1 /*
2  * This file contains the routines setting up the linux page tables.
3  *  -- paulus
4  *
5  *  Derived from arch/ppc/mm/init.c:
6  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
7  *
8  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
9  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
10  *    Copyright (C) 1996 Paul Mackerras
11  *
12  *  Derived from "arch/i386/mm/init.c"
13  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
14  *
15  *  This program is free software; you can redistribute it and/or
16  *  modify it under the terms of the GNU General Public License
17  *  as published by the Free Software Foundation; either version
18  *  2 of the License, or (at your option) any later version.
19  *
20  */
21 
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/types.h>
25 #include <linux/mm.h>
26 #include <linux/vmalloc.h>
27 #include <linux/init.h>
28 #include <linux/highmem.h>
29 #include <linux/memblock.h>
30 #include <linux/slab.h>
31 
32 #include <asm/pgtable.h>
33 #include <asm/pgalloc.h>
34 #include <asm/fixmap.h>
35 #include <asm/io.h>
36 #include <asm/setup.h>
37 
38 #include "mmu_decl.h"
39 
40 unsigned long ioremap_base;
41 unsigned long ioremap_bot;
42 EXPORT_SYMBOL(ioremap_bot);	/* aka VMALLOC_END */
43 
44 #if defined(CONFIG_6xx) || defined(CONFIG_POWER3)
45 #define HAVE_BATS	1
46 #endif
47 
48 #if defined(CONFIG_FSL_BOOKE)
49 #define HAVE_TLBCAM	1
50 #endif
51 
52 extern char etext[], _stext[];
53 
54 #ifdef HAVE_BATS
55 extern phys_addr_t v_mapped_by_bats(unsigned long va);
56 extern unsigned long p_mapped_by_bats(phys_addr_t pa);
57 void setbat(int index, unsigned long virt, phys_addr_t phys,
58 	    unsigned int size, int flags);
59 
60 #else /* !HAVE_BATS */
61 #define v_mapped_by_bats(x)	(0UL)
62 #define p_mapped_by_bats(x)	(0UL)
63 #endif /* HAVE_BATS */
64 
65 #ifdef HAVE_TLBCAM
66 extern unsigned int tlbcam_index;
67 extern phys_addr_t v_mapped_by_tlbcam(unsigned long va);
68 extern unsigned long p_mapped_by_tlbcam(phys_addr_t pa);
69 #else /* !HAVE_TLBCAM */
70 #define v_mapped_by_tlbcam(x)	(0UL)
71 #define p_mapped_by_tlbcam(x)	(0UL)
72 #endif /* HAVE_TLBCAM */
73 
74 #define PGDIR_ORDER	(32 + PGD_T_LOG2 - PGDIR_SHIFT)
75 
pgd_alloc(struct mm_struct * mm)76 pgd_t *pgd_alloc(struct mm_struct *mm)
77 {
78 	pgd_t *ret;
79 
80 	/* pgdir take page or two with 4K pages and a page fraction otherwise */
81 #ifndef CONFIG_PPC_4K_PAGES
82 	ret = kzalloc(1 << PGDIR_ORDER, GFP_KERNEL);
83 #else
84 	ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
85 			PGDIR_ORDER - PAGE_SHIFT);
86 #endif
87 	return ret;
88 }
89 
pgd_free(struct mm_struct * mm,pgd_t * pgd)90 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
91 {
92 #ifndef CONFIG_PPC_4K_PAGES
93 	kfree((void *)pgd);
94 #else
95 	free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT);
96 #endif
97 }
98 
pte_alloc_one_kernel(struct mm_struct * mm,unsigned long address)99 __init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
100 {
101 	pte_t *pte;
102 	extern int mem_init_done;
103 	extern void *early_get_page(void);
104 
105 	if (mem_init_done) {
106 		pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
107 	} else {
108 		pte = (pte_t *)early_get_page();
109 		if (pte)
110 			clear_page(pte);
111 	}
112 	return pte;
113 }
114 
pte_alloc_one(struct mm_struct * mm,unsigned long address)115 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
116 {
117 	struct page *ptepage;
118 
119 	gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO;
120 
121 	ptepage = alloc_pages(flags, 0);
122 	if (!ptepage)
123 		return NULL;
124 	pgtable_page_ctor(ptepage);
125 	return ptepage;
126 }
127 
128 void __iomem *
ioremap(phys_addr_t addr,unsigned long size)129 ioremap(phys_addr_t addr, unsigned long size)
130 {
131 	return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
132 				__builtin_return_address(0));
133 }
134 EXPORT_SYMBOL(ioremap);
135 
136 void __iomem *
ioremap_wc(phys_addr_t addr,unsigned long size)137 ioremap_wc(phys_addr_t addr, unsigned long size)
138 {
139 	return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
140 				__builtin_return_address(0));
141 }
142 EXPORT_SYMBOL(ioremap_wc);
143 
144 void __iomem *
ioremap_prot(phys_addr_t addr,unsigned long size,unsigned long flags)145 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
146 {
147 	/* writeable implies dirty for kernel addresses */
148 	if (flags & _PAGE_RW)
149 		flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
150 
151 	/* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
152 	flags &= ~(_PAGE_USER | _PAGE_EXEC);
153 
154 #ifdef _PAGE_BAP_SR
155 	/* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
156 	 * which means that we just cleared supervisor access... oops ;-) This
157 	 * restores it
158 	 */
159 	flags |= _PAGE_BAP_SR;
160 #endif
161 
162 	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
163 }
164 EXPORT_SYMBOL(ioremap_prot);
165 
166 void __iomem *
__ioremap(phys_addr_t addr,unsigned long size,unsigned long flags)167 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
168 {
169 	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
170 }
171 
172 void __iomem *
__ioremap_caller(phys_addr_t addr,unsigned long size,unsigned long flags,void * caller)173 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
174 		 void *caller)
175 {
176 	unsigned long v, i;
177 	phys_addr_t p;
178 	int err;
179 
180 	/* Make sure we have the base flags */
181 	if ((flags & _PAGE_PRESENT) == 0)
182 		flags |= PAGE_KERNEL;
183 
184 	/* Non-cacheable page cannot be coherent */
185 	if (flags & _PAGE_NO_CACHE)
186 		flags &= ~_PAGE_COHERENT;
187 
188 	/*
189 	 * Choose an address to map it to.
190 	 * Once the vmalloc system is running, we use it.
191 	 * Before then, we use space going down from ioremap_base
192 	 * (ioremap_bot records where we're up to).
193 	 */
194 	p = addr & PAGE_MASK;
195 	size = PAGE_ALIGN(addr + size) - p;
196 
197 	/*
198 	 * If the address lies within the first 16 MB, assume it's in ISA
199 	 * memory space
200 	 */
201 	if (p < 16*1024*1024)
202 		p += _ISA_MEM_BASE;
203 
204 #ifndef CONFIG_CRASH_DUMP
205 	/*
206 	 * Don't allow anybody to remap normal RAM that we're using.
207 	 * mem_init() sets high_memory so only do the check after that.
208 	 */
209 	if (mem_init_done && (p < virt_to_phys(high_memory)) &&
210 	    !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
211 		printk("__ioremap(): phys addr 0x%llx is RAM lr %pf\n",
212 		       (unsigned long long)p, __builtin_return_address(0));
213 		return NULL;
214 	}
215 #endif
216 
217 	if (size == 0)
218 		return NULL;
219 
220 	/*
221 	 * Is it already mapped?  Perhaps overlapped by a previous
222 	 * BAT mapping.  If the whole area is mapped then we're done,
223 	 * otherwise remap it since we want to keep the virt addrs for
224 	 * each request contiguous.
225 	 *
226 	 * We make the assumption here that if the bottom and top
227 	 * of the range we want are mapped then it's mapped to the
228 	 * same virt address (and this is contiguous).
229 	 *  -- Cort
230 	 */
231 	if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
232 		goto out;
233 
234 	if ((v = p_mapped_by_tlbcam(p)))
235 		goto out;
236 
237 	if (mem_init_done) {
238 		struct vm_struct *area;
239 		area = get_vm_area_caller(size, VM_IOREMAP, caller);
240 		if (area == 0)
241 			return NULL;
242 		area->phys_addr = p;
243 		v = (unsigned long) area->addr;
244 	} else {
245 		v = (ioremap_bot -= size);
246 	}
247 
248 	/*
249 	 * Should check if it is a candidate for a BAT mapping
250 	 */
251 
252 	err = 0;
253 	for (i = 0; i < size && err == 0; i += PAGE_SIZE)
254 		err = map_page(v+i, p+i, flags);
255 	if (err) {
256 		if (mem_init_done)
257 			vunmap((void *)v);
258 		return NULL;
259 	}
260 
261 out:
262 	return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
263 }
264 EXPORT_SYMBOL(__ioremap);
265 
iounmap(volatile void __iomem * addr)266 void iounmap(volatile void __iomem *addr)
267 {
268 	/*
269 	 * If mapped by BATs then there is nothing to do.
270 	 * Calling vfree() generates a benign warning.
271 	 */
272 	if (v_mapped_by_bats((unsigned long)addr)) return;
273 
274 	if (addr > high_memory && (unsigned long) addr < ioremap_bot)
275 		vunmap((void *) (PAGE_MASK & (unsigned long)addr));
276 }
277 EXPORT_SYMBOL(iounmap);
278 
map_page(unsigned long va,phys_addr_t pa,int flags)279 int map_page(unsigned long va, phys_addr_t pa, int flags)
280 {
281 	pmd_t *pd;
282 	pte_t *pg;
283 	int err = -ENOMEM;
284 
285 	/* Use upper 10 bits of VA to index the first level map */
286 	pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
287 	/* Use middle 10 bits of VA to index the second-level map */
288 	pg = pte_alloc_kernel(pd, va);
289 	if (pg != 0) {
290 		err = 0;
291 		/* The PTE should never be already set nor present in the
292 		 * hash table
293 		 */
294 		BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
295 		       flags);
296 		set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
297 						     __pgprot(flags)));
298 	}
299 	return err;
300 }
301 
302 /*
303  * Map in a chunk of physical memory starting at start.
304  */
__mapin_ram_chunk(unsigned long offset,unsigned long top)305 void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
306 {
307 	unsigned long v, s, f;
308 	phys_addr_t p;
309 	int ktext;
310 
311 	s = offset;
312 	v = PAGE_OFFSET + s;
313 	p = memstart_addr + s;
314 	for (; s < top; s += PAGE_SIZE) {
315 		ktext = ((char *) v >= _stext && (char *) v < etext);
316 		f = ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL;
317 		map_page(v, p, f);
318 #ifdef CONFIG_PPC_STD_MMU_32
319 		if (ktext)
320 			hash_preload(&init_mm, v, 0, 0x300);
321 #endif
322 		v += PAGE_SIZE;
323 		p += PAGE_SIZE;
324 	}
325 }
326 
mapin_ram(void)327 void __init mapin_ram(void)
328 {
329 	unsigned long s, top;
330 
331 #ifndef CONFIG_WII
332 	top = total_lowmem;
333 	s = mmu_mapin_ram(top);
334 	__mapin_ram_chunk(s, top);
335 #else
336 	if (!wii_hole_size) {
337 		s = mmu_mapin_ram(total_lowmem);
338 		__mapin_ram_chunk(s, total_lowmem);
339 	} else {
340 		top = wii_hole_start;
341 		s = mmu_mapin_ram(top);
342 		__mapin_ram_chunk(s, top);
343 
344 		top = memblock_end_of_DRAM();
345 		s = wii_mmu_mapin_mem2(top);
346 		__mapin_ram_chunk(s, top);
347 	}
348 #endif
349 }
350 
351 /* Scan the real Linux page tables and return a PTE pointer for
352  * a virtual address in a context.
353  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
354  * the PTE pointer is unmodified if PTE is not found.
355  */
356 int
get_pteptr(struct mm_struct * mm,unsigned long addr,pte_t ** ptep,pmd_t ** pmdp)357 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
358 {
359         pgd_t	*pgd;
360 	pud_t	*pud;
361         pmd_t	*pmd;
362         pte_t	*pte;
363         int     retval = 0;
364 
365         pgd = pgd_offset(mm, addr & PAGE_MASK);
366         if (pgd) {
367 		pud = pud_offset(pgd, addr & PAGE_MASK);
368 		if (pud && pud_present(*pud)) {
369 			pmd = pmd_offset(pud, addr & PAGE_MASK);
370 			if (pmd_present(*pmd)) {
371 				pte = pte_offset_map(pmd, addr & PAGE_MASK);
372 				if (pte) {
373 					retval = 1;
374 					*ptep = pte;
375 					if (pmdp)
376 						*pmdp = pmd;
377 					/* XXX caller needs to do pte_unmap, yuck */
378 				}
379 			}
380 		}
381         }
382         return(retval);
383 }
384 
385 #ifdef CONFIG_DEBUG_PAGEALLOC
386 
__change_page_attr(struct page * page,pgprot_t prot)387 static int __change_page_attr(struct page *page, pgprot_t prot)
388 {
389 	pte_t *kpte;
390 	pmd_t *kpmd;
391 	unsigned long address;
392 
393 	BUG_ON(PageHighMem(page));
394 	address = (unsigned long)page_address(page);
395 
396 	if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address))
397 		return 0;
398 	if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
399 		return -EINVAL;
400 	__set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
401 	wmb();
402 	flush_tlb_page(NULL, address);
403 	pte_unmap(kpte);
404 
405 	return 0;
406 }
407 
408 /*
409  * Change the page attributes of an page in the linear mapping.
410  *
411  * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY
412  */
change_page_attr(struct page * page,int numpages,pgprot_t prot)413 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
414 {
415 	int i, err = 0;
416 	unsigned long flags;
417 
418 	local_irq_save(flags);
419 	for (i = 0; i < numpages; i++, page++) {
420 		err = __change_page_attr(page, prot);
421 		if (err)
422 			break;
423 	}
424 	local_irq_restore(flags);
425 	return err;
426 }
427 
428 
kernel_map_pages(struct page * page,int numpages,int enable)429 void kernel_map_pages(struct page *page, int numpages, int enable)
430 {
431 	if (PageHighMem(page))
432 		return;
433 
434 	change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
435 }
436 #endif /* CONFIG_DEBUG_PAGEALLOC */
437 
438 static int fixmaps;
439 
__set_fixmap(enum fixed_addresses idx,phys_addr_t phys,pgprot_t flags)440 void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags)
441 {
442 	unsigned long address = __fix_to_virt(idx);
443 
444 	if (idx >= __end_of_fixed_addresses) {
445 		BUG();
446 		return;
447 	}
448 
449 	map_page(address, phys, pgprot_val(flags));
450 	fixmaps++;
451 }
452 
__this_fixmap_does_not_exist(void)453 void __this_fixmap_does_not_exist(void)
454 {
455 	WARN_ON(1);
456 }
457