xref: /illumos-gate/usr/src/uts/sun4/os/startup.c (revision c97ad5cdc75eb73e3cc38542ca3ba783574b0a7a)
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 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/machsystm.h>
30 #include <sys/archsystm.h>
31 #include <sys/vm.h>
32 #include <sys/cpu.h>
33 #include <sys/atomic.h>
34 #include <sys/reboot.h>
35 #include <sys/kdi.h>
36 #include <sys/bootconf.h>
37 #include <sys/memlist_plat.h>
38 #include <sys/memlist_impl.h>
39 #include <sys/prom_plat.h>
40 #include <sys/prom_isa.h>
41 #include <sys/autoconf.h>
42 #include <sys/intreg.h>
43 #include <sys/ivintr.h>
44 #include <sys/fpu/fpusystm.h>
45 #include <sys/iommutsb.h>
46 #include <vm/vm_dep.h>
47 #include <vm/seg_dev.h>
48 #include <vm/seg_kmem.h>
49 #include <vm/seg_kpm.h>
50 #include <vm/seg_map.h>
51 #include <vm/seg_kp.h>
52 #include <sys/sysconf.h>
53 #include <vm/hat_sfmmu.h>
54 #include <sys/kobj.h>
55 #include <sys/sun4asi.h>
56 #include <sys/clconf.h>
57 #include <sys/platform_module.h>
58 #include <sys/panic.h>
59 #include <sys/cpu_sgnblk_defs.h>
60 #include <sys/clock.h>
61 #include <sys/cmn_err.h>
62 #include <sys/promif.h>
63 #include <sys/prom_debug.h>
64 #include <sys/traptrace.h>
65 #include <sys/memnode.h>
66 #include <sys/mem_cage.h>
67 #include <sys/mmu.h>
68 
69 extern void setup_trap_table(void);
70 extern void cpu_intrq_setup(struct cpu *);
71 extern void cpu_intrq_register(struct cpu *);
72 extern void contig_mem_init(void);
73 extern void mach_dump_buffer_init(void);
74 extern void mach_descrip_init(void);
75 extern void mach_descrip_startup_fini(void);
76 extern void mach_memscrub(void);
77 extern void mach_fpras(void);
78 extern void mach_cpu_halt_idle(void);
79 extern void mach_hw_copy_limit(void);
80 extern void load_mach_drivers(void);
81 extern void load_tod_module(void);
82 #pragma weak load_tod_module
83 
84 extern int ndata_alloc_mmfsa(struct memlist *ndata);
85 #pragma weak ndata_alloc_mmfsa
86 
87 extern void cif_init(void);
88 #pragma weak cif_init
89 
90 extern void parse_idprom(void);
91 extern void add_vx_handler(char *, int, void (*)(cell_t *));
92 extern void mem_config_init(void);
93 extern void memseg_remap_init(void);
94 
95 extern void mach_kpm_init(void);
96 
97 /*
98  * External Data:
99  */
100 extern int vac_size;	/* cache size in bytes */
101 extern uint_t vac_mask;	/* VAC alignment consistency mask */
102 extern uint_t vac_colors;
103 
104 /*
105  * Global Data Definitions:
106  */
107 
108 /*
109  * XXX - Don't port this to new architectures
110  * A 3rd party volume manager driver (vxdm) depends on the symbol romp.
111  * 'romp' has no use with a prom with an IEEE 1275 client interface.
112  * The driver doesn't use the value, but it depends on the symbol.
113  */
114 void *romp;		/* veritas driver won't load without romp 4154976 */
115 /*
116  * Declare these as initialized data so we can patch them.
117  */
118 pgcnt_t physmem = 0;	/* memory size in pages, patch if you want less */
119 pgcnt_t segkpsize =
120     btop(SEGKPDEFSIZE);	/* size of segkp segment in pages */
121 uint_t segmap_percent = 12; /* Size of segmap segment */
122 
123 int use_cache = 1;		/* cache not reliable (605 bugs) with MP */
124 int vac_copyback = 1;
125 char *cache_mode = NULL;
126 int use_mix = 1;
127 int prom_debug = 0;
128 
129 struct bootops *bootops = 0;	/* passed in from boot in %o2 */
130 caddr_t boot_tba;		/* %tba at boot - used by kmdb */
131 uint_t	tba_taken_over = 0;
132 
133 caddr_t s_text;			/* start of kernel text segment */
134 caddr_t e_text;			/* end of kernel text segment */
135 caddr_t s_data;			/* start of kernel data segment */
136 caddr_t e_data;			/* end of kernel data segment */
137 
138 caddr_t modtext;		/* beginning of module text */
139 size_t	modtext_sz;		/* size of module text */
140 caddr_t moddata;		/* beginning of module data reserve */
141 caddr_t e_moddata;		/* end of module data reserve */
142 
143 /*
144  * End of first block of contiguous kernel in 32-bit virtual address space
145  */
146 caddr_t		econtig32;	/* end of first blk of contiguous kernel */
147 
148 caddr_t		ncbase;		/* beginning of non-cached segment */
149 caddr_t		ncend;		/* end of non-cached segment */
150 caddr_t		sdata;		/* beginning of data segment */
151 
152 caddr_t		extra_etva;	/* beginning of unused nucleus text */
153 pgcnt_t		extra_etpg;	/* number of pages of unused nucleus text */
154 
155 size_t	ndata_remain_sz;	/* bytes from end of data to 4MB boundary */
156 caddr_t	nalloc_base;		/* beginning of nucleus allocation */
157 caddr_t nalloc_end;		/* end of nucleus allocatable memory */
158 caddr_t valloc_base;		/* beginning of kvalloc segment	*/
159 
160 caddr_t kmem64_base;		/* base of kernel mem segment in 64-bit space */
161 caddr_t kmem64_end;		/* end of kernel mem segment in 64-bit space */
162 caddr_t kmem64_aligned_end;	/* end of large page, overmaps 64-bit space */
163 int	kmem64_alignsize;	/* page size for mem segment in 64-bit space */
164 int	kmem64_szc;		/* page size code */
165 uint64_t kmem64_pabase = (uint64_t)-1;	/* physical address of kmem64_base */
166 
167 uintptr_t shm_alignment;	/* VAC address consistency modulus */
168 struct memlist *phys_install;	/* Total installed physical memory */
169 struct memlist *phys_avail;	/* Available (unreserved) physical memory */
170 struct memlist *virt_avail;	/* Available (unmapped?) virtual memory */
171 struct memlist ndata;		/* memlist of nucleus allocatable memory */
172 int memexp_flag;		/* memory expansion card flag */
173 uint64_t ecache_flushaddr;	/* physical address used for flushing E$ */
174 pgcnt_t obp_pages;		/* Physical pages used by OBP */
175 
176 /*
177  * VM data structures
178  */
179 long page_hashsz;		/* Size of page hash table (power of two) */
180 struct page *pp_base;		/* Base of system page struct array */
181 size_t pp_sz;			/* Size in bytes of page struct array */
182 struct page **page_hash;	/* Page hash table */
183 struct seg ktextseg;		/* Segment used for kernel executable image */
184 struct seg kvalloc;		/* Segment used for "valloc" mapping */
185 struct seg kpseg;		/* Segment used for pageable kernel virt mem */
186 struct seg ktexthole;		/* Segment used for nucleus text hole */
187 struct seg kmapseg;		/* Segment used for generic kernel mappings */
188 struct seg kpmseg;		/* Segment used for physical mapping */
189 struct seg kdebugseg;		/* Segment used for the kernel debugger */
190 
191 uintptr_t kpm_pp_base;		/* Base of system kpm_page array */
192 size_t	kpm_pp_sz;		/* Size of system kpm_page array */
193 pgcnt_t	kpm_npages;		/* How many kpm pages are managed */
194 
195 struct seg *segkp = &kpseg;	/* Pageable kernel virtual memory segment */
196 struct seg *segkmap = &kmapseg;	/* Kernel generic mapping segment */
197 struct seg *segkpm = &kpmseg;	/* 64bit kernel physical mapping segment */
198 
199 int segzio_fromheap = 0;	/* zio allocations occur from heap */
200 caddr_t segzio_base;		/* Base address of segzio */
201 pgcnt_t segziosize = 0;		/* size of zio segment in pages */
202 
203 /*
204  * debugger pages (if allocated)
205  */
206 struct vnode kdebugvp;
207 
208 /*
209  * VA range available to the debugger
210  */
211 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
212 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
213 
214 /*
215  * Segment for relocated kernel structures in 64-bit large RAM kernels
216  */
217 struct seg kmem64;
218 
219 struct memseg *memseg_base;
220 size_t memseg_sz;		/* Used to translate a va to page */
221 struct vnode unused_pages_vp;
222 
223 /*
224  * VM data structures allocated early during boot.
225  */
226 size_t pagehash_sz;
227 uint64_t memlist_sz;
228 
229 char tbr_wr_addr_inited = 0;
230 
231 
232 /*
233  * Static Routines:
234  */
235 static void memlist_add(uint64_t, uint64_t, struct memlist **,
236 	struct memlist **);
237 static void kphysm_init(page_t *, struct memseg *, pgcnt_t, uintptr_t,
238 	pgcnt_t);
239 static void kvm_init(void);
240 
241 static void startup_init(void);
242 static void startup_memlist(void);
243 static void startup_modules(void);
244 static void startup_bop_gone(void);
245 static void startup_vm(void);
246 static void startup_end(void);
247 static void setup_cage_params(void);
248 static void startup_create_io_node(void);
249 
250 static pgcnt_t npages;
251 static struct memlist *memlist;
252 void *memlist_end;
253 
254 static pgcnt_t bop_alloc_pages;
255 static caddr_t hblk_base;
256 uint_t hblk_alloc_dynamic = 0;
257 uint_t hblk1_min = H1MIN;
258 
259 
260 /*
261  * Hooks for unsupported platforms and down-rev firmware
262  */
263 int iam_positron(void);
264 #pragma weak iam_positron
265 static void do_prom_version_check(void);
266 static void kpm_init(void);
267 static void kpm_npages_setup(int);
268 static void kpm_memseg_init(void);
269 
270 /*
271  * After receiving a thermal interrupt, this is the number of seconds
272  * to delay before shutting off the system, assuming
273  * shutdown fails.  Use /etc/system to change the delay if this isn't
274  * large enough.
275  */
276 int thermal_powerdown_delay = 1200;
277 
278 /*
279  * Used to hold off page relocations into the cage until OBP has completed
280  * its boot-time handoff of its resources to the kernel.
281  */
282 int page_relocate_ready = 0;
283 
284 /*
285  * Enable some debugging messages concerning memory usage...
286  */
287 #ifdef  DEBUGGING_MEM
288 static int debugging_mem;
289 static void
290 printmemlist(char *title, struct memlist *list)
291 {
292 	if (!debugging_mem)
293 		return;
294 
295 	printf("%s\n", title);
296 
297 	while (list) {
298 		prom_printf("\taddr = 0x%x %8x, size = 0x%x %8x\n",
299 		    (uint32_t)(list->address >> 32), (uint32_t)list->address,
300 		    (uint32_t)(list->size >> 32), (uint32_t)(list->size));
301 		list = list->next;
302 	}
303 }
304 
305 void
306 printmemseg(struct memseg *memseg)
307 {
308 	if (!debugging_mem)
309 		return;
310 
311 	printf("memseg\n");
312 
313 	while (memseg) {
314 		prom_printf("\tpage = 0x%p, epage = 0x%p, "
315 		    "pfn = 0x%x, epfn = 0x%x\n",
316 		    memseg->pages, memseg->epages,
317 		    memseg->pages_base, memseg->pages_end);
318 		memseg = memseg->next;
319 	}
320 }
321 
322 #define	debug_pause(str)	halt((str))
323 #define	MPRINTF(str)		if (debugging_mem) prom_printf((str))
324 #define	MPRINTF1(str, a)	if (debugging_mem) prom_printf((str), (a))
325 #define	MPRINTF2(str, a, b)	if (debugging_mem) prom_printf((str), (a), (b))
326 #define	MPRINTF3(str, a, b, c) \
327 	if (debugging_mem) prom_printf((str), (a), (b), (c))
328 #else	/* DEBUGGING_MEM */
329 #define	MPRINTF(str)
330 #define	MPRINTF1(str, a)
331 #define	MPRINTF2(str, a, b)
332 #define	MPRINTF3(str, a, b, c)
333 #endif	/* DEBUGGING_MEM */
334 
335 /* Simple message to indicate that the bootops pointer has been zeroed */
336 #ifdef DEBUG
337 static int bootops_gone_on = 0;
338 #define	BOOTOPS_GONE() \
339 	if (bootops_gone_on) \
340 		prom_printf("The bootops vec is zeroed now!\n");
341 #else
342 #define	BOOTOPS_GONE()
343 #endif /* DEBUG */
344 
345 /*
346  * Monitor pages may not be where this says they are.
347  * and the debugger may not be there either.
348  *
349  * Note that 'pages' here are *physical* pages, which are 8k on sun4u.
350  *
351  *                        Physical memory layout
352  *                     (not necessarily contiguous)
353  *                       (THIS IS SOMEWHAT WRONG)
354  *                       /-----------------------\
355  *                       |       monitor pages   |
356  *             availmem -|-----------------------|
357  *                       |                       |
358  *                       |       page pool       |
359  *                       |                       |
360  *                       |-----------------------|
361  *                       |   configured tables   |
362  *                       |       buffers         |
363  *            firstaddr -|-----------------------|
364  *                       |   hat data structures |
365  *                       |-----------------------|
366  *                       |    kernel data, bss   |
367  *                       |-----------------------|
368  *                       |    interrupt stack    |
369  *                       |-----------------------|
370  *                       |    kernel text (RO)   |
371  *                       |-----------------------|
372  *                       |    trap table (4k)    |
373  *                       |-----------------------|
374  *               page 1  |      panicbuf         |
375  *                       |-----------------------|
376  *               page 0  |       reclaimed       |
377  *                       |_______________________|
378  *
379  *
380  *
381  *                    Kernel's Virtual Memory Layout.
382  *                       /-----------------------\
383  * 0xFFFFFFFF.FFFFFFFF  -|                       |-
384  *                       |   OBP's virtual page  |
385  *                       |        tables         |
386  * 0xFFFFFFFC.00000000  -|-----------------------|-
387  *                       :                       :
388  *                       :                       :
389  *                      -|-----------------------|-
390  *                       |       segzio          | (base and size vary)
391  * 0xFFFFFE00.00000000  -|-----------------------|-
392  *                       |                       |  Ultrasparc I/II support
393  *                       |    segkpm segment     |  up to 2TB of physical
394  *                       | (64-bit kernel ONLY)  |  memory, VAC has 2 colors
395  *                       |                       |
396  * 0xFFFFFA00.00000000  -|-----------------------|- 2TB segkpm alignment
397  *                       :                       :
398  *                       :                       :
399  * 0xFFFFF810.00000000  -|-----------------------|- hole_end
400  *                       |                       |      ^
401  *                       |  UltraSPARC I/II call |      |
402  *                       | bug requires an extra |      |
403  *                       | 4 GB of space between |      |
404  *                       |   hole and used RAM   |	|
405  *                       |                       |      |
406  * 0xFFFFF800.00000000  -|-----------------------|-     |
407  *                       |                       |      |
408  *                       | Virtual Address Hole  |   UltraSPARC
409  *                       |  on UltraSPARC I/II   |  I/II * ONLY *
410  *                       |                       |      |
411  * 0x00000800.00000000  -|-----------------------|-     |
412  *                       |                       |      |
413  *                       |  UltraSPARC I/II call |      |
414  *                       | bug requires an extra |      |
415  *                       | 4 GB of space between |      |
416  *                       |   hole and used RAM   |      |
417  *                       |                       |      v
418  * 0x000007FF.00000000  -|-----------------------|- hole_start -----
419  *                       :                       :		   ^
420  *                       :                       :		   |
421  * 0x00000XXX.XXX00000  -|-----------------------|- kmem64_	   |
422  *                       | overmapped area       |   alignend_end  |
423  *                       | (kmem64_alignsize     |		   |
424  *                       |  boundary)            |		   |
425  * 0x00000XXX.XXXXXXXX  -|-----------------------|- kmem64_end	   |
426  *                       |                       |		   |
427  *                       |   64-bit kernel ONLY  |		   |
428  *                       |                       |		   |
429  *                       |    kmem64 segment     |		   |
430  *                       |                       |		   |
431  *                       | (Relocated extra HME  |	     Approximately
432  *                       |   block allocations,  |	    1 TB of virtual
433  *                       |   memnode freelists,  |	     address space
434  *                       |    HME hash buckets,  |		   |
435  *                       | mml_table, kpmp_table,|		   |
436  *                       |  page_t array and     |		   |
437  *                       |  hashblock pool to    |		   |
438  *                       |   avoid hard-coded    |		   |
439  *                       |     32-bit vaddr      |		   |
440  *                       |     limitations)      |		   |
441  *                       |                       |		   v
442  * 0x00000700.00000000  -|-----------------------|- SYSLIMIT (kmem64_base)
443  *                       |                       |
444  *                       |  segkmem segment      | (SYSLIMIT - SYSBASE = 4TB)
445  *                       |                       |
446  * 0x00000300.00000000  -|-----------------------|- SYSBASE
447  *                       :                       :
448  *                       :                       :
449  *                      -|-----------------------|-
450  *                       |                       |
451  *                       |  segmap segment       |   SEGMAPSIZE (1/8th physmem,
452  *                       |                       |               256G MAX)
453  * 0x000002a7.50000000  -|-----------------------|- SEGMAPBASE
454  *                       :                       :
455  *                       :                       :
456  *                      -|-----------------------|-
457  *                       |                       |
458  *                       |       segkp           |    SEGKPSIZE (2GB)
459  *                       |                       |
460  *                       |                       |
461  * 0x000002a1.00000000  -|-----------------------|- SEGKPBASE
462  *                       |                       |
463  * 0x000002a0.00000000  -|-----------------------|- MEMSCRUBBASE
464  *                       |                       |       (SEGKPBASE - 0x400000)
465  * 0x0000029F.FFE00000  -|-----------------------|- ARGSBASE
466  *                       |                       |       (MEMSCRUBBASE - NCARGS)
467  * 0x0000029F.FFD80000  -|-----------------------|- PPMAPBASE
468  *                       |                       |       (ARGSBASE - PPMAPSIZE)
469  * 0x0000029F.FFD00000  -|-----------------------|- PPMAP_FAST_BASE
470  *                       |                       |
471  * 0x0000029F.FF980000  -|-----------------------|- PIOMAPBASE
472  *                       |                       |
473  * 0x0000029F.FF580000  -|-----------------------|- NARG_BASE
474  *                       :                       :
475  *                       :                       :
476  * 0x00000000.FFFFFFFF  -|-----------------------|- OFW_END_ADDR
477  *                       |                       |
478  *                       |         OBP           |
479  *                       |                       |
480  * 0x00000000.F0000000  -|-----------------------|- OFW_START_ADDR
481  *                       |         kmdb          |
482  * 0x00000000.EDD00000  -|-----------------------|- SEGDEBUGBASE
483  *                       :                       :
484  *                       :                       :
485  * 0x00000000.7c000000  -|-----------------------|- SYSLIMIT32
486  *                       |                       |
487  *                       |  segkmem32 segment    | (SYSLIMIT32 - SYSBASE32 =
488  *                       |                       |    ~64MB)
489  * 0x00000000.78002000  -|-----------------------|
490  *                       |     panicbuf          |
491  * 0x00000000.78000000  -|-----------------------|- SYSBASE32
492  *                       :                       :
493  *                       :                       :
494  *                       |                       |
495  *                       |-----------------------|- econtig32
496  *                       |    vm structures      |
497  * 0x00000000.01C00000   |-----------------------|- nalloc_end
498  *                       |         TSBs          |
499  *                       |-----------------------|- end/nalloc_base
500  *                       |   kernel data & bss   |
501  * 0x00000000.01800000  -|-----------------------|
502  *                       :   nucleus text hole   :
503  * 0x00000000.01400000  -|-----------------------|
504  *                       :                       :
505  *                       |-----------------------|
506  *                       |      module text      |
507  *                       |-----------------------|- e_text/modtext
508  *                       |      kernel text      |
509  *                       |-----------------------|
510  *                       |    trap table (48k)   |
511  * 0x00000000.01000000  -|-----------------------|- KERNELBASE
512  *                       | reserved for trapstat |} TSTAT_TOTAL_SIZE
513  *                       |-----------------------|
514  *                       |                       |
515  *                       |        invalid        |
516  *                       |                       |
517  * 0x00000000.00000000  _|_______________________|
518  *
519  *
520  *
521  *                   32-bit User Virtual Memory Layout.
522  *                       /-----------------------\
523  *                       |                       |
524  *                       |        invalid        |
525  *                       |                       |
526  *          0xFFC00000  -|-----------------------|- USERLIMIT
527  *                       |       user stack      |
528  *                       :                       :
529  *                       :                       :
530  *                       :                       :
531  *                       |       user data       |
532  *                      -|-----------------------|-
533  *                       |       user text       |
534  *          0x00002000  -|-----------------------|-
535  *                       |       invalid         |
536  *          0x00000000  _|_______________________|
537  *
538  *
539  *
540  *                   64-bit User Virtual Memory Layout.
541  *                       /-----------------------\
542  *                       |                       |
543  *                       |        invalid        |
544  *                       |                       |
545  *  0xFFFFFFFF.80000000 -|-----------------------|- USERLIMIT
546  *                       |       user stack      |
547  *                       :                       :
548  *                       :                       :
549  *                       :                       :
550  *                       |       user data       |
551  *                      -|-----------------------|-
552  *                       |       user text       |
553  *  0x00000000.00100000 -|-----------------------|-
554  *                       |       invalid         |
555  *  0x00000000.00000000 _|_______________________|
556  */
557 
558 extern caddr_t ecache_init_scrub_flush_area(caddr_t alloc_base);
559 extern uint64_t ecache_flush_address(void);
560 
561 #pragma weak load_platform_modules
562 #pragma weak plat_startup_memlist
563 #pragma weak ecache_init_scrub_flush_area
564 #pragma weak ecache_flush_address
565 
566 
567 /*
568  * By default the DR Cage is enabled for maximum OS
569  * MPSS performance.  Users needing to disable the cage mechanism
570  * can set this variable to zero via /etc/system.
571  * Disabling the cage on systems supporting Dynamic Reconfiguration (DR)
572  * will result in loss of DR functionality.
573  * Platforms wishing to disable kernel Cage by default
574  * should do so in their set_platform_defaults() routine.
575  */
576 int	kernel_cage_enable = 1;
577 
578 static void
579 setup_cage_params(void)
580 {
581 	void (*func)(void);
582 
583 	func = (void (*)(void))kobj_getsymvalue("set_platform_cage_params", 0);
584 	if (func != NULL) {
585 		(*func)();
586 		return;
587 	}
588 
589 	if (kernel_cage_enable == 0) {
590 		return;
591 	}
592 	kcage_range_lock();
593 	if (kcage_range_init(phys_avail, 1) == 0) {
594 		kcage_init(total_pages / 256);
595 	}
596 	kcage_range_unlock();
597 
598 	if (kcage_on) {
599 		cmn_err(CE_NOTE, "!Kernel Cage is ENABLED");
600 	} else {
601 		cmn_err(CE_NOTE, "!Kernel Cage is DISABLED");
602 	}
603 
604 }
605 
606 /*
607  * Machine-dependent startup code
608  */
609 void
610 startup(void)
611 {
612 	startup_init();
613 	if (&startup_platform)
614 		startup_platform();
615 	startup_memlist();
616 	startup_modules();
617 	setup_cage_params();
618 	startup_bop_gone();
619 	startup_vm();
620 	startup_end();
621 }
622 
623 struct regs sync_reg_buf;
624 uint64_t sync_tt;
625 
626 void
627 sync_handler(void)
628 {
629 	struct  trap_info 	ti;
630 	int i;
631 
632 	/*
633 	 * Prevent trying to talk to the other CPUs since they are
634 	 * sitting in the prom and won't reply.
635 	 */
636 	for (i = 0; i < NCPU; i++) {
637 		if ((i != CPU->cpu_id) && CPU_XCALL_READY(i)) {
638 			cpu[i]->cpu_flags &= ~CPU_READY;
639 			cpu[i]->cpu_flags |= CPU_QUIESCED;
640 			CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
641 		}
642 	}
643 
644 	/*
645 	 * We've managed to get here without going through the
646 	 * normal panic code path. Try and save some useful
647 	 * information.
648 	 */
649 	if (!panicstr && (curthread->t_panic_trap == NULL)) {
650 		ti.trap_type = sync_tt;
651 		ti.trap_regs = &sync_reg_buf;
652 		ti.trap_addr = NULL;
653 		ti.trap_mmu_fsr = 0x0;
654 
655 		curthread->t_panic_trap = &ti;
656 	}
657 
658 	/*
659 	 * If we're re-entering the panic path, update the signature
660 	 * block so that the SC knows we're in the second part of panic.
661 	 */
662 	if (panicstr)
663 		CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
664 
665 	nopanicdebug = 1; /* do not perform debug_enter() prior to dump */
666 	panic("sync initiated");
667 }
668 
669 
670 static void
671 startup_init(void)
672 {
673 	/*
674 	 * We want to save the registers while we're still in OBP
675 	 * so that we know they haven't been fiddled with since.
676 	 * (In principle, OBP can't change them just because it
677 	 * makes a callback, but we'd rather not depend on that
678 	 * behavior.)
679 	 */
680 	char		sync_str[] =
681 		"warning @ warning off : sync "
682 		"%%tl-c %%tstate h# %p x! "
683 		"%%g1 h# %p x! %%g2 h# %p x! %%g3 h# %p x! "
684 		"%%g4 h# %p x! %%g5 h# %p x! %%g6 h# %p x! "
685 		"%%g7 h# %p x! %%o0 h# %p x! %%o1 h# %p x! "
686 		"%%o2 h# %p x! %%o3 h# %p x! %%o4 h# %p x! "
687 		"%%o5 h# %p x! %%o6 h# %p x! %%o7 h# %p x! "
688 		"%%tl-c %%tpc h# %p x! %%tl-c %%tnpc h# %p x! "
689 		"%%y h# %p l! %%tl-c %%tt h# %p x! "
690 		"sync ; warning !";
691 
692 	/*
693 	 * 20 == num of %p substrings
694 	 * 16 == max num of chars %p will expand to.
695 	 */
696 	char 		bp[sizeof (sync_str) + 16 * 20];
697 
698 	(void) check_boot_version(BOP_GETVERSION(bootops));
699 
700 	/*
701 	 * Initialize ptl1 stack for the 1st CPU.
702 	 */
703 	ptl1_init_cpu(&cpu0);
704 
705 	/*
706 	 * Initialize the address map for cache consistent mappings
707 	 * to random pages; must be done after vac_size is set.
708 	 */
709 	ppmapinit();
710 
711 	/*
712 	 * Initialize the PROM callback handler.
713 	 */
714 	init_vx_handler();
715 
716 	/*
717 	 * have prom call sync_callback() to handle the sync and
718 	 * save some useful information which will be stored in the
719 	 * core file later.
720 	 */
721 	(void) sprintf((char *)bp, sync_str,
722 		(void *)&sync_reg_buf.r_tstate, (void *)&sync_reg_buf.r_g1,
723 		(void *)&sync_reg_buf.r_g2, (void *)&sync_reg_buf.r_g3,
724 		(void *)&sync_reg_buf.r_g4, (void *)&sync_reg_buf.r_g5,
725 		(void *)&sync_reg_buf.r_g6, (void *)&sync_reg_buf.r_g7,
726 		(void *)&sync_reg_buf.r_o0, (void *)&sync_reg_buf.r_o1,
727 		(void *)&sync_reg_buf.r_o2, (void *)&sync_reg_buf.r_o3,
728 		(void *)&sync_reg_buf.r_o4, (void *)&sync_reg_buf.r_o5,
729 		(void *)&sync_reg_buf.r_o6, (void *)&sync_reg_buf.r_o7,
730 		(void *)&sync_reg_buf.r_pc, (void *)&sync_reg_buf.r_npc,
731 		(void *)&sync_reg_buf.r_y, (void *)&sync_tt);
732 	prom_interpret(bp, 0, 0, 0, 0, 0);
733 	add_vx_handler("sync", 1, (void (*)(cell_t *))sync_handler);
734 }
735 
736 static u_longlong_t *boot_physinstalled, *boot_physavail, *boot_virtavail;
737 static size_t boot_physinstalled_len, boot_physavail_len, boot_virtavail_len;
738 
739 #define	IVSIZE	((MAXIVNUM * sizeof (intr_vec_t *)) + \
740 		(MAX_RSVD_IV * sizeof (intr_vec_t)) + \
741 		(MAX_RSVD_IVX * sizeof (intr_vecx_t)))
742 
743 #if !defined(C_OBP)
744 /*
745  * Install a temporary tte handler in OBP for kmem64 area.
746  *
747  * We map kmem64 area with large pages before the trap table is taken
748  * over. Since OBP makes 8K mappings, it can create 8K tlb entries in
749  * the same area. Duplicate tlb entries with different page sizes
750  * cause unpredicatble behavior.  To avoid this, we don't create
751  * kmem64 mappings via BOP_ALLOC (ends up as prom_alloc() call to
752  * OBP).  Instead, we manage translations with a temporary va>tte-data
753  * handler (kmem64-tte).  This handler is replaced by unix-tte when
754  * the trap table is taken over.
755  *
756  * The temporary handler knows the physical address of the kmem64
757  * area. It uses the prom's pgmap@ Forth word for other addresses.
758  *
759  * We have to use BOP_ALLOC() method for C-OBP platforms because
760  * pgmap@ is not defined in C-OBP. C-OBP is only used on serengeti
761  * sun4u platforms. On sun4u we flush tlb after trap table is taken
762  * over if we use large pages for kernel heap and kmem64. Since sun4u
763  * prom (unlike sun4v) calls va>tte-data first for client address
764  * translation prom's ttes for kmem64 can't get into TLB even if we
765  * later switch to prom's trap table again. C-OBP uses 4M pages for
766  * client mappings when possible so on all platforms we get the
767  * benefit from large mappings for kmem64 area immediately during
768  * boot.
769  *
770  * pseudo code:
771  * if (context != 0) {
772  * 	return false
773  * } else if (miss_va in range[kmem64_base, kmem64_end)) {
774  *	tte = tte_template +
775  *		(((miss_va & pagemask) - kmem64_base));
776  *	return tte, true
777  * } else {
778  *	return pgmap@ result
779  * }
780  */
781 char kmem64_obp_str[] =
782 	"h# %lx constant kmem64_base "
783 	"h# %lx constant kmem64_end "
784 	"h# %lx constant kmem64_pagemask "
785 	"h# %lx constant kmem64_template "
786 
787 	": kmem64-tte ( addr cnum -- false | tte-data true ) "
788 	"    if                                       ( addr ) "
789 	"       drop false exit then                  ( false ) "
790 	"    dup  kmem64_base kmem64_end  within  if  ( addr ) "
791 	"	kmem64_pagemask and                   ( addr' ) "
792 	"	kmem64_base -                         ( addr' ) "
793 	"	kmem64_template +                     ( tte ) "
794 	"	true                                  ( tte true ) "
795 	"    else                                     ( addr ) "
796 	"	pgmap@                                ( tte ) "
797 	"       dup 0< if true else drop false then   ( tte true  |  false ) "
798 	"    then                                     ( tte true  |  false ) "
799 	"; "
800 
801 	"' kmem64-tte is va>tte-data "
802 ;
803 
804 void
805 install_kmem64_tte()
806 {
807 	char b[sizeof (kmem64_obp_str) + (4 * 16)];
808 	tte_t tte;
809 
810 	PRM_DEBUG(kmem64_pabase);
811 	PRM_DEBUG(kmem64_szc);
812 	sfmmu_memtte(&tte, kmem64_pabase >> MMU_PAGESHIFT,
813 	    PROC_DATA | HAT_NOSYNC, kmem64_szc);
814 	PRM_DEBUG(tte.ll);
815 	(void) sprintf(b, kmem64_obp_str,
816 	    kmem64_base, kmem64_end, TTE_PAGEMASK(kmem64_szc), tte.ll);
817 	ASSERT(strlen(b) < sizeof (b));
818 	prom_interpret(b, 0, 0, 0, 0, 0);
819 }
820 #endif	/* !C_OBP */
821 
822 /*
823  * As OBP takes up some RAM when the system boots, pages will already be "lost"
824  * to the system and reflected in npages by the time we see it.
825  *
826  * We only want to allocate kernel structures in the 64-bit virtual address
827  * space on systems with enough RAM to make the overhead of keeping track of
828  * an extra kernel memory segment worthwhile.
829  *
830  * Since OBP has already performed its memory allocations by this point, if we
831  * have more than MINMOVE_RAM_MB MB of RAM left free, go ahead and map
832  * memory in the 64-bit virtual address space; otherwise keep allocations
833  * contiguous with we've mapped so far in the 32-bit virtual address space.
834  */
835 #define	MINMOVE_RAM_MB	((size_t)1900)
836 #define	MB_TO_BYTES(mb)	((mb) * 1048576ul)
837 
838 pgcnt_t	tune_npages = (pgcnt_t)
839 	(MB_TO_BYTES(MINMOVE_RAM_MB)/ (size_t)MMU_PAGESIZE);
840 
841 #pragma weak page_set_colorequiv_arr_cpu
842 extern void page_set_colorequiv_arr_cpu(void);
843 
844 static void
845 startup_memlist(void)
846 {
847 	size_t alloc_sz;
848 	size_t ctrs_sz;
849 	caddr_t alloc_base;
850 	caddr_t ctrs_base, ctrs_end;
851 	caddr_t memspace;
852 	caddr_t va;
853 	int memblocks = 0;
854 	struct memlist *cur;
855 	size_t syslimit = (size_t)SYSLIMIT;
856 	size_t sysbase = (size_t)SYSBASE;
857 	int alloc_alignsize = ecache_alignsize;
858 	int i;
859 	extern void page_coloring_init(void);
860 	extern void page_set_colorequiv_arr(void);
861 
862 	/*
863 	 * Initialize enough of the system to allow kmem_alloc to work by
864 	 * calling boot to allocate its memory until the time that
865 	 * kvm_init is completed.  The page structs are allocated after
866 	 * rounding up end to the nearest page boundary; the memsegs are
867 	 * initialized and the space they use comes from the kernel heap.
868 	 * With appropriate initialization, they can be reallocated later
869 	 * to a size appropriate for the machine's configuration.
870 	 *
871 	 * At this point, memory is allocated for things that will never
872 	 * need to be freed, this used to be "valloced".  This allows a
873 	 * savings as the pages don't need page structures to describe
874 	 * them because them will not be managed by the vm system.
875 	 */
876 
877 	/*
878 	 * We're loaded by boot with the following configuration (as
879 	 * specified in the sun4u/conf/Mapfile):
880 	 *
881 	 * 	text:		4 MB chunk aligned on a 4MB boundary
882 	 * 	data & bss:	4 MB chunk aligned on a 4MB boundary
883 	 *
884 	 * These two chunks will eventually be mapped by 2 locked 4MB
885 	 * ttes and will represent the nucleus of the kernel.  This gives
886 	 * us some free space that is already allocated, some or all of
887 	 * which is made available to kernel module text.
888 	 *
889 	 * The free space in the data-bss chunk is used for nucleus
890 	 * allocatable data structures and we reserve it using the
891 	 * nalloc_base and nalloc_end variables.  This space is currently
892 	 * being used for hat data structures required for tlb miss
893 	 * handling operations.  We align nalloc_base to a l2 cache
894 	 * linesize because this is the line size the hardware uses to
895 	 * maintain cache coherency.
896 	 * 256K is carved out for module data.
897 	 */
898 
899 	nalloc_base = (caddr_t)roundup((uintptr_t)e_data, MMU_PAGESIZE);
900 	moddata = nalloc_base;
901 	e_moddata = nalloc_base + MODDATA;
902 	nalloc_base = e_moddata;
903 
904 	nalloc_end = (caddr_t)roundup((uintptr_t)nalloc_base, MMU_PAGESIZE4M);
905 	valloc_base = nalloc_base;
906 
907 	/*
908 	 * Calculate the start of the data segment.
909 	 */
910 	sdata = (caddr_t)((uintptr_t)e_data & MMU_PAGEMASK4M);
911 
912 	PRM_DEBUG(moddata);
913 	PRM_DEBUG(nalloc_base);
914 	PRM_DEBUG(nalloc_end);
915 	PRM_DEBUG(sdata);
916 
917 	/*
918 	 * Remember any slop after e_text so we can give it to the modules.
919 	 */
920 	PRM_DEBUG(e_text);
921 	modtext = (caddr_t)roundup((uintptr_t)e_text, MMU_PAGESIZE);
922 	if (((uintptr_t)e_text & MMU_PAGEMASK4M) != (uintptr_t)s_text)
923 		prom_panic("nucleus text overflow");
924 	modtext_sz = (caddr_t)roundup((uintptr_t)modtext, MMU_PAGESIZE4M) -
925 	    modtext;
926 	PRM_DEBUG(modtext);
927 	PRM_DEBUG(modtext_sz);
928 
929 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
930 	    &boot_physavail, &boot_physavail_len,
931 	    &boot_virtavail, &boot_virtavail_len);
932 	/*
933 	 * Remember what the physically available highest page is
934 	 * so that dumpsys works properly, and find out how much
935 	 * memory is installed.
936 	 */
937 	installed_top_size_memlist_array(boot_physinstalled,
938 	    boot_physinstalled_len, &physmax, &physinstalled);
939 	PRM_DEBUG(physinstalled);
940 	PRM_DEBUG(physmax);
941 
942 	/* Fill out memory nodes config structure */
943 	startup_build_mem_nodes(boot_physinstalled, boot_physinstalled_len);
944 
945 	/*
946 	 * Get the list of physically available memory to size
947 	 * the number of page structures needed.
948 	 */
949 	size_physavail(boot_physavail, boot_physavail_len, &npages, &memblocks);
950 	/*
951 	 * This first snap shot of npages can represent the pages used
952 	 * by OBP's text and data approximately. This is used in the
953 	 * the calculation of the kernel size
954 	 */
955 	obp_pages = physinstalled - npages;
956 
957 
958 	/*
959 	 * On small-memory systems (<MODTEXT_SM_SIZE MB, currently 256MB), the
960 	 * in-nucleus module text is capped to MODTEXT_SM_CAP bytes (currently
961 	 * 2MB) and any excess pages are put on physavail.  The assumption is
962 	 * that small-memory systems will need more pages more than they'll
963 	 * need efficiently-mapped module texts.
964 	 */
965 	if ((physinstalled < mmu_btop(MODTEXT_SM_SIZE << 20)) &&
966 	    modtext_sz > MODTEXT_SM_CAP) {
967 		extra_etpg = mmu_btop(modtext_sz - MODTEXT_SM_CAP);
968 		modtext_sz = MODTEXT_SM_CAP;
969 		extra_etva = modtext + modtext_sz;
970 	}
971 
972 	PRM_DEBUG(extra_etpg);
973 	PRM_DEBUG(modtext_sz);
974 	PRM_DEBUG(extra_etva);
975 
976 	/*
977 	 * Account for any pages after e_text and e_data.
978 	 */
979 	npages += extra_etpg;
980 	npages += mmu_btopr(nalloc_end - nalloc_base);
981 	PRM_DEBUG(npages);
982 
983 	/*
984 	 * npages is the maximum of available physical memory possible.
985 	 * (ie. it will never be more than this)
986 	 */
987 
988 	/*
989 	 * initialize the nucleus memory allocator.
990 	 */
991 	ndata_alloc_init(&ndata, (uintptr_t)nalloc_base, (uintptr_t)nalloc_end);
992 
993 	/*
994 	 * Allocate mmu fault status area from the nucleus data area.
995 	 */
996 	if ((&ndata_alloc_mmfsa != NULL) && (ndata_alloc_mmfsa(&ndata) != 0))
997 		cmn_err(CE_PANIC, "no more nucleus memory after mfsa alloc");
998 
999 	/*
1000 	 * Allocate kernel TSBs from the nucleus data area.
1001 	 */
1002 	if (ndata_alloc_tsbs(&ndata, npages) != 0)
1003 		cmn_err(CE_PANIC, "no more nucleus memory after tsbs alloc");
1004 
1005 	/*
1006 	 * Allocate dmv dispatch table from the nucleus data area.
1007 	 */
1008 	if (ndata_alloc_dmv(&ndata) != 0)
1009 		cmn_err(CE_PANIC, "no more nucleus memory after dmv alloc");
1010 
1011 
1012 	page_coloring_init();
1013 
1014 	/*
1015 	 * Allocate page_freelists bin headers for memnode 0 from the
1016 	 * nucleus data area.
1017 	 */
1018 	if (ndata_alloc_page_freelists(&ndata, 0) != 0)
1019 		cmn_err(CE_PANIC,
1020 		    "no more nucleus memory after page free lists alloc");
1021 
1022 	if (kpm_enable) {
1023 		kpm_init();
1024 		/*
1025 		 * kpm page space -- Update kpm_npages and make the
1026 		 * same assumption about fragmenting as it is done
1027 		 * for memseg_sz.
1028 		 */
1029 		kpm_npages_setup(memblocks + 4);
1030 	}
1031 
1032 	/*
1033 	 * Allocate hat related structs from the nucleus data area.
1034 	 */
1035 	if (ndata_alloc_hat(&ndata, npages, kpm_npages) != 0)
1036 		cmn_err(CE_PANIC, "no more nucleus memory after hat alloc");
1037 
1038 	/*
1039 	 * We want to do the BOP_ALLOCs before the real allocation of page
1040 	 * structs in order to not have to allocate page structs for this
1041 	 * memory.  We need to calculate a virtual address because we want
1042 	 * the page structs to come before other allocations in virtual address
1043 	 * space.  This is so some (if not all) of page structs can actually
1044 	 * live in the nucleus.
1045 	 */
1046 
1047 	/*
1048 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1049 	 *
1050 	 * There are comments all over the SFMMU code warning of dire
1051 	 * consequences if the TSBs are moved out of 32-bit space.  This
1052 	 * is largely because the asm code uses "sethi %hi(addr)"-type
1053 	 * instructions which will not provide the expected result if the
1054 	 * address is a 64-bit one.
1055 	 *
1056 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1057 	 */
1058 	alloc_base = (caddr_t)roundup((uintptr_t)nalloc_end, MMU_PAGESIZE);
1059 	alloc_base = sfmmu_ktsb_alloc(alloc_base);
1060 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1061 	PRM_DEBUG(alloc_base);
1062 
1063 	/*
1064 	 * Allocate IOMMU TSB array.  We do this here so that the physical
1065 	 * memory gets deducted from the PROM's physical memory list.
1066 	 */
1067 	alloc_base = iommu_tsb_init(alloc_base);
1068 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1069 	    ecache_alignsize);
1070 	PRM_DEBUG(alloc_base);
1071 
1072 	/*
1073 	 * Platforms like Starcat and OPL need special structures assigned in
1074 	 * 32-bit virtual address space because their probing routines execute
1075 	 * FCode, and FCode can't handle 64-bit virtual addresses...
1076 	 */
1077 	if (&plat_startup_memlist) {
1078 		alloc_base = plat_startup_memlist(alloc_base);
1079 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1080 		    ecache_alignsize);
1081 		PRM_DEBUG(alloc_base);
1082 	}
1083 
1084 	/*
1085 	 * Save off where the contiguous allocations to date have ended
1086 	 * in econtig32.
1087 	 */
1088 	econtig32 = alloc_base;
1089 	PRM_DEBUG(econtig32);
1090 
1091 	if (econtig32 > (caddr_t)KERNEL_LIMIT32)
1092 		cmn_err(CE_PANIC, "econtig32 too big");
1093 
1094 	/*
1095 	 * To avoid memory allocation collisions in the 32-bit virtual address
1096 	 * space, make allocations from this point forward in 64-bit virtual
1097 	 * address space starting at syslimit and working up.
1098 	 *
1099 	 * All this is needed because on large memory systems, the default
1100 	 * Solaris allocations will collide with SYSBASE32, which is hard
1101 	 * coded to be at the virtual address 0x78000000.  Therefore, on 64-bit
1102 	 * kernels, move the allocations to a location in the 64-bit virtual
1103 	 * address space space, allowing those structures to grow without
1104 	 * worry.
1105 	 *
1106 	 * On current CPUs we'll run out of physical memory address bits before
1107 	 * we need to worry about the allocations running into anything else in
1108 	 * VM or the virtual address holes on US-I and II, as there's currently
1109 	 * about 1 TB of addressable space before the US-I/II VA hole.
1110 	 */
1111 	kmem64_base = (caddr_t)syslimit;
1112 	PRM_DEBUG(kmem64_base);
1113 
1114 	/*
1115 	 * Allocate addresses, but not physical memory. None of these locations
1116 	 * can be touched until physical memory is allocated below.
1117 	 */
1118 	alloc_base = kmem64_base;
1119 
1120 	/*
1121 	 * If KHME and/or UHME hash buckets won't fit in the nucleus, allocate
1122 	 * them here.
1123 	 */
1124 	if (khme_hash == NULL || uhme_hash == NULL) {
1125 		/*
1126 		 * alloc_hme_buckets() will align alloc_base properly before
1127 		 * assigning the hash buckets, so we don't need to do it
1128 		 * before the call...
1129 		 */
1130 		alloc_base = alloc_hme_buckets(alloc_base, alloc_alignsize);
1131 
1132 		PRM_DEBUG(alloc_base);
1133 		PRM_DEBUG(khme_hash);
1134 		PRM_DEBUG(uhme_hash);
1135 	}
1136 
1137 	/*
1138 	 * Allocate the remaining page freelists.  NUMA systems can
1139 	 * have lots of page freelists, one per node, which quickly
1140 	 * outgrow the amount of nucleus memory available.
1141 	 */
1142 	if (max_mem_nodes > 1) {
1143 		int mnode;
1144 
1145 		for (mnode = 1; mnode < max_mem_nodes; mnode++) {
1146 			alloc_base = alloc_page_freelists(mnode, alloc_base,
1147 				ecache_alignsize);
1148 		}
1149 		PRM_DEBUG(alloc_base);
1150 	}
1151 
1152 	if (!mml_table) {
1153 		size_t mmltable_sz;
1154 
1155 		/*
1156 		 * We need to allocate the mml_table here because there
1157 		 * was not enough space within the nucleus.
1158 		 */
1159 		mmltable_sz = sizeof (kmutex_t) * mml_table_sz;
1160 		alloc_sz = roundup(mmltable_sz, alloc_alignsize);
1161 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1162 		    alloc_alignsize);
1163 		mml_table = (kmutex_t *)alloc_base;
1164 		alloc_base += alloc_sz;
1165 		PRM_DEBUG(mml_table);
1166 		PRM_DEBUG(alloc_base);
1167 	}
1168 
1169 	if (kpm_enable && !(kpmp_table || kpmp_stable)) {
1170 		size_t kpmptable_sz;
1171 		caddr_t table;
1172 
1173 		/*
1174 		 * We need to allocate either kpmp_table or kpmp_stable here
1175 		 * because there was not enough space within the nucleus.
1176 		 */
1177 		kpmptable_sz = (kpm_smallpages == 0) ?
1178 				sizeof (kpm_hlk_t) * kpmp_table_sz :
1179 				sizeof (kpm_shlk_t) * kpmp_stable_sz;
1180 
1181 		alloc_sz = roundup(kpmptable_sz, alloc_alignsize);
1182 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1183 		    alloc_alignsize);
1184 
1185 		table = alloc_base;
1186 
1187 		if (kpm_smallpages == 0) {
1188 			kpmp_table = (kpm_hlk_t *)table;
1189 			PRM_DEBUG(kpmp_table);
1190 		} else {
1191 			kpmp_stable = (kpm_shlk_t *)table;
1192 			PRM_DEBUG(kpmp_stable);
1193 		}
1194 
1195 		alloc_base += alloc_sz;
1196 		PRM_DEBUG(alloc_base);
1197 	}
1198 
1199 	if (&ecache_init_scrub_flush_area) {
1200 		/*
1201 		 * Pass alloc_base directly, as the routine itself is
1202 		 * responsible for any special alignment requirements...
1203 		 */
1204 		alloc_base = ecache_init_scrub_flush_area(alloc_base);
1205 		PRM_DEBUG(alloc_base);
1206 	}
1207 
1208 	/*
1209 	 * Take the most current snapshot we can by calling mem-update.
1210 	 */
1211 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1212 	    &boot_physavail, &boot_physavail_len,
1213 	    &boot_virtavail, &boot_virtavail_len);
1214 
1215 	/*
1216 	 * Reset npages and memblocks based on boot_physavail list.
1217 	 */
1218 	size_physavail(boot_physavail, boot_physavail_len, &npages, &memblocks);
1219 	PRM_DEBUG(npages);
1220 
1221 	/*
1222 	 * Account for extra memory after e_text.
1223 	 */
1224 	npages += extra_etpg;
1225 
1226 	/*
1227 	 * Calculate the largest free memory chunk in the nucleus data area.
1228 	 * We need to figure out if page structs can fit in there or not.
1229 	 * We also make sure enough page structs get created for any physical
1230 	 * memory we might be returning to the system.
1231 	 */
1232 	ndata_remain_sz = ndata_maxsize(&ndata);
1233 	PRM_DEBUG(ndata_remain_sz);
1234 
1235 	pp_sz = sizeof (struct page) * npages;
1236 
1237 	/*
1238 	 * Here's a nice bit of code based on somewhat recursive logic:
1239 	 *
1240 	 * If the page array would fit within the nucleus, we want to
1241 	 * add npages to cover any extra memory we may be returning back
1242 	 * to the system.
1243 	 *
1244 	 * HOWEVER, the page array is sized by calculating the size of
1245 	 * (struct page * npages), as are the pagehash table, ctrs and
1246 	 * memseg_list, so the very act of performing the calculation below may
1247 	 * in fact make the array large enough that it no longer fits in the
1248 	 * nucleus, meaning there would now be a much larger area of the
1249 	 * nucleus free that should really be added to npages, which would
1250 	 * make the page array that much larger, and so on.
1251 	 *
1252 	 * This also ignores the memory possibly used in the nucleus for the
1253 	 * the page hash, ctrs and memseg list and the fact that whether they
1254 	 * fit there or not varies with the npages calculation below, but we
1255 	 * don't even factor them into the equation at this point; perhaps we
1256 	 * should or perhaps we should just take the approach that the few
1257 	 * extra pages we could add via this calculation REALLY aren't worth
1258 	 * the hassle...
1259 	 */
1260 	if (ndata_remain_sz > pp_sz) {
1261 		size_t spare = ndata_spare(&ndata, pp_sz, ecache_alignsize);
1262 
1263 		npages += mmu_btop(spare);
1264 
1265 		pp_sz = npages * sizeof (struct page);
1266 
1267 		pp_base = ndata_alloc(&ndata, pp_sz, ecache_alignsize);
1268 	}
1269 
1270 	/*
1271 	 * If physmem is patched to be non-zero, use it instead of
1272 	 * the monitor value unless physmem is larger than the total
1273 	 * amount of memory on hand.
1274 	 */
1275 	if (physmem == 0 || physmem > npages)
1276 		physmem = npages;
1277 
1278 	/*
1279 	 * If pp_base is NULL that means the routines above have determined
1280 	 * the page array will not fit in the nucleus; we'll have to
1281 	 * BOP_ALLOC() ourselves some space for them.
1282 	 */
1283 	if (pp_base == NULL) {
1284 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1285 		    alloc_alignsize);
1286 		alloc_sz = roundup(pp_sz, alloc_alignsize);
1287 
1288 		pp_base = (struct page *)alloc_base;
1289 
1290 		alloc_base += alloc_sz;
1291 	}
1292 
1293 	/*
1294 	 * The page structure hash table size is a power of 2
1295 	 * such that the average hash chain length is PAGE_HASHAVELEN.
1296 	 */
1297 	page_hashsz = npages / PAGE_HASHAVELEN;
1298 	page_hashsz = 1 << highbit((ulong_t)page_hashsz);
1299 	pagehash_sz = sizeof (struct page *) * page_hashsz;
1300 
1301 	/*
1302 	 * We want to TRY to fit the page structure hash table,
1303 	 * the page size free list counters, the memseg list and
1304 	 * and the kpm page space in the nucleus if possible.
1305 	 *
1306 	 * alloc_sz counts how much memory needs to be allocated by
1307 	 * BOP_ALLOC().
1308 	 */
1309 	page_hash = ndata_alloc(&ndata, pagehash_sz, ecache_alignsize);
1310 
1311 	alloc_sz = (page_hash == NULL ? pagehash_sz : 0);
1312 
1313 	/*
1314 	 * Size up per page size free list counters.
1315 	 */
1316 	ctrs_sz = page_ctrs_sz();
1317 	ctrs_base = ndata_alloc(&ndata, ctrs_sz, ecache_alignsize);
1318 
1319 	if (ctrs_base == NULL)
1320 		alloc_sz = roundup(alloc_sz, ecache_alignsize) + ctrs_sz;
1321 
1322 	/*
1323 	 * The memseg list is for the chunks of physical memory that
1324 	 * will be managed by the vm system.  The number calculated is
1325 	 * a guess as boot may fragment it more when memory allocations
1326 	 * are made before kphysm_init().  Currently, there are two
1327 	 * allocations before then, so we assume each causes fragmen-
1328 	 * tation, and add a couple more for good measure.
1329 	 */
1330 	memseg_sz = sizeof (struct memseg) * (memblocks + 4);
1331 	memseg_base = ndata_alloc(&ndata, memseg_sz, ecache_alignsize);
1332 
1333 	if (memseg_base == NULL)
1334 		alloc_sz = roundup(alloc_sz, ecache_alignsize) + memseg_sz;
1335 
1336 
1337 	if (kpm_enable) {
1338 		/*
1339 		 * kpm page space -- Update kpm_npages and make the
1340 		 * same assumption about fragmenting as it is done
1341 		 * for memseg_sz above.
1342 		 */
1343 		kpm_npages_setup(memblocks + 4);
1344 		kpm_pp_sz = (kpm_smallpages == 0) ?
1345 		    kpm_npages * sizeof (kpm_page_t):
1346 		    kpm_npages * sizeof (kpm_spage_t);
1347 
1348 		kpm_pp_base = (uintptr_t)ndata_alloc(&ndata, kpm_pp_sz,
1349 		    ecache_alignsize);
1350 
1351 		if (kpm_pp_base == NULL)
1352 			alloc_sz = roundup(alloc_sz, ecache_alignsize) +
1353 			    kpm_pp_sz;
1354 	}
1355 
1356 	if (alloc_sz > 0) {
1357 		uintptr_t bop_base;
1358 
1359 		/*
1360 		 * We need extra memory allocated through BOP_ALLOC.
1361 		 */
1362 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1363 		    alloc_alignsize);
1364 
1365 		alloc_sz = roundup(alloc_sz, alloc_alignsize);
1366 
1367 		bop_base = (uintptr_t)alloc_base;
1368 
1369 		alloc_base += alloc_sz;
1370 
1371 		if (page_hash == NULL) {
1372 			page_hash = (struct page **)bop_base;
1373 			bop_base = roundup(bop_base + pagehash_sz,
1374 			    ecache_alignsize);
1375 		}
1376 
1377 		if (ctrs_base == NULL) {
1378 			ctrs_base = (caddr_t)bop_base;
1379 			bop_base = roundup(bop_base + ctrs_sz,
1380 			    ecache_alignsize);
1381 		}
1382 
1383 		if (memseg_base == NULL) {
1384 			memseg_base = (struct memseg *)bop_base;
1385 			bop_base = roundup(bop_base + memseg_sz,
1386 			    ecache_alignsize);
1387 		}
1388 
1389 		if (kpm_enable && kpm_pp_base == NULL) {
1390 			kpm_pp_base = (uintptr_t)bop_base;
1391 			bop_base = roundup(bop_base + kpm_pp_sz,
1392 			    ecache_alignsize);
1393 		}
1394 
1395 		ASSERT(bop_base <= (uintptr_t)alloc_base);
1396 	}
1397 
1398 	PRM_DEBUG(page_hash);
1399 	PRM_DEBUG(memseg_base);
1400 	PRM_DEBUG(kpm_pp_base);
1401 	PRM_DEBUG(kpm_pp_sz);
1402 	PRM_DEBUG(pp_base);
1403 	PRM_DEBUG(pp_sz);
1404 	PRM_DEBUG(alloc_base);
1405 
1406 #ifdef	TRAPTRACE
1407 	alloc_base = trap_trace_alloc(alloc_base);
1408 	PRM_DEBUG(alloc_base);
1409 #endif	/* TRAPTRACE */
1410 
1411 	/*
1412 	 * In theory it's possible that kmem64 chunk is 0 sized
1413 	 * (on very small machines). Check for that.
1414 	 */
1415 	if (alloc_base == kmem64_base) {
1416 		kmem64_base = NULL;
1417 		kmem64_end = NULL;
1418 		kmem64_aligned_end = NULL;
1419 		goto kmem64_alloced;
1420 	}
1421 
1422 	/*
1423 	 * Allocate kmem64 memory.
1424 	 * Round up to end of large page and overmap.
1425 	 * kmem64_end..kmem64_aligned_end is added to memory list for reuse
1426 	 */
1427 	kmem64_end = (caddr_t)roundup((uintptr_t)alloc_base,
1428 	    MMU_PAGESIZE);
1429 
1430 	/*
1431 	 * Make one large memory alloc after figuring out the 64-bit size. This
1432 	 * will enable use of the largest page size appropriate for the system
1433 	 * architecture.
1434 	 */
1435 	ASSERT(mmu_exported_pagesize_mask & (1 << TTE8K));
1436 	ASSERT(IS_P2ALIGNED(kmem64_base, TTEBYTES(max_bootlp_tteszc)));
1437 	for (i = max_bootlp_tteszc; i >= TTE8K; i--) {
1438 		size_t asize;
1439 #if !defined(C_OBP)
1440 		unsigned long long pa;
1441 #endif	/* !C_OBP */
1442 
1443 		if ((mmu_exported_pagesize_mask & (1 << i)) == 0)
1444 			continue;
1445 		kmem64_alignsize = TTEBYTES(i);
1446 		kmem64_szc = i;
1447 
1448 		/* limit page size for small memory */
1449 		if (mmu_btop(kmem64_alignsize) > (npages >> 2))
1450 			continue;
1451 
1452 		kmem64_aligned_end = (caddr_t)roundup((uintptr_t)kmem64_end,
1453 		    kmem64_alignsize);
1454 		asize = kmem64_aligned_end - kmem64_base;
1455 #if !defined(C_OBP)
1456 		if (prom_allocate_phys(asize, kmem64_alignsize, &pa) == 0) {
1457 			if (prom_claim_virt(asize, kmem64_base) !=
1458 			    (caddr_t)-1) {
1459 				kmem64_pabase = pa;
1460 				install_kmem64_tte();
1461 				break;
1462 			} else {
1463 				prom_free_phys(asize, pa);
1464 			}
1465 		}
1466 #else	/* !C_OBP */
1467 		if ((caddr_t)BOP_ALLOC(bootops, kmem64_base, asize,
1468 		    kmem64_alignsize) == kmem64_base) {
1469 			kmem64_pabase = va_to_pa(kmem64_base);
1470 			break;
1471 		}
1472 #endif	/* !C_OBP */
1473 		if (i == TTE8K) {
1474 			prom_panic("kmem64 allocation failure");
1475 		}
1476 	}
1477 
1478 	PRM_DEBUG(kmem64_base);
1479 	PRM_DEBUG(kmem64_end);
1480 	PRM_DEBUG(kmem64_aligned_end);
1481 	PRM_DEBUG(kmem64_alignsize);
1482 
1483 	/*
1484 	 * Now set pa using saved va from above.
1485 	 */
1486 	if (&ecache_init_scrub_flush_area) {
1487 		(void) ecache_init_scrub_flush_area(NULL);
1488 	}
1489 
1490 kmem64_alloced:
1491 
1492 	/*
1493 	 * Initialize per page size free list counters.
1494 	 */
1495 	ctrs_end = page_ctrs_alloc(ctrs_base);
1496 	ASSERT(ctrs_base + ctrs_sz >= ctrs_end);
1497 
1498 	/*
1499 	 * Allocate space for the interrupt vector table and also for the
1500 	 * reserved interrupt vector data structures.
1501 	 */
1502 	memspace = (caddr_t)BOP_ALLOC(bootops, (caddr_t)intr_vec_table,
1503 	    IVSIZE, MMU_PAGESIZE);
1504 	if (memspace != (caddr_t)intr_vec_table)
1505 		prom_panic("interrupt vector table allocation failure");
1506 
1507 	/*
1508 	 * The memory lists from boot are allocated from the heap arena
1509 	 * so that later they can be freed and/or reallocated.
1510 	 */
1511 	if (BOP_GETPROP(bootops, "extent", &memlist_sz) == -1)
1512 		prom_panic("could not retrieve property \"extent\"");
1513 
1514 	/*
1515 	 * Between now and when we finish copying in the memory lists,
1516 	 * allocations happen so the space gets fragmented and the
1517 	 * lists longer.  Leave enough space for lists twice as long
1518 	 * as what boot says it has now; roundup to a pagesize.
1519 	 * Also add space for the final phys-avail copy in the fixup
1520 	 * routine.
1521 	 */
1522 	va = (caddr_t)(sysbase + PAGESIZE + PANICBUFSIZE +
1523 	    roundup(IVSIZE, MMU_PAGESIZE));
1524 	memlist_sz *= 4;
1525 	memlist_sz = roundup(memlist_sz, MMU_PAGESIZE);
1526 	memspace = (caddr_t)BOP_ALLOC(bootops, va, memlist_sz, BO_NO_ALIGN);
1527 	if (memspace == NULL)
1528 		halt("Boot allocation failed.");
1529 
1530 	memlist = (struct memlist *)memspace;
1531 	memlist_end = (char *)memspace + memlist_sz;
1532 
1533 	PRM_DEBUG(memlist);
1534 	PRM_DEBUG(memlist_end);
1535 	PRM_DEBUG(sysbase);
1536 	PRM_DEBUG(syslimit);
1537 
1538 	kernelheap_init((void *)sysbase, (void *)syslimit,
1539 	    (caddr_t)sysbase + PAGESIZE, NULL, NULL);
1540 
1541 	/*
1542 	 * Take the most current snapshot we can by calling mem-update.
1543 	 */
1544 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1545 	    &boot_physavail, &boot_physavail_len,
1546 	    &boot_virtavail, &boot_virtavail_len);
1547 
1548 	/*
1549 	 * Remove the space used by BOP_ALLOC from the kernel heap
1550 	 * plus the area actually used by the OBP (if any)
1551 	 * ignoring virtual addresses in virt_avail, above syslimit.
1552 	 */
1553 	virt_avail = memlist;
1554 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1555 
1556 	for (cur = virt_avail; cur->next; cur = cur->next) {
1557 		uint64_t range_base, range_size;
1558 
1559 		if ((range_base = cur->address + cur->size) < (uint64_t)sysbase)
1560 			continue;
1561 		if (range_base >= (uint64_t)syslimit)
1562 			break;
1563 		/*
1564 		 * Limit the range to end at syslimit.
1565 		 */
1566 		range_size = MIN(cur->next->address,
1567 		    (uint64_t)syslimit) - range_base;
1568 		(void) vmem_xalloc(heap_arena, (size_t)range_size, PAGESIZE,
1569 		    0, 0, (void *)range_base, (void *)(range_base + range_size),
1570 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1571 	}
1572 
1573 	phys_avail = memlist;
1574 	(void) copy_physavail(boot_physavail, boot_physavail_len,
1575 	    &memlist, 0, 0);
1576 
1577 	/*
1578 	 * Add any unused kmem64 memory from overmapped page
1579 	 * (Note: va_to_pa does not work for kmem64_end)
1580 	 */
1581 	if (kmem64_end < kmem64_aligned_end) {
1582 		uint64_t overlap_size = kmem64_aligned_end - kmem64_end;
1583 		uint64_t overlap_pa = kmem64_pabase +
1584 		    (kmem64_end - kmem64_base);
1585 
1586 		PRM_DEBUG(overlap_pa);
1587 		PRM_DEBUG(overlap_size);
1588 		memlist_add(overlap_pa, overlap_size, &memlist, &phys_avail);
1589 	}
1590 
1591 	/*
1592 	 * Add any extra memory after e_text to the phys_avail list, as long
1593 	 * as there's at least a page to add.
1594 	 */
1595 	if (extra_etpg)
1596 		memlist_add(va_to_pa(extra_etva), mmu_ptob(extra_etpg),
1597 		    &memlist, &phys_avail);
1598 
1599 	/*
1600 	 * Add any extra memory after e_data to the phys_avail list as long
1601 	 * as there's at least a page to add.  Usually, there isn't any,
1602 	 * since extra HME blocks typically get allocated there first before
1603 	 * using RAM elsewhere.
1604 	 */
1605 	if ((nalloc_base = ndata_extra_base(&ndata, MMU_PAGESIZE)) == NULL)
1606 		nalloc_base = nalloc_end;
1607 	ndata_remain_sz = nalloc_end - nalloc_base;
1608 
1609 	if (ndata_remain_sz >= MMU_PAGESIZE)
1610 		memlist_add(va_to_pa(nalloc_base),
1611 		    (uint64_t)ndata_remain_sz, &memlist, &phys_avail);
1612 
1613 	PRM_DEBUG(memlist);
1614 	PRM_DEBUG(memlist_sz);
1615 	PRM_DEBUG(memspace);
1616 
1617 	if ((caddr_t)memlist > (memspace + memlist_sz))
1618 		prom_panic("memlist overflow");
1619 
1620 	PRM_DEBUG(pp_base);
1621 	PRM_DEBUG(memseg_base);
1622 	PRM_DEBUG(npages);
1623 
1624 	/*
1625 	 * Initialize the page structures from the memory lists.
1626 	 */
1627 	kphysm_init(pp_base, memseg_base, npages, kpm_pp_base, kpm_npages);
1628 
1629 	availrmem_initial = availrmem = freemem;
1630 	PRM_DEBUG(availrmem);
1631 
1632 	/*
1633 	 * Some of the locks depend on page_hashsz being set!
1634 	 * kmem_init() depends on this; so, keep it here.
1635 	 */
1636 	page_lock_init();
1637 
1638 	/*
1639 	 * Initialize kernel memory allocator.
1640 	 */
1641 	kmem_init();
1642 
1643 	/*
1644 	 * Factor in colorequiv to check additional 'equivalent' bins
1645 	 */
1646 	if (&page_set_colorequiv_arr_cpu != NULL)
1647 		page_set_colorequiv_arr_cpu();
1648 	else
1649 		page_set_colorequiv_arr();
1650 
1651 	/*
1652 	 * Initialize bp_mapin().
1653 	 */
1654 	bp_init(shm_alignment, HAT_STRICTORDER);
1655 
1656 	/*
1657 	 * Reserve space for panicbuf, intr_vec_table and reserved interrupt
1658 	 * vector data structures from the 32-bit heap.
1659 	 */
1660 	(void) vmem_xalloc(heap32_arena, PANICBUFSIZE, PAGESIZE, 0, 0,
1661 	    panicbuf, panicbuf + PANICBUFSIZE,
1662 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1663 
1664 	(void) vmem_xalloc(heap32_arena, IVSIZE, PAGESIZE, 0, 0,
1665 	    intr_vec_table, (caddr_t)intr_vec_table + IVSIZE,
1666 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1667 
1668 	mem_config_init();
1669 }
1670 
1671 static void
1672 startup_modules(void)
1673 {
1674 	int proplen, nhblk1, nhblk8;
1675 	size_t  nhblksz;
1676 	pgcnt_t pages_per_hblk;
1677 	size_t hme8blk_sz, hme1blk_sz;
1678 
1679 	/*
1680 	 * Log any optional messages from the boot program
1681 	 */
1682 	proplen = (size_t)BOP_GETPROPLEN(bootops, "boot-message");
1683 	if (proplen > 0) {
1684 		char *msg;
1685 		size_t len = (size_t)proplen;
1686 
1687 		msg = kmem_zalloc(len, KM_SLEEP);
1688 		(void) BOP_GETPROP(bootops, "boot-message", msg);
1689 		cmn_err(CE_CONT, "?%s\n", msg);
1690 		kmem_free(msg, len);
1691 	}
1692 
1693 	/*
1694 	 * Let the platforms have a chance to change default
1695 	 * values before reading system file.
1696 	 */
1697 	if (&set_platform_defaults)
1698 		set_platform_defaults();
1699 
1700 	/*
1701 	 * Calculate default settings of system parameters based upon
1702 	 * maxusers, yet allow to be overridden via the /etc/system file.
1703 	 */
1704 	param_calc(0);
1705 
1706 	mod_setup();
1707 
1708 	/*
1709 	 * If this is a positron, complain and halt.
1710 	 */
1711 	if (&iam_positron && iam_positron()) {
1712 		cmn_err(CE_WARN, "This hardware platform is not supported"
1713 		    " by this release of Solaris.\n");
1714 #ifdef DEBUG
1715 		prom_enter_mon();	/* Type 'go' to resume */
1716 		cmn_err(CE_WARN, "Booting an unsupported platform.\n");
1717 		cmn_err(CE_WARN, "Booting with down-rev firmware.\n");
1718 
1719 #else /* DEBUG */
1720 		halt(0);
1721 #endif /* DEBUG */
1722 	}
1723 
1724 	/*
1725 	 * If we are running firmware that isn't 64-bit ready
1726 	 * then complain and halt.
1727 	 */
1728 	do_prom_version_check();
1729 
1730 	/*
1731 	 * Initialize system parameters
1732 	 */
1733 	param_init();
1734 
1735 	/*
1736 	 * maxmem is the amount of physical memory we're playing with.
1737 	 */
1738 	maxmem = physmem;
1739 
1740 	/* Set segkp limits. */
1741 	ncbase = kdi_segdebugbase;
1742 	ncend = kdi_segdebugbase;
1743 
1744 	/*
1745 	 * Initialize the hat layer.
1746 	 */
1747 	hat_init();
1748 
1749 	/*
1750 	 * Initialize segment management stuff.
1751 	 */
1752 	seg_init();
1753 
1754 	/*
1755 	 * Create the va>tte handler, so the prom can understand
1756 	 * kernel translations.  The handler is installed later, just
1757 	 * as we are about to take over the trap table from the prom.
1758 	 */
1759 	create_va_to_tte();
1760 
1761 	/*
1762 	 * Load the forthdebugger (optional)
1763 	 */
1764 	forthdebug_init();
1765 
1766 	/*
1767 	 * Create OBP node for console input callbacks
1768 	 * if it is needed.
1769 	 */
1770 	startup_create_io_node();
1771 
1772 	if (modloadonly("fs", "specfs") == -1)
1773 		halt("Can't load specfs");
1774 
1775 	if (modloadonly("fs", "devfs") == -1)
1776 		halt("Can't load devfs");
1777 
1778 	if (modloadonly("misc", "swapgeneric") == -1)
1779 		halt("Can't load swapgeneric");
1780 
1781 	(void) modloadonly("sys", "lbl_edition");
1782 
1783 	dispinit();
1784 
1785 	/*
1786 	 * Infer meanings to the members of the idprom buffer.
1787 	 */
1788 	parse_idprom();
1789 
1790 	/* Read cluster configuration data. */
1791 	clconf_init();
1792 
1793 	setup_ddi();
1794 
1795 	/*
1796 	 * Lets take this opportunity to load the root device.
1797 	 */
1798 	if (loadrootmodules() != 0)
1799 		debug_enter("Can't load the root filesystem");
1800 
1801 	/*
1802 	 * Load tod driver module for the tod part found on this system.
1803 	 * Recompute the cpu frequency/delays based on tod as tod part
1804 	 * tends to keep time more accurately.
1805 	 */
1806 	if (&load_tod_module)
1807 		load_tod_module();
1808 
1809 	/*
1810 	 * Allow platforms to load modules which might
1811 	 * be needed after bootops are gone.
1812 	 */
1813 	if (&load_platform_modules)
1814 		load_platform_modules();
1815 
1816 	setcpudelay();
1817 
1818 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1819 	    &boot_physavail, &boot_physavail_len,
1820 	    &boot_virtavail, &boot_virtavail_len);
1821 
1822 	/*
1823 	 * Calculation and allocation of hmeblks needed to remap
1824 	 * the memory allocated by PROM till now.
1825 	 * Overestimate the number of hblk1 elements by assuming
1826 	 * worst case of TTE64K mappings.
1827 	 * sfmmu_hblk_alloc will panic if this calculation is wrong.
1828 	 */
1829 	bop_alloc_pages = btopr(kmem64_end - kmem64_base);
1830 	pages_per_hblk = btop(HMEBLK_SPAN(TTE64K));
1831 	bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1832 	nhblk1 = bop_alloc_pages / pages_per_hblk + hblk1_min;
1833 
1834 	bop_alloc_pages = size_virtalloc(boot_virtavail, boot_virtavail_len);
1835 
1836 	/* sfmmu_init_nucleus_hblks expects properly aligned data structures */
1837 	hme8blk_sz = roundup(HME8BLK_SZ, sizeof (int64_t));
1838 	hme1blk_sz = roundup(HME1BLK_SZ, sizeof (int64_t));
1839 
1840 	bop_alloc_pages += btopr(nhblk1 * hme1blk_sz);
1841 
1842 	pages_per_hblk = btop(HMEBLK_SPAN(TTE8K));
1843 	nhblk8 = 0;
1844 	while (bop_alloc_pages > 1) {
1845 		bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1846 		nhblk8 += bop_alloc_pages /= pages_per_hblk;
1847 		bop_alloc_pages *= hme8blk_sz;
1848 		bop_alloc_pages = btopr(bop_alloc_pages);
1849 	}
1850 	nhblk8 += 2;
1851 
1852 	/*
1853 	 * Since hblk8's can hold up to 64k of mappings aligned on a 64k
1854 	 * boundary, the number of hblk8's needed to map the entries in the
1855 	 * boot_virtavail list needs to be adjusted to take this into
1856 	 * consideration.  Thus, we need to add additional hblk8's since it
1857 	 * is possible that an hblk8 will not have all 8 slots used due to
1858 	 * alignment constraints.  Since there were boot_virtavail_len entries
1859 	 * in that list, we need to add that many hblk8's to the number
1860 	 * already calculated to make sure we don't underestimate.
1861 	 */
1862 	nhblk8 += boot_virtavail_len;
1863 	nhblksz = nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz;
1864 
1865 	/* Allocate in pagesize chunks */
1866 	nhblksz = roundup(nhblksz, MMU_PAGESIZE);
1867 	hblk_base = kmem_zalloc(nhblksz, KM_SLEEP);
1868 	sfmmu_init_nucleus_hblks(hblk_base, nhblksz, nhblk8, nhblk1);
1869 }
1870 
1871 static void
1872 startup_bop_gone(void)
1873 {
1874 	extern int bop_io_quiesced;
1875 
1876 	/*
1877 	 * Destroy the MD initialized at startup
1878 	 * The startup initializes the MD framework
1879 	 * using prom and BOP alloc free it now.
1880 	 */
1881 	mach_descrip_startup_fini();
1882 
1883 	/*
1884 	 * Call back into boot and release boots resources.
1885 	 */
1886 	BOP_QUIESCE_IO(bootops);
1887 	bop_io_quiesced = 1;
1888 
1889 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1890 	    &boot_physavail, &boot_physavail_len,
1891 	    &boot_virtavail, &boot_virtavail_len);
1892 	/*
1893 	 * Copy physinstalled list into kernel space.
1894 	 */
1895 	phys_install = memlist;
1896 	copy_memlist(boot_physinstalled, boot_physinstalled_len, &memlist);
1897 
1898 	/*
1899 	 * setup physically contiguous area twice as large as the ecache.
1900 	 * this is used while doing displacement flush of ecaches
1901 	 */
1902 	if (&ecache_flush_address) {
1903 		ecache_flushaddr = ecache_flush_address();
1904 		if (ecache_flushaddr == (uint64_t)-1) {
1905 			cmn_err(CE_PANIC,
1906 			    "startup: no memory to set ecache_flushaddr");
1907 		}
1908 	}
1909 
1910 	/*
1911 	 * Virtual available next.
1912 	 */
1913 	ASSERT(virt_avail != NULL);
1914 	memlist_free_list(virt_avail);
1915 	virt_avail = memlist;
1916 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1917 
1918 	/*
1919 	 * Last chance to ask our booter questions ..
1920 	 */
1921 }
1922 
1923 
1924 /*
1925  * startup_fixup_physavail - called from mach_sfmmu.c after the final
1926  * allocations have been performed.  We can't call it in startup_bop_gone
1927  * since later operations can cause obp to allocate more memory.
1928  */
1929 void
1930 startup_fixup_physavail(void)
1931 {
1932 	struct memlist *cur;
1933 	size_t kmem64_overmap_size = kmem64_aligned_end - kmem64_end;
1934 
1935 	PRM_DEBUG(kmem64_overmap_size);
1936 
1937 	/*
1938 	 * take the most current snapshot we can by calling mem-update
1939 	 */
1940 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1941 	    &boot_physavail, &boot_physavail_len,
1942 	    &boot_virtavail, &boot_virtavail_len);
1943 
1944 	/*
1945 	 * Copy phys_avail list, again.
1946 	 * Both the kernel/boot and the prom have been allocating
1947 	 * from the original list we copied earlier.
1948 	 */
1949 	cur = memlist;
1950 	(void) copy_physavail(boot_physavail, boot_physavail_len,
1951 	    &memlist, 0, 0);
1952 
1953 	/*
1954 	 * Add any unused kmem64 memory from overmapped page
1955 	 * (Note: va_to_pa does not work for kmem64_end)
1956 	 */
1957 	if (kmem64_overmap_size) {
1958 		memlist_add(kmem64_pabase + (kmem64_end - kmem64_base),
1959 		    kmem64_overmap_size,
1960 		    &memlist, &cur);
1961 	}
1962 
1963 	/*
1964 	 * Add any extra memory after e_text we added to the phys_avail list
1965 	 * back to the old list.
1966 	 */
1967 	if (extra_etpg)
1968 		memlist_add(va_to_pa(extra_etva), mmu_ptob(extra_etpg),
1969 		    &memlist, &cur);
1970 	if (ndata_remain_sz >= MMU_PAGESIZE)
1971 		memlist_add(va_to_pa(nalloc_base),
1972 		    (uint64_t)ndata_remain_sz, &memlist, &cur);
1973 
1974 	/*
1975 	 * There isn't any bounds checking on the memlist area
1976 	 * so ensure it hasn't overgrown.
1977 	 */
1978 	if ((caddr_t)memlist > (caddr_t)memlist_end)
1979 		cmn_err(CE_PANIC, "startup: memlist size exceeded");
1980 
1981 	/*
1982 	 * The kernel removes the pages that were allocated for it from
1983 	 * the freelist, but we now have to find any -extra- pages that
1984 	 * the prom has allocated for it's own book-keeping, and remove
1985 	 * them from the freelist too. sigh.
1986 	 */
1987 	fix_prom_pages(phys_avail, cur);
1988 
1989 	ASSERT(phys_avail != NULL);
1990 	memlist_free_list(phys_avail);
1991 	phys_avail = cur;
1992 
1993 	/*
1994 	 * We're done with boot.  Just after this point in time, boot
1995 	 * gets unmapped, so we can no longer rely on its services.
1996 	 * Zero the bootops to indicate this fact.
1997 	 */
1998 	bootops = (struct bootops *)NULL;
1999 	BOOTOPS_GONE();
2000 }
2001 
2002 static void
2003 startup_vm(void)
2004 {
2005 	size_t	i;
2006 	struct segmap_crargs a;
2007 	struct segkpm_crargs b;
2008 
2009 	uint64_t avmem;
2010 	caddr_t va;
2011 	pgcnt_t	max_phys_segkp;
2012 	int	mnode;
2013 
2014 	extern int use_brk_lpg, use_stk_lpg;
2015 
2016 	/*
2017 	 * get prom's mappings, create hments for them and switch
2018 	 * to the kernel context.
2019 	 */
2020 	hat_kern_setup();
2021 
2022 	/*
2023 	 * Take over trap table
2024 	 */
2025 	setup_trap_table();
2026 
2027 	/*
2028 	 * Install the va>tte handler, so that the prom can handle
2029 	 * misses and understand the kernel table layout in case
2030 	 * we need call into the prom.
2031 	 */
2032 	install_va_to_tte();
2033 
2034 	/*
2035 	 * Set a flag to indicate that the tba has been taken over.
2036 	 */
2037 	tba_taken_over = 1;
2038 
2039 	/* initialize MMU primary context register */
2040 	mmu_init_kcontext();
2041 
2042 	/*
2043 	 * The boot cpu can now take interrupts, x-calls, x-traps
2044 	 */
2045 	CPUSET_ADD(cpu_ready_set, CPU->cpu_id);
2046 	CPU->cpu_flags |= (CPU_READY | CPU_ENABLE | CPU_EXISTS);
2047 
2048 	/*
2049 	 * Set a flag to tell write_scb_int() that it can access V_TBR_WR_ADDR.
2050 	 */
2051 	tbr_wr_addr_inited = 1;
2052 
2053 	/*
2054 	 * Initialize VM system, and map kernel address space.
2055 	 */
2056 	kvm_init();
2057 
2058 	/*
2059 	 * XXX4U: previously, we initialized and turned on
2060 	 * the caches at this point. But of course we have
2061 	 * nothing to do, as the prom has already done this
2062 	 * for us -- main memory must be E$able at all times.
2063 	 */
2064 
2065 	/*
2066 	 * If the following is true, someone has patched
2067 	 * phsymem to be less than the number of pages that
2068 	 * the system actually has.  Remove pages until system
2069 	 * memory is limited to the requested amount.  Since we
2070 	 * have allocated page structures for all pages, we
2071 	 * correct the amount of memory we want to remove
2072 	 * by the size of the memory used to hold page structures
2073 	 * for the non-used pages.
2074 	 */
2075 	if (physmem < npages) {
2076 		pgcnt_t diff, off;
2077 		struct page *pp;
2078 		struct seg kseg;
2079 
2080 		cmn_err(CE_WARN, "limiting physmem to %ld pages", physmem);
2081 
2082 		off = 0;
2083 		diff = npages - physmem;
2084 		diff -= mmu_btopr(diff * sizeof (struct page));
2085 		kseg.s_as = &kas;
2086 		while (diff--) {
2087 			pp = page_create_va(&unused_pages_vp, (offset_t)off,
2088 			    MMU_PAGESIZE, PG_WAIT | PG_EXCL,
2089 			    &kseg, (caddr_t)off);
2090 			if (pp == NULL)
2091 				cmn_err(CE_PANIC, "limited physmem too much!");
2092 			page_io_unlock(pp);
2093 			page_downgrade(pp);
2094 			availrmem--;
2095 			off += MMU_PAGESIZE;
2096 		}
2097 	}
2098 
2099 	/*
2100 	 * When printing memory, show the total as physmem less
2101 	 * that stolen by a debugger.
2102 	 */
2103 	cmn_err(CE_CONT, "?mem = %ldK (0x%lx000)\n",
2104 	    (ulong_t)(physinstalled) << (PAGESHIFT - 10),
2105 	    (ulong_t)(physinstalled) << (PAGESHIFT - 12));
2106 
2107 	avmem = (uint64_t)freemem << PAGESHIFT;
2108 	cmn_err(CE_CONT, "?avail mem = %lld\n", (unsigned long long)avmem);
2109 
2110 	/*
2111 	 * For small memory systems disable automatic large pages.
2112 	 */
2113 	if (physmem < privm_lpg_min_physmem) {
2114 		use_brk_lpg = 0;
2115 		use_stk_lpg = 0;
2116 	}
2117 
2118 	/*
2119 	 * Perform platform specific freelist processing
2120 	 */
2121 	if (&plat_freelist_process) {
2122 		for (mnode = 0; mnode < max_mem_nodes; mnode++)
2123 			if (mem_node_config[mnode].exists)
2124 				plat_freelist_process(mnode);
2125 	}
2126 
2127 	/*
2128 	 * Initialize the segkp segment type.  We position it
2129 	 * after the configured tables and buffers (whose end
2130 	 * is given by econtig) and before V_WKBASE_ADDR.
2131 	 * Also in this area is segkmap (size SEGMAPSIZE).
2132 	 */
2133 
2134 	/* XXX - cache alignment? */
2135 	va = (caddr_t)SEGKPBASE;
2136 	ASSERT(((uintptr_t)va & PAGEOFFSET) == 0);
2137 
2138 	max_phys_segkp = (physmem * 2);
2139 
2140 	if (segkpsize < btop(SEGKPMINSIZE) || segkpsize > btop(SEGKPMAXSIZE)) {
2141 		segkpsize = btop(SEGKPDEFSIZE);
2142 		cmn_err(CE_WARN, "Illegal value for segkpsize. "
2143 		    "segkpsize has been reset to %ld pages", segkpsize);
2144 	}
2145 
2146 	i = ptob(MIN(segkpsize, max_phys_segkp));
2147 
2148 	rw_enter(&kas.a_lock, RW_WRITER);
2149 	if (seg_attach(&kas, va, i, segkp) < 0)
2150 		cmn_err(CE_PANIC, "startup: cannot attach segkp");
2151 	if (segkp_create(segkp) != 0)
2152 		cmn_err(CE_PANIC, "startup: segkp_create failed");
2153 	rw_exit(&kas.a_lock);
2154 
2155 	/*
2156 	 * kpm segment
2157 	 */
2158 	segmap_kpm = kpm_enable &&
2159 		segmap_kpm && PAGESIZE == MAXBSIZE;
2160 
2161 	if (kpm_enable) {
2162 		rw_enter(&kas.a_lock, RW_WRITER);
2163 
2164 		/*
2165 		 * The segkpm virtual range range is larger than the
2166 		 * actual physical memory size and also covers gaps in
2167 		 * the physical address range for the following reasons:
2168 		 * . keep conversion between segkpm and physical addresses
2169 		 *   simple, cheap and unambiguous.
2170 		 * . avoid extension/shrink of the the segkpm in case of DR.
2171 		 * . avoid complexity for handling of virtual addressed
2172 		 *   caches, segkpm and the regular mapping scheme must be
2173 		 *   kept in sync wrt. the virtual color of mapped pages.
2174 		 * Any accesses to virtual segkpm ranges not backed by
2175 		 * physical memory will fall through the memseg pfn hash
2176 		 * and will be handled in segkpm_fault.
2177 		 * Additional kpm_size spaces needed for vac alias prevention.
2178 		 */
2179 		if (seg_attach(&kas, kpm_vbase, kpm_size * vac_colors,
2180 		    segkpm) < 0)
2181 			cmn_err(CE_PANIC, "cannot attach segkpm");
2182 
2183 		b.prot = PROT_READ | PROT_WRITE;
2184 		b.nvcolors = shm_alignment >> MMU_PAGESHIFT;
2185 
2186 		if (segkpm_create(segkpm, (caddr_t)&b) != 0)
2187 			panic("segkpm_create segkpm");
2188 
2189 		rw_exit(&kas.a_lock);
2190 
2191 		mach_kpm_init();
2192 	}
2193 
2194 	if (!segzio_fromheap) {
2195 		size_t size;
2196 		size_t physmem_b = mmu_ptob(physmem);
2197 
2198 		/* size is in bytes, segziosize is in pages */
2199 		if (segziosize == 0) {
2200 			size = physmem_b;
2201 		} else {
2202 			size = mmu_ptob(segziosize);
2203 		}
2204 
2205 		if (size < SEGZIOMINSIZE) {
2206 			size = SEGZIOMINSIZE;
2207 		} else if (size > SEGZIOMAXSIZE) {
2208 			size = SEGZIOMAXSIZE;
2209 			/*
2210 			 * On 64-bit x86, we only have 2TB of KVA.  This exists
2211 			 * for parity with x86.
2212 			 *
2213 			 * SEGZIOMAXSIZE is capped at 512gb so that segzio
2214 			 * doesn't consume all of KVA.  However, if we have a
2215 			 * system that has more thant 512gb of physical memory,
2216 			 * we can actually consume about half of the difference
2217 			 * between 512gb and the rest of the available physical
2218 			 * memory.
2219 			 */
2220 			if (physmem_b > SEGZIOMAXSIZE) {
2221 				size += (physmem_b - SEGZIOMAXSIZE) / 2;
2222 		}
2223 		}
2224 		segziosize = mmu_btop(roundup(size, MMU_PAGESIZE));
2225 		/* put the base of the ZIO segment after the kpm segment */
2226 		segzio_base = kpm_vbase + (kpm_size * vac_colors);
2227 		PRM_DEBUG(segziosize);
2228 		PRM_DEBUG(segzio_base);
2229 
2230 		/*
2231 		 * On some platforms, kvm_init is called after the kpm
2232 		 * sizes have been determined.  On SPARC, kvm_init is called
2233 		 * before, so we have to attach the kzioseg after kvm is
2234 		 * initialized, otherwise we'll try to allocate from the boot
2235 		 * area since the kernel heap hasn't yet been configured.
2236 		 */
2237 		rw_enter(&kas.a_lock, RW_WRITER);
2238 
2239 		(void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2240 		    &kzioseg);
2241 		(void) segkmem_zio_create(&kzioseg);
2242 
2243 		/* create zio area covering new segment */
2244 		segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2245 
2246 		rw_exit(&kas.a_lock);
2247 	}
2248 
2249 
2250 	/*
2251 	 * Now create generic mapping segment.  This mapping
2252 	 * goes SEGMAPSIZE beyond SEGMAPBASE.  But if the total
2253 	 * virtual address is greater than the amount of free
2254 	 * memory that is available, then we trim back the
2255 	 * segment size to that amount
2256 	 */
2257 	va = (caddr_t)SEGMAPBASE;
2258 
2259 	/*
2260 	 * 1201049: segkmap base address must be MAXBSIZE aligned
2261 	 */
2262 	ASSERT(((uintptr_t)va & MAXBOFFSET) == 0);
2263 
2264 	/*
2265 	 * Set size of segmap to percentage of freemem at boot,
2266 	 * but stay within the allowable range
2267 	 * Note we take percentage  before converting from pages
2268 	 * to bytes to avoid an overflow on 32-bit kernels.
2269 	 */
2270 	i = mmu_ptob((freemem * segmap_percent) / 100);
2271 
2272 	if (i < MINMAPSIZE)
2273 		i = MINMAPSIZE;
2274 
2275 	if (i > MIN(SEGMAPSIZE, mmu_ptob(freemem)))
2276 		i = MIN(SEGMAPSIZE, mmu_ptob(freemem));
2277 
2278 	i &= MAXBMASK;	/* 1201049: segkmap size must be MAXBSIZE aligned */
2279 
2280 	rw_enter(&kas.a_lock, RW_WRITER);
2281 	if (seg_attach(&kas, va, i, segkmap) < 0)
2282 		cmn_err(CE_PANIC, "cannot attach segkmap");
2283 
2284 	a.prot = PROT_READ | PROT_WRITE;
2285 	a.shmsize = shm_alignment;
2286 	a.nfreelist = 0;	/* use segmap driver defaults */
2287 
2288 	if (segmap_create(segkmap, (caddr_t)&a) != 0)
2289 		panic("segmap_create segkmap");
2290 	rw_exit(&kas.a_lock);
2291 
2292 	segdev_init();
2293 }
2294 
2295 static void
2296 startup_end(void)
2297 {
2298 	if ((caddr_t)memlist > (caddr_t)memlist_end)
2299 		panic("memlist overflow 2");
2300 	memlist_free_block((caddr_t)memlist,
2301 	    ((caddr_t)memlist_end - (caddr_t)memlist));
2302 	memlist = NULL;
2303 
2304 	/* enable page_relocation since OBP is now done */
2305 	page_relocate_ready = 1;
2306 
2307 	/*
2308 	 * Perform tasks that get done after most of the VM
2309 	 * initialization has been done but before the clock
2310 	 * and other devices get started.
2311 	 */
2312 	kern_setup1();
2313 
2314 	/*
2315 	 * Intialize the VM arenas for allocating physically
2316 	 * contiguus memory chunk for interrupt queues snd
2317 	 * allocate/register boot cpu's queues, if any and
2318 	 * allocate dump buffer for sun4v systems to store
2319 	 * extra crash information during crash dump
2320 	 */
2321 	contig_mem_init();
2322 	mach_descrip_init();
2323 	cpu_intrq_setup(CPU);
2324 	cpu_intrq_register(CPU);
2325 	mach_htraptrace_setup(CPU->cpu_id);
2326 	mach_htraptrace_configure(CPU->cpu_id);
2327 	mach_dump_buffer_init();
2328 
2329 	/*
2330 	 * Initialize interrupt related stuff
2331 	 */
2332 	cpu_intr_alloc(CPU, NINTR_THREADS);
2333 
2334 	(void) splzs();			/* allow hi clock ints but not zs */
2335 
2336 	/*
2337 	 * Initialize errors.
2338 	 */
2339 	error_init();
2340 
2341 	/*
2342 	 * Note that we may have already used kernel bcopy before this
2343 	 * point - but if you really care about this, adb the use_hw_*
2344 	 * variables to 0 before rebooting.
2345 	 */
2346 	mach_hw_copy_limit();
2347 
2348 	/*
2349 	 * Install the "real" preemption guards before DDI services
2350 	 * are available.
2351 	 */
2352 	(void) prom_set_preprom(kern_preprom);
2353 	(void) prom_set_postprom(kern_postprom);
2354 	CPU->cpu_m.mutex_ready = 1;
2355 
2356 	/*
2357 	 * Initialize segnf (kernel support for non-faulting loads).
2358 	 */
2359 	segnf_init();
2360 
2361 	/*
2362 	 * Configure the root devinfo node.
2363 	 */
2364 	configure();		/* set up devices */
2365 	mach_cpu_halt_idle();
2366 }
2367 
2368 
2369 void
2370 post_startup(void)
2371 {
2372 #ifdef	PTL1_PANIC_DEBUG
2373 	extern void init_ptl1_thread(void);
2374 #endif	/* PTL1_PANIC_DEBUG */
2375 	extern void abort_sequence_init(void);
2376 
2377 	/*
2378 	 * Set the system wide, processor-specific flags to be passed
2379 	 * to userland via the aux vector for performance hints and
2380 	 * instruction set extensions.
2381 	 */
2382 	bind_hwcap();
2383 
2384 	/*
2385 	 * Startup memory scrubber (if any)
2386 	 */
2387 	mach_memscrub();
2388 
2389 	/*
2390 	 * Allocate soft interrupt to handle abort sequence.
2391 	 */
2392 	abort_sequence_init();
2393 
2394 	/*
2395 	 * Configure the rest of the system.
2396 	 * Perform forceloading tasks for /etc/system.
2397 	 */
2398 	(void) mod_sysctl(SYS_FORCELOAD, NULL);
2399 	/*
2400 	 * ON4.0: Force /proc module in until clock interrupt handle fixed
2401 	 * ON4.0: This must be fixed or restated in /etc/systems.
2402 	 */
2403 	(void) modload("fs", "procfs");
2404 
2405 	/* load machine class specific drivers */
2406 	load_mach_drivers();
2407 
2408 	/* load platform specific drivers */
2409 	if (&load_platform_drivers)
2410 		load_platform_drivers();
2411 
2412 	/* load vis simulation module, if we are running w/fpu off */
2413 	if (!fpu_exists) {
2414 		if (modload("misc", "vis") == -1)
2415 			halt("Can't load vis");
2416 	}
2417 
2418 	mach_fpras();
2419 
2420 	maxmem = freemem;
2421 
2422 #ifdef	PTL1_PANIC_DEBUG
2423 	init_ptl1_thread();
2424 #endif	/* PTL1_PANIC_DEBUG */
2425 
2426 	if (&cif_init)
2427 		cif_init();
2428 }
2429 
2430 #ifdef	PTL1_PANIC_DEBUG
2431 int		ptl1_panic_test = 0;
2432 int		ptl1_panic_xc_one_test = 0;
2433 int		ptl1_panic_xc_all_test = 0;
2434 int		ptl1_panic_xt_one_test = 0;
2435 int		ptl1_panic_xt_all_test = 0;
2436 kthread_id_t	ptl1_thread_p = NULL;
2437 kcondvar_t	ptl1_cv;
2438 kmutex_t	ptl1_mutex;
2439 int		ptl1_recurse_count_threshold = 0x40;
2440 int		ptl1_recurse_trap_threshold = 0x3d;
2441 extern void	ptl1_recurse(int, int);
2442 extern void	ptl1_panic_xt(int, int);
2443 
2444 /*
2445  * Called once per second by timeout() to wake up
2446  * the ptl1_panic thread to see if it should cause
2447  * a trap to the ptl1_panic() code.
2448  */
2449 /* ARGSUSED */
2450 static void
2451 ptl1_wakeup(void *arg)
2452 {
2453 	mutex_enter(&ptl1_mutex);
2454 	cv_signal(&ptl1_cv);
2455 	mutex_exit(&ptl1_mutex);
2456 }
2457 
2458 /*
2459  * ptl1_panic cross call function:
2460  *     Needed because xc_one() and xc_some() can pass
2461  *	64 bit args but ptl1_recurse() expects ints.
2462  */
2463 static void
2464 ptl1_panic_xc(void)
2465 {
2466 	ptl1_recurse(ptl1_recurse_count_threshold,
2467 	    ptl1_recurse_trap_threshold);
2468 }
2469 
2470 /*
2471  * The ptl1 thread waits for a global flag to be set
2472  * and uses the recurse thresholds to set the stack depth
2473  * to cause a ptl1_panic() directly via a call to ptl1_recurse
2474  * or indirectly via the cross call and cross trap functions.
2475  *
2476  * This is useful testing stack overflows and normal
2477  * ptl1_panic() states with a know stack frame.
2478  *
2479  * ptl1_recurse() is an asm function in ptl1_panic.s that
2480  * sets the {In, Local, Out, and Global} registers to a
2481  * know state on the stack and just prior to causing a
2482  * test ptl1_panic trap.
2483  */
2484 static void
2485 ptl1_thread(void)
2486 {
2487 	mutex_enter(&ptl1_mutex);
2488 	while (ptl1_thread_p) {
2489 		cpuset_t	other_cpus;
2490 		int		cpu_id;
2491 		int		my_cpu_id;
2492 		int		target_cpu_id;
2493 		int		target_found;
2494 
2495 		if (ptl1_panic_test) {
2496 			ptl1_recurse(ptl1_recurse_count_threshold,
2497 			    ptl1_recurse_trap_threshold);
2498 		}
2499 
2500 		/*
2501 		 * Find potential targets for x-call and x-trap,
2502 		 * if any exist while preempt is disabled we
2503 		 * start a ptl1_panic if requested via a
2504 		 * globals.
2505 		 */
2506 		kpreempt_disable();
2507 		my_cpu_id = CPU->cpu_id;
2508 		other_cpus = cpu_ready_set;
2509 		CPUSET_DEL(other_cpus, CPU->cpu_id);
2510 		target_found = 0;
2511 		if (!CPUSET_ISNULL(other_cpus)) {
2512 			/*
2513 			 * Pick the first one
2514 			 */
2515 			for (cpu_id = 0; cpu_id < NCPU; cpu_id++) {
2516 				if (cpu_id == my_cpu_id)
2517 					continue;
2518 
2519 				if (CPU_XCALL_READY(cpu_id)) {
2520 					target_cpu_id = cpu_id;
2521 					target_found = 1;
2522 					break;
2523 				}
2524 			}
2525 			ASSERT(target_found);
2526 
2527 			if (ptl1_panic_xc_one_test) {
2528 				xc_one(target_cpu_id,
2529 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2530 			}
2531 			if (ptl1_panic_xc_all_test) {
2532 				xc_some(other_cpus,
2533 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2534 			}
2535 			if (ptl1_panic_xt_one_test) {
2536 				xt_one(target_cpu_id,
2537 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2538 			}
2539 			if (ptl1_panic_xt_all_test) {
2540 				xt_some(other_cpus,
2541 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2542 			}
2543 		}
2544 		kpreempt_enable();
2545 		(void) timeout(ptl1_wakeup, NULL, hz);
2546 		(void) cv_wait(&ptl1_cv, &ptl1_mutex);
2547 	}
2548 	mutex_exit(&ptl1_mutex);
2549 }
2550 
2551 /*
2552  * Called during early startup to create the ptl1_thread
2553  */
2554 void
2555 init_ptl1_thread(void)
2556 {
2557 	ptl1_thread_p = thread_create(NULL, 0, ptl1_thread, NULL, 0,
2558 	    &p0, TS_RUN, 0);
2559 }
2560 #endif	/* PTL1_PANIC_DEBUG */
2561 
2562 
2563 /*
2564  * Add to a memory list.
2565  * start = start of new memory segment
2566  * len = length of new memory segment in bytes
2567  * memlistp = pointer to array of available memory segment structures
2568  * curmemlistp = memory list to which to add segment.
2569  */
2570 static void
2571 memlist_add(uint64_t start, uint64_t len, struct memlist **memlistp,
2572 	struct memlist **curmemlistp)
2573 {
2574 	struct memlist *new;
2575 
2576 	new = *memlistp;
2577 	new->address = start;
2578 	new->size = len;
2579 	*memlistp = new + 1;
2580 
2581 	memlist_insert(new, curmemlistp);
2582 }
2583 
2584 /*
2585  * In the case of architectures that support dynamic addition of
2586  * memory at run-time there are two cases where memsegs need to
2587  * be initialized and added to the memseg list.
2588  * 1) memsegs that are constructed at startup.
2589  * 2) memsegs that are constructed at run-time on
2590  *    hot-plug capable architectures.
2591  * This code was originally part of the function kphysm_init().
2592  */
2593 
2594 static void
2595 memseg_list_add(struct memseg *memsegp)
2596 {
2597 	struct memseg **prev_memsegp;
2598 	pgcnt_t num;
2599 
2600 	/* insert in memseg list, decreasing number of pages order */
2601 
2602 	num = MSEG_NPAGES(memsegp);
2603 
2604 	for (prev_memsegp = &memsegs; *prev_memsegp;
2605 	    prev_memsegp = &((*prev_memsegp)->next)) {
2606 		if (num > MSEG_NPAGES(*prev_memsegp))
2607 			break;
2608 	}
2609 
2610 	memsegp->next = *prev_memsegp;
2611 	*prev_memsegp = memsegp;
2612 
2613 	if (kpm_enable) {
2614 		memsegp->nextpa = (memsegp->next) ?
2615 			va_to_pa(memsegp->next) : MSEG_NULLPTR_PA;
2616 
2617 		if (prev_memsegp != &memsegs) {
2618 			struct memseg *msp;
2619 			msp = (struct memseg *)((caddr_t)prev_memsegp -
2620 				offsetof(struct memseg, next));
2621 			msp->nextpa = va_to_pa(memsegp);
2622 		} else {
2623 			memsegspa = va_to_pa(memsegs);
2624 		}
2625 	}
2626 }
2627 
2628 /*
2629  * PSM add_physmem_cb(). US-II and newer processors have some
2630  * flavor of the prefetch capability implemented. We exploit
2631  * this capability for optimum performance.
2632  */
2633 #define	PREFETCH_BYTES	64
2634 
2635 void
2636 add_physmem_cb(page_t *pp, pfn_t pnum)
2637 {
2638 	extern void	 prefetch_page_w(void *);
2639 
2640 	pp->p_pagenum = pnum;
2641 
2642 	/*
2643 	 * Prefetch one more page_t into E$. To prevent future
2644 	 * mishaps with the sizeof(page_t) changing on us, we
2645 	 * catch this on debug kernels if we can't bring in the
2646 	 * entire hpage with 2 PREFETCH_BYTES reads. See
2647 	 * also, sun4u/cpu/cpu_module.c
2648 	 */
2649 	/*LINTED*/
2650 	ASSERT(sizeof (page_t) <= 2*PREFETCH_BYTES);
2651 	prefetch_page_w((char *)pp);
2652 }
2653 
2654 /*
2655  * kphysm_init() tackles the problem of initializing physical memory.
2656  * The old startup made some assumptions about the kernel living in
2657  * physically contiguous space which is no longer valid.
2658  */
2659 static void
2660 kphysm_init(page_t *pp, struct memseg *memsegp, pgcnt_t npages,
2661 	uintptr_t kpm_pp, pgcnt_t kpm_npages)
2662 {
2663 	struct memlist	*pmem;
2664 	struct memseg	*msp;
2665 	pfn_t		 base;
2666 	pgcnt_t		 num;
2667 	pfn_t		 lastseg_pages_end = 0;
2668 	pgcnt_t		 nelem_used = 0;
2669 
2670 	ASSERT(page_hash != NULL && page_hashsz != 0);
2671 
2672 	msp = memsegp;
2673 	for (pmem = phys_avail; pmem && npages; pmem = pmem->next) {
2674 
2675 		/*
2676 		 * Build the memsegs entry
2677 		 */
2678 		num = btop(pmem->size);
2679 		if (num > npages)
2680 			num = npages;
2681 		npages -= num;
2682 		base = btop(pmem->address);
2683 
2684 		msp->pages = pp;
2685 		msp->epages = pp + num;
2686 		msp->pages_base = base;
2687 		msp->pages_end = base + num;
2688 
2689 		if (kpm_enable) {
2690 			pfn_t pbase_a;
2691 			pfn_t pend_a;
2692 			pfn_t prev_pend_a;
2693 			pgcnt_t	nelem;
2694 
2695 			msp->pagespa = va_to_pa(pp);
2696 			msp->epagespa = va_to_pa(pp + num);
2697 			pbase_a = kpmptop(ptokpmp(base));
2698 			pend_a = kpmptop(ptokpmp(base + num - 1)) + kpmpnpgs;
2699 			nelem = ptokpmp(pend_a - pbase_a);
2700 			msp->kpm_nkpmpgs = nelem;
2701 			msp->kpm_pbase = pbase_a;
2702 			if (lastseg_pages_end) {
2703 				/*
2704 				 * Assume phys_avail is in ascending order
2705 				 * of physical addresses.
2706 				 */
2707 				ASSERT(base + num > lastseg_pages_end);
2708 				prev_pend_a = kpmptop(
2709 				    ptokpmp(lastseg_pages_end - 1)) + kpmpnpgs;
2710 
2711 				if (prev_pend_a > pbase_a) {
2712 					/*
2713 					 * Overlap, more than one memseg may
2714 					 * point to the same kpm_page range.
2715 					 */
2716 					if (kpm_smallpages == 0) {
2717 						msp->kpm_pages =
2718 						    (kpm_page_t *)kpm_pp - 1;
2719 						kpm_pp = (uintptr_t)
2720 							((kpm_page_t *)kpm_pp
2721 							+ nelem - 1);
2722 					} else {
2723 						msp->kpm_spages =
2724 						    (kpm_spage_t *)kpm_pp - 1;
2725 						kpm_pp = (uintptr_t)
2726 							((kpm_spage_t *)kpm_pp
2727 							+ nelem - 1);
2728 					}
2729 					nelem_used += nelem - 1;
2730 
2731 				} else {
2732 					if (kpm_smallpages == 0) {
2733 						msp->kpm_pages =
2734 						    (kpm_page_t *)kpm_pp;
2735 						kpm_pp = (uintptr_t)
2736 							((kpm_page_t *)kpm_pp
2737 							+ nelem);
2738 					} else {
2739 						msp->kpm_spages =
2740 						    (kpm_spage_t *)kpm_pp;
2741 						kpm_pp = (uintptr_t)
2742 							((kpm_spage_t *)
2743 							kpm_pp + nelem);
2744 					}
2745 					nelem_used += nelem;
2746 				}
2747 
2748 			} else {
2749 				if (kpm_smallpages == 0) {
2750 					msp->kpm_pages = (kpm_page_t *)kpm_pp;
2751 					kpm_pp = (uintptr_t)
2752 						((kpm_page_t *)kpm_pp + nelem);
2753 				} else {
2754 					msp->kpm_spages = (kpm_spage_t *)kpm_pp;
2755 					kpm_pp = (uintptr_t)
2756 						((kpm_spage_t *)kpm_pp + nelem);
2757 				}
2758 				nelem_used = nelem;
2759 			}
2760 
2761 			if (nelem_used > kpm_npages)
2762 				panic("kphysm_init: kpm_pp overflow\n");
2763 
2764 			msp->kpm_pagespa = va_to_pa(msp->kpm_pages);
2765 			lastseg_pages_end = msp->pages_end;
2766 		}
2767 
2768 		memseg_list_add(msp);
2769 
2770 		/*
2771 		 * add_physmem() initializes the PSM part of the page
2772 		 * struct by calling the PSM back with add_physmem_cb().
2773 		 * In addition it coalesces pages into larger pages as
2774 		 * it initializes them.
2775 		 */
2776 		add_physmem(pp, num, base);
2777 		pp += num;
2778 		msp++;
2779 	}
2780 
2781 	build_pfn_hash();
2782 }
2783 
2784 /*
2785  * Kernel VM initialization.
2786  * Assumptions about kernel address space ordering:
2787  *	(1) gap (user space)
2788  *	(2) kernel text
2789  *	(3) kernel data/bss
2790  *	(4) gap
2791  *	(5) kernel data structures
2792  *	(6) gap
2793  *	(7) debugger (optional)
2794  *	(8) monitor
2795  *	(9) gap (possibly null)
2796  *	(10) dvma
2797  *	(11) devices
2798  */
2799 static void
2800 kvm_init(void)
2801 {
2802 	/*
2803 	 * Put the kernel segments in kernel address space.
2804 	 */
2805 	rw_enter(&kas.a_lock, RW_WRITER);
2806 	as_avlinit(&kas);
2807 
2808 	(void) seg_attach(&kas, (caddr_t)KERNELBASE,
2809 	    (size_t)(e_moddata - KERNELBASE), &ktextseg);
2810 	(void) segkmem_create(&ktextseg);
2811 
2812 	(void) seg_attach(&kas, (caddr_t)(KERNELBASE + MMU_PAGESIZE4M),
2813 	    (size_t)(MMU_PAGESIZE4M), &ktexthole);
2814 	(void) segkmem_create(&ktexthole);
2815 
2816 	(void) seg_attach(&kas, (caddr_t)valloc_base,
2817 	    (size_t)(econtig32 - valloc_base), &kvalloc);
2818 	(void) segkmem_create(&kvalloc);
2819 
2820 	if (kmem64_base) {
2821 	    (void) seg_attach(&kas, (caddr_t)kmem64_base,
2822 		(size_t)(kmem64_end - kmem64_base), &kmem64);
2823 	    (void) segkmem_create(&kmem64);
2824 	}
2825 
2826 	/*
2827 	 * We're about to map out /boot.  This is the beginning of the
2828 	 * system resource management transition. We can no longer
2829 	 * call into /boot for I/O or memory allocations.
2830 	 */
2831 	(void) seg_attach(&kas, kernelheap, ekernelheap - kernelheap, &kvseg);
2832 	(void) segkmem_create(&kvseg);
2833 	hblk_alloc_dynamic = 1;
2834 
2835 	/*
2836 	 * we need to preallocate pages for DR operations before enabling large
2837 	 * page kernel heap because of memseg_remap_init() hat_unload() hack.
2838 	 */
2839 	memseg_remap_init();
2840 
2841 	/* at this point we are ready to use large page heap */
2842 	segkmem_heap_lp_init();
2843 
2844 	(void) seg_attach(&kas, (caddr_t)SYSBASE32, SYSLIMIT32 - SYSBASE32,
2845 	    &kvseg32);
2846 	(void) segkmem_create(&kvseg32);
2847 
2848 	/*
2849 	 * Create a segment for the debugger.
2850 	 */
2851 	(void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2852 	(void) segkmem_create(&kdebugseg);
2853 
2854 	rw_exit(&kas.a_lock);
2855 }
2856 
2857 char obp_tte_str[] =
2858 	"h# %x constant MMU_PAGESHIFT "
2859 	"h# %x constant TTE8K "
2860 	"h# %x constant SFHME_SIZE "
2861 	"h# %x constant SFHME_TTE "
2862 	"h# %x constant HMEBLK_TAG "
2863 	"h# %x constant HMEBLK_NEXT "
2864 	"h# %x constant HMEBLK_MISC "
2865 	"h# %x constant HMEBLK_HME1 "
2866 	"h# %x constant NHMENTS "
2867 	"h# %x constant HBLK_SZMASK "
2868 	"h# %x constant HBLK_RANGE_SHIFT "
2869 	"h# %x constant HMEBP_HBLK "
2870 	"h# %x constant HMEBUCKET_SIZE "
2871 	"h# %x constant HTAG_SFMMUPSZ "
2872 	"h# %x constant HTAG_REHASHSZ "
2873 	"h# %x constant mmu_hashcnt "
2874 	"h# %p constant uhme_hash "
2875 	"h# %p constant khme_hash "
2876 	"h# %x constant UHMEHASH_SZ "
2877 	"h# %x constant KHMEHASH_SZ "
2878 	"h# %p constant KCONTEXT "
2879 	"h# %p constant KHATID "
2880 	"h# %x constant ASI_MEM "
2881 
2882 	": PHYS-X@ ( phys -- data ) "
2883 	"   ASI_MEM spacex@ "
2884 	"; "
2885 
2886 	": PHYS-W@ ( phys -- data ) "
2887 	"   ASI_MEM spacew@ "
2888 	"; "
2889 
2890 	": PHYS-L@ ( phys -- data ) "
2891 	"   ASI_MEM spaceL@ "
2892 	"; "
2893 
2894 	": TTE_PAGE_SHIFT ( ttesz -- hmeshift ) "
2895 	"   3 * MMU_PAGESHIFT + "
2896 	"; "
2897 
2898 	": TTE_IS_VALID ( ttep -- flag ) "
2899 	"   PHYS-X@ 0< "
2900 	"; "
2901 
2902 	": HME_HASH_SHIFT ( ttesz -- hmeshift ) "
2903 	"   dup TTE8K =  if "
2904 	"      drop HBLK_RANGE_SHIFT "
2905 	"   else "
2906 	"      TTE_PAGE_SHIFT "
2907 	"   then "
2908 	"; "
2909 
2910 	": HME_HASH_BSPAGE ( addr hmeshift -- bspage ) "
2911 	"   tuck >> swap MMU_PAGESHIFT - << "
2912 	"; "
2913 
2914 	": HME_HASH_FUNCTION ( sfmmup addr hmeshift -- hmebp ) "
2915 	"   >> over xor swap                    ( hash sfmmup ) "
2916 	"   KHATID <>  if                       ( hash ) "
2917 	"      UHMEHASH_SZ and                  ( bucket ) "
2918 	"      HMEBUCKET_SIZE * uhme_hash +     ( hmebp ) "
2919 	"   else                                ( hash ) "
2920 	"      KHMEHASH_SZ and                  ( bucket ) "
2921 	"      HMEBUCKET_SIZE * khme_hash +     ( hmebp ) "
2922 	"   then                                ( hmebp ) "
2923 	"; "
2924 
2925 	": HME_HASH_TABLE_SEARCH "
2926 	"       ( sfmmup hmebp hblktag --  sfmmup null | sfmmup hmeblkp ) "
2927 	"   >r hmebp_hblk + phys-x@ begin ( sfmmup hmeblkp ) ( r: hblktag ) "
2928 	"      dup if   		( sfmmup hmeblkp ) ( r: hblktag ) "
2929 	"         dup hmeblk_tag + phys-x@ r@ = if ( sfmmup hmeblkp )	  "
2930 	"	     dup hmeblk_tag + 8 + phys-x@ 2 pick = if		  "
2931 	"		  true 	( sfmmup hmeblkp true ) ( r: hblktag )	  "
2932 	"	     else						  "
2933 	"	     	  hmeblk_next + phys-x@ false 			  "
2934 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2935 	"	     then  						  "
2936 	"	  else							  "
2937 	"	     hmeblk_next + phys-x@ false 			  "
2938 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2939 	"	  then 							  "
2940 	"      else							  "
2941 	"         true 							  "
2942 	"      then  							  "
2943 	"   until r> drop 						  "
2944 	"; "
2945 
2946 	": HME_HASH_TAG ( sfmmup rehash addr -- hblktag ) "
2947 	"   over HME_HASH_SHIFT HME_HASH_BSPAGE      ( sfmmup rehash bspage ) "
2948 	"   HTAG_REHASHSZ << or nip		     ( hblktag ) "
2949 	"; "
2950 
2951 	": HBLK_TO_TTEP ( hmeblkp addr -- ttep ) "
2952 	"   over HMEBLK_MISC + PHYS-L@ HBLK_SZMASK and  ( hmeblkp addr ttesz ) "
2953 	"   TTE8K =  if                            ( hmeblkp addr ) "
2954 	"      MMU_PAGESHIFT >> NHMENTS 1- and     ( hmeblkp hme-index ) "
2955 	"   else                                   ( hmeblkp addr ) "
2956 	"      drop 0                              ( hmeblkp 0 ) "
2957 	"   then                                   ( hmeblkp hme-index ) "
2958 	"   SFHME_SIZE * + HMEBLK_HME1 +           ( hmep ) "
2959 	"   SFHME_TTE +                            ( ttep ) "
2960 	"; "
2961 
2962 	": unix-tte ( addr cnum -- false | tte-data true ) "
2963 	"    KCONTEXT = if                   ( addr ) "
2964 	"	KHATID                       ( addr khatid ) "
2965 	"    else                            ( addr ) "
2966 	"       drop false exit              ( false ) "
2967 	"    then "
2968 	"      ( addr khatid ) "
2969 	"      mmu_hashcnt 1+ 1  do           ( addr sfmmup ) "
2970 	"         2dup swap i HME_HASH_SHIFT  "
2971 					"( addr sfmmup sfmmup addr hmeshift ) "
2972 	"         HME_HASH_FUNCTION           ( addr sfmmup hmebp ) "
2973 	"         over i 4 pick               "
2974 				"( addr sfmmup hmebp sfmmup rehash addr ) "
2975 	"         HME_HASH_TAG                ( addr sfmmup hmebp hblktag ) "
2976 	"         HME_HASH_TABLE_SEARCH       "
2977 					"( addr sfmmup { null | hmeblkp } ) "
2978 	"         ?dup  if                    ( addr sfmmup hmeblkp ) "
2979 	"            nip swap HBLK_TO_TTEP    ( ttep ) "
2980 	"            dup TTE_IS_VALID  if     ( valid-ttep ) "
2981 	"               PHYS-X@ true          ( tte-data true ) "
2982 	"            else                     ( invalid-tte ) "
2983 	"               drop false            ( false ) "
2984 	"            then                     ( false | tte-data true ) "
2985 	"            unloop exit              ( false | tte-data true ) "
2986 	"         then                        ( addr sfmmup ) "
2987 	"      loop                           ( addr sfmmup ) "
2988 	"      2drop false                    ( false ) "
2989 	"; "
2990 ;
2991 
2992 void
2993 create_va_to_tte(void)
2994 {
2995 	char *bp;
2996 	extern int khmehash_num, uhmehash_num;
2997 	extern struct hmehash_bucket *khme_hash, *uhme_hash;
2998 
2999 #define	OFFSET(type, field)	((uintptr_t)(&((type *)0)->field))
3000 
3001 	bp = (char *)kobj_zalloc(MMU_PAGESIZE, KM_SLEEP);
3002 
3003 	/*
3004 	 * Teach obp how to parse our sw ttes.
3005 	 */
3006 	(void) sprintf(bp, obp_tte_str,
3007 	    MMU_PAGESHIFT,
3008 	    TTE8K,
3009 	    sizeof (struct sf_hment),
3010 	    OFFSET(struct sf_hment, hme_tte),
3011 	    OFFSET(struct hme_blk, hblk_tag),
3012 	    OFFSET(struct hme_blk, hblk_nextpa),
3013 	    OFFSET(struct hme_blk, hblk_misc),
3014 	    OFFSET(struct hme_blk, hblk_hme),
3015 	    NHMENTS,
3016 	    HBLK_SZMASK,
3017 	    HBLK_RANGE_SHIFT,
3018 	    OFFSET(struct hmehash_bucket, hmeh_nextpa),
3019 	    sizeof (struct hmehash_bucket),
3020 	    HTAG_SFMMUPSZ,
3021 	    HTAG_REHASHSZ,
3022 	    mmu_hashcnt,
3023 	    (caddr_t)va_to_pa((caddr_t)uhme_hash),
3024 	    (caddr_t)va_to_pa((caddr_t)khme_hash),
3025 	    UHMEHASH_SZ,
3026 	    KHMEHASH_SZ,
3027 	    KCONTEXT,
3028 	    KHATID,
3029 	    ASI_MEM);
3030 	prom_interpret(bp, 0, 0, 0, 0, 0);
3031 
3032 	kobj_free(bp, MMU_PAGESIZE);
3033 }
3034 
3035 void
3036 install_va_to_tte(void)
3037 {
3038 	/*
3039 	 * advise prom that he can use unix-tte
3040 	 */
3041 	prom_interpret("' unix-tte is va>tte-data", 0, 0, 0, 0, 0);
3042 }
3043 
3044 
3045 /*
3046  * Because kmdb links prom_stdout_is_framebuffer into its own
3047  * module, we add "device-type=display" here for /os-io node, so that
3048  * prom_stdout_is_framebuffer still works corrrectly  after /os-io node
3049  * is registered into OBP.
3050  */
3051 static char *create_node =
3052 	"\" /\" find-device "
3053 	"new-device "
3054 	"\" os-io\" device-name "
3055 	"\" display\" device-type "
3056 	": cb-r/w  ( adr,len method$ -- #read/#written ) "
3057 	"   2>r swap 2 2r> ['] $callback  catch  if "
3058 	"      2drop 3drop 0 "
3059 	"   then "
3060 	"; "
3061 	": read ( adr,len -- #read ) "
3062 	"       \" read\" ['] cb-r/w catch  if  2drop 2drop -2 exit then "
3063 	"       ( retN ... ret1 N ) "
3064 	"       ?dup  if "
3065 	"               swap >r 1-  0  ?do  drop  loop  r> "
3066 	"       else "
3067 	"               -2 "
3068 	"       then "
3069 	";    "
3070 	": write ( adr,len -- #written ) "
3071 	"       \" write\" ['] cb-r/w catch  if  2drop 2drop 0 exit  then "
3072 	"       ( retN ... ret1 N ) "
3073 	"       ?dup  if "
3074 	"               swap >r 1-  0  ?do  drop  loop  r> "
3075 	"        else "
3076 	"               0 "
3077 	"       then "
3078 	"; "
3079 	": poll-tty ( -- ) ; "
3080 	": install-abort  ( -- )  ['] poll-tty d# 10 alarm ; "
3081 	": remove-abort ( -- )  ['] poll-tty 0 alarm ; "
3082 	": cb-give/take ( $method -- ) "
3083 	"       0 -rot ['] $callback catch  ?dup  if "
3084 	"               >r 2drop 2drop r> throw "
3085 	"       else "
3086 	"               0  ?do  drop  loop "
3087 	"       then "
3088 	"; "
3089 	": give ( -- )  \" exit-input\" cb-give/take ; "
3090 	": take ( -- )  \" enter-input\" cb-give/take ; "
3091 	": open ( -- ok? )  true ; "
3092 	": close ( -- ) ; "
3093 	"finish-device "
3094 	"device-end ";
3095 
3096 /*
3097  * Create the OBP input/output node (FCode serial driver).
3098  * It is needed for both USB console keyboard and for
3099  * the kernel terminal emulator.  It is too early to check for a
3100  * kernel console compatible framebuffer now, so we create this
3101  * so that we're ready if we need to enable kernel terminal emulation.
3102  *
3103  * When the USB software takes over the input device at the time
3104  * consconfig runs, OBP's stdin is redirected to this node.
3105  * Whenever the FORTH user interface is used after this switch,
3106  * the node will call back into the kernel for console input.
3107  * If a serial device such as ttya or a UART with a Type 5 keyboard
3108  * attached is used, OBP takes over the serial device when the system
3109  * goes to the debugger after the system is booted.  This sharing
3110  * of the relatively simple serial device is difficult but possible.
3111  * Sharing the USB host controller is impossible due its complexity.
3112  *
3113  * Similarly to USB keyboard input redirection, after consconfig_dacf
3114  * configures a kernel console framebuffer as the standard output
3115  * device, OBP's stdout is switched to to vector through the
3116  * /os-io node into the kernel terminal emulator.
3117  */
3118 static void
3119 startup_create_io_node(void)
3120 {
3121 	prom_interpret(create_node, 0, 0, 0, 0, 0);
3122 }
3123 
3124 
3125 static void
3126 do_prom_version_check(void)
3127 {
3128 	int i;
3129 	pnode_t node;
3130 	char buf[64];
3131 	static char drev[] = "Down-rev firmware detected%s\n"
3132 		"\tPlease upgrade to the following minimum version:\n"
3133 		"\t\t%s\n";
3134 
3135 	i = prom_version_check(buf, sizeof (buf), &node);
3136 
3137 	if (i == PROM_VER64_OK)
3138 		return;
3139 
3140 	if (i == PROM_VER64_UPGRADE) {
3141 		cmn_err(CE_WARN, drev, "", buf);
3142 
3143 #ifdef	DEBUG
3144 		prom_enter_mon();	/* Type 'go' to continue */
3145 		cmn_err(CE_WARN, "Booting with down-rev firmware\n");
3146 		return;
3147 #else
3148 		halt(0);
3149 #endif
3150 	}
3151 
3152 	/*
3153 	 * The other possibility is that this is a server running
3154 	 * good firmware, but down-rev firmware was detected on at
3155 	 * least one other cpu board. We just complain if we see
3156 	 * that.
3157 	 */
3158 	cmn_err(CE_WARN, drev, " on one or more CPU boards", buf);
3159 }
3160 
3161 static void
3162 kpm_init()
3163 {
3164 	kpm_pgshft = (kpm_smallpages == 0) ? MMU_PAGESHIFT4M : MMU_PAGESHIFT;
3165 	kpm_pgsz = 1ull << kpm_pgshft;
3166 	kpm_pgoff = kpm_pgsz - 1;
3167 	kpmp2pshft = kpm_pgshft - PAGESHIFT;
3168 	kpmpnpgs = 1 << kpmp2pshft;
3169 	ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0);
3170 }
3171 
3172 void
3173 kpm_npages_setup(int memblocks)
3174 {
3175 	/*
3176 	 * npages can be scattered in a maximum of 'memblocks'
3177 	 */
3178 	kpm_npages = ptokpmpr(npages) + memblocks;
3179 }
3180 
3181 /*
3182  * Must be defined in platform dependent code.
3183  */
3184 extern caddr_t modtext;
3185 extern size_t modtext_sz;
3186 extern caddr_t moddata;
3187 
3188 #define	HEAPTEXT_ARENA(addr)	\
3189 	((uintptr_t)(addr) < KERNELBASE + 2 * MMU_PAGESIZE4M ? 0 : \
3190 	(((uintptr_t)(addr) - HEAPTEXT_BASE) / \
3191 	(HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) + 1))
3192 
3193 #define	HEAPTEXT_OVERSIZED(addr)	\
3194 	((uintptr_t)(addr) >= HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE)
3195 
3196 vmem_t *texthole_source[HEAPTEXT_NARENAS];
3197 vmem_t *texthole_arena[HEAPTEXT_NARENAS];
3198 kmutex_t texthole_lock;
3199 
3200 char kern_bootargs[OBP_MAXPATHLEN];
3201 
3202 void
3203 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3204 {
3205 	uintptr_t addr, limit;
3206 
3207 	addr = HEAPTEXT_BASE;
3208 	limit = addr + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE;
3209 
3210 	/*
3211 	 * Before we initialize the text_arena, we want to punch holes in the
3212 	 * underlying heaptext_arena.  This guarantees that for any text
3213 	 * address we can find a text hole less than HEAPTEXT_MAPPED away.
3214 	 */
3215 	for (; addr + HEAPTEXT_UNMAPPED <= limit;
3216 	    addr += HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) {
3217 		(void) vmem_xalloc(heaptext_arena, HEAPTEXT_UNMAPPED, PAGESIZE,
3218 		    0, 0, (void *)addr, (void *)(addr + HEAPTEXT_UNMAPPED),
3219 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3220 	}
3221 
3222 	/*
3223 	 * Allocate one page at the oversize to break up the text region
3224 	 * from the oversized region.
3225 	 */
3226 	(void) vmem_xalloc(heaptext_arena, PAGESIZE, PAGESIZE, 0, 0,
3227 	    (void *)limit, (void *)(limit + PAGESIZE),
3228 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3229 
3230 	*text_arena = vmem_create("module_text", modtext_sz ? modtext : NULL,
3231 	    modtext_sz, sizeof (uintptr_t), segkmem_alloc, segkmem_free,
3232 	    heaptext_arena, 0, VM_SLEEP);
3233 	*data_arena = vmem_create("module_data", moddata, MODDATA, 1,
3234 	    segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3235 }
3236 
3237 caddr_t
3238 kobj_text_alloc(vmem_t *arena, size_t size)
3239 {
3240 	caddr_t rval, better;
3241 
3242 	/*
3243 	 * First, try a sleeping allocation.
3244 	 */
3245 	rval = vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT);
3246 
3247 	if (size >= HEAPTEXT_MAPPED || !HEAPTEXT_OVERSIZED(rval))
3248 		return (rval);
3249 
3250 	/*
3251 	 * We didn't get the area that we wanted.  We're going to try to do an
3252 	 * allocation with explicit constraints.
3253 	 */
3254 	better = vmem_xalloc(arena, size, sizeof (uintptr_t), 0, 0, NULL,
3255 	    (void *)(HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE),
3256 	    VM_NOSLEEP | VM_BESTFIT);
3257 
3258 	if (better != NULL) {
3259 		/*
3260 		 * That worked.  Free our first attempt and return.
3261 		 */
3262 		vmem_free(arena, rval, size);
3263 		return (better);
3264 	}
3265 
3266 	/*
3267 	 * That didn't work; we'll have to return our first attempt.
3268 	 */
3269 	return (rval);
3270 }
3271 
3272 caddr_t
3273 kobj_texthole_alloc(caddr_t addr, size_t size)
3274 {
3275 	int arena = HEAPTEXT_ARENA(addr);
3276 	char c[30];
3277 	uintptr_t base;
3278 
3279 	if (HEAPTEXT_OVERSIZED(addr)) {
3280 		/*
3281 		 * If this is an oversized allocation, there is no text hole
3282 		 * available for it; return NULL.
3283 		 */
3284 		return (NULL);
3285 	}
3286 
3287 	mutex_enter(&texthole_lock);
3288 
3289 	if (texthole_arena[arena] == NULL) {
3290 		ASSERT(texthole_source[arena] == NULL);
3291 
3292 		if (arena == 0) {
3293 			texthole_source[0] = vmem_create("module_text_holesrc",
3294 			    (void *)(KERNELBASE + MMU_PAGESIZE4M),
3295 			    MMU_PAGESIZE4M, PAGESIZE, NULL, NULL, NULL,
3296 			    0, VM_SLEEP);
3297 		} else {
3298 			base = HEAPTEXT_BASE +
3299 			    (arena - 1) * (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED);
3300 
3301 			(void) snprintf(c, sizeof (c),
3302 			    "heaptext_holesrc_%d", arena);
3303 
3304 			texthole_source[arena] = vmem_create(c, (void *)base,
3305 			    HEAPTEXT_UNMAPPED, PAGESIZE, NULL, NULL, NULL,
3306 			    0, VM_SLEEP);
3307 		}
3308 
3309 		(void) snprintf(c, sizeof (c), "heaptext_hole_%d", arena);
3310 
3311 		texthole_arena[arena] = vmem_create(c, NULL, 0,
3312 		    sizeof (uint32_t), segkmem_alloc_permanent, segkmem_free,
3313 		    texthole_source[arena], 0, VM_SLEEP);
3314 	}
3315 
3316 	mutex_exit(&texthole_lock);
3317 
3318 	ASSERT(texthole_arena[arena] != NULL);
3319 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3320 	return (vmem_alloc(texthole_arena[arena], size,
3321 	    VM_BESTFIT | VM_NOSLEEP));
3322 }
3323 
3324 void
3325 kobj_texthole_free(caddr_t addr, size_t size)
3326 {
3327 	int arena = HEAPTEXT_ARENA(addr);
3328 
3329 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3330 	ASSERT(texthole_arena[arena] != NULL);
3331 	vmem_free(texthole_arena[arena], addr, size);
3332 }
3333