xref: /openbmc/linux/arch/powerpc/mm/pgtable_32.c (revision 68198dca)
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 	return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
80 				__builtin_return_address(0));
81 }
82 EXPORT_SYMBOL(ioremap);
83 
84 void __iomem *
85 ioremap_wc(phys_addr_t addr, unsigned long size)
86 {
87 	return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
88 				__builtin_return_address(0));
89 }
90 EXPORT_SYMBOL(ioremap_wc);
91 
92 void __iomem *
93 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
94 {
95 	/* writeable implies dirty for kernel addresses */
96 	if ((flags & (_PAGE_RW | _PAGE_RO)) != _PAGE_RO)
97 		flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
98 
99 	/* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
100 	flags &= ~(_PAGE_USER | _PAGE_EXEC);
101 
102 #ifdef _PAGE_BAP_SR
103 	/* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
104 	 * which means that we just cleared supervisor access... oops ;-) This
105 	 * restores it
106 	 */
107 	flags |= _PAGE_BAP_SR;
108 #endif
109 
110 	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
111 }
112 EXPORT_SYMBOL(ioremap_prot);
113 
114 void __iomem *
115 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
116 {
117 	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
118 }
119 
120 void __iomem *
121 __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
122 		 void *caller)
123 {
124 	unsigned long v, i;
125 	phys_addr_t p;
126 	int err;
127 
128 	/* Make sure we have the base flags */
129 	if ((flags & _PAGE_PRESENT) == 0)
130 		flags |= pgprot_val(PAGE_KERNEL);
131 
132 	/* Non-cacheable page cannot be coherent */
133 	if (flags & _PAGE_NO_CACHE)
134 		flags &= ~_PAGE_COHERENT;
135 
136 	/*
137 	 * Choose an address to map it to.
138 	 * Once the vmalloc system is running, we use it.
139 	 * Before then, we use space going down from IOREMAP_TOP
140 	 * (ioremap_bot records where we're up to).
141 	 */
142 	p = addr & PAGE_MASK;
143 	size = PAGE_ALIGN(addr + size) - p;
144 
145 	/*
146 	 * If the address lies within the first 16 MB, assume it's in ISA
147 	 * memory space
148 	 */
149 	if (p < 16*1024*1024)
150 		p += _ISA_MEM_BASE;
151 
152 #ifndef CONFIG_CRASH_DUMP
153 	/*
154 	 * Don't allow anybody to remap normal RAM that we're using.
155 	 * mem_init() sets high_memory so only do the check after that.
156 	 */
157 	if (slab_is_available() && (p < virt_to_phys(high_memory)) &&
158 	    !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
159 		printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n",
160 		       (unsigned long long)p, __builtin_return_address(0));
161 		return NULL;
162 	}
163 #endif
164 
165 	if (size == 0)
166 		return NULL;
167 
168 	/*
169 	 * Is it already mapped?  Perhaps overlapped by a previous
170 	 * mapping.
171 	 */
172 	v = p_block_mapped(p);
173 	if (v)
174 		goto out;
175 
176 	if (slab_is_available()) {
177 		struct vm_struct *area;
178 		area = get_vm_area_caller(size, VM_IOREMAP, caller);
179 		if (area == 0)
180 			return NULL;
181 		area->phys_addr = p;
182 		v = (unsigned long) area->addr;
183 	} else {
184 		v = (ioremap_bot -= size);
185 	}
186 
187 	/*
188 	 * Should check if it is a candidate for a BAT mapping
189 	 */
190 
191 	err = 0;
192 	for (i = 0; i < size && err == 0; i += PAGE_SIZE)
193 		err = map_kernel_page(v+i, p+i, flags);
194 	if (err) {
195 		if (slab_is_available())
196 			vunmap((void *)v);
197 		return NULL;
198 	}
199 
200 out:
201 	return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
202 }
203 EXPORT_SYMBOL(__ioremap);
204 
205 void iounmap(volatile void __iomem *addr)
206 {
207 	/*
208 	 * If mapped by BATs then there is nothing to do.
209 	 * Calling vfree() generates a benign warning.
210 	 */
211 	if (v_block_mapped((unsigned long)addr))
212 		return;
213 
214 	if (addr > high_memory && (unsigned long) addr < ioremap_bot)
215 		vunmap((void *) (PAGE_MASK & (unsigned long)addr));
216 }
217 EXPORT_SYMBOL(iounmap);
218 
219 int map_kernel_page(unsigned long va, phys_addr_t pa, int flags)
220 {
221 	pmd_t *pd;
222 	pte_t *pg;
223 	int err = -ENOMEM;
224 
225 	/* Use upper 10 bits of VA to index the first level map */
226 	pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
227 	/* Use middle 10 bits of VA to index the second-level map */
228 	pg = pte_alloc_kernel(pd, va);
229 	if (pg != 0) {
230 		err = 0;
231 		/* The PTE should never be already set nor present in the
232 		 * hash table
233 		 */
234 		BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
235 		       flags);
236 		set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
237 						     __pgprot(flags)));
238 	}
239 	smp_wmb();
240 	return err;
241 }
242 
243 /*
244  * Map in a chunk of physical memory starting at start.
245  */
246 static void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
247 {
248 	unsigned long v, s, f;
249 	phys_addr_t p;
250 	int ktext;
251 
252 	s = offset;
253 	v = PAGE_OFFSET + s;
254 	p = memstart_addr + s;
255 	for (; s < top; s += PAGE_SIZE) {
256 		ktext = ((char *)v >= _stext && (char *)v < etext) ||
257 			((char *)v >= _sinittext && (char *)v < _einittext);
258 		f = ktext ? pgprot_val(PAGE_KERNEL_TEXT) : pgprot_val(PAGE_KERNEL);
259 		map_kernel_page(v, p, f);
260 #ifdef CONFIG_PPC_STD_MMU_32
261 		if (ktext)
262 			hash_preload(&init_mm, v, 0, 0x300);
263 #endif
264 		v += PAGE_SIZE;
265 		p += PAGE_SIZE;
266 	}
267 }
268 
269 void __init mapin_ram(void)
270 {
271 	unsigned long s, top;
272 
273 #ifndef CONFIG_WII
274 	top = total_lowmem;
275 	s = mmu_mapin_ram(top);
276 	__mapin_ram_chunk(s, top);
277 #else
278 	if (!wii_hole_size) {
279 		s = mmu_mapin_ram(total_lowmem);
280 		__mapin_ram_chunk(s, total_lowmem);
281 	} else {
282 		top = wii_hole_start;
283 		s = mmu_mapin_ram(top);
284 		__mapin_ram_chunk(s, top);
285 
286 		top = memblock_end_of_DRAM();
287 		s = wii_mmu_mapin_mem2(top);
288 		__mapin_ram_chunk(s, top);
289 	}
290 #endif
291 }
292 
293 /* Scan the real Linux page tables and return a PTE pointer for
294  * a virtual address in a context.
295  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
296  * the PTE pointer is unmodified if PTE is not found.
297  */
298 static int
299 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
300 {
301         pgd_t	*pgd;
302 	pud_t	*pud;
303         pmd_t	*pmd;
304         pte_t	*pte;
305         int     retval = 0;
306 
307         pgd = pgd_offset(mm, addr & PAGE_MASK);
308         if (pgd) {
309 		pud = pud_offset(pgd, addr & PAGE_MASK);
310 		if (pud && pud_present(*pud)) {
311 			pmd = pmd_offset(pud, addr & PAGE_MASK);
312 			if (pmd_present(*pmd)) {
313 				pte = pte_offset_map(pmd, addr & PAGE_MASK);
314 				if (pte) {
315 					retval = 1;
316 					*ptep = pte;
317 					if (pmdp)
318 						*pmdp = pmd;
319 					/* XXX caller needs to do pte_unmap, yuck */
320 				}
321 			}
322 		}
323         }
324         return(retval);
325 }
326 
327 static int __change_page_attr_noflush(struct page *page, pgprot_t prot)
328 {
329 	pte_t *kpte;
330 	pmd_t *kpmd;
331 	unsigned long address;
332 
333 	BUG_ON(PageHighMem(page));
334 	address = (unsigned long)page_address(page);
335 
336 	if (v_block_mapped(address))
337 		return 0;
338 	if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
339 		return -EINVAL;
340 	__set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
341 	pte_unmap(kpte);
342 
343 	return 0;
344 }
345 
346 /*
347  * Change the page attributes of an page in the linear mapping.
348  *
349  * THIS DOES NOTHING WITH BAT MAPPINGS, DEBUG USE ONLY
350  */
351 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
352 {
353 	int i, err = 0;
354 	unsigned long flags;
355 	struct page *start = page;
356 
357 	local_irq_save(flags);
358 	for (i = 0; i < numpages; i++, page++) {
359 		err = __change_page_attr_noflush(page, prot);
360 		if (err)
361 			break;
362 	}
363 	wmb();
364 	local_irq_restore(flags);
365 	flush_tlb_kernel_range((unsigned long)page_address(start),
366 			       (unsigned long)page_address(page));
367 	return err;
368 }
369 
370 void mark_initmem_nx(void)
371 {
372 	struct page *page = virt_to_page(_sinittext);
373 	unsigned long numpages = PFN_UP((unsigned long)_einittext) -
374 				 PFN_DOWN((unsigned long)_sinittext);
375 
376 	change_page_attr(page, numpages, PAGE_KERNEL);
377 }
378 
379 #ifdef CONFIG_STRICT_KERNEL_RWX
380 void mark_rodata_ro(void)
381 {
382 	struct page *page;
383 	unsigned long numpages;
384 
385 	page = virt_to_page(_stext);
386 	numpages = PFN_UP((unsigned long)_etext) -
387 		   PFN_DOWN((unsigned long)_stext);
388 
389 	change_page_attr(page, numpages, PAGE_KERNEL_ROX);
390 	/*
391 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
392 	 * to cover NOTES and EXCEPTION_TABLE.
393 	 */
394 	page = virt_to_page(__start_rodata);
395 	numpages = PFN_UP((unsigned long)__init_begin) -
396 		   PFN_DOWN((unsigned long)__start_rodata);
397 
398 	change_page_attr(page, numpages, PAGE_KERNEL_RO);
399 }
400 #endif
401 
402 #ifdef CONFIG_DEBUG_PAGEALLOC
403 void __kernel_map_pages(struct page *page, int numpages, int enable)
404 {
405 	if (PageHighMem(page))
406 		return;
407 
408 	change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
409 }
410 #endif /* CONFIG_DEBUG_PAGEALLOC */
411