xref: /openbmc/qemu/hw/i386/xen/xen-hvm.c (revision 53422040)
1 /*
2  * Copyright (C) 2010       Citrix Ltd.
3  *
4  * This work is licensed under the terms of the GNU GPL, version 2.  See
5  * the COPYING file in the top-level directory.
6  *
7  * Contributions after 2012-01-13 are licensed under the terms of the
8  * GNU GPL, version 2 or (at your option) any later version.
9  */
10 
11 #include "qemu/osdep.h"
12 
13 #include "cpu.h"
14 #include "hw/pci/pci.h"
15 #include "hw/pci/pci_host.h"
16 #include "hw/i386/pc.h"
17 #include "hw/southbridge/piix.h"
18 #include "hw/irq.h"
19 #include "hw/hw.h"
20 #include "hw/i386/apic-msidef.h"
21 #include "hw/xen/xen_common.h"
22 #include "hw/xen/xen-legacy-backend.h"
23 #include "hw/xen/xen-bus.h"
24 #include "qapi/error.h"
25 #include "qapi/qapi-commands-misc.h"
26 #include "qemu/error-report.h"
27 #include "qemu/main-loop.h"
28 #include "qemu/range.h"
29 #include "sysemu/runstate.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/xen-mapcache.h"
32 #include "trace.h"
33 #include "exec/address-spaces.h"
34 
35 #include <xen/hvm/ioreq.h>
36 #include <xen/hvm/e820.h>
37 
38 //#define DEBUG_XEN_HVM
39 
40 #ifdef DEBUG_XEN_HVM
41 #define DPRINTF(fmt, ...) \
42     do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
43 #else
44 #define DPRINTF(fmt, ...) \
45     do { } while (0)
46 #endif
47 
48 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
49 static MemoryRegion *framebuffer;
50 static bool xen_in_migration;
51 
52 /* Compatibility with older version */
53 
54 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
55  * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
56  * needs to be included before this block and hw/xen/xen_common.h needs to
57  * be included before xen/hvm/ioreq.h
58  */
59 #ifndef IOREQ_TYPE_VMWARE_PORT
60 #define IOREQ_TYPE_VMWARE_PORT  3
61 struct vmware_regs {
62     uint32_t esi;
63     uint32_t edi;
64     uint32_t ebx;
65     uint32_t ecx;
66     uint32_t edx;
67 };
68 typedef struct vmware_regs vmware_regs_t;
69 
70 struct shared_vmport_iopage {
71     struct vmware_regs vcpu_vmport_regs[1];
72 };
73 typedef struct shared_vmport_iopage shared_vmport_iopage_t;
74 #endif
75 
76 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
77 {
78     return shared_page->vcpu_ioreq[i].vp_eport;
79 }
80 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
81 {
82     return &shared_page->vcpu_ioreq[vcpu];
83 }
84 
85 #define BUFFER_IO_MAX_DELAY  100
86 
87 typedef struct XenPhysmap {
88     hwaddr start_addr;
89     ram_addr_t size;
90     const char *name;
91     hwaddr phys_offset;
92 
93     QLIST_ENTRY(XenPhysmap) list;
94 } XenPhysmap;
95 
96 static QLIST_HEAD(, XenPhysmap) xen_physmap;
97 
98 typedef struct XenPciDevice {
99     PCIDevice *pci_dev;
100     uint32_t sbdf;
101     QLIST_ENTRY(XenPciDevice) entry;
102 } XenPciDevice;
103 
104 typedef struct XenIOState {
105     ioservid_t ioservid;
106     shared_iopage_t *shared_page;
107     shared_vmport_iopage_t *shared_vmport_page;
108     buffered_iopage_t *buffered_io_page;
109     QEMUTimer *buffered_io_timer;
110     CPUState **cpu_by_vcpu_id;
111     /* the evtchn port for polling the notification, */
112     evtchn_port_t *ioreq_local_port;
113     /* evtchn remote and local ports for buffered io */
114     evtchn_port_t bufioreq_remote_port;
115     evtchn_port_t bufioreq_local_port;
116     /* the evtchn fd for polling */
117     xenevtchn_handle *xce_handle;
118     /* which vcpu we are serving */
119     int send_vcpu;
120 
121     struct xs_handle *xenstore;
122     MemoryListener memory_listener;
123     MemoryListener io_listener;
124     QLIST_HEAD(, XenPciDevice) dev_list;
125     DeviceListener device_listener;
126     hwaddr free_phys_offset;
127     const XenPhysmap *log_for_dirtybit;
128     /* Buffer used by xen_sync_dirty_bitmap */
129     unsigned long *dirty_bitmap;
130 
131     Notifier exit;
132     Notifier suspend;
133     Notifier wakeup;
134 } XenIOState;
135 
136 /* Xen specific function for piix pci */
137 
138 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
139 {
140     return irq_num + ((pci_dev->devfn >> 3) << 2);
141 }
142 
143 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
144 {
145     xen_set_pci_intx_level(xen_domid, 0, 0, irq_num >> 2,
146                            irq_num & 3, level);
147 }
148 
149 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
150 {
151     int i;
152 
153     /* Scan for updates to PCI link routes (0x60-0x63). */
154     for (i = 0; i < len; i++) {
155         uint8_t v = (val >> (8 * i)) & 0xff;
156         if (v & 0x80) {
157             v = 0;
158         }
159         v &= 0xf;
160         if (((address + i) >= PIIX_PIRQCA) && ((address + i) <= PIIX_PIRQCD)) {
161             xen_set_pci_link_route(xen_domid, address + i - PIIX_PIRQCA, v);
162         }
163     }
164 }
165 
166 int xen_is_pirq_msi(uint32_t msi_data)
167 {
168     /* If vector is 0, the msi is remapped into a pirq, passed as
169      * dest_id.
170      */
171     return ((msi_data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT) == 0;
172 }
173 
174 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
175 {
176     xen_inject_msi(xen_domid, addr, data);
177 }
178 
179 static void xen_suspend_notifier(Notifier *notifier, void *data)
180 {
181     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
182 }
183 
184 /* Xen Interrupt Controller */
185 
186 static void xen_set_irq(void *opaque, int irq, int level)
187 {
188     xen_set_isa_irq_level(xen_domid, irq, level);
189 }
190 
191 qemu_irq *xen_interrupt_controller_init(void)
192 {
193     return qemu_allocate_irqs(xen_set_irq, NULL, 16);
194 }
195 
196 /* Memory Ops */
197 
198 static void xen_ram_init(PCMachineState *pcms,
199                          ram_addr_t ram_size, MemoryRegion **ram_memory_p)
200 {
201     X86MachineState *x86ms = X86_MACHINE(pcms);
202     MemoryRegion *sysmem = get_system_memory();
203     ram_addr_t block_len;
204     uint64_t user_lowmem =
205         object_property_get_uint(qdev_get_machine(),
206                                  X86_MACHINE_MAX_RAM_BELOW_4G,
207                                  &error_abort);
208 
209     /* Handle the machine opt max-ram-below-4g.  It is basically doing
210      * min(xen limit, user limit).
211      */
212     if (!user_lowmem) {
213         user_lowmem = HVM_BELOW_4G_RAM_END; /* default */
214     }
215     if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
216         user_lowmem = HVM_BELOW_4G_RAM_END;
217     }
218 
219     if (ram_size >= user_lowmem) {
220         x86ms->above_4g_mem_size = ram_size - user_lowmem;
221         x86ms->below_4g_mem_size = user_lowmem;
222     } else {
223         x86ms->above_4g_mem_size = 0;
224         x86ms->below_4g_mem_size = ram_size;
225     }
226     if (!x86ms->above_4g_mem_size) {
227         block_len = ram_size;
228     } else {
229         /*
230          * Xen does not allocate the memory continuously, it keeps a
231          * hole of the size computed above or passed in.
232          */
233         block_len = (1ULL << 32) + x86ms->above_4g_mem_size;
234     }
235     memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
236                            &error_fatal);
237     *ram_memory_p = &ram_memory;
238 
239     memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
240                              &ram_memory, 0, 0xa0000);
241     memory_region_add_subregion(sysmem, 0, &ram_640k);
242     /* Skip of the VGA IO memory space, it will be registered later by the VGA
243      * emulated device.
244      *
245      * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
246      * the Options ROM, so it is registered here as RAM.
247      */
248     memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
249                              &ram_memory, 0xc0000,
250                              x86ms->below_4g_mem_size - 0xc0000);
251     memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
252     if (x86ms->above_4g_mem_size > 0) {
253         memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
254                                  &ram_memory, 0x100000000ULL,
255                                  x86ms->above_4g_mem_size);
256         memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
257     }
258 }
259 
260 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr,
261                    Error **errp)
262 {
263     unsigned long nr_pfn;
264     xen_pfn_t *pfn_list;
265     int i;
266 
267     if (runstate_check(RUN_STATE_INMIGRATE)) {
268         /* RAM already populated in Xen */
269         fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
270                 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
271                 __func__, size, ram_addr);
272         return;
273     }
274 
275     if (mr == &ram_memory) {
276         return;
277     }
278 
279     trace_xen_ram_alloc(ram_addr, size);
280 
281     nr_pfn = size >> TARGET_PAGE_BITS;
282     pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
283 
284     for (i = 0; i < nr_pfn; i++) {
285         pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
286     }
287 
288     if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
289         error_setg(errp, "xen: failed to populate ram at " RAM_ADDR_FMT,
290                    ram_addr);
291     }
292 
293     g_free(pfn_list);
294 }
295 
296 static XenPhysmap *get_physmapping(hwaddr start_addr, ram_addr_t size)
297 {
298     XenPhysmap *physmap = NULL;
299 
300     start_addr &= TARGET_PAGE_MASK;
301 
302     QLIST_FOREACH(physmap, &xen_physmap, list) {
303         if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
304             return physmap;
305         }
306     }
307     return NULL;
308 }
309 
310 static hwaddr xen_phys_offset_to_gaddr(hwaddr phys_offset, ram_addr_t size)
311 {
312     hwaddr addr = phys_offset & TARGET_PAGE_MASK;
313     XenPhysmap *physmap = NULL;
314 
315     QLIST_FOREACH(physmap, &xen_physmap, list) {
316         if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
317             return physmap->start_addr + (phys_offset - physmap->phys_offset);
318         }
319     }
320 
321     return phys_offset;
322 }
323 
324 #ifdef XEN_COMPAT_PHYSMAP
325 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
326 {
327     char path[80], value[17];
328 
329     snprintf(path, sizeof(path),
330             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
331             xen_domid, (uint64_t)physmap->phys_offset);
332     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->start_addr);
333     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
334         return -1;
335     }
336     snprintf(path, sizeof(path),
337             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
338             xen_domid, (uint64_t)physmap->phys_offset);
339     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->size);
340     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
341         return -1;
342     }
343     if (physmap->name) {
344         snprintf(path, sizeof(path),
345                 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
346                 xen_domid, (uint64_t)physmap->phys_offset);
347         if (!xs_write(state->xenstore, 0, path,
348                       physmap->name, strlen(physmap->name))) {
349             return -1;
350         }
351     }
352     return 0;
353 }
354 #else
355 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
356 {
357     return 0;
358 }
359 #endif
360 
361 static int xen_add_to_physmap(XenIOState *state,
362                               hwaddr start_addr,
363                               ram_addr_t size,
364                               MemoryRegion *mr,
365                               hwaddr offset_within_region)
366 {
367     unsigned long nr_pages;
368     int rc = 0;
369     XenPhysmap *physmap = NULL;
370     hwaddr pfn, start_gpfn;
371     hwaddr phys_offset = memory_region_get_ram_addr(mr);
372     const char *mr_name;
373 
374     if (get_physmapping(start_addr, size)) {
375         return 0;
376     }
377     if (size <= 0) {
378         return -1;
379     }
380 
381     /* Xen can only handle a single dirty log region for now and we want
382      * the linear framebuffer to be that region.
383      * Avoid tracking any regions that is not videoram and avoid tracking
384      * the legacy vga region. */
385     if (mr == framebuffer && start_addr > 0xbffff) {
386         goto go_physmap;
387     }
388     return -1;
389 
390 go_physmap:
391     DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
392             start_addr, start_addr + size);
393 
394     mr_name = memory_region_name(mr);
395 
396     physmap = g_malloc(sizeof(XenPhysmap));
397 
398     physmap->start_addr = start_addr;
399     physmap->size = size;
400     physmap->name = mr_name;
401     physmap->phys_offset = phys_offset;
402 
403     QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
404 
405     if (runstate_check(RUN_STATE_INMIGRATE)) {
406         /* Now when we have a physmap entry we can replace a dummy mapping with
407          * a real one of guest foreign memory. */
408         uint8_t *p = xen_replace_cache_entry(phys_offset, start_addr, size);
409         assert(p && p == memory_region_get_ram_ptr(mr));
410 
411         return 0;
412     }
413 
414     pfn = phys_offset >> TARGET_PAGE_BITS;
415     start_gpfn = start_addr >> TARGET_PAGE_BITS;
416     nr_pages = size >> TARGET_PAGE_BITS;
417     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, nr_pages, pfn,
418                                         start_gpfn);
419     if (rc) {
420         int saved_errno = errno;
421 
422         error_report("relocate_memory %lu pages from GFN %"HWADDR_PRIx
423                      " to GFN %"HWADDR_PRIx" failed: %s",
424                      nr_pages, pfn, start_gpfn, strerror(saved_errno));
425         errno = saved_errno;
426         return -1;
427     }
428 
429     rc = xendevicemodel_pin_memory_cacheattr(xen_dmod, xen_domid,
430                                    start_addr >> TARGET_PAGE_BITS,
431                                    (start_addr + size - 1) >> TARGET_PAGE_BITS,
432                                    XEN_DOMCTL_MEM_CACHEATTR_WB);
433     if (rc) {
434         error_report("pin_memory_cacheattr failed: %s", strerror(errno));
435     }
436     return xen_save_physmap(state, physmap);
437 }
438 
439 static int xen_remove_from_physmap(XenIOState *state,
440                                    hwaddr start_addr,
441                                    ram_addr_t size)
442 {
443     int rc = 0;
444     XenPhysmap *physmap = NULL;
445     hwaddr phys_offset = 0;
446 
447     physmap = get_physmapping(start_addr, size);
448     if (physmap == NULL) {
449         return -1;
450     }
451 
452     phys_offset = physmap->phys_offset;
453     size = physmap->size;
454 
455     DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
456             "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
457 
458     size >>= TARGET_PAGE_BITS;
459     start_addr >>= TARGET_PAGE_BITS;
460     phys_offset >>= TARGET_PAGE_BITS;
461     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, size, start_addr,
462                                         phys_offset);
463     if (rc) {
464         int saved_errno = errno;
465 
466         error_report("relocate_memory "RAM_ADDR_FMT" pages"
467                      " from GFN %"HWADDR_PRIx
468                      " to GFN %"HWADDR_PRIx" failed: %s",
469                      size, start_addr, phys_offset, strerror(saved_errno));
470         errno = saved_errno;
471         return -1;
472     }
473 
474     QLIST_REMOVE(physmap, list);
475     if (state->log_for_dirtybit == physmap) {
476         state->log_for_dirtybit = NULL;
477         g_free(state->dirty_bitmap);
478         state->dirty_bitmap = NULL;
479     }
480     g_free(physmap);
481 
482     return 0;
483 }
484 
485 static void xen_set_memory(struct MemoryListener *listener,
486                            MemoryRegionSection *section,
487                            bool add)
488 {
489     XenIOState *state = container_of(listener, XenIOState, memory_listener);
490     hwaddr start_addr = section->offset_within_address_space;
491     ram_addr_t size = int128_get64(section->size);
492     bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA);
493     hvmmem_type_t mem_type;
494 
495     if (section->mr == &ram_memory) {
496         return;
497     } else {
498         if (add) {
499             xen_map_memory_section(xen_domid, state->ioservid,
500                                    section);
501         } else {
502             xen_unmap_memory_section(xen_domid, state->ioservid,
503                                      section);
504         }
505     }
506 
507     if (!memory_region_is_ram(section->mr)) {
508         return;
509     }
510 
511     if (log_dirty != add) {
512         return;
513     }
514 
515     trace_xen_client_set_memory(start_addr, size, log_dirty);
516 
517     start_addr &= TARGET_PAGE_MASK;
518     size = TARGET_PAGE_ALIGN(size);
519 
520     if (add) {
521         if (!memory_region_is_rom(section->mr)) {
522             xen_add_to_physmap(state, start_addr, size,
523                                section->mr, section->offset_within_region);
524         } else {
525             mem_type = HVMMEM_ram_ro;
526             if (xen_set_mem_type(xen_domid, mem_type,
527                                  start_addr >> TARGET_PAGE_BITS,
528                                  size >> TARGET_PAGE_BITS)) {
529                 DPRINTF("xen_set_mem_type error, addr: "TARGET_FMT_plx"\n",
530                         start_addr);
531             }
532         }
533     } else {
534         if (xen_remove_from_physmap(state, start_addr, size) < 0) {
535             DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
536         }
537     }
538 }
539 
540 static void xen_region_add(MemoryListener *listener,
541                            MemoryRegionSection *section)
542 {
543     memory_region_ref(section->mr);
544     xen_set_memory(listener, section, true);
545 }
546 
547 static void xen_region_del(MemoryListener *listener,
548                            MemoryRegionSection *section)
549 {
550     xen_set_memory(listener, section, false);
551     memory_region_unref(section->mr);
552 }
553 
554 static void xen_io_add(MemoryListener *listener,
555                        MemoryRegionSection *section)
556 {
557     XenIOState *state = container_of(listener, XenIOState, io_listener);
558     MemoryRegion *mr = section->mr;
559 
560     if (mr->ops == &unassigned_io_ops) {
561         return;
562     }
563 
564     memory_region_ref(mr);
565 
566     xen_map_io_section(xen_domid, state->ioservid, section);
567 }
568 
569 static void xen_io_del(MemoryListener *listener,
570                        MemoryRegionSection *section)
571 {
572     XenIOState *state = container_of(listener, XenIOState, io_listener);
573     MemoryRegion *mr = section->mr;
574 
575     if (mr->ops == &unassigned_io_ops) {
576         return;
577     }
578 
579     xen_unmap_io_section(xen_domid, state->ioservid, section);
580 
581     memory_region_unref(mr);
582 }
583 
584 static void xen_device_realize(DeviceListener *listener,
585                                DeviceState *dev)
586 {
587     XenIOState *state = container_of(listener, XenIOState, device_listener);
588 
589     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
590         PCIDevice *pci_dev = PCI_DEVICE(dev);
591         XenPciDevice *xendev = g_new(XenPciDevice, 1);
592 
593         xendev->pci_dev = pci_dev;
594         xendev->sbdf = PCI_BUILD_BDF(pci_dev_bus_num(pci_dev),
595                                      pci_dev->devfn);
596         QLIST_INSERT_HEAD(&state->dev_list, xendev, entry);
597 
598         xen_map_pcidev(xen_domid, state->ioservid, pci_dev);
599     }
600 }
601 
602 static void xen_device_unrealize(DeviceListener *listener,
603                                  DeviceState *dev)
604 {
605     XenIOState *state = container_of(listener, XenIOState, device_listener);
606 
607     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
608         PCIDevice *pci_dev = PCI_DEVICE(dev);
609         XenPciDevice *xendev, *next;
610 
611         xen_unmap_pcidev(xen_domid, state->ioservid, pci_dev);
612 
613         QLIST_FOREACH_SAFE(xendev, &state->dev_list, entry, next) {
614             if (xendev->pci_dev == pci_dev) {
615                 QLIST_REMOVE(xendev, entry);
616                 g_free(xendev);
617                 break;
618             }
619         }
620     }
621 }
622 
623 static void xen_sync_dirty_bitmap(XenIOState *state,
624                                   hwaddr start_addr,
625                                   ram_addr_t size)
626 {
627     hwaddr npages = size >> TARGET_PAGE_BITS;
628     const int width = sizeof(unsigned long) * 8;
629     size_t bitmap_size = DIV_ROUND_UP(npages, width);
630     int rc, i, j;
631     const XenPhysmap *physmap = NULL;
632 
633     physmap = get_physmapping(start_addr, size);
634     if (physmap == NULL) {
635         /* not handled */
636         return;
637     }
638 
639     if (state->log_for_dirtybit == NULL) {
640         state->log_for_dirtybit = physmap;
641         state->dirty_bitmap = g_new(unsigned long, bitmap_size);
642     } else if (state->log_for_dirtybit != physmap) {
643         /* Only one range for dirty bitmap can be tracked. */
644         return;
645     }
646 
647     rc = xen_track_dirty_vram(xen_domid, start_addr >> TARGET_PAGE_BITS,
648                               npages, state->dirty_bitmap);
649     if (rc < 0) {
650 #ifndef ENODATA
651 #define ENODATA  ENOENT
652 #endif
653         if (errno == ENODATA) {
654             memory_region_set_dirty(framebuffer, 0, size);
655             DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
656                     ", 0x" TARGET_FMT_plx "): %s\n",
657                     start_addr, start_addr + size, strerror(errno));
658         }
659         return;
660     }
661 
662     for (i = 0; i < bitmap_size; i++) {
663         unsigned long map = state->dirty_bitmap[i];
664         while (map != 0) {
665             j = ctzl(map);
666             map &= ~(1ul << j);
667             memory_region_set_dirty(framebuffer,
668                                     (i * width + j) * TARGET_PAGE_SIZE,
669                                     TARGET_PAGE_SIZE);
670         };
671     }
672 }
673 
674 static void xen_log_start(MemoryListener *listener,
675                           MemoryRegionSection *section,
676                           int old, int new)
677 {
678     XenIOState *state = container_of(listener, XenIOState, memory_listener);
679 
680     if (new & ~old & (1 << DIRTY_MEMORY_VGA)) {
681         xen_sync_dirty_bitmap(state, section->offset_within_address_space,
682                               int128_get64(section->size));
683     }
684 }
685 
686 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section,
687                          int old, int new)
688 {
689     XenIOState *state = container_of(listener, XenIOState, memory_listener);
690 
691     if (old & ~new & (1 << DIRTY_MEMORY_VGA)) {
692         state->log_for_dirtybit = NULL;
693         g_free(state->dirty_bitmap);
694         state->dirty_bitmap = NULL;
695         /* Disable dirty bit tracking */
696         xen_track_dirty_vram(xen_domid, 0, 0, NULL);
697     }
698 }
699 
700 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
701 {
702     XenIOState *state = container_of(listener, XenIOState, memory_listener);
703 
704     xen_sync_dirty_bitmap(state, section->offset_within_address_space,
705                           int128_get64(section->size));
706 }
707 
708 static void xen_log_global_start(MemoryListener *listener)
709 {
710     if (xen_enabled()) {
711         xen_in_migration = true;
712     }
713 }
714 
715 static void xen_log_global_stop(MemoryListener *listener)
716 {
717     xen_in_migration = false;
718 }
719 
720 static MemoryListener xen_memory_listener = {
721     .region_add = xen_region_add,
722     .region_del = xen_region_del,
723     .log_start = xen_log_start,
724     .log_stop = xen_log_stop,
725     .log_sync = xen_log_sync,
726     .log_global_start = xen_log_global_start,
727     .log_global_stop = xen_log_global_stop,
728     .priority = 10,
729 };
730 
731 static MemoryListener xen_io_listener = {
732     .region_add = xen_io_add,
733     .region_del = xen_io_del,
734     .priority = 10,
735 };
736 
737 static DeviceListener xen_device_listener = {
738     .realize = xen_device_realize,
739     .unrealize = xen_device_unrealize,
740 };
741 
742 /* get the ioreq packets from share mem */
743 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
744 {
745     ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
746 
747     if (req->state != STATE_IOREQ_READY) {
748         DPRINTF("I/O request not ready: "
749                 "%x, ptr: %x, port: %"PRIx64", "
750                 "data: %"PRIx64", count: %u, size: %u\n",
751                 req->state, req->data_is_ptr, req->addr,
752                 req->data, req->count, req->size);
753         return NULL;
754     }
755 
756     xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
757 
758     req->state = STATE_IOREQ_INPROCESS;
759     return req;
760 }
761 
762 /* use poll to get the port notification */
763 /* ioreq_vec--out,the */
764 /* retval--the number of ioreq packet */
765 static ioreq_t *cpu_get_ioreq(XenIOState *state)
766 {
767     MachineState *ms = MACHINE(qdev_get_machine());
768     unsigned int max_cpus = ms->smp.max_cpus;
769     int i;
770     evtchn_port_t port;
771 
772     port = xenevtchn_pending(state->xce_handle);
773     if (port == state->bufioreq_local_port) {
774         timer_mod(state->buffered_io_timer,
775                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
776         return NULL;
777     }
778 
779     if (port != -1) {
780         for (i = 0; i < max_cpus; i++) {
781             if (state->ioreq_local_port[i] == port) {
782                 break;
783             }
784         }
785 
786         if (i == max_cpus) {
787             hw_error("Fatal error while trying to get io event!\n");
788         }
789 
790         /* unmask the wanted port again */
791         xenevtchn_unmask(state->xce_handle, port);
792 
793         /* get the io packet from shared memory */
794         state->send_vcpu = i;
795         return cpu_get_ioreq_from_shared_memory(state, i);
796     }
797 
798     /* read error or read nothing */
799     return NULL;
800 }
801 
802 static uint32_t do_inp(uint32_t addr, unsigned long size)
803 {
804     switch (size) {
805         case 1:
806             return cpu_inb(addr);
807         case 2:
808             return cpu_inw(addr);
809         case 4:
810             return cpu_inl(addr);
811         default:
812             hw_error("inp: bad size: %04x %lx", addr, size);
813     }
814 }
815 
816 static void do_outp(uint32_t addr,
817         unsigned long size, uint32_t val)
818 {
819     switch (size) {
820         case 1:
821             return cpu_outb(addr, val);
822         case 2:
823             return cpu_outw(addr, val);
824         case 4:
825             return cpu_outl(addr, val);
826         default:
827             hw_error("outp: bad size: %04x %lx", addr, size);
828     }
829 }
830 
831 /*
832  * Helper functions which read/write an object from/to physical guest
833  * memory, as part of the implementation of an ioreq.
834  *
835  * Equivalent to
836  *   cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
837  *                          val, req->size, 0/1)
838  * except without the integer overflow problems.
839  */
840 static void rw_phys_req_item(hwaddr addr,
841                              ioreq_t *req, uint32_t i, void *val, int rw)
842 {
843     /* Do everything unsigned so overflow just results in a truncated result
844      * and accesses to undesired parts of guest memory, which is up
845      * to the guest */
846     hwaddr offset = (hwaddr)req->size * i;
847     if (req->df) {
848         addr -= offset;
849     } else {
850         addr += offset;
851     }
852     cpu_physical_memory_rw(addr, val, req->size, rw);
853 }
854 
855 static inline void read_phys_req_item(hwaddr addr,
856                                       ioreq_t *req, uint32_t i, void *val)
857 {
858     rw_phys_req_item(addr, req, i, val, 0);
859 }
860 static inline void write_phys_req_item(hwaddr addr,
861                                        ioreq_t *req, uint32_t i, void *val)
862 {
863     rw_phys_req_item(addr, req, i, val, 1);
864 }
865 
866 
867 static void cpu_ioreq_pio(ioreq_t *req)
868 {
869     uint32_t i;
870 
871     trace_cpu_ioreq_pio(req, req->dir, req->df, req->data_is_ptr, req->addr,
872                          req->data, req->count, req->size);
873 
874     if (req->size > sizeof(uint32_t)) {
875         hw_error("PIO: bad size (%u)", req->size);
876     }
877 
878     if (req->dir == IOREQ_READ) {
879         if (!req->data_is_ptr) {
880             req->data = do_inp(req->addr, req->size);
881             trace_cpu_ioreq_pio_read_reg(req, req->data, req->addr,
882                                          req->size);
883         } else {
884             uint32_t tmp;
885 
886             for (i = 0; i < req->count; i++) {
887                 tmp = do_inp(req->addr, req->size);
888                 write_phys_req_item(req->data, req, i, &tmp);
889             }
890         }
891     } else if (req->dir == IOREQ_WRITE) {
892         if (!req->data_is_ptr) {
893             trace_cpu_ioreq_pio_write_reg(req, req->data, req->addr,
894                                           req->size);
895             do_outp(req->addr, req->size, req->data);
896         } else {
897             for (i = 0; i < req->count; i++) {
898                 uint32_t tmp = 0;
899 
900                 read_phys_req_item(req->data, req, i, &tmp);
901                 do_outp(req->addr, req->size, tmp);
902             }
903         }
904     }
905 }
906 
907 static void cpu_ioreq_move(ioreq_t *req)
908 {
909     uint32_t i;
910 
911     trace_cpu_ioreq_move(req, req->dir, req->df, req->data_is_ptr, req->addr,
912                          req->data, req->count, req->size);
913 
914     if (req->size > sizeof(req->data)) {
915         hw_error("MMIO: bad size (%u)", req->size);
916     }
917 
918     if (!req->data_is_ptr) {
919         if (req->dir == IOREQ_READ) {
920             for (i = 0; i < req->count; i++) {
921                 read_phys_req_item(req->addr, req, i, &req->data);
922             }
923         } else if (req->dir == IOREQ_WRITE) {
924             for (i = 0; i < req->count; i++) {
925                 write_phys_req_item(req->addr, req, i, &req->data);
926             }
927         }
928     } else {
929         uint64_t tmp;
930 
931         if (req->dir == IOREQ_READ) {
932             for (i = 0; i < req->count; i++) {
933                 read_phys_req_item(req->addr, req, i, &tmp);
934                 write_phys_req_item(req->data, req, i, &tmp);
935             }
936         } else if (req->dir == IOREQ_WRITE) {
937             for (i = 0; i < req->count; i++) {
938                 read_phys_req_item(req->data, req, i, &tmp);
939                 write_phys_req_item(req->addr, req, i, &tmp);
940             }
941         }
942     }
943 }
944 
945 static void cpu_ioreq_config(XenIOState *state, ioreq_t *req)
946 {
947     uint32_t sbdf = req->addr >> 32;
948     uint32_t reg = req->addr;
949     XenPciDevice *xendev;
950 
951     if (req->size != sizeof(uint8_t) && req->size != sizeof(uint16_t) &&
952         req->size != sizeof(uint32_t)) {
953         hw_error("PCI config access: bad size (%u)", req->size);
954     }
955 
956     if (req->count != 1) {
957         hw_error("PCI config access: bad count (%u)", req->count);
958     }
959 
960     QLIST_FOREACH(xendev, &state->dev_list, entry) {
961         if (xendev->sbdf != sbdf) {
962             continue;
963         }
964 
965         if (!req->data_is_ptr) {
966             if (req->dir == IOREQ_READ) {
967                 req->data = pci_host_config_read_common(
968                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
969                     req->size);
970                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
971                                             req->size, req->data);
972             } else if (req->dir == IOREQ_WRITE) {
973                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
974                                              req->size, req->data);
975                 pci_host_config_write_common(
976                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
977                     req->data, req->size);
978             }
979         } else {
980             uint32_t tmp;
981 
982             if (req->dir == IOREQ_READ) {
983                 tmp = pci_host_config_read_common(
984                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
985                     req->size);
986                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
987                                             req->size, tmp);
988                 write_phys_req_item(req->data, req, 0, &tmp);
989             } else if (req->dir == IOREQ_WRITE) {
990                 read_phys_req_item(req->data, req, 0, &tmp);
991                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
992                                              req->size, tmp);
993                 pci_host_config_write_common(
994                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
995                     tmp, req->size);
996             }
997         }
998     }
999 }
1000 
1001 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
1002 {
1003     X86CPU *cpu;
1004     CPUX86State *env;
1005 
1006     cpu = X86_CPU(current_cpu);
1007     env = &cpu->env;
1008     env->regs[R_EAX] = req->data;
1009     env->regs[R_EBX] = vmport_regs->ebx;
1010     env->regs[R_ECX] = vmport_regs->ecx;
1011     env->regs[R_EDX] = vmport_regs->edx;
1012     env->regs[R_ESI] = vmport_regs->esi;
1013     env->regs[R_EDI] = vmport_regs->edi;
1014 }
1015 
1016 static void regs_from_cpu(vmware_regs_t *vmport_regs)
1017 {
1018     X86CPU *cpu = X86_CPU(current_cpu);
1019     CPUX86State *env = &cpu->env;
1020 
1021     vmport_regs->ebx = env->regs[R_EBX];
1022     vmport_regs->ecx = env->regs[R_ECX];
1023     vmport_regs->edx = env->regs[R_EDX];
1024     vmport_regs->esi = env->regs[R_ESI];
1025     vmport_regs->edi = env->regs[R_EDI];
1026 }
1027 
1028 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
1029 {
1030     vmware_regs_t *vmport_regs;
1031 
1032     assert(state->shared_vmport_page);
1033     vmport_regs =
1034         &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
1035     QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
1036 
1037     current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
1038     regs_to_cpu(vmport_regs, req);
1039     cpu_ioreq_pio(req);
1040     regs_from_cpu(vmport_regs);
1041     current_cpu = NULL;
1042 }
1043 
1044 static void handle_ioreq(XenIOState *state, ioreq_t *req)
1045 {
1046     trace_handle_ioreq(req, req->type, req->dir, req->df, req->data_is_ptr,
1047                        req->addr, req->data, req->count, req->size);
1048 
1049     if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
1050             (req->size < sizeof (target_ulong))) {
1051         req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
1052     }
1053 
1054     if (req->dir == IOREQ_WRITE)
1055         trace_handle_ioreq_write(req, req->type, req->df, req->data_is_ptr,
1056                                  req->addr, req->data, req->count, req->size);
1057 
1058     switch (req->type) {
1059         case IOREQ_TYPE_PIO:
1060             cpu_ioreq_pio(req);
1061             break;
1062         case IOREQ_TYPE_COPY:
1063             cpu_ioreq_move(req);
1064             break;
1065         case IOREQ_TYPE_VMWARE_PORT:
1066             handle_vmport_ioreq(state, req);
1067             break;
1068         case IOREQ_TYPE_TIMEOFFSET:
1069             break;
1070         case IOREQ_TYPE_INVALIDATE:
1071             xen_invalidate_map_cache();
1072             break;
1073         case IOREQ_TYPE_PCI_CONFIG:
1074             cpu_ioreq_config(state, req);
1075             break;
1076         default:
1077             hw_error("Invalid ioreq type 0x%x\n", req->type);
1078     }
1079     if (req->dir == IOREQ_READ) {
1080         trace_handle_ioreq_read(req, req->type, req->df, req->data_is_ptr,
1081                                 req->addr, req->data, req->count, req->size);
1082     }
1083 }
1084 
1085 static int handle_buffered_iopage(XenIOState *state)
1086 {
1087     buffered_iopage_t *buf_page = state->buffered_io_page;
1088     buf_ioreq_t *buf_req = NULL;
1089     ioreq_t req;
1090     int qw;
1091 
1092     if (!buf_page) {
1093         return 0;
1094     }
1095 
1096     memset(&req, 0x00, sizeof(req));
1097     req.state = STATE_IOREQ_READY;
1098     req.count = 1;
1099     req.dir = IOREQ_WRITE;
1100 
1101     for (;;) {
1102         uint32_t rdptr = buf_page->read_pointer, wrptr;
1103 
1104         xen_rmb();
1105         wrptr = buf_page->write_pointer;
1106         xen_rmb();
1107         if (rdptr != buf_page->read_pointer) {
1108             continue;
1109         }
1110         if (rdptr == wrptr) {
1111             break;
1112         }
1113         buf_req = &buf_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
1114         req.size = 1U << buf_req->size;
1115         req.addr = buf_req->addr;
1116         req.data = buf_req->data;
1117         req.type = buf_req->type;
1118         xen_rmb();
1119         qw = (req.size == 8);
1120         if (qw) {
1121             if (rdptr + 1 == wrptr) {
1122                 hw_error("Incomplete quad word buffered ioreq");
1123             }
1124             buf_req = &buf_page->buf_ioreq[(rdptr + 1) %
1125                                            IOREQ_BUFFER_SLOT_NUM];
1126             req.data |= ((uint64_t)buf_req->data) << 32;
1127             xen_rmb();
1128         }
1129 
1130         handle_ioreq(state, &req);
1131 
1132         /* Only req.data may get updated by handle_ioreq(), albeit even that
1133          * should not happen as such data would never make it to the guest (we
1134          * can only usefully see writes here after all).
1135          */
1136         assert(req.state == STATE_IOREQ_READY);
1137         assert(req.count == 1);
1138         assert(req.dir == IOREQ_WRITE);
1139         assert(!req.data_is_ptr);
1140 
1141         atomic_add(&buf_page->read_pointer, qw + 1);
1142     }
1143 
1144     return req.count;
1145 }
1146 
1147 static void handle_buffered_io(void *opaque)
1148 {
1149     XenIOState *state = opaque;
1150 
1151     if (handle_buffered_iopage(state)) {
1152         timer_mod(state->buffered_io_timer,
1153                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
1154     } else {
1155         timer_del(state->buffered_io_timer);
1156         xenevtchn_unmask(state->xce_handle, state->bufioreq_local_port);
1157     }
1158 }
1159 
1160 static void cpu_handle_ioreq(void *opaque)
1161 {
1162     XenIOState *state = opaque;
1163     ioreq_t *req = cpu_get_ioreq(state);
1164 
1165     handle_buffered_iopage(state);
1166     if (req) {
1167         ioreq_t copy = *req;
1168 
1169         xen_rmb();
1170         handle_ioreq(state, &copy);
1171         req->data = copy.data;
1172 
1173         if (req->state != STATE_IOREQ_INPROCESS) {
1174             fprintf(stderr, "Badness in I/O request ... not in service?!: "
1175                     "%x, ptr: %x, port: %"PRIx64", "
1176                     "data: %"PRIx64", count: %u, size: %u, type: %u\n",
1177                     req->state, req->data_is_ptr, req->addr,
1178                     req->data, req->count, req->size, req->type);
1179             destroy_hvm_domain(false);
1180             return;
1181         }
1182 
1183         xen_wmb(); /* Update ioreq contents /then/ update state. */
1184 
1185         /*
1186          * We do this before we send the response so that the tools
1187          * have the opportunity to pick up on the reset before the
1188          * guest resumes and does a hlt with interrupts disabled which
1189          * causes Xen to powerdown the domain.
1190          */
1191         if (runstate_is_running()) {
1192             ShutdownCause request;
1193 
1194             if (qemu_shutdown_requested_get()) {
1195                 destroy_hvm_domain(false);
1196             }
1197             request = qemu_reset_requested_get();
1198             if (request) {
1199                 qemu_system_reset(request);
1200                 destroy_hvm_domain(true);
1201             }
1202         }
1203 
1204         req->state = STATE_IORESP_READY;
1205         xenevtchn_notify(state->xce_handle,
1206                          state->ioreq_local_port[state->send_vcpu]);
1207     }
1208 }
1209 
1210 static void xen_main_loop_prepare(XenIOState *state)
1211 {
1212     int evtchn_fd = -1;
1213 
1214     if (state->xce_handle != NULL) {
1215         evtchn_fd = xenevtchn_fd(state->xce_handle);
1216     }
1217 
1218     state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
1219                                                  state);
1220 
1221     if (evtchn_fd != -1) {
1222         CPUState *cpu_state;
1223 
1224         DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
1225         CPU_FOREACH(cpu_state) {
1226             DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1227                     __func__, cpu_state->cpu_index, cpu_state);
1228             state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
1229         }
1230         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
1231     }
1232 }
1233 
1234 
1235 static void xen_hvm_change_state_handler(void *opaque, int running,
1236                                          RunState rstate)
1237 {
1238     XenIOState *state = opaque;
1239 
1240     if (running) {
1241         xen_main_loop_prepare(state);
1242     }
1243 
1244     xen_set_ioreq_server_state(xen_domid,
1245                                state->ioservid,
1246                                (rstate == RUN_STATE_RUNNING));
1247 }
1248 
1249 static void xen_exit_notifier(Notifier *n, void *data)
1250 {
1251     XenIOState *state = container_of(n, XenIOState, exit);
1252 
1253     xen_destroy_ioreq_server(xen_domid, state->ioservid);
1254 
1255     xenevtchn_close(state->xce_handle);
1256     xs_daemon_close(state->xenstore);
1257 }
1258 
1259 #ifdef XEN_COMPAT_PHYSMAP
1260 static void xen_read_physmap(XenIOState *state)
1261 {
1262     XenPhysmap *physmap = NULL;
1263     unsigned int len, num, i;
1264     char path[80], *value = NULL;
1265     char **entries = NULL;
1266 
1267     snprintf(path, sizeof(path),
1268             "/local/domain/0/device-model/%d/physmap", xen_domid);
1269     entries = xs_directory(state->xenstore, 0, path, &num);
1270     if (entries == NULL)
1271         return;
1272 
1273     for (i = 0; i < num; i++) {
1274         physmap = g_malloc(sizeof (XenPhysmap));
1275         physmap->phys_offset = strtoull(entries[i], NULL, 16);
1276         snprintf(path, sizeof(path),
1277                 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1278                 xen_domid, entries[i]);
1279         value = xs_read(state->xenstore, 0, path, &len);
1280         if (value == NULL) {
1281             g_free(physmap);
1282             continue;
1283         }
1284         physmap->start_addr = strtoull(value, NULL, 16);
1285         free(value);
1286 
1287         snprintf(path, sizeof(path),
1288                 "/local/domain/0/device-model/%d/physmap/%s/size",
1289                 xen_domid, entries[i]);
1290         value = xs_read(state->xenstore, 0, path, &len);
1291         if (value == NULL) {
1292             g_free(physmap);
1293             continue;
1294         }
1295         physmap->size = strtoull(value, NULL, 16);
1296         free(value);
1297 
1298         snprintf(path, sizeof(path),
1299                 "/local/domain/0/device-model/%d/physmap/%s/name",
1300                 xen_domid, entries[i]);
1301         physmap->name = xs_read(state->xenstore, 0, path, &len);
1302 
1303         QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
1304     }
1305     free(entries);
1306 }
1307 #else
1308 static void xen_read_physmap(XenIOState *state)
1309 {
1310 }
1311 #endif
1312 
1313 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1314 {
1315     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1316 }
1317 
1318 static int xen_map_ioreq_server(XenIOState *state)
1319 {
1320     void *addr = NULL;
1321     xenforeignmemory_resource_handle *fres;
1322     xen_pfn_t ioreq_pfn;
1323     xen_pfn_t bufioreq_pfn;
1324     evtchn_port_t bufioreq_evtchn;
1325     int rc;
1326 
1327     /*
1328      * Attempt to map using the resource API and fall back to normal
1329      * foreign mapping if this is not supported.
1330      */
1331     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_bufioreq != 0);
1332     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_ioreq(0) != 1);
1333     fres = xenforeignmemory_map_resource(xen_fmem, xen_domid,
1334                                          XENMEM_resource_ioreq_server,
1335                                          state->ioservid, 0, 2,
1336                                          &addr,
1337                                          PROT_READ | PROT_WRITE, 0);
1338     if (fres != NULL) {
1339         trace_xen_map_resource_ioreq(state->ioservid, addr);
1340         state->buffered_io_page = addr;
1341         state->shared_page = addr + TARGET_PAGE_SIZE;
1342     } else if (errno != EOPNOTSUPP) {
1343         error_report("failed to map ioreq server resources: error %d handle=%p",
1344                      errno, xen_xc);
1345         return -1;
1346     }
1347 
1348     rc = xen_get_ioreq_server_info(xen_domid, state->ioservid,
1349                                    (state->shared_page == NULL) ?
1350                                    &ioreq_pfn : NULL,
1351                                    (state->buffered_io_page == NULL) ?
1352                                    &bufioreq_pfn : NULL,
1353                                    &bufioreq_evtchn);
1354     if (rc < 0) {
1355         error_report("failed to get ioreq server info: error %d handle=%p",
1356                      errno, xen_xc);
1357         return rc;
1358     }
1359 
1360     if (state->shared_page == NULL) {
1361         DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1362 
1363         state->shared_page = xenforeignmemory_map(xen_fmem, xen_domid,
1364                                                   PROT_READ | PROT_WRITE,
1365                                                   1, &ioreq_pfn, NULL);
1366         if (state->shared_page == NULL) {
1367             error_report("map shared IO page returned error %d handle=%p",
1368                          errno, xen_xc);
1369         }
1370     }
1371 
1372     if (state->buffered_io_page == NULL) {
1373         DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
1374 
1375         state->buffered_io_page = xenforeignmemory_map(xen_fmem, xen_domid,
1376                                                        PROT_READ | PROT_WRITE,
1377                                                        1, &bufioreq_pfn,
1378                                                        NULL);
1379         if (state->buffered_io_page == NULL) {
1380             error_report("map buffered IO page returned error %d", errno);
1381             return -1;
1382         }
1383     }
1384 
1385     if (state->shared_page == NULL || state->buffered_io_page == NULL) {
1386         return -1;
1387     }
1388 
1389     DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
1390 
1391     state->bufioreq_remote_port = bufioreq_evtchn;
1392 
1393     return 0;
1394 }
1395 
1396 void xen_hvm_init(PCMachineState *pcms, MemoryRegion **ram_memory)
1397 {
1398     MachineState *ms = MACHINE(pcms);
1399     unsigned int max_cpus = ms->smp.max_cpus;
1400     int i, rc;
1401     xen_pfn_t ioreq_pfn;
1402     XenIOState *state;
1403 
1404     state = g_malloc0(sizeof (XenIOState));
1405 
1406     state->xce_handle = xenevtchn_open(NULL, 0);
1407     if (state->xce_handle == NULL) {
1408         perror("xen: event channel open");
1409         goto err;
1410     }
1411 
1412     state->xenstore = xs_daemon_open();
1413     if (state->xenstore == NULL) {
1414         perror("xen: xenstore open");
1415         goto err;
1416     }
1417 
1418     xen_create_ioreq_server(xen_domid, &state->ioservid);
1419 
1420     state->exit.notify = xen_exit_notifier;
1421     qemu_add_exit_notifier(&state->exit);
1422 
1423     state->suspend.notify = xen_suspend_notifier;
1424     qemu_register_suspend_notifier(&state->suspend);
1425 
1426     state->wakeup.notify = xen_wakeup_notifier;
1427     qemu_register_wakeup_notifier(&state->wakeup);
1428 
1429     /*
1430      * Register wake-up support in QMP query-current-machine API
1431      */
1432     qemu_register_wakeup_support();
1433 
1434     rc = xen_map_ioreq_server(state);
1435     if (rc < 0) {
1436         goto err;
1437     }
1438 
1439     rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1440     if (!rc) {
1441         DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1442         state->shared_vmport_page =
1443             xenforeignmemory_map(xen_fmem, xen_domid, PROT_READ|PROT_WRITE,
1444                                  1, &ioreq_pfn, NULL);
1445         if (state->shared_vmport_page == NULL) {
1446             error_report("map shared vmport IO page returned error %d handle=%p",
1447                          errno, xen_xc);
1448             goto err;
1449         }
1450     } else if (rc != -ENOSYS) {
1451         error_report("get vmport regs pfn returned error %d, rc=%d",
1452                      errno, rc);
1453         goto err;
1454     }
1455 
1456     /* Note: cpus is empty at this point in init */
1457     state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1458 
1459     rc = xen_set_ioreq_server_state(xen_domid, state->ioservid, true);
1460     if (rc < 0) {
1461         error_report("failed to enable ioreq server info: error %d handle=%p",
1462                      errno, xen_xc);
1463         goto err;
1464     }
1465 
1466     state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1467 
1468     /* FIXME: how about if we overflow the page here? */
1469     for (i = 0; i < max_cpus; i++) {
1470         rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1471                                         xen_vcpu_eport(state->shared_page, i));
1472         if (rc == -1) {
1473             error_report("shared evtchn %d bind error %d", i, errno);
1474             goto err;
1475         }
1476         state->ioreq_local_port[i] = rc;
1477     }
1478 
1479     rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1480                                     state->bufioreq_remote_port);
1481     if (rc == -1) {
1482         error_report("buffered evtchn bind error %d", errno);
1483         goto err;
1484     }
1485     state->bufioreq_local_port = rc;
1486 
1487     /* Init RAM management */
1488 #ifdef XEN_COMPAT_PHYSMAP
1489     xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1490 #else
1491     xen_map_cache_init(NULL, state);
1492 #endif
1493     xen_ram_init(pcms, ram_size, ram_memory);
1494 
1495     qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1496 
1497     state->memory_listener = xen_memory_listener;
1498     memory_listener_register(&state->memory_listener, &address_space_memory);
1499     state->log_for_dirtybit = NULL;
1500 
1501     state->io_listener = xen_io_listener;
1502     memory_listener_register(&state->io_listener, &address_space_io);
1503 
1504     state->device_listener = xen_device_listener;
1505     QLIST_INIT(&state->dev_list);
1506     device_listener_register(&state->device_listener);
1507 
1508     xen_bus_init();
1509 
1510     /* Initialize backend core & drivers */
1511     if (xen_be_init() != 0) {
1512         error_report("xen backend core setup failed");
1513         goto err;
1514     }
1515     xen_be_register_common();
1516 
1517     QLIST_INIT(&xen_physmap);
1518     xen_read_physmap(state);
1519 
1520     /* Disable ACPI build because Xen handles it */
1521     pcms->acpi_build_enabled = false;
1522 
1523     return;
1524 
1525 err:
1526     error_report("xen hardware virtual machine initialisation failed");
1527     exit(1);
1528 }
1529 
1530 void destroy_hvm_domain(bool reboot)
1531 {
1532     xc_interface *xc_handle;
1533     int sts;
1534     int rc;
1535 
1536     unsigned int reason = reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff;
1537 
1538     if (xen_dmod) {
1539         rc = xendevicemodel_shutdown(xen_dmod, xen_domid, reason);
1540         if (!rc) {
1541             return;
1542         }
1543         if (errno != ENOTTY /* old Xen */) {
1544             perror("xendevicemodel_shutdown failed");
1545         }
1546         /* well, try the old thing then */
1547     }
1548 
1549     xc_handle = xc_interface_open(0, 0, 0);
1550     if (xc_handle == NULL) {
1551         fprintf(stderr, "Cannot acquire xenctrl handle\n");
1552     } else {
1553         sts = xc_domain_shutdown(xc_handle, xen_domid, reason);
1554         if (sts != 0) {
1555             fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1556                     "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1557                     sts, strerror(errno));
1558         } else {
1559             fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1560                     reboot ? "reboot" : "poweroff");
1561         }
1562         xc_interface_close(xc_handle);
1563     }
1564 }
1565 
1566 void xen_register_framebuffer(MemoryRegion *mr)
1567 {
1568     framebuffer = mr;
1569 }
1570 
1571 void xen_shutdown_fatal_error(const char *fmt, ...)
1572 {
1573     va_list ap;
1574 
1575     va_start(ap, fmt);
1576     vfprintf(stderr, fmt, ap);
1577     va_end(ap);
1578     fprintf(stderr, "Will destroy the domain.\n");
1579     /* destroy the domain */
1580     qemu_system_shutdown_request(SHUTDOWN_CAUSE_HOST_ERROR);
1581 }
1582 
1583 void xen_hvm_modified_memory(ram_addr_t start, ram_addr_t length)
1584 {
1585     if (unlikely(xen_in_migration)) {
1586         int rc;
1587         ram_addr_t start_pfn, nb_pages;
1588 
1589         start = xen_phys_offset_to_gaddr(start, length);
1590 
1591         if (length == 0) {
1592             length = TARGET_PAGE_SIZE;
1593         }
1594         start_pfn = start >> TARGET_PAGE_BITS;
1595         nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1596             - start_pfn;
1597         rc = xen_modified_memory(xen_domid, start_pfn, nb_pages);
1598         if (rc) {
1599             fprintf(stderr,
1600                     "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1601                     __func__, start, nb_pages, errno, strerror(errno));
1602         }
1603     }
1604 }
1605 
1606 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1607 {
1608     if (enable) {
1609         memory_global_dirty_log_start();
1610     } else {
1611         memory_global_dirty_log_stop();
1612     }
1613 }
1614