1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2012 DEY Storage Systems, Inc.  All rights reserved.
  25  * Copyright 2017 Nexenta Systems, Inc.
  26  * Copyright 2015 Joyent, Inc.
  27  * Copyright (c) 2015 by Delphix. All rights reserved.
  28  */
  29 
  30 /*
  31  * Copyright (c) 2010, Intel Corporation.
  32  * All rights reserved.
  33  */
  34 
  35 #include <sys/types.h>
  36 #include <sys/t_lock.h>
  37 #include <sys/param.h>
  38 #include <sys/sysmacros.h>
  39 #include <sys/signal.h>
  40 #include <sys/systm.h>
  41 #include <sys/user.h>
  42 #include <sys/mman.h>
  43 #include <sys/vm.h>
  44 #include <sys/conf.h>
  45 #include <sys/avintr.h>
  46 #include <sys/autoconf.h>
  47 #include <sys/disp.h>
  48 #include <sys/class.h>
  49 #include <sys/bitmap.h>
  50 
  51 #include <sys/privregs.h>
  52 
  53 #include <sys/proc.h>
  54 #include <sys/buf.h>
  55 #include <sys/kmem.h>
  56 #include <sys/mem.h>
  57 #include <sys/kstat.h>
  58 
  59 #include <sys/reboot.h>
  60 
  61 #include <sys/cred.h>
  62 #include <sys/vnode.h>
  63 #include <sys/file.h>
  64 
  65 #include <sys/procfs.h>
  66 
  67 #include <sys/vfs.h>
  68 #include <sys/cmn_err.h>
  69 #include <sys/utsname.h>
  70 #include <sys/debug.h>
  71 #include <sys/kdi.h>
  72 
  73 #include <sys/dumphdr.h>
  74 #include <sys/bootconf.h>
  75 #include <sys/memlist_plat.h>
  76 #include <sys/varargs.h>
  77 #include <sys/promif.h>
  78 #include <sys/modctl.h>
  79 
  80 #include <sys/sunddi.h>
  81 #include <sys/sunndi.h>
  82 #include <sys/ndi_impldefs.h>
  83 #include <sys/ddidmareq.h>
  84 #include <sys/psw.h>
  85 #include <sys/regset.h>
  86 #include <sys/clock.h>
  87 #include <sys/pte.h>
  88 #include <sys/tss.h>
  89 #include <sys/stack.h>
  90 #include <sys/trap.h>
  91 #include <sys/fp.h>
  92 #include <vm/kboot_mmu.h>
  93 #include <vm/anon.h>
  94 #include <vm/as.h>
  95 #include <vm/page.h>
  96 #include <vm/seg.h>
  97 #include <vm/seg_dev.h>
  98 #include <vm/seg_kmem.h>
  99 #include <vm/seg_kpm.h>
 100 #include <vm/seg_map.h>
 101 #include <vm/seg_vn.h>
 102 #include <vm/seg_kp.h>
 103 #include <sys/memnode.h>
 104 #include <vm/vm_dep.h>
 105 #include <sys/thread.h>
 106 #include <sys/sysconf.h>
 107 #include <sys/vm_machparam.h>
 108 #include <sys/archsystm.h>
 109 #include <sys/machsystm.h>
 110 #include <vm/hat.h>
 111 #include <vm/hat_i86.h>
 112 #include <sys/pmem.h>
 113 #include <sys/smp_impldefs.h>
 114 #include <sys/x86_archext.h>
 115 #include <sys/cpuvar.h>
 116 #include <sys/segments.h>
 117 #include <sys/clconf.h>
 118 #include <sys/kobj.h>
 119 #include <sys/kobj_lex.h>
 120 #include <sys/cpc_impl.h>
 121 #include <sys/cpu_module.h>
 122 #include <sys/smbios.h>
 123 #include <sys/debug_info.h>
 124 #include <sys/bootinfo.h>
 125 #include <sys/ddi_periodic.h>
 126 #include <sys/systeminfo.h>
 127 #include <sys/multiboot.h>
 128 #include <sys/ramdisk.h>
 129 #include <sys/framebuffer.h>
 130 
 131 #ifdef  __xpv
 132 
 133 #include <sys/hypervisor.h>
 134 #include <sys/xen_mmu.h>
 135 #include <sys/evtchn_impl.h>
 136 #include <sys/gnttab.h>
 137 #include <sys/xpv_panic.h>
 138 #include <xen/sys/xenbus_comms.h>
 139 #include <xen/public/physdev.h>
 140 
 141 extern void xen_late_startup(void);
 142 
 143 struct xen_evt_data cpu0_evt_data;
 144 
 145 #else   /* __xpv */
 146 #include <sys/memlist_impl.h>
 147 
 148 extern void mem_config_init(void);
 149 #endif /* __xpv */
 150 
 151 extern void progressbar_init(void);
 152 extern void brand_init(void);
 153 extern void pcf_init(void);
 154 extern void pg_init(void);
 155 extern void ssp_init(void);
 156 
 157 extern int size_pse_array(pgcnt_t, int);
 158 
 159 #if defined(_SOFT_HOSTID)
 160 
 161 #include <sys/rtc.h>
 162 
 163 static int32_t set_soft_hostid(void);
 164 static char hostid_file[] = "/etc/hostid";
 165 
 166 #endif
 167 
 168 void *gfx_devinfo_list;
 169 
 170 #if defined(__amd64) && !defined(__xpv)
 171 extern void immu_startup(void);
 172 #endif
 173 
 174 /*
 175  * XXX make declaration below "static" when drivers no longer use this
 176  * interface.
 177  */
 178 extern caddr_t p0_va;   /* Virtual address for accessing physical page 0 */
 179 
 180 /*
 181  * segkp
 182  */
 183 extern int segkp_fromheap;
 184 
 185 static void kvm_init(void);
 186 static void startup_init(void);
 187 static void startup_memlist(void);
 188 static void startup_kmem(void);
 189 static void startup_modules(void);
 190 static void startup_vm(void);
 191 static void startup_end(void);
 192 static void layout_kernel_va(void);
 193 
 194 /*
 195  * Declare these as initialized data so we can patch them.
 196  */
 197 #ifdef __i386
 198 
 199 /*
 200  * Due to virtual address space limitations running in 32 bit mode, restrict
 201  * the amount of physical memory configured to a max of PHYSMEM pages (16g).
 202  *
 203  * If the physical max memory size of 64g were allowed to be configured, the
 204  * size of user virtual address space will be less than 1g. A limited user
 205  * address space greatly reduces the range of applications that can run.
 206  *
 207  * If more physical memory than PHYSMEM is required, users should preferably
 208  * run in 64 bit mode which has far looser virtual address space limitations.
 209  *
 210  * If 64 bit mode is not available (as in IA32) and/or more physical memory
 211  * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired
 212  * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase
 213  * should also be carefully tuned to balance out the need of the user
 214  * application while minimizing the risk of kernel heap exhaustion due to
 215  * kernelbase being set too high.
 216  */
 217 #define PHYSMEM 0x400000
 218 
 219 #else /* __amd64 */
 220 
 221 /*
 222  * For now we can handle memory with physical addresses up to about
 223  * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly
 224  * half the VA space for seg_kpm. When systems get bigger than 64TB this
 225  * code will need revisiting. There is an implicit assumption that there
 226  * are no *huge* holes in the physical address space too.
 227  */
 228 #define TERABYTE                (1ul << 40)
 229 #define PHYSMEM_MAX64           mmu_btop(64 * TERABYTE)
 230 #define PHYSMEM                 PHYSMEM_MAX64
 231 #define AMD64_VA_HOLE_END       0xFFFF800000000000ul
 232 
 233 #endif /* __amd64 */
 234 
 235 volatile pgcnt_t physmem = PHYSMEM;
 236 pgcnt_t obp_pages;      /* Memory used by PROM for its text and data */
 237 
 238 char *kobj_file_buf;
 239 int kobj_file_bufsize;  /* set in /etc/system */
 240 
 241 /* Global variables for MP support. Used in mp_startup */
 242 caddr_t rm_platter_va = 0;
 243 uint32_t rm_platter_pa;
 244 
 245 int     auto_lpg_disable = 1;
 246 
 247 /*
 248  * Some CPUs have holes in the middle of the 64-bit virtual address range.
 249  */
 250 uintptr_t hole_start, hole_end;
 251 
 252 /*
 253  * kpm mapping window
 254  */
 255 caddr_t kpm_vbase;
 256 size_t  kpm_size;
 257 static int kpm_desired;
 258 #ifdef __amd64
 259 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE;
 260 #endif
 261 
 262 /*
 263  * Configuration parameters set at boot time.
 264  */
 265 
 266 caddr_t econtig;                /* end of first block of contiguous kernel */
 267 
 268 struct bootops          *bootops = 0;   /* passed in from boot */
 269 struct bootops          **bootopsp;
 270 struct boot_syscalls    *sysp;          /* passed in from boot */
 271 
 272 char bootblock_fstype[16];
 273 
 274 char kern_bootargs[OBP_MAXPATHLEN];
 275 char kern_bootfile[OBP_MAXPATHLEN];
 276 
 277 /*
 278  * ZFS zio segment.  This allows us to exclude large portions of ZFS data that
 279  * gets cached in kmem caches on the heap.  If this is set to zero, we allocate
 280  * zio buffers from their own segment, otherwise they are allocated from the
 281  * heap.  The optimization of allocating zio buffers from their own segment is
 282  * only valid on 64-bit kernels.
 283  */
 284 #if defined(__amd64)
 285 int segzio_fromheap = 0;
 286 #else
 287 int segzio_fromheap = 1;
 288 #endif
 289 
 290 /*
 291  * Give folks an escape hatch for disabling SMAP via kmdb. Doesn't work
 292  * post-boot.
 293  */
 294 int disable_smap = 0;
 295 
 296 /*
 297  * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this
 298  * depends on number of BOP_ALLOC calls made and requested size, memory size
 299  * combination and whether boot.bin memory needs to be freed.
 300  */
 301 #define POSS_NEW_FRAGMENTS      12
 302 
 303 /*
 304  * VM data structures
 305  */
 306 long page_hashsz;               /* Size of page hash table (power of two) */
 307 unsigned int page_hashsz_shift; /* log2(page_hashsz) */
 308 struct page *pp_base;           /* Base of initial system page struct array */
 309 struct page **page_hash;        /* Page hash table */
 310 pad_mutex_t *pse_mutex;         /* Locks protecting pp->p_selock */
 311 size_t pse_table_size;          /* Number of mutexes in pse_mutex[] */
 312 int pse_shift;                  /* log2(pse_table_size) */
 313 struct seg ktextseg;            /* Segment used for kernel executable image */
 314 struct seg kvalloc;             /* Segment used for "valloc" mapping */
 315 struct seg kpseg;               /* Segment used for pageable kernel virt mem */
 316 struct seg kmapseg;             /* Segment used for generic kernel mappings */
 317 struct seg kdebugseg;           /* Segment used for the kernel debugger */
 318 
 319 struct seg *segkmap = &kmapseg;     /* Kernel generic mapping segment */
 320 static struct seg *segmap = &kmapseg;       /* easier to use name for in here */
 321 
 322 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */
 323 
 324 #if defined(__amd64)
 325 struct seg kvseg_core;          /* Segment used for the core heap */
 326 struct seg kpmseg;              /* Segment used for physical mapping */
 327 struct seg *segkpm = &kpmseg;       /* 64bit kernel physical mapping segment */
 328 #else
 329 struct seg *segkpm = NULL;      /* Unused on IA32 */
 330 #endif
 331 
 332 caddr_t segkp_base;             /* Base address of segkp */
 333 caddr_t segzio_base;            /* Base address of segzio */
 334 #if defined(__amd64)
 335 volatile pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in */
 336                                                 /* pages */
 337 #else
 338 volatile pgcnt_t segkpsize = 0;
 339 #endif
 340 pgcnt_t segziosize = 0;         /* size of zio segment in pages */
 341 
 342 /*
 343  * A static DR page_t VA map is reserved that can map the page structures
 344  * for a domain's entire RA space. The pages that back this space are
 345  * dynamically allocated and need not be physically contiguous.  The DR
 346  * map size is derived from KPM size.
 347  * This mechanism isn't used by x86 yet, so just stubs here.
 348  */
 349 int ppvm_enable = 0;            /* Static virtual map for page structs */
 350 page_t *ppvm_base = NULL;       /* Base of page struct map */
 351 pgcnt_t ppvm_size = 0;          /* Size of page struct map */
 352 
 353 /*
 354  * VA range available to the debugger
 355  */
 356 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
 357 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
 358 
 359 struct memseg *memseg_base;
 360 struct vnode unused_pages_vp;
 361 
 362 #define FOURGB  0x100000000LL
 363 
 364 struct memlist *memlist;
 365 
 366 caddr_t s_text;         /* start of kernel text segment */
 367 caddr_t e_text;         /* end of kernel text segment */
 368 caddr_t s_data;         /* start of kernel data segment */
 369 caddr_t e_data;         /* end of kernel data segment */
 370 caddr_t modtext;        /* start of loadable module text reserved */
 371 caddr_t e_modtext;      /* end of loadable module text reserved */
 372 caddr_t moddata;        /* start of loadable module data reserved */
 373 caddr_t e_moddata;      /* end of loadable module data reserved */
 374 
 375 struct memlist *phys_install;   /* Total installed physical memory */
 376 struct memlist *phys_avail;     /* Total available physical memory */
 377 struct memlist *bios_rsvd;      /* Bios reserved memory */
 378 
 379 /*
 380  * kphysm_init returns the number of pages that were processed
 381  */
 382 static pgcnt_t kphysm_init(page_t *, pgcnt_t);
 383 
 384 #define IO_PROP_SIZE    64      /* device property size */
 385 
 386 /*
 387  * a couple useful roundup macros
 388  */
 389 #define ROUND_UP_PAGE(x)        \
 390         ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE))
 391 #define ROUND_UP_LPAGE(x)       \
 392         ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1]))
 393 #define ROUND_UP_4MEG(x)        \
 394         ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG))
 395 #define ROUND_UP_TOPLEVEL(x)    \
 396         ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level]))
 397 
 398 /*
 399  *      32-bit Kernel's Virtual memory layout.
 400  *              +-----------------------+
 401  *              |                       |
 402  * 0xFFC00000  -|-----------------------|- ARGSBASE
 403  *              |       debugger        |
 404  * 0xFF800000  -|-----------------------|- SEGDEBUGBASE
 405  *              |      Kernel Data      |
 406  * 0xFEC00000  -|-----------------------|
 407  *              |      Kernel Text      |
 408  * 0xFE800000  -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen)
 409  *              |---       GDT       ---|- GDT page (GDT_VA)
 410  *              |---    debug info   ---|- debug info (DEBUG_INFO_VA)
 411  *              |                       |
 412  *              |   page_t structures   |
 413  *              |   memsegs, memlists,  |
 414  *              |   page hash, etc.     |
 415  * ---         -|-----------------------|- ekernelheap, valloc_base (floating)
 416  *              |                       |  (segkp is just an arena in the heap)
 417  *              |                       |
 418  *              |       kvseg           |
 419  *              |                       |
 420  *              |                       |
 421  * ---         -|-----------------------|- kernelheap (floating)
 422  *              |        Segkmap        |
 423  * 0xC3002000  -|-----------------------|- segmap_start (floating)
 424  *              |       Red Zone        |
 425  * 0xC3000000  -|-----------------------|- kernelbase / userlimit (floating)
 426  *              |                       |                       ||
 427  *              |     Shared objects    |                       \/
 428  *              |                       |
 429  *              :                       :
 430  *              |       user data       |
 431  *              |-----------------------|
 432  *              |       user text       |
 433  * 0x08048000  -|-----------------------|
 434  *              |       user stack      |
 435  *              :                       :
 436  *              |       invalid         |
 437  * 0x00000000   +-----------------------+
 438  *
 439  *
 440  *              64-bit Kernel's Virtual memory layout. (assuming 64 bit app)
 441  *                      +-----------------------+
 442  *                      |                       |
 443  * 0xFFFFFFFF.FFC00000  |-----------------------|- ARGSBASE
 444  *                      |       debugger (?)    |
 445  * 0xFFFFFFFF.FF800000  |-----------------------|- SEGDEBUGBASE
 446  *                      |      unused           |
 447  *                      +-----------------------+
 448  *                      |      Kernel Data      |
 449  * 0xFFFFFFFF.FBC00000  |-----------------------|
 450  *                      |      Kernel Text      |
 451  * 0xFFFFFFFF.FB800000  |-----------------------|- KERNEL_TEXT
 452  *                      |---       GDT       ---|- GDT page (GDT_VA)
 453  *                      |---    debug info   ---|- debug info (DEBUG_INFO_VA)
 454  *                      |                       |
 455  *                      |      Core heap        | (used for loadable modules)
 456  * 0xFFFFFFFF.C0000000  |-----------------------|- core_base / ekernelheap
 457  *                      |        Kernel         |
 458  *                      |         heap          |
 459  * 0xFFFFFXXX.XXX00000  |-----------------------|- kernelheap (floating)
 460  *                      |        segmap         |
 461  * 0xFFFFFXXX.XXX00000  |-----------------------|- segmap_start (floating)
 462  *                      |    device mappings    |
 463  * 0xFFFFFXXX.XXX00000  |-----------------------|- toxic_addr (floating)
 464  *                      |         segzio        |
 465  * 0xFFFFFXXX.XXX00000  |-----------------------|- segzio_base (floating)
 466  *                      |         segkp         |
 467  * ---                  |-----------------------|- segkp_base (floating)
 468  *                      |   page_t structures   |  valloc_base + valloc_sz
 469  *                      |   memsegs, memlists,  |
 470  *                      |   page hash, etc.     |
 471  * 0xFFFFFF00.00000000  |-----------------------|- valloc_base (lower if >256GB)
 472  *                      |        segkpm         |
 473  * 0xFFFFFE00.00000000  |-----------------------|
 474  *                      |       Red Zone        |
 475  * 0xFFFFFD80.00000000  |-----------------------|- KERNELBASE (lower if >256GB)
 476  *                      |     User stack        |- User space memory
 477  *                      |                       |
 478  *                      | shared objects, etc   |       (grows downwards)
 479  *                      :                       :
 480  *                      |                       |
 481  * 0xFFFF8000.00000000  |-----------------------|
 482  *                      |                       |
 483  *                      | VA Hole / unused      |
 484  *                      |                       |
 485  * 0x00008000.00000000  |-----------------------|
 486  *                      |                       |
 487  *                      |                       |
 488  *                      :                       :
 489  *                      |       user heap       |       (grows upwards)
 490  *                      |                       |
 491  *                      |       user data       |
 492  *                      |-----------------------|
 493  *                      |       user text       |
 494  * 0x00000000.04000000  |-----------------------|
 495  *                      |       invalid         |
 496  * 0x00000000.00000000  +-----------------------+
 497  *
 498  * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit
 499  * kernel, except that userlimit is raised to 0xfe000000
 500  *
 501  * Floating values:
 502  *
 503  * valloc_base: start of the kernel's memory management/tracking data
 504  * structures.  This region contains page_t structures for
 505  * physical memory, memsegs, memlists, and the page hash.
 506  *
 507  * core_base: start of the kernel's "core" heap area on 64-bit systems.
 508  * This area is intended to be used for global data as well as for module
 509  * text/data that does not fit into the nucleus pages.  The core heap is
 510  * restricted to a 2GB range, allowing every address within it to be
 511  * accessed using rip-relative addressing
 512  *
 513  * ekernelheap: end of kernelheap and start of segmap.
 514  *
 515  * kernelheap: start of kernel heap.  On 32-bit systems, this starts right
 516  * above a red zone that separates the user's address space from the
 517  * kernel's.  On 64-bit systems, it sits above segkp and segkpm.
 518  *
 519  * segmap_start: start of segmap. The length of segmap can be modified
 520  * through eeprom. The default length is 16MB on 32-bit systems and 64MB
 521  * on 64-bit systems.
 522  *
 523  * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be
 524  * decreased by 2X the size required for page_t.  This allows the kernel
 525  * heap to grow in size with physical memory.  With sizeof(page_t) == 80
 526  * bytes, the following shows the values of kernelbase and kernel heap
 527  * sizes for different memory configurations (assuming default segmap and
 528  * segkp sizes).
 529  *
 530  *      mem     size for        kernelbase      kernel heap
 531  *      size    page_t's                        size
 532  *      ----    ---------       ----------      -----------
 533  *      1gb     0x01400000      0xd1800000      684MB
 534  *      2gb     0x02800000      0xcf000000      704MB
 535  *      4gb     0x05000000      0xca000000      744MB
 536  *      6gb     0x07800000      0xc5000000      784MB
 537  *      8gb     0x0a000000      0xc0000000      824MB
 538  *      16gb    0x14000000      0xac000000      984MB
 539  *      32gb    0x28000000      0x84000000      1304MB
 540  *      64gb    0x50000000      0x34000000      1944MB (*)
 541  *
 542  * kernelbase is less than the abi minimum of 0xc0000000 for memory
 543  * configurations above 8gb.
 544  *
 545  * (*) support for memory configurations above 32gb will require manual tuning
 546  * of kernelbase to balance out the need of user applications.
 547  */
 548 
 549 /* real-time-clock initialization parameters */
 550 extern time_t process_rtc_config_file(void);
 551 
 552 uintptr_t       kernelbase;
 553 uintptr_t       postbootkernelbase;     /* not set till boot loader is gone */
 554 uintptr_t       eprom_kernelbase;
 555 size_t          segmapsize;
 556 uintptr_t       segmap_start;
 557 int             segmapfreelists;
 558 pgcnt_t         npages;
 559 pgcnt_t         orig_npages;
 560 size_t          core_size;              /* size of "core" heap */
 561 uintptr_t       core_base;              /* base address of "core" heap */
 562 
 563 /*
 564  * List of bootstrap pages. We mark these as allocated in startup.
 565  * release_bootstrap() will free them when we're completely done with
 566  * the bootstrap.
 567  */
 568 static page_t *bootpages;
 569 
 570 /*
 571  * boot time pages that have a vnode from the ramdisk will keep that forever.
 572  */
 573 static page_t *rd_pages;
 574 
 575 /*
 576  * Lower 64K
 577  */
 578 static page_t *lower_pages = NULL;
 579 static int lower_pages_count = 0;
 580 
 581 struct system_hardware system_hardware;
 582 
 583 /*
 584  * Enable some debugging messages concerning memory usage...
 585  */
 586 static void
 587 print_memlist(char *title, struct memlist *mp)
 588 {
 589         prom_printf("MEMLIST: %s:\n", title);
 590         while (mp != NULL)  {
 591                 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n",
 592                     mp->ml_address, mp->ml_size);
 593                 mp = mp->ml_next;
 594         }
 595 }
 596 
 597 /*
 598  * XX64 need a comment here.. are these just default values, surely
 599  * we read the "cpuid" type information to figure this out.
 600  */
 601 int     l2cache_sz = 0x80000;
 602 int     l2cache_linesz = 0x40;
 603 int     l2cache_assoc = 1;
 604 
 605 static size_t   textrepl_min_gb = 10;
 606 
 607 /*
 608  * on 64 bit we use a predifined VA range for mapping devices in the kernel
 609  * on 32 bit the mappings are intermixed in the heap, so we use a bit map
 610  */
 611 #ifdef __amd64
 612 
 613 vmem_t          *device_arena;
 614 uintptr_t       toxic_addr = (uintptr_t)NULL;
 615 size_t          toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */
 616 
 617 #else   /* __i386 */
 618 
 619 ulong_t         *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */
 620 size_t          toxic_bit_map_len = 0;  /* in bits */
 621 
 622 #endif  /* __i386 */
 623 
 624 /*
 625  * Simple boot time debug facilities
 626  */
 627 static char *prm_dbg_str[] = {
 628         "%s:%d: '%s' is 0x%x\n",
 629         "%s:%d: '%s' is 0x%llx\n"
 630 };
 631 
 632 int prom_debug;
 633 
 634 #define PRM_DEBUG(q)    if (prom_debug)         \
 635         prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q);
 636 #define PRM_POINT(q)    if (prom_debug)         \
 637         prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q);
 638 
 639 /*
 640  * This structure is used to keep track of the intial allocations
 641  * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to
 642  * be >= the number of ADD_TO_ALLOCATIONS() executed in the code.
 643  */
 644 #define NUM_ALLOCATIONS 8
 645 int num_allocations = 0;
 646 struct {
 647         void **al_ptr;
 648         size_t al_size;
 649 } allocations[NUM_ALLOCATIONS];
 650 size_t valloc_sz = 0;
 651 uintptr_t valloc_base;
 652 
 653 #define ADD_TO_ALLOCATIONS(ptr, size) {                                 \
 654                 size = ROUND_UP_PAGE(size);                             \
 655                 if (num_allocations == NUM_ALLOCATIONS)                 \
 656                         panic("too many ADD_TO_ALLOCATIONS()");         \
 657                 allocations[num_allocations].al_ptr = (void**)&ptr; \
 658                 allocations[num_allocations].al_size = size;            \
 659                 valloc_sz += size;                                      \
 660                 ++num_allocations;                                      \
 661         }
 662 
 663 /*
 664  * Allocate all the initial memory needed by the page allocator.
 665  */
 666 static void
 667 perform_allocations(void)
 668 {
 669         caddr_t mem;
 670         int i;
 671         int valloc_align;
 672 
 673         PRM_DEBUG(valloc_base);
 674         PRM_DEBUG(valloc_sz);
 675         valloc_align = mmu.level_size[mmu.max_page_level > 0];
 676         mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align);
 677         if (mem != (caddr_t)valloc_base)
 678                 panic("BOP_ALLOC() failed");
 679         bzero(mem, valloc_sz);
 680         for (i = 0; i < num_allocations; ++i) {
 681                 *allocations[i].al_ptr = (void *)mem;
 682                 mem += allocations[i].al_size;
 683         }
 684 }
 685 
 686 /*
 687  * Set up and enable SMAP now before we start other CPUs, but after the kernel's
 688  * VM has been set up so we can use hot_patch_kernel_text().
 689  *
 690  * We can only patch 1, 2, or 4 bytes, but not three bytes. So instead, we
 691  * replace the four byte word at the patch point. See uts/intel/ia32/ml/copy.s
 692  * for more information on what's going on here.
 693  */
 694 static void
 695 startup_smap(void)
 696 {
 697         int i;
 698         uint32_t inst;
 699         uint8_t *instp;
 700         char sym[128];
 701 
 702         extern int _smap_enable_patch_count;
 703         extern int _smap_disable_patch_count;
 704 
 705         if (disable_smap != 0)
 706                 remove_x86_feature(x86_featureset, X86FSET_SMAP);
 707 
 708         if (is_x86_feature(x86_featureset, X86FSET_SMAP) == B_FALSE)
 709                 return;
 710 
 711         for (i = 0; i < _smap_enable_patch_count; i++) {
 712                 int sizep;
 713 
 714                 VERIFY3U(i, <, _smap_enable_patch_count);
 715                 VERIFY(snprintf(sym, sizeof (sym), "_smap_enable_patch_%d", i) <
 716                     sizeof (sym));
 717                 instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep);
 718                 VERIFY(instp != 0);
 719                 inst = (instp[3] << 24) | (SMAP_CLAC_INSTR & 0x00ffffff);
 720                 hot_patch_kernel_text((caddr_t)instp, inst, 4);
 721         }
 722 
 723         for (i = 0; i < _smap_disable_patch_count; i++) {
 724                 int sizep;
 725 
 726                 VERIFY(snprintf(sym, sizeof (sym), "_smap_disable_patch_%d",
 727                     i) < sizeof (sym));
 728                 instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep);
 729                 VERIFY(instp != 0);
 730                 inst = (instp[3] << 24) | (SMAP_STAC_INSTR & 0x00ffffff);
 731                 hot_patch_kernel_text((caddr_t)instp, inst, 4);
 732         }
 733 
 734         hot_patch_kernel_text((caddr_t)smap_enable, SMAP_CLAC_INSTR, 4);
 735         hot_patch_kernel_text((caddr_t)smap_disable, SMAP_STAC_INSTR, 4);
 736         setcr4(getcr4() | CR4_SMAP);
 737         smap_enable();
 738 }
 739 
 740 /*
 741  * Our world looks like this at startup time.
 742  *
 743  * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data
 744  * at 0xfec00000.  On a 64-bit OS, kernel text and data are loaded at
 745  * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively.  Those
 746  * addresses are fixed in the binary at link time.
 747  *
 748  * On the text page:
 749  * unix/genunix/krtld/module text loads.
 750  *
 751  * On the data page:
 752  * unix/genunix/krtld/module data loads.
 753  *
 754  * Machine-dependent startup code
 755  */
 756 void
 757 startup(void)
 758 {
 759 #if !defined(__xpv)
 760         extern void startup_pci_bios(void);
 761 #endif
 762         extern cpuset_t cpu_ready_set;
 763 
 764         /*
 765          * Make sure that nobody tries to use sekpm until we have
 766          * initialized it properly.
 767          */
 768 #if defined(__amd64)
 769         kpm_desired = 1;
 770 #endif
 771         kpm_enable = 0;
 772         CPUSET_ONLY(cpu_ready_set, 0);  /* cpu 0 is boot cpu */
 773 
 774 #if defined(__xpv)      /* XXPV fix me! */
 775         {
 776                 extern int segvn_use_regions;
 777                 segvn_use_regions = 0;
 778         }
 779 #endif
 780         ssp_init();
 781         progressbar_init();
 782         startup_init();
 783 #if defined(__xpv)
 784         startup_xen_version();
 785 #endif
 786         startup_memlist();
 787         startup_kmem();
 788         startup_vm();
 789 #if !defined(__xpv)
 790         /*
 791          * Note we need to do this even on fast reboot in order to access
 792          * the irq routing table (used for pci labels).
 793          */
 794         startup_pci_bios();
 795         startup_smap();
 796 #endif
 797 #if defined(__xpv)
 798         startup_xen_mca();
 799 #endif
 800         startup_modules();
 801 
 802         startup_end();
 803 }
 804 
 805 static void
 806 startup_init()
 807 {
 808         PRM_POINT("startup_init() starting...");
 809 
 810         /*
 811          * Complete the extraction of cpuid data
 812          */
 813         cpuid_pass2(CPU);
 814 
 815         (void) check_boot_version(BOP_GETVERSION(bootops));
 816 
 817         /*
 818          * Check for prom_debug in boot environment
 819          */
 820         if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) {
 821                 ++prom_debug;
 822                 PRM_POINT("prom_debug found in boot enviroment");
 823         }
 824 
 825         /*
 826          * Collect node, cpu and memory configuration information.
 827          */
 828         get_system_configuration();
 829 
 830         /*
 831          * Halt if this is an unsupported processor.
 832          */
 833         if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) {
 834                 printf("\n486 processor (\"%s\") detected.\n",
 835                     CPU->cpu_brandstr);
 836                 halt("This processor is not supported by this release "
 837                     "of Solaris.");
 838         }
 839 
 840         PRM_POINT("startup_init() done");
 841 }
 842 
 843 /*
 844  * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie.
 845  * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it
 846  * also filters out physical page zero.  There is some reliance on the
 847  * boot loader allocating only a few contiguous physical memory chunks.
 848  */
 849 static void
 850 avail_filter(uint64_t *addr, uint64_t *size)
 851 {
 852         uintptr_t va;
 853         uintptr_t next_va;
 854         pfn_t pfn;
 855         uint64_t pfn_addr;
 856         uint64_t pfn_eaddr;
 857         uint_t prot;
 858         size_t len;
 859         uint_t change;
 860 
 861         if (prom_debug)
 862                 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n",
 863                     *addr, *size);
 864 
 865         /*
 866          * page zero is required for BIOS.. never make it available
 867          */
 868         if (*addr == 0) {
 869                 *addr += MMU_PAGESIZE;
 870                 *size -= MMU_PAGESIZE;
 871         }
 872 
 873         /*
 874          * First we trim from the front of the range. Since kbm_probe()
 875          * walks ranges in virtual order, but addr/size are physical, we need
 876          * to the list until no changes are seen.  This deals with the case
 877          * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w
 878          * but w < v.
 879          */
 880         do {
 881                 change = 0;
 882                 for (va = KERNEL_TEXT;
 883                     *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
 884                     va = next_va) {
 885 
 886                         next_va = va + len;
 887                         pfn_addr = pfn_to_pa(pfn);
 888                         pfn_eaddr = pfn_addr + len;
 889 
 890                         if (pfn_addr <= *addr && pfn_eaddr > *addr) {
 891                                 change = 1;
 892                                 while (*size > 0 && len > 0) {
 893                                         *addr += MMU_PAGESIZE;
 894                                         *size -= MMU_PAGESIZE;
 895                                         len -= MMU_PAGESIZE;
 896                                 }
 897                         }
 898                 }
 899                 if (change && prom_debug)
 900                         prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n",
 901                             *addr, *size);
 902         } while (change);
 903 
 904         /*
 905          * Trim pages from the end of the range.
 906          */
 907         for (va = KERNEL_TEXT;
 908             *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
 909             va = next_va) {
 910 
 911                 next_va = va + len;
 912                 pfn_addr = pfn_to_pa(pfn);
 913 
 914                 if (pfn_addr >= *addr && pfn_addr < *addr + *size)
 915                         *size = pfn_addr - *addr;
 916         }
 917 
 918         if (prom_debug)
 919                 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n",
 920                     *addr, *size);
 921 }
 922 
 923 static void
 924 kpm_init()
 925 {
 926         struct segkpm_crargs b;
 927 
 928         /*
 929          * These variables were all designed for sfmmu in which segkpm is
 930          * mapped using a single pagesize - either 8KB or 4MB.  On x86, we
 931          * might use 2+ page sizes on a single machine, so none of these
 932          * variables have a single correct value.  They are set up as if we
 933          * always use a 4KB pagesize, which should do no harm.  In the long
 934          * run, we should get rid of KPM's assumption that only a single
 935          * pagesize is used.
 936          */
 937         kpm_pgshft = MMU_PAGESHIFT;
 938         kpm_pgsz =  MMU_PAGESIZE;
 939         kpm_pgoff = MMU_PAGEOFFSET;
 940         kpmp2pshft = 0;
 941         kpmpnpgs = 1;
 942         ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0);
 943 
 944         PRM_POINT("about to create segkpm");
 945         rw_enter(&kas.a_lock, RW_WRITER);
 946 
 947         if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0)
 948                 panic("cannot attach segkpm");
 949 
 950         b.prot = PROT_READ | PROT_WRITE;
 951         b.nvcolors = 1;
 952 
 953         if (segkpm_create(segkpm, (caddr_t)&b) != 0)
 954                 panic("segkpm_create segkpm");
 955 
 956         rw_exit(&kas.a_lock);
 957 }
 958 
 959 /*
 960  * The debug info page provides enough information to allow external
 961  * inspectors (e.g. when running under a hypervisor) to bootstrap
 962  * themselves into allowing full-blown kernel debugging.
 963  */
 964 static void
 965 init_debug_info(void)
 966 {
 967         caddr_t mem;
 968         debug_info_t *di;
 969 
 970 #ifndef __lint
 971         ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE);
 972 #endif
 973 
 974         mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE,
 975             MMU_PAGESIZE);
 976 
 977         if (mem != (caddr_t)DEBUG_INFO_VA)
 978                 panic("BOP_ALLOC() failed");
 979         bzero(mem, MMU_PAGESIZE);
 980 
 981         di = (debug_info_t *)mem;
 982 
 983         di->di_magic = DEBUG_INFO_MAGIC;
 984         di->di_version = DEBUG_INFO_VERSION;
 985         di->di_modules = (uintptr_t)&modules;
 986         di->di_s_text = (uintptr_t)s_text;
 987         di->di_e_text = (uintptr_t)e_text;
 988         di->di_s_data = (uintptr_t)s_data;
 989         di->di_e_data = (uintptr_t)e_data;
 990         di->di_hat_htable_off = offsetof(hat_t, hat_htable);
 991         di->di_ht_pfn_off = offsetof(htable_t, ht_pfn);
 992 }
 993 
 994 /*
 995  * Build the memlists and other kernel essential memory system data structures.
 996  * This is everything at valloc_base.
 997  */
 998 static void
 999 startup_memlist(void)
1000 {
1001         size_t memlist_sz;
1002         size_t memseg_sz;
1003         size_t pagehash_sz;
1004         size_t pp_sz;
1005         uintptr_t va;
1006         size_t len;
1007         uint_t prot;
1008         pfn_t pfn;
1009         int memblocks;
1010         pfn_t rsvd_high_pfn;
1011         pgcnt_t rsvd_pgcnt;
1012         size_t rsvdmemlist_sz;
1013         int rsvdmemblocks;
1014         caddr_t pagecolor_mem;
1015         size_t pagecolor_memsz;
1016         caddr_t page_ctrs_mem;
1017         size_t page_ctrs_size;
1018         size_t pse_table_alloc_size;
1019         struct memlist *current;
1020         extern void startup_build_mem_nodes(struct memlist *);
1021 
1022         /* XX64 fix these - they should be in include files */
1023         extern size_t page_coloring_init(uint_t, int, int);
1024         extern void page_coloring_setup(caddr_t);
1025 
1026         PRM_POINT("startup_memlist() starting...");
1027 
1028         /*
1029          * Use leftover large page nucleus text/data space for loadable modules.
1030          * Use at most MODTEXT/MODDATA.
1031          */
1032         len = kbm_nucleus_size;
1033         ASSERT(len > MMU_PAGESIZE);
1034 
1035         moddata = (caddr_t)ROUND_UP_PAGE(e_data);
1036         e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len);
1037         if (e_moddata - moddata > MODDATA)
1038                 e_moddata = moddata + MODDATA;
1039 
1040         modtext = (caddr_t)ROUND_UP_PAGE(e_text);
1041         e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len);
1042         if (e_modtext - modtext > MODTEXT)
1043                 e_modtext = modtext + MODTEXT;
1044 
1045         econtig = e_moddata;
1046 
1047         PRM_DEBUG(modtext);
1048         PRM_DEBUG(e_modtext);
1049         PRM_DEBUG(moddata);
1050         PRM_DEBUG(e_moddata);
1051         PRM_DEBUG(econtig);
1052 
1053         /*
1054          * Examine the boot loader physical memory map to find out:
1055          * - total memory in system - physinstalled
1056          * - the max physical address - physmax
1057          * - the number of discontiguous segments of memory.
1058          */
1059         if (prom_debug)
1060                 print_memlist("boot physinstalled",
1061                     bootops->boot_mem->physinstalled);
1062         installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax,
1063             &physinstalled, &memblocks);
1064         PRM_DEBUG(physmax);
1065         PRM_DEBUG(physinstalled);
1066         PRM_DEBUG(memblocks);
1067 
1068         /*
1069          * Compute maximum physical address for memory DR operations.
1070          * Memory DR operations are unsupported on xpv or 32bit OSes.
1071          */
1072 #ifdef  __amd64
1073         if (plat_dr_support_memory()) {
1074                 if (plat_dr_physmax == 0) {
1075                         uint_t pabits = UINT_MAX;
1076 
1077                         cpuid_get_addrsize(CPU, &pabits, NULL);
1078                         plat_dr_physmax = btop(1ULL << pabits);
1079                 }
1080                 if (plat_dr_physmax > PHYSMEM_MAX64)
1081                         plat_dr_physmax = PHYSMEM_MAX64;
1082         } else
1083 #endif
1084                 plat_dr_physmax = 0;
1085 
1086         /*
1087          * Examine the bios reserved memory to find out:
1088          * - the number of discontiguous segments of memory.
1089          */
1090         if (prom_debug)
1091                 print_memlist("boot reserved mem",
1092                     bootops->boot_mem->rsvdmem);
1093         installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn,
1094             &rsvd_pgcnt, &rsvdmemblocks);
1095         PRM_DEBUG(rsvd_high_pfn);
1096         PRM_DEBUG(rsvd_pgcnt);
1097         PRM_DEBUG(rsvdmemblocks);
1098 
1099         /*
1100          * Initialize hat's mmu parameters.
1101          * Check for enforce-prot-exec in boot environment. It's used to
1102          * enable/disable support for the page table entry NX bit.
1103          * The default is to enforce PROT_EXEC on processors that support NX.
1104          * Boot seems to round up the "len", but 8 seems to be big enough.
1105          */
1106         mmu_init();
1107 
1108 #ifdef  __i386
1109         /*
1110          * physmax is lowered if there is more memory than can be
1111          * physically addressed in 32 bit (PAE/non-PAE) modes.
1112          */
1113         if (mmu.pae_hat) {
1114                 if (PFN_ABOVE64G(physmax)) {
1115                         physinstalled -= (physmax - (PFN_64G - 1));
1116                         physmax = PFN_64G - 1;
1117                 }
1118         } else {
1119                 if (PFN_ABOVE4G(physmax)) {
1120                         physinstalled -= (physmax - (PFN_4G - 1));
1121                         physmax = PFN_4G - 1;
1122                 }
1123         }
1124 #endif
1125 
1126         startup_build_mem_nodes(bootops->boot_mem->physinstalled);
1127 
1128         if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) {
1129                 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec");
1130                 char value[8];
1131 
1132                 if (len < 8)
1133                         (void) BOP_GETPROP(bootops, "enforce-prot-exec", value);
1134                 else
1135                         (void) strcpy(value, "");
1136                 if (strcmp(value, "off") == 0)
1137                         mmu.pt_nx = 0;
1138         }
1139         PRM_DEBUG(mmu.pt_nx);
1140 
1141         /*
1142          * We will need page_t's for every page in the system, except for
1143          * memory mapped at or above above the start of the kernel text segment.
1144          *
1145          * pages above e_modtext are attributed to kernel debugger (obp_pages)
1146          */
1147         npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */
1148         obp_pages = 0;
1149         va = KERNEL_TEXT;
1150         while (kbm_probe(&va, &len, &pfn, &prot) != 0) {
1151                 npages -= len >> MMU_PAGESHIFT;
1152                 if (va >= (uintptr_t)e_moddata)
1153                         obp_pages += len >> MMU_PAGESHIFT;
1154                 va += len;
1155         }
1156         PRM_DEBUG(npages);
1157         PRM_DEBUG(obp_pages);
1158 
1159         /*
1160          * If physmem is patched to be non-zero, use it instead of the computed
1161          * value unless it is larger than the actual amount of memory on hand.
1162          */
1163         if (physmem == 0 || physmem > npages) {
1164                 physmem = npages;
1165         } else if (physmem < npages) {
1166                 orig_npages = npages;
1167                 npages = physmem;
1168         }
1169         PRM_DEBUG(physmem);
1170 
1171         /*
1172          * We now compute the sizes of all the  initial allocations for
1173          * structures the kernel needs in order do kmem_alloc(). These
1174          * include:
1175          *      memsegs
1176          *      memlists
1177          *      page hash table
1178          *      page_t's
1179          *      page coloring data structs
1180          */
1181         memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS);
1182         ADD_TO_ALLOCATIONS(memseg_base, memseg_sz);
1183         PRM_DEBUG(memseg_sz);
1184 
1185         /*
1186          * Reserve space for memlists. There's no real good way to know exactly
1187          * how much room we'll need, but this should be a good upper bound.
1188          */
1189         memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1190             (memblocks + POSS_NEW_FRAGMENTS));
1191         ADD_TO_ALLOCATIONS(memlist, memlist_sz);
1192         PRM_DEBUG(memlist_sz);
1193 
1194         /*
1195          * Reserve space for bios reserved memlists.
1196          */
1197         rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1198             (rsvdmemblocks + POSS_NEW_FRAGMENTS));
1199         ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz);
1200         PRM_DEBUG(rsvdmemlist_sz);
1201 
1202         /* LINTED */
1203         ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page)));
1204         /*
1205          * The page structure hash table size is a power of 2
1206          * such that the average hash chain length is PAGE_HASHAVELEN.
1207          */
1208         page_hashsz = npages / PAGE_HASHAVELEN;
1209         page_hashsz_shift = highbit(page_hashsz);
1210         page_hashsz = 1 << page_hashsz_shift;
1211         pagehash_sz = sizeof (struct page *) * page_hashsz;
1212         ADD_TO_ALLOCATIONS(page_hash, pagehash_sz);
1213         PRM_DEBUG(pagehash_sz);
1214 
1215         /*
1216          * Set aside room for the page structures themselves.
1217          */
1218         PRM_DEBUG(npages);
1219         pp_sz = sizeof (struct page) * npages;
1220         ADD_TO_ALLOCATIONS(pp_base, pp_sz);
1221         PRM_DEBUG(pp_sz);
1222 
1223         /*
1224          * determine l2 cache info and memory size for page coloring
1225          */
1226         (void) getl2cacheinfo(CPU,
1227             &l2cache_sz, &l2cache_linesz, &l2cache_assoc);
1228         pagecolor_memsz =
1229             page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc);
1230         ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz);
1231         PRM_DEBUG(pagecolor_memsz);
1232 
1233         page_ctrs_size = page_ctrs_sz();
1234         ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size);
1235         PRM_DEBUG(page_ctrs_size);
1236 
1237         /*
1238          * Allocate the array that protects pp->p_selock.
1239          */
1240         pse_shift = size_pse_array(physmem, max_ncpus);
1241         pse_table_size = 1 << pse_shift;
1242         pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t);
1243         ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size);
1244 
1245 #if defined(__amd64)
1246         valloc_sz = ROUND_UP_LPAGE(valloc_sz);
1247         valloc_base = VALLOC_BASE;
1248 
1249         /*
1250          * The default values of VALLOC_BASE and SEGKPM_BASE should work
1251          * for values of physmax up to 256GB (1/4 TB). They need adjusting when
1252          * memory is at addresses above 256GB. When adjusted, segkpm_base must
1253          * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte).
1254          *
1255          * In the general case (>256GB), we use (4 * physmem) for the
1256          * kernel's virtual addresses, which is divided approximately
1257          * as follows:
1258          *  - 1 * physmem for segkpm
1259          *  - 1.5 * physmem for segzio
1260          *  - 1.5 * physmem for heap
1261          * Total: 4.0 * physmem
1262          *
1263          * Note that the segzio and heap sizes are more than physmem so that
1264          * VA fragmentation does not prevent either of them from being
1265          * able to use nearly all of physmem.  The value of 1.5x is determined
1266          * experimentally and may need to change if the workload changes.
1267          */
1268         if (physmax + 1 > mmu_btop(TERABYTE / 4) ||
1269             plat_dr_physmax > mmu_btop(TERABYTE / 4)) {
1270                 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1);
1271 
1272                 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) {
1273                         kpm_resv_amount = mmu_ptob(plat_dr_physmax);
1274                 }
1275 
1276                 /*
1277                  * This is what actually controls the KVA : UVA split.
1278                  * The kernel uses high VA, and this is lowering the
1279                  * boundary, thus increasing the amount of VA for the kernel.
1280                  * This gives the kernel 4 * (amount of physical memory) VA.
1281                  *
1282                  * The maximum VA is UINT64_MAX and we are using
1283                  * 64-bit 2's complement math, so e.g. if you have 512GB
1284                  * of memory, segkpm_base = -(4 * 512GB) == -2TB ==
1285                  * UINT64_MAX - 2TB (approximately).  So the kernel's
1286                  * VA is [UINT64_MAX-2TB to UINT64_MAX].
1287                  */
1288                 segkpm_base = -(P2ROUNDUP((4 * kpm_resv_amount),
1289                     KERNEL_REDZONE_SIZE));
1290 
1291                 /* make sure we leave some space for user apps above hole */
1292                 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE);
1293                 if (segkpm_base > SEGKPM_BASE)
1294                         segkpm_base = SEGKPM_BASE;
1295                 PRM_DEBUG(segkpm_base);
1296 
1297                 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG);
1298                 if (valloc_base < segkpm_base)
1299                         panic("not enough kernel VA to support memory size");
1300                 PRM_DEBUG(valloc_base);
1301         }
1302 #else   /* __i386 */
1303         valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz);
1304         valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]);
1305         PRM_DEBUG(valloc_base);
1306 #endif  /* __i386 */
1307 
1308         /*
1309          * do all the initial allocations
1310          */
1311         perform_allocations();
1312 
1313         /*
1314          * Build phys_install and phys_avail in kernel memspace.
1315          * - phys_install should be all memory in the system.
1316          * - phys_avail is phys_install minus any memory mapped before this
1317          *    point above KERNEL_TEXT.
1318          */
1319         current = phys_install = memlist;
1320         copy_memlist_filter(bootops->boot_mem->physinstalled, &current, NULL);
1321         if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1322                 panic("physinstalled was too big!");
1323         if (prom_debug)
1324                 print_memlist("phys_install", phys_install);
1325 
1326         phys_avail = current;
1327         PRM_POINT("Building phys_avail:\n");
1328         copy_memlist_filter(bootops->boot_mem->physinstalled, &current,
1329             avail_filter);
1330         if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1331                 panic("physavail was too big!");
1332         if (prom_debug)
1333                 print_memlist("phys_avail", phys_avail);
1334 #ifndef __xpv
1335         /*
1336          * Free unused memlist items, which may be used by memory DR driver
1337          * at runtime.
1338          */
1339         if ((caddr_t)current < (caddr_t)memlist + memlist_sz) {
1340                 memlist_free_block((caddr_t)current,
1341                     (caddr_t)memlist + memlist_sz - (caddr_t)current);
1342         }
1343 #endif
1344 
1345         /*
1346          * Build bios reserved memspace
1347          */
1348         current = bios_rsvd;
1349         copy_memlist_filter(bootops->boot_mem->rsvdmem, &current, NULL);
1350         if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz)
1351                 panic("bios_rsvd was too big!");
1352         if (prom_debug)
1353                 print_memlist("bios_rsvd", bios_rsvd);
1354 #ifndef __xpv
1355         /*
1356          * Free unused memlist items, which may be used by memory DR driver
1357          * at runtime.
1358          */
1359         if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) {
1360                 memlist_free_block((caddr_t)current,
1361                     (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current);
1362         }
1363 #endif
1364 
1365         /*
1366          * setup page coloring
1367          */
1368         page_coloring_setup(pagecolor_mem);
1369         page_lock_init();       /* currently a no-op */
1370 
1371         /*
1372          * free page list counters
1373          */
1374         (void) page_ctrs_alloc(page_ctrs_mem);
1375 
1376         /*
1377          * Size the pcf array based on the number of cpus in the box at
1378          * boot time.
1379          */
1380 
1381         pcf_init();
1382 
1383         /*
1384          * Initialize the page structures from the memory lists.
1385          */
1386         availrmem_initial = availrmem = freemem = 0;
1387         PRM_POINT("Calling kphysm_init()...");
1388         npages = kphysm_init(pp_base, npages);
1389         PRM_POINT("kphysm_init() done");
1390         PRM_DEBUG(npages);
1391 
1392         init_debug_info();
1393 
1394         /*
1395          * Now that page_t's have been initialized, remove all the
1396          * initial allocation pages from the kernel free page lists.
1397          */
1398         boot_mapin((caddr_t)valloc_base, valloc_sz);
1399         boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE);
1400         PRM_POINT("startup_memlist() done");
1401 
1402         PRM_DEBUG(valloc_sz);
1403 
1404 #if defined(__amd64)
1405         if ((availrmem >> (30 - MMU_PAGESHIFT)) >=
1406             textrepl_min_gb && l2cache_sz <= 2 << 20) {
1407                 extern size_t textrepl_size_thresh;
1408                 textrepl_size_thresh = (16 << 20) - 1;
1409         }
1410 #endif
1411 }
1412 
1413 /*
1414  * Layout the kernel's part of address space and initialize kmem allocator.
1415  */
1416 static void
1417 startup_kmem(void)
1418 {
1419         extern void page_set_colorequiv_arr(void);
1420 
1421         PRM_POINT("startup_kmem() starting...");
1422 
1423 #if defined(__amd64)
1424         if (eprom_kernelbase && eprom_kernelbase != KERNELBASE)
1425                 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit "
1426                     "systems.");
1427         kernelbase = segkpm_base - KERNEL_REDZONE_SIZE;
1428         core_base = (uintptr_t)COREHEAP_BASE;
1429         core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE;
1430 #else   /* __i386 */
1431         /*
1432          * We configure kernelbase based on:
1433          *
1434          * 1. user specified kernelbase via eeprom command. Value cannot exceed
1435          *    KERNELBASE_MAX. we large page align eprom_kernelbase
1436          *
1437          * 2. Default to KERNELBASE and adjust to 2X less the size for page_t.
1438          *    On large memory systems we must lower kernelbase to allow
1439          *    enough room for page_t's for all of memory.
1440          *
1441          * The value set here, might be changed a little later.
1442          */
1443         if (eprom_kernelbase) {
1444                 kernelbase = eprom_kernelbase & mmu.level_mask[1];
1445                 if (kernelbase > KERNELBASE_MAX)
1446                         kernelbase = KERNELBASE_MAX;
1447         } else {
1448                 kernelbase = (uintptr_t)KERNELBASE;
1449                 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz);
1450         }
1451         ASSERT((kernelbase & mmu.level_offset[1]) == 0);
1452         core_base = valloc_base;
1453         core_size = 0;
1454 #endif  /* __i386 */
1455 
1456         PRM_DEBUG(core_base);
1457         PRM_DEBUG(core_size);
1458         PRM_DEBUG(kernelbase);
1459 
1460 #if defined(__i386)
1461         segkp_fromheap = 1;
1462 #endif  /* __i386 */
1463 
1464         ekernelheap = (char *)core_base;
1465         PRM_DEBUG(ekernelheap);
1466 
1467         /*
1468          * Now that we know the real value of kernelbase,
1469          * update variables that were initialized with a value of
1470          * KERNELBASE (in common/conf/param.c).
1471          *
1472          * XXX  The problem with this sort of hackery is that the
1473          *      compiler just may feel like putting the const declarations
1474          *      (in param.c) into the .text section.  Perhaps they should
1475          *      just be declared as variables there?
1476          */
1477 
1478         *(uintptr_t *)&_kernelbase = kernelbase;
1479         *(uintptr_t *)&_userlimit = kernelbase;
1480 #if defined(__amd64)
1481         *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT;
1482 #else
1483         *(uintptr_t *)&_userlimit32 = _userlimit;
1484 #endif
1485         PRM_DEBUG(_kernelbase);
1486         PRM_DEBUG(_userlimit);
1487         PRM_DEBUG(_userlimit32);
1488 
1489         layout_kernel_va();
1490 
1491 #if defined(__i386)
1492         /*
1493          * If segmap is too large we can push the bottom of the kernel heap
1494          * higher than the base.  Or worse, it could exceed the top of the
1495          * VA space entirely, causing it to wrap around.
1496          */
1497         if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase)
1498                 panic("too little address space available for kernelheap,"
1499                     " use eeprom for lower kernelbase or smaller segmapsize");
1500 #endif  /* __i386 */
1501 
1502         /*
1503          * Initialize the kernel heap. Note 3rd argument must be > 1st.
1504          */
1505         kernelheap_init(kernelheap, ekernelheap,
1506             kernelheap + MMU_PAGESIZE,
1507             (void *)core_base, (void *)(core_base + core_size));
1508 
1509 #if defined(__xpv)
1510         /*
1511          * Link pending events struct into cpu struct
1512          */
1513         CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data;
1514 #endif
1515         /*
1516          * Initialize kernel memory allocator.
1517          */
1518         kmem_init();
1519 
1520         /*
1521          * Factor in colorequiv to check additional 'equivalent' bins
1522          */
1523         page_set_colorequiv_arr();
1524 
1525         /*
1526          * print this out early so that we know what's going on
1527          */
1528         print_x86_featureset(x86_featureset);
1529 
1530         /*
1531          * Initialize bp_mapin().
1532          */
1533         bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK);
1534 
1535         /*
1536          * orig_npages is non-zero if physmem has been configured for less
1537          * than the available memory.
1538          */
1539         if (orig_npages) {
1540                 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages",
1541                     (npages == PHYSMEM ? "Due to virtual address space " : ""),
1542                     npages, orig_npages);
1543         }
1544 #if defined(__i386)
1545         if (eprom_kernelbase && (eprom_kernelbase != kernelbase))
1546                 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, "
1547                     "System using 0x%lx",
1548                     (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase);
1549 #endif
1550 
1551 #ifdef  KERNELBASE_ABI_MIN
1552         if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) {
1553                 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not "
1554                     "i386 ABI compliant.", (uintptr_t)kernelbase);
1555         }
1556 #endif
1557 
1558 #ifndef __xpv
1559         if (plat_dr_support_memory()) {
1560                 mem_config_init();
1561         }
1562 #else   /* __xpv */
1563         /*
1564          * Some of the xen start information has to be relocated up
1565          * into the kernel's permanent address space.
1566          */
1567         PRM_POINT("calling xen_relocate_start_info()");
1568         xen_relocate_start_info();
1569         PRM_POINT("xen_relocate_start_info() done");
1570 
1571         /*
1572          * (Update the vcpu pointer in our cpu structure to point into
1573          * the relocated shared info.)
1574          */
1575         CPU->cpu_m.mcpu_vcpu_info =
1576             &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id];
1577 #endif  /* __xpv */
1578 
1579         PRM_POINT("startup_kmem() done");
1580 }
1581 
1582 #ifndef __xpv
1583 /*
1584  * If we have detected that we are running in an HVM environment, we need
1585  * to prepend the PV driver directory to the module search path.
1586  */
1587 #define HVM_MOD_DIR "/platform/i86hvm/kernel"
1588 static void
1589 update_default_path()
1590 {
1591         char *current, *newpath;
1592         int newlen;
1593 
1594         /*
1595          * We are about to resync with krtld.  krtld will reset its
1596          * internal module search path iff Solaris has set default_path.
1597          * We want to be sure we're prepending this new directory to the
1598          * right search path.
1599          */
1600         current = (default_path == NULL) ? kobj_module_path : default_path;
1601 
1602         newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2;
1603         newpath = kmem_alloc(newlen, KM_SLEEP);
1604         (void) strcpy(newpath, HVM_MOD_DIR);
1605         (void) strcat(newpath, " ");
1606         (void) strcat(newpath, current);
1607 
1608         default_path = newpath;
1609 }
1610 #endif
1611 
1612 static void
1613 startup_modules(void)
1614 {
1615         int cnt;
1616         extern void prom_setup(void);
1617         int32_t v, h;
1618         char d[11];
1619         char *cp;
1620         cmi_hdl_t hdl;
1621 
1622         PRM_POINT("startup_modules() starting...");
1623 
1624 #ifndef __xpv
1625         /*
1626          * Initialize ten-micro second timer so that drivers will
1627          * not get short changed in their init phase. This was
1628          * not getting called until clkinit which, on fast cpu's
1629          * caused the drv_usecwait to be way too short.
1630          */
1631         microfind();
1632 
1633         if ((get_hwenv() & HW_XEN_HVM) != 0)
1634                 update_default_path();
1635 #endif
1636 
1637         /*
1638          * Read the GMT lag from /etc/rtc_config.
1639          */
1640         sgmtl(process_rtc_config_file());
1641 
1642         /*
1643          * Calculate default settings of system parameters based upon
1644          * maxusers, yet allow to be overridden via the /etc/system file.
1645          */
1646         param_calc(0);
1647 
1648         mod_setup();
1649 
1650         /*
1651          * Initialize system parameters.
1652          */
1653         param_init();
1654 
1655         /*
1656          * Initialize the default brands
1657          */
1658         brand_init();
1659 
1660         /*
1661          * maxmem is the amount of physical memory we're playing with.
1662          */
1663         maxmem = physmem;
1664 
1665         /*
1666          * Initialize segment management stuff.
1667          */
1668         seg_init();
1669 
1670         if (modload("fs", "specfs") == -1)
1671                 halt("Can't load specfs");
1672 
1673         if (modload("fs", "devfs") == -1)
1674                 halt("Can't load devfs");
1675 
1676         if (modload("fs", "dev") == -1)
1677                 halt("Can't load dev");
1678 
1679         if (modload("fs", "procfs") == -1)
1680                 halt("Can't load procfs");
1681 
1682         (void) modloadonly("sys", "lbl_edition");
1683 
1684         dispinit();
1685 
1686         /* Read cluster configuration data. */
1687         clconf_init();
1688 
1689 #if defined(__xpv)
1690         (void) ec_init();
1691         gnttab_init();
1692         (void) xs_early_init();
1693 #endif /* __xpv */
1694 
1695         /*
1696          * Create a kernel device tree. First, create rootnex and
1697          * then invoke bus specific code to probe devices.
1698          */
1699         setup_ddi();
1700 
1701 #ifdef __xpv
1702         if (DOMAIN_IS_INITDOMAIN(xen_info))
1703 #endif
1704         {
1705                 id_t smid;
1706                 smbios_system_t smsys;
1707                 smbios_info_t sminfo;
1708                 char *mfg;
1709                 /*
1710                  * Load the System Management BIOS into the global ksmbios
1711                  * handle, if an SMBIOS is present on this system.
1712                  * Also set "si-hw-provider" property, if not already set.
1713                  */
1714                 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL);
1715                 if (ksmbios != NULL &&
1716                     ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) &&
1717                     (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) {
1718                         mfg = (char *)sminfo.smbi_manufacturer;
1719                         if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) {
1720                                 extern char hw_provider[];
1721                                 int i;
1722                                 for (i = 0; i < SYS_NMLN; i++) {
1723                                         if (isprint(mfg[i]))
1724                                                 hw_provider[i] = mfg[i];
1725                                         else {
1726                                                 hw_provider[i] = '\0';
1727                                                 break;
1728                                         }
1729                                 }
1730                                 hw_provider[SYS_NMLN - 1] = '\0';
1731                         }
1732                 }
1733         }
1734 
1735 
1736         /*
1737          * Originally clconf_init() apparently needed the hostid.  But
1738          * this no longer appears to be true - it uses its own nodeid.
1739          * By placing the hostid logic here, we are able to make use of
1740          * the SMBIOS UUID.
1741          */
1742         if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) {
1743                 cmn_err(CE_WARN, "Unable to set hostid");
1744         } else {
1745                 for (v = h, cnt = 0; cnt < 10; cnt++) {
1746                         d[cnt] = (char)(v % 10);
1747                         v /= 10;
1748                         if (v == 0)
1749                                 break;
1750                 }
1751                 for (cp = hw_serial; cnt >= 0; cnt--)
1752                         *cp++ = d[cnt] + '0';
1753                 *cp = 0;
1754         }
1755 
1756         /*
1757          * Set up the CPU module subsystem for the boot cpu in the native
1758          * case, and all physical cpu resource in the xpv dom0 case.
1759          * Modifies the device tree, so this must be done after
1760          * setup_ddi().
1761          */
1762 #ifdef __xpv
1763         /*
1764          * If paravirtualized and on dom0 then we initialize all physical
1765          * cpu handles now;  if paravirtualized on a domU then do not
1766          * initialize.
1767          */
1768         if (DOMAIN_IS_INITDOMAIN(xen_info)) {
1769                 xen_mc_lcpu_cookie_t cpi;
1770 
1771                 for (cpi = xen_physcpu_next(NULL); cpi != NULL;
1772                     cpi = xen_physcpu_next(cpi)) {
1773                         if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA,
1774                             xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi),
1775                             xen_physcpu_strandid(cpi))) != NULL &&
1776                             is_x86_feature(x86_featureset, X86FSET_MCA))
1777                                 cmi_mca_init(hdl);
1778                 }
1779         }
1780 #else
1781         /*
1782          * Initialize a handle for the boot cpu - others will initialize
1783          * as they startup.
1784          */
1785         if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
1786             cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL) {
1787                 if (is_x86_feature(x86_featureset, X86FSET_MCA))
1788                         cmi_mca_init(hdl);
1789                 CPU->cpu_m.mcpu_cmi_hdl = hdl;
1790         }
1791 #endif  /* __xpv */
1792 
1793         /*
1794          * Fake a prom tree such that /dev/openprom continues to work
1795          */
1796         PRM_POINT("startup_modules: calling prom_setup...");
1797         prom_setup();
1798         PRM_POINT("startup_modules: done");
1799 
1800         /*
1801          * Load all platform specific modules
1802          */
1803         PRM_POINT("startup_modules: calling psm_modload...");
1804         psm_modload();
1805 
1806         PRM_POINT("startup_modules() done");
1807 }
1808 
1809 /*
1810  * claim a "setaside" boot page for use in the kernel
1811  */
1812 page_t *
1813 boot_claim_page(pfn_t pfn)
1814 {
1815         page_t *pp;
1816 
1817         pp = page_numtopp_nolock(pfn);
1818         ASSERT(pp != NULL);
1819 
1820         if (PP_ISBOOTPAGES(pp)) {
1821                 if (pp->p_next != NULL)
1822                         pp->p_next->p_prev = pp->p_prev;
1823                 if (pp->p_prev == NULL)
1824                         bootpages = pp->p_next;
1825                 else
1826                         pp->p_prev->p_next = pp->p_next;
1827         } else {
1828                 /*
1829                  * htable_attach() expects a base pagesize page
1830                  */
1831                 if (pp->p_szc != 0)
1832                         page_boot_demote(pp);
1833                 pp = page_numtopp(pfn, SE_EXCL);
1834         }
1835         return (pp);
1836 }
1837 
1838 /*
1839  * Walk through the pagetables looking for pages mapped in by boot.  If the
1840  * setaside flag is set the pages are expected to be returned to the
1841  * kernel later in boot, so we add them to the bootpages list.
1842  */
1843 static void
1844 protect_boot_range(uintptr_t low, uintptr_t high, int setaside)
1845 {
1846         uintptr_t va = low;
1847         size_t len;
1848         uint_t prot;
1849         pfn_t pfn;
1850         page_t *pp;
1851         pgcnt_t boot_protect_cnt = 0;
1852 
1853         while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) {
1854                 if (va + len >= high)
1855                         panic("0x%lx byte mapping at 0x%p exceeds boot's "
1856                             "legal range.", len, (void *)va);
1857 
1858                 while (len > 0) {
1859                         pp = page_numtopp_alloc(pfn);
1860                         if (pp != NULL) {
1861                                 if (setaside == 0)
1862                                         panic("Unexpected mapping by boot.  "
1863                                             "addr=%p pfn=%lx\n",
1864                                             (void *)va, pfn);
1865 
1866                                 pp->p_next = bootpages;
1867                                 pp->p_prev = NULL;
1868                                 PP_SETBOOTPAGES(pp);
1869                                 if (bootpages != NULL) {
1870                                         bootpages->p_prev = pp;
1871                                 }
1872                                 bootpages = pp;
1873                                 ++boot_protect_cnt;
1874                         }
1875 
1876                         ++pfn;
1877                         len -= MMU_PAGESIZE;
1878                         va += MMU_PAGESIZE;
1879                 }
1880         }
1881         PRM_DEBUG(boot_protect_cnt);
1882 }
1883 
1884 /*
1885  *
1886  */
1887 static void
1888 layout_kernel_va(void)
1889 {
1890         PRM_POINT("layout_kernel_va() starting...");
1891         /*
1892          * Establish the final size of the kernel's heap, size of segmap,
1893          * segkp, etc.
1894          */
1895 
1896 #if defined(__amd64)
1897 
1898         kpm_vbase = (caddr_t)segkpm_base;
1899         if (physmax + 1 < plat_dr_physmax) {
1900                 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax));
1901         } else {
1902                 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1));
1903         }
1904         if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base)
1905                 panic("not enough room for kpm!");
1906         PRM_DEBUG(kpm_size);
1907         PRM_DEBUG(kpm_vbase);
1908 
1909         /*
1910          * By default we create a seg_kp in 64 bit kernels, it's a little
1911          * faster to access than embedding it in the heap.
1912          */
1913         segkp_base = (caddr_t)valloc_base + valloc_sz;
1914         if (!segkp_fromheap) {
1915                 size_t sz = mmu_ptob(segkpsize);
1916 
1917                 /*
1918                  * determine size of segkp
1919                  */
1920                 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) {
1921                         sz = SEGKPDEFSIZE;
1922                         cmn_err(CE_WARN, "!Illegal value for segkpsize. "
1923                             "segkpsize has been reset to %ld pages",
1924                             mmu_btop(sz));
1925                 }
1926                 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem)));
1927 
1928                 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz));
1929         }
1930         PRM_DEBUG(segkp_base);
1931         PRM_DEBUG(segkpsize);
1932 
1933         /*
1934          * segzio is used for ZFS cached data. It uses a distinct VA
1935          * segment (from kernel heap) so that we can easily tell not to
1936          * include it in kernel crash dumps on 64 bit kernels. The trick is
1937          * to give it lots of VA, but not constrain the kernel heap.
1938          * We can use 1.5x physmem for segzio, leaving approximately
1939          * another 1.5x physmem for heap.  See also the comment in
1940          * startup_memlist().
1941          */
1942         segzio_base = segkp_base + mmu_ptob(segkpsize);
1943         if (segzio_fromheap) {
1944                 segziosize = 0;
1945         } else {
1946                 size_t physmem_size = mmu_ptob(physmem);
1947                 size_t size = (segziosize == 0) ?
1948                     physmem_size * 3 / 2 : mmu_ptob(segziosize);
1949 
1950                 if (size < SEGZIOMINSIZE)
1951                         size = SEGZIOMINSIZE;
1952                 segziosize = mmu_btop(ROUND_UP_LPAGE(size));
1953         }
1954         PRM_DEBUG(segziosize);
1955         PRM_DEBUG(segzio_base);
1956 
1957         /*
1958          * Put the range of VA for device mappings next, kmdb knows to not
1959          * grep in this range of addresses.
1960          */
1961         toxic_addr =
1962             ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize));
1963         PRM_DEBUG(toxic_addr);
1964         segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size);
1965 #else /* __i386 */
1966         segmap_start = ROUND_UP_LPAGE(kernelbase);
1967 #endif /* __i386 */
1968         PRM_DEBUG(segmap_start);
1969 
1970         /*
1971          * Users can change segmapsize through eeprom. If the variable
1972          * is tuned through eeprom, there is no upper bound on the
1973          * size of segmap.
1974          */
1975         segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT);
1976 
1977 #if defined(__i386)
1978         /*
1979          * 32-bit systems don't have segkpm or segkp, so segmap appears at
1980          * the bottom of the kernel's address range.  Set aside space for a
1981          * small red zone just below the start of segmap.
1982          */
1983         segmap_start += KERNEL_REDZONE_SIZE;
1984         segmapsize -= KERNEL_REDZONE_SIZE;
1985 #endif
1986 
1987         PRM_DEBUG(segmap_start);
1988         PRM_DEBUG(segmapsize);
1989         kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize);
1990         PRM_DEBUG(kernelheap);
1991         PRM_POINT("layout_kernel_va() done...");
1992 }
1993 
1994 /*
1995  * Finish initializing the VM system, now that we are no longer
1996  * relying on the boot time memory allocators.
1997  */
1998 static void
1999 startup_vm(void)
2000 {
2001         struct segmap_crargs a;
2002 
2003         extern int use_brk_lpg, use_stk_lpg;
2004 
2005         PRM_POINT("startup_vm() starting...");
2006 
2007         /*
2008          * Initialize the hat layer.
2009          */
2010         hat_init();
2011 
2012         /*
2013          * Do final allocations of HAT data structures that need to
2014          * be allocated before quiescing the boot loader.
2015          */
2016         PRM_POINT("Calling hat_kern_alloc()...");
2017         hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap);
2018         PRM_POINT("hat_kern_alloc() done");
2019 
2020 #ifndef __xpv
2021         /*
2022          * Setup Page Attribute Table
2023          */
2024         pat_sync();
2025 #endif
2026 
2027         /*
2028          * The next two loops are done in distinct steps in order
2029          * to be sure that any page that is doubly mapped (both above
2030          * KERNEL_TEXT and below kernelbase) is dealt with correctly.
2031          * Note this may never happen, but it might someday.
2032          */
2033         bootpages = NULL;
2034         PRM_POINT("Protecting boot pages");
2035 
2036         /*
2037          * Protect any pages mapped above KERNEL_TEXT that somehow have
2038          * page_t's. This can only happen if something weird allocated
2039          * in this range (like kadb/kmdb).
2040          */
2041         protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0);
2042 
2043         /*
2044          * Before we can take over memory allocation/mapping from the boot
2045          * loader we must remove from our free page lists any boot allocated
2046          * pages that stay mapped until release_bootstrap().
2047          */
2048         protect_boot_range(0, kernelbase, 1);
2049 
2050         /*
2051          * Switch to running on regular HAT (not boot_mmu)
2052          */
2053         PRM_POINT("Calling hat_kern_setup()...");
2054         hat_kern_setup();
2055 
2056         /*
2057          * It is no longer safe to call BOP_ALLOC(), so make sure we don't.
2058          */
2059         bop_no_more_mem();
2060 
2061         PRM_POINT("hat_kern_setup() done");
2062 
2063         hat_cpu_online(CPU);
2064 
2065         /*
2066          * Initialize VM system
2067          */
2068         PRM_POINT("Calling kvm_init()...");
2069         kvm_init();
2070         PRM_POINT("kvm_init() done");
2071 
2072         /*
2073          * Tell kmdb that the VM system is now working
2074          */
2075         if (boothowto & RB_DEBUG)
2076                 kdi_dvec_vmready();
2077 
2078 #if defined(__xpv)
2079         /*
2080          * Populate the I/O pool on domain 0
2081          */
2082         if (DOMAIN_IS_INITDOMAIN(xen_info)) {
2083                 extern long populate_io_pool(void);
2084                 long init_io_pool_cnt;
2085 
2086                 PRM_POINT("Populating reserve I/O page pool");
2087                 init_io_pool_cnt = populate_io_pool();
2088                 PRM_DEBUG(init_io_pool_cnt);
2089         }
2090 #endif
2091         /*
2092          * Mangle the brand string etc.
2093          */
2094         cpuid_pass3(CPU);
2095 
2096 #if defined(__amd64)
2097 
2098         /*
2099          * Create the device arena for toxic (to dtrace/kmdb) mappings.
2100          */
2101         device_arena = vmem_create("device", (void *)toxic_addr,
2102             toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP);
2103 
2104 #else   /* __i386 */
2105 
2106         /*
2107          * allocate the bit map that tracks toxic pages
2108          */
2109         toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase));
2110         PRM_DEBUG(toxic_bit_map_len);
2111         toxic_bit_map =
2112             kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP);
2113         ASSERT(toxic_bit_map != NULL);
2114         PRM_DEBUG(toxic_bit_map);
2115 
2116 #endif  /* __i386 */
2117 
2118 
2119         /*
2120          * Now that we've got more VA, as well as the ability to allocate from
2121          * it, tell the debugger.
2122          */
2123         if (boothowto & RB_DEBUG)
2124                 kdi_dvec_memavail();
2125 
2126         /*
2127          * The following code installs a special page fault handler (#pf)
2128          * to work around a pentium bug.
2129          */
2130 #if !defined(__amd64) && !defined(__xpv)
2131         if (x86_type == X86_TYPE_P5) {
2132                 desctbr_t idtr;
2133                 gate_desc_t *newidt;
2134 
2135                 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL)
2136                         panic("failed to install pentium_pftrap");
2137 
2138                 bcopy(idt0, newidt, NIDT * sizeof (*idt0));
2139                 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap,
2140                     KCS_SEL, SDT_SYSIGT, TRP_KPL, 0);
2141 
2142                 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE,
2143                     PROT_READ | PROT_EXEC);
2144 
2145                 CPU->cpu_idt = newidt;
2146                 idtr.dtr_base = (uintptr_t)CPU->cpu_idt;
2147                 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1;
2148                 wr_idtr(&idtr);
2149         }
2150 #endif  /* !__amd64 */
2151 
2152 #if !defined(__xpv)
2153         /*
2154          * Map page pfn=0 for drivers, such as kd, that need to pick up
2155          * parameters left there by controllers/BIOS.
2156          */
2157         PRM_POINT("setup up p0_va");
2158         p0_va = i86devmap(0, 1, PROT_READ);
2159         PRM_DEBUG(p0_va);
2160 #endif
2161 
2162         cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n",
2163             physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled));
2164 
2165         /*
2166          * disable automatic large pages for small memory systems or
2167          * when the disable flag is set.
2168          *
2169          * Do not yet consider page sizes larger than 2m/4m.
2170          */
2171         if (!auto_lpg_disable && mmu.max_page_level > 0) {
2172                 max_uheap_lpsize = LEVEL_SIZE(1);
2173                 max_ustack_lpsize = LEVEL_SIZE(1);
2174                 max_privmap_lpsize = LEVEL_SIZE(1);
2175                 max_uidata_lpsize = LEVEL_SIZE(1);
2176                 max_utext_lpsize = LEVEL_SIZE(1);
2177                 max_shm_lpsize = LEVEL_SIZE(1);
2178         }
2179         if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 ||
2180             auto_lpg_disable) {
2181                 use_brk_lpg = 0;
2182                 use_stk_lpg = 0;
2183         }
2184         mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level);
2185 
2186         PRM_POINT("Calling hat_init_finish()...");
2187         hat_init_finish();
2188         PRM_POINT("hat_init_finish() done");
2189 
2190         /*
2191          * Initialize the segkp segment type.
2192          */
2193         rw_enter(&kas.a_lock, RW_WRITER);
2194         PRM_POINT("Attaching segkp");
2195         if (segkp_fromheap) {
2196                 segkp->s_as = &kas;
2197         } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize),
2198             segkp) < 0) {
2199                 panic("startup: cannot attach segkp");
2200                 /*NOTREACHED*/
2201         }
2202         PRM_POINT("Doing segkp_create()");
2203         if (segkp_create(segkp) != 0) {
2204                 panic("startup: segkp_create failed");
2205                 /*NOTREACHED*/
2206         }
2207         PRM_DEBUG(segkp);
2208         rw_exit(&kas.a_lock);
2209 
2210         /*
2211          * kpm segment
2212          */
2213         segmap_kpm = 0;
2214         if (kpm_desired) {
2215                 kpm_init();
2216                 kpm_enable = 1;
2217         }
2218 
2219         /*
2220          * Now create segmap segment.
2221          */
2222         rw_enter(&kas.a_lock, RW_WRITER);
2223         if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) {
2224                 panic("cannot attach segmap");
2225                 /*NOTREACHED*/
2226         }
2227         PRM_DEBUG(segmap);
2228 
2229         a.prot = PROT_READ | PROT_WRITE;
2230         a.shmsize = 0;
2231         a.nfreelist = segmapfreelists;
2232 
2233         if (segmap_create(segmap, (caddr_t)&a) != 0)
2234                 panic("segmap_create segmap");
2235         rw_exit(&kas.a_lock);
2236 
2237         setup_vaddr_for_ppcopy(CPU);
2238 
2239         segdev_init();
2240 #if defined(__xpv)
2241         if (DOMAIN_IS_INITDOMAIN(xen_info))
2242 #endif
2243                 pmem_init();
2244 
2245         PRM_POINT("startup_vm() done");
2246 }
2247 
2248 /*
2249  * Load a tod module for the non-standard tod part found on this system.
2250  */
2251 static void
2252 load_tod_module(char *todmod)
2253 {
2254         if (modload("tod", todmod) == -1)
2255                 halt("Can't load TOD module");
2256 }
2257 
2258 static void
2259 startup_end(void)
2260 {
2261         int i;
2262         extern void setx86isalist(void);
2263         extern void cpu_event_init(void);
2264 
2265         PRM_POINT("startup_end() starting...");
2266 
2267         /*
2268          * Perform tasks that get done after most of the VM
2269          * initialization has been done but before the clock
2270          * and other devices get started.
2271          */
2272         kern_setup1();
2273 
2274         /*
2275          * Perform CPC initialization for this CPU.
2276          */
2277         kcpc_hw_init(CPU);
2278 
2279         /*
2280          * Initialize cpu event framework.
2281          */
2282         cpu_event_init();
2283 
2284 #if defined(OPTERON_WORKAROUND_6323525)
2285         if (opteron_workaround_6323525)
2286                 patch_workaround_6323525();
2287 #endif
2288         /*
2289          * If needed, load TOD module now so that ddi_get_time(9F) etc. work
2290          * (For now, "needed" is defined as set tod_module_name in /etc/system)
2291          */
2292         if (tod_module_name != NULL) {
2293                 PRM_POINT("load_tod_module()");
2294                 load_tod_module(tod_module_name);
2295         }
2296 
2297 #if defined(__xpv)
2298         /*
2299          * Forceload interposing TOD module for the hypervisor.
2300          */
2301         PRM_POINT("load_tod_module()");
2302         load_tod_module("xpvtod");
2303 #endif
2304 
2305         /*
2306          * Configure the system.
2307          */
2308         PRM_POINT("Calling configure()...");
2309         configure();            /* set up devices */
2310         PRM_POINT("configure() done");
2311 
2312         /*
2313          * We can now setup for XSAVE because fpu_probe is done in configure().
2314          */
2315         if (fp_save_mech == FP_XSAVE) {
2316                 xsave_setup_msr(CPU);
2317         }
2318 
2319         /*
2320          * Set the isa_list string to the defined instruction sets we
2321          * support.
2322          */
2323         setx86isalist();
2324         cpu_intr_alloc(CPU, NINTR_THREADS);
2325         psm_install();
2326 
2327         /*
2328          * We're done with bootops.  We don't unmap the bootstrap yet because
2329          * we're still using bootsvcs.
2330          */
2331         PRM_POINT("NULLing out bootops");
2332         *bootopsp = (struct bootops *)NULL;
2333         bootops = (struct bootops *)NULL;
2334 
2335 #if defined(__xpv)
2336         ec_init_debug_irq();
2337         xs_domu_init();
2338 #endif
2339 
2340 #if defined(__amd64) && !defined(__xpv)
2341         /*
2342          * Intel IOMMU has been setup/initialized in ddi_impl.c
2343          * Start it up now.
2344          */
2345         immu_startup();
2346 #endif
2347 
2348         PRM_POINT("Enabling interrupts");
2349         (*picinitf)();
2350         sti();
2351 #if defined(__xpv)
2352         ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0);
2353         xen_late_startup();
2354 #endif
2355 
2356         (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1,
2357             "softlevel1", NULL, NULL); /* XXX to be moved later */
2358 
2359         /*
2360          * Register software interrupt handlers for ddi_periodic_add(9F).
2361          * Software interrupts up to the level 10 are supported.
2362          */
2363         for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) {
2364                 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i,
2365                     (avfunc)ddi_periodic_softintr, "ddi_periodic",
2366                     (caddr_t)(uintptr_t)i, NULL);
2367         }
2368 
2369 #if !defined(__xpv)
2370         if (modload("drv", "amd_iommu") < 0) {
2371                 PRM_POINT("No AMD IOMMU present\n");
2372         } else if (ddi_hold_installed_driver(ddi_name_to_major(
2373             "amd_iommu")) == NULL) {
2374                 prom_printf("ERROR: failed to attach AMD IOMMU\n");
2375         }
2376 #endif
2377         post_startup_cpu_fixups();
2378 
2379         PRM_POINT("startup_end() done");
2380 }
2381 
2382 /*
2383  * Don't remove the following 2 variables.  They are necessary
2384  * for reading the hostid from the legacy file (/kernel/misc/sysinit).
2385  */
2386 char *_hs1107 = hw_serial;
2387 ulong_t  _bdhs34;
2388 
2389 void
2390 post_startup(void)
2391 {
2392         extern void cpupm_init(cpu_t *);
2393         extern void cpu_event_init_cpu(cpu_t *);
2394 
2395         /*
2396          * Set the system wide, processor-specific flags to be passed
2397          * to userland via the aux vector for performance hints and
2398          * instruction set extensions.
2399          */
2400         bind_hwcap();
2401 
2402 #ifdef __xpv
2403         if (DOMAIN_IS_INITDOMAIN(xen_info))
2404 #endif
2405         {
2406 #if defined(__xpv)
2407                 xpv_panic_init();
2408 #else
2409                 /*
2410                  * Startup the memory scrubber.
2411                  * XXPV This should be running somewhere ..
2412                  */
2413                 if ((get_hwenv() & HW_VIRTUAL) == 0)
2414                         memscrub_init();
2415 #endif
2416         }
2417 
2418         /*
2419          * Complete CPU module initialization
2420          */
2421         cmi_post_startup();
2422 
2423         /*
2424          * Perform forceloading tasks for /etc/system.
2425          */
2426         (void) mod_sysctl(SYS_FORCELOAD, NULL);
2427 
2428         /*
2429          * ON4.0: Force /proc module in until clock interrupt handle fixed
2430          * ON4.0: This must be fixed or restated in /etc/systems.
2431          */
2432         (void) modload("fs", "procfs");
2433 
2434         (void) i_ddi_attach_hw_nodes("pit_beep");
2435 
2436 #if defined(__i386)
2437         /*
2438          * Check for required functional Floating Point hardware,
2439          * unless FP hardware explicitly disabled.
2440          */
2441         if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO))
2442                 halt("No working FP hardware found");
2443 #endif
2444 
2445         maxmem = freemem;
2446 
2447         cpu_event_init_cpu(CPU);
2448         cpupm_init(CPU);
2449         (void) mach_cpu_create_device_node(CPU, NULL);
2450 
2451         pg_init();
2452 }
2453 
2454 static int
2455 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr)
2456 {
2457         return ((pp->p_pagenum >= btop(low_addr)) &&
2458             (pp->p_pagenum < btopr(high_addr)));
2459 }
2460 
2461 static int
2462 pp_in_module(page_t *pp, const rd_existing_t *modranges)
2463 {
2464         uint_t i;
2465 
2466         for (i = 0; modranges[i].phys != 0; i++) {
2467                 if (pp_in_range(pp, modranges[i].phys,
2468                     modranges[i].phys + modranges[i].size))
2469                         return (1);
2470         }
2471 
2472         return (0);
2473 }
2474 
2475 void
2476 release_bootstrap(void)
2477 {
2478         int root_is_ramdisk;
2479         page_t *pp;
2480         extern void kobj_boot_unmountroot(void);
2481         extern dev_t rootdev;
2482         uint_t i;
2483         char propname[32];
2484         rd_existing_t *modranges;
2485 #if !defined(__xpv)
2486         pfn_t   pfn;
2487 #endif
2488 
2489         /*
2490          * Save the bootfs module ranges so that we can reserve them below
2491          * for the real bootfs.
2492          */
2493         modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES,
2494             KM_SLEEP);
2495         for (i = 0; ; i++) {
2496                 uint64_t start, size;
2497 
2498                 modranges[i].phys = 0;
2499 
2500                 (void) snprintf(propname, sizeof (propname),
2501                     "module-addr-%u", i);
2502                 if (do_bsys_getproplen(NULL, propname) <= 0)
2503                         break;
2504                 (void) do_bsys_getprop(NULL, propname, &start);
2505 
2506                 (void) snprintf(propname, sizeof (propname),
2507                     "module-size-%u", i);
2508                 if (do_bsys_getproplen(NULL, propname) <= 0)
2509                         break;
2510                 (void) do_bsys_getprop(NULL, propname, &size);
2511 
2512                 modranges[i].phys = start;
2513                 modranges[i].size = size;
2514         }
2515 
2516         /* unmount boot ramdisk and release kmem usage */
2517         kobj_boot_unmountroot();
2518 
2519         /*
2520          * We're finished using the boot loader so free its pages.
2521          */
2522         PRM_POINT("Unmapping lower boot pages");
2523 
2524         clear_boot_mappings(0, _userlimit);
2525 
2526 #if 0
2527         if (fb_info.paddr != 0 && fb_info.fb_type != FB_TYPE_EGA_TEXT) {
2528                 clear_boot_mappings(fb_info.paddr,
2529                     P2ROUNDUP(fb_info.paddr + fb_info.fb_size, MMU_PAGESIZE));
2530                 clear_boot_mappings((uintptr_t)fb_info.fb,
2531                     P2ROUNDUP((uintptr_t)fb_info.fb + fb_info.fb_size,
2532                     MMU_PAGESIZE));
2533         }
2534 #endif
2535 
2536         postbootkernelbase = kernelbase;
2537 
2538         /*
2539          * If root isn't on ramdisk, destroy the hardcoded
2540          * ramdisk node now and release the memory. Else,
2541          * ramdisk memory is kept in rd_pages.
2542          */
2543         root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk"));
2544         if (!root_is_ramdisk) {
2545                 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0);
2546                 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node());
2547                 ndi_rele_devi(dip);     /* held from ddi_find_devinfo */
2548                 (void) ddi_remove_child(dip, 0);
2549         }
2550 
2551         PRM_POINT("Releasing boot pages");
2552         while (bootpages) {
2553                 extern uint64_t ramdisk_start, ramdisk_end;
2554                 pp = bootpages;
2555                 bootpages = pp->p_next;
2556 
2557 
2558                 /* Keep pages for the lower 64K */
2559                 if (pp_in_range(pp, 0, 0x40000)) {
2560                         pp->p_next = lower_pages;
2561                         lower_pages = pp;
2562                         lower_pages_count++;
2563                         continue;
2564                 }
2565 
2566                 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start,
2567                     ramdisk_end) || pp_in_module(pp, modranges)) {
2568                         pp->p_next = rd_pages;
2569                         rd_pages = pp;
2570                         continue;
2571                 }
2572                 pp->p_next = (struct page *)0;
2573                 pp->p_prev = (struct page *)0;
2574                 PP_CLRBOOTPAGES(pp);
2575                 page_free(pp, 1);
2576         }
2577         PRM_POINT("Boot pages released");
2578 
2579         kmem_free(modranges, sizeof (rd_existing_t) * 99);
2580 
2581 #if !defined(__xpv)
2582 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */
2583         /*
2584          * Find 1 page below 1 MB so that other processors can boot up or
2585          * so that any processor can resume.
2586          * Make sure it has a kernel VA as well as a 1:1 mapping.
2587          * We should have just free'd one up.
2588          */
2589 
2590         /*
2591          * 0x10 pages is 64K.  Leave the bottom 64K alone
2592          * for BIOS.
2593          */
2594         for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) {
2595                 if (page_numtopp_alloc(pfn) == NULL)
2596                         continue;
2597                 rm_platter_va = i86devmap(pfn, 1,
2598                     PROT_READ | PROT_WRITE | PROT_EXEC);
2599                 rm_platter_pa = ptob(pfn);
2600                 break;
2601         }
2602         if (pfn == btop(1*1024*1024) && use_mp)
2603                 panic("No page below 1M available for starting "
2604                     "other processors or for resuming from system-suspend");
2605 #endif  /* !__xpv */
2606 }
2607 
2608 /*
2609  * Initialize the platform-specific parts of a page_t.
2610  */
2611 void
2612 add_physmem_cb(page_t *pp, pfn_t pnum)
2613 {
2614         pp->p_pagenum = pnum;
2615         pp->p_mapping = NULL;
2616         pp->p_embed = 0;
2617         pp->p_share = 0;
2618         pp->p_mlentry = 0;
2619 }
2620 
2621 /*
2622  * kphysm_init() initializes physical memory.
2623  */
2624 static pgcnt_t
2625 kphysm_init(
2626         page_t *pp,
2627         pgcnt_t npages)
2628 {
2629         struct memlist  *pmem;
2630         struct memseg   *cur_memseg;
2631         pfn_t           base_pfn;
2632         pfn_t           end_pfn;
2633         pgcnt_t         num;
2634         pgcnt_t         pages_done = 0;
2635         uint64_t        addr;
2636         uint64_t        size;
2637         extern pfn_t    ddiphysmin;
2638         extern int      mnode_xwa;
2639         int             ms = 0, me = 0;
2640 
2641         ASSERT(page_hash != NULL && page_hashsz != 0);
2642 
2643         cur_memseg = memseg_base;
2644         for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) {
2645                 /*
2646                  * In a 32 bit kernel can't use higher memory if we're
2647                  * not booting in PAE mode. This check takes care of that.
2648                  */
2649                 addr = pmem->ml_address;
2650                 size = pmem->ml_size;
2651                 if (btop(addr) > physmax)
2652                         continue;
2653 
2654                 /*
2655                  * align addr and size - they may not be at page boundaries
2656                  */
2657                 if ((addr & MMU_PAGEOFFSET) != 0) {
2658                         addr += MMU_PAGEOFFSET;
2659                         addr &= ~(uint64_t)MMU_PAGEOFFSET;
2660                         size -= addr - pmem->ml_address;
2661                 }
2662 
2663                 /* only process pages below or equal to physmax */
2664                 if ((btop(addr + size) - 1) > physmax)
2665                         size = ptob(physmax - btop(addr) + 1);
2666 
2667                 num = btop(size);
2668                 if (num == 0)
2669                         continue;
2670 
2671                 if (num > npages)
2672                         num = npages;
2673 
2674                 npages -= num;
2675                 pages_done += num;
2676                 base_pfn = btop(addr);
2677 
2678                 if (prom_debug)
2679                         prom_printf("MEMSEG addr=0x%" PRIx64
2680                             " pgs=0x%lx pfn 0x%lx-0x%lx\n",
2681                             addr, num, base_pfn, base_pfn + num);
2682 
2683                 /*
2684                  * Ignore pages below ddiphysmin to simplify ddi memory
2685                  * allocation with non-zero addr_lo requests.
2686                  */
2687                 if (base_pfn < ddiphysmin) {
2688                         if (base_pfn + num <= ddiphysmin)
2689                                 continue;
2690                         pp += (ddiphysmin - base_pfn);
2691                         num -= (ddiphysmin - base_pfn);
2692                         base_pfn = ddiphysmin;
2693                 }
2694 
2695                 /*
2696                  * mnode_xwa is greater than 1 when large pages regions can
2697                  * cross memory node boundaries. To prevent the formation
2698                  * of these large pages, configure the memsegs based on the
2699                  * memory node ranges which had been made non-contiguous.
2700                  */
2701                 if (mnode_xwa > 1) {
2702 
2703                         end_pfn = base_pfn + num - 1;
2704                         ms = PFN_2_MEM_NODE(base_pfn);
2705                         me = PFN_2_MEM_NODE(end_pfn);
2706 
2707                         if (ms != me) {
2708                                 /*
2709                                  * current range spans more than 1 memory node.
2710                                  * Set num to only the pfn range in the start
2711                                  * memory node.
2712                                  */
2713                                 num = mem_node_config[ms].physmax - base_pfn
2714                                     + 1;
2715                                 ASSERT(end_pfn > mem_node_config[ms].physmax);
2716                         }
2717                 }
2718 
2719                 for (;;) {
2720                         /*
2721                          * Build the memsegs entry
2722                          */
2723                         cur_memseg->pages = pp;
2724                         cur_memseg->epages = pp + num;
2725                         cur_memseg->pages_base = base_pfn;
2726                         cur_memseg->pages_end = base_pfn + num;
2727 
2728                         /*
2729                          * Insert into memseg list in decreasing pfn range
2730                          * order. Low memory is typically more fragmented such
2731                          * that this ordering keeps the larger ranges at the
2732                          * front of the list for code that searches memseg.
2733                          * This ASSERTS that the memsegs coming in from boot
2734                          * are in increasing physical address order and not
2735                          * contiguous.
2736                          */
2737                         if (memsegs != NULL) {
2738                                 ASSERT(cur_memseg->pages_base >=
2739                                     memsegs->pages_end);
2740                                 cur_memseg->next = memsegs;
2741                         }
2742                         memsegs = cur_memseg;
2743 
2744                         /*
2745                          * add_physmem() initializes the PSM part of the page
2746                          * struct by calling the PSM back with add_physmem_cb().
2747                          * In addition it coalesces pages into larger pages as
2748                          * it initializes them.
2749                          */
2750                         add_physmem(pp, num, base_pfn);
2751                         cur_memseg++;
2752                         availrmem_initial += num;
2753                         availrmem += num;
2754 
2755                         pp += num;
2756                         if (ms >= me)
2757                                 break;
2758 
2759                         /* process next memory node range */
2760                         ms++;
2761                         base_pfn = mem_node_config[ms].physbase;
2762                         num = MIN(mem_node_config[ms].physmax,
2763                             end_pfn) - base_pfn + 1;
2764                 }
2765         }
2766 
2767         PRM_DEBUG(availrmem_initial);
2768         PRM_DEBUG(availrmem);
2769         PRM_DEBUG(freemem);
2770         build_pfn_hash();
2771         return (pages_done);
2772 }
2773 
2774 /*
2775  * Kernel VM initialization.
2776  */
2777 static void
2778 kvm_init(void)
2779 {
2780         ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0);
2781 
2782         /*
2783          * Put the kernel segments in kernel address space.
2784          */
2785         rw_enter(&kas.a_lock, RW_WRITER);
2786         as_avlinit(&kas);
2787 
2788         (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg);
2789         (void) segkmem_create(&ktextseg);
2790 
2791         (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc);
2792         (void) segkmem_create(&kvalloc);
2793 
2794         (void) seg_attach(&kas, kernelheap,
2795             ekernelheap - kernelheap, &kvseg);
2796         (void) segkmem_create(&kvseg);
2797 
2798         if (core_size > 0) {
2799                 PRM_POINT("attaching kvseg_core");
2800                 (void) seg_attach(&kas, (caddr_t)core_base, core_size,
2801                     &kvseg_core);
2802                 (void) segkmem_create(&kvseg_core);
2803         }
2804 
2805         if (segziosize > 0) {
2806                 PRM_POINT("attaching segzio");
2807                 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2808                     &kzioseg);
2809                 (void) segkmem_zio_create(&kzioseg);
2810 
2811                 /* create zio area covering new segment */
2812                 segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2813         }
2814 
2815         (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2816         (void) segkmem_create(&kdebugseg);
2817 
2818         rw_exit(&kas.a_lock);
2819 
2820         /*
2821          * Ensure that the red zone at kernelbase is never accessible.
2822          */
2823         PRM_POINT("protecting redzone");
2824         (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0);
2825 
2826         /*
2827          * Make the text writable so that it can be hot patched by DTrace.
2828          */
2829         (void) as_setprot(&kas, s_text, e_modtext - s_text,
2830             PROT_READ | PROT_WRITE | PROT_EXEC);
2831 
2832         /*
2833          * Make data writable until end.
2834          */
2835         (void) as_setprot(&kas, s_data, e_moddata - s_data,
2836             PROT_READ | PROT_WRITE | PROT_EXEC);
2837 }
2838 
2839 #ifndef __xpv
2840 /*
2841  * Solaris adds an entry for Write Combining caching to the PAT
2842  */
2843 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE;
2844 
2845 void
2846 pat_sync(void)
2847 {
2848         ulong_t cr0, cr0_orig, cr4;
2849 
2850         if (!is_x86_feature(x86_featureset, X86FSET_PAT))
2851                 return;
2852         cr0_orig = cr0 = getcr0();
2853         cr4 = getcr4();
2854 
2855         /* disable caching and flush all caches and TLBs */
2856         cr0 |= CR0_CD;
2857         cr0 &= ~CR0_NW;
2858         setcr0(cr0);
2859         invalidate_cache();
2860         if (cr4 & CR4_PGE) {
2861                 setcr4(cr4 & ~(ulong_t)CR4_PGE);
2862                 setcr4(cr4);
2863         } else {
2864                 reload_cr3();
2865         }
2866 
2867         /* add our entry to the PAT */
2868         wrmsr(REG_PAT, pat_attr_reg);
2869 
2870         /* flush TLBs and cache again, then reenable cr0 caching */
2871         if (cr4 & CR4_PGE) {
2872                 setcr4(cr4 & ~(ulong_t)CR4_PGE);
2873                 setcr4(cr4);
2874         } else {
2875                 reload_cr3();
2876         }
2877         invalidate_cache();
2878         setcr0(cr0_orig);
2879 }
2880 
2881 #endif /* !__xpv */
2882 
2883 #if defined(_SOFT_HOSTID)
2884 /*
2885  * On platforms that do not have a hardware serial number, attempt
2886  * to set one based on the contents of /etc/hostid.  If this file does
2887  * not exist, assume that we are to generate a new hostid and set
2888  * it in the kernel, for subsequent saving by a userland process
2889  * once the system is up and the root filesystem is mounted r/w.
2890  *
2891  * In order to gracefully support upgrade on OpenSolaris, if
2892  * /etc/hostid does not exist, we will attempt to get a serial number
2893  * using the legacy method (/kernel/misc/sysinit).
2894  *
2895  * If that isn't present, we attempt to use an SMBIOS UUID, which is
2896  * a hardware serial number.  Note that we don't automatically trust
2897  * all SMBIOS UUIDs (some older platforms are defective and ship duplicate
2898  * UUIDs in violation of the standard), we check against a blacklist.
2899  *
2900  * In an attempt to make the hostid less prone to abuse
2901  * (for license circumvention, etc), we store it in /etc/hostid
2902  * in rot47 format.
2903  */
2904 extern volatile unsigned long tenmicrodata;
2905 static int atoi(char *);
2906 
2907 /*
2908  * Set this to non-zero in /etc/system if you think your SMBIOS returns a
2909  * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist
2910  * array can be updated.)
2911  */
2912 int smbios_broken_uuid = 0;
2913 
2914 /*
2915  * List of known bad UUIDs.  This is just the lower 32-bit values, since
2916  * that's what we use for the host id.  If your hostid falls here, you need
2917  * to contact your hardware OEM for a fix for your BIOS.
2918  */
2919 static unsigned char
2920 smbios_uuid_blacklist[][16] = {
2921 
2922         {       /* Reported bad UUID (Google search) */
2923                 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05,
2924                 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09,
2925         },
2926         {       /* Known bad DELL UUID */
2927                 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10,
2928                 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20,
2929         },
2930         {       /* Uninitialized flash */
2931                 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
2932                 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
2933         },
2934         {       /* All zeros */
2935                 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
2936                 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
2937         },
2938 };
2939 
2940 static int32_t
2941 uuid_to_hostid(const uint8_t *uuid)
2942 {
2943         /*
2944          * Although the UUIDs are 128-bits, they may not distribute entropy
2945          * evenly.  We would like to use SHA or MD5, but those are located
2946          * in loadable modules and not available this early in boot.  As we
2947          * don't need the values to be cryptographically strong, we just
2948          * generate 32-bit vaue by xor'ing the various sequences together,
2949          * which ensures that the entire UUID contributes to the hostid.
2950          */
2951         uint32_t        id = 0;
2952 
2953         /* first check against the blacklist */
2954         for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) {
2955                 if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) {
2956                         cmn_err(CE_CONT, "?Broken SMBIOS UUID. "
2957                             "Contact BIOS manufacturer for repair.\n");
2958                         return ((int32_t)HW_INVALID_HOSTID);
2959                 }
2960         }
2961 
2962         for (int i = 0; i < 16; i++)
2963                 id ^= ((uuid[i]) << (8 * (i % sizeof (id))));
2964 
2965         /* Make sure return value is positive */
2966         return (id & 0x7fffffff);
2967 }
2968 
2969 static int32_t
2970 set_soft_hostid(void)
2971 {
2972         struct _buf *file;
2973         char tokbuf[MAXNAMELEN];
2974         token_t token;
2975         int done = 0;
2976         u_longlong_t tmp;
2977         int i;
2978         int32_t hostid = (int32_t)HW_INVALID_HOSTID;
2979         unsigned char *c;
2980         hrtime_t tsc;
2981         smbios_system_t smsys;
2982 
2983         /*
2984          * If /etc/hostid file not found, we'd like to get a pseudo
2985          * random number to use at the hostid.  A nice way to do this
2986          * is to read the real time clock.  To remain xen-compatible,
2987          * we can't poke the real hardware, so we use tsc_read() to
2988          * read the real time clock.  However, there is an ominous
2989          * warning in tsc_read that says it can return zero, so we
2990          * deal with that possibility by falling back to using the
2991          * (hopefully random enough) value in tenmicrodata.
2992          */
2993 
2994         if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) {
2995                 /*
2996                  * hostid file not found - try to load sysinit module
2997                  * and see if it has a nonzero hostid value...use that
2998                  * instead of generating a new hostid here if so.
2999                  */
3000                 if ((i = modload("misc", "sysinit")) != -1) {
3001                         if (strlen(hw_serial) > 0)
3002                                 hostid = (int32_t)atoi(hw_serial);
3003                         (void) modunload(i);
3004                 }
3005 
3006                 /*
3007                  * We try to use the SMBIOS UUID. But not if it is blacklisted
3008                  * in /etc/system.
3009                  */
3010                 if ((hostid == HW_INVALID_HOSTID) &&
3011                     (smbios_broken_uuid == 0) &&
3012                     (ksmbios != NULL) &&
3013                     (smbios_info_system(ksmbios, &smsys) != SMB_ERR) &&
3014                     (smsys.smbs_uuidlen >= 16)) {
3015                         hostid = uuid_to_hostid(smsys.smbs_uuid);
3016                 }
3017 
3018                 /*
3019                  * Generate a "random" hostid using the clock.  These
3020                  * hostids will change on each boot if the value is not
3021                  * saved to a persistent /etc/hostid file.
3022                  */
3023                 if (hostid == HW_INVALID_HOSTID) {
3024                         tsc = tsc_read();
3025                         if (tsc == 0)   /* tsc_read can return zero sometimes */
3026                                 hostid = (int32_t)tenmicrodata & 0x0CFFFFF;
3027                         else
3028                                 hostid = (int32_t)tsc & 0x0CFFFFF;
3029                 }
3030         } else {
3031                 /* hostid file found */
3032                 while (!done) {
3033                         token = kobj_lex(file, tokbuf, sizeof (tokbuf));
3034 
3035                         switch (token) {
3036                         case POUND:
3037                                 /*
3038                                  * skip comments
3039                                  */
3040                                 kobj_find_eol(file);
3041                                 break;
3042                         case STRING:
3043                                 /*
3044                                  * un-rot47 - obviously this
3045                                  * nonsense is ascii-specific
3046                                  */
3047                                 for (c = (unsigned char *)tokbuf;
3048                                     *c != '\0'; c++) {
3049                                         *c += 47;
3050                                         if (*c > '~')
3051                                                 *c -= 94;
3052                                         else if (*c < '!')
3053                                                 *c += 94;
3054                                 }
3055                                 /*
3056                                  * now we should have a real number
3057                                  */
3058 
3059                                 if (kobj_getvalue(tokbuf, &tmp) != 0)
3060                                         kobj_file_err(CE_WARN, file,
3061                                             "Bad value %s for hostid",
3062                                             tokbuf);
3063                                 else
3064                                         hostid = (int32_t)tmp;
3065 
3066                                 break;
3067                         case EOF:
3068                                 done = 1;
3069                                 /* FALLTHROUGH */
3070                         case NEWLINE:
3071                                 kobj_newline(file);
3072                                 break;
3073                         default:
3074                                 break;
3075 
3076                         }
3077                 }
3078                 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */
3079                         kobj_file_err(CE_WARN, file,
3080                             "hostid missing or corrupt");
3081 
3082                 kobj_close_file(file);
3083         }
3084         /*
3085          * hostid is now the value read from /etc/hostid, or the
3086          * new hostid we generated in this routine or HW_INVALID_HOSTID if not
3087          * set.
3088          */
3089         return (hostid);
3090 }
3091 
3092 static int
3093 atoi(char *p)
3094 {
3095         int i = 0;
3096 
3097         while (*p != '\0')
3098                 i = 10 * i + (*p++ - '0');
3099 
3100         return (i);
3101 }
3102 
3103 #endif /* _SOFT_HOSTID */
3104 
3105 void
3106 get_system_configuration(void)
3107 {
3108         char    prop[32];
3109         u_longlong_t nodes_ll, cpus_pernode_ll, lvalue;
3110 
3111         if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) ||
3112             BOP_GETPROP(bootops, "nodes", prop) < 0 ||
3113             kobj_getvalue(prop, &nodes_ll) == -1 ||
3114             nodes_ll > MAXNODES ||
3115             BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) ||
3116             BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 ||
3117             kobj_getvalue(prop, &cpus_pernode_ll) == -1) {
3118                 system_hardware.hd_nodes = 1;
3119                 system_hardware.hd_cpus_per_node = 0;
3120         } else {
3121                 system_hardware.hd_nodes = (int)nodes_ll;
3122                 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll;
3123         }
3124 
3125         if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) ||
3126             BOP_GETPROP(bootops, "kernelbase", prop) < 0 ||
3127             kobj_getvalue(prop, &lvalue) == -1)
3128                 eprom_kernelbase = NULL;
3129         else
3130                 eprom_kernelbase = (uintptr_t)lvalue;
3131 
3132         if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) ||
3133             BOP_GETPROP(bootops, "segmapsize", prop) < 0 ||
3134             kobj_getvalue(prop, &lvalue) == -1)
3135                 segmapsize = SEGMAPDEFAULT;
3136         else
3137                 segmapsize = (uintptr_t)lvalue;
3138 
3139         if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) ||
3140             BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 ||
3141             kobj_getvalue(prop, &lvalue) == -1)
3142                 segmapfreelists = 0;    /* use segmap driver default */
3143         else
3144                 segmapfreelists = (int)lvalue;
3145 
3146         /* physmem used to be here, but moved much earlier to fakebop.c */
3147 }
3148 
3149 /*
3150  * Add to a memory list.
3151  * start = start of new memory segment
3152  * len = length of new memory segment in bytes
3153  * new = pointer to a new struct memlist
3154  * memlistp = memory list to which to add segment.
3155  */
3156 void
3157 memlist_add(
3158         uint64_t start,
3159         uint64_t len,
3160         struct memlist *new,
3161         struct memlist **memlistp)
3162 {
3163         struct memlist *cur;
3164         uint64_t end = start + len;
3165 
3166         new->ml_address = start;
3167         new->ml_size = len;
3168 
3169         cur = *memlistp;
3170 
3171         while (cur) {
3172                 if (cur->ml_address >= end) {
3173                         new->ml_next = cur;
3174                         *memlistp = new;
3175                         new->ml_prev = cur->ml_prev;
3176                         cur->ml_prev = new;
3177                         return;
3178                 }
3179                 ASSERT(cur->ml_address + cur->ml_size <= start);
3180                 if (cur->ml_next == NULL) {
3181                         cur->ml_next = new;
3182                         new->ml_prev = cur;
3183                         new->ml_next = NULL;
3184                         return;
3185                 }
3186                 memlistp = &cur->ml_next;
3187                 cur = cur->ml_next;
3188         }
3189 }
3190 
3191 void
3192 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3193 {
3194         size_t tsize = e_modtext - modtext;
3195         size_t dsize = e_moddata - moddata;
3196 
3197         *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize,
3198             1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP);
3199         *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize,
3200             1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3201 }
3202 
3203 caddr_t
3204 kobj_text_alloc(vmem_t *arena, size_t size)
3205 {
3206         return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT));
3207 }
3208 
3209 /*ARGSUSED*/
3210 caddr_t
3211 kobj_texthole_alloc(caddr_t addr, size_t size)
3212 {
3213         panic("unexpected call to kobj_texthole_alloc()");
3214         /*NOTREACHED*/
3215         return (0);
3216 }
3217 
3218 /*ARGSUSED*/
3219 void
3220 kobj_texthole_free(caddr_t addr, size_t size)
3221 {
3222         panic("unexpected call to kobj_texthole_free()");
3223 }
3224 
3225 /*
3226  * This is called just after configure() in startup().
3227  *
3228  * The ISALIST concept is a bit hopeless on Intel, because
3229  * there's no guarantee of an ever-more-capable processor
3230  * given that various parts of the instruction set may appear
3231  * and disappear between different implementations.
3232  *
3233  * While it would be possible to correct it and even enhance
3234  * it somewhat, the explicit hardware capability bitmask allows
3235  * more flexibility.
3236  *
3237  * So, we just leave this alone.
3238  */
3239 void
3240 setx86isalist(void)
3241 {
3242         char *tp;
3243         size_t len;
3244         extern char *isa_list;
3245 
3246 #define TBUFSIZE        1024
3247 
3248         tp = kmem_alloc(TBUFSIZE, KM_SLEEP);
3249         *tp = '\0';
3250 
3251 #if defined(__amd64)
3252         (void) strcpy(tp, "amd64 ");
3253 #endif
3254 
3255         switch (x86_vendor) {
3256         case X86_VENDOR_Intel:
3257         case X86_VENDOR_AMD:
3258         case X86_VENDOR_TM:
3259                 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) {
3260                         /*
3261                          * Pentium Pro or later
3262                          */
3263                         (void) strcat(tp, "pentium_pro");
3264                         (void) strcat(tp,
3265                             is_x86_feature(x86_featureset, X86FSET_MMX) ?
3266                             "+mmx pentium_pro " : " ");
3267                 }
3268                 /*FALLTHROUGH*/
3269         case X86_VENDOR_Cyrix:
3270                 /*
3271                  * The Cyrix 6x86 does not have any Pentium features
3272                  * accessible while not at privilege level 0.
3273                  */
3274                 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) {
3275                         (void) strcat(tp, "pentium");
3276                         (void) strcat(tp,
3277                             is_x86_feature(x86_featureset, X86FSET_MMX) ?
3278                             "+mmx pentium " : " ");
3279                 }
3280                 break;
3281         default:
3282                 break;
3283         }
3284         (void) strcat(tp, "i486 i386 i86");
3285         len = strlen(tp) + 1;   /* account for NULL at end of string */
3286         isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp);
3287         kmem_free(tp, TBUFSIZE);
3288 
3289 #undef TBUFSIZE
3290 }
3291 
3292 
3293 #ifdef __amd64
3294 
3295 void *
3296 device_arena_alloc(size_t size, int vm_flag)
3297 {
3298         return (vmem_alloc(device_arena, size, vm_flag));
3299 }
3300 
3301 void
3302 device_arena_free(void *vaddr, size_t size)
3303 {
3304         vmem_free(device_arena, vaddr, size);
3305 }
3306 
3307 #else /* __i386 */
3308 
3309 void *
3310 device_arena_alloc(size_t size, int vm_flag)
3311 {
3312         caddr_t vaddr;
3313         uintptr_t v;
3314         size_t  start;
3315         size_t  end;
3316 
3317         vaddr = vmem_alloc(heap_arena, size, vm_flag);
3318         if (vaddr == NULL)
3319                 return (NULL);
3320 
3321         v = (uintptr_t)vaddr;
3322         ASSERT(v >= kernelbase);
3323         ASSERT(v + size <= valloc_base);
3324 
3325         start = btop(v - kernelbase);
3326         end = btop(v + size - 1 - kernelbase);
3327         ASSERT(start < toxic_bit_map_len);
3328         ASSERT(end < toxic_bit_map_len);
3329 
3330         while (start <= end) {
3331                 BT_ATOMIC_SET(toxic_bit_map, start);
3332                 ++start;
3333         }
3334         return (vaddr);
3335 }
3336 
3337 void
3338 device_arena_free(void *vaddr, size_t size)
3339 {
3340         uintptr_t v = (uintptr_t)vaddr;
3341         size_t  start;
3342         size_t  end;
3343 
3344         ASSERT(v >= kernelbase);
3345         ASSERT(v + size <= valloc_base);
3346 
3347         start = btop(v - kernelbase);
3348         end = btop(v + size - 1 - kernelbase);
3349         ASSERT(start < toxic_bit_map_len);
3350         ASSERT(end < toxic_bit_map_len);
3351 
3352         while (start <= end) {
3353                 ASSERT(BT_TEST(toxic_bit_map, start) != 0);
3354                 BT_ATOMIC_CLEAR(toxic_bit_map, start);
3355                 ++start;
3356         }
3357         vmem_free(heap_arena, vaddr, size);
3358 }
3359 
3360 /*
3361  * returns 1st address in range that is in device arena, or NULL
3362  * if len is not NULL it returns the length of the toxic range
3363  */
3364 void *
3365 device_arena_contains(void *vaddr, size_t size, size_t *len)
3366 {
3367         uintptr_t v = (uintptr_t)vaddr;
3368         uintptr_t eaddr = v + size;
3369         size_t start;
3370         size_t end;
3371 
3372         /*
3373          * if called very early by kmdb, just return NULL
3374          */
3375         if (toxic_bit_map == NULL)
3376                 return (NULL);
3377 
3378         /*
3379          * First check if we're completely outside the bitmap range.
3380          */
3381         if (v >= valloc_base || eaddr < kernelbase)
3382                 return (NULL);
3383 
3384         /*
3385          * Trim ends of search to look at only what the bitmap covers.
3386          */
3387         if (v < kernelbase)
3388                 v = kernelbase;
3389         start = btop(v - kernelbase);
3390         end = btop(eaddr - kernelbase);
3391         if (end >= toxic_bit_map_len)
3392                 end = toxic_bit_map_len;
3393 
3394         if (bt_range(toxic_bit_map, &start, &end, end) == 0)
3395                 return (NULL);
3396 
3397         v = kernelbase + ptob(start);
3398         if (len != NULL)
3399                 *len = ptob(end - start);
3400         return ((void *)v);
3401 }
3402 
3403 #endif  /* __i386 */