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"
19 #error "Unsupported architecture"
23 * XEN "SYSTEM CALLS" (a.k.a. HYPERCALLS).
27 * x86_32: EAX = vector; EBX, ECX, EDX, ESI, EDI = args 1, 2, 3, 4, 5.
29 * (argument registers may be clobbered on return)
30 * x86_64: RAX = vector; RDI, RSI, RDX, R10, R8, R9 = args 1, 2, 3, 4, 5, 6.
32 * (argument registers not clobbered on return; RCX, R11 are)
34 #define __HYPERVISOR_set_trap_table 0
35 #define __HYPERVISOR_mmu_update 1
36 #define __HYPERVISOR_set_gdt 2
37 #define __HYPERVISOR_stack_switch 3
38 #define __HYPERVISOR_set_callbacks 4
39 #define __HYPERVISOR_fpu_taskswitch 5
40 #define __HYPERVISOR_sched_op 6
41 #define __HYPERVISOR_dom0_op 7
42 #define __HYPERVISOR_set_debugreg 8
43 #define __HYPERVISOR_get_debugreg 9
44 #define __HYPERVISOR_update_descriptor 10
45 #define __HYPERVISOR_set_fast_trap 11 /* x86/32 only */
46 #define __HYPERVISOR_dom_mem_op 12
47 #define __HYPERVISOR_multicall 13
48 #define __HYPERVISOR_update_va_mapping 14
49 #define __HYPERVISOR_set_timer_op 15
50 #define __HYPERVISOR_event_channel_op 16
51 #define __HYPERVISOR_xen_version 17
52 #define __HYPERVISOR_console_io 18
53 #define __HYPERVISOR_physdev_op 19
54 #define __HYPERVISOR_grant_table_op 20
55 #define __HYPERVISOR_vm_assist 21
56 #define __HYPERVISOR_update_va_mapping_otherdomain 22
57 #define __HYPERVISOR_switch_vm86 23 /* x86/32 only */
58 #define __HYPERVISOR_switch_to_user 23 /* x86/64 only */
59 #define __HYPERVISOR_boot_vcpu 24
60 #define __HYPERVISOR_set_segment_base 25 /* x86/64 only */
61 #define __HYPERVISOR_mmuext_op 26
66 * Multicalls are listed in an array, with each element being a fixed size
67 * (BYTES_PER_MULTICALL_ENTRY). Each is of the form (op, arg1, ..., argN)
68 * where each element of the tuple is a machine word.
70 #define ARGS_PER_MULTICALL_ENTRY 8
76 * Virtual interrupts that a guest OS may receive from Xen.
78 #define VIRQ_TIMER 0 /* Timebase update, and/or requested timeout. */
79 #define VIRQ_DEBUG 1 /* Request guest to dump debug info. */
80 #define VIRQ_CONSOLE 2 /* (DOM0) bytes received on emergency console. */
81 #define VIRQ_DOM_EXC 3 /* (DOM0) Exceptional event for some domain. */
82 #define VIRQ_PARITY_ERR 4 /* (DOM0) NMI parity error. */
83 #define VIRQ_IO_ERR 5 /* (DOM0) NMI I/O error. */
89 * HYPERVISOR_mmu_update() accepts a list of (ptr, val) pairs.
90 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
91 * Where the FD has some effect, it is described below.
92 * ptr[1:0] specifies the appropriate MMU_* command.
94 * ptr[1:0] == MMU_NORMAL_PT_UPDATE:
95 * Updates an entry in a page table. If updating an L1 table, and the new
96 * table entry is valid/present, the mapped frame must belong to the FD, if
97 * an FD has been specified. If attempting to map an I/O page then the
98 * caller assumes the privilege of the FD.
99 * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller.
100 * FD == DOMID_XEN: Map restricted areas of Xen's heap space.
101 * ptr[:2] -- Machine address of the page-table entry to modify.
102 * val -- Value to write.
104 * ptr[1:0] == MMU_MACHPHYS_UPDATE:
105 * Updates an entry in the machine->pseudo-physical mapping table.
106 * ptr[:2] -- Machine address within the frame whose mapping to modify.
107 * The frame must belong to the FD, if one is specified.
108 * val -- Value to write into the mapping entry.
110 #define MMU_NORMAL_PT_UPDATE 0 /* checked '*ptr = val'. ptr is MA. */
111 #define MMU_MACHPHYS_UPDATE 1 /* ptr = MA of frame to modify entry for */
114 * MMU EXTENDED OPERATIONS
116 * HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures.
117 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
118 * Where the FD has some effect, it is described below.
120 * cmd: MMUEXT_(UN)PIN_*_TABLE
121 * mfn: Machine frame number to be (un)pinned as a p.t. page.
122 * The frame must belong to the FD, if one is specified.
124 * cmd: MMUEXT_NEW_BASEPTR
125 * mfn: Machine frame number of new page-table base to install in MMU.
127 * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only]
128 * mfn: Machine frame number of new page-table base to install in MMU
129 * when in user space.
131 * cmd: MMUEXT_TLB_FLUSH_LOCAL
132 * No additional arguments. Flushes local TLB.
134 * cmd: MMUEXT_INVLPG_LOCAL
135 * linear_addr: Linear address to be flushed from the local TLB.
137 * cmd: MMUEXT_TLB_FLUSH_MULTI
138 * cpuset: Pointer to bitmap of VCPUs to be flushed.
140 * cmd: MMUEXT_INVLPG_MULTI
141 * linear_addr: Linear address to be flushed.
142 * cpuset: Pointer to bitmap of VCPUs to be flushed.
144 * cmd: MMUEXT_TLB_FLUSH_ALL
145 * No additional arguments. Flushes all VCPUs' TLBs.
147 * cmd: MMUEXT_INVLPG_ALL
148 * linear_addr: Linear address to be flushed from all VCPUs' TLBs.
150 * cmd: MMUEXT_FLUSH_CACHE
151 * No additional arguments. Writes back and flushes cache contents.
153 * cmd: MMUEXT_SET_LDT
154 * linear_addr: Linear address of LDT base (NB. must be page-aligned).
155 * nr_ents: Number of entries in LDT.
157 * cmd: MMUEXT_REASSIGN_PAGE
158 * mfn: Machine frame number to be reassigned to the FD.
159 * (NB. page must currently belong to the calling domain).
161 #define MMUEXT_PIN_L1_TABLE 0
162 #define MMUEXT_PIN_L2_TABLE 1
163 #define MMUEXT_PIN_L3_TABLE 2
164 #define MMUEXT_PIN_L4_TABLE 3
165 #define MMUEXT_UNPIN_TABLE 4
166 #define MMUEXT_NEW_BASEPTR 5
167 #define MMUEXT_TLB_FLUSH_LOCAL 6
168 #define MMUEXT_INVLPG_LOCAL 7
169 #define MMUEXT_TLB_FLUSH_MULTI 8
170 #define MMUEXT_INVLPG_MULTI 9
171 #define MMUEXT_TLB_FLUSH_ALL 10
172 #define MMUEXT_INVLPG_ALL 11
173 #define MMUEXT_FLUSH_CACHE 12
174 #define MMUEXT_SET_LDT 13
175 #define MMUEXT_REASSIGN_PAGE 14
176 #define MMUEXT_NEW_USER_BASEPTR 15
182 /* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR, REASSIGN_PAGE */
184 /* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */
185 memory_t linear_addr;
189 unsigned int nr_ents;
190 /* TLB_FLUSH_MULTI, INVLPG_MULTI */
196 /* These are passed as 'flags' to update_va_mapping. They can be ORed. */
197 /* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap. */
198 /* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer. */
199 #define UVMF_NONE (0UL) /* No flushing at all. */
200 #define UVMF_TLB_FLUSH (1UL<<0) /* Flush entire TLB(s). */
201 #define UVMF_INVLPG (2UL<<0) /* Flush only one entry. */
202 #define UVMF_FLUSHTYPE_MASK (3UL<<0)
203 #define UVMF_MULTI (0UL<<1) /* Flush subset of TLBs. */
204 #define UVMF_LOCAL (0UL<<2) /* Flush local TLB. */
205 #define UVMF_ALL (1UL<<2) /* Flush all TLBs. */
208 * Commands to HYPERVISOR_sched_op().
210 #define SCHEDOP_yield 0 /* Give up the CPU voluntarily. */
211 #define SCHEDOP_block 1 /* Block until an event is received. */
212 #define SCHEDOP_shutdown 2 /* Stop executing this domain. */
213 #define SCHEDOP_cmdmask 255 /* 8-bit command. */
214 #define SCHEDOP_reasonshift 8 /* 8-bit reason code. (SCHEDOP_shutdown) */
217 * Reason codes for SCHEDOP_shutdown. These may be interpreted by control
218 * software to determine the appropriate action. For the most part, Xen does
219 * not care about the shutdown code (SHUTDOWN_crash excepted).
221 #define SHUTDOWN_poweroff 0 /* Domain exited normally. Clean up and kill. */
222 #define SHUTDOWN_reboot 1 /* Clean up, kill, and then restart. */
223 #define SHUTDOWN_suspend 2 /* Clean up, save suspend info, kill. */
224 #define SHUTDOWN_crash 3 /* Tell controller we've crashed. */
227 * Commands to HYPERVISOR_console_io().
229 #define CONSOLEIO_write 0
230 #define CONSOLEIO_read 1
233 * Commands to HYPERVISOR_dom_mem_op().
235 #define MEMOP_increase_reservation 0
236 #define MEMOP_decrease_reservation 1
239 * Commands to HYPERVISOR_vm_assist().
241 #define VMASST_CMD_enable 0
242 #define VMASST_CMD_disable 1
243 #define VMASST_TYPE_4gb_segments 0
244 #define VMASST_TYPE_4gb_segments_notify 1
245 #define VMASST_TYPE_writable_pagetables 2
246 #define MAX_VMASST_TYPE 2
252 /* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */
253 #define DOMID_FIRST_RESERVED (0x7FF0U)
255 /* DOMID_SELF is used in certain contexts to refer to oneself. */
256 #define DOMID_SELF (0x7FF0U)
259 * DOMID_IO is used to restrict page-table updates to mapping I/O memory.
260 * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO
261 * is useful to ensure that no mappings to the OS's own heap are accidentally
262 * installed. (e.g., in Linux this could cause havoc as reference counts
263 * aren't adjusted on the I/O-mapping code path).
264 * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can
265 * be specified by any calling domain.
267 #define DOMID_IO (0x7FF1U)
270 * DOMID_XEN is used to allow privileged domains to map restricted parts of
271 * Xen's heap space (e.g., the machine_to_phys table).
272 * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if
273 * the caller is privileged.
275 #define DOMID_XEN (0x7FF2U)
278 * Send an array of these to HYPERVISOR_mmu_update().
279 * NB. The fields are natural pointer/address size for this architecture.
283 memory_t ptr; /* Machine address of PTE. */
284 memory_t val; /* New contents of PTE. */
285 } PACKED mmu_update_t;
288 * Send an array of these to HYPERVISOR_multicall().
289 * NB. The fields are natural register size for this architecture.
295 } PACKED multicall_entry_t;
297 /* Event channel endpoints per domain. */
298 #define NR_EVENT_CHANNELS 1024
300 /* Support for multi-processor guests. */
301 #define MAX_VIRT_CPUS 32
304 * Per-VCPU information goes here. This will be cleaned up more when Xen
305 * actually supports multi-VCPU guests.
310 * 'evtchn_upcall_pending' is written non-zero by Xen to indicate
311 * a pending notification for a particular VCPU. It is then cleared
312 * by the guest OS /before/ checking for pending work, thus avoiding
313 * a set-and-check race. Note that the mask is only accessed by Xen
314 * on the CPU that is currently hosting the VCPU. This means that the
315 * pending and mask flags can be updated by the guest without special
316 * synchronisation (i.e., no need for the x86 LOCK prefix).
317 * This may seem suboptimal because if the pending flag is set by
318 * a different CPU then an IPI may be scheduled even when the mask
319 * is set. However, note:
320 * 1. The task of 'interrupt holdoff' is covered by the per-event-
321 * channel mask bits. A 'noisy' event that is continually being
322 * triggered can be masked at source at this very precise
324 * 2. The main purpose of the per-VCPU mask is therefore to restrict
325 * reentrant execution: whether for concurrency control, or to
326 * prevent unbounded stack usage. Whatever the purpose, we expect
327 * that the mask will be asserted only for short periods at a time,
328 * and so the likelihood of a 'spurious' IPI is suitably small.
329 * The mask is read before making an event upcall to the guest: a
330 * non-zero mask therefore guarantees that the VCPU will not receive
331 * an upcall activation. The mask is cleared when the VCPU requests
332 * to block: this avoids wakeup-waiting races.
334 u8 evtchn_upcall_pending; /* 0 */
335 u8 evtchn_upcall_mask; /* 1 */
337 u32 evtchn_pending_sel; /* 4 */
338 arch_vcpu_info_t arch; /* 8 */
339 } PACKED vcpu_info_t; /* 8 + arch */
342 * Xen/kernel shared data -- pointer provided in start_info.
343 * NB. We expect that this struct is smaller than a page.
345 typedef struct shared_info_st
347 vcpu_info_t vcpu_data[MAX_VIRT_CPUS]; /* 0 */
352 * A domain can have up to 1024 "event channels" on which it can send
353 * and receive asynchronous event notifications. There are three classes
354 * of event that are delivered by this mechanism:
355 * 1. Bi-directional inter- and intra-domain connections. Domains must
356 * arrange out-of-band to set up a connection (usually the setup
357 * is initiated and organised by a privileged third party such as
358 * software running in domain 0).
359 * 2. Physical interrupts. A domain with suitable hardware-access
360 * privileges can bind an event-channel port to a physical interrupt
362 * 3. Virtual interrupts ('events'). A domain can bind an event-channel
363 * port to a virtual interrupt source, such as the virtual-timer
364 * device or the emergency console.
366 * Event channels are addressed by a "port index" between 0 and 1023.
367 * Each channel is associated with two bits of information:
368 * 1. PENDING -- notifies the domain that there is a pending notification
369 * to be processed. This bit is cleared by the guest.
370 * 2. MASK -- if this bit is clear then a 0->1 transition of PENDING
371 * will cause an asynchronous upcall to be scheduled. This bit is only
372 * updated by the guest. It is read-only within Xen. If a channel
373 * becomes pending while the channel is masked then the 'edge' is lost
374 * (i.e., when the channel is unmasked, the guest must manually handle
375 * pending notifications as no upcall will be scheduled by Xen).
377 * To expedite scanning of pending notifications, any 0->1 pending
378 * transition on an unmasked channel causes a corresponding bit in a
379 * 32-bit selector to be set. Each bit in the selector covers a 32-bit
380 * word in the PENDING bitfield array.
382 u32 evtchn_pending[32]; /* 4 */
383 u32 evtchn_mask[32]; /* 136 */
386 * Time: The following abstractions are exposed: System Time, Clock Time,
387 * Domain Virtual Time. Domains can access Cycle counter time directly.
389 u64 cpu_freq; /* 264: CPU frequency (Hz). */
392 * The following values are updated periodically (and not necessarily
393 * atomically!). The guest OS detects this because 'time_version1' is
394 * incremented just before updating these values, and 'time_version2' is
395 * incremented immediately after. See the Xen-specific Linux code for an
396 * example of how to read these values safely (arch/xen/kernel/time.c).
398 u32 time_version1; /* 272 */
399 u32 time_version2; /* 276 */
400 tsc_timestamp_t tsc_timestamp; /* TSC at last update of time vals. */
401 u64 system_time; /* Time, in nanosecs, since boot. */
402 u32 wc_sec; /* Secs 00:00:00 UTC, Jan 1, 1970. */
403 u32 wc_usec; /* Usecs 00:00:00 UTC, Jan 1, 1970. */
404 u64 domain_time; /* Domain virtual time, in nanosecs. */
408 * Allow a domain to specify a timeout value in system time and
409 * domain virtual time.
411 u64 wall_timeout; /* 312 */
412 u64 domain_timeout; /* 320 */
414 arch_shared_info_t arch;
416 } PACKED shared_info_t;
419 * Start-of-day memory layout for the initial domain (DOM0):
420 * 1. The domain is started within contiguous virtual-memory region.
421 * 2. The contiguous region begins and ends on an aligned 4MB boundary.
422 * 3. The region start corresponds to the load address of the OS image.
423 * If the load address is not 4MB aligned then the address is rounded down.
424 * 4. This the order of bootstrap elements in the initial virtual region:
425 * a. relocated kernel image
426 * b. initial ram disk [mod_start, mod_len]
427 * c. list of allocated page frames [mfn_list, nr_pages]
428 * d. bootstrap page tables [pt_base, CR3 (x86)]
429 * e. start_info_t structure [register ESI (x86)]
430 * f. bootstrap stack [register ESP (x86)]
431 * 5. Bootstrap elements are packed together, but each is 4kB-aligned.
432 * 6. The initial ram disk may be omitted.
433 * 7. The list of page frames forms a contiguous 'pseudo-physical' memory
434 * layout for the domain. In particular, the bootstrap virtual-memory
435 * region is a 1:1 mapping to the first section of the pseudo-physical map.
436 * 8. All bootstrap elements are mapped read-writable for the guest OS. The
437 * only exception is the bootstrap page table, which is mapped read-only.
438 * 9. There is guaranteed to be at least 512kB padding after the final
439 * bootstrap element. If necessary, the bootstrap virtual region is
440 * extended by an extra 4MB to ensure this.
443 #define MAX_CMDLINE 256
445 /* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME. */
446 memory_t nr_pages; /* 0: Total pages allocated to this domain. */
448 memory_t shared_info; /* 8: MACHINE address of shared info struct.*/
450 u32 flags; /* 16: SIF_xxx flags. */
451 u16 domain_controller_evtchn; /* 20 */
453 /* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME). */
454 memory_t pt_base; /* 24: VIRTUAL address of page directory. */
456 memory_t nr_pt_frames; /* 32: Number of bootstrap p.t. frames. */
458 memory_t mfn_list; /* 40: VIRTUAL address of page-frame list. */
460 memory_t mod_start; /* 48: VIRTUAL address of pre-loaded module. */
462 memory_t mod_len; /* 56: Size (bytes) of pre-loaded module. */
464 s8 cmd_line[MAX_CMDLINE]; /* 64 */
465 } PACKED start_info_t; /* 320 bytes */
467 /* These flags are passed in the 'flags' field of start_info_t. */
468 #define SIF_PRIVILEGED (1<<0) /* Is the domain privileged? */
469 #define SIF_INITDOMAIN (1<<1) /* Is this the initial control domain? */
470 #define SIF_BLK_BE_DOMAIN (1<<4) /* Is this a block backend domain? */
471 #define SIF_NET_BE_DOMAIN (1<<5) /* Is this a net backend domain? */
472 #define SIF_USB_BE_DOMAIN (1<<6) /* Is this a usb backend domain? */
473 /* For use in guest OSes. */
474 extern shared_info_t *HYPERVISOR_shared_info;
476 #endif /* !__ASSEMBLY__ */
478 #endif /* __XEN_PUBLIC_XEN_H__ */