1 /******************************************************************************
4 * Guest OS interface to Xen.
6 * Copyright (c) 2004, K A Fraser
9 #ifndef __XEN_PUBLIC_XEN_H__
10 #define __XEN_PUBLIC_XEN_H__
13 #include "arch-x86_32.h"
14 #elif defined(__x86_64__)
15 #include "arch-x86_64.h"
16 #elif defined(__ia64__)
17 #include "arch-ia64.h"
18 #elif defined(__powerpc__)
19 #include "arch-powerpc.h"
21 #error "Unsupported architecture"
25 * XEN "SYSTEM CALLS" (a.k.a. HYPERCALLS).
29 * x86_32: EAX = vector; EBX, ECX, EDX, ESI, EDI = args 1, 2, 3, 4, 5.
31 * (argument registers may be clobbered on return)
32 * x86_64: RAX = vector; RDI, RSI, RDX, R10, R8, R9 = args 1, 2, 3, 4, 5, 6.
34 * (argument registers not clobbered on return; RCX, R11 are)
36 #define __HYPERVISOR_set_trap_table 0
37 #define __HYPERVISOR_mmu_update 1
38 #define __HYPERVISOR_set_gdt 2
39 #define __HYPERVISOR_stack_switch 3
40 #define __HYPERVISOR_set_callbacks 4
41 #define __HYPERVISOR_fpu_taskswitch 5
42 #define __HYPERVISOR_sched_op_compat 6 /* compat since 0x00030101 */
43 #define __HYPERVISOR_dom0_op 7
44 #define __HYPERVISOR_set_debugreg 8
45 #define __HYPERVISOR_get_debugreg 9
46 #define __HYPERVISOR_update_descriptor 10
47 #define __HYPERVISOR_memory_op 12
48 #define __HYPERVISOR_multicall 13
49 #define __HYPERVISOR_update_va_mapping 14
50 #define __HYPERVISOR_set_timer_op 15
51 #define __HYPERVISOR_event_channel_op_compat 16 /* compat since 0x00030202 */
52 #define __HYPERVISOR_xen_version 17
53 #define __HYPERVISOR_console_io 18
54 #define __HYPERVISOR_physdev_op_compat 19 /* compat since 0x00030202 */
55 #define __HYPERVISOR_grant_table_op 20
56 #define __HYPERVISOR_vm_assist 21
57 #define __HYPERVISOR_update_va_mapping_otherdomain 22
58 #define __HYPERVISOR_iret 23 /* x86 only */
59 #define __HYPERVISOR_vcpu_op 24
60 #define __HYPERVISOR_set_segment_base 25 /* x86/64 only */
61 #define __HYPERVISOR_mmuext_op 26
62 #define __HYPERVISOR_acm_op 27
63 #define __HYPERVISOR_nmi_op 28
64 #define __HYPERVISOR_sched_op 29
65 #define __HYPERVISOR_callback_op 30
66 #define __HYPERVISOR_xenoprof_op 31
67 #define __HYPERVISOR_event_channel_op 32
68 #define __HYPERVISOR_physdev_op 33
69 #define __HYPERVISOR_hvm_op 34
71 /* Architecture-specific hypercall definitions. */
72 #define __HYPERVISOR_arch_0 48
73 #define __HYPERVISOR_arch_1 49
74 #define __HYPERVISOR_arch_2 50
75 #define __HYPERVISOR_arch_3 51
76 #define __HYPERVISOR_arch_4 52
77 #define __HYPERVISOR_arch_5 53
78 #define __HYPERVISOR_arch_6 54
79 #define __HYPERVISOR_arch_7 55
84 * Virtual interrupts that a guest OS may receive from Xen.
86 * In the side comments, 'V.' denotes a per-VCPU VIRQ while 'G.' denotes a
87 * global VIRQ. The former can be bound once per VCPU and cannot be re-bound.
88 * The latter can be allocated only once per guest: they must initially be
89 * allocated to VCPU0 but can subsequently be re-bound.
91 #define VIRQ_TIMER 0 /* V. Timebase update, and/or requested timeout. */
92 #define VIRQ_DEBUG 1 /* V. Request guest to dump debug info. */
93 #define VIRQ_CONSOLE 2 /* G. (DOM0) Bytes received on emergency console. */
94 #define VIRQ_DOM_EXC 3 /* G. (DOM0) Exceptional event for some domain. */
95 #define VIRQ_TBUF 4 /* G. (DOM0) Trace buffer has records available. */
96 #define VIRQ_DEBUGGER 6 /* G. (DOM0) A domain has paused for debugging. */
97 #define VIRQ_XENOPROF 7 /* V. XenOprofile interrupt: new sample available */
99 /* Architecture-specific VIRQ definitions. */
100 #define VIRQ_ARCH_0 16
101 #define VIRQ_ARCH_1 17
102 #define VIRQ_ARCH_2 18
103 #define VIRQ_ARCH_3 19
104 #define VIRQ_ARCH_4 20
105 #define VIRQ_ARCH_5 21
106 #define VIRQ_ARCH_6 22
107 #define VIRQ_ARCH_7 23
112 * MMU-UPDATE REQUESTS
114 * HYPERVISOR_mmu_update() accepts a list of (ptr, val) pairs.
115 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
116 * Where the FD has some effect, it is described below.
117 * ptr[1:0] specifies the appropriate MMU_* command.
119 * ptr[1:0] == MMU_NORMAL_PT_UPDATE:
120 * Updates an entry in a page table. If updating an L1 table, and the new
121 * table entry is valid/present, the mapped frame must belong to the FD, if
122 * an FD has been specified. If attempting to map an I/O page then the
123 * caller assumes the privilege of the FD.
124 * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller.
125 * FD == DOMID_XEN: Map restricted areas of Xen's heap space.
126 * ptr[:2] -- Machine address of the page-table entry to modify.
127 * val -- Value to write.
129 * ptr[1:0] == MMU_MACHPHYS_UPDATE:
130 * Updates an entry in the machine->pseudo-physical mapping table.
131 * ptr[:2] -- Machine address within the frame whose mapping to modify.
132 * The frame must belong to the FD, if one is specified.
133 * val -- Value to write into the mapping entry.
135 #define MMU_NORMAL_PT_UPDATE 0 /* checked '*ptr = val'. ptr is MA. */
136 #define MMU_MACHPHYS_UPDATE 1 /* ptr = MA of frame to modify entry for */
139 * MMU EXTENDED OPERATIONS
141 * HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures.
142 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
143 * Where the FD has some effect, it is described below.
145 * cmd: MMUEXT_(UN)PIN_*_TABLE
146 * mfn: Machine frame number to be (un)pinned as a p.t. page.
147 * The frame must belong to the FD, if one is specified.
149 * cmd: MMUEXT_NEW_BASEPTR
150 * mfn: Machine frame number of new page-table base to install in MMU.
152 * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only]
153 * mfn: Machine frame number of new page-table base to install in MMU
154 * when in user space.
156 * cmd: MMUEXT_TLB_FLUSH_LOCAL
157 * No additional arguments. Flushes local TLB.
159 * cmd: MMUEXT_INVLPG_LOCAL
160 * linear_addr: Linear address to be flushed from the local TLB.
162 * cmd: MMUEXT_TLB_FLUSH_MULTI
163 * vcpumask: Pointer to bitmap of VCPUs to be flushed.
165 * cmd: MMUEXT_INVLPG_MULTI
166 * linear_addr: Linear address to be flushed.
167 * vcpumask: Pointer to bitmap of VCPUs to be flushed.
169 * cmd: MMUEXT_TLB_FLUSH_ALL
170 * No additional arguments. Flushes all VCPUs' TLBs.
172 * cmd: MMUEXT_INVLPG_ALL
173 * linear_addr: Linear address to be flushed from all VCPUs' TLBs.
175 * cmd: MMUEXT_FLUSH_CACHE
176 * No additional arguments. Writes back and flushes cache contents.
178 * cmd: MMUEXT_SET_LDT
179 * linear_addr: Linear address of LDT base (NB. must be page-aligned).
180 * nr_ents: Number of entries in LDT.
182 #define MMUEXT_PIN_L1_TABLE 0
183 #define MMUEXT_PIN_L2_TABLE 1
184 #define MMUEXT_PIN_L3_TABLE 2
185 #define MMUEXT_PIN_L4_TABLE 3
186 #define MMUEXT_UNPIN_TABLE 4
187 #define MMUEXT_NEW_BASEPTR 5
188 #define MMUEXT_TLB_FLUSH_LOCAL 6
189 #define MMUEXT_INVLPG_LOCAL 7
190 #define MMUEXT_TLB_FLUSH_MULTI 8
191 #define MMUEXT_INVLPG_MULTI 9
192 #define MMUEXT_TLB_FLUSH_ALL 10
193 #define MMUEXT_INVLPG_ALL 11
194 #define MMUEXT_FLUSH_CACHE 12
195 #define MMUEXT_SET_LDT 13
196 #define MMUEXT_NEW_USER_BASEPTR 15
202 /* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR */
204 /* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */
205 unsigned long linear_addr;
209 unsigned int nr_ents;
210 /* TLB_FLUSH_MULTI, INVLPG_MULTI */
214 typedef struct mmuext_op mmuext_op_t;
215 DEFINE_XEN_GUEST_HANDLE(mmuext_op_t);
218 /* These are passed as 'flags' to update_va_mapping. They can be ORed. */
219 /* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap. */
220 /* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer. */
221 #define UVMF_NONE (0UL<<0) /* No flushing at all. */
222 #define UVMF_TLB_FLUSH (1UL<<0) /* Flush entire TLB(s). */
223 #define UVMF_INVLPG (2UL<<0) /* Flush only one entry. */
224 #define UVMF_FLUSHTYPE_MASK (3UL<<0)
225 #define UVMF_MULTI (0UL<<2) /* Flush subset of TLBs. */
226 #define UVMF_LOCAL (0UL<<2) /* Flush local TLB. */
227 #define UVMF_ALL (1UL<<2) /* Flush all TLBs. */
230 * Commands to HYPERVISOR_console_io().
232 #define CONSOLEIO_write 0
233 #define CONSOLEIO_read 1
236 * Commands to HYPERVISOR_vm_assist().
238 #define VMASST_CMD_enable 0
239 #define VMASST_CMD_disable 1
241 /* x86/32 guests: simulate full 4GB segment limits. */
242 #define VMASST_TYPE_4gb_segments 0
244 /* x86/32 guests: trap (vector 15) whenever above vmassist is used. */
245 #define VMASST_TYPE_4gb_segments_notify 1
248 * x86 guests: support writes to bottom-level PTEs.
249 * NB1. Page-directory entries cannot be written.
250 * NB2. Guest must continue to remove all writable mappings of PTEs.
252 #define VMASST_TYPE_writable_pagetables 2
254 /* x86/PAE guests: support PDPTs above 4GB. */
255 #define VMASST_TYPE_pae_extended_cr3 3
257 #define MAX_VMASST_TYPE 3
261 typedef uint16_t domid_t;
263 /* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */
264 #define DOMID_FIRST_RESERVED (0x7FF0U)
266 /* DOMID_SELF is used in certain contexts to refer to oneself. */
267 #define DOMID_SELF (0x7FF0U)
270 * DOMID_IO is used to restrict page-table updates to mapping I/O memory.
271 * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO
272 * is useful to ensure that no mappings to the OS's own heap are accidentally
273 * installed. (e.g., in Linux this could cause havoc as reference counts
274 * aren't adjusted on the I/O-mapping code path).
275 * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can
276 * be specified by any calling domain.
278 #define DOMID_IO (0x7FF1U)
281 * DOMID_XEN is used to allow privileged domains to map restricted parts of
282 * Xen's heap space (e.g., the machine_to_phys table).
283 * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if
284 * the caller is privileged.
286 #define DOMID_XEN (0x7FF2U)
289 * Send an array of these to HYPERVISOR_mmu_update().
290 * NB. The fields are natural pointer/address size for this architecture.
293 uint64_t ptr; /* Machine address of PTE. */
294 uint64_t val; /* New contents of PTE. */
296 typedef struct mmu_update mmu_update_t;
297 DEFINE_XEN_GUEST_HANDLE(mmu_update_t);
300 * Send an array of these to HYPERVISOR_multicall().
301 * NB. The fields are natural register size for this architecture.
303 struct multicall_entry {
304 unsigned long op, result;
305 unsigned long args[6];
307 typedef struct multicall_entry multicall_entry_t;
308 DEFINE_XEN_GUEST_HANDLE(multicall_entry_t);
311 * Event channel endpoints per domain:
312 * 1024 if a long is 32 bits; 4096 if a long is 64 bits.
314 #define NR_EVENT_CHANNELS (sizeof(unsigned long) * sizeof(unsigned long) * 64)
316 struct vcpu_time_info {
318 * Updates to the following values are preceded and followed by an
319 * increment of 'version'. The guest can therefore detect updates by
320 * looking for changes to 'version'. If the least-significant bit of
321 * the version number is set then an update is in progress and the guest
322 * must wait to read a consistent set of values.
323 * The correct way to interact with the version number is similar to
324 * Linux's seqlock: see the implementations of read_seqbegin/read_seqretry.
328 uint64_t tsc_timestamp; /* TSC at last update of time vals. */
329 uint64_t system_time; /* Time, in nanosecs, since boot. */
331 * Current system time:
333 * ((((tsc - tsc_timestamp) << tsc_shift) * tsc_to_system_mul) >> 32)
334 * CPU frequency (Hz):
335 * ((10^9 << 32) / tsc_to_system_mul) >> tsc_shift
337 uint32_t tsc_to_system_mul;
341 typedef struct vcpu_time_info vcpu_time_info_t;
345 * 'evtchn_upcall_pending' is written non-zero by Xen to indicate
346 * a pending notification for a particular VCPU. It is then cleared
347 * by the guest OS /before/ checking for pending work, thus avoiding
348 * a set-and-check race. Note that the mask is only accessed by Xen
349 * on the CPU that is currently hosting the VCPU. This means that the
350 * pending and mask flags can be updated by the guest without special
351 * synchronisation (i.e., no need for the x86 LOCK prefix).
352 * This may seem suboptimal because if the pending flag is set by
353 * a different CPU then an IPI may be scheduled even when the mask
354 * is set. However, note:
355 * 1. The task of 'interrupt holdoff' is covered by the per-event-
356 * channel mask bits. A 'noisy' event that is continually being
357 * triggered can be masked at source at this very precise
359 * 2. The main purpose of the per-VCPU mask is therefore to restrict
360 * reentrant execution: whether for concurrency control, or to
361 * prevent unbounded stack usage. Whatever the purpose, we expect
362 * that the mask will be asserted only for short periods at a time,
363 * and so the likelihood of a 'spurious' IPI is suitably small.
364 * The mask is read before making an event upcall to the guest: a
365 * non-zero mask therefore guarantees that the VCPU will not receive
366 * an upcall activation. The mask is cleared when the VCPU requests
367 * to block: this avoids wakeup-waiting races.
369 uint8_t evtchn_upcall_pending;
370 uint8_t evtchn_upcall_mask;
371 unsigned long evtchn_pending_sel;
372 struct arch_vcpu_info arch;
373 struct vcpu_time_info time;
374 }; /* 64 bytes (x86) */
375 typedef struct vcpu_info vcpu_info_t;
378 * Xen/kernel shared data -- pointer provided in start_info.
380 * This structure is defined to be both smaller than a page, and the
381 * only data on the shared page, but may vary in actual size even within
382 * compatible Xen versions; guests should not rely on the size
383 * of this structure remaining constant.
386 struct vcpu_info vcpu_info[MAX_VIRT_CPUS];
389 * A domain can create "event channels" on which it can send and receive
390 * asynchronous event notifications. There are three classes of event that
391 * are delivered by this mechanism:
392 * 1. Bi-directional inter- and intra-domain connections. Domains must
393 * arrange out-of-band to set up a connection (usually by allocating
394 * an unbound 'listener' port and avertising that via a storage service
396 * 2. Physical interrupts. A domain with suitable hardware-access
397 * privileges can bind an event-channel port to a physical interrupt
399 * 3. Virtual interrupts ('events'). A domain can bind an event-channel
400 * port to a virtual interrupt source, such as the virtual-timer
401 * device or the emergency console.
403 * Event channels are addressed by a "port index". Each channel is
404 * associated with two bits of information:
405 * 1. PENDING -- notifies the domain that there is a pending notification
406 * to be processed. This bit is cleared by the guest.
407 * 2. MASK -- if this bit is clear then a 0->1 transition of PENDING
408 * will cause an asynchronous upcall to be scheduled. This bit is only
409 * updated by the guest. It is read-only within Xen. If a channel
410 * becomes pending while the channel is masked then the 'edge' is lost
411 * (i.e., when the channel is unmasked, the guest must manually handle
412 * pending notifications as no upcall will be scheduled by Xen).
414 * To expedite scanning of pending notifications, any 0->1 pending
415 * transition on an unmasked channel causes a corresponding bit in a
416 * per-vcpu selector word to be set. Each bit in the selector covers a
417 * 'C long' in the PENDING bitfield array.
419 unsigned long evtchn_pending[sizeof(unsigned long) * 8];
420 unsigned long evtchn_mask[sizeof(unsigned long) * 8];
423 * Wallclock time: updated only by control software. Guests should base
424 * their gettimeofday() syscall on this wallclock-base value.
426 uint32_t wc_version; /* Version counter: see vcpu_time_info_t. */
427 uint32_t wc_sec; /* Secs 00:00:00 UTC, Jan 1, 1970. */
428 uint32_t wc_nsec; /* Nsecs 00:00:00 UTC, Jan 1, 1970. */
430 struct arch_shared_info arch;
433 typedef struct shared_info shared_info_t;
436 * Start-of-day memory layout for the initial domain (DOM0):
437 * 1. The domain is started within contiguous virtual-memory region.
438 * 2. The contiguous region begins and ends on an aligned 4MB boundary.
439 * 3. The region start corresponds to the load address of the OS image.
440 * If the load address is not 4MB aligned then the address is rounded down.
441 * 4. This the order of bootstrap elements in the initial virtual region:
442 * a. relocated kernel image
443 * b. initial ram disk [mod_start, mod_len]
444 * c. list of allocated page frames [mfn_list, nr_pages]
445 * d. start_info_t structure [register ESI (x86)]
446 * e. bootstrap page tables [pt_base, CR3 (x86)]
447 * f. bootstrap stack [register ESP (x86)]
448 * 5. Bootstrap elements are packed together, but each is 4kB-aligned.
449 * 6. The initial ram disk may be omitted.
450 * 7. The list of page frames forms a contiguous 'pseudo-physical' memory
451 * layout for the domain. In particular, the bootstrap virtual-memory
452 * region is a 1:1 mapping to the first section of the pseudo-physical map.
453 * 8. All bootstrap elements are mapped read-writable for the guest OS. The
454 * only exception is the bootstrap page table, which is mapped read-only.
455 * 9. There is guaranteed to be at least 512kB padding after the final
456 * bootstrap element. If necessary, the bootstrap virtual region is
457 * extended by an extra 4MB to ensure this.
460 #define MAX_GUEST_CMDLINE 1024
462 /* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME. */
463 char magic[32]; /* "xen-<version>-<platform>". */
464 unsigned long nr_pages; /* Total pages allocated to this domain. */
465 unsigned long shared_info; /* MACHINE address of shared info struct. */
466 uint32_t flags; /* SIF_xxx flags. */
467 xen_pfn_t store_mfn; /* MACHINE page number of shared page. */
468 uint32_t store_evtchn; /* Event channel for store communication. */
469 xen_pfn_t console_mfn; /* MACHINE page number of console page. */
470 uint32_t console_evtchn; /* Event channel for console messages. */
471 /* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME). */
472 unsigned long pt_base; /* VIRTUAL address of page directory. */
473 unsigned long nr_pt_frames; /* Number of bootstrap p.t. frames. */
474 unsigned long mfn_list; /* VIRTUAL address of page-frame list. */
475 unsigned long mod_start; /* VIRTUAL address of pre-loaded module. */
476 unsigned long mod_len; /* Size (bytes) of pre-loaded module. */
477 int8_t cmd_line[MAX_GUEST_CMDLINE];
479 typedef struct start_info start_info_t;
481 /* These flags are passed in the 'flags' field of start_info_t. */
482 #define SIF_PRIVILEGED (1<<0) /* Is the domain privileged? */
483 #define SIF_INITDOMAIN (1<<1) /* Is this the initial control domain? */
485 typedef uint64_t cpumap_t;
487 typedef uint8_t xen_domain_handle_t[16];
489 /* Turn a plain number into a C unsigned long constant. */
490 #define __mk_unsigned_long(x) x ## UL
491 #define mk_unsigned_long(x) __mk_unsigned_long(x)
493 #else /* __ASSEMBLY__ */
495 /* In assembly code we cannot use C numeric constant suffixes. */
496 #define mk_unsigned_long(x) x
498 #endif /* !__ASSEMBLY__ */
500 #include "xen-compat.h"
502 #endif /* __XEN_PUBLIC_XEN_H__ */
510 * indent-tabs-mode: nil