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 #include <asm/pvclock-abi.h> 14 15 /* 16 * XEN "SYSTEM CALLS" (a.k.a. HYPERCALLS). 17 */ 18 19 /* 20 * x86_32: EAX = vector; EBX, ECX, EDX, ESI, EDI = args 1, 2, 3, 4, 5. 21 * EAX = return value 22 * (argument registers may be clobbered on return) 23 * x86_64: RAX = vector; RDI, RSI, RDX, R10, R8, R9 = args 1, 2, 3, 4, 5, 6. 24 * RAX = return value 25 * (argument registers not clobbered on return; RCX, R11 are) 26 */ 27 #define __HYPERVISOR_set_trap_table 0 28 #define __HYPERVISOR_mmu_update 1 29 #define __HYPERVISOR_set_gdt 2 30 #define __HYPERVISOR_stack_switch 3 31 #define __HYPERVISOR_set_callbacks 4 32 #define __HYPERVISOR_fpu_taskswitch 5 33 #define __HYPERVISOR_sched_op_compat 6 34 #define __HYPERVISOR_dom0_op 7 35 #define __HYPERVISOR_set_debugreg 8 36 #define __HYPERVISOR_get_debugreg 9 37 #define __HYPERVISOR_update_descriptor 10 38 #define __HYPERVISOR_memory_op 12 39 #define __HYPERVISOR_multicall 13 40 #define __HYPERVISOR_update_va_mapping 14 41 #define __HYPERVISOR_set_timer_op 15 42 #define __HYPERVISOR_event_channel_op_compat 16 43 #define __HYPERVISOR_xen_version 17 44 #define __HYPERVISOR_console_io 18 45 #define __HYPERVISOR_physdev_op_compat 19 46 #define __HYPERVISOR_grant_table_op 20 47 #define __HYPERVISOR_vm_assist 21 48 #define __HYPERVISOR_update_va_mapping_otherdomain 22 49 #define __HYPERVISOR_iret 23 /* x86 only */ 50 #define __HYPERVISOR_vcpu_op 24 51 #define __HYPERVISOR_set_segment_base 25 /* x86/64 only */ 52 #define __HYPERVISOR_mmuext_op 26 53 #define __HYPERVISOR_acm_op 27 54 #define __HYPERVISOR_nmi_op 28 55 #define __HYPERVISOR_sched_op 29 56 #define __HYPERVISOR_callback_op 30 57 #define __HYPERVISOR_xenoprof_op 31 58 #define __HYPERVISOR_event_channel_op 32 59 #define __HYPERVISOR_physdev_op 33 60 #define __HYPERVISOR_hvm_op 34 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 83 /* Architecture-specific VIRQ definitions. */ 84 #define VIRQ_ARCH_0 16 85 #define VIRQ_ARCH_1 17 86 #define VIRQ_ARCH_2 18 87 #define VIRQ_ARCH_3 19 88 #define VIRQ_ARCH_4 20 89 #define VIRQ_ARCH_5 21 90 #define VIRQ_ARCH_6 22 91 #define VIRQ_ARCH_7 23 92 93 #define NR_VIRQS 24 94 /* 95 * MMU-UPDATE REQUESTS 96 * 97 * HYPERVISOR_mmu_update() accepts a list of (ptr, val) pairs. 98 * A foreigndom (FD) can be specified (or DOMID_SELF for none). 99 * Where the FD has some effect, it is described below. 100 * ptr[1:0] specifies the appropriate MMU_* command. 101 * 102 * ptr[1:0] == MMU_NORMAL_PT_UPDATE: 103 * Updates an entry in a page table. If updating an L1 table, and the new 104 * table entry is valid/present, the mapped frame must belong to the FD, if 105 * an FD has been specified. If attempting to map an I/O page then the 106 * caller assumes the privilege of the FD. 107 * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller. 108 * FD == DOMID_XEN: Map restricted areas of Xen's heap space. 109 * ptr[:2] -- Machine address of the page-table entry to modify. 110 * val -- Value to write. 111 * 112 * ptr[1:0] == MMU_MACHPHYS_UPDATE: 113 * Updates an entry in the machine->pseudo-physical mapping table. 114 * ptr[:2] -- Machine address within the frame whose mapping to modify. 115 * The frame must belong to the FD, if one is specified. 116 * val -- Value to write into the mapping entry. 117 * 118 * ptr[1:0] == MMU_PT_UPDATE_PRESERVE_AD: 119 * As MMU_NORMAL_PT_UPDATE above, but A/D bits currently in the PTE are ORed 120 * with those in @val. 121 */ 122 #define MMU_NORMAL_PT_UPDATE 0 /* checked '*ptr = val'. ptr is MA. */ 123 #define MMU_MACHPHYS_UPDATE 1 /* ptr = MA of frame to modify entry for */ 124 #define MMU_PT_UPDATE_PRESERVE_AD 2 /* atomically: *ptr = val | (*ptr&(A|D)) */ 125 126 /* 127 * MMU EXTENDED OPERATIONS 128 * 129 * HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures. 130 * A foreigndom (FD) can be specified (or DOMID_SELF for none). 131 * Where the FD has some effect, it is described below. 132 * 133 * cmd: MMUEXT_(UN)PIN_*_TABLE 134 * mfn: Machine frame number to be (un)pinned as a p.t. page. 135 * The frame must belong to the FD, if one is specified. 136 * 137 * cmd: MMUEXT_NEW_BASEPTR 138 * mfn: Machine frame number of new page-table base to install in MMU. 139 * 140 * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only] 141 * mfn: Machine frame number of new page-table base to install in MMU 142 * when in user space. 143 * 144 * cmd: MMUEXT_TLB_FLUSH_LOCAL 145 * No additional arguments. Flushes local TLB. 146 * 147 * cmd: MMUEXT_INVLPG_LOCAL 148 * linear_addr: Linear address to be flushed from the local TLB. 149 * 150 * cmd: MMUEXT_TLB_FLUSH_MULTI 151 * vcpumask: Pointer to bitmap of VCPUs to be flushed. 152 * 153 * cmd: MMUEXT_INVLPG_MULTI 154 * linear_addr: Linear address to be flushed. 155 * vcpumask: Pointer to bitmap of VCPUs to be flushed. 156 * 157 * cmd: MMUEXT_TLB_FLUSH_ALL 158 * No additional arguments. Flushes all VCPUs' TLBs. 159 * 160 * cmd: MMUEXT_INVLPG_ALL 161 * linear_addr: Linear address to be flushed from all VCPUs' TLBs. 162 * 163 * cmd: MMUEXT_FLUSH_CACHE 164 * No additional arguments. Writes back and flushes cache contents. 165 * 166 * cmd: MMUEXT_SET_LDT 167 * linear_addr: Linear address of LDT base (NB. must be page-aligned). 168 * nr_ents: Number of entries in LDT. 169 */ 170 #define MMUEXT_PIN_L1_TABLE 0 171 #define MMUEXT_PIN_L2_TABLE 1 172 #define MMUEXT_PIN_L3_TABLE 2 173 #define MMUEXT_PIN_L4_TABLE 3 174 #define MMUEXT_UNPIN_TABLE 4 175 #define MMUEXT_NEW_BASEPTR 5 176 #define MMUEXT_TLB_FLUSH_LOCAL 6 177 #define MMUEXT_INVLPG_LOCAL 7 178 #define MMUEXT_TLB_FLUSH_MULTI 8 179 #define MMUEXT_INVLPG_MULTI 9 180 #define MMUEXT_TLB_FLUSH_ALL 10 181 #define MMUEXT_INVLPG_ALL 11 182 #define MMUEXT_FLUSH_CACHE 12 183 #define MMUEXT_SET_LDT 13 184 #define MMUEXT_NEW_USER_BASEPTR 15 185 186 #ifndef __ASSEMBLY__ 187 struct mmuext_op { 188 unsigned int cmd; 189 union { 190 /* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR */ 191 unsigned long mfn; 192 /* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */ 193 unsigned long linear_addr; 194 } arg1; 195 union { 196 /* SET_LDT */ 197 unsigned int nr_ents; 198 /* TLB_FLUSH_MULTI, INVLPG_MULTI */ 199 void *vcpumask; 200 } arg2; 201 }; 202 DEFINE_GUEST_HANDLE_STRUCT(mmuext_op); 203 #endif 204 205 /* These are passed as 'flags' to update_va_mapping. They can be ORed. */ 206 /* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap. */ 207 /* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer. */ 208 #define UVMF_NONE (0UL<<0) /* No flushing at all. */ 209 #define UVMF_TLB_FLUSH (1UL<<0) /* Flush entire TLB(s). */ 210 #define UVMF_INVLPG (2UL<<0) /* Flush only one entry. */ 211 #define UVMF_FLUSHTYPE_MASK (3UL<<0) 212 #define UVMF_MULTI (0UL<<2) /* Flush subset of TLBs. */ 213 #define UVMF_LOCAL (0UL<<2) /* Flush local TLB. */ 214 #define UVMF_ALL (1UL<<2) /* Flush all TLBs. */ 215 216 /* 217 * Commands to HYPERVISOR_console_io(). 218 */ 219 #define CONSOLEIO_write 0 220 #define CONSOLEIO_read 1 221 222 /* 223 * Commands to HYPERVISOR_vm_assist(). 224 */ 225 #define VMASST_CMD_enable 0 226 #define VMASST_CMD_disable 1 227 #define VMASST_TYPE_4gb_segments 0 228 #define VMASST_TYPE_4gb_segments_notify 1 229 #define VMASST_TYPE_writable_pagetables 2 230 #define VMASST_TYPE_pae_extended_cr3 3 231 #define MAX_VMASST_TYPE 3 232 233 #ifndef __ASSEMBLY__ 234 235 typedef uint16_t domid_t; 236 237 /* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */ 238 #define DOMID_FIRST_RESERVED (0x7FF0U) 239 240 /* DOMID_SELF is used in certain contexts to refer to oneself. */ 241 #define DOMID_SELF (0x7FF0U) 242 243 /* 244 * DOMID_IO is used to restrict page-table updates to mapping I/O memory. 245 * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO 246 * is useful to ensure that no mappings to the OS's own heap are accidentally 247 * installed. (e.g., in Linux this could cause havoc as reference counts 248 * aren't adjusted on the I/O-mapping code path). 249 * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can 250 * be specified by any calling domain. 251 */ 252 #define DOMID_IO (0x7FF1U) 253 254 /* 255 * DOMID_XEN is used to allow privileged domains to map restricted parts of 256 * Xen's heap space (e.g., the machine_to_phys table). 257 * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if 258 * the caller is privileged. 259 */ 260 #define DOMID_XEN (0x7FF2U) 261 262 /* 263 * Send an array of these to HYPERVISOR_mmu_update(). 264 * NB. The fields are natural pointer/address size for this architecture. 265 */ 266 struct mmu_update { 267 uint64_t ptr; /* Machine address of PTE. */ 268 uint64_t val; /* New contents of PTE. */ 269 }; 270 DEFINE_GUEST_HANDLE_STRUCT(mmu_update); 271 272 /* 273 * Send an array of these to HYPERVISOR_multicall(). 274 * NB. The fields are natural register size for this architecture. 275 */ 276 struct multicall_entry { 277 unsigned long op; 278 long result; 279 unsigned long args[6]; 280 }; 281 DEFINE_GUEST_HANDLE_STRUCT(multicall_entry); 282 283 /* 284 * Event channel endpoints per domain: 285 * 1024 if a long is 32 bits; 4096 if a long is 64 bits. 286 */ 287 #define NR_EVENT_CHANNELS (sizeof(unsigned long) * sizeof(unsigned long) * 64) 288 289 struct vcpu_time_info { 290 /* 291 * Updates to the following values are preceded and followed 292 * by an increment of 'version'. The guest can therefore 293 * detect updates by looking for changes to 'version'. If the 294 * least-significant bit of the version number is set then an 295 * update is in progress and the guest must wait to read a 296 * consistent set of values. The correct way to interact with 297 * the version number is similar to Linux's seqlock: see the 298 * implementations of read_seqbegin/read_seqretry. 299 */ 300 uint32_t version; 301 uint32_t pad0; 302 uint64_t tsc_timestamp; /* TSC at last update of time vals. */ 303 uint64_t system_time; /* Time, in nanosecs, since boot. */ 304 /* 305 * Current system time: 306 * system_time + ((tsc - tsc_timestamp) << tsc_shift) * tsc_to_system_mul 307 * CPU frequency (Hz): 308 * ((10^9 << 32) / tsc_to_system_mul) >> tsc_shift 309 */ 310 uint32_t tsc_to_system_mul; 311 int8_t tsc_shift; 312 int8_t pad1[3]; 313 }; /* 32 bytes */ 314 315 struct vcpu_info { 316 /* 317 * 'evtchn_upcall_pending' is written non-zero by Xen to indicate 318 * a pending notification for a particular VCPU. It is then cleared 319 * by the guest OS /before/ checking for pending work, thus avoiding 320 * a set-and-check race. Note that the mask is only accessed by Xen 321 * on the CPU that is currently hosting the VCPU. This means that the 322 * pending and mask flags can be updated by the guest without special 323 * synchronisation (i.e., no need for the x86 LOCK prefix). 324 * This may seem suboptimal because if the pending flag is set by 325 * a different CPU then an IPI may be scheduled even when the mask 326 * is set. However, note: 327 * 1. The task of 'interrupt holdoff' is covered by the per-event- 328 * channel mask bits. A 'noisy' event that is continually being 329 * triggered can be masked at source at this very precise 330 * granularity. 331 * 2. The main purpose of the per-VCPU mask is therefore to restrict 332 * reentrant execution: whether for concurrency control, or to 333 * prevent unbounded stack usage. Whatever the purpose, we expect 334 * that the mask will be asserted only for short periods at a time, 335 * and so the likelihood of a 'spurious' IPI is suitably small. 336 * The mask is read before making an event upcall to the guest: a 337 * non-zero mask therefore guarantees that the VCPU will not receive 338 * an upcall activation. The mask is cleared when the VCPU requests 339 * to block: this avoids wakeup-waiting races. 340 */ 341 uint8_t evtchn_upcall_pending; 342 uint8_t evtchn_upcall_mask; 343 unsigned long evtchn_pending_sel; 344 struct arch_vcpu_info arch; 345 struct pvclock_vcpu_time_info time; 346 }; /* 64 bytes (x86) */ 347 348 /* 349 * Xen/kernel shared data -- pointer provided in start_info. 350 * NB. We expect that this struct is smaller than a page. 351 */ 352 struct shared_info { 353 struct vcpu_info vcpu_info[MAX_VIRT_CPUS]; 354 355 /* 356 * A domain can create "event channels" on which it can send and receive 357 * asynchronous event notifications. There are three classes of event that 358 * are delivered by this mechanism: 359 * 1. Bi-directional inter- and intra-domain connections. Domains must 360 * arrange out-of-band to set up a connection (usually by allocating 361 * an unbound 'listener' port and avertising that via a storage service 362 * such as xenstore). 363 * 2. Physical interrupts. A domain with suitable hardware-access 364 * privileges can bind an event-channel port to a physical interrupt 365 * source. 366 * 3. Virtual interrupts ('events'). A domain can bind an event-channel 367 * port to a virtual interrupt source, such as the virtual-timer 368 * device or the emergency console. 369 * 370 * Event channels are addressed by a "port index". Each channel is 371 * associated with two bits of information: 372 * 1. PENDING -- notifies the domain that there is a pending notification 373 * to be processed. This bit is cleared by the guest. 374 * 2. MASK -- if this bit is clear then a 0->1 transition of PENDING 375 * will cause an asynchronous upcall to be scheduled. This bit is only 376 * updated by the guest. It is read-only within Xen. If a channel 377 * becomes pending while the channel is masked then the 'edge' is lost 378 * (i.e., when the channel is unmasked, the guest must manually handle 379 * pending notifications as no upcall will be scheduled by Xen). 380 * 381 * To expedite scanning of pending notifications, any 0->1 pending 382 * transition on an unmasked channel causes a corresponding bit in a 383 * per-vcpu selector word to be set. Each bit in the selector covers a 384 * 'C long' in the PENDING bitfield array. 385 */ 386 unsigned long evtchn_pending[sizeof(unsigned long) * 8]; 387 unsigned long evtchn_mask[sizeof(unsigned long) * 8]; 388 389 /* 390 * Wallclock time: updated only by control software. Guests should base 391 * their gettimeofday() syscall on this wallclock-base value. 392 */ 393 struct pvclock_wall_clock wc; 394 395 struct arch_shared_info arch; 396 397 }; 398 399 /* 400 * Start-of-day memory layout for the initial domain (DOM0): 401 * 1. The domain is started within contiguous virtual-memory region. 402 * 2. The contiguous region begins and ends on an aligned 4MB boundary. 403 * 3. The region start corresponds to the load address of the OS image. 404 * If the load address is not 4MB aligned then the address is rounded down. 405 * 4. This the order of bootstrap elements in the initial virtual region: 406 * a. relocated kernel image 407 * b. initial ram disk [mod_start, mod_len] 408 * c. list of allocated page frames [mfn_list, nr_pages] 409 * d. start_info_t structure [register ESI (x86)] 410 * e. bootstrap page tables [pt_base, CR3 (x86)] 411 * f. bootstrap stack [register ESP (x86)] 412 * 5. Bootstrap elements are packed together, but each is 4kB-aligned. 413 * 6. The initial ram disk may be omitted. 414 * 7. The list of page frames forms a contiguous 'pseudo-physical' memory 415 * layout for the domain. In particular, the bootstrap virtual-memory 416 * region is a 1:1 mapping to the first section of the pseudo-physical map. 417 * 8. All bootstrap elements are mapped read-writable for the guest OS. The 418 * only exception is the bootstrap page table, which is mapped read-only. 419 * 9. There is guaranteed to be at least 512kB padding after the final 420 * bootstrap element. If necessary, the bootstrap virtual region is 421 * extended by an extra 4MB to ensure this. 422 */ 423 424 #define MAX_GUEST_CMDLINE 1024 425 struct start_info { 426 /* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME. */ 427 char magic[32]; /* "xen-<version>-<platform>". */ 428 unsigned long nr_pages; /* Total pages allocated to this domain. */ 429 unsigned long shared_info; /* MACHINE address of shared info struct. */ 430 uint32_t flags; /* SIF_xxx flags. */ 431 unsigned long store_mfn; /* MACHINE page number of shared page. */ 432 uint32_t store_evtchn; /* Event channel for store communication. */ 433 union { 434 struct { 435 unsigned long mfn; /* MACHINE page number of console page. */ 436 uint32_t evtchn; /* Event channel for console page. */ 437 } domU; 438 struct { 439 uint32_t info_off; /* Offset of console_info struct. */ 440 uint32_t info_size; /* Size of console_info struct from start.*/ 441 } dom0; 442 } console; 443 /* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME). */ 444 unsigned long pt_base; /* VIRTUAL address of page directory. */ 445 unsigned long nr_pt_frames; /* Number of bootstrap p.t. frames. */ 446 unsigned long mfn_list; /* VIRTUAL address of page-frame list. */ 447 unsigned long mod_start; /* VIRTUAL address of pre-loaded module. */ 448 unsigned long mod_len; /* Size (bytes) of pre-loaded module. */ 449 int8_t cmd_line[MAX_GUEST_CMDLINE]; 450 }; 451 452 /* These flags are passed in the 'flags' field of start_info_t. */ 453 #define SIF_PRIVILEGED (1<<0) /* Is the domain privileged? */ 454 #define SIF_INITDOMAIN (1<<1) /* Is this the initial control domain? */ 455 456 typedef uint64_t cpumap_t; 457 458 typedef uint8_t xen_domain_handle_t[16]; 459 460 /* Turn a plain number into a C unsigned long constant. */ 461 #define __mk_unsigned_long(x) x ## UL 462 #define mk_unsigned_long(x) __mk_unsigned_long(x) 463 464 #else /* __ASSEMBLY__ */ 465 466 /* In assembly code we cannot use C numeric constant suffixes. */ 467 #define mk_unsigned_long(x) x 468 469 #endif /* !__ASSEMBLY__ */ 470 471 #endif /* __XEN_PUBLIC_XEN_H__ */ 472