sys/ia64/ia64/machdep.c

   1 /*-
   2  * Copyright (c) 2003,2004 Marcel Moolenaar
   3  * Copyright (c) 2000,2001 Doug Rabson
   4  * All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  25  * SUCH DAMAGE.
  26  */
  27
  28 #include <sys/cdefs.h>
  29 __FBSDID("$FreeBSD$");
  30
  31 #include "opt_compat.h"
  32 #include "opt_ddb.h"
  33 #include "opt_kstack_pages.h"
  34 #include "opt_sched.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/proc.h>
  38 #include <sys/systm.h>
  39 #include <sys/bio.h>
  40 #include <sys/buf.h>
  41 #include <sys/bus.h>
  42 #include <sys/cons.h>
  43 #include <sys/cpu.h>
  44 #include <sys/eventhandler.h>
  45 #include <sys/exec.h>
  46 #include <sys/imgact.h>
  47 #include <sys/kdb.h>
  48 #include <sys/kernel.h>
  49 #include <sys/linker.h>
  50 #include <sys/lock.h>
  51 #include <sys/malloc.h>
  52 #include <sys/mbuf.h>
  53 #include <sys/msgbuf.h>
  54 #include <sys/pcpu.h>
  55 #include <sys/ptrace.h>
  56 #include <sys/random.h>
  57 #include <sys/reboot.h>
  58 #include <sys/sched.h>
  59 #include <sys/signalvar.h>
  60 #include <sys/syscall.h>
  61 #include <sys/syscallsubr.h>
  62 #include <sys/sysctl.h>
  63 #include <sys/sysproto.h>
  64 #include <sys/ucontext.h>
  65 #include <sys/uio.h>
  66 #include <sys/uuid.h>
  67 #include <sys/vmmeter.h>
  68 #include <sys/vnode.h>
  69
  70 #include <ddb/ddb.h>
  71
  72 #include <net/netisr.h>
  73
  74 #include <vm/vm.h>
  75 #include <vm/vm_extern.h>
  76 #include <vm/vm_kern.h>
  77 #include <vm/vm_page.h>
  78 #include <vm/vm_map.h>
  79 #include <vm/vm_object.h>
  80 #include <vm/vm_pager.h>
  81
  82 #include <machine/bootinfo.h>
  83 #include <machine/cpu.h>
  84 #include <machine/efi.h>
  85 #include <machine/elf.h>
  86 #include <machine/fpu.h>
  87 #include <machine/intr.h>
  88 #include <machine/mca.h>
  89 #include <machine/md_var.h>
  90 #include <machine/pal.h>
  91 #include <machine/pcb.h>
  92 #include <machine/reg.h>
  93 #include <machine/sal.h>
  94 #include <machine/sigframe.h>
  95 #ifdef SMP
  96 #include <machine/smp.h>
  97 #endif
  98 #include <machine/unwind.h>
  99 #include <machine/vmparam.h>
 100
 101 SYSCTL_NODE(_hw, OID_AUTO, freq, CTLFLAG_RD, 0, "");
 102 SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, "");
 103
 104 static u_int bus_freq;
 105 SYSCTL_UINT(_hw_freq, OID_AUTO, bus, CTLFLAG_RD, &bus_freq, 0,
 106     "Bus clock frequency");
 107
 108 static u_int cpu_freq;
 109 SYSCTL_UINT(_hw_freq, OID_AUTO, cpu, CTLFLAG_RD, &cpu_freq, 0,
 110     "CPU clock frequency");
 111
 112 static u_int itc_freq;
 113 SYSCTL_UINT(_hw_freq, OID_AUTO, itc, CTLFLAG_RD, &itc_freq, 0,
 114     "ITC frequency");
 115
 116 int cold = 1;
 117
 118 struct bootinfo *bootinfo;
 119
 120 struct pcpu pcpu0;
 121
 122 extern u_int64_t kernel_text[], _end[];
 123
 124 extern u_int64_t ia64_gateway_page[];
 125 extern u_int64_t break_sigtramp[];
 126 extern u_int64_t epc_sigtramp[];
 127
 128 struct fpswa_iface *fpswa_iface;
 129
 130 vm_size_t ia64_pal_size;
 131 vm_paddr_t ia64_pal_base;
 132 vm_offset_t ia64_port_base;
 133
 134 u_int64_t ia64_lapic_addr = PAL_PIB_DEFAULT_ADDR;
 135
 136 struct ia64_pib *ia64_pib;
 137
 138 static int ia64_sync_icache_needed;
 139
 140 char machine[] = MACHINE;
 141 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
 142
 143 static char cpu_model[64];
 144 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
 145     "The CPU model name");
 146
 147 static char cpu_family[64];
 148 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
 149     "The CPU family name");
 150
 151 #ifdef DDB
 152 extern vm_offset_t ksym_start, ksym_end;
 153 #endif
 154
 155
 156 struct msgbuf *msgbufp = NULL;
 157
 158 /* Other subsystems (e.g., ACPI) can hook this later. */
 159 void (*cpu_idle_hook)(void) = NULL;
 160
 161 long Maxmem = 0;
 162 long realmem = 0;
 163
 164 #define PHYSMAP_SIZE    (2 * VM_PHYSSEG_MAX)
 165
 166 vm_paddr_t phys_avail[PHYSMAP_SIZE + 2];
 167
 168 /* must be 2 less so 0 0 can signal end of chunks */
 169 #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2)
 170
 171 struct kva_md_info kmi;
 172
 173 #define Mhz     1000000L
 174 #define Ghz     (1000L*Mhz)
 175
 176 static void
 177 identifycpu(void)
 178 {
 179         char vendor[17];
 180         char *family_name, *model_name;
 181         u_int64_t features, tmp;
 182         int number, revision, model, family, archrev;
 183
 184         /*
 185          * Assumes little-endian.
 186          */
 187         *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
 188         *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
 189         vendor[16] = '\0';
 190
 191         tmp = ia64_get_cpuid(3);
 192         number = (tmp >> 0) & 0xff;
 193         revision = (tmp >> 8) & 0xff;
 194         model = (tmp >> 16) & 0xff;
 195         family = (tmp >> 24) & 0xff;
 196         archrev = (tmp >> 32) & 0xff;
 197
 198         family_name = model_name = "unknown";
 199         switch (family) {
 200         case 0x07:
 201                 family_name = "Itanium";
 202                 model_name = "Merced";
 203                 break;
 204         case 0x1f:
 205                 family_name = "Itanium 2";
 206                 switch (model) {
 207                 case 0x00:
 208                         model_name = "McKinley";
 209                         break;
 210                 case 0x01:
 211                         /*
 212                          * Deerfield is a low-voltage variant based on the
 213                          * Madison core. We need circumstantial evidence
 214                          * (i.e. the clock frequency) to identify those.
 215                          * Allow for roughly 1% error margin.
 216                          */
 217                         if (cpu_freq > 990 && cpu_freq < 1010)
 218                                 model_name = "Deerfield";
 219                         else
 220                                 model_name = "Madison";
 221                         break;
 222                 case 0x02:
 223                         model_name = "Madison II";
 224                         break;
 225                 }
 226                 break;
 227         case 0x20:
 228                 ia64_sync_icache_needed = 1;
 229
 230                 family_name = "Itanium 2";
 231                 switch (model) {
 232                 case 0x00:
 233                         model_name = "Montecito";
 234                         break;
 235                 case 0x01:
 236                         model_name = "Montvale";
 237                         break;
 238                 }
 239                 break;
 240         }
 241         snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
 242         snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
 243
 244         features = ia64_get_cpuid(4);
 245
 246         printf("CPU: %s (", model_name);
 247         if (cpu_freq)
 248                 printf("%u Mhz ", cpu_freq);
 249         printf("%s)\n", family_name);
 250         printf("  Origin = \"%s\"  Revision = %d\n", vendor, revision);
 251         printf("  Features = 0x%b\n", (u_int32_t) features,
 252             "\020"
 253             "\001LB"    /* long branch (brl) instruction. */
 254             "\002SD"    /* Spontaneous deferral. */
 255             "\003AO"    /* 16-byte atomic operations (ld, st, cmpxchg). */ );
 256 }
 257
 258 static void
 259 cpu_startup(void *dummy)
 260 {
 261         char nodename[16];
 262         struct pcpu *pc;
 263         struct pcpu_stats *pcs;
 264
 265         /*
 266          * Good {morning,afternoon,evening,night}.
 267          */
 268         identifycpu();
 269
 270 #ifdef PERFMON
 271         perfmon_init();
 272 #endif
 273         printf("real memory  = %ld (%ld MB)\n", ia64_ptob(Maxmem),
 274             ia64_ptob(Maxmem) / 1048576);
 275         realmem = Maxmem;
 276
 277         /*
 278          * Display any holes after the first chunk of extended memory.
 279          */
 280         if (bootverbose) {
 281                 int indx;
 282
 283                 printf("Physical memory chunk(s):\n");
 284                 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
 285                         long size1 = phys_avail[indx + 1] - phys_avail[indx];
 286
 287                         printf("0x%08lx - 0x%08lx, %ld bytes (%ld pages)\n",
 288                             phys_avail[indx], phys_avail[indx + 1] - 1, size1,
 289                             size1 >> PAGE_SHIFT);
 290                 }
 291         }
 292
 293         vm_ksubmap_init(&kmi);
 294
 295         printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count),
 296             ptoa(cnt.v_free_count) / 1048576);
 297
 298         if (fpswa_iface == NULL)
 299                 printf("Warning: no FPSWA package supplied\n");
 300         else
 301                 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
 302                     (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa);
 303
 304         /*
 305          * Set up buffers, so they can be used to read disk labels.
 306          */
 307         bufinit();
 308         vm_pager_bufferinit();
 309
 310         /*
 311          * Traverse the MADT to discover IOSAPIC and Local SAPIC
 312          * information.
 313          */
 314         ia64_probe_sapics();
 315         ia64_pib = pmap_mapdev(ia64_lapic_addr, sizeof(*ia64_pib));
 316
 317         ia64_mca_init();
 318
 319         /*
 320          * Create sysctl tree for per-CPU information.
 321          */
 322         STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
 323                 snprintf(nodename, sizeof(nodename), "%u", pc->pc_cpuid);
 324                 sysctl_ctx_init(&pc->pc_md.sysctl_ctx);
 325                 pc->pc_md.sysctl_tree = SYSCTL_ADD_NODE(&pc->pc_md.sysctl_ctx,
 326                     SYSCTL_STATIC_CHILDREN(_machdep_cpu), OID_AUTO, nodename,
 327                     CTLFLAG_RD, NULL, "");
 328                 if (pc->pc_md.sysctl_tree == NULL)
 329                         continue;
 330
 331                 pcs = &pc->pc_md.stats;
 332
 333                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 334                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 335                     "nasts", CTLFLAG_RD, &pcs->pcs_nasts,
 336                     "Number of IPI_AST interrupts");
 337
 338                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 339                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 340                     "nclks", CTLFLAG_RD, &pcs->pcs_nclks,
 341                     "Number of clock interrupts");
 342
 343                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 344                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 345                     "nextints", CTLFLAG_RD, &pcs->pcs_nextints,
 346                     "Number of ExtINT interrupts");
 347
 348                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 349                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 350                     "nhardclocks", CTLFLAG_RD, &pcs->pcs_nhardclocks,
 351                     "Number of IPI_HARDCLOCK interrupts");
 352
 353                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 354                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 355                     "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps,
 356                     "Number of IPI_HIGH_FP interrupts");
 357
 358                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 359                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 360                     "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints,
 361                     "Number of hardware (device) interrupts");
 362
 363                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 364                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 365                     "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts,
 366                     "Number of IPI_PREEMPT interrupts");
 367
 368                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 369                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 370                     "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs,
 371                     "Number of IPI_RENDEZVOUS interrupts");
 372
 373                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 374                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 375                     "nstops", CTLFLAG_RD, &pcs->pcs_nstops,
 376                     "Number of IPI_STOP interrupts");
 377
 378                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 379                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 380                     "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays,
 381                     "Number of stray interrupts");
 382         }
 383 }
 384 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
 385
 386 void
 387 cpu_flush_dcache(void *ptr, size_t len)
 388 {
 389         vm_offset_t lim, va;
 390
 391         va = (uintptr_t)ptr & ~31;
 392         lim = (uintptr_t)ptr + len;
 393         while (va < lim) {
 394                 ia64_fc(va);
 395                 va += 32;
 396         }
 397
 398         ia64_srlz_d();
 399 }
 400
 401 /* Get current clock frequency for the given cpu id. */
 402 int
 403 cpu_est_clockrate(int cpu_id, uint64_t *rate)
 404 {
 405
 406         if (pcpu_find(cpu_id) == NULL || rate == NULL)
 407                 return (EINVAL);
 408         *rate = (u_long)cpu_freq * 1000000ul;
 409         return (0);
 410 }
 411
 412 void
 413 cpu_halt()
 414 {
 415
 416         efi_reset_system();
 417 }
 418
 419 void
 420 cpu_idle(int busy)
 421 {
 422         register_t ie;
 423
 424         if (!busy) {
 425                 critical_enter();
 426                 cpu_idleclock();
 427         }
 428
 429         ie = intr_disable();
 430         KASSERT(ie != 0, ("%s called with interrupts disabled\n", __func__));
 431
 432         if (sched_runnable())
 433                 ia64_enable_intr();
 434         else if (cpu_idle_hook != NULL) {
 435                 (*cpu_idle_hook)();
 436                 /* The hook must enable interrupts! */
 437         } else {
 438                 ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
 439                 ia64_enable_intr();
 440         }
 441
 442         if (!busy) {
 443                 cpu_activeclock();
 444                 critical_exit();
 445         }
 446 }
 447
 448 int
 449 cpu_idle_wakeup(int cpu)
 450 {
 451
 452         return (0);
 453 }
 454
 455 void
 456 cpu_reset()
 457 {
 458
 459         efi_reset_system();
 460 }
 461
 462 void
 463 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
 464 {
 465         struct pcb *oldpcb, *newpcb;
 466
 467         oldpcb = old->td_pcb;
 468 #ifdef COMPAT_FREEBSD32
 469         ia32_savectx(oldpcb);
 470 #endif
 471         if (PCPU_GET(fpcurthread) == old)
 472                 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
 473         if (!savectx(oldpcb)) {
 474                 newpcb = new->td_pcb;
 475                 oldpcb->pcb_current_pmap =
 476                     pmap_switch(newpcb->pcb_current_pmap);
 477
 478                 atomic_store_rel_ptr(&old->td_lock, mtx);
 479
 480 #if defined(SCHED_ULE) && defined(SMP)
 481                 while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
 482                         cpu_spinwait();
 483 #endif
 484
 485                 PCPU_SET(curthread, new);
 486
 487 #ifdef COMPAT_FREEBSD32
 488                 ia32_restorectx(newpcb);
 489 #endif
 490
 491                 if (PCPU_GET(fpcurthread) == new)
 492                         new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
 493                 restorectx(newpcb);
 494                 /* We should not get here. */
 495                 panic("cpu_switch: restorectx() returned");
 496                 /* NOTREACHED */
 497         }
 498 }
 499
 500 void
 501 cpu_throw(struct thread *old __unused, struct thread *new)
 502 {
 503         struct pcb *newpcb;
 504
 505         newpcb = new->td_pcb;
 506         (void)pmap_switch(newpcb->pcb_current_pmap);
 507
 508 #if defined(SCHED_ULE) && defined(SMP)
 509         while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
 510                 cpu_spinwait();
 511 #endif
 512
 513         PCPU_SET(curthread, new);
 514
 515 #ifdef COMPAT_FREEBSD32
 516         ia32_restorectx(newpcb);
 517 #endif
 518
 519         restorectx(newpcb);
 520         /* We should not get here. */
 521         panic("cpu_throw: restorectx() returned");
 522         /* NOTREACHED */
 523 }
 524
 525 void
 526 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
 527 {
 528
 529         /*
 530          * Set pc_acpi_id to "uninitialized".
 531          * See sys/dev/acpica/acpi_cpu.c
 532          */
 533         pcpu->pc_acpi_id = 0xffffffff;
 534 }
 535
 536 void
 537 spinlock_enter(void)
 538 {
 539         struct thread *td;
 540         int intr;
 541
 542         td = curthread;
 543         if (td->td_md.md_spinlock_count == 0) {
 544                 intr = intr_disable();
 545                 td->td_md.md_spinlock_count = 1;
 546                 td->td_md.md_saved_intr = intr;
 547         } else
 548                 td->td_md.md_spinlock_count++;
 549         critical_enter();
 550 }
 551
 552 void
 553 spinlock_exit(void)
 554 {
 555         struct thread *td;
 556         int intr;
 557
 558         td = curthread;
 559         critical_exit();
 560         intr = td->td_md.md_saved_intr;
 561         td->td_md.md_spinlock_count--;
 562         if (td->td_md.md_spinlock_count == 0)
 563                 intr_restore(intr);
 564 }
 565
 566 void
 567 map_vhpt(uintptr_t vhpt)
 568 {
 569         pt_entry_t pte;
 570         uint64_t psr;
 571
 572         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 573             PTE_PL_KERN | PTE_AR_RW;
 574         pte |= vhpt & PTE_PPN_MASK;
 575
 576         __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
 577             "r"(pmap_vhpt_log2size << 2));
 578
 579         __asm __volatile("mov   %0=psr" : "=r"(psr));
 580         __asm __volatile("rsm   psr.ic|psr.i");
 581         ia64_srlz_i();
 582         ia64_set_ifa(vhpt);
 583         ia64_set_itir(pmap_vhpt_log2size << 2);
 584         ia64_srlz_d();
 585         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
 586         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 587         ia64_srlz_i();
 588 }
 589
 590 void
 591 map_pal_code(void)
 592 {
 593         pt_entry_t pte;
 594         vm_offset_t va;
 595         vm_size_t sz;
 596         uint64_t psr;
 597         u_int shft;
 598
 599         if (ia64_pal_size == 0)
 600                 return;
 601
 602         va = IA64_PHYS_TO_RR7(ia64_pal_base);
 603
 604         sz = ia64_pal_size;
 605         shft = 0;
 606         while (sz > 1) {
 607                 shft++;
 608                 sz >>= 1;
 609         }
 610
 611         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 612             PTE_PL_KERN | PTE_AR_RWX;
 613         pte |= ia64_pal_base & PTE_PPN_MASK;
 614
 615         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" :: "r"(va), "r"(shft<<2));
 616
 617         __asm __volatile("mov   %0=psr" : "=r"(psr));
 618         __asm __volatile("rsm   psr.ic|psr.i");
 619         ia64_srlz_i();
 620         ia64_set_ifa(va);
 621         ia64_set_itir(shft << 2);
 622         ia64_srlz_d();
 623         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(4), "r"(pte));
 624         ia64_srlz_d();
 625         __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
 626         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 627         ia64_srlz_i();
 628 }
 629
 630 void
 631 map_gateway_page(void)
 632 {
 633         pt_entry_t pte;
 634         uint64_t psr;
 635
 636         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 637             PTE_PL_KERN | PTE_AR_X_RX;
 638         pte |= ia64_tpa((uint64_t)ia64_gateway_page) & PTE_PPN_MASK;
 639
 640         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
 641             "r"(VM_MAXUSER_ADDRESS), "r"(PAGE_SHIFT << 2));
 642
 643         __asm __volatile("mov   %0=psr" : "=r"(psr));
 644         __asm __volatile("rsm   psr.ic|psr.i");
 645         ia64_srlz_i();
 646         ia64_set_ifa(VM_MAXUSER_ADDRESS);
 647         ia64_set_itir(PAGE_SHIFT << 2);
 648         ia64_srlz_d();
 649         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(5), "r"(pte));
 650         ia64_srlz_d();
 651         __asm __volatile("itr.i itr[%0]=%1" :: "r"(2), "r"(pte));
 652         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 653         ia64_srlz_i();
 654
 655         /* Expose the mapping to userland in ar.k5 */
 656         ia64_set_k5(VM_MAXUSER_ADDRESS);
 657 }
 658
 659 static u_int
 660 freq_ratio(u_long base, u_long ratio)
 661 {
 662         u_long f;
 663
 664         f = (base * (ratio >> 32)) / (ratio & 0xfffffffful);
 665         return ((f + 500000) / 1000000);
 666 }
 667
 668 static void
 669 calculate_frequencies(void)
 670 {
 671         struct ia64_sal_result sal;
 672         struct ia64_pal_result pal;
 673         register_t ie;
 674
 675         ie = intr_disable();
 676         sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
 677         pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
 678         intr_restore(ie);
 679
 680         if (sal.sal_status == 0 && pal.pal_status == 0) {
 681                 if (bootverbose) {
 682                         printf("Platform clock frequency %ld Hz\n",
 683                                sal.sal_result[0]);
 684                         printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
 685                                "ITC ratio %ld/%ld\n",
 686                                pal.pal_result[0] >> 32,
 687                                pal.pal_result[0] & ((1L << 32) - 1),
 688                                pal.pal_result[1] >> 32,
 689                                pal.pal_result[1] & ((1L << 32) - 1),
 690                                pal.pal_result[2] >> 32,
 691                                pal.pal_result[2] & ((1L << 32) - 1));
 692                 }
 693                 cpu_freq = freq_ratio(sal.sal_result[0], pal.pal_result[0]);
 694                 bus_freq = freq_ratio(sal.sal_result[0], pal.pal_result[1]);
 695                 itc_freq = freq_ratio(sal.sal_result[0], pal.pal_result[2]);
 696         }
 697 }
 698
 699 struct ia64_init_return
 700 ia64_init(void)
 701 {
 702         struct ia64_init_return ret;
 703         int phys_avail_cnt;
 704         vm_offset_t kernstart, kernend;
 705         vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1;
 706         char *p;
 707         struct efi_md *md;
 708         int metadata_missing;
 709
 710         /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */
 711
 712         /*
 713          * TODO: Disable interrupts, floating point etc.
 714          * Maybe flush cache and tlb
 715          */
 716         ia64_set_fpsr(IA64_FPSR_DEFAULT);
 717
 718         /*
 719          * TODO: Get critical system information (if possible, from the
 720          * information provided by the boot program).
 721          */
 722
 723         /*
 724          * Look for the I/O ports first - we need them for console
 725          * probing.
 726          */
 727         for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
 728                 switch (md->md_type) {
 729                 case EFI_MD_TYPE_IOPORT:
 730                         ia64_port_base = (uintptr_t)pmap_mapdev(md->md_phys,
 731                             md->md_pages * EFI_PAGE_SIZE);
 732                         break;
 733                 case EFI_MD_TYPE_PALCODE:
 734                         ia64_pal_size = md->md_pages * EFI_PAGE_SIZE;
 735                         ia64_pal_base = md->md_phys;
 736                         break;
 737                 }
 738         }
 739
 740         metadata_missing = 0;
 741         if (bootinfo->bi_modulep)
 742                 preload_metadata = (caddr_t)bootinfo->bi_modulep;
 743         else
 744                 metadata_missing = 1;
 745
 746         if (envmode == 0 && bootinfo->bi_envp)
 747                 kern_envp = (caddr_t)bootinfo->bi_envp;
 748         else
 749                 kern_envp = static_env;
 750
 751         /*
 752          * Look at arguments passed to us and compute boothowto.
 753          */
 754         boothowto = bootinfo->bi_boothowto;
 755
 756         if (boothowto & RB_VERBOSE)
 757                 bootverbose = 1;
 758
 759         /*
 760          * Find the beginning and end of the kernel.
 761          */
 762         kernstart = trunc_page(kernel_text);
 763 #ifdef DDB
 764         ksym_start = bootinfo->bi_symtab;
 765         ksym_end = bootinfo->bi_esymtab;
 766         kernend = (vm_offset_t)round_page(ksym_end);
 767 #else
 768         kernend = (vm_offset_t)round_page(_end);
 769 #endif
 770         /* But if the bootstrap tells us otherwise, believe it! */
 771         if (bootinfo->bi_kernend)
 772                 kernend = round_page(bootinfo->bi_kernend);
 773
 774         /*
 775          * Region 6 is direct mapped UC and region 7 is direct mapped
 776          * WC. The details of this is controlled by the Alt {I,D}TLB
 777          * handlers. Here we just make sure that they have the largest
 778          * possible page size to minimise TLB usage.
 779          */
 780         ia64_set_rr(IA64_RR_BASE(6), (6 << 8) | (PAGE_SHIFT << 2));
 781         ia64_set_rr(IA64_RR_BASE(7), (7 << 8) | (PAGE_SHIFT << 2));
 782         ia64_srlz_d();
 783
 784         /*
 785          * Wire things up so we can call the firmware.
 786          */
 787         map_pal_code();
 788         efi_boot_minimal(bootinfo->bi_systab);
 789         ia64_xiv_init();
 790         ia64_sal_init();
 791         calculate_frequencies();
 792
 793         set_cputicker(ia64_get_itc, (u_long)itc_freq * 1000000, 0);
 794
 795         /*
 796          * Setup the PCPU data for the bootstrap processor. It is needed
 797          * by printf(). Also, since printf() has critical sections, we
 798          * need to initialize at least pc_curthread.
 799          */
 800         pcpup = &pcpu0;
 801         ia64_set_k4((u_int64_t)pcpup);
 802         pcpu_init(pcpup, 0, sizeof(pcpu0));
 803         dpcpu_init((void *)kernend, 0);
 804         PCPU_SET(md.lid, ia64_get_lid());
 805         kernend += DPCPU_SIZE;
 806         PCPU_SET(curthread, &thread0);
 807
 808         /*
 809          * Initialize the console before we print anything out.
 810          */
 811         cninit();
 812
 813         /* OUTPUT NOW ALLOWED */
 814
 815         if (metadata_missing)
 816                 printf("WARNING: loader(8) metadata is missing!\n");
 817
 818         /* Get FPSWA interface */
 819         fpswa_iface = (bootinfo->bi_fpswa == 0) ? NULL :
 820             (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo->bi_fpswa);
 821
 822         /* Init basic tunables, including hz */
 823         init_param1();
 824
 825         p = getenv("kernelname");
 826         if (p != NULL) {
 827                 strlcpy(kernelname, p, sizeof(kernelname));
 828                 freeenv(p);
 829         }
 830
 831         kernstartpfn = atop(IA64_RR_MASK(kernstart));
 832         kernendpfn = atop(IA64_RR_MASK(kernend));
 833
 834         /*
 835          * Size the memory regions and load phys_avail[] with the results.
 836          */
 837
 838         /*
 839          * Find out how much memory is available, by looking at
 840          * the memory descriptors.
 841          */
 842
 843 #ifdef DEBUG_MD
 844         printf("Memory descriptor count: %d\n", mdcount);
 845 #endif
 846
 847         phys_avail_cnt = 0;
 848         for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
 849 #ifdef DEBUG_MD
 850                 printf("MD %p: type %d pa 0x%lx cnt 0x%lx\n", md,
 851                     md->md_type, md->md_phys, md->md_pages);
 852 #endif
 853
 854                 pfn0 = ia64_btop(round_page(md->md_phys));
 855                 pfn1 = ia64_btop(trunc_page(md->md_phys + md->md_pages * 4096));
 856                 if (pfn1 <= pfn0)
 857                         continue;
 858
 859                 if (md->md_type != EFI_MD_TYPE_FREE)
 860                         continue;
 861
 862                 /*
 863                  * We have a memory descriptor that describes conventional
 864                  * memory that is for general use. We must determine if the
 865                  * loader has put the kernel in this region.
 866                  */
 867                 physmem += (pfn1 - pfn0);
 868                 if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) {
 869                         /*
 870                          * Must compute the location of the kernel
 871                          * within the segment.
 872                          */
 873 #ifdef DEBUG_MD
 874                         printf("Descriptor %p contains kernel\n", mp);
 875 #endif
 876                         if (pfn0 < kernstartpfn) {
 877                                 /*
 878                                  * There is a chunk before the kernel.
 879                                  */
 880 #ifdef DEBUG_MD
 881                                 printf("Loading chunk before kernel: "
 882                                        "0x%lx / 0x%lx\n", pfn0, kernstartpfn);
 883 #endif
 884                                 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
 885                                 phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn);
 886                                 phys_avail_cnt += 2;
 887                         }
 888                         if (kernendpfn < pfn1) {
 889                                 /*
 890                                  * There is a chunk after the kernel.
 891                                  */
 892 #ifdef DEBUG_MD
 893                                 printf("Loading chunk after kernel: "
 894                                        "0x%lx / 0x%lx\n", kernendpfn, pfn1);
 895 #endif
 896                                 phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn);
 897                                 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
 898                                 phys_avail_cnt += 2;
 899                         }
 900                 } else {
 901                         /*
 902                          * Just load this cluster as one chunk.
 903                          */
 904 #ifdef DEBUG_MD
 905                         printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i,
 906                                pfn0, pfn1);
 907 #endif
 908                         phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
 909                         phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
 910                         phys_avail_cnt += 2;
 911
 912                 }
 913         }
 914         phys_avail[phys_avail_cnt] = 0;
 915
 916         Maxmem = physmem;
 917         init_param2(physmem);
 918
 919         /*
 920          * Initialize error message buffer (at end of core).
 921          */
 922         msgbufp = (struct msgbuf *)pmap_steal_memory(msgbufsize);
 923         msgbufinit(msgbufp, msgbufsize);
 924
 925         proc_linkup0(&proc0, &thread0);
 926         /*
 927          * Init mapping for kernel stack for proc 0
 928          */
 929         thread0.td_kstack = pmap_steal_memory(KSTACK_PAGES * PAGE_SIZE);
 930         thread0.td_kstack_pages = KSTACK_PAGES;
 931
 932         mutex_init();
 933
 934         /*
 935          * Initialize the rest of proc 0's PCB.
 936          *
 937          * Set the kernel sp, reserving space for an (empty) trapframe,
 938          * and make proc0's trapframe pointer point to it for sanity.
 939          * Initialise proc0's backing store to start after u area.
 940          */
 941         cpu_thread_alloc(&thread0);
 942         thread0.td_frame->tf_flags = FRAME_SYSCALL;
 943         thread0.td_pcb->pcb_special.sp =
 944             (u_int64_t)thread0.td_frame - 16;
 945         thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
 946
 947         /*
 948          * Initialize the virtual memory system.
 949          */
 950         pmap_bootstrap();
 951
 952         /*
 953          * Initialize debuggers, and break into them if appropriate.
 954          */
 955         kdb_init();
 956
 957 #ifdef KDB
 958         if (boothowto & RB_KDB)
 959                 kdb_enter(KDB_WHY_BOOTFLAGS,
 960                     "Boot flags requested debugger\n");
 961 #endif
 962
 963         ia64_set_tpr(0);
 964         ia64_srlz_d();
 965
 966         ret.bspstore = thread0.td_pcb->pcb_special.bspstore;
 967         ret.sp = thread0.td_pcb->pcb_special.sp;
 968         return (ret);
 969 }
 970
 971 uint64_t
 972 ia64_get_hcdp(void)
 973 {
 974
 975         return (bootinfo->bi_hcdp);
 976 }
 977
 978 void
 979 bzero(void *buf, size_t len)
 980 {
 981         caddr_t p = buf;
 982
 983         while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
 984                 *p++ = 0;
 985                 len--;
 986         }
 987         while (len >= sizeof(u_long) * 8) {
 988                 *(u_long*) p = 0;
 989                 *((u_long*) p + 1) = 0;
 990                 *((u_long*) p + 2) = 0;
 991                 *((u_long*) p + 3) = 0;
 992                 len -= sizeof(u_long) * 8;
 993                 *((u_long*) p + 4) = 0;
 994                 *((u_long*) p + 5) = 0;
 995                 *((u_long*) p + 6) = 0;
 996                 *((u_long*) p + 7) = 0;
 997                 p += sizeof(u_long) * 8;
 998         }
 999         while (len >= sizeof(u_long)) {
1000                 *(u_long*) p = 0;
1001                 len -= sizeof(u_long);
1002                 p += sizeof(u_long);
1003         }
1004         while (len) {
1005                 *p++ = 0;
1006                 len--;
1007         }
1008 }
1009
1010 u_int
1011 ia64_itc_freq(void)
1012 {
1013
1014         return (itc_freq);
1015 }
1016
1017 void
1018 DELAY(int n)
1019 {
1020         u_int64_t start, end, now;
1021
1022         sched_pin();
1023
1024         start = ia64_get_itc();
1025         end = start + itc_freq * n;
1026         /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
1027         do {
1028                 now = ia64_get_itc();
1029         } while (now < end || (now > start && end < start));
1030
1031         sched_unpin();
1032 }
1033
1034 /*
1035  * Send an interrupt (signal) to a process.
1036  */
1037 void
1038 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
1039 {
1040         struct proc *p;
1041         struct thread *td;
1042         struct trapframe *tf;
1043         struct sigacts *psp;
1044         struct sigframe sf, *sfp;
1045         u_int64_t sbs, sp;
1046         int oonstack;
1047         int sig;
1048         u_long code;
1049
1050         td = curthread;
1051         p = td->td_proc;
1052         PROC_LOCK_ASSERT(p, MA_OWNED);
1053         sig = ksi->ksi_signo;
1054         code = ksi->ksi_code;
1055         psp = p->p_sigacts;
1056         mtx_assert(&psp->ps_mtx, MA_OWNED);
1057         tf = td->td_frame;
1058         sp = tf->tf_special.sp;
1059         oonstack = sigonstack(sp);
1060         sbs = 0;
1061
1062         /* save user context */
1063         bzero(&sf, sizeof(struct sigframe));
1064         sf.sf_uc.uc_sigmask = *mask;
1065         sf.sf_uc.uc_stack = td->td_sigstk;
1066         sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1067             ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1068
1069         /*
1070          * Allocate and validate space for the signal handler
1071          * context. Note that if the stack is in P0 space, the
1072          * call to grow() is a nop, and the useracc() check
1073          * will fail if the process has not already allocated
1074          * the space with a `brk'.
1075          */
1076         if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1077             SIGISMEMBER(psp->ps_sigonstack, sig)) {
1078                 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1079                 sbs = (sbs + 15) & ~15;
1080                 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1081 #if defined(COMPAT_43)
1082                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1083 #endif
1084         } else
1085                 sfp = (struct sigframe *)sp;
1086         sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1087
1088         /* Fill in the siginfo structure for POSIX handlers. */
1089         if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1090                 sf.sf_si = ksi->ksi_info;
1091                 sf.sf_si.si_signo = sig;
1092                 /*
1093                  * XXX this shouldn't be here after code in trap.c
1094                  * is fixed
1095                  */
1096                 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1097                 code = (u_int64_t)&sfp->sf_si;
1098         }
1099
1100         mtx_unlock(&psp->ps_mtx);
1101         PROC_UNLOCK(p);
1102
1103         get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1104
1105         /* Copy the frame out to userland. */
1106         if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1107                 /*
1108                  * Process has trashed its stack; give it an illegal
1109                  * instruction to halt it in its tracks.
1110                  */
1111                 PROC_LOCK(p);
1112                 sigexit(td, SIGILL);
1113                 return;
1114         }
1115
1116         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1117                 tf->tf_special.psr &= ~IA64_PSR_RI;
1118                 tf->tf_special.iip = ia64_get_k5() +
1119                     ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1120         } else
1121                 tf->tf_special.iip = ia64_get_k5() +
1122                     ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1123
1124         /*
1125          * Setup the trapframe to return to the signal trampoline. We pass
1126          * information to the trampoline in the following registers:
1127          *
1128          *      gp      new backing store or NULL
1129          *      r8      signal number
1130          *      r9      signal code or siginfo pointer
1131          *      r10     signal handler (function descriptor)
1132          */
1133         tf->tf_special.sp = (u_int64_t)sfp - 16;
1134         tf->tf_special.gp = sbs;
1135         tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1136         tf->tf_special.ndirty = 0;
1137         tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1138         tf->tf_scratch.gr8 = sig;
1139         tf->tf_scratch.gr9 = code;
1140         tf->tf_scratch.gr10 = (u_int64_t)catcher;
1141
1142         PROC_LOCK(p);
1143         mtx_lock(&psp->ps_mtx);
1144 }
1145
1146 /*
1147  * System call to cleanup state after a signal
1148  * has been taken.  Reset signal mask and
1149  * stack state from context left by sendsig (above).
1150  * Return to previous pc and psl as specified by
1151  * context left by sendsig. Check carefully to
1152  * make sure that the user has not modified the
1153  * state to gain improper privileges.
1154  *
1155  * MPSAFE
1156  */
1157 int
1158 sys_sigreturn(struct thread *td,
1159         struct sigreturn_args /* {
1160                 ucontext_t *sigcntxp;
1161         } */ *uap)
1162 {
1163         ucontext_t uc;
1164         struct trapframe *tf;
1165         struct pcb *pcb;
1166
1167         tf = td->td_frame;
1168         pcb = td->td_pcb;
1169
1170         /*
1171          * Fetch the entire context structure at once for speed.
1172          * We don't use a normal argument to simplify RSE handling.
1173          */
1174         if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1175                 return (EFAULT);
1176
1177         set_mcontext(td, &uc.uc_mcontext);
1178
1179 #if defined(COMPAT_43)
1180         if (sigonstack(tf->tf_special.sp))
1181                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1182         else
1183                 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1184 #endif
1185         kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
1186
1187         return (EJUSTRETURN);
1188 }
1189
1190 #ifdef COMPAT_FREEBSD4
1191 int
1192 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1193 {
1194
1195         return sys_sigreturn(td, (struct sigreturn_args *)uap);
1196 }
1197 #endif
1198
1199 /*
1200  * Construct a PCB from a trapframe. This is called from kdb_trap() where
1201  * we want to start a backtrace from the function that caused us to enter
1202  * the debugger. We have the context in the trapframe, but base the trace
1203  * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1204  * enough for a backtrace.
1205  */
1206 void
1207 makectx(struct trapframe *tf, struct pcb *pcb)
1208 {
1209
1210         pcb->pcb_special = tf->tf_special;
1211         pcb->pcb_special.__spare = ~0UL;        /* XXX see unwind.c */
1212         save_callee_saved(&pcb->pcb_preserved);
1213         save_callee_saved_fp(&pcb->pcb_preserved_fp);
1214 }
1215
1216 int
1217 ia64_flush_dirty(struct thread *td, struct _special *r)
1218 {
1219         struct iovec iov;
1220         struct uio uio;
1221         uint64_t bspst, kstk, rnat;
1222         int error, locked;
1223
1224         if (r->ndirty == 0)
1225                 return (0);
1226
1227         kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1228         if (td == curthread) {
1229                 __asm __volatile("mov   ar.rsc=0;;");
1230                 __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1231                 /* Make sure we have all the user registers written out. */
1232                 if (bspst - kstk < r->ndirty) {
1233                         __asm __volatile("flushrs;;");
1234                         __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1235                 }
1236                 __asm __volatile("mov   %0=ar.rnat;;" : "=r"(rnat));
1237                 __asm __volatile("mov   ar.rsc=3");
1238                 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1239                 kstk += r->ndirty;
1240                 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1241                     ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1242         } else {
1243                 locked = PROC_LOCKED(td->td_proc);
1244                 if (!locked)
1245                         PHOLD(td->td_proc);
1246                 iov.iov_base = (void*)(uintptr_t)kstk;
1247                 iov.iov_len = r->ndirty;
1248                 uio.uio_iov = &iov;
1249                 uio.uio_iovcnt = 1;
1250                 uio.uio_offset = r->bspstore;
1251                 uio.uio_resid = r->ndirty;
1252                 uio.uio_segflg = UIO_SYSSPACE;
1253                 uio.uio_rw = UIO_WRITE;
1254                 uio.uio_td = td;
1255                 error = proc_rwmem(td->td_proc, &uio);
1256                 /*
1257                  * XXX proc_rwmem() doesn't currently return ENOSPC,
1258                  * so I think it can bogusly return 0. Neither do
1259                  * we allow short writes.
1260                  */
1261                 if (uio.uio_resid != 0 && error == 0)
1262                         error = ENOSPC;
1263                 if (!locked)
1264                         PRELE(td->td_proc);
1265         }
1266
1267         r->bspstore += r->ndirty;
1268         r->ndirty = 0;
1269         return (error);
1270 }
1271
1272 int
1273 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1274 {
1275         struct trapframe *tf;
1276         int error;
1277
1278         tf = td->td_frame;
1279         bzero(mc, sizeof(*mc));
1280         mc->mc_special = tf->tf_special;
1281         error = ia64_flush_dirty(td, &mc->mc_special);
1282         if (tf->tf_flags & FRAME_SYSCALL) {
1283                 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1284                 mc->mc_scratch = tf->tf_scratch;
1285                 if (flags & GET_MC_CLEAR_RET) {
1286                         mc->mc_scratch.gr8 = 0;
1287                         mc->mc_scratch.gr9 = 0;
1288                         mc->mc_scratch.gr10 = 0;
1289                         mc->mc_scratch.gr11 = 0;
1290                 }
1291         } else {
1292                 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1293                 mc->mc_scratch = tf->tf_scratch;
1294                 mc->mc_scratch_fp = tf->tf_scratch_fp;
1295                 /*
1296                  * XXX If the thread never used the high FP registers, we
1297                  * probably shouldn't waste time saving them.
1298                  */
1299                 ia64_highfp_save(td);
1300                 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1301                 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1302         }
1303         save_callee_saved(&mc->mc_preserved);
1304         save_callee_saved_fp(&mc->mc_preserved_fp);
1305         return (error);
1306 }
1307
1308 int
1309 set_mcontext(struct thread *td, const mcontext_t *mc)
1310 {
1311         struct _special s;
1312         struct trapframe *tf;
1313         uint64_t psrmask;
1314
1315         tf = td->td_frame;
1316
1317         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1318             ("Whoa there! We have more than 8KB of dirty registers!"));
1319
1320         s = mc->mc_special;
1321         /*
1322          * Only copy the user mask and the restart instruction bit from
1323          * the new context.
1324          */
1325         psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1326             IA64_PSR_MFH | IA64_PSR_RI;
1327         s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1328         /* We don't have any dirty registers of the new context. */
1329         s.ndirty = 0;
1330         if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1331                 /*
1332                  * We can get an async context passed to us while we
1333                  * entered the kernel through a syscall: sigreturn(2)
1334                  * takes contexts that could previously be the result of
1335                  * a trap or interrupt.
1336                  * Hence, we cannot assert that the trapframe is not
1337                  * a syscall frame, but we can assert that it's at
1338                  * least an expected syscall.
1339                  */
1340                 if (tf->tf_flags & FRAME_SYSCALL) {
1341                         KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1342                         tf->tf_flags &= ~FRAME_SYSCALL;
1343                 }
1344                 tf->tf_scratch = mc->mc_scratch;
1345                 tf->tf_scratch_fp = mc->mc_scratch_fp;
1346                 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1347                         td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1348         } else {
1349                 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1350                 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1351                         s.cfm = tf->tf_special.cfm;
1352                         s.iip = tf->tf_special.iip;
1353                         tf->tf_scratch.gr15 = 0;        /* Clear syscall nr. */
1354                 } else
1355                         tf->tf_scratch = mc->mc_scratch;
1356         }
1357         tf->tf_special = s;
1358         restore_callee_saved(&mc->mc_preserved);
1359         restore_callee_saved_fp(&mc->mc_preserved_fp);
1360
1361         return (0);
1362 }
1363
1364 /*
1365  * Clear registers on exec.
1366  */
1367 void
1368 exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
1369 {
1370         struct trapframe *tf;
1371         uint64_t *ksttop, *kst;
1372
1373         tf = td->td_frame;
1374         ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1375             (tf->tf_special.bspstore & 0x1ffUL));
1376
1377         /*
1378          * We can ignore up to 8KB of dirty registers by masking off the
1379          * lower 13 bits in exception_restore() or epc_syscall(). This
1380          * should be enough for a couple of years, but if there are more
1381          * than 8KB of dirty registers, we lose track of the bottom of
1382          * the kernel stack. The solution is to copy the active part of
1383          * the kernel stack down 1 page (or 2, but not more than that)
1384          * so that we always have less than 8KB of dirty registers.
1385          */
1386         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1387             ("Whoa there! We have more than 8KB of dirty registers!"));
1388
1389         bzero(&tf->tf_special, sizeof(tf->tf_special));
1390         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {      /* break syscalls. */
1391                 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1392                 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1393                 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1394                 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1395                 /*
1396                  * Copy the arguments onto the kernel register stack so that
1397                  * they get loaded by the loadrs instruction. Skip over the
1398                  * NaT collection points.
1399                  */
1400                 kst = ksttop - 1;
1401                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1402                         *kst-- = 0;
1403                 *kst-- = 0;
1404                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1405                         *kst-- = 0;
1406                 *kst-- = imgp->ps_strings;
1407                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1408                         *kst-- = 0;
1409                 *kst = stack;
1410                 tf->tf_special.ndirty = (ksttop - kst) << 3;
1411         } else {                                /* epc syscalls (default). */
1412                 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1413                 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1414                 /*
1415                  * Write values for out0, out1 and out2 to the user's backing
1416                  * store and arrange for them to be restored into the user's
1417                  * initial register frame.
1418                  * Assumes that (bspstore & 0x1f8) < 0x1e0.
1419                  */
1420                 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1421                 suword((caddr_t)tf->tf_special.bspstore - 16, imgp->ps_strings);
1422                 suword((caddr_t)tf->tf_special.bspstore -  8, 0);
1423         }
1424
1425         tf->tf_special.iip = imgp->entry_addr;
1426         tf->tf_special.sp = (stack & ~15) - 16;
1427         tf->tf_special.rsc = 0xf;
1428         tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1429         tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1430             IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1431             IA64_PSR_CPL_USER;
1432 }
1433
1434 int
1435 ptrace_set_pc(struct thread *td, unsigned long addr)
1436 {
1437         uint64_t slot;
1438
1439         switch (addr & 0xFUL) {
1440         case 0:
1441                 slot = IA64_PSR_RI_0;
1442                 break;
1443         case 1:
1444                 /* XXX we need to deal with MLX bundles here */
1445                 slot = IA64_PSR_RI_1;
1446                 break;
1447         case 2:
1448                 slot = IA64_PSR_RI_2;
1449                 break;
1450         default:
1451                 return (EINVAL);
1452         }
1453
1454         td->td_frame->tf_special.iip = addr & ~0x0FULL;
1455         td->td_frame->tf_special.psr =
1456             (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1457         return (0);
1458 }
1459
1460 int
1461 ptrace_single_step(struct thread *td)
1462 {
1463         struct trapframe *tf;
1464
1465         /*
1466          * There's no way to set single stepping when we're leaving the
1467          * kernel through the EPC syscall path. The way we solve this is
1468          * by enabling the lower-privilege trap so that we re-enter the
1469          * kernel as soon as the privilege level changes. See trap.c for
1470          * how we proceed from there.
1471          */
1472         tf = td->td_frame;
1473         if (tf->tf_flags & FRAME_SYSCALL)
1474                 tf->tf_special.psr |= IA64_PSR_LP;
1475         else
1476                 tf->tf_special.psr |= IA64_PSR_SS;
1477         return (0);
1478 }
1479
1480 int
1481 ptrace_clear_single_step(struct thread *td)
1482 {
1483         struct trapframe *tf;
1484
1485         /*
1486          * Clear any and all status bits we may use to implement single
1487          * stepping.
1488          */
1489         tf = td->td_frame;
1490         tf->tf_special.psr &= ~IA64_PSR_SS;
1491         tf->tf_special.psr &= ~IA64_PSR_LP;
1492         tf->tf_special.psr &= ~IA64_PSR_TB;
1493         return (0);
1494 }
1495
1496 int
1497 fill_regs(struct thread *td, struct reg *regs)
1498 {
1499         struct trapframe *tf;
1500
1501         tf = td->td_frame;
1502         regs->r_special = tf->tf_special;
1503         regs->r_scratch = tf->tf_scratch;
1504         save_callee_saved(&regs->r_preserved);
1505         return (0);
1506 }
1507
1508 int
1509 set_regs(struct thread *td, struct reg *regs)
1510 {
1511         struct trapframe *tf;
1512         int error;
1513
1514         tf = td->td_frame;
1515         error = ia64_flush_dirty(td, &tf->tf_special);
1516         if (!error) {
1517                 tf->tf_special = regs->r_special;
1518                 tf->tf_special.bspstore += tf->tf_special.ndirty;
1519                 tf->tf_special.ndirty = 0;
1520                 tf->tf_scratch = regs->r_scratch;
1521                 restore_callee_saved(&regs->r_preserved);
1522         }
1523         return (error);
1524 }
1525
1526 int
1527 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1528 {
1529
1530         return (ENOSYS);
1531 }
1532
1533 int
1534 set_dbregs(struct thread *td, struct dbreg *dbregs)
1535 {
1536
1537         return (ENOSYS);
1538 }
1539
1540 int
1541 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1542 {
1543         struct trapframe *frame = td->td_frame;
1544         struct pcb *pcb = td->td_pcb;
1545
1546         /* Save the high FP registers. */
1547         ia64_highfp_save(td);
1548
1549         fpregs->fpr_scratch = frame->tf_scratch_fp;
1550         save_callee_saved_fp(&fpregs->fpr_preserved);
1551         fpregs->fpr_high = pcb->pcb_high_fp;
1552         return (0);
1553 }
1554
1555 int
1556 set_fpregs(struct thread *td, struct fpreg *fpregs)
1557 {
1558         struct trapframe *frame = td->td_frame;
1559         struct pcb *pcb = td->td_pcb;
1560
1561         /* Throw away the high FP registers (should be redundant). */
1562         ia64_highfp_drop(td);
1563
1564         frame->tf_scratch_fp = fpregs->fpr_scratch;
1565         restore_callee_saved_fp(&fpregs->fpr_preserved);
1566         pcb->pcb_high_fp = fpregs->fpr_high;
1567         return (0);
1568 }
1569
1570 void
1571 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1572 {
1573         vm_offset_t lim;
1574
1575         if (!ia64_sync_icache_needed)
1576                 return;
1577
1578         lim = va + sz;
1579         while (va < lim) {
1580                 ia64_fc_i(va);
1581                 va += 32;       /* XXX */
1582         }
1583
1584         ia64_sync_i();
1585         ia64_srlz_i();
1586 }