xref: /openbmc/linux/arch/powerpc/mm/pgtable_32.c (revision 2209fda3)
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 #include <asm/sections.h>
38 
39 #include "mmu_decl.h"
40 
41 unsigned long ioremap_bot;
42 EXPORT_SYMBOL(ioremap_bot);	/* aka VMALLOC_END */
43 
44 extern char etext[], _stext[], _sinittext[], _einittext[];
45 
46 __ref pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
47 {
48 	pte_t *pte;
49 
50 	if (slab_is_available()) {
51 		pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
52 	} else {
53 		pte = __va(memblock_alloc(PAGE_SIZE, PAGE_SIZE));
54 		if (pte)
55 			clear_page(pte);
56 	}
57 	return pte;
58 }
59 
60 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
61 {
62 	struct page *ptepage;
63 
64 	gfp_t flags = GFP_KERNEL | __GFP_ZERO | __GFP_ACCOUNT;
65 
66 	ptepage = alloc_pages(flags, 0);
67 	if (!ptepage)
68 		return NULL;
69 	if (!pgtable_page_ctor(ptepage)) {
70 		__free_page(ptepage);
71 		return NULL;
72 	}
73 	return ptepage;
74 }
75 
76 void __iomem *
77 ioremap(phys_addr_t addr, unsigned long size)
78 {
79 	pgprot_t prot = pgprot_noncached(PAGE_KERNEL);
80 
81 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
82 }
83 EXPORT_SYMBOL(ioremap);
84 
85 void __iomem *
86 ioremap_wc(phys_addr_t addr, unsigned long size)
87 {
88 	pgprot_t prot = pgprot_noncached_wc(PAGE_KERNEL);
89 
90 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
91 }
92 EXPORT_SYMBOL(ioremap_wc);
93 
94 void __iomem *
95 ioremap_wt(phys_addr_t addr, unsigned long size)
96 {
97 	pgprot_t prot = pgprot_cached_wthru(PAGE_KERNEL);
98 
99 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
100 }
101 EXPORT_SYMBOL(ioremap_wt);
102 
103 void __iomem *
104 ioremap_coherent(phys_addr_t addr, unsigned long size)
105 {
106 	pgprot_t prot = pgprot_cached(PAGE_KERNEL);
107 
108 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
109 }
110 EXPORT_SYMBOL(ioremap_coherent);
111 
112 void __iomem *
113 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
114 {
115 	pte_t pte = __pte(flags);
116 
117 	/* writeable implies dirty for kernel addresses */
118 	if (pte_write(pte))
119 		pte = pte_mkdirty(pte);
120 
121 	/* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
122 	pte = pte_exprotect(pte);
123 	pte = pte_mkprivileged(pte);
124 
125 	return __ioremap_caller(addr, size, pte_pgprot(pte), __builtin_return_address(0));
126 }
127 EXPORT_SYMBOL(ioremap_prot);
128 
129 void __iomem *
130 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
131 {
132 	return __ioremap_caller(addr, size, __pgprot(flags), __builtin_return_address(0));
133 }
134 
135 void __iomem *
136 __ioremap_caller(phys_addr_t addr, unsigned long size, pgprot_t prot, void *caller)
137 {
138 	unsigned long v, i;
139 	phys_addr_t p;
140 	int err;
141 
142 	/*
143 	 * Choose an address to map it to.
144 	 * Once the vmalloc system is running, we use it.
145 	 * Before then, we use space going down from IOREMAP_TOP
146 	 * (ioremap_bot records where we're up to).
147 	 */
148 	p = addr & PAGE_MASK;
149 	size = PAGE_ALIGN(addr + size) - p;
150 
151 	/*
152 	 * If the address lies within the first 16 MB, assume it's in ISA
153 	 * memory space
154 	 */
155 	if (p < 16*1024*1024)
156 		p += _ISA_MEM_BASE;
157 
158 #ifndef CONFIG_CRASH_DUMP
159 	/*
160 	 * Don't allow anybody to remap normal RAM that we're using.
161 	 * mem_init() sets high_memory so only do the check after that.
162 	 */
163 	if (slab_is_available() && (p < virt_to_phys(high_memory)) &&
164 	    page_is_ram(__phys_to_pfn(p))) {
165 		printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n",
166 		       (unsigned long long)p, __builtin_return_address(0));
167 		return NULL;
168 	}
169 #endif
170 
171 	if (size == 0)
172 		return NULL;
173 
174 	/*
175 	 * Is it already mapped?  Perhaps overlapped by a previous
176 	 * mapping.
177 	 */
178 	v = p_block_mapped(p);
179 	if (v)
180 		goto out;
181 
182 	if (slab_is_available()) {
183 		struct vm_struct *area;
184 		area = get_vm_area_caller(size, VM_IOREMAP, caller);
185 		if (area == 0)
186 			return NULL;
187 		area->phys_addr = p;
188 		v = (unsigned long) area->addr;
189 	} else {
190 		v = (ioremap_bot -= size);
191 	}
192 
193 	/*
194 	 * Should check if it is a candidate for a BAT mapping
195 	 */
196 
197 	err = 0;
198 	for (i = 0; i < size && err == 0; i += PAGE_SIZE)
199 		err = map_kernel_page(v + i, p + i, prot);
200 	if (err) {
201 		if (slab_is_available())
202 			vunmap((void *)v);
203 		return NULL;
204 	}
205 
206 out:
207 	return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
208 }
209 EXPORT_SYMBOL(__ioremap);
210 
211 void iounmap(volatile void __iomem *addr)
212 {
213 	/*
214 	 * If mapped by BATs then there is nothing to do.
215 	 * Calling vfree() generates a benign warning.
216 	 */
217 	if (v_block_mapped((unsigned long)addr))
218 		return;
219 
220 	if (addr > high_memory && (unsigned long) addr < ioremap_bot)
221 		vunmap((void *) (PAGE_MASK & (unsigned long)addr));
222 }
223 EXPORT_SYMBOL(iounmap);
224 
225 int map_kernel_page(unsigned long va, phys_addr_t pa, pgprot_t prot)
226 {
227 	pmd_t *pd;
228 	pte_t *pg;
229 	int err = -ENOMEM;
230 
231 	/* Use upper 10 bits of VA to index the first level map */
232 	pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
233 	/* Use middle 10 bits of VA to index the second-level map */
234 	pg = pte_alloc_kernel(pd, va);
235 	if (pg != 0) {
236 		err = 0;
237 		/* The PTE should never be already set nor present in the
238 		 * hash table
239 		 */
240 		BUG_ON((pte_present(*pg) | pte_hashpte(*pg)) && pgprot_val(prot));
241 		set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, prot));
242 	}
243 	smp_wmb();
244 	return err;
245 }
246 
247 /*
248  * Map in a chunk of physical memory starting at start.
249  */
250 static void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
251 {
252 	unsigned long v, s;
253 	phys_addr_t p;
254 	int ktext;
255 
256 	s = offset;
257 	v = PAGE_OFFSET + s;
258 	p = memstart_addr + s;
259 	for (; s < top; s += PAGE_SIZE) {
260 		ktext = ((char *)v >= _stext && (char *)v < etext) ||
261 			((char *)v >= _sinittext && (char *)v < _einittext);
262 		map_kernel_page(v, p, ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL);
263 #ifdef CONFIG_PPC_STD_MMU_32
264 		if (ktext)
265 			hash_preload(&init_mm, v, false, 0x300);
266 #endif
267 		v += PAGE_SIZE;
268 		p += PAGE_SIZE;
269 	}
270 }
271 
272 void __init mapin_ram(void)
273 {
274 	unsigned long s, top;
275 
276 #ifndef CONFIG_WII
277 	top = total_lowmem;
278 	s = mmu_mapin_ram(top);
279 	__mapin_ram_chunk(s, top);
280 #else
281 	if (!wii_hole_size) {
282 		s = mmu_mapin_ram(total_lowmem);
283 		__mapin_ram_chunk(s, total_lowmem);
284 	} else {
285 		top = wii_hole_start;
286 		s = mmu_mapin_ram(top);
287 		__mapin_ram_chunk(s, top);
288 
289 		top = memblock_end_of_DRAM();
290 		s = wii_mmu_mapin_mem2(top);
291 		__mapin_ram_chunk(s, top);
292 	}
293 #endif
294 }
295 
296 /* Scan the real Linux page tables and return a PTE pointer for
297  * a virtual address in a context.
298  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
299  * the PTE pointer is unmodified if PTE is not found.
300  */
301 static int
302 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
303 {
304         pgd_t	*pgd;
305 	pud_t	*pud;
306         pmd_t	*pmd;
307         pte_t	*pte;
308         int     retval = 0;
309 
310         pgd = pgd_offset(mm, addr & PAGE_MASK);
311         if (pgd) {
312 		pud = pud_offset(pgd, addr & PAGE_MASK);
313 		if (pud && pud_present(*pud)) {
314 			pmd = pmd_offset(pud, addr & PAGE_MASK);
315 			if (pmd_present(*pmd)) {
316 				pte = pte_offset_map(pmd, addr & PAGE_MASK);
317 				if (pte) {
318 					retval = 1;
319 					*ptep = pte;
320 					if (pmdp)
321 						*pmdp = pmd;
322 					/* XXX caller needs to do pte_unmap, yuck */
323 				}
324 			}
325 		}
326         }
327         return(retval);
328 }
329 
330 static int __change_page_attr_noflush(struct page *page, pgprot_t prot)
331 {
332 	pte_t *kpte;
333 	pmd_t *kpmd;
334 	unsigned long address;
335 
336 	BUG_ON(PageHighMem(page));
337 	address = (unsigned long)page_address(page);
338 
339 	if (v_block_mapped(address))
340 		return 0;
341 	if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
342 		return -EINVAL;
343 	__set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
344 	pte_unmap(kpte);
345 
346 	return 0;
347 }
348 
349 /*
350  * Change the page attributes of an page in the linear mapping.
351  *
352  * THIS DOES NOTHING WITH BAT MAPPINGS, DEBUG USE ONLY
353  */
354 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
355 {
356 	int i, err = 0;
357 	unsigned long flags;
358 	struct page *start = page;
359 
360 	local_irq_save(flags);
361 	for (i = 0; i < numpages; i++, page++) {
362 		err = __change_page_attr_noflush(page, prot);
363 		if (err)
364 			break;
365 	}
366 	wmb();
367 	local_irq_restore(flags);
368 	flush_tlb_kernel_range((unsigned long)page_address(start),
369 			       (unsigned long)page_address(page));
370 	return err;
371 }
372 
373 void mark_initmem_nx(void)
374 {
375 	struct page *page = virt_to_page(_sinittext);
376 	unsigned long numpages = PFN_UP((unsigned long)_einittext) -
377 				 PFN_DOWN((unsigned long)_sinittext);
378 
379 	change_page_attr(page, numpages, PAGE_KERNEL);
380 }
381 
382 #ifdef CONFIG_STRICT_KERNEL_RWX
383 void mark_rodata_ro(void)
384 {
385 	struct page *page;
386 	unsigned long numpages;
387 
388 	page = virt_to_page(_stext);
389 	numpages = PFN_UP((unsigned long)_etext) -
390 		   PFN_DOWN((unsigned long)_stext);
391 
392 	change_page_attr(page, numpages, PAGE_KERNEL_ROX);
393 	/*
394 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
395 	 * to cover NOTES and EXCEPTION_TABLE.
396 	 */
397 	page = virt_to_page(__start_rodata);
398 	numpages = PFN_UP((unsigned long)__init_begin) -
399 		   PFN_DOWN((unsigned long)__start_rodata);
400 
401 	change_page_attr(page, numpages, PAGE_KERNEL_RO);
402 }
403 #endif
404 
405 #ifdef CONFIG_DEBUG_PAGEALLOC
406 void __kernel_map_pages(struct page *page, int numpages, int enable)
407 {
408 	if (PageHighMem(page))
409 		return;
410 
411 	change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
412 }
413 #endif /* CONFIG_DEBUG_PAGEALLOC */
414