xref: /openbmc/qemu/hw/i386/xen/xen-hvm.c (revision 697b66d006676620a56fb5b79720ce457158204b)
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/i386/pc.h"
16 #include "hw/i386/apic-msidef.h"
17 #include "hw/xen/xen_common.h"
18 #include "hw/xen/xen_backend.h"
19 #include "qmp-commands.h"
20 
21 #include "qemu/error-report.h"
22 #include "qemu/range.h"
23 #include "sysemu/xen-mapcache.h"
24 #include "trace.h"
25 #include "exec/address-spaces.h"
26 
27 #include <xen/hvm/ioreq.h>
28 #include <xen/hvm/params.h>
29 #include <xen/hvm/e820.h>
30 
31 //#define DEBUG_XEN_HVM
32 
33 #ifdef DEBUG_XEN_HVM
34 #define DPRINTF(fmt, ...) \
35     do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
36 #else
37 #define DPRINTF(fmt, ...) \
38     do { } while (0)
39 #endif
40 
41 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
42 static MemoryRegion *framebuffer;
43 static bool xen_in_migration;
44 
45 /* Compatibility with older version */
46 
47 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
48  * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
49  * needs to be included before this block and hw/xen/xen_common.h needs to
50  * be included before xen/hvm/ioreq.h
51  */
52 #ifndef IOREQ_TYPE_VMWARE_PORT
53 #define IOREQ_TYPE_VMWARE_PORT  3
54 struct vmware_regs {
55     uint32_t esi;
56     uint32_t edi;
57     uint32_t ebx;
58     uint32_t ecx;
59     uint32_t edx;
60 };
61 typedef struct vmware_regs vmware_regs_t;
62 
63 struct shared_vmport_iopage {
64     struct vmware_regs vcpu_vmport_regs[1];
65 };
66 typedef struct shared_vmport_iopage shared_vmport_iopage_t;
67 #endif
68 
69 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
70 {
71     return shared_page->vcpu_ioreq[i].vp_eport;
72 }
73 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
74 {
75     return &shared_page->vcpu_ioreq[vcpu];
76 }
77 
78 #define BUFFER_IO_MAX_DELAY  100
79 
80 typedef struct XenPhysmap {
81     hwaddr start_addr;
82     ram_addr_t size;
83     const char *name;
84     hwaddr phys_offset;
85 
86     QLIST_ENTRY(XenPhysmap) list;
87 } XenPhysmap;
88 
89 typedef struct XenIOState {
90     ioservid_t ioservid;
91     shared_iopage_t *shared_page;
92     shared_vmport_iopage_t *shared_vmport_page;
93     buffered_iopage_t *buffered_io_page;
94     QEMUTimer *buffered_io_timer;
95     CPUState **cpu_by_vcpu_id;
96     /* the evtchn port for polling the notification, */
97     evtchn_port_t *ioreq_local_port;
98     /* evtchn local port for buffered io */
99     evtchn_port_t bufioreq_local_port;
100     /* the evtchn fd for polling */
101     xenevtchn_handle *xce_handle;
102     /* which vcpu we are serving */
103     int send_vcpu;
104 
105     struct xs_handle *xenstore;
106     MemoryListener memory_listener;
107     MemoryListener io_listener;
108     DeviceListener device_listener;
109     QLIST_HEAD(, XenPhysmap) physmap;
110     hwaddr free_phys_offset;
111     const XenPhysmap *log_for_dirtybit;
112 
113     Notifier exit;
114     Notifier suspend;
115     Notifier wakeup;
116 } XenIOState;
117 
118 /* Xen specific function for piix pci */
119 
120 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
121 {
122     return irq_num + ((pci_dev->devfn >> 3) << 2);
123 }
124 
125 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
126 {
127     xen_set_pci_intx_level(xen_domid, 0, 0, irq_num >> 2,
128                            irq_num & 3, level);
129 }
130 
131 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
132 {
133     int i;
134 
135     /* Scan for updates to PCI link routes (0x60-0x63). */
136     for (i = 0; i < len; i++) {
137         uint8_t v = (val >> (8 * i)) & 0xff;
138         if (v & 0x80) {
139             v = 0;
140         }
141         v &= 0xf;
142         if (((address + i) >= 0x60) && ((address + i) <= 0x63)) {
143             xen_set_pci_link_route(xen_domid, address + i - 0x60, v);
144         }
145     }
146 }
147 
148 int xen_is_pirq_msi(uint32_t msi_data)
149 {
150     /* If vector is 0, the msi is remapped into a pirq, passed as
151      * dest_id.
152      */
153     return ((msi_data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT) == 0;
154 }
155 
156 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
157 {
158     xen_inject_msi(xen_domid, addr, data);
159 }
160 
161 static void xen_suspend_notifier(Notifier *notifier, void *data)
162 {
163     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
164 }
165 
166 /* Xen Interrupt Controller */
167 
168 static void xen_set_irq(void *opaque, int irq, int level)
169 {
170     xen_set_isa_irq_level(xen_domid, irq, level);
171 }
172 
173 qemu_irq *xen_interrupt_controller_init(void)
174 {
175     return qemu_allocate_irqs(xen_set_irq, NULL, 16);
176 }
177 
178 /* Memory Ops */
179 
180 static void xen_ram_init(PCMachineState *pcms,
181                          ram_addr_t ram_size, MemoryRegion **ram_memory_p)
182 {
183     MemoryRegion *sysmem = get_system_memory();
184     ram_addr_t block_len;
185     uint64_t user_lowmem = object_property_get_uint(qdev_get_machine(),
186                                                     PC_MACHINE_MAX_RAM_BELOW_4G,
187                                                     &error_abort);
188 
189     /* Handle the machine opt max-ram-below-4g.  It is basically doing
190      * min(xen limit, user limit).
191      */
192     if (!user_lowmem) {
193         user_lowmem = HVM_BELOW_4G_RAM_END; /* default */
194     }
195     if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
196         user_lowmem = HVM_BELOW_4G_RAM_END;
197     }
198 
199     if (ram_size >= user_lowmem) {
200         pcms->above_4g_mem_size = ram_size - user_lowmem;
201         pcms->below_4g_mem_size = user_lowmem;
202     } else {
203         pcms->above_4g_mem_size = 0;
204         pcms->below_4g_mem_size = ram_size;
205     }
206     if (!pcms->above_4g_mem_size) {
207         block_len = ram_size;
208     } else {
209         /*
210          * Xen does not allocate the memory continuously, it keeps a
211          * hole of the size computed above or passed in.
212          */
213         block_len = (1ULL << 32) + pcms->above_4g_mem_size;
214     }
215     memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
216                            &error_fatal);
217     *ram_memory_p = &ram_memory;
218 
219     memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
220                              &ram_memory, 0, 0xa0000);
221     memory_region_add_subregion(sysmem, 0, &ram_640k);
222     /* Skip of the VGA IO memory space, it will be registered later by the VGA
223      * emulated device.
224      *
225      * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
226      * the Options ROM, so it is registered here as RAM.
227      */
228     memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
229                              &ram_memory, 0xc0000,
230                              pcms->below_4g_mem_size - 0xc0000);
231     memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
232     if (pcms->above_4g_mem_size > 0) {
233         memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
234                                  &ram_memory, 0x100000000ULL,
235                                  pcms->above_4g_mem_size);
236         memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
237     }
238 }
239 
240 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr,
241                    Error **errp)
242 {
243     unsigned long nr_pfn;
244     xen_pfn_t *pfn_list;
245     int i;
246 
247     if (runstate_check(RUN_STATE_INMIGRATE)) {
248         /* RAM already populated in Xen */
249         fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
250                 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
251                 __func__, size, ram_addr);
252         return;
253     }
254 
255     if (mr == &ram_memory) {
256         return;
257     }
258 
259     trace_xen_ram_alloc(ram_addr, size);
260 
261     nr_pfn = size >> TARGET_PAGE_BITS;
262     pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
263 
264     for (i = 0; i < nr_pfn; i++) {
265         pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
266     }
267 
268     if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
269         error_setg(errp, "xen: failed to populate ram at " RAM_ADDR_FMT,
270                    ram_addr);
271     }
272 
273     g_free(pfn_list);
274 }
275 
276 static XenPhysmap *get_physmapping(XenIOState *state,
277                                    hwaddr start_addr, ram_addr_t size)
278 {
279     XenPhysmap *physmap = NULL;
280 
281     start_addr &= TARGET_PAGE_MASK;
282 
283     QLIST_FOREACH(physmap, &state->physmap, list) {
284         if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
285             return physmap;
286         }
287     }
288     return NULL;
289 }
290 
291 static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
292                                                    ram_addr_t size, void *opaque)
293 {
294     hwaddr addr = start_addr & TARGET_PAGE_MASK;
295     XenIOState *xen_io_state = opaque;
296     XenPhysmap *physmap = NULL;
297 
298     QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
299         if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
300             return physmap->start_addr;
301         }
302     }
303 
304     return start_addr;
305 }
306 
307 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
308 {
309     char path[80], value[17];
310 
311     snprintf(path, sizeof(path),
312             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
313             xen_domid, (uint64_t)physmap->phys_offset);
314     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->start_addr);
315     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
316         return -1;
317     }
318     snprintf(path, sizeof(path),
319             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
320             xen_domid, (uint64_t)physmap->phys_offset);
321     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->size);
322     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
323         return -1;
324     }
325     if (physmap->name) {
326         snprintf(path, sizeof(path),
327                 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
328                 xen_domid, (uint64_t)physmap->phys_offset);
329         if (!xs_write(state->xenstore, 0, path,
330                       physmap->name, strlen(physmap->name))) {
331             return -1;
332         }
333     }
334     return 0;
335 }
336 
337 static int xen_add_to_physmap(XenIOState *state,
338                               hwaddr start_addr,
339                               ram_addr_t size,
340                               MemoryRegion *mr,
341                               hwaddr offset_within_region)
342 {
343     unsigned long i = 0;
344     int rc = 0;
345     XenPhysmap *physmap = NULL;
346     hwaddr pfn, start_gpfn;
347     hwaddr phys_offset = memory_region_get_ram_addr(mr);
348     const char *mr_name;
349 
350     if (get_physmapping(state, start_addr, size)) {
351         return 0;
352     }
353     if (size <= 0) {
354         return -1;
355     }
356 
357     /* Xen can only handle a single dirty log region for now and we want
358      * the linear framebuffer to be that region.
359      * Avoid tracking any regions that is not videoram and avoid tracking
360      * the legacy vga region. */
361     if (mr == framebuffer && start_addr > 0xbffff) {
362         goto go_physmap;
363     }
364     return -1;
365 
366 go_physmap:
367     DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
368             start_addr, start_addr + size);
369 
370     pfn = phys_offset >> TARGET_PAGE_BITS;
371     start_gpfn = start_addr >> TARGET_PAGE_BITS;
372     for (i = 0; i < size >> TARGET_PAGE_BITS; i++) {
373         unsigned long idx = pfn + i;
374         xen_pfn_t gpfn = start_gpfn + i;
375 
376         rc = xen_xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
377         if (rc) {
378             DPRINTF("add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
379                     PRI_xen_pfn" failed: %d (errno: %d)\n", idx, gpfn, rc, errno);
380             return -rc;
381         }
382     }
383 
384     mr_name = memory_region_name(mr);
385 
386     physmap = g_malloc(sizeof (XenPhysmap));
387 
388     physmap->start_addr = start_addr;
389     physmap->size = size;
390     physmap->name = mr_name;
391     physmap->phys_offset = phys_offset;
392 
393     QLIST_INSERT_HEAD(&state->physmap, physmap, list);
394 
395     xc_domain_pin_memory_cacheattr(xen_xc, xen_domid,
396                                    start_addr >> TARGET_PAGE_BITS,
397                                    (start_addr + size - 1) >> TARGET_PAGE_BITS,
398                                    XEN_DOMCTL_MEM_CACHEATTR_WB);
399     return xen_save_physmap(state, physmap);
400 }
401 
402 static int xen_remove_from_physmap(XenIOState *state,
403                                    hwaddr start_addr,
404                                    ram_addr_t size)
405 {
406     unsigned long i = 0;
407     int rc = 0;
408     XenPhysmap *physmap = NULL;
409     hwaddr phys_offset = 0;
410 
411     physmap = get_physmapping(state, start_addr, size);
412     if (physmap == NULL) {
413         return -1;
414     }
415 
416     phys_offset = physmap->phys_offset;
417     size = physmap->size;
418 
419     DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
420             "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
421 
422     size >>= TARGET_PAGE_BITS;
423     start_addr >>= TARGET_PAGE_BITS;
424     phys_offset >>= TARGET_PAGE_BITS;
425     for (i = 0; i < size; i++) {
426         xen_pfn_t idx = start_addr + i;
427         xen_pfn_t gpfn = phys_offset + i;
428 
429         rc = xen_xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
430         if (rc) {
431             fprintf(stderr, "add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
432                     PRI_xen_pfn" failed: %d (errno: %d)\n", idx, gpfn, rc, errno);
433             return -rc;
434         }
435     }
436 
437     QLIST_REMOVE(physmap, list);
438     if (state->log_for_dirtybit == physmap) {
439         state->log_for_dirtybit = NULL;
440     }
441     g_free(physmap);
442 
443     return 0;
444 }
445 
446 static void xen_set_memory(struct MemoryListener *listener,
447                            MemoryRegionSection *section,
448                            bool add)
449 {
450     XenIOState *state = container_of(listener, XenIOState, memory_listener);
451     hwaddr start_addr = section->offset_within_address_space;
452     ram_addr_t size = int128_get64(section->size);
453     bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA);
454     hvmmem_type_t mem_type;
455 
456     if (section->mr == &ram_memory) {
457         return;
458     } else {
459         if (add) {
460             xen_map_memory_section(xen_domid, state->ioservid,
461                                    section);
462         } else {
463             xen_unmap_memory_section(xen_domid, state->ioservid,
464                                      section);
465         }
466     }
467 
468     if (!memory_region_is_ram(section->mr)) {
469         return;
470     }
471 
472     if (log_dirty != add) {
473         return;
474     }
475 
476     trace_xen_client_set_memory(start_addr, size, log_dirty);
477 
478     start_addr &= TARGET_PAGE_MASK;
479     size = TARGET_PAGE_ALIGN(size);
480 
481     if (add) {
482         if (!memory_region_is_rom(section->mr)) {
483             xen_add_to_physmap(state, start_addr, size,
484                                section->mr, section->offset_within_region);
485         } else {
486             mem_type = HVMMEM_ram_ro;
487             if (xen_set_mem_type(xen_domid, mem_type,
488                                  start_addr >> TARGET_PAGE_BITS,
489                                  size >> TARGET_PAGE_BITS)) {
490                 DPRINTF("xen_set_mem_type error, addr: "TARGET_FMT_plx"\n",
491                         start_addr);
492             }
493         }
494     } else {
495         if (xen_remove_from_physmap(state, start_addr, size) < 0) {
496             DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
497         }
498     }
499 }
500 
501 static void xen_region_add(MemoryListener *listener,
502                            MemoryRegionSection *section)
503 {
504     memory_region_ref(section->mr);
505     xen_set_memory(listener, section, true);
506 }
507 
508 static void xen_region_del(MemoryListener *listener,
509                            MemoryRegionSection *section)
510 {
511     xen_set_memory(listener, section, false);
512     memory_region_unref(section->mr);
513 }
514 
515 static void xen_io_add(MemoryListener *listener,
516                        MemoryRegionSection *section)
517 {
518     XenIOState *state = container_of(listener, XenIOState, io_listener);
519     MemoryRegion *mr = section->mr;
520 
521     if (mr->ops == &unassigned_io_ops) {
522         return;
523     }
524 
525     memory_region_ref(mr);
526 
527     xen_map_io_section(xen_domid, state->ioservid, section);
528 }
529 
530 static void xen_io_del(MemoryListener *listener,
531                        MemoryRegionSection *section)
532 {
533     XenIOState *state = container_of(listener, XenIOState, io_listener);
534     MemoryRegion *mr = section->mr;
535 
536     if (mr->ops == &unassigned_io_ops) {
537         return;
538     }
539 
540     xen_unmap_io_section(xen_domid, state->ioservid, section);
541 
542     memory_region_unref(mr);
543 }
544 
545 static void xen_device_realize(DeviceListener *listener,
546 			       DeviceState *dev)
547 {
548     XenIOState *state = container_of(listener, XenIOState, device_listener);
549 
550     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
551         PCIDevice *pci_dev = PCI_DEVICE(dev);
552 
553         xen_map_pcidev(xen_domid, state->ioservid, pci_dev);
554     }
555 }
556 
557 static void xen_device_unrealize(DeviceListener *listener,
558 				 DeviceState *dev)
559 {
560     XenIOState *state = container_of(listener, XenIOState, device_listener);
561 
562     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
563         PCIDevice *pci_dev = PCI_DEVICE(dev);
564 
565         xen_unmap_pcidev(xen_domid, state->ioservid, pci_dev);
566     }
567 }
568 
569 static void xen_sync_dirty_bitmap(XenIOState *state,
570                                   hwaddr start_addr,
571                                   ram_addr_t size)
572 {
573     hwaddr npages = size >> TARGET_PAGE_BITS;
574     const int width = sizeof(unsigned long) * 8;
575     unsigned long bitmap[DIV_ROUND_UP(npages, width)];
576     int rc, i, j;
577     const XenPhysmap *physmap = NULL;
578 
579     physmap = get_physmapping(state, start_addr, size);
580     if (physmap == NULL) {
581         /* not handled */
582         return;
583     }
584 
585     if (state->log_for_dirtybit == NULL) {
586         state->log_for_dirtybit = physmap;
587     } else if (state->log_for_dirtybit != physmap) {
588         /* Only one range for dirty bitmap can be tracked. */
589         return;
590     }
591 
592     rc = xen_track_dirty_vram(xen_domid, start_addr >> TARGET_PAGE_BITS,
593                               npages, bitmap);
594     if (rc < 0) {
595 #ifndef ENODATA
596 #define ENODATA  ENOENT
597 #endif
598         if (errno == ENODATA) {
599             memory_region_set_dirty(framebuffer, 0, size);
600             DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
601                     ", 0x" TARGET_FMT_plx "): %s\n",
602                     start_addr, start_addr + size, strerror(errno));
603         }
604         return;
605     }
606 
607     for (i = 0; i < ARRAY_SIZE(bitmap); i++) {
608         unsigned long map = bitmap[i];
609         while (map != 0) {
610             j = ctzl(map);
611             map &= ~(1ul << j);
612             memory_region_set_dirty(framebuffer,
613                                     (i * width + j) * TARGET_PAGE_SIZE,
614                                     TARGET_PAGE_SIZE);
615         };
616     }
617 }
618 
619 static void xen_log_start(MemoryListener *listener,
620                           MemoryRegionSection *section,
621                           int old, int new)
622 {
623     XenIOState *state = container_of(listener, XenIOState, memory_listener);
624 
625     if (new & ~old & (1 << DIRTY_MEMORY_VGA)) {
626         xen_sync_dirty_bitmap(state, section->offset_within_address_space,
627                               int128_get64(section->size));
628     }
629 }
630 
631 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section,
632                          int old, int new)
633 {
634     XenIOState *state = container_of(listener, XenIOState, memory_listener);
635 
636     if (old & ~new & (1 << DIRTY_MEMORY_VGA)) {
637         state->log_for_dirtybit = NULL;
638         /* Disable dirty bit tracking */
639         xen_track_dirty_vram(xen_domid, 0, 0, NULL);
640     }
641 }
642 
643 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
644 {
645     XenIOState *state = container_of(listener, XenIOState, memory_listener);
646 
647     xen_sync_dirty_bitmap(state, section->offset_within_address_space,
648                           int128_get64(section->size));
649 }
650 
651 static void xen_log_global_start(MemoryListener *listener)
652 {
653     if (xen_enabled()) {
654         xen_in_migration = true;
655     }
656 }
657 
658 static void xen_log_global_stop(MemoryListener *listener)
659 {
660     xen_in_migration = false;
661 }
662 
663 static MemoryListener xen_memory_listener = {
664     .region_add = xen_region_add,
665     .region_del = xen_region_del,
666     .log_start = xen_log_start,
667     .log_stop = xen_log_stop,
668     .log_sync = xen_log_sync,
669     .log_global_start = xen_log_global_start,
670     .log_global_stop = xen_log_global_stop,
671     .priority = 10,
672 };
673 
674 static MemoryListener xen_io_listener = {
675     .region_add = xen_io_add,
676     .region_del = xen_io_del,
677     .priority = 10,
678 };
679 
680 static DeviceListener xen_device_listener = {
681     .realize = xen_device_realize,
682     .unrealize = xen_device_unrealize,
683 };
684 
685 /* get the ioreq packets from share mem */
686 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
687 {
688     ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
689 
690     if (req->state != STATE_IOREQ_READY) {
691         DPRINTF("I/O request not ready: "
692                 "%x, ptr: %x, port: %"PRIx64", "
693                 "data: %"PRIx64", count: %u, size: %u\n",
694                 req->state, req->data_is_ptr, req->addr,
695                 req->data, req->count, req->size);
696         return NULL;
697     }
698 
699     xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
700 
701     req->state = STATE_IOREQ_INPROCESS;
702     return req;
703 }
704 
705 /* use poll to get the port notification */
706 /* ioreq_vec--out,the */
707 /* retval--the number of ioreq packet */
708 static ioreq_t *cpu_get_ioreq(XenIOState *state)
709 {
710     int i;
711     evtchn_port_t port;
712 
713     port = xenevtchn_pending(state->xce_handle);
714     if (port == state->bufioreq_local_port) {
715         timer_mod(state->buffered_io_timer,
716                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
717         return NULL;
718     }
719 
720     if (port != -1) {
721         for (i = 0; i < max_cpus; i++) {
722             if (state->ioreq_local_port[i] == port) {
723                 break;
724             }
725         }
726 
727         if (i == max_cpus) {
728             hw_error("Fatal error while trying to get io event!\n");
729         }
730 
731         /* unmask the wanted port again */
732         xenevtchn_unmask(state->xce_handle, port);
733 
734         /* get the io packet from shared memory */
735         state->send_vcpu = i;
736         return cpu_get_ioreq_from_shared_memory(state, i);
737     }
738 
739     /* read error or read nothing */
740     return NULL;
741 }
742 
743 static uint32_t do_inp(uint32_t addr, unsigned long size)
744 {
745     switch (size) {
746         case 1:
747             return cpu_inb(addr);
748         case 2:
749             return cpu_inw(addr);
750         case 4:
751             return cpu_inl(addr);
752         default:
753             hw_error("inp: bad size: %04x %lx", addr, size);
754     }
755 }
756 
757 static void do_outp(uint32_t addr,
758         unsigned long size, uint32_t val)
759 {
760     switch (size) {
761         case 1:
762             return cpu_outb(addr, val);
763         case 2:
764             return cpu_outw(addr, val);
765         case 4:
766             return cpu_outl(addr, val);
767         default:
768             hw_error("outp: bad size: %04x %lx", addr, size);
769     }
770 }
771 
772 /*
773  * Helper functions which read/write an object from/to physical guest
774  * memory, as part of the implementation of an ioreq.
775  *
776  * Equivalent to
777  *   cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
778  *                          val, req->size, 0/1)
779  * except without the integer overflow problems.
780  */
781 static void rw_phys_req_item(hwaddr addr,
782                              ioreq_t *req, uint32_t i, void *val, int rw)
783 {
784     /* Do everything unsigned so overflow just results in a truncated result
785      * and accesses to undesired parts of guest memory, which is up
786      * to the guest */
787     hwaddr offset = (hwaddr)req->size * i;
788     if (req->df) {
789         addr -= offset;
790     } else {
791         addr += offset;
792     }
793     cpu_physical_memory_rw(addr, val, req->size, rw);
794 }
795 
796 static inline void read_phys_req_item(hwaddr addr,
797                                       ioreq_t *req, uint32_t i, void *val)
798 {
799     rw_phys_req_item(addr, req, i, val, 0);
800 }
801 static inline void write_phys_req_item(hwaddr addr,
802                                        ioreq_t *req, uint32_t i, void *val)
803 {
804     rw_phys_req_item(addr, req, i, val, 1);
805 }
806 
807 
808 static void cpu_ioreq_pio(ioreq_t *req)
809 {
810     uint32_t i;
811 
812     trace_cpu_ioreq_pio(req, req->dir, req->df, req->data_is_ptr, req->addr,
813                          req->data, req->count, req->size);
814 
815     if (req->size > sizeof(uint32_t)) {
816         hw_error("PIO: bad size (%u)", req->size);
817     }
818 
819     if (req->dir == IOREQ_READ) {
820         if (!req->data_is_ptr) {
821             req->data = do_inp(req->addr, req->size);
822             trace_cpu_ioreq_pio_read_reg(req, req->data, req->addr,
823                                          req->size);
824         } else {
825             uint32_t tmp;
826 
827             for (i = 0; i < req->count; i++) {
828                 tmp = do_inp(req->addr, req->size);
829                 write_phys_req_item(req->data, req, i, &tmp);
830             }
831         }
832     } else if (req->dir == IOREQ_WRITE) {
833         if (!req->data_is_ptr) {
834             trace_cpu_ioreq_pio_write_reg(req, req->data, req->addr,
835                                           req->size);
836             do_outp(req->addr, req->size, req->data);
837         } else {
838             for (i = 0; i < req->count; i++) {
839                 uint32_t tmp = 0;
840 
841                 read_phys_req_item(req->data, req, i, &tmp);
842                 do_outp(req->addr, req->size, tmp);
843             }
844         }
845     }
846 }
847 
848 static void cpu_ioreq_move(ioreq_t *req)
849 {
850     uint32_t i;
851 
852     trace_cpu_ioreq_move(req, req->dir, req->df, req->data_is_ptr, req->addr,
853                          req->data, req->count, req->size);
854 
855     if (req->size > sizeof(req->data)) {
856         hw_error("MMIO: bad size (%u)", req->size);
857     }
858 
859     if (!req->data_is_ptr) {
860         if (req->dir == IOREQ_READ) {
861             for (i = 0; i < req->count; i++) {
862                 read_phys_req_item(req->addr, req, i, &req->data);
863             }
864         } else if (req->dir == IOREQ_WRITE) {
865             for (i = 0; i < req->count; i++) {
866                 write_phys_req_item(req->addr, req, i, &req->data);
867             }
868         }
869     } else {
870         uint64_t tmp;
871 
872         if (req->dir == IOREQ_READ) {
873             for (i = 0; i < req->count; i++) {
874                 read_phys_req_item(req->addr, req, i, &tmp);
875                 write_phys_req_item(req->data, req, i, &tmp);
876             }
877         } else if (req->dir == IOREQ_WRITE) {
878             for (i = 0; i < req->count; i++) {
879                 read_phys_req_item(req->data, req, i, &tmp);
880                 write_phys_req_item(req->addr, req, i, &tmp);
881             }
882         }
883     }
884 }
885 
886 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
887 {
888     X86CPU *cpu;
889     CPUX86State *env;
890 
891     cpu = X86_CPU(current_cpu);
892     env = &cpu->env;
893     env->regs[R_EAX] = req->data;
894     env->regs[R_EBX] = vmport_regs->ebx;
895     env->regs[R_ECX] = vmport_regs->ecx;
896     env->regs[R_EDX] = vmport_regs->edx;
897     env->regs[R_ESI] = vmport_regs->esi;
898     env->regs[R_EDI] = vmport_regs->edi;
899 }
900 
901 static void regs_from_cpu(vmware_regs_t *vmport_regs)
902 {
903     X86CPU *cpu = X86_CPU(current_cpu);
904     CPUX86State *env = &cpu->env;
905 
906     vmport_regs->ebx = env->regs[R_EBX];
907     vmport_regs->ecx = env->regs[R_ECX];
908     vmport_regs->edx = env->regs[R_EDX];
909     vmport_regs->esi = env->regs[R_ESI];
910     vmport_regs->edi = env->regs[R_EDI];
911 }
912 
913 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
914 {
915     vmware_regs_t *vmport_regs;
916 
917     assert(state->shared_vmport_page);
918     vmport_regs =
919         &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
920     QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
921 
922     current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
923     regs_to_cpu(vmport_regs, req);
924     cpu_ioreq_pio(req);
925     regs_from_cpu(vmport_regs);
926     current_cpu = NULL;
927 }
928 
929 static void handle_ioreq(XenIOState *state, ioreq_t *req)
930 {
931     trace_handle_ioreq(req, req->type, req->dir, req->df, req->data_is_ptr,
932                        req->addr, req->data, req->count, req->size);
933 
934     if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
935             (req->size < sizeof (target_ulong))) {
936         req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
937     }
938 
939     if (req->dir == IOREQ_WRITE)
940         trace_handle_ioreq_write(req, req->type, req->df, req->data_is_ptr,
941                                  req->addr, req->data, req->count, req->size);
942 
943     switch (req->type) {
944         case IOREQ_TYPE_PIO:
945             cpu_ioreq_pio(req);
946             break;
947         case IOREQ_TYPE_COPY:
948             cpu_ioreq_move(req);
949             break;
950         case IOREQ_TYPE_VMWARE_PORT:
951             handle_vmport_ioreq(state, req);
952             break;
953         case IOREQ_TYPE_TIMEOFFSET:
954             break;
955         case IOREQ_TYPE_INVALIDATE:
956             xen_invalidate_map_cache();
957             break;
958         case IOREQ_TYPE_PCI_CONFIG: {
959             uint32_t sbdf = req->addr >> 32;
960             uint32_t val;
961 
962             /* Fake a write to port 0xCF8 so that
963              * the config space access will target the
964              * correct device model.
965              */
966             val = (1u << 31) |
967                   ((req->addr & 0x0f00) << 16) |
968                   ((sbdf & 0xffff) << 8) |
969                   (req->addr & 0xfc);
970             do_outp(0xcf8, 4, val);
971 
972             /* Now issue the config space access via
973              * port 0xCFC
974              */
975             req->addr = 0xcfc | (req->addr & 0x03);
976             cpu_ioreq_pio(req);
977             break;
978         }
979         default:
980             hw_error("Invalid ioreq type 0x%x\n", req->type);
981     }
982     if (req->dir == IOREQ_READ) {
983         trace_handle_ioreq_read(req, req->type, req->df, req->data_is_ptr,
984                                 req->addr, req->data, req->count, req->size);
985     }
986 }
987 
988 static int handle_buffered_iopage(XenIOState *state)
989 {
990     buffered_iopage_t *buf_page = state->buffered_io_page;
991     buf_ioreq_t *buf_req = NULL;
992     ioreq_t req;
993     int qw;
994 
995     if (!buf_page) {
996         return 0;
997     }
998 
999     memset(&req, 0x00, sizeof(req));
1000     req.state = STATE_IOREQ_READY;
1001     req.count = 1;
1002     req.dir = IOREQ_WRITE;
1003 
1004     for (;;) {
1005         uint32_t rdptr = buf_page->read_pointer, wrptr;
1006 
1007         xen_rmb();
1008         wrptr = buf_page->write_pointer;
1009         xen_rmb();
1010         if (rdptr != buf_page->read_pointer) {
1011             continue;
1012         }
1013         if (rdptr == wrptr) {
1014             break;
1015         }
1016         buf_req = &buf_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
1017         req.size = 1U << buf_req->size;
1018         req.addr = buf_req->addr;
1019         req.data = buf_req->data;
1020         req.type = buf_req->type;
1021         xen_rmb();
1022         qw = (req.size == 8);
1023         if (qw) {
1024             if (rdptr + 1 == wrptr) {
1025                 hw_error("Incomplete quad word buffered ioreq");
1026             }
1027             buf_req = &buf_page->buf_ioreq[(rdptr + 1) %
1028                                            IOREQ_BUFFER_SLOT_NUM];
1029             req.data |= ((uint64_t)buf_req->data) << 32;
1030             xen_rmb();
1031         }
1032 
1033         handle_ioreq(state, &req);
1034 
1035         /* Only req.data may get updated by handle_ioreq(), albeit even that
1036          * should not happen as such data would never make it to the guest (we
1037          * can only usefully see writes here after all).
1038          */
1039         assert(req.state == STATE_IOREQ_READY);
1040         assert(req.count == 1);
1041         assert(req.dir == IOREQ_WRITE);
1042         assert(!req.data_is_ptr);
1043 
1044         atomic_add(&buf_page->read_pointer, qw + 1);
1045     }
1046 
1047     return req.count;
1048 }
1049 
1050 static void handle_buffered_io(void *opaque)
1051 {
1052     XenIOState *state = opaque;
1053 
1054     if (handle_buffered_iopage(state)) {
1055         timer_mod(state->buffered_io_timer,
1056                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
1057     } else {
1058         timer_del(state->buffered_io_timer);
1059         xenevtchn_unmask(state->xce_handle, state->bufioreq_local_port);
1060     }
1061 }
1062 
1063 static void cpu_handle_ioreq(void *opaque)
1064 {
1065     XenIOState *state = opaque;
1066     ioreq_t *req = cpu_get_ioreq(state);
1067 
1068     handle_buffered_iopage(state);
1069     if (req) {
1070         ioreq_t copy = *req;
1071 
1072         xen_rmb();
1073         handle_ioreq(state, &copy);
1074         req->data = copy.data;
1075 
1076         if (req->state != STATE_IOREQ_INPROCESS) {
1077             fprintf(stderr, "Badness in I/O request ... not in service?!: "
1078                     "%x, ptr: %x, port: %"PRIx64", "
1079                     "data: %"PRIx64", count: %u, size: %u, type: %u\n",
1080                     req->state, req->data_is_ptr, req->addr,
1081                     req->data, req->count, req->size, req->type);
1082             destroy_hvm_domain(false);
1083             return;
1084         }
1085 
1086         xen_wmb(); /* Update ioreq contents /then/ update state. */
1087 
1088         /*
1089          * We do this before we send the response so that the tools
1090          * have the opportunity to pick up on the reset before the
1091          * guest resumes and does a hlt with interrupts disabled which
1092          * causes Xen to powerdown the domain.
1093          */
1094         if (runstate_is_running()) {
1095             ShutdownCause request;
1096 
1097             if (qemu_shutdown_requested_get()) {
1098                 destroy_hvm_domain(false);
1099             }
1100             request = qemu_reset_requested_get();
1101             if (request) {
1102                 qemu_system_reset(request);
1103                 destroy_hvm_domain(true);
1104             }
1105         }
1106 
1107         req->state = STATE_IORESP_READY;
1108         xenevtchn_notify(state->xce_handle,
1109                          state->ioreq_local_port[state->send_vcpu]);
1110     }
1111 }
1112 
1113 static void xen_main_loop_prepare(XenIOState *state)
1114 {
1115     int evtchn_fd = -1;
1116 
1117     if (state->xce_handle != NULL) {
1118         evtchn_fd = xenevtchn_fd(state->xce_handle);
1119     }
1120 
1121     state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
1122                                                  state);
1123 
1124     if (evtchn_fd != -1) {
1125         CPUState *cpu_state;
1126 
1127         DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
1128         CPU_FOREACH(cpu_state) {
1129             DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1130                     __func__, cpu_state->cpu_index, cpu_state);
1131             state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
1132         }
1133         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
1134     }
1135 }
1136 
1137 
1138 static void xen_hvm_change_state_handler(void *opaque, int running,
1139                                          RunState rstate)
1140 {
1141     XenIOState *state = opaque;
1142 
1143     if (running) {
1144         xen_main_loop_prepare(state);
1145     }
1146 
1147     xen_set_ioreq_server_state(xen_domid,
1148                                state->ioservid,
1149                                (rstate == RUN_STATE_RUNNING));
1150 }
1151 
1152 static void xen_exit_notifier(Notifier *n, void *data)
1153 {
1154     XenIOState *state = container_of(n, XenIOState, exit);
1155 
1156     xenevtchn_close(state->xce_handle);
1157     xs_daemon_close(state->xenstore);
1158 }
1159 
1160 static void xen_read_physmap(XenIOState *state)
1161 {
1162     XenPhysmap *physmap = NULL;
1163     unsigned int len, num, i;
1164     char path[80], *value = NULL;
1165     char **entries = NULL;
1166 
1167     snprintf(path, sizeof(path),
1168             "/local/domain/0/device-model/%d/physmap", xen_domid);
1169     entries = xs_directory(state->xenstore, 0, path, &num);
1170     if (entries == NULL)
1171         return;
1172 
1173     for (i = 0; i < num; i++) {
1174         physmap = g_malloc(sizeof (XenPhysmap));
1175         physmap->phys_offset = strtoull(entries[i], NULL, 16);
1176         snprintf(path, sizeof(path),
1177                 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1178                 xen_domid, entries[i]);
1179         value = xs_read(state->xenstore, 0, path, &len);
1180         if (value == NULL) {
1181             g_free(physmap);
1182             continue;
1183         }
1184         physmap->start_addr = strtoull(value, NULL, 16);
1185         free(value);
1186 
1187         snprintf(path, sizeof(path),
1188                 "/local/domain/0/device-model/%d/physmap/%s/size",
1189                 xen_domid, entries[i]);
1190         value = xs_read(state->xenstore, 0, path, &len);
1191         if (value == NULL) {
1192             g_free(physmap);
1193             continue;
1194         }
1195         physmap->size = strtoull(value, NULL, 16);
1196         free(value);
1197 
1198         snprintf(path, sizeof(path),
1199                 "/local/domain/0/device-model/%d/physmap/%s/name",
1200                 xen_domid, entries[i]);
1201         physmap->name = xs_read(state->xenstore, 0, path, &len);
1202 
1203         QLIST_INSERT_HEAD(&state->physmap, physmap, list);
1204     }
1205     free(entries);
1206 }
1207 
1208 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1209 {
1210     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1211 }
1212 
1213 void xen_hvm_init(PCMachineState *pcms, MemoryRegion **ram_memory)
1214 {
1215     int i, rc;
1216     xen_pfn_t ioreq_pfn;
1217     xen_pfn_t bufioreq_pfn;
1218     evtchn_port_t bufioreq_evtchn;
1219     XenIOState *state;
1220 
1221     state = g_malloc0(sizeof (XenIOState));
1222 
1223     state->xce_handle = xenevtchn_open(NULL, 0);
1224     if (state->xce_handle == NULL) {
1225         perror("xen: event channel open");
1226         goto err;
1227     }
1228 
1229     state->xenstore = xs_daemon_open();
1230     if (state->xenstore == NULL) {
1231         perror("xen: xenstore open");
1232         goto err;
1233     }
1234 
1235     if (xen_domid_restrict) {
1236         rc = xen_restrict(xen_domid);
1237         if (rc < 0) {
1238             error_report("failed to restrict: error %d", errno);
1239             goto err;
1240         }
1241     }
1242 
1243     xen_create_ioreq_server(xen_domid, &state->ioservid);
1244 
1245     state->exit.notify = xen_exit_notifier;
1246     qemu_add_exit_notifier(&state->exit);
1247 
1248     state->suspend.notify = xen_suspend_notifier;
1249     qemu_register_suspend_notifier(&state->suspend);
1250 
1251     state->wakeup.notify = xen_wakeup_notifier;
1252     qemu_register_wakeup_notifier(&state->wakeup);
1253 
1254     rc = xen_get_ioreq_server_info(xen_domid, state->ioservid,
1255                                    &ioreq_pfn, &bufioreq_pfn,
1256                                    &bufioreq_evtchn);
1257     if (rc < 0) {
1258         error_report("failed to get ioreq server info: error %d handle=%p",
1259                      errno, xen_xc);
1260         goto err;
1261     }
1262 
1263     DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1264     DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
1265     DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
1266 
1267     state->shared_page = xenforeignmemory_map(xen_fmem, xen_domid,
1268                                               PROT_READ|PROT_WRITE,
1269                                               1, &ioreq_pfn, NULL);
1270     if (state->shared_page == NULL) {
1271         error_report("map shared IO page returned error %d handle=%p",
1272                      errno, xen_xc);
1273         goto err;
1274     }
1275 
1276     rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1277     if (!rc) {
1278         DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1279         state->shared_vmport_page =
1280             xenforeignmemory_map(xen_fmem, xen_domid, PROT_READ|PROT_WRITE,
1281                                  1, &ioreq_pfn, NULL);
1282         if (state->shared_vmport_page == NULL) {
1283             error_report("map shared vmport IO page returned error %d handle=%p",
1284                          errno, xen_xc);
1285             goto err;
1286         }
1287     } else if (rc != -ENOSYS) {
1288         error_report("get vmport regs pfn returned error %d, rc=%d",
1289                      errno, rc);
1290         goto err;
1291     }
1292 
1293     state->buffered_io_page = xenforeignmemory_map(xen_fmem, xen_domid,
1294                                                    PROT_READ|PROT_WRITE,
1295                                                    1, &bufioreq_pfn, NULL);
1296     if (state->buffered_io_page == NULL) {
1297         error_report("map buffered IO page returned error %d", errno);
1298         goto err;
1299     }
1300 
1301     /* Note: cpus is empty at this point in init */
1302     state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1303 
1304     rc = xen_set_ioreq_server_state(xen_domid, state->ioservid, true);
1305     if (rc < 0) {
1306         error_report("failed to enable ioreq server info: error %d handle=%p",
1307                      errno, xen_xc);
1308         goto err;
1309     }
1310 
1311     state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1312 
1313     /* FIXME: how about if we overflow the page here? */
1314     for (i = 0; i < max_cpus; i++) {
1315         rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1316                                         xen_vcpu_eport(state->shared_page, i));
1317         if (rc == -1) {
1318             error_report("shared evtchn %d bind error %d", i, errno);
1319             goto err;
1320         }
1321         state->ioreq_local_port[i] = rc;
1322     }
1323 
1324     rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1325                                     bufioreq_evtchn);
1326     if (rc == -1) {
1327         error_report("buffered evtchn bind error %d", errno);
1328         goto err;
1329     }
1330     state->bufioreq_local_port = rc;
1331 
1332     /* Init RAM management */
1333     xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1334     xen_ram_init(pcms, ram_size, ram_memory);
1335 
1336     qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1337 
1338     state->memory_listener = xen_memory_listener;
1339     QLIST_INIT(&state->physmap);
1340     memory_listener_register(&state->memory_listener, &address_space_memory);
1341     state->log_for_dirtybit = NULL;
1342 
1343     state->io_listener = xen_io_listener;
1344     memory_listener_register(&state->io_listener, &address_space_io);
1345 
1346     state->device_listener = xen_device_listener;
1347     device_listener_register(&state->device_listener);
1348 
1349     /* Initialize backend core & drivers */
1350     if (xen_be_init() != 0) {
1351         error_report("xen backend core setup failed");
1352         goto err;
1353     }
1354     xen_be_register_common();
1355     xen_read_physmap(state);
1356 
1357     /* Disable ACPI build because Xen handles it */
1358     pcms->acpi_build_enabled = false;
1359 
1360     return;
1361 
1362 err:
1363     error_report("xen hardware virtual machine initialisation failed");
1364     exit(1);
1365 }
1366 
1367 void destroy_hvm_domain(bool reboot)
1368 {
1369     xc_interface *xc_handle;
1370     int sts;
1371 
1372     xc_handle = xc_interface_open(0, 0, 0);
1373     if (xc_handle == NULL) {
1374         fprintf(stderr, "Cannot acquire xenctrl handle\n");
1375     } else {
1376         sts = xc_domain_shutdown(xc_handle, xen_domid,
1377                                  reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
1378         if (sts != 0) {
1379             fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1380                     "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1381                     sts, strerror(errno));
1382         } else {
1383             fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1384                     reboot ? "reboot" : "poweroff");
1385         }
1386         xc_interface_close(xc_handle);
1387     }
1388 }
1389 
1390 void xen_register_framebuffer(MemoryRegion *mr)
1391 {
1392     framebuffer = mr;
1393 }
1394 
1395 void xen_shutdown_fatal_error(const char *fmt, ...)
1396 {
1397     va_list ap;
1398 
1399     va_start(ap, fmt);
1400     vfprintf(stderr, fmt, ap);
1401     va_end(ap);
1402     fprintf(stderr, "Will destroy the domain.\n");
1403     /* destroy the domain */
1404     qemu_system_shutdown_request(SHUTDOWN_CAUSE_HOST_ERROR);
1405 }
1406 
1407 void xen_hvm_modified_memory(ram_addr_t start, ram_addr_t length)
1408 {
1409     if (unlikely(xen_in_migration)) {
1410         int rc;
1411         ram_addr_t start_pfn, nb_pages;
1412 
1413         if (length == 0) {
1414             length = TARGET_PAGE_SIZE;
1415         }
1416         start_pfn = start >> TARGET_PAGE_BITS;
1417         nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1418             - start_pfn;
1419         rc = xen_modified_memory(xen_domid, start_pfn, nb_pages);
1420         if (rc) {
1421             fprintf(stderr,
1422                     "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1423                     __func__, start, nb_pages, rc, strerror(-rc));
1424         }
1425     }
1426 }
1427 
1428 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1429 {
1430     if (enable) {
1431         memory_global_dirty_log_start();
1432     } else {
1433         memory_global_dirty_log_stop();
1434     }
1435 }
1436