s390x: fix generation of event information crw
[qemu.git] / xen-hvm.c
blob590f66dcf8bb2bfd19fccaf203ef3323c4416524
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/error-report.h"
21 #include "qemu/range.h"
22 #include "sysemu/xen-mapcache.h"
23 #include "trace.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
32 #ifdef DEBUG_XEN_HVM
33 #define DPRINTF(fmt, ...) \
34 do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
35 #else
36 #define DPRINTF(fmt, ...) \
37 do { } while (0)
38 #endif
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
53 struct vmware_regs {
54 uint32_t esi;
55 uint32_t edi;
56 uint32_t ebx;
57 uint32_t ecx;
58 uint32_t edx;
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;
66 #endif
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
78 #else
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"
88 #endif
90 #define BUFFER_IO_MAX_DELAY 100
92 typedef struct XenPhysmap {
93 hwaddr start_addr;
94 ram_addr_t size;
95 const char *name;
96 hwaddr phys_offset;
98 QLIST_ENTRY(XenPhysmap) list;
99 } XenPhysmap;
101 typedef struct XenIOState {
102 ioservid_t ioservid;
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_handle *xce_handle;
114 /* which vcpu we are serving */
115 int send_vcpu;
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;
125 Notifier exit;
126 Notifier suspend;
127 Notifier wakeup;
128 } XenIOState;
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,
140 irq_num & 3, level);
143 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
145 int i;
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;
150 if (v & 0x80) {
151 v = 0;
153 v &= 0xf;
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);
182 /* Memory Ops */
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,
191 &error_abort);
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;
203 } else {
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;
209 } else {
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,
217 &error_fatal);
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
225 * emulated device.
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,
243 Error **errp)
245 unsigned long nr_pfn;
246 xen_pfn_t *pfn_list;
247 int i;
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);
254 return;
257 if (mr == &ram_memory) {
258 return;
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 error_setg(errp, "xen: failed to populate ram at " RAM_ADDR_FMT,
272 ram_addr);
275 g_free(pfn_list);
278 static XenPhysmap *get_physmapping(XenIOState *state,
279 hwaddr start_addr, ram_addr_t size)
281 XenPhysmap *physmap = NULL;
283 start_addr &= TARGET_PAGE_MASK;
285 QLIST_FOREACH(physmap, &state->physmap, list) {
286 if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) {
287 return physmap;
290 return NULL;
293 static hwaddr xen_phys_offset_to_gaddr(hwaddr start_addr,
294 ram_addr_t size, void *opaque)
296 hwaddr addr = start_addr & TARGET_PAGE_MASK;
297 XenIOState *xen_io_state = opaque;
298 XenPhysmap *physmap = NULL;
300 QLIST_FOREACH(physmap, &xen_io_state->physmap, list) {
301 if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) {
302 return physmap->start_addr;
306 return start_addr;
309 #if CONFIG_XEN_CTRL_INTERFACE_VERSION >= 340
310 static int xen_add_to_physmap(XenIOState *state,
311 hwaddr start_addr,
312 ram_addr_t size,
313 MemoryRegion *mr,
314 hwaddr offset_within_region)
316 unsigned long i = 0;
317 int rc = 0;
318 XenPhysmap *physmap = NULL;
319 hwaddr pfn, start_gpfn;
320 hwaddr phys_offset = memory_region_get_ram_addr(mr);
321 char path[80], value[17];
322 const char *mr_name;
324 if (get_physmapping(state, start_addr, size)) {
325 return 0;
327 if (size <= 0) {
328 return -1;
331 /* Xen can only handle a single dirty log region for now and we want
332 * the linear framebuffer to be that region.
333 * Avoid tracking any regions that is not videoram and avoid tracking
334 * the legacy vga region. */
335 if (mr == framebuffer && start_addr > 0xbffff) {
336 goto go_physmap;
338 return -1;
340 go_physmap:
341 DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n",
342 start_addr, start_addr + size);
344 pfn = phys_offset >> TARGET_PAGE_BITS;
345 start_gpfn = start_addr >> TARGET_PAGE_BITS;
346 for (i = 0; i < size >> TARGET_PAGE_BITS; i++) {
347 unsigned long idx = pfn + i;
348 xen_pfn_t gpfn = start_gpfn + i;
350 rc = xen_xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
351 if (rc) {
352 DPRINTF("add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
353 PRI_xen_pfn" failed: %d (errno: %d)\n", idx, gpfn, rc, errno);
354 return -rc;
358 mr_name = memory_region_name(mr);
360 physmap = g_malloc(sizeof (XenPhysmap));
362 physmap->start_addr = start_addr;
363 physmap->size = size;
364 physmap->name = mr_name;
365 physmap->phys_offset = phys_offset;
367 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
369 xc_domain_pin_memory_cacheattr(xen_xc, xen_domid,
370 start_addr >> TARGET_PAGE_BITS,
371 (start_addr + size - 1) >> TARGET_PAGE_BITS,
372 XEN_DOMCTL_MEM_CACHEATTR_WB);
374 snprintf(path, sizeof(path),
375 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr",
376 xen_domid, (uint64_t)phys_offset);
377 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)start_addr);
378 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
379 return -1;
381 snprintf(path, sizeof(path),
382 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size",
383 xen_domid, (uint64_t)phys_offset);
384 snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)size);
385 if (!xs_write(state->xenstore, 0, path, value, strlen(value))) {
386 return -1;
388 if (mr_name) {
389 snprintf(path, sizeof(path),
390 "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name",
391 xen_domid, (uint64_t)phys_offset);
392 if (!xs_write(state->xenstore, 0, path, mr_name, strlen(mr_name))) {
393 return -1;
397 return 0;
400 static int xen_remove_from_physmap(XenIOState *state,
401 hwaddr start_addr,
402 ram_addr_t size)
404 unsigned long i = 0;
405 int rc = 0;
406 XenPhysmap *physmap = NULL;
407 hwaddr phys_offset = 0;
409 physmap = get_physmapping(state, start_addr, size);
410 if (physmap == NULL) {
411 return -1;
414 phys_offset = physmap->phys_offset;
415 size = physmap->size;
417 DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at "
418 "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset);
420 size >>= TARGET_PAGE_BITS;
421 start_addr >>= TARGET_PAGE_BITS;
422 phys_offset >>= TARGET_PAGE_BITS;
423 for (i = 0; i < size; i++) {
424 xen_pfn_t idx = start_addr + i;
425 xen_pfn_t gpfn = phys_offset + i;
427 rc = xen_xc_domain_add_to_physmap(xen_xc, xen_domid, XENMAPSPACE_gmfn, idx, gpfn);
428 if (rc) {
429 fprintf(stderr, "add_to_physmap MFN %"PRI_xen_pfn" to PFN %"
430 PRI_xen_pfn" failed: %d (errno: %d)\n", idx, gpfn, rc, errno);
431 return -rc;
435 QLIST_REMOVE(physmap, list);
436 if (state->log_for_dirtybit == physmap) {
437 state->log_for_dirtybit = NULL;
439 g_free(physmap);
441 return 0;
444 #else
445 static int xen_add_to_physmap(XenIOState *state,
446 hwaddr start_addr,
447 ram_addr_t size,
448 MemoryRegion *mr,
449 hwaddr offset_within_region)
451 return -ENOSYS;
454 static int xen_remove_from_physmap(XenIOState *state,
455 hwaddr start_addr,
456 ram_addr_t size)
458 return -ENOSYS;
460 #endif
462 static void xen_set_memory(struct MemoryListener *listener,
463 MemoryRegionSection *section,
464 bool add)
466 XenIOState *state = container_of(listener, XenIOState, memory_listener);
467 hwaddr start_addr = section->offset_within_address_space;
468 ram_addr_t size = int128_get64(section->size);
469 bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA);
470 hvmmem_type_t mem_type;
472 if (section->mr == &ram_memory) {
473 return;
474 } else {
475 if (add) {
476 xen_map_memory_section(xen_xc, xen_domid, state->ioservid,
477 section);
478 } else {
479 xen_unmap_memory_section(xen_xc, xen_domid, state->ioservid,
480 section);
484 if (!memory_region_is_ram(section->mr)) {
485 return;
488 if (log_dirty != add) {
489 return;
492 trace_xen_client_set_memory(start_addr, size, log_dirty);
494 start_addr &= TARGET_PAGE_MASK;
495 size = TARGET_PAGE_ALIGN(size);
497 if (add) {
498 if (!memory_region_is_rom(section->mr)) {
499 xen_add_to_physmap(state, start_addr, size,
500 section->mr, section->offset_within_region);
501 } else {
502 mem_type = HVMMEM_ram_ro;
503 if (xc_hvm_set_mem_type(xen_xc, xen_domid, mem_type,
504 start_addr >> TARGET_PAGE_BITS,
505 size >> TARGET_PAGE_BITS)) {
506 DPRINTF("xc_hvm_set_mem_type error, addr: "TARGET_FMT_plx"\n",
507 start_addr);
510 } else {
511 if (xen_remove_from_physmap(state, start_addr, size) < 0) {
512 DPRINTF("physmapping does not exist at "TARGET_FMT_plx"\n", start_addr);
517 static void xen_region_add(MemoryListener *listener,
518 MemoryRegionSection *section)
520 memory_region_ref(section->mr);
521 xen_set_memory(listener, section, true);
524 static void xen_region_del(MemoryListener *listener,
525 MemoryRegionSection *section)
527 xen_set_memory(listener, section, false);
528 memory_region_unref(section->mr);
531 static void xen_io_add(MemoryListener *listener,
532 MemoryRegionSection *section)
534 XenIOState *state = container_of(listener, XenIOState, io_listener);
536 memory_region_ref(section->mr);
538 xen_map_io_section(xen_xc, xen_domid, state->ioservid, section);
541 static void xen_io_del(MemoryListener *listener,
542 MemoryRegionSection *section)
544 XenIOState *state = container_of(listener, XenIOState, io_listener);
546 xen_unmap_io_section(xen_xc, xen_domid, state->ioservid, section);
548 memory_region_unref(section->mr);
551 static void xen_device_realize(DeviceListener *listener,
552 DeviceState *dev)
554 XenIOState *state = container_of(listener, XenIOState, device_listener);
556 if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
557 PCIDevice *pci_dev = PCI_DEVICE(dev);
559 xen_map_pcidev(xen_xc, xen_domid, state->ioservid, pci_dev);
563 static void xen_device_unrealize(DeviceListener *listener,
564 DeviceState *dev)
566 XenIOState *state = container_of(listener, XenIOState, device_listener);
568 if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
569 PCIDevice *pci_dev = PCI_DEVICE(dev);
571 xen_unmap_pcidev(xen_xc, xen_domid, state->ioservid, pci_dev);
575 static void xen_sync_dirty_bitmap(XenIOState *state,
576 hwaddr start_addr,
577 ram_addr_t size)
579 hwaddr npages = size >> TARGET_PAGE_BITS;
580 const int width = sizeof(unsigned long) * 8;
581 unsigned long bitmap[(npages + width - 1) / width];
582 int rc, i, j;
583 const XenPhysmap *physmap = NULL;
585 physmap = get_physmapping(state, start_addr, size);
586 if (physmap == NULL) {
587 /* not handled */
588 return;
591 if (state->log_for_dirtybit == NULL) {
592 state->log_for_dirtybit = physmap;
593 } else if (state->log_for_dirtybit != physmap) {
594 /* Only one range for dirty bitmap can be tracked. */
595 return;
598 rc = xc_hvm_track_dirty_vram(xen_xc, xen_domid,
599 start_addr >> TARGET_PAGE_BITS, npages,
600 bitmap);
601 if (rc < 0) {
602 #ifndef ENODATA
603 #define ENODATA ENOENT
604 #endif
605 if (errno == ENODATA) {
606 memory_region_set_dirty(framebuffer, 0, size);
607 DPRINTF("xen: track_dirty_vram failed (0x" TARGET_FMT_plx
608 ", 0x" TARGET_FMT_plx "): %s\n",
609 start_addr, start_addr + size, strerror(errno));
611 return;
614 for (i = 0; i < ARRAY_SIZE(bitmap); i++) {
615 unsigned long map = bitmap[i];
616 while (map != 0) {
617 j = ctzl(map);
618 map &= ~(1ul << j);
619 memory_region_set_dirty(framebuffer,
620 (i * width + j) * TARGET_PAGE_SIZE,
621 TARGET_PAGE_SIZE);
626 static void xen_log_start(MemoryListener *listener,
627 MemoryRegionSection *section,
628 int old, int new)
630 XenIOState *state = container_of(listener, XenIOState, memory_listener);
632 if (new & ~old & (1 << DIRTY_MEMORY_VGA)) {
633 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
634 int128_get64(section->size));
638 static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section,
639 int old, int new)
641 XenIOState *state = container_of(listener, XenIOState, memory_listener);
643 if (old & ~new & (1 << DIRTY_MEMORY_VGA)) {
644 state->log_for_dirtybit = NULL;
645 /* Disable dirty bit tracking */
646 xc_hvm_track_dirty_vram(xen_xc, xen_domid, 0, 0, NULL);
650 static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section)
652 XenIOState *state = container_of(listener, XenIOState, memory_listener);
654 xen_sync_dirty_bitmap(state, section->offset_within_address_space,
655 int128_get64(section->size));
658 static void xen_log_global_start(MemoryListener *listener)
660 if (xen_enabled()) {
661 xen_in_migration = true;
665 static void xen_log_global_stop(MemoryListener *listener)
667 xen_in_migration = false;
670 static MemoryListener xen_memory_listener = {
671 .region_add = xen_region_add,
672 .region_del = xen_region_del,
673 .log_start = xen_log_start,
674 .log_stop = xen_log_stop,
675 .log_sync = xen_log_sync,
676 .log_global_start = xen_log_global_start,
677 .log_global_stop = xen_log_global_stop,
678 .priority = 10,
681 static MemoryListener xen_io_listener = {
682 .region_add = xen_io_add,
683 .region_del = xen_io_del,
684 .priority = 10,
687 static DeviceListener xen_device_listener = {
688 .realize = xen_device_realize,
689 .unrealize = xen_device_unrealize,
692 /* get the ioreq packets from share mem */
693 static ioreq_t *cpu_get_ioreq_from_shared_memory(XenIOState *state, int vcpu)
695 ioreq_t *req = xen_vcpu_ioreq(state->shared_page, vcpu);
697 if (req->state != STATE_IOREQ_READY) {
698 DPRINTF("I/O request not ready: "
699 "%x, ptr: %x, port: %"PRIx64", "
700 "data: %"PRIx64", count: %" FMT_ioreq_size ", size: %" FMT_ioreq_size "\n",
701 req->state, req->data_is_ptr, req->addr,
702 req->data, req->count, req->size);
703 return NULL;
706 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
708 req->state = STATE_IOREQ_INPROCESS;
709 return req;
712 /* use poll to get the port notification */
713 /* ioreq_vec--out,the */
714 /* retval--the number of ioreq packet */
715 static ioreq_t *cpu_get_ioreq(XenIOState *state)
717 int i;
718 evtchn_port_t port;
720 port = xenevtchn_pending(state->xce_handle);
721 if (port == state->bufioreq_local_port) {
722 timer_mod(state->buffered_io_timer,
723 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
724 return NULL;
727 if (port != -1) {
728 for (i = 0; i < max_cpus; i++) {
729 if (state->ioreq_local_port[i] == port) {
730 break;
734 if (i == max_cpus) {
735 hw_error("Fatal error while trying to get io event!\n");
738 /* unmask the wanted port again */
739 xenevtchn_unmask(state->xce_handle, port);
741 /* get the io packet from shared memory */
742 state->send_vcpu = i;
743 return cpu_get_ioreq_from_shared_memory(state, i);
746 /* read error or read nothing */
747 return NULL;
750 static uint32_t do_inp(pio_addr_t addr, unsigned long size)
752 switch (size) {
753 case 1:
754 return cpu_inb(addr);
755 case 2:
756 return cpu_inw(addr);
757 case 4:
758 return cpu_inl(addr);
759 default:
760 hw_error("inp: bad size: %04"FMT_pioaddr" %lx", addr, size);
764 static void do_outp(pio_addr_t addr,
765 unsigned long size, uint32_t val)
767 switch (size) {
768 case 1:
769 return cpu_outb(addr, val);
770 case 2:
771 return cpu_outw(addr, val);
772 case 4:
773 return cpu_outl(addr, val);
774 default:
775 hw_error("outp: bad size: %04"FMT_pioaddr" %lx", addr, size);
780 * Helper functions which read/write an object from/to physical guest
781 * memory, as part of the implementation of an ioreq.
783 * Equivalent to
784 * cpu_physical_memory_rw(addr + (req->df ? -1 : +1) * req->size * i,
785 * val, req->size, 0/1)
786 * except without the integer overflow problems.
788 static void rw_phys_req_item(hwaddr addr,
789 ioreq_t *req, uint32_t i, void *val, int rw)
791 /* Do everything unsigned so overflow just results in a truncated result
792 * and accesses to undesired parts of guest memory, which is up
793 * to the guest */
794 hwaddr offset = (hwaddr)req->size * i;
795 if (req->df) {
796 addr -= offset;
797 } else {
798 addr += offset;
800 cpu_physical_memory_rw(addr, val, req->size, rw);
803 static inline void read_phys_req_item(hwaddr addr,
804 ioreq_t *req, uint32_t i, void *val)
806 rw_phys_req_item(addr, req, i, val, 0);
808 static inline void write_phys_req_item(hwaddr addr,
809 ioreq_t *req, uint32_t i, void *val)
811 rw_phys_req_item(addr, req, i, val, 1);
815 static void cpu_ioreq_pio(ioreq_t *req)
817 uint32_t i;
819 trace_cpu_ioreq_pio(req, req->dir, req->df, req->data_is_ptr, req->addr,
820 req->data, req->count, req->size);
822 if (req->dir == IOREQ_READ) {
823 if (!req->data_is_ptr) {
824 req->data = do_inp(req->addr, req->size);
825 trace_cpu_ioreq_pio_read_reg(req, req->data, req->addr,
826 req->size);
827 } else {
828 uint32_t tmp;
830 for (i = 0; i < req->count; i++) {
831 tmp = do_inp(req->addr, req->size);
832 write_phys_req_item(req->data, req, i, &tmp);
835 } else if (req->dir == IOREQ_WRITE) {
836 if (!req->data_is_ptr) {
837 trace_cpu_ioreq_pio_write_reg(req, req->data, req->addr,
838 req->size);
839 do_outp(req->addr, req->size, req->data);
840 } else {
841 for (i = 0; i < req->count; i++) {
842 uint32_t tmp = 0;
844 read_phys_req_item(req->data, req, i, &tmp);
845 do_outp(req->addr, req->size, tmp);
851 static void cpu_ioreq_move(ioreq_t *req)
853 uint32_t i;
855 trace_cpu_ioreq_move(req, req->dir, req->df, req->data_is_ptr, req->addr,
856 req->data, req->count, req->size);
858 if (!req->data_is_ptr) {
859 if (req->dir == IOREQ_READ) {
860 for (i = 0; i < req->count; i++) {
861 read_phys_req_item(req->addr, req, i, &req->data);
863 } else if (req->dir == IOREQ_WRITE) {
864 for (i = 0; i < req->count; i++) {
865 write_phys_req_item(req->addr, req, i, &req->data);
868 } else {
869 uint64_t tmp;
871 if (req->dir == IOREQ_READ) {
872 for (i = 0; i < req->count; i++) {
873 read_phys_req_item(req->addr, req, i, &tmp);
874 write_phys_req_item(req->data, req, i, &tmp);
876 } else if (req->dir == IOREQ_WRITE) {
877 for (i = 0; i < req->count; i++) {
878 read_phys_req_item(req->data, req, i, &tmp);
879 write_phys_req_item(req->addr, req, i, &tmp);
885 static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req)
887 X86CPU *cpu;
888 CPUX86State *env;
890 cpu = X86_CPU(current_cpu);
891 env = &cpu->env;
892 env->regs[R_EAX] = req->data;
893 env->regs[R_EBX] = vmport_regs->ebx;
894 env->regs[R_ECX] = vmport_regs->ecx;
895 env->regs[R_EDX] = vmport_regs->edx;
896 env->regs[R_ESI] = vmport_regs->esi;
897 env->regs[R_EDI] = vmport_regs->edi;
900 static void regs_from_cpu(vmware_regs_t *vmport_regs)
902 X86CPU *cpu = X86_CPU(current_cpu);
903 CPUX86State *env = &cpu->env;
905 vmport_regs->ebx = env->regs[R_EBX];
906 vmport_regs->ecx = env->regs[R_ECX];
907 vmport_regs->edx = env->regs[R_EDX];
908 vmport_regs->esi = env->regs[R_ESI];
909 vmport_regs->edi = env->regs[R_EDI];
912 static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req)
914 vmware_regs_t *vmport_regs;
916 assert(state->shared_vmport_page);
917 vmport_regs =
918 &state->shared_vmport_page->vcpu_vmport_regs[state->send_vcpu];
919 QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs));
921 current_cpu = state->cpu_by_vcpu_id[state->send_vcpu];
922 regs_to_cpu(vmport_regs, req);
923 cpu_ioreq_pio(req);
924 regs_from_cpu(vmport_regs);
925 current_cpu = NULL;
928 static void handle_ioreq(XenIOState *state, ioreq_t *req)
930 trace_handle_ioreq(req, req->type, req->dir, req->df, req->data_is_ptr,
931 req->addr, req->data, req->count, req->size);
933 if (!req->data_is_ptr && (req->dir == IOREQ_WRITE) &&
934 (req->size < sizeof (target_ulong))) {
935 req->data &= ((target_ulong) 1 << (8 * req->size)) - 1;
938 if (req->dir == IOREQ_WRITE)
939 trace_handle_ioreq_write(req, req->type, req->df, req->data_is_ptr,
940 req->addr, req->data, req->count, req->size);
942 switch (req->type) {
943 case IOREQ_TYPE_PIO:
944 cpu_ioreq_pio(req);
945 break;
946 case IOREQ_TYPE_COPY:
947 cpu_ioreq_move(req);
948 break;
949 case IOREQ_TYPE_VMWARE_PORT:
950 handle_vmport_ioreq(state, req);
951 break;
952 case IOREQ_TYPE_TIMEOFFSET:
953 break;
954 case IOREQ_TYPE_INVALIDATE:
955 xen_invalidate_map_cache();
956 break;
957 case IOREQ_TYPE_PCI_CONFIG: {
958 uint32_t sbdf = req->addr >> 32;
959 uint32_t val;
961 /* Fake a write to port 0xCF8 so that
962 * the config space access will target the
963 * correct device model.
965 val = (1u << 31) |
966 ((req->addr & 0x0f00) << 16) |
967 ((sbdf & 0xffff) << 8) |
968 (req->addr & 0xfc);
969 do_outp(0xcf8, 4, val);
971 /* Now issue the config space access via
972 * port 0xCFC
974 req->addr = 0xcfc | (req->addr & 0x03);
975 cpu_ioreq_pio(req);
976 break;
978 default:
979 hw_error("Invalid ioreq type 0x%x\n", req->type);
981 if (req->dir == IOREQ_READ) {
982 trace_handle_ioreq_read(req, req->type, req->df, req->data_is_ptr,
983 req->addr, req->data, req->count, req->size);
987 static int handle_buffered_iopage(XenIOState *state)
989 buffered_iopage_t *buf_page = state->buffered_io_page;
990 buf_ioreq_t *buf_req = NULL;
991 ioreq_t req;
992 int qw;
994 if (!buf_page) {
995 return 0;
998 memset(&req, 0x00, sizeof(req));
1000 for (;;) {
1001 uint32_t rdptr = buf_page->read_pointer, wrptr;
1003 xen_rmb();
1004 wrptr = buf_page->write_pointer;
1005 xen_rmb();
1006 if (rdptr != buf_page->read_pointer) {
1007 continue;
1009 if (rdptr == wrptr) {
1010 break;
1012 buf_req = &buf_page->buf_ioreq[rdptr % IOREQ_BUFFER_SLOT_NUM];
1013 req.size = 1UL << buf_req->size;
1014 req.count = 1;
1015 req.addr = buf_req->addr;
1016 req.data = buf_req->data;
1017 req.state = STATE_IOREQ_READY;
1018 req.dir = buf_req->dir;
1019 req.df = 1;
1020 req.type = buf_req->type;
1021 req.data_is_ptr = 0;
1022 qw = (req.size == 8);
1023 if (qw) {
1024 buf_req = &buf_page->buf_ioreq[(rdptr + 1) %
1025 IOREQ_BUFFER_SLOT_NUM];
1026 req.data |= ((uint64_t)buf_req->data) << 32;
1029 handle_ioreq(state, &req);
1031 atomic_add(&buf_page->read_pointer, qw + 1);
1034 return req.count;
1037 static void handle_buffered_io(void *opaque)
1039 XenIOState *state = opaque;
1041 if (handle_buffered_iopage(state)) {
1042 timer_mod(state->buffered_io_timer,
1043 BUFFER_IO_MAX_DELAY + qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
1044 } else {
1045 timer_del(state->buffered_io_timer);
1046 xenevtchn_unmask(state->xce_handle, state->bufioreq_local_port);
1050 static void cpu_handle_ioreq(void *opaque)
1052 XenIOState *state = opaque;
1053 ioreq_t *req = cpu_get_ioreq(state);
1055 handle_buffered_iopage(state);
1056 if (req) {
1057 handle_ioreq(state, req);
1059 if (req->state != STATE_IOREQ_INPROCESS) {
1060 fprintf(stderr, "Badness in I/O request ... not in service?!: "
1061 "%x, ptr: %x, port: %"PRIx64", "
1062 "data: %"PRIx64", count: %" FMT_ioreq_size
1063 ", size: %" FMT_ioreq_size
1064 ", type: %"FMT_ioreq_size"\n",
1065 req->state, req->data_is_ptr, req->addr,
1066 req->data, req->count, req->size, req->type);
1067 destroy_hvm_domain(false);
1068 return;
1071 xen_wmb(); /* Update ioreq contents /then/ update state. */
1074 * We do this before we send the response so that the tools
1075 * have the opportunity to pick up on the reset before the
1076 * guest resumes and does a hlt with interrupts disabled which
1077 * causes Xen to powerdown the domain.
1079 if (runstate_is_running()) {
1080 if (qemu_shutdown_requested_get()) {
1081 destroy_hvm_domain(false);
1083 if (qemu_reset_requested_get()) {
1084 qemu_system_reset(VMRESET_REPORT);
1085 destroy_hvm_domain(true);
1089 req->state = STATE_IORESP_READY;
1090 xenevtchn_notify(state->xce_handle,
1091 state->ioreq_local_port[state->send_vcpu]);
1095 static void xen_main_loop_prepare(XenIOState *state)
1097 int evtchn_fd = -1;
1099 if (state->xce_handle != NULL) {
1100 evtchn_fd = xenevtchn_fd(state->xce_handle);
1103 state->buffered_io_timer = timer_new_ms(QEMU_CLOCK_REALTIME, handle_buffered_io,
1104 state);
1106 if (evtchn_fd != -1) {
1107 CPUState *cpu_state;
1109 DPRINTF("%s: Init cpu_by_vcpu_id\n", __func__);
1110 CPU_FOREACH(cpu_state) {
1111 DPRINTF("%s: cpu_by_vcpu_id[%d]=%p\n",
1112 __func__, cpu_state->cpu_index, cpu_state);
1113 state->cpu_by_vcpu_id[cpu_state->cpu_index] = cpu_state;
1115 qemu_set_fd_handler(evtchn_fd, cpu_handle_ioreq, NULL, state);
1120 static void xen_hvm_change_state_handler(void *opaque, int running,
1121 RunState rstate)
1123 XenIOState *state = opaque;
1125 if (running) {
1126 xen_main_loop_prepare(state);
1129 xen_set_ioreq_server_state(xen_xc, xen_domid,
1130 state->ioservid,
1131 (rstate == RUN_STATE_RUNNING));
1134 static void xen_exit_notifier(Notifier *n, void *data)
1136 XenIOState *state = container_of(n, XenIOState, exit);
1138 xenevtchn_close(state->xce_handle);
1139 xs_daemon_close(state->xenstore);
1142 static void xen_read_physmap(XenIOState *state)
1144 XenPhysmap *physmap = NULL;
1145 unsigned int len, num, i;
1146 char path[80], *value = NULL;
1147 char **entries = NULL;
1149 snprintf(path, sizeof(path),
1150 "/local/domain/0/device-model/%d/physmap", xen_domid);
1151 entries = xs_directory(state->xenstore, 0, path, &num);
1152 if (entries == NULL)
1153 return;
1155 for (i = 0; i < num; i++) {
1156 physmap = g_malloc(sizeof (XenPhysmap));
1157 physmap->phys_offset = strtoull(entries[i], NULL, 16);
1158 snprintf(path, sizeof(path),
1159 "/local/domain/0/device-model/%d/physmap/%s/start_addr",
1160 xen_domid, entries[i]);
1161 value = xs_read(state->xenstore, 0, path, &len);
1162 if (value == NULL) {
1163 g_free(physmap);
1164 continue;
1166 physmap->start_addr = strtoull(value, NULL, 16);
1167 free(value);
1169 snprintf(path, sizeof(path),
1170 "/local/domain/0/device-model/%d/physmap/%s/size",
1171 xen_domid, entries[i]);
1172 value = xs_read(state->xenstore, 0, path, &len);
1173 if (value == NULL) {
1174 g_free(physmap);
1175 continue;
1177 physmap->size = strtoull(value, NULL, 16);
1178 free(value);
1180 snprintf(path, sizeof(path),
1181 "/local/domain/0/device-model/%d/physmap/%s/name",
1182 xen_domid, entries[i]);
1183 physmap->name = xs_read(state->xenstore, 0, path, &len);
1185 QLIST_INSERT_HEAD(&state->physmap, physmap, list);
1187 free(entries);
1190 static void xen_wakeup_notifier(Notifier *notifier, void *data)
1192 xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0);
1195 void xen_hvm_init(PCMachineState *pcms, MemoryRegion **ram_memory)
1197 int i, rc;
1198 xen_pfn_t ioreq_pfn;
1199 xen_pfn_t bufioreq_pfn;
1200 evtchn_port_t bufioreq_evtchn;
1201 XenIOState *state;
1203 state = g_malloc0(sizeof (XenIOState));
1205 state->xce_handle = xenevtchn_open(NULL, 0);
1206 if (state->xce_handle == NULL) {
1207 perror("xen: event channel open");
1208 goto err;
1211 state->xenstore = xs_daemon_open();
1212 if (state->xenstore == NULL) {
1213 perror("xen: xenstore open");
1214 goto err;
1217 rc = xen_create_ioreq_server(xen_xc, xen_domid, &state->ioservid);
1218 if (rc < 0) {
1219 perror("xen: ioreq server create");
1220 goto err;
1223 state->exit.notify = xen_exit_notifier;
1224 qemu_add_exit_notifier(&state->exit);
1226 state->suspend.notify = xen_suspend_notifier;
1227 qemu_register_suspend_notifier(&state->suspend);
1229 state->wakeup.notify = xen_wakeup_notifier;
1230 qemu_register_wakeup_notifier(&state->wakeup);
1232 rc = xen_get_ioreq_server_info(xen_xc, xen_domid, state->ioservid,
1233 &ioreq_pfn, &bufioreq_pfn,
1234 &bufioreq_evtchn);
1235 if (rc < 0) {
1236 error_report("failed to get ioreq server info: error %d handle=" XC_INTERFACE_FMT,
1237 errno, xen_xc);
1238 goto err;
1241 DPRINTF("shared page at pfn %lx\n", ioreq_pfn);
1242 DPRINTF("buffered io page at pfn %lx\n", bufioreq_pfn);
1243 DPRINTF("buffered io evtchn is %x\n", bufioreq_evtchn);
1245 state->shared_page = xenforeignmemory_map(xen_fmem, xen_domid,
1246 PROT_READ|PROT_WRITE,
1247 1, &ioreq_pfn, NULL);
1248 if (state->shared_page == NULL) {
1249 error_report("map shared IO page returned error %d handle=" XC_INTERFACE_FMT,
1250 errno, xen_xc);
1251 goto err;
1254 rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn);
1255 if (!rc) {
1256 DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn);
1257 state->shared_vmport_page =
1258 xenforeignmemory_map(xen_fmem, xen_domid, PROT_READ|PROT_WRITE,
1259 1, &ioreq_pfn, NULL);
1260 if (state->shared_vmport_page == NULL) {
1261 error_report("map shared vmport IO page returned error %d handle="
1262 XC_INTERFACE_FMT, errno, xen_xc);
1263 goto err;
1265 } else if (rc != -ENOSYS) {
1266 error_report("get vmport regs pfn returned error %d, rc=%d",
1267 errno, rc);
1268 goto err;
1271 state->buffered_io_page = xenforeignmemory_map(xen_fmem, xen_domid,
1272 PROT_READ|PROT_WRITE,
1273 1, &bufioreq_pfn, NULL);
1274 if (state->buffered_io_page == NULL) {
1275 error_report("map buffered IO page returned error %d", errno);
1276 goto err;
1279 /* Note: cpus is empty at this point in init */
1280 state->cpu_by_vcpu_id = g_malloc0(max_cpus * sizeof(CPUState *));
1282 rc = xen_set_ioreq_server_state(xen_xc, xen_domid, state->ioservid, true);
1283 if (rc < 0) {
1284 error_report("failed to enable ioreq server info: error %d handle=" XC_INTERFACE_FMT,
1285 errno, xen_xc);
1286 goto err;
1289 state->ioreq_local_port = g_malloc0(max_cpus * sizeof (evtchn_port_t));
1291 /* FIXME: how about if we overflow the page here? */
1292 for (i = 0; i < max_cpus; i++) {
1293 rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1294 xen_vcpu_eport(state->shared_page, i));
1295 if (rc == -1) {
1296 error_report("shared evtchn %d bind error %d", i, errno);
1297 goto err;
1299 state->ioreq_local_port[i] = rc;
1302 rc = xenevtchn_bind_interdomain(state->xce_handle, xen_domid,
1303 bufioreq_evtchn);
1304 if (rc == -1) {
1305 error_report("buffered evtchn bind error %d", errno);
1306 goto err;
1308 state->bufioreq_local_port = rc;
1310 /* Init RAM management */
1311 xen_map_cache_init(xen_phys_offset_to_gaddr, state);
1312 xen_ram_init(pcms, ram_size, ram_memory);
1314 qemu_add_vm_change_state_handler(xen_hvm_change_state_handler, state);
1316 state->memory_listener = xen_memory_listener;
1317 QLIST_INIT(&state->physmap);
1318 memory_listener_register(&state->memory_listener, &address_space_memory);
1319 state->log_for_dirtybit = NULL;
1321 state->io_listener = xen_io_listener;
1322 memory_listener_register(&state->io_listener, &address_space_io);
1324 state->device_listener = xen_device_listener;
1325 device_listener_register(&state->device_listener);
1327 /* Initialize backend core & drivers */
1328 if (xen_be_init() != 0) {
1329 error_report("xen backend core setup failed");
1330 goto err;
1332 xen_be_register("console", &xen_console_ops);
1333 xen_be_register("vkbd", &xen_kbdmouse_ops);
1334 xen_be_register("qdisk", &xen_blkdev_ops);
1335 xen_read_physmap(state);
1336 return;
1338 err:
1339 error_report("xen hardware virtual machine initialisation failed");
1340 exit(1);
1343 void destroy_hvm_domain(bool reboot)
1345 XenXC xc_handle;
1346 int sts;
1348 xc_handle = xen_xc_interface_open(0, 0, 0);
1349 if (xc_handle == XC_HANDLER_INITIAL_VALUE) {
1350 fprintf(stderr, "Cannot acquire xenctrl handle\n");
1351 } else {
1352 sts = xc_domain_shutdown(xc_handle, xen_domid,
1353 reboot ? SHUTDOWN_reboot : SHUTDOWN_poweroff);
1354 if (sts != 0) {
1355 fprintf(stderr, "xc_domain_shutdown failed to issue %s, "
1356 "sts %d, %s\n", reboot ? "reboot" : "poweroff",
1357 sts, strerror(errno));
1358 } else {
1359 fprintf(stderr, "Issued domain %d %s\n", xen_domid,
1360 reboot ? "reboot" : "poweroff");
1362 xc_interface_close(xc_handle);
1366 void xen_register_framebuffer(MemoryRegion *mr)
1368 framebuffer = mr;
1371 void xen_shutdown_fatal_error(const char *fmt, ...)
1373 va_list ap;
1375 va_start(ap, fmt);
1376 vfprintf(stderr, fmt, ap);
1377 va_end(ap);
1378 fprintf(stderr, "Will destroy the domain.\n");
1379 /* destroy the domain */
1380 qemu_system_shutdown_request();
1383 void xen_modified_memory(ram_addr_t start, ram_addr_t length)
1385 if (unlikely(xen_in_migration)) {
1386 int rc;
1387 ram_addr_t start_pfn, nb_pages;
1389 if (length == 0) {
1390 length = TARGET_PAGE_SIZE;
1392 start_pfn = start >> TARGET_PAGE_BITS;
1393 nb_pages = ((start + length + TARGET_PAGE_SIZE - 1) >> TARGET_PAGE_BITS)
1394 - start_pfn;
1395 rc = xc_hvm_modified_memory(xen_xc, xen_domid, start_pfn, nb_pages);
1396 if (rc) {
1397 fprintf(stderr,
1398 "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n",
1399 __func__, start, nb_pages, rc, strerror(-rc));
1404 void qmp_xen_set_global_dirty_log(bool enable, Error **errp)
1406 if (enable) {
1407 memory_global_dirty_log_start();
1408 } else {
1409 memory_global_dirty_log_stop();