ia64/xen-unstable

view xen/include/public/xen.h @ 9597:8f7aad20b4a5

Backtrack on the new interface for reserved event-channel
ports, as binding them in user space via the evtchn driver
would be a pain. Instead extend VIRQs so they can be
classified as 'global' or 'per vcpu'. The former can only
be allocated once per guest, but can be re-bound to
an arbitrary VCPU.

Signed-off-by: Keir Fraser <keir@xensource.com>
author kaf24@firebug.cl.cam.ac.uk
date Wed Apr 05 19:30:02 2006 +0100 (2006-04-05)
parents 9316fe0c9c4f
children 67de34c062b5
line source
1 /******************************************************************************
2 * xen.h
3 *
4 * Guest OS interface to Xen.
5 *
6 * Copyright (c) 2004, K A Fraser
7 */
9 #ifndef __XEN_PUBLIC_XEN_H__
10 #define __XEN_PUBLIC_XEN_H__
12 #if defined(__i386__)
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 #else
19 #error "Unsupported architecture"
20 #endif
22 /*
23 * XEN "SYSTEM CALLS" (a.k.a. HYPERCALLS).
24 */
26 /*
27 * x86_32: EAX = vector; EBX, ECX, EDX, ESI, EDI = args 1, 2, 3, 4, 5.
28 * EAX = return value
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.
31 * RAX = return value
32 * (argument registers not clobbered on return; RCX, R11 are)
33 */
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_compat 6 /* compat as of 0x00030101 */
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_memory_op 12
46 #define __HYPERVISOR_multicall 13
47 #define __HYPERVISOR_update_va_mapping 14
48 #define __HYPERVISOR_set_timer_op 15
49 #define __HYPERVISOR_event_channel_op 16
50 #define __HYPERVISOR_xen_version 17
51 #define __HYPERVISOR_console_io 18
52 #define __HYPERVISOR_physdev_op 19
53 #define __HYPERVISOR_grant_table_op 20
54 #define __HYPERVISOR_vm_assist 21
55 #define __HYPERVISOR_update_va_mapping_otherdomain 22
56 #define __HYPERVISOR_iret 23 /* x86 only */
57 #define __HYPERVISOR_vcpu_op 24
58 #define __HYPERVISOR_set_segment_base 25 /* x86/64 only */
59 #define __HYPERVISOR_mmuext_op 26
60 #define __HYPERVISOR_acm_op 27
61 #define __HYPERVISOR_nmi_op 28
62 #define __HYPERVISOR_sched_op 29
64 /*
65 * VIRTUAL INTERRUPTS
66 *
67 * Virtual interrupts that a guest OS may receive from Xen.
68 *
69 * In the side comments, 'V.' denotes a per-VCPU VIRQ while 'G.' denotes a
70 * global VIRQ. The former can be bound once per VCPU and cannot be re-bound.
71 * The latter can be allocated only once per guest: they must initially be
72 * allocated to VCPU0 but can subsequently be re-bound.
73 */
74 #define VIRQ_TIMER 0 /* V. Timebase update, and/or requested timeout. */
75 #define VIRQ_DEBUG 1 /* V. Request guest to dump debug info. */
76 #define VIRQ_CONSOLE 2 /* G. (DOM0) Bytes received on emergency console. */
77 #define VIRQ_DOM_EXC 3 /* G. (DOM0) Exceptional event for some domain. */
78 #define VIRQ_DEBUGGER 6 /* G. (DOM0) A domain has paused for debugging. */
79 #define NR_VIRQS 8
81 /*
82 * MMU-UPDATE REQUESTS
83 *
84 * HYPERVISOR_mmu_update() accepts a list of (ptr, val) pairs.
85 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
86 * Where the FD has some effect, it is described below.
87 * ptr[1:0] specifies the appropriate MMU_* command.
88 *
89 * ptr[1:0] == MMU_NORMAL_PT_UPDATE:
90 * Updates an entry in a page table. If updating an L1 table, and the new
91 * table entry is valid/present, the mapped frame must belong to the FD, if
92 * an FD has been specified. If attempting to map an I/O page then the
93 * caller assumes the privilege of the FD.
94 * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller.
95 * FD == DOMID_XEN: Map restricted areas of Xen's heap space.
96 * ptr[:2] -- Machine address of the page-table entry to modify.
97 * val -- Value to write.
98 *
99 * ptr[1:0] == MMU_MACHPHYS_UPDATE:
100 * Updates an entry in the machine->pseudo-physical mapping table.
101 * ptr[:2] -- Machine address within the frame whose mapping to modify.
102 * The frame must belong to the FD, if one is specified.
103 * val -- Value to write into the mapping entry.
104 */
105 #define MMU_NORMAL_PT_UPDATE 0 /* checked '*ptr = val'. ptr is MA. */
106 #define MMU_MACHPHYS_UPDATE 1 /* ptr = MA of frame to modify entry for */
108 /*
109 * MMU EXTENDED OPERATIONS
110 *
111 * HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures.
112 * A foreigndom (FD) can be specified (or DOMID_SELF for none).
113 * Where the FD has some effect, it is described below.
114 *
115 * cmd: MMUEXT_(UN)PIN_*_TABLE
116 * mfn: Machine frame number to be (un)pinned as a p.t. page.
117 * The frame must belong to the FD, if one is specified.
118 *
119 * cmd: MMUEXT_NEW_BASEPTR
120 * mfn: Machine frame number of new page-table base to install in MMU.
121 *
122 * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only]
123 * mfn: Machine frame number of new page-table base to install in MMU
124 * when in user space.
125 *
126 * cmd: MMUEXT_TLB_FLUSH_LOCAL
127 * No additional arguments. Flushes local TLB.
128 *
129 * cmd: MMUEXT_INVLPG_LOCAL
130 * linear_addr: Linear address to be flushed from the local TLB.
131 *
132 * cmd: MMUEXT_TLB_FLUSH_MULTI
133 * vcpumask: Pointer to bitmap of VCPUs to be flushed.
134 *
135 * cmd: MMUEXT_INVLPG_MULTI
136 * linear_addr: Linear address to be flushed.
137 * vcpumask: Pointer to bitmap of VCPUs to be flushed.
138 *
139 * cmd: MMUEXT_TLB_FLUSH_ALL
140 * No additional arguments. Flushes all VCPUs' TLBs.
141 *
142 * cmd: MMUEXT_INVLPG_ALL
143 * linear_addr: Linear address to be flushed from all VCPUs' TLBs.
144 *
145 * cmd: MMUEXT_FLUSH_CACHE
146 * No additional arguments. Writes back and flushes cache contents.
147 *
148 * cmd: MMUEXT_SET_LDT
149 * linear_addr: Linear address of LDT base (NB. must be page-aligned).
150 * nr_ents: Number of entries in LDT.
151 */
152 #define MMUEXT_PIN_L1_TABLE 0
153 #define MMUEXT_PIN_L2_TABLE 1
154 #define MMUEXT_PIN_L3_TABLE 2
155 #define MMUEXT_PIN_L4_TABLE 3
156 #define MMUEXT_UNPIN_TABLE 4
157 #define MMUEXT_NEW_BASEPTR 5
158 #define MMUEXT_TLB_FLUSH_LOCAL 6
159 #define MMUEXT_INVLPG_LOCAL 7
160 #define MMUEXT_TLB_FLUSH_MULTI 8
161 #define MMUEXT_INVLPG_MULTI 9
162 #define MMUEXT_TLB_FLUSH_ALL 10
163 #define MMUEXT_INVLPG_ALL 11
164 #define MMUEXT_FLUSH_CACHE 12
165 #define MMUEXT_SET_LDT 13
166 #define MMUEXT_NEW_USER_BASEPTR 15
168 #ifndef __ASSEMBLY__
169 typedef struct mmuext_op {
170 unsigned int cmd;
171 union {
172 /* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR */
173 unsigned long mfn;
174 /* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */
175 unsigned long linear_addr;
176 } arg1;
177 union {
178 /* SET_LDT */
179 unsigned int nr_ents;
180 /* TLB_FLUSH_MULTI, INVLPG_MULTI */
181 void *vcpumask;
182 } arg2;
183 } mmuext_op_t;
184 DEFINE_GUEST_HANDLE(mmuext_op_t);
185 #endif
187 /* These are passed as 'flags' to update_va_mapping. They can be ORed. */
188 /* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap. */
189 /* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer. */
190 #define UVMF_NONE (0UL<<0) /* No flushing at all. */
191 #define UVMF_TLB_FLUSH (1UL<<0) /* Flush entire TLB(s). */
192 #define UVMF_INVLPG (2UL<<0) /* Flush only one entry. */
193 #define UVMF_FLUSHTYPE_MASK (3UL<<0)
194 #define UVMF_MULTI (0UL<<2) /* Flush subset of TLBs. */
195 #define UVMF_LOCAL (0UL<<2) /* Flush local TLB. */
196 #define UVMF_ALL (1UL<<2) /* Flush all TLBs. */
198 /*
199 * Commands to HYPERVISOR_console_io().
200 */
201 #define CONSOLEIO_write 0
202 #define CONSOLEIO_read 1
204 /*
205 * Commands to HYPERVISOR_vm_assist().
206 */
207 #define VMASST_CMD_enable 0
208 #define VMASST_CMD_disable 1
209 #define VMASST_TYPE_4gb_segments 0
210 #define VMASST_TYPE_4gb_segments_notify 1
211 #define VMASST_TYPE_writable_pagetables 2
212 #define MAX_VMASST_TYPE 2
214 #ifndef __ASSEMBLY__
216 typedef uint16_t domid_t;
218 /* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */
219 #define DOMID_FIRST_RESERVED (0x7FF0U)
221 /* DOMID_SELF is used in certain contexts to refer to oneself. */
222 #define DOMID_SELF (0x7FF0U)
224 /*
225 * DOMID_IO is used to restrict page-table updates to mapping I/O memory.
226 * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO
227 * is useful to ensure that no mappings to the OS's own heap are accidentally
228 * installed. (e.g., in Linux this could cause havoc as reference counts
229 * aren't adjusted on the I/O-mapping code path).
230 * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can
231 * be specified by any calling domain.
232 */
233 #define DOMID_IO (0x7FF1U)
235 /*
236 * DOMID_XEN is used to allow privileged domains to map restricted parts of
237 * Xen's heap space (e.g., the machine_to_phys table).
238 * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if
239 * the caller is privileged.
240 */
241 #define DOMID_XEN (0x7FF2U)
243 /*
244 * Send an array of these to HYPERVISOR_mmu_update().
245 * NB. The fields are natural pointer/address size for this architecture.
246 */
247 typedef struct mmu_update {
248 uint64_t ptr; /* Machine address of PTE. */
249 uint64_t val; /* New contents of PTE. */
250 } mmu_update_t;
251 DEFINE_GUEST_HANDLE(mmu_update_t);
253 /*
254 * Send an array of these to HYPERVISOR_multicall().
255 * NB. The fields are natural register size for this architecture.
256 */
257 typedef struct multicall_entry {
258 unsigned long op, result;
259 unsigned long args[6];
260 } multicall_entry_t;
261 DEFINE_GUEST_HANDLE(multicall_entry_t);
263 /*
264 * Event channel endpoints per domain:
265 * 1024 if a long is 32 bits; 4096 if a long is 64 bits.
266 */
267 #define NR_EVENT_CHANNELS (sizeof(unsigned long) * sizeof(unsigned long) * 64)
269 typedef struct vcpu_time_info {
270 /*
271 * Updates to the following values are preceded and followed by an
272 * increment of 'version'. The guest can therefore detect updates by
273 * looking for changes to 'version'. If the least-significant bit of
274 * the version number is set then an update is in progress and the guest
275 * must wait to read a consistent set of values.
276 * The correct way to interact with the version number is similar to
277 * Linux's seqlock: see the implementations of read_seqbegin/read_seqretry.
278 */
279 uint32_t version;
280 uint32_t pad0;
281 uint64_t tsc_timestamp; /* TSC at last update of time vals. */
282 uint64_t system_time; /* Time, in nanosecs, since boot. */
283 /*
284 * Current system time:
285 * system_time + ((tsc - tsc_timestamp) << tsc_shift) * tsc_to_system_mul
286 * CPU frequency (Hz):
287 * ((10^9 << 32) / tsc_to_system_mul) >> tsc_shift
288 */
289 uint32_t tsc_to_system_mul;
290 int8_t tsc_shift;
291 int8_t pad1[3];
292 } vcpu_time_info_t; /* 32 bytes */
294 typedef struct vcpu_info {
295 /*
296 * 'evtchn_upcall_pending' is written non-zero by Xen to indicate
297 * a pending notification for a particular VCPU. It is then cleared
298 * by the guest OS /before/ checking for pending work, thus avoiding
299 * a set-and-check race. Note that the mask is only accessed by Xen
300 * on the CPU that is currently hosting the VCPU. This means that the
301 * pending and mask flags can be updated by the guest without special
302 * synchronisation (i.e., no need for the x86 LOCK prefix).
303 * This may seem suboptimal because if the pending flag is set by
304 * a different CPU then an IPI may be scheduled even when the mask
305 * is set. However, note:
306 * 1. The task of 'interrupt holdoff' is covered by the per-event-
307 * channel mask bits. A 'noisy' event that is continually being
308 * triggered can be masked at source at this very precise
309 * granularity.
310 * 2. The main purpose of the per-VCPU mask is therefore to restrict
311 * reentrant execution: whether for concurrency control, or to
312 * prevent unbounded stack usage. Whatever the purpose, we expect
313 * that the mask will be asserted only for short periods at a time,
314 * and so the likelihood of a 'spurious' IPI is suitably small.
315 * The mask is read before making an event upcall to the guest: a
316 * non-zero mask therefore guarantees that the VCPU will not receive
317 * an upcall activation. The mask is cleared when the VCPU requests
318 * to block: this avoids wakeup-waiting races.
319 */
320 uint8_t evtchn_upcall_pending;
321 uint8_t evtchn_upcall_mask;
322 unsigned long evtchn_pending_sel;
323 arch_vcpu_info_t arch;
324 vcpu_time_info_t time;
325 } vcpu_info_t; /* 64 bytes (x86) */
327 /*
328 * Xen/kernel shared data -- pointer provided in start_info.
329 * NB. We expect that this struct is smaller than a page.
330 */
331 typedef struct shared_info {
332 vcpu_info_t vcpu_info[MAX_VIRT_CPUS];
334 /*
335 * A domain can create "event channels" on which it can send and receive
336 * asynchronous event notifications. There are three classes of event that
337 * are delivered by this mechanism:
338 * 1. Bi-directional inter- and intra-domain connections. Domains must
339 * arrange out-of-band to set up a connection (usually by allocating
340 * an unbound 'listener' port and avertising that via a storage service
341 * such as xenstore).
342 * 2. Physical interrupts. A domain with suitable hardware-access
343 * privileges can bind an event-channel port to a physical interrupt
344 * source.
345 * 3. Virtual interrupts ('events'). A domain can bind an event-channel
346 * port to a virtual interrupt source, such as the virtual-timer
347 * device or the emergency console.
348 *
349 * Event channels are addressed by a "port index". Each channel is
350 * associated with two bits of information:
351 * 1. PENDING -- notifies the domain that there is a pending notification
352 * to be processed. This bit is cleared by the guest.
353 * 2. MASK -- if this bit is clear then a 0->1 transition of PENDING
354 * will cause an asynchronous upcall to be scheduled. This bit is only
355 * updated by the guest. It is read-only within Xen. If a channel
356 * becomes pending while the channel is masked then the 'edge' is lost
357 * (i.e., when the channel is unmasked, the guest must manually handle
358 * pending notifications as no upcall will be scheduled by Xen).
359 *
360 * To expedite scanning of pending notifications, any 0->1 pending
361 * transition on an unmasked channel causes a corresponding bit in a
362 * per-vcpu selector word to be set. Each bit in the selector covers a
363 * 'C long' in the PENDING bitfield array.
364 */
365 unsigned long evtchn_pending[sizeof(unsigned long) * 8];
366 unsigned long evtchn_mask[sizeof(unsigned long) * 8];
368 /*
369 * Wallclock time: updated only by control software. Guests should base
370 * their gettimeofday() syscall on this wallclock-base value.
371 */
372 uint32_t wc_version; /* Version counter: see vcpu_time_info_t. */
373 uint32_t wc_sec; /* Secs 00:00:00 UTC, Jan 1, 1970. */
374 uint32_t wc_nsec; /* Nsecs 00:00:00 UTC, Jan 1, 1970. */
376 arch_shared_info_t arch;
378 } shared_info_t;
380 /*
381 * Start-of-day memory layout for the initial domain (DOM0):
382 * 1. The domain is started within contiguous virtual-memory region.
383 * 2. The contiguous region begins and ends on an aligned 4MB boundary.
384 * 3. The region start corresponds to the load address of the OS image.
385 * If the load address is not 4MB aligned then the address is rounded down.
386 * 4. This the order of bootstrap elements in the initial virtual region:
387 * a. relocated kernel image
388 * b. initial ram disk [mod_start, mod_len]
389 * c. list of allocated page frames [mfn_list, nr_pages]
390 * d. start_info_t structure [register ESI (x86)]
391 * e. bootstrap page tables [pt_base, CR3 (x86)]
392 * f. bootstrap stack [register ESP (x86)]
393 * 5. Bootstrap elements are packed together, but each is 4kB-aligned.
394 * 6. The initial ram disk may be omitted.
395 * 7. The list of page frames forms a contiguous 'pseudo-physical' memory
396 * layout for the domain. In particular, the bootstrap virtual-memory
397 * region is a 1:1 mapping to the first section of the pseudo-physical map.
398 * 8. All bootstrap elements are mapped read-writable for the guest OS. The
399 * only exception is the bootstrap page table, which is mapped read-only.
400 * 9. There is guaranteed to be at least 512kB padding after the final
401 * bootstrap element. If necessary, the bootstrap virtual region is
402 * extended by an extra 4MB to ensure this.
403 */
405 #define MAX_GUEST_CMDLINE 1024
406 typedef struct start_info {
407 /* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME. */
408 char magic[32]; /* "xen-<version>-<platform>". */
409 unsigned long nr_pages; /* Total pages allocated to this domain. */
410 unsigned long shared_info; /* MACHINE address of shared info struct. */
411 uint32_t flags; /* SIF_xxx flags. */
412 unsigned long store_mfn; /* MACHINE page number of shared page. */
413 uint32_t store_evtchn; /* Event channel for store communication. */
414 unsigned long console_mfn; /* MACHINE address of console page. */
415 uint32_t console_evtchn; /* Event channel for console messages. */
416 /* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME). */
417 unsigned long pt_base; /* VIRTUAL address of page directory. */
418 unsigned long nr_pt_frames; /* Number of bootstrap p.t. frames. */
419 unsigned long mfn_list; /* VIRTUAL address of page-frame list. */
420 unsigned long mod_start; /* VIRTUAL address of pre-loaded module. */
421 unsigned long mod_len; /* Size (bytes) of pre-loaded module. */
422 int8_t cmd_line[MAX_GUEST_CMDLINE];
423 } start_info_t;
425 /* These flags are passed in the 'flags' field of start_info_t. */
426 #define SIF_PRIVILEGED (1<<0) /* Is the domain privileged? */
427 #define SIF_INITDOMAIN (1<<1) /* Is this the initial control domain? */
429 typedef uint64_t cpumap_t;
431 typedef uint8_t xen_domain_handle_t[16];
433 /* Turn a plain number into a C unsigned long constant. */
434 #define __mk_unsigned_long(x) x ## UL
435 #define mk_unsigned_long(x) __mk_unsigned_long(x)
437 #else /* __ASSEMBLY__ */
439 /* In assembly code we cannot use C numeric constant suffixes. */
440 #define mk_unsigned_long(x) x
442 #endif /* !__ASSEMBLY__ */
444 #include "xen-compat.h"
446 #endif /* __XEN_PUBLIC_XEN_H__ */
448 /*
449 * Local variables:
450 * mode: C
451 * c-set-style: "BSD"
452 * c-basic-offset: 4
453 * tab-width: 4
454 * indent-tabs-mode: nil
455 * End:
456 */