sys/ia64/ia64/machdep.c

   1 /*-
   2  * Copyright (c) 2003,2004 Marcel Moolenaar
   3  * Copyright (c) 2000,2001 Doug Rabson
   4  * All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  25  * SUCH DAMAGE.
  26  */
  27
  28 #include <sys/cdefs.h>
  29 __FBSDID("$FreeBSD$");
  30
  31 #include "opt_compat.h"
  32 #include "opt_ddb.h"
  33 #include "opt_kstack_pages.h"
  34 #include "opt_sched.h"
  35
  36 #include <sys/param.h>
  37 #include <sys/proc.h>
  38 #include <sys/systm.h>
  39 #include <sys/bio.h>
  40 #include <sys/buf.h>
  41 #include <sys/bus.h>
  42 #include <sys/cons.h>
  43 #include <sys/cpu.h>
  44 #include <sys/eventhandler.h>
  45 #include <sys/exec.h>
  46 #include <sys/imgact.h>
  47 #include <sys/kdb.h>
  48 #include <sys/kernel.h>
  49 #include <sys/linker.h>
  50 #include <sys/lock.h>
  51 #include <sys/malloc.h>
  52 #include <sys/mbuf.h>
  53 #include <sys/msgbuf.h>
  54 #include <sys/pcpu.h>
  55 #include <sys/ptrace.h>
  56 #include <sys/random.h>
  57 #include <sys/reboot.h>
  58 #include <sys/sched.h>
  59 #include <sys/signalvar.h>
  60 #include <sys/syscall.h>
  61 #include <sys/sysctl.h>
  62 #include <sys/sysproto.h>
  63 #include <sys/ucontext.h>
  64 #include <sys/uio.h>
  65 #include <sys/uuid.h>
  66 #include <sys/vmmeter.h>
  67 #include <sys/vnode.h>
  68
  69 #include <ddb/ddb.h>
  70
  71 #include <net/netisr.h>
  72
  73 #include <vm/vm.h>
  74 #include <vm/vm_extern.h>
  75 #include <vm/vm_kern.h>
  76 #include <vm/vm_page.h>
  77 #include <vm/vm_map.h>
  78 #include <vm/vm_object.h>
  79 #include <vm/vm_pager.h>
  80
  81 #include <machine/bootinfo.h>
  82 #include <machine/cpu.h>
  83 #include <machine/efi.h>
  84 #include <machine/elf.h>
  85 #include <machine/fpu.h>
  86 #include <machine/intr.h>
  87 #include <machine/mca.h>
  88 #include <machine/md_var.h>
  89 #include <machine/mutex.h>
  90 #include <machine/pal.h>
  91 #include <machine/pcb.h>
  92 #include <machine/reg.h>
  93 #include <machine/sal.h>
  94 #include <machine/sigframe.h>
  95 #ifdef SMP
  96 #include <machine/smp.h>
  97 #endif
  98 #include <machine/unwind.h>
  99 #include <machine/vmparam.h>
 100
 101 SYSCTL_NODE(_hw, OID_AUTO, freq, CTLFLAG_RD, 0, "");
 102 SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, "");
 103
 104 static u_int bus_freq;
 105 SYSCTL_UINT(_hw_freq, OID_AUTO, bus, CTLFLAG_RD, &bus_freq, 0,
 106     "Bus clock frequency");
 107
 108 static u_int cpu_freq;
 109 SYSCTL_UINT(_hw_freq, OID_AUTO, cpu, CTLFLAG_RD, &cpu_freq, 0,
 110     "CPU clock frequency");
 111
 112 static u_int itc_freq;
 113 SYSCTL_UINT(_hw_freq, OID_AUTO, itc, CTLFLAG_RD, &itc_freq, 0,
 114     "ITC frequency");
 115
 116 int cold = 1;
 117
 118 u_int64_t pa_bootinfo;
 119 struct bootinfo bootinfo;
 120
 121 struct pcpu pcpu0;
 122
 123 extern u_int64_t kernel_text[], _end[];
 124
 125 extern u_int64_t ia64_gateway_page[];
 126 extern u_int64_t break_sigtramp[];
 127 extern u_int64_t epc_sigtramp[];
 128
 129 struct fpswa_iface *fpswa_iface;
 130
 131 u_int64_t ia64_pal_base;
 132 u_int64_t ia64_port_base;
 133
 134 u_int64_t ia64_lapic_addr = PAL_PIB_DEFAULT_ADDR;
 135
 136 struct ia64_pib *ia64_pib;
 137
 138 static int ia64_sync_icache_needed;
 139
 140 char machine[] = MACHINE;
 141 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "");
 142
 143 static char cpu_model[64];
 144 SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0,
 145     "The CPU model name");
 146
 147 static char cpu_family[64];
 148 SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0,
 149     "The CPU family name");
 150
 151 #ifdef DDB
 152 extern vm_offset_t ksym_start, ksym_end;
 153 #endif
 154
 155
 156 struct msgbuf *msgbufp = NULL;
 157
 158 /* Other subsystems (e.g., ACPI) can hook this later. */
 159 void (*cpu_idle_hook)(void) = NULL;
 160
 161 long Maxmem = 0;
 162 long realmem = 0;
 163
 164 #define PHYSMAP_SIZE    (2 * VM_PHYSSEG_MAX)
 165
 166 vm_paddr_t phys_avail[PHYSMAP_SIZE + 2];
 167
 168 /* must be 2 less so 0 0 can signal end of chunks */
 169 #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2)
 170
 171 struct kva_md_info kmi;
 172
 173 #define Mhz     1000000L
 174 #define Ghz     (1000L*Mhz)
 175
 176 static void
 177 identifycpu(void)
 178 {
 179         char vendor[17];
 180         char *family_name, *model_name;
 181         u_int64_t features, tmp;
 182         int number, revision, model, family, archrev;
 183
 184         /*
 185          * Assumes little-endian.
 186          */
 187         *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0);
 188         *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1);
 189         vendor[16] = '\0';
 190
 191         tmp = ia64_get_cpuid(3);
 192         number = (tmp >> 0) & 0xff;
 193         revision = (tmp >> 8) & 0xff;
 194         model = (tmp >> 16) & 0xff;
 195         family = (tmp >> 24) & 0xff;
 196         archrev = (tmp >> 32) & 0xff;
 197
 198         family_name = model_name = "unknown";
 199         switch (family) {
 200         case 0x07:
 201                 family_name = "Itanium";
 202                 model_name = "Merced";
 203                 break;
 204         case 0x1f:
 205                 family_name = "Itanium 2";
 206                 switch (model) {
 207                 case 0x00:
 208                         model_name = "McKinley";
 209                         break;
 210                 case 0x01:
 211                         /*
 212                          * Deerfield is a low-voltage variant based on the
 213                          * Madison core. We need circumstantial evidence
 214                          * (i.e. the clock frequency) to identify those.
 215                          * Allow for roughly 1% error margin.
 216                          */
 217                         if (cpu_freq > 990 && cpu_freq < 1010)
 218                                 model_name = "Deerfield";
 219                         else
 220                                 model_name = "Madison";
 221                         break;
 222                 case 0x02:
 223                         model_name = "Madison II";
 224                         break;
 225                 }
 226                 break;
 227         case 0x20:
 228                 ia64_sync_icache_needed = 1;
 229
 230                 family_name = "Itanium 2";
 231                 switch (model) {
 232                 case 0x00:
 233                         model_name = "Montecito";
 234                         break;
 235                 }
 236                 break;
 237         }
 238         snprintf(cpu_family, sizeof(cpu_family), "%s", family_name);
 239         snprintf(cpu_model, sizeof(cpu_model), "%s", model_name);
 240
 241         features = ia64_get_cpuid(4);
 242
 243         printf("CPU: %s (", model_name);
 244         if (cpu_freq)
 245                 printf("%u Mhz ", cpu_freq);
 246         printf("%s)\n", family_name);
 247         printf("  Origin = \"%s\"  Revision = %d\n", vendor, revision);
 248         printf("  Features = 0x%b\n", (u_int32_t) features,
 249             "\020"
 250             "\001LB"    /* long branch (brl) instruction. */
 251             "\002SD"    /* Spontaneous deferral. */
 252             "\003AO"    /* 16-byte atomic operations (ld, st, cmpxchg). */ );
 253 }
 254
 255 static void
 256 cpu_startup(void *dummy)
 257 {
 258         char nodename[16];
 259         struct pcpu *pc;
 260         struct pcpu_stats *pcs;
 261
 262         /*
 263          * Good {morning,afternoon,evening,night}.
 264          */
 265         identifycpu();
 266
 267 #ifdef PERFMON
 268         perfmon_init();
 269 #endif
 270         printf("real memory  = %ld (%ld MB)\n", ia64_ptob(Maxmem),
 271             ia64_ptob(Maxmem) / 1048576);
 272         realmem = Maxmem;
 273
 274         /*
 275          * Display any holes after the first chunk of extended memory.
 276          */
 277         if (bootverbose) {
 278                 int indx;
 279
 280                 printf("Physical memory chunk(s):\n");
 281                 for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
 282                         long size1 = phys_avail[indx + 1] - phys_avail[indx];
 283
 284                         printf("0x%08lx - 0x%08lx, %ld bytes (%ld pages)\n",
 285                             phys_avail[indx], phys_avail[indx + 1] - 1, size1,
 286                             size1 >> PAGE_SHIFT);
 287                 }
 288         }
 289
 290         vm_ksubmap_init(&kmi);
 291
 292         printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count),
 293             ptoa(cnt.v_free_count) / 1048576);
 294
 295         if (fpswa_iface == NULL)
 296                 printf("Warning: no FPSWA package supplied\n");
 297         else
 298                 printf("FPSWA Revision = 0x%lx, Entry = %p\n",
 299                     (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa);
 300
 301         /*
 302          * Set up buffers, so they can be used to read disk labels.
 303          */
 304         bufinit();
 305         vm_pager_bufferinit();
 306
 307         /*
 308          * Traverse the MADT to discover IOSAPIC and Local SAPIC
 309          * information.
 310          */
 311         ia64_probe_sapics();
 312         ia64_pib = pmap_mapdev(ia64_lapic_addr, sizeof(*ia64_pib));
 313
 314         ia64_mca_init();
 315
 316         /*
 317          * Create sysctl tree for per-CPU information.
 318          */
 319         SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
 320                 snprintf(nodename, sizeof(nodename), "%u", pc->pc_cpuid);
 321                 sysctl_ctx_init(&pc->pc_md.sysctl_ctx);
 322                 pc->pc_md.sysctl_tree = SYSCTL_ADD_NODE(&pc->pc_md.sysctl_ctx,
 323                     SYSCTL_STATIC_CHILDREN(_machdep_cpu), OID_AUTO, nodename,
 324                     CTLFLAG_RD, NULL, "");
 325                 if (pc->pc_md.sysctl_tree == NULL)
 326                         continue;
 327
 328                 pcs = &pc->pc_md.stats;
 329
 330                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 331                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 332                     "nasts", CTLFLAG_RD, &pcs->pcs_nasts,
 333                     "Number of IPI_AST interrupts");
 334
 335                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 336                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 337                     "nclks", CTLFLAG_RD, &pcs->pcs_nclks,
 338                     "Number of clock interrupts");
 339
 340                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 341                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 342                     "nextints", CTLFLAG_RD, &pcs->pcs_nextints,
 343                     "Number of ExtINT interrupts");
 344
 345                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 346                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 347                     "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps,
 348                     "Number of IPI_HIGH_FP interrupts");
 349
 350                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 351                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 352                     "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints,
 353                     "Number of hardware (device) interrupts");
 354
 355                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 356                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 357                     "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts,
 358                     "Number of IPI_PREEMPT interrupts");
 359
 360                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 361                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 362                     "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs,
 363                     "Number of IPI_RENDEZVOUS interrupts");
 364
 365                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 366                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 367                     "nstops", CTLFLAG_RD, &pcs->pcs_nstops,
 368                     "Number of IPI_STOP interrupts");
 369
 370                 SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx,
 371                     SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO,
 372                     "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays,
 373                     "Number of stray interrupts");
 374         }
 375 }
 376 SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL);
 377
 378 void
 379 cpu_flush_dcache(void *ptr, size_t len)
 380 {
 381         vm_offset_t lim, va;
 382
 383         va = (uintptr_t)ptr & ~31;
 384         lim = (uintptr_t)ptr + len;
 385         while (va < lim) {
 386                 ia64_fc(va);
 387                 va += 32;
 388         }
 389
 390         ia64_srlz_d();
 391 }
 392
 393 /* Get current clock frequency for the given cpu id. */
 394 int
 395 cpu_est_clockrate(int cpu_id, uint64_t *rate)
 396 {
 397
 398         if (pcpu_find(cpu_id) == NULL || rate == NULL)
 399                 return (EINVAL);
 400         *rate = (u_long)cpu_freq * 1000000ul;
 401         return (0);
 402 }
 403
 404 void
 405 cpu_halt()
 406 {
 407
 408         efi_reset_system();
 409 }
 410
 411 void
 412 cpu_idle(int busy)
 413 {
 414         struct ia64_pal_result res;
 415
 416         if (cpu_idle_hook != NULL)
 417                 (*cpu_idle_hook)();
 418         else
 419                 res = ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
 420 }
 421
 422 int
 423 cpu_idle_wakeup(int cpu)
 424 {
 425
 426         return (0);
 427 }
 428
 429 void
 430 cpu_reset()
 431 {
 432
 433         efi_reset_system();
 434 }
 435
 436 void
 437 cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx)
 438 {
 439         struct pcb *oldpcb, *newpcb;
 440
 441         oldpcb = old->td_pcb;
 442 #ifdef COMPAT_FREEBSD32
 443         ia32_savectx(oldpcb);
 444 #endif
 445         if (PCPU_GET(fpcurthread) == old)
 446                 old->td_frame->tf_special.psr |= IA64_PSR_DFH;
 447         if (!savectx(oldpcb)) {
 448                 atomic_store_rel_ptr(&old->td_lock, mtx);
 449
 450                 newpcb = new->td_pcb;
 451                 oldpcb->pcb_current_pmap =
 452                     pmap_switch(newpcb->pcb_current_pmap);
 453
 454 #if defined(SCHED_ULE) && defined(SMP)
 455                 while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
 456                         cpu_spinwait();
 457 #endif
 458
 459                 PCPU_SET(curthread, new);
 460
 461 #ifdef COMPAT_FREEBSD32
 462                 ia32_restorectx(newpcb);
 463 #endif
 464
 465                 if (PCPU_GET(fpcurthread) == new)
 466                         new->td_frame->tf_special.psr &= ~IA64_PSR_DFH;
 467                 restorectx(newpcb);
 468                 /* We should not get here. */
 469                 panic("cpu_switch: restorectx() returned");
 470                 /* NOTREACHED */
 471         }
 472 }
 473
 474 void
 475 cpu_throw(struct thread *old __unused, struct thread *new)
 476 {
 477         struct pcb *newpcb;
 478
 479         newpcb = new->td_pcb;
 480         (void)pmap_switch(newpcb->pcb_current_pmap);
 481
 482 #if defined(SCHED_ULE) && defined(SMP)
 483         while (atomic_load_acq_ptr(&new->td_lock) == &blocked_lock)
 484                 cpu_spinwait();
 485 #endif
 486
 487         PCPU_SET(curthread, new);
 488
 489 #ifdef COMPAT_FREEBSD32
 490         ia32_restorectx(newpcb);
 491 #endif
 492
 493         restorectx(newpcb);
 494         /* We should not get here. */
 495         panic("cpu_throw: restorectx() returned");
 496         /* NOTREACHED */
 497 }
 498
 499 void
 500 cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
 501 {
 502
 503         /*
 504          * Set pc_acpi_id to "uninitialized".
 505          * See sys/dev/acpica/acpi_cpu.c
 506          */
 507         pcpu->pc_acpi_id = 0xffffffff;
 508 }
 509
 510 void
 511 spinlock_enter(void)
 512 {
 513         struct thread *td;
 514         int intr;
 515
 516         td = curthread;
 517         if (td->td_md.md_spinlock_count == 0) {
 518                 intr = intr_disable();
 519                 td->td_md.md_spinlock_count = 1;
 520                 td->td_md.md_saved_intr = intr;
 521         } else
 522                 td->td_md.md_spinlock_count++;
 523         critical_enter();
 524 }
 525
 526 void
 527 spinlock_exit(void)
 528 {
 529         struct thread *td;
 530         int intr;
 531
 532         td = curthread;
 533         critical_exit();
 534         intr = td->td_md.md_saved_intr;
 535         td->td_md.md_spinlock_count--;
 536         if (td->td_md.md_spinlock_count == 0)
 537                 intr_restore(intr);
 538 }
 539
 540 void
 541 map_vhpt(uintptr_t vhpt)
 542 {
 543         pt_entry_t pte;
 544         uint64_t psr;
 545
 546         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 547             PTE_PL_KERN | PTE_AR_RW;
 548         pte |= vhpt & PTE_PPN_MASK;
 549
 550         __asm __volatile("ptr.d %0,%1" :: "r"(vhpt),
 551             "r"(IA64_ID_PAGE_SHIFT<<2));
 552
 553         __asm __volatile("mov   %0=psr" : "=r"(psr));
 554         __asm __volatile("rsm   psr.ic|psr.i");
 555         ia64_srlz_i();
 556         ia64_set_ifa(vhpt);
 557         ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
 558         ia64_srlz_d();
 559         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(2), "r"(pte));
 560         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 561         ia64_srlz_i();
 562 }
 563
 564 void
 565 map_pal_code(void)
 566 {
 567         pt_entry_t pte;
 568         uint64_t psr;
 569
 570         if (ia64_pal_base == 0)
 571                 return;
 572
 573         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 574             PTE_PL_KERN | PTE_AR_RWX;
 575         pte |= ia64_pal_base & PTE_PPN_MASK;
 576
 577         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
 578             "r"(IA64_PHYS_TO_RR7(ia64_pal_base)), "r"(IA64_ID_PAGE_SHIFT<<2));
 579
 580         __asm __volatile("mov   %0=psr" : "=r"(psr));
 581         __asm __volatile("rsm   psr.ic|psr.i");
 582         ia64_srlz_i();
 583         ia64_set_ifa(IA64_PHYS_TO_RR7(ia64_pal_base));
 584         ia64_set_itir(IA64_ID_PAGE_SHIFT << 2);
 585         ia64_srlz_d();
 586         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(1), "r"(pte));
 587         ia64_srlz_d();
 588         __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte));
 589         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 590         ia64_srlz_i();
 591 }
 592
 593 void
 594 map_gateway_page(void)
 595 {
 596         pt_entry_t pte;
 597         uint64_t psr;
 598
 599         pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY |
 600             PTE_PL_KERN | PTE_AR_X_RX;
 601         pte |= (uint64_t)ia64_gateway_page & PTE_PPN_MASK;
 602
 603         __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" ::
 604             "r"(VM_MAX_ADDRESS), "r"(PAGE_SHIFT << 2));
 605
 606         __asm __volatile("mov   %0=psr" : "=r"(psr));
 607         __asm __volatile("rsm   psr.ic|psr.i");
 608         ia64_srlz_i();
 609         ia64_set_ifa(VM_MAX_ADDRESS);
 610         ia64_set_itir(PAGE_SHIFT << 2);
 611         ia64_srlz_d();
 612         __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte));
 613         ia64_srlz_d();
 614         __asm __volatile("itr.i itr[%0]=%1" :: "r"(3), "r"(pte));
 615         __asm __volatile("mov   psr.l=%0" :: "r" (psr));
 616         ia64_srlz_i();
 617
 618         /* Expose the mapping to userland in ar.k5 */
 619         ia64_set_k5(VM_MAX_ADDRESS);
 620 }
 621
 622 static u_int
 623 freq_ratio(u_long base, u_long ratio)
 624 {
 625         u_long f;
 626
 627         f = (base * (ratio >> 32)) / (ratio & 0xfffffffful);
 628         return ((f + 500000) / 1000000);
 629 }
 630
 631 static void
 632 calculate_frequencies(void)
 633 {
 634         struct ia64_sal_result sal;
 635         struct ia64_pal_result pal;
 636
 637         sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0);
 638         pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0);
 639
 640         if (sal.sal_status == 0 && pal.pal_status == 0) {
 641                 if (bootverbose) {
 642                         printf("Platform clock frequency %ld Hz\n",
 643                                sal.sal_result[0]);
 644                         printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, "
 645                                "ITC ratio %ld/%ld\n",
 646                                pal.pal_result[0] >> 32,
 647                                pal.pal_result[0] & ((1L << 32) - 1),
 648                                pal.pal_result[1] >> 32,
 649                                pal.pal_result[1] & ((1L << 32) - 1),
 650                                pal.pal_result[2] >> 32,
 651                                pal.pal_result[2] & ((1L << 32) - 1));
 652                 }
 653                 cpu_freq = freq_ratio(sal.sal_result[0], pal.pal_result[0]);
 654                 bus_freq = freq_ratio(sal.sal_result[0], pal.pal_result[1]);
 655                 itc_freq = freq_ratio(sal.sal_result[0], pal.pal_result[2]);
 656         }
 657 }
 658
 659 struct ia64_init_return
 660 ia64_init(void)
 661 {
 662         struct ia64_init_return ret;
 663         int phys_avail_cnt;
 664         vm_offset_t kernstart, kernend;
 665         vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1;
 666         char *p;
 667         struct efi_md *md;
 668         int metadata_missing;
 669
 670         /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */
 671
 672         /*
 673          * TODO: Disable interrupts, floating point etc.
 674          * Maybe flush cache and tlb
 675          */
 676         ia64_set_fpsr(IA64_FPSR_DEFAULT);
 677
 678         /*
 679          * TODO: Get critical system information (if possible, from the
 680          * information provided by the boot program).
 681          */
 682
 683         /*
 684          * pa_bootinfo is the physical address of the bootinfo block as
 685          * passed to us by the loader and set in locore.s.
 686          */
 687         bootinfo = *(struct bootinfo *)(IA64_PHYS_TO_RR7(pa_bootinfo));
 688
 689         if (bootinfo.bi_magic != BOOTINFO_MAGIC || bootinfo.bi_version != 1) {
 690                 bzero(&bootinfo, sizeof(bootinfo));
 691                 bootinfo.bi_kernend = (vm_offset_t) round_page(_end);
 692         }
 693
 694         /*
 695          * Look for the I/O ports first - we need them for console
 696          * probing.
 697          */
 698         for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
 699                 switch (md->md_type) {
 700                 case EFI_MD_TYPE_IOPORT:
 701                         ia64_port_base = (uintptr_t)pmap_mapdev(md->md_phys,
 702                             md->md_pages * EFI_PAGE_SIZE);
 703                         break;
 704                 case EFI_MD_TYPE_PALCODE:
 705                         ia64_pal_base = md->md_phys;
 706                         break;
 707                 }
 708         }
 709
 710         metadata_missing = 0;
 711         if (bootinfo.bi_modulep)
 712                 preload_metadata = (caddr_t)bootinfo.bi_modulep;
 713         else
 714                 metadata_missing = 1;
 715
 716         if (envmode == 0 && bootinfo.bi_envp)
 717                 kern_envp = (caddr_t)bootinfo.bi_envp;
 718         else
 719                 kern_envp = static_env;
 720
 721         /*
 722          * Look at arguments passed to us and compute boothowto.
 723          */
 724         boothowto = bootinfo.bi_boothowto;
 725
 726         if (boothowto & RB_VERBOSE)
 727                 bootverbose = 1;
 728
 729         /*
 730          * Find the beginning and end of the kernel.
 731          */
 732         kernstart = trunc_page(kernel_text);
 733 #ifdef DDB
 734         ksym_start = bootinfo.bi_symtab;
 735         ksym_end = bootinfo.bi_esymtab;
 736         kernend = (vm_offset_t)round_page(ksym_end);
 737 #else
 738         kernend = (vm_offset_t)round_page(_end);
 739 #endif
 740         /* But if the bootstrap tells us otherwise, believe it! */
 741         if (bootinfo.bi_kernend)
 742                 kernend = round_page(bootinfo.bi_kernend);
 743
 744         /*
 745          * Setup the PCPU data for the bootstrap processor. It is needed
 746          * by printf(). Also, since printf() has critical sections, we
 747          * need to initialize at least pc_curthread.
 748          */
 749         pcpup = &pcpu0;
 750         ia64_set_k4((u_int64_t)pcpup);
 751         pcpu_init(pcpup, 0, sizeof(pcpu0));
 752         dpcpu_init((void *)kernend, 0);
 753         kernend += DPCPU_SIZE;
 754         PCPU_SET(curthread, &thread0);
 755
 756         /*
 757          * Initialize the console before we print anything out.
 758          */
 759         cninit();
 760
 761         /* OUTPUT NOW ALLOWED */
 762
 763         if (ia64_pal_base != 0) {
 764                 ia64_pal_base &= ~IA64_ID_PAGE_MASK;
 765                 /*
 766                  * We use a TR to map the first 256M of memory - this might
 767                  * cover the palcode too.
 768                  */
 769                 if (ia64_pal_base == 0)
 770                         printf("PAL code mapped by the kernel's TR\n");
 771         } else
 772                 printf("PAL code not found\n");
 773
 774         /*
 775          * Wire things up so we can call the firmware.
 776          */
 777         map_pal_code();
 778         efi_boot_minimal(bootinfo.bi_systab);
 779         ia64_xiv_init();
 780         ia64_sal_init();
 781         calculate_frequencies();
 782
 783         if (metadata_missing)
 784                 printf("WARNING: loader(8) metadata is missing!\n");
 785
 786         /* Get FPSWA interface */
 787         fpswa_iface = (bootinfo.bi_fpswa == 0) ? NULL :
 788             (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo.bi_fpswa);
 789
 790         /* Init basic tunables, including hz */
 791         init_param1();
 792
 793         p = getenv("kernelname");
 794         if (p != NULL) {
 795                 strncpy(kernelname, p, sizeof(kernelname) - 1);
 796                 freeenv(p);
 797         }
 798
 799         kernstartpfn = atop(IA64_RR_MASK(kernstart));
 800         kernendpfn = atop(IA64_RR_MASK(kernend));
 801
 802         /*
 803          * Size the memory regions and load phys_avail[] with the results.
 804          */
 805
 806         /*
 807          * Find out how much memory is available, by looking at
 808          * the memory descriptors.
 809          */
 810
 811 #ifdef DEBUG_MD
 812         printf("Memory descriptor count: %d\n", mdcount);
 813 #endif
 814
 815         phys_avail_cnt = 0;
 816         for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) {
 817 #ifdef DEBUG_MD
 818                 printf("MD %p: type %d pa 0x%lx cnt 0x%lx\n", md,
 819                     md->md_type, md->md_phys, md->md_pages);
 820 #endif
 821
 822                 pfn0 = ia64_btop(round_page(md->md_phys));
 823                 pfn1 = ia64_btop(trunc_page(md->md_phys + md->md_pages * 4096));
 824                 if (pfn1 <= pfn0)
 825                         continue;
 826
 827                 if (md->md_type != EFI_MD_TYPE_FREE)
 828                         continue;
 829
 830                 /*
 831                  * We have a memory descriptor that describes conventional
 832                  * memory that is for general use. We must determine if the
 833                  * loader has put the kernel in this region.
 834                  */
 835                 physmem += (pfn1 - pfn0);
 836                 if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) {
 837                         /*
 838                          * Must compute the location of the kernel
 839                          * within the segment.
 840                          */
 841 #ifdef DEBUG_MD
 842                         printf("Descriptor %p contains kernel\n", mp);
 843 #endif
 844                         if (pfn0 < kernstartpfn) {
 845                                 /*
 846                                  * There is a chunk before the kernel.
 847                                  */
 848 #ifdef DEBUG_MD
 849                                 printf("Loading chunk before kernel: "
 850                                        "0x%lx / 0x%lx\n", pfn0, kernstartpfn);
 851 #endif
 852                                 phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
 853                                 phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn);
 854                                 phys_avail_cnt += 2;
 855                         }
 856                         if (kernendpfn < pfn1) {
 857                                 /*
 858                                  * There is a chunk after the kernel.
 859                                  */
 860 #ifdef DEBUG_MD
 861                                 printf("Loading chunk after kernel: "
 862                                        "0x%lx / 0x%lx\n", kernendpfn, pfn1);
 863 #endif
 864                                 phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn);
 865                                 phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
 866                                 phys_avail_cnt += 2;
 867                         }
 868                 } else {
 869                         /*
 870                          * Just load this cluster as one chunk.
 871                          */
 872 #ifdef DEBUG_MD
 873                         printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i,
 874                                pfn0, pfn1);
 875 #endif
 876                         phys_avail[phys_avail_cnt] = ia64_ptob(pfn0);
 877                         phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1);
 878                         phys_avail_cnt += 2;
 879
 880                 }
 881         }
 882         phys_avail[phys_avail_cnt] = 0;
 883
 884         Maxmem = physmem;
 885         init_param2(physmem);
 886
 887         /*
 888          * Initialize error message buffer (at end of core).
 889          */
 890         msgbufp = (struct msgbuf *)pmap_steal_memory(msgbufsize);
 891         msgbufinit(msgbufp, msgbufsize);
 892
 893         proc_linkup0(&proc0, &thread0);
 894         /*
 895          * Init mapping for kernel stack for proc 0
 896          */
 897         thread0.td_kstack = pmap_steal_memory(KSTACK_PAGES * PAGE_SIZE);
 898         thread0.td_kstack_pages = KSTACK_PAGES;
 899
 900         mutex_init();
 901
 902         /*
 903          * Initialize the rest of proc 0's PCB.
 904          *
 905          * Set the kernel sp, reserving space for an (empty) trapframe,
 906          * and make proc0's trapframe pointer point to it for sanity.
 907          * Initialise proc0's backing store to start after u area.
 908          */
 909         cpu_thread_alloc(&thread0);
 910         thread0.td_frame->tf_flags = FRAME_SYSCALL;
 911         thread0.td_pcb->pcb_special.sp =
 912             (u_int64_t)thread0.td_frame - 16;
 913         thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack;
 914
 915         /*
 916          * Initialize the virtual memory system.
 917          */
 918         pmap_bootstrap();
 919
 920         /*
 921          * Initialize debuggers, and break into them if appropriate.
 922          */
 923         kdb_init();
 924
 925 #ifdef KDB
 926         if (boothowto & RB_KDB)
 927                 kdb_enter(KDB_WHY_BOOTFLAGS,
 928                     "Boot flags requested debugger\n");
 929 #endif
 930
 931         ia64_set_tpr(0);
 932         ia64_srlz_d();
 933
 934         ret.bspstore = thread0.td_pcb->pcb_special.bspstore;
 935         ret.sp = thread0.td_pcb->pcb_special.sp;
 936         return (ret);
 937 }
 938
 939 uint64_t
 940 ia64_get_hcdp(void)
 941 {
 942
 943         return (bootinfo.bi_hcdp);
 944 }
 945
 946 void
 947 bzero(void *buf, size_t len)
 948 {
 949         caddr_t p = buf;
 950
 951         while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) {
 952                 *p++ = 0;
 953                 len--;
 954         }
 955         while (len >= sizeof(u_long) * 8) {
 956                 *(u_long*) p = 0;
 957                 *((u_long*) p + 1) = 0;
 958                 *((u_long*) p + 2) = 0;
 959                 *((u_long*) p + 3) = 0;
 960                 len -= sizeof(u_long) * 8;
 961                 *((u_long*) p + 4) = 0;
 962                 *((u_long*) p + 5) = 0;
 963                 *((u_long*) p + 6) = 0;
 964                 *((u_long*) p + 7) = 0;
 965                 p += sizeof(u_long) * 8;
 966         }
 967         while (len >= sizeof(u_long)) {
 968                 *(u_long*) p = 0;
 969                 len -= sizeof(u_long);
 970                 p += sizeof(u_long);
 971         }
 972         while (len) {
 973                 *p++ = 0;
 974                 len--;
 975         }
 976 }
 977
 978 u_int
 979 ia64_itc_freq(void)
 980 {
 981
 982         return (itc_freq);
 983 }
 984
 985 void
 986 DELAY(int n)
 987 {
 988         u_int64_t start, end, now;
 989
 990         sched_pin();
 991
 992         start = ia64_get_itc();
 993         end = start + itc_freq * n;
 994         /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */
 995         do {
 996                 now = ia64_get_itc();
 997         } while (now < end || (now > start && end < start));
 998
 999         sched_unpin();
1000 }
1001
1002 /*
1003  * Send an interrupt (signal) to a process.
1004  */
1005 void
1006 sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
1007 {
1008         struct proc *p;
1009         struct thread *td;
1010         struct trapframe *tf;
1011         struct sigacts *psp;
1012         struct sigframe sf, *sfp;
1013         u_int64_t sbs, sp;
1014         int oonstack;
1015         int sig;
1016         u_long code;
1017
1018         td = curthread;
1019         p = td->td_proc;
1020         PROC_LOCK_ASSERT(p, MA_OWNED);
1021         sig = ksi->ksi_signo;
1022         code = ksi->ksi_code;
1023         psp = p->p_sigacts;
1024         mtx_assert(&psp->ps_mtx, MA_OWNED);
1025         tf = td->td_frame;
1026         sp = tf->tf_special.sp;
1027         oonstack = sigonstack(sp);
1028         sbs = 0;
1029
1030         /* save user context */
1031         bzero(&sf, sizeof(struct sigframe));
1032         sf.sf_uc.uc_sigmask = *mask;
1033         sf.sf_uc.uc_stack = td->td_sigstk;
1034         sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
1035             ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1036
1037         /*
1038          * Allocate and validate space for the signal handler
1039          * context. Note that if the stack is in P0 space, the
1040          * call to grow() is a nop, and the useracc() check
1041          * will fail if the process has not already allocated
1042          * the space with a `brk'.
1043          */
1044         if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
1045             SIGISMEMBER(psp->ps_sigonstack, sig)) {
1046                 sbs = (u_int64_t)td->td_sigstk.ss_sp;
1047                 sbs = (sbs + 15) & ~15;
1048                 sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size);
1049 #if defined(COMPAT_43)
1050                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1051 #endif
1052         } else
1053                 sfp = (struct sigframe *)sp;
1054         sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15);
1055
1056         /* Fill in the siginfo structure for POSIX handlers. */
1057         if (SIGISMEMBER(psp->ps_siginfo, sig)) {
1058                 sf.sf_si = ksi->ksi_info;
1059                 sf.sf_si.si_signo = sig;
1060                 /*
1061                  * XXX this shouldn't be here after code in trap.c
1062                  * is fixed
1063                  */
1064                 sf.sf_si.si_addr = (void*)tf->tf_special.ifa;
1065                 code = (u_int64_t)&sfp->sf_si;
1066         }
1067
1068         mtx_unlock(&psp->ps_mtx);
1069         PROC_UNLOCK(p);
1070
1071         get_mcontext(td, &sf.sf_uc.uc_mcontext, 0);
1072
1073         /* Copy the frame out to userland. */
1074         if (copyout(&sf, sfp, sizeof(sf)) != 0) {
1075                 /*
1076                  * Process has trashed its stack; give it an illegal
1077                  * instruction to halt it in its tracks.
1078                  */
1079                 PROC_LOCK(p);
1080                 sigexit(td, SIGILL);
1081                 return;
1082         }
1083
1084         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {
1085                 tf->tf_special.psr &= ~IA64_PSR_RI;
1086                 tf->tf_special.iip = ia64_get_k5() +
1087                     ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page);
1088         } else
1089                 tf->tf_special.iip = ia64_get_k5() +
1090                     ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page);
1091
1092         /*
1093          * Setup the trapframe to return to the signal trampoline. We pass
1094          * information to the trampoline in the following registers:
1095          *
1096          *      gp      new backing store or NULL
1097          *      r8      signal number
1098          *      r9      signal code or siginfo pointer
1099          *      r10     signal handler (function descriptor)
1100          */
1101         tf->tf_special.sp = (u_int64_t)sfp - 16;
1102         tf->tf_special.gp = sbs;
1103         tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore;
1104         tf->tf_special.ndirty = 0;
1105         tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat;
1106         tf->tf_scratch.gr8 = sig;
1107         tf->tf_scratch.gr9 = code;
1108         tf->tf_scratch.gr10 = (u_int64_t)catcher;
1109
1110         PROC_LOCK(p);
1111         mtx_lock(&psp->ps_mtx);
1112 }
1113
1114 /*
1115  * System call to cleanup state after a signal
1116  * has been taken.  Reset signal mask and
1117  * stack state from context left by sendsig (above).
1118  * Return to previous pc and psl as specified by
1119  * context left by sendsig. Check carefully to
1120  * make sure that the user has not modified the
1121  * state to gain improper privileges.
1122  *
1123  * MPSAFE
1124  */
1125 int
1126 sigreturn(struct thread *td,
1127         struct sigreturn_args /* {
1128                 ucontext_t *sigcntxp;
1129         } */ *uap)
1130 {
1131         ucontext_t uc;
1132         struct trapframe *tf;
1133         struct pcb *pcb;
1134
1135         tf = td->td_frame;
1136         pcb = td->td_pcb;
1137
1138         /*
1139          * Fetch the entire context structure at once for speed.
1140          * We don't use a normal argument to simplify RSE handling.
1141          */
1142         if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc)))
1143                 return (EFAULT);
1144
1145         set_mcontext(td, &uc.uc_mcontext);
1146
1147 #if defined(COMPAT_43)
1148         if (sigonstack(tf->tf_special.sp))
1149                 td->td_sigstk.ss_flags |= SS_ONSTACK;
1150         else
1151                 td->td_sigstk.ss_flags &= ~SS_ONSTACK;
1152 #endif
1153         kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
1154
1155         return (EJUSTRETURN);
1156 }
1157
1158 #ifdef COMPAT_FREEBSD4
1159 int
1160 freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap)
1161 {
1162
1163         return sigreturn(td, (struct sigreturn_args *)uap);
1164 }
1165 #endif
1166
1167 /*
1168  * Construct a PCB from a trapframe. This is called from kdb_trap() where
1169  * we want to start a backtrace from the function that caused us to enter
1170  * the debugger. We have the context in the trapframe, but base the trace
1171  * on the PCB. The PCB doesn't have to be perfect, as long as it contains
1172  * enough for a backtrace.
1173  */
1174 void
1175 makectx(struct trapframe *tf, struct pcb *pcb)
1176 {
1177
1178         pcb->pcb_special = tf->tf_special;
1179         pcb->pcb_special.__spare = ~0UL;        /* XXX see unwind.c */
1180         save_callee_saved(&pcb->pcb_preserved);
1181         save_callee_saved_fp(&pcb->pcb_preserved_fp);
1182 }
1183
1184 int
1185 ia64_flush_dirty(struct thread *td, struct _special *r)
1186 {
1187         struct iovec iov;
1188         struct uio uio;
1189         uint64_t bspst, kstk, rnat;
1190         int error, locked;
1191
1192         if (r->ndirty == 0)
1193                 return (0);
1194
1195         kstk = td->td_kstack + (r->bspstore & 0x1ffUL);
1196         if (td == curthread) {
1197                 __asm __volatile("mov   ar.rsc=0;;");
1198                 __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1199                 /* Make sure we have all the user registers written out. */
1200                 if (bspst - kstk < r->ndirty) {
1201                         __asm __volatile("flushrs;;");
1202                         __asm __volatile("mov   %0=ar.bspstore" : "=r"(bspst));
1203                 }
1204                 __asm __volatile("mov   %0=ar.rnat;;" : "=r"(rnat));
1205                 __asm __volatile("mov   ar.rsc=3");
1206                 error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty);
1207                 kstk += r->ndirty;
1208                 r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL))
1209                     ? *(uint64_t*)(kstk | 0x1f8L) : rnat;
1210         } else {
1211                 locked = PROC_LOCKED(td->td_proc);
1212                 if (!locked)
1213                         PHOLD(td->td_proc);
1214                 iov.iov_base = (void*)(uintptr_t)kstk;
1215                 iov.iov_len = r->ndirty;
1216                 uio.uio_iov = &iov;
1217                 uio.uio_iovcnt = 1;
1218                 uio.uio_offset = r->bspstore;
1219                 uio.uio_resid = r->ndirty;
1220                 uio.uio_segflg = UIO_SYSSPACE;
1221                 uio.uio_rw = UIO_WRITE;
1222                 uio.uio_td = td;
1223                 error = proc_rwmem(td->td_proc, &uio);
1224                 /*
1225                  * XXX proc_rwmem() doesn't currently return ENOSPC,
1226                  * so I think it can bogusly return 0. Neither do
1227                  * we allow short writes.
1228                  */
1229                 if (uio.uio_resid != 0 && error == 0)
1230                         error = ENOSPC;
1231                 if (!locked)
1232                         PRELE(td->td_proc);
1233         }
1234
1235         r->bspstore += r->ndirty;
1236         r->ndirty = 0;
1237         return (error);
1238 }
1239
1240 int
1241 get_mcontext(struct thread *td, mcontext_t *mc, int flags)
1242 {
1243         struct trapframe *tf;
1244         int error;
1245
1246         tf = td->td_frame;
1247         bzero(mc, sizeof(*mc));
1248         mc->mc_special = tf->tf_special;
1249         error = ia64_flush_dirty(td, &mc->mc_special);
1250         if (tf->tf_flags & FRAME_SYSCALL) {
1251                 mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT;
1252                 mc->mc_scratch = tf->tf_scratch;
1253                 if (flags & GET_MC_CLEAR_RET) {
1254                         mc->mc_scratch.gr8 = 0;
1255                         mc->mc_scratch.gr9 = 0;
1256                         mc->mc_scratch.gr10 = 0;
1257                         mc->mc_scratch.gr11 = 0;
1258                 }
1259         } else {
1260                 mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT;
1261                 mc->mc_scratch = tf->tf_scratch;
1262                 mc->mc_scratch_fp = tf->tf_scratch_fp;
1263                 /*
1264                  * XXX If the thread never used the high FP registers, we
1265                  * probably shouldn't waste time saving them.
1266                  */
1267                 ia64_highfp_save(td);
1268                 mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID;
1269                 mc->mc_high_fp = td->td_pcb->pcb_high_fp;
1270         }
1271         save_callee_saved(&mc->mc_preserved);
1272         save_callee_saved_fp(&mc->mc_preserved_fp);
1273         return (error);
1274 }
1275
1276 int
1277 set_mcontext(struct thread *td, const mcontext_t *mc)
1278 {
1279         struct _special s;
1280         struct trapframe *tf;
1281         uint64_t psrmask;
1282
1283         tf = td->td_frame;
1284
1285         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1286             ("Whoa there! We have more than 8KB of dirty registers!"));
1287
1288         s = mc->mc_special;
1289         /*
1290          * Only copy the user mask and the restart instruction bit from
1291          * the new context.
1292          */
1293         psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL |
1294             IA64_PSR_MFH | IA64_PSR_RI;
1295         s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask);
1296         /* We don't have any dirty registers of the new context. */
1297         s.ndirty = 0;
1298         if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) {
1299                 /*
1300                  * We can get an async context passed to us while we
1301                  * entered the kernel through a syscall: sigreturn(2)
1302                  * takes contexts that could previously be the result of
1303                  * a trap or interrupt.
1304                  * Hence, we cannot assert that the trapframe is not
1305                  * a syscall frame, but we can assert that it's at
1306                  * least an expected syscall.
1307                  */
1308                 if (tf->tf_flags & FRAME_SYSCALL) {
1309                         KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo"));
1310                         tf->tf_flags &= ~FRAME_SYSCALL;
1311                 }
1312                 tf->tf_scratch = mc->mc_scratch;
1313                 tf->tf_scratch_fp = mc->mc_scratch_fp;
1314                 if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID)
1315                         td->td_pcb->pcb_high_fp = mc->mc_high_fp;
1316         } else {
1317                 KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo"));
1318                 if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) {
1319                         s.cfm = tf->tf_special.cfm;
1320                         s.iip = tf->tf_special.iip;
1321                         tf->tf_scratch.gr15 = 0;        /* Clear syscall nr. */
1322                 } else
1323                         tf->tf_scratch = mc->mc_scratch;
1324         }
1325         tf->tf_special = s;
1326         restore_callee_saved(&mc->mc_preserved);
1327         restore_callee_saved_fp(&mc->mc_preserved_fp);
1328
1329         return (0);
1330 }
1331
1332 /*
1333  * Clear registers on exec.
1334  */
1335 void
1336 exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings)
1337 {
1338         struct trapframe *tf;
1339         uint64_t *ksttop, *kst;
1340
1341         tf = td->td_frame;
1342         ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty +
1343             (tf->tf_special.bspstore & 0x1ffUL));
1344
1345         /*
1346          * We can ignore up to 8KB of dirty registers by masking off the
1347          * lower 13 bits in exception_restore() or epc_syscall(). This
1348          * should be enough for a couple of years, but if there are more
1349          * than 8KB of dirty registers, we lose track of the bottom of
1350          * the kernel stack. The solution is to copy the active part of
1351          * the kernel stack down 1 page (or 2, but not more than that)
1352          * so that we always have less than 8KB of dirty registers.
1353          */
1354         KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0,
1355             ("Whoa there! We have more than 8KB of dirty registers!"));
1356
1357         bzero(&tf->tf_special, sizeof(tf->tf_special));
1358         if ((tf->tf_flags & FRAME_SYSCALL) == 0) {      /* break syscalls. */
1359                 bzero(&tf->tf_scratch, sizeof(tf->tf_scratch));
1360                 bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp));
1361                 tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL;
1362                 tf->tf_special.bspstore = IA64_BACKINGSTORE;
1363                 /*
1364                  * Copy the arguments onto the kernel register stack so that
1365                  * they get loaded by the loadrs instruction. Skip over the
1366                  * NaT collection points.
1367                  */
1368                 kst = ksttop - 1;
1369                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1370                         *kst-- = 0;
1371                 *kst-- = 0;
1372                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1373                         *kst-- = 0;
1374                 *kst-- = ps_strings;
1375                 if (((uintptr_t)kst & 0x1ff) == 0x1f8)
1376                         *kst-- = 0;
1377                 *kst = stack;
1378                 tf->tf_special.ndirty = (ksttop - kst) << 3;
1379         } else {                                /* epc syscalls (default). */
1380                 tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL;
1381                 tf->tf_special.bspstore = IA64_BACKINGSTORE + 24;
1382                 /*
1383                  * Write values for out0, out1 and out2 to the user's backing
1384                  * store and arrange for them to be restored into the user's
1385                  * initial register frame.
1386                  * Assumes that (bspstore & 0x1f8) < 0x1e0.
1387                  */
1388                 suword((caddr_t)tf->tf_special.bspstore - 24, stack);
1389                 suword((caddr_t)tf->tf_special.bspstore - 16, ps_strings);
1390                 suword((caddr_t)tf->tf_special.bspstore -  8, 0);
1391         }
1392
1393         tf->tf_special.iip = entry;
1394         tf->tf_special.sp = (stack & ~15) - 16;
1395         tf->tf_special.rsc = 0xf;
1396         tf->tf_special.fpsr = IA64_FPSR_DEFAULT;
1397         tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT |
1398             IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN |
1399             IA64_PSR_CPL_USER;
1400 }
1401
1402 int
1403 ptrace_set_pc(struct thread *td, unsigned long addr)
1404 {
1405         uint64_t slot;
1406
1407         switch (addr & 0xFUL) {
1408         case 0:
1409                 slot = IA64_PSR_RI_0;
1410                 break;
1411         case 1:
1412                 /* XXX we need to deal with MLX bundles here */
1413                 slot = IA64_PSR_RI_1;
1414                 break;
1415         case 2:
1416                 slot = IA64_PSR_RI_2;
1417                 break;
1418         default:
1419                 return (EINVAL);
1420         }
1421
1422         td->td_frame->tf_special.iip = addr & ~0x0FULL;
1423         td->td_frame->tf_special.psr =
1424             (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot;
1425         return (0);
1426 }
1427
1428 int
1429 ptrace_single_step(struct thread *td)
1430 {
1431         struct trapframe *tf;
1432
1433         /*
1434          * There's no way to set single stepping when we're leaving the
1435          * kernel through the EPC syscall path. The way we solve this is
1436          * by enabling the lower-privilege trap so that we re-enter the
1437          * kernel as soon as the privilege level changes. See trap.c for
1438          * how we proceed from there.
1439          */
1440         tf = td->td_frame;
1441         if (tf->tf_flags & FRAME_SYSCALL)
1442                 tf->tf_special.psr |= IA64_PSR_LP;
1443         else
1444                 tf->tf_special.psr |= IA64_PSR_SS;
1445         return (0);
1446 }
1447
1448 int
1449 ptrace_clear_single_step(struct thread *td)
1450 {
1451         struct trapframe *tf;
1452
1453         /*
1454          * Clear any and all status bits we may use to implement single
1455          * stepping.
1456          */
1457         tf = td->td_frame;
1458         tf->tf_special.psr &= ~IA64_PSR_SS;
1459         tf->tf_special.psr &= ~IA64_PSR_LP;
1460         tf->tf_special.psr &= ~IA64_PSR_TB;
1461         return (0);
1462 }
1463
1464 int
1465 fill_regs(struct thread *td, struct reg *regs)
1466 {
1467         struct trapframe *tf;
1468
1469         tf = td->td_frame;
1470         regs->r_special = tf->tf_special;
1471         regs->r_scratch = tf->tf_scratch;
1472         save_callee_saved(&regs->r_preserved);
1473         return (0);
1474 }
1475
1476 int
1477 set_regs(struct thread *td, struct reg *regs)
1478 {
1479         struct trapframe *tf;
1480         int error;
1481
1482         tf = td->td_frame;
1483         error = ia64_flush_dirty(td, &tf->tf_special);
1484         if (!error) {
1485                 tf->tf_special = regs->r_special;
1486                 tf->tf_special.bspstore += tf->tf_special.ndirty;
1487                 tf->tf_special.ndirty = 0;
1488                 tf->tf_scratch = regs->r_scratch;
1489                 restore_callee_saved(&regs->r_preserved);
1490         }
1491         return (error);
1492 }
1493
1494 int
1495 fill_dbregs(struct thread *td, struct dbreg *dbregs)
1496 {
1497
1498         return (ENOSYS);
1499 }
1500
1501 int
1502 set_dbregs(struct thread *td, struct dbreg *dbregs)
1503 {
1504
1505         return (ENOSYS);
1506 }
1507
1508 int
1509 fill_fpregs(struct thread *td, struct fpreg *fpregs)
1510 {
1511         struct trapframe *frame = td->td_frame;
1512         struct pcb *pcb = td->td_pcb;
1513
1514         /* Save the high FP registers. */
1515         ia64_highfp_save(td);
1516
1517         fpregs->fpr_scratch = frame->tf_scratch_fp;
1518         save_callee_saved_fp(&fpregs->fpr_preserved);
1519         fpregs->fpr_high = pcb->pcb_high_fp;
1520         return (0);
1521 }
1522
1523 int
1524 set_fpregs(struct thread *td, struct fpreg *fpregs)
1525 {
1526         struct trapframe *frame = td->td_frame;
1527         struct pcb *pcb = td->td_pcb;
1528
1529         /* Throw away the high FP registers (should be redundant). */
1530         ia64_highfp_drop(td);
1531
1532         frame->tf_scratch_fp = fpregs->fpr_scratch;
1533         restore_callee_saved_fp(&fpregs->fpr_preserved);
1534         pcb->pcb_high_fp = fpregs->fpr_high;
1535         return (0);
1536 }
1537
1538 void
1539 ia64_sync_icache(vm_offset_t va, vm_offset_t sz)
1540 {
1541         vm_offset_t lim;
1542
1543         if (!ia64_sync_icache_needed)
1544                 return;
1545
1546         lim = va + sz;
1547         while (va < lim) {
1548                 ia64_fc_i(va);
1549                 va += 32;       /* XXX */
1550         }
1551
1552         ia64_sync_i();
1553         ia64_srlz_i();
1554 }