1 /*
2 * Copyright IBM Corp. 2007, 2011
3 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
4 */
5
6 #include <linux/sched.h>
7 #include <linux/kernel.h>
8 #include <linux/errno.h>
9 #include <linux/gfp.h>
10 #include <linux/mm.h>
11 #include <linux/swap.h>
12 #include <linux/smp.h>
13 #include <linux/spinlock.h>
14 #include <linux/rcupdate.h>
15 #include <linux/slab.h>
16 #include <linux/swapops.h>
17 #include <linux/sysctl.h>
18 #include <linux/ksm.h>
19 #include <linux/mman.h>
20
21 #include <asm/pgtable.h>
22 #include <asm/pgalloc.h>
23 #include <asm/tlb.h>
24 #include <asm/tlbflush.h>
25 #include <asm/mmu_context.h>
26
ptep_flush_direct(struct mm_struct * mm,unsigned long addr,pte_t * ptep)27 static inline pte_t ptep_flush_direct(struct mm_struct *mm,
28 unsigned long addr, pte_t *ptep)
29 {
30 pte_t old;
31
32 old = *ptep;
33 if (unlikely(pte_val(old) & _PAGE_INVALID))
34 return old;
35 atomic_inc(&mm->context.flush_count);
36 if (MACHINE_HAS_TLB_LC &&
37 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
38 __ptep_ipte(addr, ptep, IPTE_LOCAL);
39 else
40 __ptep_ipte(addr, ptep, IPTE_GLOBAL);
41 atomic_dec(&mm->context.flush_count);
42 return old;
43 }
44
ptep_flush_lazy(struct mm_struct * mm,unsigned long addr,pte_t * ptep)45 static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
46 unsigned long addr, pte_t *ptep)
47 {
48 pte_t old;
49
50 old = *ptep;
51 if (unlikely(pte_val(old) & _PAGE_INVALID))
52 return old;
53 atomic_inc(&mm->context.flush_count);
54 if (cpumask_equal(&mm->context.cpu_attach_mask,
55 cpumask_of(smp_processor_id()))) {
56 pte_val(*ptep) |= _PAGE_INVALID;
57 mm->context.flush_mm = 1;
58 } else
59 __ptep_ipte(addr, ptep, IPTE_GLOBAL);
60 atomic_dec(&mm->context.flush_count);
61 return old;
62 }
63
pgste_get_lock(pte_t * ptep)64 static inline pgste_t pgste_get_lock(pte_t *ptep)
65 {
66 unsigned long new = 0;
67 #ifdef CONFIG_PGSTE
68 unsigned long old;
69
70 asm(
71 " lg %0,%2\n"
72 "0: lgr %1,%0\n"
73 " nihh %0,0xff7f\n" /* clear PCL bit in old */
74 " oihh %1,0x0080\n" /* set PCL bit in new */
75 " csg %0,%1,%2\n"
76 " jl 0b\n"
77 : "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
78 : "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
79 #endif
80 return __pgste(new);
81 }
82
pgste_set_unlock(pte_t * ptep,pgste_t pgste)83 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
84 {
85 #ifdef CONFIG_PGSTE
86 asm(
87 " nihh %1,0xff7f\n" /* clear PCL bit */
88 " stg %1,%0\n"
89 : "=Q" (ptep[PTRS_PER_PTE])
90 : "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
91 : "cc", "memory");
92 #endif
93 }
94
pgste_get(pte_t * ptep)95 static inline pgste_t pgste_get(pte_t *ptep)
96 {
97 unsigned long pgste = 0;
98 #ifdef CONFIG_PGSTE
99 pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
100 #endif
101 return __pgste(pgste);
102 }
103
pgste_set(pte_t * ptep,pgste_t pgste)104 static inline void pgste_set(pte_t *ptep, pgste_t pgste)
105 {
106 #ifdef CONFIG_PGSTE
107 *(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
108 #endif
109 }
110
pgste_update_all(pte_t pte,pgste_t pgste,struct mm_struct * mm)111 static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
112 struct mm_struct *mm)
113 {
114 #ifdef CONFIG_PGSTE
115 unsigned long address, bits, skey;
116
117 if (!mm_use_skey(mm) || pte_val(pte) & _PAGE_INVALID)
118 return pgste;
119 address = pte_val(pte) & PAGE_MASK;
120 skey = (unsigned long) page_get_storage_key(address);
121 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
122 /* Transfer page changed & referenced bit to guest bits in pgste */
123 pgste_val(pgste) |= bits << 48; /* GR bit & GC bit */
124 /* Copy page access key and fetch protection bit to pgste */
125 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
126 pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
127 #endif
128 return pgste;
129
130 }
131
pgste_set_key(pte_t * ptep,pgste_t pgste,pte_t entry,struct mm_struct * mm)132 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
133 struct mm_struct *mm)
134 {
135 #ifdef CONFIG_PGSTE
136 unsigned long address;
137 unsigned long nkey;
138
139 if (!mm_use_skey(mm) || pte_val(entry) & _PAGE_INVALID)
140 return;
141 VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
142 address = pte_val(entry) & PAGE_MASK;
143 /*
144 * Set page access key and fetch protection bit from pgste.
145 * The guest C/R information is still in the PGSTE, set real
146 * key C/R to 0.
147 */
148 nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
149 nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
150 page_set_storage_key(address, nkey, 0);
151 #endif
152 }
153
pgste_set_pte(pte_t * ptep,pgste_t pgste,pte_t entry)154 static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
155 {
156 #ifdef CONFIG_PGSTE
157 if ((pte_val(entry) & _PAGE_PRESENT) &&
158 (pte_val(entry) & _PAGE_WRITE) &&
159 !(pte_val(entry) & _PAGE_INVALID)) {
160 if (!MACHINE_HAS_ESOP) {
161 /*
162 * Without enhanced suppression-on-protection force
163 * the dirty bit on for all writable ptes.
164 */
165 pte_val(entry) |= _PAGE_DIRTY;
166 pte_val(entry) &= ~_PAGE_PROTECT;
167 }
168 if (!(pte_val(entry) & _PAGE_PROTECT))
169 /* This pte allows write access, set user-dirty */
170 pgste_val(pgste) |= PGSTE_UC_BIT;
171 }
172 #endif
173 *ptep = entry;
174 return pgste;
175 }
176
pgste_pte_notify(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pgste_t pgste)177 static inline pgste_t pgste_pte_notify(struct mm_struct *mm,
178 unsigned long addr,
179 pte_t *ptep, pgste_t pgste)
180 {
181 #ifdef CONFIG_PGSTE
182 unsigned long bits;
183
184 bits = pgste_val(pgste) & (PGSTE_IN_BIT | PGSTE_VSIE_BIT);
185 if (bits) {
186 pgste_val(pgste) ^= bits;
187 ptep_notify(mm, addr, ptep, bits);
188 }
189 #endif
190 return pgste;
191 }
192
ptep_xchg_start(struct mm_struct * mm,unsigned long addr,pte_t * ptep)193 static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
194 unsigned long addr, pte_t *ptep)
195 {
196 pgste_t pgste = __pgste(0);
197
198 if (mm_has_pgste(mm)) {
199 pgste = pgste_get_lock(ptep);
200 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
201 }
202 return pgste;
203 }
204
ptep_xchg_commit(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pgste_t pgste,pte_t old,pte_t new)205 static inline pte_t ptep_xchg_commit(struct mm_struct *mm,
206 unsigned long addr, pte_t *ptep,
207 pgste_t pgste, pte_t old, pte_t new)
208 {
209 if (mm_has_pgste(mm)) {
210 if (pte_val(old) & _PAGE_INVALID)
211 pgste_set_key(ptep, pgste, new, mm);
212 if (pte_val(new) & _PAGE_INVALID) {
213 pgste = pgste_update_all(old, pgste, mm);
214 if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
215 _PGSTE_GPS_USAGE_UNUSED)
216 pte_val(old) |= _PAGE_UNUSED;
217 }
218 pgste = pgste_set_pte(ptep, pgste, new);
219 pgste_set_unlock(ptep, pgste);
220 } else {
221 *ptep = new;
222 }
223 return old;
224 }
225
ptep_xchg_direct(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t new)226 pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
227 pte_t *ptep, pte_t new)
228 {
229 pgste_t pgste;
230 pte_t old;
231
232 preempt_disable();
233 pgste = ptep_xchg_start(mm, addr, ptep);
234 old = ptep_flush_direct(mm, addr, ptep);
235 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
236 preempt_enable();
237 return old;
238 }
239 EXPORT_SYMBOL(ptep_xchg_direct);
240
ptep_xchg_lazy(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t new)241 pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
242 pte_t *ptep, pte_t new)
243 {
244 pgste_t pgste;
245 pte_t old;
246
247 preempt_disable();
248 pgste = ptep_xchg_start(mm, addr, ptep);
249 old = ptep_flush_lazy(mm, addr, ptep);
250 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
251 preempt_enable();
252 return old;
253 }
254 EXPORT_SYMBOL(ptep_xchg_lazy);
255
ptep_modify_prot_start(struct mm_struct * mm,unsigned long addr,pte_t * ptep)256 pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
257 pte_t *ptep)
258 {
259 pgste_t pgste;
260 pte_t old;
261
262 preempt_disable();
263 pgste = ptep_xchg_start(mm, addr, ptep);
264 old = ptep_flush_lazy(mm, addr, ptep);
265 if (mm_has_pgste(mm)) {
266 pgste = pgste_update_all(old, pgste, mm);
267 pgste_set(ptep, pgste);
268 }
269 return old;
270 }
271 EXPORT_SYMBOL(ptep_modify_prot_start);
272
ptep_modify_prot_commit(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte)273 void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
274 pte_t *ptep, pte_t pte)
275 {
276 pgste_t pgste;
277
278 if (mm_has_pgste(mm)) {
279 pgste = pgste_get(ptep);
280 pgste_set_key(ptep, pgste, pte, mm);
281 pgste = pgste_set_pte(ptep, pgste, pte);
282 pgste_set_unlock(ptep, pgste);
283 } else {
284 *ptep = pte;
285 }
286 preempt_enable();
287 }
288 EXPORT_SYMBOL(ptep_modify_prot_commit);
289
pmdp_flush_direct(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp)290 static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
291 unsigned long addr, pmd_t *pmdp)
292 {
293 pmd_t old;
294
295 old = *pmdp;
296 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
297 return old;
298 if (!MACHINE_HAS_IDTE) {
299 __pmdp_csp(pmdp);
300 return old;
301 }
302 atomic_inc(&mm->context.flush_count);
303 if (MACHINE_HAS_TLB_LC &&
304 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
305 __pmdp_idte(addr, pmdp, IDTE_LOCAL);
306 else
307 __pmdp_idte(addr, pmdp, IDTE_GLOBAL);
308 atomic_dec(&mm->context.flush_count);
309 return old;
310 }
311
pmdp_flush_lazy(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp)312 static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
313 unsigned long addr, pmd_t *pmdp)
314 {
315 pmd_t old;
316
317 old = *pmdp;
318 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
319 return old;
320 atomic_inc(&mm->context.flush_count);
321 if (cpumask_equal(&mm->context.cpu_attach_mask,
322 cpumask_of(smp_processor_id()))) {
323 pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
324 mm->context.flush_mm = 1;
325 } else if (MACHINE_HAS_IDTE)
326 __pmdp_idte(addr, pmdp, IDTE_GLOBAL);
327 else
328 __pmdp_csp(pmdp);
329 atomic_dec(&mm->context.flush_count);
330 return old;
331 }
332
pmdp_xchg_direct(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp,pmd_t new)333 pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
334 pmd_t *pmdp, pmd_t new)
335 {
336 pmd_t old;
337
338 preempt_disable();
339 old = pmdp_flush_direct(mm, addr, pmdp);
340 *pmdp = new;
341 preempt_enable();
342 return old;
343 }
344 EXPORT_SYMBOL(pmdp_xchg_direct);
345
pmdp_xchg_lazy(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp,pmd_t new)346 pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
347 pmd_t *pmdp, pmd_t new)
348 {
349 pmd_t old;
350
351 preempt_disable();
352 old = pmdp_flush_lazy(mm, addr, pmdp);
353 *pmdp = new;
354 preempt_enable();
355 return old;
356 }
357 EXPORT_SYMBOL(pmdp_xchg_lazy);
358
pudp_flush_direct(struct mm_struct * mm,unsigned long addr,pud_t * pudp)359 static inline pud_t pudp_flush_direct(struct mm_struct *mm,
360 unsigned long addr, pud_t *pudp)
361 {
362 pud_t old;
363
364 old = *pudp;
365 if (pud_val(old) & _REGION_ENTRY_INVALID)
366 return old;
367 if (!MACHINE_HAS_IDTE) {
368 /*
369 * Invalid bit position is the same for pmd and pud, so we can
370 * re-use _pmd_csp() here
371 */
372 __pmdp_csp((pmd_t *) pudp);
373 return old;
374 }
375 atomic_inc(&mm->context.flush_count);
376 if (MACHINE_HAS_TLB_LC &&
377 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
378 __pudp_idte(addr, pudp, IDTE_LOCAL);
379 else
380 __pudp_idte(addr, pudp, IDTE_GLOBAL);
381 atomic_dec(&mm->context.flush_count);
382 return old;
383 }
384
pudp_xchg_direct(struct mm_struct * mm,unsigned long addr,pud_t * pudp,pud_t new)385 pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
386 pud_t *pudp, pud_t new)
387 {
388 pud_t old;
389
390 preempt_disable();
391 old = pudp_flush_direct(mm, addr, pudp);
392 *pudp = new;
393 preempt_enable();
394 return old;
395 }
396 EXPORT_SYMBOL(pudp_xchg_direct);
397
398 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
pgtable_trans_huge_deposit(struct mm_struct * mm,pmd_t * pmdp,pgtable_t pgtable)399 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
400 pgtable_t pgtable)
401 {
402 struct list_head *lh = (struct list_head *) pgtable;
403
404 assert_spin_locked(pmd_lockptr(mm, pmdp));
405
406 /* FIFO */
407 if (!pmd_huge_pte(mm, pmdp))
408 INIT_LIST_HEAD(lh);
409 else
410 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
411 pmd_huge_pte(mm, pmdp) = pgtable;
412 }
413
pgtable_trans_huge_withdraw(struct mm_struct * mm,pmd_t * pmdp)414 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
415 {
416 struct list_head *lh;
417 pgtable_t pgtable;
418 pte_t *ptep;
419
420 assert_spin_locked(pmd_lockptr(mm, pmdp));
421
422 /* FIFO */
423 pgtable = pmd_huge_pte(mm, pmdp);
424 lh = (struct list_head *) pgtable;
425 if (list_empty(lh))
426 pmd_huge_pte(mm, pmdp) = NULL;
427 else {
428 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
429 list_del(lh);
430 }
431 ptep = (pte_t *) pgtable;
432 pte_val(*ptep) = _PAGE_INVALID;
433 ptep++;
434 pte_val(*ptep) = _PAGE_INVALID;
435 return pgtable;
436 }
437 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
438
439 #ifdef CONFIG_PGSTE
ptep_set_pte_at(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t entry)440 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
441 pte_t *ptep, pte_t entry)
442 {
443 pgste_t pgste;
444
445 /* the mm_has_pgste() check is done in set_pte_at() */
446 preempt_disable();
447 pgste = pgste_get_lock(ptep);
448 pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
449 pgste_set_key(ptep, pgste, entry, mm);
450 pgste = pgste_set_pte(ptep, pgste, entry);
451 pgste_set_unlock(ptep, pgste);
452 preempt_enable();
453 }
454
ptep_set_notify(struct mm_struct * mm,unsigned long addr,pte_t * ptep)455 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
456 {
457 pgste_t pgste;
458
459 preempt_disable();
460 pgste = pgste_get_lock(ptep);
461 pgste_val(pgste) |= PGSTE_IN_BIT;
462 pgste_set_unlock(ptep, pgste);
463 preempt_enable();
464 }
465
466 /**
467 * ptep_force_prot - change access rights of a locked pte
468 * @mm: pointer to the process mm_struct
469 * @addr: virtual address in the guest address space
470 * @ptep: pointer to the page table entry
471 * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
472 * @bit: pgste bit to set (e.g. for notification)
473 *
474 * Returns 0 if the access rights were changed and -EAGAIN if the current
475 * and requested access rights are incompatible.
476 */
ptep_force_prot(struct mm_struct * mm,unsigned long addr,pte_t * ptep,int prot,unsigned long bit)477 int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
478 pte_t *ptep, int prot, unsigned long bit)
479 {
480 pte_t entry;
481 pgste_t pgste;
482 int pte_i, pte_p;
483
484 pgste = pgste_get_lock(ptep);
485 entry = *ptep;
486 /* Check pte entry after all locks have been acquired */
487 pte_i = pte_val(entry) & _PAGE_INVALID;
488 pte_p = pte_val(entry) & _PAGE_PROTECT;
489 if ((pte_i && (prot != PROT_NONE)) ||
490 (pte_p && (prot & PROT_WRITE))) {
491 pgste_set_unlock(ptep, pgste);
492 return -EAGAIN;
493 }
494 /* Change access rights and set pgste bit */
495 if (prot == PROT_NONE && !pte_i) {
496 ptep_flush_direct(mm, addr, ptep);
497 pgste = pgste_update_all(entry, pgste, mm);
498 pte_val(entry) |= _PAGE_INVALID;
499 }
500 if (prot == PROT_READ && !pte_p) {
501 ptep_flush_direct(mm, addr, ptep);
502 pte_val(entry) &= ~_PAGE_INVALID;
503 pte_val(entry) |= _PAGE_PROTECT;
504 }
505 pgste_val(pgste) |= bit;
506 pgste = pgste_set_pte(ptep, pgste, entry);
507 pgste_set_unlock(ptep, pgste);
508 return 0;
509 }
510
ptep_shadow_pte(struct mm_struct * mm,unsigned long saddr,pte_t * sptep,pte_t * tptep,pte_t pte)511 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
512 pte_t *sptep, pte_t *tptep, pte_t pte)
513 {
514 pgste_t spgste, tpgste;
515 pte_t spte, tpte;
516 int rc = -EAGAIN;
517
518 if (!(pte_val(*tptep) & _PAGE_INVALID))
519 return 0; /* already shadowed */
520 spgste = pgste_get_lock(sptep);
521 spte = *sptep;
522 if (!(pte_val(spte) & _PAGE_INVALID) &&
523 !((pte_val(spte) & _PAGE_PROTECT) &&
524 !(pte_val(pte) & _PAGE_PROTECT))) {
525 pgste_val(spgste) |= PGSTE_VSIE_BIT;
526 tpgste = pgste_get_lock(tptep);
527 pte_val(tpte) = (pte_val(spte) & PAGE_MASK) |
528 (pte_val(pte) & _PAGE_PROTECT);
529 /* don't touch the storage key - it belongs to parent pgste */
530 tpgste = pgste_set_pte(tptep, tpgste, tpte);
531 pgste_set_unlock(tptep, tpgste);
532 rc = 1;
533 }
534 pgste_set_unlock(sptep, spgste);
535 return rc;
536 }
537
ptep_unshadow_pte(struct mm_struct * mm,unsigned long saddr,pte_t * ptep)538 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
539 {
540 pgste_t pgste;
541
542 pgste = pgste_get_lock(ptep);
543 /* notifier is called by the caller */
544 ptep_flush_direct(mm, saddr, ptep);
545 /* don't touch the storage key - it belongs to parent pgste */
546 pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
547 pgste_set_unlock(ptep, pgste);
548 }
549
ptep_zap_swap_entry(struct mm_struct * mm,swp_entry_t entry)550 static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
551 {
552 if (!non_swap_entry(entry))
553 dec_mm_counter(mm, MM_SWAPENTS);
554 else if (is_migration_entry(entry)) {
555 struct page *page = migration_entry_to_page(entry);
556
557 dec_mm_counter(mm, mm_counter(page));
558 }
559 free_swap_and_cache(entry);
560 }
561
ptep_zap_unused(struct mm_struct * mm,unsigned long addr,pte_t * ptep,int reset)562 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
563 pte_t *ptep, int reset)
564 {
565 unsigned long pgstev;
566 pgste_t pgste;
567 pte_t pte;
568
569 /* Zap unused and logically-zero pages */
570 preempt_disable();
571 pgste = pgste_get_lock(ptep);
572 pgstev = pgste_val(pgste);
573 pte = *ptep;
574 if (!reset && pte_swap(pte) &&
575 ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
576 (pgstev & _PGSTE_GPS_ZERO))) {
577 ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
578 pte_clear(mm, addr, ptep);
579 }
580 if (reset)
581 pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
582 pgste_set_unlock(ptep, pgste);
583 preempt_enable();
584 }
585
ptep_zap_key(struct mm_struct * mm,unsigned long addr,pte_t * ptep)586 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
587 {
588 unsigned long ptev;
589 pgste_t pgste;
590
591 /* Clear storage key */
592 preempt_disable();
593 pgste = pgste_get_lock(ptep);
594 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT |
595 PGSTE_GR_BIT | PGSTE_GC_BIT);
596 ptev = pte_val(*ptep);
597 if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
598 page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 1);
599 pgste_set_unlock(ptep, pgste);
600 preempt_enable();
601 }
602
603 /*
604 * Test and reset if a guest page is dirty
605 */
test_and_clear_guest_dirty(struct mm_struct * mm,unsigned long addr)606 bool test_and_clear_guest_dirty(struct mm_struct *mm, unsigned long addr)
607 {
608 spinlock_t *ptl;
609 pgd_t *pgd;
610 pud_t *pud;
611 pmd_t *pmd;
612 pgste_t pgste;
613 pte_t *ptep;
614 pte_t pte;
615 bool dirty;
616
617 pgd = pgd_offset(mm, addr);
618 pud = pud_alloc(mm, pgd, addr);
619 if (!pud)
620 return false;
621 pmd = pmd_alloc(mm, pud, addr);
622 if (!pmd)
623 return false;
624 /* We can't run guests backed by huge pages, but userspace can
625 * still set them up and then try to migrate them without any
626 * migration support.
627 */
628 if (pmd_large(*pmd))
629 return true;
630
631 ptep = pte_alloc_map_lock(mm, pmd, addr, &ptl);
632 if (unlikely(!ptep))
633 return false;
634
635 pgste = pgste_get_lock(ptep);
636 dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
637 pgste_val(pgste) &= ~PGSTE_UC_BIT;
638 pte = *ptep;
639 if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
640 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
641 __ptep_ipte(addr, ptep, IPTE_GLOBAL);
642 if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
643 pte_val(pte) |= _PAGE_PROTECT;
644 else
645 pte_val(pte) |= _PAGE_INVALID;
646 *ptep = pte;
647 }
648 pgste_set_unlock(ptep, pgste);
649
650 spin_unlock(ptl);
651 return dirty;
652 }
653 EXPORT_SYMBOL_GPL(test_and_clear_guest_dirty);
654
set_guest_storage_key(struct mm_struct * mm,unsigned long addr,unsigned char key,bool nq)655 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
656 unsigned char key, bool nq)
657 {
658 unsigned long keyul;
659 spinlock_t *ptl;
660 pgste_t old, new;
661 pte_t *ptep;
662
663 ptep = get_locked_pte(mm, addr, &ptl);
664 if (unlikely(!ptep))
665 return -EFAULT;
666
667 new = old = pgste_get_lock(ptep);
668 pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
669 PGSTE_ACC_BITS | PGSTE_FP_BIT);
670 keyul = (unsigned long) key;
671 pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
672 pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
673 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
674 unsigned long address, bits, skey;
675
676 address = pte_val(*ptep) & PAGE_MASK;
677 skey = (unsigned long) page_get_storage_key(address);
678 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
679 skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
680 /* Set storage key ACC and FP */
681 page_set_storage_key(address, skey, !nq);
682 /* Merge host changed & referenced into pgste */
683 pgste_val(new) |= bits << 52;
684 }
685 /* changing the guest storage key is considered a change of the page */
686 if ((pgste_val(new) ^ pgste_val(old)) &
687 (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
688 pgste_val(new) |= PGSTE_UC_BIT;
689
690 pgste_set_unlock(ptep, new);
691 pte_unmap_unlock(ptep, ptl);
692 return 0;
693 }
694 EXPORT_SYMBOL(set_guest_storage_key);
695
696 /**
697 * Conditionally set a guest storage key (handling csske).
698 * oldkey will be updated when either mr or mc is set and a pointer is given.
699 *
700 * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
701 * storage key was updated and -EFAULT on access errors.
702 */
cond_set_guest_storage_key(struct mm_struct * mm,unsigned long addr,unsigned char key,unsigned char * oldkey,bool nq,bool mr,bool mc)703 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
704 unsigned char key, unsigned char *oldkey,
705 bool nq, bool mr, bool mc)
706 {
707 unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
708 int rc;
709
710 /* we can drop the pgste lock between getting and setting the key */
711 if (mr | mc) {
712 rc = get_guest_storage_key(current->mm, addr, &tmp);
713 if (rc)
714 return rc;
715 if (oldkey)
716 *oldkey = tmp;
717 if (!mr)
718 mask |= _PAGE_REFERENCED;
719 if (!mc)
720 mask |= _PAGE_CHANGED;
721 if (!((tmp ^ key) & mask))
722 return 0;
723 }
724 rc = set_guest_storage_key(current->mm, addr, key, nq);
725 return rc < 0 ? rc : 1;
726 }
727 EXPORT_SYMBOL(cond_set_guest_storage_key);
728
729 /**
730 * Reset a guest reference bit (rrbe), returning the reference and changed bit.
731 *
732 * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
733 */
reset_guest_reference_bit(struct mm_struct * mm,unsigned long addr)734 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
735 {
736 spinlock_t *ptl;
737 pgste_t old, new;
738 pte_t *ptep;
739 int cc = 0;
740
741 ptep = get_locked_pte(mm, addr, &ptl);
742 if (unlikely(!ptep))
743 return -EFAULT;
744
745 new = old = pgste_get_lock(ptep);
746 /* Reset guest reference bit only */
747 pgste_val(new) &= ~PGSTE_GR_BIT;
748
749 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
750 cc = page_reset_referenced(pte_val(*ptep) & PAGE_MASK);
751 /* Merge real referenced bit into host-set */
752 pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
753 }
754 /* Reflect guest's logical view, not physical */
755 cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
756 /* Changing the guest storage key is considered a change of the page */
757 if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
758 pgste_val(new) |= PGSTE_UC_BIT;
759
760 pgste_set_unlock(ptep, new);
761 pte_unmap_unlock(ptep, ptl);
762 return 0;
763 }
764 EXPORT_SYMBOL(reset_guest_reference_bit);
765
get_guest_storage_key(struct mm_struct * mm,unsigned long addr,unsigned char * key)766 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
767 unsigned char *key)
768 {
769 spinlock_t *ptl;
770 pgste_t pgste;
771 pte_t *ptep;
772
773 ptep = get_locked_pte(mm, addr, &ptl);
774 if (unlikely(!ptep))
775 return -EFAULT;
776
777 pgste = pgste_get_lock(ptep);
778 *key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
779 if (!(pte_val(*ptep) & _PAGE_INVALID))
780 *key = page_get_storage_key(pte_val(*ptep) & PAGE_MASK);
781 /* Reflect guest's logical view, not physical */
782 *key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
783 pgste_set_unlock(ptep, pgste);
784 pte_unmap_unlock(ptep, ptl);
785 return 0;
786 }
787 EXPORT_SYMBOL(get_guest_storage_key);
788 #endif
789