xref: /linux/arch/s390/mm/pageattr.c (revision a8b70ccf10e38775785d9cb12ead916474549f99)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright IBM Corp. 2011
4  * Author(s): Jan Glauber <jang@linux.vnet.ibm.com>
5  */
6 #include <linux/hugetlb.h>
7 #include <linux/mm.h>
8 #include <asm/cacheflush.h>
9 #include <asm/facility.h>
10 #include <asm/pgtable.h>
11 #include <asm/pgalloc.h>
12 #include <asm/page.h>
13 #include <asm/set_memory.h>
14 
15 static inline unsigned long sske_frame(unsigned long addr, unsigned char skey)
16 {
17 	asm volatile(".insn rrf,0xb22b0000,%[skey],%[addr],9,0"
18 		     : [addr] "+a" (addr) : [skey] "d" (skey));
19 	return addr;
20 }
21 
22 void __storage_key_init_range(unsigned long start, unsigned long end)
23 {
24 	unsigned long boundary, size;
25 
26 	if (!PAGE_DEFAULT_KEY)
27 		return;
28 	while (start < end) {
29 		if (MACHINE_HAS_EDAT1) {
30 			/* set storage keys for a 1MB frame */
31 			size = 1UL << 20;
32 			boundary = (start + size) & ~(size - 1);
33 			if (boundary <= end) {
34 				do {
35 					start = sske_frame(start, PAGE_DEFAULT_KEY);
36 				} while (start < boundary);
37 				continue;
38 			}
39 		}
40 		page_set_storage_key(start, PAGE_DEFAULT_KEY, 0);
41 		start += PAGE_SIZE;
42 	}
43 }
44 
45 #ifdef CONFIG_PROC_FS
46 atomic_long_t direct_pages_count[PG_DIRECT_MAP_MAX];
47 
48 void arch_report_meminfo(struct seq_file *m)
49 {
50 	seq_printf(m, "DirectMap4k:    %8lu kB\n",
51 		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_4K]) << 2);
52 	seq_printf(m, "DirectMap1M:    %8lu kB\n",
53 		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_1M]) << 10);
54 	seq_printf(m, "DirectMap2G:    %8lu kB\n",
55 		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_2G]) << 21);
56 }
57 #endif /* CONFIG_PROC_FS */
58 
59 static void pgt_set(unsigned long *old, unsigned long new, unsigned long addr,
60 		    unsigned long dtt)
61 {
62 	unsigned long table, mask;
63 
64 	mask = 0;
65 	if (MACHINE_HAS_EDAT2) {
66 		switch (dtt) {
67 		case CRDTE_DTT_REGION3:
68 			mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1);
69 			break;
70 		case CRDTE_DTT_SEGMENT:
71 			mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
72 			break;
73 		case CRDTE_DTT_PAGE:
74 			mask = ~(PTRS_PER_PTE * sizeof(pte_t) - 1);
75 			break;
76 		}
77 		table = (unsigned long)old & mask;
78 		crdte(*old, new, table, dtt, addr, S390_lowcore.kernel_asce);
79 	} else if (MACHINE_HAS_IDTE) {
80 		cspg(old, *old, new);
81 	} else {
82 		csp((unsigned int *)old + 1, *old, new);
83 	}
84 }
85 
86 static int walk_pte_level(pmd_t *pmdp, unsigned long addr, unsigned long end,
87 			  unsigned long flags)
88 {
89 	pte_t *ptep, new;
90 
91 	ptep = pte_offset(pmdp, addr);
92 	do {
93 		new = *ptep;
94 		if (pte_none(new))
95 			return -EINVAL;
96 		if (flags & SET_MEMORY_RO)
97 			new = pte_wrprotect(new);
98 		else if (flags & SET_MEMORY_RW)
99 			new = pte_mkwrite(pte_mkdirty(new));
100 		if (flags & SET_MEMORY_NX)
101 			pte_val(new) |= _PAGE_NOEXEC;
102 		else if (flags & SET_MEMORY_X)
103 			pte_val(new) &= ~_PAGE_NOEXEC;
104 		pgt_set((unsigned long *)ptep, pte_val(new), addr, CRDTE_DTT_PAGE);
105 		ptep++;
106 		addr += PAGE_SIZE;
107 		cond_resched();
108 	} while (addr < end);
109 	return 0;
110 }
111 
112 static int split_pmd_page(pmd_t *pmdp, unsigned long addr)
113 {
114 	unsigned long pte_addr, prot;
115 	pte_t *pt_dir, *ptep;
116 	pmd_t new;
117 	int i, ro, nx;
118 
119 	pt_dir = vmem_pte_alloc();
120 	if (!pt_dir)
121 		return -ENOMEM;
122 	pte_addr = pmd_pfn(*pmdp) << PAGE_SHIFT;
123 	ro = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT);
124 	nx = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_NOEXEC);
125 	prot = pgprot_val(ro ? PAGE_KERNEL_RO : PAGE_KERNEL);
126 	if (!nx)
127 		prot &= ~_PAGE_NOEXEC;
128 	ptep = pt_dir;
129 	for (i = 0; i < PTRS_PER_PTE; i++) {
130 		pte_val(*ptep) = pte_addr | prot;
131 		pte_addr += PAGE_SIZE;
132 		ptep++;
133 	}
134 	pmd_val(new) = __pa(pt_dir) | _SEGMENT_ENTRY;
135 	pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT);
136 	update_page_count(PG_DIRECT_MAP_4K, PTRS_PER_PTE);
137 	update_page_count(PG_DIRECT_MAP_1M, -1);
138 	return 0;
139 }
140 
141 static void modify_pmd_page(pmd_t *pmdp, unsigned long addr,
142 			    unsigned long flags)
143 {
144 	pmd_t new = *pmdp;
145 
146 	if (flags & SET_MEMORY_RO)
147 		new = pmd_wrprotect(new);
148 	else if (flags & SET_MEMORY_RW)
149 		new = pmd_mkwrite(pmd_mkdirty(new));
150 	if (flags & SET_MEMORY_NX)
151 		pmd_val(new) |= _SEGMENT_ENTRY_NOEXEC;
152 	else if (flags & SET_MEMORY_X)
153 		pmd_val(new) &= ~_SEGMENT_ENTRY_NOEXEC;
154 	pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT);
155 }
156 
157 static int walk_pmd_level(pud_t *pudp, unsigned long addr, unsigned long end,
158 			  unsigned long flags)
159 {
160 	unsigned long next;
161 	pmd_t *pmdp;
162 	int rc = 0;
163 
164 	pmdp = pmd_offset(pudp, addr);
165 	do {
166 		if (pmd_none(*pmdp))
167 			return -EINVAL;
168 		next = pmd_addr_end(addr, end);
169 		if (pmd_large(*pmdp)) {
170 			if (addr & ~PMD_MASK || addr + PMD_SIZE > next) {
171 				rc = split_pmd_page(pmdp, addr);
172 				if (rc)
173 					return rc;
174 				continue;
175 			}
176 			modify_pmd_page(pmdp, addr, flags);
177 		} else {
178 			rc = walk_pte_level(pmdp, addr, next, flags);
179 			if (rc)
180 				return rc;
181 		}
182 		pmdp++;
183 		addr = next;
184 		cond_resched();
185 	} while (addr < end);
186 	return rc;
187 }
188 
189 static int split_pud_page(pud_t *pudp, unsigned long addr)
190 {
191 	unsigned long pmd_addr, prot;
192 	pmd_t *pm_dir, *pmdp;
193 	pud_t new;
194 	int i, ro, nx;
195 
196 	pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY);
197 	if (!pm_dir)
198 		return -ENOMEM;
199 	pmd_addr = pud_pfn(*pudp) << PAGE_SHIFT;
200 	ro = !!(pud_val(*pudp) & _REGION_ENTRY_PROTECT);
201 	nx = !!(pud_val(*pudp) & _REGION_ENTRY_NOEXEC);
202 	prot = pgprot_val(ro ? SEGMENT_KERNEL_RO : SEGMENT_KERNEL);
203 	if (!nx)
204 		prot &= ~_SEGMENT_ENTRY_NOEXEC;
205 	pmdp = pm_dir;
206 	for (i = 0; i < PTRS_PER_PMD; i++) {
207 		pmd_val(*pmdp) = pmd_addr | prot;
208 		pmd_addr += PMD_SIZE;
209 		pmdp++;
210 	}
211 	pud_val(new) = __pa(pm_dir) | _REGION3_ENTRY;
212 	pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3);
213 	update_page_count(PG_DIRECT_MAP_1M, PTRS_PER_PMD);
214 	update_page_count(PG_DIRECT_MAP_2G, -1);
215 	return 0;
216 }
217 
218 static void modify_pud_page(pud_t *pudp, unsigned long addr,
219 			    unsigned long flags)
220 {
221 	pud_t new = *pudp;
222 
223 	if (flags & SET_MEMORY_RO)
224 		new = pud_wrprotect(new);
225 	else if (flags & SET_MEMORY_RW)
226 		new = pud_mkwrite(pud_mkdirty(new));
227 	if (flags & SET_MEMORY_NX)
228 		pud_val(new) |= _REGION_ENTRY_NOEXEC;
229 	else if (flags & SET_MEMORY_X)
230 		pud_val(new) &= ~_REGION_ENTRY_NOEXEC;
231 	pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3);
232 }
233 
234 static int walk_pud_level(p4d_t *p4d, unsigned long addr, unsigned long end,
235 			  unsigned long flags)
236 {
237 	unsigned long next;
238 	pud_t *pudp;
239 	int rc = 0;
240 
241 	pudp = pud_offset(p4d, addr);
242 	do {
243 		if (pud_none(*pudp))
244 			return -EINVAL;
245 		next = pud_addr_end(addr, end);
246 		if (pud_large(*pudp)) {
247 			if (addr & ~PUD_MASK || addr + PUD_SIZE > next) {
248 				rc = split_pud_page(pudp, addr);
249 				if (rc)
250 					break;
251 				continue;
252 			}
253 			modify_pud_page(pudp, addr, flags);
254 		} else {
255 			rc = walk_pmd_level(pudp, addr, next, flags);
256 		}
257 		pudp++;
258 		addr = next;
259 		cond_resched();
260 	} while (addr < end && !rc);
261 	return rc;
262 }
263 
264 static int walk_p4d_level(pgd_t *pgd, unsigned long addr, unsigned long end,
265 			  unsigned long flags)
266 {
267 	unsigned long next;
268 	p4d_t *p4dp;
269 	int rc = 0;
270 
271 	p4dp = p4d_offset(pgd, addr);
272 	do {
273 		if (p4d_none(*p4dp))
274 			return -EINVAL;
275 		next = p4d_addr_end(addr, end);
276 		rc = walk_pud_level(p4dp, addr, next, flags);
277 		p4dp++;
278 		addr = next;
279 		cond_resched();
280 	} while (addr < end && !rc);
281 	return rc;
282 }
283 
284 static DEFINE_MUTEX(cpa_mutex);
285 
286 static int change_page_attr(unsigned long addr, unsigned long end,
287 			    unsigned long flags)
288 {
289 	unsigned long next;
290 	int rc = -EINVAL;
291 	pgd_t *pgdp;
292 
293 	if (addr == end)
294 		return 0;
295 	if (end >= MODULES_END)
296 		return -EINVAL;
297 	mutex_lock(&cpa_mutex);
298 	pgdp = pgd_offset_k(addr);
299 	do {
300 		if (pgd_none(*pgdp))
301 			break;
302 		next = pgd_addr_end(addr, end);
303 		rc = walk_p4d_level(pgdp, addr, next, flags);
304 		if (rc)
305 			break;
306 		cond_resched();
307 	} while (pgdp++, addr = next, addr < end && !rc);
308 	mutex_unlock(&cpa_mutex);
309 	return rc;
310 }
311 
312 int __set_memory(unsigned long addr, int numpages, unsigned long flags)
313 {
314 	if (!MACHINE_HAS_NX)
315 		flags &= ~(SET_MEMORY_NX | SET_MEMORY_X);
316 	if (!flags)
317 		return 0;
318 	addr &= PAGE_MASK;
319 	return change_page_attr(addr, addr + numpages * PAGE_SIZE, flags);
320 }
321 
322 #ifdef CONFIG_DEBUG_PAGEALLOC
323 
324 static void ipte_range(pte_t *pte, unsigned long address, int nr)
325 {
326 	int i;
327 
328 	if (test_facility(13)) {
329 		__ptep_ipte_range(address, nr - 1, pte, IPTE_GLOBAL);
330 		return;
331 	}
332 	for (i = 0; i < nr; i++) {
333 		__ptep_ipte(address, pte, 0, 0, IPTE_GLOBAL);
334 		address += PAGE_SIZE;
335 		pte++;
336 	}
337 }
338 
339 void __kernel_map_pages(struct page *page, int numpages, int enable)
340 {
341 	unsigned long address;
342 	int nr, i, j;
343 	pgd_t *pgd;
344 	p4d_t *p4d;
345 	pud_t *pud;
346 	pmd_t *pmd;
347 	pte_t *pte;
348 
349 	for (i = 0; i < numpages;) {
350 		address = page_to_phys(page + i);
351 		pgd = pgd_offset_k(address);
352 		p4d = p4d_offset(pgd, address);
353 		pud = pud_offset(p4d, address);
354 		pmd = pmd_offset(pud, address);
355 		pte = pte_offset_kernel(pmd, address);
356 		nr = (unsigned long)pte >> ilog2(sizeof(long));
357 		nr = PTRS_PER_PTE - (nr & (PTRS_PER_PTE - 1));
358 		nr = min(numpages - i, nr);
359 		if (enable) {
360 			for (j = 0; j < nr; j++) {
361 				pte_val(*pte) &= ~_PAGE_INVALID;
362 				address += PAGE_SIZE;
363 				pte++;
364 			}
365 		} else {
366 			ipte_range(pte, address, nr);
367 		}
368 		i += nr;
369 	}
370 }
371 
372 #ifdef CONFIG_HIBERNATION
373 bool kernel_page_present(struct page *page)
374 {
375 	unsigned long addr;
376 	int cc;
377 
378 	addr = page_to_phys(page);
379 	asm volatile(
380 		"	lra	%1,0(%1)\n"
381 		"	ipm	%0\n"
382 		"	srl	%0,28"
383 		: "=d" (cc), "+a" (addr) : : "cc");
384 	return cc == 0;
385 }
386 #endif /* CONFIG_HIBERNATION */
387 
388 #endif /* CONFIG_DEBUG_PAGEALLOC */
389