2 * Copyright (C) 2010 Citrix Ltd.
4 * This work is licensed under the terms of the GNU GPL, version 2. See
5 * the COPYING file in the top-level directory.
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.
13 #include "hw/pci/pci.h"
14 #include "hw/i386/pc.h"
15 #include "hw/xen/xen_common.h"
16 #include "hw/xen/xen_backend.h"
17 #include "qmp-commands.h"
19 #include "sysemu/char.h"
20 #include "qemu/error-report.h"
21 #include "qemu/range.h"
22 #include "sysemu/xen-mapcache.h"
24 #include "exec/address-spaces.h"
26 #include <xen/hvm/ioreq.h>
27 #include <xen/hvm/params.h>
28 #include <xen/hvm/e820.h>
30 //#define DEBUG_XEN_HVM
33 #define DPRINTF(fmt, ...) \
34 do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
36 #define DPRINTF(fmt, ...) \
40 static MemoryRegion ram_memory
, ram_640k
, ram_lo
, ram_hi
;
41 static MemoryRegion
*framebuffer
;
42 static bool xen_in_migration
;
44 /* Compatibility with older version */
46 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
47 * installed. This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
48 * needs to be included before this block and hw/xen/xen_common.h needs to
49 * be included before xen/hvm/ioreq.h
51 #ifndef IOREQ_TYPE_VMWARE_PORT
52 #define IOREQ_TYPE_VMWARE_PORT 3
60 typedef struct vmware_regs vmware_regs_t
;
62 struct shared_vmport_iopage
{
63 struct vmware_regs vcpu_vmport_regs
[1];
65 typedef struct shared_vmport_iopage shared_vmport_iopage_t
;
68 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
69 static inline uint32_t xen_vcpu_eport(shared_iopage_t
*shared_page
, int i
)
71 return shared_page
->vcpu_iodata
[i
].vp_eport
;
73 static inline ioreq_t
*xen_vcpu_ioreq(shared_iopage_t
*shared_page
, int vcpu
)
75 return &shared_page
->vcpu_iodata
[vcpu
].vp_ioreq
;
77 # define FMT_ioreq_size PRIx64
79 static inline uint32_t xen_vcpu_eport(shared_iopage_t
*shared_page
, int i
)
81 return shared_page
->vcpu_ioreq
[i
].vp_eport
;
83 static inline ioreq_t
*xen_vcpu_ioreq(shared_iopage_t
*shared_page
, int vcpu
)
85 return &shared_page
->vcpu_ioreq
[vcpu
];
87 # define FMT_ioreq_size "u"
90 #define BUFFER_IO_MAX_DELAY 100
92 typedef struct XenPhysmap
{
98 QLIST_ENTRY(XenPhysmap
) list
;
101 typedef struct XenIOState
{
103 shared_iopage_t
*shared_page
;
104 shared_vmport_iopage_t
*shared_vmport_page
;
105 buffered_iopage_t
*buffered_io_page
;
106 QEMUTimer
*buffered_io_timer
;
107 CPUState
**cpu_by_vcpu_id
;
108 /* the evtchn port for polling the notification, */
109 evtchn_port_t
*ioreq_local_port
;
110 /* evtchn local port for buffered io */
111 evtchn_port_t bufioreq_local_port
;
112 /* the evtchn fd for polling */
113 XenEvtchn xce_handle
;
114 /* which vcpu we are serving */
117 struct xs_handle
*xenstore
;
118 MemoryListener memory_listener
;
119 MemoryListener io_listener
;
120 DeviceListener device_listener
;
121 QLIST_HEAD(, XenPhysmap
) physmap
;
122 hwaddr free_phys_offset
;
123 const XenPhysmap
*log_for_dirtybit
;
130 /* Xen specific function for piix pci */
132 int xen_pci_slot_get_pirq(PCIDevice
*pci_dev
, int irq_num
)
134 return irq_num
+ ((pci_dev
->devfn
>> 3) << 2);
137 void xen_piix3_set_irq(void *opaque
, int irq_num
, int level
)
139 xc_hvm_set_pci_intx_level(xen_xc
, xen_domid
, 0, 0, irq_num
>> 2,
143 void xen_piix_pci_write_config_client(uint32_t address
, uint32_t val
, int len
)
147 /* Scan for updates to PCI link routes (0x60-0x63). */
148 for (i
= 0; i
< len
; i
++) {
149 uint8_t v
= (val
>> (8 * i
)) & 0xff;
154 if (((address
+ i
) >= 0x60) && ((address
+ i
) <= 0x63)) {
155 xc_hvm_set_pci_link_route(xen_xc
, xen_domid
, address
+ i
- 0x60, v
);
160 void xen_hvm_inject_msi(uint64_t addr
, uint32_t data
)
162 xen_xc_hvm_inject_msi(xen_xc
, xen_domid
, addr
, data
);
165 static void xen_suspend_notifier(Notifier
*notifier
, void *data
)
167 xc_set_hvm_param(xen_xc
, xen_domid
, HVM_PARAM_ACPI_S_STATE
, 3);
170 /* Xen Interrupt Controller */
172 static void xen_set_irq(void *opaque
, int irq
, int level
)
174 xc_hvm_set_isa_irq_level(xen_xc
, xen_domid
, irq
, level
);
177 qemu_irq
*xen_interrupt_controller_init(void)
179 return qemu_allocate_irqs(xen_set_irq
, NULL
, 16);
184 static void xen_ram_init(PCMachineState
*pcms
,
185 ram_addr_t ram_size
, MemoryRegion
**ram_memory_p
)
187 MemoryRegion
*sysmem
= get_system_memory();
188 ram_addr_t block_len
;
189 uint64_t user_lowmem
= object_property_get_int(qdev_get_machine(),
190 PC_MACHINE_MAX_RAM_BELOW_4G
,
193 /* Handle the machine opt max-ram-below-4g. It is basically doing
194 * min(xen limit, user limit).
196 if (HVM_BELOW_4G_RAM_END
<= user_lowmem
) {
197 user_lowmem
= HVM_BELOW_4G_RAM_END
;
200 if (ram_size
>= user_lowmem
) {
201 pcms
->above_4g_mem_size
= ram_size
- user_lowmem
;
202 pcms
->below_4g_mem_size
= user_lowmem
;
204 pcms
->above_4g_mem_size
= 0;
205 pcms
->below_4g_mem_size
= ram_size
;
207 if (!pcms
->above_4g_mem_size
) {
208 block_len
= ram_size
;
211 * Xen does not allocate the memory continuously, it keeps a
212 * hole of the size computed above or passed in.
214 block_len
= (1ULL << 32) + pcms
->above_4g_mem_size
;
216 memory_region_init_ram(&ram_memory
, NULL
, "xen.ram", block_len
,
218 *ram_memory_p
= &ram_memory
;
219 vmstate_register_ram_global(&ram_memory
);
221 memory_region_init_alias(&ram_640k
, NULL
, "xen.ram.640k",
222 &ram_memory
, 0, 0xa0000);
223 memory_region_add_subregion(sysmem
, 0, &ram_640k
);
224 /* Skip of the VGA IO memory space, it will be registered later by the VGA
227 * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
228 * the Options ROM, so it is registered here as RAM.
230 memory_region_init_alias(&ram_lo
, NULL
, "xen.ram.lo",
231 &ram_memory
, 0xc0000,
232 pcms
->below_4g_mem_size
- 0xc0000);
233 memory_region_add_subregion(sysmem
, 0xc0000, &ram_lo
);
234 if (pcms
->above_4g_mem_size
> 0) {
235 memory_region_init_alias(&ram_hi
, NULL
, "xen.ram.hi",
236 &ram_memory
, 0x100000000ULL
,
237 pcms
->above_4g_mem_size
);
238 memory_region_add_subregion(sysmem
, 0x100000000ULL
, &ram_hi
);
242 void xen_ram_alloc(ram_addr_t ram_addr
, ram_addr_t size
, MemoryRegion
*mr
)
244 /* FIXME caller ram_block_add() wants error_setg() on failure */
245 unsigned long nr_pfn
;
249 if (runstate_check(RUN_STATE_INMIGRATE
)) {
250 /* RAM already populated in Xen */
251 fprintf(stderr
, "%s: do not alloc "RAM_ADDR_FMT
252 " bytes of ram at "RAM_ADDR_FMT
" when runstate is INMIGRATE\n",
253 __func__
, size
, ram_addr
);
257 if (mr
== &ram_memory
) {
261 trace_xen_ram_alloc(ram_addr
, size
);
263 nr_pfn
= size
>> TARGET_PAGE_BITS
;
264 pfn_list
= g_malloc(sizeof (*pfn_list
) * nr_pfn
);
266 for (i
= 0; i
< nr_pfn
; i
++) {
267 pfn_list
[i
] = (ram_addr
>> TARGET_PAGE_BITS
) + i
;
270 if (xc_domain_populate_physmap_exact(xen_xc
, xen_domid
, nr_pfn
, 0, 0, pfn_list
)) {
271 hw_error("xen: failed to populate ram at " RAM_ADDR_FMT
, ram_addr
);
277 static XenPhysmap
*get_physmapping(XenIOState
*state
,
278 hwaddr start_addr
, ram_addr_t size
)
280 XenPhysmap
*physmap
= NULL
;
282 start_addr
&= TARGET_PAGE_MASK
;
284 QLIST_FOREACH(physmap
, &state
->physmap
, list
) {
285 if (range_covers_byte(physmap
->start_addr
, physmap
->size
, start_addr
)) {
292 static hwaddr
xen_phys_offset_to_gaddr(hwaddr start_addr
,
293 ram_addr_t size
, void *opaque
)
295 hwaddr addr
= start_addr
& TARGET_PAGE_MASK
;
296 XenIOState
*xen_io_state
= opaque
;
297 XenPhysmap
*physmap
= NULL
;
299 QLIST_FOREACH(physmap
, &xen_io_state
->physmap
, list
) {
300 if (range_covers_byte(physmap
->phys_offset
, physmap
->size
, addr
)) {
301 return physmap
->start_addr
;
308 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
309 static int xen_add_to_physmap(XenIOState
*state
,
313 hwaddr offset_within_region
)
317 XenPhysmap
*physmap
= NULL
;
318 hwaddr pfn
, start_gpfn
;
319 hwaddr phys_offset
= memory_region_get_ram_addr(mr
);
320 char path
[80], value
[17];
323 if (get_physmapping(state
, start_addr
, size
)) {
330 /* Xen can only handle a single dirty log region for now and we want
331 * the linear framebuffer to be that region.
332 * Avoid tracking any regions that is not videoram and avoid tracking
333 * the legacy vga region. */
334 if (mr
== framebuffer
&& start_addr
> 0xbffff) {
340 DPRINTF("mapping vram to %"HWADDR_PRIx
" - %"HWADDR_PRIx
"\n",
341 start_addr
, start_addr
+ size
);
343 pfn
= phys_offset
>> TARGET_PAGE_BITS
;
344 start_gpfn
= start_addr
>> TARGET_PAGE_BITS
;
345 for (i
= 0; i
< size
>> TARGET_PAGE_BITS
; i
++) {
346 unsigned long idx
= pfn
+ i
;
347 xen_pfn_t gpfn
= start_gpfn
+ i
;
349 rc
= xen_xc_domain_add_to_physmap(xen_xc
, xen_domid
, XENMAPSPACE_gmfn
, idx
, gpfn
);
351 DPRINTF("add_to_physmap MFN %"PRI_xen_pfn
" to PFN %"
352 PRI_xen_pfn
" failed: %d (errno: %d)\n", idx
, gpfn
, rc
, errno
);
357 mr_name
= memory_region_name(mr
);
359 physmap
= g_malloc(sizeof (XenPhysmap
));
361 physmap
->start_addr
= start_addr
;
362 physmap
->size
= size
;
363 physmap
->name
= mr_name
;
364 physmap
->phys_offset
= phys_offset
;
366 QLIST_INSERT_HEAD(&state
->physmap
, physmap
, list
);
368 xc_domain_pin_memory_cacheattr(xen_xc
, xen_domid
,
369 start_addr
>> TARGET_PAGE_BITS
,
370 (start_addr
+ size
- 1) >> TARGET_PAGE_BITS
,
371 XEN_DOMCTL_MEM_CACHEATTR_WB
);
373 snprintf(path
, sizeof(path
),
374 "/local/domain/0/device-model/%d/physmap/%"PRIx64
"/start_addr",
375 xen_domid
, (uint64_t)phys_offset
);
376 snprintf(value
, sizeof(value
), "%"PRIx64
, (uint64_t)start_addr
);
377 if (!xs_write(state
->xenstore
, 0, path
, value
, strlen(value
))) {
380 snprintf(path
, sizeof(path
),
381 "/local/domain/0/device-model/%d/physmap/%"PRIx64
"/size",
382 xen_domid
, (uint64_t)phys_offset
);
383 snprintf(value
, sizeof(value
), "%"PRIx64
, (uint64_t)size
);
384 if (!xs_write(state
->xenstore
, 0, path
, value
, strlen(value
))) {
388 snprintf(path
, sizeof(path
),
389 "/local/domain/0/device-model/%d/physmap/%"PRIx64
"/name",
390 xen_domid
, (uint64_t)phys_offset
);
391 if (!xs_write(state
->xenstore
, 0, path
, mr_name
, strlen(mr_name
))) {
399 static int xen_remove_from_physmap(XenIOState
*state
,
405 XenPhysmap
*physmap
= NULL
;
406 hwaddr phys_offset
= 0;
408 physmap
= get_physmapping(state
, start_addr
, size
);
409 if (physmap
== NULL
) {
413 phys_offset
= physmap
->phys_offset
;
414 size
= physmap
->size
;
416 DPRINTF("unmapping vram to %"HWADDR_PRIx
" - %"HWADDR_PRIx
", at "
417 "%"HWADDR_PRIx
"\n", start_addr
, start_addr
+ size
, phys_offset
);
419 size
>>= TARGET_PAGE_BITS
;
420 start_addr
>>= TARGET_PAGE_BITS
;
421 phys_offset
>>= TARGET_PAGE_BITS
;
422 for (i
= 0; i
< size
; i
++) {
423 xen_pfn_t idx
= start_addr
+ i
;
424 xen_pfn_t gpfn
= phys_offset
+ i
;
426 rc
= xen_xc_domain_add_to_physmap(xen_xc
, xen_domid
, XENMAPSPACE_gmfn
, idx
, gpfn
);
428 fprintf(stderr
, "add_to_physmap MFN %"PRI_xen_pfn
" to PFN %"
429 PRI_xen_pfn
" failed: %d (errno: %d)\n", idx
, gpfn
, rc
, errno
);
434 QLIST_REMOVE(physmap
, list
);
435 if (state
->log_for_dirtybit
== physmap
) {
436 state
->log_for_dirtybit
= NULL
;
444 static int xen_add_to_physmap(XenIOState
*state
,
448 hwaddr offset_within_region
)
453 static int xen_remove_from_physmap(XenIOState
*state
,
461 static void xen_set_memory(struct MemoryListener
*listener
,
462 MemoryRegionSection
*section
,
465 XenIOState
*state
= container_of(listener
, XenIOState
, memory_listener
);
466 hwaddr start_addr
= section
->offset_within_address_space
;
467 ram_addr_t size
= int128_get64(section
->size
);
468 bool log_dirty
= memory_region_is_logging(section
->mr
, DIRTY_MEMORY_VGA
);
469 hvmmem_type_t mem_type
;
471 if (section
->mr
== &ram_memory
) {
475 xen_map_memory_section(xen_xc
, xen_domid
, state
->ioservid
,
478 xen_unmap_memory_section(xen_xc
, xen_domid
, state
->ioservid
,
483 if (!memory_region_is_ram(section
->mr
)) {
487 if (log_dirty
!= add
) {
491 trace_xen_client_set_memory(start_addr
, size
, log_dirty
);
493 start_addr
&= TARGET_PAGE_MASK
;
494 size
= TARGET_PAGE_ALIGN(size
);
497 if (!memory_region_is_rom(section
->mr
)) {
498 xen_add_to_physmap(state
, start_addr
, size
,
499 section
->mr
, section
->offset_within_region
);
501 mem_type
= HVMMEM_ram_ro
;
502 if (xc_hvm_set_mem_type(xen_xc
, xen_domid
, mem_type
,
503 start_addr
>> TARGET_PAGE_BITS
,
504 size
>> TARGET_PAGE_BITS
)) {
505 DPRINTF("xc_hvm_set_mem_type error, addr: "TARGET_FMT_plx
"\n",
510 if (xen_remove_from_physmap(state
, start_addr
, size
) < 0) {
511 DPRINTF("physmapping does not exist at "TARGET_FMT_plx
"\n", start_addr
);
516 static void xen_region_add(MemoryListener
*listener
,
517 MemoryRegionSection
*section
)
519 memory_region_ref(section
->mr
);
520 xen_set_memory(listener
, section
, true);
523 static void xen_region_del(MemoryListener
*listener
,
524 MemoryRegionSection
*section
)
526 xen_set_memory(listener
, section
, false);
527 memory_region_unref(section
->mr
);
530 static void xen_io_add(MemoryListener
*listener
,
531 MemoryRegionSection
*section
)
533 XenIOState
*state
= container_of(listener
, XenIOState
, io_listener
);
535 memory_region_ref(section
->mr
);
537 xen_map_io_section(xen_xc
, xen_domid
, state
->ioservid
, section
);
540 static void xen_io_del(MemoryListener
*listener
,
541 MemoryRegionSection
*section
)
543 XenIOState
*state
= container_of(listener
, XenIOState
, io_listener
);
545 xen_unmap_io_section(xen_xc
, xen_domid
, state
->ioservid
, section
);
547 memory_region_unref(section
->mr
);
550 static void xen_device_realize(DeviceListener
*listener
,
553 XenIOState
*state
= container_of(listener
, XenIOState
, device_listener
);
555 if (object_dynamic_cast(OBJECT(dev
), TYPE_PCI_DEVICE
)) {
556 PCIDevice
*pci_dev
= PCI_DEVICE(dev
);
558 xen_map_pcidev(xen_xc
, xen_domid
, state
->ioservid
, pci_dev
);
562 static void xen_device_unrealize(DeviceListener
*listener
,
565 XenIOState
*state
= container_of(listener
, XenIOState
, device_listener
);
567 if (object_dynamic_cast(OBJECT(dev
), TYPE_PCI_DEVICE
)) {
568 PCIDevice
*pci_dev
= PCI_DEVICE(dev
);
570 xen_unmap_pcidev(xen_xc
, xen_domid
, state
->ioservid
, pci_dev
);
574 static void xen_sync_dirty_bitmap(XenIOState
*state
,
578 hwaddr npages
= size
>> TARGET_PAGE_BITS
;
579 const int width
= sizeof(unsigned long) * 8;
580 unsigned long bitmap
[(npages
+ width
- 1) / width
];
582 const XenPhysmap
*physmap
= NULL
;
584 physmap
= get_physmapping(state
, start_addr
, size
);
585 if (physmap
== NULL
) {
590 if (state
->log_for_dirtybit
== NULL
) {
591 state
->log_for_dirtybit
= physmap
;
592 } else if (state
->log_for_dirtybit
!= physmap
) {
593 /* Only one range for dirty bitmap can be tracked. */
597 rc
= xc_hvm_track_dirty_vram(xen_xc
, xen_domid
,
598 start_addr
>> TARGET_PAGE_BITS
, npages
,
602 #define ENODATA ENOENT
604 if (errno
== ENODATA
) {
605 memory_region_set_dirty(framebuffer
, 0, size
);
606 DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
607 ", 0x" TARGET_FMT_plx
"): %s\n",
608 start_addr
, start_addr
+ size
, strerror(errno
));
613 for (i
= 0; i
< ARRAY_SIZE(bitmap
); i
++) {
614 unsigned long map
= bitmap
[i
];
618 memory_region_set_dirty(framebuffer
,
619 (i
* width
+ j
) * TARGET_PAGE_SIZE
,
625 static void xen_log_start(MemoryListener
*listener
,
626 MemoryRegionSection
*section
,
629 XenIOState
*state
= container_of(listener
, XenIOState
, memory_listener
);
631 if (new & ~old
& (1 << DIRTY_MEMORY_VGA
)) {
632 xen_sync_dirty_bitmap(state
, section
->offset_within_address_space
,
633 int128_get64(section
->size
));
637 static void xen_log_stop(MemoryListener
*listener
, MemoryRegionSection
*section
,
640 XenIOState
*state
= container_of(listener
, XenIOState
, memory_listener
);
642 if (old
& ~new & (1 << DIRTY_MEMORY_VGA
)) {
643 state
->log_for_dirtybit
= NULL
;
644 /* Disable dirty bit tracking */
645 xc_hvm_track_dirty_vram(xen_xc
, xen_domid
, 0, 0, NULL
);
649 static void xen_log_sync(MemoryListener
*listener
, MemoryRegionSection
*section
)
651 XenIOState
*state
= container_of(listener
, XenIOState
, memory_listener
);
653 xen_sync_dirty_bitmap(state
, section
->offset_within_address_space
,
654 int128_get64(section
->size
));
657 static void xen_log_global_start(MemoryListener
*listener
)
660 xen_in_migration
= true;
664 static void xen_log_global_stop(MemoryListener
*listener
)
666 xen_in_migration
= false;
669 static MemoryListener xen_memory_listener
= {
670 .region_add
= xen_region_add
,
671 .region_del
= xen_region_del
,
672 .log_start
= xen_log_start
,
673 .log_stop
= xen_log_stop
,
674 .log_sync
= xen_log_sync
,
675 .log_global_start
= xen_log_global_start
,
676 .log_global_stop
= xen_log_global_stop
,
680 static MemoryListener xen_io_listener
= {
681 .region_add
= xen_io_add
,
682 .region_del
= xen_io_del
,
686 static DeviceListener xen_device_listener
= {
687 .realize
= xen_device_realize
,
688 .unrealize
= xen_device_unrealize
,
691 /* get the ioreq packets from share mem */
692 static ioreq_t
*cpu_get_ioreq_from_shared_memory(XenIOState
*state
, int vcpu
)
694 ioreq_t
*req
= xen_vcpu_ioreq(state
->shared_page
, vcpu
);
696 if (req
->state
!= STATE_IOREQ_READY
) {
697 DPRINTF("I/O request not ready: "
698 "%x, ptr: %x, port: %"PRIx64
", "
699 "data: %"PRIx64
", count: %" FMT_ioreq_size
", size: %" FMT_ioreq_size
"\n",
700 req
->state
, req
->data_is_ptr
, req
->addr
,
701 req
->data
, req
->count
, req
->size
);
705 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
707 req
->state
= STATE_IOREQ_INPROCESS
;
711 /* use poll to get the port notification */
712 /* ioreq_vec--out,the */
713 /* retval--the number of ioreq packet */
714 static ioreq_t
*cpu_get_ioreq(XenIOState
*state
)
719 port
= xc_evtchn_pending(state
->xce_handle
);
720 if (port
== state
->bufioreq_local_port
) {
721 timer_mod(state
->buffered_io_timer
,
722 BUFFER_IO_MAX_DELAY
+ qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
727 for (i
= 0; i
< max_cpus
; i
++) {
728 if (state
->ioreq_local_port
[i
] == port
) {
734 hw_error("Fatal error while trying to get io event!\n");
737 /* unmask the wanted port again */
738 xc_evtchn_unmask(state
->xce_handle
, port
);
740 /* get the io packet from shared memory */
741 state
->send_vcpu
= i
;
742 return cpu_get_ioreq_from_shared_memory(state
, i
);
745 /* read error or read nothing */
749 static uint32_t do_inp(pio_addr_t addr
, unsigned long size
)
753 return cpu_inb(addr
);
755 return cpu_inw(addr
);
757 return cpu_inl(addr
);
759 hw_error("inp: bad size: %04"FMT_pioaddr
" %lx", addr
, size
);
763 static void do_outp(pio_addr_t addr
,
764 unsigned long size
, uint32_t val
)
768 return cpu_outb(addr
, val
);
770 return cpu_outw(addr
, val
);
772 return cpu_outl(addr
, val
);
774 hw_error("outp: bad size: %04"FMT_pioaddr
" %lx", addr
, size
);
779 * Helper functions which read/write an object from/to physical guest
780 * memory, as part of the implementation of an ioreq.
783 * cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
784 * val, req->size, 0/1)
785 * except without the integer overflow problems.
787 static void rw_phys_req_item(hwaddr addr
,
788 ioreq_t
*req
, uint32_t i
, void *val
, int rw
)
790 /* Do everything unsigned so overflow just results in a truncated result
791 * and accesses to undesired parts of guest memory, which is up
793 hwaddr offset
= (hwaddr
)req
->size
* i
;
799 cpu_physical_memory_rw(addr
, val
, req
->size
, rw
);
802 static inline void read_phys_req_item(hwaddr addr
,
803 ioreq_t
*req
, uint32_t i
, void *val
)
805 rw_phys_req_item(addr
, req
, i
, val
, 0);
807 static inline void write_phys_req_item(hwaddr addr
,
808 ioreq_t
*req
, uint32_t i
, void *val
)
810 rw_phys_req_item(addr
, req
, i
, val
, 1);
814 static void cpu_ioreq_pio(ioreq_t
*req
)
818 trace_cpu_ioreq_pio(req
, req
->dir
, req
->df
, req
->data_is_ptr
, req
->addr
,
819 req
->data
, req
->count
, req
->size
);
821 if (req
->dir
== IOREQ_READ
) {
822 if (!req
->data_is_ptr
) {
823 req
->data
= do_inp(req
->addr
, req
->size
);
824 trace_cpu_ioreq_pio_read_reg(req
, req
->data
, req
->addr
,
829 for (i
= 0; i
< req
->count
; i
++) {
830 tmp
= do_inp(req
->addr
, req
->size
);
831 write_phys_req_item(req
->data
, req
, i
, &tmp
);
834 } else if (req
->dir
== IOREQ_WRITE
) {
835 if (!req
->data_is_ptr
) {
836 trace_cpu_ioreq_pio_write_reg(req
, req
->data
, req
->addr
,
838 do_outp(req
->addr
, req
->size
, req
->data
);
840 for (i
= 0; i
< req
->count
; i
++) {
843 read_phys_req_item(req
->data
, req
, i
, &tmp
);
844 do_outp(req
->addr
, req
->size
, tmp
);
850 static void cpu_ioreq_move(ioreq_t
*req
)
854 trace_cpu_ioreq_move(req
, req
->dir
, req
->df
, req
->data_is_ptr
, req
->addr
,
855 req
->data
, req
->count
, req
->size
);
857 if (!req
->data_is_ptr
) {
858 if (req
->dir
== IOREQ_READ
) {
859 for (i
= 0; i
< req
->count
; i
++) {
860 read_phys_req_item(req
->addr
, req
, i
, &req
->data
);
862 } else if (req
->dir
== IOREQ_WRITE
) {
863 for (i
= 0; i
< req
->count
; i
++) {
864 write_phys_req_item(req
->addr
, req
, i
, &req
->data
);
870 if (req
->dir
== IOREQ_READ
) {
871 for (i
= 0; i
< req
->count
; i
++) {
872 read_phys_req_item(req
->addr
, req
, i
, &tmp
);
873 write_phys_req_item(req
->data
, req
, i
, &tmp
);
875 } else if (req
->dir
== IOREQ_WRITE
) {
876 for (i
= 0; i
< req
->count
; i
++) {
877 read_phys_req_item(req
->data
, req
, i
, &tmp
);
878 write_phys_req_item(req
->addr
, req
, i
, &tmp
);
884 static void regs_to_cpu(vmware_regs_t
*vmport_regs
, ioreq_t
*req
)
889 cpu
= X86_CPU(current_cpu
);
891 env
->regs
[R_EAX
] = req
->data
;
892 env
->regs
[R_EBX
] = vmport_regs
->ebx
;
893 env
->regs
[R_ECX
] = vmport_regs
->ecx
;
894 env
->regs
[R_EDX
] = vmport_regs
->edx
;
895 env
->regs
[R_ESI
] = vmport_regs
->esi
;
896 env
->regs
[R_EDI
] = vmport_regs
->edi
;
899 static void regs_from_cpu(vmware_regs_t
*vmport_regs
)
901 X86CPU
*cpu
= X86_CPU(current_cpu
);
902 CPUX86State
*env
= &cpu
->env
;
904 vmport_regs
->ebx
= env
->regs
[R_EBX
];
905 vmport_regs
->ecx
= env
->regs
[R_ECX
];
906 vmport_regs
->edx
= env
->regs
[R_EDX
];
907 vmport_regs
->esi
= env
->regs
[R_ESI
];
908 vmport_regs
->edi
= env
->regs
[R_EDI
];
911 static void handle_vmport_ioreq(XenIOState
*state
, ioreq_t
*req
)
913 vmware_regs_t
*vmport_regs
;
915 assert(state
->shared_vmport_page
);
917 &state
->shared_vmport_page
->vcpu_vmport_regs
[state
->send_vcpu
];
918 QEMU_BUILD_BUG_ON(sizeof(*req
) < sizeof(*vmport_regs
));
920 current_cpu
= state
->cpu_by_vcpu_id
[state
->send_vcpu
];
921 regs_to_cpu(vmport_regs
, req
);
923 regs_from_cpu(vmport_regs
);
927 static void handle_ioreq(XenIOState
*state
, ioreq_t
*req
)
929 trace_handle_ioreq(req
, req
->type
, req
->dir
, req
->df
, req
->data_is_ptr
,
930 req
->addr
, req
->data
, req
->count
, req
->size
);
932 if (!req
->data_is_ptr
&& (req
->dir
== IOREQ_WRITE
) &&
933 (req
->size
< sizeof (target_ulong
))) {
934 req
->data
&= ((target_ulong
) 1 << (8 * req
->size
)) - 1;
937 if (req
->dir
== IOREQ_WRITE
)
938 trace_handle_ioreq_write(req
, req
->type
, req
->df
, req
->data_is_ptr
,
939 req
->addr
, req
->data
, req
->count
, req
->size
);
945 case IOREQ_TYPE_COPY
:
948 case IOREQ_TYPE_VMWARE_PORT
:
949 handle_vmport_ioreq(state
, req
);
951 case IOREQ_TYPE_TIMEOFFSET
:
953 case IOREQ_TYPE_INVALIDATE
:
954 xen_invalidate_map_cache();
956 case IOREQ_TYPE_PCI_CONFIG
: {
957 uint32_t sbdf
= req
->addr
>> 32;
960 /* Fake a write to port 0xCF8 so that
961 * the config space access will target the
962 * correct device model.
965 ((req
->addr
& 0x0f00) << 16) |
966 ((sbdf
& 0xffff) << 8) |
968 do_outp(0xcf8, 4, val
);
970 /* Now issue the config space access via
973 req
->addr
= 0xcfc | (req
->addr
& 0x03);
978 hw_error("Invalid ioreq type 0x%x\n", req
->type
);
980 if (req
->dir
== IOREQ_READ
) {
981 trace_handle_ioreq_read(req
, req
->type
, req
->df
, req
->data_is_ptr
,
982 req
->addr
, req
->data
, req
->count
, req
->size
);
986 static int handle_buffered_iopage(XenIOState
*state
)
988 buffered_iopage_t
*buf_page
= state
->buffered_io_page
;
989 buf_ioreq_t
*buf_req
= NULL
;
997 memset(&req
, 0x00, sizeof(req
));
1000 uint32_t rdptr
= buf_page
->read_pointer
, wrptr
;
1003 wrptr
= buf_page
->write_pointer
;
1005 if (rdptr
!= buf_page
->read_pointer
) {
1008 if (rdptr
== wrptr
) {
1011 buf_req
= &buf_page
->buf_ioreq
[rdptr
% IOREQ_BUFFER_SLOT_NUM
];
1012 req
.size
= 1UL << buf_req
->size
;
1014 req
.addr
= buf_req
->addr
;
1015 req
.data
= buf_req
->data
;
1016 req
.state
= STATE_IOREQ_READY
;
1017 req
.dir
= buf_req
->dir
;
1019 req
.type
= buf_req
->type
;
1020 req
.data_is_ptr
= 0;
1021 qw
= (req
.size
== 8);
1023 buf_req
= &buf_page
->buf_ioreq
[(rdptr
+ 1) %
1024 IOREQ_BUFFER_SLOT_NUM
];
1025 req
.data
|= ((uint64_t)buf_req
->data
) << 32;
1028 handle_ioreq(state
, &req
);
1030 atomic_add(&buf_page
->read_pointer
, qw
+ 1);
1036 static void handle_buffered_io(void *opaque
)
1038 XenIOState
*state
= opaque
;
1040 if (handle_buffered_iopage(state
)) {
1041 timer_mod(state
->buffered_io_timer
,
1042 BUFFER_IO_MAX_DELAY
+ qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
1044 timer_del(state
->buffered_io_timer
);
1045 xc_evtchn_unmask(state
->xce_handle
, state
->bufioreq_local_port
);
1049 static void cpu_handle_ioreq(void *opaque
)
1051 XenIOState
*state
= opaque
;
1052 ioreq_t
*req
= cpu_get_ioreq(state
);
1054 handle_buffered_iopage(state
);
1056 handle_ioreq(state
, req
);
1058 if (req
->state
!= STATE_IOREQ_INPROCESS
) {
1059 fprintf(stderr
, "Badness in I/O request ... not in service?!: "
1060 "%x, ptr: %x, port: %"PRIx64
", "
1061 "data: %"PRIx64
", count: %" FMT_ioreq_size
1062 ", size: %" FMT_ioreq_size
1063 ", type: %"FMT_ioreq_size
"\n",
1064 req
->state
, req
->data_is_ptr
, req
->addr
,
1065 req
->data
, req
->count
, req
->size
, req
->type
);
1066 destroy_hvm_domain(false);
1070 xen_wmb(); /* Update ioreq contents /then/ update state. */
1073 * We do this before we send the response so that the tools
1074 * have the opportunity to pick up on the reset before the
1075 * guest resumes and does a hlt with interrupts disabled which
1076 * causes Xen to powerdown the domain.
1078 if (runstate_is_running()) {
1079 if (qemu_shutdown_requested_get()) {
1080 destroy_hvm_domain(false);
1082 if (qemu_reset_requested_get()) {
1083 qemu_system_reset(VMRESET_REPORT
);
1084 destroy_hvm_domain(true);
1088 req
->state
= STATE_IORESP_READY
;
1089 xc_evtchn_notify(state
->xce_handle
, state
->ioreq_local_port
[state
->send_vcpu
]);
1093 static void xen_main_loop_prepare(XenIOState
*state
)
1097 if (state
->xce_handle
!= XC_HANDLER_INITIAL_VALUE
) {
1098 evtchn_fd
= xc_evtchn_fd(state
->xce_handle
);
1101 state
->buffered_io_timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, handle_buffered_io
,
1104 if (evtchn_fd
!= -1) {
1105 CPUState
*cpu_state
;
1107 DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__
);
1108 CPU_FOREACH(cpu_state
) {
1109 DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1110 __func__
, cpu_state
->cpu_index
, cpu_state
);
1111 state
->cpu_by_vcpu_id
[cpu_state
->cpu_index
] = cpu_state
;
1113 qemu_set_fd_handler(evtchn_fd
, cpu_handle_ioreq
, NULL
, state
);
1118 static void xen_hvm_change_state_handler(void *opaque
, int running
,
1121 XenIOState
*state
= opaque
;
1124 xen_main_loop_prepare(state
);
1127 xen_set_ioreq_server_state(xen_xc
, xen_domid
,
1129 (rstate
== RUN_STATE_RUNNING
));
1132 static void xen_exit_notifier(Notifier
*n
, void *data
)
1134 XenIOState
*state
= container_of(n
, XenIOState
, exit
);
1136 xc_evtchn_close(state
->xce_handle
);
1137 xs_daemon_close(state
->xenstore
);
1140 static void xen_read_physmap(XenIOState
*state
)
1142 XenPhysmap
*physmap
= NULL
;
1143 unsigned int len
, num
, i
;
1144 char path
[80], *value
= NULL
;
1145 char **entries
= NULL
;
1147 snprintf(path
, sizeof(path
),
1148 "/local/domain/0/device-model/%d/physmap", xen_domid
);
1149 entries
= xs_directory(state
->xenstore
, 0, path
, &num
);
1150 if (entries
== NULL
)
1153 for (i
= 0; i
< num
; i
++) {
1154 physmap
= g_malloc(sizeof (XenPhysmap
));
1155 physmap
->phys_offset
= strtoull(entries
[i
], NULL
, 16);
1156 snprintf(path
, sizeof(path
),
1157 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1158 xen_domid
, entries
[i
]);
1159 value
= xs_read(state
->xenstore
, 0, path
, &len
);
1160 if (value
== NULL
) {
1164 physmap
->start_addr
= strtoull(value
, NULL
, 16);
1167 snprintf(path
, sizeof(path
),
1168 "/local/domain/0/device-model/%d/physmap/%s/size",
1169 xen_domid
, entries
[i
]);
1170 value
= xs_read(state
->xenstore
, 0, path
, &len
);
1171 if (value
== NULL
) {
1175 physmap
->size
= strtoull(value
, NULL
, 16);
1178 snprintf(path
, sizeof(path
),
1179 "/local/domain/0/device-model/%d/physmap/%s/name",
1180 xen_domid
, entries
[i
]);
1181 physmap
->name
= xs_read(state
->xenstore
, 0, path
, &len
);
1183 QLIST_INSERT_HEAD(&state
->physmap
, physmap
, list
);
1188 static void xen_wakeup_notifier(Notifier
*notifier
, void *data
)
1190 xc_set_hvm_param(xen_xc
, xen_domid
, HVM_PARAM_ACPI_S_STATE
, 0);
1193 void xen_hvm_init(PCMachineState
*pcms
, MemoryRegion
**ram_memory
)
1196 xen_pfn_t ioreq_pfn
;
1197 xen_pfn_t bufioreq_pfn
;
1198 evtchn_port_t bufioreq_evtchn
;
1201 state
= g_malloc0(sizeof (XenIOState
));
1203 state
->xce_handle
= xen_xc_evtchn_open(NULL
, 0);
1204 if (state
->xce_handle
== XC_HANDLER_INITIAL_VALUE
) {
1205 perror("xen: event channel open");
1209 state
->xenstore
= xs_daemon_open();
1210 if (state
->xenstore
== NULL
) {
1211 perror("xen: xenstore open");
1215 rc
= xen_create_ioreq_server(xen_xc
, xen_domid
, &state
->ioservid
);
1217 perror("xen: ioreq server create");
1221 state
->exit
.notify
= xen_exit_notifier
;
1222 qemu_add_exit_notifier(&state
->exit
);
1224 state
->suspend
.notify
= xen_suspend_notifier
;
1225 qemu_register_suspend_notifier(&state
->suspend
);
1227 state
->wakeup
.notify
= xen_wakeup_notifier
;
1228 qemu_register_wakeup_notifier(&state
->wakeup
);
1230 rc
= xen_get_ioreq_server_info(xen_xc
, xen_domid
, state
->ioservid
,
1231 &ioreq_pfn
, &bufioreq_pfn
,
1234 error_report("failed to get ioreq server info: error %d handle=" XC_INTERFACE_FMT
,
1239 DPRINTF("shared page at pfn %lx\n", ioreq_pfn
);
1240 DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn
);
1241 DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn
);
1243 state
->shared_page
= xc_map_foreign_range(xen_xc
, xen_domid
, XC_PAGE_SIZE
,
1244 PROT_READ
|PROT_WRITE
, ioreq_pfn
);
1245 if (state
->shared_page
== NULL
) {
1246 error_report("map shared IO page returned error %d handle=" XC_INTERFACE_FMT
,
1251 rc
= xen_get_vmport_regs_pfn(xen_xc
, xen_domid
, &ioreq_pfn
);
1253 DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn
);
1254 state
->shared_vmport_page
=
1255 xc_map_foreign_range(xen_xc
, xen_domid
, XC_PAGE_SIZE
,
1256 PROT_READ
|PROT_WRITE
, ioreq_pfn
);
1257 if (state
->shared_vmport_page
== NULL
) {
1258 error_report("map shared vmport IO page returned error %d handle="
1259 XC_INTERFACE_FMT
, errno
, xen_xc
);
1262 } else if (rc
!= -ENOSYS
) {
1263 error_report("get vmport regs pfn returned error %d, rc=%d",
1268 state
->buffered_io_page
= xc_map_foreign_range(xen_xc
, xen_domid
,
1270 PROT_READ
|PROT_WRITE
,
1272 if (state
->buffered_io_page
== NULL
) {
1273 error_report("map buffered IO page returned error %d", errno
);
1277 /* Note: cpus is empty at this point in init */
1278 state
->cpu_by_vcpu_id
= g_malloc0(max_cpus
* sizeof(CPUState
*));
1280 rc
= xen_set_ioreq_server_state(xen_xc
, xen_domid
, state
->ioservid
, true);
1282 error_report("failed to enable ioreq server info: error %d handle=" XC_INTERFACE_FMT
,
1287 state
->ioreq_local_port
= g_malloc0(max_cpus
* sizeof (evtchn_port_t
));
1289 /* FIXME: how about if we overflow the page here? */
1290 for (i
= 0; i
< max_cpus
; i
++) {
1291 rc
= xc_evtchn_bind_interdomain(state
->xce_handle
, xen_domid
,
1292 xen_vcpu_eport(state
->shared_page
, i
));
1294 error_report("shared evtchn %d bind error %d", i
, errno
);
1297 state
->ioreq_local_port
[i
] = rc
;
1300 rc
= xc_evtchn_bind_interdomain(state
->xce_handle
, xen_domid
,
1303 error_report("buffered evtchn bind error %d", errno
);
1306 state
->bufioreq_local_port
= rc
;
1308 /* Init RAM management */
1309 xen_map_cache_init(xen_phys_offset_to_gaddr
, state
);
1310 xen_ram_init(pcms
, ram_size
, ram_memory
);
1312 qemu_add_vm_change_state_handler(xen_hvm_change_state_handler
, state
);
1314 state
->memory_listener
= xen_memory_listener
;
1315 QLIST_INIT(&state
->physmap
);
1316 memory_listener_register(&state
->memory_listener
, &address_space_memory
);
1317 state
->log_for_dirtybit
= NULL
;
1319 state
->io_listener
= xen_io_listener
;
1320 memory_listener_register(&state
->io_listener
, &address_space_io
);
1322 state
->device_listener
= xen_device_listener
;
1323 device_listener_register(&state
->device_listener
);
1325 /* Initialize backend core & drivers */
1326 if (xen_be_init() != 0) {
1327 error_report("xen backend core setup failed");
1330 xen_be_register("console", &xen_console_ops
);
1331 xen_be_register("vkbd", &xen_kbdmouse_ops
);
1332 xen_be_register("qdisk", &xen_blkdev_ops
);
1333 xen_read_physmap(state
);
1337 error_report("xen hardware virtual machine initialisation failed");
1341 void destroy_hvm_domain(bool reboot
)
1346 xc_handle
= xen_xc_interface_open(0, 0, 0);
1347 if (xc_handle
== XC_HANDLER_INITIAL_VALUE
) {
1348 fprintf(stderr
, "Cannot acquire xenctrl handle\n");
1350 sts
= xc_domain_shutdown(xc_handle
, xen_domid
,
1351 reboot
? SHUTDOWN_reboot
: SHUTDOWN_poweroff
);
1353 fprintf(stderr
, "xc_domain_shutdown failed to issue %s, "
1354 "sts %d, %s\n", reboot
? "reboot" : "poweroff",
1355 sts
, strerror(errno
));
1357 fprintf(stderr
, "Issued domain %d %s\n", xen_domid
,
1358 reboot
? "reboot" : "poweroff");
1360 xc_interface_close(xc_handle
);
1364 void xen_register_framebuffer(MemoryRegion
*mr
)
1369 void xen_shutdown_fatal_error(const char *fmt
, ...)
1374 vfprintf(stderr
, fmt
, ap
);
1376 fprintf(stderr
, "Will destroy the domain.\n");
1377 /* destroy the domain */
1378 qemu_system_shutdown_request();
1381 void xen_modified_memory(ram_addr_t start
, ram_addr_t length
)
1383 if (unlikely(xen_in_migration
)) {
1385 ram_addr_t start_pfn
, nb_pages
;
1388 length
= TARGET_PAGE_SIZE
;
1390 start_pfn
= start
>> TARGET_PAGE_BITS
;
1391 nb_pages
= ((start
+ length
+ TARGET_PAGE_SIZE
- 1) >> TARGET_PAGE_BITS
)
1393 rc
= xc_hvm_modified_memory(xen_xc
, xen_domid
, start_pfn
, nb_pages
);
1396 "%s failed for "RAM_ADDR_FMT
" ("RAM_ADDR_FMT
"): %i, %s\n",
1397 __func__
, start
, nb_pages
, rc
, strerror(-rc
));
1402 void qmp_xen_set_global_dirty_log(bool enable
, Error
**errp
)
1405 memory_global_dirty_log_start();
1407 memory_global_dirty_log_stop();