1 /****************************************************************************** 2 * xen.h 3 * 4 * Guest OS interface to Xen. 5 * 6 * Copyright (c) 2004, K A Fraser 7 */ 8 9 #ifndef __XEN_PUBLIC_XEN_H__ 10 #define __XEN_PUBLIC_XEN_H__ 11 12 #include <asm/xen/interface.h> 13 14 /* 15 * XEN "SYSTEM CALLS" (a.k.a. HYPERCALLS). 16 */ 17 18 /* 19 * x86_32: EAX = vector; EBX, ECX, EDX, ESI, EDI = args 1, 2, 3, 4, 5. 20 * EAX = return value 21 * (argument registers may be clobbered on return) 22 * x86_64: RAX = vector; RDI, RSI, RDX, R10, R8, R9 = args 1, 2, 3, 4, 5, 6. 23 * RAX = return value 24 * (argument registers not clobbered on return; RCX, R11 are) 25 */ 26 #define __HYPERVISOR_set_trap_table 0 27 #define __HYPERVISOR_mmu_update 1 28 #define __HYPERVISOR_set_gdt 2 29 #define __HYPERVISOR_stack_switch 3 30 #define __HYPERVISOR_set_callbacks 4 31 #define __HYPERVISOR_fpu_taskswitch 5 32 #define __HYPERVISOR_sched_op_compat 6 33 #define __HYPERVISOR_dom0_op 7 34 #define __HYPERVISOR_set_debugreg 8 35 #define __HYPERVISOR_get_debugreg 9 36 #define __HYPERVISOR_update_descriptor 10 37 #define __HYPERVISOR_memory_op 12 38 #define __HYPERVISOR_multicall 13 39 #define __HYPERVISOR_update_va_mapping 14 40 #define __HYPERVISOR_set_timer_op 15 41 #define __HYPERVISOR_event_channel_op_compat 16 42 #define __HYPERVISOR_xen_version 17 43 #define __HYPERVISOR_console_io 18 44 #define __HYPERVISOR_physdev_op_compat 19 45 #define __HYPERVISOR_grant_table_op 20 46 #define __HYPERVISOR_vm_assist 21 47 #define __HYPERVISOR_update_va_mapping_otherdomain 22 48 #define __HYPERVISOR_iret 23 /* x86 only */ 49 #define __HYPERVISOR_vcpu_op 24 50 #define __HYPERVISOR_set_segment_base 25 /* x86/64 only */ 51 #define __HYPERVISOR_mmuext_op 26 52 #define __HYPERVISOR_acm_op 27 53 #define __HYPERVISOR_nmi_op 28 54 #define __HYPERVISOR_sched_op 29 55 #define __HYPERVISOR_callback_op 30 56 #define __HYPERVISOR_xenoprof_op 31 57 #define __HYPERVISOR_event_channel_op 32 58 #define __HYPERVISOR_physdev_op 33 59 #define __HYPERVISOR_hvm_op 34 60 #define __HYPERVISOR_tmem_op 38 61 62 /* Architecture-specific hypercall definitions. */ 63 #define __HYPERVISOR_arch_0 48 64 #define __HYPERVISOR_arch_1 49 65 #define __HYPERVISOR_arch_2 50 66 #define __HYPERVISOR_arch_3 51 67 #define __HYPERVISOR_arch_4 52 68 #define __HYPERVISOR_arch_5 53 69 #define __HYPERVISOR_arch_6 54 70 #define __HYPERVISOR_arch_7 55 71 72 /* 73 * VIRTUAL INTERRUPTS 74 * 75 * Virtual interrupts that a guest OS may receive from Xen. 76 */ 77 #define VIRQ_TIMER 0 /* Timebase update, and/or requested timeout. */ 78 #define VIRQ_DEBUG 1 /* Request guest to dump debug info. */ 79 #define VIRQ_CONSOLE 2 /* (DOM0) Bytes received on emergency console. */ 80 #define VIRQ_DOM_EXC 3 /* (DOM0) Exceptional event for some domain. */ 81 #define VIRQ_DEBUGGER 6 /* (DOM0) A domain has paused for debugging. */ 82 #define VIRQ_PCPU_STATE 9 /* (DOM0) PCPU state changed */ 83 84 /* Architecture-specific VIRQ definitions. */ 85 #define VIRQ_ARCH_0 16 86 #define VIRQ_ARCH_1 17 87 #define VIRQ_ARCH_2 18 88 #define VIRQ_ARCH_3 19 89 #define VIRQ_ARCH_4 20 90 #define VIRQ_ARCH_5 21 91 #define VIRQ_ARCH_6 22 92 #define VIRQ_ARCH_7 23 93 94 #define NR_VIRQS 24 95 /* 96 * MMU-UPDATE REQUESTS 97 * 98 * HYPERVISOR_mmu_update() accepts a list of (ptr, val) pairs. 99 * A foreigndom (FD) can be specified (or DOMID_SELF for none). 100 * Where the FD has some effect, it is described below. 101 * ptr[1:0] specifies the appropriate MMU_* command. 102 * 103 * ptr[1:0] == MMU_NORMAL_PT_UPDATE: 104 * Updates an entry in a page table. If updating an L1 table, and the new 105 * table entry is valid/present, the mapped frame must belong to the FD, if 106 * an FD has been specified. If attempting to map an I/O page then the 107 * caller assumes the privilege of the FD. 108 * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller. 109 * FD == DOMID_XEN: Map restricted areas of Xen's heap space. 110 * ptr[:2] -- Machine address of the page-table entry to modify. 111 * val -- Value to write. 112 * 113 * ptr[1:0] == MMU_MACHPHYS_UPDATE: 114 * Updates an entry in the machine->pseudo-physical mapping table. 115 * ptr[:2] -- Machine address within the frame whose mapping to modify. 116 * The frame must belong to the FD, if one is specified. 117 * val -- Value to write into the mapping entry. 118 * 119 * ptr[1:0] == MMU_PT_UPDATE_PRESERVE_AD: 120 * As MMU_NORMAL_PT_UPDATE above, but A/D bits currently in the PTE are ORed 121 * with those in @val. 122 */ 123 #define MMU_NORMAL_PT_UPDATE 0 /* checked '*ptr = val'. ptr is MA. */ 124 #define MMU_MACHPHYS_UPDATE 1 /* ptr = MA of frame to modify entry for */ 125 #define MMU_PT_UPDATE_PRESERVE_AD 2 /* atomically: *ptr = val | (*ptr&(A|D)) */ 126 127 /* 128 * MMU EXTENDED OPERATIONS 129 * 130 * HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures. 131 * A foreigndom (FD) can be specified (or DOMID_SELF for none). 132 * Where the FD has some effect, it is described below. 133 * 134 * cmd: MMUEXT_(UN)PIN_*_TABLE 135 * mfn: Machine frame number to be (un)pinned as a p.t. page. 136 * The frame must belong to the FD, if one is specified. 137 * 138 * cmd: MMUEXT_NEW_BASEPTR 139 * mfn: Machine frame number of new page-table base to install in MMU. 140 * 141 * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only] 142 * mfn: Machine frame number of new page-table base to install in MMU 143 * when in user space. 144 * 145 * cmd: MMUEXT_TLB_FLUSH_LOCAL 146 * No additional arguments. Flushes local TLB. 147 * 148 * cmd: MMUEXT_INVLPG_LOCAL 149 * linear_addr: Linear address to be flushed from the local TLB. 150 * 151 * cmd: MMUEXT_TLB_FLUSH_MULTI 152 * vcpumask: Pointer to bitmap of VCPUs to be flushed. 153 * 154 * cmd: MMUEXT_INVLPG_MULTI 155 * linear_addr: Linear address to be flushed. 156 * vcpumask: Pointer to bitmap of VCPUs to be flushed. 157 * 158 * cmd: MMUEXT_TLB_FLUSH_ALL 159 * No additional arguments. Flushes all VCPUs' TLBs. 160 * 161 * cmd: MMUEXT_INVLPG_ALL 162 * linear_addr: Linear address to be flushed from all VCPUs' TLBs. 163 * 164 * cmd: MMUEXT_FLUSH_CACHE 165 * No additional arguments. Writes back and flushes cache contents. 166 * 167 * cmd: MMUEXT_SET_LDT 168 * linear_addr: Linear address of LDT base (NB. must be page-aligned). 169 * nr_ents: Number of entries in LDT. 170 */ 171 #define MMUEXT_PIN_L1_TABLE 0 172 #define MMUEXT_PIN_L2_TABLE 1 173 #define MMUEXT_PIN_L3_TABLE 2 174 #define MMUEXT_PIN_L4_TABLE 3 175 #define MMUEXT_UNPIN_TABLE 4 176 #define MMUEXT_NEW_BASEPTR 5 177 #define MMUEXT_TLB_FLUSH_LOCAL 6 178 #define MMUEXT_INVLPG_LOCAL 7 179 #define MMUEXT_TLB_FLUSH_MULTI 8 180 #define MMUEXT_INVLPG_MULTI 9 181 #define MMUEXT_TLB_FLUSH_ALL 10 182 #define MMUEXT_INVLPG_ALL 11 183 #define MMUEXT_FLUSH_CACHE 12 184 #define MMUEXT_SET_LDT 13 185 #define MMUEXT_NEW_USER_BASEPTR 15 186 187 #ifndef __ASSEMBLY__ 188 struct mmuext_op { 189 unsigned int cmd; 190 union { 191 /* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR */ 192 xen_pfn_t mfn; 193 /* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */ 194 unsigned long linear_addr; 195 } arg1; 196 union { 197 /* SET_LDT */ 198 unsigned int nr_ents; 199 /* TLB_FLUSH_MULTI, INVLPG_MULTI */ 200 void *vcpumask; 201 } arg2; 202 }; 203 DEFINE_GUEST_HANDLE_STRUCT(mmuext_op); 204 #endif 205 206 /* These are passed as 'flags' to update_va_mapping. They can be ORed. */ 207 /* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap. */ 208 /* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer. */ 209 #define UVMF_NONE (0UL<<0) /* No flushing at all. */ 210 #define UVMF_TLB_FLUSH (1UL<<0) /* Flush entire TLB(s). */ 211 #define UVMF_INVLPG (2UL<<0) /* Flush only one entry. */ 212 #define UVMF_FLUSHTYPE_MASK (3UL<<0) 213 #define UVMF_MULTI (0UL<<2) /* Flush subset of TLBs. */ 214 #define UVMF_LOCAL (0UL<<2) /* Flush local TLB. */ 215 #define UVMF_ALL (1UL<<2) /* Flush all TLBs. */ 216 217 /* 218 * Commands to HYPERVISOR_console_io(). 219 */ 220 #define CONSOLEIO_write 0 221 #define CONSOLEIO_read 1 222 223 /* 224 * Commands to HYPERVISOR_vm_assist(). 225 */ 226 #define VMASST_CMD_enable 0 227 #define VMASST_CMD_disable 1 228 #define VMASST_TYPE_4gb_segments 0 229 #define VMASST_TYPE_4gb_segments_notify 1 230 #define VMASST_TYPE_writable_pagetables 2 231 #define VMASST_TYPE_pae_extended_cr3 3 232 #define MAX_VMASST_TYPE 3 233 234 #ifndef __ASSEMBLY__ 235 236 typedef uint16_t domid_t; 237 238 /* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */ 239 #define DOMID_FIRST_RESERVED (0x7FF0U) 240 241 /* DOMID_SELF is used in certain contexts to refer to oneself. */ 242 #define DOMID_SELF (0x7FF0U) 243 244 /* 245 * DOMID_IO is used to restrict page-table updates to mapping I/O memory. 246 * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO 247 * is useful to ensure that no mappings to the OS's own heap are accidentally 248 * installed. (e.g., in Linux this could cause havoc as reference counts 249 * aren't adjusted on the I/O-mapping code path). 250 * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can 251 * be specified by any calling domain. 252 */ 253 #define DOMID_IO (0x7FF1U) 254 255 /* 256 * DOMID_XEN is used to allow privileged domains to map restricted parts of 257 * Xen's heap space (e.g., the machine_to_phys table). 258 * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if 259 * the caller is privileged. 260 */ 261 #define DOMID_XEN (0x7FF2U) 262 263 /* 264 * Send an array of these to HYPERVISOR_mmu_update(). 265 * NB. The fields are natural pointer/address size for this architecture. 266 */ 267 struct mmu_update { 268 uint64_t ptr; /* Machine address of PTE. */ 269 uint64_t val; /* New contents of PTE. */ 270 }; 271 DEFINE_GUEST_HANDLE_STRUCT(mmu_update); 272 273 /* 274 * Send an array of these to HYPERVISOR_multicall(). 275 * NB. The fields are natural register size for this architecture. 276 */ 277 struct multicall_entry { 278 xen_ulong_t op; 279 xen_long_t result; 280 xen_ulong_t args[6]; 281 }; 282 DEFINE_GUEST_HANDLE_STRUCT(multicall_entry); 283 284 struct vcpu_time_info { 285 /* 286 * Updates to the following values are preceded and followed 287 * by an increment of 'version'. The guest can therefore 288 * detect updates by looking for changes to 'version'. If the 289 * least-significant bit of the version number is set then an 290 * update is in progress and the guest must wait to read a 291 * consistent set of values. The correct way to interact with 292 * the version number is similar to Linux's seqlock: see the 293 * implementations of read_seqbegin/read_seqretry. 294 */ 295 uint32_t version; 296 uint32_t pad0; 297 uint64_t tsc_timestamp; /* TSC at last update of time vals. */ 298 uint64_t system_time; /* Time, in nanosecs, since boot. */ 299 /* 300 * Current system time: 301 * system_time + ((tsc - tsc_timestamp) << tsc_shift) * tsc_to_system_mul 302 * CPU frequency (Hz): 303 * ((10^9 << 32) / tsc_to_system_mul) >> tsc_shift 304 */ 305 uint32_t tsc_to_system_mul; 306 int8_t tsc_shift; 307 int8_t pad1[3]; 308 }; /* 32 bytes */ 309 310 struct vcpu_info { 311 /* 312 * 'evtchn_upcall_pending' is written non-zero by Xen to indicate 313 * a pending notification for a particular VCPU. It is then cleared 314 * by the guest OS /before/ checking for pending work, thus avoiding 315 * a set-and-check race. Note that the mask is only accessed by Xen 316 * on the CPU that is currently hosting the VCPU. This means that the 317 * pending and mask flags can be updated by the guest without special 318 * synchronisation (i.e., no need for the x86 LOCK prefix). 319 * This may seem suboptimal because if the pending flag is set by 320 * a different CPU then an IPI may be scheduled even when the mask 321 * is set. However, note: 322 * 1. The task of 'interrupt holdoff' is covered by the per-event- 323 * channel mask bits. A 'noisy' event that is continually being 324 * triggered can be masked at source at this very precise 325 * granularity. 326 * 2. The main purpose of the per-VCPU mask is therefore to restrict 327 * reentrant execution: whether for concurrency control, or to 328 * prevent unbounded stack usage. Whatever the purpose, we expect 329 * that the mask will be asserted only for short periods at a time, 330 * and so the likelihood of a 'spurious' IPI is suitably small. 331 * The mask is read before making an event upcall to the guest: a 332 * non-zero mask therefore guarantees that the VCPU will not receive 333 * an upcall activation. The mask is cleared when the VCPU requests 334 * to block: this avoids wakeup-waiting races. 335 */ 336 uint8_t evtchn_upcall_pending; 337 uint8_t evtchn_upcall_mask; 338 xen_ulong_t evtchn_pending_sel; 339 struct arch_vcpu_info arch; 340 struct pvclock_vcpu_time_info time; 341 }; /* 64 bytes (x86) */ 342 343 /* 344 * Xen/kernel shared data -- pointer provided in start_info. 345 * NB. We expect that this struct is smaller than a page. 346 */ 347 struct shared_info { 348 struct vcpu_info vcpu_info[MAX_VIRT_CPUS]; 349 350 /* 351 * A domain can create "event channels" on which it can send and receive 352 * asynchronous event notifications. There are three classes of event that 353 * are delivered by this mechanism: 354 * 1. Bi-directional inter- and intra-domain connections. Domains must 355 * arrange out-of-band to set up a connection (usually by allocating 356 * an unbound 'listener' port and avertising that via a storage service 357 * such as xenstore). 358 * 2. Physical interrupts. A domain with suitable hardware-access 359 * privileges can bind an event-channel port to a physical interrupt 360 * source. 361 * 3. Virtual interrupts ('events'). A domain can bind an event-channel 362 * port to a virtual interrupt source, such as the virtual-timer 363 * device or the emergency console. 364 * 365 * Event channels are addressed by a "port index". Each channel is 366 * associated with two bits of information: 367 * 1. PENDING -- notifies the domain that there is a pending notification 368 * to be processed. This bit is cleared by the guest. 369 * 2. MASK -- if this bit is clear then a 0->1 transition of PENDING 370 * will cause an asynchronous upcall to be scheduled. This bit is only 371 * updated by the guest. It is read-only within Xen. If a channel 372 * becomes pending while the channel is masked then the 'edge' is lost 373 * (i.e., when the channel is unmasked, the guest must manually handle 374 * pending notifications as no upcall will be scheduled by Xen). 375 * 376 * To expedite scanning of pending notifications, any 0->1 pending 377 * transition on an unmasked channel causes a corresponding bit in a 378 * per-vcpu selector word to be set. Each bit in the selector covers a 379 * 'C long' in the PENDING bitfield array. 380 */ 381 xen_ulong_t evtchn_pending[sizeof(xen_ulong_t) * 8]; 382 xen_ulong_t evtchn_mask[sizeof(xen_ulong_t) * 8]; 383 384 /* 385 * Wallclock time: updated only by control software. Guests should base 386 * their gettimeofday() syscall on this wallclock-base value. 387 */ 388 struct pvclock_wall_clock wc; 389 390 struct arch_shared_info arch; 391 392 }; 393 394 /* 395 * Start-of-day memory layout for the initial domain (DOM0): 396 * 1. The domain is started within contiguous virtual-memory region. 397 * 2. The contiguous region begins and ends on an aligned 4MB boundary. 398 * 3. The region start corresponds to the load address of the OS image. 399 * If the load address is not 4MB aligned then the address is rounded down. 400 * 4. This the order of bootstrap elements in the initial virtual region: 401 * a. relocated kernel image 402 * b. initial ram disk [mod_start, mod_len] 403 * c. list of allocated page frames [mfn_list, nr_pages] 404 * d. start_info_t structure [register ESI (x86)] 405 * e. bootstrap page tables [pt_base, CR3 (x86)] 406 * f. bootstrap stack [register ESP (x86)] 407 * 5. Bootstrap elements are packed together, but each is 4kB-aligned. 408 * 6. The initial ram disk may be omitted. 409 * 7. The list of page frames forms a contiguous 'pseudo-physical' memory 410 * layout for the domain. In particular, the bootstrap virtual-memory 411 * region is a 1:1 mapping to the first section of the pseudo-physical map. 412 * 8. All bootstrap elements are mapped read-writable for the guest OS. The 413 * only exception is the bootstrap page table, which is mapped read-only. 414 * 9. There is guaranteed to be at least 512kB padding after the final 415 * bootstrap element. If necessary, the bootstrap virtual region is 416 * extended by an extra 4MB to ensure this. 417 */ 418 419 #define MAX_GUEST_CMDLINE 1024 420 struct start_info { 421 /* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME. */ 422 char magic[32]; /* "xen-<version>-<platform>". */ 423 unsigned long nr_pages; /* Total pages allocated to this domain. */ 424 unsigned long shared_info; /* MACHINE address of shared info struct. */ 425 uint32_t flags; /* SIF_xxx flags. */ 426 xen_pfn_t store_mfn; /* MACHINE page number of shared page. */ 427 uint32_t store_evtchn; /* Event channel for store communication. */ 428 union { 429 struct { 430 xen_pfn_t mfn; /* MACHINE page number of console page. */ 431 uint32_t evtchn; /* Event channel for console page. */ 432 } domU; 433 struct { 434 uint32_t info_off; /* Offset of console_info struct. */ 435 uint32_t info_size; /* Size of console_info struct from start.*/ 436 } dom0; 437 } console; 438 /* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME). */ 439 unsigned long pt_base; /* VIRTUAL address of page directory. */ 440 unsigned long nr_pt_frames; /* Number of bootstrap p.t. frames. */ 441 unsigned long mfn_list; /* VIRTUAL address of page-frame list. */ 442 unsigned long mod_start; /* VIRTUAL address of pre-loaded module. */ 443 unsigned long mod_len; /* Size (bytes) of pre-loaded module. */ 444 int8_t cmd_line[MAX_GUEST_CMDLINE]; 445 }; 446 447 struct dom0_vga_console_info { 448 uint8_t video_type; 449 #define XEN_VGATYPE_TEXT_MODE_3 0x03 450 #define XEN_VGATYPE_VESA_LFB 0x23 451 #define XEN_VGATYPE_EFI_LFB 0x70 452 453 union { 454 struct { 455 /* Font height, in pixels. */ 456 uint16_t font_height; 457 /* Cursor location (column, row). */ 458 uint16_t cursor_x, cursor_y; 459 /* Number of rows and columns (dimensions in characters). */ 460 uint16_t rows, columns; 461 } text_mode_3; 462 463 struct { 464 /* Width and height, in pixels. */ 465 uint16_t width, height; 466 /* Bytes per scan line. */ 467 uint16_t bytes_per_line; 468 /* Bits per pixel. */ 469 uint16_t bits_per_pixel; 470 /* LFB physical address, and size (in units of 64kB). */ 471 uint32_t lfb_base; 472 uint32_t lfb_size; 473 /* RGB mask offsets and sizes, as defined by VBE 1.2+ */ 474 uint8_t red_pos, red_size; 475 uint8_t green_pos, green_size; 476 uint8_t blue_pos, blue_size; 477 uint8_t rsvd_pos, rsvd_size; 478 479 /* VESA capabilities (offset 0xa, VESA command 0x4f00). */ 480 uint32_t gbl_caps; 481 /* Mode attributes (offset 0x0, VESA command 0x4f01). */ 482 uint16_t mode_attrs; 483 } vesa_lfb; 484 } u; 485 }; 486 487 /* These flags are passed in the 'flags' field of start_info_t. */ 488 #define SIF_PRIVILEGED (1<<0) /* Is the domain privileged? */ 489 #define SIF_INITDOMAIN (1<<1) /* Is this the initial control domain? */ 490 #define SIF_PM_MASK (0xFF<<8) /* reserve 1 byte for xen-pm options */ 491 492 typedef uint64_t cpumap_t; 493 494 typedef uint8_t xen_domain_handle_t[16]; 495 496 /* Turn a plain number into a C unsigned long constant. */ 497 #define __mk_unsigned_long(x) x ## UL 498 #define mk_unsigned_long(x) __mk_unsigned_long(x) 499 500 #define TMEM_SPEC_VERSION 1 501 502 struct tmem_op { 503 uint32_t cmd; 504 int32_t pool_id; 505 union { 506 struct { /* for cmd == TMEM_NEW_POOL */ 507 uint64_t uuid[2]; 508 uint32_t flags; 509 } new; 510 struct { 511 uint64_t oid[3]; 512 uint32_t index; 513 uint32_t tmem_offset; 514 uint32_t pfn_offset; 515 uint32_t len; 516 GUEST_HANDLE(void) gmfn; /* guest machine page frame */ 517 } gen; 518 } u; 519 }; 520 521 DEFINE_GUEST_HANDLE(u64); 522 523 #else /* __ASSEMBLY__ */ 524 525 /* In assembly code we cannot use C numeric constant suffixes. */ 526 #define mk_unsigned_long(x) x 527 528 #endif /* !__ASSEMBLY__ */ 529 530 #endif /* __XEN_PUBLIC_XEN_H__ */ 531