fastboot.c revision 10559:6030a9284200
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27/*
28 * This file contains the functions for performing Fast Reboot -- a
29 * reboot which bypasses the firmware and bootloader, considerably
30 * reducing downtime.
31 *
32 * fastboot_load_kernel(): This function is invoked by mdpreboot() in the
33 * reboot path.  It loads the new kernel and boot archive into memory, builds
34 * the data structure containing sufficient information about the new
35 * kernel and boot archive to be passed to the fast reboot switcher
36 * (see fb_swtch_src.s for details).  When invoked the switcher relocates
37 * the new kernel and boot archive to physically contiguous low memory,
38 * similar to where the boot loader would have loaded them, and jumps to
39 * the new kernel.
40 *
41 * If fastreboot_onpanic is enabled, fastboot_load_kernel() is called
42 * by fastreboot_post_startup() to load the back up kernel in case of
43 * panic.
44 *
45 * The physical addresses of the memory allocated for the new kernel, boot
46 * archive and their page tables must be above where the boot archive ends
47 * after it has been relocated by the switcher, otherwise the new files
48 * and their page tables could be overridden during relocation.
49 *
50 * fast_reboot(): This function is invoked by mdboot() once it's determined
51 * that the system is capable of fast reboot.  It jumps to the fast reboot
52 * switcher with the data structure built by fastboot_load_kernel() as the
53 * argument.
54 */
55
56#include <sys/types.h>
57#include <sys/param.h>
58#include <sys/segments.h>
59#include <sys/sysmacros.h>
60#include <sys/vm.h>
61
62#include <sys/proc.h>
63#include <sys/buf.h>
64#include <sys/kmem.h>
65
66#include <sys/reboot.h>
67#include <sys/uadmin.h>
68
69#include <sys/cred.h>
70#include <sys/vnode.h>
71#include <sys/file.h>
72
73#include <sys/cmn_err.h>
74#include <sys/dumphdr.h>
75#include <sys/bootconf.h>
76#include <sys/ddidmareq.h>
77#include <sys/varargs.h>
78#include <sys/promif.h>
79#include <sys/modctl.h>
80
81#include <vm/hat.h>
82#include <vm/as.h>
83#include <vm/page.h>
84#include <vm/seg.h>
85#include <vm/hat_i86.h>
86#include <sys/vm_machparam.h>
87#include <sys/archsystm.h>
88#include <sys/machsystm.h>
89#include <sys/mman.h>
90#include <sys/x86_archext.h>
91#include <sys/smp_impldefs.h>
92#include <sys/spl.h>
93
94#include <sys/fastboot.h>
95#include <sys/machelf.h>
96#include <sys/kobj.h>
97#include <sys/multiboot.h>
98#include <sys/kobj_lex.h>
99
100/*
101 * Macro to determine how many pages are needed for PTEs to map a particular
102 * file.  Allocate one extra page table entry for terminating the list.
103 */
104#define	FASTBOOT_PTE_LIST_SIZE(fsize)	\
105	P2ROUNDUP((((fsize) >> PAGESHIFT) + 1) * sizeof (x86pte_t), PAGESIZE)
106
107/*
108 * Data structure containing necessary information for the fast reboot
109 * switcher to jump to the new kernel.
110 */
111fastboot_info_t newkernel = { 0 };
112char		fastboot_args[OBP_MAXPATHLEN];
113
114static char fastboot_filename[2][OBP_MAXPATHLEN] = { { 0 }, { 0 }};
115static x86pte_t ptp_bits = PT_VALID | PT_REF | PT_USER | PT_WRITABLE;
116static x86pte_t pte_bits =
117    PT_VALID | PT_REF | PT_MOD | PT_NOCONSIST | PT_WRITABLE;
118static uint_t fastboot_shift_amt_pae[] = {12, 21, 30, 39};
119
120int fastboot_debug = 0;
121int fastboot_contig = 0;
122
123/*
124 * Fake starting va for new kernel and boot archive.
125 */
126static uintptr_t fake_va = FASTBOOT_FAKE_VA;
127
128/*
129 * Reserve memory below PA 1G in preparation of fast reboot.
130 *
131 * This variable is only checked when fastreboot_capable is set, but
132 * fastreboot_onpanic is not set.  The amount of memory reserved
133 * is negligible, but just in case we are really short of low memory,
134 * this variable will give us a backdoor to not consume memory at all.
135 */
136int reserve_mem_enabled = 1;
137
138/*
139 * Mutex to protect fastreboot_onpanic.
140 */
141kmutex_t fastreboot_config_mutex;
142
143/*
144 * Amount of memory below PA 1G to reserve for constructing the multiboot
145 * data structure and the page tables as we tend to run out of those
146 * when more drivers are loaded.
147 */
148static size_t fastboot_mbi_size = 0x2000;	/* 8K */
149static size_t fastboot_pagetable_size = 0x5000;	/* 20K */
150
151/*
152 * Use below 1G for page tables as
153 *	1. we are only doing 1:1 mapping of the bottom 1G of physical memory.
154 *	2. we are using 2G as the fake virtual address for the new kernel and
155 *	boot archive.
156 */
157static ddi_dma_attr_t fastboot_below_1G_dma_attr = {
158	DMA_ATTR_V0,
159	0x0000000008000000ULL,	/* dma_attr_addr_lo: 128MB */
160	0x000000003FFFFFFFULL,	/* dma_attr_addr_hi: 1G */
161	0x00000000FFFFFFFFULL,	/* dma_attr_count_max */
162	0x0000000000001000ULL,	/* dma_attr_align: 4KB */
163	1,			/* dma_attr_burstsize */
164	1,			/* dma_attr_minxfer */
165	0x00000000FFFFFFFFULL,	/* dma_attr_maxxfer */
166	0x00000000FFFFFFFFULL,	/* dma_attr_seg */
167	1,			/* dma_attr_sgllen */
168	0x1000ULL,		/* dma_attr_granular */
169	0,			/* dma_attr_flags */
170};
171
172static ddi_dma_attr_t fastboot_dma_attr = {
173	DMA_ATTR_V0,
174	0x0000000008000000ULL,	/* dma_attr_addr_lo: 128MB */
175#ifdef	__amd64
176	0xFFFFFFFFFFFFFFFFULL,	/* dma_attr_addr_hi: 2^64B */
177#else
178	0x0000000FFFFFFFFFULL,	/* dma_attr_addr_hi: 64GB */
179#endif	/* __amd64 */
180	0x00000000FFFFFFFFULL,	/* dma_attr_count_max */
181	0x0000000000001000ULL,	/* dma_attr_align: 4KB */
182	1,			/* dma_attr_burstsize */
183	1,			/* dma_attr_minxfer */
184	0x00000000FFFFFFFFULL,	/* dma_attr_maxxfer */
185	0x00000000FFFFFFFFULL,	/* dma_attr_seg */
186	1,			/* dma_attr_sgllen */
187	0x1000ULL,		/* dma_attr_granular */
188	0,			/* dma_attr_flags */
189};
190
191/*
192 * Various information saved from the previous boot to reconstruct
193 * multiboot_info.
194 */
195extern multiboot_info_t saved_mbi;
196extern mb_memory_map_t saved_mmap[FASTBOOT_SAVED_MMAP_COUNT];
197extern uint8_t saved_drives[FASTBOOT_SAVED_DRIVES_SIZE];
198extern char saved_cmdline[FASTBOOT_SAVED_CMDLINE_LEN];
199extern int saved_cmdline_len;
200extern size_t saved_file_size[];
201
202extern void* contig_alloc(size_t size, ddi_dma_attr_t *attr,
203    uintptr_t align, int cansleep);
204extern void contig_free(void *addr, size_t size);
205
206
207/* PRINTLIKE */
208extern void vprintf(const char *, va_list);
209
210
211/*
212 * Need to be able to get boot_archives from other places
213 */
214#define	BOOTARCHIVE64	"/platform/i86pc/amd64/boot_archive"
215#define	BOOTARCHIVE32	"/platform/i86pc/boot_archive"
216#define	BOOTARCHIVE32_FAILSAFE	"/boot/x86.miniroot-safe"
217#define	BOOTARCHIVE64_FAILSAFE	"/boot/amd64/x86.miniroot-safe"
218#define	FAILSAFE_BOOTFILE32	"/boot/platform/i86pc/kernel/unix"
219#define	FAILSAFE_BOOTFILE64	"/boot/platform/i86pc/kernel/amd64/unix"
220
221static uint_t fastboot_vatoindex(fastboot_info_t *, uintptr_t, int);
222static void fastboot_map_with_size(fastboot_info_t *, uintptr_t,
223    paddr_t, size_t, int);
224static void fastboot_build_pagetables(fastboot_info_t *);
225static int fastboot_build_mbi(char *, fastboot_info_t *);
226static void fastboot_free_file(fastboot_file_t *);
227
228static const char fastboot_enomem_msg[] = "Fastboot: Couldn't allocate 0x%"
229	PRIx64" bytes below %s to do fast reboot";
230
231static void
232dprintf(char *fmt, ...)
233{
234	va_list adx;
235
236	if (!fastboot_debug)
237		return;
238
239	va_start(adx, fmt);
240	vprintf(fmt, adx);
241	va_end(adx);
242}
243
244
245/*
246 * Return the index corresponding to a virt address at a given page table level.
247 */
248static uint_t
249fastboot_vatoindex(fastboot_info_t *nk, uintptr_t va, int level)
250{
251	return ((va >> nk->fi_shift_amt[level]) & (nk->fi_ptes_per_table - 1));
252}
253
254
255/*
256 * Add mapping from vstart to pstart for the specified size.
257 * vstart, pstart and size should all have been aligned at 2M boundaries.
258 */
259static void
260fastboot_map_with_size(fastboot_info_t *nk, uintptr_t vstart, paddr_t pstart,
261    size_t size, int level)
262{
263	x86pte_t	pteval, *table;
264	uintptr_t	vaddr;
265	paddr_t		paddr;
266	int		index, l;
267
268	table = (x86pte_t *)(nk->fi_pagetable_va);
269
270	for (l = nk->fi_top_level; l >= level; l--) {
271
272		index = fastboot_vatoindex(nk, vstart, l);
273
274		if (l == level) {
275			/*
276			 * Last level.  Program the page table entries.
277			 */
278			for (vaddr = vstart, paddr = pstart;
279			    vaddr < vstart + size;
280			    vaddr += (1ULL << nk->fi_shift_amt[l]),
281			    paddr += (1ULL << nk->fi_shift_amt[l])) {
282
283				uint_t index = fastboot_vatoindex(nk, vaddr, l);
284
285				if (l > 0)
286					pteval = paddr | pte_bits | PT_PAGESIZE;
287				else
288					pteval = paddr | pte_bits;
289
290				table[index] = pteval;
291			}
292		} else if (table[index] & PT_VALID) {
293
294			table = (x86pte_t *)
295			    ((uintptr_t)(((paddr_t)table[index] & MMU_PAGEMASK)
296			    - nk->fi_pagetable_pa) + nk->fi_pagetable_va);
297		} else {
298			/*
299			 * Intermediate levels.
300			 * Program with either valid bit or PTP bits.
301			 */
302			if (l == nk->fi_top_level) {
303#ifdef	__amd64
304				ASSERT(nk->fi_top_level == 3);
305				table[index] = nk->fi_next_table_pa | ptp_bits;
306#else
307				table[index] = nk->fi_next_table_pa | PT_VALID;
308#endif	/* __amd64 */
309			} else {
310				table[index] = nk->fi_next_table_pa | ptp_bits;
311			}
312			table = (x86pte_t *)(nk->fi_next_table_va);
313			nk->fi_next_table_va += MMU_PAGESIZE;
314			nk->fi_next_table_pa += MMU_PAGESIZE;
315		}
316	}
317}
318
319/*
320 * Build page tables for the lower 1G of physical memory using 2M
321 * pages, and prepare page tables for mapping new kernel and boot
322 * archive pages using 4K pages.
323 */
324static void
325fastboot_build_pagetables(fastboot_info_t *nk)
326{
327	/*
328	 * Map lower 1G physical memory.  Use large pages.
329	 */
330	fastboot_map_with_size(nk, 0, 0, ONE_GIG, 1);
331
332	/*
333	 * Map one 4K page to get the middle page tables set up.
334	 */
335	fake_va = P2ALIGN_TYPED(fake_va, nk->fi_lpagesize, uintptr_t);
336	fastboot_map_with_size(nk, fake_va,
337	    nk->fi_files[0].fb_pte_list_va[0] & MMU_PAGEMASK, PAGESIZE, 0);
338}
339
340
341/*
342 * Sanity check.  Look for dboot offset.
343 */
344static int
345fastboot_elf64_find_dboot_load_offset(void *img, off_t imgsz, uint32_t *offp)
346{
347	Elf64_Ehdr	*ehdr = (Elf64_Ehdr *)img;
348	Elf64_Phdr	*phdr;
349	uint8_t		*phdrbase;
350	int		i;
351
352	if ((ehdr->e_phoff + ehdr->e_phnum * ehdr->e_phentsize) >= imgsz)
353		return (-1);
354
355	phdrbase = (uint8_t *)img + ehdr->e_phoff;
356
357	for (i = 0; i < ehdr->e_phnum; i++) {
358		phdr = (Elf64_Phdr *)(phdrbase + ehdr->e_phentsize * i);
359
360		if (phdr->p_type == PT_LOAD) {
361			if (phdr->p_vaddr == phdr->p_paddr &&
362			    phdr->p_vaddr == DBOOT_ENTRY_ADDRESS) {
363				ASSERT(phdr->p_offset <= UINT32_MAX);
364				*offp = (uint32_t)phdr->p_offset;
365				return (0);
366			}
367		}
368	}
369
370	return (-1);
371}
372
373
374/*
375 * Initialize text and data section information for 32-bit kernel.
376 * sectcntp - is both input/output parameter.
377 * On entry, *sectcntp contains maximum allowable number of sections;
378 * on return, it contains the actual number of sections filled.
379 */
380static int
381fastboot_elf32_find_loadables(void *img, off_t imgsz, fastboot_section_t *sectp,
382    int *sectcntp, uint32_t *offp)
383{
384	Elf32_Ehdr	*ehdr = (Elf32_Ehdr *)img;
385	Elf32_Phdr	*phdr;
386	uint8_t		*phdrbase;
387	int		i;
388	int		used_sections = 0;
389	const int	max_sectcnt = *sectcntp;
390
391	if ((ehdr->e_phoff + ehdr->e_phnum * ehdr->e_phentsize) >= imgsz)
392		return (-1);
393
394	phdrbase = (uint8_t *)img + ehdr->e_phoff;
395
396	for (i = 0; i < ehdr->e_phnum; i++) {
397		phdr = (Elf32_Phdr *)(phdrbase + ehdr->e_phentsize * i);
398
399		if (phdr->p_type == PT_INTERP)
400			return (-1);
401
402		if (phdr->p_type != PT_LOAD)
403			continue;
404
405		if (phdr->p_vaddr == phdr->p_paddr &&
406		    phdr->p_paddr == DBOOT_ENTRY_ADDRESS) {
407			*offp = (uint32_t)phdr->p_offset;
408		} else {
409			if (max_sectcnt <= used_sections)
410				return (-1);
411
412			sectp[used_sections].fb_sec_offset = phdr->p_offset;
413			sectp[used_sections].fb_sec_paddr = phdr->p_paddr;
414			sectp[used_sections].fb_sec_size = phdr->p_filesz;
415			sectp[used_sections].fb_sec_bss_size =
416			    (phdr->p_filesz < phdr->p_memsz) ?
417			    (phdr->p_memsz - phdr->p_filesz) : 0;
418
419			/* Extra sanity check for the input object file */
420			if (sectp[used_sections].fb_sec_paddr +
421			    sectp[used_sections].fb_sec_size +
422			    sectp[used_sections].fb_sec_bss_size >=
423			    DBOOT_ENTRY_ADDRESS)
424				return (-1);
425
426			used_sections++;
427		}
428	}
429
430	*sectcntp = used_sections;
431	return (0);
432}
433
434/*
435 * Create multiboot info structure (mbi) base on the saved mbi.
436 * Recalculate values of the pointer type fields in the data
437 * structure based on the new starting physical address of the
438 * data structure.
439 */
440static int
441fastboot_build_mbi(char *mdep, fastboot_info_t *nk)
442{
443	mb_module_t	*mbp;
444	multiboot_info_t	*mbi;	/* pointer to multiboot structure */
445	uintptr_t	start_addr_va;	/* starting VA of mbi */
446	uintptr_t	start_addr_pa;	/* starting PA of mbi */
447	size_t		offs = 0;	/* offset from the starting address */
448	size_t		arglen;		/* length of the command line arg */
449	size_t		size;	/* size of the memory reserved for mbi */
450	size_t		mdnsz;	/* length of the boot archive name */
451
452	/*
453	 * If mdep is not NULL or empty, use the length of mdep + 1
454	 * (for NULL terminating) as the length of the new command
455	 * line; else use the saved command line length as the
456	 * length for the new command line.
457	 */
458	if (mdep != NULL && strlen(mdep) != 0) {
459		arglen = strlen(mdep) + 1;
460	} else {
461		arglen = saved_cmdline_len;
462	}
463
464	/*
465	 * Allocate memory for the new multiboot info structure (mbi).
466	 * If we have reserved memory for mbi but it's not enough,
467	 * free it and reallocate.
468	 */
469	size = PAGESIZE + P2ROUNDUP(arglen, PAGESIZE);
470	if (nk->fi_mbi_size && nk->fi_mbi_size < size) {
471		contig_free((void *)nk->fi_new_mbi_va, nk->fi_mbi_size);
472		nk->fi_mbi_size = 0;
473	}
474
475	if (nk->fi_mbi_size == 0) {
476		if ((nk->fi_new_mbi_va =
477		    (uintptr_t)contig_alloc(size, &fastboot_below_1G_dma_attr,
478		    PAGESIZE, 0)) == NULL) {
479			cmn_err(CE_WARN, fastboot_enomem_msg,
480			    (uint64_t)size, "1G");
481			return (-1);
482		}
483		/*
484		 * fi_mbi_size must be set after the allocation succeeds
485		 * as it's used to determine how much memory to free.
486		 */
487		nk->fi_mbi_size = size;
488	}
489
490	/*
491	 * Initalize memory
492	 */
493	bzero((void *)nk->fi_new_mbi_va, nk->fi_mbi_size);
494
495	/*
496	 * Get PA for the new mbi
497	 */
498	start_addr_va = nk->fi_new_mbi_va;
499	start_addr_pa = mmu_ptob((uint64_t)hat_getpfnum(kas.a_hat,
500	    (caddr_t)start_addr_va));
501	nk->fi_new_mbi_pa = (paddr_t)start_addr_pa;
502
503	/*
504	 * Populate the rest of the fields in the data structure
505	 */
506
507	/*
508	 * Copy from the saved mbi to preserve all non-pointer type fields.
509	 */
510	mbi = (multiboot_info_t *)start_addr_va;
511	bcopy(&saved_mbi, mbi, sizeof (*mbi));
512
513	/*
514	 * Recalculate mods_addr.  Set mod_start and mod_end based on
515	 * the physical address of the new boot archive.  Set mod_name
516	 * to the name of the new boto archive.
517	 */
518	offs += sizeof (multiboot_info_t);
519	mbi->mods_addr = start_addr_pa + offs;
520	mbp = (mb_module_t *)(start_addr_va + offs);
521	mbp->mod_start = nk->fi_files[FASTBOOT_BOOTARCHIVE].fb_dest_pa;
522	mbp->mod_end = nk->fi_files[FASTBOOT_BOOTARCHIVE].fb_next_pa;
523
524	offs += sizeof (mb_module_t);
525	mdnsz = strlen(fastboot_filename[FASTBOOT_NAME_BOOTARCHIVE]) + 1;
526	bcopy(fastboot_filename[FASTBOOT_NAME_BOOTARCHIVE],
527	    (void *)(start_addr_va + offs), mdnsz);
528	mbp->mod_name = start_addr_pa + offs;
529	mbp->reserved = 0;
530
531	/*
532	 * Make sure the offset is 16-byte aligned to avoid unaligned access.
533	 */
534	offs += mdnsz;
535	offs = P2ROUNDUP_TYPED(offs, 16, size_t);
536
537	/*
538	 * Recalculate mmap_addr
539	 */
540	mbi->mmap_addr = start_addr_pa + offs;
541	bcopy((void *)(uintptr_t)saved_mmap, (void *)(start_addr_va + offs),
542	    saved_mbi.mmap_length);
543	offs += saved_mbi.mmap_length;
544
545	/*
546	 * Recalculate drives_addr
547	 */
548	mbi->drives_addr = start_addr_pa + offs;
549	bcopy((void *)(uintptr_t)saved_drives, (void *)(start_addr_va + offs),
550	    saved_mbi.drives_length);
551	offs += saved_mbi.drives_length;
552
553	/*
554	 * Recalculate the address of cmdline.  Set cmdline to contain the
555	 * new boot argument.
556	 */
557	mbi->cmdline = start_addr_pa + offs;
558
559	if (mdep != NULL && strlen(mdep) != 0) {
560		bcopy(mdep, (void *)(start_addr_va + offs), arglen);
561	} else {
562		bcopy((void *)saved_cmdline, (void *)(start_addr_va + offs),
563		    arglen);
564	}
565
566	/* clear fields and flags that are not copied */
567	bzero(&mbi->config_table,
568	    sizeof (*mbi) - offsetof(multiboot_info_t, config_table));
569	mbi->flags &= ~(MB_INFO_CONFIG_TABLE | MB_INFO_BOOT_LOADER_NAME |
570	    MB_INFO_APM_TABLE | MB_INFO_VIDEO_INFO);
571
572	return (0);
573}
574
575/*
576 * Initialize HAT related fields
577 */
578static void
579fastboot_init_fields(fastboot_info_t *nk)
580{
581	if (x86_feature & X86_PAE) {
582		nk->fi_has_pae = 1;
583		nk->fi_shift_amt = fastboot_shift_amt_pae;
584		nk->fi_ptes_per_table = 512;
585		nk->fi_lpagesize = (2 << 20);	/* 2M */
586#ifdef	__amd64
587		nk->fi_top_level = 3;
588#else
589		nk->fi_top_level = 2;
590#endif	/* __amd64 */
591	}
592}
593
594/*
595 * Process boot argument
596 */
597static void
598fastboot_parse_mdep(char *mdep, char *kern_bootpath, int *bootpath_len,
599    char *bootargs)
600{
601	int	i;
602
603	/*
604	 * If mdep is not NULL, it comes in the format of
605	 *	mountpoint unix args
606	 */
607	if (mdep != NULL && strlen(mdep) != 0) {
608		if (mdep[0] != '-') {
609			/* First get the root argument */
610			i = 0;
611			while (mdep[i] != '\0' && mdep[i] != ' ') {
612				i++;
613			}
614
615			if (i < 4 || strncmp(&mdep[i-4], "unix", 4) != 0) {
616				/* mount point */
617				bcopy(mdep, kern_bootpath, i);
618				kern_bootpath[i] = '\0';
619				*bootpath_len = i;
620
621				/*
622				 * Get the next argument. It should be unix as
623				 * we have validated in in halt.c.
624				 */
625				if (strlen(mdep) > i) {
626					mdep += (i + 1);
627					i = 0;
628					while (mdep[i] != '\0' &&
629					    mdep[i] != ' ') {
630						i++;
631					}
632				}
633
634			}
635			bcopy(mdep, kern_bootfile, i);
636			kern_bootfile[i] = '\0';
637			bcopy(mdep, bootargs, strlen(mdep));
638		} else {
639			int off = strlen(kern_bootfile);
640			bcopy(kern_bootfile, bootargs, off);
641			bcopy(" ", &bootargs[off++], 1);
642			bcopy(mdep, &bootargs[off], strlen(mdep));
643			off += strlen(mdep);
644			bootargs[off] = '\0';
645		}
646	}
647}
648
649/*
650 * Reserve memory under PA 1G for mapping the new kernel and boot archive.
651 * This function is only called if fastreboot_onpanic is *not* set.
652 */
653static void
654fastboot_reserve_mem(fastboot_info_t *nk)
655{
656	int i;
657
658	/*
659	 * A valid kernel is in place.  No need to reserve any memory.
660	 */
661	if (nk->fi_valid)
662		return;
663
664	/*
665	 * Reserve memory under PA 1G for PTE lists.
666	 */
667	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
668		fastboot_file_t *fb = &nk->fi_files[i];
669		size_t fsize_roundup, size;
670
671		fsize_roundup = P2ROUNDUP_TYPED(saved_file_size[i],
672		    PAGESIZE, size_t);
673		size = FASTBOOT_PTE_LIST_SIZE(fsize_roundup);
674		if ((fb->fb_pte_list_va = contig_alloc(size,
675		    &fastboot_below_1G_dma_attr, PAGESIZE, 0)) == NULL) {
676			return;
677		}
678		fb->fb_pte_list_size = size;
679	}
680
681	/*
682	 * Reserve memory under PA 1G for page tables.
683	 */
684	if ((nk->fi_pagetable_va =
685	    (uintptr_t)contig_alloc(fastboot_pagetable_size,
686	    &fastboot_below_1G_dma_attr, PAGESIZE, 0)) == NULL) {
687		return;
688	}
689	nk->fi_pagetable_size = fastboot_pagetable_size;
690
691	/*
692	 * Reserve memory under PA 1G for multiboot structure.
693	 */
694	if ((nk->fi_new_mbi_va = (uintptr_t)contig_alloc(fastboot_mbi_size,
695	    &fastboot_below_1G_dma_attr, PAGESIZE, 0)) == NULL) {
696		return;
697	}
698	nk->fi_mbi_size = fastboot_mbi_size;
699}
700
701/*
702 * Calculate MD5 digest for the given fastboot_file.
703 * Assumes that the file is allready loaded properly.
704 */
705static void
706fastboot_cksum_file(fastboot_file_t *fb, uchar_t *md5_hash)
707{
708	MD5_CTX md5_ctx;
709
710	MD5Init(&md5_ctx);
711	MD5Update(&md5_ctx, (void *)fb->fb_va, fb->fb_size);
712	MD5Final(md5_hash, &md5_ctx);
713}
714
715/*
716 * Free up the memory we have allocated for a file
717 */
718static void
719fastboot_free_file(fastboot_file_t *fb)
720{
721	size_t	fsize_roundup;
722
723	fsize_roundup = P2ROUNDUP_TYPED(fb->fb_size, PAGESIZE, size_t);
724	if (fsize_roundup) {
725		contig_free((void *)fb->fb_va, fsize_roundup);
726		fb->fb_va = NULL;
727		fb->fb_size = 0;
728	}
729}
730
731/*
732 * Free up memory used by the PTEs for a file.
733 */
734static void
735fastboot_free_file_pte(fastboot_file_t *fb, uint64_t endaddr)
736{
737	if (fb->fb_pte_list_size && fb->fb_pte_list_pa < endaddr) {
738		contig_free((void *)fb->fb_pte_list_va, fb->fb_pte_list_size);
739		fb->fb_pte_list_va = 0;
740		fb->fb_pte_list_pa = 0;
741		fb->fb_pte_list_size = 0;
742	}
743}
744
745/*
746 * Free up all the memory used for representing a kernel with
747 * fastboot_info_t.
748 */
749static void
750fastboot_free_mem(fastboot_info_t *nk, uint64_t endaddr)
751{
752	int i;
753
754	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
755		fastboot_free_file(nk->fi_files + i);
756		fastboot_free_file_pte(nk->fi_files + i, endaddr);
757	}
758
759	if (nk->fi_pagetable_size && nk->fi_pagetable_pa < endaddr) {
760		contig_free((void *)nk->fi_pagetable_va, nk->fi_pagetable_size);
761		nk->fi_pagetable_va = 0;
762		nk->fi_pagetable_pa = 0;
763		nk->fi_pagetable_size = 0;
764	}
765
766	if (nk->fi_mbi_size && nk->fi_new_mbi_pa < endaddr) {
767		contig_free((void *)nk->fi_new_mbi_va, nk->fi_mbi_size);
768		nk->fi_new_mbi_va = 0;
769		nk->fi_new_mbi_pa = 0;
770		nk->fi_mbi_size = 0;
771	}
772}
773
774/*
775 * Only free up the memory allocated for the kernel and boot archive,
776 * but not for the page tables.
777 */
778void
779fastboot_free_newkernel(fastboot_info_t *nk)
780{
781	int i;
782
783	nk->fi_valid = 0;
784	/*
785	 * Free the memory we have allocated
786	 */
787	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
788		fastboot_free_file(&(nk->fi_files[i]));
789	}
790}
791
792static void
793fastboot_cksum_cdata(fastboot_info_t *nk, uchar_t *md5_hash)
794{
795	int i;
796	MD5_CTX md5_ctx;
797
798	MD5Init(&md5_ctx);
799	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
800		MD5Update(&md5_ctx, nk->fi_files[i].fb_pte_list_va,
801		    nk->fi_files[i].fb_pte_list_size);
802	}
803	MD5Update(&md5_ctx, (void *)nk->fi_pagetable_va, nk->fi_pagetable_size);
804	MD5Update(&md5_ctx, (void *)nk->fi_new_mbi_va, nk->fi_mbi_size);
805
806	MD5Final(md5_hash, &md5_ctx);
807}
808
809/*
810 * Generate MD5 checksum of the given kernel.
811 */
812static void
813fastboot_cksum_generate(fastboot_info_t *nk)
814{
815	int i;
816
817	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
818		fastboot_cksum_file(nk->fi_files + i, nk->fi_md5_hash[i]);
819	}
820	fastboot_cksum_cdata(nk, nk->fi_md5_hash[i]);
821}
822
823/*
824 * Calculate MD5 checksum of the given kernel and verify that
825 * it matches with what was calculated before.
826 */
827int
828fastboot_cksum_verify(fastboot_info_t *nk)
829{
830	int i;
831	uchar_t md5_hash[MD5_DIGEST_LENGTH];
832
833	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
834		fastboot_cksum_file(nk->fi_files + i, md5_hash);
835		if (bcmp(nk->fi_md5_hash[i], md5_hash,
836		    sizeof (nk->fi_md5_hash[i])) != 0)
837			return (i + 1);
838	}
839
840	fastboot_cksum_cdata(nk, md5_hash);
841	if (bcmp(nk->fi_md5_hash[i], md5_hash,
842	    sizeof (nk->fi_md5_hash[i])) != 0)
843		return (i + 1);
844
845	return (0);
846}
847
848/*
849 * This function performs the following tasks:
850 * - Read the sizes of the new kernel and boot archive.
851 * - Allocate memory for the new kernel and boot archive.
852 * - Allocate memory for page tables necessary for mapping the memory
853 *   allocated for the files.
854 * - Read the new kernel and boot archive into memory.
855 * - Map in the fast reboot switcher.
856 * - Load the fast reboot switcher to FASTBOOT_SWTCH_PA.
857 * - Build the new multiboot_info structure
858 * - Build page tables for the low 1G of physical memory.
859 * - Mark the data structure as valid if all steps have succeeded.
860 */
861void
862fastboot_load_kernel(char *mdep)
863{
864	void		*buf = NULL;
865	int		i;
866	fastboot_file_t	*fb;
867	uint32_t	dboot_start_offset;
868	char		kern_bootpath[OBP_MAXPATHLEN];
869	extern uintptr_t postbootkernelbase;
870	uintptr_t	saved_kernelbase;
871	int		bootpath_len = 0;
872	int		is_failsafe = 0;
873	int		is_retry = 0;
874	uint64_t	end_addr;
875
876	if (!fastreboot_capable)
877		return;
878
879	if (newkernel.fi_valid)
880		fastboot_free_newkernel(&newkernel);
881
882	saved_kernelbase = postbootkernelbase;
883
884	postbootkernelbase = 0;
885
886	/*
887	 * Initialize various HAT related fields in the data structure
888	 */
889	fastboot_init_fields(&newkernel);
890
891	bzero(kern_bootpath, OBP_MAXPATHLEN);
892
893	/*
894	 * Process the boot argument
895	 */
896	bzero(fastboot_args, OBP_MAXPATHLEN);
897	fastboot_parse_mdep(mdep, kern_bootpath, &bootpath_len, fastboot_args);
898
899	/*
900	 * Make sure we get the null character
901	 */
902	bcopy(kern_bootpath, fastboot_filename[FASTBOOT_NAME_UNIX],
903	    bootpath_len);
904	bcopy(kern_bootfile,
905	    &fastboot_filename[FASTBOOT_NAME_UNIX][bootpath_len],
906	    strlen(kern_bootfile) + 1);
907
908	bcopy(kern_bootpath, fastboot_filename[FASTBOOT_NAME_BOOTARCHIVE],
909	    bootpath_len);
910
911	if (bcmp(kern_bootfile, FAILSAFE_BOOTFILE32,
912	    (sizeof (FAILSAFE_BOOTFILE32) - 1)) == 0 ||
913	    bcmp(kern_bootfile, FAILSAFE_BOOTFILE64,
914	    (sizeof (FAILSAFE_BOOTFILE64) - 1)) == 0) {
915		is_failsafe = 1;
916	}
917
918load_kernel_retry:
919	/*
920	 * Read in unix and boot_archive
921	 */
922	end_addr = DBOOT_ENTRY_ADDRESS;
923	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
924		struct _buf	*file;
925		uintptr_t	va;
926		uint64_t	fsize;
927		size_t		fsize_roundup, pt_size;
928		int		page_index;
929		uintptr_t	offset;
930		ddi_dma_attr_t dma_attr = fastboot_dma_attr;
931
932
933		dprintf("fastboot_filename[%d] = %s\n",
934		    i, fastboot_filename[i]);
935
936		if ((file = kobj_open_file(fastboot_filename[i])) ==
937		    (struct _buf *)-1) {
938			cmn_err(CE_WARN, "Fastboot: Couldn't open %s",
939			    fastboot_filename[i]);
940			goto err_out;
941		}
942
943		if (kobj_get_filesize(file, &fsize) != 0) {
944			cmn_err(CE_WARN,
945			    "Fastboot: Couldn't get filesize for %s",
946			    fastboot_filename[i]);
947			goto err_out;
948		}
949
950		fsize_roundup = P2ROUNDUP_TYPED(fsize, PAGESIZE, size_t);
951
952		/*
953		 * Where the files end in physical memory after being
954		 * relocated by the fast boot switcher.
955		 */
956		end_addr += fsize_roundup;
957		if (end_addr > fastboot_below_1G_dma_attr.dma_attr_addr_hi) {
958			cmn_err(CE_WARN, "Fastboot: boot archive is too big");
959			goto err_out;
960		}
961
962		/*
963		 * Adjust dma_attr_addr_lo so that the new kernel and boot
964		 * archive will not be overridden during relocation.
965		 */
966		if (end_addr > fastboot_dma_attr.dma_attr_addr_lo ||
967		    end_addr > fastboot_below_1G_dma_attr.dma_attr_addr_lo) {
968
969			if (is_retry) {
970				/*
971				 * If we have already tried and didn't succeed,
972				 * just give up.
973				 */
974				cmn_err(CE_WARN,
975				    "Fastboot: boot archive is too big");
976				goto err_out;
977			} else {
978				/* Set the flag so we don't keep retrying */
979				is_retry++;
980
981				/* Adjust dma_attr_addr_lo */
982				fastboot_dma_attr.dma_attr_addr_lo = end_addr;
983				fastboot_below_1G_dma_attr.dma_attr_addr_lo =
984				    end_addr;
985
986				/*
987				 * Free the memory we have already allocated
988				 * whose physical addresses might not fit
989				 * the new lo and hi constraints.
990				 */
991				fastboot_free_mem(&newkernel, end_addr);
992				goto load_kernel_retry;
993			}
994		}
995
996
997		if (!fastboot_contig)
998			dma_attr.dma_attr_sgllen = (fsize / PAGESIZE) +
999			    (((fsize % PAGESIZE) == 0) ? 0 : 1);
1000
1001		if ((buf = contig_alloc(fsize, &dma_attr, PAGESIZE, 0))
1002		    == NULL) {
1003			cmn_err(CE_WARN, fastboot_enomem_msg, fsize, "64G");
1004			goto err_out;
1005		}
1006
1007		va = P2ROUNDUP_TYPED((uintptr_t)buf, PAGESIZE, uintptr_t);
1008
1009		if (kobj_read_file(file, (char *)va, fsize, 0) < 0) {
1010			cmn_err(CE_WARN, "Fastboot: Couldn't read %s",
1011			    fastboot_filename[i]);
1012			goto err_out;
1013		}
1014
1015		fb = &newkernel.fi_files[i];
1016		fb->fb_va = va;
1017		fb->fb_size = fsize;
1018		fb->fb_sectcnt = 0;
1019
1020		pt_size = FASTBOOT_PTE_LIST_SIZE(fsize_roundup);
1021
1022		/*
1023		 * If we have reserved memory but it not enough, free it.
1024		 */
1025		if (fb->fb_pte_list_size && fb->fb_pte_list_size < pt_size) {
1026			contig_free((void *)fb->fb_pte_list_va,
1027			    fb->fb_pte_list_size);
1028			fb->fb_pte_list_size = 0;
1029		}
1030
1031		if (fb->fb_pte_list_size == 0) {
1032			if ((fb->fb_pte_list_va =
1033			    (x86pte_t *)contig_alloc(pt_size,
1034			    &fastboot_below_1G_dma_attr, PAGESIZE, 0))
1035			    == NULL) {
1036				cmn_err(CE_WARN, fastboot_enomem_msg,
1037				    (uint64_t)pt_size, "1G");
1038				goto err_out;
1039			}
1040			/*
1041			 * fb_pte_list_size must be set after the allocation
1042			 * succeeds as it's used to determine how much memory to
1043			 * free.
1044			 */
1045			fb->fb_pte_list_size = pt_size;
1046		}
1047
1048		bzero((void *)(fb->fb_pte_list_va), fb->fb_pte_list_size);
1049
1050		fb->fb_pte_list_pa = mmu_ptob((uint64_t)hat_getpfnum(kas.a_hat,
1051		    (caddr_t)fb->fb_pte_list_va));
1052
1053		for (page_index = 0, offset = 0; offset < fb->fb_size;
1054		    offset += PAGESIZE) {
1055			uint64_t paddr;
1056
1057			paddr = mmu_ptob((uint64_t)hat_getpfnum(kas.a_hat,
1058			    (caddr_t)fb->fb_va + offset));
1059
1060			ASSERT(paddr >= fastboot_dma_attr.dma_attr_addr_lo);
1061
1062			/*
1063			 * Include the pte_bits so we don't have to make
1064			 * it in assembly.
1065			 */
1066			fb->fb_pte_list_va[page_index++] = (x86pte_t)
1067			    (paddr | pte_bits);
1068		}
1069
1070		fb->fb_pte_list_va[page_index] = FASTBOOT_TERMINATE;
1071
1072		if (i == FASTBOOT_UNIX) {
1073			Ehdr	*ehdr = (Ehdr *)va;
1074			int	j;
1075
1076			/*
1077			 * Sanity checks:
1078			 */
1079			for (j = 0; j < SELFMAG; j++) {
1080				if (ehdr->e_ident[j] != ELFMAG[j]) {
1081					cmn_err(CE_WARN, "Fastboot: Bad ELF "
1082					    "signature");
1083					goto err_out;
1084				}
1085			}
1086
1087			if (ehdr->e_ident[EI_CLASS] == ELFCLASS32 &&
1088			    ehdr->e_ident[EI_DATA] == ELFDATA2LSB &&
1089			    ehdr->e_machine == EM_386) {
1090
1091				fb->fb_sectcnt = sizeof (fb->fb_sections) /
1092				    sizeof (fb->fb_sections[0]);
1093
1094				if (fastboot_elf32_find_loadables((void *)va,
1095				    fsize, &fb->fb_sections[0],
1096				    &fb->fb_sectcnt, &dboot_start_offset) < 0) {
1097					cmn_err(CE_WARN, "Fastboot: ELF32 "
1098					    "program section failure");
1099					goto err_out;
1100				}
1101
1102				if (fb->fb_sectcnt == 0) {
1103					cmn_err(CE_WARN, "Fastboot: No ELF32 "
1104					    "program sections found");
1105					goto err_out;
1106				}
1107
1108				if (is_failsafe) {
1109					/* Failsafe boot_archive */
1110					bcopy(BOOTARCHIVE32_FAILSAFE,
1111					    &fastboot_filename
1112					    [FASTBOOT_NAME_BOOTARCHIVE]
1113					    [bootpath_len],
1114					    sizeof (BOOTARCHIVE32_FAILSAFE));
1115				} else {
1116					bcopy(BOOTARCHIVE32,
1117					    &fastboot_filename
1118					    [FASTBOOT_NAME_BOOTARCHIVE]
1119					    [bootpath_len],
1120					    sizeof (BOOTARCHIVE32));
1121				}
1122
1123			} else if (ehdr->e_ident[EI_CLASS] == ELFCLASS64 &&
1124			    ehdr->e_ident[EI_DATA] == ELFDATA2LSB &&
1125			    ehdr->e_machine == EM_AMD64) {
1126
1127				if (fastboot_elf64_find_dboot_load_offset(
1128				    (void *)va, fsize, &dboot_start_offset)
1129				    != 0) {
1130					cmn_err(CE_WARN, "Fastboot: Couldn't "
1131					    "find ELF64 dboot entry offset");
1132					goto err_out;
1133				}
1134
1135				if ((x86_feature & X86_64) == 0 ||
1136				    (x86_feature & X86_PAE) == 0) {
1137					cmn_err(CE_WARN, "Fastboot: Cannot "
1138					    "reboot to %s: "
1139					    "not a 64-bit capable system",
1140					    kern_bootfile);
1141					goto err_out;
1142				}
1143
1144				if (is_failsafe) {
1145					/* Failsafe boot_archive */
1146					bcopy(BOOTARCHIVE64_FAILSAFE,
1147					    &fastboot_filename
1148					    [FASTBOOT_NAME_BOOTARCHIVE]
1149					    [bootpath_len],
1150					    sizeof (BOOTARCHIVE64_FAILSAFE));
1151				} else {
1152					bcopy(BOOTARCHIVE64,
1153					    &fastboot_filename
1154					    [FASTBOOT_NAME_BOOTARCHIVE]
1155					    [bootpath_len],
1156					    sizeof (BOOTARCHIVE64));
1157				}
1158			} else {
1159				cmn_err(CE_WARN, "Fastboot: Unknown ELF type");
1160				goto err_out;
1161			}
1162
1163			fb->fb_dest_pa = DBOOT_ENTRY_ADDRESS -
1164			    dboot_start_offset;
1165
1166			fb->fb_next_pa = DBOOT_ENTRY_ADDRESS + fsize_roundup;
1167		} else {
1168			fb->fb_dest_pa = newkernel.fi_files[i - 1].fb_next_pa;
1169			fb->fb_next_pa = fb->fb_dest_pa + fsize_roundup;
1170		}
1171
1172		kobj_close_file(file);
1173
1174	}
1175
1176	/*
1177	 * Add the function that will switch us to 32-bit protected mode
1178	 */
1179	fb = &newkernel.fi_files[FASTBOOT_SWTCH];
1180	fb->fb_va = fb->fb_dest_pa = FASTBOOT_SWTCH_PA;
1181	fb->fb_size = MMU_PAGESIZE;
1182
1183	hat_devload(kas.a_hat, (caddr_t)fb->fb_va,
1184	    MMU_PAGESIZE, mmu_btop(fb->fb_dest_pa),
1185	    PROT_READ | PROT_WRITE | PROT_EXEC,
1186	    HAT_LOAD_NOCONSIST | HAT_LOAD_LOCK);
1187
1188	/*
1189	 * Build the new multiboot_info structure
1190	 */
1191	if (fastboot_build_mbi(fastboot_args, &newkernel) != 0) {
1192		goto err_out;
1193	}
1194
1195	/*
1196	 * Build page table for low 1G physical memory. Use big pages.
1197	 * Allocate 4 (5 for amd64) pages for the page tables.
1198	 *    1 page for PML4 (amd64)
1199	 *    1 page for Page-Directory-Pointer Table
1200	 *    2 pages for Page Directory
1201	 *    1 page for Page Table.
1202	 * The page table entry will be rewritten to map the physical
1203	 * address as we do the copying.
1204	 */
1205	if (newkernel.fi_has_pae) {
1206#ifdef	__amd64
1207		size_t size = MMU_PAGESIZE * 5;
1208#else
1209		size_t size = MMU_PAGESIZE * 4;
1210#endif	/* __amd64 */
1211
1212		if (newkernel.fi_pagetable_size && newkernel.fi_pagetable_size
1213		    < size) {
1214			contig_free((void *)newkernel.fi_pagetable_va,
1215			    newkernel.fi_pagetable_size);
1216			newkernel.fi_pagetable_size = 0;
1217		}
1218
1219		if (newkernel.fi_pagetable_size == 0) {
1220			if ((newkernel.fi_pagetable_va = (uintptr_t)
1221			    contig_alloc(size, &fastboot_below_1G_dma_attr,
1222			    MMU_PAGESIZE, 0)) == NULL) {
1223				cmn_err(CE_WARN, fastboot_enomem_msg,
1224				    (uint64_t)size, "1G");
1225				goto err_out;
1226			}
1227			/*
1228			 * fi_pagetable_size must be set after the allocation
1229			 * succeeds as it's used to determine how much memory to
1230			 * free.
1231			 */
1232			newkernel.fi_pagetable_size = size;
1233		}
1234
1235		bzero((void *)(newkernel.fi_pagetable_va), size);
1236
1237		newkernel.fi_pagetable_pa =
1238		    mmu_ptob((uint64_t)hat_getpfnum(kas.a_hat,
1239		    (caddr_t)newkernel.fi_pagetable_va));
1240
1241		newkernel.fi_last_table_pa = newkernel.fi_pagetable_pa +
1242		    size - MMU_PAGESIZE;
1243
1244		newkernel.fi_next_table_va = newkernel.fi_pagetable_va +
1245		    MMU_PAGESIZE;
1246		newkernel.fi_next_table_pa = newkernel.fi_pagetable_pa +
1247		    MMU_PAGESIZE;
1248
1249		fastboot_build_pagetables(&newkernel);
1250	}
1251
1252
1253	/* Generate MD5 checksums */
1254	fastboot_cksum_generate(&newkernel);
1255
1256	/* Mark it as valid */
1257	newkernel.fi_valid = 1;
1258	newkernel.fi_magic = FASTBOOT_MAGIC;
1259
1260	postbootkernelbase = saved_kernelbase;
1261	return;
1262
1263err_out:
1264	postbootkernelbase = saved_kernelbase;
1265	newkernel.fi_valid = 0;
1266	fastboot_free_newkernel(&newkernel);
1267}
1268
1269
1270/* ARGSUSED */
1271static int
1272fastboot_xc_func(fastboot_info_t *nk, xc_arg_t unused2, xc_arg_t unused3)
1273{
1274	void (*fastboot_func)(fastboot_info_t *);
1275	fastboot_file_t	*fb = &nk->fi_files[FASTBOOT_SWTCH];
1276	fastboot_func = (void (*)())(fb->fb_va);
1277	kthread_t *t_intr = curthread->t_intr;
1278
1279	if (&kas != curproc->p_as) {
1280		hat_devload(curproc->p_as->a_hat, (caddr_t)fb->fb_va,
1281		    MMU_PAGESIZE, mmu_btop(fb->fb_dest_pa),
1282		    PROT_READ | PROT_WRITE | PROT_EXEC,
1283		    HAT_LOAD_NOCONSIST | HAT_LOAD_LOCK);
1284	}
1285
1286	/*
1287	 * If we have pinned a thread, make sure the address is mapped
1288	 * in the address space of the pinned thread.
1289	 */
1290	if (t_intr && t_intr->t_procp->p_as->a_hat != curproc->p_as->a_hat &&
1291	    t_intr->t_procp->p_as != &kas)
1292		hat_devload(t_intr->t_procp->p_as->a_hat, (caddr_t)fb->fb_va,
1293		    MMU_PAGESIZE, mmu_btop(fb->fb_dest_pa),
1294		    PROT_READ | PROT_WRITE | PROT_EXEC,
1295		    HAT_LOAD_NOCONSIST | HAT_LOAD_LOCK);
1296
1297	(*psm_shutdownf)(A_SHUTDOWN, AD_FASTREBOOT);
1298	(*fastboot_func)(nk);
1299
1300	/*NOTREACHED*/
1301	return (0);
1302}
1303
1304/*
1305 * Jump to the fast reboot switcher.  This function never returns.
1306 */
1307void
1308fast_reboot()
1309{
1310	processorid_t bootcpuid = 0;
1311	extern uintptr_t postbootkernelbase;
1312	extern char	fb_swtch_image[];
1313	fastboot_file_t	*fb;
1314	int i;
1315
1316	postbootkernelbase = 0;
1317
1318	fb = &newkernel.fi_files[FASTBOOT_SWTCH];
1319
1320	/*
1321	 * Map the address into both the current proc's address
1322	 * space and the kernel's address space in case the panic
1323	 * is forced by kmdb.
1324	 */
1325	if (&kas != curproc->p_as) {
1326		hat_devload(curproc->p_as->a_hat, (caddr_t)fb->fb_va,
1327		    MMU_PAGESIZE, mmu_btop(fb->fb_dest_pa),
1328		    PROT_READ | PROT_WRITE | PROT_EXEC,
1329		    HAT_LOAD_NOCONSIST | HAT_LOAD_LOCK);
1330	}
1331
1332	bcopy((void *)fb_swtch_image, (void *)fb->fb_va, fb->fb_size);
1333
1334
1335	/*
1336	 * Set fb_va to fake_va
1337	 */
1338	for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
1339		newkernel.fi_files[i].fb_va = fake_va;
1340
1341	}
1342
1343	if (panicstr && CPU->cpu_id != bootcpuid &&
1344	    CPU_ACTIVE(cpu_get(bootcpuid))) {
1345		extern void panic_idle(void);
1346		cpuset_t cpuset;
1347
1348		CPUSET_ZERO(cpuset);
1349		CPUSET_ADD(cpuset, bootcpuid);
1350		xc_priority((xc_arg_t)&newkernel, 0, 0, CPUSET2BV(cpuset),
1351		    (xc_func_t)fastboot_xc_func);
1352
1353		panic_idle();
1354	} else
1355		(void) fastboot_xc_func(&newkernel, 0, 0);
1356}
1357
1358
1359/*
1360 * Get boot property value for fastreboot_onpanic.
1361 *
1362 * NOTE: If fastreboot_onpanic is set to non-zero in /etc/system,
1363 * new setting passed in via "-B fastreboot_onpanic" is ignored.
1364 * This order of precedence is to enable developers debugging panics
1365 * that occur early in boot to utilize Fast Reboot on panic.
1366 */
1367static void
1368fastboot_get_bootprop(void)
1369{
1370	int		val = 0xaa, len, ret;
1371	dev_info_t	*devi;
1372	char		*propstr = NULL;
1373
1374	devi = ddi_root_node();
1375
1376	ret = ddi_prop_lookup_string(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS,
1377	    FASTREBOOT_ONPANIC, &propstr);
1378
1379	if (ret == DDI_PROP_SUCCESS) {
1380		if (FASTREBOOT_ONPANIC_NOTSET(propstr))
1381			val = 0;
1382		else if (FASTREBOOT_ONPANIC_ISSET(propstr))
1383			val = UA_FASTREBOOT_ONPANIC;
1384
1385		/*
1386		 * Only set fastreboot_onpanic to the value passed in
1387		 * if it's not already set to non-zero, and the value
1388		 * has indeed been passed in via command line.
1389		 */
1390		if (!fastreboot_onpanic && val != 0xaa)
1391			fastreboot_onpanic = val;
1392		ddi_prop_free(propstr);
1393	} else if (ret != DDI_PROP_NOT_FOUND && ret != DDI_PROP_UNDEFINED) {
1394		cmn_err(CE_WARN, "%s value is invalid, will be ignored",
1395		    FASTREBOOT_ONPANIC);
1396	}
1397
1398	len = sizeof (fastreboot_onpanic_cmdline);
1399	ret = ddi_getlongprop_buf(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS,
1400	    FASTREBOOT_ONPANIC_CMDLINE, fastreboot_onpanic_cmdline, &len);
1401
1402	if (ret == DDI_PROP_BUF_TOO_SMALL)
1403		cmn_err(CE_WARN, "%s value is too long, will be ignored",
1404		    FASTREBOOT_ONPANIC_CMDLINE);
1405}
1406
1407/*
1408 * This function is called by main() to either load the backup kernel for panic
1409 * fast reboot, or to reserve low physical memory for fast reboot.
1410 */
1411void
1412fastboot_post_startup()
1413{
1414	if (!fastreboot_capable)
1415		return;
1416
1417	mutex_enter(&fastreboot_config_mutex);
1418
1419	fastboot_get_bootprop();
1420
1421	if (fastreboot_onpanic)
1422		fastboot_load_kernel(fastreboot_onpanic_cmdline);
1423	else if (reserve_mem_enabled)
1424		fastboot_reserve_mem(&newkernel);
1425
1426	mutex_exit(&fastreboot_config_mutex);
1427}
1428
1429/*
1430 * Update boot configuration settings.
1431 * If the new fastreboot_onpanic setting is false, and a kernel has
1432 * been preloaded, free the memory;
1433 * if the new fastreboot_onpanic setting is true and newkernel is
1434 * not valid, load the new kernel.
1435 */
1436void
1437fastboot_update_config(const char *mdep)
1438{
1439	uint8_t boot_config = (uint8_t)*mdep;
1440	int cur_fastreboot_onpanic;
1441
1442	if (!fastreboot_capable)
1443		return;
1444
1445	mutex_enter(&fastreboot_config_mutex);
1446
1447	cur_fastreboot_onpanic = fastreboot_onpanic;
1448	fastreboot_onpanic = boot_config & UA_FASTREBOOT_ONPANIC;
1449
1450	if (fastreboot_onpanic && (!cur_fastreboot_onpanic ||
1451	    !newkernel.fi_valid))
1452		fastboot_load_kernel(fastreboot_onpanic_cmdline);
1453	if (cur_fastreboot_onpanic && !fastreboot_onpanic)
1454		fastboot_free_newkernel(&newkernel);
1455
1456	mutex_exit(&fastreboot_config_mutex);
1457}
1458
1459/*
1460 * This is the interface to be called by other kernel components to
1461 * disable fastreboot_onpanic.
1462 */
1463void
1464fastreboot_disable()
1465{
1466	uint8_t boot_config = (uint8_t)(~UA_FASTREBOOT_ONPANIC);
1467	fastboot_update_config((const char *)&boot_config);
1468}
1469
1470/*
1471 * This is the interface to be called by fm_panic() in case FMA has diagnosed
1472 * a terminal machine check exception.  It does not free up memory allocated
1473 * for the backup kernel.  General disabling fastreboot_onpanic in a
1474 * non-panicking situation must go through fastboot_update_config().
1475 */
1476void
1477fastreboot_disable_highpil()
1478{
1479	fastreboot_onpanic = 0;
1480}
1481
1482
1483/*
1484 * A simplified interface for uadmin to call to update the configuration
1485 * setting and load a new kernel if necessary.
1486 */
1487void
1488fastboot_update_and_load(int fcn, char *mdep)
1489{
1490	if (fcn != AD_FASTREBOOT) {
1491		/*
1492		 * If user has explicitly requested reboot to prom,
1493		 * or uadmin(1M) was invoked with other functions,
1494		 * don't try to fast reboot after dumping.
1495		 */
1496		fastreboot_disable();
1497	}
1498
1499	mutex_enter(&fastreboot_config_mutex);
1500
1501	if (fastreboot_onpanic)
1502		fastboot_load_kernel(mdep);
1503
1504	mutex_exit(&fastreboot_config_mutex);
1505}
1506