xref: /linux/arch/s390/mm/pgtable.c (revision 856e7c4b619af622d56b3b454f7bec32a170ac99)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *    Copyright IBM Corp. 2007, 2011
4  *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
5  */
6 
7 #include <linux/sched.h>
8 #include <linux/kernel.h>
9 #include <linux/errno.h>
10 #include <linux/gfp.h>
11 #include <linux/mm.h>
12 #include <linux/swap.h>
13 #include <linux/smp.h>
14 #include <linux/spinlock.h>
15 #include <linux/rcupdate.h>
16 #include <linux/slab.h>
17 #include <linux/swapops.h>
18 #include <linux/sysctl.h>
19 #include <linux/ksm.h>
20 #include <linux/mman.h>
21 
22 #include <asm/pgtable.h>
23 #include <asm/pgalloc.h>
24 #include <asm/tlb.h>
25 #include <asm/tlbflush.h>
26 #include <asm/mmu_context.h>
27 #include <asm/page-states.h>
28 
29 static inline void ptep_ipte_local(struct mm_struct *mm, unsigned long addr,
30 				   pte_t *ptep, int nodat)
31 {
32 	unsigned long opt, asce;
33 
34 	if (MACHINE_HAS_TLB_GUEST) {
35 		opt = 0;
36 		asce = READ_ONCE(mm->context.gmap_asce);
37 		if (asce == 0UL || nodat)
38 			opt |= IPTE_NODAT;
39 		if (asce != -1UL) {
40 			asce = asce ? : mm->context.asce;
41 			opt |= IPTE_GUEST_ASCE;
42 		}
43 		__ptep_ipte(addr, ptep, opt, asce, IPTE_LOCAL);
44 	} else {
45 		__ptep_ipte(addr, ptep, 0, 0, IPTE_LOCAL);
46 	}
47 }
48 
49 static inline void ptep_ipte_global(struct mm_struct *mm, unsigned long addr,
50 				    pte_t *ptep, int nodat)
51 {
52 	unsigned long opt, asce;
53 
54 	if (MACHINE_HAS_TLB_GUEST) {
55 		opt = 0;
56 		asce = READ_ONCE(mm->context.gmap_asce);
57 		if (asce == 0UL || nodat)
58 			opt |= IPTE_NODAT;
59 		if (asce != -1UL) {
60 			asce = asce ? : mm->context.asce;
61 			opt |= IPTE_GUEST_ASCE;
62 		}
63 		__ptep_ipte(addr, ptep, opt, asce, IPTE_GLOBAL);
64 	} else {
65 		__ptep_ipte(addr, ptep, 0, 0, IPTE_GLOBAL);
66 	}
67 }
68 
69 static inline pte_t ptep_flush_direct(struct mm_struct *mm,
70 				      unsigned long addr, pte_t *ptep,
71 				      int nodat)
72 {
73 	pte_t old;
74 
75 	old = *ptep;
76 	if (unlikely(pte_val(old) & _PAGE_INVALID))
77 		return old;
78 	atomic_inc(&mm->context.flush_count);
79 	if (MACHINE_HAS_TLB_LC &&
80 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
81 		ptep_ipte_local(mm, addr, ptep, nodat);
82 	else
83 		ptep_ipte_global(mm, addr, ptep, nodat);
84 	atomic_dec(&mm->context.flush_count);
85 	return old;
86 }
87 
88 static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
89 				    unsigned long addr, pte_t *ptep,
90 				    int nodat)
91 {
92 	pte_t old;
93 
94 	old = *ptep;
95 	if (unlikely(pte_val(old) & _PAGE_INVALID))
96 		return old;
97 	atomic_inc(&mm->context.flush_count);
98 	if (cpumask_equal(&mm->context.cpu_attach_mask,
99 			  cpumask_of(smp_processor_id()))) {
100 		pte_val(*ptep) |= _PAGE_INVALID;
101 		mm->context.flush_mm = 1;
102 	} else
103 		ptep_ipte_global(mm, addr, ptep, nodat);
104 	atomic_dec(&mm->context.flush_count);
105 	return old;
106 }
107 
108 static inline pgste_t pgste_get_lock(pte_t *ptep)
109 {
110 	unsigned long new = 0;
111 #ifdef CONFIG_PGSTE
112 	unsigned long old;
113 
114 	asm(
115 		"	lg	%0,%2\n"
116 		"0:	lgr	%1,%0\n"
117 		"	nihh	%0,0xff7f\n"	/* clear PCL bit in old */
118 		"	oihh	%1,0x0080\n"	/* set PCL bit in new */
119 		"	csg	%0,%1,%2\n"
120 		"	jl	0b\n"
121 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
122 		: "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
123 #endif
124 	return __pgste(new);
125 }
126 
127 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
128 {
129 #ifdef CONFIG_PGSTE
130 	asm(
131 		"	nihh	%1,0xff7f\n"	/* clear PCL bit */
132 		"	stg	%1,%0\n"
133 		: "=Q" (ptep[PTRS_PER_PTE])
134 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
135 		: "cc", "memory");
136 #endif
137 }
138 
139 static inline pgste_t pgste_get(pte_t *ptep)
140 {
141 	unsigned long pgste = 0;
142 #ifdef CONFIG_PGSTE
143 	pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
144 #endif
145 	return __pgste(pgste);
146 }
147 
148 static inline void pgste_set(pte_t *ptep, pgste_t pgste)
149 {
150 #ifdef CONFIG_PGSTE
151 	*(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
152 #endif
153 }
154 
155 static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
156 				       struct mm_struct *mm)
157 {
158 #ifdef CONFIG_PGSTE
159 	unsigned long address, bits, skey;
160 
161 	if (!mm_uses_skeys(mm) || pte_val(pte) & _PAGE_INVALID)
162 		return pgste;
163 	address = pte_val(pte) & PAGE_MASK;
164 	skey = (unsigned long) page_get_storage_key(address);
165 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
166 	/* Transfer page changed & referenced bit to guest bits in pgste */
167 	pgste_val(pgste) |= bits << 48;		/* GR bit & GC bit */
168 	/* Copy page access key and fetch protection bit to pgste */
169 	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
170 	pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
171 #endif
172 	return pgste;
173 
174 }
175 
176 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
177 				 struct mm_struct *mm)
178 {
179 #ifdef CONFIG_PGSTE
180 	unsigned long address;
181 	unsigned long nkey;
182 
183 	if (!mm_uses_skeys(mm) || pte_val(entry) & _PAGE_INVALID)
184 		return;
185 	VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
186 	address = pte_val(entry) & PAGE_MASK;
187 	/*
188 	 * Set page access key and fetch protection bit from pgste.
189 	 * The guest C/R information is still in the PGSTE, set real
190 	 * key C/R to 0.
191 	 */
192 	nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
193 	nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
194 	page_set_storage_key(address, nkey, 0);
195 #endif
196 }
197 
198 static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
199 {
200 #ifdef CONFIG_PGSTE
201 	if ((pte_val(entry) & _PAGE_PRESENT) &&
202 	    (pte_val(entry) & _PAGE_WRITE) &&
203 	    !(pte_val(entry) & _PAGE_INVALID)) {
204 		if (!MACHINE_HAS_ESOP) {
205 			/*
206 			 * Without enhanced suppression-on-protection force
207 			 * the dirty bit on for all writable ptes.
208 			 */
209 			pte_val(entry) |= _PAGE_DIRTY;
210 			pte_val(entry) &= ~_PAGE_PROTECT;
211 		}
212 		if (!(pte_val(entry) & _PAGE_PROTECT))
213 			/* This pte allows write access, set user-dirty */
214 			pgste_val(pgste) |= PGSTE_UC_BIT;
215 	}
216 #endif
217 	*ptep = entry;
218 	return pgste;
219 }
220 
221 static inline pgste_t pgste_pte_notify(struct mm_struct *mm,
222 				       unsigned long addr,
223 				       pte_t *ptep, pgste_t pgste)
224 {
225 #ifdef CONFIG_PGSTE
226 	unsigned long bits;
227 
228 	bits = pgste_val(pgste) & (PGSTE_IN_BIT | PGSTE_VSIE_BIT);
229 	if (bits) {
230 		pgste_val(pgste) ^= bits;
231 		ptep_notify(mm, addr, ptep, bits);
232 	}
233 #endif
234 	return pgste;
235 }
236 
237 static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
238 				      unsigned long addr, pte_t *ptep)
239 {
240 	pgste_t pgste = __pgste(0);
241 
242 	if (mm_has_pgste(mm)) {
243 		pgste = pgste_get_lock(ptep);
244 		pgste = pgste_pte_notify(mm, addr, ptep, pgste);
245 	}
246 	return pgste;
247 }
248 
249 static inline pte_t ptep_xchg_commit(struct mm_struct *mm,
250 				    unsigned long addr, pte_t *ptep,
251 				    pgste_t pgste, pte_t old, pte_t new)
252 {
253 	if (mm_has_pgste(mm)) {
254 		if (pte_val(old) & _PAGE_INVALID)
255 			pgste_set_key(ptep, pgste, new, mm);
256 		if (pte_val(new) & _PAGE_INVALID) {
257 			pgste = pgste_update_all(old, pgste, mm);
258 			if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
259 			    _PGSTE_GPS_USAGE_UNUSED)
260 				pte_val(old) |= _PAGE_UNUSED;
261 		}
262 		pgste = pgste_set_pte(ptep, pgste, new);
263 		pgste_set_unlock(ptep, pgste);
264 	} else {
265 		*ptep = new;
266 	}
267 	return old;
268 }
269 
270 pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
271 		       pte_t *ptep, pte_t new)
272 {
273 	pgste_t pgste;
274 	pte_t old;
275 	int nodat;
276 
277 	preempt_disable();
278 	pgste = ptep_xchg_start(mm, addr, ptep);
279 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
280 	old = ptep_flush_direct(mm, addr, ptep, nodat);
281 	old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
282 	preempt_enable();
283 	return old;
284 }
285 EXPORT_SYMBOL(ptep_xchg_direct);
286 
287 pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
288 		     pte_t *ptep, pte_t new)
289 {
290 	pgste_t pgste;
291 	pte_t old;
292 	int nodat;
293 
294 	preempt_disable();
295 	pgste = ptep_xchg_start(mm, addr, ptep);
296 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
297 	old = ptep_flush_lazy(mm, addr, ptep, nodat);
298 	old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
299 	preempt_enable();
300 	return old;
301 }
302 EXPORT_SYMBOL(ptep_xchg_lazy);
303 
304 pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
305 			     pte_t *ptep)
306 {
307 	pgste_t pgste;
308 	pte_t old;
309 	int nodat;
310 
311 	preempt_disable();
312 	pgste = ptep_xchg_start(mm, addr, ptep);
313 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
314 	old = ptep_flush_lazy(mm, addr, ptep, nodat);
315 	if (mm_has_pgste(mm)) {
316 		pgste = pgste_update_all(old, pgste, mm);
317 		pgste_set(ptep, pgste);
318 	}
319 	return old;
320 }
321 EXPORT_SYMBOL(ptep_modify_prot_start);
322 
323 void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
324 			     pte_t *ptep, pte_t pte)
325 {
326 	pgste_t pgste;
327 
328 	if (!MACHINE_HAS_NX)
329 		pte_val(pte) &= ~_PAGE_NOEXEC;
330 	if (mm_has_pgste(mm)) {
331 		pgste = pgste_get(ptep);
332 		pgste_set_key(ptep, pgste, pte, mm);
333 		pgste = pgste_set_pte(ptep, pgste, pte);
334 		pgste_set_unlock(ptep, pgste);
335 	} else {
336 		*ptep = pte;
337 	}
338 	preempt_enable();
339 }
340 EXPORT_SYMBOL(ptep_modify_prot_commit);
341 
342 static inline void pmdp_idte_local(struct mm_struct *mm,
343 				   unsigned long addr, pmd_t *pmdp)
344 {
345 	if (MACHINE_HAS_TLB_GUEST)
346 		__pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
347 			    mm->context.asce, IDTE_LOCAL);
348 	else
349 		__pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
350 }
351 
352 static inline void pmdp_idte_global(struct mm_struct *mm,
353 				    unsigned long addr, pmd_t *pmdp)
354 {
355 	if (MACHINE_HAS_TLB_GUEST)
356 		__pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
357 			    mm->context.asce, IDTE_GLOBAL);
358 	else if (MACHINE_HAS_IDTE)
359 		__pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
360 	else
361 		__pmdp_csp(pmdp);
362 }
363 
364 static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
365 				      unsigned long addr, pmd_t *pmdp)
366 {
367 	pmd_t old;
368 
369 	old = *pmdp;
370 	if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
371 		return old;
372 	atomic_inc(&mm->context.flush_count);
373 	if (MACHINE_HAS_TLB_LC &&
374 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
375 		pmdp_idte_local(mm, addr, pmdp);
376 	else
377 		pmdp_idte_global(mm, addr, pmdp);
378 	atomic_dec(&mm->context.flush_count);
379 	return old;
380 }
381 
382 static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
383 				    unsigned long addr, pmd_t *pmdp)
384 {
385 	pmd_t old;
386 
387 	old = *pmdp;
388 	if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
389 		return old;
390 	atomic_inc(&mm->context.flush_count);
391 	if (cpumask_equal(&mm->context.cpu_attach_mask,
392 			  cpumask_of(smp_processor_id()))) {
393 		pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
394 		mm->context.flush_mm = 1;
395 	} else {
396 		pmdp_idte_global(mm, addr, pmdp);
397 	}
398 	atomic_dec(&mm->context.flush_count);
399 	return old;
400 }
401 
402 pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
403 		       pmd_t *pmdp, pmd_t new)
404 {
405 	pmd_t old;
406 
407 	preempt_disable();
408 	old = pmdp_flush_direct(mm, addr, pmdp);
409 	*pmdp = new;
410 	preempt_enable();
411 	return old;
412 }
413 EXPORT_SYMBOL(pmdp_xchg_direct);
414 
415 pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
416 		     pmd_t *pmdp, pmd_t new)
417 {
418 	pmd_t old;
419 
420 	preempt_disable();
421 	old = pmdp_flush_lazy(mm, addr, pmdp);
422 	*pmdp = new;
423 	preempt_enable();
424 	return old;
425 }
426 EXPORT_SYMBOL(pmdp_xchg_lazy);
427 
428 static inline void pudp_idte_local(struct mm_struct *mm,
429 				   unsigned long addr, pud_t *pudp)
430 {
431 	if (MACHINE_HAS_TLB_GUEST)
432 		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
433 			    mm->context.asce, IDTE_LOCAL);
434 	else
435 		__pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
436 }
437 
438 static inline void pudp_idte_global(struct mm_struct *mm,
439 				    unsigned long addr, pud_t *pudp)
440 {
441 	if (MACHINE_HAS_TLB_GUEST)
442 		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
443 			    mm->context.asce, IDTE_GLOBAL);
444 	else if (MACHINE_HAS_IDTE)
445 		__pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
446 	else
447 		/*
448 		 * Invalid bit position is the same for pmd and pud, so we can
449 		 * re-use _pmd_csp() here
450 		 */
451 		__pmdp_csp((pmd_t *) pudp);
452 }
453 
454 static inline pud_t pudp_flush_direct(struct mm_struct *mm,
455 				      unsigned long addr, pud_t *pudp)
456 {
457 	pud_t old;
458 
459 	old = *pudp;
460 	if (pud_val(old) & _REGION_ENTRY_INVALID)
461 		return old;
462 	atomic_inc(&mm->context.flush_count);
463 	if (MACHINE_HAS_TLB_LC &&
464 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
465 		pudp_idte_local(mm, addr, pudp);
466 	else
467 		pudp_idte_global(mm, addr, pudp);
468 	atomic_dec(&mm->context.flush_count);
469 	return old;
470 }
471 
472 pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
473 		       pud_t *pudp, pud_t new)
474 {
475 	pud_t old;
476 
477 	preempt_disable();
478 	old = pudp_flush_direct(mm, addr, pudp);
479 	*pudp = new;
480 	preempt_enable();
481 	return old;
482 }
483 EXPORT_SYMBOL(pudp_xchg_direct);
484 
485 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
486 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
487 				pgtable_t pgtable)
488 {
489 	struct list_head *lh = (struct list_head *) pgtable;
490 
491 	assert_spin_locked(pmd_lockptr(mm, pmdp));
492 
493 	/* FIFO */
494 	if (!pmd_huge_pte(mm, pmdp))
495 		INIT_LIST_HEAD(lh);
496 	else
497 		list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
498 	pmd_huge_pte(mm, pmdp) = pgtable;
499 }
500 
501 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
502 {
503 	struct list_head *lh;
504 	pgtable_t pgtable;
505 	pte_t *ptep;
506 
507 	assert_spin_locked(pmd_lockptr(mm, pmdp));
508 
509 	/* FIFO */
510 	pgtable = pmd_huge_pte(mm, pmdp);
511 	lh = (struct list_head *) pgtable;
512 	if (list_empty(lh))
513 		pmd_huge_pte(mm, pmdp) = NULL;
514 	else {
515 		pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
516 		list_del(lh);
517 	}
518 	ptep = (pte_t *) pgtable;
519 	pte_val(*ptep) = _PAGE_INVALID;
520 	ptep++;
521 	pte_val(*ptep) = _PAGE_INVALID;
522 	return pgtable;
523 }
524 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
525 
526 #ifdef CONFIG_PGSTE
527 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
528 		     pte_t *ptep, pte_t entry)
529 {
530 	pgste_t pgste;
531 
532 	/* the mm_has_pgste() check is done in set_pte_at() */
533 	preempt_disable();
534 	pgste = pgste_get_lock(ptep);
535 	pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
536 	pgste_set_key(ptep, pgste, entry, mm);
537 	pgste = pgste_set_pte(ptep, pgste, entry);
538 	pgste_set_unlock(ptep, pgste);
539 	preempt_enable();
540 }
541 
542 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
543 {
544 	pgste_t pgste;
545 
546 	preempt_disable();
547 	pgste = pgste_get_lock(ptep);
548 	pgste_val(pgste) |= PGSTE_IN_BIT;
549 	pgste_set_unlock(ptep, pgste);
550 	preempt_enable();
551 }
552 
553 /**
554  * ptep_force_prot - change access rights of a locked pte
555  * @mm: pointer to the process mm_struct
556  * @addr: virtual address in the guest address space
557  * @ptep: pointer to the page table entry
558  * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
559  * @bit: pgste bit to set (e.g. for notification)
560  *
561  * Returns 0 if the access rights were changed and -EAGAIN if the current
562  * and requested access rights are incompatible.
563  */
564 int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
565 		    pte_t *ptep, int prot, unsigned long bit)
566 {
567 	pte_t entry;
568 	pgste_t pgste;
569 	int pte_i, pte_p, nodat;
570 
571 	pgste = pgste_get_lock(ptep);
572 	entry = *ptep;
573 	/* Check pte entry after all locks have been acquired */
574 	pte_i = pte_val(entry) & _PAGE_INVALID;
575 	pte_p = pte_val(entry) & _PAGE_PROTECT;
576 	if ((pte_i && (prot != PROT_NONE)) ||
577 	    (pte_p && (prot & PROT_WRITE))) {
578 		pgste_set_unlock(ptep, pgste);
579 		return -EAGAIN;
580 	}
581 	/* Change access rights and set pgste bit */
582 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
583 	if (prot == PROT_NONE && !pte_i) {
584 		ptep_flush_direct(mm, addr, ptep, nodat);
585 		pgste = pgste_update_all(entry, pgste, mm);
586 		pte_val(entry) |= _PAGE_INVALID;
587 	}
588 	if (prot == PROT_READ && !pte_p) {
589 		ptep_flush_direct(mm, addr, ptep, nodat);
590 		pte_val(entry) &= ~_PAGE_INVALID;
591 		pte_val(entry) |= _PAGE_PROTECT;
592 	}
593 	pgste_val(pgste) |= bit;
594 	pgste = pgste_set_pte(ptep, pgste, entry);
595 	pgste_set_unlock(ptep, pgste);
596 	return 0;
597 }
598 
599 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
600 		    pte_t *sptep, pte_t *tptep, pte_t pte)
601 {
602 	pgste_t spgste, tpgste;
603 	pte_t spte, tpte;
604 	int rc = -EAGAIN;
605 
606 	if (!(pte_val(*tptep) & _PAGE_INVALID))
607 		return 0;	/* already shadowed */
608 	spgste = pgste_get_lock(sptep);
609 	spte = *sptep;
610 	if (!(pte_val(spte) & _PAGE_INVALID) &&
611 	    !((pte_val(spte) & _PAGE_PROTECT) &&
612 	      !(pte_val(pte) & _PAGE_PROTECT))) {
613 		pgste_val(spgste) |= PGSTE_VSIE_BIT;
614 		tpgste = pgste_get_lock(tptep);
615 		pte_val(tpte) = (pte_val(spte) & PAGE_MASK) |
616 				(pte_val(pte) & _PAGE_PROTECT);
617 		/* don't touch the storage key - it belongs to parent pgste */
618 		tpgste = pgste_set_pte(tptep, tpgste, tpte);
619 		pgste_set_unlock(tptep, tpgste);
620 		rc = 1;
621 	}
622 	pgste_set_unlock(sptep, spgste);
623 	return rc;
624 }
625 
626 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
627 {
628 	pgste_t pgste;
629 	int nodat;
630 
631 	pgste = pgste_get_lock(ptep);
632 	/* notifier is called by the caller */
633 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
634 	ptep_flush_direct(mm, saddr, ptep, nodat);
635 	/* don't touch the storage key - it belongs to parent pgste */
636 	pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
637 	pgste_set_unlock(ptep, pgste);
638 }
639 
640 static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
641 {
642 	if (!non_swap_entry(entry))
643 		dec_mm_counter(mm, MM_SWAPENTS);
644 	else if (is_migration_entry(entry)) {
645 		struct page *page = migration_entry_to_page(entry);
646 
647 		dec_mm_counter(mm, mm_counter(page));
648 	}
649 	free_swap_and_cache(entry);
650 }
651 
652 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
653 		     pte_t *ptep, int reset)
654 {
655 	unsigned long pgstev;
656 	pgste_t pgste;
657 	pte_t pte;
658 
659 	/* Zap unused and logically-zero pages */
660 	preempt_disable();
661 	pgste = pgste_get_lock(ptep);
662 	pgstev = pgste_val(pgste);
663 	pte = *ptep;
664 	if (!reset && pte_swap(pte) &&
665 	    ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
666 	     (pgstev & _PGSTE_GPS_ZERO))) {
667 		ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
668 		pte_clear(mm, addr, ptep);
669 	}
670 	if (reset)
671 		pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
672 	pgste_set_unlock(ptep, pgste);
673 	preempt_enable();
674 }
675 
676 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
677 {
678 	unsigned long ptev;
679 	pgste_t pgste;
680 
681 	/* Clear storage key ACC and F, but set R/C */
682 	preempt_disable();
683 	pgste = pgste_get_lock(ptep);
684 	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
685 	pgste_val(pgste) |= PGSTE_GR_BIT | PGSTE_GC_BIT;
686 	ptev = pte_val(*ptep);
687 	if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
688 		page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 1);
689 	pgste_set_unlock(ptep, pgste);
690 	preempt_enable();
691 }
692 
693 /*
694  * Test and reset if a guest page is dirty
695  */
696 bool test_and_clear_guest_dirty(struct mm_struct *mm, unsigned long addr)
697 {
698 	spinlock_t *ptl;
699 	pgd_t *pgd;
700 	p4d_t *p4d;
701 	pud_t *pud;
702 	pmd_t *pmd;
703 	pgste_t pgste;
704 	pte_t *ptep;
705 	pte_t pte;
706 	bool dirty;
707 	int nodat;
708 
709 	pgd = pgd_offset(mm, addr);
710 	p4d = p4d_alloc(mm, pgd, addr);
711 	if (!p4d)
712 		return false;
713 	pud = pud_alloc(mm, p4d, addr);
714 	if (!pud)
715 		return false;
716 	pmd = pmd_alloc(mm, pud, addr);
717 	if (!pmd)
718 		return false;
719 	/* We can't run guests backed by huge pages, but userspace can
720 	 * still set them up and then try to migrate them without any
721 	 * migration support.
722 	 */
723 	if (pmd_large(*pmd))
724 		return true;
725 
726 	ptep = pte_alloc_map_lock(mm, pmd, addr, &ptl);
727 	if (unlikely(!ptep))
728 		return false;
729 
730 	pgste = pgste_get_lock(ptep);
731 	dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
732 	pgste_val(pgste) &= ~PGSTE_UC_BIT;
733 	pte = *ptep;
734 	if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
735 		pgste = pgste_pte_notify(mm, addr, ptep, pgste);
736 		nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
737 		ptep_ipte_global(mm, addr, ptep, nodat);
738 		if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
739 			pte_val(pte) |= _PAGE_PROTECT;
740 		else
741 			pte_val(pte) |= _PAGE_INVALID;
742 		*ptep = pte;
743 	}
744 	pgste_set_unlock(ptep, pgste);
745 
746 	spin_unlock(ptl);
747 	return dirty;
748 }
749 EXPORT_SYMBOL_GPL(test_and_clear_guest_dirty);
750 
751 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
752 			  unsigned char key, bool nq)
753 {
754 	unsigned long keyul;
755 	spinlock_t *ptl;
756 	pgste_t old, new;
757 	pte_t *ptep;
758 
759 	ptep = get_locked_pte(mm, addr, &ptl);
760 	if (unlikely(!ptep))
761 		return -EFAULT;
762 
763 	new = old = pgste_get_lock(ptep);
764 	pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
765 			    PGSTE_ACC_BITS | PGSTE_FP_BIT);
766 	keyul = (unsigned long) key;
767 	pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
768 	pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
769 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
770 		unsigned long address, bits, skey;
771 
772 		address = pte_val(*ptep) & PAGE_MASK;
773 		skey = (unsigned long) page_get_storage_key(address);
774 		bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
775 		skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
776 		/* Set storage key ACC and FP */
777 		page_set_storage_key(address, skey, !nq);
778 		/* Merge host changed & referenced into pgste  */
779 		pgste_val(new) |= bits << 52;
780 	}
781 	/* changing the guest storage key is considered a change of the page */
782 	if ((pgste_val(new) ^ pgste_val(old)) &
783 	    (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
784 		pgste_val(new) |= PGSTE_UC_BIT;
785 
786 	pgste_set_unlock(ptep, new);
787 	pte_unmap_unlock(ptep, ptl);
788 	return 0;
789 }
790 EXPORT_SYMBOL(set_guest_storage_key);
791 
792 /**
793  * Conditionally set a guest storage key (handling csske).
794  * oldkey will be updated when either mr or mc is set and a pointer is given.
795  *
796  * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
797  * storage key was updated and -EFAULT on access errors.
798  */
799 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
800 			       unsigned char key, unsigned char *oldkey,
801 			       bool nq, bool mr, bool mc)
802 {
803 	unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
804 	int rc;
805 
806 	/* we can drop the pgste lock between getting and setting the key */
807 	if (mr | mc) {
808 		rc = get_guest_storage_key(current->mm, addr, &tmp);
809 		if (rc)
810 			return rc;
811 		if (oldkey)
812 			*oldkey = tmp;
813 		if (!mr)
814 			mask |= _PAGE_REFERENCED;
815 		if (!mc)
816 			mask |= _PAGE_CHANGED;
817 		if (!((tmp ^ key) & mask))
818 			return 0;
819 	}
820 	rc = set_guest_storage_key(current->mm, addr, key, nq);
821 	return rc < 0 ? rc : 1;
822 }
823 EXPORT_SYMBOL(cond_set_guest_storage_key);
824 
825 /**
826  * Reset a guest reference bit (rrbe), returning the reference and changed bit.
827  *
828  * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
829  */
830 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
831 {
832 	spinlock_t *ptl;
833 	pgste_t old, new;
834 	pte_t *ptep;
835 	int cc = 0;
836 
837 	ptep = get_locked_pte(mm, addr, &ptl);
838 	if (unlikely(!ptep))
839 		return -EFAULT;
840 
841 	new = old = pgste_get_lock(ptep);
842 	/* Reset guest reference bit only */
843 	pgste_val(new) &= ~PGSTE_GR_BIT;
844 
845 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
846 		cc = page_reset_referenced(pte_val(*ptep) & PAGE_MASK);
847 		/* Merge real referenced bit into host-set */
848 		pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
849 	}
850 	/* Reflect guest's logical view, not physical */
851 	cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
852 	/* Changing the guest storage key is considered a change of the page */
853 	if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
854 		pgste_val(new) |= PGSTE_UC_BIT;
855 
856 	pgste_set_unlock(ptep, new);
857 	pte_unmap_unlock(ptep, ptl);
858 	return cc;
859 }
860 EXPORT_SYMBOL(reset_guest_reference_bit);
861 
862 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
863 			  unsigned char *key)
864 {
865 	spinlock_t *ptl;
866 	pgste_t pgste;
867 	pte_t *ptep;
868 
869 	ptep = get_locked_pte(mm, addr, &ptl);
870 	if (unlikely(!ptep))
871 		return -EFAULT;
872 
873 	pgste = pgste_get_lock(ptep);
874 	*key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
875 	if (!(pte_val(*ptep) & _PAGE_INVALID))
876 		*key = page_get_storage_key(pte_val(*ptep) & PAGE_MASK);
877 	/* Reflect guest's logical view, not physical */
878 	*key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
879 	pgste_set_unlock(ptep, pgste);
880 	pte_unmap_unlock(ptep, ptl);
881 	return 0;
882 }
883 EXPORT_SYMBOL(get_guest_storage_key);
884 
885 /**
886  * pgste_perform_essa - perform ESSA actions on the PGSTE.
887  * @mm: the memory context. It must have PGSTEs, no check is performed here!
888  * @hva: the host virtual address of the page whose PGSTE is to be processed
889  * @orc: the specific action to perform, see the ESSA_SET_* macros.
890  * @oldpte: the PTE will be saved there if the pointer is not NULL.
891  * @oldpgste: the old PGSTE will be saved there if the pointer is not NULL.
892  *
893  * Return: 1 if the page is to be added to the CBRL, otherwise 0,
894  *	   or < 0 in case of error. -EINVAL is returned for invalid values
895  *	   of orc, -EFAULT for invalid addresses.
896  */
897 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
898 			unsigned long *oldpte, unsigned long *oldpgste)
899 {
900 	unsigned long pgstev;
901 	spinlock_t *ptl;
902 	pgste_t pgste;
903 	pte_t *ptep;
904 	int res = 0;
905 
906 	WARN_ON_ONCE(orc > ESSA_MAX);
907 	if (unlikely(orc > ESSA_MAX))
908 		return -EINVAL;
909 	ptep = get_locked_pte(mm, hva, &ptl);
910 	if (unlikely(!ptep))
911 		return -EFAULT;
912 	pgste = pgste_get_lock(ptep);
913 	pgstev = pgste_val(pgste);
914 	if (oldpte)
915 		*oldpte = pte_val(*ptep);
916 	if (oldpgste)
917 		*oldpgste = pgstev;
918 
919 	switch (orc) {
920 	case ESSA_GET_STATE:
921 		break;
922 	case ESSA_SET_STABLE:
923 		pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
924 		pgstev |= _PGSTE_GPS_USAGE_STABLE;
925 		break;
926 	case ESSA_SET_UNUSED:
927 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
928 		pgstev |= _PGSTE_GPS_USAGE_UNUSED;
929 		if (pte_val(*ptep) & _PAGE_INVALID)
930 			res = 1;
931 		break;
932 	case ESSA_SET_VOLATILE:
933 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
934 		pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
935 		if (pte_val(*ptep) & _PAGE_INVALID)
936 			res = 1;
937 		break;
938 	case ESSA_SET_POT_VOLATILE:
939 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
940 		if (!(pte_val(*ptep) & _PAGE_INVALID)) {
941 			pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
942 			break;
943 		}
944 		if (pgstev & _PGSTE_GPS_ZERO) {
945 			pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
946 			break;
947 		}
948 		if (!(pgstev & PGSTE_GC_BIT)) {
949 			pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
950 			res = 1;
951 			break;
952 		}
953 		break;
954 	case ESSA_SET_STABLE_RESIDENT:
955 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
956 		pgstev |= _PGSTE_GPS_USAGE_STABLE;
957 		/*
958 		 * Since the resident state can go away any time after this
959 		 * call, we will not make this page resident. We can revisit
960 		 * this decision if a guest will ever start using this.
961 		 */
962 		break;
963 	case ESSA_SET_STABLE_IF_RESIDENT:
964 		if (!(pte_val(*ptep) & _PAGE_INVALID)) {
965 			pgstev &= ~_PGSTE_GPS_USAGE_MASK;
966 			pgstev |= _PGSTE_GPS_USAGE_STABLE;
967 		}
968 		break;
969 	case ESSA_SET_STABLE_NODAT:
970 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
971 		pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
972 		break;
973 	default:
974 		/* we should never get here! */
975 		break;
976 	}
977 	/* If we are discarding a page, set it to logical zero */
978 	if (res)
979 		pgstev |= _PGSTE_GPS_ZERO;
980 
981 	pgste_val(pgste) = pgstev;
982 	pgste_set_unlock(ptep, pgste);
983 	pte_unmap_unlock(ptep, ptl);
984 	return res;
985 }
986 EXPORT_SYMBOL(pgste_perform_essa);
987 
988 /**
989  * set_pgste_bits - set specific PGSTE bits.
990  * @mm: the memory context. It must have PGSTEs, no check is performed here!
991  * @hva: the host virtual address of the page whose PGSTE is to be processed
992  * @bits: a bitmask representing the bits that will be touched
993  * @value: the values of the bits to be written. Only the bits in the mask
994  *	   will be written.
995  *
996  * Return: 0 on success, < 0 in case of error.
997  */
998 int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
999 			unsigned long bits, unsigned long value)
1000 {
1001 	spinlock_t *ptl;
1002 	pgste_t new;
1003 	pte_t *ptep;
1004 
1005 	ptep = get_locked_pte(mm, hva, &ptl);
1006 	if (unlikely(!ptep))
1007 		return -EFAULT;
1008 	new = pgste_get_lock(ptep);
1009 
1010 	pgste_val(new) &= ~bits;
1011 	pgste_val(new) |= value & bits;
1012 
1013 	pgste_set_unlock(ptep, new);
1014 	pte_unmap_unlock(ptep, ptl);
1015 	return 0;
1016 }
1017 EXPORT_SYMBOL(set_pgste_bits);
1018 
1019 /**
1020  * get_pgste - get the current PGSTE for the given address.
1021  * @mm: the memory context. It must have PGSTEs, no check is performed here!
1022  * @hva: the host virtual address of the page whose PGSTE is to be processed
1023  * @pgstep: will be written with the current PGSTE for the given address.
1024  *
1025  * Return: 0 on success, < 0 in case of error.
1026  */
1027 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1028 {
1029 	spinlock_t *ptl;
1030 	pte_t *ptep;
1031 
1032 	ptep = get_locked_pte(mm, hva, &ptl);
1033 	if (unlikely(!ptep))
1034 		return -EFAULT;
1035 	*pgstep = pgste_val(pgste_get(ptep));
1036 	pte_unmap_unlock(ptep, ptl);
1037 	return 0;
1038 }
1039 EXPORT_SYMBOL(get_pgste);
1040 #endif
1041