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;
165 block_len = ram_size;
166 if (ram_size >= HVM_BELOW_4G_RAM_END) {
167 /* Xen does not allocate the memory continuously, and keep a hole at
168 * HVM_BELOW_4G_MMIO_START of HVM_BELOW_4G_MMIO_LENGTH
170 block_len += HVM_BELOW_4G_MMIO_LENGTH;
172 memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len);
173 *ram_memory_p = &ram_memory;
174 vmstate_register_ram_global(&ram_memory);
176 if (ram_size >= HVM_BELOW_4G_RAM_END) {
177 *above_4g_mem_size = ram_size - HVM_BELOW_4G_RAM_END;
178 *below_4g_mem_size = HVM_BELOW_4G_RAM_END;
180 *above_4g_mem_size = 0;
181 *below_4g_mem_size = ram_size;
184 memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
185 &ram_memory, 0, 0xa0000);
186 memory_region_add_subregion(sysmem, 0, &ram_640k);
187 /* Skip of the VGA IO memory space, it will be registered later by the VGA
190 * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
191 * the Options ROM, so it is registered here as RAM.
193 memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
194 &ram_memory, 0xc0000,
195 *below_4g_mem_size - 0xc0000);
196 memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
197 if (*above_4g_mem_size > 0) {
198 memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
199 &ram_memory, 0x100000000ULL,
201 memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
205 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
207 unsigned long nr_pfn;
211 if (runstate_check(RUN_STATE_INMIGRATE)) {
212 /* RAM already populated in Xen */
213 fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
214 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
215 __func__, size, ram_addr);
219 if (mr == &ram_memory) {
223 trace_xen_ram_alloc(ram_addr, size);
225 nr_pfn = size >> TARGET_PAGE_BITS;
226 pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
228 for (i = 0; i < nr_pfn; i++) {
229 pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
232 if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
233 hw_error("xen: failed to populate ram at " RAM_ADDR_FMT, ram_addr);
239 static XenPhysmap *get_physmapping(XenIOState *state,
240 hwaddr start_addr, ram_addr_t size)
242 XenPhysmap *physmap = NULL;
244 start_addr &= TARGET_PAGE_MASK;
246 QLIST_FOREACH(physmap, &state->physmap, list) {
247 if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
254 static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
255 ram_addr_t size, void *opaque)
257 hwaddr addr = start_addr & TARGET_PAGE_MASK;
258 XenIOState *xen_io_state = opaque;
259 XenPhysmap *physmap = NULL;
261 QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
262 if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
263 return physmap->start_addr;
270 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
271 static int xen_add_to_physmap(XenIOState *state,
275 hwaddr offset_within_region)
279 XenPhysmap *physmap = NULL;
280 hwaddr pfn, start_gpfn;
281 hwaddr phys_offset = memory_region_get_ram_addr(mr);
282 char path[80], value[17];
284 if (get_physmapping(state, start_addr, size)) {
291 /* Xen can only handle a single dirty log region for now and we want
292 * the linear framebuffer to be that region.
293 * Avoid tracking any regions that is not videoram and avoid tracking
294 * the legacy vga region. */
295 if (mr == framebuffer && start_addr > 0xbffff) {
301 DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
302 start_addr, start_addr + size);
304 pfn = phys_offset >> TARGET_PAGE_BITS;
305 start_gpfn = start_addr >> TARGET_PAGE_BITS;
306 for (i = 0; i < size >> TARGET_PAGE_BITS; i++) {
307 unsigned long idx = pfn + i;
308 xen_pfn_t gpfn = start_gpfn + i;
310 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
312 DPRINTF("add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
313 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
318 physmap = g_malloc(sizeof (XenPhysmap));
320 physmap->start_addr = start_addr;
321 physmap->size = size;
322 physmap->name = (char *)mr->name;
323 physmap->phys_offset = phys_offset;
325 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
327 xc_domain_pin_memory_cacheattr(xen_xc, xen_domid,
328 start_addr >> TARGET_PAGE_BITS,
329 (start_addr + size - 1) >> TARGET_PAGE_BITS,
330 XEN_DOMCTL_MEM_CACHEATTR_WB);
332 snprintf(path, sizeof(path),
333 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
334 xen_domid, (uint64_t)phys_offset);
335 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)start_addr);
336 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
339 snprintf(path, sizeof(path),
340 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
341 xen_domid, (uint64_t)phys_offset);
342 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)size);
343 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
347 snprintf(path, sizeof(path),
348 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
349 xen_domid, (uint64_t)phys_offset);
350 if (!xs_write(state->xenstore, 0, path, mr->name, strlen(mr->name))) {
358 static int xen_remove_from_physmap(XenIOState *state,
364 XenPhysmap *physmap = NULL;
365 hwaddr phys_offset = 0;
367 physmap = get_physmapping(state, start_addr, size);
368 if (physmap == NULL) {
372 phys_offset = physmap->phys_offset;
373 size = physmap->size;
375 DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
376 "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
378 size >>= TARGET_PAGE_BITS;
379 start_addr >>= TARGET_PAGE_BITS;
380 phys_offset >>= TARGET_PAGE_BITS;
381 for (i = 0; i < size; i++) {
382 unsigned long idx = start_addr + i;
383 xen_pfn_t gpfn = phys_offset + i;
385 rc = xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
387 fprintf(stderr, "add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
388 PRI_xen_pfn" failed: %d\n", idx, gpfn, rc);
393 QLIST_REMOVE(physmap, list);
394 if (state->log_for_dirtybit == physmap) {
395 state->log_for_dirtybit = NULL;
403 static int xen_add_to_physmap(XenIOState *state,
407 hwaddr offset_within_region)
412 static int xen_remove_from_physmap(XenIOState *state,
420 static void xen_set_memory(struct MemoryListener *listener,
421 MemoryRegionSection *section,
424 XenIOState *state = container_of(listener, XenIOState, memory_listener);
425 hwaddr start_addr = section->offset_within_address_space;
426 ram_addr_t size = int128_get64(section->size);
427 bool log_dirty = memory_region_is_logging(section->mr);
428 hvmmem_type_t mem_type;
430 if (!memory_region_is_ram(section->mr)) {
434 if (!(section->mr != &ram_memory
435 && ( (log_dirty && add) || (!log_dirty && !add)))) {
439 trace_xen_client_set_memory(start_addr, size, log_dirty);
441 start_addr &= TARGET_PAGE_MASK;
442 size = TARGET_PAGE_ALIGN(size);
445 if (!memory_region_is_rom(section->mr)) {
446 xen_add_to_physmap(state, start_addr, size,
447 section->mr, section->offset_within_region);
449 mem_type = HVMMEM_ram_ro;
450 if (xc_hvm_set_mem_type(xen_xc, xen_domid, mem_type,
451 start_addr >> TARGET_PAGE_BITS,
452 size >> TARGET_PAGE_BITS)) {
453 DPRINTF("xc_hvm_set_mem_type error, addr: "TARGET_FMT_plx"\n",
458 if (xen_remove_from_physmap(state, start_addr, size) < 0) {
459 DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
464 static void xen_region_add(MemoryListener *listener,
465 MemoryRegionSection *section)
467 memory_region_ref(section->mr);
468 xen_set_memory(listener, section, true);
471 static void xen_region_del(MemoryListener *listener,
472 MemoryRegionSection *section)
474 xen_set_memory(listener, section, false);
475 memory_region_unref(section->mr);
478 static void xen_sync_dirty_bitmap(XenIOState *state,
482 hwaddr npages = size >> TARGET_PAGE_BITS;
483 const int width = sizeof(unsigned long) * 8;
484 unsigned long bitmap[(npages + width - 1) / width];
486 const XenPhysmap *physmap = NULL;
488 physmap = get_physmapping(state, start_addr, size);
489 if (physmap == NULL) {
494 if (state->log_for_dirtybit == NULL) {
495 state->log_for_dirtybit = physmap;
496 } else if (state->log_for_dirtybit != physmap) {
497 /* Only one range for dirty bitmap can be tracked. */
501 rc = xc_hvm_track_dirty_vram(xen_xc, xen_domid,
502 start_addr >> TARGET_PAGE_BITS, npages,
505 if (rc != -ENODATA) {
506 memory_region_set_dirty(framebuffer, 0, size);
507 DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
508 ", 0x" TARGET_FMT_plx "): %s\n",
509 start_addr, start_addr + size, strerror(-rc));
514 for (i = 0; i < ARRAY_SIZE(bitmap); i++) {
515 unsigned long map = bitmap[i];
519 memory_region_set_dirty(framebuffer,
520 (i * width + j) * TARGET_PAGE_SIZE,
526 static void xen_log_start(MemoryListener *listener,
527 MemoryRegionSection *section)
529 XenIOState *state = container_of(listener, XenIOState, memory_listener);
531 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
532 int128_get64(section->size));
535 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section)
537 XenIOState *state = container_of(listener, XenIOState, memory_listener);
539 state->log_for_dirtybit = NULL;
540 /* Disable dirty bit tracking */
541 xc_hvm_track_dirty_vram(xen_xc, xen_domid, 0, 0, NULL);
544 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
546 XenIOState *state = container_of(listener, XenIOState, memory_listener);
548 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
549 int128_get64(section->size));
552 static void xen_log_global_start(MemoryListener *listener)
555 xen_in_migration = true;
559 static void xen_log_global_stop(MemoryListener *listener)
561 xen_in_migration = false;
564 static MemoryListener xen_memory_listener = {
565 .region_add = xen_region_add,
566 .region_del = xen_region_del,
567 .log_start = xen_log_start,
568 .log_stop = xen_log_stop,
569 .log_sync = xen_log_sync,
570 .log_global_start = xen_log_global_start,
571 .log_global_stop = xen_log_global_stop,
575 /* get the ioreq packets from share mem */
576 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
578 ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
580 if (req->state != STATE_IOREQ_READY) {
581 DPRINTF("I/O request not ready: "
582 "%x, ptr: %x, port: %"PRIx64", "
583 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
584 req->state, req->data_is_ptr, req->addr,
585 req->data, req->count, req->size);
589 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
591 req->state = STATE_IOREQ_INPROCESS;
595 /* use poll to get the port notification */
596 /* ioreq_vec--out,the */
597 /* retval--the number of ioreq packet */
598 static ioreq_t *cpu_get_ioreq(XenIOState *state)
603 port = xc_evtchn_pending(state->xce_handle);
604 if (port == state->bufioreq_local_port) {
605 timer_mod(state->buffered_io_timer,
606 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
611 for (i = 0; i < max_cpus; i++) {
612 if (state->ioreq_local_port[i] == port) {
618 hw_error("Fatal error while trying to get io event!\n");
621 /* unmask the wanted port again */
622 xc_evtchn_unmask(state->xce_handle, port);
624 /* get the io packet from shared memory */
625 state->send_vcpu = i;
626 return cpu_get_ioreq_from_shared_memory(state, i);
629 /* read error or read nothing */
633 static uint32_t do_inp(pio_addr_t addr, unsigned long size)
637 return cpu_inb(addr);
639 return cpu_inw(addr);
641 return cpu_inl(addr);
643 hw_error("inp: bad size: %04"FMT_pioaddr" %lx", addr, size);
647 static void do_outp(pio_addr_t addr,
648 unsigned long size, uint32_t val)
652 return cpu_outb(addr, val);
654 return cpu_outw(addr, val);
656 return cpu_outl(addr, val);
658 hw_error("outp: bad size: %04"FMT_pioaddr" %lx", addr, size);
663 * Helper functions which read/write an object from/to physical guest
664 * memory, as part of the implementation of an ioreq.
667 * cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
668 * val, req->size, 0/1)
669 * except without the integer overflow problems.
671 static void rw_phys_req_item(hwaddr addr,
672 ioreq_t *req, uint32_t i, void *val, int rw)
674 /* Do everything unsigned so overflow just results in a truncated result
675 * and accesses to undesired parts of guest memory, which is up
677 hwaddr offset = (hwaddr)req->size * i;
683 cpu_physical_memory_rw(addr, val, req->size, rw);
686 static inline void read_phys_req_item(hwaddr addr,
687 ioreq_t *req, uint32_t i, void *val)
689 rw_phys_req_item(addr, req, i, val, 0);
691 static inline void write_phys_req_item(hwaddr addr,
692 ioreq_t *req, uint32_t i, void *val)
694 rw_phys_req_item(addr, req, i, val, 1);
698 static void cpu_ioreq_pio(ioreq_t *req)
702 if (req->dir == IOREQ_READ) {
703 if (!req->data_is_ptr) {
704 req->data = do_inp(req->addr, req->size);
708 for (i = 0; i < req->count; i++) {
709 tmp = do_inp(req->addr, req->size);
710 write_phys_req_item(req->data, req, i, &tmp);
713 } else if (req->dir == IOREQ_WRITE) {
714 if (!req->data_is_ptr) {
715 do_outp(req->addr, req->size, req->data);
717 for (i = 0; i < req->count; i++) {
720 read_phys_req_item(req->data, req, i, &tmp);
721 do_outp(req->addr, req->size, tmp);
727 static void cpu_ioreq_move(ioreq_t *req)
731 if (!req->data_is_ptr) {
732 if (req->dir == IOREQ_READ) {
733 for (i = 0; i < req->count; i++) {
734 read_phys_req_item(req->addr, req, i, &req->data);
736 } else if (req->dir == IOREQ_WRITE) {
737 for (i = 0; i < req->count; i++) {
738 write_phys_req_item(req->addr, req, i, &req->data);
744 if (req->dir == IOREQ_READ) {
745 for (i = 0; i < req->count; i++) {
746 read_phys_req_item(req->addr, req, i, &tmp);
747 write_phys_req_item(req->data, req, i, &tmp);
749 } else if (req->dir == IOREQ_WRITE) {
750 for (i = 0; i < req->count; i++) {
751 read_phys_req_item(req->data, req, i, &tmp);
752 write_phys_req_item(req->addr, req, i, &tmp);
758 static void handle_ioreq(ioreq_t *req)
760 if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
761 (req->size < sizeof (target_ulong))) {
762 req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
769 case IOREQ_TYPE_COPY:
772 case IOREQ_TYPE_TIMEOFFSET:
774 case IOREQ_TYPE_INVALIDATE:
775 xen_invalidate_map_cache();
778 hw_error("Invalid ioreq type 0x%x\n", req->type);
782 static int handle_buffered_iopage(XenIOState *state)
784 buf_ioreq_t *buf_req = NULL;
788 if (!state->buffered_io_page) {
792 memset(&req, 0x00, sizeof(req));
794 while (state->buffered_io_page->read_pointer != state->buffered_io_page->write_pointer) {
795 buf_req = &state->buffered_io_page->buf_ioreq[
796 state->buffered_io_page->read_pointer % IOREQ_BUFFER_SLOT_NUM];
797 req.size = 1UL << buf_req->size;
799 req.addr = buf_req->addr;
800 req.data = buf_req->data;
801 req.state = STATE_IOREQ_READY;
802 req.dir = buf_req->dir;
804 req.type = buf_req->type;
806 qw = (req.size == 8);
808 buf_req = &state->buffered_io_page->buf_ioreq[
809 (state->buffered_io_page->read_pointer + 1) % IOREQ_BUFFER_SLOT_NUM];
810 req.data |= ((uint64_t)buf_req->data) << 32;
816 state->buffered_io_page->read_pointer += qw ? 2 : 1;
822 static void handle_buffered_io(void *opaque)
824 XenIOState *state = opaque;
826 if (handle_buffered_iopage(state)) {
827 timer_mod(state->buffered_io_timer,
828 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
830 timer_del(state->buffered_io_timer);
831 xc_evtchn_unmask(state->xce_handle, state->bufioreq_local_port);
835 static void cpu_handle_ioreq(void *opaque)
837 XenIOState *state = opaque;
838 ioreq_t *req = cpu_get_ioreq(state);
840 handle_buffered_iopage(state);
844 if (req->state != STATE_IOREQ_INPROCESS) {
845 fprintf(stderr, "Badness in I/O request ... not in service?!: "
846 "%x, ptr: %x, port: %"PRIx64", "
847 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
848 req->state, req->data_is_ptr, req->addr,
849 req->data, req->count, req->size);
850 destroy_hvm_domain(false);
854 xen_wmb(); /* Update ioreq contents /then/ update state. */
857 * We do this before we send the response so that the tools
858 * have the opportunity to pick up on the reset before the
859 * guest resumes and does a hlt with interrupts disabled which
860 * causes Xen to powerdown the domain.
862 if (runstate_is_running()) {
863 if (qemu_shutdown_requested_get()) {
864 destroy_hvm_domain(false);
866 if (qemu_reset_requested_get()) {
867 qemu_system_reset(VMRESET_REPORT);
868 destroy_hvm_domain(true);
872 req->state = STATE_IORESP_READY;
873 xc_evtchn_notify(state->xce_handle, state->ioreq_local_port[state->send_vcpu]);
877 static void xen_main_loop_prepare(XenIOState *state)
881 if (state->xce_handle != XC_HANDLER_INITIAL_VALUE) {
882 evtchn_fd = xc_evtchn_fd(state->xce_handle);
885 state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
888 if (evtchn_fd != -1) {
889 qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
894 static void xen_hvm_change_state_handler(void *opaque, int running,
897 XenIOState *xstate = opaque;
899 xen_main_loop_prepare(xstate);
903 static void xen_exit_notifier(Notifier *n, void *data)
905 XenIOState *state = container_of(n, XenIOState, exit);
907 xc_evtchn_close(state->xce_handle);
908 xs_daemon_close(state->xenstore);
911 static void xen_read_physmap(XenIOState *state)
913 XenPhysmap *physmap = NULL;
914 unsigned int len, num, i;
915 char path[80], *value = NULL;
916 char **entries = NULL;
918 snprintf(path, sizeof(path),
919 "/local/domain/0/device-model/%d/physmap", xen_domid);
920 entries = xs_directory(state->xenstore, 0, path, &num);
924 for (i = 0; i < num; i++) {
925 physmap = g_malloc(sizeof (XenPhysmap));
926 physmap->phys_offset = strtoull(entries[i], NULL, 16);
927 snprintf(path, sizeof(path),
928 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
929 xen_domid, entries[i]);
930 value = xs_read(state->xenstore, 0, path, &len);
935 physmap->start_addr = strtoull(value, NULL, 16);
938 snprintf(path, sizeof(path),
939 "/local/domain/0/device-model/%d/physmap/%s/size",
940 xen_domid, entries[i]);
941 value = xs_read(state->xenstore, 0, path, &len);
946 physmap->size = strtoull(value, NULL, 16);
949 snprintf(path, sizeof(path),
950 "/local/domain/0/device-model/%d/physmap/%s/name",
951 xen_domid, entries[i]);
952 physmap->name = xs_read(state->xenstore, 0, path, &len);
954 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
959 static void xen_wakeup_notifier(Notifier *notifier, void *data)
961 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
964 int xen_hvm_init(ram_addr_t *below_4g_mem_size, ram_addr_t *above_4g_mem_size,
965 MemoryRegion **ram_memory)
968 unsigned long ioreq_pfn;
969 unsigned long bufioreq_evtchn;
972 state = g_malloc0(sizeof (XenIOState));
974 state->xce_handle = xen_xc_evtchn_open(NULL, 0);
975 if (state->xce_handle == XC_HANDLER_INITIAL_VALUE) {
976 perror("xen: event channel open");
981 state->xenstore = xs_daemon_open();
982 if (state->xenstore == NULL) {
983 perror("xen: xenstore open");
988 state->exit.notify = xen_exit_notifier;
989 qemu_add_exit_notifier(&state->exit);
991 state->suspend.notify = xen_suspend_notifier;
992 qemu_register_suspend_notifier(&state->suspend);
994 state->wakeup.notify = xen_wakeup_notifier;
995 qemu_register_wakeup_notifier(&state->wakeup);
997 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_IOREQ_PFN, &ioreq_pfn);
998 DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
999 state->shared_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1000 PROT_READ|PROT_WRITE, ioreq_pfn);
1001 if (state->shared_page == NULL) {
1002 hw_error("map shared IO page returned error %d handle=" XC_INTERFACE_FMT,
1006 xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_PFN, &ioreq_pfn);
1007 DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn);
1008 state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1009 PROT_READ|PROT_WRITE, ioreq_pfn);
1010 if (state->buffered_io_page == NULL) {
1011 hw_error("map buffered IO page returned error %d", errno);
1014 state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1016 /* FIXME: how about if we overflow the page here? */
1017 for (i = 0; i < max_cpus; i++) {
1018 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1019 xen_vcpu_eport(state->shared_page, i));
1021 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1024 state->ioreq_local_port[i] = rc;
1027 rc = xc_get_hvm_param(xen_xc, xen_domid, HVM_PARAM_BUFIOREQ_EVTCHN,
1030 fprintf(stderr, "failed to get HVM_PARAM_BUFIOREQ_EVTCHN\n");
1033 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1034 (uint32_t)bufioreq_evtchn);
1036 fprintf(stderr, "bind interdomain ioctl error %d\n", errno);
1039 state->bufioreq_local_port = rc;
1041 /* Init RAM management */
1042 xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1043 xen_ram_init(below_4g_mem_size, above_4g_mem_size, ram_size, ram_memory);
1045 qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1047 state->memory_listener = xen_memory_listener;
1048 QLIST_INIT(&state->physmap);
1049 memory_listener_register(&state->memory_listener, &address_space_memory);
1050 state->log_for_dirtybit = NULL;
1052 /* Initialize backend core & drivers */
1053 if (xen_be_init() != 0) {
1054 fprintf(stderr, "%s: xen backend core setup failed\n", __FUNCTION__);
1057 xen_be_register("console", &xen_console_ops);
1058 xen_be_register("vkbd", &xen_kbdmouse_ops);
1059 xen_be_register("qdisk", &xen_blkdev_ops);
1060 xen_read_physmap(state);
1065 void destroy_hvm_domain(bool reboot)
1070 xc_handle = xen_xc_interface_open(0, 0, 0);
1071 if (xc_handle == XC_HANDLER_INITIAL_VALUE) {
1072 fprintf(stderr, "Cannot acquire xenctrl handle\n");
1074 sts = xc_domain_shutdown(xc_handle, xen_domid,
1075 reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
1077 fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1078 "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1079 sts, strerror(errno));
1081 fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1082 reboot ? "reboot" : "poweroff");
1084 xc_interface_close(xc_handle);
1088 void xen_register_framebuffer(MemoryRegion *mr)
1093 void xen_shutdown_fatal_error(const char *fmt, ...)
1098 vfprintf(stderr, fmt, ap);
1100 fprintf(stderr, "Will destroy the domain.\n");
1101 /* destroy the domain */
1102 qemu_system_shutdown_request();
1105 void xen_modified_memory(ram_addr_t start, ram_addr_t length)
1107 if (unlikely(xen_in_migration)) {
1109 ram_addr_t start_pfn, nb_pages;
1112 length = TARGET_PAGE_SIZE;
1114 start_pfn = start >> TARGET_PAGE_BITS;
1115 nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1117 rc = xc_hvm_modified_memory(xen_xc, xen_domid, start_pfn, nb_pages);
1120 "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1121 __func__, start, nb_pages, rc, strerror(-rc));
1126 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1129 memory_global_dirty_log_start();
1131 memory_global_dirty_log_stop();