blob: 518a5bf79ed34becbf0b67e822421387124da534 [file] [log] [blame]
Jeremy Fitzhardingea42089d2007-07-17 18:37:04 -07001/******************************************************************************
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/*
62 * VIRTUAL INTERRUPTS
63 *
64 * Virtual interrupts that a guest OS may receive from Xen.
65 */
66#define VIRQ_TIMER 0 /* Timebase update, and/or requested timeout. */
67#define VIRQ_DEBUG 1 /* Request guest to dump debug info. */
68#define VIRQ_CONSOLE 2 /* (DOM0) Bytes received on emergency console. */
69#define VIRQ_DOM_EXC 3 /* (DOM0) Exceptional event for some domain. */
70#define VIRQ_DEBUGGER 6 /* (DOM0) A domain has paused for debugging. */
71#define NR_VIRQS 8
72
73/*
74 * MMU-UPDATE REQUESTS
75 *
76 * HYPERVISOR_mmu_update() accepts a list of (ptr, val) pairs.
77 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
78 * Where the FD has some effect, it is described below.
79 * ptr[1:0] specifies the appropriate MMU_* command.
80 *
81 * ptr[1:0] == MMU_NORMAL_PT_UPDATE:
82 * Updates an entry in a page table. If updating an L1 table, and the new
83 * table entry is valid/present, the mapped frame must belong to the FD, if
84 * an FD has been specified. If attempting to map an I/O page then the
85 * caller assumes the privilege of the FD.
86 * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller.
87 * FD == DOMID_XEN: Map restricted areas of Xen's heap space.
88 * ptr[:2] -- Machine address of the page-table entry to modify.
89 * val -- Value to write.
90 *
91 * ptr[1:0] == MMU_MACHPHYS_UPDATE:
92 * Updates an entry in the machine->pseudo-physical mapping table.
93 * ptr[:2] -- Machine address within the frame whose mapping to modify.
94 * The frame must belong to the FD, if one is specified.
95 * val -- Value to write into the mapping entry.
96 */
97#define MMU_NORMAL_PT_UPDATE 0 /* checked '*ptr = val'. ptr is MA. */
98#define MMU_MACHPHYS_UPDATE 1 /* ptr = MA of frame to modify entry for */
99
100/*
101 * MMU EXTENDED OPERATIONS
102 *
103 * HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures.
104 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
105 * Where the FD has some effect, it is described below.
106 *
107 * cmd: MMUEXT_(UN)PIN_*_TABLE
108 * mfn: Machine frame number to be (un)pinned as a p.t. page.
109 * The frame must belong to the FD, if one is specified.
110 *
111 * cmd: MMUEXT_NEW_BASEPTR
112 * mfn: Machine frame number of new page-table base to install in MMU.
113 *
114 * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only]
115 * mfn: Machine frame number of new page-table base to install in MMU
116 * when in user space.
117 *
118 * cmd: MMUEXT_TLB_FLUSH_LOCAL
119 * No additional arguments. Flushes local TLB.
120 *
121 * cmd: MMUEXT_INVLPG_LOCAL
122 * linear_addr: Linear address to be flushed from the local TLB.
123 *
124 * cmd: MMUEXT_TLB_FLUSH_MULTI
125 * vcpumask: Pointer to bitmap of VCPUs to be flushed.
126 *
127 * cmd: MMUEXT_INVLPG_MULTI
128 * linear_addr: Linear address to be flushed.
129 * vcpumask: Pointer to bitmap of VCPUs to be flushed.
130 *
131 * cmd: MMUEXT_TLB_FLUSH_ALL
132 * No additional arguments. Flushes all VCPUs' TLBs.
133 *
134 * cmd: MMUEXT_INVLPG_ALL
135 * linear_addr: Linear address to be flushed from all VCPUs' TLBs.
136 *
137 * cmd: MMUEXT_FLUSH_CACHE
138 * No additional arguments. Writes back and flushes cache contents.
139 *
140 * cmd: MMUEXT_SET_LDT
141 * linear_addr: Linear address of LDT base (NB. must be page-aligned).
142 * nr_ents: Number of entries in LDT.
143 */
144#define MMUEXT_PIN_L1_TABLE 0
145#define MMUEXT_PIN_L2_TABLE 1
146#define MMUEXT_PIN_L3_TABLE 2
147#define MMUEXT_PIN_L4_TABLE 3
148#define MMUEXT_UNPIN_TABLE 4
149#define MMUEXT_NEW_BASEPTR 5
150#define MMUEXT_TLB_FLUSH_LOCAL 6
151#define MMUEXT_INVLPG_LOCAL 7
152#define MMUEXT_TLB_FLUSH_MULTI 8
153#define MMUEXT_INVLPG_MULTI 9
154#define MMUEXT_TLB_FLUSH_ALL 10
155#define MMUEXT_INVLPG_ALL 11
156#define MMUEXT_FLUSH_CACHE 12
157#define MMUEXT_SET_LDT 13
158#define MMUEXT_NEW_USER_BASEPTR 15
159
160#ifndef __ASSEMBLY__
161struct mmuext_op {
162 unsigned int cmd;
163 union {
164 /* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR */
165 unsigned long mfn;
166 /* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */
167 unsigned long linear_addr;
168 } arg1;
169 union {
170 /* SET_LDT */
171 unsigned int nr_ents;
172 /* TLB_FLUSH_MULTI, INVLPG_MULTI */
173 void *vcpumask;
174 } arg2;
175};
176DEFINE_GUEST_HANDLE_STRUCT(mmuext_op);
177#endif
178
179/* These are passed as 'flags' to update_va_mapping. They can be ORed. */
180/* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap. */
181/* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer. */
182#define UVMF_NONE (0UL<<0) /* No flushing at all. */
183#define UVMF_TLB_FLUSH (1UL<<0) /* Flush entire TLB(s). */
184#define UVMF_INVLPG (2UL<<0) /* Flush only one entry. */
185#define UVMF_FLUSHTYPE_MASK (3UL<<0)
186#define UVMF_MULTI (0UL<<2) /* Flush subset of TLBs. */
187#define UVMF_LOCAL (0UL<<2) /* Flush local TLB. */
188#define UVMF_ALL (1UL<<2) /* Flush all TLBs. */
189
190/*
191 * Commands to HYPERVISOR_console_io().
192 */
193#define CONSOLEIO_write 0
194#define CONSOLEIO_read 1
195
196/*
197 * Commands to HYPERVISOR_vm_assist().
198 */
199#define VMASST_CMD_enable 0
200#define VMASST_CMD_disable 1
201#define VMASST_TYPE_4gb_segments 0
202#define VMASST_TYPE_4gb_segments_notify 1
203#define VMASST_TYPE_writable_pagetables 2
204#define VMASST_TYPE_pae_extended_cr3 3
205#define MAX_VMASST_TYPE 3
206
207#ifndef __ASSEMBLY__
208
209typedef uint16_t domid_t;
210
211/* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */
212#define DOMID_FIRST_RESERVED (0x7FF0U)
213
214/* DOMID_SELF is used in certain contexts to refer to oneself. */
215#define DOMID_SELF (0x7FF0U)
216
217/*
218 * DOMID_IO is used to restrict page-table updates to mapping I/O memory.
219 * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO
220 * is useful to ensure that no mappings to the OS's own heap are accidentally
221 * installed. (e.g., in Linux this could cause havoc as reference counts
222 * aren't adjusted on the I/O-mapping code path).
223 * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can
224 * be specified by any calling domain.
225 */
226#define DOMID_IO (0x7FF1U)
227
228/*
229 * DOMID_XEN is used to allow privileged domains to map restricted parts of
230 * Xen's heap space (e.g., the machine_to_phys table).
231 * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if
232 * the caller is privileged.
233 */
234#define DOMID_XEN (0x7FF2U)
235
236/*
237 * Send an array of these to HYPERVISOR_mmu_update().
238 * NB. The fields are natural pointer/address size for this architecture.
239 */
240struct mmu_update {
241 uint64_t ptr; /* Machine address of PTE. */
242 uint64_t val; /* New contents of PTE. */
243};
244DEFINE_GUEST_HANDLE_STRUCT(mmu_update);
245
246/*
247 * Send an array of these to HYPERVISOR_multicall().
248 * NB. The fields are natural register size for this architecture.
249 */
250struct multicall_entry {
251 unsigned long op;
252 long result;
253 unsigned long args[6];
254};
255DEFINE_GUEST_HANDLE_STRUCT(multicall_entry);
256
257/*
258 * Event channel endpoints per domain:
259 * 1024 if a long is 32 bits; 4096 if a long is 64 bits.
260 */
261#define NR_EVENT_CHANNELS (sizeof(unsigned long) * sizeof(unsigned long) * 64)
262
263struct vcpu_time_info {
264 /*
265 * Updates to the following values are preceded and followed
266 * by an increment of 'version'. The guest can therefore
267 * detect updates by looking for changes to 'version'. If the
268 * least-significant bit of the version number is set then an
269 * update is in progress and the guest must wait to read a
270 * consistent set of values. The correct way to interact with
271 * the version number is similar to Linux's seqlock: see the
272 * implementations of read_seqbegin/read_seqretry.
273 */
274 uint32_t version;
275 uint32_t pad0;
276 uint64_t tsc_timestamp; /* TSC at last update of time vals. */
277 uint64_t system_time; /* Time, in nanosecs, since boot. */
278 /*
279 * Current system time:
280 * system_time + ((tsc - tsc_timestamp) << tsc_shift) * tsc_to_system_mul
281 * CPU frequency (Hz):
282 * ((10^9 << 32) / tsc_to_system_mul) >> tsc_shift
283 */
284 uint32_t tsc_to_system_mul;
285 int8_t tsc_shift;
286 int8_t pad1[3];
287}; /* 32 bytes */
288
289struct vcpu_info {
290 /*
291 * 'evtchn_upcall_pending' is written non-zero by Xen to indicate
292 * a pending notification for a particular VCPU. It is then cleared
293 * by the guest OS /before/ checking for pending work, thus avoiding
294 * a set-and-check race. Note that the mask is only accessed by Xen
295 * on the CPU that is currently hosting the VCPU. This means that the
296 * pending and mask flags can be updated by the guest without special
297 * synchronisation (i.e., no need for the x86 LOCK prefix).
298 * This may seem suboptimal because if the pending flag is set by
299 * a different CPU then an IPI may be scheduled even when the mask
300 * is set. However, note:
301 * 1. The task of 'interrupt holdoff' is covered by the per-event-
302 * channel mask bits. A 'noisy' event that is continually being
303 * triggered can be masked at source at this very precise
304 * granularity.
305 * 2. The main purpose of the per-VCPU mask is therefore to restrict
306 * reentrant execution: whether for concurrency control, or to
307 * prevent unbounded stack usage. Whatever the purpose, we expect
308 * that the mask will be asserted only for short periods at a time,
309 * and so the likelihood of a 'spurious' IPI is suitably small.
310 * The mask is read before making an event upcall to the guest: a
311 * non-zero mask therefore guarantees that the VCPU will not receive
312 * an upcall activation. The mask is cleared when the VCPU requests
313 * to block: this avoids wakeup-waiting races.
314 */
315 uint8_t evtchn_upcall_pending;
316 uint8_t evtchn_upcall_mask;
317 unsigned long evtchn_pending_sel;
318 struct arch_vcpu_info arch;
319 struct vcpu_time_info time;
320}; /* 64 bytes (x86) */
321
322/*
323 * Xen/kernel shared data -- pointer provided in start_info.
324 * NB. We expect that this struct is smaller than a page.
325 */
326struct shared_info {
327 struct vcpu_info vcpu_info[MAX_VIRT_CPUS];
328
329 /*
330 * A domain can create "event channels" on which it can send and receive
331 * asynchronous event notifications. There are three classes of event that
332 * are delivered by this mechanism:
333 * 1. Bi-directional inter- and intra-domain connections. Domains must
334 * arrange out-of-band to set up a connection (usually by allocating
335 * an unbound 'listener' port and avertising that via a storage service
336 * such as xenstore).
337 * 2. Physical interrupts. A domain with suitable hardware-access
338 * privileges can bind an event-channel port to a physical interrupt
339 * source.
340 * 3. Virtual interrupts ('events'). A domain can bind an event-channel
341 * port to a virtual interrupt source, such as the virtual-timer
342 * device or the emergency console.
343 *
344 * Event channels are addressed by a "port index". Each channel is
345 * associated with two bits of information:
346 * 1. PENDING -- notifies the domain that there is a pending notification
347 * to be processed. This bit is cleared by the guest.
348 * 2. MASK -- if this bit is clear then a 0->1 transition of PENDING
349 * will cause an asynchronous upcall to be scheduled. This bit is only
350 * updated by the guest. It is read-only within Xen. If a channel
351 * becomes pending while the channel is masked then the 'edge' is lost
352 * (i.e., when the channel is unmasked, the guest must manually handle
353 * pending notifications as no upcall will be scheduled by Xen).
354 *
355 * To expedite scanning of pending notifications, any 0->1 pending
356 * transition on an unmasked channel causes a corresponding bit in a
357 * per-vcpu selector word to be set. Each bit in the selector covers a
358 * 'C long' in the PENDING bitfield array.
359 */
360 unsigned long evtchn_pending[sizeof(unsigned long) * 8];
361 unsigned long evtchn_mask[sizeof(unsigned long) * 8];
362
363 /*
364 * Wallclock time: updated only by control software. Guests should base
365 * their gettimeofday() syscall on this wallclock-base value.
366 */
367 uint32_t wc_version; /* Version counter: see vcpu_time_info_t. */
368 uint32_t wc_sec; /* Secs 00:00:00 UTC, Jan 1, 1970. */
369 uint32_t wc_nsec; /* Nsecs 00:00:00 UTC, Jan 1, 1970. */
370
371 struct arch_shared_info arch;
372
373};
374
375/*
376 * Start-of-day memory layout for the initial domain (DOM0):
377 * 1. The domain is started within contiguous virtual-memory region.
378 * 2. The contiguous region begins and ends on an aligned 4MB boundary.
379 * 3. The region start corresponds to the load address of the OS image.
380 * If the load address is not 4MB aligned then the address is rounded down.
381 * 4. This the order of bootstrap elements in the initial virtual region:
382 * a. relocated kernel image
383 * b. initial ram disk [mod_start, mod_len]
384 * c. list of allocated page frames [mfn_list, nr_pages]
385 * d. start_info_t structure [register ESI (x86)]
386 * e. bootstrap page tables [pt_base, CR3 (x86)]
387 * f. bootstrap stack [register ESP (x86)]
388 * 5. Bootstrap elements are packed together, but each is 4kB-aligned.
389 * 6. The initial ram disk may be omitted.
390 * 7. The list of page frames forms a contiguous 'pseudo-physical' memory
391 * layout for the domain. In particular, the bootstrap virtual-memory
392 * region is a 1:1 mapping to the first section of the pseudo-physical map.
393 * 8. All bootstrap elements are mapped read-writable for the guest OS. The
394 * only exception is the bootstrap page table, which is mapped read-only.
395 * 9. There is guaranteed to be at least 512kB padding after the final
396 * bootstrap element. If necessary, the bootstrap virtual region is
397 * extended by an extra 4MB to ensure this.
398 */
399
400#define MAX_GUEST_CMDLINE 1024
401struct start_info {
402 /* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME. */
403 char magic[32]; /* "xen-<version>-<platform>". */
404 unsigned long nr_pages; /* Total pages allocated to this domain. */
405 unsigned long shared_info; /* MACHINE address of shared info struct. */
406 uint32_t flags; /* SIF_xxx flags. */
407 unsigned long store_mfn; /* MACHINE page number of shared page. */
408 uint32_t store_evtchn; /* Event channel for store communication. */
409 union {
410 struct {
411 unsigned long mfn; /* MACHINE page number of console page. */
412 uint32_t evtchn; /* Event channel for console page. */
413 } domU;
414 struct {
415 uint32_t info_off; /* Offset of console_info struct. */
416 uint32_t info_size; /* Size of console_info struct from start.*/
417 } dom0;
418 } console;
419 /* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME). */
420 unsigned long pt_base; /* VIRTUAL address of page directory. */
421 unsigned long nr_pt_frames; /* Number of bootstrap p.t. frames. */
422 unsigned long mfn_list; /* VIRTUAL address of page-frame list. */
423 unsigned long mod_start; /* VIRTUAL address of pre-loaded module. */
424 unsigned long mod_len; /* Size (bytes) of pre-loaded module. */
425 int8_t cmd_line[MAX_GUEST_CMDLINE];
426};
427
428/* These flags are passed in the 'flags' field of start_info_t. */
429#define SIF_PRIVILEGED (1<<0) /* Is the domain privileged? */
430#define SIF_INITDOMAIN (1<<1) /* Is this the initial control domain? */
431
432typedef uint64_t cpumap_t;
433
434typedef uint8_t xen_domain_handle_t[16];
435
436/* Turn a plain number into a C unsigned long constant. */
437#define __mk_unsigned_long(x) x ## UL
438#define mk_unsigned_long(x) __mk_unsigned_long(x)
439
440#else /* __ASSEMBLY__ */
441
442/* In assembly code we cannot use C numeric constant suffixes. */
443#define mk_unsigned_long(x) x
444
445#endif /* !__ASSEMBLY__ */
446
447#endif /* __XEN_PUBLIC_XEN_H__ */