-06-05 23:15:52 -07002514 NULL, TARGET_PAGE_SIZE);
Avi Kivityb3b00c72012-01-02 13:20:11 +02002515 mmio->iomem.subpage = true;
blueswir1db7b5422007-05-26 17:36:03 +00002516#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002517 printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__,
2518 mmio, base, TARGET_PAGE_SIZE);
blueswir1db7b5422007-05-26 17:36:03 +00002519#endif
Liu Ping Fanb41aac42013-05-29 11:09:17 +02002520 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED);
blueswir1db7b5422007-05-26 17:36:03 +00002521
2522 return mmio;
2523}
2524
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002525static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as,
2526 MemoryRegion *mr)
Avi Kivity5312bd82012-02-12 18:32:55 +02002527{
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002528 assert(as);
Avi Kivity5312bd82012-02-12 18:32:55 +02002529 MemoryRegionSection section = {
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002530 .address_space = as,
Avi Kivity5312bd82012-02-12 18:32:55 +02002531 .mr = mr,
2532 .offset_within_address_space = 0,
2533 .offset_within_region = 0,
Paolo Bonzini052e87b2013-05-27 10:08:27 +02002534 .size = int128_2_64(),
Avi Kivity5312bd82012-02-12 18:32:55 +02002535 };
2536
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002537 return phys_section_add(map, &section);
Avi Kivity5312bd82012-02-12 18:32:55 +02002538}
2539
Peter Maydella54c87b2016-01-21 14:15:05 +00002540MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index, MemTxAttrs attrs)
Avi Kivityaa102232012-03-08 17:06:55 +02002541{
Peter Maydella54c87b2016-01-21 14:15:05 +00002542 int asidx = cpu_asidx_from_attrs(cpu, attrs);
2543 CPUAddressSpace *cpuas = &cpu->cpu_ases[asidx];
Peter Maydell32857f42015-10-01 15:29:50 +01002544 AddressSpaceDispatch *d = atomic_rcu_read(&cpuas->memory_dispatch);
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002545 MemoryRegionSection *sections = d->map.sections;
Paolo Bonzini9d82b5a2013-08-16 08:26:30 +02002546
2547 return sections[index & ~TARGET_PAGE_MASK].mr;
Avi Kivityaa102232012-03-08 17:06:55 +02002548}
2549
Avi Kivitye9179ce2009-06-14 11:38:52 +03002550static void io_mem_init(void)
2551{
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002552 memory_region_init_io(&io_mem_rom, NULL, &unassigned_mem_ops, NULL, NULL, UINT64_MAX);
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002553 memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002554 NULL, UINT64_MAX);
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002555
2556 /* io_mem_notdirty calls tb_invalidate_phys_page_fast,
2557 * which can be called without the iothread mutex.
2558 */
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002559 memory_region_init_io(&io_mem_notdirty, NULL, &notdirty_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002560 NULL, UINT64_MAX);
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002561 memory_region_clear_global_locking(&io_mem_notdirty);
2562
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002563 memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002564 NULL, UINT64_MAX);
Avi Kivitye9179ce2009-06-14 11:38:52 +03002565}
2566
Avi Kivityac1970f2012-10-03 16:22:53 +02002567static void mem_begin(MemoryListener *listener)
2568{
Paolo Bonzini89ae3372013-06-02 10:39:07 +02002569 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002570 AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1);
2571 uint16_t n;
2572
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002573 n = dummy_section(&d->map, as, &io_mem_unassigned);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002574 assert(n == PHYS_SECTION_UNASSIGNED);
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002575 n = dummy_section(&d->map, as, &io_mem_notdirty);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002576 assert(n == PHYS_SECTION_NOTDIRTY);
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002577 n = dummy_section(&d->map, as, &io_mem_rom);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002578 assert(n == PHYS_SECTION_ROM);
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002579 n = dummy_section(&d->map, as, &io_mem_watch);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002580 assert(n == PHYS_SECTION_WATCH);
Paolo Bonzini00752702013-05-29 12:13:54 +02002581
Michael S. Tsirkin9736e552013-11-11 14:42:43 +02002582 d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 };
Paolo Bonzini00752702013-05-29 12:13:54 +02002583 d->as = as;
2584 as->next_dispatch = d;
2585}
2586
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002587static void address_space_dispatch_free(AddressSpaceDispatch *d)
2588{
2589 phys_sections_free(&d->map);
2590 g_free(d);
2591}
2592
Paolo Bonzini00752702013-05-29 12:13:54 +02002593static void mem_commit(MemoryListener *listener)
2594{
2595 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
Paolo Bonzini0475d942013-05-29 12:28:21 +02002596 AddressSpaceDispatch *cur = as->dispatch;
2597 AddressSpaceDispatch *next = as->next_dispatch;
Avi Kivityac1970f2012-10-03 16:22:53 +02002598
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002599 phys_page_compact_all(next, next->map.nodes_nb);
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +02002600
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002601 atomic_rcu_set(&as->dispatch, next);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002602 if (cur) {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002603 call_rcu(cur, address_space_dispatch_free, rcu);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002604 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02002605}
2606
Avi Kivity1d711482012-10-02 18:54:45 +02002607static void tcg_commit(MemoryListener *listener)
Avi Kivity50c1e142012-02-08 21:36:02 +02002608{
Peter Maydell32857f42015-10-01 15:29:50 +01002609 CPUAddressSpace *cpuas;
2610 AddressSpaceDispatch *d;
Avi Kivity117712c2012-02-12 21:23:17 +02002611
2612 /* since each CPU stores ram addresses in its TLB cache, we must
2613 reset the modified entries */
Peter Maydell32857f42015-10-01 15:29:50 +01002614 cpuas = container_of(listener, CPUAddressSpace, tcg_as_listener);
2615 cpu_reloading_memory_map();
2616 /* The CPU and TLB are protected by the iothread lock.
2617 * We reload the dispatch pointer now because cpu_reloading_memory_map()
2618 * may have split the RCU critical section.
2619 */
2620 d = atomic_rcu_read(&cpuas->as->dispatch);
Alex Bennéef35e44e2016-10-21 16:34:18 +01002621 atomic_rcu_set(&cpuas->memory_dispatch, d);
Alex Bennéed10eb082016-11-14 14:17:28 +00002622 tlb_flush(cpuas->cpu);
Avi Kivity50c1e142012-02-08 21:36:02 +02002623}
2624
Avi Kivityac1970f2012-10-03 16:22:53 +02002625void address_space_init_dispatch(AddressSpace *as)
2626{
Paolo Bonzini00752702013-05-29 12:13:54 +02002627 as->dispatch = NULL;
Paolo Bonzini89ae3372013-06-02 10:39:07 +02002628 as->dispatch_listener = (MemoryListener) {
Avi Kivityac1970f2012-10-03 16:22:53 +02002629 .begin = mem_begin,
Paolo Bonzini00752702013-05-29 12:13:54 +02002630 .commit = mem_commit,
Avi Kivityac1970f2012-10-03 16:22:53 +02002631 .region_add = mem_add,
2632 .region_nop = mem_add,
2633 .priority = 0,
2634 };
Paolo Bonzini89ae3372013-06-02 10:39:07 +02002635 memory_listener_register(&as->dispatch_listener, as);
Avi Kivityac1970f2012-10-03 16:22:53 +02002636}
2637
Paolo Bonzini6e48e8f2015-02-10 10:25:44 -07002638void address_space_unregister(AddressSpace *as)
2639{
2640 memory_listener_unregister(&as->dispatch_listener);
2641}
2642
Avi Kivity83f3c252012-10-07 12:59:55 +02002643void address_space_destroy_dispatch(AddressSpace *as)
2644{
2645 AddressSpaceDispatch *d = as->dispatch;
2646
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002647 atomic_rcu_set(&as->dispatch, NULL);
2648 if (d) {
2649 call_rcu(d, address_space_dispatch_free, rcu);
2650 }
Avi Kivity83f3c252012-10-07 12:59:55 +02002651}
2652
Avi Kivity62152b82011-07-26 14:26:14 +03002653static void memory_map_init(void)
2654{
Anthony Liguori7267c092011-08-20 22:09:37 -05002655 system_memory = g_malloc(sizeof(*system_memory));
Paolo Bonzini03f49952013-11-07 17:14:36 +01002656
Paolo Bonzini57271d62013-11-07 17:14:37 +01002657 memory_region_init(system_memory, NULL, "system", UINT64_MAX);
Alexey Kardashevskiy7dca8042013-04-29 16:25:51 +00002658 address_space_init(&address_space_memory, system_memory, "memory");
Avi Kivity309cb472011-08-08 16:09:03 +03002659
Anthony Liguori7267c092011-08-20 22:09:37 -05002660 system_io = g_malloc(sizeof(*system_io));
Jan Kiszka3bb28b72013-09-02 18:43:30 +02002661 memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io",
2662 65536);
Alexey Kardashevskiy7dca8042013-04-29 16:25:51 +00002663 address_space_init(&address_space_io, system_io, "I/O");
Avi Kivity62152b82011-07-26 14:26:14 +03002664}
2665
2666MemoryRegion *get_system_memory(void)
2667{
2668 return system_memory;
2669}
2670
Avi Kivity309cb472011-08-08 16:09:03 +03002671MemoryRegion *get_system_io(void)
2672{
2673 return system_io;
2674}
2675
pbrooke2eef172008-06-08 01:09:01 +00002676#endif /* !defined(CONFIG_USER_ONLY) */
2677
bellard13eb76e2004-01-24 15:23:36 +00002678/* physical memory access (slow version, mainly for debug) */
2679#if defined(CONFIG_USER_ONLY)
Andreas Färberf17ec442013-06-29 19:40:58 +02002680int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
Paul Brooka68fe892010-03-01 00:08:59 +00002681 uint8_t *buf, int len, int is_write)
bellard13eb76e2004-01-24 15:23:36 +00002682{
2683 int l, flags;
2684 target_ulong page;
pbrook53a59602006-03-25 19:31:22 +00002685 void * p;
bellard13eb76e2004-01-24 15:23:36 +00002686
2687 while (len > 0) {
2688 page = addr & TARGET_PAGE_MASK;
2689 l = (page + TARGET_PAGE_SIZE) - addr;
2690 if (l > len)
2691 l = len;
2692 flags = page_get_flags(page);
2693 if (!(flags & PAGE_VALID))
Paul Brooka68fe892010-03-01 00:08:59 +00002694 return -1;
bellard13eb76e2004-01-24 15:23:36 +00002695 if (is_write) {
2696 if (!(flags & PAGE_WRITE))
Paul Brooka68fe892010-03-01 00:08:59 +00002697 return -1;
bellard579a97f2007-11-11 14:26:47 +00002698 /* XXX: this code should not depend on lock_user */
aurel3272fb7da2008-04-27 23:53:45 +00002699 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
Paul Brooka68fe892010-03-01 00:08:59 +00002700 return -1;
aurel3272fb7da2008-04-27 23:53:45 +00002701 memcpy(p, buf, l);
2702 unlock_user(p, addr, l);
bellard13eb76e2004-01-24 15:23:36 +00002703 } else {
2704 if (!(flags & PAGE_READ))
Paul Brooka68fe892010-03-01 00:08:59 +00002705 return -1;
bellard579a97f2007-11-11 14:26:47 +00002706 /* XXX: this code should not depend on lock_user */
aurel3272fb7da2008-04-27 23:53:45 +00002707 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
Paul Brooka68fe892010-03-01 00:08:59 +00002708 return -1;
aurel3272fb7da2008-04-27 23:53:45 +00002709 memcpy(buf, p, l);
aurel325b257572008-04-28 08:54:59 +00002710 unlock_user(p, addr, 0);
bellard13eb76e2004-01-24 15:23:36 +00002711 }
2712 len -= l;
2713 buf += l;
2714 addr += l;
2715 }
Paul Brooka68fe892010-03-01 00:08:59 +00002716 return 0;
bellard13eb76e2004-01-24 15:23:36 +00002717}
bellard8df1cd02005-01-28 22:37:22 +00002718
bellard13eb76e2004-01-24 15:23:36 +00002719#else
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002720
Paolo Bonzini845b6212015-03-23 11:45:53 +01002721static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr,
Avi Kivitya8170e52012-10-23 12:30:10 +02002722 hwaddr length)
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002723{
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002724 uint8_t dirty_log_mask = memory_region_get_dirty_log_mask(mr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002725 addr += memory_region_get_ram_addr(mr);
2726
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002727 /* No early return if dirty_log_mask is or becomes 0, because
2728 * cpu_physical_memory_set_dirty_range will still call
2729 * xen_modified_memory.
2730 */
2731 if (dirty_log_mask) {
2732 dirty_log_mask =
2733 cpu_physical_memory_range_includes_clean(addr, length, dirty_log_mask);
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002734 }
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002735 if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) {
Alex Bennéeba051fb2016-10-27 16:10:16 +01002736 tb_lock();
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002737 tb_invalidate_phys_range(addr, addr + length);
Alex Bennéeba051fb2016-10-27 16:10:16 +01002738 tb_unlock();
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002739 dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE);
2740 }
2741 cpu_physical_memory_set_dirty_range(addr, length, dirty_log_mask);
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002742}
2743
Richard Henderson23326162013-07-08 14:55:59 -07002744static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr)
Paolo Bonzini82f25632013-05-24 11:59:43 +02002745{
Paolo Bonzinie1622f42013-07-17 13:17:41 +02002746 unsigned access_size_max = mr->ops->valid.max_access_size;
Richard Henderson23326162013-07-08 14:55:59 -07002747
2748 /* Regions are assumed to support 1-4 byte accesses unless
2749 otherwise specified. */
Richard Henderson23326162013-07-08 14:55:59 -07002750 if (access_size_max == 0) {
2751 access_size_max = 4;
Paolo Bonzini82f25632013-05-24 11:59:43 +02002752 }
Richard Henderson23326162013-07-08 14:55:59 -07002753
2754 /* Bound the maximum access by the alignment of the address. */
2755 if (!mr->ops->impl.unaligned) {
2756 unsigned align_size_max = addr & -addr;
2757 if (align_size_max != 0 && align_size_max < access_size_max) {
2758 access_size_max = align_size_max;
2759 }
2760 }
2761
2762 /* Don't attempt accesses larger than the maximum. */
2763 if (l > access_size_max) {
2764 l = access_size_max;
2765 }
Peter Maydell6554f5c2015-07-24 13:33:10 +01002766 l = pow2floor(l);
Richard Henderson23326162013-07-08 14:55:59 -07002767
2768 return l;
Paolo Bonzini82f25632013-05-24 11:59:43 +02002769}
2770
Jan Kiszka4840f102015-06-18 18:47:22 +02002771static bool prepare_mmio_access(MemoryRegion *mr)
Paolo Bonzini125b3802015-06-18 18:47:21 +02002772{
Jan Kiszka4840f102015-06-18 18:47:22 +02002773 bool unlocked = !qemu_mutex_iothread_locked();
2774 bool release_lock = false;
2775
2776 if (unlocked && mr->global_locking) {
2777 qemu_mutex_lock_iothread();
2778 unlocked = false;
2779 release_lock = true;
Paolo Bonzini125b3802015-06-18 18:47:21 +02002780 }
Jan Kiszka4840f102015-06-18 18:47:22 +02002781 if (mr->flush_coalesced_mmio) {
2782 if (unlocked) {
2783 qemu_mutex_lock_iothread();
2784 }
2785 qemu_flush_coalesced_mmio_buffer();
2786 if (unlocked) {
2787 qemu_mutex_unlock_iothread();
2788 }
2789 }
2790
2791 return release_lock;
Paolo Bonzini125b3802015-06-18 18:47:21 +02002792}
2793
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002794/* Called within RCU critical section. */
2795static MemTxResult address_space_write_continue(AddressSpace *as, hwaddr addr,
2796 MemTxAttrs attrs,
2797 const uint8_t *buf,
2798 int len, hwaddr addr1,
2799 hwaddr l, MemoryRegion *mr)
bellard13eb76e2004-01-24 15:23:36 +00002800{
bellard13eb76e2004-01-24 15:23:36 +00002801 uint8_t *ptr;
Paolo Bonzini791af8c2013-05-24 16:10:39 +02002802 uint64_t val;
Peter Maydell3b643492015-04-26 16:49:23 +01002803 MemTxResult result = MEMTX_OK;
Jan Kiszka4840f102015-06-18 18:47:22 +02002804 bool release_lock = false;
ths3b46e622007-09-17 08:09:54 +00002805
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002806 for (;;) {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002807 if (!memory_access_is_direct(mr, true)) {
2808 release_lock |= prepare_mmio_access(mr);
2809 l = memory_access_size(mr, l, addr1);
2810 /* XXX: could force current_cpu to NULL to avoid
2811 potential bugs */
2812 switch (l) {
2813 case 8:
2814 /* 64 bit write access */
2815 val = ldq_p(buf);
2816 result |= memory_region_dispatch_write(mr, addr1, val, 8,
2817 attrs);
2818 break;
2819 case 4:
2820 /* 32 bit write access */
Ladi Prosek6da67de2017-01-26 15:22:37 +01002821 val = (uint32_t)ldl_p(buf);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002822 result |= memory_region_dispatch_write(mr, addr1, val, 4,
2823 attrs);
2824 break;
2825 case 2:
2826 /* 16 bit write access */
2827 val = lduw_p(buf);
2828 result |= memory_region_dispatch_write(mr, addr1, val, 2,
2829 attrs);
2830 break;
2831 case 1:
2832 /* 8 bit write access */
2833 val = ldub_p(buf);
2834 result |= memory_region_dispatch_write(mr, addr1, val, 1,
2835 attrs);
2836 break;
2837 default:
2838 abort();
bellard13eb76e2004-01-24 15:23:36 +00002839 }
2840 } else {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002841 /* RAM case */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002842 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002843 memcpy(ptr, buf, l);
2844 invalidate_and_set_dirty(mr, addr1, l);
bellard13eb76e2004-01-24 15:23:36 +00002845 }
Jan Kiszka4840f102015-06-18 18:47:22 +02002846
2847 if (release_lock) {
2848 qemu_mutex_unlock_iothread();
2849 release_lock = false;
2850 }
2851
bellard13eb76e2004-01-24 15:23:36 +00002852 len -= l;
2853 buf += l;
2854 addr += l;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002855
2856 if (!len) {
2857 break;
2858 }
2859
2860 l = len;
2861 mr = address_space_translate(as, addr, &addr1, &l, true);
bellard13eb76e2004-01-24 15:23:36 +00002862 }
Paolo Bonzinifd8aaa72013-05-21 09:56:55 +02002863
Peter Maydell3b643492015-04-26 16:49:23 +01002864 return result;
bellard13eb76e2004-01-24 15:23:36 +00002865}
bellard8df1cd02005-01-28 22:37:22 +00002866
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002867MemTxResult address_space_write(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
2868 const uint8_t *buf, int len)
Avi Kivityac1970f2012-10-03 16:22:53 +02002869{
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002870 hwaddr l;
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002871 hwaddr addr1;
2872 MemoryRegion *mr;
2873 MemTxResult result = MEMTX_OK;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002874
2875 if (len > 0) {
2876 rcu_read_lock();
2877 l = len;
2878 mr = address_space_translate(as, addr, &addr1, &l, true);
2879 result = address_space_write_continue(as, addr, attrs, buf, len,
2880 addr1, l, mr);
2881 rcu_read_unlock();
2882 }
2883
2884 return result;
2885}
2886
2887/* Called within RCU critical section. */
2888MemTxResult address_space_read_continue(AddressSpace *as, hwaddr addr,
2889 MemTxAttrs attrs, uint8_t *buf,
2890 int len, hwaddr addr1, hwaddr l,
2891 MemoryRegion *mr)
2892{
2893 uint8_t *ptr;
2894 uint64_t val;
2895 MemTxResult result = MEMTX_OK;
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002896 bool release_lock = false;
2897
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002898 for (;;) {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002899 if (!memory_access_is_direct(mr, false)) {
2900 /* I/O case */
2901 release_lock |= prepare_mmio_access(mr);
2902 l = memory_access_size(mr, l, addr1);
2903 switch (l) {
2904 case 8:
2905 /* 64 bit read access */
2906 result |= memory_region_dispatch_read(mr, addr1, &val, 8,
2907 attrs);
2908 stq_p(buf, val);
2909 break;
2910 case 4:
2911 /* 32 bit read access */
2912 result |= memory_region_dispatch_read(mr, addr1, &val, 4,
2913 attrs);
2914 stl_p(buf, val);
2915 break;
2916 case 2:
2917 /* 16 bit read access */
2918 result |= memory_region_dispatch_read(mr, addr1, &val, 2,
2919 attrs);
2920 stw_p(buf, val);
2921 break;
2922 case 1:
2923 /* 8 bit read access */
2924 result |= memory_region_dispatch_read(mr, addr1, &val, 1,
2925 attrs);
2926 stb_p(buf, val);
2927 break;
2928 default:
2929 abort();
2930 }
2931 } else {
2932 /* RAM case */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002933 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002934 memcpy(buf, ptr, l);
2935 }
2936
2937 if (release_lock) {
2938 qemu_mutex_unlock_iothread();
2939 release_lock = false;
2940 }
2941
2942 len -= l;
2943 buf += l;
2944 addr += l;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002945
2946 if (!len) {
2947 break;
2948 }
2949
2950 l = len;
2951 mr = address_space_translate(as, addr, &addr1, &l, false);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002952 }
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002953
2954 return result;
2955}
2956
Paolo Bonzini3cc8f882015-12-09 10:34:13 +01002957MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
2958 MemTxAttrs attrs, uint8_t *buf, int len)
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002959{
2960 hwaddr l;
2961 hwaddr addr1;
2962 MemoryRegion *mr;
2963 MemTxResult result = MEMTX_OK;
2964
2965 if (len > 0) {
2966 rcu_read_lock();
2967 l = len;
2968 mr = address_space_translate(as, addr, &addr1, &l, false);
2969 result = address_space_read_continue(as, addr, attrs, buf, len,
2970 addr1, l, mr);
2971 rcu_read_unlock();
2972 }
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002973
2974 return result;
Avi Kivityac1970f2012-10-03 16:22:53 +02002975}
2976
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002977MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
2978 uint8_t *buf, int len, bool is_write)
2979{
2980 if (is_write) {
2981 return address_space_write(as, addr, attrs, (uint8_t *)buf, len);
2982 } else {
2983 return address_space_read(as, addr, attrs, (uint8_t *)buf, len);
2984 }
2985}
Avi Kivityac1970f2012-10-03 16:22:53 +02002986
Avi Kivitya8170e52012-10-23 12:30:10 +02002987void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf,
Avi Kivityac1970f2012-10-03 16:22:53 +02002988 int len, int is_write)
2989{
Peter Maydell5c9eb022015-04-26 16:49:24 +01002990 address_space_rw(&address_space_memory, addr, MEMTXATTRS_UNSPECIFIED,
2991 buf, len, is_write);
Avi Kivityac1970f2012-10-03 16:22:53 +02002992}
2993
Alexander Graf582b55a2013-12-11 14:17:44 +01002994enum write_rom_type {
2995 WRITE_DATA,
2996 FLUSH_CACHE,
2997};
2998
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10002999static inline void cpu_physical_memory_write_rom_internal(AddressSpace *as,
Alexander Graf582b55a2013-12-11 14:17:44 +01003000 hwaddr addr, const uint8_t *buf, int len, enum write_rom_type type)
bellardd0ecd2a2006-04-23 17:14:48 +00003001{
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003002 hwaddr l;
bellardd0ecd2a2006-04-23 17:14:48 +00003003 uint8_t *ptr;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003004 hwaddr addr1;
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003005 MemoryRegion *mr;
ths3b46e622007-09-17 08:09:54 +00003006
Paolo Bonzini41063e12015-03-18 14:21:43 +01003007 rcu_read_lock();
bellardd0ecd2a2006-04-23 17:14:48 +00003008 while (len > 0) {
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003009 l = len;
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003010 mr = address_space_translate(as, addr, &addr1, &l, true);
ths3b46e622007-09-17 08:09:54 +00003011
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003012 if (!(memory_region_is_ram(mr) ||
3013 memory_region_is_romd(mr))) {
Paolo Bonzinib242e0e2015-07-04 00:24:51 +02003014 l = memory_access_size(mr, l, addr1);
bellardd0ecd2a2006-04-23 17:14:48 +00003015 } else {
bellardd0ecd2a2006-04-23 17:14:48 +00003016 /* ROM/RAM case */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01003017 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
Alexander Graf582b55a2013-12-11 14:17:44 +01003018 switch (type) {
3019 case WRITE_DATA:
3020 memcpy(ptr, buf, l);
Paolo Bonzini845b6212015-03-23 11:45:53 +01003021 invalidate_and_set_dirty(mr, addr1, l);
Alexander Graf582b55a2013-12-11 14:17:44 +01003022 break;
3023 case FLUSH_CACHE:
3024 flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l);
3025 break;
3026 }
bellardd0ecd2a2006-04-23 17:14:48 +00003027 }
3028 len -= l;
3029 buf += l;
3030 addr += l;
3031 }
Paolo Bonzini41063e12015-03-18 14:21:43 +01003032 rcu_read_unlock();
bellardd0ecd2a2006-04-23 17:14:48 +00003033}
3034
Alexander Graf582b55a2013-12-11 14:17:44 +01003035/* used for ROM loading : can write in RAM and ROM */
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003036void cpu_physical_memory_write_rom(AddressSpace *as, hwaddr addr,
Alexander Graf582b55a2013-12-11 14:17:44 +01003037 const uint8_t *buf, int len)
3038{
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003039 cpu_physical_memory_write_rom_internal(as, addr, buf, len, WRITE_DATA);
Alexander Graf582b55a2013-12-11 14:17:44 +01003040}
3041
3042void cpu_flush_icache_range(hwaddr start, int len)
3043{
3044 /*
3045 * This function should do the same thing as an icache flush that was
3046 * triggered from within the guest. For TCG we are always cache coherent,
3047 * so there is no need to flush anything. For KVM / Xen we need to flush
3048 * the host's instruction cache at least.
3049 */
3050 if (tcg_enabled()) {
3051 return;
3052 }
3053
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003054 cpu_physical_memory_write_rom_internal(&address_space_memory,
3055 start, NULL, len, FLUSH_CACHE);
Alexander Graf582b55a2013-12-11 14:17:44 +01003056}
3057
aliguori6d16c2f2009-01-22 16:59:11 +00003058typedef struct {
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003059 MemoryRegion *mr;
aliguori6d16c2f2009-01-22 16:59:11 +00003060 void *buffer;
Avi Kivitya8170e52012-10-23 12:30:10 +02003061 hwaddr addr;
3062 hwaddr len;
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003063 bool in_use;
aliguori6d16c2f2009-01-22 16:59:11 +00003064} BounceBuffer;
3065
3066static BounceBuffer bounce;
3067
aliguoriba223c22009-01-22 16:59:16 +00003068typedef struct MapClient {
Fam Zhenge95205e2015-03-16 17:03:37 +08003069 QEMUBH *bh;
Blue Swirl72cf2d42009-09-12 07:36:22 +00003070 QLIST_ENTRY(MapClient) link;
aliguoriba223c22009-01-22 16:59:16 +00003071} MapClient;
3072
Fam Zheng38e047b2015-03-16 17:03:35 +08003073QemuMutex map_client_list_lock;
Blue Swirl72cf2d42009-09-12 07:36:22 +00003074static QLIST_HEAD(map_client_list, MapClient) map_client_list
3075 = QLIST_HEAD_INITIALIZER(map_client_list);
aliguoriba223c22009-01-22 16:59:16 +00003076
Fam Zhenge95205e2015-03-16 17:03:37 +08003077static void cpu_unregister_map_client_do(MapClient *client)
aliguoriba223c22009-01-22 16:59:16 +00003078{
Blue Swirl72cf2d42009-09-12 07:36:22 +00003079 QLIST_REMOVE(client, link);
Anthony Liguori7267c092011-08-20 22:09:37 -05003080 g_free(client);
aliguoriba223c22009-01-22 16:59:16 +00003081}
3082
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003083static void cpu_notify_map_clients_locked(void)
aliguoriba223c22009-01-22 16:59:16 +00003084{
3085 MapClient *client;
3086
Blue Swirl72cf2d42009-09-12 07:36:22 +00003087 while (!QLIST_EMPTY(&map_client_list)) {
3088 client = QLIST_FIRST(&map_client_list);
Fam Zhenge95205e2015-03-16 17:03:37 +08003089 qemu_bh_schedule(client->bh);
3090 cpu_unregister_map_client_do(client);
aliguoriba223c22009-01-22 16:59:16 +00003091 }
3092}
3093
Fam Zhenge95205e2015-03-16 17:03:37 +08003094void cpu_register_map_client(QEMUBH *bh)
bellardd0ecd2a2006-04-23 17:14:48 +00003095{
3096 MapClient *client = g_malloc(sizeof(*client));
3097
Fam Zheng38e047b2015-03-16 17:03:35 +08003098 qemu_mutex_lock(&map_client_list_lock);
Fam Zhenge95205e2015-03-16 17:03:37 +08003099 client->bh = bh;
bellardd0ecd2a2006-04-23 17:14:48 +00003100 QLIST_INSERT_HEAD(&map_client_list, client, link);
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003101 if (!atomic_read(&bounce.in_use)) {
3102 cpu_notify_map_clients_locked();
3103 }
Fam Zheng38e047b2015-03-16 17:03:35 +08003104 qemu_mutex_unlock(&map_client_list_lock);
bellardd0ecd2a2006-04-23 17:14:48 +00003105}
3106
Fam Zheng38e047b2015-03-16 17:03:35 +08003107void cpu_exec_init_all(void)
3108{
3109 qemu_mutex_init(&ram_list.mutex);
Peter Maydell20bccb82016-10-24 16:26:49 +01003110 /* The data structures we set up here depend on knowing the page size,
3111 * so no more changes can be made after this point.
3112 * In an ideal world, nothing we did before we had finished the
3113 * machine setup would care about the target page size, and we could
3114 * do this much later, rather than requiring board models to state
3115 * up front what their requirements are.
3116 */
3117 finalize_target_page_bits();
Fam Zheng38e047b2015-03-16 17:03:35 +08003118 io_mem_init();
Paolo Bonzini680a4782015-11-02 09:23:52 +01003119 memory_map_init();
Fam Zheng38e047b2015-03-16 17:03:35 +08003120 qemu_mutex_init(&map_client_list_lock);
3121}
3122
Fam Zhenge95205e2015-03-16 17:03:37 +08003123void cpu_unregister_map_client(QEMUBH *bh)
bellardd0ecd2a2006-04-23 17:14:48 +00003124{
Fam Zhenge95205e2015-03-16 17:03:37 +08003125 MapClient *client;
bellardd0ecd2a2006-04-23 17:14:48 +00003126
Fam Zhenge95205e2015-03-16 17:03:37 +08003127 qemu_mutex_lock(&map_client_list_lock);
3128 QLIST_FOREACH(client, &map_client_list, link) {
3129 if (client->bh == bh) {
3130 cpu_unregister_map_client_do(client);
3131 break;
3132 }
3133 }
3134 qemu_mutex_unlock(&map_client_list_lock);
bellardd0ecd2a2006-04-23 17:14:48 +00003135}
3136
3137static void cpu_notify_map_clients(void)
3138{
Fam Zheng38e047b2015-03-16 17:03:35 +08003139 qemu_mutex_lock(&map_client_list_lock);
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003140 cpu_notify_map_clients_locked();
Fam Zheng38e047b2015-03-16 17:03:35 +08003141 qemu_mutex_unlock(&map_client_list_lock);
aliguori6d16c2f2009-01-22 16:59:11 +00003142}
3143
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003144bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write)
3145{
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003146 MemoryRegion *mr;
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003147 hwaddr l, xlat;
3148
Paolo Bonzini41063e12015-03-18 14:21:43 +01003149 rcu_read_lock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003150 while (len > 0) {
3151 l = len;
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003152 mr = address_space_translate(as, addr, &xlat, &l, is_write);
3153 if (!memory_access_is_direct(mr, is_write)) {
3154 l = memory_access_size(mr, l, addr);
3155 if (!memory_region_access_valid(mr, xlat, l, is_write)) {
Roman Kapl5ad4a2b2017-01-09 12:09:21 +01003156 rcu_read_unlock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003157 return false;
3158 }
3159 }
3160
3161 len -= l;
3162 addr += l;
3163 }
Paolo Bonzini41063e12015-03-18 14:21:43 +01003164 rcu_read_unlock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003165 return true;
3166}
3167
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003168static hwaddr
3169address_space_extend_translation(AddressSpace *as, hwaddr addr, hwaddr target_len,
3170 MemoryRegion *mr, hwaddr base, hwaddr len,
3171 bool is_write)
3172{
3173 hwaddr done = 0;
3174 hwaddr xlat;
3175 MemoryRegion *this_mr;
3176
3177 for (;;) {
3178 target_len -= len;
3179 addr += len;
3180 done += len;
3181 if (target_len == 0) {
3182 return done;
3183 }
3184
3185 len = target_len;
3186 this_mr = address_space_translate(as, addr, &xlat, &len, is_write);
3187 if (this_mr != mr || xlat != base + done) {
3188 return done;
3189 }
3190 }
3191}
3192
aliguori6d16c2f2009-01-22 16:59:11 +00003193/* Map a physical memory region into a host virtual address.
3194 * May map a subset of the requested range, given by and returned in *plen.
3195 * May return NULL if resources needed to perform the mapping are exhausted.
3196 * Use only for reads OR writes - not for read-modify-write operations.
aliguoriba223c22009-01-22 16:59:16 +00003197 * Use cpu_register_map_client() to know when retrying the map operation is
3198 * likely to succeed.
aliguori6d16c2f2009-01-22 16:59:11 +00003199 */
Avi Kivityac1970f2012-10-03 16:22:53 +02003200void *address_space_map(AddressSpace *as,
Avi Kivitya8170e52012-10-23 12:30:10 +02003201 hwaddr addr,
3202 hwaddr *plen,
Avi Kivityac1970f2012-10-03 16:22:53 +02003203 bool is_write)
aliguori6d16c2f2009-01-22 16:59:11 +00003204{
Avi Kivitya8170e52012-10-23 12:30:10 +02003205 hwaddr len = *plen;
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003206 hwaddr l, xlat;
3207 MemoryRegion *mr;
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01003208 void *ptr;
aliguori6d16c2f2009-01-22 16:59:11 +00003209
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003210 if (len == 0) {
3211 return NULL;
3212 }
aliguori6d16c2f2009-01-22 16:59:11 +00003213
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003214 l = len;
Paolo Bonzini41063e12015-03-18 14:21:43 +01003215 rcu_read_lock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003216 mr = address_space_translate(as, addr, &xlat, &l, is_write);
Paolo Bonzini41063e12015-03-18 14:21:43 +01003217
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003218 if (!memory_access_is_direct(mr, is_write)) {
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003219 if (atomic_xchg(&bounce.in_use, true)) {
Paolo Bonzini41063e12015-03-18 14:21:43 +01003220 rcu_read_unlock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003221 return NULL;
aliguori6d16c2f2009-01-22 16:59:11 +00003222 }
Kevin Wolfe85d9db2013-07-22 14:30:23 +02003223 /* Avoid unbounded allocations */
3224 l = MIN(l, TARGET_PAGE_SIZE);
3225 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l);
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003226 bounce.addr = addr;
3227 bounce.len = l;
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003228
3229 memory_region_ref(mr);
3230 bounce.mr = mr;
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003231 if (!is_write) {
Peter Maydell5c9eb022015-04-26 16:49:24 +01003232 address_space_read(as, addr, MEMTXATTRS_UNSPECIFIED,
3233 bounce.buffer, l);
Stefano Stabellini8ab934f2011-06-27 18:26:06 +01003234 }
aliguori6d16c2f2009-01-22 16:59:11 +00003235
Paolo Bonzini41063e12015-03-18 14:21:43 +01003236 rcu_read_unlock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003237 *plen = l;
3238 return bounce.buffer;
3239 }
3240
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003241
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003242 memory_region_ref(mr);
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003243 *plen = address_space_extend_translation(as, addr, len, mr, xlat, l, is_write);
3244 ptr = qemu_ram_ptr_length(mr->ram_block, xlat, plen);
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01003245 rcu_read_unlock();
3246
3247 return ptr;
aliguori6d16c2f2009-01-22 16:59:11 +00003248}
3249
Avi Kivityac1970f2012-10-03 16:22:53 +02003250/* Unmaps a memory region previously mapped by address_space_map().
aliguori6d16c2f2009-01-22 16:59:11 +00003251 * Will also mark the memory as dirty if is_write == 1. access_len gives
3252 * the amount of memory that was actually read or written by the caller.
3253 */
Avi Kivitya8170e52012-10-23 12:30:10 +02003254void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
3255 int is_write, hwaddr access_len)
aliguori6d16c2f2009-01-22 16:59:11 +00003256{
3257 if (buffer != bounce.buffer) {
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003258 MemoryRegion *mr;
3259 ram_addr_t addr1;
3260
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01003261 mr = memory_region_from_host(buffer, &addr1);
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003262 assert(mr != NULL);
aliguori6d16c2f2009-01-22 16:59:11 +00003263 if (is_write) {
Paolo Bonzini845b6212015-03-23 11:45:53 +01003264 invalidate_and_set_dirty(mr, addr1, access_len);
aliguori6d16c2f2009-01-22 16:59:11 +00003265 }
Jan Kiszka868bb332011-06-21 22:59:09 +02003266 if (xen_enabled()) {
Jan Kiszkae41d7c62011-06-21 22:59:08 +02003267 xen_invalidate_map_cache_entry(buffer);
Anthony PERARD050a0dd2010-09-16 13:57:49 +01003268 }
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003269 memory_region_unref(mr);
aliguori6d16c2f2009-01-22 16:59:11 +00003270 return;
3271 }
3272 if (is_write) {
Peter Maydell5c9eb022015-04-26 16:49:24 +01003273 address_space_write(as, bounce.addr, MEMTXATTRS_UNSPECIFIED,
3274 bounce.buffer, access_len);
aliguori6d16c2f2009-01-22 16:59:11 +00003275 }
Herve Poussineauf8a83242010-01-24 21:23:56 +00003276 qemu_vfree(bounce.buffer);
aliguori6d16c2f2009-01-22 16:59:11 +00003277 bounce.buffer = NULL;
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003278 memory_region_unref(bounce.mr);
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003279 atomic_mb_set(&bounce.in_use, false);
aliguoriba223c22009-01-22 16:59:16 +00003280 cpu_notify_map_clients();
aliguori6d16c2f2009-01-22 16:59:11 +00003281}
bellardd0ecd2a2006-04-23 17:14:48 +00003282
Avi Kivitya8170e52012-10-23 12:30:10 +02003283void *cpu_physical_memory_map(hwaddr addr,
3284 hwaddr *plen,
Avi Kivityac1970f2012-10-03 16:22:53 +02003285 int is_write)
3286{
3287 return address_space_map(&address_space_memory, addr, plen, is_write);
3288}
3289
Avi Kivitya8170e52012-10-23 12:30:10 +02003290void cpu_physical_memory_unmap(void *buffer, hwaddr len,
3291 int is_write, hwaddr access_len)
Avi Kivityac1970f2012-10-03 16:22:53 +02003292{
3293 return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len);
3294}
3295
Paolo Bonzini0ce265f2016-11-22 11:34:02 +01003296#define ARG1_DECL AddressSpace *as
3297#define ARG1 as
3298#define SUFFIX
3299#define TRANSLATE(...) address_space_translate(as, __VA_ARGS__)
3300#define IS_DIRECT(mr, is_write) memory_access_is_direct(mr, is_write)
3301#define MAP_RAM(mr, ofs) qemu_map_ram_ptr((mr)->ram_block, ofs)
3302#define INVALIDATE(mr, ofs, len) invalidate_and_set_dirty(mr, ofs, len)
3303#define RCU_READ_LOCK(...) rcu_read_lock()
3304#define RCU_READ_UNLOCK(...) rcu_read_unlock()
3305#include "memory_ldst.inc.c"
Alexander Graf1e78bcc2011-07-06 09:09:23 +02003306
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003307int64_t address_space_cache_init(MemoryRegionCache *cache,
3308 AddressSpace *as,
3309 hwaddr addr,
3310 hwaddr len,
3311 bool is_write)
3312{
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003313 cache->len = len;
3314 cache->as = as;
3315 cache->xlat = addr;
3316 return len;
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003317}
3318
3319void address_space_cache_invalidate(MemoryRegionCache *cache,
3320 hwaddr addr,
3321 hwaddr access_len)
3322{
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003323}
3324
3325void address_space_cache_destroy(MemoryRegionCache *cache)
3326{
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003327 cache->as = NULL;
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003328}
3329
3330#define ARG1_DECL MemoryRegionCache *cache
3331#define ARG1 cache
3332#define SUFFIX _cached
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003333#define TRANSLATE(addr, ...) \
3334 address_space_translate(cache->as, cache->xlat + (addr), __VA_ARGS__)
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003335#define IS_DIRECT(mr, is_write) true
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003336#define MAP_RAM(mr, ofs) qemu_map_ram_ptr((mr)->ram_block, ofs)
3337#define INVALIDATE(mr, ofs, len) invalidate_and_set_dirty(mr, ofs, len)
3338#define RCU_READ_LOCK() rcu_read_lock()
3339#define RCU_READ_UNLOCK() rcu_read_unlock()
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003340#include "memory_ldst.inc.c"
3341
aliguori5e2972f2009-03-28 17:51:36 +00003342/* virtual memory access for debug (includes writing to ROM) */
Andreas Färberf17ec442013-06-29 19:40:58 +02003343int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
bellardb448f2f2004-02-25 23:24:04 +00003344 uint8_t *buf, int len, int is_write)
bellard13eb76e2004-01-24 15:23:36 +00003345{
3346 int l;
Avi Kivitya8170e52012-10-23 12:30:10 +02003347 hwaddr phys_addr;
j_mayer9b3c35e2007-04-07 11:21:28 +00003348 target_ulong page;
bellard13eb76e2004-01-24 15:23:36 +00003349
Christian Borntraeger79ca7a12017-03-07 15:19:08 +01003350 cpu_synchronize_state(cpu);
bellard13eb76e2004-01-24 15:23:36 +00003351 while (len > 0) {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003352 int asidx;
3353 MemTxAttrs attrs;
3354
bellard13eb76e2004-01-24 15:23:36 +00003355 page = addr & TARGET_PAGE_MASK;
Peter Maydell5232e4c2016-01-21 14:15:06 +00003356 phys_addr = cpu_get_phys_page_attrs_debug(cpu, page, &attrs);
3357 asidx = cpu_asidx_from_attrs(cpu, attrs);
bellard13eb76e2004-01-24 15:23:36 +00003358 /* if no physical page mapped, return an error */
3359 if (phys_addr == -1)
3360 return -1;
3361 l = (page + TARGET_PAGE_SIZE) - addr;
3362 if (l > len)
3363 l = len;
aliguori5e2972f2009-03-28 17:51:36 +00003364 phys_addr += (addr & ~TARGET_PAGE_MASK);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003365 if (is_write) {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003366 cpu_physical_memory_write_rom(cpu->cpu_ases[asidx].as,
3367 phys_addr, buf, l);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003368 } else {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003369 address_space_rw(cpu->cpu_ases[asidx].as, phys_addr,
3370 MEMTXATTRS_UNSPECIFIED,
Peter Maydell5c9eb022015-04-26 16:49:24 +01003371 buf, l, 0);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003372 }
bellard13eb76e2004-01-24 15:23:36 +00003373 len -= l;
3374 buf += l;
3375 addr += l;
3376 }
3377 return 0;
3378}
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003379
3380/*
3381 * Allows code that needs to deal with migration bitmaps etc to still be built
3382 * target independent.
3383 */
Juan Quintela20afaed2017-03-21 09:09:14 +01003384size_t qemu_target_page_size(void)
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003385{
Juan Quintela20afaed2017-03-21 09:09:14 +01003386 return TARGET_PAGE_SIZE;
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003387}
3388
Paul Brooka68fe892010-03-01 00:08:59 +00003389#endif
bellard13eb76e2004-01-24 15:23:36 +00003390
Blue Swirl8e4a4242013-01-06 18:30:17 +00003391/*
3392 * A helper function for the _utterly broken_ virtio device model to find out if
3393 * it's running on a big endian machine. Don't do this at home kids!
3394 */
Greg Kurz98ed8ec2014-06-24 19:26:29 +02003395bool target_words_bigendian(void);
3396bool target_words_bigendian(void)
Blue Swirl8e4a4242013-01-06 18:30:17 +00003397{
3398#if defined(TARGET_WORDS_BIGENDIAN)
3399 return true;
3400#else
3401 return false;
3402#endif
3403}
3404
Wen Congyang76f35532012-05-07 12:04:18 +08003405#ifndef CONFIG_USER_ONLY
Avi Kivitya8170e52012-10-23 12:30:10 +02003406bool cpu_physical_memory_is_io(hwaddr phys_addr)
Wen Congyang76f35532012-05-07 12:04:18 +08003407{
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003408 MemoryRegion*mr;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003409 hwaddr l = 1;
Paolo Bonzini41063e12015-03-18 14:21:43 +01003410 bool res;
Wen Congyang76f35532012-05-07 12:04:18 +08003411
Paolo Bonzini41063e12015-03-18 14:21:43 +01003412 rcu_read_lock();
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003413 mr = address_space_translate(&address_space_memory,
3414 phys_addr, &phys_addr, &l, false);
Wen Congyang76f35532012-05-07 12:04:18 +08003415
Paolo Bonzini41063e12015-03-18 14:21:43 +01003416 res = !(memory_region_is_ram(mr) || memory_region_is_romd(mr));
3417 rcu_read_unlock();
3418 return res;
Wen Congyang76f35532012-05-07 12:04:18 +08003419}
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003420
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003421int qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003422{
3423 RAMBlock *block;
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003424 int ret = 0;
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003425
Mike Day0dc3f442013-09-05 14:41:35 -04003426 rcu_read_lock();
3427 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003428 ret = func(block->idstr, block->host, block->offset,
3429 block->used_length, opaque);
3430 if (ret) {
3431 break;
3432 }
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003433 }
Mike Day0dc3f442013-09-05 14:41:35 -04003434 rcu_read_unlock();
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003435 return ret;
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003436}
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003437
3438/*
3439 * Unmap pages of memory from start to start+length such that
3440 * they a) read as 0, b) Trigger whatever fault mechanism
3441 * the OS provides for postcopy.
3442 * The pages must be unmapped by the end of the function.
3443 * Returns: 0 on success, none-0 on failure
3444 *
3445 */
3446int ram_block_discard_range(RAMBlock *rb, uint64_t start, size_t length)
3447{
3448 int ret = -1;
3449
3450 uint8_t *host_startaddr = rb->host + start;
3451
3452 if ((uintptr_t)host_startaddr & (rb->page_size - 1)) {
3453 error_report("ram_block_discard_range: Unaligned start address: %p",
3454 host_startaddr);
3455 goto err;
3456 }
3457
3458 if ((start + length) <= rb->used_length) {
3459 uint8_t *host_endaddr = host_startaddr + length;
3460 if ((uintptr_t)host_endaddr & (rb->page_size - 1)) {
3461 error_report("ram_block_discard_range: Unaligned end address: %p",
3462 host_endaddr);
3463 goto err;
3464 }
3465
3466 errno = ENOTSUP; /* If we are missing MADVISE etc */
3467
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003468 if (rb->page_size == qemu_host_page_size) {
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003469#if defined(CONFIG_MADVISE)
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003470 /* Note: We need the madvise MADV_DONTNEED behaviour of definitely
3471 * freeing the page.
3472 */
3473 ret = madvise(host_startaddr, length, MADV_DONTNEED);
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003474#endif
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003475 } else {
3476 /* Huge page case - unfortunately it can't do DONTNEED, but
3477 * it can do the equivalent by FALLOC_FL_PUNCH_HOLE in the
3478 * huge page file.
3479 */
3480#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
3481 ret = fallocate(rb->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
3482 start, length);
3483#endif
3484 }
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003485 if (ret) {
3486 ret = -errno;
3487 error_report("ram_block_discard_range: Failed to discard range "
3488 "%s:%" PRIx64 " +%zx (%d)",
3489 rb->idstr, start, length, ret);
3490 }
3491 } else {
3492 error_report("ram_block_discard_range: Overrun block '%s' (%" PRIu64
3493 "/%zx/" RAM_ADDR_FMT")",
3494 rb->idstr, start, length, rb->used_length);
3495 }
3496
3497err:
3498 return ret;
3499}
3500
Peter Maydellec3f8c92013-06-27 20:53:38 +01003501#endif