Home | History | Annotate | Line # | Download | only in arm32
arm32_kvminit.c revision 1.32
      1 /*	$NetBSD: arm32_kvminit.c,v 1.32 2014/10/29 14:14:14 skrll Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2002, 2003, 2005  Genetec Corporation.  All rights reserved.
      5  * Written by Hiroyuki Bessho for Genetec Corporation.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. The name of Genetec Corporation may not be used to endorse or
     16  *    promote products derived from this software without specific prior
     17  *    written permission.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY GENETEC CORPORATION ``AS IS'' AND
     20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL GENETEC CORPORATION
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  *
     31  * Copyright (c) 2001 Wasabi Systems, Inc.
     32  * All rights reserved.
     33  *
     34  * Written by Jason R. Thorpe for Wasabi Systems, Inc.
     35  *
     36  * Redistribution and use in source and binary forms, with or without
     37  * modification, are permitted provided that the following conditions
     38  * are met:
     39  * 1. Redistributions of source code must retain the above copyright
     40  *    notice, this list of conditions and the following disclaimer.
     41  * 2. Redistributions in binary form must reproduce the above copyright
     42  *    notice, this list of conditions and the following disclaimer in the
     43  *    documentation and/or other materials provided with the distribution.
     44  * 3. All advertising materials mentioning features or use of this software
     45  *    must display the following acknowledgement:
     46  *	This product includes software developed for the NetBSD Project by
     47  *	Wasabi Systems, Inc.
     48  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
     49  *    or promote products derived from this software without specific prior
     50  *    written permission.
     51  *
     52  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
     53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     54  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     55  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
     56  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     57  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     58  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     59  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     60  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     61  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     62  * POSSIBILITY OF SUCH DAMAGE.
     63  *
     64  * Copyright (c) 1997,1998 Mark Brinicombe.
     65  * Copyright (c) 1997,1998 Causality Limited.
     66  * All rights reserved.
     67  *
     68  * Redistribution and use in source and binary forms, with or without
     69  * modification, are permitted provided that the following conditions
     70  * are met:
     71  * 1. Redistributions of source code must retain the above copyright
     72  *    notice, this list of conditions and the following disclaimer.
     73  * 2. Redistributions in binary form must reproduce the above copyright
     74  *    notice, this list of conditions and the following disclaimer in the
     75  *    documentation and/or other materials provided with the distribution.
     76  * 3. All advertising materials mentioning features or use of this software
     77  *    must display the following acknowledgement:
     78  *	This product includes software developed by Mark Brinicombe
     79  *	for the NetBSD Project.
     80  * 4. The name of the company nor the name of the author may be used to
     81  *    endorse or promote products derived from this software without specific
     82  *    prior written permission.
     83  *
     84  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
     85  * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
     86  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     87  * IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
     88  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
     89  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     90  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     91  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     92  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     93  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     94  * SUCH DAMAGE.
     95  *
     96  * Copyright (c) 2007 Microsoft
     97  * All rights reserved.
     98  *
     99  * Redistribution and use in source and binary forms, with or without
    100  * modification, are permitted provided that the following conditions
    101  * are met:
    102  * 1. Redistributions of source code must retain the above copyright
    103  *    notice, this list of conditions and the following disclaimer.
    104  * 2. Redistributions in binary form must reproduce the above copyright
    105  *    notice, this list of conditions and the following disclaimer in the
    106  *    documentation and/or other materials provided with the distribution.
    107  * 3. All advertising materials mentioning features or use of this software
    108  *    must display the following acknowledgement:
    109  *	This product includes software developed by Microsoft
    110  *
    111  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
    112  * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
    113  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
    114  * IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTERS BE LIABLE FOR ANY DIRECT,
    115  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
    116  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
    117  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
    118  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
    119  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
    120  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
    121  * SUCH DAMAGE.
    122  */
    123 
    124 #include "opt_multiprocessor.h"
    125 
    126 #include <sys/cdefs.h>
    127 __KERNEL_RCSID(0, "$NetBSD: arm32_kvminit.c,v 1.32 2014/10/29 14:14:14 skrll Exp $");
    128 
    129 #include <sys/param.h>
    130 #include <sys/device.h>
    131 #include <sys/kernel.h>
    132 #include <sys/reboot.h>
    133 #include <sys/bus.h>
    134 
    135 #include <dev/cons.h>
    136 
    137 #include <uvm/uvm_extern.h>
    138 
    139 #include <arm/locore.h>
    140 #include <arm/db_machdep.h>
    141 #include <arm/undefined.h>
    142 #include <arm/bootconfig.h>
    143 #include <arm/arm32/machdep.h>
    144 
    145 struct bootmem_info bootmem_info;
    146 
    147 extern void *msgbufaddr;
    148 paddr_t msgbufphys;
    149 paddr_t physical_start;
    150 paddr_t physical_end;
    151 
    152 extern char etext[];
    153 extern char __data_start[], _edata[];
    154 extern char __bss_start[], __bss_end__[];
    155 extern char _end[];
    156 
    157 /* Page tables for mapping kernel VM */
    158 #define KERNEL_L2PT_VMDATA_NUM	8	/* start with 32MB of KVM */
    159 
    160 /*
    161  * Macros to translate between physical and virtual for a subset of the
    162  * kernel address space.  *Not* for general use.
    163  */
    164 #if defined(KERNEL_BASE_VOFFSET)
    165 #define KERN_VTOPHYS(bmi, va) \
    166 	((paddr_t)((vaddr_t)(va) - KERNEL_BASE_VOFFSET))
    167 #define KERN_PHYSTOV(bmi, pa) \
    168 	((vaddr_t)((paddr_t)(pa) + KERNEL_BASE_VOFFSET))
    169 #elif defined(ARM_MMU_EXTENDED) && defined(__HAVE_MM_MD_DIRECT_MAPPED_PHYS)
    170 #define KERN_VTOPHYS(bmi, va) \
    171 	((paddr_t)((vaddr_t)(va) - pmap_directbase + (bmi)->bmi_start))
    172 #define KERN_PHYSTOV(bmi, pa) \
    173 	((vaddr_t)((paddr_t)(pa) - (bmi)->bmi_start + pmap_directbase))
    174 #else
    175 #define KERN_VTOPHYS(bmi, va) \
    176 	((paddr_t)((vaddr_t)(va) - KERNEL_BASE + (bmi)->bmi_start))
    177 #define KERN_PHYSTOV(bmi, pa) \
    178 	((vaddr_t)((paddr_t)(pa) - (bmi)->bmi_start + KERNEL_BASE))
    179 #endif
    180 
    181 void
    182 arm32_bootmem_init(paddr_t memstart, psize_t memsize, vsize_t kernelstart)
    183 {
    184 	struct bootmem_info * const bmi = &bootmem_info;
    185 	pv_addr_t *pv = bmi->bmi_freeblocks;
    186 
    187 #ifdef VERBOSE_INIT_ARM
    188 	printf("%s: memstart=%#lx, memsize=%#lx, kernelstart=%#lx\n",
    189 	    __func__, memstart, memsize, kernelstart);
    190 #endif
    191 
    192 	physical_start = bmi->bmi_start = memstart;
    193 	physical_end = bmi->bmi_end = memstart + memsize;
    194 	physmem = memsize / PAGE_SIZE;
    195 
    196 	/*
    197 	 * Let's record where the kernel lives.
    198 	 */
    199 	bmi->bmi_kernelstart = kernelstart;
    200 	bmi->bmi_kernelend = KERN_VTOPHYS(bmi, round_page((vaddr_t)_end));
    201 
    202 #ifdef VERBOSE_INIT_ARM
    203 	printf("%s: kernelend=%#lx\n", __func__, bmi->bmi_kernelend);
    204 #endif
    205 
    206 	/*
    207 	 * Now the rest of the free memory must be after the kernel.
    208 	 */
    209 	pv->pv_pa = bmi->bmi_kernelend;
    210 	pv->pv_va = KERN_PHYSTOV(bmi, pv->pv_pa);
    211 	pv->pv_size = bmi->bmi_end - bmi->bmi_kernelend;
    212 	bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
    213 #ifdef VERBOSE_INIT_ARM
    214 	printf("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
    215 	    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
    216 	    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
    217 #endif
    218 	pv++;
    219 
    220 	/*
    221 	 * Add a free block for any memory before the kernel.
    222 	 */
    223 	if (bmi->bmi_start < bmi->bmi_kernelstart) {
    224 		pv->pv_pa = bmi->bmi_start;
    225 #if defined(ARM_MMU_EXTENDED) && defined(__HAVE_MM_MD_DIRECT_MAPPED_PHYS)
    226 		pv->pv_va = pmap_directbase;
    227 #else
    228 		pv->pv_va = KERNEL_BASE;
    229 #endif
    230 		pv->pv_size = bmi->bmi_kernelstart - bmi->bmi_start;
    231 		bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
    232 #ifdef VERBOSE_INIT_ARM
    233 		printf("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
    234 		    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
    235 		    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
    236 #endif
    237 		pv++;
    238 	}
    239 
    240 	bmi->bmi_nfreeblocks = pv - bmi->bmi_freeblocks;
    241 
    242 	SLIST_INIT(&bmi->bmi_freechunks);
    243 	SLIST_INIT(&bmi->bmi_chunks);
    244 }
    245 
    246 static bool
    247 concat_pvaddr(pv_addr_t *acc_pv, pv_addr_t *pv)
    248 {
    249 	if (acc_pv->pv_pa + acc_pv->pv_size == pv->pv_pa
    250 	    && acc_pv->pv_va + acc_pv->pv_size == pv->pv_va
    251 	    && acc_pv->pv_prot == pv->pv_prot
    252 	    && acc_pv->pv_cache == pv->pv_cache) {
    253 #ifdef VERBOSE_INIT_ARMX
    254 		printf("%s: appending pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    255 		    __func__, pv, pv->pv_pa, pv->pv_pa + pv->pv_size + 1,
    256 		    acc_pv->pv_pa, acc_pv->pv_pa + acc_pv->pv_size + 1);
    257 #endif
    258 		acc_pv->pv_size += pv->pv_size;
    259 		return true;
    260 	}
    261 
    262 	return false;
    263 }
    264 
    265 static void
    266 add_pages(struct bootmem_info *bmi, pv_addr_t *pv)
    267 {
    268 	pv_addr_t **pvp = &SLIST_FIRST(&bmi->bmi_chunks);
    269 	while ((*pvp) != NULL && (*pvp)->pv_va <= pv->pv_va) {
    270 		pv_addr_t * const pv0 = (*pvp);
    271 		KASSERT(SLIST_NEXT(pv0, pv_list) == NULL || pv0->pv_pa < SLIST_NEXT(pv0, pv_list)->pv_pa);
    272 		if (concat_pvaddr(pv0, pv)) {
    273 #ifdef VERBOSE_INIT_ARM
    274 			printf("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    275 			    __func__, "appending", pv,
    276 			    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    277 			    pv0->pv_pa, pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
    278 #endif
    279 			pv = SLIST_NEXT(pv0, pv_list);
    280 			if (pv != NULL && concat_pvaddr(pv0, pv)) {
    281 #ifdef VERBOSE_INIT_ARM
    282 				printf("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    283 				    __func__, "merging", pv,
    284 				    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    285 				    pv0->pv_pa,
    286 				    pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
    287 #endif
    288 				SLIST_REMOVE_AFTER(pv0, pv_list);
    289 				SLIST_INSERT_HEAD(&bmi->bmi_freechunks, pv, pv_list);
    290 			}
    291 			return;
    292 		}
    293 		KASSERT(pv->pv_va != (*pvp)->pv_va);
    294 		pvp = &SLIST_NEXT(*pvp, pv_list);
    295 	}
    296 	KASSERT((*pvp) == NULL || pv->pv_va < (*pvp)->pv_va);
    297 	pv_addr_t * const new_pv = SLIST_FIRST(&bmi->bmi_freechunks);
    298 	KASSERT(new_pv != NULL);
    299 	SLIST_REMOVE_HEAD(&bmi->bmi_freechunks, pv_list);
    300 	*new_pv = *pv;
    301 	SLIST_NEXT(new_pv, pv_list) = *pvp;
    302 	(*pvp) = new_pv;
    303 #ifdef VERBOSE_INIT_ARM
    304 	printf("%s: adding pv %p (pa %#lx, va %#lx, %lu pages) ",
    305 	    __func__, new_pv, new_pv->pv_pa, new_pv->pv_va,
    306 	    new_pv->pv_size / PAGE_SIZE);
    307 	if (SLIST_NEXT(new_pv, pv_list))
    308 		printf("before pa %#lx\n", SLIST_NEXT(new_pv, pv_list)->pv_pa);
    309 	else
    310 		printf("at tail\n");
    311 #endif
    312 }
    313 
    314 static void
    315 valloc_pages(struct bootmem_info *bmi, pv_addr_t *pv, size_t npages,
    316 	int prot, int cache, bool zero_p)
    317 {
    318 	size_t nbytes = npages * PAGE_SIZE;
    319 	pv_addr_t *free_pv = bmi->bmi_freeblocks;
    320 	size_t free_idx = 0;
    321 	static bool l1pt_found;
    322 
    323 	KASSERT(npages > 0);
    324 
    325 	/*
    326 	 * If we haven't allocated the kernel L1 page table and we are aligned
    327 	 * at a L1 table boundary, alloc the memory for it.
    328 	 */
    329 	if (!l1pt_found
    330 	    && (free_pv->pv_pa & (L1_TABLE_SIZE - 1)) == 0
    331 	    && free_pv->pv_size >= L1_TABLE_SIZE) {
    332 		l1pt_found = true;
    333 		valloc_pages(bmi, &kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE,
    334 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    335 		add_pages(bmi, &kernel_l1pt);
    336 	}
    337 
    338 	while (nbytes > free_pv->pv_size) {
    339 		free_pv++;
    340 		free_idx++;
    341 		if (free_idx == bmi->bmi_nfreeblocks) {
    342 			panic("%s: could not allocate %zu bytes",
    343 			    __func__, nbytes);
    344 		}
    345 	}
    346 
    347 	/*
    348 	 * As we allocate the memory, make sure that we don't walk over
    349 	 * our current first level translation table.
    350 	 */
    351 	KASSERT((armreg_ttbr_read() & ~(L1_TABLE_SIZE - 1)) != free_pv->pv_pa);
    352 
    353 	pv->pv_pa = free_pv->pv_pa;
    354 	pv->pv_va = free_pv->pv_va;
    355 	pv->pv_size = nbytes;
    356 	pv->pv_prot = prot;
    357 	pv->pv_cache = cache;
    358 
    359 	/*
    360 	 * If PTE_PAGETABLE uses the same cache modes as PTE_CACHE
    361 	 * just use PTE_CACHE.
    362 	 */
    363 	if (cache == PTE_PAGETABLE
    364 	    && pte_l1_s_cache_mode == pte_l1_s_cache_mode_pt
    365 	    && pte_l2_l_cache_mode == pte_l2_l_cache_mode_pt
    366 	    && pte_l2_s_cache_mode == pte_l2_s_cache_mode_pt)
    367 		pv->pv_cache = PTE_CACHE;
    368 
    369 	free_pv->pv_pa += nbytes;
    370 	free_pv->pv_va += nbytes;
    371 	free_pv->pv_size -= nbytes;
    372 	if (free_pv->pv_size == 0) {
    373 		--bmi->bmi_nfreeblocks;
    374 		for (; free_idx < bmi->bmi_nfreeblocks; free_idx++) {
    375 			free_pv[0] = free_pv[1];
    376 		}
    377 	}
    378 
    379 	bmi->bmi_freepages -= npages;
    380 
    381 	if (zero_p)
    382 		memset((void *)pv->pv_va, 0, nbytes);
    383 }
    384 
    385 void
    386 arm32_kernel_vm_init(vaddr_t kernel_vm_base, vaddr_t vectors, vaddr_t iovbase,
    387 	const struct pmap_devmap *devmap, bool mapallmem_p)
    388 {
    389 	struct bootmem_info * const bmi = &bootmem_info;
    390 #ifdef MULTIPROCESSOR
    391 	const size_t cpu_num = arm_cpu_max;
    392 #else
    393 	const size_t cpu_num = 1;
    394 #endif
    395 #ifdef ARM_HAS_VBAR
    396 	const bool map_vectors_p = false;
    397 #elif defined(CPU_ARMV7) || defined(CPU_ARM11)
    398 	const bool map_vectors_p = vectors == ARM_VECTORS_HIGH
    399 	    || (armreg_pfr1_read() & ARM_PFR1_SEC_MASK) == 0;
    400 #else
    401 	const bool map_vectors_p = true;
    402 #endif
    403 
    404 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS
    405 	KASSERT(mapallmem_p);
    406 #ifdef ARM_MMU_EXTENDED
    407 	/*
    408 	 * We can only use address beneath kernel_vm_base to map physical
    409 	 * memory.
    410 	 */
    411 	const psize_t physical_size =
    412 	    roundup(physical_end - physical_start, L1_SS_SIZE);
    413 	KASSERT(kernel_vm_base >= physical_size);
    414 	/*
    415 	 * If we don't have enough memory via TTBR1, we have use addresses
    416 	 * from TTBR0 to map some of the physical memory.  But try to use as
    417 	 * much high memory space as possible.
    418 	 */
    419 	if (kernel_vm_base - KERNEL_BASE < physical_size) {
    420 		pmap_directbase = kernel_vm_base - physical_size;
    421 		printf("%s: changing pmap_directbase to %#lx\n", __func__,
    422 		    pmap_directbase);
    423 	}
    424 #else
    425 	KASSERT(kernel_vm_base - KERNEL_BASE >= physical_end - physical_start);
    426 #endif /* ARM_MMU_EXTENDED */
    427 #endif /* __HAVE_MM_MD_DIRECT_MAPPED_PHYS */
    428 
    429 	/*
    430 	 * Calculate the number of L2 pages needed for mapping the
    431 	 * kernel + data + stuff.  Assume 2 L2 pages for kernel, 1 for vectors,
    432 	 * and 1 for IO
    433 	 */
    434 	size_t kernel_size = bmi->bmi_kernelend;
    435 	kernel_size -= (bmi->bmi_kernelstart & -L2_S_SEGSIZE);
    436 	kernel_size += L1_TABLE_SIZE_REAL;
    437 	kernel_size += PAGE_SIZE * KERNEL_L2PT_VMDATA_NUM;
    438 	if (map_vectors_p) {
    439 		kernel_size += PAGE_SIZE;	/* L2PT for VECTORS */
    440 	}
    441 	if (iovbase) {
    442 		kernel_size += PAGE_SIZE;	/* L2PT for IO */
    443 	}
    444 	kernel_size +=
    445 	    cpu_num * (ABT_STACK_SIZE + FIQ_STACK_SIZE + IRQ_STACK_SIZE
    446 	    + UND_STACK_SIZE + UPAGES) * PAGE_SIZE;
    447 	kernel_size += round_page(MSGBUFSIZE);
    448 	kernel_size += 0x10000;	/* slop */
    449 	if (!mapallmem_p) {
    450 		kernel_size += PAGE_SIZE
    451 		    * ((kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE);
    452 	}
    453 	kernel_size = round_page(kernel_size);
    454 
    455 	/*
    456 	 * Now we know how many L2 pages it will take.  If we've mapped
    457 	 * all of memory, then it won't take any.
    458 	 */
    459 	const size_t KERNEL_L2PT_KERNEL_NUM = mapallmem_p
    460 	    ? 0 : round_page(kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE;
    461 
    462 #ifdef VERBOSE_INIT_ARM
    463 	printf("%s: %zu L2 pages are needed to map %#zx kernel bytes\n",
    464 	    __func__, KERNEL_L2PT_KERNEL_NUM, kernel_size);
    465 #endif
    466 
    467 	KASSERT(KERNEL_L2PT_KERNEL_NUM + KERNEL_L2PT_VMDATA_NUM < __arraycount(bmi->bmi_l2pts));
    468 	pv_addr_t * const kernel_l2pt = bmi->bmi_l2pts;
    469 	pv_addr_t * const vmdata_l2pt = kernel_l2pt + KERNEL_L2PT_KERNEL_NUM;
    470 	pv_addr_t msgbuf;
    471 	pv_addr_t text;
    472 	pv_addr_t data;
    473 	pv_addr_t chunks[KERNEL_L2PT_KERNEL_NUM+KERNEL_L2PT_VMDATA_NUM+11];
    474 #if ARM_MMU_XSCALE == 1
    475 	pv_addr_t minidataclean;
    476 #endif
    477 
    478 	/*
    479 	 * We need to allocate some fixed page tables to get the kernel going.
    480 	 *
    481 	 * We are going to allocate our bootstrap pages from the beginning of
    482 	 * the free space that we just calculated.  We allocate one page
    483 	 * directory and a number of page tables and store the physical
    484 	 * addresses in the bmi_l2pts array in bootmem_info.
    485 	 *
    486 	 * The kernel page directory must be on a 16K boundary.  The page
    487 	 * tables must be on 4K boundaries.  What we do is allocate the
    488 	 * page directory on the first 16K boundary that we encounter, and
    489 	 * the page tables on 4K boundaries otherwise.  Since we allocate
    490 	 * at least 3 L2 page tables, we are guaranteed to encounter at
    491 	 * least one 16K aligned region.
    492 	 */
    493 
    494 #ifdef VERBOSE_INIT_ARM
    495 	printf("%s: allocating page tables for", __func__);
    496 #endif
    497 	for (size_t i = 0; i < __arraycount(chunks); i++) {
    498 		SLIST_INSERT_HEAD(&bmi->bmi_freechunks, &chunks[i], pv_list);
    499 	}
    500 
    501 	kernel_l1pt.pv_pa = 0;
    502 	kernel_l1pt.pv_va = 0;
    503 
    504 	/*
    505 	 * Allocate the L2 pages, but if we get to a page that is aligned for
    506 	 * an L1 page table, we will allocate the pages for it first and then
    507 	 * allocate the L2 page.
    508 	 */
    509 
    510 	if (map_vectors_p) {
    511 		/*
    512 		 * First allocate L2 page for the vectors.
    513 		 */
    514 #ifdef VERBOSE_INIT_ARM
    515 		printf(" vector");
    516 #endif
    517 		valloc_pages(bmi, &bmi->bmi_vector_l2pt, 1,
    518 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    519 		add_pages(bmi, &bmi->bmi_vector_l2pt);
    520 	}
    521 
    522 	/*
    523 	 * Now allocate L2 pages for the kernel
    524 	 */
    525 #ifdef VERBOSE_INIT_ARM
    526 	printf(" kernel");
    527 #endif
    528 	KASSERT(mapallmem_p || KERNEL_L2PT_KERNEL_NUM > 0);
    529 	KASSERT(!mapallmem_p || KERNEL_L2PT_KERNEL_NUM == 0);
    530 	for (size_t idx = 0; idx < KERNEL_L2PT_KERNEL_NUM; ++idx) {
    531 		valloc_pages(bmi, &kernel_l2pt[idx], 1,
    532 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    533 		add_pages(bmi, &kernel_l2pt[idx]);
    534 	}
    535 
    536 	/*
    537 	 * Now allocate L2 pages for the initial kernel VA space.
    538 	 */
    539 #ifdef VERBOSE_INIT_ARM
    540 	printf(" vm");
    541 #endif
    542 	for (size_t idx = 0; idx < KERNEL_L2PT_VMDATA_NUM; ++idx) {
    543 		valloc_pages(bmi, &vmdata_l2pt[idx], 1,
    544 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    545 		add_pages(bmi, &vmdata_l2pt[idx]);
    546 	}
    547 
    548 	/*
    549 	 * If someone wanted a L2 page for I/O, allocate it now.
    550 	 */
    551 	if (iovbase) {
    552 #ifdef VERBOSE_INIT_ARM
    553 		printf(" io");
    554 #endif
    555 		valloc_pages(bmi, &bmi->bmi_io_l2pt, 1,
    556 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    557 		add_pages(bmi, &bmi->bmi_io_l2pt);
    558 	}
    559 
    560 #ifdef VERBOSE_INIT_ARM
    561 	printf("%s: allocating stacks\n", __func__);
    562 #endif
    563 
    564 	/* Allocate stacks for all modes and CPUs */
    565 	valloc_pages(bmi, &abtstack, ABT_STACK_SIZE * cpu_num,
    566 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    567 	add_pages(bmi, &abtstack);
    568 	valloc_pages(bmi, &fiqstack, FIQ_STACK_SIZE * cpu_num,
    569 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    570 	add_pages(bmi, &fiqstack);
    571 	valloc_pages(bmi, &irqstack, IRQ_STACK_SIZE * cpu_num,
    572 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    573 	add_pages(bmi, &irqstack);
    574 	valloc_pages(bmi, &undstack, UND_STACK_SIZE * cpu_num,
    575 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    576 	add_pages(bmi, &undstack);
    577 	valloc_pages(bmi, &idlestack, UPAGES * cpu_num,		/* SVC32 */
    578 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    579 	add_pages(bmi, &idlestack);
    580 	valloc_pages(bmi, &kernelstack, UPAGES,			/* SVC32 */
    581 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    582 	add_pages(bmi, &kernelstack);
    583 
    584 	/* Allocate the message buffer from the end of memory. */
    585 	const size_t msgbuf_pgs = round_page(MSGBUFSIZE) / PAGE_SIZE;
    586 	valloc_pages(bmi, &msgbuf, msgbuf_pgs,
    587 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, false);
    588 	add_pages(bmi, &msgbuf);
    589 	msgbufphys = msgbuf.pv_pa;
    590 	msgbufaddr = (void *)msgbuf.pv_va;
    591 
    592 	if (map_vectors_p) {
    593 		/*
    594 		 * Allocate a page for the system vector page.
    595 		 * This page will just contain the system vectors and can be
    596 		 * shared by all processes.
    597 		 */
    598 		valloc_pages(bmi, &systempage, 1, VM_PROT_READ|VM_PROT_WRITE,
    599 		    PTE_CACHE, true);
    600 	}
    601 	systempage.pv_va = vectors;
    602 
    603 	/*
    604 	 * If the caller needed a few extra pages for some reason, allocate
    605 	 * them now.
    606 	 */
    607 #if ARM_MMU_XSCALE == 1
    608 #if (ARM_NMMUS > 1)
    609 	if (xscale_use_minidata)
    610 #endif
    611 		valloc_pages(bmi, &minidataclean, 1,
    612 		    VM_PROT_READ|VM_PROT_WRITE, 0, true);
    613 #endif
    614 
    615 	/*
    616 	 * Ok we have allocated physical pages for the primary kernel
    617 	 * page tables and stacks.  Let's just confirm that.
    618 	 */
    619 	if (kernel_l1pt.pv_va == 0
    620 	    && (!kernel_l1pt.pv_pa || (kernel_l1pt.pv_pa & (L1_TABLE_SIZE - 1)) != 0))
    621 		panic("%s: Failed to allocate or align the kernel "
    622 		    "page directory", __func__);
    623 
    624 
    625 #ifdef VERBOSE_INIT_ARM
    626 	printf("Creating L1 page table at 0x%08lx\n", kernel_l1pt.pv_pa);
    627 #endif
    628 
    629 	/*
    630 	 * Now we start construction of the L1 page table
    631 	 * We start by mapping the L2 page tables into the L1.
    632 	 * This means that we can replace L1 mappings later on if necessary
    633 	 */
    634 	vaddr_t l1pt_va = kernel_l1pt.pv_va;
    635 	paddr_t l1pt_pa = kernel_l1pt.pv_pa;
    636 
    637 	if (map_vectors_p) {
    638 		/* Map the L2 pages tables in the L1 page table */
    639 		pmap_link_l2pt(l1pt_va, systempage.pv_va & -L2_S_SEGSIZE,
    640 		    &bmi->bmi_vector_l2pt);
    641 #ifdef VERBOSE_INIT_ARM
    642 		printf("%s: adding L2 pt (VA %#lx, PA %#lx) "
    643 		    "for VA %#lx\n (vectors)",
    644 		    __func__, bmi->bmi_vector_l2pt.pv_va,
    645 		    bmi->bmi_vector_l2pt.pv_pa, systempage.pv_va);
    646 #endif
    647 	}
    648 
    649 	const vaddr_t kernel_base =
    650 	    KERN_PHYSTOV(bmi, bmi->bmi_kernelstart & -L2_S_SEGSIZE);
    651 	for (size_t idx = 0; idx < KERNEL_L2PT_KERNEL_NUM; idx++) {
    652 		pmap_link_l2pt(l1pt_va, kernel_base + idx * L2_S_SEGSIZE,
    653 		    &kernel_l2pt[idx]);
    654 #ifdef VERBOSE_INIT_ARM
    655 		printf("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (kernel)\n",
    656 		    __func__, kernel_l2pt[idx].pv_va,
    657 		    kernel_l2pt[idx].pv_pa, kernel_base + idx * L2_S_SEGSIZE);
    658 #endif
    659 	}
    660 
    661 	for (size_t idx = 0; idx < KERNEL_L2PT_VMDATA_NUM; idx++) {
    662 		pmap_link_l2pt(l1pt_va, kernel_vm_base + idx * L2_S_SEGSIZE,
    663 		    &vmdata_l2pt[idx]);
    664 #ifdef VERBOSE_INIT_ARM
    665 		printf("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (vm)\n",
    666 		    __func__, vmdata_l2pt[idx].pv_va, vmdata_l2pt[idx].pv_pa,
    667 		    kernel_vm_base + idx * L2_S_SEGSIZE);
    668 #endif
    669 	}
    670 	if (iovbase) {
    671 		pmap_link_l2pt(l1pt_va, iovbase & -L2_S_SEGSIZE, &bmi->bmi_io_l2pt);
    672 #ifdef VERBOSE_INIT_ARM
    673 		printf("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (io)\n",
    674 		    __func__, bmi->bmi_io_l2pt.pv_va, bmi->bmi_io_l2pt.pv_pa,
    675 		    iovbase & -L2_S_SEGSIZE);
    676 #endif
    677 	}
    678 
    679 	/* update the top of the kernel VM */
    680 	pmap_curmaxkvaddr =
    681 	    kernel_vm_base + (KERNEL_L2PT_VMDATA_NUM * L2_S_SEGSIZE);
    682 
    683 #ifdef VERBOSE_INIT_ARM
    684 	printf("Mapping kernel\n");
    685 #endif
    686 
    687 	extern char etext[], _end[];
    688 	size_t totalsize = bmi->bmi_kernelend - bmi->bmi_kernelstart;
    689 	size_t textsize = KERN_VTOPHYS(bmi, (uintptr_t)etext) - bmi->bmi_kernelstart;
    690 
    691 	textsize = (textsize + PGOFSET) & ~PGOFSET;
    692 
    693 	/* start at offset of kernel in RAM */
    694 
    695 	text.pv_pa = bmi->bmi_kernelstart;
    696 	text.pv_va = KERN_PHYSTOV(bmi, bmi->bmi_kernelstart);
    697 	text.pv_size = textsize;
    698 	text.pv_prot = VM_PROT_READ|VM_PROT_WRITE; /* XXX VM_PROT_EXECUTE */
    699 	text.pv_cache = PTE_CACHE;
    700 
    701 #ifdef VERBOSE_INIT_ARM
    702 	printf("%s: adding chunk for kernel text %#lx..%#lx (VA %#lx)\n",
    703 	    __func__, text.pv_pa, text.pv_pa + text.pv_size - 1, text.pv_va);
    704 #endif
    705 
    706 	add_pages(bmi, &text);
    707 
    708 	data.pv_pa = text.pv_pa + textsize;
    709 	data.pv_va = text.pv_va + textsize;
    710 	data.pv_size = totalsize - textsize;
    711 	data.pv_prot = VM_PROT_READ|VM_PROT_WRITE;
    712 	data.pv_cache = PTE_CACHE;
    713 
    714 #ifdef VERBOSE_INIT_ARM
    715 	printf("%s: adding chunk for kernel data/bss %#lx..%#lx (VA %#lx)\n",
    716 	    __func__, data.pv_pa, data.pv_pa + data.pv_size - 1, data.pv_va);
    717 #endif
    718 
    719 	add_pages(bmi, &data);
    720 
    721 #ifdef VERBOSE_INIT_ARM
    722 	printf("Listing Chunks\n");
    723 
    724 	pv_addr_t *lpv;
    725 	SLIST_FOREACH(lpv, &bmi->bmi_chunks, pv_list) {
    726 		printf("%s: pv %p: chunk VA %#lx..%#lx "
    727 		    "(PA %#lx, prot %d, cache %d)\n",
    728 		    __func__, lpv, lpv->pv_va, lpv->pv_va + lpv->pv_size - 1,
    729 		    lpv->pv_pa, lpv->pv_prot, lpv->pv_cache);
    730 	}
    731 	printf("\nMapping Chunks\n");
    732 #endif
    733 
    734 	pv_addr_t cur_pv;
    735 	pv_addr_t *pv = SLIST_FIRST(&bmi->bmi_chunks);
    736 	if (!mapallmem_p || pv->pv_pa == bmi->bmi_start) {
    737 		cur_pv = *pv;
    738 		pv = SLIST_NEXT(pv, pv_list);
    739 	} else {
    740 #if defined(ARM_MMU_EXTENDED) && defined(__HAVE_MM_MD_DIRECT_MAPPED_PHYS)
    741 		cur_pv.pv_va = pmap_directbase;
    742 #else
    743 		cur_pv.pv_va = KERNEL_BASE;
    744 #endif
    745 		cur_pv.pv_pa = bmi->bmi_start;
    746 		cur_pv.pv_size = pv->pv_pa - bmi->bmi_start;
    747 		cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    748 		cur_pv.pv_cache = PTE_CACHE;
    749 	}
    750 	while (pv != NULL) {
    751 		if (mapallmem_p) {
    752 			if (concat_pvaddr(&cur_pv, pv)) {
    753 				pv = SLIST_NEXT(pv, pv_list);
    754 				continue;
    755 			}
    756 			if (cur_pv.pv_pa + cur_pv.pv_size < pv->pv_pa) {
    757 				/*
    758 				 * See if we can extend the current pv to emcompass the
    759 				 * hole, and if so do it and retry the concatenation.
    760 				 */
    761 				if (cur_pv.pv_prot == (VM_PROT_READ|VM_PROT_WRITE)
    762 				    && cur_pv.pv_cache == PTE_CACHE) {
    763 					cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
    764 					continue;
    765 				}
    766 
    767 				/*
    768 				 * We couldn't so emit the current chunk and then
    769 				 */
    770 #ifdef VERBOSE_INIT_ARM
    771 				printf("%s: mapping chunk VA %#lx..%#lx "
    772 				    "(PA %#lx, prot %d, cache %d)\n",
    773 				    __func__,
    774 				    cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    775 				    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    776 #endif
    777 				pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    778 				    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    779 
    780 				/*
    781 				 * set the current chunk to the hole and try again.
    782 				 */
    783 				cur_pv.pv_pa += cur_pv.pv_size;
    784 				cur_pv.pv_va += cur_pv.pv_size;
    785 				cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
    786 				cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    787 				cur_pv.pv_cache = PTE_CACHE;
    788 				continue;
    789 			}
    790 		}
    791 
    792 		/*
    793 		 * The new pv didn't concatenate so emit the current one
    794 		 * and use the new pv as the current pv.
    795 		 */
    796 #ifdef VERBOSE_INIT_ARM
    797 		printf("%s: mapping chunk VA %#lx..%#lx "
    798 		    "(PA %#lx, prot %d, cache %d)\n",
    799 		    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    800 		    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    801 #endif
    802 		pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    803 		    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    804 		cur_pv = *pv;
    805 		pv = SLIST_NEXT(pv, pv_list);
    806 	}
    807 
    808 	/*
    809 	 * If we are mapping all of memory, let's map the rest of memory.
    810 	 */
    811 	if (mapallmem_p && cur_pv.pv_pa + cur_pv.pv_size < bmi->bmi_end) {
    812 		if (cur_pv.pv_prot == (VM_PROT_READ | VM_PROT_WRITE)
    813 		    && cur_pv.pv_cache == PTE_CACHE) {
    814 			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
    815 		} else {
    816 #ifdef VERBOSE_INIT_ARM
    817 			printf("%s: mapping chunk VA %#lx..%#lx "
    818 			    "(PA %#lx, prot %d, cache %d)\n",
    819 			    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    820 			    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    821 #endif
    822 			pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    823 			    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    824 			cur_pv.pv_pa += cur_pv.pv_size;
    825 			cur_pv.pv_va += cur_pv.pv_size;
    826 			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
    827 			cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    828 			cur_pv.pv_cache = PTE_CACHE;
    829 		}
    830 	}
    831 
    832 	/*
    833 	 * Now we map the final chunk.
    834 	 */
    835 #ifdef VERBOSE_INIT_ARM
    836 	printf("%s: mapping last chunk VA %#lx..%#lx (PA %#lx, prot %d, cache %d)\n",
    837 	    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    838 	    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    839 #endif
    840 	pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    841 	    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    842 
    843 	/*
    844 	 * Now we map the stuff that isn't directly after the kernel
    845 	 */
    846 
    847 	if (map_vectors_p) {
    848 		/* Map the vector page. */
    849 		pmap_map_entry(l1pt_va, systempage.pv_va, systempage.pv_pa,
    850 		    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
    851 	}
    852 
    853 	/* Map the Mini-Data cache clean area. */
    854 #if ARM_MMU_XSCALE == 1
    855 #if (ARM_NMMUS > 1)
    856 	if (xscale_use_minidata)
    857 #endif
    858 		xscale_setup_minidata(l1pt_va, minidataclean.pv_va,
    859 		    minidataclean.pv_pa);
    860 #endif
    861 
    862 	/*
    863 	 * Map integrated peripherals at same address in first level page
    864 	 * table so that we can continue to use console.
    865 	 */
    866 	if (devmap)
    867 		pmap_devmap_bootstrap(l1pt_va, devmap);
    868 
    869 #ifdef VERBOSE_INIT_ARM
    870 	/* Tell the user about where all the bits and pieces live. */
    871 	printf("%22s       Physical              Virtual        Num\n", " ");
    872 	printf("%22s Starting    Ending    Starting    Ending   Pages\n", " ");
    873 
    874 	static const char mem_fmt[] =
    875 	    "%20s: 0x%08lx 0x%08lx 0x%08lx 0x%08lx %u\n";
    876 	static const char mem_fmt_nov[] =
    877 	    "%20s: 0x%08lx 0x%08lx                       %zu\n";
    878 
    879 	printf(mem_fmt, "SDRAM", bmi->bmi_start, bmi->bmi_end - 1,
    880 	    KERN_PHYSTOV(bmi, bmi->bmi_start), KERN_PHYSTOV(bmi, bmi->bmi_end - 1),
    881 	    physmem);
    882 	printf(mem_fmt, "text section",
    883 	       text.pv_pa, text.pv_pa + text.pv_size - 1,
    884 	       text.pv_va, text.pv_va + text.pv_size - 1,
    885 	       (int)(text.pv_size / PAGE_SIZE));
    886 	printf(mem_fmt, "data section",
    887 	       KERN_VTOPHYS(bmi, __data_start), KERN_VTOPHYS(bmi, _edata),
    888 	       (vaddr_t)__data_start, (vaddr_t)_edata,
    889 	       (int)((round_page((vaddr_t)_edata)
    890 		      - trunc_page((vaddr_t)__data_start)) / PAGE_SIZE));
    891 	printf(mem_fmt, "bss section",
    892 	       KERN_VTOPHYS(bmi, __bss_start), KERN_VTOPHYS(bmi, __bss_end__),
    893 	       (vaddr_t)__bss_start, (vaddr_t)__bss_end__,
    894 	       (int)((round_page((vaddr_t)__bss_end__)
    895 		      - trunc_page((vaddr_t)__bss_start)) / PAGE_SIZE));
    896 	printf(mem_fmt, "L1 page directory",
    897 	    kernel_l1pt.pv_pa, kernel_l1pt.pv_pa + L1_TABLE_SIZE - 1,
    898 	    kernel_l1pt.pv_va, kernel_l1pt.pv_va + L1_TABLE_SIZE - 1,
    899 	    L1_TABLE_SIZE / PAGE_SIZE);
    900 	printf(mem_fmt, "ABT stack (CPU 0)",
    901 	    abtstack.pv_pa, abtstack.pv_pa + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
    902 	    abtstack.pv_va, abtstack.pv_va + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
    903 	    ABT_STACK_SIZE);
    904 	printf(mem_fmt, "FIQ stack (CPU 0)",
    905 	    fiqstack.pv_pa, fiqstack.pv_pa + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
    906 	    fiqstack.pv_va, fiqstack.pv_va + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
    907 	    FIQ_STACK_SIZE);
    908 	printf(mem_fmt, "IRQ stack (CPU 0)",
    909 	    irqstack.pv_pa, irqstack.pv_pa + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
    910 	    irqstack.pv_va, irqstack.pv_va + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
    911 	    IRQ_STACK_SIZE);
    912 	printf(mem_fmt, "UND stack (CPU 0)",
    913 	    undstack.pv_pa, undstack.pv_pa + (UND_STACK_SIZE * PAGE_SIZE) - 1,
    914 	    undstack.pv_va, undstack.pv_va + (UND_STACK_SIZE * PAGE_SIZE) - 1,
    915 	    UND_STACK_SIZE);
    916 	printf(mem_fmt, "IDLE stack (CPU 0)",
    917 	    idlestack.pv_pa, idlestack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
    918 	    idlestack.pv_va, idlestack.pv_va + (UPAGES * PAGE_SIZE) - 1,
    919 	    UPAGES);
    920 	printf(mem_fmt, "SVC stack",
    921 	    kernelstack.pv_pa, kernelstack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
    922 	    kernelstack.pv_va, kernelstack.pv_va + (UPAGES * PAGE_SIZE) - 1,
    923 	    UPAGES);
    924 	printf(mem_fmt, "Message Buffer",
    925 	    msgbuf.pv_pa, msgbuf.pv_pa + (msgbuf_pgs * PAGE_SIZE) - 1,
    926 	    msgbuf.pv_va, msgbuf.pv_va + (msgbuf_pgs * PAGE_SIZE) - 1,
    927 	    (int)msgbuf_pgs);
    928 	if (map_vectors_p) {
    929 		printf(mem_fmt, "Exception Vectors",
    930 		    systempage.pv_pa, systempage.pv_pa + PAGE_SIZE - 1,
    931 		    systempage.pv_va, systempage.pv_va + PAGE_SIZE - 1,
    932 		    1);
    933 	}
    934 	for (size_t i = 0; i < bmi->bmi_nfreeblocks; i++) {
    935 		pv = &bmi->bmi_freeblocks[i];
    936 
    937 		printf(mem_fmt_nov, "Free Memory",
    938 		    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    939 		    pv->pv_size / PAGE_SIZE);
    940 	}
    941 #endif
    942 	/*
    943 	 * Now we have the real page tables in place so we can switch to them.
    944 	 * Once this is done we will be running with the REAL kernel page
    945 	 * tables.
    946 	 */
    947 
    948 #if defined(VERBOSE_INIT_ARM)
    949 	printf("TTBR0=%#x", armreg_ttbr_read());
    950 #ifdef _ARM_ARCH_6
    951 	printf(" TTBR1=%#x TTBCR=%#x CONTEXTIDR=%#x",
    952 	    armreg_ttbr1_read(), armreg_ttbcr_read(),
    953 	    armreg_contextidr_read());
    954 #endif
    955 	printf("\n");
    956 #endif
    957 
    958 	/* Switch tables */
    959 #ifdef VERBOSE_INIT_ARM
    960 	printf("switching to new L1 page table @%#lx...", l1pt_pa);
    961 #endif
    962 
    963 #ifdef ARM_MMU_EXTENDED
    964 	cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2))
    965 	    | (DOMAIN_CLIENT << (PMAP_DOMAIN_USER*2)));
    966 #else
    967 	cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT);
    968 #endif
    969 	cpu_idcache_wbinv_all();
    970 #ifdef VERBOSE_INIT_ARM
    971 	printf(" ttb");
    972 #endif
    973 #ifdef ARM_MMU_EXTENDED
    974 	/*
    975 	 * TTBCR should have been initialized by the MD start code.
    976 	 */
    977 	KASSERT((armreg_contextidr_read() & 0xff) == 0);
    978 	KASSERT(armreg_ttbcr_read() == __SHIFTIN(1, TTBCR_S_N));
    979 	/*
    980 	 * Disable lookups via TTBR0 until there is an activated pmap.
    981 	 */
    982 	armreg_ttbcr_write(armreg_ttbcr_read() | TTBCR_S_PD0);
    983 	cpu_setttb(l1pt_pa, KERNEL_PID);
    984 	arm_isb();
    985 #else
    986 	cpu_setttb(l1pt_pa, true);
    987 	cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2));
    988 #endif
    989 	cpu_tlb_flushID();
    990 
    991 #ifdef VERBOSE_INIT_ARM
    992 #ifdef ARM_MMU_EXTENDED
    993 	printf(" (TTBCR=%#x TTBR0=%#x TTBR1=%#x)",
    994 	    armreg_ttbcr_read(), armreg_ttbr_read(), armreg_ttbr1_read());
    995 #else
    996 	printf(" (TTBR0=%#x)", armreg_ttbr_read());
    997 #endif
    998 #endif
    999 
   1000 #ifdef MULTIPROCESSOR
   1001 	/*
   1002 	 * Kick the secondaries to load the TTB.  After which they'll go
   1003 	 * back to sleep to wait for the final kick so they will hatch.
   1004 	 */
   1005 #ifdef VERBOSE_INIT_ARM
   1006 	printf(" hatchlings");
   1007 #endif
   1008 	cpu_boot_secondary_processors();
   1009 #endif
   1010 
   1011 #ifdef VERBOSE_INIT_ARM
   1012 	printf(" OK\n");
   1013 #endif
   1014 }
   1015