include/xen/interface/xen.h

   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              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_new         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
  61 /* Architecture-specific hypercall definitions. */
  62 #define __HYPERVISOR_arch_0               48
  63 #define __HYPERVISOR_arch_1               49
  64 #define __HYPERVISOR_arch_2               50
  65 #define __HYPERVISOR_arch_3               51
  66 #define __HYPERVISOR_arch_4               52
  67 #define __HYPERVISOR_arch_5               53
  68 #define __HYPERVISOR_arch_6               54
  69 #define __HYPERVISOR_arch_7               55
  70
  71 /*
  72  * VIRTUAL INTERRUPTS
  73  *
  74  * Virtual interrupts that a guest OS may receive from Xen.
  75  */
  76 #define VIRQ_TIMER      0  /* Timebase update, and/or requested timeout.  */
  77 #define VIRQ_DEBUG      1  /* Request guest to dump debug info.           */
  78 #define VIRQ_CONSOLE    2  /* (DOM0) Bytes received on emergency console. */
  79 #define VIRQ_DOM_EXC    3  /* (DOM0) Exceptional event for some domain.   */
  80 #define VIRQ_DEBUGGER   6  /* (DOM0) A domain has paused for debugging.   */
  81
  82 /* Architecture-specific VIRQ definitions. */
  83 #define VIRQ_ARCH_0    16
  84 #define VIRQ_ARCH_1    17
  85 #define VIRQ_ARCH_2    18
  86 #define VIRQ_ARCH_3    19
  87 #define VIRQ_ARCH_4    20
  88 #define VIRQ_ARCH_5    21
  89 #define VIRQ_ARCH_6    22
  90 #define VIRQ_ARCH_7    23
  91
  92 #define NR_VIRQS       24
  93 /*
  94  * MMU-UPDATE REQUESTS
  95  *
  96  * HYPERVISOR_mmu_update() accepts a list of (ptr, val) pairs.
  97  * A foreigndom (FD) can be specified (or DOMID_SELF for none).
  98  * Where the FD has some effect, it is described below.
  99  * ptr[1:0] specifies the appropriate MMU_* command.
 100  *
 101  * ptr[1:0] == MMU_NORMAL_PT_UPDATE:
 102  * Updates an entry in a page table. If updating an L1 table, and the new
 103  * table entry is valid/present, the mapped frame must belong to the FD, if
 104  * an FD has been specified. If attempting to map an I/O page then the
 105  * caller assumes the privilege of the FD.
 106  * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller.
 107  * FD == DOMID_XEN: Map restricted areas of Xen's heap space.
 108  * ptr[:2]  -- Machine address of the page-table entry to modify.
 109  * val      -- Value to write.
 110  *
 111  * ptr[1:0] == MMU_MACHPHYS_UPDATE:
 112  * Updates an entry in the machine->pseudo-physical mapping table.
 113  * ptr[:2]  -- Machine address within the frame whose mapping to modify.
 114  *             The frame must belong to the FD, if one is specified.
 115  * val      -- Value to write into the mapping entry.
 116  */
 117 #define MMU_NORMAL_PT_UPDATE     0 /* checked '*ptr = val'. ptr is MA.       */
 118 #define MMU_MACHPHYS_UPDATE      1 /* ptr = MA of frame to modify entry for  */
 119
 120 /*
 121  * MMU EXTENDED OPERATIONS
 122  *
 123  * HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures.
 124  * A foreigndom (FD) can be specified (or DOMID_SELF for none).
 125  * Where the FD has some effect, it is described below.
 126  *
 127  * cmd: MMUEXT_(UN)PIN_*_TABLE
 128  * mfn: Machine frame number to be (un)pinned as a p.t. page.
 129  *      The frame must belong to the FD, if one is specified.
 130  *
 131  * cmd: MMUEXT_NEW_BASEPTR
 132  * mfn: Machine frame number of new page-table base to install in MMU.
 133  *
 134  * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only]
 135  * mfn: Machine frame number of new page-table base to install in MMU
 136  *      when in user space.
 137  *
 138  * cmd: MMUEXT_TLB_FLUSH_LOCAL
 139  * No additional arguments. Flushes local TLB.
 140  *
 141  * cmd: MMUEXT_INVLPG_LOCAL
 142  * linear_addr: Linear address to be flushed from the local TLB.
 143  *
 144  * cmd: MMUEXT_TLB_FLUSH_MULTI
 145  * vcpumask: Pointer to bitmap of VCPUs to be flushed.
 146  *
 147  * cmd: MMUEXT_INVLPG_MULTI
 148  * linear_addr: Linear address to be flushed.
 149  * vcpumask: Pointer to bitmap of VCPUs to be flushed.
 150  *
 151  * cmd: MMUEXT_TLB_FLUSH_ALL
 152  * No additional arguments. Flushes all VCPUs' TLBs.
 153  *
 154  * cmd: MMUEXT_INVLPG_ALL
 155  * linear_addr: Linear address to be flushed from all VCPUs' TLBs.
 156  *
 157  * cmd: MMUEXT_FLUSH_CACHE
 158  * No additional arguments. Writes back and flushes cache contents.
 159  *
 160  * cmd: MMUEXT_SET_LDT
 161  * linear_addr: Linear address of LDT base (NB. must be page-aligned).
 162  * nr_ents: Number of entries in LDT.
 163  */
 164 #define MMUEXT_PIN_L1_TABLE      0
 165 #define MMUEXT_PIN_L2_TABLE      1
 166 #define MMUEXT_PIN_L3_TABLE      2
 167 #define MMUEXT_PIN_L4_TABLE      3
 168 #define MMUEXT_UNPIN_TABLE       4
 169 #define MMUEXT_NEW_BASEPTR       5
 170 #define MMUEXT_TLB_FLUSH_LOCAL   6
 171 #define MMUEXT_INVLPG_LOCAL      7
 172 #define MMUEXT_TLB_FLUSH_MULTI   8
 173 #define MMUEXT_INVLPG_MULTI      9
 174 #define MMUEXT_TLB_FLUSH_ALL    10
 175 #define MMUEXT_INVLPG_ALL       11
 176 #define MMUEXT_FLUSH_CACHE      12
 177 #define MMUEXT_SET_LDT          13
 178 #define MMUEXT_NEW_USER_BASEPTR 15
 179
 180 #ifndef __ASSEMBLY__
 181 struct mmuext_op {
 182         unsigned int cmd;
 183         union {
 184                 /* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR */
 185                 unsigned long mfn;
 186                 /* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */
 187                 unsigned long linear_addr;
 188         } arg1;
 189         union {
 190                 /* SET_LDT */
 191                 unsigned int nr_ents;
 192                 /* TLB_FLUSH_MULTI, INVLPG_MULTI */
 193                 void *vcpumask;
 194         } arg2;
 195 };
 196 DEFINE_GUEST_HANDLE_STRUCT(mmuext_op);
 197 #endif
 198
 199 /* These are passed as 'flags' to update_va_mapping. They can be ORed. */
 200 /* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap.   */
 201 /* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer.         */
 202 #define UVMF_NONE               (0UL<<0) /* No flushing at all.   */
 203 #define UVMF_TLB_FLUSH          (1UL<<0) /* Flush entire TLB(s).  */
 204 #define UVMF_INVLPG             (2UL<<0) /* Flush only one entry. */
 205 #define UVMF_FLUSHTYPE_MASK     (3UL<<0)
 206 #define UVMF_MULTI              (0UL<<2) /* Flush subset of TLBs. */
 207 #define UVMF_LOCAL              (0UL<<2) /* Flush local TLB.      */
 208 #define UVMF_ALL                (1UL<<2) /* Flush all TLBs.       */
 209
 210 /*
 211  * Commands to HYPERVISOR_console_io().
 212  */
 213 #define CONSOLEIO_write         0
 214 #define CONSOLEIO_read          1
 215
 216 /*
 217  * Commands to HYPERVISOR_vm_assist().
 218  */
 219 #define VMASST_CMD_enable                0
 220 #define VMASST_CMD_disable               1
 221 #define VMASST_TYPE_4gb_segments         0
 222 #define VMASST_TYPE_4gb_segments_notify  1
 223 #define VMASST_TYPE_writable_pagetables  2
 224 #define VMASST_TYPE_pae_extended_cr3     3
 225 #define MAX_VMASST_TYPE 3
 226
 227 #ifndef __ASSEMBLY__
 228
 229 typedef uint16_t domid_t;
 230
 231 /* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */
 232 #define DOMID_FIRST_RESERVED (0x7FF0U)
 233
 234 /* DOMID_SELF is used in certain contexts to refer to oneself. */
 235 #define DOMID_SELF (0x7FF0U)
 236
 237 /*
 238  * DOMID_IO is used to restrict page-table updates to mapping I/O memory.
 239  * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO
 240  * is useful to ensure that no mappings to the OS's own heap are accidentally
 241  * installed. (e.g., in Linux this could cause havoc as reference counts
 242  * aren't adjusted on the I/O-mapping code path).
 243  * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can
 244  * be specified by any calling domain.
 245  */
 246 #define DOMID_IO   (0x7FF1U)
 247
 248 /*
 249  * DOMID_XEN is used to allow privileged domains to map restricted parts of
 250  * Xen's heap space (e.g., the machine_to_phys table).
 251  * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if
 252  * the caller is privileged.
 253  */
 254 #define DOMID_XEN  (0x7FF2U)
 255
 256 /*
 257  * Send an array of these to HYPERVISOR_mmu_update().
 258  * NB. The fields are natural pointer/address size for this architecture.
 259  */
 260 struct mmu_update {
 261     uint64_t ptr;       /* Machine address of PTE. */
 262     uint64_t val;       /* New contents of PTE.    */
 263 };
 264 DEFINE_GUEST_HANDLE_STRUCT(mmu_update);
 265
 266 /*
 267  * Send an array of these to HYPERVISOR_multicall().
 268  * NB. The fields are natural register size for this architecture.
 269  */
 270 struct multicall_entry {
 271     unsigned long op;
 272     long result;
 273     unsigned long args[6];
 274 };
 275 DEFINE_GUEST_HANDLE_STRUCT(multicall_entry);
 276
 277 /*
 278  * Event channel endpoints per domain:
 279  *  1024 if a long is 32 bits; 4096 if a long is 64 bits.
 280  */
 281 #define NR_EVENT_CHANNELS (sizeof(unsigned long) * sizeof(unsigned long) * 64)
 282
 283 struct vcpu_time_info {
 284         /*
 285          * Updates to the following values are preceded and followed
 286          * by an increment of 'version'. The guest can therefore
 287          * detect updates by looking for changes to 'version'. If the
 288          * least-significant bit of the version number is set then an
 289          * update is in progress and the guest must wait to read a
 290          * consistent set of values.  The correct way to interact with
 291          * the version number is similar to Linux's seqlock: see the
 292          * implementations of read_seqbegin/read_seqretry.
 293          */
 294         uint32_t version;
 295         uint32_t pad0;
 296         uint64_t tsc_timestamp;   /* TSC at last update of time vals.  */
 297         uint64_t system_time;     /* Time, in nanosecs, since boot.    */
 298         /*
 299          * Current system time:
 300          *   system_time + ((tsc - tsc_timestamp) << tsc_shift) * tsc_to_system_mul
 301          * CPU frequency (Hz):
 302          *   ((10^9 << 32) / tsc_to_system_mul) >> tsc_shift
 303          */
 304         uint32_t tsc_to_system_mul;
 305         int8_t   tsc_shift;
 306         int8_t   pad1[3];
 307 }; /* 32 bytes */
 308
 309 struct vcpu_info {
 310         /*
 311          * 'evtchn_upcall_pending' is written non-zero by Xen to indicate
 312          * a pending notification for a particular VCPU. It is then cleared
 313          * by the guest OS /before/ checking for pending work, thus avoiding
 314          * a set-and-check race. Note that the mask is only accessed by Xen
 315          * on the CPU that is currently hosting the VCPU. This means that the
 316          * pending and mask flags can be updated by the guest without special
 317          * synchronisation (i.e., no need for the x86 LOCK prefix).
 318          * This may seem suboptimal because if the pending flag is set by
 319          * a different CPU then an IPI may be scheduled even when the mask
 320          * is set. However, note:
 321          *  1. The task of 'interrupt holdoff' is covered by the per-event-
 322          *     channel mask bits. A 'noisy' event that is continually being
 323          *     triggered can be masked at source at this very precise
 324          *     granularity.
 325          *  2. The main purpose of the per-VCPU mask is therefore to restrict
 326          *     reentrant execution: whether for concurrency control, or to
 327          *     prevent unbounded stack usage. Whatever the purpose, we expect
 328          *     that the mask will be asserted only for short periods at a time,
 329          *     and so the likelihood of a 'spurious' IPI is suitably small.
 330          * The mask is read before making an event upcall to the guest: a
 331          * non-zero mask therefore guarantees that the VCPU will not receive
 332          * an upcall activation. The mask is cleared when the VCPU requests
 333          * to block: this avoids wakeup-waiting races.
 334          */
 335         uint8_t evtchn_upcall_pending;
 336         uint8_t evtchn_upcall_mask;
 337         unsigned long evtchn_pending_sel;
 338         struct arch_vcpu_info arch;
 339         struct vcpu_time_info time;
 340 }; /* 64 bytes (x86) */
 341
 342 /*
 343  * Xen/kernel shared data -- pointer provided in start_info.
 344  * NB. We expect that this struct is smaller than a page.
 345  */
 346 struct shared_info {
 347         struct vcpu_info vcpu_info[MAX_VIRT_CPUS];
 348
 349         /*
 350          * A domain can create "event channels" on which it can send and receive
 351          * asynchronous event notifications. There are three classes of event that
 352          * are delivered by this mechanism:
 353          *  1. Bi-directional inter- and intra-domain connections. Domains must
 354          *     arrange out-of-band to set up a connection (usually by allocating
 355          *     an unbound 'listener' port and avertising that via a storage service
 356          *     such as xenstore).
 357          *  2. Physical interrupts. A domain with suitable hardware-access
 358          *     privileges can bind an event-channel port to a physical interrupt
 359          *     source.
 360          *  3. Virtual interrupts ('events'). A domain can bind an event-channel
 361          *     port to a virtual interrupt source, such as the virtual-timer
 362          *     device or the emergency console.
 363          *
 364          * Event channels are addressed by a "port index". Each channel is
 365          * associated with two bits of information:
 366          *  1. PENDING -- notifies the domain that there is a pending notification
 367          *     to be processed. This bit is cleared by the guest.
 368          *  2. MASK -- if this bit is clear then a 0->1 transition of PENDING
 369          *     will cause an asynchronous upcall to be scheduled. This bit is only
 370          *     updated by the guest. It is read-only within Xen. If a channel
 371          *     becomes pending while the channel is masked then the 'edge' is lost
 372          *     (i.e., when the channel is unmasked, the guest must manually handle
 373          *     pending notifications as no upcall will be scheduled by Xen).
 374          *
 375          * To expedite scanning of pending notifications, any 0->1 pending
 376          * transition on an unmasked channel causes a corresponding bit in a
 377          * per-vcpu selector word to be set. Each bit in the selector covers a
 378          * 'C long' in the PENDING bitfield array.
 379          */
 380         unsigned long evtchn_pending[sizeof(unsigned long) * 8];
 381         unsigned long evtchn_mask[sizeof(unsigned long) * 8];
 382
 383         /*
 384          * Wallclock time: updated only by control software. Guests should base
 385          * their gettimeofday() syscall on this wallclock-base value.
 386          */
 387         uint32_t wc_version;      /* Version counter: see vcpu_time_info_t. */
 388         uint32_t wc_sec;          /* Secs  00:00:00 UTC, Jan 1, 1970.  */
 389         uint32_t wc_nsec;         /* Nsecs 00:00:00 UTC, Jan 1, 1970.  */
 390
 391         struct arch_shared_info arch;
 392
 393 };
 394
 395 /*
 396  * Start-of-day memory layout for the initial domain (DOM0):
 397  *  1. The domain is started within contiguous virtual-memory region.
 398  *  2. The contiguous region begins and ends on an aligned 4MB boundary.
 399  *  3. The region start corresponds to the load address of the OS image.
 400  *     If the load address is not 4MB aligned then the address is rounded down.
 401  *  4. This the order of bootstrap elements in the initial virtual region:
 402  *      a. relocated kernel image
 403  *      b. initial ram disk              [mod_start, mod_len]
 404  *      c. list of allocated page frames [mfn_list, nr_pages]
 405  *      d. start_info_t structure        [register ESI (x86)]
 406  *      e. bootstrap page tables         [pt_base, CR3 (x86)]
 407  *      f. bootstrap stack               [register ESP (x86)]
 408  *  5. Bootstrap elements are packed together, but each is 4kB-aligned.
 409  *  6. The initial ram disk may be omitted.
 410  *  7. The list of page frames forms a contiguous 'pseudo-physical' memory
 411  *     layout for the domain. In particular, the bootstrap virtual-memory
 412  *     region is a 1:1 mapping to the first section of the pseudo-physical map.
 413  *  8. All bootstrap elements are mapped read-writable for the guest OS. The
 414  *     only exception is the bootstrap page table, which is mapped read-only.
 415  *  9. There is guaranteed to be at least 512kB padding after the final
 416  *     bootstrap element. If necessary, the bootstrap virtual region is
 417  *     extended by an extra 4MB to ensure this.
 418  */
 419
 420 #define MAX_GUEST_CMDLINE 1024
 421 struct start_info {
 422         /* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME.    */
 423         char magic[32];             /* "xen-<version>-<platform>".            */
 424         unsigned long nr_pages;     /* Total pages allocated to this domain.  */
 425         unsigned long shared_info;  /* MACHINE address of shared info struct. */
 426         uint32_t flags;             /* SIF_xxx flags.                         */
 427         unsigned long store_mfn;    /* MACHINE page number of shared page.    */
 428         uint32_t store_evtchn;      /* Event channel for store communication. */
 429         union {
 430                 struct {
 431                         unsigned long mfn;  /* MACHINE page number of console page.   */
 432                         uint32_t  evtchn;   /* Event channel for console page.        */
 433                 } domU;
 434                 struct {
 435                         uint32_t info_off;  /* Offset of console_info struct.         */
 436                         uint32_t info_size; /* Size of console_info struct from start.*/
 437                 } dom0;
 438         } console;
 439         /* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME).     */
 440         unsigned long pt_base;      /* VIRTUAL address of page directory.     */
 441         unsigned long nr_pt_frames; /* Number of bootstrap p.t. frames.       */
 442         unsigned long mfn_list;     /* VIRTUAL address of page-frame list.    */
 443         unsigned long mod_start;    /* VIRTUAL address of pre-loaded module.  */
 444         unsigned long mod_len;      /* Size (bytes) of pre-loaded module.     */
 445         int8_t cmd_line[MAX_GUEST_CMDLINE];
 446 };
 447
 448 /* These flags are passed in the 'flags' field of start_info_t. */
 449 #define SIF_PRIVILEGED    (1<<0)  /* Is the domain privileged? */
 450 #define SIF_INITDOMAIN    (1<<1)  /* Is this the initial control domain? */
 451
 452 typedef uint64_t cpumap_t;
 453
 454 typedef uint8_t xen_domain_handle_t[16];
 455
 456 /* Turn a plain number into a C unsigned long constant. */
 457 #define __mk_unsigned_long(x) x ## UL
 458 #define mk_unsigned_long(x) __mk_unsigned_long(x)
 459
 460 #else /* __ASSEMBLY__ */
 461
 462 /* In assembly code we cannot use C numeric constant suffixes. */
 463 #define mk_unsigned_long(x) x
 464
 465 #endif /* !__ASSEMBLY__ */
 466
 467 #endif /* __XEN_PUBLIC_XEN_H__ */