arm32_kvminit.c revision 1.7
1/*	$NetBSD: arm32_kvminit.c,v 1.7 2012/10/15 12:26:06 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 <sys/cdefs.h>
125__KERNEL_RCSID(0, "$NetBSD: arm32_kvminit.c,v 1.7 2012/10/15 12:26:06 skrll Exp $");
126
127#include <sys/param.h>
128#include <sys/device.h>
129#include <sys/kernel.h>
130#include <sys/reboot.h>
131#include <sys/bus.h>
132
133#include <dev/cons.h>
134
135#include <uvm/uvm_extern.h>
136
137#include <arm/db_machdep.h>
138#include <arm/undefined.h>
139#include <arm/bootconfig.h>
140#include <arm/arm32/machdep.h>
141
142#include "ksyms.h"
143
144struct bootmem_info bootmem_info;
145
146paddr_t msgbufphys;
147paddr_t physical_start;
148paddr_t physical_end;
149
150extern char etext[];
151extern char __data_start[], _edata[];
152extern char __bss_start[], __bss_end__[];
153extern char _end[];
154
155/* Page tables for mapping kernel VM */
156#define KERNEL_L2PT_VMDATA_NUM	8	/* start with 32MB of KVM */
157
158/*
159 * Macros to translate between physical and virtual for a subset of the
160 * kernel address space.  *Not* for general use.
161 */
162#define KERN_VTOPHYS(bmi, va) \
163	((paddr_t)((vaddr_t)(va) - KERNEL_BASE + (bmi)->bmi_start))
164#define KERN_PHYSTOV(bmi, pa) \
165	((vaddr_t)((paddr_t)(pa) - (bmi)->bmi_start + KERNEL_BASE))
166
167void
168arm32_bootmem_init(paddr_t memstart, psize_t memsize, vsize_t kernelstart)
169{
170	struct bootmem_info * const bmi = &bootmem_info;
171	pv_addr_t *pv = bmi->bmi_freeblocks;
172
173#ifdef VERBOSE_INIT_ARM
174	printf("%s: memstart=%#lx, memsize=%#lx, kernelstart=%#lx\n",
175	    __func__, memstart, memsize, kernelstart);
176#endif
177
178	physical_start = bmi->bmi_start = memstart;
179	physical_end = bmi->bmi_end = memstart + memsize;
180	physmem = memsize / PAGE_SIZE;
181
182	/*
183	 * Let's record where the kernel lives.
184	 */
185	bmi->bmi_kernelstart = kernelstart;
186	bmi->bmi_kernelend = KERN_VTOPHYS(bmi, round_page((vaddr_t)_end));
187
188#ifdef VERBOSE_INIT_ARM
189	printf("%s: kernelend=%#lx\n", __func__, bmi->bmi_kernelend);
190#endif
191
192	/*
193	 * Now the rest of the free memory must be after the kernel.
194	 */
195	pv->pv_pa = bmi->bmi_kernelend;
196	pv->pv_va = KERN_PHYSTOV(bmi, pv->pv_pa);
197	pv->pv_size = bmi->bmi_end - bmi->bmi_kernelend;
198	bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
199#ifdef VERBOSE_INIT_ARM
200	printf("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
201	    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
202	    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
203#endif
204	pv++;
205
206	/*
207	 * Add a free block for any memory before the kernel.
208	 */
209	if (bmi->bmi_start < bmi->bmi_kernelstart) {
210		pv->pv_pa = bmi->bmi_start;
211		pv->pv_va = KERNEL_BASE;
212		pv->pv_size = bmi->bmi_kernelstart - bmi->bmi_start;
213		bmi->bmi_freepages += pv->pv_size / PAGE_SIZE;
214#ifdef VERBOSE_INIT_ARM
215		printf("%s: adding %lu free pages: [%#lx..%#lx] (VA %#lx)\n",
216		    __func__, pv->pv_size / PAGE_SIZE, pv->pv_pa,
217		    pv->pv_pa + pv->pv_size - 1, pv->pv_va);
218#endif
219		pv++;
220	}
221
222	bmi->bmi_nfreeblocks = pv - bmi->bmi_freeblocks;
223
224	SLIST_INIT(&bmi->bmi_freechunks);
225	SLIST_INIT(&bmi->bmi_chunks);
226}
227
228static bool
229concat_pvaddr(pv_addr_t *acc_pv, pv_addr_t *pv)
230{
231	if (acc_pv->pv_pa + acc_pv->pv_size == pv->pv_pa
232	    && acc_pv->pv_va + acc_pv->pv_size == pv->pv_va
233	    && acc_pv->pv_prot == pv->pv_prot
234	    && acc_pv->pv_cache == pv->pv_cache) {
235#ifdef VERBOSE_INIT_ARMX
236		printf("%s: appending pv %p (%#lx..%#lx) to %#lx..%#lx\n",
237		    __func__, pv, pv->pv_pa, pv->pv_pa + pv->pv_size + 1,
238		    acc_pv->pv_pa, acc_pv->pv_pa + acc_pv->pv_size + 1);
239#endif
240		acc_pv->pv_size += pv->pv_size;
241		return true;
242	}
243
244	return false;
245}
246
247static void
248add_pages(struct bootmem_info *bmi, pv_addr_t *pv)
249{
250	pv_addr_t **pvp = &SLIST_FIRST(&bmi->bmi_chunks);
251	while ((*pvp) != 0 && (*pvp)->pv_va <= pv->pv_va) {
252		pv_addr_t * const pv0 = (*pvp);
253		KASSERT(SLIST_NEXT(pv0, pv_list) == NULL || pv0->pv_pa < SLIST_NEXT(pv0, pv_list)->pv_pa);
254		if (concat_pvaddr(pv0, pv)) {
255#ifdef VERBOSE_INIT_ARM
256			printf("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
257			    __func__, "appending", pv,
258			    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
259			    pv0->pv_pa, pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
260#endif
261			pv = SLIST_NEXT(pv0, pv_list);
262			if (pv != NULL && concat_pvaddr(pv0, pv)) {
263#ifdef VERBOSE_INIT_ARM
264				printf("%s: %s pv %p (%#lx..%#lx) to %#lx..%#lx\n",
265				    __func__, "merging", pv,
266				    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
267				    pv0->pv_pa,
268				    pv0->pv_pa + pv0->pv_size - pv->pv_size - 1);
269#endif
270				SLIST_REMOVE_AFTER(pv0, pv_list);
271				SLIST_INSERT_HEAD(&bmi->bmi_freechunks, pv, pv_list);
272			}
273			return;
274		}
275		KASSERT(pv->pv_va != (*pvp)->pv_va);
276		pvp = &SLIST_NEXT(*pvp, pv_list);
277	}
278	KASSERT((*pvp) == NULL || pv->pv_va < (*pvp)->pv_va);
279	pv_addr_t * const new_pv = SLIST_FIRST(&bmi->bmi_freechunks);
280	KASSERT(new_pv != NULL);
281	SLIST_REMOVE_HEAD(&bmi->bmi_freechunks, pv_list);
282	*new_pv = *pv;
283	SLIST_NEXT(new_pv, pv_list) = *pvp;
284	(*pvp) = new_pv;
285#ifdef VERBOSE_INIT_ARM
286	printf("%s: adding pv %p (pa %#lx, va %#lx, %lu pages) ",
287	    __func__, new_pv, new_pv->pv_pa, new_pv->pv_va,
288	    new_pv->pv_size / PAGE_SIZE);
289	if (SLIST_NEXT(new_pv, pv_list))
290		printf("before pa %#lx\n", SLIST_NEXT(new_pv, pv_list)->pv_pa);
291	else
292		printf("at tail\n");
293#endif
294}
295
296static void
297valloc_pages(struct bootmem_info *bmi, pv_addr_t *pv, size_t npages,
298	int prot, int cache)
299{
300	size_t nbytes = npages * PAGE_SIZE;
301	pv_addr_t *free_pv = bmi->bmi_freeblocks;
302	size_t free_idx = 0;
303	static bool l1pt_found;
304
305	/*
306	 * If we haven't allocated the kernel L1 page table and we are aligned
307	 * at a L1 table boundary, alloc the memory for it.
308	 */
309	if (!l1pt_found
310	    && (free_pv->pv_pa & (L1_TABLE_SIZE - 1)) == 0
311	    && free_pv->pv_size >= L1_TABLE_SIZE) {
312		l1pt_found = true;
313		valloc_pages(bmi, &kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE,
314		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
315		add_pages(bmi, &kernel_l1pt);
316	}
317
318	while (nbytes > free_pv->pv_size) {
319		free_pv++;
320		free_idx++;
321		if (free_idx == bmi->bmi_nfreeblocks) {
322			panic("%s: could not allocate %zu bytes",
323			    __func__, nbytes);
324		}
325	}
326
327	pv->pv_pa = free_pv->pv_pa;
328	pv->pv_va = free_pv->pv_va;
329	pv->pv_size = nbytes;
330	pv->pv_prot = prot;
331	pv->pv_cache = cache;
332
333	/*
334	 * If PTE_PAGETABLE uses the same cache modes as PTE_CACHE
335	 * just use PTE_CACHE.
336	 */
337	if (cache == PTE_PAGETABLE
338	    && pte_l1_s_cache_mode == pte_l1_s_cache_mode_pt
339	    && pte_l2_l_cache_mode == pte_l2_l_cache_mode_pt
340	    && pte_l2_s_cache_mode == pte_l2_s_cache_mode_pt)
341		pv->pv_cache = PTE_CACHE;
342
343	free_pv->pv_pa += nbytes;
344	free_pv->pv_va += nbytes;
345	free_pv->pv_size -= nbytes;
346	if (free_pv->pv_size == 0) {
347		--bmi->bmi_nfreeblocks;
348		for (; free_idx < bmi->bmi_nfreeblocks; free_idx++) {
349			free_pv[0] = free_pv[1];
350		}
351	}
352
353	bmi->bmi_freepages -= npages;
354
355	memset((void *)pv->pv_va, 0, nbytes);
356}
357
358void
359arm32_kernel_vm_init(vaddr_t kernel_vm_base, vaddr_t vectors, vaddr_t iovbase,
360	const struct pmap_devmap *devmap, bool mapallmem_p)
361{
362	struct bootmem_info * const bmi = &bootmem_info;
363#ifdef MULTIPROCESSOR
364	const size_t cpu_num = arm_cpu_max + 1;
365#else
366	const size_t cpu_num = 1;
367#endif
368
369	/*
370	 * Calculate the number of L2 pages needed for mapping the
371	 * kernel + data + stuff
372	 */
373	size_t kernel_size = bmi->bmi_kernelend;
374	kernel_size -= (bmi->bmi_kernelstart & -L2_S_SEGSIZE);
375	kernel_size += L1_TABLE_SIZE;
376	kernel_size += round_page(MSGBUFSIZE);
377	kernel_size +=
378	    cpu_num * (ABT_STACK_SIZE + FIQ_STACK_SIZE + IRQ_STACK_SIZE
379	    + UND_STACK_SIZE + UPAGES) * PAGE_SIZE;
380	kernel_size += 0x10000;	/* slop */
381	kernel_size += (kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE;
382	kernel_size = round_page(kernel_size);
383
384	/*
385	 * Now we know how many L2 pages it will take.
386	 */
387	const size_t KERNEL_L2PT_KERNEL_NUM =
388	    (kernel_size + L2_S_SEGSIZE - 1) / L2_S_SEGSIZE;
389
390#ifdef VERBOSE_INIT_ARM
391	printf("%s: %zu L2 pages are needed to map %#zx kernel bytes\n",
392	    __func__, KERNEL_L2PT_KERNEL_NUM, kernel_size);
393#endif
394
395	KASSERT(KERNEL_L2PT_KERNEL_NUM + KERNEL_L2PT_VMDATA_NUM < __arraycount(bmi->bmi_l2pts));
396	pv_addr_t * const kernel_l2pt = bmi->bmi_l2pts;
397	pv_addr_t * const vmdata_l2pt = kernel_l2pt + KERNEL_L2PT_KERNEL_NUM;
398	pv_addr_t msgbuf;
399	pv_addr_t text;
400	pv_addr_t data;
401	pv_addr_t chunks[KERNEL_L2PT_KERNEL_NUM+KERNEL_L2PT_VMDATA_NUM+11];
402#if ARM_MMU_XSCALE == 1
403	pv_addr_t minidataclean;
404#endif
405
406	/*
407	 * We need to allocate some fixed page tables to get the kernel going.
408	 *
409	 * We are going to allocate our bootstrap pages from the beginning of
410	 * the free space that we just calculated.  We allocate one page
411	 * directory and a number of page tables and store the physical
412	 * addresses in the kernel_l2pt_table array.
413	 *
414	 * The kernel page directory must be on a 16K boundary.  The page
415	 * tables must be on 4K boundaries.  What we do is allocate the
416	 * page directory on the first 16K boundary that we encounter, and
417	 * the page tables on 4K boundaries otherwise.  Since we allocate
418	 * at least 3 L2 page tables, we are guaranteed to encounter at
419	 * least one 16K aligned region.
420	 */
421
422#ifdef VERBOSE_INIT_ARM
423	printf("%s: allocating page tables for", __func__);
424#endif
425	for (size_t i = 0; i < __arraycount(chunks); i++) {
426		SLIST_INSERT_HEAD(&bmi->bmi_freechunks, &chunks[i], pv_list);
427	}
428
429	/*
430	 * As we allocate the memory, make sure that we don't walk over
431	 * our temporary first level translation table.
432	 */
433
434	kernel_l1pt.pv_pa = 0;
435	kernel_l1pt.pv_va = 0;
436
437	/*
438	 * First allocate L2 page for the vectors.
439	 */
440#ifdef VERBOSE_INIT_ARM
441	printf(" vector");
442#endif
443	valloc_pages(bmi, &bmi->bmi_vector_l2pt, L2_TABLE_SIZE / PAGE_SIZE,
444	    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
445	add_pages(bmi, &bmi->bmi_vector_l2pt);
446
447	/*
448	 * Allocate the L2 pages, but if we get to a page that aligned for a
449	 * L1 page table, we will allocate pages for it first and allocate
450	 * L2 page.
451	 */
452#ifdef VERBOSE_INIT_ARM
453	printf(" kernel");
454#endif
455	for (size_t idx = 0; idx <= KERNEL_L2PT_KERNEL_NUM; ++idx) {
456		valloc_pages(bmi, &kernel_l2pt[idx], L2_TABLE_SIZE / PAGE_SIZE,
457		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
458		add_pages(bmi, &kernel_l2pt[idx]);
459	}
460#ifdef VERBOSE_INIT_ARM
461	printf(" vm");
462#endif
463	for (size_t idx = 0; idx <= KERNEL_L2PT_VMDATA_NUM; ++idx) {
464		valloc_pages(bmi, &vmdata_l2pt[idx], L2_TABLE_SIZE / PAGE_SIZE,
465		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
466		add_pages(bmi, &vmdata_l2pt[idx]);
467	}
468
469	/*
470	 * If someone wanted a L2 page for I/O, allocate it now.
471	 */
472	if (iovbase != 0) {
473#ifdef VERBOSE_INIT_ARM
474		printf(" io");
475#endif
476		valloc_pages(bmi, &bmi->bmi_io_l2pt, L2_TABLE_SIZE / PAGE_SIZE,
477		    VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
478		add_pages(bmi, &bmi->bmi_io_l2pt);
479	}
480
481#ifdef VERBOSE_ARM_INIT
482	printf("%s: allocating stacks\n", __func__);
483#endif
484
485	/* Allocate stacks for all modes */
486	valloc_pages(bmi, &abtstack, ABT_STACK_SIZE * cpu_num,
487	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
488	add_pages(bmi, &abtstack);
489	valloc_pages(bmi, &fiqstack, FIQ_STACK_SIZE * cpu_num,
490	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
491	add_pages(bmi, &fiqstack);
492	valloc_pages(bmi, &irqstack, IRQ_STACK_SIZE * cpu_num,
493	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
494	add_pages(bmi, &irqstack);
495	valloc_pages(bmi, &undstack, UND_STACK_SIZE * cpu_num,
496	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
497	add_pages(bmi, &undstack);
498	valloc_pages(bmi, &idlestack, UPAGES * cpu_num,		/* SVC32 */
499	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
500	add_pages(bmi, &idlestack);
501	valloc_pages(bmi, &kernelstack, UPAGES,			/* SVC32 */
502	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
503	add_pages(bmi, &kernelstack);
504
505	/* Allocate the message buffer from the end of memory. */
506	const size_t msgbuf_pgs = round_page(MSGBUFSIZE) / PAGE_SIZE;
507	valloc_pages(bmi, &msgbuf, msgbuf_pgs,
508	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
509	add_pages(bmi, &msgbuf);
510	msgbufphys = msgbuf.pv_pa;
511
512	/*
513	 * Allocate a page for the system vector page.
514	 * This page will just contain the system vectors and can be
515	 * shared by all processes.
516	 */
517	valloc_pages(bmi, &systempage, 1, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
518	systempage.pv_va = vectors;
519
520	/*
521	 * If the caller needed a few extra pages for some reason, allocate
522	 * them now.
523	 */
524#if ARM_MMU_XSCALE == 1
525#if (ARM_NMMUS > 1)
526	if (xscale_use_minidata)
527#endif
528		valloc_pages(bmi, extrapv, nextrapages,
529		    VM_PROT_READ|VM_PROT_WRITE, 0);
530#endif
531
532	/*
533	 * Ok we have allocated physical pages for the primary kernel
534	 * page tables and stacks.  Let's just confirm that.
535	 */
536	if (kernel_l1pt.pv_va == 0
537	    && (!kernel_l1pt.pv_pa || (kernel_l1pt.pv_pa & (L1_TABLE_SIZE - 1)) != 0))
538		panic("%s: Failed to allocate or align the kernel "
539		    "page directory", __func__);
540
541
542#ifdef VERBOSE_INIT_ARM
543	printf("Creating L1 page table at 0x%08lx\n", kernel_l1pt.pv_pa);
544#endif
545
546	/*
547	 * Now we start construction of the L1 page table
548	 * We start by mapping the L2 page tables into the L1.
549	 * This means that we can replace L1 mappings later on if necessary
550	 */
551	vaddr_t l1pt_va = kernel_l1pt.pv_va;
552	paddr_t l1pt_pa = kernel_l1pt.pv_pa;
553
554	/* Map the L2 pages tables in the L1 page table */
555	pmap_link_l2pt(l1pt_va, systempage.pv_va & -L2_S_SEGSIZE,
556	    &bmi->bmi_vector_l2pt);
557#ifdef VERBOSE_INIT_ARM
558	printf("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx\n",
559	    __func__, bmi->bmi_vector_l2pt.pv_va, bmi->bmi_vector_l2pt.pv_pa,
560	    systempage.pv_va);
561#endif
562
563	const vaddr_t kernel_base =
564	    KERN_PHYSTOV(bmi, bmi->bmi_kernelstart & -L2_S_SEGSIZE);
565	for (size_t idx = 0; idx < KERNEL_L2PT_KERNEL_NUM; idx++) {
566		pmap_link_l2pt(l1pt_va, kernel_base + idx * L2_S_SEGSIZE,
567		    &kernel_l2pt[idx]);
568#ifdef VERBOSE_INIT_ARM
569		printf("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (kernel)\n",
570		    __func__, kernel_l2pt[idx].pv_va, kernel_l2pt[idx].pv_pa,
571		    kernel_base + idx * L2_S_SEGSIZE);
572#endif
573	}
574
575	for (size_t idx = 0; idx < KERNEL_L2PT_VMDATA_NUM; idx++) {
576		pmap_link_l2pt(l1pt_va, kernel_vm_base + idx * L2_S_SEGSIZE,
577		    &vmdata_l2pt[idx]);
578#ifdef VERBOSE_INIT_ARM
579		printf("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (vm)\n",
580		    __func__, vmdata_l2pt[idx].pv_va, vmdata_l2pt[idx].pv_pa,
581		    kernel_vm_base + idx * L2_S_SEGSIZE);
582#endif
583	}
584	if (iovbase) {
585		pmap_link_l2pt(l1pt_va, iovbase & -L2_S_SEGSIZE, &bmi->bmi_io_l2pt);
586#ifdef VERBOSE_INIT_ARM
587		printf("%s: adding L2 pt (VA %#lx, PA %#lx) for VA %#lx (io)\n",
588		    __func__, bmi->bmi_io_l2pt.pv_va, bmi->bmi_io_l2pt.pv_pa,
589		    iovbase & -L2_S_SEGSIZE);
590#endif
591	}
592
593	/* update the top of the kernel VM */
594	pmap_curmaxkvaddr =
595	    kernel_vm_base + (KERNEL_L2PT_VMDATA_NUM * L2_S_SEGSIZE);
596
597#ifdef VERBOSE_INIT_ARM
598	printf("Mapping kernel\n");
599#endif
600
601	extern char etext[], _end[];
602	size_t totalsize = bmi->bmi_kernelend - bmi->bmi_kernelstart;
603	size_t textsize = KERN_VTOPHYS(bmi, (uintptr_t)etext) - bmi->bmi_kernelstart;
604
605	textsize = (textsize + PGOFSET) & ~PGOFSET;
606
607	/* start at offset of kernel in RAM */
608
609	text.pv_pa = bmi->bmi_kernelstart;
610	text.pv_va = KERN_PHYSTOV(bmi, bmi->bmi_kernelstart);
611	text.pv_size = textsize;
612	text.pv_prot = VM_PROT_READ|VM_PROT_WRITE; /* XXX VM_PROT_EXECUTE */
613	text.pv_cache = PTE_CACHE;
614
615#ifdef VERBOSE_INIT_ARM
616	printf("%s: adding chunk for kernel text %#lx..%#lx (VA %#lx)\n",
617	    __func__, text.pv_pa, text.pv_pa + text.pv_size - 1, text.pv_va);
618#endif
619
620	add_pages(bmi, &text);
621
622	data.pv_pa = text.pv_pa + textsize;
623	data.pv_va = text.pv_va + textsize;
624	data.pv_size = totalsize - textsize;
625	data.pv_prot = VM_PROT_READ|VM_PROT_WRITE;
626	data.pv_cache = PTE_CACHE;
627
628#ifdef VERBOSE_INIT_ARM
629	printf("%s: adding chunk for kernel data/bss %#lx..%#lx (VA %#lx)\n",
630	    __func__, data.pv_pa, data.pv_pa + data.pv_size - 1, data.pv_va);
631#endif
632
633	add_pages(bmi, &data);
634
635#ifdef VERBOSE_INIT_ARM
636	printf("Listing Chunks\n");
637	{
638		pv_addr_t *pv;
639		SLIST_FOREACH(pv, &bmi->bmi_chunks, pv_list) {
640			printf("%s: pv %p: chunk VA %#lx..%#lx "
641			    "(PA %#lx, prot %d, cache %d)\n",
642			    __func__, pv, pv->pv_va, pv->pv_va + pv->pv_size - 1,
643			    pv->pv_pa, pv->pv_prot, pv->pv_cache);
644		}
645	}
646	printf("\nMapping Chunks\n");
647#endif
648
649	pv_addr_t cur_pv;
650	pv_addr_t *pv = SLIST_FIRST(&bmi->bmi_chunks);
651	if (!mapallmem_p || pv->pv_pa == bmi->bmi_start) {
652		cur_pv = *pv;
653		pv = SLIST_NEXT(pv, pv_list);
654	} else {
655		cur_pv.pv_va = kernel_base;
656		cur_pv.pv_pa = bmi->bmi_start;
657		cur_pv.pv_size = pv->pv_pa - bmi->bmi_start;
658		cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
659		cur_pv.pv_cache = PTE_CACHE;
660	}
661	while (pv != NULL) {
662		if (mapallmem_p) {
663			if (concat_pvaddr(&cur_pv, pv)) {
664				pv = SLIST_NEXT(pv, pv_list);
665				continue;
666			}
667			if (cur_pv.pv_pa + cur_pv.pv_size < pv->pv_pa) {
668				/*
669				 * See if we can extend the current pv to emcompass the
670				 * hole, and if so do it and retry the concatenation.
671				 */
672				if (cur_pv.pv_prot == (VM_PROT_READ|VM_PROT_WRITE)
673				    && cur_pv.pv_cache == PTE_CACHE) {
674					cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
675					continue;
676				}
677
678				/*
679				 * We couldn't so emit the current chunk and then
680				 */
681#ifdef VERBOSE_INIT_ARM
682				printf("%s: mapping chunk VA %#lx..%#lx "
683				    "(PA %#lx, prot %d, cache %d)\n",
684				    __func__,
685				    cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
686				    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
687#endif
688				pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
689				    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
690
691				/*
692				 * set the current chunk to the hole and try again.
693				 */
694				cur_pv.pv_pa += cur_pv.pv_size;
695				cur_pv.pv_va += cur_pv.pv_size;
696				cur_pv.pv_size = pv->pv_pa - cur_pv.pv_va;
697				cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
698				cur_pv.pv_cache = PTE_CACHE;
699				continue;
700			}
701		}
702
703		/*
704		 * The new pv didn't concatenate so emit the current one
705		 * and use the new pv as the current pv.
706		 */
707#ifdef VERBOSE_INIT_ARM
708		printf("%s: mapping chunk VA %#lx..%#lx "
709		    "(PA %#lx, prot %d, cache %d)\n",
710		    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
711		    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
712#endif
713		pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
714		    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
715		cur_pv = *pv;
716		pv = SLIST_NEXT(pv, pv_list);
717	}
718
719	/*
720	 * If we are mapping all of memory, let's map the rest of memory.
721	 */
722	if (mapallmem_p && cur_pv.pv_pa + cur_pv.pv_size < bmi->bmi_end) {
723		if (cur_pv.pv_prot == (VM_PROT_READ | VM_PROT_WRITE)
724		    && cur_pv.pv_cache == PTE_CACHE) {
725			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
726		} else {
727#ifdef VERBOSE_INIT_ARM
728			printf("%s: mapping chunk VA %#lx..%#lx "
729			    "(PA %#lx, prot %d, cache %d)\n",
730			    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
731			    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
732#endif
733			pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
734			    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
735			cur_pv.pv_pa += cur_pv.pv_size;
736			cur_pv.pv_va += cur_pv.pv_size;
737			cur_pv.pv_size = bmi->bmi_end - cur_pv.pv_pa;
738			cur_pv.pv_prot = VM_PROT_READ | VM_PROT_WRITE;
739			cur_pv.pv_cache = PTE_CACHE;
740		}
741	}
742
743	/*
744	 * Now we map the final chunk.
745	 */
746#ifdef VERBOSE_INIT_ARM
747	printf("%s: mapping last chunk VA %#lx..%#lx (PA %#lx, prot %d, cache %d)\n",
748	    __func__, cur_pv.pv_va, cur_pv.pv_va + cur_pv.pv_size - 1,
749	    cur_pv.pv_pa, cur_pv.pv_prot, cur_pv.pv_cache);
750#endif
751	pmap_map_chunk(l1pt_va, cur_pv.pv_va, cur_pv.pv_pa,
752	    cur_pv.pv_size, cur_pv.pv_prot, cur_pv.pv_cache);
753
754	/*
755	 * Now we map the stuff that isn't directly after the kernel
756	 */
757
758	/* Map the vector page. */
759	pmap_map_entry(l1pt_va, systempage.pv_va, systempage.pv_pa,
760	    VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
761
762	/* Map the Mini-Data cache clean area. */
763#if ARM_MMU_XSCALE == 1
764#if (ARM_NMMUS > 1)
765	if (xscale_use_minidata)
766#endif
767		xscale_setup_minidata(l1_va, minidataclean.pv_va,
768		    minidataclean.pv_pa);
769#endif
770
771	/*
772	 * Map integrated peripherals at same address in first level page
773	 * table so that we can continue to use console.
774	 */
775	if (devmap)
776		pmap_devmap_bootstrap(l1pt_va, devmap);
777
778#ifdef VERBOSE_INIT_ARM
779	/* Tell the user about where all the bits and pieces live. */
780	printf("%22s       Physical              Virtual        Num\n", " ");
781	printf("%22s Starting    Ending    Starting    Ending   Pages\n", " ");
782
783	static const char mem_fmt[] =
784	    "%20s: 0x%08lx 0x%08lx 0x%08lx 0x%08lx %u\n";
785	static const char mem_fmt_nov[] =
786	    "%20s: 0x%08lx 0x%08lx                       %zu\n";
787
788	printf(mem_fmt, "SDRAM", bmi->bmi_start, bmi->bmi_end - 1,
789	    KERN_PHYSTOV(bmi, bmi->bmi_start), KERN_PHYSTOV(bmi, bmi->bmi_end - 1),
790	    physmem);
791	printf(mem_fmt, "text section",
792	       text.pv_pa, text.pv_pa + text.pv_size - 1,
793	       text.pv_va, text.pv_va + text.pv_size - 1,
794	       (int)(text.pv_size / PAGE_SIZE));
795	printf(mem_fmt, "data section",
796	       KERN_VTOPHYS(bmi, __data_start), KERN_VTOPHYS(bmi, _edata),
797	       (vaddr_t)__data_start, (vaddr_t)_edata,
798	       (int)((round_page((vaddr_t)_edata)
799		      - trunc_page((vaddr_t)__data_start)) / PAGE_SIZE));
800	printf(mem_fmt, "bss section",
801	       KERN_VTOPHYS(bmi, __bss_start), KERN_VTOPHYS(bmi, __bss_end__),
802	       (vaddr_t)__bss_start, (vaddr_t)__bss_end__,
803	       (int)((round_page((vaddr_t)__bss_end__)
804		      - trunc_page((vaddr_t)__bss_start)) / PAGE_SIZE));
805	printf(mem_fmt, "L1 page directory",
806	    kernel_l1pt.pv_pa, kernel_l1pt.pv_pa + L1_TABLE_SIZE - 1,
807	    kernel_l1pt.pv_va, kernel_l1pt.pv_va + L1_TABLE_SIZE - 1,
808	    L1_TABLE_SIZE / PAGE_SIZE);
809	printf(mem_fmt, "ABT stack (CPU 0)",
810	    abtstack.pv_pa, abtstack.pv_pa + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
811	    abtstack.pv_va, abtstack.pv_va + (ABT_STACK_SIZE * PAGE_SIZE) - 1,
812	    ABT_STACK_SIZE);
813	printf(mem_fmt, "FIQ stack (CPU 0)",
814	    fiqstack.pv_pa, fiqstack.pv_pa + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
815	    fiqstack.pv_va, fiqstack.pv_va + (FIQ_STACK_SIZE * PAGE_SIZE) - 1,
816	    FIQ_STACK_SIZE);
817	printf(mem_fmt, "IRQ stack (CPU 0)",
818	    irqstack.pv_pa, irqstack.pv_pa + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
819	    irqstack.pv_va, irqstack.pv_va + (IRQ_STACK_SIZE * PAGE_SIZE) - 1,
820	    IRQ_STACK_SIZE);
821	printf(mem_fmt, "UND stack (CPU 0)",
822	    undstack.pv_pa, undstack.pv_pa + (UND_STACK_SIZE * PAGE_SIZE) - 1,
823	    undstack.pv_va, undstack.pv_va + (UND_STACK_SIZE * PAGE_SIZE) - 1,
824	    UND_STACK_SIZE);
825	printf(mem_fmt, "IDLE stack (CPU 0)",
826	    idlestack.pv_pa, idlestack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
827	    idlestack.pv_va, idlestack.pv_va + (UPAGES * PAGE_SIZE) - 1,
828	    UPAGES);
829	printf(mem_fmt, "SVC stack",
830	    kernelstack.pv_pa, kernelstack.pv_pa + (UPAGES * PAGE_SIZE) - 1,
831	    kernelstack.pv_va, kernelstack.pv_va + (UPAGES * PAGE_SIZE) - 1,
832	    UPAGES);
833	printf(mem_fmt_nov, "Message Buffer",
834	    msgbufphys, msgbufphys + msgbuf_pgs * PAGE_SIZE - 1, msgbuf_pgs);
835	printf(mem_fmt, "Exception Vectors",
836	    systempage.pv_pa, systempage.pv_pa + PAGE_SIZE - 1,
837	    systempage.pv_va, systempage.pv_va + PAGE_SIZE - 1,
838	    1);
839	for (size_t i = 0; i < bmi->bmi_nfreeblocks; i++) {
840		pv = &bmi->bmi_freeblocks[i];
841
842		printf(mem_fmt_nov, "Free Memory",
843		    pv->pv_pa, pv->pv_pa + pv->pv_size - 1,
844		    pv->pv_size / PAGE_SIZE);
845	}
846#endif
847	/*
848	 * Now we have the real page tables in place so we can switch to them.
849	 * Once this is done we will be running with the REAL kernel page
850	 * tables.
851	 */
852
853#if defined(VERBOSE_INIT_ARM) && 0
854	printf("TTBR0=%#x", armreg_ttbr_read());
855#ifdef _ARM_ARCH_6
856	printf(" TTBR1=%#x TTBCR=%#x",
857	    armreg_ttbr1_read(), armreg_ttbcr_read());
858#endif
859	printf("\n");
860#endif
861
862	/* Switch tables */
863#ifdef VERBOSE_INIT_ARM
864	printf("switching to new L1 page table @%#lx...", l1pt_pa);
865#endif
866
867	cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT);
868	cpu_idcache_wbinv_all();
869	cpu_setttb(l1pt_pa, true);
870	cpu_tlb_flushID();
871	cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2));
872
873#ifdef VERBOSE_INIT_ARM
874	printf("TTBR0=%#x OK", armreg_ttbr_read());
875#endif
876}
877