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 * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2012 DEY Storage Systems, Inc. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
25 */
26 /*
27 * Copyright (c) 2010, Intel Corporation.
28 * All rights reserved.
29 */
30
31 #include <sys/types.h>
32 #include <sys/t_lock.h>
33 #include <sys/param.h>
34 #include <sys/sysmacros.h>
35 #include <sys/signal.h>
36 #include <sys/systm.h>
37 #include <sys/user.h>
38 #include <sys/mman.h>
39 #include <sys/vm.h>
40 #include <sys/conf.h>
41 #include <sys/avintr.h>
42 #include <sys/autoconf.h>
43 #include <sys/disp.h>
44 #include <sys/class.h>
45 #include <sys/bitmap.h>
46
47 #include <sys/privregs.h>
48
49 #include <sys/proc.h>
50 #include <sys/buf.h>
51 #include <sys/kmem.h>
52 #include <sys/mem.h>
53 #include <sys/kstat.h>
54
55 #include <sys/reboot.h>
56
57 #include <sys/cred.h>
58 #include <sys/vnode.h>
59 #include <sys/file.h>
60
61 #include <sys/procfs.h>
62
63 #include <sys/vfs.h>
64 #include <sys/cmn_err.h>
65 #include <sys/utsname.h>
66 #include <sys/debug.h>
67 #include <sys/kdi.h>
68
69 #include <sys/dumphdr.h>
70 #include <sys/bootconf.h>
71 #include <sys/memlist_plat.h>
72 #include <sys/varargs.h>
73 #include <sys/promif.h>
74 #include <sys/modctl.h>
75
76 #include <sys/sunddi.h>
77 #include <sys/sunndi.h>
78 #include <sys/ndi_impldefs.h>
79 #include <sys/ddidmareq.h>
80 #include <sys/psw.h>
81 #include <sys/regset.h>
82 #include <sys/clock.h>
83 #include <sys/pte.h>
84 #include <sys/tss.h>
85 #include <sys/stack.h>
86 #include <sys/trap.h>
87 #include <sys/fp.h>
88 #include <vm/kboot_mmu.h>
89 #include <vm/anon.h>
90 #include <vm/as.h>
91 #include <vm/page.h>
92 #include <vm/seg.h>
93 #include <vm/seg_dev.h>
94 #include <vm/seg_kmem.h>
95 #include <vm/seg_kpm.h>
96 #include <vm/seg_map.h>
97 #include <vm/seg_vn.h>
98 #include <vm/seg_kp.h>
99 #include <sys/memnode.h>
100 #include <vm/vm_dep.h>
101 #include <sys/thread.h>
102 #include <sys/sysconf.h>
103 #include <sys/vm_machparam.h>
104 #include <sys/archsystm.h>
105 #include <sys/machsystm.h>
106 #include <vm/hat.h>
107 #include <vm/hat_i86.h>
108 #include <sys/pmem.h>
109 #include <sys/smp_impldefs.h>
110 #include <sys/x86_archext.h>
111 #include <sys/cpuvar.h>
112 #include <sys/segments.h>
113 #include <sys/clconf.h>
114 #include <sys/kobj.h>
115 #include <sys/kobj_lex.h>
116 #include <sys/cpc_impl.h>
117 #include <sys/cpu_module.h>
118 #include <sys/smbios.h>
119 #include <sys/debug_info.h>
120 #include <sys/bootinfo.h>
121 #include <sys/ddi_timer.h>
122 #include <sys/systeminfo.h>
123 #include <sys/multiboot.h>
124
125 #ifdef __xpv
126
127 #include <sys/hypervisor.h>
128 #include <sys/xen_mmu.h>
129 #include <sys/evtchn_impl.h>
130 #include <sys/gnttab.h>
131 #include <sys/xpv_panic.h>
132 #include <xen/sys/xenbus_comms.h>
133 #include <xen/public/physdev.h>
134
135 extern void xen_late_startup(void);
136
137 struct xen_evt_data cpu0_evt_data;
138
139 #else /* __xpv */
140 #include <sys/memlist_impl.h>
141
142 extern void mem_config_init(void);
143 #endif /* __xpv */
144
145 extern void progressbar_init(void);
146 extern void brand_init(void);
147 extern void pcf_init(void);
148 extern void pg_init(void);
149
150 extern int size_pse_array(pgcnt_t, int);
151
152 #if defined(_SOFT_HOSTID)
153
154 #include <sys/rtc.h>
155
156 static int32_t set_soft_hostid(void);
157 static char hostid_file[] = "/etc/hostid";
158
159 #endif
160
161 void *gfx_devinfo_list;
162
163 #if defined(__amd64) && !defined(__xpv)
164 extern void immu_startup(void);
165 #endif
166
167 /*
168 * XXX make declaration below "static" when drivers no longer use this
169 * interface.
170 */
171 extern caddr_t p0_va; /* Virtual address for accessing physical page 0 */
172
173 /*
174 * segkp
175 */
176 extern int segkp_fromheap;
177
178 static void kvm_init(void);
179 static void startup_init(void);
180 static void startup_memlist(void);
181 static void startup_kmem(void);
182 static void startup_modules(void);
183 static void startup_vm(void);
184 static void startup_end(void);
185 static void layout_kernel_va(void);
186
187 /*
188 * Declare these as initialized data so we can patch them.
189 */
190 #ifdef __i386
191
192 /*
193 * Due to virtual address space limitations running in 32 bit mode, restrict
194 * the amount of physical memory configured to a max of PHYSMEM pages (16g).
195 *
196 * If the physical max memory size of 64g were allowed to be configured, the
197 * size of user virtual address space will be less than 1g. A limited user
198 * address space greatly reduces the range of applications that can run.
199 *
200 * If more physical memory than PHYSMEM is required, users should preferably
201 * run in 64 bit mode which has far looser virtual address space limitations.
202 *
203 * If 64 bit mode is not available (as in IA32) and/or more physical memory
204 * than PHYSMEM is required in 32 bit mode, physmem can be set to the desired
205 * value or to 0 (to configure all available memory) via eeprom(1M). kernelbase
206 * should also be carefully tuned to balance out the need of the user
207 * application while minimizing the risk of kernel heap exhaustion due to
208 * kernelbase being set too high.
209 */
210 #define PHYSMEM 0x400000
211
212 #else /* __amd64 */
213
214 /*
215 * For now we can handle memory with physical addresses up to about
216 * 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly
217 * half the VA space for seg_kpm. When systems get bigger than 64TB this
218 * code will need revisiting. There is an implicit assumption that there
219 * are no *huge* holes in the physical address space too.
220 */
221 #define TERABYTE (1ul << 40)
222 #define PHYSMEM_MAX64 mmu_btop(64 * TERABYTE)
223 #define PHYSMEM PHYSMEM_MAX64
224 #define AMD64_VA_HOLE_END 0xFFFF800000000000ul
225
226 #endif /* __amd64 */
227
228 pgcnt_t physmem = PHYSMEM;
229 pgcnt_t obp_pages; /* Memory used by PROM for its text and data */
230
231 char *kobj_file_buf;
232 int kobj_file_bufsize; /* set in /etc/system */
233
234 /* Global variables for MP support. Used in mp_startup */
235 caddr_t rm_platter_va = 0;
236 uint32_t rm_platter_pa;
237
238 int auto_lpg_disable = 1;
239
240 /*
241 * Some CPUs have holes in the middle of the 64-bit virtual address range.
242 */
243 uintptr_t hole_start, hole_end;
244
245 /*
246 * kpm mapping window
247 */
248 caddr_t kpm_vbase;
249 size_t kpm_size;
250 static int kpm_desired;
251 #ifdef __amd64
252 static uintptr_t segkpm_base = (uintptr_t)SEGKPM_BASE;
253 #endif
254
255 /*
256 * Configuration parameters set at boot time.
257 */
258
259 caddr_t econtig; /* end of first block of contiguous kernel */
260
261 struct bootops *bootops = 0; /* passed in from boot */
262 struct bootops **bootopsp;
263 struct boot_syscalls *sysp; /* passed in from boot */
264
265 char bootblock_fstype[16];
266
267 char kern_bootargs[OBP_MAXPATHLEN];
268 char kern_bootfile[OBP_MAXPATHLEN];
269
270 /*
271 * ZFS zio segment. This allows us to exclude large portions of ZFS data that
272 * gets cached in kmem caches on the heap. If this is set to zero, we allocate
273 * zio buffers from their own segment, otherwise they are allocated from the
274 * heap. The optimization of allocating zio buffers from their own segment is
275 * only valid on 64-bit kernels.
276 */
277 #if defined(__amd64)
278 int segzio_fromheap = 0;
279 #else
280 int segzio_fromheap = 1;
281 #endif
282
283 /*
284 * new memory fragmentations are possible in startup() due to BOP_ALLOCs. this
285 * depends on number of BOP_ALLOC calls made and requested size, memory size
286 * combination and whether boot.bin memory needs to be freed.
287 */
288 #define POSS_NEW_FRAGMENTS 12
289
290 /*
291 * VM data structures
292 */
293 long page_hashsz; /* Size of page hash table (power of two) */
294 unsigned int page_hashsz_shift; /* log2(page_hashsz) */
295 struct page *pp_base; /* Base of initial system page struct array */
296 struct page **page_hash; /* Page hash table */
297 pad_mutex_t *pse_mutex; /* Locks protecting pp->p_selock */
298 size_t pse_table_size; /* Number of mutexes in pse_mutex[] */
299 int pse_shift; /* log2(pse_table_size) */
300 struct seg ktextseg; /* Segment used for kernel executable image */
301 struct seg kvalloc; /* Segment used for "valloc" mapping */
302 struct seg kpseg; /* Segment used for pageable kernel virt mem */
303 struct seg kmapseg; /* Segment used for generic kernel mappings */
304 struct seg kdebugseg; /* Segment used for the kernel debugger */
305
306 struct seg *segkmap = &kmapseg; /* Kernel generic mapping segment */
307 static struct seg *segmap = &kmapseg; /* easier to use name for in here */
308
309 struct seg *segkp = &kpseg; /* Pageable kernel virtual memory segment */
310
311 #if defined(__amd64)
312 struct seg kvseg_core; /* Segment used for the core heap */
313 struct seg kpmseg; /* Segment used for physical mapping */
314 struct seg *segkpm = &kpmseg; /* 64bit kernel physical mapping segment */
315 #else
316 struct seg *segkpm = NULL; /* Unused on IA32 */
317 #endif
318
319 caddr_t segkp_base; /* Base address of segkp */
320 caddr_t segzio_base; /* Base address of segzio */
321 #if defined(__amd64)
322 pgcnt_t segkpsize = btop(SEGKPDEFSIZE); /* size of segkp segment in pages */
323 #else
324 pgcnt_t segkpsize = 0;
325 #endif
326 pgcnt_t segziosize = 0; /* size of zio segment in pages */
327
328 /*
329 * A static DR page_t VA map is reserved that can map the page structures
330 * for a domain's entire RA space. The pages that back this space are
331 * dynamically allocated and need not be physically contiguous. The DR
332 * map size is derived from KPM size.
333 * This mechanism isn't used by x86 yet, so just stubs here.
334 */
335 int ppvm_enable = 0; /* Static virtual map for page structs */
336 page_t *ppvm_base = NULL; /* Base of page struct map */
337 pgcnt_t ppvm_size = 0; /* Size of page struct map */
338
339 /*
340 * VA range available to the debugger
341 */
342 const caddr_t kdi_segdebugbase = (const caddr_t)SEGDEBUGBASE;
343 const size_t kdi_segdebugsize = SEGDEBUGSIZE;
344
345 struct memseg *memseg_base;
346 struct vnode unused_pages_vp;
347
348 #define FOURGB 0x100000000LL
349
350 struct memlist *memlist;
351
352 caddr_t s_text; /* start of kernel text segment */
353 caddr_t e_text; /* end of kernel text segment */
354 caddr_t s_data; /* start of kernel data segment */
355 caddr_t e_data; /* end of kernel data segment */
356 caddr_t modtext; /* start of loadable module text reserved */
357 caddr_t e_modtext; /* end of loadable module text reserved */
358 caddr_t moddata; /* start of loadable module data reserved */
359 caddr_t e_moddata; /* end of loadable module data reserved */
360
361 struct memlist *phys_install; /* Total installed physical memory */
362 struct memlist *phys_avail; /* Total available physical memory */
363 struct memlist *bios_rsvd; /* Bios reserved memory */
364
365 /*
366 * kphysm_init returns the number of pages that were processed
367 */
368 static pgcnt_t kphysm_init(page_t *, pgcnt_t);
369
370 #define IO_PROP_SIZE 64 /* device property size */
371
372 /*
373 * a couple useful roundup macros
374 */
375 #define ROUND_UP_PAGE(x) \
376 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)MMU_PAGESIZE))
377 #define ROUND_UP_LPAGE(x) \
378 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[1]))
379 #define ROUND_UP_4MEG(x) \
380 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), (uintptr_t)FOUR_MEG))
381 #define ROUND_UP_TOPLEVEL(x) \
382 ((uintptr_t)P2ROUNDUP((uintptr_t)(x), mmu.level_size[mmu.max_level]))
383
384 /*
385 * 32-bit Kernel's Virtual memory layout.
386 * +-----------------------+
387 * | |
388 * 0xFFC00000 -|-----------------------|- ARGSBASE
389 * | debugger |
390 * 0xFF800000 -|-----------------------|- SEGDEBUGBASE
391 * | Kernel Data |
392 * 0xFEC00000 -|-----------------------|
393 * | Kernel Text |
394 * 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen)
395 * |--- GDT ---|- GDT page (GDT_VA)
396 * |--- debug info ---|- debug info (DEBUG_INFO_VA)
397 * | |
398 * | page_t structures |
399 * | memsegs, memlists, |
400 * | page hash, etc. |
401 * --- -|-----------------------|- ekernelheap, valloc_base (floating)
402 * | | (segkp is just an arena in the heap)
403 * | |
404 * | kvseg |
405 * | |
406 * | |
407 * --- -|-----------------------|- kernelheap (floating)
408 * | Segkmap |
409 * 0xC3002000 -|-----------------------|- segmap_start (floating)
410 * | Red Zone |
411 * 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating)
412 * | | ||
413 * | Shared objects | \/
414 * | |
415 * : :
416 * | user data |
417 * |-----------------------|
418 * | user text |
419 * 0x08048000 -|-----------------------|
420 * | user stack |
421 * : :
422 * | invalid |
423 * 0x00000000 +-----------------------+
424 *
425 *
426 * 64-bit Kernel's Virtual memory layout. (assuming 64 bit app)
427 * +-----------------------+
428 * | |
429 * 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE
430 * | debugger (?) |
431 * 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE
432 * | unused |
433 * +-----------------------+
434 * | Kernel Data |
435 * 0xFFFFFFFF.FBC00000 |-----------------------|
436 * | Kernel Text |
437 * 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT
438 * |--- GDT ---|- GDT page (GDT_VA)
439 * |--- debug info ---|- debug info (DEBUG_INFO_VA)
440 * | |
441 * | Core heap | (used for loadable modules)
442 * 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap
443 * | Kernel |
444 * | heap |
445 * 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating)
446 * | segmap |
447 * 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating)
448 * | device mappings |
449 * 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating)
450 * | segzio |
451 * 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating)
452 * | segkp |
453 * --- |-----------------------|- segkp_base (floating)
454 * | page_t structures | valloc_base + valloc_sz
455 * | memsegs, memlists, |
456 * | page hash, etc. |
457 * 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB)
458 * | segkpm |
459 * 0xFFFFFE00.00000000 |-----------------------|
460 * | Red Zone |
461 * 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB)
462 * | User stack |- User space memory
463 * | |
464 * | shared objects, etc | (grows downwards)
465 * : :
466 * | |
467 * 0xFFFF8000.00000000 |-----------------------|
468 * | |
469 * | VA Hole / unused |
470 * | |
471 * 0x00008000.00000000 |-----------------------|
472 * | |
473 * | |
474 * : :
475 * | user heap | (grows upwards)
476 * | |
477 * | user data |
478 * |-----------------------|
479 * | user text |
480 * 0x00000000.04000000 |-----------------------|
481 * | invalid |
482 * 0x00000000.00000000 +-----------------------+
483 *
484 * A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit
485 * kernel, except that userlimit is raised to 0xfe000000
486 *
487 * Floating values:
488 *
489 * valloc_base: start of the kernel's memory management/tracking data
490 * structures. This region contains page_t structures for
491 * physical memory, memsegs, memlists, and the page hash.
492 *
493 * core_base: start of the kernel's "core" heap area on 64-bit systems.
494 * This area is intended to be used for global data as well as for module
495 * text/data that does not fit into the nucleus pages. The core heap is
496 * restricted to a 2GB range, allowing every address within it to be
497 * accessed using rip-relative addressing
498 *
499 * ekernelheap: end of kernelheap and start of segmap.
500 *
501 * kernelheap: start of kernel heap. On 32-bit systems, this starts right
502 * above a red zone that separates the user's address space from the
503 * kernel's. On 64-bit systems, it sits above segkp and segkpm.
504 *
505 * segmap_start: start of segmap. The length of segmap can be modified
506 * through eeprom. The default length is 16MB on 32-bit systems and 64MB
507 * on 64-bit systems.
508 *
509 * kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be
510 * decreased by 2X the size required for page_t. This allows the kernel
511 * heap to grow in size with physical memory. With sizeof(page_t) == 80
512 * bytes, the following shows the values of kernelbase and kernel heap
513 * sizes for different memory configurations (assuming default segmap and
514 * segkp sizes).
515 *
516 * mem size for kernelbase kernel heap
517 * size page_t's size
518 * ---- --------- ---------- -----------
519 * 1gb 0x01400000 0xd1800000 684MB
520 * 2gb 0x02800000 0xcf000000 704MB
521 * 4gb 0x05000000 0xca000000 744MB
522 * 6gb 0x07800000 0xc5000000 784MB
523 * 8gb 0x0a000000 0xc0000000 824MB
524 * 16gb 0x14000000 0xac000000 984MB
525 * 32gb 0x28000000 0x84000000 1304MB
526 * 64gb 0x50000000 0x34000000 1944MB (*)
527 *
528 * kernelbase is less than the abi minimum of 0xc0000000 for memory
529 * configurations above 8gb.
530 *
531 * (*) support for memory configurations above 32gb will require manual tuning
532 * of kernelbase to balance out the need of user applications.
533 */
534
535 /* real-time-clock initialization parameters */
536 extern time_t process_rtc_config_file(void);
537
538 uintptr_t kernelbase;
539 uintptr_t postbootkernelbase; /* not set till boot loader is gone */
540 uintptr_t eprom_kernelbase;
541 size_t segmapsize;
542 uintptr_t segmap_start;
543 int segmapfreelists;
544 pgcnt_t npages;
545 pgcnt_t orig_npages;
546 size_t core_size; /* size of "core" heap */
547 uintptr_t core_base; /* base address of "core" heap */
548
549 /*
550 * List of bootstrap pages. We mark these as allocated in startup.
551 * release_bootstrap() will free them when we're completely done with
552 * the bootstrap.
553 */
554 static page_t *bootpages;
555
556 /*
557 * boot time pages that have a vnode from the ramdisk will keep that forever.
558 */
559 static page_t *rd_pages;
560
561 /*
562 * Lower 64K
563 */
564 static page_t *lower_pages = NULL;
565 static int lower_pages_count = 0;
566
567 struct system_hardware system_hardware;
568
569 /*
570 * Enable some debugging messages concerning memory usage...
571 */
572 static void
573 print_memlist(char *title, struct memlist *mp)
574 {
575 prom_printf("MEMLIST: %s:\n", title);
576 while (mp != NULL) {
577 prom_printf("\tAddress 0x%" PRIx64 ", size 0x%" PRIx64 "\n",
578 mp->ml_address, mp->ml_size);
579 mp = mp->ml_next;
580 }
581 }
582
583 /*
584 * XX64 need a comment here.. are these just default values, surely
585 * we read the "cpuid" type information to figure this out.
586 */
587 int l2cache_sz = 0x80000;
588 int l2cache_linesz = 0x40;
589 int l2cache_assoc = 1;
590
591 static size_t textrepl_min_gb = 10;
592
593 /*
594 * on 64 bit we use a predifined VA range for mapping devices in the kernel
595 * on 32 bit the mappings are intermixed in the heap, so we use a bit map
596 */
597 #ifdef __amd64
598
599 vmem_t *device_arena;
600 uintptr_t toxic_addr = (uintptr_t)NULL;
601 size_t toxic_size = 1024 * 1024 * 1024; /* Sparc uses 1 gig too */
602
603 #else /* __i386 */
604
605 ulong_t *toxic_bit_map; /* one bit for each 4k of VA in heap_arena */
606 size_t toxic_bit_map_len = 0; /* in bits */
607
608 #endif /* __i386 */
609
610 /*
611 * Simple boot time debug facilities
612 */
613 static char *prm_dbg_str[] = {
614 "%s:%d: '%s' is 0x%x\n",
615 "%s:%d: '%s' is 0x%llx\n"
616 };
617
618 int prom_debug;
619
620 #define PRM_DEBUG(q) if (prom_debug) \
621 prom_printf(prm_dbg_str[sizeof (q) >> 3], "startup.c", __LINE__, #q, q);
622 #define PRM_POINT(q) if (prom_debug) \
623 prom_printf("%s:%d: %s\n", "startup.c", __LINE__, q);
624
625 /*
626 * This structure is used to keep track of the intial allocations
627 * done in startup_memlist(). The value of NUM_ALLOCATIONS needs to
628 * be >= the number of ADD_TO_ALLOCATIONS() executed in the code.
629 */
630 #define NUM_ALLOCATIONS 8
631 int num_allocations = 0;
632 struct {
633 void **al_ptr;
634 size_t al_size;
635 } allocations[NUM_ALLOCATIONS];
636 size_t valloc_sz = 0;
637 uintptr_t valloc_base;
638
639 #define ADD_TO_ALLOCATIONS(ptr, size) { \
640 size = ROUND_UP_PAGE(size); \
641 if (num_allocations == NUM_ALLOCATIONS) \
642 panic("too many ADD_TO_ALLOCATIONS()"); \
643 allocations[num_allocations].al_ptr = (void**)&ptr; \
644 allocations[num_allocations].al_size = size; \
645 valloc_sz += size; \
646 ++num_allocations; \
647 }
648
649 /*
650 * Allocate all the initial memory needed by the page allocator.
651 */
652 static void
653 perform_allocations(void)
654 {
655 caddr_t mem;
656 int i;
657 int valloc_align;
658
659 PRM_DEBUG(valloc_base);
660 PRM_DEBUG(valloc_sz);
661 valloc_align = mmu.level_size[mmu.max_page_level > 0];
662 mem = BOP_ALLOC(bootops, (caddr_t)valloc_base, valloc_sz, valloc_align);
663 if (mem != (caddr_t)valloc_base)
664 panic("BOP_ALLOC() failed");
665 bzero(mem, valloc_sz);
666 for (i = 0; i < num_allocations; ++i) {
667 *allocations[i].al_ptr = (void *)mem;
668 mem += allocations[i].al_size;
669 }
670 }
671
672 /*
673 * Our world looks like this at startup time.
674 *
675 * In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data
676 * at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at
677 * 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those
678 * addresses are fixed in the binary at link time.
679 *
680 * On the text page:
681 * unix/genunix/krtld/module text loads.
682 *
683 * On the data page:
684 * unix/genunix/krtld/module data loads.
685 *
686 * Machine-dependent startup code
687 */
688 void
689 startup(void)
690 {
691 #if !defined(__xpv)
692 extern void startup_pci_bios(void);
693 #endif
694 extern cpuset_t cpu_ready_set;
695
696 /*
697 * Make sure that nobody tries to use sekpm until we have
698 * initialized it properly.
699 */
700 #if defined(__amd64)
701 kpm_desired = 1;
702 #endif
703 kpm_enable = 0;
704 CPUSET_ONLY(cpu_ready_set, 0); /* cpu 0 is boot cpu */
705
706 #if defined(__xpv) /* XXPV fix me! */
707 {
708 extern int segvn_use_regions;
709 segvn_use_regions = 0;
710 }
711 #endif
712 progressbar_init();
713 startup_init();
714 #if defined(__xpv)
715 startup_xen_version();
716 #endif
717 startup_memlist();
718 startup_kmem();
719 startup_vm();
720 #if !defined(__xpv)
721 /*
722 * Note we need to do this even on fast reboot in order to access
723 * the irq routing table (used for pci labels).
724 */
725 startup_pci_bios();
726 #endif
727 #if defined(__xpv)
728 startup_xen_mca();
729 #endif
730 startup_modules();
731
732 startup_end();
733 }
734
735 static void
736 startup_init()
737 {
738 PRM_POINT("startup_init() starting...");
739
740 /*
741 * Complete the extraction of cpuid data
742 */
743 cpuid_pass2(CPU);
744
745 (void) check_boot_version(BOP_GETVERSION(bootops));
746
747 /*
748 * Check for prom_debug in boot environment
749 */
750 if (BOP_GETPROPLEN(bootops, "prom_debug") >= 0) {
751 ++prom_debug;
752 PRM_POINT("prom_debug found in boot enviroment");
753 }
754
755 /*
756 * Collect node, cpu and memory configuration information.
757 */
758 get_system_configuration();
759
760 /*
761 * Halt if this is an unsupported processor.
762 */
763 if (x86_type == X86_TYPE_486 || x86_type == X86_TYPE_CYRIX_486) {
764 printf("\n486 processor (\"%s\") detected.\n",
765 CPU->cpu_brandstr);
766 halt("This processor is not supported by this release "
767 "of Solaris.");
768 }
769
770 PRM_POINT("startup_init() done");
771 }
772
773 /*
774 * Callback for copy_memlist_filter() to filter nucleus, kadb/kmdb, (ie.
775 * everything mapped above KERNEL_TEXT) pages from phys_avail. Note it
776 * also filters out physical page zero. There is some reliance on the
777 * boot loader allocating only a few contiguous physical memory chunks.
778 */
779 static void
780 avail_filter(uint64_t *addr, uint64_t *size)
781 {
782 uintptr_t va;
783 uintptr_t next_va;
784 pfn_t pfn;
785 uint64_t pfn_addr;
786 uint64_t pfn_eaddr;
787 uint_t prot;
788 size_t len;
789 uint_t change;
790
791 if (prom_debug)
792 prom_printf("\tFilter: in: a=%" PRIx64 ", s=%" PRIx64 "\n",
793 *addr, *size);
794
795 /*
796 * page zero is required for BIOS.. never make it available
797 */
798 if (*addr == 0) {
799 *addr += MMU_PAGESIZE;
800 *size -= MMU_PAGESIZE;
801 }
802
803 /*
804 * First we trim from the front of the range. Since kbm_probe()
805 * walks ranges in virtual order, but addr/size are physical, we need
806 * to the list until no changes are seen. This deals with the case
807 * where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w
808 * but w < v.
809 */
810 do {
811 change = 0;
812 for (va = KERNEL_TEXT;
813 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
814 va = next_va) {
815
816 next_va = va + len;
817 pfn_addr = pfn_to_pa(pfn);
818 pfn_eaddr = pfn_addr + len;
819
820 if (pfn_addr <= *addr && pfn_eaddr > *addr) {
821 change = 1;
822 while (*size > 0 && len > 0) {
823 *addr += MMU_PAGESIZE;
824 *size -= MMU_PAGESIZE;
825 len -= MMU_PAGESIZE;
826 }
827 }
828 }
829 if (change && prom_debug)
830 prom_printf("\t\ttrim: a=%" PRIx64 ", s=%" PRIx64 "\n",
831 *addr, *size);
832 } while (change);
833
834 /*
835 * Trim pages from the end of the range.
836 */
837 for (va = KERNEL_TEXT;
838 *size > 0 && kbm_probe(&va, &len, &pfn, &prot) != 0;
839 va = next_va) {
840
841 next_va = va + len;
842 pfn_addr = pfn_to_pa(pfn);
843
844 if (pfn_addr >= *addr && pfn_addr < *addr + *size)
845 *size = pfn_addr - *addr;
846 }
847
848 if (prom_debug)
849 prom_printf("\tFilter out: a=%" PRIx64 ", s=%" PRIx64 "\n",
850 *addr, *size);
851 }
852
853 static void
854 kpm_init()
855 {
856 struct segkpm_crargs b;
857
858 /*
859 * These variables were all designed for sfmmu in which segkpm is
860 * mapped using a single pagesize - either 8KB or 4MB. On x86, we
861 * might use 2+ page sizes on a single machine, so none of these
862 * variables have a single correct value. They are set up as if we
863 * always use a 4KB pagesize, which should do no harm. In the long
864 * run, we should get rid of KPM's assumption that only a single
865 * pagesize is used.
866 */
867 kpm_pgshft = MMU_PAGESHIFT;
868 kpm_pgsz = MMU_PAGESIZE;
869 kpm_pgoff = MMU_PAGEOFFSET;
870 kpmp2pshft = 0;
871 kpmpnpgs = 1;
872 ASSERT(((uintptr_t)kpm_vbase & (kpm_pgsz - 1)) == 0);
873
874 PRM_POINT("about to create segkpm");
875 rw_enter(&kas.a_lock, RW_WRITER);
876
877 if (seg_attach(&kas, kpm_vbase, kpm_size, segkpm) < 0)
878 panic("cannot attach segkpm");
879
880 b.prot = PROT_READ | PROT_WRITE;
881 b.nvcolors = 1;
882
883 if (segkpm_create(segkpm, (caddr_t)&b) != 0)
884 panic("segkpm_create segkpm");
885
886 rw_exit(&kas.a_lock);
887 }
888
889 /*
890 * The debug info page provides enough information to allow external
891 * inspectors (e.g. when running under a hypervisor) to bootstrap
892 * themselves into allowing full-blown kernel debugging.
893 */
894 static void
895 init_debug_info(void)
896 {
897 caddr_t mem;
898 debug_info_t *di;
899
900 #ifndef __lint
901 ASSERT(sizeof (debug_info_t) < MMU_PAGESIZE);
902 #endif
903
904 mem = BOP_ALLOC(bootops, (caddr_t)DEBUG_INFO_VA, MMU_PAGESIZE,
905 MMU_PAGESIZE);
906
907 if (mem != (caddr_t)DEBUG_INFO_VA)
908 panic("BOP_ALLOC() failed");
909 bzero(mem, MMU_PAGESIZE);
910
911 di = (debug_info_t *)mem;
912
913 di->di_magic = DEBUG_INFO_MAGIC;
914 di->di_version = DEBUG_INFO_VERSION;
915 di->di_modules = (uintptr_t)&modules;
916 di->di_s_text = (uintptr_t)s_text;
917 di->di_e_text = (uintptr_t)e_text;
918 di->di_s_data = (uintptr_t)s_data;
919 di->di_e_data = (uintptr_t)e_data;
920 di->di_hat_htable_off = offsetof(hat_t, hat_htable);
921 di->di_ht_pfn_off = offsetof(htable_t, ht_pfn);
922 }
923
924 /*
925 * Build the memlists and other kernel essential memory system data structures.
926 * This is everything at valloc_base.
927 */
928 static void
929 startup_memlist(void)
930 {
931 size_t memlist_sz;
932 size_t memseg_sz;
933 size_t pagehash_sz;
934 size_t pp_sz;
935 uintptr_t va;
936 size_t len;
937 uint_t prot;
938 pfn_t pfn;
939 int memblocks;
940 pfn_t rsvd_high_pfn;
941 pgcnt_t rsvd_pgcnt;
942 size_t rsvdmemlist_sz;
943 int rsvdmemblocks;
944 caddr_t pagecolor_mem;
945 size_t pagecolor_memsz;
946 caddr_t page_ctrs_mem;
947 size_t page_ctrs_size;
948 size_t pse_table_alloc_size;
949 struct memlist *current;
950 extern void startup_build_mem_nodes(struct memlist *);
951
952 /* XX64 fix these - they should be in include files */
953 extern size_t page_coloring_init(uint_t, int, int);
954 extern void page_coloring_setup(caddr_t);
955
956 PRM_POINT("startup_memlist() starting...");
957
958 /*
959 * Use leftover large page nucleus text/data space for loadable modules.
960 * Use at most MODTEXT/MODDATA.
961 */
962 len = kbm_nucleus_size;
963 ASSERT(len > MMU_PAGESIZE);
964
965 moddata = (caddr_t)ROUND_UP_PAGE(e_data);
966 e_moddata = (caddr_t)P2ROUNDUP((uintptr_t)e_data, (uintptr_t)len);
967 if (e_moddata - moddata > MODDATA)
968 e_moddata = moddata + MODDATA;
969
970 modtext = (caddr_t)ROUND_UP_PAGE(e_text);
971 e_modtext = (caddr_t)P2ROUNDUP((uintptr_t)e_text, (uintptr_t)len);
972 if (e_modtext - modtext > MODTEXT)
973 e_modtext = modtext + MODTEXT;
974
975 econtig = e_moddata;
976
977 PRM_DEBUG(modtext);
978 PRM_DEBUG(e_modtext);
979 PRM_DEBUG(moddata);
980 PRM_DEBUG(e_moddata);
981 PRM_DEBUG(econtig);
982
983 /*
984 * Examine the boot loader physical memory map to find out:
985 * - total memory in system - physinstalled
986 * - the max physical address - physmax
987 * - the number of discontiguous segments of memory.
988 */
989 if (prom_debug)
990 print_memlist("boot physinstalled",
991 bootops->boot_mem->physinstalled);
992 installed_top_size_ex(bootops->boot_mem->physinstalled, &physmax,
993 &physinstalled, &memblocks);
994 PRM_DEBUG(physmax);
995 PRM_DEBUG(physinstalled);
996 PRM_DEBUG(memblocks);
997
998 /*
999 * Compute maximum physical address for memory DR operations.
1000 * Memory DR operations are unsupported on xpv or 32bit OSes.
1001 */
1002 #ifdef __amd64
1003 if (plat_dr_support_memory()) {
1004 if (plat_dr_physmax == 0) {
1005 uint_t pabits = UINT_MAX;
1006
1007 cpuid_get_addrsize(CPU, &pabits, NULL);
1008 plat_dr_physmax = btop(1ULL << pabits);
1009 }
1010 if (plat_dr_physmax > PHYSMEM_MAX64)
1011 plat_dr_physmax = PHYSMEM_MAX64;
1012 } else
1013 #endif
1014 plat_dr_physmax = 0;
1015
1016 /*
1017 * Examine the bios reserved memory to find out:
1018 * - the number of discontiguous segments of memory.
1019 */
1020 if (prom_debug)
1021 print_memlist("boot reserved mem",
1022 bootops->boot_mem->rsvdmem);
1023 installed_top_size_ex(bootops->boot_mem->rsvdmem, &rsvd_high_pfn,
1024 &rsvd_pgcnt, &rsvdmemblocks);
1025 PRM_DEBUG(rsvd_high_pfn);
1026 PRM_DEBUG(rsvd_pgcnt);
1027 PRM_DEBUG(rsvdmemblocks);
1028
1029 /*
1030 * Initialize hat's mmu parameters.
1031 * Check for enforce-prot-exec in boot environment. It's used to
1032 * enable/disable support for the page table entry NX bit.
1033 * The default is to enforce PROT_EXEC on processors that support NX.
1034 * Boot seems to round up the "len", but 8 seems to be big enough.
1035 */
1036 mmu_init();
1037
1038 #ifdef __i386
1039 /*
1040 * physmax is lowered if there is more memory than can be
1041 * physically addressed in 32 bit (PAE/non-PAE) modes.
1042 */
1043 if (mmu.pae_hat) {
1044 if (PFN_ABOVE64G(physmax)) {
1045 physinstalled -= (physmax - (PFN_64G - 1));
1046 physmax = PFN_64G - 1;
1047 }
1048 } else {
1049 if (PFN_ABOVE4G(physmax)) {
1050 physinstalled -= (physmax - (PFN_4G - 1));
1051 physmax = PFN_4G - 1;
1052 }
1053 }
1054 #endif
1055
1056 startup_build_mem_nodes(bootops->boot_mem->physinstalled);
1057
1058 if (BOP_GETPROPLEN(bootops, "enforce-prot-exec") >= 0) {
1059 int len = BOP_GETPROPLEN(bootops, "enforce-prot-exec");
1060 char value[8];
1061
1062 if (len < 8)
1063 (void) BOP_GETPROP(bootops, "enforce-prot-exec", value);
1064 else
1065 (void) strcpy(value, "");
1066 if (strcmp(value, "off") == 0)
1067 mmu.pt_nx = 0;
1068 }
1069 PRM_DEBUG(mmu.pt_nx);
1070
1071 /*
1072 * We will need page_t's for every page in the system, except for
1073 * memory mapped at or above above the start of the kernel text segment.
1074 *
1075 * pages above e_modtext are attributed to kernel debugger (obp_pages)
1076 */
1077 npages = physinstalled - 1; /* avail_filter() skips page 0, so "- 1" */
1078 obp_pages = 0;
1079 va = KERNEL_TEXT;
1080 while (kbm_probe(&va, &len, &pfn, &prot) != 0) {
1081 npages -= len >> MMU_PAGESHIFT;
1082 if (va >= (uintptr_t)e_moddata)
1083 obp_pages += len >> MMU_PAGESHIFT;
1084 va += len;
1085 }
1086 PRM_DEBUG(npages);
1087 PRM_DEBUG(obp_pages);
1088
1089 /*
1090 * If physmem is patched to be non-zero, use it instead of the computed
1091 * value unless it is larger than the actual amount of memory on hand.
1092 */
1093 if (physmem == 0 || physmem > npages) {
1094 physmem = npages;
1095 } else if (physmem < npages) {
1096 orig_npages = npages;
1097 npages = physmem;
1098 }
1099 PRM_DEBUG(physmem);
1100
1101 /*
1102 * We now compute the sizes of all the initial allocations for
1103 * structures the kernel needs in order do kmem_alloc(). These
1104 * include:
1105 * memsegs
1106 * memlists
1107 * page hash table
1108 * page_t's
1109 * page coloring data structs
1110 */
1111 memseg_sz = sizeof (struct memseg) * (memblocks + POSS_NEW_FRAGMENTS);
1112 ADD_TO_ALLOCATIONS(memseg_base, memseg_sz);
1113 PRM_DEBUG(memseg_sz);
1114
1115 /*
1116 * Reserve space for memlists. There's no real good way to know exactly
1117 * how much room we'll need, but this should be a good upper bound.
1118 */
1119 memlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1120 (memblocks + POSS_NEW_FRAGMENTS));
1121 ADD_TO_ALLOCATIONS(memlist, memlist_sz);
1122 PRM_DEBUG(memlist_sz);
1123
1124 /*
1125 * Reserve space for bios reserved memlists.
1126 */
1127 rsvdmemlist_sz = ROUND_UP_PAGE(2 * sizeof (struct memlist) *
1128 (rsvdmemblocks + POSS_NEW_FRAGMENTS));
1129 ADD_TO_ALLOCATIONS(bios_rsvd, rsvdmemlist_sz);
1130 PRM_DEBUG(rsvdmemlist_sz);
1131
1132 /* LINTED */
1133 ASSERT(P2SAMEHIGHBIT((1 << PP_SHIFT), sizeof (struct page)));
1134 /*
1135 * The page structure hash table size is a power of 2
1136 * such that the average hash chain length is PAGE_HASHAVELEN.
1137 */
1138 page_hashsz = npages / PAGE_HASHAVELEN;
1139 page_hashsz_shift = highbit(page_hashsz);
1140 page_hashsz = 1 << page_hashsz_shift;
1141 pagehash_sz = sizeof (struct page *) * page_hashsz;
1142 ADD_TO_ALLOCATIONS(page_hash, pagehash_sz);
1143 PRM_DEBUG(pagehash_sz);
1144
1145 /*
1146 * Set aside room for the page structures themselves.
1147 */
1148 PRM_DEBUG(npages);
1149 pp_sz = sizeof (struct page) * npages;
1150 ADD_TO_ALLOCATIONS(pp_base, pp_sz);
1151 PRM_DEBUG(pp_sz);
1152
1153 /*
1154 * determine l2 cache info and memory size for page coloring
1155 */
1156 (void) getl2cacheinfo(CPU,
1157 &l2cache_sz, &l2cache_linesz, &l2cache_assoc);
1158 pagecolor_memsz =
1159 page_coloring_init(l2cache_sz, l2cache_linesz, l2cache_assoc);
1160 ADD_TO_ALLOCATIONS(pagecolor_mem, pagecolor_memsz);
1161 PRM_DEBUG(pagecolor_memsz);
1162
1163 page_ctrs_size = page_ctrs_sz();
1164 ADD_TO_ALLOCATIONS(page_ctrs_mem, page_ctrs_size);
1165 PRM_DEBUG(page_ctrs_size);
1166
1167 /*
1168 * Allocate the array that protects pp->p_selock.
1169 */
1170 pse_shift = size_pse_array(physmem, max_ncpus);
1171 pse_table_size = 1 << pse_shift;
1172 pse_table_alloc_size = pse_table_size * sizeof (pad_mutex_t);
1173 ADD_TO_ALLOCATIONS(pse_mutex, pse_table_alloc_size);
1174
1175 #if defined(__amd64)
1176 valloc_sz = ROUND_UP_LPAGE(valloc_sz);
1177 valloc_base = VALLOC_BASE;
1178
1179 /*
1180 * The default values of VALLOC_BASE and SEGKPM_BASE should work
1181 * for values of physmax up to 1 Terabyte. They need adjusting when
1182 * memory is at addresses above 1 TB. When adjusted, segkpm_base must
1183 * be aligned on KERNEL_REDZONE_SIZE boundary (span of top level pte).
1184 */
1185 if (physmax + 1 > mmu_btop(TERABYTE) ||
1186 plat_dr_physmax > mmu_btop(TERABYTE)) {
1187 uint64_t kpm_resv_amount = mmu_ptob(physmax + 1);
1188
1189 if (kpm_resv_amount < mmu_ptob(plat_dr_physmax)) {
1190 kpm_resv_amount = mmu_ptob(plat_dr_physmax);
1191 }
1192
1193 segkpm_base = -(P2ROUNDUP((2 * kpm_resv_amount),
1194 KERNEL_REDZONE_SIZE)); /* down from top VA */
1195
1196 /* make sure we leave some space for user apps above hole */
1197 segkpm_base = MAX(segkpm_base, AMD64_VA_HOLE_END + TERABYTE);
1198 if (segkpm_base > SEGKPM_BASE)
1199 segkpm_base = SEGKPM_BASE;
1200 PRM_DEBUG(segkpm_base);
1201
1202 valloc_base = segkpm_base + P2ROUNDUP(kpm_resv_amount, ONE_GIG);
1203 if (valloc_base < segkpm_base)
1204 panic("not enough kernel VA to support memory size");
1205 PRM_DEBUG(valloc_base);
1206 }
1207 #else /* __i386 */
1208 valloc_base = (uintptr_t)(MISC_VA_BASE - valloc_sz);
1209 valloc_base = P2ALIGN(valloc_base, mmu.level_size[1]);
1210 PRM_DEBUG(valloc_base);
1211 #endif /* __i386 */
1212
1213 /*
1214 * do all the initial allocations
1215 */
1216 perform_allocations();
1217
1218 /*
1219 * Build phys_install and phys_avail in kernel memspace.
1220 * - phys_install should be all memory in the system.
1221 * - phys_avail is phys_install minus any memory mapped before this
1222 * point above KERNEL_TEXT.
1223 */
1224 current = phys_install = memlist;
1225 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t, NULL);
1226 if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1227 panic("physinstalled was too big!");
1228 if (prom_debug)
1229 print_memlist("phys_install", phys_install);
1230
1231 phys_avail = current;
1232 PRM_POINT("Building phys_avail:\n");
1233 copy_memlist_filter(bootops->boot_mem->physinstalled, ¤t,
1234 avail_filter);
1235 if ((caddr_t)current > (caddr_t)memlist + memlist_sz)
1236 panic("physavail was too big!");
1237 if (prom_debug)
1238 print_memlist("phys_avail", phys_avail);
1239 #ifndef __xpv
1240 /*
1241 * Free unused memlist items, which may be used by memory DR driver
1242 * at runtime.
1243 */
1244 if ((caddr_t)current < (caddr_t)memlist + memlist_sz) {
1245 memlist_free_block((caddr_t)current,
1246 (caddr_t)memlist + memlist_sz - (caddr_t)current);
1247 }
1248 #endif
1249
1250 /*
1251 * Build bios reserved memspace
1252 */
1253 current = bios_rsvd;
1254 copy_memlist_filter(bootops->boot_mem->rsvdmem, ¤t, NULL);
1255 if ((caddr_t)current > (caddr_t)bios_rsvd + rsvdmemlist_sz)
1256 panic("bios_rsvd was too big!");
1257 if (prom_debug)
1258 print_memlist("bios_rsvd", bios_rsvd);
1259 #ifndef __xpv
1260 /*
1261 * Free unused memlist items, which may be used by memory DR driver
1262 * at runtime.
1263 */
1264 if ((caddr_t)current < (caddr_t)bios_rsvd + rsvdmemlist_sz) {
1265 memlist_free_block((caddr_t)current,
1266 (caddr_t)bios_rsvd + rsvdmemlist_sz - (caddr_t)current);
1267 }
1268 #endif
1269
1270 /*
1271 * setup page coloring
1272 */
1273 page_coloring_setup(pagecolor_mem);
1274 page_lock_init(); /* currently a no-op */
1275
1276 /*
1277 * free page list counters
1278 */
1279 (void) page_ctrs_alloc(page_ctrs_mem);
1280
1281 /*
1282 * Size the pcf array based on the number of cpus in the box at
1283 * boot time.
1284 */
1285
1286 pcf_init();
1287
1288 /*
1289 * Initialize the page structures from the memory lists.
1290 */
1291 availrmem_initial = availrmem = freemem = 0;
1292 PRM_POINT("Calling kphysm_init()...");
1293 npages = kphysm_init(pp_base, npages);
1294 PRM_POINT("kphysm_init() done");
1295 PRM_DEBUG(npages);
1296
1297 init_debug_info();
1298
1299 /*
1300 * Now that page_t's have been initialized, remove all the
1301 * initial allocation pages from the kernel free page lists.
1302 */
1303 boot_mapin((caddr_t)valloc_base, valloc_sz);
1304 boot_mapin((caddr_t)MISC_VA_BASE, MISC_VA_SIZE);
1305 PRM_POINT("startup_memlist() done");
1306
1307 PRM_DEBUG(valloc_sz);
1308
1309 #if defined(__amd64)
1310 if ((availrmem >> (30 - MMU_PAGESHIFT)) >=
1311 textrepl_min_gb && l2cache_sz <= 2 << 20) {
1312 extern size_t textrepl_size_thresh;
1313 textrepl_size_thresh = (16 << 20) - 1;
1314 }
1315 #endif
1316 }
1317
1318 /*
1319 * Layout the kernel's part of address space and initialize kmem allocator.
1320 */
1321 static void
1322 startup_kmem(void)
1323 {
1324 extern void page_set_colorequiv_arr(void);
1325
1326 PRM_POINT("startup_kmem() starting...");
1327
1328 #if defined(__amd64)
1329 if (eprom_kernelbase && eprom_kernelbase != KERNELBASE)
1330 cmn_err(CE_NOTE, "!kernelbase cannot be changed on 64-bit "
1331 "systems.");
1332 kernelbase = segkpm_base - KERNEL_REDZONE_SIZE;
1333 core_base = (uintptr_t)COREHEAP_BASE;
1334 core_size = (size_t)MISC_VA_BASE - COREHEAP_BASE;
1335 #else /* __i386 */
1336 /*
1337 * We configure kernelbase based on:
1338 *
1339 * 1. user specified kernelbase via eeprom command. Value cannot exceed
1340 * KERNELBASE_MAX. we large page align eprom_kernelbase
1341 *
1342 * 2. Default to KERNELBASE and adjust to 2X less the size for page_t.
1343 * On large memory systems we must lower kernelbase to allow
1344 * enough room for page_t's for all of memory.
1345 *
1346 * The value set here, might be changed a little later.
1347 */
1348 if (eprom_kernelbase) {
1349 kernelbase = eprom_kernelbase & mmu.level_mask[1];
1350 if (kernelbase > KERNELBASE_MAX)
1351 kernelbase = KERNELBASE_MAX;
1352 } else {
1353 kernelbase = (uintptr_t)KERNELBASE;
1354 kernelbase -= ROUND_UP_4MEG(2 * valloc_sz);
1355 }
1356 ASSERT((kernelbase & mmu.level_offset[1]) == 0);
1357 core_base = valloc_base;
1358 core_size = 0;
1359 #endif /* __i386 */
1360
1361 PRM_DEBUG(core_base);
1362 PRM_DEBUG(core_size);
1363 PRM_DEBUG(kernelbase);
1364
1365 #if defined(__i386)
1366 segkp_fromheap = 1;
1367 #endif /* __i386 */
1368
1369 ekernelheap = (char *)core_base;
1370 PRM_DEBUG(ekernelheap);
1371
1372 /*
1373 * Now that we know the real value of kernelbase,
1374 * update variables that were initialized with a value of
1375 * KERNELBASE (in common/conf/param.c).
1376 *
1377 * XXX The problem with this sort of hackery is that the
1378 * compiler just may feel like putting the const declarations
1379 * (in param.c) into the .text section. Perhaps they should
1380 * just be declared as variables there?
1381 */
1382
1383 *(uintptr_t *)&_kernelbase = kernelbase;
1384 *(uintptr_t *)&_userlimit = kernelbase;
1385 #if defined(__amd64)
1386 *(uintptr_t *)&_userlimit -= KERNELBASE - USERLIMIT;
1387 #else
1388 *(uintptr_t *)&_userlimit32 = _userlimit;
1389 #endif
1390 PRM_DEBUG(_kernelbase);
1391 PRM_DEBUG(_userlimit);
1392 PRM_DEBUG(_userlimit32);
1393
1394 layout_kernel_va();
1395
1396 #if defined(__i386)
1397 /*
1398 * If segmap is too large we can push the bottom of the kernel heap
1399 * higher than the base. Or worse, it could exceed the top of the
1400 * VA space entirely, causing it to wrap around.
1401 */
1402 if (kernelheap >= ekernelheap || (uintptr_t)kernelheap < kernelbase)
1403 panic("too little address space available for kernelheap,"
1404 " use eeprom for lower kernelbase or smaller segmapsize");
1405 #endif /* __i386 */
1406
1407 /*
1408 * Initialize the kernel heap. Note 3rd argument must be > 1st.
1409 */
1410 kernelheap_init(kernelheap, ekernelheap,
1411 kernelheap + MMU_PAGESIZE,
1412 (void *)core_base, (void *)(core_base + core_size));
1413
1414 #if defined(__xpv)
1415 /*
1416 * Link pending events struct into cpu struct
1417 */
1418 CPU->cpu_m.mcpu_evt_pend = &cpu0_evt_data;
1419 #endif
1420 /*
1421 * Initialize kernel memory allocator.
1422 */
1423 kmem_init();
1424
1425 /*
1426 * Factor in colorequiv to check additional 'equivalent' bins
1427 */
1428 page_set_colorequiv_arr();
1429
1430 /*
1431 * print this out early so that we know what's going on
1432 */
1433 print_x86_featureset(x86_featureset);
1434
1435 /*
1436 * Initialize bp_mapin().
1437 */
1438 bp_init(MMU_PAGESIZE, HAT_STORECACHING_OK);
1439
1440 /*
1441 * orig_npages is non-zero if physmem has been configured for less
1442 * than the available memory.
1443 */
1444 if (orig_npages) {
1445 cmn_err(CE_WARN, "!%slimiting physmem to 0x%lx of 0x%lx pages",
1446 (npages == PHYSMEM ? "Due to virtual address space " : ""),
1447 npages, orig_npages);
1448 }
1449 #if defined(__i386)
1450 if (eprom_kernelbase && (eprom_kernelbase != kernelbase))
1451 cmn_err(CE_WARN, "kernelbase value, User specified 0x%lx, "
1452 "System using 0x%lx",
1453 (uintptr_t)eprom_kernelbase, (uintptr_t)kernelbase);
1454 #endif
1455
1456 #ifdef KERNELBASE_ABI_MIN
1457 if (kernelbase < (uintptr_t)KERNELBASE_ABI_MIN) {
1458 cmn_err(CE_NOTE, "!kernelbase set to 0x%lx, system is not "
1459 "i386 ABI compliant.", (uintptr_t)kernelbase);
1460 }
1461 #endif
1462
1463 #ifndef __xpv
1464 if (plat_dr_support_memory()) {
1465 mem_config_init();
1466 }
1467 #else /* __xpv */
1468 /*
1469 * Some of the xen start information has to be relocated up
1470 * into the kernel's permanent address space.
1471 */
1472 PRM_POINT("calling xen_relocate_start_info()");
1473 xen_relocate_start_info();
1474 PRM_POINT("xen_relocate_start_info() done");
1475
1476 /*
1477 * (Update the vcpu pointer in our cpu structure to point into
1478 * the relocated shared info.)
1479 */
1480 CPU->cpu_m.mcpu_vcpu_info =
1481 &HYPERVISOR_shared_info->vcpu_info[CPU->cpu_id];
1482 #endif /* __xpv */
1483
1484 PRM_POINT("startup_kmem() done");
1485 }
1486
1487 #ifndef __xpv
1488 /*
1489 * If we have detected that we are running in an HVM environment, we need
1490 * to prepend the PV driver directory to the module search path.
1491 */
1492 #define HVM_MOD_DIR "/platform/i86hvm/kernel"
1493 static void
1494 update_default_path()
1495 {
1496 char *current, *newpath;
1497 int newlen;
1498
1499 /*
1500 * We are about to resync with krtld. krtld will reset its
1501 * internal module search path iff Solaris has set default_path.
1502 * We want to be sure we're prepending this new directory to the
1503 * right search path.
1504 */
1505 current = (default_path == NULL) ? kobj_module_path : default_path;
1506
1507 newlen = strlen(HVM_MOD_DIR) + strlen(current) + 2;
1508 newpath = kmem_alloc(newlen, KM_SLEEP);
1509 (void) strcpy(newpath, HVM_MOD_DIR);
1510 (void) strcat(newpath, " ");
1511 (void) strcat(newpath, current);
1512
1513 default_path = newpath;
1514 }
1515 #endif
1516
1517 static void
1518 startup_modules(void)
1519 {
1520 int cnt;
1521 extern void prom_setup(void);
1522 int32_t v, h;
1523 char d[11];
1524 char *cp;
1525 cmi_hdl_t hdl;
1526
1527 PRM_POINT("startup_modules() starting...");
1528
1529 #ifndef __xpv
1530 /*
1531 * Initialize ten-micro second timer so that drivers will
1532 * not get short changed in their init phase. This was
1533 * not getting called until clkinit which, on fast cpu's
1534 * caused the drv_usecwait to be way too short.
1535 */
1536 microfind();
1537
1538 if ((get_hwenv() & HW_XEN_HVM) != 0)
1539 update_default_path();
1540 #endif
1541
1542 /*
1543 * Read the GMT lag from /etc/rtc_config.
1544 */
1545 sgmtl(process_rtc_config_file());
1546
1547 /*
1548 * Calculate default settings of system parameters based upon
1549 * maxusers, yet allow to be overridden via the /etc/system file.
1550 */
1551 param_calc(0);
1552
1553 mod_setup();
1554
1555 /*
1556 * Initialize system parameters.
1557 */
1558 param_init();
1559
1560 /*
1561 * Initialize the default brands
1562 */
1563 brand_init();
1564
1565 /*
1566 * maxmem is the amount of physical memory we're playing with.
1567 */
1568 maxmem = physmem;
1569
1570 /*
1571 * Initialize segment management stuff.
1572 */
1573 seg_init();
1574
1575 if (modload("fs", "specfs") == -1)
1576 halt("Can't load specfs");
1577
1578 if (modload("fs", "devfs") == -1)
1579 halt("Can't load devfs");
1580
1581 if (modload("fs", "dev") == -1)
1582 halt("Can't load dev");
1583
1584 if (modload("fs", "procfs") == -1)
1585 halt("Can't load procfs");
1586
1587 (void) modloadonly("sys", "lbl_edition");
1588
1589 dispinit();
1590
1591 /* Read cluster configuration data. */
1592 clconf_init();
1593
1594 #if defined(__xpv)
1595 (void) ec_init();
1596 gnttab_init();
1597 (void) xs_early_init();
1598 #endif /* __xpv */
1599
1600 /*
1601 * Create a kernel device tree. First, create rootnex and
1602 * then invoke bus specific code to probe devices.
1603 */
1604 setup_ddi();
1605
1606 #ifdef __xpv
1607 if (DOMAIN_IS_INITDOMAIN(xen_info))
1608 #endif
1609 {
1610 /*
1611 * Load the System Management BIOS into the global ksmbios
1612 * handle, if an SMBIOS is present on this system.
1613 */
1614 ksmbios = smbios_open(NULL, SMB_VERSION, ksmbios_flags, NULL);
1615 }
1616
1617
1618 /*
1619 * Originally clconf_init() apparently needed the hostid. But
1620 * this no longer appears to be true - it uses its own nodeid.
1621 * By placing the hostid logic here, we are able to make use of
1622 * the SMBIOS UUID.
1623 */
1624 if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) {
1625 cmn_err(CE_WARN, "Unable to set hostid");
1626 } else {
1627 for (v = h, cnt = 0; cnt < 10; cnt++) {
1628 d[cnt] = (char)(v % 10);
1629 v /= 10;
1630 if (v == 0)
1631 break;
1632 }
1633 for (cp = hw_serial; cnt >= 0; cnt--)
1634 *cp++ = d[cnt] + '0';
1635 *cp = 0;
1636 }
1637
1638 /*
1639 * Set up the CPU module subsystem for the boot cpu in the native
1640 * case, and all physical cpu resource in the xpv dom0 case.
1641 * Modifies the device tree, so this must be done after
1642 * setup_ddi().
1643 */
1644 #ifdef __xpv
1645 /*
1646 * If paravirtualized and on dom0 then we initialize all physical
1647 * cpu handles now; if paravirtualized on a domU then do not
1648 * initialize.
1649 */
1650 if (DOMAIN_IS_INITDOMAIN(xen_info)) {
1651 xen_mc_lcpu_cookie_t cpi;
1652
1653 for (cpi = xen_physcpu_next(NULL); cpi != NULL;
1654 cpi = xen_physcpu_next(cpi)) {
1655 if ((hdl = cmi_init(CMI_HDL_SOLARIS_xVM_MCA,
1656 xen_physcpu_chipid(cpi), xen_physcpu_coreid(cpi),
1657 xen_physcpu_strandid(cpi))) != NULL &&
1658 is_x86_feature(x86_featureset, X86FSET_MCA))
1659 cmi_mca_init(hdl);
1660 }
1661 }
1662 #else
1663 /*
1664 * Initialize a handle for the boot cpu - others will initialize
1665 * as they startup. Do not do this if we know we are in an HVM domU.
1666 */
1667 if ((get_hwenv() & HW_XEN_HVM) == 0 &&
1668 (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
1669 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL &&
1670 is_x86_feature(x86_featureset, X86FSET_MCA)) {
1671 cmi_mca_init(hdl);
1672 CPU->cpu_m.mcpu_cmi_hdl = hdl;
1673 }
1674 #endif /* __xpv */
1675
1676 /*
1677 * Fake a prom tree such that /dev/openprom continues to work
1678 */
1679 PRM_POINT("startup_modules: calling prom_setup...");
1680 prom_setup();
1681 PRM_POINT("startup_modules: done");
1682
1683 /*
1684 * Load all platform specific modules
1685 */
1686 PRM_POINT("startup_modules: calling psm_modload...");
1687 psm_modload();
1688
1689 PRM_POINT("startup_modules() done");
1690 }
1691
1692 /*
1693 * claim a "setaside" boot page for use in the kernel
1694 */
1695 page_t *
1696 boot_claim_page(pfn_t pfn)
1697 {
1698 page_t *pp;
1699
1700 pp = page_numtopp_nolock(pfn);
1701 ASSERT(pp != NULL);
1702
1703 if (PP_ISBOOTPAGES(pp)) {
1704 if (pp->p_next != NULL)
1705 pp->p_next->p_prev = pp->p_prev;
1706 if (pp->p_prev == NULL)
1707 bootpages = pp->p_next;
1708 else
1709 pp->p_prev->p_next = pp->p_next;
1710 } else {
1711 /*
1712 * htable_attach() expects a base pagesize page
1713 */
1714 if (pp->p_szc != 0)
1715 page_boot_demote(pp);
1716 pp = page_numtopp(pfn, SE_EXCL);
1717 }
1718 return (pp);
1719 }
1720
1721 /*
1722 * Walk through the pagetables looking for pages mapped in by boot. If the
1723 * setaside flag is set the pages are expected to be returned to the
1724 * kernel later in boot, so we add them to the bootpages list.
1725 */
1726 static void
1727 protect_boot_range(uintptr_t low, uintptr_t high, int setaside)
1728 {
1729 uintptr_t va = low;
1730 size_t len;
1731 uint_t prot;
1732 pfn_t pfn;
1733 page_t *pp;
1734 pgcnt_t boot_protect_cnt = 0;
1735
1736 while (kbm_probe(&va, &len, &pfn, &prot) != 0 && va < high) {
1737 if (va + len >= high)
1738 panic("0x%lx byte mapping at 0x%p exceeds boot's "
1739 "legal range.", len, (void *)va);
1740
1741 while (len > 0) {
1742 pp = page_numtopp_alloc(pfn);
1743 if (pp != NULL) {
1744 if (setaside == 0)
1745 panic("Unexpected mapping by boot. "
1746 "addr=%p pfn=%lx\n",
1747 (void *)va, pfn);
1748
1749 pp->p_next = bootpages;
1750 pp->p_prev = NULL;
1751 PP_SETBOOTPAGES(pp);
1752 if (bootpages != NULL) {
1753 bootpages->p_prev = pp;
1754 }
1755 bootpages = pp;
1756 ++boot_protect_cnt;
1757 }
1758
1759 ++pfn;
1760 len -= MMU_PAGESIZE;
1761 va += MMU_PAGESIZE;
1762 }
1763 }
1764 PRM_DEBUG(boot_protect_cnt);
1765 }
1766
1767 /*
1768 *
1769 */
1770 static void
1771 layout_kernel_va(void)
1772 {
1773 PRM_POINT("layout_kernel_va() starting...");
1774 /*
1775 * Establish the final size of the kernel's heap, size of segmap,
1776 * segkp, etc.
1777 */
1778
1779 #if defined(__amd64)
1780
1781 kpm_vbase = (caddr_t)segkpm_base;
1782 if (physmax + 1 < plat_dr_physmax) {
1783 kpm_size = ROUND_UP_LPAGE(mmu_ptob(plat_dr_physmax));
1784 } else {
1785 kpm_size = ROUND_UP_LPAGE(mmu_ptob(physmax + 1));
1786 }
1787 if ((uintptr_t)kpm_vbase + kpm_size > (uintptr_t)valloc_base)
1788 panic("not enough room for kpm!");
1789 PRM_DEBUG(kpm_size);
1790 PRM_DEBUG(kpm_vbase);
1791
1792 /*
1793 * By default we create a seg_kp in 64 bit kernels, it's a little
1794 * faster to access than embedding it in the heap.
1795 */
1796 segkp_base = (caddr_t)valloc_base + valloc_sz;
1797 if (!segkp_fromheap) {
1798 size_t sz = mmu_ptob(segkpsize);
1799
1800 /*
1801 * determine size of segkp
1802 */
1803 if (sz < SEGKPMINSIZE || sz > SEGKPMAXSIZE) {
1804 sz = SEGKPDEFSIZE;
1805 cmn_err(CE_WARN, "!Illegal value for segkpsize. "
1806 "segkpsize has been reset to %ld pages",
1807 mmu_btop(sz));
1808 }
1809 sz = MIN(sz, MAX(SEGKPMINSIZE, mmu_ptob(physmem)));
1810
1811 segkpsize = mmu_btop(ROUND_UP_LPAGE(sz));
1812 }
1813 PRM_DEBUG(segkp_base);
1814 PRM_DEBUG(segkpsize);
1815
1816 /*
1817 * segzio is used for ZFS cached data. It uses a distinct VA
1818 * segment (from kernel heap) so that we can easily tell not to
1819 * include it in kernel crash dumps on 64 bit kernels. The trick is
1820 * to give it lots of VA, but not constrain the kernel heap.
1821 * We scale the size of segzio linearly with physmem up to
1822 * SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem.
1823 */
1824 segzio_base = segkp_base + mmu_ptob(segkpsize);
1825 if (segzio_fromheap) {
1826 segziosize = 0;
1827 } else {
1828 size_t physmem_size = mmu_ptob(physmem);
1829 size_t size = (segziosize == 0) ?
1830 physmem_size : mmu_ptob(segziosize);
1831
1832 if (size < SEGZIOMINSIZE)
1833 size = SEGZIOMINSIZE;
1834 if (size > SEGZIOMAXSIZE) {
1835 size = SEGZIOMAXSIZE;
1836 if (physmem_size > size)
1837 size += (physmem_size - size) / 2;
1838 }
1839 segziosize = mmu_btop(ROUND_UP_LPAGE(size));
1840 }
1841 PRM_DEBUG(segziosize);
1842 PRM_DEBUG(segzio_base);
1843
1844 /*
1845 * Put the range of VA for device mappings next, kmdb knows to not
1846 * grep in this range of addresses.
1847 */
1848 toxic_addr =
1849 ROUND_UP_LPAGE((uintptr_t)segzio_base + mmu_ptob(segziosize));
1850 PRM_DEBUG(toxic_addr);
1851 segmap_start = ROUND_UP_LPAGE(toxic_addr + toxic_size);
1852 #else /* __i386 */
1853 segmap_start = ROUND_UP_LPAGE(kernelbase);
1854 #endif /* __i386 */
1855 PRM_DEBUG(segmap_start);
1856
1857 /*
1858 * Users can change segmapsize through eeprom. If the variable
1859 * is tuned through eeprom, there is no upper bound on the
1860 * size of segmap.
1861 */
1862 segmapsize = MAX(ROUND_UP_LPAGE(segmapsize), SEGMAPDEFAULT);
1863
1864 #if defined(__i386)
1865 /*
1866 * 32-bit systems don't have segkpm or segkp, so segmap appears at
1867 * the bottom of the kernel's address range. Set aside space for a
1868 * small red zone just below the start of segmap.
1869 */
1870 segmap_start += KERNEL_REDZONE_SIZE;
1871 segmapsize -= KERNEL_REDZONE_SIZE;
1872 #endif
1873
1874 PRM_DEBUG(segmap_start);
1875 PRM_DEBUG(segmapsize);
1876 kernelheap = (caddr_t)ROUND_UP_LPAGE(segmap_start + segmapsize);
1877 PRM_DEBUG(kernelheap);
1878 PRM_POINT("layout_kernel_va() done...");
1879 }
1880
1881 /*
1882 * Finish initializing the VM system, now that we are no longer
1883 * relying on the boot time memory allocators.
1884 */
1885 static void
1886 startup_vm(void)
1887 {
1888 struct segmap_crargs a;
1889
1890 extern int use_brk_lpg, use_stk_lpg;
1891
1892 PRM_POINT("startup_vm() starting...");
1893
1894 /*
1895 * Initialize the hat layer.
1896 */
1897 hat_init();
1898
1899 /*
1900 * Do final allocations of HAT data structures that need to
1901 * be allocated before quiescing the boot loader.
1902 */
1903 PRM_POINT("Calling hat_kern_alloc()...");
1904 hat_kern_alloc((caddr_t)segmap_start, segmapsize, ekernelheap);
1905 PRM_POINT("hat_kern_alloc() done");
1906
1907 #ifndef __xpv
1908 /*
1909 * Setup Page Attribute Table
1910 */
1911 pat_sync();
1912 #endif
1913
1914 /*
1915 * The next two loops are done in distinct steps in order
1916 * to be sure that any page that is doubly mapped (both above
1917 * KERNEL_TEXT and below kernelbase) is dealt with correctly.
1918 * Note this may never happen, but it might someday.
1919 */
1920 bootpages = NULL;
1921 PRM_POINT("Protecting boot pages");
1922
1923 /*
1924 * Protect any pages mapped above KERNEL_TEXT that somehow have
1925 * page_t's. This can only happen if something weird allocated
1926 * in this range (like kadb/kmdb).
1927 */
1928 protect_boot_range(KERNEL_TEXT, (uintptr_t)-1, 0);
1929
1930 /*
1931 * Before we can take over memory allocation/mapping from the boot
1932 * loader we must remove from our free page lists any boot allocated
1933 * pages that stay mapped until release_bootstrap().
1934 */
1935 protect_boot_range(0, kernelbase, 1);
1936
1937
1938 /*
1939 * Switch to running on regular HAT (not boot_mmu)
1940 */
1941 PRM_POINT("Calling hat_kern_setup()...");
1942 hat_kern_setup();
1943
1944 /*
1945 * It is no longer safe to call BOP_ALLOC(), so make sure we don't.
1946 */
1947 bop_no_more_mem();
1948
1949 PRM_POINT("hat_kern_setup() done");
1950
1951 hat_cpu_online(CPU);
1952
1953 /*
1954 * Initialize VM system
1955 */
1956 PRM_POINT("Calling kvm_init()...");
1957 kvm_init();
1958 PRM_POINT("kvm_init() done");
1959
1960 /*
1961 * Tell kmdb that the VM system is now working
1962 */
1963 if (boothowto & RB_DEBUG)
1964 kdi_dvec_vmready();
1965
1966 #if defined(__xpv)
1967 /*
1968 * Populate the I/O pool on domain 0
1969 */
1970 if (DOMAIN_IS_INITDOMAIN(xen_info)) {
1971 extern long populate_io_pool(void);
1972 long init_io_pool_cnt;
1973
1974 PRM_POINT("Populating reserve I/O page pool");
1975 init_io_pool_cnt = populate_io_pool();
1976 PRM_DEBUG(init_io_pool_cnt);
1977 }
1978 #endif
1979 /*
1980 * Mangle the brand string etc.
1981 */
1982 cpuid_pass3(CPU);
1983
1984 #if defined(__amd64)
1985
1986 /*
1987 * Create the device arena for toxic (to dtrace/kmdb) mappings.
1988 */
1989 device_arena = vmem_create("device", (void *)toxic_addr,
1990 toxic_size, MMU_PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP);
1991
1992 #else /* __i386 */
1993
1994 /*
1995 * allocate the bit map that tracks toxic pages
1996 */
1997 toxic_bit_map_len = btop((ulong_t)(valloc_base - kernelbase));
1998 PRM_DEBUG(toxic_bit_map_len);
1999 toxic_bit_map =
2000 kmem_zalloc(BT_SIZEOFMAP(toxic_bit_map_len), KM_NOSLEEP);
2001 ASSERT(toxic_bit_map != NULL);
2002 PRM_DEBUG(toxic_bit_map);
2003
2004 #endif /* __i386 */
2005
2006
2007 /*
2008 * Now that we've got more VA, as well as the ability to allocate from
2009 * it, tell the debugger.
2010 */
2011 if (boothowto & RB_DEBUG)
2012 kdi_dvec_memavail();
2013
2014 /*
2015 * The following code installs a special page fault handler (#pf)
2016 * to work around a pentium bug.
2017 */
2018 #if !defined(__amd64) && !defined(__xpv)
2019 if (x86_type == X86_TYPE_P5) {
2020 desctbr_t idtr;
2021 gate_desc_t *newidt;
2022
2023 if ((newidt = kmem_zalloc(MMU_PAGESIZE, KM_NOSLEEP)) == NULL)
2024 panic("failed to install pentium_pftrap");
2025
2026 bcopy(idt0, newidt, NIDT * sizeof (*idt0));
2027 set_gatesegd(&newidt[T_PGFLT], &pentium_pftrap,
2028 KCS_SEL, SDT_SYSIGT, TRP_KPL, 0);
2029
2030 (void) as_setprot(&kas, (caddr_t)newidt, MMU_PAGESIZE,
2031 PROT_READ | PROT_EXEC);
2032
2033 CPU->cpu_idt = newidt;
2034 idtr.dtr_base = (uintptr_t)CPU->cpu_idt;
2035 idtr.dtr_limit = (NIDT * sizeof (*idt0)) - 1;
2036 wr_idtr(&idtr);
2037 }
2038 #endif /* !__amd64 */
2039
2040 #if !defined(__xpv)
2041 /*
2042 * Map page pfn=0 for drivers, such as kd, that need to pick up
2043 * parameters left there by controllers/BIOS.
2044 */
2045 PRM_POINT("setup up p0_va");
2046 p0_va = i86devmap(0, 1, PROT_READ);
2047 PRM_DEBUG(p0_va);
2048 #endif
2049
2050 cmn_err(CE_CONT, "?mem = %luK (0x%lx)\n",
2051 physinstalled << (MMU_PAGESHIFT - 10), ptob(physinstalled));
2052
2053 /*
2054 * disable automatic large pages for small memory systems or
2055 * when the disable flag is set.
2056 *
2057 * Do not yet consider page sizes larger than 2m/4m.
2058 */
2059 if (!auto_lpg_disable && mmu.max_page_level > 0) {
2060 max_uheap_lpsize = LEVEL_SIZE(1);
2061 max_ustack_lpsize = LEVEL_SIZE(1);
2062 max_privmap_lpsize = LEVEL_SIZE(1);
2063 max_uidata_lpsize = LEVEL_SIZE(1);
2064 max_utext_lpsize = LEVEL_SIZE(1);
2065 max_shm_lpsize = LEVEL_SIZE(1);
2066 }
2067 if (physmem < privm_lpg_min_physmem || mmu.max_page_level == 0 ||
2068 auto_lpg_disable) {
2069 use_brk_lpg = 0;
2070 use_stk_lpg = 0;
2071 }
2072 mcntl0_lpsize = LEVEL_SIZE(mmu.umax_page_level);
2073
2074 PRM_POINT("Calling hat_init_finish()...");
2075 hat_init_finish();
2076 PRM_POINT("hat_init_finish() done");
2077
2078 /*
2079 * Initialize the segkp segment type.
2080 */
2081 rw_enter(&kas.a_lock, RW_WRITER);
2082 PRM_POINT("Attaching segkp");
2083 if (segkp_fromheap) {
2084 segkp->s_as = &kas;
2085 } else if (seg_attach(&kas, (caddr_t)segkp_base, mmu_ptob(segkpsize),
2086 segkp) < 0) {
2087 panic("startup: cannot attach segkp");
2088 /*NOTREACHED*/
2089 }
2090 PRM_POINT("Doing segkp_create()");
2091 if (segkp_create(segkp) != 0) {
2092 panic("startup: segkp_create failed");
2093 /*NOTREACHED*/
2094 }
2095 PRM_DEBUG(segkp);
2096 rw_exit(&kas.a_lock);
2097
2098 /*
2099 * kpm segment
2100 */
2101 segmap_kpm = 0;
2102 if (kpm_desired) {
2103 kpm_init();
2104 kpm_enable = 1;
2105 }
2106
2107 /*
2108 * Now create segmap segment.
2109 */
2110 rw_enter(&kas.a_lock, RW_WRITER);
2111 if (seg_attach(&kas, (caddr_t)segmap_start, segmapsize, segmap) < 0) {
2112 panic("cannot attach segmap");
2113 /*NOTREACHED*/
2114 }
2115 PRM_DEBUG(segmap);
2116
2117 a.prot = PROT_READ | PROT_WRITE;
2118 a.shmsize = 0;
2119 a.nfreelist = segmapfreelists;
2120
2121 if (segmap_create(segmap, (caddr_t)&a) != 0)
2122 panic("segmap_create segmap");
2123 rw_exit(&kas.a_lock);
2124
2125 setup_vaddr_for_ppcopy(CPU);
2126
2127 segdev_init();
2128 #if defined(__xpv)
2129 if (DOMAIN_IS_INITDOMAIN(xen_info))
2130 #endif
2131 pmem_init();
2132
2133 PRM_POINT("startup_vm() done");
2134 }
2135
2136 /*
2137 * Load a tod module for the non-standard tod part found on this system.
2138 */
2139 static void
2140 load_tod_module(char *todmod)
2141 {
2142 if (modload("tod", todmod) == -1)
2143 halt("Can't load TOD module");
2144 }
2145
2146 static void
2147 startup_end(void)
2148 {
2149 int i;
2150 extern void setx86isalist(void);
2151 extern void cpu_event_init(void);
2152
2153 PRM_POINT("startup_end() starting...");
2154
2155 /*
2156 * Perform tasks that get done after most of the VM
2157 * initialization has been done but before the clock
2158 * and other devices get started.
2159 */
2160 kern_setup1();
2161
2162 /*
2163 * Perform CPC initialization for this CPU.
2164 */
2165 kcpc_hw_init(CPU);
2166
2167 /*
2168 * Initialize cpu event framework.
2169 */
2170 cpu_event_init();
2171
2172 #if defined(OPTERON_WORKAROUND_6323525)
2173 if (opteron_workaround_6323525)
2174 patch_workaround_6323525();
2175 #endif
2176 /*
2177 * If needed, load TOD module now so that ddi_get_time(9F) etc. work
2178 * (For now, "needed" is defined as set tod_module_name in /etc/system)
2179 */
2180 if (tod_module_name != NULL) {
2181 PRM_POINT("load_tod_module()");
2182 load_tod_module(tod_module_name);
2183 }
2184
2185 #if defined(__xpv)
2186 /*
2187 * Forceload interposing TOD module for the hypervisor.
2188 */
2189 PRM_POINT("load_tod_module()");
2190 load_tod_module("xpvtod");
2191 #endif
2192
2193 /*
2194 * Configure the system.
2195 */
2196 PRM_POINT("Calling configure()...");
2197 configure(); /* set up devices */
2198 PRM_POINT("configure() done");
2199
2200 /*
2201 * We can now setup for XSAVE because fpu_probe is done in configure().
2202 */
2203 if (fp_save_mech == FP_XSAVE) {
2204 xsave_setup_msr(CPU);
2205 }
2206
2207 /*
2208 * Set the isa_list string to the defined instruction sets we
2209 * support.
2210 */
2211 setx86isalist();
2212 cpu_intr_alloc(CPU, NINTR_THREADS);
2213 psm_install();
2214
2215 /*
2216 * We're done with bootops. We don't unmap the bootstrap yet because
2217 * we're still using bootsvcs.
2218 */
2219 PRM_POINT("NULLing out bootops");
2220 *bootopsp = (struct bootops *)NULL;
2221 bootops = (struct bootops *)NULL;
2222
2223 #if defined(__xpv)
2224 ec_init_debug_irq();
2225 xs_domu_init();
2226 #endif
2227
2228 #if defined(__amd64) && !defined(__xpv)
2229 /*
2230 * Intel IOMMU has been setup/initialized in ddi_impl.c
2231 * Start it up now.
2232 */
2233 immu_startup();
2234 #endif
2235
2236 PRM_POINT("Enabling interrupts");
2237 (*picinitf)();
2238 sti();
2239 #if defined(__xpv)
2240 ASSERT(CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0);
2241 xen_late_startup();
2242 #endif
2243
2244 (void) add_avsoftintr((void *)&softlevel1_hdl, 1, softlevel1,
2245 "softlevel1", NULL, NULL); /* XXX to be moved later */
2246
2247 /*
2248 * Register these software interrupts for ddi timer.
2249 * Software interrupts up to the level 10 are supported.
2250 */
2251 for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) {
2252 char name[sizeof ("timer_softintr") + 2];
2253 (void) sprintf(name, "timer_softintr%02d", i);
2254 (void) add_avsoftintr((void *)&softlevel_hdl[i-1], i,
2255 (avfunc)timer_softintr, name, (caddr_t)(uintptr_t)i, NULL);
2256 }
2257
2258 #if !defined(__xpv)
2259 if (modload("drv", "amd_iommu") < 0) {
2260 PRM_POINT("No AMD IOMMU present\n");
2261 } else if (ddi_hold_installed_driver(ddi_name_to_major(
2262 "amd_iommu")) == NULL) {
2263 prom_printf("ERROR: failed to attach AMD IOMMU\n");
2264 }
2265 #endif
2266 post_startup_cpu_fixups();
2267
2268 PRM_POINT("startup_end() done");
2269 }
2270
2271 /*
2272 * Don't remove the following 2 variables. They are necessary
2273 * for reading the hostid from the legacy file (/kernel/misc/sysinit).
2274 */
2275 char *_hs1107 = hw_serial;
2276 ulong_t _bdhs34;
2277
2278 void
2279 post_startup(void)
2280 {
2281 extern void cpupm_init(cpu_t *);
2282 extern void cpu_event_init_cpu(cpu_t *);
2283
2284 /*
2285 * Set the system wide, processor-specific flags to be passed
2286 * to userland via the aux vector for performance hints and
2287 * instruction set extensions.
2288 */
2289 bind_hwcap();
2290
2291 #ifdef __xpv
2292 if (DOMAIN_IS_INITDOMAIN(xen_info))
2293 #endif
2294 {
2295 #if defined(__xpv)
2296 xpv_panic_init();
2297 #else
2298 /*
2299 * Startup the memory scrubber.
2300 * XXPV This should be running somewhere ..
2301 */
2302 if ((get_hwenv() & HW_VIRTUAL) == 0)
2303 memscrub_init();
2304 #endif
2305 }
2306
2307 /*
2308 * Complete CPU module initialization
2309 */
2310 cmi_post_startup();
2311
2312 /*
2313 * Perform forceloading tasks for /etc/system.
2314 */
2315 (void) mod_sysctl(SYS_FORCELOAD, NULL);
2316
2317 /*
2318 * ON4.0: Force /proc module in until clock interrupt handle fixed
2319 * ON4.0: This must be fixed or restated in /etc/systems.
2320 */
2321 (void) modload("fs", "procfs");
2322
2323 (void) i_ddi_attach_hw_nodes("pit_beep");
2324
2325 #if defined(__i386)
2326 /*
2327 * Check for required functional Floating Point hardware,
2328 * unless FP hardware explicitly disabled.
2329 */
2330 if (fpu_exists && (fpu_pentium_fdivbug || fp_kind == FP_NO))
2331 halt("No working FP hardware found");
2332 #endif
2333
2334 maxmem = freemem;
2335
2336 cpu_event_init_cpu(CPU);
2337 cpupm_init(CPU);
2338 (void) mach_cpu_create_device_node(CPU, NULL);
2339
2340 pg_init();
2341 }
2342
2343 static int
2344 pp_in_range(page_t *pp, uint64_t low_addr, uint64_t high_addr)
2345 {
2346 return ((pp->p_pagenum >= btop(low_addr)) &&
2347 (pp->p_pagenum < btopr(high_addr)));
2348 }
2349
2350 void
2351 release_bootstrap(void)
2352 {
2353 int root_is_ramdisk;
2354 page_t *pp;
2355 extern void kobj_boot_unmountroot(void);
2356 extern dev_t rootdev;
2357 #if !defined(__xpv)
2358 pfn_t pfn;
2359 #endif
2360
2361 /* unmount boot ramdisk and release kmem usage */
2362 kobj_boot_unmountroot();
2363
2364 /*
2365 * We're finished using the boot loader so free its pages.
2366 */
2367 PRM_POINT("Unmapping lower boot pages");
2368
2369 clear_boot_mappings(0, _userlimit);
2370
2371 postbootkernelbase = kernelbase;
2372
2373 /*
2374 * If root isn't on ramdisk, destroy the hardcoded
2375 * ramdisk node now and release the memory. Else,
2376 * ramdisk memory is kept in rd_pages.
2377 */
2378 root_is_ramdisk = (getmajor(rootdev) == ddi_name_to_major("ramdisk"));
2379 if (!root_is_ramdisk) {
2380 dev_info_t *dip = ddi_find_devinfo("ramdisk", -1, 0);
2381 ASSERT(dip && ddi_get_parent(dip) == ddi_root_node());
2382 ndi_rele_devi(dip); /* held from ddi_find_devinfo */
2383 (void) ddi_remove_child(dip, 0);
2384 }
2385
2386 PRM_POINT("Releasing boot pages");
2387 while (bootpages) {
2388 extern uint64_t ramdisk_start, ramdisk_end;
2389 pp = bootpages;
2390 bootpages = pp->p_next;
2391
2392
2393 /* Keep pages for the lower 64K */
2394 if (pp_in_range(pp, 0, 0x40000)) {
2395 pp->p_next = lower_pages;
2396 lower_pages = pp;
2397 lower_pages_count++;
2398 continue;
2399 }
2400
2401
2402 if (root_is_ramdisk && pp_in_range(pp, ramdisk_start,
2403 ramdisk_end)) {
2404 pp->p_next = rd_pages;
2405 rd_pages = pp;
2406 continue;
2407 }
2408 pp->p_next = (struct page *)0;
2409 pp->p_prev = (struct page *)0;
2410 PP_CLRBOOTPAGES(pp);
2411 page_free(pp, 1);
2412 }
2413 PRM_POINT("Boot pages released");
2414
2415 #if !defined(__xpv)
2416 /* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */
2417 /*
2418 * Find 1 page below 1 MB so that other processors can boot up or
2419 * so that any processor can resume.
2420 * Make sure it has a kernel VA as well as a 1:1 mapping.
2421 * We should have just free'd one up.
2422 */
2423
2424 /*
2425 * 0x10 pages is 64K. Leave the bottom 64K alone
2426 * for BIOS.
2427 */
2428 for (pfn = 0x10; pfn < btop(1*1024*1024); pfn++) {
2429 if (page_numtopp_alloc(pfn) == NULL)
2430 continue;
2431 rm_platter_va = i86devmap(pfn, 1,
2432 PROT_READ | PROT_WRITE | PROT_EXEC);
2433 rm_platter_pa = ptob(pfn);
2434 break;
2435 }
2436 if (pfn == btop(1*1024*1024) && use_mp)
2437 panic("No page below 1M available for starting "
2438 "other processors or for resuming from system-suspend");
2439 #endif /* !__xpv */
2440 }
2441
2442 /*
2443 * Initialize the platform-specific parts of a page_t.
2444 */
2445 void
2446 add_physmem_cb(page_t *pp, pfn_t pnum)
2447 {
2448 pp->p_pagenum = pnum;
2449 pp->p_mapping = NULL;
2450 pp->p_embed = 0;
2451 pp->p_share = 0;
2452 pp->p_mlentry = 0;
2453 }
2454
2455 /*
2456 * kphysm_init() initializes physical memory.
2457 */
2458 static pgcnt_t
2459 kphysm_init(
2460 page_t *pp,
2461 pgcnt_t npages)
2462 {
2463 struct memlist *pmem;
2464 struct memseg *cur_memseg;
2465 pfn_t base_pfn;
2466 pfn_t end_pfn;
2467 pgcnt_t num;
2468 pgcnt_t pages_done = 0;
2469 uint64_t addr;
2470 uint64_t size;
2471 extern pfn_t ddiphysmin;
2472 extern int mnode_xwa;
2473 int ms = 0, me = 0;
2474
2475 ASSERT(page_hash != NULL && page_hashsz != 0);
2476
2477 cur_memseg = memseg_base;
2478 for (pmem = phys_avail; pmem && npages; pmem = pmem->ml_next) {
2479 /*
2480 * In a 32 bit kernel can't use higher memory if we're
2481 * not booting in PAE mode. This check takes care of that.
2482 */
2483 addr = pmem->ml_address;
2484 size = pmem->ml_size;
2485 if (btop(addr) > physmax)
2486 continue;
2487
2488 /*
2489 * align addr and size - they may not be at page boundaries
2490 */
2491 if ((addr & MMU_PAGEOFFSET) != 0) {
2492 addr += MMU_PAGEOFFSET;
2493 addr &= ~(uint64_t)MMU_PAGEOFFSET;
2494 size -= addr - pmem->ml_address;
2495 }
2496
2497 /* only process pages below or equal to physmax */
2498 if ((btop(addr + size) - 1) > physmax)
2499 size = ptob(physmax - btop(addr) + 1);
2500
2501 num = btop(size);
2502 if (num == 0)
2503 continue;
2504
2505 if (num > npages)
2506 num = npages;
2507
2508 npages -= num;
2509 pages_done += num;
2510 base_pfn = btop(addr);
2511
2512 if (prom_debug)
2513 prom_printf("MEMSEG addr=0x%" PRIx64
2514 " pgs=0x%lx pfn 0x%lx-0x%lx\n",
2515 addr, num, base_pfn, base_pfn + num);
2516
2517 /*
2518 * Ignore pages below ddiphysmin to simplify ddi memory
2519 * allocation with non-zero addr_lo requests.
2520 */
2521 if (base_pfn < ddiphysmin) {
2522 if (base_pfn + num <= ddiphysmin)
2523 continue;
2524 pp += (ddiphysmin - base_pfn);
2525 num -= (ddiphysmin - base_pfn);
2526 base_pfn = ddiphysmin;
2527 }
2528
2529 /*
2530 * mnode_xwa is greater than 1 when large pages regions can
2531 * cross memory node boundaries. To prevent the formation
2532 * of these large pages, configure the memsegs based on the
2533 * memory node ranges which had been made non-contiguous.
2534 */
2535 if (mnode_xwa > 1) {
2536
2537 end_pfn = base_pfn + num - 1;
2538 ms = PFN_2_MEM_NODE(base_pfn);
2539 me = PFN_2_MEM_NODE(end_pfn);
2540
2541 if (ms != me) {
2542 /*
2543 * current range spans more than 1 memory node.
2544 * Set num to only the pfn range in the start
2545 * memory node.
2546 */
2547 num = mem_node_config[ms].physmax - base_pfn
2548 + 1;
2549 ASSERT(end_pfn > mem_node_config[ms].physmax);
2550 }
2551 }
2552
2553 for (;;) {
2554 /*
2555 * Build the memsegs entry
2556 */
2557 cur_memseg->pages = pp;
2558 cur_memseg->epages = pp + num;
2559 cur_memseg->pages_base = base_pfn;
2560 cur_memseg->pages_end = base_pfn + num;
2561
2562 /*
2563 * Insert into memseg list in decreasing pfn range
2564 * order. Low memory is typically more fragmented such
2565 * that this ordering keeps the larger ranges at the
2566 * front of the list for code that searches memseg.
2567 * This ASSERTS that the memsegs coming in from boot
2568 * are in increasing physical address order and not
2569 * contiguous.
2570 */
2571 if (memsegs != NULL) {
2572 ASSERT(cur_memseg->pages_base >=
2573 memsegs->pages_end);
2574 cur_memseg->next = memsegs;
2575 }
2576 memsegs = cur_memseg;
2577
2578 /*
2579 * add_physmem() initializes the PSM part of the page
2580 * struct by calling the PSM back with add_physmem_cb().
2581 * In addition it coalesces pages into larger pages as
2582 * it initializes them.
2583 */
2584 add_physmem(pp, num, base_pfn);
2585 cur_memseg++;
2586 availrmem_initial += num;
2587 availrmem += num;
2588
2589 pp += num;
2590 if (ms >= me)
2591 break;
2592
2593 /* process next memory node range */
2594 ms++;
2595 base_pfn = mem_node_config[ms].physbase;
2596 num = MIN(mem_node_config[ms].physmax,
2597 end_pfn) - base_pfn + 1;
2598 }
2599 }
2600
2601 PRM_DEBUG(availrmem_initial);
2602 PRM_DEBUG(availrmem);
2603 PRM_DEBUG(freemem);
2604 build_pfn_hash();
2605 return (pages_done);
2606 }
2607
2608 /*
2609 * Kernel VM initialization.
2610 */
2611 static void
2612 kvm_init(void)
2613 {
2614 ASSERT((((uintptr_t)s_text) & MMU_PAGEOFFSET) == 0);
2615
2616 /*
2617 * Put the kernel segments in kernel address space.
2618 */
2619 rw_enter(&kas.a_lock, RW_WRITER);
2620 as_avlinit(&kas);
2621
2622 (void) seg_attach(&kas, s_text, e_moddata - s_text, &ktextseg);
2623 (void) segkmem_create(&ktextseg);
2624
2625 (void) seg_attach(&kas, (caddr_t)valloc_base, valloc_sz, &kvalloc);
2626 (void) segkmem_create(&kvalloc);
2627
2628 (void) seg_attach(&kas, kernelheap,
2629 ekernelheap - kernelheap, &kvseg);
2630 (void) segkmem_create(&kvseg);
2631
2632 if (core_size > 0) {
2633 PRM_POINT("attaching kvseg_core");
2634 (void) seg_attach(&kas, (caddr_t)core_base, core_size,
2635 &kvseg_core);
2636 (void) segkmem_create(&kvseg_core);
2637 }
2638
2639 if (segziosize > 0) {
2640 PRM_POINT("attaching segzio");
2641 (void) seg_attach(&kas, segzio_base, mmu_ptob(segziosize),
2642 &kzioseg);
2643 (void) segkmem_zio_create(&kzioseg);
2644
2645 /* create zio area covering new segment */
2646 segkmem_zio_init(segzio_base, mmu_ptob(segziosize));
2647 }
2648
2649 (void) seg_attach(&kas, kdi_segdebugbase, kdi_segdebugsize, &kdebugseg);
2650 (void) segkmem_create(&kdebugseg);
2651
2652 rw_exit(&kas.a_lock);
2653
2654 /*
2655 * Ensure that the red zone at kernelbase is never accessible.
2656 */
2657 PRM_POINT("protecting redzone");
2658 (void) as_setprot(&kas, (caddr_t)kernelbase, KERNEL_REDZONE_SIZE, 0);
2659
2660 /*
2661 * Make the text writable so that it can be hot patched by DTrace.
2662 */
2663 (void) as_setprot(&kas, s_text, e_modtext - s_text,
2664 PROT_READ | PROT_WRITE | PROT_EXEC);
2665
2666 /*
2667 * Make data writable until end.
2668 */
2669 (void) as_setprot(&kas, s_data, e_moddata - s_data,
2670 PROT_READ | PROT_WRITE | PROT_EXEC);
2671 }
2672
2673 #ifndef __xpv
2674 /*
2675 * Solaris adds an entry for Write Combining caching to the PAT
2676 */
2677 static uint64_t pat_attr_reg = PAT_DEFAULT_ATTRIBUTE;
2678
2679 void
2680 pat_sync(void)
2681 {
2682 ulong_t cr0, cr0_orig, cr4;
2683
2684 if (!is_x86_feature(x86_featureset, X86FSET_PAT))
2685 return;
2686 cr0_orig = cr0 = getcr0();
2687 cr4 = getcr4();
2688
2689 /* disable caching and flush all caches and TLBs */
2690 cr0 |= CR0_CD;
2691 cr0 &= ~CR0_NW;
2692 setcr0(cr0);
2693 invalidate_cache();
2694 if (cr4 & CR4_PGE) {
2695 setcr4(cr4 & ~(ulong_t)CR4_PGE);
2696 setcr4(cr4);
2697 } else {
2698 reload_cr3();
2699 }
2700
2701 /* add our entry to the PAT */
2702 wrmsr(REG_PAT, pat_attr_reg);
2703
2704 /* flush TLBs and cache again, then reenable cr0 caching */
2705 if (cr4 & CR4_PGE) {
2706 setcr4(cr4 & ~(ulong_t)CR4_PGE);
2707 setcr4(cr4);
2708 } else {
2709 reload_cr3();
2710 }
2711 invalidate_cache();
2712 setcr0(cr0_orig);
2713 }
2714
2715 #endif /* !__xpv */
2716
2717 #if defined(_SOFT_HOSTID)
2718 /*
2719 * On platforms that do not have a hardware serial number, attempt
2720 * to set one based on the contents of /etc/hostid. If this file does
2721 * not exist, assume that we are to generate a new hostid and set
2722 * it in the kernel, for subsequent saving by a userland process
2723 * once the system is up and the root filesystem is mounted r/w.
2724 *
2725 * In order to gracefully support upgrade on OpenSolaris, if
2726 * /etc/hostid does not exist, we will attempt to get a serial number
2727 * using the legacy method (/kernel/misc/sysinit).
2728 *
2729 * If that isn't present, we attempt to use an SMBIOS UUID, which is
2730 * a hardware serial number. Note that we don't automatically trust
2731 * all SMBIOS UUIDs (some older platforms are defective and ship duplicate
2732 * UUIDs in violation of the standard), we check against a blacklist.
2733 *
2734 * In an attempt to make the hostid less prone to abuse
2735 * (for license circumvention, etc), we store it in /etc/hostid
2736 * in rot47 format.
2737 */
2738 extern volatile unsigned long tenmicrodata;
2739 static int atoi(char *);
2740
2741 /*
2742 * Set this to non-zero in /etc/system if you think your SMBIOS returns a
2743 * UUID that is not unique. (Also report it so that the smbios_uuid_blacklist
2744 * array can be updated.)
2745 */
2746 int smbios_broken_uuid = 0;
2747
2748 /*
2749 * List of known bad UUIDs. This is just the lower 32-bit values, since
2750 * that's what we use for the host id. If your hostid falls here, you need
2751 * to contact your hardware OEM for a fix for your BIOS.
2752 */
2753 static unsigned char
2754 smbios_uuid_blacklist[][16] = {
2755
2756 { /* Reported bad UUID (Google search) */
2757 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05,
2758 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00, 0x09,
2759 },
2760 { /* Known bad DELL UUID */
2761 0x4C, 0x4C, 0x45, 0x44, 0x00, 0x00, 0x20, 0x10,
2762 0x80, 0x20, 0x80, 0xC0, 0x4F, 0x20, 0x20, 0x20,
2763 },
2764 { /* Uninitialized flash */
2765 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
2766 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
2767 },
2768 { /* All zeros */
2769 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
2770 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
2771 },
2772 };
2773
2774 static int32_t
2775 uuid_to_hostid(const uint8_t *uuid)
2776 {
2777 /*
2778 * Although the UUIDs are 128-bits, they may not distribute entropy
2779 * evenly. We would like to use SHA or MD5, but those are located
2780 * in loadable modules and not available this early in boot. As we
2781 * don't need the values to be cryptographically strong, we just
2782 * generate 32-bit vaue by xor'ing the various sequences together,
2783 * which ensures that the enire UUID contributes to the hostid.
2784 */
2785 int32_t id = 0;
2786
2787 /* first check against the blacklist */
2788 for (int i = 0; i < (sizeof (smbios_uuid_blacklist) / 16); i++) {
2789 if (bcmp(smbios_uuid_blacklist[0], uuid, 16) == 0) {
2790 cmn_err(CE_CONT, "?Broken SMBIOS UUID. "
2791 "Contact BIOS manufacturer for repair.\n");
2792 return ((int32_t)HW_INVALID_HOSTID);
2793 }
2794 }
2795
2796 for (int i = 0; i < 16; i++)
2797 id ^= ((uuid[i]) << (8 * (i % sizeof (id))));
2798
2799 return (id);
2800 }
2801
2802 static int32_t
2803 set_soft_hostid(void)
2804 {
2805 struct _buf *file;
2806 char tokbuf[MAXNAMELEN];
2807 token_t token;
2808 int done = 0;
2809 u_longlong_t tmp;
2810 int i;
2811 int32_t hostid = (int32_t)HW_INVALID_HOSTID;
2812 unsigned char *c;
2813 hrtime_t tsc;
2814 smbios_system_t smsys;
2815
2816 /*
2817 * If /etc/hostid file not found, we'd like to get a pseudo
2818 * random number to use at the hostid. A nice way to do this
2819 * is to read the real time clock. To remain xen-compatible,
2820 * we can't poke the real hardware, so we use tsc_read() to
2821 * read the real time clock. However, there is an ominous
2822 * warning in tsc_read that says it can return zero, so we
2823 * deal with that possibility by falling back to using the
2824 * (hopefully random enough) value in tenmicrodata.
2825 */
2826
2827 if ((file = kobj_open_file(hostid_file)) == (struct _buf *)-1) {
2828 /*
2829 * hostid file not found - try to load sysinit module
2830 * and see if it has a nonzero hostid value...use that
2831 * instead of generating a new hostid here if so.
2832 */
2833 if ((i = modload("misc", "sysinit")) != -1) {
2834 if (strlen(hw_serial) > 0)
2835 hostid = (int32_t)atoi(hw_serial);
2836 (void) modunload(i);
2837 }
2838
2839 /*
2840 * We try to use the SMBIOS UUID. But not if it is blacklisted
2841 * in /etc/system.
2842 */
2843 if ((hostid == HW_INVALID_HOSTID) &&
2844 (smbios_broken_uuid == 0) &&
2845 (ksmbios != NULL) &&
2846 (smbios_info_system(ksmbios, &smsys) != SMB_ERR) &&
2847 (smsys.smbs_uuidlen >= 16)) {
2848 hostid = uuid_to_hostid(smsys.smbs_uuid);
2849 }
2850
2851 /*
2852 * Generate a "random" hostid using the clock. These
2853 * hostids will change on each boot if the value is not
2854 * saved to a persistent /etc/hostid file.
2855 */
2856 if (hostid == HW_INVALID_HOSTID) {
2857 tsc = tsc_read();
2858 if (tsc == 0) /* tsc_read can return zero sometimes */
2859 hostid = (int32_t)tenmicrodata & 0x0CFFFFF;
2860 else
2861 hostid = (int32_t)tsc & 0x0CFFFFF;
2862 }
2863 } else {
2864 /* hostid file found */
2865 while (!done) {
2866 token = kobj_lex(file, tokbuf, sizeof (tokbuf));
2867
2868 switch (token) {
2869 case POUND:
2870 /*
2871 * skip comments
2872 */
2873 kobj_find_eol(file);
2874 break;
2875 case STRING:
2876 /*
2877 * un-rot47 - obviously this
2878 * nonsense is ascii-specific
2879 */
2880 for (c = (unsigned char *)tokbuf;
2881 *c != '\0'; c++) {
2882 *c += 47;
2883 if (*c > '~')
2884 *c -= 94;
2885 else if (*c < '!')
2886 *c += 94;
2887 }
2888 /*
2889 * now we should have a real number
2890 */
2891
2892 if (kobj_getvalue(tokbuf, &tmp) != 0)
2893 kobj_file_err(CE_WARN, file,
2894 "Bad value %s for hostid",
2895 tokbuf);
2896 else
2897 hostid = (int32_t)tmp;
2898
2899 break;
2900 case EOF:
2901 done = 1;
2902 /* FALLTHROUGH */
2903 case NEWLINE:
2904 kobj_newline(file);
2905 break;
2906 default:
2907 break;
2908
2909 }
2910 }
2911 if (hostid == HW_INVALID_HOSTID) /* didn't find a hostid */
2912 kobj_file_err(CE_WARN, file,
2913 "hostid missing or corrupt");
2914
2915 kobj_close_file(file);
2916 }
2917 /*
2918 * hostid is now the value read from /etc/hostid, or the
2919 * new hostid we generated in this routine or HW_INVALID_HOSTID if not
2920 * set.
2921 */
2922 return (hostid);
2923 }
2924
2925 static int
2926 atoi(char *p)
2927 {
2928 int i = 0;
2929
2930 while (*p != '\0')
2931 i = 10 * i + (*p++ - '0');
2932
2933 return (i);
2934 }
2935
2936 #endif /* _SOFT_HOSTID */
2937
2938 void
2939 get_system_configuration(void)
2940 {
2941 char prop[32];
2942 u_longlong_t nodes_ll, cpus_pernode_ll, lvalue;
2943
2944 if (BOP_GETPROPLEN(bootops, "nodes") > sizeof (prop) ||
2945 BOP_GETPROP(bootops, "nodes", prop) < 0 ||
2946 kobj_getvalue(prop, &nodes_ll) == -1 ||
2947 nodes_ll > MAXNODES ||
2948 BOP_GETPROPLEN(bootops, "cpus_pernode") > sizeof (prop) ||
2949 BOP_GETPROP(bootops, "cpus_pernode", prop) < 0 ||
2950 kobj_getvalue(prop, &cpus_pernode_ll) == -1) {
2951 system_hardware.hd_nodes = 1;
2952 system_hardware.hd_cpus_per_node = 0;
2953 } else {
2954 system_hardware.hd_nodes = (int)nodes_ll;
2955 system_hardware.hd_cpus_per_node = (int)cpus_pernode_ll;
2956 }
2957
2958 if (BOP_GETPROPLEN(bootops, "kernelbase") > sizeof (prop) ||
2959 BOP_GETPROP(bootops, "kernelbase", prop) < 0 ||
2960 kobj_getvalue(prop, &lvalue) == -1)
2961 eprom_kernelbase = NULL;
2962 else
2963 eprom_kernelbase = (uintptr_t)lvalue;
2964
2965 if (BOP_GETPROPLEN(bootops, "segmapsize") > sizeof (prop) ||
2966 BOP_GETPROP(bootops, "segmapsize", prop) < 0 ||
2967 kobj_getvalue(prop, &lvalue) == -1)
2968 segmapsize = SEGMAPDEFAULT;
2969 else
2970 segmapsize = (uintptr_t)lvalue;
2971
2972 if (BOP_GETPROPLEN(bootops, "segmapfreelists") > sizeof (prop) ||
2973 BOP_GETPROP(bootops, "segmapfreelists", prop) < 0 ||
2974 kobj_getvalue(prop, &lvalue) == -1)
2975 segmapfreelists = 0; /* use segmap driver default */
2976 else
2977 segmapfreelists = (int)lvalue;
2978
2979 /* physmem used to be here, but moved much earlier to fakebop.c */
2980 }
2981
2982 /*
2983 * Add to a memory list.
2984 * start = start of new memory segment
2985 * len = length of new memory segment in bytes
2986 * new = pointer to a new struct memlist
2987 * memlistp = memory list to which to add segment.
2988 */
2989 void
2990 memlist_add(
2991 uint64_t start,
2992 uint64_t len,
2993 struct memlist *new,
2994 struct memlist **memlistp)
2995 {
2996 struct memlist *cur;
2997 uint64_t end = start + len;
2998
2999 new->ml_address = start;
3000 new->ml_size = len;
3001
3002 cur = *memlistp;
3003
3004 while (cur) {
3005 if (cur->ml_address >= end) {
3006 new->ml_next = cur;
3007 *memlistp = new;
3008 new->ml_prev = cur->ml_prev;
3009 cur->ml_prev = new;
3010 return;
3011 }
3012 ASSERT(cur->ml_address + cur->ml_size <= start);
3013 if (cur->ml_next == NULL) {
3014 cur->ml_next = new;
3015 new->ml_prev = cur;
3016 new->ml_next = NULL;
3017 return;
3018 }
3019 memlistp = &cur->ml_next;
3020 cur = cur->ml_next;
3021 }
3022 }
3023
3024 void
3025 kobj_vmem_init(vmem_t **text_arena, vmem_t **data_arena)
3026 {
3027 size_t tsize = e_modtext - modtext;
3028 size_t dsize = e_moddata - moddata;
3029
3030 *text_arena = vmem_create("module_text", tsize ? modtext : NULL, tsize,
3031 1, segkmem_alloc, segkmem_free, heaptext_arena, 0, VM_SLEEP);
3032 *data_arena = vmem_create("module_data", dsize ? moddata : NULL, dsize,
3033 1, segkmem_alloc, segkmem_free, heap32_arena, 0, VM_SLEEP);
3034 }
3035
3036 caddr_t
3037 kobj_text_alloc(vmem_t *arena, size_t size)
3038 {
3039 return (vmem_alloc(arena, size, VM_SLEEP | VM_BESTFIT));
3040 }
3041
3042 /*ARGSUSED*/
3043 caddr_t
3044 kobj_texthole_alloc(caddr_t addr, size_t size)
3045 {
3046 panic("unexpected call to kobj_texthole_alloc()");
3047 /*NOTREACHED*/
3048 return (0);
3049 }
3050
3051 /*ARGSUSED*/
3052 void
3053 kobj_texthole_free(caddr_t addr, size_t size)
3054 {
3055 panic("unexpected call to kobj_texthole_free()");
3056 }
3057
3058 /*
3059 * This is called just after configure() in startup().
3060 *
3061 * The ISALIST concept is a bit hopeless on Intel, because
3062 * there's no guarantee of an ever-more-capable processor
3063 * given that various parts of the instruction set may appear
3064 * and disappear between different implementations.
3065 *
3066 * While it would be possible to correct it and even enhance
3067 * it somewhat, the explicit hardware capability bitmask allows
3068 * more flexibility.
3069 *
3070 * So, we just leave this alone.
3071 */
3072 void
3073 setx86isalist(void)
3074 {
3075 char *tp;
3076 size_t len;
3077 extern char *isa_list;
3078
3079 #define TBUFSIZE 1024
3080
3081 tp = kmem_alloc(TBUFSIZE, KM_SLEEP);
3082 *tp = '\0';
3083
3084 #if defined(__amd64)
3085 (void) strcpy(tp, "amd64 ");
3086 #endif
3087
3088 switch (x86_vendor) {
3089 case X86_VENDOR_Intel:
3090 case X86_VENDOR_AMD:
3091 case X86_VENDOR_TM:
3092 if (is_x86_feature(x86_featureset, X86FSET_CMOV)) {
3093 /*
3094 * Pentium Pro or later
3095 */
3096 (void) strcat(tp, "pentium_pro");
3097 (void) strcat(tp,
3098 is_x86_feature(x86_featureset, X86FSET_MMX) ?
3099 "+mmx pentium_pro " : " ");
3100 }
3101 /*FALLTHROUGH*/
3102 case X86_VENDOR_Cyrix:
3103 /*
3104 * The Cyrix 6x86 does not have any Pentium features
3105 * accessible while not at privilege level 0.
3106 */
3107 if (is_x86_feature(x86_featureset, X86FSET_CPUID)) {
3108 (void) strcat(tp, "pentium");
3109 (void) strcat(tp,
3110 is_x86_feature(x86_featureset, X86FSET_MMX) ?
3111 "+mmx pentium " : " ");
3112 }
3113 break;
3114 default:
3115 break;
3116 }
3117 (void) strcat(tp, "i486 i386 i86");
3118 len = strlen(tp) + 1; /* account for NULL at end of string */
3119 isa_list = strcpy(kmem_alloc(len, KM_SLEEP), tp);
3120 kmem_free(tp, TBUFSIZE);
3121
3122 #undef TBUFSIZE
3123 }
3124
3125
3126 #ifdef __amd64
3127
3128 void *
3129 device_arena_alloc(size_t size, int vm_flag)
3130 {
3131 return (vmem_alloc(device_arena, size, vm_flag));
3132 }
3133
3134 void
3135 device_arena_free(void *vaddr, size_t size)
3136 {
3137 vmem_free(device_arena, vaddr, size);
3138 }
3139
3140 #else /* __i386 */
3141
3142 void *
3143 device_arena_alloc(size_t size, int vm_flag)
3144 {
3145 caddr_t vaddr;
3146 uintptr_t v;
3147 size_t start;
3148 size_t end;
3149
3150 vaddr = vmem_alloc(heap_arena, size, vm_flag);
3151 if (vaddr == NULL)
3152 return (NULL);
3153
3154 v = (uintptr_t)vaddr;
3155 ASSERT(v >= kernelbase);
3156 ASSERT(v + size <= valloc_base);
3157
3158 start = btop(v - kernelbase);
3159 end = btop(v + size - 1 - kernelbase);
3160 ASSERT(start < toxic_bit_map_len);
3161 ASSERT(end < toxic_bit_map_len);
3162
3163 while (start <= end) {
3164 BT_ATOMIC_SET(toxic_bit_map, start);
3165 ++start;
3166 }
3167 return (vaddr);
3168 }
3169
3170 void
3171 device_arena_free(void *vaddr, size_t size)
3172 {
3173 uintptr_t v = (uintptr_t)vaddr;
3174 size_t start;
3175 size_t end;
3176
3177 ASSERT(v >= kernelbase);
3178 ASSERT(v + size <= valloc_base);
3179
3180 start = btop(v - kernelbase);
3181 end = btop(v + size - 1 - kernelbase);
3182 ASSERT(start < toxic_bit_map_len);
3183 ASSERT(end < toxic_bit_map_len);
3184
3185 while (start <= end) {
3186 ASSERT(BT_TEST(toxic_bit_map, start) != 0);
3187 BT_ATOMIC_CLEAR(toxic_bit_map, start);
3188 ++start;
3189 }
3190 vmem_free(heap_arena, vaddr, size);
3191 }
3192
3193 /*
3194 * returns 1st address in range that is in device arena, or NULL
3195 * if len is not NULL it returns the length of the toxic range
3196 */
3197 void *
3198 device_arena_contains(void *vaddr, size_t size, size_t *len)
3199 {
3200 uintptr_t v = (uintptr_t)vaddr;
3201 uintptr_t eaddr = v + size;
3202 size_t start;
3203 size_t end;
3204
3205 /*
3206 * if called very early by kmdb, just return NULL
3207 */
3208 if (toxic_bit_map == NULL)
3209 return (NULL);
3210
3211 /*
3212 * First check if we're completely outside the bitmap range.
3213 */
3214 if (v >= valloc_base || eaddr < kernelbase)
3215 return (NULL);
3216
3217 /*
3218 * Trim ends of search to look at only what the bitmap covers.
3219 */
3220 if (v < kernelbase)
3221 v = kernelbase;
3222 start = btop(v - kernelbase);
3223 end = btop(eaddr - kernelbase);
3224 if (end >= toxic_bit_map_len)
3225 end = toxic_bit_map_len;
3226
3227 if (bt_range(toxic_bit_map, &start, &end, end) == 0)
3228 return (NULL);
3229
3230 v = kernelbase + ptob(start);
3231 if (len != NULL)
3232 *len = ptob(end - start);
3233 return ((void *)v);
3234 }
3235
3236 #endif /* __i386 */