xref: /titanic_41/usr/src/uts/sun4/os/startup.c (revision 24fe0b3bf671e123467ce1df0b67cadd3614c8e4)
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 2009 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/intreg.h>
41 #include <sys/ivintr.h>
42 #include <sys/fpu/fpusystm.h>
43 #include <sys/iommutsb.h>
44 #include <vm/vm_dep.h>
45 #include <vm/seg_dev.h>
46 #include <vm/seg_kmem.h>
47 #include <vm/seg_kpm.h>
48 #include <vm/seg_map.h>
49 #include <vm/seg_kp.h>
50 #include <sys/sysconf.h>
51 #include <vm/hat_sfmmu.h>
52 #include <sys/kobj.h>
53 #include <sys/sun4asi.h>
54 #include <sys/clconf.h>
55 #include <sys/platform_module.h>
56 #include <sys/panic.h>
57 #include <sys/cpu_sgnblk_defs.h>
58 #include <sys/clock.h>
59 #include <sys/cmn_err.h>
60 #include <sys/promif.h>
61 #include <sys/prom_debug.h>
62 #include <sys/traptrace.h>
63 #include <sys/memnode.h>
64 #include <sys/mem_cage.h>
65 #include <sys/mmu.h>
66 
67 extern void setup_trap_table(void);
68 extern int cpu_intrq_setup(struct cpu *);
69 extern void cpu_intrq_register(struct cpu *);
70 extern void contig_mem_init(void);
71 extern caddr_t contig_mem_prealloc(caddr_t, pgcnt_t);
72 extern void mach_dump_buffer_init(void);
73 extern void mach_descrip_init(void);
74 extern void mach_descrip_startup_fini(void);
75 extern void mach_memscrub(void);
76 extern void mach_fpras(void);
77 extern void mach_cpu_halt_idle(void);
78 extern void mach_hw_copy_limit(void);
79 extern void load_mach_drivers(void);
80 extern void load_tod_module(void);
81 #pragma weak load_tod_module
82 
83 extern int ndata_alloc_mmfsa(struct memlist *ndata);
84 #pragma weak ndata_alloc_mmfsa
85 
86 extern void cif_init(void);
87 #pragma weak cif_init
88 
89 extern void parse_idprom(void);
90 extern void add_vx_handler(char *, int, void (*)(cell_t *));
91 extern void mem_config_init(void);
92 extern void memseg_remap_init(void);
93 
94 extern void mach_kpm_init(void);
95 extern void pcf_init();
96 extern int size_pse_array(pgcnt_t, int);
97 extern void pg_init();
98 
99 /*
100  * External Data:
101  */
102 extern int vac_size;	/* cache size in bytes */
103 extern uint_t vac_mask;	/* VAC alignment consistency mask */
104 extern uint_t vac_colors;
105 
106 /*
107  * Global Data Definitions:
108  */
109 
110 /*
111  * XXX - Don't port this to new architectures
112  * A 3rd party volume manager driver (vxdm) depends on the symbol romp.
113  * 'romp' has no use with a prom with an IEEE 1275 client interface.
114  * The driver doesn't use the value, but it depends on the symbol.
115  */
116 void *romp;		/* veritas driver won't load without romp 4154976 */
117 /*
118  * Declare these as initialized data so we can patch them.
119  */
120 pgcnt_t physmem = 0;	/* memory size in pages, patch if you want less */
121 pgcnt_t segkpsize =
122     btop(SEGKPDEFSIZE);	/* size of segkp segment in pages */
123 uint_t segmap_percent = 6; /* Size of segmap segment */
124 
125 int use_cache = 1;		/* cache not reliable (605 bugs) with MP */
126 int vac_copyback = 1;
127 char *cache_mode = NULL;
128 int use_mix = 1;
129 int prom_debug = 0;
130 
131 caddr_t boot_tba;		/* %tba at boot - used by kmdb */
132 uint_t	tba_taken_over = 0;
133 
134 caddr_t s_text;			/* start of kernel text segment */
135 caddr_t e_text;			/* end of kernel text segment */
136 caddr_t s_data;			/* start of kernel data segment */
137 caddr_t e_data;			/* end of kernel data segment */
138 
139 caddr_t modtext;		/* beginning of module text */
140 size_t	modtext_sz;		/* size of module text */
141 caddr_t moddata;		/* beginning of module data reserve */
142 caddr_t e_moddata;		/* end of module data reserve */
143 
144 /*
145  * End of first block of contiguous kernel in 32-bit virtual address space
146  */
147 caddr_t		econtig32;	/* end of first blk of contiguous kernel */
148 
149 caddr_t		ncbase;		/* beginning of non-cached segment */
150 caddr_t		ncend;		/* end of non-cached segment */
151 
152 size_t	ndata_remain_sz;	/* bytes from end of data to 4MB boundary */
153 caddr_t	nalloc_base;		/* beginning of nucleus allocation */
154 caddr_t nalloc_end;		/* end of nucleus allocatable memory */
155 caddr_t valloc_base;		/* beginning of kvalloc segment	*/
156 
157 caddr_t kmem64_base;		/* base of kernel mem segment in 64-bit space */
158 caddr_t kmem64_end;		/* end of kernel mem segment in 64-bit space */
159 size_t	kmem64_sz;		/* bytes in kernel mem segment, 64-bit space */
160 caddr_t kmem64_aligned_end;	/* end of large page, overmaps 64-bit space */
161 int	kmem64_szc;		/* page size code */
162 uint64_t kmem64_pabase = (uint64_t)-1;	/* physical address of kmem64_base */
163 
164 uintptr_t shm_alignment;	/* VAC address consistency modulus */
165 struct memlist *phys_install;	/* Total installed physical memory */
166 struct memlist *phys_avail;	/* Available (unreserved) physical memory */
167 struct memlist *virt_avail;	/* Available (unmapped?) virtual memory */
168 struct memlist *nopp_list;	/* pages with no backing page structs */
169 struct memlist ndata;		/* memlist of nucleus allocatable memory */
170 int memexp_flag;		/* memory expansion card flag */
171 uint64_t ecache_flushaddr;	/* physical address used for flushing E$ */
172 pgcnt_t obp_pages;		/* Physical pages used by OBP */
173 
174 /*
175  * VM data structures
176  */
177 long page_hashsz;		/* Size of page hash table (power of two) */
178 struct page *pp_base;		/* Base of system page struct array */
179 size_t pp_sz;			/* Size in bytes of page struct array */
180 struct page **page_hash;	/* Page hash table */
181 pad_mutex_t *pse_mutex;		/* Locks protecting pp->p_selock */
182 size_t pse_table_size;		/* Number of mutexes in pse_mutex[] */
183 int pse_shift;			/* log2(pse_table_size) */
184 struct seg ktextseg;		/* Segment used for kernel executable image */
185 struct seg kvalloc;		/* Segment used for "valloc" mapping */
186 struct seg kpseg;		/* Segment used for pageable kernel virt mem */
187 struct seg ktexthole;		/* Segment used for nucleus text hole */
188 struct seg kmapseg;		/* Segment used for generic kernel mappings */
189 struct seg kpmseg;		/* Segment used for physical mapping */
190 struct seg kdebugseg;		/* Segment used for the kernel debugger */
191 
192 void *kpm_pp_base;		/* Base of system kpm_page array */
193 size_t	kpm_pp_sz;		/* Size of system kpm_page array */
194 pgcnt_t	kpm_npages;		/* How many kpm pages are managed */
195 
196 struct seg *segkp = &kpseg;	/* Pageable kernel virtual memory segment */
197 struct seg *segkmap = &kmapseg;	/* Kernel generic mapping segment */
198 struct seg *segkpm = &kpmseg;	/* 64bit kernel physical mapping segment */
199 
200 int segzio_fromheap = 0;	/* zio allocations occur from heap */
201 caddr_t segzio_base;		/* Base address of segzio */
202 pgcnt_t segziosize = 0;		/* size of zio segment in pages */
203 
204 /*
205  * A static DR page_t VA map is reserved that can map the page structures
206  * for a domain's entire RA space. The pages that backs this space are
207  * dynamically allocated and need not be physically contiguous.  The DR
208  * map size is derived from KPM size.
209  */
210 int ppvm_enable = 0;		/* Static virtual map for page structs */
211 page_t *ppvm_base;		/* Base of page struct map */
212 pgcnt_t ppvm_size = 0;		/* Size of page struct map */
213 
214 /*
215  * debugger pages (if allocated)
216  */
217 struct vnode kdebugvp;
218 
219 /*
220  * VA range available to the debugger
221  */
222 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
223 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
224 
225 /*
226  * Segment for relocated kernel structures in 64-bit large RAM kernels
227  */
228 struct seg kmem64;
229 
230 struct memseg *memseg_free;
231 
232 struct vnode unused_pages_vp;
233 
234 /*
235  * VM data structures allocated early during boot.
236  */
237 size_t pagehash_sz;
238 uint64_t memlist_sz;
239 
240 char tbr_wr_addr_inited = 0;
241 
242 caddr_t	mpo_heap32_buf = NULL;
243 size_t	mpo_heap32_bufsz = 0;
244 
245 /*
246  * Static Routines:
247  */
248 static int ndata_alloc_memseg(struct memlist *, size_t);
249 static void memlist_new(uint64_t, uint64_t, struct memlist **);
250 static void memlist_add(uint64_t, uint64_t,
251 	struct memlist **, struct memlist **);
252 static void kphysm_init(void);
253 static void kvm_init(void);
254 static void install_kmem64_tte(void);
255 
256 static void startup_init(void);
257 static void startup_memlist(void);
258 static void startup_modules(void);
259 static void startup_bop_gone(void);
260 static void startup_vm(void);
261 static void startup_end(void);
262 static void setup_cage_params(void);
263 static void startup_create_io_node(void);
264 
265 static pgcnt_t npages;
266 static struct memlist *memlist;
267 void *memlist_end;
268 
269 static pgcnt_t bop_alloc_pages;
270 static caddr_t hblk_base;
271 uint_t hblk_alloc_dynamic = 0;
272 uint_t hblk1_min = H1MIN;
273 
274 
275 /*
276  * Hooks for unsupported platforms and down-rev firmware
277  */
278 int iam_positron(void);
279 #pragma weak iam_positron
280 static void do_prom_version_check(void);
281 
282 /*
283  * After receiving a thermal interrupt, this is the number of seconds
284  * to delay before shutting off the system, assuming
285  * shutdown fails.  Use /etc/system to change the delay if this isn't
286  * large enough.
287  */
288 int thermal_powerdown_delay = 1200;
289 
290 /*
291  * Used to hold off page relocations into the cage until OBP has completed
292  * its boot-time handoff of its resources to the kernel.
293  */
294 int page_relocate_ready = 0;
295 
296 /*
297  * Indicate if kmem64 allocation was done in small chunks
298  */
299 int kmem64_smchunks = 0;
300 
301 /*
302  * Enable some debugging messages concerning memory usage...
303  */
304 #ifdef  DEBUGGING_MEM
305 static int debugging_mem;
306 static void
307 printmemlist(char *title, struct memlist *list)
308 {
309 	if (!debugging_mem)
310 		return;
311 
312 	printf("%s\n", title);
313 
314 	while (list) {
315 		prom_printf("\taddr = 0x%x %8x, size = 0x%x %8x\n",
316 		    (uint32_t)(list->address >> 32), (uint32_t)list->address,
317 		    (uint32_t)(list->size >> 32), (uint32_t)(list->size));
318 		list = list->next;
319 	}
320 }
321 
322 void
323 printmemseg(struct memseg *memseg)
324 {
325 	if (!debugging_mem)
326 		return;
327 
328 	printf("memseg\n");
329 
330 	while (memseg) {
331 		prom_printf("\tpage = 0x%p, epage = 0x%p, "
332 		    "pfn = 0x%x, epfn = 0x%x\n",
333 		    memseg->pages, memseg->epages,
334 		    memseg->pages_base, memseg->pages_end);
335 		memseg = memseg->next;
336 	}
337 }
338 
339 #define	debug_pause(str)	halt((str))
340 #define	MPRINTF(str)		if (debugging_mem) prom_printf((str))
341 #define	MPRINTF1(str, a)	if (debugging_mem) prom_printf((str), (a))
342 #define	MPRINTF2(str, a, b)	if (debugging_mem) prom_printf((str), (a), (b))
343 #define	MPRINTF3(str, a, b, c) \
344 	if (debugging_mem) prom_printf((str), (a), (b), (c))
345 #else	/* DEBUGGING_MEM */
346 #define	MPRINTF(str)
347 #define	MPRINTF1(str, a)
348 #define	MPRINTF2(str, a, b)
349 #define	MPRINTF3(str, a, b, c)
350 #endif	/* DEBUGGING_MEM */
351 
352 
353 /*
354  *
355  *                    Kernel's Virtual Memory Layout.
356  *                       /-----------------------\
357  * 0xFFFFFFFF.FFFFFFFF  -|                       |-
358  *                       |   OBP's virtual page  |
359  *                       |        tables         |
360  * 0xFFFFFFFC.00000000  -|-----------------------|-
361  *                       :                       :
362  *                       :                       :
363  *                      -|-----------------------|-
364  *                       |       segzio          | (base and size vary)
365  * 0xFFFFFE00.00000000  -|-----------------------|-
366  *                       |                       |  Ultrasparc I/II support
367  *                       |    segkpm segment     |  up to 2TB of physical
368  *                       | (64-bit kernel ONLY)  |  memory, VAC has 2 colors
369  *                       |                       |
370  * 0xFFFFFA00.00000000  -|-----------------------|- 2TB segkpm alignment
371  *                       :                       :
372  *                       :                       :
373  * 0xFFFFF810.00000000  -|-----------------------|- hole_end
374  *                       |                       |      ^
375  *                       |  UltraSPARC I/II call |      |
376  *                       | bug requires an extra |      |
377  *                       | 4 GB of space between |      |
378  *                       |   hole and used RAM   |	|
379  *                       |                       |      |
380  * 0xFFFFF800.00000000  -|-----------------------|-     |
381  *                       |                       |      |
382  *                       | Virtual Address Hole  |   UltraSPARC
383  *                       |  on UltraSPARC I/II   |  I/II * ONLY *
384  *                       |                       |      |
385  * 0x00000800.00000000  -|-----------------------|-     |
386  *                       |                       |      |
387  *                       |  UltraSPARC I/II call |      |
388  *                       | bug requires an extra |      |
389  *                       | 4 GB of space between |      |
390  *                       |   hole and used RAM   |      |
391  *                       |                       |      v
392  * 0x000007FF.00000000  -|-----------------------|- hole_start -----
393  *                       :                       :		   ^
394  *                       :                       :		   |
395  *                       |-----------------------|                 |
396  *                       |                       |                 |
397  *                       |  ecache flush area    |                 |
398  *                       |  (twice largest e$)   |                 |
399  *                       |                       |                 |
400  * 0x00000XXX.XXX00000  -|-----------------------|- kmem64_	   |
401  *                       | overmapped area       |   alignend_end  |
402  *                       | (kmem64_alignsize     |		   |
403  *                       |  boundary)            |		   |
404  * 0x00000XXX.XXXXXXXX  -|-----------------------|- kmem64_end	   |
405  *                       |                       |		   |
406  *                       |   64-bit kernel ONLY  |		   |
407  *                       |                       |		   |
408  *                       |    kmem64 segment     |		   |
409  *                       |                       |		   |
410  *                       | (Relocated extra HME  |	     Approximately
411  *                       |   block allocations,  |	    1 TB of virtual
412  *                       |   memnode freelists,  |	     address space
413  *                       |    HME hash buckets,  |		   |
414  *                       | mml_table, kpmp_table,|		   |
415  *                       |  page_t array and     |		   |
416  *                       |  hashblock pool to    |		   |
417  *                       |   avoid hard-coded    |		   |
418  *                       |     32-bit vaddr      |		   |
419  *                       |     limitations)      |		   |
420  *                       |                       |		   v
421  * 0x00000700.00000000  -|-----------------------|- SYSLIMIT (kmem64_base)
422  *                       |                       |
423  *                       |  segkmem segment      | (SYSLIMIT - SYSBASE = 4TB)
424  *                       |                       |
425  * 0x00000300.00000000  -|-----------------------|- SYSBASE
426  *                       :                       :
427  *                       :                       :
428  *                      -|-----------------------|-
429  *                       |                       |
430  *                       |  segmap segment       |   SEGMAPSIZE (1/8th physmem,
431  *                       |                       |               256G MAX)
432  * 0x000002a7.50000000  -|-----------------------|- SEGMAPBASE
433  *                       :                       :
434  *                       :                       :
435  *                      -|-----------------------|-
436  *                       |                       |
437  *                       |       segkp           |    SEGKPSIZE (2GB)
438  *                       |                       |
439  *                       |                       |
440  * 0x000002a1.00000000  -|-----------------------|- SEGKPBASE
441  *                       |                       |
442  * 0x000002a0.00000000  -|-----------------------|- MEMSCRUBBASE
443  *                       |                       |       (SEGKPBASE - 0x400000)
444  * 0x0000029F.FFE00000  -|-----------------------|- ARGSBASE
445  *                       |                       |       (MEMSCRUBBASE - NCARGS)
446  * 0x0000029F.FFD80000  -|-----------------------|- PPMAPBASE
447  *                       |                       |       (ARGSBASE - PPMAPSIZE)
448  * 0x0000029F.FFD00000  -|-----------------------|- PPMAP_FAST_BASE
449  *                       |                       |
450  * 0x0000029F.FF980000  -|-----------------------|- PIOMAPBASE
451  *                       |                       |
452  * 0x0000029F.FF580000  -|-----------------------|- NARG_BASE
453  *                       :                       :
454  *                       :                       :
455  * 0x00000000.FFFFFFFF  -|-----------------------|- OFW_END_ADDR
456  *                       |                       |
457  *                       |         OBP           |
458  *                       |                       |
459  * 0x00000000.F0000000  -|-----------------------|- OFW_START_ADDR
460  *                       |         kmdb          |
461  * 0x00000000.EDD00000  -|-----------------------|- SEGDEBUGBASE
462  *                       :                       :
463  *                       :                       :
464  * 0x00000000.7c000000  -|-----------------------|- SYSLIMIT32
465  *                       |                       |
466  *                       |  segkmem32 segment    | (SYSLIMIT32 - SYSBASE32 =
467  *                       |                       |    ~64MB)
468  * 0x00000000.70002000  -|-----------------------|
469  *                       |     panicbuf          |
470  * 0x00000000.70000000  -|-----------------------|- SYSBASE32
471  *                       |       boot-time       |
472  *                       |    temporary space    |
473  * 0x00000000.4C000000  -|-----------------------|- BOOTTMPBASE
474  *                       :                       :
475  *                       :                       :
476  *                       |                       |
477  *                       |-----------------------|- econtig32
478  *                       |    vm structures      |
479  * 0x00000000.01C00000   |-----------------------|- nalloc_end
480  *                       |         TSBs          |
481  *                       |-----------------------|- end/nalloc_base
482  *                       |   kernel data & bss   |
483  * 0x00000000.01800000  -|-----------------------|
484  *                       :   nucleus text hole   :
485  * 0x00000000.01400000  -|-----------------------|
486  *                       :                       :
487  *                       |-----------------------|
488  *                       |      module text      |
489  *                       |-----------------------|- e_text/modtext
490  *                       |      kernel text      |
491  *                       |-----------------------|
492  *                       |    trap table (48k)   |
493  * 0x00000000.01000000  -|-----------------------|- KERNELBASE
494  *                       | reserved for trapstat |} TSTAT_TOTAL_SIZE
495  *                       |-----------------------|
496  *                       |                       |
497  *                       |        invalid        |
498  *                       |                       |
499  * 0x00000000.00000000  _|_______________________|
500  *
501  *
502  *
503  *                   32-bit User Virtual Memory Layout.
504  *                       /-----------------------\
505  *                       |                       |
506  *                       |        invalid        |
507  *                       |                       |
508  *          0xFFC00000  -|-----------------------|- USERLIMIT
509  *                       |       user stack      |
510  *                       :                       :
511  *                       :                       :
512  *                       :                       :
513  *                       |       user data       |
514  *                      -|-----------------------|-
515  *                       |       user text       |
516  *          0x00002000  -|-----------------------|-
517  *                       |       invalid         |
518  *          0x00000000  _|_______________________|
519  *
520  *
521  *
522  *                   64-bit User Virtual Memory Layout.
523  *                       /-----------------------\
524  *                       |                       |
525  *                       |        invalid        |
526  *                       |                       |
527  *  0xFFFFFFFF.80000000 -|-----------------------|- USERLIMIT
528  *                       |       user stack      |
529  *                       :                       :
530  *                       :                       :
531  *                       :                       :
532  *                       |       user data       |
533  *                      -|-----------------------|-
534  *                       |       user text       |
535  *  0x00000000.01000000 -|-----------------------|-
536  *                       |       invalid         |
537  *  0x00000000.00000000 _|_______________________|
538  */
539 
540 extern caddr_t ecache_init_scrub_flush_area(caddr_t alloc_base);
541 extern uint64_t ecache_flush_address(void);
542 
543 #pragma weak load_platform_modules
544 #pragma weak plat_startup_memlist
545 #pragma weak ecache_init_scrub_flush_area
546 #pragma weak ecache_flush_address
547 
548 
549 /*
550  * By default the DR Cage is enabled for maximum OS
551  * MPSS performance.  Users needing to disable the cage mechanism
552  * can set this variable to zero via /etc/system.
553  * Disabling the cage on systems supporting Dynamic Reconfiguration (DR)
554  * will result in loss of DR functionality.
555  * Platforms wishing to disable kernel Cage by default
556  * should do so in their set_platform_defaults() routine.
557  */
558 int	kernel_cage_enable = 1;
559 
560 static void
561 setup_cage_params(void)
562 {
563 	void (*func)(void);
564 
565 	func = (void (*)(void))kobj_getsymvalue("set_platform_cage_params", 0);
566 	if (func != NULL) {
567 		(*func)();
568 		return;
569 	}
570 
571 	if (kernel_cage_enable == 0) {
572 		return;
573 	}
574 	kcage_range_init(phys_avail, KCAGE_DOWN, total_pages / 256);
575 
576 	if (kcage_on) {
577 		cmn_err(CE_NOTE, "!Kernel Cage is ENABLED");
578 	} else {
579 		cmn_err(CE_NOTE, "!Kernel Cage is DISABLED");
580 	}
581 
582 }
583 
584 /*
585  * Machine-dependent startup code
586  */
587 void
588 startup(void)
589 {
590 	startup_init();
591 	if (&startup_platform)
592 		startup_platform();
593 	startup_memlist();
594 	startup_modules();
595 	setup_cage_params();
596 	startup_bop_gone();
597 	startup_vm();
598 	startup_end();
599 }
600 
601 struct regs sync_reg_buf;
602 uint64_t sync_tt;
603 
604 void
605 sync_handler(void)
606 {
607 	struct  panic_trap_info 	ti;
608 	int i;
609 
610 	/*
611 	 * Prevent trying to talk to the other CPUs since they are
612 	 * sitting in the prom and won't reply.
613 	 */
614 	for (i = 0; i < NCPU; i++) {
615 		if ((i != CPU->cpu_id) && CPU_XCALL_READY(i)) {
616 			cpu[i]->cpu_flags &= ~CPU_READY;
617 			cpu[i]->cpu_flags |= CPU_QUIESCED;
618 			CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
619 		}
620 	}
621 
622 	/*
623 	 * We've managed to get here without going through the
624 	 * normal panic code path. Try and save some useful
625 	 * information.
626 	 */
627 	if (!panicstr && (curthread->t_panic_trap == NULL)) {
628 		ti.trap_type = sync_tt;
629 		ti.trap_regs = &sync_reg_buf;
630 		ti.trap_addr = NULL;
631 		ti.trap_mmu_fsr = 0x0;
632 
633 		curthread->t_panic_trap = &ti;
634 	}
635 
636 	/*
637 	 * If we're re-entering the panic path, update the signature
638 	 * block so that the SC knows we're in the second part of panic.
639 	 */
640 	if (panicstr)
641 		CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
642 
643 	nopanicdebug = 1; /* do not perform debug_enter() prior to dump */
644 	panic("sync initiated");
645 }
646 
647 
648 static void
649 startup_init(void)
650 {
651 	/*
652 	 * We want to save the registers while we're still in OBP
653 	 * so that we know they haven't been fiddled with since.
654 	 * (In principle, OBP can't change them just because it
655 	 * makes a callback, but we'd rather not depend on that
656 	 * behavior.)
657 	 */
658 	char		sync_str[] =
659 	    "warning @ warning off : sync "
660 	    "%%tl-c %%tstate h# %p x! "
661 	    "%%g1 h# %p x! %%g2 h# %p x! %%g3 h# %p x! "
662 	    "%%g4 h# %p x! %%g5 h# %p x! %%g6 h# %p x! "
663 	    "%%g7 h# %p x! %%o0 h# %p x! %%o1 h# %p x! "
664 	    "%%o2 h# %p x! %%o3 h# %p x! %%o4 h# %p x! "
665 	    "%%o5 h# %p x! %%o6 h# %p x! %%o7 h# %p x! "
666 	    "%%tl-c %%tpc h# %p x! %%tl-c %%tnpc h# %p x! "
667 	    "%%y h# %p l! %%tl-c %%tt h# %p x! "
668 	    "sync ; warning !";
669 
670 	/*
671 	 * 20 == num of %p substrings
672 	 * 16 == max num of chars %p will expand to.
673 	 */
674 	char 		bp[sizeof (sync_str) + 16 * 20];
675 
676 	/*
677 	 * Initialize ptl1 stack for the 1st CPU.
678 	 */
679 	ptl1_init_cpu(&cpu0);
680 
681 	/*
682 	 * Initialize the address map for cache consistent mappings
683 	 * to random pages; must be done after vac_size is set.
684 	 */
685 	ppmapinit();
686 
687 	/*
688 	 * Initialize the PROM callback handler.
689 	 */
690 	init_vx_handler();
691 
692 	/*
693 	 * have prom call sync_callback() to handle the sync and
694 	 * save some useful information which will be stored in the
695 	 * core file later.
696 	 */
697 	(void) sprintf((char *)bp, sync_str,
698 	    (void *)&sync_reg_buf.r_tstate, (void *)&sync_reg_buf.r_g1,
699 	    (void *)&sync_reg_buf.r_g2, (void *)&sync_reg_buf.r_g3,
700 	    (void *)&sync_reg_buf.r_g4, (void *)&sync_reg_buf.r_g5,
701 	    (void *)&sync_reg_buf.r_g6, (void *)&sync_reg_buf.r_g7,
702 	    (void *)&sync_reg_buf.r_o0, (void *)&sync_reg_buf.r_o1,
703 	    (void *)&sync_reg_buf.r_o2, (void *)&sync_reg_buf.r_o3,
704 	    (void *)&sync_reg_buf.r_o4, (void *)&sync_reg_buf.r_o5,
705 	    (void *)&sync_reg_buf.r_o6, (void *)&sync_reg_buf.r_o7,
706 	    (void *)&sync_reg_buf.r_pc, (void *)&sync_reg_buf.r_npc,
707 	    (void *)&sync_reg_buf.r_y, (void *)&sync_tt);
708 	prom_interpret(bp, 0, 0, 0, 0, 0);
709 	add_vx_handler("sync", 1, (void (*)(cell_t *))sync_handler);
710 }
711 
712 
713 size_t
714 calc_pp_sz(pgcnt_t npages)
715 {
716 
717 	return (npages * sizeof (struct page));
718 }
719 
720 size_t
721 calc_kpmpp_sz(pgcnt_t npages)
722 {
723 
724 	kpm_pgshft = (kpm_smallpages == 0) ? MMU_PAGESHIFT4M : MMU_PAGESHIFT;
725 	kpm_pgsz = 1ull << kpm_pgshft;
726 	kpm_pgoff = kpm_pgsz - 1;
727 	kpmp2pshft = kpm_pgshft - PAGESHIFT;
728 	kpmpnpgs = 1 << kpmp2pshft;
729 
730 	if (kpm_smallpages == 0) {
731 		/*
732 		 * Avoid fragmentation problems in kphysm_init()
733 		 * by allocating for all of physical memory
734 		 */
735 		kpm_npages = ptokpmpr(physinstalled);
736 		return (kpm_npages * sizeof (kpm_page_t));
737 	} else {
738 		kpm_npages = npages;
739 		return (kpm_npages * sizeof (kpm_spage_t));
740 	}
741 }
742 
743 size_t
744 calc_pagehash_sz(pgcnt_t npages)
745 {
746 
747 	/*
748 	 * The page structure hash table size is a power of 2
749 	 * such that the average hash chain length is PAGE_HASHAVELEN.
750 	 */
751 	page_hashsz = npages / PAGE_HASHAVELEN;
752 	page_hashsz = 1 << highbit(page_hashsz);
753 	return (page_hashsz * sizeof (struct page *));
754 }
755 
756 int testkmem64_smchunks = 0;
757 
758 int
759 alloc_kmem64(caddr_t base, caddr_t end)
760 {
761 	int i;
762 	caddr_t aligned_end = NULL;
763 
764 	if (testkmem64_smchunks)
765 		return (1);
766 
767 	/*
768 	 * Make one large memory alloc after figuring out the 64-bit size. This
769 	 * will enable use of the largest page size appropriate for the system
770 	 * architecture.
771 	 */
772 	ASSERT(mmu_exported_pagesize_mask & (1 << TTE8K));
773 	ASSERT(IS_P2ALIGNED(base, TTEBYTES(max_bootlp_tteszc)));
774 	for (i = max_bootlp_tteszc; i >= TTE8K; i--) {
775 		size_t alloc_size, alignsize;
776 #if !defined(C_OBP)
777 		unsigned long long pa;
778 #endif	/* !C_OBP */
779 
780 		if ((mmu_exported_pagesize_mask & (1 << i)) == 0)
781 			continue;
782 		alignsize = TTEBYTES(i);
783 		kmem64_szc = i;
784 
785 		/* limit page size for small memory */
786 		if (mmu_btop(alignsize) > (npages >> 2))
787 			continue;
788 
789 		aligned_end = (caddr_t)roundup((uintptr_t)end, alignsize);
790 		alloc_size = aligned_end - base;
791 #if !defined(C_OBP)
792 		if (prom_allocate_phys(alloc_size, alignsize, &pa) == 0) {
793 			if (prom_claim_virt(alloc_size, base) != (caddr_t)-1) {
794 				kmem64_pabase = pa;
795 				kmem64_aligned_end = aligned_end;
796 				install_kmem64_tte();
797 				break;
798 			} else {
799 				prom_free_phys(alloc_size, pa);
800 			}
801 		}
802 #else	/* !C_OBP */
803 		if (prom_alloc(base, alloc_size, alignsize) == base) {
804 			kmem64_pabase = va_to_pa(kmem64_base);
805 			kmem64_aligned_end = aligned_end;
806 			break;
807 		}
808 #endif	/* !C_OBP */
809 		if (i == TTE8K) {
810 #ifdef sun4v
811 			/* return failure to try small allocations */
812 			return (1);
813 #else
814 			prom_panic("kmem64 allocation failure");
815 #endif
816 		}
817 	}
818 	ASSERT(aligned_end != NULL);
819 	return (0);
820 }
821 
822 static prom_memlist_t *boot_physinstalled, *boot_physavail, *boot_virtavail;
823 static size_t boot_physinstalled_len, boot_physavail_len, boot_virtavail_len;
824 
825 #define	IVSIZE	roundup(((MAXIVNUM * sizeof (intr_vec_t *)) + \
826 			(MAX_RSVD_IV * sizeof (intr_vec_t)) + \
827 			(MAX_RSVD_IVX * sizeof (intr_vecx_t))), PAGESIZE)
828 
829 #if !defined(C_OBP)
830 /*
831  * Install a temporary tte handler in OBP for kmem64 area.
832  *
833  * We map kmem64 area with large pages before the trap table is taken
834  * over. Since OBP makes 8K mappings, it can create 8K tlb entries in
835  * the same area. Duplicate tlb entries with different page sizes
836  * cause unpredicatble behavior.  To avoid this, we don't create
837  * kmem64 mappings via BOP_ALLOC (ends up as prom_alloc() call to
838  * OBP).  Instead, we manage translations with a temporary va>tte-data
839  * handler (kmem64-tte).  This handler is replaced by unix-tte when
840  * the trap table is taken over.
841  *
842  * The temporary handler knows the physical address of the kmem64
843  * area. It uses the prom's pgmap@ Forth word for other addresses.
844  *
845  * We have to use BOP_ALLOC() method for C-OBP platforms because
846  * pgmap@ is not defined in C-OBP. C-OBP is only used on serengeti
847  * sun4u platforms. On sun4u we flush tlb after trap table is taken
848  * over if we use large pages for kernel heap and kmem64. Since sun4u
849  * prom (unlike sun4v) calls va>tte-data first for client address
850  * translation prom's ttes for kmem64 can't get into TLB even if we
851  * later switch to prom's trap table again. C-OBP uses 4M pages for
852  * client mappings when possible so on all platforms we get the
853  * benefit from large mappings for kmem64 area immediately during
854  * boot.
855  *
856  * pseudo code:
857  * if (context != 0) {
858  * 	return false
859  * } else if (miss_va in range[kmem64_base, kmem64_end)) {
860  *	tte = tte_template +
861  *		(((miss_va & pagemask) - kmem64_base));
862  *	return tte, true
863  * } else {
864  *	return pgmap@ result
865  * }
866  */
867 char kmem64_obp_str[] =
868 	"h# %lx constant kmem64-base "
869 	"h# %lx constant kmem64-end "
870 	"h# %lx constant kmem64-pagemask "
871 	"h# %lx constant kmem64-template "
872 
873 	": kmem64-tte ( addr cnum -- false | tte-data true ) "
874 	"    if                                       ( addr ) "
875 	"       drop false exit then                  ( false ) "
876 	"    dup  kmem64-base kmem64-end  within  if  ( addr ) "
877 	"	kmem64-pagemask and                   ( addr' ) "
878 	"	kmem64-base -                         ( addr' ) "
879 	"	kmem64-template +                     ( tte ) "
880 	"	true                                  ( tte true ) "
881 	"    else                                     ( addr ) "
882 	"	pgmap@                                ( tte ) "
883 	"       dup 0< if true else drop false then   ( tte true  |  false ) "
884 	"    then                                     ( tte true  |  false ) "
885 	"; "
886 
887 	"' kmem64-tte is va>tte-data "
888 ;
889 
890 static void
891 install_kmem64_tte()
892 {
893 	char b[sizeof (kmem64_obp_str) + (4 * 16)];
894 	tte_t tte;
895 
896 	PRM_DEBUG(kmem64_pabase);
897 	PRM_DEBUG(kmem64_szc);
898 	sfmmu_memtte(&tte, kmem64_pabase >> MMU_PAGESHIFT,
899 	    PROC_DATA | HAT_NOSYNC | HAT_ATTR_NOSOFTEXEC, kmem64_szc);
900 	PRM_DEBUG(tte.ll);
901 	(void) sprintf(b, kmem64_obp_str,
902 	    kmem64_base, kmem64_end, TTE_PAGEMASK(kmem64_szc), tte.ll);
903 	ASSERT(strlen(b) < sizeof (b));
904 	prom_interpret(b, 0, 0, 0, 0, 0);
905 }
906 #endif	/* !C_OBP */
907 
908 /*
909  * As OBP takes up some RAM when the system boots, pages will already be "lost"
910  * to the system and reflected in npages by the time we see it.
911  *
912  * We only want to allocate kernel structures in the 64-bit virtual address
913  * space on systems with enough RAM to make the overhead of keeping track of
914  * an extra kernel memory segment worthwhile.
915  *
916  * Since OBP has already performed its memory allocations by this point, if we
917  * have more than MINMOVE_RAM_MB MB of RAM left free, go ahead and map
918  * memory in the 64-bit virtual address space; otherwise keep allocations
919  * contiguous with we've mapped so far in the 32-bit virtual address space.
920  */
921 #define	MINMOVE_RAM_MB	((size_t)1900)
922 #define	MB_TO_BYTES(mb)	((mb) * 1048576ul)
923 #define	BYTES_TO_MB(b) ((b) / 1048576ul)
924 
925 pgcnt_t	tune_npages = (pgcnt_t)
926 	(MB_TO_BYTES(MINMOVE_RAM_MB)/ (size_t)MMU_PAGESIZE);
927 
928 #pragma weak page_set_colorequiv_arr_cpu
929 extern void page_set_colorequiv_arr_cpu(void);
930 extern void page_set_colorequiv_arr(void);
931 
932 static pgcnt_t ramdisk_npages;
933 static struct memlist *old_phys_avail;
934 
935 kcage_dir_t kcage_startup_dir = KCAGE_DOWN;
936 
937 static void
938 startup_memlist(void)
939 {
940 	size_t hmehash_sz, pagelist_sz, tt_sz;
941 	size_t psetable_sz;
942 	caddr_t alloc_base;
943 	caddr_t memspace;
944 	struct memlist *cur;
945 	size_t syslimit = (size_t)SYSLIMIT;
946 	size_t sysbase = (size_t)SYSBASE;
947 
948 	/*
949 	 * Initialize enough of the system to allow kmem_alloc to work by
950 	 * calling boot to allocate its memory until the time that
951 	 * kvm_init is completed.  The page structs are allocated after
952 	 * rounding up end to the nearest page boundary; the memsegs are
953 	 * initialized and the space they use comes from the kernel heap.
954 	 * With appropriate initialization, they can be reallocated later
955 	 * to a size appropriate for the machine's configuration.
956 	 *
957 	 * At this point, memory is allocated for things that will never
958 	 * need to be freed, this used to be "valloced".  This allows a
959 	 * savings as the pages don't need page structures to describe
960 	 * them because them will not be managed by the vm system.
961 	 */
962 
963 	/*
964 	 * We're loaded by boot with the following configuration (as
965 	 * specified in the sun4u/conf/Mapfile):
966 	 *
967 	 * 	text:		4 MB chunk aligned on a 4MB boundary
968 	 * 	data & bss:	4 MB chunk aligned on a 4MB boundary
969 	 *
970 	 * These two chunks will eventually be mapped by 2 locked 4MB
971 	 * ttes and will represent the nucleus of the kernel.  This gives
972 	 * us some free space that is already allocated, some or all of
973 	 * which is made available to kernel module text.
974 	 *
975 	 * The free space in the data-bss chunk is used for nucleus
976 	 * allocatable data structures and we reserve it using the
977 	 * nalloc_base and nalloc_end variables.  This space is currently
978 	 * being used for hat data structures required for tlb miss
979 	 * handling operations.  We align nalloc_base to a l2 cache
980 	 * linesize because this is the line size the hardware uses to
981 	 * maintain cache coherency.
982 	 * 512K is carved out for module data.
983 	 */
984 
985 	moddata = (caddr_t)roundup((uintptr_t)e_data, MMU_PAGESIZE);
986 	e_moddata = moddata + MODDATA;
987 	nalloc_base = e_moddata;
988 
989 	nalloc_end = (caddr_t)roundup((uintptr_t)nalloc_base, MMU_PAGESIZE4M);
990 	valloc_base = nalloc_base;
991 
992 	/*
993 	 * Calculate the start of the data segment.
994 	 */
995 	if (((uintptr_t)e_moddata & MMU_PAGEMASK4M) != (uintptr_t)s_data)
996 		prom_panic("nucleus data overflow");
997 
998 	PRM_DEBUG(moddata);
999 	PRM_DEBUG(nalloc_base);
1000 	PRM_DEBUG(nalloc_end);
1001 
1002 	/*
1003 	 * Remember any slop after e_text so we can give it to the modules.
1004 	 */
1005 	PRM_DEBUG(e_text);
1006 	modtext = (caddr_t)roundup((uintptr_t)e_text, MMU_PAGESIZE);
1007 	if (((uintptr_t)e_text & MMU_PAGEMASK4M) != (uintptr_t)s_text)
1008 		prom_panic("nucleus text overflow");
1009 	modtext_sz = (caddr_t)roundup((uintptr_t)modtext, MMU_PAGESIZE4M) -
1010 	    modtext;
1011 	PRM_DEBUG(modtext);
1012 	PRM_DEBUG(modtext_sz);
1013 
1014 	init_boot_memlists();
1015 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1016 	    &boot_physavail, &boot_physavail_len,
1017 	    &boot_virtavail, &boot_virtavail_len);
1018 
1019 	/*
1020 	 * Remember what the physically available highest page is
1021 	 * so that dumpsys works properly, and find out how much
1022 	 * memory is installed.
1023 	 */
1024 	installed_top_size_memlist_array(boot_physinstalled,
1025 	    boot_physinstalled_len, &physmax, &physinstalled);
1026 	PRM_DEBUG(physinstalled);
1027 	PRM_DEBUG(physmax);
1028 
1029 	/* Fill out memory nodes config structure */
1030 	startup_build_mem_nodes(boot_physinstalled, boot_physinstalled_len);
1031 
1032 	/*
1033 	 * npages is the maximum of available physical memory possible.
1034 	 * (ie. it will never be more than this)
1035 	 *
1036 	 * When we boot from a ramdisk, the ramdisk memory isn't free, so
1037 	 * using phys_avail will underestimate what will end up being freed.
1038 	 * A better initial guess is just total memory minus the kernel text
1039 	 */
1040 	npages = physinstalled - btop(MMU_PAGESIZE4M);
1041 
1042 	/*
1043 	 * First allocate things that can go in the nucleus data page
1044 	 * (fault status, TSBs, dmv, CPUs)
1045 	 */
1046 	ndata_alloc_init(&ndata, (uintptr_t)nalloc_base, (uintptr_t)nalloc_end);
1047 
1048 	if ((&ndata_alloc_mmfsa != NULL) && (ndata_alloc_mmfsa(&ndata) != 0))
1049 		cmn_err(CE_PANIC, "no more nucleus memory after mfsa alloc");
1050 
1051 	if (ndata_alloc_tsbs(&ndata, npages) != 0)
1052 		cmn_err(CE_PANIC, "no more nucleus memory after tsbs alloc");
1053 
1054 	if (ndata_alloc_dmv(&ndata) != 0)
1055 		cmn_err(CE_PANIC, "no more nucleus memory after dmv alloc");
1056 
1057 	if (ndata_alloc_page_mutexs(&ndata) != 0)
1058 		cmn_err(CE_PANIC,
1059 		    "no more nucleus memory after page free lists alloc");
1060 
1061 	if (ndata_alloc_hat(&ndata, npages) != 0)
1062 		cmn_err(CE_PANIC, "no more nucleus memory after hat alloc");
1063 
1064 	if (ndata_alloc_memseg(&ndata, boot_physavail_len) != 0)
1065 		cmn_err(CE_PANIC, "no more nucleus memory after memseg alloc");
1066 
1067 	/*
1068 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1069 	 *
1070 	 * There are comments all over the SFMMU code warning of dire
1071 	 * consequences if the TSBs are moved out of 32-bit space.  This
1072 	 * is largely because the asm code uses "sethi %hi(addr)"-type
1073 	 * instructions which will not provide the expected result if the
1074 	 * address is a 64-bit one.
1075 	 *
1076 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1077 	 */
1078 	alloc_base = (caddr_t)roundup((uintptr_t)nalloc_end, MMU_PAGESIZE);
1079 	PRM_DEBUG(alloc_base);
1080 
1081 	alloc_base = sfmmu_ktsb_alloc(alloc_base);
1082 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1083 	PRM_DEBUG(alloc_base);
1084 
1085 	/*
1086 	 * Allocate IOMMU TSB array.  We do this here so that the physical
1087 	 * memory gets deducted from the PROM's physical memory list.
1088 	 */
1089 	alloc_base = iommu_tsb_init(alloc_base);
1090 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1091 	PRM_DEBUG(alloc_base);
1092 
1093 	/*
1094 	 * Allow for an early allocation of physically contiguous memory.
1095 	 */
1096 	alloc_base = contig_mem_prealloc(alloc_base, npages);
1097 
1098 	/*
1099 	 * Platforms like Starcat and OPL need special structures assigned in
1100 	 * 32-bit virtual address space because their probing routines execute
1101 	 * FCode, and FCode can't handle 64-bit virtual addresses...
1102 	 */
1103 	if (&plat_startup_memlist) {
1104 		alloc_base = plat_startup_memlist(alloc_base);
1105 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1106 		    ecache_alignsize);
1107 		PRM_DEBUG(alloc_base);
1108 	}
1109 
1110 	/*
1111 	 * Save off where the contiguous allocations to date have ended
1112 	 * in econtig32.
1113 	 */
1114 	econtig32 = alloc_base;
1115 	PRM_DEBUG(econtig32);
1116 	if (econtig32 > (caddr_t)KERNEL_LIMIT32)
1117 		cmn_err(CE_PANIC, "econtig32 too big");
1118 
1119 	pp_sz = calc_pp_sz(npages);
1120 	PRM_DEBUG(pp_sz);
1121 	if (kpm_enable) {
1122 		kpm_pp_sz = calc_kpmpp_sz(npages);
1123 		PRM_DEBUG(kpm_pp_sz);
1124 	}
1125 
1126 	hmehash_sz = calc_hmehash_sz(npages);
1127 	PRM_DEBUG(hmehash_sz);
1128 
1129 	pagehash_sz = calc_pagehash_sz(npages);
1130 	PRM_DEBUG(pagehash_sz);
1131 
1132 	pagelist_sz = calc_free_pagelist_sz();
1133 	PRM_DEBUG(pagelist_sz);
1134 
1135 #ifdef	TRAPTRACE
1136 	tt_sz = calc_traptrace_sz();
1137 	PRM_DEBUG(tt_sz);
1138 #else
1139 	tt_sz = 0;
1140 #endif	/* TRAPTRACE */
1141 
1142 	/*
1143 	 * Place the array that protects pp->p_selock in the kmem64 wad.
1144 	 */
1145 	pse_shift = size_pse_array(npages, max_ncpus);
1146 	PRM_DEBUG(pse_shift);
1147 	pse_table_size = 1 << pse_shift;
1148 	PRM_DEBUG(pse_table_size);
1149 	psetable_sz = roundup(
1150 	    pse_table_size * sizeof (pad_mutex_t), ecache_alignsize);
1151 	PRM_DEBUG(psetable_sz);
1152 
1153 	/*
1154 	 * Now allocate the whole wad
1155 	 */
1156 	kmem64_sz = pp_sz + kpm_pp_sz + hmehash_sz + pagehash_sz +
1157 	    pagelist_sz + tt_sz + psetable_sz;
1158 	kmem64_sz = roundup(kmem64_sz, PAGESIZE);
1159 	kmem64_base = (caddr_t)syslimit;
1160 	kmem64_end = kmem64_base + kmem64_sz;
1161 	if (alloc_kmem64(kmem64_base, kmem64_end)) {
1162 		/*
1163 		 * Attempt for kmem64 to allocate one big
1164 		 * contiguous chunk of memory failed.
1165 		 * We get here because we are sun4v.
1166 		 * We will proceed by breaking up
1167 		 * the allocation into two attempts.
1168 		 * First, we allocate kpm_pp_sz, hmehash_sz,
1169 		 * pagehash_sz, pagelist_sz, tt_sz & psetable_sz as
1170 		 * one contiguous chunk. This is a much smaller
1171 		 * chunk and we should get it, if not we panic.
1172 		 * Note that hmehash and tt need to be physically
1173 		 * (in the real address sense) contiguous.
1174 		 * Next, we use bop_alloc_chunk() to
1175 		 * to allocate the page_t structures.
1176 		 * This will allow the page_t to be allocated
1177 		 * in multiple smaller chunks.
1178 		 * In doing so, the assumption that page_t is
1179 		 * physically contiguous no longer hold, this is ok
1180 		 * for sun4v but not for sun4u.
1181 		 */
1182 		size_t  tmp_size;
1183 		caddr_t tmp_base;
1184 
1185 		pp_sz  = roundup(pp_sz, PAGESIZE);
1186 
1187 		/*
1188 		 * Allocate kpm_pp_sz, hmehash_sz,
1189 		 * pagehash_sz, pagelist_sz, tt_sz & psetable_sz
1190 		 */
1191 		tmp_base = kmem64_base + pp_sz;
1192 		tmp_size = roundup(kpm_pp_sz + hmehash_sz + pagehash_sz +
1193 		    pagelist_sz + tt_sz + psetable_sz, PAGESIZE);
1194 		if (prom_alloc(tmp_base, tmp_size, PAGESIZE) == 0)
1195 			prom_panic("kmem64 prom_alloc contig failed");
1196 		PRM_DEBUG(tmp_base);
1197 		PRM_DEBUG(tmp_size);
1198 
1199 		/*
1200 		 * Allocate the page_ts
1201 		 */
1202 		if (bop_alloc_chunk(kmem64_base, pp_sz, PAGESIZE) == 0)
1203 			prom_panic("kmem64 bop_alloc_chunk page_t failed");
1204 		PRM_DEBUG(kmem64_base);
1205 		PRM_DEBUG(pp_sz);
1206 
1207 		kmem64_aligned_end = kmem64_base + pp_sz + tmp_size;
1208 		ASSERT(kmem64_aligned_end >= kmem64_end);
1209 
1210 		kmem64_smchunks = 1;
1211 	} else {
1212 
1213 		/*
1214 		 * We need to adjust pp_sz for the normal
1215 		 * case where kmem64 can allocate one large chunk
1216 		 */
1217 		if (kpm_smallpages == 0) {
1218 			npages -= kmem64_sz / (PAGESIZE + sizeof (struct page));
1219 		} else {
1220 			npages -= kmem64_sz / (PAGESIZE + sizeof (struct page) +
1221 			    sizeof (kpm_spage_t));
1222 		}
1223 		pp_sz = npages * sizeof (struct page);
1224 	}
1225 
1226 	if (kmem64_aligned_end > (hole_start ? hole_start : kpm_vbase))
1227 		cmn_err(CE_PANIC, "not enough kmem64 space");
1228 	PRM_DEBUG(kmem64_base);
1229 	PRM_DEBUG(kmem64_end);
1230 	PRM_DEBUG(kmem64_aligned_end);
1231 
1232 	/*
1233 	 * ... and divy it up
1234 	 */
1235 	alloc_base = kmem64_base;
1236 
1237 	pp_base = (page_t *)alloc_base;
1238 	alloc_base += pp_sz;
1239 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1240 	PRM_DEBUG(pp_base);
1241 	PRM_DEBUG(npages);
1242 
1243 	if (kpm_enable) {
1244 		kpm_pp_base = alloc_base;
1245 		if (kpm_smallpages == 0) {
1246 			/* kpm_npages based on physinstalled, don't reset */
1247 			kpm_pp_sz = kpm_npages * sizeof (kpm_page_t);
1248 		} else {
1249 			kpm_npages = ptokpmpr(npages);
1250 			kpm_pp_sz = kpm_npages * sizeof (kpm_spage_t);
1251 		}
1252 		alloc_base += kpm_pp_sz;
1253 		alloc_base =
1254 		    (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1255 		PRM_DEBUG(kpm_pp_base);
1256 	}
1257 
1258 	alloc_base = alloc_hmehash(alloc_base);
1259 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1260 	PRM_DEBUG(alloc_base);
1261 
1262 	page_hash = (page_t **)alloc_base;
1263 	alloc_base += pagehash_sz;
1264 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1265 	PRM_DEBUG(page_hash);
1266 
1267 	alloc_base = alloc_page_freelists(alloc_base);
1268 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1269 	PRM_DEBUG(alloc_base);
1270 
1271 #ifdef	TRAPTRACE
1272 	ttrace_buf = alloc_base;
1273 	alloc_base += tt_sz;
1274 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1275 	PRM_DEBUG(alloc_base);
1276 #endif	/* TRAPTRACE */
1277 
1278 	pse_mutex = (pad_mutex_t *)alloc_base;
1279 	alloc_base += psetable_sz;
1280 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1281 	PRM_DEBUG(alloc_base);
1282 
1283 	/*
1284 	 * Note that if we use small chunk allocations for
1285 	 * kmem64, we need to ensure kmem64_end is the same as
1286 	 * kmem64_aligned_end to prevent subsequent logic from
1287 	 * trying to reuse the overmapping.
1288 	 * Otherwise we adjust kmem64_end to what we really allocated.
1289 	 */
1290 	if (kmem64_smchunks) {
1291 		kmem64_end = kmem64_aligned_end;
1292 	} else {
1293 		kmem64_end = (caddr_t)roundup((uintptr_t)alloc_base, PAGESIZE);
1294 	}
1295 	kmem64_sz = kmem64_end - kmem64_base;
1296 
1297 	if (&ecache_init_scrub_flush_area) {
1298 		alloc_base = ecache_init_scrub_flush_area(kmem64_aligned_end);
1299 		ASSERT(alloc_base <= (hole_start ? hole_start : kpm_vbase));
1300 	}
1301 
1302 	/*
1303 	 * If physmem is patched to be non-zero, use it instead of
1304 	 * the monitor value unless physmem is larger than the total
1305 	 * amount of memory on hand.
1306 	 */
1307 	if (physmem == 0 || physmem > npages)
1308 		physmem = npages;
1309 
1310 	/*
1311 	 * root_is_ramdisk is set via /etc/system when the ramdisk miniroot
1312 	 * is mounted as root. This memory is held down by OBP and unlike
1313 	 * the stub boot_archive is never released.
1314 	 *
1315 	 * In order to get things sized correctly on lower memory
1316 	 * machines (where the memory used by the ramdisk represents
1317 	 * a significant portion of memory), physmem is adjusted.
1318 	 *
1319 	 * This is done by subtracting the ramdisk_size which is set
1320 	 * to the size of the ramdisk (in Kb) in /etc/system at the
1321 	 * time the miniroot archive is constructed.
1322 	 */
1323 	if (root_is_ramdisk == B_TRUE) {
1324 		ramdisk_npages = (ramdisk_size * 1024) / PAGESIZE;
1325 		physmem -= ramdisk_npages;
1326 	}
1327 
1328 	if (kpm_enable && (ndata_alloc_kpm(&ndata, kpm_npages) != 0))
1329 		cmn_err(CE_PANIC, "no more nucleus memory after kpm alloc");
1330 
1331 	/*
1332 	 * Allocate space for the interrupt vector table.
1333 	 */
1334 	memspace = prom_alloc((caddr_t)intr_vec_table, IVSIZE, MMU_PAGESIZE);
1335 	if (memspace != (caddr_t)intr_vec_table)
1336 		prom_panic("interrupt vector table allocation failure");
1337 
1338 	/*
1339 	 * Between now and when we finish copying in the memory lists,
1340 	 * allocations happen so the space gets fragmented and the
1341 	 * lists longer.  Leave enough space for lists twice as
1342 	 * long as we have now; then roundup to a pagesize.
1343 	 */
1344 	memlist_sz = sizeof (struct memlist) * (prom_phys_installed_len() +
1345 	    prom_phys_avail_len() + prom_virt_avail_len());
1346 	memlist_sz *= 2;
1347 	memlist_sz = roundup(memlist_sz, PAGESIZE);
1348 	memspace = ndata_alloc(&ndata, memlist_sz, ecache_alignsize);
1349 	if (memspace == NULL)
1350 		cmn_err(CE_PANIC, "no more nucleus memory after memlist alloc");
1351 
1352 	memlist = (struct memlist *)memspace;
1353 	memlist_end = (char *)memspace + memlist_sz;
1354 	PRM_DEBUG(memlist);
1355 	PRM_DEBUG(memlist_end);
1356 
1357 	PRM_DEBUG(sysbase);
1358 	PRM_DEBUG(syslimit);
1359 	kernelheap_init((void *)sysbase, (void *)syslimit,
1360 	    (caddr_t)sysbase + PAGESIZE, NULL, NULL);
1361 
1362 	/*
1363 	 * Take the most current snapshot we can by calling mem-update.
1364 	 */
1365 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1366 	    &boot_physavail, &boot_physavail_len,
1367 	    &boot_virtavail, &boot_virtavail_len);
1368 
1369 	/*
1370 	 * Remove the space used by prom_alloc from the kernel heap
1371 	 * plus the area actually used by the OBP (if any)
1372 	 * ignoring virtual addresses in virt_avail, above syslimit.
1373 	 */
1374 	virt_avail = memlist;
1375 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1376 
1377 	for (cur = virt_avail; cur->next; cur = cur->next) {
1378 		uint64_t range_base, range_size;
1379 
1380 		if ((range_base = cur->address + cur->size) < (uint64_t)sysbase)
1381 			continue;
1382 		if (range_base >= (uint64_t)syslimit)
1383 			break;
1384 		/*
1385 		 * Limit the range to end at syslimit.
1386 		 */
1387 		range_size = MIN(cur->next->address,
1388 		    (uint64_t)syslimit) - range_base;
1389 		(void) vmem_xalloc(heap_arena, (size_t)range_size, PAGESIZE,
1390 		    0, 0, (void *)range_base, (void *)(range_base + range_size),
1391 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1392 	}
1393 
1394 	phys_avail = memlist;
1395 	copy_memlist(boot_physavail, boot_physavail_len, &memlist);
1396 
1397 	/*
1398 	 * Add any extra memory at the end of the ndata region if there's at
1399 	 * least a page to add.  There might be a few more pages available in
1400 	 * the middle of the ndata region, but for now they are ignored.
1401 	 */
1402 	nalloc_base = ndata_extra_base(&ndata, MMU_PAGESIZE, nalloc_end);
1403 	if (nalloc_base == NULL)
1404 		nalloc_base = nalloc_end;
1405 	ndata_remain_sz = nalloc_end - nalloc_base;
1406 
1407 	/*
1408 	 * Copy physinstalled list into kernel space.
1409 	 */
1410 	phys_install = memlist;
1411 	copy_memlist(boot_physinstalled, boot_physinstalled_len, &memlist);
1412 
1413 	/*
1414 	 * Create list of physical addrs we don't need pp's for:
1415 	 * kernel text 4M page
1416 	 * kernel data 4M page - ndata_remain_sz
1417 	 * kmem64 pages
1418 	 *
1419 	 * NB if adding any pages here, make sure no kpm page
1420 	 * overlaps can occur (see ASSERTs in kphysm_memsegs)
1421 	 */
1422 	nopp_list = memlist;
1423 	memlist_new(va_to_pa(s_text), MMU_PAGESIZE4M, &memlist);
1424 	memlist_add(va_to_pa(s_data), MMU_PAGESIZE4M - ndata_remain_sz,
1425 	    &memlist, &nopp_list);
1426 
1427 	/* Don't add to nopp_list if kmem64 was allocated in smchunks */
1428 	if (!kmem64_smchunks)
1429 		memlist_add(kmem64_pabase, kmem64_sz, &memlist, &nopp_list);
1430 
1431 	if ((caddr_t)memlist > (memspace + memlist_sz))
1432 		prom_panic("memlist overflow");
1433 
1434 	/*
1435 	 * Size the pcf array based on the number of cpus in the box at
1436 	 * boot time.
1437 	 */
1438 	pcf_init();
1439 
1440 	/*
1441 	 * Initialize the page structures from the memory lists.
1442 	 */
1443 	kphysm_init();
1444 
1445 	availrmem_initial = availrmem = freemem;
1446 	PRM_DEBUG(availrmem);
1447 
1448 	/*
1449 	 * Some of the locks depend on page_hashsz being set!
1450 	 * kmem_init() depends on this; so, keep it here.
1451 	 */
1452 	page_lock_init();
1453 
1454 	/*
1455 	 * Initialize kernel memory allocator.
1456 	 */
1457 	kmem_init();
1458 
1459 	/*
1460 	 * Factor in colorequiv to check additional 'equivalent' bins
1461 	 */
1462 	if (&page_set_colorequiv_arr_cpu != NULL)
1463 		page_set_colorequiv_arr_cpu();
1464 	else
1465 		page_set_colorequiv_arr();
1466 
1467 	/*
1468 	 * Initialize bp_mapin().
1469 	 */
1470 	bp_init(shm_alignment, HAT_STRICTORDER);
1471 
1472 	/*
1473 	 * Reserve space for panicbuf, intr_vec_table, reserved interrupt
1474 	 * vector data structures and MPO mblock structs from the 32-bit heap.
1475 	 */
1476 	(void) vmem_xalloc(heap32_arena, PANICBUFSIZE, PAGESIZE, 0, 0,
1477 	    panicbuf, panicbuf + PANICBUFSIZE,
1478 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1479 
1480 	(void) vmem_xalloc(heap32_arena, IVSIZE, PAGESIZE, 0, 0,
1481 	    intr_vec_table, (caddr_t)intr_vec_table + IVSIZE,
1482 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1483 
1484 	if (mpo_heap32_bufsz > (size_t)0) {
1485 		(void) vmem_xalloc(heap32_arena, mpo_heap32_bufsz,
1486 		    PAGESIZE, 0, 0, mpo_heap32_buf,
1487 		    mpo_heap32_buf + mpo_heap32_bufsz,
1488 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1489 	}
1490 	mem_config_init();
1491 }
1492 
1493 static void
1494 startup_modules(void)
1495 {
1496 	int nhblk1, nhblk8;
1497 	size_t  nhblksz;
1498 	pgcnt_t pages_per_hblk;
1499 	size_t hme8blk_sz, hme1blk_sz;
1500 
1501 	/*
1502 	 * Let the platforms have a chance to change default
1503 	 * values before reading system file.
1504 	 */
1505 	if (&set_platform_defaults)
1506 		set_platform_defaults();
1507 
1508 	/*
1509 	 * Calculate default settings of system parameters based upon
1510 	 * maxusers, yet allow to be overridden via the /etc/system file.
1511 	 */
1512 	param_calc(0);
1513 
1514 	mod_setup();
1515 
1516 	/*
1517 	 * If this is a positron, complain and halt.
1518 	 */
1519 	if (&iam_positron && iam_positron()) {
1520 		cmn_err(CE_WARN, "This hardware platform is not supported"
1521 		    " by this release of Solaris.\n");
1522 #ifdef DEBUG
1523 		prom_enter_mon();	/* Type 'go' to resume */
1524 		cmn_err(CE_WARN, "Booting an unsupported platform.\n");
1525 		cmn_err(CE_WARN, "Booting with down-rev firmware.\n");
1526 
1527 #else /* DEBUG */
1528 		halt(0);
1529 #endif /* DEBUG */
1530 	}
1531 
1532 	/*
1533 	 * If we are running firmware that isn't 64-bit ready
1534 	 * then complain and halt.
1535 	 */
1536 	do_prom_version_check();
1537 
1538 	/*
1539 	 * Initialize system parameters
1540 	 */
1541 	param_init();
1542 
1543 	/*
1544 	 * maxmem is the amount of physical memory we're playing with.
1545 	 */
1546 	maxmem = physmem;
1547 
1548 	/* Set segkp limits. */
1549 	ncbase = kdi_segdebugbase;
1550 	ncend = kdi_segdebugbase;
1551 
1552 	/*
1553 	 * Initialize the hat layer.
1554 	 */
1555 	hat_init();
1556 
1557 	/*
1558 	 * Initialize segment management stuff.
1559 	 */
1560 	seg_init();
1561 
1562 	/*
1563 	 * Create the va>tte handler, so the prom can understand
1564 	 * kernel translations.  The handler is installed later, just
1565 	 * as we are about to take over the trap table from the prom.
1566 	 */
1567 	create_va_to_tte();
1568 
1569 	/*
1570 	 * Load the forthdebugger (optional)
1571 	 */
1572 	forthdebug_init();
1573 
1574 	/*
1575 	 * Create OBP node for console input callbacks
1576 	 * if it is needed.
1577 	 */
1578 	startup_create_io_node();
1579 
1580 	if (modloadonly("fs", "specfs") == -1)
1581 		halt("Can't load specfs");
1582 
1583 	if (modloadonly("fs", "devfs") == -1)
1584 		halt("Can't load devfs");
1585 
1586 	if (modloadonly("misc", "swapgeneric") == -1)
1587 		halt("Can't load swapgeneric");
1588 
1589 	(void) modloadonly("sys", "lbl_edition");
1590 
1591 	dispinit();
1592 
1593 	/*
1594 	 * Infer meanings to the members of the idprom buffer.
1595 	 */
1596 	parse_idprom();
1597 
1598 	/* Read cluster configuration data. */
1599 	clconf_init();
1600 
1601 	setup_ddi();
1602 
1603 	/*
1604 	 * Lets take this opportunity to load the root device.
1605 	 */
1606 	if (loadrootmodules() != 0)
1607 		debug_enter("Can't load the root filesystem");
1608 
1609 	/*
1610 	 * Load tod driver module for the tod part found on this system.
1611 	 * Recompute the cpu frequency/delays based on tod as tod part
1612 	 * tends to keep time more accurately.
1613 	 */
1614 	if (&load_tod_module)
1615 		load_tod_module();
1616 
1617 	/*
1618 	 * Allow platforms to load modules which might
1619 	 * be needed after bootops are gone.
1620 	 */
1621 	if (&load_platform_modules)
1622 		load_platform_modules();
1623 
1624 	setcpudelay();
1625 
1626 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1627 	    &boot_physavail, &boot_physavail_len,
1628 	    &boot_virtavail, &boot_virtavail_len);
1629 
1630 	/*
1631 	 * Calculation and allocation of hmeblks needed to remap
1632 	 * the memory allocated by PROM till now.
1633 	 * Overestimate the number of hblk1 elements by assuming
1634 	 * worst case of TTE64K mappings.
1635 	 * sfmmu_hblk_alloc will panic if this calculation is wrong.
1636 	 */
1637 	bop_alloc_pages = btopr(kmem64_end - kmem64_base);
1638 	pages_per_hblk = btop(HMEBLK_SPAN(TTE64K));
1639 	bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1640 	nhblk1 = bop_alloc_pages / pages_per_hblk + hblk1_min;
1641 
1642 	bop_alloc_pages = size_virtalloc(boot_virtavail, boot_virtavail_len);
1643 
1644 	/* sfmmu_init_nucleus_hblks expects properly aligned data structures */
1645 	hme8blk_sz = roundup(HME8BLK_SZ, sizeof (int64_t));
1646 	hme1blk_sz = roundup(HME1BLK_SZ, sizeof (int64_t));
1647 
1648 	bop_alloc_pages += btopr(nhblk1 * hme1blk_sz);
1649 
1650 	pages_per_hblk = btop(HMEBLK_SPAN(TTE8K));
1651 	nhblk8 = 0;
1652 	while (bop_alloc_pages > 1) {
1653 		bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1654 		nhblk8 += bop_alloc_pages /= pages_per_hblk;
1655 		bop_alloc_pages *= hme8blk_sz;
1656 		bop_alloc_pages = btopr(bop_alloc_pages);
1657 	}
1658 	nhblk8 += 2;
1659 
1660 	/*
1661 	 * Since hblk8's can hold up to 64k of mappings aligned on a 64k
1662 	 * boundary, the number of hblk8's needed to map the entries in the
1663 	 * boot_virtavail list needs to be adjusted to take this into
1664 	 * consideration.  Thus, we need to add additional hblk8's since it
1665 	 * is possible that an hblk8 will not have all 8 slots used due to
1666 	 * alignment constraints.  Since there were boot_virtavail_len entries
1667 	 * in that list, we need to add that many hblk8's to the number
1668 	 * already calculated to make sure we don't underestimate.
1669 	 */
1670 	nhblk8 += boot_virtavail_len;
1671 	nhblksz = nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz;
1672 
1673 	/* Allocate in pagesize chunks */
1674 	nhblksz = roundup(nhblksz, MMU_PAGESIZE);
1675 	hblk_base = kmem_zalloc(nhblksz, KM_SLEEP);
1676 	sfmmu_init_nucleus_hblks(hblk_base, nhblksz, nhblk8, nhblk1);
1677 }
1678 
1679 static void
1680 startup_bop_gone(void)
1681 {
1682 
1683 	/*
1684 	 * Destroy the MD initialized at startup
1685 	 * The startup initializes the MD framework
1686 	 * using prom and BOP alloc free it now.
1687 	 */
1688 	mach_descrip_startup_fini();
1689 
1690 	/*
1691 	 * We're done with prom allocations.
1692 	 */
1693 	bop_fini();
1694 
1695 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1696 	    &boot_physavail, &boot_physavail_len,
1697 	    &boot_virtavail, &boot_virtavail_len);
1698 
1699 	/*
1700 	 * setup physically contiguous area twice as large as the ecache.
1701 	 * this is used while doing displacement flush of ecaches
1702 	 */
1703 	if (&ecache_flush_address) {
1704 		ecache_flushaddr = ecache_flush_address();
1705 		if (ecache_flushaddr == (uint64_t)-1) {
1706 			cmn_err(CE_PANIC,
1707 			    "startup: no memory to set ecache_flushaddr");
1708 		}
1709 	}
1710 
1711 	/*
1712 	 * Virtual available next.
1713 	 */
1714 	ASSERT(virt_avail != NULL);
1715 	memlist_free_list(virt_avail);
1716 	virt_avail = memlist;
1717 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1718 
1719 }
1720 
1721 
1722 /*
1723  * startup_fixup_physavail - called from mach_sfmmu.c after the final
1724  * allocations have been performed.  We can't call it in startup_bop_gone
1725  * since later operations can cause obp to allocate more memory.
1726  */
1727 void
1728 startup_fixup_physavail(void)
1729 {
1730 	struct memlist *cur;
1731 	size_t kmem64_overmap_size = kmem64_aligned_end - kmem64_end;
1732 
1733 	PRM_DEBUG(kmem64_overmap_size);
1734 
1735 	/*
1736 	 * take the most current snapshot we can by calling mem-update
1737 	 */
1738 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1739 	    &boot_physavail, &boot_physavail_len,
1740 	    &boot_virtavail, &boot_virtavail_len);
1741 
1742 	/*
1743 	 * Copy phys_avail list, again.
1744 	 * Both the kernel/boot and the prom have been allocating
1745 	 * from the original list we copied earlier.
1746 	 */
1747 	cur = memlist;
1748 	copy_memlist(boot_physavail, boot_physavail_len, &memlist);
1749 
1750 	/*
1751 	 * Add any unused kmem64 memory from overmapped page
1752 	 * (Note: va_to_pa does not work for kmem64_end)
1753 	 */
1754 	if (kmem64_overmap_size) {
1755 		memlist_add(kmem64_pabase + (kmem64_end - kmem64_base),
1756 		    kmem64_overmap_size, &memlist, &cur);
1757 	}
1758 
1759 	/*
1760 	 * Add any extra memory after e_data we added to the phys_avail list
1761 	 * back to the old list.
1762 	 */
1763 	if (ndata_remain_sz >= MMU_PAGESIZE)
1764 		memlist_add(va_to_pa(nalloc_base),
1765 		    (uint64_t)ndata_remain_sz, &memlist, &cur);
1766 
1767 	/*
1768 	 * There isn't any bounds checking on the memlist area
1769 	 * so ensure it hasn't overgrown.
1770 	 */
1771 	if ((caddr_t)memlist > (caddr_t)memlist_end)
1772 		cmn_err(CE_PANIC, "startup: memlist size exceeded");
1773 
1774 	/*
1775 	 * The kernel removes the pages that were allocated for it from
1776 	 * the freelist, but we now have to find any -extra- pages that
1777 	 * the prom has allocated for it's own book-keeping, and remove
1778 	 * them from the freelist too. sigh.
1779 	 */
1780 	sync_memlists(phys_avail, cur);
1781 
1782 	ASSERT(phys_avail != NULL);
1783 
1784 	old_phys_avail = phys_avail;
1785 	phys_avail = cur;
1786 }
1787 
1788 void
1789 update_kcage_ranges(uint64_t addr, uint64_t len)
1790 {
1791 	pfn_t base = btop(addr);
1792 	pgcnt_t num = btop(len);
1793 	int rv;
1794 
1795 	rv = kcage_range_add(base, num, kcage_startup_dir);
1796 
1797 	if (rv == ENOMEM) {
1798 		cmn_err(CE_WARN, "%ld megabytes not available to kernel cage",
1799 		    (len == 0 ? 0 : BYTES_TO_MB(len)));
1800 	} else if (rv != 0) {
1801 		/* catch this in debug kernels */
1802 		ASSERT(0);
1803 
1804 		cmn_err(CE_WARN, "unexpected kcage_range_add"
1805 		    " return value %d", rv);
1806 	}
1807 }
1808 
1809 static void
1810 startup_vm(void)
1811 {
1812 	size_t	i;
1813 	struct segmap_crargs a;
1814 	struct segkpm_crargs b;
1815 
1816 	uint64_t avmem;
1817 	caddr_t va;
1818 	pgcnt_t	max_phys_segkp;
1819 	int	mnode;
1820 
1821 	extern int use_brk_lpg, use_stk_lpg;
1822 
1823 	/*
1824 	 * get prom's mappings, create hments for them and switch
1825 	 * to the kernel context.
1826 	 */
1827 	hat_kern_setup();
1828 
1829 	/*
1830 	 * Take over trap table
1831 	 */
1832 	setup_trap_table();
1833 
1834 	/*
1835 	 * Install the va>tte handler, so that the prom can handle
1836 	 * misses and understand the kernel table layout in case
1837 	 * we need call into the prom.
1838 	 */
1839 	install_va_to_tte();
1840 
1841 	/*
1842 	 * Set a flag to indicate that the tba has been taken over.
1843 	 */
1844 	tba_taken_over = 1;
1845 
1846 	/* initialize MMU primary context register */
1847 	mmu_init_kcontext();
1848 
1849 	/*
1850 	 * The boot cpu can now take interrupts, x-calls, x-traps
1851 	 */
1852 	CPUSET_ADD(cpu_ready_set, CPU->cpu_id);
1853 	CPU->cpu_flags |= (CPU_READY | CPU_ENABLE | CPU_EXISTS);
1854 
1855 	/*
1856 	 * Set a flag to tell write_scb_int() that it can access V_TBR_WR_ADDR.
1857 	 */
1858 	tbr_wr_addr_inited = 1;
1859 
1860 	/*
1861 	 * Initialize VM system, and map kernel address space.
1862 	 */
1863 	kvm_init();
1864 
1865 	ASSERT(old_phys_avail != NULL && phys_avail != NULL);
1866 	if (kernel_cage_enable) {
1867 		diff_memlists(phys_avail, old_phys_avail, update_kcage_ranges);
1868 	}
1869 	memlist_free_list(old_phys_avail);
1870 
1871 	/*
1872 	 * If the following is true, someone has patched
1873 	 * phsymem to be less than the number of pages that
1874 	 * the system actually has.  Remove pages until system
1875 	 * memory is limited to the requested amount.  Since we
1876 	 * have allocated page structures for all pages, we
1877 	 * correct the amount of memory we want to remove
1878 	 * by the size of the memory used to hold page structures
1879 	 * for the non-used pages.
1880 	 */
1881 	if (physmem + ramdisk_npages < npages) {
1882 		pgcnt_t diff, off;
1883 		struct page *pp;
1884 		struct seg kseg;
1885 
1886 		cmn_err(CE_WARN, "limiting physmem to %ld pages", physmem);
1887 
1888 		off = 0;
1889 		diff = npages - (physmem + ramdisk_npages);
1890 		diff -= mmu_btopr(diff * sizeof (struct page));
1891 		kseg.s_as = &kas;
1892 		while (diff--) {
1893 			pp = page_create_va(&unused_pages_vp, (offset_t)off,
1894 			    MMU_PAGESIZE, PG_WAIT | PG_EXCL,
1895 			    &kseg, (caddr_t)off);
1896 			if (pp == NULL)
1897 				cmn_err(CE_PANIC, "limited physmem too much!");
1898 			page_io_unlock(pp);
1899 			page_downgrade(pp);
1900 			availrmem--;
1901 			off += MMU_PAGESIZE;
1902 		}
1903 	}
1904 
1905 	/*
1906 	 * When printing memory, show the total as physmem less
1907 	 * that stolen by a debugger.
1908 	 */
1909 	cmn_err(CE_CONT, "?mem = %ldK (0x%lx000)\n",
1910 	    (ulong_t)(physinstalled) << (PAGESHIFT - 10),
1911 	    (ulong_t)(physinstalled) << (PAGESHIFT - 12));
1912 
1913 	avmem = (uint64_t)freemem << PAGESHIFT;
1914 	cmn_err(CE_CONT, "?avail mem = %lld\n", (unsigned long long)avmem);
1915 
1916 	/*
1917 	 * For small memory systems disable automatic large pages.
1918 	 */
1919 	if (physmem < privm_lpg_min_physmem) {
1920 		use_brk_lpg = 0;
1921 		use_stk_lpg = 0;
1922 	}
1923 
1924 	/*
1925 	 * Perform platform specific freelist processing
1926 	 */
1927 	if (&plat_freelist_process) {
1928 		for (mnode = 0; mnode < max_mem_nodes; mnode++)
1929 			if (mem_node_config[mnode].exists)
1930 				plat_freelist_process(mnode);
1931 	}
1932 
1933 	/*
1934 	 * Initialize the segkp segment type.  We position it
1935 	 * after the configured tables and buffers (whose end
1936 	 * is given by econtig) and before V_WKBASE_ADDR.
1937 	 * Also in this area is segkmap (size SEGMAPSIZE).
1938 	 */
1939 
1940 	/* XXX - cache alignment? */
1941 	va = (caddr_t)SEGKPBASE;
1942 	ASSERT(((uintptr_t)va & PAGEOFFSET) == 0);
1943 
1944 	max_phys_segkp = (physmem * 2);
1945 
1946 	if (segkpsize < btop(SEGKPMINSIZE) || segkpsize > btop(SEGKPMAXSIZE)) {
1947 		segkpsize = btop(SEGKPDEFSIZE);
1948 		cmn_err(CE_WARN, "Illegal value for segkpsize. "
1949 		    "segkpsize has been reset to %ld pages", segkpsize);
1950 	}
1951 
1952 	i = ptob(MIN(segkpsize, max_phys_segkp));
1953 
1954 	rw_enter(&kas.a_lock, RW_WRITER);
1955 	if (seg_attach(&kas, va, i, segkp) < 0)
1956 		cmn_err(CE_PANIC, "startup: cannot attach segkp");
1957 	if (segkp_create(segkp) != 0)
1958 		cmn_err(CE_PANIC, "startup: segkp_create failed");
1959 	rw_exit(&kas.a_lock);
1960 
1961 	/*
1962 	 * kpm segment
1963 	 */
1964 	segmap_kpm = kpm_enable &&
1965 	    segmap_kpm && PAGESIZE == MAXBSIZE;
1966 
1967 	if (kpm_enable) {
1968 		rw_enter(&kas.a_lock, RW_WRITER);
1969 
1970 		/*
1971 		 * The segkpm virtual range range is larger than the
1972 		 * actual physical memory size and also covers gaps in
1973 		 * the physical address range for the following reasons:
1974 		 * . keep conversion between segkpm and physical addresses
1975 		 *   simple, cheap and unambiguous.
1976 		 * . avoid extension/shrink of the the segkpm in case of DR.
1977 		 * . avoid complexity for handling of virtual addressed
1978 		 *   caches, segkpm and the regular mapping scheme must be
1979 		 *   kept in sync wrt. the virtual color of mapped pages.
1980 		 * Any accesses to virtual segkpm ranges not backed by
1981 		 * physical memory will fall through the memseg pfn hash
1982 		 * and will be handled in segkpm_fault.
1983 		 * Additional kpm_size spaces needed for vac alias prevention.
1984 		 */
1985 		if (seg_attach(&kas, kpm_vbase, kpm_size * vac_colors,
1986 		    segkpm) < 0)
1987 			cmn_err(CE_PANIC, "cannot attach segkpm");
1988 
1989 		b.prot = PROT_READ | PROT_WRITE;
1990 		b.nvcolors = shm_alignment >> MMU_PAGESHIFT;
1991 
1992 		if (segkpm_create(segkpm, (caddr_t)&b) != 0)
1993 			panic("segkpm_create segkpm");
1994 
1995 		rw_exit(&kas.a_lock);
1996 
1997 		mach_kpm_init();
1998 	}
1999 
2000 	va = kpm_vbase + (kpm_size * vac_colors);
2001 
2002 	if (!segzio_fromheap) {
2003 		size_t size;
2004 		size_t physmem_b = mmu_ptob(physmem);
2005 
2006 		/* size is in bytes, segziosize is in pages */
2007 		if (segziosize == 0) {
2008 			size = physmem_b;
2009 		} else {
2010 			size = mmu_ptob(segziosize);
2011 		}
2012 
2013 		if (size < SEGZIOMINSIZE) {
2014 			size = SEGZIOMINSIZE;
2015 		} else if (size > SEGZIOMAXSIZE) {
2016 			size = SEGZIOMAXSIZE;
2017 			/*
2018 			 * On 64-bit x86, we only have 2TB of KVA.  This exists
2019 			 * for parity with x86.
2020 			 *
2021 			 * SEGZIOMAXSIZE is capped at 512gb so that segzio
2022 			 * doesn't consume all of KVA.  However, if we have a
2023 			 * system that has more thant 512gb of physical memory,
2024 			 * we can actually consume about half of the difference
2025 			 * between 512gb and the rest of the available physical
2026 			 * memory.
2027 			 */
2028 			if (physmem_b > SEGZIOMAXSIZE) {
2029 				size += (physmem_b - SEGZIOMAXSIZE) / 2;
2030 		}
2031 		}
2032 		segziosize = mmu_btop(roundup(size, MMU_PAGESIZE));
2033 		/* put the base of the ZIO segment after the kpm segment */
2034 		segzio_base = va;
2035 		va += mmu_ptob(segziosize);
2036 		PRM_DEBUG(segziosize);
2037 		PRM_DEBUG(segzio_base);
2038 
2039 		/*
2040 		 * On some platforms, kvm_init is called after the kpm
2041 		 * sizes have been determined.  On SPARC, kvm_init is called
2042 		 * before, so we have to attach the kzioseg after kvm is
2043 		 * initialized, otherwise we'll try to allocate from the boot
2044 		 * area since the kernel heap hasn't yet been configured.
2045 		 */
2046 		rw_enter(&kas.a_lock, RW_WRITER);
2047 
2048 		(void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2049 		    &kzioseg);
2050 		(void) segkmem_zio_create(&kzioseg);
2051 
2052 		/* create zio area covering new segment */
2053 		segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2054 
2055 		rw_exit(&kas.a_lock);
2056 	}
2057 
2058 	if (ppvm_enable) {
2059 		caddr_t ppvm_max;
2060 
2061 		/*
2062 		 * ppvm refers to the static VA space used to map
2063 		 * the page_t's for dynamically added memory.
2064 		 *
2065 		 * ppvm_base should not cross a potential VA hole.
2066 		 *
2067 		 * ppvm_size should be large enough to map the
2068 		 * page_t's needed to manage all of KPM range.
2069 		 */
2070 		ppvm_size =
2071 		    roundup(mmu_btop(kpm_size * vac_colors) * sizeof (page_t),
2072 		    MMU_PAGESIZE);
2073 		ppvm_max = (caddr_t)(0ull - ppvm_size);
2074 		ppvm_base = (page_t *)va;
2075 
2076 		if ((caddr_t)ppvm_base <= hole_end) {
2077 			cmn_err(CE_WARN,
2078 			    "Memory DR disabled: invalid DR map base: 0x%p\n",
2079 			    (void *)ppvm_base);
2080 			ppvm_enable = 0;
2081 		} else if ((caddr_t)ppvm_base > ppvm_max) {
2082 			uint64_t diff = (caddr_t)ppvm_base - ppvm_max;
2083 
2084 			cmn_err(CE_WARN,
2085 			    "Memory DR disabled: insufficient DR map size:"
2086 			    " 0x%lx (needed 0x%lx)\n",
2087 			    ppvm_size - diff, ppvm_size);
2088 			ppvm_enable = 0;
2089 		}
2090 		PRM_DEBUG(ppvm_size);
2091 		PRM_DEBUG(ppvm_base);
2092 	}
2093 
2094 	/*
2095 	 * Now create generic mapping segment.  This mapping
2096 	 * goes SEGMAPSIZE beyond SEGMAPBASE.  But if the total
2097 	 * virtual address is greater than the amount of free
2098 	 * memory that is available, then we trim back the
2099 	 * segment size to that amount
2100 	 */
2101 	va = (caddr_t)SEGMAPBASE;
2102 
2103 	/*
2104 	 * 1201049: segkmap base address must be MAXBSIZE aligned
2105 	 */
2106 	ASSERT(((uintptr_t)va & MAXBOFFSET) == 0);
2107 
2108 	/*
2109 	 * Set size of segmap to percentage of freemem at boot,
2110 	 * but stay within the allowable range
2111 	 * Note we take percentage  before converting from pages
2112 	 * to bytes to avoid an overflow on 32-bit kernels.
2113 	 */
2114 	i = mmu_ptob((freemem * segmap_percent) / 100);
2115 
2116 	if (i < MINMAPSIZE)
2117 		i = MINMAPSIZE;
2118 
2119 	if (i > MIN(SEGMAPSIZE, mmu_ptob(freemem)))
2120 		i = MIN(SEGMAPSIZE, mmu_ptob(freemem));
2121 
2122 	i &= MAXBMASK;	/* 1201049: segkmap size must be MAXBSIZE aligned */
2123 
2124 	rw_enter(&kas.a_lock, RW_WRITER);
2125 	if (seg_attach(&kas, va, i, segkmap) < 0)
2126 		cmn_err(CE_PANIC, "cannot attach segkmap");
2127 
2128 	a.prot = PROT_READ | PROT_WRITE;
2129 	a.shmsize = shm_alignment;
2130 	a.nfreelist = 0;	/* use segmap driver defaults */
2131 
2132 	if (segmap_create(segkmap, (caddr_t)&a) != 0)
2133 		panic("segmap_create segkmap");
2134 	rw_exit(&kas.a_lock);
2135 
2136 	segdev_init();
2137 }
2138 
2139 static void
2140 startup_end(void)
2141 {
2142 	if ((caddr_t)memlist > (caddr_t)memlist_end)
2143 		panic("memlist overflow 2");
2144 	memlist_free_block((caddr_t)memlist,
2145 	    ((caddr_t)memlist_end - (caddr_t)memlist));
2146 	memlist = NULL;
2147 
2148 	/* enable page_relocation since OBP is now done */
2149 	page_relocate_ready = 1;
2150 
2151 	/*
2152 	 * Perform tasks that get done after most of the VM
2153 	 * initialization has been done but before the clock
2154 	 * and other devices get started.
2155 	 */
2156 	kern_setup1();
2157 
2158 	/*
2159 	 * Perform CPC initialization for this CPU.
2160 	 */
2161 	kcpc_hw_init();
2162 
2163 	/*
2164 	 * Intialize the VM arenas for allocating physically
2165 	 * contiguus memory chunk for interrupt queues snd
2166 	 * allocate/register boot cpu's queues, if any and
2167 	 * allocate dump buffer for sun4v systems to store
2168 	 * extra crash information during crash dump
2169 	 */
2170 	contig_mem_init();
2171 	mach_descrip_init();
2172 
2173 	if (cpu_intrq_setup(CPU)) {
2174 		cmn_err(CE_PANIC, "cpu%d: setup failed", CPU->cpu_id);
2175 	}
2176 	cpu_intrq_register(CPU);
2177 	mach_htraptrace_setup(CPU->cpu_id);
2178 	mach_htraptrace_configure(CPU->cpu_id);
2179 	mach_dump_buffer_init();
2180 
2181 	/*
2182 	 * Initialize interrupt related stuff
2183 	 */
2184 	cpu_intr_alloc(CPU, NINTR_THREADS);
2185 
2186 	(void) splzs();			/* allow hi clock ints but not zs */
2187 
2188 	/*
2189 	 * Initialize errors.
2190 	 */
2191 	error_init();
2192 
2193 	/*
2194 	 * Note that we may have already used kernel bcopy before this
2195 	 * point - but if you really care about this, adb the use_hw_*
2196 	 * variables to 0 before rebooting.
2197 	 */
2198 	mach_hw_copy_limit();
2199 
2200 	/*
2201 	 * Install the "real" preemption guards before DDI services
2202 	 * are available.
2203 	 */
2204 	(void) prom_set_preprom(kern_preprom);
2205 	(void) prom_set_postprom(kern_postprom);
2206 	CPU->cpu_m.mutex_ready = 1;
2207 
2208 	/*
2209 	 * Initialize segnf (kernel support for non-faulting loads).
2210 	 */
2211 	segnf_init();
2212 
2213 	/*
2214 	 * Configure the root devinfo node.
2215 	 */
2216 	configure();		/* set up devices */
2217 	mach_cpu_halt_idle();
2218 }
2219 
2220 
2221 void
2222 post_startup(void)
2223 {
2224 #ifdef	PTL1_PANIC_DEBUG
2225 	extern void init_ptl1_thread(void);
2226 #endif	/* PTL1_PANIC_DEBUG */
2227 	extern void abort_sequence_init(void);
2228 
2229 	/*
2230 	 * Set the system wide, processor-specific flags to be passed
2231 	 * to userland via the aux vector for performance hints and
2232 	 * instruction set extensions.
2233 	 */
2234 	bind_hwcap();
2235 
2236 	/*
2237 	 * Startup memory scrubber (if any)
2238 	 */
2239 	mach_memscrub();
2240 
2241 	/*
2242 	 * Allocate soft interrupt to handle abort sequence.
2243 	 */
2244 	abort_sequence_init();
2245 
2246 	/*
2247 	 * Configure the rest of the system.
2248 	 * Perform forceloading tasks for /etc/system.
2249 	 */
2250 	(void) mod_sysctl(SYS_FORCELOAD, NULL);
2251 	/*
2252 	 * ON4.0: Force /proc module in until clock interrupt handle fixed
2253 	 * ON4.0: This must be fixed or restated in /etc/systems.
2254 	 */
2255 	(void) modload("fs", "procfs");
2256 
2257 	/* load machine class specific drivers */
2258 	load_mach_drivers();
2259 
2260 	/* load platform specific drivers */
2261 	if (&load_platform_drivers)
2262 		load_platform_drivers();
2263 
2264 	/* load vis simulation module, if we are running w/fpu off */
2265 	if (!fpu_exists) {
2266 		if (modload("misc", "vis") == -1)
2267 			halt("Can't load vis");
2268 	}
2269 
2270 	mach_fpras();
2271 
2272 	maxmem = freemem;
2273 
2274 	pg_init();
2275 
2276 #ifdef	PTL1_PANIC_DEBUG
2277 	init_ptl1_thread();
2278 #endif	/* PTL1_PANIC_DEBUG */
2279 }
2280 
2281 #ifdef	PTL1_PANIC_DEBUG
2282 int		ptl1_panic_test = 0;
2283 int		ptl1_panic_xc_one_test = 0;
2284 int		ptl1_panic_xc_all_test = 0;
2285 int		ptl1_panic_xt_one_test = 0;
2286 int		ptl1_panic_xt_all_test = 0;
2287 kthread_id_t	ptl1_thread_p = NULL;
2288 kcondvar_t	ptl1_cv;
2289 kmutex_t	ptl1_mutex;
2290 int		ptl1_recurse_count_threshold = 0x40;
2291 int		ptl1_recurse_trap_threshold = 0x3d;
2292 extern void	ptl1_recurse(int, int);
2293 extern void	ptl1_panic_xt(int, int);
2294 
2295 /*
2296  * Called once per second by timeout() to wake up
2297  * the ptl1_panic thread to see if it should cause
2298  * a trap to the ptl1_panic() code.
2299  */
2300 /* ARGSUSED */
2301 static void
2302 ptl1_wakeup(void *arg)
2303 {
2304 	mutex_enter(&ptl1_mutex);
2305 	cv_signal(&ptl1_cv);
2306 	mutex_exit(&ptl1_mutex);
2307 }
2308 
2309 /*
2310  * ptl1_panic cross call function:
2311  *     Needed because xc_one() and xc_some() can pass
2312  *	64 bit args but ptl1_recurse() expects ints.
2313  */
2314 static void
2315 ptl1_panic_xc(void)
2316 {
2317 	ptl1_recurse(ptl1_recurse_count_threshold,
2318 	    ptl1_recurse_trap_threshold);
2319 }
2320 
2321 /*
2322  * The ptl1 thread waits for a global flag to be set
2323  * and uses the recurse thresholds to set the stack depth
2324  * to cause a ptl1_panic() directly via a call to ptl1_recurse
2325  * or indirectly via the cross call and cross trap functions.
2326  *
2327  * This is useful testing stack overflows and normal
2328  * ptl1_panic() states with a know stack frame.
2329  *
2330  * ptl1_recurse() is an asm function in ptl1_panic.s that
2331  * sets the {In, Local, Out, and Global} registers to a
2332  * know state on the stack and just prior to causing a
2333  * test ptl1_panic trap.
2334  */
2335 static void
2336 ptl1_thread(void)
2337 {
2338 	mutex_enter(&ptl1_mutex);
2339 	while (ptl1_thread_p) {
2340 		cpuset_t	other_cpus;
2341 		int		cpu_id;
2342 		int		my_cpu_id;
2343 		int		target_cpu_id;
2344 		int		target_found;
2345 
2346 		if (ptl1_panic_test) {
2347 			ptl1_recurse(ptl1_recurse_count_threshold,
2348 			    ptl1_recurse_trap_threshold);
2349 		}
2350 
2351 		/*
2352 		 * Find potential targets for x-call and x-trap,
2353 		 * if any exist while preempt is disabled we
2354 		 * start a ptl1_panic if requested via a
2355 		 * globals.
2356 		 */
2357 		kpreempt_disable();
2358 		my_cpu_id = CPU->cpu_id;
2359 		other_cpus = cpu_ready_set;
2360 		CPUSET_DEL(other_cpus, CPU->cpu_id);
2361 		target_found = 0;
2362 		if (!CPUSET_ISNULL(other_cpus)) {
2363 			/*
2364 			 * Pick the first one
2365 			 */
2366 			for (cpu_id = 0; cpu_id < NCPU; cpu_id++) {
2367 				if (cpu_id == my_cpu_id)
2368 					continue;
2369 
2370 				if (CPU_XCALL_READY(cpu_id)) {
2371 					target_cpu_id = cpu_id;
2372 					target_found = 1;
2373 					break;
2374 				}
2375 			}
2376 			ASSERT(target_found);
2377 
2378 			if (ptl1_panic_xc_one_test) {
2379 				xc_one(target_cpu_id,
2380 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2381 			}
2382 			if (ptl1_panic_xc_all_test) {
2383 				xc_some(other_cpus,
2384 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2385 			}
2386 			if (ptl1_panic_xt_one_test) {
2387 				xt_one(target_cpu_id,
2388 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2389 			}
2390 			if (ptl1_panic_xt_all_test) {
2391 				xt_some(other_cpus,
2392 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2393 			}
2394 		}
2395 		kpreempt_enable();
2396 		(void) timeout(ptl1_wakeup, NULL, hz);
2397 		(void) cv_wait(&ptl1_cv, &ptl1_mutex);
2398 	}
2399 	mutex_exit(&ptl1_mutex);
2400 }
2401 
2402 /*
2403  * Called during early startup to create the ptl1_thread
2404  */
2405 void
2406 init_ptl1_thread(void)
2407 {
2408 	ptl1_thread_p = thread_create(NULL, 0, ptl1_thread, NULL, 0,
2409 	    &p0, TS_RUN, 0);
2410 }
2411 #endif	/* PTL1_PANIC_DEBUG */
2412 
2413 
2414 static void
2415 memlist_new(uint64_t start, uint64_t len, struct memlist **memlistp)
2416 {
2417 	struct memlist *new;
2418 
2419 	new = *memlistp;
2420 	new->address = start;
2421 	new->size = len;
2422 	*memlistp = new + 1;
2423 }
2424 
2425 /*
2426  * Add to a memory list.
2427  * start = start of new memory segment
2428  * len = length of new memory segment in bytes
2429  * memlistp = pointer to array of available memory segment structures
2430  * curmemlistp = memory list to which to add segment.
2431  */
2432 static void
2433 memlist_add(uint64_t start, uint64_t len, struct memlist **memlistp,
2434 	struct memlist **curmemlistp)
2435 {
2436 	struct memlist *new = *memlistp;
2437 
2438 	memlist_new(start, len, memlistp);
2439 	memlist_insert(new, curmemlistp);
2440 }
2441 
2442 static int
2443 ndata_alloc_memseg(struct memlist *ndata, size_t avail)
2444 {
2445 	int nseg;
2446 	size_t memseg_sz;
2447 	struct memseg *msp;
2448 
2449 	/*
2450 	 * The memseg list is for the chunks of physical memory that
2451 	 * will be managed by the vm system.  The number calculated is
2452 	 * a guess as boot may fragment it more when memory allocations
2453 	 * are made before kphysm_init().
2454 	 */
2455 	memseg_sz = (avail + 10) * sizeof (struct memseg);
2456 	memseg_sz = roundup(memseg_sz, PAGESIZE);
2457 	nseg = memseg_sz / sizeof (struct memseg);
2458 	msp = ndata_alloc(ndata, memseg_sz, ecache_alignsize);
2459 	if (msp == NULL)
2460 		return (1);
2461 	PRM_DEBUG(memseg_free);
2462 
2463 	while (nseg--) {
2464 		msp->next = memseg_free;
2465 		memseg_free = msp;
2466 		msp++;
2467 	}
2468 	return (0);
2469 }
2470 
2471 /*
2472  * In the case of architectures that support dynamic addition of
2473  * memory at run-time there are two cases where memsegs need to
2474  * be initialized and added to the memseg list.
2475  * 1) memsegs that are constructed at startup.
2476  * 2) memsegs that are constructed at run-time on
2477  *    hot-plug capable architectures.
2478  * This code was originally part of the function kphysm_init().
2479  */
2480 
2481 static void
2482 memseg_list_add(struct memseg *memsegp)
2483 {
2484 	struct memseg **prev_memsegp;
2485 	pgcnt_t num;
2486 
2487 	/* insert in memseg list, decreasing number of pages order */
2488 
2489 	num = MSEG_NPAGES(memsegp);
2490 
2491 	for (prev_memsegp = &memsegs; *prev_memsegp;
2492 	    prev_memsegp = &((*prev_memsegp)->next)) {
2493 		if (num > MSEG_NPAGES(*prev_memsegp))
2494 			break;
2495 	}
2496 
2497 	memsegp->next = *prev_memsegp;
2498 	*prev_memsegp = memsegp;
2499 
2500 	if (kpm_enable) {
2501 		memsegp->nextpa = (memsegp->next) ?
2502 		    va_to_pa(memsegp->next) : MSEG_NULLPTR_PA;
2503 
2504 		if (prev_memsegp != &memsegs) {
2505 			struct memseg *msp;
2506 			msp = (struct memseg *)((caddr_t)prev_memsegp -
2507 			    offsetof(struct memseg, next));
2508 			msp->nextpa = va_to_pa(memsegp);
2509 		} else {
2510 			memsegspa = va_to_pa(memsegs);
2511 		}
2512 	}
2513 }
2514 
2515 /*
2516  * PSM add_physmem_cb(). US-II and newer processors have some
2517  * flavor of the prefetch capability implemented. We exploit
2518  * this capability for optimum performance.
2519  */
2520 #define	PREFETCH_BYTES	64
2521 
2522 void
2523 add_physmem_cb(page_t *pp, pfn_t pnum)
2524 {
2525 	extern void	 prefetch_page_w(void *);
2526 
2527 	pp->p_pagenum = pnum;
2528 
2529 	/*
2530 	 * Prefetch one more page_t into E$. To prevent future
2531 	 * mishaps with the sizeof(page_t) changing on us, we
2532 	 * catch this on debug kernels if we can't bring in the
2533 	 * entire hpage with 2 PREFETCH_BYTES reads. See
2534 	 * also, sun4u/cpu/cpu_module.c
2535 	 */
2536 	/*LINTED*/
2537 	ASSERT(sizeof (page_t) <= 2*PREFETCH_BYTES);
2538 	prefetch_page_w((char *)pp);
2539 }
2540 
2541 /*
2542  * Find memseg with given pfn
2543  */
2544 static struct memseg *
2545 memseg_find(pfn_t base, pfn_t *next)
2546 {
2547 	struct memseg *seg;
2548 
2549 	if (next != NULL)
2550 		*next = LONG_MAX;
2551 	for (seg = memsegs; seg != NULL; seg = seg->next) {
2552 		if (base >= seg->pages_base && base < seg->pages_end)
2553 			return (seg);
2554 		if (next != NULL && seg->pages_base > base &&
2555 		    seg->pages_base < *next)
2556 			*next = seg->pages_base;
2557 	}
2558 	return (NULL);
2559 }
2560 
2561 extern struct vnode prom_ppages;
2562 
2563 /*
2564  * Put page allocated by OBP on prom_ppages
2565  */
2566 static void
2567 kphysm_erase(uint64_t addr, uint64_t len)
2568 {
2569 	struct page *pp;
2570 	struct memseg *seg;
2571 	pfn_t base = btop(addr), next;
2572 	pgcnt_t num = btop(len);
2573 
2574 	while (num != 0) {
2575 		pgcnt_t off, left;
2576 
2577 		seg = memseg_find(base, &next);
2578 		if (seg == NULL) {
2579 			if (next == LONG_MAX)
2580 				break;
2581 			left = MIN(next - base, num);
2582 			base += left, num -= left;
2583 			continue;
2584 		}
2585 		off = base - seg->pages_base;
2586 		pp = seg->pages + off;
2587 		left = num - MIN(num, (seg->pages_end - seg->pages_base) - off);
2588 		while (num != left) {
2589 			/*
2590 			 * init it, lock it, and hashin on prom_pages vp.
2591 			 *
2592 			 * Mark it as NONRELOC to let DR know the page
2593 			 * is locked long term, otherwise DR hangs when
2594 			 * trying to remove those pages.
2595 			 *
2596 			 * XXX	vnode offsets on the prom_ppages vnode
2597 			 *	are page numbers (gack) for >32 bit
2598 			 *	physical memory machines.
2599 			 */
2600 			PP_SETNORELOC(pp);
2601 			add_physmem_cb(pp, base);
2602 			if (page_trylock(pp, SE_EXCL) == 0)
2603 				cmn_err(CE_PANIC, "prom page locked");
2604 			(void) page_hashin(pp, &prom_ppages,
2605 			    (offset_t)base, NULL);
2606 			(void) page_pp_lock(pp, 0, 1);
2607 			pp++, base++, num--;
2608 		}
2609 	}
2610 }
2611 
2612 static page_t *ppnext;
2613 static pgcnt_t ppleft;
2614 
2615 static void *kpm_ppnext;
2616 static pgcnt_t kpm_ppleft;
2617 
2618 /*
2619  * Create a memseg
2620  */
2621 static void
2622 kphysm_memseg(uint64_t addr, uint64_t len)
2623 {
2624 	pfn_t base = btop(addr);
2625 	pgcnt_t num = btop(len);
2626 	struct memseg *seg;
2627 
2628 	seg = memseg_free;
2629 	memseg_free = seg->next;
2630 	ASSERT(seg != NULL);
2631 
2632 	seg->pages = ppnext;
2633 	seg->epages = ppnext + num;
2634 	seg->pages_base = base;
2635 	seg->pages_end = base + num;
2636 	ppnext += num;
2637 	ppleft -= num;
2638 
2639 	if (kpm_enable) {
2640 		pgcnt_t kpnum = ptokpmpr(num);
2641 
2642 		if (kpnum > kpm_ppleft)
2643 			panic("kphysm_memseg: kpm_pp overflow");
2644 		seg->pagespa = va_to_pa(seg->pages);
2645 		seg->epagespa = va_to_pa(seg->epages);
2646 		seg->kpm_pbase = kpmptop(ptokpmp(base));
2647 		seg->kpm_nkpmpgs = kpnum;
2648 		/*
2649 		 * In the kpm_smallpage case, the kpm array
2650 		 * is 1-1 wrt the page array
2651 		 */
2652 		if (kpm_smallpages) {
2653 			kpm_spage_t *kpm_pp = kpm_ppnext;
2654 
2655 			kpm_ppnext = kpm_pp + kpnum;
2656 			seg->kpm_spages = kpm_pp;
2657 			seg->kpm_pagespa = va_to_pa(seg->kpm_spages);
2658 		} else {
2659 			kpm_page_t *kpm_pp = kpm_ppnext;
2660 
2661 			kpm_ppnext = kpm_pp + kpnum;
2662 			seg->kpm_pages = kpm_pp;
2663 			seg->kpm_pagespa = va_to_pa(seg->kpm_pages);
2664 			/* ASSERT no kpm overlaps */
2665 			ASSERT(
2666 			    memseg_find(base - pmodkpmp(base), NULL) == NULL);
2667 			ASSERT(memseg_find(
2668 			    roundup(base + num, kpmpnpgs) - 1, NULL) == NULL);
2669 		}
2670 		kpm_ppleft -= kpnum;
2671 	}
2672 
2673 	memseg_list_add(seg);
2674 }
2675 
2676 /*
2677  * Add range to free list
2678  */
2679 void
2680 kphysm_add(uint64_t addr, uint64_t len, int reclaim)
2681 {
2682 	struct page *pp;
2683 	struct memseg *seg;
2684 	pfn_t base = btop(addr);
2685 	pgcnt_t num = btop(len);
2686 
2687 	seg = memseg_find(base, NULL);
2688 	ASSERT(seg != NULL);
2689 	pp = seg->pages + (base - seg->pages_base);
2690 
2691 	if (reclaim) {
2692 		struct page *rpp = pp;
2693 		struct page *lpp = pp + num;
2694 
2695 		/*
2696 		 * page should be locked on prom_ppages
2697 		 * unhash and unlock it
2698 		 */
2699 		while (rpp < lpp) {
2700 			ASSERT(PAGE_EXCL(rpp) && rpp->p_vnode == &prom_ppages);
2701 			ASSERT(PP_ISNORELOC(rpp));
2702 			PP_CLRNORELOC(rpp);
2703 			page_pp_unlock(rpp, 0, 1);
2704 			page_hashout(rpp, NULL);
2705 			page_unlock(rpp);
2706 			rpp++;
2707 		}
2708 	}
2709 
2710 	/*
2711 	 * add_physmem() initializes the PSM part of the page
2712 	 * struct by calling the PSM back with add_physmem_cb().
2713 	 * In addition it coalesces pages into larger pages as
2714 	 * it initializes them.
2715 	 */
2716 	add_physmem(pp, num, base);
2717 }
2718 
2719 /*
2720  * kphysm_init() tackles the problem of initializing physical memory.
2721  */
2722 static void
2723 kphysm_init(void)
2724 {
2725 	struct memlist *pmem;
2726 
2727 	ASSERT(page_hash != NULL && page_hashsz != 0);
2728 
2729 	ppnext = pp_base;
2730 	ppleft = npages;
2731 	kpm_ppnext = kpm_pp_base;
2732 	kpm_ppleft = kpm_npages;
2733 
2734 	/*
2735 	 * installed pages not on nopp_memlist go in memseg list
2736 	 */
2737 	diff_memlists(phys_install, nopp_list, kphysm_memseg);
2738 
2739 	/*
2740 	 * Free the avail list
2741 	 */
2742 	for (pmem = phys_avail; pmem != NULL; pmem = pmem->next)
2743 		kphysm_add(pmem->address, pmem->size, 0);
2744 
2745 	/*
2746 	 * Erase pages that aren't available
2747 	 */
2748 	diff_memlists(phys_install, phys_avail, kphysm_erase);
2749 
2750 	build_pfn_hash();
2751 }
2752 
2753 /*
2754  * Kernel VM initialization.
2755  * Assumptions about kernel address space ordering:
2756  *	(1) gap (user space)
2757  *	(2) kernel text
2758  *	(3) kernel data/bss
2759  *	(4) gap
2760  *	(5) kernel data structures
2761  *	(6) gap
2762  *	(7) debugger (optional)
2763  *	(8) monitor
2764  *	(9) gap (possibly null)
2765  *	(10) dvma
2766  *	(11) devices
2767  */
2768 static void
2769 kvm_init(void)
2770 {
2771 	/*
2772 	 * Put the kernel segments in kernel address space.
2773 	 */
2774 	rw_enter(&kas.a_lock, RW_WRITER);
2775 	as_avlinit(&kas);
2776 
2777 	(void) seg_attach(&kas, (caddr_t)KERNELBASE,
2778 	    (size_t)(e_moddata - KERNELBASE), &ktextseg);
2779 	(void) segkmem_create(&ktextseg);
2780 
2781 	(void) seg_attach(&kas, (caddr_t)(KERNELBASE + MMU_PAGESIZE4M),
2782 	    (size_t)(MMU_PAGESIZE4M), &ktexthole);
2783 	(void) segkmem_create(&ktexthole);
2784 
2785 	(void) seg_attach(&kas, (caddr_t)valloc_base,
2786 	    (size_t)(econtig32 - valloc_base), &kvalloc);
2787 	(void) segkmem_create(&kvalloc);
2788 
2789 	if (kmem64_base) {
2790 		(void) seg_attach(&kas, (caddr_t)kmem64_base,
2791 		    (size_t)(kmem64_end - kmem64_base), &kmem64);
2792 		(void) segkmem_create(&kmem64);
2793 	}
2794 
2795 	/*
2796 	 * We're about to map out /boot.  This is the beginning of the
2797 	 * system resource management transition. We can no longer
2798 	 * call into /boot for I/O or memory allocations.
2799 	 */
2800 	(void) seg_attach(&kas, kernelheap, ekernelheap - kernelheap, &kvseg);
2801 	(void) segkmem_create(&kvseg);
2802 	hblk_alloc_dynamic = 1;
2803 
2804 	/*
2805 	 * we need to preallocate pages for DR operations before enabling large
2806 	 * page kernel heap because of memseg_remap_init() hat_unload() hack.
2807 	 */
2808 	memseg_remap_init();
2809 
2810 	/* at this point we are ready to use large page heap */
2811 	segkmem_heap_lp_init();
2812 
2813 	(void) seg_attach(&kas, (caddr_t)SYSBASE32, SYSLIMIT32 - SYSBASE32,
2814 	    &kvseg32);
2815 	(void) segkmem_create(&kvseg32);
2816 
2817 	/*
2818 	 * Create a segment for the debugger.
2819 	 */
2820 	(void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2821 	(void) segkmem_create(&kdebugseg);
2822 
2823 	rw_exit(&kas.a_lock);
2824 }
2825 
2826 char obp_tte_str[] =
2827 	"h# %x constant MMU_PAGESHIFT "
2828 	"h# %x constant TTE8K "
2829 	"h# %x constant SFHME_SIZE "
2830 	"h# %x constant SFHME_TTE "
2831 	"h# %x constant HMEBLK_TAG "
2832 	"h# %x constant HMEBLK_NEXT "
2833 	"h# %x constant HMEBLK_MISC "
2834 	"h# %x constant HMEBLK_HME1 "
2835 	"h# %x constant NHMENTS "
2836 	"h# %x constant HBLK_SZMASK "
2837 	"h# %x constant HBLK_RANGE_SHIFT "
2838 	"h# %x constant HMEBP_HBLK "
2839 	"h# %x constant HMEBLK_ENDPA "
2840 	"h# %x constant HMEBUCKET_SIZE "
2841 	"h# %x constant HTAG_SFMMUPSZ "
2842 	"h# %x constant HTAG_BSPAGE_SHIFT "
2843 	"h# %x constant HTAG_REHASH_SHIFT "
2844 	"h# %x constant SFMMU_INVALID_SHMERID "
2845 	"h# %x constant mmu_hashcnt "
2846 	"h# %p constant uhme_hash "
2847 	"h# %p constant khme_hash "
2848 	"h# %x constant UHMEHASH_SZ "
2849 	"h# %x constant KHMEHASH_SZ "
2850 	"h# %p constant KCONTEXT "
2851 	"h# %p constant KHATID "
2852 	"h# %x constant ASI_MEM "
2853 	"h# %x constant SOFTEXEC "
2854 	"h# %x constant EXECPRM "
2855 
2856 	": PHYS-X@ ( phys -- data ) "
2857 	"   ASI_MEM spacex@ "
2858 	"; "
2859 
2860 	": PHYS-W@ ( phys -- data ) "
2861 	"   ASI_MEM spacew@ "
2862 	"; "
2863 
2864 	": PHYS-L@ ( phys -- data ) "
2865 	"   ASI_MEM spaceL@ "
2866 	"; "
2867 
2868 	": TTE_PAGE_SHIFT ( ttesz -- hmeshift ) "
2869 	"   3 * MMU_PAGESHIFT + "
2870 	"; "
2871 
2872 	": TTE_IS_VALID ( ttep -- flag ) "
2873 	"   PHYS-X@ 0< "
2874 	"; "
2875 
2876 	": HME_HASH_SHIFT ( ttesz -- hmeshift ) "
2877 	"   dup TTE8K =  if "
2878 	"      drop HBLK_RANGE_SHIFT "
2879 	"   else "
2880 	"      TTE_PAGE_SHIFT "
2881 	"   then "
2882 	"; "
2883 
2884 	": HME_HASH_BSPAGE ( addr hmeshift -- bspage ) "
2885 	"   tuck >> swap MMU_PAGESHIFT - << "
2886 	"; "
2887 
2888 	": HME_HASH_FUNCTION ( sfmmup addr hmeshift -- hmebp ) "
2889 	"   >> over xor swap                    ( hash sfmmup ) "
2890 	"   KHATID <>  if                       ( hash ) "
2891 	"      UHMEHASH_SZ and                  ( bucket ) "
2892 	"      HMEBUCKET_SIZE * uhme_hash +     ( hmebp ) "
2893 	"   else                                ( hash ) "
2894 	"      KHMEHASH_SZ and                  ( bucket ) "
2895 	"      HMEBUCKET_SIZE * khme_hash +     ( hmebp ) "
2896 	"   then                                ( hmebp ) "
2897 	"; "
2898 
2899 	": HME_HASH_TABLE_SEARCH "
2900 	"       ( sfmmup hmebp hblktag --  sfmmup null | sfmmup hmeblkp ) "
2901 	"   >r hmebp_hblk + phys-x@ begin ( sfmmup hmeblkp ) ( r: hblktag ) "
2902 	"      dup HMEBLK_ENDPA <> if     ( sfmmup hmeblkp ) ( r: hblktag ) "
2903 	"         dup hmeblk_tag + phys-x@ r@ = if ( sfmmup hmeblkp )	  "
2904 	"	     dup hmeblk_tag + 8 + phys-x@ 2 pick = if		  "
2905 	"		  true 	( sfmmup hmeblkp true ) ( r: hblktag )	  "
2906 	"	     else						  "
2907 	"	     	  hmeblk_next + phys-x@ false 			  "
2908 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2909 	"	     then  						  "
2910 	"	  else							  "
2911 	"	     hmeblk_next + phys-x@ false 			  "
2912 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2913 	"	  then 							  "
2914 	"      else							  "
2915 	"         drop 0 true 						  "
2916 	"      then  							  "
2917 	"   until r> drop 						  "
2918 	"; "
2919 
2920 	": HME_HASH_TAG ( sfmmup rehash addr -- hblktag ) "
2921 	"   over HME_HASH_SHIFT HME_HASH_BSPAGE  ( sfmmup rehash bspage ) "
2922 	"   HTAG_BSPAGE_SHIFT <<		 ( sfmmup rehash htag-bspage )"
2923 	"   swap HTAG_REHASH_SHIFT << or	 ( sfmmup htag-bspage-rehash )"
2924 	"   SFMMU_INVALID_SHMERID or nip	 ( hblktag ) "
2925 	"; "
2926 
2927 	": HBLK_TO_TTEP ( hmeblkp addr -- ttep ) "
2928 	"   over HMEBLK_MISC + PHYS-L@ HBLK_SZMASK and  ( hmeblkp addr ttesz ) "
2929 	"   TTE8K =  if                            ( hmeblkp addr ) "
2930 	"      MMU_PAGESHIFT >> NHMENTS 1- and     ( hmeblkp hme-index ) "
2931 	"   else                                   ( hmeblkp addr ) "
2932 	"      drop 0                              ( hmeblkp 0 ) "
2933 	"   then                                   ( hmeblkp hme-index ) "
2934 	"   SFHME_SIZE * + HMEBLK_HME1 +           ( hmep ) "
2935 	"   SFHME_TTE +                            ( ttep ) "
2936 	"; "
2937 
2938 	": unix-tte ( addr cnum -- false | tte-data true ) "
2939 	"    KCONTEXT = if                   ( addr ) "
2940 	"	KHATID                       ( addr khatid ) "
2941 	"    else                            ( addr ) "
2942 	"       drop false exit              ( false ) "
2943 	"    then "
2944 	"      ( addr khatid ) "
2945 	"      mmu_hashcnt 1+ 1  do           ( addr sfmmup ) "
2946 	"         2dup swap i HME_HASH_SHIFT  "
2947 					"( addr sfmmup sfmmup addr hmeshift ) "
2948 	"         HME_HASH_FUNCTION           ( addr sfmmup hmebp ) "
2949 	"         over i 4 pick               "
2950 				"( addr sfmmup hmebp sfmmup rehash addr ) "
2951 	"         HME_HASH_TAG                ( addr sfmmup hmebp hblktag ) "
2952 	"         HME_HASH_TABLE_SEARCH       "
2953 					"( addr sfmmup { null | hmeblkp } ) "
2954 	"         ?dup  if                    ( addr sfmmup hmeblkp ) "
2955 	"            nip swap HBLK_TO_TTEP    ( ttep ) "
2956 	"            dup TTE_IS_VALID  if     ( valid-ttep ) "
2957 	"               PHYS-X@               ( tte-data ) "
2958 	"               dup SOFTEXEC and 0> if  ( tte-data ) "
2959 	"                 SOFTEXEC - EXECPRM or ( tte-data ) "
2960 	"               then                    ( tte-data ) "
2961 	"               true                  ( tte-data true ) "
2962 	"            else                     ( invalid-tte ) "
2963 	"               drop false            ( false ) "
2964 	"            then                     ( false | tte-data true ) "
2965 	"            unloop exit              ( false | tte-data true ) "
2966 	"         then                        ( addr sfmmup ) "
2967 	"      loop                           ( addr sfmmup ) "
2968 	"      2drop false                    ( false ) "
2969 	"; "
2970 ;
2971 
2972 void
2973 create_va_to_tte(void)
2974 {
2975 	char *bp;
2976 	extern int khmehash_num, uhmehash_num;
2977 	extern struct hmehash_bucket *khme_hash, *uhme_hash;
2978 
2979 #define	OFFSET(type, field)	((uintptr_t)(&((type *)0)->field))
2980 
2981 	bp = (char *)kobj_zalloc(MMU_PAGESIZE, KM_SLEEP);
2982 
2983 	/*
2984 	 * Teach obp how to parse our sw ttes.
2985 	 */
2986 	(void) sprintf(bp, obp_tte_str,
2987 	    MMU_PAGESHIFT,
2988 	    TTE8K,
2989 	    sizeof (struct sf_hment),
2990 	    OFFSET(struct sf_hment, hme_tte),
2991 	    OFFSET(struct hme_blk, hblk_tag),
2992 	    OFFSET(struct hme_blk, hblk_nextpa),
2993 	    OFFSET(struct hme_blk, hblk_misc),
2994 	    OFFSET(struct hme_blk, hblk_hme),
2995 	    NHMENTS,
2996 	    HBLK_SZMASK,
2997 	    HBLK_RANGE_SHIFT,
2998 	    OFFSET(struct hmehash_bucket, hmeh_nextpa),
2999 	    HMEBLK_ENDPA,
3000 	    sizeof (struct hmehash_bucket),
3001 	    HTAG_SFMMUPSZ,
3002 	    HTAG_BSPAGE_SHIFT,
3003 	    HTAG_REHASH_SHIFT,
3004 	    SFMMU_INVALID_SHMERID,
3005 	    mmu_hashcnt,
3006 	    (caddr_t)va_to_pa((caddr_t)uhme_hash),
3007 	    (caddr_t)va_to_pa((caddr_t)khme_hash),
3008 	    UHMEHASH_SZ,
3009 	    KHMEHASH_SZ,
3010 	    KCONTEXT,
3011 	    KHATID,
3012 	    ASI_MEM,
3013 	    icache_is_coherent ? 0 : TTE_SOFTEXEC_INT,
3014 	    TTE_EXECPRM_INT);
3015 	prom_interpret(bp, 0, 0, 0, 0, 0);
3016 
3017 	kobj_free(bp, MMU_PAGESIZE);
3018 }
3019 
3020 void
3021 install_va_to_tte(void)
3022 {
3023 	/*
3024 	 * advise prom that he can use unix-tte
3025 	 */
3026 	prom_interpret("' unix-tte is va>tte-data", 0, 0, 0, 0, 0);
3027 }
3028 
3029 /*
3030  * Here we add "device-type=console" for /os-io node, for currently
3031  * our kernel console output only supports displaying text and
3032  * performing cursor-positioning operations (through kernel framebuffer
3033  * driver) and it doesn't support other functionalities required for a
3034  * standard "display" device as specified in 1275 spec. The main missing
3035  * interface defined by the 1275 spec is "draw-logo".
3036  * also see the comments above prom_stdout_is_framebuffer().
3037  */
3038 static char *create_node =
3039 	"\" /\" find-device "
3040 	"new-device "
3041 	"\" os-io\" device-name "
3042 	"\" "OBP_DISPLAY_CONSOLE"\" device-type "
3043 	": cb-r/w  ( adr,len method$ -- #read/#written ) "
3044 	"   2>r swap 2 2r> ['] $callback  catch  if "
3045 	"      2drop 3drop 0 "
3046 	"   then "
3047 	"; "
3048 	": read ( adr,len -- #read ) "
3049 	"       \" read\" ['] cb-r/w catch  if  2drop 2drop -2 exit then "
3050 	"       ( retN ... ret1 N ) "
3051 	"       ?dup  if "
3052 	"               swap >r 1-  0  ?do  drop  loop  r> "
3053 	"       else "
3054 	"               -2 "
3055 	"       then "
3056 	";    "
3057 	": write ( adr,len -- #written ) "
3058 	"       \" write\" ['] cb-r/w catch  if  2drop 2drop 0 exit  then "
3059 	"       ( retN ... ret1 N ) "
3060 	"       ?dup  if "
3061 	"               swap >r 1-  0  ?do  drop  loop  r> "
3062 	"        else "
3063 	"               0 "
3064 	"       then "
3065 	"; "
3066 	": poll-tty ( -- ) ; "
3067 	": install-abort  ( -- )  ['] poll-tty d# 10 alarm ; "
3068 	": remove-abort ( -- )  ['] poll-tty 0 alarm ; "
3069 	": cb-give/take ( $method -- ) "
3070 	"       0 -rot ['] $callback catch  ?dup  if "
3071 	"               >r 2drop 2drop r> throw "
3072 	"       else "
3073 	"               0  ?do  drop  loop "
3074 	"       then "
3075 	"; "
3076 	": give ( -- )  \" exit-input\" cb-give/take ; "
3077 	": take ( -- )  \" enter-input\" cb-give/take ; "
3078 	": open ( -- ok? )  true ; "
3079 	": close ( -- ) ; "
3080 	"finish-device "
3081 	"device-end ";
3082 
3083 /*
3084  * Create the OBP input/output node (FCode serial driver).
3085  * It is needed for both USB console keyboard and for
3086  * the kernel terminal emulator.  It is too early to check for a
3087  * kernel console compatible framebuffer now, so we create this
3088  * so that we're ready if we need to enable kernel terminal emulation.
3089  *
3090  * When the USB software takes over the input device at the time
3091  * consconfig runs, OBP's stdin is redirected to this node.
3092  * Whenever the FORTH user interface is used after this switch,
3093  * the node will call back into the kernel for console input.
3094  * If a serial device such as ttya or a UART with a Type 5 keyboard
3095  * attached is used, OBP takes over the serial device when the system
3096  * goes to the debugger after the system is booted.  This sharing
3097  * of the relatively simple serial device is difficult but possible.
3098  * Sharing the USB host controller is impossible due its complexity.
3099  *
3100  * Similarly to USB keyboard input redirection, after consconfig_dacf
3101  * configures a kernel console framebuffer as the standard output
3102  * device, OBP's stdout is switched to to vector through the
3103  * /os-io node into the kernel terminal emulator.
3104  */
3105 static void
3106 startup_create_io_node(void)
3107 {
3108 	prom_interpret(create_node, 0, 0, 0, 0, 0);
3109 }
3110 
3111 
3112 static void
3113 do_prom_version_check(void)
3114 {
3115 	int i;
3116 	pnode_t node;
3117 	char buf[64];
3118 	static char drev[] = "Down-rev firmware detected%s\n"
3119 	    "\tPlease upgrade to the following minimum version:\n"
3120 	    "\t\t%s\n";
3121 
3122 	i = prom_version_check(buf, sizeof (buf), &node);
3123 
3124 	if (i == PROM_VER64_OK)
3125 		return;
3126 
3127 	if (i == PROM_VER64_UPGRADE) {
3128 		cmn_err(CE_WARN, drev, "", buf);
3129 
3130 #ifdef	DEBUG
3131 		prom_enter_mon();	/* Type 'go' to continue */
3132 		cmn_err(CE_WARN, "Booting with down-rev firmware\n");
3133 		return;
3134 #else
3135 		halt(0);
3136 #endif
3137 	}
3138 
3139 	/*
3140 	 * The other possibility is that this is a server running
3141 	 * good firmware, but down-rev firmware was detected on at
3142 	 * least one other cpu board. We just complain if we see
3143 	 * that.
3144 	 */
3145 	cmn_err(CE_WARN, drev, " on one or more CPU boards", buf);
3146 }
3147 
3148 
3149 /*
3150  * Must be defined in platform dependent code.
3151  */
3152 extern caddr_t modtext;
3153 extern size_t modtext_sz;
3154 extern caddr_t moddata;
3155 
3156 #define	HEAPTEXT_ARENA(addr)	\
3157 	((uintptr_t)(addr) < KERNELBASE + 2 * MMU_PAGESIZE4M ? 0 : \
3158 	(((uintptr_t)(addr) - HEAPTEXT_BASE) / \
3159 	(HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) + 1))
3160 
3161 #define	HEAPTEXT_OVERSIZED(addr)	\
3162 	((uintptr_t)(addr) >= HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE)
3163 
3164 #define	HEAPTEXT_IN_NUCLEUSDATA(addr) \
3165 	(((uintptr_t)(addr) >= KERNELBASE + 2 * MMU_PAGESIZE4M) && \
3166 	((uintptr_t)(addr) < KERNELBASE + 3 * MMU_PAGESIZE4M))
3167 
3168 vmem_t *texthole_source[HEAPTEXT_NARENAS];
3169 vmem_t *texthole_arena[HEAPTEXT_NARENAS];
3170 kmutex_t texthole_lock;
3171 
3172 char kern_bootargs[OBP_MAXPATHLEN];
3173 char kern_bootfile[OBP_MAXPATHLEN];
3174 
3175 void
3176 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3177 {
3178 	uintptr_t addr, limit;
3179 
3180 	addr = HEAPTEXT_BASE;
3181 	limit = addr + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE;
3182 
3183 	/*
3184 	 * Before we initialize the text_arena, we want to punch holes in the
3185 	 * underlying heaptext_arena.  This guarantees that for any text
3186 	 * address we can find a text hole less than HEAPTEXT_MAPPED away.
3187 	 */
3188 	for (; addr + HEAPTEXT_UNMAPPED <= limit;
3189 	    addr += HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) {
3190 		(void) vmem_xalloc(heaptext_arena, HEAPTEXT_UNMAPPED, PAGESIZE,
3191 		    0, 0, (void *)addr, (void *)(addr + HEAPTEXT_UNMAPPED),
3192 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3193 	}
3194 
3195 	/*
3196 	 * Allocate one page at the oversize to break up the text region
3197 	 * from the oversized region.
3198 	 */
3199 	(void) vmem_xalloc(heaptext_arena, PAGESIZE, PAGESIZE, 0, 0,
3200 	    (void *)limit, (void *)(limit + PAGESIZE),
3201 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3202 
3203 	*text_arena = vmem_create("module_text", modtext_sz ? modtext : NULL,
3204 	    modtext_sz, sizeof (uintptr_t), segkmem_alloc, segkmem_free,
3205 	    heaptext_arena, 0, VM_SLEEP);
3206 	*data_arena = vmem_create("module_data", moddata, MODDATA, 1,
3207 	    segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3208 }
3209 
3210 caddr_t
3211 kobj_text_alloc(vmem_t *arena, size_t size)
3212 {
3213 	caddr_t rval, better;
3214 
3215 	/*
3216 	 * First, try a sleeping allocation.
3217 	 */
3218 	rval = vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT);
3219 
3220 	if (size >= HEAPTEXT_MAPPED || !HEAPTEXT_OVERSIZED(rval))
3221 		return (rval);
3222 
3223 	/*
3224 	 * We didn't get the area that we wanted.  We're going to try to do an
3225 	 * allocation with explicit constraints.
3226 	 */
3227 	better = vmem_xalloc(arena, size, sizeof (uintptr_t), 0, 0, NULL,
3228 	    (void *)(HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE),
3229 	    VM_NOSLEEP | VM_BESTFIT);
3230 
3231 	if (better != NULL) {
3232 		/*
3233 		 * That worked.  Free our first attempt and return.
3234 		 */
3235 		vmem_free(arena, rval, size);
3236 		return (better);
3237 	}
3238 
3239 	/*
3240 	 * That didn't work; we'll have to return our first attempt.
3241 	 */
3242 	return (rval);
3243 }
3244 
3245 caddr_t
3246 kobj_texthole_alloc(caddr_t addr, size_t size)
3247 {
3248 	int arena = HEAPTEXT_ARENA(addr);
3249 	char c[30];
3250 	uintptr_t base;
3251 
3252 	if (HEAPTEXT_OVERSIZED(addr) || HEAPTEXT_IN_NUCLEUSDATA(addr)) {
3253 		/*
3254 		 * If this is an oversized allocation or it is allocated in
3255 		 * the nucleus data page, there is no text hole available for
3256 		 * it; return NULL.
3257 		 */
3258 		return (NULL);
3259 	}
3260 
3261 	mutex_enter(&texthole_lock);
3262 
3263 	if (texthole_arena[arena] == NULL) {
3264 		ASSERT(texthole_source[arena] == NULL);
3265 
3266 		if (arena == 0) {
3267 			texthole_source[0] = vmem_create("module_text_holesrc",
3268 			    (void *)(KERNELBASE + MMU_PAGESIZE4M),
3269 			    MMU_PAGESIZE4M, PAGESIZE, NULL, NULL, NULL,
3270 			    0, VM_SLEEP);
3271 		} else {
3272 			base = HEAPTEXT_BASE +
3273 			    (arena - 1) * (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED);
3274 
3275 			(void) snprintf(c, sizeof (c),
3276 			    "heaptext_holesrc_%d", arena);
3277 
3278 			texthole_source[arena] = vmem_create(c, (void *)base,
3279 			    HEAPTEXT_UNMAPPED, PAGESIZE, NULL, NULL, NULL,
3280 			    0, VM_SLEEP);
3281 		}
3282 
3283 		(void) snprintf(c, sizeof (c), "heaptext_hole_%d", arena);
3284 
3285 		texthole_arena[arena] = vmem_create(c, NULL, 0,
3286 		    sizeof (uint32_t), segkmem_alloc_permanent, segkmem_free,
3287 		    texthole_source[arena], 0, VM_SLEEP);
3288 	}
3289 
3290 	mutex_exit(&texthole_lock);
3291 
3292 	ASSERT(texthole_arena[arena] != NULL);
3293 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3294 	return (vmem_alloc(texthole_arena[arena], size,
3295 	    VM_BESTFIT | VM_NOSLEEP));
3296 }
3297 
3298 void
3299 kobj_texthole_free(caddr_t addr, size_t size)
3300 {
3301 	int arena = HEAPTEXT_ARENA(addr);
3302 
3303 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3304 	ASSERT(texthole_arena[arena] != NULL);
3305 	vmem_free(texthole_arena[arena], addr, size);
3306 }
3307 
3308 void
3309 release_bootstrap(void)
3310 {
3311 	if (&cif_init)
3312 		cif_init();
3313 }
3314