osfmk/i386/mp_desc.c

   1 /*
   2  * Copyright (c) 2000-2019 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  * @OSF_COPYRIGHT@
  30  */
  31 /*
  32  * Mach Operating System
  33  * Copyright (c) 1991,1990 Carnegie Mellon University
  34  * All Rights Reserved.
  35  *
  36  * Permission to use, copy, modify and distribute this software and its
  37  * documentation is hereby granted, provided that both the copyright
  38  * notice and this permission notice appear in all copies of the
  39  * software, derivative works or modified versions, and any portions
  40  * thereof, and that both notices appear in supporting documentation.
  41  *
  42  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  43  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  44  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  45  *
  46  * Carnegie Mellon requests users of this software to return to
  47  *
  48  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  49  *  School of Computer Science
  50  *  Carnegie Mellon University
  51  *  Pittsburgh PA 15213-3890
  52  *
  53  * any improvements or extensions that they make and grant Carnegie Mellon
  54  * the rights to redistribute these changes.
  55  */
  56
  57 /*
  58  */
  59
  60 #include <kern/cpu_number.h>
  61 #include <kern/kalloc.h>
  62 #include <kern/cpu_data.h>
  63 #include <mach/mach_types.h>
  64 #include <mach/machine.h>
  65 #include <mach/vm_map.h>
  66 #include <mach/machine/vm_param.h>
  67 #include <vm/vm_kern.h>
  68 #include <vm/vm_map.h>
  69
  70 #include <i386/bit_routines.h>
  71 #include <i386/mp_desc.h>
  72 #include <i386/misc_protos.h>
  73 #include <i386/mp.h>
  74 #include <i386/pmap.h>
  75 #include <i386/postcode.h>
  76 #include <i386/pmap_internal.h>
  77 #if CONFIG_MCA
  78 #include <i386/machine_check.h>
  79 #endif
  80
  81 #include <kern/misc_protos.h>
  82
  83 #if MONOTONIC
  84 #include <kern/monotonic.h>
  85 #endif /* MONOTONIC */
  86 #include <san/kasan.h>
  87
  88 #define K_INTR_GATE (ACC_P|ACC_PL_K|ACC_INTR_GATE)
  89 #define U_INTR_GATE (ACC_P|ACC_PL_U|ACC_INTR_GATE)
  90
  91 // Declare macros that will declare the externs
  92 #define TRAP(n, name)           extern void *name ;
  93 #define TRAP_ERR(n, name)       extern void *name ;
  94 #define TRAP_SPC(n, name)       extern void *name ;
  95 #define TRAP_IST1(n, name)      extern void *name ;
  96 #define TRAP_IST2(n, name)      extern void *name ;
  97 #define INTERRUPT(n)            extern void *_intr_ ## n ;
  98 #define USER_TRAP(n, name)      extern void *name ;
  99 #define USER_TRAP_SPC(n, name)  extern void *name ;
 100
 101 // Include the table to declare the externs
 102 #include "../x86_64/idt_table.h"
 103
 104 // Undef the macros, then redefine them so we can declare the table
 105 #undef TRAP
 106 #undef TRAP_ERR
 107 #undef TRAP_SPC
 108 #undef TRAP_IST1
 109 #undef TRAP_IST2
 110 #undef INTERRUPT
 111 #undef USER_TRAP
 112 #undef USER_TRAP_SPC
 113
 114 #define TRAP(n, name)                   \
 115         [n] = {                         \
 116                 (uintptr_t)&name,       \
 117                 KERNEL64_CS,            \
 118                 0,                      \
 119                 K_INTR_GATE,            \
 120                 0                       \
 121         },
 122
 123 #define TRAP_ERR TRAP
 124 #define TRAP_SPC TRAP
 125
 126 #define TRAP_IST1(n, name) \
 127         [n] = {                         \
 128                 (uintptr_t)&name,       \
 129                 KERNEL64_CS,            \
 130                 1,                      \
 131                 K_INTR_GATE,            \
 132                 0                       \
 133         },
 134
 135 #define TRAP_IST2(n, name) \
 136         [n] = {                         \
 137                 (uintptr_t)&name,       \
 138                 KERNEL64_CS,            \
 139                 2,                      \
 140                 K_INTR_GATE,            \
 141                 0                       \
 142         },
 143
 144 #define INTERRUPT(n) \
 145         [n] = {                         \
 146                 (uintptr_t)&_intr_ ## n,\
 147                 KERNEL64_CS,            \
 148                 0,                      \
 149                 K_INTR_GATE,            \
 150                 0                       \
 151         },
 152
 153 #define USER_TRAP(n, name) \
 154         [n] = {                         \
 155                 (uintptr_t)&name,       \
 156                 KERNEL64_CS,            \
 157                 0,                      \
 158                 U_INTR_GATE,            \
 159                 0                       \
 160         },
 161
 162 #define USER_TRAP_SPC USER_TRAP
 163
 164 // Declare the table using the macros we just set up
 165 struct fake_descriptor64 master_idt64[IDTSZ]
 166 __attribute__ ((section("__HIB,__desc")))
 167 __attribute__ ((aligned(PAGE_SIZE))) = {
 168 #include "../x86_64/idt_table.h"
 169 };
 170
 171 /*
 172  * First cpu`s interrupt stack.
 173  */
 174 extern uint32_t         low_intstack[];         /* bottom */
 175 extern uint32_t         low_eintstack[];        /* top */
 176
 177 /*
 178  * Per-cpu data area pointers.
 179  */
 180 cpu_data_t cpshadows[MAX_CPUS] __attribute__((aligned(64))) __attribute__((section("__HIB, __desc")));
 181 cpu_data_t scdatas[MAX_CPUS] __attribute__((aligned(64))) = {
 182         [0].cpu_this = &scdatas[0],
 183         [0].cpu_nanotime = &pal_rtc_nanotime_info,
 184         [0].cpu_int_stack_top = (vm_offset_t) low_eintstack,
 185         [0].cd_shadow = &cpshadows[0]
 186 };
 187 cpu_data_t *cpu_data_master = &scdatas[0];
 188
 189 cpu_data_t      *cpu_data_ptr[MAX_CPUS] = {[0] = &scdatas[0] };
 190
 191 decl_simple_lock_data(, ncpus_lock);     /* protects real_ncpus */
 192 unsigned int    real_ncpus = 1;
 193 unsigned int    max_ncpus = MAX_CPUS;
 194
 195 extern void hi64_sysenter(void);
 196 extern void hi64_syscall(void);
 197
 198 typedef struct {
 199         struct real_descriptor pcldts[LDTSZ];
 200 } cldt_t;
 201
 202 cpu_desc_table64_t scdtables[MAX_CPUS] __attribute__((aligned(64))) __attribute__((section("__HIB, __desc")));
 203 cpu_fault_stack_t scfstks[MAX_CPUS] __attribute__((aligned(64))) __attribute__((section("__HIB, __desc")));
 204
 205 cldt_t *dyn_ldts;
 206
 207 /*
 208  * Multiprocessor i386/i486 systems use a separate copy of the
 209  * GDT, IDT, LDT, and kernel TSS per processor.  The first three
 210  * are separate to avoid lock contention: the i386 uses locked
 211  * memory cycles to access the descriptor tables.  The TSS is
 212  * separate since each processor needs its own kernel stack,
 213  * and since using a TSS marks it busy.
 214  */
 215
 216 /*
 217  * Allocate and initialize the per-processor descriptor tables.
 218  */
 219
 220 /*
 221  * This is the expanded, 64-bit variant of the kernel LDT descriptor.
 222  * When switching to 64-bit mode this replaces KERNEL_LDT entry
 223  * and the following empty slot. This enables the LDT to be referenced
 224  * in the uber-space remapping window on the kernel.
 225  */
 226 struct fake_descriptor64 kernel_ldt_desc64 = {
 227         .offset64 = 0,
 228         .lim_or_seg = LDTSZ_MIN * sizeof(struct fake_descriptor) - 1,
 229         .size_or_IST = 0,
 230         .access = ACC_P | ACC_PL_K | ACC_LDT,
 231         .reserved = 0
 232 };
 233
 234 /*
 235  * This is the expanded, 64-bit variant of the kernel TSS descriptor.
 236  * It is follows pattern of the KERNEL_LDT.
 237  */
 238 struct fake_descriptor64 kernel_tss_desc64 = {
 239         .offset64 = 0,
 240         .lim_or_seg = sizeof(struct x86_64_tss) - 1,
 241         .size_or_IST = 0,
 242         .access = ACC_P | ACC_PL_K | ACC_TSS,
 243         .reserved = 0
 244 };
 245
 246 /*
 247  * Convert a descriptor from fake to real format.
 248  *
 249  * Fake descriptor format:
 250  *      bytes 0..3              base 31..0
 251  *      bytes 4..5              limit 15..0
 252  *      byte  6                 access byte 2 | limit 19..16
 253  *      byte  7                 access byte 1
 254  *
 255  * Real descriptor format:
 256  *      bytes 0..1              limit 15..0
 257  *      bytes 2..3              base 15..0
 258  *      byte  4                 base 23..16
 259  *      byte  5                 access byte 1
 260  *      byte  6                 access byte 2 | limit 19..16
 261  *      byte  7                 base 31..24
 262  *
 263  * Fake gate format:
 264  *      bytes 0..3              offset
 265  *      bytes 4..5              selector
 266  *      byte  6                 word count << 4 (to match fake descriptor)
 267  *      byte  7                 access byte 1
 268  *
 269  * Real gate format:
 270  *      bytes 0..1              offset 15..0
 271  *      bytes 2..3              selector
 272  *      byte  4                 word count
 273  *      byte  5                 access byte 1
 274  *      bytes 6..7              offset 31..16
 275  */
 276 void
 277 fix_desc(void *d, int num_desc)
 278 {
 279         uint8_t *desc = (uint8_t*) d;
 280
 281         do {
 282                 if ((desc[7] & 0x14) == 0x04) { /* gate */
 283                         uint32_t offset;
 284                         uint16_t selector;
 285                         uint8_t wordcount;
 286                         uint8_t acc;
 287
 288                         offset = *((uint32_t*)(desc));
 289                         selector = *((uint32_t*)(desc + 4));
 290                         wordcount = desc[6] >> 4;
 291                         acc = desc[7];
 292
 293                         *((uint16_t*)desc) = offset & 0xFFFF;
 294                         *((uint16_t*)(desc + 2)) = selector;
 295                         desc[4] = wordcount;
 296                         desc[5] = acc;
 297                         *((uint16_t*)(desc + 6)) = offset >> 16;
 298                 } else { /* descriptor */
 299                         uint32_t base;
 300                         uint16_t limit;
 301                         uint8_t acc1, acc2;
 302
 303                         base = *((uint32_t*)(desc));
 304                         limit = *((uint16_t*)(desc + 4));
 305                         acc2 = desc[6];
 306                         acc1 = desc[7];
 307
 308                         *((uint16_t*)(desc)) = limit;
 309                         *((uint16_t*)(desc + 2)) = base & 0xFFFF;
 310                         desc[4] = (base >> 16) & 0xFF;
 311                         desc[5] = acc1;
 312                         desc[6] = acc2;
 313                         desc[7] = base >> 24;
 314                 }
 315                 desc += 8;
 316         } while (--num_desc);
 317 }
 318
 319 void
 320 fix_desc64(void *descp, int count)
 321 {
 322         struct fake_descriptor64        *fakep;
 323         union {
 324                 struct real_gate64              gate;
 325                 struct real_descriptor64        desc;
 326         }                               real;
 327         int                             i;
 328
 329         fakep = (struct fake_descriptor64 *) descp;
 330
 331         for (i = 0; i < count; i++, fakep++) {
 332                 /*
 333                  * Construct the real decriptor locally.
 334                  */
 335
 336                 bzero((void *) &real, sizeof(real));
 337
 338                 switch (fakep->access & ACC_TYPE) {
 339                 case 0:
 340                         break;
 341                 case ACC_CALL_GATE:
 342                 case ACC_INTR_GATE:
 343                 case ACC_TRAP_GATE:
 344                         real.gate.offset_low16 = (uint16_t)(fakep->offset64 & 0xFFFF);
 345                         real.gate.selector16 = fakep->lim_or_seg & 0xFFFF;
 346                         real.gate.IST = fakep->size_or_IST & 0x7;
 347                         real.gate.access8 = fakep->access;
 348                         real.gate.offset_high16 = (uint16_t)((fakep->offset64 >> 16) & 0xFFFF);
 349                         real.gate.offset_top32 = (uint32_t)(fakep->offset64 >> 32);
 350                         break;
 351                 default:        /* Otherwise */
 352                         real.desc.limit_low16 = fakep->lim_or_seg & 0xFFFF;
 353                         real.desc.base_low16 = (uint16_t)(fakep->offset64 & 0xFFFF);
 354                         real.desc.base_med8 = (uint8_t)((fakep->offset64 >> 16) & 0xFF);
 355                         real.desc.access8 = fakep->access;
 356                         real.desc.limit_high4 = (fakep->lim_or_seg >> 16) & 0xFF;
 357                         real.desc.granularity4 = fakep->size_or_IST;
 358                         real.desc.base_high8 = (uint8_t)((fakep->offset64 >> 24) & 0xFF);
 359                         real.desc.base_top32 = (uint32_t)(fakep->offset64 >> 32);
 360                 }
 361
 362                 /*
 363                  * Now copy back over the fake structure.
 364                  */
 365                 bcopy((void *) &real, (void *) fakep, sizeof(real));
 366         }
 367 }
 368
 369 extern unsigned mldtsz;
 370 void
 371 cpu_desc_init(cpu_data_t *cdp)
 372 {
 373         cpu_desc_index_t        *cdi = &cdp->cpu_desc_index;
 374
 375         if (cdp == cpu_data_master) {
 376                 /*
 377                  * Populate the double-mapped 'u' and base 'b' fields in the
 378                  * KTSS with I/G/LDT and sysenter stack data.
 379                  */
 380                 cdi->cdi_ktssu = (void *)DBLMAP(&master_ktss64);
 381                 cdi->cdi_ktssb = (void *)&master_ktss64;
 382                 cdi->cdi_sstku = (vm_offset_t) DBLMAP(&master_sstk.top);
 383                 cdi->cdi_sstkb = (vm_offset_t) &master_sstk.top;
 384
 385                 cdi->cdi_gdtu.ptr = (void *)DBLMAP((uintptr_t) &master_gdt);
 386                 cdi->cdi_gdtb.ptr = (void *)&master_gdt;
 387                 cdi->cdi_idtu.ptr  = (void *)DBLMAP((uintptr_t) &master_idt64);
 388                 cdi->cdi_idtb.ptr  = (void *)((uintptr_t) &master_idt64);
 389                 cdi->cdi_ldtu  = (struct real_descriptor *)DBLMAP((uintptr_t)&master_ldt[0]);
 390                 cdi->cdi_ldtb  = &master_ldt[0];
 391
 392                 /* Replace the expanded LDTs and TSS slots in the GDT */
 393                 kernel_ldt_desc64.offset64 = (uintptr_t) cdi->cdi_ldtu;
 394                 *(struct fake_descriptor64 *) &master_gdt[sel_idx(KERNEL_LDT)] =
 395                     kernel_ldt_desc64;
 396                 *(struct fake_descriptor64 *) &master_gdt[sel_idx(USER_LDT)] =
 397                     kernel_ldt_desc64;
 398                 kernel_tss_desc64.offset64 = (uintptr_t) DBLMAP(&master_ktss64);
 399                 *(struct fake_descriptor64 *) &master_gdt[sel_idx(KERNEL_TSS)] =
 400                     kernel_tss_desc64;
 401
 402                 /* Fix up the expanded descriptors for 64-bit. */
 403                 fix_desc64((void *) &master_idt64, IDTSZ);
 404                 fix_desc64((void *) &master_gdt[sel_idx(KERNEL_LDT)], 1);
 405                 fix_desc64((void *) &master_gdt[sel_idx(USER_LDT)], 1);
 406                 fix_desc64((void *) &master_gdt[sel_idx(KERNEL_TSS)], 1);
 407
 408                 /*
 409                  * Set the NMI/fault stacks as IST2/IST1 in the 64-bit TSS
 410                  */
 411                 master_ktss64.ist2 = (uintptr_t) low_eintstack;
 412                 master_ktss64.ist1 = (uintptr_t) low_eintstack - sizeof(x86_64_intr_stack_frame_t);
 413         } else if (cdi->cdi_ktssu == NULL) {    /* Skipping re-init on wake */
 414                 cpu_desc_table64_t      *cdt = (cpu_desc_table64_t *) cdp->cpu_desc_tablep;
 415
 416                 cdi->cdi_idtu.ptr  = (void *)DBLMAP((uintptr_t) &master_idt64);
 417
 418                 cdi->cdi_ktssu = (void *)DBLMAP(&cdt->ktss);
 419                 cdi->cdi_ktssb = (void *)(&cdt->ktss);
 420                 cdi->cdi_sstku = (vm_offset_t)DBLMAP(&cdt->sstk.top);
 421                 cdi->cdi_sstkb = (vm_offset_t)(&cdt->sstk.top);
 422                 cdi->cdi_ldtu  = (void *)LDTALIAS(cdp->cpu_ldtp);
 423                 cdi->cdi_ldtb  = (void *)(cdp->cpu_ldtp);
 424
 425                 /*
 426                  * Copy the tables
 427                  */
 428                 bcopy((char *)master_gdt, (char *)cdt->gdt, sizeof(master_gdt));
 429                 bcopy((char *)master_ldt, (char *)cdp->cpu_ldtp, mldtsz);
 430                 bcopy((char *)&master_ktss64, (char *)&cdt->ktss, sizeof(struct x86_64_tss));
 431                 cdi->cdi_gdtu.ptr  = (void *)DBLMAP(cdt->gdt);
 432                 cdi->cdi_gdtb.ptr  = (void *)(cdt->gdt);
 433                 /*
 434                  * Fix up the entries in the GDT to point to
 435                  * this LDT and this TSS.
 436                  * Note reuse of global 'kernel_ldt_desc64, which is not
 437                  * concurrency-safe. Higher level synchronization is expected
 438                  */
 439                 kernel_ldt_desc64.offset64 = (uintptr_t) cdi->cdi_ldtu;
 440                 *(struct fake_descriptor64 *) &cdt->gdt[sel_idx(KERNEL_LDT)] =
 441                     kernel_ldt_desc64;
 442                 fix_desc64(&cdt->gdt[sel_idx(KERNEL_LDT)], 1);
 443
 444                 kernel_ldt_desc64.offset64 = (uintptr_t) cdi->cdi_ldtu;
 445                 *(struct fake_descriptor64 *) &cdt->gdt[sel_idx(USER_LDT)] =
 446                     kernel_ldt_desc64;
 447                 fix_desc64(&cdt->gdt[sel_idx(USER_LDT)], 1);
 448
 449                 kernel_tss_desc64.offset64 = (uintptr_t) cdi->cdi_ktssu;
 450                 *(struct fake_descriptor64 *) &cdt->gdt[sel_idx(KERNEL_TSS)] =
 451                     kernel_tss_desc64;
 452                 fix_desc64(&cdt->gdt[sel_idx(KERNEL_TSS)], 1);
 453
 454                 /* Set (zeroed) fault stack as IST1, NMI intr stack IST2 */
 455                 uint8_t *cfstk = &scfstks[cdp->cpu_number].fstk[0];
 456                 cdt->fstkp = cfstk;
 457                 bzero((void *) cfstk, FSTK_SZ);
 458                 cdt->ktss.ist2 = DBLMAP((uint64_t)cdt->fstkp + FSTK_SZ);
 459                 cdt->ktss.ist1 = cdt->ktss.ist2 - sizeof(x86_64_intr_stack_frame_t);
 460         }
 461
 462         /* Require that the top of the sysenter stack is 16-byte aligned */
 463         if ((cdi->cdi_sstku % 16) != 0) {
 464                 panic("cpu_desc_init() sysenter stack not 16-byte aligned");
 465         }
 466 }
 467 void
 468 cpu_desc_load(cpu_data_t *cdp)
 469 {
 470         cpu_desc_index_t        *cdi = &cdp->cpu_desc_index;
 471
 472         postcode(CPU_DESC_LOAD_ENTRY);
 473
 474         /* Stuff the kernel per-cpu data area address into the MSRs */
 475         postcode(CPU_DESC_LOAD_GS_BASE);
 476         wrmsr64(MSR_IA32_GS_BASE, (uintptr_t) cdp);
 477         postcode(CPU_DESC_LOAD_KERNEL_GS_BASE);
 478         wrmsr64(MSR_IA32_KERNEL_GS_BASE, (uintptr_t) cdp);
 479
 480         /*
 481          * Ensure the TSS segment's busy bit is clear. This is required
 482          * for the case of reloading descriptors at wake to avoid
 483          * their complete re-initialization.
 484          */
 485         gdt_desc_p(KERNEL_TSS)->access &= ~ACC_TSS_BUSY;
 486
 487         /* Load the GDT, LDT, IDT and TSS */
 488         cdi->cdi_gdtb.size = sizeof(struct real_descriptor) * GDTSZ - 1;
 489         cdi->cdi_gdtu.size = cdi->cdi_gdtb.size;
 490         cdi->cdi_idtb.size = 0x1000 + cdp->cpu_number;
 491         cdi->cdi_idtu.size = cdi->cdi_idtb.size;
 492
 493         postcode(CPU_DESC_LOAD_GDT);
 494         lgdt((uintptr_t *) &cdi->cdi_gdtu);
 495         postcode(CPU_DESC_LOAD_IDT);
 496         lidt((uintptr_t *) &cdi->cdi_idtu);
 497         postcode(CPU_DESC_LOAD_LDT);
 498         lldt(KERNEL_LDT);
 499         postcode(CPU_DESC_LOAD_TSS);
 500         set_tr(KERNEL_TSS);
 501
 502         postcode(CPU_DESC_LOAD_EXIT);
 503 }
 504
 505 /*
 506  * Set MSRs for sysenter/sysexit and syscall/sysret for 64-bit.
 507  */
 508 void
 509 cpu_syscall_init(cpu_data_t *cdp)
 510 {
 511 #pragma unused(cdp)
 512
 513         wrmsr64(MSR_IA32_SYSENTER_CS, SYSENTER_CS);
 514         wrmsr64(MSR_IA32_SYSENTER_EIP, DBLMAP((uintptr_t) hi64_sysenter));
 515         wrmsr64(MSR_IA32_SYSENTER_ESP, current_cpu_datap()->cpu_desc_index.cdi_sstku);
 516         /* Enable syscall/sysret */
 517         wrmsr64(MSR_IA32_EFER, rdmsr64(MSR_IA32_EFER) | MSR_IA32_EFER_SCE);
 518
 519         /*
 520          * MSRs for 64-bit syscall/sysret
 521          * Note USER_CS because sysret uses this + 16 when returning to
 522          * 64-bit code.
 523          */
 524         wrmsr64(MSR_IA32_LSTAR, DBLMAP((uintptr_t) hi64_syscall));
 525         wrmsr64(MSR_IA32_STAR, (((uint64_t)USER_CS) << 48) | (((uint64_t)KERNEL64_CS) << 32));
 526         /*
 527          * Emulate eflags cleared by sysenter but note that
 528          * we also clear the trace trap to avoid the complications
 529          * of single-stepping into a syscall. The nested task bit
 530          * is also cleared to avoid a spurious "task switch"
 531          * should we choose to return via an IRET.
 532          */
 533         wrmsr64(MSR_IA32_FMASK, EFL_DF | EFL_IF | EFL_TF | EFL_NT);
 534 }
 535 extern vm_offset_t dyn_dblmap(vm_offset_t, vm_offset_t);
 536 uint64_t ldt_alias_offset;
 537
 538 cpu_data_t *
 539 cpu_data_alloc(boolean_t is_boot_cpu)
 540 {
 541         int             ret;
 542         cpu_data_t      *cdp;
 543
 544         if (is_boot_cpu) {
 545                 assert(real_ncpus == 1);
 546                 cdp = cpu_datap(0);
 547                 if (cdp->cpu_processor == NULL) {
 548                         simple_lock_init(&ncpus_lock, 0);
 549                         cdp->cpu_processor = cpu_processor_alloc(TRUE);
 550 #if NCOPY_WINDOWS > 0
 551                         cdp->cpu_pmap = pmap_cpu_alloc(TRUE);
 552 #endif
 553                 }
 554                 return cdp;
 555         }
 556
 557         boolean_t do_ldt_alloc = FALSE;
 558         simple_lock(&ncpus_lock, LCK_GRP_NULL);
 559         int cnum = real_ncpus;
 560         real_ncpus++;
 561         if (dyn_ldts == NULL) {
 562                 do_ldt_alloc = TRUE;
 563         }
 564         simple_unlock(&ncpus_lock);
 565
 566         /*
 567          * Allocate per-cpu data:
 568          */
 569
 570         cdp = &scdatas[cnum];
 571         bzero((void*) cdp, sizeof(cpu_data_t));
 572         cdp->cpu_this = cdp;
 573         cdp->cpu_number = cnum;
 574         cdp->cd_shadow = &cpshadows[cnum];
 575         /*
 576          * Allocate interrupt stack:
 577          */
 578         ret = kmem_alloc(kernel_map,
 579             (vm_offset_t *) &cdp->cpu_int_stack_top,
 580             INTSTACK_SIZE, VM_KERN_MEMORY_CPU);
 581         if (ret != KERN_SUCCESS) {
 582                 panic("cpu_data_alloc() int stack failed, ret=%d\n", ret);
 583         }
 584         bzero((void*) cdp->cpu_int_stack_top, INTSTACK_SIZE);
 585         cdp->cpu_int_stack_top += INTSTACK_SIZE;
 586
 587         /*
 588          * Allocate descriptor table:
 589          */
 590
 591         cdp->cpu_desc_tablep = (struct cpu_desc_table *) &scdtables[cnum];
 592         /*
 593          * Allocate LDT
 594          */
 595         if (do_ldt_alloc) {
 596                 boolean_t do_ldt_free = FALSE;
 597                 vm_offset_t sldtoffset = 0;
 598                 /*
 599                  * Allocate LDT
 600                  */
 601                 vm_offset_t ldtalloc = 0, ldtallocsz = round_page_64(MAX_CPUS * sizeof(struct real_descriptor) * LDTSZ);
 602                 ret = kmem_alloc(kernel_map, (vm_offset_t *) &ldtalloc, ldtallocsz, VM_KERN_MEMORY_CPU);
 603                 if (ret != KERN_SUCCESS) {
 604                         panic("cpu_data_alloc() ldt failed, kmem_alloc=%d\n", ret);
 605                 }
 606
 607                 simple_lock(&ncpus_lock, LCK_GRP_NULL);
 608                 if (dyn_ldts == NULL) {
 609                         dyn_ldts = (cldt_t *)ldtalloc;
 610                 } else {
 611                         do_ldt_free = TRUE;
 612                 }
 613                 simple_unlock(&ncpus_lock);
 614
 615                 if (do_ldt_free) {
 616                         kmem_free(kernel_map, ldtalloc, ldtallocsz);
 617                 } else {
 618                         /* CPU registration and startup are expected to execute
 619                          * serially, as invoked by the platform driver.
 620                          * Create trampoline alias of LDT region.
 621                          */
 622                         sldtoffset = dyn_dblmap(ldtalloc, ldtallocsz);
 623                         ldt_alias_offset = sldtoffset;
 624                 }
 625         }
 626         cdp->cpu_ldtp = &dyn_ldts[cnum].pcldts[0];
 627
 628 #if CONFIG_MCA
 629         /* Machine-check shadow register allocation. */
 630         mca_cpu_alloc(cdp);
 631 #endif
 632
 633         /*
 634          * Before this cpu has been assigned a real thread context,
 635          * we give it a fake, unique, non-zero thread id which the locking
 636          * primitives use as their lock value.
 637          * Note that this does not apply to the boot processor, cpu 0, which
 638          * transitions to a thread context well before other processors are
 639          * started.
 640          */
 641         cdp->cpu_active_thread = (thread_t) (uintptr_t) cdp->cpu_number;
 642         cdp->cpu_NMI_acknowledged = TRUE;
 643         cdp->cpu_nanotime = &pal_rtc_nanotime_info;
 644
 645         kprintf("cpu_data_alloc(%d) %p desc_table: %p "
 646             "ldt: %p "
 647             "int_stack: 0x%lx-0x%lx\n",
 648             cdp->cpu_number, cdp, cdp->cpu_desc_tablep, cdp->cpu_ldtp,
 649             (long)(cdp->cpu_int_stack_top - INTSTACK_SIZE), (long)(cdp->cpu_int_stack_top));
 650         cpu_data_ptr[cnum] = cdp;
 651
 652         return cdp;
 653 }
 654
 655 boolean_t
 656 valid_user_data_selector(uint16_t selector)
 657 {
 658         sel_t       sel = selector_to_sel(selector);
 659
 660         if (selector == 0) {
 661                 return TRUE;
 662         }
 663
 664         if (sel.ti == SEL_LDT) {
 665                 return TRUE;
 666         } else if (sel.index < GDTSZ) {
 667                 if ((gdt_desc_p(selector)->access & ACC_PL_U) == ACC_PL_U) {
 668                         return TRUE;
 669                 }
 670         }
 671         return FALSE;
 672 }
 673
 674 boolean_t
 675 valid_user_code_selector(uint16_t selector)
 676 {
 677         sel_t       sel = selector_to_sel(selector);
 678
 679         if (selector == 0) {
 680                 return FALSE;
 681         }
 682
 683         if (sel.ti == SEL_LDT) {
 684                 if (sel.rpl == USER_PRIV) {
 685                         return TRUE;
 686                 }
 687         } else if (sel.index < GDTSZ && sel.rpl == USER_PRIV) {
 688                 if ((gdt_desc_p(selector)->access & ACC_PL_U) == ACC_PL_U) {
 689                         return TRUE;
 690                 }
 691                 /* Explicitly validate the system code selectors
 692                  * even if not instantaneously privileged,
 693                  * since they are dynamically re-privileged
 694                  * at context switch
 695                  */
 696                 if ((selector == USER_CS) || (selector == USER64_CS)) {
 697                         return TRUE;
 698                 }
 699         }
 700
 701         return FALSE;
 702 }
 703
 704 boolean_t
 705 valid_user_stack_selector(uint16_t selector)
 706 {
 707         sel_t       sel = selector_to_sel(selector);
 708
 709         if (selector == 0) {
 710                 return FALSE;
 711         }
 712
 713         if (sel.ti == SEL_LDT) {
 714                 if (sel.rpl == USER_PRIV) {
 715                         return TRUE;
 716                 }
 717         } else if (sel.index < GDTSZ && sel.rpl == USER_PRIV) {
 718                 if ((gdt_desc_p(selector)->access & ACC_PL_U) == ACC_PL_U) {
 719                         return TRUE;
 720                 }
 721         }
 722
 723         return FALSE;
 724 }
 725
 726 boolean_t
 727 valid_user_segment_selectors(uint16_t cs,
 728     uint16_t ss,
 729     uint16_t ds,
 730     uint16_t es,
 731     uint16_t fs,
 732     uint16_t gs)
 733 {
 734         return valid_user_code_selector(cs) &&
 735                valid_user_stack_selector(ss) &&
 736                valid_user_data_selector(ds) &&
 737                valid_user_data_selector(es) &&
 738                valid_user_data_selector(fs) &&
 739                valid_user_data_selector(gs);
 740 }
 741
 742 #if NCOPY_WINDOWS > 0
 743
 744 static vm_offset_t user_window_base = 0;
 745
 746 void
 747 cpu_userwindow_init(int cpu)
 748 {
 749         cpu_data_t              *cdp = cpu_data_ptr[cpu];
 750         vm_offset_t             user_window;
 751         vm_offset_t             vaddr;
 752         int                     num_cpus;
 753
 754         num_cpus = ml_get_max_cpus();
 755
 756         if (cpu >= num_cpus) {
 757                 panic("cpu_userwindow_init: cpu > num_cpus");
 758         }
 759
 760         if (user_window_base == 0) {
 761                 if (vm_allocate(kernel_map, &vaddr,
 762                     (NBPDE * NCOPY_WINDOWS * num_cpus) + NBPDE,
 763                     VM_FLAGS_ANYWHERE | VM_MAKE_TAG(VM_KERN_MEMORY_CPU)) != KERN_SUCCESS) {
 764                         panic("cpu_userwindow_init: "
 765                             "couldn't allocate user map window");
 766                 }
 767
 768                 /*
 769                  * window must start on a page table boundary
 770                  * in the virtual address space
 771                  */
 772                 user_window_base = (vaddr + (NBPDE - 1)) & ~(NBPDE - 1);
 773
 774                 /*
 775                  * get rid of any allocation leading up to our
 776                  * starting boundary
 777                  */
 778                 vm_deallocate(kernel_map, vaddr, user_window_base - vaddr);
 779
 780                 /*
 781                  * get rid of tail that we don't need
 782                  */
 783                 user_window = user_window_base +
 784                     (NBPDE * NCOPY_WINDOWS * num_cpus);
 785
 786                 vm_deallocate(kernel_map, user_window,
 787                     (vaddr +
 788                     ((NBPDE * NCOPY_WINDOWS * num_cpus) + NBPDE)) -
 789                     user_window);
 790         }
 791
 792         user_window = user_window_base + (cpu * NCOPY_WINDOWS * NBPDE);
 793
 794         cdp->cpu_copywindow_base = user_window;
 795         /*
 796          * Abuse this pdp entry, the pdp now actually points to
 797          * an array of copy windows addresses.
 798          */
 799         cdp->cpu_copywindow_pdp  = pmap_pde(kernel_pmap, user_window);
 800 }
 801
 802 void
 803 cpu_physwindow_init(int cpu)
 804 {
 805         cpu_data_t              *cdp = cpu_data_ptr[cpu];
 806         vm_offset_t             phys_window = cdp->cpu_physwindow_base;
 807
 808         if (phys_window == 0) {
 809                 if (vm_allocate(kernel_map, &phys_window,
 810                     PAGE_SIZE, VM_FLAGS_ANYWHERE | VM_MAKE_TAG(VM_KERN_MEMORY_CPU))
 811                     != KERN_SUCCESS) {
 812                         panic("cpu_physwindow_init: "
 813                             "couldn't allocate phys map window");
 814                 }
 815
 816                 /*
 817                  * make sure the page that encompasses the
 818                  * pte pointer we're interested in actually
 819                  * exists in the page table
 820                  */
 821                 pmap_expand(kernel_pmap, phys_window, PMAP_EXPAND_OPTIONS_NONE);
 822
 823                 cdp->cpu_physwindow_base = phys_window;
 824                 cdp->cpu_physwindow_ptep = vtopte(phys_window);
 825         }
 826 }
 827 #endif /* NCOPY_WINDOWS > 0 */
 828
 829 /*
 830  * Allocate a new interrupt stack for the boot processor from the
 831  * heap rather than continue to use the statically allocated space.
 832  * Also switch to a dynamically allocated cpu data area.
 833  */
 834 void
 835 cpu_data_realloc(void)
 836 {
 837         int             ret;
 838         vm_offset_t     istk;
 839         cpu_data_t      *cdp;
 840         boolean_t       istate;
 841
 842         ret = kmem_alloc(kernel_map, &istk, INTSTACK_SIZE, VM_KERN_MEMORY_CPU);
 843         if (ret != KERN_SUCCESS) {
 844                 panic("cpu_data_realloc() stack alloc, ret=%d\n", ret);
 845         }
 846         bzero((void*) istk, INTSTACK_SIZE);
 847         istk += INTSTACK_SIZE;
 848
 849         cdp = &scdatas[0];
 850
 851         /* Copy old contents into new area and make fix-ups */
 852         assert(cpu_number() == 0);
 853         bcopy((void *) cpu_data_ptr[0], (void*) cdp, sizeof(cpu_data_t));
 854         cdp->cpu_this = cdp;
 855         cdp->cpu_int_stack_top = istk;
 856         timer_call_queue_init(&cdp->rtclock_timer.queue);
 857         cdp->cpu_desc_tablep = (struct cpu_desc_table *) &scdtables[0];
 858         cpu_desc_table64_t      *cdt = (cpu_desc_table64_t *) cdp->cpu_desc_tablep;
 859
 860         uint8_t *cfstk = &scfstks[cdp->cpu_number].fstk[0];
 861         cdt->fstkp = cfstk;
 862         cfstk += FSTK_SZ;
 863
 864         /*
 865          * With interrupts disabled commmit the new areas.
 866          */
 867         istate = ml_set_interrupts_enabled(FALSE);
 868         cpu_data_ptr[0] = cdp;
 869         master_ktss64.ist2 = DBLMAP((uintptr_t) cfstk);
 870         master_ktss64.ist1 = DBLMAP((uintptr_t) cfstk - sizeof(x86_64_intr_stack_frame_t));
 871         wrmsr64(MSR_IA32_GS_BASE, (uintptr_t) cdp);
 872         wrmsr64(MSR_IA32_KERNEL_GS_BASE, (uintptr_t) cdp);
 873         (void) ml_set_interrupts_enabled(istate);
 874
 875         kprintf("Reallocated master cpu data: %p,"
 876             " interrupt stack: %p, fault stack: %p\n",
 877             (void *) cdp, (void *) istk, (void *) cfstk);
 878 }