xref: /openbmc/linux/arch/s390/mm/pgtable.c (revision 95298d63)
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/pgalloc.h>
23 #include <asm/tlb.h>
24 #include <asm/tlbflush.h>
25 #include <asm/mmu_context.h>
26 #include <asm/page-states.h>
27 
28 static inline void ptep_ipte_local(struct mm_struct *mm, unsigned long addr,
29 				   pte_t *ptep, int nodat)
30 {
31 	unsigned long opt, asce;
32 
33 	if (MACHINE_HAS_TLB_GUEST) {
34 		opt = 0;
35 		asce = READ_ONCE(mm->context.gmap_asce);
36 		if (asce == 0UL || nodat)
37 			opt |= IPTE_NODAT;
38 		if (asce != -1UL) {
39 			asce = asce ? : mm->context.asce;
40 			opt |= IPTE_GUEST_ASCE;
41 		}
42 		__ptep_ipte(addr, ptep, opt, asce, IPTE_LOCAL);
43 	} else {
44 		__ptep_ipte(addr, ptep, 0, 0, IPTE_LOCAL);
45 	}
46 }
47 
48 static inline void ptep_ipte_global(struct mm_struct *mm, unsigned long addr,
49 				    pte_t *ptep, int nodat)
50 {
51 	unsigned long opt, asce;
52 
53 	if (MACHINE_HAS_TLB_GUEST) {
54 		opt = 0;
55 		asce = READ_ONCE(mm->context.gmap_asce);
56 		if (asce == 0UL || nodat)
57 			opt |= IPTE_NODAT;
58 		if (asce != -1UL) {
59 			asce = asce ? : mm->context.asce;
60 			opt |= IPTE_GUEST_ASCE;
61 		}
62 		__ptep_ipte(addr, ptep, opt, asce, IPTE_GLOBAL);
63 	} else {
64 		__ptep_ipte(addr, ptep, 0, 0, IPTE_GLOBAL);
65 	}
66 }
67 
68 static inline pte_t ptep_flush_direct(struct mm_struct *mm,
69 				      unsigned long addr, pte_t *ptep,
70 				      int nodat)
71 {
72 	pte_t old;
73 
74 	old = *ptep;
75 	if (unlikely(pte_val(old) & _PAGE_INVALID))
76 		return old;
77 	atomic_inc(&mm->context.flush_count);
78 	if (MACHINE_HAS_TLB_LC &&
79 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
80 		ptep_ipte_local(mm, addr, ptep, nodat);
81 	else
82 		ptep_ipte_global(mm, addr, ptep, nodat);
83 	atomic_dec(&mm->context.flush_count);
84 	return old;
85 }
86 
87 static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
88 				    unsigned long addr, pte_t *ptep,
89 				    int nodat)
90 {
91 	pte_t old;
92 
93 	old = *ptep;
94 	if (unlikely(pte_val(old) & _PAGE_INVALID))
95 		return old;
96 	atomic_inc(&mm->context.flush_count);
97 	if (cpumask_equal(&mm->context.cpu_attach_mask,
98 			  cpumask_of(smp_processor_id()))) {
99 		pte_val(*ptep) |= _PAGE_INVALID;
100 		mm->context.flush_mm = 1;
101 	} else
102 		ptep_ipte_global(mm, addr, ptep, nodat);
103 	atomic_dec(&mm->context.flush_count);
104 	return old;
105 }
106 
107 static inline pgste_t pgste_get_lock(pte_t *ptep)
108 {
109 	unsigned long new = 0;
110 #ifdef CONFIG_PGSTE
111 	unsigned long old;
112 
113 	asm(
114 		"	lg	%0,%2\n"
115 		"0:	lgr	%1,%0\n"
116 		"	nihh	%0,0xff7f\n"	/* clear PCL bit in old */
117 		"	oihh	%1,0x0080\n"	/* set PCL bit in new */
118 		"	csg	%0,%1,%2\n"
119 		"	jl	0b\n"
120 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
121 		: "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
122 #endif
123 	return __pgste(new);
124 }
125 
126 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
127 {
128 #ifdef CONFIG_PGSTE
129 	asm(
130 		"	nihh	%1,0xff7f\n"	/* clear PCL bit */
131 		"	stg	%1,%0\n"
132 		: "=Q" (ptep[PTRS_PER_PTE])
133 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
134 		: "cc", "memory");
135 #endif
136 }
137 
138 static inline pgste_t pgste_get(pte_t *ptep)
139 {
140 	unsigned long pgste = 0;
141 #ifdef CONFIG_PGSTE
142 	pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
143 #endif
144 	return __pgste(pgste);
145 }
146 
147 static inline void pgste_set(pte_t *ptep, pgste_t pgste)
148 {
149 #ifdef CONFIG_PGSTE
150 	*(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
151 #endif
152 }
153 
154 static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
155 				       struct mm_struct *mm)
156 {
157 #ifdef CONFIG_PGSTE
158 	unsigned long address, bits, skey;
159 
160 	if (!mm_uses_skeys(mm) || pte_val(pte) & _PAGE_INVALID)
161 		return pgste;
162 	address = pte_val(pte) & PAGE_MASK;
163 	skey = (unsigned long) page_get_storage_key(address);
164 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
165 	/* Transfer page changed & referenced bit to guest bits in pgste */
166 	pgste_val(pgste) |= bits << 48;		/* GR bit & GC bit */
167 	/* Copy page access key and fetch protection bit to pgste */
168 	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
169 	pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
170 #endif
171 	return pgste;
172 
173 }
174 
175 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
176 				 struct mm_struct *mm)
177 {
178 #ifdef CONFIG_PGSTE
179 	unsigned long address;
180 	unsigned long nkey;
181 
182 	if (!mm_uses_skeys(mm) || pte_val(entry) & _PAGE_INVALID)
183 		return;
184 	VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
185 	address = pte_val(entry) & PAGE_MASK;
186 	/*
187 	 * Set page access key and fetch protection bit from pgste.
188 	 * The guest C/R information is still in the PGSTE, set real
189 	 * key C/R to 0.
190 	 */
191 	nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
192 	nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
193 	page_set_storage_key(address, nkey, 0);
194 #endif
195 }
196 
197 static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
198 {
199 #ifdef CONFIG_PGSTE
200 	if ((pte_val(entry) & _PAGE_PRESENT) &&
201 	    (pte_val(entry) & _PAGE_WRITE) &&
202 	    !(pte_val(entry) & _PAGE_INVALID)) {
203 		if (!MACHINE_HAS_ESOP) {
204 			/*
205 			 * Without enhanced suppression-on-protection force
206 			 * the dirty bit on for all writable ptes.
207 			 */
208 			pte_val(entry) |= _PAGE_DIRTY;
209 			pte_val(entry) &= ~_PAGE_PROTECT;
210 		}
211 		if (!(pte_val(entry) & _PAGE_PROTECT))
212 			/* This pte allows write access, set user-dirty */
213 			pgste_val(pgste) |= PGSTE_UC_BIT;
214 	}
215 #endif
216 	*ptep = entry;
217 	return pgste;
218 }
219 
220 static inline pgste_t pgste_pte_notify(struct mm_struct *mm,
221 				       unsigned long addr,
222 				       pte_t *ptep, pgste_t pgste)
223 {
224 #ifdef CONFIG_PGSTE
225 	unsigned long bits;
226 
227 	bits = pgste_val(pgste) & (PGSTE_IN_BIT | PGSTE_VSIE_BIT);
228 	if (bits) {
229 		pgste_val(pgste) ^= bits;
230 		ptep_notify(mm, addr, ptep, bits);
231 	}
232 #endif
233 	return pgste;
234 }
235 
236 static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
237 				      unsigned long addr, pte_t *ptep)
238 {
239 	pgste_t pgste = __pgste(0);
240 
241 	if (mm_has_pgste(mm)) {
242 		pgste = pgste_get_lock(ptep);
243 		pgste = pgste_pte_notify(mm, addr, ptep, pgste);
244 	}
245 	return pgste;
246 }
247 
248 static inline pte_t ptep_xchg_commit(struct mm_struct *mm,
249 				    unsigned long addr, pte_t *ptep,
250 				    pgste_t pgste, pte_t old, pte_t new)
251 {
252 	if (mm_has_pgste(mm)) {
253 		if (pte_val(old) & _PAGE_INVALID)
254 			pgste_set_key(ptep, pgste, new, mm);
255 		if (pte_val(new) & _PAGE_INVALID) {
256 			pgste = pgste_update_all(old, pgste, mm);
257 			if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
258 			    _PGSTE_GPS_USAGE_UNUSED)
259 				pte_val(old) |= _PAGE_UNUSED;
260 		}
261 		pgste = pgste_set_pte(ptep, pgste, new);
262 		pgste_set_unlock(ptep, pgste);
263 	} else {
264 		*ptep = new;
265 	}
266 	return old;
267 }
268 
269 pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
270 		       pte_t *ptep, pte_t new)
271 {
272 	pgste_t pgste;
273 	pte_t old;
274 	int nodat;
275 
276 	preempt_disable();
277 	pgste = ptep_xchg_start(mm, addr, ptep);
278 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
279 	old = ptep_flush_direct(mm, addr, ptep, nodat);
280 	old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
281 	preempt_enable();
282 	return old;
283 }
284 EXPORT_SYMBOL(ptep_xchg_direct);
285 
286 pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
287 		     pte_t *ptep, pte_t new)
288 {
289 	pgste_t pgste;
290 	pte_t old;
291 	int nodat;
292 
293 	preempt_disable();
294 	pgste = ptep_xchg_start(mm, addr, ptep);
295 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
296 	old = ptep_flush_lazy(mm, addr, ptep, nodat);
297 	old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
298 	preempt_enable();
299 	return old;
300 }
301 EXPORT_SYMBOL(ptep_xchg_lazy);
302 
303 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr,
304 			     pte_t *ptep)
305 {
306 	pgste_t pgste;
307 	pte_t old;
308 	int nodat;
309 	struct mm_struct *mm = vma->vm_mm;
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 
322 void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
323 			     pte_t *ptep, pte_t old_pte, pte_t pte)
324 {
325 	pgste_t pgste;
326 	struct mm_struct *mm = vma->vm_mm;
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 
341 static inline void pmdp_idte_local(struct mm_struct *mm,
342 				   unsigned long addr, pmd_t *pmdp)
343 {
344 	if (MACHINE_HAS_TLB_GUEST)
345 		__pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
346 			    mm->context.asce, IDTE_LOCAL);
347 	else
348 		__pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
349 	if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
350 		gmap_pmdp_idte_local(mm, addr);
351 }
352 
353 static inline void pmdp_idte_global(struct mm_struct *mm,
354 				    unsigned long addr, pmd_t *pmdp)
355 {
356 	if (MACHINE_HAS_TLB_GUEST) {
357 		__pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
358 			    mm->context.asce, IDTE_GLOBAL);
359 		if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
360 			gmap_pmdp_idte_global(mm, addr);
361 	} else if (MACHINE_HAS_IDTE) {
362 		__pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
363 		if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
364 			gmap_pmdp_idte_global(mm, addr);
365 	} else {
366 		__pmdp_csp(pmdp);
367 		if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
368 			gmap_pmdp_csp(mm, addr);
369 	}
370 }
371 
372 static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
373 				      unsigned long addr, pmd_t *pmdp)
374 {
375 	pmd_t old;
376 
377 	old = *pmdp;
378 	if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
379 		return old;
380 	atomic_inc(&mm->context.flush_count);
381 	if (MACHINE_HAS_TLB_LC &&
382 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
383 		pmdp_idte_local(mm, addr, pmdp);
384 	else
385 		pmdp_idte_global(mm, addr, pmdp);
386 	atomic_dec(&mm->context.flush_count);
387 	return old;
388 }
389 
390 static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
391 				    unsigned long addr, pmd_t *pmdp)
392 {
393 	pmd_t old;
394 
395 	old = *pmdp;
396 	if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
397 		return old;
398 	atomic_inc(&mm->context.flush_count);
399 	if (cpumask_equal(&mm->context.cpu_attach_mask,
400 			  cpumask_of(smp_processor_id()))) {
401 		pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
402 		mm->context.flush_mm = 1;
403 		if (mm_has_pgste(mm))
404 			gmap_pmdp_invalidate(mm, addr);
405 	} else {
406 		pmdp_idte_global(mm, addr, pmdp);
407 	}
408 	atomic_dec(&mm->context.flush_count);
409 	return old;
410 }
411 
412 #ifdef CONFIG_PGSTE
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 #endif
431 
432 pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
433 		       pmd_t *pmdp, pmd_t new)
434 {
435 	pmd_t old;
436 
437 	preempt_disable();
438 	old = pmdp_flush_direct(mm, addr, pmdp);
439 	*pmdp = new;
440 	preempt_enable();
441 	return old;
442 }
443 EXPORT_SYMBOL(pmdp_xchg_direct);
444 
445 pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
446 		     pmd_t *pmdp, pmd_t new)
447 {
448 	pmd_t old;
449 
450 	preempt_disable();
451 	old = pmdp_flush_lazy(mm, addr, pmdp);
452 	*pmdp = new;
453 	preempt_enable();
454 	return old;
455 }
456 EXPORT_SYMBOL(pmdp_xchg_lazy);
457 
458 static inline void pudp_idte_local(struct mm_struct *mm,
459 				   unsigned long addr, pud_t *pudp)
460 {
461 	if (MACHINE_HAS_TLB_GUEST)
462 		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
463 			    mm->context.asce, IDTE_LOCAL);
464 	else
465 		__pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
466 }
467 
468 static inline void pudp_idte_global(struct mm_struct *mm,
469 				    unsigned long addr, pud_t *pudp)
470 {
471 	if (MACHINE_HAS_TLB_GUEST)
472 		__pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
473 			    mm->context.asce, IDTE_GLOBAL);
474 	else if (MACHINE_HAS_IDTE)
475 		__pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
476 	else
477 		/*
478 		 * Invalid bit position is the same for pmd and pud, so we can
479 		 * re-use _pmd_csp() here
480 		 */
481 		__pmdp_csp((pmd_t *) pudp);
482 }
483 
484 static inline pud_t pudp_flush_direct(struct mm_struct *mm,
485 				      unsigned long addr, pud_t *pudp)
486 {
487 	pud_t old;
488 
489 	old = *pudp;
490 	if (pud_val(old) & _REGION_ENTRY_INVALID)
491 		return old;
492 	atomic_inc(&mm->context.flush_count);
493 	if (MACHINE_HAS_TLB_LC &&
494 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
495 		pudp_idte_local(mm, addr, pudp);
496 	else
497 		pudp_idte_global(mm, addr, pudp);
498 	atomic_dec(&mm->context.flush_count);
499 	return old;
500 }
501 
502 pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
503 		       pud_t *pudp, pud_t new)
504 {
505 	pud_t old;
506 
507 	preempt_disable();
508 	old = pudp_flush_direct(mm, addr, pudp);
509 	*pudp = new;
510 	preempt_enable();
511 	return old;
512 }
513 EXPORT_SYMBOL(pudp_xchg_direct);
514 
515 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
516 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
517 				pgtable_t pgtable)
518 {
519 	struct list_head *lh = (struct list_head *) pgtable;
520 
521 	assert_spin_locked(pmd_lockptr(mm, pmdp));
522 
523 	/* FIFO */
524 	if (!pmd_huge_pte(mm, pmdp))
525 		INIT_LIST_HEAD(lh);
526 	else
527 		list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
528 	pmd_huge_pte(mm, pmdp) = pgtable;
529 }
530 
531 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
532 {
533 	struct list_head *lh;
534 	pgtable_t pgtable;
535 	pte_t *ptep;
536 
537 	assert_spin_locked(pmd_lockptr(mm, pmdp));
538 
539 	/* FIFO */
540 	pgtable = pmd_huge_pte(mm, pmdp);
541 	lh = (struct list_head *) pgtable;
542 	if (list_empty(lh))
543 		pmd_huge_pte(mm, pmdp) = NULL;
544 	else {
545 		pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
546 		list_del(lh);
547 	}
548 	ptep = (pte_t *) pgtable;
549 	pte_val(*ptep) = _PAGE_INVALID;
550 	ptep++;
551 	pte_val(*ptep) = _PAGE_INVALID;
552 	return pgtable;
553 }
554 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
555 
556 #ifdef CONFIG_PGSTE
557 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
558 		     pte_t *ptep, pte_t entry)
559 {
560 	pgste_t pgste;
561 
562 	/* the mm_has_pgste() check is done in set_pte_at() */
563 	preempt_disable();
564 	pgste = pgste_get_lock(ptep);
565 	pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
566 	pgste_set_key(ptep, pgste, entry, mm);
567 	pgste = pgste_set_pte(ptep, pgste, entry);
568 	pgste_set_unlock(ptep, pgste);
569 	preempt_enable();
570 }
571 
572 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
573 {
574 	pgste_t pgste;
575 
576 	preempt_disable();
577 	pgste = pgste_get_lock(ptep);
578 	pgste_val(pgste) |= PGSTE_IN_BIT;
579 	pgste_set_unlock(ptep, pgste);
580 	preempt_enable();
581 }
582 
583 /**
584  * ptep_force_prot - change access rights of a locked pte
585  * @mm: pointer to the process mm_struct
586  * @addr: virtual address in the guest address space
587  * @ptep: pointer to the page table entry
588  * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
589  * @bit: pgste bit to set (e.g. for notification)
590  *
591  * Returns 0 if the access rights were changed and -EAGAIN if the current
592  * and requested access rights are incompatible.
593  */
594 int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
595 		    pte_t *ptep, int prot, unsigned long bit)
596 {
597 	pte_t entry;
598 	pgste_t pgste;
599 	int pte_i, pte_p, nodat;
600 
601 	pgste = pgste_get_lock(ptep);
602 	entry = *ptep;
603 	/* Check pte entry after all locks have been acquired */
604 	pte_i = pte_val(entry) & _PAGE_INVALID;
605 	pte_p = pte_val(entry) & _PAGE_PROTECT;
606 	if ((pte_i && (prot != PROT_NONE)) ||
607 	    (pte_p && (prot & PROT_WRITE))) {
608 		pgste_set_unlock(ptep, pgste);
609 		return -EAGAIN;
610 	}
611 	/* Change access rights and set pgste bit */
612 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
613 	if (prot == PROT_NONE && !pte_i) {
614 		ptep_flush_direct(mm, addr, ptep, nodat);
615 		pgste = pgste_update_all(entry, pgste, mm);
616 		pte_val(entry) |= _PAGE_INVALID;
617 	}
618 	if (prot == PROT_READ && !pte_p) {
619 		ptep_flush_direct(mm, addr, ptep, nodat);
620 		pte_val(entry) &= ~_PAGE_INVALID;
621 		pte_val(entry) |= _PAGE_PROTECT;
622 	}
623 	pgste_val(pgste) |= bit;
624 	pgste = pgste_set_pte(ptep, pgste, entry);
625 	pgste_set_unlock(ptep, pgste);
626 	return 0;
627 }
628 
629 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
630 		    pte_t *sptep, pte_t *tptep, pte_t pte)
631 {
632 	pgste_t spgste, tpgste;
633 	pte_t spte, tpte;
634 	int rc = -EAGAIN;
635 
636 	if (!(pte_val(*tptep) & _PAGE_INVALID))
637 		return 0;	/* already shadowed */
638 	spgste = pgste_get_lock(sptep);
639 	spte = *sptep;
640 	if (!(pte_val(spte) & _PAGE_INVALID) &&
641 	    !((pte_val(spte) & _PAGE_PROTECT) &&
642 	      !(pte_val(pte) & _PAGE_PROTECT))) {
643 		pgste_val(spgste) |= PGSTE_VSIE_BIT;
644 		tpgste = pgste_get_lock(tptep);
645 		pte_val(tpte) = (pte_val(spte) & PAGE_MASK) |
646 				(pte_val(pte) & _PAGE_PROTECT);
647 		/* don't touch the storage key - it belongs to parent pgste */
648 		tpgste = pgste_set_pte(tptep, tpgste, tpte);
649 		pgste_set_unlock(tptep, tpgste);
650 		rc = 1;
651 	}
652 	pgste_set_unlock(sptep, spgste);
653 	return rc;
654 }
655 
656 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
657 {
658 	pgste_t pgste;
659 	int nodat;
660 
661 	pgste = pgste_get_lock(ptep);
662 	/* notifier is called by the caller */
663 	nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
664 	ptep_flush_direct(mm, saddr, ptep, nodat);
665 	/* don't touch the storage key - it belongs to parent pgste */
666 	pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
667 	pgste_set_unlock(ptep, pgste);
668 }
669 
670 static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
671 {
672 	if (!non_swap_entry(entry))
673 		dec_mm_counter(mm, MM_SWAPENTS);
674 	else if (is_migration_entry(entry)) {
675 		struct page *page = migration_entry_to_page(entry);
676 
677 		dec_mm_counter(mm, mm_counter(page));
678 	}
679 	free_swap_and_cache(entry);
680 }
681 
682 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
683 		     pte_t *ptep, int reset)
684 {
685 	unsigned long pgstev;
686 	pgste_t pgste;
687 	pte_t pte;
688 
689 	/* Zap unused and logically-zero pages */
690 	preempt_disable();
691 	pgste = pgste_get_lock(ptep);
692 	pgstev = pgste_val(pgste);
693 	pte = *ptep;
694 	if (!reset && pte_swap(pte) &&
695 	    ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
696 	     (pgstev & _PGSTE_GPS_ZERO))) {
697 		ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
698 		pte_clear(mm, addr, ptep);
699 	}
700 	if (reset)
701 		pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
702 	pgste_set_unlock(ptep, pgste);
703 	preempt_enable();
704 }
705 
706 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
707 {
708 	unsigned long ptev;
709 	pgste_t pgste;
710 
711 	/* Clear storage key ACC and F, but set R/C */
712 	preempt_disable();
713 	pgste = pgste_get_lock(ptep);
714 	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
715 	pgste_val(pgste) |= PGSTE_GR_BIT | PGSTE_GC_BIT;
716 	ptev = pte_val(*ptep);
717 	if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
718 		page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 1);
719 	pgste_set_unlock(ptep, pgste);
720 	preempt_enable();
721 }
722 
723 /*
724  * Test and reset if a guest page is dirty
725  */
726 bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long addr,
727 		       pte_t *ptep)
728 {
729 	pgste_t pgste;
730 	pte_t pte;
731 	bool dirty;
732 	int nodat;
733 
734 	pgste = pgste_get_lock(ptep);
735 	dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
736 	pgste_val(pgste) &= ~PGSTE_UC_BIT;
737 	pte = *ptep;
738 	if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
739 		pgste = pgste_pte_notify(mm, addr, ptep, pgste);
740 		nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
741 		ptep_ipte_global(mm, addr, ptep, nodat);
742 		if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
743 			pte_val(pte) |= _PAGE_PROTECT;
744 		else
745 			pte_val(pte) |= _PAGE_INVALID;
746 		*ptep = pte;
747 	}
748 	pgste_set_unlock(ptep, pgste);
749 	return dirty;
750 }
751 EXPORT_SYMBOL_GPL(ptep_test_and_clear_uc);
752 
753 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
754 			  unsigned char key, bool nq)
755 {
756 	unsigned long keyul, paddr;
757 	spinlock_t *ptl;
758 	pgste_t old, new;
759 	pmd_t *pmdp;
760 	pte_t *ptep;
761 
762 	pmdp = pmd_alloc_map(mm, addr);
763 	if (unlikely(!pmdp))
764 		return -EFAULT;
765 
766 	ptl = pmd_lock(mm, pmdp);
767 	if (!pmd_present(*pmdp)) {
768 		spin_unlock(ptl);
769 		return -EFAULT;
770 	}
771 
772 	if (pmd_large(*pmdp)) {
773 		paddr = pmd_val(*pmdp) & HPAGE_MASK;
774 		paddr |= addr & ~HPAGE_MASK;
775 		/*
776 		 * Huge pmds need quiescing operations, they are
777 		 * always mapped.
778 		 */
779 		page_set_storage_key(paddr, key, 1);
780 		spin_unlock(ptl);
781 		return 0;
782 	}
783 	spin_unlock(ptl);
784 
785 	ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
786 	if (unlikely(!ptep))
787 		return -EFAULT;
788 
789 	new = old = pgste_get_lock(ptep);
790 	pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
791 			    PGSTE_ACC_BITS | PGSTE_FP_BIT);
792 	keyul = (unsigned long) key;
793 	pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
794 	pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
795 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
796 		unsigned long bits, skey;
797 
798 		paddr = pte_val(*ptep) & PAGE_MASK;
799 		skey = (unsigned long) page_get_storage_key(paddr);
800 		bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
801 		skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
802 		/* Set storage key ACC and FP */
803 		page_set_storage_key(paddr, skey, !nq);
804 		/* Merge host changed & referenced into pgste  */
805 		pgste_val(new) |= bits << 52;
806 	}
807 	/* changing the guest storage key is considered a change of the page */
808 	if ((pgste_val(new) ^ pgste_val(old)) &
809 	    (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
810 		pgste_val(new) |= PGSTE_UC_BIT;
811 
812 	pgste_set_unlock(ptep, new);
813 	pte_unmap_unlock(ptep, ptl);
814 	return 0;
815 }
816 EXPORT_SYMBOL(set_guest_storage_key);
817 
818 /**
819  * Conditionally set a guest storage key (handling csske).
820  * oldkey will be updated when either mr or mc is set and a pointer is given.
821  *
822  * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
823  * storage key was updated and -EFAULT on access errors.
824  */
825 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
826 			       unsigned char key, unsigned char *oldkey,
827 			       bool nq, bool mr, bool mc)
828 {
829 	unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
830 	int rc;
831 
832 	/* we can drop the pgste lock between getting and setting the key */
833 	if (mr | mc) {
834 		rc = get_guest_storage_key(current->mm, addr, &tmp);
835 		if (rc)
836 			return rc;
837 		if (oldkey)
838 			*oldkey = tmp;
839 		if (!mr)
840 			mask |= _PAGE_REFERENCED;
841 		if (!mc)
842 			mask |= _PAGE_CHANGED;
843 		if (!((tmp ^ key) & mask))
844 			return 0;
845 	}
846 	rc = set_guest_storage_key(current->mm, addr, key, nq);
847 	return rc < 0 ? rc : 1;
848 }
849 EXPORT_SYMBOL(cond_set_guest_storage_key);
850 
851 /**
852  * Reset a guest reference bit (rrbe), returning the reference and changed bit.
853  *
854  * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
855  */
856 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
857 {
858 	spinlock_t *ptl;
859 	unsigned long paddr;
860 	pgste_t old, new;
861 	pmd_t *pmdp;
862 	pte_t *ptep;
863 	int cc = 0;
864 
865 	pmdp = pmd_alloc_map(mm, addr);
866 	if (unlikely(!pmdp))
867 		return -EFAULT;
868 
869 	ptl = pmd_lock(mm, pmdp);
870 	if (!pmd_present(*pmdp)) {
871 		spin_unlock(ptl);
872 		return -EFAULT;
873 	}
874 
875 	if (pmd_large(*pmdp)) {
876 		paddr = pmd_val(*pmdp) & HPAGE_MASK;
877 		paddr |= addr & ~HPAGE_MASK;
878 		cc = page_reset_referenced(paddr);
879 		spin_unlock(ptl);
880 		return cc;
881 	}
882 	spin_unlock(ptl);
883 
884 	ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
885 	if (unlikely(!ptep))
886 		return -EFAULT;
887 
888 	new = old = pgste_get_lock(ptep);
889 	/* Reset guest reference bit only */
890 	pgste_val(new) &= ~PGSTE_GR_BIT;
891 
892 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
893 		paddr = pte_val(*ptep) & PAGE_MASK;
894 		cc = page_reset_referenced(paddr);
895 		/* Merge real referenced bit into host-set */
896 		pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
897 	}
898 	/* Reflect guest's logical view, not physical */
899 	cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
900 	/* Changing the guest storage key is considered a change of the page */
901 	if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
902 		pgste_val(new) |= PGSTE_UC_BIT;
903 
904 	pgste_set_unlock(ptep, new);
905 	pte_unmap_unlock(ptep, ptl);
906 	return cc;
907 }
908 EXPORT_SYMBOL(reset_guest_reference_bit);
909 
910 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
911 			  unsigned char *key)
912 {
913 	unsigned long paddr;
914 	spinlock_t *ptl;
915 	pgste_t pgste;
916 	pmd_t *pmdp;
917 	pte_t *ptep;
918 
919 	pmdp = pmd_alloc_map(mm, addr);
920 	if (unlikely(!pmdp))
921 		return -EFAULT;
922 
923 	ptl = pmd_lock(mm, pmdp);
924 	if (!pmd_present(*pmdp)) {
925 		/* Not yet mapped memory has a zero key */
926 		spin_unlock(ptl);
927 		*key = 0;
928 		return 0;
929 	}
930 
931 	if (pmd_large(*pmdp)) {
932 		paddr = pmd_val(*pmdp) & HPAGE_MASK;
933 		paddr |= addr & ~HPAGE_MASK;
934 		*key = page_get_storage_key(paddr);
935 		spin_unlock(ptl);
936 		return 0;
937 	}
938 	spin_unlock(ptl);
939 
940 	ptep = pte_alloc_map_lock(mm, pmdp, addr, &ptl);
941 	if (unlikely(!ptep))
942 		return -EFAULT;
943 
944 	pgste = pgste_get_lock(ptep);
945 	*key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
946 	paddr = pte_val(*ptep) & PAGE_MASK;
947 	if (!(pte_val(*ptep) & _PAGE_INVALID))
948 		*key = page_get_storage_key(paddr);
949 	/* Reflect guest's logical view, not physical */
950 	*key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
951 	pgste_set_unlock(ptep, pgste);
952 	pte_unmap_unlock(ptep, ptl);
953 	return 0;
954 }
955 EXPORT_SYMBOL(get_guest_storage_key);
956 
957 /**
958  * pgste_perform_essa - perform ESSA actions on the PGSTE.
959  * @mm: the memory context. It must have PGSTEs, no check is performed here!
960  * @hva: the host virtual address of the page whose PGSTE is to be processed
961  * @orc: the specific action to perform, see the ESSA_SET_* macros.
962  * @oldpte: the PTE will be saved there if the pointer is not NULL.
963  * @oldpgste: the old PGSTE will be saved there if the pointer is not NULL.
964  *
965  * Return: 1 if the page is to be added to the CBRL, otherwise 0,
966  *	   or < 0 in case of error. -EINVAL is returned for invalid values
967  *	   of orc, -EFAULT for invalid addresses.
968  */
969 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
970 			unsigned long *oldpte, unsigned long *oldpgste)
971 {
972 	unsigned long pgstev;
973 	spinlock_t *ptl;
974 	pgste_t pgste;
975 	pte_t *ptep;
976 	int res = 0;
977 
978 	WARN_ON_ONCE(orc > ESSA_MAX);
979 	if (unlikely(orc > ESSA_MAX))
980 		return -EINVAL;
981 	ptep = get_locked_pte(mm, hva, &ptl);
982 	if (unlikely(!ptep))
983 		return -EFAULT;
984 	pgste = pgste_get_lock(ptep);
985 	pgstev = pgste_val(pgste);
986 	if (oldpte)
987 		*oldpte = pte_val(*ptep);
988 	if (oldpgste)
989 		*oldpgste = pgstev;
990 
991 	switch (orc) {
992 	case ESSA_GET_STATE:
993 		break;
994 	case ESSA_SET_STABLE:
995 		pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
996 		pgstev |= _PGSTE_GPS_USAGE_STABLE;
997 		break;
998 	case ESSA_SET_UNUSED:
999 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1000 		pgstev |= _PGSTE_GPS_USAGE_UNUSED;
1001 		if (pte_val(*ptep) & _PAGE_INVALID)
1002 			res = 1;
1003 		break;
1004 	case ESSA_SET_VOLATILE:
1005 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1006 		pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1007 		if (pte_val(*ptep) & _PAGE_INVALID)
1008 			res = 1;
1009 		break;
1010 	case ESSA_SET_POT_VOLATILE:
1011 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1012 		if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1013 			pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
1014 			break;
1015 		}
1016 		if (pgstev & _PGSTE_GPS_ZERO) {
1017 			pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1018 			break;
1019 		}
1020 		if (!(pgstev & PGSTE_GC_BIT)) {
1021 			pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1022 			res = 1;
1023 			break;
1024 		}
1025 		break;
1026 	case ESSA_SET_STABLE_RESIDENT:
1027 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1028 		pgstev |= _PGSTE_GPS_USAGE_STABLE;
1029 		/*
1030 		 * Since the resident state can go away any time after this
1031 		 * call, we will not make this page resident. We can revisit
1032 		 * this decision if a guest will ever start using this.
1033 		 */
1034 		break;
1035 	case ESSA_SET_STABLE_IF_RESIDENT:
1036 		if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1037 			pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1038 			pgstev |= _PGSTE_GPS_USAGE_STABLE;
1039 		}
1040 		break;
1041 	case ESSA_SET_STABLE_NODAT:
1042 		pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1043 		pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
1044 		break;
1045 	default:
1046 		/* we should never get here! */
1047 		break;
1048 	}
1049 	/* If we are discarding a page, set it to logical zero */
1050 	if (res)
1051 		pgstev |= _PGSTE_GPS_ZERO;
1052 
1053 	pgste_val(pgste) = pgstev;
1054 	pgste_set_unlock(ptep, pgste);
1055 	pte_unmap_unlock(ptep, ptl);
1056 	return res;
1057 }
1058 EXPORT_SYMBOL(pgste_perform_essa);
1059 
1060 /**
1061  * set_pgste_bits - set specific PGSTE bits.
1062  * @mm: the memory context. It must have PGSTEs, no check is performed here!
1063  * @hva: the host virtual address of the page whose PGSTE is to be processed
1064  * @bits: a bitmask representing the bits that will be touched
1065  * @value: the values of the bits to be written. Only the bits in the mask
1066  *	   will be written.
1067  *
1068  * Return: 0 on success, < 0 in case of error.
1069  */
1070 int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
1071 			unsigned long bits, unsigned long value)
1072 {
1073 	spinlock_t *ptl;
1074 	pgste_t new;
1075 	pte_t *ptep;
1076 
1077 	ptep = get_locked_pte(mm, hva, &ptl);
1078 	if (unlikely(!ptep))
1079 		return -EFAULT;
1080 	new = pgste_get_lock(ptep);
1081 
1082 	pgste_val(new) &= ~bits;
1083 	pgste_val(new) |= value & bits;
1084 
1085 	pgste_set_unlock(ptep, new);
1086 	pte_unmap_unlock(ptep, ptl);
1087 	return 0;
1088 }
1089 EXPORT_SYMBOL(set_pgste_bits);
1090 
1091 /**
1092  * get_pgste - get the current PGSTE for the given address.
1093  * @mm: the memory context. It must have PGSTEs, no check is performed here!
1094  * @hva: the host virtual address of the page whose PGSTE is to be processed
1095  * @pgstep: will be written with the current PGSTE for the given address.
1096  *
1097  * Return: 0 on success, < 0 in case of error.
1098  */
1099 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1100 {
1101 	spinlock_t *ptl;
1102 	pte_t *ptep;
1103 
1104 	ptep = get_locked_pte(mm, hva, &ptl);
1105 	if (unlikely(!ptep))
1106 		return -EFAULT;
1107 	*pgstep = pgste_val(pgste_get(ptep));
1108 	pte_unmap_unlock(ptep, ptl);
1109 	return 0;
1110 }
1111 EXPORT_SYMBOL(get_pgste);
1112 #endif
1113