Home | History | Annotate | Line # | Download | only in arm32
arm32_kvminit.c revision 1.51
      1 /*	$NetBSD: arm32_kvminit.c,v 1.51 2019/02/06 13:22:54 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_arm_debug.h"
    125 #include "opt_arm_start.h"
    126 #include "opt_fdt.h"
    127 #include "opt_multiprocessor.h"
    128 
    129 #include <sys/cdefs.h>
    130 __KERNEL_RCSID(0, "$NetBSD: arm32_kvminit.c,v 1.51 2019/02/06 13:22:54 skrll Exp $");
    131 
    132 #include <sys/param.h>
    133 #include <sys/device.h>
    134 #include <sys/kernel.h>
    135 #include <sys/reboot.h>
    136 #include <sys/bus.h>
    137 
    138 #include <dev/cons.h>
    139 
    140 #include <uvm/uvm_extern.h>
    141 
    142 #include <arm/locore.h>
    143 #include <arm/db_machdep.h>
    144 #include <arm/undefined.h>
    145 #include <arm/bootconfig.h>
    146 #include <arm/arm32/machdep.h>
    147 
    148 #if defined(FDT)
    149 #include <arch/evbarm/fdt/platform.h>
    150 #include <arm/fdt/arm_fdtvar.h>
    151 #endif
    152 
    153 #ifdef MULTIPROCESSOR
    154 #ifndef __HAVE_CPU_UAREA_ALLOC_IDLELWP
    155 #error __HAVE_CPU_UAREA_ALLOC_IDLELWP required to not waste pages for idlestack
    156 #endif
    157 #endif
    158 
    159 #ifdef VERBOSE_INIT_ARM
    160 #define VPRINTF(...)	printf(__VA_ARGS__)
    161 #else
    162 #define VPRINTF(...)	__nothing
    163 #endif
    164 
    165 struct bootmem_info bootmem_info;
    166 
    167 extern void *msgbufaddr;
    168 paddr_t msgbufphys;
    169 paddr_t physical_start;
    170 paddr_t physical_end;
    171 
    172 extern char etext[];
    173 extern char __data_start[], _edata[];
    174 extern char __bss_start[], __bss_end__[];
    175 extern char _end[];
    176 
    177 /* Page tables for mapping kernel VM */
    178 #define KERNEL_L2PT_VMDATA_NUM	8	/* start with 32MB of KVM */
    179 
    180 u_long kern_vtopdiff __attribute__((__section__(".data")));
    181 
    182 void
    183 arm32_bootmem_init(paddr_t memstart, psize_t memsize, vsize_t kernelstart)
    184 {
    185 	struct bootmem_info * const bmi = &bootmem_info;
    186 	pv_addr_t *pv = bmi->bmi_freeblocks;
    187 
    188 	/*
    189 	 * FDT/generic start fills in kern_vtopdiff early
    190 	 */
    191 #if defined(__HAVE_GENERIC_START)
    192 	extern char KERNEL_BASE_virt[];
    193 	extern char ARM_BOOTSTRAP_LxPT[];
    194 
    195 	VPRINTF("%s: kern_vtopdiff=%#lx\n", __func__, kern_vtopdiff);
    196 
    197 	vaddr_t kstartva = trunc_page((vaddr_t)KERNEL_BASE_virt);
    198 	vaddr_t kendva = round_page((vaddr_t)ARM_BOOTSTRAP_LxPT + L1_TABLE_SIZE);
    199 
    200 	kernelstart = KERN_VTOPHYS(kstartva);
    201 
    202 	VPRINTF("%s: kstartva=%#lx, kernelstart=%#lx\n", __func__, kstartva, kernelstart);
    203 #else
    204 	vaddr_t kendva = round_page((vaddr_t)_end);
    205 
    206 #if defined(KERNEL_BASE_VOFFSET)
    207 	kern_vtopdiff = KERNEL_BASE_VOFFSET;
    208 #else
    209 	KASSERT(memstart == kernelstart);
    210 	kern_vtopdiff = KERNEL_BASE + memstart;
    211 #endif
    212 #endif
    213 	paddr_t kernelend = KERN_VTOPHYS(kendva);
    214 
    215 	VPRINTF("%s: memstart=%#lx, memsize=%#lx\n", __func__,
    216 	    memstart, memsize);
    217 	VPRINTF("%s: kernelstart=%#lx, kernelend=%#lx\n", __func__,
    218 	    kernelstart, kernelend);
    219 
    220 	physical_start = bmi->bmi_start = memstart;
    221 	physical_end = bmi->bmi_end = memstart + memsize;
    222 #ifndef ARM_HAS_LPAE
    223 	if (physical_end == 0) {
    224 		physical_end = -PAGE_SIZE;
    225 		memsize -= PAGE_SIZE;
    226 		bmi->bmi_end -= PAGE_SIZE;
    227 		VPRINTF("%s: memsize shrunk by a page to avoid ending at 4GB\n",
    228 		    __func__);
    229 	}
    230 #endif
    231 	physmem = memsize / PAGE_SIZE;
    232 
    233 	/*
    234 	 * Let's record where the kernel lives.
    235 	 */
    236 
    237 	bmi->bmi_kernelstart = kernelstart;
    238 	bmi->bmi_kernelend = kernelend;
    239 
    240 #if defined(FDT)
    241 	fdt_add_reserved_memory_range(bmi->bmi_kernelstart,
    242 	    bmi->bmi_kernelend - bmi->bmi_kernelstart);
    243 #endif
    244 
    245 	VPRINTF("%s: kernel phys start %#lx end %#lx\n", __func__, kernelstart,
    246 	    kernelend);
    247 
    248 #if 0
    249 	// XXX Makes RPI abort
    250 	KASSERT((kernelstart & (L2_S_SEGSIZE - 1)) == 0);
    251 #endif
    252 	/*
    253 	 * Now the rest of the free memory must be after the kernel.
    254 	 */
    255 	pv->pv_pa = bmi->bmi_kernelend;
    256 	pv->pv_va = KERN_PHYSTOV(pv->pv_pa);
    257 	pv->pv_size = bmi->bmi_end - bmi->bmi_kernelend;
    258 	bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
    259 	VPRINTF("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
    260 	    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
    261 	    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
    262 	pv++;
    263 
    264 	/*
    265 	 * Add a free block for any memory before the kernel.
    266 	 */
    267 	if (bmi->bmi_start < bmi->bmi_kernelstart) {
    268 		pv->pv_pa = bmi->bmi_start;
    269 		pv->pv_va = KERN_PHYSTOV(pv->pv_pa);
    270 		pv->pv_size = bmi->bmi_kernelstart - pv->pv_pa;
    271 		bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
    272 		VPRINTF("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
    273 		    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
    274 		    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
    275 		pv++;
    276 	}
    277 
    278 	bmi->bmi_nfreeblocks = pv - bmi->bmi_freeblocks;
    279 
    280 	SLIST_INIT(&bmi->bmi_freechunks);
    281 	SLIST_INIT(&bmi->bmi_chunks);
    282 }
    283 
    284 static bool
    285 concat_pvaddr(pv_addr_t *acc_pv, pv_addr_t *pv)
    286 {
    287 	if (acc_pv->pv_pa + acc_pv->pv_size == pv->pv_pa
    288 	    && acc_pv->pv_va + acc_pv->pv_size == pv->pv_va
    289 	    && acc_pv->pv_prot == pv->pv_prot
    290 	    && acc_pv->pv_cache == pv->pv_cache) {
    291 #if 0
    292 		VPRINTF("%s: appending pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    293 		    __func__, pv, pv->pv_pa, pv->pv_pa + pv->pv_size,
    294 		    acc_pv->pv_pa, acc_pv->pv_pa + acc_pv->pv_size);
    295 #endif
    296 		acc_pv->pv_size += pv->pv_size;
    297 		return true;
    298 	}
    299 
    300 	return false;
    301 }
    302 
    303 static void
    304 add_pages(struct bootmem_info *bmi, pv_addr_t *pv)
    305 {
    306 	pv_addr_t **pvp = &SLIST_FIRST(&bmi->bmi_chunks);
    307 	while ((*pvp) != NULL && (*pvp)->pv_va <= pv->pv_va) {
    308 		pv_addr_t * const pv0 = (*pvp);
    309 		KASSERT(SLIST_NEXT(pv0, pv_list) == NULL || pv0->pv_pa < SLIST_NEXT(pv0, pv_list)->pv_pa);
    310 		if (concat_pvaddr(pv0, pv)) {
    311 			VPRINTF("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    312 			    __func__, "appending", pv,
    313 			    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    314 			    pv0->pv_pa, pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
    315 			pv = SLIST_NEXT(pv0, pv_list);
    316 			if (pv != NULL && concat_pvaddr(pv0, pv)) {
    317 				VPRINTF("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    318 				    __func__, "merging", pv,
    319 				    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    320 				    pv0->pv_pa,
    321 				    pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
    322 				SLIST_REMOVE_AFTER(pv0, pv_list);
    323 				SLIST_INSERT_HEAD(&bmi->bmi_freechunks, pv, pv_list);
    324 			}
    325 			return;
    326 		}
    327 		KASSERT(pv->pv_va != (*pvp)->pv_va);
    328 		pvp = &SLIST_NEXT(*pvp, pv_list);
    329 	}
    330 	KASSERT((*pvp) == NULL || pv->pv_va < (*pvp)->pv_va);
    331 	pv_addr_t * const new_pv = SLIST_FIRST(&bmi->bmi_freechunks);
    332 	KASSERT(new_pv != NULL);
    333 	SLIST_REMOVE_HEAD(&bmi->bmi_freechunks, pv_list);
    334 	*new_pv = *pv;
    335 	SLIST_NEXT(new_pv, pv_list) = *pvp;
    336 	(*pvp) = new_pv;
    337 
    338 	VPRINTF("%s: adding pv %p (pa %#lx, va %#lx, %lu pages) ",
    339 	    __func__, new_pv, new_pv->pv_pa, new_pv->pv_va,
    340 	    new_pv->pv_size / PAGE_SIZE);
    341 	if (SLIST_NEXT(new_pv, pv_list)) {
    342 		VPRINTF("before pa %#lx\n", SLIST_NEXT(new_pv, pv_list)->pv_pa);
    343 	} else {
    344 		VPRINTF("at tail\n");
    345 	}
    346 }
    347 
    348 static void
    349 valloc_pages(struct bootmem_info *bmi, pv_addr_t *pv, size_t npages,
    350 	int prot, int cache, bool zero_p)
    351 {
    352 	size_t nbytes = npages * PAGE_SIZE;
    353 	pv_addr_t *free_pv = bmi->bmi_freeblocks;
    354 	size_t free_idx = 0;
    355 	static bool l1pt_found;
    356 
    357 	KASSERT(npages > 0);
    358 
    359 	/*
    360 	 * If we haven't allocated the kernel L1 page table and we are aligned
    361 	 * at a L1 table boundary, alloc the memory for it.
    362 	 */
    363 	if (!l1pt_found
    364 	    && (free_pv->pv_pa & (L1_TABLE_SIZE - 1)) == 0
    365 	    && free_pv->pv_size >= L1_TABLE_SIZE) {
    366 		l1pt_found = true;
    367 		VPRINTF(" l1pt");
    368 
    369 		valloc_pages(bmi, &kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE,
    370 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    371 		add_pages(bmi, &kernel_l1pt);
    372 	}
    373 
    374 	while (nbytes > free_pv->pv_size) {
    375 		free_pv++;
    376 		free_idx++;
    377 		if (free_idx == bmi->bmi_nfreeblocks) {
    378 			panic("%s: could not allocate %zu bytes",
    379 			    __func__, nbytes);
    380 		}
    381 	}
    382 
    383 	/*
    384 	 * As we allocate the memory, make sure that we don't walk over
    385 	 * our current first level translation table.
    386 	 */
    387 	KASSERT((armreg_ttbr_read() & ~(L1_TABLE_SIZE - 1)) != free_pv->pv_pa);
    388 
    389 #if defined(FDT)
    390 	fdt_add_reserved_memory_range(free_pv->pv_pa, nbytes);
    391 #endif
    392 	pv->pv_pa = free_pv->pv_pa;
    393 	pv->pv_va = free_pv->pv_va;
    394 	pv->pv_size = nbytes;
    395 	pv->pv_prot = prot;
    396 	pv->pv_cache = cache;
    397 
    398 	/*
    399 	 * If PTE_PAGETABLE uses the same cache modes as PTE_CACHE
    400 	 * just use PTE_CACHE.
    401 	 */
    402 	if (cache == PTE_PAGETABLE
    403 	    && pte_l1_s_cache_mode == pte_l1_s_cache_mode_pt
    404 	    && pte_l2_l_cache_mode == pte_l2_l_cache_mode_pt
    405 	    && pte_l2_s_cache_mode == pte_l2_s_cache_mode_pt)
    406 		pv->pv_cache = PTE_CACHE;
    407 
    408 	free_pv->pv_pa += nbytes;
    409 	free_pv->pv_va += nbytes;
    410 	free_pv->pv_size -= nbytes;
    411 	if (free_pv->pv_size == 0) {
    412 		--bmi->bmi_nfreeblocks;
    413 		for (; free_idx < bmi->bmi_nfreeblocks; free_idx++) {
    414 			free_pv[0] = free_pv[1];
    415 		}
    416 	}
    417 
    418 	bmi->bmi_freepages -= npages;
    419 
    420 	if (zero_p)
    421 		memset((void *)pv->pv_va, 0, nbytes);
    422 }
    423 
    424 void
    425 arm32_kernel_vm_init(vaddr_t kernel_vm_base, vaddr_t vectors, vaddr_t iovbase,
    426 	const struct pmap_devmap *devmap, bool mapallmem_p)
    427 {
    428 	struct bootmem_info * const bmi = &bootmem_info;
    429 #ifdef MULTIPROCESSOR
    430 	const size_t cpu_num = arm_cpu_max;
    431 #else
    432 	const size_t cpu_num = 1;
    433 #endif
    434 
    435 #ifdef ARM_HAS_VBAR
    436 	const bool map_vectors_p = false;
    437 #elif defined(CPU_ARMV7) || defined(CPU_ARM11)
    438 	const bool map_vectors_p = vectors == ARM_VECTORS_HIGH
    439 	    || (armreg_pfr1_read() & ARM_PFR1_SEC_MASK) == 0;
    440 #else
    441 	const bool map_vectors_p = true;
    442 #endif
    443 
    444 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS
    445 	KASSERT(mapallmem_p);
    446 #ifdef ARM_MMU_EXTENDED
    447 	/*
    448 	 * The direct map VA space ends at the start of the kernel VM space.
    449 	 */
    450 	pmap_directlimit = kernel_vm_base;
    451 #else
    452 	KASSERT(kernel_vm_base - KERNEL_BASE >= physical_end - physical_start);
    453 #endif /* ARM_MMU_EXTENDED */
    454 #endif /* __HAVE_MM_MD_DIRECT_MAPPED_PHYS */
    455 
    456 	/*
    457 	 * Calculate the number of L2 pages needed for mapping the
    458 	 * kernel + data + stuff.  Assume 2 L2 pages for kernel, 1 for vectors,
    459 	 * and 1 for IO
    460 	 */
    461 	size_t kernel_size = bmi->bmi_kernelend;
    462 	kernel_size -= (bmi->bmi_kernelstart & -L2_S_SEGSIZE);
    463 	kernel_size += L1_TABLE_SIZE_REAL;
    464 	kernel_size += PAGE_SIZE * KERNEL_L2PT_VMDATA_NUM;
    465 	if (map_vectors_p) {
    466 		kernel_size += PAGE_SIZE;	/* L2PT for VECTORS */
    467 	}
    468 	if (iovbase) {
    469 		kernel_size += PAGE_SIZE;	/* L2PT for IO */
    470 	}
    471 	kernel_size +=
    472 	    cpu_num * (ABT_STACK_SIZE + FIQ_STACK_SIZE + IRQ_STACK_SIZE
    473 	    + UND_STACK_SIZE + UPAGES) * PAGE_SIZE;
    474 	kernel_size += round_page(MSGBUFSIZE);
    475 	kernel_size += 0x10000;	/* slop */
    476 	if (!mapallmem_p) {
    477 		kernel_size += PAGE_SIZE
    478 		    * ((kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE);
    479 	}
    480 	kernel_size = round_page(kernel_size);
    481 
    482 	/*
    483 	 * Now we know how many L2 pages it will take.
    484 	 */
    485 	const size_t KERNEL_L2PT_KERNEL_NUM =
    486 	    round_page(kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE;
    487 
    488 	VPRINTF("%s: %zu L2 pages are needed to map %#zx kernel bytes\n",
    489 	    __func__, KERNEL_L2PT_KERNEL_NUM, kernel_size);
    490 
    491 	KASSERT(KERNEL_L2PT_KERNEL_NUM + KERNEL_L2PT_VMDATA_NUM < __arraycount(bmi->bmi_l2pts));
    492 	pv_addr_t * const kernel_l2pt = bmi->bmi_l2pts;
    493 	pv_addr_t * const vmdata_l2pt = kernel_l2pt + KERNEL_L2PT_KERNEL_NUM;
    494 	pv_addr_t msgbuf;
    495 	pv_addr_t text;
    496 	pv_addr_t data;
    497 	pv_addr_t chunks[KERNEL_L2PT_KERNEL_NUM+KERNEL_L2PT_VMDATA_NUM+11];
    498 #if ARM_MMU_XSCALE == 1
    499 	pv_addr_t minidataclean;
    500 #endif
    501 
    502 	/*
    503 	 * We need to allocate some fixed page tables to get the kernel going.
    504 	 *
    505 	 * We are going to allocate our bootstrap pages from the beginning of
    506 	 * the free space that we just calculated.  We allocate one page
    507 	 * directory and a number of page tables and store the physical
    508 	 * addresses in the bmi_l2pts array in bootmem_info.
    509 	 *
    510 	 * The kernel page directory must be on a 16K boundary.  The page
    511 	 * tables must be on 4K boundaries.  What we do is allocate the
    512 	 * page directory on the first 16K boundary that we encounter, and
    513 	 * the page tables on 4K boundaries otherwise.  Since we allocate
    514 	 * at least 3 L2 page tables, we are guaranteed to encounter at
    515 	 * least one 16K aligned region.
    516 	 */
    517 
    518 	VPRINTF("%s: allocating page tables for", __func__);
    519 	for (size_t i = 0; i < __arraycount(chunks); i++) {
    520 		SLIST_INSERT_HEAD(&bmi->bmi_freechunks, &chunks[i], pv_list);
    521 	}
    522 
    523 	kernel_l1pt.pv_pa = 0;
    524 	kernel_l1pt.pv_va = 0;
    525 
    526 	/*
    527 	 * Allocate the L2 pages, but if we get to a page that is aligned for
    528 	 * an L1 page table, we will allocate the pages for it first and then
    529 	 * allocate the L2 page.
    530 	 */
    531 
    532 	if (map_vectors_p) {
    533 		/*
    534 		 * First allocate L2 page for the vectors.
    535 		 */
    536 		VPRINTF(" vector");
    537 		valloc_pages(bmi, &bmi->bmi_vector_l2pt, 1,
    538 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    539 		add_pages(bmi, &bmi->bmi_vector_l2pt);
    540 	}
    541 
    542 	/*
    543 	 * Now allocate L2 pages for the kernel
    544 	 */
    545 	VPRINTF(" kernel");
    546 	for (size_t idx = 0; idx < KERNEL_L2PT_KERNEL_NUM; ++idx) {
    547 		valloc_pages(bmi, &kernel_l2pt[idx], 1,
    548 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    549 		add_pages(bmi, &kernel_l2pt[idx]);
    550 	}
    551 
    552 	/*
    553 	 * Now allocate L2 pages for the initial kernel VA space.
    554 	 */
    555 	VPRINTF(" vm");
    556 	for (size_t idx = 0; idx < KERNEL_L2PT_VMDATA_NUM; ++idx) {
    557 		valloc_pages(bmi, &vmdata_l2pt[idx], 1,
    558 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    559 		add_pages(bmi, &vmdata_l2pt[idx]);
    560 	}
    561 
    562 	/*
    563 	 * If someone wanted a L2 page for I/O, allocate it now.
    564 	 */
    565 	if (iovbase) {
    566 		VPRINTF(" io");
    567 		valloc_pages(bmi, &bmi->bmi_io_l2pt, 1,
    568 		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE, true);
    569 		add_pages(bmi, &bmi->bmi_io_l2pt);
    570 	}
    571 
    572 	VPRINTF("%s: allocating stacks\n", __func__);
    573 
    574 	/* Allocate stacks for all modes and CPUs */
    575 	valloc_pages(bmi, &abtstack, ABT_STACK_SIZE * cpu_num,
    576 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    577 	add_pages(bmi, &abtstack);
    578 	valloc_pages(bmi, &fiqstack, FIQ_STACK_SIZE * cpu_num,
    579 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    580 	add_pages(bmi, &fiqstack);
    581 	valloc_pages(bmi, &irqstack, IRQ_STACK_SIZE * cpu_num,
    582 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    583 	add_pages(bmi, &irqstack);
    584 	valloc_pages(bmi, &undstack, UND_STACK_SIZE * cpu_num,
    585 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    586 	add_pages(bmi, &undstack);
    587 	valloc_pages(bmi, &idlestack, UPAGES * cpu_num,		/* SVC32 */
    588 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    589 	add_pages(bmi, &idlestack);
    590 	valloc_pages(bmi, &kernelstack, UPAGES,			/* SVC32 */
    591 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, true);
    592 	add_pages(bmi, &kernelstack);
    593 
    594 	/* Allocate the message buffer from the end of memory. */
    595 	const size_t msgbuf_pgs = round_page(MSGBUFSIZE) / PAGE_SIZE;
    596 	valloc_pages(bmi, &msgbuf, msgbuf_pgs,
    597 	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE, false);
    598 	add_pages(bmi, &msgbuf);
    599 	msgbufphys = msgbuf.pv_pa;
    600 	msgbufaddr = (void *)msgbuf.pv_va;
    601 
    602 	if (map_vectors_p) {
    603 		/*
    604 		 * Allocate a page for the system vector page.
    605 		 * This page will just contain the system vectors and can be
    606 		 * shared by all processes.
    607 		 */
    608 		VPRINTF(" vector");
    609 
    610 		valloc_pages(bmi, &systempage, 1, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE,
    611 		    PTE_CACHE, true);
    612 	}
    613 	systempage.pv_va = vectors;
    614 
    615 	/*
    616 	 * If the caller needed a few extra pages for some reason, allocate
    617 	 * them now.
    618 	 */
    619 #if ARM_MMU_XSCALE == 1
    620 #if (ARM_NMMUS > 1)
    621 	if (xscale_use_minidata)
    622 #endif
    623 		valloc_pages(bmi, &minidataclean, 1,
    624 		    VM_PROT_READ|VM_PROT_WRITE, 0, true);
    625 #endif
    626 
    627 	/*
    628 	 * Ok we have allocated physical pages for the primary kernel
    629 	 * page tables and stacks.  Let's just confirm that.
    630 	 */
    631 	if (kernel_l1pt.pv_va == 0
    632 	    && (!kernel_l1pt.pv_pa || (kernel_l1pt.pv_pa & (L1_TABLE_SIZE - 1)) != 0))
    633 		panic("%s: Failed to allocate or align the kernel "
    634 		    "page directory", __func__);
    635 
    636 	VPRINTF("Creating L1 page table at 0x%08lx/0x%08lx\n",
    637 	    kernel_l1pt.pv_va, kernel_l1pt.pv_pa);
    638 
    639 	/*
    640 	 * Now we start construction of the L1 page table
    641 	 * We start by mapping the L2 page tables into the L1.
    642 	 * This means that we can replace L1 mappings later on if necessary
    643 	 */
    644 	vaddr_t l1pt_va = kernel_l1pt.pv_va;
    645 	paddr_t l1pt_pa = kernel_l1pt.pv_pa;
    646 
    647 	if (map_vectors_p) {
    648 		/* Map the L2 pages tables in the L1 page table */
    649 		pmap_link_l2pt(l1pt_va, systempage.pv_va & -L2_S_SEGSIZE,
    650 		    &bmi->bmi_vector_l2pt);
    651 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) "
    652 		    "for VA %#lx\n (vectors)",
    653 		    __func__, bmi->bmi_vector_l2pt.pv_va,
    654 		    bmi->bmi_vector_l2pt.pv_pa, systempage.pv_va);
    655 	}
    656 
    657 	/*
    658 	 * This enforces a alignment requirement of L2_S_SEGSIZE for kernel
    659 	 * start PA
    660 	 */
    661 	const vaddr_t kernel_base =
    662 	    KERN_PHYSTOV(bmi->bmi_kernelstart & -L2_S_SEGSIZE);
    663 
    664 	VPRINTF("%s: kernel_base %lx KERNEL_L2PT_KERNEL_NUM %zu\n", __func__,
    665 	    kernel_base, KERNEL_L2PT_KERNEL_NUM);
    666 
    667 	for (size_t idx = 0; idx < KERNEL_L2PT_KERNEL_NUM; idx++) {
    668 		pmap_link_l2pt(l1pt_va, kernel_base + idx * L2_S_SEGSIZE,
    669 		    &kernel_l2pt[idx]);
    670 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (kernel)\n",
    671 		    __func__, kernel_l2pt[idx].pv_va,
    672 		    kernel_l2pt[idx].pv_pa, kernel_base + idx * L2_S_SEGSIZE);
    673 	}
    674 
    675 	VPRINTF("%s: kernel_vm_base %lx KERNEL_L2PT_VMDATA_NUM %d\n", __func__,
    676 	    kernel_vm_base, KERNEL_L2PT_VMDATA_NUM);
    677 
    678 	for (size_t idx = 0; idx < KERNEL_L2PT_VMDATA_NUM; idx++) {
    679 		pmap_link_l2pt(l1pt_va, kernel_vm_base + idx * L2_S_SEGSIZE,
    680 		    &vmdata_l2pt[idx]);
    681 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (vm)\n",
    682 		    __func__, vmdata_l2pt[idx].pv_va, vmdata_l2pt[idx].pv_pa,
    683 		    kernel_vm_base + idx * L2_S_SEGSIZE);
    684 	}
    685 	if (iovbase) {
    686 		pmap_link_l2pt(l1pt_va, iovbase & -L2_S_SEGSIZE, &bmi->bmi_io_l2pt);
    687 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (io)\n",
    688 		    __func__, bmi->bmi_io_l2pt.pv_va, bmi->bmi_io_l2pt.pv_pa,
    689 		    iovbase & -L2_S_SEGSIZE);
    690 	}
    691 
    692 	/* update the top of the kernel VM */
    693 	pmap_curmaxkvaddr =
    694 	    kernel_vm_base + (KERNEL_L2PT_VMDATA_NUM * L2_S_SEGSIZE);
    695 
    696 	// This could be done earlier and then the kernel data and pages
    697 	// allocated above would get merged (concatentated)
    698 
    699 	VPRINTF("Mapping kernel\n");
    700 
    701 	extern char etext[];
    702 	size_t totalsize = bmi->bmi_kernelend - bmi->bmi_kernelstart;
    703 	size_t textsize = KERN_VTOPHYS((uintptr_t)etext) - bmi->bmi_kernelstart;
    704 
    705 	textsize = (textsize + PGOFSET) & ~PGOFSET;
    706 
    707 	/* start at offset of kernel in RAM */
    708 
    709 	text.pv_pa = bmi->bmi_kernelstart;
    710 	text.pv_va = KERN_PHYSTOV(bmi->bmi_kernelstart);
    711 	text.pv_size = textsize;
    712 	text.pv_prot = VM_PROT_READ | VM_PROT_EXECUTE;
    713 	text.pv_cache = PTE_CACHE;
    714 
    715 	VPRINTF("%s: adding chunk for kernel text %#lx..%#lx (VA %#lx)\n",
    716 	    __func__, text.pv_pa, text.pv_pa + text.pv_size - 1, text.pv_va);
    717 
    718 	add_pages(bmi, &text);
    719 
    720 	data.pv_pa = text.pv_pa + textsize;
    721 	data.pv_va = text.pv_va + textsize;
    722 	data.pv_size = totalsize - textsize;
    723 	data.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    724 	data.pv_cache = PTE_CACHE;
    725 
    726 	VPRINTF("%s: adding chunk for kernel data/bss %#lx..%#lx (VA %#lx)\n",
    727 	    __func__, data.pv_pa, data.pv_pa + data.pv_size - 1, data.pv_va);
    728 
    729 	add_pages(bmi, &data);
    730 
    731 	VPRINTF("Listing Chunks\n");
    732 
    733 	pv_addr_t *lpv;
    734 	SLIST_FOREACH(lpv, &bmi->bmi_chunks, pv_list) {
    735 		VPRINTF("%s: pv %p: chunk VA %#lx..%#lx "
    736 		    "(PA %#lx, prot %d, cache %d)\n",
    737 		    __func__, lpv, lpv->pv_va, lpv->pv_va + lpv->pv_size - 1,
    738 		    lpv->pv_pa, lpv->pv_prot, lpv->pv_cache);
    739 	}
    740 	VPRINTF("\nMapping Chunks\n");
    741 
    742 	pv_addr_t cur_pv;
    743 	pv_addr_t *pv = SLIST_FIRST(&bmi->bmi_chunks);
    744 	if (!mapallmem_p || pv->pv_pa == bmi->bmi_start) {
    745 		cur_pv = *pv;
    746 		KASSERTMSG(cur_pv.pv_va >= KERNEL_BASE, "%#lx", cur_pv.pv_va);
    747 		pv = SLIST_NEXT(pv, pv_list);
    748 	} else {
    749 		cur_pv.pv_va = KERNEL_BASE;
    750 		cur_pv.pv_pa = KERN_VTOPHYS(cur_pv.pv_va);
    751 		cur_pv.pv_size = pv->pv_pa - cur_pv.pv_pa;
    752 		cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    753 		cur_pv.pv_cache = PTE_CACHE;
    754 	}
    755 	while (pv != NULL) {
    756 		if (mapallmem_p) {
    757 			if (concat_pvaddr(&cur_pv, pv)) {
    758 				pv = SLIST_NEXT(pv, pv_list);
    759 				continue;
    760 			}
    761 			if (cur_pv.pv_pa + cur_pv.pv_size < pv->pv_pa) {
    762 				/*
    763 				 * See if we can extend the current pv to emcompass the
    764 				 * hole, and if so do it and retry the concatenation.
    765 				 */
    766 				if (cur_pv.pv_prot == (VM_PROT_READ|VM_PROT_WRITE)
    767 				    && cur_pv.pv_cache == PTE_CACHE) {
    768 					cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
    769 					continue;
    770 				}
    771 
    772 				/*
    773 				 * We couldn't so emit the current chunk and then
    774 				 */
    775 				VPRINTF("%s: mapping chunk VA %#lx..%#lx "
    776 				    "(PA %#lx, prot %d, cache %d)\n",
    777 				    __func__,
    778 				    cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    779 				    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    780 				pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    781 				    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    782 
    783 				/*
    784 				 * set the current chunk to the hole and try again.
    785 				 */
    786 				cur_pv.pv_pa += cur_pv.pv_size;
    787 				cur_pv.pv_va += cur_pv.pv_size;
    788 				cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
    789 				cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    790 				cur_pv.pv_cache = PTE_CACHE;
    791 				continue;
    792 			}
    793 		}
    794 
    795 		/*
    796 		 * The new pv didn't concatenate so emit the current one
    797 		 * and use the new pv as the current pv.
    798 		 */
    799 		VPRINTF("%s: mapping chunk VA %#lx..%#lx "
    800 		    "(PA %#lx, prot %d, cache %d)\n",
    801 		    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    802 		    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    803 		pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    804 		    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    805 		cur_pv = *pv;
    806 		pv = SLIST_NEXT(pv, pv_list);
    807 	}
    808 
    809 	/*
    810 	 * If we are mapping all of memory, let's map the rest of memory.
    811 	 */
    812 	if (mapallmem_p && cur_pv.pv_pa + cur_pv.pv_size < bmi->bmi_end) {
    813 		if (cur_pv.pv_prot == (VM_PROT_READ | VM_PROT_WRITE)
    814 		    && cur_pv.pv_cache == PTE_CACHE) {
    815 			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
    816 		} else {
    817 			KASSERTMSG(cur_pv.pv_va + cur_pv.pv_size <= kernel_vm_base,
    818 			    "%#lx >= %#lx", cur_pv.pv_va + cur_pv.pv_size,
    819 			    kernel_vm_base);
    820 			VPRINTF("%s: mapping chunk VA %#lx..%#lx "
    821 			    "(PA %#lx, prot %d, cache %d)\n",
    822 			    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    823 			    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    824 			pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    825 			    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    826 			cur_pv.pv_pa += cur_pv.pv_size;
    827 			cur_pv.pv_va += cur_pv.pv_size;
    828 			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
    829 			cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    830 			cur_pv.pv_cache = PTE_CACHE;
    831 		}
    832 	}
    833 
    834 	/*
    835 	 * The amount we can direct map is limited by the start of the
    836 	 * virtual part of the kernel address space.  Don't overrun
    837 	 * into it.
    838 	 */
    839 	if (mapallmem_p && cur_pv.pv_va + cur_pv.pv_size > kernel_vm_base) {
    840 		cur_pv.pv_size = kernel_vm_base - cur_pv.pv_va;
    841 	}
    842 
    843 	/*
    844 	 * Now we map the final chunk.
    845 	 */
    846 	VPRINTF("%s: mapping last chunk VA %#lx..%#lx (PA %#lx, prot %d, cache %d)\n",
    847 	    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    848 	    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    849 	pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    850 	    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    851 
    852 	/*
    853 	 * Now we map the stuff that isn't directly after the kernel
    854 	 */
    855 	if (map_vectors_p) {
    856 		/* Map the vector page. */
    857 		pmap_map_entry(l1pt_va, systempage.pv_va, systempage.pv_pa,
    858 		    VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE, PTE_CACHE);
    859 	}
    860 
    861 	/* Map the Mini-Data cache clean area. */
    862 #if ARM_MMU_XSCALE == 1
    863 #if (ARM_NMMUS > 1)
    864 	if (xscale_use_minidata)
    865 #endif
    866 		xscale_setup_minidata(l1pt_va, minidataclean.pv_va,
    867 		    minidataclean.pv_pa);
    868 #endif
    869 
    870 	/*
    871 	 * Map integrated peripherals at same address in first level page
    872 	 * table so that we can continue to use console.
    873 	 */
    874 	if (devmap)
    875 		pmap_devmap_bootstrap(l1pt_va, devmap);
    876 
    877 	/* Tell the user about where all the bits and pieces live. */
    878 	VPRINTF("%22s       Physical              Virtual        Num\n", " ");
    879 	VPRINTF("%22s Starting    Ending    Starting    Ending   Pages\n", " ");
    880 
    881 #ifdef VERBOSE_INIT_ARM
    882 	static const char mem_fmt[] =
    883 	    "%20s: 0x%08lx 0x%08lx 0x%08lx 0x%08lx %u\n";
    884 	static const char mem_fmt_nov[] =
    885 	    "%20s: 0x%08lx 0x%08lx                       %zu\n";
    886 #endif
    887 
    888 #if 0
    889 	// XXX Doesn't make sense if kernel not at bottom of RAM
    890 	VPRINTF(mem_fmt, "SDRAM", bmi->bmi_start, bmi->bmi_end - 1,
    891 	    KERN_PHYSTOV(bmi->bmi_start), KERN_PHYSTOV(bmi->bmi_end - 1),
    892 	    (int)physmem);
    893 #endif
    894 	VPRINTF(mem_fmt, "text section",
    895 	       text.pv_pa, text.pv_pa + text.pv_size - 1,
    896 	       text.pv_va, text.pv_va + text.pv_size - 1,
    897 	       (int)(text.pv_size / PAGE_SIZE));
    898 	VPRINTF(mem_fmt, "data section",
    899 	       KERN_VTOPHYS((vaddr_t)__data_start), KERN_VTOPHYS((vaddr_t)_edata),
    900 	       (vaddr_t)__data_start, (vaddr_t)_edata,
    901 	       (int)((round_page((vaddr_t)_edata)
    902 		      - trunc_page((vaddr_t)__data_start)) / PAGE_SIZE));
    903 	VPRINTF(mem_fmt, "bss section",
    904 	       KERN_VTOPHYS((vaddr_t)__bss_start), KERN_VTOPHYS((vaddr_t)__bss_end__),
    905 	       (vaddr_t)__bss_start, (vaddr_t)__bss_end__,
    906 	       (int)((round_page((vaddr_t)__bss_end__)
    907 		      - trunc_page((vaddr_t)__bss_start)) / PAGE_SIZE));
    908 	VPRINTF(mem_fmt, "L1 page directory",
    909 	    kernel_l1pt.pv_pa, kernel_l1pt.pv_pa + L1_TABLE_SIZE - 1,
    910 	    kernel_l1pt.pv_va, kernel_l1pt.pv_va + L1_TABLE_SIZE - 1,
    911 	    L1_TABLE_SIZE / PAGE_SIZE);
    912 	VPRINTF(mem_fmt, "ABT stack (CPU 0)",
    913 	    abtstack.pv_pa, abtstack.pv_pa + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
    914 	    abtstack.pv_va, abtstack.pv_va + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
    915 	    ABT_STACK_SIZE);
    916 	VPRINTF(mem_fmt, "FIQ stack (CPU 0)",
    917 	    fiqstack.pv_pa, fiqstack.pv_pa + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
    918 	    fiqstack.pv_va, fiqstack.pv_va + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
    919 	    FIQ_STACK_SIZE);
    920 	VPRINTF(mem_fmt, "IRQ stack (CPU 0)",
    921 	    irqstack.pv_pa, irqstack.pv_pa + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
    922 	    irqstack.pv_va, irqstack.pv_va + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
    923 	    IRQ_STACK_SIZE);
    924 	VPRINTF(mem_fmt, "UND stack (CPU 0)",
    925 	    undstack.pv_pa, undstack.pv_pa + (UND_STACK_SIZE * PAGE_SIZE) - 1,
    926 	    undstack.pv_va, undstack.pv_va + (UND_STACK_SIZE * PAGE_SIZE) - 1,
    927 	    UND_STACK_SIZE);
    928 	VPRINTF(mem_fmt, "IDLE stack (CPU 0)",
    929 	    idlestack.pv_pa, idlestack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
    930 	    idlestack.pv_va, idlestack.pv_va + (UPAGES * PAGE_SIZE) - 1,
    931 	    UPAGES);
    932 	VPRINTF(mem_fmt, "SVC stack",
    933 	    kernelstack.pv_pa, kernelstack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
    934 	    kernelstack.pv_va, kernelstack.pv_va + (UPAGES * PAGE_SIZE) - 1,
    935 	    UPAGES);
    936 	VPRINTF(mem_fmt, "Message Buffer",
    937 	    msgbuf.pv_pa, msgbuf.pv_pa + (msgbuf_pgs * PAGE_SIZE) - 1,
    938 	    msgbuf.pv_va, msgbuf.pv_va + (msgbuf_pgs * PAGE_SIZE) - 1,
    939 	    (int)msgbuf_pgs);
    940 	if (map_vectors_p) {
    941 		VPRINTF(mem_fmt, "Exception Vectors",
    942 		    systempage.pv_pa, systempage.pv_pa + PAGE_SIZE - 1,
    943 		    systempage.pv_va, systempage.pv_va + PAGE_SIZE - 1,
    944 		    1);
    945 	}
    946 	for (size_t i = 0; i < bmi->bmi_nfreeblocks; i++) {
    947 		pv = &bmi->bmi_freeblocks[i];
    948 
    949 		VPRINTF(mem_fmt_nov, "Free Memory",
    950 		    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    951 		    pv->pv_size / PAGE_SIZE);
    952 	}
    953 	/*
    954 	 * Now we have the real page tables in place so we can switch to them.
    955 	 * Once this is done we will be running with the REAL kernel page
    956 	 * tables.
    957 	 */
    958 
    959 	VPRINTF("TTBR0=%#x", armreg_ttbr_read());
    960 #ifdef _ARM_ARCH_6
    961 	VPRINTF(" TTBR1=%#x TTBCR=%#x CONTEXTIDR=%#x",
    962 	    armreg_ttbr1_read(), armreg_ttbcr_read(),
    963 	    armreg_contextidr_read());
    964 #endif
    965 	VPRINTF("\n");
    966 
    967 	/* Switch tables */
    968 	VPRINTF("switching to new L1 page table @%#lx...\n", l1pt_pa);
    969 
    970 	cpu_ttb = l1pt_pa;
    971 
    972 	cpu_domains(DOMAIN_DEFAULT);
    973 
    974 	cpu_idcache_wbinv_all();
    975 
    976 #ifdef __HAVE_GENERIC_START
    977 
    978 	/*
    979 	 * Turn on caches and set SCTLR/ACTLR
    980 	 */
    981 	cpu_setup(boot_args);
    982 #endif
    983 
    984 	VPRINTF(" ttb");
    985 
    986 #ifdef ARM_MMU_EXTENDED
    987 	/*
    988 	 * TTBCR should have been initialized by the MD start code.
    989 	 */
    990 	KASSERT((armreg_contextidr_read() & 0xff) == 0);
    991 	KASSERT(armreg_ttbcr_read() == __SHIFTIN(1, TTBCR_S_N));
    992 	/*
    993 	 * Disable lookups via TTBR0 until there is an activated pmap.
    994 	 */
    995 	armreg_ttbcr_write(armreg_ttbcr_read() | TTBCR_S_PD0);
    996 	cpu_setttb(l1pt_pa, KERNEL_PID);
    997 	arm_isb();
    998 #else
    999 	cpu_setttb(l1pt_pa, true);
   1000 #endif
   1001 
   1002 	cpu_tlb_flushID();
   1003 
   1004 #ifdef ARM_MMU_EXTENDED
   1005 	VPRINTF("\nsctlr=%#x actlr=%#x\n",
   1006 	    armreg_sctlr_read(), armreg_auxctl_read());
   1007 #else
   1008 	VPRINTF(" (TTBR0=%#x)", armreg_ttbr_read());
   1009 #endif
   1010 
   1011 #ifdef MULTIPROCESSOR
   1012 #ifndef __HAVE_GENERIC_START
   1013 	/*
   1014 	 * Kick the secondaries to load the TTB.  After which they'll go
   1015 	 * back to sleep to wait for the final kick so they will hatch.
   1016 	 */
   1017 	VPRINTF(" hatchlings");
   1018 	cpu_boot_secondary_processors();
   1019 #endif
   1020 #endif
   1021 
   1022 	VPRINTF(" OK\n");
   1023 }
   1024