pageattr.c

  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/cpufeature.h>
  7#include <linux/hugetlb.h>
  8#include <linux/proc_fs.h>
  9#include <linux/vmalloc.h>
 10#include <linux/mm.h>
 11#include <asm/cacheflush.h>
 12#include <asm/facility.h>
 13#include <asm/pgalloc.h>
 14#include <asm/kfence.h>
 15#include <asm/page.h>
 16#include <asm/asm.h>
 17#include <asm/set_memory.h>
 18
 19static inline unsigned long sske_frame(unsigned long addr, unsigned char skey)
 20{
 21	asm volatile(".insn rrf,0xb22b0000,%[skey],%[addr],1,0"
 22		     : [addr] "+a" (addr) : [skey] "d" (skey));
 23	return addr;
 24}
 25
 26void __storage_key_init_range(unsigned long start, unsigned long end)
 27{
 28	unsigned long boundary, size;
 29
 30	while (start < end) {
 31		if (cpu_has_edat1()) {
 32			/* set storage keys for a 1MB frame */
 33			size = 1UL << 20;
 34			boundary = (start + size) & ~(size - 1);
 35			if (boundary <= end) {
 36				do {
 37					start = sske_frame(start, PAGE_DEFAULT_KEY);
 38				} while (start < boundary);
 39				continue;
 40			}
 41		}
 42		page_set_storage_key(start, PAGE_DEFAULT_KEY, 1);
 43		start += PAGE_SIZE;
 44	}
 45}
 46
 47#ifdef CONFIG_PROC_FS
 48atomic_long_t __bootdata_preserved(direct_pages_count[PG_DIRECT_MAP_MAX]);
 49
 50void arch_report_meminfo(struct seq_file *m)
 51{
 52	seq_printf(m, "DirectMap4k:    %8lu kB\n",
 53		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_4K]) << 2);
 54	seq_printf(m, "DirectMap1M:    %8lu kB\n",
 55		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_1M]) << 10);
 56	seq_printf(m, "DirectMap2G:    %8lu kB\n",
 57		   atomic_long_read(&direct_pages_count[PG_DIRECT_MAP_2G]) << 21);
 58}
 59#endif /* CONFIG_PROC_FS */
 60
 61static void pgt_set(unsigned long *old, unsigned long new, unsigned long addr,
 62		    unsigned long dtt)
 63{
 64	unsigned long *table, mask;
 65
 66	mask = 0;
 67	if (cpu_has_edat2()) {
 68		switch (dtt) {
 69		case CRDTE_DTT_REGION3:
 70			mask = ~(PTRS_PER_PUD * sizeof(pud_t) - 1);
 71			break;
 72		case CRDTE_DTT_SEGMENT:
 73			mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
 74			break;
 75		case CRDTE_DTT_PAGE:
 76			mask = ~(PTRS_PER_PTE * sizeof(pte_t) - 1);
 77			break;
 78		}
 79		table = (unsigned long *)((unsigned long)old & mask);
 80		crdte(*old, new, table, dtt, addr, get_lowcore()->kernel_asce.val);
 81	} else {
 82		cspg(old, *old, new);
 83	}
 84}
 85
 86static 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	if (flags == SET_MEMORY_4K)
 92		return 0;
 93	ptep = pte_offset_kernel(pmdp, addr);
 94	do {
 95		new = *ptep;
 96		if (pte_none(new))
 97			return -EINVAL;
 98		if (flags & SET_MEMORY_RO)
 99			new = pte_wrprotect(new);
100		else if (flags & SET_MEMORY_RW)
101			new = pte_mkwrite_novma(pte_mkdirty(new));
102		if (flags & SET_MEMORY_NX)
103			new = set_pte_bit(new, __pgprot(_PAGE_NOEXEC));
104		else if (flags & SET_MEMORY_X)
105			new = clear_pte_bit(new, __pgprot(_PAGE_NOEXEC));
106		if (flags & SET_MEMORY_INV) {
107			new = set_pte_bit(new, __pgprot(_PAGE_INVALID));
108		} else if (flags & SET_MEMORY_DEF) {
109			new = __pte(pte_val(new) & PAGE_MASK);
110			new = set_pte_bit(new, PAGE_KERNEL);
111		}
112		pgt_set((unsigned long *)ptep, pte_val(new), addr, CRDTE_DTT_PAGE);
113		ptep++;
114		addr += PAGE_SIZE;
115		cond_resched();
116	} while (addr < end);
117	return 0;
118}
119
120static int split_pmd_page(pmd_t *pmdp, unsigned long addr)
121{
122	unsigned long pte_addr, prot;
123	pte_t *pt_dir, *ptep;
124	pmd_t new;
125	int i, ro, nx;
126
127	pt_dir = vmem_pte_alloc();
128	if (!pt_dir)
129		return -ENOMEM;
130	pte_addr = pmd_pfn(*pmdp) << PAGE_SHIFT;
131	ro = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT);
132	nx = !!(pmd_val(*pmdp) & _SEGMENT_ENTRY_NOEXEC);
133	prot = pgprot_val(ro ? PAGE_KERNEL_RO : PAGE_KERNEL);
134	if (!nx)
135		prot &= ~_PAGE_NOEXEC;
136	ptep = pt_dir;
137	for (i = 0; i < PTRS_PER_PTE; i++) {
138		set_pte(ptep, __pte(pte_addr | prot));
139		pte_addr += PAGE_SIZE;
140		ptep++;
141	}
142	new = __pmd(__pa(pt_dir) | _SEGMENT_ENTRY);
143	pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT);
144	update_page_count(PG_DIRECT_MAP_4K, PTRS_PER_PTE);
145	update_page_count(PG_DIRECT_MAP_1M, -1);
146	return 0;
147}
148
149static void modify_pmd_page(pmd_t *pmdp, unsigned long addr,
150			    unsigned long flags)
151{
152	pmd_t new = *pmdp;
153
154	if (flags & SET_MEMORY_RO)
155		new = pmd_wrprotect(new);
156	else if (flags & SET_MEMORY_RW)
157		new = pmd_mkwrite_novma(pmd_mkdirty(new));
158	if (flags & SET_MEMORY_NX)
159		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_NOEXEC));
160	else if (flags & SET_MEMORY_X)
161		new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_NOEXEC));
162	if (flags & SET_MEMORY_INV) {
163		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
164	} else if (flags & SET_MEMORY_DEF) {
165		new = __pmd(pmd_val(new) & PMD_MASK);
166		new = set_pmd_bit(new, SEGMENT_KERNEL);
167	}
168	pgt_set((unsigned long *)pmdp, pmd_val(new), addr, CRDTE_DTT_SEGMENT);
169}
170
171static int walk_pmd_level(pud_t *pudp, unsigned long addr, unsigned long end,
172			  unsigned long flags)
173{
174	unsigned long next;
175	int need_split;
176	pmd_t *pmdp;
177	int rc = 0;
178
179	pmdp = pmd_offset(pudp, addr);
180	do {
181		if (pmd_none(*pmdp))
182			return -EINVAL;
183		next = pmd_addr_end(addr, end);
184		if (pmd_leaf(*pmdp)) {
185			need_split  = !!(flags & SET_MEMORY_4K);
186			need_split |= !!(addr & ~PMD_MASK);
187			need_split |= !!(addr + PMD_SIZE > next);
188			if (need_split) {
189				rc = split_pmd_page(pmdp, addr);
190				if (rc)
191					return rc;
192				continue;
193			}
194			modify_pmd_page(pmdp, addr, flags);
195		} else {
196			rc = walk_pte_level(pmdp, addr, next, flags);
197			if (rc)
198				return rc;
199		}
200		pmdp++;
201		addr = next;
202		cond_resched();
203	} while (addr < end);
204	return rc;
205}
206
207int split_pud_page(pud_t *pudp, unsigned long addr)
208{
209	unsigned long pmd_addr, prot;
210	pmd_t *pm_dir, *pmdp;
211	pud_t new;
212	int i, ro, nx;
213
214	pm_dir = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY);
215	if (!pm_dir)
216		return -ENOMEM;
217	pmd_addr = pud_pfn(*pudp) << PAGE_SHIFT;
218	ro = !!(pud_val(*pudp) & _REGION_ENTRY_PROTECT);
219	nx = !!(pud_val(*pudp) & _REGION_ENTRY_NOEXEC);
220	prot = pgprot_val(ro ? SEGMENT_KERNEL_RO : SEGMENT_KERNEL);
221	if (!nx)
222		prot &= ~_SEGMENT_ENTRY_NOEXEC;
223	pmdp = pm_dir;
224	for (i = 0; i < PTRS_PER_PMD; i++) {
225		set_pmd(pmdp, __pmd(pmd_addr | prot));
226		pmd_addr += PMD_SIZE;
227		pmdp++;
228	}
229	new = __pud(__pa(pm_dir) | _REGION3_ENTRY);
230	pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3);
231	update_page_count(PG_DIRECT_MAP_1M, PTRS_PER_PMD);
232	update_page_count(PG_DIRECT_MAP_2G, -1);
233	return 0;
234}
235
236static void modify_pud_page(pud_t *pudp, unsigned long addr,
237			    unsigned long flags)
238{
239	pud_t new = *pudp;
240
241	if (flags & SET_MEMORY_RO)
242		new = pud_wrprotect(new);
243	else if (flags & SET_MEMORY_RW)
244		new = pud_mkwrite(pud_mkdirty(new));
245	if (flags & SET_MEMORY_NX)
246		new = set_pud_bit(new, __pgprot(_REGION_ENTRY_NOEXEC));
247	else if (flags & SET_MEMORY_X)
248		new = clear_pud_bit(new, __pgprot(_REGION_ENTRY_NOEXEC));
249	if (flags & SET_MEMORY_INV) {
250		new = set_pud_bit(new, __pgprot(_REGION_ENTRY_INVALID));
251	} else if (flags & SET_MEMORY_DEF) {
252		new = __pud(pud_val(new) & PUD_MASK);
253		new = set_pud_bit(new, REGION3_KERNEL);
254	}
255	pgt_set((unsigned long *)pudp, pud_val(new), addr, CRDTE_DTT_REGION3);
256}
257
258static int walk_pud_level(p4d_t *p4d, unsigned long addr, unsigned long end,
259			  unsigned long flags)
260{
261	unsigned long next;
262	int need_split;
263	pud_t *pudp;
264	int rc = 0;
265
266	pudp = pud_offset(p4d, addr);
267	do {
268		if (pud_none(*pudp))
269			return -EINVAL;
270		next = pud_addr_end(addr, end);
271		if (pud_leaf(*pudp)) {
272			need_split  = !!(flags & SET_MEMORY_4K);
273			need_split |= !!(addr & ~PUD_MASK);
274			need_split |= !!(addr + PUD_SIZE > next);
275			if (need_split) {
276				rc = split_pud_page(pudp, addr);
277				if (rc)
278					break;
279				continue;
280			}
281			modify_pud_page(pudp, addr, flags);
282		} else {
283			rc = walk_pmd_level(pudp, addr, next, flags);
284		}
285		pudp++;
286		addr = next;
287		cond_resched();
288	} while (addr < end && !rc);
289	return rc;
290}
291
292static int walk_p4d_level(pgd_t *pgd, unsigned long addr, unsigned long end,
293			  unsigned long flags)
294{
295	unsigned long next;
296	p4d_t *p4dp;
297	int rc = 0;
298
299	p4dp = p4d_offset(pgd, addr);
300	do {
301		if (p4d_none(*p4dp))
302			return -EINVAL;
303		next = p4d_addr_end(addr, end);
304		rc = walk_pud_level(p4dp, addr, next, flags);
305		p4dp++;
306		addr = next;
307		cond_resched();
308	} while (addr < end && !rc);
309	return rc;
310}
311
312DEFINE_MUTEX(cpa_mutex);
313
314static int change_page_attr(unsigned long addr, unsigned long end,
315			    unsigned long flags)
316{
317	unsigned long next;
318	int rc = -EINVAL;
319	pgd_t *pgdp;
320
321	pgdp = pgd_offset_k(addr);
322	do {
323		if (pgd_none(*pgdp))
324			break;
325		next = pgd_addr_end(addr, end);
326		rc = walk_p4d_level(pgdp, addr, next, flags);
327		if (rc)
328			break;
329		cond_resched();
330	} while (pgdp++, addr = next, addr < end && !rc);
331	return rc;
332}
333
334static int change_page_attr_alias(unsigned long addr, unsigned long end,
335				  unsigned long flags)
336{
337	unsigned long alias, offset, va_start, va_end;
338	struct vm_struct *area;
339	int rc = 0;
340
341	/*
342	 * Changes to read-only permissions on kernel VA mappings are also
343	 * applied to the kernel direct mapping. Execute permissions are
344	 * intentionally not transferred to keep all allocated pages within
345	 * the direct mapping non-executable.
346	 */
347	flags &= SET_MEMORY_RO | SET_MEMORY_RW;
348	if (!flags)
349		return 0;
350	area = NULL;
351	while (addr < end) {
352		if (!area)
353			area = find_vm_area((void *)addr);
354		if (!area || !(area->flags & VM_ALLOC))
355			return 0;
356		va_start = (unsigned long)area->addr;
357		va_end = va_start + area->nr_pages * PAGE_SIZE;
358		offset = (addr - va_start) >> PAGE_SHIFT;
359		alias = (unsigned long)page_address(area->pages[offset]);
360		rc = change_page_attr(alias, alias + PAGE_SIZE, flags);
361		if (rc)
362			break;
363		addr += PAGE_SIZE;
364		if (addr >= va_end)
365			area = NULL;
366	}
367	return rc;
368}
369
370int __set_memory(unsigned long addr, unsigned long numpages, unsigned long flags)
371{
372	unsigned long end;
373	int rc;
374
375	if (!cpu_has_nx())
376		flags &= ~(SET_MEMORY_NX | SET_MEMORY_X);
377	if (!flags)
378		return 0;
379	if (!numpages)
380		return 0;
381	addr &= PAGE_MASK;
382	end = addr + numpages * PAGE_SIZE;
383	mutex_lock(&cpa_mutex);
384	rc = change_page_attr(addr, end, flags);
385	if (rc)
386		goto out;
387	rc = change_page_attr_alias(addr, end, flags);
388out:
389	mutex_unlock(&cpa_mutex);
390	return rc;
391}
392
393int set_direct_map_invalid_noflush(struct page *page)
394{
395	return __set_memory((unsigned long)page_to_virt(page), 1, SET_MEMORY_INV);
396}
397
398int set_direct_map_default_noflush(struct page *page)
399{
400	return __set_memory((unsigned long)page_to_virt(page), 1, SET_MEMORY_DEF);
401}
402
403int set_direct_map_valid_noflush(struct page *page, unsigned nr, bool valid)
404{
405	unsigned long flags;
406
407	if (valid)
408		flags = SET_MEMORY_DEF;
409	else
410		flags = SET_MEMORY_INV;
411
412	return __set_memory((unsigned long)page_to_virt(page), nr, flags);
413}
414
415bool kernel_page_present(struct page *page)
416{
417	unsigned long addr;
418	unsigned int cc;
419
420	addr = (unsigned long)page_address(page);
421	asm volatile(
422		"	lra	%[addr],0(%[addr])\n"
423		CC_IPM(cc)
424		: CC_OUT(cc, cc), [addr] "+a" (addr)
425		:
426		: CC_CLOBBER);
427	return CC_TRANSFORM(cc) == 0;
428}
429
430#if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KFENCE)
431
432static void ipte_range(pte_t *pte, unsigned long address, int nr)
433{
434	int i;
435
436	if (test_facility(13)) {
437		__ptep_ipte_range(address, nr - 1, pte, IPTE_GLOBAL);
438		return;
439	}
440	for (i = 0; i < nr; i++) {
441		__ptep_ipte(address, pte, 0, 0, IPTE_GLOBAL);
442		address += PAGE_SIZE;
443		pte++;
444	}
445}
446
447void __kernel_map_pages(struct page *page, int numpages, int enable)
448{
449	unsigned long address;
450	pte_t *ptep, pte;
451	int nr, i, j;
452
453	for (i = 0; i < numpages;) {
454		address = (unsigned long)page_to_virt(page + i);
455		ptep = virt_to_kpte(address);
456		nr = (unsigned long)ptep >> ilog2(sizeof(long));
457		nr = PTRS_PER_PTE - (nr & (PTRS_PER_PTE - 1));
458		nr = min(numpages - i, nr);
459		if (enable) {
460			for (j = 0; j < nr; j++) {
461				pte = clear_pte_bit(*ptep, __pgprot(_PAGE_INVALID));
462				set_pte(ptep, pte);
463				address += PAGE_SIZE;
464				ptep++;
465			}
466		} else {
467			ipte_range(ptep, address, nr);
468		}
469		i += nr;
470	}
471}
472
473#endif /* CONFIG_DEBUG_PAGEALLOC */