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_msgbuf.h"
  35 #include "opt_sched.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/proc.h>
  39 #include <sys/systm.h>
  40 #include <sys/bio.h>
  41 #include <sys/buf.h>
  42 #include <sys/bus.h>
  43 #include <sys/cons.h>
  44 #include <sys/cpu.h>
  45 #include <sys/eventhandler.h>
  46 #include <sys/exec.h>
  47 #include <sys/imgact.h>
  48 #include <sys/kdb.h>
  49 #include <sys/kernel.h>
  50 #include <sys/linker.h>
  51 #include <sys/lock.h>
  52 #include <sys/malloc.h>
  53 #include <sys/mbuf.h>
  54 #include <sys/msgbuf.h>
  55 #include <sys/pcpu.h>
  56 #include <sys/ptrace.h>
  57 #include <sys/random.h>
  58 #include <sys/reboot.h>
  59 #include <sys/sched.h>
  60 #include <sys/signalvar.h>
  61 #include <sys/syscall.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/clock.h>
  84 #include <machine/cpu.h>
  85 #include <machine/efi.h>
  86 #include <machine/elf.h>
  87 #include <machine/fpu.h>
  88 #include <machine/mca.h>
  89 #include <machine/md_var.h>
  90 #include <machine/mutex.h>
  91 #include <machine/pal.h>
  92 #include <machine/pcb.h>
  93 #include <machine/reg.h>
  94 #include <machine/sal.h>
  95 #include <machine/sigframe.h>
  96 #ifdef SMP
  97 #include <machine/smp.h>
  98 #endif
  99 #include <machine/unwind.h>
 100 #include <machine/vmparam.h>
 101
 102 #include <i386/include/specialreg.h>
 103
 104 SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, "");
 105
 106 u_int64_t processor_frequency;
 107 u_int64_t bus_frequency;
 108 u_int64_t itc_frequency;
 109 int cold = 1;
 110
 111 u_int64_t pa_bootinfo;
 112 struct bootinfo bootinfo;
 113
 114 struct pcpu pcpu0;
 115
 116 extern u_int64_t kernel_text[], _end[];
 117
 118 extern u_int64_t ia64_gateway_page[];
 119 extern u_int64_t break_sigtramp[];
 120 extern u_int64_t epc_sigtramp[];
 121
 122 struct fpswa_iface *fpswa_iface;
 123
 124 u_int64_t ia64_pal_base;
 125 u_int64_t ia64_port_base;
 126
 127 static int ia64_sync_icache_needed;
 128
 129 char machine[] = MACHINE;
 130 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
 131
 132 static char cpu_model[64];
 133 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
 134     "The CPU model name");
 135
 136 static char cpu_family[64];
 137 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
 138     "The CPU family name");
 139
 140 #ifdef DDB
 141 extern vm_offset_t ksym_start, ksym_end;
 142 #endif
 143
 144
 145 struct msgbuf *msgbufp = NULL;
 146
 147 /* Other subsystems (e.g., ACPI) can hook this later. */
 148 void (*cpu_idle_hook)(void) = NULL;
 149
 150 long Maxmem = 0;
 151 long realmem = 0;
 152
 153 #define PHYSMAP_SIZE    (2 * VM_PHYSSEG_MAX)
 154
 155 vm_paddr_t phys_avail[PHYSMAP_SIZE + 2];
 156
 157 /* must be 2 less so 0 0 can signal end of chunks */
 158 #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2)
 159
 160 struct kva_md_info kmi;
 161
 162 #define Mhz     1000000L
 163 #define Ghz     (1000L*Mhz)
 164
 165 static void
 166 identifycpu(void)
 167 {
 168         char vendor[17];
 169         char *family_name, *model_name;
 170         u_int64_t features, tmp;
 171         int number, revision, model, family, archrev;
 172
 173         /*
 174          * Assumes little-endian.
 175          */
 176         *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
 177         *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
 178         vendor[16] = '\0';
 179
 180         tmp = ia64_get_cpuid(3);
 181         number = (tmp >> 0) & 0xff;
 182         revision = (tmp >> 8) & 0xff;
 183         model = (tmp >> 16) & 0xff;
 184         family = (tmp >> 24) & 0xff;
 185         archrev = (tmp >> 32) & 0xff;
 186
 187         family_name = model_name = "unknown";
 188         switch (family) {
 189         case 0x07:
 190                 family_name = "Itanium";
 191                 model_name = "Merced";
 192                 break;
 193         case 0x1f:
 194                 family_name = "Itanium 2";
 195                 switch (model) {
 196                 case 0x00:
 197                         model_name = "McKinley";
 198                         break;
 199                 case 0x01:
 200                         /*
 201                          * Deerfield is a low-voltage variant based on the
 202                          * Madison core. We need circumstantial evidence
 203                          * (i.e. the clock frequency) to identify those.
 204                          * Allow for roughly 1% error margin.
 205                          */
 206                         tmp = processor_frequency >> 7;
 207                         if ((processor_frequency - tmp) < 1*Ghz &&
 208                             (processor_frequency + tmp) >= 1*Ghz)
 209                                 model_name = "Deerfield";
 210                         else
 211                                 model_name = "Madison";
 212                         break;
 213                 case 0x02:
 214                         model_name = "Madison II";
 215                         break;
 216                 }
 217                 break;
 218         case 0x20:
 219                 ia64_sync_icache_needed = 1;
 220
 221                 family_name = "Itanium 2";
 222                 switch (model) {
 223                 case 0x00:
 224                         model_name = "Montecito";
 225                         break;
 226                 }
 227                 break;
 228         }
 229         snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
 230         snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
 231
 232         features = ia64_get_cpuid(4);
 233
 234         printf("CPU: %s (", model_name);
 235         if (processor_frequency) {
 236                 printf("%ld.%02ld-Mhz ",
 237                     (processor_frequency + 4999) / Mhz,
 238                     ((processor_frequency + 4999) / (Mhz/100)) % 100);
 239         }
 240         printf("%s)\n", family_name);
 241         printf("  Origin = \"%s\"  Revision = %d\n", vendor, revision);
 242         printf("  Features = 0x%b\n", (u_int32_t) features,
 243             "\020"
 244             "\001LB"    /* long branch (brl) instruction. */
 245             "\002SD"    /* Spontaneous deferral. */
 246             "\003AO"    /* 16-byte atomic operations (ld, st, cmpxchg). */ );
 247 }
 248
 249 static void
 250 cpu_startup(void *dummy)
 251 {
 252         char nodename[16];
 253         struct pcpu *pc;
 254         struct pcpu_stats *pcs;
 255
 256         /*
 257          * Good {morning,afternoon,evening,night}.
 258          */
 259         identifycpu();
 260
 261 #ifdef PERFMON
 262         perfmon_init();
 263 #endif
 264         printf("real memory  = %ld (%ld MB)\n", ia64_ptob(Maxmem),
 265             ia64_ptob(Maxmem) / 1048576);
 266         realmem = Maxmem;
 267
 268         /*
 269          * Display any holes after the first chunk of extended memory.
 270          */
 271         if (bootverbose) {
 272                 int indx;
 273
 274                 printf("Physical memory chunk(s):\n");
 275                 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
 276                         long size1 = phys_avail[indx + 1] - phys_avail[indx];
 277
 278                         printf("0x%08lx - 0x%08lx, %ld bytes (%ld pages)\n",
 279                             phys_avail[indx], phys_avail[indx + 1] - 1, size1,
 280                             size1 >> PAGE_SHIFT);
 281                 }
 282         }
 283
 284         vm_ksubmap_init(&kmi);
 285
 286         printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count),
 287             ptoa(cnt.v_free_count) / 1048576);
 288
 289         if (fpswa_iface == NULL)
 290                 printf("Warning: no FPSWA package supplied\n");
 291         else
 292                 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
 293                     (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa);
 294
 295         /*
 296          * Set up buffers, so they can be used to read disk labels.
 297          */
 298         bufinit();
 299         vm_pager_bufferinit();
 300
 301         /*
 302          * Traverse the MADT to discover IOSAPIC and Local SAPIC
 303          * information.
 304          */
 305         ia64_probe_sapics();
 306         ia64_mca_init();
 307
 308         /*
 309          * Create sysctl tree for per-CPU information.
 310          */
 311         SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
 312                 snprintf(nodename, sizeof(nodename), "%u", pc->pc_cpuid);
 313                 sysctl_ctx_init(&pc->pc_md.sysctl_ctx);
 314                 pc->pc_md.sysctl_tree = SYSCTL_ADD_NODE(&pc->pc_md.sysctl_ctx,
 315                     SYSCTL_STATIC_CHILDREN(_machdep_cpu), OID_AUTO, nodename,
 316                     CTLFLAG_RD, NULL, "");
 317                 if (pc->pc_md.sysctl_tree == NULL)
 318                         continue;
 319
 320                 pcs = &pc->pc_md.stats;
 321
 322                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 323                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 324                     "nasts", CTLFLAG_RD, &pcs->pcs_nasts,
 325                     "Number of IPI_AST interrupts");
 326
 327                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 328                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 329                     "nclks", CTLFLAG_RD, &pcs->pcs_nclks,
 330                     "Number of clock interrupts");
 331
 332                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 333                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 334                     "nextints", CTLFLAG_RD, &pcs->pcs_nextints,
 335                     "Number of ExtINT interrupts");
 336
 337                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 338                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 339                     "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps,
 340                     "Number of IPI_HIGH_FP interrupts");
 341
 342                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 343                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 344                     "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints,
 345                     "Number of hardware (device) interrupts");
 346
 347                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 348                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 349                     "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts,
 350                     "Number of IPI_PREEMPT interrupts");
 351
 352                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 353                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 354                     "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs,
 355                     "Number of IPI_RENDEZVOUS interrupts");
 356
 357                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 358                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 359                     "nstops", CTLFLAG_RD, &pcs->pcs_nstops,
 360                     "Number of IPI_STOP interrupts");
 361
 362                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 363                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 364                     "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays,
 365                     "Number of stray vectors");
 366         }
 367 }
 368 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
 369
 370 void
 371 cpu_boot(int howto)
 372 {
 373
 374         efi_reset_system();
 375 }
 376
 377 void
 378 cpu_flush_dcache(void *ptr, size_t len)
 379 {
 380         vm_offset_t lim, va;
 381
 382         va = (uintptr_t)ptr & ~31;
 383         lim = (uintptr_t)ptr + len;
 384         while (va < lim) {
 385                 ia64_fc(va);
 386                 va += 32;
 387         }
 388
 389         ia64_srlz_d();
 390 }
 391
 392 /* Get current clock frequency for the given cpu id. */
 393 int
 394 cpu_est_clockrate(int cpu_id, uint64_t *rate)
 395 {
 396
 397         if (pcpu_find(cpu_id) == NULL || rate == NULL)
 398                 return (EINVAL);
 399         *rate = processor_frequency;
 400         return (0);
 401 }
 402
 403 void
 404 cpu_halt()
 405 {
 406
 407         efi_reset_system();
 408 }
 409
 410 void
 411 cpu_idle(int busy)
 412 {
 413         struct ia64_pal_result res;
 414
 415         if (cpu_idle_hook != NULL)
 416                 (*cpu_idle_hook)();
 417         else
 418                 res = ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
 419 }
 420
 421 int
 422 cpu_idle_wakeup(int cpu)
 423 {
 424
 425         return (0);
 426 }
 427
 428 void
 429 cpu_reset()
 430 {
 431
 432         cpu_boot(0);
 433 }
 434
 435 void
 436 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
 437 {
 438         struct pcb *oldpcb, *newpcb;
 439
 440         oldpcb = old->td_pcb;
 441 #ifdef COMPAT_IA32
 442         ia32_savectx(oldpcb);
 443 #endif
 444         if (PCPU_GET(fpcurthread) == old)
 445                 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
 446         if (!savectx(oldpcb)) {
 447                 old->td_lock = mtx;
 448 #if defined(SCHED_ULE) && defined(SMP)
 449                 /* td_lock is volatile */
 450                 while (new->td_lock == &blocked_lock)
 451                         ;
 452 #endif
 453                 newpcb = new->td_pcb;
 454                 oldpcb->pcb_current_pmap =
 455                     pmap_switch(newpcb->pcb_current_pmap);
 456                 PCPU_SET(curthread, new);
 457 #ifdef COMPAT_IA32
 458                 ia32_restorectx(newpcb);
 459 #endif
 460                 if (PCPU_GET(fpcurthread) == new)
 461                         new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
 462                 restorectx(newpcb);
 463                 /* We should not get here. */
 464                 panic("cpu_switch: restorectx() returned");
 465                 /* NOTREACHED */
 466         }
 467 }
 468
 469 void
 470 cpu_throw(struct thread *old __unused, struct thread *new)
 471 {
 472         struct pcb *newpcb;
 473
 474         newpcb = new->td_pcb;
 475         (void)pmap_switch(newpcb->pcb_current_pmap);
 476         PCPU_SET(curthread, new);
 477 #ifdef COMPAT_IA32
 478         ia32_restorectx(newpcb);
 479 #endif
 480         restorectx(newpcb);
 481         /* We should not get here. */
 482         panic("cpu_throw: restorectx() returned");
 483         /* NOTREACHED */
 484 }
 485
 486 void
 487 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
 488 {
 489
 490         /*
 491          * Set pc_acpi_id to "uninitialized".
 492          * See sys/dev/acpica/acpi_cpu.c
 493          */
 494         pcpu->pc_acpi_id = 0xffffffff;
 495 }
 496
 497 void
 498 spinlock_enter(void)
 499 {
 500         struct thread *td;
 501
 502         td = curthread;
 503         if (td->td_md.md_spinlock_count == 0)
 504                 td->td_md.md_saved_intr = intr_disable();
 505         td->td_md.md_spinlock_count++;
 506         critical_enter();
 507 }
 508
 509 void
 510 spinlock_exit(void)
 511 {
 512         struct thread *td;
 513
 514         td = curthread;
 515         critical_exit();
 516         td->td_md.md_spinlock_count--;
 517         if (td->td_md.md_spinlock_count == 0)
 518                 intr_restore(td->td_md.md_saved_intr);
 519 }
 520
 521 void
 522 map_vhpt(uintptr_t vhpt)
 523 {
 524         pt_entry_t pte;
 525         uint64_t psr;
 526
 527         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 528             PTE_PL_KERN | PTE_AR_RW;
 529         pte |= vhpt & PTE_PPN_MASK;
 530
 531         __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
 532             "r"(IA64_ID_PAGE_SHIFT<<2));
 533
 534         __asm __volatile("mov   %0=psr" : "=r"(psr));
 535         __asm __volatile("rsm   psr.ic|psr.i");
 536         ia64_srlz_i();
 537         ia64_set_ifa(vhpt);
 538         ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
 539         ia64_srlz_d();
 540         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(2), "r"(pte));
 541         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 542         ia64_srlz_i();
 543 }
 544
 545 void
 546 map_pal_code(void)
 547 {
 548         pt_entry_t pte;
 549         uint64_t psr;
 550
 551         if (ia64_pal_base == 0)
 552                 return;
 553
 554         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 555             PTE_PL_KERN | PTE_AR_RWX;
 556         pte |= ia64_pal_base & PTE_PPN_MASK;
 557
 558         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
 559             "r"(IA64_PHYS_TO_RR7(ia64_pal_base)), "r"(IA64_ID_PAGE_SHIFT<<2));
 560
 561         __asm __volatile("mov   %0=psr" : "=r"(psr));
 562         __asm __volatile("rsm   psr.ic|psr.i");
 563         ia64_srlz_i();
 564         ia64_set_ifa(IA64_PHYS_TO_RR7(ia64_pal_base));
 565         ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
 566         ia64_srlz_d();
 567         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(1), "r"(pte));
 568         ia64_srlz_d();
 569         __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
 570         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 571         ia64_srlz_i();
 572 }
 573
 574 void
 575 map_gateway_page(void)
 576 {
 577         pt_entry_t pte;
 578         uint64_t psr;
 579
 580         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 581             PTE_PL_KERN | PTE_AR_X_RX;
 582         pte |= (uint64_t)ia64_gateway_page & PTE_PPN_MASK;
 583
 584         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
 585             "r"(VM_MAX_ADDRESS), "r"(PAGE_SHIFT << 2));
 586
 587         __asm __volatile("mov   %0=psr" : "=r"(psr));
 588         __asm __volatile("rsm   psr.ic|psr.i");
 589         ia64_srlz_i();
 590         ia64_set_ifa(VM_MAX_ADDRESS);
 591         ia64_set_itir(PAGE_SHIFT << 2);
 592         ia64_srlz_d();
 593         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
 594         ia64_srlz_d();
 595         __asm __volatile("itr.i itr[%0]=%1" :: "r"(3), "r"(pte));
 596         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 597         ia64_srlz_i();
 598
 599         /* Expose the mapping to userland in ar.k5 */
 600         ia64_set_k5(VM_MAX_ADDRESS);
 601 }
 602
 603 static void
 604 calculate_frequencies(void)
 605 {
 606         struct ia64_sal_result sal;
 607         struct ia64_pal_result pal;
 608
 609         sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
 610         pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
 611
 612         if (sal.sal_status == 0 && pal.pal_status == 0) {
 613                 if (bootverbose) {
 614                         printf("Platform clock frequency %ld Hz\n",
 615                                sal.sal_result[0]);
 616                         printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
 617                                "ITC ratio %ld/%ld\n",
 618                                pal.pal_result[0] >> 32,
 619                                pal.pal_result[0] & ((1L << 32) - 1),
 620                                pal.pal_result[1] >> 32,
 621                                pal.pal_result[1] & ((1L << 32) - 1),
 622                                pal.pal_result[2] >> 32,
 623                                pal.pal_result[2] & ((1L << 32) - 1));
 624                 }
 625                 processor_frequency =
 626                         sal.sal_result[0] * (pal.pal_result[0] >> 32)
 627                         / (pal.pal_result[0] & ((1L << 32) - 1));
 628                 bus_frequency =
 629                         sal.sal_result[0] * (pal.pal_result[1] >> 32)
 630                         / (pal.pal_result[1] & ((1L << 32) - 1));
 631                 itc_frequency =
 632                         sal.sal_result[0] * (pal.pal_result[2] >> 32)
 633                         / (pal.pal_result[2] & ((1L << 32) - 1));
 634         }
 635 }
 636
 637 struct ia64_init_return
 638 ia64_init(void)
 639 {
 640         struct ia64_init_return ret;
 641         int phys_avail_cnt;
 642         vm_offset_t kernstart, kernend;
 643         vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1;
 644         char *p;
 645         struct efi_md *md;
 646         int metadata_missing;
 647
 648         /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */
 649
 650         /*
 651          * TODO: Disable interrupts, floating point etc.
 652          * Maybe flush cache and tlb
 653          */
 654         ia64_set_fpsr(IA64_FPSR_DEFAULT);
 655
 656         /*
 657          * TODO: Get critical system information (if possible, from the
 658          * information provided by the boot program).
 659          */
 660
 661         /*
 662          * pa_bootinfo is the physical address of the bootinfo block as
 663          * passed to us by the loader and set in locore.s.
 664          */
 665         bootinfo = *(struct bootinfo *)(IA64_PHYS_TO_RR7(pa_bootinfo));
 666
 667         if (bootinfo.bi_magic != BOOTINFO_MAGIC || bootinfo.bi_version != 1) {
 668                 bzero(&bootinfo, sizeof(bootinfo));
 669                 bootinfo.bi_kernend = (vm_offset_t) round_page(_end);
 670         }
 671
 672         /*
 673          * Look for the I/O ports first - we need them for console
 674          * probing.
 675          */
 676         for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
 677                 switch (md->md_type) {
 678                 case EFI_MD_TYPE_IOPORT:
 679                         ia64_port_base = IA64_PHYS_TO_RR6(md->md_phys);
 680                         break;
 681                 case EFI_MD_TYPE_PALCODE:
 682                         ia64_pal_base = md->md_phys;
 683                         break;
 684                 }
 685         }
 686
 687         metadata_missing = 0;
 688         if (bootinfo.bi_modulep)
 689                 preload_metadata = (caddr_t)bootinfo.bi_modulep;
 690         else
 691                 metadata_missing = 1;
 692
 693         if (envmode == 0 && bootinfo.bi_envp)
 694                 kern_envp = (caddr_t)bootinfo.bi_envp;
 695         else
 696                 kern_envp = static_env;
 697
 698         /*
 699          * Look at arguments passed to us and compute boothowto.
 700          */
 701         boothowto = bootinfo.bi_boothowto;
 702
 703         /*
 704          * Catch case of boot_verbose set in environment.
 705          */
 706         if ((p = getenv("boot_verbose")) != NULL) {
 707                 if (strcmp(p, "yes") == 0 || strcmp(p, "YES") == 0) {
 708                         boothowto |= RB_VERBOSE;
 709                 }
 710                 freeenv(p);
 711         }
 712
 713         if (boothowto & RB_VERBOSE)
 714                 bootverbose = 1;
 715
 716         /*
 717          * Find the beginning and end of the kernel.
 718          */
 719         kernstart = trunc_page(kernel_text);
 720 #ifdef DDB
 721         ksym_start = bootinfo.bi_symtab;
 722         ksym_end = bootinfo.bi_esymtab;
 723         kernend = (vm_offset_t)round_page(ksym_end);
 724 #else
 725         kernend = (vm_offset_t)round_page(_end);
 726 #endif
 727         /* But if the bootstrap tells us otherwise, believe it! */
 728         if (bootinfo.bi_kernend)
 729                 kernend = round_page(bootinfo.bi_kernend);
 730
 731         /*
 732          * Setup the PCPU data for the bootstrap processor. It is needed
 733          * by printf(). Also, since printf() has critical sections, we
 734          * need to initialize at least pc_curthread.
 735          */
 736         pcpup = &pcpu0;
 737         ia64_set_k4((u_int64_t)pcpup);
 738         pcpu_init(pcpup, 0, sizeof(pcpu0));
 739         dpcpu_init((void *)kernend, 0);
 740         kernend += DPCPU_SIZE;
 741         PCPU_SET(curthread, &thread0);
 742
 743         /*
 744          * Initialize the console before we print anything out.
 745          */
 746         cninit();
 747
 748         /* OUTPUT NOW ALLOWED */
 749
 750         if (ia64_pal_base != 0) {
 751                 ia64_pal_base &= ~IA64_ID_PAGE_MASK;
 752                 /*
 753                  * We use a TR to map the first 256M of memory - this might
 754                  * cover the palcode too.
 755                  */
 756                 if (ia64_pal_base == 0)
 757                         printf("PAL code mapped by the kernel's TR\n");
 758         } else
 759                 printf("PAL code not found\n");
 760
 761         /*
 762          * Wire things up so we can call the firmware.
 763          */
 764         map_pal_code();
 765         efi_boot_minimal(bootinfo.bi_systab);
 766         ia64_sal_init();
 767         calculate_frequencies();
 768
 769         if (metadata_missing)
 770                 printf("WARNING: loader(8) metadata is missing!\n");
 771
 772         /* Get FPSWA interface */
 773         fpswa_iface = (bootinfo.bi_fpswa == 0) ? NULL :
 774             (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo.bi_fpswa);
 775
 776         /* Init basic tunables, including hz */
 777         init_param1();
 778
 779         p = getenv("kernelname");
 780         if (p) {
 781                 strncpy(kernelname, p, sizeof(kernelname) - 1);
 782                 freeenv(p);
 783         }
 784
 785         kernstartpfn = atop(IA64_RR_MASK(kernstart));
 786         kernendpfn = atop(IA64_RR_MASK(kernend));
 787
 788         /*
 789          * Size the memory regions and load phys_avail[] with the results.
 790          */
 791
 792         /*
 793          * Find out how much memory is available, by looking at
 794          * the memory descriptors.
 795          */
 796
 797 #ifdef DEBUG_MD
 798         printf("Memory descriptor count: %d\n", mdcount);
 799 #endif
 800
 801         phys_avail_cnt = 0;
 802         for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
 803 #ifdef DEBUG_MD
 804                 printf("MD %p: type %d pa 0x%lx cnt 0x%lx\n", md,
 805                     md->md_type, md->md_phys, md->md_pages);
 806 #endif
 807
 808                 pfn0 = ia64_btop(round_page(md->md_phys));
 809                 pfn1 = ia64_btop(trunc_page(md->md_phys + md->md_pages * 4096));
 810                 if (pfn1 <= pfn0)
 811                         continue;
 812
 813                 if (md->md_type != EFI_MD_TYPE_FREE)
 814                         continue;
 815
 816                 /*
 817                  * We have a memory descriptor that describes conventional
 818                  * memory that is for general use. We must determine if the
 819                  * loader has put the kernel in this region.
 820                  */
 821                 physmem += (pfn1 - pfn0);
 822                 if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) {
 823                         /*
 824                          * Must compute the location of the kernel
 825                          * within the segment.
 826                          */
 827 #ifdef DEBUG_MD
 828                         printf("Descriptor %p contains kernel\n", mp);
 829 #endif
 830                         if (pfn0 < kernstartpfn) {
 831                                 /*
 832                                  * There is a chunk before the kernel.
 833                                  */
 834 #ifdef DEBUG_MD
 835                                 printf("Loading chunk before kernel: "
 836                                        "0x%lx / 0x%lx\n", pfn0, kernstartpfn);
 837 #endif
 838                                 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
 839                                 phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn);
 840                                 phys_avail_cnt += 2;
 841                         }
 842                         if (kernendpfn < pfn1) {
 843                                 /*
 844                                  * There is a chunk after the kernel.
 845                                  */
 846 #ifdef DEBUG_MD
 847                                 printf("Loading chunk after kernel: "
 848                                        "0x%lx / 0x%lx\n", kernendpfn, pfn1);
 849 #endif
 850                                 phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn);
 851                                 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
 852                                 phys_avail_cnt += 2;
 853                         }
 854                 } else {
 855                         /*
 856                          * Just load this cluster as one chunk.
 857                          */
 858 #ifdef DEBUG_MD
 859                         printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i,
 860                                pfn0, pfn1);
 861 #endif
 862                         phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
 863                         phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
 864                         phys_avail_cnt += 2;
 865
 866                 }
 867         }
 868         phys_avail[phys_avail_cnt] = 0;
 869
 870         Maxmem = physmem;
 871         init_param2(physmem);
 872
 873         /*
 874          * Initialize error message buffer (at end of core).
 875          */
 876         msgbufp = (struct msgbuf *)pmap_steal_memory(MSGBUF_SIZE);
 877         msgbufinit(msgbufp, MSGBUF_SIZE);
 878
 879         proc_linkup0(&proc0, &thread0);
 880         /*
 881          * Init mapping for kernel stack for proc 0
 882          */
 883         thread0.td_kstack = pmap_steal_memory(KSTACK_PAGES * PAGE_SIZE);
 884         thread0.td_kstack_pages = KSTACK_PAGES;
 885
 886         mutex_init();
 887
 888         /*
 889          * Initialize the rest of proc 0's PCB.
 890          *
 891          * Set the kernel sp, reserving space for an (empty) trapframe,
 892          * and make proc0's trapframe pointer point to it for sanity.
 893          * Initialise proc0's backing store to start after u area.
 894          */
 895         cpu_thread_alloc(&thread0);
 896         thread0.td_frame->tf_flags = FRAME_SYSCALL;
 897         thread0.td_pcb->pcb_special.sp =
 898             (u_int64_t)thread0.td_frame - 16;
 899         thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
 900
 901         /*
 902          * Initialize the virtual memory system.
 903          */
 904         pmap_bootstrap();
 905
 906         /*
 907          * Initialize debuggers, and break into them if appropriate.
 908          */
 909         kdb_init();
 910
 911 #ifdef KDB
 912         if (boothowto & RB_KDB)
 913                 kdb_enter(KDB_WHY_BOOTFLAGS,
 914                     "Boot flags requested debugger\n");
 915 #endif
 916
 917         ia64_set_tpr(0);
 918         ia64_srlz_d();
 919
 920         ret.bspstore = thread0.td_pcb->pcb_special.bspstore;
 921         ret.sp = thread0.td_pcb->pcb_special.sp;
 922         return (ret);
 923 }
 924
 925 void *
 926 ia64_ioport_address(u_int port)
 927 {
 928         uint64_t addr;
 929
 930         addr = (port > 0xffff) ? IA64_PHYS_TO_RR6((uint64_t)port) :
 931             ia64_port_base | ((port & 0xfffc) << 10) | (port & 0xFFF);
 932         return ((void *)addr);
 933 }
 934
 935 uint64_t
 936 ia64_get_hcdp(void)
 937 {
 938
 939         return (bootinfo.bi_hcdp);
 940 }
 941
 942 void
 943 bzero(void *buf, size_t len)
 944 {
 945         caddr_t p = buf;
 946
 947         while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
 948                 *p++ = 0;
 949                 len--;
 950         }
 951         while (len >= sizeof(u_long) * 8) {
 952                 *(u_long*) p = 0;
 953                 *((u_long*) p + 1) = 0;
 954                 *((u_long*) p + 2) = 0;
 955                 *((u_long*) p + 3) = 0;
 956                 len -= sizeof(u_long) * 8;
 957                 *((u_long*) p + 4) = 0;
 958                 *((u_long*) p + 5) = 0;
 959                 *((u_long*) p + 6) = 0;
 960                 *((u_long*) p + 7) = 0;
 961                 p += sizeof(u_long) * 8;
 962         }
 963         while (len >= sizeof(u_long)) {
 964                 *(u_long*) p = 0;
 965                 len -= sizeof(u_long);
 966                 p += sizeof(u_long);
 967         }
 968         while (len) {
 969                 *p++ = 0;
 970                 len--;
 971         }
 972 }
 973
 974 void
 975 DELAY(int n)
 976 {
 977         u_int64_t start, end, now;
 978
 979         sched_pin();
 980
 981         start = ia64_get_itc();
 982         end = start + (itc_frequency * n) / 1000000;
 983         /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
 984         do {
 985                 now = ia64_get_itc();
 986         } while (now < end || (now > start && end < start));
 987
 988         sched_unpin();
 989 }
 990
 991 /*
 992  * Send an interrupt (signal) to a process.
 993  */
 994 void
 995 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
 996 {
 997         struct proc *p;
 998         struct thread *td;
 999         struct trapframe *tf;
1000         struct sigacts *psp;
1001         struct sigframe sf, *sfp;
1002         u_int64_t sbs, sp;
1003         int oonstack;
1004         int sig;
1005         u_long code;
1006
1007         td = curthread;
1008         p = td->td_proc;
1009         PROC_LOCK_ASSERT(p, MA_OWNED);
1010         sig = ksi->ksi_signo;
1011         code = ksi->ksi_code;
1012         psp = p->p_sigacts;
1013         mtx_assert(&psp->ps_mtx, MA_OWNED);
1014         tf = td->td_frame;
1015         sp = tf->tf_special.sp;
1016         oonstack = sigonstack(sp);
1017         sbs = 0;
1018
1019         /* save user context */
1020         bzero(&sf, sizeof(struct sigframe));
1021         sf.sf_uc.uc_sigmask = *mask;
1022         sf.sf_uc.uc_stack = td->td_sigstk;
1023         sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1024             ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1025
1026         /*
1027          * Allocate and validate space for the signal handler
1028          * context. Note that if the stack is in P0 space, the
1029          * call to grow() is a nop, and the useracc() check
1030          * will fail if the process has not already allocated
1031          * the space with a `brk'.
1032          */
1033         if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1034             SIGISMEMBER(psp->ps_sigonstack, sig)) {
1035                 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1036                 sbs = (sbs + 15) & ~15;
1037                 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1038 #if defined(COMPAT_43)
1039                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1040 #endif
1041         } else
1042                 sfp = (struct sigframe *)sp;
1043         sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1044
1045         /* Fill in the siginfo structure for POSIX handlers. */
1046         if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1047                 sf.sf_si = ksi->ksi_info;
1048                 sf.sf_si.si_signo = sig;
1049                 /*
1050                  * XXX this shouldn't be here after code in trap.c
1051                  * is fixed
1052                  */
1053                 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1054                 code = (u_int64_t)&sfp->sf_si;
1055         }
1056
1057         mtx_unlock(&psp->ps_mtx);
1058         PROC_UNLOCK(p);
1059
1060         get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1061
1062         /* Copy the frame out to userland. */
1063         if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1064                 /*
1065                  * Process has trashed its stack; give it an illegal
1066                  * instruction to halt it in its tracks.
1067                  */
1068                 PROC_LOCK(p);
1069                 sigexit(td, SIGILL);
1070                 return;
1071         }
1072
1073         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1074                 tf->tf_special.psr &= ~IA64_PSR_RI;
1075                 tf->tf_special.iip = ia64_get_k5() +
1076                     ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1077         } else
1078                 tf->tf_special.iip = ia64_get_k5() +
1079                     ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1080
1081         /*
1082          * Setup the trapframe to return to the signal trampoline. We pass
1083          * information to the trampoline in the following registers:
1084          *
1085          *      gp      new backing store or NULL
1086          *      r8      signal number
1087          *      r9      signal code or siginfo pointer
1088          *      r10     signal handler (function descriptor)
1089          */
1090         tf->tf_special.sp = (u_int64_t)sfp - 16;
1091         tf->tf_special.gp = sbs;
1092         tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1093         tf->tf_special.ndirty = 0;
1094         tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1095         tf->tf_scratch.gr8 = sig;
1096         tf->tf_scratch.gr9 = code;
1097         tf->tf_scratch.gr10 = (u_int64_t)catcher;
1098
1099         PROC_LOCK(p);
1100         mtx_lock(&psp->ps_mtx);
1101 }
1102
1103 /*
1104  * System call to cleanup state after a signal
1105  * has been taken.  Reset signal mask and
1106  * stack state from context left by sendsig (above).
1107  * Return to previous pc and psl as specified by
1108  * context left by sendsig. Check carefully to
1109  * make sure that the user has not modified the
1110  * state to gain improper privileges.
1111  *
1112  * MPSAFE
1113  */
1114 int
1115 sigreturn(struct thread *td,
1116         struct sigreturn_args /* {
1117                 ucontext_t *sigcntxp;
1118         } */ *uap)
1119 {
1120         ucontext_t uc;
1121         struct trapframe *tf;
1122         struct proc *p;
1123         struct pcb *pcb;
1124
1125         tf = td->td_frame;
1126         p = td->td_proc;
1127         pcb = td->td_pcb;
1128
1129         /*
1130          * Fetch the entire context structure at once for speed.
1131          * We don't use a normal argument to simplify RSE handling.
1132          */
1133         if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1134                 return (EFAULT);
1135
1136         set_mcontext(td, &uc.uc_mcontext);
1137
1138         PROC_LOCK(p);
1139 #if defined(COMPAT_43)
1140         if (sigonstack(tf->tf_special.sp))
1141                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1142         else
1143                 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1144 #endif
1145         td->td_sigmask = uc.uc_sigmask;
1146         SIG_CANTMASK(td->td_sigmask);
1147         signotify(td);
1148         PROC_UNLOCK(p);
1149
1150         return (EJUSTRETURN);
1151 }
1152
1153 #ifdef COMPAT_FREEBSD4
1154 int
1155 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1156 {
1157
1158         return sigreturn(td, (struct sigreturn_args *)uap);
1159 }
1160 #endif
1161
1162 /*
1163  * Construct a PCB from a trapframe. This is called from kdb_trap() where
1164  * we want to start a backtrace from the function that caused us to enter
1165  * the debugger. We have the context in the trapframe, but base the trace
1166  * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1167  * enough for a backtrace.
1168  */
1169 void
1170 makectx(struct trapframe *tf, struct pcb *pcb)
1171 {
1172
1173         pcb->pcb_special = tf->tf_special;
1174         pcb->pcb_special.__spare = ~0UL;        /* XXX see unwind.c */
1175         save_callee_saved(&pcb->pcb_preserved);
1176         save_callee_saved_fp(&pcb->pcb_preserved_fp);
1177 }
1178
1179 int
1180 ia64_flush_dirty(struct thread *td, struct _special *r)
1181 {
1182         struct iovec iov;
1183         struct uio uio;
1184         uint64_t bspst, kstk, rnat;
1185         int error, locked;
1186
1187         if (r->ndirty == 0)
1188                 return (0);
1189
1190         kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1191         if (td == curthread) {
1192                 __asm __volatile("mov   ar.rsc=0;;");
1193                 __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1194                 /* Make sure we have all the user registers written out. */
1195                 if (bspst - kstk < r->ndirty) {
1196                         __asm __volatile("flushrs;;");
1197                         __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1198                 }
1199                 __asm __volatile("mov   %0=ar.rnat;;" : "=r"(rnat));
1200                 __asm __volatile("mov   ar.rsc=3");
1201                 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1202                 kstk += r->ndirty;
1203                 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1204                     ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1205         } else {
1206                 locked = PROC_LOCKED(td->td_proc);
1207                 if (!locked)
1208                         PHOLD(td->td_proc);
1209                 iov.iov_base = (void*)(uintptr_t)kstk;
1210                 iov.iov_len = r->ndirty;
1211                 uio.uio_iov = &iov;
1212                 uio.uio_iovcnt = 1;
1213                 uio.uio_offset = r->bspstore;
1214                 uio.uio_resid = r->ndirty;
1215                 uio.uio_segflg = UIO_SYSSPACE;
1216                 uio.uio_rw = UIO_WRITE;
1217                 uio.uio_td = td;
1218                 error = proc_rwmem(td->td_proc, &uio);
1219                 /*
1220                  * XXX proc_rwmem() doesn't currently return ENOSPC,
1221                  * so I think it can bogusly return 0. Neither do
1222                  * we allow short writes.
1223                  */
1224                 if (uio.uio_resid != 0 && error == 0)
1225                         error = ENOSPC;
1226                 if (!locked)
1227                         PRELE(td->td_proc);
1228         }
1229
1230         r->bspstore += r->ndirty;
1231         r->ndirty = 0;
1232         return (error);
1233 }
1234
1235 int
1236 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1237 {
1238         struct trapframe *tf;
1239         int error;
1240
1241         tf = td->td_frame;
1242         bzero(mc, sizeof(*mc));
1243         mc->mc_special = tf->tf_special;
1244         error = ia64_flush_dirty(td, &mc->mc_special);
1245         if (tf->tf_flags & FRAME_SYSCALL) {
1246                 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1247                 mc->mc_scratch = tf->tf_scratch;
1248                 if (flags & GET_MC_CLEAR_RET) {
1249                         mc->mc_scratch.gr8 = 0;
1250                         mc->mc_scratch.gr9 = 0;
1251                         mc->mc_scratch.gr10 = 0;
1252                         mc->mc_scratch.gr11 = 0;
1253                 }
1254         } else {
1255                 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1256                 mc->mc_scratch = tf->tf_scratch;
1257                 mc->mc_scratch_fp = tf->tf_scratch_fp;
1258                 /*
1259                  * XXX If the thread never used the high FP registers, we
1260                  * probably shouldn't waste time saving them.
1261                  */
1262                 ia64_highfp_save(td);
1263                 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1264                 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1265         }
1266         save_callee_saved(&mc->mc_preserved);
1267         save_callee_saved_fp(&mc->mc_preserved_fp);
1268         return (error);
1269 }
1270
1271 int
1272 set_mcontext(struct thread *td, const mcontext_t *mc)
1273 {
1274         struct _special s;
1275         struct trapframe *tf;
1276         uint64_t psrmask;
1277
1278         tf = td->td_frame;
1279
1280         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1281             ("Whoa there! We have more than 8KB of dirty registers!"));
1282
1283         s = mc->mc_special;
1284         /*
1285          * Only copy the user mask and the restart instruction bit from
1286          * the new context.
1287          */
1288         psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1289             IA64_PSR_MFH | IA64_PSR_RI;
1290         s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1291         /* We don't have any dirty registers of the new context. */
1292         s.ndirty = 0;
1293         if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1294                 /*
1295                  * We can get an async context passed to us while we
1296                  * entered the kernel through a syscall: sigreturn(2)
1297                  * takes contexts that could previously be the result of
1298                  * a trap or interrupt.
1299                  * Hence, we cannot assert that the trapframe is not
1300                  * a syscall frame, but we can assert that it's at
1301                  * least an expected syscall.
1302                  */
1303                 if (tf->tf_flags & FRAME_SYSCALL) {
1304                         KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1305                         tf->tf_flags &= ~FRAME_SYSCALL;
1306                 }
1307                 tf->tf_scratch = mc->mc_scratch;
1308                 tf->tf_scratch_fp = mc->mc_scratch_fp;
1309                 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1310                         td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1311         } else {
1312                 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1313                 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1314                         s.cfm = tf->tf_special.cfm;
1315                         s.iip = tf->tf_special.iip;
1316                         tf->tf_scratch.gr15 = 0;        /* Clear syscall nr. */
1317                 } else
1318                         tf->tf_scratch = mc->mc_scratch;
1319         }
1320         tf->tf_special = s;
1321         restore_callee_saved(&mc->mc_preserved);
1322         restore_callee_saved_fp(&mc->mc_preserved_fp);
1323
1324         return (0);
1325 }
1326
1327 /*
1328  * Clear registers on exec.
1329  */
1330 void
1331 exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
1332 {
1333         struct trapframe *tf;
1334         uint64_t *ksttop, *kst;
1335
1336         tf = td->td_frame;
1337         ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1338             (tf->tf_special.bspstore & 0x1ffUL));
1339
1340         /*
1341          * We can ignore up to 8KB of dirty registers by masking off the
1342          * lower 13 bits in exception_restore() or epc_syscall(). This
1343          * should be enough for a couple of years, but if there are more
1344          * than 8KB of dirty registers, we lose track of the bottom of
1345          * the kernel stack. The solution is to copy the active part of
1346          * the kernel stack down 1 page (or 2, but not more than that)
1347          * so that we always have less than 8KB of dirty registers.
1348          */
1349         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1350             ("Whoa there! We have more than 8KB of dirty registers!"));
1351
1352         bzero(&tf->tf_special, sizeof(tf->tf_special));
1353         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {      /* break syscalls. */
1354                 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1355                 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1356                 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1357                 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1358                 /*
1359                  * Copy the arguments onto the kernel register stack so that
1360                  * they get loaded by the loadrs instruction. Skip over the
1361                  * NaT collection points.
1362                  */
1363                 kst = ksttop - 1;
1364                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1365                         *kst-- = 0;
1366                 *kst-- = 0;
1367                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1368                         *kst-- = 0;
1369                 *kst-- = ps_strings;
1370                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1371                         *kst-- = 0;
1372                 *kst = stack;
1373                 tf->tf_special.ndirty = (ksttop - kst) << 3;
1374         } else {                                /* epc syscalls (default). */
1375                 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1376                 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1377                 /*
1378                  * Write values for out0, out1 and out2 to the user's backing
1379                  * store and arrange for them to be restored into the user's
1380                  * initial register frame.
1381                  * Assumes that (bspstore & 0x1f8) < 0x1e0.
1382                  */
1383                 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1384                 suword((caddr_t)tf->tf_special.bspstore - 16, ps_strings);
1385                 suword((caddr_t)tf->tf_special.bspstore -  8, 0);
1386         }
1387
1388         tf->tf_special.iip = entry;
1389         tf->tf_special.sp = (stack & ~15) - 16;
1390         tf->tf_special.rsc = 0xf;
1391         tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1392         tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1393             IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1394             IA64_PSR_CPL_USER;
1395 }
1396
1397 int
1398 ptrace_set_pc(struct thread *td, unsigned long addr)
1399 {
1400         uint64_t slot;
1401
1402         switch (addr & 0xFUL) {
1403         case 0:
1404                 slot = IA64_PSR_RI_0;
1405                 break;
1406         case 1:
1407                 /* XXX we need to deal with MLX bundles here */
1408                 slot = IA64_PSR_RI_1;
1409                 break;
1410         case 2:
1411                 slot = IA64_PSR_RI_2;
1412                 break;
1413         default:
1414                 return (EINVAL);
1415         }
1416
1417         td->td_frame->tf_special.iip = addr & ~0x0FULL;
1418         td->td_frame->tf_special.psr =
1419             (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1420         return (0);
1421 }
1422
1423 int
1424 ptrace_single_step(struct thread *td)
1425 {
1426         struct trapframe *tf;
1427
1428         /*
1429          * There's no way to set single stepping when we're leaving the
1430          * kernel through the EPC syscall path. The way we solve this is
1431          * by enabling the lower-privilege trap so that we re-enter the
1432          * kernel as soon as the privilege level changes. See trap.c for
1433          * how we proceed from there.
1434          */
1435         tf = td->td_frame;
1436         if (tf->tf_flags & FRAME_SYSCALL)
1437                 tf->tf_special.psr |= IA64_PSR_LP;
1438         else
1439                 tf->tf_special.psr |= IA64_PSR_SS;
1440         return (0);
1441 }
1442
1443 int
1444 ptrace_clear_single_step(struct thread *td)
1445 {
1446         struct trapframe *tf;
1447
1448         /*
1449          * Clear any and all status bits we may use to implement single
1450          * stepping.
1451          */
1452         tf = td->td_frame;
1453         tf->tf_special.psr &= ~IA64_PSR_SS;
1454         tf->tf_special.psr &= ~IA64_PSR_LP;
1455         tf->tf_special.psr &= ~IA64_PSR_TB;
1456         return (0);
1457 }
1458
1459 int
1460 fill_regs(struct thread *td, struct reg *regs)
1461 {
1462         struct trapframe *tf;
1463
1464         tf = td->td_frame;
1465         regs->r_special = tf->tf_special;
1466         regs->r_scratch = tf->tf_scratch;
1467         save_callee_saved(&regs->r_preserved);
1468         return (0);
1469 }
1470
1471 int
1472 set_regs(struct thread *td, struct reg *regs)
1473 {
1474         struct trapframe *tf;
1475         int error;
1476
1477         tf = td->td_frame;
1478         error = ia64_flush_dirty(td, &tf->tf_special);
1479         if (!error) {
1480                 tf->tf_special = regs->r_special;
1481                 tf->tf_special.bspstore += tf->tf_special.ndirty;
1482                 tf->tf_special.ndirty = 0;
1483                 tf->tf_scratch = regs->r_scratch;
1484                 restore_callee_saved(&regs->r_preserved);
1485         }
1486         return (error);
1487 }
1488
1489 int
1490 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1491 {
1492
1493         return (ENOSYS);
1494 }
1495
1496 int
1497 set_dbregs(struct thread *td, struct dbreg *dbregs)
1498 {
1499
1500         return (ENOSYS);
1501 }
1502
1503 int
1504 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1505 {
1506         struct trapframe *frame = td->td_frame;
1507         struct pcb *pcb = td->td_pcb;
1508
1509         /* Save the high FP registers. */
1510         ia64_highfp_save(td);
1511
1512         fpregs->fpr_scratch = frame->tf_scratch_fp;
1513         save_callee_saved_fp(&fpregs->fpr_preserved);
1514         fpregs->fpr_high = pcb->pcb_high_fp;
1515         return (0);
1516 }
1517
1518 int
1519 set_fpregs(struct thread *td, struct fpreg *fpregs)
1520 {
1521         struct trapframe *frame = td->td_frame;
1522         struct pcb *pcb = td->td_pcb;
1523
1524         /* Throw away the high FP registers (should be redundant). */
1525         ia64_highfp_drop(td);
1526
1527         frame->tf_scratch_fp = fpregs->fpr_scratch;
1528         restore_callee_saved_fp(&fpregs->fpr_preserved);
1529         pcb->pcb_high_fp = fpregs->fpr_high;
1530         return (0);
1531 }
1532
1533 void
1534 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1535 {
1536         vm_offset_t lim;
1537
1538         if (!ia64_sync_icache_needed)
1539                 return;
1540
1541         lim = va + sz;
1542         while (va < lim) {
1543                 ia64_fc_i(va);
1544                 va += 32;       /* XXX */
1545         }
1546
1547         ia64_sync_i();
1548         ia64_srlz_i();
1549 }