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 #include "opt_xtrace.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/rwlock.h>
  60 #include <sys/sched.h>
  61 #include <sys/signalvar.h>
  62 #include <sys/syscall.h>
  63 #include <sys/syscallsubr.h>
  64 #include <sys/sysctl.h>
  65 #include <sys/sysproto.h>
  66 #include <sys/ucontext.h>
  67 #include <sys/uio.h>
  68 #include <sys/uuid.h>
  69 #include <sys/vmmeter.h>
  70 #include <sys/vnode.h>
  71
  72 #include <ddb/ddb.h>
  73
  74 #include <net/netisr.h>
  75
  76 #include <vm/vm.h>
  77 #include <vm/vm_extern.h>
  78 #include <vm/vm_kern.h>
  79 #include <vm/vm_page.h>
  80 #include <vm/vm_map.h>
  81 #include <vm/vm_object.h>
  82 #include <vm/vm_pager.h>
  83
  84 #include <machine/bootinfo.h>
  85 #include <machine/cpu.h>
  86 #include <machine/efi.h>
  87 #include <machine/elf.h>
  88 #include <machine/fpu.h>
  89 #include <machine/intr.h>
  90 #include <machine/kdb.h>
  91 #include <machine/mca.h>
  92 #include <machine/md_var.h>
  93 #include <machine/pal.h>
  94 #include <machine/pcb.h>
  95 #include <machine/reg.h>
  96 #include <machine/sal.h>
  97 #include <machine/sigframe.h>
  98 #ifdef SMP
  99 #include <machine/smp.h>
 100 #endif
 101 #include <machine/unwind.h>
 102 #include <machine/vmparam.h>
 103
 104 /*
 105  * For atomicity reasons, we demand that pc_curthread is the first
 106  * field in the struct pcpu. It allows us to read the pointer with
 107  * a single atomic instruction:
 108  *      ld8 %curthread = [r13]
 109  * Otherwise we would first have to calculate the load address and
 110  * store the result in a temporary register and that for the load:
 111  *      add %temp = %offsetof(struct pcpu), r13
 112  *      ld8 %curthread = [%temp]
 113  * A context switch inbetween the add and the ld8 could have the
 114  * thread migrate to a different core. In that case,  %curthread
 115  * would be the thread running on the original core and not actually
 116  * the current thread.
 117  */
 118 CTASSERT(offsetof(struct pcpu, pc_curthread) == 0);
 119
 120 static SYSCTL_NODE(_hw, OID_AUTO, freq, CTLFLAG_RD, 0, "");
 121 static SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, "");
 122
 123 static u_int bus_freq;
 124 SYSCTL_UINT(_hw_freq, OID_AUTO, bus, CTLFLAG_RD, &bus_freq, 0,
 125     "Bus clock frequency");
 126
 127 static u_int cpu_freq;
 128 SYSCTL_UINT(_hw_freq, OID_AUTO, cpu, CTLFLAG_RD, &cpu_freq, 0,
 129     "CPU clock frequency");
 130
 131 static u_int itc_freq;
 132 SYSCTL_UINT(_hw_freq, OID_AUTO, itc, CTLFLAG_RD, &itc_freq, 0,
 133     "ITC frequency");
 134
 135 int cold = 1;
 136 int unmapped_buf_allowed = 0;
 137
 138 struct bootinfo *bootinfo;
 139
 140 struct pcpu pcpu0;
 141
 142 extern u_int64_t kernel_text[], _end[];
 143
 144 extern u_int64_t ia64_gateway_page[];
 145 extern u_int64_t break_sigtramp[];
 146 extern u_int64_t epc_sigtramp[];
 147
 148 struct fpswa_iface *fpswa_iface;
 149
 150 vm_size_t ia64_pal_size;
 151 vm_paddr_t ia64_pal_base;
 152 vm_offset_t ia64_port_base;
 153
 154 u_int64_t ia64_lapic_addr = PAL_PIB_DEFAULT_ADDR;
 155
 156 struct ia64_pib *ia64_pib;
 157
 158 static int ia64_sync_icache_needed;
 159
 160 char machine[] = MACHINE;
 161 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
 162
 163 static char cpu_model[64];
 164 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
 165     "The CPU model name");
 166
 167 static char cpu_family[64];
 168 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
 169     "The CPU family name");
 170
 171 #ifdef DDB
 172 extern vm_offset_t ksym_start, ksym_end;
 173 #endif
 174
 175 struct msgbuf *msgbufp = NULL;
 176
 177 /* Other subsystems (e.g., ACPI) can hook this later. */
 178 void (*cpu_idle_hook)(sbintime_t) = NULL;
 179
 180 struct kva_md_info kmi;
 181
 182 static void
 183 identifycpu(void)
 184 {
 185         char vendor[17];
 186         char *family_name, *model_name;
 187         u_int64_t features, tmp;
 188         int number, revision, model, family, archrev;
 189
 190         /*
 191          * Assumes little-endian.
 192          */
 193         *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
 194         *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
 195         vendor[16] = '\0';
 196
 197         tmp = ia64_get_cpuid(3);
 198         number = (tmp >> 0) & 0xff;
 199         revision = (tmp >> 8) & 0xff;
 200         model = (tmp >> 16) & 0xff;
 201         family = (tmp >> 24) & 0xff;
 202         archrev = (tmp >> 32) & 0xff;
 203
 204         family_name = model_name = "unknown";
 205         switch (family) {
 206         case 0x07:
 207                 family_name = "Itanium";
 208                 model_name = "Merced";
 209                 break;
 210         case 0x1f:
 211                 family_name = "Itanium 2";
 212                 switch (model) {
 213                 case 0x00:
 214                         model_name = "McKinley";
 215                         break;
 216                 case 0x01:
 217                         /*
 218                          * Deerfield is a low-voltage variant based on the
 219                          * Madison core. We need circumstantial evidence
 220                          * (i.e. the clock frequency) to identify those.
 221                          * Allow for roughly 1% error margin.
 222                          */
 223                         if (cpu_freq > 990 && cpu_freq < 1010)
 224                                 model_name = "Deerfield";
 225                         else
 226                                 model_name = "Madison";
 227                         break;
 228                 case 0x02:
 229                         model_name = "Madison II";
 230                         break;
 231                 }
 232                 break;
 233         case 0x20:
 234                 ia64_sync_icache_needed = 1;
 235
 236                 family_name = "Itanium 2";
 237                 switch (model) {
 238                 case 0x00:
 239                         model_name = "Montecito";
 240                         break;
 241                 case 0x01:
 242                         model_name = "Montvale";
 243                         break;
 244                 }
 245                 break;
 246         }
 247         snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
 248         snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
 249
 250         features = ia64_get_cpuid(4);
 251
 252         printf("CPU: %s (", model_name);
 253         if (cpu_freq)
 254                 printf("%u MHz ", cpu_freq);
 255         printf("%s)\n", family_name);
 256         printf("  Origin = \"%s\"  Revision = %d\n", vendor, revision);
 257         printf("  Features = 0x%b\n", (u_int32_t) features,
 258             "\020"
 259             "\001LB"    /* long branch (brl) instruction. */
 260             "\002SD"    /* Spontaneous deferral. */
 261             "\003AO"    /* 16-byte atomic operations (ld, st, cmpxchg). */ );
 262 }
 263
 264 static void
 265 cpu_startup(void *dummy)
 266 {
 267         char nodename[16];
 268         struct pcpu *pc;
 269         struct pcpu_stats *pcs;
 270
 271         /*
 272          * Good {morning,afternoon,evening,night}.
 273          */
 274         identifycpu();
 275
 276 #ifdef PERFMON
 277         perfmon_init();
 278 #endif
 279         printf("real memory  = %ld (%ld MB)\n", ptoa(realmem),
 280             ptoa(realmem) / 1048576);
 281
 282         vm_ksubmap_init(&kmi);
 283
 284         printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count),
 285             ptoa(cnt.v_free_count) / 1048576);
 286
 287         if (fpswa_iface == NULL)
 288                 printf("Warning: no FPSWA package supplied\n");
 289         else
 290                 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
 291                     (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa);
 292
 293         /*
 294          * Set up buffers, so they can be used to read disk labels.
 295          */
 296         bufinit();
 297         vm_pager_bufferinit();
 298
 299         /*
 300          * Traverse the MADT to discover IOSAPIC and Local SAPIC
 301          * information.
 302          */
 303         ia64_probe_sapics();
 304         ia64_pib = pmap_mapdev(ia64_lapic_addr, sizeof(*ia64_pib));
 305
 306         ia64_mca_init();
 307
 308         /*
 309          * Create sysctl tree for per-CPU information.
 310          */
 311         STAILQ_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                     "nhardclocks", CTLFLAG_RD, &pcs->pcs_nhardclocks,
 340                     "Number of IPI_HARDCLOCK interrupts");
 341
 342                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 343                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 344                     "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps,
 345                     "Number of IPI_HIGH_FP interrupts");
 346
 347                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 348                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 349                     "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints,
 350                     "Number of hardware (device) interrupts");
 351
 352                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 353                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 354                     "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts,
 355                     "Number of IPI_PREEMPT interrupts");
 356
 357                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 358                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 359                     "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs,
 360                     "Number of IPI_RENDEZVOUS interrupts");
 361
 362                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 363                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 364                     "nstops", CTLFLAG_RD, &pcs->pcs_nstops,
 365                     "Number of IPI_STOP interrupts");
 366
 367                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 368                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 369                     "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays,
 370                     "Number of stray interrupts");
 371         }
 372 }
 373 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
 374
 375 void
 376 cpu_flush_dcache(void *ptr, size_t len)
 377 {
 378         vm_offset_t lim, va;
 379
 380         va = (uintptr_t)ptr & ~31;
 381         lim = (uintptr_t)ptr + len;
 382         while (va < lim) {
 383                 ia64_fc(va);
 384                 va += 32;
 385         }
 386
 387         ia64_srlz_d();
 388 }
 389
 390 /* Get current clock frequency for the given cpu id. */
 391 int
 392 cpu_est_clockrate(int cpu_id, uint64_t *rate)
 393 {
 394
 395         if (pcpu_find(cpu_id) == NULL || rate == NULL)
 396                 return (EINVAL);
 397         *rate = (u_long)cpu_freq * 1000000ul;
 398         return (0);
 399 }
 400
 401 void
 402 cpu_halt()
 403 {
 404
 405         efi_reset_system();
 406 }
 407
 408 void
 409 cpu_idle(int busy)
 410 {
 411         register_t ie;
 412         sbintime_t sbt = -1;
 413
 414         if (!busy) {
 415                 critical_enter();
 416                 sbt = cpu_idleclock();
 417         }
 418
 419         ie = intr_disable();
 420         KASSERT(ie != 0, ("%s called with interrupts disabled\n", __func__));
 421
 422         if (sched_runnable())
 423                 ia64_enable_intr();
 424         else if (cpu_idle_hook != NULL) {
 425                 (*cpu_idle_hook)(sbt);
 426                 /* The hook must enable interrupts! */
 427         } else {
 428                 ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
 429                 ia64_enable_intr();
 430         }
 431
 432         if (!busy) {
 433                 cpu_activeclock();
 434                 critical_exit();
 435         }
 436 }
 437
 438 int
 439 cpu_idle_wakeup(int cpu)
 440 {
 441
 442         return (0);
 443 }
 444
 445 void
 446 cpu_reset()
 447 {
 448
 449         efi_reset_system();
 450 }
 451
 452 void
 453 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
 454 {
 455         struct pcb *oldpcb, *newpcb;
 456
 457         oldpcb = old->td_pcb;
 458 #ifdef COMPAT_FREEBSD32
 459         ia32_savectx(oldpcb);
 460 #endif
 461         if (PCPU_GET(fpcurthread) == old)
 462                 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
 463         if (!savectx(oldpcb)) {
 464                 newpcb = new->td_pcb;
 465                 oldpcb->pcb_current_pmap =
 466                     pmap_switch(newpcb->pcb_current_pmap);
 467
 468                 atomic_store_rel_ptr(&old->td_lock, mtx);
 469
 470 #if defined(SCHED_ULE) && defined(SMP)
 471                 while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
 472                         cpu_spinwait();
 473 #endif
 474
 475                 PCPU_SET(curthread, new);
 476
 477 #ifdef COMPAT_FREEBSD32
 478                 ia32_restorectx(newpcb);
 479 #endif
 480
 481                 if (PCPU_GET(fpcurthread) == new)
 482                         new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
 483                 restorectx(newpcb);
 484                 /* We should not get here. */
 485                 panic("cpu_switch: restorectx() returned");
 486                 /* NOTREACHED */
 487         }
 488 }
 489
 490 void
 491 cpu_throw(struct thread *old __unused, struct thread *new)
 492 {
 493         struct pcb *newpcb;
 494
 495         newpcb = new->td_pcb;
 496         (void)pmap_switch(newpcb->pcb_current_pmap);
 497
 498 #if defined(SCHED_ULE) && defined(SMP)
 499         while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
 500                 cpu_spinwait();
 501 #endif
 502
 503         PCPU_SET(curthread, new);
 504
 505 #ifdef COMPAT_FREEBSD32
 506         ia32_restorectx(newpcb);
 507 #endif
 508
 509         restorectx(newpcb);
 510         /* We should not get here. */
 511         panic("cpu_throw: restorectx() returned");
 512         /* NOTREACHED */
 513 }
 514
 515 void
 516 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
 517 {
 518
 519         /*
 520          * Set pc_acpi_id to "uninitialized".
 521          * See sys/dev/acpica/acpi_cpu.c
 522          */
 523         pcpu->pc_acpi_id = 0xffffffff;
 524 }
 525
 526 void
 527 cpu_pcpu_setup(struct pcpu *pc, u_int acpi_id, u_int sapic_id)
 528 {
 529
 530         pc->pc_acpi_id = acpi_id;
 531         pc->pc_md.lid = IA64_LID_SET_SAPIC_ID(sapic_id);
 532 }
 533
 534 void
 535 spinlock_enter(void)
 536 {
 537         struct thread *td;
 538         int intr;
 539
 540         td = curthread;
 541         if (td->td_md.md_spinlock_count == 0) {
 542                 intr = intr_disable();
 543                 td->td_md.md_spinlock_count = 1;
 544                 td->td_md.md_saved_intr = intr;
 545         } else
 546                 td->td_md.md_spinlock_count++;
 547         critical_enter();
 548 }
 549
 550 void
 551 spinlock_exit(void)
 552 {
 553         struct thread *td;
 554         int intr;
 555
 556         td = curthread;
 557         critical_exit();
 558         intr = td->td_md.md_saved_intr;
 559         td->td_md.md_spinlock_count--;
 560         if (td->td_md.md_spinlock_count == 0)
 561                 intr_restore(intr);
 562 }
 563
 564 void
 565 kdb_cpu_trap(int vector, int code __unused)
 566 {
 567
 568 #ifdef XTRACE
 569         ia64_xtrace_stop();
 570 #endif
 571         __asm __volatile("flushrs;;");
 572
 573         /* Restart after the break instruction. */
 574         if (vector == IA64_VEC_BREAK &&
 575             kdb_frame->tf_special.ifa == IA64_FIXED_BREAK)
 576                 kdb_frame->tf_special.psr += IA64_PSR_RI_1;
 577 }
 578
 579 void
 580 map_vhpt(uintptr_t vhpt)
 581 {
 582         pt_entry_t pte;
 583         uint64_t psr;
 584
 585         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 586             PTE_PL_KERN | PTE_AR_RW;
 587         pte |= vhpt & PTE_PPN_MASK;
 588
 589         __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
 590             "r"(pmap_vhpt_log2size << 2));
 591
 592         __asm __volatile("mov   %0=psr" : "=r"(psr));
 593         __asm __volatile("rsm   psr.ic|psr.i");
 594         ia64_srlz_i();
 595         ia64_set_ifa(vhpt);
 596         ia64_set_itir(pmap_vhpt_log2size << 2);
 597         ia64_srlz_d();
 598         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
 599         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 600         ia64_srlz_i();
 601 }
 602
 603 void
 604 map_pal_code(void)
 605 {
 606         pt_entry_t pte;
 607         vm_offset_t va;
 608         vm_size_t sz;
 609         uint64_t psr;
 610         u_int shft;
 611
 612         if (ia64_pal_size == 0)
 613                 return;
 614
 615         va = IA64_PHYS_TO_RR7(ia64_pal_base);
 616
 617         sz = ia64_pal_size;
 618         shft = 0;
 619         while (sz > 1) {
 620                 shft++;
 621                 sz >>= 1;
 622         }
 623
 624         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 625             PTE_PL_KERN | PTE_AR_RWX;
 626         pte |= ia64_pal_base & PTE_PPN_MASK;
 627
 628         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" :: "r"(va), "r"(shft<<2));
 629
 630         __asm __volatile("mov   %0=psr" : "=r"(psr));
 631         __asm __volatile("rsm   psr.ic|psr.i");
 632         ia64_srlz_i();
 633         ia64_set_ifa(va);
 634         ia64_set_itir(shft << 2);
 635         ia64_srlz_d();
 636         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(4), "r"(pte));
 637         ia64_srlz_d();
 638         __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
 639         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 640         ia64_srlz_i();
 641 }
 642
 643 void
 644 map_gateway_page(void)
 645 {
 646         pt_entry_t pte;
 647         uint64_t psr;
 648
 649         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 650             PTE_PL_KERN | PTE_AR_X_RX;
 651         pte |= ia64_tpa((uint64_t)ia64_gateway_page) & PTE_PPN_MASK;
 652
 653         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
 654             "r"(VM_MAXUSER_ADDRESS), "r"(PAGE_SHIFT << 2));
 655
 656         __asm __volatile("mov   %0=psr" : "=r"(psr));
 657         __asm __volatile("rsm   psr.ic|psr.i");
 658         ia64_srlz_i();
 659         ia64_set_ifa(VM_MAXUSER_ADDRESS);
 660         ia64_set_itir(PAGE_SHIFT << 2);
 661         ia64_srlz_d();
 662         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(5), "r"(pte));
 663         ia64_srlz_d();
 664         __asm __volatile("itr.i itr[%0]=%1" :: "r"(2), "r"(pte));
 665         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 666         ia64_srlz_i();
 667
 668         /* Expose the mapping to userland in ar.k5 */
 669         ia64_set_k5(VM_MAXUSER_ADDRESS);
 670 }
 671
 672 static u_int
 673 freq_ratio(u_long base, u_long ratio)
 674 {
 675         u_long f;
 676
 677         f = (base * (ratio >> 32)) / (ratio & 0xfffffffful);
 678         return ((f + 500000) / 1000000);
 679 }
 680
 681 static void
 682 calculate_frequencies(void)
 683 {
 684         struct ia64_sal_result sal;
 685         struct ia64_pal_result pal;
 686         register_t ie;
 687
 688         ie = intr_disable();
 689         sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
 690         pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
 691         intr_restore(ie);
 692
 693         if (sal.sal_status == 0 && pal.pal_status == 0) {
 694                 if (bootverbose) {
 695                         printf("Platform clock frequency %ld Hz\n",
 696                                sal.sal_result[0]);
 697                         printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
 698                                "ITC ratio %ld/%ld\n",
 699                                pal.pal_result[0] >> 32,
 700                                pal.pal_result[0] & ((1L << 32) - 1),
 701                                pal.pal_result[1] >> 32,
 702                                pal.pal_result[1] & ((1L << 32) - 1),
 703                                pal.pal_result[2] >> 32,
 704                                pal.pal_result[2] & ((1L << 32) - 1));
 705                 }
 706                 cpu_freq = freq_ratio(sal.sal_result[0], pal.pal_result[0]);
 707                 bus_freq = freq_ratio(sal.sal_result[0], pal.pal_result[1]);
 708                 itc_freq = freq_ratio(sal.sal_result[0], pal.pal_result[2]);
 709         }
 710 }
 711
 712 struct ia64_init_return
 713 ia64_init(void)
 714 {
 715         struct ia64_init_return ret;
 716         struct efi_md *md;
 717         pt_entry_t *pbvm_pgtbl_ent, *pbvm_pgtbl_lim;
 718         char *p;
 719         vm_size_t mdlen;
 720         int metadata_missing;
 721
 722         /*
 723          * NO OUTPUT ALLOWED UNTIL FURTHER NOTICE.
 724          */
 725
 726         ia64_set_fpsr(IA64_FPSR_DEFAULT);
 727
 728         /*
 729          * Region 6 is direct mapped UC and region 7 is direct mapped
 730          * WC. The details of this is controlled by the Alt {I,D}TLB
 731          * handlers. Here we just make sure that they have the largest
 732          * possible page size to minimise TLB usage.
 733          */
 734         ia64_set_rr(IA64_RR_BASE(6), (6 << 8) | (LOG2_ID_PAGE_SIZE << 2));
 735         ia64_set_rr(IA64_RR_BASE(7), (7 << 8) | (LOG2_ID_PAGE_SIZE << 2));
 736         ia64_srlz_d();
 737
 738         /* Initialize/setup physical memory datastructures */
 739         ia64_physmem_init();
 740
 741         /*
 742          * Process the memory map. This gives us the PAL locations,
 743          * the I/O port base address, the available memory regions
 744          * for initializing the physical memory map.
 745          */
 746         for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
 747                 mdlen = md->md_pages * EFI_PAGE_SIZE;
 748                 switch (md->md_type) {
 749                 case EFI_MD_TYPE_IOPORT:
 750                         ia64_port_base = pmap_mapdev_priv(md->md_phys,
 751                             mdlen, VM_MEMATTR_UNCACHEABLE);
 752                         break;
 753                 case EFI_MD_TYPE_PALCODE:
 754                         ia64_pal_base = md->md_phys;
 755                         ia64_pal_size = mdlen;
 756                         /*FALLTHROUGH*/
 757                 case EFI_MD_TYPE_BAD:
 758                 case EFI_MD_TYPE_FIRMWARE:
 759                 case EFI_MD_TYPE_RECLAIM:
 760                 case EFI_MD_TYPE_RT_CODE:
 761                 case EFI_MD_TYPE_RT_DATA:
 762                         /* Don't use these memory regions. */
 763                         ia64_physmem_track(md->md_phys, mdlen);
 764                         break;
 765                 case EFI_MD_TYPE_BS_CODE:
 766                 case EFI_MD_TYPE_BS_DATA:
 767                 case EFI_MD_TYPE_CODE:
 768                 case EFI_MD_TYPE_DATA:
 769                 case EFI_MD_TYPE_FREE:
 770                         /* These are ok to use. */
 771                         ia64_physmem_add(md->md_phys, mdlen);
 772                         break;
 773                 }
 774         }
 775
 776         /*
 777          * Remove the PBVM and its page table from phys_avail. The loader
 778          * passes the physical address of the page table to us. The virtual
 779          * address of the page table is fixed.
 780          * Track and the PBVM limit for later use.
 781          */
 782         ia64_physmem_delete(bootinfo->bi_pbvm_pgtbl, bootinfo->bi_pbvm_pgtblsz);
 783         pbvm_pgtbl_ent = (void *)IA64_PBVM_PGTBL;
 784         pbvm_pgtbl_lim = (void *)(IA64_PBVM_PGTBL + bootinfo->bi_pbvm_pgtblsz);
 785         while (pbvm_pgtbl_ent < pbvm_pgtbl_lim) {
 786                 if ((*pbvm_pgtbl_ent & PTE_PRESENT) == 0)
 787                         break;
 788                 ia64_physmem_delete(*pbvm_pgtbl_ent & PTE_PPN_MASK,
 789                     IA64_PBVM_PAGE_SIZE);
 790                 pbvm_pgtbl_ent++;
 791         }
 792
 793         /* Finalize physical memory datastructures */
 794         ia64_physmem_fini();
 795
 796         metadata_missing = 0;
 797         if (bootinfo->bi_modulep)
 798                 preload_metadata = (caddr_t)bootinfo->bi_modulep;
 799         else
 800                 metadata_missing = 1;
 801
 802         if (envmode == 0 && bootinfo->bi_envp)
 803                 kern_envp = (caddr_t)bootinfo->bi_envp;
 804         else
 805                 kern_envp = static_env;
 806
 807         /*
 808          * Look at arguments passed to us and compute boothowto.
 809          */
 810         boothowto = bootinfo->bi_boothowto;
 811
 812         if (boothowto & RB_VERBOSE)
 813                 bootverbose = 1;
 814
 815         /*
 816          * Wire things up so we can call the firmware.
 817          */
 818         map_pal_code();
 819         efi_boot_minimal(bootinfo->bi_systab);
 820         ia64_xiv_init();
 821         ia64_sal_init();
 822         calculate_frequencies();
 823
 824         set_cputicker(ia64_get_itc, (u_long)itc_freq * 1000000, 0);
 825
 826         /*
 827          * Setup the PCPU data for the bootstrap processor. It is needed
 828          * by printf(). Also, since printf() has critical sections, we
 829          * need to initialize at least pc_curthread.
 830          */
 831         pcpup = &pcpu0;
 832         ia64_set_k4((u_int64_t)pcpup);
 833         pcpu_init(pcpup, 0, sizeof(pcpu0));
 834         dpcpu_init(ia64_physmem_alloc(DPCPU_SIZE, PAGE_SIZE), 0);
 835         cpu_pcpu_setup(pcpup, ~0U, ia64_get_lid());
 836         PCPU_SET(curthread, &thread0);
 837
 838         /*
 839          * Initialize the console before we print anything out.
 840          */
 841         cninit();
 842
 843         /* OUTPUT NOW ALLOWED */
 844
 845         if (metadata_missing)
 846                 printf("WARNING: loader(8) metadata is missing!\n");
 847
 848         /* Get FPSWA interface */
 849         fpswa_iface = (bootinfo->bi_fpswa == 0) ? NULL :
 850             (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo->bi_fpswa);
 851
 852         /* Init basic tunables, including hz */
 853         init_param1();
 854
 855         p = getenv("kernelname");
 856         if (p != NULL) {
 857                 strlcpy(kernelname, p, sizeof(kernelname));
 858                 freeenv(p);
 859         }
 860
 861         init_param2(physmem);
 862
 863         /*
 864          * Initialize error message buffer (at end of core).
 865          */
 866         msgbufp = ia64_physmem_alloc(msgbufsize, PAGE_SIZE);
 867         msgbufinit(msgbufp, msgbufsize);
 868
 869         proc_linkup0(&proc0, &thread0);
 870         /*
 871          * Init mapping for kernel stack for proc 0
 872          */
 873         p = ia64_physmem_alloc(KSTACK_PAGES * PAGE_SIZE, PAGE_SIZE);
 874         thread0.td_kstack = (uintptr_t)p;
 875         thread0.td_kstack_pages = KSTACK_PAGES;
 876
 877         mutex_init();
 878
 879         /*
 880          * Initialize the rest of proc 0's PCB.
 881          *
 882          * Set the kernel sp, reserving space for an (empty) trapframe,
 883          * and make proc0's trapframe pointer point to it for sanity.
 884          * Initialise proc0's backing store to start after u area.
 885          */
 886         cpu_thread_alloc(&thread0);
 887         thread0.td_frame->tf_flags = FRAME_SYSCALL;
 888         thread0.td_pcb->pcb_special.sp =
 889             (u_int64_t)thread0.td_frame - 16;
 890         thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
 891
 892         /*
 893          * Initialize the virtual memory system.
 894          */
 895         pmap_bootstrap();
 896
 897 #ifdef XTRACE
 898         ia64_xtrace_init_bsp();
 899 #endif
 900
 901         /*
 902          * Initialize debuggers, and break into them if appropriate.
 903          */
 904 #ifdef DDB
 905         ksym_start = bootinfo->bi_symtab;
 906         ksym_end = bootinfo->bi_esymtab;
 907 #endif
 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 uint64_t
 926 ia64_get_hcdp(void)
 927 {
 928
 929         return (bootinfo->bi_hcdp);
 930 }
 931
 932 void
 933 bzero(void *buf, size_t len)
 934 {
 935         caddr_t p = buf;
 936
 937         while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
 938                 *p++ = 0;
 939                 len--;
 940         }
 941         while (len >= sizeof(u_long) * 8) {
 942                 *(u_long*) p = 0;
 943                 *((u_long*) p + 1) = 0;
 944                 *((u_long*) p + 2) = 0;
 945                 *((u_long*) p + 3) = 0;
 946                 len -= sizeof(u_long) * 8;
 947                 *((u_long*) p + 4) = 0;
 948                 *((u_long*) p + 5) = 0;
 949                 *((u_long*) p + 6) = 0;
 950                 *((u_long*) p + 7) = 0;
 951                 p += sizeof(u_long) * 8;
 952         }
 953         while (len >= sizeof(u_long)) {
 954                 *(u_long*) p = 0;
 955                 len -= sizeof(u_long);
 956                 p += sizeof(u_long);
 957         }
 958         while (len) {
 959                 *p++ = 0;
 960                 len--;
 961         }
 962 }
 963
 964 u_int
 965 ia64_itc_freq(void)
 966 {
 967
 968         return (itc_freq);
 969 }
 970
 971 void
 972 DELAY(int n)
 973 {
 974         u_int64_t start, end, now;
 975
 976         sched_pin();
 977
 978         start = ia64_get_itc();
 979         end = start + itc_freq * n;
 980         /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
 981         do {
 982                 now = ia64_get_itc();
 983         } while (now < end || (now > start && end < start));
 984
 985         sched_unpin();
 986 }
 987
 988 /*
 989  * Send an interrupt (signal) to a process.
 990  */
 991 void
 992 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
 993 {
 994         struct proc *p;
 995         struct thread *td;
 996         struct trapframe *tf;
 997         struct sigacts *psp;
 998         struct sigframe sf, *sfp;
 999         u_int64_t sbs, sp;
1000         int oonstack;
1001         int sig;
1002         u_long code;
1003
1004         td = curthread;
1005         p = td->td_proc;
1006         PROC_LOCK_ASSERT(p, MA_OWNED);
1007         sig = ksi->ksi_signo;
1008         code = ksi->ksi_code;
1009         psp = p->p_sigacts;
1010         mtx_assert(&psp->ps_mtx, MA_OWNED);
1011         tf = td->td_frame;
1012         sp = tf->tf_special.sp;
1013         oonstack = sigonstack(sp);
1014         sbs = 0;
1015
1016         /* save user context */
1017         bzero(&sf, sizeof(struct sigframe));
1018         sf.sf_uc.uc_sigmask = *mask;
1019         sf.sf_uc.uc_stack = td->td_sigstk;
1020         sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1021             ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1022
1023         /*
1024          * Allocate and validate space for the signal handler
1025          * context. Note that if the stack is in P0 space, the
1026          * call to grow() is a nop, and the useracc() check
1027          * will fail if the process has not already allocated
1028          * the space with a `brk'.
1029          */
1030         if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1031             SIGISMEMBER(psp->ps_sigonstack, sig)) {
1032                 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1033                 sbs = (sbs + 15) & ~15;
1034                 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1035 #if defined(COMPAT_43)
1036                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1037 #endif
1038         } else
1039                 sfp = (struct sigframe *)sp;
1040         sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1041
1042         /* Fill in the siginfo structure for POSIX handlers. */
1043         if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1044                 sf.sf_si = ksi->ksi_info;
1045                 sf.sf_si.si_signo = sig;
1046                 /*
1047                  * XXX this shouldn't be here after code in trap.c
1048                  * is fixed
1049                  */
1050                 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1051                 code = (u_int64_t)&sfp->sf_si;
1052         }
1053
1054         mtx_unlock(&psp->ps_mtx);
1055         PROC_UNLOCK(p);
1056
1057         get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1058
1059         /* Copy the frame out to userland. */
1060         if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1061                 /*
1062                  * Process has trashed its stack; give it an illegal
1063                  * instruction to halt it in its tracks.
1064                  */
1065                 PROC_LOCK(p);
1066                 sigexit(td, SIGILL);
1067                 return;
1068         }
1069
1070         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1071                 tf->tf_special.psr &= ~IA64_PSR_RI;
1072                 tf->tf_special.iip = ia64_get_k5() +
1073                     ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1074         } else
1075                 tf->tf_special.iip = ia64_get_k5() +
1076                     ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1077
1078         /*
1079          * Setup the trapframe to return to the signal trampoline. We pass
1080          * information to the trampoline in the following registers:
1081          *
1082          *      gp      new backing store or NULL
1083          *      r8      signal number
1084          *      r9      signal code or siginfo pointer
1085          *      r10     signal handler (function descriptor)
1086          */
1087         tf->tf_special.sp = (u_int64_t)sfp - 16;
1088         tf->tf_special.gp = sbs;
1089         tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1090         tf->tf_special.ndirty = 0;
1091         tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1092         tf->tf_scratch.gr8 = sig;
1093         tf->tf_scratch.gr9 = code;
1094         tf->tf_scratch.gr10 = (u_int64_t)catcher;
1095
1096         PROC_LOCK(p);
1097         mtx_lock(&psp->ps_mtx);
1098 }
1099
1100 /*
1101  * System call to cleanup state after a signal
1102  * has been taken.  Reset signal mask and
1103  * stack state from context left by sendsig (above).
1104  * Return to previous pc and psl as specified by
1105  * context left by sendsig. Check carefully to
1106  * make sure that the user has not modified the
1107  * state to gain improper privileges.
1108  *
1109  * MPSAFE
1110  */
1111 int
1112 sys_sigreturn(struct thread *td,
1113         struct sigreturn_args /* {
1114                 ucontext_t *sigcntxp;
1115         } */ *uap)
1116 {
1117         ucontext_t uc;
1118         struct trapframe *tf;
1119         struct pcb *pcb;
1120
1121         tf = td->td_frame;
1122         pcb = td->td_pcb;
1123
1124         /*
1125          * Fetch the entire context structure at once for speed.
1126          * We don't use a normal argument to simplify RSE handling.
1127          */
1128         if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1129                 return (EFAULT);
1130
1131         set_mcontext(td, &uc.uc_mcontext);
1132
1133 #if defined(COMPAT_43)
1134         if (sigonstack(tf->tf_special.sp))
1135                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1136         else
1137                 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1138 #endif
1139         kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
1140
1141         return (EJUSTRETURN);
1142 }
1143
1144 #ifdef COMPAT_FREEBSD4
1145 int
1146 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1147 {
1148
1149         return sys_sigreturn(td, (struct sigreturn_args *)uap);
1150 }
1151 #endif
1152
1153 /*
1154  * Construct a PCB from a trapframe. This is called from kdb_trap() where
1155  * we want to start a backtrace from the function that caused us to enter
1156  * the debugger. We have the context in the trapframe, but base the trace
1157  * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1158  * enough for a backtrace.
1159  */
1160 void
1161 makectx(struct trapframe *tf, struct pcb *pcb)
1162 {
1163
1164         pcb->pcb_special = tf->tf_special;
1165         pcb->pcb_special.__spare = ~0UL;        /* XXX see unwind.c */
1166         save_callee_saved(&pcb->pcb_preserved);
1167         save_callee_saved_fp(&pcb->pcb_preserved_fp);
1168 }
1169
1170 int
1171 ia64_flush_dirty(struct thread *td, struct _special *r)
1172 {
1173         struct iovec iov;
1174         struct uio uio;
1175         uint64_t bspst, kstk, rnat;
1176         int error, locked;
1177
1178         if (r->ndirty == 0)
1179                 return (0);
1180
1181         kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1182         if (td == curthread) {
1183                 __asm __volatile("mov   ar.rsc=0;;");
1184                 __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1185                 /* Make sure we have all the user registers written out. */
1186                 if (bspst - kstk < r->ndirty) {
1187                         __asm __volatile("flushrs;;");
1188                         __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1189                 }
1190                 __asm __volatile("mov   %0=ar.rnat;;" : "=r"(rnat));
1191                 __asm __volatile("mov   ar.rsc=3");
1192                 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1193                 kstk += r->ndirty;
1194                 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1195                     ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1196         } else {
1197                 locked = PROC_LOCKED(td->td_proc);
1198                 if (!locked)
1199                         PHOLD(td->td_proc);
1200                 iov.iov_base = (void*)(uintptr_t)kstk;
1201                 iov.iov_len = r->ndirty;
1202                 uio.uio_iov = &iov;
1203                 uio.uio_iovcnt = 1;
1204                 uio.uio_offset = r->bspstore;
1205                 uio.uio_resid = r->ndirty;
1206                 uio.uio_segflg = UIO_SYSSPACE;
1207                 uio.uio_rw = UIO_WRITE;
1208                 uio.uio_td = td;
1209                 error = proc_rwmem(td->td_proc, &uio);
1210                 /*
1211                  * XXX proc_rwmem() doesn't currently return ENOSPC,
1212                  * so I think it can bogusly return 0. Neither do
1213                  * we allow short writes.
1214                  */
1215                 if (uio.uio_resid != 0 && error == 0)
1216                         error = ENOSPC;
1217                 if (!locked)
1218                         PRELE(td->td_proc);
1219         }
1220
1221         r->bspstore += r->ndirty;
1222         r->ndirty = 0;
1223         return (error);
1224 }
1225
1226 int
1227 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1228 {
1229         struct trapframe *tf;
1230         int error;
1231
1232         tf = td->td_frame;
1233         bzero(mc, sizeof(*mc));
1234         mc->mc_special = tf->tf_special;
1235         error = ia64_flush_dirty(td, &mc->mc_special);
1236         if (tf->tf_flags & FRAME_SYSCALL) {
1237                 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1238                 mc->mc_scratch = tf->tf_scratch;
1239                 if (flags & GET_MC_CLEAR_RET) {
1240                         mc->mc_scratch.gr8 = 0;
1241                         mc->mc_scratch.gr9 = 0;
1242                         mc->mc_scratch.gr10 = 0;
1243                         mc->mc_scratch.gr11 = 0;
1244                 }
1245         } else {
1246                 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1247                 mc->mc_scratch = tf->tf_scratch;
1248                 mc->mc_scratch_fp = tf->tf_scratch_fp;
1249                 /*
1250                  * XXX If the thread never used the high FP registers, we
1251                  * probably shouldn't waste time saving them.
1252                  */
1253                 ia64_highfp_save(td);
1254                 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1255                 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1256         }
1257         save_callee_saved(&mc->mc_preserved);
1258         save_callee_saved_fp(&mc->mc_preserved_fp);
1259         return (error);
1260 }
1261
1262 int
1263 set_mcontext(struct thread *td, const mcontext_t *mc)
1264 {
1265         struct _special s;
1266         struct trapframe *tf;
1267         uint64_t psrmask;
1268
1269         tf = td->td_frame;
1270
1271         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1272             ("Whoa there! We have more than 8KB of dirty registers!"));
1273
1274         s = mc->mc_special;
1275         /*
1276          * Only copy the user mask and the restart instruction bit from
1277          * the new context.
1278          */
1279         psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1280             IA64_PSR_MFH | IA64_PSR_RI;
1281         s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1282         /* We don't have any dirty registers of the new context. */
1283         s.ndirty = 0;
1284         if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1285                 /*
1286                  * We can get an async context passed to us while we
1287                  * entered the kernel through a syscall: sigreturn(2)
1288                  * takes contexts that could previously be the result of
1289                  * a trap or interrupt.
1290                  * Hence, we cannot assert that the trapframe is not
1291                  * a syscall frame, but we can assert that it's at
1292                  * least an expected syscall.
1293                  */
1294                 if (tf->tf_flags & FRAME_SYSCALL) {
1295                         KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1296                         tf->tf_flags &= ~FRAME_SYSCALL;
1297                 }
1298                 tf->tf_scratch = mc->mc_scratch;
1299                 tf->tf_scratch_fp = mc->mc_scratch_fp;
1300                 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1301                         td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1302         } else {
1303                 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1304                 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1305                         s.cfm = tf->tf_special.cfm;
1306                         s.iip = tf->tf_special.iip;
1307                         tf->tf_scratch.gr15 = 0;        /* Clear syscall nr. */
1308                 } else
1309                         tf->tf_scratch = mc->mc_scratch;
1310         }
1311         tf->tf_special = s;
1312         restore_callee_saved(&mc->mc_preserved);
1313         restore_callee_saved_fp(&mc->mc_preserved_fp);
1314
1315         return (0);
1316 }
1317
1318 /*
1319  * Clear registers on exec.
1320  */
1321 void
1322 exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
1323 {
1324         struct trapframe *tf;
1325         uint64_t *ksttop, *kst;
1326
1327         tf = td->td_frame;
1328         ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1329             (tf->tf_special.bspstore & 0x1ffUL));
1330
1331         /*
1332          * We can ignore up to 8KB of dirty registers by masking off the
1333          * lower 13 bits in exception_restore() or epc_syscall(). This
1334          * should be enough for a couple of years, but if there are more
1335          * than 8KB of dirty registers, we lose track of the bottom of
1336          * the kernel stack. The solution is to copy the active part of
1337          * the kernel stack down 1 page (or 2, but not more than that)
1338          * so that we always have less than 8KB of dirty registers.
1339          */
1340         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1341             ("Whoa there! We have more than 8KB of dirty registers!"));
1342
1343         bzero(&tf->tf_special, sizeof(tf->tf_special));
1344         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {      /* break syscalls. */
1345                 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1346                 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1347                 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1348                 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1349                 /*
1350                  * Copy the arguments onto the kernel register stack so that
1351                  * they get loaded by the loadrs instruction. Skip over the
1352                  * NaT collection points.
1353                  */
1354                 kst = ksttop - 1;
1355                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1356                         *kst-- = 0;
1357                 *kst-- = 0;
1358                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1359                         *kst-- = 0;
1360                 *kst-- = imgp->ps_strings;
1361                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1362                         *kst-- = 0;
1363                 *kst = stack;
1364                 tf->tf_special.ndirty = (ksttop - kst) << 3;
1365         } else {                                /* epc syscalls (default). */
1366                 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1367                 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1368                 /*
1369                  * Write values for out0, out1 and out2 to the user's backing
1370                  * store and arrange for them to be restored into the user's
1371                  * initial register frame.
1372                  * Assumes that (bspstore & 0x1f8) < 0x1e0.
1373                  */
1374                 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1375                 suword((caddr_t)tf->tf_special.bspstore - 16, imgp->ps_strings);
1376                 suword((caddr_t)tf->tf_special.bspstore -  8, 0);
1377         }
1378
1379         tf->tf_special.iip = imgp->entry_addr;
1380         tf->tf_special.sp = (stack & ~15) - 16;
1381         tf->tf_special.rsc = 0xf;
1382         tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1383         tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1384             IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1385             IA64_PSR_CPL_USER;
1386 }
1387
1388 int
1389 ptrace_set_pc(struct thread *td, unsigned long addr)
1390 {
1391         uint64_t slot;
1392
1393         switch (addr & 0xFUL) {
1394         case 0:
1395                 slot = IA64_PSR_RI_0;
1396                 break;
1397         case 1:
1398                 /* XXX we need to deal with MLX bundles here */
1399                 slot = IA64_PSR_RI_1;
1400                 break;
1401         case 2:
1402                 slot = IA64_PSR_RI_2;
1403                 break;
1404         default:
1405                 return (EINVAL);
1406         }
1407
1408         td->td_frame->tf_special.iip = addr & ~0x0FULL;
1409         td->td_frame->tf_special.psr =
1410             (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1411         return (0);
1412 }
1413
1414 int
1415 ptrace_single_step(struct thread *td)
1416 {
1417         struct trapframe *tf;
1418
1419         /*
1420          * There's no way to set single stepping when we're leaving the
1421          * kernel through the EPC syscall path. The way we solve this is
1422          * by enabling the lower-privilege trap so that we re-enter the
1423          * kernel as soon as the privilege level changes. See trap.c for
1424          * how we proceed from there.
1425          */
1426         tf = td->td_frame;
1427         if (tf->tf_flags & FRAME_SYSCALL)
1428                 tf->tf_special.psr |= IA64_PSR_LP;
1429         else
1430                 tf->tf_special.psr |= IA64_PSR_SS;
1431         return (0);
1432 }
1433
1434 int
1435 ptrace_clear_single_step(struct thread *td)
1436 {
1437         struct trapframe *tf;
1438
1439         /*
1440          * Clear any and all status bits we may use to implement single
1441          * stepping.
1442          */
1443         tf = td->td_frame;
1444         tf->tf_special.psr &= ~IA64_PSR_SS;
1445         tf->tf_special.psr &= ~IA64_PSR_LP;
1446         tf->tf_special.psr &= ~IA64_PSR_TB;
1447         return (0);
1448 }
1449
1450 int
1451 fill_regs(struct thread *td, struct reg *regs)
1452 {
1453         struct trapframe *tf;
1454
1455         tf = td->td_frame;
1456         regs->r_special = tf->tf_special;
1457         regs->r_scratch = tf->tf_scratch;
1458         save_callee_saved(&regs->r_preserved);
1459         return (0);
1460 }
1461
1462 int
1463 set_regs(struct thread *td, struct reg *regs)
1464 {
1465         struct trapframe *tf;
1466         int error;
1467
1468         tf = td->td_frame;
1469         error = ia64_flush_dirty(td, &tf->tf_special);
1470         if (!error) {
1471                 tf->tf_special = regs->r_special;
1472                 tf->tf_special.bspstore += tf->tf_special.ndirty;
1473                 tf->tf_special.ndirty = 0;
1474                 tf->tf_scratch = regs->r_scratch;
1475                 restore_callee_saved(&regs->r_preserved);
1476         }
1477         return (error);
1478 }
1479
1480 int
1481 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1482 {
1483
1484         return (ENOSYS);
1485 }
1486
1487 int
1488 set_dbregs(struct thread *td, struct dbreg *dbregs)
1489 {
1490
1491         return (ENOSYS);
1492 }
1493
1494 int
1495 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1496 {
1497         struct trapframe *frame = td->td_frame;
1498         struct pcb *pcb = td->td_pcb;
1499
1500         /* Save the high FP registers. */
1501         ia64_highfp_save(td);
1502
1503         fpregs->fpr_scratch = frame->tf_scratch_fp;
1504         save_callee_saved_fp(&fpregs->fpr_preserved);
1505         fpregs->fpr_high = pcb->pcb_high_fp;
1506         return (0);
1507 }
1508
1509 int
1510 set_fpregs(struct thread *td, struct fpreg *fpregs)
1511 {
1512         struct trapframe *frame = td->td_frame;
1513         struct pcb *pcb = td->td_pcb;
1514
1515         /* Throw away the high FP registers (should be redundant). */
1516         ia64_highfp_drop(td);
1517
1518         frame->tf_scratch_fp = fpregs->fpr_scratch;
1519         restore_callee_saved_fp(&fpregs->fpr_preserved);
1520         pcb->pcb_high_fp = fpregs->fpr_high;
1521         return (0);
1522 }
1523
1524 void
1525 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1526 {
1527         vm_offset_t lim;
1528
1529         if (!ia64_sync_icache_needed)
1530                 return;
1531
1532         lim = va + sz;
1533         while (va < lim) {
1534                 ia64_fc_i(va);
1535                 va += 32;       /* XXX */
1536         }
1537
1538         ia64_sync_i();
1539         ia64_srlz_i();
1540 }