xref: /openbmc/linux/arch/s390/mm/pgtable.c (revision d003d772)
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 vm_area_struct *vma, unsigned long addr,
305 			     pte_t *ptep)
306 {
307 	pgste_t pgste;
308 	pte_t old;
309 	int nodat;
310 	struct mm_struct *mm = vma->vm_mm;
311 
312 	preempt_disable();
313 	pgste = ptep_xchg_start(mm, addr, ptep);
314 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
315 	old = ptep_flush_lazy(mm, addr, ptep, nodat);
316 	if (mm_has_pgste(mm)) {
317 		pgste = pgste_update_all(old, pgste, mm);
318 		pgste_set(ptep, pgste);
319 	}
320 	return old;
321 }
322 
323 void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
324 			     pte_t *ptep, pte_t old_pte, pte_t pte)
325 {
326 	pgste_t pgste;
327 	struct mm_struct *mm = vma->vm_mm;
328 
329 	if (!MACHINE_HAS_NX)
330 		pte_val(pte) &= ~_PAGE_NOEXEC;
331 	if (mm_has_pgste(mm)) {
332 		pgste = pgste_get(ptep);
333 		pgste_set_key(ptep, pgste, pte, mm);
334 		pgste = pgste_set_pte(ptep, pgste, pte);
335 		pgste_set_unlock(ptep, pgste);
336 	} else {
337 		*ptep = pte;
338 	}
339 	preempt_enable();
340 }
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 	if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
351 		gmap_pmdp_idte_local(mm, addr);
352 }
353 
354 static inline void pmdp_idte_global(struct mm_struct *mm,
355 				    unsigned long addr, pmd_t *pmdp)
356 {
357 	if (MACHINE_HAS_TLB_GUEST) {
358 		__pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
359 			    mm->context.asce, IDTE_GLOBAL);
360 		if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
361 			gmap_pmdp_idte_global(mm, addr);
362 	} else if (MACHINE_HAS_IDTE) {
363 		__pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
364 		if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
365 			gmap_pmdp_idte_global(mm, addr);
366 	} else {
367 		__pmdp_csp(pmdp);
368 		if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
369 			gmap_pmdp_csp(mm, addr);
370 	}
371 }
372 
373 static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
374 				      unsigned long addr, pmd_t *pmdp)
375 {
376 	pmd_t old;
377 
378 	old = *pmdp;
379 	if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
380 		return old;
381 	atomic_inc(&mm->context.flush_count);
382 	if (MACHINE_HAS_TLB_LC &&
383 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
384 		pmdp_idte_local(mm, addr, pmdp);
385 	else
386 		pmdp_idte_global(mm, addr, pmdp);
387 	atomic_dec(&mm->context.flush_count);
388 	return old;
389 }
390 
391 static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
392 				    unsigned long addr, pmd_t *pmdp)
393 {
394 	pmd_t old;
395 
396 	old = *pmdp;
397 	if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
398 		return old;
399 	atomic_inc(&mm->context.flush_count);
400 	if (cpumask_equal(&mm->context.cpu_attach_mask,
401 			  cpumask_of(smp_processor_id()))) {
402 		pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
403 		mm->context.flush_mm = 1;
404 		if (mm_has_pgste(mm))
405 			gmap_pmdp_invalidate(mm, addr);
406 	} else {
407 		pmdp_idte_global(mm, addr, pmdp);
408 	}
409 	atomic_dec(&mm->context.flush_count);
410 	return old;
411 }
412 
413 static pmd_t *pmd_alloc_map(struct mm_struct *mm, unsigned long addr)
414 {
415 	pgd_t *pgd;
416 	p4d_t *p4d;
417 	pud_t *pud;
418 	pmd_t *pmd;
419 
420 	pgd = pgd_offset(mm, addr);
421 	p4d = p4d_alloc(mm, pgd, addr);
422 	if (!p4d)
423 		return NULL;
424 	pud = pud_alloc(mm, p4d, addr);
425 	if (!pud)
426 		return NULL;
427 	pmd = pmd_alloc(mm, pud, addr);
428 	return pmd;
429 }
430 
431 pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
432 		       pmd_t *pmdp, pmd_t new)
433 {
434 	pmd_t old;
435 
436 	preempt_disable();
437 	old = pmdp_flush_direct(mm, addr, pmdp);
438 	*pmdp = new;
439 	preempt_enable();
440 	return old;
441 }
442 EXPORT_SYMBOL(pmdp_xchg_direct);
443 
444 pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
445 		     pmd_t *pmdp, pmd_t new)
446 {
447 	pmd_t old;
448 
449 	preempt_disable();
450 	old = pmdp_flush_lazy(mm, addr, pmdp);
451 	*pmdp = new;
452 	preempt_enable();
453 	return old;
454 }
455 EXPORT_SYMBOL(pmdp_xchg_lazy);
456 
457 static inline void pudp_idte_local(struct mm_struct *mm,
458 				   unsigned long addr, pud_t *pudp)
459 {
460 	if (MACHINE_HAS_TLB_GUEST)
461 		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
462 			    mm->context.asce, IDTE_LOCAL);
463 	else
464 		__pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
465 }
466 
467 static inline void pudp_idte_global(struct mm_struct *mm,
468 				    unsigned long addr, pud_t *pudp)
469 {
470 	if (MACHINE_HAS_TLB_GUEST)
471 		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
472 			    mm->context.asce, IDTE_GLOBAL);
473 	else if (MACHINE_HAS_IDTE)
474 		__pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
475 	else
476 		/*
477 		 * Invalid bit position is the same for pmd and pud, so we can
478 		 * re-use _pmd_csp() here
479 		 */
480 		__pmdp_csp((pmd_t *) pudp);
481 }
482 
483 static inline pud_t pudp_flush_direct(struct mm_struct *mm,
484 				      unsigned long addr, pud_t *pudp)
485 {
486 	pud_t old;
487 
488 	old = *pudp;
489 	if (pud_val(old) & _REGION_ENTRY_INVALID)
490 		return old;
491 	atomic_inc(&mm->context.flush_count);
492 	if (MACHINE_HAS_TLB_LC &&
493 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
494 		pudp_idte_local(mm, addr, pudp);
495 	else
496 		pudp_idte_global(mm, addr, pudp);
497 	atomic_dec(&mm->context.flush_count);
498 	return old;
499 }
500 
501 pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
502 		       pud_t *pudp, pud_t new)
503 {
504 	pud_t old;
505 
506 	preempt_disable();
507 	old = pudp_flush_direct(mm, addr, pudp);
508 	*pudp = new;
509 	preempt_enable();
510 	return old;
511 }
512 EXPORT_SYMBOL(pudp_xchg_direct);
513 
514 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
515 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
516 				pgtable_t pgtable)
517 {
518 	struct list_head *lh = (struct list_head *) pgtable;
519 
520 	assert_spin_locked(pmd_lockptr(mm, pmdp));
521 
522 	/* FIFO */
523 	if (!pmd_huge_pte(mm, pmdp))
524 		INIT_LIST_HEAD(lh);
525 	else
526 		list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
527 	pmd_huge_pte(mm, pmdp) = pgtable;
528 }
529 
530 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
531 {
532 	struct list_head *lh;
533 	pgtable_t pgtable;
534 	pte_t *ptep;
535 
536 	assert_spin_locked(pmd_lockptr(mm, pmdp));
537 
538 	/* FIFO */
539 	pgtable = pmd_huge_pte(mm, pmdp);
540 	lh = (struct list_head *) pgtable;
541 	if (list_empty(lh))
542 		pmd_huge_pte(mm, pmdp) = NULL;
543 	else {
544 		pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
545 		list_del(lh);
546 	}
547 	ptep = (pte_t *) pgtable;
548 	pte_val(*ptep) = _PAGE_INVALID;
549 	ptep++;
550 	pte_val(*ptep) = _PAGE_INVALID;
551 	return pgtable;
552 }
553 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
554 
555 #ifdef CONFIG_PGSTE
556 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
557 		     pte_t *ptep, pte_t entry)
558 {
559 	pgste_t pgste;
560 
561 	/* the mm_has_pgste() check is done in set_pte_at() */
562 	preempt_disable();
563 	pgste = pgste_get_lock(ptep);
564 	pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
565 	pgste_set_key(ptep, pgste, entry, mm);
566 	pgste = pgste_set_pte(ptep, pgste, entry);
567 	pgste_set_unlock(ptep, pgste);
568 	preempt_enable();
569 }
570 
571 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
572 {
573 	pgste_t pgste;
574 
575 	preempt_disable();
576 	pgste = pgste_get_lock(ptep);
577 	pgste_val(pgste) |= PGSTE_IN_BIT;
578 	pgste_set_unlock(ptep, pgste);
579 	preempt_enable();
580 }
581 
582 /**
583  * ptep_force_prot - change access rights of a locked pte
584  * @mm: pointer to the process mm_struct
585  * @addr: virtual address in the guest address space
586  * @ptep: pointer to the page table entry
587  * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
588  * @bit: pgste bit to set (e.g. for notification)
589  *
590  * Returns 0 if the access rights were changed and -EAGAIN if the current
591  * and requested access rights are incompatible.
592  */
593 int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
594 		    pte_t *ptep, int prot, unsigned long bit)
595 {
596 	pte_t entry;
597 	pgste_t pgste;
598 	int pte_i, pte_p, nodat;
599 
600 	pgste = pgste_get_lock(ptep);
601 	entry = *ptep;
602 	/* Check pte entry after all locks have been acquired */
603 	pte_i = pte_val(entry) & _PAGE_INVALID;
604 	pte_p = pte_val(entry) & _PAGE_PROTECT;
605 	if ((pte_i && (prot != PROT_NONE)) ||
606 	    (pte_p && (prot & PROT_WRITE))) {
607 		pgste_set_unlock(ptep, pgste);
608 		return -EAGAIN;
609 	}
610 	/* Change access rights and set pgste bit */
611 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
612 	if (prot == PROT_NONE && !pte_i) {
613 		ptep_flush_direct(mm, addr, ptep, nodat);
614 		pgste = pgste_update_all(entry, pgste, mm);
615 		pte_val(entry) |= _PAGE_INVALID;
616 	}
617 	if (prot == PROT_READ && !pte_p) {
618 		ptep_flush_direct(mm, addr, ptep, nodat);
619 		pte_val(entry) &= ~_PAGE_INVALID;
620 		pte_val(entry) |= _PAGE_PROTECT;
621 	}
622 	pgste_val(pgste) |= bit;
623 	pgste = pgste_set_pte(ptep, pgste, entry);
624 	pgste_set_unlock(ptep, pgste);
625 	return 0;
626 }
627 
628 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
629 		    pte_t *sptep, pte_t *tptep, pte_t pte)
630 {
631 	pgste_t spgste, tpgste;
632 	pte_t spte, tpte;
633 	int rc = -EAGAIN;
634 
635 	if (!(pte_val(*tptep) & _PAGE_INVALID))
636 		return 0;	/* already shadowed */
637 	spgste = pgste_get_lock(sptep);
638 	spte = *sptep;
639 	if (!(pte_val(spte) & _PAGE_INVALID) &&
640 	    !((pte_val(spte) & _PAGE_PROTECT) &&
641 	      !(pte_val(pte) & _PAGE_PROTECT))) {
642 		pgste_val(spgste) |= PGSTE_VSIE_BIT;
643 		tpgste = pgste_get_lock(tptep);
644 		pte_val(tpte) = (pte_val(spte) & PAGE_MASK) |
645 				(pte_val(pte) & _PAGE_PROTECT);
646 		/* don't touch the storage key - it belongs to parent pgste */
647 		tpgste = pgste_set_pte(tptep, tpgste, tpte);
648 		pgste_set_unlock(tptep, tpgste);
649 		rc = 1;
650 	}
651 	pgste_set_unlock(sptep, spgste);
652 	return rc;
653 }
654 
655 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
656 {
657 	pgste_t pgste;
658 	int nodat;
659 
660 	pgste = pgste_get_lock(ptep);
661 	/* notifier is called by the caller */
662 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
663 	ptep_flush_direct(mm, saddr, ptep, nodat);
664 	/* don't touch the storage key - it belongs to parent pgste */
665 	pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
666 	pgste_set_unlock(ptep, pgste);
667 }
668 
669 static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
670 {
671 	if (!non_swap_entry(entry))
672 		dec_mm_counter(mm, MM_SWAPENTS);
673 	else if (is_migration_entry(entry)) {
674 		struct page *page = migration_entry_to_page(entry);
675 
676 		dec_mm_counter(mm, mm_counter(page));
677 	}
678 	free_swap_and_cache(entry);
679 }
680 
681 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
682 		     pte_t *ptep, int reset)
683 {
684 	unsigned long pgstev;
685 	pgste_t pgste;
686 	pte_t pte;
687 
688 	/* Zap unused and logically-zero pages */
689 	preempt_disable();
690 	pgste = pgste_get_lock(ptep);
691 	pgstev = pgste_val(pgste);
692 	pte = *ptep;
693 	if (!reset && pte_swap(pte) &&
694 	    ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
695 	     (pgstev & _PGSTE_GPS_ZERO))) {
696 		ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
697 		pte_clear(mm, addr, ptep);
698 	}
699 	if (reset)
700 		pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
701 	pgste_set_unlock(ptep, pgste);
702 	preempt_enable();
703 }
704 
705 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
706 {
707 	unsigned long ptev;
708 	pgste_t pgste;
709 
710 	/* Clear storage key ACC and F, but set R/C */
711 	preempt_disable();
712 	pgste = pgste_get_lock(ptep);
713 	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
714 	pgste_val(pgste) |= PGSTE_GR_BIT | PGSTE_GC_BIT;
715 	ptev = pte_val(*ptep);
716 	if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
717 		page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 1);
718 	pgste_set_unlock(ptep, pgste);
719 	preempt_enable();
720 }
721 
722 /*
723  * Test and reset if a guest page is dirty
724  */
725 bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long addr,
726 		       pte_t *ptep)
727 {
728 	pgste_t pgste;
729 	pte_t pte;
730 	bool dirty;
731 	int nodat;
732 
733 	pgste = pgste_get_lock(ptep);
734 	dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
735 	pgste_val(pgste) &= ~PGSTE_UC_BIT;
736 	pte = *ptep;
737 	if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
738 		pgste = pgste_pte_notify(mm, addr, ptep, pgste);
739 		nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
740 		ptep_ipte_global(mm, addr, ptep, nodat);
741 		if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
742 			pte_val(pte) |= _PAGE_PROTECT;
743 		else
744 			pte_val(pte) |= _PAGE_INVALID;
745 		*ptep = pte;
746 	}
747 	pgste_set_unlock(ptep, pgste);
748 	return dirty;
749 }
750 EXPORT_SYMBOL_GPL(ptep_test_and_clear_uc);
751 
752 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
753 			  unsigned char key, bool nq)
754 {
755 	unsigned long keyul, paddr;
756 	spinlock_t *ptl;
757 	pgste_t old, new;
758 	pmd_t *pmdp;
759 	pte_t *ptep;
760 
761 	pmdp = pmd_alloc_map(mm, addr);
762 	if (unlikely(!pmdp))
763 		return -EFAULT;
764 
765 	ptl = pmd_lock(mm, pmdp);
766 	if (!pmd_present(*pmdp)) {
767 		spin_unlock(ptl);
768 		return -EFAULT;
769 	}
770 
771 	if (pmd_large(*pmdp)) {
772 		paddr = pmd_val(*pmdp) & HPAGE_MASK;
773 		paddr |= addr & ~HPAGE_MASK;
774 		/*
775 		 * Huge pmds need quiescing operations, they are
776 		 * always mapped.
777 		 */
778 		page_set_storage_key(paddr, key, 1);
779 		spin_unlock(ptl);
780 		return 0;
781 	}
782 	spin_unlock(ptl);
783 
784 	ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
785 	if (unlikely(!ptep))
786 		return -EFAULT;
787 
788 	new = old = pgste_get_lock(ptep);
789 	pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
790 			    PGSTE_ACC_BITS | PGSTE_FP_BIT);
791 	keyul = (unsigned long) key;
792 	pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
793 	pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
794 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
795 		unsigned long bits, skey;
796 
797 		paddr = pte_val(*ptep) & PAGE_MASK;
798 		skey = (unsigned long) page_get_storage_key(paddr);
799 		bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
800 		skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
801 		/* Set storage key ACC and FP */
802 		page_set_storage_key(paddr, skey, !nq);
803 		/* Merge host changed & referenced into pgste  */
804 		pgste_val(new) |= bits << 52;
805 	}
806 	/* changing the guest storage key is considered a change of the page */
807 	if ((pgste_val(new) ^ pgste_val(old)) &
808 	    (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
809 		pgste_val(new) |= PGSTE_UC_BIT;
810 
811 	pgste_set_unlock(ptep, new);
812 	pte_unmap_unlock(ptep, ptl);
813 	return 0;
814 }
815 EXPORT_SYMBOL(set_guest_storage_key);
816 
817 /**
818  * Conditionally set a guest storage key (handling csske).
819  * oldkey will be updated when either mr or mc is set and a pointer is given.
820  *
821  * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
822  * storage key was updated and -EFAULT on access errors.
823  */
824 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
825 			       unsigned char key, unsigned char *oldkey,
826 			       bool nq, bool mr, bool mc)
827 {
828 	unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
829 	int rc;
830 
831 	/* we can drop the pgste lock between getting and setting the key */
832 	if (mr | mc) {
833 		rc = get_guest_storage_key(current->mm, addr, &tmp);
834 		if (rc)
835 			return rc;
836 		if (oldkey)
837 			*oldkey = tmp;
838 		if (!mr)
839 			mask |= _PAGE_REFERENCED;
840 		if (!mc)
841 			mask |= _PAGE_CHANGED;
842 		if (!((tmp ^ key) & mask))
843 			return 0;
844 	}
845 	rc = set_guest_storage_key(current->mm, addr, key, nq);
846 	return rc < 0 ? rc : 1;
847 }
848 EXPORT_SYMBOL(cond_set_guest_storage_key);
849 
850 /**
851  * Reset a guest reference bit (rrbe), returning the reference and changed bit.
852  *
853  * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
854  */
855 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
856 {
857 	spinlock_t *ptl;
858 	unsigned long paddr;
859 	pgste_t old, new;
860 	pmd_t *pmdp;
861 	pte_t *ptep;
862 	int cc = 0;
863 
864 	pmdp = pmd_alloc_map(mm, addr);
865 	if (unlikely(!pmdp))
866 		return -EFAULT;
867 
868 	ptl = pmd_lock(mm, pmdp);
869 	if (!pmd_present(*pmdp)) {
870 		spin_unlock(ptl);
871 		return -EFAULT;
872 	}
873 
874 	if (pmd_large(*pmdp)) {
875 		paddr = pmd_val(*pmdp) & HPAGE_MASK;
876 		paddr |= addr & ~HPAGE_MASK;
877 		cc = page_reset_referenced(paddr);
878 		spin_unlock(ptl);
879 		return cc;
880 	}
881 	spin_unlock(ptl);
882 
883 	ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
884 	if (unlikely(!ptep))
885 		return -EFAULT;
886 
887 	new = old = pgste_get_lock(ptep);
888 	/* Reset guest reference bit only */
889 	pgste_val(new) &= ~PGSTE_GR_BIT;
890 
891 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
892 		paddr = pte_val(*ptep) & PAGE_MASK;
893 		cc = page_reset_referenced(paddr);
894 		/* Merge real referenced bit into host-set */
895 		pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
896 	}
897 	/* Reflect guest's logical view, not physical */
898 	cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
899 	/* Changing the guest storage key is considered a change of the page */
900 	if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
901 		pgste_val(new) |= PGSTE_UC_BIT;
902 
903 	pgste_set_unlock(ptep, new);
904 	pte_unmap_unlock(ptep, ptl);
905 	return cc;
906 }
907 EXPORT_SYMBOL(reset_guest_reference_bit);
908 
909 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
910 			  unsigned char *key)
911 {
912 	unsigned long paddr;
913 	spinlock_t *ptl;
914 	pgste_t pgste;
915 	pmd_t *pmdp;
916 	pte_t *ptep;
917 
918 	pmdp = pmd_alloc_map(mm, addr);
919 	if (unlikely(!pmdp))
920 		return -EFAULT;
921 
922 	ptl = pmd_lock(mm, pmdp);
923 	if (!pmd_present(*pmdp)) {
924 		/* Not yet mapped memory has a zero key */
925 		spin_unlock(ptl);
926 		*key = 0;
927 		return 0;
928 	}
929 
930 	if (pmd_large(*pmdp)) {
931 		paddr = pmd_val(*pmdp) & HPAGE_MASK;
932 		paddr |= addr & ~HPAGE_MASK;
933 		*key = page_get_storage_key(paddr);
934 		spin_unlock(ptl);
935 		return 0;
936 	}
937 	spin_unlock(ptl);
938 
939 	ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
940 	if (unlikely(!ptep))
941 		return -EFAULT;
942 
943 	pgste = pgste_get_lock(ptep);
944 	*key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
945 	paddr = pte_val(*ptep) & PAGE_MASK;
946 	if (!(pte_val(*ptep) & _PAGE_INVALID))
947 		*key = page_get_storage_key(paddr);
948 	/* Reflect guest's logical view, not physical */
949 	*key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
950 	pgste_set_unlock(ptep, pgste);
951 	pte_unmap_unlock(ptep, ptl);
952 	return 0;
953 }
954 EXPORT_SYMBOL(get_guest_storage_key);
955 
956 /**
957  * pgste_perform_essa - perform ESSA actions on the PGSTE.
958  * @mm: the memory context. It must have PGSTEs, no check is performed here!
959  * @hva: the host virtual address of the page whose PGSTE is to be processed
960  * @orc: the specific action to perform, see the ESSA_SET_* macros.
961  * @oldpte: the PTE will be saved there if the pointer is not NULL.
962  * @oldpgste: the old PGSTE will be saved there if the pointer is not NULL.
963  *
964  * Return: 1 if the page is to be added to the CBRL, otherwise 0,
965  *	   or < 0 in case of error. -EINVAL is returned for invalid values
966  *	   of orc, -EFAULT for invalid addresses.
967  */
968 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
969 			unsigned long *oldpte, unsigned long *oldpgste)
970 {
971 	unsigned long pgstev;
972 	spinlock_t *ptl;
973 	pgste_t pgste;
974 	pte_t *ptep;
975 	int res = 0;
976 
977 	WARN_ON_ONCE(orc > ESSA_MAX);
978 	if (unlikely(orc > ESSA_MAX))
979 		return -EINVAL;
980 	ptep = get_locked_pte(mm, hva, &ptl);
981 	if (unlikely(!ptep))
982 		return -EFAULT;
983 	pgste = pgste_get_lock(ptep);
984 	pgstev = pgste_val(pgste);
985 	if (oldpte)
986 		*oldpte = pte_val(*ptep);
987 	if (oldpgste)
988 		*oldpgste = pgstev;
989 
990 	switch (orc) {
991 	case ESSA_GET_STATE:
992 		break;
993 	case ESSA_SET_STABLE:
994 		pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
995 		pgstev |= _PGSTE_GPS_USAGE_STABLE;
996 		break;
997 	case ESSA_SET_UNUSED:
998 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
999 		pgstev |= _PGSTE_GPS_USAGE_UNUSED;
1000 		if (pte_val(*ptep) & _PAGE_INVALID)
1001 			res = 1;
1002 		break;
1003 	case ESSA_SET_VOLATILE:
1004 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1005 		pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1006 		if (pte_val(*ptep) & _PAGE_INVALID)
1007 			res = 1;
1008 		break;
1009 	case ESSA_SET_POT_VOLATILE:
1010 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1011 		if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1012 			pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
1013 			break;
1014 		}
1015 		if (pgstev & _PGSTE_GPS_ZERO) {
1016 			pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1017 			break;
1018 		}
1019 		if (!(pgstev & PGSTE_GC_BIT)) {
1020 			pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1021 			res = 1;
1022 			break;
1023 		}
1024 		break;
1025 	case ESSA_SET_STABLE_RESIDENT:
1026 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1027 		pgstev |= _PGSTE_GPS_USAGE_STABLE;
1028 		/*
1029 		 * Since the resident state can go away any time after this
1030 		 * call, we will not make this page resident. We can revisit
1031 		 * this decision if a guest will ever start using this.
1032 		 */
1033 		break;
1034 	case ESSA_SET_STABLE_IF_RESIDENT:
1035 		if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1036 			pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1037 			pgstev |= _PGSTE_GPS_USAGE_STABLE;
1038 		}
1039 		break;
1040 	case ESSA_SET_STABLE_NODAT:
1041 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1042 		pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
1043 		break;
1044 	default:
1045 		/* we should never get here! */
1046 		break;
1047 	}
1048 	/* If we are discarding a page, set it to logical zero */
1049 	if (res)
1050 		pgstev |= _PGSTE_GPS_ZERO;
1051 
1052 	pgste_val(pgste) = pgstev;
1053 	pgste_set_unlock(ptep, pgste);
1054 	pte_unmap_unlock(ptep, ptl);
1055 	return res;
1056 }
1057 EXPORT_SYMBOL(pgste_perform_essa);
1058 
1059 /**
1060  * set_pgste_bits - set specific PGSTE bits.
1061  * @mm: the memory context. It must have PGSTEs, no check is performed here!
1062  * @hva: the host virtual address of the page whose PGSTE is to be processed
1063  * @bits: a bitmask representing the bits that will be touched
1064  * @value: the values of the bits to be written. Only the bits in the mask
1065  *	   will be written.
1066  *
1067  * Return: 0 on success, < 0 in case of error.
1068  */
1069 int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
1070 			unsigned long bits, unsigned long value)
1071 {
1072 	spinlock_t *ptl;
1073 	pgste_t new;
1074 	pte_t *ptep;
1075 
1076 	ptep = get_locked_pte(mm, hva, &ptl);
1077 	if (unlikely(!ptep))
1078 		return -EFAULT;
1079 	new = pgste_get_lock(ptep);
1080 
1081 	pgste_val(new) &= ~bits;
1082 	pgste_val(new) |= value & bits;
1083 
1084 	pgste_set_unlock(ptep, new);
1085 	pte_unmap_unlock(ptep, ptl);
1086 	return 0;
1087 }
1088 EXPORT_SYMBOL(set_pgste_bits);
1089 
1090 /**
1091  * get_pgste - get the current PGSTE for the given address.
1092  * @mm: the memory context. It must have PGSTEs, no check is performed here!
1093  * @hva: the host virtual address of the page whose PGSTE is to be processed
1094  * @pgstep: will be written with the current PGSTE for the given address.
1095  *
1096  * Return: 0 on success, < 0 in case of error.
1097  */
1098 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1099 {
1100 	spinlock_t *ptl;
1101 	pte_t *ptep;
1102 
1103 	ptep = get_locked_pte(mm, hva, &ptl);
1104 	if (unlikely(!ptep))
1105 		return -EFAULT;
1106 	*pgstep = pgste_val(pgste_get(ptep));
1107 	pte_unmap_unlock(ptep, ptl);
1108 	return 0;
1109 }
1110 EXPORT_SYMBOL(get_pgste);
1111 #endif
1112