xref: /openbmc/qemu/hw/i386/xen/xen-hvm.c (revision 1da79ecc)
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 #include "qemu/units.h"
13 
14 #include "cpu.h"
15 #include "hw/pci/pci.h"
16 #include "hw/pci/pci_host.h"
17 #include "hw/i386/pc.h"
18 #include "hw/southbridge/piix.h"
19 #include "hw/irq.h"
20 #include "hw/hw.h"
21 #include "hw/i386/apic-msidef.h"
22 #include "hw/xen/xen_common.h"
23 #include "hw/xen/xen-legacy-backend.h"
24 #include "hw/xen/xen-bus.h"
25 #include "hw/xen/xen-x86.h"
26 #include "qapi/error.h"
27 #include "qapi/qapi-commands-migration.h"
28 #include "qemu/error-report.h"
29 #include "qemu/main-loop.h"
30 #include "qemu/range.h"
31 #include "sysemu/runstate.h"
32 #include "sysemu/sysemu.h"
33 #include "sysemu/xen.h"
34 #include "sysemu/xen-mapcache.h"
35 #include "trace.h"
36 #include "exec/address-spaces.h"
37 
38 #include <xen/hvm/ioreq.h>
39 #include <xen/hvm/e820.h>
40 
41 //#define DEBUG_XEN_HVM
42 
43 #ifdef DEBUG_XEN_HVM
44 #define DPRINTF(fmt, ...) \
45     do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
46 #else
47 #define DPRINTF(fmt, ...) \
48     do { } while (0)
49 #endif
50 
51 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
52 static MemoryRegion *framebuffer;
53 static bool xen_in_migration;
54 
55 /* Compatibility with older version */
56 
57 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
58  * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
59  * needs to be included before this block and hw/xen/xen_common.h needs to
60  * be included before xen/hvm/ioreq.h
61  */
62 #ifndef IOREQ_TYPE_VMWARE_PORT
63 #define IOREQ_TYPE_VMWARE_PORT  3
64 struct vmware_regs {
65     uint32_t esi;
66     uint32_t edi;
67     uint32_t ebx;
68     uint32_t ecx;
69     uint32_t edx;
70 };
71 typedef struct vmware_regs vmware_regs_t;
72 
73 struct shared_vmport_iopage {
74     struct vmware_regs vcpu_vmport_regs[1];
75 };
76 typedef struct shared_vmport_iopage shared_vmport_iopage_t;
77 #endif
78 
79 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
80 {
81     return shared_page->vcpu_ioreq[i].vp_eport;
82 }
83 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
84 {
85     return &shared_page->vcpu_ioreq[vcpu];
86 }
87 
88 #define BUFFER_IO_MAX_DELAY  100
89 
90 typedef struct XenPhysmap {
91     hwaddr start_addr;
92     ram_addr_t size;
93     const char *name;
94     hwaddr phys_offset;
95 
96     QLIST_ENTRY(XenPhysmap) list;
97 } XenPhysmap;
98 
99 static QLIST_HEAD(, XenPhysmap) xen_physmap;
100 
101 typedef struct XenPciDevice {
102     PCIDevice *pci_dev;
103     uint32_t sbdf;
104     QLIST_ENTRY(XenPciDevice) entry;
105 } XenPciDevice;
106 
107 typedef struct XenIOState {
108     ioservid_t ioservid;
109     shared_iopage_t *shared_page;
110     shared_vmport_iopage_t *shared_vmport_page;
111     buffered_iopage_t *buffered_io_page;
112     QEMUTimer *buffered_io_timer;
113     CPUState **cpu_by_vcpu_id;
114     /* the evtchn port for polling the notification, */
115     evtchn_port_t *ioreq_local_port;
116     /* evtchn remote and local ports for buffered io */
117     evtchn_port_t bufioreq_remote_port;
118     evtchn_port_t bufioreq_local_port;
119     /* the evtchn fd for polling */
120     xenevtchn_handle *xce_handle;
121     /* which vcpu we are serving */
122     int send_vcpu;
123 
124     struct xs_handle *xenstore;
125     MemoryListener memory_listener;
126     MemoryListener io_listener;
127     QLIST_HEAD(, XenPciDevice) dev_list;
128     DeviceListener device_listener;
129     hwaddr free_phys_offset;
130     const XenPhysmap *log_for_dirtybit;
131     /* Buffer used by xen_sync_dirty_bitmap */
132     unsigned long *dirty_bitmap;
133 
134     Notifier exit;
135     Notifier suspend;
136     Notifier wakeup;
137 } XenIOState;
138 
139 /* Xen specific function for piix pci */
140 
141 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
142 {
143     return irq_num + (PCI_SLOT(pci_dev->devfn) << 2);
144 }
145 
146 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
147 {
148     xen_set_pci_intx_level(xen_domid, 0, 0, irq_num >> 2,
149                            irq_num & 3, level);
150 }
151 
152 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
153 {
154     int i;
155 
156     /* Scan for updates to PCI link routes (0x60-0x63). */
157     for (i = 0; i < len; i++) {
158         uint8_t v = (val >> (8 * i)) & 0xff;
159         if (v & 0x80) {
160             v = 0;
161         }
162         v &= 0xf;
163         if (((address + i) >= PIIX_PIRQCA) && ((address + i) <= PIIX_PIRQCD)) {
164             xen_set_pci_link_route(xen_domid, address + i - PIIX_PIRQCA, v);
165         }
166     }
167 }
168 
169 int xen_is_pirq_msi(uint32_t msi_data)
170 {
171     /* If vector is 0, the msi is remapped into a pirq, passed as
172      * dest_id.
173      */
174     return ((msi_data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT) == 0;
175 }
176 
177 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
178 {
179     xen_inject_msi(xen_domid, addr, data);
180 }
181 
182 static void xen_suspend_notifier(Notifier *notifier, void *data)
183 {
184     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
185 }
186 
187 /* Xen Interrupt Controller */
188 
189 static void xen_set_irq(void *opaque, int irq, int level)
190 {
191     xen_set_isa_irq_level(xen_domid, irq, level);
192 }
193 
194 qemu_irq *xen_interrupt_controller_init(void)
195 {
196     return qemu_allocate_irqs(xen_set_irq, NULL, 16);
197 }
198 
199 /* Memory Ops */
200 
201 static void xen_ram_init(PCMachineState *pcms,
202                          ram_addr_t ram_size, MemoryRegion **ram_memory_p)
203 {
204     X86MachineState *x86ms = X86_MACHINE(pcms);
205     MemoryRegion *sysmem = get_system_memory();
206     ram_addr_t block_len;
207     uint64_t user_lowmem =
208         object_property_get_uint(qdev_get_machine(),
209                                  PC_MACHINE_MAX_RAM_BELOW_4G,
210                                  &error_abort);
211 
212     /* Handle the machine opt max-ram-below-4g.  It is basically doing
213      * min(xen limit, user limit).
214      */
215     if (!user_lowmem) {
216         user_lowmem = HVM_BELOW_4G_RAM_END; /* default */
217     }
218     if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
219         user_lowmem = HVM_BELOW_4G_RAM_END;
220     }
221 
222     if (ram_size >= user_lowmem) {
223         x86ms->above_4g_mem_size = ram_size - user_lowmem;
224         x86ms->below_4g_mem_size = user_lowmem;
225     } else {
226         x86ms->above_4g_mem_size = 0;
227         x86ms->below_4g_mem_size = ram_size;
228     }
229     if (!x86ms->above_4g_mem_size) {
230         block_len = ram_size;
231     } else {
232         /*
233          * Xen does not allocate the memory continuously, it keeps a
234          * hole of the size computed above or passed in.
235          */
236         block_len = (4 * GiB) + x86ms->above_4g_mem_size;
237     }
238     memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
239                            &error_fatal);
240     *ram_memory_p = &ram_memory;
241 
242     memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
243                              &ram_memory, 0, 0xa0000);
244     memory_region_add_subregion(sysmem, 0, &ram_640k);
245     /* Skip of the VGA IO memory space, it will be registered later by the VGA
246      * emulated device.
247      *
248      * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
249      * the Options ROM, so it is registered here as RAM.
250      */
251     memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
252                              &ram_memory, 0xc0000,
253                              x86ms->below_4g_mem_size - 0xc0000);
254     memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
255     if (x86ms->above_4g_mem_size > 0) {
256         memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
257                                  &ram_memory, 0x100000000ULL,
258                                  x86ms->above_4g_mem_size);
259         memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
260     }
261 }
262 
263 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr,
264                    Error **errp)
265 {
266     unsigned long nr_pfn;
267     xen_pfn_t *pfn_list;
268     int i;
269 
270     if (runstate_check(RUN_STATE_INMIGRATE)) {
271         /* RAM already populated in Xen */
272         fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
273                 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
274                 __func__, size, ram_addr);
275         return;
276     }
277 
278     if (mr == &ram_memory) {
279         return;
280     }
281 
282     trace_xen_ram_alloc(ram_addr, size);
283 
284     nr_pfn = size >> TARGET_PAGE_BITS;
285     pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
286 
287     for (i = 0; i < nr_pfn; i++) {
288         pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
289     }
290 
291     if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
292         error_setg(errp, "xen: failed to populate ram at " RAM_ADDR_FMT,
293                    ram_addr);
294     }
295 
296     g_free(pfn_list);
297 }
298 
299 static XenPhysmap *get_physmapping(hwaddr start_addr, ram_addr_t size)
300 {
301     XenPhysmap *physmap = NULL;
302 
303     start_addr &= TARGET_PAGE_MASK;
304 
305     QLIST_FOREACH(physmap, &xen_physmap, list) {
306         if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
307             return physmap;
308         }
309     }
310     return NULL;
311 }
312 
313 static hwaddr xen_phys_offset_to_gaddr(hwaddr phys_offset, ram_addr_t size)
314 {
315     hwaddr addr = phys_offset & TARGET_PAGE_MASK;
316     XenPhysmap *physmap = NULL;
317 
318     QLIST_FOREACH(physmap, &xen_physmap, list) {
319         if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
320             return physmap->start_addr + (phys_offset - physmap->phys_offset);
321         }
322     }
323 
324     return phys_offset;
325 }
326 
327 #ifdef XEN_COMPAT_PHYSMAP
328 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
329 {
330     char path[80], value[17];
331 
332     snprintf(path, sizeof(path),
333             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
334             xen_domid, (uint64_t)physmap->phys_offset);
335     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->start_addr);
336     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
337         return -1;
338     }
339     snprintf(path, sizeof(path),
340             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
341             xen_domid, (uint64_t)physmap->phys_offset);
342     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->size);
343     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
344         return -1;
345     }
346     if (physmap->name) {
347         snprintf(path, sizeof(path),
348                 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
349                 xen_domid, (uint64_t)physmap->phys_offset);
350         if (!xs_write(state->xenstore, 0, path,
351                       physmap->name, strlen(physmap->name))) {
352             return -1;
353         }
354     }
355     return 0;
356 }
357 #else
358 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
359 {
360     return 0;
361 }
362 #endif
363 
364 static int xen_add_to_physmap(XenIOState *state,
365                               hwaddr start_addr,
366                               ram_addr_t size,
367                               MemoryRegion *mr,
368                               hwaddr offset_within_region)
369 {
370     unsigned long nr_pages;
371     int rc = 0;
372     XenPhysmap *physmap = NULL;
373     hwaddr pfn, start_gpfn;
374     hwaddr phys_offset = memory_region_get_ram_addr(mr);
375     const char *mr_name;
376 
377     if (get_physmapping(start_addr, size)) {
378         return 0;
379     }
380     if (size <= 0) {
381         return -1;
382     }
383 
384     /* Xen can only handle a single dirty log region for now and we want
385      * the linear framebuffer to be that region.
386      * Avoid tracking any regions that is not videoram and avoid tracking
387      * the legacy vga region. */
388     if (mr == framebuffer && start_addr > 0xbffff) {
389         goto go_physmap;
390     }
391     return -1;
392 
393 go_physmap:
394     DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
395             start_addr, start_addr + size);
396 
397     mr_name = memory_region_name(mr);
398 
399     physmap = g_malloc(sizeof(XenPhysmap));
400 
401     physmap->start_addr = start_addr;
402     physmap->size = size;
403     physmap->name = mr_name;
404     physmap->phys_offset = phys_offset;
405 
406     QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
407 
408     if (runstate_check(RUN_STATE_INMIGRATE)) {
409         /* Now when we have a physmap entry we can replace a dummy mapping with
410          * a real one of guest foreign memory. */
411         uint8_t *p = xen_replace_cache_entry(phys_offset, start_addr, size);
412         assert(p && p == memory_region_get_ram_ptr(mr));
413 
414         return 0;
415     }
416 
417     pfn = phys_offset >> TARGET_PAGE_BITS;
418     start_gpfn = start_addr >> TARGET_PAGE_BITS;
419     nr_pages = size >> TARGET_PAGE_BITS;
420     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, nr_pages, pfn,
421                                         start_gpfn);
422     if (rc) {
423         int saved_errno = errno;
424 
425         error_report("relocate_memory %lu pages from GFN %"HWADDR_PRIx
426                      " to GFN %"HWADDR_PRIx" failed: %s",
427                      nr_pages, pfn, start_gpfn, strerror(saved_errno));
428         errno = saved_errno;
429         return -1;
430     }
431 
432     rc = xendevicemodel_pin_memory_cacheattr(xen_dmod, xen_domid,
433                                    start_addr >> TARGET_PAGE_BITS,
434                                    (start_addr + size - 1) >> TARGET_PAGE_BITS,
435                                    XEN_DOMCTL_MEM_CACHEATTR_WB);
436     if (rc) {
437         error_report("pin_memory_cacheattr failed: %s", strerror(errno));
438     }
439     return xen_save_physmap(state, physmap);
440 }
441 
442 static int xen_remove_from_physmap(XenIOState *state,
443                                    hwaddr start_addr,
444                                    ram_addr_t size)
445 {
446     int rc = 0;
447     XenPhysmap *physmap = NULL;
448     hwaddr phys_offset = 0;
449 
450     physmap = get_physmapping(start_addr, size);
451     if (physmap == NULL) {
452         return -1;
453     }
454 
455     phys_offset = physmap->phys_offset;
456     size = physmap->size;
457 
458     DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
459             "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
460 
461     size >>= TARGET_PAGE_BITS;
462     start_addr >>= TARGET_PAGE_BITS;
463     phys_offset >>= TARGET_PAGE_BITS;
464     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, size, start_addr,
465                                         phys_offset);
466     if (rc) {
467         int saved_errno = errno;
468 
469         error_report("relocate_memory "RAM_ADDR_FMT" pages"
470                      " from GFN %"HWADDR_PRIx
471                      " to GFN %"HWADDR_PRIx" failed: %s",
472                      size, start_addr, phys_offset, strerror(saved_errno));
473         errno = saved_errno;
474         return -1;
475     }
476 
477     QLIST_REMOVE(physmap, list);
478     if (state->log_for_dirtybit == physmap) {
479         state->log_for_dirtybit = NULL;
480         g_free(state->dirty_bitmap);
481         state->dirty_bitmap = NULL;
482     }
483     g_free(physmap);
484 
485     return 0;
486 }
487 
488 static void xen_set_memory(struct MemoryListener *listener,
489                            MemoryRegionSection *section,
490                            bool add)
491 {
492     XenIOState *state = container_of(listener, XenIOState, memory_listener);
493     hwaddr start_addr = section->offset_within_address_space;
494     ram_addr_t size = int128_get64(section->size);
495     bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA);
496     hvmmem_type_t mem_type;
497 
498     if (section->mr == &ram_memory) {
499         return;
500     } else {
501         if (add) {
502             xen_map_memory_section(xen_domid, state->ioservid,
503                                    section);
504         } else {
505             xen_unmap_memory_section(xen_domid, state->ioservid,
506                                      section);
507         }
508     }
509 
510     if (!memory_region_is_ram(section->mr)) {
511         return;
512     }
513 
514     if (log_dirty != add) {
515         return;
516     }
517 
518     trace_xen_client_set_memory(start_addr, size, log_dirty);
519 
520     start_addr &= TARGET_PAGE_MASK;
521     size = TARGET_PAGE_ALIGN(size);
522 
523     if (add) {
524         if (!memory_region_is_rom(section->mr)) {
525             xen_add_to_physmap(state, start_addr, size,
526                                section->mr, section->offset_within_region);
527         } else {
528             mem_type = HVMMEM_ram_ro;
529             if (xen_set_mem_type(xen_domid, mem_type,
530                                  start_addr >> TARGET_PAGE_BITS,
531                                  size >> TARGET_PAGE_BITS)) {
532                 DPRINTF("xen_set_mem_type error, addr: "TARGET_FMT_plx"\n",
533                         start_addr);
534             }
535         }
536     } else {
537         if (xen_remove_from_physmap(state, start_addr, size) < 0) {
538             DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
539         }
540     }
541 }
542 
543 static void xen_region_add(MemoryListener *listener,
544                            MemoryRegionSection *section)
545 {
546     memory_region_ref(section->mr);
547     xen_set_memory(listener, section, true);
548 }
549 
550 static void xen_region_del(MemoryListener *listener,
551                            MemoryRegionSection *section)
552 {
553     xen_set_memory(listener, section, false);
554     memory_region_unref(section->mr);
555 }
556 
557 static void xen_io_add(MemoryListener *listener,
558                        MemoryRegionSection *section)
559 {
560     XenIOState *state = container_of(listener, XenIOState, io_listener);
561     MemoryRegion *mr = section->mr;
562 
563     if (mr->ops == &unassigned_io_ops) {
564         return;
565     }
566 
567     memory_region_ref(mr);
568 
569     xen_map_io_section(xen_domid, state->ioservid, section);
570 }
571 
572 static void xen_io_del(MemoryListener *listener,
573                        MemoryRegionSection *section)
574 {
575     XenIOState *state = container_of(listener, XenIOState, io_listener);
576     MemoryRegion *mr = section->mr;
577 
578     if (mr->ops == &unassigned_io_ops) {
579         return;
580     }
581 
582     xen_unmap_io_section(xen_domid, state->ioservid, section);
583 
584     memory_region_unref(mr);
585 }
586 
587 static void xen_device_realize(DeviceListener *listener,
588                                DeviceState *dev)
589 {
590     XenIOState *state = container_of(listener, XenIOState, device_listener);
591 
592     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
593         PCIDevice *pci_dev = PCI_DEVICE(dev);
594         XenPciDevice *xendev = g_new(XenPciDevice, 1);
595 
596         xendev->pci_dev = pci_dev;
597         xendev->sbdf = PCI_BUILD_BDF(pci_dev_bus_num(pci_dev),
598                                      pci_dev->devfn);
599         QLIST_INSERT_HEAD(&state->dev_list, xendev, entry);
600 
601         xen_map_pcidev(xen_domid, state->ioservid, pci_dev);
602     }
603 }
604 
605 static void xen_device_unrealize(DeviceListener *listener,
606                                  DeviceState *dev)
607 {
608     XenIOState *state = container_of(listener, XenIOState, device_listener);
609 
610     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
611         PCIDevice *pci_dev = PCI_DEVICE(dev);
612         XenPciDevice *xendev, *next;
613 
614         xen_unmap_pcidev(xen_domid, state->ioservid, pci_dev);
615 
616         QLIST_FOREACH_SAFE(xendev, &state->dev_list, entry, next) {
617             if (xendev->pci_dev == pci_dev) {
618                 QLIST_REMOVE(xendev, entry);
619                 g_free(xendev);
620                 break;
621             }
622         }
623     }
624 }
625 
626 static void xen_sync_dirty_bitmap(XenIOState *state,
627                                   hwaddr start_addr,
628                                   ram_addr_t size)
629 {
630     hwaddr npages = size >> TARGET_PAGE_BITS;
631     const int width = sizeof(unsigned long) * 8;
632     size_t bitmap_size = DIV_ROUND_UP(npages, width);
633     int rc, i, j;
634     const XenPhysmap *physmap = NULL;
635 
636     physmap = get_physmapping(start_addr, size);
637     if (physmap == NULL) {
638         /* not handled */
639         return;
640     }
641 
642     if (state->log_for_dirtybit == NULL) {
643         state->log_for_dirtybit = physmap;
644         state->dirty_bitmap = g_new(unsigned long, bitmap_size);
645     } else if (state->log_for_dirtybit != physmap) {
646         /* Only one range for dirty bitmap can be tracked. */
647         return;
648     }
649 
650     rc = xen_track_dirty_vram(xen_domid, start_addr >> TARGET_PAGE_BITS,
651                               npages, state->dirty_bitmap);
652     if (rc < 0) {
653 #ifndef ENODATA
654 #define ENODATA  ENOENT
655 #endif
656         if (errno == ENODATA) {
657             memory_region_set_dirty(framebuffer, 0, size);
658             DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
659                     ", 0x" TARGET_FMT_plx "): %s\n",
660                     start_addr, start_addr + size, strerror(errno));
661         }
662         return;
663     }
664 
665     for (i = 0; i < bitmap_size; i++) {
666         unsigned long map = state->dirty_bitmap[i];
667         while (map != 0) {
668             j = ctzl(map);
669             map &= ~(1ul << j);
670             memory_region_set_dirty(framebuffer,
671                                     (i * width + j) * TARGET_PAGE_SIZE,
672                                     TARGET_PAGE_SIZE);
673         };
674     }
675 }
676 
677 static void xen_log_start(MemoryListener *listener,
678                           MemoryRegionSection *section,
679                           int old, int new)
680 {
681     XenIOState *state = container_of(listener, XenIOState, memory_listener);
682 
683     if (new & ~old & (1 << DIRTY_MEMORY_VGA)) {
684         xen_sync_dirty_bitmap(state, section->offset_within_address_space,
685                               int128_get64(section->size));
686     }
687 }
688 
689 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section,
690                          int old, int new)
691 {
692     XenIOState *state = container_of(listener, XenIOState, memory_listener);
693 
694     if (old & ~new & (1 << DIRTY_MEMORY_VGA)) {
695         state->log_for_dirtybit = NULL;
696         g_free(state->dirty_bitmap);
697         state->dirty_bitmap = NULL;
698         /* Disable dirty bit tracking */
699         xen_track_dirty_vram(xen_domid, 0, 0, NULL);
700     }
701 }
702 
703 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
704 {
705     XenIOState *state = container_of(listener, XenIOState, memory_listener);
706 
707     xen_sync_dirty_bitmap(state, section->offset_within_address_space,
708                           int128_get64(section->size));
709 }
710 
711 static void xen_log_global_start(MemoryListener *listener)
712 {
713     if (xen_enabled()) {
714         xen_in_migration = true;
715     }
716 }
717 
718 static void xen_log_global_stop(MemoryListener *listener)
719 {
720     xen_in_migration = false;
721 }
722 
723 static MemoryListener xen_memory_listener = {
724     .region_add = xen_region_add,
725     .region_del = xen_region_del,
726     .log_start = xen_log_start,
727     .log_stop = xen_log_stop,
728     .log_sync = xen_log_sync,
729     .log_global_start = xen_log_global_start,
730     .log_global_stop = xen_log_global_stop,
731     .priority = 10,
732 };
733 
734 static MemoryListener xen_io_listener = {
735     .region_add = xen_io_add,
736     .region_del = xen_io_del,
737     .priority = 10,
738 };
739 
740 static DeviceListener xen_device_listener = {
741     .realize = xen_device_realize,
742     .unrealize = xen_device_unrealize,
743 };
744 
745 /* get the ioreq packets from share mem */
746 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
747 {
748     ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
749 
750     if (req->state != STATE_IOREQ_READY) {
751         DPRINTF("I/O request not ready: "
752                 "%x, ptr: %x, port: %"PRIx64", "
753                 "data: %"PRIx64", count: %u, size: %u\n",
754                 req->state, req->data_is_ptr, req->addr,
755                 req->data, req->count, req->size);
756         return NULL;
757     }
758 
759     xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
760 
761     req->state = STATE_IOREQ_INPROCESS;
762     return req;
763 }
764 
765 /* use poll to get the port notification */
766 /* ioreq_vec--out,the */
767 /* retval--the number of ioreq packet */
768 static ioreq_t *cpu_get_ioreq(XenIOState *state)
769 {
770     MachineState *ms = MACHINE(qdev_get_machine());
771     unsigned int max_cpus = ms->smp.max_cpus;
772     int i;
773     evtchn_port_t port;
774 
775     port = xenevtchn_pending(state->xce_handle);
776     if (port == state->bufioreq_local_port) {
777         timer_mod(state->buffered_io_timer,
778                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
779         return NULL;
780     }
781 
782     if (port != -1) {
783         for (i = 0; i < max_cpus; i++) {
784             if (state->ioreq_local_port[i] == port) {
785                 break;
786             }
787         }
788 
789         if (i == max_cpus) {
790             hw_error("Fatal error while trying to get io event!\n");
791         }
792 
793         /* unmask the wanted port again */
794         xenevtchn_unmask(state->xce_handle, port);
795 
796         /* get the io packet from shared memory */
797         state->send_vcpu = i;
798         return cpu_get_ioreq_from_shared_memory(state, i);
799     }
800 
801     /* read error or read nothing */
802     return NULL;
803 }
804 
805 static uint32_t do_inp(uint32_t addr, unsigned long size)
806 {
807     switch (size) {
808         case 1:
809             return cpu_inb(addr);
810         case 2:
811             return cpu_inw(addr);
812         case 4:
813             return cpu_inl(addr);
814         default:
815             hw_error("inp: bad size: %04x %lx", addr, size);
816     }
817 }
818 
819 static void do_outp(uint32_t addr,
820         unsigned long size, uint32_t val)
821 {
822     switch (size) {
823         case 1:
824             return cpu_outb(addr, val);
825         case 2:
826             return cpu_outw(addr, val);
827         case 4:
828             return cpu_outl(addr, val);
829         default:
830             hw_error("outp: bad size: %04x %lx", addr, size);
831     }
832 }
833 
834 /*
835  * Helper functions which read/write an object from/to physical guest
836  * memory, as part of the implementation of an ioreq.
837  *
838  * Equivalent to
839  *   cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
840  *                          val, req->size, 0/1)
841  * except without the integer overflow problems.
842  */
843 static void rw_phys_req_item(hwaddr addr,
844                              ioreq_t *req, uint32_t i, void *val, int rw)
845 {
846     /* Do everything unsigned so overflow just results in a truncated result
847      * and accesses to undesired parts of guest memory, which is up
848      * to the guest */
849     hwaddr offset = (hwaddr)req->size * i;
850     if (req->df) {
851         addr -= offset;
852     } else {
853         addr += offset;
854     }
855     cpu_physical_memory_rw(addr, val, req->size, rw);
856 }
857 
858 static inline void read_phys_req_item(hwaddr addr,
859                                       ioreq_t *req, uint32_t i, void *val)
860 {
861     rw_phys_req_item(addr, req, i, val, 0);
862 }
863 static inline void write_phys_req_item(hwaddr addr,
864                                        ioreq_t *req, uint32_t i, void *val)
865 {
866     rw_phys_req_item(addr, req, i, val, 1);
867 }
868 
869 
870 static void cpu_ioreq_pio(ioreq_t *req)
871 {
872     uint32_t i;
873 
874     trace_cpu_ioreq_pio(req, req->dir, req->df, req->data_is_ptr, req->addr,
875                          req->data, req->count, req->size);
876 
877     if (req->size > sizeof(uint32_t)) {
878         hw_error("PIO: bad size (%u)", req->size);
879     }
880 
881     if (req->dir == IOREQ_READ) {
882         if (!req->data_is_ptr) {
883             req->data = do_inp(req->addr, req->size);
884             trace_cpu_ioreq_pio_read_reg(req, req->data, req->addr,
885                                          req->size);
886         } else {
887             uint32_t tmp;
888 
889             for (i = 0; i < req->count; i++) {
890                 tmp = do_inp(req->addr, req->size);
891                 write_phys_req_item(req->data, req, i, &tmp);
892             }
893         }
894     } else if (req->dir == IOREQ_WRITE) {
895         if (!req->data_is_ptr) {
896             trace_cpu_ioreq_pio_write_reg(req, req->data, req->addr,
897                                           req->size);
898             do_outp(req->addr, req->size, req->data);
899         } else {
900             for (i = 0; i < req->count; i++) {
901                 uint32_t tmp = 0;
902 
903                 read_phys_req_item(req->data, req, i, &tmp);
904                 do_outp(req->addr, req->size, tmp);
905             }
906         }
907     }
908 }
909 
910 static void cpu_ioreq_move(ioreq_t *req)
911 {
912     uint32_t i;
913 
914     trace_cpu_ioreq_move(req, req->dir, req->df, req->data_is_ptr, req->addr,
915                          req->data, req->count, req->size);
916 
917     if (req->size > sizeof(req->data)) {
918         hw_error("MMIO: bad size (%u)", req->size);
919     }
920 
921     if (!req->data_is_ptr) {
922         if (req->dir == IOREQ_READ) {
923             for (i = 0; i < req->count; i++) {
924                 read_phys_req_item(req->addr, req, i, &req->data);
925             }
926         } else if (req->dir == IOREQ_WRITE) {
927             for (i = 0; i < req->count; i++) {
928                 write_phys_req_item(req->addr, req, i, &req->data);
929             }
930         }
931     } else {
932         uint64_t tmp;
933 
934         if (req->dir == IOREQ_READ) {
935             for (i = 0; i < req->count; i++) {
936                 read_phys_req_item(req->addr, req, i, &tmp);
937                 write_phys_req_item(req->data, req, i, &tmp);
938             }
939         } else if (req->dir == IOREQ_WRITE) {
940             for (i = 0; i < req->count; i++) {
941                 read_phys_req_item(req->data, req, i, &tmp);
942                 write_phys_req_item(req->addr, req, i, &tmp);
943             }
944         }
945     }
946 }
947 
948 static void cpu_ioreq_config(XenIOState *state, ioreq_t *req)
949 {
950     uint32_t sbdf = req->addr >> 32;
951     uint32_t reg = req->addr;
952     XenPciDevice *xendev;
953 
954     if (req->size != sizeof(uint8_t) && req->size != sizeof(uint16_t) &&
955         req->size != sizeof(uint32_t)) {
956         hw_error("PCI config access: bad size (%u)", req->size);
957     }
958 
959     if (req->count != 1) {
960         hw_error("PCI config access: bad count (%u)", req->count);
961     }
962 
963     QLIST_FOREACH(xendev, &state->dev_list, entry) {
964         if (xendev->sbdf != sbdf) {
965             continue;
966         }
967 
968         if (!req->data_is_ptr) {
969             if (req->dir == IOREQ_READ) {
970                 req->data = pci_host_config_read_common(
971                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
972                     req->size);
973                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
974                                             req->size, req->data);
975             } else if (req->dir == IOREQ_WRITE) {
976                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
977                                              req->size, req->data);
978                 pci_host_config_write_common(
979                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
980                     req->data, req->size);
981             }
982         } else {
983             uint32_t tmp;
984 
985             if (req->dir == IOREQ_READ) {
986                 tmp = pci_host_config_read_common(
987                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
988                     req->size);
989                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
990                                             req->size, tmp);
991                 write_phys_req_item(req->data, req, 0, &tmp);
992             } else if (req->dir == IOREQ_WRITE) {
993                 read_phys_req_item(req->data, req, 0, &tmp);
994                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
995                                              req->size, tmp);
996                 pci_host_config_write_common(
997                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
998                     tmp, req->size);
999             }
1000         }
1001     }
1002 }
1003 
1004 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
1005 {
1006     X86CPU *cpu;
1007     CPUX86State *env;
1008 
1009     cpu = X86_CPU(current_cpu);
1010     env = &cpu->env;
1011     env->regs[R_EAX] = req->data;
1012     env->regs[R_EBX] = vmport_regs->ebx;
1013     env->regs[R_ECX] = vmport_regs->ecx;
1014     env->regs[R_EDX] = vmport_regs->edx;
1015     env->regs[R_ESI] = vmport_regs->esi;
1016     env->regs[R_EDI] = vmport_regs->edi;
1017 }
1018 
1019 static void regs_from_cpu(vmware_regs_t *vmport_regs)
1020 {
1021     X86CPU *cpu = X86_CPU(current_cpu);
1022     CPUX86State *env = &cpu->env;
1023 
1024     vmport_regs->ebx = env->regs[R_EBX];
1025     vmport_regs->ecx = env->regs[R_ECX];
1026     vmport_regs->edx = env->regs[R_EDX];
1027     vmport_regs->esi = env->regs[R_ESI];
1028     vmport_regs->edi = env->regs[R_EDI];
1029 }
1030 
1031 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
1032 {
1033     vmware_regs_t *vmport_regs;
1034 
1035     assert(state->shared_vmport_page);
1036     vmport_regs =
1037         &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
1038     QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
1039 
1040     current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
1041     regs_to_cpu(vmport_regs, req);
1042     cpu_ioreq_pio(req);
1043     regs_from_cpu(vmport_regs);
1044     current_cpu = NULL;
1045 }
1046 
1047 static void handle_ioreq(XenIOState *state, ioreq_t *req)
1048 {
1049     trace_handle_ioreq(req, req->type, req->dir, req->df, req->data_is_ptr,
1050                        req->addr, req->data, req->count, req->size);
1051 
1052     if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
1053             (req->size < sizeof (target_ulong))) {
1054         req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
1055     }
1056 
1057     if (req->dir == IOREQ_WRITE)
1058         trace_handle_ioreq_write(req, req->type, req->df, req->data_is_ptr,
1059                                  req->addr, req->data, req->count, req->size);
1060 
1061     switch (req->type) {
1062         case IOREQ_TYPE_PIO:
1063             cpu_ioreq_pio(req);
1064             break;
1065         case IOREQ_TYPE_COPY:
1066             cpu_ioreq_move(req);
1067             break;
1068         case IOREQ_TYPE_VMWARE_PORT:
1069             handle_vmport_ioreq(state, req);
1070             break;
1071         case IOREQ_TYPE_TIMEOFFSET:
1072             break;
1073         case IOREQ_TYPE_INVALIDATE:
1074             xen_invalidate_map_cache();
1075             break;
1076         case IOREQ_TYPE_PCI_CONFIG:
1077             cpu_ioreq_config(state, req);
1078             break;
1079         default:
1080             hw_error("Invalid ioreq type 0x%x\n", req->type);
1081     }
1082     if (req->dir == IOREQ_READ) {
1083         trace_handle_ioreq_read(req, req->type, req->df, req->data_is_ptr,
1084                                 req->addr, req->data, req->count, req->size);
1085     }
1086 }
1087 
1088 static int handle_buffered_iopage(XenIOState *state)
1089 {
1090     buffered_iopage_t *buf_page = state->buffered_io_page;
1091     buf_ioreq_t *buf_req = NULL;
1092     ioreq_t req;
1093     int qw;
1094 
1095     if (!buf_page) {
1096         return 0;
1097     }
1098 
1099     memset(&req, 0x00, sizeof(req));
1100     req.state = STATE_IOREQ_READY;
1101     req.count = 1;
1102     req.dir = IOREQ_WRITE;
1103 
1104     for (;;) {
1105         uint32_t rdptr = buf_page->read_pointer, wrptr;
1106 
1107         xen_rmb();
1108         wrptr = buf_page->write_pointer;
1109         xen_rmb();
1110         if (rdptr != buf_page->read_pointer) {
1111             continue;
1112         }
1113         if (rdptr == wrptr) {
1114             break;
1115         }
1116         buf_req = &buf_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
1117         req.size = 1U << buf_req->size;
1118         req.addr = buf_req->addr;
1119         req.data = buf_req->data;
1120         req.type = buf_req->type;
1121         xen_rmb();
1122         qw = (req.size == 8);
1123         if (qw) {
1124             if (rdptr + 1 == wrptr) {
1125                 hw_error("Incomplete quad word buffered ioreq");
1126             }
1127             buf_req = &buf_page->buf_ioreq[(rdptr + 1) %
1128                                            IOREQ_BUFFER_SLOT_NUM];
1129             req.data |= ((uint64_t)buf_req->data) << 32;
1130             xen_rmb();
1131         }
1132 
1133         handle_ioreq(state, &req);
1134 
1135         /* Only req.data may get updated by handle_ioreq(), albeit even that
1136          * should not happen as such data would never make it to the guest (we
1137          * can only usefully see writes here after all).
1138          */
1139         assert(req.state == STATE_IOREQ_READY);
1140         assert(req.count == 1);
1141         assert(req.dir == IOREQ_WRITE);
1142         assert(!req.data_is_ptr);
1143 
1144         qatomic_add(&buf_page->read_pointer, qw + 1);
1145     }
1146 
1147     return req.count;
1148 }
1149 
1150 static void handle_buffered_io(void *opaque)
1151 {
1152     XenIOState *state = opaque;
1153 
1154     if (handle_buffered_iopage(state)) {
1155         timer_mod(state->buffered_io_timer,
1156                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
1157     } else {
1158         timer_del(state->buffered_io_timer);
1159         xenevtchn_unmask(state->xce_handle, state->bufioreq_local_port);
1160     }
1161 }
1162 
1163 static void cpu_handle_ioreq(void *opaque)
1164 {
1165     XenIOState *state = opaque;
1166     ioreq_t *req = cpu_get_ioreq(state);
1167 
1168     handle_buffered_iopage(state);
1169     if (req) {
1170         ioreq_t copy = *req;
1171 
1172         xen_rmb();
1173         handle_ioreq(state, &copy);
1174         req->data = copy.data;
1175 
1176         if (req->state != STATE_IOREQ_INPROCESS) {
1177             fprintf(stderr, "Badness in I/O request ... not in service?!: "
1178                     "%x, ptr: %x, port: %"PRIx64", "
1179                     "data: %"PRIx64", count: %u, size: %u, type: %u\n",
1180                     req->state, req->data_is_ptr, req->addr,
1181                     req->data, req->count, req->size, req->type);
1182             destroy_hvm_domain(false);
1183             return;
1184         }
1185 
1186         xen_wmb(); /* Update ioreq contents /then/ update state. */
1187 
1188         /*
1189          * We do this before we send the response so that the tools
1190          * have the opportunity to pick up on the reset before the
1191          * guest resumes and does a hlt with interrupts disabled which
1192          * causes Xen to powerdown the domain.
1193          */
1194         if (runstate_is_running()) {
1195             ShutdownCause request;
1196 
1197             if (qemu_shutdown_requested_get()) {
1198                 destroy_hvm_domain(false);
1199             }
1200             request = qemu_reset_requested_get();
1201             if (request) {
1202                 qemu_system_reset(request);
1203                 destroy_hvm_domain(true);
1204             }
1205         }
1206 
1207         req->state = STATE_IORESP_READY;
1208         xenevtchn_notify(state->xce_handle,
1209                          state->ioreq_local_port[state->send_vcpu]);
1210     }
1211 }
1212 
1213 static void xen_main_loop_prepare(XenIOState *state)
1214 {
1215     int evtchn_fd = -1;
1216 
1217     if (state->xce_handle != NULL) {
1218         evtchn_fd = xenevtchn_fd(state->xce_handle);
1219     }
1220 
1221     state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
1222                                                  state);
1223 
1224     if (evtchn_fd != -1) {
1225         CPUState *cpu_state;
1226 
1227         DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
1228         CPU_FOREACH(cpu_state) {
1229             DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1230                     __func__, cpu_state->cpu_index, cpu_state);
1231             state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
1232         }
1233         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
1234     }
1235 }
1236 
1237 
1238 static void xen_hvm_change_state_handler(void *opaque, int running,
1239                                          RunState rstate)
1240 {
1241     XenIOState *state = opaque;
1242 
1243     if (running) {
1244         xen_main_loop_prepare(state);
1245     }
1246 
1247     xen_set_ioreq_server_state(xen_domid,
1248                                state->ioservid,
1249                                (rstate == RUN_STATE_RUNNING));
1250 }
1251 
1252 static void xen_exit_notifier(Notifier *n, void *data)
1253 {
1254     XenIOState *state = container_of(n, XenIOState, exit);
1255 
1256     xen_destroy_ioreq_server(xen_domid, state->ioservid);
1257 
1258     xenevtchn_close(state->xce_handle);
1259     xs_daemon_close(state->xenstore);
1260 }
1261 
1262 #ifdef XEN_COMPAT_PHYSMAP
1263 static void xen_read_physmap(XenIOState *state)
1264 {
1265     XenPhysmap *physmap = NULL;
1266     unsigned int len, num, i;
1267     char path[80], *value = NULL;
1268     char **entries = NULL;
1269 
1270     snprintf(path, sizeof(path),
1271             "/local/domain/0/device-model/%d/physmap", xen_domid);
1272     entries = xs_directory(state->xenstore, 0, path, &num);
1273     if (entries == NULL)
1274         return;
1275 
1276     for (i = 0; i < num; i++) {
1277         physmap = g_malloc(sizeof (XenPhysmap));
1278         physmap->phys_offset = strtoull(entries[i], NULL, 16);
1279         snprintf(path, sizeof(path),
1280                 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1281                 xen_domid, entries[i]);
1282         value = xs_read(state->xenstore, 0, path, &len);
1283         if (value == NULL) {
1284             g_free(physmap);
1285             continue;
1286         }
1287         physmap->start_addr = strtoull(value, NULL, 16);
1288         free(value);
1289 
1290         snprintf(path, sizeof(path),
1291                 "/local/domain/0/device-model/%d/physmap/%s/size",
1292                 xen_domid, entries[i]);
1293         value = xs_read(state->xenstore, 0, path, &len);
1294         if (value == NULL) {
1295             g_free(physmap);
1296             continue;
1297         }
1298         physmap->size = strtoull(value, NULL, 16);
1299         free(value);
1300 
1301         snprintf(path, sizeof(path),
1302                 "/local/domain/0/device-model/%d/physmap/%s/name",
1303                 xen_domid, entries[i]);
1304         physmap->name = xs_read(state->xenstore, 0, path, &len);
1305 
1306         QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
1307     }
1308     free(entries);
1309 }
1310 #else
1311 static void xen_read_physmap(XenIOState *state)
1312 {
1313 }
1314 #endif
1315 
1316 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1317 {
1318     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1319 }
1320 
1321 static int xen_map_ioreq_server(XenIOState *state)
1322 {
1323     void *addr = NULL;
1324     xenforeignmemory_resource_handle *fres;
1325     xen_pfn_t ioreq_pfn;
1326     xen_pfn_t bufioreq_pfn;
1327     evtchn_port_t bufioreq_evtchn;
1328     int rc;
1329 
1330     /*
1331      * Attempt to map using the resource API and fall back to normal
1332      * foreign mapping if this is not supported.
1333      */
1334     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_bufioreq != 0);
1335     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_ioreq(0) != 1);
1336     fres = xenforeignmemory_map_resource(xen_fmem, xen_domid,
1337                                          XENMEM_resource_ioreq_server,
1338                                          state->ioservid, 0, 2,
1339                                          &addr,
1340                                          PROT_READ | PROT_WRITE, 0);
1341     if (fres != NULL) {
1342         trace_xen_map_resource_ioreq(state->ioservid, addr);
1343         state->buffered_io_page = addr;
1344         state->shared_page = addr + TARGET_PAGE_SIZE;
1345     } else if (errno != EOPNOTSUPP) {
1346         error_report("failed to map ioreq server resources: error %d handle=%p",
1347                      errno, xen_xc);
1348         return -1;
1349     }
1350 
1351     rc = xen_get_ioreq_server_info(xen_domid, state->ioservid,
1352                                    (state->shared_page == NULL) ?
1353                                    &ioreq_pfn : NULL,
1354                                    (state->buffered_io_page == NULL) ?
1355                                    &bufioreq_pfn : NULL,
1356                                    &bufioreq_evtchn);
1357     if (rc < 0) {
1358         error_report("failed to get ioreq server info: error %d handle=%p",
1359                      errno, xen_xc);
1360         return rc;
1361     }
1362 
1363     if (state->shared_page == NULL) {
1364         DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1365 
1366         state->shared_page = xenforeignmemory_map(xen_fmem, xen_domid,
1367                                                   PROT_READ | PROT_WRITE,
1368                                                   1, &ioreq_pfn, NULL);
1369         if (state->shared_page == NULL) {
1370             error_report("map shared IO page returned error %d handle=%p",
1371                          errno, xen_xc);
1372         }
1373     }
1374 
1375     if (state->buffered_io_page == NULL) {
1376         DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
1377 
1378         state->buffered_io_page = xenforeignmemory_map(xen_fmem, xen_domid,
1379                                                        PROT_READ | PROT_WRITE,
1380                                                        1, &bufioreq_pfn,
1381                                                        NULL);
1382         if (state->buffered_io_page == NULL) {
1383             error_report("map buffered IO page returned error %d", errno);
1384             return -1;
1385         }
1386     }
1387 
1388     if (state->shared_page == NULL || state->buffered_io_page == NULL) {
1389         return -1;
1390     }
1391 
1392     DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
1393 
1394     state->bufioreq_remote_port = bufioreq_evtchn;
1395 
1396     return 0;
1397 }
1398 
1399 void xen_hvm_init_pc(PCMachineState *pcms, MemoryRegion **ram_memory)
1400 {
1401     MachineState *ms = MACHINE(pcms);
1402     unsigned int max_cpus = ms->smp.max_cpus;
1403     int i, rc;
1404     xen_pfn_t ioreq_pfn;
1405     XenIOState *state;
1406 
1407     state = g_malloc0(sizeof (XenIOState));
1408 
1409     state->xce_handle = xenevtchn_open(NULL, 0);
1410     if (state->xce_handle == NULL) {
1411         perror("xen: event channel open");
1412         goto err;
1413     }
1414 
1415     state->xenstore = xs_daemon_open();
1416     if (state->xenstore == NULL) {
1417         perror("xen: xenstore open");
1418         goto err;
1419     }
1420 
1421     xen_create_ioreq_server(xen_domid, &state->ioservid);
1422 
1423     state->exit.notify = xen_exit_notifier;
1424     qemu_add_exit_notifier(&state->exit);
1425 
1426     state->suspend.notify = xen_suspend_notifier;
1427     qemu_register_suspend_notifier(&state->suspend);
1428 
1429     state->wakeup.notify = xen_wakeup_notifier;
1430     qemu_register_wakeup_notifier(&state->wakeup);
1431 
1432     /*
1433      * Register wake-up support in QMP query-current-machine API
1434      */
1435     qemu_register_wakeup_support();
1436 
1437     rc = xen_map_ioreq_server(state);
1438     if (rc < 0) {
1439         goto err;
1440     }
1441 
1442     rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1443     if (!rc) {
1444         DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1445         state->shared_vmport_page =
1446             xenforeignmemory_map(xen_fmem, xen_domid, PROT_READ|PROT_WRITE,
1447                                  1, &ioreq_pfn, NULL);
1448         if (state->shared_vmport_page == NULL) {
1449             error_report("map shared vmport IO page returned error %d handle=%p",
1450                          errno, xen_xc);
1451             goto err;
1452         }
1453     } else if (rc != -ENOSYS) {
1454         error_report("get vmport regs pfn returned error %d, rc=%d",
1455                      errno, rc);
1456         goto err;
1457     }
1458 
1459     /* Note: cpus is empty at this point in init */
1460     state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1461 
1462     rc = xen_set_ioreq_server_state(xen_domid, state->ioservid, true);
1463     if (rc < 0) {
1464         error_report("failed to enable ioreq server info: error %d handle=%p",
1465                      errno, xen_xc);
1466         goto err;
1467     }
1468 
1469     state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1470 
1471     /* FIXME: how about if we overflow the page here? */
1472     for (i = 0; i < max_cpus; i++) {
1473         rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1474                                         xen_vcpu_eport(state->shared_page, i));
1475         if (rc == -1) {
1476             error_report("shared evtchn %d bind error %d", i, errno);
1477             goto err;
1478         }
1479         state->ioreq_local_port[i] = rc;
1480     }
1481 
1482     rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1483                                     state->bufioreq_remote_port);
1484     if (rc == -1) {
1485         error_report("buffered evtchn bind error %d", errno);
1486         goto err;
1487     }
1488     state->bufioreq_local_port = rc;
1489 
1490     /* Init RAM management */
1491 #ifdef XEN_COMPAT_PHYSMAP
1492     xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1493 #else
1494     xen_map_cache_init(NULL, state);
1495 #endif
1496     xen_ram_init(pcms, ms->ram_size, ram_memory);
1497 
1498     qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1499 
1500     state->memory_listener = xen_memory_listener;
1501     memory_listener_register(&state->memory_listener, &address_space_memory);
1502     state->log_for_dirtybit = NULL;
1503 
1504     state->io_listener = xen_io_listener;
1505     memory_listener_register(&state->io_listener, &address_space_io);
1506 
1507     state->device_listener = xen_device_listener;
1508     QLIST_INIT(&state->dev_list);
1509     device_listener_register(&state->device_listener);
1510 
1511     xen_bus_init();
1512 
1513     /* Initialize backend core & drivers */
1514     if (xen_be_init() != 0) {
1515         error_report("xen backend core setup failed");
1516         goto err;
1517     }
1518     xen_be_register_common();
1519 
1520     QLIST_INIT(&xen_physmap);
1521     xen_read_physmap(state);
1522 
1523     /* Disable ACPI build because Xen handles it */
1524     pcms->acpi_build_enabled = false;
1525 
1526     return;
1527 
1528 err:
1529     error_report("xen hardware virtual machine initialisation failed");
1530     exit(1);
1531 }
1532 
1533 void destroy_hvm_domain(bool reboot)
1534 {
1535     xc_interface *xc_handle;
1536     int sts;
1537     int rc;
1538 
1539     unsigned int reason = reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff;
1540 
1541     if (xen_dmod) {
1542         rc = xendevicemodel_shutdown(xen_dmod, xen_domid, reason);
1543         if (!rc) {
1544             return;
1545         }
1546         if (errno != ENOTTY /* old Xen */) {
1547             perror("xendevicemodel_shutdown failed");
1548         }
1549         /* well, try the old thing then */
1550     }
1551 
1552     xc_handle = xc_interface_open(0, 0, 0);
1553     if (xc_handle == NULL) {
1554         fprintf(stderr, "Cannot acquire xenctrl handle\n");
1555     } else {
1556         sts = xc_domain_shutdown(xc_handle, xen_domid, reason);
1557         if (sts != 0) {
1558             fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1559                     "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1560                     sts, strerror(errno));
1561         } else {
1562             fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1563                     reboot ? "reboot" : "poweroff");
1564         }
1565         xc_interface_close(xc_handle);
1566     }
1567 }
1568 
1569 void xen_register_framebuffer(MemoryRegion *mr)
1570 {
1571     framebuffer = mr;
1572 }
1573 
1574 void xen_shutdown_fatal_error(const char *fmt, ...)
1575 {
1576     va_list ap;
1577 
1578     va_start(ap, fmt);
1579     vfprintf(stderr, fmt, ap);
1580     va_end(ap);
1581     fprintf(stderr, "Will destroy the domain.\n");
1582     /* destroy the domain */
1583     qemu_system_shutdown_request(SHUTDOWN_CAUSE_HOST_ERROR);
1584 }
1585 
1586 void xen_hvm_modified_memory(ram_addr_t start, ram_addr_t length)
1587 {
1588     if (unlikely(xen_in_migration)) {
1589         int rc;
1590         ram_addr_t start_pfn, nb_pages;
1591 
1592         start = xen_phys_offset_to_gaddr(start, length);
1593 
1594         if (length == 0) {
1595             length = TARGET_PAGE_SIZE;
1596         }
1597         start_pfn = start >> TARGET_PAGE_BITS;
1598         nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1599             - start_pfn;
1600         rc = xen_modified_memory(xen_domid, start_pfn, nb_pages);
1601         if (rc) {
1602             fprintf(stderr,
1603                     "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1604                     __func__, start, nb_pages, errno, strerror(errno));
1605         }
1606     }
1607 }
1608 
1609 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1610 {
1611     if (enable) {
1612         memory_global_dirty_log_start();
1613     } else {
1614         memory_global_dirty_log_stop();
1615     }
1616 }
1617