xref: /illumos-gate/usr/src/uts/i86pc/os/startup.c (revision 35e1ee58e7a1d8c7874269affd084e732ad5a1eb)
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 (c) 2018 Joyent, Inc.
27  * Copyright (c) 2015 by Delphix. All rights reserved.
28  */
29 /*
30  * Copyright (c) 2010, Intel Corporation.
31  * All rights reserved.
32  */
33 
34 #include <sys/types.h>
35 #include <sys/t_lock.h>
36 #include <sys/param.h>
37 #include <sys/sysmacros.h>
38 #include <sys/signal.h>
39 #include <sys/systm.h>
40 #include <sys/user.h>
41 #include <sys/mman.h>
42 #include <sys/vm.h>
43 #include <sys/conf.h>
44 #include <sys/avintr.h>
45 #include <sys/autoconf.h>
46 #include <sys/disp.h>
47 #include <sys/class.h>
48 #include <sys/bitmap.h>
49 
50 #include <sys/privregs.h>
51 
52 #include <sys/proc.h>
53 #include <sys/buf.h>
54 #include <sys/kmem.h>
55 #include <sys/mem.h>
56 #include <sys/kstat.h>
57 
58 #include <sys/reboot.h>
59 
60 #include <sys/cred.h>
61 #include <sys/vnode.h>
62 #include <sys/file.h>
63 
64 #include <sys/procfs.h>
65 
66 #include <sys/vfs.h>
67 #include <sys/cmn_err.h>
68 #include <sys/utsname.h>
69 #include <sys/debug.h>
70 #include <sys/kdi.h>
71 
72 #include <sys/dumphdr.h>
73 #include <sys/bootconf.h>
74 #include <sys/memlist_plat.h>
75 #include <sys/varargs.h>
76 #include <sys/promif.h>
77 #include <sys/modctl.h>
78 
79 #include <sys/sunddi.h>
80 #include <sys/sunndi.h>
81 #include <sys/ndi_impldefs.h>
82 #include <sys/ddidmareq.h>
83 #include <sys/psw.h>
84 #include <sys/regset.h>
85 #include <sys/clock.h>
86 #include <sys/pte.h>
87 #include <sys/tss.h>
88 #include <sys/stack.h>
89 #include <sys/trap.h>
90 #include <sys/fp.h>
91 #include <vm/kboot_mmu.h>
92 #include <vm/anon.h>
93 #include <vm/as.h>
94 #include <vm/page.h>
95 #include <vm/seg.h>
96 #include <vm/seg_dev.h>
97 #include <vm/seg_kmem.h>
98 #include <vm/seg_kpm.h>
99 #include <vm/seg_map.h>
100 #include <vm/seg_vn.h>
101 #include <vm/seg_kp.h>
102 #include <sys/memnode.h>
103 #include <vm/vm_dep.h>
104 #include <sys/thread.h>
105 #include <sys/sysconf.h>
106 #include <sys/vm_machparam.h>
107 #include <sys/archsystm.h>
108 #include <sys/machsystm.h>
109 #include <vm/hat.h>
110 #include <vm/hat_i86.h>
111 #include <sys/pmem.h>
112 #include <sys/smp_impldefs.h>
113 #include <sys/x86_archext.h>
114 #include <sys/cpuvar.h>
115 #include <sys/segments.h>
116 #include <sys/clconf.h>
117 #include <sys/kobj.h>
118 #include <sys/kobj_lex.h>
119 #include <sys/cpc_impl.h>
120 #include <sys/cpu_module.h>
121 #include <sys/smbios.h>
122 #include <sys/debug_info.h>
123 #include <sys/bootinfo.h>
124 #include <sys/ddi_periodic.h>
125 #include <sys/systeminfo.h>
126 #include <sys/multiboot.h>
127 #include <sys/ramdisk.h>
128 
129 #ifdef	__xpv
130 
131 #include <sys/hypervisor.h>
132 #include <sys/xen_mmu.h>
133 #include <sys/evtchn_impl.h>
134 #include <sys/gnttab.h>
135 #include <sys/xpv_panic.h>
136 #include <xen/sys/xenbus_comms.h>
137 #include <xen/public/physdev.h>
138 
139 extern void xen_late_startup(void);
140 
141 struct xen_evt_data cpu0_evt_data;
142 
143 #else	/* __xpv */
144 #include <sys/memlist_impl.h>
145 
146 extern void mem_config_init(void);
147 #endif /* __xpv */
148 
149 extern void progressbar_init(void);
150 extern void brand_init(void);
151 extern void pcf_init(void);
152 extern void pg_init(void);
153 extern void ssp_init(void);
154 
155 extern int size_pse_array(pgcnt_t, int);
156 
157 #if defined(_SOFT_HOSTID)
158 
159 #include <sys/rtc.h>
160 
161 static int32_t set_soft_hostid(void);
162 static char hostid_file[] = "/etc/hostid";
163 
164 #endif
165 
166 void *gfx_devinfo_list;
167 
168 #if defined(__amd64) && !defined(__xpv)
169 extern void immu_startup(void);
170 #endif
171 
172 /*
173  * XXX make declaration below "static" when drivers no longer use this
174  * interface.
175  */
176 extern caddr_t p0_va;	/* Virtual address for accessing physical page 0 */
177 
178 /*
179  * segkp
180  */
181 extern int segkp_fromheap;
182 
183 static void kvm_init(void);
184 static void startup_init(void);
185 static void startup_memlist(void);
186 static void startup_kmem(void);
187 static void startup_modules(void);
188 static void startup_vm(void);
189 static void startup_end(void);
190 static void layout_kernel_va(void);
191 
192 /*
193  * Declare these as initialized data so we can patch them.
194  */
195 #ifdef __i386
196 
197 /*
198  * Due to virtual address space limitations running in 32 bit mode, restrict
199  * the amount of physical memory configured to a max of PHYSMEM pages (16g).
200  *
201  * If the physical max memory size of 64g were allowed to be configured, the
202  * size of user virtual address space will be less than 1g. A limited user
203  * address space greatly reduces the range of applications that can run.
204  *
205  * If more physical memory than PHYSMEM is required, users should preferably
206  * run in 64 bit mode which has far looser virtual address space limitations.
207  *
208  * If 64 bit mode is not available (as in IA32) and/or more physical memory
209  * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired
210  * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase
211  * should also be carefully tuned to balance out the need of the user
212  * application while minimizing the risk of kernel heap exhaustion due to
213  * kernelbase being set too high.
214  */
215 #define	PHYSMEM	0x400000
216 
217 #else /* __amd64 */
218 
219 /*
220  * For now we can handle memory with physical addresses up to about
221  * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly
222  * half the VA space for seg_kpm. When systems get bigger than 64TB this
223  * code will need revisiting. There is an implicit assumption that there
224  * are no *huge* holes in the physical address space too.
225  */
226 #define	TERABYTE		(1ul << 40)
227 #define	PHYSMEM_MAX64		mmu_btop(64 * TERABYTE)
228 #define	PHYSMEM			PHYSMEM_MAX64
229 #define	AMD64_VA_HOLE_END	0xFFFF800000000000ul
230 
231 #endif /* __amd64 */
232 
233 pgcnt_t physmem = PHYSMEM;
234 pgcnt_t obp_pages;	/* Memory used by PROM for its text and data */
235 
236 char *kobj_file_buf;
237 int kobj_file_bufsize;	/* set in /etc/system */
238 
239 /* Global variables for MP support. Used in mp_startup */
240 caddr_t	rm_platter_va = 0;
241 uint32_t rm_platter_pa;
242 
243 int	auto_lpg_disable = 1;
244 
245 /*
246  * Some CPUs have holes in the middle of the 64-bit virtual address range.
247  */
248 uintptr_t hole_start, hole_end;
249 
250 /*
251  * kpm mapping window
252  */
253 caddr_t kpm_vbase;
254 size_t  kpm_size;
255 static int kpm_desired;
256 #ifdef __amd64
257 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE;
258 #endif
259 
260 /*
261  * Configuration parameters set at boot time.
262  */
263 
264 caddr_t econtig;		/* end of first block of contiguous kernel */
265 
266 struct bootops		*bootops = 0;	/* passed in from boot */
267 struct bootops		**bootopsp;
268 struct boot_syscalls	*sysp;		/* passed in from boot */
269 
270 char bootblock_fstype[16];
271 
272 char kern_bootargs[OBP_MAXPATHLEN];
273 char kern_bootfile[OBP_MAXPATHLEN];
274 
275 /*
276  * ZFS zio segment.  This allows us to exclude large portions of ZFS data that
277  * gets cached in kmem caches on the heap.  If this is set to zero, we allocate
278  * zio buffers from their own segment, otherwise they are allocated from the
279  * heap.  The optimization of allocating zio buffers from their own segment is
280  * only valid on 64-bit kernels.
281  */
282 #if defined(__amd64)
283 int segzio_fromheap = 0;
284 #else
285 int segzio_fromheap = 1;
286 #endif
287 
288 /*
289  * Give folks an escape hatch for disabling SMAP via kmdb. Doesn't work
290  * post-boot.
291  */
292 int disable_smap = 0;
293 
294 /*
295  * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this
296  * depends on number of BOP_ALLOC calls made and requested size, memory size
297  * combination and whether boot.bin memory needs to be freed.
298  */
299 #define	POSS_NEW_FRAGMENTS	12
300 
301 /*
302  * VM data structures
303  */
304 long page_hashsz;		/* Size of page hash table (power of two) */
305 unsigned int page_hashsz_shift;	/* log2(page_hashsz) */
306 struct page *pp_base;		/* Base of initial system page struct array */
307 struct page **page_hash;	/* Page hash table */
308 pad_mutex_t *pse_mutex;		/* Locks protecting pp->p_selock */
309 size_t pse_table_size;		/* Number of mutexes in pse_mutex[] */
310 int pse_shift;			/* log2(pse_table_size) */
311 struct seg ktextseg;		/* Segment used for kernel executable image */
312 struct seg kvalloc;		/* Segment used for "valloc" mapping */
313 struct seg kpseg;		/* Segment used for pageable kernel virt mem */
314 struct seg kmapseg;		/* Segment used for generic kernel mappings */
315 struct seg kdebugseg;		/* Segment used for the kernel debugger */
316 
317 struct seg *segkmap = &kmapseg;	/* Kernel generic mapping segment */
318 static struct seg *segmap = &kmapseg;	/* easier to use name for in here */
319 
320 struct seg *segkp = &kpseg;	/* Pageable kernel virtual memory segment */
321 
322 #if defined(__amd64)
323 struct seg kvseg_core;		/* Segment used for the core heap */
324 struct seg kpmseg;		/* Segment used for physical mapping */
325 struct seg *segkpm = &kpmseg;	/* 64bit kernel physical mapping segment */
326 #else
327 struct seg *segkpm = NULL;	/* Unused on IA32 */
328 #endif
329 
330 caddr_t segkp_base;		/* Base address of segkp */
331 caddr_t segzio_base;		/* Base address of segzio */
332 #if defined(__amd64)
333 pgcnt_t segkpsize = btop(SEGKPDEFSIZE);	/* size of segkp segment in pages */
334 #else
335 pgcnt_t segkpsize = 0;
336 #endif
337 pgcnt_t segziosize = 0;		/* size of zio segment in pages */
338 
339 /*
340  * A static DR page_t VA map is reserved that can map the page structures
341  * for a domain's entire RA space. The pages that back this space are
342  * dynamically allocated and need not be physically contiguous.  The DR
343  * map size is derived from KPM size.
344  * This mechanism isn't used by x86 yet, so just stubs here.
345  */
346 int ppvm_enable = 0;		/* Static virtual map for page structs */
347 page_t *ppvm_base = NULL;	/* Base of page struct map */
348 pgcnt_t ppvm_size = 0;		/* Size of page struct map */
349 
350 /*
351  * VA range available to the debugger
352  */
353 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
354 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
355 
356 struct memseg *memseg_base;
357 struct vnode unused_pages_vp;
358 
359 #define	FOURGB	0x100000000LL
360 
361 struct memlist *memlist;
362 
363 caddr_t s_text;		/* start of kernel text segment */
364 caddr_t e_text;		/* end of kernel text segment */
365 caddr_t s_data;		/* start of kernel data segment */
366 caddr_t e_data;		/* end of kernel data segment */
367 caddr_t modtext;	/* start of loadable module text reserved */
368 caddr_t e_modtext;	/* end of loadable module text reserved */
369 caddr_t moddata;	/* start of loadable module data reserved */
370 caddr_t e_moddata;	/* end of loadable module data reserved */
371 
372 struct memlist *phys_install;	/* Total installed physical memory */
373 struct memlist *phys_avail;	/* Total available physical memory */
374 struct memlist *bios_rsvd;	/* Bios reserved memory */
375 
376 /*
377  * kphysm_init returns the number of pages that were processed
378  */
379 static pgcnt_t kphysm_init(page_t *, pgcnt_t);
380 
381 #define	IO_PROP_SIZE	64	/* device property size */
382 
383 /*
384  * a couple useful roundup macros
385  */
386 #define	ROUND_UP_PAGE(x)	\
387 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE))
388 #define	ROUND_UP_LPAGE(x)	\
389 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1]))
390 #define	ROUND_UP_4MEG(x)	\
391 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG))
392 #define	ROUND_UP_TOPLEVEL(x)	\
393 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level]))
394 
395 /*
396  *	32-bit Kernel's Virtual memory layout.
397  *		+-----------------------+
398  *		|			|
399  * 0xFFC00000  -|-----------------------|- ARGSBASE
400  *		|	debugger	|
401  * 0xFF800000  -|-----------------------|- SEGDEBUGBASE
402  *		|      Kernel Data	|
403  * 0xFEC00000  -|-----------------------|
404  *              |      Kernel Text	|
405  * 0xFE800000  -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen)
406  *		|---       GDT       ---|- GDT page (GDT_VA)
407  *		|---    debug info   ---|- debug info (DEBUG_INFO_VA)
408  *		|			|
409  *		|   page_t structures	|
410  *		|   memsegs, memlists,	|
411  *		|   page hash, etc.	|
412  * ---	       -|-----------------------|- ekernelheap, valloc_base (floating)
413  *		|			|  (segkp is just an arena in the heap)
414  *		|			|
415  *		|	kvseg		|
416  *		|			|
417  *		|			|
418  * ---         -|-----------------------|- kernelheap (floating)
419  *		|        Segkmap	|
420  * 0xC3002000  -|-----------------------|- segmap_start (floating)
421  *		|	Red Zone	|
422  * 0xC3000000  -|-----------------------|- kernelbase / userlimit (floating)
423  *		|			|			||
424  *		|     Shared objects	|			\/
425  *		|			|
426  *		:			:
427  *		|	user data	|
428  *		|-----------------------|
429  *		|	user text	|
430  * 0x08048000  -|-----------------------|
431  *		|	user stack	|
432  *		:			:
433  *		|	invalid		|
434  * 0x00000000	+-----------------------+
435  *
436  *
437  *		64-bit Kernel's Virtual memory layout. (assuming 64 bit app)
438  *			+-----------------------+
439  *			|			|
440  * 0xFFFFFFFF.FFC00000  |-----------------------|- ARGSBASE
441  *			|	debugger (?)	|
442  * 0xFFFFFFFF.FF800000  |-----------------------|- SEGDEBUGBASE
443  *			|      unused		|
444  *			+-----------------------+
445  *			|      Kernel Data	|
446  * 0xFFFFFFFF.FBC00000  |-----------------------|
447  *			|      Kernel Text	|
448  * 0xFFFFFFFF.FB800000  |-----------------------|- KERNEL_TEXT
449  *			|---    debug info   ---|- debug info (DEBUG_INFO_VA)
450  *			|---       GDT       ---|- GDT page (GDT_VA)
451  *			|---       IDT       ---|- IDT page (IDT_VA)
452  *			|---       LDT       ---|- LDT pages (LDT_VA)
453  *			|			|
454  *			|      Core heap	| (used for loadable modules)
455  * 0xFFFFFFFF.C0000000  |-----------------------|- core_base / ekernelheap
456  *			|	 Kernel		|
457  *			|	  heap		|
458  * 0xFFFFFXXX.XXX00000  |-----------------------|- kernelheap (floating)
459  *			|	 segmap		|
460  * 0xFFFFFXXX.XXX00000  |-----------------------|- segmap_start (floating)
461  *			|    device mappings	|
462  * 0xFFFFFXXX.XXX00000  |-----------------------|- toxic_addr (floating)
463  *			|	  segzio	|
464  * 0xFFFFFXXX.XXX00000  |-----------------------|- segzio_base (floating)
465  *			|	  segkp		|
466  * ---                  |-----------------------|- segkp_base (floating)
467  *			|   page_t structures	|  valloc_base + valloc_sz
468  *			|   memsegs, memlists,	|
469  *			|   page hash, etc.	|
470  * 0xFFFFFF00.00000000  |-----------------------|- valloc_base (lower if >256GB)
471  *			|	 segkpm		|
472  * 0xFFFFFE00.00000000  |-----------------------|
473  *			|	Red Zone	|
474  * 0xFFFFFD80.00000000  |-----------------------|- KERNELBASE (lower if >256GB)
475  *			|     User stack	|- User space memory
476  *			|			|
477  *			| shared objects, etc	|	(grows downwards)
478  *			:			:
479  *			|			|
480  * 0xFFFF8000.00000000  |-----------------------|
481  *			|			|
482  *			| VA Hole / unused	|
483  *			|			|
484  * 0x00008000.00000000  |-----------------------|
485  *			|			|
486  *			|			|
487  *			:			:
488  *			|	user heap	|	(grows upwards)
489  *			|			|
490  *			|	user data	|
491  *			|-----------------------|
492  *			|	user text	|
493  * 0x00000000.04000000  |-----------------------|
494  *			|	invalid		|
495  * 0x00000000.00000000	+-----------------------+
496  *
497  * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit
498  * kernel, except that userlimit is raised to 0xfe000000
499  *
500  * Floating values:
501  *
502  * valloc_base: start of the kernel's memory management/tracking data
503  * structures.  This region contains page_t structures for
504  * physical memory, memsegs, memlists, and the page hash.
505  *
506  * core_base: start of the kernel's "core" heap area on 64-bit systems.
507  * This area is intended to be used for global data as well as for module
508  * text/data that does not fit into the nucleus pages.  The core heap is
509  * restricted to a 2GB range, allowing every address within it to be
510  * accessed using rip-relative addressing
511  *
512  * ekernelheap: end of kernelheap and start of segmap.
513  *
514  * kernelheap: start of kernel heap.  On 32-bit systems, this starts right
515  * above a red zone that separates the user's address space from the
516  * kernel's.  On 64-bit systems, it sits above segkp and segkpm.
517  *
518  * segmap_start: start of segmap. The length of segmap can be modified
519  * through eeprom. The default length is 16MB on 32-bit systems and 64MB
520  * on 64-bit systems.
521  *
522  * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be
523  * decreased by 2X the size required for page_t.  This allows the kernel
524  * heap to grow in size with physical memory.  With sizeof(page_t) == 80
525  * bytes, the following shows the values of kernelbase and kernel heap
526  * sizes for different memory configurations (assuming default segmap and
527  * segkp sizes).
528  *
529  *	mem	size for	kernelbase	kernel heap
530  *	size	page_t's			size
531  *	----	---------	----------	-----------
532  *	1gb	0x01400000	0xd1800000	684MB
533  *	2gb	0x02800000	0xcf000000	704MB
534  *	4gb	0x05000000	0xca000000	744MB
535  *	6gb	0x07800000	0xc5000000	784MB
536  *	8gb	0x0a000000	0xc0000000	824MB
537  *	16gb	0x14000000	0xac000000	984MB
538  *	32gb	0x28000000	0x84000000	1304MB
539  *	64gb	0x50000000	0x34000000	1944MB (*)
540  *
541  * kernelbase is less than the abi minimum of 0xc0000000 for memory
542  * configurations above 8gb.
543  *
544  * (*) support for memory configurations above 32gb will require manual tuning
545  * of kernelbase to balance out the need of user applications.
546  */
547 
548 /* real-time-clock initialization parameters */
549 extern time_t process_rtc_config_file(void);
550 
551 uintptr_t	kernelbase;
552 uintptr_t	postbootkernelbase;	/* not set till boot loader is gone */
553 uintptr_t	eprom_kernelbase;
554 size_t		segmapsize;
555 uintptr_t	segmap_start;
556 int		segmapfreelists;
557 pgcnt_t		npages;
558 pgcnt_t		orig_npages;
559 size_t		core_size;		/* size of "core" heap */
560 uintptr_t	core_base;		/* base address of "core" heap */
561 
562 /*
563  * List of bootstrap pages. We mark these as allocated in startup.
564  * release_bootstrap() will free them when we're completely done with
565  * the bootstrap.
566  */
567 static page_t *bootpages;
568 
569 /*
570  * boot time pages that have a vnode from the ramdisk will keep that forever.
571  */
572 static page_t *rd_pages;
573 
574 /*
575  * Lower 64K
576  */
577 static page_t *lower_pages = NULL;
578 static int lower_pages_count = 0;
579 
580 struct system_hardware system_hardware;
581 
582 /*
583  * Enable some debugging messages concerning memory usage...
584  */
585 static void
586 print_memlist(char *title, struct memlist *mp)
587 {
588 	prom_printf("MEMLIST: %s:\n", title);
589 	while (mp != NULL)  {
590 		prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n",
591 		    mp->ml_address, mp->ml_size);
592 		mp = mp->ml_next;
593 	}
594 }
595 
596 /*
597  * XX64 need a comment here.. are these just default values, surely
598  * we read the "cpuid" type information to figure this out.
599  */
600 int	l2cache_sz = 0x80000;
601 int	l2cache_linesz = 0x40;
602 int	l2cache_assoc = 1;
603 
604 static size_t	textrepl_min_gb = 10;
605 
606 /*
607  * on 64 bit we use a predifined VA range for mapping devices in the kernel
608  * on 32 bit the mappings are intermixed in the heap, so we use a bit map
609  */
610 #ifdef __amd64
611 
612 vmem_t		*device_arena;
613 uintptr_t	toxic_addr = (uintptr_t)NULL;
614 size_t		toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */
615 
616 #else	/* __i386 */
617 
618 ulong_t		*toxic_bit_map;	/* one bit for each 4k of VA in heap_arena */
619 size_t		toxic_bit_map_len = 0;	/* in bits */
620 
621 #endif	/* __i386 */
622 
623 /*
624  * Simple boot time debug facilities
625  */
626 static char *prm_dbg_str[] = {
627 	"%s:%d: '%s' is 0x%x\n",
628 	"%s:%d: '%s' is 0x%llx\n"
629 };
630 
631 int prom_debug;
632 
633 #define	PRM_DEBUG(q)	if (prom_debug)		\
634 	prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q);
635 #define	PRM_POINT(q)	if (prom_debug)		\
636 	prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q);
637 
638 /*
639  * This structure is used to keep track of the intial allocations
640  * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to
641  * be >= the number of ADD_TO_ALLOCATIONS() executed in the code.
642  */
643 #define	NUM_ALLOCATIONS 8
644 int num_allocations = 0;
645 struct {
646 	void **al_ptr;
647 	size_t al_size;
648 } allocations[NUM_ALLOCATIONS];
649 size_t valloc_sz = 0;
650 uintptr_t valloc_base;
651 
652 #define	ADD_TO_ALLOCATIONS(ptr, size) {					\
653 		size = ROUND_UP_PAGE(size);				\
654 		if (num_allocations == NUM_ALLOCATIONS)			\
655 			panic("too many ADD_TO_ALLOCATIONS()");		\
656 		allocations[num_allocations].al_ptr = (void**)&ptr;	\
657 		allocations[num_allocations].al_size = size;		\
658 		valloc_sz += size;					\
659 		++num_allocations;					\
660 	}
661 
662 /*
663  * Allocate all the initial memory needed by the page allocator.
664  */
665 static void
666 perform_allocations(void)
667 {
668 	caddr_t mem;
669 	int i;
670 	int valloc_align;
671 
672 	PRM_DEBUG(valloc_base);
673 	PRM_DEBUG(valloc_sz);
674 	valloc_align = mmu.level_size[mmu.max_page_level > 0];
675 	mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align);
676 	if (mem != (caddr_t)valloc_base)
677 		panic("BOP_ALLOC() failed");
678 	bzero(mem, valloc_sz);
679 	for (i = 0; i < num_allocations; ++i) {
680 		*allocations[i].al_ptr = (void *)mem;
681 		mem += allocations[i].al_size;
682 	}
683 }
684 
685 /*
686  * Set up and enable SMAP now before we start other CPUs, but after the kernel's
687  * VM has been set up so we can use hot_patch_kernel_text().
688  *
689  * We can only patch 1, 2, or 4 bytes, but not three bytes. So instead, we
690  * replace the four byte word at the patch point. See uts/intel/ia32/ml/copy.s
691  * for more information on what's going on here.
692  */
693 static void
694 startup_smap(void)
695 {
696 	int i;
697 	uint32_t inst;
698 	uint8_t *instp;
699 	char sym[128];
700 	struct modctl *modp;
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 	/*
735 	 * Hotinline calls to smap_enable and smap_disable within
736 	 * unix module. Hotinlines in other modules are done on
737 	 * mod_load().
738 	 */
739 	modp = mod_hold_by_name("unix");
740 	do_hotinlines(modp->mod_mp);
741 	mod_release_mod(modp);
742 
743 	setcr4(getcr4() | CR4_SMAP);
744 	smap_enable();
745 }
746 
747 /*
748  * Our world looks like this at startup time.
749  *
750  * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data
751  * at 0xfec00000.  On a 64-bit OS, kernel text and data are loaded at
752  * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively.  Those
753  * addresses are fixed in the binary at link time.
754  *
755  * On the text page:
756  * unix/genunix/krtld/module text loads.
757  *
758  * On the data page:
759  * unix/genunix/krtld/module data loads.
760  *
761  * Machine-dependent startup code
762  */
763 void
764 startup(void)
765 {
766 #if !defined(__xpv)
767 	extern void startup_pci_bios(void);
768 #endif
769 	extern cpuset_t cpu_ready_set;
770 
771 	/*
772 	 * Make sure that nobody tries to use sekpm until we have
773 	 * initialized it properly.
774 	 */
775 #if defined(__amd64)
776 	kpm_desired = 1;
777 #endif
778 	kpm_enable = 0;
779 	CPUSET_ONLY(cpu_ready_set, 0);	/* cpu 0 is boot cpu */
780 
781 #if defined(__xpv)	/* XXPV fix me! */
782 	{
783 		extern int segvn_use_regions;
784 		segvn_use_regions = 0;
785 	}
786 #endif
787 	ssp_init();
788 	progressbar_init();
789 	startup_init();
790 #if defined(__xpv)
791 	startup_xen_version();
792 #endif
793 	startup_memlist();
794 	startup_kmem();
795 	startup_vm();
796 #if !defined(__xpv)
797 	/*
798 	 * Note we need to do this even on fast reboot in order to access
799 	 * the irq routing table (used for pci labels).
800 	 */
801 	startup_pci_bios();
802 	startup_smap();
803 #endif
804 #if defined(__xpv)
805 	startup_xen_mca();
806 #endif
807 	startup_modules();
808 
809 	startup_end();
810 }
811 
812 static void
813 startup_init()
814 {
815 	PRM_POINT("startup_init() starting...");
816 
817 	/*
818 	 * Complete the extraction of cpuid data
819 	 */
820 	cpuid_pass2(CPU);
821 
822 	(void) check_boot_version(BOP_GETVERSION(bootops));
823 
824 	/*
825 	 * Check for prom_debug in boot environment
826 	 */
827 	if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) {
828 		++prom_debug;
829 		PRM_POINT("prom_debug found in boot enviroment");
830 	}
831 
832 	/*
833 	 * Collect node, cpu and memory configuration information.
834 	 */
835 	get_system_configuration();
836 
837 	/*
838 	 * Halt if this is an unsupported processor.
839 	 */
840 	if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) {
841 		printf("\n486 processor (\"%s\") detected.\n",
842 		    CPU->cpu_brandstr);
843 		halt("This processor is not supported by this release "
844 		    "of Solaris.");
845 	}
846 
847 	PRM_POINT("startup_init() done");
848 }
849 
850 /*
851  * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie.
852  * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it
853  * also filters out physical page zero.  There is some reliance on the
854  * boot loader allocating only a few contiguous physical memory chunks.
855  */
856 static void
857 avail_filter(uint64_t *addr, uint64_t *size)
858 {
859 	uintptr_t va;
860 	uintptr_t next_va;
861 	pfn_t pfn;
862 	uint64_t pfn_addr;
863 	uint64_t pfn_eaddr;
864 	uint_t prot;
865 	size_t len;
866 	uint_t change;
867 
868 	if (prom_debug)
869 		prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n",
870 		    *addr, *size);
871 
872 	/*
873 	 * page zero is required for BIOS.. never make it available
874 	 */
875 	if (*addr == 0) {
876 		*addr += MMU_PAGESIZE;
877 		*size -= MMU_PAGESIZE;
878 	}
879 
880 	/*
881 	 * First we trim from the front of the range. Since kbm_probe()
882 	 * walks ranges in virtual order, but addr/size are physical, we need
883 	 * to the list until no changes are seen.  This deals with the case
884 	 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w
885 	 * but w < v.
886 	 */
887 	do {
888 		change = 0;
889 		for (va = KERNEL_TEXT;
890 		    *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
891 		    va = next_va) {
892 
893 			next_va = va + len;
894 			pfn_addr = pfn_to_pa(pfn);
895 			pfn_eaddr = pfn_addr + len;
896 
897 			if (pfn_addr <= *addr && pfn_eaddr > *addr) {
898 				change = 1;
899 				while (*size > 0 && len > 0) {
900 					*addr += MMU_PAGESIZE;
901 					*size -= MMU_PAGESIZE;
902 					len -= MMU_PAGESIZE;
903 				}
904 			}
905 		}
906 		if (change && prom_debug)
907 			prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n",
908 			    *addr, *size);
909 	} while (change);
910 
911 	/*
912 	 * Trim pages from the end of the range.
913 	 */
914 	for (va = KERNEL_TEXT;
915 	    *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
916 	    va = next_va) {
917 
918 		next_va = va + len;
919 		pfn_addr = pfn_to_pa(pfn);
920 
921 		if (pfn_addr >= *addr && pfn_addr < *addr + *size)
922 			*size = pfn_addr - *addr;
923 	}
924 
925 	if (prom_debug)
926 		prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n",
927 		    *addr, *size);
928 }
929 
930 static void
931 kpm_init()
932 {
933 	struct segkpm_crargs b;
934 
935 	/*
936 	 * These variables were all designed for sfmmu in which segkpm is
937 	 * mapped using a single pagesize - either 8KB or 4MB.  On x86, we
938 	 * might use 2+ page sizes on a single machine, so none of these
939 	 * variables have a single correct value.  They are set up as if we
940 	 * always use a 4KB pagesize, which should do no harm.  In the long
941 	 * run, we should get rid of KPM's assumption that only a single
942 	 * pagesize is used.
943 	 */
944 	kpm_pgshft = MMU_PAGESHIFT;
945 	kpm_pgsz =  MMU_PAGESIZE;
946 	kpm_pgoff = MMU_PAGEOFFSET;
947 	kpmp2pshft = 0;
948 	kpmpnpgs = 1;
949 	ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0);
950 
951 	PRM_POINT("about to create segkpm");
952 	rw_enter(&kas.a_lock, RW_WRITER);
953 
954 	if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0)
955 		panic("cannot attach segkpm");
956 
957 	b.prot = PROT_READ | PROT_WRITE;
958 	b.nvcolors = 1;
959 
960 	if (segkpm_create(segkpm, (caddr_t)&b) != 0)
961 		panic("segkpm_create segkpm");
962 
963 	rw_exit(&kas.a_lock);
964 
965 	kpm_enable = 1;
966 
967 	/*
968 	 * As the KPM was disabled while setting up the system, go back and fix
969 	 * CPU zero's access to its user page table. This is a bit gross, but
970 	 * we have a chicken and egg problem otherwise.
971 	 */
972 	ASSERT(CPU->cpu_hat_info->hci_user_l3ptes == NULL);
973 	CPU->cpu_hat_info->hci_user_l3ptes =
974 	    (x86pte_t *)hat_kpm_mapin_pfn(CPU->cpu_hat_info->hci_user_l3pfn);
975 }
976 
977 /*
978  * The debug info page provides enough information to allow external
979  * inspectors (e.g. when running under a hypervisor) to bootstrap
980  * themselves into allowing full-blown kernel debugging.
981  */
982 static void
983 init_debug_info(void)
984 {
985 	caddr_t mem;
986 	debug_info_t *di;
987 
988 #ifndef __lint
989 	ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE);
990 #endif
991 
992 	mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE,
993 	    MMU_PAGESIZE);
994 
995 	if (mem != (caddr_t)DEBUG_INFO_VA)
996 		panic("BOP_ALLOC() failed");
997 	bzero(mem, MMU_PAGESIZE);
998 
999 	di = (debug_info_t *)mem;
1000 
1001 	di->di_magic = DEBUG_INFO_MAGIC;
1002 	di->di_version = DEBUG_INFO_VERSION;
1003 	di->di_modules = (uintptr_t)&modules;
1004 	di->di_s_text = (uintptr_t)s_text;
1005 	di->di_e_text = (uintptr_t)e_text;
1006 	di->di_s_data = (uintptr_t)s_data;
1007 	di->di_e_data = (uintptr_t)e_data;
1008 	di->di_hat_htable_off = offsetof(hat_t, hat_htable);
1009 	di->di_ht_pfn_off = offsetof(htable_t, ht_pfn);
1010 }
1011 
1012 /*
1013  * Build the memlists and other kernel essential memory system data structures.
1014  * This is everything at valloc_base.
1015  */
1016 static void
1017 startup_memlist(void)
1018 {
1019 	size_t memlist_sz;
1020 	size_t memseg_sz;
1021 	size_t pagehash_sz;
1022 	size_t pp_sz;
1023 	uintptr_t va;
1024 	size_t len;
1025 	uint_t prot;
1026 	pfn_t pfn;
1027 	int memblocks;
1028 	pfn_t rsvd_high_pfn;
1029 	pgcnt_t rsvd_pgcnt;
1030 	size_t rsvdmemlist_sz;
1031 	int rsvdmemblocks;
1032 	caddr_t pagecolor_mem;
1033 	size_t pagecolor_memsz;
1034 	caddr_t page_ctrs_mem;
1035 	size_t page_ctrs_size;
1036 	size_t pse_table_alloc_size;
1037 	struct memlist *current;
1038 	extern void startup_build_mem_nodes(struct memlist *);
1039 
1040 	/* XX64 fix these - they should be in include files */
1041 	extern size_t page_coloring_init(uint_t, int, int);
1042 	extern void page_coloring_setup(caddr_t);
1043 
1044 	PRM_POINT("startup_memlist() starting...");
1045 
1046 	/*
1047 	 * Use leftover large page nucleus text/data space for loadable modules.
1048 	 * Use at most MODTEXT/MODDATA.
1049 	 */
1050 	len = kbm_nucleus_size;
1051 	ASSERT(len > MMU_PAGESIZE);
1052 
1053 	moddata = (caddr_t)ROUND_UP_PAGE(e_data);
1054 	e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len);
1055 	if (e_moddata - moddata > MODDATA)
1056 		e_moddata = moddata + MODDATA;
1057 
1058 	modtext = (caddr_t)ROUND_UP_PAGE(e_text);
1059 	e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len);
1060 	if (e_modtext - modtext > MODTEXT)
1061 		e_modtext = modtext + MODTEXT;
1062 
1063 	econtig = e_moddata;
1064 
1065 	PRM_DEBUG(modtext);
1066 	PRM_DEBUG(e_modtext);
1067 	PRM_DEBUG(moddata);
1068 	PRM_DEBUG(e_moddata);
1069 	PRM_DEBUG(econtig);
1070 
1071 	/*
1072 	 * Examine the boot loader physical memory map to find out:
1073 	 * - total memory in system - physinstalled
1074 	 * - the max physical address - physmax
1075 	 * - the number of discontiguous segments of memory.
1076 	 */
1077 	if (prom_debug)
1078 		print_memlist("boot physinstalled",
1079 		    bootops->boot_mem->physinstalled);
1080 	installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax,
1081 	    &physinstalled, &memblocks);
1082 	PRM_DEBUG(physmax);
1083 	PRM_DEBUG(physinstalled);
1084 	PRM_DEBUG(memblocks);
1085 
1086 	/*
1087 	 * Compute maximum physical address for memory DR operations.
1088 	 * Memory DR operations are unsupported on xpv or 32bit OSes.
1089 	 */
1090 #ifdef	__amd64
1091 	if (plat_dr_support_memory()) {
1092 		if (plat_dr_physmax == 0) {
1093 			uint_t pabits = UINT_MAX;
1094 
1095 			cpuid_get_addrsize(CPU, &pabits, NULL);
1096 			plat_dr_physmax = btop(1ULL << pabits);
1097 		}
1098 		if (plat_dr_physmax > PHYSMEM_MAX64)
1099 			plat_dr_physmax = PHYSMEM_MAX64;
1100 	} else
1101 #endif
1102 		plat_dr_physmax = 0;
1103 
1104 	/*
1105 	 * Examine the bios reserved memory to find out:
1106 	 * - the number of discontiguous segments of memory.
1107 	 */
1108 	if (prom_debug)
1109 		print_memlist("boot reserved mem",
1110 		    bootops->boot_mem->rsvdmem);
1111 	installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn,
1112 	    &rsvd_pgcnt, &rsvdmemblocks);
1113 	PRM_DEBUG(rsvd_high_pfn);
1114 	PRM_DEBUG(rsvd_pgcnt);
1115 	PRM_DEBUG(rsvdmemblocks);
1116 
1117 	/*
1118 	 * Initialize hat's mmu parameters.
1119 	 * Check for enforce-prot-exec in boot environment. It's used to
1120 	 * enable/disable support for the page table entry NX bit.
1121 	 * The default is to enforce PROT_EXEC on processors that support NX.
1122 	 * Boot seems to round up the "len", but 8 seems to be big enough.
1123 	 */
1124 	mmu_init();
1125 
1126 #ifdef	__i386
1127 	/*
1128 	 * physmax is lowered if there is more memory than can be
1129 	 * physically addressed in 32 bit (PAE/non-PAE) modes.
1130 	 */
1131 	if (mmu.pae_hat) {
1132 		if (PFN_ABOVE64G(physmax)) {
1133 			physinstalled -= (physmax - (PFN_64G - 1));
1134 			physmax = PFN_64G - 1;
1135 		}
1136 	} else {
1137 		if (PFN_ABOVE4G(physmax)) {
1138 			physinstalled -= (physmax - (PFN_4G - 1));
1139 			physmax = PFN_4G - 1;
1140 		}
1141 	}
1142 #endif
1143 
1144 	startup_build_mem_nodes(bootops->boot_mem->physinstalled);
1145 
1146 	if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) {
1147 		int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec");
1148 		char value[8];
1149 
1150 		if (len < 8)
1151 			(void) BOP_GETPROP(bootops, "enforce-prot-exec", value);
1152 		else
1153 			(void) strcpy(value, "");
1154 		if (strcmp(value, "off") == 0)
1155 			mmu.pt_nx = 0;
1156 	}
1157 	PRM_DEBUG(mmu.pt_nx);
1158 
1159 	/*
1160 	 * We will need page_t's for every page in the system, except for
1161 	 * memory mapped at or above above the start of the kernel text segment.
1162 	 *
1163 	 * pages above e_modtext are attributed to kernel debugger (obp_pages)
1164 	 */
1165 	npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */
1166 	obp_pages = 0;
1167 	va = KERNEL_TEXT;
1168 	while (kbm_probe(&va, &len, &pfn, &prot) != 0) {
1169 		npages -= len >> MMU_PAGESHIFT;
1170 		if (va >= (uintptr_t)e_moddata)
1171 			obp_pages += len >> MMU_PAGESHIFT;
1172 		va += len;
1173 	}
1174 	PRM_DEBUG(npages);
1175 	PRM_DEBUG(obp_pages);
1176 
1177 	/*
1178 	 * If physmem is patched to be non-zero, use it instead of the computed
1179 	 * value unless it is larger than the actual amount of memory on hand.
1180 	 */
1181 	if (physmem == 0 || physmem > npages) {
1182 		physmem = npages;
1183 	} else if (physmem < npages) {
1184 		orig_npages = npages;
1185 		npages = physmem;
1186 	}
1187 	PRM_DEBUG(physmem);
1188 
1189 	/*
1190 	 * We now compute the sizes of all the  initial allocations for
1191 	 * structures the kernel needs in order do kmem_alloc(). These
1192 	 * include:
1193 	 *	memsegs
1194 	 *	memlists
1195 	 *	page hash table
1196 	 *	page_t's
1197 	 *	page coloring data structs
1198 	 */
1199 	memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS);
1200 	ADD_TO_ALLOCATIONS(memseg_base, memseg_sz);
1201 	PRM_DEBUG(memseg_sz);
1202 
1203 	/*
1204 	 * Reserve space for memlists. There's no real good way to know exactly
1205 	 * how much room we'll need, but this should be a good upper bound.
1206 	 */
1207 	memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1208 	    (memblocks + POSS_NEW_FRAGMENTS));
1209 	ADD_TO_ALLOCATIONS(memlist, memlist_sz);
1210 	PRM_DEBUG(memlist_sz);
1211 
1212 	/*
1213 	 * Reserve space for bios reserved memlists.
1214 	 */
1215 	rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1216 	    (rsvdmemblocks + POSS_NEW_FRAGMENTS));
1217 	ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz);
1218 	PRM_DEBUG(rsvdmemlist_sz);
1219 
1220 	/* LINTED */
1221 	ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page)));
1222 	/*
1223 	 * The page structure hash table size is a power of 2
1224 	 * such that the average hash chain length is PAGE_HASHAVELEN.
1225 	 */
1226 	page_hashsz = npages / PAGE_HASHAVELEN;
1227 	page_hashsz_shift = highbit(page_hashsz);
1228 	page_hashsz = 1 << page_hashsz_shift;
1229 	pagehash_sz = sizeof (struct page *) * page_hashsz;
1230 	ADD_TO_ALLOCATIONS(page_hash, pagehash_sz);
1231 	PRM_DEBUG(pagehash_sz);
1232 
1233 	/*
1234 	 * Set aside room for the page structures themselves.
1235 	 */
1236 	PRM_DEBUG(npages);
1237 	pp_sz = sizeof (struct page) * npages;
1238 	ADD_TO_ALLOCATIONS(pp_base, pp_sz);
1239 	PRM_DEBUG(pp_sz);
1240 
1241 	/*
1242 	 * determine l2 cache info and memory size for page coloring
1243 	 */
1244 	(void) getl2cacheinfo(CPU,
1245 	    &l2cache_sz, &l2cache_linesz, &l2cache_assoc);
1246 	pagecolor_memsz =
1247 	    page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc);
1248 	ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz);
1249 	PRM_DEBUG(pagecolor_memsz);
1250 
1251 	page_ctrs_size = page_ctrs_sz();
1252 	ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size);
1253 	PRM_DEBUG(page_ctrs_size);
1254 
1255 	/*
1256 	 * Allocate the array that protects pp->p_selock.
1257 	 */
1258 	pse_shift = size_pse_array(physmem, max_ncpus);
1259 	pse_table_size = 1 << pse_shift;
1260 	pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t);
1261 	ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size);
1262 
1263 #if defined(__amd64)
1264 	valloc_sz = ROUND_UP_LPAGE(valloc_sz);
1265 	valloc_base = VALLOC_BASE;
1266 
1267 	/*
1268 	 * The default values of VALLOC_BASE and SEGKPM_BASE should work
1269 	 * for values of physmax up to 256GB (1/4 TB). They need adjusting when
1270 	 * memory is at addresses above 256GB. When adjusted, segkpm_base must
1271 	 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte).
1272 	 *
1273 	 * In the general case (>256GB), we use (4 * physmem) for the
1274 	 * kernel's virtual addresses, which is divided approximately
1275 	 * as follows:
1276 	 *  - 1 * physmem for segkpm
1277 	 *  - 1.5 * physmem for segzio
1278 	 *  - 1.5 * physmem for heap
1279 	 * Total: 4.0 * physmem
1280 	 *
1281 	 * Note that the segzio and heap sizes are more than physmem so that
1282 	 * VA fragmentation does not prevent either of them from being
1283 	 * able to use nearly all of physmem.  The value of 1.5x is determined
1284 	 * experimentally and may need to change if the workload changes.
1285 	 */
1286 	if (physmax + 1 > mmu_btop(TERABYTE / 4) ||
1287 	    plat_dr_physmax > mmu_btop(TERABYTE / 4)) {
1288 		uint64_t kpm_resv_amount = mmu_ptob(physmax + 1);
1289 
1290 		if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) {
1291 			kpm_resv_amount = mmu_ptob(plat_dr_physmax);
1292 		}
1293 
1294 		/*
1295 		 * This is what actually controls the KVA : UVA split.
1296 		 * The kernel uses high VA, and this is lowering the
1297 		 * boundary, thus increasing the amount of VA for the kernel.
1298 		 * This gives the kernel 4 * (amount of physical memory) VA.
1299 		 *
1300 		 * The maximum VA is UINT64_MAX and we are using
1301 		 * 64-bit 2's complement math, so e.g. if you have 512GB
1302 		 * of memory, segkpm_base = -(4 * 512GB) == -2TB ==
1303 		 * UINT64_MAX - 2TB (approximately).  So the kernel's
1304 		 * VA is [UINT64_MAX-2TB to UINT64_MAX].
1305 		 */
1306 		segkpm_base = -(P2ROUNDUP((4 * kpm_resv_amount),
1307 		    KERNEL_REDZONE_SIZE));
1308 
1309 		/* make sure we leave some space for user apps above hole */
1310 		segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE);
1311 		if (segkpm_base > SEGKPM_BASE)
1312 			segkpm_base = SEGKPM_BASE;
1313 		PRM_DEBUG(segkpm_base);
1314 
1315 		valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG);
1316 		if (valloc_base < segkpm_base)
1317 			panic("not enough kernel VA to support memory size");
1318 		PRM_DEBUG(valloc_base);
1319 	}
1320 #else	/* __i386 */
1321 	valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz);
1322 	valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]);
1323 	PRM_DEBUG(valloc_base);
1324 #endif	/* __i386 */
1325 
1326 	/*
1327 	 * do all the initial allocations
1328 	 */
1329 	perform_allocations();
1330 
1331 	/*
1332 	 * Build phys_install and phys_avail in kernel memspace.
1333 	 * - phys_install should be all memory in the system.
1334 	 * - phys_avail is phys_install minus any memory mapped before this
1335 	 *    point above KERNEL_TEXT.
1336 	 */
1337 	current = phys_install = memlist;
1338 	copy_memlist_filter(bootops->boot_mem->physinstalled, &current, NULL);
1339 	if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1340 		panic("physinstalled was too big!");
1341 	if (prom_debug)
1342 		print_memlist("phys_install", phys_install);
1343 
1344 	phys_avail = current;
1345 	PRM_POINT("Building phys_avail:\n");
1346 	copy_memlist_filter(bootops->boot_mem->physinstalled, &current,
1347 	    avail_filter);
1348 	if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1349 		panic("physavail was too big!");
1350 	if (prom_debug)
1351 		print_memlist("phys_avail", phys_avail);
1352 #ifndef	__xpv
1353 	/*
1354 	 * Free unused memlist items, which may be used by memory DR driver
1355 	 * at runtime.
1356 	 */
1357 	if ((caddr_t)current < (caddr_t)memlist + memlist_sz) {
1358 		memlist_free_block((caddr_t)current,
1359 		    (caddr_t)memlist + memlist_sz - (caddr_t)current);
1360 	}
1361 #endif
1362 
1363 	/*
1364 	 * Build bios reserved memspace
1365 	 */
1366 	current = bios_rsvd;
1367 	copy_memlist_filter(bootops->boot_mem->rsvdmem, &current, NULL);
1368 	if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz)
1369 		panic("bios_rsvd was too big!");
1370 	if (prom_debug)
1371 		print_memlist("bios_rsvd", bios_rsvd);
1372 #ifndef	__xpv
1373 	/*
1374 	 * Free unused memlist items, which may be used by memory DR driver
1375 	 * at runtime.
1376 	 */
1377 	if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) {
1378 		memlist_free_block((caddr_t)current,
1379 		    (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current);
1380 	}
1381 #endif
1382 
1383 	/*
1384 	 * setup page coloring
1385 	 */
1386 	page_coloring_setup(pagecolor_mem);
1387 	page_lock_init();	/* currently a no-op */
1388 
1389 	/*
1390 	 * free page list counters
1391 	 */
1392 	(void) page_ctrs_alloc(page_ctrs_mem);
1393 
1394 	/*
1395 	 * Size the pcf array based on the number of cpus in the box at
1396 	 * boot time.
1397 	 */
1398 
1399 	pcf_init();
1400 
1401 	/*
1402 	 * Initialize the page structures from the memory lists.
1403 	 */
1404 	availrmem_initial = availrmem = freemem = 0;
1405 	PRM_POINT("Calling kphysm_init()...");
1406 	npages = kphysm_init(pp_base, npages);
1407 	PRM_POINT("kphysm_init() done");
1408 	PRM_DEBUG(npages);
1409 
1410 	init_debug_info();
1411 
1412 	/*
1413 	 * Now that page_t's have been initialized, remove all the
1414 	 * initial allocation pages from the kernel free page lists.
1415 	 */
1416 	boot_mapin((caddr_t)valloc_base, valloc_sz);
1417 	boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE);
1418 	PRM_POINT("startup_memlist() done");
1419 
1420 	PRM_DEBUG(valloc_sz);
1421 
1422 #if defined(__amd64)
1423 	if ((availrmem >> (30 - MMU_PAGESHIFT)) >=
1424 	    textrepl_min_gb && l2cache_sz <= 2 << 20) {
1425 		extern size_t textrepl_size_thresh;
1426 		textrepl_size_thresh = (16 << 20) - 1;
1427 	}
1428 #endif
1429 }
1430 
1431 /*
1432  * Layout the kernel's part of address space and initialize kmem allocator.
1433  */
1434 static void
1435 startup_kmem(void)
1436 {
1437 	extern void page_set_colorequiv_arr(void);
1438 #if !defined(__xpv)
1439 	extern uint64_t kpti_kbase;
1440 #endif
1441 
1442 	PRM_POINT("startup_kmem() starting...");
1443 
1444 #if defined(__amd64)
1445 	if (eprom_kernelbase && eprom_kernelbase != KERNELBASE)
1446 		cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit "
1447 		    "systems.");
1448 	kernelbase = segkpm_base - KERNEL_REDZONE_SIZE;
1449 	core_base = (uintptr_t)COREHEAP_BASE;
1450 	core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE;
1451 #else	/* __i386 */
1452 	/*
1453 	 * We configure kernelbase based on:
1454 	 *
1455 	 * 1. user specified kernelbase via eeprom command. Value cannot exceed
1456 	 *    KERNELBASE_MAX. we large page align eprom_kernelbase
1457 	 *
1458 	 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t.
1459 	 *    On large memory systems we must lower kernelbase to allow
1460 	 *    enough room for page_t's for all of memory.
1461 	 *
1462 	 * The value set here, might be changed a little later.
1463 	 */
1464 	if (eprom_kernelbase) {
1465 		kernelbase = eprom_kernelbase & mmu.level_mask[1];
1466 		if (kernelbase > KERNELBASE_MAX)
1467 			kernelbase = KERNELBASE_MAX;
1468 	} else {
1469 		kernelbase = (uintptr_t)KERNELBASE;
1470 		kernelbase -= ROUND_UP_4MEG(2 * valloc_sz);
1471 	}
1472 	ASSERT((kernelbase & mmu.level_offset[1]) == 0);
1473 	core_base = valloc_base;
1474 	core_size = 0;
1475 #endif	/* __i386 */
1476 
1477 	PRM_DEBUG(core_base);
1478 	PRM_DEBUG(core_size);
1479 	PRM_DEBUG(kernelbase);
1480 
1481 #if defined(__i386)
1482 	segkp_fromheap = 1;
1483 #endif	/* __i386 */
1484 
1485 	ekernelheap = (char *)core_base;
1486 	PRM_DEBUG(ekernelheap);
1487 
1488 	/*
1489 	 * Now that we know the real value of kernelbase,
1490 	 * update variables that were initialized with a value of
1491 	 * KERNELBASE (in common/conf/param.c).
1492 	 *
1493 	 * XXX	The problem with this sort of hackery is that the
1494 	 *	compiler just may feel like putting the const declarations
1495 	 *	(in param.c) into the .text section.  Perhaps they should
1496 	 *	just be declared as variables there?
1497 	 */
1498 
1499 	*(uintptr_t *)&_kernelbase = kernelbase;
1500 	*(uintptr_t *)&_userlimit = kernelbase;
1501 #if defined(__amd64)
1502 	*(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT;
1503 #if !defined(__xpv)
1504 	kpti_kbase = kernelbase;
1505 #endif
1506 #else
1507 	*(uintptr_t *)&_userlimit32 = _userlimit;
1508 #endif
1509 	PRM_DEBUG(_kernelbase);
1510 	PRM_DEBUG(_userlimit);
1511 	PRM_DEBUG(_userlimit32);
1512 
1513 	/* We have to re-do this now that we've modified _userlimit. */
1514 	mmu_calc_user_slots();
1515 
1516 	layout_kernel_va();
1517 
1518 #if defined(__i386)
1519 	/*
1520 	 * If segmap is too large we can push the bottom of the kernel heap
1521 	 * higher than the base.  Or worse, it could exceed the top of the
1522 	 * VA space entirely, causing it to wrap around.
1523 	 */
1524 	if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase)
1525 		panic("too little address space available for kernelheap,"
1526 		    " use eeprom for lower kernelbase or smaller segmapsize");
1527 #endif	/* __i386 */
1528 
1529 	/*
1530 	 * Initialize the kernel heap. Note 3rd argument must be > 1st.
1531 	 */
1532 	kernelheap_init(kernelheap, ekernelheap,
1533 	    kernelheap + MMU_PAGESIZE,
1534 	    (void *)core_base, (void *)(core_base + core_size));
1535 
1536 #if defined(__xpv)
1537 	/*
1538 	 * Link pending events struct into cpu struct
1539 	 */
1540 	CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data;
1541 #endif
1542 	/*
1543 	 * Initialize kernel memory allocator.
1544 	 */
1545 	kmem_init();
1546 
1547 	/*
1548 	 * Factor in colorequiv to check additional 'equivalent' bins
1549 	 */
1550 	page_set_colorequiv_arr();
1551 
1552 	/*
1553 	 * print this out early so that we know what's going on
1554 	 */
1555 	print_x86_featureset(x86_featureset);
1556 
1557 	/*
1558 	 * Initialize bp_mapin().
1559 	 */
1560 	bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK);
1561 
1562 	/*
1563 	 * orig_npages is non-zero if physmem has been configured for less
1564 	 * than the available memory.
1565 	 */
1566 	if (orig_npages) {
1567 		cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages",
1568 		    (npages == PHYSMEM ? "Due to virtual address space " : ""),
1569 		    npages, orig_npages);
1570 	}
1571 #if defined(__i386)
1572 	if (eprom_kernelbase && (eprom_kernelbase != kernelbase))
1573 		cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, "
1574 		    "System using 0x%lx",
1575 		    (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase);
1576 #endif
1577 
1578 #ifdef	KERNELBASE_ABI_MIN
1579 	if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) {
1580 		cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not "
1581 		    "i386 ABI compliant.", (uintptr_t)kernelbase);
1582 	}
1583 #endif
1584 
1585 #ifndef __xpv
1586 	if (plat_dr_support_memory()) {
1587 		mem_config_init();
1588 	}
1589 #else	/* __xpv */
1590 	/*
1591 	 * Some of the xen start information has to be relocated up
1592 	 * into the kernel's permanent address space.
1593 	 */
1594 	PRM_POINT("calling xen_relocate_start_info()");
1595 	xen_relocate_start_info();
1596 	PRM_POINT("xen_relocate_start_info() done");
1597 
1598 	/*
1599 	 * (Update the vcpu pointer in our cpu structure to point into
1600 	 * the relocated shared info.)
1601 	 */
1602 	CPU->cpu_m.mcpu_vcpu_info =
1603 	    &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id];
1604 #endif	/* __xpv */
1605 
1606 	PRM_POINT("startup_kmem() done");
1607 }
1608 
1609 #ifndef __xpv
1610 /*
1611  * If we have detected that we are running in an HVM environment, we need
1612  * to prepend the PV driver directory to the module search path.
1613  */
1614 #define	HVM_MOD_DIR "/platform/i86hvm/kernel"
1615 static void
1616 update_default_path()
1617 {
1618 	char *current, *newpath;
1619 	int newlen;
1620 
1621 	/*
1622 	 * We are about to resync with krtld.  krtld will reset its
1623 	 * internal module search path iff Solaris has set default_path.
1624 	 * We want to be sure we're prepending this new directory to the
1625 	 * right search path.
1626 	 */
1627 	current = (default_path == NULL) ? kobj_module_path : default_path;
1628 
1629 	newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2;
1630 	newpath = kmem_alloc(newlen, KM_SLEEP);
1631 	(void) strcpy(newpath, HVM_MOD_DIR);
1632 	(void) strcat(newpath, " ");
1633 	(void) strcat(newpath, current);
1634 
1635 	default_path = newpath;
1636 }
1637 #endif
1638 
1639 static void
1640 startup_modules(void)
1641 {
1642 	int cnt;
1643 	extern void prom_setup(void);
1644 	int32_t v, h;
1645 	char d[11];
1646 	char *cp;
1647 	cmi_hdl_t hdl;
1648 
1649 	PRM_POINT("startup_modules() starting...");
1650 
1651 #ifndef __xpv
1652 	/*
1653 	 * Initialize ten-micro second timer so that drivers will
1654 	 * not get short changed in their init phase. This was
1655 	 * not getting called until clkinit which, on fast cpu's
1656 	 * caused the drv_usecwait to be way too short.
1657 	 */
1658 	microfind();
1659 
1660 	if ((get_hwenv() & HW_XEN_HVM) != 0)
1661 		update_default_path();
1662 #endif
1663 
1664 	/*
1665 	 * Read the GMT lag from /etc/rtc_config.
1666 	 */
1667 	sgmtl(process_rtc_config_file());
1668 
1669 	/*
1670 	 * Calculate default settings of system parameters based upon
1671 	 * maxusers, yet allow to be overridden via the /etc/system file.
1672 	 */
1673 	param_calc(0);
1674 
1675 	mod_setup();
1676 
1677 	/*
1678 	 * Initialize system parameters.
1679 	 */
1680 	param_init();
1681 
1682 	/*
1683 	 * Initialize the default brands
1684 	 */
1685 	brand_init();
1686 
1687 	/*
1688 	 * maxmem is the amount of physical memory we're playing with.
1689 	 */
1690 	maxmem = physmem;
1691 
1692 	/*
1693 	 * Initialize segment management stuff.
1694 	 */
1695 	seg_init();
1696 
1697 	if (modload("fs", "specfs") == -1)
1698 		halt("Can't load specfs");
1699 
1700 	if (modload("fs", "devfs") == -1)
1701 		halt("Can't load devfs");
1702 
1703 	if (modload("fs", "dev") == -1)
1704 		halt("Can't load dev");
1705 
1706 	if (modload("fs", "procfs") == -1)
1707 		halt("Can't load procfs");
1708 
1709 	(void) modloadonly("sys", "lbl_edition");
1710 
1711 	dispinit();
1712 
1713 	/* Read cluster configuration data. */
1714 	clconf_init();
1715 
1716 #if defined(__xpv)
1717 	(void) ec_init();
1718 	gnttab_init();
1719 	(void) xs_early_init();
1720 #endif /* __xpv */
1721 
1722 	/*
1723 	 * Create a kernel device tree. First, create rootnex and
1724 	 * then invoke bus specific code to probe devices.
1725 	 */
1726 	setup_ddi();
1727 
1728 #ifdef __xpv
1729 	if (DOMAIN_IS_INITDOMAIN(xen_info))
1730 #endif
1731 	{
1732 		id_t smid;
1733 		smbios_system_t smsys;
1734 		smbios_info_t sminfo;
1735 		char *mfg;
1736 		/*
1737 		 * Load the System Management BIOS into the global ksmbios
1738 		 * handle, if an SMBIOS is present on this system.
1739 		 * Also set "si-hw-provider" property, if not already set.
1740 		 */
1741 		ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL);
1742 		if (ksmbios != NULL &&
1743 		    ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) &&
1744 		    (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) {
1745 			mfg = (char *)sminfo.smbi_manufacturer;
1746 			if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) {
1747 				extern char hw_provider[];
1748 				int i;
1749 				for (i = 0; i < SYS_NMLN; i++) {
1750 					if (isprint(mfg[i]))
1751 						hw_provider[i] = mfg[i];
1752 					else {
1753 						hw_provider[i] = '\0';
1754 						break;
1755 					}
1756 				}
1757 				hw_provider[SYS_NMLN - 1] = '\0';
1758 			}
1759 		}
1760 	}
1761 
1762 
1763 	/*
1764 	 * Originally clconf_init() apparently needed the hostid.  But
1765 	 * this no longer appears to be true - it uses its own nodeid.
1766 	 * By placing the hostid logic here, we are able to make use of
1767 	 * the SMBIOS UUID.
1768 	 */
1769 	if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) {
1770 		cmn_err(CE_WARN, "Unable to set hostid");
1771 	} else {
1772 		for (v = h, cnt = 0; cnt < 10; cnt++) {
1773 			d[cnt] = (char)(v % 10);
1774 			v /= 10;
1775 			if (v == 0)
1776 				break;
1777 		}
1778 		for (cp = hw_serial; cnt >= 0; cnt--)
1779 			*cp++ = d[cnt] + '0';
1780 		*cp = 0;
1781 	}
1782 
1783 	/*
1784 	 * Set up the CPU module subsystem for the boot cpu in the native
1785 	 * case, and all physical cpu resource in the xpv dom0 case.
1786 	 * Modifies the device tree, so this must be done after
1787 	 * setup_ddi().
1788 	 */
1789 #ifdef __xpv
1790 	/*
1791 	 * If paravirtualized and on dom0 then we initialize all physical
1792 	 * cpu handles now;  if paravirtualized on a domU then do not
1793 	 * initialize.
1794 	 */
1795 	if (DOMAIN_IS_INITDOMAIN(xen_info)) {
1796 		xen_mc_lcpu_cookie_t cpi;
1797 
1798 		for (cpi = xen_physcpu_next(NULL); cpi != NULL;
1799 		    cpi = xen_physcpu_next(cpi)) {
1800 			if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA,
1801 			    xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi),
1802 			    xen_physcpu_strandid(cpi))) != NULL &&
1803 			    is_x86_feature(x86_featureset, X86FSET_MCA))
1804 				cmi_mca_init(hdl);
1805 		}
1806 	}
1807 #else
1808 	/*
1809 	 * Initialize a handle for the boot cpu - others will initialize
1810 	 * as they startup.
1811 	 */
1812 	if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
1813 	    cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL) {
1814 		if (is_x86_feature(x86_featureset, X86FSET_MCA))
1815 			cmi_mca_init(hdl);
1816 		CPU->cpu_m.mcpu_cmi_hdl = hdl;
1817 	}
1818 #endif	/* __xpv */
1819 
1820 	/*
1821 	 * Fake a prom tree such that /dev/openprom continues to work
1822 	 */
1823 	PRM_POINT("startup_modules: calling prom_setup...");
1824 	prom_setup();
1825 	PRM_POINT("startup_modules: done");
1826 
1827 	/*
1828 	 * Load all platform specific modules
1829 	 */
1830 	PRM_POINT("startup_modules: calling psm_modload...");
1831 	psm_modload();
1832 
1833 	PRM_POINT("startup_modules() done");
1834 }
1835 
1836 /*
1837  * claim a "setaside" boot page for use in the kernel
1838  */
1839 page_t *
1840 boot_claim_page(pfn_t pfn)
1841 {
1842 	page_t *pp;
1843 
1844 	pp = page_numtopp_nolock(pfn);
1845 	ASSERT(pp != NULL);
1846 
1847 	if (PP_ISBOOTPAGES(pp)) {
1848 		if (pp->p_next != NULL)
1849 			pp->p_next->p_prev = pp->p_prev;
1850 		if (pp->p_prev == NULL)
1851 			bootpages = pp->p_next;
1852 		else
1853 			pp->p_prev->p_next = pp->p_next;
1854 	} else {
1855 		/*
1856 		 * htable_attach() expects a base pagesize page
1857 		 */
1858 		if (pp->p_szc != 0)
1859 			page_boot_demote(pp);
1860 		pp = page_numtopp(pfn, SE_EXCL);
1861 	}
1862 	return (pp);
1863 }
1864 
1865 /*
1866  * Walk through the pagetables looking for pages mapped in by boot.  If the
1867  * setaside flag is set the pages are expected to be returned to the
1868  * kernel later in boot, so we add them to the bootpages list.
1869  */
1870 static void
1871 protect_boot_range(uintptr_t low, uintptr_t high, int setaside)
1872 {
1873 	uintptr_t va = low;
1874 	size_t len;
1875 	uint_t prot;
1876 	pfn_t pfn;
1877 	page_t *pp;
1878 	pgcnt_t boot_protect_cnt = 0;
1879 
1880 	while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) {
1881 		if (va + len >= high)
1882 			panic("0x%lx byte mapping at 0x%p exceeds boot's "
1883 			    "legal range.", len, (void *)va);
1884 
1885 		while (len > 0) {
1886 			pp = page_numtopp_alloc(pfn);
1887 			if (pp != NULL) {
1888 				if (setaside == 0)
1889 					panic("Unexpected mapping by boot.  "
1890 					    "addr=%p pfn=%lx\n",
1891 					    (void *)va, pfn);
1892 
1893 				pp->p_next = bootpages;
1894 				pp->p_prev = NULL;
1895 				PP_SETBOOTPAGES(pp);
1896 				if (bootpages != NULL) {
1897 					bootpages->p_prev = pp;
1898 				}
1899 				bootpages = pp;
1900 				++boot_protect_cnt;
1901 			}
1902 
1903 			++pfn;
1904 			len -= MMU_PAGESIZE;
1905 			va += MMU_PAGESIZE;
1906 		}
1907 	}
1908 	PRM_DEBUG(boot_protect_cnt);
1909 }
1910 
1911 /*
1912  *
1913  */
1914 static void
1915 layout_kernel_va(void)
1916 {
1917 	PRM_POINT("layout_kernel_va() starting...");
1918 	/*
1919 	 * Establish the final size of the kernel's heap, size of segmap,
1920 	 * segkp, etc.
1921 	 */
1922 
1923 #if defined(__amd64)
1924 
1925 	kpm_vbase = (caddr_t)segkpm_base;
1926 	if (physmax + 1 < plat_dr_physmax) {
1927 		kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax));
1928 	} else {
1929 		kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1));
1930 	}
1931 	if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base)
1932 		panic("not enough room for kpm!");
1933 	PRM_DEBUG(kpm_size);
1934 	PRM_DEBUG(kpm_vbase);
1935 
1936 	/*
1937 	 * By default we create a seg_kp in 64 bit kernels, it's a little
1938 	 * faster to access than embedding it in the heap.
1939 	 */
1940 	segkp_base = (caddr_t)valloc_base + valloc_sz;
1941 	if (!segkp_fromheap) {
1942 		size_t sz = mmu_ptob(segkpsize);
1943 
1944 		/*
1945 		 * determine size of segkp
1946 		 */
1947 		if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) {
1948 			sz = SEGKPDEFSIZE;
1949 			cmn_err(CE_WARN, "!Illegal value for segkpsize. "
1950 			    "segkpsize has been reset to %ld pages",
1951 			    mmu_btop(sz));
1952 		}
1953 		sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem)));
1954 
1955 		segkpsize = mmu_btop(ROUND_UP_LPAGE(sz));
1956 	}
1957 	PRM_DEBUG(segkp_base);
1958 	PRM_DEBUG(segkpsize);
1959 
1960 	/*
1961 	 * segzio is used for ZFS cached data. It uses a distinct VA
1962 	 * segment (from kernel heap) so that we can easily tell not to
1963 	 * include it in kernel crash dumps on 64 bit kernels. The trick is
1964 	 * to give it lots of VA, but not constrain the kernel heap.
1965 	 * We can use 1.5x physmem for segzio, leaving approximately
1966 	 * another 1.5x physmem for heap.  See also the comment in
1967 	 * startup_memlist().
1968 	 */
1969 	segzio_base = segkp_base + mmu_ptob(segkpsize);
1970 	if (segzio_fromheap) {
1971 		segziosize = 0;
1972 	} else {
1973 		size_t physmem_size = mmu_ptob(physmem);
1974 		size_t size = (segziosize == 0) ?
1975 		    physmem_size * 3 / 2 : mmu_ptob(segziosize);
1976 
1977 		if (size < SEGZIOMINSIZE)
1978 			size = SEGZIOMINSIZE;
1979 		segziosize = mmu_btop(ROUND_UP_LPAGE(size));
1980 	}
1981 	PRM_DEBUG(segziosize);
1982 	PRM_DEBUG(segzio_base);
1983 
1984 	/*
1985 	 * Put the range of VA for device mappings next, kmdb knows to not
1986 	 * grep in this range of addresses.
1987 	 */
1988 	toxic_addr =
1989 	    ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize));
1990 	PRM_DEBUG(toxic_addr);
1991 	segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size);
1992 #else /* __i386 */
1993 	segmap_start = ROUND_UP_LPAGE(kernelbase);
1994 #endif /* __i386 */
1995 	PRM_DEBUG(segmap_start);
1996 
1997 	/*
1998 	 * Users can change segmapsize through eeprom. If the variable
1999 	 * is tuned through eeprom, there is no upper bound on the
2000 	 * size of segmap.
2001 	 */
2002 	segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT);
2003 
2004 #if defined(__i386)
2005 	/*
2006 	 * 32-bit systems don't have segkpm or segkp, so segmap appears at
2007 	 * the bottom of the kernel's address range.  Set aside space for a
2008 	 * small red zone just below the start of segmap.
2009 	 */
2010 	segmap_start += KERNEL_REDZONE_SIZE;
2011 	segmapsize -= KERNEL_REDZONE_SIZE;
2012 #endif
2013 
2014 	PRM_DEBUG(segmap_start);
2015 	PRM_DEBUG(segmapsize);
2016 	kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize);
2017 	PRM_DEBUG(kernelheap);
2018 	PRM_POINT("layout_kernel_va() done...");
2019 }
2020 
2021 /*
2022  * Finish initializing the VM system, now that we are no longer
2023  * relying on the boot time memory allocators.
2024  */
2025 static void
2026 startup_vm(void)
2027 {
2028 	struct segmap_crargs a;
2029 
2030 	extern int use_brk_lpg, use_stk_lpg;
2031 
2032 	PRM_POINT("startup_vm() starting...");
2033 
2034 	/*
2035 	 * Initialize the hat layer.
2036 	 */
2037 	hat_init();
2038 
2039 	/*
2040 	 * Do final allocations of HAT data structures that need to
2041 	 * be allocated before quiescing the boot loader.
2042 	 */
2043 	PRM_POINT("Calling hat_kern_alloc()...");
2044 	hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap);
2045 	PRM_POINT("hat_kern_alloc() done");
2046 
2047 #ifndef __xpv
2048 	/*
2049 	 * Setup Page Attribute Table
2050 	 */
2051 	pat_sync();
2052 #endif
2053 
2054 	/*
2055 	 * The next two loops are done in distinct steps in order
2056 	 * to be sure that any page that is doubly mapped (both above
2057 	 * KERNEL_TEXT and below kernelbase) is dealt with correctly.
2058 	 * Note this may never happen, but it might someday.
2059 	 */
2060 	bootpages = NULL;
2061 	PRM_POINT("Protecting boot pages");
2062 
2063 	/*
2064 	 * Protect any pages mapped above KERNEL_TEXT that somehow have
2065 	 * page_t's. This can only happen if something weird allocated
2066 	 * in this range (like kadb/kmdb).
2067 	 */
2068 	protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0);
2069 
2070 	/*
2071 	 * Before we can take over memory allocation/mapping from the boot
2072 	 * loader we must remove from our free page lists any boot allocated
2073 	 * pages that stay mapped until release_bootstrap().
2074 	 */
2075 	protect_boot_range(0, kernelbase, 1);
2076 
2077 
2078 	/*
2079 	 * Switch to running on regular HAT (not boot_mmu)
2080 	 */
2081 	PRM_POINT("Calling hat_kern_setup()...");
2082 	hat_kern_setup();
2083 
2084 	/*
2085 	 * It is no longer safe to call BOP_ALLOC(), so make sure we don't.
2086 	 */
2087 	bop_no_more_mem();
2088 
2089 	PRM_POINT("hat_kern_setup() done");
2090 
2091 	hat_cpu_online(CPU);
2092 
2093 	/*
2094 	 * Initialize VM system
2095 	 */
2096 	PRM_POINT("Calling kvm_init()...");
2097 	kvm_init();
2098 	PRM_POINT("kvm_init() done");
2099 
2100 	/*
2101 	 * Tell kmdb that the VM system is now working
2102 	 */
2103 	if (boothowto & RB_DEBUG)
2104 		kdi_dvec_vmready();
2105 
2106 #if defined(__xpv)
2107 	/*
2108 	 * Populate the I/O pool on domain 0
2109 	 */
2110 	if (DOMAIN_IS_INITDOMAIN(xen_info)) {
2111 		extern long populate_io_pool(void);
2112 		long init_io_pool_cnt;
2113 
2114 		PRM_POINT("Populating reserve I/O page pool");
2115 		init_io_pool_cnt = populate_io_pool();
2116 		PRM_DEBUG(init_io_pool_cnt);
2117 	}
2118 #endif
2119 	/*
2120 	 * Mangle the brand string etc.
2121 	 */
2122 	cpuid_pass3(CPU);
2123 
2124 #if defined(__amd64)
2125 
2126 	/*
2127 	 * Create the device arena for toxic (to dtrace/kmdb) mappings.
2128 	 */
2129 	device_arena = vmem_create("device", (void *)toxic_addr,
2130 	    toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP);
2131 
2132 #else	/* __i386 */
2133 
2134 	/*
2135 	 * allocate the bit map that tracks toxic pages
2136 	 */
2137 	toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase));
2138 	PRM_DEBUG(toxic_bit_map_len);
2139 	toxic_bit_map =
2140 	    kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP);
2141 	ASSERT(toxic_bit_map != NULL);
2142 	PRM_DEBUG(toxic_bit_map);
2143 
2144 #endif	/* __i386 */
2145 
2146 
2147 	/*
2148 	 * Now that we've got more VA, as well as the ability to allocate from
2149 	 * it, tell the debugger.
2150 	 */
2151 	if (boothowto & RB_DEBUG)
2152 		kdi_dvec_memavail();
2153 
2154 #if !defined(__xpv)
2155 	/*
2156 	 * Map page pfn=0 for drivers, such as kd, that need to pick up
2157 	 * parameters left there by controllers/BIOS.
2158 	 */
2159 	PRM_POINT("setup up p0_va");
2160 	p0_va = i86devmap(0, 1, PROT_READ);
2161 	PRM_DEBUG(p0_va);
2162 #endif
2163 
2164 	cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n",
2165 	    physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled));
2166 
2167 	/*
2168 	 * disable automatic large pages for small memory systems or
2169 	 * when the disable flag is set.
2170 	 *
2171 	 * Do not yet consider page sizes larger than 2m/4m.
2172 	 */
2173 	if (!auto_lpg_disable && mmu.max_page_level > 0) {
2174 		max_uheap_lpsize = LEVEL_SIZE(1);
2175 		max_ustack_lpsize = LEVEL_SIZE(1);
2176 		max_privmap_lpsize = LEVEL_SIZE(1);
2177 		max_uidata_lpsize = LEVEL_SIZE(1);
2178 		max_utext_lpsize = LEVEL_SIZE(1);
2179 		max_shm_lpsize = LEVEL_SIZE(1);
2180 	}
2181 	if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 ||
2182 	    auto_lpg_disable) {
2183 		use_brk_lpg = 0;
2184 		use_stk_lpg = 0;
2185 	}
2186 	mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level);
2187 
2188 	PRM_POINT("Calling hat_init_finish()...");
2189 	hat_init_finish();
2190 	PRM_POINT("hat_init_finish() done");
2191 
2192 	/*
2193 	 * Initialize the segkp segment type.
2194 	 */
2195 	rw_enter(&kas.a_lock, RW_WRITER);
2196 	PRM_POINT("Attaching segkp");
2197 	if (segkp_fromheap) {
2198 		segkp->s_as = &kas;
2199 	} else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize),
2200 	    segkp) < 0) {
2201 		panic("startup: cannot attach segkp");
2202 		/*NOTREACHED*/
2203 	}
2204 	PRM_POINT("Doing segkp_create()");
2205 	if (segkp_create(segkp) != 0) {
2206 		panic("startup: segkp_create failed");
2207 		/*NOTREACHED*/
2208 	}
2209 	PRM_DEBUG(segkp);
2210 	rw_exit(&kas.a_lock);
2211 
2212 	/*
2213 	 * kpm segment
2214 	 */
2215 	segmap_kpm = 0;
2216 	if (kpm_desired)
2217 		kpm_init();
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(__xpv)
2341 	/*
2342 	 * Intel IOMMU has been setup/initialized in ddi_impl.c
2343 	 * Start it up now.
2344 	 */
2345 	immu_startup();
2346 
2347 	/*
2348 	 * Now that we're no longer going to drop into real mode for a BIOS call
2349 	 * via bootops, we can enable PCID (which requires CR0.PG).
2350 	 */
2351 	enable_pcid();
2352 #endif
2353 
2354 	PRM_POINT("Enabling interrupts");
2355 	(*picinitf)();
2356 	sti();
2357 #if defined(__xpv)
2358 	ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0);
2359 	xen_late_startup();
2360 #endif
2361 
2362 	(void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1,
2363 	    "softlevel1", NULL, NULL); /* XXX to be moved later */
2364 
2365 	/*
2366 	 * Register software interrupt handlers for ddi_periodic_add(9F).
2367 	 * Software interrupts up to the level 10 are supported.
2368 	 */
2369 	for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) {
2370 		(void) add_avsoftintr((void *)&softlevel_hdl[i-1], i,
2371 		    (avfunc)(uintptr_t)ddi_periodic_softintr, "ddi_periodic",
2372 		    (caddr_t)(uintptr_t)i, NULL);
2373 	}
2374 
2375 #if !defined(__xpv)
2376 	if (modload("drv", "amd_iommu") < 0) {
2377 		PRM_POINT("No AMD IOMMU present\n");
2378 	} else if (ddi_hold_installed_driver(ddi_name_to_major(
2379 	    "amd_iommu")) == NULL) {
2380 		prom_printf("ERROR: failed to attach AMD IOMMU\n");
2381 	}
2382 #endif
2383 	post_startup_cpu_fixups();
2384 
2385 	PRM_POINT("startup_end() done");
2386 }
2387 
2388 /*
2389  * Don't remove the following 2 variables.  They are necessary
2390  * for reading the hostid from the legacy file (/kernel/misc/sysinit).
2391  */
2392 char *_hs1107 = hw_serial;
2393 ulong_t  _bdhs34;
2394 
2395 void
2396 post_startup(void)
2397 {
2398 	extern void cpupm_init(cpu_t *);
2399 	extern void cpu_event_init_cpu(cpu_t *);
2400 
2401 	/*
2402 	 * Set the system wide, processor-specific flags to be passed
2403 	 * to userland via the aux vector for performance hints and
2404 	 * instruction set extensions.
2405 	 */
2406 	bind_hwcap();
2407 
2408 #ifdef __xpv
2409 	if (DOMAIN_IS_INITDOMAIN(xen_info))
2410 #endif
2411 	{
2412 #if defined(__xpv)
2413 		xpv_panic_init();
2414 #else
2415 		/*
2416 		 * Startup the memory scrubber.
2417 		 * XXPV	This should be running somewhere ..
2418 		 */
2419 		if ((get_hwenv() & HW_VIRTUAL) == 0)
2420 			memscrub_init();
2421 #endif
2422 	}
2423 
2424 	/*
2425 	 * Complete CPU module initialization
2426 	 */
2427 	cmi_post_startup();
2428 
2429 	/*
2430 	 * Perform forceloading tasks for /etc/system.
2431 	 */
2432 	(void) mod_sysctl(SYS_FORCELOAD, NULL);
2433 
2434 	/*
2435 	 * ON4.0: Force /proc module in until clock interrupt handle fixed
2436 	 * ON4.0: This must be fixed or restated in /etc/systems.
2437 	 */
2438 	(void) modload("fs", "procfs");
2439 
2440 	(void) i_ddi_attach_hw_nodes("pit_beep");
2441 
2442 #if defined(__i386)
2443 	/*
2444 	 * Check for required functional Floating Point hardware,
2445 	 * unless FP hardware explicitly disabled.
2446 	 */
2447 	if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO))
2448 		halt("No working FP hardware found");
2449 #endif
2450 
2451 	maxmem = freemem;
2452 
2453 	cpu_event_init_cpu(CPU);
2454 	cpupm_init(CPU);
2455 	(void) mach_cpu_create_device_node(CPU, NULL);
2456 
2457 	pg_init();
2458 }
2459 
2460 static int
2461 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr)
2462 {
2463 	return ((pp->p_pagenum >= btop(low_addr)) &&
2464 	    (pp->p_pagenum < btopr(high_addr)));
2465 }
2466 
2467 static int
2468 pp_in_module(page_t *pp, const rd_existing_t *modranges)
2469 {
2470 	uint_t i;
2471 
2472 	for (i = 0; modranges[i].phys != 0; i++) {
2473 		if (pp_in_range(pp, modranges[i].phys,
2474 		    modranges[i].phys + modranges[i].size))
2475 			return (1);
2476 	}
2477 
2478 	return (0);
2479 }
2480 
2481 void
2482 release_bootstrap(void)
2483 {
2484 	int root_is_ramdisk;
2485 	page_t *pp;
2486 	extern void kobj_boot_unmountroot(void);
2487 	extern dev_t rootdev;
2488 	uint_t i;
2489 	char propname[32];
2490 	rd_existing_t *modranges;
2491 #if !defined(__xpv)
2492 	pfn_t	pfn;
2493 #endif
2494 
2495 	/*
2496 	 * Save the bootfs module ranges so that we can reserve them below
2497 	 * for the real bootfs.
2498 	 */
2499 	modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES,
2500 	    KM_SLEEP);
2501 	for (i = 0; ; i++) {
2502 		uint64_t start, size;
2503 
2504 		modranges[i].phys = 0;
2505 
2506 		(void) snprintf(propname, sizeof (propname),
2507 		    "module-addr-%u", i);
2508 		if (do_bsys_getproplen(NULL, propname) <= 0)
2509 			break;
2510 		(void) do_bsys_getprop(NULL, propname, &start);
2511 
2512 		(void) snprintf(propname, sizeof (propname),
2513 		    "module-size-%u", i);
2514 		if (do_bsys_getproplen(NULL, propname) <= 0)
2515 			break;
2516 		(void) do_bsys_getprop(NULL, propname, &size);
2517 
2518 		modranges[i].phys = start;
2519 		modranges[i].size = size;
2520 	}
2521 
2522 	/* unmount boot ramdisk and release kmem usage */
2523 	kobj_boot_unmountroot();
2524 
2525 	/*
2526 	 * We're finished using the boot loader so free its pages.
2527 	 */
2528 	PRM_POINT("Unmapping lower boot pages");
2529 
2530 	clear_boot_mappings(0, _userlimit);
2531 
2532 	postbootkernelbase = kernelbase;
2533 
2534 	/*
2535 	 * If root isn't on ramdisk, destroy the hardcoded
2536 	 * ramdisk node now and release the memory. Else,
2537 	 * ramdisk memory is kept in rd_pages.
2538 	 */
2539 	root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk"));
2540 	if (!root_is_ramdisk) {
2541 		dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0);
2542 		ASSERT(dip && ddi_get_parent(dip) == ddi_root_node());
2543 		ndi_rele_devi(dip);	/* held from ddi_find_devinfo */
2544 		(void) ddi_remove_child(dip, 0);
2545 	}
2546 
2547 	PRM_POINT("Releasing boot pages");
2548 	while (bootpages) {
2549 		extern uint64_t ramdisk_start, ramdisk_end;
2550 		pp = bootpages;
2551 		bootpages = pp->p_next;
2552 
2553 
2554 		/* Keep pages for the lower 64K */
2555 		if (pp_in_range(pp, 0, 0x40000)) {
2556 			pp->p_next = lower_pages;
2557 			lower_pages = pp;
2558 			lower_pages_count++;
2559 			continue;
2560 		}
2561 
2562 		if (root_is_ramdisk && pp_in_range(pp, ramdisk_start,
2563 		    ramdisk_end) || pp_in_module(pp, modranges)) {
2564 			pp->p_next = rd_pages;
2565 			rd_pages = pp;
2566 			continue;
2567 		}
2568 		pp->p_next = (struct page *)0;
2569 		pp->p_prev = (struct page *)0;
2570 		PP_CLRBOOTPAGES(pp);
2571 		page_free(pp, 1);
2572 	}
2573 	PRM_POINT("Boot pages released");
2574 
2575 	kmem_free(modranges, sizeof (rd_existing_t) * 99);
2576 
2577 #if !defined(__xpv)
2578 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */
2579 	/*
2580 	 * Find 1 page below 1 MB so that other processors can boot up or
2581 	 * so that any processor can resume.
2582 	 * Make sure it has a kernel VA as well as a 1:1 mapping.
2583 	 * We should have just free'd one up.
2584 	 */
2585 
2586 	/*
2587 	 * 0x10 pages is 64K.  Leave the bottom 64K alone
2588 	 * for BIOS.
2589 	 */
2590 	for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) {
2591 		if (page_numtopp_alloc(pfn) == NULL)
2592 			continue;
2593 		rm_platter_va = i86devmap(pfn, 1,
2594 		    PROT_READ | PROT_WRITE | PROT_EXEC);
2595 		rm_platter_pa = ptob(pfn);
2596 		break;
2597 	}
2598 	if (pfn == btop(1*1024*1024) && use_mp)
2599 		panic("No page below 1M available for starting "
2600 		    "other processors or for resuming from system-suspend");
2601 #endif	/* !__xpv */
2602 }
2603 
2604 /*
2605  * Initialize the platform-specific parts of a page_t.
2606  */
2607 void
2608 add_physmem_cb(page_t *pp, pfn_t pnum)
2609 {
2610 	pp->p_pagenum = pnum;
2611 	pp->p_mapping = NULL;
2612 	pp->p_embed = 0;
2613 	pp->p_share = 0;
2614 	pp->p_mlentry = 0;
2615 }
2616 
2617 /*
2618  * kphysm_init() initializes physical memory.
2619  */
2620 static pgcnt_t
2621 kphysm_init(page_t *pp, pgcnt_t npages)
2622 {
2623 	struct memlist	*pmem;
2624 	struct memseg	*cur_memseg;
2625 	pfn_t		base_pfn;
2626 	pfn_t		end_pfn;
2627 	pgcnt_t		num;
2628 	pgcnt_t		pages_done = 0;
2629 	uint64_t	addr;
2630 	uint64_t	size;
2631 	extern pfn_t	ddiphysmin;
2632 	extern int	mnode_xwa;
2633 	int		ms = 0, me = 0;
2634 
2635 	ASSERT(page_hash != NULL && page_hashsz != 0);
2636 
2637 	cur_memseg = memseg_base;
2638 	for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) {
2639 		/*
2640 		 * In a 32 bit kernel can't use higher memory if we're
2641 		 * not booting in PAE mode. This check takes care of that.
2642 		 */
2643 		addr = pmem->ml_address;
2644 		size = pmem->ml_size;
2645 		if (btop(addr) > physmax)
2646 			continue;
2647 
2648 		/*
2649 		 * align addr and size - they may not be at page boundaries
2650 		 */
2651 		if ((addr & MMU_PAGEOFFSET) != 0) {
2652 			addr += MMU_PAGEOFFSET;
2653 			addr &= ~(uint64_t)MMU_PAGEOFFSET;
2654 			size -= addr - pmem->ml_address;
2655 		}
2656 
2657 		/* only process pages below or equal to physmax */
2658 		if ((btop(addr + size) - 1) > physmax)
2659 			size = ptob(physmax - btop(addr) + 1);
2660 
2661 		num = btop(size);
2662 		if (num == 0)
2663 			continue;
2664 
2665 		if (num > npages)
2666 			num = npages;
2667 
2668 		npages -= num;
2669 		pages_done += num;
2670 		base_pfn = btop(addr);
2671 
2672 		if (prom_debug)
2673 			prom_printf("MEMSEG addr=0x%" PRIx64
2674 			    " pgs=0x%lx pfn 0x%lx-0x%lx\n",
2675 			    addr, num, base_pfn, base_pfn + num);
2676 
2677 		/*
2678 		 * Ignore pages below ddiphysmin to simplify ddi memory
2679 		 * allocation with non-zero addr_lo requests.
2680 		 */
2681 		if (base_pfn < ddiphysmin) {
2682 			if (base_pfn + num <= ddiphysmin)
2683 				continue;
2684 			pp += (ddiphysmin - base_pfn);
2685 			num -= (ddiphysmin - base_pfn);
2686 			base_pfn = ddiphysmin;
2687 		}
2688 
2689 		/*
2690 		 * mnode_xwa is greater than 1 when large pages regions can
2691 		 * cross memory node boundaries. To prevent the formation
2692 		 * of these large pages, configure the memsegs based on the
2693 		 * memory node ranges which had been made non-contiguous.
2694 		 */
2695 		end_pfn = base_pfn + num - 1;
2696 		if (mnode_xwa > 1) {
2697 			ms = PFN_2_MEM_NODE(base_pfn);
2698 			me = PFN_2_MEM_NODE(end_pfn);
2699 
2700 			if (ms != me) {
2701 				/*
2702 				 * current range spans more than 1 memory node.
2703 				 * Set num to only the pfn range in the start
2704 				 * memory node.
2705 				 */
2706 				num = mem_node_config[ms].physmax - base_pfn
2707 				    + 1;
2708 				ASSERT(end_pfn > mem_node_config[ms].physmax);
2709 			}
2710 		}
2711 
2712 		for (;;) {
2713 			/*
2714 			 * Build the memsegs entry
2715 			 */
2716 			cur_memseg->pages = pp;
2717 			cur_memseg->epages = pp + num;
2718 			cur_memseg->pages_base = base_pfn;
2719 			cur_memseg->pages_end = base_pfn + num;
2720 
2721 			/*
2722 			 * Insert into memseg list in decreasing pfn range
2723 			 * order. Low memory is typically more fragmented such
2724 			 * that this ordering keeps the larger ranges at the
2725 			 * front of the list for code that searches memseg.
2726 			 * This ASSERTS that the memsegs coming in from boot
2727 			 * are in increasing physical address order and not
2728 			 * contiguous.
2729 			 */
2730 			if (memsegs != NULL) {
2731 				ASSERT(cur_memseg->pages_base >=
2732 				    memsegs->pages_end);
2733 				cur_memseg->next = memsegs;
2734 			}
2735 			memsegs = cur_memseg;
2736 
2737 			/*
2738 			 * add_physmem() initializes the PSM part of the page
2739 			 * struct by calling the PSM back with add_physmem_cb().
2740 			 * In addition it coalesces pages into larger pages as
2741 			 * it initializes them.
2742 			 */
2743 			add_physmem(pp, num, base_pfn);
2744 			cur_memseg++;
2745 			availrmem_initial += num;
2746 			availrmem += num;
2747 
2748 			pp += num;
2749 			if (ms >= me)
2750 				break;
2751 
2752 			/* process next memory node range */
2753 			ms++;
2754 			base_pfn = mem_node_config[ms].physbase;
2755 
2756 			if (mnode_xwa > 1) {
2757 				num = MIN(mem_node_config[ms].physmax,
2758 				    end_pfn) - base_pfn + 1;
2759 			} else {
2760 				num = mem_node_config[ms].physmax -
2761 				    base_pfn + 1;
2762 			}
2763 		}
2764 	}
2765 
2766 	PRM_DEBUG(availrmem_initial);
2767 	PRM_DEBUG(availrmem);
2768 	PRM_DEBUG(freemem);
2769 	build_pfn_hash();
2770 	return (pages_done);
2771 }
2772 
2773 /*
2774  * Kernel VM initialization.
2775  */
2776 static void
2777 kvm_init(void)
2778 {
2779 	ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0);
2780 
2781 	/*
2782 	 * Put the kernel segments in kernel address space.
2783 	 */
2784 	rw_enter(&kas.a_lock, RW_WRITER);
2785 	as_avlinit(&kas);
2786 
2787 	(void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg);
2788 	(void) segkmem_create(&ktextseg);
2789 
2790 	(void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc);
2791 	(void) segkmem_create(&kvalloc);
2792 
2793 	(void) seg_attach(&kas, kernelheap,
2794 	    ekernelheap - kernelheap, &kvseg);
2795 	(void) segkmem_create(&kvseg);
2796 
2797 	if (core_size > 0) {
2798 		PRM_POINT("attaching kvseg_core");
2799 		(void) seg_attach(&kas, (caddr_t)core_base, core_size,
2800 		    &kvseg_core);
2801 		(void) segkmem_create(&kvseg_core);
2802 	}
2803 
2804 	if (segziosize > 0) {
2805 		PRM_POINT("attaching segzio");
2806 		(void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2807 		    &kzioseg);
2808 		(void) segkmem_zio_create(&kzioseg);
2809 
2810 		/* create zio area covering new segment */
2811 		segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2812 	}
2813 
2814 	(void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2815 	(void) segkmem_create(&kdebugseg);
2816 
2817 	rw_exit(&kas.a_lock);
2818 
2819 	/*
2820 	 * Ensure that the red zone at kernelbase is never accessible.
2821 	 */
2822 	PRM_POINT("protecting redzone");
2823 	(void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0);
2824 
2825 	/*
2826 	 * Make the text writable so that it can be hot patched by DTrace.
2827 	 */
2828 	(void) as_setprot(&kas, s_text, e_modtext - s_text,
2829 	    PROT_READ | PROT_WRITE | PROT_EXEC);
2830 
2831 	/*
2832 	 * Make data writable until end.
2833 	 */
2834 	(void) as_setprot(&kas, s_data, e_moddata - s_data,
2835 	    PROT_READ | PROT_WRITE | PROT_EXEC);
2836 }
2837 
2838 #ifndef __xpv
2839 /*
2840  * Solaris adds an entry for Write Combining caching to the PAT
2841  */
2842 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE;
2843 
2844 void
2845 pat_sync(void)
2846 {
2847 	ulong_t	cr0, cr0_orig, cr4;
2848 
2849 	if (!is_x86_feature(x86_featureset, X86FSET_PAT))
2850 		return;
2851 	cr0_orig = cr0 = getcr0();
2852 	cr4 = getcr4();
2853 
2854 	/* disable caching and flush all caches and TLBs */
2855 	cr0 |= CR0_CD;
2856 	cr0 &= ~CR0_NW;
2857 	setcr0(cr0);
2858 	invalidate_cache();
2859 	if (cr4 & CR4_PGE) {
2860 		setcr4(cr4 & ~(ulong_t)CR4_PGE);
2861 		setcr4(cr4);
2862 	} else {
2863 		reload_cr3();
2864 	}
2865 
2866 	/* add our entry to the PAT */
2867 	wrmsr(REG_PAT, pat_attr_reg);
2868 
2869 	/* flush TLBs and cache again, then reenable cr0 caching */
2870 	if (cr4 & CR4_PGE) {
2871 		setcr4(cr4 & ~(ulong_t)CR4_PGE);
2872 		setcr4(cr4);
2873 	} else {
2874 		reload_cr3();
2875 	}
2876 	invalidate_cache();
2877 	setcr0(cr0_orig);
2878 }
2879 
2880 #endif /* !__xpv */
2881 
2882 #if defined(_SOFT_HOSTID)
2883 /*
2884  * On platforms that do not have a hardware serial number, attempt
2885  * to set one based on the contents of /etc/hostid.  If this file does
2886  * not exist, assume that we are to generate a new hostid and set
2887  * it in the kernel, for subsequent saving by a userland process
2888  * once the system is up and the root filesystem is mounted r/w.
2889  *
2890  * In order to gracefully support upgrade on OpenSolaris, if
2891  * /etc/hostid does not exist, we will attempt to get a serial number
2892  * using the legacy method (/kernel/misc/sysinit).
2893  *
2894  * If that isn't present, we attempt to use an SMBIOS UUID, which is
2895  * a hardware serial number.  Note that we don't automatically trust
2896  * all SMBIOS UUIDs (some older platforms are defective and ship duplicate
2897  * UUIDs in violation of the standard), we check against a blacklist.
2898  *
2899  * In an attempt to make the hostid less prone to abuse
2900  * (for license circumvention, etc), we store it in /etc/hostid
2901  * in rot47 format.
2902  */
2903 extern volatile unsigned long tenmicrodata;
2904 static int atoi(char *);
2905 
2906 /*
2907  * Set this to non-zero in /etc/system if you think your SMBIOS returns a
2908  * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist
2909  * array can be updated.)
2910  */
2911 int smbios_broken_uuid = 0;
2912 
2913 /*
2914  * List of known bad UUIDs.  This is just the lower 32-bit values, since
2915  * that's what we use for the host id.  If your hostid falls here, you need
2916  * to contact your hardware OEM for a fix for your BIOS.
2917  */
2918 static unsigned char
2919 smbios_uuid_blacklist[][16] = {
2920 
2921 	{	/* Reported bad UUID (Google search) */
2922 		0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05,
2923 		0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09,
2924 	},
2925 	{	/* Known bad DELL UUID */
2926 		0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10,
2927 		0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20,
2928 	},
2929 	{	/* Uninitialized flash */
2930 		0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
2931 		0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
2932 	},
2933 	{	/* All zeros */
2934 		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
2935 		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
2936 	},
2937 };
2938 
2939 static int32_t
2940 uuid_to_hostid(const uint8_t *uuid)
2941 {
2942 	/*
2943 	 * Although the UUIDs are 128-bits, they may not distribute entropy
2944 	 * evenly.  We would like to use SHA or MD5, but those are located
2945 	 * in loadable modules and not available this early in boot.  As we
2946 	 * don't need the values to be cryptographically strong, we just
2947 	 * generate 32-bit vaue by xor'ing the various sequences together,
2948 	 * which ensures that the entire UUID contributes to the hostid.
2949 	 */
2950 	uint32_t	id = 0;
2951 
2952 	/* first check against the blacklist */
2953 	for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) {
2954 		if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) {
2955 			cmn_err(CE_CONT, "?Broken SMBIOS UUID. "
2956 			    "Contact BIOS manufacturer for repair.\n");
2957 			return ((int32_t)HW_INVALID_HOSTID);
2958 		}
2959 	}
2960 
2961 	for (int i = 0; i < 16; i++)
2962 		id ^= ((uuid[i]) << (8 * (i % sizeof (id))));
2963 
2964 	/* Make sure return value is positive */
2965 	return (id & 0x7fffffff);
2966 }
2967 
2968 static int32_t
2969 set_soft_hostid(void)
2970 {
2971 	struct _buf *file;
2972 	char tokbuf[MAXNAMELEN];
2973 	token_t token;
2974 	int done = 0;
2975 	u_longlong_t tmp;
2976 	int i;
2977 	int32_t hostid = (int32_t)HW_INVALID_HOSTID;
2978 	unsigned char *c;
2979 	hrtime_t tsc;
2980 	smbios_system_t smsys;
2981 
2982 	/*
2983 	 * If /etc/hostid file not found, we'd like to get a pseudo
2984 	 * random number to use at the hostid.  A nice way to do this
2985 	 * is to read the real time clock.  To remain xen-compatible,
2986 	 * we can't poke the real hardware, so we use tsc_read() to
2987 	 * read the real time clock.  However, there is an ominous
2988 	 * warning in tsc_read that says it can return zero, so we
2989 	 * deal with that possibility by falling back to using the
2990 	 * (hopefully random enough) value in tenmicrodata.
2991 	 */
2992 
2993 	if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) {
2994 		/*
2995 		 * hostid file not found - try to load sysinit module
2996 		 * and see if it has a nonzero hostid value...use that
2997 		 * instead of generating a new hostid here if so.
2998 		 */
2999 		if ((i = modload("misc", "sysinit")) != -1) {
3000 			if (strlen(hw_serial) > 0)
3001 				hostid = (int32_t)atoi(hw_serial);
3002 			(void) modunload(i);
3003 		}
3004 
3005 		/*
3006 		 * We try to use the SMBIOS UUID. But not if it is blacklisted
3007 		 * in /etc/system.
3008 		 */
3009 		if ((hostid == HW_INVALID_HOSTID) &&
3010 		    (smbios_broken_uuid == 0) &&
3011 		    (ksmbios != NULL) &&
3012 		    (smbios_info_system(ksmbios, &smsys) != SMB_ERR) &&
3013 		    (smsys.smbs_uuidlen >= 16)) {
3014 			hostid = uuid_to_hostid(smsys.smbs_uuid);
3015 		}
3016 
3017 		/*
3018 		 * Generate a "random" hostid using the clock.  These
3019 		 * hostids will change on each boot if the value is not
3020 		 * saved to a persistent /etc/hostid file.
3021 		 */
3022 		if (hostid == HW_INVALID_HOSTID) {
3023 			tsc = tsc_read();
3024 			if (tsc == 0)	/* tsc_read can return zero sometimes */
3025 				hostid = (int32_t)tenmicrodata & 0x0CFFFFF;
3026 			else
3027 				hostid = (int32_t)tsc & 0x0CFFFFF;
3028 		}
3029 	} else {
3030 		/* hostid file found */
3031 		while (!done) {
3032 			token = kobj_lex(file, tokbuf, sizeof (tokbuf));
3033 
3034 			switch (token) {
3035 			case POUND:
3036 				/*
3037 				 * skip comments
3038 				 */
3039 				kobj_find_eol(file);
3040 				break;
3041 			case STRING:
3042 				/*
3043 				 * un-rot47 - obviously this
3044 				 * nonsense is ascii-specific
3045 				 */
3046 				for (c = (unsigned char *)tokbuf;
3047 				    *c != '\0'; c++) {
3048 					*c += 47;
3049 					if (*c > '~')
3050 						*c -= 94;
3051 					else if (*c < '!')
3052 						*c += 94;
3053 				}
3054 				/*
3055 				 * now we should have a real number
3056 				 */
3057 
3058 				if (kobj_getvalue(tokbuf, &tmp) != 0)
3059 					kobj_file_err(CE_WARN, file,
3060 					    "Bad value %s for hostid",
3061 					    tokbuf);
3062 				else
3063 					hostid = (int32_t)tmp;
3064 
3065 				break;
3066 			case EOF:
3067 				done = 1;
3068 				/* FALLTHROUGH */
3069 			case NEWLINE:
3070 				kobj_newline(file);
3071 				break;
3072 			default:
3073 				break;
3074 
3075 			}
3076 		}
3077 		if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */
3078 			kobj_file_err(CE_WARN, file,
3079 			    "hostid missing or corrupt");
3080 
3081 		kobj_close_file(file);
3082 	}
3083 	/*
3084 	 * hostid is now the value read from /etc/hostid, or the
3085 	 * new hostid we generated in this routine or HW_INVALID_HOSTID if not
3086 	 * set.
3087 	 */
3088 	return (hostid);
3089 }
3090 
3091 static int
3092 atoi(char *p)
3093 {
3094 	int i = 0;
3095 
3096 	while (*p != '\0')
3097 		i = 10 * i + (*p++ - '0');
3098 
3099 	return (i);
3100 }
3101 
3102 #endif /* _SOFT_HOSTID */
3103 
3104 void
3105 get_system_configuration(void)
3106 {
3107 	char	prop[32];
3108 	u_longlong_t nodes_ll, cpus_pernode_ll, lvalue;
3109 
3110 	if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) ||
3111 	    BOP_GETPROP(bootops, "nodes", prop) < 0 ||
3112 	    kobj_getvalue(prop, &nodes_ll) == -1 ||
3113 	    nodes_ll > MAXNODES ||
3114 	    BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) ||
3115 	    BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 ||
3116 	    kobj_getvalue(prop, &cpus_pernode_ll) == -1) {
3117 		system_hardware.hd_nodes = 1;
3118 		system_hardware.hd_cpus_per_node = 0;
3119 	} else {
3120 		system_hardware.hd_nodes = (int)nodes_ll;
3121 		system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll;
3122 	}
3123 
3124 	if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) ||
3125 	    BOP_GETPROP(bootops, "kernelbase", prop) < 0 ||
3126 	    kobj_getvalue(prop, &lvalue) == -1)
3127 		eprom_kernelbase = 0;
3128 	else
3129 		eprom_kernelbase = (uintptr_t)lvalue;
3130 
3131 	if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) ||
3132 	    BOP_GETPROP(bootops, "segmapsize", prop) < 0 ||
3133 	    kobj_getvalue(prop, &lvalue) == -1)
3134 		segmapsize = SEGMAPDEFAULT;
3135 	else
3136 		segmapsize = (uintptr_t)lvalue;
3137 
3138 	if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) ||
3139 	    BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 ||
3140 	    kobj_getvalue(prop, &lvalue) == -1)
3141 		segmapfreelists = 0;	/* use segmap driver default */
3142 	else
3143 		segmapfreelists = (int)lvalue;
3144 
3145 	/* physmem used to be here, but moved much earlier to fakebop.c */
3146 }
3147 
3148 /*
3149  * Add to a memory list.
3150  * start = start of new memory segment
3151  * len = length of new memory segment in bytes
3152  * new = pointer to a new struct memlist
3153  * memlistp = memory list to which to add segment.
3154  */
3155 void
3156 memlist_add(
3157 	uint64_t start,
3158 	uint64_t len,
3159 	struct memlist *new,
3160 	struct memlist **memlistp)
3161 {
3162 	struct memlist *cur;
3163 	uint64_t end = start + len;
3164 
3165 	new->ml_address = start;
3166 	new->ml_size = len;
3167 
3168 	cur = *memlistp;
3169 
3170 	while (cur) {
3171 		if (cur->ml_address >= end) {
3172 			new->ml_next = cur;
3173 			*memlistp = new;
3174 			new->ml_prev = cur->ml_prev;
3175 			cur->ml_prev = new;
3176 			return;
3177 		}
3178 		ASSERT(cur->ml_address + cur->ml_size <= start);
3179 		if (cur->ml_next == NULL) {
3180 			cur->ml_next = new;
3181 			new->ml_prev = cur;
3182 			new->ml_next = NULL;
3183 			return;
3184 		}
3185 		memlistp = &cur->ml_next;
3186 		cur = cur->ml_next;
3187 	}
3188 }
3189 
3190 void
3191 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3192 {
3193 	size_t tsize = e_modtext - modtext;
3194 	size_t dsize = e_moddata - moddata;
3195 
3196 	*text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize,
3197 	    1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP);
3198 	*data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize,
3199 	    1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3200 }
3201 
3202 caddr_t
3203 kobj_text_alloc(vmem_t *arena, size_t size)
3204 {
3205 	return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT));
3206 }
3207 
3208 /*ARGSUSED*/
3209 caddr_t
3210 kobj_texthole_alloc(caddr_t addr, size_t size)
3211 {
3212 	panic("unexpected call to kobj_texthole_alloc()");
3213 	/*NOTREACHED*/
3214 	return (0);
3215 }
3216 
3217 /*ARGSUSED*/
3218 void
3219 kobj_texthole_free(caddr_t addr, size_t size)
3220 {
3221 	panic("unexpected call to kobj_texthole_free()");
3222 }
3223 
3224 /*
3225  * This is called just after configure() in startup().
3226  *
3227  * The ISALIST concept is a bit hopeless on Intel, because
3228  * there's no guarantee of an ever-more-capable processor
3229  * given that various parts of the instruction set may appear
3230  * and disappear between different implementations.
3231  *
3232  * While it would be possible to correct it and even enhance
3233  * it somewhat, the explicit hardware capability bitmask allows
3234  * more flexibility.
3235  *
3236  * So, we just leave this alone.
3237  */
3238 void
3239 setx86isalist(void)
3240 {
3241 	char *tp;
3242 	size_t len;
3243 	extern char *isa_list;
3244 
3245 #define	TBUFSIZE	1024
3246 
3247 	tp = kmem_alloc(TBUFSIZE, KM_SLEEP);
3248 	*tp = '\0';
3249 
3250 #if defined(__amd64)
3251 	(void) strcpy(tp, "amd64 ");
3252 #endif
3253 
3254 	switch (x86_vendor) {
3255 	case X86_VENDOR_Intel:
3256 	case X86_VENDOR_AMD:
3257 	case X86_VENDOR_TM:
3258 		if (is_x86_feature(x86_featureset, X86FSET_CMOV)) {
3259 			/*
3260 			 * Pentium Pro or later
3261 			 */
3262 			(void) strcat(tp, "pentium_pro");
3263 			(void) strcat(tp,
3264 			    is_x86_feature(x86_featureset, X86FSET_MMX) ?
3265 			    "+mmx pentium_pro " : " ");
3266 		}
3267 		/*FALLTHROUGH*/
3268 	case X86_VENDOR_Cyrix:
3269 		/*
3270 		 * The Cyrix 6x86 does not have any Pentium features
3271 		 * accessible while not at privilege level 0.
3272 		 */
3273 		if (is_x86_feature(x86_featureset, X86FSET_CPUID)) {
3274 			(void) strcat(tp, "pentium");
3275 			(void) strcat(tp,
3276 			    is_x86_feature(x86_featureset, X86FSET_MMX) ?
3277 			    "+mmx pentium " : " ");
3278 		}
3279 		break;
3280 	default:
3281 		break;
3282 	}
3283 	(void) strcat(tp, "i486 i386 i86");
3284 	len = strlen(tp) + 1;   /* account for NULL at end of string */
3285 	isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp);
3286 	kmem_free(tp, TBUFSIZE);
3287 
3288 #undef TBUFSIZE
3289 }
3290 
3291 
3292 #ifdef __amd64
3293 
3294 void *
3295 device_arena_alloc(size_t size, int vm_flag)
3296 {
3297 	return (vmem_alloc(device_arena, size, vm_flag));
3298 }
3299 
3300 void
3301 device_arena_free(void *vaddr, size_t size)
3302 {
3303 	vmem_free(device_arena, vaddr, size);
3304 }
3305 
3306 #else /* __i386 */
3307 
3308 void *
3309 device_arena_alloc(size_t size, int vm_flag)
3310 {
3311 	caddr_t	vaddr;
3312 	uintptr_t v;
3313 	size_t	start;
3314 	size_t	end;
3315 
3316 	vaddr = vmem_alloc(heap_arena, size, vm_flag);
3317 	if (vaddr == NULL)
3318 		return (NULL);
3319 
3320 	v = (uintptr_t)vaddr;
3321 	ASSERT(v >= kernelbase);
3322 	ASSERT(v + size <= valloc_base);
3323 
3324 	start = btop(v - kernelbase);
3325 	end = btop(v + size - 1 - kernelbase);
3326 	ASSERT(start < toxic_bit_map_len);
3327 	ASSERT(end < toxic_bit_map_len);
3328 
3329 	while (start <= end) {
3330 		BT_ATOMIC_SET(toxic_bit_map, start);
3331 		++start;
3332 	}
3333 	return (vaddr);
3334 }
3335 
3336 void
3337 device_arena_free(void *vaddr, size_t size)
3338 {
3339 	uintptr_t v = (uintptr_t)vaddr;
3340 	size_t	start;
3341 	size_t	end;
3342 
3343 	ASSERT(v >= kernelbase);
3344 	ASSERT(v + size <= valloc_base);
3345 
3346 	start = btop(v - kernelbase);
3347 	end = btop(v + size - 1 - kernelbase);
3348 	ASSERT(start < toxic_bit_map_len);
3349 	ASSERT(end < toxic_bit_map_len);
3350 
3351 	while (start <= end) {
3352 		ASSERT(BT_TEST(toxic_bit_map, start) != 0);
3353 		BT_ATOMIC_CLEAR(toxic_bit_map, start);
3354 		++start;
3355 	}
3356 	vmem_free(heap_arena, vaddr, size);
3357 }
3358 
3359 /*
3360  * returns 1st address in range that is in device arena, or NULL
3361  * if len is not NULL it returns the length of the toxic range
3362  */
3363 void *
3364 device_arena_contains(void *vaddr, size_t size, size_t *len)
3365 {
3366 	uintptr_t v = (uintptr_t)vaddr;
3367 	uintptr_t eaddr = v + size;
3368 	size_t start;
3369 	size_t end;
3370 
3371 	/*
3372 	 * if called very early by kmdb, just return NULL
3373 	 */
3374 	if (toxic_bit_map == NULL)
3375 		return (NULL);
3376 
3377 	/*
3378 	 * First check if we're completely outside the bitmap range.
3379 	 */
3380 	if (v >= valloc_base || eaddr < kernelbase)
3381 		return (NULL);
3382 
3383 	/*
3384 	 * Trim ends of search to look at only what the bitmap covers.
3385 	 */
3386 	if (v < kernelbase)
3387 		v = kernelbase;
3388 	start = btop(v - kernelbase);
3389 	end = btop(eaddr - kernelbase);
3390 	if (end >= toxic_bit_map_len)
3391 		end = toxic_bit_map_len;
3392 
3393 	if (bt_range(toxic_bit_map, &start, &end, end) == 0)
3394 		return (NULL);
3395 
3396 	v = kernelbase + ptob(start);
3397 	if (len != NULL)
3398 		*len = ptob(end - start);
3399 	return ((void *)v);
3400 }
3401 
3402 #endif	/* __i386 */
3403