xref: /openbmc/qemu/hw/i386/xen/xen-hvm.c (revision 02e5d7d7)
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 "qapi/error.h"
26 #include "qapi/qapi-commands-misc.h"
27 #include "qemu/error-report.h"
28 #include "qemu/main-loop.h"
29 #include "qemu/range.h"
30 #include "sysemu/runstate.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/xen.h"
33 #include "sysemu/xen-mapcache.h"
34 #include "trace.h"
35 #include "exec/address-spaces.h"
36 
37 #include <xen/hvm/ioreq.h>
38 #include <xen/hvm/e820.h>
39 
40 //#define DEBUG_XEN_HVM
41 
42 #ifdef DEBUG_XEN_HVM
43 #define DPRINTF(fmt, ...) \
44     do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
45 #else
46 #define DPRINTF(fmt, ...) \
47     do { } while (0)
48 #endif
49 
50 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
51 static MemoryRegion *framebuffer;
52 static bool xen_in_migration;
53 
54 /* Compatibility with older version */
55 
56 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
57  * installed.  This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
58  * needs to be included before this block and hw/xen/xen_common.h needs to
59  * be included before xen/hvm/ioreq.h
60  */
61 #ifndef IOREQ_TYPE_VMWARE_PORT
62 #define IOREQ_TYPE_VMWARE_PORT  3
63 struct vmware_regs {
64     uint32_t esi;
65     uint32_t edi;
66     uint32_t ebx;
67     uint32_t ecx;
68     uint32_t edx;
69 };
70 typedef struct vmware_regs vmware_regs_t;
71 
72 struct shared_vmport_iopage {
73     struct vmware_regs vcpu_vmport_regs[1];
74 };
75 typedef struct shared_vmport_iopage shared_vmport_iopage_t;
76 #endif
77 
78 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
79 {
80     return shared_page->vcpu_ioreq[i].vp_eport;
81 }
82 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
83 {
84     return &shared_page->vcpu_ioreq[vcpu];
85 }
86 
87 #define BUFFER_IO_MAX_DELAY  100
88 
89 typedef struct XenPhysmap {
90     hwaddr start_addr;
91     ram_addr_t size;
92     const char *name;
93     hwaddr phys_offset;
94 
95     QLIST_ENTRY(XenPhysmap) list;
96 } XenPhysmap;
97 
98 static QLIST_HEAD(, XenPhysmap) xen_physmap;
99 
100 typedef struct XenPciDevice {
101     PCIDevice *pci_dev;
102     uint32_t sbdf;
103     QLIST_ENTRY(XenPciDevice) entry;
104 } XenPciDevice;
105 
106 typedef struct XenIOState {
107     ioservid_t ioservid;
108     shared_iopage_t *shared_page;
109     shared_vmport_iopage_t *shared_vmport_page;
110     buffered_iopage_t *buffered_io_page;
111     QEMUTimer *buffered_io_timer;
112     CPUState **cpu_by_vcpu_id;
113     /* the evtchn port for polling the notification, */
114     evtchn_port_t *ioreq_local_port;
115     /* evtchn remote and local ports for buffered io */
116     evtchn_port_t bufioreq_remote_port;
117     evtchn_port_t bufioreq_local_port;
118     /* the evtchn fd for polling */
119     xenevtchn_handle *xce_handle;
120     /* which vcpu we are serving */
121     int send_vcpu;
122 
123     struct xs_handle *xenstore;
124     MemoryListener memory_listener;
125     MemoryListener io_listener;
126     QLIST_HEAD(, XenPciDevice) dev_list;
127     DeviceListener device_listener;
128     hwaddr free_phys_offset;
129     const XenPhysmap *log_for_dirtybit;
130     /* Buffer used by xen_sync_dirty_bitmap */
131     unsigned long *dirty_bitmap;
132 
133     Notifier exit;
134     Notifier suspend;
135     Notifier wakeup;
136 } XenIOState;
137 
138 /* Xen specific function for piix pci */
139 
140 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
141 {
142     return irq_num + ((pci_dev->devfn >> 3) << 2);
143 }
144 
145 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
146 {
147     xen_set_pci_intx_level(xen_domid, 0, 0, irq_num >> 2,
148                            irq_num & 3, level);
149 }
150 
151 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
152 {
153     int i;
154 
155     /* Scan for updates to PCI link routes (0x60-0x63). */
156     for (i = 0; i < len; i++) {
157         uint8_t v = (val >> (8 * i)) & 0xff;
158         if (v & 0x80) {
159             v = 0;
160         }
161         v &= 0xf;
162         if (((address + i) >= PIIX_PIRQCA) && ((address + i) <= PIIX_PIRQCD)) {
163             xen_set_pci_link_route(xen_domid, address + i - PIIX_PIRQCA, v);
164         }
165     }
166 }
167 
168 int xen_is_pirq_msi(uint32_t msi_data)
169 {
170     /* If vector is 0, the msi is remapped into a pirq, passed as
171      * dest_id.
172      */
173     return ((msi_data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT) == 0;
174 }
175 
176 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
177 {
178     xen_inject_msi(xen_domid, addr, data);
179 }
180 
181 static void xen_suspend_notifier(Notifier *notifier, void *data)
182 {
183     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
184 }
185 
186 /* Xen Interrupt Controller */
187 
188 static void xen_set_irq(void *opaque, int irq, int level)
189 {
190     xen_set_isa_irq_level(xen_domid, irq, level);
191 }
192 
193 qemu_irq *xen_interrupt_controller_init(void)
194 {
195     return qemu_allocate_irqs(xen_set_irq, NULL, 16);
196 }
197 
198 /* Memory Ops */
199 
200 static void xen_ram_init(PCMachineState *pcms,
201                          ram_addr_t ram_size, MemoryRegion **ram_memory_p)
202 {
203     X86MachineState *x86ms = X86_MACHINE(pcms);
204     MemoryRegion *sysmem = get_system_memory();
205     ram_addr_t block_len;
206     uint64_t user_lowmem =
207         object_property_get_uint(qdev_get_machine(),
208                                  PC_MACHINE_MAX_RAM_BELOW_4G,
209                                  &error_abort);
210 
211     /* Handle the machine opt max-ram-below-4g.  It is basically doing
212      * min(xen limit, user limit).
213      */
214     if (!user_lowmem) {
215         user_lowmem = HVM_BELOW_4G_RAM_END; /* default */
216     }
217     if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
218         user_lowmem = HVM_BELOW_4G_RAM_END;
219     }
220 
221     if (ram_size >= user_lowmem) {
222         x86ms->above_4g_mem_size = ram_size - user_lowmem;
223         x86ms->below_4g_mem_size = user_lowmem;
224     } else {
225         x86ms->above_4g_mem_size = 0;
226         x86ms->below_4g_mem_size = ram_size;
227     }
228     if (!x86ms->above_4g_mem_size) {
229         block_len = ram_size;
230     } else {
231         /*
232          * Xen does not allocate the memory continuously, it keeps a
233          * hole of the size computed above or passed in.
234          */
235         block_len = (4 * GiB) + x86ms->above_4g_mem_size;
236     }
237     memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
238                            &error_fatal);
239     *ram_memory_p = &ram_memory;
240 
241     memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
242                              &ram_memory, 0, 0xa0000);
243     memory_region_add_subregion(sysmem, 0, &ram_640k);
244     /* Skip of the VGA IO memory space, it will be registered later by the VGA
245      * emulated device.
246      *
247      * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
248      * the Options ROM, so it is registered here as RAM.
249      */
250     memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
251                              &ram_memory, 0xc0000,
252                              x86ms->below_4g_mem_size - 0xc0000);
253     memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
254     if (x86ms->above_4g_mem_size > 0) {
255         memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
256                                  &ram_memory, 0x100000000ULL,
257                                  x86ms->above_4g_mem_size);
258         memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
259     }
260 }
261 
262 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr,
263                    Error **errp)
264 {
265     unsigned long nr_pfn;
266     xen_pfn_t *pfn_list;
267     int i;
268 
269     if (runstate_check(RUN_STATE_INMIGRATE)) {
270         /* RAM already populated in Xen */
271         fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
272                 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
273                 __func__, size, ram_addr);
274         return;
275     }
276 
277     if (mr == &ram_memory) {
278         return;
279     }
280 
281     trace_xen_ram_alloc(ram_addr, size);
282 
283     nr_pfn = size >> TARGET_PAGE_BITS;
284     pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
285 
286     for (i = 0; i < nr_pfn; i++) {
287         pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
288     }
289 
290     if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
291         error_setg(errp, "xen: failed to populate ram at " RAM_ADDR_FMT,
292                    ram_addr);
293     }
294 
295     g_free(pfn_list);
296 }
297 
298 static XenPhysmap *get_physmapping(hwaddr start_addr, ram_addr_t size)
299 {
300     XenPhysmap *physmap = NULL;
301 
302     start_addr &= TARGET_PAGE_MASK;
303 
304     QLIST_FOREACH(physmap, &xen_physmap, list) {
305         if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
306             return physmap;
307         }
308     }
309     return NULL;
310 }
311 
312 static hwaddr xen_phys_offset_to_gaddr(hwaddr phys_offset, ram_addr_t size)
313 {
314     hwaddr addr = phys_offset & TARGET_PAGE_MASK;
315     XenPhysmap *physmap = NULL;
316 
317     QLIST_FOREACH(physmap, &xen_physmap, list) {
318         if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
319             return physmap->start_addr + (phys_offset - physmap->phys_offset);
320         }
321     }
322 
323     return phys_offset;
324 }
325 
326 #ifdef XEN_COMPAT_PHYSMAP
327 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
328 {
329     char path[80], value[17];
330 
331     snprintf(path, sizeof(path),
332             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
333             xen_domid, (uint64_t)physmap->phys_offset);
334     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->start_addr);
335     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
336         return -1;
337     }
338     snprintf(path, sizeof(path),
339             "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
340             xen_domid, (uint64_t)physmap->phys_offset);
341     snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->size);
342     if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
343         return -1;
344     }
345     if (physmap->name) {
346         snprintf(path, sizeof(path),
347                 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
348                 xen_domid, (uint64_t)physmap->phys_offset);
349         if (!xs_write(state->xenstore, 0, path,
350                       physmap->name, strlen(physmap->name))) {
351             return -1;
352         }
353     }
354     return 0;
355 }
356 #else
357 static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap)
358 {
359     return 0;
360 }
361 #endif
362 
363 static int xen_add_to_physmap(XenIOState *state,
364                               hwaddr start_addr,
365                               ram_addr_t size,
366                               MemoryRegion *mr,
367                               hwaddr offset_within_region)
368 {
369     unsigned long nr_pages;
370     int rc = 0;
371     XenPhysmap *physmap = NULL;
372     hwaddr pfn, start_gpfn;
373     hwaddr phys_offset = memory_region_get_ram_addr(mr);
374     const char *mr_name;
375 
376     if (get_physmapping(start_addr, size)) {
377         return 0;
378     }
379     if (size <= 0) {
380         return -1;
381     }
382 
383     /* Xen can only handle a single dirty log region for now and we want
384      * the linear framebuffer to be that region.
385      * Avoid tracking any regions that is not videoram and avoid tracking
386      * the legacy vga region. */
387     if (mr == framebuffer && start_addr > 0xbffff) {
388         goto go_physmap;
389     }
390     return -1;
391 
392 go_physmap:
393     DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
394             start_addr, start_addr + size);
395 
396     mr_name = memory_region_name(mr);
397 
398     physmap = g_malloc(sizeof(XenPhysmap));
399 
400     physmap->start_addr = start_addr;
401     physmap->size = size;
402     physmap->name = mr_name;
403     physmap->phys_offset = phys_offset;
404 
405     QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
406 
407     if (runstate_check(RUN_STATE_INMIGRATE)) {
408         /* Now when we have a physmap entry we can replace a dummy mapping with
409          * a real one of guest foreign memory. */
410         uint8_t *p = xen_replace_cache_entry(phys_offset, start_addr, size);
411         assert(p && p == memory_region_get_ram_ptr(mr));
412 
413         return 0;
414     }
415 
416     pfn = phys_offset >> TARGET_PAGE_BITS;
417     start_gpfn = start_addr >> TARGET_PAGE_BITS;
418     nr_pages = size >> TARGET_PAGE_BITS;
419     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, nr_pages, pfn,
420                                         start_gpfn);
421     if (rc) {
422         int saved_errno = errno;
423 
424         error_report("relocate_memory %lu pages from GFN %"HWADDR_PRIx
425                      " to GFN %"HWADDR_PRIx" failed: %s",
426                      nr_pages, pfn, start_gpfn, strerror(saved_errno));
427         errno = saved_errno;
428         return -1;
429     }
430 
431     rc = xendevicemodel_pin_memory_cacheattr(xen_dmod, xen_domid,
432                                    start_addr >> TARGET_PAGE_BITS,
433                                    (start_addr + size - 1) >> TARGET_PAGE_BITS,
434                                    XEN_DOMCTL_MEM_CACHEATTR_WB);
435     if (rc) {
436         error_report("pin_memory_cacheattr failed: %s", strerror(errno));
437     }
438     return xen_save_physmap(state, physmap);
439 }
440 
441 static int xen_remove_from_physmap(XenIOState *state,
442                                    hwaddr start_addr,
443                                    ram_addr_t size)
444 {
445     int rc = 0;
446     XenPhysmap *physmap = NULL;
447     hwaddr phys_offset = 0;
448 
449     physmap = get_physmapping(start_addr, size);
450     if (physmap == NULL) {
451         return -1;
452     }
453 
454     phys_offset = physmap->phys_offset;
455     size = physmap->size;
456 
457     DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
458             "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
459 
460     size >>= TARGET_PAGE_BITS;
461     start_addr >>= TARGET_PAGE_BITS;
462     phys_offset >>= TARGET_PAGE_BITS;
463     rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, size, start_addr,
464                                         phys_offset);
465     if (rc) {
466         int saved_errno = errno;
467 
468         error_report("relocate_memory "RAM_ADDR_FMT" pages"
469                      " from GFN %"HWADDR_PRIx
470                      " to GFN %"HWADDR_PRIx" failed: %s",
471                      size, start_addr, phys_offset, strerror(saved_errno));
472         errno = saved_errno;
473         return -1;
474     }
475 
476     QLIST_REMOVE(physmap, list);
477     if (state->log_for_dirtybit == physmap) {
478         state->log_for_dirtybit = NULL;
479         g_free(state->dirty_bitmap);
480         state->dirty_bitmap = NULL;
481     }
482     g_free(physmap);
483 
484     return 0;
485 }
486 
487 static void xen_set_memory(struct MemoryListener *listener,
488                            MemoryRegionSection *section,
489                            bool add)
490 {
491     XenIOState *state = container_of(listener, XenIOState, memory_listener);
492     hwaddr start_addr = section->offset_within_address_space;
493     ram_addr_t size = int128_get64(section->size);
494     bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA);
495     hvmmem_type_t mem_type;
496 
497     if (section->mr == &ram_memory) {
498         return;
499     } else {
500         if (add) {
501             xen_map_memory_section(xen_domid, state->ioservid,
502                                    section);
503         } else {
504             xen_unmap_memory_section(xen_domid, state->ioservid,
505                                      section);
506         }
507     }
508 
509     if (!memory_region_is_ram(section->mr)) {
510         return;
511     }
512 
513     if (log_dirty != add) {
514         return;
515     }
516 
517     trace_xen_client_set_memory(start_addr, size, log_dirty);
518 
519     start_addr &= TARGET_PAGE_MASK;
520     size = TARGET_PAGE_ALIGN(size);
521 
522     if (add) {
523         if (!memory_region_is_rom(section->mr)) {
524             xen_add_to_physmap(state, start_addr, size,
525                                section->mr, section->offset_within_region);
526         } else {
527             mem_type = HVMMEM_ram_ro;
528             if (xen_set_mem_type(xen_domid, mem_type,
529                                  start_addr >> TARGET_PAGE_BITS,
530                                  size >> TARGET_PAGE_BITS)) {
531                 DPRINTF("xen_set_mem_type error, addr: "TARGET_FMT_plx"\n",
532                         start_addr);
533             }
534         }
535     } else {
536         if (xen_remove_from_physmap(state, start_addr, size) < 0) {
537             DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
538         }
539     }
540 }
541 
542 static void xen_region_add(MemoryListener *listener,
543                            MemoryRegionSection *section)
544 {
545     memory_region_ref(section->mr);
546     xen_set_memory(listener, section, true);
547 }
548 
549 static void xen_region_del(MemoryListener *listener,
550                            MemoryRegionSection *section)
551 {
552     xen_set_memory(listener, section, false);
553     memory_region_unref(section->mr);
554 }
555 
556 static void xen_io_add(MemoryListener *listener,
557                        MemoryRegionSection *section)
558 {
559     XenIOState *state = container_of(listener, XenIOState, io_listener);
560     MemoryRegion *mr = section->mr;
561 
562     if (mr->ops == &unassigned_io_ops) {
563         return;
564     }
565 
566     memory_region_ref(mr);
567 
568     xen_map_io_section(xen_domid, state->ioservid, section);
569 }
570 
571 static void xen_io_del(MemoryListener *listener,
572                        MemoryRegionSection *section)
573 {
574     XenIOState *state = container_of(listener, XenIOState, io_listener);
575     MemoryRegion *mr = section->mr;
576 
577     if (mr->ops == &unassigned_io_ops) {
578         return;
579     }
580 
581     xen_unmap_io_section(xen_domid, state->ioservid, section);
582 
583     memory_region_unref(mr);
584 }
585 
586 static void xen_device_realize(DeviceListener *listener,
587                                DeviceState *dev)
588 {
589     XenIOState *state = container_of(listener, XenIOState, device_listener);
590 
591     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
592         PCIDevice *pci_dev = PCI_DEVICE(dev);
593         XenPciDevice *xendev = g_new(XenPciDevice, 1);
594 
595         xendev->pci_dev = pci_dev;
596         xendev->sbdf = PCI_BUILD_BDF(pci_dev_bus_num(pci_dev),
597                                      pci_dev->devfn);
598         QLIST_INSERT_HEAD(&state->dev_list, xendev, entry);
599 
600         xen_map_pcidev(xen_domid, state->ioservid, pci_dev);
601     }
602 }
603 
604 static void xen_device_unrealize(DeviceListener *listener,
605                                  DeviceState *dev)
606 {
607     XenIOState *state = container_of(listener, XenIOState, device_listener);
608 
609     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
610         PCIDevice *pci_dev = PCI_DEVICE(dev);
611         XenPciDevice *xendev, *next;
612 
613         xen_unmap_pcidev(xen_domid, state->ioservid, pci_dev);
614 
615         QLIST_FOREACH_SAFE(xendev, &state->dev_list, entry, next) {
616             if (xendev->pci_dev == pci_dev) {
617                 QLIST_REMOVE(xendev, entry);
618                 g_free(xendev);
619                 break;
620             }
621         }
622     }
623 }
624 
625 static void xen_sync_dirty_bitmap(XenIOState *state,
626                                   hwaddr start_addr,
627                                   ram_addr_t size)
628 {
629     hwaddr npages = size >> TARGET_PAGE_BITS;
630     const int width = sizeof(unsigned long) * 8;
631     size_t bitmap_size = DIV_ROUND_UP(npages, width);
632     int rc, i, j;
633     const XenPhysmap *physmap = NULL;
634 
635     physmap = get_physmapping(start_addr, size);
636     if (physmap == NULL) {
637         /* not handled */
638         return;
639     }
640 
641     if (state->log_for_dirtybit == NULL) {
642         state->log_for_dirtybit = physmap;
643         state->dirty_bitmap = g_new(unsigned long, bitmap_size);
644     } else if (state->log_for_dirtybit != physmap) {
645         /* Only one range for dirty bitmap can be tracked. */
646         return;
647     }
648 
649     rc = xen_track_dirty_vram(xen_domid, start_addr >> TARGET_PAGE_BITS,
650                               npages, state->dirty_bitmap);
651     if (rc < 0) {
652 #ifndef ENODATA
653 #define ENODATA  ENOENT
654 #endif
655         if (errno == ENODATA) {
656             memory_region_set_dirty(framebuffer, 0, size);
657             DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
658                     ", 0x" TARGET_FMT_plx "): %s\n",
659                     start_addr, start_addr + size, strerror(errno));
660         }
661         return;
662     }
663 
664     for (i = 0; i < bitmap_size; i++) {
665         unsigned long map = state->dirty_bitmap[i];
666         while (map != 0) {
667             j = ctzl(map);
668             map &= ~(1ul << j);
669             memory_region_set_dirty(framebuffer,
670                                     (i * width + j) * TARGET_PAGE_SIZE,
671                                     TARGET_PAGE_SIZE);
672         };
673     }
674 }
675 
676 static void xen_log_start(MemoryListener *listener,
677                           MemoryRegionSection *section,
678                           int old, int new)
679 {
680     XenIOState *state = container_of(listener, XenIOState, memory_listener);
681 
682     if (new & ~old & (1 << DIRTY_MEMORY_VGA)) {
683         xen_sync_dirty_bitmap(state, section->offset_within_address_space,
684                               int128_get64(section->size));
685     }
686 }
687 
688 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section,
689                          int old, int new)
690 {
691     XenIOState *state = container_of(listener, XenIOState, memory_listener);
692 
693     if (old & ~new & (1 << DIRTY_MEMORY_VGA)) {
694         state->log_for_dirtybit = NULL;
695         g_free(state->dirty_bitmap);
696         state->dirty_bitmap = NULL;
697         /* Disable dirty bit tracking */
698         xen_track_dirty_vram(xen_domid, 0, 0, NULL);
699     }
700 }
701 
702 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
703 {
704     XenIOState *state = container_of(listener, XenIOState, memory_listener);
705 
706     xen_sync_dirty_bitmap(state, section->offset_within_address_space,
707                           int128_get64(section->size));
708 }
709 
710 static void xen_log_global_start(MemoryListener *listener)
711 {
712     if (xen_enabled()) {
713         xen_in_migration = true;
714     }
715 }
716 
717 static void xen_log_global_stop(MemoryListener *listener)
718 {
719     xen_in_migration = false;
720 }
721 
722 static MemoryListener xen_memory_listener = {
723     .region_add = xen_region_add,
724     .region_del = xen_region_del,
725     .log_start = xen_log_start,
726     .log_stop = xen_log_stop,
727     .log_sync = xen_log_sync,
728     .log_global_start = xen_log_global_start,
729     .log_global_stop = xen_log_global_stop,
730     .priority = 10,
731 };
732 
733 static MemoryListener xen_io_listener = {
734     .region_add = xen_io_add,
735     .region_del = xen_io_del,
736     .priority = 10,
737 };
738 
739 static DeviceListener xen_device_listener = {
740     .realize = xen_device_realize,
741     .unrealize = xen_device_unrealize,
742 };
743 
744 /* get the ioreq packets from share mem */
745 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
746 {
747     ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
748 
749     if (req->state != STATE_IOREQ_READY) {
750         DPRINTF("I/O request not ready: "
751                 "%x, ptr: %x, port: %"PRIx64", "
752                 "data: %"PRIx64", count: %u, size: %u\n",
753                 req->state, req->data_is_ptr, req->addr,
754                 req->data, req->count, req->size);
755         return NULL;
756     }
757 
758     xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
759 
760     req->state = STATE_IOREQ_INPROCESS;
761     return req;
762 }
763 
764 /* use poll to get the port notification */
765 /* ioreq_vec--out,the */
766 /* retval--the number of ioreq packet */
767 static ioreq_t *cpu_get_ioreq(XenIOState *state)
768 {
769     MachineState *ms = MACHINE(qdev_get_machine());
770     unsigned int max_cpus = ms->smp.max_cpus;
771     int i;
772     evtchn_port_t port;
773 
774     port = xenevtchn_pending(state->xce_handle);
775     if (port == state->bufioreq_local_port) {
776         timer_mod(state->buffered_io_timer,
777                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
778         return NULL;
779     }
780 
781     if (port != -1) {
782         for (i = 0; i < max_cpus; i++) {
783             if (state->ioreq_local_port[i] == port) {
784                 break;
785             }
786         }
787 
788         if (i == max_cpus) {
789             hw_error("Fatal error while trying to get io event!\n");
790         }
791 
792         /* unmask the wanted port again */
793         xenevtchn_unmask(state->xce_handle, port);
794 
795         /* get the io packet from shared memory */
796         state->send_vcpu = i;
797         return cpu_get_ioreq_from_shared_memory(state, i);
798     }
799 
800     /* read error or read nothing */
801     return NULL;
802 }
803 
804 static uint32_t do_inp(uint32_t addr, unsigned long size)
805 {
806     switch (size) {
807         case 1:
808             return cpu_inb(addr);
809         case 2:
810             return cpu_inw(addr);
811         case 4:
812             return cpu_inl(addr);
813         default:
814             hw_error("inp: bad size: %04x %lx", addr, size);
815     }
816 }
817 
818 static void do_outp(uint32_t addr,
819         unsigned long size, uint32_t val)
820 {
821     switch (size) {
822         case 1:
823             return cpu_outb(addr, val);
824         case 2:
825             return cpu_outw(addr, val);
826         case 4:
827             return cpu_outl(addr, val);
828         default:
829             hw_error("outp: bad size: %04x %lx", addr, size);
830     }
831 }
832 
833 /*
834  * Helper functions which read/write an object from/to physical guest
835  * memory, as part of the implementation of an ioreq.
836  *
837  * Equivalent to
838  *   cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
839  *                          val, req->size, 0/1)
840  * except without the integer overflow problems.
841  */
842 static void rw_phys_req_item(hwaddr addr,
843                              ioreq_t *req, uint32_t i, void *val, int rw)
844 {
845     /* Do everything unsigned so overflow just results in a truncated result
846      * and accesses to undesired parts of guest memory, which is up
847      * to the guest */
848     hwaddr offset = (hwaddr)req->size * i;
849     if (req->df) {
850         addr -= offset;
851     } else {
852         addr += offset;
853     }
854     cpu_physical_memory_rw(addr, val, req->size, rw);
855 }
856 
857 static inline void read_phys_req_item(hwaddr addr,
858                                       ioreq_t *req, uint32_t i, void *val)
859 {
860     rw_phys_req_item(addr, req, i, val, 0);
861 }
862 static inline void write_phys_req_item(hwaddr addr,
863                                        ioreq_t *req, uint32_t i, void *val)
864 {
865     rw_phys_req_item(addr, req, i, val, 1);
866 }
867 
868 
869 static void cpu_ioreq_pio(ioreq_t *req)
870 {
871     uint32_t i;
872 
873     trace_cpu_ioreq_pio(req, req->dir, req->df, req->data_is_ptr, req->addr,
874                          req->data, req->count, req->size);
875 
876     if (req->size > sizeof(uint32_t)) {
877         hw_error("PIO: bad size (%u)", req->size);
878     }
879 
880     if (req->dir == IOREQ_READ) {
881         if (!req->data_is_ptr) {
882             req->data = do_inp(req->addr, req->size);
883             trace_cpu_ioreq_pio_read_reg(req, req->data, req->addr,
884                                          req->size);
885         } else {
886             uint32_t tmp;
887 
888             for (i = 0; i < req->count; i++) {
889                 tmp = do_inp(req->addr, req->size);
890                 write_phys_req_item(req->data, req, i, &tmp);
891             }
892         }
893     } else if (req->dir == IOREQ_WRITE) {
894         if (!req->data_is_ptr) {
895             trace_cpu_ioreq_pio_write_reg(req, req->data, req->addr,
896                                           req->size);
897             do_outp(req->addr, req->size, req->data);
898         } else {
899             for (i = 0; i < req->count; i++) {
900                 uint32_t tmp = 0;
901 
902                 read_phys_req_item(req->data, req, i, &tmp);
903                 do_outp(req->addr, req->size, tmp);
904             }
905         }
906     }
907 }
908 
909 static void cpu_ioreq_move(ioreq_t *req)
910 {
911     uint32_t i;
912 
913     trace_cpu_ioreq_move(req, req->dir, req->df, req->data_is_ptr, req->addr,
914                          req->data, req->count, req->size);
915 
916     if (req->size > sizeof(req->data)) {
917         hw_error("MMIO: bad size (%u)", req->size);
918     }
919 
920     if (!req->data_is_ptr) {
921         if (req->dir == IOREQ_READ) {
922             for (i = 0; i < req->count; i++) {
923                 read_phys_req_item(req->addr, req, i, &req->data);
924             }
925         } else if (req->dir == IOREQ_WRITE) {
926             for (i = 0; i < req->count; i++) {
927                 write_phys_req_item(req->addr, req, i, &req->data);
928             }
929         }
930     } else {
931         uint64_t tmp;
932 
933         if (req->dir == IOREQ_READ) {
934             for (i = 0; i < req->count; i++) {
935                 read_phys_req_item(req->addr, req, i, &tmp);
936                 write_phys_req_item(req->data, req, i, &tmp);
937             }
938         } else if (req->dir == IOREQ_WRITE) {
939             for (i = 0; i < req->count; i++) {
940                 read_phys_req_item(req->data, req, i, &tmp);
941                 write_phys_req_item(req->addr, req, i, &tmp);
942             }
943         }
944     }
945 }
946 
947 static void cpu_ioreq_config(XenIOState *state, ioreq_t *req)
948 {
949     uint32_t sbdf = req->addr >> 32;
950     uint32_t reg = req->addr;
951     XenPciDevice *xendev;
952 
953     if (req->size != sizeof(uint8_t) && req->size != sizeof(uint16_t) &&
954         req->size != sizeof(uint32_t)) {
955         hw_error("PCI config access: bad size (%u)", req->size);
956     }
957 
958     if (req->count != 1) {
959         hw_error("PCI config access: bad count (%u)", req->count);
960     }
961 
962     QLIST_FOREACH(xendev, &state->dev_list, entry) {
963         if (xendev->sbdf != sbdf) {
964             continue;
965         }
966 
967         if (!req->data_is_ptr) {
968             if (req->dir == IOREQ_READ) {
969                 req->data = pci_host_config_read_common(
970                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
971                     req->size);
972                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
973                                             req->size, req->data);
974             } else if (req->dir == IOREQ_WRITE) {
975                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
976                                              req->size, req->data);
977                 pci_host_config_write_common(
978                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
979                     req->data, req->size);
980             }
981         } else {
982             uint32_t tmp;
983 
984             if (req->dir == IOREQ_READ) {
985                 tmp = pci_host_config_read_common(
986                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
987                     req->size);
988                 trace_cpu_ioreq_config_read(req, xendev->sbdf, reg,
989                                             req->size, tmp);
990                 write_phys_req_item(req->data, req, 0, &tmp);
991             } else if (req->dir == IOREQ_WRITE) {
992                 read_phys_req_item(req->data, req, 0, &tmp);
993                 trace_cpu_ioreq_config_write(req, xendev->sbdf, reg,
994                                              req->size, tmp);
995                 pci_host_config_write_common(
996                     xendev->pci_dev, reg, PCI_CONFIG_SPACE_SIZE,
997                     tmp, req->size);
998             }
999         }
1000     }
1001 }
1002 
1003 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
1004 {
1005     X86CPU *cpu;
1006     CPUX86State *env;
1007 
1008     cpu = X86_CPU(current_cpu);
1009     env = &cpu->env;
1010     env->regs[R_EAX] = req->data;
1011     env->regs[R_EBX] = vmport_regs->ebx;
1012     env->regs[R_ECX] = vmport_regs->ecx;
1013     env->regs[R_EDX] = vmport_regs->edx;
1014     env->regs[R_ESI] = vmport_regs->esi;
1015     env->regs[R_EDI] = vmport_regs->edi;
1016 }
1017 
1018 static void regs_from_cpu(vmware_regs_t *vmport_regs)
1019 {
1020     X86CPU *cpu = X86_CPU(current_cpu);
1021     CPUX86State *env = &cpu->env;
1022 
1023     vmport_regs->ebx = env->regs[R_EBX];
1024     vmport_regs->ecx = env->regs[R_ECX];
1025     vmport_regs->edx = env->regs[R_EDX];
1026     vmport_regs->esi = env->regs[R_ESI];
1027     vmport_regs->edi = env->regs[R_EDI];
1028 }
1029 
1030 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
1031 {
1032     vmware_regs_t *vmport_regs;
1033 
1034     assert(state->shared_vmport_page);
1035     vmport_regs =
1036         &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
1037     QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
1038 
1039     current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
1040     regs_to_cpu(vmport_regs, req);
1041     cpu_ioreq_pio(req);
1042     regs_from_cpu(vmport_regs);
1043     current_cpu = NULL;
1044 }
1045 
1046 static void handle_ioreq(XenIOState *state, ioreq_t *req)
1047 {
1048     trace_handle_ioreq(req, req->type, req->dir, req->df, req->data_is_ptr,
1049                        req->addr, req->data, req->count, req->size);
1050 
1051     if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
1052             (req->size < sizeof (target_ulong))) {
1053         req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
1054     }
1055 
1056     if (req->dir == IOREQ_WRITE)
1057         trace_handle_ioreq_write(req, req->type, req->df, req->data_is_ptr,
1058                                  req->addr, req->data, req->count, req->size);
1059 
1060     switch (req->type) {
1061         case IOREQ_TYPE_PIO:
1062             cpu_ioreq_pio(req);
1063             break;
1064         case IOREQ_TYPE_COPY:
1065             cpu_ioreq_move(req);
1066             break;
1067         case IOREQ_TYPE_VMWARE_PORT:
1068             handle_vmport_ioreq(state, req);
1069             break;
1070         case IOREQ_TYPE_TIMEOFFSET:
1071             break;
1072         case IOREQ_TYPE_INVALIDATE:
1073             xen_invalidate_map_cache();
1074             break;
1075         case IOREQ_TYPE_PCI_CONFIG:
1076             cpu_ioreq_config(state, req);
1077             break;
1078         default:
1079             hw_error("Invalid ioreq type 0x%x\n", req->type);
1080     }
1081     if (req->dir == IOREQ_READ) {
1082         trace_handle_ioreq_read(req, req->type, req->df, req->data_is_ptr,
1083                                 req->addr, req->data, req->count, req->size);
1084     }
1085 }
1086 
1087 static int handle_buffered_iopage(XenIOState *state)
1088 {
1089     buffered_iopage_t *buf_page = state->buffered_io_page;
1090     buf_ioreq_t *buf_req = NULL;
1091     ioreq_t req;
1092     int qw;
1093 
1094     if (!buf_page) {
1095         return 0;
1096     }
1097 
1098     memset(&req, 0x00, sizeof(req));
1099     req.state = STATE_IOREQ_READY;
1100     req.count = 1;
1101     req.dir = IOREQ_WRITE;
1102 
1103     for (;;) {
1104         uint32_t rdptr = buf_page->read_pointer, wrptr;
1105 
1106         xen_rmb();
1107         wrptr = buf_page->write_pointer;
1108         xen_rmb();
1109         if (rdptr != buf_page->read_pointer) {
1110             continue;
1111         }
1112         if (rdptr == wrptr) {
1113             break;
1114         }
1115         buf_req = &buf_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
1116         req.size = 1U << buf_req->size;
1117         req.addr = buf_req->addr;
1118         req.data = buf_req->data;
1119         req.type = buf_req->type;
1120         xen_rmb();
1121         qw = (req.size == 8);
1122         if (qw) {
1123             if (rdptr + 1 == wrptr) {
1124                 hw_error("Incomplete quad word buffered ioreq");
1125             }
1126             buf_req = &buf_page->buf_ioreq[(rdptr + 1) %
1127                                            IOREQ_BUFFER_SLOT_NUM];
1128             req.data |= ((uint64_t)buf_req->data) << 32;
1129             xen_rmb();
1130         }
1131 
1132         handle_ioreq(state, &req);
1133 
1134         /* Only req.data may get updated by handle_ioreq(), albeit even that
1135          * should not happen as such data would never make it to the guest (we
1136          * can only usefully see writes here after all).
1137          */
1138         assert(req.state == STATE_IOREQ_READY);
1139         assert(req.count == 1);
1140         assert(req.dir == IOREQ_WRITE);
1141         assert(!req.data_is_ptr);
1142 
1143         atomic_add(&buf_page->read_pointer, qw + 1);
1144     }
1145 
1146     return req.count;
1147 }
1148 
1149 static void handle_buffered_io(void *opaque)
1150 {
1151     XenIOState *state = opaque;
1152 
1153     if (handle_buffered_iopage(state)) {
1154         timer_mod(state->buffered_io_timer,
1155                 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
1156     } else {
1157         timer_del(state->buffered_io_timer);
1158         xenevtchn_unmask(state->xce_handle, state->bufioreq_local_port);
1159     }
1160 }
1161 
1162 static void cpu_handle_ioreq(void *opaque)
1163 {
1164     XenIOState *state = opaque;
1165     ioreq_t *req = cpu_get_ioreq(state);
1166 
1167     handle_buffered_iopage(state);
1168     if (req) {
1169         ioreq_t copy = *req;
1170 
1171         xen_rmb();
1172         handle_ioreq(state, &copy);
1173         req->data = copy.data;
1174 
1175         if (req->state != STATE_IOREQ_INPROCESS) {
1176             fprintf(stderr, "Badness in I/O request ... not in service?!: "
1177                     "%x, ptr: %x, port: %"PRIx64", "
1178                     "data: %"PRIx64", count: %u, size: %u, type: %u\n",
1179                     req->state, req->data_is_ptr, req->addr,
1180                     req->data, req->count, req->size, req->type);
1181             destroy_hvm_domain(false);
1182             return;
1183         }
1184 
1185         xen_wmb(); /* Update ioreq contents /then/ update state. */
1186 
1187         /*
1188          * We do this before we send the response so that the tools
1189          * have the opportunity to pick up on the reset before the
1190          * guest resumes and does a hlt with interrupts disabled which
1191          * causes Xen to powerdown the domain.
1192          */
1193         if (runstate_is_running()) {
1194             ShutdownCause request;
1195 
1196             if (qemu_shutdown_requested_get()) {
1197                 destroy_hvm_domain(false);
1198             }
1199             request = qemu_reset_requested_get();
1200             if (request) {
1201                 qemu_system_reset(request);
1202                 destroy_hvm_domain(true);
1203             }
1204         }
1205 
1206         req->state = STATE_IORESP_READY;
1207         xenevtchn_notify(state->xce_handle,
1208                          state->ioreq_local_port[state->send_vcpu]);
1209     }
1210 }
1211 
1212 static void xen_main_loop_prepare(XenIOState *state)
1213 {
1214     int evtchn_fd = -1;
1215 
1216     if (state->xce_handle != NULL) {
1217         evtchn_fd = xenevtchn_fd(state->xce_handle);
1218     }
1219 
1220     state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
1221                                                  state);
1222 
1223     if (evtchn_fd != -1) {
1224         CPUState *cpu_state;
1225 
1226         DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
1227         CPU_FOREACH(cpu_state) {
1228             DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1229                     __func__, cpu_state->cpu_index, cpu_state);
1230             state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
1231         }
1232         qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
1233     }
1234 }
1235 
1236 
1237 static void xen_hvm_change_state_handler(void *opaque, int running,
1238                                          RunState rstate)
1239 {
1240     XenIOState *state = opaque;
1241 
1242     if (running) {
1243         xen_main_loop_prepare(state);
1244     }
1245 
1246     xen_set_ioreq_server_state(xen_domid,
1247                                state->ioservid,
1248                                (rstate == RUN_STATE_RUNNING));
1249 }
1250 
1251 static void xen_exit_notifier(Notifier *n, void *data)
1252 {
1253     XenIOState *state = container_of(n, XenIOState, exit);
1254 
1255     xen_destroy_ioreq_server(xen_domid, state->ioservid);
1256 
1257     xenevtchn_close(state->xce_handle);
1258     xs_daemon_close(state->xenstore);
1259 }
1260 
1261 #ifdef XEN_COMPAT_PHYSMAP
1262 static void xen_read_physmap(XenIOState *state)
1263 {
1264     XenPhysmap *physmap = NULL;
1265     unsigned int len, num, i;
1266     char path[80], *value = NULL;
1267     char **entries = NULL;
1268 
1269     snprintf(path, sizeof(path),
1270             "/local/domain/0/device-model/%d/physmap", xen_domid);
1271     entries = xs_directory(state->xenstore, 0, path, &num);
1272     if (entries == NULL)
1273         return;
1274 
1275     for (i = 0; i < num; i++) {
1276         physmap = g_malloc(sizeof (XenPhysmap));
1277         physmap->phys_offset = strtoull(entries[i], NULL, 16);
1278         snprintf(path, sizeof(path),
1279                 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1280                 xen_domid, entries[i]);
1281         value = xs_read(state->xenstore, 0, path, &len);
1282         if (value == NULL) {
1283             g_free(physmap);
1284             continue;
1285         }
1286         physmap->start_addr = strtoull(value, NULL, 16);
1287         free(value);
1288 
1289         snprintf(path, sizeof(path),
1290                 "/local/domain/0/device-model/%d/physmap/%s/size",
1291                 xen_domid, entries[i]);
1292         value = xs_read(state->xenstore, 0, path, &len);
1293         if (value == NULL) {
1294             g_free(physmap);
1295             continue;
1296         }
1297         physmap->size = strtoull(value, NULL, 16);
1298         free(value);
1299 
1300         snprintf(path, sizeof(path),
1301                 "/local/domain/0/device-model/%d/physmap/%s/name",
1302                 xen_domid, entries[i]);
1303         physmap->name = xs_read(state->xenstore, 0, path, &len);
1304 
1305         QLIST_INSERT_HEAD(&xen_physmap, physmap, list);
1306     }
1307     free(entries);
1308 }
1309 #else
1310 static void xen_read_physmap(XenIOState *state)
1311 {
1312 }
1313 #endif
1314 
1315 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1316 {
1317     xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1318 }
1319 
1320 static int xen_map_ioreq_server(XenIOState *state)
1321 {
1322     void *addr = NULL;
1323     xenforeignmemory_resource_handle *fres;
1324     xen_pfn_t ioreq_pfn;
1325     xen_pfn_t bufioreq_pfn;
1326     evtchn_port_t bufioreq_evtchn;
1327     int rc;
1328 
1329     /*
1330      * Attempt to map using the resource API and fall back to normal
1331      * foreign mapping if this is not supported.
1332      */
1333     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_bufioreq != 0);
1334     QEMU_BUILD_BUG_ON(XENMEM_resource_ioreq_server_frame_ioreq(0) != 1);
1335     fres = xenforeignmemory_map_resource(xen_fmem, xen_domid,
1336                                          XENMEM_resource_ioreq_server,
1337                                          state->ioservid, 0, 2,
1338                                          &addr,
1339                                          PROT_READ | PROT_WRITE, 0);
1340     if (fres != NULL) {
1341         trace_xen_map_resource_ioreq(state->ioservid, addr);
1342         state->buffered_io_page = addr;
1343         state->shared_page = addr + TARGET_PAGE_SIZE;
1344     } else if (errno != EOPNOTSUPP) {
1345         error_report("failed to map ioreq server resources: error %d handle=%p",
1346                      errno, xen_xc);
1347         return -1;
1348     }
1349 
1350     rc = xen_get_ioreq_server_info(xen_domid, state->ioservid,
1351                                    (state->shared_page == NULL) ?
1352                                    &ioreq_pfn : NULL,
1353                                    (state->buffered_io_page == NULL) ?
1354                                    &bufioreq_pfn : NULL,
1355                                    &bufioreq_evtchn);
1356     if (rc < 0) {
1357         error_report("failed to get ioreq server info: error %d handle=%p",
1358                      errno, xen_xc);
1359         return rc;
1360     }
1361 
1362     if (state->shared_page == NULL) {
1363         DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1364 
1365         state->shared_page = xenforeignmemory_map(xen_fmem, xen_domid,
1366                                                   PROT_READ | PROT_WRITE,
1367                                                   1, &ioreq_pfn, NULL);
1368         if (state->shared_page == NULL) {
1369             error_report("map shared IO page returned error %d handle=%p",
1370                          errno, xen_xc);
1371         }
1372     }
1373 
1374     if (state->buffered_io_page == NULL) {
1375         DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
1376 
1377         state->buffered_io_page = xenforeignmemory_map(xen_fmem, xen_domid,
1378                                                        PROT_READ | PROT_WRITE,
1379                                                        1, &bufioreq_pfn,
1380                                                        NULL);
1381         if (state->buffered_io_page == NULL) {
1382             error_report("map buffered IO page returned error %d", errno);
1383             return -1;
1384         }
1385     }
1386 
1387     if (state->shared_page == NULL || state->buffered_io_page == NULL) {
1388         return -1;
1389     }
1390 
1391     DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
1392 
1393     state->bufioreq_remote_port = bufioreq_evtchn;
1394 
1395     return 0;
1396 }
1397 
1398 void xen_hvm_init(PCMachineState *pcms, MemoryRegion **ram_memory)
1399 {
1400     MachineState *ms = MACHINE(pcms);
1401     unsigned int max_cpus = ms->smp.max_cpus;
1402     int i, rc;
1403     xen_pfn_t ioreq_pfn;
1404     XenIOState *state;
1405 
1406     state = g_malloc0(sizeof (XenIOState));
1407 
1408     state->xce_handle = xenevtchn_open(NULL, 0);
1409     if (state->xce_handle == NULL) {
1410         perror("xen: event channel open");
1411         goto err;
1412     }
1413 
1414     state->xenstore = xs_daemon_open();
1415     if (state->xenstore == NULL) {
1416         perror("xen: xenstore open");
1417         goto err;
1418     }
1419 
1420     xen_create_ioreq_server(xen_domid, &state->ioservid);
1421 
1422     state->exit.notify = xen_exit_notifier;
1423     qemu_add_exit_notifier(&state->exit);
1424 
1425     state->suspend.notify = xen_suspend_notifier;
1426     qemu_register_suspend_notifier(&state->suspend);
1427 
1428     state->wakeup.notify = xen_wakeup_notifier;
1429     qemu_register_wakeup_notifier(&state->wakeup);
1430 
1431     /*
1432      * Register wake-up support in QMP query-current-machine API
1433      */
1434     qemu_register_wakeup_support();
1435 
1436     rc = xen_map_ioreq_server(state);
1437     if (rc < 0) {
1438         goto err;
1439     }
1440 
1441     rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1442     if (!rc) {
1443         DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1444         state->shared_vmport_page =
1445             xenforeignmemory_map(xen_fmem, xen_domid, PROT_READ|PROT_WRITE,
1446                                  1, &ioreq_pfn, NULL);
1447         if (state->shared_vmport_page == NULL) {
1448             error_report("map shared vmport IO page returned error %d handle=%p",
1449                          errno, xen_xc);
1450             goto err;
1451         }
1452     } else if (rc != -ENOSYS) {
1453         error_report("get vmport regs pfn returned error %d, rc=%d",
1454                      errno, rc);
1455         goto err;
1456     }
1457 
1458     /* Note: cpus is empty at this point in init */
1459     state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1460 
1461     rc = xen_set_ioreq_server_state(xen_domid, state->ioservid, true);
1462     if (rc < 0) {
1463         error_report("failed to enable ioreq server info: error %d handle=%p",
1464                      errno, xen_xc);
1465         goto err;
1466     }
1467 
1468     state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1469 
1470     /* FIXME: how about if we overflow the page here? */
1471     for (i = 0; i < max_cpus; i++) {
1472         rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1473                                         xen_vcpu_eport(state->shared_page, i));
1474         if (rc == -1) {
1475             error_report("shared evtchn %d bind error %d", i, errno);
1476             goto err;
1477         }
1478         state->ioreq_local_port[i] = rc;
1479     }
1480 
1481     rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1482                                     state->bufioreq_remote_port);
1483     if (rc == -1) {
1484         error_report("buffered evtchn bind error %d", errno);
1485         goto err;
1486     }
1487     state->bufioreq_local_port = rc;
1488 
1489     /* Init RAM management */
1490 #ifdef XEN_COMPAT_PHYSMAP
1491     xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1492 #else
1493     xen_map_cache_init(NULL, state);
1494 #endif
1495     xen_ram_init(pcms, ram_size, ram_memory);
1496 
1497     qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1498 
1499     state->memory_listener = xen_memory_listener;
1500     memory_listener_register(&state->memory_listener, &address_space_memory);
1501     state->log_for_dirtybit = NULL;
1502 
1503     state->io_listener = xen_io_listener;
1504     memory_listener_register(&state->io_listener, &address_space_io);
1505 
1506     state->device_listener = xen_device_listener;
1507     QLIST_INIT(&state->dev_list);
1508     device_listener_register(&state->device_listener);
1509 
1510     xen_bus_init();
1511 
1512     /* Initialize backend core & drivers */
1513     if (xen_be_init() != 0) {
1514         error_report("xen backend core setup failed");
1515         goto err;
1516     }
1517     xen_be_register_common();
1518 
1519     QLIST_INIT(&xen_physmap);
1520     xen_read_physmap(state);
1521 
1522     /* Disable ACPI build because Xen handles it */
1523     pcms->acpi_build_enabled = false;
1524 
1525     return;
1526 
1527 err:
1528     error_report("xen hardware virtual machine initialisation failed");
1529     exit(1);
1530 }
1531 
1532 void destroy_hvm_domain(bool reboot)
1533 {
1534     xc_interface *xc_handle;
1535     int sts;
1536     int rc;
1537 
1538     unsigned int reason = reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff;
1539 
1540     if (xen_dmod) {
1541         rc = xendevicemodel_shutdown(xen_dmod, xen_domid, reason);
1542         if (!rc) {
1543             return;
1544         }
1545         if (errno != ENOTTY /* old Xen */) {
1546             perror("xendevicemodel_shutdown failed");
1547         }
1548         /* well, try the old thing then */
1549     }
1550 
1551     xc_handle = xc_interface_open(0, 0, 0);
1552     if (xc_handle == NULL) {
1553         fprintf(stderr, "Cannot acquire xenctrl handle\n");
1554     } else {
1555         sts = xc_domain_shutdown(xc_handle, xen_domid, reason);
1556         if (sts != 0) {
1557             fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1558                     "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1559                     sts, strerror(errno));
1560         } else {
1561             fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1562                     reboot ? "reboot" : "poweroff");
1563         }
1564         xc_interface_close(xc_handle);
1565     }
1566 }
1567 
1568 void xen_register_framebuffer(MemoryRegion *mr)
1569 {
1570     framebuffer = mr;
1571 }
1572 
1573 void xen_shutdown_fatal_error(const char *fmt, ...)
1574 {
1575     va_list ap;
1576 
1577     va_start(ap, fmt);
1578     vfprintf(stderr, fmt, ap);
1579     va_end(ap);
1580     fprintf(stderr, "Will destroy the domain.\n");
1581     /* destroy the domain */
1582     qemu_system_shutdown_request(SHUTDOWN_CAUSE_HOST_ERROR);
1583 }
1584 
1585 void xen_hvm_modified_memory(ram_addr_t start, ram_addr_t length)
1586 {
1587     if (unlikely(xen_in_migration)) {
1588         int rc;
1589         ram_addr_t start_pfn, nb_pages;
1590 
1591         start = xen_phys_offset_to_gaddr(start, length);
1592 
1593         if (length == 0) {
1594             length = TARGET_PAGE_SIZE;
1595         }
1596         start_pfn = start >> TARGET_PAGE_BITS;
1597         nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1598             - start_pfn;
1599         rc = xen_modified_memory(xen_domid, start_pfn, nb_pages);
1600         if (rc) {
1601             fprintf(stderr,
1602                     "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1603                     __func__, start, nb_pages, errno, strerror(errno));
1604         }
1605     }
1606 }
1607 
1608 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1609 {
1610     if (enable) {
1611         memory_global_dirty_log_start();
1612     } else {
1613         memory_global_dirty_log_stop();
1614     }
1615 }
1616