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/range.h"
21 #include "sysemu/xen-mapcache.h"
23 #include "exec/address-spaces.h"
25 #include <xen/hvm/ioreq.h>
26 #include <xen/hvm/params.h>
27 #include <xen/hvm/e820.h>
29 //#define DEBUG_XEN_HVM
32 #define DPRINTF(fmt, ...) \
33 do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
35 #define DPRINTF(fmt, ...) \
39 static MemoryRegion ram_memory, ram_640k, ram_lo, ram_hi;
40 static MemoryRegion *framebuffer;
41 static bool xen_in_migration;
43 /* Compatibility with older version */
44 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
45 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
47 return shared_page->vcpu_iodata[i].vp_eport;
49 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
51 return &shared_page->vcpu_iodata[vcpu].vp_ioreq;
53 # define FMT_ioreq_size PRIx64
55 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
57 return shared_page->vcpu_ioreq[i].vp_eport;
59 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
61 return &shared_page->vcpu_ioreq[vcpu];
63 # define FMT_ioreq_size "u"
65 #ifndef HVM_PARAM_BUFIOREQ_EVTCHN
66 #define HVM_PARAM_BUFIOREQ_EVTCHN 26
69 #define BUFFER_IO_MAX_DELAY 100
71 typedef struct XenPhysmap {
77 QLIST_ENTRY(XenPhysmap) list;
80 typedef struct XenIOState {
81 shared_iopage_t *shared_page;
82 buffered_iopage_t *buffered_io_page;
83 QEMUTimer *buffered_io_timer;
84 /* the evtchn port for polling the notification, */
85 evtchn_port_t *ioreq_local_port;
86 /* evtchn local port for buffered io */
87 evtchn_port_t bufioreq_local_port;
88 /* the evtchn fd for polling */
90 /* which vcpu we are serving */
93 struct xs_handle *xenstore;
94 MemoryListener memory_listener;
95 QLIST_HEAD(, XenPhysmap) physmap;
96 hwaddr free_phys_offset;
97 const XenPhysmap *log_for_dirtybit;
104 /* Xen specific function for piix pci */
106 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
108 return irq_num + ((pci_dev->devfn >> 3) << 2);
111 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
113 xc_hvm_set_pci_intx_level(xen_xc, xen_domid, 0, 0, irq_num >> 2,
117 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
121 /* Scan for updates to PCI link routes (0x60-0x63). */
122 for (i = 0; i < len; i++) {
123 uint8_t v = (val >> (8 * i)) & 0xff;
128 if (((address + i) >= 0x60) && ((address + i) <= 0x63)) {
129 xc_hvm_set_pci_link_route(xen_xc, xen_domid, address + i - 0x60, v);
134 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
136 xen_xc_hvm_inject_msi(xen_xc, xen_domid, addr, data);
139 static void xen_suspend_notifier(Notifier *notifier, void *data)
141 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
144 /* Xen Interrupt Controller */
146 static void xen_set_irq(void *opaque, int irq, int level)
148 xc_hvm_set_isa_irq_level(xen_xc, xen_domid, irq, level);
151 qemu_irq *xen_interrupt_controller_init(void)
153 return qemu_allocate_irqs(xen_set_irq, NULL, 16);
158 static void xen_ram_init(ram_addr_t *below_4g_mem_size,
159 ram_addr_t *above_4g_mem_size,
160 ram_addr_t ram_size, MemoryRegion **ram_memory_p)
162 MemoryRegion *sysmem = get_system_memory();
163 ram_addr_t block_len;
164 uint64_t user_lowmem = object_property_get_int(qdev_get_machine(),
165 PC_MACHINE_MAX_RAM_BELOW_4G,
168 /* Handle the machine opt max-ram-below-4g. It is basically doing
169 * min(xen limit, user limit).
171 if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
172 user_lowmem = HVM_BELOW_4G_RAM_END;
175 if (ram_size >= user_lowmem) {
176 *above_4g_mem_size = ram_size - user_lowmem;
177 *below_4g_mem_size = user_lowmem;
179 *above_4g_mem_size = 0;
180 *below_4g_mem_size = ram_size;
182 if (!*above_4g_mem_size) {
183 block_len = ram_size;
186 * Xen does not allocate the memory continuously, it keeps a
187 * hole of the size computed above or passed in.
189 block_len = (1ULL << 32) + *above_4g_mem_size;
191 memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
193 *ram_memory_p = &ram_memory;
194 vmstate_register_ram_global(&ram_memory);
196 memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
197 &ram_memory, 0, 0xa0000);
198 memory_region_add_subregion(sysmem, 0, &ram_640k);
199 /* Skip of the VGA IO memory space, it will be registered later by the VGA
202 * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
203 * the Options ROM, so it is registered here as RAM.
205 memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
206 &ram_memory, 0xc0000,
207 *below_4g_mem_size - 0xc0000);
208 memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
209 if (*above_4g_mem_size > 0) {
210 memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
211 &ram_memory, 0x100000000ULL,
213 memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
217 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
219 unsigned long nr_pfn;
223 if (runstate_check(RUN_STATE_INMIGRATE)) {
224 /* RAM already populated in Xen */
225 fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
226 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
227 __func__, size, ram_addr);
231 if (mr == &ram_memory) {
235 trace_xen_ram_alloc(ram_addr, size);
237 nr_pfn = size >> TARGET_PAGE_BITS;
238 pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
240 for (i = 0; i < nr_pfn; i++) {
241 pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
244 if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
245 hw_error("xen: failed to populate ram at " RAM_ADDR_FMT, ram_addr);
251 static XenPhysmap *get_physmapping(XenIOState *state,
252 hwaddr start_addr, ram_addr_t size)
254 XenPhysmap *physmap = NULL;
256 start_addr &= TARGET_PAGE_MASK;
258 QLIST_FOREACH(physmap, &state->physmap, list) {
259 if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
266 static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
267 ram_addr_t size, void *opaque)
269 hwaddr addr = start_addr & TARGET_PAGE_MASK;
270 XenIOState *xen_io_state = opaque;
271 XenPhysmap *physmap = NULL;
273 QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
274 if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
275 return physmap->start_addr;
282 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
283 static int xen_add_to_physmap(XenIOState *state,
287 hwaddr offset_within_region)
291 XenPhysmap *physmap = NULL;
292 hwaddr pfn, start_gpfn;
293 hwaddr phys_offset = memory_region_get_ram_addr(mr);
294 char path[80], value[17];
297 if (get_physmapping(state, start_addr, size)) {
304 /* Xen can only handle a single dirty log region for now and we want
305 * the linear framebuffer to be that region.
306 * Avoid tracking any regions that is not videoram and avoid tracking
307 * the legacy vga region. */
308 if (mr == framebuffer && start_addr > 0xbffff) {
314 DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
315 start_addr, start_addr + size);
317 pfn = phys_offset >> TARGET_PAGE_BITS;
318 start_gpfn = start_addr >> TARGET_PAGE_BITS;
319 for (i = 0; i < size >> TARGET_PAGE_BITS; i++) {
320 unsigned long idx = pfn + i;
321 xen_pfn_t gpfn = start_gpfn + i;
323 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
325 DPRINTF("add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
326 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
331 mr_name = memory_region_name(mr);
333 physmap = g_malloc(sizeof (XenPhysmap));
335 physmap->start_addr = start_addr;
336 physmap->size = size;
337 physmap->name = mr_name;
338 physmap->phys_offset = phys_offset;
340 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
342 xc_domain_pin_memory_cacheattr(xen_xc, xen_domid,
343 start_addr >> TARGET_PAGE_BITS,
344 (start_addr + size - 1) >> TARGET_PAGE_BITS,
345 XEN_DOMCTL_MEM_CACHEATTR_WB);
347 snprintf(path, sizeof(path),
348 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
349 xen_domid, (uint64_t)phys_offset);
350 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)start_addr);
351 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
354 snprintf(path, sizeof(path),
355 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
356 xen_domid, (uint64_t)phys_offset);
357 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)size);
358 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
362 snprintf(path, sizeof(path),
363 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
364 xen_domid, (uint64_t)phys_offset);
365 if (!xs_write(state->xenstore, 0, path, mr_name, strlen(mr_name))) {
373 static int xen_remove_from_physmap(XenIOState *state,
379 XenPhysmap *physmap = NULL;
380 hwaddr phys_offset = 0;
382 physmap = get_physmapping(state, start_addr, size);
383 if (physmap == NULL) {
387 phys_offset = physmap->phys_offset;
388 size = physmap->size;
390 DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
391 "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
393 size >>= TARGET_PAGE_BITS;
394 start_addr >>= TARGET_PAGE_BITS;
395 phys_offset >>= TARGET_PAGE_BITS;
396 for (i = 0; i < size; i++) {
397 xen_pfn_t idx = start_addr + i;
398 xen_pfn_t gpfn = phys_offset + i;
400 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
402 fprintf(stderr, "add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
403 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
408 QLIST_REMOVE(physmap, list);
409 if (state->log_for_dirtybit == physmap) {
410 state->log_for_dirtybit = NULL;
418 static int xen_add_to_physmap(XenIOState *state,
422 hwaddr offset_within_region)
427 static int xen_remove_from_physmap(XenIOState *state,
435 static void xen_set_memory(struct MemoryListener *listener,
436 MemoryRegionSection *section,
439 XenIOState *state = container_of(listener, XenIOState, memory_listener);
440 hwaddr start_addr = section->offset_within_address_space;
441 ram_addr_t size = int128_get64(section->size);
442 bool log_dirty = memory_region_is_logging(section->mr);
443 hvmmem_type_t mem_type;
445 if (!memory_region_is_ram(section->mr)) {
449 if (!(section->mr != &ram_memory
450 && ( (log_dirty && add) || (!log_dirty && !add)))) {
454 trace_xen_client_set_memory(start_addr, size, log_dirty);
456 start_addr &= TARGET_PAGE_MASK;
457 size = TARGET_PAGE_ALIGN(size);
460 if (!memory_region_is_rom(section->mr)) {
461 xen_add_to_physmap(state, start_addr, size,
462 section->mr, section->offset_within_region);
464 mem_type = HVMMEM_ram_ro;
465 if (xc_hvm_set_mem_type(xen_xc, xen_domid, mem_type,
466 start_addr >> TARGET_PAGE_BITS,
467 size >> TARGET_PAGE_BITS)) {
468 DPRINTF("xc_hvm_set_mem_type error, addr: "TARGET_FMT_plx"\n",
473 if (xen_remove_from_physmap(state, start_addr, size) < 0) {
474 DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
479 static void xen_region_add(MemoryListener *listener,
480 MemoryRegionSection *section)
482 memory_region_ref(section->mr);
483 xen_set_memory(listener, section, true);
486 static void xen_region_del(MemoryListener *listener,
487 MemoryRegionSection *section)
489 xen_set_memory(listener, section, false);
490 memory_region_unref(section->mr);
493 static void xen_sync_dirty_bitmap(XenIOState *state,
497 hwaddr npages = size >> TARGET_PAGE_BITS;
498 const int width = sizeof(unsigned long) * 8;
499 unsigned long bitmap[(npages + width - 1) / width];
501 const XenPhysmap *physmap = NULL;
503 physmap = get_physmapping(state, start_addr, size);
504 if (physmap == NULL) {
509 if (state->log_for_dirtybit == NULL) {
510 state->log_for_dirtybit = physmap;
511 } else if (state->log_for_dirtybit != physmap) {
512 /* Only one range for dirty bitmap can be tracked. */
516 rc = xc_hvm_track_dirty_vram(xen_xc, xen_domid,
517 start_addr >> TARGET_PAGE_BITS, npages,
521 #define ENODATA ENOENT
523 if (errno == ENODATA) {
524 memory_region_set_dirty(framebuffer, 0, size);
525 DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
526 ", 0x" TARGET_FMT_plx "): %s\n",
527 start_addr, start_addr + size, strerror(errno));
532 for (i = 0; i < ARRAY_SIZE(bitmap); i++) {
533 unsigned long map = bitmap[i];
537 memory_region_set_dirty(framebuffer,
538 (i * width + j) * TARGET_PAGE_SIZE,
544 static void xen_log_start(MemoryListener *listener,
545 MemoryRegionSection *section)
547 XenIOState *state = container_of(listener, XenIOState, memory_listener);
549 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
550 int128_get64(section->size));
553 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section)
555 XenIOState *state = container_of(listener, XenIOState, memory_listener);
557 state->log_for_dirtybit = NULL;
558 /* Disable dirty bit tracking */
559 xc_hvm_track_dirty_vram(xen_xc, xen_domid, 0, 0, NULL);
562 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
564 XenIOState *state = container_of(listener, XenIOState, memory_listener);
566 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
567 int128_get64(section->size));
570 static void xen_log_global_start(MemoryListener *listener)
573 xen_in_migration = true;
577 static void xen_log_global_stop(MemoryListener *listener)
579 xen_in_migration = false;
582 static MemoryListener xen_memory_listener = {
583 .region_add = xen_region_add,
584 .region_del = xen_region_del,
585 .log_start = xen_log_start,
586 .log_stop = xen_log_stop,
587 .log_sync = xen_log_sync,
588 .log_global_start = xen_log_global_start,
589 .log_global_stop = xen_log_global_stop,
593 /* get the ioreq packets from share mem */
594 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
596 ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
598 if (req->state != STATE_IOREQ_READY) {
599 DPRINTF("I/O request not ready: "
600 "%x, ptr: %x, port: %"PRIx64", "
601 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
602 req->state, req->data_is_ptr, req->addr,
603 req->data, req->count, req->size);
607 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
609 req->state = STATE_IOREQ_INPROCESS;
613 /* use poll to get the port notification */
614 /* ioreq_vec--out,the */
615 /* retval--the number of ioreq packet */
616 static ioreq_t *cpu_get_ioreq(XenIOState *state)
621 port = xc_evtchn_pending(state->xce_handle);
622 if (port == state->bufioreq_local_port) {
623 timer_mod(state->buffered_io_timer,
624 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
629 for (i = 0; i < max_cpus; i++) {
630 if (state->ioreq_local_port[i] == port) {
636 hw_error("Fatal error while trying to get io event!\n");
639 /* unmask the wanted port again */
640 xc_evtchn_unmask(state->xce_handle, port);
642 /* get the io packet from shared memory */
643 state->send_vcpu = i;
644 return cpu_get_ioreq_from_shared_memory(state, i);
647 /* read error or read nothing */
651 static uint32_t do_inp(pio_addr_t addr, unsigned long size)
655 return cpu_inb(addr);
657 return cpu_inw(addr);
659 return cpu_inl(addr);
661 hw_error("inp: bad size: %04"FMT_pioaddr" %lx", addr, size);
665 static void do_outp(pio_addr_t addr,
666 unsigned long size, uint32_t val)
670 return cpu_outb(addr, val);
672 return cpu_outw(addr, val);
674 return cpu_outl(addr, val);
676 hw_error("outp: bad size: %04"FMT_pioaddr" %lx", addr, size);
681 * Helper functions which read/write an object from/to physical guest
682 * memory, as part of the implementation of an ioreq.
685 * cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
686 * val, req->size, 0/1)
687 * except without the integer overflow problems.
689 static void rw_phys_req_item(hwaddr addr,
690 ioreq_t *req, uint32_t i, void *val, int rw)
692 /* Do everything unsigned so overflow just results in a truncated result
693 * and accesses to undesired parts of guest memory, which is up
695 hwaddr offset = (hwaddr)req->size * i;
701 cpu_physical_memory_rw(addr, val, req->size, rw);
704 static inline void read_phys_req_item(hwaddr addr,
705 ioreq_t *req, uint32_t i, void *val)
707 rw_phys_req_item(addr, req, i, val, 0);
709 static inline void write_phys_req_item(hwaddr addr,
710 ioreq_t *req, uint32_t i, void *val)
712 rw_phys_req_item(addr, req, i, val, 1);
716 static void cpu_ioreq_pio(ioreq_t *req)
720 if (req->dir == IOREQ_READ) {
721 if (!req->data_is_ptr) {
722 req->data = do_inp(req->addr, req->size);
726 for (i = 0; i < req->count; i++) {
727 tmp = do_inp(req->addr, req->size);
728 write_phys_req_item(req->data, req, i, &tmp);
731 } else if (req->dir == IOREQ_WRITE) {
732 if (!req->data_is_ptr) {
733 do_outp(req->addr, req->size, req->data);
735 for (i = 0; i < req->count; i++) {
738 read_phys_req_item(req->data, req, i, &tmp);
739 do_outp(req->addr, req->size, tmp);
745 static void cpu_ioreq_move(ioreq_t *req)
749 if (!req->data_is_ptr) {
750 if (req->dir == IOREQ_READ) {
751 for (i = 0; i < req->count; i++) {
752 read_phys_req_item(req->addr, req, i, &req->data);
754 } else if (req->dir == IOREQ_WRITE) {
755 for (i = 0; i < req->count; i++) {
756 write_phys_req_item(req->addr, req, i, &req->data);
762 if (req->dir == IOREQ_READ) {
763 for (i = 0; i < req->count; i++) {
764 read_phys_req_item(req->addr, req, i, &tmp);
765 write_phys_req_item(req->data, req, i, &tmp);
767 } else if (req->dir == IOREQ_WRITE) {
768 for (i = 0; i < req->count; i++) {
769 read_phys_req_item(req->data, req, i, &tmp);
770 write_phys_req_item(req->addr, req, i, &tmp);
776 static void handle_ioreq(ioreq_t *req)
778 if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
779 (req->size < sizeof (target_ulong))) {
780 req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
787 case IOREQ_TYPE_COPY:
790 case IOREQ_TYPE_TIMEOFFSET:
792 case IOREQ_TYPE_INVALIDATE:
793 xen_invalidate_map_cache();
796 hw_error("Invalid ioreq type 0x%x\n", req->type);
800 static int handle_buffered_iopage(XenIOState *state)
802 buf_ioreq_t *buf_req = NULL;
806 if (!state->buffered_io_page) {
810 memset(&req, 0x00, sizeof(req));
812 while (state->buffered_io_page->read_pointer != state->buffered_io_page->write_pointer) {
813 buf_req = &state->buffered_io_page->buf_ioreq[
814 state->buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
815 req.size = 1UL << buf_req->size;
817 req.addr = buf_req->addr;
818 req.data = buf_req->data;
819 req.state = STATE_IOREQ_READY;
820 req.dir = buf_req->dir;
822 req.type = buf_req->type;
824 qw = (req.size == 8);
826 buf_req = &state->buffered_io_page->buf_ioreq[
827 (state->buffered_io_page->read_pointer + 1) % IOREQ_BUFFER_SLOT_NUM];
828 req.data |= ((uint64_t)buf_req->data) << 32;
834 state->buffered_io_page->read_pointer += qw ? 2 : 1;
840 static void handle_buffered_io(void *opaque)
842 XenIOState *state = opaque;
844 if (handle_buffered_iopage(state)) {
845 timer_mod(state->buffered_io_timer,
846 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
848 timer_del(state->buffered_io_timer);
849 xc_evtchn_unmask(state->xce_handle, state->bufioreq_local_port);
853 static void cpu_handle_ioreq(void *opaque)
855 XenIOState *state = opaque;
856 ioreq_t *req = cpu_get_ioreq(state);
858 handle_buffered_iopage(state);
862 if (req->state != STATE_IOREQ_INPROCESS) {
863 fprintf(stderr, "Badness in I/O request ... not in service?!: "
864 "%x, ptr: %x, port: %"PRIx64", "
865 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
866 req->state, req->data_is_ptr, req->addr,
867 req->data, req->count, req->size);
868 destroy_hvm_domain(false);
872 xen_wmb(); /* Update ioreq contents /then/ update state. */
875 * We do this before we send the response so that the tools
876 * have the opportunity to pick up on the reset before the
877 * guest resumes and does a hlt with interrupts disabled which
878 * causes Xen to powerdown the domain.
880 if (runstate_is_running()) {
881 if (qemu_shutdown_requested_get()) {
882 destroy_hvm_domain(false);
884 if (qemu_reset_requested_get()) {
885 qemu_system_reset(VMRESET_REPORT);
886 destroy_hvm_domain(true);
890 req->state = STATE_IORESP_READY;
891 xc_evtchn_notify(state->xce_handle, state->ioreq_local_port[state->send_vcpu]);
895 static void xen_main_loop_prepare(XenIOState *state)
899 if (state->xce_handle != XC_HANDLER_INITIAL_VALUE) {
900 evtchn_fd = xc_evtchn_fd(state->xce_handle);
903 state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
906 if (evtchn_fd != -1) {
907 qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
912 static void xen_hvm_change_state_handler(void *opaque, int running,
915 XenIOState *xstate = opaque;
917 xen_main_loop_prepare(xstate);
921 static void xen_exit_notifier(Notifier *n, void *data)
923 XenIOState *state = container_of(n, XenIOState, exit);
925 xc_evtchn_close(state->xce_handle);
926 xs_daemon_close(state->xenstore);
929 static void xen_read_physmap(XenIOState *state)
931 XenPhysmap *physmap = NULL;
932 unsigned int len, num, i;
933 char path[80], *value = NULL;
934 char **entries = NULL;
936 snprintf(path, sizeof(path),
937 "/local/domain/0/device-model/%d/physmap", xen_domid);
938 entries = xs_directory(state->xenstore, 0, path, &num);
942 for (i = 0; i < num; i++) {
943 physmap = g_malloc(sizeof (XenPhysmap));
944 physmap->phys_offset = strtoull(entries[i], NULL, 16);
945 snprintf(path, sizeof(path),
946 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
947 xen_domid, entries[i]);
948 value = xs_read(state->xenstore, 0, path, &len);
953 physmap->start_addr = strtoull(value, NULL, 16);
956 snprintf(path, sizeof(path),
957 "/local/domain/0/device-model/%d/physmap/%s/size",
958 xen_domid, entries[i]);
959 value = xs_read(state->xenstore, 0, path, &len);
964 physmap->size = strtoull(value, NULL, 16);
967 snprintf(path, sizeof(path),
968 "/local/domain/0/device-model/%d/physmap/%s/name",
969 xen_domid, entries[i]);
970 physmap->name = xs_read(state->xenstore, 0, path, &len);
972 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
977 static void xen_wakeup_notifier(Notifier *notifier, void *data)
979 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
982 /* return 0 means OK, or -1 means critical issue -- will exit(1) */
983 int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
984 MemoryRegion **ram_memory)
987 unsigned long ioreq_pfn;
988 unsigned long bufioreq_evtchn;
991 state = g_malloc0(sizeof (XenIOState));
993 state->xce_handle = xen_xc_evtchn_open(NULL, 0);
994 if (state->xce_handle == XC_HANDLER_INITIAL_VALUE) {
995 perror("xen: event channel open");
999 state->xenstore = xs_daemon_open();
1000 if (state->xenstore == NULL) {
1001 perror("xen: xenstore open");
1005 state->exit.notify = xen_exit_notifier;
1006 qemu_add_exit_notifier(&state->exit);
1008 state->suspend.notify = xen_suspend_notifier;
1009 qemu_register_suspend_notifier(&state->suspend);
1011 state->wakeup.notify = xen_wakeup_notifier;
1012 qemu_register_wakeup_notifier(&state->wakeup);
1014 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_IOREQ_PFN, &ioreq_pfn);
1015 DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1016 state->shared_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1017 PROT_READ|PROT_WRITE, ioreq_pfn);
1018 if (state->shared_page == NULL) {
1019 hw_error("map shared IO page returned error %d handle=" XC_INTERFACE_FMT,
1023 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
1024 DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
1025 state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1026 PROT_READ|PROT_WRITE, ioreq_pfn);
1027 if (state->buffered_io_page == NULL) {
1028 hw_error("map buffered IO page returned error %d", errno);
1031 state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1033 /* FIXME: how about if we overflow the page here? */
1034 for (i = 0; i < max_cpus; i++) {
1035 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1036 xen_vcpu_eport(state->shared_page, i));
1038 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1041 state->ioreq_local_port[i] = rc;
1044 rc = xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_EVTCHN,
1047 fprintf(stderr, "failed to get HVM_PARAM_BUFIOREQ_EVTCHN\n");
1050 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1051 (uint32_t)bufioreq_evtchn);
1053 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1056 state->bufioreq_local_port = rc;
1058 /* Init RAM management */
1059 xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1060 xen_ram_init(below_4g_mem_size, above_4g_mem_size, ram_size, ram_memory);
1062 qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1064 state->memory_listener = xen_memory_listener;
1065 QLIST_INIT(&state->physmap);
1066 memory_listener_register(&state->memory_listener, &address_space_memory);
1067 state->log_for_dirtybit = NULL;
1069 /* Initialize backend core & drivers */
1070 if (xen_be_init() != 0) {
1071 fprintf(stderr, "%s: xen backend core setup failed\n", __FUNCTION__);
1074 xen_be_register("console", &xen_console_ops);
1075 xen_be_register("vkbd", &xen_kbdmouse_ops);
1076 xen_be_register("qdisk", &xen_blkdev_ops);
1077 xen_read_physmap(state);
1082 void destroy_hvm_domain(bool reboot)
1087 xc_handle = xen_xc_interface_open(0, 0, 0);
1088 if (xc_handle == XC_HANDLER_INITIAL_VALUE) {
1089 fprintf(stderr, "Cannot acquire xenctrl handle\n");
1091 sts = xc_domain_shutdown(xc_handle, xen_domid,
1092 reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
1094 fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1095 "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1096 sts, strerror(errno));
1098 fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1099 reboot ? "reboot" : "poweroff");
1101 xc_interface_close(xc_handle);
1105 void xen_register_framebuffer(MemoryRegion *mr)
1110 void xen_shutdown_fatal_error(const char *fmt, ...)
1115 vfprintf(stderr, fmt, ap);
1117 fprintf(stderr, "Will destroy the domain.\n");
1118 /* destroy the domain */
1119 qemu_system_shutdown_request();
1122 void xen_modified_memory(ram_addr_t start, ram_addr_t length)
1124 if (unlikely(xen_in_migration)) {
1126 ram_addr_t start_pfn, nb_pages;
1129 length = TARGET_PAGE_SIZE;
1131 start_pfn = start >> TARGET_PAGE_BITS;
1132 nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1134 rc = xc_hvm_modified_memory(xen_xc, xen_domid, start_pfn, nb_pages);
1137 "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1138 __func__, start, nb_pages, rc, strerror(-rc));
1143 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1146 memory_global_dirty_log_start();
1148 memory_global_dirty_log_stop();