| 1 | /* |
| 2 | * Copyright (C) 2010 Citrix Ltd. |
| 3 | * |
| 4 | * This work is licensed under the terms of the GNU GPL, version 2. See |
| 5 | * the COPYING file in the top-level directory. |
| 6 | * |
| 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 | */ |
| 10 | |
| 11 | #include "qemu/osdep.h" |
| 12 | #include "qemu/units.h" |
| 13 | #include "qemu/error-report.h" |
| 14 | #include "qapi/error.h" |
| 15 | #include "qapi/qapi-commands-migration.h" |
| 16 | #include "trace.h" |
| 17 | |
| 18 | #include "hw/core/hw-error.h" |
| 19 | #include "hw/i386/pc.h" |
| 20 | #include "hw/core/irq.h" |
| 21 | #include "hw/i386/apic-msidef.h" |
| 22 | #include "hw/xen/xen-x86.h" |
| 23 | #include "qemu/range.h" |
| 24 | |
| 25 | #include "hw/xen/xen-hvm-common.h" |
| 26 | #include "hw/xen/arch_hvm.h" |
| 27 | #include <xen/hvm/e820.h> |
| 28 | #include "exec/target_page.h" |
| 29 | #include "target/i386/cpu.h" |
| 30 | #include "system/runstate.h" |
| 31 | #include "system/xen-mapcache.h" |
| 32 | #include "system/xen.h" |
| 33 | |
| 34 | static MemoryRegion ram_640k, ram_lo, ram_hi; |
| 35 | static MemoryRegion *framebuffer; |
| 36 | static bool xen_in_migration; |
| 37 | |
| 38 | /* Compatibility with older version */ |
| 39 | |
| 40 | /* |
| 41 | * This allows QEMU to build on a system that has Xen 4.5 or earlier installed. |
| 42 | * This is here (not in hw/xen/xen_native.h) because xen/hvm/ioreq.h needs to |
| 43 | * be included before this block and hw/xen/xen_native.h needs to be included |
| 44 | * before xen/hvm/ioreq.h |
| 45 | */ |
| 46 | #ifndef IOREQ_TYPE_VMWARE_PORT |
| 47 | #define IOREQ_TYPE_VMWARE_PORT 3 |
| 48 | struct vmware_regs { |
| 49 | uint32_t esi; |
| 50 | uint32_t edi; |
| 51 | uint32_t ebx; |
| 52 | uint32_t ecx; |
| 53 | uint32_t edx; |
| 54 | }; |
| 55 | typedef struct vmware_regs vmware_regs_t; |
| 56 | |
| 57 | struct shared_vmport_iopage { |
| 58 | struct vmware_regs vcpu_vmport_regs[1]; |
| 59 | }; |
| 60 | typedef struct shared_vmport_iopage shared_vmport_iopage_t; |
| 61 | #endif |
| 62 | |
| 63 | static shared_vmport_iopage_t *shared_vmport_page; |
| 64 | |
| 65 | static QLIST_HEAD(, XenPhysmap) xen_physmap; |
| 66 | static const XenPhysmap *log_for_dirtybit; |
| 67 | /* Buffer used by xen_sync_dirty_bitmap */ |
| 68 | static unsigned long *dirty_bitmap; |
| 69 | static Notifier suspend; |
| 70 | static Notifier wakeup; |
| 71 | |
| 72 | /* Xen specific function for piix pci */ |
| 73 | |
| 74 | int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num) |
| 75 | { |
| 76 | return irq_num + (PCI_SLOT(pci_dev->devfn) << 2); |
| 77 | } |
| 78 | |
| 79 | void xen_intx_set_irq(void *opaque, int irq_num, int level) |
| 80 | { |
| 81 | xen_set_pci_intx_level(xen_domid, 0, 0, irq_num >> 2, |
| 82 | irq_num & 3, level); |
| 83 | } |
| 84 | |
| 85 | int xen_set_pci_link_route(uint8_t link, uint8_t irq) |
| 86 | { |
| 87 | return xendevicemodel_set_pci_link_route(xen_dmod, xen_domid, link, irq); |
| 88 | } |
| 89 | |
| 90 | int xen_is_pirq_msi(uint32_t msi_data) |
| 91 | { |
| 92 | /* If vector is 0, the msi is remapped into a pirq, passed as |
| 93 | * dest_id. |
| 94 | */ |
| 95 | return ((msi_data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT) == 0; |
| 96 | } |
| 97 | |
| 98 | void xen_hvm_inject_msi(uint64_t addr, uint32_t data) |
| 99 | { |
| 100 | xen_inject_msi(xen_domid, addr, data); |
| 101 | } |
| 102 | |
| 103 | static void xen_suspend_notifier(Notifier *notifier, void *data) |
| 104 | { |
| 105 | xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 3); |
| 106 | } |
| 107 | |
| 108 | /* Xen Interrupt Controller */ |
| 109 | |
| 110 | static void xen_set_irq(void *opaque, int irq, int level) |
| 111 | { |
| 112 | xen_set_isa_irq_level(xen_domid, irq, level); |
| 113 | } |
| 114 | |
| 115 | qemu_irq *xen_interrupt_controller_init(void) |
| 116 | { |
| 117 | return qemu_allocate_irqs(xen_set_irq, NULL, 16); |
| 118 | } |
| 119 | |
| 120 | /* Memory Ops */ |
| 121 | |
| 122 | static void xen_ram_init(PCMachineState *pcms, |
| 123 | ram_addr_t ram_size, MemoryRegion **ram_memory_p) |
| 124 | { |
| 125 | X86MachineState *x86ms = X86_MACHINE(pcms); |
| 126 | MemoryRegion *sysmem = get_system_memory(); |
| 127 | ram_addr_t block_len; |
| 128 | uint64_t user_lowmem = |
| 129 | object_property_get_uint(qdev_get_machine(), |
| 130 | PC_MACHINE_MAX_RAM_BELOW_4G, |
| 131 | &error_abort); |
| 132 | |
| 133 | /* Handle the machine opt max-ram-below-4g. It is basically doing |
| 134 | * min(xen limit, user limit). |
| 135 | */ |
| 136 | if (!user_lowmem) { |
| 137 | user_lowmem = HVM_BELOW_4G_RAM_END; /* default */ |
| 138 | } |
| 139 | if (HVM_BELOW_4G_RAM_END <= user_lowmem) { |
| 140 | user_lowmem = HVM_BELOW_4G_RAM_END; |
| 141 | } |
| 142 | |
| 143 | if (ram_size >= user_lowmem) { |
| 144 | x86ms->above_4g_mem_size = ram_size - user_lowmem; |
| 145 | x86ms->below_4g_mem_size = user_lowmem; |
| 146 | } else { |
| 147 | x86ms->above_4g_mem_size = 0; |
| 148 | x86ms->below_4g_mem_size = ram_size; |
| 149 | } |
| 150 | if (!x86ms->above_4g_mem_size) { |
| 151 | block_len = ram_size; |
| 152 | } else { |
| 153 | /* |
| 154 | * Xen does not allocate the memory continuously, it keeps a |
| 155 | * hole of the size computed above or passed in. |
| 156 | */ |
| 157 | block_len = (4 * GiB) + x86ms->above_4g_mem_size; |
| 158 | } |
| 159 | memory_region_init_ram(&xen_memory, NULL, "xen.ram", block_len, |
| 160 | &error_fatal); |
| 161 | *ram_memory_p = &xen_memory; |
| 162 | |
| 163 | memory_region_init_alias(&ram_640k, NULL, "xen.ram.640k", |
| 164 | &xen_memory, 0, 0xa0000); |
| 165 | memory_region_add_subregion(sysmem, 0, &ram_640k); |
| 166 | /* Skip of the VGA IO memory space, it will be registered later by the VGA |
| 167 | * emulated device. |
| 168 | * |
| 169 | * The area between 0xc0000 and 0x100000 will be used by SeaBIOS to load |
| 170 | * the Options ROM, so it is registered here as RAM. |
| 171 | */ |
| 172 | memory_region_init_alias(&ram_lo, NULL, "xen.ram.lo", |
| 173 | &xen_memory, 0xc0000, |
| 174 | x86ms->below_4g_mem_size - 0xc0000); |
| 175 | memory_region_add_subregion(sysmem, 0xc0000, &ram_lo); |
| 176 | if (x86ms->above_4g_mem_size > 0) { |
| 177 | memory_region_init_alias(&ram_hi, NULL, "xen.ram.hi", |
| 178 | &xen_memory, 0x100000000ULL, |
| 179 | x86ms->above_4g_mem_size); |
| 180 | memory_region_add_subregion(sysmem, 0x100000000ULL, &ram_hi); |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | static XenPhysmap *get_physmapping(hwaddr start_addr, ram_addr_t size, |
| 185 | int page_mask) |
| 186 | { |
| 187 | XenPhysmap *physmap; |
| 188 | |
| 189 | start_addr &= page_mask; |
| 190 | |
| 191 | QLIST_FOREACH(physmap, &xen_physmap, list) { |
| 192 | if (range_covers_byte(physmap->start_addr, physmap->size, start_addr)) { |
| 193 | return physmap; |
| 194 | } |
| 195 | } |
| 196 | return NULL; |
| 197 | } |
| 198 | |
| 199 | static hwaddr xen_phys_offset_to_gaddr(hwaddr phys_offset, ram_addr_t size, |
| 200 | int page_mask) |
| 201 | { |
| 202 | hwaddr addr = phys_offset & page_mask; |
| 203 | XenPhysmap *physmap; |
| 204 | |
| 205 | QLIST_FOREACH(physmap, &xen_physmap, list) { |
| 206 | if (range_covers_byte(physmap->phys_offset, physmap->size, addr)) { |
| 207 | return physmap->start_addr + (phys_offset - physmap->phys_offset); |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | return phys_offset; |
| 212 | } |
| 213 | |
| 214 | #ifdef XEN_COMPAT_PHYSMAP |
| 215 | static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap) |
| 216 | { |
| 217 | char path[80], value[17]; |
| 218 | |
| 219 | snprintf(path, sizeof(path), |
| 220 | "/local/domain/0/device-model/%d/physmap/%"PRIx64"/start_addr", |
| 221 | xen_domid, (uint64_t)physmap->phys_offset); |
| 222 | snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->start_addr); |
| 223 | if (!xs_write(state->xenstore, 0, path, value, strlen(value))) { |
| 224 | return -1; |
| 225 | } |
| 226 | snprintf(path, sizeof(path), |
| 227 | "/local/domain/0/device-model/%d/physmap/%"PRIx64"/size", |
| 228 | xen_domid, (uint64_t)physmap->phys_offset); |
| 229 | snprintf(value, sizeof(value), "%"PRIx64, (uint64_t)physmap->size); |
| 230 | if (!xs_write(state->xenstore, 0, path, value, strlen(value))) { |
| 231 | return -1; |
| 232 | } |
| 233 | if (physmap->name) { |
| 234 | snprintf(path, sizeof(path), |
| 235 | "/local/domain/0/device-model/%d/physmap/%"PRIx64"/name", |
| 236 | xen_domid, (uint64_t)physmap->phys_offset); |
| 237 | if (!xs_write(state->xenstore, 0, path, |
| 238 | physmap->name, strlen(physmap->name))) { |
| 239 | return -1; |
| 240 | } |
| 241 | } |
| 242 | return 0; |
| 243 | } |
| 244 | #else |
| 245 | static int xen_save_physmap(XenIOState *state, XenPhysmap *physmap) |
| 246 | { |
| 247 | return 0; |
| 248 | } |
| 249 | #endif |
| 250 | |
| 251 | static int xen_add_to_physmap(XenIOState *state, |
| 252 | hwaddr start_addr, |
| 253 | ram_addr_t size, |
| 254 | MemoryRegion *mr, |
| 255 | hwaddr offset_within_region) |
| 256 | { |
| 257 | unsigned target_page_bits = qemu_target_page_bits(); |
| 258 | int page_size = qemu_target_page_size(); |
| 259 | int page_mask = -page_size; |
| 260 | unsigned long nr_pages; |
| 261 | int rc = 0; |
| 262 | XenPhysmap *physmap = NULL; |
| 263 | hwaddr pfn, start_gpfn; |
| 264 | hwaddr phys_offset = memory_region_get_ram_addr(mr); |
| 265 | const char *mr_name; |
| 266 | |
| 267 | if (get_physmapping(start_addr, size, page_mask)) { |
| 268 | return 0; |
| 269 | } |
| 270 | if (size <= 0) { |
| 271 | return -1; |
| 272 | } |
| 273 | |
| 274 | /* Xen can only handle a single dirty log region for now and we want |
| 275 | * the linear framebuffer to be that region. |
| 276 | * Avoid tracking any regions that is not videoram and avoid tracking |
| 277 | * the legacy vga region. */ |
| 278 | if (mr == framebuffer && start_addr > 0xbffff) { |
| 279 | goto go_physmap; |
| 280 | } |
| 281 | return -1; |
| 282 | |
| 283 | go_physmap: |
| 284 | DPRINTF("mapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx"\n", |
| 285 | start_addr, start_addr + size); |
| 286 | |
| 287 | mr_name = memory_region_name(mr); |
| 288 | |
| 289 | physmap = g_new(XenPhysmap, 1); |
| 290 | |
| 291 | physmap->start_addr = start_addr; |
| 292 | physmap->size = size; |
| 293 | physmap->name = mr_name; |
| 294 | physmap->phys_offset = phys_offset; |
| 295 | |
| 296 | QLIST_INSERT_HEAD(&xen_physmap, physmap, list); |
| 297 | |
| 298 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
| 299 | /* Now when we have a physmap entry we can replace a dummy mapping with |
| 300 | * a real one of guest foreign memory. */ |
| 301 | uint8_t *p = xen_replace_cache_entry(phys_offset, start_addr, size); |
| 302 | assert(p && p == memory_region_get_ram_ptr(mr)); |
| 303 | |
| 304 | return 0; |
| 305 | } |
| 306 | |
| 307 | pfn = phys_offset >> target_page_bits; |
| 308 | start_gpfn = start_addr >> target_page_bits; |
| 309 | nr_pages = size >> target_page_bits; |
| 310 | rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, nr_pages, pfn, |
| 311 | start_gpfn); |
| 312 | if (rc) { |
| 313 | int saved_errno = errno; |
| 314 | |
| 315 | error_report("relocate_memory %lu pages from GFN %"HWADDR_PRIx |
| 316 | " to GFN %"HWADDR_PRIx" failed: %s", |
| 317 | nr_pages, pfn, start_gpfn, strerror(saved_errno)); |
| 318 | errno = saved_errno; |
| 319 | return -1; |
| 320 | } |
| 321 | |
| 322 | rc = xendevicemodel_pin_memory_cacheattr(xen_dmod, xen_domid, |
| 323 | start_addr >> target_page_bits, |
| 324 | (start_addr + size - 1) >> target_page_bits, |
| 325 | XEN_DOMCTL_MEM_CACHEATTR_WB); |
| 326 | if (rc) { |
| 327 | error_report("pin_memory_cacheattr failed: %s", strerror(errno)); |
| 328 | } |
| 329 | return xen_save_physmap(state, physmap); |
| 330 | } |
| 331 | |
| 332 | static int xen_remove_from_physmap(XenIOState *state, |
| 333 | hwaddr start_addr, |
| 334 | ram_addr_t size) |
| 335 | { |
| 336 | unsigned target_page_bits = qemu_target_page_bits(); |
| 337 | int page_size = qemu_target_page_size(); |
| 338 | int page_mask = -page_size; |
| 339 | int rc = 0; |
| 340 | XenPhysmap *physmap = NULL; |
| 341 | hwaddr phys_offset = 0; |
| 342 | |
| 343 | physmap = get_physmapping(start_addr, size, page_mask); |
| 344 | if (physmap == NULL) { |
| 345 | return -1; |
| 346 | } |
| 347 | |
| 348 | phys_offset = physmap->phys_offset; |
| 349 | size = physmap->size; |
| 350 | |
| 351 | DPRINTF("unmapping vram to %"HWADDR_PRIx" - %"HWADDR_PRIx", at " |
| 352 | "%"HWADDR_PRIx"\n", start_addr, start_addr + size, phys_offset); |
| 353 | |
| 354 | size >>= target_page_bits; |
| 355 | start_addr >>= target_page_bits; |
| 356 | phys_offset >>= target_page_bits; |
| 357 | rc = xendevicemodel_relocate_memory(xen_dmod, xen_domid, size, start_addr, |
| 358 | phys_offset); |
| 359 | if (rc) { |
| 360 | int saved_errno = errno; |
| 361 | |
| 362 | error_report("relocate_memory "RAM_ADDR_FMT" pages" |
| 363 | " from GFN %"HWADDR_PRIx |
| 364 | " to GFN %"HWADDR_PRIx" failed: %s", |
| 365 | size, start_addr, phys_offset, strerror(saved_errno)); |
| 366 | errno = saved_errno; |
| 367 | return -1; |
| 368 | } |
| 369 | |
| 370 | QLIST_REMOVE(physmap, list); |
| 371 | if (log_for_dirtybit == physmap) { |
| 372 | log_for_dirtybit = NULL; |
| 373 | g_free(dirty_bitmap); |
| 374 | dirty_bitmap = NULL; |
| 375 | } |
| 376 | g_free(physmap); |
| 377 | |
| 378 | return 0; |
| 379 | } |
| 380 | |
| 381 | static void xen_sync_dirty_bitmap(XenIOState *state, |
| 382 | hwaddr start_addr, |
| 383 | ram_addr_t size) |
| 384 | { |
| 385 | unsigned target_page_bits = qemu_target_page_bits(); |
| 386 | int page_size = qemu_target_page_size(); |
| 387 | int page_mask = -page_size; |
| 388 | hwaddr npages = size >> target_page_bits; |
| 389 | const int width = sizeof(unsigned long) * 8; |
| 390 | size_t bitmap_size = DIV_ROUND_UP(npages, width); |
| 391 | int rc, i, j; |
| 392 | const XenPhysmap *physmap = NULL; |
| 393 | |
| 394 | physmap = get_physmapping(start_addr, size, page_mask); |
| 395 | if (physmap == NULL) { |
| 396 | /* not handled */ |
| 397 | return; |
| 398 | } |
| 399 | |
| 400 | if (log_for_dirtybit == NULL) { |
| 401 | log_for_dirtybit = physmap; |
| 402 | dirty_bitmap = g_new(unsigned long, bitmap_size); |
| 403 | } else if (log_for_dirtybit != physmap) { |
| 404 | /* Only one range for dirty bitmap can be tracked. */ |
| 405 | return; |
| 406 | } |
| 407 | |
| 408 | rc = xen_track_dirty_vram(xen_domid, start_addr >> target_page_bits, |
| 409 | npages, dirty_bitmap); |
| 410 | if (rc < 0) { |
| 411 | #ifndef ENODATA |
| 412 | #define ENODATA ENOENT |
| 413 | #endif |
| 414 | if (errno == ENODATA) { |
| 415 | memory_region_set_dirty(framebuffer, 0, size); |
| 416 | DPRINTF("xen: track_dirty_vram failed (0x" HWADDR_FMT_plx |
| 417 | ", 0x" HWADDR_FMT_plx "): %s\n", |
| 418 | start_addr, start_addr + size, strerror(errno)); |
| 419 | } |
| 420 | return; |
| 421 | } |
| 422 | |
| 423 | for (i = 0; i < bitmap_size; i++) { |
| 424 | unsigned long map = dirty_bitmap[i]; |
| 425 | while (map != 0) { |
| 426 | j = ctzl(map); |
| 427 | map &= ~(1ul << j); |
| 428 | memory_region_set_dirty(framebuffer, |
| 429 | (i * width + j) * page_size, page_size); |
| 430 | }; |
| 431 | } |
| 432 | } |
| 433 | |
| 434 | static void xen_log_start(MemoryListener *listener, |
| 435 | MemoryRegionSection *section, |
| 436 | int old, int new) |
| 437 | { |
| 438 | XenIOState *state = container_of(listener, XenIOState, memory_listener); |
| 439 | |
| 440 | if (new & ~old & (1 << DIRTY_MEMORY_VGA)) { |
| 441 | xen_sync_dirty_bitmap(state, section->offset_within_address_space, |
| 442 | int128_get64(section->size)); |
| 443 | } |
| 444 | } |
| 445 | |
| 446 | static void xen_log_stop(MemoryListener *listener, MemoryRegionSection *section, |
| 447 | int old, int new) |
| 448 | { |
| 449 | if (old & ~new & (1 << DIRTY_MEMORY_VGA)) { |
| 450 | log_for_dirtybit = NULL; |
| 451 | g_free(dirty_bitmap); |
| 452 | dirty_bitmap = NULL; |
| 453 | /* Disable dirty bit tracking */ |
| 454 | xen_track_dirty_vram(xen_domid, 0, 0, NULL); |
| 455 | } |
| 456 | } |
| 457 | |
| 458 | static void xen_log_sync(MemoryListener *listener, MemoryRegionSection *section) |
| 459 | { |
| 460 | XenIOState *state = container_of(listener, XenIOState, memory_listener); |
| 461 | |
| 462 | xen_sync_dirty_bitmap(state, section->offset_within_address_space, |
| 463 | int128_get64(section->size)); |
| 464 | } |
| 465 | |
| 466 | static bool xen_log_global_start(MemoryListener *listener, Error **errp) |
| 467 | { |
| 468 | if (xen_enabled()) { |
| 469 | xen_in_migration = true; |
| 470 | } |
| 471 | return true; |
| 472 | } |
| 473 | |
| 474 | static void xen_log_global_stop(MemoryListener *listener) |
| 475 | { |
| 476 | xen_in_migration = false; |
| 477 | } |
| 478 | |
| 479 | static const MemoryListener xen_memory_listener = { |
| 480 | .name = "xen-memory", |
| 481 | .region_add = xen_region_add, |
| 482 | .region_del = xen_region_del, |
| 483 | .log_start = xen_log_start, |
| 484 | .log_stop = xen_log_stop, |
| 485 | .log_sync = xen_log_sync, |
| 486 | .log_global_start = xen_log_global_start, |
| 487 | .log_global_stop = xen_log_global_stop, |
| 488 | .priority = MEMORY_LISTENER_PRIORITY_ACCEL, |
| 489 | }; |
| 490 | |
| 491 | static void regs_to_cpu(vmware_regs_t *vmport_regs, ioreq_t *req) |
| 492 | { |
| 493 | X86CPU *cpu; |
| 494 | CPUX86State *env; |
| 495 | |
| 496 | cpu = X86_CPU(current_cpu); |
| 497 | env = &cpu->env; |
| 498 | env->regs[R_EAX] = req->data; |
| 499 | env->regs[R_EBX] = vmport_regs->ebx; |
| 500 | env->regs[R_ECX] = vmport_regs->ecx; |
| 501 | env->regs[R_EDX] = vmport_regs->edx; |
| 502 | env->regs[R_ESI] = vmport_regs->esi; |
| 503 | env->regs[R_EDI] = vmport_regs->edi; |
| 504 | } |
| 505 | |
| 506 | static void regs_from_cpu(vmware_regs_t *vmport_regs) |
| 507 | { |
| 508 | X86CPU *cpu = X86_CPU(current_cpu); |
| 509 | CPUX86State *env = &cpu->env; |
| 510 | |
| 511 | vmport_regs->ebx = env->regs[R_EBX]; |
| 512 | vmport_regs->ecx = env->regs[R_ECX]; |
| 513 | vmport_regs->edx = env->regs[R_EDX]; |
| 514 | vmport_regs->esi = env->regs[R_ESI]; |
| 515 | vmport_regs->edi = env->regs[R_EDI]; |
| 516 | } |
| 517 | |
| 518 | static void handle_vmport_ioreq(XenIOState *state, ioreq_t *req) |
| 519 | { |
| 520 | vmware_regs_t *vmport_regs; |
| 521 | |
| 522 | assert(shared_vmport_page); |
| 523 | vmport_regs = |
| 524 | &shared_vmport_page->vcpu_vmport_regs[state->send_vcpu]; |
| 525 | QEMU_BUILD_BUG_ON(sizeof(*req) < sizeof(*vmport_regs)); |
| 526 | |
| 527 | current_cpu = state->cpu_by_vcpu_id[state->send_vcpu]; |
| 528 | regs_to_cpu(vmport_regs, req); |
| 529 | cpu_ioreq_pio(req); |
| 530 | regs_from_cpu(vmport_regs); |
| 531 | current_cpu = NULL; |
| 532 | } |
| 533 | |
| 534 | #ifdef XEN_COMPAT_PHYSMAP |
| 535 | static void xen_read_physmap(XenIOState *state) |
| 536 | { |
| 537 | XenPhysmap *physmap = NULL; |
| 538 | unsigned int len, num, i; |
| 539 | char path[80], *value = NULL; |
| 540 | char **entries = NULL; |
| 541 | |
| 542 | snprintf(path, sizeof(path), |
| 543 | "/local/domain/0/device-model/%d/physmap", xen_domid); |
| 544 | entries = xs_directory(state->xenstore, 0, path, &num); |
| 545 | if (entries == NULL) |
| 546 | return; |
| 547 | |
| 548 | for (i = 0; i < num; i++) { |
| 549 | physmap = g_new(XenPhysmap, 1); |
| 550 | physmap->phys_offset = strtoull(entries[i], NULL, 16); |
| 551 | snprintf(path, sizeof(path), |
| 552 | "/local/domain/0/device-model/%d/physmap/%s/start_addr", |
| 553 | xen_domid, entries[i]); |
| 554 | value = xs_read(state->xenstore, 0, path, &len); |
| 555 | if (value == NULL) { |
| 556 | g_free(physmap); |
| 557 | continue; |
| 558 | } |
| 559 | physmap->start_addr = strtoull(value, NULL, 16); |
| 560 | free(value); |
| 561 | |
| 562 | snprintf(path, sizeof(path), |
| 563 | "/local/domain/0/device-model/%d/physmap/%s/size", |
| 564 | xen_domid, entries[i]); |
| 565 | value = xs_read(state->xenstore, 0, path, &len); |
| 566 | if (value == NULL) { |
| 567 | g_free(physmap); |
| 568 | continue; |
| 569 | } |
| 570 | physmap->size = strtoull(value, NULL, 16); |
| 571 | free(value); |
| 572 | |
| 573 | snprintf(path, sizeof(path), |
| 574 | "/local/domain/0/device-model/%d/physmap/%s/name", |
| 575 | xen_domid, entries[i]); |
| 576 | physmap->name = xs_read(state->xenstore, 0, path, &len); |
| 577 | |
| 578 | QLIST_INSERT_HEAD(&xen_physmap, physmap, list); |
| 579 | } |
| 580 | free(entries); |
| 581 | } |
| 582 | #else |
| 583 | static void xen_read_physmap(XenIOState *state) |
| 584 | { |
| 585 | } |
| 586 | #endif |
| 587 | |
| 588 | static void xen_wakeup_notifier(Notifier *notifier, void *data) |
| 589 | { |
| 590 | xc_set_hvm_param(xen_xc, xen_domid, HVM_PARAM_ACPI_S_STATE, 0); |
| 591 | } |
| 592 | |
| 593 | static bool xen_check_stubdomain(struct xs_handle *xsh) |
| 594 | { |
| 595 | char *dm_path = g_strdup_printf( |
| 596 | "/local/domain/%d/image/device-model-domid", xen_domid); |
| 597 | char *val; |
| 598 | int32_t dm_domid; |
| 599 | bool is_stubdom = false; |
| 600 | |
| 601 | val = xs_read(xsh, 0, dm_path, NULL); |
| 602 | if (val) { |
| 603 | if (sscanf(val, "%d", &dm_domid) == 1) { |
| 604 | is_stubdom = dm_domid != 0; |
| 605 | } |
| 606 | free(val); |
| 607 | } |
| 608 | |
| 609 | g_free(dm_path); |
| 610 | return is_stubdom; |
| 611 | } |
| 612 | |
| 613 | void xen_hvm_init_pc(PCMachineState *pcms, MemoryRegion **ram_memory) |
| 614 | { |
| 615 | MachineState *ms = MACHINE(pcms); |
| 616 | unsigned int max_cpus = ms->smp.max_cpus; |
| 617 | int rc; |
| 618 | xen_pfn_t ioreq_pfn; |
| 619 | XenIOState *state; |
| 620 | |
| 621 | state = g_new0(XenIOState, 1); |
| 622 | |
| 623 | xen_register_ioreq(state, max_cpus, |
| 624 | HVM_IOREQSRV_BUFIOREQ_ATOMIC, |
| 625 | &xen_memory_listener, true); |
| 626 | |
| 627 | xen_is_stubdomain = xen_check_stubdomain(state->xenstore); |
| 628 | |
| 629 | QLIST_INIT(&xen_physmap); |
| 630 | xen_read_physmap(state); |
| 631 | |
| 632 | suspend.notify = xen_suspend_notifier; |
| 633 | qemu_register_suspend_notifier(&suspend); |
| 634 | |
| 635 | wakeup.notify = xen_wakeup_notifier; |
| 636 | qemu_register_wakeup_notifier(&wakeup); |
| 637 | |
| 638 | rc = xen_get_vmport_regs_pfn(xen_xc, xen_domid, &ioreq_pfn); |
| 639 | if (!rc) { |
| 640 | DPRINTF("shared vmport page at pfn %lx\n", ioreq_pfn); |
| 641 | shared_vmport_page = |
| 642 | xenforeignmemory_map(xen_fmem, xen_domid, PROT_READ|PROT_WRITE, |
| 643 | 1, &ioreq_pfn, NULL); |
| 644 | if (shared_vmport_page == NULL) { |
| 645 | error_report("map shared vmport IO page returned error %d handle=%p", |
| 646 | errno, xen_xc); |
| 647 | goto err; |
| 648 | } |
| 649 | } else if (rc != -ENOSYS) { |
| 650 | error_report("get vmport regs pfn returned error %d, rc=%d", |
| 651 | errno, rc); |
| 652 | goto err; |
| 653 | } |
| 654 | |
| 655 | xen_ram_init(pcms, ms->ram_size, ram_memory); |
| 656 | |
| 657 | /* Disable ACPI build because Xen handles it */ |
| 658 | pcms->acpi_build_enabled = false; |
| 659 | |
| 660 | return; |
| 661 | |
| 662 | err: |
| 663 | error_report("xen hardware virtual machine initialisation failed"); |
| 664 | exit(1); |
| 665 | } |
| 666 | |
| 667 | void xen_register_framebuffer(MemoryRegion *mr) |
| 668 | { |
| 669 | framebuffer = mr; |
| 670 | } |
| 671 | |
| 672 | void xen_hvm_modified_memory(ram_addr_t start, ram_addr_t length) |
| 673 | { |
| 674 | unsigned target_page_bits = qemu_target_page_bits(); |
| 675 | int page_size = qemu_target_page_size(); |
| 676 | int page_mask = -page_size; |
| 677 | |
| 678 | if (unlikely(xen_in_migration)) { |
| 679 | int rc; |
| 680 | ram_addr_t start_pfn, nb_pages; |
| 681 | |
| 682 | start = xen_phys_offset_to_gaddr(start, length, page_mask); |
| 683 | |
| 684 | if (length == 0) { |
| 685 | length = page_size; |
| 686 | } |
| 687 | start_pfn = start >> target_page_bits; |
| 688 | nb_pages = ((start + length + page_size - 1) >> target_page_bits) |
| 689 | - start_pfn; |
| 690 | rc = xen_modified_memory(xen_domid, start_pfn, nb_pages); |
| 691 | if (rc) { |
| 692 | fprintf(stderr, |
| 693 | "%s failed for "RAM_ADDR_FMT" ("RAM_ADDR_FMT"): %i, %s\n", |
| 694 | __func__, start, nb_pages, errno, strerror(errno)); |
| 695 | } |
| 696 | } |
| 697 | } |
| 698 | |
| 699 | void qmp_xen_set_global_dirty_log(bool enable, Error **errp) |
| 700 | { |
| 701 | if (enable) { |
| 702 | memory_global_dirty_log_start(GLOBAL_DIRTY_MIGRATION, errp); |
| 703 | } else { |
| 704 | memory_global_dirty_log_stop(GLOBAL_DIRTY_MIGRATION); |
| 705 | } |
| 706 | } |
| 707 | |
| 708 | void arch_xen_set_memory(XenIOState *state, MemoryRegionSection *section, |
| 709 | bool add) |
| 710 | { |
| 711 | unsigned target_page_bits = qemu_target_page_bits(); |
| 712 | int page_size = qemu_target_page_size(); |
| 713 | int page_mask = -page_size; |
| 714 | hwaddr start_addr = section->offset_within_address_space; |
| 715 | ram_addr_t size = int128_get64(section->size); |
| 716 | bool log_dirty = memory_region_is_logging(section->mr, DIRTY_MEMORY_VGA); |
| 717 | hvmmem_type_t mem_type; |
| 718 | |
| 719 | if (!memory_region_is_ram(section->mr)) { |
| 720 | return; |
| 721 | } |
| 722 | |
| 723 | if (log_dirty != add && |
| 724 | !(section->mr == framebuffer && start_addr > 0xbffff)) { |
| 725 | return; |
| 726 | } |
| 727 | |
| 728 | trace_xen_client_set_memory(start_addr, size, log_dirty); |
| 729 | |
| 730 | start_addr &= page_mask; |
| 731 | size = ROUND_UP(size, page_size); |
| 732 | |
| 733 | if (add) { |
| 734 | if (!memory_region_is_rom(section->mr)) { |
| 735 | xen_add_to_physmap(state, start_addr, size, |
| 736 | section->mr, section->offset_within_region); |
| 737 | } else { |
| 738 | mem_type = HVMMEM_ram_ro; |
| 739 | if (xen_set_mem_type(xen_domid, mem_type, |
| 740 | start_addr >> target_page_bits, |
| 741 | size >> target_page_bits)) { |
| 742 | DPRINTF("xen_set_mem_type error, addr: "HWADDR_FMT_plx"\n", |
| 743 | start_addr); |
| 744 | } |
| 745 | } |
| 746 | } else { |
| 747 | if (xen_remove_from_physmap(state, start_addr, size) < 0) { |
| 748 | DPRINTF("physmapping does not exist at "HWADDR_FMT_plx"\n", start_addr); |
| 749 | } |
| 750 | } |
| 751 | } |
| 752 | |
| 753 | void arch_handle_ioreq(XenIOState *state, ioreq_t *req) |
| 754 | { |
| 755 | switch (req->type) { |
| 756 | case IOREQ_TYPE_VMWARE_PORT: |
| 757 | handle_vmport_ioreq(state, req); |
| 758 | break; |
| 759 | default: |
| 760 | hw_error("Invalid ioreq type 0x%x\n", req->type); |
| 761 | } |
| 762 | } |