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 #ifdef INVARIANTS
 219 static struct mmu_log xpq_queue_log[XPQUEUE_SIZE];
 220 #endif
 221 static int xpq_idx = 0;
 222
 223 #define XPQ_QUEUE_LOG xpq_queue_log
 224 #define XPQ_QUEUE xpq_queue
 225 #define XPQ_IDX xpq_idx
 226 #define SET_VCPU()
 227 #endif /* !SMP */
 228
 229 #define XPQ_IDX_INC atomic_add_int(&XPQ_IDX, 1);
 230
 231 #if 0
 232 static void
 233 xen_dump_queue(void)
 234 {
 235         int _xpq_idx = XPQ_IDX;
 236         int i;
 237
 238         if (_xpq_idx <= 1)
 239                 return;
 240
 241         printk("xen_dump_queue(): %u entries\n", _xpq_idx);
 242         for (i = 0; i < _xpq_idx; i++) {
 243                 printk(" val: %llx ptr: %llx\n", XPQ_QUEUE[i].val, XPQ_QUEUE[i].ptr);
 244         }
 245 }
 246 #endif
 247
 248
 249 static __inline void
 250 _xen_flush_queue(void)
 251 {
 252         SET_VCPU();
 253         int _xpq_idx = XPQ_IDX;
 254         int error, i;
 255
 256 #ifdef INVARIANTS
 257         if (__predict_true(gdtset))
 258                 CRITICAL_ASSERT(curthread);
 259 #endif
 260
 261         XPQ_IDX = 0;
 262         /* Make sure index is cleared first to avoid double updates. */
 263         error = HYPERVISOR_mmu_update((mmu_update_t *)&XPQ_QUEUE,
 264                                       _xpq_idx, NULL, DOMID_SELF);
 265
 266 #if 0
 267         if (__predict_true(gdtset))
 268         for (i = _xpq_idx; i > 0;) {
 269                 if (i >= 3) {
 270                         CTR6(KTR_PMAP, "mmu:val: %lx ptr: %lx val: %lx "
 271                             "ptr: %lx val: %lx ptr: %lx",
 272                             (XPQ_QUEUE[i-1].val & 0xffffffff),
 273                             (XPQ_QUEUE[i-1].ptr & 0xffffffff),
 274                             (XPQ_QUEUE[i-2].val & 0xffffffff),
 275                             (XPQ_QUEUE[i-2].ptr & 0xffffffff),
 276                             (XPQ_QUEUE[i-3].val & 0xffffffff),
 277                             (XPQ_QUEUE[i-3].ptr & 0xffffffff));
 278                             i -= 3;
 279                 } else if (i == 2) {
 280                         CTR4(KTR_PMAP, "mmu: val: %lx ptr: %lx val: %lx ptr: %lx",
 281                             (XPQ_QUEUE[i-1].val & 0xffffffff),
 282                             (XPQ_QUEUE[i-1].ptr & 0xffffffff),
 283                             (XPQ_QUEUE[i-2].val & 0xffffffff),
 284                             (XPQ_QUEUE[i-2].ptr & 0xffffffff));
 285                         i = 0;
 286                 } else {
 287                         CTR2(KTR_PMAP, "mmu: val: %lx ptr: %lx",
 288                             (XPQ_QUEUE[i-1].val & 0xffffffff),
 289                             (XPQ_QUEUE[i-1].ptr & 0xffffffff));
 290                         i = 0;
 291                 }
 292         }
 293 #endif
 294         if (__predict_false(error < 0)) {
 295                 for (i = 0; i < _xpq_idx; i++)
 296                         printf("val: %llx ptr: %llx\n",
 297                             XPQ_QUEUE[i].val, XPQ_QUEUE[i].ptr);
 298                 panic("Failed to execute MMU updates: %d", error);
 299         }
 300
 301 }
 302
 303 void
 304 xen_flush_queue(void)
 305 {
 306         SET_VCPU();
 307
 308         if (__predict_true(gdtset))
 309                 critical_enter();
 310         if (XPQ_IDX != 0) _xen_flush_queue();
 311         if (__predict_true(gdtset))
 312                 critical_exit();
 313 }
 314
 315 static __inline void
 316 xen_increment_idx(void)
 317 {
 318         SET_VCPU();
 319
 320         XPQ_IDX++;
 321         if (__predict_false(XPQ_IDX == XPQUEUE_SIZE))
 322                 xen_flush_queue();
 323 }
 324
 325 void
 326 xen_check_queue(void)
 327 {
 328 #ifdef INVARIANTS
 329         SET_VCPU();
 330
 331         KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX));
 332 #endif
 333 }
 334
 335 void
 336 xen_invlpg(vm_offset_t va)
 337 {
 338         struct mmuext_op op;
 339         op.cmd = MMUEXT_INVLPG_ALL;
 340         op.arg1.linear_addr = va & ~PAGE_MASK;
 341         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 342 }
 343
 344 void
 345 xen_load_cr3(u_int val)
 346 {
 347         struct mmuext_op op;
 348 #ifdef INVARIANTS
 349         SET_VCPU();
 350
 351         KASSERT(XPQ_IDX == 0, ("pending operations XPQ_IDX=%d", XPQ_IDX));
 352 #endif
 353         op.cmd = MMUEXT_NEW_BASEPTR;
 354         op.arg1.mfn = xpmap_ptom(val) >> PAGE_SHIFT;
 355         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 356 }
 357
 358 #ifdef KTR
 359 static __inline u_int
 360 rebp(void)
 361 {
 362         u_int   data;
 363
 364         __asm __volatile("movl 4(%%ebp),%0" : "=r" (data));
 365         return (data);
 366 }
 367 #endif
 368
 369 u_int
 370 read_eflags(void)
 371 {
 372         vcpu_info_t *_vcpu;
 373         u_int eflags;
 374
 375         eflags = _read_eflags();
 376         _vcpu = &HYPERVISOR_shared_info->vcpu_info[smp_processor_id()];
 377         if (_vcpu->evtchn_upcall_mask)
 378                 eflags &= ~PSL_I;
 379
 380         return (eflags);
 381 }
 382
 383 void
 384 write_eflags(u_int eflags)
 385 {
 386         u_int intr;
 387
 388         CTR2(KTR_SPARE2, "%x xen_restore_flags eflags %x", rebp(), eflags);
 389         intr = ((eflags & PSL_I) == 0);
 390         __restore_flags(intr);
 391         _write_eflags(eflags);
 392 }
 393
 394 void
 395 xen_cli(void)
 396 {
 397         CTR1(KTR_SPARE2, "%x xen_cli disabling interrupts", rebp());
 398         __cli();
 399 }
 400
 401 void
 402 xen_sti(void)
 403 {
 404         CTR1(KTR_SPARE2, "%x xen_sti enabling interrupts", rebp());
 405         __sti();
 406 }
 407
 408 u_int
 409 xen_rcr2(void)
 410 {
 411
 412         return (HYPERVISOR_shared_info->vcpu_info[curcpu].arch.cr2);
 413 }
 414
 415 void
 416 _xen_machphys_update(vm_paddr_t mfn, vm_paddr_t pfn, char *file, int line)
 417 {
 418         SET_VCPU();
 419
 420         if (__predict_true(gdtset))
 421                 critical_enter();
 422         XPQ_QUEUE[XPQ_IDX].ptr = (mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE;
 423         XPQ_QUEUE[XPQ_IDX].val = pfn;
 424 #ifdef INVARIANTS
 425         XPQ_QUEUE_LOG[XPQ_IDX].file = file;
 426         XPQ_QUEUE_LOG[XPQ_IDX].line = line;
 427 #endif
 428         xen_increment_idx();
 429         if (__predict_true(gdtset))
 430                 critical_exit();
 431 }
 432
 433 void
 434 _xen_queue_pt_update(vm_paddr_t ptr, vm_paddr_t val, char *file, int line)
 435 {
 436         SET_VCPU();
 437
 438         if (__predict_true(gdtset))
 439                 mtx_assert(&vm_page_queue_mtx, MA_OWNED);
 440
 441         KASSERT((ptr & 7) == 0, ("misaligned update"));
 442
 443         if (__predict_true(gdtset))
 444                 critical_enter();
 445
 446         XPQ_QUEUE[XPQ_IDX].ptr = ((uint64_t)ptr) | MMU_NORMAL_PT_UPDATE;
 447         XPQ_QUEUE[XPQ_IDX].val = (uint64_t)val;
 448 #ifdef INVARIANTS
 449         XPQ_QUEUE_LOG[XPQ_IDX].file = file;
 450         XPQ_QUEUE_LOG[XPQ_IDX].line = line;
 451 #endif
 452         xen_increment_idx();
 453         if (__predict_true(gdtset))
 454                 critical_exit();
 455 }
 456
 457 void
 458 xen_pgdpt_pin(vm_paddr_t ma)
 459 {
 460         struct mmuext_op op;
 461         op.cmd = MMUEXT_PIN_L3_TABLE;
 462         op.arg1.mfn = ma >> PAGE_SHIFT;
 463         xen_flush_queue();
 464         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 465 }
 466
 467 void
 468 xen_pgd_pin(vm_paddr_t ma)
 469 {
 470         struct mmuext_op op;
 471         op.cmd = MMUEXT_PIN_L2_TABLE;
 472         op.arg1.mfn = ma >> PAGE_SHIFT;
 473         xen_flush_queue();
 474         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 475 }
 476
 477 void
 478 xen_pgd_unpin(vm_paddr_t ma)
 479 {
 480         struct mmuext_op op;
 481         op.cmd = MMUEXT_UNPIN_TABLE;
 482         op.arg1.mfn = ma >> PAGE_SHIFT;
 483         xen_flush_queue();
 484         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 485 }
 486
 487 void
 488 xen_pt_pin(vm_paddr_t ma)
 489 {
 490         struct mmuext_op op;
 491         op.cmd = MMUEXT_PIN_L1_TABLE;
 492         op.arg1.mfn = ma >> PAGE_SHIFT;
 493         xen_flush_queue();
 494         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 495 }
 496
 497 void
 498 xen_pt_unpin(vm_paddr_t ma)
 499 {
 500         struct mmuext_op op;
 501         op.cmd = MMUEXT_UNPIN_TABLE;
 502         op.arg1.mfn = ma >> PAGE_SHIFT;
 503         xen_flush_queue();
 504         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 505 }
 506
 507 void
 508 xen_set_ldt(vm_paddr_t ptr, unsigned long len)
 509 {
 510         struct mmuext_op op;
 511         op.cmd = MMUEXT_SET_LDT;
 512         op.arg1.linear_addr = ptr;
 513         op.arg2.nr_ents = len;
 514         xen_flush_queue();
 515         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 516 }
 517
 518 void xen_tlb_flush(void)
 519 {
 520         struct mmuext_op op;
 521         op.cmd = MMUEXT_TLB_FLUSH_LOCAL;
 522         xen_flush_queue();
 523         PANIC_IF(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 524 }
 525
 526 void
 527 xen_update_descriptor(union descriptor *table, union descriptor *entry)
 528 {
 529         vm_paddr_t pa;
 530         pt_entry_t *ptp;
 531
 532         ptp = vtopte((vm_offset_t)table);
 533         pa = (*ptp & PG_FRAME) | ((vm_offset_t)table & PAGE_MASK);
 534         if (HYPERVISOR_update_descriptor(pa, *(uint64_t *)entry))
 535                 panic("HYPERVISOR_update_descriptor failed\n");
 536 }
 537
 538
 539 #if 0
 540 /*
 541  * Bitmap is indexed by page number. If bit is set, the page is part of a
 542  * xen_create_contiguous_region() area of memory.
 543  */
 544 unsigned long *contiguous_bitmap;
 545
 546 static void
 547 contiguous_bitmap_set(unsigned long first_page, unsigned long nr_pages)
 548 {
 549         unsigned long start_off, end_off, curr_idx, end_idx;
 550
 551         curr_idx  = first_page / BITS_PER_LONG;
 552         start_off = first_page & (BITS_PER_LONG-1);
 553         end_idx   = (first_page + nr_pages) / BITS_PER_LONG;
 554         end_off   = (first_page + nr_pages) & (BITS_PER_LONG-1);
 555
 556         if (curr_idx == end_idx) {
 557                 contiguous_bitmap[curr_idx] |=
 558                         ((1UL<<end_off)-1) & -(1UL<<start_off);
 559         } else {
 560                 contiguous_bitmap[curr_idx] |= -(1UL<<start_off);
 561                 while ( ++curr_idx < end_idx )
 562                         contiguous_bitmap[curr_idx] = ~0UL;
 563                 contiguous_bitmap[curr_idx] |= (1UL<<end_off)-1;
 564         }
 565 }
 566
 567 static void
 568 contiguous_bitmap_clear(unsigned long first_page, unsigned long nr_pages)
 569 {
 570         unsigned long start_off, end_off, curr_idx, end_idx;
 571
 572         curr_idx  = first_page / BITS_PER_LONG;
 573         start_off = first_page & (BITS_PER_LONG-1);
 574         end_idx   = (first_page + nr_pages) / BITS_PER_LONG;
 575         end_off   = (first_page + nr_pages) & (BITS_PER_LONG-1);
 576
 577         if (curr_idx == end_idx) {
 578                 contiguous_bitmap[curr_idx] &=
 579                         -(1UL<<end_off) | ((1UL<<start_off)-1);
 580         } else {
 581                 contiguous_bitmap[curr_idx] &= (1UL<<start_off)-1;
 582                 while ( ++curr_idx != end_idx )
 583                         contiguous_bitmap[curr_idx] = 0;
 584                 contiguous_bitmap[curr_idx] &= -(1UL<<end_off);
 585         }
 586 }
 587 #endif
 588
 589 /* Ensure multi-page extents are contiguous in machine memory. */
 590 int
 591 xen_create_contiguous_region(vm_page_t pages, int npages)
 592 {
 593         unsigned long  mfn, i, flags;
 594         int order;
 595         struct xen_memory_reservation reservation = {
 596                 .nr_extents   = 1,
 597                 .extent_order = 0,
 598                 .domid        = DOMID_SELF
 599         };
 600         set_xen_guest_handle(reservation.extent_start, &mfn);
 601
 602         balloon_lock(flags);
 603
 604         /* can currently only handle power of two allocation */
 605         PANIC_IF(ffs(npages) != fls(npages));
 606
 607         /* 0. determine order */
 608         order = (ffs(npages) == fls(npages)) ? fls(npages) - 1 : fls(npages);
 609
 610         /* 1. give away machine pages. */
 611         for (i = 0; i < (1 << order); i++) {
 612                 int pfn;
 613                 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
 614                 mfn = PFNTOMFN(pfn);
 615                 PFNTOMFN(pfn) = INVALID_P2M_ENTRY;
 616                 PANIC_IF(HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation) != 1);
 617         }
 618
 619
 620         /* 2. Get a new contiguous memory extent. */
 621         reservation.extent_order = order;
 622         /* xenlinux hardcodes this because of aacraid - maybe set to 0 if we're not
 623          * running with a broxen driver XXXEN
 624          */
 625         reservation.address_bits = 31;
 626         if (HYPERVISOR_memory_op(XENMEM_increase_reservation, &reservation) != 1)
 627                 goto fail;
 628
 629         /* 3. Map the new extent in place of old pages. */
 630         for (i = 0; i < (1 << order); i++) {
 631                 int pfn;
 632                 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
 633                 xen_machphys_update(mfn+i, pfn);
 634                 PFNTOMFN(pfn) = mfn+i;
 635         }
 636
 637         xen_tlb_flush();
 638
 639 #if 0
 640         contiguous_bitmap_set(VM_PAGE_TO_PHYS(&pages[0]) >> PAGE_SHIFT, 1UL << order);
 641 #endif
 642
 643         balloon_unlock(flags);
 644
 645         return 0;
 646
 647  fail:
 648         reservation.extent_order = 0;
 649         reservation.address_bits = 0;
 650
 651         for (i = 0; i < (1 << order); i++) {
 652                 int pfn;
 653                 pfn = VM_PAGE_TO_PHYS(&pages[i]) >> PAGE_SHIFT;
 654                 PANIC_IF(HYPERVISOR_memory_op(
 655                         XENMEM_increase_reservation, &reservation) != 1);
 656                 xen_machphys_update(mfn, pfn);
 657                 PFNTOMFN(pfn) = mfn;
 658         }
 659
 660         xen_tlb_flush();
 661
 662         balloon_unlock(flags);
 663
 664         return ENOMEM;
 665 }
 666
 667 void
 668 xen_destroy_contiguous_region(void *addr, int npages)
 669 {
 670         unsigned long  mfn, i, flags, order, pfn0;
 671         struct xen_memory_reservation reservation = {
 672                 .nr_extents   = 1,
 673                 .extent_order = 0,
 674                 .domid        = DOMID_SELF
 675         };
 676         set_xen_guest_handle(reservation.extent_start, &mfn);
 677
 678         pfn0 = vtophys(addr) >> PAGE_SHIFT;
 679 #if 0
 680         scrub_pages(vstart, 1 << order);
 681 #endif
 682         /* can currently only handle power of two allocation */
 683         PANIC_IF(ffs(npages) != fls(npages));
 684
 685         /* 0. determine order */
 686         order = (ffs(npages) == fls(npages)) ? fls(npages) - 1 : fls(npages);
 687
 688         balloon_lock(flags);
 689
 690 #if 0
 691         contiguous_bitmap_clear(vtophys(addr) >> PAGE_SHIFT, 1UL << order);
 692 #endif
 693
 694         /* 1. Zap current PTEs, giving away the underlying pages. */
 695         for (i = 0; i < (1 << order); i++) {
 696                 int pfn;
 697                 uint64_t new_val = 0;
 698                 pfn = vtomach((char *)addr + i*PAGE_SIZE) >> PAGE_SHIFT;
 699
 700                 PANIC_IF(HYPERVISOR_update_va_mapping((vm_offset_t)((char *)addr + (i * PAGE_SIZE)), new_val, 0));
 701                 PFNTOMFN(pfn) = INVALID_P2M_ENTRY;
 702                 PANIC_IF(HYPERVISOR_memory_op(
 703                         XENMEM_decrease_reservation, &reservation) != 1);
 704         }
 705
 706         /* 2. Map new pages in place of old pages. */
 707         for (i = 0; i < (1 << order); i++) {
 708                 int pfn;
 709                 uint64_t new_val;
 710                 pfn = pfn0 + i;
 711                 PANIC_IF(HYPERVISOR_memory_op(XENMEM_increase_reservation, &reservation) != 1);
 712
 713                 new_val = mfn << PAGE_SHIFT;
 714                 PANIC_IF(HYPERVISOR_update_va_mapping((vm_offset_t)addr + (i * PAGE_SIZE),
 715                                                       new_val, PG_KERNEL));
 716                 xen_machphys_update(mfn, pfn);
 717                 PFNTOMFN(pfn) = mfn;
 718         }
 719
 720         xen_tlb_flush();
 721
 722         balloon_unlock(flags);
 723 }
 724
 725 extern unsigned long cpu0prvpage;
 726 extern unsigned long *SMPpt;
 727 extern  struct user     *proc0uarea;
 728 extern  vm_offset_t     proc0kstack;
 729 extern int vm86paddr, vm86phystk;
 730 char *bootmem_start, *bootmem_current, *bootmem_end;
 731
 732 pteinfo_t *pteinfo_list;
 733 void initvalues(start_info_t *startinfo);
 734
 735 struct xenstore_domain_interface;
 736 extern struct xenstore_domain_interface *xen_store;
 737
 738 char *console_page;
 739
 740 void *
 741 bootmem_alloc(unsigned int size)
 742 {
 743         char *retptr;
 744
 745         retptr = bootmem_current;
 746         PANIC_IF(retptr + size > bootmem_end);
 747         bootmem_current += size;
 748
 749         return retptr;
 750 }
 751
 752 void
 753 bootmem_free(void *ptr, unsigned int size)
 754 {
 755         char *tptr;
 756
 757         tptr = ptr;
 758         PANIC_IF(tptr != bootmem_current - size ||
 759                 bootmem_current - size < bootmem_start);
 760
 761         bootmem_current -= size;
 762 }
 763
 764 #if 0
 765 static vm_paddr_t
 766 xpmap_mtop2(vm_paddr_t mpa)
 767 {
 768         return ((machine_to_phys_mapping[mpa >> PAGE_SHIFT] << PAGE_SHIFT)
 769             ) | (mpa & ~PG_FRAME);
 770 }
 771
 772 static pd_entry_t
 773 xpmap_get_bootpde(vm_paddr_t va)
 774 {
 775
 776         return ((pd_entry_t *)xen_start_info->pt_base)[va >> 22];
 777 }
 778
 779 static pd_entry_t
 780 xpmap_get_vbootpde(vm_paddr_t va)
 781 {
 782         pd_entry_t pde;
 783
 784         pde = xpmap_get_bootpde(va);
 785         if ((pde & PG_V) == 0)
 786                 return (pde & ~PG_FRAME);
 787         return (pde & ~PG_FRAME) |
 788                 (xpmap_mtop2(pde & PG_FRAME) + KERNBASE);
 789 }
 790
 791 static pt_entry_t 8*
 792 xpmap_get_bootptep(vm_paddr_t va)
 793 {
 794         pd_entry_t pde;
 795
 796         pde = xpmap_get_vbootpde(va);
 797         if ((pde & PG_V) == 0)
 798                 return (void *)-1;
 799 #define PT_MASK         0x003ff000      /* page table address bits */
 800         return &(((pt_entry_t *)(pde & PG_FRAME))[(va & PT_MASK) >> PAGE_SHIFT]);
 801 }
 802
 803 static pt_entry_t
 804 xpmap_get_bootpte(vm_paddr_t va)
 805 {
 806
 807         return xpmap_get_bootptep(va)[0];
 808 }
 809 #endif
 810
 811
 812 #ifdef ADD_ISA_HOLE
 813 static void
 814 shift_phys_machine(unsigned long *phys_machine, int nr_pages)
 815 {
 816
 817         unsigned long *tmp_page, *current_page, *next_page;
 818         int i;
 819
 820         tmp_page = bootmem_alloc(PAGE_SIZE);
 821         current_page = phys_machine + nr_pages - (PAGE_SIZE/sizeof(unsigned long));
 822         next_page = current_page - (PAGE_SIZE/sizeof(unsigned long));
 823         bcopy(phys_machine, tmp_page, PAGE_SIZE);
 824
 825         while (current_page > phys_machine) {
 826                 /*  save next page */
 827                 bcopy(next_page, tmp_page, PAGE_SIZE);
 828                 /* shift down page */
 829                 bcopy(current_page, next_page, PAGE_SIZE);
 830                 /*  finish swap */
 831                 bcopy(tmp_page, current_page, PAGE_SIZE);
 832
 833                 current_page -= (PAGE_SIZE/sizeof(unsigned long));
 834                 next_page -= (PAGE_SIZE/sizeof(unsigned long));
 835         }
 836         bootmem_free(tmp_page, PAGE_SIZE);
 837
 838         for (i = 0; i < nr_pages; i++) {
 839                 xen_machphys_update(phys_machine[i], i);
 840         }
 841         memset(phys_machine, INVALID_P2M_ENTRY, PAGE_SIZE);
 842
 843 }
 844 #endif /* ADD_ISA_HOLE */
 845
 846 /*
 847  * Build a directory of the pages that make up our Physical to Machine
 848  * mapping table. The Xen suspend/restore code uses this to find our
 849  * mapping table.
 850  */
 851 static void
 852 init_frame_list_list(void *arg)
 853 {
 854         unsigned long nr_pages = xen_start_info->nr_pages;
 855 #define FPP     (PAGE_SIZE/sizeof(xen_pfn_t))
 856         int i, j, k;
 857
 858         xen_pfn_to_mfn_frame_list_list = malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK);
 859         for (i = 0, j = 0, k = -1; i < nr_pages;
 860              i += FPP, j++) {
 861                 if ((j & (FPP - 1)) == 0) {
 862                         k++;
 863                         xen_pfn_to_mfn_frame_list[k] =
 864                                 malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK);
 865                         xen_pfn_to_mfn_frame_list_list[k] =
 866                                 VTOMFN(xen_pfn_to_mfn_frame_list[k]);
 867                         j = 0;
 868                 }
 869                 xen_pfn_to_mfn_frame_list[k][j] =
 870                         VTOMFN(&xen_phys_machine[i]);
 871         }
 872
 873         HYPERVISOR_shared_info->arch.max_pfn = nr_pages;
 874         HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list
 875                 = VTOMFN(xen_pfn_to_mfn_frame_list_list);
 876 }
 877 SYSINIT(init_fll, SI_SUB_DEVFS, SI_ORDER_ANY, init_frame_list_list, NULL);
 878
 879 extern unsigned long physfree;
 880
 881 int pdir, curoffset;
 882 extern int nkpt;
 883
 884 void
 885 initvalues(start_info_t *startinfo)
 886 {
 887         int l3_pages, l2_pages, l1_pages, offset;
 888         vm_offset_t cur_space, cur_space_pt;
 889         struct physdev_set_iopl set_iopl;
 890
 891         vm_paddr_t KPTphys, IdlePTDma;
 892         vm_paddr_t console_page_ma, xen_store_ma;
 893         vm_offset_t KPTphysoff, tmpva;
 894         vm_paddr_t shinfo;
 895 #ifdef PAE
 896         vm_paddr_t IdlePDPTma, IdlePDPTnewma;
 897         vm_paddr_t IdlePTDnewma[4];
 898         pd_entry_t *IdlePDPTnew, *IdlePTDnew;
 899 #else
 900         vm_paddr_t pdir_shadow_ma;
 901 #endif
 902         unsigned long i;
 903         int ncpus = MAXCPU;
 904
 905         nkpt = min(
 906                 min(
 907                         max((startinfo->nr_pages >> NPGPTD_SHIFT), nkpt),
 908                     NPGPTD*NPDEPG - KPTDI),
 909                     (HYPERVISOR_VIRT_START - KERNBASE) >> PDRSHIFT);
 910
 911         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
 912 #ifdef notyet
 913         /*
 914          * need to install handler
 915          */
 916         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments_notify);
 917 #endif
 918         xen_start_info = startinfo;
 919         xen_phys_machine = (xen_pfn_t *)startinfo->mfn_list;
 920
 921         IdlePTD = (pd_entry_t *)((uint8_t *)startinfo->pt_base + PAGE_SIZE);
 922         l1_pages = 0;
 923
 924 #ifdef PAE
 925         l3_pages = 1;
 926         l2_pages = 0;
 927         IdlePDPT = (pd_entry_t *)startinfo->pt_base;
 928         IdlePDPTma = xpmap_ptom(VTOP(startinfo->pt_base));
 929         for (i = (KERNBASE >> 30);
 930              (i < 4) && (IdlePDPT[i] != 0); i++)
 931                         l2_pages++;
 932         /*
 933          * Note that only one page directory has been allocated at this point.
 934          * Thus, if KERNBASE
 935          */
 936 #if 0
 937         for (i = 0; i < l2_pages; i++)
 938                 IdlePTDma[i] = xpmap_ptom(VTOP(IdlePTD + i*PAGE_SIZE));
 939 #endif
 940
 941         l2_pages = (l2_pages == 0) ? 1 : l2_pages;
 942 #else
 943         l3_pages = 0;
 944         l2_pages = 1;
 945 #endif
 946         for (i = (((KERNBASE>>18) & PAGE_MASK)>>PAGE_SHIFT);
 947              (i<l2_pages*NPDEPG) && (i<(VM_MAX_KERNEL_ADDRESS>>PDRSHIFT)); i++) {
 948
 949                 if (IdlePTD[i] == 0)
 950                         break;
 951                 l1_pages++;
 952         }
 953
 954         /* number of pages allocated after the pts + 1*/;
 955         cur_space = xen_start_info->pt_base +
 956             ((xen_start_info->nr_pt_frames) + 3 )*PAGE_SIZE;
 957         printk("initvalues(): wooh - availmem=%x,%x\n", avail_space, cur_space);
 958
 959         printk("KERNBASE=%x,pt_base=%x, VTOPFN(base)=%x, nr_pt_frames=%x\n",
 960             KERNBASE,xen_start_info->pt_base, VTOPFN(xen_start_info->pt_base),
 961             xen_start_info->nr_pt_frames);
 962         xendebug_flags = 0; /* 0xffffffff; */
 963
 964         /* allocate 4 pages for bootmem allocator */
 965         bootmem_start = bootmem_current = (char *)cur_space;
 966         cur_space += (4 * PAGE_SIZE);
 967         bootmem_end = (char *)cur_space;
 968
 969         /* allocate page for gdt */
 970         gdt = (union descriptor *)cur_space;
 971         cur_space += PAGE_SIZE*ncpus;
 972
 973         /* allocate page for ldt */
 974         ldt = (union descriptor *)cur_space; cur_space += PAGE_SIZE;
 975         cur_space += PAGE_SIZE;
 976
 977         HYPERVISOR_shared_info = (shared_info_t *)cur_space;
 978         cur_space += PAGE_SIZE;
 979
 980         xen_store = (struct xenstore_domain_interface *)cur_space;
 981         cur_space += PAGE_SIZE;
 982
 983         console_page = (char *)cur_space;
 984         cur_space += PAGE_SIZE;
 985
 986 #ifdef ADD_ISA_HOLE
 987         shift_phys_machine(xen_phys_machine, xen_start_info->nr_pages);
 988 #endif
 989         /*
 990          * pre-zero unused mapped pages - mapped on 4MB boundary
 991          */
 992 #ifdef PAE
 993         IdlePDPT = (pd_entry_t *)startinfo->pt_base;
 994         IdlePDPTma = xpmap_ptom(VTOP(startinfo->pt_base));
 995         /*
 996          * Note that only one page directory has been allocated at this point.
 997          * Thus, if KERNBASE
 998          */
 999         IdlePTD = (pd_entry_t *)((uint8_t *)startinfo->pt_base + PAGE_SIZE);
1000         IdlePTDma = xpmap_ptom(VTOP(IdlePTD));
1001         l3_pages = 1;
1002 #else
1003         IdlePTD = (pd_entry_t *)startinfo->pt_base;
1004         IdlePTDma = xpmap_ptom(VTOP(startinfo->pt_base));
1005         l3_pages = 0;
1006 #endif
1007         l2_pages = 1;
1008         l1_pages = xen_start_info->nr_pt_frames - l2_pages - l3_pages;
1009
1010         KPTphysoff = (l2_pages + l3_pages)*PAGE_SIZE;
1011
1012         KPTphys = xpmap_ptom(VTOP(startinfo->pt_base + KPTphysoff));
1013         XENPRINTF("IdlePTD %p\n", IdlePTD);
1014         XENPRINTF("nr_pages: %ld shared_info: 0x%lx flags: 0x%lx pt_base: 0x%lx "
1015                   "mod_start: 0x%lx mod_len: 0x%lx\n",
1016                   xen_start_info->nr_pages, xen_start_info->shared_info,
1017                   xen_start_info->flags, xen_start_info->pt_base,
1018                   xen_start_info->mod_start, xen_start_info->mod_len);
1019         /* Map proc0's KSTACK */
1020
1021         proc0kstack = cur_space; cur_space += (KSTACK_PAGES * PAGE_SIZE);
1022         printk("proc0kstack=%u\n", proc0kstack);
1023
1024         /* vm86/bios stack */
1025         cur_space += PAGE_SIZE;
1026
1027         /* Map space for the vm86 region */
1028         vm86paddr = (vm_offset_t)cur_space;
1029         cur_space += (PAGE_SIZE * 3);
1030
1031 #ifdef PAE
1032         IdlePDPTnew = (pd_entry_t *)cur_space; cur_space += PAGE_SIZE;
1033         bzero(IdlePDPTnew, PAGE_SIZE);
1034
1035         IdlePDPTnewma =  xpmap_ptom(VTOP(IdlePDPTnew));
1036         IdlePTDnew = (pd_entry_t *)cur_space; cur_space += 4*PAGE_SIZE;
1037         bzero(IdlePTDnew, 4*PAGE_SIZE);
1038
1039         for (i = 0; i < 4; i++)
1040                 IdlePTDnewma[i] =
1041                     xpmap_ptom(VTOP((uint8_t *)IdlePTDnew + i*PAGE_SIZE));
1042         /*
1043          * L3
1044          *
1045          * Copy the 4 machine addresses of the new PTDs in to the PDPT
1046          *
1047          */
1048         for (i = 0; i < 4; i++)
1049                 IdlePDPTnew[i] = IdlePTDnewma[i] | PG_V;
1050
1051         __asm__("nop;");
1052         /*
1053          *
1054          * re-map the new PDPT read-only
1055          */
1056         PT_SET_MA(IdlePDPTnew, IdlePDPTnewma | PG_V);
1057         /*
1058          *
1059          * Unpin the current PDPT
1060          */
1061         xen_pt_unpin(IdlePDPTma);
1062
1063         for (i = 0; i < 20; i++) {
1064                 int startidx = ((KERNBASE >> 18) & PAGE_MASK) >> 3;
1065
1066                 if (IdlePTD[startidx + i] == 0) {
1067                         l1_pages = i;
1068                         break;
1069                 }
1070         }
1071
1072 #endif  /* PAE */
1073
1074         /* unmap remaining pages from initial 4MB chunk
1075          *
1076          */
1077         for (tmpva = cur_space; (tmpva & ((1<<22)-1)) != 0; tmpva += PAGE_SIZE) {
1078                 bzero((char *)tmpva, PAGE_SIZE);
1079                 PT_SET_MA(tmpva, (vm_paddr_t)0);
1080         }
1081
1082         PT_UPDATES_FLUSH();
1083
1084         memcpy(((uint8_t *)IdlePTDnew) + ((unsigned int)(KERNBASE >> 18)),
1085             ((uint8_t *)IdlePTD) + ((KERNBASE >> 18) & PAGE_MASK),
1086             l1_pages*sizeof(pt_entry_t));
1087
1088         for (i = 0; i < 4; i++) {
1089                 PT_SET_MA((uint8_t *)IdlePTDnew + i*PAGE_SIZE,
1090                     IdlePTDnewma[i] | PG_V);
1091         }
1092         xen_load_cr3(VTOP(IdlePDPTnew));
1093         xen_pgdpt_pin(xpmap_ptom(VTOP(IdlePDPTnew)));
1094
1095         /* allocate remainder of nkpt pages */
1096         cur_space_pt = cur_space;
1097         for (offset = (KERNBASE >> PDRSHIFT), i = l1_pages; i < nkpt;
1098              i++, cur_space += PAGE_SIZE) {
1099                 pdir = (offset + i) / NPDEPG;
1100                 curoffset = ((offset + i) % NPDEPG);
1101                 if (((offset + i) << PDRSHIFT) == VM_MAX_KERNEL_ADDRESS)
1102                         break;
1103
1104                 /*
1105                  * make sure that all the initial page table pages
1106                  * have been zeroed
1107                  */
1108                 PT_SET_MA(cur_space_pt,
1109                     xpmap_ptom(VTOP(cur_space)) | PG_V | PG_RW);
1110                 bzero((char *)cur_space_pt, PAGE_SIZE);
1111                 PT_SET_MA(cur_space_pt, (vm_paddr_t)0);
1112                 xen_pt_pin(xpmap_ptom(VTOP(cur_space)));
1113                 xen_queue_pt_update((vm_paddr_t)(IdlePTDnewma[pdir] +
1114                         curoffset*sizeof(vm_paddr_t)),
1115                     xpmap_ptom(VTOP(cur_space)) | PG_KERNEL);
1116                 PT_UPDATES_FLUSH();
1117         }
1118
1119         for (i = 0; i < 4; i++) {
1120                 pdir = (PTDPTDI + i) / NPDEPG;
1121                 curoffset = (PTDPTDI + i) % NPDEPG;
1122
1123                 xen_queue_pt_update((vm_paddr_t)(IdlePTDnewma[pdir] +
1124                         curoffset*sizeof(vm_paddr_t)),
1125                     IdlePTDnewma[i] | PG_V);
1126         }
1127
1128         PT_UPDATES_FLUSH();
1129
1130         IdlePTD = IdlePTDnew;
1131         IdlePDPT = IdlePDPTnew;
1132         IdlePDPTma = IdlePDPTnewma;
1133
1134         /*
1135          * shared_info is an unsigned long so this will randomly break if
1136          * it is allocated above 4GB - I guess people are used to that
1137          * sort of thing with Xen ... sigh
1138          */
1139         shinfo = xen_start_info->shared_info;
1140         PT_SET_MA(HYPERVISOR_shared_info, shinfo | PG_KERNEL);
1141
1142         printk("#4\n");
1143
1144         xen_store_ma = (((vm_paddr_t)xen_start_info->store_mfn) << PAGE_SHIFT);
1145         PT_SET_MA(xen_store, xen_store_ma | PG_KERNEL);
1146         console_page_ma = (((vm_paddr_t)xen_start_info->console.domU.mfn) << PAGE_SHIFT);
1147         PT_SET_MA(console_page, console_page_ma | PG_KERNEL);
1148
1149         printk("#5\n");
1150
1151         set_iopl.iopl = 1;
1152         PANIC_IF(HYPERVISOR_physdev_op(PHYSDEVOP_SET_IOPL, &set_iopl));
1153         printk("#6\n");
1154 #if 0
1155         /* add page table for KERNBASE */
1156         xen_queue_pt_update(IdlePTDma + KPTDI*sizeof(vm_paddr_t),
1157                             xpmap_ptom(VTOP(cur_space) | PG_KERNEL));
1158         xen_flush_queue();
1159 #ifdef PAE
1160         xen_queue_pt_update(pdir_shadow_ma[3] + KPTDI*sizeof(vm_paddr_t),
1161                             xpmap_ptom(VTOP(cur_space) | PG_V | PG_A));
1162 #else
1163         xen_queue_pt_update(pdir_shadow_ma + KPTDI*sizeof(vm_paddr_t),
1164                             xpmap_ptom(VTOP(cur_space) | PG_V | PG_A));
1165 #endif
1166         xen_flush_queue();
1167         cur_space += PAGE_SIZE;
1168         printk("#6\n");
1169 #endif /* 0 */
1170 #ifdef notyet
1171         if (xen_start_info->flags & SIF_INITDOMAIN) {
1172                 /* Map first megabyte */
1173                 for (i = 0; i < (256 << PAGE_SHIFT); i += PAGE_SIZE)
1174                         PT_SET_MA(KERNBASE + i, i | PG_KERNEL | PG_NC_PCD);
1175                 xen_flush_queue();
1176         }
1177 #endif
1178         /*
1179          * re-map kernel text read-only
1180          *
1181          */
1182         for (i = (((vm_offset_t)&btext) & ~PAGE_MASK);
1183              i < (((vm_offset_t)&etext) & ~PAGE_MASK); i += PAGE_SIZE)
1184                 PT_SET_MA(i, xpmap_ptom(VTOP(i)) | PG_V | PG_A);
1185
1186         printk("#7\n");
1187         physfree = VTOP(cur_space);
1188         init_first = physfree >> PAGE_SHIFT;
1189         IdlePTD = (pd_entry_t *)VTOP(IdlePTD);
1190         IdlePDPT = (pd_entry_t *)VTOP(IdlePDPT);
1191         setup_xen_features();
1192         printk("#8, proc0kstack=%u\n", proc0kstack);
1193 }
1194
1195
1196 trap_info_t trap_table[] = {
1197         { 0,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(div)},
1198         { 1,   0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(dbg)},
1199         { 3,   3|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(bpt)},
1200         { 4,   3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(ofl)},
1201         /* This is UPL on Linux and KPL on BSD */
1202         { 5,   3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(bnd)},
1203         { 6,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(ill)},
1204         { 7,   0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(dna)},
1205         /*
1206          * { 8,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(XXX)},
1207          *   no handler for double fault
1208          */
1209         { 9,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(fpusegm)},
1210         {10,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(tss)},
1211         {11,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(missing)},
1212         {12,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(stk)},
1213         {13,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(prot)},
1214         {14,   0|4, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(page)},
1215         {15,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(rsvd)},
1216         {16,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(fpu)},
1217         {17,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(align)},
1218         {18,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(mchk)},
1219         {19,   0, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(xmm)},
1220         {0x80, 3, GSEL(GCODE_SEL, SEL_KPL), (unsigned long) &IDTVEC(int0x80_syscall)},
1221         {  0, 0,           0, 0 }
1222 };
1223
1224 /* Perform a multicall and check that individual calls succeeded. */
1225 int
1226 HYPERVISOR_multicall(struct multicall_entry * call_list, int nr_calls)
1227 {
1228         int ret = 0;
1229         int i;
1230
1231         /* Perform the multicall. */
1232         PANIC_IF(_HYPERVISOR_multicall(call_list, nr_calls));
1233
1234         /* Check the results of individual hypercalls. */
1235         for (i = 0; i < nr_calls; i++)
1236                 if (unlikely(call_list[i].result < 0))
1237                         ret++;
1238         if (unlikely(ret > 0))
1239                 panic("%d multicall(s) failed: cpu %d\n",
1240                     ret, smp_processor_id());
1241
1242         /* If we didn't panic already, everything succeeded. */
1243         return (0);
1244 }
1245
1246 /********** CODE WORTH KEEPING ABOVE HERE *****************/
1247
1248 void xen_failsafe_handler(void);
1249
1250 void
1251 xen_failsafe_handler(void)
1252 {
1253
1254         panic("xen_failsafe_handler called!\n");
1255 }
1256
1257 void xen_handle_thread_switch(struct pcb *pcb);
1258
1259 /* This is called by cpu_switch() when switching threads. */
1260 /* The pcb arg refers to the process control block of the */
1261 /* next thread which is to run */
1262 void
1263 xen_handle_thread_switch(struct pcb *pcb)
1264 {
1265     uint32_t *a = (uint32_t *)&PCPU_GET(fsgs_gdt)[0];
1266     uint32_t *b = (uint32_t *)&pcb->pcb_fsd;
1267     multicall_entry_t mcl[3];
1268     int i = 0;
1269
1270     /* Notify Xen of task switch */
1271     mcl[i].op = __HYPERVISOR_stack_switch;
1272     mcl[i].args[0] = GSEL(GDATA_SEL, SEL_KPL);
1273     mcl[i++].args[1] = (unsigned long)pcb;
1274
1275     /* Check for update of fsd */
1276     if (*a != *b || *(a+1) != *(b+1)) {
1277         mcl[i].op = __HYPERVISOR_update_descriptor;
1278         *(uint64_t *)&mcl[i].args[0] = vtomach((vm_offset_t)a);
1279         *(uint64_t *)&mcl[i++].args[2] = *(uint64_t *)b;
1280     }
1281
1282     a += 2;
1283     b += 2;
1284
1285     /* Check for update of gsd */
1286     if (*a != *b || *(a+1) != *(b+1)) {
1287         mcl[i].op = __HYPERVISOR_update_descriptor;
1288         *(uint64_t *)&mcl[i].args[0] = vtomach((vm_offset_t)a);
1289         *(uint64_t *)&mcl[i++].args[2] = *(uint64_t *)b;
1290     }
1291
1292     (void)HYPERVISOR_multicall(mcl, i);
1293 }