Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * This file contains the routines setting up the linux page tables.
4 * -- paulus
5 *
6 * Derived from arch/ppc/mm/init.c:
7 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
8 *
9 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
10 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
11 * Copyright (C) 1996 Paul Mackerras
12 *
13 * Derived from "arch/i386/mm/init.c"
14 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
15 */
16
17#include <linux/kernel.h>
18#include <linux/module.h>
19#include <linux/types.h>
20#include <linux/mm.h>
21#include <linux/vmalloc.h>
22#include <linux/init.h>
23#include <linux/highmem.h>
24#include <linux/memblock.h>
25#include <linux/slab.h>
26#include <linux/set_memory.h>
27
28#include <asm/pgalloc.h>
29#include <asm/fixmap.h>
30#include <asm/setup.h>
31#include <asm/sections.h>
32#include <asm/early_ioremap.h>
33
34#include <mm/mmu_decl.h>
35
36static u8 early_fixmap_pagetable[FIXMAP_PTE_SIZE] __page_aligned_data;
37
38notrace void __init early_ioremap_init(void)
39{
40 unsigned long addr = ALIGN_DOWN(FIXADDR_START, PGDIR_SIZE);
41 pte_t *ptep = (pte_t *)early_fixmap_pagetable;
42 pmd_t *pmdp = pmd_off_k(addr);
43
44 for (; (s32)(FIXADDR_TOP - addr) > 0;
45 addr += PGDIR_SIZE, ptep += PTRS_PER_PTE, pmdp++)
46 pmd_populate_kernel(&init_mm, pmdp, ptep);
47
48 early_ioremap_setup();
49}
50
51static void __init *early_alloc_pgtable(unsigned long size)
52{
53 void *ptr = memblock_alloc(size, size);
54
55 if (!ptr)
56 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
57 __func__, size, size);
58
59 return ptr;
60}
61
62pte_t __init *early_pte_alloc_kernel(pmd_t *pmdp, unsigned long va)
63{
64 if (pmd_none(*pmdp)) {
65 pte_t *ptep = early_alloc_pgtable(PTE_FRAG_SIZE);
66
67 pmd_populate_kernel(&init_mm, pmdp, ptep);
68 }
69 return pte_offset_kernel(pmdp, va);
70}
71
72
73int __ref map_kernel_page(unsigned long va, phys_addr_t pa, pgprot_t prot)
74{
75 pmd_t *pd;
76 pte_t *pg;
77 int err = -ENOMEM;
78
79 /* Use upper 10 bits of VA to index the first level map */
80 pd = pmd_off_k(va);
81 /* Use middle 10 bits of VA to index the second-level map */
82 if (likely(slab_is_available()))
83 pg = pte_alloc_kernel(pd, va);
84 else
85 pg = early_pte_alloc_kernel(pd, va);
86 if (pg) {
87 err = 0;
88 /* The PTE should never be already set nor present in the
89 * hash table
90 */
91 BUG_ON((pte_present(*pg) | pte_hashpte(*pg)) && pgprot_val(prot));
92 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, prot));
93 }
94 smp_wmb();
95 return err;
96}
97
98/*
99 * Map in a chunk of physical memory starting at start.
100 */
101static void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
102{
103 unsigned long v, s;
104 phys_addr_t p;
105 bool ktext;
106
107 s = offset;
108 v = PAGE_OFFSET + s;
109 p = memstart_addr + s;
110 for (; s < top; s += PAGE_SIZE) {
111 ktext = core_kernel_text(v);
112 map_kernel_page(v, p, ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL);
113 v += PAGE_SIZE;
114 p += PAGE_SIZE;
115 }
116}
117
118void __init mapin_ram(void)
119{
120 phys_addr_t base, end;
121 u64 i;
122
123 for_each_mem_range(i, &base, &end) {
124 phys_addr_t top = min(end, total_lowmem);
125
126 if (base >= top)
127 continue;
128 base = mmu_mapin_ram(base, top);
129 __mapin_ram_chunk(base, top);
130 }
131}
132
133void mark_initmem_nx(void)
134{
135 unsigned long numpages = PFN_UP((unsigned long)_einittext) -
136 PFN_DOWN((unsigned long)_sinittext);
137
138 mmu_mark_initmem_nx();
139
140 if (!v_block_mapped((unsigned long)_sinittext)) {
141 set_memory_nx((unsigned long)_sinittext, numpages);
142 set_memory_rw((unsigned long)_sinittext, numpages);
143 }
144}
145
146#ifdef CONFIG_STRICT_KERNEL_RWX
147void mark_rodata_ro(void)
148{
149 unsigned long numpages;
150
151 if (IS_ENABLED(CONFIG_STRICT_MODULE_RWX) && mmu_has_feature(MMU_FTR_HPTE_TABLE))
152 pr_warn("This platform has HASH MMU, STRICT_MODULE_RWX won't work\n");
153
154 if (v_block_mapped((unsigned long)_stext + 1)) {
155 mmu_mark_rodata_ro();
156 ptdump_check_wx();
157 return;
158 }
159
160 /*
161 * mark text and rodata as read only. __end_rodata is set by
162 * powerpc's linker script and includes tables and data
163 * requiring relocation which are not put in RO_DATA.
164 */
165 numpages = PFN_UP((unsigned long)__end_rodata) -
166 PFN_DOWN((unsigned long)_stext);
167
168 set_memory_ro((unsigned long)_stext, numpages);
169
170 // mark_initmem_nx() should have already run by now
171 ptdump_check_wx();
172}
173#endif
174
175#if defined(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) && defined(CONFIG_DEBUG_PAGEALLOC)
176void __kernel_map_pages(struct page *page, int numpages, int enable)
177{
178 unsigned long addr = (unsigned long)page_address(page);
179
180 if (PageHighMem(page))
181 return;
182
183 if (enable)
184 set_memory_p(addr, numpages);
185 else
186 set_memory_np(addr, numpages);
187}
188#endif /* CONFIG_DEBUG_PAGEALLOC */
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
41unsigned long ioremap_bot;
42EXPORT_SYMBOL(ioremap_bot); /* aka VMALLOC_END */
43
44extern 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
60pgtable_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
76void __iomem *
77ioremap(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}
82EXPORT_SYMBOL(ioremap);
83
84void __iomem *
85ioremap_wc(phys_addr_t addr, unsigned long size)
86{
87 return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
88 __builtin_return_address(0));
89}
90EXPORT_SYMBOL(ioremap_wc);
91
92void __iomem *
93ioremap_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 flags |= _PAGE_PRIVILEGED;
102
103 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
104}
105EXPORT_SYMBOL(ioremap_prot);
106
107void __iomem *
108__ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
109{
110 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
111}
112
113void __iomem *
114__ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
115 void *caller)
116{
117 unsigned long v, i;
118 phys_addr_t p;
119 int err;
120
121 /* Make sure we have the base flags */
122 if ((flags & _PAGE_PRESENT) == 0)
123 flags |= pgprot_val(PAGE_KERNEL);
124
125 /* Non-cacheable page cannot be coherent */
126 if (flags & _PAGE_NO_CACHE)
127 flags &= ~_PAGE_COHERENT;
128
129 /*
130 * Choose an address to map it to.
131 * Once the vmalloc system is running, we use it.
132 * Before then, we use space going down from IOREMAP_TOP
133 * (ioremap_bot records where we're up to).
134 */
135 p = addr & PAGE_MASK;
136 size = PAGE_ALIGN(addr + size) - p;
137
138 /*
139 * If the address lies within the first 16 MB, assume it's in ISA
140 * memory space
141 */
142 if (p < 16*1024*1024)
143 p += _ISA_MEM_BASE;
144
145#ifndef CONFIG_CRASH_DUMP
146 /*
147 * Don't allow anybody to remap normal RAM that we're using.
148 * mem_init() sets high_memory so only do the check after that.
149 */
150 if (slab_is_available() && (p < virt_to_phys(high_memory)) &&
151 page_is_ram(__phys_to_pfn(p))) {
152 printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n",
153 (unsigned long long)p, __builtin_return_address(0));
154 return NULL;
155 }
156#endif
157
158 if (size == 0)
159 return NULL;
160
161 /*
162 * Is it already mapped? Perhaps overlapped by a previous
163 * mapping.
164 */
165 v = p_block_mapped(p);
166 if (v)
167 goto out;
168
169 if (slab_is_available()) {
170 struct vm_struct *area;
171 area = get_vm_area_caller(size, VM_IOREMAP, caller);
172 if (area == 0)
173 return NULL;
174 area->phys_addr = p;
175 v = (unsigned long) area->addr;
176 } else {
177 v = (ioremap_bot -= size);
178 }
179
180 /*
181 * Should check if it is a candidate for a BAT mapping
182 */
183
184 err = 0;
185 for (i = 0; i < size && err == 0; i += PAGE_SIZE)
186 err = map_kernel_page(v+i, p+i, flags);
187 if (err) {
188 if (slab_is_available())
189 vunmap((void *)v);
190 return NULL;
191 }
192
193out:
194 return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
195}
196EXPORT_SYMBOL(__ioremap);
197
198void iounmap(volatile void __iomem *addr)
199{
200 /*
201 * If mapped by BATs then there is nothing to do.
202 * Calling vfree() generates a benign warning.
203 */
204 if (v_block_mapped((unsigned long)addr))
205 return;
206
207 if (addr > high_memory && (unsigned long) addr < ioremap_bot)
208 vunmap((void *) (PAGE_MASK & (unsigned long)addr));
209}
210EXPORT_SYMBOL(iounmap);
211
212int map_kernel_page(unsigned long va, phys_addr_t pa, int flags)
213{
214 pmd_t *pd;
215 pte_t *pg;
216 int err = -ENOMEM;
217
218 /* Use upper 10 bits of VA to index the first level map */
219 pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
220 /* Use middle 10 bits of VA to index the second-level map */
221 pg = pte_alloc_kernel(pd, va);
222 if (pg != 0) {
223 err = 0;
224 /* The PTE should never be already set nor present in the
225 * hash table
226 */
227 BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
228 flags);
229 set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
230 __pgprot(flags)));
231 }
232 smp_wmb();
233 return err;
234}
235
236/*
237 * Map in a chunk of physical memory starting at start.
238 */
239static void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
240{
241 unsigned long v, s, f;
242 phys_addr_t p;
243 int ktext;
244
245 s = offset;
246 v = PAGE_OFFSET + s;
247 p = memstart_addr + s;
248 for (; s < top; s += PAGE_SIZE) {
249 ktext = ((char *)v >= _stext && (char *)v < etext) ||
250 ((char *)v >= _sinittext && (char *)v < _einittext);
251 f = ktext ? pgprot_val(PAGE_KERNEL_TEXT) : pgprot_val(PAGE_KERNEL);
252 map_kernel_page(v, p, f);
253#ifdef CONFIG_PPC_STD_MMU_32
254 if (ktext)
255 hash_preload(&init_mm, v, 0, 0x300);
256#endif
257 v += PAGE_SIZE;
258 p += PAGE_SIZE;
259 }
260}
261
262void __init mapin_ram(void)
263{
264 unsigned long s, top;
265
266#ifndef CONFIG_WII
267 top = total_lowmem;
268 s = mmu_mapin_ram(top);
269 __mapin_ram_chunk(s, top);
270#else
271 if (!wii_hole_size) {
272 s = mmu_mapin_ram(total_lowmem);
273 __mapin_ram_chunk(s, total_lowmem);
274 } else {
275 top = wii_hole_start;
276 s = mmu_mapin_ram(top);
277 __mapin_ram_chunk(s, top);
278
279 top = memblock_end_of_DRAM();
280 s = wii_mmu_mapin_mem2(top);
281 __mapin_ram_chunk(s, top);
282 }
283#endif
284}
285
286/* Scan the real Linux page tables and return a PTE pointer for
287 * a virtual address in a context.
288 * Returns true (1) if PTE was found, zero otherwise. The pointer to
289 * the PTE pointer is unmodified if PTE is not found.
290 */
291static int
292get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
293{
294 pgd_t *pgd;
295 pud_t *pud;
296 pmd_t *pmd;
297 pte_t *pte;
298 int retval = 0;
299
300 pgd = pgd_offset(mm, addr & PAGE_MASK);
301 if (pgd) {
302 pud = pud_offset(pgd, addr & PAGE_MASK);
303 if (pud && pud_present(*pud)) {
304 pmd = pmd_offset(pud, addr & PAGE_MASK);
305 if (pmd_present(*pmd)) {
306 pte = pte_offset_map(pmd, addr & PAGE_MASK);
307 if (pte) {
308 retval = 1;
309 *ptep = pte;
310 if (pmdp)
311 *pmdp = pmd;
312 /* XXX caller needs to do pte_unmap, yuck */
313 }
314 }
315 }
316 }
317 return(retval);
318}
319
320static int __change_page_attr_noflush(struct page *page, pgprot_t prot)
321{
322 pte_t *kpte;
323 pmd_t *kpmd;
324 unsigned long address;
325
326 BUG_ON(PageHighMem(page));
327 address = (unsigned long)page_address(page);
328
329 if (v_block_mapped(address))
330 return 0;
331 if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
332 return -EINVAL;
333 __set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
334 pte_unmap(kpte);
335
336 return 0;
337}
338
339/*
340 * Change the page attributes of an page in the linear mapping.
341 *
342 * THIS DOES NOTHING WITH BAT MAPPINGS, DEBUG USE ONLY
343 */
344static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
345{
346 int i, err = 0;
347 unsigned long flags;
348 struct page *start = page;
349
350 local_irq_save(flags);
351 for (i = 0; i < numpages; i++, page++) {
352 err = __change_page_attr_noflush(page, prot);
353 if (err)
354 break;
355 }
356 wmb();
357 local_irq_restore(flags);
358 flush_tlb_kernel_range((unsigned long)page_address(start),
359 (unsigned long)page_address(page));
360 return err;
361}
362
363void mark_initmem_nx(void)
364{
365 struct page *page = virt_to_page(_sinittext);
366 unsigned long numpages = PFN_UP((unsigned long)_einittext) -
367 PFN_DOWN((unsigned long)_sinittext);
368
369 change_page_attr(page, numpages, PAGE_KERNEL);
370}
371
372#ifdef CONFIG_STRICT_KERNEL_RWX
373void mark_rodata_ro(void)
374{
375 struct page *page;
376 unsigned long numpages;
377
378 page = virt_to_page(_stext);
379 numpages = PFN_UP((unsigned long)_etext) -
380 PFN_DOWN((unsigned long)_stext);
381
382 change_page_attr(page, numpages, PAGE_KERNEL_ROX);
383 /*
384 * mark .rodata as read only. Use __init_begin rather than __end_rodata
385 * to cover NOTES and EXCEPTION_TABLE.
386 */
387 page = virt_to_page(__start_rodata);
388 numpages = PFN_UP((unsigned long)__init_begin) -
389 PFN_DOWN((unsigned long)__start_rodata);
390
391 change_page_attr(page, numpages, PAGE_KERNEL_RO);
392}
393#endif
394
395#ifdef CONFIG_DEBUG_PAGEALLOC
396void __kernel_map_pages(struct page *page, int numpages, int enable)
397{
398 if (PageHighMem(page))
399 return;
400
401 change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
402}
403#endif /* CONFIG_DEBUG_PAGEALLOC */