]> git.proxmox.com Git - mirror_qemu.git/blame - xen-hvm.c
sdhci: Fix misuse of qemu_free_irqs()
[mirror_qemu.git] / xen-hvm.c
CommitLineData
3285cf4f
AP
1/*
2 * Copyright (C) 2010 Citrix Ltd.
3 *
4 * This work is licensed under the terms of the GNU GPL, version 2. See
5 * the COPYING file in the top-level directory.
6 *
6b620ca3
PB
7 * Contributions after 2012-01-13 are licensed under the terms of the
8 * GNU GPL, version 2 or (at your option) any later version.
3285cf4f
AP
9 */
10
9ce94e7c
AS
11#include <sys/mman.h>
12
a2cb15b0 13#include "hw/pci/pci.h"
0d09e41a
PB
14#include "hw/i386/pc.h"
15#include "hw/xen/xen_common.h"
16#include "hw/xen/xen_backend.h"
39f42439 17#include "qmp-commands.h"
3285cf4f 18
dccfcd0e 19#include "sysemu/char.h"
1de7afc9 20#include "qemu/range.h"
9c17d615 21#include "sysemu/xen-mapcache.h"
432d268c 22#include "trace.h"
022c62cb 23#include "exec/address-spaces.h"
432d268c 24
9ce94e7c
AS
25#include <xen/hvm/ioreq.h>
26#include <xen/hvm/params.h>
8a369e20 27#include <xen/hvm/e820.h>
9ce94e7c 28
04b0de0e 29//#define DEBUG_XEN_HVM
9ce94e7c 30
04b0de0e 31#ifdef DEBUG_XEN_HVM
9ce94e7c
AS
32#define DPRINTF(fmt, ...) \
33 do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
34#else
35#define DPRINTF(fmt, ...) \
36 do { } while (0)
37#endif
38
ce76b8a8 39static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
c65adf9b 40static MemoryRegion *framebuffer;
39f42439 41static bool xen_in_migration;
ce76b8a8 42
9ce94e7c
AS
43/* Compatibility with older version */
44#if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
45static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
46{
47 return shared_page->vcpu_iodata[i].vp_eport;
48}
49static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
50{
51 return &shared_page->vcpu_iodata[vcpu].vp_ioreq;
52}
53# define FMT_ioreq_size PRIx64
54#else
55static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
56{
57 return shared_page->vcpu_ioreq[i].vp_eport;
58}
59static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
60{
61 return &shared_page->vcpu_ioreq[vcpu];
62}
63# define FMT_ioreq_size "u"
64#endif
fda1f768
SS
65#ifndef HVM_PARAM_BUFIOREQ_EVTCHN
66#define HVM_PARAM_BUFIOREQ_EVTCHN 26
67#endif
9ce94e7c
AS
68
69#define BUFFER_IO_MAX_DELAY 100
70
b4dd7802 71typedef struct XenPhysmap {
a8170e5e 72 hwaddr start_addr;
b4dd7802 73 ram_addr_t size;
d1814e08 74 char *name;
a8170e5e 75 hwaddr phys_offset;
b4dd7802
AP
76
77 QLIST_ENTRY(XenPhysmap) list;
78} XenPhysmap;
79
9ce94e7c
AS
80typedef struct XenIOState {
81 shared_iopage_t *shared_page;
82 buffered_iopage_t *buffered_io_page;
83 QEMUTimer *buffered_io_timer;
84 /* the evtchn port for polling the notification, */
85 evtchn_port_t *ioreq_local_port;
fda1f768
SS
86 /* evtchn local port for buffered io */
87 evtchn_port_t bufioreq_local_port;
9ce94e7c
AS
88 /* the evtchn fd for polling */
89 XenEvtchn xce_handle;
90 /* which vcpu we are serving */
91 int send_vcpu;
92
29321335 93 struct xs_handle *xenstore;
20581d20 94 MemoryListener memory_listener;
b4dd7802 95 QLIST_HEAD(, XenPhysmap) physmap;
a8170e5e 96 hwaddr free_phys_offset;
b4dd7802 97 const XenPhysmap *log_for_dirtybit;
29321335 98
9ce94e7c 99 Notifier exit;
da98c8eb 100 Notifier suspend;
11addd0a 101 Notifier wakeup;
9ce94e7c
AS
102} XenIOState;
103
41445300
AP
104/* Xen specific function for piix pci */
105
106int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
107{
108 return irq_num + ((pci_dev->devfn >> 3) << 2);
109}
110
111void xen_piix3_set_irq(void *opaque, int irq_num, int level)
112{
113 xc_hvm_set_pci_intx_level(xen_xc, xen_domid, 0, 0, irq_num >> 2,
114 irq_num & 3, level);
115}
116
117void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
118{
119 int i;
120
121 /* Scan for updates to PCI link routes (0x60-0x63). */
122 for (i = 0; i < len; i++) {
123 uint8_t v = (val >> (8 * i)) & 0xff;
124 if (v & 0x80) {
125 v = 0;
126 }
127 v &= 0xf;
128 if (((address + i) >= 0x60) && ((address + i) <= 0x63)) {
129 xc_hvm_set_pci_link_route(xen_xc, xen_domid, address + i - 0x60, v);
130 }
131 }
132}
133
f1dbf015
WL
134void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
135{
4c9f8d1b 136 xen_xc_hvm_inject_msi(xen_xc, xen_domid, addr, data);
f1dbf015
WL
137}
138
da98c8eb 139static void xen_suspend_notifier(Notifier *notifier, void *data)
c9622478 140{
da98c8eb 141 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
c9622478
AP
142}
143
9c11a8ac
AP
144/* Xen Interrupt Controller */
145
146static void xen_set_irq(void *opaque, int irq, int level)
147{
148 xc_hvm_set_isa_irq_level(xen_xc, xen_domid, irq, level);
149}
150
151qemu_irq *xen_interrupt_controller_init(void)
152{
153 return qemu_allocate_irqs(xen_set_irq, NULL, 16);
154}
155
432d268c
JN
156/* Memory Ops */
157
3c2a9669
DS
158static void xen_ram_init(ram_addr_t *below_4g_mem_size,
159 ram_addr_t *above_4g_mem_size,
160 ram_addr_t ram_size, MemoryRegion **ram_memory_p)
432d268c 161{
ce76b8a8 162 MemoryRegion *sysmem = get_system_memory();
ce76b8a8 163 ram_addr_t block_len;
c4f5cdc5
DS
164 uint64_t user_lowmem = object_property_get_int(qdev_get_machine(),
165 PC_MACHINE_MAX_RAM_BELOW_4G,
166 &error_abort);
432d268c 167
c4f5cdc5
DS
168 /* Handle the machine opt max-ram-below-4g. It is basicly doing
169 * min(xen limit, user limit).
170 */
171 if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
172 user_lowmem = HVM_BELOW_4G_RAM_END;
8a369e20 173 }
432d268c 174
c4f5cdc5
DS
175 if (ram_size >= user_lowmem) {
176 *above_4g_mem_size = ram_size - user_lowmem;
177 *below_4g_mem_size = user_lowmem;
432d268c 178 } else {
3c2a9669
DS
179 *above_4g_mem_size = 0;
180 *below_4g_mem_size = ram_size;
432d268c 181 }
c4f5cdc5
DS
182 if (!*above_4g_mem_size) {
183 block_len = ram_size;
184 } else {
185 /*
186 * Xen does not allocate the memory continuously, it keeps a
187 * hole of the size computed above or passed in.
188 */
189 block_len = (1ULL << 32) + *above_4g_mem_size;
190 }
191 memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len);
192 *ram_memory_p = &ram_memory;
193 vmstate_register_ram_global(&ram_memory);
432d268c 194
2c9b15ca 195 memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
ce76b8a8
AK
196 &ram_memory, 0, 0xa0000);
197 memory_region_add_subregion(sysmem, 0, &ram_640k);
8a369e20
AP
198 /* Skip of the VGA IO memory space, it will be registered later by the VGA
199 * emulated device.
200 *
201 * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
202 * the Options ROM, so it is registered here as RAM.
203 */
2c9b15ca 204 memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
3c2a9669
DS
205 &ram_memory, 0xc0000,
206 *below_4g_mem_size - 0xc0000);
ce76b8a8 207 memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
3c2a9669 208 if (*above_4g_mem_size > 0) {
2c9b15ca 209 memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
ce76b8a8 210 &ram_memory, 0x100000000ULL,
3c2a9669 211 *above_4g_mem_size);
ce76b8a8 212 memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
432d268c 213 }
432d268c
JN
214}
215
fce537d4 216void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
432d268c
JN
217{
218 unsigned long nr_pfn;
219 xen_pfn_t *pfn_list;
220 int i;
221
c234572d
AP
222 if (runstate_check(RUN_STATE_INMIGRATE)) {
223 /* RAM already populated in Xen */
224 fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
225 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
226 __func__, size, ram_addr);
227 return;
228 }
229
ce76b8a8
AK
230 if (mr == &ram_memory) {
231 return;
232 }
233
432d268c
JN
234 trace_xen_ram_alloc(ram_addr, size);
235
236 nr_pfn = size >> TARGET_PAGE_BITS;
7267c094 237 pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
432d268c
JN
238
239 for (i = 0; i < nr_pfn; i++) {
240 pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
241 }
242
243 if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
f15fbc4b 244 hw_error("xen: failed to populate ram at " RAM_ADDR_FMT, ram_addr);
432d268c
JN
245 }
246
7267c094 247 g_free(pfn_list);
432d268c
JN
248}
249
b4dd7802 250static XenPhysmap *get_physmapping(XenIOState *state,
a8170e5e 251 hwaddr start_addr, ram_addr_t size)
b4dd7802
AP
252{
253 XenPhysmap *physmap = NULL;
254
255 start_addr &= TARGET_PAGE_MASK;
256
257 QLIST_FOREACH(physmap, &state->physmap, list) {
258 if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
259 return physmap;
260 }
261 }
262 return NULL;
263}
264
a8170e5e 265static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
cd1ba7de
AP
266 ram_addr_t size, void *opaque)
267{
a8170e5e 268 hwaddr addr = start_addr & TARGET_PAGE_MASK;
cd1ba7de
AP
269 XenIOState *xen_io_state = opaque;
270 XenPhysmap *physmap = NULL;
271
272 QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
273 if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
274 return physmap->start_addr;
275 }
276 }
277
278 return start_addr;
279}
280
b4dd7802
AP
281#if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
282static int xen_add_to_physmap(XenIOState *state,
a8170e5e 283 hwaddr start_addr,
b4dd7802 284 ram_addr_t size,
20581d20 285 MemoryRegion *mr,
a8170e5e 286 hwaddr offset_within_region)
b4dd7802
AP
287{
288 unsigned long i = 0;
289 int rc = 0;
290 XenPhysmap *physmap = NULL;
a8170e5e
AK
291 hwaddr pfn, start_gpfn;
292 hwaddr phys_offset = memory_region_get_ram_addr(mr);
d1814e08 293 char path[80], value[17];
b4dd7802
AP
294
295 if (get_physmapping(state, start_addr, size)) {
296 return 0;
297 }
298 if (size <= 0) {
299 return -1;
300 }
301
ebed8505
SS
302 /* Xen can only handle a single dirty log region for now and we want
303 * the linear framebuffer to be that region.
304 * Avoid tracking any regions that is not videoram and avoid tracking
305 * the legacy vga region. */
20581d20
AK
306 if (mr == framebuffer && start_addr > 0xbffff) {
307 goto go_physmap;
ebed8505
SS
308 }
309 return -1;
310
311go_physmap:
f1b8caf1
SE
312 DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
313 start_addr, start_addr + size);
b4dd7802
AP
314
315 pfn = phys_offset >> TARGET_PAGE_BITS;
316 start_gpfn = start_addr >> TARGET_PAGE_BITS;
317 for (i = 0; i < size >> TARGET_PAGE_BITS; i++) {
318 unsigned long idx = pfn + i;
319 xen_pfn_t gpfn = start_gpfn + i;
320
321 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
322 if (rc) {
323 DPRINTF("add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
324 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
325 return -rc;
326 }
327 }
328
7267c094 329 physmap = g_malloc(sizeof (XenPhysmap));
b4dd7802
AP
330
331 physmap->start_addr = start_addr;
332 physmap->size = size;
d1814e08 333 physmap->name = (char *)mr->name;
b4dd7802
AP
334 physmap->phys_offset = phys_offset;
335
336 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
337
338 xc_domain_pin_memory_cacheattr(xen_xc, xen_domid,
339 start_addr >> TARGET_PAGE_BITS,
8b6bb0ad 340 (start_addr + size - 1) >> TARGET_PAGE_BITS,
b4dd7802 341 XEN_DOMCTL_MEM_CACHEATTR_WB);
d1814e08
SS
342
343 snprintf(path, sizeof(path),
344 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
345 xen_domid, (uint64_t)phys_offset);
346 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)start_addr);
347 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
348 return -1;
349 }
350 snprintf(path, sizeof(path),
351 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
352 xen_domid, (uint64_t)phys_offset);
353 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)size);
354 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
355 return -1;
356 }
357 if (mr->name) {
358 snprintf(path, sizeof(path),
359 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
360 xen_domid, (uint64_t)phys_offset);
361 if (!xs_write(state->xenstore, 0, path, mr->name, strlen(mr->name))) {
362 return -1;
363 }
364 }
365
b4dd7802
AP
366 return 0;
367}
368
369static int xen_remove_from_physmap(XenIOState *state,
a8170e5e 370 hwaddr start_addr,
b4dd7802
AP
371 ram_addr_t size)
372{
373 unsigned long i = 0;
374 int rc = 0;
375 XenPhysmap *physmap = NULL;
a8170e5e 376 hwaddr phys_offset = 0;
b4dd7802
AP
377
378 physmap = get_physmapping(state, start_addr, size);
379 if (physmap == NULL) {
380 return -1;
381 }
382
383 phys_offset = physmap->phys_offset;
384 size = physmap->size;
385
d18e173a
WL
386 DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
387 "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
b4dd7802
AP
388
389 size >>= TARGET_PAGE_BITS;
390 start_addr >>= TARGET_PAGE_BITS;
391 phys_offset >>= TARGET_PAGE_BITS;
392 for (i = 0; i < size; i++) {
393 unsigned long idx = start_addr + i;
394 xen_pfn_t gpfn = phys_offset + i;
395
396 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
397 if (rc) {
398 fprintf(stderr, "add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
399 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
400 return -rc;
401 }
402 }
403
404 QLIST_REMOVE(physmap, list);
405 if (state->log_for_dirtybit == physmap) {
406 state->log_for_dirtybit = NULL;
407 }
c5633d99 408 g_free(physmap);
b4dd7802
AP
409
410 return 0;
411}
412
413#else
414static int xen_add_to_physmap(XenIOState *state,
a8170e5e 415 hwaddr start_addr,
b4dd7802 416 ram_addr_t size,
20581d20 417 MemoryRegion *mr,
a8170e5e 418 hwaddr offset_within_region)
b4dd7802
AP
419{
420 return -ENOSYS;
421}
422
423static int xen_remove_from_physmap(XenIOState *state,
a8170e5e 424 hwaddr start_addr,
b4dd7802
AP
425 ram_addr_t size)
426{
427 return -ENOSYS;
428}
429#endif
430
20581d20
AK
431static void xen_set_memory(struct MemoryListener *listener,
432 MemoryRegionSection *section,
433 bool add)
b4dd7802 434{
20581d20 435 XenIOState *state = container_of(listener, XenIOState, memory_listener);
a8170e5e 436 hwaddr start_addr = section->offset_within_address_space;
052e87b0 437 ram_addr_t size = int128_get64(section->size);
20581d20 438 bool log_dirty = memory_region_is_logging(section->mr);
b4dd7802
AP
439 hvmmem_type_t mem_type;
440
20581d20 441 if (!memory_region_is_ram(section->mr)) {
b4dd7802
AP
442 return;
443 }
444
20581d20
AK
445 if (!(section->mr != &ram_memory
446 && ( (log_dirty && add) || (!log_dirty && !add)))) {
447 return;
448 }
449
450 trace_xen_client_set_memory(start_addr, size, log_dirty);
b4dd7802
AP
451
452 start_addr &= TARGET_PAGE_MASK;
453 size = TARGET_PAGE_ALIGN(size);
20581d20
AK
454
455 if (add) {
456 if (!memory_region_is_rom(section->mr)) {
457 xen_add_to_physmap(state, start_addr, size,
458 section->mr, section->offset_within_region);
459 } else {
460 mem_type = HVMMEM_ram_ro;
461 if (xc_hvm_set_mem_type(xen_xc, xen_domid, mem_type,
462 start_addr >> TARGET_PAGE_BITS,
463 size >> TARGET_PAGE_BITS)) {
464 DPRINTF("xc_hvm_set_mem_type error, addr: "TARGET_FMT_plx"\n",
465 start_addr);
466 }
b4dd7802 467 }
20581d20 468 } else {
b4dd7802
AP
469 if (xen_remove_from_physmap(state, start_addr, size) < 0) {
470 DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
471 }
b4dd7802
AP
472 }
473}
474
20581d20
AK
475static void xen_region_add(MemoryListener *listener,
476 MemoryRegionSection *section)
477{
dfde4e6e 478 memory_region_ref(section->mr);
20581d20
AK
479 xen_set_memory(listener, section, true);
480}
481
482static void xen_region_del(MemoryListener *listener,
483 MemoryRegionSection *section)
484{
485 xen_set_memory(listener, section, false);
dfde4e6e 486 memory_region_unref(section->mr);
20581d20
AK
487}
488
b18620cf 489static void xen_sync_dirty_bitmap(XenIOState *state,
a8170e5e 490 hwaddr start_addr,
b18620cf 491 ram_addr_t size)
b4dd7802 492{
a8170e5e 493 hwaddr npages = size >> TARGET_PAGE_BITS;
b4dd7802
AP
494 const int width = sizeof(unsigned long) * 8;
495 unsigned long bitmap[(npages + width - 1) / width];
496 int rc, i, j;
497 const XenPhysmap *physmap = NULL;
498
499 physmap = get_physmapping(state, start_addr, size);
500 if (physmap == NULL) {
501 /* not handled */
b18620cf 502 return;
b4dd7802
AP
503 }
504
505 if (state->log_for_dirtybit == NULL) {
506 state->log_for_dirtybit = physmap;
507 } else if (state->log_for_dirtybit != physmap) {
b18620cf
AP
508 /* Only one range for dirty bitmap can be tracked. */
509 return;
b4dd7802 510 }
b4dd7802
AP
511
512 rc = xc_hvm_track_dirty_vram(xen_xc, xen_domid,
513 start_addr >> TARGET_PAGE_BITS, npages,
514 bitmap);
b18620cf
AP
515 if (rc < 0) {
516 if (rc != -ENODATA) {
8aba7dc0
AP
517 memory_region_set_dirty(framebuffer, 0, size);
518 DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
b18620cf
AP
519 ", 0x" TARGET_FMT_plx "): %s\n",
520 start_addr, start_addr + size, strerror(-rc));
521 }
522 return;
b4dd7802
AP
523 }
524
525 for (i = 0; i < ARRAY_SIZE(bitmap); i++) {
526 unsigned long map = bitmap[i];
527 while (map != 0) {
adf9d70b 528 j = ctzl(map);
b4dd7802 529 map &= ~(1ul << j);
5a97065b 530 memory_region_set_dirty(framebuffer,
fd4aa979
BS
531 (i * width + j) * TARGET_PAGE_SIZE,
532 TARGET_PAGE_SIZE);
b4dd7802
AP
533 };
534 }
b4dd7802
AP
535}
536
20581d20
AK
537static void xen_log_start(MemoryListener *listener,
538 MemoryRegionSection *section)
b4dd7802 539{
20581d20 540 XenIOState *state = container_of(listener, XenIOState, memory_listener);
b4dd7802 541
b18620cf 542 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
052e87b0 543 int128_get64(section->size));
b4dd7802
AP
544}
545
20581d20 546static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section)
b4dd7802 547{
20581d20 548 XenIOState *state = container_of(listener, XenIOState, memory_listener);
b4dd7802
AP
549
550 state->log_for_dirtybit = NULL;
551 /* Disable dirty bit tracking */
b18620cf 552 xc_hvm_track_dirty_vram(xen_xc, xen_domid, 0, 0, NULL);
b4dd7802
AP
553}
554
20581d20 555static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
b4dd7802 556{
20581d20 557 XenIOState *state = container_of(listener, XenIOState, memory_listener);
b4dd7802 558
b18620cf 559 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
052e87b0 560 int128_get64(section->size));
b4dd7802
AP
561}
562
20581d20
AK
563static void xen_log_global_start(MemoryListener *listener)
564{
39f42439
AP
565 if (xen_enabled()) {
566 xen_in_migration = true;
567 }
20581d20
AK
568}
569
570static void xen_log_global_stop(MemoryListener *listener)
b4dd7802 571{
39f42439 572 xen_in_migration = false;
b4dd7802
AP
573}
574
20581d20
AK
575static MemoryListener xen_memory_listener = {
576 .region_add = xen_region_add,
577 .region_del = xen_region_del,
b4dd7802
AP
578 .log_start = xen_log_start,
579 .log_stop = xen_log_stop,
20581d20
AK
580 .log_sync = xen_log_sync,
581 .log_global_start = xen_log_global_start,
582 .log_global_stop = xen_log_global_stop,
72e22d2f 583 .priority = 10,
b4dd7802 584};
432d268c 585
9ce94e7c
AS
586/* get the ioreq packets from share mem */
587static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
588{
589 ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
590
591 if (req->state != STATE_IOREQ_READY) {
592 DPRINTF("I/O request not ready: "
593 "%x, ptr: %x, port: %"PRIx64", "
594 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
595 req->state, req->data_is_ptr, req->addr,
596 req->data, req->count, req->size);
597 return NULL;
598 }
599
600 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
601
602 req->state = STATE_IOREQ_INPROCESS;
603 return req;
604}
605
606/* use poll to get the port notification */
607/* ioreq_vec--out,the */
608/* retval--the number of ioreq packet */
609static ioreq_t *cpu_get_ioreq(XenIOState *state)
610{
611 int i;
612 evtchn_port_t port;
613
614 port = xc_evtchn_pending(state->xce_handle);
fda1f768 615 if (port == state->bufioreq_local_port) {
bc72ad67
AB
616 timer_mod(state->buffered_io_timer,
617 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
fda1f768
SS
618 return NULL;
619 }
620
9ce94e7c 621 if (port != -1) {
1cd25a88 622 for (i = 0; i < max_cpus; i++) {
9ce94e7c
AS
623 if (state->ioreq_local_port[i] == port) {
624 break;
625 }
626 }
627
1cd25a88 628 if (i == max_cpus) {
9ce94e7c
AS
629 hw_error("Fatal error while trying to get io event!\n");
630 }
631
632 /* unmask the wanted port again */
633 xc_evtchn_unmask(state->xce_handle, port);
634
635 /* get the io packet from shared memory */
636 state->send_vcpu = i;
637 return cpu_get_ioreq_from_shared_memory(state, i);
638 }
639
640 /* read error or read nothing */
641 return NULL;
642}
643
644static uint32_t do_inp(pio_addr_t addr, unsigned long size)
645{
646 switch (size) {
647 case 1:
648 return cpu_inb(addr);
649 case 2:
650 return cpu_inw(addr);
651 case 4:
652 return cpu_inl(addr);
653 default:
654 hw_error("inp: bad size: %04"FMT_pioaddr" %lx", addr, size);
655 }
656}
657
658static void do_outp(pio_addr_t addr,
659 unsigned long size, uint32_t val)
660{
661 switch (size) {
662 case 1:
663 return cpu_outb(addr, val);
664 case 2:
665 return cpu_outw(addr, val);
666 case 4:
667 return cpu_outl(addr, val);
668 default:
669 hw_error("outp: bad size: %04"FMT_pioaddr" %lx", addr, size);
670 }
671}
672
a3864829
IJ
673/*
674 * Helper functions which read/write an object from/to physical guest
675 * memory, as part of the implementation of an ioreq.
676 *
677 * Equivalent to
678 * cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
679 * val, req->size, 0/1)
680 * except without the integer overflow problems.
681 */
682static void rw_phys_req_item(hwaddr addr,
683 ioreq_t *req, uint32_t i, void *val, int rw)
684{
685 /* Do everything unsigned so overflow just results in a truncated result
686 * and accesses to undesired parts of guest memory, which is up
687 * to the guest */
688 hwaddr offset = (hwaddr)req->size * i;
689 if (req->df) {
690 addr -= offset;
691 } else {
692 addr += offset;
693 }
694 cpu_physical_memory_rw(addr, val, req->size, rw);
695}
696
697static inline void read_phys_req_item(hwaddr addr,
698 ioreq_t *req, uint32_t i, void *val)
9ce94e7c 699{
a3864829
IJ
700 rw_phys_req_item(addr, req, i, val, 0);
701}
702static inline void write_phys_req_item(hwaddr addr,
703 ioreq_t *req, uint32_t i, void *val)
704{
705 rw_phys_req_item(addr, req, i, val, 1);
706}
9ce94e7c 707
a3864829
IJ
708
709static void cpu_ioreq_pio(ioreq_t *req)
710{
249e7e0f 711 uint32_t i;
9ce94e7c
AS
712
713 if (req->dir == IOREQ_READ) {
714 if (!req->data_is_ptr) {
715 req->data = do_inp(req->addr, req->size);
716 } else {
717 uint32_t tmp;
718
719 for (i = 0; i < req->count; i++) {
720 tmp = do_inp(req->addr, req->size);
a3864829 721 write_phys_req_item(req->data, req, i, &tmp);
9ce94e7c
AS
722 }
723 }
724 } else if (req->dir == IOREQ_WRITE) {
725 if (!req->data_is_ptr) {
726 do_outp(req->addr, req->size, req->data);
727 } else {
728 for (i = 0; i < req->count; i++) {
729 uint32_t tmp = 0;
730
a3864829 731 read_phys_req_item(req->data, req, i, &tmp);
9ce94e7c
AS
732 do_outp(req->addr, req->size, tmp);
733 }
734 }
735 }
736}
737
738static void cpu_ioreq_move(ioreq_t *req)
739{
249e7e0f 740 uint32_t i;
9ce94e7c
AS
741
742 if (!req->data_is_ptr) {
743 if (req->dir == IOREQ_READ) {
744 for (i = 0; i < req->count; i++) {
a3864829 745 read_phys_req_item(req->addr, req, i, &req->data);
9ce94e7c
AS
746 }
747 } else if (req->dir == IOREQ_WRITE) {
748 for (i = 0; i < req->count; i++) {
a3864829 749 write_phys_req_item(req->addr, req, i, &req->data);
9ce94e7c
AS
750 }
751 }
752 } else {
2b734340 753 uint64_t tmp;
9ce94e7c
AS
754
755 if (req->dir == IOREQ_READ) {
756 for (i = 0; i < req->count; i++) {
a3864829
IJ
757 read_phys_req_item(req->addr, req, i, &tmp);
758 write_phys_req_item(req->data, req, i, &tmp);
9ce94e7c
AS
759 }
760 } else if (req->dir == IOREQ_WRITE) {
761 for (i = 0; i < req->count; i++) {
a3864829
IJ
762 read_phys_req_item(req->data, req, i, &tmp);
763 write_phys_req_item(req->addr, req, i, &tmp);
9ce94e7c
AS
764 }
765 }
766 }
767}
768
769static void handle_ioreq(ioreq_t *req)
770{
771 if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
772 (req->size < sizeof (target_ulong))) {
773 req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
774 }
775
776 switch (req->type) {
777 case IOREQ_TYPE_PIO:
778 cpu_ioreq_pio(req);
779 break;
780 case IOREQ_TYPE_COPY:
781 cpu_ioreq_move(req);
782 break;
783 case IOREQ_TYPE_TIMEOFFSET:
784 break;
785 case IOREQ_TYPE_INVALIDATE:
e41d7c69 786 xen_invalidate_map_cache();
9ce94e7c
AS
787 break;
788 default:
789 hw_error("Invalid ioreq type 0x%x\n", req->type);
790 }
791}
792
fda1f768 793static int handle_buffered_iopage(XenIOState *state)
9ce94e7c
AS
794{
795 buf_ioreq_t *buf_req = NULL;
796 ioreq_t req;
797 int qw;
798
799 if (!state->buffered_io_page) {
fda1f768 800 return 0;
9ce94e7c
AS
801 }
802
fda1f768
SS
803 memset(&req, 0x00, sizeof(req));
804
9ce94e7c
AS
805 while (state->buffered_io_page->read_pointer != state->buffered_io_page->write_pointer) {
806 buf_req = &state->buffered_io_page->buf_ioreq[
807 state->buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
808 req.size = 1UL << buf_req->size;
809 req.count = 1;
810 req.addr = buf_req->addr;
811 req.data = buf_req->data;
812 req.state = STATE_IOREQ_READY;
813 req.dir = buf_req->dir;
814 req.df = 1;
815 req.type = buf_req->type;
816 req.data_is_ptr = 0;
817 qw = (req.size == 8);
818 if (qw) {
819 buf_req = &state->buffered_io_page->buf_ioreq[
820 (state->buffered_io_page->read_pointer + 1) % IOREQ_BUFFER_SLOT_NUM];
821 req.data |= ((uint64_t)buf_req->data) << 32;
822 }
823
824 handle_ioreq(&req);
825
826 xen_mb();
827 state->buffered_io_page->read_pointer += qw ? 2 : 1;
828 }
fda1f768
SS
829
830 return req.count;
9ce94e7c
AS
831}
832
833static void handle_buffered_io(void *opaque)
834{
835 XenIOState *state = opaque;
836
fda1f768 837 if (handle_buffered_iopage(state)) {
bc72ad67
AB
838 timer_mod(state->buffered_io_timer,
839 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
fda1f768 840 } else {
bc72ad67 841 timer_del(state->buffered_io_timer);
fda1f768
SS
842 xc_evtchn_unmask(state->xce_handle, state->bufioreq_local_port);
843 }
9ce94e7c
AS
844}
845
846static void cpu_handle_ioreq(void *opaque)
847{
848 XenIOState *state = opaque;
849 ioreq_t *req = cpu_get_ioreq(state);
850
851 handle_buffered_iopage(state);
852 if (req) {
853 handle_ioreq(req);
854
855 if (req->state != STATE_IOREQ_INPROCESS) {
856 fprintf(stderr, "Badness in I/O request ... not in service?!: "
857 "%x, ptr: %x, port: %"PRIx64", "
858 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
859 req->state, req->data_is_ptr, req->addr,
860 req->data, req->count, req->size);
180640ea 861 destroy_hvm_domain(false);
9ce94e7c
AS
862 return;
863 }
864
865 xen_wmb(); /* Update ioreq contents /then/ update state. */
866
867 /*
868 * We do this before we send the response so that the tools
869 * have the opportunity to pick up on the reset before the
870 * guest resumes and does a hlt with interrupts disabled which
871 * causes Xen to powerdown the domain.
872 */
1354869c 873 if (runstate_is_running()) {
9ce94e7c 874 if (qemu_shutdown_requested_get()) {
180640ea 875 destroy_hvm_domain(false);
9ce94e7c
AS
876 }
877 if (qemu_reset_requested_get()) {
e063eb1f 878 qemu_system_reset(VMRESET_REPORT);
180640ea 879 destroy_hvm_domain(true);
9ce94e7c
AS
880 }
881 }
882
883 req->state = STATE_IORESP_READY;
884 xc_evtchn_notify(state->xce_handle, state->ioreq_local_port[state->send_vcpu]);
885 }
886}
887
888static void xen_main_loop_prepare(XenIOState *state)
889{
890 int evtchn_fd = -1;
891
892 if (state->xce_handle != XC_HANDLER_INITIAL_VALUE) {
893 evtchn_fd = xc_evtchn_fd(state->xce_handle);
894 }
895
bc72ad67 896 state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
9ce94e7c 897 state);
9ce94e7c
AS
898
899 if (evtchn_fd != -1) {
900 qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
901 }
902}
903
904
1dfb4dd9
LC
905static void xen_hvm_change_state_handler(void *opaque, int running,
906 RunState rstate)
9ce94e7c 907{
1dfb4dd9 908 XenIOState *xstate = opaque;
9ce94e7c 909 if (running) {
1dfb4dd9 910 xen_main_loop_prepare(xstate);
9ce94e7c
AS
911 }
912}
913
9e8dd451 914static void xen_exit_notifier(Notifier *n, void *data)
9ce94e7c
AS
915{
916 XenIOState *state = container_of(n, XenIOState, exit);
917
918 xc_evtchn_close(state->xce_handle);
29321335 919 xs_daemon_close(state->xenstore);
9ce94e7c
AS
920}
921
d1814e08
SS
922static void xen_read_physmap(XenIOState *state)
923{
924 XenPhysmap *physmap = NULL;
925 unsigned int len, num, i;
926 char path[80], *value = NULL;
927 char **entries = NULL;
928
929 snprintf(path, sizeof(path),
930 "/local/domain/0/device-model/%d/physmap", xen_domid);
931 entries = xs_directory(state->xenstore, 0, path, &num);
932 if (entries == NULL)
933 return;
934
935 for (i = 0; i < num; i++) {
936 physmap = g_malloc(sizeof (XenPhysmap));
937 physmap->phys_offset = strtoull(entries[i], NULL, 16);
938 snprintf(path, sizeof(path),
939 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
940 xen_domid, entries[i]);
941 value = xs_read(state->xenstore, 0, path, &len);
942 if (value == NULL) {
c5633d99 943 g_free(physmap);
d1814e08
SS
944 continue;
945 }
946 physmap->start_addr = strtoull(value, NULL, 16);
947 free(value);
948
949 snprintf(path, sizeof(path),
950 "/local/domain/0/device-model/%d/physmap/%s/size",
951 xen_domid, entries[i]);
952 value = xs_read(state->xenstore, 0, path, &len);
953 if (value == NULL) {
c5633d99 954 g_free(physmap);
d1814e08
SS
955 continue;
956 }
957 physmap->size = strtoull(value, NULL, 16);
958 free(value);
959
960 snprintf(path, sizeof(path),
961 "/local/domain/0/device-model/%d/physmap/%s/name",
962 xen_domid, entries[i]);
963 physmap->name = xs_read(state->xenstore, 0, path, &len);
964
965 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
966 }
967 free(entries);
d1814e08
SS
968}
969
11addd0a
LJ
970static void xen_wakeup_notifier(Notifier *notifier, void *data)
971{
972 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
973}
974
3c2a9669
DS
975int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
976 MemoryRegion **ram_memory)
29d3ccde 977{
9ce94e7c
AS
978 int i, rc;
979 unsigned long ioreq_pfn;
fda1f768 980 unsigned long bufioreq_evtchn;
9ce94e7c
AS
981 XenIOState *state;
982
7267c094 983 state = g_malloc0(sizeof (XenIOState));
9ce94e7c
AS
984
985 state->xce_handle = xen_xc_evtchn_open(NULL, 0);
986 if (state->xce_handle == XC_HANDLER_INITIAL_VALUE) {
987 perror("xen: event channel open");
c5633d99 988 g_free(state);
9ce94e7c
AS
989 return -errno;
990 }
991
29321335
AP
992 state->xenstore = xs_daemon_open();
993 if (state->xenstore == NULL) {
994 perror("xen: xenstore open");
c5633d99 995 g_free(state);
29321335
AP
996 return -errno;
997 }
998
9ce94e7c
AS
999 state->exit.notify = xen_exit_notifier;
1000 qemu_add_exit_notifier(&state->exit);
1001
da98c8eb
GH
1002 state->suspend.notify = xen_suspend_notifier;
1003 qemu_register_suspend_notifier(&state->suspend);
1004
11addd0a
LJ
1005 state->wakeup.notify = xen_wakeup_notifier;
1006 qemu_register_wakeup_notifier(&state->wakeup);
1007
9ce94e7c
AS
1008 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_IOREQ_PFN, &ioreq_pfn);
1009 DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1010 state->shared_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1011 PROT_READ|PROT_WRITE, ioreq_pfn);
1012 if (state->shared_page == NULL) {
1013 hw_error("map shared IO page returned error %d handle=" XC_INTERFACE_FMT,
1014 errno, xen_xc);
1015 }
1016
1017 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
1018 DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
1019 state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1020 PROT_READ|PROT_WRITE, ioreq_pfn);
1021 if (state->buffered_io_page == NULL) {
1022 hw_error("map buffered IO page returned error %d", errno);
1023 }
1024
1cd25a88 1025 state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
9ce94e7c
AS
1026
1027 /* FIXME: how about if we overflow the page here? */
1cd25a88 1028 for (i = 0; i < max_cpus; i++) {
9ce94e7c
AS
1029 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1030 xen_vcpu_eport(state->shared_page, i));
1031 if (rc == -1) {
1032 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1033 return -1;
1034 }
1035 state->ioreq_local_port[i] = rc;
1036 }
1037
fda1f768
SS
1038 rc = xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_EVTCHN,
1039 &bufioreq_evtchn);
1040 if (rc < 0) {
1041 fprintf(stderr, "failed to get HVM_PARAM_BUFIOREQ_EVTCHN\n");
1042 return -1;
1043 }
1044 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1045 (uint32_t)bufioreq_evtchn);
1046 if (rc == -1) {
1047 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1048 return -1;
1049 }
1050 state->bufioreq_local_port = rc;
1051
432d268c 1052 /* Init RAM management */
cd1ba7de 1053 xen_map_cache_init(xen_phys_offset_to_gaddr, state);
3c2a9669 1054 xen_ram_init(below_4g_mem_size, above_4g_mem_size, ram_size, ram_memory);
432d268c 1055
fb4bb2b5 1056 qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
9ce94e7c 1057
20581d20 1058 state->memory_listener = xen_memory_listener;
b4dd7802 1059 QLIST_INIT(&state->physmap);
f6790af6 1060 memory_listener_register(&state->memory_listener, &address_space_memory);
b4dd7802
AP
1061 state->log_for_dirtybit = NULL;
1062
ad35a7da
SS
1063 /* Initialize backend core & drivers */
1064 if (xen_be_init() != 0) {
1065 fprintf(stderr, "%s: xen backend core setup failed\n", __FUNCTION__);
1066 exit(1);
1067 }
1068 xen_be_register("console", &xen_console_ops);
37cdfcf1 1069 xen_be_register("vkbd", &xen_kbdmouse_ops);
ad35a7da 1070 xen_be_register("qdisk", &xen_blkdev_ops);
d1814e08 1071 xen_read_physmap(state);
ad35a7da 1072
29d3ccde
AP
1073 return 0;
1074}
9ce94e7c 1075
180640ea 1076void destroy_hvm_domain(bool reboot)
9ce94e7c
AS
1077{
1078 XenXC xc_handle;
1079 int sts;
1080
1081 xc_handle = xen_xc_interface_open(0, 0, 0);
1082 if (xc_handle == XC_HANDLER_INITIAL_VALUE) {
1083 fprintf(stderr, "Cannot acquire xenctrl handle\n");
1084 } else {
180640ea
JB
1085 sts = xc_domain_shutdown(xc_handle, xen_domid,
1086 reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
9ce94e7c 1087 if (sts != 0) {
180640ea
JB
1088 fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1089 "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1090 sts, strerror(errno));
9ce94e7c 1091 } else {
180640ea
JB
1092 fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1093 reboot ? "reboot" : "poweroff");
9ce94e7c
AS
1094 }
1095 xc_interface_close(xc_handle);
1096 }
1097}
c65adf9b
AK
1098
1099void xen_register_framebuffer(MemoryRegion *mr)
1100{
1101 framebuffer = mr;
1102}
eaab4d60
AK
1103
1104void xen_shutdown_fatal_error(const char *fmt, ...)
1105{
1106 va_list ap;
1107
1108 va_start(ap, fmt);
1109 vfprintf(stderr, fmt, ap);
1110 va_end(ap);
1111 fprintf(stderr, "Will destroy the domain.\n");
1112 /* destroy the domain */
1113 qemu_system_shutdown_request();
1114}
910b38e4
AP
1115
1116void xen_modified_memory(ram_addr_t start, ram_addr_t length)
1117{
1118 if (unlikely(xen_in_migration)) {
1119 int rc;
1120 ram_addr_t start_pfn, nb_pages;
1121
1122 if (length == 0) {
1123 length = TARGET_PAGE_SIZE;
1124 }
1125 start_pfn = start >> TARGET_PAGE_BITS;
1126 nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1127 - start_pfn;
1128 rc = xc_hvm_modified_memory(xen_xc, xen_domid, start_pfn, nb_pages);
1129 if (rc) {
1130 fprintf(stderr,
1131 "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1132 __func__, start, nb_pages, rc, strerror(-rc));
1133 }
1134 }
1135}
04b0de0e
WL
1136
1137void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1138{
1139 if (enable) {
1140 memory_global_dirty_log_start();
1141 } else {
1142 memory_global_dirty_log_stop();
1143 }
1144}