xref: /openbmc/linux/arch/powerpc/mm/pgtable_64.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  This file contains ioremap and related functions for 64-bit machines.
4  *
5  *  Derived from arch/ppc64/mm/init.c
6  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
7  *
8  *  Modifications by Paul Mackerras (PowerMac) (paulus@samba.org)
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  *  Dave Engebretsen <engebret@us.ibm.com>
16  *      Rework for PPC64 port.
17  */
18 
19 #include <linux/signal.h>
20 #include <linux/sched.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/string.h>
24 #include <linux/export.h>
25 #include <linux/types.h>
26 #include <linux/mman.h>
27 #include <linux/mm.h>
28 #include <linux/swap.h>
29 #include <linux/stddef.h>
30 #include <linux/vmalloc.h>
31 #include <linux/slab.h>
32 #include <linux/hugetlb.h>
33 
34 #include <asm/pgalloc.h>
35 #include <asm/page.h>
36 #include <asm/prom.h>
37 #include <asm/io.h>
38 #include <asm/mmu_context.h>
39 #include <asm/pgtable.h>
40 #include <asm/mmu.h>
41 #include <asm/smp.h>
42 #include <asm/machdep.h>
43 #include <asm/tlb.h>
44 #include <asm/processor.h>
45 #include <asm/cputable.h>
46 #include <asm/sections.h>
47 #include <asm/firmware.h>
48 #include <asm/dma.h>
49 
50 #include <mm/mmu_decl.h>
51 
52 
53 #ifdef CONFIG_PPC_BOOK3S_64
54 /*
55  * partition table and process table for ISA 3.0
56  */
57 struct prtb_entry *process_tb;
58 struct patb_entry *partition_tb;
59 /*
60  * page table size
61  */
62 unsigned long __pte_index_size;
63 EXPORT_SYMBOL(__pte_index_size);
64 unsigned long __pmd_index_size;
65 EXPORT_SYMBOL(__pmd_index_size);
66 unsigned long __pud_index_size;
67 EXPORT_SYMBOL(__pud_index_size);
68 unsigned long __pgd_index_size;
69 EXPORT_SYMBOL(__pgd_index_size);
70 unsigned long __pud_cache_index;
71 EXPORT_SYMBOL(__pud_cache_index);
72 unsigned long __pte_table_size;
73 EXPORT_SYMBOL(__pte_table_size);
74 unsigned long __pmd_table_size;
75 EXPORT_SYMBOL(__pmd_table_size);
76 unsigned long __pud_table_size;
77 EXPORT_SYMBOL(__pud_table_size);
78 unsigned long __pgd_table_size;
79 EXPORT_SYMBOL(__pgd_table_size);
80 unsigned long __pmd_val_bits;
81 EXPORT_SYMBOL(__pmd_val_bits);
82 unsigned long __pud_val_bits;
83 EXPORT_SYMBOL(__pud_val_bits);
84 unsigned long __pgd_val_bits;
85 EXPORT_SYMBOL(__pgd_val_bits);
86 unsigned long __kernel_virt_start;
87 EXPORT_SYMBOL(__kernel_virt_start);
88 unsigned long __vmalloc_start;
89 EXPORT_SYMBOL(__vmalloc_start);
90 unsigned long __vmalloc_end;
91 EXPORT_SYMBOL(__vmalloc_end);
92 unsigned long __kernel_io_start;
93 EXPORT_SYMBOL(__kernel_io_start);
94 unsigned long __kernel_io_end;
95 struct page *vmemmap;
96 EXPORT_SYMBOL(vmemmap);
97 unsigned long __pte_frag_nr;
98 EXPORT_SYMBOL(__pte_frag_nr);
99 unsigned long __pte_frag_size_shift;
100 EXPORT_SYMBOL(__pte_frag_size_shift);
101 unsigned long ioremap_bot;
102 #else /* !CONFIG_PPC_BOOK3S_64 */
103 unsigned long ioremap_bot = IOREMAP_BASE;
104 #endif
105 
106 int __weak ioremap_range(unsigned long ea, phys_addr_t pa, unsigned long size, pgprot_t prot, int nid)
107 {
108 	unsigned long i;
109 
110 	for (i = 0; i < size; i += PAGE_SIZE) {
111 		int err = map_kernel_page(ea + i, pa + i, prot);
112 		if (err) {
113 			if (slab_is_available())
114 				unmap_kernel_range(ea, size);
115 			else
116 				WARN_ON_ONCE(1); /* Should clean up */
117 			return err;
118 		}
119 	}
120 
121 	return 0;
122 }
123 
124 /**
125  * __ioremap_at - Low level function to establish the page tables
126  *                for an IO mapping
127  */
128 void __iomem *__ioremap_at(phys_addr_t pa, void *ea, unsigned long size, pgprot_t prot)
129 {
130 	/* We don't support the 4K PFN hack with ioremap */
131 	if (pgprot_val(prot) & H_PAGE_4K_PFN)
132 		return NULL;
133 
134 	if ((ea + size) >= (void *)IOREMAP_END) {
135 		pr_warn("Outside the supported range\n");
136 		return NULL;
137 	}
138 
139 	WARN_ON(pa & ~PAGE_MASK);
140 	WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
141 	WARN_ON(size & ~PAGE_MASK);
142 
143 	if (ioremap_range((unsigned long)ea, pa, size, prot, NUMA_NO_NODE))
144 		return NULL;
145 
146 	return (void __iomem *)ea;
147 }
148 
149 /**
150  * __iounmap_from - Low level function to tear down the page tables
151  *                  for an IO mapping. This is used for mappings that
152  *                  are manipulated manually, like partial unmapping of
153  *                  PCI IOs or ISA space.
154  */
155 void __iounmap_at(void *ea, unsigned long size)
156 {
157 	WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
158 	WARN_ON(size & ~PAGE_MASK);
159 
160 	unmap_kernel_range((unsigned long)ea, size);
161 }
162 
163 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
164 				pgprot_t prot, void *caller)
165 {
166 	phys_addr_t paligned;
167 	void __iomem *ret;
168 
169 	/*
170 	 * Choose an address to map it to.
171 	 * Once the imalloc system is running, we use it.
172 	 * Before that, we map using addresses going
173 	 * up from ioremap_bot.  imalloc will use
174 	 * the addresses from ioremap_bot through
175 	 * IMALLOC_END
176 	 *
177 	 */
178 	paligned = addr & PAGE_MASK;
179 	size = PAGE_ALIGN(addr + size) - paligned;
180 
181 	if ((size == 0) || (paligned == 0))
182 		return NULL;
183 
184 	if (slab_is_available()) {
185 		struct vm_struct *area;
186 
187 		area = __get_vm_area_caller(size, VM_IOREMAP,
188 					    ioremap_bot, IOREMAP_END,
189 					    caller);
190 		if (area == NULL)
191 			return NULL;
192 
193 		area->phys_addr = paligned;
194 		ret = __ioremap_at(paligned, area->addr, size, prot);
195 	} else {
196 		ret = __ioremap_at(paligned, (void *)ioremap_bot, size, prot);
197 		if (ret)
198 			ioremap_bot += size;
199 	}
200 
201 	if (ret)
202 		ret += addr & ~PAGE_MASK;
203 	return ret;
204 }
205 
206 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
207 			 unsigned long flags)
208 {
209 	return __ioremap_caller(addr, size, __pgprot(flags), __builtin_return_address(0));
210 }
211 
212 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
213 {
214 	pgprot_t prot = pgprot_noncached(PAGE_KERNEL);
215 	void *caller = __builtin_return_address(0);
216 
217 	if (ppc_md.ioremap)
218 		return ppc_md.ioremap(addr, size, prot, caller);
219 	return __ioremap_caller(addr, size, prot, caller);
220 }
221 
222 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
223 {
224 	pgprot_t prot = pgprot_noncached_wc(PAGE_KERNEL);
225 	void *caller = __builtin_return_address(0);
226 
227 	if (ppc_md.ioremap)
228 		return ppc_md.ioremap(addr, size, prot, caller);
229 	return __ioremap_caller(addr, size, prot, caller);
230 }
231 
232 void __iomem *ioremap_coherent(phys_addr_t addr, unsigned long size)
233 {
234 	pgprot_t prot = pgprot_cached(PAGE_KERNEL);
235 	void *caller = __builtin_return_address(0);
236 
237 	if (ppc_md.ioremap)
238 		return ppc_md.ioremap(addr, size, prot, caller);
239 	return __ioremap_caller(addr, size, prot, caller);
240 }
241 
242 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
243 			     unsigned long flags)
244 {
245 	pte_t pte = __pte(flags);
246 	void *caller = __builtin_return_address(0);
247 
248 	/* writeable implies dirty for kernel addresses */
249 	if (pte_write(pte))
250 		pte = pte_mkdirty(pte);
251 
252 	/* we don't want to let _PAGE_EXEC leak out */
253 	pte = pte_exprotect(pte);
254 	/*
255 	 * Force kernel mapping.
256 	 */
257 	pte = pte_mkprivileged(pte);
258 
259 	if (ppc_md.ioremap)
260 		return ppc_md.ioremap(addr, size, pte_pgprot(pte), caller);
261 	return __ioremap_caller(addr, size, pte_pgprot(pte), caller);
262 }
263 
264 
265 /*
266  * Unmap an IO region and remove it from imalloc'd list.
267  * Access to IO memory should be serialized by driver.
268  */
269 void __iounmap(volatile void __iomem *token)
270 {
271 	void *addr;
272 
273 	if (!slab_is_available())
274 		return;
275 
276 	addr = (void *) ((unsigned long __force)
277 			 PCI_FIX_ADDR(token) & PAGE_MASK);
278 	if ((unsigned long)addr < ioremap_bot) {
279 		printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
280 		       " at 0x%p\n", addr);
281 		return;
282 	}
283 	vunmap(addr);
284 }
285 
286 void iounmap(volatile void __iomem *token)
287 {
288 	if (ppc_md.iounmap)
289 		ppc_md.iounmap(token);
290 	else
291 		__iounmap(token);
292 }
293 
294 EXPORT_SYMBOL(ioremap);
295 EXPORT_SYMBOL(ioremap_wc);
296 EXPORT_SYMBOL(ioremap_prot);
297 EXPORT_SYMBOL(__ioremap);
298 EXPORT_SYMBOL(__ioremap_at);
299 EXPORT_SYMBOL(iounmap);
300 EXPORT_SYMBOL(__iounmap);
301 EXPORT_SYMBOL(__iounmap_at);
302 
303 #ifndef __PAGETABLE_PUD_FOLDED
304 /* 4 level page table */
305 struct page *pgd_page(pgd_t pgd)
306 {
307 	if (pgd_is_leaf(pgd)) {
308 		VM_WARN_ON(!pgd_huge(pgd));
309 		return pte_page(pgd_pte(pgd));
310 	}
311 	return virt_to_page(pgd_page_vaddr(pgd));
312 }
313 #endif
314 
315 struct page *pud_page(pud_t pud)
316 {
317 	if (pud_is_leaf(pud)) {
318 		VM_WARN_ON(!pud_huge(pud));
319 		return pte_page(pud_pte(pud));
320 	}
321 	return virt_to_page(pud_page_vaddr(pud));
322 }
323 
324 /*
325  * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
326  * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
327  */
328 struct page *pmd_page(pmd_t pmd)
329 {
330 	if (pmd_is_leaf(pmd)) {
331 		VM_WARN_ON(!(pmd_large(pmd) || pmd_huge(pmd)));
332 		return pte_page(pmd_pte(pmd));
333 	}
334 	return virt_to_page(pmd_page_vaddr(pmd));
335 }
336 
337 #ifdef CONFIG_STRICT_KERNEL_RWX
338 void mark_rodata_ro(void)
339 {
340 	if (!mmu_has_feature(MMU_FTR_KERNEL_RO)) {
341 		pr_warn("Warning: Unable to mark rodata read only on this CPU.\n");
342 		return;
343 	}
344 
345 	if (radix_enabled())
346 		radix__mark_rodata_ro();
347 	else
348 		hash__mark_rodata_ro();
349 
350 	// mark_initmem_nx() should have already run by now
351 	ptdump_check_wx();
352 }
353 
354 void mark_initmem_nx(void)
355 {
356 	if (radix_enabled())
357 		radix__mark_initmem_nx();
358 	else
359 		hash__mark_initmem_nx();
360 }
361 #endif
362