xref: /titanic_50/usr/src/uts/sun4/os/startup.c (revision 97682a823d4ed139a3af5f6e1385b1de019d55d3)
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 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/machsystm.h>
30 #include <sys/archsystm.h>
31 #include <sys/vm.h>
32 #include <sys/cpu.h>
33 #include <sys/atomic.h>
34 #include <sys/reboot.h>
35 #include <sys/kdi.h>
36 #include <sys/bootconf.h>
37 #include <sys/memlist_plat.h>
38 #include <sys/memlist_impl.h>
39 #include <sys/prom_plat.h>
40 #include <sys/prom_isa.h>
41 #include <sys/autoconf.h>
42 #include <sys/intreg.h>
43 #include <sys/ivintr.h>
44 #include <sys/fpu/fpusystm.h>
45 #include <sys/iommutsb.h>
46 #include <vm/vm_dep.h>
47 #include <vm/seg_dev.h>
48 #include <vm/seg_kmem.h>
49 #include <vm/seg_kpm.h>
50 #include <vm/seg_map.h>
51 #include <vm/seg_kp.h>
52 #include <sys/sysconf.h>
53 #include <vm/hat_sfmmu.h>
54 #include <sys/kobj.h>
55 #include <sys/sun4asi.h>
56 #include <sys/clconf.h>
57 #include <sys/platform_module.h>
58 #include <sys/panic.h>
59 #include <sys/cpu_sgnblk_defs.h>
60 #include <sys/clock.h>
61 #include <sys/cmn_err.h>
62 #include <sys/promif.h>
63 #include <sys/prom_debug.h>
64 #include <sys/traptrace.h>
65 #include <sys/memnode.h>
66 #include <sys/mem_cage.h>
67 
68 extern void setup_trap_table(void);
69 extern void cpu_intrq_setup(struct cpu *);
70 extern void cpu_intrq_register(struct cpu *);
71 extern void contig_mem_init(void);
72 extern void mach_dump_buffer_init(void);
73 extern void mach_descrip_init(void);
74 extern void mach_memscrub(void);
75 extern void mach_fpras(void);
76 extern void mach_cpu_halt_idle(void);
77 extern void mach_hw_copy_limit(void);
78 extern void load_tod_module(void);
79 #pragma weak load_tod_module
80 
81 extern int ndata_alloc_mmfsa(struct memlist *ndata);
82 #pragma weak ndata_alloc_mmfsa
83 
84 extern void parse_idprom(void);
85 extern void add_vx_handler(char *, int, void (*)(cell_t *));
86 extern void mem_config_init(void);
87 extern void memseg_remap_init(void);
88 
89 /*
90  * External Data:
91  */
92 extern int vac_size;	/* cache size in bytes */
93 extern uint_t vac_mask;	/* VAC alignment consistency mask */
94 extern uint_t vac_colors;
95 
96 /*
97  * Global Data Definitions:
98  */
99 
100 /*
101  * XXX - Don't port this to new architectures
102  * A 3rd party volume manager driver (vxdm) depends on the symbol romp.
103  * 'romp' has no use with a prom with an IEEE 1275 client interface.
104  * The driver doesn't use the value, but it depends on the symbol.
105  */
106 void *romp;		/* veritas driver won't load without romp 4154976 */
107 /*
108  * Declare these as initialized data so we can patch them.
109  */
110 pgcnt_t physmem = 0;	/* memory size in pages, patch if you want less */
111 pgcnt_t segkpsize =
112     btop(SEGKPDEFSIZE);	/* size of segkp segment in pages */
113 uint_t segmap_percent = 12; /* Size of segmap segment */
114 
115 int use_cache = 1;		/* cache not reliable (605 bugs) with MP */
116 int vac_copyback = 1;
117 char *cache_mode = NULL;
118 int use_mix = 1;
119 int prom_debug = 0;
120 
121 struct bootops *bootops = 0;	/* passed in from boot in %o2 */
122 caddr_t boot_tba;		/* %tba at boot - used by kmdb */
123 uint_t	tba_taken_over = 0;
124 
125 caddr_t s_text;			/* start of kernel text segment */
126 caddr_t e_text;			/* end of kernel text segment */
127 caddr_t s_data;			/* start of kernel data segment */
128 caddr_t e_data;			/* end of kernel data segment */
129 
130 caddr_t modtext;		/* beginning of module text */
131 size_t	modtext_sz;		/* size of module text */
132 caddr_t moddata;		/* beginning of module data reserve */
133 caddr_t e_moddata;		/* end of module data reserve */
134 
135 /*
136  * End of first block of contiguous kernel in 32-bit virtual address space
137  */
138 caddr_t		econtig32;	/* end of first blk of contiguous kernel */
139 
140 caddr_t		ncbase;		/* beginning of non-cached segment */
141 caddr_t		ncend;		/* end of non-cached segment */
142 caddr_t		sdata;		/* beginning of data segment */
143 
144 caddr_t		extra_etva;	/* beginning of unused nucleus text */
145 pgcnt_t		extra_etpg;	/* number of pages of unused nucleus text */
146 
147 size_t	ndata_remain_sz;	/* bytes from end of data to 4MB boundary */
148 caddr_t	nalloc_base;		/* beginning of nucleus allocation */
149 caddr_t nalloc_end;		/* end of nucleus allocatable memory */
150 caddr_t valloc_base;		/* beginning of kvalloc segment	*/
151 
152 caddr_t kmem64_base;		/* base of kernel mem segment in 64-bit space */
153 caddr_t kmem64_end;		/* end of kernel mem segment in 64-bit space */
154 
155 uintptr_t shm_alignment = 0;	/* VAC address consistency modulus */
156 struct memlist *phys_install;	/* Total installed physical memory */
157 struct memlist *phys_avail;	/* Available (unreserved) physical memory */
158 struct memlist *virt_avail;	/* Available (unmapped?) virtual memory */
159 struct memlist ndata;		/* memlist of nucleus allocatable memory */
160 int memexp_flag;		/* memory expansion card flag */
161 uint64_t ecache_flushaddr;	/* physical address used for flushing E$ */
162 pgcnt_t obp_pages;		/* Physical pages used by OBP */
163 
164 /*
165  * VM data structures
166  */
167 long page_hashsz;		/* Size of page hash table (power of two) */
168 struct page *pp_base;		/* Base of system page struct array */
169 size_t pp_sz;			/* Size in bytes of page struct array */
170 struct page **page_hash;	/* Page hash table */
171 struct seg ktextseg;		/* Segment used for kernel executable image */
172 struct seg kvalloc;		/* Segment used for "valloc" mapping */
173 struct seg kpseg;		/* Segment used for pageable kernel virt mem */
174 struct seg ktexthole;		/* Segment used for nucleus text hole */
175 struct seg kmapseg;		/* Segment used for generic kernel mappings */
176 struct seg kpmseg;		/* Segment used for physical mapping */
177 struct seg kdebugseg;		/* Segment used for the kernel debugger */
178 
179 uintptr_t kpm_pp_base;		/* Base of system kpm_page array */
180 size_t	kpm_pp_sz;		/* Size of system kpm_page array */
181 pgcnt_t	kpm_npages;		/* How many kpm pages are managed */
182 
183 struct seg *segkp = &kpseg;	/* Pageable kernel virtual memory segment */
184 struct seg *segkmap = &kmapseg;	/* Kernel generic mapping segment */
185 struct seg *segkpm = &kpmseg;	/* 64bit kernel physical mapping segment */
186 
187 /*
188  * debugger pages (if allocated)
189  */
190 struct vnode kdebugvp;
191 
192 /*
193  * Segment for relocated kernel structures in 64-bit large RAM kernels
194  */
195 struct seg kmem64;
196 
197 struct memseg *memseg_base;
198 size_t memseg_sz;		/* Used to translate a va to page */
199 struct vnode unused_pages_vp;
200 
201 /*
202  * VM data structures allocated early during boot.
203  */
204 size_t pagehash_sz;
205 uint64_t memlist_sz;
206 
207 char tbr_wr_addr_inited = 0;
208 
209 
210 /*
211  * Static Routines:
212  */
213 static void memlist_add(uint64_t, uint64_t, struct memlist **,
214 	struct memlist **);
215 static void kphysm_init(page_t *, struct memseg *, pgcnt_t, uintptr_t,
216 	pgcnt_t);
217 static void kvm_init(void);
218 
219 static void startup_init(void);
220 static void startup_memlist(void);
221 static void startup_modules(void);
222 static void startup_bop_gone(void);
223 static void startup_vm(void);
224 static void startup_end(void);
225 static void setup_cage_params(void);
226 static void startup_create_io_node(void);
227 
228 static pgcnt_t npages;
229 static struct memlist *memlist;
230 void *memlist_end;
231 
232 static pgcnt_t bop_alloc_pages;
233 static caddr_t hblk_base;
234 uint_t hblk_alloc_dynamic = 0;
235 uint_t hblk1_min = H1MIN;
236 uint_t hblk8_min;
237 
238 
239 /*
240  * Hooks for unsupported platforms and down-rev firmware
241  */
242 int iam_positron(void);
243 #pragma weak iam_positron
244 static void do_prom_version_check(void);
245 static void kpm_init(void);
246 static void kpm_npages_setup(int);
247 static void kpm_memseg_init(void);
248 
249 /*
250  * After receiving a thermal interrupt, this is the number of seconds
251  * to delay before shutting off the system, assuming
252  * shutdown fails.  Use /etc/system to change the delay if this isn't
253  * large enough.
254  */
255 int thermal_powerdown_delay = 1200;
256 
257 /*
258  * Used to hold off page relocations into the cage until OBP has completed
259  * its boot-time handoff of its resources to the kernel.
260  */
261 int page_relocate_ready = 0;
262 
263 /*
264  * Enable some debugging messages concerning memory usage...
265  */
266 #ifdef  DEBUGGING_MEM
267 static int debugging_mem;
268 static void
269 printmemlist(char *title, struct memlist *list)
270 {
271 	if (!debugging_mem)
272 		return;
273 
274 	printf("%s\n", title);
275 
276 	while (list) {
277 		prom_printf("\taddr = 0x%x %8x, size = 0x%x %8x\n",
278 		    (uint32_t)(list->address >> 32), (uint32_t)list->address,
279 		    (uint32_t)(list->size >> 32), (uint32_t)(list->size));
280 		list = list->next;
281 	}
282 }
283 
284 void
285 printmemseg(struct memseg *memseg)
286 {
287 	if (!debugging_mem)
288 		return;
289 
290 	printf("memseg\n");
291 
292 	while (memseg) {
293 		prom_printf("\tpage = 0x%p, epage = 0x%p, "
294 		    "pfn = 0x%x, epfn = 0x%x\n",
295 		    memseg->pages, memseg->epages,
296 		    memseg->pages_base, memseg->pages_end);
297 		memseg = memseg->next;
298 	}
299 }
300 
301 #define	debug_pause(str)	halt((str))
302 #define	MPRINTF(str)		if (debugging_mem) prom_printf((str))
303 #define	MPRINTF1(str, a)	if (debugging_mem) prom_printf((str), (a))
304 #define	MPRINTF2(str, a, b)	if (debugging_mem) prom_printf((str), (a), (b))
305 #define	MPRINTF3(str, a, b, c) \
306 	if (debugging_mem) prom_printf((str), (a), (b), (c))
307 #else	/* DEBUGGING_MEM */
308 #define	MPRINTF(str)
309 #define	MPRINTF1(str, a)
310 #define	MPRINTF2(str, a, b)
311 #define	MPRINTF3(str, a, b, c)
312 #endif	/* DEBUGGING_MEM */
313 
314 /* Simple message to indicate that the bootops pointer has been zeroed */
315 #ifdef DEBUG
316 static int bootops_gone_on = 0;
317 #define	BOOTOPS_GONE() \
318 	if (bootops_gone_on) \
319 		prom_printf("The bootops vec is zeroed now!\n");
320 #else
321 #define	BOOTOPS_GONE()
322 #endif /* DEBUG */
323 
324 /*
325  * Monitor pages may not be where this says they are.
326  * and the debugger may not be there either.
327  *
328  * Note that 'pages' here are *physical* pages, which are 8k on sun4u.
329  *
330  *                        Physical memory layout
331  *                     (not necessarily contiguous)
332  *                       (THIS IS SOMEWHAT WRONG)
333  *                       /-----------------------\
334  *                       |       monitor pages   |
335  *             availmem -|-----------------------|
336  *                       |                       |
337  *                       |       page pool       |
338  *                       |                       |
339  *                       |-----------------------|
340  *                       |   configured tables   |
341  *                       |       buffers         |
342  *            firstaddr -|-----------------------|
343  *                       |   hat data structures |
344  *                       |-----------------------|
345  *                       |    kernel data, bss   |
346  *                       |-----------------------|
347  *                       |    interrupt stack    |
348  *                       |-----------------------|
349  *                       |    kernel text (RO)   |
350  *                       |-----------------------|
351  *                       |    trap table (4k)    |
352  *                       |-----------------------|
353  *               page 1  |      panicbuf         |
354  *                       |-----------------------|
355  *               page 0  |       reclaimed       |
356  *                       |_______________________|
357  *
358  *
359  *
360  *                    Kernel's Virtual Memory Layout.
361  *                       /-----------------------\
362  * 0xFFFFFFFF.FFFFFFFF  -|                       |-
363  *                       |   OBP's virtual page  |
364  *                       |        tables         |
365  * 0xFFFFFFFC.00000000  -|-----------------------|-
366  *                       :                       :
367  *                       :                       :
368  * 0xFFFFFE00.00000000  -|-----------------------|-
369  *                       |                       |  Ultrasparc I/II support
370  *                       |    segkpm segment     |  up to 2TB of physical
371  *                       | (64-bit kernel ONLY)  |  memory, VAC has 2 colors
372  *                       |                       |
373  * 0xFFFFFA00.00000000  -|-----------------------|- 2TB segkpm alignment
374  *                       :                       :
375  *                       :                       :
376  * 0xFFFFF810.00000000  -|-----------------------|- hole_end
377  *                       |                       |      ^
378  *                       |  UltraSPARC I/II call |      |
379  *                       | bug requires an extra |      |
380  *                       | 4 GB of space between |      |
381  *                       |   hole and used RAM   |	|
382  *                       |                       |      |
383  * 0xFFFFF800.00000000  -|-----------------------|-     |
384  *                       |                       |      |
385  *                       | Virtual Address Hole  |   UltraSPARC
386  *                       |  on UltraSPARC I/II   |  I/II * ONLY *
387  *                       |                       |      |
388  * 0x00000800.00000000  -|-----------------------|-     |
389  *                       |                       |      |
390  *                       |  UltraSPARC I/II call |      |
391  *                       | bug requires an extra |      |
392  *                       | 4 GB of space between |      |
393  *                       |   hole and used RAM   |      |
394  *                       |                       |      v
395  * 0x000007FF.00000000  -|-----------------------|- hole_start -----
396  *                       :                       :		   ^
397  *                       :                       :		   |
398  * 0x00000XXX.XXXXXXXX  -|-----------------------|- kmem64_end	   |
399  *                       |                       |		   |
400  *                       |   64-bit kernel ONLY  |		   |
401  *                       |                       |		   |
402  *                       |    kmem64 segment     |		   |
403  *                       |                       |		   |
404  *                       | (Relocated extra HME  |	     Approximately
405  *                       |   block allocations,  |	    1 TB of virtual
406  *                       |   memnode freelists,  |	     address space
407  *                       |    HME hash buckets,  |		   |
408  *                       | mml_table, kpmp_table,|		   |
409  *                       |  page_t array and     |		   |
410  *                       |  hashblock pool to    |		   |
411  *                       |   avoid hard-coded    |		   |
412  *                       |     32-bit vaddr      |		   |
413  *                       |     limitations)      |		   |
414  *                       |                       |		   v
415  * 0x00000700.00000000  -|-----------------------|- SYSLIMIT (kmem64_base)
416  *                       |                       |
417  *                       |  segkmem segment      | (SYSLIMIT - SYSBASE = 4TB)
418  *                       |                       |
419  * 0x00000300.00000000  -|-----------------------|- SYSBASE
420  *                       :                       :
421  *                       :                       :
422  *                      -|-----------------------|-
423  *                       |                       |
424  *                       |  segmap segment       |   SEGMAPSIZE (1/8th physmem,
425  *                       |                       |               256G MAX)
426  * 0x000002a7.50000000  -|-----------------------|- SEGMAPBASE
427  *                       :                       :
428  *                       :                       :
429  *                      -|-----------------------|-
430  *                       |                       |
431  *                       |       segkp           |    SEGKPSIZE (2GB)
432  *                       |                       |
433  *                       |                       |
434  * 0x000002a1.00000000  -|-----------------------|- SEGKPBASE
435  *                       |                       |
436  * 0x000002a0.00000000  -|-----------------------|- MEMSCRUBBASE
437  *                       |                       |       (SEGKPBASE - 0x400000)
438  * 0x0000029F.FFE00000  -|-----------------------|- ARGSBASE
439  *                       |                       |       (MEMSCRUBBASE - NCARGS)
440  * 0x0000029F.FFD80000  -|-----------------------|- PPMAPBASE
441  *                       |                       |       (ARGSBASE - PPMAPSIZE)
442  * 0x0000029F.FFD00000  -|-----------------------|- PPMAP_FAST_BASE
443  *                       |                       |
444  * 0x0000029F.FF980000  -|-----------------------|- PIOMAPBASE
445  *                       |                       |
446  * 0x0000029F.FF580000  -|-----------------------|- NARG_BASE
447  *                       :                       :
448  *                       :                       :
449  * 0x00000000.FFFFFFFF  -|-----------------------|- OFW_END_ADDR
450  *                       |                       |
451  *                       |         OBP           |
452  *                       |                       |
453  * 0x00000000.F0000000  -|-----------------------|- OFW_START_ADDR
454  *                       |         kmdb          |
455  * 0x00000000.EDD00000  -|-----------------------|- SEGDEBUGBASE
456  *                       :                       :
457  *                       :                       :
458  * 0x00000000.7c000000  -|-----------------------|- SYSLIMIT32
459  *                       |                       |
460  *                       |  segkmem32 segment    | (SYSLIMIT32 - SYSBASE32 =
461  *                       |                       |    ~64MB)
462  * 0x00000000.78002000  -|-----------------------|
463  *                       |     panicbuf          |
464  * 0x00000000.78000000  -|-----------------------|- SYSBASE32
465  *                       :                       :
466  *                       :                       :
467  *                       |                       |
468  *                       |-----------------------|- econtig32
469  *                       |    vm structures      |
470  * 0x00000000.01C00000   |-----------------------|- nalloc_end
471  *                       |         TSBs          |
472  *                       |-----------------------|- end/nalloc_base
473  *                       |   kernel data & bss   |
474  * 0x00000000.01800000  -|-----------------------|
475  *                       :   nucleus text hole   :
476  * 0x00000000.01400000  -|-----------------------|
477  *                       :                       :
478  *                       |-----------------------|
479  *                       |      module text      |
480  *                       |-----------------------|- e_text/modtext
481  *                       |      kernel text      |
482  *                       |-----------------------|
483  *                       |    trap table (48k)   |
484  * 0x00000000.01000000  -|-----------------------|- KERNELBASE
485  *                       | reserved for trapstat |} TSTAT_TOTAL_SIZE
486  *                       |-----------------------|
487  *                       |                       |
488  *                       |        invalid        |
489  *                       |                       |
490  * 0x00000000.00000000  _|_______________________|
491  *
492  *
493  *
494  *                   32-bit User Virtual Memory Layout.
495  *                       /-----------------------\
496  *                       |                       |
497  *                       |        invalid        |
498  *                       |                       |
499  *          0xFFC00000  -|-----------------------|- USERLIMIT
500  *                       |       user stack      |
501  *                       :                       :
502  *                       :                       :
503  *                       :                       :
504  *                       |       user data       |
505  *                      -|-----------------------|-
506  *                       |       user text       |
507  *          0x00002000  -|-----------------------|-
508  *                       |       invalid         |
509  *          0x00000000  _|_______________________|
510  *
511  *
512  *
513  *                   64-bit User Virtual Memory Layout.
514  *                       /-----------------------\
515  *                       |                       |
516  *                       |        invalid        |
517  *                       |                       |
518  *  0xFFFFFFFF.80000000 -|-----------------------|- USERLIMIT
519  *                       |       user stack      |
520  *                       :                       :
521  *                       :                       :
522  *                       :                       :
523  *                       |       user data       |
524  *                      -|-----------------------|-
525  *                       |       user text       |
526  *  0x00000000.00100000 -|-----------------------|-
527  *                       |       invalid         |
528  *  0x00000000.00000000 _|_______________________|
529  */
530 
531 extern caddr_t ecache_init_scrub_flush_area(caddr_t alloc_base);
532 extern uint64_t ecache_flush_address(void);
533 
534 #pragma weak load_platform_modules
535 #pragma weak starcat_startup_memlist
536 #pragma weak ecache_init_scrub_flush_area
537 #pragma weak ecache_flush_address
538 
539 
540 /*
541  * By default the DR Cage is enabled for maximum OS
542  * MPSS performance.  Users needing to disable the cage mechanism
543  * can set this variable to zero via /etc/system.
544  * Disabling the cage on systems supporting Dynamic Reconfiguration (DR)
545  * will result in loss of DR functionality.
546  * Platforms wishing to disable kernel Cage by default
547  * should do so in their set_platform_defaults() routine.
548  */
549 int	kernel_cage_enable = 1;
550 
551 static void
552 setup_cage_params(void)
553 {
554 	void (*func)(void);
555 
556 	func = (void (*)(void))kobj_getsymvalue("set_platform_cage_params", 0);
557 	if (func != NULL) {
558 		(*func)();
559 		return;
560 	}
561 
562 	if (kernel_cage_enable == 0) {
563 		return;
564 	}
565 	kcage_range_lock();
566 	if (kcage_range_init(phys_avail, 1) == 0) {
567 		kcage_init(total_pages / 256);
568 	}
569 	kcage_range_unlock();
570 
571 	if (kcage_on) {
572 		cmn_err(CE_NOTE, "!Kernel Cage is ENABLED");
573 	} else {
574 		cmn_err(CE_NOTE, "!Kernel Cage is DISABLED");
575 	}
576 
577 }
578 
579 /*
580  * Machine-dependent startup code
581  */
582 void
583 startup(void)
584 {
585 	startup_init();
586 	if (&startup_platform)
587 		startup_platform();
588 	startup_memlist();
589 	startup_modules();
590 	setup_cage_params();
591 	startup_bop_gone();
592 	startup_vm();
593 	startup_end();
594 }
595 
596 struct regs sync_reg_buf;
597 uint64_t sync_tt;
598 
599 void
600 sync_handler(void)
601 {
602 	struct  trap_info 	ti;
603 	int i;
604 
605 	/*
606 	 * Prevent trying to talk to the other CPUs since they are
607 	 * sitting in the prom and won't reply.
608 	 */
609 	for (i = 0; i < NCPU; i++) {
610 		if ((i != CPU->cpu_id) && CPU_XCALL_READY(i)) {
611 			cpu[i]->cpu_flags &= ~CPU_READY;
612 			cpu[i]->cpu_flags |= CPU_QUIESCED;
613 			CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
614 		}
615 	}
616 
617 	/*
618 	 * We've managed to get here without going through the
619 	 * normal panic code path. Try and save some useful
620 	 * information.
621 	 */
622 	if (!panicstr && (curthread->t_panic_trap == NULL)) {
623 		ti.trap_type = sync_tt;
624 		ti.trap_regs = &sync_reg_buf;
625 		ti.trap_addr = NULL;
626 		ti.trap_mmu_fsr = 0x0;
627 
628 		curthread->t_panic_trap = &ti;
629 	}
630 
631 	/*
632 	 * If we're re-entering the panic path, update the signature
633 	 * block so that the SC knows we're in the second part of panic.
634 	 */
635 	if (panicstr)
636 		CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
637 
638 	nopanicdebug = 1; /* do not perform debug_enter() prior to dump */
639 	panic("sync initiated");
640 }
641 
642 
643 static void
644 startup_init(void)
645 {
646 	/*
647 	 * We want to save the registers while we're still in OBP
648 	 * so that we know they haven't been fiddled with since.
649 	 * (In principle, OBP can't change them just because it
650 	 * makes a callback, but we'd rather not depend on that
651 	 * behavior.)
652 	 */
653 	char		sync_str[] =
654 		"warning @ warning off : sync "
655 		"%%tl-c %%tstate h# %p x! "
656 		"%%g1 h# %p x! %%g2 h# %p x! %%g3 h# %p x! "
657 		"%%g4 h# %p x! %%g5 h# %p x! %%g6 h# %p x! "
658 		"%%g7 h# %p x! %%o0 h# %p x! %%o1 h# %p x! "
659 		"%%o2 h# %p x! %%o3 h# %p x! %%o4 h# %p x! "
660 		"%%o5 h# %p x! %%o6 h# %p x! %%o7 h# %p x! "
661 		"%%tl-c %%tpc h# %p x! %%tl-c %%tnpc h# %p x! "
662 		"%%y h# %p l! %%tl-c %%tt h# %p x! "
663 		"sync ; warning !";
664 
665 	/*
666 	 * 20 == num of %p substrings
667 	 * 16 == max num of chars %p will expand to.
668 	 */
669 	char 		bp[sizeof (sync_str) + 16 * 20];
670 
671 	(void) check_boot_version(BOP_GETVERSION(bootops));
672 
673 	/*
674 	 * Initialize ptl1 stack for the 1st CPU.
675 	 */
676 	ptl1_init_cpu(&cpu0);
677 
678 	/*
679 	 * Initialize the address map for cache consistent mappings
680 	 * to random pages; must be done after vac_size is set.
681 	 */
682 	ppmapinit();
683 
684 	/*
685 	 * Initialize the PROM callback handler.
686 	 */
687 	init_vx_handler();
688 
689 	/*
690 	 * have prom call sync_callback() to handle the sync and
691 	 * save some useful information which will be stored in the
692 	 * core file later.
693 	 */
694 	(void) sprintf((char *)bp, sync_str,
695 		(void *)&sync_reg_buf.r_tstate, (void *)&sync_reg_buf.r_g1,
696 		(void *)&sync_reg_buf.r_g2, (void *)&sync_reg_buf.r_g3,
697 		(void *)&sync_reg_buf.r_g4, (void *)&sync_reg_buf.r_g5,
698 		(void *)&sync_reg_buf.r_g6, (void *)&sync_reg_buf.r_g7,
699 		(void *)&sync_reg_buf.r_o0, (void *)&sync_reg_buf.r_o1,
700 		(void *)&sync_reg_buf.r_o2, (void *)&sync_reg_buf.r_o3,
701 		(void *)&sync_reg_buf.r_o4, (void *)&sync_reg_buf.r_o5,
702 		(void *)&sync_reg_buf.r_o6, (void *)&sync_reg_buf.r_o7,
703 		(void *)&sync_reg_buf.r_pc, (void *)&sync_reg_buf.r_npc,
704 		(void *)&sync_reg_buf.r_y, (void *)&sync_tt);
705 	prom_interpret(bp, 0, 0, 0, 0, 0);
706 	add_vx_handler("sync", 1, (void (*)(cell_t *))sync_handler);
707 }
708 
709 static u_longlong_t *boot_physinstalled, *boot_physavail, *boot_virtavail;
710 static size_t boot_physinstalled_len, boot_physavail_len, boot_virtavail_len;
711 
712 #define	IVSIZE	((MAXIVNUM + 1) * sizeof (struct intr_vector))
713 
714 /*
715  * As OBP takes up some RAM when the system boots, pages will already be "lost"
716  * to the system and reflected in npages by the time we see it.
717  *
718  * We only want to allocate kernel structures in the 64-bit virtual address
719  * space on systems with enough RAM to make the overhead of keeping track of
720  * an extra kernel memory segment worthwhile.
721  *
722  * Since OBP has already performed its memory allocations by this point, if we
723  * have more than MINMOVE_RAM_MB MB of RAM left free, go ahead and map
724  * memory in the 64-bit virtual address space; otherwise keep allocations
725  * contiguous with we've mapped so far in the 32-bit virtual address space.
726  */
727 #define	MINMOVE_RAM_MB	((size_t)1900)
728 #define	MB_TO_BYTES(mb)	((mb) * 1048576ul)
729 
730 pgcnt_t	tune_npages = (pgcnt_t)
731 	(MB_TO_BYTES(MINMOVE_RAM_MB)/ (size_t)MMU_PAGESIZE);
732 
733 static void
734 startup_memlist(void)
735 {
736 	size_t alloc_sz;
737 	size_t ctrs_sz;
738 	caddr_t alloc_base;
739 	caddr_t ctrs_base, ctrs_end;
740 	caddr_t memspace;
741 	caddr_t va;
742 	int memblocks = 0;
743 	struct memlist *cur;
744 	size_t syslimit = (size_t)SYSLIMIT;
745 	size_t sysbase = (size_t)SYSBASE;
746 	int alloc_alignsize = MMU_PAGESIZE;
747 	extern void page_coloring_init(void);
748 
749 	/*
750 	 * Initialize enough of the system to allow kmem_alloc to work by
751 	 * calling boot to allocate its memory until the time that
752 	 * kvm_init is completed.  The page structs are allocated after
753 	 * rounding up end to the nearest page boundary; the memsegs are
754 	 * initialized and the space they use comes from the kernel heap.
755 	 * With appropriate initialization, they can be reallocated later
756 	 * to a size appropriate for the machine's configuration.
757 	 *
758 	 * At this point, memory is allocated for things that will never
759 	 * need to be freed, this used to be "valloced".  This allows a
760 	 * savings as the pages don't need page structures to describe
761 	 * them because them will not be managed by the vm system.
762 	 */
763 
764 	/*
765 	 * We're loaded by boot with the following configuration (as
766 	 * specified in the sun4u/conf/Mapfile):
767 	 *
768 	 * 	text:		4 MB chunk aligned on a 4MB boundary
769 	 * 	data & bss:	4 MB chunk aligned on a 4MB boundary
770 	 *
771 	 * These two chunks will eventually be mapped by 2 locked 4MB
772 	 * ttes and will represent the nucleus of the kernel.  This gives
773 	 * us some free space that is already allocated, some or all of
774 	 * which is made available to kernel module text.
775 	 *
776 	 * The free space in the data-bss chunk is used for nucleus
777 	 * allocatable data structures and we reserve it using the
778 	 * nalloc_base and nalloc_end variables.  This space is currently
779 	 * being used for hat data structures required for tlb miss
780 	 * handling operations.  We align nalloc_base to a l2 cache
781 	 * linesize because this is the line size the hardware uses to
782 	 * maintain cache coherency.
783 	 * 256K is carved out for module data.
784 	 */
785 
786 	nalloc_base = (caddr_t)roundup((uintptr_t)e_data, MMU_PAGESIZE);
787 	moddata = nalloc_base;
788 	e_moddata = nalloc_base + MODDATA;
789 	nalloc_base = e_moddata;
790 
791 	nalloc_end = (caddr_t)roundup((uintptr_t)nalloc_base, MMU_PAGESIZE4M);
792 	valloc_base = nalloc_base;
793 
794 	/*
795 	 * Calculate the start of the data segment.
796 	 */
797 	sdata = (caddr_t)((uintptr_t)e_data & MMU_PAGEMASK4M);
798 
799 	PRM_DEBUG(moddata);
800 	PRM_DEBUG(nalloc_base);
801 	PRM_DEBUG(nalloc_end);
802 	PRM_DEBUG(sdata);
803 
804 	/*
805 	 * Remember any slop after e_text so we can give it to the modules.
806 	 */
807 	PRM_DEBUG(e_text);
808 	modtext = (caddr_t)roundup((uintptr_t)e_text, MMU_PAGESIZE);
809 	if (((uintptr_t)modtext & MMU_PAGEMASK4M) != (uintptr_t)s_text)
810 		panic("nucleus text overflow");
811 	modtext_sz = (caddr_t)roundup((uintptr_t)modtext, MMU_PAGESIZE4M) -
812 	    modtext;
813 	PRM_DEBUG(modtext);
814 	PRM_DEBUG(modtext_sz);
815 
816 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
817 	    &boot_physavail, &boot_physavail_len,
818 	    &boot_virtavail, &boot_virtavail_len);
819 	/*
820 	 * Remember what the physically available highest page is
821 	 * so that dumpsys works properly, and find out how much
822 	 * memory is installed.
823 	 */
824 	installed_top_size_memlist_array(boot_physinstalled,
825 	    boot_physinstalled_len, &physmax, &physinstalled);
826 	PRM_DEBUG(physinstalled);
827 	PRM_DEBUG(physmax);
828 
829 	/* Fill out memory nodes config structure */
830 	startup_build_mem_nodes(boot_physinstalled, boot_physinstalled_len);
831 
832 	/*
833 	 * Get the list of physically available memory to size
834 	 * the number of page structures needed.
835 	 */
836 	size_physavail(boot_physavail, boot_physavail_len, &npages, &memblocks);
837 	/*
838 	 * This first snap shot of npages can represent the pages used
839 	 * by OBP's text and data approximately. This is used in the
840 	 * the calculation of the kernel size
841 	 */
842 	obp_pages = physinstalled - npages;
843 
844 
845 	/*
846 	 * On small-memory systems (<MODTEXT_SM_SIZE MB, currently 256MB), the
847 	 * in-nucleus module text is capped to MODTEXT_SM_CAP bytes (currently
848 	 * 2MB) and any excess pages are put on physavail.  The assumption is
849 	 * that small-memory systems will need more pages more than they'll
850 	 * need efficiently-mapped module texts.
851 	 */
852 	if ((physinstalled < mmu_btop(MODTEXT_SM_SIZE << 20)) &&
853 	    modtext_sz > MODTEXT_SM_CAP) {
854 		extra_etpg = mmu_btop(modtext_sz - MODTEXT_SM_CAP);
855 		modtext_sz = MODTEXT_SM_CAP;
856 	} else
857 		extra_etpg = 0;
858 	PRM_DEBUG(extra_etpg);
859 	PRM_DEBUG(modtext_sz);
860 	extra_etva = modtext + modtext_sz;
861 	PRM_DEBUG(extra_etva);
862 
863 	/*
864 	 * Account for any pages after e_text and e_data.
865 	 */
866 	npages += extra_etpg;
867 	npages += mmu_btopr(nalloc_end - nalloc_base);
868 	PRM_DEBUG(npages);
869 
870 	/*
871 	 * npages is the maximum of available physical memory possible.
872 	 * (ie. it will never be more than this)
873 	 */
874 
875 	/*
876 	 * initialize the nucleus memory allocator.
877 	 */
878 	ndata_alloc_init(&ndata, (uintptr_t)nalloc_base, (uintptr_t)nalloc_end);
879 
880 	/*
881 	 * Allocate mmu fault status area from the nucleus data area.
882 	 */
883 	if ((&ndata_alloc_mmfsa != NULL) && (ndata_alloc_mmfsa(&ndata) != 0))
884 		cmn_err(CE_PANIC, "no more nucleus memory after mfsa alloc");
885 
886 	/*
887 	 * Allocate kernel TSBs from the nucleus data area.
888 	 */
889 	if (ndata_alloc_tsbs(&ndata, npages) != 0)
890 		cmn_err(CE_PANIC, "no more nucleus memory after tsbs alloc");
891 
892 	/*
893 	 * Allocate cpus structs from the nucleus data area.
894 	 */
895 	if (ndata_alloc_cpus(&ndata) != 0)
896 		cmn_err(CE_PANIC, "no more nucleus memory after cpu alloc");
897 
898 	/*
899 	 * Allocate dmv dispatch table from the nucleus data area.
900 	 */
901 	if (ndata_alloc_dmv(&ndata) != 0)
902 		cmn_err(CE_PANIC, "no more nucleus memory after dmv alloc");
903 
904 
905 	page_coloring_init();
906 
907 	/*
908 	 * Allocate page_freelists bin headers for memnode 0 from the
909 	 * nucleus data area.
910 	 */
911 	if (ndata_alloc_page_freelists(&ndata, 0) != 0)
912 		cmn_err(CE_PANIC,
913 		    "no more nucleus memory after page free lists alloc");
914 
915 	if (kpm_enable) {
916 		kpm_init();
917 		/*
918 		 * kpm page space -- Update kpm_npages and make the
919 		 * same assumption about fragmenting as it is done
920 		 * for memseg_sz.
921 		 */
922 		kpm_npages_setup(memblocks + 4);
923 	}
924 
925 	/*
926 	 * Allocate hat related structs from the nucleus data area.
927 	 */
928 	if (ndata_alloc_hat(&ndata, npages, kpm_npages) != 0)
929 		cmn_err(CE_PANIC, "no more nucleus memory after hat alloc");
930 
931 	/*
932 	 * We want to do the BOP_ALLOCs before the real allocation of page
933 	 * structs in order to not have to allocate page structs for this
934 	 * memory.  We need to calculate a virtual address because we want
935 	 * the page structs to come before other allocations in virtual address
936 	 * space.  This is so some (if not all) of page structs can actually
937 	 * live in the nucleus.
938 	 */
939 
940 	/*
941 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
942 	 *
943 	 * There are comments all over the SFMMU code warning of dire
944 	 * consequences if the TSBs are moved out of 32-bit space.  This
945 	 * is largely because the asm code uses "sethi %hi(addr)"-type
946 	 * instructions which will not provide the expected result if the
947 	 * address is a 64-bit one.
948 	 *
949 	 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
950 	 */
951 	alloc_base = (caddr_t)roundup((uintptr_t)nalloc_end, MMU_PAGESIZE);
952 	alloc_base = sfmmu_ktsb_alloc(alloc_base);
953 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
954 	PRM_DEBUG(alloc_base);
955 
956 	/*
957 	 * Allocate IOMMU TSB array.  We do this here so that the physical
958 	 * memory gets deducted from the PROM's physical memory list.
959 	 */
960 	alloc_base = iommu_tsb_init(alloc_base);
961 	alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
962 	    ecache_alignsize);
963 	PRM_DEBUG(alloc_base);
964 
965 	/*
966 	 * Starcat needs its special structures assigned in 32-bit virtual
967 	 * address space because its probing routines execute FCode, and FCode
968 	 * can't handle 64-bit virtual addresses...
969 	 */
970 	if (&starcat_startup_memlist) {
971 		alloc_base = starcat_startup_memlist(alloc_base);
972 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
973 		    ecache_alignsize);
974 		PRM_DEBUG(alloc_base);
975 	}
976 
977 	/*
978 	 * If we have enough memory, use 4M pages for alignment because it
979 	 * greatly reduces the number of TLB misses we take albeit at the cost
980 	 * of possible RAM wastage (degenerate case of 4 MB - MMU_PAGESIZE per
981 	 * allocation.) Still, the speedup on large memory systems (e.g. > 64
982 	 * GB) is quite noticeable, so it is worth the effort to do if we can.
983 	 *
984 	 * Note, however, that this speedup will only occur if the boot PROM
985 	 * uses the largest possible MMU page size possible to map memory
986 	 * requests that are properly aligned and sized (for example, a request
987 	 * for a multiple of 4MB of memory aligned to a 4MB boundary will
988 	 * result in a mapping using a 4MB MMU page.)
989 	 *
990 	 * Even then, the large page mappings will only speed things up until
991 	 * the startup process proceeds a bit further, as when
992 	 * sfmmu_map_prom_mappings() copies page mappings from the PROM to the
993 	 * kernel it remaps everything but the TSBs using 8K pages anyway...
994 	 *
995 	 * At some point in the future, sfmmu_map_prom_mappings() will be
996 	 * rewritten to copy memory mappings to the kernel using the same MMU
997 	 * page sizes the PROM used.  When that occurs, if the PROM did use
998 	 * large MMU pages to map memory, the alignment/sizing work we're
999 	 * doing now should give us a nice extra performance boost, albeit at
1000 	 * the cost of greater RAM usage...
1001 	 */
1002 	alloc_alignsize = ((npages >= tune_npages) ? MMU_PAGESIZE4M :
1003 	    MMU_PAGESIZE);
1004 
1005 	PRM_DEBUG(tune_npages);
1006 	PRM_DEBUG(alloc_alignsize);
1007 
1008 	/*
1009 	 * Save off where the contiguous allocations to date have ended
1010 	 * in econtig32.
1011 	 */
1012 	econtig32 = alloc_base;
1013 	PRM_DEBUG(econtig32);
1014 
1015 	if (econtig32 > (caddr_t)KERNEL_LIMIT32)
1016 		cmn_err(CE_PANIC, "econtig32 too big");
1017 
1018 	/*
1019 	 * To avoid memory allocation collisions in the 32-bit virtual address
1020 	 * space, make allocations from this point forward in 64-bit virtual
1021 	 * address space starting at syslimit and working up.  Also use the
1022 	 * alignment specified by alloc_alignsize, as we may be able to save
1023 	 * ourselves TLB misses by using larger page sizes if they're
1024 	 * available.
1025 	 *
1026 	 * All this is needed because on large memory systems, the default
1027 	 * Solaris allocations will collide with SYSBASE32, which is hard
1028 	 * coded to be at the virtual address 0x78000000.  Therefore, on 64-bit
1029 	 * kernels, move the allocations to a location in the 64-bit virtual
1030 	 * address space space, allowing those structures to grow without
1031 	 * worry.
1032 	 *
1033 	 * On current CPUs we'll run out of physical memory address bits before
1034 	 * we need to worry about the allocations running into anything else in
1035 	 * VM or the virtual address holes on US-I and II, as there's currently
1036 	 * about 1 TB of addressable space before the US-I/II VA hole.
1037 	 */
1038 	kmem64_base = (caddr_t)syslimit;
1039 	PRM_DEBUG(kmem64_base);
1040 
1041 	alloc_base = (caddr_t)roundup((uintptr_t)kmem64_base, alloc_alignsize);
1042 
1043 	/*
1044 	 * If KHME and/or UHME hash buckets won't fit in the nucleus, allocate
1045 	 * them here.
1046 	 */
1047 	if (khme_hash == NULL || uhme_hash == NULL) {
1048 		/*
1049 		 * alloc_hme_buckets() will align alloc_base properly before
1050 		 * assigning the hash buckets, so we don't need to do it
1051 		 * before the call...
1052 		 */
1053 		alloc_base = alloc_hme_buckets(alloc_base, alloc_alignsize);
1054 
1055 		PRM_DEBUG(alloc_base);
1056 		PRM_DEBUG(khme_hash);
1057 		PRM_DEBUG(uhme_hash);
1058 	}
1059 
1060 	/*
1061 	 * Allocate the remaining page freelists.  NUMA systems can
1062 	 * have lots of page freelists, one per node, which quickly
1063 	 * outgrow the amount of nucleus memory available.
1064 	 */
1065 	if (max_mem_nodes > 1) {
1066 		int mnode;
1067 		caddr_t alloc_start = alloc_base;
1068 
1069 		for (mnode = 1; mnode < max_mem_nodes; mnode++) {
1070 			alloc_base = alloc_page_freelists(mnode, alloc_base,
1071 				ecache_alignsize);
1072 		}
1073 
1074 		if (alloc_base > alloc_start) {
1075 			alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1076 				alloc_alignsize);
1077 			if ((caddr_t)BOP_ALLOC(bootops, alloc_start,
1078 				alloc_base - alloc_start,
1079 				alloc_alignsize) != alloc_start)
1080 				cmn_err(CE_PANIC,
1081 					"Unable to alloc page freelists\n");
1082 		}
1083 
1084 		PRM_DEBUG(alloc_base);
1085 	}
1086 
1087 	if (!mml_table) {
1088 		size_t mmltable_sz;
1089 
1090 		/*
1091 		 * We need to allocate the mml_table here because there
1092 		 * was not enough space within the nucleus.
1093 		 */
1094 		mmltable_sz = sizeof (kmutex_t) * mml_table_sz;
1095 		alloc_sz = roundup(mmltable_sz, alloc_alignsize);
1096 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1097 		    alloc_alignsize);
1098 
1099 		if ((mml_table = (kmutex_t *)BOP_ALLOC(bootops, alloc_base,
1100 		    alloc_sz, alloc_alignsize)) != (kmutex_t *)alloc_base)
1101 			panic("mml_table alloc failure");
1102 
1103 		alloc_base += alloc_sz;
1104 		PRM_DEBUG(mml_table);
1105 		PRM_DEBUG(alloc_base);
1106 	}
1107 
1108 	if (kpm_enable && !(kpmp_table || kpmp_stable)) {
1109 		size_t kpmptable_sz;
1110 		caddr_t table;
1111 
1112 		/*
1113 		 * We need to allocate either kpmp_table or kpmp_stable here
1114 		 * because there was not enough space within the nucleus.
1115 		 */
1116 		kpmptable_sz = (kpm_smallpages == 0) ?
1117 				sizeof (kpm_hlk_t) * kpmp_table_sz :
1118 				sizeof (kpm_shlk_t) * kpmp_stable_sz;
1119 
1120 		alloc_sz = roundup(kpmptable_sz, alloc_alignsize);
1121 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1122 		    alloc_alignsize);
1123 
1124 		table = BOP_ALLOC(bootops, alloc_base, alloc_sz,
1125 				alloc_alignsize);
1126 
1127 		if (table != alloc_base)
1128 			panic("kpmp_table or kpmp_stable alloc failure");
1129 
1130 		if (kpm_smallpages == 0) {
1131 			kpmp_table = (kpm_hlk_t *)table;
1132 			PRM_DEBUG(kpmp_table);
1133 		} else {
1134 			kpmp_stable = (kpm_shlk_t *)table;
1135 			PRM_DEBUG(kpmp_stable);
1136 		}
1137 
1138 		alloc_base += alloc_sz;
1139 		PRM_DEBUG(alloc_base);
1140 	}
1141 
1142 	if (&ecache_init_scrub_flush_area) {
1143 		/*
1144 		 * Pass alloc_base directly, as the routine itself is
1145 		 * responsible for any special alignment requirements...
1146 		 */
1147 		alloc_base = ecache_init_scrub_flush_area(alloc_base);
1148 		PRM_DEBUG(alloc_base);
1149 	}
1150 
1151 	/*
1152 	 * Take the most current snapshot we can by calling mem-update.
1153 	 */
1154 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1155 	    &boot_physavail, &boot_physavail_len,
1156 	    &boot_virtavail, &boot_virtavail_len);
1157 
1158 	/*
1159 	 * Reset npages and memblocks based on boot_physavail list.
1160 	 */
1161 	size_physavail(boot_physavail, boot_physavail_len, &npages, &memblocks);
1162 	PRM_DEBUG(npages);
1163 
1164 	/*
1165 	 * Account for extra memory after e_text.
1166 	 */
1167 	npages += extra_etpg;
1168 
1169 	/*
1170 	 * Calculate the largest free memory chunk in the nucleus data area.
1171 	 * We need to figure out if page structs can fit in there or not.
1172 	 * We also make sure enough page structs get created for any physical
1173 	 * memory we might be returning to the system.
1174 	 */
1175 	ndata_remain_sz = ndata_maxsize(&ndata);
1176 	PRM_DEBUG(ndata_remain_sz);
1177 
1178 	pp_sz = sizeof (struct page) * npages;
1179 
1180 	/*
1181 	 * Here's a nice bit of code based on somewhat recursive logic:
1182 	 *
1183 	 * If the page array would fit within the nucleus, we want to
1184 	 * add npages to cover any extra memory we may be returning back
1185 	 * to the system.
1186 	 *
1187 	 * HOWEVER, the page array is sized by calculating the size of
1188 	 * (struct page * npages), as are the pagehash table, ctrs and
1189 	 * memseg_list, so the very act of performing the calculation below may
1190 	 * in fact make the array large enough that it no longer fits in the
1191 	 * nucleus, meaning there would now be a much larger area of the
1192 	 * nucleus free that should really be added to npages, which would
1193 	 * make the page array that much larger, and so on.
1194 	 *
1195 	 * This also ignores the memory possibly used in the nucleus for the
1196 	 * the page hash, ctrs and memseg list and the fact that whether they
1197 	 * fit there or not varies with the npages calculation below, but we
1198 	 * don't even factor them into the equation at this point; perhaps we
1199 	 * should or perhaps we should just take the approach that the few
1200 	 * extra pages we could add via this calculation REALLY aren't worth
1201 	 * the hassle...
1202 	 */
1203 	if (ndata_remain_sz > pp_sz) {
1204 		size_t spare = ndata_spare(&ndata, pp_sz, ecache_alignsize);
1205 
1206 		npages += mmu_btop(spare);
1207 
1208 		pp_sz = npages * sizeof (struct page);
1209 
1210 		pp_base = ndata_alloc(&ndata, pp_sz, ecache_alignsize);
1211 	}
1212 
1213 	/*
1214 	 * If physmem is patched to be non-zero, use it instead of
1215 	 * the monitor value unless physmem is larger than the total
1216 	 * amount of memory on hand.
1217 	 */
1218 	if (physmem == 0 || physmem > npages)
1219 		physmem = npages;
1220 
1221 	/*
1222 	 * If pp_base is NULL that means the routines above have determined
1223 	 * the page array will not fit in the nucleus; we'll have to
1224 	 * BOP_ALLOC() ourselves some space for them.
1225 	 */
1226 	if (pp_base == NULL) {
1227 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1228 		    alloc_alignsize);
1229 
1230 		alloc_sz = roundup(pp_sz, alloc_alignsize);
1231 
1232 		if ((pp_base = (struct page *)BOP_ALLOC(bootops,
1233 		    alloc_base, alloc_sz, alloc_alignsize)) !=
1234 		    (struct page *)alloc_base)
1235 			panic("page alloc failure");
1236 
1237 		alloc_base += alloc_sz;
1238 	}
1239 
1240 	/*
1241 	 * The page structure hash table size is a power of 2
1242 	 * such that the average hash chain length is PAGE_HASHAVELEN.
1243 	 */
1244 	page_hashsz = npages / PAGE_HASHAVELEN;
1245 	page_hashsz = 1 << highbit((ulong_t)page_hashsz);
1246 	pagehash_sz = sizeof (struct page *) * page_hashsz;
1247 
1248 	/*
1249 	 * We want to TRY to fit the page structure hash table,
1250 	 * the page size free list counters, the memseg list and
1251 	 * and the kpm page space in the nucleus if possible.
1252 	 *
1253 	 * alloc_sz counts how much memory needs to be allocated by
1254 	 * BOP_ALLOC().
1255 	 */
1256 	page_hash = ndata_alloc(&ndata, pagehash_sz, ecache_alignsize);
1257 
1258 	alloc_sz = (page_hash == NULL ? pagehash_sz : 0);
1259 
1260 	/*
1261 	 * Size up per page size free list counters.
1262 	 */
1263 	ctrs_sz = page_ctrs_sz();
1264 	ctrs_base = ndata_alloc(&ndata, ctrs_sz, ecache_alignsize);
1265 
1266 	if (ctrs_base == NULL)
1267 		alloc_sz = roundup(alloc_sz, ecache_alignsize) + ctrs_sz;
1268 
1269 	/*
1270 	 * The memseg list is for the chunks of physical memory that
1271 	 * will be managed by the vm system.  The number calculated is
1272 	 * a guess as boot may fragment it more when memory allocations
1273 	 * are made before kphysm_init().  Currently, there are two
1274 	 * allocations before then, so we assume each causes fragmen-
1275 	 * tation, and add a couple more for good measure.
1276 	 */
1277 	memseg_sz = sizeof (struct memseg) * (memblocks + 4);
1278 	memseg_base = ndata_alloc(&ndata, memseg_sz, ecache_alignsize);
1279 
1280 	if (memseg_base == NULL)
1281 		alloc_sz = roundup(alloc_sz, ecache_alignsize) + memseg_sz;
1282 
1283 
1284 	if (kpm_enable) {
1285 		/*
1286 		 * kpm page space -- Update kpm_npages and make the
1287 		 * same assumption about fragmenting as it is done
1288 		 * for memseg_sz above.
1289 		 */
1290 		kpm_npages_setup(memblocks + 4);
1291 		kpm_pp_sz = (kpm_smallpages == 0) ?
1292 				kpm_npages * sizeof (kpm_page_t):
1293 				kpm_npages * sizeof (kpm_spage_t);
1294 
1295 		kpm_pp_base = (uintptr_t)ndata_alloc(&ndata, kpm_pp_sz,
1296 		    ecache_alignsize);
1297 
1298 		if (kpm_pp_base == NULL)
1299 			alloc_sz = roundup(alloc_sz, ecache_alignsize) +
1300 			    kpm_pp_sz;
1301 	}
1302 
1303 	if (alloc_sz > 0) {
1304 		uintptr_t bop_base;
1305 
1306 		/*
1307 		 * We need extra memory allocated through BOP_ALLOC.
1308 		 */
1309 		alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1310 		    alloc_alignsize);
1311 
1312 		alloc_sz = roundup(alloc_sz, alloc_alignsize);
1313 
1314 		if ((bop_base = (uintptr_t)BOP_ALLOC(bootops, alloc_base,
1315 		    alloc_sz, alloc_alignsize)) != (uintptr_t)alloc_base)
1316 			panic("system page struct alloc failure");
1317 
1318 		alloc_base += alloc_sz;
1319 
1320 		if (page_hash == NULL) {
1321 			page_hash = (struct page **)bop_base;
1322 			bop_base = roundup(bop_base + pagehash_sz,
1323 			    ecache_alignsize);
1324 		}
1325 
1326 		if (ctrs_base == NULL) {
1327 			ctrs_base = (caddr_t)bop_base;
1328 			bop_base = roundup(bop_base + ctrs_sz,
1329 			    ecache_alignsize);
1330 		}
1331 
1332 		if (memseg_base == NULL) {
1333 			memseg_base = (struct memseg *)bop_base;
1334 			bop_base = roundup(bop_base + memseg_sz,
1335 			    ecache_alignsize);
1336 		}
1337 
1338 		if (kpm_enable && kpm_pp_base == NULL) {
1339 			kpm_pp_base = (uintptr_t)bop_base;
1340 			bop_base = roundup(bop_base + kpm_pp_sz,
1341 			    ecache_alignsize);
1342 		}
1343 
1344 		ASSERT(bop_base <= (uintptr_t)alloc_base);
1345 	}
1346 
1347 	/*
1348 	 * Initialize per page size free list counters.
1349 	 */
1350 	ctrs_end = page_ctrs_alloc(ctrs_base);
1351 	ASSERT(ctrs_base + ctrs_sz >= ctrs_end);
1352 
1353 	PRM_DEBUG(page_hash);
1354 	PRM_DEBUG(memseg_base);
1355 	PRM_DEBUG(kpm_pp_base);
1356 	PRM_DEBUG(kpm_pp_sz);
1357 	PRM_DEBUG(pp_base);
1358 	PRM_DEBUG(pp_sz);
1359 	PRM_DEBUG(alloc_base);
1360 
1361 #ifdef	TRAPTRACE
1362 	/*
1363 	 * Allocate trap trace buffer last so as not to affect
1364 	 * the 4M alignments of the allocations above on V9 SPARCs...
1365 	 */
1366 	alloc_base = trap_trace_alloc(alloc_base);
1367 	PRM_DEBUG(alloc_base);
1368 #endif	/* TRAPTRACE */
1369 
1370 	if (kmem64_base) {
1371 		/*
1372 		 * Set the end of the kmem64 segment for V9 SPARCs, if
1373 		 * appropriate...
1374 		 */
1375 		kmem64_end = (caddr_t)roundup((uintptr_t)alloc_base,
1376 		    alloc_alignsize);
1377 
1378 		PRM_DEBUG(kmem64_base);
1379 		PRM_DEBUG(kmem64_end);
1380 	}
1381 
1382 	/*
1383 	 * Allocate space for the interrupt vector table.
1384 	 */
1385 	memspace = (caddr_t)BOP_ALLOC(bootops, (caddr_t)intr_vector,
1386 	    IVSIZE, MMU_PAGESIZE);
1387 	if (memspace != (caddr_t)intr_vector)
1388 		panic("interrupt table allocation failure");
1389 
1390 	/*
1391 	 * The memory lists from boot are allocated from the heap arena
1392 	 * so that later they can be freed and/or reallocated.
1393 	 */
1394 	if (BOP_GETPROP(bootops, "extent", &memlist_sz) == -1)
1395 		panic("could not retrieve property \"extent\"");
1396 
1397 	/*
1398 	 * Between now and when we finish copying in the memory lists,
1399 	 * allocations happen so the space gets fragmented and the
1400 	 * lists longer.  Leave enough space for lists twice as long
1401 	 * as what boot says it has now; roundup to a pagesize.
1402 	 * Also add space for the final phys-avail copy in the fixup
1403 	 * routine.
1404 	 */
1405 	va = (caddr_t)(sysbase + PAGESIZE + PANICBUFSIZE +
1406 	    roundup(IVSIZE, MMU_PAGESIZE));
1407 	memlist_sz *= 4;
1408 	memlist_sz = roundup(memlist_sz, MMU_PAGESIZE);
1409 	memspace = (caddr_t)BOP_ALLOC(bootops, va, memlist_sz, BO_NO_ALIGN);
1410 	if (memspace == NULL)
1411 		halt("Boot allocation failed.");
1412 
1413 	memlist = (struct memlist *)memspace;
1414 	memlist_end = (char *)memspace + memlist_sz;
1415 
1416 	PRM_DEBUG(memlist);
1417 	PRM_DEBUG(memlist_end);
1418 	PRM_DEBUG(sysbase);
1419 	PRM_DEBUG(syslimit);
1420 
1421 	kernelheap_init((void *)sysbase, (void *)syslimit,
1422 	    (caddr_t)sysbase + PAGESIZE, NULL, NULL);
1423 
1424 	/*
1425 	 * Take the most current snapshot we can by calling mem-update.
1426 	 */
1427 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1428 	    &boot_physavail, &boot_physavail_len,
1429 	    &boot_virtavail, &boot_virtavail_len);
1430 
1431 	/*
1432 	 * Remove the space used by BOP_ALLOC from the kernel heap
1433 	 * plus the area actually used by the OBP (if any)
1434 	 * ignoring virtual addresses in virt_avail, above syslimit.
1435 	 */
1436 	virt_avail = memlist;
1437 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1438 
1439 	for (cur = virt_avail; cur->next; cur = cur->next) {
1440 		uint64_t range_base, range_size;
1441 
1442 		if ((range_base = cur->address + cur->size) < (uint64_t)sysbase)
1443 			continue;
1444 		if (range_base >= (uint64_t)syslimit)
1445 			break;
1446 		/*
1447 		 * Limit the range to end at syslimit.
1448 		 */
1449 		range_size = MIN(cur->next->address,
1450 		    (uint64_t)syslimit) - range_base;
1451 		(void) vmem_xalloc(heap_arena, (size_t)range_size, PAGESIZE,
1452 		    0, 0, (void *)range_base, (void *)(range_base + range_size),
1453 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1454 	}
1455 
1456 	phys_avail = memlist;
1457 	(void) copy_physavail(boot_physavail, boot_physavail_len,
1458 	    &memlist, 0, 0);
1459 
1460 	/*
1461 	 * Add any extra memory after e_text to the phys_avail list, as long
1462 	 * as there's at least a page to add.
1463 	 */
1464 	if (extra_etpg)
1465 		memlist_add(va_to_pa(extra_etva), mmu_ptob(extra_etpg),
1466 		    &memlist, &phys_avail);
1467 
1468 	/*
1469 	 * Add any extra memory after e_data to the phys_avail list as long
1470 	 * as there's at least a page to add.  Usually, there isn't any,
1471 	 * since extra HME blocks typically get allocated there first before
1472 	 * using RAM elsewhere.
1473 	 */
1474 	if ((nalloc_base = ndata_extra_base(&ndata, MMU_PAGESIZE)) == NULL)
1475 		nalloc_base = nalloc_end;
1476 	ndata_remain_sz = nalloc_end - nalloc_base;
1477 
1478 	if (ndata_remain_sz >= MMU_PAGESIZE)
1479 		memlist_add(va_to_pa(nalloc_base),
1480 		    (uint64_t)ndata_remain_sz, &memlist, &phys_avail);
1481 
1482 	PRM_DEBUG(memlist);
1483 	PRM_DEBUG(memlist_sz);
1484 	PRM_DEBUG(memspace);
1485 
1486 	if ((caddr_t)memlist > (memspace + memlist_sz))
1487 		panic("memlist overflow");
1488 
1489 	PRM_DEBUG(pp_base);
1490 	PRM_DEBUG(memseg_base);
1491 	PRM_DEBUG(npages);
1492 
1493 	/*
1494 	 * Initialize the page structures from the memory lists.
1495 	 */
1496 	kphysm_init(pp_base, memseg_base, npages, kpm_pp_base, kpm_npages);
1497 
1498 	availrmem_initial = availrmem = freemem;
1499 	PRM_DEBUG(availrmem);
1500 
1501 	/*
1502 	 * Some of the locks depend on page_hashsz being set!
1503 	 * kmem_init() depends on this; so, keep it here.
1504 	 */
1505 	page_lock_init();
1506 
1507 	/*
1508 	 * Initialize kernel memory allocator.
1509 	 */
1510 	kmem_init();
1511 
1512 	/*
1513 	 * Initialize bp_mapin().
1514 	 */
1515 	bp_init(shm_alignment, HAT_STRICTORDER);
1516 
1517 	/*
1518 	 * Reserve space for panicbuf and intr_vector from the 32-bit heap
1519 	 */
1520 	(void) vmem_xalloc(heap32_arena, PANICBUFSIZE, PAGESIZE, 0, 0,
1521 	    panicbuf, panicbuf + PANICBUFSIZE,
1522 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1523 
1524 	(void) vmem_xalloc(heap32_arena, IVSIZE, PAGESIZE, 0, 0,
1525 	    intr_vector, (caddr_t)intr_vector + IVSIZE,
1526 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1527 
1528 	mem_config_init();
1529 }
1530 
1531 static void
1532 startup_modules(void)
1533 {
1534 	int proplen, nhblk1, nhblk8;
1535 	size_t  nhblksz;
1536 	pgcnt_t hblk_pages, pages_per_hblk;
1537 	size_t hme8blk_sz, hme1blk_sz;
1538 
1539 	/*
1540 	 * Log any optional messages from the boot program
1541 	 */
1542 	proplen = (size_t)BOP_GETPROPLEN(bootops, "boot-message");
1543 	if (proplen > 0) {
1544 		char *msg;
1545 		size_t len = (size_t)proplen;
1546 
1547 		msg = kmem_zalloc(len, KM_SLEEP);
1548 		(void) BOP_GETPROP(bootops, "boot-message", msg);
1549 		cmn_err(CE_CONT, "?%s\n", msg);
1550 		kmem_free(msg, len);
1551 	}
1552 
1553 	/*
1554 	 * Let the platforms have a chance to change default
1555 	 * values before reading system file.
1556 	 */
1557 	if (&set_platform_defaults)
1558 		set_platform_defaults();
1559 
1560 	/*
1561 	 * Calculate default settings of system parameters based upon
1562 	 * maxusers, yet allow to be overridden via the /etc/system file.
1563 	 */
1564 	param_calc(0);
1565 
1566 	mod_setup();
1567 
1568 	/*
1569 	 * If this is a positron, complain and halt.
1570 	 */
1571 	if (&iam_positron && iam_positron()) {
1572 		cmn_err(CE_WARN, "This hardware platform is not supported"
1573 		    " by this release of Solaris.\n");
1574 #ifdef DEBUG
1575 		prom_enter_mon();	/* Type 'go' to resume */
1576 		cmn_err(CE_WARN, "Booting an unsupported platform.\n");
1577 		cmn_err(CE_WARN, "Booting with down-rev firmware.\n");
1578 
1579 #else /* DEBUG */
1580 		halt(0);
1581 #endif /* DEBUG */
1582 	}
1583 
1584 	/*
1585 	 * If we are running firmware that isn't 64-bit ready
1586 	 * then complain and halt.
1587 	 */
1588 	do_prom_version_check();
1589 
1590 	/*
1591 	 * Initialize system parameters
1592 	 */
1593 	param_init();
1594 
1595 	/*
1596 	 * maxmem is the amount of physical memory we're playing with.
1597 	 */
1598 	maxmem = physmem;
1599 
1600 	/* Set segkp limits. */
1601 	ncbase = (caddr_t)SEGDEBUGBASE;
1602 	ncend = (caddr_t)SEGDEBUGBASE;
1603 
1604 	/*
1605 	 * Initialize the hat layer.
1606 	 */
1607 	hat_init();
1608 
1609 	/*
1610 	 * Initialize segment management stuff.
1611 	 */
1612 	seg_init();
1613 
1614 	/*
1615 	 * Create the va>tte handler, so the prom can understand
1616 	 * kernel translations.  The handler is installed later, just
1617 	 * as we are about to take over the trap table from the prom.
1618 	 */
1619 	create_va_to_tte();
1620 
1621 	/*
1622 	 * Load the forthdebugger (optional)
1623 	 */
1624 	forthdebug_init();
1625 
1626 	/*
1627 	 * Create OBP node for console input callbacks
1628 	 * if it is needed.
1629 	 */
1630 	startup_create_io_node();
1631 
1632 	if (modloadonly("fs", "specfs") == -1)
1633 		halt("Can't load specfs");
1634 
1635 	if (modloadonly("fs", "devfs") == -1)
1636 		halt("Can't load devfs");
1637 
1638 	if (modloadonly("misc", "swapgeneric") == -1)
1639 		halt("Can't load swapgeneric");
1640 
1641 	dispinit();
1642 
1643 	/*
1644 	 * Infer meanings to the members of the idprom buffer.
1645 	 */
1646 	parse_idprom();
1647 
1648 	/* Read cluster configuration data. */
1649 	clconf_init();
1650 
1651 	setup_ddi();
1652 
1653 	/*
1654 	 * Lets take this opportunity to load the root device.
1655 	 */
1656 	if (loadrootmodules() != 0)
1657 		debug_enter("Can't load the root filesystem");
1658 
1659 	/*
1660 	 * Load tod driver module for the tod part found on this system.
1661 	 * Recompute the cpu frequency/delays based on tod as tod part
1662 	 * tends to keep time more accurately.
1663 	 */
1664 	if (&load_tod_module)
1665 		load_tod_module();
1666 
1667 	/*
1668 	 * Allow platforms to load modules which might
1669 	 * be needed after bootops are gone.
1670 	 */
1671 	if (&load_platform_modules)
1672 		load_platform_modules();
1673 
1674 	setcpudelay();
1675 
1676 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1677 	    &boot_physavail, &boot_physavail_len,
1678 	    &boot_virtavail, &boot_virtavail_len);
1679 
1680 	bop_alloc_pages = size_virtalloc(boot_virtavail, boot_virtavail_len);
1681 
1682 	/*
1683 	 * Calculation and allocation of hmeblks needed to remap
1684 	 * the memory allocated by PROM till now:
1685 	 *
1686 	 * (1)  calculate how much virtual memory has been bop_alloc'ed.
1687 	 * (2)  roundup this memory to span of hme8blk, i.e. 64KB
1688 	 * (3)  calculate number of hme8blk's needed to remap this memory
1689 	 * (4)  calculate amount of memory that's consumed by these hme8blk's
1690 	 * (5)  add memory calculated in steps (2) and (4) above.
1691 	 * (6)  roundup this memory to span of hme8blk, i.e. 64KB
1692 	 * (7)  calculate number of hme8blk's needed to remap this memory
1693 	 * (8)  calculate amount of memory that's consumed by these hme8blk's
1694 	 * (9)  allocate additional hme1blk's to hold large mappings.
1695 	 *	H8TOH1 determines this.  The current SWAG gives enough hblk1's
1696 	 *	to remap everything with 4M mappings.
1697 	 * (10) account for partially used hblk8's due to non-64K aligned
1698 	 *	PROM mapping entries.
1699 	 * (11) add memory calculated in steps (8), (9), and (10) above.
1700 	 * (12) kmem_zalloc the memory calculated in (11); since segkmem
1701 	 *	is not ready yet, this gets bop_alloc'ed.
1702 	 * (13) there will be very few bop_alloc's after this point before
1703 	 *	trap table takes over
1704 	 */
1705 
1706 	/* sfmmu_init_nucleus_hblks expects properly aligned data structures. */
1707 	hme8blk_sz = roundup(HME8BLK_SZ, sizeof (int64_t));
1708 	hme1blk_sz = roundup(HME1BLK_SZ, sizeof (int64_t));
1709 
1710 	pages_per_hblk = btop(HMEBLK_SPAN(TTE8K));
1711 	bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1712 	nhblk8 = bop_alloc_pages / pages_per_hblk;
1713 	nhblk1 = roundup(nhblk8, H8TOH1) / H8TOH1;
1714 	hblk_pages = btopr(nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz);
1715 	bop_alloc_pages += hblk_pages;
1716 	bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1717 	nhblk8 = bop_alloc_pages / pages_per_hblk;
1718 	nhblk1 = roundup(nhblk8, H8TOH1) / H8TOH1;
1719 	if (nhblk1 < hblk1_min)
1720 		nhblk1 = hblk1_min;
1721 	if (nhblk8 < hblk8_min)
1722 		nhblk8 = hblk8_min;
1723 
1724 	/*
1725 	 * Since hblk8's can hold up to 64k of mappings aligned on a 64k
1726 	 * boundary, the number of hblk8's needed to map the entries in the
1727 	 * boot_virtavail list needs to be adjusted to take this into
1728 	 * consideration.  Thus, we need to add additional hblk8's since it
1729 	 * is possible that an hblk8 will not have all 8 slots used due to
1730 	 * alignment constraints.  Since there were boot_virtavail_len entries
1731 	 * in that list, we need to add that many hblk8's to the number
1732 	 * already calculated to make sure we don't underestimate.
1733 	 */
1734 	nhblk8 += boot_virtavail_len;
1735 	nhblksz = nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz;
1736 
1737 	/* Allocate in pagesize chunks */
1738 	nhblksz = roundup(nhblksz, MMU_PAGESIZE);
1739 	hblk_base = kmem_zalloc(nhblksz, KM_SLEEP);
1740 	sfmmu_init_nucleus_hblks(hblk_base, nhblksz, nhblk8, nhblk1);
1741 }
1742 
1743 static void
1744 startup_bop_gone(void)
1745 {
1746 	extern int bop_io_quiesced;
1747 
1748 	/*
1749 	 * Call back into boot and release boots resources.
1750 	 */
1751 	BOP_QUIESCE_IO(bootops);
1752 	bop_io_quiesced = 1;
1753 
1754 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1755 	    &boot_physavail, &boot_physavail_len,
1756 	    &boot_virtavail, &boot_virtavail_len);
1757 	/*
1758 	 * Copy physinstalled list into kernel space.
1759 	 */
1760 	phys_install = memlist;
1761 	copy_memlist(boot_physinstalled, boot_physinstalled_len, &memlist);
1762 
1763 	/*
1764 	 * setup physically contiguous area twice as large as the ecache.
1765 	 * this is used while doing displacement flush of ecaches
1766 	 */
1767 	if (&ecache_flush_address) {
1768 		ecache_flushaddr = ecache_flush_address();
1769 		if (ecache_flushaddr == (uint64_t)-1) {
1770 			cmn_err(CE_PANIC,
1771 			    "startup: no memory to set ecache_flushaddr");
1772 		}
1773 	}
1774 
1775 	/*
1776 	 * Virtual available next.
1777 	 */
1778 	ASSERT(virt_avail != NULL);
1779 	memlist_free_list(virt_avail);
1780 	virt_avail = memlist;
1781 	copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1782 
1783 	/*
1784 	 * Last chance to ask our booter questions ..
1785 	 */
1786 }
1787 
1788 
1789 /*
1790  * startup_fixup_physavail - called from mach_sfmmu.c after the final
1791  * allocations have been performed.  We can't call it in startup_bop_gone
1792  * since later operations can cause obp to allocate more memory.
1793  */
1794 void
1795 startup_fixup_physavail(void)
1796 {
1797 	struct memlist *cur;
1798 
1799 	/*
1800 	 * take the most current snapshot we can by calling mem-update
1801 	 */
1802 	copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1803 	    &boot_physavail, &boot_physavail_len,
1804 	    &boot_virtavail, &boot_virtavail_len);
1805 
1806 	/*
1807 	 * Copy phys_avail list, again.
1808 	 * Both the kernel/boot and the prom have been allocating
1809 	 * from the original list we copied earlier.
1810 	 */
1811 	cur = memlist;
1812 	(void) copy_physavail(boot_physavail, boot_physavail_len,
1813 	    &memlist, 0, 0);
1814 
1815 	/*
1816 	 * Add any extra memory after e_text we added to the phys_avail list
1817 	 * back to the old list.
1818 	 */
1819 	if (extra_etpg)
1820 		memlist_add(va_to_pa(extra_etva), mmu_ptob(extra_etpg),
1821 		    &memlist, &cur);
1822 	if (ndata_remain_sz >= MMU_PAGESIZE)
1823 		memlist_add(va_to_pa(nalloc_base),
1824 		    (uint64_t)ndata_remain_sz, &memlist, &cur);
1825 
1826 	/*
1827 	 * There isn't any bounds checking on the memlist area
1828 	 * so ensure it hasn't overgrown.
1829 	 */
1830 	if ((caddr_t)memlist > (caddr_t)memlist_end)
1831 		cmn_err(CE_PANIC, "startup: memlist size exceeded");
1832 
1833 	/*
1834 	 * The kernel removes the pages that were allocated for it from
1835 	 * the freelist, but we now have to find any -extra- pages that
1836 	 * the prom has allocated for it's own book-keeping, and remove
1837 	 * them from the freelist too. sigh.
1838 	 */
1839 	fix_prom_pages(phys_avail, cur);
1840 
1841 	ASSERT(phys_avail != NULL);
1842 	memlist_free_list(phys_avail);
1843 	phys_avail = cur;
1844 
1845 	/*
1846 	 * We're done with boot.  Just after this point in time, boot
1847 	 * gets unmapped, so we can no longer rely on its services.
1848 	 * Zero the bootops to indicate this fact.
1849 	 */
1850 	bootops = (struct bootops *)NULL;
1851 	BOOTOPS_GONE();
1852 }
1853 
1854 static void
1855 startup_vm(void)
1856 {
1857 	size_t	i;
1858 	struct segmap_crargs a;
1859 	struct segkpm_crargs b;
1860 
1861 	uint64_t avmem;
1862 	caddr_t va;
1863 	pgcnt_t	max_phys_segkp;
1864 	int	mnode;
1865 
1866 	extern int exec_lpg_disable, use_brk_lpg, use_stk_lpg, use_zmap_lpg;
1867 
1868 	/*
1869 	 * get prom's mappings, create hments for them and switch
1870 	 * to the kernel context.
1871 	 */
1872 	hat_kern_setup();
1873 
1874 	/*
1875 	 * Take over trap table
1876 	 */
1877 	setup_trap_table();
1878 
1879 	/*
1880 	 * Install the va>tte handler, so that the prom can handle
1881 	 * misses and understand the kernel table layout in case
1882 	 * we need call into the prom.
1883 	 */
1884 	install_va_to_tte();
1885 
1886 	/*
1887 	 * Set a flag to indicate that the tba has been taken over.
1888 	 */
1889 	tba_taken_over = 1;
1890 
1891 	/* initialize MMU primary context register */
1892 	mmu_init_kcontext();
1893 
1894 	/*
1895 	 * The boot cpu can now take interrupts, x-calls, x-traps
1896 	 */
1897 	CPUSET_ADD(cpu_ready_set, CPU->cpu_id);
1898 	CPU->cpu_flags |= (CPU_READY | CPU_ENABLE | CPU_EXISTS);
1899 
1900 	/*
1901 	 * Set a flag to tell write_scb_int() that it can access V_TBR_WR_ADDR.
1902 	 */
1903 	tbr_wr_addr_inited = 1;
1904 
1905 	/*
1906 	 * Initialize VM system, and map kernel address space.
1907 	 */
1908 	kvm_init();
1909 
1910 	/*
1911 	 * XXX4U: previously, we initialized and turned on
1912 	 * the caches at this point. But of course we have
1913 	 * nothing to do, as the prom has already done this
1914 	 * for us -- main memory must be E$able at all times.
1915 	 */
1916 
1917 	/*
1918 	 * If the following is true, someone has patched
1919 	 * phsymem to be less than the number of pages that
1920 	 * the system actually has.  Remove pages until system
1921 	 * memory is limited to the requested amount.  Since we
1922 	 * have allocated page structures for all pages, we
1923 	 * correct the amount of memory we want to remove
1924 	 * by the size of the memory used to hold page structures
1925 	 * for the non-used pages.
1926 	 */
1927 	if (physmem < npages) {
1928 		pgcnt_t diff, off;
1929 		struct page *pp;
1930 		struct seg kseg;
1931 
1932 		cmn_err(CE_WARN, "limiting physmem to %ld pages", physmem);
1933 
1934 		off = 0;
1935 		diff = npages - physmem;
1936 		diff -= mmu_btopr(diff * sizeof (struct page));
1937 		kseg.s_as = &kas;
1938 		while (diff--) {
1939 			pp = page_create_va(&unused_pages_vp, (offset_t)off,
1940 			    MMU_PAGESIZE, PG_WAIT | PG_EXCL,
1941 			    &kseg, (caddr_t)off);
1942 			if (pp == NULL)
1943 				cmn_err(CE_PANIC, "limited physmem too much!");
1944 			page_io_unlock(pp);
1945 			page_downgrade(pp);
1946 			availrmem--;
1947 			off += MMU_PAGESIZE;
1948 		}
1949 	}
1950 
1951 	/*
1952 	 * When printing memory, show the total as physmem less
1953 	 * that stolen by a debugger.
1954 	 */
1955 	cmn_err(CE_CONT, "?mem = %ldK (0x%lx000)\n",
1956 	    (ulong_t)(physinstalled) << (PAGESHIFT - 10),
1957 	    (ulong_t)(physinstalled) << (PAGESHIFT - 12));
1958 
1959 	avmem = (uint64_t)freemem << PAGESHIFT;
1960 	cmn_err(CE_CONT, "?avail mem = %lld\n", (unsigned long long)avmem);
1961 
1962 	/* For small memory systems disable automatic large pages. */
1963 	if (physmem < auto_lpg_min_physmem) {
1964 		exec_lpg_disable = 1;
1965 		use_brk_lpg = 0;
1966 		use_stk_lpg = 0;
1967 		use_zmap_lpg = 0;
1968 	}
1969 
1970 	/*
1971 	 * Perform platform specific freelist processing
1972 	 */
1973 	if (&plat_freelist_process) {
1974 		for (mnode = 0; mnode < max_mem_nodes; mnode++)
1975 			if (mem_node_config[mnode].exists)
1976 				plat_freelist_process(mnode);
1977 	}
1978 
1979 	/*
1980 	 * Initialize the segkp segment type.  We position it
1981 	 * after the configured tables and buffers (whose end
1982 	 * is given by econtig) and before V_WKBASE_ADDR.
1983 	 * Also in this area is segkmap (size SEGMAPSIZE).
1984 	 */
1985 
1986 	/* XXX - cache alignment? */
1987 	va = (caddr_t)SEGKPBASE;
1988 	ASSERT(((uintptr_t)va & PAGEOFFSET) == 0);
1989 
1990 	max_phys_segkp = (physmem * 2);
1991 
1992 	if (segkpsize < btop(SEGKPMINSIZE) || segkpsize > btop(SEGKPMAXSIZE)) {
1993 		segkpsize = btop(SEGKPDEFSIZE);
1994 		cmn_err(CE_WARN, "Illegal value for segkpsize. "
1995 		    "segkpsize has been reset to %ld pages", segkpsize);
1996 	}
1997 
1998 	i = ptob(MIN(segkpsize, max_phys_segkp));
1999 
2000 	rw_enter(&kas.a_lock, RW_WRITER);
2001 	if (seg_attach(&kas, va, i, segkp) < 0)
2002 		cmn_err(CE_PANIC, "startup: cannot attach segkp");
2003 	if (segkp_create(segkp) != 0)
2004 		cmn_err(CE_PANIC, "startup: segkp_create failed");
2005 	rw_exit(&kas.a_lock);
2006 
2007 	/*
2008 	 * kpm segment
2009 	 */
2010 	segmap_kpm = kpm_enable &&
2011 		segmap_kpm && PAGESIZE == MAXBSIZE;
2012 
2013 	if (kpm_enable) {
2014 		rw_enter(&kas.a_lock, RW_WRITER);
2015 
2016 		/*
2017 		 * The segkpm virtual range range is larger than the
2018 		 * actual physical memory size and also covers gaps in
2019 		 * the physical address range for the following reasons:
2020 		 * . keep conversion between segkpm and physical addresses
2021 		 *   simple, cheap and unambiguous.
2022 		 * . avoid extension/shrink of the the segkpm in case of DR.
2023 		 * . avoid complexity for handling of virtual addressed
2024 		 *   caches, segkpm and the regular mapping scheme must be
2025 		 *   kept in sync wrt. the virtual color of mapped pages.
2026 		 * Any accesses to virtual segkpm ranges not backed by
2027 		 * physical memory will fall through the memseg pfn hash
2028 		 * and will be handled in segkpm_fault.
2029 		 * Additional kpm_size spaces needed for vac alias prevention.
2030 		 */
2031 		if (seg_attach(&kas, kpm_vbase, kpm_size * vac_colors,
2032 		    segkpm) < 0)
2033 			cmn_err(CE_PANIC, "cannot attach segkpm");
2034 
2035 		b.prot = PROT_READ | PROT_WRITE;
2036 		b.nvcolors = shm_alignment >> MMU_PAGESHIFT;
2037 
2038 		if (segkpm_create(segkpm, (caddr_t)&b) != 0)
2039 			panic("segkpm_create segkpm");
2040 
2041 		rw_exit(&kas.a_lock);
2042 	}
2043 
2044 	/*
2045 	 * Now create generic mapping segment.  This mapping
2046 	 * goes SEGMAPSIZE beyond SEGMAPBASE.  But if the total
2047 	 * virtual address is greater than the amount of free
2048 	 * memory that is available, then we trim back the
2049 	 * segment size to that amount
2050 	 */
2051 	va = (caddr_t)SEGMAPBASE;
2052 
2053 	/*
2054 	 * 1201049: segkmap base address must be MAXBSIZE aligned
2055 	 */
2056 	ASSERT(((uintptr_t)va & MAXBOFFSET) == 0);
2057 
2058 	/*
2059 	 * Set size of segmap to percentage of freemem at boot,
2060 	 * but stay within the allowable range
2061 	 * Note we take percentage  before converting from pages
2062 	 * to bytes to avoid an overflow on 32-bit kernels.
2063 	 */
2064 	i = mmu_ptob((freemem * segmap_percent) / 100);
2065 
2066 	if (i < MINMAPSIZE)
2067 		i = MINMAPSIZE;
2068 
2069 	if (i > MIN(SEGMAPSIZE, mmu_ptob(freemem)))
2070 		i = MIN(SEGMAPSIZE, mmu_ptob(freemem));
2071 
2072 	i &= MAXBMASK;	/* 1201049: segkmap size must be MAXBSIZE aligned */
2073 
2074 	rw_enter(&kas.a_lock, RW_WRITER);
2075 	if (seg_attach(&kas, va, i, segkmap) < 0)
2076 		cmn_err(CE_PANIC, "cannot attach segkmap");
2077 
2078 	a.prot = PROT_READ | PROT_WRITE;
2079 	a.shmsize = shm_alignment;
2080 	a.nfreelist = 0;	/* use segmap driver defaults */
2081 
2082 	if (segmap_create(segkmap, (caddr_t)&a) != 0)
2083 		panic("segmap_create segkmap");
2084 	rw_exit(&kas.a_lock);
2085 
2086 	segdev_init();
2087 }
2088 
2089 static void
2090 startup_end(void)
2091 {
2092 	if ((caddr_t)memlist > (caddr_t)memlist_end)
2093 		panic("memlist overflow 2");
2094 	memlist_free_block((caddr_t)memlist,
2095 	    ((caddr_t)memlist_end - (caddr_t)memlist));
2096 	memlist = NULL;
2097 
2098 	/* enable page_relocation since OBP is now done */
2099 	page_relocate_ready = 1;
2100 
2101 	/*
2102 	 * Perform tasks that get done after most of the VM
2103 	 * initialization has been done but before the clock
2104 	 * and other devices get started.
2105 	 */
2106 	kern_setup1();
2107 
2108 	/*
2109 	 * Intialize the VM arenas for allocating physically
2110 	 * contiguus memory chunk for interrupt queues snd
2111 	 * allocate/register boot cpu's queues, if any and
2112 	 * allocate dump buffer for sun4v systems to store
2113 	 * extra crash information during crash dump
2114 	 */
2115 	contig_mem_init();
2116 	mach_descrip_init();
2117 	cpu_intrq_setup(CPU);
2118 	cpu_intrq_register(CPU);
2119 	mach_htraptrace_init();
2120 	mach_htraptrace_setup(CPU->cpu_id);
2121 	mach_htraptrace_configure(CPU->cpu_id);
2122 	mach_dump_buffer_init();
2123 
2124 	/*
2125 	 * Initialize interrupt related stuff
2126 	 */
2127 	cpu_intr_alloc(CPU, NINTR_THREADS);
2128 
2129 	(void) splzs();			/* allow hi clock ints but not zs */
2130 
2131 	/*
2132 	 * Initialize errors.
2133 	 */
2134 	error_init();
2135 
2136 	/*
2137 	 * Note that we may have already used kernel bcopy before this
2138 	 * point - but if you really care about this, adb the use_hw_*
2139 	 * variables to 0 before rebooting.
2140 	 */
2141 	mach_hw_copy_limit();
2142 
2143 	/*
2144 	 * Install the "real" preemption guards before DDI services
2145 	 * are available.
2146 	 */
2147 	(void) prom_set_preprom(kern_preprom);
2148 	(void) prom_set_postprom(kern_postprom);
2149 	CPU->cpu_m.mutex_ready = 1;
2150 
2151 	/*
2152 	 * Initialize segnf (kernel support for non-faulting loads).
2153 	 */
2154 	segnf_init();
2155 
2156 	/*
2157 	 * Configure the root devinfo node.
2158 	 */
2159 	configure();		/* set up devices */
2160 	mach_cpu_halt_idle();
2161 }
2162 
2163 
2164 void
2165 post_startup(void)
2166 {
2167 #ifdef	PTL1_PANIC_DEBUG
2168 	extern void init_ptl1_thread(void);
2169 #endif	/* PTL1_PANIC_DEBUG */
2170 	extern void abort_sequence_init(void);
2171 
2172 	/*
2173 	 * Set the system wide, processor-specific flags to be passed
2174 	 * to userland via the aux vector for performance hints and
2175 	 * instruction set extensions.
2176 	 */
2177 	bind_hwcap();
2178 
2179 	/*
2180 	 * Startup memory scrubber (if any)
2181 	 */
2182 	mach_memscrub();
2183 
2184 	/*
2185 	 * Allocate soft interrupt to handle abort sequence.
2186 	 */
2187 	abort_sequence_init();
2188 
2189 	/*
2190 	 * Configure the rest of the system.
2191 	 * Perform forceloading tasks for /etc/system.
2192 	 */
2193 	(void) mod_sysctl(SYS_FORCELOAD, NULL);
2194 	/*
2195 	 * ON4.0: Force /proc module in until clock interrupt handle fixed
2196 	 * ON4.0: This must be fixed or restated in /etc/systems.
2197 	 */
2198 	(void) modload("fs", "procfs");
2199 
2200 	if (&load_platform_drivers)
2201 		load_platform_drivers();
2202 
2203 	/* load vis simulation module, if we are running w/fpu off */
2204 	if (!fpu_exists) {
2205 		if (modload("misc", "vis") == -1)
2206 			halt("Can't load vis");
2207 	}
2208 
2209 	mach_fpras();
2210 
2211 	maxmem = freemem;
2212 
2213 #ifdef	PTL1_PANIC_DEBUG
2214 	init_ptl1_thread();
2215 #endif	/* PTL1_PANIC_DEBUG */
2216 }
2217 
2218 #ifdef	PTL1_PANIC_DEBUG
2219 int		ptl1_panic_test = 0;
2220 int		ptl1_panic_xc_one_test = 0;
2221 int		ptl1_panic_xc_all_test = 0;
2222 int		ptl1_panic_xt_one_test = 0;
2223 int		ptl1_panic_xt_all_test = 0;
2224 kthread_id_t	ptl1_thread_p = NULL;
2225 kcondvar_t	ptl1_cv;
2226 kmutex_t	ptl1_mutex;
2227 int		ptl1_recurse_count_threshold = 0x40;
2228 int		ptl1_recurse_trap_threshold = 0x3d;
2229 extern void	ptl1_recurse(int, int);
2230 extern void	ptl1_panic_xt(int, int);
2231 
2232 /*
2233  * Called once per second by timeout() to wake up
2234  * the ptl1_panic thread to see if it should cause
2235  * a trap to the ptl1_panic() code.
2236  */
2237 /* ARGSUSED */
2238 static void
2239 ptl1_wakeup(void *arg)
2240 {
2241 	mutex_enter(&ptl1_mutex);
2242 	cv_signal(&ptl1_cv);
2243 	mutex_exit(&ptl1_mutex);
2244 }
2245 
2246 /*
2247  * ptl1_panic cross call function:
2248  *     Needed because xc_one() and xc_some() can pass
2249  *	64 bit args but ptl1_recurse() expects ints.
2250  */
2251 static void
2252 ptl1_panic_xc(void)
2253 {
2254 	ptl1_recurse(ptl1_recurse_count_threshold,
2255 	    ptl1_recurse_trap_threshold);
2256 }
2257 
2258 /*
2259  * The ptl1 thread waits for a global flag to be set
2260  * and uses the recurse thresholds to set the stack depth
2261  * to cause a ptl1_panic() directly via a call to ptl1_recurse
2262  * or indirectly via the cross call and cross trap functions.
2263  *
2264  * This is useful testing stack overflows and normal
2265  * ptl1_panic() states with a know stack frame.
2266  *
2267  * ptl1_recurse() is an asm function in ptl1_panic.s that
2268  * sets the {In, Local, Out, and Global} registers to a
2269  * know state on the stack and just prior to causing a
2270  * test ptl1_panic trap.
2271  */
2272 static void
2273 ptl1_thread(void)
2274 {
2275 	mutex_enter(&ptl1_mutex);
2276 	while (ptl1_thread_p) {
2277 		cpuset_t	other_cpus;
2278 		int		cpu_id;
2279 		int		my_cpu_id;
2280 		int		target_cpu_id;
2281 		int		target_found;
2282 
2283 		if (ptl1_panic_test) {
2284 			ptl1_recurse(ptl1_recurse_count_threshold,
2285 			    ptl1_recurse_trap_threshold);
2286 		}
2287 
2288 		/*
2289 		 * Find potential targets for x-call and x-trap,
2290 		 * if any exist while preempt is disabled we
2291 		 * start a ptl1_panic if requested via a
2292 		 * globals.
2293 		 */
2294 		kpreempt_disable();
2295 		my_cpu_id = CPU->cpu_id;
2296 		other_cpus = cpu_ready_set;
2297 		CPUSET_DEL(other_cpus, CPU->cpu_id);
2298 		target_found = 0;
2299 		if (!CPUSET_ISNULL(other_cpus)) {
2300 			/*
2301 			 * Pick the first one
2302 			 */
2303 			for (cpu_id = 0; cpu_id < NCPU; cpu_id++) {
2304 				if (cpu_id == my_cpu_id)
2305 					continue;
2306 
2307 				if (CPU_XCALL_READY(cpu_id)) {
2308 					target_cpu_id = cpu_id;
2309 					target_found = 1;
2310 					break;
2311 				}
2312 			}
2313 			ASSERT(target_found);
2314 
2315 			if (ptl1_panic_xc_one_test) {
2316 				xc_one(target_cpu_id,
2317 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2318 			}
2319 			if (ptl1_panic_xc_all_test) {
2320 				xc_some(other_cpus,
2321 				    (xcfunc_t *)ptl1_panic_xc, 0, 0);
2322 			}
2323 			if (ptl1_panic_xt_one_test) {
2324 				xt_one(target_cpu_id,
2325 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2326 			}
2327 			if (ptl1_panic_xt_all_test) {
2328 				xt_some(other_cpus,
2329 				    (xcfunc_t *)ptl1_panic_xt, 0, 0);
2330 			}
2331 		}
2332 		kpreempt_enable();
2333 		(void) timeout(ptl1_wakeup, NULL, hz);
2334 		(void) cv_wait(&ptl1_cv, &ptl1_mutex);
2335 	}
2336 	mutex_exit(&ptl1_mutex);
2337 }
2338 
2339 /*
2340  * Called during early startup to create the ptl1_thread
2341  */
2342 void
2343 init_ptl1_thread(void)
2344 {
2345 	ptl1_thread_p = thread_create(NULL, 0, ptl1_thread, NULL, 0,
2346 	    &p0, TS_RUN, 0);
2347 }
2348 #endif	/* PTL1_PANIC_DEBUG */
2349 
2350 
2351 /*
2352  * Add to a memory list.
2353  * start = start of new memory segment
2354  * len = length of new memory segment in bytes
2355  * memlistp = pointer to array of available memory segment structures
2356  * curmemlistp = memory list to which to add segment.
2357  */
2358 static void
2359 memlist_add(uint64_t start, uint64_t len, struct memlist **memlistp,
2360 	struct memlist **curmemlistp)
2361 {
2362 	struct memlist *new;
2363 
2364 	new = *memlistp;
2365 	new->address = start;
2366 	new->size = len;
2367 	*memlistp = new + 1;
2368 
2369 	memlist_insert(new, curmemlistp);
2370 }
2371 
2372 /*
2373  * In the case of architectures that support dynamic addition of
2374  * memory at run-time there are two cases where memsegs need to
2375  * be initialized and added to the memseg list.
2376  * 1) memsegs that are constructed at startup.
2377  * 2) memsegs that are constructed at run-time on
2378  *    hot-plug capable architectures.
2379  * This code was originally part of the function kphysm_init().
2380  */
2381 
2382 static void
2383 memseg_list_add(struct memseg *memsegp)
2384 {
2385 	struct memseg **prev_memsegp;
2386 	pgcnt_t num;
2387 
2388 	/* insert in memseg list, decreasing number of pages order */
2389 
2390 	num = MSEG_NPAGES(memsegp);
2391 
2392 	for (prev_memsegp = &memsegs; *prev_memsegp;
2393 	    prev_memsegp = &((*prev_memsegp)->next)) {
2394 		if (num > MSEG_NPAGES(*prev_memsegp))
2395 			break;
2396 	}
2397 
2398 	memsegp->next = *prev_memsegp;
2399 	*prev_memsegp = memsegp;
2400 
2401 	if (kpm_enable) {
2402 		memsegp->nextpa = (memsegp->next) ?
2403 			va_to_pa(memsegp->next) : MSEG_NULLPTR_PA;
2404 
2405 		if (prev_memsegp != &memsegs) {
2406 			struct memseg *msp;
2407 			msp = (struct memseg *)((caddr_t)prev_memsegp -
2408 				offsetof(struct memseg, next));
2409 			msp->nextpa = va_to_pa(memsegp);
2410 		} else {
2411 			memsegspa = va_to_pa(memsegs);
2412 		}
2413 	}
2414 }
2415 
2416 /*
2417  * PSM add_physmem_cb(). US-II and newer processors have some
2418  * flavor of the prefetch capability implemented. We exploit
2419  * this capability for optimum performance.
2420  */
2421 #define	PREFETCH_BYTES	64
2422 
2423 void
2424 add_physmem_cb(page_t *pp, pfn_t pnum)
2425 {
2426 	extern void	 prefetch_page_w(void *);
2427 
2428 	pp->p_pagenum = pnum;
2429 
2430 	/*
2431 	 * Prefetch one more page_t into E$. To prevent future
2432 	 * mishaps with the sizeof(page_t) changing on us, we
2433 	 * catch this on debug kernels if we can't bring in the
2434 	 * entire hpage with 2 PREFETCH_BYTES reads. See
2435 	 * also, sun4u/cpu/cpu_module.c
2436 	 */
2437 	/*LINTED*/
2438 	ASSERT(sizeof (page_t) <= 2*PREFETCH_BYTES);
2439 	prefetch_page_w((char *)pp);
2440 }
2441 
2442 /*
2443  * kphysm_init() tackles the problem of initializing physical memory.
2444  * The old startup made some assumptions about the kernel living in
2445  * physically contiguous space which is no longer valid.
2446  */
2447 static void
2448 kphysm_init(page_t *pp, struct memseg *memsegp, pgcnt_t npages,
2449 	uintptr_t kpm_pp, pgcnt_t kpm_npages)
2450 {
2451 	struct memlist	*pmem;
2452 	struct memseg	*msp;
2453 	pfn_t		 base;
2454 	pgcnt_t		 num;
2455 	pfn_t		 lastseg_pages_end = 0;
2456 	pgcnt_t		 nelem_used = 0;
2457 
2458 	ASSERT(page_hash != NULL && page_hashsz != 0);
2459 
2460 	msp = memsegp;
2461 	for (pmem = phys_avail; pmem && npages; pmem = pmem->next) {
2462 
2463 		/*
2464 		 * Build the memsegs entry
2465 		 */
2466 		num = btop(pmem->size);
2467 		if (num > npages)
2468 			num = npages;
2469 		npages -= num;
2470 		base = btop(pmem->address);
2471 
2472 		msp->pages = pp;
2473 		msp->epages = pp + num;
2474 		msp->pages_base = base;
2475 		msp->pages_end = base + num;
2476 
2477 		if (kpm_enable) {
2478 			pfn_t pbase_a;
2479 			pfn_t pend_a;
2480 			pfn_t prev_pend_a;
2481 			pgcnt_t	nelem;
2482 
2483 			msp->pagespa = va_to_pa(pp);
2484 			msp->epagespa = va_to_pa(pp + num);
2485 			pbase_a = kpmptop(ptokpmp(base));
2486 			pend_a = kpmptop(ptokpmp(base + num - 1)) + kpmpnpgs;
2487 			nelem = ptokpmp(pend_a - pbase_a);
2488 			msp->kpm_nkpmpgs = nelem;
2489 			msp->kpm_pbase = pbase_a;
2490 			if (lastseg_pages_end) {
2491 				/*
2492 				 * Assume phys_avail is in ascending order
2493 				 * of physical addresses.
2494 				 */
2495 				ASSERT(base + num > lastseg_pages_end);
2496 				prev_pend_a = kpmptop(
2497 				    ptokpmp(lastseg_pages_end - 1)) + kpmpnpgs;
2498 
2499 				if (prev_pend_a > pbase_a) {
2500 					/*
2501 					 * Overlap, more than one memseg may
2502 					 * point to the same kpm_page range.
2503 					 */
2504 					if (kpm_smallpages == 0) {
2505 						msp->kpm_pages =
2506 						    (kpm_page_t *)kpm_pp - 1;
2507 						kpm_pp = (uintptr_t)
2508 							((kpm_page_t *)kpm_pp
2509 							+ nelem - 1);
2510 					} else {
2511 						msp->kpm_spages =
2512 						    (kpm_spage_t *)kpm_pp - 1;
2513 						kpm_pp = (uintptr_t)
2514 							((kpm_spage_t *)kpm_pp
2515 							+ nelem - 1);
2516 					}
2517 					nelem_used += nelem - 1;
2518 
2519 				} else {
2520 					if (kpm_smallpages == 0) {
2521 						msp->kpm_pages =
2522 						    (kpm_page_t *)kpm_pp;
2523 						kpm_pp = (uintptr_t)
2524 							((kpm_page_t *)kpm_pp
2525 							+ nelem);
2526 					} else {
2527 						msp->kpm_spages =
2528 						    (kpm_spage_t *)kpm_pp;
2529 						kpm_pp = (uintptr_t)
2530 							((kpm_spage_t *)
2531 							kpm_pp + nelem);
2532 					}
2533 					nelem_used += nelem;
2534 				}
2535 
2536 			} else {
2537 				if (kpm_smallpages == 0) {
2538 					msp->kpm_pages = (kpm_page_t *)kpm_pp;
2539 					kpm_pp = (uintptr_t)
2540 						((kpm_page_t *)kpm_pp + nelem);
2541 				} else {
2542 					msp->kpm_spages = (kpm_spage_t *)kpm_pp;
2543 					kpm_pp = (uintptr_t)
2544 						((kpm_spage_t *)kpm_pp + nelem);
2545 				}
2546 				nelem_used = nelem;
2547 			}
2548 
2549 			if (nelem_used > kpm_npages)
2550 				panic("kphysm_init: kpm_pp overflow\n");
2551 
2552 			msp->kpm_pagespa = va_to_pa(msp->kpm_pages);
2553 			lastseg_pages_end = msp->pages_end;
2554 		}
2555 
2556 		memseg_list_add(msp);
2557 
2558 		/*
2559 		 * add_physmem() initializes the PSM part of the page
2560 		 * struct by calling the PSM back with add_physmem_cb().
2561 		 * In addition it coalesces pages into larger pages as
2562 		 * it initializes them.
2563 		 */
2564 		add_physmem(pp, num, base);
2565 		pp += num;
2566 		msp++;
2567 	}
2568 
2569 	build_pfn_hash();
2570 }
2571 
2572 /*
2573  * Kernel VM initialization.
2574  * Assumptions about kernel address space ordering:
2575  *	(1) gap (user space)
2576  *	(2) kernel text
2577  *	(3) kernel data/bss
2578  *	(4) gap
2579  *	(5) kernel data structures
2580  *	(6) gap
2581  *	(7) debugger (optional)
2582  *	(8) monitor
2583  *	(9) gap (possibly null)
2584  *	(10) dvma
2585  *	(11) devices
2586  */
2587 static void
2588 kvm_init(void)
2589 {
2590 	/*
2591 	 * Put the kernel segments in kernel address space.
2592 	 */
2593 	rw_enter(&kas.a_lock, RW_WRITER);
2594 	as_avlinit(&kas);
2595 
2596 	(void) seg_attach(&kas, (caddr_t)KERNELBASE,
2597 	    (size_t)(e_moddata - KERNELBASE), &ktextseg);
2598 	(void) segkmem_create(&ktextseg);
2599 
2600 	(void) seg_attach(&kas, (caddr_t)(KERNELBASE + MMU_PAGESIZE4M),
2601 	    (size_t)(MMU_PAGESIZE4M), &ktexthole);
2602 	(void) segkmem_create(&ktexthole);
2603 
2604 	(void) seg_attach(&kas, (caddr_t)valloc_base,
2605 	    (size_t)(econtig32 - valloc_base), &kvalloc);
2606 	(void) segkmem_create(&kvalloc);
2607 
2608 	if (kmem64_base) {
2609 	    (void) seg_attach(&kas, (caddr_t)kmem64_base,
2610 		(size_t)(kmem64_end - kmem64_base), &kmem64);
2611 	    (void) segkmem_create(&kmem64);
2612 	}
2613 
2614 	/*
2615 	 * We're about to map out /boot.  This is the beginning of the
2616 	 * system resource management transition. We can no longer
2617 	 * call into /boot for I/O or memory allocations.
2618 	 */
2619 	(void) seg_attach(&kas, kernelheap, ekernelheap - kernelheap, &kvseg);
2620 	(void) segkmem_create(&kvseg);
2621 	hblk_alloc_dynamic = 1;
2622 
2623 	/*
2624 	 * we need to preallocate pages for DR operations before enabling large
2625 	 * page kernel heap because of memseg_remap_init() hat_unload() hack.
2626 	 */
2627 	memseg_remap_init();
2628 
2629 	/* at this point we are ready to use large page heap */
2630 	segkmem_heap_lp_init();
2631 
2632 	(void) seg_attach(&kas, (caddr_t)SYSBASE32, SYSLIMIT32 - SYSBASE32,
2633 	    &kvseg32);
2634 	(void) segkmem_create(&kvseg32);
2635 
2636 	/*
2637 	 * Create a segment for the debugger.
2638 	 */
2639 	(void) seg_attach(&kas, (caddr_t)SEGDEBUGBASE, (size_t)SEGDEBUGSIZE,
2640 	    &kdebugseg);
2641 	(void) segkmem_create(&kdebugseg);
2642 
2643 	rw_exit(&kas.a_lock);
2644 }
2645 
2646 char obp_tte_str[] =
2647 	"h# %x constant MMU_PAGESHIFT "
2648 	"h# %x constant TTE8K "
2649 	"h# %x constant SFHME_SIZE "
2650 	"h# %x constant SFHME_TTE "
2651 	"h# %x constant HMEBLK_TAG "
2652 	"h# %x constant HMEBLK_NEXT "
2653 	"h# %x constant HMEBLK_MISC "
2654 	"h# %x constant HMEBLK_HME1 "
2655 	"h# %x constant NHMENTS "
2656 	"h# %x constant HBLK_SZMASK "
2657 	"h# %x constant HBLK_RANGE_SHIFT "
2658 	"h# %x constant HMEBP_HBLK "
2659 	"h# %x constant HMEBUCKET_SIZE "
2660 	"h# %x constant HTAG_SFMMUPSZ "
2661 	"h# %x constant HTAG_REHASHSZ "
2662 	"h# %x constant mmu_hashcnt "
2663 	"h# %p constant uhme_hash "
2664 	"h# %p constant khme_hash "
2665 	"h# %x constant UHMEHASH_SZ "
2666 	"h# %x constant KHMEHASH_SZ "
2667 	"h# %p constant KHATID "
2668 	"h# %x constant CTX_SIZE "
2669 	"h# %x constant CTX_SFMMU "
2670 	"h# %p constant ctxs "
2671 	"h# %x constant ASI_MEM "
2672 
2673 	": PHYS-X@ ( phys -- data ) "
2674 	"   ASI_MEM spacex@ "
2675 	"; "
2676 
2677 	": PHYS-W@ ( phys -- data ) "
2678 	"   ASI_MEM spacew@ "
2679 	"; "
2680 
2681 	": PHYS-L@ ( phys -- data ) "
2682 	"   ASI_MEM spaceL@ "
2683 	"; "
2684 
2685 	": TTE_PAGE_SHIFT ( ttesz -- hmeshift ) "
2686 	"   3 * MMU_PAGESHIFT + "
2687 	"; "
2688 
2689 	": TTE_IS_VALID ( ttep -- flag ) "
2690 	"   PHYS-X@ 0< "
2691 	"; "
2692 
2693 	": HME_HASH_SHIFT ( ttesz -- hmeshift ) "
2694 	"   dup TTE8K =  if "
2695 	"      drop HBLK_RANGE_SHIFT "
2696 	"   else "
2697 	"      TTE_PAGE_SHIFT "
2698 	"   then "
2699 	"; "
2700 
2701 	": HME_HASH_BSPAGE ( addr hmeshift -- bspage ) "
2702 	"   tuck >> swap MMU_PAGESHIFT - << "
2703 	"; "
2704 
2705 	": HME_HASH_FUNCTION ( sfmmup addr hmeshift -- hmebp ) "
2706 	"   >> over xor swap                    ( hash sfmmup ) "
2707 	"   KHATID <>  if                       ( hash ) "
2708 	"      UHMEHASH_SZ and                  ( bucket ) "
2709 	"      HMEBUCKET_SIZE * uhme_hash +     ( hmebp ) "
2710 	"   else                                ( hash ) "
2711 	"      KHMEHASH_SZ and                  ( bucket ) "
2712 	"      HMEBUCKET_SIZE * khme_hash +     ( hmebp ) "
2713 	"   then                                ( hmebp ) "
2714 	"; "
2715 
2716 	": HME_HASH_TABLE_SEARCH "
2717 	"       ( sfmmup hmebp hblktag --  sfmmup null | sfmmup hmeblkp ) "
2718 	"   >r hmebp_hblk + phys-x@ begin ( sfmmup hmeblkp ) ( r: hblktag ) "
2719 	"      dup if   		( sfmmup hmeblkp ) ( r: hblktag ) "
2720 	"         dup hmeblk_tag + phys-x@ r@ = if ( sfmmup hmeblkp )	  "
2721 	"	     dup hmeblk_tag + 8 + phys-x@ 2 pick = if		  "
2722 	"		  true 	( sfmmup hmeblkp true ) ( r: hblktag )	  "
2723 	"	     else						  "
2724 	"	     	  hmeblk_next + phys-x@ false 			  "
2725 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2726 	"	     then  						  "
2727 	"	  else							  "
2728 	"	     hmeblk_next + phys-x@ false 			  "
2729 	"			( sfmmup hmeblkp false ) ( r: hblktag )   "
2730 	"	  then 							  "
2731 	"      else							  "
2732 	"         true 							  "
2733 	"      then  							  "
2734 	"   until r> drop 						  "
2735 	"; "
2736 
2737 	": CNUM_TO_SFMMUP ( cnum -- sfmmup ) "
2738 	"   CTX_SIZE * ctxs + CTX_SFMMU + "
2739 	"x@ "
2740 	"; "
2741 
2742 	": HME_HASH_TAG ( sfmmup rehash addr -- hblktag ) "
2743 	"   over HME_HASH_SHIFT HME_HASH_BSPAGE      ( sfmmup rehash bspage ) "
2744 	"   HTAG_REHASHSZ << or nip		     ( hblktag ) "
2745 	"; "
2746 
2747 	": HBLK_TO_TTEP ( hmeblkp addr -- ttep ) "
2748 	"   over HMEBLK_MISC + PHYS-L@ HBLK_SZMASK and  ( hmeblkp addr ttesz ) "
2749 	"   TTE8K =  if                            ( hmeblkp addr ) "
2750 	"      MMU_PAGESHIFT >> NHMENTS 1- and     ( hmeblkp hme-index ) "
2751 	"   else                                   ( hmeblkp addr ) "
2752 	"      drop 0                              ( hmeblkp 0 ) "
2753 	"   then                                   ( hmeblkp hme-index ) "
2754 	"   SFHME_SIZE * + HMEBLK_HME1 +           ( hmep ) "
2755 	"   SFHME_TTE +                            ( ttep ) "
2756 	"; "
2757 
2758 	": unix-tte ( addr cnum -- false | tte-data true ) "
2759 	"      CNUM_TO_SFMMUP                 ( addr sfmmup ) "
2760 	"      mmu_hashcnt 1+ 1  do           ( addr sfmmup ) "
2761 	"         2dup swap i HME_HASH_SHIFT  "
2762 					"( addr sfmmup sfmmup addr hmeshift ) "
2763 	"         HME_HASH_FUNCTION           ( addr sfmmup hmebp ) "
2764 	"         over i 4 pick               "
2765 				"( addr sfmmup hmebp sfmmup rehash addr ) "
2766 	"         HME_HASH_TAG                ( addr sfmmup hmebp hblktag ) "
2767 	"         HME_HASH_TABLE_SEARCH       "
2768 					"( addr sfmmup { null | hmeblkp } ) "
2769 	"         ?dup  if                    ( addr sfmmup hmeblkp ) "
2770 	"            nip swap HBLK_TO_TTEP    ( ttep ) "
2771 	"            dup TTE_IS_VALID  if     ( valid-ttep ) "
2772 	"               PHYS-X@ true          ( tte-data true ) "
2773 	"            else                     ( invalid-tte ) "
2774 	"               drop false            ( false ) "
2775 	"            then                     ( false | tte-data true ) "
2776 	"            unloop exit              ( false | tte-data true ) "
2777 	"         then                        ( addr sfmmup ) "
2778 	"      loop                           ( addr sfmmup ) "
2779 	"      2drop false                    ( false ) "
2780 	"; "
2781 ;
2782 
2783 void
2784 create_va_to_tte(void)
2785 {
2786 	char *bp;
2787 	extern int khmehash_num, uhmehash_num;
2788 	extern struct hmehash_bucket *khme_hash, *uhme_hash;
2789 
2790 #define	OFFSET(type, field)	((uintptr_t)(&((type *)0)->field))
2791 
2792 	bp = (char *)kobj_zalloc(MMU_PAGESIZE, KM_SLEEP);
2793 
2794 	/*
2795 	 * Teach obp how to parse our sw ttes.
2796 	 */
2797 	(void) sprintf(bp, obp_tte_str,
2798 	    MMU_PAGESHIFT,
2799 	    TTE8K,
2800 	    sizeof (struct sf_hment),
2801 	    OFFSET(struct sf_hment, hme_tte),
2802 	    OFFSET(struct hme_blk, hblk_tag),
2803 	    OFFSET(struct hme_blk, hblk_nextpa),
2804 	    OFFSET(struct hme_blk, hblk_misc),
2805 	    OFFSET(struct hme_blk, hblk_hme),
2806 	    NHMENTS,
2807 	    HBLK_SZMASK,
2808 	    HBLK_RANGE_SHIFT,
2809 	    OFFSET(struct hmehash_bucket, hmeh_nextpa),
2810 	    sizeof (struct hmehash_bucket),
2811 	    HTAG_SFMMUPSZ,
2812 	    HTAG_REHASHSZ,
2813 	    mmu_hashcnt,
2814 	    (caddr_t)va_to_pa((caddr_t)uhme_hash),
2815 	    (caddr_t)va_to_pa((caddr_t)khme_hash),
2816 	    UHMEHASH_SZ,
2817 	    KHMEHASH_SZ,
2818 	    KHATID,
2819 	    sizeof (struct ctx),
2820 	    OFFSET(struct ctx, ctx_sfmmu),
2821 	    ctxs,
2822 	    ASI_MEM);
2823 	prom_interpret(bp, 0, 0, 0, 0, 0);
2824 
2825 	kobj_free(bp, MMU_PAGESIZE);
2826 }
2827 
2828 void
2829 install_va_to_tte(void)
2830 {
2831 	/*
2832 	 * advise prom that he can use unix-tte
2833 	 */
2834 	prom_interpret("' unix-tte is va>tte-data", 0, 0, 0, 0, 0);
2835 }
2836 
2837 
2838 /*
2839  * Because kmdb links prom_stdout_is_framebuffer into its own
2840  * module, we add "device-type=display" here for /os-io node, so that
2841  * prom_stdout_is_framebuffer still works corrrectly  after /os-io node
2842  * is registered into OBP.
2843  */
2844 static char *create_node =
2845 	"\" /\" find-device "
2846 	"new-device "
2847 	"\" os-io\" device-name "
2848 	"\" display\" device-type "
2849 	": cb-r/w  ( adr,len method$ -- #read/#written ) "
2850 	"   2>r swap 2 2r> ['] $callback  catch  if "
2851 	"      2drop 3drop 0 "
2852 	"   then "
2853 	"; "
2854 	": read ( adr,len -- #read ) "
2855 	"       \" read\" ['] cb-r/w catch  if  2drop 2drop -2 exit then "
2856 	"       ( retN ... ret1 N ) "
2857 	"       ?dup  if "
2858 	"               swap >r 1-  0  ?do  drop  loop  r> "
2859 	"       else "
2860 	"               -2 "
2861 	"       then "
2862 	";    "
2863 	": write ( adr,len -- #written ) "
2864 	"       \" write\" ['] cb-r/w catch  if  2drop 2drop 0 exit  then "
2865 	"       ( retN ... ret1 N ) "
2866 	"       ?dup  if "
2867 	"               swap >r 1-  0  ?do  drop  loop  r> "
2868 	"        else "
2869 	"               0 "
2870 	"       then "
2871 	"; "
2872 	": poll-tty ( -- ) ; "
2873 	": install-abort  ( -- )  ['] poll-tty d# 10 alarm ; "
2874 	": remove-abort ( -- )  ['] poll-tty 0 alarm ; "
2875 	": cb-give/take ( $method -- ) "
2876 	"       0 -rot ['] $callback catch  ?dup  if "
2877 	"               >r 2drop 2drop r> throw "
2878 	"       else "
2879 	"               0  ?do  drop  loop "
2880 	"       then "
2881 	"; "
2882 	": give ( -- )  \" exit-input\" cb-give/take ; "
2883 	": take ( -- )  \" enter-input\" cb-give/take ; "
2884 	": open ( -- ok? )  true ; "
2885 	": close ( -- ) ; "
2886 	"finish-device "
2887 	"device-end ";
2888 
2889 /*
2890  * Create the OBP input/output node (FCode serial driver).
2891  * It is needed for both USB console keyboard and for
2892  * the kernel terminal emulator.  It is too early to check for a
2893  * kernel console compatible framebuffer now, so we create this
2894  * so that we're ready if we need to enable kernel terminal emulation.
2895  *
2896  * When the USB software takes over the input device at the time
2897  * consconfig runs, OBP's stdin is redirected to this node.
2898  * Whenever the FORTH user interface is used after this switch,
2899  * the node will call back into the kernel for console input.
2900  * If a serial device such as ttya or a UART with a Type 5 keyboard
2901  * attached is used, OBP takes over the serial device when the system
2902  * goes to the debugger after the system is booted.  This sharing
2903  * of the relatively simple serial device is difficult but possible.
2904  * Sharing the USB host controller is impossible due its complexity.
2905  *
2906  * Similarly to USB keyboard input redirection, after consconfig_dacf
2907  * configures a kernel console framebuffer as the standard output
2908  * device, OBP's stdout is switched to to vector through the
2909  * /os-io node into the kernel terminal emulator.
2910  */
2911 static void
2912 startup_create_io_node(void)
2913 {
2914 	prom_interpret(create_node, 0, 0, 0, 0, 0);
2915 }
2916 
2917 
2918 static void
2919 do_prom_version_check(void)
2920 {
2921 	int i;
2922 	pnode_t node;
2923 	char buf[64];
2924 	static char drev[] = "Down-rev firmware detected%s\n"
2925 		"\tPlease upgrade to the following minimum version:\n"
2926 		"\t\t%s\n";
2927 
2928 	i = prom_version_check(buf, sizeof (buf), &node);
2929 
2930 	if (i == PROM_VER64_OK)
2931 		return;
2932 
2933 	if (i == PROM_VER64_UPGRADE) {
2934 		cmn_err(CE_WARN, drev, "", buf);
2935 
2936 #ifdef	DEBUG
2937 		prom_enter_mon();	/* Type 'go' to continue */
2938 		cmn_err(CE_WARN, "Booting with down-rev firmware\n");
2939 		return;
2940 #else
2941 		halt(0);
2942 #endif
2943 	}
2944 
2945 	/*
2946 	 * The other possibility is that this is a server running
2947 	 * good firmware, but down-rev firmware was detected on at
2948 	 * least one other cpu board. We just complain if we see
2949 	 * that.
2950 	 */
2951 	cmn_err(CE_WARN, drev, " on one or more CPU boards", buf);
2952 }
2953 
2954 static void
2955 kpm_init()
2956 {
2957 	kpm_pgshft = (kpm_smallpages == 0) ? MMU_PAGESHIFT4M : MMU_PAGESHIFT;
2958 	kpm_pgsz = 1ull << kpm_pgshft;
2959 	kpm_pgoff = kpm_pgsz - 1;
2960 	kpmp2pshft = kpm_pgshft - PAGESHIFT;
2961 	kpmpnpgs = 1 << kpmp2pshft;
2962 	ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0);
2963 }
2964 
2965 void
2966 kpm_npages_setup(int memblocks)
2967 {
2968 	/*
2969 	 * npages can be scattered in a maximum of 'memblocks'
2970 	 */
2971 	kpm_npages = ptokpmpr(npages) + memblocks;
2972 }
2973 
2974 /*
2975  * Must be defined in platform dependent code.
2976  */
2977 extern caddr_t modtext;
2978 extern size_t modtext_sz;
2979 extern caddr_t moddata;
2980 
2981 #define	HEAPTEXT_ARENA(addr)	\
2982 	((uintptr_t)(addr) < KERNELBASE + 2 * MMU_PAGESIZE4M ? 0 : \
2983 	(((uintptr_t)(addr) - HEAPTEXT_BASE) / \
2984 	(HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) + 1))
2985 
2986 #define	HEAPTEXT_OVERSIZED(addr)	\
2987 	((uintptr_t)(addr) >= HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE)
2988 
2989 vmem_t *texthole_source[HEAPTEXT_NARENAS];
2990 vmem_t *texthole_arena[HEAPTEXT_NARENAS];
2991 kmutex_t texthole_lock;
2992 
2993 char kern_bootargs[OBP_MAXPATHLEN];
2994 
2995 void
2996 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
2997 {
2998 	uintptr_t addr, limit;
2999 
3000 	addr = HEAPTEXT_BASE;
3001 	limit = addr + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE;
3002 
3003 	/*
3004 	 * Before we initialize the text_arena, we want to punch holes in the
3005 	 * underlying heaptext_arena.  This guarantees that for any text
3006 	 * address we can find a text hole less than HEAPTEXT_MAPPED away.
3007 	 */
3008 	for (; addr + HEAPTEXT_UNMAPPED <= limit;
3009 	    addr += HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) {
3010 		(void) vmem_xalloc(heaptext_arena, HEAPTEXT_UNMAPPED, PAGESIZE,
3011 		    0, 0, (void *)addr, (void *)(addr + HEAPTEXT_UNMAPPED),
3012 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3013 	}
3014 
3015 	/*
3016 	 * Allocate one page at the oversize to break up the text region
3017 	 * from the oversized region.
3018 	 */
3019 	(void) vmem_xalloc(heaptext_arena, PAGESIZE, PAGESIZE, 0, 0,
3020 	    (void *)limit, (void *)(limit + PAGESIZE),
3021 	    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3022 
3023 	*text_arena = vmem_create("module_text", modtext, modtext_sz,
3024 	    sizeof (uintptr_t), segkmem_alloc, segkmem_free,
3025 	    heaptext_arena, 0, VM_SLEEP);
3026 	*data_arena = vmem_create("module_data", moddata, MODDATA, 1,
3027 	    segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3028 }
3029 
3030 caddr_t
3031 kobj_text_alloc(vmem_t *arena, size_t size)
3032 {
3033 	caddr_t rval, better;
3034 
3035 	/*
3036 	 * First, try a sleeping allocation.
3037 	 */
3038 	rval = vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT);
3039 
3040 	if (size >= HEAPTEXT_MAPPED || !HEAPTEXT_OVERSIZED(rval))
3041 		return (rval);
3042 
3043 	/*
3044 	 * We didn't get the area that we wanted.  We're going to try to do an
3045 	 * allocation with explicit constraints.
3046 	 */
3047 	better = vmem_xalloc(arena, size, sizeof (uintptr_t), 0, 0, NULL,
3048 	    (void *)(HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE),
3049 	    VM_NOSLEEP | VM_BESTFIT);
3050 
3051 	if (better != NULL) {
3052 		/*
3053 		 * That worked.  Free our first attempt and return.
3054 		 */
3055 		vmem_free(arena, rval, size);
3056 		return (better);
3057 	}
3058 
3059 	/*
3060 	 * That didn't work; we'll have to return our first attempt.
3061 	 */
3062 	return (rval);
3063 }
3064 
3065 caddr_t
3066 kobj_texthole_alloc(caddr_t addr, size_t size)
3067 {
3068 	int arena = HEAPTEXT_ARENA(addr);
3069 	char c[30];
3070 	uintptr_t base;
3071 
3072 	if (HEAPTEXT_OVERSIZED(addr)) {
3073 		/*
3074 		 * If this is an oversized allocation, there is no text hole
3075 		 * available for it; return NULL.
3076 		 */
3077 		return (NULL);
3078 	}
3079 
3080 	mutex_enter(&texthole_lock);
3081 
3082 	if (texthole_arena[arena] == NULL) {
3083 		ASSERT(texthole_source[arena] == NULL);
3084 
3085 		if (arena == 0) {
3086 			texthole_source[0] = vmem_create("module_text_holesrc",
3087 			    (void *)(KERNELBASE + MMU_PAGESIZE4M),
3088 			    MMU_PAGESIZE4M, PAGESIZE, NULL, NULL, NULL,
3089 			    0, VM_SLEEP);
3090 		} else {
3091 			base = HEAPTEXT_BASE +
3092 			    (arena - 1) * (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED);
3093 
3094 			(void) snprintf(c, sizeof (c),
3095 			    "heaptext_holesrc_%d", arena);
3096 
3097 			texthole_source[arena] = vmem_create(c, (void *)base,
3098 			    HEAPTEXT_UNMAPPED, PAGESIZE, NULL, NULL, NULL,
3099 			    0, VM_SLEEP);
3100 		}
3101 
3102 		(void) snprintf(c, sizeof (c), "heaptext_hole_%d", arena);
3103 
3104 		texthole_arena[arena] = vmem_create(c, NULL, 0,
3105 		    sizeof (uint32_t), segkmem_alloc_permanent, segkmem_free,
3106 		    texthole_source[arena], 0, VM_SLEEP);
3107 	}
3108 
3109 	mutex_exit(&texthole_lock);
3110 
3111 	ASSERT(texthole_arena[arena] != NULL);
3112 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3113 	return (vmem_alloc(texthole_arena[arena], size,
3114 	    VM_BESTFIT | VM_NOSLEEP));
3115 }
3116 
3117 void
3118 kobj_texthole_free(caddr_t addr, size_t size)
3119 {
3120 	int arena = HEAPTEXT_ARENA(addr);
3121 
3122 	ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3123 	ASSERT(texthole_arena[arena] != NULL);
3124 	vmem_free(texthole_arena[arena], addr, size);
3125 }
3126