xref: /illumos-gate/usr/src/uts/i86pc/os/startup.c (revision e214b19eaa16fec1fa60a97227778103f598336f)
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 struct seg kvseg_core;		/* Segment used for the core heap */
323 struct seg kpmseg;		/* Segment used for physical mapping */
324 struct seg *segkpm = &kpmseg;	/* 64bit kernel physical mapping segment */
325 
326 caddr_t segkp_base;		/* Base address of segkp */
327 caddr_t segzio_base;		/* Base address of segzio */
328 pgcnt_t segkpsize = btop(SEGKPDEFSIZE);	/* size of segkp segment in pages */
329 caddr_t segkvmm_base;
330 pgcnt_t segkvmmsize;
331 pgcnt_t segziosize;
332 
333 /*
334  * A static DR page_t VA map is reserved that can map the page structures
335  * for a domain's entire RA space. The pages that back this space are
336  * dynamically allocated and need not be physically contiguous.  The DR
337  * map size is derived from KPM size.
338  * This mechanism isn't used by x86 yet, so just stubs here.
339  */
340 int ppvm_enable = 0;		/* Static virtual map for page structs */
341 page_t *ppvm_base = NULL;	/* Base of page struct map */
342 pgcnt_t ppvm_size = 0;		/* Size of page struct map */
343 
344 /*
345  * VA range available to the debugger
346  */
347 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
348 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
349 
350 struct memseg *memseg_base;
351 struct vnode unused_pages_vp;
352 
353 #define	FOURGB	0x100000000LL
354 
355 struct memlist *memlist;
356 
357 caddr_t s_text;		/* start of kernel text segment */
358 caddr_t e_text;		/* end of kernel text segment */
359 caddr_t s_data;		/* start of kernel data segment */
360 caddr_t e_data;		/* end of kernel data segment */
361 caddr_t modtext;	/* start of loadable module text reserved */
362 caddr_t e_modtext;	/* end of loadable module text reserved */
363 caddr_t moddata;	/* start of loadable module data reserved */
364 caddr_t e_moddata;	/* end of loadable module data reserved */
365 
366 struct memlist *phys_install;	/* Total installed physical memory */
367 struct memlist *phys_avail;	/* Total available physical memory */
368 struct memlist *bios_rsvd;	/* Bios reserved memory */
369 
370 /*
371  * kphysm_init returns the number of pages that were processed
372  */
373 static pgcnt_t kphysm_init(page_t *, pgcnt_t);
374 
375 #define	IO_PROP_SIZE	64	/* device property size */
376 
377 /*
378  * a couple useful roundup macros
379  */
380 #define	ROUND_UP_PAGE(x)	\
381 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE))
382 #define	ROUND_UP_LPAGE(x)	\
383 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1]))
384 #define	ROUND_UP_4MEG(x)	\
385 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG))
386 #define	ROUND_UP_TOPLEVEL(x)	\
387 	((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level]))
388 
389 /*
390  *	32-bit Kernel's Virtual memory layout.
391  *		+-----------------------+
392  *		|			|
393  * 0xFFC00000  -|-----------------------|- ARGSBASE
394  *		|	debugger	|
395  * 0xFF800000  -|-----------------------|- SEGDEBUGBASE
396  *		|      Kernel Data	|
397  * 0xFEC00000  -|-----------------------|
398  *              |      Kernel Text	|
399  * 0xFE800000  -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen)
400  *		|---       GDT       ---|- GDT page (GDT_VA)
401  *		|---    debug info   ---|- debug info (DEBUG_INFO_VA)
402  *		|			|
403  *		|   page_t structures	|
404  *		|   memsegs, memlists,	|
405  *		|   page hash, etc.	|
406  * ---	       -|-----------------------|- ekernelheap, valloc_base (floating)
407  *		|			|  (segkp is just an arena in the heap)
408  *		|			|
409  *		|	kvseg		|
410  *		|			|
411  *		|			|
412  * ---         -|-----------------------|- kernelheap (floating)
413  *		|        Segkmap	|
414  * 0xC3002000  -|-----------------------|- segmap_start (floating)
415  *		|	Red Zone	|
416  * 0xC3000000  -|-----------------------|- kernelbase / userlimit (floating)
417  *		|			|			||
418  *		|     Shared objects	|			\/
419  *		|			|
420  *		:			:
421  *		|	user data	|
422  *		|-----------------------|
423  *		|	user text	|
424  * 0x08048000  -|-----------------------|
425  *		|	user stack	|
426  *		:			:
427  *		|	invalid		|
428  * 0x00000000	+-----------------------+
429  *
430  *
431  *		64-bit Kernel's Virtual memory layout. (assuming 64 bit app)
432  *			+-----------------------+
433  *			|			|
434  * 0xFFFFFFFF.FFC00000  |-----------------------|- ARGSBASE
435  *			|	debugger (?)	|
436  * 0xFFFFFFFF.FF800000  |-----------------------|- SEGDEBUGBASE
437  *			|      unused		|
438  *			+-----------------------+
439  *			|      Kernel Data	|
440  * 0xFFFFFFFF.FBC00000  |-----------------------|
441  *			|      Kernel Text	|
442  * 0xFFFFFFFF.FB800000  |-----------------------|- KERNEL_TEXT
443  *			|---    debug info   ---|- debug info (DEBUG_INFO_VA)
444  *			|---       GDT       ---|- GDT page (GDT_VA)
445  *			|---       IDT       ---|- IDT page (IDT_VA)
446  *			|---       LDT       ---|- LDT pages (LDT_VA)
447  *			|			|
448  *			|      Core heap	| (used for loadable modules)
449  * 0xFFFFFFFF.C0000000  |-----------------------|- core_base / ekernelheap
450  *			|	 Kernel		|
451  *			|	  heap		|
452  *			|			|
453  *			|			|
454  * 0xFFFFFXXX.XXX00000  |-----------------------|- kernelheap (floating)
455  *			|	 segmap		|
456  * 0xFFFFFXXX.XXX00000  |-----------------------|- segmap_start (floating)
457  *			|    device mappings	|
458  * 0xFFFFFXXX.XXX00000  |-----------------------|- toxic_addr (floating)
459  *			|	 segzio		|
460  * 0xFFFFFXXX.XXX00000  |-----------------------|- segzio_base (floating)
461  *			|        segkvmm	|
462  *			|			|
463  *			|			|
464  *			|			|
465  * 0xFFFFFXXX.XXX00000  |-----------------------|- segkvmm_base (floating)
466  *			|	 segkp		|
467  * 			|-----------------------|- segkp_base (floating)
468  *			|   page_t structures	|  valloc_base + valloc_sz
469  *			|   memsegs, memlists,	|
470  *			|   page hash, etc.	|
471  * 0xFFFFFE00.00000000  |-----------------------|- valloc_base (lower if >256GB)
472  *			|	 segkpm		|
473  *			|			|
474  * 0xFFFFFD00.00000000  |-----------------------|- SEGKPM_BASE (lower if >256GB)
475  *			|	Red Zone	|
476  * 0xFFFFFC80.00000000  |-----------------------|- KERNELBASE (lower if >256GB)
477  * 0xFFFFFC7F.FFE00000  |-----------------------|- USERLIMIT (lower if >256GB)
478  *			|     User stack	|- User space memory
479  *			|			|
480  *			| shared objects, etc	|	(grows downwards)
481  *			:			:
482  *			|			|
483  * 0xFFFF8000.00000000  |-----------------------|
484  *			|			|
485  *			| VA Hole / unused	|
486  *			|			|
487  * 0x00008000.00000000  |-----------------------|
488  *			|			|
489  *			|			|
490  *			:			:
491  *			|	user heap	|	(grows upwards)
492  *			|			|
493  *			|	user data	|
494  *			|-----------------------|
495  *			|	user text	|
496  * 0x00000000.04000000  |-----------------------|
497  *			|	invalid		|
498  * 0x00000000.00000000	+-----------------------+
499  *
500  * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit
501  * kernel, except that userlimit is raised to 0xfe000000
502  *
503  * Floating values:
504  *
505  * valloc_base: start of the kernel's memory management/tracking data
506  * structures.  This region contains page_t structures for
507  * physical memory, memsegs, memlists, and the page hash.
508  *
509  * core_base: start of the kernel's "core" heap area on 64-bit systems.
510  * This area is intended to be used for global data as well as for module
511  * text/data that does not fit into the nucleus pages.  The core heap is
512  * restricted to a 2GB range, allowing every address within it to be
513  * accessed using rip-relative addressing
514  *
515  * ekernelheap: end of kernelheap and start of segmap.
516  *
517  * kernelheap: start of kernel heap.  On 32-bit systems, this starts right
518  * above a red zone that separates the user's address space from the
519  * kernel's.  On 64-bit systems, it sits above segkp and segkpm.
520  *
521  * segmap_start: start of segmap. The length of segmap can be modified
522  * through eeprom. The default length is 16MB on 32-bit systems and 64MB
523  * on 64-bit systems.
524  *
525  * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be
526  * decreased by 2X the size required for page_t.  This allows the kernel
527  * heap to grow in size with physical memory.  With sizeof(page_t) == 80
528  * bytes, the following shows the values of kernelbase and kernel heap
529  * sizes for different memory configurations (assuming default segmap and
530  * segkp sizes).
531  *
532  *	mem	size for	kernelbase	kernel heap
533  *	size	page_t's			size
534  *	----	---------	----------	-----------
535  *	1gb	0x01400000	0xd1800000	684MB
536  *	2gb	0x02800000	0xcf000000	704MB
537  *	4gb	0x05000000	0xca000000	744MB
538  *	6gb	0x07800000	0xc5000000	784MB
539  *	8gb	0x0a000000	0xc0000000	824MB
540  *	16gb	0x14000000	0xac000000	984MB
541  *	32gb	0x28000000	0x84000000	1304MB
542  *	64gb	0x50000000	0x34000000	1944MB (*)
543  *
544  * kernelbase is less than the abi minimum of 0xc0000000 for memory
545  * configurations above 8gb.
546  *
547  * (*) support for memory configurations above 32gb will require manual tuning
548  * of kernelbase to balance out the need of user applications.
549  */
550 
551 /* real-time-clock initialization parameters */
552 extern time_t process_rtc_config_file(void);
553 
554 uintptr_t	kernelbase;
555 uintptr_t	postbootkernelbase;	/* not set till boot loader is gone */
556 uintptr_t	eprom_kernelbase;
557 size_t		segmapsize;
558 uintptr_t	segmap_start;
559 int		segmapfreelists;
560 pgcnt_t		npages;
561 pgcnt_t		orig_npages;
562 size_t		core_size;		/* size of "core" heap */
563 uintptr_t	core_base;		/* base address of "core" heap */
564 
565 /*
566  * List of bootstrap pages. We mark these as allocated in startup.
567  * release_bootstrap() will free them when we're completely done with
568  * the bootstrap.
569  */
570 static page_t *bootpages;
571 
572 /*
573  * boot time pages that have a vnode from the ramdisk will keep that forever.
574  */
575 static page_t *rd_pages;
576 
577 /*
578  * Lower 64K
579  */
580 static page_t *lower_pages = NULL;
581 static int lower_pages_count = 0;
582 
583 struct system_hardware system_hardware;
584 
585 /*
586  * Enable some debugging messages concerning memory usage...
587  */
588 static void
589 print_memlist(char *title, struct memlist *mp)
590 {
591 	prom_printf("MEMLIST: %s:\n", title);
592 	while (mp != NULL)  {
593 		prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n",
594 		    mp->ml_address, mp->ml_size);
595 		mp = mp->ml_next;
596 	}
597 }
598 
599 /*
600  * XX64 need a comment here.. are these just default values, surely
601  * we read the "cpuid" type information to figure this out.
602  */
603 int	l2cache_sz = 0x80000;
604 int	l2cache_linesz = 0x40;
605 int	l2cache_assoc = 1;
606 
607 static size_t	textrepl_min_gb = 10;
608 
609 /*
610  * on 64 bit we use a predifined VA range for mapping devices in the kernel
611  * on 32 bit the mappings are intermixed in the heap, so we use a bit map
612  */
613 #ifdef __amd64
614 
615 vmem_t		*device_arena;
616 uintptr_t	toxic_addr = (uintptr_t)NULL;
617 size_t		toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */
618 
619 #else	/* __i386 */
620 
621 ulong_t		*toxic_bit_map;	/* one bit for each 4k of VA in heap_arena */
622 size_t		toxic_bit_map_len = 0;	/* in bits */
623 
624 #endif	/* __i386 */
625 
626 /*
627  * Simple boot time debug facilities
628  */
629 static char *prm_dbg_str[] = {
630 	"%s:%d: '%s' is 0x%x\n",
631 	"%s:%d: '%s' is 0x%llx\n"
632 };
633 
634 int prom_debug;
635 
636 #define	PRM_DEBUG(q)	if (prom_debug)		\
637 	prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q);
638 #define	PRM_POINT(q)	if (prom_debug)		\
639 	prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q);
640 
641 /*
642  * This structure is used to keep track of the intial allocations
643  * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to
644  * be >= the number of ADD_TO_ALLOCATIONS() executed in the code.
645  */
646 #define	NUM_ALLOCATIONS 8
647 int num_allocations = 0;
648 struct {
649 	void **al_ptr;
650 	size_t al_size;
651 } allocations[NUM_ALLOCATIONS];
652 size_t valloc_sz = 0;
653 uintptr_t valloc_base;
654 
655 #define	ADD_TO_ALLOCATIONS(ptr, size) {					\
656 		size = ROUND_UP_PAGE(size);				\
657 		if (num_allocations == NUM_ALLOCATIONS)			\
658 			panic("too many ADD_TO_ALLOCATIONS()");		\
659 		allocations[num_allocations].al_ptr = (void**)&ptr;	\
660 		allocations[num_allocations].al_size = size;		\
661 		valloc_sz += size;					\
662 		++num_allocations;					\
663 	}
664 
665 /*
666  * Allocate all the initial memory needed by the page allocator.
667  */
668 static void
669 perform_allocations(void)
670 {
671 	caddr_t mem;
672 	int i;
673 	int valloc_align;
674 
675 	PRM_DEBUG(valloc_base);
676 	PRM_DEBUG(valloc_sz);
677 	valloc_align = mmu.level_size[mmu.max_page_level > 0];
678 	mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align);
679 	if (mem != (caddr_t)valloc_base)
680 		panic("BOP_ALLOC() failed");
681 	bzero(mem, valloc_sz);
682 	for (i = 0; i < num_allocations; ++i) {
683 		*allocations[i].al_ptr = (void *)mem;
684 		mem += allocations[i].al_size;
685 	}
686 }
687 
688 /*
689  * Set up and enable SMAP now before we start other CPUs, but after the kernel's
690  * VM has been set up so we can use hot_patch_kernel_text().
691  *
692  * We can only patch 1, 2, or 4 bytes, but not three bytes. So instead, we
693  * replace the four byte word at the patch point. See uts/intel/ia32/ml/copy.s
694  * for more information on what's going on here.
695  */
696 static void
697 startup_smap(void)
698 {
699 	int i;
700 	uint32_t inst;
701 	uint8_t *instp;
702 	char sym[128];
703 	struct modctl *modp;
704 
705 	extern int _smap_enable_patch_count;
706 	extern int _smap_disable_patch_count;
707 
708 	if (disable_smap != 0)
709 		remove_x86_feature(x86_featureset, X86FSET_SMAP);
710 
711 	if (is_x86_feature(x86_featureset, X86FSET_SMAP) == B_FALSE)
712 		return;
713 
714 	for (i = 0; i < _smap_enable_patch_count; i++) {
715 		int sizep;
716 
717 		VERIFY3U(i, <, _smap_enable_patch_count);
718 		VERIFY(snprintf(sym, sizeof (sym), "_smap_enable_patch_%d", i) <
719 		    sizeof (sym));
720 		instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep);
721 		VERIFY(instp != 0);
722 		inst = (instp[3] << 24) | (SMAP_CLAC_INSTR & 0x00ffffff);
723 		hot_patch_kernel_text((caddr_t)instp, inst, 4);
724 	}
725 
726 	for (i = 0; i < _smap_disable_patch_count; i++) {
727 		int sizep;
728 
729 		VERIFY(snprintf(sym, sizeof (sym), "_smap_disable_patch_%d",
730 		    i) < sizeof (sym));
731 		instp = (uint8_t *)(void *)kobj_getelfsym(sym, NULL, &sizep);
732 		VERIFY(instp != 0);
733 		inst = (instp[3] << 24) | (SMAP_STAC_INSTR & 0x00ffffff);
734 		hot_patch_kernel_text((caddr_t)instp, inst, 4);
735 	}
736 
737 	/*
738 	 * Hotinline calls to smap_enable and smap_disable within
739 	 * unix module. Hotinlines in other modules are done on
740 	 * mod_load().
741 	 */
742 	modp = mod_hold_by_name("unix");
743 	do_hotinlines(modp->mod_mp);
744 	mod_release_mod(modp);
745 
746 	setcr4(getcr4() | CR4_SMAP);
747 	smap_enable();
748 }
749 
750 /*
751  * Our world looks like this at startup time.
752  *
753  * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data
754  * at 0xfec00000.  On a 64-bit OS, kernel text and data are loaded at
755  * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively.  Those
756  * addresses are fixed in the binary at link time.
757  *
758  * On the text page:
759  * unix/genunix/krtld/module text loads.
760  *
761  * On the data page:
762  * unix/genunix/krtld/module data loads.
763  *
764  * Machine-dependent startup code
765  */
766 void
767 startup(void)
768 {
769 #if !defined(__xpv)
770 	extern void startup_pci_bios(void);
771 #endif
772 	extern cpuset_t cpu_ready_set;
773 
774 	/*
775 	 * Make sure that nobody tries to use sekpm until we have
776 	 * initialized it properly.
777 	 */
778 #if defined(__amd64)
779 	kpm_desired = 1;
780 #endif
781 	kpm_enable = 0;
782 	CPUSET_ONLY(cpu_ready_set, 0);	/* cpu 0 is boot cpu */
783 
784 #if defined(__xpv)	/* XXPV fix me! */
785 	{
786 		extern int segvn_use_regions;
787 		segvn_use_regions = 0;
788 	}
789 #endif
790 	ssp_init();
791 	progressbar_init();
792 	startup_init();
793 #if defined(__xpv)
794 	startup_xen_version();
795 #endif
796 	startup_memlist();
797 	startup_kmem();
798 	startup_vm();
799 #if !defined(__xpv)
800 	/*
801 	 * Note we need to do this even on fast reboot in order to access
802 	 * the irq routing table (used for pci labels).
803 	 */
804 	startup_pci_bios();
805 	startup_smap();
806 #endif
807 #if defined(__xpv)
808 	startup_xen_mca();
809 #endif
810 	startup_modules();
811 
812 	startup_end();
813 }
814 
815 static void
816 startup_init()
817 {
818 	PRM_POINT("startup_init() starting...");
819 
820 	/*
821 	 * Complete the extraction of cpuid data
822 	 */
823 	cpuid_pass2(CPU);
824 
825 	(void) check_boot_version(BOP_GETVERSION(bootops));
826 
827 	/*
828 	 * Check for prom_debug in boot environment
829 	 */
830 	if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) {
831 		++prom_debug;
832 		PRM_POINT("prom_debug found in boot enviroment");
833 	}
834 
835 	/*
836 	 * Collect node, cpu and memory configuration information.
837 	 */
838 	get_system_configuration();
839 
840 	/*
841 	 * Halt if this is an unsupported processor.
842 	 */
843 	if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) {
844 		printf("\n486 processor (\"%s\") detected.\n",
845 		    CPU->cpu_brandstr);
846 		halt("This processor is not supported by this release "
847 		    "of Solaris.");
848 	}
849 
850 	PRM_POINT("startup_init() done");
851 }
852 
853 /*
854  * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie.
855  * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it
856  * also filters out physical page zero.  There is some reliance on the
857  * boot loader allocating only a few contiguous physical memory chunks.
858  */
859 static void
860 avail_filter(uint64_t *addr, uint64_t *size)
861 {
862 	uintptr_t va;
863 	uintptr_t next_va;
864 	pfn_t pfn;
865 	uint64_t pfn_addr;
866 	uint64_t pfn_eaddr;
867 	uint_t prot;
868 	size_t len;
869 	uint_t change;
870 
871 	if (prom_debug)
872 		prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n",
873 		    *addr, *size);
874 
875 	/*
876 	 * page zero is required for BIOS.. never make it available
877 	 */
878 	if (*addr == 0) {
879 		*addr += MMU_PAGESIZE;
880 		*size -= MMU_PAGESIZE;
881 	}
882 
883 	/*
884 	 * First we trim from the front of the range. Since kbm_probe()
885 	 * walks ranges in virtual order, but addr/size are physical, we need
886 	 * to the list until no changes are seen.  This deals with the case
887 	 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w
888 	 * but w < v.
889 	 */
890 	do {
891 		change = 0;
892 		for (va = KERNEL_TEXT;
893 		    *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
894 		    va = next_va) {
895 
896 			next_va = va + len;
897 			pfn_addr = pfn_to_pa(pfn);
898 			pfn_eaddr = pfn_addr + len;
899 
900 			if (pfn_addr <= *addr && pfn_eaddr > *addr) {
901 				change = 1;
902 				while (*size > 0 && len > 0) {
903 					*addr += MMU_PAGESIZE;
904 					*size -= MMU_PAGESIZE;
905 					len -= MMU_PAGESIZE;
906 				}
907 			}
908 		}
909 		if (change && prom_debug)
910 			prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n",
911 			    *addr, *size);
912 	} while (change);
913 
914 	/*
915 	 * Trim pages from the end of the range.
916 	 */
917 	for (va = KERNEL_TEXT;
918 	    *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
919 	    va = next_va) {
920 
921 		next_va = va + len;
922 		pfn_addr = pfn_to_pa(pfn);
923 
924 		if (pfn_addr >= *addr && pfn_addr < *addr + *size)
925 			*size = pfn_addr - *addr;
926 	}
927 
928 	if (prom_debug)
929 		prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n",
930 		    *addr, *size);
931 }
932 
933 static void
934 kpm_init()
935 {
936 	struct segkpm_crargs b;
937 
938 	/*
939 	 * These variables were all designed for sfmmu in which segkpm is
940 	 * mapped using a single pagesize - either 8KB or 4MB.  On x86, we
941 	 * might use 2+ page sizes on a single machine, so none of these
942 	 * variables have a single correct value.  They are set up as if we
943 	 * always use a 4KB pagesize, which should do no harm.  In the long
944 	 * run, we should get rid of KPM's assumption that only a single
945 	 * pagesize is used.
946 	 */
947 	kpm_pgshft = MMU_PAGESHIFT;
948 	kpm_pgsz =  MMU_PAGESIZE;
949 	kpm_pgoff = MMU_PAGEOFFSET;
950 	kpmp2pshft = 0;
951 	kpmpnpgs = 1;
952 	ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0);
953 
954 	PRM_POINT("about to create segkpm");
955 	rw_enter(&kas.a_lock, RW_WRITER);
956 
957 	if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0)
958 		panic("cannot attach segkpm");
959 
960 	b.prot = PROT_READ | PROT_WRITE;
961 	b.nvcolors = 1;
962 
963 	if (segkpm_create(segkpm, (caddr_t)&b) != 0)
964 		panic("segkpm_create segkpm");
965 
966 	rw_exit(&kas.a_lock);
967 
968 	kpm_enable = 1;
969 
970 	/*
971 	 * As the KPM was disabled while setting up the system, go back and fix
972 	 * CPU zero's access to its user page table. This is a bit gross, but
973 	 * we have a chicken and egg problem otherwise.
974 	 */
975 	ASSERT(CPU->cpu_hat_info->hci_user_l3ptes == NULL);
976 	CPU->cpu_hat_info->hci_user_l3ptes =
977 	    (x86pte_t *)hat_kpm_mapin_pfn(CPU->cpu_hat_info->hci_user_l3pfn);
978 }
979 
980 /*
981  * The debug info page provides enough information to allow external
982  * inspectors (e.g. when running under a hypervisor) to bootstrap
983  * themselves into allowing full-blown kernel debugging.
984  */
985 static void
986 init_debug_info(void)
987 {
988 	caddr_t mem;
989 	debug_info_t *di;
990 
991 #ifndef __lint
992 	ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE);
993 #endif
994 
995 	mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE,
996 	    MMU_PAGESIZE);
997 
998 	if (mem != (caddr_t)DEBUG_INFO_VA)
999 		panic("BOP_ALLOC() failed");
1000 	bzero(mem, MMU_PAGESIZE);
1001 
1002 	di = (debug_info_t *)mem;
1003 
1004 	di->di_magic = DEBUG_INFO_MAGIC;
1005 	di->di_version = DEBUG_INFO_VERSION;
1006 	di->di_modules = (uintptr_t)&modules;
1007 	di->di_s_text = (uintptr_t)s_text;
1008 	di->di_e_text = (uintptr_t)e_text;
1009 	di->di_s_data = (uintptr_t)s_data;
1010 	di->di_e_data = (uintptr_t)e_data;
1011 	di->di_hat_htable_off = offsetof(hat_t, hat_htable);
1012 	di->di_ht_pfn_off = offsetof(htable_t, ht_pfn);
1013 }
1014 
1015 /*
1016  * Build the memlists and other kernel essential memory system data structures.
1017  * This is everything at valloc_base.
1018  */
1019 static void
1020 startup_memlist(void)
1021 {
1022 	size_t memlist_sz;
1023 	size_t memseg_sz;
1024 	size_t pagehash_sz;
1025 	size_t pp_sz;
1026 	uintptr_t va;
1027 	size_t len;
1028 	uint_t prot;
1029 	pfn_t pfn;
1030 	int memblocks;
1031 	pfn_t rsvd_high_pfn;
1032 	pgcnt_t rsvd_pgcnt;
1033 	size_t rsvdmemlist_sz;
1034 	int rsvdmemblocks;
1035 	caddr_t pagecolor_mem;
1036 	size_t pagecolor_memsz;
1037 	caddr_t page_ctrs_mem;
1038 	size_t page_ctrs_size;
1039 	size_t pse_table_alloc_size;
1040 	struct memlist *current;
1041 	extern void startup_build_mem_nodes(struct memlist *);
1042 
1043 	/* XX64 fix these - they should be in include files */
1044 	extern size_t page_coloring_init(uint_t, int, int);
1045 	extern void page_coloring_setup(caddr_t);
1046 
1047 	PRM_POINT("startup_memlist() starting...");
1048 
1049 	/*
1050 	 * Use leftover large page nucleus text/data space for loadable modules.
1051 	 * Use at most MODTEXT/MODDATA.
1052 	 */
1053 	len = kbm_nucleus_size;
1054 	ASSERT(len > MMU_PAGESIZE);
1055 
1056 	moddata = (caddr_t)ROUND_UP_PAGE(e_data);
1057 	e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len);
1058 	if (e_moddata - moddata > MODDATA)
1059 		e_moddata = moddata + MODDATA;
1060 
1061 	modtext = (caddr_t)ROUND_UP_PAGE(e_text);
1062 	e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len);
1063 	if (e_modtext - modtext > MODTEXT)
1064 		e_modtext = modtext + MODTEXT;
1065 
1066 	econtig = e_moddata;
1067 
1068 	PRM_DEBUG(modtext);
1069 	PRM_DEBUG(e_modtext);
1070 	PRM_DEBUG(moddata);
1071 	PRM_DEBUG(e_moddata);
1072 	PRM_DEBUG(econtig);
1073 
1074 	/*
1075 	 * Examine the boot loader physical memory map to find out:
1076 	 * - total memory in system - physinstalled
1077 	 * - the max physical address - physmax
1078 	 * - the number of discontiguous segments of memory.
1079 	 */
1080 	if (prom_debug)
1081 		print_memlist("boot physinstalled",
1082 		    bootops->boot_mem->physinstalled);
1083 	installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax,
1084 	    &physinstalled, &memblocks);
1085 	PRM_DEBUG(physmax);
1086 	PRM_DEBUG(physinstalled);
1087 	PRM_DEBUG(memblocks);
1088 
1089 	/*
1090 	 * We no longer support any form of memory DR.
1091 	 */
1092 	plat_dr_physmax = 0;
1093 
1094 	/*
1095 	 * Examine the bios reserved memory to find out:
1096 	 * - the number of discontiguous segments of memory.
1097 	 */
1098 	if (prom_debug)
1099 		print_memlist("boot reserved mem",
1100 		    bootops->boot_mem->rsvdmem);
1101 	installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn,
1102 	    &rsvd_pgcnt, &rsvdmemblocks);
1103 	PRM_DEBUG(rsvd_high_pfn);
1104 	PRM_DEBUG(rsvd_pgcnt);
1105 	PRM_DEBUG(rsvdmemblocks);
1106 
1107 	/*
1108 	 * Initialize hat's mmu parameters.
1109 	 * Check for enforce-prot-exec in boot environment. It's used to
1110 	 * enable/disable support for the page table entry NX bit.
1111 	 * The default is to enforce PROT_EXEC on processors that support NX.
1112 	 * Boot seems to round up the "len", but 8 seems to be big enough.
1113 	 */
1114 	mmu_init();
1115 
1116 #ifdef	__i386
1117 	/*
1118 	 * physmax is lowered if there is more memory than can be
1119 	 * physically addressed in 32 bit (PAE/non-PAE) modes.
1120 	 */
1121 	if (mmu.pae_hat) {
1122 		if (PFN_ABOVE64G(physmax)) {
1123 			physinstalled -= (physmax - (PFN_64G - 1));
1124 			physmax = PFN_64G - 1;
1125 		}
1126 	} else {
1127 		if (PFN_ABOVE4G(physmax)) {
1128 			physinstalled -= (physmax - (PFN_4G - 1));
1129 			physmax = PFN_4G - 1;
1130 		}
1131 	}
1132 #endif
1133 
1134 	startup_build_mem_nodes(bootops->boot_mem->physinstalled);
1135 
1136 	if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) {
1137 		int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec");
1138 		char value[8];
1139 
1140 		if (len < 8)
1141 			(void) BOP_GETPROP(bootops, "enforce-prot-exec", value);
1142 		else
1143 			(void) strcpy(value, "");
1144 		if (strcmp(value, "off") == 0)
1145 			mmu.pt_nx = 0;
1146 	}
1147 	PRM_DEBUG(mmu.pt_nx);
1148 
1149 	/*
1150 	 * We will need page_t's for every page in the system, except for
1151 	 * memory mapped at or above above the start of the kernel text segment.
1152 	 *
1153 	 * pages above e_modtext are attributed to kernel debugger (obp_pages)
1154 	 */
1155 	npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */
1156 	obp_pages = 0;
1157 	va = KERNEL_TEXT;
1158 	while (kbm_probe(&va, &len, &pfn, &prot) != 0) {
1159 		npages -= len >> MMU_PAGESHIFT;
1160 		if (va >= (uintptr_t)e_moddata)
1161 			obp_pages += len >> MMU_PAGESHIFT;
1162 		va += len;
1163 	}
1164 	PRM_DEBUG(npages);
1165 	PRM_DEBUG(obp_pages);
1166 
1167 	/*
1168 	 * If physmem is patched to be non-zero, use it instead of the computed
1169 	 * value unless it is larger than the actual amount of memory on hand.
1170 	 */
1171 	if (physmem == 0 || physmem > npages) {
1172 		physmem = npages;
1173 	} else if (physmem < npages) {
1174 		orig_npages = npages;
1175 		npages = physmem;
1176 	}
1177 	PRM_DEBUG(physmem);
1178 
1179 	/*
1180 	 * We now compute the sizes of all the  initial allocations for
1181 	 * structures the kernel needs in order do kmem_alloc(). These
1182 	 * include:
1183 	 *	memsegs
1184 	 *	memlists
1185 	 *	page hash table
1186 	 *	page_t's
1187 	 *	page coloring data structs
1188 	 */
1189 	memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS);
1190 	ADD_TO_ALLOCATIONS(memseg_base, memseg_sz);
1191 	PRM_DEBUG(memseg_sz);
1192 
1193 	/*
1194 	 * Reserve space for memlists. There's no real good way to know exactly
1195 	 * how much room we'll need, but this should be a good upper bound.
1196 	 */
1197 	memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1198 	    (memblocks + POSS_NEW_FRAGMENTS));
1199 	ADD_TO_ALLOCATIONS(memlist, memlist_sz);
1200 	PRM_DEBUG(memlist_sz);
1201 
1202 	/*
1203 	 * Reserve space for bios reserved memlists.
1204 	 */
1205 	rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1206 	    (rsvdmemblocks + POSS_NEW_FRAGMENTS));
1207 	ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz);
1208 	PRM_DEBUG(rsvdmemlist_sz);
1209 
1210 	/* LINTED */
1211 	ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page)));
1212 	/*
1213 	 * The page structure hash table size is a power of 2
1214 	 * such that the average hash chain length is PAGE_HASHAVELEN.
1215 	 */
1216 	page_hashsz = npages / PAGE_HASHAVELEN;
1217 	page_hashsz_shift = highbit(page_hashsz);
1218 	page_hashsz = 1 << page_hashsz_shift;
1219 	pagehash_sz = sizeof (struct page *) * page_hashsz;
1220 	ADD_TO_ALLOCATIONS(page_hash, pagehash_sz);
1221 	PRM_DEBUG(pagehash_sz);
1222 
1223 	/*
1224 	 * Set aside room for the page structures themselves.
1225 	 */
1226 	PRM_DEBUG(npages);
1227 	pp_sz = sizeof (struct page) * npages;
1228 	ADD_TO_ALLOCATIONS(pp_base, pp_sz);
1229 	PRM_DEBUG(pp_sz);
1230 
1231 	/*
1232 	 * determine l2 cache info and memory size for page coloring
1233 	 */
1234 	(void) getl2cacheinfo(CPU,
1235 	    &l2cache_sz, &l2cache_linesz, &l2cache_assoc);
1236 	pagecolor_memsz =
1237 	    page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc);
1238 	ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz);
1239 	PRM_DEBUG(pagecolor_memsz);
1240 
1241 	page_ctrs_size = page_ctrs_sz();
1242 	ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size);
1243 	PRM_DEBUG(page_ctrs_size);
1244 
1245 	/*
1246 	 * Allocate the array that protects pp->p_selock.
1247 	 */
1248 	pse_shift = size_pse_array(physmem, max_ncpus);
1249 	pse_table_size = 1 << pse_shift;
1250 	pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t);
1251 	ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size);
1252 
1253 	valloc_sz = ROUND_UP_LPAGE(valloc_sz);
1254 	valloc_base = VALLOC_BASE;
1255 
1256 	/*
1257 	 * The signicant memory-sized regions are roughly sized as follows in
1258 	 * the default layout with max physmem:
1259 	 *  segkpm: 1x physmem allocated (but 1Tb room, below VALLOC_BASE)
1260 	 *  segzio: 1.5x physmem
1261 	 *  segkvmm: 4x physmem
1262 	 *  heap: whatever's left up to COREHEAP_BASE, at least 1.5x physmem
1263 	 *
1264 	 * The idea is that we leave enough room to avoid fragmentation issues,
1265 	 * so we would like the VA arenas to have some extra.
1266 	 *
1267 	 * Ignoring the loose change of segkp, valloc, and such, this means that
1268 	 * as COREHEAP_BASE-VALLOC_BASE=2Tb, we can accommodate a physmem up to
1269 	 * about (2Tb / 7.0), rounded down to 256Gb in the check below.
1270 	 *
1271 	 * Note that KPM lives below VALLOC_BASE, but we want to include it in
1272 	 * adjustments, hence the 8 below.
1273 	 *
1274 	 * Beyond 256Gb, we push segkpm_base (and hence kernelbase and
1275 	 * _userlimit) down to accommodate the VA requirements above.
1276 	 */
1277 	if (physmax + 1 > mmu_btop(TERABYTE / 4)) {
1278 		uint64_t physmem_bytes = mmu_ptob(physmax + 1);
1279 		uint64_t adjustment = 8 * (physmem_bytes - (TERABYTE / 4));
1280 
1281 		PRM_DEBUG(adjustment);
1282 
1283 		/*
1284 		 * segkpm_base is always aligned on a L3 PTE boundary.
1285 		 */
1286 		segkpm_base -= P2ROUNDUP(adjustment, KERNEL_REDZONE_SIZE);
1287 
1288 		/*
1289 		 * But make sure we leave some space for user apps above hole.
1290 		 */
1291 		segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE);
1292 
1293 		ASSERT(segkpm_base <= SEGKPM_BASE);
1294 
1295 		valloc_base = segkpm_base + P2ROUNDUP(physmem_bytes, ONE_GIG);
1296 		if (valloc_base < segkpm_base)
1297 			panic("not enough kernel VA to support memory size");
1298 	}
1299 
1300 	PRM_DEBUG(segkpm_base);
1301 	PRM_DEBUG(valloc_base);
1302 
1303 	/*
1304 	 * do all the initial allocations
1305 	 */
1306 	perform_allocations();
1307 
1308 	/*
1309 	 * Build phys_install and phys_avail in kernel memspace.
1310 	 * - phys_install should be all memory in the system.
1311 	 * - phys_avail is phys_install minus any memory mapped before this
1312 	 *    point above KERNEL_TEXT.
1313 	 */
1314 	current = phys_install = memlist;
1315 	copy_memlist_filter(bootops->boot_mem->physinstalled, &current, NULL);
1316 	if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1317 		panic("physinstalled was too big!");
1318 	if (prom_debug)
1319 		print_memlist("phys_install", phys_install);
1320 
1321 	phys_avail = current;
1322 	PRM_POINT("Building phys_avail:\n");
1323 	copy_memlist_filter(bootops->boot_mem->physinstalled, &current,
1324 	    avail_filter);
1325 	if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1326 		panic("physavail was too big!");
1327 	if (prom_debug)
1328 		print_memlist("phys_avail", phys_avail);
1329 #ifndef	__xpv
1330 	/*
1331 	 * Free unused memlist items, which may be used by memory DR driver
1332 	 * at runtime.
1333 	 */
1334 	if ((caddr_t)current < (caddr_t)memlist + memlist_sz) {
1335 		memlist_free_block((caddr_t)current,
1336 		    (caddr_t)memlist + memlist_sz - (caddr_t)current);
1337 	}
1338 #endif
1339 
1340 	/*
1341 	 * Build bios reserved memspace
1342 	 */
1343 	current = bios_rsvd;
1344 	copy_memlist_filter(bootops->boot_mem->rsvdmem, &current, NULL);
1345 	if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz)
1346 		panic("bios_rsvd was too big!");
1347 	if (prom_debug)
1348 		print_memlist("bios_rsvd", bios_rsvd);
1349 #ifndef	__xpv
1350 	/*
1351 	 * Free unused memlist items, which may be used by memory DR driver
1352 	 * at runtime.
1353 	 */
1354 	if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) {
1355 		memlist_free_block((caddr_t)current,
1356 		    (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current);
1357 	}
1358 #endif
1359 
1360 	/*
1361 	 * setup page coloring
1362 	 */
1363 	page_coloring_setup(pagecolor_mem);
1364 	page_lock_init();	/* currently a no-op */
1365 
1366 	/*
1367 	 * free page list counters
1368 	 */
1369 	(void) page_ctrs_alloc(page_ctrs_mem);
1370 
1371 	/*
1372 	 * Size the pcf array based on the number of cpus in the box at
1373 	 * boot time.
1374 	 */
1375 
1376 	pcf_init();
1377 
1378 	/*
1379 	 * Initialize the page structures from the memory lists.
1380 	 */
1381 	availrmem_initial = availrmem = freemem = 0;
1382 	PRM_POINT("Calling kphysm_init()...");
1383 	npages = kphysm_init(pp_base, npages);
1384 	PRM_POINT("kphysm_init() done");
1385 	PRM_DEBUG(npages);
1386 
1387 	init_debug_info();
1388 
1389 	/*
1390 	 * Now that page_t's have been initialized, remove all the
1391 	 * initial allocation pages from the kernel free page lists.
1392 	 */
1393 	boot_mapin((caddr_t)valloc_base, valloc_sz);
1394 	boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE);
1395 	PRM_POINT("startup_memlist() done");
1396 
1397 	PRM_DEBUG(valloc_sz);
1398 
1399 #if defined(__amd64)
1400 	if ((availrmem >> (30 - MMU_PAGESHIFT)) >=
1401 	    textrepl_min_gb && l2cache_sz <= 2 << 20) {
1402 		extern size_t textrepl_size_thresh;
1403 		textrepl_size_thresh = (16 << 20) - 1;
1404 	}
1405 #endif
1406 }
1407 
1408 /*
1409  * Layout the kernel's part of address space and initialize kmem allocator.
1410  */
1411 static void
1412 startup_kmem(void)
1413 {
1414 	extern void page_set_colorequiv_arr(void);
1415 #if !defined(__xpv)
1416 	extern uint64_t kpti_kbase;
1417 #endif
1418 
1419 	PRM_POINT("startup_kmem() starting...");
1420 
1421 #if defined(__amd64)
1422 	if (eprom_kernelbase && eprom_kernelbase != KERNELBASE)
1423 		cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit "
1424 		    "systems.");
1425 	kernelbase = segkpm_base - KERNEL_REDZONE_SIZE;
1426 	core_base = (uintptr_t)COREHEAP_BASE;
1427 	core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE;
1428 #else	/* __i386 */
1429 	/*
1430 	 * We configure kernelbase based on:
1431 	 *
1432 	 * 1. user specified kernelbase via eeprom command. Value cannot exceed
1433 	 *    KERNELBASE_MAX. we large page align eprom_kernelbase
1434 	 *
1435 	 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t.
1436 	 *    On large memory systems we must lower kernelbase to allow
1437 	 *    enough room for page_t's for all of memory.
1438 	 *
1439 	 * The value set here, might be changed a little later.
1440 	 */
1441 	if (eprom_kernelbase) {
1442 		kernelbase = eprom_kernelbase & mmu.level_mask[1];
1443 		if (kernelbase > KERNELBASE_MAX)
1444 			kernelbase = KERNELBASE_MAX;
1445 	} else {
1446 		kernelbase = (uintptr_t)KERNELBASE;
1447 		kernelbase -= ROUND_UP_4MEG(2 * valloc_sz);
1448 	}
1449 	ASSERT((kernelbase & mmu.level_offset[1]) == 0);
1450 	core_base = valloc_base;
1451 	core_size = 0;
1452 #endif	/* __i386 */
1453 
1454 	PRM_DEBUG(core_base);
1455 	PRM_DEBUG(core_size);
1456 	PRM_DEBUG(kernelbase);
1457 
1458 #if defined(__i386)
1459 	segkp_fromheap = 1;
1460 #endif	/* __i386 */
1461 
1462 	ekernelheap = (char *)core_base;
1463 	PRM_DEBUG(ekernelheap);
1464 
1465 	/*
1466 	 * Now that we know the real value of kernelbase,
1467 	 * update variables that were initialized with a value of
1468 	 * KERNELBASE (in common/conf/param.c).
1469 	 *
1470 	 * XXX	The problem with this sort of hackery is that the
1471 	 *	compiler just may feel like putting the const declarations
1472 	 *	(in param.c) into the .text section.  Perhaps they should
1473 	 *	just be declared as variables there?
1474 	 */
1475 
1476 	*(uintptr_t *)&_kernelbase = kernelbase;
1477 	*(uintptr_t *)&_userlimit = kernelbase;
1478 #if defined(__amd64)
1479 	*(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT;
1480 #if !defined(__xpv)
1481 	kpti_kbase = kernelbase;
1482 #endif
1483 #else
1484 	*(uintptr_t *)&_userlimit32 = _userlimit;
1485 #endif
1486 	PRM_DEBUG(_kernelbase);
1487 	PRM_DEBUG(_userlimit);
1488 	PRM_DEBUG(_userlimit32);
1489 
1490 	/* We have to re-do this now that we've modified _userlimit. */
1491 	mmu_calc_user_slots();
1492 
1493 	layout_kernel_va();
1494 
1495 #if defined(__i386)
1496 	/*
1497 	 * If segmap is too large we can push the bottom of the kernel heap
1498 	 * higher than the base.  Or worse, it could exceed the top of the
1499 	 * VA space entirely, causing it to wrap around.
1500 	 */
1501 	if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase)
1502 		panic("too little address space available for kernelheap,"
1503 		    " use eeprom for lower kernelbase or smaller segmapsize");
1504 #endif	/* __i386 */
1505 
1506 	/*
1507 	 * Initialize the kernel heap. Note 3rd argument must be > 1st.
1508 	 */
1509 	kernelheap_init(kernelheap, ekernelheap,
1510 	    kernelheap + MMU_PAGESIZE,
1511 	    (void *)core_base, (void *)(core_base + core_size));
1512 
1513 #if defined(__xpv)
1514 	/*
1515 	 * Link pending events struct into cpu struct
1516 	 */
1517 	CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data;
1518 #endif
1519 	/*
1520 	 * Initialize kernel memory allocator.
1521 	 */
1522 	kmem_init();
1523 
1524 	/*
1525 	 * Factor in colorequiv to check additional 'equivalent' bins
1526 	 */
1527 	page_set_colorequiv_arr();
1528 
1529 	/*
1530 	 * print this out early so that we know what's going on
1531 	 */
1532 	print_x86_featureset(x86_featureset);
1533 
1534 	/*
1535 	 * Initialize bp_mapin().
1536 	 */
1537 	bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK);
1538 
1539 	/*
1540 	 * orig_npages is non-zero if physmem has been configured for less
1541 	 * than the available memory.
1542 	 */
1543 	if (orig_npages) {
1544 		cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages",
1545 		    (npages == PHYSMEM ? "Due to virtual address space " : ""),
1546 		    npages, orig_npages);
1547 	}
1548 #if defined(__i386)
1549 	if (eprom_kernelbase && (eprom_kernelbase != kernelbase))
1550 		cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, "
1551 		    "System using 0x%lx",
1552 		    (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase);
1553 #endif
1554 
1555 #ifdef	KERNELBASE_ABI_MIN
1556 	if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) {
1557 		cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not "
1558 		    "i386 ABI compliant.", (uintptr_t)kernelbase);
1559 	}
1560 #endif
1561 
1562 #ifndef __xpv
1563 	if (plat_dr_support_memory()) {
1564 		mem_config_init();
1565 	}
1566 #else	/* __xpv */
1567 	/*
1568 	 * Some of the xen start information has to be relocated up
1569 	 * into the kernel's permanent address space.
1570 	 */
1571 	PRM_POINT("calling xen_relocate_start_info()");
1572 	xen_relocate_start_info();
1573 	PRM_POINT("xen_relocate_start_info() done");
1574 
1575 	/*
1576 	 * (Update the vcpu pointer in our cpu structure to point into
1577 	 * the relocated shared info.)
1578 	 */
1579 	CPU->cpu_m.mcpu_vcpu_info =
1580 	    &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id];
1581 #endif	/* __xpv */
1582 
1583 	PRM_POINT("startup_kmem() done");
1584 }
1585 
1586 #ifndef __xpv
1587 /*
1588  * If we have detected that we are running in an HVM environment, we need
1589  * to prepend the PV driver directory to the module search path.
1590  */
1591 #define	HVM_MOD_DIR "/platform/i86hvm/kernel"
1592 static void
1593 update_default_path()
1594 {
1595 	char *current, *newpath;
1596 	int newlen;
1597 
1598 	/*
1599 	 * We are about to resync with krtld.  krtld will reset its
1600 	 * internal module search path iff Solaris has set default_path.
1601 	 * We want to be sure we're prepending this new directory to the
1602 	 * right search path.
1603 	 */
1604 	current = (default_path == NULL) ? kobj_module_path : default_path;
1605 
1606 	newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2;
1607 	newpath = kmem_alloc(newlen, KM_SLEEP);
1608 	(void) strcpy(newpath, HVM_MOD_DIR);
1609 	(void) strcat(newpath, " ");
1610 	(void) strcat(newpath, current);
1611 
1612 	default_path = newpath;
1613 }
1614 #endif
1615 
1616 static void
1617 startup_modules(void)
1618 {
1619 	int cnt;
1620 	extern void prom_setup(void);
1621 	int32_t v, h;
1622 	char d[11];
1623 	char *cp;
1624 	cmi_hdl_t hdl;
1625 
1626 	PRM_POINT("startup_modules() starting...");
1627 
1628 #ifndef __xpv
1629 	/*
1630 	 * Initialize ten-micro second timer so that drivers will
1631 	 * not get short changed in their init phase. This was
1632 	 * not getting called until clkinit which, on fast cpu's
1633 	 * caused the drv_usecwait to be way too short.
1634 	 */
1635 	microfind();
1636 
1637 	if ((get_hwenv() & HW_XEN_HVM) != 0)
1638 		update_default_path();
1639 #endif
1640 
1641 	/*
1642 	 * Read the GMT lag from /etc/rtc_config.
1643 	 */
1644 	sgmtl(process_rtc_config_file());
1645 
1646 	/*
1647 	 * Calculate default settings of system parameters based upon
1648 	 * maxusers, yet allow to be overridden via the /etc/system file.
1649 	 */
1650 	param_calc(0);
1651 
1652 	mod_setup();
1653 
1654 	/*
1655 	 * Initialize system parameters.
1656 	 */
1657 	param_init();
1658 
1659 	/*
1660 	 * Initialize the default brands
1661 	 */
1662 	brand_init();
1663 
1664 	/*
1665 	 * maxmem is the amount of physical memory we're playing with.
1666 	 */
1667 	maxmem = physmem;
1668 
1669 	/*
1670 	 * Initialize segment management stuff.
1671 	 */
1672 	seg_init();
1673 
1674 	if (modload("fs", "specfs") == -1)
1675 		halt("Can't load specfs");
1676 
1677 	if (modload("fs", "devfs") == -1)
1678 		halt("Can't load devfs");
1679 
1680 	if (modload("fs", "dev") == -1)
1681 		halt("Can't load dev");
1682 
1683 	if (modload("fs", "procfs") == -1)
1684 		halt("Can't load procfs");
1685 
1686 	(void) modloadonly("sys", "lbl_edition");
1687 
1688 	dispinit();
1689 
1690 	/* Read cluster configuration data. */
1691 	clconf_init();
1692 
1693 #if defined(__xpv)
1694 	(void) ec_init();
1695 	gnttab_init();
1696 	(void) xs_early_init();
1697 #endif /* __xpv */
1698 
1699 	/*
1700 	 * Create a kernel device tree. First, create rootnex and
1701 	 * then invoke bus specific code to probe devices.
1702 	 */
1703 	setup_ddi();
1704 
1705 #ifdef __xpv
1706 	if (DOMAIN_IS_INITDOMAIN(xen_info))
1707 #endif
1708 	{
1709 		id_t smid;
1710 		smbios_system_t smsys;
1711 		smbios_info_t sminfo;
1712 		char *mfg;
1713 		/*
1714 		 * Load the System Management BIOS into the global ksmbios
1715 		 * handle, if an SMBIOS is present on this system.
1716 		 * Also set "si-hw-provider" property, if not already set.
1717 		 */
1718 		ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL);
1719 		if (ksmbios != NULL &&
1720 		    ((smid = smbios_info_system(ksmbios, &smsys)) != SMB_ERR) &&
1721 		    (smbios_info_common(ksmbios, smid, &sminfo)) != SMB_ERR) {
1722 			mfg = (char *)sminfo.smbi_manufacturer;
1723 			if (BOP_GETPROPLEN(bootops, "si-hw-provider") < 0) {
1724 				extern char hw_provider[];
1725 				int i;
1726 				for (i = 0; i < SYS_NMLN; i++) {
1727 					if (isprint(mfg[i]))
1728 						hw_provider[i] = mfg[i];
1729 					else {
1730 						hw_provider[i] = '\0';
1731 						break;
1732 					}
1733 				}
1734 				hw_provider[SYS_NMLN - 1] = '\0';
1735 			}
1736 		}
1737 	}
1738 
1739 
1740 	/*
1741 	 * Originally clconf_init() apparently needed the hostid.  But
1742 	 * this no longer appears to be true - it uses its own nodeid.
1743 	 * By placing the hostid logic here, we are able to make use of
1744 	 * the SMBIOS UUID.
1745 	 */
1746 	if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) {
1747 		cmn_err(CE_WARN, "Unable to set hostid");
1748 	} else {
1749 		for (v = h, cnt = 0; cnt < 10; cnt++) {
1750 			d[cnt] = (char)(v % 10);
1751 			v /= 10;
1752 			if (v == 0)
1753 				break;
1754 		}
1755 		for (cp = hw_serial; cnt >= 0; cnt--)
1756 			*cp++ = d[cnt] + '0';
1757 		*cp = 0;
1758 	}
1759 
1760 	/*
1761 	 * Set up the CPU module subsystem for the boot cpu in the native
1762 	 * case, and all physical cpu resource in the xpv dom0 case.
1763 	 * Modifies the device tree, so this must be done after
1764 	 * setup_ddi().
1765 	 */
1766 #ifdef __xpv
1767 	/*
1768 	 * If paravirtualized and on dom0 then we initialize all physical
1769 	 * cpu handles now;  if paravirtualized on a domU then do not
1770 	 * initialize.
1771 	 */
1772 	if (DOMAIN_IS_INITDOMAIN(xen_info)) {
1773 		xen_mc_lcpu_cookie_t cpi;
1774 
1775 		for (cpi = xen_physcpu_next(NULL); cpi != NULL;
1776 		    cpi = xen_physcpu_next(cpi)) {
1777 			if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA,
1778 			    xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi),
1779 			    xen_physcpu_strandid(cpi))) != NULL &&
1780 			    is_x86_feature(x86_featureset, X86FSET_MCA))
1781 				cmi_mca_init(hdl);
1782 		}
1783 	}
1784 #else
1785 	/*
1786 	 * Initialize a handle for the boot cpu - others will initialize
1787 	 * as they startup.
1788 	 */
1789 	if ((hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
1790 	    cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL) {
1791 		if (is_x86_feature(x86_featureset, X86FSET_MCA))
1792 			cmi_mca_init(hdl);
1793 		CPU->cpu_m.mcpu_cmi_hdl = hdl;
1794 	}
1795 #endif	/* __xpv */
1796 
1797 	/*
1798 	 * Fake a prom tree such that /dev/openprom continues to work
1799 	 */
1800 	PRM_POINT("startup_modules: calling prom_setup...");
1801 	prom_setup();
1802 	PRM_POINT("startup_modules: done");
1803 
1804 	/*
1805 	 * Load all platform specific modules
1806 	 */
1807 	PRM_POINT("startup_modules: calling psm_modload...");
1808 	psm_modload();
1809 
1810 	PRM_POINT("startup_modules() done");
1811 }
1812 
1813 /*
1814  * claim a "setaside" boot page for use in the kernel
1815  */
1816 page_t *
1817 boot_claim_page(pfn_t pfn)
1818 {
1819 	page_t *pp;
1820 
1821 	pp = page_numtopp_nolock(pfn);
1822 	ASSERT(pp != NULL);
1823 
1824 	if (PP_ISBOOTPAGES(pp)) {
1825 		if (pp->p_next != NULL)
1826 			pp->p_next->p_prev = pp->p_prev;
1827 		if (pp->p_prev == NULL)
1828 			bootpages = pp->p_next;
1829 		else
1830 			pp->p_prev->p_next = pp->p_next;
1831 	} else {
1832 		/*
1833 		 * htable_attach() expects a base pagesize page
1834 		 */
1835 		if (pp->p_szc != 0)
1836 			page_boot_demote(pp);
1837 		pp = page_numtopp(pfn, SE_EXCL);
1838 	}
1839 	return (pp);
1840 }
1841 
1842 /*
1843  * Walk through the pagetables looking for pages mapped in by boot.  If the
1844  * setaside flag is set the pages are expected to be returned to the
1845  * kernel later in boot, so we add them to the bootpages list.
1846  */
1847 static void
1848 protect_boot_range(uintptr_t low, uintptr_t high, int setaside)
1849 {
1850 	uintptr_t va = low;
1851 	size_t len;
1852 	uint_t prot;
1853 	pfn_t pfn;
1854 	page_t *pp;
1855 	pgcnt_t boot_protect_cnt = 0;
1856 
1857 	while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) {
1858 		if (va + len >= high)
1859 			panic("0x%lx byte mapping at 0x%p exceeds boot's "
1860 			    "legal range.", len, (void *)va);
1861 
1862 		while (len > 0) {
1863 			pp = page_numtopp_alloc(pfn);
1864 			if (pp != NULL) {
1865 				if (setaside == 0)
1866 					panic("Unexpected mapping by boot.  "
1867 					    "addr=%p pfn=%lx\n",
1868 					    (void *)va, pfn);
1869 
1870 				pp->p_next = bootpages;
1871 				pp->p_prev = NULL;
1872 				PP_SETBOOTPAGES(pp);
1873 				if (bootpages != NULL) {
1874 					bootpages->p_prev = pp;
1875 				}
1876 				bootpages = pp;
1877 				++boot_protect_cnt;
1878 			}
1879 
1880 			++pfn;
1881 			len -= MMU_PAGESIZE;
1882 			va += MMU_PAGESIZE;
1883 		}
1884 	}
1885 	PRM_DEBUG(boot_protect_cnt);
1886 }
1887 
1888 /*
1889  * Establish the final size of the kernel's heap, size of segmap, segkp, etc.
1890  */
1891 static void
1892 layout_kernel_va(void)
1893 {
1894 	const size_t physmem_size = mmu_ptob(physmem);
1895 	size_t size;
1896 
1897 	PRM_POINT("layout_kernel_va() starting...");
1898 
1899 	kpm_vbase = (caddr_t)segkpm_base;
1900 	kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1));
1901 	if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base)
1902 		panic("not enough room for kpm!");
1903 	PRM_DEBUG(kpm_size);
1904 	PRM_DEBUG(kpm_vbase);
1905 
1906 	segkp_base = (caddr_t)valloc_base + valloc_sz;
1907 	if (!segkp_fromheap) {
1908 		size = mmu_ptob(segkpsize);
1909 
1910 		/*
1911 		 * determine size of segkp
1912 		 */
1913 		if (size < SEGKPMINSIZE || size > SEGKPMAXSIZE) {
1914 			size = SEGKPDEFSIZE;
1915 			cmn_err(CE_WARN, "!Illegal value for segkpsize. "
1916 			    "segkpsize has been reset to %ld pages",
1917 			    mmu_btop(size));
1918 		}
1919 		size = MIN(size, MAX(SEGKPMINSIZE, physmem_size));
1920 
1921 		segkpsize = mmu_btop(ROUND_UP_LPAGE(size));
1922 	}
1923 	PRM_DEBUG(segkp_base);
1924 	PRM_DEBUG(segkpsize);
1925 
1926 	/*
1927 	 * segkvmm: backing for vmm guest memory. Like segzio, we have a
1928 	 * separate segment for two reasons: it makes it easy to skip our pages
1929 	 * on kernel crash dumps, and it helps avoid fragmentation.  With this
1930 	 * segment, we're expecting significantly-sized allocations only; we'll
1931 	 * default to 4x the size of physmem.
1932 	 */
1933 	segkvmm_base = segkp_base + mmu_ptob(segkpsize);
1934 	size = segkvmmsize != 0 ? mmu_ptob(segkvmmsize) : (physmem_size * 4);
1935 
1936 	size = MAX(size, SEGVMMMINSIZE);
1937 	segkvmmsize = mmu_btop(ROUND_UP_LPAGE(size));
1938 
1939 	PRM_DEBUG(segkvmmsize);
1940 	PRM_DEBUG(segkvmm_base);
1941 
1942 	/*
1943 	 * segzio is used for ZFS cached data.  For segzio, we use 1.5x physmem.
1944 	 */
1945 	segzio_base = segkvmm_base + mmu_ptob(segkvmmsize);
1946 	if (segzio_fromheap) {
1947 		segziosize = 0;
1948 	} else {
1949 		size = (segziosize != 0) ? mmu_ptob(segziosize) :
1950 		    (physmem_size * 3) / 2;
1951 
1952 		size = MAX(size, SEGZIOMINSIZE);
1953 		segziosize = mmu_btop(ROUND_UP_LPAGE(size));
1954 	}
1955 	PRM_DEBUG(segziosize);
1956 	PRM_DEBUG(segzio_base);
1957 
1958 	/*
1959 	 * Put the range of VA for device mappings next, kmdb knows to not
1960 	 * grep in this range of addresses.
1961 	 */
1962 	toxic_addr =
1963 	    ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize));
1964 	PRM_DEBUG(toxic_addr);
1965 	segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size);
1966 
1967 	/*
1968 	 * Users can change segmapsize through eeprom. If the variable
1969 	 * is tuned through eeprom, there is no upper bound on the
1970 	 * size of segmap.
1971 	 */
1972 	segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT);
1973 
1974 	PRM_DEBUG(segmap_start);
1975 	PRM_DEBUG(segmapsize);
1976 	kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize);
1977 	PRM_DEBUG(kernelheap);
1978 	PRM_POINT("layout_kernel_va() done...");
1979 }
1980 
1981 /*
1982  * Finish initializing the VM system, now that we are no longer
1983  * relying on the boot time memory allocators.
1984  */
1985 static void
1986 startup_vm(void)
1987 {
1988 	struct segmap_crargs a;
1989 
1990 	extern int use_brk_lpg, use_stk_lpg;
1991 
1992 	PRM_POINT("startup_vm() starting...");
1993 
1994 	/*
1995 	 * Initialize the hat layer.
1996 	 */
1997 	hat_init();
1998 
1999 	/*
2000 	 * Do final allocations of HAT data structures that need to
2001 	 * be allocated before quiescing the boot loader.
2002 	 */
2003 	PRM_POINT("Calling hat_kern_alloc()...");
2004 	hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap);
2005 	PRM_POINT("hat_kern_alloc() done");
2006 
2007 #ifndef __xpv
2008 	/*
2009 	 * Setup Page Attribute Table
2010 	 */
2011 	pat_sync();
2012 #endif
2013 
2014 	/*
2015 	 * The next two loops are done in distinct steps in order
2016 	 * to be sure that any page that is doubly mapped (both above
2017 	 * KERNEL_TEXT and below kernelbase) is dealt with correctly.
2018 	 * Note this may never happen, but it might someday.
2019 	 */
2020 	bootpages = NULL;
2021 	PRM_POINT("Protecting boot pages");
2022 
2023 	/*
2024 	 * Protect any pages mapped above KERNEL_TEXT that somehow have
2025 	 * page_t's. This can only happen if something weird allocated
2026 	 * in this range (like kadb/kmdb).
2027 	 */
2028 	protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0);
2029 
2030 	/*
2031 	 * Before we can take over memory allocation/mapping from the boot
2032 	 * loader we must remove from our free page lists any boot allocated
2033 	 * pages that stay mapped until release_bootstrap().
2034 	 */
2035 	protect_boot_range(0, kernelbase, 1);
2036 
2037 
2038 	/*
2039 	 * Switch to running on regular HAT (not boot_mmu)
2040 	 */
2041 	PRM_POINT("Calling hat_kern_setup()...");
2042 	hat_kern_setup();
2043 
2044 	/*
2045 	 * It is no longer safe to call BOP_ALLOC(), so make sure we don't.
2046 	 */
2047 	bop_no_more_mem();
2048 
2049 	PRM_POINT("hat_kern_setup() done");
2050 
2051 	hat_cpu_online(CPU);
2052 
2053 	/*
2054 	 * Initialize VM system
2055 	 */
2056 	PRM_POINT("Calling kvm_init()...");
2057 	kvm_init();
2058 	PRM_POINT("kvm_init() done");
2059 
2060 	/*
2061 	 * Tell kmdb that the VM system is now working
2062 	 */
2063 	if (boothowto & RB_DEBUG)
2064 		kdi_dvec_vmready();
2065 
2066 #if defined(__xpv)
2067 	/*
2068 	 * Populate the I/O pool on domain 0
2069 	 */
2070 	if (DOMAIN_IS_INITDOMAIN(xen_info)) {
2071 		extern long populate_io_pool(void);
2072 		long init_io_pool_cnt;
2073 
2074 		PRM_POINT("Populating reserve I/O page pool");
2075 		init_io_pool_cnt = populate_io_pool();
2076 		PRM_DEBUG(init_io_pool_cnt);
2077 	}
2078 #endif
2079 	/*
2080 	 * Mangle the brand string etc.
2081 	 */
2082 	cpuid_pass3(CPU);
2083 
2084 #if defined(__amd64)
2085 
2086 	/*
2087 	 * Create the device arena for toxic (to dtrace/kmdb) mappings.
2088 	 */
2089 	device_arena = vmem_create("device", (void *)toxic_addr,
2090 	    toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP);
2091 
2092 #else	/* __i386 */
2093 
2094 	/*
2095 	 * allocate the bit map that tracks toxic pages
2096 	 */
2097 	toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase));
2098 	PRM_DEBUG(toxic_bit_map_len);
2099 	toxic_bit_map =
2100 	    kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP);
2101 	ASSERT(toxic_bit_map != NULL);
2102 	PRM_DEBUG(toxic_bit_map);
2103 
2104 #endif	/* __i386 */
2105 
2106 
2107 	/*
2108 	 * Now that we've got more VA, as well as the ability to allocate from
2109 	 * it, tell the debugger.
2110 	 */
2111 	if (boothowto & RB_DEBUG)
2112 		kdi_dvec_memavail();
2113 
2114 #if !defined(__xpv)
2115 	/*
2116 	 * Map page pfn=0 for drivers, such as kd, that need to pick up
2117 	 * parameters left there by controllers/BIOS.
2118 	 */
2119 	PRM_POINT("setup up p0_va");
2120 	p0_va = i86devmap(0, 1, PROT_READ);
2121 	PRM_DEBUG(p0_va);
2122 #endif
2123 
2124 	cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n",
2125 	    physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled));
2126 
2127 	/*
2128 	 * disable automatic large pages for small memory systems or
2129 	 * when the disable flag is set.
2130 	 *
2131 	 * Do not yet consider page sizes larger than 2m/4m.
2132 	 */
2133 	if (!auto_lpg_disable && mmu.max_page_level > 0) {
2134 		max_uheap_lpsize = LEVEL_SIZE(1);
2135 		max_ustack_lpsize = LEVEL_SIZE(1);
2136 		max_privmap_lpsize = LEVEL_SIZE(1);
2137 		max_uidata_lpsize = LEVEL_SIZE(1);
2138 		max_utext_lpsize = LEVEL_SIZE(1);
2139 		max_shm_lpsize = LEVEL_SIZE(1);
2140 	}
2141 	if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 ||
2142 	    auto_lpg_disable) {
2143 		use_brk_lpg = 0;
2144 		use_stk_lpg = 0;
2145 	}
2146 	mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level);
2147 
2148 	PRM_POINT("Calling hat_init_finish()...");
2149 	hat_init_finish();
2150 	PRM_POINT("hat_init_finish() done");
2151 
2152 	/*
2153 	 * Initialize the segkp segment type.
2154 	 */
2155 	rw_enter(&kas.a_lock, RW_WRITER);
2156 	PRM_POINT("Attaching segkp");
2157 	if (segkp_fromheap) {
2158 		segkp->s_as = &kas;
2159 	} else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize),
2160 	    segkp) < 0) {
2161 		panic("startup: cannot attach segkp");
2162 		/*NOTREACHED*/
2163 	}
2164 	PRM_POINT("Doing segkp_create()");
2165 	if (segkp_create(segkp) != 0) {
2166 		panic("startup: segkp_create failed");
2167 		/*NOTREACHED*/
2168 	}
2169 	PRM_DEBUG(segkp);
2170 	rw_exit(&kas.a_lock);
2171 
2172 	/*
2173 	 * kpm segment
2174 	 */
2175 	segmap_kpm = 0;
2176 	if (kpm_desired)
2177 		kpm_init();
2178 
2179 	/*
2180 	 * Now create segmap segment.
2181 	 */
2182 	rw_enter(&kas.a_lock, RW_WRITER);
2183 	if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) {
2184 		panic("cannot attach segmap");
2185 		/*NOTREACHED*/
2186 	}
2187 	PRM_DEBUG(segmap);
2188 
2189 	a.prot = PROT_READ | PROT_WRITE;
2190 	a.shmsize = 0;
2191 	a.nfreelist = segmapfreelists;
2192 
2193 	if (segmap_create(segmap, (caddr_t)&a) != 0)
2194 		panic("segmap_create segmap");
2195 	rw_exit(&kas.a_lock);
2196 
2197 	setup_vaddr_for_ppcopy(CPU);
2198 
2199 	segdev_init();
2200 #if defined(__xpv)
2201 	if (DOMAIN_IS_INITDOMAIN(xen_info))
2202 #endif
2203 		pmem_init();
2204 
2205 	PRM_POINT("startup_vm() done");
2206 }
2207 
2208 /*
2209  * Load a tod module for the non-standard tod part found on this system.
2210  */
2211 static void
2212 load_tod_module(char *todmod)
2213 {
2214 	if (modload("tod", todmod) == -1)
2215 		halt("Can't load TOD module");
2216 }
2217 
2218 static void
2219 startup_end(void)
2220 {
2221 	int i;
2222 	extern void setx86isalist(void);
2223 	extern void cpu_event_init(void);
2224 
2225 	PRM_POINT("startup_end() starting...");
2226 
2227 	/*
2228 	 * Perform tasks that get done after most of the VM
2229 	 * initialization has been done but before the clock
2230 	 * and other devices get started.
2231 	 */
2232 	kern_setup1();
2233 
2234 	/*
2235 	 * Perform CPC initialization for this CPU.
2236 	 */
2237 	kcpc_hw_init(CPU);
2238 
2239 	/*
2240 	 * Initialize cpu event framework.
2241 	 */
2242 	cpu_event_init();
2243 
2244 #if defined(OPTERON_WORKAROUND_6323525)
2245 	if (opteron_workaround_6323525)
2246 		patch_workaround_6323525();
2247 #endif
2248 	/*
2249 	 * If needed, load TOD module now so that ddi_get_time(9F) etc. work
2250 	 * (For now, "needed" is defined as set tod_module_name in /etc/system)
2251 	 */
2252 	if (tod_module_name != NULL) {
2253 		PRM_POINT("load_tod_module()");
2254 		load_tod_module(tod_module_name);
2255 	}
2256 
2257 #if defined(__xpv)
2258 	/*
2259 	 * Forceload interposing TOD module for the hypervisor.
2260 	 */
2261 	PRM_POINT("load_tod_module()");
2262 	load_tod_module("xpvtod");
2263 #endif
2264 
2265 	/*
2266 	 * Configure the system.
2267 	 */
2268 	PRM_POINT("Calling configure()...");
2269 	configure();		/* set up devices */
2270 	PRM_POINT("configure() done");
2271 
2272 	/*
2273 	 * We can now setup for XSAVE because fpu_probe is done in configure().
2274 	 */
2275 	if (fp_save_mech == FP_XSAVE) {
2276 		xsave_setup_msr(CPU);
2277 	}
2278 
2279 	/*
2280 	 * Set the isa_list string to the defined instruction sets we
2281 	 * support.
2282 	 */
2283 	setx86isalist();
2284 	cpu_intr_alloc(CPU, NINTR_THREADS);
2285 	psm_install();
2286 
2287 	/*
2288 	 * We're done with bootops.  We don't unmap the bootstrap yet because
2289 	 * we're still using bootsvcs.
2290 	 */
2291 	PRM_POINT("NULLing out bootops");
2292 	*bootopsp = (struct bootops *)NULL;
2293 	bootops = (struct bootops *)NULL;
2294 
2295 #if defined(__xpv)
2296 	ec_init_debug_irq();
2297 	xs_domu_init();
2298 #endif
2299 
2300 #if !defined(__xpv)
2301 	/*
2302 	 * Intel IOMMU has been setup/initialized in ddi_impl.c
2303 	 * Start it up now.
2304 	 */
2305 	immu_startup();
2306 
2307 	/*
2308 	 * Now that we're no longer going to drop into real mode for a BIOS call
2309 	 * via bootops, we can enable PCID (which requires CR0.PG).
2310 	 */
2311 	enable_pcid();
2312 #endif
2313 
2314 	PRM_POINT("Enabling interrupts");
2315 	(*picinitf)();
2316 	sti();
2317 #if defined(__xpv)
2318 	ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0);
2319 	xen_late_startup();
2320 #endif
2321 
2322 	(void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1,
2323 	    "softlevel1", NULL, NULL); /* XXX to be moved later */
2324 
2325 	/*
2326 	 * Register software interrupt handlers for ddi_periodic_add(9F).
2327 	 * Software interrupts up to the level 10 are supported.
2328 	 */
2329 	for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) {
2330 		(void) add_avsoftintr((void *)&softlevel_hdl[i-1], i,
2331 		    (avfunc)(uintptr_t)ddi_periodic_softintr, "ddi_periodic",
2332 		    (caddr_t)(uintptr_t)i, NULL);
2333 	}
2334 
2335 #if !defined(__xpv)
2336 	if (modload("drv", "amd_iommu") < 0) {
2337 		PRM_POINT("No AMD IOMMU present\n");
2338 	} else if (ddi_hold_installed_driver(ddi_name_to_major(
2339 	    "amd_iommu")) == NULL) {
2340 		prom_printf("ERROR: failed to attach AMD IOMMU\n");
2341 	}
2342 #endif
2343 	post_startup_cpu_fixups();
2344 
2345 	PRM_POINT("startup_end() done");
2346 }
2347 
2348 /*
2349  * Don't remove the following 2 variables.  They are necessary
2350  * for reading the hostid from the legacy file (/kernel/misc/sysinit).
2351  */
2352 char *_hs1107 = hw_serial;
2353 ulong_t  _bdhs34;
2354 
2355 void
2356 post_startup(void)
2357 {
2358 	extern void cpupm_init(cpu_t *);
2359 	extern void cpu_event_init_cpu(cpu_t *);
2360 
2361 	/*
2362 	 * Set the system wide, processor-specific flags to be passed
2363 	 * to userland via the aux vector for performance hints and
2364 	 * instruction set extensions.
2365 	 */
2366 	bind_hwcap();
2367 
2368 #ifdef __xpv
2369 	if (DOMAIN_IS_INITDOMAIN(xen_info))
2370 #endif
2371 	{
2372 #if defined(__xpv)
2373 		xpv_panic_init();
2374 #else
2375 		/*
2376 		 * Startup the memory scrubber.
2377 		 * XXPV	This should be running somewhere ..
2378 		 */
2379 		if ((get_hwenv() & HW_VIRTUAL) == 0)
2380 			memscrub_init();
2381 #endif
2382 	}
2383 
2384 	/*
2385 	 * Complete CPU module initialization
2386 	 */
2387 	cmi_post_startup();
2388 
2389 	/*
2390 	 * Perform forceloading tasks for /etc/system.
2391 	 */
2392 	(void) mod_sysctl(SYS_FORCELOAD, NULL);
2393 
2394 	/*
2395 	 * ON4.0: Force /proc module in until clock interrupt handle fixed
2396 	 * ON4.0: This must be fixed or restated in /etc/systems.
2397 	 */
2398 	(void) modload("fs", "procfs");
2399 
2400 	(void) i_ddi_attach_hw_nodes("pit_beep");
2401 
2402 #if defined(__i386)
2403 	/*
2404 	 * Check for required functional Floating Point hardware,
2405 	 * unless FP hardware explicitly disabled.
2406 	 */
2407 	if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO))
2408 		halt("No working FP hardware found");
2409 #endif
2410 
2411 	maxmem = freemem;
2412 
2413 	cpu_event_init_cpu(CPU);
2414 	cpupm_init(CPU);
2415 	(void) mach_cpu_create_device_node(CPU, NULL);
2416 
2417 	pg_init();
2418 }
2419 
2420 static int
2421 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr)
2422 {
2423 	return ((pp->p_pagenum >= btop(low_addr)) &&
2424 	    (pp->p_pagenum < btopr(high_addr)));
2425 }
2426 
2427 static int
2428 pp_in_module(page_t *pp, const rd_existing_t *modranges)
2429 {
2430 	uint_t i;
2431 
2432 	for (i = 0; modranges[i].phys != 0; i++) {
2433 		if (pp_in_range(pp, modranges[i].phys,
2434 		    modranges[i].phys + modranges[i].size))
2435 			return (1);
2436 	}
2437 
2438 	return (0);
2439 }
2440 
2441 void
2442 release_bootstrap(void)
2443 {
2444 	int root_is_ramdisk;
2445 	page_t *pp;
2446 	extern void kobj_boot_unmountroot(void);
2447 	extern dev_t rootdev;
2448 	uint_t i;
2449 	char propname[32];
2450 	rd_existing_t *modranges;
2451 #if !defined(__xpv)
2452 	pfn_t	pfn;
2453 #endif
2454 
2455 	/*
2456 	 * Save the bootfs module ranges so that we can reserve them below
2457 	 * for the real bootfs.
2458 	 */
2459 	modranges = kmem_alloc(sizeof (rd_existing_t) * MAX_BOOT_MODULES,
2460 	    KM_SLEEP);
2461 	for (i = 0; ; i++) {
2462 		uint64_t start, size;
2463 
2464 		modranges[i].phys = 0;
2465 
2466 		(void) snprintf(propname, sizeof (propname),
2467 		    "module-addr-%u", i);
2468 		if (do_bsys_getproplen(NULL, propname) <= 0)
2469 			break;
2470 		(void) do_bsys_getprop(NULL, propname, &start);
2471 
2472 		(void) snprintf(propname, sizeof (propname),
2473 		    "module-size-%u", i);
2474 		if (do_bsys_getproplen(NULL, propname) <= 0)
2475 			break;
2476 		(void) do_bsys_getprop(NULL, propname, &size);
2477 
2478 		modranges[i].phys = start;
2479 		modranges[i].size = size;
2480 	}
2481 
2482 	/* unmount boot ramdisk and release kmem usage */
2483 	kobj_boot_unmountroot();
2484 
2485 	/*
2486 	 * We're finished using the boot loader so free its pages.
2487 	 */
2488 	PRM_POINT("Unmapping lower boot pages");
2489 
2490 	clear_boot_mappings(0, _userlimit);
2491 
2492 	postbootkernelbase = kernelbase;
2493 
2494 	/*
2495 	 * If root isn't on ramdisk, destroy the hardcoded
2496 	 * ramdisk node now and release the memory. Else,
2497 	 * ramdisk memory is kept in rd_pages.
2498 	 */
2499 	root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk"));
2500 	if (!root_is_ramdisk) {
2501 		dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0);
2502 		ASSERT(dip && ddi_get_parent(dip) == ddi_root_node());
2503 		ndi_rele_devi(dip);	/* held from ddi_find_devinfo */
2504 		(void) ddi_remove_child(dip, 0);
2505 	}
2506 
2507 	PRM_POINT("Releasing boot pages");
2508 	while (bootpages) {
2509 		extern uint64_t ramdisk_start, ramdisk_end;
2510 		pp = bootpages;
2511 		bootpages = pp->p_next;
2512 
2513 
2514 		/* Keep pages for the lower 64K */
2515 		if (pp_in_range(pp, 0, 0x40000)) {
2516 			pp->p_next = lower_pages;
2517 			lower_pages = pp;
2518 			lower_pages_count++;
2519 			continue;
2520 		}
2521 
2522 		if (root_is_ramdisk && pp_in_range(pp, ramdisk_start,
2523 		    ramdisk_end) || pp_in_module(pp, modranges)) {
2524 			pp->p_next = rd_pages;
2525 			rd_pages = pp;
2526 			continue;
2527 		}
2528 		pp->p_next = (struct page *)0;
2529 		pp->p_prev = (struct page *)0;
2530 		PP_CLRBOOTPAGES(pp);
2531 		page_free(pp, 1);
2532 	}
2533 	PRM_POINT("Boot pages released");
2534 
2535 	kmem_free(modranges, sizeof (rd_existing_t) * 99);
2536 
2537 #if !defined(__xpv)
2538 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */
2539 	/*
2540 	 * Find 1 page below 1 MB so that other processors can boot up or
2541 	 * so that any processor can resume.
2542 	 * Make sure it has a kernel VA as well as a 1:1 mapping.
2543 	 * We should have just free'd one up.
2544 	 */
2545 
2546 	/*
2547 	 * 0x10 pages is 64K.  Leave the bottom 64K alone
2548 	 * for BIOS.
2549 	 */
2550 	for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) {
2551 		if (page_numtopp_alloc(pfn) == NULL)
2552 			continue;
2553 		rm_platter_va = i86devmap(pfn, 1,
2554 		    PROT_READ | PROT_WRITE | PROT_EXEC);
2555 		rm_platter_pa = ptob(pfn);
2556 		break;
2557 	}
2558 	if (pfn == btop(1*1024*1024) && use_mp)
2559 		panic("No page below 1M available for starting "
2560 		    "other processors or for resuming from system-suspend");
2561 #endif	/* !__xpv */
2562 }
2563 
2564 /*
2565  * Initialize the platform-specific parts of a page_t.
2566  */
2567 void
2568 add_physmem_cb(page_t *pp, pfn_t pnum)
2569 {
2570 	pp->p_pagenum = pnum;
2571 	pp->p_mapping = NULL;
2572 	pp->p_embed = 0;
2573 	pp->p_share = 0;
2574 	pp->p_mlentry = 0;
2575 }
2576 
2577 /*
2578  * kphysm_init() initializes physical memory.
2579  */
2580 static pgcnt_t
2581 kphysm_init(page_t *pp, pgcnt_t npages)
2582 {
2583 	struct memlist	*pmem;
2584 	struct memseg	*cur_memseg;
2585 	pfn_t		base_pfn;
2586 	pfn_t		end_pfn;
2587 	pgcnt_t		num;
2588 	pgcnt_t		pages_done = 0;
2589 	uint64_t	addr;
2590 	uint64_t	size;
2591 	extern pfn_t	ddiphysmin;
2592 	extern int	mnode_xwa;
2593 	int		ms = 0, me = 0;
2594 
2595 	ASSERT(page_hash != NULL && page_hashsz != 0);
2596 
2597 	cur_memseg = memseg_base;
2598 	for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) {
2599 		/*
2600 		 * In a 32 bit kernel can't use higher memory if we're
2601 		 * not booting in PAE mode. This check takes care of that.
2602 		 */
2603 		addr = pmem->ml_address;
2604 		size = pmem->ml_size;
2605 		if (btop(addr) > physmax)
2606 			continue;
2607 
2608 		/*
2609 		 * align addr and size - they may not be at page boundaries
2610 		 */
2611 		if ((addr & MMU_PAGEOFFSET) != 0) {
2612 			addr += MMU_PAGEOFFSET;
2613 			addr &= ~(uint64_t)MMU_PAGEOFFSET;
2614 			size -= addr - pmem->ml_address;
2615 		}
2616 
2617 		/* only process pages below or equal to physmax */
2618 		if ((btop(addr + size) - 1) > physmax)
2619 			size = ptob(physmax - btop(addr) + 1);
2620 
2621 		num = btop(size);
2622 		if (num == 0)
2623 			continue;
2624 
2625 		if (num > npages)
2626 			num = npages;
2627 
2628 		npages -= num;
2629 		pages_done += num;
2630 		base_pfn = btop(addr);
2631 
2632 		if (prom_debug)
2633 			prom_printf("MEMSEG addr=0x%" PRIx64
2634 			    " pgs=0x%lx pfn 0x%lx-0x%lx\n",
2635 			    addr, num, base_pfn, base_pfn + num);
2636 
2637 		/*
2638 		 * Ignore pages below ddiphysmin to simplify ddi memory
2639 		 * allocation with non-zero addr_lo requests.
2640 		 */
2641 		if (base_pfn < ddiphysmin) {
2642 			if (base_pfn + num <= ddiphysmin)
2643 				continue;
2644 			pp += (ddiphysmin - base_pfn);
2645 			num -= (ddiphysmin - base_pfn);
2646 			base_pfn = ddiphysmin;
2647 		}
2648 
2649 		/*
2650 		 * mnode_xwa is greater than 1 when large pages regions can
2651 		 * cross memory node boundaries. To prevent the formation
2652 		 * of these large pages, configure the memsegs based on the
2653 		 * memory node ranges which had been made non-contiguous.
2654 		 */
2655 		end_pfn = base_pfn + num - 1;
2656 		if (mnode_xwa > 1) {
2657 			ms = PFN_2_MEM_NODE(base_pfn);
2658 			me = PFN_2_MEM_NODE(end_pfn);
2659 
2660 			if (ms != me) {
2661 				/*
2662 				 * current range spans more than 1 memory node.
2663 				 * Set num to only the pfn range in the start
2664 				 * memory node.
2665 				 */
2666 				num = mem_node_config[ms].physmax - base_pfn
2667 				    + 1;
2668 				ASSERT(end_pfn > mem_node_config[ms].physmax);
2669 			}
2670 		}
2671 
2672 		for (;;) {
2673 			/*
2674 			 * Build the memsegs entry
2675 			 */
2676 			cur_memseg->pages = pp;
2677 			cur_memseg->epages = pp + num;
2678 			cur_memseg->pages_base = base_pfn;
2679 			cur_memseg->pages_end = base_pfn + num;
2680 
2681 			/*
2682 			 * Insert into memseg list in decreasing pfn range
2683 			 * order. Low memory is typically more fragmented such
2684 			 * that this ordering keeps the larger ranges at the
2685 			 * front of the list for code that searches memseg.
2686 			 * This ASSERTS that the memsegs coming in from boot
2687 			 * are in increasing physical address order and not
2688 			 * contiguous.
2689 			 */
2690 			if (memsegs != NULL) {
2691 				ASSERT(cur_memseg->pages_base >=
2692 				    memsegs->pages_end);
2693 				cur_memseg->next = memsegs;
2694 			}
2695 			memsegs = cur_memseg;
2696 
2697 			/*
2698 			 * add_physmem() initializes the PSM part of the page
2699 			 * struct by calling the PSM back with add_physmem_cb().
2700 			 * In addition it coalesces pages into larger pages as
2701 			 * it initializes them.
2702 			 */
2703 			add_physmem(pp, num, base_pfn);
2704 			cur_memseg++;
2705 			availrmem_initial += num;
2706 			availrmem += num;
2707 
2708 			pp += num;
2709 			if (ms >= me)
2710 				break;
2711 
2712 			/* process next memory node range */
2713 			ms++;
2714 			base_pfn = mem_node_config[ms].physbase;
2715 
2716 			if (mnode_xwa > 1) {
2717 				num = MIN(mem_node_config[ms].physmax,
2718 				    end_pfn) - base_pfn + 1;
2719 			} else {
2720 				num = mem_node_config[ms].physmax -
2721 				    base_pfn + 1;
2722 			}
2723 		}
2724 	}
2725 
2726 	PRM_DEBUG(availrmem_initial);
2727 	PRM_DEBUG(availrmem);
2728 	PRM_DEBUG(freemem);
2729 	build_pfn_hash();
2730 	return (pages_done);
2731 }
2732 
2733 /*
2734  * Kernel VM initialization.
2735  */
2736 static void
2737 kvm_init(void)
2738 {
2739 	ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0);
2740 
2741 	/*
2742 	 * Put the kernel segments in kernel address space.
2743 	 */
2744 	rw_enter(&kas.a_lock, RW_WRITER);
2745 	as_avlinit(&kas);
2746 
2747 	(void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg);
2748 	(void) segkmem_create(&ktextseg);
2749 
2750 	(void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc);
2751 	(void) segkmem_create(&kvalloc);
2752 
2753 	(void) seg_attach(&kas, kernelheap,
2754 	    ekernelheap - kernelheap, &kvseg);
2755 	(void) segkmem_create(&kvseg);
2756 
2757 	if (core_size > 0) {
2758 		PRM_POINT("attaching kvseg_core");
2759 		(void) seg_attach(&kas, (caddr_t)core_base, core_size,
2760 		    &kvseg_core);
2761 		(void) segkmem_create(&kvseg_core);
2762 	}
2763 
2764 	PRM_POINT("attaching segkvmm");
2765 	(void) seg_attach(&kas, segkvmm_base, mmu_ptob(segkvmmsize), &kvmmseg);
2766 	(void) segkmem_create(&kvmmseg);
2767 	segkmem_kvmm_init(segkvmm_base, mmu_ptob(segkvmmsize));
2768 
2769 	if (segziosize > 0) {
2770 		PRM_POINT("attaching segzio");
2771 		(void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2772 		    &kzioseg);
2773 		(void) segkmem_create(&kzioseg);
2774 
2775 		/* create zio area covering new segment */
2776 		segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2777 	}
2778 
2779 	(void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2780 	(void) segkmem_create(&kdebugseg);
2781 
2782 	rw_exit(&kas.a_lock);
2783 
2784 	/*
2785 	 * Ensure that the red zone at kernelbase is never accessible.
2786 	 */
2787 	PRM_POINT("protecting redzone");
2788 	(void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0);
2789 
2790 	/*
2791 	 * Make the text writable so that it can be hot patched by DTrace.
2792 	 */
2793 	(void) as_setprot(&kas, s_text, e_modtext - s_text,
2794 	    PROT_READ | PROT_WRITE | PROT_EXEC);
2795 
2796 	/*
2797 	 * Make data writable until end.
2798 	 */
2799 	(void) as_setprot(&kas, s_data, e_moddata - s_data,
2800 	    PROT_READ | PROT_WRITE | PROT_EXEC);
2801 }
2802 
2803 #ifndef __xpv
2804 /*
2805  * Solaris adds an entry for Write Combining caching to the PAT
2806  */
2807 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE;
2808 
2809 void
2810 pat_sync(void)
2811 {
2812 	ulong_t	cr0, cr0_orig, cr4;
2813 
2814 	if (!is_x86_feature(x86_featureset, X86FSET_PAT))
2815 		return;
2816 	cr0_orig = cr0 = getcr0();
2817 	cr4 = getcr4();
2818 
2819 	/* disable caching and flush all caches and TLBs */
2820 	cr0 |= CR0_CD;
2821 	cr0 &= ~CR0_NW;
2822 	setcr0(cr0);
2823 	invalidate_cache();
2824 	if (cr4 & CR4_PGE) {
2825 		setcr4(cr4 & ~(ulong_t)CR4_PGE);
2826 		setcr4(cr4);
2827 	} else {
2828 		reload_cr3();
2829 	}
2830 
2831 	/* add our entry to the PAT */
2832 	wrmsr(REG_PAT, pat_attr_reg);
2833 
2834 	/* flush TLBs and cache again, then reenable cr0 caching */
2835 	if (cr4 & CR4_PGE) {
2836 		setcr4(cr4 & ~(ulong_t)CR4_PGE);
2837 		setcr4(cr4);
2838 	} else {
2839 		reload_cr3();
2840 	}
2841 	invalidate_cache();
2842 	setcr0(cr0_orig);
2843 }
2844 
2845 #endif /* !__xpv */
2846 
2847 #if defined(_SOFT_HOSTID)
2848 /*
2849  * On platforms that do not have a hardware serial number, attempt
2850  * to set one based on the contents of /etc/hostid.  If this file does
2851  * not exist, assume that we are to generate a new hostid and set
2852  * it in the kernel, for subsequent saving by a userland process
2853  * once the system is up and the root filesystem is mounted r/w.
2854  *
2855  * In order to gracefully support upgrade on OpenSolaris, if
2856  * /etc/hostid does not exist, we will attempt to get a serial number
2857  * using the legacy method (/kernel/misc/sysinit).
2858  *
2859  * If that isn't present, we attempt to use an SMBIOS UUID, which is
2860  * a hardware serial number.  Note that we don't automatically trust
2861  * all SMBIOS UUIDs (some older platforms are defective and ship duplicate
2862  * UUIDs in violation of the standard), we check against a blacklist.
2863  *
2864  * In an attempt to make the hostid less prone to abuse
2865  * (for license circumvention, etc), we store it in /etc/hostid
2866  * in rot47 format.
2867  */
2868 extern volatile unsigned long tenmicrodata;
2869 static int atoi(char *);
2870 
2871 /*
2872  * Set this to non-zero in /etc/system if you think your SMBIOS returns a
2873  * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist
2874  * array can be updated.)
2875  */
2876 int smbios_broken_uuid = 0;
2877 
2878 /*
2879  * List of known bad UUIDs.  This is just the lower 32-bit values, since
2880  * that's what we use for the host id.  If your hostid falls here, you need
2881  * to contact your hardware OEM for a fix for your BIOS.
2882  */
2883 static unsigned char
2884 smbios_uuid_blacklist[][16] = {
2885 
2886 	{	/* Reported bad UUID (Google search) */
2887 		0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05,
2888 		0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09,
2889 	},
2890 	{	/* Known bad DELL UUID */
2891 		0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10,
2892 		0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20,
2893 	},
2894 	{	/* Uninitialized flash */
2895 		0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
2896 		0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
2897 	},
2898 	{	/* All zeros */
2899 		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
2900 		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
2901 	},
2902 };
2903 
2904 static int32_t
2905 uuid_to_hostid(const uint8_t *uuid)
2906 {
2907 	/*
2908 	 * Although the UUIDs are 128-bits, they may not distribute entropy
2909 	 * evenly.  We would like to use SHA or MD5, but those are located
2910 	 * in loadable modules and not available this early in boot.  As we
2911 	 * don't need the values to be cryptographically strong, we just
2912 	 * generate 32-bit vaue by xor'ing the various sequences together,
2913 	 * which ensures that the entire UUID contributes to the hostid.
2914 	 */
2915 	uint32_t	id = 0;
2916 
2917 	/* first check against the blacklist */
2918 	for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) {
2919 		if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) {
2920 			cmn_err(CE_CONT, "?Broken SMBIOS UUID. "
2921 			    "Contact BIOS manufacturer for repair.\n");
2922 			return ((int32_t)HW_INVALID_HOSTID);
2923 		}
2924 	}
2925 
2926 	for (int i = 0; i < 16; i++)
2927 		id ^= ((uuid[i]) << (8 * (i % sizeof (id))));
2928 
2929 	/* Make sure return value is positive */
2930 	return (id & 0x7fffffff);
2931 }
2932 
2933 static int32_t
2934 set_soft_hostid(void)
2935 {
2936 	struct _buf *file;
2937 	char tokbuf[MAXNAMELEN];
2938 	token_t token;
2939 	int done = 0;
2940 	u_longlong_t tmp;
2941 	int i;
2942 	int32_t hostid = (int32_t)HW_INVALID_HOSTID;
2943 	unsigned char *c;
2944 	hrtime_t tsc;
2945 	smbios_system_t smsys;
2946 
2947 	/*
2948 	 * If /etc/hostid file not found, we'd like to get a pseudo
2949 	 * random number to use at the hostid.  A nice way to do this
2950 	 * is to read the real time clock.  To remain xen-compatible,
2951 	 * we can't poke the real hardware, so we use tsc_read() to
2952 	 * read the real time clock.  However, there is an ominous
2953 	 * warning in tsc_read that says it can return zero, so we
2954 	 * deal with that possibility by falling back to using the
2955 	 * (hopefully random enough) value in tenmicrodata.
2956 	 */
2957 
2958 	if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) {
2959 		/*
2960 		 * hostid file not found - try to load sysinit module
2961 		 * and see if it has a nonzero hostid value...use that
2962 		 * instead of generating a new hostid here if so.
2963 		 */
2964 		if ((i = modload("misc", "sysinit")) != -1) {
2965 			if (strlen(hw_serial) > 0)
2966 				hostid = (int32_t)atoi(hw_serial);
2967 			(void) modunload(i);
2968 		}
2969 
2970 		/*
2971 		 * We try to use the SMBIOS UUID. But not if it is blacklisted
2972 		 * in /etc/system.
2973 		 */
2974 		if ((hostid == HW_INVALID_HOSTID) &&
2975 		    (smbios_broken_uuid == 0) &&
2976 		    (ksmbios != NULL) &&
2977 		    (smbios_info_system(ksmbios, &smsys) != SMB_ERR) &&
2978 		    (smsys.smbs_uuidlen >= 16)) {
2979 			hostid = uuid_to_hostid(smsys.smbs_uuid);
2980 		}
2981 
2982 		/*
2983 		 * Generate a "random" hostid using the clock.  These
2984 		 * hostids will change on each boot if the value is not
2985 		 * saved to a persistent /etc/hostid file.
2986 		 */
2987 		if (hostid == HW_INVALID_HOSTID) {
2988 			tsc = tsc_read();
2989 			if (tsc == 0)	/* tsc_read can return zero sometimes */
2990 				hostid = (int32_t)tenmicrodata & 0x0CFFFFF;
2991 			else
2992 				hostid = (int32_t)tsc & 0x0CFFFFF;
2993 		}
2994 	} else {
2995 		/* hostid file found */
2996 		while (!done) {
2997 			token = kobj_lex(file, tokbuf, sizeof (tokbuf));
2998 
2999 			switch (token) {
3000 			case POUND:
3001 				/*
3002 				 * skip comments
3003 				 */
3004 				kobj_find_eol(file);
3005 				break;
3006 			case STRING:
3007 				/*
3008 				 * un-rot47 - obviously this
3009 				 * nonsense is ascii-specific
3010 				 */
3011 				for (c = (unsigned char *)tokbuf;
3012 				    *c != '\0'; c++) {
3013 					*c += 47;
3014 					if (*c > '~')
3015 						*c -= 94;
3016 					else if (*c < '!')
3017 						*c += 94;
3018 				}
3019 				/*
3020 				 * now we should have a real number
3021 				 */
3022 
3023 				if (kobj_getvalue(tokbuf, &tmp) != 0)
3024 					kobj_file_err(CE_WARN, file,
3025 					    "Bad value %s for hostid",
3026 					    tokbuf);
3027 				else
3028 					hostid = (int32_t)tmp;
3029 
3030 				break;
3031 			case EOF:
3032 				done = 1;
3033 				/* FALLTHROUGH */
3034 			case NEWLINE:
3035 				kobj_newline(file);
3036 				break;
3037 			default:
3038 				break;
3039 
3040 			}
3041 		}
3042 		if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */
3043 			kobj_file_err(CE_WARN, file,
3044 			    "hostid missing or corrupt");
3045 
3046 		kobj_close_file(file);
3047 	}
3048 	/*
3049 	 * hostid is now the value read from /etc/hostid, or the
3050 	 * new hostid we generated in this routine or HW_INVALID_HOSTID if not
3051 	 * set.
3052 	 */
3053 	return (hostid);
3054 }
3055 
3056 static int
3057 atoi(char *p)
3058 {
3059 	int i = 0;
3060 
3061 	while (*p != '\0')
3062 		i = 10 * i + (*p++ - '0');
3063 
3064 	return (i);
3065 }
3066 
3067 #endif /* _SOFT_HOSTID */
3068 
3069 void
3070 get_system_configuration(void)
3071 {
3072 	char	prop[32];
3073 	u_longlong_t nodes_ll, cpus_pernode_ll, lvalue;
3074 
3075 	if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) ||
3076 	    BOP_GETPROP(bootops, "nodes", prop) < 0 ||
3077 	    kobj_getvalue(prop, &nodes_ll) == -1 ||
3078 	    nodes_ll > MAXNODES ||
3079 	    BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) ||
3080 	    BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 ||
3081 	    kobj_getvalue(prop, &cpus_pernode_ll) == -1) {
3082 		system_hardware.hd_nodes = 1;
3083 		system_hardware.hd_cpus_per_node = 0;
3084 	} else {
3085 		system_hardware.hd_nodes = (int)nodes_ll;
3086 		system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll;
3087 	}
3088 
3089 	if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) ||
3090 	    BOP_GETPROP(bootops, "kernelbase", prop) < 0 ||
3091 	    kobj_getvalue(prop, &lvalue) == -1)
3092 		eprom_kernelbase = 0;
3093 	else
3094 		eprom_kernelbase = (uintptr_t)lvalue;
3095 
3096 	if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) ||
3097 	    BOP_GETPROP(bootops, "segmapsize", prop) < 0 ||
3098 	    kobj_getvalue(prop, &lvalue) == -1)
3099 		segmapsize = SEGMAPDEFAULT;
3100 	else
3101 		segmapsize = (uintptr_t)lvalue;
3102 
3103 	if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) ||
3104 	    BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 ||
3105 	    kobj_getvalue(prop, &lvalue) == -1)
3106 		segmapfreelists = 0;	/* use segmap driver default */
3107 	else
3108 		segmapfreelists = (int)lvalue;
3109 
3110 	/* physmem used to be here, but moved much earlier to fakebop.c */
3111 }
3112 
3113 /*
3114  * Add to a memory list.
3115  * start = start of new memory segment
3116  * len = length of new memory segment in bytes
3117  * new = pointer to a new struct memlist
3118  * memlistp = memory list to which to add segment.
3119  */
3120 void
3121 memlist_add(
3122 	uint64_t start,
3123 	uint64_t len,
3124 	struct memlist *new,
3125 	struct memlist **memlistp)
3126 {
3127 	struct memlist *cur;
3128 	uint64_t end = start + len;
3129 
3130 	new->ml_address = start;
3131 	new->ml_size = len;
3132 
3133 	cur = *memlistp;
3134 
3135 	while (cur) {
3136 		if (cur->ml_address >= end) {
3137 			new->ml_next = cur;
3138 			*memlistp = new;
3139 			new->ml_prev = cur->ml_prev;
3140 			cur->ml_prev = new;
3141 			return;
3142 		}
3143 		ASSERT(cur->ml_address + cur->ml_size <= start);
3144 		if (cur->ml_next == NULL) {
3145 			cur->ml_next = new;
3146 			new->ml_prev = cur;
3147 			new->ml_next = NULL;
3148 			return;
3149 		}
3150 		memlistp = &cur->ml_next;
3151 		cur = cur->ml_next;
3152 	}
3153 }
3154 
3155 void
3156 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3157 {
3158 	size_t tsize = e_modtext - modtext;
3159 	size_t dsize = e_moddata - moddata;
3160 
3161 	*text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize,
3162 	    1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP);
3163 	*data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize,
3164 	    1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3165 }
3166 
3167 caddr_t
3168 kobj_text_alloc(vmem_t *arena, size_t size)
3169 {
3170 	return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT));
3171 }
3172 
3173 /*ARGSUSED*/
3174 caddr_t
3175 kobj_texthole_alloc(caddr_t addr, size_t size)
3176 {
3177 	panic("unexpected call to kobj_texthole_alloc()");
3178 	/*NOTREACHED*/
3179 	return (0);
3180 }
3181 
3182 /*ARGSUSED*/
3183 void
3184 kobj_texthole_free(caddr_t addr, size_t size)
3185 {
3186 	panic("unexpected call to kobj_texthole_free()");
3187 }
3188 
3189 /*
3190  * This is called just after configure() in startup().
3191  *
3192  * The ISALIST concept is a bit hopeless on Intel, because
3193  * there's no guarantee of an ever-more-capable processor
3194  * given that various parts of the instruction set may appear
3195  * and disappear between different implementations.
3196  *
3197  * While it would be possible to correct it and even enhance
3198  * it somewhat, the explicit hardware capability bitmask allows
3199  * more flexibility.
3200  *
3201  * So, we just leave this alone.
3202  */
3203 void
3204 setx86isalist(void)
3205 {
3206 	char *tp;
3207 	size_t len;
3208 	extern char *isa_list;
3209 
3210 #define	TBUFSIZE	1024
3211 
3212 	tp = kmem_alloc(TBUFSIZE, KM_SLEEP);
3213 	*tp = '\0';
3214 
3215 #if defined(__amd64)
3216 	(void) strcpy(tp, "amd64 ");
3217 #endif
3218 
3219 	switch (x86_vendor) {
3220 	case X86_VENDOR_Intel:
3221 	case X86_VENDOR_AMD:
3222 	case X86_VENDOR_TM:
3223 		if (is_x86_feature(x86_featureset, X86FSET_CMOV)) {
3224 			/*
3225 			 * Pentium Pro or later
3226 			 */
3227 			(void) strcat(tp, "pentium_pro");
3228 			(void) strcat(tp,
3229 			    is_x86_feature(x86_featureset, X86FSET_MMX) ?
3230 			    "+mmx pentium_pro " : " ");
3231 		}
3232 		/*FALLTHROUGH*/
3233 	case X86_VENDOR_Cyrix:
3234 		/*
3235 		 * The Cyrix 6x86 does not have any Pentium features
3236 		 * accessible while not at privilege level 0.
3237 		 */
3238 		if (is_x86_feature(x86_featureset, X86FSET_CPUID)) {
3239 			(void) strcat(tp, "pentium");
3240 			(void) strcat(tp,
3241 			    is_x86_feature(x86_featureset, X86FSET_MMX) ?
3242 			    "+mmx pentium " : " ");
3243 		}
3244 		break;
3245 	default:
3246 		break;
3247 	}
3248 	(void) strcat(tp, "i486 i386 i86");
3249 	len = strlen(tp) + 1;   /* account for NULL at end of string */
3250 	isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp);
3251 	kmem_free(tp, TBUFSIZE);
3252 
3253 #undef TBUFSIZE
3254 }
3255 
3256 
3257 #ifdef __amd64
3258 
3259 void *
3260 device_arena_alloc(size_t size, int vm_flag)
3261 {
3262 	return (vmem_alloc(device_arena, size, vm_flag));
3263 }
3264 
3265 void
3266 device_arena_free(void *vaddr, size_t size)
3267 {
3268 	vmem_free(device_arena, vaddr, size);
3269 }
3270 
3271 #else /* __i386 */
3272 
3273 void *
3274 device_arena_alloc(size_t size, int vm_flag)
3275 {
3276 	caddr_t	vaddr;
3277 	uintptr_t v;
3278 	size_t	start;
3279 	size_t	end;
3280 
3281 	vaddr = vmem_alloc(heap_arena, size, vm_flag);
3282 	if (vaddr == NULL)
3283 		return (NULL);
3284 
3285 	v = (uintptr_t)vaddr;
3286 	ASSERT(v >= kernelbase);
3287 	ASSERT(v + size <= valloc_base);
3288 
3289 	start = btop(v - kernelbase);
3290 	end = btop(v + size - 1 - kernelbase);
3291 	ASSERT(start < toxic_bit_map_len);
3292 	ASSERT(end < toxic_bit_map_len);
3293 
3294 	while (start <= end) {
3295 		BT_ATOMIC_SET(toxic_bit_map, start);
3296 		++start;
3297 	}
3298 	return (vaddr);
3299 }
3300 
3301 void
3302 device_arena_free(void *vaddr, size_t size)
3303 {
3304 	uintptr_t v = (uintptr_t)vaddr;
3305 	size_t	start;
3306 	size_t	end;
3307 
3308 	ASSERT(v >= kernelbase);
3309 	ASSERT(v + size <= valloc_base);
3310 
3311 	start = btop(v - kernelbase);
3312 	end = btop(v + size - 1 - kernelbase);
3313 	ASSERT(start < toxic_bit_map_len);
3314 	ASSERT(end < toxic_bit_map_len);
3315 
3316 	while (start <= end) {
3317 		ASSERT(BT_TEST(toxic_bit_map, start) != 0);
3318 		BT_ATOMIC_CLEAR(toxic_bit_map, start);
3319 		++start;
3320 	}
3321 	vmem_free(heap_arena, vaddr, size);
3322 }
3323 
3324 /*
3325  * returns 1st address in range that is in device arena, or NULL
3326  * if len is not NULL it returns the length of the toxic range
3327  */
3328 void *
3329 device_arena_contains(void *vaddr, size_t size, size_t *len)
3330 {
3331 	uintptr_t v = (uintptr_t)vaddr;
3332 	uintptr_t eaddr = v + size;
3333 	size_t start;
3334 	size_t end;
3335 
3336 	/*
3337 	 * if called very early by kmdb, just return NULL
3338 	 */
3339 	if (toxic_bit_map == NULL)
3340 		return (NULL);
3341 
3342 	/*
3343 	 * First check if we're completely outside the bitmap range.
3344 	 */
3345 	if (v >= valloc_base || eaddr < kernelbase)
3346 		return (NULL);
3347 
3348 	/*
3349 	 * Trim ends of search to look at only what the bitmap covers.
3350 	 */
3351 	if (v < kernelbase)
3352 		v = kernelbase;
3353 	start = btop(v - kernelbase);
3354 	end = btop(eaddr - kernelbase);
3355 	if (end >= toxic_bit_map_len)
3356 		end = toxic_bit_map_len;
3357 
3358 	if (bt_range(toxic_bit_map, &start, &end, end) == 0)
3359 		return (NULL);
3360 
3361 	v = kernelbase + ptob(start);
3362 	if (len != NULL)
3363 		*len = ptob(end - start);
3364 	return ((void *)v);
3365 }
3366 
3367 #endif	/* __i386 */
3368