Home | History | Annotate | Line # | Download | only in arm32
arm32_kvminit.c revision 1.61
      1 /*	$NetBSD: arm32_kvminit.c,v 1.61 2020/07/03 06:22:48 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 1.61 2020/07/03 06:22:48 skrll Exp $");
    131 
    132 #include <sys/param.h>
    133 
    134 #include <sys/bus.h>
    135 #include <sys/device.h>
    136 #include <sys/kernel.h>
    137 #include <sys/reboot.h>
    138 
    139 #include <dev/cons.h>
    140 
    141 #include <uvm/uvm_extern.h>
    142 
    143 #include <arm/arm32/machdep.h>
    144 #include <arm/bootconfig.h>
    145 #include <arm/db_machdep.h>
    146 #include <arm/locore.h>
    147 #include <arm/undefined.h>
    148 
    149 #if defined(FDT)
    150 #include <arch/evbarm/fdt/platform.h>
    151 #include <arm/fdt/arm_fdtvar.h>
    152 #endif
    153 
    154 #ifdef MULTIPROCESSOR
    155 #ifndef __HAVE_CPU_UAREA_ALLOC_IDLELWP
    156 #error __HAVE_CPU_UAREA_ALLOC_IDLELWP required to not waste pages for idlestack
    157 #endif
    158 #endif
    159 
    160 #ifdef VERBOSE_INIT_ARM
    161 #define VPRINTF(...)	printf(__VA_ARGS__)
    162 #else
    163 #define VPRINTF(...)	__nothing
    164 #endif
    165 
    166 struct bootmem_info bootmem_info;
    167 
    168 extern void *msgbufaddr;
    169 paddr_t msgbufphys;
    170 paddr_t physical_start;
    171 paddr_t physical_end;
    172 
    173 extern char etext[];
    174 extern char __data_start[], _edata[];
    175 extern char __bss_start[], __bss_end__[];
    176 extern char _end[];
    177 
    178 /* Page tables for mapping kernel VM */
    179 #define KERNEL_L2PT_VMDATA_NUM	8	/* start with 32MB of KVM */
    180 
    181 u_long kern_vtopdiff __attribute__((__section__(".data")));
    182 
    183 void
    184 arm32_bootmem_init(paddr_t memstart, psize_t memsize, vsize_t kernelstart)
    185 {
    186 	struct bootmem_info * const bmi = &bootmem_info;
    187 	pv_addr_t *pv = bmi->bmi_freeblocks;
    188 
    189 	/*
    190 	 * FDT/generic start fills in kern_vtopdiff early
    191 	 */
    192 #if defined(__HAVE_GENERIC_START)
    193 	extern char KERNEL_BASE_virt[];
    194 	extern char const __stop__init_memory[];
    195 
    196 	VPRINTF("%s: kern_vtopdiff=%#lx\n", __func__, kern_vtopdiff);
    197 
    198 	vaddr_t kstartva = trunc_page((vaddr_t)KERNEL_BASE_virt);
    199 	vaddr_t kendva = round_page((vaddr_t)__stop__init_memory);
    200 
    201 	kernelstart = KERN_VTOPHYS(kstartva);
    202 
    203 	VPRINTF("%s: kstartva=%#lx, kernelstart=%#lx\n", __func__, kstartva, kernelstart);
    204 #else
    205 	vaddr_t kendva = round_page((vaddr_t)_end);
    206 
    207 #if defined(KERNEL_BASE_VOFFSET)
    208 	kern_vtopdiff = KERNEL_BASE_VOFFSET;
    209 #else
    210 	KASSERT(memstart == kernelstart);
    211 	kern_vtopdiff = KERNEL_BASE + memstart;
    212 #endif
    213 #endif
    214 	paddr_t kernelend = KERN_VTOPHYS(kendva);
    215 
    216 	VPRINTF("%s: memstart=%#lx, memsize=%#lx\n", __func__,
    217 	    memstart, memsize);
    218 	VPRINTF("%s: kernelstart=%#lx, kernelend=%#lx\n", __func__,
    219 	    kernelstart, kernelend);
    220 
    221 	physical_start = bmi->bmi_start = memstart;
    222 	physical_end = bmi->bmi_end = memstart + memsize;
    223 #ifndef ARM_HAS_LPAE
    224 	if (physical_end == 0) {
    225 		physical_end = -PAGE_SIZE;
    226 		memsize -= PAGE_SIZE;
    227 		bmi->bmi_end -= PAGE_SIZE;
    228 		VPRINTF("%s: memsize shrunk by a page to avoid ending at 4GB\n",
    229 		    __func__);
    230 	}
    231 #endif
    232 	physmem = memsize / PAGE_SIZE;
    233 
    234 	/*
    235 	 * Let's record where the kernel lives.
    236 	 */
    237 
    238 	bmi->bmi_kernelstart = kernelstart;
    239 	bmi->bmi_kernelend = kernelend;
    240 
    241 #if defined(FDT)
    242 	fdt_add_reserved_memory_range(bmi->bmi_kernelstart,
    243 	    bmi->bmi_kernelend - bmi->bmi_kernelstart);
    244 #endif
    245 
    246 	VPRINTF("%s: kernel phys start %#lx end %#lx\n", __func__, kernelstart,
    247 	    kernelend);
    248 
    249 #if 0
    250 	// XXX Makes RPI abort
    251 	KASSERT((kernelstart & (L2_S_SEGSIZE - 1)) == 0);
    252 #endif
    253 	/*
    254 	 * Now the rest of the free memory must be after the kernel.
    255 	 */
    256 	pv->pv_pa = bmi->bmi_kernelend;
    257 	pv->pv_va = KERN_PHYSTOV(pv->pv_pa);
    258 	pv->pv_size = bmi->bmi_end - bmi->bmi_kernelend;
    259 	bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
    260 	VPRINTF("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
    261 	    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
    262 	    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
    263 	pv++;
    264 
    265 	/*
    266 	 * Add a free block for any memory before the kernel.
    267 	 */
    268 	if (bmi->bmi_start < bmi->bmi_kernelstart) {
    269 		pv->pv_pa = bmi->bmi_start;
    270 		pv->pv_va = KERN_PHYSTOV(pv->pv_pa);
    271 		pv->pv_size = bmi->bmi_kernelstart - pv->pv_pa;
    272 		bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
    273 		VPRINTF("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
    274 		    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
    275 		    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
    276 		pv++;
    277 	}
    278 
    279 	bmi->bmi_nfreeblocks = pv - bmi->bmi_freeblocks;
    280 
    281 	SLIST_INIT(&bmi->bmi_freechunks);
    282 	SLIST_INIT(&bmi->bmi_chunks);
    283 }
    284 
    285 static bool
    286 concat_pvaddr(pv_addr_t *acc_pv, pv_addr_t *pv)
    287 {
    288 	if (acc_pv->pv_pa + acc_pv->pv_size == pv->pv_pa
    289 	    && acc_pv->pv_va + acc_pv->pv_size == pv->pv_va
    290 	    && acc_pv->pv_prot == pv->pv_prot
    291 	    && acc_pv->pv_cache == pv->pv_cache) {
    292 #if 0
    293 		VPRINTF("%s: appending pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    294 		    __func__, pv, pv->pv_pa, pv->pv_pa + pv->pv_size,
    295 		    acc_pv->pv_pa, acc_pv->pv_pa + acc_pv->pv_size);
    296 #endif
    297 		acc_pv->pv_size += pv->pv_size;
    298 		return true;
    299 	}
    300 
    301 	return false;
    302 }
    303 
    304 static void
    305 add_pages(struct bootmem_info *bmi, pv_addr_t *pv)
    306 {
    307 	pv_addr_t **pvp = &SLIST_FIRST(&bmi->bmi_chunks);
    308 	while ((*pvp) != NULL && (*pvp)->pv_va <= pv->pv_va) {
    309 		pv_addr_t * const pv0 = (*pvp);
    310 		KASSERT(SLIST_NEXT(pv0, pv_list) == NULL || pv0->pv_pa < SLIST_NEXT(pv0, pv_list)->pv_pa);
    311 		if (concat_pvaddr(pv0, pv)) {
    312 			VPRINTF("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    313 			    __func__, "appending", pv,
    314 			    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    315 			    pv0->pv_pa, pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
    316 			pv = SLIST_NEXT(pv0, pv_list);
    317 			if (pv != NULL && concat_pvaddr(pv0, pv)) {
    318 				VPRINTF("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
    319 				    __func__, "merging", pv,
    320 				    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    321 				    pv0->pv_pa,
    322 				    pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
    323 				SLIST_REMOVE_AFTER(pv0, pv_list);
    324 				SLIST_INSERT_HEAD(&bmi->bmi_freechunks, pv, pv_list);
    325 			}
    326 			return;
    327 		}
    328 		KASSERT(pv->pv_va != (*pvp)->pv_va);
    329 		pvp = &SLIST_NEXT(*pvp, pv_list);
    330 	}
    331 	KASSERT((*pvp) == NULL || pv->pv_va < (*pvp)->pv_va);
    332 	pv_addr_t * const new_pv = SLIST_FIRST(&bmi->bmi_freechunks);
    333 	KASSERT(new_pv != NULL);
    334 	SLIST_REMOVE_HEAD(&bmi->bmi_freechunks, pv_list);
    335 	*new_pv = *pv;
    336 	SLIST_NEXT(new_pv, pv_list) = *pvp;
    337 	(*pvp) = new_pv;
    338 
    339 	VPRINTF("%s: adding pv %p (pa %#lx, va %#lx, %lu pages) ",
    340 	    __func__, new_pv, new_pv->pv_pa, new_pv->pv_va,
    341 	    new_pv->pv_size / PAGE_SIZE);
    342 	if (SLIST_NEXT(new_pv, pv_list)) {
    343 		VPRINTF("before pa %#lx\n", SLIST_NEXT(new_pv, pv_list)->pv_pa);
    344 	} else {
    345 		VPRINTF("at tail\n");
    346 	}
    347 }
    348 
    349 static void
    350 valloc_pages(struct bootmem_info *bmi, pv_addr_t *pv, size_t npages,
    351     int prot, int cache, bool zero_p)
    352 {
    353 	size_t nbytes = npages * PAGE_SIZE;
    354 	pv_addr_t *free_pv = bmi->bmi_freeblocks;
    355 	size_t free_idx = 0;
    356 	static bool l1pt_found;
    357 
    358 	KASSERT(npages > 0);
    359 
    360 	/*
    361 	 * If we haven't allocated the kernel L1 page table and we are aligned
    362 	 * at a L1 table boundary, alloc the memory for it.
    363 	 */
    364 	if (!l1pt_found
    365 	    && (free_pv->pv_pa & (L1_TABLE_SIZE - 1)) == 0
    366 	    && free_pv->pv_size >= L1_TABLE_SIZE) {
    367 		l1pt_found = true;
    368 		VPRINTF(" l1pt");
    369 
    370 		valloc_pages(bmi, &kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE,
    371 		    VM_PROT_READ | VM_PROT_WRITE, PTE_PAGETABLE, true);
    372 		add_pages(bmi, &kernel_l1pt);
    373 	}
    374 
    375 	while (nbytes > free_pv->pv_size) {
    376 		free_pv++;
    377 		free_idx++;
    378 		if (free_idx == bmi->bmi_nfreeblocks) {
    379 			panic("%s: could not allocate %zu bytes",
    380 			    __func__, nbytes);
    381 		}
    382 	}
    383 
    384 	/*
    385 	 * As we allocate the memory, make sure that we don't walk over
    386 	 * our current first level translation table.
    387 	 */
    388 	KASSERT((armreg_ttbr_read() & ~(L1_TABLE_SIZE - 1)) != free_pv->pv_pa);
    389 
    390 #if defined(FDT)
    391 	fdt_add_reserved_memory_range(free_pv->pv_pa, nbytes);
    392 #endif
    393 	pv->pv_pa = free_pv->pv_pa;
    394 	pv->pv_va = free_pv->pv_va;
    395 	pv->pv_size = nbytes;
    396 	pv->pv_prot = prot;
    397 	pv->pv_cache = cache;
    398 
    399 	/*
    400 	 * If PTE_PAGETABLE uses the same cache modes as PTE_CACHE
    401 	 * just use PTE_CACHE.
    402 	 */
    403 	if (cache == PTE_PAGETABLE
    404 	    && pte_l1_s_cache_mode == pte_l1_s_cache_mode_pt
    405 	    && pte_l2_l_cache_mode == pte_l2_l_cache_mode_pt
    406 	    && pte_l2_s_cache_mode == pte_l2_s_cache_mode_pt)
    407 		pv->pv_cache = PTE_CACHE;
    408 
    409 	free_pv->pv_pa += nbytes;
    410 	free_pv->pv_va += nbytes;
    411 	free_pv->pv_size -= nbytes;
    412 	if (free_pv->pv_size == 0) {
    413 		--bmi->bmi_nfreeblocks;
    414 		for (; free_idx < bmi->bmi_nfreeblocks; free_idx++) {
    415 			free_pv[0] = free_pv[1];
    416 		}
    417 	}
    418 
    419 	bmi->bmi_freepages -= npages;
    420 
    421 	if (zero_p)
    422 		memset((void *)pv->pv_va, 0, nbytes);
    423 }
    424 
    425 void
    426 arm32_kernel_vm_init(vaddr_t kernel_vm_base, vaddr_t vectors, vaddr_t iovbase,
    427     const struct pmap_devmap *devmap, bool mapallmem_p)
    428 {
    429 	struct bootmem_info * const bmi = &bootmem_info;
    430 #ifdef MULTIPROCESSOR
    431 	const size_t cpu_num = arm_cpu_max;
    432 #else
    433 	const size_t cpu_num = 1;
    434 #endif
    435 
    436 #ifdef ARM_HAS_VBAR
    437 	const bool map_vectors_p = false;
    438 #elif defined(CPU_ARMV7) || defined(CPU_ARM11)
    439 	const bool map_vectors_p = vectors == ARM_VECTORS_HIGH
    440 	    || (armreg_pfr1_read() & ARM_PFR1_SEC_MASK) == 0;
    441 #else
    442 	const bool map_vectors_p = true;
    443 #endif
    444 
    445 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS
    446 	KASSERT(mapallmem_p);
    447 #ifdef ARM_MMU_EXTENDED
    448 	/*
    449 	 * The direct map VA space ends at the start of the kernel VM space.
    450 	 */
    451 	pmap_directlimit = kernel_vm_base;
    452 #else
    453 	KASSERT(kernel_vm_base - KERNEL_BASE >= physical_end - physical_start);
    454 #endif /* ARM_MMU_EXTENDED */
    455 #endif /* __HAVE_MM_MD_DIRECT_MAPPED_PHYS */
    456 
    457 	/*
    458 	 * Calculate the number of L2 pages needed for mapping the
    459 	 * kernel + data + stuff.  Assume 2 L2 pages for kernel, 1 for vectors,
    460 	 * and 1 for IO
    461 	 */
    462 	size_t kernel_size = bmi->bmi_kernelend;
    463 	kernel_size -= (bmi->bmi_kernelstart & -L2_S_SEGSIZE);
    464 	kernel_size += L1_TABLE_SIZE;
    465 	kernel_size += PAGE_SIZE * KERNEL_L2PT_VMDATA_NUM;
    466 	if (map_vectors_p) {
    467 		kernel_size += PAGE_SIZE;	/* L2PT for VECTORS */
    468 	}
    469 	if (iovbase) {
    470 		kernel_size += PAGE_SIZE;	/* L2PT for IO */
    471 	}
    472 	kernel_size +=
    473 	    cpu_num * (ABT_STACK_SIZE + FIQ_STACK_SIZE + IRQ_STACK_SIZE
    474 	    + UND_STACK_SIZE + UPAGES) * PAGE_SIZE;
    475 	kernel_size += round_page(MSGBUFSIZE);
    476 	kernel_size += 0x10000;	/* slop */
    477 	if (!mapallmem_p) {
    478 		kernel_size += PAGE_SIZE
    479 		    * ((kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE);
    480 	}
    481 	kernel_size = round_page(kernel_size);
    482 
    483 	/*
    484 	 * Now we know how many L2 pages it will take.
    485 	 */
    486 	const size_t KERNEL_L2PT_KERNEL_NUM =
    487 	    round_page(kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE;
    488 
    489 	VPRINTF("%s: %zu L2 pages are needed to map %#zx kernel bytes\n",
    490 	    __func__, KERNEL_L2PT_KERNEL_NUM, kernel_size);
    491 
    492 	KASSERT(KERNEL_L2PT_KERNEL_NUM + KERNEL_L2PT_VMDATA_NUM < __arraycount(bmi->bmi_l2pts));
    493 	pv_addr_t * const kernel_l2pt = bmi->bmi_l2pts;
    494 	pv_addr_t * const vmdata_l2pt = kernel_l2pt + KERNEL_L2PT_KERNEL_NUM;
    495 	pv_addr_t msgbuf;
    496 	pv_addr_t text;
    497 	pv_addr_t data;
    498 	pv_addr_t chunks[__arraycount(bmi->bmi_l2pts) + 11];
    499 #if ARM_MMU_XSCALE == 1
    500 	pv_addr_t minidataclean;
    501 #endif
    502 
    503 	/*
    504 	 * We need to allocate some fixed page tables to get the kernel going.
    505 	 *
    506 	 * We are going to allocate our bootstrap pages from the beginning of
    507 	 * the free space that we just calculated.  We allocate one page
    508 	 * directory and a number of page tables and store the physical
    509 	 * addresses in the bmi_l2pts array in bootmem_info.
    510 	 *
    511 	 * The kernel page directory must be on a 16K boundary.  The page
    512 	 * tables must be on 4K boundaries.  What we do is allocate the
    513 	 * page directory on the first 16K boundary that we encounter, and
    514 	 * the page tables on 4K boundaries otherwise.  Since we allocate
    515 	 * at least 3 L2 page tables, we are guaranteed to encounter at
    516 	 * least one 16K aligned region.
    517 	 */
    518 
    519 	VPRINTF("%s: allocating page tables for", __func__);
    520 	for (size_t i = 0; i < __arraycount(chunks); i++) {
    521 		SLIST_INSERT_HEAD(&bmi->bmi_freechunks, &chunks[i], pv_list);
    522 	}
    523 
    524 	kernel_l1pt.pv_pa = 0;
    525 	kernel_l1pt.pv_va = 0;
    526 
    527 	/*
    528 	 * Allocate the L2 pages, but if we get to a page that is aligned for
    529 	 * an L1 page table, we will allocate the pages for it first and then
    530 	 * allocate the L2 page.
    531 	 */
    532 
    533 	if (map_vectors_p) {
    534 		/*
    535 		 * First allocate L2 page for the vectors.
    536 		 */
    537 		VPRINTF(" vector");
    538 		valloc_pages(bmi, &bmi->bmi_vector_l2pt, 1,
    539 		    VM_PROT_READ | VM_PROT_WRITE, PTE_PAGETABLE, true);
    540 		add_pages(bmi, &bmi->bmi_vector_l2pt);
    541 	}
    542 
    543 	/*
    544 	 * Now allocate L2 pages for the kernel
    545 	 */
    546 	VPRINTF(" kernel");
    547 	for (size_t idx = 0; idx < KERNEL_L2PT_KERNEL_NUM; ++idx) {
    548 		valloc_pages(bmi, &kernel_l2pt[idx], 1,
    549 		    VM_PROT_READ | VM_PROT_WRITE, PTE_PAGETABLE, true);
    550 		add_pages(bmi, &kernel_l2pt[idx]);
    551 	}
    552 
    553 	/*
    554 	 * Now allocate L2 pages for the initial kernel VA space.
    555 	 */
    556 	VPRINTF(" vm");
    557 	for (size_t idx = 0; idx < KERNEL_L2PT_VMDATA_NUM; ++idx) {
    558 		valloc_pages(bmi, &vmdata_l2pt[idx], 1,
    559 		    VM_PROT_READ | VM_PROT_WRITE, PTE_PAGETABLE, true);
    560 		add_pages(bmi, &vmdata_l2pt[idx]);
    561 	}
    562 
    563 	/*
    564 	 * If someone wanted a L2 page for I/O, allocate it now.
    565 	 */
    566 	if (iovbase) {
    567 		VPRINTF(" io");
    568 		valloc_pages(bmi, &bmi->bmi_io_l2pt, 1,
    569 		    VM_PROT_READ | VM_PROT_WRITE, PTE_PAGETABLE, true);
    570 		add_pages(bmi, &bmi->bmi_io_l2pt);
    571 	}
    572 
    573 	VPRINTF("%s: allocating stacks\n", __func__);
    574 
    575 	/* Allocate stacks for all modes and CPUs */
    576 	valloc_pages(bmi, &abtstack, ABT_STACK_SIZE * cpu_num,
    577 	    VM_PROT_READ | VM_PROT_WRITE, PTE_CACHE, true);
    578 	add_pages(bmi, &abtstack);
    579 	valloc_pages(bmi, &fiqstack, FIQ_STACK_SIZE * cpu_num,
    580 	    VM_PROT_READ | VM_PROT_WRITE, PTE_CACHE, true);
    581 	add_pages(bmi, &fiqstack);
    582 	valloc_pages(bmi, &irqstack, IRQ_STACK_SIZE * cpu_num,
    583 	    VM_PROT_READ | VM_PROT_WRITE, PTE_CACHE, true);
    584 	add_pages(bmi, &irqstack);
    585 	valloc_pages(bmi, &undstack, UND_STACK_SIZE * cpu_num,
    586 	    VM_PROT_READ | VM_PROT_WRITE, PTE_CACHE, true);
    587 	add_pages(bmi, &undstack);
    588 	valloc_pages(bmi, &idlestack, UPAGES * cpu_num,		/* SVC32 */
    589 	    VM_PROT_READ | VM_PROT_WRITE, PTE_CACHE, true);
    590 	add_pages(bmi, &idlestack);
    591 	valloc_pages(bmi, &kernelstack, UPAGES,			/* SVC32 */
    592 	    VM_PROT_READ | VM_PROT_WRITE, PTE_CACHE, true);
    593 	add_pages(bmi, &kernelstack);
    594 
    595 	/* Allocate the message buffer from the end of memory. */
    596 	const size_t msgbuf_pgs = round_page(MSGBUFSIZE) / PAGE_SIZE;
    597 	valloc_pages(bmi, &msgbuf, msgbuf_pgs,
    598 	    VM_PROT_READ | VM_PROT_WRITE, PTE_CACHE, false);
    599 	add_pages(bmi, &msgbuf);
    600 	msgbufphys = msgbuf.pv_pa;
    601 	msgbufaddr = (void *)msgbuf.pv_va;
    602 
    603 	if (map_vectors_p) {
    604 		/*
    605 		 * Allocate a page for the system vector page.
    606 		 * This page will just contain the system vectors and can be
    607 		 * shared by all processes.
    608 		 */
    609 		VPRINTF(" vector");
    610 
    611 		valloc_pages(bmi, &systempage, 1,
    612 		    VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE,
    613 		    PTE_CACHE, true);
    614 	}
    615 	systempage.pv_va = vectors;
    616 
    617 	/*
    618 	 * If the caller needed a few extra pages for some reason, allocate
    619 	 * them now.
    620 	 */
    621 #if ARM_MMU_XSCALE == 1
    622 #if (ARM_NMMUS > 1)
    623 	if (xscale_use_minidata)
    624 #endif
    625 		valloc_pages(bmi, &minidataclean, 1,
    626 		    VM_PROT_READ | VM_PROT_WRITE, 0, true);
    627 #endif
    628 
    629 	/*
    630 	 * Ok we have allocated physical pages for the primary kernel
    631 	 * page tables and stacks.  Let's just confirm that.
    632 	 */
    633 	if (kernel_l1pt.pv_va == 0
    634 	    && (!kernel_l1pt.pv_pa || (kernel_l1pt.pv_pa & (L1_TABLE_SIZE - 1)) != 0))
    635 		panic("%s: Failed to allocate or align the kernel "
    636 		    "page directory", __func__);
    637 
    638 	VPRINTF("Creating L1 page table at 0x%08lx/0x%08lx\n",
    639 	    kernel_l1pt.pv_va, kernel_l1pt.pv_pa);
    640 
    641 	/*
    642 	 * Now we start construction of the L1 page table
    643 	 * We start by mapping the L2 page tables into the L1.
    644 	 * This means that we can replace L1 mappings later on if necessary
    645 	 */
    646 	vaddr_t l1pt_va = kernel_l1pt.pv_va;
    647 	paddr_t l1pt_pa = kernel_l1pt.pv_pa;
    648 
    649 	if (map_vectors_p) {
    650 		/* Map the L2 pages tables in the L1 page table */
    651 		pmap_link_l2pt(l1pt_va, systempage.pv_va & -L2_S_SEGSIZE,
    652 		    &bmi->bmi_vector_l2pt);
    653 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) "
    654 		    "for VA %#lx\n (vectors)",
    655 		    __func__, bmi->bmi_vector_l2pt.pv_va,
    656 		    bmi->bmi_vector_l2pt.pv_pa, systempage.pv_va);
    657 	}
    658 
    659 	/*
    660 	 * This enforces an alignment requirement of L2_S_SEGSIZE for kernel
    661 	 * start PA
    662 	 */
    663 	const vaddr_t kernel_base =
    664 	    KERN_PHYSTOV(bmi->bmi_kernelstart & -L2_S_SEGSIZE);
    665 
    666 	VPRINTF("%s: kernel_base %lx KERNEL_L2PT_KERNEL_NUM %zu\n", __func__,
    667 	    kernel_base, KERNEL_L2PT_KERNEL_NUM);
    668 
    669 	for (size_t idx = 0; idx < KERNEL_L2PT_KERNEL_NUM; idx++) {
    670 		pmap_link_l2pt(l1pt_va, kernel_base + idx * L2_S_SEGSIZE,
    671 		    &kernel_l2pt[idx]);
    672 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (kernel)\n",
    673 		    __func__, kernel_l2pt[idx].pv_va,
    674 		    kernel_l2pt[idx].pv_pa, kernel_base + idx * L2_S_SEGSIZE);
    675 	}
    676 
    677 	VPRINTF("%s: kernel_vm_base %lx KERNEL_L2PT_VMDATA_NUM %d\n", __func__,
    678 	    kernel_vm_base, KERNEL_L2PT_VMDATA_NUM);
    679 
    680 	for (size_t idx = 0; idx < KERNEL_L2PT_VMDATA_NUM; idx++) {
    681 		pmap_link_l2pt(l1pt_va, kernel_vm_base + idx * L2_S_SEGSIZE,
    682 		    &vmdata_l2pt[idx]);
    683 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (vm)\n",
    684 		    __func__, vmdata_l2pt[idx].pv_va, vmdata_l2pt[idx].pv_pa,
    685 		    kernel_vm_base + idx * L2_S_SEGSIZE);
    686 	}
    687 	if (iovbase) {
    688 		pmap_link_l2pt(l1pt_va, iovbase & -L2_S_SEGSIZE, &bmi->bmi_io_l2pt);
    689 		VPRINTF("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (io)\n",
    690 		    __func__, bmi->bmi_io_l2pt.pv_va, bmi->bmi_io_l2pt.pv_pa,
    691 		    iovbase & -L2_S_SEGSIZE);
    692 	}
    693 
    694 	/* update the top of the kernel VM */
    695 	pmap_curmaxkvaddr =
    696 	    kernel_vm_base + (KERNEL_L2PT_VMDATA_NUM * L2_S_SEGSIZE);
    697 
    698 	// This could be done earlier and then the kernel data and pages
    699 	// allocated above would get merged (concatentated)
    700 
    701 	VPRINTF("Mapping kernel\n");
    702 
    703 	extern char etext[];
    704 	size_t totalsize = bmi->bmi_kernelend - bmi->bmi_kernelstart;
    705 	size_t textsize = KERN_VTOPHYS((uintptr_t)etext) - bmi->bmi_kernelstart;
    706 
    707 	textsize = (textsize + PGOFSET) & ~PGOFSET;
    708 
    709 	/* start at offset of kernel in RAM */
    710 
    711 	text.pv_pa = bmi->bmi_kernelstart;
    712 	text.pv_va = KERN_PHYSTOV(bmi->bmi_kernelstart);
    713 	text.pv_size = textsize;
    714 	text.pv_prot = VM_PROT_READ | VM_PROT_EXECUTE;
    715 	text.pv_cache = PTE_CACHE;
    716 
    717 	VPRINTF("%s: adding chunk for kernel text %#lx..%#lx (VA %#lx)\n",
    718 	    __func__, text.pv_pa, text.pv_pa + text.pv_size - 1, text.pv_va);
    719 
    720 	add_pages(bmi, &text);
    721 
    722 	data.pv_pa = text.pv_pa + textsize;
    723 	data.pv_va = text.pv_va + textsize;
    724 	data.pv_size = totalsize - textsize;
    725 	data.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    726 	data.pv_cache = PTE_CACHE;
    727 
    728 	VPRINTF("%s: adding chunk for kernel data/bss %#lx..%#lx (VA %#lx)\n",
    729 	    __func__, data.pv_pa, data.pv_pa + data.pv_size - 1, data.pv_va);
    730 
    731 	add_pages(bmi, &data);
    732 
    733 	VPRINTF("Listing Chunks\n");
    734 
    735 	pv_addr_t *lpv;
    736 	SLIST_FOREACH(lpv, &bmi->bmi_chunks, pv_list) {
    737 		VPRINTF("%s: pv %p: chunk VA %#lx..%#lx "
    738 		    "(PA %#lx, prot %d, cache %d)\n",
    739 		    __func__, lpv, lpv->pv_va, lpv->pv_va + lpv->pv_size - 1,
    740 		    lpv->pv_pa, lpv->pv_prot, lpv->pv_cache);
    741 	}
    742 	VPRINTF("\nMapping Chunks\n");
    743 
    744 	pv_addr_t cur_pv;
    745 	pv_addr_t *pv = SLIST_FIRST(&bmi->bmi_chunks);
    746 	if (!mapallmem_p || pv->pv_pa == bmi->bmi_start) {
    747 		cur_pv = *pv;
    748 		KASSERTMSG(cur_pv.pv_va >= KERNEL_BASE, "%#lx", cur_pv.pv_va);
    749 		pv = SLIST_NEXT(pv, pv_list);
    750 	} else {
    751 		cur_pv.pv_va = KERNEL_BASE;
    752 		cur_pv.pv_pa = KERN_VTOPHYS(cur_pv.pv_va);
    753 		cur_pv.pv_size = pv->pv_pa - cur_pv.pv_pa;
    754 		cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    755 		cur_pv.pv_cache = PTE_CACHE;
    756 	}
    757 	while (pv != NULL) {
    758 		if (mapallmem_p) {
    759 			if (concat_pvaddr(&cur_pv, pv)) {
    760 				pv = SLIST_NEXT(pv, pv_list);
    761 				continue;
    762 			}
    763 			if (cur_pv.pv_pa + cur_pv.pv_size < pv->pv_pa) {
    764 				/*
    765 				 * See if we can extend the current pv to emcompass the
    766 				 * hole, and if so do it and retry the concatenation.
    767 				 */
    768 				if (cur_pv.pv_prot == (VM_PROT_READ | VM_PROT_WRITE)
    769 				    && cur_pv.pv_cache == PTE_CACHE) {
    770 					cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
    771 					continue;
    772 				}
    773 
    774 				/*
    775 				 * We couldn't so emit the current chunk and then
    776 				 */
    777 				VPRINTF("%s: mapping chunk VA %#lx..%#lx "
    778 				    "(PA %#lx, prot %d, cache %d)\n",
    779 				    __func__,
    780 				    cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    781 				    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    782 				pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    783 				    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    784 
    785 				/*
    786 				 * set the current chunk to the hole and try again.
    787 				 */
    788 				cur_pv.pv_pa += cur_pv.pv_size;
    789 				cur_pv.pv_va += cur_pv.pv_size;
    790 				cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
    791 				cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    792 				cur_pv.pv_cache = PTE_CACHE;
    793 				continue;
    794 			}
    795 		}
    796 
    797 		/*
    798 		 * The new pv didn't concatenate so emit the current one
    799 		 * and use the new pv as the current pv.
    800 		 */
    801 		VPRINTF("%s: mapping chunk VA %#lx..%#lx "
    802 		    "(PA %#lx, prot %d, cache %d)\n",
    803 		    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    804 		    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    805 		pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    806 		    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    807 		cur_pv = *pv;
    808 		pv = SLIST_NEXT(pv, pv_list);
    809 	}
    810 
    811 	/*
    812 	 * If we are mapping all of memory, let's map the rest of memory.
    813 	 */
    814 	if (mapallmem_p && cur_pv.pv_pa + cur_pv.pv_size < bmi->bmi_end) {
    815 		if (cur_pv.pv_prot == (VM_PROT_READ | VM_PROT_WRITE)
    816 		    && cur_pv.pv_cache == PTE_CACHE) {
    817 			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
    818 		} else {
    819 			KASSERTMSG(cur_pv.pv_va + cur_pv.pv_size <= kernel_vm_base,
    820 			    "%#lx >= %#lx", cur_pv.pv_va + cur_pv.pv_size,
    821 			    kernel_vm_base);
    822 			VPRINTF("%s: mapping chunk VA %#lx..%#lx "
    823 			    "(PA %#lx, prot %d, cache %d)\n",
    824 			    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    825 			    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    826 			pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    827 			    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    828 			cur_pv.pv_pa += cur_pv.pv_size;
    829 			cur_pv.pv_va += cur_pv.pv_size;
    830 			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
    831 			cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
    832 			cur_pv.pv_cache = PTE_CACHE;
    833 		}
    834 	}
    835 
    836 	/*
    837 	 * The amount we can direct map is limited by the start of the
    838 	 * virtual part of the kernel address space.  Don't overrun
    839 	 * into it.
    840 	 */
    841 	if (mapallmem_p && cur_pv.pv_va + cur_pv.pv_size > kernel_vm_base) {
    842 		cur_pv.pv_size = kernel_vm_base - cur_pv.pv_va;
    843 	}
    844 
    845 	/*
    846 	 * Now we map the final chunk.
    847 	 */
    848 	VPRINTF("%s: mapping last chunk VA %#lx..%#lx (PA %#lx, prot %d, cache %d)\n",
    849 	    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
    850 	    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
    851 	pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
    852 	    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
    853 
    854 	/*
    855 	 * Now we map the stuff that isn't directly after the kernel
    856 	 */
    857 	if (map_vectors_p) {
    858 		/* Map the vector page. */
    859 		pmap_map_entry(l1pt_va, systempage.pv_va, systempage.pv_pa,
    860 		    VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE, PTE_CACHE);
    861 	}
    862 
    863 	/* Map the Mini-Data cache clean area. */
    864 #if ARM_MMU_XSCALE == 1
    865 #if (ARM_NMMUS > 1)
    866 	if (xscale_use_minidata)
    867 #endif
    868 		xscale_setup_minidata(l1pt_va, minidataclean.pv_va,
    869 		    minidataclean.pv_pa);
    870 #endif
    871 
    872 	/*
    873 	 * Map integrated peripherals at same address in first level page
    874 	 * table so that we can continue to use console.
    875 	 */
    876 	if (devmap)
    877 		pmap_devmap_bootstrap(l1pt_va, devmap);
    878 
    879 	/* Tell the user about where all the bits and pieces live. */
    880 	VPRINTF("%22s       Physical              Virtual        Num\n", " ");
    881 	VPRINTF("%22s Starting    Ending    Starting    Ending   Pages\n", " ");
    882 
    883 #ifdef VERBOSE_INIT_ARM
    884 	static const char mem_fmt[] =
    885 	    "%20s: 0x%08lx 0x%08lx 0x%08lx 0x%08lx %u\n";
    886 	static const char mem_fmt_nov[] =
    887 	    "%20s: 0x%08lx 0x%08lx                       %zu\n";
    888 #endif
    889 
    890 #if 0
    891 	// XXX Doesn't make sense if kernel not at bottom of RAM
    892 	VPRINTF(mem_fmt, "SDRAM", bmi->bmi_start, bmi->bmi_end - 1,
    893 	    KERN_PHYSTOV(bmi->bmi_start), KERN_PHYSTOV(bmi->bmi_end - 1),
    894 	    (int)physmem);
    895 #endif
    896 	VPRINTF(mem_fmt, "text section",
    897 	       text.pv_pa, text.pv_pa + text.pv_size - 1,
    898 	       text.pv_va, text.pv_va + text.pv_size - 1,
    899 	       (int)(text.pv_size / PAGE_SIZE));
    900 	VPRINTF(mem_fmt, "data section",
    901 	       KERN_VTOPHYS((vaddr_t)__data_start), KERN_VTOPHYS((vaddr_t)_edata),
    902 	       (vaddr_t)__data_start, (vaddr_t)_edata,
    903 	       (int)((round_page((vaddr_t)_edata)
    904 		      - trunc_page((vaddr_t)__data_start)) / PAGE_SIZE));
    905 	VPRINTF(mem_fmt, "bss section",
    906 	       KERN_VTOPHYS((vaddr_t)__bss_start), KERN_VTOPHYS((vaddr_t)__bss_end__),
    907 	       (vaddr_t)__bss_start, (vaddr_t)__bss_end__,
    908 	       (int)((round_page((vaddr_t)__bss_end__)
    909 		      - trunc_page((vaddr_t)__bss_start)) / PAGE_SIZE));
    910 	VPRINTF(mem_fmt, "L1 page directory",
    911 	    kernel_l1pt.pv_pa, kernel_l1pt.pv_pa + L1_TABLE_SIZE - 1,
    912 	    kernel_l1pt.pv_va, kernel_l1pt.pv_va + L1_TABLE_SIZE - 1,
    913 	    L1_TABLE_SIZE / PAGE_SIZE);
    914 	VPRINTF(mem_fmt, "ABT stack (CPU 0)",
    915 	    abtstack.pv_pa, abtstack.pv_pa + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
    916 	    abtstack.pv_va, abtstack.pv_va + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
    917 	    ABT_STACK_SIZE);
    918 	VPRINTF(mem_fmt, "FIQ stack (CPU 0)",
    919 	    fiqstack.pv_pa, fiqstack.pv_pa + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
    920 	    fiqstack.pv_va, fiqstack.pv_va + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
    921 	    FIQ_STACK_SIZE);
    922 	VPRINTF(mem_fmt, "IRQ stack (CPU 0)",
    923 	    irqstack.pv_pa, irqstack.pv_pa + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
    924 	    irqstack.pv_va, irqstack.pv_va + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
    925 	    IRQ_STACK_SIZE);
    926 	VPRINTF(mem_fmt, "UND stack (CPU 0)",
    927 	    undstack.pv_pa, undstack.pv_pa + (UND_STACK_SIZE * PAGE_SIZE) - 1,
    928 	    undstack.pv_va, undstack.pv_va + (UND_STACK_SIZE * PAGE_SIZE) - 1,
    929 	    UND_STACK_SIZE);
    930 	VPRINTF(mem_fmt, "IDLE stack (CPU 0)",
    931 	    idlestack.pv_pa, idlestack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
    932 	    idlestack.pv_va, idlestack.pv_va + (UPAGES * PAGE_SIZE) - 1,
    933 	    UPAGES);
    934 	VPRINTF(mem_fmt, "SVC stack",
    935 	    kernelstack.pv_pa, kernelstack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
    936 	    kernelstack.pv_va, kernelstack.pv_va + (UPAGES * PAGE_SIZE) - 1,
    937 	    UPAGES);
    938 	VPRINTF(mem_fmt, "Message Buffer",
    939 	    msgbuf.pv_pa, msgbuf.pv_pa + (msgbuf_pgs * PAGE_SIZE) - 1,
    940 	    msgbuf.pv_va, msgbuf.pv_va + (msgbuf_pgs * PAGE_SIZE) - 1,
    941 	    (int)msgbuf_pgs);
    942 	if (map_vectors_p) {
    943 		VPRINTF(mem_fmt, "Exception Vectors",
    944 		    systempage.pv_pa, systempage.pv_pa + PAGE_SIZE - 1,
    945 		    systempage.pv_va, systempage.pv_va + PAGE_SIZE - 1,
    946 		    1);
    947 	}
    948 	for (size_t i = 0; i < bmi->bmi_nfreeblocks; i++) {
    949 		pv = &bmi->bmi_freeblocks[i];
    950 
    951 		VPRINTF(mem_fmt_nov, "Free Memory",
    952 		    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
    953 		    pv->pv_size / PAGE_SIZE);
    954 	}
    955 	/*
    956 	 * Now we have the real page tables in place so we can switch to them.
    957 	 * Once this is done we will be running with the REAL kernel page
    958 	 * tables.
    959 	 */
    960 
    961 	VPRINTF("TTBR0=%#x", armreg_ttbr_read());
    962 #ifdef _ARM_ARCH_6
    963 	VPRINTF(" TTBR1=%#x TTBCR=%#x CONTEXTIDR=%#x",
    964 	    armreg_ttbr1_read(), armreg_ttbcr_read(),
    965 	    armreg_contextidr_read());
    966 #endif
    967 	VPRINTF("\n");
    968 
    969 	/* Switch tables */
    970 	VPRINTF("switching to new L1 page table @%#lx...\n", l1pt_pa);
    971 
    972 	cpu_ttb = l1pt_pa;
    973 
    974 	cpu_domains(DOMAIN_DEFAULT);
    975 
    976 	cpu_idcache_wbinv_all();
    977 
    978 #ifdef __HAVE_GENERIC_START
    979 
    980 	/*
    981 	 * Turn on caches and set SCTLR/ACTLR
    982 	 */
    983 	cpu_setup(boot_args);
    984 #endif
    985 
    986 	VPRINTF(" ttb");
    987 
    988 #ifdef ARM_MMU_EXTENDED
    989 	/*
    990 	 * TTBCR should have been initialized by the MD start code.
    991 	 */
    992 	KASSERT((armreg_contextidr_read() & 0xff) == 0);
    993 	KASSERT(armreg_ttbcr_read() == __SHIFTIN(1, TTBCR_S_N));
    994 	/*
    995 	 * Disable lookups via TTBR0 until there is an activated pmap.
    996 	 */
    997 	armreg_ttbcr_write(armreg_ttbcr_read() | TTBCR_S_PD0);
    998 	cpu_setttb(l1pt_pa, KERNEL_PID);
    999 	arm_isb();
   1000 #else
   1001 	cpu_setttb(l1pt_pa, true);
   1002 #endif
   1003 
   1004 	cpu_tlb_flushID();
   1005 
   1006 #ifdef ARM_MMU_EXTENDED
   1007 	VPRINTF("\nsctlr=%#x actlr=%#x\n",
   1008 	    armreg_sctlr_read(), armreg_auxctl_read());
   1009 #else
   1010 	VPRINTF(" (TTBR0=%#x)", armreg_ttbr_read());
   1011 #endif
   1012 
   1013 #ifdef MULTIPROCESSOR
   1014 #ifndef __HAVE_GENERIC_START
   1015 	/*
   1016 	 * Kick the secondaries to load the TTB.  After which they'll go
   1017 	 * back to sleep to wait for the final kick so they will hatch.
   1018 	 */
   1019 	VPRINTF(" hatchlings");
   1020 	cpu_boot_secondary_processors();
   1021 #endif
   1022 #endif
   1023 
   1024 	VPRINTF(" OK\n");
   1025 }
   1026