sys/i386/xen/xen_machdep.c

   1 /*
   2  *
   3  * Copyright (c) 2004 Christian Limpach.
   4  * Copyright (c) 2004-2006,2008 Kip Macy
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. All advertising materials mentioning features or use of this software
  16  *    must display the following acknowledgement:
  17  *      This product includes software developed by Christian Limpach.
  18  * 4. The name of the author may not be used to endorse or promote products
  19  *    derived from this software without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  24  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  31  */
  32
  33 #include <sys/cdefs.h>
  34 __FBSDID("$FreeBSD$");
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/bus.h>
  39 #include <sys/ktr.h>
  40 #include <sys/lock.h>
  41 #include <sys/mount.h>
  42 #include <sys/malloc.h>
  43 #include <sys/mutex.h>
  44 #include <sys/kernel.h>
  45 #include <sys/proc.h>
  46 #include <sys/reboot.h>
  47 #include <sys/sysproto.h>
  48
  49 #include <machine/xen/xen-os.h>
  50
  51 #include <vm/vm.h>
  52 #include <vm/pmap.h>
  53 #include <machine/segments.h>
  54 #include <machine/pcb.h>
  55 #include <machine/stdarg.h>
  56 #include <machine/vmparam.h>
  57 #include <machine/cpu.h>
  58 #include <machine/intr_machdep.h>
  59 #include <machine/md_var.h>
  60 #include <machine/asmacros.h>
  61
  62
  63
  64 #include <xen/hypervisor.h>
  65 #include <machine/xen/xenvar.h>
  66 #include <machine/xen/xenfunc.h>
  67 #include <machine/xen/xenpmap.h>
  68 #include <machine/xen/xenfunc.h>
  69 #include <xen/interface/memory.h>
  70 #include <machine/xen/features.h>
  71 #ifdef SMP
  72 #include <machine/privatespace.h>
  73 #endif
  74
  75
  76 #include <vm/vm_page.h>
  77
  78
  79 #define IDTVEC(name)    __CONCAT(X,name)
  80
  81 extern inthand_t
  82 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
  83         IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
  84         IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
  85         IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align),
  86         IDTVEC(xmm), IDTVEC(lcall_syscall), IDTVEC(int0x80_syscall);
  87
  88
  89 int xendebug_flags;
  90 start_info_t *xen_start_info;
  91 shared_info_t *HYPERVISOR_shared_info;
  92 xen_pfn_t *xen_machine_phys = machine_to_phys_mapping;
  93 xen_pfn_t *xen_phys_machine;
  94 xen_pfn_t *xen_pfn_to_mfn_frame_list[16];
  95 xen_pfn_t *xen_pfn_to_mfn_frame_list_list;
  96 int preemptable, init_first;
  97 extern unsigned int avail_space;
  98
  99 void ni_cli(void);
 100 void ni_sti(void);
 101
 102
 103 void
 104 ni_cli(void)
 105 {
 106         CTR0(KTR_SPARE2, "ni_cli disabling interrupts");
 107         __asm__("pushl %edx;"
 108                 "pushl %eax;"
 109                 );
 110         __cli();
 111         __asm__("popl %eax;"
 112                 "popl %edx;"
 113                 );
 114 }
 115
 116
 117 void
 118 ni_sti(void)
 119 {
 120         __asm__("pushl %edx;"
 121                 "pushl %esi;"
 122                 "pushl %eax;"
 123                 );
 124         __sti();
 125         __asm__("popl %eax;"
 126                 "popl %esi;"
 127                 "popl %edx;"
 128                 );
 129 }
 130
 131 /*
 132  * Modify the cmd_line by converting ',' to NULLs so that it is in a  format
 133  * suitable for the static env vars.
 134  */
 135 char *
 136 xen_setbootenv(char *cmd_line)
 137 {
 138         char *cmd_line_next;
 139
 140         /* Skip leading spaces */
 141         for (; *cmd_line == ' '; cmd_line++);
 142
 143         printk("xen_setbootenv(): cmd_line='%s'\n", cmd_line);
 144
 145         for (cmd_line_next = cmd_line; strsep(&cmd_line_next, ",") != NULL;);
 146         return cmd_line;
 147 }
 148
 149 static struct
 150 {
 151         const char      *ev;
 152         int             mask;
 153 } howto_names[] = {
 154         {"boot_askname",        RB_ASKNAME},
 155         {"boot_single", RB_SINGLE},
 156         {"boot_nosync", RB_NOSYNC},
 157         {"boot_halt",   RB_ASKNAME},
 158         {"boot_serial", RB_SERIAL},
 159         {"boot_cdrom",  RB_CDROM},
 160         {"boot_gdb",    RB_GDB},
 161         {"boot_gdb_pause",      RB_RESERVED1},
 162         {"boot_verbose",        RB_VERBOSE},
 163         {"boot_multicons",      RB_MULTIPLE},
 164         {NULL,  0}
 165 };
 166
 167 int
 168 xen_boothowto(char *envp)
 169 {
 170         int i, howto = 0;
 171
 172         /* get equivalents from the environment */
 173         for (i = 0; howto_names[i].ev != NULL; i++)
 174                 if (getenv(howto_names[i].ev) != NULL)
 175                         howto |= howto_names[i].mask;
 176         return howto;
 177 }
 178
 179 #define PRINTK_BUFSIZE 1024
 180 void
 181 printk(const char *fmt, ...)
 182 {
 183         __va_list ap;
 184         int retval;
 185         static char buf[PRINTK_BUFSIZE];
 186
 187         va_start(ap, fmt);
 188         retval = vsnprintf(buf, PRINTK_BUFSIZE - 1, fmt, ap);
 189         va_end(ap);
 190         buf[retval] = 0;
 191         (void)HYPERVISOR_console_write(buf, retval);
 192 }
 193
 194
 195 #define XPQUEUE_SIZE 128
 196
 197 struct mmu_log {
 198         char *file;
 199         int line;
 200 };
 201
 202 #ifdef SMP
 203 /* per-cpu queues and indices */
 204 #ifdef INVARIANTS
 205 static struct mmu_log xpq_queue_log[MAX_VIRT_CPUS][XPQUEUE_SIZE];
 206 #endif
 207
 208 static int xpq_idx[MAX_VIRT_CPUS];
 209 static mmu_update_t xpq_queue[MAX_VIRT_CPUS][XPQUEUE_SIZE];
 210
 211 #define XPQ_QUEUE_LOG xpq_queue_log[vcpu]
 212 #define XPQ_QUEUE xpq_queue[vcpu]
 213 #define XPQ_IDX xpq_idx[vcpu]
 214 #define SET_VCPU() int vcpu = smp_processor_id()
 215 #else
 216
 217 static mmu_update_t xpq_queue[XPQUEUE_SIZE];
 218 static struct mmu_log xpq_queue_log[XPQUEUE_SIZE];
 219 static int xpq_idx = 0;
 220
 221 #define XPQ_QUEUE_LOG xpq_queue_log
 222 #define XPQ_QUEUE xpq_queue
 223 #define XPQ_IDX xpq_idx
 224 #define SET_VCPU()
 225 #endif /* !SMP */
 226
 227 #define XPQ_IDX_INC atomic_add_int(&XPQ_IDX, 1);
 228
 229 #if 0
 230 static void
 231 xen_dump_queue(void)
 232 {
 233         int _xpq_idx = XPQ_IDX;
 234         int i;
 235
 236         if (_xpq_idx <= 1)
 237                 return;
 238
 239         printk("xen_dump_queue(): %u entries\n", _xpq_idx);
 240         for (i = 0; i < _xpq_idx; i++) {
 241                 printk(" val: %llx ptr: %llx\n", XPQ_QUEUE[i].val, XPQ_QUEUE[i].ptr);
 242         }
 243 }
 244 #endif
 245
 246
 247 static __inline void
 248 _xen_flush_queue(void)
 249 {
 250         SET_VCPU();
 251         int _xpq_idx = XPQ_IDX;
 252         int error, i;
 253
 254 #ifdef INVARIANTS
 255         if (__predict_true(gdtset))
 256                 CRITICAL_ASSERT(curthread);
 257 #endif
 258
 259         XPQ_IDX = 0;
 260         /* Make sure index is cleared first to avoid double updates. */
 261         error = HYPERVISOR_mmu_update((mmu_update_t *)&XPQ_QUEUE,
 262                                       _xpq_idx, NULL, DOMID_SELF);
 263
 264 #if 0
 265         if (__predict_true(gdtset))
 266         for (i = _xpq_idx; i > 0;) {
 267                 if (i >= 3) {
 268                         CTR6(KTR_PMAP, "mmu:val: %lx ptr: %lx val: %lx "
 269                             "ptr: %lx val: %lx ptr: %lx",
 270                             (XPQ_QUEUE[i-1].val & 0xffffffff),
 271                             (XPQ_QUEUE[i-1].ptr & 0xffffffff),
 272                             (XPQ_QUEUE[i-2].val & 0xffffffff),
 273                             (XPQ_QUEUE[i-2].ptr & 0xffffffff),
 274                             (XPQ_QUEUE[i-3].val & 0xffffffff),
 275                             (XPQ_QUEUE[i-3].ptr & 0xffffffff));
 276                             i -= 3;
 277                 } else if (i == 2) {
 278                         CTR4(KTR_PMAP, "mmu: val: %lx ptr: %lx val: %lx ptr: %lx",
 279                             (XPQ_QUEUE[i-1].val & 0xffffffff),
 280                             (XPQ_QUEUE[i-1].ptr & 0xffffffff),
 281                             (XPQ_QUEUE[i-2].val & 0xffffffff),
 282                             (XPQ_QUEUE[i-2].ptr & 0xffffffff));
 283                         i = 0;
 284                 } else {
 285                         CTR2(KTR_PMAP, "mmu: val: %lx ptr: %lx",
 286                             (XPQ_QUEUE[i-1].val & 0xffffffff),
 287                             (XPQ_QUEUE[i-1].ptr & 0xffffffff));
 288                         i = 0;
 289                 }
 290         }
 291 #endif
 292         if (__predict_false(error < 0)) {
 293                 for (i = 0; i < _xpq_idx; i++)
 294                         printf("val: %llx ptr: %llx\n",
 295                             XPQ_QUEUE[i].val, XPQ_QUEUE[i].ptr);
 296                 panic("Failed to execute MMU updates: %d", error);
 297         }
 298
 299 }
 300
 301 void
 302 xen_flush_queue(void)
 303 {
 304         SET_VCPU();
 305
 306         if (__predict_true(gdtset))
 307                 critical_enter();
 308         if (XPQ_IDX != 0) _xen_flush_queue();
 309         if (__predict_true(gdtset))
 310                 critical_exit();
 311 }
 312
 313 static __inline void
 314 xen_increment_idx(void)
 315 {
 316         SET_VCPU();
 317
 318         XPQ_IDX++;
 319         if (__predict_false(XPQ_IDX == XPQUEUE_SIZE))
 320                 xen_flush_queue();
 321 }
 322
 323 void
 324 xen_check_queue(void)
 325 {
 326 #ifdef INVARIANTS
 327         SET_VCPU();
 328
 329         KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX));
 330 #endif
 331 }
 332
 333 void
 334 xen_invlpg(vm_offset_t va)
 335 {
 336         struct mmuext_op op;
 337         op.cmd = MMUEXT_INVLPG_ALL;
 338         op.arg1.linear_addr = va & ~PAGE_MASK;
 339         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 340 }
 341
 342 void
 343 xen_load_cr3(u_int val)
 344 {
 345         struct mmuext_op op;
 346 #ifdef INVARIANTS
 347         SET_VCPU();
 348
 349         KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX));
 350 #endif
 351         op.cmd = MMUEXT_NEW_BASEPTR;
 352         op.arg1.mfn = xpmap_ptom(val) >> PAGE_SHIFT;
 353         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 354 }
 355
 356 #ifdef KTR
 357 static __inline u_int
 358 rebp(void)
 359 {
 360         u_int   data;
 361
 362         __asm __volatile("movl 4(%%ebp),%0" : "=r" (data));
 363         return (data);
 364 }
 365 #endif
 366
 367 u_int
 368 read_eflags(void)
 369 {
 370         vcpu_info_t *_vcpu;
 371         u_int eflags;
 372
 373         eflags = _read_eflags();
 374         _vcpu = &HYPERVISOR_shared_info->vcpu_info[smp_processor_id()];
 375         if (_vcpu->evtchn_upcall_mask)
 376                 eflags &= ~PSL_I;
 377
 378         return (eflags);
 379 }
 380
 381 void
 382 write_eflags(u_int eflags)
 383 {
 384         u_int intr;
 385
 386         CTR2(KTR_SPARE2, "%x xen_restore_flags eflags %x", rebp(), eflags);
 387         intr = ((eflags & PSL_I) == 0);
 388         __restore_flags(intr);
 389         _write_eflags(eflags);
 390 }
 391
 392 void
 393 xen_cli(void)
 394 {
 395         CTR1(KTR_SPARE2, "%x xen_cli disabling interrupts", rebp());
 396         __cli();
 397 }
 398
 399 void
 400 xen_sti(void)
 401 {
 402         CTR1(KTR_SPARE2, "%x xen_sti enabling interrupts", rebp());
 403         __sti();
 404 }
 405
 406 u_int
 407 xen_rcr2(void)
 408 {
 409
 410         return (HYPERVISOR_shared_info->vcpu_info[curcpu].arch.cr2);
 411 }
 412
 413 void
 414 _xen_machphys_update(vm_paddr_t mfn, vm_paddr_t pfn, char *file, int line)
 415 {
 416         SET_VCPU();
 417
 418         if (__predict_true(gdtset))
 419                 critical_enter();
 420         XPQ_QUEUE[XPQ_IDX].ptr = (mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE;
 421         XPQ_QUEUE[XPQ_IDX].val = pfn;
 422 #ifdef INVARIANTS
 423         XPQ_QUEUE_LOG[XPQ_IDX].file = file;
 424         XPQ_QUEUE_LOG[XPQ_IDX].line = line;
 425 #endif
 426         xen_increment_idx();
 427         if (__predict_true(gdtset))
 428                 critical_exit();
 429 }
 430
 431 void
 432 _xen_queue_pt_update(vm_paddr_t ptr, vm_paddr_t val, char *file, int line)
 433 {
 434         SET_VCPU();
 435
 436         if (__predict_true(gdtset))
 437                 mtx_assert(&vm_page_queue_mtx, MA_OWNED);
 438
 439         KASSERT((ptr & 7) == 0, ("misaligned update"));
 440
 441         if (__predict_true(gdtset))
 442                 critical_enter();
 443
 444         XPQ_QUEUE[XPQ_IDX].ptr = ((uint64_t)ptr) | MMU_NORMAL_PT_UPDATE;
 445         XPQ_QUEUE[XPQ_IDX].val = (uint64_t)val;
 446 #ifdef INVARIANTS
 447         XPQ_QUEUE_LOG[XPQ_IDX].file = file;
 448         XPQ_QUEUE_LOG[XPQ_IDX].line = line;
 449 #endif
 450         xen_increment_idx();
 451         if (__predict_true(gdtset))
 452                 critical_exit();
 453 }
 454
 455 void
 456 xen_pgdpt_pin(vm_paddr_t ma)
 457 {
 458         struct mmuext_op op;
 459         op.cmd = MMUEXT_PIN_L3_TABLE;
 460         op.arg1.mfn = ma >> PAGE_SHIFT;
 461         xen_flush_queue();
 462         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 463 }
 464
 465 void
 466 xen_pgd_pin(vm_paddr_t ma)
 467 {
 468         struct mmuext_op op;
 469         op.cmd = MMUEXT_PIN_L2_TABLE;
 470         op.arg1.mfn = ma >> PAGE_SHIFT;
 471         xen_flush_queue();
 472         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 473 }
 474
 475 void
 476 xen_pgd_unpin(vm_paddr_t ma)
 477 {
 478         struct mmuext_op op;
 479         op.cmd = MMUEXT_UNPIN_TABLE;
 480         op.arg1.mfn = ma >> PAGE_SHIFT;
 481         xen_flush_queue();
 482         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 483 }
 484
 485 void
 486 xen_pt_pin(vm_paddr_t ma)
 487 {
 488         struct mmuext_op op;
 489         op.cmd = MMUEXT_PIN_L1_TABLE;
 490         op.arg1.mfn = ma >> PAGE_SHIFT;
 491         xen_flush_queue();
 492         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 493 }
 494
 495 void
 496 xen_pt_unpin(vm_paddr_t ma)
 497 {
 498         struct mmuext_op op;
 499         op.cmd = MMUEXT_UNPIN_TABLE;
 500         op.arg1.mfn = ma >> PAGE_SHIFT;
 501         xen_flush_queue();
 502         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 503 }
 504
 505 void
 506 xen_set_ldt(vm_paddr_t ptr, unsigned long len)
 507 {
 508         struct mmuext_op op;
 509         op.cmd = MMUEXT_SET_LDT;
 510         op.arg1.linear_addr = ptr;
 511         op.arg2.nr_ents = len;
 512         xen_flush_queue();
 513         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 514 }
 515
 516 void xen_tlb_flush(void)
 517 {
 518         struct mmuext_op op;
 519         op.cmd = MMUEXT_TLB_FLUSH_LOCAL;
 520         xen_flush_queue();
 521         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 522 }
 523
 524 void
 525 xen_update_descriptor(union descriptor *table, union descriptor *entry)
 526 {
 527         vm_paddr_t pa;
 528         pt_entry_t *ptp;
 529
 530         ptp = vtopte((vm_offset_t)table);
 531         pa = (*ptp & PG_FRAME) | ((vm_offset_t)table & PAGE_MASK);
 532         if (HYPERVISOR_update_descriptor(pa, *(uint64_t *)entry))
 533                 panic("HYPERVISOR_update_descriptor failed\n");
 534 }
 535
 536
 537 #if 0
 538 /*
 539  * Bitmap is indexed by page number. If bit is set, the page is part of a
 540  * xen_create_contiguous_region() area of memory.
 541  */
 542 unsigned long *contiguous_bitmap;
 543
 544 static void
 545 contiguous_bitmap_set(unsigned long first_page, unsigned long nr_pages)
 546 {
 547         unsigned long start_off, end_off, curr_idx, end_idx;
 548
 549         curr_idx  = first_page / BITS_PER_LONG;
 550         start_off = first_page & (BITS_PER_LONG-1);
 551         end_idx   = (first_page + nr_pages) / BITS_PER_LONG;
 552         end_off   = (first_page + nr_pages) & (BITS_PER_LONG-1);
 553
 554         if (curr_idx == end_idx) {
 555                 contiguous_bitmap[curr_idx] |=
 556                         ((1UL<<end_off)-1) & -(1UL<<start_off);
 557         } else {
 558                 contiguous_bitmap[curr_idx] |= -(1UL<<start_off);
 559                 while ( ++curr_idx < end_idx )
 560                         contiguous_bitmap[curr_idx] = ~0UL;
 561                 contiguous_bitmap[curr_idx] |= (1UL<<end_off)-1;
 562         }
 563 }
 564
 565 static void
 566 contiguous_bitmap_clear(unsigned long first_page, unsigned long nr_pages)
 567 {
 568         unsigned long start_off, end_off, curr_idx, end_idx;
 569
 570         curr_idx  = first_page / BITS_PER_LONG;
 571         start_off = first_page & (BITS_PER_LONG-1);
 572         end_idx   = (first_page + nr_pages) / BITS_PER_LONG;
 573         end_off   = (first_page + nr_pages) & (BITS_PER_LONG-1);
 574
 575         if (curr_idx == end_idx) {
 576                 contiguous_bitmap[curr_idx] &=
 577                         -(1UL<<end_off) | ((1UL<<start_off)-1);
 578         } else {
 579                 contiguous_bitmap[curr_idx] &= (1UL<<start_off)-1;
 580                 while ( ++curr_idx != end_idx )
 581                         contiguous_bitmap[curr_idx] = 0;
 582                 contiguous_bitmap[curr_idx] &= -(1UL<<end_off);
 583         }
 584 }
 585 #endif
 586
 587 /* Ensure multi-page extents are contiguous in machine memory. */
 588 int
 589 xen_create_contiguous_region(vm_page_t pages, int npages)
 590 {
 591         unsigned long  mfn, i, flags;
 592         int order;
 593         struct xen_memory_reservation reservation = {
 594                 .nr_extents   = 1,
 595                 .extent_order = 0,
 596                 .domid        = DOMID_SELF
 597         };
 598         set_xen_guest_handle(reservation.extent_start, &mfn);
 599
 600         balloon_lock(flags);
 601
 602         /* can currently only handle power of two allocation */
 603         PANIC_IF(ffs(npages) != fls(npages));
 604
 605         /* 0. determine order */
 606         order = (ffs(npages) == fls(npages)) ? fls(npages) - 1 : fls(npages);
 607
 608         /* 1. give away machine pages. */
 609         for (i = 0; i < (1 << order); i++) {
 610                 int pfn;
 611                 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
 612                 mfn = PFNTOMFN(pfn);
 613                 PFNTOMFN(pfn) = INVALID_P2M_ENTRY;
 614                 PANIC_IF(HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation) != 1);
 615         }
 616
 617
 618         /* 2. Get a new contiguous memory extent. */
 619         reservation.extent_order = order;
 620         /* xenlinux hardcodes this because of aacraid - maybe set to 0 if we're not
 621          * running with a broxen driver XXXEN
 622          */
 623         reservation.address_bits = 31;
 624         if (HYPERVISOR_memory_op(XENMEM_increase_reservation, &reservation) != 1)
 625                 goto fail;
 626
 627         /* 3. Map the new extent in place of old pages. */
 628         for (i = 0; i < (1 << order); i++) {
 629                 int pfn;
 630                 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
 631                 xen_machphys_update(mfn+i, pfn);
 632                 PFNTOMFN(pfn) = mfn+i;
 633         }
 634
 635         xen_tlb_flush();
 636
 637 #if 0
 638         contiguous_bitmap_set(VM_PAGE_TO_PHYS(&pages[0]) >> PAGE_SHIFT, 1UL << order);
 639 #endif
 640
 641         balloon_unlock(flags);
 642
 643         return 0;
 644
 645  fail:
 646         reservation.extent_order = 0;
 647         reservation.address_bits = 0;
 648
 649         for (i = 0; i < (1 << order); i++) {
 650                 int pfn;
 651                 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
 652                 PANIC_IF(HYPERVISOR_memory_op(
 653                         XENMEM_increase_reservation, &reservation) != 1);
 654                 xen_machphys_update(mfn, pfn);
 655                 PFNTOMFN(pfn) = mfn;
 656         }
 657
 658         xen_tlb_flush();
 659
 660         balloon_unlock(flags);
 661
 662         return ENOMEM;
 663 }
 664
 665 void
 666 xen_destroy_contiguous_region(void *addr, int npages)
 667 {
 668         unsigned long  mfn, i, flags, order, pfn0;
 669         struct xen_memory_reservation reservation = {
 670                 .nr_extents   = 1,
 671                 .extent_order = 0,
 672                 .domid        = DOMID_SELF
 673         };
 674         set_xen_guest_handle(reservation.extent_start, &mfn);
 675
 676         pfn0 = vtophys(addr) >> PAGE_SHIFT;
 677 #if 0
 678         scrub_pages(vstart, 1 << order);
 679 #endif
 680         /* can currently only handle power of two allocation */
 681         PANIC_IF(ffs(npages) != fls(npages));
 682
 683         /* 0. determine order */
 684         order = (ffs(npages) == fls(npages)) ? fls(npages) - 1 : fls(npages);
 685
 686         balloon_lock(flags);
 687
 688 #if 0
 689         contiguous_bitmap_clear(vtophys(addr) >> PAGE_SHIFT, 1UL << order);
 690 #endif
 691
 692         /* 1. Zap current PTEs, giving away the underlying pages. */
 693         for (i = 0; i < (1 << order); i++) {
 694                 int pfn;
 695                 uint64_t new_val = 0;
 696                 pfn = vtomach((char *)addr + i*PAGE_SIZE) >> PAGE_SHIFT;
 697
 698                 PANIC_IF(HYPERVISOR_update_va_mapping((vm_offset_t)((char *)addr + (i * PAGE_SIZE)), new_val, 0));
 699                 PFNTOMFN(pfn) = INVALID_P2M_ENTRY;
 700                 PANIC_IF(HYPERVISOR_memory_op(
 701                         XENMEM_decrease_reservation, &reservation) != 1);
 702         }
 703
 704         /* 2. Map new pages in place of old pages. */
 705         for (i = 0; i < (1 << order); i++) {
 706                 int pfn;
 707                 uint64_t new_val;
 708                 pfn = pfn0 + i;
 709                 PANIC_IF(HYPERVISOR_memory_op(XENMEM_increase_reservation, &reservation) != 1);
 710
 711                 new_val = mfn << PAGE_SHIFT;
 712                 PANIC_IF(HYPERVISOR_update_va_mapping((vm_offset_t)addr + (i * PAGE_SIZE),
 713                                                       new_val, PG_KERNEL));
 714                 xen_machphys_update(mfn, pfn);
 715                 PFNTOMFN(pfn) = mfn;
 716         }
 717
 718         xen_tlb_flush();
 719
 720         balloon_unlock(flags);
 721 }
 722
 723 extern  vm_offset_t     proc0kstack;
 724 extern int vm86paddr, vm86phystk;
 725 char *bootmem_start, *bootmem_current, *bootmem_end;
 726
 727 pteinfo_t *pteinfo_list;
 728 void initvalues(start_info_t *startinfo);
 729
 730 struct xenstore_domain_interface;
 731 extern struct xenstore_domain_interface *xen_store;
 732
 733 char *console_page;
 734
 735 void *
 736 bootmem_alloc(unsigned int size)
 737 {
 738         char *retptr;
 739
 740         retptr = bootmem_current;
 741         PANIC_IF(retptr + size > bootmem_end);
 742         bootmem_current += size;
 743
 744         return retptr;
 745 }
 746
 747 void
 748 bootmem_free(void *ptr, unsigned int size)
 749 {
 750         char *tptr;
 751
 752         tptr = ptr;
 753         PANIC_IF(tptr != bootmem_current - size ||
 754                 bootmem_current - size < bootmem_start);
 755
 756         bootmem_current -= size;
 757 }
 758
 759 #if 0
 760 static vm_paddr_t
 761 xpmap_mtop2(vm_paddr_t mpa)
 762 {
 763         return ((machine_to_phys_mapping[mpa >> PAGE_SHIFT] << PAGE_SHIFT)
 764             ) | (mpa & ~PG_FRAME);
 765 }
 766
 767 static pd_entry_t
 768 xpmap_get_bootpde(vm_paddr_t va)
 769 {
 770
 771         return ((pd_entry_t *)xen_start_info->pt_base)[va >> 22];
 772 }
 773
 774 static pd_entry_t
 775 xpmap_get_vbootpde(vm_paddr_t va)
 776 {
 777         pd_entry_t pde;
 778
 779         pde = xpmap_get_bootpde(va);
 780         if ((pde & PG_V) == 0)
 781                 return (pde & ~PG_FRAME);
 782         return (pde & ~PG_FRAME) |
 783                 (xpmap_mtop2(pde & PG_FRAME) + KERNBASE);
 784 }
 785
 786 static pt_entry_t 8*
 787 xpmap_get_bootptep(vm_paddr_t va)
 788 {
 789         pd_entry_t pde;
 790
 791         pde = xpmap_get_vbootpde(va);
 792         if ((pde & PG_V) == 0)
 793                 return (void *)-1;
 794 #define PT_MASK         0x003ff000      /* page table address bits */
 795         return &(((pt_entry_t *)(pde & PG_FRAME))[(va & PT_MASK) >> PAGE_SHIFT]);
 796 }
 797
 798 static pt_entry_t
 799 xpmap_get_bootpte(vm_paddr_t va)
 800 {
 801
 802         return xpmap_get_bootptep(va)[0];
 803 }
 804 #endif
 805
 806
 807 #ifdef ADD_ISA_HOLE
 808 static void
 809 shift_phys_machine(unsigned long *phys_machine, int nr_pages)
 810 {
 811
 812         unsigned long *tmp_page, *current_page, *next_page;
 813         int i;
 814
 815         tmp_page = bootmem_alloc(PAGE_SIZE);
 816         current_page = phys_machine + nr_pages - (PAGE_SIZE/sizeof(unsigned long));
 817         next_page = current_page - (PAGE_SIZE/sizeof(unsigned long));
 818         bcopy(phys_machine, tmp_page, PAGE_SIZE);
 819
 820         while (current_page > phys_machine) {
 821                 /*  save next page */
 822                 bcopy(next_page, tmp_page, PAGE_SIZE);
 823                 /* shift down page */
 824                 bcopy(current_page, next_page, PAGE_SIZE);
 825                 /*  finish swap */
 826                 bcopy(tmp_page, current_page, PAGE_SIZE);
 827
 828                 current_page -= (PAGE_SIZE/sizeof(unsigned long));
 829                 next_page -= (PAGE_SIZE/sizeof(unsigned long));
 830         }
 831         bootmem_free(tmp_page, PAGE_SIZE);
 832
 833         for (i = 0; i < nr_pages; i++) {
 834                 xen_machphys_update(phys_machine[i], i);
 835         }
 836         memset(phys_machine, INVALID_P2M_ENTRY, PAGE_SIZE);
 837
 838 }
 839 #endif /* ADD_ISA_HOLE */
 840
 841 /*
 842  * Build a directory of the pages that make up our Physical to Machine
 843  * mapping table. The Xen suspend/restore code uses this to find our
 844  * mapping table.
 845  */
 846 static void
 847 init_frame_list_list(void *arg)
 848 {
 849         unsigned long nr_pages = xen_start_info->nr_pages;
 850 #define FPP     (PAGE_SIZE/sizeof(xen_pfn_t))
 851         int i, j, k;
 852
 853         xen_pfn_to_mfn_frame_list_list = malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK);
 854         for (i = 0, j = 0, k = -1; i < nr_pages;
 855              i += FPP, j++) {
 856                 if ((j & (FPP - 1)) == 0) {
 857                         k++;
 858                         xen_pfn_to_mfn_frame_list[k] =
 859                                 malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK);
 860                         xen_pfn_to_mfn_frame_list_list[k] =
 861                                 VTOMFN(xen_pfn_to_mfn_frame_list[k]);
 862                         j = 0;
 863                 }
 864                 xen_pfn_to_mfn_frame_list[k][j] =
 865                         VTOMFN(&xen_phys_machine[i]);
 866         }
 867
 868         HYPERVISOR_shared_info->arch.max_pfn = nr_pages;
 869         HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list
 870                 = VTOMFN(xen_pfn_to_mfn_frame_list_list);
 871 }
 872 SYSINIT(init_fll, SI_SUB_DEVFS, SI_ORDER_ANY, init_frame_list_list, NULL);
 873
 874 extern unsigned long physfree;
 875
 876 int pdir, curoffset;
 877 extern int nkpt;
 878
 879 extern uint32_t kernbase;
 880
 881 void
 882 initvalues(start_info_t *startinfo)
 883 {
 884         vm_offset_t cur_space, cur_space_pt;
 885         struct physdev_set_iopl set_iopl;
 886
 887         int l3_pages, l2_pages, l1_pages, offset;
 888         vm_paddr_t console_page_ma, xen_store_ma;
 889         vm_offset_t tmpva;
 890         vm_paddr_t shinfo;
 891 #ifdef PAE
 892         vm_paddr_t IdlePDPTma, IdlePDPTnewma;
 893         vm_paddr_t IdlePTDnewma[4];
 894         pd_entry_t *IdlePDPTnew, *IdlePTDnew;
 895         vm_paddr_t IdlePTDma[4];
 896 #else
 897         vm_paddr_t IdlePTDma[1];
 898 #endif
 899         unsigned long i;
 900         int ncpus = MAXCPU;
 901
 902         nkpt = min(
 903                 min(
 904                         max((startinfo->nr_pages >> NPGPTD_SHIFT), nkpt),
 905                     NPGPTD*NPDEPG - KPTDI),
 906                     (HYPERVISOR_VIRT_START - KERNBASE) >> PDRSHIFT);
 907
 908         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 909 #ifdef notyet
 910         /*
 911          * need to install handler
 912          */
 913         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments_notify);
 914 #endif
 915         xen_start_info = startinfo;
 916         xen_phys_machine = (xen_pfn_t *)startinfo->mfn_list;
 917
 918         IdlePTD = (pd_entry_t *)((uint8_t *)startinfo->pt_base + PAGE_SIZE);
 919         l1_pages = 0;
 920
 921 #ifdef PAE
 922         l3_pages = 1;
 923         l2_pages = 0;
 924         IdlePDPT = (pd_entry_t *)startinfo->pt_base;
 925         IdlePDPTma = VTOM(startinfo->pt_base);
 926         for (i = (KERNBASE >> 30);
 927              (i < 4) && (IdlePDPT[i] != 0); i++)
 928                         l2_pages++;
 929         /*
 930          * Note that only one page directory has been allocated at this point.
 931          * Thus, if KERNBASE
 932          */
 933         for (i = 0; i < l2_pages; i++)
 934                 IdlePTDma[i] = VTOM(IdlePTD + i*PAGE_SIZE);
 935
 936         l2_pages = (l2_pages == 0) ? 1 : l2_pages;
 937 #else
 938         l3_pages = 0;
 939         l2_pages = 1;
 940 #endif
 941         for (i = (((KERNBASE>>18) & PAGE_MASK)>>PAGE_SHIFT);
 942              (i<l2_pages*NPDEPG) && (i<(VM_MAX_KERNEL_ADDRESS>>PDRSHIFT)); i++) {
 943
 944                 if (IdlePTD[i] == 0)
 945                         break;
 946                 l1_pages++;
 947         }
 948
 949         /* number of pages allocated after the pts + 1*/;
 950         cur_space = xen_start_info->pt_base +
 951             (l3_pages + l2_pages + l1_pages + 1)*PAGE_SIZE;
 952
 953         printk("initvalues(): wooh - availmem=%x,%x\n", avail_space, cur_space);
 954
 955         printk("KERNBASE=%x,pt_base=%x, VTOPFN(base)=%x, nr_pt_frames=%x\n",
 956             KERNBASE,xen_start_info->pt_base, VTOPFN(xen_start_info->pt_base),
 957             xen_start_info->nr_pt_frames);
 958         xendebug_flags = 0; /* 0xffffffff; */
 959
 960 #ifdef ADD_ISA_HOLE
 961         shift_phys_machine(xen_phys_machine, xen_start_info->nr_pages);
 962 #endif
 963         XENPRINTF("IdlePTD %p\n", IdlePTD);
 964         XENPRINTF("nr_pages: %ld shared_info: 0x%lx flags: 0x%lx pt_base: 0x%lx "
 965                   "mod_start: 0x%lx mod_len: 0x%lx\n",
 966                   xen_start_info->nr_pages, xen_start_info->shared_info,
 967                   xen_start_info->flags, xen_start_info->pt_base,
 968                   xen_start_info->mod_start, xen_start_info->mod_len);
 969
 970 #ifdef PAE
 971         IdlePDPTnew = (pd_entry_t *)cur_space; cur_space += PAGE_SIZE;
 972         bzero(IdlePDPTnew, PAGE_SIZE);
 973
 974         IdlePDPTnewma =  VTOM(IdlePDPTnew);
 975         IdlePTDnew = (pd_entry_t *)cur_space; cur_space += 4*PAGE_SIZE;
 976         bzero(IdlePTDnew, 4*PAGE_SIZE);
 977
 978         for (i = 0; i < 4; i++)
 979                 IdlePTDnewma[i] = VTOM((uint8_t *)IdlePTDnew + i*PAGE_SIZE);
 980         /*
 981          * L3
 982          *
 983          * Copy the 4 machine addresses of the new PTDs in to the PDPT
 984          *
 985          */
 986         for (i = 0; i < 4; i++)
 987                 IdlePDPTnew[i] = IdlePTDnewma[i] | PG_V;
 988
 989         __asm__("nop;");
 990         /*
 991          *
 992          * re-map the new PDPT read-only
 993          */
 994         PT_SET_MA(IdlePDPTnew, IdlePDPTnewma | PG_V);
 995         /*
 996          *
 997          * Unpin the current PDPT
 998          */
 999         xen_pt_unpin(IdlePDPTma);
1000
1001 #endif  /* PAE */
1002
1003         /* Map proc0's KSTACK */
1004         proc0kstack = cur_space; cur_space += (KSTACK_PAGES * PAGE_SIZE);
1005         printk("proc0kstack=%u\n", proc0kstack);
1006
1007         /* vm86/bios stack */
1008         cur_space += PAGE_SIZE;
1009
1010         /* Map space for the vm86 region */
1011         vm86paddr = (vm_offset_t)cur_space;
1012         cur_space += (PAGE_SIZE * 3);
1013
1014         /* allocate 4 pages for bootmem allocator */
1015         bootmem_start = bootmem_current = (char *)cur_space;
1016         cur_space += (4 * PAGE_SIZE);
1017         bootmem_end = (char *)cur_space;
1018
1019         /* allocate pages for gdt */
1020         gdt = (union descriptor *)cur_space;
1021         cur_space += PAGE_SIZE*ncpus;
1022
1023         /* allocate page for ldt */
1024         ldt = (union descriptor *)cur_space; cur_space += PAGE_SIZE;
1025         cur_space += PAGE_SIZE;
1026
1027         /* unmap remaining pages from initial chunk
1028          *
1029          */
1030         for (tmpva = cur_space; tmpva < (((uint32_t)&kernbase) + (l1_pages<<PDRSHIFT));
1031              tmpva += PAGE_SIZE) {
1032                 bzero((char *)tmpva, PAGE_SIZE);
1033                 PT_SET_MA(tmpva, (vm_paddr_t)0);
1034         }
1035
1036         PT_UPDATES_FLUSH();
1037
1038         memcpy(((uint8_t *)IdlePTDnew) + ((unsigned int)(KERNBASE >> 18)),
1039             ((uint8_t *)IdlePTD) + ((KERNBASE >> 18) & PAGE_MASK),
1040             l1_pages*sizeof(pt_entry_t));
1041
1042         for (i = 0; i < 4; i++) {
1043                 PT_SET_MA((uint8_t *)IdlePTDnew + i*PAGE_SIZE,
1044                     IdlePTDnewma[i] | PG_V);
1045         }
1046         xen_load_cr3(VTOP(IdlePDPTnew));
1047         xen_pgdpt_pin(VTOM(IdlePDPTnew));
1048
1049         /* allocate remainder of nkpt pages */
1050         cur_space_pt = cur_space;
1051         for (offset = (KERNBASE >> PDRSHIFT), i = l1_pages; i < nkpt;
1052              i++, cur_space += PAGE_SIZE) {
1053                 pdir = (offset + i) / NPDEPG;
1054                 curoffset = ((offset + i) % NPDEPG);
1055                 if (((offset + i) << PDRSHIFT) == VM_MAX_KERNEL_ADDRESS)
1056                         break;
1057
1058                 /*
1059                  * make sure that all the initial page table pages
1060                  * have been zeroed
1061                  */
1062                 PT_SET_MA(cur_space, VTOM(cur_space) | PG_V | PG_RW);
1063                 bzero((char *)cur_space, PAGE_SIZE);
1064                 PT_SET_MA(cur_space, (vm_paddr_t)0);
1065                 xen_pt_pin(VTOM(cur_space));
1066                 xen_queue_pt_update((vm_paddr_t)(IdlePTDnewma[pdir] +
1067                         curoffset*sizeof(vm_paddr_t)),
1068                     VTOM(cur_space) | PG_KERNEL);
1069                 PT_UPDATES_FLUSH();
1070         }
1071
1072         for (i = 0; i < 4; i++) {
1073                 pdir = (PTDPTDI + i) / NPDEPG;
1074                 curoffset = (PTDPTDI + i) % NPDEPG;
1075
1076                 xen_queue_pt_update((vm_paddr_t)(IdlePTDnewma[pdir] +
1077                         curoffset*sizeof(vm_paddr_t)),
1078                     IdlePTDnewma[i] | PG_V);
1079         }
1080
1081         PT_UPDATES_FLUSH();
1082
1083         IdlePTD = IdlePTDnew;
1084         IdlePDPT = IdlePDPTnew;
1085         IdlePDPTma = IdlePDPTnewma;
1086
1087         HYPERVISOR_shared_info = (shared_info_t *)cur_space;
1088         cur_space += PAGE_SIZE;
1089
1090         xen_store = (struct xenstore_domain_interface *)cur_space;
1091         cur_space += PAGE_SIZE;
1092
1093         console_page = (char *)cur_space;
1094         cur_space += PAGE_SIZE;
1095
1096         /*
1097          * shared_info is an unsigned long so this will randomly break if
1098          * it is allocated above 4GB - I guess people are used to that
1099          * sort of thing with Xen ... sigh
1100          */
1101         shinfo = xen_start_info->shared_info;
1102         PT_SET_MA(HYPERVISOR_shared_info, shinfo | PG_KERNEL);
1103
1104         printk("#4\n");
1105
1106         xen_store_ma = (((vm_paddr_t)xen_start_info->store_mfn) << PAGE_SHIFT);
1107         PT_SET_MA(xen_store, xen_store_ma | PG_KERNEL);
1108         console_page_ma = (((vm_paddr_t)xen_start_info->console.domU.mfn) << PAGE_SHIFT);
1109         PT_SET_MA(console_page, console_page_ma | PG_KERNEL);
1110
1111         printk("#5\n");
1112
1113         set_iopl.iopl = 1;
1114         PANIC_IF(HYPERVISOR_physdev_op(PHYSDEVOP_SET_IOPL, &set_iopl));
1115         printk("#6\n");
1116 #if 0
1117         /* add page table for KERNBASE */
1118         xen_queue_pt_update(IdlePTDma + KPTDI*sizeof(vm_paddr_t),
1119                             VTOM(cur_space) | PG_KERNEL);
1120         xen_flush_queue();
1121 #ifdef PAE
1122         xen_queue_pt_update(pdir_shadow_ma[3] + KPTDI*sizeof(vm_paddr_t),
1123                             VTOM(cur_space) | PG_V | PG_A);
1124 #else
1125         xen_queue_pt_update(pdir_shadow_ma + KPTDI*sizeof(vm_paddr_t),
1126                             VTOM(cur_space) | PG_V | PG_A);
1127 #endif
1128         xen_flush_queue();
1129         cur_space += PAGE_SIZE;
1130         printk("#6\n");
1131 #endif /* 0 */
1132 #ifdef notyet
1133         if (xen_start_info->flags & SIF_INITDOMAIN) {
1134                 /* Map first megabyte */
1135                 for (i = 0; i < (256 << PAGE_SHIFT); i += PAGE_SIZE)
1136                         PT_SET_MA(KERNBASE + i, i | PG_KERNEL | PG_NC_PCD);
1137                 xen_flush_queue();
1138         }
1139 #endif
1140         /*
1141          * re-map kernel text read-only
1142          *
1143          */
1144         for (i = (((vm_offset_t)&btext) & ~PAGE_MASK);
1145              i < (((vm_offset_t)&etext) & ~PAGE_MASK); i += PAGE_SIZE)
1146                 PT_SET_MA(i, VTOM(i) | PG_V | PG_A);
1147
1148         printk("#7\n");
1149         physfree = VTOP(cur_space);
1150         init_first = physfree >> PAGE_SHIFT;
1151         IdlePTD = (pd_entry_t *)VTOP(IdlePTD);
1152         IdlePDPT = (pd_entry_t *)VTOP(IdlePDPT);
1153         setup_xen_features();
1154         printk("#8, proc0kstack=%u\n", proc0kstack);
1155 }
1156
1157
1158 trap_info_t trap_table[] = {
1159         { 0,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(div)},
1160         { 1,   0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(dbg)},
1161         { 3,   3|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(bpt)},
1162         { 4,   3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(ofl)},
1163         /* This is UPL on Linux and KPL on BSD */
1164         { 5,   3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(bnd)},
1165         { 6,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(ill)},
1166         { 7,   0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(dna)},
1167         /*
1168          * { 8,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(XXX)},
1169          *   no handler for double fault
1170          */
1171         { 9,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(fpusegm)},
1172         {10,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(tss)},
1173         {11,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(missing)},
1174         {12,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(stk)},
1175         {13,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(prot)},
1176         {14,   0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(page)},
1177         {15,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(rsvd)},
1178         {16,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(fpu)},
1179         {17,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(align)},
1180         {18,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(mchk)},
1181         {19,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(xmm)},
1182         {0x80, 3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(int0x80_syscall)},
1183         {  0, 0,           0, 0 }
1184 };
1185
1186 /* Perform a multicall and check that individual calls succeeded. */
1187 int
1188 HYPERVISOR_multicall(struct multicall_entry * call_list, int nr_calls)
1189 {
1190         int ret = 0;
1191         int i;
1192
1193         /* Perform the multicall. */
1194         PANIC_IF(_HYPERVISOR_multicall(call_list, nr_calls));
1195
1196         /* Check the results of individual hypercalls. */
1197         for (i = 0; i < nr_calls; i++)
1198                 if (unlikely(call_list[i].result < 0))
1199                         ret++;
1200         if (unlikely(ret > 0))
1201                 panic("%d multicall(s) failed: cpu %d\n",
1202                     ret, smp_processor_id());
1203
1204         /* If we didn't panic already, everything succeeded. */
1205         return (0);
1206 }
1207
1208 /********** CODE WORTH KEEPING ABOVE HERE *****************/
1209
1210 void xen_failsafe_handler(void);
1211
1212 void
1213 xen_failsafe_handler(void)
1214 {
1215
1216         panic("xen_failsafe_handler called!\n");
1217 }
1218
1219 void xen_handle_thread_switch(struct pcb *pcb);
1220
1221 /* This is called by cpu_switch() when switching threads. */
1222 /* The pcb arg refers to the process control block of the */
1223 /* next thread which is to run */
1224 void
1225 xen_handle_thread_switch(struct pcb *pcb)
1226 {
1227     uint32_t *a = (uint32_t *)&PCPU_GET(fsgs_gdt)[0];
1228     uint32_t *b = (uint32_t *)&pcb->pcb_fsd;
1229     multicall_entry_t mcl[3];
1230     int i = 0;
1231
1232     /* Notify Xen of task switch */
1233     mcl[i].op = __HYPERVISOR_stack_switch;
1234     mcl[i].args[0] = GSEL(GDATA_SEL, SEL_KPL);
1235     mcl[i++].args[1] = (unsigned long)pcb;
1236
1237     /* Check for update of fsd */
1238     if (*a != *b || *(a+1) != *(b+1)) {
1239         mcl[i].op = __HYPERVISOR_update_descriptor;
1240         *(uint64_t *)&mcl[i].args[0] = vtomach((vm_offset_t)a);
1241         *(uint64_t *)&mcl[i++].args[2] = *(uint64_t *)b;
1242     }
1243
1244     a += 2;
1245     b += 2;
1246
1247     /* Check for update of gsd */
1248     if (*a != *b || *(a+1) != *(b+1)) {
1249         mcl[i].op = __HYPERVISOR_update_descriptor;
1250         *(uint64_t *)&mcl[i].args[0] = vtomach((vm_offset_t)a);
1251         *(uint64_t *)&mcl[i++].args[2] = *(uint64_t *)b;
1252     }
1253
1254     (void)HYPERVISOR_multicall(mcl, i);
1255 }