block: define 'clock_type' for the accounting code
[qemu.git] / xen-hvm.c
blob3d78a0c529fa485609f84afe3f6ab4b3b3bbf83a
1 /*
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.
9 */
11 #include <sys/mman.h>
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"
22 #include "trace.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
31 #ifdef DEBUG_XEN_HVM
32 #define DPRINTF(fmt, ...) \
33 do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
34 #else
35 #define DPRINTF(fmt, ...) \
36 do { } while (0)
37 #endif
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 */
45 /* This allows QEMU to build on a system that has Xen 4.5 or earlier
46 * installed. This here (not in hw/xen/xen_common.h) because xen/hvm/ioreq.h
47 * needs to be included before this block and hw/xen/xen_common.h needs to
48 * be included before xen/hvm/ioreq.h
50 #ifndef IOREQ_TYPE_VMWARE_PORT
51 #define IOREQ_TYPE_VMWARE_PORT 3
52 struct vmware_regs {
53 uint32_t esi;
54 uint32_t edi;
55 uint32_t ebx;
56 uint32_t ecx;
57 uint32_t edx;
59 typedef struct vmware_regs vmware_regs_t;
61 struct shared_vmport_iopage {
62 struct vmware_regs vcpu_vmport_regs[1];
64 typedef struct shared_vmport_iopage shared_vmport_iopage_t;
65 #endif
67 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
68 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
70 return shared_page->vcpu_iodata[i].vp_eport;
72 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
74 return &shared_page->vcpu_iodata[vcpu].vp_ioreq;
76 # define FMT_ioreq_size PRIx64
77 #else
78 static inline uint32_t xen_vcpu_eport(shared_iopage_t *shared_page, int i)
80 return shared_page->vcpu_ioreq[i].vp_eport;
82 static inline ioreq_t *xen_vcpu_ioreq(shared_iopage_t *shared_page, int vcpu)
84 return &shared_page->vcpu_ioreq[vcpu];
86 # define FMT_ioreq_size "u"
87 #endif
89 #define BUFFER_IO_MAX_DELAY 100
91 typedef struct XenPhysmap {
92 hwaddr start_addr;
93 ram_addr_t size;
94 const char *name;
95 hwaddr phys_offset;
97 QLIST_ENTRY(XenPhysmap) list;
98 } XenPhysmap;
100 typedef struct XenIOState {
101 ioservid_t ioservid;
102 shared_iopage_t *shared_page;
103 shared_vmport_iopage_t *shared_vmport_page;
104 buffered_iopage_t *buffered_io_page;
105 QEMUTimer *buffered_io_timer;
106 CPUState **cpu_by_vcpu_id;
107 /* the evtchn port for polling the notification, */
108 evtchn_port_t *ioreq_local_port;
109 /* evtchn local port for buffered io */
110 evtchn_port_t bufioreq_local_port;
111 /* the evtchn fd for polling */
112 XenEvtchn xce_handle;
113 /* which vcpu we are serving */
114 int send_vcpu;
116 struct xs_handle *xenstore;
117 MemoryListener memory_listener;
118 MemoryListener io_listener;
119 DeviceListener device_listener;
120 QLIST_HEAD(, XenPhysmap) physmap;
121 hwaddr free_phys_offset;
122 const XenPhysmap *log_for_dirtybit;
124 Notifier exit;
125 Notifier suspend;
126 Notifier wakeup;
127 } XenIOState;
129 /* Xen specific function for piix pci */
131 int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num)
133 return irq_num + ((pci_dev->devfn >> 3) << 2);
136 void xen_piix3_set_irq(void *opaque, int irq_num, int level)
138 xc_hvm_set_pci_intx_level(xen_xc, xen_domid, 0, 0, irq_num >> 2,
139 irq_num & 3, level);
142 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
144 int i;
146 /* Scan for updates to PCI link routes (0x60-0x63). */
147 for (i = 0; i < len; i++) {
148 uint8_t v = (val >> (8 * i)) & 0xff;
149 if (v & 0x80) {
150 v = 0;
152 v &= 0xf;
153 if (((address + i) >= 0x60) && ((address + i) <= 0x63)) {
154 xc_hvm_set_pci_link_route(xen_xc, xen_domid, address + i - 0x60, v);
159 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
161 xen_xc_hvm_inject_msi(xen_xc, xen_domid, addr, data);
164 static void xen_suspend_notifier(Notifier *notifier, void *data)
166 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3);
169 /* Xen Interrupt Controller */
171 static void xen_set_irq(void *opaque, int irq, int level)
173 xc_hvm_set_isa_irq_level(xen_xc, xen_domid, irq, level);
176 qemu_irq *xen_interrupt_controller_init(void)
178 return qemu_allocate_irqs(xen_set_irq, NULL, 16);
181 /* Memory Ops */
183 static void xen_ram_init(PCMachineState *pcms,
184 ram_addr_t ram_size, MemoryRegion **ram_memory_p)
186 MemoryRegion *sysmem = get_system_memory();
187 ram_addr_t block_len;
188 uint64_t user_lowmem = object_property_get_int(qdev_get_machine(),
189 PC_MACHINE_MAX_RAM_BELOW_4G,
190 &error_abort);
192 /* Handle the machine opt max-ram-below-4g. It is basically doing
193 * min(xen limit, user limit).
195 if (HVM_BELOW_4G_RAM_END <= user_lowmem) {
196 user_lowmem = HVM_BELOW_4G_RAM_END;
199 if (ram_size >= user_lowmem) {
200 pcms->above_4g_mem_size = ram_size - user_lowmem;
201 pcms->below_4g_mem_size = user_lowmem;
202 } else {
203 pcms->above_4g_mem_size = 0;
204 pcms->below_4g_mem_size = ram_size;
206 if (!pcms->above_4g_mem_size) {
207 block_len = ram_size;
208 } else {
210 * Xen does not allocate the memory continuously, it keeps a
211 * hole of the size computed above or passed in.
213 block_len = (1ULL << 32) + pcms->above_4g_mem_size;
215 memory_region_init_ram(&ram_memory, NULL, "xen.ram", block_len,
216 &error_fatal);
217 *ram_memory_p = &ram_memory;
218 vmstate_register_ram_global(&ram_memory);
220 memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k",
221 &ram_memory, 0, 0xa0000);
222 memory_region_add_subregion(sysmem, 0, &ram_640k);
223 /* Skip of the VGA IO memory space, it will be registered later by the VGA
224 * emulated device.
226 * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load
227 * the Options ROM, so it is registered here as RAM.
229 memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo",
230 &ram_memory, 0xc0000,
231 pcms->below_4g_mem_size - 0xc0000);
232 memory_region_add_subregion(sysmem, 0xc0000, &ram_lo);
233 if (pcms->above_4g_mem_size > 0) {
234 memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi",
235 &ram_memory, 0x100000000ULL,
236 pcms->above_4g_mem_size);
237 memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi);
241 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr)
243 unsigned long nr_pfn;
244 xen_pfn_t *pfn_list;
245 int i;
247 if (runstate_check(RUN_STATE_INMIGRATE)) {
248 /* RAM already populated in Xen */
249 fprintf(stderr, "%s: do not alloc "RAM_ADDR_FMT
250 " bytes of ram at "RAM_ADDR_FMT" when runstate is INMIGRATE\n",
251 __func__, size, ram_addr);
252 return;
255 if (mr == &ram_memory) {
256 return;
259 trace_xen_ram_alloc(ram_addr, size);
261 nr_pfn = size >> TARGET_PAGE_BITS;
262 pfn_list = g_malloc(sizeof (*pfn_list) * nr_pfn);
264 for (i = 0; i < nr_pfn; i++) {
265 pfn_list[i] = (ram_addr >> TARGET_PAGE_BITS) + i;
268 if (xc_domain_populate_physmap_exact(xen_xc, xen_domid, nr_pfn, 0, 0, pfn_list)) {
269 hw_error("xen: failed to populate ram at " RAM_ADDR_FMT, ram_addr);
272 g_free(pfn_list);
275 static XenPhysmap *get_physmapping(XenIOState *state,
276 hwaddr start_addr, ram_addr_t size)
278 XenPhysmap *physmap = NULL;
280 start_addr &= TARGET_PAGE_MASK;
282 QLIST_FOREACH(physmap, &state->physmap, list) {
283 if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
284 return physmap;
287 return NULL;
290 static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
291 ram_addr_t size, void *opaque)
293 hwaddr addr = start_addr & TARGET_PAGE_MASK;
294 XenIOState *xen_io_state = opaque;
295 XenPhysmap *physmap = NULL;
297 QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
298 if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
299 return physmap->start_addr;
303 return start_addr;
306 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
307 static int xen_add_to_physmap(XenIOState *state,
308 hwaddr start_addr,
309 ram_addr_t size,
310 MemoryRegion *mr,
311 hwaddr offset_within_region)
313 unsigned long i = 0;
314 int rc = 0;
315 XenPhysmap *physmap = NULL;
316 hwaddr pfn, start_gpfn;
317 hwaddr phys_offset = memory_region_get_ram_addr(mr);
318 char path[80], value[17];
319 const char *mr_name;
321 if (get_physmapping(state, start_addr, size)) {
322 return 0;
324 if (size <= 0) {
325 return -1;
328 /* Xen can only handle a single dirty log region for now and we want
329 * the linear framebuffer to be that region.
330 * Avoid tracking any regions that is not videoram and avoid tracking
331 * the legacy vga region. */
332 if (mr == framebuffer && start_addr > 0xbffff) {
333 goto go_physmap;
335 return -1;
337 go_physmap:
338 DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
339 start_addr, start_addr + size);
341 pfn = phys_offset >> TARGET_PAGE_BITS;
342 start_gpfn = start_addr >> TARGET_PAGE_BITS;
343 for (i = 0; i < size >> TARGET_PAGE_BITS; i++) {
344 unsigned long idx = pfn + i;
345 xen_pfn_t gpfn = start_gpfn + i;
347 rc = xen_xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
348 if (rc) {
349 DPRINTF("add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
350 PRI_xen_pfn" failed: %d (errno: %d)\n", idx, gpfn, rc, errno);
351 return -rc;
355 mr_name = memory_region_name(mr);
357 physmap = g_malloc(sizeof (XenPhysmap));
359 physmap->start_addr = start_addr;
360 physmap->size = size;
361 physmap->name = mr_name;
362 physmap->phys_offset = phys_offset;
364 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
366 xc_domain_pin_memory_cacheattr(xen_xc, xen_domid,
367 start_addr >> TARGET_PAGE_BITS,
368 (start_addr + size - 1) >> TARGET_PAGE_BITS,
369 XEN_DOMCTL_MEM_CACHEATTR_WB);
371 snprintf(path, sizeof(path),
372 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
373 xen_domid, (uint64_t)phys_offset);
374 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)start_addr);
375 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
376 return -1;
378 snprintf(path, sizeof(path),
379 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
380 xen_domid, (uint64_t)phys_offset);
381 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)size);
382 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
383 return -1;
385 if (mr_name) {
386 snprintf(path, sizeof(path),
387 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
388 xen_domid, (uint64_t)phys_offset);
389 if (!xs_write(state->xenstore, 0, path, mr_name, strlen(mr_name))) {
390 return -1;
394 return 0;
397 static int xen_remove_from_physmap(XenIOState *state,
398 hwaddr start_addr,
399 ram_addr_t size)
401 unsigned long i = 0;
402 int rc = 0;
403 XenPhysmap *physmap = NULL;
404 hwaddr phys_offset = 0;
406 physmap = get_physmapping(state, start_addr, size);
407 if (physmap == NULL) {
408 return -1;
411 phys_offset = physmap->phys_offset;
412 size = physmap->size;
414 DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
415 "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
417 size >>= TARGET_PAGE_BITS;
418 start_addr >>= TARGET_PAGE_BITS;
419 phys_offset >>= TARGET_PAGE_BITS;
420 for (i = 0; i < size; i++) {
421 xen_pfn_t idx = start_addr + i;
422 xen_pfn_t gpfn = phys_offset + i;
424 rc = xen_xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
425 if (rc) {
426 fprintf(stderr, "add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
427 PRI_xen_pfn" failed: %d (errno: %d)\n", idx, gpfn, rc, errno);
428 return -rc;
432 QLIST_REMOVE(physmap, list);
433 if (state->log_for_dirtybit == physmap) {
434 state->log_for_dirtybit = NULL;
436 g_free(physmap);
438 return 0;
441 #else
442 static int xen_add_to_physmap(XenIOState *state,
443 hwaddr start_addr,
444 ram_addr_t size,
445 MemoryRegion *mr,
446 hwaddr offset_within_region)
448 return -ENOSYS;
451 static int xen_remove_from_physmap(XenIOState *state,
452 hwaddr start_addr,
453 ram_addr_t size)
455 return -ENOSYS;
457 #endif
459 static void xen_set_memory(struct MemoryListener *listener,
460 MemoryRegionSection *section,
461 bool add)
463 XenIOState *state = container_of(listener, XenIOState, memory_listener);
464 hwaddr start_addr = section->offset_within_address_space;
465 ram_addr_t size = int128_get64(section->size);
466 bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA);
467 hvmmem_type_t mem_type;
469 if (section->mr == &ram_memory) {
470 return;
471 } else {
472 if (add) {
473 xen_map_memory_section(xen_xc, xen_domid, state->ioservid,
474 section);
475 } else {
476 xen_unmap_memory_section(xen_xc, xen_domid, state->ioservid,
477 section);
481 if (!memory_region_is_ram(section->mr)) {
482 return;
485 if (log_dirty != add) {
486 return;
489 trace_xen_client_set_memory(start_addr, size, log_dirty);
491 start_addr &= TARGET_PAGE_MASK;
492 size = TARGET_PAGE_ALIGN(size);
494 if (add) {
495 if (!memory_region_is_rom(section->mr)) {
496 xen_add_to_physmap(state, start_addr, size,
497 section->mr, section->offset_within_region);
498 } else {
499 mem_type = HVMMEM_ram_ro;
500 if (xc_hvm_set_mem_type(xen_xc, xen_domid, mem_type,
501 start_addr >> TARGET_PAGE_BITS,
502 size >> TARGET_PAGE_BITS)) {
503 DPRINTF("xc_hvm_set_mem_type error, addr: "TARGET_FMT_plx"\n",
504 start_addr);
507 } else {
508 if (xen_remove_from_physmap(state, start_addr, size) < 0) {
509 DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
514 static void xen_region_add(MemoryListener *listener,
515 MemoryRegionSection *section)
517 memory_region_ref(section->mr);
518 xen_set_memory(listener, section, true);
521 static void xen_region_del(MemoryListener *listener,
522 MemoryRegionSection *section)
524 xen_set_memory(listener, section, false);
525 memory_region_unref(section->mr);
528 static void xen_io_add(MemoryListener *listener,
529 MemoryRegionSection *section)
531 XenIOState *state = container_of(listener, XenIOState, io_listener);
533 memory_region_ref(section->mr);
535 xen_map_io_section(xen_xc, xen_domid, state->ioservid, section);
538 static void xen_io_del(MemoryListener *listener,
539 MemoryRegionSection *section)
541 XenIOState *state = container_of(listener, XenIOState, io_listener);
543 xen_unmap_io_section(xen_xc, xen_domid, state->ioservid, section);
545 memory_region_unref(section->mr);
548 static void xen_device_realize(DeviceListener *listener,
549 DeviceState *dev)
551 XenIOState *state = container_of(listener, XenIOState, device_listener);
553 if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
554 PCIDevice *pci_dev = PCI_DEVICE(dev);
556 xen_map_pcidev(xen_xc, xen_domid, state->ioservid, pci_dev);
560 static void xen_device_unrealize(DeviceListener *listener,
561 DeviceState *dev)
563 XenIOState *state = container_of(listener, XenIOState, device_listener);
565 if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
566 PCIDevice *pci_dev = PCI_DEVICE(dev);
568 xen_unmap_pcidev(xen_xc, xen_domid, state->ioservid, pci_dev);
572 static void xen_sync_dirty_bitmap(XenIOState *state,
573 hwaddr start_addr,
574 ram_addr_t size)
576 hwaddr npages = size >> TARGET_PAGE_BITS;
577 const int width = sizeof(unsigned long) * 8;
578 unsigned long bitmap[(npages + width - 1) / width];
579 int rc, i, j;
580 const XenPhysmap *physmap = NULL;
582 physmap = get_physmapping(state, start_addr, size);
583 if (physmap == NULL) {
584 /* not handled */
585 return;
588 if (state->log_for_dirtybit == NULL) {
589 state->log_for_dirtybit = physmap;
590 } else if (state->log_for_dirtybit != physmap) {
591 /* Only one range for dirty bitmap can be tracked. */
592 return;
595 rc = xc_hvm_track_dirty_vram(xen_xc, xen_domid,
596 start_addr >> TARGET_PAGE_BITS, npages,
597 bitmap);
598 if (rc < 0) {
599 #ifndef ENODATA
600 #define ENODATA ENOENT
601 #endif
602 if (errno == ENODATA) {
603 memory_region_set_dirty(framebuffer, 0, size);
604 DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
605 ", 0x" TARGET_FMT_plx "): %s\n",
606 start_addr, start_addr + size, strerror(errno));
608 return;
611 for (i = 0; i < ARRAY_SIZE(bitmap); i++) {
612 unsigned long map = bitmap[i];
613 while (map != 0) {
614 j = ctzl(map);
615 map &= ~(1ul << j);
616 memory_region_set_dirty(framebuffer,
617 (i * width + j) * TARGET_PAGE_SIZE,
618 TARGET_PAGE_SIZE);
623 static void xen_log_start(MemoryListener *listener,
624 MemoryRegionSection *section,
625 int old, int new)
627 XenIOState *state = container_of(listener, XenIOState, memory_listener);
629 if (new & ~old & (1 << DIRTY_MEMORY_VGA)) {
630 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
631 int128_get64(section->size));
635 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section,
636 int old, int new)
638 XenIOState *state = container_of(listener, XenIOState, memory_listener);
640 if (old & ~new & (1 << DIRTY_MEMORY_VGA)) {
641 state->log_for_dirtybit = NULL;
642 /* Disable dirty bit tracking */
643 xc_hvm_track_dirty_vram(xen_xc, xen_domid, 0, 0, NULL);
647 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
649 XenIOState *state = container_of(listener, XenIOState, memory_listener);
651 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
652 int128_get64(section->size));
655 static void xen_log_global_start(MemoryListener *listener)
657 if (xen_enabled()) {
658 xen_in_migration = true;
662 static void xen_log_global_stop(MemoryListener *listener)
664 xen_in_migration = false;
667 static MemoryListener xen_memory_listener = {
668 .region_add = xen_region_add,
669 .region_del = xen_region_del,
670 .log_start = xen_log_start,
671 .log_stop = xen_log_stop,
672 .log_sync = xen_log_sync,
673 .log_global_start = xen_log_global_start,
674 .log_global_stop = xen_log_global_stop,
675 .priority = 10,
678 static MemoryListener xen_io_listener = {
679 .region_add = xen_io_add,
680 .region_del = xen_io_del,
681 .priority = 10,
684 static DeviceListener xen_device_listener = {
685 .realize = xen_device_realize,
686 .unrealize = xen_device_unrealize,
689 /* get the ioreq packets from share mem */
690 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
692 ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
694 if (req->state != STATE_IOREQ_READY) {
695 DPRINTF("I/O request not ready: "
696 "%x, ptr: %x, port: %"PRIx64", "
697 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
698 req->state, req->data_is_ptr, req->addr,
699 req->data, req->count, req->size);
700 return NULL;
703 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
705 req->state = STATE_IOREQ_INPROCESS;
706 return req;
709 /* use poll to get the port notification */
710 /* ioreq_vec--out,the */
711 /* retval--the number of ioreq packet */
712 static ioreq_t *cpu_get_ioreq(XenIOState *state)
714 int i;
715 evtchn_port_t port;
717 port = xc_evtchn_pending(state->xce_handle);
718 if (port == state->bufioreq_local_port) {
719 timer_mod(state->buffered_io_timer,
720 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
721 return NULL;
724 if (port != -1) {
725 for (i = 0; i < max_cpus; i++) {
726 if (state->ioreq_local_port[i] == port) {
727 break;
731 if (i == max_cpus) {
732 hw_error("Fatal error while trying to get io event!\n");
735 /* unmask the wanted port again */
736 xc_evtchn_unmask(state->xce_handle, port);
738 /* get the io packet from shared memory */
739 state->send_vcpu = i;
740 return cpu_get_ioreq_from_shared_memory(state, i);
743 /* read error or read nothing */
744 return NULL;
747 static uint32_t do_inp(pio_addr_t addr, unsigned long size)
749 switch (size) {
750 case 1:
751 return cpu_inb(addr);
752 case 2:
753 return cpu_inw(addr);
754 case 4:
755 return cpu_inl(addr);
756 default:
757 hw_error("inp: bad size: %04"FMT_pioaddr" %lx", addr, size);
761 static void do_outp(pio_addr_t addr,
762 unsigned long size, uint32_t val)
764 switch (size) {
765 case 1:
766 return cpu_outb(addr, val);
767 case 2:
768 return cpu_outw(addr, val);
769 case 4:
770 return cpu_outl(addr, val);
771 default:
772 hw_error("outp: bad size: %04"FMT_pioaddr" %lx", addr, size);
777 * Helper functions which read/write an object from/to physical guest
778 * memory, as part of the implementation of an ioreq.
780 * Equivalent to
781 * cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
782 * val, req->size, 0/1)
783 * except without the integer overflow problems.
785 static void rw_phys_req_item(hwaddr addr,
786 ioreq_t *req, uint32_t i, void *val, int rw)
788 /* Do everything unsigned so overflow just results in a truncated result
789 * and accesses to undesired parts of guest memory, which is up
790 * to the guest */
791 hwaddr offset = (hwaddr)req->size * i;
792 if (req->df) {
793 addr -= offset;
794 } else {
795 addr += offset;
797 cpu_physical_memory_rw(addr, val, req->size, rw);
800 static inline void read_phys_req_item(hwaddr addr,
801 ioreq_t *req, uint32_t i, void *val)
803 rw_phys_req_item(addr, req, i, val, 0);
805 static inline void write_phys_req_item(hwaddr addr,
806 ioreq_t *req, uint32_t i, void *val)
808 rw_phys_req_item(addr, req, i, val, 1);
812 static void cpu_ioreq_pio(ioreq_t *req)
814 uint32_t i;
816 trace_cpu_ioreq_pio(req, req->dir, req->df, req->data_is_ptr, req->addr,
817 req->data, req->count, req->size);
819 if (req->dir == IOREQ_READ) {
820 if (!req->data_is_ptr) {
821 req->data = do_inp(req->addr, req->size);
822 trace_cpu_ioreq_pio_read_reg(req, req->data, req->addr,
823 req->size);
824 } else {
825 uint32_t tmp;
827 for (i = 0; i < req->count; i++) {
828 tmp = do_inp(req->addr, req->size);
829 write_phys_req_item(req->data, req, i, &tmp);
832 } else if (req->dir == IOREQ_WRITE) {
833 if (!req->data_is_ptr) {
834 trace_cpu_ioreq_pio_write_reg(req, req->data, req->addr,
835 req->size);
836 do_outp(req->addr, req->size, req->data);
837 } else {
838 for (i = 0; i < req->count; i++) {
839 uint32_t tmp = 0;
841 read_phys_req_item(req->data, req, i, &tmp);
842 do_outp(req->addr, req->size, tmp);
848 static void cpu_ioreq_move(ioreq_t *req)
850 uint32_t i;
852 trace_cpu_ioreq_move(req, req->dir, req->df, req->data_is_ptr, req->addr,
853 req->data, req->count, req->size);
855 if (!req->data_is_ptr) {
856 if (req->dir == IOREQ_READ) {
857 for (i = 0; i < req->count; i++) {
858 read_phys_req_item(req->addr, req, i, &req->data);
860 } else if (req->dir == IOREQ_WRITE) {
861 for (i = 0; i < req->count; i++) {
862 write_phys_req_item(req->addr, req, i, &req->data);
865 } else {
866 uint64_t tmp;
868 if (req->dir == IOREQ_READ) {
869 for (i = 0; i < req->count; i++) {
870 read_phys_req_item(req->addr, req, i, &tmp);
871 write_phys_req_item(req->data, req, i, &tmp);
873 } else if (req->dir == IOREQ_WRITE) {
874 for (i = 0; i < req->count; i++) {
875 read_phys_req_item(req->data, req, i, &tmp);
876 write_phys_req_item(req->addr, req, i, &tmp);
882 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
884 X86CPU *cpu;
885 CPUX86State *env;
887 cpu = X86_CPU(current_cpu);
888 env = &cpu->env;
889 env->regs[R_EAX] = req->data;
890 env->regs[R_EBX] = vmport_regs->ebx;
891 env->regs[R_ECX] = vmport_regs->ecx;
892 env->regs[R_EDX] = vmport_regs->edx;
893 env->regs[R_ESI] = vmport_regs->esi;
894 env->regs[R_EDI] = vmport_regs->edi;
897 static void regs_from_cpu(vmware_regs_t *vmport_regs)
899 X86CPU *cpu = X86_CPU(current_cpu);
900 CPUX86State *env = &cpu->env;
902 vmport_regs->ebx = env->regs[R_EBX];
903 vmport_regs->ecx = env->regs[R_ECX];
904 vmport_regs->edx = env->regs[R_EDX];
905 vmport_regs->esi = env->regs[R_ESI];
906 vmport_regs->edi = env->regs[R_EDI];
909 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
911 vmware_regs_t *vmport_regs;
913 assert(state->shared_vmport_page);
914 vmport_regs =
915 &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
916 QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
918 current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
919 regs_to_cpu(vmport_regs, req);
920 cpu_ioreq_pio(req);
921 regs_from_cpu(vmport_regs);
922 current_cpu = NULL;
925 static void handle_ioreq(XenIOState *state, ioreq_t *req)
927 trace_handle_ioreq(req, req->type, req->dir, req->df, req->data_is_ptr,
928 req->addr, req->data, req->count, req->size);
930 if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
931 (req->size < sizeof (target_ulong))) {
932 req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
935 if (req->dir == IOREQ_WRITE)
936 trace_handle_ioreq_write(req, req->type, req->df, req->data_is_ptr,
937 req->addr, req->data, req->count, req->size);
939 switch (req->type) {
940 case IOREQ_TYPE_PIO:
941 cpu_ioreq_pio(req);
942 break;
943 case IOREQ_TYPE_COPY:
944 cpu_ioreq_move(req);
945 break;
946 case IOREQ_TYPE_VMWARE_PORT:
947 handle_vmport_ioreq(state, req);
948 break;
949 case IOREQ_TYPE_TIMEOFFSET:
950 break;
951 case IOREQ_TYPE_INVALIDATE:
952 xen_invalidate_map_cache();
953 break;
954 case IOREQ_TYPE_PCI_CONFIG: {
955 uint32_t sbdf = req->addr >> 32;
956 uint32_t val;
958 /* Fake a write to port 0xCF8 so that
959 * the config space access will target the
960 * correct device model.
962 val = (1u << 31) |
963 ((req->addr & 0x0f00) << 16) |
964 ((sbdf & 0xffff) << 8) |
965 (req->addr & 0xfc);
966 do_outp(0xcf8, 4, val);
968 /* Now issue the config space access via
969 * port 0xCFC
971 req->addr = 0xcfc | (req->addr & 0x03);
972 cpu_ioreq_pio(req);
973 break;
975 default:
976 hw_error("Invalid ioreq type 0x%x\n", req->type);
978 if (req->dir == IOREQ_READ) {
979 trace_handle_ioreq_read(req, req->type, req->df, req->data_is_ptr,
980 req->addr, req->data, req->count, req->size);
984 static int handle_buffered_iopage(XenIOState *state)
986 buffered_iopage_t *buf_page = state->buffered_io_page;
987 buf_ioreq_t *buf_req = NULL;
988 ioreq_t req;
989 int qw;
991 if (!buf_page) {
992 return 0;
995 memset(&req, 0x00, sizeof(req));
997 for (;;) {
998 uint32_t rdptr = buf_page->read_pointer, wrptr;
1000 xen_rmb();
1001 wrptr = buf_page->write_pointer;
1002 xen_rmb();
1003 if (rdptr != buf_page->read_pointer) {
1004 continue;
1006 if (rdptr == wrptr) {
1007 break;
1009 buf_req = &buf_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
1010 req.size = 1UL << buf_req->size;
1011 req.count = 1;
1012 req.addr = buf_req->addr;
1013 req.data = buf_req->data;
1014 req.state = STATE_IOREQ_READY;
1015 req.dir = buf_req->dir;
1016 req.df = 1;
1017 req.type = buf_req->type;
1018 req.data_is_ptr = 0;
1019 qw = (req.size == 8);
1020 if (qw) {
1021 buf_req = &buf_page->buf_ioreq[(rdptr + 1) %
1022 IOREQ_BUFFER_SLOT_NUM];
1023 req.data |= ((uint64_t)buf_req->data) << 32;
1026 handle_ioreq(state, &req);
1028 atomic_add(&buf_page->read_pointer, qw + 1);
1031 return req.count;
1034 static void handle_buffered_io(void *opaque)
1036 XenIOState *state = opaque;
1038 if (handle_buffered_iopage(state)) {
1039 timer_mod(state->buffered_io_timer,
1040 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
1041 } else {
1042 timer_del(state->buffered_io_timer);
1043 xc_evtchn_unmask(state->xce_handle, state->bufioreq_local_port);
1047 static void cpu_handle_ioreq(void *opaque)
1049 XenIOState *state = opaque;
1050 ioreq_t *req = cpu_get_ioreq(state);
1052 handle_buffered_iopage(state);
1053 if (req) {
1054 handle_ioreq(state, req);
1056 if (req->state != STATE_IOREQ_INPROCESS) {
1057 fprintf(stderr, "Badness in I/O request ... not in service?!: "
1058 "%x, ptr: %x, port: %"PRIx64", "
1059 "data: %"PRIx64", count: %" FMT_ioreq_size
1060 ", size: %" FMT_ioreq_size
1061 ", type: %"FMT_ioreq_size"\n",
1062 req->state, req->data_is_ptr, req->addr,
1063 req->data, req->count, req->size, req->type);
1064 destroy_hvm_domain(false);
1065 return;
1068 xen_wmb(); /* Update ioreq contents /then/ update state. */
1071 * We do this before we send the response so that the tools
1072 * have the opportunity to pick up on the reset before the
1073 * guest resumes and does a hlt with interrupts disabled which
1074 * causes Xen to powerdown the domain.
1076 if (runstate_is_running()) {
1077 if (qemu_shutdown_requested_get()) {
1078 destroy_hvm_domain(false);
1080 if (qemu_reset_requested_get()) {
1081 qemu_system_reset(VMRESET_REPORT);
1082 destroy_hvm_domain(true);
1086 req->state = STATE_IORESP_READY;
1087 xc_evtchn_notify(state->xce_handle, state->ioreq_local_port[state->send_vcpu]);
1091 static void xen_main_loop_prepare(XenIOState *state)
1093 int evtchn_fd = -1;
1095 if (state->xce_handle != XC_HANDLER_INITIAL_VALUE) {
1096 evtchn_fd = xc_evtchn_fd(state->xce_handle);
1099 state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
1100 state);
1102 if (evtchn_fd != -1) {
1103 CPUState *cpu_state;
1105 DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
1106 CPU_FOREACH(cpu_state) {
1107 DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1108 __func__, cpu_state->cpu_index, cpu_state);
1109 state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
1111 qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
1116 static void xen_hvm_change_state_handler(void *opaque, int running,
1117 RunState rstate)
1119 XenIOState *state = opaque;
1121 if (running) {
1122 xen_main_loop_prepare(state);
1125 xen_set_ioreq_server_state(xen_xc, xen_domid,
1126 state->ioservid,
1127 (rstate == RUN_STATE_RUNNING));
1130 static void xen_exit_notifier(Notifier *n, void *data)
1132 XenIOState *state = container_of(n, XenIOState, exit);
1134 xc_evtchn_close(state->xce_handle);
1135 xs_daemon_close(state->xenstore);
1138 static void xen_read_physmap(XenIOState *state)
1140 XenPhysmap *physmap = NULL;
1141 unsigned int len, num, i;
1142 char path[80], *value = NULL;
1143 char **entries = NULL;
1145 snprintf(path, sizeof(path),
1146 "/local/domain/0/device-model/%d/physmap", xen_domid);
1147 entries = xs_directory(state->xenstore, 0, path, &num);
1148 if (entries == NULL)
1149 return;
1151 for (i = 0; i < num; i++) {
1152 physmap = g_malloc(sizeof (XenPhysmap));
1153 physmap->phys_offset = strtoull(entries[i], NULL, 16);
1154 snprintf(path, sizeof(path),
1155 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1156 xen_domid, entries[i]);
1157 value = xs_read(state->xenstore, 0, path, &len);
1158 if (value == NULL) {
1159 g_free(physmap);
1160 continue;
1162 physmap->start_addr = strtoull(value, NULL, 16);
1163 free(value);
1165 snprintf(path, sizeof(path),
1166 "/local/domain/0/device-model/%d/physmap/%s/size",
1167 xen_domid, entries[i]);
1168 value = xs_read(state->xenstore, 0, path, &len);
1169 if (value == NULL) {
1170 g_free(physmap);
1171 continue;
1173 physmap->size = strtoull(value, NULL, 16);
1174 free(value);
1176 snprintf(path, sizeof(path),
1177 "/local/domain/0/device-model/%d/physmap/%s/name",
1178 xen_domid, entries[i]);
1179 physmap->name = xs_read(state->xenstore, 0, path, &len);
1181 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
1183 free(entries);
1186 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1188 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1191 /* return 0 means OK, or -1 means critical issue -- will exit(1) */
1192 int xen_hvm_init(PCMachineState *pcms,
1193 MemoryRegion **ram_memory)
1195 int i, rc;
1196 xen_pfn_t ioreq_pfn;
1197 xen_pfn_t bufioreq_pfn;
1198 evtchn_port_t bufioreq_evtchn;
1199 XenIOState *state;
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");
1206 return -1;
1209 state->xenstore = xs_daemon_open();
1210 if (state->xenstore == NULL) {
1211 perror("xen: xenstore open");
1212 return -1;
1215 rc = xen_create_ioreq_server(xen_xc, xen_domid, &state->ioservid);
1216 if (rc < 0) {
1217 perror("xen: ioreq server create");
1218 return -1;
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,
1232 &bufioreq_evtchn);
1233 if (rc < 0) {
1234 hw_error("failed to get ioreq server info: error %d handle=" XC_INTERFACE_FMT,
1235 errno, xen_xc);
1238 DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1239 DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
1240 DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
1242 state->shared_page = xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1243 PROT_READ|PROT_WRITE, ioreq_pfn);
1244 if (state->shared_page == NULL) {
1245 hw_error("map shared IO page returned error %d handle=" XC_INTERFACE_FMT,
1246 errno, xen_xc);
1249 rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1250 if (!rc) {
1251 DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1252 state->shared_vmport_page =
1253 xc_map_foreign_range(xen_xc, xen_domid, XC_PAGE_SIZE,
1254 PROT_READ|PROT_WRITE, ioreq_pfn);
1255 if (state->shared_vmport_page == NULL) {
1256 hw_error("map shared vmport IO page returned error %d handle="
1257 XC_INTERFACE_FMT, errno, xen_xc);
1259 } else if (rc != -ENOSYS) {
1260 hw_error("get vmport regs pfn returned error %d, rc=%d", errno, rc);
1263 state->buffered_io_page = xc_map_foreign_range(xen_xc, xen_domid,
1264 XC_PAGE_SIZE,
1265 PROT_READ|PROT_WRITE,
1266 bufioreq_pfn);
1267 if (state->buffered_io_page == NULL) {
1268 hw_error("map buffered IO page returned error %d", errno);
1271 /* Note: cpus is empty at this point in init */
1272 state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1274 rc = xen_set_ioreq_server_state(xen_xc, xen_domid, state->ioservid, true);
1275 if (rc < 0) {
1276 hw_error("failed to enable ioreq server info: error %d handle=" XC_INTERFACE_FMT,
1277 errno, xen_xc);
1280 state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1282 /* FIXME: how about if we overflow the page here? */
1283 for (i = 0; i < max_cpus; i++) {
1284 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1285 xen_vcpu_eport(state->shared_page, i));
1286 if (rc == -1) {
1287 fprintf(stderr, "shared evtchn %d bind error %d\n", i, errno);
1288 return -1;
1290 state->ioreq_local_port[i] = rc;
1293 rc = xc_evtchn_bind_interdomain(state->xce_handle, xen_domid,
1294 bufioreq_evtchn);
1295 if (rc == -1) {
1296 fprintf(stderr, "buffered evtchn bind error %d\n", errno);
1297 return -1;
1299 state->bufioreq_local_port = rc;
1301 /* Init RAM management */
1302 xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1303 xen_ram_init(pcms, ram_size, ram_memory);
1305 qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1307 state->memory_listener = xen_memory_listener;
1308 QLIST_INIT(&state->physmap);
1309 memory_listener_register(&state->memory_listener, &address_space_memory);
1310 state->log_for_dirtybit = NULL;
1312 state->io_listener = xen_io_listener;
1313 memory_listener_register(&state->io_listener, &address_space_io);
1315 state->device_listener = xen_device_listener;
1316 device_listener_register(&state->device_listener);
1318 /* Initialize backend core & drivers */
1319 if (xen_be_init() != 0) {
1320 fprintf(stderr, "%s: xen backend core setup failed\n", __FUNCTION__);
1321 return -1;
1323 xen_be_register("console", &xen_console_ops);
1324 xen_be_register("vkbd", &xen_kbdmouse_ops);
1325 xen_be_register("qdisk", &xen_blkdev_ops);
1326 xen_read_physmap(state);
1328 return 0;
1331 void destroy_hvm_domain(bool reboot)
1333 XenXC xc_handle;
1334 int sts;
1336 xc_handle = xen_xc_interface_open(0, 0, 0);
1337 if (xc_handle == XC_HANDLER_INITIAL_VALUE) {
1338 fprintf(stderr, "Cannot acquire xenctrl handle\n");
1339 } else {
1340 sts = xc_domain_shutdown(xc_handle, xen_domid,
1341 reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
1342 if (sts != 0) {
1343 fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1344 "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1345 sts, strerror(errno));
1346 } else {
1347 fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1348 reboot ? "reboot" : "poweroff");
1350 xc_interface_close(xc_handle);
1354 void xen_register_framebuffer(MemoryRegion *mr)
1356 framebuffer = mr;
1359 void xen_shutdown_fatal_error(const char *fmt, ...)
1361 va_list ap;
1363 va_start(ap, fmt);
1364 vfprintf(stderr, fmt, ap);
1365 va_end(ap);
1366 fprintf(stderr, "Will destroy the domain.\n");
1367 /* destroy the domain */
1368 qemu_system_shutdown_request();
1371 void xen_modified_memory(ram_addr_t start, ram_addr_t length)
1373 if (unlikely(xen_in_migration)) {
1374 int rc;
1375 ram_addr_t start_pfn, nb_pages;
1377 if (length == 0) {
1378 length = TARGET_PAGE_SIZE;
1380 start_pfn = start >> TARGET_PAGE_BITS;
1381 nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1382 - start_pfn;
1383 rc = xc_hvm_modified_memory(xen_xc, xen_domid, start_pfn, nb_pages);
1384 if (rc) {
1385 fprintf(stderr,
1386 "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1387 __func__, start, nb_pages, rc, strerror(-rc));
1392 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1394 if (enable) {
1395 memory_global_dirty_log_start();
1396 } else {
1397 memory_global_dirty_log_stop();