xref: /openbmc/linux/arch/powerpc/mm/pgtable_32.c (revision a20eefae)
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 
27 #include <asm/pgtable.h>
28 #include <asm/pgalloc.h>
29 #include <asm/fixmap.h>
30 #include <asm/io.h>
31 #include <asm/setup.h>
32 #include <asm/sections.h>
33 
34 #include <mm/mmu_decl.h>
35 
36 unsigned long ioremap_bot;
37 EXPORT_SYMBOL(ioremap_bot);	/* aka VMALLOC_END */
38 
39 extern char etext[], _stext[], _sinittext[], _einittext[];
40 
41 void __iomem *
42 ioremap(phys_addr_t addr, unsigned long size)
43 {
44 	pgprot_t prot = pgprot_noncached(PAGE_KERNEL);
45 
46 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
47 }
48 EXPORT_SYMBOL(ioremap);
49 
50 void __iomem *
51 ioremap_wc(phys_addr_t addr, unsigned long size)
52 {
53 	pgprot_t prot = pgprot_noncached_wc(PAGE_KERNEL);
54 
55 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
56 }
57 EXPORT_SYMBOL(ioremap_wc);
58 
59 void __iomem *
60 ioremap_wt(phys_addr_t addr, unsigned long size)
61 {
62 	pgprot_t prot = pgprot_cached_wthru(PAGE_KERNEL);
63 
64 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
65 }
66 EXPORT_SYMBOL(ioremap_wt);
67 
68 void __iomem *
69 ioremap_coherent(phys_addr_t addr, unsigned long size)
70 {
71 	pgprot_t prot = pgprot_cached(PAGE_KERNEL);
72 
73 	return __ioremap_caller(addr, size, prot, __builtin_return_address(0));
74 }
75 EXPORT_SYMBOL(ioremap_coherent);
76 
77 void __iomem *
78 ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
79 {
80 	pte_t pte = __pte(flags);
81 
82 	/* writeable implies dirty for kernel addresses */
83 	if (pte_write(pte))
84 		pte = pte_mkdirty(pte);
85 
86 	/* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
87 	pte = pte_exprotect(pte);
88 	pte = pte_mkprivileged(pte);
89 
90 	return __ioremap_caller(addr, size, pte_pgprot(pte), __builtin_return_address(0));
91 }
92 EXPORT_SYMBOL(ioremap_prot);
93 
94 void __iomem *
95 __ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
96 {
97 	return __ioremap_caller(addr, size, __pgprot(flags), __builtin_return_address(0));
98 }
99 
100 void __iomem *
101 __ioremap_caller(phys_addr_t addr, unsigned long size, pgprot_t prot, void *caller)
102 {
103 	unsigned long v, i;
104 	phys_addr_t p;
105 	int err;
106 
107 	/*
108 	 * Choose an address to map it to.
109 	 * Once the vmalloc system is running, we use it.
110 	 * Before then, we use space going down from IOREMAP_TOP
111 	 * (ioremap_bot records where we're up to).
112 	 */
113 	p = addr & PAGE_MASK;
114 	size = PAGE_ALIGN(addr + size) - p;
115 
116 	/*
117 	 * If the address lies within the first 16 MB, assume it's in ISA
118 	 * memory space
119 	 */
120 	if (p < 16*1024*1024)
121 		p += _ISA_MEM_BASE;
122 
123 #ifndef CONFIG_CRASH_DUMP
124 	/*
125 	 * Don't allow anybody to remap normal RAM that we're using.
126 	 * mem_init() sets high_memory so only do the check after that.
127 	 */
128 	if (slab_is_available() && p <= virt_to_phys(high_memory - 1) &&
129 	    page_is_ram(__phys_to_pfn(p))) {
130 		printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n",
131 		       (unsigned long long)p, __builtin_return_address(0));
132 		return NULL;
133 	}
134 #endif
135 
136 	if (size == 0)
137 		return NULL;
138 
139 	/*
140 	 * Is it already mapped?  Perhaps overlapped by a previous
141 	 * mapping.
142 	 */
143 	v = p_block_mapped(p);
144 	if (v)
145 		goto out;
146 
147 	if (slab_is_available()) {
148 		struct vm_struct *area;
149 		area = get_vm_area_caller(size, VM_IOREMAP, caller);
150 		if (area == 0)
151 			return NULL;
152 		area->phys_addr = p;
153 		v = (unsigned long) area->addr;
154 	} else {
155 		v = (ioremap_bot -= size);
156 	}
157 
158 	/*
159 	 * Should check if it is a candidate for a BAT mapping
160 	 */
161 
162 	err = 0;
163 	for (i = 0; i < size && err == 0; i += PAGE_SIZE)
164 		err = map_kernel_page(v + i, p + i, prot);
165 	if (err) {
166 		if (slab_is_available())
167 			vunmap((void *)v);
168 		return NULL;
169 	}
170 
171 out:
172 	return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
173 }
174 EXPORT_SYMBOL(__ioremap);
175 
176 void iounmap(volatile void __iomem *addr)
177 {
178 	/*
179 	 * If mapped by BATs then there is nothing to do.
180 	 * Calling vfree() generates a benign warning.
181 	 */
182 	if (v_block_mapped((unsigned long)addr))
183 		return;
184 
185 	if (addr > high_memory && (unsigned long) addr < ioremap_bot)
186 		vunmap((void *) (PAGE_MASK & (unsigned long)addr));
187 }
188 EXPORT_SYMBOL(iounmap);
189 
190 static void __init *early_alloc_pgtable(unsigned long size)
191 {
192 	void *ptr = memblock_alloc(size, size);
193 
194 	if (!ptr)
195 		panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
196 		      __func__, size, size);
197 
198 	return ptr;
199 }
200 
201 static pte_t __init *early_pte_alloc_kernel(pmd_t *pmdp, unsigned long va)
202 {
203 	if (pmd_none(*pmdp)) {
204 		pte_t *ptep = early_alloc_pgtable(PTE_FRAG_SIZE);
205 
206 		pmd_populate_kernel(&init_mm, pmdp, ptep);
207 	}
208 	return pte_offset_kernel(pmdp, va);
209 }
210 
211 
212 int __ref map_kernel_page(unsigned long va, phys_addr_t pa, pgprot_t prot)
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 	if (likely(slab_is_available()))
222 		pg = pte_alloc_kernel(pd, va);
223 	else
224 		pg = early_pte_alloc_kernel(pd, va);
225 	if (pg != 0) {
226 		err = 0;
227 		/* The PTE should never be already set nor present in the
228 		 * hash table
229 		 */
230 		BUG_ON((pte_present(*pg) | pte_hashpte(*pg)) && pgprot_val(prot));
231 		set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, prot));
232 	}
233 	smp_wmb();
234 	return err;
235 }
236 
237 /*
238  * Map in a chunk of physical memory starting at start.
239  */
240 static void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
241 {
242 	unsigned long v, s;
243 	phys_addr_t p;
244 	int ktext;
245 
246 	s = offset;
247 	v = PAGE_OFFSET + s;
248 	p = memstart_addr + s;
249 	for (; s < top; s += PAGE_SIZE) {
250 		ktext = ((char *)v >= _stext && (char *)v < etext) ||
251 			((char *)v >= _sinittext && (char *)v < _einittext);
252 		map_kernel_page(v, p, ktext ? PAGE_KERNEL_TEXT : PAGE_KERNEL);
253 #ifdef CONFIG_PPC_BOOK3S_32
254 		if (ktext)
255 			hash_preload(&init_mm, v, false, 0x300);
256 #endif
257 		v += PAGE_SIZE;
258 		p += PAGE_SIZE;
259 	}
260 }
261 
262 void __init mapin_ram(void)
263 {
264 	struct memblock_region *reg;
265 
266 	for_each_memblock(memory, reg) {
267 		phys_addr_t base = reg->base;
268 		phys_addr_t top = min(base + reg->size, total_lowmem);
269 
270 		if (base >= top)
271 			continue;
272 		base = mmu_mapin_ram(base, top);
273 		if (IS_ENABLED(CONFIG_BDI_SWITCH))
274 			__mapin_ram_chunk(reg->base, top);
275 		else
276 			__mapin_ram_chunk(base, top);
277 	}
278 }
279 
280 /* Scan the real Linux page tables and return a PTE pointer for
281  * a virtual address in a context.
282  * Returns true (1) if PTE was found, zero otherwise.  The pointer to
283  * the PTE pointer is unmodified if PTE is not found.
284  */
285 static int
286 get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
287 {
288         pgd_t	*pgd;
289 	pud_t	*pud;
290         pmd_t	*pmd;
291         pte_t	*pte;
292         int     retval = 0;
293 
294         pgd = pgd_offset(mm, addr & PAGE_MASK);
295         if (pgd) {
296 		pud = pud_offset(pgd, addr & PAGE_MASK);
297 		if (pud && pud_present(*pud)) {
298 			pmd = pmd_offset(pud, addr & PAGE_MASK);
299 			if (pmd_present(*pmd)) {
300 				pte = pte_offset_map(pmd, addr & PAGE_MASK);
301 				if (pte) {
302 					retval = 1;
303 					*ptep = pte;
304 					if (pmdp)
305 						*pmdp = pmd;
306 					/* XXX caller needs to do pte_unmap, yuck */
307 				}
308 			}
309 		}
310         }
311         return(retval);
312 }
313 
314 static int __change_page_attr_noflush(struct page *page, pgprot_t prot)
315 {
316 	pte_t *kpte;
317 	pmd_t *kpmd;
318 	unsigned long address;
319 
320 	BUG_ON(PageHighMem(page));
321 	address = (unsigned long)page_address(page);
322 
323 	if (v_block_mapped(address))
324 		return 0;
325 	if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
326 		return -EINVAL;
327 	__set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
328 	pte_unmap(kpte);
329 
330 	return 0;
331 }
332 
333 /*
334  * Change the page attributes of an page in the linear mapping.
335  *
336  * THIS DOES NOTHING WITH BAT MAPPINGS, DEBUG USE ONLY
337  */
338 static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
339 {
340 	int i, err = 0;
341 	unsigned long flags;
342 	struct page *start = page;
343 
344 	local_irq_save(flags);
345 	for (i = 0; i < numpages; i++, page++) {
346 		err = __change_page_attr_noflush(page, prot);
347 		if (err)
348 			break;
349 	}
350 	wmb();
351 	local_irq_restore(flags);
352 	flush_tlb_kernel_range((unsigned long)page_address(start),
353 			       (unsigned long)page_address(page));
354 	return err;
355 }
356 
357 void mark_initmem_nx(void)
358 {
359 	struct page *page = virt_to_page(_sinittext);
360 	unsigned long numpages = PFN_UP((unsigned long)_einittext) -
361 				 PFN_DOWN((unsigned long)_sinittext);
362 
363 	if (v_block_mapped((unsigned long)_stext + 1))
364 		mmu_mark_initmem_nx();
365 	else
366 		change_page_attr(page, numpages, PAGE_KERNEL);
367 }
368 
369 #ifdef CONFIG_STRICT_KERNEL_RWX
370 void mark_rodata_ro(void)
371 {
372 	struct page *page;
373 	unsigned long numpages;
374 
375 	if (v_block_mapped((unsigned long)_sinittext)) {
376 		mmu_mark_rodata_ro();
377 		return;
378 	}
379 
380 	page = virt_to_page(_stext);
381 	numpages = PFN_UP((unsigned long)_etext) -
382 		   PFN_DOWN((unsigned long)_stext);
383 
384 	change_page_attr(page, numpages, PAGE_KERNEL_ROX);
385 	/*
386 	 * mark .rodata as read only. Use __init_begin rather than __end_rodata
387 	 * to cover NOTES and EXCEPTION_TABLE.
388 	 */
389 	page = virt_to_page(__start_rodata);
390 	numpages = PFN_UP((unsigned long)__init_begin) -
391 		   PFN_DOWN((unsigned long)__start_rodata);
392 
393 	change_page_attr(page, numpages, PAGE_KERNEL_RO);
394 
395 	// mark_initmem_nx() should have already run by now
396 	ptdump_check_wx();
397 }
398 #endif
399 
400 #ifdef CONFIG_DEBUG_PAGEALLOC
401 void __kernel_map_pages(struct page *page, int numpages, int enable)
402 {
403 	if (PageHighMem(page))
404 		return;
405 
406 	change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
407 }
408 #endif /* CONFIG_DEBUG_PAGEALLOC */
409