xref: /linux/arch/s390/mm/pageattr.c (revision 1aea9b3f921003f0880f0676ae85d87c9f1cb4a2)
1 /*
2  * Copyright IBM Corp. 2011
3  * Author(s): Jan Glauber <jang@linux.vnet.ibm.com>
4  */
5 #include <linux/hugetlb.h>
6 #include <linux/mm.h>
7 #include <asm/cacheflush.h>
8 #include <asm/facility.h>
9 #include <asm/pgtable.h>
10 #include <asm/page.h>
11 #include <asm/set_memory.h>
12 
13 static inline unsigned long sske_frame(unsigned long addr, unsigned char skey)
14 {
15 	asm volatile(".insn rrf,0xb22b0000,%[skey],%[addr],9,0"
16 		     : [addr] "+a" (addr) : [skey] "d" (skey));
17 	return addr;
18 }
19 
20 void __storage_key_init_range(unsigned long start, unsigned long end)
21 {
22 	unsigned long boundary, size;
23 
24 	if (!PAGE_DEFAULT_KEY)
25 		return;
26 	while (start < end) {
27 		if (MACHINE_HAS_EDAT1) {
28 			/* set storage keys for a 1MB frame */
29 			size = 1UL << 20;
30 			boundary = (start + size) & ~(size - 1);
31 			if (boundary <= end) {
32 				do {
33 					start = sske_frame(start, PAGE_DEFAULT_KEY);
34 				} while (start < boundary);
35 				continue;
36 			}
37 		}
38 		page_set_storage_key(start, PAGE_DEFAULT_KEY, 0);
39 		start += PAGE_SIZE;
40 	}
41 }
42 
43 #ifdef CONFIG_PROC_FS
44 atomic_long_t direct_pages_count[PG_DIRECT_MAP_MAX];
45 
46 void arch_report_meminfo(struct seq_file *m)
47 {
48 	seq_printf(m, "DirectMap4k:    %8lu kB\n",
49 		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_4K]) << 2);
50 	seq_printf(m, "DirectMap1M:    %8lu kB\n",
51 		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_1M]) << 10);
52 	seq_printf(m, "DirectMap2G:    %8lu kB\n",
53 		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_2G]) << 21);
54 }
55 #endif /* CONFIG_PROC_FS */
56 
57 static void pgt_set(unsigned long *old, unsigned long new, unsigned long addr,
58 		    unsigned long dtt)
59 {
60 	unsigned long table, mask;
61 
62 	mask = 0;
63 	if (MACHINE_HAS_EDAT2) {
64 		switch (dtt) {
65 		case CRDTE_DTT_REGION3:
66 			mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1);
67 			break;
68 		case CRDTE_DTT_SEGMENT:
69 			mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
70 			break;
71 		case CRDTE_DTT_PAGE:
72 			mask = ~(PTRS_PER_PTE * sizeof(pte_t) - 1);
73 			break;
74 		}
75 		table = (unsigned long)old & mask;
76 		crdte(*old, new, table, dtt, addr, S390_lowcore.kernel_asce);
77 	} else if (MACHINE_HAS_IDTE) {
78 		cspg(old, *old, new);
79 	} else {
80 		csp((unsigned int *)old + 1, *old, new);
81 	}
82 }
83 
84 static int walk_pte_level(pmd_t *pmdp, unsigned long addr, unsigned long end,
85 			  unsigned long flags)
86 {
87 	pte_t *ptep, new;
88 
89 	ptep = pte_offset(pmdp, addr);
90 	do {
91 		new = *ptep;
92 		if (pte_none(new))
93 			return -EINVAL;
94 		if (flags & SET_MEMORY_RO)
95 			new = pte_wrprotect(new);
96 		else if (flags & SET_MEMORY_RW)
97 			new = pte_mkwrite(pte_mkdirty(new));
98 		if (flags & SET_MEMORY_NX)
99 			pte_val(new) |= _PAGE_NOEXEC;
100 		else if (flags & SET_MEMORY_X)
101 			pte_val(new) &= ~_PAGE_NOEXEC;
102 		pgt_set((unsigned long *)ptep, pte_val(new), addr, CRDTE_DTT_PAGE);
103 		ptep++;
104 		addr += PAGE_SIZE;
105 		cond_resched();
106 	} while (addr < end);
107 	return 0;
108 }
109 
110 static int split_pmd_page(pmd_t *pmdp, unsigned long addr)
111 {
112 	unsigned long pte_addr, prot;
113 	pte_t *pt_dir, *ptep;
114 	pmd_t new;
115 	int i, ro, nx;
116 
117 	pt_dir = vmem_pte_alloc();
118 	if (!pt_dir)
119 		return -ENOMEM;
120 	pte_addr = pmd_pfn(*pmdp) << PAGE_SHIFT;
121 	ro = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT);
122 	nx = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_NOEXEC);
123 	prot = pgprot_val(ro ? PAGE_KERNEL_RO : PAGE_KERNEL);
124 	if (!nx)
125 		prot &= ~_PAGE_NOEXEC;
126 	ptep = pt_dir;
127 	for (i = 0; i < PTRS_PER_PTE; i++) {
128 		pte_val(*ptep) = pte_addr | prot;
129 		pte_addr += PAGE_SIZE;
130 		ptep++;
131 	}
132 	pmd_val(new) = __pa(pt_dir) | _SEGMENT_ENTRY;
133 	pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT);
134 	update_page_count(PG_DIRECT_MAP_4K, PTRS_PER_PTE);
135 	update_page_count(PG_DIRECT_MAP_1M, -1);
136 	return 0;
137 }
138 
139 static void modify_pmd_page(pmd_t *pmdp, unsigned long addr,
140 			    unsigned long flags)
141 {
142 	pmd_t new = *pmdp;
143 
144 	if (flags & SET_MEMORY_RO)
145 		new = pmd_wrprotect(new);
146 	else if (flags & SET_MEMORY_RW)
147 		new = pmd_mkwrite(pmd_mkdirty(new));
148 	if (flags & SET_MEMORY_NX)
149 		pmd_val(new) |= _SEGMENT_ENTRY_NOEXEC;
150 	else if (flags & SET_MEMORY_X)
151 		pmd_val(new) &= ~_SEGMENT_ENTRY_NOEXEC;
152 	pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT);
153 }
154 
155 static int walk_pmd_level(pud_t *pudp, unsigned long addr, unsigned long end,
156 			  unsigned long flags)
157 {
158 	unsigned long next;
159 	pmd_t *pmdp;
160 	int rc = 0;
161 
162 	pmdp = pmd_offset(pudp, addr);
163 	do {
164 		if (pmd_none(*pmdp))
165 			return -EINVAL;
166 		next = pmd_addr_end(addr, end);
167 		if (pmd_large(*pmdp)) {
168 			if (addr & ~PMD_MASK || addr + PMD_SIZE > next) {
169 				rc = split_pmd_page(pmdp, addr);
170 				if (rc)
171 					return rc;
172 				continue;
173 			}
174 			modify_pmd_page(pmdp, addr, flags);
175 		} else {
176 			rc = walk_pte_level(pmdp, addr, next, flags);
177 			if (rc)
178 				return rc;
179 		}
180 		pmdp++;
181 		addr = next;
182 		cond_resched();
183 	} while (addr < end);
184 	return rc;
185 }
186 
187 static int split_pud_page(pud_t *pudp, unsigned long addr)
188 {
189 	unsigned long pmd_addr, prot;
190 	pmd_t *pm_dir, *pmdp;
191 	pud_t new;
192 	int i, ro, nx;
193 
194 	pm_dir = vmem_pmd_alloc();
195 	if (!pm_dir)
196 		return -ENOMEM;
197 	pmd_addr = pud_pfn(*pudp) << PAGE_SHIFT;
198 	ro = !!(pud_val(*pudp) & _REGION_ENTRY_PROTECT);
199 	nx = !!(pud_val(*pudp) & _REGION_ENTRY_NOEXEC);
200 	prot = pgprot_val(ro ? SEGMENT_KERNEL_RO : SEGMENT_KERNEL);
201 	if (!nx)
202 		prot &= ~_SEGMENT_ENTRY_NOEXEC;
203 	pmdp = pm_dir;
204 	for (i = 0; i < PTRS_PER_PMD; i++) {
205 		pmd_val(*pmdp) = pmd_addr | prot;
206 		pmd_addr += PMD_SIZE;
207 		pmdp++;
208 	}
209 	pud_val(new) = __pa(pm_dir) | _REGION3_ENTRY;
210 	pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3);
211 	update_page_count(PG_DIRECT_MAP_1M, PTRS_PER_PMD);
212 	update_page_count(PG_DIRECT_MAP_2G, -1);
213 	return 0;
214 }
215 
216 static void modify_pud_page(pud_t *pudp, unsigned long addr,
217 			    unsigned long flags)
218 {
219 	pud_t new = *pudp;
220 
221 	if (flags & SET_MEMORY_RO)
222 		new = pud_wrprotect(new);
223 	else if (flags & SET_MEMORY_RW)
224 		new = pud_mkwrite(pud_mkdirty(new));
225 	if (flags & SET_MEMORY_NX)
226 		pud_val(new) |= _REGION_ENTRY_NOEXEC;
227 	else if (flags & SET_MEMORY_X)
228 		pud_val(new) &= ~_REGION_ENTRY_NOEXEC;
229 	pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3);
230 }
231 
232 static int walk_pud_level(p4d_t *p4d, unsigned long addr, unsigned long end,
233 			  unsigned long flags)
234 {
235 	unsigned long next;
236 	pud_t *pudp;
237 	int rc = 0;
238 
239 	pudp = pud_offset(p4d, addr);
240 	do {
241 		if (pud_none(*pudp))
242 			return -EINVAL;
243 		next = pud_addr_end(addr, end);
244 		if (pud_large(*pudp)) {
245 			if (addr & ~PUD_MASK || addr + PUD_SIZE > next) {
246 				rc = split_pud_page(pudp, addr);
247 				if (rc)
248 					break;
249 				continue;
250 			}
251 			modify_pud_page(pudp, addr, flags);
252 		} else {
253 			rc = walk_pmd_level(pudp, addr, next, flags);
254 		}
255 		pudp++;
256 		addr = next;
257 		cond_resched();
258 	} while (addr < end && !rc);
259 	return rc;
260 }
261 
262 static int walk_p4d_level(pgd_t *pgd, unsigned long addr, unsigned long end,
263 			  unsigned long flags)
264 {
265 	unsigned long next;
266 	p4d_t *p4dp;
267 	int rc = 0;
268 
269 	p4dp = p4d_offset(pgd, addr);
270 	do {
271 		if (p4d_none(*p4dp))
272 			return -EINVAL;
273 		next = p4d_addr_end(addr, end);
274 		rc = walk_pud_level(p4dp, addr, next, flags);
275 		p4dp++;
276 		addr = next;
277 		cond_resched();
278 	} while (addr < end && !rc);
279 	return rc;
280 }
281 
282 static DEFINE_MUTEX(cpa_mutex);
283 
284 static int change_page_attr(unsigned long addr, unsigned long end,
285 			    unsigned long flags)
286 {
287 	unsigned long next;
288 	int rc = -EINVAL;
289 	pgd_t *pgdp;
290 
291 	if (addr == end)
292 		return 0;
293 	if (end >= MODULES_END)
294 		return -EINVAL;
295 	mutex_lock(&cpa_mutex);
296 	pgdp = pgd_offset_k(addr);
297 	do {
298 		if (pgd_none(*pgdp))
299 			break;
300 		next = pgd_addr_end(addr, end);
301 		rc = walk_p4d_level(pgdp, addr, next, flags);
302 		if (rc)
303 			break;
304 		cond_resched();
305 	} while (pgdp++, addr = next, addr < end && !rc);
306 	mutex_unlock(&cpa_mutex);
307 	return rc;
308 }
309 
310 int __set_memory(unsigned long addr, int numpages, unsigned long flags)
311 {
312 	if (!MACHINE_HAS_NX)
313 		flags &= ~(SET_MEMORY_NX | SET_MEMORY_X);
314 	if (!flags)
315 		return 0;
316 	addr &= PAGE_MASK;
317 	return change_page_attr(addr, addr + numpages * PAGE_SIZE, flags);
318 }
319 
320 #ifdef CONFIG_DEBUG_PAGEALLOC
321 
322 static void ipte_range(pte_t *pte, unsigned long address, int nr)
323 {
324 	int i;
325 
326 	if (test_facility(13)) {
327 		__ptep_ipte_range(address, nr - 1, pte, IPTE_GLOBAL);
328 		return;
329 	}
330 	for (i = 0; i < nr; i++) {
331 		__ptep_ipte(address, pte, IPTE_GLOBAL);
332 		address += PAGE_SIZE;
333 		pte++;
334 	}
335 }
336 
337 void __kernel_map_pages(struct page *page, int numpages, int enable)
338 {
339 	unsigned long address;
340 	int nr, i, j;
341 	pgd_t *pgd;
342 	p4d_t *p4d;
343 	pud_t *pud;
344 	pmd_t *pmd;
345 	pte_t *pte;
346 
347 	for (i = 0; i < numpages;) {
348 		address = page_to_phys(page + i);
349 		pgd = pgd_offset_k(address);
350 		p4d = p4d_offset(pgd, address);
351 		pud = pud_offset(p4d, address);
352 		pmd = pmd_offset(pud, address);
353 		pte = pte_offset_kernel(pmd, address);
354 		nr = (unsigned long)pte >> ilog2(sizeof(long));
355 		nr = PTRS_PER_PTE - (nr & (PTRS_PER_PTE - 1));
356 		nr = min(numpages - i, nr);
357 		if (enable) {
358 			for (j = 0; j < nr; j++) {
359 				pte_val(*pte) &= ~_PAGE_INVALID;
360 				address += PAGE_SIZE;
361 				pte++;
362 			}
363 		} else {
364 			ipte_range(pte, address, nr);
365 		}
366 		i += nr;
367 	}
368 }
369 
370 #ifdef CONFIG_HIBERNATION
371 bool kernel_page_present(struct page *page)
372 {
373 	unsigned long addr;
374 	int cc;
375 
376 	addr = page_to_phys(page);
377 	asm volatile(
378 		"	lra	%1,0(%1)\n"
379 		"	ipm	%0\n"
380 		"	srl	%0,28"
381 		: "=d" (cc), "+a" (addr) : : "cc");
382 	return cc == 0;
383 }
384 #endif /* CONFIG_HIBERNATION */
385 
386 #endif /* CONFIG_DEBUG_PAGEALLOC */
387