1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2016 by Delphix. All rights reserved.
25 * Copyright 2018 Nexenta Systems, Inc. All rights reserved.
26 */
27
28 #include <sys/machsystm.h>
29 #include <sys/archsystm.h>
30 #include <sys/vm.h>
31 #include <sys/cpu.h>
32 #include <sys/atomic.h>
33 #include <sys/reboot.h>
34 #include <sys/kdi.h>
35 #include <sys/bootconf.h>
36 #include <sys/memlist_plat.h>
37 #include <sys/memlist_impl.h>
38 #include <sys/prom_plat.h>
39 #include <sys/prom_isa.h>
40 #include <sys/autoconf.h>
41 #include <sys/ivintr.h>
42 #include <sys/fpu/fpusystm.h>
43 #include <sys/iommutsb.h>
44 #include <vm/vm_dep.h>
45 #include <vm/seg_dev.h>
46 #include <vm/seg_kmem.h>
47 #include <vm/seg_kpm.h>
48 #include <vm/seg_map.h>
49 #include <vm/seg_kp.h>
50 #include <sys/sysconf.h>
51 #include <vm/hat_sfmmu.h>
52 #include <sys/kobj.h>
53 #include <sys/sun4asi.h>
54 #include <sys/clconf.h>
55 #include <sys/platform_module.h>
56 #include <sys/panic.h>
57 #include <sys/cpu_sgnblk_defs.h>
58 #include <sys/clock.h>
59 #include <sys/cmn_err.h>
60 #include <sys/dumphdr.h>
61 #include <sys/promif.h>
62 #include <sys/prom_debug.h>
63 #include <sys/traptrace.h>
64 #include <sys/memnode.h>
65 #include <sys/mem_cage.h>
66 #include <sys/mmu.h>
67 #include <sys/swap.h>
68
69 extern void setup_trap_table(void);
70 extern int cpu_intrq_setup(struct cpu *);
71 extern void cpu_intrq_register(struct cpu *);
72 extern void contig_mem_init(void);
73 extern caddr_t contig_mem_prealloc(caddr_t, pgcnt_t);
74 extern void mach_dump_buffer_init(void);
75 extern void mach_descrip_init(void);
76 extern void mach_descrip_startup_fini(void);
77 extern void mach_memscrub(void);
78 extern void mach_fpras(void);
79 extern void mach_cpu_halt_idle(void);
80 extern void mach_hw_copy_limit(void);
81 extern void load_mach_drivers(void);
82 extern void load_tod_module(void);
83 #pragma weak load_tod_module
84
85 extern int ndata_alloc_mmfsa(struct memlist *ndata);
86 #pragma weak ndata_alloc_mmfsa
87
88 extern void cif_init(void);
89 #pragma weak cif_init
90
91 extern void parse_idprom(void);
92 extern void add_vx_handler(char *, int, void (*)(cell_t *));
93 extern void mem_config_init(void);
94 extern void memseg_remap_init(void);
95
96 extern void mach_kpm_init(void);
97 extern void pcf_init();
98 extern int size_pse_array(pgcnt_t, int);
99 extern void pg_init();
100
101 /*
102 * External Data:
103 */
104 extern int vac_size; /* cache size in bytes */
105 extern uint_t vac_mask; /* VAC alignment consistency mask */
106 extern uint_t vac_colors;
107
108 /*
109 * Global Data Definitions:
110 */
111
112 /*
113 * XXX - Don't port this to new architectures
114 * A 3rd party volume manager driver (vxdm) depends on the symbol romp.
115 * 'romp' has no use with a prom with an IEEE 1275 client interface.
116 * The driver doesn't use the value, but it depends on the symbol.
117 */
118 void *romp; /* veritas driver won't load without romp 4154976 */
119 /*
120 * Declare these as initialized data so we can patch them.
121 */
122 pgcnt_t physmem = 0; /* memory size in pages, patch if you want less */
123 pgcnt_t segkpsize =
124 btop(SEGKPDEFSIZE); /* size of segkp segment in pages */
125 uint_t segmap_percent = 6; /* Size of segmap segment */
126
127 int use_cache = 1; /* cache not reliable (605 bugs) with MP */
128 int vac_copyback = 1;
129 char *cache_mode = NULL;
130 int use_mix = 1;
131 int prom_debug = 0;
132
133 caddr_t boot_tba; /* %tba at boot - used by kmdb */
134 uint_t tba_taken_over = 0;
135
136 caddr_t s_text; /* start of kernel text segment */
137 caddr_t e_text; /* end of kernel text segment */
138 caddr_t s_data; /* start of kernel data segment */
139 caddr_t e_data; /* end of kernel data segment */
140
141 caddr_t modtext; /* beginning of module text */
142 size_t modtext_sz; /* size of module text */
143 caddr_t moddata; /* beginning of module data reserve */
144 caddr_t e_moddata; /* end of module data reserve */
145
146 /*
147 * End of first block of contiguous kernel in 32-bit virtual address space
148 */
149 caddr_t econtig32; /* end of first blk of contiguous kernel */
150
151 caddr_t ncbase; /* beginning of non-cached segment */
152 caddr_t ncend; /* end of non-cached segment */
153
154 size_t ndata_remain_sz; /* bytes from end of data to 4MB boundary */
155 caddr_t nalloc_base; /* beginning of nucleus allocation */
156 caddr_t nalloc_end; /* end of nucleus allocatable memory */
157 caddr_t valloc_base; /* beginning of kvalloc segment */
158
159 caddr_t kmem64_base; /* base of kernel mem segment in 64-bit space */
160 caddr_t kmem64_end; /* end of kernel mem segment in 64-bit space */
161 size_t kmem64_sz; /* bytes in kernel mem segment, 64-bit space */
162 caddr_t kmem64_aligned_end; /* end of large page, overmaps 64-bit space */
163 int kmem64_szc; /* page size code */
164 uint64_t kmem64_pabase = (uint64_t)-1; /* physical address of kmem64_base */
165
166 uintptr_t shm_alignment; /* VAC address consistency modulus */
167 struct memlist *phys_install; /* Total installed physical memory */
168 struct memlist *phys_avail; /* Available (unreserved) physical memory */
169 struct memlist *virt_avail; /* Available (unmapped?) virtual memory */
170 struct memlist *nopp_list; /* pages with no backing page structs */
171 struct memlist ndata; /* memlist of nucleus allocatable memory */
172 int memexp_flag; /* memory expansion card flag */
173 uint64_t ecache_flushaddr; /* physical address used for flushing E$ */
174 pgcnt_t obp_pages; /* Physical pages used by OBP */
175
176 /*
177 * VM data structures
178 */
179 long page_hashsz; /* Size of page hash table (power of two) */
180 unsigned int page_hashsz_shift; /* log2(page_hashsz) */
181 struct page *pp_base; /* Base of system page struct array */
182 size_t pp_sz; /* Size in bytes of page struct array */
183 struct page **page_hash; /* Page hash table */
184 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */
185 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */
186 int pse_shift; /* log2(pse_table_size) */
187 struct seg ktextseg; /* Segment used for kernel executable image */
188 struct seg kvalloc; /* Segment used for "valloc" mapping */
189 struct seg kpseg; /* Segment used for pageable kernel virt mem */
190 struct seg ktexthole; /* Segment used for nucleus text hole */
191 struct seg kmapseg; /* Segment used for generic kernel mappings */
192 struct seg kpmseg; /* Segment used for physical mapping */
193 struct seg kdebugseg; /* Segment used for the kernel debugger */
194
195 void *kpm_pp_base; /* Base of system kpm_page array */
196 size_t kpm_pp_sz; /* Size of system kpm_page array */
197 pgcnt_t kpm_npages; /* How many kpm pages are managed */
198
199 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */
200 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */
201 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */
202
203 int segzio_fromheap = 0; /* zio allocations occur from heap */
204 caddr_t segzio_base; /* Base address of segzio */
205 pgcnt_t segziosize = 0; /* size of zio segment in pages */
206
207 /*
208 * A static DR page_t VA map is reserved that can map the page structures
209 * for a domain's entire RA space. The pages that backs this space are
210 * dynamically allocated and need not be physically contiguous. The DR
211 * map size is derived from KPM size.
212 */
213 int ppvm_enable = 0; /* Static virtual map for page structs */
214 page_t *ppvm_base; /* Base of page struct map */
215 pgcnt_t ppvm_size = 0; /* Size of page struct map */
216
217 /*
218 * debugger pages (if allocated)
219 */
220 struct vnode kdebugvp;
221
222 /*
223 * VA range available to the debugger
224 */
225 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
226 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
227
228 /*
229 * Segment for relocated kernel structures in 64-bit large RAM kernels
230 */
231 struct seg kmem64;
232
233 struct memseg *memseg_free;
234
235 struct vnode unused_pages_vp;
236
237 /*
238 * VM data structures allocated early during boot.
239 */
240 size_t pagehash_sz;
241 uint64_t memlist_sz;
242
243 char tbr_wr_addr_inited = 0;
244
245 caddr_t mpo_heap32_buf = NULL;
246 size_t mpo_heap32_bufsz = 0;
247
248 /*
249 * Static Routines:
250 */
251 static int ndata_alloc_memseg(struct memlist *, size_t);
252 static void memlist_new(uint64_t, uint64_t, struct memlist **);
253 static void memlist_add(uint64_t, uint64_t,
254 struct memlist **, struct memlist **);
255 static void kphysm_init(void);
256 static void kvm_init(void);
257 static void install_kmem64_tte(void);
258
259 static void startup_init(void);
260 static void startup_memlist(void);
261 static void startup_modules(void);
262 static void startup_bop_gone(void);
263 static void startup_vm(void);
264 static void startup_end(void);
265 static void setup_cage_params(void);
266 static void startup_create_io_node(void);
267
268 static pgcnt_t npages;
269 static struct memlist *memlist;
270 void *memlist_end;
271
272 static pgcnt_t bop_alloc_pages;
273 static caddr_t hblk_base;
274 uint_t hblk_alloc_dynamic = 0;
275 uint_t hblk1_min = H1MIN;
276
277
278 /*
279 * Hooks for unsupported platforms and down-rev firmware
280 */
281 int iam_positron(void);
282 #pragma weak iam_positron
283 static void do_prom_version_check(void);
284
285 /*
286 * After receiving a thermal interrupt, this is the number of seconds
287 * to delay before shutting off the system, assuming
288 * shutdown fails. Use /etc/system to change the delay if this isn't
289 * large enough.
290 */
291 int thermal_powerdown_delay = 1200;
292
293 /*
294 * Used to hold off page relocations into the cage until OBP has completed
295 * its boot-time handoff of its resources to the kernel.
296 */
297 int page_relocate_ready = 0;
298
299 /*
300 * Indicate if kmem64 allocation was done in small chunks
301 */
302 int kmem64_smchunks = 0;
303
304 /*
305 * Enable some debugging messages concerning memory usage...
306 */
307 #ifdef DEBUGGING_MEM
308 static int debugging_mem;
309 static void
310 printmemlist(char *title, struct memlist *list)
311 {
312 if (!debugging_mem)
313 return;
314
315 printf("%s\n", title);
316
317 while (list) {
318 prom_printf("\taddr = 0x%x %8x, size = 0x%x %8x\n",
319 (uint32_t)(list->ml_address >> 32),
320 (uint32_t)list->ml_address,
321 (uint32_t)(list->ml_size >> 32),
322 (uint32_t)(list->ml_size));
323 list = list->ml_next;
324 }
325 }
326
327 void
328 printmemseg(struct memseg *memseg)
329 {
330 if (!debugging_mem)
331 return;
332
333 printf("memseg\n");
334
335 while (memseg) {
336 prom_printf("\tpage = 0x%p, epage = 0x%p, "
337 "pfn = 0x%x, epfn = 0x%x\n",
338 memseg->pages, memseg->epages,
339 memseg->pages_base, memseg->pages_end);
340 memseg = memseg->next;
341 }
342 }
343
344 #define debug_pause(str) halt((str))
345 #define MPRINTF(str) if (debugging_mem) prom_printf((str))
346 #define MPRINTF1(str, a) if (debugging_mem) prom_printf((str), (a))
347 #define MPRINTF2(str, a, b) if (debugging_mem) prom_printf((str), (a), (b))
348 #define MPRINTF3(str, a, b, c) \
349 if (debugging_mem) prom_printf((str), (a), (b), (c))
350 #else /* DEBUGGING_MEM */
351 #define MPRINTF(str)
352 #define MPRINTF1(str, a)
353 #define MPRINTF2(str, a, b)
354 #define MPRINTF3(str, a, b, c)
355 #endif /* DEBUGGING_MEM */
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 * -|-----------------------|-
369 * | segzio | (base and size vary)
370 * 0xFFFFFE00.00000000 -|-----------------------|-
371 * | | Ultrasparc I/II support
372 * | segkpm segment | up to 2TB of physical
373 * | (64-bit kernel ONLY) | memory, VAC has 2 colors
374 * | |
375 * 0xFFFFFA00.00000000 -|-----------------------|- 2TB segkpm alignment
376 * : :
377 * : :
378 * 0xFFFFF810.00000000 -|-----------------------|- hole_end
379 * | | ^
380 * | UltraSPARC I/II call | |
381 * | bug requires an extra | |
382 * | 4 GB of space between | |
383 * | hole and used RAM | |
384 * | | |
385 * 0xFFFFF800.00000000 -|-----------------------|- |
386 * | | |
387 * | Virtual Address Hole | UltraSPARC
388 * | on UltraSPARC I/II | I/II * ONLY *
389 * | | |
390 * 0x00000800.00000000 -|-----------------------|- |
391 * | | |
392 * | UltraSPARC I/II call | |
393 * | bug requires an extra | |
394 * | 4 GB of space between | |
395 * | hole and used RAM | |
396 * | | v
397 * 0x000007FF.00000000 -|-----------------------|- hole_start -----
398 * : : ^
399 * : : |
400 * |-----------------------| |
401 * | | |
402 * | ecache flush area | |
403 * | (twice largest e$) | |
404 * | | |
405 * 0x00000XXX.XXX00000 -|-----------------------|- kmem64_ |
406 * | overmapped area | alignend_end |
407 * | (kmem64_alignsize | |
408 * | boundary) | |
409 * 0x00000XXX.XXXXXXXX -|-----------------------|- kmem64_end |
410 * | | |
411 * | 64-bit kernel ONLY | |
412 * | | |
413 * | kmem64 segment | |
414 * | | |
415 * | (Relocated extra HME | Approximately
416 * | block allocations, | 1 TB of virtual
417 * | memnode freelists, | address space
418 * | HME hash buckets, | |
419 * | mml_table, kpmp_table,| |
420 * | page_t array and | |
421 * | hashblock pool to | |
422 * | avoid hard-coded | |
423 * | 32-bit vaddr | |
424 * | limitations) | |
425 * | | v
426 * 0x00000700.00000000 -|-----------------------|- SYSLIMIT (kmem64_base)
427 * | |
428 * | segkmem segment | (SYSLIMIT - SYSBASE = 4TB)
429 * | |
430 * 0x00000300.00000000 -|-----------------------|- SYSBASE
431 * : :
432 * : :
433 * -|-----------------------|-
434 * | |
435 * | segmap segment | SEGMAPSIZE (1/8th physmem,
436 * | | 256G MAX)
437 * 0x000002a7.50000000 -|-----------------------|- SEGMAPBASE
438 * : :
439 * : :
440 * -|-----------------------|-
441 * | |
442 * | segkp | SEGKPSIZE (2GB)
443 * | |
444 * | |
445 * 0x000002a1.00000000 -|-----------------------|- SEGKPBASE
446 * | |
447 * 0x000002a0.00000000 -|-----------------------|- MEMSCRUBBASE
448 * | | (SEGKPBASE - 0x400000)
449 * 0x0000029F.FFE00000 -|-----------------------|- ARGSBASE
450 * | | (MEMSCRUBBASE - NCARGS)
451 * 0x0000029F.FFD80000 -|-----------------------|- PPMAPBASE
452 * | | (ARGSBASE - PPMAPSIZE)
453 * 0x0000029F.FFD00000 -|-----------------------|- PPMAP_FAST_BASE
454 * | |
455 * 0x0000029F.FF980000 -|-----------------------|- PIOMAPBASE
456 * | |
457 * 0x0000029F.FF580000 -|-----------------------|- NARG_BASE
458 * : :
459 * : :
460 * 0x00000000.FFFFFFFF -|-----------------------|- OFW_END_ADDR
461 * | |
462 * | OBP |
463 * | |
464 * 0x00000000.F0000000 -|-----------------------|- OFW_START_ADDR
465 * | kmdb |
466 * 0x00000000.EDD00000 -|-----------------------|- SEGDEBUGBASE
467 * : :
468 * : :
469 * 0x00000000.7c000000 -|-----------------------|- SYSLIMIT32
470 * | |
471 * | segkmem32 segment | (SYSLIMIT32 - SYSBASE32 =
472 * | | ~64MB)
473 * -|-----------------------|
474 * | IVSIZE |
475 * 0x00000000.70004000 -|-----------------------|
476 * | panicbuf |
477 * 0x00000000.70002000 -|-----------------------|
478 * | PAGESIZE |
479 * 0x00000000.70000000 -|-----------------------|- SYSBASE32
480 * | boot-time |
481 * | temporary space |
482 * 0x00000000.4C000000 -|-----------------------|- BOOTTMPBASE
483 * : :
484 * : :
485 * | |
486 * |-----------------------|- econtig32
487 * | vm structures |
488 * 0x00000000.01C00000 |-----------------------|- nalloc_end
489 * | TSBs |
490 * |-----------------------|- end/nalloc_base
491 * | kernel data & bss |
492 * 0x00000000.01800000 -|-----------------------|
493 * : nucleus text hole :
494 * 0x00000000.01400000 -|-----------------------|
495 * : :
496 * |-----------------------|
497 * | module text |
498 * |-----------------------|- e_text/modtext
499 * | kernel text |
500 * |-----------------------|
501 * | trap table (48k) |
502 * 0x00000000.01000000 -|-----------------------|- KERNELBASE
503 * | reserved for trapstat |} TSTAT_TOTAL_SIZE
504 * |-----------------------|
505 * | |
506 * | invalid |
507 * | |
508 * 0x00000000.00000000 _|_______________________|
509 *
510 *
511 *
512 * 32-bit User Virtual Memory Layout.
513 * /-----------------------\
514 * | |
515 * | invalid |
516 * | |
517 * 0xFFC00000 -|-----------------------|- USERLIMIT
518 * | user stack |
519 * : :
520 * : :
521 * : :
522 * | user data |
523 * -|-----------------------|-
524 * | user text |
525 * 0x00002000 -|-----------------------|-
526 * | invalid |
527 * 0x00000000 _|_______________________|
528 *
529 *
530 *
531 * 64-bit User Virtual Memory Layout.
532 * /-----------------------\
533 * | |
534 * | invalid |
535 * | |
536 * 0xFFFFFFFF.80000000 -|-----------------------|- USERLIMIT
537 * | user stack |
538 * : :
539 * : :
540 * : :
541 * | user data |
542 * -|-----------------------|-
543 * | user text |
544 * 0x00000000.01000000 -|-----------------------|-
545 * | invalid |
546 * 0x00000000.00000000 _|_______________________|
547 */
548
549 extern caddr_t ecache_init_scrub_flush_area(caddr_t alloc_base);
550 extern uint64_t ecache_flush_address(void);
551
552 #pragma weak load_platform_modules
553 #pragma weak plat_startup_memlist
554 #pragma weak ecache_init_scrub_flush_area
555 #pragma weak ecache_flush_address
556
557
558 /*
559 * By default the DR Cage is enabled for maximum OS
560 * MPSS performance. Users needing to disable the cage mechanism
561 * can set this variable to zero via /etc/system.
562 * Disabling the cage on systems supporting Dynamic Reconfiguration (DR)
563 * will result in loss of DR functionality.
564 * Platforms wishing to disable kernel Cage by default
565 * should do so in their set_platform_defaults() routine.
566 */
567 int kernel_cage_enable = 1;
568
569 static void
570 setup_cage_params(void)
571 {
572 void (*func)(void);
573
574 func = (void (*)(void))kobj_getsymvalue("set_platform_cage_params", 0);
575 if (func != NULL) {
576 (*func)();
577 return;
578 }
579
580 if (kernel_cage_enable == 0) {
581 return;
582 }
583 kcage_range_init(phys_avail, KCAGE_DOWN, total_pages / 256);
584
585 if (kcage_on) {
586 cmn_err(CE_NOTE, "!Kernel Cage is ENABLED");
587 } else {
588 cmn_err(CE_NOTE, "!Kernel Cage is DISABLED");
589 }
590
591 }
592
593 /*
594 * Machine-dependent startup code
595 */
596 void
597 startup(void)
598 {
599 startup_init();
600 if (&startup_platform)
601 startup_platform();
602 startup_memlist();
603 startup_modules();
604 setup_cage_params();
605 startup_bop_gone();
606 startup_vm();
607 startup_end();
608 }
609
610 struct regs sync_reg_buf;
611 uint64_t sync_tt;
612
613 void
614 sync_handler(void)
615 {
616 struct panic_trap_info ti;
617 int i;
618
619 /*
620 * Prevent trying to talk to the other CPUs since they are
621 * sitting in the prom and won't reply.
622 */
623 for (i = 0; i < NCPU; i++) {
624 if ((i != CPU->cpu_id) && CPU_XCALL_READY(i)) {
625 cpu[i]->cpu_flags &= ~CPU_READY;
626 cpu[i]->cpu_flags |= CPU_QUIESCED;
627 CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
628 }
629 }
630
631 /*
632 * Force a serial dump, since there are no CPUs to help.
633 */
634 dump_ncpu_low = 0;
635
636 /*
637 * We've managed to get here without going through the
638 * normal panic code path. Try and save some useful
639 * information.
640 */
641 if (!panicstr && (curthread->t_panic_trap == NULL)) {
642 ti.trap_type = sync_tt;
643 ti.trap_regs = &sync_reg_buf;
644 ti.trap_addr = NULL;
645 ti.trap_mmu_fsr = 0x0;
646
647 curthread->t_panic_trap = &ti;
648 }
649
650 /*
651 * If we're re-entering the panic path, update the signature
652 * block so that the SC knows we're in the second part of panic.
653 */
654 if (panicstr)
655 CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
656
657 nopanicdebug = 1; /* do not perform debug_enter() prior to dump */
658 panic("sync initiated");
659 }
660
661
662 static void
663 startup_init(void)
664 {
665 /*
666 * We want to save the registers while we're still in OBP
667 * so that we know they haven't been fiddled with since.
668 * (In principle, OBP can't change them just because it
669 * makes a callback, but we'd rather not depend on that
670 * behavior.)
671 */
672 char sync_str[] =
673 "warning @ warning off : sync "
674 "%%tl-c %%tstate h# %p x! "
675 "%%g1 h# %p x! %%g2 h# %p x! %%g3 h# %p x! "
676 "%%g4 h# %p x! %%g5 h# %p x! %%g6 h# %p x! "
677 "%%g7 h# %p x! %%o0 h# %p x! %%o1 h# %p x! "
678 "%%o2 h# %p x! %%o3 h# %p x! %%o4 h# %p x! "
679 "%%o5 h# %p x! %%o6 h# %p x! %%o7 h# %p x! "
680 "%%tl-c %%tpc h# %p x! %%tl-c %%tnpc h# %p x! "
681 "%%y h# %p l! %%tl-c %%tt h# %p x! "
682 "sync ; warning !";
683
684 /*
685 * 20 == num of %p substrings
686 * 16 == max num of chars %p will expand to.
687 */
688 char bp[sizeof (sync_str) + 16 * 20];
689
690 /*
691 * Initialize ptl1 stack for the 1st CPU.
692 */
693 ptl1_init_cpu(&cpu0);
694
695 /*
696 * Initialize the address map for cache consistent mappings
697 * to random pages; must be done after vac_size is set.
698 */
699 ppmapinit();
700
701 /*
702 * Initialize the PROM callback handler.
703 */
704 init_vx_handler();
705
706 /*
707 * have prom call sync_callback() to handle the sync and
708 * save some useful information which will be stored in the
709 * core file later.
710 */
711 (void) sprintf((char *)bp, sync_str,
712 (void *)&sync_reg_buf.r_tstate, (void *)&sync_reg_buf.r_g1,
713 (void *)&sync_reg_buf.r_g2, (void *)&sync_reg_buf.r_g3,
714 (void *)&sync_reg_buf.r_g4, (void *)&sync_reg_buf.r_g5,
715 (void *)&sync_reg_buf.r_g6, (void *)&sync_reg_buf.r_g7,
716 (void *)&sync_reg_buf.r_o0, (void *)&sync_reg_buf.r_o1,
717 (void *)&sync_reg_buf.r_o2, (void *)&sync_reg_buf.r_o3,
718 (void *)&sync_reg_buf.r_o4, (void *)&sync_reg_buf.r_o5,
719 (void *)&sync_reg_buf.r_o6, (void *)&sync_reg_buf.r_o7,
720 (void *)&sync_reg_buf.r_pc, (void *)&sync_reg_buf.r_npc,
721 (void *)&sync_reg_buf.r_y, (void *)&sync_tt);
722 prom_interpret(bp, 0, 0, 0, 0, 0);
723 add_vx_handler("sync", 1, (void (*)(cell_t *))sync_handler);
724 }
725
726
727 size_t
728 calc_pp_sz(pgcnt_t npages)
729 {
730
731 return (npages * sizeof (struct page));
732 }
733
734 size_t
735 calc_kpmpp_sz(pgcnt_t npages)
736 {
737
738 kpm_pgshft = (kpm_smallpages == 0) ? MMU_PAGESHIFT4M : MMU_PAGESHIFT;
739 kpm_pgsz = 1ull << kpm_pgshft;
740 kpm_pgoff = kpm_pgsz - 1;
741 kpmp2pshft = kpm_pgshft - PAGESHIFT;
742 kpmpnpgs = 1 << kpmp2pshft;
743
744 if (kpm_smallpages == 0) {
745 /*
746 * Avoid fragmentation problems in kphysm_init()
747 * by allocating for all of physical memory
748 */
749 kpm_npages = ptokpmpr(physinstalled);
750 return (kpm_npages * sizeof (kpm_page_t));
751 } else {
752 kpm_npages = npages;
753 return (kpm_npages * sizeof (kpm_spage_t));
754 }
755 }
756
757 size_t
758 calc_pagehash_sz(pgcnt_t npages)
759 {
760 /* LINTED */
761 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), (sizeof (struct page))));
762 /*
763 * The page structure hash table size is a power of 2
764 * such that the average hash chain length is PAGE_HASHAVELEN.
765 */
766 page_hashsz = npages / PAGE_HASHAVELEN;
767 page_hashsz_shift = MAX((AN_VPSHIFT + VNODE_ALIGN_LOG2 + 1),
768 highbit(page_hashsz));
769 page_hashsz = 1 << page_hashsz_shift;
770 return (page_hashsz * sizeof (struct page *));
771 }
772
773 int testkmem64_smchunks = 0;
774
775 int
776 alloc_kmem64(caddr_t base, caddr_t end)
777 {
778 int i;
779 caddr_t aligned_end = NULL;
780
781 if (testkmem64_smchunks)
782 return (1);
783
784 /*
785 * Make one large memory alloc after figuring out the 64-bit size. This
786 * will enable use of the largest page size appropriate for the system
787 * architecture.
788 */
789 ASSERT(mmu_exported_pagesize_mask & (1 << TTE8K));
790 ASSERT(IS_P2ALIGNED(base, TTEBYTES(max_bootlp_tteszc)));
791 for (i = max_bootlp_tteszc; i >= TTE8K; i--) {
792 size_t alloc_size, alignsize;
793 #if !defined(C_OBP)
794 unsigned long long pa;
795 #endif /* !C_OBP */
796
797 if ((mmu_exported_pagesize_mask & (1 << i)) == 0)
798 continue;
799 alignsize = TTEBYTES(i);
800 kmem64_szc = i;
801
802 /* limit page size for small memory */
803 if (mmu_btop(alignsize) > (npages >> 2))
804 continue;
805
806 aligned_end = (caddr_t)roundup((uintptr_t)end, alignsize);
807 alloc_size = aligned_end - base;
808 #if !defined(C_OBP)
809 if (prom_allocate_phys(alloc_size, alignsize, &pa) == 0) {
810 if (prom_claim_virt(alloc_size, base) != (caddr_t)-1) {
811 kmem64_pabase = pa;
812 kmem64_aligned_end = aligned_end;
813 install_kmem64_tte();
814 break;
815 } else {
816 prom_free_phys(alloc_size, pa);
817 }
818 }
819 #else /* !C_OBP */
820 if (prom_alloc(base, alloc_size, alignsize) == base) {
821 kmem64_pabase = va_to_pa(kmem64_base);
822 kmem64_aligned_end = aligned_end;
823 break;
824 }
825 #endif /* !C_OBP */
826 if (i == TTE8K) {
827 #ifdef sun4v
828 /* return failure to try small allocations */
829 return (1);
830 #else
831 prom_panic("kmem64 allocation failure");
832 #endif
833 }
834 }
835 ASSERT(aligned_end != NULL);
836 return (0);
837 }
838
839 static prom_memlist_t *boot_physinstalled, *boot_physavail, *boot_virtavail;
840 static size_t boot_physinstalled_len, boot_physavail_len, boot_virtavail_len;
841
842 #if !defined(C_OBP)
843 /*
844 * Install a temporary tte handler in OBP for kmem64 area.
845 *
846 * We map kmem64 area with large pages before the trap table is taken
847 * over. Since OBP makes 8K mappings, it can create 8K tlb entries in
848 * the same area. Duplicate tlb entries with different page sizes
849 * cause unpredicatble behavior. To avoid this, we don't create
850 * kmem64 mappings via BOP_ALLOC (ends up as prom_alloc() call to
851 * OBP). Instead, we manage translations with a temporary va>tte-data
852 * handler (kmem64-tte). This handler is replaced by unix-tte when
853 * the trap table is taken over.
854 *
855 * The temporary handler knows the physical address of the kmem64
856 * area. It uses the prom's pgmap@ Forth word for other addresses.
857 *
858 * We have to use BOP_ALLOC() method for C-OBP platforms because
859 * pgmap@ is not defined in C-OBP. C-OBP is only used on serengeti
860 * sun4u platforms. On sun4u we flush tlb after trap table is taken
861 * over if we use large pages for kernel heap and kmem64. Since sun4u
862 * prom (unlike sun4v) calls va>tte-data first for client address
863 * translation prom's ttes for kmem64 can't get into TLB even if we
864 * later switch to prom's trap table again. C-OBP uses 4M pages for
865 * client mappings when possible so on all platforms we get the
866 * benefit from large mappings for kmem64 area immediately during
867 * boot.
868 *
869 * pseudo code:
870 * if (context != 0) {
871 * return false
872 * } else if (miss_va in range[kmem64_base, kmem64_end)) {
873 * tte = tte_template +
874 * (((miss_va & pagemask) - kmem64_base));
875 * return tte, true
876 * } else {
877 * return pgmap@ result
878 * }
879 */
880 char kmem64_obp_str[] =
881 "h# %lx constant kmem64-base "
882 "h# %lx constant kmem64-end "
883 "h# %lx constant kmem64-pagemask "
884 "h# %lx constant kmem64-template "
885
886 ": kmem64-tte ( addr cnum -- false | tte-data true ) "
887 " if ( addr ) "
888 " drop false exit then ( false ) "
889 " dup kmem64-base kmem64-end within if ( addr ) "
890 " kmem64-pagemask and ( addr' ) "
891 " kmem64-base - ( addr' ) "
892 " kmem64-template + ( tte ) "
893 " true ( tte true ) "
894 " else ( addr ) "
895 " pgmap@ ( tte ) "
896 " dup 0< if true else drop false then ( tte true | false ) "
897 " then ( tte true | false ) "
898 "; "
899
900 "' kmem64-tte is va>tte-data "
901 ;
902
903 static void
904 install_kmem64_tte()
905 {
906 char b[sizeof (kmem64_obp_str) + (4 * 16)];
907 tte_t tte;
908
909 PRM_DEBUG(kmem64_pabase);
910 PRM_DEBUG(kmem64_szc);
911 sfmmu_memtte(&tte, kmem64_pabase >> MMU_PAGESHIFT,
912 PROC_DATA | HAT_NOSYNC, kmem64_szc);
913 PRM_DEBUG(tte.ll);
914 (void) sprintf(b, kmem64_obp_str,
915 kmem64_base, kmem64_end, TTE_PAGEMASK(kmem64_szc), tte.ll);
916 ASSERT(strlen(b) < sizeof (b));
917 prom_interpret(b, 0, 0, 0, 0, 0);
918 }
919 #endif /* !C_OBP */
920
921 /*
922 * As OBP takes up some RAM when the system boots, pages will already be "lost"
923 * to the system and reflected in npages by the time we see it.
924 *
925 * We only want to allocate kernel structures in the 64-bit virtual address
926 * space on systems with enough RAM to make the overhead of keeping track of
927 * an extra kernel memory segment worthwhile.
928 *
929 * Since OBP has already performed its memory allocations by this point, if we
930 * have more than MINMOVE_RAM_MB MB of RAM left free, go ahead and map
931 * memory in the 64-bit virtual address space; otherwise keep allocations
932 * contiguous with we've mapped so far in the 32-bit virtual address space.
933 */
934 #define MINMOVE_RAM_MB ((size_t)1900)
935 #define MB_TO_BYTES(mb) ((mb) * 1048576ul)
936 #define BYTES_TO_MB(b) ((b) / 1048576ul)
937
938 pgcnt_t tune_npages = (pgcnt_t)
939 (MB_TO_BYTES(MINMOVE_RAM_MB)/ (size_t)MMU_PAGESIZE);
940
941 #pragma weak page_set_colorequiv_arr_cpu
942 extern void page_set_colorequiv_arr_cpu(void);
943 extern void page_set_colorequiv_arr(void);
944
945 static pgcnt_t ramdisk_npages;
946 static struct memlist *old_phys_avail;
947
948 kcage_dir_t kcage_startup_dir = KCAGE_DOWN;
949
950 static void
951 startup_memlist(void)
952 {
953 size_t hmehash_sz, pagelist_sz, tt_sz;
954 size_t psetable_sz;
955 caddr_t alloc_base;
956 caddr_t memspace;
957 struct memlist *cur;
958 size_t syslimit = (size_t)SYSLIMIT;
959 size_t sysbase = (size_t)SYSBASE;
960
961 /*
962 * Initialize enough of the system to allow kmem_alloc to work by
963 * calling boot to allocate its memory until the time that
964 * kvm_init is completed. The page structs are allocated after
965 * rounding up end to the nearest page boundary; the memsegs are
966 * initialized and the space they use comes from the kernel heap.
967 * With appropriate initialization, they can be reallocated later
968 * to a size appropriate for the machine's configuration.
969 *
970 * At this point, memory is allocated for things that will never
971 * need to be freed, this used to be "valloced". This allows a
972 * savings as the pages don't need page structures to describe
973 * them because them will not be managed by the vm system.
974 */
975
976 /*
977 * We're loaded by boot with the following configuration (as
978 * specified in the sun4u/conf/Mapfile):
979 *
980 * text: 4 MB chunk aligned on a 4MB boundary
981 * data & bss: 4 MB chunk aligned on a 4MB boundary
982 *
983 * These two chunks will eventually be mapped by 2 locked 4MB
984 * ttes and will represent the nucleus of the kernel. This gives
985 * us some free space that is already allocated, some or all of
986 * which is made available to kernel module text.
987 *
988 * The free space in the data-bss chunk is used for nucleus
989 * allocatable data structures and we reserve it using the
990 * nalloc_base and nalloc_end variables. This space is currently
991 * being used for hat data structures required for tlb miss
992 * handling operations. We align nalloc_base to a l2 cache
993 * linesize because this is the line size the hardware uses to
994 * maintain cache coherency.
995 * 512K is carved out for module data.
996 */
997
998 moddata = (caddr_t)roundup((uintptr_t)e_data, MMU_PAGESIZE);
999 e_moddata = moddata + MODDATA;
1000 nalloc_base = e_moddata;
1001
1002 nalloc_end = (caddr_t)roundup((uintptr_t)nalloc_base, MMU_PAGESIZE4M);
1003 valloc_base = nalloc_base;
1004
1005 /*
1006 * Calculate the start of the data segment.
1007 */
1008 if (((uintptr_t)e_moddata & MMU_PAGEMASK4M) != (uintptr_t)s_data)
1009 prom_panic("nucleus data overflow");
1010
1011 PRM_DEBUG(moddata);
1012 PRM_DEBUG(nalloc_base);
1013 PRM_DEBUG(nalloc_end);
1014
1015 /*
1016 * Remember any slop after e_text so we can give it to the modules.
1017 */
1018 PRM_DEBUG(e_text);
1019 modtext = (caddr_t)roundup((uintptr_t)e_text, MMU_PAGESIZE);
1020 if (((uintptr_t)e_text & MMU_PAGEMASK4M) != (uintptr_t)s_text)
1021 prom_panic("nucleus text overflow");
1022 modtext_sz = (caddr_t)roundup((uintptr_t)modtext, MMU_PAGESIZE4M) -
1023 modtext;
1024 PRM_DEBUG(modtext);
1025 PRM_DEBUG(modtext_sz);
1026
1027 init_boot_memlists();
1028 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1029 &boot_physavail, &boot_physavail_len,
1030 &boot_virtavail, &boot_virtavail_len);
1031
1032 /*
1033 * Remember what the physically available highest page is
1034 * so that dumpsys works properly, and find out how much
1035 * memory is installed.
1036 */
1037 installed_top_size_memlist_array(boot_physinstalled,
1038 boot_physinstalled_len, &physmax, &physinstalled);
1039 PRM_DEBUG(physinstalled);
1040 PRM_DEBUG(physmax);
1041
1042 /* Fill out memory nodes config structure */
1043 startup_build_mem_nodes(boot_physinstalled, boot_physinstalled_len);
1044
1045 /*
1046 * npages is the maximum of available physical memory possible.
1047 * (ie. it will never be more than this)
1048 *
1049 * When we boot from a ramdisk, the ramdisk memory isn't free, so
1050 * using phys_avail will underestimate what will end up being freed.
1051 * A better initial guess is just total memory minus the kernel text
1052 */
1053 npages = physinstalled - btop(MMU_PAGESIZE4M);
1054
1055 /*
1056 * First allocate things that can go in the nucleus data page
1057 * (fault status, TSBs, dmv, CPUs)
1058 */
1059 ndata_alloc_init(&ndata, (uintptr_t)nalloc_base, (uintptr_t)nalloc_end);
1060
1061 if ((&ndata_alloc_mmfsa != NULL) && (ndata_alloc_mmfsa(&ndata) != 0))
1062 cmn_err(CE_PANIC, "no more nucleus memory after mfsa alloc");
1063
1064 if (ndata_alloc_tsbs(&ndata, npages) != 0)
1065 cmn_err(CE_PANIC, "no more nucleus memory after tsbs alloc");
1066
1067 if (ndata_alloc_dmv(&ndata) != 0)
1068 cmn_err(CE_PANIC, "no more nucleus memory after dmv alloc");
1069
1070 if (ndata_alloc_page_mutexs(&ndata) != 0)
1071 cmn_err(CE_PANIC,
1072 "no more nucleus memory after page free lists alloc");
1073
1074 if (ndata_alloc_hat(&ndata) != 0)
1075 cmn_err(CE_PANIC, "no more nucleus memory after hat alloc");
1076
1077 if (ndata_alloc_memseg(&ndata, boot_physavail_len) != 0)
1078 cmn_err(CE_PANIC, "no more nucleus memory after memseg alloc");
1079
1080 /*
1081 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1082 *
1083 * There are comments all over the SFMMU code warning of dire
1084 * consequences if the TSBs are moved out of 32-bit space. This
1085 * is largely because the asm code uses "sethi %hi(addr)"-type
1086 * instructions which will not provide the expected result if the
1087 * address is a 64-bit one.
1088 *
1089 * WARNING WARNING WARNING WARNING WARNING WARNING WARNING
1090 */
1091 alloc_base = (caddr_t)roundup((uintptr_t)nalloc_end, MMU_PAGESIZE);
1092 PRM_DEBUG(alloc_base);
1093
1094 alloc_base = sfmmu_ktsb_alloc(alloc_base);
1095 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1096 PRM_DEBUG(alloc_base);
1097
1098 /*
1099 * Allocate IOMMU TSB array. We do this here so that the physical
1100 * memory gets deducted from the PROM's physical memory list.
1101 */
1102 alloc_base = iommu_tsb_init(alloc_base);
1103 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1104 PRM_DEBUG(alloc_base);
1105
1106 /*
1107 * Allow for an early allocation of physically contiguous memory.
1108 */
1109 alloc_base = contig_mem_prealloc(alloc_base, npages);
1110
1111 /*
1112 * Platforms like Starcat and OPL need special structures assigned in
1113 * 32-bit virtual address space because their probing routines execute
1114 * FCode, and FCode can't handle 64-bit virtual addresses...
1115 */
1116 if (&plat_startup_memlist) {
1117 alloc_base = plat_startup_memlist(alloc_base);
1118 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base,
1119 ecache_alignsize);
1120 PRM_DEBUG(alloc_base);
1121 }
1122
1123 /*
1124 * Save off where the contiguous allocations to date have ended
1125 * in econtig32.
1126 */
1127 econtig32 = alloc_base;
1128 PRM_DEBUG(econtig32);
1129 if (econtig32 > (caddr_t)KERNEL_LIMIT32)
1130 cmn_err(CE_PANIC, "econtig32 too big");
1131
1132 pp_sz = calc_pp_sz(npages);
1133 PRM_DEBUG(pp_sz);
1134 if (kpm_enable) {
1135 kpm_pp_sz = calc_kpmpp_sz(npages);
1136 PRM_DEBUG(kpm_pp_sz);
1137 }
1138
1139 hmehash_sz = calc_hmehash_sz(npages);
1140 PRM_DEBUG(hmehash_sz);
1141
1142 pagehash_sz = calc_pagehash_sz(npages);
1143 PRM_DEBUG(pagehash_sz);
1144
1145 pagelist_sz = calc_free_pagelist_sz();
1146 PRM_DEBUG(pagelist_sz);
1147
1148 #ifdef TRAPTRACE
1149 tt_sz = calc_traptrace_sz();
1150 PRM_DEBUG(tt_sz);
1151 #else
1152 tt_sz = 0;
1153 #endif /* TRAPTRACE */
1154
1155 /*
1156 * Place the array that protects pp->p_selock in the kmem64 wad.
1157 */
1158 pse_shift = size_pse_array(npages, max_ncpus);
1159 PRM_DEBUG(pse_shift);
1160 pse_table_size = 1 << pse_shift;
1161 PRM_DEBUG(pse_table_size);
1162 psetable_sz = roundup(
1163 pse_table_size * sizeof (pad_mutex_t), ecache_alignsize);
1164 PRM_DEBUG(psetable_sz);
1165
1166 /*
1167 * Now allocate the whole wad
1168 */
1169 kmem64_sz = pp_sz + kpm_pp_sz + hmehash_sz + pagehash_sz +
1170 pagelist_sz + tt_sz + psetable_sz;
1171 kmem64_sz = roundup(kmem64_sz, PAGESIZE);
1172 kmem64_base = (caddr_t)syslimit;
1173 kmem64_end = kmem64_base + kmem64_sz;
1174 if (alloc_kmem64(kmem64_base, kmem64_end)) {
1175 /*
1176 * Attempt for kmem64 to allocate one big
1177 * contiguous chunk of memory failed.
1178 * We get here because we are sun4v.
1179 * We will proceed by breaking up
1180 * the allocation into two attempts.
1181 * First, we allocate kpm_pp_sz, hmehash_sz,
1182 * pagehash_sz, pagelist_sz, tt_sz & psetable_sz as
1183 * one contiguous chunk. This is a much smaller
1184 * chunk and we should get it, if not we panic.
1185 * Note that hmehash and tt need to be physically
1186 * (in the real address sense) contiguous.
1187 * Next, we use bop_alloc_chunk() to
1188 * to allocate the page_t structures.
1189 * This will allow the page_t to be allocated
1190 * in multiple smaller chunks.
1191 * In doing so, the assumption that page_t is
1192 * physically contiguous no longer hold, this is ok
1193 * for sun4v but not for sun4u.
1194 */
1195 size_t tmp_size;
1196 caddr_t tmp_base;
1197
1198 pp_sz = roundup(pp_sz, PAGESIZE);
1199
1200 /*
1201 * Allocate kpm_pp_sz, hmehash_sz,
1202 * pagehash_sz, pagelist_sz, tt_sz & psetable_sz
1203 */
1204 tmp_base = kmem64_base + pp_sz;
1205 tmp_size = roundup(kpm_pp_sz + hmehash_sz + pagehash_sz +
1206 pagelist_sz + tt_sz + psetable_sz, PAGESIZE);
1207 if (prom_alloc(tmp_base, tmp_size, PAGESIZE) == 0)
1208 prom_panic("kmem64 prom_alloc contig failed");
1209 PRM_DEBUG(tmp_base);
1210 PRM_DEBUG(tmp_size);
1211
1212 /*
1213 * Allocate the page_ts
1214 */
1215 if (bop_alloc_chunk(kmem64_base, pp_sz, PAGESIZE) == 0)
1216 prom_panic("kmem64 bop_alloc_chunk page_t failed");
1217 PRM_DEBUG(kmem64_base);
1218 PRM_DEBUG(pp_sz);
1219
1220 kmem64_aligned_end = kmem64_base + pp_sz + tmp_size;
1221 ASSERT(kmem64_aligned_end >= kmem64_end);
1222
1223 kmem64_smchunks = 1;
1224 } else {
1225
1226 /*
1227 * We need to adjust pp_sz for the normal
1228 * case where kmem64 can allocate one large chunk
1229 */
1230 if (kpm_smallpages == 0) {
1231 npages -= kmem64_sz / (PAGESIZE + sizeof (struct page));
1232 } else {
1233 npages -= kmem64_sz / (PAGESIZE + sizeof (struct page) +
1234 sizeof (kpm_spage_t));
1235 }
1236 pp_sz = npages * sizeof (struct page);
1237 }
1238
1239 if (kmem64_aligned_end > (hole_start ? hole_start : kpm_vbase))
1240 cmn_err(CE_PANIC, "not enough kmem64 space");
1241 PRM_DEBUG(kmem64_base);
1242 PRM_DEBUG(kmem64_end);
1243 PRM_DEBUG(kmem64_aligned_end);
1244
1245 /*
1246 * ... and divy it up
1247 */
1248 alloc_base = kmem64_base;
1249
1250 pp_base = (page_t *)alloc_base;
1251 alloc_base += pp_sz;
1252 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1253 PRM_DEBUG(pp_base);
1254 PRM_DEBUG(npages);
1255
1256 if (kpm_enable) {
1257 kpm_pp_base = alloc_base;
1258 if (kpm_smallpages == 0) {
1259 /* kpm_npages based on physinstalled, don't reset */
1260 kpm_pp_sz = kpm_npages * sizeof (kpm_page_t);
1261 } else {
1262 kpm_npages = ptokpmpr(npages);
1263 kpm_pp_sz = kpm_npages * sizeof (kpm_spage_t);
1264 }
1265 alloc_base += kpm_pp_sz;
1266 alloc_base =
1267 (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1268 PRM_DEBUG(kpm_pp_base);
1269 }
1270
1271 alloc_base = alloc_hmehash(alloc_base);
1272 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1273 PRM_DEBUG(alloc_base);
1274
1275 page_hash = (page_t **)alloc_base;
1276 alloc_base += pagehash_sz;
1277 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1278 PRM_DEBUG(page_hash);
1279
1280 alloc_base = alloc_page_freelists(alloc_base);
1281 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1282 PRM_DEBUG(alloc_base);
1283
1284 #ifdef TRAPTRACE
1285 ttrace_buf = alloc_base;
1286 alloc_base += tt_sz;
1287 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1288 PRM_DEBUG(alloc_base);
1289 #endif /* TRAPTRACE */
1290
1291 pse_mutex = (pad_mutex_t *)alloc_base;
1292 alloc_base += psetable_sz;
1293 alloc_base = (caddr_t)roundup((uintptr_t)alloc_base, ecache_alignsize);
1294 PRM_DEBUG(alloc_base);
1295
1296 /*
1297 * Note that if we use small chunk allocations for
1298 * kmem64, we need to ensure kmem64_end is the same as
1299 * kmem64_aligned_end to prevent subsequent logic from
1300 * trying to reuse the overmapping.
1301 * Otherwise we adjust kmem64_end to what we really allocated.
1302 */
1303 if (kmem64_smchunks) {
1304 kmem64_end = kmem64_aligned_end;
1305 } else {
1306 kmem64_end = (caddr_t)roundup((uintptr_t)alloc_base, PAGESIZE);
1307 }
1308 kmem64_sz = kmem64_end - kmem64_base;
1309
1310 if (&ecache_init_scrub_flush_area) {
1311 alloc_base = ecache_init_scrub_flush_area(kmem64_aligned_end);
1312 ASSERT(alloc_base <= (hole_start ? hole_start : kpm_vbase));
1313 }
1314
1315 /*
1316 * If physmem is patched to be non-zero, use it instead of
1317 * the monitor value unless physmem is larger than the total
1318 * amount of memory on hand.
1319 */
1320 if (physmem == 0 || physmem > npages)
1321 physmem = npages;
1322
1323 /*
1324 * root_is_ramdisk is set via /etc/system when the ramdisk miniroot
1325 * is mounted as root. This memory is held down by OBP and unlike
1326 * the stub boot_archive is never released.
1327 *
1328 * In order to get things sized correctly on lower memory
1329 * machines (where the memory used by the ramdisk represents
1330 * a significant portion of memory), physmem is adjusted.
1331 *
1332 * This is done by subtracting the ramdisk_size which is set
1333 * to the size of the ramdisk (in Kb) in /etc/system at the
1334 * time the miniroot archive is constructed.
1335 */
1336 if (root_is_ramdisk == B_TRUE) {
1337 ramdisk_npages = (ramdisk_size * 1024) / PAGESIZE;
1338 physmem -= ramdisk_npages;
1339 }
1340
1341 if (kpm_enable && (ndata_alloc_kpm(&ndata, kpm_npages) != 0))
1342 cmn_err(CE_PANIC, "no more nucleus memory after kpm alloc");
1343
1344 /*
1345 * Allocate space for the interrupt vector table.
1346 */
1347 memspace = prom_alloc((caddr_t)intr_vec_table, IVSIZE, MMU_PAGESIZE);
1348 if (memspace != (caddr_t)intr_vec_table)
1349 prom_panic("interrupt vector table allocation failure");
1350
1351 /*
1352 * Between now and when we finish copying in the memory lists,
1353 * allocations happen so the space gets fragmented and the
1354 * lists longer. Leave enough space for lists twice as
1355 * long as we have now; then roundup to a pagesize.
1356 */
1357 memlist_sz = sizeof (struct memlist) * (prom_phys_installed_len() +
1358 prom_phys_avail_len() + prom_virt_avail_len());
1359 memlist_sz *= 2;
1360 memlist_sz = roundup(memlist_sz, PAGESIZE);
1361 memspace = ndata_alloc(&ndata, memlist_sz, ecache_alignsize);
1362 if (memspace == NULL)
1363 cmn_err(CE_PANIC, "no more nucleus memory after memlist alloc");
1364
1365 memlist = (struct memlist *)memspace;
1366 memlist_end = (char *)memspace + memlist_sz;
1367 PRM_DEBUG(memlist);
1368 PRM_DEBUG(memlist_end);
1369
1370 PRM_DEBUG(sysbase);
1371 PRM_DEBUG(syslimit);
1372 kernelheap_init((void *)sysbase, (void *)syslimit,
1373 (caddr_t)sysbase + PAGESIZE, NULL, NULL);
1374
1375 /*
1376 * Take the most current snapshot we can by calling mem-update.
1377 */
1378 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1379 &boot_physavail, &boot_physavail_len,
1380 &boot_virtavail, &boot_virtavail_len);
1381
1382 /*
1383 * Remove the space used by prom_alloc from the kernel heap
1384 * plus the area actually used by the OBP (if any)
1385 * ignoring virtual addresses in virt_avail, above syslimit.
1386 */
1387 virt_avail = memlist;
1388 copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1389
1390 for (cur = virt_avail; cur->ml_next; cur = cur->ml_next) {
1391 uint64_t range_base, range_size;
1392
1393 if ((range_base = cur->ml_address + cur->ml_size) <
1394 (uint64_t)sysbase)
1395 continue;
1396 if (range_base >= (uint64_t)syslimit)
1397 break;
1398 /*
1399 * Limit the range to end at syslimit.
1400 */
1401 range_size = MIN(cur->ml_next->ml_address,
1402 (uint64_t)syslimit) - range_base;
1403 (void) vmem_xalloc(heap_arena, (size_t)range_size, PAGESIZE,
1404 0, 0, (void *)range_base, (void *)(range_base + range_size),
1405 VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1406 }
1407
1408 phys_avail = memlist;
1409 copy_memlist(boot_physavail, boot_physavail_len, &memlist);
1410
1411 /*
1412 * Add any extra memory at the end of the ndata region if there's at
1413 * least a page to add. There might be a few more pages available in
1414 * the middle of the ndata region, but for now they are ignored.
1415 */
1416 nalloc_base = ndata_extra_base(&ndata, MMU_PAGESIZE, nalloc_end);
1417 if (nalloc_base == NULL)
1418 nalloc_base = nalloc_end;
1419 ndata_remain_sz = nalloc_end - nalloc_base;
1420
1421 /*
1422 * Copy physinstalled list into kernel space.
1423 */
1424 phys_install = memlist;
1425 copy_memlist(boot_physinstalled, boot_physinstalled_len, &memlist);
1426
1427 /*
1428 * Create list of physical addrs we don't need pp's for:
1429 * kernel text 4M page
1430 * kernel data 4M page - ndata_remain_sz
1431 * kmem64 pages
1432 *
1433 * NB if adding any pages here, make sure no kpm page
1434 * overlaps can occur (see ASSERTs in kphysm_memsegs)
1435 */
1436 nopp_list = memlist;
1437 memlist_new(va_to_pa(s_text), MMU_PAGESIZE4M, &memlist);
1438 memlist_add(va_to_pa(s_data), MMU_PAGESIZE4M - ndata_remain_sz,
1439 &memlist, &nopp_list);
1440
1441 /* Don't add to nopp_list if kmem64 was allocated in smchunks */
1442 if (!kmem64_smchunks)
1443 memlist_add(kmem64_pabase, kmem64_sz, &memlist, &nopp_list);
1444
1445 if ((caddr_t)memlist > (memspace + memlist_sz))
1446 prom_panic("memlist overflow");
1447
1448 /*
1449 * Size the pcf array based on the number of cpus in the box at
1450 * boot time.
1451 */
1452 pcf_init();
1453
1454 /*
1455 * Initialize the page structures from the memory lists.
1456 */
1457 kphysm_init();
1458
1459 availrmem_initial = availrmem = freemem;
1460 PRM_DEBUG(availrmem);
1461
1462 /*
1463 * Some of the locks depend on page_hashsz being set!
1464 * kmem_init() depends on this; so, keep it here.
1465 */
1466 page_lock_init();
1467
1468 /*
1469 * Initialize kernel memory allocator.
1470 */
1471 kmem_init();
1472
1473 /*
1474 * Factor in colorequiv to check additional 'equivalent' bins
1475 */
1476 if (&page_set_colorequiv_arr_cpu != NULL)
1477 page_set_colorequiv_arr_cpu();
1478 else
1479 page_set_colorequiv_arr();
1480
1481 /*
1482 * Initialize bp_mapin().
1483 */
1484 bp_init(shm_alignment, HAT_STRICTORDER);
1485
1486 /*
1487 * Reserve space for MPO mblock structs from the 32-bit heap.
1488 */
1489
1490 if (mpo_heap32_bufsz > (size_t)0) {
1491 (void) vmem_xalloc(heap32_arena, mpo_heap32_bufsz,
1492 PAGESIZE, 0, 0, mpo_heap32_buf,
1493 mpo_heap32_buf + mpo_heap32_bufsz,
1494 VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1495 }
1496 mem_config_init();
1497 }
1498
1499 static void
1500 startup_modules(void)
1501 {
1502 int nhblk1, nhblk8;
1503 size_t nhblksz;
1504 pgcnt_t pages_per_hblk;
1505 size_t hme8blk_sz, hme1blk_sz;
1506
1507 /*
1508 * The system file /etc/system was read already under startup_memlist.
1509 */
1510 if (&set_platform_defaults)
1511 set_platform_defaults();
1512
1513 /*
1514 * Calculate default settings of system parameters based upon
1515 * maxusers, yet allow to be overridden via the /etc/system file.
1516 */
1517 param_calc(0);
1518
1519 mod_setup();
1520
1521 /*
1522 * If this is a positron, complain and halt.
1523 */
1524 if (&iam_positron && iam_positron()) {
1525 cmn_err(CE_WARN, "This hardware platform is not supported"
1526 " by this release of Solaris.\n");
1527 #ifdef DEBUG
1528 prom_enter_mon(); /* Type 'go' to resume */
1529 cmn_err(CE_WARN, "Booting an unsupported platform.\n");
1530 cmn_err(CE_WARN, "Booting with down-rev firmware.\n");
1531
1532 #else /* DEBUG */
1533 halt(0);
1534 #endif /* DEBUG */
1535 }
1536
1537 /*
1538 * If we are running firmware that isn't 64-bit ready
1539 * then complain and halt.
1540 */
1541 do_prom_version_check();
1542
1543 /*
1544 * Initialize system parameters
1545 */
1546 param_init();
1547
1548 /*
1549 * maxmem is the amount of physical memory we're playing with.
1550 */
1551 maxmem = physmem;
1552
1553 /* Set segkp limits. */
1554 ncbase = kdi_segdebugbase;
1555 ncend = kdi_segdebugbase;
1556
1557 /*
1558 * Initialize the hat layer.
1559 */
1560 hat_init();
1561
1562 /*
1563 * Initialize segment management stuff.
1564 */
1565 seg_init();
1566
1567 /*
1568 * Create the va>tte handler, so the prom can understand
1569 * kernel translations. The handler is installed later, just
1570 * as we are about to take over the trap table from the prom.
1571 */
1572 create_va_to_tte();
1573
1574 /*
1575 * Load the forthdebugger (optional)
1576 */
1577 forthdebug_init();
1578
1579 /*
1580 * Create OBP node for console input callbacks
1581 * if it is needed.
1582 */
1583 startup_create_io_node();
1584
1585 if (modloadonly("fs", "specfs") == -1)
1586 halt("Can't load specfs");
1587
1588 if (modloadonly("fs", "devfs") == -1)
1589 halt("Can't load devfs");
1590
1591 if (modloadonly("fs", "procfs") == -1)
1592 halt("Can't load procfs");
1593
1594 if (modloadonly("misc", "swapgeneric") == -1)
1595 halt("Can't load swapgeneric");
1596
1597 (void) modloadonly("sys", "lbl_edition");
1598
1599 dispinit();
1600
1601 /*
1602 * Infer meanings to the members of the idprom buffer.
1603 */
1604 parse_idprom();
1605
1606 /* Read cluster configuration data. */
1607 clconf_init();
1608
1609 setup_ddi();
1610
1611 /*
1612 * Lets take this opportunity to load the root device.
1613 */
1614 if (loadrootmodules() != 0)
1615 debug_enter("Can't load the root filesystem");
1616
1617 /*
1618 * Load tod driver module for the tod part found on this system.
1619 * Recompute the cpu frequency/delays based on tod as tod part
1620 * tends to keep time more accurately.
1621 */
1622 if (&load_tod_module)
1623 load_tod_module();
1624
1625 /*
1626 * Allow platforms to load modules which might
1627 * be needed after bootops are gone.
1628 */
1629 if (&load_platform_modules)
1630 load_platform_modules();
1631
1632 setcpudelay();
1633
1634 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1635 &boot_physavail, &boot_physavail_len,
1636 &boot_virtavail, &boot_virtavail_len);
1637
1638 /*
1639 * Calculation and allocation of hmeblks needed to remap
1640 * the memory allocated by PROM till now.
1641 * Overestimate the number of hblk1 elements by assuming
1642 * worst case of TTE64K mappings.
1643 * sfmmu_hblk_alloc will panic if this calculation is wrong.
1644 */
1645 bop_alloc_pages = btopr(kmem64_end - kmem64_base);
1646 pages_per_hblk = btop(HMEBLK_SPAN(TTE64K));
1647 bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1648 nhblk1 = bop_alloc_pages / pages_per_hblk + hblk1_min;
1649
1650 bop_alloc_pages = size_virtalloc(boot_virtavail, boot_virtavail_len);
1651
1652 /* sfmmu_init_nucleus_hblks expects properly aligned data structures */
1653 hme8blk_sz = roundup(HME8BLK_SZ, sizeof (int64_t));
1654 hme1blk_sz = roundup(HME1BLK_SZ, sizeof (int64_t));
1655
1656 bop_alloc_pages += btopr(nhblk1 * hme1blk_sz);
1657
1658 pages_per_hblk = btop(HMEBLK_SPAN(TTE8K));
1659 nhblk8 = 0;
1660 while (bop_alloc_pages > 1) {
1661 bop_alloc_pages = roundup(bop_alloc_pages, pages_per_hblk);
1662 nhblk8 += bop_alloc_pages /= pages_per_hblk;
1663 bop_alloc_pages *= hme8blk_sz;
1664 bop_alloc_pages = btopr(bop_alloc_pages);
1665 }
1666 nhblk8 += 2;
1667
1668 /*
1669 * Since hblk8's can hold up to 64k of mappings aligned on a 64k
1670 * boundary, the number of hblk8's needed to map the entries in the
1671 * boot_virtavail list needs to be adjusted to take this into
1672 * consideration. Thus, we need to add additional hblk8's since it
1673 * is possible that an hblk8 will not have all 8 slots used due to
1674 * alignment constraints. Since there were boot_virtavail_len entries
1675 * in that list, we need to add that many hblk8's to the number
1676 * already calculated to make sure we don't underestimate.
1677 */
1678 nhblk8 += boot_virtavail_len;
1679 nhblksz = nhblk8 * hme8blk_sz + nhblk1 * hme1blk_sz;
1680
1681 /* Allocate in pagesize chunks */
1682 nhblksz = roundup(nhblksz, MMU_PAGESIZE);
1683 hblk_base = kmem_zalloc(nhblksz, KM_SLEEP);
1684 sfmmu_init_nucleus_hblks(hblk_base, nhblksz, nhblk8, nhblk1);
1685 }
1686
1687 static void
1688 startup_bop_gone(void)
1689 {
1690
1691 /*
1692 * Destroy the MD initialized at startup
1693 * The startup initializes the MD framework
1694 * using prom and BOP alloc free it now.
1695 */
1696 mach_descrip_startup_fini();
1697
1698 /*
1699 * We're done with prom allocations.
1700 */
1701 bop_fini();
1702
1703 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1704 &boot_physavail, &boot_physavail_len,
1705 &boot_virtavail, &boot_virtavail_len);
1706
1707 /*
1708 * setup physically contiguous area twice as large as the ecache.
1709 * this is used while doing displacement flush of ecaches
1710 */
1711 if (&ecache_flush_address) {
1712 ecache_flushaddr = ecache_flush_address();
1713 if (ecache_flushaddr == (uint64_t)-1) {
1714 cmn_err(CE_PANIC,
1715 "startup: no memory to set ecache_flushaddr");
1716 }
1717 }
1718
1719 /*
1720 * Virtual available next.
1721 */
1722 ASSERT(virt_avail != NULL);
1723 memlist_free_list(virt_avail);
1724 virt_avail = memlist;
1725 copy_memlist(boot_virtavail, boot_virtavail_len, &memlist);
1726
1727 }
1728
1729
1730 /*
1731 * startup_fixup_physavail - called from mach_sfmmu.c after the final
1732 * allocations have been performed. We can't call it in startup_bop_gone
1733 * since later operations can cause obp to allocate more memory.
1734 */
1735 void
1736 startup_fixup_physavail(void)
1737 {
1738 struct memlist *cur;
1739 size_t kmem64_overmap_size = kmem64_aligned_end - kmem64_end;
1740
1741 PRM_DEBUG(kmem64_overmap_size);
1742
1743 /*
1744 * take the most current snapshot we can by calling mem-update
1745 */
1746 copy_boot_memlists(&boot_physinstalled, &boot_physinstalled_len,
1747 &boot_physavail, &boot_physavail_len,
1748 &boot_virtavail, &boot_virtavail_len);
1749
1750 /*
1751 * Copy phys_avail list, again.
1752 * Both the kernel/boot and the prom have been allocating
1753 * from the original list we copied earlier.
1754 */
1755 cur = memlist;
1756 copy_memlist(boot_physavail, boot_physavail_len, &memlist);
1757
1758 /*
1759 * Add any unused kmem64 memory from overmapped page
1760 * (Note: va_to_pa does not work for kmem64_end)
1761 */
1762 if (kmem64_overmap_size) {
1763 memlist_add(kmem64_pabase + (kmem64_end - kmem64_base),
1764 kmem64_overmap_size, &memlist, &cur);
1765 }
1766
1767 /*
1768 * Add any extra memory after e_data we added to the phys_avail list
1769 * back to the old list.
1770 */
1771 if (ndata_remain_sz >= MMU_PAGESIZE)
1772 memlist_add(va_to_pa(nalloc_base),
1773 (uint64_t)ndata_remain_sz, &memlist, &cur);
1774
1775 /*
1776 * There isn't any bounds checking on the memlist area
1777 * so ensure it hasn't overgrown.
1778 */
1779 if ((caddr_t)memlist > (caddr_t)memlist_end)
1780 cmn_err(CE_PANIC, "startup: memlist size exceeded");
1781
1782 /*
1783 * The kernel removes the pages that were allocated for it from
1784 * the freelist, but we now have to find any -extra- pages that
1785 * the prom has allocated for it's own book-keeping, and remove
1786 * them from the freelist too. sigh.
1787 */
1788 sync_memlists(phys_avail, cur);
1789
1790 ASSERT(phys_avail != NULL);
1791
1792 old_phys_avail = phys_avail;
1793 phys_avail = cur;
1794 }
1795
1796 void
1797 update_kcage_ranges(uint64_t addr, uint64_t len)
1798 {
1799 pfn_t base = btop(addr);
1800 pgcnt_t num = btop(len);
1801 int rv;
1802
1803 rv = kcage_range_add(base, num, kcage_startup_dir);
1804
1805 if (rv == ENOMEM) {
1806 cmn_err(CE_WARN, "%ld megabytes not available to kernel cage",
1807 (len == 0 ? 0 : BYTES_TO_MB(len)));
1808 } else if (rv != 0) {
1809 /* catch this in debug kernels */
1810 ASSERT(0);
1811
1812 cmn_err(CE_WARN, "unexpected kcage_range_add"
1813 " return value %d", rv);
1814 }
1815 }
1816
1817 static void
1818 startup_vm(void)
1819 {
1820 size_t i;
1821 struct segmap_crargs a;
1822 struct segkpm_crargs b;
1823
1824 uint64_t avmem;
1825 caddr_t va;
1826 pgcnt_t max_phys_segkp;
1827 int mnode;
1828
1829 extern int use_brk_lpg, use_stk_lpg;
1830
1831 /*
1832 * get prom's mappings, create hments for them and switch
1833 * to the kernel context.
1834 */
1835 hat_kern_setup();
1836
1837 /*
1838 * Take over trap table
1839 */
1840 setup_trap_table();
1841
1842 /*
1843 * Install the va>tte handler, so that the prom can handle
1844 * misses and understand the kernel table layout in case
1845 * we need call into the prom.
1846 */
1847 install_va_to_tte();
1848
1849 /*
1850 * Set a flag to indicate that the tba has been taken over.
1851 */
1852 tba_taken_over = 1;
1853
1854 /* initialize MMU primary context register */
1855 mmu_init_kcontext();
1856
1857 /*
1858 * The boot cpu can now take interrupts, x-calls, x-traps
1859 */
1860 CPUSET_ADD(cpu_ready_set, CPU->cpu_id);
1861 CPU->cpu_flags |= (CPU_READY | CPU_ENABLE | CPU_EXISTS);
1862
1863 /*
1864 * Set a flag to tell write_scb_int() that it can access V_TBR_WR_ADDR.
1865 */
1866 tbr_wr_addr_inited = 1;
1867
1868 /*
1869 * Initialize VM system, and map kernel address space.
1870 */
1871 kvm_init();
1872
1873 ASSERT(old_phys_avail != NULL && phys_avail != NULL);
1874 if (kernel_cage_enable) {
1875 diff_memlists(phys_avail, old_phys_avail, update_kcage_ranges);
1876 }
1877 memlist_free_list(old_phys_avail);
1878
1879 /*
1880 * If the following is true, someone has patched
1881 * phsymem to be less than the number of pages that
1882 * the system actually has. Remove pages until system
1883 * memory is limited to the requested amount. Since we
1884 * have allocated page structures for all pages, we
1885 * correct the amount of memory we want to remove
1886 * by the size of the memory used to hold page structures
1887 * for the non-used pages.
1888 */
1889 if (physmem + ramdisk_npages < npages) {
1890 pgcnt_t diff, off;
1891 struct page *pp;
1892 struct seg kseg;
1893
1894 cmn_err(CE_WARN, "limiting physmem to %ld pages", physmem);
1895
1896 off = 0;
1897 diff = npages - (physmem + ramdisk_npages);
1898 diff -= mmu_btopr(diff * sizeof (struct page));
1899 kseg.s_as = &kas;
1900 while (diff--) {
1901 pp = page_create_va(&unused_pages_vp, (offset_t)off,
1902 MMU_PAGESIZE, PG_WAIT | PG_EXCL,
1903 &kseg, (caddr_t)off);
1904 if (pp == NULL)
1905 cmn_err(CE_PANIC, "limited physmem too much!");
1906 page_io_unlock(pp);
1907 page_downgrade(pp);
1908 availrmem--;
1909 off += MMU_PAGESIZE;
1910 }
1911 }
1912
1913 /*
1914 * When printing memory, show the total as physmem less
1915 * that stolen by a debugger.
1916 */
1917 cmn_err(CE_CONT, "?mem = %ldK (0x%lx000)\n",
1918 (ulong_t)(physinstalled) << (PAGESHIFT - 10),
1919 (ulong_t)(physinstalled) << (PAGESHIFT - 12));
1920
1921 avmem = (uint64_t)freemem << PAGESHIFT;
1922 cmn_err(CE_CONT, "?avail mem = %lld\n", (unsigned long long)avmem);
1923
1924 /*
1925 * For small memory systems disable automatic large pages.
1926 */
1927 if (physmem < privm_lpg_min_physmem) {
1928 use_brk_lpg = 0;
1929 use_stk_lpg = 0;
1930 }
1931
1932 /*
1933 * Perform platform specific freelist processing
1934 */
1935 if (&plat_freelist_process) {
1936 for (mnode = 0; mnode < max_mem_nodes; mnode++)
1937 if (mem_node_config[mnode].exists)
1938 plat_freelist_process(mnode);
1939 }
1940
1941 /*
1942 * Initialize the segkp segment type. We position it
1943 * after the configured tables and buffers (whose end
1944 * is given by econtig) and before V_WKBASE_ADDR.
1945 * Also in this area is segkmap (size SEGMAPSIZE).
1946 */
1947
1948 /* XXX - cache alignment? */
1949 va = (caddr_t)SEGKPBASE;
1950 ASSERT(((uintptr_t)va & PAGEOFFSET) == 0);
1951
1952 max_phys_segkp = (physmem * 2);
1953
1954 if (segkpsize < btop(SEGKPMINSIZE) || segkpsize > btop(SEGKPMAXSIZE)) {
1955 segkpsize = btop(SEGKPDEFSIZE);
1956 cmn_err(CE_WARN, "Illegal value for segkpsize. "
1957 "segkpsize has been reset to %ld pages", segkpsize);
1958 }
1959
1960 i = ptob(MIN(segkpsize, max_phys_segkp));
1961
1962 rw_enter(&kas.a_lock, RW_WRITER);
1963 if (seg_attach(&kas, va, i, segkp) < 0)
1964 cmn_err(CE_PANIC, "startup: cannot attach segkp");
1965 if (segkp_create(segkp) != 0)
1966 cmn_err(CE_PANIC, "startup: segkp_create failed");
1967 rw_exit(&kas.a_lock);
1968
1969 /*
1970 * kpm segment
1971 */
1972 segmap_kpm = kpm_enable &&
1973 segmap_kpm && PAGESIZE == MAXBSIZE;
1974
1975 if (kpm_enable) {
1976 rw_enter(&kas.a_lock, RW_WRITER);
1977
1978 /*
1979 * The segkpm virtual range range is larger than the
1980 * actual physical memory size and also covers gaps in
1981 * the physical address range for the following reasons:
1982 * . keep conversion between segkpm and physical addresses
1983 * simple, cheap and unambiguous.
1984 * . avoid extension/shrink of the the segkpm in case of DR.
1985 * . avoid complexity for handling of virtual addressed
1986 * caches, segkpm and the regular mapping scheme must be
1987 * kept in sync wrt. the virtual color of mapped pages.
1988 * Any accesses to virtual segkpm ranges not backed by
1989 * physical memory will fall through the memseg pfn hash
1990 * and will be handled in segkpm_fault.
1991 * Additional kpm_size spaces needed for vac alias prevention.
1992 */
1993 if (seg_attach(&kas, kpm_vbase, kpm_size * vac_colors,
1994 segkpm) < 0)
1995 cmn_err(CE_PANIC, "cannot attach segkpm");
1996
1997 b.prot = PROT_READ | PROT_WRITE;
1998 b.nvcolors = shm_alignment >> MMU_PAGESHIFT;
1999
2000 if (segkpm_create(segkpm, (caddr_t)&b) != 0)
2001 panic("segkpm_create segkpm");
2002
2003 rw_exit(&kas.a_lock);
2004
2005 mach_kpm_init();
2006 }
2007
2008 va = kpm_vbase + (kpm_size * vac_colors);
2009
2010 if (!segzio_fromheap) {
2011 size_t size;
2012 size_t physmem_b = mmu_ptob(physmem);
2013
2014 /* size is in bytes, segziosize is in pages */
2015 if (segziosize == 0) {
2016 size = physmem_b;
2017 } else {
2018 size = mmu_ptob(segziosize);
2019 }
2020
2021 if (size < SEGZIOMINSIZE) {
2022 size = SEGZIOMINSIZE;
2023 } else if (size > SEGZIOMAXSIZE) {
2024 size = SEGZIOMAXSIZE;
2025 /*
2026 * On 64-bit x86, we only have 2TB of KVA. This exists
2027 * for parity with x86.
2028 *
2029 * SEGZIOMAXSIZE is capped at 512gb so that segzio
2030 * doesn't consume all of KVA. However, if we have a
2031 * system that has more thant 512gb of physical memory,
2032 * we can actually consume about half of the difference
2033 * between 512gb and the rest of the available physical
2034 * memory.
2035 */
2036 if (physmem_b > SEGZIOMAXSIZE) {
2037 size += (physmem_b - SEGZIOMAXSIZE) / 2;
2038 }
2039 }
2040 segziosize = mmu_btop(roundup(size, MMU_PAGESIZE));
2041 /* put the base of the ZIO segment after the kpm segment */
2042 segzio_base = va;
2043 va += mmu_ptob(segziosize);
2044 PRM_DEBUG(segziosize);
2045 PRM_DEBUG(segzio_base);
2046
2047 /*
2048 * On some platforms, kvm_init is called after the kpm
2049 * sizes have been determined. On SPARC, kvm_init is called
2050 * before, so we have to attach the kzioseg after kvm is
2051 * initialized, otherwise we'll try to allocate from the boot
2052 * area since the kernel heap hasn't yet been configured.
2053 */
2054 rw_enter(&kas.a_lock, RW_WRITER);
2055
2056 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2057 &kzioseg);
2058 (void) segkmem_zio_create(&kzioseg);
2059
2060 /* create zio area covering new segment */
2061 segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2062
2063 rw_exit(&kas.a_lock);
2064 }
2065
2066 if (ppvm_enable) {
2067 caddr_t ppvm_max;
2068
2069 /*
2070 * ppvm refers to the static VA space used to map
2071 * the page_t's for dynamically added memory.
2072 *
2073 * ppvm_base should not cross a potential VA hole.
2074 *
2075 * ppvm_size should be large enough to map the
2076 * page_t's needed to manage all of KPM range.
2077 */
2078 ppvm_size =
2079 roundup(mmu_btop(kpm_size * vac_colors) * sizeof (page_t),
2080 MMU_PAGESIZE);
2081 ppvm_max = (caddr_t)(0ull - ppvm_size);
2082 ppvm_base = (page_t *)va;
2083
2084 if ((caddr_t)ppvm_base <= hole_end) {
2085 cmn_err(CE_WARN,
2086 "Memory DR disabled: invalid DR map base: 0x%p\n",
2087 (void *)ppvm_base);
2088 ppvm_enable = 0;
2089 } else if ((caddr_t)ppvm_base > ppvm_max) {
2090 uint64_t diff = (caddr_t)ppvm_base - ppvm_max;
2091
2092 cmn_err(CE_WARN,
2093 "Memory DR disabled: insufficient DR map size:"
2094 " 0x%lx (needed 0x%lx)\n",
2095 ppvm_size - diff, ppvm_size);
2096 ppvm_enable = 0;
2097 }
2098 PRM_DEBUG(ppvm_size);
2099 PRM_DEBUG(ppvm_base);
2100 }
2101
2102 /*
2103 * Now create generic mapping segment. This mapping
2104 * goes SEGMAPSIZE beyond SEGMAPBASE. But if the total
2105 * virtual address is greater than the amount of free
2106 * memory that is available, then we trim back the
2107 * segment size to that amount
2108 */
2109 va = (caddr_t)SEGMAPBASE;
2110
2111 /*
2112 * 1201049: segkmap base address must be MAXBSIZE aligned
2113 */
2114 ASSERT(((uintptr_t)va & MAXBOFFSET) == 0);
2115
2116 /*
2117 * Set size of segmap to percentage of freemem at boot,
2118 * but stay within the allowable range
2119 * Note we take percentage before converting from pages
2120 * to bytes to avoid an overflow on 32-bit kernels.
2121 */
2122 i = mmu_ptob((freemem * segmap_percent) / 100);
2123
2124 if (i < MINMAPSIZE)
2125 i = MINMAPSIZE;
2126
2127 if (i > MIN(SEGMAPSIZE, mmu_ptob(freemem)))
2128 i = MIN(SEGMAPSIZE, mmu_ptob(freemem));
2129
2130 i &= MAXBMASK; /* 1201049: segkmap size must be MAXBSIZE aligned */
2131
2132 rw_enter(&kas.a_lock, RW_WRITER);
2133 if (seg_attach(&kas, va, i, segkmap) < 0)
2134 cmn_err(CE_PANIC, "cannot attach segkmap");
2135
2136 a.prot = PROT_READ | PROT_WRITE;
2137 a.shmsize = shm_alignment;
2138 a.nfreelist = 0; /* use segmap driver defaults */
2139
2140 if (segmap_create(segkmap, (caddr_t)&a) != 0)
2141 panic("segmap_create segkmap");
2142 rw_exit(&kas.a_lock);
2143
2144 segdev_init();
2145 }
2146
2147 static void
2148 startup_end(void)
2149 {
2150 if ((caddr_t)memlist > (caddr_t)memlist_end)
2151 panic("memlist overflow 2");
2152 memlist_free_block((caddr_t)memlist,
2153 ((caddr_t)memlist_end - (caddr_t)memlist));
2154 memlist = NULL;
2155
2156 /* enable page_relocation since OBP is now done */
2157 page_relocate_ready = 1;
2158
2159 /*
2160 * Perform tasks that get done after most of the VM
2161 * initialization has been done but before the clock
2162 * and other devices get started.
2163 */
2164 kern_setup1();
2165
2166 /*
2167 * Perform CPC initialization for this CPU.
2168 */
2169 kcpc_hw_init();
2170
2171 /*
2172 * Intialize the VM arenas for allocating physically
2173 * contiguus memory chunk for interrupt queues snd
2174 * allocate/register boot cpu's queues, if any and
2175 * allocate dump buffer for sun4v systems to store
2176 * extra crash information during crash dump
2177 */
2178 contig_mem_init();
2179 mach_descrip_init();
2180
2181 if (cpu_intrq_setup(CPU)) {
2182 cmn_err(CE_PANIC, "cpu%d: setup failed", CPU->cpu_id);
2183 }
2184 cpu_intrq_register(CPU);
2185 mach_htraptrace_setup(CPU->cpu_id);
2186 mach_htraptrace_configure(CPU->cpu_id);
2187 mach_dump_buffer_init();
2188
2189 /*
2190 * Initialize interrupt related stuff
2191 */
2192 cpu_intr_alloc(CPU, NINTR_THREADS);
2193
2194 (void) splzs(); /* allow hi clock ints but not zs */
2195
2196 /*
2197 * Initialize errors.
2198 */
2199 error_init();
2200
2201 /*
2202 * Note that we may have already used kernel bcopy before this
2203 * point - but if you really care about this, adb the use_hw_*
2204 * variables to 0 before rebooting.
2205 */
2206 mach_hw_copy_limit();
2207
2208 /*
2209 * Install the "real" preemption guards before DDI services
2210 * are available.
2211 */
2212 (void) prom_set_preprom(kern_preprom);
2213 (void) prom_set_postprom(kern_postprom);
2214 CPU->cpu_m.mutex_ready = 1;
2215
2216 /*
2217 * Initialize segnf (kernel support for non-faulting loads).
2218 */
2219 segnf_init();
2220
2221 /*
2222 * Configure the root devinfo node.
2223 */
2224 configure(); /* set up devices */
2225 mach_cpu_halt_idle();
2226 }
2227
2228
2229 void
2230 post_startup(void)
2231 {
2232 #ifdef PTL1_PANIC_DEBUG
2233 extern void init_ptl1_thread(void);
2234 #endif /* PTL1_PANIC_DEBUG */
2235 extern void abort_sequence_init(void);
2236
2237 /*
2238 * Set the system wide, processor-specific flags to be passed
2239 * to userland via the aux vector for performance hints and
2240 * instruction set extensions.
2241 */
2242 bind_hwcap();
2243
2244 /*
2245 * Startup memory scrubber (if any)
2246 */
2247 mach_memscrub();
2248
2249 /*
2250 * Allocate soft interrupt to handle abort sequence.
2251 */
2252 abort_sequence_init();
2253
2254 /*
2255 * Configure the rest of the system.
2256 * Perform forceloading tasks for /etc/system.
2257 */
2258 (void) mod_sysctl(SYS_FORCELOAD, NULL);
2259 /*
2260 * ON4.0: Force /proc module in until clock interrupt handle fixed
2261 * ON4.0: This must be fixed or restated in /etc/systems.
2262 */
2263 (void) modload("fs", "procfs");
2264
2265 /* load machine class specific drivers */
2266 load_mach_drivers();
2267
2268 /* load platform specific drivers */
2269 if (&load_platform_drivers)
2270 load_platform_drivers();
2271
2272 /* load vis simulation module, if we are running w/fpu off */
2273 if (!fpu_exists) {
2274 if (modload("misc", "vis") == -1)
2275 halt("Can't load vis");
2276 }
2277
2278 mach_fpras();
2279
2280 maxmem = freemem;
2281
2282 pg_init();
2283
2284 #ifdef PTL1_PANIC_DEBUG
2285 init_ptl1_thread();
2286 #endif /* PTL1_PANIC_DEBUG */
2287 }
2288
2289 #ifdef PTL1_PANIC_DEBUG
2290 int ptl1_panic_test = 0;
2291 int ptl1_panic_xc_one_test = 0;
2292 int ptl1_panic_xc_all_test = 0;
2293 int ptl1_panic_xt_one_test = 0;
2294 int ptl1_panic_xt_all_test = 0;
2295 kthread_id_t ptl1_thread_p = NULL;
2296 kcondvar_t ptl1_cv;
2297 kmutex_t ptl1_mutex;
2298 int ptl1_recurse_count_threshold = 0x40;
2299 int ptl1_recurse_trap_threshold = 0x3d;
2300 extern void ptl1_recurse(int, int);
2301 extern void ptl1_panic_xt(int, int);
2302
2303 /*
2304 * Called once per second by timeout() to wake up
2305 * the ptl1_panic thread to see if it should cause
2306 * a trap to the ptl1_panic() code.
2307 */
2308 /* ARGSUSED */
2309 static void
2310 ptl1_wakeup(void *arg)
2311 {
2312 mutex_enter(&ptl1_mutex);
2313 cv_signal(&ptl1_cv);
2314 mutex_exit(&ptl1_mutex);
2315 }
2316
2317 /*
2318 * ptl1_panic cross call function:
2319 * Needed because xc_one() and xc_some() can pass
2320 * 64 bit args but ptl1_recurse() expects ints.
2321 */
2322 static void
2323 ptl1_panic_xc(void)
2324 {
2325 ptl1_recurse(ptl1_recurse_count_threshold,
2326 ptl1_recurse_trap_threshold);
2327 }
2328
2329 /*
2330 * The ptl1 thread waits for a global flag to be set
2331 * and uses the recurse thresholds to set the stack depth
2332 * to cause a ptl1_panic() directly via a call to ptl1_recurse
2333 * or indirectly via the cross call and cross trap functions.
2334 *
2335 * This is useful testing stack overflows and normal
2336 * ptl1_panic() states with a know stack frame.
2337 *
2338 * ptl1_recurse() is an asm function in ptl1_panic.s that
2339 * sets the {In, Local, Out, and Global} registers to a
2340 * know state on the stack and just prior to causing a
2341 * test ptl1_panic trap.
2342 */
2343 static void
2344 ptl1_thread(void)
2345 {
2346 mutex_enter(&ptl1_mutex);
2347 while (ptl1_thread_p) {
2348 cpuset_t other_cpus;
2349 int cpu_id;
2350 int my_cpu_id;
2351 int target_cpu_id;
2352 int target_found;
2353
2354 if (ptl1_panic_test) {
2355 ptl1_recurse(ptl1_recurse_count_threshold,
2356 ptl1_recurse_trap_threshold);
2357 }
2358
2359 /*
2360 * Find potential targets for x-call and x-trap,
2361 * if any exist while preempt is disabled we
2362 * start a ptl1_panic if requested via a
2363 * globals.
2364 */
2365 kpreempt_disable();
2366 my_cpu_id = CPU->cpu_id;
2367 other_cpus = cpu_ready_set;
2368 CPUSET_DEL(other_cpus, CPU->cpu_id);
2369 target_found = 0;
2370 if (!CPUSET_ISNULL(other_cpus)) {
2371 /*
2372 * Pick the first one
2373 */
2374 for (cpu_id = 0; cpu_id < NCPU; cpu_id++) {
2375 if (cpu_id == my_cpu_id)
2376 continue;
2377
2378 if (CPU_XCALL_READY(cpu_id)) {
2379 target_cpu_id = cpu_id;
2380 target_found = 1;
2381 break;
2382 }
2383 }
2384 ASSERT(target_found);
2385
2386 if (ptl1_panic_xc_one_test) {
2387 xc_one(target_cpu_id,
2388 (xcfunc_t *)ptl1_panic_xc, 0, 0);
2389 }
2390 if (ptl1_panic_xc_all_test) {
2391 xc_some(other_cpus,
2392 (xcfunc_t *)ptl1_panic_xc, 0, 0);
2393 }
2394 if (ptl1_panic_xt_one_test) {
2395 xt_one(target_cpu_id,
2396 (xcfunc_t *)ptl1_panic_xt, 0, 0);
2397 }
2398 if (ptl1_panic_xt_all_test) {
2399 xt_some(other_cpus,
2400 (xcfunc_t *)ptl1_panic_xt, 0, 0);
2401 }
2402 }
2403 kpreempt_enable();
2404 (void) timeout(ptl1_wakeup, NULL, hz);
2405 (void) cv_wait(&ptl1_cv, &ptl1_mutex);
2406 }
2407 mutex_exit(&ptl1_mutex);
2408 }
2409
2410 /*
2411 * Called during early startup to create the ptl1_thread
2412 */
2413 void
2414 init_ptl1_thread(void)
2415 {
2416 ptl1_thread_p = thread_create(NULL, 0, ptl1_thread, NULL, 0,
2417 &p0, TS_RUN, 0);
2418 }
2419 #endif /* PTL1_PANIC_DEBUG */
2420
2421
2422 static void
2423 memlist_new(uint64_t start, uint64_t len, struct memlist **memlistp)
2424 {
2425 struct memlist *new;
2426
2427 new = *memlistp;
2428 new->ml_address = start;
2429 new->ml_size = len;
2430 *memlistp = new + 1;
2431 }
2432
2433 /*
2434 * Add to a memory list.
2435 * start = start of new memory segment
2436 * len = length of new memory segment in bytes
2437 * memlistp = pointer to array of available memory segment structures
2438 * curmemlistp = memory list to which to add segment.
2439 */
2440 static void
2441 memlist_add(uint64_t start, uint64_t len, struct memlist **memlistp,
2442 struct memlist **curmemlistp)
2443 {
2444 struct memlist *new = *memlistp;
2445
2446 memlist_new(start, len, memlistp);
2447 memlist_insert(new, curmemlistp);
2448 }
2449
2450 static int
2451 ndata_alloc_memseg(struct memlist *ndata, size_t avail)
2452 {
2453 int nseg;
2454 size_t memseg_sz;
2455 struct memseg *msp;
2456
2457 /*
2458 * The memseg list is for the chunks of physical memory that
2459 * will be managed by the vm system. The number calculated is
2460 * a guess as boot may fragment it more when memory allocations
2461 * are made before kphysm_init().
2462 */
2463 memseg_sz = (avail + 10) * sizeof (struct memseg);
2464 memseg_sz = roundup(memseg_sz, PAGESIZE);
2465 nseg = memseg_sz / sizeof (struct memseg);
2466 msp = ndata_alloc(ndata, memseg_sz, ecache_alignsize);
2467 if (msp == NULL)
2468 return (1);
2469 PRM_DEBUG(memseg_free);
2470
2471 while (nseg--) {
2472 msp->next = memseg_free;
2473 memseg_free = msp;
2474 msp++;
2475 }
2476 return (0);
2477 }
2478
2479 /*
2480 * In the case of architectures that support dynamic addition of
2481 * memory at run-time there are two cases where memsegs need to
2482 * be initialized and added to the memseg list.
2483 * 1) memsegs that are constructed at startup.
2484 * 2) memsegs that are constructed at run-time on
2485 * hot-plug capable architectures.
2486 * This code was originally part of the function kphysm_init().
2487 */
2488
2489 static void
2490 memseg_list_add(struct memseg *memsegp)
2491 {
2492 struct memseg **prev_memsegp;
2493 pgcnt_t num;
2494
2495 /* insert in memseg list, decreasing number of pages order */
2496
2497 num = MSEG_NPAGES(memsegp);
2498
2499 for (prev_memsegp = &memsegs; *prev_memsegp;
2500 prev_memsegp = &((*prev_memsegp)->next)) {
2501 if (num > MSEG_NPAGES(*prev_memsegp))
2502 break;
2503 }
2504
2505 memsegp->next = *prev_memsegp;
2506 *prev_memsegp = memsegp;
2507
2508 if (kpm_enable) {
2509 memsegp->nextpa = (memsegp->next) ?
2510 va_to_pa(memsegp->next) : MSEG_NULLPTR_PA;
2511
2512 if (prev_memsegp != &memsegs) {
2513 struct memseg *msp;
2514 msp = (struct memseg *)((caddr_t)prev_memsegp -
2515 offsetof(struct memseg, next));
2516 msp->nextpa = va_to_pa(memsegp);
2517 } else {
2518 memsegspa = va_to_pa(memsegs);
2519 }
2520 }
2521 }
2522
2523 /*
2524 * PSM add_physmem_cb(). US-II and newer processors have some
2525 * flavor of the prefetch capability implemented. We exploit
2526 * this capability for optimum performance.
2527 */
2528 #define PREFETCH_BYTES 64
2529
2530 void
2531 add_physmem_cb(page_t *pp, pfn_t pnum)
2532 {
2533 extern void prefetch_page_w(void *);
2534
2535 pp->p_pagenum = pnum;
2536
2537 /*
2538 * Prefetch one more page_t into E$. To prevent future
2539 * mishaps with the sizeof(page_t) changing on us, we
2540 * catch this on debug kernels if we can't bring in the
2541 * entire hpage with 2 PREFETCH_BYTES reads. See
2542 * also, sun4u/cpu/cpu_module.c
2543 */
2544 /*LINTED*/
2545 ASSERT(sizeof (page_t) <= 2*PREFETCH_BYTES);
2546 prefetch_page_w((char *)pp);
2547 }
2548
2549 /*
2550 * Find memseg with given pfn
2551 */
2552 static struct memseg *
2553 memseg_find(pfn_t base, pfn_t *next)
2554 {
2555 struct memseg *seg;
2556
2557 if (next != NULL)
2558 *next = LONG_MAX;
2559 for (seg = memsegs; seg != NULL; seg = seg->next) {
2560 if (base >= seg->pages_base && base < seg->pages_end)
2561 return (seg);
2562 if (next != NULL && seg->pages_base > base &&
2563 seg->pages_base < *next)
2564 *next = seg->pages_base;
2565 }
2566 return (NULL);
2567 }
2568
2569 /*
2570 * Put page allocated by OBP on prom_ppages
2571 */
2572 static void
2573 kphysm_erase(uint64_t addr, uint64_t len)
2574 {
2575 struct page *pp;
2576 struct memseg *seg;
2577 pfn_t base = btop(addr), next;
2578 pgcnt_t num = btop(len);
2579
2580 while (num != 0) {
2581 pgcnt_t off, left;
2582
2583 seg = memseg_find(base, &next);
2584 if (seg == NULL) {
2585 if (next == LONG_MAX)
2586 break;
2587 left = MIN(next - base, num);
2588 base += left, num -= left;
2589 continue;
2590 }
2591 off = base - seg->pages_base;
2592 pp = seg->pages + off;
2593 left = num - MIN(num, (seg->pages_end - seg->pages_base) - off);
2594 while (num != left) {
2595 /*
2596 * init it, lock it, and hashin on prom_pages vp.
2597 *
2598 * Mark it as NONRELOC to let DR know the page
2599 * is locked long term, otherwise DR hangs when
2600 * trying to remove those pages.
2601 *
2602 * XXX vnode offsets on the prom_ppages vnode
2603 * are page numbers (gack) for >32 bit
2604 * physical memory machines.
2605 */
2606 PP_SETNORELOC(pp);
2607 add_physmem_cb(pp, base);
2608 if (page_trylock(pp, SE_EXCL) == 0)
2609 cmn_err(CE_PANIC, "prom page locked");
2610 (void) page_hashin(pp, &promvp,
2611 (offset_t)base, NULL);
2612 (void) page_pp_lock(pp, 0, 1);
2613 pp++, base++, num--;
2614 }
2615 }
2616 }
2617
2618 static page_t *ppnext;
2619 static pgcnt_t ppleft;
2620
2621 static void *kpm_ppnext;
2622 static pgcnt_t kpm_ppleft;
2623
2624 /*
2625 * Create a memseg
2626 */
2627 static void
2628 kphysm_memseg(uint64_t addr, uint64_t len)
2629 {
2630 pfn_t base = btop(addr);
2631 pgcnt_t num = btop(len);
2632 struct memseg *seg;
2633
2634 seg = memseg_free;
2635 memseg_free = seg->next;
2636 ASSERT(seg != NULL);
2637
2638 seg->pages = ppnext;
2639 seg->epages = ppnext + num;
2640 seg->pages_base = base;
2641 seg->pages_end = base + num;
2642 ppnext += num;
2643 ppleft -= num;
2644
2645 if (kpm_enable) {
2646 pgcnt_t kpnum = ptokpmpr(num);
2647
2648 if (kpnum > kpm_ppleft)
2649 panic("kphysm_memseg: kpm_pp overflow");
2650 seg->pagespa = va_to_pa(seg->pages);
2651 seg->epagespa = va_to_pa(seg->epages);
2652 seg->kpm_pbase = kpmptop(ptokpmp(base));
2653 seg->kpm_nkpmpgs = kpnum;
2654 /*
2655 * In the kpm_smallpage case, the kpm array
2656 * is 1-1 wrt the page array
2657 */
2658 if (kpm_smallpages) {
2659 kpm_spage_t *kpm_pp = kpm_ppnext;
2660
2661 kpm_ppnext = kpm_pp + kpnum;
2662 seg->kpm_spages = kpm_pp;
2663 seg->kpm_pagespa = va_to_pa(seg->kpm_spages);
2664 } else {
2665 kpm_page_t *kpm_pp = kpm_ppnext;
2666
2667 kpm_ppnext = kpm_pp + kpnum;
2668 seg->kpm_pages = kpm_pp;
2669 seg->kpm_pagespa = va_to_pa(seg->kpm_pages);
2670 /* ASSERT no kpm overlaps */
2671 ASSERT(
2672 memseg_find(base - pmodkpmp(base), NULL) == NULL);
2673 ASSERT(memseg_find(
2674 roundup(base + num, kpmpnpgs) - 1, NULL) == NULL);
2675 }
2676 kpm_ppleft -= kpnum;
2677 }
2678
2679 memseg_list_add(seg);
2680 }
2681
2682 /*
2683 * Add range to free list
2684 */
2685 void
2686 kphysm_add(uint64_t addr, uint64_t len, int reclaim)
2687 {
2688 struct page *pp;
2689 struct memseg *seg;
2690 pfn_t base = btop(addr);
2691 pgcnt_t num = btop(len);
2692
2693 seg = memseg_find(base, NULL);
2694 ASSERT(seg != NULL);
2695 pp = seg->pages + (base - seg->pages_base);
2696
2697 if (reclaim) {
2698 struct page *rpp = pp;
2699 struct page *lpp = pp + num;
2700
2701 /*
2702 * page should be locked on prom_ppages
2703 * unhash and unlock it
2704 */
2705 while (rpp < lpp) {
2706 ASSERT(PAGE_EXCL(rpp) && rpp->p_vnode == &promvp);
2707 ASSERT(PP_ISNORELOC(rpp));
2708 PP_CLRNORELOC(rpp);
2709 page_pp_unlock(rpp, 0, 1);
2710 page_hashout(rpp, NULL);
2711 page_unlock(rpp);
2712 rpp++;
2713 }
2714 }
2715
2716 /*
2717 * add_physmem() initializes the PSM part of the page
2718 * struct by calling the PSM back with add_physmem_cb().
2719 * In addition it coalesces pages into larger pages as
2720 * it initializes them.
2721 */
2722 add_physmem(pp, num, base);
2723 }
2724
2725 /*
2726 * kphysm_init() tackles the problem of initializing physical memory.
2727 */
2728 static void
2729 kphysm_init(void)
2730 {
2731 struct memlist *pmem;
2732
2733 ASSERT(page_hash != NULL && page_hashsz != 0);
2734
2735 ppnext = pp_base;
2736 ppleft = npages;
2737 kpm_ppnext = kpm_pp_base;
2738 kpm_ppleft = kpm_npages;
2739
2740 /*
2741 * installed pages not on nopp_memlist go in memseg list
2742 */
2743 diff_memlists(phys_install, nopp_list, kphysm_memseg);
2744
2745 /*
2746 * Free the avail list
2747 */
2748 for (pmem = phys_avail; pmem != NULL; pmem = pmem->ml_next)
2749 kphysm_add(pmem->ml_address, pmem->ml_size, 0);
2750
2751 /*
2752 * Erase pages that aren't available
2753 */
2754 diff_memlists(phys_install, phys_avail, kphysm_erase);
2755
2756 build_pfn_hash();
2757 }
2758
2759 /*
2760 * Kernel VM initialization.
2761 * Assumptions about kernel address space ordering:
2762 * (1) gap (user space)
2763 * (2) kernel text
2764 * (3) kernel data/bss
2765 * (4) gap
2766 * (5) kernel data structures
2767 * (6) gap
2768 * (7) debugger (optional)
2769 * (8) monitor
2770 * (9) gap (possibly null)
2771 * (10) dvma
2772 * (11) devices
2773 */
2774 static void
2775 kvm_init(void)
2776 {
2777 /*
2778 * Put the kernel segments in kernel address space.
2779 */
2780 rw_enter(&kas.a_lock, RW_WRITER);
2781 as_avlinit(&kas);
2782
2783 (void) seg_attach(&kas, (caddr_t)KERNELBASE,
2784 (size_t)(e_moddata - KERNELBASE), &ktextseg);
2785 (void) segkmem_create(&ktextseg);
2786
2787 (void) seg_attach(&kas, (caddr_t)(KERNELBASE + MMU_PAGESIZE4M),
2788 (size_t)(MMU_PAGESIZE4M), &ktexthole);
2789 (void) segkmem_create(&ktexthole);
2790
2791 (void) seg_attach(&kas, (caddr_t)valloc_base,
2792 (size_t)(econtig32 - valloc_base), &kvalloc);
2793 (void) segkmem_create(&kvalloc);
2794
2795 if (kmem64_base) {
2796 (void) seg_attach(&kas, (caddr_t)kmem64_base,
2797 (size_t)(kmem64_end - kmem64_base), &kmem64);
2798 (void) segkmem_create(&kmem64);
2799 }
2800
2801 /*
2802 * We're about to map out /boot. This is the beginning of the
2803 * system resource management transition. We can no longer
2804 * call into /boot for I/O or memory allocations.
2805 */
2806 (void) seg_attach(&kas, kernelheap, ekernelheap - kernelheap, &kvseg);
2807 (void) segkmem_create(&kvseg);
2808 hblk_alloc_dynamic = 1;
2809
2810 /*
2811 * we need to preallocate pages for DR operations before enabling large
2812 * page kernel heap because of memseg_remap_init() hat_unload() hack.
2813 */
2814 memseg_remap_init();
2815
2816 /* at this point we are ready to use large page heap */
2817 segkmem_heap_lp_init();
2818
2819 (void) seg_attach(&kas, (caddr_t)SYSBASE32, SYSLIMIT32 - SYSBASE32,
2820 &kvseg32);
2821 (void) segkmem_create(&kvseg32);
2822
2823 /*
2824 * Create a segment for the debugger.
2825 */
2826 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2827 (void) segkmem_create(&kdebugseg);
2828
2829 rw_exit(&kas.a_lock);
2830 }
2831
2832 char obp_tte_str[] =
2833 "h# %x constant MMU_PAGESHIFT "
2834 "h# %x constant TTE8K "
2835 "h# %x constant SFHME_SIZE "
2836 "h# %x constant SFHME_TTE "
2837 "h# %x constant HMEBLK_TAG "
2838 "h# %x constant HMEBLK_NEXT "
2839 "h# %x constant HMEBLK_MISC "
2840 "h# %x constant HMEBLK_HME1 "
2841 "h# %x constant NHMENTS "
2842 "h# %x constant HBLK_SZMASK "
2843 "h# %x constant HBLK_RANGE_SHIFT "
2844 "h# %x constant HMEBP_HBLK "
2845 "h# %x constant HMEBLK_ENDPA "
2846 "h# %x constant HMEBUCKET_SIZE "
2847 "h# %x constant HTAG_SFMMUPSZ "
2848 "h# %x constant HTAG_BSPAGE_SHIFT "
2849 "h# %x constant HTAG_REHASH_SHIFT "
2850 "h# %x constant SFMMU_INVALID_SHMERID "
2851 "h# %x constant mmu_hashcnt "
2852 "h# %p constant uhme_hash "
2853 "h# %p constant khme_hash "
2854 "h# %x constant UHMEHASH_SZ "
2855 "h# %x constant KHMEHASH_SZ "
2856 "h# %p constant KCONTEXT "
2857 "h# %p constant KHATID "
2858 "h# %x constant ASI_MEM "
2859
2860 ": PHYS-X@ ( phys -- data ) "
2861 " ASI_MEM spacex@ "
2862 "; "
2863
2864 ": PHYS-W@ ( phys -- data ) "
2865 " ASI_MEM spacew@ "
2866 "; "
2867
2868 ": PHYS-L@ ( phys -- data ) "
2869 " ASI_MEM spaceL@ "
2870 "; "
2871
2872 ": TTE_PAGE_SHIFT ( ttesz -- hmeshift ) "
2873 " 3 * MMU_PAGESHIFT + "
2874 "; "
2875
2876 ": TTE_IS_VALID ( ttep -- flag ) "
2877 " PHYS-X@ 0< "
2878 "; "
2879
2880 ": HME_HASH_SHIFT ( ttesz -- hmeshift ) "
2881 " dup TTE8K = if "
2882 " drop HBLK_RANGE_SHIFT "
2883 " else "
2884 " TTE_PAGE_SHIFT "
2885 " then "
2886 "; "
2887
2888 ": HME_HASH_BSPAGE ( addr hmeshift -- bspage ) "
2889 " tuck >> swap MMU_PAGESHIFT - << "
2890 "; "
2891
2892 ": HME_HASH_FUNCTION ( sfmmup addr hmeshift -- hmebp ) "
2893 " >> over xor swap ( hash sfmmup ) "
2894 " KHATID <> if ( hash ) "
2895 " UHMEHASH_SZ and ( bucket ) "
2896 " HMEBUCKET_SIZE * uhme_hash + ( hmebp ) "
2897 " else ( hash ) "
2898 " KHMEHASH_SZ and ( bucket ) "
2899 " HMEBUCKET_SIZE * khme_hash + ( hmebp ) "
2900 " then ( hmebp ) "
2901 "; "
2902
2903 ": HME_HASH_TABLE_SEARCH "
2904 " ( sfmmup hmebp hblktag -- sfmmup null | sfmmup hmeblkp ) "
2905 " >r hmebp_hblk + phys-x@ begin ( sfmmup hmeblkp ) ( r: hblktag ) "
2906 " dup HMEBLK_ENDPA <> if ( sfmmup hmeblkp ) ( r: hblktag ) "
2907 " dup hmeblk_tag + phys-x@ r@ = if ( sfmmup hmeblkp ) "
2908 " dup hmeblk_tag + 8 + phys-x@ 2 pick = if "
2909 " true ( sfmmup hmeblkp true ) ( r: hblktag ) "
2910 " else "
2911 " hmeblk_next + phys-x@ false "
2912 " ( sfmmup hmeblkp false ) ( r: hblktag ) "
2913 " then "
2914 " else "
2915 " hmeblk_next + phys-x@ false "
2916 " ( sfmmup hmeblkp false ) ( r: hblktag ) "
2917 " then "
2918 " else "
2919 " drop 0 true "
2920 " then "
2921 " until r> drop "
2922 "; "
2923
2924 ": HME_HASH_TAG ( sfmmup rehash addr -- hblktag ) "
2925 " over HME_HASH_SHIFT HME_HASH_BSPAGE ( sfmmup rehash bspage ) "
2926 " HTAG_BSPAGE_SHIFT << ( sfmmup rehash htag-bspage )"
2927 " swap HTAG_REHASH_SHIFT << or ( sfmmup htag-bspage-rehash )"
2928 " SFMMU_INVALID_SHMERID or nip ( hblktag ) "
2929 "; "
2930
2931 ": HBLK_TO_TTEP ( hmeblkp addr -- ttep ) "
2932 " over HMEBLK_MISC + PHYS-L@ HBLK_SZMASK and ( hmeblkp addr ttesz ) "
2933 " TTE8K = if ( hmeblkp addr ) "
2934 " MMU_PAGESHIFT >> NHMENTS 1- and ( hmeblkp hme-index ) "
2935 " else ( hmeblkp addr ) "
2936 " drop 0 ( hmeblkp 0 ) "
2937 " then ( hmeblkp hme-index ) "
2938 " SFHME_SIZE * + HMEBLK_HME1 + ( hmep ) "
2939 " SFHME_TTE + ( ttep ) "
2940 "; "
2941
2942 ": unix-tte ( addr cnum -- false | tte-data true ) "
2943 " KCONTEXT = if ( addr ) "
2944 " KHATID ( addr khatid ) "
2945 " else ( addr ) "
2946 " drop false exit ( false ) "
2947 " then "
2948 " ( addr khatid ) "
2949 " mmu_hashcnt 1+ 1 do ( addr sfmmup ) "
2950 " 2dup swap i HME_HASH_SHIFT "
2951 "( addr sfmmup sfmmup addr hmeshift ) "
2952 " HME_HASH_FUNCTION ( addr sfmmup hmebp ) "
2953 " over i 4 pick "
2954 "( addr sfmmup hmebp sfmmup rehash addr ) "
2955 " HME_HASH_TAG ( addr sfmmup hmebp hblktag ) "
2956 " HME_HASH_TABLE_SEARCH "
2957 "( addr sfmmup { null | hmeblkp } ) "
2958 " ?dup if ( addr sfmmup hmeblkp ) "
2959 " nip swap HBLK_TO_TTEP ( ttep ) "
2960 " dup TTE_IS_VALID if ( valid-ttep ) "
2961 " PHYS-X@ true ( tte-data true ) "
2962 " else ( invalid-tte ) "
2963 " drop false ( false ) "
2964 " then ( false | tte-data true ) "
2965 " unloop exit ( false | tte-data true ) "
2966 " then ( addr sfmmup ) "
2967 " loop ( addr sfmmup ) "
2968 " 2drop false ( false ) "
2969 "; "
2970 ;
2971
2972 void
2973 create_va_to_tte(void)
2974 {
2975 char *bp;
2976 extern int khmehash_num, uhmehash_num;
2977 extern struct hmehash_bucket *khme_hash, *uhme_hash;
2978
2979 #define OFFSET(type, field) ((uintptr_t)(&((type *)0)->field))
2980
2981 bp = (char *)kobj_zalloc(MMU_PAGESIZE, KM_SLEEP);
2982
2983 /*
2984 * Teach obp how to parse our sw ttes.
2985 */
2986 (void) sprintf(bp, obp_tte_str,
2987 MMU_PAGESHIFT,
2988 TTE8K,
2989 sizeof (struct sf_hment),
2990 OFFSET(struct sf_hment, hme_tte),
2991 OFFSET(struct hme_blk, hblk_tag),
2992 OFFSET(struct hme_blk, hblk_nextpa),
2993 OFFSET(struct hme_blk, hblk_misc),
2994 OFFSET(struct hme_blk, hblk_hme),
2995 NHMENTS,
2996 HBLK_SZMASK,
2997 HBLK_RANGE_SHIFT,
2998 OFFSET(struct hmehash_bucket, hmeh_nextpa),
2999 HMEBLK_ENDPA,
3000 sizeof (struct hmehash_bucket),
3001 HTAG_SFMMUPSZ,
3002 HTAG_BSPAGE_SHIFT,
3003 HTAG_REHASH_SHIFT,
3004 SFMMU_INVALID_SHMERID,
3005 mmu_hashcnt,
3006 (caddr_t)va_to_pa((caddr_t)uhme_hash),
3007 (caddr_t)va_to_pa((caddr_t)khme_hash),
3008 UHMEHASH_SZ,
3009 KHMEHASH_SZ,
3010 KCONTEXT,
3011 KHATID,
3012 ASI_MEM);
3013 prom_interpret(bp, 0, 0, 0, 0, 0);
3014
3015 kobj_free(bp, MMU_PAGESIZE);
3016 }
3017
3018 void
3019 install_va_to_tte(void)
3020 {
3021 /*
3022 * advise prom that it can use unix-tte
3023 */
3024 prom_interpret("' unix-tte is va>tte-data", 0, 0, 0, 0, 0);
3025 }
3026
3027 /*
3028 * Here we add "device-type=console" for /os-io node, for currently
3029 * our kernel console output only supports displaying text and
3030 * performing cursor-positioning operations (through kernel framebuffer
3031 * driver) and it doesn't support other functionalities required for a
3032 * standard "display" device as specified in 1275 spec. The main missing
3033 * interface defined by the 1275 spec is "draw-logo".
3034 * also see the comments above prom_stdout_is_framebuffer().
3035 */
3036 static char *create_node =
3037 "\" /\" find-device "
3038 "new-device "
3039 "\" os-io\" device-name "
3040 "\" "OBP_DISPLAY_CONSOLE"\" device-type "
3041 ": cb-r/w ( adr,len method$ -- #read/#written ) "
3042 " 2>r swap 2 2r> ['] $callback catch if "
3043 " 2drop 3drop 0 "
3044 " then "
3045 "; "
3046 ": read ( adr,len -- #read ) "
3047 " \" read\" ['] cb-r/w catch if 2drop 2drop -2 exit then "
3048 " ( retN ... ret1 N ) "
3049 " ?dup if "
3050 " swap >r 1- 0 ?do drop loop r> "
3051 " else "
3052 " -2 "
3053 " then "
3054 "; "
3055 ": write ( adr,len -- #written ) "
3056 " \" write\" ['] cb-r/w catch if 2drop 2drop 0 exit then "
3057 " ( retN ... ret1 N ) "
3058 " ?dup if "
3059 " swap >r 1- 0 ?do drop loop r> "
3060 " else "
3061 " 0 "
3062 " then "
3063 "; "
3064 ": poll-tty ( -- ) ; "
3065 ": install-abort ( -- ) ['] poll-tty d# 10 alarm ; "
3066 ": remove-abort ( -- ) ['] poll-tty 0 alarm ; "
3067 ": cb-give/take ( $method -- ) "
3068 " 0 -rot ['] $callback catch ?dup if "
3069 " >r 2drop 2drop r> throw "
3070 " else "
3071 " 0 ?do drop loop "
3072 " then "
3073 "; "
3074 ": give ( -- ) \" exit-input\" cb-give/take ; "
3075 ": take ( -- ) \" enter-input\" cb-give/take ; "
3076 ": open ( -- ok? ) true ; "
3077 ": close ( -- ) ; "
3078 "finish-device "
3079 "device-end ";
3080
3081 /*
3082 * Create the OBP input/output node (FCode serial driver).
3083 * It is needed for both USB console keyboard and for
3084 * the kernel terminal emulator. It is too early to check for a
3085 * kernel console compatible framebuffer now, so we create this
3086 * so that we're ready if we need to enable kernel terminal emulation.
3087 *
3088 * When the USB software takes over the input device at the time
3089 * consconfig runs, OBP's stdin is redirected to this node.
3090 * Whenever the FORTH user interface is used after this switch,
3091 * the node will call back into the kernel for console input.
3092 * If a serial device such as ttya or a UART with a Type 5 keyboard
3093 * attached is used, OBP takes over the serial device when the system
3094 * goes to the debugger after the system is booted. This sharing
3095 * of the relatively simple serial device is difficult but possible.
3096 * Sharing the USB host controller is impossible due its complexity.
3097 *
3098 * Similarly to USB keyboard input redirection, after consconfig_dacf
3099 * configures a kernel console framebuffer as the standard output
3100 * device, OBP's stdout is switched to to vector through the
3101 * /os-io node into the kernel terminal emulator.
3102 */
3103 static void
3104 startup_create_io_node(void)
3105 {
3106 prom_interpret(create_node, 0, 0, 0, 0, 0);
3107 }
3108
3109
3110 static void
3111 do_prom_version_check(void)
3112 {
3113 int i;
3114 pnode_t node;
3115 char buf[64];
3116 static char drev[] = "Down-rev firmware detected%s\n"
3117 "\tPlease upgrade to the following minimum version:\n"
3118 "\t\t%s\n";
3119
3120 i = prom_version_check(buf, sizeof (buf), &node);
3121
3122 if (i == PROM_VER64_OK)
3123 return;
3124
3125 if (i == PROM_VER64_UPGRADE) {
3126 cmn_err(CE_WARN, drev, "", buf);
3127
3128 #ifdef DEBUG
3129 prom_enter_mon(); /* Type 'go' to continue */
3130 cmn_err(CE_WARN, "Booting with down-rev firmware\n");
3131 return;
3132 #else
3133 halt(0);
3134 #endif
3135 }
3136
3137 /*
3138 * The other possibility is that this is a server running
3139 * good firmware, but down-rev firmware was detected on at
3140 * least one other cpu board. We just complain if we see
3141 * that.
3142 */
3143 cmn_err(CE_WARN, drev, " on one or more CPU boards", buf);
3144 }
3145
3146
3147 /*
3148 * Must be defined in platform dependent code.
3149 */
3150 extern caddr_t modtext;
3151 extern size_t modtext_sz;
3152 extern caddr_t moddata;
3153
3154 #define HEAPTEXT_ARENA(addr) \
3155 ((uintptr_t)(addr) < KERNELBASE + 2 * MMU_PAGESIZE4M ? 0 : \
3156 (((uintptr_t)(addr) - HEAPTEXT_BASE) / \
3157 (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) + 1))
3158
3159 #define HEAPTEXT_OVERSIZED(addr) \
3160 ((uintptr_t)(addr) >= HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE)
3161
3162 #define HEAPTEXT_IN_NUCLEUSDATA(addr) \
3163 (((uintptr_t)(addr) >= KERNELBASE + 2 * MMU_PAGESIZE4M) && \
3164 ((uintptr_t)(addr) < KERNELBASE + 3 * MMU_PAGESIZE4M))
3165
3166 vmem_t *texthole_source[HEAPTEXT_NARENAS];
3167 vmem_t *texthole_arena[HEAPTEXT_NARENAS];
3168 kmutex_t texthole_lock;
3169
3170 char kern_bootargs[OBP_MAXPATHLEN];
3171 char kern_bootfile[OBP_MAXPATHLEN];
3172
3173 void
3174 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3175 {
3176 uintptr_t addr, limit;
3177
3178 addr = HEAPTEXT_BASE;
3179 limit = addr + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE;
3180
3181 /*
3182 * Before we initialize the text_arena, we want to punch holes in the
3183 * underlying heaptext_arena. This guarantees that for any text
3184 * address we can find a text hole less than HEAPTEXT_MAPPED away.
3185 */
3186 for (; addr + HEAPTEXT_UNMAPPED <= limit;
3187 addr += HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED) {
3188 (void) vmem_xalloc(heaptext_arena, HEAPTEXT_UNMAPPED, PAGESIZE,
3189 0, 0, (void *)addr, (void *)(addr + HEAPTEXT_UNMAPPED),
3190 VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3191 }
3192
3193 /*
3194 * Allocate one page at the oversize to break up the text region
3195 * from the oversized region.
3196 */
3197 (void) vmem_xalloc(heaptext_arena, PAGESIZE, PAGESIZE, 0, 0,
3198 (void *)limit, (void *)(limit + PAGESIZE),
3199 VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
3200
3201 *text_arena = vmem_create("module_text", modtext_sz ? modtext : NULL,
3202 modtext_sz, sizeof (uintptr_t), segkmem_alloc, segkmem_free,
3203 heaptext_arena, 0, VM_SLEEP);
3204 *data_arena = vmem_create("module_data", moddata, MODDATA, 1,
3205 segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3206 }
3207
3208 caddr_t
3209 kobj_text_alloc(vmem_t *arena, size_t size)
3210 {
3211 caddr_t rval, better;
3212
3213 /*
3214 * First, try a sleeping allocation.
3215 */
3216 rval = vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT);
3217
3218 if (size >= HEAPTEXT_MAPPED || !HEAPTEXT_OVERSIZED(rval))
3219 return (rval);
3220
3221 /*
3222 * We didn't get the area that we wanted. We're going to try to do an
3223 * allocation with explicit constraints.
3224 */
3225 better = vmem_xalloc(arena, size, sizeof (uintptr_t), 0, 0, NULL,
3226 (void *)(HEAPTEXT_BASE + HEAPTEXT_SIZE - HEAPTEXT_OVERSIZE),
3227 VM_NOSLEEP | VM_BESTFIT);
3228
3229 if (better != NULL) {
3230 /*
3231 * That worked. Free our first attempt and return.
3232 */
3233 vmem_free(arena, rval, size);
3234 return (better);
3235 }
3236
3237 /*
3238 * That didn't work; we'll have to return our first attempt.
3239 */
3240 return (rval);
3241 }
3242
3243 caddr_t
3244 kobj_texthole_alloc(caddr_t addr, size_t size)
3245 {
3246 int arena = HEAPTEXT_ARENA(addr);
3247 char c[30];
3248 uintptr_t base;
3249
3250 if (HEAPTEXT_OVERSIZED(addr) || HEAPTEXT_IN_NUCLEUSDATA(addr)) {
3251 /*
3252 * If this is an oversized allocation or it is allocated in
3253 * the nucleus data page, there is no text hole available for
3254 * it; return NULL.
3255 */
3256 return (NULL);
3257 }
3258
3259 mutex_enter(&texthole_lock);
3260
3261 if (texthole_arena[arena] == NULL) {
3262 ASSERT(texthole_source[arena] == NULL);
3263
3264 if (arena == 0) {
3265 texthole_source[0] = vmem_create("module_text_holesrc",
3266 (void *)(KERNELBASE + MMU_PAGESIZE4M),
3267 MMU_PAGESIZE4M, PAGESIZE, NULL, NULL, NULL,
3268 0, VM_SLEEP);
3269 } else {
3270 base = HEAPTEXT_BASE +
3271 (arena - 1) * (HEAPTEXT_MAPPED + HEAPTEXT_UNMAPPED);
3272
3273 (void) snprintf(c, sizeof (c),
3274 "heaptext_holesrc_%d", arena);
3275
3276 texthole_source[arena] = vmem_create(c, (void *)base,
3277 HEAPTEXT_UNMAPPED, PAGESIZE, NULL, NULL, NULL,
3278 0, VM_SLEEP);
3279 }
3280
3281 (void) snprintf(c, sizeof (c), "heaptext_hole_%d", arena);
3282
3283 texthole_arena[arena] = vmem_create(c, NULL, 0,
3284 sizeof (uint32_t), segkmem_alloc_permanent, segkmem_free,
3285 texthole_source[arena], 0, VM_SLEEP);
3286 }
3287
3288 mutex_exit(&texthole_lock);
3289
3290 ASSERT(texthole_arena[arena] != NULL);
3291 ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3292 return (vmem_alloc(texthole_arena[arena], size,
3293 VM_BESTFIT | VM_NOSLEEP));
3294 }
3295
3296 void
3297 kobj_texthole_free(caddr_t addr, size_t size)
3298 {
3299 int arena = HEAPTEXT_ARENA(addr);
3300
3301 ASSERT(arena >= 0 && arena < HEAPTEXT_NARENAS);
3302 ASSERT(texthole_arena[arena] != NULL);
3303 vmem_free(texthole_arena[arena], addr, size);
3304 }
3305
3306 void
3307 release_bootstrap(void)
3308 {
3309 if (&cif_init)
3310 cif_init();
3311 }