xref: /illumos-gate/usr/src/uts/sun4/os/startup.c (revision 08dcd69c259e91563982b9e7715366f99fee816e)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <sys/machsystm.h>
28 #include <sys/archsystm.h>
29 #include <sys/vm.h>
30 #include <sys/cpu.h>
31 #include <sys/atomic.h>
32 #include <sys/reboot.h>
33 #include <sys/kdi.h>
34 #include <sys/bootconf.h>
35 #include <sys/memlist_plat.h>
36 #include <sys/memlist_impl.h>
37 #include <sys/prom_plat.h>
38 #include <sys/prom_isa.h>
39 #include <sys/autoconf.h>
40 #include <sys/ivintr.h>
41 #include <sys/fpu/fpusystm.h>
42 #include <sys/iommutsb.h>
43 #include <vm/vm_dep.h>
44 #include <vm/seg_dev.h>
45 #include <vm/seg_kmem.h>
46 #include <vm/seg_kpm.h>
47 #include <vm/seg_map.h>
48 #include <vm/seg_kp.h>
49 #include <sys/sysconf.h>
50 #include <vm/hat_sfmmu.h>
51 #include <sys/kobj.h>
52 #include <sys/sun4asi.h>
53 #include <sys/clconf.h>
54 #include <sys/platform_module.h>
55 #include <sys/panic.h>
56 #include <sys/cpu_sgnblk_defs.h>
57 #include <sys/clock.h>
58 #include <sys/cmn_err.h>
59 #include <sys/promif.h>
60 #include <sys/prom_debug.h>
61 #include <sys/traptrace.h>
62 #include <sys/memnode.h>
63 #include <sys/mem_cage.h>
64 #include <sys/mmu.h>
65 
66 extern void setup_trap_table(void);
67 extern int cpu_intrq_setup(struct cpu *);
68 extern void cpu_intrq_register(struct cpu *);
69 extern void contig_mem_init(void);
70 extern caddr_t contig_mem_prealloc(caddr_t, pgcnt_t);
71 extern void mach_dump_buffer_init(void);
72 extern void mach_descrip_init(void);
73 extern void mach_descrip_startup_fini(void);
74 extern void mach_memscrub(void);
75 extern void mach_fpras(void);
76 extern void mach_cpu_halt_idle(void);
77 extern void mach_hw_copy_limit(void);
78 extern void load_mach_drivers(void);
79 extern void load_tod_module(void);
80 #pragma weak load_tod_module
81 
82 extern int ndata_alloc_mmfsa(struct memlist *ndata);
83 #pragma weak ndata_alloc_mmfsa
84 
85 extern void cif_init(void);
86 #pragma weak cif_init
87 
88 extern void parse_idprom(void);
89 extern void add_vx_handler(char *, int, void (*)(cell_t *));
90 extern void mem_config_init(void);
91 extern void memseg_remap_init(void);
92 
93 extern void mach_kpm_init(void);
94 extern void pcf_init();
95 extern int size_pse_array(pgcnt_t, int);
96 extern void pg_init();
97 
98 /*
99  * External Data:
100  */
101 extern int vac_size;	/* cache size in bytes */
102 extern uint_t vac_mask;	/* VAC alignment consistency mask */
103 extern uint_t vac_colors;
104 
105 /*
106  * Global Data Definitions:
107  */
108 
109 /*
110  * XXX - Don't port this to new architectures
111  * A 3rd party volume manager driver (vxdm) depends on the symbol romp.
112  * 'romp' has no use with a prom with an IEEE 1275 client interface.
113  * The driver doesn't use the value, but it depends on the symbol.
114  */
115 void *romp;		/* veritas driver won't load without romp 4154976 */
116 /*
117  * Declare these as initialized data so we can patch them.
118  */
119 pgcnt_t physmem = 0;	/* memory size in pages, patch if you want less */
120 pgcnt_t segkpsize =
121     btop(SEGKPDEFSIZE);	/* size of segkp segment in pages */
122 uint_t segmap_percent = 6; /* Size of segmap segment */
123 
124 int use_cache = 1;		/* cache not reliable (605 bugs) with MP */
125 int vac_copyback = 1;
126 char *cache_mode = NULL;
127 int use_mix = 1;
128 int prom_debug = 0;
129 
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 
151 size_t	ndata_remain_sz;	/* bytes from end of data to 4MB boundary */
152 caddr_t	nalloc_base;		/* beginning of nucleus allocation */
153 caddr_t nalloc_end;		/* end of nucleus allocatable memory */
154 caddr_t valloc_base;		/* beginning of kvalloc segment	*/
155 
156 caddr_t kmem64_base;		/* base of kernel mem segment in 64-bit space */
157 caddr_t kmem64_end;		/* end of kernel mem segment in 64-bit space */
158 size_t	kmem64_sz;		/* bytes in kernel mem segment, 64-bit space */
159 caddr_t kmem64_aligned_end;	/* end of large page, overmaps 64-bit space */
160 int	kmem64_szc;		/* page size code */
161 uint64_t kmem64_pabase = (uint64_t)-1;	/* physical address of kmem64_base */
162 
163 uintptr_t shm_alignment;	/* VAC address consistency modulus */
164 struct memlist *phys_install;	/* Total installed physical memory */
165 struct memlist *phys_avail;	/* Available (unreserved) physical memory */
166 struct memlist *virt_avail;	/* Available (unmapped?) virtual memory */
167 struct memlist *nopp_list;	/* pages with no backing page structs */
168 struct memlist ndata;		/* memlist of nucleus allocatable memory */
169 int memexp_flag;		/* memory expansion card flag */
170 uint64_t ecache_flushaddr;	/* physical address used for flushing E$ */
171 pgcnt_t obp_pages;		/* Physical pages used by OBP */
172 
173 /*
174  * VM data structures
175  */
176 long page_hashsz;		/* Size of page hash table (power of two) */
177 struct page *pp_base;		/* Base of system page struct array */
178 size_t pp_sz;			/* Size in bytes of page struct array */
179 struct page **page_hash;	/* Page hash table */
180 pad_mutex_t *pse_mutex;		/* Locks protecting pp->p_selock */
181 size_t pse_table_size;		/* Number of mutexes in pse_mutex[] */
182 int pse_shift;			/* log2(pse_table_size) */
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 void *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  * A static DR page_t VA map is reserved that can map the page structures
205  * for a domain's entire RA space. The pages that backs this space are
206  * dynamically allocated and need not be physically contiguous.  The DR
207  * map size is derived from KPM size.
208  */
209 int ppvm_enable = 0;		/* Static virtual map for page structs */
210 page_t *ppvm_base;		/* Base of page struct map */
211 pgcnt_t ppvm_size = 0;		/* Size of page struct map */
212 
213 /*
214  * debugger pages (if allocated)
215  */
216 struct vnode kdebugvp;
217 
218 /*
219  * VA range available to the debugger
220  */
221 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
222 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
223 
224 /*
225  * Segment for relocated kernel structures in 64-bit large RAM kernels
226  */
227 struct seg kmem64;
228 
229 struct memseg *memseg_free;
230 
231 struct vnode unused_pages_vp;
232 
233 /*
234  * VM data structures allocated early during boot.
235  */
236 size_t pagehash_sz;
237 uint64_t memlist_sz;
238 
239 char tbr_wr_addr_inited = 0;
240 
241 caddr_t	mpo_heap32_buf = NULL;
242 size_t	mpo_heap32_bufsz = 0;
243 
244 /*
245  * Static Routines:
246  */
247 static int ndata_alloc_memseg(struct memlist *, size_t);
248 static void memlist_new(uint64_t, uint64_t, struct memlist **);
249 static void memlist_add(uint64_t, uint64_t,
250 	struct memlist **, struct memlist **);
251 static void kphysm_init(void);
252 static void kvm_init(void);
253 static void install_kmem64_tte(void);
254 
255 static void startup_init(void);
256 static void startup_memlist(void);
257 static void startup_modules(void);
258 static void startup_bop_gone(void);
259 static void startup_vm(void);
260 static void startup_end(void);
261 static void setup_cage_params(void);
262 static void startup_create_io_node(void);
263 
264 static pgcnt_t npages;
265 static struct memlist *memlist;
266 void *memlist_end;
267 
268 static pgcnt_t bop_alloc_pages;
269 static caddr_t hblk_base;
270 uint_t hblk_alloc_dynamic = 0;
271 uint_t hblk1_min = H1MIN;
272 
273 
274 /*
275  * Hooks for unsupported platforms and down-rev firmware
276  */
277 int iam_positron(void);
278 #pragma weak iam_positron
279 static void do_prom_version_check(void);
280 
281 /*
282  * After receiving a thermal interrupt, this is the number of seconds
283  * to delay before shutting off the system, assuming
284  * shutdown fails.  Use /etc/system to change the delay if this isn't
285  * large enough.
286  */
287 int thermal_powerdown_delay = 1200;
288 
289 /*
290  * Used to hold off page relocations into the cage until OBP has completed
291  * its boot-time handoff of its resources to the kernel.
292  */
293 int page_relocate_ready = 0;
294 
295 /*
296  * Indicate if kmem64 allocation was done in small chunks
297  */
298 int kmem64_smchunks = 0;
299 
300 /*
301  * Enable some debugging messages concerning memory usage...
302  */
303 #ifdef  DEBUGGING_MEM
304 static int debugging_mem;
305 static void
306 printmemlist(char *title, struct memlist *list)
307 {
308 	if (!debugging_mem)
309 		return;
310 
311 	printf("%s\n", title);
312 
313 	while (list) {
314 		prom_printf("\taddr = 0x%x %8x, size = 0x%x %8x\n",
315 		    (uint32_t)(list->ml_address >> 32),
316 		    (uint32_t)list->ml_address,
317 		    (uint32_t)(list->ml_size >> 32),
318 		    (uint32_t)(list->ml_size));
319 		list = list->ml_next;
320 	}
321 }
322 
323 void
324 printmemseg(struct memseg *memseg)
325 {
326 	if (!debugging_mem)
327 		return;
328 
329 	printf("memseg\n");
330 
331 	while (memseg) {
332 		prom_printf("\tpage = 0x%p, epage = 0x%p, "
333 		    "pfn = 0x%x, epfn = 0x%x\n",
334 		    memseg->pages, memseg->epages,
335 		    memseg->pages_base, memseg->pages_end);
336 		memseg = memseg->next;
337 	}
338 }
339 
340 #define	debug_pause(str)	halt((str))
341 #define	MPRINTF(str)		if (debugging_mem) prom_printf((str))
342 #define	MPRINTF1(str, a)	if (debugging_mem) prom_printf((str), (a))
343 #define	MPRINTF2(str, a, b)	if (debugging_mem) prom_printf((str), (a), (b))
344 #define	MPRINTF3(str, a, b, c) \
345 	if (debugging_mem) prom_printf((str), (a), (b), (c))
346 #else	/* DEBUGGING_MEM */
347 #define	MPRINTF(str)
348 #define	MPRINTF1(str, a)
349 #define	MPRINTF2(str, a, b)
350 #define	MPRINTF3(str, a, b, c)
351 #endif	/* DEBUGGING_MEM */
352 
353 
354 /*
355  *
356  *                    Kernel's Virtual Memory Layout.
357  *                       /-----------------------\
358  * 0xFFFFFFFF.FFFFFFFF  -|                       |-
359  *                       |   OBP's virtual page  |
360  *                       |        tables         |
361  * 0xFFFFFFFC.00000000  -|-----------------------|-
362  *                       :                       :
363  *                       :                       :
364  *                      -|-----------------------|-
365  *                       |       segzio          | (base and size vary)
366  * 0xFFFFFE00.00000000  -|-----------------------|-
367  *                       |                       |  Ultrasparc I/II support
368  *                       |    segkpm segment     |  up to 2TB of physical
369  *                       | (64-bit kernel ONLY)  |  memory, VAC has 2 colors
370  *                       |                       |
371  * 0xFFFFFA00.00000000  -|-----------------------|- 2TB segkpm alignment
372  *                       :                       :
373  *                       :                       :
374  * 0xFFFFF810.00000000  -|-----------------------|- hole_end
375  *                       |                       |      ^
376  *                       |  UltraSPARC I/II call |      |
377  *                       | bug requires an extra |      |
378  *                       | 4 GB of space between |      |
379  *                       |   hole and used RAM   |	|
380  *                       |                       |      |
381  * 0xFFFFF800.00000000  -|-----------------------|-     |
382  *                       |                       |      |
383  *                       | Virtual Address Hole  |   UltraSPARC
384  *                       |  on UltraSPARC I/II   |  I/II * ONLY *
385  *                       |                       |      |
386  * 0x00000800.00000000  -|-----------------------|-     |
387  *                       |                       |      |
388  *                       |  UltraSPARC I/II call |      |
389  *                       | bug requires an extra |      |
390  *                       | 4 GB of space between |      |
391  *                       |   hole and used RAM   |      |
392  *                       |                       |      v
393  * 0x000007FF.00000000  -|-----------------------|- hole_start -----
394  *                       :                       :		   ^
395  *                       :                       :		   |
396  *                       |-----------------------|                 |
397  *                       |                       |                 |
398  *                       |  ecache flush area    |                 |
399  *                       |  (twice largest e$)   |                 |
400  *                       |                       |                 |
401  * 0x00000XXX.XXX00000  -|-----------------------|- kmem64_	   |
402  *                       | overmapped area       |   alignend_end  |
403  *                       | (kmem64_alignsize     |		   |
404  *                       |  boundary)            |		   |
405  * 0x00000XXX.XXXXXXXX  -|-----------------------|- kmem64_end	   |
406  *                       |                       |		   |
407  *                       |   64-bit kernel ONLY  |		   |
408  *                       |                       |		   |
409  *                       |    kmem64 segment     |		   |
410  *                       |                       |		   |
411  *                       | (Relocated extra HME  |	     Approximately
412  *                       |   block allocations,  |	    1 TB of virtual
413  *                       |   memnode freelists,  |	     address space
414  *                       |    HME hash buckets,  |		   |
415  *                       | mml_table, kpmp_table,|		   |
416  *                       |  page_t array and     |		   |
417  *                       |  hashblock pool to    |		   |
418  *                       |   avoid hard-coded    |		   |
419  *                       |     32-bit vaddr      |		   |
420  *                       |     limitations)      |		   |
421  *                       |                       |		   v
422  * 0x00000700.00000000  -|-----------------------|- SYSLIMIT (kmem64_base)
423  *                       |                       |
424  *                       |  segkmem segment      | (SYSLIMIT - SYSBASE = 4TB)
425  *                       |                       |
426  * 0x00000300.00000000  -|-----------------------|- SYSBASE
427  *                       :                       :
428  *                       :                       :
429  *                      -|-----------------------|-
430  *                       |                       |
431  *                       |  segmap segment       |   SEGMAPSIZE (1/8th physmem,
432  *                       |                       |               256G MAX)
433  * 0x000002a7.50000000  -|-----------------------|- SEGMAPBASE
434  *                       :                       :
435  *                       :                       :
436  *                      -|-----------------------|-
437  *                       |                       |
438  *                       |       segkp           |    SEGKPSIZE (2GB)
439  *                       |                       |
440  *                       |                       |
441  * 0x000002a1.00000000  -|-----------------------|- SEGKPBASE
442  *                       |                       |
443  * 0x000002a0.00000000  -|-----------------------|- MEMSCRUBBASE
444  *                       |                       |       (SEGKPBASE - 0x400000)
445  * 0x0000029F.FFE00000  -|-----------------------|- ARGSBASE
446  *                       |                       |       (MEMSCRUBBASE - NCARGS)
447  * 0x0000029F.FFD80000  -|-----------------------|- PPMAPBASE
448  *                       |                       |       (ARGSBASE - PPMAPSIZE)
449  * 0x0000029F.FFD00000  -|-----------------------|- PPMAP_FAST_BASE
450  *                       |                       |
451  * 0x0000029F.FF980000  -|-----------------------|- PIOMAPBASE
452  *                       |                       |
453  * 0x0000029F.FF580000  -|-----------------------|- NARG_BASE
454  *                       :                       :
455  *                       :                       :
456  * 0x00000000.FFFFFFFF  -|-----------------------|- OFW_END_ADDR
457  *                       |                       |
458  *                       |         OBP           |
459  *                       |                       |
460  * 0x00000000.F0000000  -|-----------------------|- OFW_START_ADDR
461  *                       |         kmdb          |
462  * 0x00000000.EDD00000  -|-----------------------|- SEGDEBUGBASE
463  *                       :                       :
464  *                       :                       :
465  * 0x00000000.7c000000  -|-----------------------|- SYSLIMIT32
466  *                       |                       |
467  *                       |  segkmem32 segment    | (SYSLIMIT32 - SYSBASE32 =
468  *                       |                       |    ~64MB)
469  *			-|-----------------------|
470  *			 |	IVSIZE		 |
471  * 0x00000000.70004000  -|-----------------------|
472  *                       |     panicbuf          |
473  * 0x00000000.70002000	-|-----------------------|
474  *			 |	PAGESIZE	 |
475  * 0x00000000.70000000  -|-----------------------|- SYSBASE32
476  *                       |       boot-time       |
477  *                       |    temporary space    |
478  * 0x00000000.4C000000  -|-----------------------|- BOOTTMPBASE
479  *                       :                       :
480  *                       :                       :
481  *                       |                       |
482  *                       |-----------------------|- econtig32
483  *                       |    vm structures      |
484  * 0x00000000.01C00000   |-----------------------|- nalloc_end
485  *                       |         TSBs          |
486  *                       |-----------------------|- end/nalloc_base
487  *                       |   kernel data & bss   |
488  * 0x00000000.01800000  -|-----------------------|
489  *                       :   nucleus text hole   :
490  * 0x00000000.01400000  -|-----------------------|
491  *                       :                       :
492  *                       |-----------------------|
493  *                       |      module text      |
494  *                       |-----------------------|- e_text/modtext
495  *                       |      kernel text      |
496  *                       |-----------------------|
497  *                       |    trap table (48k)   |
498  * 0x00000000.01000000  -|-----------------------|- KERNELBASE
499  *                       | reserved for trapstat |} TSTAT_TOTAL_SIZE
500  *                       |-----------------------|
501  *                       |                       |
502  *                       |        invalid        |
503  *                       |                       |
504  * 0x00000000.00000000  _|_______________________|
505  *
506  *
507  *
508  *                   32-bit User Virtual Memory Layout.
509  *                       /-----------------------\
510  *                       |                       |
511  *                       |        invalid        |
512  *                       |                       |
513  *          0xFFC00000  -|-----------------------|- USERLIMIT
514  *                       |       user stack      |
515  *                       :                       :
516  *                       :                       :
517  *                       :                       :
518  *                       |       user data       |
519  *                      -|-----------------------|-
520  *                       |       user text       |
521  *          0x00002000  -|-----------------------|-
522  *                       |       invalid         |
523  *          0x00000000  _|_______________________|
524  *
525  *
526  *
527  *                   64-bit User Virtual Memory Layout.
528  *                       /-----------------------\
529  *                       |                       |
530  *                       |        invalid        |
531  *                       |                       |
532  *  0xFFFFFFFF.80000000 -|-----------------------|- USERLIMIT
533  *                       |       user stack      |
534  *                       :                       :
535  *                       :                       :
536  *                       :                       :
537  *                       |       user data       |
538  *                      -|-----------------------|-
539  *                       |       user text       |
540  *  0x00000000.01000000 -|-----------------------|-
541  *                       |       invalid         |
542  *  0x00000000.00000000 _|_______________________|
543  */
544 
545 extern caddr_t ecache_init_scrub_flush_area(caddr_t alloc_base);
546 extern uint64_t ecache_flush_address(void);
547 
548 #pragma weak load_platform_modules
549 #pragma weak plat_startup_memlist
550 #pragma weak ecache_init_scrub_flush_area
551 #pragma weak ecache_flush_address
552 
553 
554 /*
555  * By default the DR Cage is enabled for maximum OS
556  * MPSS performance.  Users needing to disable the cage mechanism
557  * can set this variable to zero via /etc/system.
558  * Disabling the cage on systems supporting Dynamic Reconfiguration (DR)
559  * will result in loss of DR functionality.
560  * Platforms wishing to disable kernel Cage by default
561  * should do so in their set_platform_defaults() routine.
562  */
563 int	kernel_cage_enable = 1;
564 
565 static void
566 setup_cage_params(void)
567 {
568 	void (*func)(void);
569 
570 	func = (void (*)(void))kobj_getsymvalue("set_platform_cage_params", 0);
571 	if (func != NULL) {
572 		(*func)();
573 		return;
574 	}
575 
576 	if (kernel_cage_enable == 0) {
577 		return;
578 	}
579 	kcage_range_init(phys_avail, KCAGE_DOWN, total_pages / 256);
580 
581 	if (kcage_on) {
582 		cmn_err(CE_NOTE, "!Kernel Cage is ENABLED");
583 	} else {
584 		cmn_err(CE_NOTE, "!Kernel Cage is DISABLED");
585 	}
586 
587 }
588 
589 /*
590  * Machine-dependent startup code
591  */
592 void
593 startup(void)
594 {
595 	startup_init();
596 	if (&startup_platform)
597 		startup_platform();
598 	startup_memlist();
599 	startup_modules();
600 	setup_cage_params();
601 	startup_bop_gone();
602 	startup_vm();
603 	startup_end();
604 }
605 
606 struct regs sync_reg_buf;
607 uint64_t sync_tt;
608 
609 void
610 sync_handler(void)
611 {
612 	struct  panic_trap_info 	ti;
613 	int i;
614 
615 	/*
616 	 * Prevent trying to talk to the other CPUs since they are
617 	 * sitting in the prom and won't reply.
618 	 */
619 	for (i = 0; i < NCPU; i++) {
620 		if ((i != CPU->cpu_id) && CPU_XCALL_READY(i)) {
621 			cpu[i]->cpu_flags &= ~CPU_READY;
622 			cpu[i]->cpu_flags |= CPU_QUIESCED;
623 			CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
624 		}
625 	}
626 
627 	/*
628 	 * We've managed to get here without going through the
629 	 * normal panic code path. Try and save some useful
630 	 * information.
631 	 */
632 	if (!panicstr && (curthread->t_panic_trap == NULL)) {
633 		ti.trap_type = sync_tt;
634 		ti.trap_regs = &sync_reg_buf;
635 		ti.trap_addr = NULL;
636 		ti.trap_mmu_fsr = 0x0;
637 
638 		curthread->t_panic_trap = &ti;
639 	}
640 
641 	/*
642 	 * If we're re-entering the panic path, update the signature
643 	 * block so that the SC knows we're in the second part of panic.
644 	 */
645 	if (panicstr)
646 		CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
647 
648 	nopanicdebug = 1; /* do not perform debug_enter() prior to dump */
649 	panic("sync initiated");
650 }
651 
652 
653 static void
654 startup_init(void)
655 {
656 	/*
657 	 * We want to save the registers while we're still in OBP
658 	 * so that we know they haven't been fiddled with since.
659 	 * (In principle, OBP can't change them just because it
660 	 * makes a callback, but we'd rather not depend on that
661 	 * behavior.)
662 	 */
663 	char		sync_str[] =
664 	    "warning @ warning off : sync "
665 	    "%%tl-c %%tstate h# %p x! "
666 	    "%%g1 h# %p x! %%g2 h# %p x! %%g3 h# %p x! "
667 	    "%%g4 h# %p x! %%g5 h# %p x! %%g6 h# %p x! "
668 	    "%%g7 h# %p x! %%o0 h# %p x! %%o1 h# %p x! "
669 	    "%%o2 h# %p x! %%o3 h# %p x! %%o4 h# %p x! "
670 	    "%%o5 h# %p x! %%o6 h# %p x! %%o7 h# %p x! "
671 	    "%%tl-c %%tpc h# %p x! %%tl-c %%tnpc h# %p x! "
672 	    "%%y h# %p l! %%tl-c %%tt h# %p x! "
673 	    "sync ; warning !";
674 
675 	/*
676 	 * 20 == num of %p substrings
677 	 * 16 == max num of chars %p will expand to.
678 	 */
679 	char 		bp[sizeof (sync_str) + 16 * 20];
680 
681 	/*
682 	 * Initialize ptl1 stack for the 1st CPU.
683 	 */
684 	ptl1_init_cpu(&cpu0);
685 
686 	/*
687 	 * Initialize the address map for cache consistent mappings
688 	 * to random pages; must be done after vac_size is set.
689 	 */
690 	ppmapinit();
691 
692 	/*
693 	 * Initialize the PROM callback handler.
694 	 */
695 	init_vx_handler();
696 
697 	/*
698 	 * have prom call sync_callback() to handle the sync and
699 	 * save some useful information which will be stored in the
700 	 * core file later.
701 	 */
702 	(void) sprintf((char *)bp, sync_str,
703 	    (void *)&sync_reg_buf.r_tstate, (void *)&sync_reg_buf.r_g1,
704 	    (void *)&sync_reg_buf.r_g2, (void *)&sync_reg_buf.r_g3,
705 	    (void *)&sync_reg_buf.r_g4, (void *)&sync_reg_buf.r_g5,
706 	    (void *)&sync_reg_buf.r_g6, (void *)&sync_reg_buf.r_g7,
707 	    (void *)&sync_reg_buf.r_o0, (void *)&sync_reg_buf.r_o1,
708 	    (void *)&sync_reg_buf.r_o2, (void *)&sync_reg_buf.r_o3,
709 	    (void *)&sync_reg_buf.r_o4, (void *)&sync_reg_buf.r_o5,
710 	    (void *)&sync_reg_buf.r_o6, (void *)&sync_reg_buf.r_o7,
711 	    (void *)&sync_reg_buf.r_pc, (void *)&sync_reg_buf.r_npc,
712 	    (void *)&sync_reg_buf.r_y, (void *)&sync_tt);
713 	prom_interpret(bp, 0, 0, 0, 0, 0);
714 	add_vx_handler("sync", 1, (void (*)(cell_t *))sync_handler);
715 }
716 
717 
718 size_t
719 calc_pp_sz(pgcnt_t npages)
720 {
721 
722 	return (npages * sizeof (struct page));
723 }
724 
725 size_t
726 calc_kpmpp_sz(pgcnt_t npages)
727 {
728 
729 	kpm_pgshft = (kpm_smallpages == 0) ? MMU_PAGESHIFT4M : MMU_PAGESHIFT;
730 	kpm_pgsz = 1ull << kpm_pgshft;
731 	kpm_pgoff = kpm_pgsz - 1;
732 	kpmp2pshft = kpm_pgshft - PAGESHIFT;
733 	kpmpnpgs = 1 << kpmp2pshft;
734 
735 	if (kpm_smallpages == 0) {
736 		/*
737 		 * Avoid fragmentation problems in kphysm_init()
738 		 * by allocating for all of physical memory
739 		 */
740 		kpm_npages = ptokpmpr(physinstalled);
741 		return (kpm_npages * sizeof (kpm_page_t));
742 	} else {
743 		kpm_npages = npages;
744 		return (kpm_npages * sizeof (kpm_spage_t));
745 	}
746 }
747 
748 size_t
749 calc_pagehash_sz(pgcnt_t npages)
750 {
751 
752 	/*
753 	 * The page structure hash table size is a power of 2
754 	 * such that the average hash chain length is PAGE_HASHAVELEN.
755 	 */
756 	page_hashsz = npages / PAGE_HASHAVELEN;
757 	page_hashsz = 1 << highbit(page_hashsz);
758 	return (page_hashsz * sizeof (struct page *));
759 }
760 
761 int testkmem64_smchunks = 0;
762 
763 int
764 alloc_kmem64(caddr_t base, caddr_t end)
765 {
766 	int i;
767 	caddr_t aligned_end = NULL;
768 
769 	if (testkmem64_smchunks)
770 		return (1);
771 
772 	/*
773 	 * Make one large memory alloc after figuring out the 64-bit size. This
774 	 * will enable use of the largest page size appropriate for the system
775 	 * architecture.
776 	 */
777 	ASSERT(mmu_exported_pagesize_mask & (1 << TTE8K));
778 	ASSERT(IS_P2ALIGNED(base, TTEBYTES(max_bootlp_tteszc)));
779 	for (i = max_bootlp_tteszc; i >= TTE8K; i--) {
780 		size_t alloc_size, alignsize;
781 #if !defined(C_OBP)
782 		unsigned long long pa;
783 #endif	/* !C_OBP */
784 
785 		if ((mmu_exported_pagesize_mask & (1 << i)) == 0)
786 			continue;
787 		alignsize = TTEBYTES(i);
788 		kmem64_szc = i;
789 
790 		/* limit page size for small memory */
791 		if (mmu_btop(alignsize) > (npages >> 2))
792 			continue;
793 
794 		aligned_end = (caddr_t)roundup((uintptr_t)end, alignsize);
795 		alloc_size = aligned_end - base;
796 #if !defined(C_OBP)
797 		if (prom_allocate_phys(alloc_size, alignsize, &pa) == 0) {
798 			if (prom_claim_virt(alloc_size, base) != (caddr_t)-1) {
799 				kmem64_pabase = pa;
800 				kmem64_aligned_end = aligned_end;
801 				install_kmem64_tte();
802 				break;
803 			} else {
804 				prom_free_phys(alloc_size, pa);
805 			}
806 		}
807 #else	/* !C_OBP */
808 		if (prom_alloc(base, alloc_size, alignsize) == base) {
809 			kmem64_pabase = va_to_pa(kmem64_base);
810 			kmem64_aligned_end = aligned_end;
811 			break;
812 		}
813 #endif	/* !C_OBP */
814 		if (i == TTE8K) {
815 #ifdef sun4v
816 			/* return failure to try small allocations */
817 			return (1);
818 #else
819 			prom_panic("kmem64 allocation failure");
820 #endif
821 		}
822 	}
823 	ASSERT(aligned_end != NULL);
824 	return (0);
825 }
826 
827 static prom_memlist_t *boot_physinstalled, *boot_physavail, *boot_virtavail;
828 static size_t boot_physinstalled_len, boot_physavail_len, boot_virtavail_len;
829 
830 #if !defined(C_OBP)
831 /*
832  * Install a temporary tte handler in OBP for kmem64 area.
833  *
834  * We map kmem64 area with large pages before the trap table is taken
835  * over. Since OBP makes 8K mappings, it can create 8K tlb entries in
836  * the same area. Duplicate tlb entries with different page sizes
837  * cause unpredicatble behavior.  To avoid this, we don't create
838  * kmem64 mappings via BOP_ALLOC (ends up as prom_alloc() call to
839  * OBP).  Instead, we manage translations with a temporary va>tte-data
840  * handler (kmem64-tte).  This handler is replaced by unix-tte when
841  * the trap table is taken over.
842  *
843  * The temporary handler knows the physical address of the kmem64
844  * area. It uses the prom's pgmap@ Forth word for other addresses.
845  *
846  * We have to use BOP_ALLOC() method for C-OBP platforms because
847  * pgmap@ is not defined in C-OBP. C-OBP is only used on serengeti
848  * sun4u platforms. On sun4u we flush tlb after trap table is taken
849  * over if we use large pages for kernel heap and kmem64. Since sun4u
850  * prom (unlike sun4v) calls va>tte-data first for client address
851  * translation prom's ttes for kmem64 can't get into TLB even if we
852  * later switch to prom's trap table again. C-OBP uses 4M pages for
853  * client mappings when possible so on all platforms we get the
854  * benefit from large mappings for kmem64 area immediately during
855  * boot.
856  *
857  * pseudo code:
858  * if (context != 0) {
859  * 	return false
860  * } else if (miss_va in range[kmem64_base, kmem64_end)) {
861  *	tte = tte_template +
862  *		(((miss_va & pagemask) - kmem64_base));
863  *	return tte, true
864  * } else {
865  *	return pgmap@ result
866  * }
867  */
868 char kmem64_obp_str[] =
869 	"h# %lx constant kmem64-base "
870 	"h# %lx constant kmem64-end "
871 	"h# %lx constant kmem64-pagemask "
872 	"h# %lx constant kmem64-template "
873 
874 	": kmem64-tte ( addr cnum -- false | tte-data true ) "
875 	"    if                                       ( addr ) "
876 	"       drop false exit then                  ( false ) "
877 	"    dup  kmem64-base kmem64-end  within  if  ( addr ) "
878 	"	kmem64-pagemask and                   ( addr' ) "
879 	"	kmem64-base -                         ( addr' ) "
880 	"	kmem64-template +                     ( tte ) "
881 	"	true                                  ( tte true ) "
882 	"    else                                     ( addr ) "
883 	"	pgmap@                                ( tte ) "
884 	"       dup 0< if true else drop false then   ( tte true  |  false ) "
885 	"    then                                     ( tte true  |  false ) "
886 	"; "
887 
888 	"' kmem64-tte is va>tte-data "
889 ;
890 
891 static void
892 install_kmem64_tte()
893 {
894 	char b[sizeof (kmem64_obp_str) + (4 * 16)];
895 	tte_t tte;
896 
897 	PRM_DEBUG(kmem64_pabase);
898 	PRM_DEBUG(kmem64_szc);
899 	sfmmu_memtte(&tte, kmem64_pabase >> MMU_PAGESHIFT,
900 	    PROC_DATA | HAT_NOSYNC, kmem64_szc);
901 	PRM_DEBUG(tte.ll);
902 	(void) sprintf(b, kmem64_obp_str,
903 	    kmem64_base, kmem64_end, TTE_PAGEMASK(kmem64_szc), tte.ll);
904 	ASSERT(strlen(b) < sizeof (b));
905 	prom_interpret(b, 0, 0, 0, 0, 0);
906 }
907 #endif	/* !C_OBP */
908 
909 /*
910  * As OBP takes up some RAM when the system boots, pages will already be "lost"
911  * to the system and reflected in npages by the time we see it.
912  *
913  * We only want to allocate kernel structures in the 64-bit virtual address
914  * space on systems with enough RAM to make the overhead of keeping track of
915  * an extra kernel memory segment worthwhile.
916  *
917  * Since OBP has already performed its memory allocations by this point, if we
918  * have more than MINMOVE_RAM_MB MB of RAM left free, go ahead and map
919  * memory in the 64-bit virtual address space; otherwise keep allocations
920  * contiguous with we've mapped so far in the 32-bit virtual address space.
921  */
922 #define	MINMOVE_RAM_MB	((size_t)1900)
923 #define	MB_TO_BYTES(mb)	((mb) * 1048576ul)
924 #define	BYTES_TO_MB(b) ((b) / 1048576ul)
925 
926 pgcnt_t	tune_npages = (pgcnt_t)
927 	(MB_TO_BYTES(MINMOVE_RAM_MB)/ (size_t)MMU_PAGESIZE);
928 
929 #pragma weak page_set_colorequiv_arr_cpu
930 extern void page_set_colorequiv_arr_cpu(void);
931 extern void page_set_colorequiv_arr(void);
932 
933 static pgcnt_t ramdisk_npages;
934 static struct memlist *old_phys_avail;
935 
936 kcage_dir_t kcage_startup_dir = KCAGE_DOWN;
937 
938 static void
939 startup_memlist(void)
940 {
941 	size_t hmehash_sz, pagelist_sz, tt_sz;
942 	size_t psetable_sz;
943 	caddr_t alloc_base;
944 	caddr_t memspace;
945 	struct memlist *cur;
946 	size_t syslimit = (size_t)SYSLIMIT;
947 	size_t sysbase = (size_t)SYSBASE;
948 
949 	/*
950 	 * Initialize enough of the system to allow kmem_alloc to work by
951 	 * calling boot to allocate its memory until the time that
952 	 * kvm_init is completed.  The page structs are allocated after
953 	 * rounding up end to the nearest page boundary; the memsegs are
954 	 * initialized and the space they use comes from the kernel heap.
955 	 * With appropriate initialization, they can be reallocated later
956 	 * to a size appropriate for the machine's configuration.
957 	 *
958 	 * At this point, memory is allocated for things that will never
959 	 * need to be freed, this used to be "valloced".  This allows a
960 	 * savings as the pages don't need page structures to describe
961 	 * them because them will not be managed by the vm system.
962 	 */
963 
964 	/*
965 	 * We're loaded by boot with the following configuration (as
966 	 * specified in the sun4u/conf/Mapfile):
967 	 *
968 	 * 	text:		4 MB chunk aligned on a 4MB boundary
969 	 * 	data & bss:	4 MB chunk aligned on a 4MB boundary
970 	 *
971 	 * These two chunks will eventually be mapped by 2 locked 4MB
972 	 * ttes and will represent the nucleus of the kernel.  This gives
973 	 * us some free space that is already allocated, some or all of
974 	 * which is made available to kernel module text.
975 	 *
976 	 * The free space in the data-bss chunk is used for nucleus
977 	 * allocatable data structures and we reserve it using the
978 	 * nalloc_base and nalloc_end variables.  This space is currently
979 	 * being used for hat data structures required for tlb miss
980 	 * handling operations.  We align nalloc_base to a l2 cache
981 	 * linesize because this is the line size the hardware uses to
982 	 * maintain cache coherency.
983 	 * 512K is carved out for module data.
984 	 */
985 
986 	moddata = (caddr_t)roundup((uintptr_t)e_data, MMU_PAGESIZE);
987 	e_moddata = moddata + MODDATA;
988 	nalloc_base = e_moddata;
989 
990 	nalloc_end = (caddr_t)roundup((uintptr_t)nalloc_base, MMU_PAGESIZE4M);
991 	valloc_base = nalloc_base;
992 
993 	/*
994 	 * Calculate the start of the data segment.
995 	 */
996 	if (((uintptr_t)e_moddata & MMU_PAGEMASK4M) != (uintptr_t)s_data)
997 		prom_panic("nucleus data overflow");
998 
999 	PRM_DEBUG(moddata);
1000 	PRM_DEBUG(nalloc_base);
1001 	PRM_DEBUG(nalloc_end);
1002 
1003 	/*
1004 	 * Remember any slop after e_text so we can give it to the modules.
1005 	 */
1006 	PRM_DEBUG(e_text);
1007 	modtext = (caddr_t)roundup((uintptr_t)e_text, MMU_PAGESIZE);
1008 	if (((uintptr_t)e_text & MMU_PAGEMASK4M) != (uintptr_t)s_text)
1009 		prom_panic("nucleus text overflow");
1010 	modtext_sz = (caddr_t)roundup((uintptr_t)modtext, MMU_PAGESIZE4M) -
1011 	    modtext;
1012 	PRM_DEBUG(modtext);
1013 	PRM_DEBUG(modtext_sz);
1014 
1015 	init_boot_memlists();
1016 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1017 	    &boot_physavail, &boot_physavail_len,
1018 	    &boot_virtavail, &boot_virtavail_len);
1019 
1020 	/*
1021 	 * Remember what the physically available highest page is
1022 	 * so that dumpsys works properly, and find out how much
1023 	 * memory is installed.
1024 	 */
1025 	installed_top_size_memlist_array(boot_physinstalled,
1026 	    boot_physinstalled_len, &physmax, &physinstalled);
1027 	PRM_DEBUG(physinstalled);
1028 	PRM_DEBUG(physmax);
1029 
1030 	/* Fill out memory nodes config structure */
1031 	startup_build_mem_nodes(boot_physinstalled, boot_physinstalled_len);
1032 
1033 	/*
1034 	 * npages is the maximum of available physical memory possible.
1035 	 * (ie. it will never be more than this)
1036 	 *
1037 	 * When we boot from a ramdisk, the ramdisk memory isn't free, so
1038 	 * using phys_avail will underestimate what will end up being freed.
1039 	 * A better initial guess is just total memory minus the kernel text
1040 	 */
1041 	npages = physinstalled - btop(MMU_PAGESIZE4M);
1042 
1043 	/*
1044 	 * First allocate things that can go in the nucleus data page
1045 	 * (fault status, TSBs, dmv, CPUs)
1046 	 */
1047 	ndata_alloc_init(&ndata, (uintptr_t)nalloc_base, (uintptr_t)nalloc_end);
1048 
1049 	if ((&ndata_alloc_mmfsa != NULL) && (ndata_alloc_mmfsa(&ndata) != 0))
1050 		cmn_err(CE_PANIC, "no more nucleus memory after mfsa alloc");
1051 
1052 	if (ndata_alloc_tsbs(&ndata, npages) != 0)
1053 		cmn_err(CE_PANIC, "no more nucleus memory after tsbs alloc");
1054 
1055 	if (ndata_alloc_dmv(&ndata) != 0)
1056 		cmn_err(CE_PANIC, "no more nucleus memory after dmv alloc");
1057 
1058 	if (ndata_alloc_page_mutexs(&ndata) != 0)
1059 		cmn_err(CE_PANIC,
1060 		    "no more nucleus memory after page free lists alloc");
1061 
1062 	if (ndata_alloc_hat(&ndata, npages) != 0)
1063 		cmn_err(CE_PANIC, "no more nucleus memory after hat alloc");
1064 
1065 	if (ndata_alloc_memseg(&ndata, boot_physavail_len) != 0)
1066 		cmn_err(CE_PANIC, "no more nucleus memory after memseg alloc");
1067 
1068 	/*
1069 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1070 	 *
1071 	 * There are comments all over the SFMMU code warning of dire
1072 	 * consequences if the TSBs are moved out of 32-bit space.  This
1073 	 * is largely because the asm code uses "sethi %hi(addr)"-type
1074 	 * instructions which will not provide the expected result if the
1075 	 * address is a 64-bit one.
1076 	 *
1077 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1078 	 */
1079 	alloc_base = (caddr_t)roundup((uintptr_t)nalloc_end, MMU_PAGESIZE);
1080 	PRM_DEBUG(alloc_base);
1081 
1082 	alloc_base = sfmmu_ktsb_alloc(alloc_base);
1083 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1084 	PRM_DEBUG(alloc_base);
1085 
1086 	/*
1087 	 * Allocate IOMMU TSB array.  We do this here so that the physical
1088 	 * memory gets deducted from the PROM's physical memory list.
1089 	 */
1090 	alloc_base = iommu_tsb_init(alloc_base);
1091 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1092 	PRM_DEBUG(alloc_base);
1093 
1094 	/*
1095 	 * Allow for an early allocation of physically contiguous memory.
1096 	 */
1097 	alloc_base = contig_mem_prealloc(alloc_base, npages);
1098 
1099 	/*
1100 	 * Platforms like Starcat and OPL need special structures assigned in
1101 	 * 32-bit virtual address space because their probing routines execute
1102 	 * FCode, and FCode can't handle 64-bit virtual addresses...
1103 	 */
1104 	if (&plat_startup_memlist) {
1105 		alloc_base = plat_startup_memlist(alloc_base);
1106 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1107 		    ecache_alignsize);
1108 		PRM_DEBUG(alloc_base);
1109 	}
1110 
1111 	/*
1112 	 * Save off where the contiguous allocations to date have ended
1113 	 * in econtig32.
1114 	 */
1115 	econtig32 = alloc_base;
1116 	PRM_DEBUG(econtig32);
1117 	if (econtig32 > (caddr_t)KERNEL_LIMIT32)
1118 		cmn_err(CE_PANIC, "econtig32 too big");
1119 
1120 	pp_sz = calc_pp_sz(npages);
1121 	PRM_DEBUG(pp_sz);
1122 	if (kpm_enable) {
1123 		kpm_pp_sz = calc_kpmpp_sz(npages);
1124 		PRM_DEBUG(kpm_pp_sz);
1125 	}
1126 
1127 	hmehash_sz = calc_hmehash_sz(npages);
1128 	PRM_DEBUG(hmehash_sz);
1129 
1130 	pagehash_sz = calc_pagehash_sz(npages);
1131 	PRM_DEBUG(pagehash_sz);
1132 
1133 	pagelist_sz = calc_free_pagelist_sz();
1134 	PRM_DEBUG(pagelist_sz);
1135 
1136 #ifdef	TRAPTRACE
1137 	tt_sz = calc_traptrace_sz();
1138 	PRM_DEBUG(tt_sz);
1139 #else
1140 	tt_sz = 0;
1141 #endif	/* TRAPTRACE */
1142 
1143 	/*
1144 	 * Place the array that protects pp->p_selock in the kmem64 wad.
1145 	 */
1146 	pse_shift = size_pse_array(npages, max_ncpus);
1147 	PRM_DEBUG(pse_shift);
1148 	pse_table_size = 1 << pse_shift;
1149 	PRM_DEBUG(pse_table_size);
1150 	psetable_sz = roundup(
1151 	    pse_table_size * sizeof (pad_mutex_t), ecache_alignsize);
1152 	PRM_DEBUG(psetable_sz);
1153 
1154 	/*
1155 	 * Now allocate the whole wad
1156 	 */
1157 	kmem64_sz = pp_sz + kpm_pp_sz + hmehash_sz + pagehash_sz +
1158 	    pagelist_sz + tt_sz + psetable_sz;
1159 	kmem64_sz = roundup(kmem64_sz, PAGESIZE);
1160 	kmem64_base = (caddr_t)syslimit;
1161 	kmem64_end = kmem64_base + kmem64_sz;
1162 	if (alloc_kmem64(kmem64_base, kmem64_end)) {
1163 		/*
1164 		 * Attempt for kmem64 to allocate one big
1165 		 * contiguous chunk of memory failed.
1166 		 * We get here because we are sun4v.
1167 		 * We will proceed by breaking up
1168 		 * the allocation into two attempts.
1169 		 * First, we allocate kpm_pp_sz, hmehash_sz,
1170 		 * pagehash_sz, pagelist_sz, tt_sz & psetable_sz as
1171 		 * one contiguous chunk. This is a much smaller
1172 		 * chunk and we should get it, if not we panic.
1173 		 * Note that hmehash and tt need to be physically
1174 		 * (in the real address sense) contiguous.
1175 		 * Next, we use bop_alloc_chunk() to
1176 		 * to allocate the page_t structures.
1177 		 * This will allow the page_t to be allocated
1178 		 * in multiple smaller chunks.
1179 		 * In doing so, the assumption that page_t is
1180 		 * physically contiguous no longer hold, this is ok
1181 		 * for sun4v but not for sun4u.
1182 		 */
1183 		size_t  tmp_size;
1184 		caddr_t tmp_base;
1185 
1186 		pp_sz  = roundup(pp_sz, PAGESIZE);
1187 
1188 		/*
1189 		 * Allocate kpm_pp_sz, hmehash_sz,
1190 		 * pagehash_sz, pagelist_sz, tt_sz & psetable_sz
1191 		 */
1192 		tmp_base = kmem64_base + pp_sz;
1193 		tmp_size = roundup(kpm_pp_sz + hmehash_sz + pagehash_sz +
1194 		    pagelist_sz + tt_sz + psetable_sz, PAGESIZE);
1195 		if (prom_alloc(tmp_base, tmp_size, PAGESIZE) == 0)
1196 			prom_panic("kmem64 prom_alloc contig failed");
1197 		PRM_DEBUG(tmp_base);
1198 		PRM_DEBUG(tmp_size);
1199 
1200 		/*
1201 		 * Allocate the page_ts
1202 		 */
1203 		if (bop_alloc_chunk(kmem64_base, pp_sz, PAGESIZE) == 0)
1204 			prom_panic("kmem64 bop_alloc_chunk page_t failed");
1205 		PRM_DEBUG(kmem64_base);
1206 		PRM_DEBUG(pp_sz);
1207 
1208 		kmem64_aligned_end = kmem64_base + pp_sz + tmp_size;
1209 		ASSERT(kmem64_aligned_end >= kmem64_end);
1210 
1211 		kmem64_smchunks = 1;
1212 	} else {
1213 
1214 		/*
1215 		 * We need to adjust pp_sz for the normal
1216 		 * case where kmem64 can allocate one large chunk
1217 		 */
1218 		if (kpm_smallpages == 0) {
1219 			npages -= kmem64_sz / (PAGESIZE + sizeof (struct page));
1220 		} else {
1221 			npages -= kmem64_sz / (PAGESIZE + sizeof (struct page) +
1222 			    sizeof (kpm_spage_t));
1223 		}
1224 		pp_sz = npages * sizeof (struct page);
1225 	}
1226 
1227 	if (kmem64_aligned_end > (hole_start ? hole_start : kpm_vbase))
1228 		cmn_err(CE_PANIC, "not enough kmem64 space");
1229 	PRM_DEBUG(kmem64_base);
1230 	PRM_DEBUG(kmem64_end);
1231 	PRM_DEBUG(kmem64_aligned_end);
1232 
1233 	/*
1234 	 * ... and divy it up
1235 	 */
1236 	alloc_base = kmem64_base;
1237 
1238 	pp_base = (page_t *)alloc_base;
1239 	alloc_base += pp_sz;
1240 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1241 	PRM_DEBUG(pp_base);
1242 	PRM_DEBUG(npages);
1243 
1244 	if (kpm_enable) {
1245 		kpm_pp_base = alloc_base;
1246 		if (kpm_smallpages == 0) {
1247 			/* kpm_npages based on physinstalled, don't reset */
1248 			kpm_pp_sz = kpm_npages * sizeof (kpm_page_t);
1249 		} else {
1250 			kpm_npages = ptokpmpr(npages);
1251 			kpm_pp_sz = kpm_npages * sizeof (kpm_spage_t);
1252 		}
1253 		alloc_base += kpm_pp_sz;
1254 		alloc_base =
1255 		    (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1256 		PRM_DEBUG(kpm_pp_base);
1257 	}
1258 
1259 	alloc_base = alloc_hmehash(alloc_base);
1260 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1261 	PRM_DEBUG(alloc_base);
1262 
1263 	page_hash = (page_t **)alloc_base;
1264 	alloc_base += pagehash_sz;
1265 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1266 	PRM_DEBUG(page_hash);
1267 
1268 	alloc_base = alloc_page_freelists(alloc_base);
1269 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1270 	PRM_DEBUG(alloc_base);
1271 
1272 #ifdef	TRAPTRACE
1273 	ttrace_buf = alloc_base;
1274 	alloc_base += tt_sz;
1275 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1276 	PRM_DEBUG(alloc_base);
1277 #endif	/* TRAPTRACE */
1278 
1279 	pse_mutex = (pad_mutex_t *)alloc_base;
1280 	alloc_base += psetable_sz;
1281 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1282 	PRM_DEBUG(alloc_base);
1283 
1284 	/*
1285 	 * Note that if we use small chunk allocations for
1286 	 * kmem64, we need to ensure kmem64_end is the same as
1287 	 * kmem64_aligned_end to prevent subsequent logic from
1288 	 * trying to reuse the overmapping.
1289 	 * Otherwise we adjust kmem64_end to what we really allocated.
1290 	 */
1291 	if (kmem64_smchunks) {
1292 		kmem64_end = kmem64_aligned_end;
1293 	} else {
1294 		kmem64_end = (caddr_t)roundup((uintptr_t)alloc_base, PAGESIZE);
1295 	}
1296 	kmem64_sz = kmem64_end - kmem64_base;
1297 
1298 	if (&ecache_init_scrub_flush_area) {
1299 		alloc_base = ecache_init_scrub_flush_area(kmem64_aligned_end);
1300 		ASSERT(alloc_base <= (hole_start ? hole_start : kpm_vbase));
1301 	}
1302 
1303 	/*
1304 	 * If physmem is patched to be non-zero, use it instead of
1305 	 * the monitor value unless physmem is larger than the total
1306 	 * amount of memory on hand.
1307 	 */
1308 	if (physmem == 0 || physmem > npages)
1309 		physmem = npages;
1310 
1311 	/*
1312 	 * root_is_ramdisk is set via /etc/system when the ramdisk miniroot
1313 	 * is mounted as root. This memory is held down by OBP and unlike
1314 	 * the stub boot_archive is never released.
1315 	 *
1316 	 * In order to get things sized correctly on lower memory
1317 	 * machines (where the memory used by the ramdisk represents
1318 	 * a significant portion of memory), physmem is adjusted.
1319 	 *
1320 	 * This is done by subtracting the ramdisk_size which is set
1321 	 * to the size of the ramdisk (in Kb) in /etc/system at the
1322 	 * time the miniroot archive is constructed.
1323 	 */
1324 	if (root_is_ramdisk == B_TRUE) {
1325 		ramdisk_npages = (ramdisk_size * 1024) / PAGESIZE;
1326 		physmem -= ramdisk_npages;
1327 	}
1328 
1329 	if (kpm_enable && (ndata_alloc_kpm(&ndata, kpm_npages) != 0))
1330 		cmn_err(CE_PANIC, "no more nucleus memory after kpm alloc");
1331 
1332 	/*
1333 	 * Allocate space for the interrupt vector table.
1334 	 */
1335 	memspace = prom_alloc((caddr_t)intr_vec_table, IVSIZE, MMU_PAGESIZE);
1336 	if (memspace != (caddr_t)intr_vec_table)
1337 		prom_panic("interrupt vector table allocation failure");
1338 
1339 	/*
1340 	 * Between now and when we finish copying in the memory lists,
1341 	 * allocations happen so the space gets fragmented and the
1342 	 * lists longer.  Leave enough space for lists twice as
1343 	 * long as we have now; then roundup to a pagesize.
1344 	 */
1345 	memlist_sz = sizeof (struct memlist) * (prom_phys_installed_len() +
1346 	    prom_phys_avail_len() + prom_virt_avail_len());
1347 	memlist_sz *= 2;
1348 	memlist_sz = roundup(memlist_sz, PAGESIZE);
1349 	memspace = ndata_alloc(&ndata, memlist_sz, ecache_alignsize);
1350 	if (memspace == NULL)
1351 		cmn_err(CE_PANIC, "no more nucleus memory after memlist alloc");
1352 
1353 	memlist = (struct memlist *)memspace;
1354 	memlist_end = (char *)memspace + memlist_sz;
1355 	PRM_DEBUG(memlist);
1356 	PRM_DEBUG(memlist_end);
1357 
1358 	PRM_DEBUG(sysbase);
1359 	PRM_DEBUG(syslimit);
1360 	kernelheap_init((void *)sysbase, (void *)syslimit,
1361 	    (caddr_t)sysbase + PAGESIZE, NULL, NULL);
1362 
1363 	/*
1364 	 * Take the most current snapshot we can by calling mem-update.
1365 	 */
1366 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1367 	    &boot_physavail, &boot_physavail_len,
1368 	    &boot_virtavail, &boot_virtavail_len);
1369 
1370 	/*
1371 	 * Remove the space used by prom_alloc from the kernel heap
1372 	 * plus the area actually used by the OBP (if any)
1373 	 * ignoring virtual addresses in virt_avail, above syslimit.
1374 	 */
1375 	virt_avail = memlist;
1376 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1377 
1378 	for (cur = virt_avail; cur->ml_next; cur = cur->ml_next) {
1379 		uint64_t range_base, range_size;
1380 
1381 		if ((range_base = cur->ml_address + cur->ml_size) <
1382 		    (uint64_t)sysbase)
1383 			continue;
1384 		if (range_base >= (uint64_t)syslimit)
1385 			break;
1386 		/*
1387 		 * Limit the range to end at syslimit.
1388 		 */
1389 		range_size = MIN(cur->ml_next->ml_address,
1390 		    (uint64_t)syslimit) - range_base;
1391 		(void) vmem_xalloc(heap_arena, (size_t)range_size, PAGESIZE,
1392 		    0, 0, (void *)range_base, (void *)(range_base + range_size),
1393 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1394 	}
1395 
1396 	phys_avail = memlist;
1397 	copy_memlist(boot_physavail, boot_physavail_len, &memlist);
1398 
1399 	/*
1400 	 * Add any extra memory at the end of the ndata region if there's at
1401 	 * least a page to add.  There might be a few more pages available in
1402 	 * the middle of the ndata region, but for now they are ignored.
1403 	 */
1404 	nalloc_base = ndata_extra_base(&ndata, MMU_PAGESIZE, nalloc_end);
1405 	if (nalloc_base == NULL)
1406 		nalloc_base = nalloc_end;
1407 	ndata_remain_sz = nalloc_end - nalloc_base;
1408 
1409 	/*
1410 	 * Copy physinstalled list into kernel space.
1411 	 */
1412 	phys_install = memlist;
1413 	copy_memlist(boot_physinstalled, boot_physinstalled_len, &memlist);
1414 
1415 	/*
1416 	 * Create list of physical addrs we don't need pp's for:
1417 	 * kernel text 4M page
1418 	 * kernel data 4M page - ndata_remain_sz
1419 	 * kmem64 pages
1420 	 *
1421 	 * NB if adding any pages here, make sure no kpm page
1422 	 * overlaps can occur (see ASSERTs in kphysm_memsegs)
1423 	 */
1424 	nopp_list = memlist;
1425 	memlist_new(va_to_pa(s_text), MMU_PAGESIZE4M, &memlist);
1426 	memlist_add(va_to_pa(s_data), MMU_PAGESIZE4M - ndata_remain_sz,
1427 	    &memlist, &nopp_list);
1428 
1429 	/* Don't add to nopp_list if kmem64 was allocated in smchunks */
1430 	if (!kmem64_smchunks)
1431 		memlist_add(kmem64_pabase, kmem64_sz, &memlist, &nopp_list);
1432 
1433 	if ((caddr_t)memlist > (memspace + memlist_sz))
1434 		prom_panic("memlist overflow");
1435 
1436 	/*
1437 	 * Size the pcf array based on the number of cpus in the box at
1438 	 * boot time.
1439 	 */
1440 	pcf_init();
1441 
1442 	/*
1443 	 * Initialize the page structures from the memory lists.
1444 	 */
1445 	kphysm_init();
1446 
1447 	availrmem_initial = availrmem = freemem;
1448 	PRM_DEBUG(availrmem);
1449 
1450 	/*
1451 	 * Some of the locks depend on page_hashsz being set!
1452 	 * kmem_init() depends on this; so, keep it here.
1453 	 */
1454 	page_lock_init();
1455 
1456 	/*
1457 	 * Initialize kernel memory allocator.
1458 	 */
1459 	kmem_init();
1460 
1461 	/*
1462 	 * Factor in colorequiv to check additional 'equivalent' bins
1463 	 */
1464 	if (&page_set_colorequiv_arr_cpu != NULL)
1465 		page_set_colorequiv_arr_cpu();
1466 	else
1467 		page_set_colorequiv_arr();
1468 
1469 	/*
1470 	 * Initialize bp_mapin().
1471 	 */
1472 	bp_init(shm_alignment, HAT_STRICTORDER);
1473 
1474 	/*
1475 	 * Reserve space for MPO mblock structs from the 32-bit heap.
1476 	 */
1477 
1478 	if (mpo_heap32_bufsz > (size_t)0) {
1479 		(void) vmem_xalloc(heap32_arena, mpo_heap32_bufsz,
1480 		    PAGESIZE, 0, 0, mpo_heap32_buf,
1481 		    mpo_heap32_buf + mpo_heap32_bufsz,
1482 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1483 	}
1484 	mem_config_init();
1485 }
1486 
1487 static void
1488 startup_modules(void)
1489 {
1490 	int nhblk1, nhblk8;
1491 	size_t  nhblksz;
1492 	pgcnt_t pages_per_hblk;
1493 	size_t hme8blk_sz, hme1blk_sz;
1494 
1495 	/*
1496 	 * Let the platforms have a chance to change default
1497 	 * values before reading system file.
1498 	 */
1499 	if (&set_platform_defaults)
1500 		set_platform_defaults();
1501 
1502 	/*
1503 	 * Calculate default settings of system parameters based upon
1504 	 * maxusers, yet allow to be overridden via the /etc/system file.
1505 	 */
1506 	param_calc(0);
1507 
1508 	mod_setup();
1509 
1510 	/*
1511 	 * If this is a positron, complain and halt.
1512 	 */
1513 	if (&iam_positron && iam_positron()) {
1514 		cmn_err(CE_WARN, "This hardware platform is not supported"
1515 		    " by this release of Solaris.\n");
1516 #ifdef DEBUG
1517 		prom_enter_mon();	/* Type 'go' to resume */
1518 		cmn_err(CE_WARN, "Booting an unsupported platform.\n");
1519 		cmn_err(CE_WARN, "Booting with down-rev firmware.\n");
1520 
1521 #else /* DEBUG */
1522 		halt(0);
1523 #endif /* DEBUG */
1524 	}
1525 
1526 	/*
1527 	 * If we are running firmware that isn't 64-bit ready
1528 	 * then complain and halt.
1529 	 */
1530 	do_prom_version_check();
1531 
1532 	/*
1533 	 * Initialize system parameters
1534 	 */
1535 	param_init();
1536 
1537 	/*
1538 	 * maxmem is the amount of physical memory we're playing with.
1539 	 */
1540 	maxmem = physmem;
1541 
1542 	/* Set segkp limits. */
1543 	ncbase = kdi_segdebugbase;
1544 	ncend = kdi_segdebugbase;
1545 
1546 	/*
1547 	 * Initialize the hat layer.
1548 	 */
1549 	hat_init();
1550 
1551 	/*
1552 	 * Initialize segment management stuff.
1553 	 */
1554 	seg_init();
1555 
1556 	/*
1557 	 * Create the va>tte handler, so the prom can understand
1558 	 * kernel translations.  The handler is installed later, just
1559 	 * as we are about to take over the trap table from the prom.
1560 	 */
1561 	create_va_to_tte();
1562 
1563 	/*
1564 	 * Load the forthdebugger (optional)
1565 	 */
1566 	forthdebug_init();
1567 
1568 	/*
1569 	 * Create OBP node for console input callbacks
1570 	 * if it is needed.
1571 	 */
1572 	startup_create_io_node();
1573 
1574 	if (modloadonly("fs", "specfs") == -1)
1575 		halt("Can't load specfs");
1576 
1577 	if (modloadonly("fs", "devfs") == -1)
1578 		halt("Can't load devfs");
1579 
1580 	if (modloadonly("fs", "procfs") == -1)
1581 		halt("Can't load procfs");
1582 
1583 	if (modloadonly("misc", "swapgeneric") == -1)
1584 		halt("Can't load swapgeneric");
1585 
1586 	(void) modloadonly("sys", "lbl_edition");
1587 
1588 	dispinit();
1589 
1590 	/*
1591 	 * Infer meanings to the members of the idprom buffer.
1592 	 */
1593 	parse_idprom();
1594 
1595 	/* Read cluster configuration data. */
1596 	clconf_init();
1597 
1598 	setup_ddi();
1599 
1600 	/*
1601 	 * Lets take this opportunity to load the root device.
1602 	 */
1603 	if (loadrootmodules() != 0)
1604 		debug_enter("Can't load the root filesystem");
1605 
1606 	/*
1607 	 * Load tod driver module for the tod part found on this system.
1608 	 * Recompute the cpu frequency/delays based on tod as tod part
1609 	 * tends to keep time more accurately.
1610 	 */
1611 	if (&load_tod_module)
1612 		load_tod_module();
1613 
1614 	/*
1615 	 * Allow platforms to load modules which might
1616 	 * be needed after bootops are gone.
1617 	 */
1618 	if (&load_platform_modules)
1619 		load_platform_modules();
1620 
1621 	setcpudelay();
1622 
1623 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1624 	    &boot_physavail, &boot_physavail_len,
1625 	    &boot_virtavail, &boot_virtavail_len);
1626 
1627 	/*
1628 	 * Calculation and allocation of hmeblks needed to remap
1629 	 * the memory allocated by PROM till now.
1630 	 * Overestimate the number of hblk1 elements by assuming
1631 	 * worst case of TTE64K mappings.
1632 	 * sfmmu_hblk_alloc will panic if this calculation is wrong.
1633 	 */
1634 	bop_alloc_pages = btopr(kmem64_end - kmem64_base);
1635 	pages_per_hblk = btop(HMEBLK_SPAN(TTE64K));
1636 	bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1637 	nhblk1 = bop_alloc_pages / pages_per_hblk + hblk1_min;
1638 
1639 	bop_alloc_pages = size_virtalloc(boot_virtavail, boot_virtavail_len);
1640 
1641 	/* sfmmu_init_nucleus_hblks expects properly aligned data structures */
1642 	hme8blk_sz = roundup(HME8BLK_SZ, sizeof (int64_t));
1643 	hme1blk_sz = roundup(HME1BLK_SZ, sizeof (int64_t));
1644 
1645 	bop_alloc_pages += btopr(nhblk1 * hme1blk_sz);
1646 
1647 	pages_per_hblk = btop(HMEBLK_SPAN(TTE8K));
1648 	nhblk8 = 0;
1649 	while (bop_alloc_pages > 1) {
1650 		bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1651 		nhblk8 += bop_alloc_pages /= pages_per_hblk;
1652 		bop_alloc_pages *= hme8blk_sz;
1653 		bop_alloc_pages = btopr(bop_alloc_pages);
1654 	}
1655 	nhblk8 += 2;
1656 
1657 	/*
1658 	 * Since hblk8's can hold up to 64k of mappings aligned on a 64k
1659 	 * boundary, the number of hblk8's needed to map the entries in the
1660 	 * boot_virtavail list needs to be adjusted to take this into
1661 	 * consideration.  Thus, we need to add additional hblk8's since it
1662 	 * is possible that an hblk8 will not have all 8 slots used due to
1663 	 * alignment constraints.  Since there were boot_virtavail_len entries
1664 	 * in that list, we need to add that many hblk8's to the number
1665 	 * already calculated to make sure we don't underestimate.
1666 	 */
1667 	nhblk8 += boot_virtavail_len;
1668 	nhblksz = nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz;
1669 
1670 	/* Allocate in pagesize chunks */
1671 	nhblksz = roundup(nhblksz, MMU_PAGESIZE);
1672 	hblk_base = kmem_zalloc(nhblksz, KM_SLEEP);
1673 	sfmmu_init_nucleus_hblks(hblk_base, nhblksz, nhblk8, nhblk1);
1674 }
1675 
1676 static void
1677 startup_bop_gone(void)
1678 {
1679 
1680 	/*
1681 	 * Destroy the MD initialized at startup
1682 	 * The startup initializes the MD framework
1683 	 * using prom and BOP alloc free it now.
1684 	 */
1685 	mach_descrip_startup_fini();
1686 
1687 	/*
1688 	 * We're done with prom allocations.
1689 	 */
1690 	bop_fini();
1691 
1692 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1693 	    &boot_physavail, &boot_physavail_len,
1694 	    &boot_virtavail, &boot_virtavail_len);
1695 
1696 	/*
1697 	 * setup physically contiguous area twice as large as the ecache.
1698 	 * this is used while doing displacement flush of ecaches
1699 	 */
1700 	if (&ecache_flush_address) {
1701 		ecache_flushaddr = ecache_flush_address();
1702 		if (ecache_flushaddr == (uint64_t)-1) {
1703 			cmn_err(CE_PANIC,
1704 			    "startup: no memory to set ecache_flushaddr");
1705 		}
1706 	}
1707 
1708 	/*
1709 	 * Virtual available next.
1710 	 */
1711 	ASSERT(virt_avail != NULL);
1712 	memlist_free_list(virt_avail);
1713 	virt_avail = memlist;
1714 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1715 
1716 }
1717 
1718 
1719 /*
1720  * startup_fixup_physavail - called from mach_sfmmu.c after the final
1721  * allocations have been performed.  We can't call it in startup_bop_gone
1722  * since later operations can cause obp to allocate more memory.
1723  */
1724 void
1725 startup_fixup_physavail(void)
1726 {
1727 	struct memlist *cur;
1728 	size_t kmem64_overmap_size = kmem64_aligned_end - kmem64_end;
1729 
1730 	PRM_DEBUG(kmem64_overmap_size);
1731 
1732 	/*
1733 	 * take the most current snapshot we can by calling mem-update
1734 	 */
1735 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1736 	    &boot_physavail, &boot_physavail_len,
1737 	    &boot_virtavail, &boot_virtavail_len);
1738 
1739 	/*
1740 	 * Copy phys_avail list, again.
1741 	 * Both the kernel/boot and the prom have been allocating
1742 	 * from the original list we copied earlier.
1743 	 */
1744 	cur = memlist;
1745 	copy_memlist(boot_physavail, boot_physavail_len, &memlist);
1746 
1747 	/*
1748 	 * Add any unused kmem64 memory from overmapped page
1749 	 * (Note: va_to_pa does not work for kmem64_end)
1750 	 */
1751 	if (kmem64_overmap_size) {
1752 		memlist_add(kmem64_pabase + (kmem64_end - kmem64_base),
1753 		    kmem64_overmap_size, &memlist, &cur);
1754 	}
1755 
1756 	/*
1757 	 * Add any extra memory after e_data we added to the phys_avail list
1758 	 * back to the old list.
1759 	 */
1760 	if (ndata_remain_sz >= MMU_PAGESIZE)
1761 		memlist_add(va_to_pa(nalloc_base),
1762 		    (uint64_t)ndata_remain_sz, &memlist, &cur);
1763 
1764 	/*
1765 	 * There isn't any bounds checking on the memlist area
1766 	 * so ensure it hasn't overgrown.
1767 	 */
1768 	if ((caddr_t)memlist > (caddr_t)memlist_end)
1769 		cmn_err(CE_PANIC, "startup: memlist size exceeded");
1770 
1771 	/*
1772 	 * The kernel removes the pages that were allocated for it from
1773 	 * the freelist, but we now have to find any -extra- pages that
1774 	 * the prom has allocated for it's own book-keeping, and remove
1775 	 * them from the freelist too. sigh.
1776 	 */
1777 	sync_memlists(phys_avail, cur);
1778 
1779 	ASSERT(phys_avail != NULL);
1780 
1781 	old_phys_avail = phys_avail;
1782 	phys_avail = cur;
1783 }
1784 
1785 void
1786 update_kcage_ranges(uint64_t addr, uint64_t len)
1787 {
1788 	pfn_t base = btop(addr);
1789 	pgcnt_t num = btop(len);
1790 	int rv;
1791 
1792 	rv = kcage_range_add(base, num, kcage_startup_dir);
1793 
1794 	if (rv == ENOMEM) {
1795 		cmn_err(CE_WARN, "%ld megabytes not available to kernel cage",
1796 		    (len == 0 ? 0 : BYTES_TO_MB(len)));
1797 	} else if (rv != 0) {
1798 		/* catch this in debug kernels */
1799 		ASSERT(0);
1800 
1801 		cmn_err(CE_WARN, "unexpected kcage_range_add"
1802 		    " return value %d", rv);
1803 	}
1804 }
1805 
1806 static void
1807 startup_vm(void)
1808 {
1809 	size_t	i;
1810 	struct segmap_crargs a;
1811 	struct segkpm_crargs b;
1812 
1813 	uint64_t avmem;
1814 	caddr_t va;
1815 	pgcnt_t	max_phys_segkp;
1816 	int	mnode;
1817 
1818 	extern int use_brk_lpg, use_stk_lpg;
1819 
1820 	/*
1821 	 * get prom's mappings, create hments for them and switch
1822 	 * to the kernel context.
1823 	 */
1824 	hat_kern_setup();
1825 
1826 	/*
1827 	 * Take over trap table
1828 	 */
1829 	setup_trap_table();
1830 
1831 	/*
1832 	 * Install the va>tte handler, so that the prom can handle
1833 	 * misses and understand the kernel table layout in case
1834 	 * we need call into the prom.
1835 	 */
1836 	install_va_to_tte();
1837 
1838 	/*
1839 	 * Set a flag to indicate that the tba has been taken over.
1840 	 */
1841 	tba_taken_over = 1;
1842 
1843 	/* initialize MMU primary context register */
1844 	mmu_init_kcontext();
1845 
1846 	/*
1847 	 * The boot cpu can now take interrupts, x-calls, x-traps
1848 	 */
1849 	CPUSET_ADD(cpu_ready_set, CPU->cpu_id);
1850 	CPU->cpu_flags |= (CPU_READY | CPU_ENABLE | CPU_EXISTS);
1851 
1852 	/*
1853 	 * Set a flag to tell write_scb_int() that it can access V_TBR_WR_ADDR.
1854 	 */
1855 	tbr_wr_addr_inited = 1;
1856 
1857 	/*
1858 	 * Initialize VM system, and map kernel address space.
1859 	 */
1860 	kvm_init();
1861 
1862 	ASSERT(old_phys_avail != NULL && phys_avail != NULL);
1863 	if (kernel_cage_enable) {
1864 		diff_memlists(phys_avail, old_phys_avail, update_kcage_ranges);
1865 	}
1866 	memlist_free_list(old_phys_avail);
1867 
1868 	/*
1869 	 * If the following is true, someone has patched
1870 	 * phsymem to be less than the number of pages that
1871 	 * the system actually has.  Remove pages until system
1872 	 * memory is limited to the requested amount.  Since we
1873 	 * have allocated page structures for all pages, we
1874 	 * correct the amount of memory we want to remove
1875 	 * by the size of the memory used to hold page structures
1876 	 * for the non-used pages.
1877 	 */
1878 	if (physmem + ramdisk_npages < npages) {
1879 		pgcnt_t diff, off;
1880 		struct page *pp;
1881 		struct seg kseg;
1882 
1883 		cmn_err(CE_WARN, "limiting physmem to %ld pages", physmem);
1884 
1885 		off = 0;
1886 		diff = npages - (physmem + ramdisk_npages);
1887 		diff -= mmu_btopr(diff * sizeof (struct page));
1888 		kseg.s_as = &kas;
1889 		while (diff--) {
1890 			pp = page_create_va(&unused_pages_vp, (offset_t)off,
1891 			    MMU_PAGESIZE, PG_WAIT | PG_EXCL,
1892 			    &kseg, (caddr_t)off);
1893 			if (pp == NULL)
1894 				cmn_err(CE_PANIC, "limited physmem too much!");
1895 			page_io_unlock(pp);
1896 			page_downgrade(pp);
1897 			availrmem--;
1898 			off += MMU_PAGESIZE;
1899 		}
1900 	}
1901 
1902 	/*
1903 	 * When printing memory, show the total as physmem less
1904 	 * that stolen by a debugger.
1905 	 */
1906 	cmn_err(CE_CONT, "?mem = %ldK (0x%lx000)\n",
1907 	    (ulong_t)(physinstalled) << (PAGESHIFT - 10),
1908 	    (ulong_t)(physinstalled) << (PAGESHIFT - 12));
1909 
1910 	avmem = (uint64_t)freemem << PAGESHIFT;
1911 	cmn_err(CE_CONT, "?avail mem = %lld\n", (unsigned long long)avmem);
1912 
1913 	/*
1914 	 * For small memory systems disable automatic large pages.
1915 	 */
1916 	if (physmem < privm_lpg_min_physmem) {
1917 		use_brk_lpg = 0;
1918 		use_stk_lpg = 0;
1919 	}
1920 
1921 	/*
1922 	 * Perform platform specific freelist processing
1923 	 */
1924 	if (&plat_freelist_process) {
1925 		for (mnode = 0; mnode < max_mem_nodes; mnode++)
1926 			if (mem_node_config[mnode].exists)
1927 				plat_freelist_process(mnode);
1928 	}
1929 
1930 	/*
1931 	 * Initialize the segkp segment type.  We position it
1932 	 * after the configured tables and buffers (whose end
1933 	 * is given by econtig) and before V_WKBASE_ADDR.
1934 	 * Also in this area is segkmap (size SEGMAPSIZE).
1935 	 */
1936 
1937 	/* XXX - cache alignment? */
1938 	va = (caddr_t)SEGKPBASE;
1939 	ASSERT(((uintptr_t)va & PAGEOFFSET) == 0);
1940 
1941 	max_phys_segkp = (physmem * 2);
1942 
1943 	if (segkpsize < btop(SEGKPMINSIZE) || segkpsize > btop(SEGKPMAXSIZE)) {
1944 		segkpsize = btop(SEGKPDEFSIZE);
1945 		cmn_err(CE_WARN, "Illegal value for segkpsize. "
1946 		    "segkpsize has been reset to %ld pages", segkpsize);
1947 	}
1948 
1949 	i = ptob(MIN(segkpsize, max_phys_segkp));
1950 
1951 	rw_enter(&kas.a_lock, RW_WRITER);
1952 	if (seg_attach(&kas, va, i, segkp) < 0)
1953 		cmn_err(CE_PANIC, "startup: cannot attach segkp");
1954 	if (segkp_create(segkp) != 0)
1955 		cmn_err(CE_PANIC, "startup: segkp_create failed");
1956 	rw_exit(&kas.a_lock);
1957 
1958 	/*
1959 	 * kpm segment
1960 	 */
1961 	segmap_kpm = kpm_enable &&
1962 	    segmap_kpm && PAGESIZE == MAXBSIZE;
1963 
1964 	if (kpm_enable) {
1965 		rw_enter(&kas.a_lock, RW_WRITER);
1966 
1967 		/*
1968 		 * The segkpm virtual range range is larger than the
1969 		 * actual physical memory size and also covers gaps in
1970 		 * the physical address range for the following reasons:
1971 		 * . keep conversion between segkpm and physical addresses
1972 		 *   simple, cheap and unambiguous.
1973 		 * . avoid extension/shrink of the the segkpm in case of DR.
1974 		 * . avoid complexity for handling of virtual addressed
1975 		 *   caches, segkpm and the regular mapping scheme must be
1976 		 *   kept in sync wrt. the virtual color of mapped pages.
1977 		 * Any accesses to virtual segkpm ranges not backed by
1978 		 * physical memory will fall through the memseg pfn hash
1979 		 * and will be handled in segkpm_fault.
1980 		 * Additional kpm_size spaces needed for vac alias prevention.
1981 		 */
1982 		if (seg_attach(&kas, kpm_vbase, kpm_size * vac_colors,
1983 		    segkpm) < 0)
1984 			cmn_err(CE_PANIC, "cannot attach segkpm");
1985 
1986 		b.prot = PROT_READ | PROT_WRITE;
1987 		b.nvcolors = shm_alignment >> MMU_PAGESHIFT;
1988 
1989 		if (segkpm_create(segkpm, (caddr_t)&b) != 0)
1990 			panic("segkpm_create segkpm");
1991 
1992 		rw_exit(&kas.a_lock);
1993 
1994 		mach_kpm_init();
1995 	}
1996 
1997 	va = kpm_vbase + (kpm_size * vac_colors);
1998 
1999 	if (!segzio_fromheap) {
2000 		size_t size;
2001 		size_t physmem_b = mmu_ptob(physmem);
2002 
2003 		/* size is in bytes, segziosize is in pages */
2004 		if (segziosize == 0) {
2005 			size = physmem_b;
2006 		} else {
2007 			size = mmu_ptob(segziosize);
2008 		}
2009 
2010 		if (size < SEGZIOMINSIZE) {
2011 			size = SEGZIOMINSIZE;
2012 		} else if (size > SEGZIOMAXSIZE) {
2013 			size = SEGZIOMAXSIZE;
2014 			/*
2015 			 * On 64-bit x86, we only have 2TB of KVA.  This exists
2016 			 * for parity with x86.
2017 			 *
2018 			 * SEGZIOMAXSIZE is capped at 512gb so that segzio
2019 			 * doesn't consume all of KVA.  However, if we have a
2020 			 * system that has more thant 512gb of physical memory,
2021 			 * we can actually consume about half of the difference
2022 			 * between 512gb and the rest of the available physical
2023 			 * memory.
2024 			 */
2025 			if (physmem_b > SEGZIOMAXSIZE) {
2026 				size += (physmem_b - SEGZIOMAXSIZE) / 2;
2027 		}
2028 		}
2029 		segziosize = mmu_btop(roundup(size, MMU_PAGESIZE));
2030 		/* put the base of the ZIO segment after the kpm segment */
2031 		segzio_base = va;
2032 		va += mmu_ptob(segziosize);
2033 		PRM_DEBUG(segziosize);
2034 		PRM_DEBUG(segzio_base);
2035 
2036 		/*
2037 		 * On some platforms, kvm_init is called after the kpm
2038 		 * sizes have been determined.  On SPARC, kvm_init is called
2039 		 * before, so we have to attach the kzioseg after kvm is
2040 		 * initialized, otherwise we'll try to allocate from the boot
2041 		 * area since the kernel heap hasn't yet been configured.
2042 		 */
2043 		rw_enter(&kas.a_lock, RW_WRITER);
2044 
2045 		(void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2046 		    &kzioseg);
2047 		(void) segkmem_zio_create(&kzioseg);
2048 
2049 		/* create zio area covering new segment */
2050 		segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2051 
2052 		rw_exit(&kas.a_lock);
2053 	}
2054 
2055 	if (ppvm_enable) {
2056 		caddr_t ppvm_max;
2057 
2058 		/*
2059 		 * ppvm refers to the static VA space used to map
2060 		 * the page_t's for dynamically added memory.
2061 		 *
2062 		 * ppvm_base should not cross a potential VA hole.
2063 		 *
2064 		 * ppvm_size should be large enough to map the
2065 		 * page_t's needed to manage all of KPM range.
2066 		 */
2067 		ppvm_size =
2068 		    roundup(mmu_btop(kpm_size * vac_colors) * sizeof (page_t),
2069 		    MMU_PAGESIZE);
2070 		ppvm_max = (caddr_t)(0ull - ppvm_size);
2071 		ppvm_base = (page_t *)va;
2072 
2073 		if ((caddr_t)ppvm_base <= hole_end) {
2074 			cmn_err(CE_WARN,
2075 			    "Memory DR disabled: invalid DR map base: 0x%p\n",
2076 			    (void *)ppvm_base);
2077 			ppvm_enable = 0;
2078 		} else if ((caddr_t)ppvm_base > ppvm_max) {
2079 			uint64_t diff = (caddr_t)ppvm_base - ppvm_max;
2080 
2081 			cmn_err(CE_WARN,
2082 			    "Memory DR disabled: insufficient DR map size:"
2083 			    " 0x%lx (needed 0x%lx)\n",
2084 			    ppvm_size - diff, ppvm_size);
2085 			ppvm_enable = 0;
2086 		}
2087 		PRM_DEBUG(ppvm_size);
2088 		PRM_DEBUG(ppvm_base);
2089 	}
2090 
2091 	/*
2092 	 * Now create generic mapping segment.  This mapping
2093 	 * goes SEGMAPSIZE beyond SEGMAPBASE.  But if the total
2094 	 * virtual address is greater than the amount of free
2095 	 * memory that is available, then we trim back the
2096 	 * segment size to that amount
2097 	 */
2098 	va = (caddr_t)SEGMAPBASE;
2099 
2100 	/*
2101 	 * 1201049: segkmap base address must be MAXBSIZE aligned
2102 	 */
2103 	ASSERT(((uintptr_t)va & MAXBOFFSET) == 0);
2104 
2105 	/*
2106 	 * Set size of segmap to percentage of freemem at boot,
2107 	 * but stay within the allowable range
2108 	 * Note we take percentage  before converting from pages
2109 	 * to bytes to avoid an overflow on 32-bit kernels.
2110 	 */
2111 	i = mmu_ptob((freemem * segmap_percent) / 100);
2112 
2113 	if (i < MINMAPSIZE)
2114 		i = MINMAPSIZE;
2115 
2116 	if (i > MIN(SEGMAPSIZE, mmu_ptob(freemem)))
2117 		i = MIN(SEGMAPSIZE, mmu_ptob(freemem));
2118 
2119 	i &= MAXBMASK;	/* 1201049: segkmap size must be MAXBSIZE aligned */
2120 
2121 	rw_enter(&kas.a_lock, RW_WRITER);
2122 	if (seg_attach(&kas, va, i, segkmap) < 0)
2123 		cmn_err(CE_PANIC, "cannot attach segkmap");
2124 
2125 	a.prot = PROT_READ | PROT_WRITE;
2126 	a.shmsize = shm_alignment;
2127 	a.nfreelist = 0;	/* use segmap driver defaults */
2128 
2129 	if (segmap_create(segkmap, (caddr_t)&a) != 0)
2130 		panic("segmap_create segkmap");
2131 	rw_exit(&kas.a_lock);
2132 
2133 	segdev_init();
2134 }
2135 
2136 static void
2137 startup_end(void)
2138 {
2139 	if ((caddr_t)memlist > (caddr_t)memlist_end)
2140 		panic("memlist overflow 2");
2141 	memlist_free_block((caddr_t)memlist,
2142 	    ((caddr_t)memlist_end - (caddr_t)memlist));
2143 	memlist = NULL;
2144 
2145 	/* enable page_relocation since OBP is now done */
2146 	page_relocate_ready = 1;
2147 
2148 	/*
2149 	 * Perform tasks that get done after most of the VM
2150 	 * initialization has been done but before the clock
2151 	 * and other devices get started.
2152 	 */
2153 	kern_setup1();
2154 
2155 	/*
2156 	 * Perform CPC initialization for this CPU.
2157 	 */
2158 	kcpc_hw_init();
2159 
2160 	/*
2161 	 * Intialize the VM arenas for allocating physically
2162 	 * contiguus memory chunk for interrupt queues snd
2163 	 * allocate/register boot cpu's queues, if any and
2164 	 * allocate dump buffer for sun4v systems to store
2165 	 * extra crash information during crash dump
2166 	 */
2167 	contig_mem_init();
2168 	mach_descrip_init();
2169 
2170 	if (cpu_intrq_setup(CPU)) {
2171 		cmn_err(CE_PANIC, "cpu%d: setup failed", CPU->cpu_id);
2172 	}
2173 	cpu_intrq_register(CPU);
2174 	mach_htraptrace_setup(CPU->cpu_id);
2175 	mach_htraptrace_configure(CPU->cpu_id);
2176 	mach_dump_buffer_init();
2177 
2178 	/*
2179 	 * Initialize interrupt related stuff
2180 	 */
2181 	cpu_intr_alloc(CPU, NINTR_THREADS);
2182 
2183 	(void) splzs();			/* allow hi clock ints but not zs */
2184 
2185 	/*
2186 	 * Initialize errors.
2187 	 */
2188 	error_init();
2189 
2190 	/*
2191 	 * Note that we may have already used kernel bcopy before this
2192 	 * point - but if you really care about this, adb the use_hw_*
2193 	 * variables to 0 before rebooting.
2194 	 */
2195 	mach_hw_copy_limit();
2196 
2197 	/*
2198 	 * Install the "real" preemption guards before DDI services
2199 	 * are available.
2200 	 */
2201 	(void) prom_set_preprom(kern_preprom);
2202 	(void) prom_set_postprom(kern_postprom);
2203 	CPU->cpu_m.mutex_ready = 1;
2204 
2205 	/*
2206 	 * Initialize segnf (kernel support for non-faulting loads).
2207 	 */
2208 	segnf_init();
2209 
2210 	/*
2211 	 * Configure the root devinfo node.
2212 	 */
2213 	configure();		/* set up devices */
2214 	mach_cpu_halt_idle();
2215 }
2216 
2217 
2218 void
2219 post_startup(void)
2220 {
2221 #ifdef	PTL1_PANIC_DEBUG
2222 	extern void init_ptl1_thread(void);
2223 #endif	/* PTL1_PANIC_DEBUG */
2224 	extern void abort_sequence_init(void);
2225 
2226 	/*
2227 	 * Set the system wide, processor-specific flags to be passed
2228 	 * to userland via the aux vector for performance hints and
2229 	 * instruction set extensions.
2230 	 */
2231 	bind_hwcap();
2232 
2233 	/*
2234 	 * Startup memory scrubber (if any)
2235 	 */
2236 	mach_memscrub();
2237 
2238 	/*
2239 	 * Allocate soft interrupt to handle abort sequence.
2240 	 */
2241 	abort_sequence_init();
2242 
2243 	/*
2244 	 * Configure the rest of the system.
2245 	 * Perform forceloading tasks for /etc/system.
2246 	 */
2247 	(void) mod_sysctl(SYS_FORCELOAD, NULL);
2248 	/*
2249 	 * ON4.0: Force /proc module in until clock interrupt handle fixed
2250 	 * ON4.0: This must be fixed or restated in /etc/systems.
2251 	 */
2252 	(void) modload("fs", "procfs");
2253 
2254 	/* load machine class specific drivers */
2255 	load_mach_drivers();
2256 
2257 	/* load platform specific drivers */
2258 	if (&load_platform_drivers)
2259 		load_platform_drivers();
2260 
2261 	/* load vis simulation module, if we are running w/fpu off */
2262 	if (!fpu_exists) {
2263 		if (modload("misc", "vis") == -1)
2264 			halt("Can't load vis");
2265 	}
2266 
2267 	mach_fpras();
2268 
2269 	maxmem = freemem;
2270 
2271 	pg_init();
2272 
2273 #ifdef	PTL1_PANIC_DEBUG
2274 	init_ptl1_thread();
2275 #endif	/* PTL1_PANIC_DEBUG */
2276 }
2277 
2278 #ifdef	PTL1_PANIC_DEBUG
2279 int		ptl1_panic_test = 0;
2280 int		ptl1_panic_xc_one_test = 0;
2281 int		ptl1_panic_xc_all_test = 0;
2282 int		ptl1_panic_xt_one_test = 0;
2283 int		ptl1_panic_xt_all_test = 0;
2284 kthread_id_t	ptl1_thread_p = NULL;
2285 kcondvar_t	ptl1_cv;
2286 kmutex_t	ptl1_mutex;
2287 int		ptl1_recurse_count_threshold = 0x40;
2288 int		ptl1_recurse_trap_threshold = 0x3d;
2289 extern void	ptl1_recurse(int, int);
2290 extern void	ptl1_panic_xt(int, int);
2291 
2292 /*
2293  * Called once per second by timeout() to wake up
2294  * the ptl1_panic thread to see if it should cause
2295  * a trap to the ptl1_panic() code.
2296  */
2297 /* ARGSUSED */
2298 static void
2299 ptl1_wakeup(void *arg)
2300 {
2301 	mutex_enter(&ptl1_mutex);
2302 	cv_signal(&ptl1_cv);
2303 	mutex_exit(&ptl1_mutex);
2304 }
2305 
2306 /*
2307  * ptl1_panic cross call function:
2308  *     Needed because xc_one() and xc_some() can pass
2309  *	64 bit args but ptl1_recurse() expects ints.
2310  */
2311 static void
2312 ptl1_panic_xc(void)
2313 {
2314 	ptl1_recurse(ptl1_recurse_count_threshold,
2315 	    ptl1_recurse_trap_threshold);
2316 }
2317 
2318 /*
2319  * The ptl1 thread waits for a global flag to be set
2320  * and uses the recurse thresholds to set the stack depth
2321  * to cause a ptl1_panic() directly via a call to ptl1_recurse
2322  * or indirectly via the cross call and cross trap functions.
2323  *
2324  * This is useful testing stack overflows and normal
2325  * ptl1_panic() states with a know stack frame.
2326  *
2327  * ptl1_recurse() is an asm function in ptl1_panic.s that
2328  * sets the {In, Local, Out, and Global} registers to a
2329  * know state on the stack and just prior to causing a
2330  * test ptl1_panic trap.
2331  */
2332 static void
2333 ptl1_thread(void)
2334 {
2335 	mutex_enter(&ptl1_mutex);
2336 	while (ptl1_thread_p) {
2337 		cpuset_t	other_cpus;
2338 		int		cpu_id;
2339 		int		my_cpu_id;
2340 		int		target_cpu_id;
2341 		int		target_found;
2342 
2343 		if (ptl1_panic_test) {
2344 			ptl1_recurse(ptl1_recurse_count_threshold,
2345 			    ptl1_recurse_trap_threshold);
2346 		}
2347 
2348 		/*
2349 		 * Find potential targets for x-call and x-trap,
2350 		 * if any exist while preempt is disabled we
2351 		 * start a ptl1_panic if requested via a
2352 		 * globals.
2353 		 */
2354 		kpreempt_disable();
2355 		my_cpu_id = CPU->cpu_id;
2356 		other_cpus = cpu_ready_set;
2357 		CPUSET_DEL(other_cpus, CPU->cpu_id);
2358 		target_found = 0;
2359 		if (!CPUSET_ISNULL(other_cpus)) {
2360 			/*
2361 			 * Pick the first one
2362 			 */
2363 			for (cpu_id = 0; cpu_id < NCPU; cpu_id++) {
2364 				if (cpu_id == my_cpu_id)
2365 					continue;
2366 
2367 				if (CPU_XCALL_READY(cpu_id)) {
2368 					target_cpu_id = cpu_id;
2369 					target_found = 1;
2370 					break;
2371 				}
2372 			}
2373 			ASSERT(target_found);
2374 
2375 			if (ptl1_panic_xc_one_test) {
2376 				xc_one(target_cpu_id,
2377 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2378 			}
2379 			if (ptl1_panic_xc_all_test) {
2380 				xc_some(other_cpus,
2381 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2382 			}
2383 			if (ptl1_panic_xt_one_test) {
2384 				xt_one(target_cpu_id,
2385 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2386 			}
2387 			if (ptl1_panic_xt_all_test) {
2388 				xt_some(other_cpus,
2389 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2390 			}
2391 		}
2392 		kpreempt_enable();
2393 		(void) timeout(ptl1_wakeup, NULL, hz);
2394 		(void) cv_wait(&ptl1_cv, &ptl1_mutex);
2395 	}
2396 	mutex_exit(&ptl1_mutex);
2397 }
2398 
2399 /*
2400  * Called during early startup to create the ptl1_thread
2401  */
2402 void
2403 init_ptl1_thread(void)
2404 {
2405 	ptl1_thread_p = thread_create(NULL, 0, ptl1_thread, NULL, 0,
2406 	    &p0, TS_RUN, 0);
2407 }
2408 #endif	/* PTL1_PANIC_DEBUG */
2409 
2410 
2411 static void
2412 memlist_new(uint64_t start, uint64_t len, struct memlist **memlistp)
2413 {
2414 	struct memlist *new;
2415 
2416 	new = *memlistp;
2417 	new->ml_address = start;
2418 	new->ml_size = len;
2419 	*memlistp = new + 1;
2420 }
2421 
2422 /*
2423  * Add to a memory list.
2424  * start = start of new memory segment
2425  * len = length of new memory segment in bytes
2426  * memlistp = pointer to array of available memory segment structures
2427  * curmemlistp = memory list to which to add segment.
2428  */
2429 static void
2430 memlist_add(uint64_t start, uint64_t len, struct memlist **memlistp,
2431 	struct memlist **curmemlistp)
2432 {
2433 	struct memlist *new = *memlistp;
2434 
2435 	memlist_new(start, len, memlistp);
2436 	memlist_insert(new, curmemlistp);
2437 }
2438 
2439 static int
2440 ndata_alloc_memseg(struct memlist *ndata, size_t avail)
2441 {
2442 	int nseg;
2443 	size_t memseg_sz;
2444 	struct memseg *msp;
2445 
2446 	/*
2447 	 * The memseg list is for the chunks of physical memory that
2448 	 * will be managed by the vm system.  The number calculated is
2449 	 * a guess as boot may fragment it more when memory allocations
2450 	 * are made before kphysm_init().
2451 	 */
2452 	memseg_sz = (avail + 10) * sizeof (struct memseg);
2453 	memseg_sz = roundup(memseg_sz, PAGESIZE);
2454 	nseg = memseg_sz / sizeof (struct memseg);
2455 	msp = ndata_alloc(ndata, memseg_sz, ecache_alignsize);
2456 	if (msp == NULL)
2457 		return (1);
2458 	PRM_DEBUG(memseg_free);
2459 
2460 	while (nseg--) {
2461 		msp->next = memseg_free;
2462 		memseg_free = msp;
2463 		msp++;
2464 	}
2465 	return (0);
2466 }
2467 
2468 /*
2469  * In the case of architectures that support dynamic addition of
2470  * memory at run-time there are two cases where memsegs need to
2471  * be initialized and added to the memseg list.
2472  * 1) memsegs that are constructed at startup.
2473  * 2) memsegs that are constructed at run-time on
2474  *    hot-plug capable architectures.
2475  * This code was originally part of the function kphysm_init().
2476  */
2477 
2478 static void
2479 memseg_list_add(struct memseg *memsegp)
2480 {
2481 	struct memseg **prev_memsegp;
2482 	pgcnt_t num;
2483 
2484 	/* insert in memseg list, decreasing number of pages order */
2485 
2486 	num = MSEG_NPAGES(memsegp);
2487 
2488 	for (prev_memsegp = &memsegs; *prev_memsegp;
2489 	    prev_memsegp = &((*prev_memsegp)->next)) {
2490 		if (num > MSEG_NPAGES(*prev_memsegp))
2491 			break;
2492 	}
2493 
2494 	memsegp->next = *prev_memsegp;
2495 	*prev_memsegp = memsegp;
2496 
2497 	if (kpm_enable) {
2498 		memsegp->nextpa = (memsegp->next) ?
2499 		    va_to_pa(memsegp->next) : MSEG_NULLPTR_PA;
2500 
2501 		if (prev_memsegp != &memsegs) {
2502 			struct memseg *msp;
2503 			msp = (struct memseg *)((caddr_t)prev_memsegp -
2504 			    offsetof(struct memseg, next));
2505 			msp->nextpa = va_to_pa(memsegp);
2506 		} else {
2507 			memsegspa = va_to_pa(memsegs);
2508 		}
2509 	}
2510 }
2511 
2512 /*
2513  * PSM add_physmem_cb(). US-II and newer processors have some
2514  * flavor of the prefetch capability implemented. We exploit
2515  * this capability for optimum performance.
2516  */
2517 #define	PREFETCH_BYTES	64
2518 
2519 void
2520 add_physmem_cb(page_t *pp, pfn_t pnum)
2521 {
2522 	extern void	 prefetch_page_w(void *);
2523 
2524 	pp->p_pagenum = pnum;
2525 
2526 	/*
2527 	 * Prefetch one more page_t into E$. To prevent future
2528 	 * mishaps with the sizeof(page_t) changing on us, we
2529 	 * catch this on debug kernels if we can't bring in the
2530 	 * entire hpage with 2 PREFETCH_BYTES reads. See
2531 	 * also, sun4u/cpu/cpu_module.c
2532 	 */
2533 	/*LINTED*/
2534 	ASSERT(sizeof (page_t) <= 2*PREFETCH_BYTES);
2535 	prefetch_page_w((char *)pp);
2536 }
2537 
2538 /*
2539  * Find memseg with given pfn
2540  */
2541 static struct memseg *
2542 memseg_find(pfn_t base, pfn_t *next)
2543 {
2544 	struct memseg *seg;
2545 
2546 	if (next != NULL)
2547 		*next = LONG_MAX;
2548 	for (seg = memsegs; seg != NULL; seg = seg->next) {
2549 		if (base >= seg->pages_base && base < seg->pages_end)
2550 			return (seg);
2551 		if (next != NULL && seg->pages_base > base &&
2552 		    seg->pages_base < *next)
2553 			*next = seg->pages_base;
2554 	}
2555 	return (NULL);
2556 }
2557 
2558 /*
2559  * Put page allocated by OBP on prom_ppages
2560  */
2561 static void
2562 kphysm_erase(uint64_t addr, uint64_t len)
2563 {
2564 	struct page *pp;
2565 	struct memseg *seg;
2566 	pfn_t base = btop(addr), next;
2567 	pgcnt_t num = btop(len);
2568 
2569 	while (num != 0) {
2570 		pgcnt_t off, left;
2571 
2572 		seg = memseg_find(base, &next);
2573 		if (seg == NULL) {
2574 			if (next == LONG_MAX)
2575 				break;
2576 			left = MIN(next - base, num);
2577 			base += left, num -= left;
2578 			continue;
2579 		}
2580 		off = base - seg->pages_base;
2581 		pp = seg->pages + off;
2582 		left = num - MIN(num, (seg->pages_end - seg->pages_base) - off);
2583 		while (num != left) {
2584 			/*
2585 			 * init it, lock it, and hashin on prom_pages vp.
2586 			 *
2587 			 * Mark it as NONRELOC to let DR know the page
2588 			 * is locked long term, otherwise DR hangs when
2589 			 * trying to remove those pages.
2590 			 *
2591 			 * XXX	vnode offsets on the prom_ppages vnode
2592 			 *	are page numbers (gack) for >32 bit
2593 			 *	physical memory machines.
2594 			 */
2595 			PP_SETNORELOC(pp);
2596 			add_physmem_cb(pp, base);
2597 			if (page_trylock(pp, SE_EXCL) == 0)
2598 				cmn_err(CE_PANIC, "prom page locked");
2599 			(void) page_hashin(pp, &promvp,
2600 			    (offset_t)base, NULL);
2601 			(void) page_pp_lock(pp, 0, 1);
2602 			pp++, base++, num--;
2603 		}
2604 	}
2605 }
2606 
2607 static page_t *ppnext;
2608 static pgcnt_t ppleft;
2609 
2610 static void *kpm_ppnext;
2611 static pgcnt_t kpm_ppleft;
2612 
2613 /*
2614  * Create a memseg
2615  */
2616 static void
2617 kphysm_memseg(uint64_t addr, uint64_t len)
2618 {
2619 	pfn_t base = btop(addr);
2620 	pgcnt_t num = btop(len);
2621 	struct memseg *seg;
2622 
2623 	seg = memseg_free;
2624 	memseg_free = seg->next;
2625 	ASSERT(seg != NULL);
2626 
2627 	seg->pages = ppnext;
2628 	seg->epages = ppnext + num;
2629 	seg->pages_base = base;
2630 	seg->pages_end = base + num;
2631 	ppnext += num;
2632 	ppleft -= num;
2633 
2634 	if (kpm_enable) {
2635 		pgcnt_t kpnum = ptokpmpr(num);
2636 
2637 		if (kpnum > kpm_ppleft)
2638 			panic("kphysm_memseg: kpm_pp overflow");
2639 		seg->pagespa = va_to_pa(seg->pages);
2640 		seg->epagespa = va_to_pa(seg->epages);
2641 		seg->kpm_pbase = kpmptop(ptokpmp(base));
2642 		seg->kpm_nkpmpgs = kpnum;
2643 		/*
2644 		 * In the kpm_smallpage case, the kpm array
2645 		 * is 1-1 wrt the page array
2646 		 */
2647 		if (kpm_smallpages) {
2648 			kpm_spage_t *kpm_pp = kpm_ppnext;
2649 
2650 			kpm_ppnext = kpm_pp + kpnum;
2651 			seg->kpm_spages = kpm_pp;
2652 			seg->kpm_pagespa = va_to_pa(seg->kpm_spages);
2653 		} else {
2654 			kpm_page_t *kpm_pp = kpm_ppnext;
2655 
2656 			kpm_ppnext = kpm_pp + kpnum;
2657 			seg->kpm_pages = kpm_pp;
2658 			seg->kpm_pagespa = va_to_pa(seg->kpm_pages);
2659 			/* ASSERT no kpm overlaps */
2660 			ASSERT(
2661 			    memseg_find(base - pmodkpmp(base), NULL) == NULL);
2662 			ASSERT(memseg_find(
2663 			    roundup(base + num, kpmpnpgs) - 1, NULL) == NULL);
2664 		}
2665 		kpm_ppleft -= kpnum;
2666 	}
2667 
2668 	memseg_list_add(seg);
2669 }
2670 
2671 /*
2672  * Add range to free list
2673  */
2674 void
2675 kphysm_add(uint64_t addr, uint64_t len, int reclaim)
2676 {
2677 	struct page *pp;
2678 	struct memseg *seg;
2679 	pfn_t base = btop(addr);
2680 	pgcnt_t num = btop(len);
2681 
2682 	seg = memseg_find(base, NULL);
2683 	ASSERT(seg != NULL);
2684 	pp = seg->pages + (base - seg->pages_base);
2685 
2686 	if (reclaim) {
2687 		struct page *rpp = pp;
2688 		struct page *lpp = pp + num;
2689 
2690 		/*
2691 		 * page should be locked on prom_ppages
2692 		 * unhash and unlock it
2693 		 */
2694 		while (rpp < lpp) {
2695 			ASSERT(PAGE_EXCL(rpp) && rpp->p_vnode == &promvp);
2696 			ASSERT(PP_ISNORELOC(rpp));
2697 			PP_CLRNORELOC(rpp);
2698 			page_pp_unlock(rpp, 0, 1);
2699 			page_hashout(rpp, NULL);
2700 			page_unlock(rpp);
2701 			rpp++;
2702 		}
2703 	}
2704 
2705 	/*
2706 	 * add_physmem() initializes the PSM part of the page
2707 	 * struct by calling the PSM back with add_physmem_cb().
2708 	 * In addition it coalesces pages into larger pages as
2709 	 * it initializes them.
2710 	 */
2711 	add_physmem(pp, num, base);
2712 }
2713 
2714 /*
2715  * kphysm_init() tackles the problem of initializing physical memory.
2716  */
2717 static void
2718 kphysm_init(void)
2719 {
2720 	struct memlist *pmem;
2721 
2722 	ASSERT(page_hash != NULL && page_hashsz != 0);
2723 
2724 	ppnext = pp_base;
2725 	ppleft = npages;
2726 	kpm_ppnext = kpm_pp_base;
2727 	kpm_ppleft = kpm_npages;
2728 
2729 	/*
2730 	 * installed pages not on nopp_memlist go in memseg list
2731 	 */
2732 	diff_memlists(phys_install, nopp_list, kphysm_memseg);
2733 
2734 	/*
2735 	 * Free the avail list
2736 	 */
2737 	for (pmem = phys_avail; pmem != NULL; pmem = pmem->ml_next)
2738 		kphysm_add(pmem->ml_address, pmem->ml_size, 0);
2739 
2740 	/*
2741 	 * Erase pages that aren't available
2742 	 */
2743 	diff_memlists(phys_install, phys_avail, kphysm_erase);
2744 
2745 	build_pfn_hash();
2746 }
2747 
2748 /*
2749  * Kernel VM initialization.
2750  * Assumptions about kernel address space ordering:
2751  *	(1) gap (user space)
2752  *	(2) kernel text
2753  *	(3) kernel data/bss
2754  *	(4) gap
2755  *	(5) kernel data structures
2756  *	(6) gap
2757  *	(7) debugger (optional)
2758  *	(8) monitor
2759  *	(9) gap (possibly null)
2760  *	(10) dvma
2761  *	(11) devices
2762  */
2763 static void
2764 kvm_init(void)
2765 {
2766 	/*
2767 	 * Put the kernel segments in kernel address space.
2768 	 */
2769 	rw_enter(&kas.a_lock, RW_WRITER);
2770 	as_avlinit(&kas);
2771 
2772 	(void) seg_attach(&kas, (caddr_t)KERNELBASE,
2773 	    (size_t)(e_moddata - KERNELBASE), &ktextseg);
2774 	(void) segkmem_create(&ktextseg);
2775 
2776 	(void) seg_attach(&kas, (caddr_t)(KERNELBASE + MMU_PAGESIZE4M),
2777 	    (size_t)(MMU_PAGESIZE4M), &ktexthole);
2778 	(void) segkmem_create(&ktexthole);
2779 
2780 	(void) seg_attach(&kas, (caddr_t)valloc_base,
2781 	    (size_t)(econtig32 - valloc_base), &kvalloc);
2782 	(void) segkmem_create(&kvalloc);
2783 
2784 	if (kmem64_base) {
2785 		(void) seg_attach(&kas, (caddr_t)kmem64_base,
2786 		    (size_t)(kmem64_end - kmem64_base), &kmem64);
2787 		(void) segkmem_create(&kmem64);
2788 	}
2789 
2790 	/*
2791 	 * We're about to map out /boot.  This is the beginning of the
2792 	 * system resource management transition. We can no longer
2793 	 * call into /boot for I/O or memory allocations.
2794 	 */
2795 	(void) seg_attach(&kas, kernelheap, ekernelheap - kernelheap, &kvseg);
2796 	(void) segkmem_create(&kvseg);
2797 	hblk_alloc_dynamic = 1;
2798 
2799 	/*
2800 	 * we need to preallocate pages for DR operations before enabling large
2801 	 * page kernel heap because of memseg_remap_init() hat_unload() hack.
2802 	 */
2803 	memseg_remap_init();
2804 
2805 	/* at this point we are ready to use large page heap */
2806 	segkmem_heap_lp_init();
2807 
2808 	(void) seg_attach(&kas, (caddr_t)SYSBASE32, SYSLIMIT32 - SYSBASE32,
2809 	    &kvseg32);
2810 	(void) segkmem_create(&kvseg32);
2811 
2812 	/*
2813 	 * Create a segment for the debugger.
2814 	 */
2815 	(void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2816 	(void) segkmem_create(&kdebugseg);
2817 
2818 	rw_exit(&kas.a_lock);
2819 }
2820 
2821 char obp_tte_str[] =
2822 	"h# %x constant MMU_PAGESHIFT "
2823 	"h# %x constant TTE8K "
2824 	"h# %x constant SFHME_SIZE "
2825 	"h# %x constant SFHME_TTE "
2826 	"h# %x constant HMEBLK_TAG "
2827 	"h# %x constant HMEBLK_NEXT "
2828 	"h# %x constant HMEBLK_MISC "
2829 	"h# %x constant HMEBLK_HME1 "
2830 	"h# %x constant NHMENTS "
2831 	"h# %x constant HBLK_SZMASK "
2832 	"h# %x constant HBLK_RANGE_SHIFT "
2833 	"h# %x constant HMEBP_HBLK "
2834 	"h# %x constant HMEBLK_ENDPA "
2835 	"h# %x constant HMEBUCKET_SIZE "
2836 	"h# %x constant HTAG_SFMMUPSZ "
2837 	"h# %x constant HTAG_BSPAGE_SHIFT "
2838 	"h# %x constant HTAG_REHASH_SHIFT "
2839 	"h# %x constant SFMMU_INVALID_SHMERID "
2840 	"h# %x constant mmu_hashcnt "
2841 	"h# %p constant uhme_hash "
2842 	"h# %p constant khme_hash "
2843 	"h# %x constant UHMEHASH_SZ "
2844 	"h# %x constant KHMEHASH_SZ "
2845 	"h# %p constant KCONTEXT "
2846 	"h# %p constant KHATID "
2847 	"h# %x constant ASI_MEM "
2848 
2849 	": PHYS-X@ ( phys -- data ) "
2850 	"   ASI_MEM spacex@ "
2851 	"; "
2852 
2853 	": PHYS-W@ ( phys -- data ) "
2854 	"   ASI_MEM spacew@ "
2855 	"; "
2856 
2857 	": PHYS-L@ ( phys -- data ) "
2858 	"   ASI_MEM spaceL@ "
2859 	"; "
2860 
2861 	": TTE_PAGE_SHIFT ( ttesz -- hmeshift ) "
2862 	"   3 * MMU_PAGESHIFT + "
2863 	"; "
2864 
2865 	": TTE_IS_VALID ( ttep -- flag ) "
2866 	"   PHYS-X@ 0< "
2867 	"; "
2868 
2869 	": HME_HASH_SHIFT ( ttesz -- hmeshift ) "
2870 	"   dup TTE8K =  if "
2871 	"      drop HBLK_RANGE_SHIFT "
2872 	"   else "
2873 	"      TTE_PAGE_SHIFT "
2874 	"   then "
2875 	"; "
2876 
2877 	": HME_HASH_BSPAGE ( addr hmeshift -- bspage ) "
2878 	"   tuck >> swap MMU_PAGESHIFT - << "
2879 	"; "
2880 
2881 	": HME_HASH_FUNCTION ( sfmmup addr hmeshift -- hmebp ) "
2882 	"   >> over xor swap                    ( hash sfmmup ) "
2883 	"   KHATID <>  if                       ( hash ) "
2884 	"      UHMEHASH_SZ and                  ( bucket ) "
2885 	"      HMEBUCKET_SIZE * uhme_hash +     ( hmebp ) "
2886 	"   else                                ( hash ) "
2887 	"      KHMEHASH_SZ and                  ( bucket ) "
2888 	"      HMEBUCKET_SIZE * khme_hash +     ( hmebp ) "
2889 	"   then                                ( hmebp ) "
2890 	"; "
2891 
2892 	": HME_HASH_TABLE_SEARCH "
2893 	"       ( sfmmup hmebp hblktag --  sfmmup null | sfmmup hmeblkp ) "
2894 	"   >r hmebp_hblk + phys-x@ begin ( sfmmup hmeblkp ) ( r: hblktag ) "
2895 	"      dup HMEBLK_ENDPA <> if     ( sfmmup hmeblkp ) ( r: hblktag ) "
2896 	"         dup hmeblk_tag + phys-x@ r@ = if ( sfmmup hmeblkp )	  "
2897 	"	     dup hmeblk_tag + 8 + phys-x@ 2 pick = if		  "
2898 	"		  true 	( sfmmup hmeblkp true ) ( r: hblktag )	  "
2899 	"	     else						  "
2900 	"	     	  hmeblk_next + phys-x@ false 			  "
2901 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2902 	"	     then  						  "
2903 	"	  else							  "
2904 	"	     hmeblk_next + phys-x@ false 			  "
2905 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2906 	"	  then 							  "
2907 	"      else							  "
2908 	"         drop 0 true 						  "
2909 	"      then  							  "
2910 	"   until r> drop 						  "
2911 	"; "
2912 
2913 	": HME_HASH_TAG ( sfmmup rehash addr -- hblktag ) "
2914 	"   over HME_HASH_SHIFT HME_HASH_BSPAGE  ( sfmmup rehash bspage ) "
2915 	"   HTAG_BSPAGE_SHIFT <<		 ( sfmmup rehash htag-bspage )"
2916 	"   swap HTAG_REHASH_SHIFT << or	 ( sfmmup htag-bspage-rehash )"
2917 	"   SFMMU_INVALID_SHMERID or nip	 ( hblktag ) "
2918 	"; "
2919 
2920 	": HBLK_TO_TTEP ( hmeblkp addr -- ttep ) "
2921 	"   over HMEBLK_MISC + PHYS-L@ HBLK_SZMASK and  ( hmeblkp addr ttesz ) "
2922 	"   TTE8K =  if                            ( hmeblkp addr ) "
2923 	"      MMU_PAGESHIFT >> NHMENTS 1- and     ( hmeblkp hme-index ) "
2924 	"   else                                   ( hmeblkp addr ) "
2925 	"      drop 0                              ( hmeblkp 0 ) "
2926 	"   then                                   ( hmeblkp hme-index ) "
2927 	"   SFHME_SIZE * + HMEBLK_HME1 +           ( hmep ) "
2928 	"   SFHME_TTE +                            ( ttep ) "
2929 	"; "
2930 
2931 	": unix-tte ( addr cnum -- false | tte-data true ) "
2932 	"    KCONTEXT = if                   ( addr ) "
2933 	"	KHATID                       ( addr khatid ) "
2934 	"    else                            ( addr ) "
2935 	"       drop false exit              ( false ) "
2936 	"    then "
2937 	"      ( addr khatid ) "
2938 	"      mmu_hashcnt 1+ 1  do           ( addr sfmmup ) "
2939 	"         2dup swap i HME_HASH_SHIFT  "
2940 					"( addr sfmmup sfmmup addr hmeshift ) "
2941 	"         HME_HASH_FUNCTION           ( addr sfmmup hmebp ) "
2942 	"         over i 4 pick               "
2943 				"( addr sfmmup hmebp sfmmup rehash addr ) "
2944 	"         HME_HASH_TAG                ( addr sfmmup hmebp hblktag ) "
2945 	"         HME_HASH_TABLE_SEARCH       "
2946 					"( addr sfmmup { null | hmeblkp } ) "
2947 	"         ?dup  if                    ( addr sfmmup hmeblkp ) "
2948 	"            nip swap HBLK_TO_TTEP    ( ttep ) "
2949 	"            dup TTE_IS_VALID  if     ( valid-ttep ) "
2950 	"               PHYS-X@ true          ( tte-data true ) "
2951 	"            else                     ( invalid-tte ) "
2952 	"               drop false            ( false ) "
2953 	"            then                     ( false | tte-data true ) "
2954 	"            unloop exit              ( false | tte-data true ) "
2955 	"         then                        ( addr sfmmup ) "
2956 	"      loop                           ( addr sfmmup ) "
2957 	"      2drop false                    ( false ) "
2958 	"; "
2959 ;
2960 
2961 void
2962 create_va_to_tte(void)
2963 {
2964 	char *bp;
2965 	extern int khmehash_num, uhmehash_num;
2966 	extern struct hmehash_bucket *khme_hash, *uhme_hash;
2967 
2968 #define	OFFSET(type, field)	((uintptr_t)(&((type *)0)->field))
2969 
2970 	bp = (char *)kobj_zalloc(MMU_PAGESIZE, KM_SLEEP);
2971 
2972 	/*
2973 	 * Teach obp how to parse our sw ttes.
2974 	 */
2975 	(void) sprintf(bp, obp_tte_str,
2976 	    MMU_PAGESHIFT,
2977 	    TTE8K,
2978 	    sizeof (struct sf_hment),
2979 	    OFFSET(struct sf_hment, hme_tte),
2980 	    OFFSET(struct hme_blk, hblk_tag),
2981 	    OFFSET(struct hme_blk, hblk_nextpa),
2982 	    OFFSET(struct hme_blk, hblk_misc),
2983 	    OFFSET(struct hme_blk, hblk_hme),
2984 	    NHMENTS,
2985 	    HBLK_SZMASK,
2986 	    HBLK_RANGE_SHIFT,
2987 	    OFFSET(struct hmehash_bucket, hmeh_nextpa),
2988 	    HMEBLK_ENDPA,
2989 	    sizeof (struct hmehash_bucket),
2990 	    HTAG_SFMMUPSZ,
2991 	    HTAG_BSPAGE_SHIFT,
2992 	    HTAG_REHASH_SHIFT,
2993 	    SFMMU_INVALID_SHMERID,
2994 	    mmu_hashcnt,
2995 	    (caddr_t)va_to_pa((caddr_t)uhme_hash),
2996 	    (caddr_t)va_to_pa((caddr_t)khme_hash),
2997 	    UHMEHASH_SZ,
2998 	    KHMEHASH_SZ,
2999 	    KCONTEXT,
3000 	    KHATID,
3001 	    ASI_MEM);
3002 	prom_interpret(bp, 0, 0, 0, 0, 0);
3003 
3004 	kobj_free(bp, MMU_PAGESIZE);
3005 }
3006 
3007 void
3008 install_va_to_tte(void)
3009 {
3010 	/*
3011 	 * advise prom that he can use unix-tte
3012 	 */
3013 	prom_interpret("' unix-tte is va>tte-data", 0, 0, 0, 0, 0);
3014 }
3015 
3016 /*
3017  * Here we add "device-type=console" for /os-io node, for currently
3018  * our kernel console output only supports displaying text and
3019  * performing cursor-positioning operations (through kernel framebuffer
3020  * driver) and it doesn't support other functionalities required for a
3021  * standard "display" device as specified in 1275 spec. The main missing
3022  * interface defined by the 1275 spec is "draw-logo".
3023  * also see the comments above prom_stdout_is_framebuffer().
3024  */
3025 static char *create_node =
3026 	"\" /\" find-device "
3027 	"new-device "
3028 	"\" os-io\" device-name "
3029 	"\" "OBP_DISPLAY_CONSOLE"\" device-type "
3030 	": cb-r/w  ( adr,len method$ -- #read/#written ) "
3031 	"   2>r swap 2 2r> ['] $callback  catch  if "
3032 	"      2drop 3drop 0 "
3033 	"   then "
3034 	"; "
3035 	": read ( adr,len -- #read ) "
3036 	"       \" read\" ['] cb-r/w catch  if  2drop 2drop -2 exit then "
3037 	"       ( retN ... ret1 N ) "
3038 	"       ?dup  if "
3039 	"               swap >r 1-  0  ?do  drop  loop  r> "
3040 	"       else "
3041 	"               -2 "
3042 	"       then "
3043 	";    "
3044 	": write ( adr,len -- #written ) "
3045 	"       \" write\" ['] cb-r/w catch  if  2drop 2drop 0 exit  then "
3046 	"       ( retN ... ret1 N ) "
3047 	"       ?dup  if "
3048 	"               swap >r 1-  0  ?do  drop  loop  r> "
3049 	"        else "
3050 	"               0 "
3051 	"       then "
3052 	"; "
3053 	": poll-tty ( -- ) ; "
3054 	": install-abort  ( -- )  ['] poll-tty d# 10 alarm ; "
3055 	": remove-abort ( -- )  ['] poll-tty 0 alarm ; "
3056 	": cb-give/take ( $method -- ) "
3057 	"       0 -rot ['] $callback catch  ?dup  if "
3058 	"               >r 2drop 2drop r> throw "
3059 	"       else "
3060 	"               0  ?do  drop  loop "
3061 	"       then "
3062 	"; "
3063 	": give ( -- )  \" exit-input\" cb-give/take ; "
3064 	": take ( -- )  \" enter-input\" cb-give/take ; "
3065 	": open ( -- ok? )  true ; "
3066 	": close ( -- ) ; "
3067 	"finish-device "
3068 	"device-end ";
3069 
3070 /*
3071  * Create the OBP input/output node (FCode serial driver).
3072  * It is needed for both USB console keyboard and for
3073  * the kernel terminal emulator.  It is too early to check for a
3074  * kernel console compatible framebuffer now, so we create this
3075  * so that we're ready if we need to enable kernel terminal emulation.
3076  *
3077  * When the USB software takes over the input device at the time
3078  * consconfig runs, OBP's stdin is redirected to this node.
3079  * Whenever the FORTH user interface is used after this switch,
3080  * the node will call back into the kernel for console input.
3081  * If a serial device such as ttya or a UART with a Type 5 keyboard
3082  * attached is used, OBP takes over the serial device when the system
3083  * goes to the debugger after the system is booted.  This sharing
3084  * of the relatively simple serial device is difficult but possible.
3085  * Sharing the USB host controller is impossible due its complexity.
3086  *
3087  * Similarly to USB keyboard input redirection, after consconfig_dacf
3088  * configures a kernel console framebuffer as the standard output
3089  * device, OBP's stdout is switched to to vector through the
3090  * /os-io node into the kernel terminal emulator.
3091  */
3092 static void
3093 startup_create_io_node(void)
3094 {
3095 	prom_interpret(create_node, 0, 0, 0, 0, 0);
3096 }
3097 
3098 
3099 static void
3100 do_prom_version_check(void)
3101 {
3102 	int i;
3103 	pnode_t node;
3104 	char buf[64];
3105 	static char drev[] = "Down-rev firmware detected%s\n"
3106 	    "\tPlease upgrade to the following minimum version:\n"
3107 	    "\t\t%s\n";
3108 
3109 	i = prom_version_check(buf, sizeof (buf), &node);
3110 
3111 	if (i == PROM_VER64_OK)
3112 		return;
3113 
3114 	if (i == PROM_VER64_UPGRADE) {
3115 		cmn_err(CE_WARN, drev, "", buf);
3116 
3117 #ifdef	DEBUG
3118 		prom_enter_mon();	/* Type 'go' to continue */
3119 		cmn_err(CE_WARN, "Booting with down-rev firmware\n");
3120 		return;
3121 #else
3122 		halt(0);
3123 #endif
3124 	}
3125 
3126 	/*
3127 	 * The other possibility is that this is a server running
3128 	 * good firmware, but down-rev firmware was detected on at
3129 	 * least one other cpu board. We just complain if we see
3130 	 * that.
3131 	 */
3132 	cmn_err(CE_WARN, drev, " on one or more CPU boards", buf);
3133 }
3134 
3135 
3136 /*
3137  * Must be defined in platform dependent code.
3138  */
3139 extern caddr_t modtext;
3140 extern size_t modtext_sz;
3141 extern caddr_t moddata;
3142 
3143 #define	HEAPTEXT_ARENA(addr)	\
3144 	((uintptr_t)(addr) < KERNELBASE + 2 * MMU_PAGESIZE4M ? 0 : \
3145 	(((uintptr_t)(addr) - HEAPTEXT_BASE) / \
3146 	(HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) + 1))
3147 
3148 #define	HEAPTEXT_OVERSIZED(addr)	\
3149 	((uintptr_t)(addr) >= HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE)
3150 
3151 #define	HEAPTEXT_IN_NUCLEUSDATA(addr) \
3152 	(((uintptr_t)(addr) >= KERNELBASE + 2 * MMU_PAGESIZE4M) && \
3153 	((uintptr_t)(addr) < KERNELBASE + 3 * MMU_PAGESIZE4M))
3154 
3155 vmem_t *texthole_source[HEAPTEXT_NARENAS];
3156 vmem_t *texthole_arena[HEAPTEXT_NARENAS];
3157 kmutex_t texthole_lock;
3158 
3159 char kern_bootargs[OBP_MAXPATHLEN];
3160 char kern_bootfile[OBP_MAXPATHLEN];
3161 
3162 void
3163 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3164 {
3165 	uintptr_t addr, limit;
3166 
3167 	addr = HEAPTEXT_BASE;
3168 	limit = addr + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE;
3169 
3170 	/*
3171 	 * Before we initialize the text_arena, we want to punch holes in the
3172 	 * underlying heaptext_arena.  This guarantees that for any text
3173 	 * address we can find a text hole less than HEAPTEXT_MAPPED away.
3174 	 */
3175 	for (; addr + HEAPTEXT_UNMAPPED <= limit;
3176 	    addr += HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) {
3177 		(void) vmem_xalloc(heaptext_arena, HEAPTEXT_UNMAPPED, PAGESIZE,
3178 		    0, 0, (void *)addr, (void *)(addr + HEAPTEXT_UNMAPPED),
3179 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3180 	}
3181 
3182 	/*
3183 	 * Allocate one page at the oversize to break up the text region
3184 	 * from the oversized region.
3185 	 */
3186 	(void) vmem_xalloc(heaptext_arena, PAGESIZE, PAGESIZE, 0, 0,
3187 	    (void *)limit, (void *)(limit + PAGESIZE),
3188 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3189 
3190 	*text_arena = vmem_create("module_text", modtext_sz ? modtext : NULL,
3191 	    modtext_sz, sizeof (uintptr_t), segkmem_alloc, segkmem_free,
3192 	    heaptext_arena, 0, VM_SLEEP);
3193 	*data_arena = vmem_create("module_data", moddata, MODDATA, 1,
3194 	    segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3195 }
3196 
3197 caddr_t
3198 kobj_text_alloc(vmem_t *arena, size_t size)
3199 {
3200 	caddr_t rval, better;
3201 
3202 	/*
3203 	 * First, try a sleeping allocation.
3204 	 */
3205 	rval = vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT);
3206 
3207 	if (size >= HEAPTEXT_MAPPED || !HEAPTEXT_OVERSIZED(rval))
3208 		return (rval);
3209 
3210 	/*
3211 	 * We didn't get the area that we wanted.  We're going to try to do an
3212 	 * allocation with explicit constraints.
3213 	 */
3214 	better = vmem_xalloc(arena, size, sizeof (uintptr_t), 0, 0, NULL,
3215 	    (void *)(HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE),
3216 	    VM_NOSLEEP | VM_BESTFIT);
3217 
3218 	if (better != NULL) {
3219 		/*
3220 		 * That worked.  Free our first attempt and return.
3221 		 */
3222 		vmem_free(arena, rval, size);
3223 		return (better);
3224 	}
3225 
3226 	/*
3227 	 * That didn't work; we'll have to return our first attempt.
3228 	 */
3229 	return (rval);
3230 }
3231 
3232 caddr_t
3233 kobj_texthole_alloc(caddr_t addr, size_t size)
3234 {
3235 	int arena = HEAPTEXT_ARENA(addr);
3236 	char c[30];
3237 	uintptr_t base;
3238 
3239 	if (HEAPTEXT_OVERSIZED(addr) || HEAPTEXT_IN_NUCLEUSDATA(addr)) {
3240 		/*
3241 		 * If this is an oversized allocation or it is allocated in
3242 		 * the nucleus data page, there is no text hole available for
3243 		 * it; return NULL.
3244 		 */
3245 		return (NULL);
3246 	}
3247 
3248 	mutex_enter(&texthole_lock);
3249 
3250 	if (texthole_arena[arena] == NULL) {
3251 		ASSERT(texthole_source[arena] == NULL);
3252 
3253 		if (arena == 0) {
3254 			texthole_source[0] = vmem_create("module_text_holesrc",
3255 			    (void *)(KERNELBASE + MMU_PAGESIZE4M),
3256 			    MMU_PAGESIZE4M, PAGESIZE, NULL, NULL, NULL,
3257 			    0, VM_SLEEP);
3258 		} else {
3259 			base = HEAPTEXT_BASE +
3260 			    (arena - 1) * (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED);
3261 
3262 			(void) snprintf(c, sizeof (c),
3263 			    "heaptext_holesrc_%d", arena);
3264 
3265 			texthole_source[arena] = vmem_create(c, (void *)base,
3266 			    HEAPTEXT_UNMAPPED, PAGESIZE, NULL, NULL, NULL,
3267 			    0, VM_SLEEP);
3268 		}
3269 
3270 		(void) snprintf(c, sizeof (c), "heaptext_hole_%d", arena);
3271 
3272 		texthole_arena[arena] = vmem_create(c, NULL, 0,
3273 		    sizeof (uint32_t), segkmem_alloc_permanent, segkmem_free,
3274 		    texthole_source[arena], 0, VM_SLEEP);
3275 	}
3276 
3277 	mutex_exit(&texthole_lock);
3278 
3279 	ASSERT(texthole_arena[arena] != NULL);
3280 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3281 	return (vmem_alloc(texthole_arena[arena], size,
3282 	    VM_BESTFIT | VM_NOSLEEP));
3283 }
3284 
3285 void
3286 kobj_texthole_free(caddr_t addr, size_t size)
3287 {
3288 	int arena = HEAPTEXT_ARENA(addr);
3289 
3290 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3291 	ASSERT(texthole_arena[arena] != NULL);
3292 	vmem_free(texthole_arena[arena], addr, size);
3293 }
3294 
3295 void
3296 release_bootstrap(void)
3297 {
3298 	if (&cif_init)
3299 		cif_init();
3300 }
3301