xref: /openbmc/qemu/hw/i386/xen/xen-hvm.c (revision db725815985654007ade0fd53590d613fd657208)
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  *
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.
9  */
10 
11 #include "qemu/osdep.h"
12 
13 #include "cpu.h"
14 #include "hw/pci/pci.h"
15 #include "hw/pci/pci_host.h"
16 #include "hw/i386/pc.h"
17 #include "hw/irq.h"
18 #include "hw/hw.h"
19 #include "hw/i386/apic-msidef.h"
20 #include "hw/xen/xen_common.h"
21 #include "hw/xen/xen-legacy-backend.h"
22 #include "hw/xen/xen-bus.h"
23 #include "qapi/error.h"
24 #include "qapi/qapi-commands-misc.h"
25 #include "qemu/error-report.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/range.h"
28 #include "sysemu/xen-mapcache.h"
29 #include "trace.h"
30 #include "exec/address-spaces.h"
31 
32 #include <xen/hvm/ioreq.h>
33 #include <xen/hvm/e820.h>
34 
35 //#define DEBUG_XEN_HVM
36 
37 #ifdef DEBUG_XEN_HVM
38 #define DPRINTF(fmt, ...) \
39     do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
40 #else
41 #define DPRINTF(fmt, ...) \
42     do { } while (0)
43 #endif
44 
45 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
46 static MemoryRegion *framebuffer;
47 static bool xen_in_migration;
48 
49 /* Compatibility with older version */
50 
51 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
52  * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
53  * needs to be included before this block and hw/xen/xen_common.h needs to
54  * be included before xen/hvm/ioreq.h
55  */
56 #ifndef IOREQ_TYPE_VMWARE_PORT
57 #define IOREQ_TYPE_VMWARE_PORT  3
58 struct vmware_regs {
59     uint32_t esi;
60     uint32_t edi;
61     uint32_t ebx;
62     uint32_t ecx;
63     uint32_t edx;
64 };
65 typedef struct vmware_regs vmware_regs_t;
66 
67 struct shared_vmport_iopage {
68     struct vmware_regs vcpu_vmport_regs[1];
69 };
70 typedef struct shared_vmport_iopage shared_vmport_iopage_t;
71 #endif
72 
73 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
74 {
75     return shared_page->vcpu_ioreq[i].vp_eport;
76 }
77 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
78 {
79     return &shared_page->vcpu_ioreq[vcpu];
80 }
81 
82 #define BUFFER_IO_MAX_DELAY  100
83 
84 typedef struct XenPhysmap {
85     hwaddr start_addr;
86     ram_addr_t size;
87     const char *name;
88     hwaddr phys_offset;
89 
90     QLIST_ENTRY(XenPhysmap) list;
91 } XenPhysmap;
92 
93 static QLIST_HEAD(, XenPhysmap) xen_physmap;
94 
95 typedef struct XenPciDevice {
96     PCIDevice *pci_dev;
97     uint32_t sbdf;
98     QLIST_ENTRY(XenPciDevice) entry;
99 } XenPciDevice;
100 
101 typedef struct XenIOState {
102     ioservid_t ioservid;
103     shared_iopage_t *shared_page;
104     shared_vmport_iopage_t *shared_vmport_page;
105     buffered_iopage_t *buffered_io_page;
106     QEMUTimer *buffered_io_timer;
107     CPUState **cpu_by_vcpu_id;
108     /* the evtchn port for polling the notification, */
109     evtchn_port_t *ioreq_local_port;
110     /* evtchn remote and local ports for buffered io */
111     evtchn_port_t bufioreq_remote_port;
112     evtchn_port_t bufioreq_local_port;
113     /* the evtchn fd for polling */
114     xenevtchn_handle *xce_handle;
115     /* which vcpu we are serving */
116     int send_vcpu;
117 
118     struct xs_handle *xenstore;
119     MemoryListener memory_listener;
120     MemoryListener io_listener;
121     QLIST_HEAD(, XenPciDevice) dev_list;
122     DeviceListener device_listener;
123     hwaddr free_phys_offset;
124     const XenPhysmap *log_for_dirtybit;
125     /* Buffer used by xen_sync_dirty_bitmap */
126     unsigned long *dirty_bitmap;
127 
128     Notifier exit;
129     Notifier suspend;
130     Notifier wakeup;
131 } XenIOState;
132 
133 /* Xen specific function for piix pci */
134 
135 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
136 {
137     return irq_num + ((pci_dev->devfn >> 3) << 2);
138 }
139 
140 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
141 {
142     xen_set_pci_intx_level(xen_domid, 0, 0, irq_num >> 2,
143                            irq_num & 3, level);
144 }
145 
146 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
147 {
148     int i;
149 
150     /* Scan for updates to PCI link routes (0x60-0x63). */
151     for (i = 0; i < len; i++) {
152         uint8_t v = (val >> (8 * i)) & 0xff;
153         if (v & 0x80) {
154             v = 0;
155         }
156         v &= 0xf;
157         if (((address + i) >= 0x60) && ((address + i) <= 0x63)) {
158             xen_set_pci_link_route(xen_domid, address + i - 0x60, v);
159         }
160     }
161 }
162 
163 int xen_is_pirq_msi(uint32_t msi_data)
164 {
165     /* If vector is 0, the msi is remapped into a pirq, passed as
166      * dest_id.
167      */
168     return ((msi_data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT) == 0;
169 }
170 
171 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
172 {
173     xen_inject_msi(xen_domid, addr, data);
174 }
175 
176 static void xen_suspend_notifier(Notifier *notifier, void *data)
177 {
178     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
179 }
180 
181 /* Xen Interrupt Controller */
182 
183 static void xen_set_irq(void *opaque, int irq, int level)
184 {
185     xen_set_isa_irq_level(xen_domid, irq, level);
186 }
187 
188 qemu_irq *xen_interrupt_controller_init(void)
189 {
190     return qemu_allocate_irqs(xen_set_irq, NULL, 16);
191 }
192 
193 /* Memory Ops */
194 
195 static void xen_ram_init(PCMachineState *pcms,
196                          ram_addr_t ram_size, MemoryRegion **ram_memory_p)
197 {
198     MemoryRegion *sysmem = get_system_memory();
199     ram_addr_t block_len;
200     uint64_t user_lowmem = object_property_get_uint(qdev_get_machine(),
201                                                     PC_MACHINE_MAX_RAM_BELOW_4G,
202                                                     &error_abort);
203 
204     /* Handle the machine opt max-ram-below-4g.  It is basically doing
205      * min(xen limit, user limit).
206      */
207     if (!user_lowmem) {
208         user_lowmem = HVM_BELOW_4G_RAM_END; /* default */
209     }
210     if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
211         user_lowmem = HVM_BELOW_4G_RAM_END;
212     }
213 
214     if (ram_size >= user_lowmem) {
215         pcms->above_4g_mem_size = ram_size - user_lowmem;
216         pcms->below_4g_mem_size = user_lowmem;
217     } else {
218         pcms->above_4g_mem_size = 0;
219         pcms->below_4g_mem_size = ram_size;
220     }
221     if (!pcms->above_4g_mem_size) {
222         block_len = ram_size;
223     } else {
224         /*
225          * Xen does not allocate the memory continuously, it keeps a
226          * hole of the size computed above or passed in.
227          */
228         block_len = (1ULL << 32) + pcms->above_4g_mem_size;
229     }
230     memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
231                            &error_fatal);
232     *ram_memory_p = &ram_memory;
233 
234     memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
235                              &ram_memory, 0, 0xa0000);
236     memory_region_add_subregion(sysmem, 0, &ram_640k);
237     /* Skip of the VGA IO memory space, it will be registered later by the VGA
238      * emulated device.
239      *
240      * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
241      * the Options ROM, so it is registered here as RAM.
242      */
243     memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
244                              &ram_memory, 0xc0000,
245                              pcms->below_4g_mem_size - 0xc0000);
246     memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
247     if (pcms->above_4g_mem_size > 0) {
248         memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
249                                  &ram_memory, 0x100000000ULL,
250                                  pcms->above_4g_mem_size);
251         memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
252     }
253 }
254 
255 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr,
256                    Error **errp)
257 {
258     unsigned long nr_pfn;
259     xen_pfn_t *pfn_list;
260     int i;
261 
262     if (runstate_check(RUN_STATE_INMIGRATE)) {
263         /* RAM already populated in Xen */
264         fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
265                 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
266                 __func__, size, ram_addr);
267         return;
268     }
269 
270     if (mr == &ram_memory) {
271         return;
272     }
273 
274     trace_xen_ram_alloc(ram_addr, size);
275 
276     nr_pfn = size >> TARGET_PAGE_BITS;
277     pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
278 
279     for (i = 0; i < nr_pfn; i++) {
280         pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
281     }
282 
283     if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
284         error_setg(errp, "xen: failed to populate ram at " RAM_ADDR_FMT,
285                    ram_addr);
286     }
287 
288     g_free(pfn_list);
289 }
290 
291 static XenPhysmap *get_physmapping(hwaddr start_addr, ram_addr_t size)
292 {
293     XenPhysmap *physmap = NULL;
294 
295     start_addr &= TARGET_PAGE_MASK;
296 
297     QLIST_FOREACH(physmap, &xen_physmap, list) {
298         if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
299             return physmap;
300         }
301     }
302     return NULL;
303 }
304 
305 static hwaddr xen_phys_offset_to_gaddr(hwaddr phys_offset, ram_addr_t size)
306 {
307     hwaddr addr = phys_offset & TARGET_PAGE_MASK;
308     XenPhysmap *physmap = NULL;
309 
310     QLIST_FOREACH(physmap, &xen_physmap, list) {
311         if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
312             return physmap->start_addr + (phys_offset - physmap->phys_offset);
313         }
314     }
315 
316     return phys_offset;
317 }
318 
319 #ifdef XEN_COMPAT_PHYSMAP
320 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
321 {
322     char path[80], value[17];
323 
324     snprintf(path, sizeof(path),
325             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
326             xen_domid, (uint64_t)physmap->phys_offset);
327     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->start_addr);
328     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
329         return -1;
330     }
331     snprintf(path, sizeof(path),
332             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
333             xen_domid, (uint64_t)physmap->phys_offset);
334     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->size);
335     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
336         return -1;
337     }
338     if (physmap->name) {
339         snprintf(path, sizeof(path),
340                 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
341                 xen_domid, (uint64_t)physmap->phys_offset);
342         if (!xs_write(state->xenstore, 0, path,
343                       physmap->name, strlen(physmap->name))) {
344             return -1;
345         }
346     }
347     return 0;
348 }
349 #else
350 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
351 {
352     return 0;
353 }
354 #endif
355 
356 static int xen_add_to_physmap(XenIOState *state,
357                               hwaddr start_addr,
358                               ram_addr_t size,
359                               MemoryRegion *mr,
360                               hwaddr offset_within_region)
361 {
362     unsigned long nr_pages;
363     int rc = 0;
364     XenPhysmap *physmap = NULL;
365     hwaddr pfn, start_gpfn;
366     hwaddr phys_offset = memory_region_get_ram_addr(mr);
367     const char *mr_name;
368 
369     if (get_physmapping(start_addr, size)) {
370         return 0;
371     }
372     if (size <= 0) {
373         return -1;
374     }
375 
376     /* Xen can only handle a single dirty log region for now and we want
377      * the linear framebuffer to be that region.
378      * Avoid tracking any regions that is not videoram and avoid tracking
379      * the legacy vga region. */
380     if (mr == framebuffer && start_addr > 0xbffff) {
381         goto go_physmap;
382     }
383     return -1;
384 
385 go_physmap:
386     DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
387             start_addr, start_addr + size);
388 
389     mr_name = memory_region_name(mr);
390 
391     physmap = g_malloc(sizeof(XenPhysmap));
392 
393     physmap->start_addr = start_addr;
394     physmap->size = size;
395     physmap->name = mr_name;
396     physmap->phys_offset = phys_offset;
397 
398     QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
399 
400     if (runstate_check(RUN_STATE_INMIGRATE)) {
401         /* Now when we have a physmap entry we can replace a dummy mapping with
402          * a real one of guest foreign memory. */
403         uint8_t *p = xen_replace_cache_entry(phys_offset, start_addr, size);
404         assert(p && p == memory_region_get_ram_ptr(mr));
405 
406         return 0;
407     }
408 
409     pfn = phys_offset >> TARGET_PAGE_BITS;
410     start_gpfn = start_addr >> TARGET_PAGE_BITS;
411     nr_pages = size >> TARGET_PAGE_BITS;
412     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, nr_pages, pfn,
413                                         start_gpfn);
414     if (rc) {
415         int saved_errno = errno;
416 
417         error_report("relocate_memory %lu pages from GFN %"HWADDR_PRIx
418                      " to GFN %"HWADDR_PRIx" failed: %s",
419                      nr_pages, pfn, start_gpfn, strerror(saved_errno));
420         errno = saved_errno;
421         return -1;
422     }
423 
424     rc = xendevicemodel_pin_memory_cacheattr(xen_dmod, xen_domid,
425                                    start_addr >> TARGET_PAGE_BITS,
426                                    (start_addr + size - 1) >> TARGET_PAGE_BITS,
427                                    XEN_DOMCTL_MEM_CACHEATTR_WB);
428     if (rc) {
429         error_report("pin_memory_cacheattr failed: %s", strerror(errno));
430     }
431     return xen_save_physmap(state, physmap);
432 }
433 
434 static int xen_remove_from_physmap(XenIOState *state,
435                                    hwaddr start_addr,
436                                    ram_addr_t size)
437 {
438     int rc = 0;
439     XenPhysmap *physmap = NULL;
440     hwaddr phys_offset = 0;
441 
442     physmap = get_physmapping(start_addr, size);
443     if (physmap == NULL) {
444         return -1;
445     }
446 
447     phys_offset = physmap->phys_offset;
448     size = physmap->size;
449 
450     DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
451             "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
452 
453     size >>= TARGET_PAGE_BITS;
454     start_addr >>= TARGET_PAGE_BITS;
455     phys_offset >>= TARGET_PAGE_BITS;
456     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, size, start_addr,
457                                         phys_offset);
458     if (rc) {
459         int saved_errno = errno;
460 
461         error_report("relocate_memory "RAM_ADDR_FMT" pages"
462                      " from GFN %"HWADDR_PRIx
463                      " to GFN %"HWADDR_PRIx" failed: %s",
464                      size, start_addr, phys_offset, strerror(saved_errno));
465         errno = saved_errno;
466         return -1;
467     }
468 
469     QLIST_REMOVE(physmap, list);
470     if (state->log_for_dirtybit == physmap) {
471         state->log_for_dirtybit = NULL;
472         g_free(state->dirty_bitmap);
473         state->dirty_bitmap = NULL;
474     }
475     g_free(physmap);
476 
477     return 0;
478 }
479 
480 static void xen_set_memory(struct MemoryListener *listener,
481                            MemoryRegionSection *section,
482                            bool add)
483 {
484     XenIOState *state = container_of(listener, XenIOState, memory_listener);
485     hwaddr start_addr = section->offset_within_address_space;
486     ram_addr_t size = int128_get64(section->size);
487     bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA);
488     hvmmem_type_t mem_type;
489 
490     if (section->mr == &ram_memory) {
491         return;
492     } else {
493         if (add) {
494             xen_map_memory_section(xen_domid, state->ioservid,
495                                    section);
496         } else {
497             xen_unmap_memory_section(xen_domid, state->ioservid,
498                                      section);
499         }
500     }
501 
502     if (!memory_region_is_ram(section->mr)) {
503         return;
504     }
505 
506     if (log_dirty != add) {
507         return;
508     }
509 
510     trace_xen_client_set_memory(start_addr, size, log_dirty);
511 
512     start_addr &= TARGET_PAGE_MASK;
513     size = TARGET_PAGE_ALIGN(size);
514 
515     if (add) {
516         if (!memory_region_is_rom(section->mr)) {
517             xen_add_to_physmap(state, start_addr, size,
518                                section->mr, section->offset_within_region);
519         } else {
520             mem_type = HVMMEM_ram_ro;
521             if (xen_set_mem_type(xen_domid, mem_type,
522                                  start_addr >> TARGET_PAGE_BITS,
523                                  size >> TARGET_PAGE_BITS)) {
524                 DPRINTF("xen_set_mem_type error, addr: "TARGET_FMT_plx"\n",
525                         start_addr);
526             }
527         }
528     } else {
529         if (xen_remove_from_physmap(state, start_addr, size) < 0) {
530             DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
531         }
532     }
533 }
534 
535 static void xen_region_add(MemoryListener *listener,
536                            MemoryRegionSection *section)
537 {
538     memory_region_ref(section->mr);
539     xen_set_memory(listener, section, true);
540 }
541 
542 static void xen_region_del(MemoryListener *listener,
543                            MemoryRegionSection *section)
544 {
545     xen_set_memory(listener, section, false);
546     memory_region_unref(section->mr);
547 }
548 
549 static void xen_io_add(MemoryListener *listener,
550                        MemoryRegionSection *section)
551 {
552     XenIOState *state = container_of(listener, XenIOState, io_listener);
553     MemoryRegion *mr = section->mr;
554 
555     if (mr->ops == &unassigned_io_ops) {
556         return;
557     }
558 
559     memory_region_ref(mr);
560 
561     xen_map_io_section(xen_domid, state->ioservid, section);
562 }
563 
564 static void xen_io_del(MemoryListener *listener,
565                        MemoryRegionSection *section)
566 {
567     XenIOState *state = container_of(listener, XenIOState, io_listener);
568     MemoryRegion *mr = section->mr;
569 
570     if (mr->ops == &unassigned_io_ops) {
571         return;
572     }
573 
574     xen_unmap_io_section(xen_domid, state->ioservid, section);
575 
576     memory_region_unref(mr);
577 }
578 
579 static void xen_device_realize(DeviceListener *listener,
580                                DeviceState *dev)
581 {
582     XenIOState *state = container_of(listener, XenIOState, device_listener);
583 
584     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
585         PCIDevice *pci_dev = PCI_DEVICE(dev);
586         XenPciDevice *xendev = g_new(XenPciDevice, 1);
587 
588         xendev->pci_dev = pci_dev;
589         xendev->sbdf = PCI_BUILD_BDF(pci_dev_bus_num(pci_dev),
590                                      pci_dev->devfn);
591         QLIST_INSERT_HEAD(&state->dev_list, xendev, entry);
592 
593         xen_map_pcidev(xen_domid, state->ioservid, pci_dev);
594     }
595 }
596 
597 static void xen_device_unrealize(DeviceListener *listener,
598                                  DeviceState *dev)
599 {
600     XenIOState *state = container_of(listener, XenIOState, device_listener);
601 
602     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
603         PCIDevice *pci_dev = PCI_DEVICE(dev);
604         XenPciDevice *xendev, *next;
605 
606         xen_unmap_pcidev(xen_domid, state->ioservid, pci_dev);
607 
608         QLIST_FOREACH_SAFE(xendev, &state->dev_list, entry, next) {
609             if (xendev->pci_dev == pci_dev) {
610                 QLIST_REMOVE(xendev, entry);
611                 g_free(xendev);
612                 break;
613             }
614         }
615     }
616 }
617 
618 static void xen_sync_dirty_bitmap(XenIOState *state,
619                                   hwaddr start_addr,
620                                   ram_addr_t size)
621 {
622     hwaddr npages = size >> TARGET_PAGE_BITS;
623     const int width = sizeof(unsigned long) * 8;
624     size_t bitmap_size = DIV_ROUND_UP(npages, width);
625     int rc, i, j;
626     const XenPhysmap *physmap = NULL;
627 
628     physmap = get_physmapping(start_addr, size);
629     if (physmap == NULL) {
630         /* not handled */
631         return;
632     }
633 
634     if (state->log_for_dirtybit == NULL) {
635         state->log_for_dirtybit = physmap;
636         state->dirty_bitmap = g_new(unsigned long, bitmap_size);
637     } else if (state->log_for_dirtybit != physmap) {
638         /* Only one range for dirty bitmap can be tracked. */
639         return;
640     }
641 
642     rc = xen_track_dirty_vram(xen_domid, start_addr >> TARGET_PAGE_BITS,
643                               npages, state->dirty_bitmap);
644     if (rc < 0) {
645 #ifndef ENODATA
646 #define ENODATA  ENOENT
647 #endif
648         if (errno == ENODATA) {
649             memory_region_set_dirty(framebuffer, 0, size);
650             DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
651                     ", 0x" TARGET_FMT_plx "): %s\n",
652                     start_addr, start_addr + size, strerror(errno));
653         }
654         return;
655     }
656 
657     for (i = 0; i < bitmap_size; i++) {
658         unsigned long map = state->dirty_bitmap[i];
659         while (map != 0) {
660             j = ctzl(map);
661             map &= ~(1ul << j);
662             memory_region_set_dirty(framebuffer,
663                                     (i * width + j) * TARGET_PAGE_SIZE,
664                                     TARGET_PAGE_SIZE);
665         };
666     }
667 }
668 
669 static void xen_log_start(MemoryListener *listener,
670                           MemoryRegionSection *section,
671                           int old, int new)
672 {
673     XenIOState *state = container_of(listener, XenIOState, memory_listener);
674 
675     if (new & ~old & (1 << DIRTY_MEMORY_VGA)) {
676         xen_sync_dirty_bitmap(state, section->offset_within_address_space,
677                               int128_get64(section->size));
678     }
679 }
680 
681 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section,
682                          int old, int new)
683 {
684     XenIOState *state = container_of(listener, XenIOState, memory_listener);
685 
686     if (old & ~new & (1 << DIRTY_MEMORY_VGA)) {
687         state->log_for_dirtybit = NULL;
688         g_free(state->dirty_bitmap);
689         state->dirty_bitmap = NULL;
690         /* Disable dirty bit tracking */
691         xen_track_dirty_vram(xen_domid, 0, 0, NULL);
692     }
693 }
694 
695 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
696 {
697     XenIOState *state = container_of(listener, XenIOState, memory_listener);
698 
699     xen_sync_dirty_bitmap(state, section->offset_within_address_space,
700                           int128_get64(section->size));
701 }
702 
703 static void xen_log_global_start(MemoryListener *listener)
704 {
705     if (xen_enabled()) {
706         xen_in_migration = true;
707     }
708 }
709 
710 static void xen_log_global_stop(MemoryListener *listener)
711 {
712     xen_in_migration = false;
713 }
714 
715 static MemoryListener xen_memory_listener = {
716     .region_add = xen_region_add,
717     .region_del = xen_region_del,
718     .log_start = xen_log_start,
719     .log_stop = xen_log_stop,
720     .log_sync = xen_log_sync,
721     .log_global_start = xen_log_global_start,
722     .log_global_stop = xen_log_global_stop,
723     .priority = 10,
724 };
725 
726 static MemoryListener xen_io_listener = {
727     .region_add = xen_io_add,
728     .region_del = xen_io_del,
729     .priority = 10,
730 };
731 
732 static DeviceListener xen_device_listener = {
733     .realize = xen_device_realize,
734     .unrealize = xen_device_unrealize,
735 };
736 
737 /* get the ioreq packets from share mem */
738 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
739 {
740     ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
741 
742     if (req->state != STATE_IOREQ_READY) {
743         DPRINTF("I/O request not ready: "
744                 "%x, ptr: %x, port: %"PRIx64", "
745                 "data: %"PRIx64", count: %u, size: %u\n",
746                 req->state, req->data_is_ptr, req->addr,
747                 req->data, req->count, req->size);
748         return NULL;
749     }
750 
751     xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
752 
753     req->state = STATE_IOREQ_INPROCESS;
754     return req;
755 }
756 
757 /* use poll to get the port notification */
758 /* ioreq_vec--out,the */
759 /* retval--the number of ioreq packet */
760 static ioreq_t *cpu_get_ioreq(XenIOState *state)
761 {
762     MachineState *ms = MACHINE(qdev_get_machine());
763     unsigned int max_cpus = ms->smp.max_cpus;
764     int i;
765     evtchn_port_t port;
766 
767     port = xenevtchn_pending(state->xce_handle);
768     if (port == state->bufioreq_local_port) {
769         timer_mod(state->buffered_io_timer,
770                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
771         return NULL;
772     }
773 
774     if (port != -1) {
775         for (i = 0; i < max_cpus; i++) {
776             if (state->ioreq_local_port[i] == port) {
777                 break;
778             }
779         }
780 
781         if (i == max_cpus) {
782             hw_error("Fatal error while trying to get io event!\n");
783         }
784 
785         /* unmask the wanted port again */
786         xenevtchn_unmask(state->xce_handle, port);
787 
788         /* get the io packet from shared memory */
789         state->send_vcpu = i;
790         return cpu_get_ioreq_from_shared_memory(state, i);
791     }
792 
793     /* read error or read nothing */
794     return NULL;
795 }
796 
797 static uint32_t do_inp(uint32_t addr, unsigned long size)
798 {
799     switch (size) {
800         case 1:
801             return cpu_inb(addr);
802         case 2:
803             return cpu_inw(addr);
804         case 4:
805             return cpu_inl(addr);
806         default:
807             hw_error("inp: bad size: %04x %lx", addr, size);
808     }
809 }
810 
811 static void do_outp(uint32_t addr,
812         unsigned long size, uint32_t val)
813 {
814     switch (size) {
815         case 1:
816             return cpu_outb(addr, val);
817         case 2:
818             return cpu_outw(addr, val);
819         case 4:
820             return cpu_outl(addr, val);
821         default:
822             hw_error("outp: bad size: %04x %lx", addr, size);
823     }
824 }
825 
826 /*
827  * Helper functions which read/write an object from/to physical guest
828  * memory, as part of the implementation of an ioreq.
829  *
830  * Equivalent to
831  *   cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
832  *                          val, req->size, 0/1)
833  * except without the integer overflow problems.
834  */
835 static void rw_phys_req_item(hwaddr addr,
836                              ioreq_t *req, uint32_t i, void *val, int rw)
837 {
838     /* Do everything unsigned so overflow just results in a truncated result
839      * and accesses to undesired parts of guest memory, which is up
840      * to the guest */
841     hwaddr offset = (hwaddr)req->size * i;
842     if (req->df) {
843         addr -= offset;
844     } else {
845         addr += offset;
846     }
847     cpu_physical_memory_rw(addr, val, req->size, rw);
848 }
849 
850 static inline void read_phys_req_item(hwaddr addr,
851                                       ioreq_t *req, uint32_t i, void *val)
852 {
853     rw_phys_req_item(addr, req, i, val, 0);
854 }
855 static inline void write_phys_req_item(hwaddr addr,
856                                        ioreq_t *req, uint32_t i, void *val)
857 {
858     rw_phys_req_item(addr, req, i, val, 1);
859 }
860 
861 
862 static void cpu_ioreq_pio(ioreq_t *req)
863 {
864     uint32_t i;
865 
866     trace_cpu_ioreq_pio(req, req->dir, req->df, req->data_is_ptr, req->addr,
867                          req->data, req->count, req->size);
868 
869     if (req->size > sizeof(uint32_t)) {
870         hw_error("PIO: bad size (%u)", req->size);
871     }
872 
873     if (req->dir == IOREQ_READ) {
874         if (!req->data_is_ptr) {
875             req->data = do_inp(req->addr, req->size);
876             trace_cpu_ioreq_pio_read_reg(req, req->data, req->addr,
877                                          req->size);
878         } else {
879             uint32_t tmp;
880 
881             for (i = 0; i < req->count; i++) {
882                 tmp = do_inp(req->addr, req->size);
883                 write_phys_req_item(req->data, req, i, &tmp);
884             }
885         }
886     } else if (req->dir == IOREQ_WRITE) {
887         if (!req->data_is_ptr) {
888             trace_cpu_ioreq_pio_write_reg(req, req->data, req->addr,
889                                           req->size);
890             do_outp(req->addr, req->size, req->data);
891         } else {
892             for (i = 0; i < req->count; i++) {
893                 uint32_t tmp = 0;
894 
895                 read_phys_req_item(req->data, req, i, &tmp);
896                 do_outp(req->addr, req->size, tmp);
897             }
898         }
899     }
900 }
901 
902 static void cpu_ioreq_move(ioreq_t *req)
903 {
904     uint32_t i;
905 
906     trace_cpu_ioreq_move(req, req->dir, req->df, req->data_is_ptr, req->addr,
907                          req->data, req->count, req->size);
908 
909     if (req->size > sizeof(req->data)) {
910         hw_error("MMIO: bad size (%u)", req->size);
911     }
912 
913     if (!req->data_is_ptr) {
914         if (req->dir == IOREQ_READ) {
915             for (i = 0; i < req->count; i++) {
916                 read_phys_req_item(req->addr, req, i, &req->data);
917             }
918         } else if (req->dir == IOREQ_WRITE) {
919             for (i = 0; i < req->count; i++) {
920                 write_phys_req_item(req->addr, req, i, &req->data);
921             }
922         }
923     } else {
924         uint64_t tmp;
925 
926         if (req->dir == IOREQ_READ) {
927             for (i = 0; i < req->count; i++) {
928                 read_phys_req_item(req->addr, req, i, &tmp);
929                 write_phys_req_item(req->data, req, i, &tmp);
930             }
931         } else if (req->dir == IOREQ_WRITE) {
932             for (i = 0; i < req->count; i++) {
933                 read_phys_req_item(req->data, req, i, &tmp);
934                 write_phys_req_item(req->addr, req, i, &tmp);
935             }
936         }
937     }
938 }
939 
940 static void cpu_ioreq_config(XenIOState *state, ioreq_t *req)
941 {
942     uint32_t sbdf = req->addr >> 32;
943     uint32_t reg = req->addr;
944     XenPciDevice *xendev;
945 
946     if (req->size != sizeof(uint8_t) && req->size != sizeof(uint16_t) &&
947         req->size != sizeof(uint32_t)) {
948         hw_error("PCI config access: bad size (%u)", req->size);
949     }
950 
951     if (req->count != 1) {
952         hw_error("PCI config access: bad count (%u)", req->count);
953     }
954 
955     QLIST_FOREACH(xendev, &state->dev_list, entry) {
956         if (xendev->sbdf != sbdf) {
957             continue;
958         }
959 
960         if (!req->data_is_ptr) {
961             if (req->dir == IOREQ_READ) {
962                 req->data = pci_host_config_read_common(
963                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
964                     req->size);
965                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
966                                             req->size, req->data);
967             } else if (req->dir == IOREQ_WRITE) {
968                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
969                                              req->size, req->data);
970                 pci_host_config_write_common(
971                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
972                     req->data, req->size);
973             }
974         } else {
975             uint32_t tmp;
976 
977             if (req->dir == IOREQ_READ) {
978                 tmp = pci_host_config_read_common(
979                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
980                     req->size);
981                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
982                                             req->size, tmp);
983                 write_phys_req_item(req->data, req, 0, &tmp);
984             } else if (req->dir == IOREQ_WRITE) {
985                 read_phys_req_item(req->data, req, 0, &tmp);
986                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
987                                              req->size, tmp);
988                 pci_host_config_write_common(
989                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
990                     tmp, req->size);
991             }
992         }
993     }
994 }
995 
996 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
997 {
998     X86CPU *cpu;
999     CPUX86State *env;
1000 
1001     cpu = X86_CPU(current_cpu);
1002     env = &cpu->env;
1003     env->regs[R_EAX] = req->data;
1004     env->regs[R_EBX] = vmport_regs->ebx;
1005     env->regs[R_ECX] = vmport_regs->ecx;
1006     env->regs[R_EDX] = vmport_regs->edx;
1007     env->regs[R_ESI] = vmport_regs->esi;
1008     env->regs[R_EDI] = vmport_regs->edi;
1009 }
1010 
1011 static void regs_from_cpu(vmware_regs_t *vmport_regs)
1012 {
1013     X86CPU *cpu = X86_CPU(current_cpu);
1014     CPUX86State *env = &cpu->env;
1015 
1016     vmport_regs->ebx = env->regs[R_EBX];
1017     vmport_regs->ecx = env->regs[R_ECX];
1018     vmport_regs->edx = env->regs[R_EDX];
1019     vmport_regs->esi = env->regs[R_ESI];
1020     vmport_regs->edi = env->regs[R_EDI];
1021 }
1022 
1023 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
1024 {
1025     vmware_regs_t *vmport_regs;
1026 
1027     assert(state->shared_vmport_page);
1028     vmport_regs =
1029         &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
1030     QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
1031 
1032     current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
1033     regs_to_cpu(vmport_regs, req);
1034     cpu_ioreq_pio(req);
1035     regs_from_cpu(vmport_regs);
1036     current_cpu = NULL;
1037 }
1038 
1039 static void handle_ioreq(XenIOState *state, ioreq_t *req)
1040 {
1041     trace_handle_ioreq(req, req->type, req->dir, req->df, req->data_is_ptr,
1042                        req->addr, req->data, req->count, req->size);
1043 
1044     if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
1045             (req->size < sizeof (target_ulong))) {
1046         req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
1047     }
1048 
1049     if (req->dir == IOREQ_WRITE)
1050         trace_handle_ioreq_write(req, req->type, req->df, req->data_is_ptr,
1051                                  req->addr, req->data, req->count, req->size);
1052 
1053     switch (req->type) {
1054         case IOREQ_TYPE_PIO:
1055             cpu_ioreq_pio(req);
1056             break;
1057         case IOREQ_TYPE_COPY:
1058             cpu_ioreq_move(req);
1059             break;
1060         case IOREQ_TYPE_VMWARE_PORT:
1061             handle_vmport_ioreq(state, req);
1062             break;
1063         case IOREQ_TYPE_TIMEOFFSET:
1064             break;
1065         case IOREQ_TYPE_INVALIDATE:
1066             xen_invalidate_map_cache();
1067             break;
1068         case IOREQ_TYPE_PCI_CONFIG:
1069             cpu_ioreq_config(state, req);
1070             break;
1071         default:
1072             hw_error("Invalid ioreq type 0x%x\n", req->type);
1073     }
1074     if (req->dir == IOREQ_READ) {
1075         trace_handle_ioreq_read(req, req->type, req->df, req->data_is_ptr,
1076                                 req->addr, req->data, req->count, req->size);
1077     }
1078 }
1079 
1080 static int handle_buffered_iopage(XenIOState *state)
1081 {
1082     buffered_iopage_t *buf_page = state->buffered_io_page;
1083     buf_ioreq_t *buf_req = NULL;
1084     ioreq_t req;
1085     int qw;
1086 
1087     if (!buf_page) {
1088         return 0;
1089     }
1090 
1091     memset(&req, 0x00, sizeof(req));
1092     req.state = STATE_IOREQ_READY;
1093     req.count = 1;
1094     req.dir = IOREQ_WRITE;
1095 
1096     for (;;) {
1097         uint32_t rdptr = buf_page->read_pointer, wrptr;
1098 
1099         xen_rmb();
1100         wrptr = buf_page->write_pointer;
1101         xen_rmb();
1102         if (rdptr != buf_page->read_pointer) {
1103             continue;
1104         }
1105         if (rdptr == wrptr) {
1106             break;
1107         }
1108         buf_req = &buf_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
1109         req.size = 1U << buf_req->size;
1110         req.addr = buf_req->addr;
1111         req.data = buf_req->data;
1112         req.type = buf_req->type;
1113         xen_rmb();
1114         qw = (req.size == 8);
1115         if (qw) {
1116             if (rdptr + 1 == wrptr) {
1117                 hw_error("Incomplete quad word buffered ioreq");
1118             }
1119             buf_req = &buf_page->buf_ioreq[(rdptr + 1) %
1120                                            IOREQ_BUFFER_SLOT_NUM];
1121             req.data |= ((uint64_t)buf_req->data) << 32;
1122             xen_rmb();
1123         }
1124 
1125         handle_ioreq(state, &req);
1126 
1127         /* Only req.data may get updated by handle_ioreq(), albeit even that
1128          * should not happen as such data would never make it to the guest (we
1129          * can only usefully see writes here after all).
1130          */
1131         assert(req.state == STATE_IOREQ_READY);
1132         assert(req.count == 1);
1133         assert(req.dir == IOREQ_WRITE);
1134         assert(!req.data_is_ptr);
1135 
1136         atomic_add(&buf_page->read_pointer, qw + 1);
1137     }
1138 
1139     return req.count;
1140 }
1141 
1142 static void handle_buffered_io(void *opaque)
1143 {
1144     XenIOState *state = opaque;
1145 
1146     if (handle_buffered_iopage(state)) {
1147         timer_mod(state->buffered_io_timer,
1148                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
1149     } else {
1150         timer_del(state->buffered_io_timer);
1151         xenevtchn_unmask(state->xce_handle, state->bufioreq_local_port);
1152     }
1153 }
1154 
1155 static void cpu_handle_ioreq(void *opaque)
1156 {
1157     XenIOState *state = opaque;
1158     ioreq_t *req = cpu_get_ioreq(state);
1159 
1160     handle_buffered_iopage(state);
1161     if (req) {
1162         ioreq_t copy = *req;
1163 
1164         xen_rmb();
1165         handle_ioreq(state, &copy);
1166         req->data = copy.data;
1167 
1168         if (req->state != STATE_IOREQ_INPROCESS) {
1169             fprintf(stderr, "Badness in I/O request ... not in service?!: "
1170                     "%x, ptr: %x, port: %"PRIx64", "
1171                     "data: %"PRIx64", count: %u, size: %u, type: %u\n",
1172                     req->state, req->data_is_ptr, req->addr,
1173                     req->data, req->count, req->size, req->type);
1174             destroy_hvm_domain(false);
1175             return;
1176         }
1177 
1178         xen_wmb(); /* Update ioreq contents /then/ update state. */
1179 
1180         /*
1181          * We do this before we send the response so that the tools
1182          * have the opportunity to pick up on the reset before the
1183          * guest resumes and does a hlt with interrupts disabled which
1184          * causes Xen to powerdown the domain.
1185          */
1186         if (runstate_is_running()) {
1187             ShutdownCause request;
1188 
1189             if (qemu_shutdown_requested_get()) {
1190                 destroy_hvm_domain(false);
1191             }
1192             request = qemu_reset_requested_get();
1193             if (request) {
1194                 qemu_system_reset(request);
1195                 destroy_hvm_domain(true);
1196             }
1197         }
1198 
1199         req->state = STATE_IORESP_READY;
1200         xenevtchn_notify(state->xce_handle,
1201                          state->ioreq_local_port[state->send_vcpu]);
1202     }
1203 }
1204 
1205 static void xen_main_loop_prepare(XenIOState *state)
1206 {
1207     int evtchn_fd = -1;
1208 
1209     if (state->xce_handle != NULL) {
1210         evtchn_fd = xenevtchn_fd(state->xce_handle);
1211     }
1212 
1213     state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
1214                                                  state);
1215 
1216     if (evtchn_fd != -1) {
1217         CPUState *cpu_state;
1218 
1219         DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
1220         CPU_FOREACH(cpu_state) {
1221             DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1222                     __func__, cpu_state->cpu_index, cpu_state);
1223             state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
1224         }
1225         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
1226     }
1227 }
1228 
1229 
1230 static void xen_hvm_change_state_handler(void *opaque, int running,
1231                                          RunState rstate)
1232 {
1233     XenIOState *state = opaque;
1234 
1235     if (running) {
1236         xen_main_loop_prepare(state);
1237     }
1238 
1239     xen_set_ioreq_server_state(xen_domid,
1240                                state->ioservid,
1241                                (rstate == RUN_STATE_RUNNING));
1242 }
1243 
1244 static void xen_exit_notifier(Notifier *n, void *data)
1245 {
1246     XenIOState *state = container_of(n, XenIOState, exit);
1247 
1248     xenevtchn_close(state->xce_handle);
1249     xs_daemon_close(state->xenstore);
1250 }
1251 
1252 #ifdef XEN_COMPAT_PHYSMAP
1253 static void xen_read_physmap(XenIOState *state)
1254 {
1255     XenPhysmap *physmap = NULL;
1256     unsigned int len, num, i;
1257     char path[80], *value = NULL;
1258     char **entries = NULL;
1259 
1260     snprintf(path, sizeof(path),
1261             "/local/domain/0/device-model/%d/physmap", xen_domid);
1262     entries = xs_directory(state->xenstore, 0, path, &num);
1263     if (entries == NULL)
1264         return;
1265 
1266     for (i = 0; i < num; i++) {
1267         physmap = g_malloc(sizeof (XenPhysmap));
1268         physmap->phys_offset = strtoull(entries[i], NULL, 16);
1269         snprintf(path, sizeof(path),
1270                 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1271                 xen_domid, entries[i]);
1272         value = xs_read(state->xenstore, 0, path, &len);
1273         if (value == NULL) {
1274             g_free(physmap);
1275             continue;
1276         }
1277         physmap->start_addr = strtoull(value, NULL, 16);
1278         free(value);
1279 
1280         snprintf(path, sizeof(path),
1281                 "/local/domain/0/device-model/%d/physmap/%s/size",
1282                 xen_domid, entries[i]);
1283         value = xs_read(state->xenstore, 0, path, &len);
1284         if (value == NULL) {
1285             g_free(physmap);
1286             continue;
1287         }
1288         physmap->size = strtoull(value, NULL, 16);
1289         free(value);
1290 
1291         snprintf(path, sizeof(path),
1292                 "/local/domain/0/device-model/%d/physmap/%s/name",
1293                 xen_domid, entries[i]);
1294         physmap->name = xs_read(state->xenstore, 0, path, &len);
1295 
1296         QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
1297     }
1298     free(entries);
1299 }
1300 #else
1301 static void xen_read_physmap(XenIOState *state)
1302 {
1303 }
1304 #endif
1305 
1306 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1307 {
1308     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1309 }
1310 
1311 static int xen_map_ioreq_server(XenIOState *state)
1312 {
1313     void *addr = NULL;
1314     xenforeignmemory_resource_handle *fres;
1315     xen_pfn_t ioreq_pfn;
1316     xen_pfn_t bufioreq_pfn;
1317     evtchn_port_t bufioreq_evtchn;
1318     int rc;
1319 
1320     /*
1321      * Attempt to map using the resource API and fall back to normal
1322      * foreign mapping if this is not supported.
1323      */
1324     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_bufioreq != 0);
1325     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_ioreq(0) != 1);
1326     fres = xenforeignmemory_map_resource(xen_fmem, xen_domid,
1327                                          XENMEM_resource_ioreq_server,
1328                                          state->ioservid, 0, 2,
1329                                          &addr,
1330                                          PROT_READ | PROT_WRITE, 0);
1331     if (fres != NULL) {
1332         trace_xen_map_resource_ioreq(state->ioservid, addr);
1333         state->buffered_io_page = addr;
1334         state->shared_page = addr + TARGET_PAGE_SIZE;
1335     } else if (errno != EOPNOTSUPP) {
1336         error_report("failed to map ioreq server resources: error %d handle=%p",
1337                      errno, xen_xc);
1338         return -1;
1339     }
1340 
1341     rc = xen_get_ioreq_server_info(xen_domid, state->ioservid,
1342                                    (state->shared_page == NULL) ?
1343                                    &ioreq_pfn : NULL,
1344                                    (state->buffered_io_page == NULL) ?
1345                                    &bufioreq_pfn : NULL,
1346                                    &bufioreq_evtchn);
1347     if (rc < 0) {
1348         error_report("failed to get ioreq server info: error %d handle=%p",
1349                      errno, xen_xc);
1350         return rc;
1351     }
1352 
1353     if (state->shared_page == NULL) {
1354         DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1355 
1356         state->shared_page = xenforeignmemory_map(xen_fmem, xen_domid,
1357                                                   PROT_READ | PROT_WRITE,
1358                                                   1, &ioreq_pfn, NULL);
1359         if (state->shared_page == NULL) {
1360             error_report("map shared IO page returned error %d handle=%p",
1361                          errno, xen_xc);
1362         }
1363     }
1364 
1365     if (state->buffered_io_page == NULL) {
1366         DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
1367 
1368         state->buffered_io_page = xenforeignmemory_map(xen_fmem, xen_domid,
1369                                                        PROT_READ | PROT_WRITE,
1370                                                        1, &bufioreq_pfn,
1371                                                        NULL);
1372         if (state->buffered_io_page == NULL) {
1373             error_report("map buffered IO page returned error %d", errno);
1374             return -1;
1375         }
1376     }
1377 
1378     if (state->shared_page == NULL || state->buffered_io_page == NULL) {
1379         return -1;
1380     }
1381 
1382     DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
1383 
1384     state->bufioreq_remote_port = bufioreq_evtchn;
1385 
1386     return 0;
1387 }
1388 
1389 void xen_hvm_init(PCMachineState *pcms, MemoryRegion **ram_memory)
1390 {
1391     MachineState *ms = MACHINE(pcms);
1392     unsigned int max_cpus = ms->smp.max_cpus;
1393     int i, rc;
1394     xen_pfn_t ioreq_pfn;
1395     XenIOState *state;
1396 
1397     state = g_malloc0(sizeof (XenIOState));
1398 
1399     state->xce_handle = xenevtchn_open(NULL, 0);
1400     if (state->xce_handle == NULL) {
1401         perror("xen: event channel open");
1402         goto err;
1403     }
1404 
1405     state->xenstore = xs_daemon_open();
1406     if (state->xenstore == NULL) {
1407         perror("xen: xenstore open");
1408         goto err;
1409     }
1410 
1411     xen_create_ioreq_server(xen_domid, &state->ioservid);
1412 
1413     state->exit.notify = xen_exit_notifier;
1414     qemu_add_exit_notifier(&state->exit);
1415 
1416     state->suspend.notify = xen_suspend_notifier;
1417     qemu_register_suspend_notifier(&state->suspend);
1418 
1419     state->wakeup.notify = xen_wakeup_notifier;
1420     qemu_register_wakeup_notifier(&state->wakeup);
1421 
1422     /*
1423      * Register wake-up support in QMP query-current-machine API
1424      */
1425     qemu_register_wakeup_support();
1426 
1427     rc = xen_map_ioreq_server(state);
1428     if (rc < 0) {
1429         goto err;
1430     }
1431 
1432     rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1433     if (!rc) {
1434         DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1435         state->shared_vmport_page =
1436             xenforeignmemory_map(xen_fmem, xen_domid, PROT_READ|PROT_WRITE,
1437                                  1, &ioreq_pfn, NULL);
1438         if (state->shared_vmport_page == NULL) {
1439             error_report("map shared vmport IO page returned error %d handle=%p",
1440                          errno, xen_xc);
1441             goto err;
1442         }
1443     } else if (rc != -ENOSYS) {
1444         error_report("get vmport regs pfn returned error %d, rc=%d",
1445                      errno, rc);
1446         goto err;
1447     }
1448 
1449     /* Note: cpus is empty at this point in init */
1450     state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1451 
1452     rc = xen_set_ioreq_server_state(xen_domid, state->ioservid, true);
1453     if (rc < 0) {
1454         error_report("failed to enable ioreq server info: error %d handle=%p",
1455                      errno, xen_xc);
1456         goto err;
1457     }
1458 
1459     state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1460 
1461     /* FIXME: how about if we overflow the page here? */
1462     for (i = 0; i < max_cpus; i++) {
1463         rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1464                                         xen_vcpu_eport(state->shared_page, i));
1465         if (rc == -1) {
1466             error_report("shared evtchn %d bind error %d", i, errno);
1467             goto err;
1468         }
1469         state->ioreq_local_port[i] = rc;
1470     }
1471 
1472     rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1473                                     state->bufioreq_remote_port);
1474     if (rc == -1) {
1475         error_report("buffered evtchn bind error %d", errno);
1476         goto err;
1477     }
1478     state->bufioreq_local_port = rc;
1479 
1480     /* Init RAM management */
1481 #ifdef XEN_COMPAT_PHYSMAP
1482     xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1483 #else
1484     xen_map_cache_init(NULL, state);
1485 #endif
1486     xen_ram_init(pcms, ram_size, ram_memory);
1487 
1488     qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1489 
1490     state->memory_listener = xen_memory_listener;
1491     memory_listener_register(&state->memory_listener, &address_space_memory);
1492     state->log_for_dirtybit = NULL;
1493 
1494     state->io_listener = xen_io_listener;
1495     memory_listener_register(&state->io_listener, &address_space_io);
1496 
1497     state->device_listener = xen_device_listener;
1498     QLIST_INIT(&state->dev_list);
1499     device_listener_register(&state->device_listener);
1500 
1501     xen_bus_init();
1502 
1503     /* Initialize backend core & drivers */
1504     if (xen_be_init() != 0) {
1505         error_report("xen backend core setup failed");
1506         goto err;
1507     }
1508     xen_be_register_common();
1509 
1510     QLIST_INIT(&xen_physmap);
1511     xen_read_physmap(state);
1512 
1513     /* Disable ACPI build because Xen handles it */
1514     pcms->acpi_build_enabled = false;
1515 
1516     return;
1517 
1518 err:
1519     error_report("xen hardware virtual machine initialisation failed");
1520     exit(1);
1521 }
1522 
1523 void destroy_hvm_domain(bool reboot)
1524 {
1525     xc_interface *xc_handle;
1526     int sts;
1527     int rc;
1528 
1529     unsigned int reason = reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff;
1530 
1531     if (xen_dmod) {
1532         rc = xendevicemodel_shutdown(xen_dmod, xen_domid, reason);
1533         if (!rc) {
1534             return;
1535         }
1536         if (errno != ENOTTY /* old Xen */) {
1537             perror("xendevicemodel_shutdown failed");
1538         }
1539         /* well, try the old thing then */
1540     }
1541 
1542     xc_handle = xc_interface_open(0, 0, 0);
1543     if (xc_handle == NULL) {
1544         fprintf(stderr, "Cannot acquire xenctrl handle\n");
1545     } else {
1546         sts = xc_domain_shutdown(xc_handle, xen_domid, reason);
1547         if (sts != 0) {
1548             fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1549                     "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1550                     sts, strerror(errno));
1551         } else {
1552             fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1553                     reboot ? "reboot" : "poweroff");
1554         }
1555         xc_interface_close(xc_handle);
1556     }
1557 }
1558 
1559 void xen_register_framebuffer(MemoryRegion *mr)
1560 {
1561     framebuffer = mr;
1562 }
1563 
1564 void xen_shutdown_fatal_error(const char *fmt, ...)
1565 {
1566     va_list ap;
1567 
1568     va_start(ap, fmt);
1569     vfprintf(stderr, fmt, ap);
1570     va_end(ap);
1571     fprintf(stderr, "Will destroy the domain.\n");
1572     /* destroy the domain */
1573     qemu_system_shutdown_request(SHUTDOWN_CAUSE_HOST_ERROR);
1574 }
1575 
1576 void xen_hvm_modified_memory(ram_addr_t start, ram_addr_t length)
1577 {
1578     if (unlikely(xen_in_migration)) {
1579         int rc;
1580         ram_addr_t start_pfn, nb_pages;
1581 
1582         start = xen_phys_offset_to_gaddr(start, length);
1583 
1584         if (length == 0) {
1585             length = TARGET_PAGE_SIZE;
1586         }
1587         start_pfn = start >> TARGET_PAGE_BITS;
1588         nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1589             - start_pfn;
1590         rc = xen_modified_memory(xen_domid, start_pfn, nb_pages);
1591         if (rc) {
1592             fprintf(stderr,
1593                     "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1594                     __func__, start, nb_pages, errno, strerror(errno));
1595         }
1596     }
1597 }
1598 
1599 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1600 {
1601     if (enable) {
1602         memory_global_dirty_log_start();
1603     } else {
1604         memory_global_dirty_log_stop();
1605     }
1606 }
1607