blob: b3c080f731f2a959c259bbf920d0f2bebfe55950 [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"
Juan Quintela51180422017-04-24 20:50:19 +020027#include "exec/target_page.h"
bellardb67d9a52008-05-23 09:57:34 +000028#include "tcg.h"
Paolo Bonzini741da0d2014-06-27 08:40:04 +020029#include "hw/qdev-core.h"
Michael S. Tsirkin4485bd22015-03-11 07:56:34 +010030#if !defined(CONFIG_USER_ONLY)
Marcel Apfelbaum47c8ca52015-02-04 17:43:54 +020031#include "hw/boards.h"
Paolo Bonzini33c11872016-03-15 16:58:45 +010032#include "hw/xen/xen.h"
Michael S. Tsirkin4485bd22015-03-11 07:56:34 +010033#endif
Paolo Bonzini9c17d612012-12-17 18:20:04 +010034#include "sysemu/kvm.h"
Markus Armbruster2ff3de62013-07-04 15:09:22 +020035#include "sysemu/sysemu.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010036#include "qemu/timer.h"
37#include "qemu/config-file.h"
Andreas Färber75a34032013-09-02 16:57:02 +020038#include "qemu/error-report.h"
pbrook53a59602006-03-25 19:31:22 +000039#if defined(CONFIG_USER_ONLY)
Markus Armbrustera9c94272016-06-22 19:11:19 +020040#include "qemu.h"
Jun Nakajima432d2682010-08-31 16:41:25 +010041#else /* !CONFIG_USER_ONLY */
Paolo Bonzini741da0d2014-06-27 08:40:04 +020042#include "hw/hw.h"
43#include "exec/memory.h"
Paolo Bonzinidf43d492016-03-16 10:24:54 +010044#include "exec/ioport.h"
Paolo Bonzini741da0d2014-06-27 08:40:04 +020045#include "sysemu/dma.h"
Alexey Kardashevskiy9c607662017-03-02 13:36:11 +110046#include "sysemu/numa.h"
Christian Borntraeger79ca7a12017-03-07 15:19:08 +010047#include "sysemu/hw_accel.h"
Paolo Bonzini741da0d2014-06-27 08:40:04 +020048#include "exec/address-spaces.h"
Paolo Bonzini9c17d612012-12-17 18:20:04 +010049#include "sysemu/xen-mapcache.h"
Daniel P. Berrange0ab8ed12017-01-25 16:14:15 +000050#include "trace-root.h"
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +000051
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +000052#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
53#include <fcntl.h>
54#include <linux/falloc.h>
55#endif
56
pbrook53a59602006-03-25 19:31:22 +000057#endif
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +010058#include "exec/cpu-all.h"
Mike Day0dc3f442013-09-05 14:41:35 -040059#include "qemu/rcu_queue.h"
Jan Kiszka4840f102015-06-18 18:47:22 +020060#include "qemu/main-loop.h"
Blue Swirl5b6dd862012-12-02 16:04:43 +000061#include "translate-all.h"
Pavel Dovgalyuk76159362015-09-17 19:25:07 +030062#include "sysemu/replay.h"
Blue Swirl0cac1b62012-04-09 16:50:52 +000063
Paolo Bonzini022c62c2012-12-17 18:19:49 +010064#include "exec/memory-internal.h"
Juan Quintela220c3eb2013-10-14 17:13:59 +020065#include "exec/ram_addr.h"
Paolo Bonzini508127e2016-01-07 16:55:28 +030066#include "exec/log.h"
Avi Kivity67d95c12011-12-15 15:25:22 +020067
Bharata B Rao9dfeca72016-05-12 09:18:12 +053068#include "migration/vmstate.h"
69
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +020070#include "qemu/range.h"
Michael S. Tsirkin794e8f32015-09-24 14:41:17 +030071#ifndef _WIN32
72#include "qemu/mmap-alloc.h"
73#endif
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +020074
Peter Xube9b23c2017-05-12 12:17:41 +080075#include "monitor/monitor.h"
76
blueswir1db7b5422007-05-26 17:36:03 +000077//#define DEBUG_SUBPAGE
ths1196be32007-03-17 15:17:58 +000078
pbrook99773bd2006-04-16 15:14:59 +000079#if !defined(CONFIG_USER_ONLY)
Mike Day0dc3f442013-09-05 14:41:35 -040080/* ram_list is read under rcu_read_lock()/rcu_read_unlock(). Writes
81 * are protected by the ramlist lock.
82 */
Mike Day0d53d9f2015-01-21 13:45:24 +010083RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) };
Avi Kivity62152b82011-07-26 14:26:14 +030084
85static MemoryRegion *system_memory;
Avi Kivity309cb472011-08-08 16:09:03 +030086static MemoryRegion *system_io;
Avi Kivity62152b82011-07-26 14:26:14 +030087
Avi Kivityf6790af2012-10-02 20:13:51 +020088AddressSpace address_space_io;
89AddressSpace address_space_memory;
Avi Kivity2673a5d2012-10-02 18:49:28 +020090
Paolo Bonzini0844e002013-05-24 14:37:28 +020091MemoryRegion io_mem_rom, io_mem_notdirty;
Jan Kiszkaacc9d802013-05-26 21:55:37 +020092static MemoryRegion io_mem_unassigned;
Avi Kivity0e0df1e2012-01-02 00:32:15 +020093
Paolo Bonzini7bd4f432014-05-14 17:43:22 +080094/* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
95#define RAM_PREALLOC (1 << 0)
96
Paolo Bonzinidbcb8982014-06-10 19:15:24 +080097/* RAM is mmap-ed with MAP_SHARED */
98#define RAM_SHARED (1 << 1)
99
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +0200100/* Only a portion of RAM (used_length) is actually used, and migrated.
101 * This used_length size can change across reboots.
102 */
103#define RAM_RESIZEABLE (1 << 2)
104
pbrooke2eef172008-06-08 01:09:01 +0000105#endif
bellard9fa3e852004-01-04 18:06:42 +0000106
Peter Maydell20bccb82016-10-24 16:26:49 +0100107#ifdef TARGET_PAGE_BITS_VARY
108int target_page_bits;
109bool target_page_bits_decided;
110#endif
111
Andreas Färberbdc44642013-06-24 23:50:24 +0200112struct CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus);
bellard6a00d602005-11-21 23:25:50 +0000113/* current CPU in the current thread. It is only valid inside
114 cpu_exec() */
Paolo Bonzinif240eb62015-08-26 00:17:58 +0200115__thread CPUState *current_cpu;
pbrook2e70f6e2008-06-29 01:03:05 +0000116/* 0 = Do not count executed instructions.
thsbf20dc02008-06-30 17:22:19 +0000117 1 = Precise instruction counting.
pbrook2e70f6e2008-06-29 01:03:05 +0000118 2 = Adaptive rate instruction counting. */
Paolo Bonzini5708fc62012-11-26 15:36:40 +0100119int use_icount;
bellard6a00d602005-11-21 23:25:50 +0000120
Yang Zhonga0be0c52017-07-03 18:12:13 +0800121uintptr_t qemu_host_page_size;
122intptr_t qemu_host_page_mask;
123uintptr_t qemu_real_host_page_size;
124intptr_t qemu_real_host_page_mask;
125
Peter Maydell20bccb82016-10-24 16:26:49 +0100126bool set_preferred_target_page_bits(int bits)
127{
128 /* The target page size is the lowest common denominator for all
129 * the CPUs in the system, so we can only make it smaller, never
130 * larger. And we can't make it smaller once we've committed to
131 * a particular size.
132 */
133#ifdef TARGET_PAGE_BITS_VARY
134 assert(bits >= TARGET_PAGE_BITS_MIN);
135 if (target_page_bits == 0 || target_page_bits > bits) {
136 if (target_page_bits_decided) {
137 return false;
138 }
139 target_page_bits = bits;
140 }
141#endif
142 return true;
143}
144
pbrooke2eef172008-06-08 01:09:01 +0000145#if !defined(CONFIG_USER_ONLY)
Avi Kivity4346ae32012-02-10 17:00:01 +0200146
Peter Maydell20bccb82016-10-24 16:26:49 +0100147static void finalize_target_page_bits(void)
148{
149#ifdef TARGET_PAGE_BITS_VARY
150 if (target_page_bits == 0) {
151 target_page_bits = TARGET_PAGE_BITS_MIN;
152 }
153 target_page_bits_decided = true;
154#endif
155}
156
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200157typedef struct PhysPageEntry PhysPageEntry;
158
159struct PhysPageEntry {
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200160 /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200161 uint32_t skip : 6;
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200162 /* index into phys_sections (!skip) or phys_map_nodes (skip) */
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200163 uint32_t ptr : 26;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200164};
165
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200166#define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6)
167
Paolo Bonzini03f49952013-11-07 17:14:36 +0100168/* Size of the L2 (and L3, etc) page tables. */
Paolo Bonzini57271d62013-11-07 17:14:37 +0100169#define ADDR_SPACE_BITS 64
Paolo Bonzini03f49952013-11-07 17:14:36 +0100170
Michael S. Tsirkin026736c2013-11-13 20:13:03 +0200171#define P_L2_BITS 9
Paolo Bonzini03f49952013-11-07 17:14:36 +0100172#define P_L2_SIZE (1 << P_L2_BITS)
173
174#define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1)
175
176typedef PhysPageEntry Node[P_L2_SIZE];
Paolo Bonzini0475d942013-05-29 12:28:21 +0200177
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200178typedef struct PhysPageMap {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100179 struct rcu_head rcu;
180
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200181 unsigned sections_nb;
182 unsigned sections_nb_alloc;
183 unsigned nodes_nb;
184 unsigned nodes_nb_alloc;
185 Node *nodes;
186 MemoryRegionSection *sections;
187} PhysPageMap;
188
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200189struct AddressSpaceDispatch {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100190 struct rcu_head rcu;
191
Fam Zheng729633c2016-03-01 14:18:24 +0800192 MemoryRegionSection *mru_section;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200193 /* This is a multi-level map on the physical address space.
194 * The bottom level has pointers to MemoryRegionSections.
195 */
196 PhysPageEntry phys_map;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200197 PhysPageMap map;
Jan Kiszkaacc9d802013-05-26 21:55:37 +0200198 AddressSpace *as;
Paolo Bonzini1db8abb2013-05-21 12:07:21 +0200199};
200
Jan Kiszka90260c62013-05-26 21:46:51 +0200201#define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK)
202typedef struct subpage_t {
203 MemoryRegion iomem;
Jan Kiszkaacc9d802013-05-26 21:55:37 +0200204 AddressSpace *as;
Jan Kiszka90260c62013-05-26 21:46:51 +0200205 hwaddr base;
Vijaya Kumar K2615fab2016-10-24 16:26:49 +0100206 uint16_t sub_section[];
Jan Kiszka90260c62013-05-26 21:46:51 +0200207} subpage_t;
208
Liu Ping Fanb41aac42013-05-29 11:09:17 +0200209#define PHYS_SECTION_UNASSIGNED 0
210#define PHYS_SECTION_NOTDIRTY 1
211#define PHYS_SECTION_ROM 2
212#define PHYS_SECTION_WATCH 3
Avi Kivity5312bd82012-02-12 18:32:55 +0200213
pbrooke2eef172008-06-08 01:09:01 +0000214static void io_mem_init(void);
Avi Kivity62152b82011-07-26 14:26:14 +0300215static void memory_map_init(void);
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000216static void tcg_commit(MemoryListener *listener);
pbrooke2eef172008-06-08 01:09:01 +0000217
Avi Kivity1ec9b902012-01-02 12:47:48 +0200218static MemoryRegion io_mem_watch;
Peter Maydell32857f42015-10-01 15:29:50 +0100219
220/**
221 * CPUAddressSpace: all the information a CPU needs about an AddressSpace
222 * @cpu: the CPU whose AddressSpace this is
223 * @as: the AddressSpace itself
224 * @memory_dispatch: its dispatch pointer (cached, RCU protected)
225 * @tcg_as_listener: listener for tracking changes to the AddressSpace
226 */
227struct CPUAddressSpace {
228 CPUState *cpu;
229 AddressSpace *as;
230 struct AddressSpaceDispatch *memory_dispatch;
231 MemoryListener tcg_as_listener;
232};
233
Gerd Hoffmann8deaf122017-04-21 11:16:25 +0200234struct DirtyBitmapSnapshot {
235 ram_addr_t start;
236 ram_addr_t end;
237 unsigned long dirty[];
238};
239
pbrook6658ffb2007-03-16 23:58:11 +0000240#endif
bellard54936002003-05-13 00:25:15 +0000241
Paul Brook6d9a1302010-02-28 23:55:53 +0000242#if !defined(CONFIG_USER_ONLY)
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200243
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200244static void phys_map_node_reserve(PhysPageMap *map, unsigned nodes)
Avi Kivityf7bf5462012-02-13 20:12:05 +0200245{
Peter Lieven101420b2016-07-15 12:03:50 +0200246 static unsigned alloc_hint = 16;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200247 if (map->nodes_nb + nodes > map->nodes_nb_alloc) {
Peter Lieven101420b2016-07-15 12:03:50 +0200248 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, alloc_hint);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200249 map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, map->nodes_nb + nodes);
250 map->nodes = g_renew(Node, map->nodes, map->nodes_nb_alloc);
Peter Lieven101420b2016-07-15 12:03:50 +0200251 alloc_hint = map->nodes_nb_alloc;
Avi Kivityf7bf5462012-02-13 20:12:05 +0200252 }
253}
254
Paolo Bonzinidb946042015-05-21 15:12:29 +0200255static uint32_t phys_map_node_alloc(PhysPageMap *map, bool leaf)
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200256{
257 unsigned i;
Michael S. Tsirkin8b795762013-11-11 14:51:56 +0200258 uint32_t ret;
Paolo Bonzinidb946042015-05-21 15:12:29 +0200259 PhysPageEntry e;
260 PhysPageEntry *p;
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200261
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200262 ret = map->nodes_nb++;
Paolo Bonzinidb946042015-05-21 15:12:29 +0200263 p = map->nodes[ret];
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200264 assert(ret != PHYS_MAP_NODE_NIL);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200265 assert(ret != map->nodes_nb_alloc);
Paolo Bonzinidb946042015-05-21 15:12:29 +0200266
267 e.skip = leaf ? 0 : 1;
268 e.ptr = leaf ? PHYS_SECTION_UNASSIGNED : PHYS_MAP_NODE_NIL;
Paolo Bonzini03f49952013-11-07 17:14:36 +0100269 for (i = 0; i < P_L2_SIZE; ++i) {
Paolo Bonzinidb946042015-05-21 15:12:29 +0200270 memcpy(&p[i], &e, sizeof(e));
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200271 }
Avi Kivityf7bf5462012-02-13 20:12:05 +0200272 return ret;
Avi Kivityd6f2ea22012-02-12 20:12:49 +0200273}
274
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200275static void phys_page_set_level(PhysPageMap *map, PhysPageEntry *lp,
276 hwaddr *index, hwaddr *nb, uint16_t leaf,
Avi Kivity29990972012-02-13 20:21:20 +0200277 int level)
Avi Kivityf7bf5462012-02-13 20:12:05 +0200278{
279 PhysPageEntry *p;
Paolo Bonzini03f49952013-11-07 17:14:36 +0100280 hwaddr step = (hwaddr)1 << (level * P_L2_BITS);
Avi Kivityf7bf5462012-02-13 20:12:05 +0200281
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200282 if (lp->skip && lp->ptr == PHYS_MAP_NODE_NIL) {
Paolo Bonzinidb946042015-05-21 15:12:29 +0200283 lp->ptr = phys_map_node_alloc(map, level == 0);
Avi Kivityf7bf5462012-02-13 20:12:05 +0200284 }
Paolo Bonzinidb946042015-05-21 15:12:29 +0200285 p = map->nodes[lp->ptr];
Paolo Bonzini03f49952013-11-07 17:14:36 +0100286 lp = &p[(*index >> (level * P_L2_BITS)) & (P_L2_SIZE - 1)];
Avi Kivityf7bf5462012-02-13 20:12:05 +0200287
Paolo Bonzini03f49952013-11-07 17:14:36 +0100288 while (*nb && lp < &p[P_L2_SIZE]) {
Avi Kivity07f07b32012-02-13 20:45:32 +0200289 if ((*index & (step - 1)) == 0 && *nb >= step) {
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200290 lp->skip = 0;
Avi Kivityc19e8802012-02-13 20:25:31 +0200291 lp->ptr = leaf;
Avi Kivity07f07b32012-02-13 20:45:32 +0200292 *index += step;
293 *nb -= step;
Avi Kivity29990972012-02-13 20:21:20 +0200294 } else {
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200295 phys_page_set_level(map, lp, index, nb, leaf, level - 1);
Avi Kivity29990972012-02-13 20:21:20 +0200296 }
297 ++lp;
Avi Kivityf7bf5462012-02-13 20:12:05 +0200298 }
299}
300
Avi Kivityac1970f2012-10-03 16:22:53 +0200301static void phys_page_set(AddressSpaceDispatch *d,
Avi Kivitya8170e52012-10-23 12:30:10 +0200302 hwaddr index, hwaddr nb,
Avi Kivity29990972012-02-13 20:21:20 +0200303 uint16_t leaf)
bellard92e873b2004-05-21 14:52:29 +0000304{
Avi Kivity29990972012-02-13 20:21:20 +0200305 /* Wildly overreserve - it doesn't matter much. */
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200306 phys_map_node_reserve(&d->map, 3 * P_L2_LEVELS);
bellard92e873b2004-05-21 14:52:29 +0000307
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200308 phys_page_set_level(&d->map, &d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1);
bellard92e873b2004-05-21 14:52:29 +0000309}
310
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200311/* Compact a non leaf page entry. Simply detect that the entry has a single child,
312 * and update our entry so we can skip it and go directly to the destination.
313 */
Marc-André Lureauefee6782016-09-28 16:37:20 +0400314static void phys_page_compact(PhysPageEntry *lp, Node *nodes)
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200315{
316 unsigned valid_ptr = P_L2_SIZE;
317 int valid = 0;
318 PhysPageEntry *p;
319 int i;
320
321 if (lp->ptr == PHYS_MAP_NODE_NIL) {
322 return;
323 }
324
325 p = nodes[lp->ptr];
326 for (i = 0; i < P_L2_SIZE; i++) {
327 if (p[i].ptr == PHYS_MAP_NODE_NIL) {
328 continue;
329 }
330
331 valid_ptr = i;
332 valid++;
333 if (p[i].skip) {
Marc-André Lureauefee6782016-09-28 16:37:20 +0400334 phys_page_compact(&p[i], nodes);
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200335 }
336 }
337
338 /* We can only compress if there's only one child. */
339 if (valid != 1) {
340 return;
341 }
342
343 assert(valid_ptr < P_L2_SIZE);
344
345 /* Don't compress if it won't fit in the # of bits we have. */
346 if (lp->skip + p[valid_ptr].skip >= (1 << 3)) {
347 return;
348 }
349
350 lp->ptr = p[valid_ptr].ptr;
351 if (!p[valid_ptr].skip) {
352 /* If our only child is a leaf, make this a leaf. */
353 /* By design, we should have made this node a leaf to begin with so we
354 * should never reach here.
355 * But since it's so simple to handle this, let's do it just in case we
356 * change this rule.
357 */
358 lp->skip = 0;
359 } else {
360 lp->skip += p[valid_ptr].skip;
361 }
362}
363
364static void phys_page_compact_all(AddressSpaceDispatch *d, int nodes_nb)
365{
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200366 if (d->phys_map.skip) {
Marc-André Lureauefee6782016-09-28 16:37:20 +0400367 phys_page_compact(&d->phys_map, d->map.nodes);
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200368 }
369}
370
Fam Zheng29cb5332016-03-01 14:18:23 +0800371static inline bool section_covers_addr(const MemoryRegionSection *section,
372 hwaddr addr)
373{
374 /* Memory topology clips a memory region to [0, 2^64); size.hi > 0 means
375 * the section must cover the entire address space.
376 */
Richard Henderson258dfaa2016-06-29 15:48:03 -0700377 return int128_gethi(section->size) ||
Fam Zheng29cb5332016-03-01 14:18:23 +0800378 range_covers_byte(section->offset_within_address_space,
Richard Henderson258dfaa2016-06-29 15:48:03 -0700379 int128_getlo(section->size), addr);
Fam Zheng29cb5332016-03-01 14:18:23 +0800380}
381
Peter Xu003a0cf2017-05-15 16:50:57 +0800382static MemoryRegionSection *phys_page_find(AddressSpaceDispatch *d, hwaddr addr)
bellard92e873b2004-05-21 14:52:29 +0000383{
Peter Xu003a0cf2017-05-15 16:50:57 +0800384 PhysPageEntry lp = d->phys_map, *p;
385 Node *nodes = d->map.nodes;
386 MemoryRegionSection *sections = d->map.sections;
Michael S. Tsirkin97115a82013-11-13 20:08:19 +0200387 hwaddr index = addr >> TARGET_PAGE_BITS;
Avi Kivity31ab2b42012-02-13 16:44:19 +0200388 int i;
Avi Kivityf1f6e3b2011-11-20 17:52:22 +0200389
Michael S. Tsirkin9736e552013-11-11 14:42:43 +0200390 for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) {
Avi Kivityc19e8802012-02-13 20:25:31 +0200391 if (lp.ptr == PHYS_MAP_NODE_NIL) {
Paolo Bonzini9affd6f2013-05-29 12:09:47 +0200392 return &sections[PHYS_SECTION_UNASSIGNED];
Avi Kivity31ab2b42012-02-13 16:44:19 +0200393 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +0200394 p = nodes[lp.ptr];
Paolo Bonzini03f49952013-11-07 17:14:36 +0100395 lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)];
Avi Kivityf1f6e3b2011-11-20 17:52:22 +0200396 }
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200397
Fam Zheng29cb5332016-03-01 14:18:23 +0800398 if (section_covers_addr(&sections[lp.ptr], addr)) {
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +0200399 return &sections[lp.ptr];
400 } else {
401 return &sections[PHYS_SECTION_UNASSIGNED];
402 }
Avi Kivityf3705d52012-03-08 16:16:34 +0200403}
404
Blue Swirle5548612012-04-21 13:08:33 +0000405bool memory_region_is_unassigned(MemoryRegion *mr)
406{
Paolo Bonzini2a8e7492013-05-24 14:34:08 +0200407 return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device
Blue Swirle5548612012-04-21 13:08:33 +0000408 && mr != &io_mem_watch;
409}
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200410
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100411/* Called from RCU critical section */
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200412static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
Jan Kiszka90260c62013-05-26 21:46:51 +0200413 hwaddr addr,
414 bool resolve_subpage)
Jan Kiszka9f029602013-05-06 16:48:02 +0200415{
Fam Zheng729633c2016-03-01 14:18:24 +0800416 MemoryRegionSection *section = atomic_read(&d->mru_section);
Jan Kiszka90260c62013-05-26 21:46:51 +0200417 subpage_t *subpage;
Fam Zheng729633c2016-03-01 14:18:24 +0800418 bool update;
Jan Kiszka90260c62013-05-26 21:46:51 +0200419
Fam Zheng729633c2016-03-01 14:18:24 +0800420 if (section && section != &d->map.sections[PHYS_SECTION_UNASSIGNED] &&
421 section_covers_addr(section, addr)) {
422 update = false;
423 } else {
Peter Xu003a0cf2017-05-15 16:50:57 +0800424 section = phys_page_find(d, addr);
Fam Zheng729633c2016-03-01 14:18:24 +0800425 update = true;
426 }
Jan Kiszka90260c62013-05-26 21:46:51 +0200427 if (resolve_subpage && section->mr->subpage) {
428 subpage = container_of(section->mr, subpage_t, iomem);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +0200429 section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]];
Jan Kiszka90260c62013-05-26 21:46:51 +0200430 }
Fam Zheng729633c2016-03-01 14:18:24 +0800431 if (update) {
432 atomic_set(&d->mru_section, section);
433 }
Jan Kiszka90260c62013-05-26 21:46:51 +0200434 return section;
Jan Kiszka9f029602013-05-06 16:48:02 +0200435}
436
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100437/* Called from RCU critical section */
Jan Kiszka90260c62013-05-26 21:46:51 +0200438static MemoryRegionSection *
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200439address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
Jan Kiszka90260c62013-05-26 21:46:51 +0200440 hwaddr *plen, bool resolve_subpage)
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200441{
442 MemoryRegionSection *section;
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200443 MemoryRegion *mr;
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100444 Int128 diff;
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200445
Paolo Bonzinic7086b42013-06-02 15:27:39 +0200446 section = address_space_lookup_region(d, addr, resolve_subpage);
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200447 /* Compute offset within MemoryRegionSection */
448 addr -= section->offset_within_address_space;
449
450 /* Compute offset within MemoryRegion */
451 *xlat = addr + section->offset_within_region;
452
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200453 mr = section->mr;
Paolo Bonzinib242e0e2015-07-04 00:24:51 +0200454
455 /* MMIO registers can be expected to perform full-width accesses based only
456 * on their address, without considering adjacent registers that could
457 * decode to completely different MemoryRegions. When such registers
458 * exist (e.g. I/O ports 0xcf8 and 0xcf9 on most PC chipsets), MMIO
459 * regions overlap wildly. For this reason we cannot clamp the accesses
460 * here.
461 *
462 * If the length is small (as is the case for address_space_ldl/stl),
463 * everything works fine. If the incoming length is large, however,
464 * the caller really has to do the clamping through memory_access_size.
465 */
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200466 if (memory_region_is_ram(mr)) {
Paolo Bonzinie4a511f2015-06-17 10:36:54 +0200467 diff = int128_sub(section->size, int128_make64(addr));
Paolo Bonzini965eb2f2015-06-17 10:40:27 +0200468 *plen = int128_get64(int128_min(diff, int128_make64(*plen)));
469 }
Paolo Bonzini149f54b2013-05-24 12:59:37 +0200470 return section;
471}
Jan Kiszka90260c62013-05-26 21:46:51 +0200472
Paolo Bonzini41063e12015-03-18 14:21:43 +0100473/* Called from RCU critical section */
Peter Xua7640402017-05-17 16:57:42 +0800474static MemoryRegionSection address_space_do_translate(AddressSpace *as,
475 hwaddr addr,
476 hwaddr *xlat,
477 hwaddr *plen,
478 bool is_write,
479 bool is_mmio)
Jan Kiszka90260c62013-05-26 21:46:51 +0200480{
Avi Kivity30951152012-10-30 13:47:46 +0200481 IOMMUTLBEntry iotlb;
482 MemoryRegionSection *section;
Alexey Kardashevskiy3df9d742017-07-11 13:56:19 +1000483 IOMMUMemoryRegion *iommu_mr;
Alexey Kardashevskiy1221a472017-07-11 13:56:20 +1000484 IOMMUMemoryRegionClass *imrc;
Avi Kivity30951152012-10-30 13:47:46 +0200485
486 for (;;) {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100487 AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
Peter Xua7640402017-05-17 16:57:42 +0800488 section = address_space_translate_internal(d, addr, &addr, plen, is_mmio);
Avi Kivity30951152012-10-30 13:47:46 +0200489
Alexey Kardashevskiy3df9d742017-07-11 13:56:19 +1000490 iommu_mr = memory_region_get_iommu(section->mr);
491 if (!iommu_mr) {
Avi Kivity30951152012-10-30 13:47:46 +0200492 break;
493 }
Alexey Kardashevskiy1221a472017-07-11 13:56:20 +1000494 imrc = memory_region_get_iommu_class_nocheck(iommu_mr);
Avi Kivity30951152012-10-30 13:47:46 +0200495
Alexey Kardashevskiy1221a472017-07-11 13:56:20 +1000496 iotlb = imrc->translate(iommu_mr, addr, is_write ?
497 IOMMU_WO : IOMMU_RO);
Avi Kivity30951152012-10-30 13:47:46 +0200498 addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
499 | (addr & iotlb.addr_mask));
Peter Crosthwaite23820db2015-03-16 22:35:54 -0700500 *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1);
Avi Kivity30951152012-10-30 13:47:46 +0200501 if (!(iotlb.perm & (1 << is_write))) {
Peter Xua7640402017-05-17 16:57:42 +0800502 goto translate_fail;
Avi Kivity30951152012-10-30 13:47:46 +0200503 }
504
505 as = iotlb.target_as;
506 }
507
Peter Xua7640402017-05-17 16:57:42 +0800508 *xlat = addr;
509
510 return *section;
511
512translate_fail:
513 return (MemoryRegionSection) { .mr = &io_mem_unassigned };
514}
515
516/* Called from RCU critical section */
517IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr,
518 bool is_write)
519{
520 MemoryRegionSection section;
521 hwaddr xlat, plen;
522
523 /* Try to get maximum page mask during translation. */
524 plen = (hwaddr)-1;
525
526 /* This can never be MMIO. */
527 section = address_space_do_translate(as, addr, &xlat, &plen,
528 is_write, false);
529
530 /* Illegal translation */
531 if (section.mr == &io_mem_unassigned) {
532 goto iotlb_fail;
533 }
534
535 /* Convert memory region offset into address space offset */
536 xlat += section.offset_within_address_space -
537 section.offset_within_region;
538
539 if (plen == (hwaddr)-1) {
540 /*
541 * We use default page size here. Logically it only happens
542 * for identity mappings.
543 */
544 plen = TARGET_PAGE_SIZE;
545 }
546
547 /* Convert to address mask */
548 plen -= 1;
549
550 return (IOMMUTLBEntry) {
551 .target_as = section.address_space,
552 .iova = addr & ~plen,
553 .translated_addr = xlat & ~plen,
554 .addr_mask = plen,
555 /* IOTLBs are for DMAs, and DMA only allows on RAMs. */
556 .perm = IOMMU_RW,
557 };
558
559iotlb_fail:
560 return (IOMMUTLBEntry) {0};
561}
562
563/* Called from RCU critical section */
564MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
565 hwaddr *xlat, hwaddr *plen,
566 bool is_write)
567{
568 MemoryRegion *mr;
569 MemoryRegionSection section;
570
571 /* This can be MMIO, so setup MMIO bit. */
572 section = address_space_do_translate(as, addr, xlat, plen, is_write, true);
573 mr = section.mr;
574
Alexey Kardashevskiyfe680d02014-05-07 13:40:39 +0000575 if (xen_enabled() && memory_access_is_direct(mr, is_write)) {
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100576 hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr;
Peter Crosthwaite23820db2015-03-16 22:35:54 -0700577 *plen = MIN(page, *plen);
Paolo Bonzinia87f3952014-02-07 15:47:46 +0100578 }
579
Avi Kivity30951152012-10-30 13:47:46 +0200580 return mr;
Jan Kiszka90260c62013-05-26 21:46:51 +0200581}
582
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +0100583/* Called from RCU critical section */
Jan Kiszka90260c62013-05-26 21:46:51 +0200584MemoryRegionSection *
Peter Maydelld7898cd2016-01-21 14:15:05 +0000585address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr,
Paolo Bonzini9d82b5a2013-08-16 08:26:30 +0200586 hwaddr *xlat, hwaddr *plen)
Jan Kiszka90260c62013-05-26 21:46:51 +0200587{
Avi Kivity30951152012-10-30 13:47:46 +0200588 MemoryRegionSection *section;
Alex Bennéef35e44e2016-10-21 16:34:18 +0100589 AddressSpaceDispatch *d = atomic_rcu_read(&cpu->cpu_ases[asidx].memory_dispatch);
Peter Maydelld7898cd2016-01-21 14:15:05 +0000590
591 section = address_space_translate_internal(d, addr, xlat, plen, false);
Avi Kivity30951152012-10-30 13:47:46 +0200592
Alexey Kardashevskiy3df9d742017-07-11 13:56:19 +1000593 assert(!memory_region_is_iommu(section->mr));
Avi Kivity30951152012-10-30 13:47:46 +0200594 return section;
Jan Kiszka90260c62013-05-26 21:46:51 +0200595}
bellard9fa3e852004-01-04 18:06:42 +0000596#endif
bellardfd6ce8f2003-05-14 19:00:11 +0000597
Andreas Färberb170fce2013-01-20 20:23:22 +0100598#if !defined(CONFIG_USER_ONLY)
pbrook9656f322008-07-01 20:01:19 +0000599
Juan Quintelae59fb372009-09-29 22:48:21 +0200600static int cpu_common_post_load(void *opaque, int version_id)
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200601{
Andreas Färber259186a2013-01-17 18:51:17 +0100602 CPUState *cpu = opaque;
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200603
aurel323098dba2009-03-07 21:28:24 +0000604 /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the
605 version_id is increased. */
Andreas Färber259186a2013-01-17 18:51:17 +0100606 cpu->interrupt_request &= ~0x01;
Alex Bennéed10eb082016-11-14 14:17:28 +0000607 tlb_flush(cpu);
pbrook9656f322008-07-01 20:01:19 +0000608
609 return 0;
610}
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200611
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400612static int cpu_common_pre_load(void *opaque)
613{
614 CPUState *cpu = opaque;
615
Paolo Bonziniadee6422014-12-19 12:53:14 +0100616 cpu->exception_index = -1;
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400617
618 return 0;
619}
620
621static bool cpu_common_exception_index_needed(void *opaque)
622{
623 CPUState *cpu = opaque;
624
Paolo Bonziniadee6422014-12-19 12:53:14 +0100625 return tcg_enabled() && cpu->exception_index != -1;
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400626}
627
628static const VMStateDescription vmstate_cpu_common_exception_index = {
629 .name = "cpu_common/exception_index",
630 .version_id = 1,
631 .minimum_version_id = 1,
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200632 .needed = cpu_common_exception_index_needed,
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400633 .fields = (VMStateField[]) {
634 VMSTATE_INT32(exception_index, CPUState),
635 VMSTATE_END_OF_LIST()
636 }
637};
638
Andrey Smetaninbac05aa2015-07-03 15:01:44 +0300639static bool cpu_common_crash_occurred_needed(void *opaque)
640{
641 CPUState *cpu = opaque;
642
643 return cpu->crash_occurred;
644}
645
646static const VMStateDescription vmstate_cpu_common_crash_occurred = {
647 .name = "cpu_common/crash_occurred",
648 .version_id = 1,
649 .minimum_version_id = 1,
650 .needed = cpu_common_crash_occurred_needed,
651 .fields = (VMStateField[]) {
652 VMSTATE_BOOL(crash_occurred, CPUState),
653 VMSTATE_END_OF_LIST()
654 }
655};
656
Andreas Färber1a1562f2013-06-17 04:09:11 +0200657const VMStateDescription vmstate_cpu_common = {
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200658 .name = "cpu_common",
659 .version_id = 1,
660 .minimum_version_id = 1,
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400661 .pre_load = cpu_common_pre_load,
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200662 .post_load = cpu_common_post_load,
Juan Quintela35d08452014-04-16 16:01:33 +0200663 .fields = (VMStateField[]) {
Andreas Färber259186a2013-01-17 18:51:17 +0100664 VMSTATE_UINT32(halted, CPUState),
665 VMSTATE_UINT32(interrupt_request, CPUState),
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200666 VMSTATE_END_OF_LIST()
Pavel Dovgaluk6c3bff02014-07-31 09:41:17 +0400667 },
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200668 .subsections = (const VMStateDescription*[]) {
669 &vmstate_cpu_common_exception_index,
Andrey Smetaninbac05aa2015-07-03 15:01:44 +0300670 &vmstate_cpu_common_crash_occurred,
Juan Quintela5cd8cad2014-09-23 14:09:54 +0200671 NULL
Juan Quintelae7f4eff2009-09-10 03:04:33 +0200672 }
673};
Andreas Färber1a1562f2013-06-17 04:09:11 +0200674
pbrook9656f322008-07-01 20:01:19 +0000675#endif
676
Andreas Färber38d8f5c2012-12-17 19:47:15 +0100677CPUState *qemu_get_cpu(int index)
Glauber Costa950f1472009-06-09 12:15:18 -0400678{
Andreas Färberbdc44642013-06-24 23:50:24 +0200679 CPUState *cpu;
Glauber Costa950f1472009-06-09 12:15:18 -0400680
Andreas Färberbdc44642013-06-24 23:50:24 +0200681 CPU_FOREACH(cpu) {
Andreas Färber55e5c282012-12-17 06:18:02 +0100682 if (cpu->cpu_index == index) {
Andreas Färberbdc44642013-06-24 23:50:24 +0200683 return cpu;
Andreas Färber55e5c282012-12-17 06:18:02 +0100684 }
Glauber Costa950f1472009-06-09 12:15:18 -0400685 }
686
Andreas Färberbdc44642013-06-24 23:50:24 +0200687 return NULL;
Glauber Costa950f1472009-06-09 12:15:18 -0400688}
689
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000690#if !defined(CONFIG_USER_ONLY)
Peter Maydell56943e82016-01-21 14:15:04 +0000691void cpu_address_space_init(CPUState *cpu, AddressSpace *as, int asidx)
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000692{
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000693 CPUAddressSpace *newas;
694
695 /* Target code should have set num_ases before calling us */
696 assert(asidx < cpu->num_ases);
697
Peter Maydell56943e82016-01-21 14:15:04 +0000698 if (asidx == 0) {
699 /* address space 0 gets the convenience alias */
700 cpu->as = as;
701 }
702
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000703 /* KVM cannot currently support multiple address spaces. */
704 assert(asidx == 0 || !kvm_enabled());
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000705
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000706 if (!cpu->cpu_ases) {
707 cpu->cpu_ases = g_new0(CPUAddressSpace, cpu->num_ases);
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000708 }
Peter Maydell32857f42015-10-01 15:29:50 +0100709
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000710 newas = &cpu->cpu_ases[asidx];
711 newas->cpu = cpu;
712 newas->as = as;
Peter Maydell56943e82016-01-21 14:15:04 +0000713 if (tcg_enabled()) {
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000714 newas->tcg_as_listener.commit = tcg_commit;
715 memory_listener_register(&newas->tcg_as_listener, as);
Peter Maydell56943e82016-01-21 14:15:04 +0000716 }
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000717}
Peter Maydell651a5bc2016-01-21 14:15:05 +0000718
719AddressSpace *cpu_get_address_space(CPUState *cpu, int asidx)
720{
721 /* Return the AddressSpace corresponding to the specified index */
722 return cpu->cpu_ases[asidx].as;
723}
Edgar E. Iglesias09daed82013-12-17 13:06:51 +1000724#endif
725
Laurent Vivier7bbc1242016-10-20 13:26:04 +0200726void cpu_exec_unrealizefn(CPUState *cpu)
Bharata B Rao1c59eb32016-05-12 09:18:11 +0530727{
Bharata B Rao9dfeca72016-05-12 09:18:12 +0530728 CPUClass *cc = CPU_GET_CLASS(cpu);
729
Paolo Bonzini267f6852016-08-28 03:45:14 +0200730 cpu_list_remove(cpu);
Bharata B Rao9dfeca72016-05-12 09:18:12 +0530731
732 if (cc->vmsd != NULL) {
733 vmstate_unregister(NULL, cc->vmsd, cpu);
734 }
735 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
736 vmstate_unregister(NULL, &vmstate_cpu_common, cpu);
737 }
Bharata B Rao1c59eb32016-05-12 09:18:11 +0530738}
739
Laurent Vivier39e329e2016-10-20 13:26:02 +0200740void cpu_exec_initfn(CPUState *cpu)
bellardfd6ce8f2003-05-14 19:00:11 +0000741{
Peter Maydell56943e82016-01-21 14:15:04 +0000742 cpu->as = NULL;
Peter Maydell12ebc9a2016-01-21 14:15:04 +0000743 cpu->num_ases = 0;
Peter Maydell56943e82016-01-21 14:15:04 +0000744
Eduardo Habkost291135b2015-04-27 17:00:33 -0300745#ifndef CONFIG_USER_ONLY
Eduardo Habkost291135b2015-04-27 17:00:33 -0300746 cpu->thread_id = qemu_get_thread_id();
Peter Crosthwaite6731d862016-01-21 14:15:06 +0000747
748 /* This is a softmmu CPU object, so create a property for it
749 * so users can wire up its memory. (This can't go in qom/cpu.c
750 * because that file is compiled only once for both user-mode
751 * and system builds.) The default if no link is set up is to use
752 * the system address space.
753 */
754 object_property_add_link(OBJECT(cpu), "memory", TYPE_MEMORY_REGION,
755 (Object **)&cpu->memory,
756 qdev_prop_allow_set_link_before_realize,
757 OBJ_PROP_LINK_UNREF_ON_RELEASE,
758 &error_abort);
759 cpu->memory = system_memory;
760 object_ref(OBJECT(cpu->memory));
Eduardo Habkost291135b2015-04-27 17:00:33 -0300761#endif
Laurent Vivier39e329e2016-10-20 13:26:02 +0200762}
763
Laurent Vivierce5b1bb2016-10-20 13:26:03 +0200764void cpu_exec_realizefn(CPUState *cpu, Error **errp)
Laurent Vivier39e329e2016-10-20 13:26:02 +0200765{
766 CPUClass *cc ATTRIBUTE_UNUSED = CPU_GET_CLASS(cpu);
Eduardo Habkost291135b2015-04-27 17:00:33 -0300767
Paolo Bonzini267f6852016-08-28 03:45:14 +0200768 cpu_list_add(cpu);
Igor Mammedov1bc7e522016-07-25 11:59:19 +0200769
770#ifndef CONFIG_USER_ONLY
Andreas Färbere0d47942013-07-29 04:07:50 +0200771 if (qdev_get_vmsd(DEVICE(cpu)) == NULL) {
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200772 vmstate_register(NULL, cpu->cpu_index, &vmstate_cpu_common, cpu);
Andreas Färbere0d47942013-07-29 04:07:50 +0200773 }
Andreas Färberb170fce2013-01-20 20:23:22 +0100774 if (cc->vmsd != NULL) {
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200775 vmstate_register(NULL, cpu->cpu_index, cc->vmsd, cpu);
Andreas Färberb170fce2013-01-20 20:23:22 +0100776 }
Paolo Bonzini741da0d2014-06-27 08:40:04 +0200777#endif
bellardfd6ce8f2003-05-14 19:00:11 +0000778}
779
Pranith Kumar406bc332017-07-12 17:51:42 -0400780#if defined(CONFIG_USER_ONLY)
Andreas Färber00b941e2013-06-29 18:55:54 +0200781static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
Paul Brook94df27f2010-02-28 23:47:45 +0000782{
Pranith Kumar406bc332017-07-12 17:51:42 -0400783 mmap_lock();
784 tb_lock();
785 tb_invalidate_phys_page_range(pc, pc + 1, 0);
786 tb_unlock();
787 mmap_unlock();
Paul Brook94df27f2010-02-28 23:47:45 +0000788}
Pranith Kumar406bc332017-07-12 17:51:42 -0400789#else
790static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
791{
792 MemTxAttrs attrs;
793 hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs);
794 int asidx = cpu_asidx_from_attrs(cpu, attrs);
795 if (phys != -1) {
796 /* Locks grabbed by tb_invalidate_phys_addr */
797 tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as,
798 phys | (pc & ~TARGET_PAGE_MASK));
799 }
800}
801#endif
bellardd720b932004-04-25 17:57:43 +0000802
Paul Brookc527ee82010-03-01 03:31:14 +0000803#if defined(CONFIG_USER_ONLY)
Andreas Färber75a34032013-09-02 16:57:02 +0200804void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
Paul Brookc527ee82010-03-01 03:31:14 +0000805
806{
807}
808
Peter Maydell3ee887e2014-09-12 14:06:48 +0100809int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
810 int flags)
811{
812 return -ENOSYS;
813}
814
815void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
816{
817}
818
Andreas Färber75a34032013-09-02 16:57:02 +0200819int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
Paul Brookc527ee82010-03-01 03:31:14 +0000820 int flags, CPUWatchpoint **watchpoint)
821{
822 return -ENOSYS;
823}
824#else
pbrook6658ffb2007-03-16 23:58:11 +0000825/* Add a watchpoint. */
Andreas Färber75a34032013-09-02 16:57:02 +0200826int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len,
aliguoria1d1bb32008-11-18 20:07:32 +0000827 int flags, CPUWatchpoint **watchpoint)
pbrook6658ffb2007-03-16 23:58:11 +0000828{
aliguoric0ce9982008-11-25 22:13:57 +0000829 CPUWatchpoint *wp;
pbrook6658ffb2007-03-16 23:58:11 +0000830
Peter Maydell05068c02014-09-12 14:06:48 +0100831 /* forbid ranges which are empty or run off the end of the address space */
Max Filippov07e28632014-09-17 22:03:36 -0700832 if (len == 0 || (addr + len - 1) < addr) {
Andreas Färber75a34032013-09-02 16:57:02 +0200833 error_report("tried to set invalid watchpoint at %"
834 VADDR_PRIx ", len=%" VADDR_PRIu, addr, len);
aliguorib4051332008-11-18 20:14:20 +0000835 return -EINVAL;
836 }
Anthony Liguori7267c092011-08-20 22:09:37 -0500837 wp = g_malloc(sizeof(*wp));
pbrook6658ffb2007-03-16 23:58:11 +0000838
aliguoria1d1bb32008-11-18 20:07:32 +0000839 wp->vaddr = addr;
Peter Maydell05068c02014-09-12 14:06:48 +0100840 wp->len = len;
aliguoria1d1bb32008-11-18 20:07:32 +0000841 wp->flags = flags;
842
aliguori2dc9f412008-11-18 20:56:59 +0000843 /* keep all GDB-injected watchpoints in front */
Andreas Färberff4700b2013-08-26 18:23:18 +0200844 if (flags & BP_GDB) {
845 QTAILQ_INSERT_HEAD(&cpu->watchpoints, wp, entry);
846 } else {
847 QTAILQ_INSERT_TAIL(&cpu->watchpoints, wp, entry);
848 }
aliguoria1d1bb32008-11-18 20:07:32 +0000849
Andreas Färber31b030d2013-09-04 01:29:02 +0200850 tlb_flush_page(cpu, addr);
aliguoria1d1bb32008-11-18 20:07:32 +0000851
852 if (watchpoint)
853 *watchpoint = wp;
854 return 0;
pbrook6658ffb2007-03-16 23:58:11 +0000855}
856
aliguoria1d1bb32008-11-18 20:07:32 +0000857/* Remove a specific watchpoint. */
Andreas Färber75a34032013-09-02 16:57:02 +0200858int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len,
aliguoria1d1bb32008-11-18 20:07:32 +0000859 int flags)
pbrook6658ffb2007-03-16 23:58:11 +0000860{
aliguoria1d1bb32008-11-18 20:07:32 +0000861 CPUWatchpoint *wp;
pbrook6658ffb2007-03-16 23:58:11 +0000862
Andreas Färberff4700b2013-08-26 18:23:18 +0200863 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +0100864 if (addr == wp->vaddr && len == wp->len
aliguori6e140f22008-11-18 20:37:55 +0000865 && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) {
Andreas Färber75a34032013-09-02 16:57:02 +0200866 cpu_watchpoint_remove_by_ref(cpu, wp);
pbrook6658ffb2007-03-16 23:58:11 +0000867 return 0;
868 }
869 }
aliguoria1d1bb32008-11-18 20:07:32 +0000870 return -ENOENT;
pbrook6658ffb2007-03-16 23:58:11 +0000871}
872
aliguoria1d1bb32008-11-18 20:07:32 +0000873/* Remove a specific watchpoint by reference. */
Andreas Färber75a34032013-09-02 16:57:02 +0200874void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint)
aliguoria1d1bb32008-11-18 20:07:32 +0000875{
Andreas Färberff4700b2013-08-26 18:23:18 +0200876 QTAILQ_REMOVE(&cpu->watchpoints, watchpoint, entry);
edgar_igl7d03f822008-05-17 18:58:29 +0000877
Andreas Färber31b030d2013-09-04 01:29:02 +0200878 tlb_flush_page(cpu, watchpoint->vaddr);
aliguoria1d1bb32008-11-18 20:07:32 +0000879
Anthony Liguori7267c092011-08-20 22:09:37 -0500880 g_free(watchpoint);
edgar_igl7d03f822008-05-17 18:58:29 +0000881}
882
aliguoria1d1bb32008-11-18 20:07:32 +0000883/* Remove all matching watchpoints. */
Andreas Färber75a34032013-09-02 16:57:02 +0200884void cpu_watchpoint_remove_all(CPUState *cpu, int mask)
aliguoria1d1bb32008-11-18 20:07:32 +0000885{
aliguoric0ce9982008-11-25 22:13:57 +0000886 CPUWatchpoint *wp, *next;
aliguoria1d1bb32008-11-18 20:07:32 +0000887
Andreas Färberff4700b2013-08-26 18:23:18 +0200888 QTAILQ_FOREACH_SAFE(wp, &cpu->watchpoints, entry, next) {
Andreas Färber75a34032013-09-02 16:57:02 +0200889 if (wp->flags & mask) {
890 cpu_watchpoint_remove_by_ref(cpu, wp);
891 }
aliguoric0ce9982008-11-25 22:13:57 +0000892 }
aliguoria1d1bb32008-11-18 20:07:32 +0000893}
Peter Maydell05068c02014-09-12 14:06:48 +0100894
895/* Return true if this watchpoint address matches the specified
896 * access (ie the address range covered by the watchpoint overlaps
897 * partially or completely with the address range covered by the
898 * access).
899 */
900static inline bool cpu_watchpoint_address_matches(CPUWatchpoint *wp,
901 vaddr addr,
902 vaddr len)
903{
904 /* We know the lengths are non-zero, but a little caution is
905 * required to avoid errors in the case where the range ends
906 * exactly at the top of the address space and so addr + len
907 * wraps round to zero.
908 */
909 vaddr wpend = wp->vaddr + wp->len - 1;
910 vaddr addrend = addr + len - 1;
911
912 return !(addr > wpend || wp->vaddr > addrend);
913}
914
Paul Brookc527ee82010-03-01 03:31:14 +0000915#endif
aliguoria1d1bb32008-11-18 20:07:32 +0000916
917/* Add a breakpoint. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200918int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags,
aliguoria1d1bb32008-11-18 20:07:32 +0000919 CPUBreakpoint **breakpoint)
bellard4c3a88a2003-07-26 12:06:08 +0000920{
aliguoric0ce9982008-11-25 22:13:57 +0000921 CPUBreakpoint *bp;
ths3b46e622007-09-17 08:09:54 +0000922
Anthony Liguori7267c092011-08-20 22:09:37 -0500923 bp = g_malloc(sizeof(*bp));
aliguoria1d1bb32008-11-18 20:07:32 +0000924
925 bp->pc = pc;
926 bp->flags = flags;
927
aliguori2dc9f412008-11-18 20:56:59 +0000928 /* keep all GDB-injected breakpoints in front */
Andreas Färber00b941e2013-06-29 18:55:54 +0200929 if (flags & BP_GDB) {
Andreas Färberf0c3c502013-08-26 21:22:53 +0200930 QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry);
Andreas Färber00b941e2013-06-29 18:55:54 +0200931 } else {
Andreas Färberf0c3c502013-08-26 21:22:53 +0200932 QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry);
Andreas Färber00b941e2013-06-29 18:55:54 +0200933 }
aliguoria1d1bb32008-11-18 20:07:32 +0000934
Andreas Färberf0c3c502013-08-26 21:22:53 +0200935 breakpoint_invalidate(cpu, pc);
aliguoria1d1bb32008-11-18 20:07:32 +0000936
Andreas Färber00b941e2013-06-29 18:55:54 +0200937 if (breakpoint) {
aliguoria1d1bb32008-11-18 20:07:32 +0000938 *breakpoint = bp;
Andreas Färber00b941e2013-06-29 18:55:54 +0200939 }
aliguoria1d1bb32008-11-18 20:07:32 +0000940 return 0;
aliguoria1d1bb32008-11-18 20:07:32 +0000941}
942
943/* Remove a specific breakpoint. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200944int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags)
aliguoria1d1bb32008-11-18 20:07:32 +0000945{
aliguoria1d1bb32008-11-18 20:07:32 +0000946 CPUBreakpoint *bp;
947
Andreas Färberf0c3c502013-08-26 21:22:53 +0200948 QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) {
aliguoria1d1bb32008-11-18 20:07:32 +0000949 if (bp->pc == pc && bp->flags == flags) {
Andreas Färberb3310ab2013-09-02 17:26:20 +0200950 cpu_breakpoint_remove_by_ref(cpu, bp);
bellard4c3a88a2003-07-26 12:06:08 +0000951 return 0;
aliguoria1d1bb32008-11-18 20:07:32 +0000952 }
bellard4c3a88a2003-07-26 12:06:08 +0000953 }
aliguoria1d1bb32008-11-18 20:07:32 +0000954 return -ENOENT;
bellard4c3a88a2003-07-26 12:06:08 +0000955}
956
aliguoria1d1bb32008-11-18 20:07:32 +0000957/* Remove a specific breakpoint by reference. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200958void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *breakpoint)
bellard4c3a88a2003-07-26 12:06:08 +0000959{
Andreas Färberf0c3c502013-08-26 21:22:53 +0200960 QTAILQ_REMOVE(&cpu->breakpoints, breakpoint, entry);
961
962 breakpoint_invalidate(cpu, breakpoint->pc);
aliguoria1d1bb32008-11-18 20:07:32 +0000963
Anthony Liguori7267c092011-08-20 22:09:37 -0500964 g_free(breakpoint);
aliguoria1d1bb32008-11-18 20:07:32 +0000965}
966
967/* Remove all matching breakpoints. */
Andreas Färberb3310ab2013-09-02 17:26:20 +0200968void cpu_breakpoint_remove_all(CPUState *cpu, int mask)
aliguoria1d1bb32008-11-18 20:07:32 +0000969{
aliguoric0ce9982008-11-25 22:13:57 +0000970 CPUBreakpoint *bp, *next;
aliguoria1d1bb32008-11-18 20:07:32 +0000971
Andreas Färberf0c3c502013-08-26 21:22:53 +0200972 QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) {
Andreas Färberb3310ab2013-09-02 17:26:20 +0200973 if (bp->flags & mask) {
974 cpu_breakpoint_remove_by_ref(cpu, bp);
975 }
aliguoric0ce9982008-11-25 22:13:57 +0000976 }
bellard4c3a88a2003-07-26 12:06:08 +0000977}
978
bellardc33a3462003-07-29 20:50:33 +0000979/* enable or disable single step mode. EXCP_DEBUG is returned by the
980 CPU loop after each instruction */
Andreas Färber3825b282013-06-24 18:41:06 +0200981void cpu_single_step(CPUState *cpu, int enabled)
bellardc33a3462003-07-29 20:50:33 +0000982{
Andreas Färbered2803d2013-06-21 20:20:45 +0200983 if (cpu->singlestep_enabled != enabled) {
984 cpu->singlestep_enabled = enabled;
985 if (kvm_enabled()) {
Stefan Weil38e478e2013-07-25 20:50:21 +0200986 kvm_update_guest_debug(cpu, 0);
Andreas Färbered2803d2013-06-21 20:20:45 +0200987 } else {
Stuart Bradyccbb4d42009-05-03 12:15:06 +0100988 /* must flush all the translated code to avoid inconsistencies */
aliguorie22a25c2009-03-12 20:12:48 +0000989 /* XXX: only flush what is necessary */
Peter Crosthwaitebbd77c12015-06-23 19:31:15 -0700990 tb_flush(cpu);
aliguorie22a25c2009-03-12 20:12:48 +0000991 }
bellardc33a3462003-07-29 20:50:33 +0000992 }
bellardc33a3462003-07-29 20:50:33 +0000993}
994
Andreas Färbera47dddd2013-09-03 17:38:47 +0200995void cpu_abort(CPUState *cpu, const char *fmt, ...)
bellard75012672003-06-21 13:11:07 +0000996{
997 va_list ap;
pbrook493ae1f2007-11-23 16:53:59 +0000998 va_list ap2;
bellard75012672003-06-21 13:11:07 +0000999
1000 va_start(ap, fmt);
pbrook493ae1f2007-11-23 16:53:59 +00001001 va_copy(ap2, ap);
bellard75012672003-06-21 13:11:07 +00001002 fprintf(stderr, "qemu: fatal: ");
1003 vfprintf(stderr, fmt, ap);
1004 fprintf(stderr, "\n");
Andreas Färber878096e2013-05-27 01:33:50 +02001005 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP);
Paolo Bonzini013a2942015-11-13 13:16:27 +01001006 if (qemu_log_separate()) {
Richard Henderson1ee73212016-09-22 15:17:10 -07001007 qemu_log_lock();
aliguori93fcfe32009-01-15 22:34:14 +00001008 qemu_log("qemu: fatal: ");
1009 qemu_log_vprintf(fmt, ap2);
1010 qemu_log("\n");
Andreas Färbera0762852013-06-16 07:28:50 +02001011 log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP);
aliguori31b1a7b2009-01-15 22:35:09 +00001012 qemu_log_flush();
Richard Henderson1ee73212016-09-22 15:17:10 -07001013 qemu_log_unlock();
aliguori93fcfe32009-01-15 22:34:14 +00001014 qemu_log_close();
balrog924edca2007-06-10 14:07:13 +00001015 }
pbrook493ae1f2007-11-23 16:53:59 +00001016 va_end(ap2);
j_mayerf9373292007-09-29 12:18:20 +00001017 va_end(ap);
Pavel Dovgalyuk76159362015-09-17 19:25:07 +03001018 replay_finish();
Riku Voipiofd052bf2010-01-25 14:30:49 +02001019#if defined(CONFIG_USER_ONLY)
1020 {
1021 struct sigaction act;
1022 sigfillset(&act.sa_mask);
1023 act.sa_handler = SIG_DFL;
1024 sigaction(SIGABRT, &act, NULL);
1025 }
1026#endif
bellard75012672003-06-21 13:11:07 +00001027 abort();
1028}
1029
bellard01243112004-01-04 15:48:17 +00001030#if !defined(CONFIG_USER_ONLY)
Mike Day0dc3f442013-09-05 14:41:35 -04001031/* Called from RCU critical section */
Paolo Bonzini041603f2013-09-09 17:49:45 +02001032static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
1033{
1034 RAMBlock *block;
1035
Paolo Bonzini43771532013-09-09 17:58:40 +02001036 block = atomic_rcu_read(&ram_list.mru_block);
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001037 if (block && addr - block->offset < block->max_length) {
Paolo Bonzini68851b92015-10-22 13:51:30 +02001038 return block;
Paolo Bonzini041603f2013-09-09 17:49:45 +02001039 }
Peter Xu99e15582017-05-12 12:17:39 +08001040 RAMBLOCK_FOREACH(block) {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001041 if (addr - block->offset < block->max_length) {
Paolo Bonzini041603f2013-09-09 17:49:45 +02001042 goto found;
1043 }
1044 }
1045
1046 fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr);
1047 abort();
1048
1049found:
Paolo Bonzini43771532013-09-09 17:58:40 +02001050 /* It is safe to write mru_block outside the iothread lock. This
1051 * is what happens:
1052 *
1053 * mru_block = xxx
1054 * rcu_read_unlock()
1055 * xxx removed from list
1056 * rcu_read_lock()
1057 * read mru_block
1058 * mru_block = NULL;
1059 * call_rcu(reclaim_ramblock, xxx);
1060 * rcu_read_unlock()
1061 *
1062 * atomic_rcu_set is not needed here. The block was already published
1063 * when it was placed into the list. Here we're just making an extra
1064 * copy of the pointer.
1065 */
Paolo Bonzini041603f2013-09-09 17:49:45 +02001066 ram_list.mru_block = block;
1067 return block;
1068}
1069
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001070static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length)
bellard1ccde1c2004-02-06 19:46:14 +00001071{
Peter Crosthwaite9a135652015-09-10 22:39:41 -07001072 CPUState *cpu;
Paolo Bonzini041603f2013-09-09 17:49:45 +02001073 ram_addr_t start1;
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001074 RAMBlock *block;
1075 ram_addr_t end;
1076
1077 end = TARGET_PAGE_ALIGN(start + length);
1078 start &= TARGET_PAGE_MASK;
bellardf23db162005-08-21 19:12:28 +00001079
Mike Day0dc3f442013-09-05 14:41:35 -04001080 rcu_read_lock();
Paolo Bonzini041603f2013-09-09 17:49:45 +02001081 block = qemu_get_ram_block(start);
1082 assert(block == qemu_get_ram_block(end - 1));
Michael S. Tsirkin1240be22014-11-12 11:44:41 +02001083 start1 = (uintptr_t)ramblock_ptr(block, start - block->offset);
Peter Crosthwaite9a135652015-09-10 22:39:41 -07001084 CPU_FOREACH(cpu) {
1085 tlb_reset_dirty(cpu, start1, length);
1086 }
Mike Day0dc3f442013-09-05 14:41:35 -04001087 rcu_read_unlock();
Juan Quintelad24981d2012-05-22 00:42:40 +02001088}
1089
1090/* Note: start and end must be within the same ram block. */
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001091bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start,
1092 ram_addr_t length,
1093 unsigned client)
Juan Quintelad24981d2012-05-22 00:42:40 +02001094{
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001095 DirtyMemoryBlocks *blocks;
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001096 unsigned long end, page;
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001097 bool dirty = false;
Juan Quintelad24981d2012-05-22 00:42:40 +02001098
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001099 if (length == 0) {
1100 return false;
1101 }
1102
1103 end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
1104 page = start >> TARGET_PAGE_BITS;
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001105
1106 rcu_read_lock();
1107
1108 blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
1109
1110 while (page < end) {
1111 unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE;
1112 unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE;
1113 unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset);
1114
1115 dirty |= bitmap_test_and_clear_atomic(blocks->blocks[idx],
1116 offset, num);
1117 page += num;
1118 }
1119
1120 rcu_read_unlock();
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001121
1122 if (dirty && tcg_enabled()) {
Juan Quintelaa2f4d5b2013-10-10 11:49:53 +02001123 tlb_reset_dirty_range_all(start, length);
Juan Quintelad24981d2012-05-22 00:42:40 +02001124 }
Stefan Hajnoczi03eebc92014-12-02 11:23:18 +00001125
1126 return dirty;
bellard1ccde1c2004-02-06 19:46:14 +00001127}
1128
Gerd Hoffmann8deaf122017-04-21 11:16:25 +02001129DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty
1130 (ram_addr_t start, ram_addr_t length, unsigned client)
1131{
1132 DirtyMemoryBlocks *blocks;
1133 unsigned long align = 1UL << (TARGET_PAGE_BITS + BITS_PER_LEVEL);
1134 ram_addr_t first = QEMU_ALIGN_DOWN(start, align);
1135 ram_addr_t last = QEMU_ALIGN_UP(start + length, align);
1136 DirtyBitmapSnapshot *snap;
1137 unsigned long page, end, dest;
1138
1139 snap = g_malloc0(sizeof(*snap) +
1140 ((last - first) >> (TARGET_PAGE_BITS + 3)));
1141 snap->start = first;
1142 snap->end = last;
1143
1144 page = first >> TARGET_PAGE_BITS;
1145 end = last >> TARGET_PAGE_BITS;
1146 dest = 0;
1147
1148 rcu_read_lock();
1149
1150 blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
1151
1152 while (page < end) {
1153 unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE;
1154 unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE;
1155 unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset);
1156
1157 assert(QEMU_IS_ALIGNED(offset, (1 << BITS_PER_LEVEL)));
1158 assert(QEMU_IS_ALIGNED(num, (1 << BITS_PER_LEVEL)));
1159 offset >>= BITS_PER_LEVEL;
1160
1161 bitmap_copy_and_clear_atomic(snap->dirty + dest,
1162 blocks->blocks[idx] + offset,
1163 num);
1164 page += num;
1165 dest += num >> BITS_PER_LEVEL;
1166 }
1167
1168 rcu_read_unlock();
1169
1170 if (tcg_enabled()) {
1171 tlb_reset_dirty_range_all(start, length);
1172 }
1173
1174 return snap;
1175}
1176
1177bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap,
1178 ram_addr_t start,
1179 ram_addr_t length)
1180{
1181 unsigned long page, end;
1182
1183 assert(start >= snap->start);
1184 assert(start + length <= snap->end);
1185
1186 end = TARGET_PAGE_ALIGN(start + length - snap->start) >> TARGET_PAGE_BITS;
1187 page = (start - snap->start) >> TARGET_PAGE_BITS;
1188
1189 while (page < end) {
1190 if (test_bit(page, snap->dirty)) {
1191 return true;
1192 }
1193 page++;
1194 }
1195 return false;
1196}
1197
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01001198/* Called from RCU critical section */
Andreas Färberbb0e6272013-09-03 13:32:01 +02001199hwaddr memory_region_section_get_iotlb(CPUState *cpu,
Paolo Bonzini149f54b2013-05-24 12:59:37 +02001200 MemoryRegionSection *section,
1201 target_ulong vaddr,
1202 hwaddr paddr, hwaddr xlat,
1203 int prot,
1204 target_ulong *address)
Blue Swirle5548612012-04-21 13:08:33 +00001205{
Avi Kivitya8170e52012-10-23 12:30:10 +02001206 hwaddr iotlb;
Blue Swirle5548612012-04-21 13:08:33 +00001207 CPUWatchpoint *wp;
1208
Blue Swirlcc5bea62012-04-14 14:56:48 +00001209 if (memory_region_is_ram(section->mr)) {
Blue Swirle5548612012-04-21 13:08:33 +00001210 /* Normal RAM. */
Paolo Bonzinie4e69792016-03-01 10:44:50 +01001211 iotlb = memory_region_get_ram_addr(section->mr) + xlat;
Blue Swirle5548612012-04-21 13:08:33 +00001212 if (!section->readonly) {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001213 iotlb |= PHYS_SECTION_NOTDIRTY;
Blue Swirle5548612012-04-21 13:08:33 +00001214 } else {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001215 iotlb |= PHYS_SECTION_ROM;
Blue Swirle5548612012-04-21 13:08:33 +00001216 }
1217 } else {
Peter Maydell0b8e2c12015-07-20 12:27:16 +01001218 AddressSpaceDispatch *d;
1219
1220 d = atomic_rcu_read(&section->address_space->dispatch);
1221 iotlb = section - d->map.sections;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02001222 iotlb += xlat;
Blue Swirle5548612012-04-21 13:08:33 +00001223 }
1224
1225 /* Make accesses to pages with watchpoints go via the
1226 watchpoint trap routines. */
Andreas Färberff4700b2013-08-26 18:23:18 +02001227 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +01001228 if (cpu_watchpoint_address_matches(wp, vaddr, TARGET_PAGE_SIZE)) {
Blue Swirle5548612012-04-21 13:08:33 +00001229 /* Avoid trapping reads of pages with a write breakpoint. */
1230 if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) {
Liu Ping Fanb41aac42013-05-29 11:09:17 +02001231 iotlb = PHYS_SECTION_WATCH + paddr;
Blue Swirle5548612012-04-21 13:08:33 +00001232 *address |= TLB_MMIO;
1233 break;
1234 }
1235 }
1236 }
1237
1238 return iotlb;
1239}
bellard9fa3e852004-01-04 18:06:42 +00001240#endif /* defined(CONFIG_USER_ONLY) */
1241
pbrooke2eef172008-06-08 01:09:01 +00001242#if !defined(CONFIG_USER_ONLY)
pbrook8da3ff12008-12-01 18:59:50 +00001243
Anthony Liguoric227f092009-10-01 16:12:16 -05001244static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
Avi Kivity5312bd82012-02-12 18:32:55 +02001245 uint16_t section);
Jan Kiszkaacc9d802013-05-26 21:55:37 +02001246static subpage_t *subpage_init(AddressSpace *as, hwaddr base);
Avi Kivity54688b12012-02-09 17:34:32 +02001247
Igor Mammedova2b257d2014-10-31 16:38:37 +00001248static void *(*phys_mem_alloc)(size_t size, uint64_t *align) =
1249 qemu_anon_ram_alloc;
Markus Armbruster91138032013-07-31 15:11:08 +02001250
1251/*
1252 * Set a custom physical guest memory alloator.
1253 * Accelerators with unusual needs may need this. Hopefully, we can
1254 * get rid of it eventually.
1255 */
Igor Mammedova2b257d2014-10-31 16:38:37 +00001256void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align))
Markus Armbruster91138032013-07-31 15:11:08 +02001257{
1258 phys_mem_alloc = alloc;
1259}
1260
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001261static uint16_t phys_section_add(PhysPageMap *map,
1262 MemoryRegionSection *section)
Avi Kivity5312bd82012-02-12 18:32:55 +02001263{
Paolo Bonzini68f3f652013-05-07 11:30:23 +02001264 /* The physical section number is ORed with a page-aligned
1265 * pointer to produce the iotlb entries. Thus it should
1266 * never overflow into the page-aligned value.
1267 */
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001268 assert(map->sections_nb < TARGET_PAGE_SIZE);
Paolo Bonzini68f3f652013-05-07 11:30:23 +02001269
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001270 if (map->sections_nb == map->sections_nb_alloc) {
1271 map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16);
1272 map->sections = g_renew(MemoryRegionSection, map->sections,
1273 map->sections_nb_alloc);
Avi Kivity5312bd82012-02-12 18:32:55 +02001274 }
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001275 map->sections[map->sections_nb] = *section;
Paolo Bonzinidfde4e62013-05-06 10:46:11 +02001276 memory_region_ref(section->mr);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001277 return map->sections_nb++;
Avi Kivity5312bd82012-02-12 18:32:55 +02001278}
1279
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001280static void phys_section_destroy(MemoryRegion *mr)
1281{
Don Slutz55b4e802015-11-30 17:11:04 -05001282 bool have_sub_page = mr->subpage;
1283
Paolo Bonzinidfde4e62013-05-06 10:46:11 +02001284 memory_region_unref(mr);
1285
Don Slutz55b4e802015-11-30 17:11:04 -05001286 if (have_sub_page) {
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001287 subpage_t *subpage = container_of(mr, subpage_t, iomem);
Peter Crosthwaiteb4fefef2014-06-05 23:15:52 -07001288 object_unref(OBJECT(&subpage->iomem));
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001289 g_free(subpage);
1290 }
1291}
1292
Paolo Bonzini60926662013-05-29 12:30:26 +02001293static void phys_sections_free(PhysPageMap *map)
Avi Kivity5312bd82012-02-12 18:32:55 +02001294{
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02001295 while (map->sections_nb > 0) {
1296 MemoryRegionSection *section = &map->sections[--map->sections_nb];
Paolo Bonzini058bc4b2013-06-25 09:30:48 +02001297 phys_section_destroy(section->mr);
1298 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02001299 g_free(map->sections);
1300 g_free(map->nodes);
Avi Kivity5312bd82012-02-12 18:32:55 +02001301}
1302
Avi Kivityac1970f2012-10-03 16:22:53 +02001303static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section)
Avi Kivity0f0cb162012-02-13 17:14:32 +02001304{
1305 subpage_t *subpage;
Avi Kivitya8170e52012-10-23 12:30:10 +02001306 hwaddr base = section->offset_within_address_space
Avi Kivity0f0cb162012-02-13 17:14:32 +02001307 & TARGET_PAGE_MASK;
Peter Xu003a0cf2017-05-15 16:50:57 +08001308 MemoryRegionSection *existing = phys_page_find(d, base);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001309 MemoryRegionSection subsection = {
1310 .offset_within_address_space = base,
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001311 .size = int128_make64(TARGET_PAGE_SIZE),
Avi Kivity0f0cb162012-02-13 17:14:32 +02001312 };
Avi Kivitya8170e52012-10-23 12:30:10 +02001313 hwaddr start, end;
Avi Kivity0f0cb162012-02-13 17:14:32 +02001314
Avi Kivityf3705d52012-03-08 16:16:34 +02001315 assert(existing->mr->subpage || existing->mr == &io_mem_unassigned);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001316
Avi Kivityf3705d52012-03-08 16:16:34 +02001317 if (!(existing->mr->subpage)) {
Jan Kiszkaacc9d802013-05-26 21:55:37 +02001318 subpage = subpage_init(d->as, base);
Edgar E. Iglesias3be91e82013-11-07 18:42:51 +01001319 subsection.address_space = d->as;
Avi Kivity0f0cb162012-02-13 17:14:32 +02001320 subsection.mr = &subpage->iomem;
Avi Kivityac1970f2012-10-03 16:22:53 +02001321 phys_page_set(d, base >> TARGET_PAGE_BITS, 1,
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001322 phys_section_add(&d->map, &subsection));
Avi Kivity0f0cb162012-02-13 17:14:32 +02001323 } else {
Avi Kivityf3705d52012-03-08 16:16:34 +02001324 subpage = container_of(existing->mr, subpage_t, iomem);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001325 }
1326 start = section->offset_within_address_space & ~TARGET_PAGE_MASK;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001327 end = start + int128_get64(section->size) - 1;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001328 subpage_register(subpage, start, end,
1329 phys_section_add(&d->map, section));
Avi Kivity0f0cb162012-02-13 17:14:32 +02001330}
1331
1332
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001333static void register_multipage(AddressSpaceDispatch *d,
1334 MemoryRegionSection *section)
bellard33417e72003-08-10 21:47:01 +00001335{
Avi Kivitya8170e52012-10-23 12:30:10 +02001336 hwaddr start_addr = section->offset_within_address_space;
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02001337 uint16_t section_index = phys_section_add(&d->map, section);
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001338 uint64_t num_pages = int128_get64(int128_rshift(section->size,
1339 TARGET_PAGE_BITS));
Avi Kivitydd811242012-01-02 12:17:03 +02001340
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001341 assert(num_pages);
1342 phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index);
bellard33417e72003-08-10 21:47:01 +00001343}
1344
Avi Kivityac1970f2012-10-03 16:22:53 +02001345static void mem_add(MemoryListener *listener, MemoryRegionSection *section)
Avi Kivity0f0cb162012-02-13 17:14:32 +02001346{
Paolo Bonzini89ae3372013-06-02 10:39:07 +02001347 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
Paolo Bonzini00752702013-05-29 12:13:54 +02001348 AddressSpaceDispatch *d = as->next_dispatch;
Paolo Bonzini99b9cc02013-05-27 13:18:01 +02001349 MemoryRegionSection now = *section, remain = *section;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001350 Int128 page_size = int128_make64(TARGET_PAGE_SIZE);
Avi Kivity0f0cb162012-02-13 17:14:32 +02001351
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001352 if (now.offset_within_address_space & ~TARGET_PAGE_MASK) {
1353 uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space)
1354 - now.offset_within_address_space;
1355
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001356 now.size = int128_min(int128_make64(left), now.size);
Avi Kivityac1970f2012-10-03 16:22:53 +02001357 register_subpage(d, &now);
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001358 } else {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001359 now.size = int128_zero();
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001360 }
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001361 while (int128_ne(remain.size, now.size)) {
1362 remain.size = int128_sub(remain.size, now.size);
1363 remain.offset_within_address_space += int128_get64(now.size);
1364 remain.offset_within_region += int128_get64(now.size);
Tyler Hall69b67642012-07-25 18:45:04 -04001365 now = remain;
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001366 if (int128_lt(remain.size, page_size)) {
Paolo Bonzini733d5ef2013-05-27 10:47:10 +02001367 register_subpage(d, &now);
Hu Tao88266242013-08-29 18:21:16 +08001368 } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001369 now.size = page_size;
Avi Kivityac1970f2012-10-03 16:22:53 +02001370 register_subpage(d, &now);
Tyler Hall69b67642012-07-25 18:45:04 -04001371 } else {
Paolo Bonzini052e87b2013-05-27 10:08:27 +02001372 now.size = int128_and(now.size, int128_neg(page_size));
Avi Kivityac1970f2012-10-03 16:22:53 +02001373 register_multipage(d, &now);
Tyler Hall69b67642012-07-25 18:45:04 -04001374 }
Avi Kivity0f0cb162012-02-13 17:14:32 +02001375 }
1376}
1377
Sheng Yang62a27442010-01-26 19:21:16 +08001378void qemu_flush_coalesced_mmio_buffer(void)
1379{
1380 if (kvm_enabled())
1381 kvm_flush_coalesced_mmio_buffer();
1382}
1383
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001384void qemu_mutex_lock_ramlist(void)
1385{
1386 qemu_mutex_lock(&ram_list.mutex);
1387}
1388
1389void qemu_mutex_unlock_ramlist(void)
1390{
1391 qemu_mutex_unlock(&ram_list.mutex);
1392}
1393
Peter Xube9b23c2017-05-12 12:17:41 +08001394void ram_block_dump(Monitor *mon)
1395{
1396 RAMBlock *block;
1397 char *psize;
1398
1399 rcu_read_lock();
1400 monitor_printf(mon, "%24s %8s %18s %18s %18s\n",
1401 "Block Name", "PSize", "Offset", "Used", "Total");
1402 RAMBLOCK_FOREACH(block) {
1403 psize = size_to_str(block->page_size);
1404 monitor_printf(mon, "%24s %8s 0x%016" PRIx64 " 0x%016" PRIx64
1405 " 0x%016" PRIx64 "\n", block->idstr, psize,
1406 (uint64_t)block->offset,
1407 (uint64_t)block->used_length,
1408 (uint64_t)block->max_length);
1409 g_free(psize);
1410 }
1411 rcu_read_unlock();
1412}
1413
Markus Armbrustere1e84ba2013-07-31 15:11:10 +02001414#ifdef __linux__
Alexey Kardashevskiy9c607662017-03-02 13:36:11 +11001415/*
1416 * FIXME TOCTTOU: this iterates over memory backends' mem-path, which
1417 * may or may not name the same files / on the same filesystem now as
1418 * when we actually open and map them. Iterate over the file
1419 * descriptors instead, and use qemu_fd_getpagesize().
1420 */
1421static int find_max_supported_pagesize(Object *obj, void *opaque)
1422{
1423 char *mem_path;
1424 long *hpsize_min = opaque;
1425
1426 if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) {
1427 mem_path = object_property_get_str(obj, "mem-path", NULL);
1428 if (mem_path) {
1429 long hpsize = qemu_mempath_getpagesize(mem_path);
1430 if (hpsize < *hpsize_min) {
1431 *hpsize_min = hpsize;
1432 }
1433 } else {
1434 *hpsize_min = getpagesize();
1435 }
1436 }
1437
1438 return 0;
1439}
1440
1441long qemu_getrampagesize(void)
1442{
1443 long hpsize = LONG_MAX;
1444 long mainrampagesize;
1445 Object *memdev_root;
1446
1447 if (mem_path) {
1448 mainrampagesize = qemu_mempath_getpagesize(mem_path);
1449 } else {
1450 mainrampagesize = getpagesize();
1451 }
1452
1453 /* it's possible we have memory-backend objects with
1454 * hugepage-backed RAM. these may get mapped into system
1455 * address space via -numa parameters or memory hotplug
1456 * hooks. we want to take these into account, but we
1457 * also want to make sure these supported hugepage
1458 * sizes are applicable across the entire range of memory
1459 * we may boot from, so we take the min across all
1460 * backends, and assume normal pages in cases where a
1461 * backend isn't backed by hugepages.
1462 */
1463 memdev_root = object_resolve_path("/objects", NULL);
1464 if (memdev_root) {
1465 object_child_foreach(memdev_root, find_max_supported_pagesize, &hpsize);
1466 }
1467 if (hpsize == LONG_MAX) {
1468 /* No additional memory regions found ==> Report main RAM page size */
1469 return mainrampagesize;
1470 }
1471
1472 /* If NUMA is disabled or the NUMA nodes are not backed with a
1473 * memory-backend, then there is at least one node using "normal" RAM,
1474 * so if its page size is smaller we have got to report that size instead.
1475 */
1476 if (hpsize > mainrampagesize &&
1477 (nb_numa_nodes == 0 || numa_info[0].node_memdev == NULL)) {
1478 static bool warned;
1479 if (!warned) {
1480 error_report("Huge page support disabled (n/a for main memory).");
1481 warned = true;
1482 }
1483 return mainrampagesize;
1484 }
1485
1486 return hpsize;
1487}
1488#else
1489long qemu_getrampagesize(void)
1490{
1491 return getpagesize();
1492}
1493#endif
1494
1495#ifdef __linux__
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001496static int64_t get_file_size(int fd)
1497{
1498 int64_t size = lseek(fd, 0, SEEK_END);
1499 if (size < 0) {
1500 return -errno;
1501 }
1502 return size;
1503}
1504
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001505static int file_ram_open(const char *path,
1506 const char *region_name,
1507 bool *created,
1508 Error **errp)
Marcelo Tosattic9027602010-03-01 20:25:08 -03001509{
1510 char *filename;
Peter Feiner8ca761f2013-03-04 13:54:25 -05001511 char *sanitized_name;
1512 char *c;
Paolo Bonzini5c3ece72016-03-17 15:53:13 +01001513 int fd = -1;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001514
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001515 *created = false;
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001516 for (;;) {
1517 fd = open(path, O_RDWR);
1518 if (fd >= 0) {
1519 /* @path names an existing file, use it */
1520 break;
1521 }
1522 if (errno == ENOENT) {
1523 /* @path names a file that doesn't exist, create it */
1524 fd = open(path, O_RDWR | O_CREAT | O_EXCL, 0644);
1525 if (fd >= 0) {
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001526 *created = true;
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001527 break;
1528 }
1529 } else if (errno == EISDIR) {
1530 /* @path names a directory, create a file there */
1531 /* Make name safe to use with mkstemp by replacing '/' with '_'. */
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001532 sanitized_name = g_strdup(region_name);
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001533 for (c = sanitized_name; *c != '\0'; c++) {
1534 if (*c == '/') {
1535 *c = '_';
1536 }
1537 }
1538
1539 filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path,
1540 sanitized_name);
1541 g_free(sanitized_name);
1542
1543 fd = mkstemp(filename);
1544 if (fd >= 0) {
1545 unlink(filename);
1546 g_free(filename);
1547 break;
1548 }
1549 g_free(filename);
1550 }
1551 if (errno != EEXIST && errno != EINTR) {
1552 error_setg_errno(errp, errno,
1553 "can't open backing store %s for guest RAM",
1554 path);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001555 return -1;
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001556 }
1557 /*
1558 * Try again on EINTR and EEXIST. The latter happens when
1559 * something else creates the file between our two open().
1560 */
1561 }
1562
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001563 return fd;
1564}
1565
1566static void *file_ram_alloc(RAMBlock *block,
1567 ram_addr_t memory,
1568 int fd,
1569 bool truncate,
1570 Error **errp)
1571{
1572 void *area;
1573
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001574 block->page_size = qemu_fd_getpagesize(fd);
Haozhong Zhang83606682016-10-24 20:49:37 +08001575 block->mr->align = block->page_size;
1576#if defined(__s390x__)
1577 if (kvm_enabled()) {
1578 block->mr->align = MAX(block->mr->align, QEMU_VMALLOC_ALIGN);
1579 }
1580#endif
Marcelo Tosattic9027602010-03-01 20:25:08 -03001581
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001582 if (memory < block->page_size) {
Hu Tao557529d2014-09-09 13:28:00 +08001583 error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to "
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001584 "or larger than page size 0x%zx",
1585 memory, block->page_size);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001586 return NULL;
Haozhong Zhang1775f112016-11-02 09:05:51 +08001587 }
1588
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001589 memory = ROUND_UP(memory, block->page_size);
Marcelo Tosattic9027602010-03-01 20:25:08 -03001590
1591 /*
1592 * ftruncate is not supported by hugetlbfs in older
1593 * hosts, so don't bother bailing out on errors.
1594 * If anything goes wrong with it under other filesystems,
1595 * mmap will fail.
Haozhong Zhangd6af99c2016-10-27 12:22:58 +08001596 *
1597 * Do not truncate the non-empty backend file to avoid corrupting
1598 * the existing data in the file. Disabling shrinking is not
1599 * enough. For example, the current vNVDIMM implementation stores
1600 * the guest NVDIMM labels at the end of the backend file. If the
1601 * backend file is later extended, QEMU will not be able to find
1602 * those labels. Therefore, extending the non-empty backend file
1603 * is disabled as well.
Marcelo Tosattic9027602010-03-01 20:25:08 -03001604 */
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001605 if (truncate && ftruncate(fd, memory)) {
Yoshiaki Tamura9742bf22010-08-18 13:30:13 +09001606 perror("ftruncate");
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001607 }
Marcelo Tosattic9027602010-03-01 20:25:08 -03001608
Dominik Dingeld2f39ad2016-04-25 13:55:38 +02001609 area = qemu_ram_mmap(fd, memory, block->mr->align,
1610 block->flags & RAM_SHARED);
Marcelo Tosattic9027602010-03-01 20:25:08 -03001611 if (area == MAP_FAILED) {
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001612 error_setg_errno(errp, errno,
Markus Armbrusterfd97fd42016-03-07 20:25:13 +01001613 "unable to map backing store for guest RAM");
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001614 return NULL;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001615 }
Marcelo Tosattief36fa12013-10-28 18:51:46 -02001616
1617 if (mem_prealloc) {
Jitendra Kolhe1e356fc2017-02-24 09:01:43 +05301618 os_mem_prealloc(fd, area, memory, smp_cpus, errp);
Igor Mammedov056b68a2016-07-20 11:54:03 +02001619 if (errp && *errp) {
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001620 qemu_ram_munmap(area, memory);
1621 return NULL;
Igor Mammedov056b68a2016-07-20 11:54:03 +02001622 }
Marcelo Tosattief36fa12013-10-28 18:51:46 -02001623 }
1624
Alex Williamson04b16652010-07-02 11:13:17 -06001625 block->fd = fd;
Marcelo Tosattic9027602010-03-01 20:25:08 -03001626 return area;
1627}
1628#endif
1629
Mike Day0dc3f442013-09-05 14:41:35 -04001630/* Called with the ramlist lock held. */
Alex Williamsond17b5282010-06-25 11:08:38 -06001631static ram_addr_t find_ram_offset(ram_addr_t size)
1632{
Alex Williamson04b16652010-07-02 11:13:17 -06001633 RAMBlock *block, *next_block;
Alex Williamson3e837b22011-10-31 08:54:09 -06001634 ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX;
Alex Williamson04b16652010-07-02 11:13:17 -06001635
Stefan Hajnoczi49cd9ac2013-03-11 10:20:21 +01001636 assert(size != 0); /* it would hand out same offset multiple times */
1637
Mike Day0dc3f442013-09-05 14:41:35 -04001638 if (QLIST_EMPTY_RCU(&ram_list.blocks)) {
Alex Williamson04b16652010-07-02 11:13:17 -06001639 return 0;
Mike Day0d53d9f2015-01-21 13:45:24 +01001640 }
Alex Williamson04b16652010-07-02 11:13:17 -06001641
Peter Xu99e15582017-05-12 12:17:39 +08001642 RAMBLOCK_FOREACH(block) {
Anthony PERARDf15fbc42011-07-20 08:17:42 +00001643 ram_addr_t end, next = RAM_ADDR_MAX;
Alex Williamson04b16652010-07-02 11:13:17 -06001644
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001645 end = block->offset + block->max_length;
Alex Williamson04b16652010-07-02 11:13:17 -06001646
Peter Xu99e15582017-05-12 12:17:39 +08001647 RAMBLOCK_FOREACH(next_block) {
Alex Williamson04b16652010-07-02 11:13:17 -06001648 if (next_block->offset >= end) {
1649 next = MIN(next, next_block->offset);
1650 }
1651 }
1652 if (next - end >= size && next - end < mingap) {
Alex Williamson3e837b22011-10-31 08:54:09 -06001653 offset = end;
Alex Williamson04b16652010-07-02 11:13:17 -06001654 mingap = next - end;
1655 }
1656 }
Alex Williamson3e837b22011-10-31 08:54:09 -06001657
1658 if (offset == RAM_ADDR_MAX) {
1659 fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n",
1660 (uint64_t)size);
1661 abort();
1662 }
1663
Alex Williamson04b16652010-07-02 11:13:17 -06001664 return offset;
1665}
1666
Juan Quintelab8c48992017-03-21 17:44:30 +01001667unsigned long last_ram_page(void)
Alex Williamson04b16652010-07-02 11:13:17 -06001668{
Alex Williamsond17b5282010-06-25 11:08:38 -06001669 RAMBlock *block;
1670 ram_addr_t last = 0;
1671
Mike Day0dc3f442013-09-05 14:41:35 -04001672 rcu_read_lock();
Peter Xu99e15582017-05-12 12:17:39 +08001673 RAMBLOCK_FOREACH(block) {
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001674 last = MAX(last, block->offset + block->max_length);
Mike Day0d53d9f2015-01-21 13:45:24 +01001675 }
Mike Day0dc3f442013-09-05 14:41:35 -04001676 rcu_read_unlock();
Juan Quintelab8c48992017-03-21 17:44:30 +01001677 return last >> TARGET_PAGE_BITS;
Alex Williamsond17b5282010-06-25 11:08:38 -06001678}
1679
Jason Baronddb97f12012-08-02 15:44:16 -04001680static void qemu_ram_setup_dump(void *addr, ram_addr_t size)
1681{
1682 int ret;
Jason Baronddb97f12012-08-02 15:44:16 -04001683
1684 /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */
Marcel Apfelbaum47c8ca52015-02-04 17:43:54 +02001685 if (!machine_dump_guest_core(current_machine)) {
Jason Baronddb97f12012-08-02 15:44:16 -04001686 ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP);
1687 if (ret) {
1688 perror("qemu_madvise");
1689 fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, "
1690 "but dump_guest_core=off specified\n");
1691 }
1692 }
1693}
1694
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00001695const char *qemu_ram_get_idstr(RAMBlock *rb)
1696{
1697 return rb->idstr;
1698}
1699
Dr. David Alan Gilbert463a4ac2017-03-07 18:36:36 +00001700bool qemu_ram_is_shared(RAMBlock *rb)
1701{
1702 return rb->flags & RAM_SHARED;
1703}
1704
Mike Dayae3a7042013-09-05 14:41:35 -04001705/* Called with iothread lock held. */
Gongleifa53a0e2016-05-10 10:04:59 +08001706void qemu_ram_set_idstr(RAMBlock *new_block, const char *name, DeviceState *dev)
Hu Tao20cfe882014-04-02 15:13:26 +08001707{
Gongleifa53a0e2016-05-10 10:04:59 +08001708 RAMBlock *block;
Hu Tao20cfe882014-04-02 15:13:26 +08001709
Avi Kivityc5705a72011-12-20 15:59:12 +02001710 assert(new_block);
1711 assert(!new_block->idstr[0]);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001712
Anthony Liguori09e5ab62012-02-03 12:28:43 -06001713 if (dev) {
1714 char *id = qdev_get_dev_path(dev);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001715 if (id) {
1716 snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id);
Anthony Liguori7267c092011-08-20 22:09:37 -05001717 g_free(id);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001718 }
1719 }
1720 pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
1721
Gongleiab0a9952016-05-10 10:05:00 +08001722 rcu_read_lock();
Peter Xu99e15582017-05-12 12:17:39 +08001723 RAMBLOCK_FOREACH(block) {
Gongleifa53a0e2016-05-10 10:04:59 +08001724 if (block != new_block &&
1725 !strcmp(block->idstr, new_block->idstr)) {
Cam Macdonell84b89d72010-07-26 18:10:57 -06001726 fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
1727 new_block->idstr);
1728 abort();
1729 }
1730 }
Mike Day0dc3f442013-09-05 14:41:35 -04001731 rcu_read_unlock();
Avi Kivityc5705a72011-12-20 15:59:12 +02001732}
1733
Mike Dayae3a7042013-09-05 14:41:35 -04001734/* Called with iothread lock held. */
Gongleifa53a0e2016-05-10 10:04:59 +08001735void qemu_ram_unset_idstr(RAMBlock *block)
Hu Tao20cfe882014-04-02 15:13:26 +08001736{
Mike Dayae3a7042013-09-05 14:41:35 -04001737 /* FIXME: arch_init.c assumes that this is not called throughout
1738 * migration. Ignore the problem since hot-unplug during migration
1739 * does not work anyway.
1740 */
Hu Tao20cfe882014-04-02 15:13:26 +08001741 if (block) {
1742 memset(block->idstr, 0, sizeof(block->idstr));
1743 }
1744}
1745
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01001746size_t qemu_ram_pagesize(RAMBlock *rb)
1747{
1748 return rb->page_size;
1749}
1750
Dr. David Alan Gilbert67f11b52017-02-24 18:28:34 +00001751/* Returns the largest size of page in use */
1752size_t qemu_ram_pagesize_largest(void)
1753{
1754 RAMBlock *block;
1755 size_t largest = 0;
1756
Peter Xu99e15582017-05-12 12:17:39 +08001757 RAMBLOCK_FOREACH(block) {
Dr. David Alan Gilbert67f11b52017-02-24 18:28:34 +00001758 largest = MAX(largest, qemu_ram_pagesize(block));
1759 }
1760
1761 return largest;
1762}
1763
Luiz Capitulino8490fc72012-09-05 16:50:16 -03001764static int memory_try_enable_merging(void *addr, size_t len)
1765{
Marcel Apfelbaum75cc7f02015-02-04 17:43:55 +02001766 if (!machine_mem_merge(current_machine)) {
Luiz Capitulino8490fc72012-09-05 16:50:16 -03001767 /* disabled by the user */
1768 return 0;
1769 }
1770
1771 return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE);
1772}
1773
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001774/* Only legal before guest might have detected the memory size: e.g. on
1775 * incoming migration, or right after reset.
1776 *
1777 * As memory core doesn't know how is memory accessed, it is up to
1778 * resize callback to update device state and/or add assertions to detect
1779 * misuse, if necessary.
1780 */
Gongleifa53a0e2016-05-10 10:04:59 +08001781int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp)
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001782{
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001783 assert(block);
1784
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00001785 newsize = HOST_PAGE_ALIGN(newsize);
Michael S. Tsirkin129ddaf2015-02-17 10:15:30 +01001786
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001787 if (block->used_length == newsize) {
1788 return 0;
1789 }
1790
1791 if (!(block->flags & RAM_RESIZEABLE)) {
1792 error_setg_errno(errp, EINVAL,
1793 "Length mismatch: %s: 0x" RAM_ADDR_FMT
1794 " in != 0x" RAM_ADDR_FMT, block->idstr,
1795 newsize, block->used_length);
1796 return -EINVAL;
1797 }
1798
1799 if (block->max_length < newsize) {
1800 error_setg_errno(errp, EINVAL,
1801 "Length too large: %s: 0x" RAM_ADDR_FMT
1802 " > 0x" RAM_ADDR_FMT, block->idstr,
1803 newsize, block->max_length);
1804 return -EINVAL;
1805 }
1806
1807 cpu_physical_memory_clear_dirty_range(block->offset, block->used_length);
1808 block->used_length = newsize;
Paolo Bonzini58d27072015-03-23 11:56:01 +01001809 cpu_physical_memory_set_dirty_range(block->offset, block->used_length,
1810 DIRTY_CLIENTS_ALL);
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02001811 memory_region_set_size(block->mr, newsize);
1812 if (block->resized) {
1813 block->resized(block->idstr, newsize, block->host);
1814 }
1815 return 0;
1816}
1817
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001818/* Called with ram_list.mutex held */
1819static void dirty_memory_extend(ram_addr_t old_ram_size,
1820 ram_addr_t new_ram_size)
1821{
1822 ram_addr_t old_num_blocks = DIV_ROUND_UP(old_ram_size,
1823 DIRTY_MEMORY_BLOCK_SIZE);
1824 ram_addr_t new_num_blocks = DIV_ROUND_UP(new_ram_size,
1825 DIRTY_MEMORY_BLOCK_SIZE);
1826 int i;
1827
1828 /* Only need to extend if block count increased */
1829 if (new_num_blocks <= old_num_blocks) {
1830 return;
1831 }
1832
1833 for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
1834 DirtyMemoryBlocks *old_blocks;
1835 DirtyMemoryBlocks *new_blocks;
1836 int j;
1837
1838 old_blocks = atomic_rcu_read(&ram_list.dirty_memory[i]);
1839 new_blocks = g_malloc(sizeof(*new_blocks) +
1840 sizeof(new_blocks->blocks[0]) * new_num_blocks);
1841
1842 if (old_num_blocks) {
1843 memcpy(new_blocks->blocks, old_blocks->blocks,
1844 old_num_blocks * sizeof(old_blocks->blocks[0]));
1845 }
1846
1847 for (j = old_num_blocks; j < new_num_blocks; j++) {
1848 new_blocks->blocks[j] = bitmap_new(DIRTY_MEMORY_BLOCK_SIZE);
1849 }
1850
1851 atomic_rcu_set(&ram_list.dirty_memory[i], new_blocks);
1852
1853 if (old_blocks) {
1854 g_free_rcu(old_blocks, rcu);
1855 }
1856 }
1857}
1858
Fam Zheng528f46a2016-03-01 14:18:18 +08001859static void ram_block_add(RAMBlock *new_block, Error **errp)
Avi Kivityc5705a72011-12-20 15:59:12 +02001860{
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001861 RAMBlock *block;
Mike Day0d53d9f2015-01-21 13:45:24 +01001862 RAMBlock *last_block = NULL;
Juan Quintela2152f5c2013-10-08 13:52:02 +02001863 ram_addr_t old_ram_size, new_ram_size;
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001864 Error *err = NULL;
Juan Quintela2152f5c2013-10-08 13:52:02 +02001865
Juan Quintelab8c48992017-03-21 17:44:30 +01001866 old_ram_size = last_ram_page();
Avi Kivityc5705a72011-12-20 15:59:12 +02001867
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001868 qemu_mutex_lock_ramlist();
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001869 new_block->offset = find_ram_offset(new_block->max_length);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001870
1871 if (!new_block->host) {
1872 if (xen_enabled()) {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001873 xen_ram_alloc(new_block->offset, new_block->max_length,
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001874 new_block->mr, &err);
1875 if (err) {
1876 error_propagate(errp, err);
1877 qemu_mutex_unlock_ramlist();
Paolo Bonzini39c350e2016-03-09 18:14:01 +01001878 return;
Markus Armbruster37aa7a02016-01-14 16:09:39 +01001879 }
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001880 } else {
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001881 new_block->host = phys_mem_alloc(new_block->max_length,
Igor Mammedova2b257d2014-10-31 16:38:37 +00001882 &new_block->mr->align);
Markus Armbruster39228252013-07-31 15:11:11 +02001883 if (!new_block->host) {
Hu Taoef701d72014-09-09 13:27:54 +08001884 error_setg_errno(errp, errno,
1885 "cannot set up guest memory '%s'",
1886 memory_region_name(new_block->mr));
1887 qemu_mutex_unlock_ramlist();
Paolo Bonzini39c350e2016-03-09 18:14:01 +01001888 return;
Markus Armbruster39228252013-07-31 15:11:11 +02001889 }
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001890 memory_try_enable_merging(new_block->host, new_block->max_length);
Yoshiaki Tamura6977dfe2010-08-18 15:41:49 +09001891 }
1892 }
Cam Macdonell84b89d72010-07-26 18:10:57 -06001893
Li Zhijiandd631692015-07-02 20:18:06 +08001894 new_ram_size = MAX(old_ram_size,
1895 (new_block->offset + new_block->max_length) >> TARGET_PAGE_BITS);
1896 if (new_ram_size > old_ram_size) {
Stefan Hajnoczi5b82b702016-01-25 13:33:20 +00001897 dirty_memory_extend(old_ram_size, new_ram_size);
Li Zhijiandd631692015-07-02 20:18:06 +08001898 }
Mike Day0d53d9f2015-01-21 13:45:24 +01001899 /* Keep the list sorted from biggest to smallest block. Unlike QTAILQ,
1900 * QLIST (which has an RCU-friendly variant) does not have insertion at
1901 * tail, so save the last element in last_block.
1902 */
Peter Xu99e15582017-05-12 12:17:39 +08001903 RAMBLOCK_FOREACH(block) {
Mike Day0d53d9f2015-01-21 13:45:24 +01001904 last_block = block;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001905 if (block->max_length < new_block->max_length) {
Paolo Bonziniabb26d62012-11-14 16:00:51 +01001906 break;
1907 }
1908 }
1909 if (block) {
Mike Day0dc3f442013-09-05 14:41:35 -04001910 QLIST_INSERT_BEFORE_RCU(block, new_block, next);
Mike Day0d53d9f2015-01-21 13:45:24 +01001911 } else if (last_block) {
Mike Day0dc3f442013-09-05 14:41:35 -04001912 QLIST_INSERT_AFTER_RCU(last_block, new_block, next);
Mike Day0d53d9f2015-01-21 13:45:24 +01001913 } else { /* list is empty */
Mike Day0dc3f442013-09-05 14:41:35 -04001914 QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next);
Paolo Bonziniabb26d62012-11-14 16:00:51 +01001915 }
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01001916 ram_list.mru_block = NULL;
Cam Macdonell84b89d72010-07-26 18:10:57 -06001917
Mike Day0dc3f442013-09-05 14:41:35 -04001918 /* Write list before version */
1919 smp_wmb();
Umesh Deshpandef798b072011-08-18 11:41:17 -07001920 ram_list.version++;
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07001921 qemu_mutex_unlock_ramlist();
Umesh Deshpandef798b072011-08-18 11:41:17 -07001922
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001923 cpu_physical_memory_set_dirty_range(new_block->offset,
Paolo Bonzini58d27072015-03-23 11:56:01 +01001924 new_block->used_length,
1925 DIRTY_CLIENTS_ALL);
Cam Macdonell84b89d72010-07-26 18:10:57 -06001926
Paolo Bonzinia904c912015-01-21 16:18:35 +01001927 if (new_block->host) {
1928 qemu_ram_setup_dump(new_block->host, new_block->max_length);
1929 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_HUGEPAGE);
Cao jinc2cd6272016-09-12 14:34:56 +08001930 /* MADV_DONTFORK is also needed by KVM in absence of synchronous MMU */
Paolo Bonzinia904c912015-01-21 16:18:35 +01001931 qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_DONTFORK);
Paolo Bonzini0987d732016-12-21 00:31:36 +08001932 ram_block_notify_add(new_block->host, new_block->max_length);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001933 }
Cam Macdonell84b89d72010-07-26 18:10:57 -06001934}
1935
Paolo Bonzini0b183fc2014-05-14 17:43:19 +08001936#ifdef __linux__
Marc-André Lureau38b33622017-06-02 18:12:23 +04001937RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, MemoryRegion *mr,
1938 bool share, int fd,
1939 Error **errp)
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001940{
1941 RAMBlock *new_block;
Hu Taoef701d72014-09-09 13:27:54 +08001942 Error *local_err = NULL;
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001943 int64_t file_size;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001944
1945 if (xen_enabled()) {
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001946 error_setg(errp, "-mem-path not supported with Xen");
Fam Zheng528f46a2016-03-01 14:18:18 +08001947 return NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001948 }
1949
Marc-André Lureaue45e7ae2017-06-02 18:12:21 +04001950 if (kvm_enabled() && !kvm_has_sync_mmu()) {
1951 error_setg(errp,
1952 "host lacks kvm mmu notifiers, -mem-path unsupported");
1953 return NULL;
1954 }
1955
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001956 if (phys_mem_alloc != qemu_anon_ram_alloc) {
1957 /*
1958 * file_ram_alloc() needs to allocate just like
1959 * phys_mem_alloc, but we haven't bothered to provide
1960 * a hook there.
1961 */
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001962 error_setg(errp,
1963 "-mem-path not supported with this accelerator");
Fam Zheng528f46a2016-03-01 14:18:18 +08001964 return NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001965 }
1966
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00001967 size = HOST_PAGE_ALIGN(size);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001968 file_size = get_file_size(fd);
1969 if (file_size > 0 && file_size < size) {
1970 error_setg(errp, "backing store %s size 0x%" PRIx64
1971 " does not match 'size' option 0x" RAM_ADDR_FMT,
1972 mem_path, file_size, size);
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001973 return NULL;
1974 }
1975
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08001976 new_block = g_malloc0(sizeof(*new_block));
1977 new_block->mr = mr;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02001978 new_block->used_length = size;
1979 new_block->max_length = size;
Paolo Bonzinidbcb8982014-06-10 19:15:24 +08001980 new_block->flags = share ? RAM_SHARED : 0;
Marc-André Lureau8d37b032017-06-02 18:12:22 +04001981 new_block->host = file_ram_alloc(new_block, size, fd, !file_size, errp);
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001982 if (!new_block->host) {
1983 g_free(new_block);
Fam Zheng528f46a2016-03-01 14:18:18 +08001984 return NULL;
Paolo Bonzini7f56e742014-05-14 17:43:20 +08001985 }
1986
Fam Zheng528f46a2016-03-01 14:18:18 +08001987 ram_block_add(new_block, &local_err);
Hu Taoef701d72014-09-09 13:27:54 +08001988 if (local_err) {
1989 g_free(new_block);
1990 error_propagate(errp, local_err);
Fam Zheng528f46a2016-03-01 14:18:18 +08001991 return NULL;
Hu Taoef701d72014-09-09 13:27:54 +08001992 }
Fam Zheng528f46a2016-03-01 14:18:18 +08001993 return new_block;
Marc-André Lureau38b33622017-06-02 18:12:23 +04001994
1995}
1996
1997
1998RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
1999 bool share, const char *mem_path,
2000 Error **errp)
2001{
2002 int fd;
2003 bool created;
2004 RAMBlock *block;
2005
2006 fd = file_ram_open(mem_path, memory_region_name(mr), &created, errp);
2007 if (fd < 0) {
2008 return NULL;
2009 }
2010
2011 block = qemu_ram_alloc_from_fd(size, mr, share, fd, errp);
2012 if (!block) {
2013 if (created) {
2014 unlink(mem_path);
2015 }
2016 close(fd);
2017 return NULL;
2018 }
2019
2020 return block;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002021}
Paolo Bonzini0b183fc2014-05-14 17:43:19 +08002022#endif
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002023
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002024static
Fam Zheng528f46a2016-03-01 14:18:18 +08002025RAMBlock *qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size,
2026 void (*resized)(const char*,
2027 uint64_t length,
2028 void *host),
2029 void *host, bool resizeable,
2030 MemoryRegion *mr, Error **errp)
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002031{
2032 RAMBlock *new_block;
Hu Taoef701d72014-09-09 13:27:54 +08002033 Error *local_err = NULL;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002034
Dr. David Alan Gilbert4ed023c2015-11-05 18:11:16 +00002035 size = HOST_PAGE_ALIGN(size);
2036 max_size = HOST_PAGE_ALIGN(max_size);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002037 new_block = g_malloc0(sizeof(*new_block));
2038 new_block->mr = mr;
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002039 new_block->resized = resized;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002040 new_block->used_length = size;
2041 new_block->max_length = max_size;
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002042 assert(max_size >= size);
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002043 new_block->fd = -1;
Dr. David Alan Gilbert863e9622016-09-29 20:09:37 +01002044 new_block->page_size = getpagesize();
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002045 new_block->host = host;
2046 if (host) {
Paolo Bonzini7bd4f432014-05-14 17:43:22 +08002047 new_block->flags |= RAM_PREALLOC;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002048 }
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002049 if (resizeable) {
2050 new_block->flags |= RAM_RESIZEABLE;
2051 }
Fam Zheng528f46a2016-03-01 14:18:18 +08002052 ram_block_add(new_block, &local_err);
Hu Taoef701d72014-09-09 13:27:54 +08002053 if (local_err) {
2054 g_free(new_block);
2055 error_propagate(errp, local_err);
Fam Zheng528f46a2016-03-01 14:18:18 +08002056 return NULL;
Hu Taoef701d72014-09-09 13:27:54 +08002057 }
Fam Zheng528f46a2016-03-01 14:18:18 +08002058 return new_block;
Paolo Bonzinie1c57ab2014-05-14 17:43:18 +08002059}
2060
Fam Zheng528f46a2016-03-01 14:18:18 +08002061RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002062 MemoryRegion *mr, Error **errp)
2063{
2064 return qemu_ram_alloc_internal(size, size, NULL, host, false, mr, errp);
2065}
2066
Fam Zheng528f46a2016-03-01 14:18:18 +08002067RAMBlock *qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr, Error **errp)
pbrook94a6b542009-04-11 17:15:54 +00002068{
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002069 return qemu_ram_alloc_internal(size, size, NULL, NULL, false, mr, errp);
2070}
2071
Fam Zheng528f46a2016-03-01 14:18:18 +08002072RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t maxsz,
Michael S. Tsirkin62be4e32014-11-12 14:27:41 +02002073 void (*resized)(const char*,
2074 uint64_t length,
2075 void *host),
2076 MemoryRegion *mr, Error **errp)
2077{
2078 return qemu_ram_alloc_internal(size, maxsz, resized, NULL, true, mr, errp);
pbrook94a6b542009-04-11 17:15:54 +00002079}
bellarde9a1ab12007-02-08 23:08:38 +00002080
Paolo Bonzini43771532013-09-09 17:58:40 +02002081static void reclaim_ramblock(RAMBlock *block)
2082{
2083 if (block->flags & RAM_PREALLOC) {
2084 ;
2085 } else if (xen_enabled()) {
2086 xen_invalidate_map_cache_entry(block->host);
2087#ifndef _WIN32
2088 } else if (block->fd >= 0) {
Eduardo Habkost2f3a2bb2015-11-06 20:11:21 -02002089 qemu_ram_munmap(block->host, block->max_length);
Paolo Bonzini43771532013-09-09 17:58:40 +02002090 close(block->fd);
2091#endif
2092 } else {
2093 qemu_anon_ram_free(block->host, block->max_length);
2094 }
2095 g_free(block);
2096}
2097
Fam Zhengf1060c52016-03-01 14:18:22 +08002098void qemu_ram_free(RAMBlock *block)
bellarde9a1ab12007-02-08 23:08:38 +00002099{
Marc-André Lureau85bc2a12016-03-29 13:20:51 +02002100 if (!block) {
2101 return;
2102 }
2103
Paolo Bonzini0987d732016-12-21 00:31:36 +08002104 if (block->host) {
2105 ram_block_notify_remove(block->host, block->max_length);
2106 }
2107
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07002108 qemu_mutex_lock_ramlist();
Fam Zhengf1060c52016-03-01 14:18:22 +08002109 QLIST_REMOVE_RCU(block, next);
2110 ram_list.mru_block = NULL;
2111 /* Write list before version */
2112 smp_wmb();
2113 ram_list.version++;
2114 call_rcu(block, reclaim_ramblock, rcu);
Umesh Deshpandeb2a86582011-08-17 00:01:33 -07002115 qemu_mutex_unlock_ramlist();
bellarde9a1ab12007-02-08 23:08:38 +00002116}
2117
Huang Yingcd19cfa2011-03-02 08:56:19 +01002118#ifndef _WIN32
2119void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
2120{
2121 RAMBlock *block;
2122 ram_addr_t offset;
2123 int flags;
2124 void *area, *vaddr;
2125
Peter Xu99e15582017-05-12 12:17:39 +08002126 RAMBLOCK_FOREACH(block) {
Huang Yingcd19cfa2011-03-02 08:56:19 +01002127 offset = addr - block->offset;
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002128 if (offset < block->max_length) {
Michael S. Tsirkin1240be22014-11-12 11:44:41 +02002129 vaddr = ramblock_ptr(block, offset);
Paolo Bonzini7bd4f432014-05-14 17:43:22 +08002130 if (block->flags & RAM_PREALLOC) {
Huang Yingcd19cfa2011-03-02 08:56:19 +01002131 ;
Markus Armbrusterdfeaf2a2013-07-31 15:11:05 +02002132 } else if (xen_enabled()) {
2133 abort();
Huang Yingcd19cfa2011-03-02 08:56:19 +01002134 } else {
2135 flags = MAP_FIXED;
Markus Armbruster3435f392013-07-31 15:11:07 +02002136 if (block->fd >= 0) {
Paolo Bonzinidbcb8982014-06-10 19:15:24 +08002137 flags |= (block->flags & RAM_SHARED ?
2138 MAP_SHARED : MAP_PRIVATE);
Markus Armbruster3435f392013-07-31 15:11:07 +02002139 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2140 flags, block->fd, offset);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002141 } else {
Markus Armbruster2eb9fba2013-07-31 15:11:09 +02002142 /*
2143 * Remap needs to match alloc. Accelerators that
2144 * set phys_mem_alloc never remap. If they did,
2145 * we'd need a remap hook here.
2146 */
2147 assert(phys_mem_alloc == qemu_anon_ram_alloc);
2148
Huang Yingcd19cfa2011-03-02 08:56:19 +01002149 flags |= MAP_PRIVATE | MAP_ANONYMOUS;
2150 area = mmap(vaddr, length, PROT_READ | PROT_WRITE,
2151 flags, -1, 0);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002152 }
2153 if (area != vaddr) {
Anthony PERARDf15fbc42011-07-20 08:17:42 +00002154 fprintf(stderr, "Could not remap addr: "
2155 RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n",
Huang Yingcd19cfa2011-03-02 08:56:19 +01002156 length, addr);
2157 exit(1);
2158 }
Luiz Capitulino8490fc72012-09-05 16:50:16 -03002159 memory_try_enable_merging(vaddr, length);
Jason Baronddb97f12012-08-02 15:44:16 -04002160 qemu_ram_setup_dump(vaddr, length);
Huang Yingcd19cfa2011-03-02 08:56:19 +01002161 }
Huang Yingcd19cfa2011-03-02 08:56:19 +01002162 }
2163 }
2164}
2165#endif /* !_WIN32 */
2166
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002167/* Return a host pointer to ram allocated with qemu_ram_alloc.
Mike Dayae3a7042013-09-05 14:41:35 -04002168 * This should not be used for general purpose DMA. Use address_space_map
2169 * or address_space_rw instead. For local memory (e.g. video ram) that the
2170 * device owns, use memory_region_get_ram_ptr.
Mike Day0dc3f442013-09-05 14:41:35 -04002171 *
Paolo Bonzini49b24af2015-12-16 10:30:47 +01002172 * Called within RCU critical section.
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002173 */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002174void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr)
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002175{
Gonglei3655cb92016-02-20 10:35:20 +08002176 RAMBlock *block = ram_block;
2177
2178 if (block == NULL) {
2179 block = qemu_get_ram_block(addr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002180 addr -= block->offset;
Gonglei3655cb92016-02-20 10:35:20 +08002181 }
Mike Dayae3a7042013-09-05 14:41:35 -04002182
2183 if (xen_enabled() && block->host == NULL) {
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002184 /* We need to check if the requested address is in the RAM
2185 * because we don't want to map the entire memory in QEMU.
2186 * In that case just map until the end of the page.
2187 */
2188 if (block->offset == 0) {
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002189 return xen_map_cache(addr, 0, 0, false);
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002190 }
Mike Dayae3a7042013-09-05 14:41:35 -04002191
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002192 block->host = xen_map_cache(block->offset, block->max_length, 1, false);
Paolo Bonzini0d6d3c82012-11-14 15:45:02 +01002193 }
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002194 return ramblock_ptr(block, addr);
pbrookdc828ca2009-04-09 22:21:07 +00002195}
2196
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002197/* Return a host pointer to guest's ram. Similar to qemu_map_ram_ptr
Mike Dayae3a7042013-09-05 14:41:35 -04002198 * but takes a size argument.
Mike Day0dc3f442013-09-05 14:41:35 -04002199 *
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002200 * Called within RCU critical section.
Mike Dayae3a7042013-09-05 14:41:35 -04002201 */
Gonglei3655cb92016-02-20 10:35:20 +08002202static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr,
2203 hwaddr *size)
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002204{
Gonglei3655cb92016-02-20 10:35:20 +08002205 RAMBlock *block = ram_block;
Stefano Stabellini8ab934f2011-06-27 18:26:06 +01002206 if (*size == 0) {
2207 return NULL;
2208 }
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002209
Gonglei3655cb92016-02-20 10:35:20 +08002210 if (block == NULL) {
2211 block = qemu_get_ram_block(addr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002212 addr -= block->offset;
Gonglei3655cb92016-02-20 10:35:20 +08002213 }
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002214 *size = MIN(*size, block->max_length - addr);
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002215
2216 if (xen_enabled() && block->host == NULL) {
2217 /* We need to check if the requested address is in the RAM
2218 * because we don't want to map the entire memory in QEMU.
2219 * In that case just map the requested area.
2220 */
2221 if (block->offset == 0) {
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002222 return xen_map_cache(addr, *size, 1, true);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002223 }
2224
Stefano Stabellini1ff7c592017-05-03 14:00:35 -07002225 block->host = xen_map_cache(block->offset, block->max_length, 1, true);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002226 }
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01002227
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002228 return ramblock_ptr(block, addr);
Stefano Stabellini38bee5d2011-05-19 18:35:45 +01002229}
2230
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002231/*
2232 * Translates a host ptr back to a RAMBlock, a ram_addr and an offset
2233 * in that RAMBlock.
2234 *
2235 * ptr: Host pointer to look up
2236 * round_offset: If true round the result offset down to a page boundary
2237 * *ram_addr: set to result ram_addr
2238 * *offset: set to result offset within the RAMBlock
2239 *
2240 * Returns: RAMBlock (or NULL if not found)
Mike Dayae3a7042013-09-05 14:41:35 -04002241 *
2242 * By the time this function returns, the returned pointer is not protected
2243 * by RCU anymore. If the caller is not within an RCU critical section and
2244 * does not hold the iothread lock, it must have other means of protecting the
2245 * pointer, such as a reference to the region that includes the incoming
2246 * ram_addr_t.
2247 */
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002248RAMBlock *qemu_ram_block_from_host(void *ptr, bool round_offset,
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002249 ram_addr_t *offset)
pbrook5579c7f2009-04-11 14:47:08 +00002250{
pbrook94a6b542009-04-11 17:15:54 +00002251 RAMBlock *block;
2252 uint8_t *host = ptr;
2253
Jan Kiszka868bb332011-06-21 22:59:09 +02002254 if (xen_enabled()) {
Paolo Bonzinif615f392016-05-26 10:07:50 +02002255 ram_addr_t ram_addr;
Mike Day0dc3f442013-09-05 14:41:35 -04002256 rcu_read_lock();
Paolo Bonzinif615f392016-05-26 10:07:50 +02002257 ram_addr = xen_ram_addr_from_mapcache(ptr);
2258 block = qemu_get_ram_block(ram_addr);
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002259 if (block) {
Anthony PERARDd6b6aec2016-06-09 16:56:17 +01002260 *offset = ram_addr - block->offset;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002261 }
Mike Day0dc3f442013-09-05 14:41:35 -04002262 rcu_read_unlock();
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002263 return block;
Stefano Stabellini712c2b42011-05-19 18:35:46 +01002264 }
2265
Mike Day0dc3f442013-09-05 14:41:35 -04002266 rcu_read_lock();
2267 block = atomic_rcu_read(&ram_list.mru_block);
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002268 if (block && block->host && host - block->host < block->max_length) {
Paolo Bonzini23887b72013-05-06 14:28:39 +02002269 goto found;
2270 }
2271
Peter Xu99e15582017-05-12 12:17:39 +08002272 RAMBLOCK_FOREACH(block) {
Jun Nakajima432d2682010-08-31 16:41:25 +01002273 /* This case append when the block is not mapped. */
2274 if (block->host == NULL) {
2275 continue;
2276 }
Michael S. Tsirkin9b8424d2014-12-15 22:55:32 +02002277 if (host - block->host < block->max_length) {
Paolo Bonzini23887b72013-05-06 14:28:39 +02002278 goto found;
Alex Williamsonf471a172010-06-11 11:11:42 -06002279 }
pbrook94a6b542009-04-11 17:15:54 +00002280 }
Jun Nakajima432d2682010-08-31 16:41:25 +01002281
Mike Day0dc3f442013-09-05 14:41:35 -04002282 rcu_read_unlock();
Paolo Bonzini1b5ec232013-05-06 14:36:15 +02002283 return NULL;
Paolo Bonzini23887b72013-05-06 14:28:39 +02002284
2285found:
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002286 *offset = (host - block->host);
2287 if (round_offset) {
2288 *offset &= TARGET_PAGE_MASK;
2289 }
Mike Day0dc3f442013-09-05 14:41:35 -04002290 rcu_read_unlock();
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002291 return block;
2292}
2293
Dr. David Alan Gilberte3dd7492015-11-05 18:10:33 +00002294/*
2295 * Finds the named RAMBlock
2296 *
2297 * name: The name of RAMBlock to find
2298 *
2299 * Returns: RAMBlock (or NULL if not found)
2300 */
2301RAMBlock *qemu_ram_block_by_name(const char *name)
2302{
2303 RAMBlock *block;
2304
Peter Xu99e15582017-05-12 12:17:39 +08002305 RAMBLOCK_FOREACH(block) {
Dr. David Alan Gilberte3dd7492015-11-05 18:10:33 +00002306 if (!strcmp(name, block->idstr)) {
2307 return block;
2308 }
2309 }
2310
2311 return NULL;
2312}
2313
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002314/* Some of the softmmu routines need to translate from a host pointer
2315 (typically a TLB entry) back to a ram offset. */
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002316ram_addr_t qemu_ram_addr_from_host(void *ptr)
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002317{
2318 RAMBlock *block;
Paolo Bonzinif615f392016-05-26 10:07:50 +02002319 ram_addr_t offset;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002320
Paolo Bonzinif615f392016-05-26 10:07:50 +02002321 block = qemu_ram_block_from_host(ptr, false, &offset);
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002322 if (!block) {
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002323 return RAM_ADDR_INVALID;
Dr. David Alan Gilbert422148d2015-11-05 18:10:32 +00002324 }
2325
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01002326 return block->offset + offset;
Marcelo Tosattie8902612010-10-11 15:31:19 -03002327}
Alex Williamsonf471a172010-06-11 11:11:42 -06002328
Paolo Bonzini49b24af2015-12-16 10:30:47 +01002329/* Called within RCU critical section. */
Avi Kivitya8170e52012-10-23 12:30:10 +02002330static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002331 uint64_t val, unsigned size)
bellard1ccde1c2004-02-06 19:46:14 +00002332{
Alex Bennéeba051fb2016-10-27 16:10:16 +01002333 bool locked = false;
2334
Paolo Bonzini5aa1ef72017-07-03 17:50:40 +02002335 assert(tcg_enabled());
Juan Quintela52159192013-10-08 12:44:04 +02002336 if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) {
Alex Bennéeba051fb2016-10-27 16:10:16 +01002337 locked = true;
2338 tb_lock();
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002339 tb_invalidate_phys_page_fast(ram_addr, size);
bellard3a7d9292005-08-21 09:26:42 +00002340 }
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002341 switch (size) {
2342 case 1:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002343 stb_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002344 break;
2345 case 2:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002346 stw_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002347 break;
2348 case 4:
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002349 stl_p(qemu_map_ram_ptr(NULL, ram_addr), val);
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002350 break;
2351 default:
2352 abort();
2353 }
Alex Bennéeba051fb2016-10-27 16:10:16 +01002354
2355 if (locked) {
2356 tb_unlock();
2357 }
2358
Paolo Bonzini58d27072015-03-23 11:56:01 +01002359 /* Set both VGA and migration bits for simplicity and to remove
2360 * the notdirty callback faster.
2361 */
2362 cpu_physical_memory_set_dirty_range(ram_addr, size,
2363 DIRTY_CLIENTS_NOCODE);
bellardf23db162005-08-21 19:12:28 +00002364 /* we remove the notdirty callback only if the code has been
2365 flushed */
Juan Quintelaa2cd8c82013-10-10 11:20:22 +02002366 if (!cpu_physical_memory_is_clean(ram_addr)) {
Peter Crosthwaitebcae01e2015-09-10 22:39:42 -07002367 tlb_set_dirty(current_cpu, current_cpu->mem_io_vaddr);
Andreas Färber4917cf42013-05-27 05:17:50 +02002368 }
bellard1ccde1c2004-02-06 19:46:14 +00002369}
2370
Paolo Bonzinib018ddf2013-05-24 14:48:38 +02002371static bool notdirty_mem_accepts(void *opaque, hwaddr addr,
2372 unsigned size, bool is_write)
2373{
2374 return is_write;
2375}
2376
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002377static const MemoryRegionOps notdirty_mem_ops = {
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002378 .write = notdirty_mem_write,
Paolo Bonzinib018ddf2013-05-24 14:48:38 +02002379 .valid.accepts = notdirty_mem_accepts,
Avi Kivity0e0df1e2012-01-02 00:32:15 +02002380 .endianness = DEVICE_NATIVE_ENDIAN,
bellard1ccde1c2004-02-06 19:46:14 +00002381};
2382
pbrook0f459d12008-06-09 00:20:13 +00002383/* Generate a debug exception if a watchpoint has been hit. */
Peter Maydell66b9b432015-04-26 16:49:24 +01002384static void check_watchpoint(int offset, int len, MemTxAttrs attrs, int flags)
pbrook0f459d12008-06-09 00:20:13 +00002385{
Andreas Färber93afead2013-08-26 03:41:01 +02002386 CPUState *cpu = current_cpu;
Sergey Fedorov568496c2016-02-11 11:17:32 +00002387 CPUClass *cc = CPU_GET_CLASS(cpu);
Andreas Färber93afead2013-08-26 03:41:01 +02002388 CPUArchState *env = cpu->env_ptr;
aliguori06d55cc2008-11-18 20:24:06 +00002389 target_ulong pc, cs_base;
pbrook0f459d12008-06-09 00:20:13 +00002390 target_ulong vaddr;
aliguoria1d1bb32008-11-18 20:07:32 +00002391 CPUWatchpoint *wp;
Emilio G. Cota89fee742016-04-07 13:19:22 -04002392 uint32_t cpu_flags;
pbrook0f459d12008-06-09 00:20:13 +00002393
Paolo Bonzini5aa1ef72017-07-03 17:50:40 +02002394 assert(tcg_enabled());
Andreas Färberff4700b2013-08-26 18:23:18 +02002395 if (cpu->watchpoint_hit) {
aliguori06d55cc2008-11-18 20:24:06 +00002396 /* We re-entered the check after replacing the TB. Now raise
2397 * the debug interrupt so that is will trigger after the
2398 * current instruction. */
Andreas Färber93afead2013-08-26 03:41:01 +02002399 cpu_interrupt(cpu, CPU_INTERRUPT_DEBUG);
aliguori06d55cc2008-11-18 20:24:06 +00002400 return;
2401 }
Andreas Färber93afead2013-08-26 03:41:01 +02002402 vaddr = (cpu->mem_io_vaddr & TARGET_PAGE_MASK) + offset;
Julian Brown40612002017-02-07 18:29:59 +00002403 vaddr = cc->adjust_watchpoint_address(cpu, vaddr, len);
Andreas Färberff4700b2013-08-26 18:23:18 +02002404 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
Peter Maydell05068c02014-09-12 14:06:48 +01002405 if (cpu_watchpoint_address_matches(wp, vaddr, len)
2406 && (wp->flags & flags)) {
Peter Maydell08225672014-09-12 14:06:48 +01002407 if (flags == BP_MEM_READ) {
2408 wp->flags |= BP_WATCHPOINT_HIT_READ;
2409 } else {
2410 wp->flags |= BP_WATCHPOINT_HIT_WRITE;
2411 }
2412 wp->hitaddr = vaddr;
Peter Maydell66b9b432015-04-26 16:49:24 +01002413 wp->hitattrs = attrs;
Andreas Färberff4700b2013-08-26 18:23:18 +02002414 if (!cpu->watchpoint_hit) {
Sergey Fedorov568496c2016-02-11 11:17:32 +00002415 if (wp->flags & BP_CPU &&
2416 !cc->debug_check_watchpoint(cpu, wp)) {
2417 wp->flags &= ~BP_WATCHPOINT_HIT;
2418 continue;
2419 }
Andreas Färberff4700b2013-08-26 18:23:18 +02002420 cpu->watchpoint_hit = wp;
KONRAD Frederica5e99822016-10-27 16:10:06 +01002421
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002422 /* Both tb_lock and iothread_mutex will be reset when
2423 * cpu_loop_exit or cpu_loop_exit_noexc longjmp
2424 * back into the cpu_exec main loop.
KONRAD Frederica5e99822016-10-27 16:10:06 +01002425 */
2426 tb_lock();
Andreas Färber239c51a2013-09-01 17:12:23 +02002427 tb_check_watchpoint(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002428 if (wp->flags & BP_STOP_BEFORE_ACCESS) {
Andreas Färber27103422013-08-26 08:31:06 +02002429 cpu->exception_index = EXCP_DEBUG;
Andreas Färber5638d182013-08-27 17:52:12 +02002430 cpu_loop_exit(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002431 } else {
2432 cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags);
Andreas Färber648f0342013-09-01 17:43:17 +02002433 tb_gen_code(cpu, pc, cs_base, cpu_flags, 1);
Peter Maydell6886b982016-05-17 15:18:04 +01002434 cpu_loop_exit_noexc(cpu);
aliguori6e140f22008-11-18 20:37:55 +00002435 }
aliguori06d55cc2008-11-18 20:24:06 +00002436 }
aliguori6e140f22008-11-18 20:37:55 +00002437 } else {
2438 wp->flags &= ~BP_WATCHPOINT_HIT;
pbrook0f459d12008-06-09 00:20:13 +00002439 }
2440 }
2441}
2442
pbrook6658ffb2007-03-16 23:58:11 +00002443/* Watchpoint access routines. Watchpoints are inserted using TLB tricks,
2444 so these check for a hit then pass through to the normal out-of-line
2445 phys routines. */
Peter Maydell66b9b432015-04-26 16:49:24 +01002446static MemTxResult watch_mem_read(void *opaque, hwaddr addr, uint64_t *pdata,
2447 unsigned size, MemTxAttrs attrs)
pbrook6658ffb2007-03-16 23:58:11 +00002448{
Peter Maydell66b9b432015-04-26 16:49:24 +01002449 MemTxResult res;
2450 uint64_t data;
Peter Maydell79ed0412016-01-21 14:15:06 +00002451 int asidx = cpu_asidx_from_attrs(current_cpu, attrs);
2452 AddressSpace *as = current_cpu->cpu_ases[asidx].as;
pbrook6658ffb2007-03-16 23:58:11 +00002453
Peter Maydell66b9b432015-04-26 16:49:24 +01002454 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_READ);
Avi Kivity1ec9b902012-01-02 12:47:48 +02002455 switch (size) {
Max Filippov67364152012-01-29 00:01:40 +04002456 case 1:
Peter Maydell79ed0412016-01-21 14:15:06 +00002457 data = address_space_ldub(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002458 break;
2459 case 2:
Peter Maydell79ed0412016-01-21 14:15:06 +00002460 data = address_space_lduw(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002461 break;
2462 case 4:
Peter Maydell79ed0412016-01-21 14:15:06 +00002463 data = address_space_ldl(as, addr, attrs, &res);
Max Filippov67364152012-01-29 00:01:40 +04002464 break;
Avi Kivity1ec9b902012-01-02 12:47:48 +02002465 default: abort();
2466 }
Peter Maydell66b9b432015-04-26 16:49:24 +01002467 *pdata = data;
2468 return res;
2469}
2470
2471static MemTxResult watch_mem_write(void *opaque, hwaddr addr,
2472 uint64_t val, unsigned size,
2473 MemTxAttrs attrs)
2474{
2475 MemTxResult res;
Peter Maydell79ed0412016-01-21 14:15:06 +00002476 int asidx = cpu_asidx_from_attrs(current_cpu, attrs);
2477 AddressSpace *as = current_cpu->cpu_ases[asidx].as;
Peter Maydell66b9b432015-04-26 16:49:24 +01002478
2479 check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_WRITE);
2480 switch (size) {
2481 case 1:
Peter Maydell79ed0412016-01-21 14:15:06 +00002482 address_space_stb(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002483 break;
2484 case 2:
Peter Maydell79ed0412016-01-21 14:15:06 +00002485 address_space_stw(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002486 break;
2487 case 4:
Peter Maydell79ed0412016-01-21 14:15:06 +00002488 address_space_stl(as, addr, val, attrs, &res);
Peter Maydell66b9b432015-04-26 16:49:24 +01002489 break;
2490 default: abort();
2491 }
2492 return res;
pbrook6658ffb2007-03-16 23:58:11 +00002493}
2494
Avi Kivity1ec9b902012-01-02 12:47:48 +02002495static const MemoryRegionOps watch_mem_ops = {
Peter Maydell66b9b432015-04-26 16:49:24 +01002496 .read_with_attrs = watch_mem_read,
2497 .write_with_attrs = watch_mem_write,
Avi Kivity1ec9b902012-01-02 12:47:48 +02002498 .endianness = DEVICE_NATIVE_ENDIAN,
pbrook6658ffb2007-03-16 23:58:11 +00002499};
pbrook6658ffb2007-03-16 23:58:11 +00002500
Peter Maydellf25a49e2015-04-26 16:49:24 +01002501static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data,
2502 unsigned len, MemTxAttrs attrs)
blueswir1db7b5422007-05-26 17:36:03 +00002503{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002504 subpage_t *subpage = opaque;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002505 uint8_t buf[8];
Peter Maydell5c9eb022015-04-26 16:49:24 +01002506 MemTxResult res;
Paolo Bonzini791af8c2013-05-24 16:10:39 +02002507
blueswir1db7b5422007-05-26 17:36:03 +00002508#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002509 printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__,
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002510 subpage, len, addr);
blueswir1db7b5422007-05-26 17:36:03 +00002511#endif
Peter Maydell5c9eb022015-04-26 16:49:24 +01002512 res = address_space_read(subpage->as, addr + subpage->base,
2513 attrs, buf, len);
2514 if (res) {
2515 return res;
Peter Maydellf25a49e2015-04-26 16:49:24 +01002516 }
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002517 switch (len) {
2518 case 1:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002519 *data = ldub_p(buf);
2520 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002521 case 2:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002522 *data = lduw_p(buf);
2523 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002524 case 4:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002525 *data = ldl_p(buf);
2526 return MEMTX_OK;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002527 case 8:
Peter Maydellf25a49e2015-04-26 16:49:24 +01002528 *data = ldq_p(buf);
2529 return MEMTX_OK;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002530 default:
2531 abort();
2532 }
blueswir1db7b5422007-05-26 17:36:03 +00002533}
2534
Peter Maydellf25a49e2015-04-26 16:49:24 +01002535static MemTxResult subpage_write(void *opaque, hwaddr addr,
2536 uint64_t value, unsigned len, MemTxAttrs attrs)
blueswir1db7b5422007-05-26 17:36:03 +00002537{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002538 subpage_t *subpage = opaque;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002539 uint8_t buf[8];
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002540
blueswir1db7b5422007-05-26 17:36:03 +00002541#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002542 printf("%s: subpage %p len %u addr " TARGET_FMT_plx
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002543 " value %"PRIx64"\n",
2544 __func__, subpage, len, addr, value);
blueswir1db7b5422007-05-26 17:36:03 +00002545#endif
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002546 switch (len) {
2547 case 1:
2548 stb_p(buf, value);
2549 break;
2550 case 2:
2551 stw_p(buf, value);
2552 break;
2553 case 4:
2554 stl_p(buf, value);
2555 break;
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002556 case 8:
2557 stq_p(buf, value);
2558 break;
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002559 default:
2560 abort();
2561 }
Peter Maydell5c9eb022015-04-26 16:49:24 +01002562 return address_space_write(subpage->as, addr + subpage->base,
2563 attrs, buf, len);
blueswir1db7b5422007-05-26 17:36:03 +00002564}
2565
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002566static bool subpage_accepts(void *opaque, hwaddr addr,
Amos Kong016e9d62013-09-27 09:25:38 +08002567 unsigned len, bool is_write)
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002568{
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002569 subpage_t *subpage = opaque;
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002570#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002571 printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n",
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002572 __func__, subpage, is_write ? 'w' : 'r', len, addr);
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002573#endif
2574
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002575 return address_space_access_valid(subpage->as, addr + subpage->base,
Amos Kong016e9d62013-09-27 09:25:38 +08002576 len, is_write);
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002577}
2578
Avi Kivity70c68e42012-01-02 12:32:48 +02002579static const MemoryRegionOps subpage_ops = {
Peter Maydellf25a49e2015-04-26 16:49:24 +01002580 .read_with_attrs = subpage_read,
2581 .write_with_attrs = subpage_write,
Paolo Bonziniff6cff72014-12-22 13:11:39 +01002582 .impl.min_access_size = 1,
2583 .impl.max_access_size = 8,
2584 .valid.min_access_size = 1,
2585 .valid.max_access_size = 8,
Paolo Bonzinic353e4c2013-05-24 14:02:39 +02002586 .valid.accepts = subpage_accepts,
Avi Kivity70c68e42012-01-02 12:32:48 +02002587 .endianness = DEVICE_NATIVE_ENDIAN,
blueswir1db7b5422007-05-26 17:36:03 +00002588};
2589
Anthony Liguoric227f092009-10-01 16:12:16 -05002590static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end,
Avi Kivity5312bd82012-02-12 18:32:55 +02002591 uint16_t section)
blueswir1db7b5422007-05-26 17:36:03 +00002592{
2593 int idx, eidx;
2594
2595 if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE)
2596 return -1;
2597 idx = SUBPAGE_IDX(start);
2598 eidx = SUBPAGE_IDX(end);
2599#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002600 printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n",
2601 __func__, mmio, start, end, idx, eidx, section);
blueswir1db7b5422007-05-26 17:36:03 +00002602#endif
blueswir1db7b5422007-05-26 17:36:03 +00002603 for (; idx <= eidx; idx++) {
Avi Kivity5312bd82012-02-12 18:32:55 +02002604 mmio->sub_section[idx] = section;
blueswir1db7b5422007-05-26 17:36:03 +00002605 }
2606
2607 return 0;
2608}
2609
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002610static subpage_t *subpage_init(AddressSpace *as, hwaddr base)
blueswir1db7b5422007-05-26 17:36:03 +00002611{
Anthony Liguoric227f092009-10-01 16:12:16 -05002612 subpage_t *mmio;
blueswir1db7b5422007-05-26 17:36:03 +00002613
Vijaya Kumar K2615fab2016-10-24 16:26:49 +01002614 mmio = g_malloc0(sizeof(subpage_t) + TARGET_PAGE_SIZE * sizeof(uint16_t));
Jan Kiszkaacc9d802013-05-26 21:55:37 +02002615 mmio->as = as;
aliguori1eec6142009-02-05 22:06:18 +00002616 mmio->base = base;
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002617 memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio,
Peter Crosthwaiteb4fefef2014-06-05 23:15:52 -07002618 NULL, TARGET_PAGE_SIZE);
Avi Kivityb3b00c72012-01-02 13:20:11 +02002619 mmio->iomem.subpage = true;
blueswir1db7b5422007-05-26 17:36:03 +00002620#if defined(DEBUG_SUBPAGE)
Amos Kong016e9d62013-09-27 09:25:38 +08002621 printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__,
2622 mmio, base, TARGET_PAGE_SIZE);
blueswir1db7b5422007-05-26 17:36:03 +00002623#endif
Liu Ping Fanb41aac42013-05-29 11:09:17 +02002624 subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED);
blueswir1db7b5422007-05-26 17:36:03 +00002625
2626 return mmio;
2627}
2628
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002629static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as,
2630 MemoryRegion *mr)
Avi Kivity5312bd82012-02-12 18:32:55 +02002631{
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002632 assert(as);
Avi Kivity5312bd82012-02-12 18:32:55 +02002633 MemoryRegionSection section = {
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002634 .address_space = as,
Avi Kivity5312bd82012-02-12 18:32:55 +02002635 .mr = mr,
2636 .offset_within_address_space = 0,
2637 .offset_within_region = 0,
Paolo Bonzini052e87b2013-05-27 10:08:27 +02002638 .size = int128_2_64(),
Avi Kivity5312bd82012-02-12 18:32:55 +02002639 };
2640
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002641 return phys_section_add(map, &section);
Avi Kivity5312bd82012-02-12 18:32:55 +02002642}
2643
Peter Maydella54c87b2016-01-21 14:15:05 +00002644MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index, MemTxAttrs attrs)
Avi Kivityaa102232012-03-08 17:06:55 +02002645{
Peter Maydella54c87b2016-01-21 14:15:05 +00002646 int asidx = cpu_asidx_from_attrs(cpu, attrs);
2647 CPUAddressSpace *cpuas = &cpu->cpu_ases[asidx];
Peter Maydell32857f42015-10-01 15:29:50 +01002648 AddressSpaceDispatch *d = atomic_rcu_read(&cpuas->memory_dispatch);
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002649 MemoryRegionSection *sections = d->map.sections;
Paolo Bonzini9d82b5a2013-08-16 08:26:30 +02002650
2651 return sections[index & ~TARGET_PAGE_MASK].mr;
Avi Kivityaa102232012-03-08 17:06:55 +02002652}
2653
Avi Kivitye9179ce2009-06-14 11:38:52 +03002654static void io_mem_init(void)
2655{
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002656 memory_region_init_io(&io_mem_rom, NULL, &unassigned_mem_ops, NULL, NULL, UINT64_MAX);
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002657 memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002658 NULL, UINT64_MAX);
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002659
2660 /* io_mem_notdirty calls tb_invalidate_phys_page_fast,
2661 * which can be called without the iothread mutex.
2662 */
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002663 memory_region_init_io(&io_mem_notdirty, NULL, &notdirty_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002664 NULL, UINT64_MAX);
Jan Kiszka8d04fb52017-02-23 18:29:11 +00002665 memory_region_clear_global_locking(&io_mem_notdirty);
2666
Paolo Bonzini2c9b15c2013-06-06 05:41:28 -04002667 memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL,
Paolo Bonzini1f6245e2014-06-13 10:48:06 +02002668 NULL, UINT64_MAX);
Avi Kivitye9179ce2009-06-14 11:38:52 +03002669}
2670
Avi Kivityac1970f2012-10-03 16:22:53 +02002671static void mem_begin(MemoryListener *listener)
2672{
Paolo Bonzini89ae3372013-06-02 10:39:07 +02002673 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002674 AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1);
2675 uint16_t n;
2676
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002677 n = dummy_section(&d->map, as, &io_mem_unassigned);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002678 assert(n == PHYS_SECTION_UNASSIGNED);
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002679 n = dummy_section(&d->map, as, &io_mem_notdirty);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002680 assert(n == PHYS_SECTION_NOTDIRTY);
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002681 n = dummy_section(&d->map, as, &io_mem_rom);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002682 assert(n == PHYS_SECTION_ROM);
Peter Crosthwaitea656e222014-06-02 19:08:44 -07002683 n = dummy_section(&d->map, as, &io_mem_watch);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002684 assert(n == PHYS_SECTION_WATCH);
Paolo Bonzini00752702013-05-29 12:13:54 +02002685
Michael S. Tsirkin9736e552013-11-11 14:42:43 +02002686 d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 };
Paolo Bonzini00752702013-05-29 12:13:54 +02002687 d->as = as;
2688 as->next_dispatch = d;
2689}
2690
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002691static void address_space_dispatch_free(AddressSpaceDispatch *d)
2692{
2693 phys_sections_free(&d->map);
2694 g_free(d);
2695}
2696
Paolo Bonzini00752702013-05-29 12:13:54 +02002697static void mem_commit(MemoryListener *listener)
2698{
2699 AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
Paolo Bonzini0475d942013-05-29 12:28:21 +02002700 AddressSpaceDispatch *cur = as->dispatch;
2701 AddressSpaceDispatch *next = as->next_dispatch;
Avi Kivityac1970f2012-10-03 16:22:53 +02002702
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002703 phys_page_compact_all(next, next->map.nodes_nb);
Michael S. Tsirkinb35ba302013-11-11 17:52:07 +02002704
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002705 atomic_rcu_set(&as->dispatch, next);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002706 if (cur) {
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002707 call_rcu(cur, address_space_dispatch_free, rcu);
Marcel Apfelbaum53cb28c2013-12-01 14:02:23 +02002708 }
Paolo Bonzini9affd6f2013-05-29 12:09:47 +02002709}
2710
Avi Kivity1d711482012-10-02 18:54:45 +02002711static void tcg_commit(MemoryListener *listener)
Avi Kivity50c1e142012-02-08 21:36:02 +02002712{
Peter Maydell32857f42015-10-01 15:29:50 +01002713 CPUAddressSpace *cpuas;
2714 AddressSpaceDispatch *d;
Avi Kivity117712c2012-02-12 21:23:17 +02002715
2716 /* since each CPU stores ram addresses in its TLB cache, we must
2717 reset the modified entries */
Peter Maydell32857f42015-10-01 15:29:50 +01002718 cpuas = container_of(listener, CPUAddressSpace, tcg_as_listener);
2719 cpu_reloading_memory_map();
2720 /* The CPU and TLB are protected by the iothread lock.
2721 * We reload the dispatch pointer now because cpu_reloading_memory_map()
2722 * may have split the RCU critical section.
2723 */
2724 d = atomic_rcu_read(&cpuas->as->dispatch);
Alex Bennéef35e44e2016-10-21 16:34:18 +01002725 atomic_rcu_set(&cpuas->memory_dispatch, d);
Alex Bennéed10eb082016-11-14 14:17:28 +00002726 tlb_flush(cpuas->cpu);
Avi Kivity50c1e142012-02-08 21:36:02 +02002727}
2728
Avi Kivityac1970f2012-10-03 16:22:53 +02002729void address_space_init_dispatch(AddressSpace *as)
2730{
Paolo Bonzini00752702013-05-29 12:13:54 +02002731 as->dispatch = NULL;
Paolo Bonzini89ae3372013-06-02 10:39:07 +02002732 as->dispatch_listener = (MemoryListener) {
Avi Kivityac1970f2012-10-03 16:22:53 +02002733 .begin = mem_begin,
Paolo Bonzini00752702013-05-29 12:13:54 +02002734 .commit = mem_commit,
Avi Kivityac1970f2012-10-03 16:22:53 +02002735 .region_add = mem_add,
2736 .region_nop = mem_add,
2737 .priority = 0,
2738 };
Paolo Bonzini89ae3372013-06-02 10:39:07 +02002739 memory_listener_register(&as->dispatch_listener, as);
Avi Kivityac1970f2012-10-03 16:22:53 +02002740}
2741
Paolo Bonzini6e48e8f2015-02-10 10:25:44 -07002742void address_space_unregister(AddressSpace *as)
2743{
2744 memory_listener_unregister(&as->dispatch_listener);
2745}
2746
Avi Kivity83f3c252012-10-07 12:59:55 +02002747void address_space_destroy_dispatch(AddressSpace *as)
2748{
2749 AddressSpaceDispatch *d = as->dispatch;
2750
Paolo Bonzini79e2b9a2015-01-21 12:09:14 +01002751 atomic_rcu_set(&as->dispatch, NULL);
2752 if (d) {
2753 call_rcu(d, address_space_dispatch_free, rcu);
2754 }
Avi Kivity83f3c252012-10-07 12:59:55 +02002755}
2756
Avi Kivity62152b82011-07-26 14:26:14 +03002757static void memory_map_init(void)
2758{
Anthony Liguori7267c092011-08-20 22:09:37 -05002759 system_memory = g_malloc(sizeof(*system_memory));
Paolo Bonzini03f49952013-11-07 17:14:36 +01002760
Paolo Bonzini57271d62013-11-07 17:14:37 +01002761 memory_region_init(system_memory, NULL, "system", UINT64_MAX);
Alexey Kardashevskiy7dca8042013-04-29 16:25:51 +00002762 address_space_init(&address_space_memory, system_memory, "memory");
Avi Kivity309cb472011-08-08 16:09:03 +03002763
Anthony Liguori7267c092011-08-20 22:09:37 -05002764 system_io = g_malloc(sizeof(*system_io));
Jan Kiszka3bb28b72013-09-02 18:43:30 +02002765 memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io",
2766 65536);
Alexey Kardashevskiy7dca8042013-04-29 16:25:51 +00002767 address_space_init(&address_space_io, system_io, "I/O");
Avi Kivity62152b82011-07-26 14:26:14 +03002768}
2769
2770MemoryRegion *get_system_memory(void)
2771{
2772 return system_memory;
2773}
2774
Avi Kivity309cb472011-08-08 16:09:03 +03002775MemoryRegion *get_system_io(void)
2776{
2777 return system_io;
2778}
2779
pbrooke2eef172008-06-08 01:09:01 +00002780#endif /* !defined(CONFIG_USER_ONLY) */
2781
bellard13eb76e2004-01-24 15:23:36 +00002782/* physical memory access (slow version, mainly for debug) */
2783#if defined(CONFIG_USER_ONLY)
Andreas Färberf17ec442013-06-29 19:40:58 +02002784int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
Paul Brooka68fe892010-03-01 00:08:59 +00002785 uint8_t *buf, int len, int is_write)
bellard13eb76e2004-01-24 15:23:36 +00002786{
2787 int l, flags;
2788 target_ulong page;
pbrook53a59602006-03-25 19:31:22 +00002789 void * p;
bellard13eb76e2004-01-24 15:23:36 +00002790
2791 while (len > 0) {
2792 page = addr & TARGET_PAGE_MASK;
2793 l = (page + TARGET_PAGE_SIZE) - addr;
2794 if (l > len)
2795 l = len;
2796 flags = page_get_flags(page);
2797 if (!(flags & PAGE_VALID))
Paul Brooka68fe892010-03-01 00:08:59 +00002798 return -1;
bellard13eb76e2004-01-24 15:23:36 +00002799 if (is_write) {
2800 if (!(flags & PAGE_WRITE))
Paul Brooka68fe892010-03-01 00:08:59 +00002801 return -1;
bellard579a97f2007-11-11 14:26:47 +00002802 /* XXX: this code should not depend on lock_user */
aurel3272fb7da2008-04-27 23:53:45 +00002803 if (!(p = lock_user(VERIFY_WRITE, addr, l, 0)))
Paul Brooka68fe892010-03-01 00:08:59 +00002804 return -1;
aurel3272fb7da2008-04-27 23:53:45 +00002805 memcpy(p, buf, l);
2806 unlock_user(p, addr, l);
bellard13eb76e2004-01-24 15:23:36 +00002807 } else {
2808 if (!(flags & PAGE_READ))
Paul Brooka68fe892010-03-01 00:08:59 +00002809 return -1;
bellard579a97f2007-11-11 14:26:47 +00002810 /* XXX: this code should not depend on lock_user */
aurel3272fb7da2008-04-27 23:53:45 +00002811 if (!(p = lock_user(VERIFY_READ, addr, l, 1)))
Paul Brooka68fe892010-03-01 00:08:59 +00002812 return -1;
aurel3272fb7da2008-04-27 23:53:45 +00002813 memcpy(buf, p, l);
aurel325b257572008-04-28 08:54:59 +00002814 unlock_user(p, addr, 0);
bellard13eb76e2004-01-24 15:23:36 +00002815 }
2816 len -= l;
2817 buf += l;
2818 addr += l;
2819 }
Paul Brooka68fe892010-03-01 00:08:59 +00002820 return 0;
bellard13eb76e2004-01-24 15:23:36 +00002821}
bellard8df1cd02005-01-28 22:37:22 +00002822
bellard13eb76e2004-01-24 15:23:36 +00002823#else
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002824
Paolo Bonzini845b6212015-03-23 11:45:53 +01002825static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr,
Avi Kivitya8170e52012-10-23 12:30:10 +02002826 hwaddr length)
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002827{
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002828 uint8_t dirty_log_mask = memory_region_get_dirty_log_mask(mr);
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01002829 addr += memory_region_get_ram_addr(mr);
2830
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002831 /* No early return if dirty_log_mask is or becomes 0, because
2832 * cpu_physical_memory_set_dirty_range will still call
2833 * xen_modified_memory.
2834 */
2835 if (dirty_log_mask) {
2836 dirty_log_mask =
2837 cpu_physical_memory_range_includes_clean(addr, length, dirty_log_mask);
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002838 }
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002839 if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) {
Paolo Bonzini5aa1ef72017-07-03 17:50:40 +02002840 assert(tcg_enabled());
Alex Bennéeba051fb2016-10-27 16:10:16 +01002841 tb_lock();
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002842 tb_invalidate_phys_range(addr, addr + length);
Alex Bennéeba051fb2016-10-27 16:10:16 +01002843 tb_unlock();
Paolo Bonzinie87f7772015-03-25 15:21:39 +01002844 dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE);
2845 }
2846 cpu_physical_memory_set_dirty_range(addr, length, dirty_log_mask);
Anthony PERARD51d7a9e2012-10-03 13:49:05 +00002847}
2848
Richard Henderson23326162013-07-08 14:55:59 -07002849static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr)
Paolo Bonzini82f25632013-05-24 11:59:43 +02002850{
Paolo Bonzinie1622f42013-07-17 13:17:41 +02002851 unsigned access_size_max = mr->ops->valid.max_access_size;
Richard Henderson23326162013-07-08 14:55:59 -07002852
2853 /* Regions are assumed to support 1-4 byte accesses unless
2854 otherwise specified. */
Richard Henderson23326162013-07-08 14:55:59 -07002855 if (access_size_max == 0) {
2856 access_size_max = 4;
Paolo Bonzini82f25632013-05-24 11:59:43 +02002857 }
Richard Henderson23326162013-07-08 14:55:59 -07002858
2859 /* Bound the maximum access by the alignment of the address. */
2860 if (!mr->ops->impl.unaligned) {
2861 unsigned align_size_max = addr & -addr;
2862 if (align_size_max != 0 && align_size_max < access_size_max) {
2863 access_size_max = align_size_max;
2864 }
2865 }
2866
2867 /* Don't attempt accesses larger than the maximum. */
2868 if (l > access_size_max) {
2869 l = access_size_max;
2870 }
Peter Maydell6554f5c2015-07-24 13:33:10 +01002871 l = pow2floor(l);
Richard Henderson23326162013-07-08 14:55:59 -07002872
2873 return l;
Paolo Bonzini82f25632013-05-24 11:59:43 +02002874}
2875
Jan Kiszka4840f102015-06-18 18:47:22 +02002876static bool prepare_mmio_access(MemoryRegion *mr)
Paolo Bonzini125b3802015-06-18 18:47:21 +02002877{
Jan Kiszka4840f102015-06-18 18:47:22 +02002878 bool unlocked = !qemu_mutex_iothread_locked();
2879 bool release_lock = false;
2880
2881 if (unlocked && mr->global_locking) {
2882 qemu_mutex_lock_iothread();
2883 unlocked = false;
2884 release_lock = true;
Paolo Bonzini125b3802015-06-18 18:47:21 +02002885 }
Jan Kiszka4840f102015-06-18 18:47:22 +02002886 if (mr->flush_coalesced_mmio) {
2887 if (unlocked) {
2888 qemu_mutex_lock_iothread();
2889 }
2890 qemu_flush_coalesced_mmio_buffer();
2891 if (unlocked) {
2892 qemu_mutex_unlock_iothread();
2893 }
2894 }
2895
2896 return release_lock;
Paolo Bonzini125b3802015-06-18 18:47:21 +02002897}
2898
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002899/* Called within RCU critical section. */
2900static MemTxResult address_space_write_continue(AddressSpace *as, hwaddr addr,
2901 MemTxAttrs attrs,
2902 const uint8_t *buf,
2903 int len, hwaddr addr1,
2904 hwaddr l, MemoryRegion *mr)
bellard13eb76e2004-01-24 15:23:36 +00002905{
bellard13eb76e2004-01-24 15:23:36 +00002906 uint8_t *ptr;
Paolo Bonzini791af8c2013-05-24 16:10:39 +02002907 uint64_t val;
Peter Maydell3b643492015-04-26 16:49:23 +01002908 MemTxResult result = MEMTX_OK;
Jan Kiszka4840f102015-06-18 18:47:22 +02002909 bool release_lock = false;
ths3b46e622007-09-17 08:09:54 +00002910
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002911 for (;;) {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002912 if (!memory_access_is_direct(mr, true)) {
2913 release_lock |= prepare_mmio_access(mr);
2914 l = memory_access_size(mr, l, addr1);
2915 /* XXX: could force current_cpu to NULL to avoid
2916 potential bugs */
2917 switch (l) {
2918 case 8:
2919 /* 64 bit write access */
2920 val = ldq_p(buf);
2921 result |= memory_region_dispatch_write(mr, addr1, val, 8,
2922 attrs);
2923 break;
2924 case 4:
2925 /* 32 bit write access */
Ladi Prosek6da67de2017-01-26 15:22:37 +01002926 val = (uint32_t)ldl_p(buf);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002927 result |= memory_region_dispatch_write(mr, addr1, val, 4,
2928 attrs);
2929 break;
2930 case 2:
2931 /* 16 bit write access */
2932 val = lduw_p(buf);
2933 result |= memory_region_dispatch_write(mr, addr1, val, 2,
2934 attrs);
2935 break;
2936 case 1:
2937 /* 8 bit write access */
2938 val = ldub_p(buf);
2939 result |= memory_region_dispatch_write(mr, addr1, val, 1,
2940 attrs);
2941 break;
2942 default:
2943 abort();
bellard13eb76e2004-01-24 15:23:36 +00002944 }
2945 } else {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002946 /* RAM case */
Prasad J Pandit04bf2522017-07-12 18:08:40 +05302947 ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002948 memcpy(ptr, buf, l);
2949 invalidate_and_set_dirty(mr, addr1, l);
bellard13eb76e2004-01-24 15:23:36 +00002950 }
Jan Kiszka4840f102015-06-18 18:47:22 +02002951
2952 if (release_lock) {
2953 qemu_mutex_unlock_iothread();
2954 release_lock = false;
2955 }
2956
bellard13eb76e2004-01-24 15:23:36 +00002957 len -= l;
2958 buf += l;
2959 addr += l;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002960
2961 if (!len) {
2962 break;
2963 }
2964
2965 l = len;
2966 mr = address_space_translate(as, addr, &addr1, &l, true);
bellard13eb76e2004-01-24 15:23:36 +00002967 }
Paolo Bonzinifd8aaa72013-05-21 09:56:55 +02002968
Peter Maydell3b643492015-04-26 16:49:23 +01002969 return result;
bellard13eb76e2004-01-24 15:23:36 +00002970}
bellard8df1cd02005-01-28 22:37:22 +00002971
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002972MemTxResult address_space_write(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
2973 const uint8_t *buf, int len)
Avi Kivityac1970f2012-10-03 16:22:53 +02002974{
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002975 hwaddr l;
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01002976 hwaddr addr1;
2977 MemoryRegion *mr;
2978 MemTxResult result = MEMTX_OK;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01002979
2980 if (len > 0) {
2981 rcu_read_lock();
2982 l = len;
2983 mr = address_space_translate(as, addr, &addr1, &l, true);
2984 result = address_space_write_continue(as, addr, attrs, buf, len,
2985 addr1, l, mr);
2986 rcu_read_unlock();
2987 }
2988
2989 return result;
2990}
2991
2992/* Called within RCU critical section. */
2993MemTxResult address_space_read_continue(AddressSpace *as, hwaddr addr,
2994 MemTxAttrs attrs, uint8_t *buf,
2995 int len, hwaddr addr1, hwaddr l,
2996 MemoryRegion *mr)
2997{
2998 uint8_t *ptr;
2999 uint64_t val;
3000 MemTxResult result = MEMTX_OK;
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003001 bool release_lock = false;
3002
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003003 for (;;) {
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003004 if (!memory_access_is_direct(mr, false)) {
3005 /* I/O case */
3006 release_lock |= prepare_mmio_access(mr);
3007 l = memory_access_size(mr, l, addr1);
3008 switch (l) {
3009 case 8:
3010 /* 64 bit read access */
3011 result |= memory_region_dispatch_read(mr, addr1, &val, 8,
3012 attrs);
3013 stq_p(buf, val);
3014 break;
3015 case 4:
3016 /* 32 bit read access */
3017 result |= memory_region_dispatch_read(mr, addr1, &val, 4,
3018 attrs);
3019 stl_p(buf, val);
3020 break;
3021 case 2:
3022 /* 16 bit read access */
3023 result |= memory_region_dispatch_read(mr, addr1, &val, 2,
3024 attrs);
3025 stw_p(buf, val);
3026 break;
3027 case 1:
3028 /* 8 bit read access */
3029 result |= memory_region_dispatch_read(mr, addr1, &val, 1,
3030 attrs);
3031 stb_p(buf, val);
3032 break;
3033 default:
3034 abort();
3035 }
3036 } else {
3037 /* RAM case */
Prasad J Pandit04bf2522017-07-12 18:08:40 +05303038 ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003039 memcpy(buf, ptr, l);
3040 }
3041
3042 if (release_lock) {
3043 qemu_mutex_unlock_iothread();
3044 release_lock = false;
3045 }
3046
3047 len -= l;
3048 buf += l;
3049 addr += l;
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003050
3051 if (!len) {
3052 break;
3053 }
3054
3055 l = len;
3056 mr = address_space_translate(as, addr, &addr1, &l, false);
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003057 }
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003058
3059 return result;
3060}
3061
Paolo Bonzini3cc8f882015-12-09 10:34:13 +01003062MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr,
3063 MemTxAttrs attrs, uint8_t *buf, int len)
Paolo Bonzinia203ac72015-12-09 10:18:57 +01003064{
3065 hwaddr l;
3066 hwaddr addr1;
3067 MemoryRegion *mr;
3068 MemTxResult result = MEMTX_OK;
3069
3070 if (len > 0) {
3071 rcu_read_lock();
3072 l = len;
3073 mr = address_space_translate(as, addr, &addr1, &l, false);
3074 result = address_space_read_continue(as, addr, attrs, buf, len,
3075 addr1, l, mr);
3076 rcu_read_unlock();
3077 }
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003078
3079 return result;
Avi Kivityac1970f2012-10-03 16:22:53 +02003080}
3081
Paolo Bonzinieb7eeb82015-12-09 10:06:31 +01003082MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, MemTxAttrs attrs,
3083 uint8_t *buf, int len, bool is_write)
3084{
3085 if (is_write) {
3086 return address_space_write(as, addr, attrs, (uint8_t *)buf, len);
3087 } else {
3088 return address_space_read(as, addr, attrs, (uint8_t *)buf, len);
3089 }
3090}
Avi Kivityac1970f2012-10-03 16:22:53 +02003091
Avi Kivitya8170e52012-10-23 12:30:10 +02003092void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf,
Avi Kivityac1970f2012-10-03 16:22:53 +02003093 int len, int is_write)
3094{
Peter Maydell5c9eb022015-04-26 16:49:24 +01003095 address_space_rw(&address_space_memory, addr, MEMTXATTRS_UNSPECIFIED,
3096 buf, len, is_write);
Avi Kivityac1970f2012-10-03 16:22:53 +02003097}
3098
Alexander Graf582b55a2013-12-11 14:17:44 +01003099enum write_rom_type {
3100 WRITE_DATA,
3101 FLUSH_CACHE,
3102};
3103
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003104static inline void cpu_physical_memory_write_rom_internal(AddressSpace *as,
Alexander Graf582b55a2013-12-11 14:17:44 +01003105 hwaddr addr, const uint8_t *buf, int len, enum write_rom_type type)
bellardd0ecd2a2006-04-23 17:14:48 +00003106{
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003107 hwaddr l;
bellardd0ecd2a2006-04-23 17:14:48 +00003108 uint8_t *ptr;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003109 hwaddr addr1;
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003110 MemoryRegion *mr;
ths3b46e622007-09-17 08:09:54 +00003111
Paolo Bonzini41063e12015-03-18 14:21:43 +01003112 rcu_read_lock();
bellardd0ecd2a2006-04-23 17:14:48 +00003113 while (len > 0) {
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003114 l = len;
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003115 mr = address_space_translate(as, addr, &addr1, &l, true);
ths3b46e622007-09-17 08:09:54 +00003116
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003117 if (!(memory_region_is_ram(mr) ||
3118 memory_region_is_romd(mr))) {
Paolo Bonzinib242e0e2015-07-04 00:24:51 +02003119 l = memory_access_size(mr, l, addr1);
bellardd0ecd2a2006-04-23 17:14:48 +00003120 } else {
bellardd0ecd2a2006-04-23 17:14:48 +00003121 /* ROM/RAM case */
Paolo Bonzini0878d0e2016-02-22 11:02:12 +01003122 ptr = qemu_map_ram_ptr(mr->ram_block, addr1);
Alexander Graf582b55a2013-12-11 14:17:44 +01003123 switch (type) {
3124 case WRITE_DATA:
3125 memcpy(ptr, buf, l);
Paolo Bonzini845b6212015-03-23 11:45:53 +01003126 invalidate_and_set_dirty(mr, addr1, l);
Alexander Graf582b55a2013-12-11 14:17:44 +01003127 break;
3128 case FLUSH_CACHE:
3129 flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l);
3130 break;
3131 }
bellardd0ecd2a2006-04-23 17:14:48 +00003132 }
3133 len -= l;
3134 buf += l;
3135 addr += l;
3136 }
Paolo Bonzini41063e12015-03-18 14:21:43 +01003137 rcu_read_unlock();
bellardd0ecd2a2006-04-23 17:14:48 +00003138}
3139
Alexander Graf582b55a2013-12-11 14:17:44 +01003140/* used for ROM loading : can write in RAM and ROM */
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003141void cpu_physical_memory_write_rom(AddressSpace *as, hwaddr addr,
Alexander Graf582b55a2013-12-11 14:17:44 +01003142 const uint8_t *buf, int len)
3143{
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003144 cpu_physical_memory_write_rom_internal(as, addr, buf, len, WRITE_DATA);
Alexander Graf582b55a2013-12-11 14:17:44 +01003145}
3146
3147void cpu_flush_icache_range(hwaddr start, int len)
3148{
3149 /*
3150 * This function should do the same thing as an icache flush that was
3151 * triggered from within the guest. For TCG we are always cache coherent,
3152 * so there is no need to flush anything. For KVM / Xen we need to flush
3153 * the host's instruction cache at least.
3154 */
3155 if (tcg_enabled()) {
3156 return;
3157 }
3158
Edgar E. Iglesias2a221652013-12-13 16:28:52 +10003159 cpu_physical_memory_write_rom_internal(&address_space_memory,
3160 start, NULL, len, FLUSH_CACHE);
Alexander Graf582b55a2013-12-11 14:17:44 +01003161}
3162
aliguori6d16c2f2009-01-22 16:59:11 +00003163typedef struct {
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003164 MemoryRegion *mr;
aliguori6d16c2f2009-01-22 16:59:11 +00003165 void *buffer;
Avi Kivitya8170e52012-10-23 12:30:10 +02003166 hwaddr addr;
3167 hwaddr len;
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003168 bool in_use;
aliguori6d16c2f2009-01-22 16:59:11 +00003169} BounceBuffer;
3170
3171static BounceBuffer bounce;
3172
aliguoriba223c22009-01-22 16:59:16 +00003173typedef struct MapClient {
Fam Zhenge95205e2015-03-16 17:03:37 +08003174 QEMUBH *bh;
Blue Swirl72cf2d42009-09-12 07:36:22 +00003175 QLIST_ENTRY(MapClient) link;
aliguoriba223c22009-01-22 16:59:16 +00003176} MapClient;
3177
Fam Zheng38e047b2015-03-16 17:03:35 +08003178QemuMutex map_client_list_lock;
Blue Swirl72cf2d42009-09-12 07:36:22 +00003179static QLIST_HEAD(map_client_list, MapClient) map_client_list
3180 = QLIST_HEAD_INITIALIZER(map_client_list);
aliguoriba223c22009-01-22 16:59:16 +00003181
Fam Zhenge95205e2015-03-16 17:03:37 +08003182static void cpu_unregister_map_client_do(MapClient *client)
aliguoriba223c22009-01-22 16:59:16 +00003183{
Blue Swirl72cf2d42009-09-12 07:36:22 +00003184 QLIST_REMOVE(client, link);
Anthony Liguori7267c092011-08-20 22:09:37 -05003185 g_free(client);
aliguoriba223c22009-01-22 16:59:16 +00003186}
3187
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003188static void cpu_notify_map_clients_locked(void)
aliguoriba223c22009-01-22 16:59:16 +00003189{
3190 MapClient *client;
3191
Blue Swirl72cf2d42009-09-12 07:36:22 +00003192 while (!QLIST_EMPTY(&map_client_list)) {
3193 client = QLIST_FIRST(&map_client_list);
Fam Zhenge95205e2015-03-16 17:03:37 +08003194 qemu_bh_schedule(client->bh);
3195 cpu_unregister_map_client_do(client);
aliguoriba223c22009-01-22 16:59:16 +00003196 }
3197}
3198
Fam Zhenge95205e2015-03-16 17:03:37 +08003199void cpu_register_map_client(QEMUBH *bh)
bellardd0ecd2a2006-04-23 17:14:48 +00003200{
3201 MapClient *client = g_malloc(sizeof(*client));
3202
Fam Zheng38e047b2015-03-16 17:03:35 +08003203 qemu_mutex_lock(&map_client_list_lock);
Fam Zhenge95205e2015-03-16 17:03:37 +08003204 client->bh = bh;
bellardd0ecd2a2006-04-23 17:14:48 +00003205 QLIST_INSERT_HEAD(&map_client_list, client, link);
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003206 if (!atomic_read(&bounce.in_use)) {
3207 cpu_notify_map_clients_locked();
3208 }
Fam Zheng38e047b2015-03-16 17:03:35 +08003209 qemu_mutex_unlock(&map_client_list_lock);
bellardd0ecd2a2006-04-23 17:14:48 +00003210}
3211
Fam Zheng38e047b2015-03-16 17:03:35 +08003212void cpu_exec_init_all(void)
3213{
3214 qemu_mutex_init(&ram_list.mutex);
Peter Maydell20bccb82016-10-24 16:26:49 +01003215 /* The data structures we set up here depend on knowing the page size,
3216 * so no more changes can be made after this point.
3217 * In an ideal world, nothing we did before we had finished the
3218 * machine setup would care about the target page size, and we could
3219 * do this much later, rather than requiring board models to state
3220 * up front what their requirements are.
3221 */
3222 finalize_target_page_bits();
Fam Zheng38e047b2015-03-16 17:03:35 +08003223 io_mem_init();
Paolo Bonzini680a4782015-11-02 09:23:52 +01003224 memory_map_init();
Fam Zheng38e047b2015-03-16 17:03:35 +08003225 qemu_mutex_init(&map_client_list_lock);
3226}
3227
Fam Zhenge95205e2015-03-16 17:03:37 +08003228void cpu_unregister_map_client(QEMUBH *bh)
bellardd0ecd2a2006-04-23 17:14:48 +00003229{
Fam Zhenge95205e2015-03-16 17:03:37 +08003230 MapClient *client;
bellardd0ecd2a2006-04-23 17:14:48 +00003231
Fam Zhenge95205e2015-03-16 17:03:37 +08003232 qemu_mutex_lock(&map_client_list_lock);
3233 QLIST_FOREACH(client, &map_client_list, link) {
3234 if (client->bh == bh) {
3235 cpu_unregister_map_client_do(client);
3236 break;
3237 }
3238 }
3239 qemu_mutex_unlock(&map_client_list_lock);
bellardd0ecd2a2006-04-23 17:14:48 +00003240}
3241
3242static void cpu_notify_map_clients(void)
3243{
Fam Zheng38e047b2015-03-16 17:03:35 +08003244 qemu_mutex_lock(&map_client_list_lock);
Fam Zheng33b6c2e2015-03-16 17:03:36 +08003245 cpu_notify_map_clients_locked();
Fam Zheng38e047b2015-03-16 17:03:35 +08003246 qemu_mutex_unlock(&map_client_list_lock);
aliguori6d16c2f2009-01-22 16:59:11 +00003247}
3248
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003249bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write)
3250{
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003251 MemoryRegion *mr;
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003252 hwaddr l, xlat;
3253
Paolo Bonzini41063e12015-03-18 14:21:43 +01003254 rcu_read_lock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003255 while (len > 0) {
3256 l = len;
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003257 mr = address_space_translate(as, addr, &xlat, &l, is_write);
3258 if (!memory_access_is_direct(mr, is_write)) {
3259 l = memory_access_size(mr, l, addr);
3260 if (!memory_region_access_valid(mr, xlat, l, is_write)) {
Roman Kapl5ad4a2b2017-01-09 12:09:21 +01003261 rcu_read_unlock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003262 return false;
3263 }
3264 }
3265
3266 len -= l;
3267 addr += l;
3268 }
Paolo Bonzini41063e12015-03-18 14:21:43 +01003269 rcu_read_unlock();
Paolo Bonzini51644ab2013-04-11 15:40:59 +02003270 return true;
3271}
3272
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003273static hwaddr
3274address_space_extend_translation(AddressSpace *as, hwaddr addr, hwaddr target_len,
3275 MemoryRegion *mr, hwaddr base, hwaddr len,
3276 bool is_write)
3277{
3278 hwaddr done = 0;
3279 hwaddr xlat;
3280 MemoryRegion *this_mr;
3281
3282 for (;;) {
3283 target_len -= len;
3284 addr += len;
3285 done += len;
3286 if (target_len == 0) {
3287 return done;
3288 }
3289
3290 len = target_len;
3291 this_mr = address_space_translate(as, addr, &xlat, &len, is_write);
3292 if (this_mr != mr || xlat != base + done) {
3293 return done;
3294 }
3295 }
3296}
3297
aliguori6d16c2f2009-01-22 16:59:11 +00003298/* Map a physical memory region into a host virtual address.
3299 * May map a subset of the requested range, given by and returned in *plen.
3300 * May return NULL if resources needed to perform the mapping are exhausted.
3301 * Use only for reads OR writes - not for read-modify-write operations.
aliguoriba223c22009-01-22 16:59:16 +00003302 * Use cpu_register_map_client() to know when retrying the map operation is
3303 * likely to succeed.
aliguori6d16c2f2009-01-22 16:59:11 +00003304 */
Avi Kivityac1970f2012-10-03 16:22:53 +02003305void *address_space_map(AddressSpace *as,
Avi Kivitya8170e52012-10-23 12:30:10 +02003306 hwaddr addr,
3307 hwaddr *plen,
Avi Kivityac1970f2012-10-03 16:22:53 +02003308 bool is_write)
aliguori6d16c2f2009-01-22 16:59:11 +00003309{
Avi Kivitya8170e52012-10-23 12:30:10 +02003310 hwaddr len = *plen;
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003311 hwaddr l, xlat;
3312 MemoryRegion *mr;
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01003313 void *ptr;
aliguori6d16c2f2009-01-22 16:59:11 +00003314
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003315 if (len == 0) {
3316 return NULL;
3317 }
aliguori6d16c2f2009-01-22 16:59:11 +00003318
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003319 l = len;
Paolo Bonzini41063e12015-03-18 14:21:43 +01003320 rcu_read_lock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003321 mr = address_space_translate(as, addr, &xlat, &l, is_write);
Paolo Bonzini41063e12015-03-18 14:21:43 +01003322
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003323 if (!memory_access_is_direct(mr, is_write)) {
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003324 if (atomic_xchg(&bounce.in_use, true)) {
Paolo Bonzini41063e12015-03-18 14:21:43 +01003325 rcu_read_unlock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003326 return NULL;
aliguori6d16c2f2009-01-22 16:59:11 +00003327 }
Kevin Wolfe85d9db2013-07-22 14:30:23 +02003328 /* Avoid unbounded allocations */
3329 l = MIN(l, TARGET_PAGE_SIZE);
3330 bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l);
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003331 bounce.addr = addr;
3332 bounce.len = l;
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003333
3334 memory_region_ref(mr);
3335 bounce.mr = mr;
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003336 if (!is_write) {
Peter Maydell5c9eb022015-04-26 16:49:24 +01003337 address_space_read(as, addr, MEMTXATTRS_UNSPECIFIED,
3338 bounce.buffer, l);
Stefano Stabellini8ab934f2011-06-27 18:26:06 +01003339 }
aliguori6d16c2f2009-01-22 16:59:11 +00003340
Paolo Bonzini41063e12015-03-18 14:21:43 +01003341 rcu_read_unlock();
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003342 *plen = l;
3343 return bounce.buffer;
3344 }
3345
Paolo Bonzinie3127ae2013-06-28 17:29:27 +02003346
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003347 memory_region_ref(mr);
Paolo Bonzini715c31e2016-11-22 12:04:31 +01003348 *plen = address_space_extend_translation(as, addr, len, mr, xlat, l, is_write);
3349 ptr = qemu_ram_ptr_length(mr->ram_block, xlat, plen);
Paolo Bonzinie81bcda2015-12-16 10:31:26 +01003350 rcu_read_unlock();
3351
3352 return ptr;
aliguori6d16c2f2009-01-22 16:59:11 +00003353}
3354
Avi Kivityac1970f2012-10-03 16:22:53 +02003355/* Unmaps a memory region previously mapped by address_space_map().
aliguori6d16c2f2009-01-22 16:59:11 +00003356 * Will also mark the memory as dirty if is_write == 1. access_len gives
3357 * the amount of memory that was actually read or written by the caller.
3358 */
Avi Kivitya8170e52012-10-23 12:30:10 +02003359void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len,
3360 int is_write, hwaddr access_len)
aliguori6d16c2f2009-01-22 16:59:11 +00003361{
3362 if (buffer != bounce.buffer) {
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003363 MemoryRegion *mr;
3364 ram_addr_t addr1;
3365
Paolo Bonzini07bdaa42016-03-25 12:55:08 +01003366 mr = memory_region_from_host(buffer, &addr1);
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003367 assert(mr != NULL);
aliguori6d16c2f2009-01-22 16:59:11 +00003368 if (is_write) {
Paolo Bonzini845b6212015-03-23 11:45:53 +01003369 invalidate_and_set_dirty(mr, addr1, access_len);
aliguori6d16c2f2009-01-22 16:59:11 +00003370 }
Jan Kiszka868bb332011-06-21 22:59:09 +02003371 if (xen_enabled()) {
Jan Kiszkae41d7c62011-06-21 22:59:08 +02003372 xen_invalidate_map_cache_entry(buffer);
Anthony PERARD050a0dd2010-09-16 13:57:49 +01003373 }
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003374 memory_region_unref(mr);
aliguori6d16c2f2009-01-22 16:59:11 +00003375 return;
3376 }
3377 if (is_write) {
Peter Maydell5c9eb022015-04-26 16:49:24 +01003378 address_space_write(as, bounce.addr, MEMTXATTRS_UNSPECIFIED,
3379 bounce.buffer, access_len);
aliguori6d16c2f2009-01-22 16:59:11 +00003380 }
Herve Poussineauf8a83242010-01-24 21:23:56 +00003381 qemu_vfree(bounce.buffer);
aliguori6d16c2f2009-01-22 16:59:11 +00003382 bounce.buffer = NULL;
Paolo Bonzinid3e71552013-06-28 17:33:29 +02003383 memory_region_unref(bounce.mr);
Fam Zhengc2cba0f2015-03-16 17:03:33 +08003384 atomic_mb_set(&bounce.in_use, false);
aliguoriba223c22009-01-22 16:59:16 +00003385 cpu_notify_map_clients();
aliguori6d16c2f2009-01-22 16:59:11 +00003386}
bellardd0ecd2a2006-04-23 17:14:48 +00003387
Avi Kivitya8170e52012-10-23 12:30:10 +02003388void *cpu_physical_memory_map(hwaddr addr,
3389 hwaddr *plen,
Avi Kivityac1970f2012-10-03 16:22:53 +02003390 int is_write)
3391{
3392 return address_space_map(&address_space_memory, addr, plen, is_write);
3393}
3394
Avi Kivitya8170e52012-10-23 12:30:10 +02003395void cpu_physical_memory_unmap(void *buffer, hwaddr len,
3396 int is_write, hwaddr access_len)
Avi Kivityac1970f2012-10-03 16:22:53 +02003397{
3398 return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len);
3399}
3400
Paolo Bonzini0ce265f2016-11-22 11:34:02 +01003401#define ARG1_DECL AddressSpace *as
3402#define ARG1 as
3403#define SUFFIX
3404#define TRANSLATE(...) address_space_translate(as, __VA_ARGS__)
3405#define IS_DIRECT(mr, is_write) memory_access_is_direct(mr, is_write)
3406#define MAP_RAM(mr, ofs) qemu_map_ram_ptr((mr)->ram_block, ofs)
3407#define INVALIDATE(mr, ofs, len) invalidate_and_set_dirty(mr, ofs, len)
3408#define RCU_READ_LOCK(...) rcu_read_lock()
3409#define RCU_READ_UNLOCK(...) rcu_read_unlock()
3410#include "memory_ldst.inc.c"
Alexander Graf1e78bcc2011-07-06 09:09:23 +02003411
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003412int64_t address_space_cache_init(MemoryRegionCache *cache,
3413 AddressSpace *as,
3414 hwaddr addr,
3415 hwaddr len,
3416 bool is_write)
3417{
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003418 cache->len = len;
3419 cache->as = as;
3420 cache->xlat = addr;
3421 return len;
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003422}
3423
3424void address_space_cache_invalidate(MemoryRegionCache *cache,
3425 hwaddr addr,
3426 hwaddr access_len)
3427{
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003428}
3429
3430void address_space_cache_destroy(MemoryRegionCache *cache)
3431{
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003432 cache->as = NULL;
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003433}
3434
3435#define ARG1_DECL MemoryRegionCache *cache
3436#define ARG1 cache
3437#define SUFFIX _cached
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003438#define TRANSLATE(addr, ...) \
3439 address_space_translate(cache->as, cache->xlat + (addr), __VA_ARGS__)
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003440#define IS_DIRECT(mr, is_write) true
Paolo Bonzini90c4fe52017-04-03 13:41:28 +02003441#define MAP_RAM(mr, ofs) qemu_map_ram_ptr((mr)->ram_block, ofs)
3442#define INVALIDATE(mr, ofs, len) invalidate_and_set_dirty(mr, ofs, len)
3443#define RCU_READ_LOCK() rcu_read_lock()
3444#define RCU_READ_UNLOCK() rcu_read_unlock()
Paolo Bonzini1f4e4962016-11-22 12:04:52 +01003445#include "memory_ldst.inc.c"
3446
aliguori5e2972f2009-03-28 17:51:36 +00003447/* virtual memory access for debug (includes writing to ROM) */
Andreas Färberf17ec442013-06-29 19:40:58 +02003448int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr,
bellardb448f2f2004-02-25 23:24:04 +00003449 uint8_t *buf, int len, int is_write)
bellard13eb76e2004-01-24 15:23:36 +00003450{
3451 int l;
Avi Kivitya8170e52012-10-23 12:30:10 +02003452 hwaddr phys_addr;
j_mayer9b3c35e2007-04-07 11:21:28 +00003453 target_ulong page;
bellard13eb76e2004-01-24 15:23:36 +00003454
Christian Borntraeger79ca7a12017-03-07 15:19:08 +01003455 cpu_synchronize_state(cpu);
bellard13eb76e2004-01-24 15:23:36 +00003456 while (len > 0) {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003457 int asidx;
3458 MemTxAttrs attrs;
3459
bellard13eb76e2004-01-24 15:23:36 +00003460 page = addr & TARGET_PAGE_MASK;
Peter Maydell5232e4c2016-01-21 14:15:06 +00003461 phys_addr = cpu_get_phys_page_attrs_debug(cpu, page, &attrs);
3462 asidx = cpu_asidx_from_attrs(cpu, attrs);
bellard13eb76e2004-01-24 15:23:36 +00003463 /* if no physical page mapped, return an error */
3464 if (phys_addr == -1)
3465 return -1;
3466 l = (page + TARGET_PAGE_SIZE) - addr;
3467 if (l > len)
3468 l = len;
aliguori5e2972f2009-03-28 17:51:36 +00003469 phys_addr += (addr & ~TARGET_PAGE_MASK);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003470 if (is_write) {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003471 cpu_physical_memory_write_rom(cpu->cpu_ases[asidx].as,
3472 phys_addr, buf, l);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003473 } else {
Peter Maydell5232e4c2016-01-21 14:15:06 +00003474 address_space_rw(cpu->cpu_ases[asidx].as, phys_addr,
3475 MEMTXATTRS_UNSPECIFIED,
Peter Maydell5c9eb022015-04-26 16:49:24 +01003476 buf, l, 0);
Edgar E. Iglesias2e388472013-12-13 16:31:02 +10003477 }
bellard13eb76e2004-01-24 15:23:36 +00003478 len -= l;
3479 buf += l;
3480 addr += l;
3481 }
3482 return 0;
3483}
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003484
3485/*
3486 * Allows code that needs to deal with migration bitmaps etc to still be built
3487 * target independent.
3488 */
Juan Quintela20afaed2017-03-21 09:09:14 +01003489size_t qemu_target_page_size(void)
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003490{
Juan Quintela20afaed2017-03-21 09:09:14 +01003491 return TARGET_PAGE_SIZE;
Dr. David Alan Gilbert038629a2015-11-05 18:10:29 +00003492}
3493
Juan Quintela46d702b2017-04-24 21:03:48 +02003494int qemu_target_page_bits(void)
3495{
3496 return TARGET_PAGE_BITS;
3497}
3498
3499int qemu_target_page_bits_min(void)
3500{
3501 return TARGET_PAGE_BITS_MIN;
3502}
Paul Brooka68fe892010-03-01 00:08:59 +00003503#endif
bellard13eb76e2004-01-24 15:23:36 +00003504
Blue Swirl8e4a4242013-01-06 18:30:17 +00003505/*
3506 * A helper function for the _utterly broken_ virtio device model to find out if
3507 * it's running on a big endian machine. Don't do this at home kids!
3508 */
Greg Kurz98ed8ec2014-06-24 19:26:29 +02003509bool target_words_bigendian(void);
3510bool target_words_bigendian(void)
Blue Swirl8e4a4242013-01-06 18:30:17 +00003511{
3512#if defined(TARGET_WORDS_BIGENDIAN)
3513 return true;
3514#else
3515 return false;
3516#endif
3517}
3518
Wen Congyang76f35532012-05-07 12:04:18 +08003519#ifndef CONFIG_USER_ONLY
Avi Kivitya8170e52012-10-23 12:30:10 +02003520bool cpu_physical_memory_is_io(hwaddr phys_addr)
Wen Congyang76f35532012-05-07 12:04:18 +08003521{
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003522 MemoryRegion*mr;
Paolo Bonzini149f54b2013-05-24 12:59:37 +02003523 hwaddr l = 1;
Paolo Bonzini41063e12015-03-18 14:21:43 +01003524 bool res;
Wen Congyang76f35532012-05-07 12:04:18 +08003525
Paolo Bonzini41063e12015-03-18 14:21:43 +01003526 rcu_read_lock();
Paolo Bonzini5c8a00c2013-05-29 12:42:00 +02003527 mr = address_space_translate(&address_space_memory,
3528 phys_addr, &phys_addr, &l, false);
Wen Congyang76f35532012-05-07 12:04:18 +08003529
Paolo Bonzini41063e12015-03-18 14:21:43 +01003530 res = !(memory_region_is_ram(mr) || memory_region_is_romd(mr));
3531 rcu_read_unlock();
3532 return res;
Wen Congyang76f35532012-05-07 12:04:18 +08003533}
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003534
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003535int qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003536{
3537 RAMBlock *block;
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003538 int ret = 0;
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003539
Mike Day0dc3f442013-09-05 14:41:35 -04003540 rcu_read_lock();
Peter Xu99e15582017-05-12 12:17:39 +08003541 RAMBLOCK_FOREACH(block) {
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003542 ret = func(block->idstr, block->host, block->offset,
3543 block->used_length, opaque);
3544 if (ret) {
3545 break;
3546 }
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003547 }
Mike Day0dc3f442013-09-05 14:41:35 -04003548 rcu_read_unlock();
Dr. David Alan Gilberte3807052015-05-21 13:24:13 +01003549 return ret;
Michael R. Hinesbd2fa512013-06-25 21:35:34 -04003550}
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003551
3552/*
3553 * Unmap pages of memory from start to start+length such that
3554 * they a) read as 0, b) Trigger whatever fault mechanism
3555 * the OS provides for postcopy.
3556 * The pages must be unmapped by the end of the function.
3557 * Returns: 0 on success, none-0 on failure
3558 *
3559 */
3560int ram_block_discard_range(RAMBlock *rb, uint64_t start, size_t length)
3561{
3562 int ret = -1;
3563
3564 uint8_t *host_startaddr = rb->host + start;
3565
3566 if ((uintptr_t)host_startaddr & (rb->page_size - 1)) {
3567 error_report("ram_block_discard_range: Unaligned start address: %p",
3568 host_startaddr);
3569 goto err;
3570 }
3571
3572 if ((start + length) <= rb->used_length) {
3573 uint8_t *host_endaddr = host_startaddr + length;
3574 if ((uintptr_t)host_endaddr & (rb->page_size - 1)) {
3575 error_report("ram_block_discard_range: Unaligned end address: %p",
3576 host_endaddr);
3577 goto err;
3578 }
3579
3580 errno = ENOTSUP; /* If we are missing MADVISE etc */
3581
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003582 if (rb->page_size == qemu_host_page_size) {
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003583#if defined(CONFIG_MADVISE)
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003584 /* Note: We need the madvise MADV_DONTNEED behaviour of definitely
3585 * freeing the page.
3586 */
3587 ret = madvise(host_startaddr, length, MADV_DONTNEED);
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003588#endif
Dr. David Alan Gilberte2fa71f2017-02-24 18:28:33 +00003589 } else {
3590 /* Huge page case - unfortunately it can't do DONTNEED, but
3591 * it can do the equivalent by FALLOC_FL_PUNCH_HOLE in the
3592 * huge page file.
3593 */
3594#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
3595 ret = fallocate(rb->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
3596 start, length);
3597#endif
3598 }
Dr. David Alan Gilbertd3a50382017-02-24 18:28:32 +00003599 if (ret) {
3600 ret = -errno;
3601 error_report("ram_block_discard_range: Failed to discard range "
3602 "%s:%" PRIx64 " +%zx (%d)",
3603 rb->idstr, start, length, ret);
3604 }
3605 } else {
3606 error_report("ram_block_discard_range: Overrun block '%s' (%" PRIu64
3607 "/%zx/" RAM_ADDR_FMT")",
3608 rb->idstr, start, length, rb->used_length);
3609 }
3610
3611err:
3612 return ret;
3613}
3614
Peter Maydellec3f8c92013-06-27 20:53:38 +01003615#endif
Yang Zhonga0be0c52017-07-03 18:12:13 +08003616
3617void page_size_init(void)
3618{
3619 /* NOTE: we can always suppose that qemu_host_page_size >=
3620 TARGET_PAGE_SIZE */
3621 qemu_real_host_page_size = getpagesize();
3622 qemu_real_host_page_mask = -(intptr_t)qemu_real_host_page_size;
3623 if (qemu_host_page_size == 0) {
3624 qemu_host_page_size = qemu_real_host_page_size;
3625 }
3626 if (qemu_host_page_size < TARGET_PAGE_SIZE) {
3627 qemu_host_page_size = TARGET_PAGE_SIZE;
3628 }
3629 qemu_host_page_mask = -(intptr_t)qemu_host_page_size;
3630}