xref: /linux/include/xen/interface/xen.h (revision 1b62f19c3c0722d9b3ac20f41225f582f2277d81)
1 /******************************************************************************
2  * xen.h
3  *
4  * Guest OS interface to Xen.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to
8  * deal in the Software without restriction, including without limitation the
9  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10  * sell copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22  * DEALINGS IN THE SOFTWARE.
23  *
24  * Copyright (c) 2004, K A Fraser
25  */
26 
27 #ifndef __XEN_PUBLIC_XEN_H__
28 #define __XEN_PUBLIC_XEN_H__
29 
30 #include <asm/xen/interface.h>
31 
32 /*
33  * XEN "SYSTEM CALLS" (a.k.a. HYPERCALLS).
34  */
35 
36 /*
37  * x86_32: EAX = vector; EBX, ECX, EDX, ESI, EDI = args 1, 2, 3, 4, 5.
38  *         EAX = return value
39  *         (argument registers may be clobbered on return)
40  * x86_64: RAX = vector; RDI, RSI, RDX, R10, R8, R9 = args 1, 2, 3, 4, 5, 6.
41  *         RAX = return value
42  *         (argument registers not clobbered on return; RCX, R11 are)
43  */
44 #define __HYPERVISOR_set_trap_table        0
45 #define __HYPERVISOR_mmu_update            1
46 #define __HYPERVISOR_set_gdt               2
47 #define __HYPERVISOR_stack_switch          3
48 #define __HYPERVISOR_set_callbacks         4
49 #define __HYPERVISOR_fpu_taskswitch        5
50 #define __HYPERVISOR_sched_op_compat       6
51 #define __HYPERVISOR_dom0_op               7
52 #define __HYPERVISOR_set_debugreg          8
53 #define __HYPERVISOR_get_debugreg          9
54 #define __HYPERVISOR_update_descriptor    10
55 #define __HYPERVISOR_memory_op            12
56 #define __HYPERVISOR_multicall            13
57 #define __HYPERVISOR_update_va_mapping    14
58 #define __HYPERVISOR_set_timer_op         15
59 #define __HYPERVISOR_event_channel_op_compat 16
60 #define __HYPERVISOR_xen_version          17
61 #define __HYPERVISOR_console_io           18
62 #define __HYPERVISOR_physdev_op_compat    19
63 #define __HYPERVISOR_grant_table_op       20
64 #define __HYPERVISOR_vm_assist            21
65 #define __HYPERVISOR_update_va_mapping_otherdomain 22
66 #define __HYPERVISOR_iret                 23 /* x86 only */
67 #define __HYPERVISOR_vcpu_op              24
68 #define __HYPERVISOR_set_segment_base     25 /* x86/64 only */
69 #define __HYPERVISOR_mmuext_op            26
70 #define __HYPERVISOR_acm_op               27
71 #define __HYPERVISOR_nmi_op               28
72 #define __HYPERVISOR_sched_op             29
73 #define __HYPERVISOR_callback_op          30
74 #define __HYPERVISOR_xenoprof_op          31
75 #define __HYPERVISOR_event_channel_op     32
76 #define __HYPERVISOR_physdev_op           33
77 #define __HYPERVISOR_hvm_op               34
78 #define __HYPERVISOR_tmem_op              38
79 
80 /* Architecture-specific hypercall definitions. */
81 #define __HYPERVISOR_arch_0               48
82 #define __HYPERVISOR_arch_1               49
83 #define __HYPERVISOR_arch_2               50
84 #define __HYPERVISOR_arch_3               51
85 #define __HYPERVISOR_arch_4               52
86 #define __HYPERVISOR_arch_5               53
87 #define __HYPERVISOR_arch_6               54
88 #define __HYPERVISOR_arch_7               55
89 
90 /*
91  * VIRTUAL INTERRUPTS
92  *
93  * Virtual interrupts that a guest OS may receive from Xen.
94  * In the side comments, 'V.' denotes a per-VCPU VIRQ while 'G.' denotes a
95  * global VIRQ. The former can be bound once per VCPU and cannot be re-bound.
96  * The latter can be allocated only once per guest: they must initially be
97  * allocated to VCPU0 but can subsequently be re-bound.
98  */
99 #define VIRQ_TIMER      0  /* V. Timebase update, and/or requested timeout.  */
100 #define VIRQ_DEBUG      1  /* V. Request guest to dump debug info.           */
101 #define VIRQ_CONSOLE    2  /* G. (DOM0) Bytes received on emergency console. */
102 #define VIRQ_DOM_EXC    3  /* G. (DOM0) Exceptional event for some domain.   */
103 #define VIRQ_TBUF       4  /* G. (DOM0) Trace buffer has records available.  */
104 #define VIRQ_DEBUGGER   6  /* G. (DOM0) A domain has paused for debugging.   */
105 #define VIRQ_XENOPROF   7  /* V. XenOprofile interrupt: new sample available */
106 #define VIRQ_CON_RING   8  /* G. (DOM0) Bytes received on console            */
107 #define VIRQ_PCPU_STATE 9  /* G. (DOM0) PCPU state changed                   */
108 #define VIRQ_MEM_EVENT  10 /* G. (DOM0) A memory event has occured           */
109 #define VIRQ_XC_RESERVED 11 /* G. Reserved for XenClient                     */
110 #define VIRQ_ENOMEM     12 /* G. (DOM0) Low on heap memory       */
111 
112 /* Architecture-specific VIRQ definitions. */
113 #define VIRQ_ARCH_0    16
114 #define VIRQ_ARCH_1    17
115 #define VIRQ_ARCH_2    18
116 #define VIRQ_ARCH_3    19
117 #define VIRQ_ARCH_4    20
118 #define VIRQ_ARCH_5    21
119 #define VIRQ_ARCH_6    22
120 #define VIRQ_ARCH_7    23
121 
122 #define NR_VIRQS       24
123 
124 /*
125  * enum neg_errnoval HYPERVISOR_mmu_update(const struct mmu_update reqs[],
126  *                                         unsigned count, unsigned *done_out,
127  *                                         unsigned foreigndom)
128  * @reqs is an array of mmu_update_t structures ((ptr, val) pairs).
129  * @count is the length of the above array.
130  * @pdone is an output parameter indicating number of completed operations
131  * @foreigndom[15:0]: FD, the expected owner of data pages referenced in this
132  *                    hypercall invocation. Can be DOMID_SELF.
133  * @foreigndom[31:16]: PFD, the expected owner of pagetable pages referenced
134  *                     in this hypercall invocation. The value of this field
135  *                     (x) encodes the PFD as follows:
136  *                     x == 0 => PFD == DOMID_SELF
137  *                     x != 0 => PFD == x - 1
138  *
139  * Sub-commands: ptr[1:0] specifies the appropriate MMU_* command.
140  * -------------
141  * ptr[1:0] == MMU_NORMAL_PT_UPDATE:
142  * Updates an entry in a page table belonging to PFD. If updating an L1 table,
143  * and the new table entry is valid/present, the mapped frame must belong to
144  * FD. If attempting to map an I/O page then the caller assumes the privilege
145  * of the FD.
146  * FD == DOMID_IO: Permit /only/ I/O mappings, at the priv level of the caller.
147  * FD == DOMID_XEN: Map restricted areas of Xen's heap space.
148  * ptr[:2]  -- Machine address of the page-table entry to modify.
149  * val      -- Value to write.
150  *
151  * There also certain implicit requirements when using this hypercall. The
152  * pages that make up a pagetable must be mapped read-only in the guest.
153  * This prevents uncontrolled guest updates to the pagetable. Xen strictly
154  * enforces this, and will disallow any pagetable update which will end up
155  * mapping pagetable page RW, and will disallow using any writable page as a
156  * pagetable. In practice it means that when constructing a page table for a
157  * process, thread, etc, we MUST be very dilligient in following these rules:
158  *  1). Start with top-level page (PGD or in Xen language: L4). Fill out
159  *      the entries.
160  *  2). Keep on going, filling out the upper (PUD or L3), and middle (PMD
161  *      or L2).
162  *  3). Start filling out the PTE table (L1) with the PTE entries. Once
163  *      done, make sure to set each of those entries to RO (so writeable bit
164  *      is unset). Once that has been completed, set the PMD (L2) for this
165  *      PTE table as RO.
166  *  4). When completed with all of the PMD (L2) entries, and all of them have
167  *      been set to RO, make sure to set RO the PUD (L3). Do the same
168  *      operation on PGD (L4) pagetable entries that have a PUD (L3) entry.
169  *  5). Now before you can use those pages (so setting the cr3), you MUST also
170  *      pin them so that the hypervisor can verify the entries. This is done
171  *      via the HYPERVISOR_mmuext_op(MMUEXT_PIN_L4_TABLE, guest physical frame
172  *      number of the PGD (L4)). And this point the HYPERVISOR_mmuext_op(
173  *      MMUEXT_NEW_BASEPTR, guest physical frame number of the PGD (L4)) can be
174  *      issued.
175  * For 32-bit guests, the L4 is not used (as there is less pagetables), so
176  * instead use L3.
177  * At this point the pagetables can be modified using the MMU_NORMAL_PT_UPDATE
178  * hypercall. Also if so desired the OS can also try to write to the PTE
179  * and be trapped by the hypervisor (as the PTE entry is RO).
180  *
181  * To deallocate the pages, the operations are the reverse of the steps
182  * mentioned above. The argument is MMUEXT_UNPIN_TABLE for all levels and the
183  * pagetable MUST not be in use (meaning that the cr3 is not set to it).
184  *
185  * ptr[1:0] == MMU_MACHPHYS_UPDATE:
186  * Updates an entry in the machine->pseudo-physical mapping table.
187  * ptr[:2]  -- Machine address within the frame whose mapping to modify.
188  *             The frame must belong to the FD, if one is specified.
189  * val      -- Value to write into the mapping entry.
190  *
191  * ptr[1:0] == MMU_PT_UPDATE_PRESERVE_AD:
192  * As MMU_NORMAL_PT_UPDATE above, but A/D bits currently in the PTE are ORed
193  * with those in @val.
194  *
195  * @val is usually the machine frame number along with some attributes.
196  * The attributes by default follow the architecture defined bits. Meaning that
197  * if this is a X86_64 machine and four page table layout is used, the layout
198  * of val is:
199  *  - 63 if set means No execute (NX)
200  *  - 46-13 the machine frame number
201  *  - 12 available for guest
202  *  - 11 available for guest
203  *  - 10 available for guest
204  *  - 9 available for guest
205  *  - 8 global
206  *  - 7 PAT (PSE is disabled, must use hypercall to make 4MB or 2MB pages)
207  *  - 6 dirty
208  *  - 5 accessed
209  *  - 4 page cached disabled
210  *  - 3 page write through
211  *  - 2 userspace accessible
212  *  - 1 writeable
213  *  - 0 present
214  *
215  *  The one bits that does not fit with the default layout is the PAGE_PSE
216  *  also called PAGE_PAT). The MMUEXT_[UN]MARK_SUPER arguments to the
217  *  HYPERVISOR_mmuext_op serve as mechanism to set a pagetable to be 4MB
218  *  (or 2MB) instead of using the PAGE_PSE bit.
219  *
220  *  The reason that the PAGE_PSE (bit 7) is not being utilized is due to Xen
221  *  using it as the Page Attribute Table (PAT) bit - for details on it please
222  *  refer to Intel SDM 10.12. The PAT allows to set the caching attributes of
223  *  pages instead of using MTRRs.
224  *
225  *  The PAT MSR is as follows (it is a 64-bit value, each entry is 8 bits):
226  *                    PAT4                 PAT0
227  *  +-----+-----+----+----+----+-----+----+----+
228  *  | UC  | UC- | WC | WB | UC | UC- | WC | WB |  <= Linux
229  *  +-----+-----+----+----+----+-----+----+----+
230  *  | UC  | UC- | WT | WB | UC | UC- | WT | WB |  <= BIOS (default when machine boots)
231  *  +-----+-----+----+----+----+-----+----+----+
232  *  | rsv | rsv | WP | WC | UC | UC- | WT | WB |  <= Xen
233  *  +-----+-----+----+----+----+-----+----+----+
234  *
235  *  The lookup of this index table translates to looking up
236  *  Bit 7, Bit 4, and Bit 3 of val entry:
237  *
238  *  PAT/PSE (bit 7) ... PCD (bit 4) .. PWT (bit 3).
239  *
240  *  If all bits are off, then we are using PAT0. If bit 3 turned on,
241  *  then we are using PAT1, if bit 3 and bit 4, then PAT2..
242  *
243  *  As you can see, the Linux PAT1 translates to PAT4 under Xen. Which means
244  *  that if a guest that follows Linux's PAT setup and would like to set Write
245  *  Combined on pages it MUST use PAT4 entry. Meaning that Bit 7 (PAGE_PAT) is
246  *  set. For example, under Linux it only uses PAT0, PAT1, and PAT2 for the
247  *  caching as:
248  *
249  *   WB = none (so PAT0)
250  *   WC = PWT (bit 3 on)
251  *   UC = PWT | PCD (bit 3 and 4 are on).
252  *
253  * To make it work with Xen, it needs to translate the WC bit as so:
254  *
255  *  PWT (so bit 3 on) --> PAT (so bit 7 is on) and clear bit 3
256  *
257  * And to translate back it would:
258  *
259  * PAT (bit 7 on) --> PWT (bit 3 on) and clear bit 7.
260  */
261 #define MMU_NORMAL_PT_UPDATE      0 /* checked '*ptr = val'. ptr is MA.       */
262 #define MMU_MACHPHYS_UPDATE       1 /* ptr = MA of frame to modify entry for  */
263 #define MMU_PT_UPDATE_PRESERVE_AD 2 /* atomically: *ptr = val | (*ptr&(A|D)) */
264 
265 /*
266  * MMU EXTENDED OPERATIONS
267  *
268  * enum neg_errnoval HYPERVISOR_mmuext_op(mmuext_op_t uops[],
269  *                                        unsigned int count,
270  *                                        unsigned int *pdone,
271  *                                        unsigned int foreigndom)
272  */
273 /* HYPERVISOR_mmuext_op() accepts a list of mmuext_op structures.
274  * A foreigndom (FD) can be specified (or DOMID_SELF for none).
275  * Where the FD has some effect, it is described below.
276  *
277  * cmd: MMUEXT_(UN)PIN_*_TABLE
278  * mfn: Machine frame number to be (un)pinned as a p.t. page.
279  *      The frame must belong to the FD, if one is specified.
280  *
281  * cmd: MMUEXT_NEW_BASEPTR
282  * mfn: Machine frame number of new page-table base to install in MMU.
283  *
284  * cmd: MMUEXT_NEW_USER_BASEPTR [x86/64 only]
285  * mfn: Machine frame number of new page-table base to install in MMU
286  *      when in user space.
287  *
288  * cmd: MMUEXT_TLB_FLUSH_LOCAL
289  * No additional arguments. Flushes local TLB.
290  *
291  * cmd: MMUEXT_INVLPG_LOCAL
292  * linear_addr: Linear address to be flushed from the local TLB.
293  *
294  * cmd: MMUEXT_TLB_FLUSH_MULTI
295  * vcpumask: Pointer to bitmap of VCPUs to be flushed.
296  *
297  * cmd: MMUEXT_INVLPG_MULTI
298  * linear_addr: Linear address to be flushed.
299  * vcpumask: Pointer to bitmap of VCPUs to be flushed.
300  *
301  * cmd: MMUEXT_TLB_FLUSH_ALL
302  * No additional arguments. Flushes all VCPUs' TLBs.
303  *
304  * cmd: MMUEXT_INVLPG_ALL
305  * linear_addr: Linear address to be flushed from all VCPUs' TLBs.
306  *
307  * cmd: MMUEXT_FLUSH_CACHE
308  * No additional arguments. Writes back and flushes cache contents.
309  *
310  * cmd: MMUEXT_FLUSH_CACHE_GLOBAL
311  * No additional arguments. Writes back and flushes cache contents
312  * on all CPUs in the system.
313  *
314  * cmd: MMUEXT_SET_LDT
315  * linear_addr: Linear address of LDT base (NB. must be page-aligned).
316  * nr_ents: Number of entries in LDT.
317  *
318  * cmd: MMUEXT_CLEAR_PAGE
319  * mfn: Machine frame number to be cleared.
320  *
321  * cmd: MMUEXT_COPY_PAGE
322  * mfn: Machine frame number of the destination page.
323  * src_mfn: Machine frame number of the source page.
324  *
325  * cmd: MMUEXT_[UN]MARK_SUPER
326  * mfn: Machine frame number of head of superpage to be [un]marked.
327  */
328 #define MMUEXT_PIN_L1_TABLE      0
329 #define MMUEXT_PIN_L2_TABLE      1
330 #define MMUEXT_PIN_L3_TABLE      2
331 #define MMUEXT_PIN_L4_TABLE      3
332 #define MMUEXT_UNPIN_TABLE       4
333 #define MMUEXT_NEW_BASEPTR       5
334 #define MMUEXT_TLB_FLUSH_LOCAL   6
335 #define MMUEXT_INVLPG_LOCAL      7
336 #define MMUEXT_TLB_FLUSH_MULTI   8
337 #define MMUEXT_INVLPG_MULTI      9
338 #define MMUEXT_TLB_FLUSH_ALL    10
339 #define MMUEXT_INVLPG_ALL       11
340 #define MMUEXT_FLUSH_CACHE      12
341 #define MMUEXT_SET_LDT          13
342 #define MMUEXT_NEW_USER_BASEPTR 15
343 #define MMUEXT_CLEAR_PAGE       16
344 #define MMUEXT_COPY_PAGE        17
345 #define MMUEXT_FLUSH_CACHE_GLOBAL 18
346 #define MMUEXT_MARK_SUPER       19
347 #define MMUEXT_UNMARK_SUPER     20
348 
349 #ifndef __ASSEMBLY__
350 struct mmuext_op {
351 	unsigned int cmd;
352 	union {
353 		/* [UN]PIN_TABLE, NEW_BASEPTR, NEW_USER_BASEPTR
354 		 * CLEAR_PAGE, COPY_PAGE, [UN]MARK_SUPER */
355 		xen_pfn_t mfn;
356 		/* INVLPG_LOCAL, INVLPG_ALL, SET_LDT */
357 		unsigned long linear_addr;
358 	} arg1;
359 	union {
360 		/* SET_LDT */
361 		unsigned int nr_ents;
362 		/* TLB_FLUSH_MULTI, INVLPG_MULTI */
363 		void *vcpumask;
364 		/* COPY_PAGE */
365 		xen_pfn_t src_mfn;
366 	} arg2;
367 };
368 DEFINE_GUEST_HANDLE_STRUCT(mmuext_op);
369 #endif
370 
371 /* These are passed as 'flags' to update_va_mapping. They can be ORed. */
372 /* When specifying UVMF_MULTI, also OR in a pointer to a CPU bitmap.   */
373 /* UVMF_LOCAL is merely UVMF_MULTI with a NULL bitmap pointer.         */
374 #define UVMF_NONE               (0UL<<0) /* No flushing at all.   */
375 #define UVMF_TLB_FLUSH          (1UL<<0) /* Flush entire TLB(s).  */
376 #define UVMF_INVLPG             (2UL<<0) /* Flush only one entry. */
377 #define UVMF_FLUSHTYPE_MASK     (3UL<<0)
378 #define UVMF_MULTI              (0UL<<2) /* Flush subset of TLBs. */
379 #define UVMF_LOCAL              (0UL<<2) /* Flush local TLB.      */
380 #define UVMF_ALL                (1UL<<2) /* Flush all TLBs.       */
381 
382 /*
383  * Commands to HYPERVISOR_console_io().
384  */
385 #define CONSOLEIO_write         0
386 #define CONSOLEIO_read          1
387 
388 /*
389  * Commands to HYPERVISOR_vm_assist().
390  */
391 #define VMASST_CMD_enable                0
392 #define VMASST_CMD_disable               1
393 
394 /* x86/32 guests: simulate full 4GB segment limits. */
395 #define VMASST_TYPE_4gb_segments         0
396 
397 /* x86/32 guests: trap (vector 15) whenever above vmassist is used. */
398 #define VMASST_TYPE_4gb_segments_notify  1
399 
400 /*
401  * x86 guests: support writes to bottom-level PTEs.
402  * NB1. Page-directory entries cannot be written.
403  * NB2. Guest must continue to remove all writable mappings of PTEs.
404  */
405 #define VMASST_TYPE_writable_pagetables  2
406 
407 /* x86/PAE guests: support PDPTs above 4GB. */
408 #define VMASST_TYPE_pae_extended_cr3     3
409 
410 #define MAX_VMASST_TYPE 3
411 
412 #ifndef __ASSEMBLY__
413 
414 typedef uint16_t domid_t;
415 
416 /* Domain ids >= DOMID_FIRST_RESERVED cannot be used for ordinary domains. */
417 #define DOMID_FIRST_RESERVED (0x7FF0U)
418 
419 /* DOMID_SELF is used in certain contexts to refer to oneself. */
420 #define DOMID_SELF (0x7FF0U)
421 
422 /*
423  * DOMID_IO is used to restrict page-table updates to mapping I/O memory.
424  * Although no Foreign Domain need be specified to map I/O pages, DOMID_IO
425  * is useful to ensure that no mappings to the OS's own heap are accidentally
426  * installed. (e.g., in Linux this could cause havoc as reference counts
427  * aren't adjusted on the I/O-mapping code path).
428  * This only makes sense in MMUEXT_SET_FOREIGNDOM, but in that context can
429  * be specified by any calling domain.
430  */
431 #define DOMID_IO   (0x7FF1U)
432 
433 /*
434  * DOMID_XEN is used to allow privileged domains to map restricted parts of
435  * Xen's heap space (e.g., the machine_to_phys table).
436  * This only makes sense in MMUEXT_SET_FOREIGNDOM, and is only permitted if
437  * the caller is privileged.
438  */
439 #define DOMID_XEN  (0x7FF2U)
440 
441 /* DOMID_COW is used as the owner of sharable pages */
442 #define DOMID_COW  (0x7FF3U)
443 
444 /* DOMID_INVALID is used to identify pages with unknown owner. */
445 #define DOMID_INVALID (0x7FF4U)
446 
447 /* Idle domain. */
448 #define DOMID_IDLE (0x7FFFU)
449 
450 /*
451  * Send an array of these to HYPERVISOR_mmu_update().
452  * NB. The fields are natural pointer/address size for this architecture.
453  */
454 struct mmu_update {
455     uint64_t ptr;       /* Machine address of PTE. */
456     uint64_t val;       /* New contents of PTE.    */
457 };
458 DEFINE_GUEST_HANDLE_STRUCT(mmu_update);
459 
460 /*
461  * Send an array of these to HYPERVISOR_multicall().
462  * NB. The fields are logically the natural register size for this
463  * architecture. In cases where xen_ulong_t is larger than this then
464  * any unused bits in the upper portion must be zero.
465  */
466 struct multicall_entry {
467     xen_ulong_t op;
468     xen_long_t result;
469     xen_ulong_t args[6];
470 };
471 DEFINE_GUEST_HANDLE_STRUCT(multicall_entry);
472 
473 struct vcpu_time_info {
474 	/*
475 	 * Updates to the following values are preceded and followed
476 	 * by an increment of 'version'. The guest can therefore
477 	 * detect updates by looking for changes to 'version'. If the
478 	 * least-significant bit of the version number is set then an
479 	 * update is in progress and the guest must wait to read a
480 	 * consistent set of values.  The correct way to interact with
481 	 * the version number is similar to Linux's seqlock: see the
482 	 * implementations of read_seqbegin/read_seqretry.
483 	 */
484 	uint32_t version;
485 	uint32_t pad0;
486 	uint64_t tsc_timestamp;   /* TSC at last update of time vals.  */
487 	uint64_t system_time;     /* Time, in nanosecs, since boot.    */
488 	/*
489 	 * Current system time:
490 	 *   system_time + ((tsc - tsc_timestamp) << tsc_shift) * tsc_to_system_mul
491 	 * CPU frequency (Hz):
492 	 *   ((10^9 << 32) / tsc_to_system_mul) >> tsc_shift
493 	 */
494 	uint32_t tsc_to_system_mul;
495 	int8_t   tsc_shift;
496 	int8_t   pad1[3];
497 }; /* 32 bytes */
498 
499 struct vcpu_info {
500 	/*
501 	 * 'evtchn_upcall_pending' is written non-zero by Xen to indicate
502 	 * a pending notification for a particular VCPU. It is then cleared
503 	 * by the guest OS /before/ checking for pending work, thus avoiding
504 	 * a set-and-check race. Note that the mask is only accessed by Xen
505 	 * on the CPU that is currently hosting the VCPU. This means that the
506 	 * pending and mask flags can be updated by the guest without special
507 	 * synchronisation (i.e., no need for the x86 LOCK prefix).
508 	 * This may seem suboptimal because if the pending flag is set by
509 	 * a different CPU then an IPI may be scheduled even when the mask
510 	 * is set. However, note:
511 	 *  1. The task of 'interrupt holdoff' is covered by the per-event-
512 	 *     channel mask bits. A 'noisy' event that is continually being
513 	 *     triggered can be masked at source at this very precise
514 	 *     granularity.
515 	 *  2. The main purpose of the per-VCPU mask is therefore to restrict
516 	 *     reentrant execution: whether for concurrency control, or to
517 	 *     prevent unbounded stack usage. Whatever the purpose, we expect
518 	 *     that the mask will be asserted only for short periods at a time,
519 	 *     and so the likelihood of a 'spurious' IPI is suitably small.
520 	 * The mask is read before making an event upcall to the guest: a
521 	 * non-zero mask therefore guarantees that the VCPU will not receive
522 	 * an upcall activation. The mask is cleared when the VCPU requests
523 	 * to block: this avoids wakeup-waiting races.
524 	 */
525 	uint8_t evtchn_upcall_pending;
526 	uint8_t evtchn_upcall_mask;
527 	xen_ulong_t evtchn_pending_sel;
528 	struct arch_vcpu_info arch;
529 	struct pvclock_vcpu_time_info time;
530 }; /* 64 bytes (x86) */
531 
532 /*
533  * Xen/kernel shared data -- pointer provided in start_info.
534  * NB. We expect that this struct is smaller than a page.
535  */
536 struct shared_info {
537 	struct vcpu_info vcpu_info[MAX_VIRT_CPUS];
538 
539 	/*
540 	 * A domain can create "event channels" on which it can send and receive
541 	 * asynchronous event notifications. There are three classes of event that
542 	 * are delivered by this mechanism:
543 	 *  1. Bi-directional inter- and intra-domain connections. Domains must
544 	 *     arrange out-of-band to set up a connection (usually by allocating
545 	 *     an unbound 'listener' port and avertising that via a storage service
546 	 *     such as xenstore).
547 	 *  2. Physical interrupts. A domain with suitable hardware-access
548 	 *     privileges can bind an event-channel port to a physical interrupt
549 	 *     source.
550 	 *  3. Virtual interrupts ('events'). A domain can bind an event-channel
551 	 *     port to a virtual interrupt source, such as the virtual-timer
552 	 *     device or the emergency console.
553 	 *
554 	 * Event channels are addressed by a "port index". Each channel is
555 	 * associated with two bits of information:
556 	 *  1. PENDING -- notifies the domain that there is a pending notification
557 	 *     to be processed. This bit is cleared by the guest.
558 	 *  2. MASK -- if this bit is clear then a 0->1 transition of PENDING
559 	 *     will cause an asynchronous upcall to be scheduled. This bit is only
560 	 *     updated by the guest. It is read-only within Xen. If a channel
561 	 *     becomes pending while the channel is masked then the 'edge' is lost
562 	 *     (i.e., when the channel is unmasked, the guest must manually handle
563 	 *     pending notifications as no upcall will be scheduled by Xen).
564 	 *
565 	 * To expedite scanning of pending notifications, any 0->1 pending
566 	 * transition on an unmasked channel causes a corresponding bit in a
567 	 * per-vcpu selector word to be set. Each bit in the selector covers a
568 	 * 'C long' in the PENDING bitfield array.
569 	 */
570 	xen_ulong_t evtchn_pending[sizeof(xen_ulong_t) * 8];
571 	xen_ulong_t evtchn_mask[sizeof(xen_ulong_t) * 8];
572 
573 	/*
574 	 * Wallclock time: updated only by control software. Guests should base
575 	 * their gettimeofday() syscall on this wallclock-base value.
576 	 */
577 	struct pvclock_wall_clock wc;
578 
579 	struct arch_shared_info arch;
580 
581 };
582 
583 /*
584  * Start-of-day memory layout for the initial domain (DOM0):
585  *  1. The domain is started within contiguous virtual-memory region.
586  *  2. The contiguous region begins and ends on an aligned 4MB boundary.
587  *  3. The region start corresponds to the load address of the OS image.
588  *     If the load address is not 4MB aligned then the address is rounded down.
589  *  4. This the order of bootstrap elements in the initial virtual region:
590  *      a. relocated kernel image
591  *      b. initial ram disk              [mod_start, mod_len]
592  *      c. list of allocated page frames [mfn_list, nr_pages]
593  *      d. start_info_t structure        [register ESI (x86)]
594  *      e. bootstrap page tables         [pt_base, CR3 (x86)]
595  *      f. bootstrap stack               [register ESP (x86)]
596  *  5. Bootstrap elements are packed together, but each is 4kB-aligned.
597  *  6. The initial ram disk may be omitted.
598  *  7. The list of page frames forms a contiguous 'pseudo-physical' memory
599  *     layout for the domain. In particular, the bootstrap virtual-memory
600  *     region is a 1:1 mapping to the first section of the pseudo-physical map.
601  *  8. All bootstrap elements are mapped read-writable for the guest OS. The
602  *     only exception is the bootstrap page table, which is mapped read-only.
603  *  9. There is guaranteed to be at least 512kB padding after the final
604  *     bootstrap element. If necessary, the bootstrap virtual region is
605  *     extended by an extra 4MB to ensure this.
606  */
607 
608 #define MAX_GUEST_CMDLINE 1024
609 struct start_info {
610 	/* THE FOLLOWING ARE FILLED IN BOTH ON INITIAL BOOT AND ON RESUME.    */
611 	char magic[32];             /* "xen-<version>-<platform>".            */
612 	unsigned long nr_pages;     /* Total pages allocated to this domain.  */
613 	unsigned long shared_info;  /* MACHINE address of shared info struct. */
614 	uint32_t flags;             /* SIF_xxx flags.                         */
615 	xen_pfn_t store_mfn;        /* MACHINE page number of shared page.    */
616 	uint32_t store_evtchn;      /* Event channel for store communication. */
617 	union {
618 		struct {
619 			xen_pfn_t mfn;      /* MACHINE page number of console page.   */
620 			uint32_t  evtchn;   /* Event channel for console page.        */
621 		} domU;
622 		struct {
623 			uint32_t info_off;  /* Offset of console_info struct.         */
624 			uint32_t info_size; /* Size of console_info struct from start.*/
625 		} dom0;
626 	} console;
627 	/* THE FOLLOWING ARE ONLY FILLED IN ON INITIAL BOOT (NOT RESUME).     */
628 	unsigned long pt_base;      /* VIRTUAL address of page directory.     */
629 	unsigned long nr_pt_frames; /* Number of bootstrap p.t. frames.       */
630 	unsigned long mfn_list;     /* VIRTUAL address of page-frame list.    */
631 	unsigned long mod_start;    /* VIRTUAL address of pre-loaded module.  */
632 	unsigned long mod_len;      /* Size (bytes) of pre-loaded module.     */
633 	int8_t cmd_line[MAX_GUEST_CMDLINE];
634 	/* The pfn range here covers both page table and p->m table frames.   */
635 	unsigned long first_p2m_pfn;/* 1st pfn forming initial P->M table.    */
636 	unsigned long nr_p2m_frames;/* # of pfns forming initial P->M table.  */
637 };
638 
639 /* These flags are passed in the 'flags' field of start_info_t. */
640 #define SIF_PRIVILEGED    (1<<0)  /* Is the domain privileged? */
641 #define SIF_INITDOMAIN    (1<<1)  /* Is this the initial control domain? */
642 #define SIF_MULTIBOOT_MOD (1<<2)  /* Is mod_start a multiboot module? */
643 #define SIF_MOD_START_PFN (1<<3)  /* Is mod_start a PFN? */
644 #define SIF_PM_MASK       (0xFF<<8) /* reserve 1 byte for xen-pm options */
645 
646 /*
647  * A multiboot module is a package containing modules very similar to a
648  * multiboot module array. The only differences are:
649  * - the array of module descriptors is by convention simply at the beginning
650  *   of the multiboot module,
651  * - addresses in the module descriptors are based on the beginning of the
652  *   multiboot module,
653  * - the number of modules is determined by a termination descriptor that has
654  *   mod_start == 0.
655  *
656  * This permits to both build it statically and reference it in a configuration
657  * file, and let the PV guest easily rebase the addresses to virtual addresses
658  * and at the same time count the number of modules.
659  */
660 struct xen_multiboot_mod_list {
661 	/* Address of first byte of the module */
662 	uint32_t mod_start;
663 	/* Address of last byte of the module (inclusive) */
664 	uint32_t mod_end;
665 	/* Address of zero-terminated command line */
666 	uint32_t cmdline;
667 	/* Unused, must be zero */
668 	uint32_t pad;
669 };
670 /*
671  * The console structure in start_info.console.dom0
672  *
673  * This structure includes a variety of information required to
674  * have a working VGA/VESA console.
675  */
676 struct dom0_vga_console_info {
677 	uint8_t video_type;
678 #define XEN_VGATYPE_TEXT_MODE_3 0x03
679 #define XEN_VGATYPE_VESA_LFB    0x23
680 #define XEN_VGATYPE_EFI_LFB     0x70
681 
682 	union {
683 		struct {
684 			/* Font height, in pixels. */
685 			uint16_t font_height;
686 			/* Cursor location (column, row). */
687 			uint16_t cursor_x, cursor_y;
688 			/* Number of rows and columns (dimensions in characters). */
689 			uint16_t rows, columns;
690 		} text_mode_3;
691 
692 		struct {
693 			/* Width and height, in pixels. */
694 			uint16_t width, height;
695 			/* Bytes per scan line. */
696 			uint16_t bytes_per_line;
697 			/* Bits per pixel. */
698 			uint16_t bits_per_pixel;
699 			/* LFB physical address, and size (in units of 64kB). */
700 			uint32_t lfb_base;
701 			uint32_t lfb_size;
702 			/* RGB mask offsets and sizes, as defined by VBE 1.2+ */
703 			uint8_t  red_pos, red_size;
704 			uint8_t  green_pos, green_size;
705 			uint8_t  blue_pos, blue_size;
706 			uint8_t  rsvd_pos, rsvd_size;
707 
708 			/* VESA capabilities (offset 0xa, VESA command 0x4f00). */
709 			uint32_t gbl_caps;
710 			/* Mode attributes (offset 0x0, VESA command 0x4f01). */
711 			uint16_t mode_attrs;
712 		} vesa_lfb;
713 	} u;
714 };
715 
716 typedef uint64_t cpumap_t;
717 
718 typedef uint8_t xen_domain_handle_t[16];
719 
720 /* Turn a plain number into a C unsigned long constant. */
721 #define __mk_unsigned_long(x) x ## UL
722 #define mk_unsigned_long(x) __mk_unsigned_long(x)
723 
724 #define TMEM_SPEC_VERSION 1
725 
726 struct tmem_op {
727 	uint32_t cmd;
728 	int32_t pool_id;
729 	union {
730 		struct {  /* for cmd == TMEM_NEW_POOL */
731 			uint64_t uuid[2];
732 			uint32_t flags;
733 		} new;
734 		struct {
735 			uint64_t oid[3];
736 			uint32_t index;
737 			uint32_t tmem_offset;
738 			uint32_t pfn_offset;
739 			uint32_t len;
740 			GUEST_HANDLE(void) gmfn; /* guest machine page frame */
741 		} gen;
742 	} u;
743 };
744 
745 DEFINE_GUEST_HANDLE(u64);
746 
747 #else /* __ASSEMBLY__ */
748 
749 /* In assembly code we cannot use C numeric constant suffixes. */
750 #define mk_unsigned_long(x) x
751 
752 #endif /* !__ASSEMBLY__ */
753 
754 #endif /* __XEN_PUBLIC_XEN_H__ */
755