Loading...
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/tlb.h>
23#include <asm/tlbflush.h>
24#include <asm/mmu_context.h>
25#include <asm/page-states.h>
26
27pgprot_t pgprot_writecombine(pgprot_t prot)
28{
29 /*
30 * mio_wb_bit_mask may be set on a different CPU, but it is only set
31 * once at init and only read afterwards.
32 */
33 return __pgprot(pgprot_val(prot) | mio_wb_bit_mask);
34}
35EXPORT_SYMBOL_GPL(pgprot_writecombine);
36
37pgprot_t pgprot_writethrough(pgprot_t prot)
38{
39 /*
40 * mio_wb_bit_mask may be set on a different CPU, but it is only set
41 * once at init and only read afterwards.
42 */
43 return __pgprot(pgprot_val(prot) & ~mio_wb_bit_mask);
44}
45EXPORT_SYMBOL_GPL(pgprot_writethrough);
46
47static inline void ptep_ipte_local(struct mm_struct *mm, unsigned long addr,
48 pte_t *ptep, int nodat)
49{
50 unsigned long opt, asce;
51
52 if (MACHINE_HAS_TLB_GUEST) {
53 opt = 0;
54 asce = READ_ONCE(mm->context.gmap_asce);
55 if (asce == 0UL || nodat)
56 opt |= IPTE_NODAT;
57 if (asce != -1UL) {
58 asce = asce ? : mm->context.asce;
59 opt |= IPTE_GUEST_ASCE;
60 }
61 __ptep_ipte(addr, ptep, opt, asce, IPTE_LOCAL);
62 } else {
63 __ptep_ipte(addr, ptep, 0, 0, IPTE_LOCAL);
64 }
65}
66
67static inline void ptep_ipte_global(struct mm_struct *mm, unsigned long addr,
68 pte_t *ptep, int nodat)
69{
70 unsigned long opt, asce;
71
72 if (MACHINE_HAS_TLB_GUEST) {
73 opt = 0;
74 asce = READ_ONCE(mm->context.gmap_asce);
75 if (asce == 0UL || nodat)
76 opt |= IPTE_NODAT;
77 if (asce != -1UL) {
78 asce = asce ? : mm->context.asce;
79 opt |= IPTE_GUEST_ASCE;
80 }
81 __ptep_ipte(addr, ptep, opt, asce, IPTE_GLOBAL);
82 } else {
83 __ptep_ipte(addr, ptep, 0, 0, IPTE_GLOBAL);
84 }
85}
86
87static inline pte_t ptep_flush_direct(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 (MACHINE_HAS_TLB_LC &&
98 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
99 ptep_ipte_local(mm, addr, ptep, nodat);
100 else
101 ptep_ipte_global(mm, addr, ptep, nodat);
102 atomic_dec(&mm->context.flush_count);
103 return old;
104}
105
106static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
107 unsigned long addr, pte_t *ptep,
108 int nodat)
109{
110 pte_t old;
111
112 old = *ptep;
113 if (unlikely(pte_val(old) & _PAGE_INVALID))
114 return old;
115 atomic_inc(&mm->context.flush_count);
116 if (cpumask_equal(&mm->context.cpu_attach_mask,
117 cpumask_of(smp_processor_id()))) {
118 set_pte(ptep, set_pte_bit(*ptep, __pgprot(_PAGE_INVALID)));
119 mm->context.flush_mm = 1;
120 } else
121 ptep_ipte_global(mm, addr, ptep, nodat);
122 atomic_dec(&mm->context.flush_count);
123 return old;
124}
125
126static inline pgste_t pgste_get_lock(pte_t *ptep)
127{
128 unsigned long new = 0;
129#ifdef CONFIG_PGSTE
130 unsigned long old;
131
132 asm(
133 " lg %0,%2\n"
134 "0: lgr %1,%0\n"
135 " nihh %0,0xff7f\n" /* clear PCL bit in old */
136 " oihh %1,0x0080\n" /* set PCL bit in new */
137 " csg %0,%1,%2\n"
138 " jl 0b\n"
139 : "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
140 : "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
141#endif
142 return __pgste(new);
143}
144
145static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
146{
147#ifdef CONFIG_PGSTE
148 asm(
149 " nihh %1,0xff7f\n" /* clear PCL bit */
150 " stg %1,%0\n"
151 : "=Q" (ptep[PTRS_PER_PTE])
152 : "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
153 : "cc", "memory");
154#endif
155}
156
157static inline pgste_t pgste_get(pte_t *ptep)
158{
159 unsigned long pgste = 0;
160#ifdef CONFIG_PGSTE
161 pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
162#endif
163 return __pgste(pgste);
164}
165
166static inline void pgste_set(pte_t *ptep, pgste_t pgste)
167{
168#ifdef CONFIG_PGSTE
169 *(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
170#endif
171}
172
173static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
174 struct mm_struct *mm)
175{
176#ifdef CONFIG_PGSTE
177 unsigned long address, bits, skey;
178
179 if (!mm_uses_skeys(mm) || pte_val(pte) & _PAGE_INVALID)
180 return pgste;
181 address = pte_val(pte) & PAGE_MASK;
182 skey = (unsigned long) page_get_storage_key(address);
183 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
184 /* Transfer page changed & referenced bit to guest bits in pgste */
185 pgste_val(pgste) |= bits << 48; /* GR bit & GC bit */
186 /* Copy page access key and fetch protection bit to pgste */
187 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
188 pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
189#endif
190 return pgste;
191
192}
193
194static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
195 struct mm_struct *mm)
196{
197#ifdef CONFIG_PGSTE
198 unsigned long address;
199 unsigned long nkey;
200
201 if (!mm_uses_skeys(mm) || pte_val(entry) & _PAGE_INVALID)
202 return;
203 VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
204 address = pte_val(entry) & PAGE_MASK;
205 /*
206 * Set page access key and fetch protection bit from pgste.
207 * The guest C/R information is still in the PGSTE, set real
208 * key C/R to 0.
209 */
210 nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
211 nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
212 page_set_storage_key(address, nkey, 0);
213#endif
214}
215
216static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
217{
218#ifdef CONFIG_PGSTE
219 if ((pte_val(entry) & _PAGE_PRESENT) &&
220 (pte_val(entry) & _PAGE_WRITE) &&
221 !(pte_val(entry) & _PAGE_INVALID)) {
222 if (!MACHINE_HAS_ESOP) {
223 /*
224 * Without enhanced suppression-on-protection force
225 * the dirty bit on for all writable ptes.
226 */
227 entry = set_pte_bit(entry, __pgprot(_PAGE_DIRTY));
228 entry = clear_pte_bit(entry, __pgprot(_PAGE_PROTECT));
229 }
230 if (!(pte_val(entry) & _PAGE_PROTECT))
231 /* This pte allows write access, set user-dirty */
232 pgste_val(pgste) |= PGSTE_UC_BIT;
233 }
234#endif
235 set_pte(ptep, entry);
236 return pgste;
237}
238
239static inline pgste_t pgste_pte_notify(struct mm_struct *mm,
240 unsigned long addr,
241 pte_t *ptep, pgste_t pgste)
242{
243#ifdef CONFIG_PGSTE
244 unsigned long bits;
245
246 bits = pgste_val(pgste) & (PGSTE_IN_BIT | PGSTE_VSIE_BIT);
247 if (bits) {
248 pgste_val(pgste) ^= bits;
249 ptep_notify(mm, addr, ptep, bits);
250 }
251#endif
252 return pgste;
253}
254
255static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
256 unsigned long addr, pte_t *ptep)
257{
258 pgste_t pgste = __pgste(0);
259
260 if (mm_has_pgste(mm)) {
261 pgste = pgste_get_lock(ptep);
262 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
263 }
264 return pgste;
265}
266
267static inline pte_t ptep_xchg_commit(struct mm_struct *mm,
268 unsigned long addr, pte_t *ptep,
269 pgste_t pgste, pte_t old, pte_t new)
270{
271 if (mm_has_pgste(mm)) {
272 if (pte_val(old) & _PAGE_INVALID)
273 pgste_set_key(ptep, pgste, new, mm);
274 if (pte_val(new) & _PAGE_INVALID) {
275 pgste = pgste_update_all(old, pgste, mm);
276 if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
277 _PGSTE_GPS_USAGE_UNUSED)
278 old = set_pte_bit(old, __pgprot(_PAGE_UNUSED));
279 }
280 pgste = pgste_set_pte(ptep, pgste, new);
281 pgste_set_unlock(ptep, pgste);
282 } else {
283 set_pte(ptep, new);
284 }
285 return old;
286}
287
288pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
289 pte_t *ptep, pte_t new)
290{
291 pgste_t pgste;
292 pte_t old;
293 int nodat;
294
295 preempt_disable();
296 pgste = ptep_xchg_start(mm, addr, ptep);
297 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
298 old = ptep_flush_direct(mm, addr, ptep, nodat);
299 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
300 preempt_enable();
301 return old;
302}
303EXPORT_SYMBOL(ptep_xchg_direct);
304
305pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
306 pte_t *ptep, pte_t new)
307{
308 pgste_t pgste;
309 pte_t old;
310 int nodat;
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 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
317 preempt_enable();
318 return old;
319}
320EXPORT_SYMBOL(ptep_xchg_lazy);
321
322pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr,
323 pte_t *ptep)
324{
325 pgste_t pgste;
326 pte_t old;
327 int nodat;
328 struct mm_struct *mm = vma->vm_mm;
329
330 preempt_disable();
331 pgste = ptep_xchg_start(mm, addr, ptep);
332 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
333 old = ptep_flush_lazy(mm, addr, ptep, nodat);
334 if (mm_has_pgste(mm)) {
335 pgste = pgste_update_all(old, pgste, mm);
336 pgste_set(ptep, pgste);
337 }
338 return old;
339}
340
341void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
342 pte_t *ptep, pte_t old_pte, pte_t pte)
343{
344 pgste_t pgste;
345 struct mm_struct *mm = vma->vm_mm;
346
347 if (!MACHINE_HAS_NX)
348 pte = clear_pte_bit(pte, __pgprot(_PAGE_NOEXEC));
349 if (mm_has_pgste(mm)) {
350 pgste = pgste_get(ptep);
351 pgste_set_key(ptep, pgste, pte, mm);
352 pgste = pgste_set_pte(ptep, pgste, pte);
353 pgste_set_unlock(ptep, pgste);
354 } else {
355 set_pte(ptep, pte);
356 }
357 preempt_enable();
358}
359
360static inline void pmdp_idte_local(struct mm_struct *mm,
361 unsigned long addr, pmd_t *pmdp)
362{
363 if (MACHINE_HAS_TLB_GUEST)
364 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
365 mm->context.asce, IDTE_LOCAL);
366 else
367 __pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
368 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
369 gmap_pmdp_idte_local(mm, addr);
370}
371
372static inline void pmdp_idte_global(struct mm_struct *mm,
373 unsigned long addr, pmd_t *pmdp)
374{
375 if (MACHINE_HAS_TLB_GUEST) {
376 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
377 mm->context.asce, IDTE_GLOBAL);
378 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
379 gmap_pmdp_idte_global(mm, addr);
380 } else if (MACHINE_HAS_IDTE) {
381 __pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
382 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
383 gmap_pmdp_idte_global(mm, addr);
384 } else {
385 __pmdp_csp(pmdp);
386 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
387 gmap_pmdp_csp(mm, addr);
388 }
389}
390
391static inline pmd_t pmdp_flush_direct(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 (MACHINE_HAS_TLB_LC &&
401 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
402 pmdp_idte_local(mm, addr, pmdp);
403 else
404 pmdp_idte_global(mm, addr, pmdp);
405 atomic_dec(&mm->context.flush_count);
406 return old;
407}
408
409static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
410 unsigned long addr, pmd_t *pmdp)
411{
412 pmd_t old;
413
414 old = *pmdp;
415 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
416 return old;
417 atomic_inc(&mm->context.flush_count);
418 if (cpumask_equal(&mm->context.cpu_attach_mask,
419 cpumask_of(smp_processor_id()))) {
420 set_pmd(pmdp, set_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_INVALID)));
421 mm->context.flush_mm = 1;
422 if (mm_has_pgste(mm))
423 gmap_pmdp_invalidate(mm, addr);
424 } else {
425 pmdp_idte_global(mm, addr, pmdp);
426 }
427 atomic_dec(&mm->context.flush_count);
428 return old;
429}
430
431#ifdef CONFIG_PGSTE
432static int pmd_lookup(struct mm_struct *mm, unsigned long addr, pmd_t **pmdp)
433{
434 struct vm_area_struct *vma;
435 pgd_t *pgd;
436 p4d_t *p4d;
437 pud_t *pud;
438
439 /* We need a valid VMA, otherwise this is clearly a fault. */
440 vma = vma_lookup(mm, addr);
441 if (!vma)
442 return -EFAULT;
443
444 pgd = pgd_offset(mm, addr);
445 if (!pgd_present(*pgd))
446 return -ENOENT;
447
448 p4d = p4d_offset(pgd, addr);
449 if (!p4d_present(*p4d))
450 return -ENOENT;
451
452 pud = pud_offset(p4d, addr);
453 if (!pud_present(*pud))
454 return -ENOENT;
455
456 /* Large PUDs are not supported yet. */
457 if (pud_large(*pud))
458 return -EFAULT;
459
460 *pmdp = pmd_offset(pud, addr);
461 return 0;
462}
463#endif
464
465pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
466 pmd_t *pmdp, pmd_t new)
467{
468 pmd_t old;
469
470 preempt_disable();
471 old = pmdp_flush_direct(mm, addr, pmdp);
472 set_pmd(pmdp, new);
473 preempt_enable();
474 return old;
475}
476EXPORT_SYMBOL(pmdp_xchg_direct);
477
478pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
479 pmd_t *pmdp, pmd_t new)
480{
481 pmd_t old;
482
483 preempt_disable();
484 old = pmdp_flush_lazy(mm, addr, pmdp);
485 set_pmd(pmdp, new);
486 preempt_enable();
487 return old;
488}
489EXPORT_SYMBOL(pmdp_xchg_lazy);
490
491static inline void pudp_idte_local(struct mm_struct *mm,
492 unsigned long addr, pud_t *pudp)
493{
494 if (MACHINE_HAS_TLB_GUEST)
495 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
496 mm->context.asce, IDTE_LOCAL);
497 else
498 __pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
499}
500
501static inline void pudp_idte_global(struct mm_struct *mm,
502 unsigned long addr, pud_t *pudp)
503{
504 if (MACHINE_HAS_TLB_GUEST)
505 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
506 mm->context.asce, IDTE_GLOBAL);
507 else if (MACHINE_HAS_IDTE)
508 __pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
509 else
510 /*
511 * Invalid bit position is the same for pmd and pud, so we can
512 * re-use _pmd_csp() here
513 */
514 __pmdp_csp((pmd_t *) pudp);
515}
516
517static inline pud_t pudp_flush_direct(struct mm_struct *mm,
518 unsigned long addr, pud_t *pudp)
519{
520 pud_t old;
521
522 old = *pudp;
523 if (pud_val(old) & _REGION_ENTRY_INVALID)
524 return old;
525 atomic_inc(&mm->context.flush_count);
526 if (MACHINE_HAS_TLB_LC &&
527 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
528 pudp_idte_local(mm, addr, pudp);
529 else
530 pudp_idte_global(mm, addr, pudp);
531 atomic_dec(&mm->context.flush_count);
532 return old;
533}
534
535pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
536 pud_t *pudp, pud_t new)
537{
538 pud_t old;
539
540 preempt_disable();
541 old = pudp_flush_direct(mm, addr, pudp);
542 set_pud(pudp, new);
543 preempt_enable();
544 return old;
545}
546EXPORT_SYMBOL(pudp_xchg_direct);
547
548#ifdef CONFIG_TRANSPARENT_HUGEPAGE
549void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
550 pgtable_t pgtable)
551{
552 struct list_head *lh = (struct list_head *) pgtable;
553
554 assert_spin_locked(pmd_lockptr(mm, pmdp));
555
556 /* FIFO */
557 if (!pmd_huge_pte(mm, pmdp))
558 INIT_LIST_HEAD(lh);
559 else
560 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
561 pmd_huge_pte(mm, pmdp) = pgtable;
562}
563
564pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
565{
566 struct list_head *lh;
567 pgtable_t pgtable;
568 pte_t *ptep;
569
570 assert_spin_locked(pmd_lockptr(mm, pmdp));
571
572 /* FIFO */
573 pgtable = pmd_huge_pte(mm, pmdp);
574 lh = (struct list_head *) pgtable;
575 if (list_empty(lh))
576 pmd_huge_pte(mm, pmdp) = NULL;
577 else {
578 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
579 list_del(lh);
580 }
581 ptep = (pte_t *) pgtable;
582 set_pte(ptep, __pte(_PAGE_INVALID));
583 ptep++;
584 set_pte(ptep, __pte(_PAGE_INVALID));
585 return pgtable;
586}
587#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
588
589#ifdef CONFIG_PGSTE
590void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
591 pte_t *ptep, pte_t entry)
592{
593 pgste_t pgste;
594
595 /* the mm_has_pgste() check is done in set_pte_at() */
596 preempt_disable();
597 pgste = pgste_get_lock(ptep);
598 pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
599 pgste_set_key(ptep, pgste, entry, mm);
600 pgste = pgste_set_pte(ptep, pgste, entry);
601 pgste_set_unlock(ptep, pgste);
602 preempt_enable();
603}
604
605void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
606{
607 pgste_t pgste;
608
609 preempt_disable();
610 pgste = pgste_get_lock(ptep);
611 pgste_val(pgste) |= PGSTE_IN_BIT;
612 pgste_set_unlock(ptep, pgste);
613 preempt_enable();
614}
615
616/**
617 * ptep_force_prot - change access rights of a locked pte
618 * @mm: pointer to the process mm_struct
619 * @addr: virtual address in the guest address space
620 * @ptep: pointer to the page table entry
621 * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
622 * @bit: pgste bit to set (e.g. for notification)
623 *
624 * Returns 0 if the access rights were changed and -EAGAIN if the current
625 * and requested access rights are incompatible.
626 */
627int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
628 pte_t *ptep, int prot, unsigned long bit)
629{
630 pte_t entry;
631 pgste_t pgste;
632 int pte_i, pte_p, nodat;
633
634 pgste = pgste_get_lock(ptep);
635 entry = *ptep;
636 /* Check pte entry after all locks have been acquired */
637 pte_i = pte_val(entry) & _PAGE_INVALID;
638 pte_p = pte_val(entry) & _PAGE_PROTECT;
639 if ((pte_i && (prot != PROT_NONE)) ||
640 (pte_p && (prot & PROT_WRITE))) {
641 pgste_set_unlock(ptep, pgste);
642 return -EAGAIN;
643 }
644 /* Change access rights and set pgste bit */
645 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
646 if (prot == PROT_NONE && !pte_i) {
647 ptep_flush_direct(mm, addr, ptep, nodat);
648 pgste = pgste_update_all(entry, pgste, mm);
649 entry = set_pte_bit(entry, __pgprot(_PAGE_INVALID));
650 }
651 if (prot == PROT_READ && !pte_p) {
652 ptep_flush_direct(mm, addr, ptep, nodat);
653 entry = clear_pte_bit(entry, __pgprot(_PAGE_INVALID));
654 entry = set_pte_bit(entry, __pgprot(_PAGE_PROTECT));
655 }
656 pgste_val(pgste) |= bit;
657 pgste = pgste_set_pte(ptep, pgste, entry);
658 pgste_set_unlock(ptep, pgste);
659 return 0;
660}
661
662int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
663 pte_t *sptep, pte_t *tptep, pte_t pte)
664{
665 pgste_t spgste, tpgste;
666 pte_t spte, tpte;
667 int rc = -EAGAIN;
668
669 if (!(pte_val(*tptep) & _PAGE_INVALID))
670 return 0; /* already shadowed */
671 spgste = pgste_get_lock(sptep);
672 spte = *sptep;
673 if (!(pte_val(spte) & _PAGE_INVALID) &&
674 !((pte_val(spte) & _PAGE_PROTECT) &&
675 !(pte_val(pte) & _PAGE_PROTECT))) {
676 pgste_val(spgste) |= PGSTE_VSIE_BIT;
677 tpgste = pgste_get_lock(tptep);
678 tpte = __pte((pte_val(spte) & PAGE_MASK) |
679 (pte_val(pte) & _PAGE_PROTECT));
680 /* don't touch the storage key - it belongs to parent pgste */
681 tpgste = pgste_set_pte(tptep, tpgste, tpte);
682 pgste_set_unlock(tptep, tpgste);
683 rc = 1;
684 }
685 pgste_set_unlock(sptep, spgste);
686 return rc;
687}
688
689void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
690{
691 pgste_t pgste;
692 int nodat;
693
694 pgste = pgste_get_lock(ptep);
695 /* notifier is called by the caller */
696 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
697 ptep_flush_direct(mm, saddr, ptep, nodat);
698 /* don't touch the storage key - it belongs to parent pgste */
699 pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
700 pgste_set_unlock(ptep, pgste);
701}
702
703static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
704{
705 if (!non_swap_entry(entry))
706 dec_mm_counter(mm, MM_SWAPENTS);
707 else if (is_migration_entry(entry)) {
708 struct page *page = pfn_swap_entry_to_page(entry);
709
710 dec_mm_counter(mm, mm_counter(page));
711 }
712 free_swap_and_cache(entry);
713}
714
715void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
716 pte_t *ptep, int reset)
717{
718 unsigned long pgstev;
719 pgste_t pgste;
720 pte_t pte;
721
722 /* Zap unused and logically-zero pages */
723 preempt_disable();
724 pgste = pgste_get_lock(ptep);
725 pgstev = pgste_val(pgste);
726 pte = *ptep;
727 if (!reset && pte_swap(pte) &&
728 ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
729 (pgstev & _PGSTE_GPS_ZERO))) {
730 ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
731 pte_clear(mm, addr, ptep);
732 }
733 if (reset)
734 pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
735 pgste_set_unlock(ptep, pgste);
736 preempt_enable();
737}
738
739void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
740{
741 unsigned long ptev;
742 pgste_t pgste;
743
744 /* Clear storage key ACC and F, but set R/C */
745 preempt_disable();
746 pgste = pgste_get_lock(ptep);
747 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
748 pgste_val(pgste) |= PGSTE_GR_BIT | PGSTE_GC_BIT;
749 ptev = pte_val(*ptep);
750 if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
751 page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 0);
752 pgste_set_unlock(ptep, pgste);
753 preempt_enable();
754}
755
756/*
757 * Test and reset if a guest page is dirty
758 */
759bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long addr,
760 pte_t *ptep)
761{
762 pgste_t pgste;
763 pte_t pte;
764 bool dirty;
765 int nodat;
766
767 pgste = pgste_get_lock(ptep);
768 dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
769 pgste_val(pgste) &= ~PGSTE_UC_BIT;
770 pte = *ptep;
771 if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
772 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
773 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
774 ptep_ipte_global(mm, addr, ptep, nodat);
775 if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
776 pte = set_pte_bit(pte, __pgprot(_PAGE_PROTECT));
777 else
778 pte = set_pte_bit(pte, __pgprot(_PAGE_INVALID));
779 set_pte(ptep, pte);
780 }
781 pgste_set_unlock(ptep, pgste);
782 return dirty;
783}
784EXPORT_SYMBOL_GPL(ptep_test_and_clear_uc);
785
786int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
787 unsigned char key, bool nq)
788{
789 unsigned long keyul, paddr;
790 spinlock_t *ptl;
791 pgste_t old, new;
792 pmd_t *pmdp;
793 pte_t *ptep;
794
795 /*
796 * If we don't have a PTE table and if there is no huge page mapped,
797 * we can ignore attempts to set the key to 0, because it already is 0.
798 */
799 switch (pmd_lookup(mm, addr, &pmdp)) {
800 case -ENOENT:
801 return key ? -EFAULT : 0;
802 case 0:
803 break;
804 default:
805 return -EFAULT;
806 }
807
808 ptl = pmd_lock(mm, pmdp);
809 if (!pmd_present(*pmdp)) {
810 spin_unlock(ptl);
811 return key ? -EFAULT : 0;
812 }
813
814 if (pmd_large(*pmdp)) {
815 paddr = pmd_val(*pmdp) & HPAGE_MASK;
816 paddr |= addr & ~HPAGE_MASK;
817 /*
818 * Huge pmds need quiescing operations, they are
819 * always mapped.
820 */
821 page_set_storage_key(paddr, key, 1);
822 spin_unlock(ptl);
823 return 0;
824 }
825 spin_unlock(ptl);
826
827 ptep = pte_offset_map_lock(mm, pmdp, addr, &ptl);
828 new = old = pgste_get_lock(ptep);
829 pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
830 PGSTE_ACC_BITS | PGSTE_FP_BIT);
831 keyul = (unsigned long) key;
832 pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
833 pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
834 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
835 unsigned long bits, skey;
836
837 paddr = pte_val(*ptep) & PAGE_MASK;
838 skey = (unsigned long) page_get_storage_key(paddr);
839 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
840 skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
841 /* Set storage key ACC and FP */
842 page_set_storage_key(paddr, skey, !nq);
843 /* Merge host changed & referenced into pgste */
844 pgste_val(new) |= bits << 52;
845 }
846 /* changing the guest storage key is considered a change of the page */
847 if ((pgste_val(new) ^ pgste_val(old)) &
848 (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
849 pgste_val(new) |= PGSTE_UC_BIT;
850
851 pgste_set_unlock(ptep, new);
852 pte_unmap_unlock(ptep, ptl);
853 return 0;
854}
855EXPORT_SYMBOL(set_guest_storage_key);
856
857/*
858 * Conditionally set a guest storage key (handling csske).
859 * oldkey will be updated when either mr or mc is set and a pointer is given.
860 *
861 * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
862 * storage key was updated and -EFAULT on access errors.
863 */
864int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
865 unsigned char key, unsigned char *oldkey,
866 bool nq, bool mr, bool mc)
867{
868 unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
869 int rc;
870
871 /* we can drop the pgste lock between getting and setting the key */
872 if (mr | mc) {
873 rc = get_guest_storage_key(current->mm, addr, &tmp);
874 if (rc)
875 return rc;
876 if (oldkey)
877 *oldkey = tmp;
878 if (!mr)
879 mask |= _PAGE_REFERENCED;
880 if (!mc)
881 mask |= _PAGE_CHANGED;
882 if (!((tmp ^ key) & mask))
883 return 0;
884 }
885 rc = set_guest_storage_key(current->mm, addr, key, nq);
886 return rc < 0 ? rc : 1;
887}
888EXPORT_SYMBOL(cond_set_guest_storage_key);
889
890/*
891 * Reset a guest reference bit (rrbe), returning the reference and changed bit.
892 *
893 * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
894 */
895int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
896{
897 spinlock_t *ptl;
898 unsigned long paddr;
899 pgste_t old, new;
900 pmd_t *pmdp;
901 pte_t *ptep;
902 int cc = 0;
903
904 /*
905 * If we don't have a PTE table and if there is no huge page mapped,
906 * the storage key is 0 and there is nothing for us to do.
907 */
908 switch (pmd_lookup(mm, addr, &pmdp)) {
909 case -ENOENT:
910 return 0;
911 case 0:
912 break;
913 default:
914 return -EFAULT;
915 }
916
917 ptl = pmd_lock(mm, pmdp);
918 if (!pmd_present(*pmdp)) {
919 spin_unlock(ptl);
920 return 0;
921 }
922
923 if (pmd_large(*pmdp)) {
924 paddr = pmd_val(*pmdp) & HPAGE_MASK;
925 paddr |= addr & ~HPAGE_MASK;
926 cc = page_reset_referenced(paddr);
927 spin_unlock(ptl);
928 return cc;
929 }
930 spin_unlock(ptl);
931
932 ptep = pte_offset_map_lock(mm, pmdp, addr, &ptl);
933 new = old = pgste_get_lock(ptep);
934 /* Reset guest reference bit only */
935 pgste_val(new) &= ~PGSTE_GR_BIT;
936
937 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
938 paddr = pte_val(*ptep) & PAGE_MASK;
939 cc = page_reset_referenced(paddr);
940 /* Merge real referenced bit into host-set */
941 pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
942 }
943 /* Reflect guest's logical view, not physical */
944 cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
945 /* Changing the guest storage key is considered a change of the page */
946 if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
947 pgste_val(new) |= PGSTE_UC_BIT;
948
949 pgste_set_unlock(ptep, new);
950 pte_unmap_unlock(ptep, ptl);
951 return cc;
952}
953EXPORT_SYMBOL(reset_guest_reference_bit);
954
955int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
956 unsigned char *key)
957{
958 unsigned long paddr;
959 spinlock_t *ptl;
960 pgste_t pgste;
961 pmd_t *pmdp;
962 pte_t *ptep;
963
964 /*
965 * If we don't have a PTE table and if there is no huge page mapped,
966 * the storage key is 0.
967 */
968 *key = 0;
969
970 switch (pmd_lookup(mm, addr, &pmdp)) {
971 case -ENOENT:
972 return 0;
973 case 0:
974 break;
975 default:
976 return -EFAULT;
977 }
978
979 ptl = pmd_lock(mm, pmdp);
980 if (!pmd_present(*pmdp)) {
981 spin_unlock(ptl);
982 return 0;
983 }
984
985 if (pmd_large(*pmdp)) {
986 paddr = pmd_val(*pmdp) & HPAGE_MASK;
987 paddr |= addr & ~HPAGE_MASK;
988 *key = page_get_storage_key(paddr);
989 spin_unlock(ptl);
990 return 0;
991 }
992 spin_unlock(ptl);
993
994 ptep = pte_offset_map_lock(mm, pmdp, addr, &ptl);
995 pgste = pgste_get_lock(ptep);
996 *key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
997 paddr = pte_val(*ptep) & PAGE_MASK;
998 if (!(pte_val(*ptep) & _PAGE_INVALID))
999 *key = page_get_storage_key(paddr);
1000 /* Reflect guest's logical view, not physical */
1001 *key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
1002 pgste_set_unlock(ptep, pgste);
1003 pte_unmap_unlock(ptep, ptl);
1004 return 0;
1005}
1006EXPORT_SYMBOL(get_guest_storage_key);
1007
1008/**
1009 * pgste_perform_essa - perform ESSA actions on the PGSTE.
1010 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1011 * @hva: the host virtual address of the page whose PGSTE is to be processed
1012 * @orc: the specific action to perform, see the ESSA_SET_* macros.
1013 * @oldpte: the PTE will be saved there if the pointer is not NULL.
1014 * @oldpgste: the old PGSTE will be saved there if the pointer is not NULL.
1015 *
1016 * Return: 1 if the page is to be added to the CBRL, otherwise 0,
1017 * or < 0 in case of error. -EINVAL is returned for invalid values
1018 * of orc, -EFAULT for invalid addresses.
1019 */
1020int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
1021 unsigned long *oldpte, unsigned long *oldpgste)
1022{
1023 struct vm_area_struct *vma;
1024 unsigned long pgstev;
1025 spinlock_t *ptl;
1026 pgste_t pgste;
1027 pte_t *ptep;
1028 int res = 0;
1029
1030 WARN_ON_ONCE(orc > ESSA_MAX);
1031 if (unlikely(orc > ESSA_MAX))
1032 return -EINVAL;
1033
1034 vma = vma_lookup(mm, hva);
1035 if (!vma || is_vm_hugetlb_page(vma))
1036 return -EFAULT;
1037 ptep = get_locked_pte(mm, hva, &ptl);
1038 if (unlikely(!ptep))
1039 return -EFAULT;
1040 pgste = pgste_get_lock(ptep);
1041 pgstev = pgste_val(pgste);
1042 if (oldpte)
1043 *oldpte = pte_val(*ptep);
1044 if (oldpgste)
1045 *oldpgste = pgstev;
1046
1047 switch (orc) {
1048 case ESSA_GET_STATE:
1049 break;
1050 case ESSA_SET_STABLE:
1051 pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
1052 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1053 break;
1054 case ESSA_SET_UNUSED:
1055 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1056 pgstev |= _PGSTE_GPS_USAGE_UNUSED;
1057 if (pte_val(*ptep) & _PAGE_INVALID)
1058 res = 1;
1059 break;
1060 case ESSA_SET_VOLATILE:
1061 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1062 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1063 if (pte_val(*ptep) & _PAGE_INVALID)
1064 res = 1;
1065 break;
1066 case ESSA_SET_POT_VOLATILE:
1067 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1068 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1069 pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
1070 break;
1071 }
1072 if (pgstev & _PGSTE_GPS_ZERO) {
1073 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1074 break;
1075 }
1076 if (!(pgstev & PGSTE_GC_BIT)) {
1077 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1078 res = 1;
1079 break;
1080 }
1081 break;
1082 case ESSA_SET_STABLE_RESIDENT:
1083 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1084 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1085 /*
1086 * Since the resident state can go away any time after this
1087 * call, we will not make this page resident. We can revisit
1088 * this decision if a guest will ever start using this.
1089 */
1090 break;
1091 case ESSA_SET_STABLE_IF_RESIDENT:
1092 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1093 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1094 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1095 }
1096 break;
1097 case ESSA_SET_STABLE_NODAT:
1098 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1099 pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
1100 break;
1101 default:
1102 /* we should never get here! */
1103 break;
1104 }
1105 /* If we are discarding a page, set it to logical zero */
1106 if (res)
1107 pgstev |= _PGSTE_GPS_ZERO;
1108
1109 pgste_val(pgste) = pgstev;
1110 pgste_set_unlock(ptep, pgste);
1111 pte_unmap_unlock(ptep, ptl);
1112 return res;
1113}
1114EXPORT_SYMBOL(pgste_perform_essa);
1115
1116/**
1117 * set_pgste_bits - set specific PGSTE bits.
1118 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1119 * @hva: the host virtual address of the page whose PGSTE is to be processed
1120 * @bits: a bitmask representing the bits that will be touched
1121 * @value: the values of the bits to be written. Only the bits in the mask
1122 * will be written.
1123 *
1124 * Return: 0 on success, < 0 in case of error.
1125 */
1126int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
1127 unsigned long bits, unsigned long value)
1128{
1129 struct vm_area_struct *vma;
1130 spinlock_t *ptl;
1131 pgste_t new;
1132 pte_t *ptep;
1133
1134 vma = vma_lookup(mm, hva);
1135 if (!vma || is_vm_hugetlb_page(vma))
1136 return -EFAULT;
1137 ptep = get_locked_pte(mm, hva, &ptl);
1138 if (unlikely(!ptep))
1139 return -EFAULT;
1140 new = pgste_get_lock(ptep);
1141
1142 pgste_val(new) &= ~bits;
1143 pgste_val(new) |= value & bits;
1144
1145 pgste_set_unlock(ptep, new);
1146 pte_unmap_unlock(ptep, ptl);
1147 return 0;
1148}
1149EXPORT_SYMBOL(set_pgste_bits);
1150
1151/**
1152 * get_pgste - get the current PGSTE for the given address.
1153 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1154 * @hva: the host virtual address of the page whose PGSTE is to be processed
1155 * @pgstep: will be written with the current PGSTE for the given address.
1156 *
1157 * Return: 0 on success, < 0 in case of error.
1158 */
1159int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1160{
1161 struct vm_area_struct *vma;
1162 spinlock_t *ptl;
1163 pte_t *ptep;
1164
1165 vma = vma_lookup(mm, hva);
1166 if (!vma || is_vm_hugetlb_page(vma))
1167 return -EFAULT;
1168 ptep = get_locked_pte(mm, hva, &ptl);
1169 if (unlikely(!ptep))
1170 return -EFAULT;
1171 *pgstep = pgste_val(pgste_get(ptep));
1172 pte_unmap_unlock(ptep, ptl);
1173 return 0;
1174}
1175EXPORT_SYMBOL(get_pgste);
1176#endif
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
29static 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
49static 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
69static 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
88static 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
108static 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
127static 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
139static 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
148static 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
155static 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_use_skey(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
176static 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_use_skey(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
198static 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
221static 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
237static 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
249static 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
270pte_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}
285EXPORT_SYMBOL(ptep_xchg_direct);
286
287pte_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}
302EXPORT_SYMBOL(ptep_xchg_lazy);
303
304pte_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}
321EXPORT_SYMBOL(ptep_modify_prot_start);
322
323void 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}
340EXPORT_SYMBOL(ptep_modify_prot_commit);
341
342static 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
352static 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
364static 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
382static 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
402pmd_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}
413EXPORT_SYMBOL(pmdp_xchg_direct);
414
415pmd_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}
426EXPORT_SYMBOL(pmdp_xchg_lazy);
427
428static 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
438static 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
454static 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
472pud_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}
483EXPORT_SYMBOL(pudp_xchg_direct);
484
485#ifdef CONFIG_TRANSPARENT_HUGEPAGE
486void 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
501pgtable_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
527void 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
542void 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 */
564int 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
599int 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
626void 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
640static 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
652void 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
676void 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 */
696bool 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}
749EXPORT_SYMBOL_GPL(test_and_clear_guest_dirty);
750
751int 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}
790EXPORT_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 */
799int 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}
823EXPORT_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 */
830int 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}
860EXPORT_SYMBOL(reset_guest_reference_bit);
861
862int 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}
883EXPORT_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 */
897int 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}
986EXPORT_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 */
998int 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}
1017EXPORT_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 */
1027int 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}
1039EXPORT_SYMBOL(get_pgste);
1040#endif